super-intel.c

   1 /*
   2  * mdadm - Intel(R) Matrix Storage Manager Support
   3  *
   4  * Copyright (C) 2002-2008 Intel Corporation
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms and conditions of the GNU General Public License,
   8  * version 2, as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along with
  16  * this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19
  20 #define HAVE_STDINT_H 1
  21 #include "mdadm.h"
  22 #include "mdmon.h"
  23 #include "sha1.h"
  24 #include "platform-intel.h"
  25 #include <values.h>
  26 #include <scsi/sg.h>
  27 #include <ctype.h>
  28 #include <dirent.h>
  29
  30 /* MPB == Metadata Parameter Block */
  31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
  32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
  33 #define MPB_VERSION_RAID0 "1.0.00"
  34 #define MPB_VERSION_RAID1 "1.1.00"
  35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
  36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
  37 #define MPB_VERSION_RAID5 "1.2.02"
  38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
  39 #define MPB_VERSION_CNG "1.2.06"
  40 #define MPB_VERSION_ATTRIBS "1.3.00"
  41 #define MAX_SIGNATURE_LENGTH  32
  42 #define MAX_RAID_SERIAL_LEN   16
  43
  44 /* supports RAID0 */
  45 #define MPB_ATTRIB_RAID0                __cpu_to_le32(0x00000001)
  46 /* supports RAID1 */
  47 #define MPB_ATTRIB_RAID1                __cpu_to_le32(0x00000002)
  48 /* supports RAID10 */
  49 #define MPB_ATTRIB_RAID10               __cpu_to_le32(0x00000004)
  50 /* supports RAID1E */
  51 #define MPB_ATTRIB_RAID1E               __cpu_to_le32(0x00000008)
  52 /* supports RAID5 */
  53 #define MPB_ATTRIB_RAID5                __cpu_to_le32(0x00000010)
  54 /* supports RAID CNG */
  55 #define MPB_ATTRIB_RAIDCNG              __cpu_to_le32(0x00000020)
  56 /* supports expanded stripe sizes of  256K, 512K and 1MB */
  57 #define MPB_ATTRIB_EXP_STRIPE_SIZE      __cpu_to_le32(0x00000040)
  58
  59 /* The OROM Support RST Caching of Volumes */
  60 #define MPB_ATTRIB_NVM                  __cpu_to_le32(0x02000000)
  61 /* The OROM supports creating disks greater than 2TB */
  62 #define MPB_ATTRIB_2TB_DISK             __cpu_to_le32(0x04000000)
  63 /* The OROM supports Bad Block Management */
  64 #define MPB_ATTRIB_BBM                  __cpu_to_le32(0x08000000)
  65
  66 /* THe OROM Supports NVM Caching of Volumes */
  67 #define MPB_ATTRIB_NEVER_USE2           __cpu_to_le32(0x10000000)
  68 /* The OROM supports creating volumes greater than 2TB */
  69 #define MPB_ATTRIB_2TB                  __cpu_to_le32(0x20000000)
  70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
  71 #define MPB_ATTRIB_NEVER_USE            __cpu_to_le32(0x40000000)
  72 /* Verify MPB contents against checksum after reading MPB */
  73 #define MPB_ATTRIB_CHECKSUM_VERIFY      __cpu_to_le32(0x80000000)
  74
  75 /* Define all supported attributes that have to be accepted by mdadm
  76  */
  77 #define MPB_ATTRIB_SUPPORTED           (MPB_ATTRIB_CHECKSUM_VERIFY | \
  78                                         MPB_ATTRIB_2TB             | \
  79                                         MPB_ATTRIB_2TB_DISK        | \
  80                                         MPB_ATTRIB_RAID0           | \
  81                                         MPB_ATTRIB_RAID1           | \
  82                                         MPB_ATTRIB_RAID10          | \
  83                                         MPB_ATTRIB_RAID5           | \
  84                                         MPB_ATTRIB_EXP_STRIPE_SIZE)
  85
  86 /* Define attributes that are unused but not harmful */
  87 #define MPB_ATTRIB_IGNORED              (MPB_ATTRIB_NEVER_USE)
  88
  89 #define MPB_SECTOR_CNT 2210
  90 #define IMSM_RESERVED_SECTORS 4096
  91 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
  92 #define SECT_PER_MB_SHIFT 11
  93 #define MAX_SECTOR_SIZE 4096
  94
  95 /* Disk configuration info. */
  96 #define IMSM_MAX_DEVICES 255
  97 struct imsm_disk {
  98         __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
  99         __u32 total_blocks_lo;           /* 0xE8 - 0xEB total blocks lo */
 100         __u32 scsi_id;                   /* 0xEC - 0xEF scsi ID */
 101 #define SPARE_DISK      __cpu_to_le32(0x01)  /* Spare */
 102 #define CONFIGURED_DISK __cpu_to_le32(0x02)  /* Member of some RaidDev */
 103 #define FAILED_DISK     __cpu_to_le32(0x04)  /* Permanent failure */
 104         __u32 status;                    /* 0xF0 - 0xF3 */
 105         __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
 106         __u32 total_blocks_hi;           /* 0xF4 - 0xF5 total blocks hi */
 107 #define IMSM_DISK_FILLERS       3
 108         __u32 filler[IMSM_DISK_FILLERS]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
 109 };
 110
 111 /* map selector for map managment
 112  */
 113 #define MAP_0           0
 114 #define MAP_1           1
 115 #define MAP_X           -1
 116
 117 /* RAID map configuration infos. */
 118 struct imsm_map {
 119         __u32 pba_of_lba0_lo;   /* start address of partition */
 120         __u32 blocks_per_member_lo;/* blocks per member */
 121         __u32 num_data_stripes_lo;      /* number of data stripes */
 122         __u16 blocks_per_strip;
 123         __u8  map_state;        /* Normal, Uninitialized, Degraded, Failed */
 124 #define IMSM_T_STATE_NORMAL 0
 125 #define IMSM_T_STATE_UNINITIALIZED 1
 126 #define IMSM_T_STATE_DEGRADED 2
 127 #define IMSM_T_STATE_FAILED 3
 128         __u8  raid_level;
 129 #define IMSM_T_RAID0 0
 130 #define IMSM_T_RAID1 1
 131 #define IMSM_T_RAID5 5          /* since metadata version 1.2.02 ? */
 132         __u8  num_members;      /* number of member disks */
 133         __u8  num_domains;      /* number of parity domains */
 134         __u8  failed_disk_num;  /* valid only when state is degraded */
 135         __u8  ddf;
 136         __u32 pba_of_lba0_hi;
 137         __u32 blocks_per_member_hi;
 138         __u32 num_data_stripes_hi;
 139         __u32 filler[4];        /* expansion area */
 140 #define IMSM_ORD_REBUILD (1 << 24)
 141         __u32 disk_ord_tbl[1];  /* disk_ord_tbl[num_members],
 142                                  * top byte contains some flags
 143                                  */
 144 } __attribute__ ((packed));
 145
 146 struct imsm_vol {
 147         __u32 curr_migr_unit;
 148         __u32 checkpoint_id;    /* id to access curr_migr_unit */
 149         __u8  migr_state;       /* Normal or Migrating */
 150 #define MIGR_INIT 0
 151 #define MIGR_REBUILD 1
 152 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
 153 #define MIGR_GEN_MIGR 3
 154 #define MIGR_STATE_CHANGE 4
 155 #define MIGR_REPAIR 5
 156         __u8  migr_type;        /* Initializing, Rebuilding, ... */
 157         __u8  dirty;
 158         __u8  fs_state;         /* fast-sync state for CnG (0xff == disabled) */
 159         __u16 verify_errors;    /* number of mismatches */
 160         __u16 bad_blocks;       /* number of bad blocks during verify */
 161         __u32 filler[4];
 162         struct imsm_map map[1];
 163         /* here comes another one if migr_state */
 164 } __attribute__ ((packed));
 165
 166 struct imsm_dev {
 167         __u8  volume[MAX_RAID_SERIAL_LEN];
 168         __u32 size_low;
 169         __u32 size_high;
 170 #define DEV_BOOTABLE            __cpu_to_le32(0x01)
 171 #define DEV_BOOT_DEVICE         __cpu_to_le32(0x02)
 172 #define DEV_READ_COALESCING     __cpu_to_le32(0x04)
 173 #define DEV_WRITE_COALESCING    __cpu_to_le32(0x08)
 174 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
 175 #define DEV_HIDDEN_AT_BOOT      __cpu_to_le32(0x20)
 176 #define DEV_CURRENTLY_HIDDEN    __cpu_to_le32(0x40)
 177 #define DEV_VERIFY_AND_FIX      __cpu_to_le32(0x80)
 178 #define DEV_MAP_STATE_UNINIT    __cpu_to_le32(0x100)
 179 #define DEV_NO_AUTO_RECOVERY    __cpu_to_le32(0x200)
 180 #define DEV_CLONE_N_GO          __cpu_to_le32(0x400)
 181 #define DEV_CLONE_MAN_SYNC      __cpu_to_le32(0x800)
 182 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
 183         __u32 status;   /* Persistent RaidDev status */
 184         __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
 185         __u8  migr_priority;
 186         __u8  num_sub_vols;
 187         __u8  tid;
 188         __u8  cng_master_disk;
 189         __u16 cache_policy;
 190         __u8  cng_state;
 191         __u8  cng_sub_state;
 192 #define IMSM_DEV_FILLERS 10
 193         __u32 filler[IMSM_DEV_FILLERS];
 194         struct imsm_vol vol;
 195 } __attribute__ ((packed));
 196
 197 struct imsm_super {
 198         __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
 199         __u32 check_sum;                /* 0x20 - 0x23 MPB Checksum */
 200         __u32 mpb_size;                 /* 0x24 - 0x27 Size of MPB */
 201         __u32 family_num;               /* 0x28 - 0x2B Checksum from first time this config was written */
 202         __u32 generation_num;           /* 0x2C - 0x2F Incremented each time this array's MPB is written */
 203         __u32 error_log_size;           /* 0x30 - 0x33 in bytes */
 204         __u32 attributes;               /* 0x34 - 0x37 */
 205         __u8 num_disks;                 /* 0x38 Number of configured disks */
 206         __u8 num_raid_devs;             /* 0x39 Number of configured volumes */
 207         __u8 error_log_pos;             /* 0x3A  */
 208         __u8 fill[1];                   /* 0x3B */
 209         __u32 cache_size;               /* 0x3c - 0x40 in mb */
 210         __u32 orig_family_num;          /* 0x40 - 0x43 original family num */
 211         __u32 pwr_cycle_count;          /* 0x44 - 0x47 simulated power cycle count for array */
 212         __u32 bbm_log_size;             /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
 213 #define IMSM_FILLERS 35
 214         __u32 filler[IMSM_FILLERS];     /* 0x4C - 0xD7 RAID_MPB_FILLERS */
 215         struct imsm_disk disk[1];       /* 0xD8 diskTbl[numDisks] */
 216         /* here comes imsm_dev[num_raid_devs] */
 217         /* here comes BBM logs */
 218 } __attribute__ ((packed));
 219
 220 #define BBM_LOG_MAX_ENTRIES 254
 221
 222 struct bbm_log_entry {
 223         __u64 defective_block_start;
 224 #define UNREADABLE 0xFFFFFFFF
 225         __u32 spare_block_offset;
 226         __u16 remapped_marked_count;
 227         __u16 disk_ordinal;
 228 } __attribute__ ((__packed__));
 229
 230 struct bbm_log {
 231         __u32 signature; /* 0xABADB10C */
 232         __u32 entry_count;
 233         __u32 reserved_spare_block_count; /* 0 */
 234         __u32 reserved; /* 0xFFFF */
 235         __u64 first_spare_lba;
 236         struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
 237 } __attribute__ ((__packed__));
 238
 239 #ifndef MDASSEMBLE
 240 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
 241 #endif
 242
 243 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
 244
 245 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
 246
 247 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
 248 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
 249                                * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
 250                                */
 251
 252 #define UNIT_SRC_NORMAL     0   /* Source data for curr_migr_unit must
 253                                  *  be recovered using srcMap */
 254 #define UNIT_SRC_IN_CP_AREA 1   /* Source data for curr_migr_unit has
 255                                  *  already been migrated and must
 256                                  *  be recovered from checkpoint area */
 257 struct migr_record {
 258         __u32 rec_status;           /* Status used to determine how to restart
 259                                      * migration in case it aborts
 260                                      * in some fashion */
 261         __u32 curr_migr_unit;       /* 0..numMigrUnits-1 */
 262         __u32 family_num;           /* Family number of MPB
 263                                      * containing the RaidDev
 264                                      * that is migrating */
 265         __u32 ascending_migr;       /* True if migrating in increasing
 266                                      * order of lbas */
 267         __u32 blocks_per_unit;      /* Num disk blocks per unit of operation */
 268         __u32 dest_depth_per_unit;  /* Num member blocks each destMap
 269                                      * member disk
 270                                      * advances per unit-of-operation */
 271         __u32 ckpt_area_pba;        /* Pba of first block of ckpt copy area */
 272         __u32 dest_1st_member_lba;  /* First member lba on first
 273                                      * stripe of destination */
 274         __u32 num_migr_units;       /* Total num migration units-of-op */
 275         __u32 post_migr_vol_cap;    /* Size of volume after
 276                                      * migration completes */
 277         __u32 post_migr_vol_cap_hi; /* Expansion space for LBA64 */
 278         __u32 ckpt_read_disk_num;   /* Which member disk in destSubMap[0] the
 279                                      * migration ckpt record was read from
 280                                      * (for recovered migrations) */
 281 } __attribute__ ((__packed__));
 282
 283 struct md_list {
 284         /* usage marker:
 285          *  1: load metadata
 286          *  2: metadata does not match
 287          *  4: already checked
 288          */
 289         int   used;
 290         char  *devname;
 291         int   found;
 292         int   container;
 293         dev_t st_rdev;
 294         struct md_list *next;
 295 };
 296
 297 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
 298
 299 static __u8 migr_type(struct imsm_dev *dev)
 300 {
 301         if (dev->vol.migr_type == MIGR_VERIFY &&
 302             dev->status & DEV_VERIFY_AND_FIX)
 303                 return MIGR_REPAIR;
 304         else
 305                 return dev->vol.migr_type;
 306 }
 307
 308 static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
 309 {
 310         /* for compatibility with older oroms convert MIGR_REPAIR, into
 311          * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
 312          */
 313         if (migr_type == MIGR_REPAIR) {
 314                 dev->vol.migr_type = MIGR_VERIFY;
 315                 dev->status |= DEV_VERIFY_AND_FIX;
 316         } else {
 317                 dev->vol.migr_type = migr_type;
 318                 dev->status &= ~DEV_VERIFY_AND_FIX;
 319         }
 320 }
 321
 322 static unsigned int sector_count(__u32 bytes, unsigned int sector_size)
 323 {
 324         return ROUND_UP(bytes, sector_size) / sector_size;
 325 }
 326
 327 static unsigned int mpb_sectors(struct imsm_super *mpb,
 328                                         unsigned int sector_size)
 329 {
 330         return sector_count(__le32_to_cpu(mpb->mpb_size), sector_size);
 331 }
 332
 333 struct intel_dev {
 334         struct imsm_dev *dev;
 335         struct intel_dev *next;
 336         unsigned index;
 337 };
 338
 339 struct intel_hba {
 340         enum sys_dev_type type;
 341         char *path;
 342         char *pci_id;
 343         struct intel_hba *next;
 344 };
 345
 346 enum action {
 347         DISK_REMOVE = 1,
 348         DISK_ADD
 349 };
 350 /* internal representation of IMSM metadata */
 351 struct intel_super {
 352         union {
 353                 void *buf; /* O_DIRECT buffer for reading/writing metadata */
 354                 struct imsm_super *anchor; /* immovable parameters */
 355         };
 356         union {
 357                 void *migr_rec_buf; /* buffer for I/O operations */
 358                 struct migr_record *migr_rec; /* migration record */
 359         };
 360         int clean_migration_record_by_mdmon; /* when reshape is switched to next
 361                 array, it indicates that mdmon is allowed to clean migration
 362                 record */
 363         size_t len; /* size of the 'buf' allocation */
 364         void *next_buf; /* for realloc'ing buf from the manager */
 365         size_t next_len;
 366         int updates_pending; /* count of pending updates for mdmon */
 367         int current_vol; /* index of raid device undergoing creation */
 368         unsigned long long create_offset; /* common start for 'current_vol' */
 369         __u32 random; /* random data for seeding new family numbers */
 370         struct intel_dev *devlist;
 371         unsigned int sector_size; /* sector size of used member drives */
 372         struct dl {
 373                 struct dl *next;
 374                 int index;
 375                 __u8 serial[MAX_RAID_SERIAL_LEN];
 376                 int major, minor;
 377                 char *devname;
 378                 struct imsm_disk disk;
 379                 int fd;
 380                 int extent_cnt;
 381                 struct extent *e; /* for determining freespace @ create */
 382                 int raiddisk; /* slot to fill in autolayout */
 383                 enum action action;
 384         } *disks, *current_disk;
 385         struct dl *disk_mgmt_list; /* list of disks to add/remove while mdmon
 386                                       active */
 387         struct dl *missing; /* disks removed while we weren't looking */
 388         struct bbm_log *bbm_log;
 389         struct intel_hba *hba; /* device path of the raid controller for this metadata */
 390         const struct imsm_orom *orom; /* platform firmware support */
 391         struct intel_super *next; /* (temp) list for disambiguating family_num */
 392 };
 393
 394 struct intel_disk {
 395         struct imsm_disk disk;
 396         #define IMSM_UNKNOWN_OWNER (-1)
 397         int owner;
 398         struct intel_disk *next;
 399 };
 400
 401 struct extent {
 402         unsigned long long start, size;
 403 };
 404
 405 /* definitions of reshape process types */
 406 enum imsm_reshape_type {
 407         CH_TAKEOVER,
 408         CH_MIGRATION,
 409         CH_ARRAY_SIZE,
 410 };
 411
 412 /* definition of messages passed to imsm_process_update */
 413 enum imsm_update_type {
 414         update_activate_spare,
 415         update_create_array,
 416         update_kill_array,
 417         update_rename_array,
 418         update_add_remove_disk,
 419         update_reshape_container_disks,
 420         update_reshape_migration,
 421         update_takeover,
 422         update_general_migration_checkpoint,
 423         update_size_change,
 424 };
 425
 426 struct imsm_update_activate_spare {
 427         enum imsm_update_type type;
 428         struct dl *dl;
 429         int slot;
 430         int array;
 431         struct imsm_update_activate_spare *next;
 432 };
 433
 434 struct geo_params {
 435         char devnm[32];
 436         char *dev_name;
 437         unsigned long long size;
 438         int level;
 439         int layout;
 440         int chunksize;
 441         int raid_disks;
 442 };
 443
 444 enum takeover_direction {
 445         R10_TO_R0,
 446         R0_TO_R10
 447 };
 448 struct imsm_update_takeover {
 449         enum imsm_update_type type;
 450         int subarray;
 451         enum takeover_direction direction;
 452 };
 453
 454 struct imsm_update_reshape {
 455         enum imsm_update_type type;
 456         int old_raid_disks;
 457         int new_raid_disks;
 458
 459         int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */
 460 };
 461
 462 struct imsm_update_reshape_migration {
 463         enum imsm_update_type type;
 464         int old_raid_disks;
 465         int new_raid_disks;
 466         /* fields for array migration changes
 467          */
 468         int subdev;
 469         int new_level;
 470         int new_layout;
 471         int new_chunksize;
 472
 473         int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */
 474 };
 475
 476 struct imsm_update_size_change {
 477         enum imsm_update_type type;
 478         int subdev;
 479         long long new_size;
 480 };
 481
 482 struct imsm_update_general_migration_checkpoint {
 483         enum imsm_update_type type;
 484         __u32 curr_migr_unit;
 485 };
 486
 487 struct disk_info {
 488         __u8 serial[MAX_RAID_SERIAL_LEN];
 489 };
 490
 491 struct imsm_update_create_array {
 492         enum imsm_update_type type;
 493         int dev_idx;
 494         struct imsm_dev dev;
 495 };
 496
 497 struct imsm_update_kill_array {
 498         enum imsm_update_type type;
 499         int dev_idx;
 500 };
 501
 502 struct imsm_update_rename_array {
 503         enum imsm_update_type type;
 504         __u8 name[MAX_RAID_SERIAL_LEN];
 505         int dev_idx;
 506 };
 507
 508 struct imsm_update_add_remove_disk {
 509         enum imsm_update_type type;
 510 };
 511
 512 static const char *_sys_dev_type[] = {
 513         [SYS_DEV_UNKNOWN] = "Unknown",
 514         [SYS_DEV_SAS] = "SAS",
 515         [SYS_DEV_SATA] = "SATA",
 516         [SYS_DEV_NVME] = "NVMe",
 517         [SYS_DEV_VMD] = "VMD"
 518 };
 519
 520 const char *get_sys_dev_type(enum sys_dev_type type)
 521 {
 522         if (type >= SYS_DEV_MAX)
 523                 type = SYS_DEV_UNKNOWN;
 524
 525         return _sys_dev_type[type];
 526 }
 527
 528 static struct intel_hba * alloc_intel_hba(struct sys_dev *device)
 529 {
 530         struct intel_hba *result = xmalloc(sizeof(*result));
 531
 532         result->type = device->type;
 533         result->path = xstrdup(device->path);
 534         result->next = NULL;
 535         if (result->path && (result->pci_id = strrchr(result->path, '/')) != NULL)
 536                 result->pci_id++;
 537
 538         return result;
 539 }
 540
 541 static struct intel_hba * find_intel_hba(struct intel_hba *hba, struct sys_dev *device)
 542 {
 543         struct intel_hba *result;
 544
 545         for (result = hba; result; result = result->next) {
 546                 if (result->type == device->type && strcmp(result->path, device->path) == 0)
 547                         break;
 548         }
 549         return result;
 550 }
 551
 552 static int attach_hba_to_super(struct intel_super *super, struct sys_dev *device)
 553 {
 554         struct intel_hba *hba;
 555
 556         /* check if disk attached to Intel HBA */
 557         hba = find_intel_hba(super->hba, device);
 558         if (hba != NULL)
 559                 return 1;
 560         /* Check if HBA is already attached to super */
 561         if (super->hba == NULL) {
 562                 super->hba = alloc_intel_hba(device);
 563                 return 1;
 564         }
 565
 566         hba = super->hba;
 567         /* Intel metadata allows for all disks attached to the same type HBA.
 568          * Do not support HBA types mixing
 569          */
 570         if (device->type != hba->type)
 571                 return 2;
 572
 573         /* Multiple same type HBAs can be used if they share the same OROM */
 574         const struct imsm_orom *device_orom = get_orom_by_device_id(device->dev_id);
 575
 576         if (device_orom != super->orom)
 577                 return 2;
 578
 579         while (hba->next)
 580                 hba = hba->next;
 581
 582         hba->next = alloc_intel_hba(device);
 583         return 1;
 584 }
 585
 586 static struct sys_dev* find_disk_attached_hba(int fd, const char *devname)
 587 {
 588         struct sys_dev *list, *elem;
 589         char *disk_path;
 590
 591         if ((list = find_intel_devices()) == NULL)
 592                 return 0;
 593
 594         if (fd < 0)
 595                 disk_path  = (char *) devname;
 596         else
 597                 disk_path = diskfd_to_devpath(fd);
 598
 599         if (!disk_path)
 600                 return 0;
 601
 602         for (elem = list; elem; elem = elem->next)
 603                 if (path_attached_to_hba(disk_path, elem->path))
 604                         return elem;
 605
 606         if (disk_path != devname)
 607                 free(disk_path);
 608
 609         return NULL;
 610 }
 611
 612 static int find_intel_hba_capability(int fd, struct intel_super *super,
 613                                      char *devname);
 614
 615 static struct supertype *match_metadata_desc_imsm(char *arg)
 616 {
 617         struct supertype *st;
 618
 619         if (strcmp(arg, "imsm") != 0 &&
 620             strcmp(arg, "default") != 0
 621                 )
 622                 return NULL;
 623
 624         st = xcalloc(1, sizeof(*st));
 625         st->ss = &super_imsm;
 626         st->max_devs = IMSM_MAX_DEVICES;
 627         st->minor_version = 0;
 628         st->sb = NULL;
 629         return st;
 630 }
 631
 632 #ifndef MDASSEMBLE
 633 static __u8 *get_imsm_version(struct imsm_super *mpb)
 634 {
 635         return &mpb->sig[MPB_SIG_LEN];
 636 }
 637 #endif
 638
 639 /* retrieve a disk directly from the anchor when the anchor is known to be
 640  * up-to-date, currently only at load time
 641  */
 642 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
 643 {
 644         if (index >= mpb->num_disks)
 645                 return NULL;
 646         return &mpb->disk[index];
 647 }
 648
 649 /* retrieve the disk description based on a index of the disk
 650  * in the sub-array
 651  */
 652 static struct dl *get_imsm_dl_disk(struct intel_super *super, __u8 index)
 653 {
 654         struct dl *d;
 655
 656         for (d = super->disks; d; d = d->next)
 657                 if (d->index == index)
 658                         return d;
 659
 660         return NULL;
 661 }
 662 /* retrieve a disk from the parsed metadata */
 663 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
 664 {
 665         struct dl *dl;
 666
 667         dl = get_imsm_dl_disk(super, index);
 668         if (dl)
 669                 return &dl->disk;
 670
 671         return NULL;
 672 }
 673
 674 /* generate a checksum directly from the anchor when the anchor is known to be
 675  * up-to-date, currently only at load or write_super after coalescing
 676  */
 677 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
 678 {
 679         __u32 end = mpb->mpb_size / sizeof(end);
 680         __u32 *p = (__u32 *) mpb;
 681         __u32 sum = 0;
 682
 683         while (end--) {
 684                 sum += __le32_to_cpu(*p);
 685                 p++;
 686         }
 687
 688         return sum - __le32_to_cpu(mpb->check_sum);
 689 }
 690
 691 static size_t sizeof_imsm_map(struct imsm_map *map)
 692 {
 693         return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
 694 }
 695
 696 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
 697 {
 698         /* A device can have 2 maps if it is in the middle of a migration.
 699          * If second_map is:
 700          *    MAP_0 - we return the first map
 701          *    MAP_1 - we return the second map if it exists, else NULL
 702          *    MAP_X - we return the second map if it exists, else the first
 703          */
 704         struct imsm_map *map = &dev->vol.map[0];
 705         struct imsm_map *map2 = NULL;
 706
 707         if (dev->vol.migr_state)
 708                 map2 = (void *)map + sizeof_imsm_map(map);
 709
 710         switch (second_map) {
 711         case MAP_0:
 712                 break;
 713         case MAP_1:
 714                 map = map2;
 715                 break;
 716         case MAP_X:
 717                 if (map2)
 718                         map = map2;
 719                 break;
 720         default:
 721                 map = NULL;
 722         }
 723         return map;
 724
 725 }
 726
 727 /* return the size of the device.
 728  * migr_state increases the returned size if map[0] were to be duplicated
 729  */
 730 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
 731 {
 732         size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
 733                       sizeof_imsm_map(get_imsm_map(dev, MAP_0));
 734
 735         /* migrating means an additional map */
 736         if (dev->vol.migr_state)
 737                 size += sizeof_imsm_map(get_imsm_map(dev, MAP_1));
 738         else if (migr_state)
 739                 size += sizeof_imsm_map(get_imsm_map(dev, MAP_0));
 740
 741         return size;
 742 }
 743
 744 #ifndef MDASSEMBLE
 745 /* retrieve disk serial number list from a metadata update */
 746 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
 747 {
 748         void *u = update;
 749         struct disk_info *inf;
 750
 751         inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
 752               sizeof_imsm_dev(&update->dev, 0);
 753
 754         return inf;
 755 }
 756 #endif
 757
 758 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
 759 {
 760         int offset;
 761         int i;
 762         void *_mpb = mpb;
 763
 764         if (index >= mpb->num_raid_devs)
 765                 return NULL;
 766
 767         /* devices start after all disks */
 768         offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
 769
 770         for (i = 0; i <= index; i++)
 771                 if (i == index)
 772                         return _mpb + offset;
 773                 else
 774                         offset += sizeof_imsm_dev(_mpb + offset, 0);
 775
 776         return NULL;
 777 }
 778
 779 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
 780 {
 781         struct intel_dev *dv;
 782
 783         if (index >= super->anchor->num_raid_devs)
 784                 return NULL;
 785         for (dv = super->devlist; dv; dv = dv->next)
 786                 if (dv->index == index)
 787                         return dv->dev;
 788         return NULL;
 789 }
 790
 791 /*
 792  * for second_map:
 793  *  == MAP_0 get first map
 794  *  == MAP_1 get second map
 795  *  == MAP_X than get map according to the current migr_state
 796  */
 797 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev,
 798                                   int slot,
 799                                   int second_map)
 800 {
 801         struct imsm_map *map;
 802
 803         map = get_imsm_map(dev, second_map);
 804
 805         /* top byte identifies disk under rebuild */
 806         return __le32_to_cpu(map->disk_ord_tbl[slot]);
 807 }
 808
 809 #define ord_to_idx(ord) (((ord) << 8) >> 8)
 810 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot, int second_map)
 811 {
 812         __u32 ord = get_imsm_ord_tbl_ent(dev, slot, second_map);
 813
 814         return ord_to_idx(ord);
 815 }
 816
 817 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
 818 {
 819         map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
 820 }
 821
 822 static int get_imsm_disk_slot(struct imsm_map *map, unsigned idx)
 823 {
 824         int slot;
 825         __u32 ord;
 826
 827         for (slot = 0; slot < map->num_members; slot++) {
 828                 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
 829                 if (ord_to_idx(ord) == idx)
 830                         return slot;
 831         }
 832
 833         return -1;
 834 }
 835
 836 static int get_imsm_raid_level(struct imsm_map *map)
 837 {
 838         if (map->raid_level == 1) {
 839                 if (map->num_members == 2)
 840                         return 1;
 841                 else
 842                         return 10;
 843         }
 844
 845         return map->raid_level;
 846 }
 847
 848 static int cmp_extent(const void *av, const void *bv)
 849 {
 850         const struct extent *a = av;
 851         const struct extent *b = bv;
 852         if (a->start < b->start)
 853                 return -1;
 854         if (a->start > b->start)
 855                 return 1;
 856         return 0;
 857 }
 858
 859 static int count_memberships(struct dl *dl, struct intel_super *super)
 860 {
 861         int memberships = 0;
 862         int i;
 863
 864         for (i = 0; i < super->anchor->num_raid_devs; i++) {
 865                 struct imsm_dev *dev = get_imsm_dev(super, i);
 866                 struct imsm_map *map = get_imsm_map(dev, MAP_0);
 867
 868                 if (get_imsm_disk_slot(map, dl->index) >= 0)
 869                         memberships++;
 870         }
 871
 872         return memberships;
 873 }
 874
 875 static __u32 imsm_min_reserved_sectors(struct intel_super *super);
 876
 877 static int split_ull(unsigned long long n, __u32 *lo, __u32 *hi)
 878 {
 879         if (lo == 0 || hi == 0)
 880                 return 1;
 881         *lo = __le32_to_cpu((unsigned)n);
 882         *hi = __le32_to_cpu((unsigned)(n >> 32));
 883         return 0;
 884 }
 885
 886 static unsigned long long join_u32(__u32 lo, __u32 hi)
 887 {
 888         return (unsigned long long)__le32_to_cpu(lo) |
 889                (((unsigned long long)__le32_to_cpu(hi)) << 32);
 890 }
 891
 892 static unsigned long long total_blocks(struct imsm_disk *disk)
 893 {
 894         if (disk == NULL)
 895                 return 0;
 896         return join_u32(disk->total_blocks_lo, disk->total_blocks_hi);
 897 }
 898
 899 static unsigned long long pba_of_lba0(struct imsm_map *map)
 900 {
 901         if (map == NULL)
 902                 return 0;
 903         return join_u32(map->pba_of_lba0_lo, map->pba_of_lba0_hi);
 904 }
 905
 906 static unsigned long long blocks_per_member(struct imsm_map *map)
 907 {
 908         if (map == NULL)
 909                 return 0;
 910         return join_u32(map->blocks_per_member_lo, map->blocks_per_member_hi);
 911 }
 912
 913 #ifndef MDASSEMBLE
 914 static unsigned long long num_data_stripes(struct imsm_map *map)
 915 {
 916         if (map == NULL)
 917                 return 0;
 918         return join_u32(map->num_data_stripes_lo, map->num_data_stripes_hi);
 919 }
 920 #endif
 921
 922 static void set_total_blocks(struct imsm_disk *disk, unsigned long long n)
 923 {
 924         split_ull(n, &disk->total_blocks_lo, &disk->total_blocks_hi);
 925 }
 926
 927 static void set_pba_of_lba0(struct imsm_map *map, unsigned long long n)
 928 {
 929         split_ull(n, &map->pba_of_lba0_lo, &map->pba_of_lba0_hi);
 930 }
 931
 932 static void set_blocks_per_member(struct imsm_map *map, unsigned long long n)
 933 {
 934         split_ull(n, &map->blocks_per_member_lo, &map->blocks_per_member_hi);
 935 }
 936
 937 static void set_num_data_stripes(struct imsm_map *map, unsigned long long n)
 938 {
 939         split_ull(n, &map->num_data_stripes_lo, &map->num_data_stripes_hi);
 940 }
 941
 942 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
 943 {
 944         /* find a list of used extents on the given physical device */
 945         struct extent *rv, *e;
 946         int i;
 947         int memberships = count_memberships(dl, super);
 948         __u32 reservation;
 949
 950         /* trim the reserved area for spares, so they can join any array
 951          * regardless of whether the OROM has assigned sectors from the
 952          * IMSM_RESERVED_SECTORS region
 953          */
 954         if (dl->index == -1)
 955                 reservation = imsm_min_reserved_sectors(super);
 956         else
 957                 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
 958
 959         rv = xcalloc(sizeof(struct extent), (memberships + 1));
 960         e = rv;
 961
 962         for (i = 0; i < super->anchor->num_raid_devs; i++) {
 963                 struct imsm_dev *dev = get_imsm_dev(super, i);
 964                 struct imsm_map *map = get_imsm_map(dev, MAP_0);
 965
 966                 if (get_imsm_disk_slot(map, dl->index) >= 0) {
 967                         e->start = pba_of_lba0(map);
 968                         e->size = blocks_per_member(map);
 969                         e++;
 970                 }
 971         }
 972         qsort(rv, memberships, sizeof(*rv), cmp_extent);
 973
 974         /* determine the start of the metadata
 975          * when no raid devices are defined use the default
 976          * ...otherwise allow the metadata to truncate the value
 977          * as is the case with older versions of imsm
 978          */
 979         if (memberships) {
 980                 struct extent *last = &rv[memberships - 1];
 981                 unsigned long long remainder;
 982
 983                 remainder = total_blocks(&dl->disk) - (last->start + last->size);
 984                 /* round down to 1k block to satisfy precision of the kernel
 985                  * 'size' interface
 986                  */
 987                 remainder &= ~1UL;
 988                 /* make sure remainder is still sane */
 989                 if (remainder < (unsigned)ROUND_UP(super->len, 512) >> 9)
 990                         remainder = ROUND_UP(super->len, 512) >> 9;
 991                 if (reservation > remainder)
 992                         reservation = remainder;
 993         }
 994         e->start = total_blocks(&dl->disk) - reservation;
 995         e->size = 0;
 996         return rv;
 997 }
 998
 999 /* try to determine how much space is reserved for metadata from
1000  * the last get_extents() entry, otherwise fallback to the
1001  * default
1002  */
1003 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
1004 {
1005         struct extent *e;
1006         int i;
1007         __u32 rv;
1008
1009         /* for spares just return a minimal reservation which will grow
1010          * once the spare is picked up by an array
1011          */
1012         if (dl->index == -1)
1013                 return MPB_SECTOR_CNT;
1014
1015         e = get_extents(super, dl);
1016         if (!e)
1017                 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
1018
1019         /* scroll to last entry */
1020         for (i = 0; e[i].size; i++)
1021                 continue;
1022
1023         rv = total_blocks(&dl->disk) - e[i].start;
1024
1025         free(e);
1026
1027         return rv;
1028 }
1029
1030 static int is_spare(struct imsm_disk *disk)
1031 {
1032         return (disk->status & SPARE_DISK) == SPARE_DISK;
1033 }
1034
1035 static int is_configured(struct imsm_disk *disk)
1036 {
1037         return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
1038 }
1039
1040 static int is_failed(struct imsm_disk *disk)
1041 {
1042         return (disk->status & FAILED_DISK) == FAILED_DISK;
1043 }
1044
1045 /* try to determine how much space is reserved for metadata from
1046  * the last get_extents() entry on the smallest active disk,
1047  * otherwise fallback to the default
1048  */
1049 static __u32 imsm_min_reserved_sectors(struct intel_super *super)
1050 {
1051         struct extent *e;
1052         int i;
1053         unsigned long long min_active;
1054         __u32 remainder;
1055         __u32 rv = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
1056         struct dl *dl, *dl_min = NULL;
1057
1058         if (!super)
1059                 return rv;
1060
1061         min_active = 0;
1062         for (dl = super->disks; dl; dl = dl->next) {
1063                 if (dl->index < 0)
1064                         continue;
1065                 unsigned long long blocks = total_blocks(&dl->disk);
1066                 if (blocks < min_active || min_active == 0) {
1067                         dl_min = dl;
1068                         min_active = blocks;
1069                 }
1070         }
1071         if (!dl_min)
1072                 return rv;
1073
1074         /* find last lba used by subarrays on the smallest active disk */
1075         e = get_extents(super, dl_min);
1076         if (!e)
1077                 return rv;
1078         for (i = 0; e[i].size; i++)
1079                 continue;
1080
1081         remainder = min_active - e[i].start;
1082         free(e);
1083
1084         /* to give priority to recovery we should not require full
1085            IMSM_RESERVED_SECTORS from the spare */
1086         rv = MPB_SECTOR_CNT + NUM_BLOCKS_DIRTY_STRIPE_REGION;
1087
1088         /* if real reservation is smaller use that value */
1089         return  (remainder < rv) ? remainder : rv;
1090 }
1091
1092 /* Return minimum size of a spare that can be used in this array*/
1093 static unsigned long long min_acceptable_spare_size_imsm(struct supertype *st)
1094 {
1095         struct intel_super *super = st->sb;
1096         struct dl *dl;
1097         struct extent *e;
1098         int i;
1099         unsigned long long rv = 0;
1100
1101         if (!super)
1102                 return rv;
1103         /* find first active disk in array */
1104         dl = super->disks;
1105         while (dl && (is_failed(&dl->disk) || dl->index == -1))
1106                 dl = dl->next;
1107         if (!dl)
1108                 return rv;
1109         /* find last lba used by subarrays */
1110         e = get_extents(super, dl);
1111         if (!e)
1112                 return rv;
1113         for (i = 0; e[i].size; i++)
1114                 continue;
1115         if (i > 0)
1116                 rv = e[i-1].start + e[i-1].size;
1117         free(e);
1118
1119         /* add the amount of space needed for metadata */
1120         rv = rv + imsm_min_reserved_sectors(super);
1121
1122         return rv * 512;
1123 }
1124
1125 static int is_gen_migration(struct imsm_dev *dev);
1126
1127 #define IMSM_4K_DIV 8
1128
1129 #ifndef MDASSEMBLE
1130 static __u64 blocks_per_migr_unit(struct intel_super *super,
1131                                   struct imsm_dev *dev);
1132
1133 static void print_imsm_dev(struct intel_super *super,
1134                            struct imsm_dev *dev,
1135                            char *uuid,
1136                            int disk_idx)
1137 {
1138         __u64 sz;
1139         int slot, i;
1140         struct imsm_map *map = get_imsm_map(dev, MAP_0);
1141         struct imsm_map *map2 = get_imsm_map(dev, MAP_1);
1142         __u32 ord;
1143
1144         printf("\n");
1145         printf("[%.16s]:\n", dev->volume);
1146         printf("           UUID : %s\n", uuid);
1147         printf("     RAID Level : %d", get_imsm_raid_level(map));
1148         if (map2)
1149                 printf(" <-- %d", get_imsm_raid_level(map2));
1150         printf("\n");
1151         printf("        Members : %d", map->num_members);
1152         if (map2)
1153                 printf(" <-- %d", map2->num_members);
1154         printf("\n");
1155         printf("          Slots : [");
1156         for (i = 0; i < map->num_members; i++) {
1157                 ord = get_imsm_ord_tbl_ent(dev, i, MAP_0);
1158                 printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
1159         }
1160         printf("]");
1161         if (map2) {
1162                 printf(" <-- [");
1163                 for (i = 0; i < map2->num_members; i++) {
1164                         ord = get_imsm_ord_tbl_ent(dev, i, MAP_1);
1165                         printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
1166                 }
1167                 printf("]");
1168         }
1169         printf("\n");
1170         printf("    Failed disk : ");
1171         if (map->failed_disk_num == 0xff)
1172                 printf("none");
1173         else
1174                 printf("%i", map->failed_disk_num);
1175         printf("\n");
1176         slot = get_imsm_disk_slot(map, disk_idx);
1177         if (slot >= 0) {
1178                 ord = get_imsm_ord_tbl_ent(dev, slot, MAP_X);
1179                 printf("      This Slot : %d%s\n", slot,
1180                        ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
1181         } else
1182                 printf("      This Slot : ?\n");
1183         sz = __le32_to_cpu(dev->size_high);
1184         sz <<= 32;
1185         sz += __le32_to_cpu(dev->size_low);
1186         printf("     Array Size : %llu%s\n", (unsigned long long)sz,
1187                human_size(sz * 512));
1188         sz = blocks_per_member(map);
1189         printf("   Per Dev Size : %llu%s\n", (unsigned long long)sz,
1190                human_size(sz * 512));
1191         printf("  Sector Offset : %llu\n",
1192                 pba_of_lba0(map));
1193         printf("    Num Stripes : %llu\n",
1194                 num_data_stripes(map));
1195         printf("     Chunk Size : %u KiB",
1196                 __le16_to_cpu(map->blocks_per_strip) / 2);
1197         if (map2)
1198                 printf(" <-- %u KiB",
1199                         __le16_to_cpu(map2->blocks_per_strip) / 2);
1200         printf("\n");
1201         printf("       Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
1202         printf("  Migrate State : ");
1203         if (dev->vol.migr_state) {
1204                 if (migr_type(dev) == MIGR_INIT)
1205                         printf("initialize\n");
1206                 else if (migr_type(dev) == MIGR_REBUILD)
1207                         printf("rebuild\n");
1208                 else if (migr_type(dev) == MIGR_VERIFY)
1209                         printf("check\n");
1210                 else if (migr_type(dev) == MIGR_GEN_MIGR)
1211                         printf("general migration\n");
1212                 else if (migr_type(dev) == MIGR_STATE_CHANGE)
1213                         printf("state change\n");
1214                 else if (migr_type(dev) == MIGR_REPAIR)
1215                         printf("repair\n");
1216                 else
1217                         printf("<unknown:%d>\n", migr_type(dev));
1218         } else
1219                 printf("idle\n");
1220         printf("      Map State : %s", map_state_str[map->map_state]);
1221         if (dev->vol.migr_state) {
1222                 struct imsm_map *map = get_imsm_map(dev, MAP_1);
1223
1224                 printf(" <-- %s", map_state_str[map->map_state]);
1225                 printf("\n     Checkpoint : %u ",
1226                            __le32_to_cpu(dev->vol.curr_migr_unit));
1227                 if (is_gen_migration(dev) && (slot > 1 || slot < 0))
1228                         printf("(N/A)");
1229                 else
1230                         printf("(%llu)", (unsigned long long)
1231                                    blocks_per_migr_unit(super, dev));
1232         }
1233         printf("\n");
1234         printf("    Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
1235 }
1236
1237 static void print_imsm_disk(struct imsm_disk *disk, int index, __u32 reserved)
1238 {
1239         char str[MAX_RAID_SERIAL_LEN + 1];
1240         __u64 sz;
1241
1242         if (index < -1 || !disk)
1243                 return;
1244
1245         printf("\n");
1246         snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
1247         if (index >= 0)
1248                 printf("  Disk%02d Serial : %s\n", index, str);
1249         else
1250                 printf("    Disk Serial : %s\n", str);
1251         printf("          State :%s%s%s\n", is_spare(disk) ? " spare" : "",
1252                                             is_configured(disk) ? " active" : "",
1253                                             is_failed(disk) ? " failed" : "");
1254         printf("             Id : %08x\n", __le32_to_cpu(disk->scsi_id));
1255         sz = total_blocks(disk) - reserved;
1256         printf("    Usable Size : %llu%s\n", (unsigned long long)sz,
1257                human_size(sz * 512));
1258 }
1259
1260 void convert_to_4k_imsm_migr_rec(struct intel_super *super)
1261 {
1262         struct migr_record *migr_rec = super->migr_rec;
1263
1264         migr_rec->blocks_per_unit /= IMSM_4K_DIV;
1265         migr_rec->ckpt_area_pba /= IMSM_4K_DIV;
1266         migr_rec->dest_1st_member_lba /= IMSM_4K_DIV;
1267         migr_rec->dest_depth_per_unit /= IMSM_4K_DIV;
1268         split_ull((join_u32(migr_rec->post_migr_vol_cap,
1269                  migr_rec->post_migr_vol_cap_hi) / IMSM_4K_DIV),
1270                  &migr_rec->post_migr_vol_cap, &migr_rec->post_migr_vol_cap_hi);
1271 }
1272
1273 void convert_to_4k_imsm_disk(struct imsm_disk *disk)
1274 {
1275         set_total_blocks(disk, (total_blocks(disk)/IMSM_4K_DIV));
1276 }
1277
1278 void convert_to_4k(struct intel_super *super)
1279 {
1280         struct imsm_super *mpb = super->anchor;
1281         struct imsm_disk *disk;
1282         int i;
1283
1284         for (i = 0; i < mpb->num_disks ; i++) {
1285                 disk = __get_imsm_disk(mpb, i);
1286                 /* disk */
1287                 convert_to_4k_imsm_disk(disk);
1288         }
1289         for (i = 0; i < mpb->num_raid_devs; i++) {
1290                 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
1291                 struct imsm_map *map = get_imsm_map(dev, MAP_0);
1292                 /* dev */
1293                 split_ull((join_u32(dev->size_low, dev->size_high)/IMSM_4K_DIV),
1294                                  &dev->size_low, &dev->size_high);
1295                 dev->vol.curr_migr_unit /= IMSM_4K_DIV;
1296
1297                 /* map0 */
1298                 set_blocks_per_member(map, blocks_per_member(map)/IMSM_4K_DIV);
1299                 map->blocks_per_strip /= IMSM_4K_DIV;
1300                 set_pba_of_lba0(map, pba_of_lba0(map)/IMSM_4K_DIV);
1301
1302                 if (dev->vol.migr_state) {
1303                         /* map1 */
1304                         map = get_imsm_map(dev, MAP_1);
1305                         set_blocks_per_member(map,
1306                             blocks_per_member(map)/IMSM_4K_DIV);
1307                         map->blocks_per_strip /= IMSM_4K_DIV;
1308                         set_pba_of_lba0(map, pba_of_lba0(map)/IMSM_4K_DIV);
1309                 }
1310         }
1311
1312         mpb->check_sum = __gen_imsm_checksum(mpb);
1313 }
1314
1315 void examine_migr_rec_imsm(struct intel_super *super)
1316 {
1317         struct migr_record *migr_rec = super->migr_rec;
1318         struct imsm_super *mpb = super->anchor;
1319         int i;
1320
1321         for (i = 0; i < mpb->num_raid_devs; i++) {
1322                 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
1323                 struct imsm_map *map;
1324                 int slot = -1;
1325
1326                 if (is_gen_migration(dev) == 0)
1327                                 continue;
1328
1329                 printf("\nMigration Record Information:");
1330
1331                 /* first map under migration */
1332                 map = get_imsm_map(dev, MAP_0);
1333                 if (map)
1334                         slot = get_imsm_disk_slot(map, super->disks->index);
1335                 if (map == NULL || slot > 1 || slot < 0) {
1336                         printf(" Empty\n                              ");
1337                         printf("Examine one of first two disks in array\n");
1338                         break;
1339                 }
1340                 printf("\n                     Status : ");
1341                 if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL)
1342                         printf("Normal\n");
1343                 else
1344                         printf("Contains Data\n");
1345                 printf("               Current Unit : %u\n",
1346                        __le32_to_cpu(migr_rec->curr_migr_unit));
1347                 printf("                     Family : %u\n",
1348                        __le32_to_cpu(migr_rec->family_num));
1349                 printf("                  Ascending : %u\n",
1350                        __le32_to_cpu(migr_rec->ascending_migr));
1351                 printf("            Blocks Per Unit : %u\n",
1352                        __le32_to_cpu(migr_rec->blocks_per_unit));
1353                 printf("       Dest. Depth Per Unit : %u\n",
1354                        __le32_to_cpu(migr_rec->dest_depth_per_unit));
1355                 printf("        Checkpoint Area pba : %u\n",
1356                        __le32_to_cpu(migr_rec->ckpt_area_pba));
1357                 printf("           First member lba : %u\n",
1358                        __le32_to_cpu(migr_rec->dest_1st_member_lba));
1359                 printf("      Total Number of Units : %u\n",
1360                        __le32_to_cpu(migr_rec->num_migr_units));
1361                 printf("             Size of volume : %u\n",
1362                        __le32_to_cpu(migr_rec->post_migr_vol_cap));
1363                 printf("  Expansion space for LBA64 : %u\n",
1364                        __le32_to_cpu(migr_rec->post_migr_vol_cap_hi));
1365                 printf("       Record was read from : %u\n",
1366                        __le32_to_cpu(migr_rec->ckpt_read_disk_num));
1367
1368                 break;
1369         }
1370 }
1371 #endif /* MDASSEMBLE */
1372
1373 void convert_from_4k_imsm_migr_rec(struct intel_super *super)
1374 {
1375         struct migr_record *migr_rec = super->migr_rec;
1376
1377         migr_rec->blocks_per_unit *= IMSM_4K_DIV;
1378         migr_rec->ckpt_area_pba *= IMSM_4K_DIV;
1379         migr_rec->dest_1st_member_lba *= IMSM_4K_DIV;
1380         migr_rec->dest_depth_per_unit *= IMSM_4K_DIV;
1381         split_ull((join_u32(migr_rec->post_migr_vol_cap,
1382                  migr_rec->post_migr_vol_cap_hi) * IMSM_4K_DIV),
1383                  &migr_rec->post_migr_vol_cap,
1384                  &migr_rec->post_migr_vol_cap_hi);
1385 }
1386
1387 void convert_from_4k(struct intel_super *super)
1388 {
1389         struct imsm_super *mpb = super->anchor;
1390         struct imsm_disk *disk;
1391         int i;
1392
1393         for (i = 0; i < mpb->num_disks ; i++) {
1394                 disk = __get_imsm_disk(mpb, i);
1395                 /* disk */
1396                 set_total_blocks(disk, (total_blocks(disk)*IMSM_4K_DIV));
1397         }
1398
1399         for (i = 0; i < mpb->num_raid_devs; i++) {
1400                 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
1401                 struct imsm_map *map = get_imsm_map(dev, MAP_0);
1402                 /* dev */
1403                 split_ull((join_u32(dev->size_low, dev->size_high)*IMSM_4K_DIV),
1404                                  &dev->size_low, &dev->size_high);
1405                 dev->vol.curr_migr_unit *= IMSM_4K_DIV;
1406
1407                 /* map0 */
1408                 set_blocks_per_member(map, blocks_per_member(map)*IMSM_4K_DIV);
1409                 map->blocks_per_strip *= IMSM_4K_DIV;
1410                 set_pba_of_lba0(map, pba_of_lba0(map)*IMSM_4K_DIV);
1411
1412                 if (dev->vol.migr_state) {
1413                         /* map1 */
1414                         map = get_imsm_map(dev, MAP_1);
1415                         set_blocks_per_member(map,
1416                             blocks_per_member(map)*IMSM_4K_DIV);
1417                         map->blocks_per_strip *= IMSM_4K_DIV;
1418                         set_pba_of_lba0(map, pba_of_lba0(map)*IMSM_4K_DIV);
1419                 }
1420         }
1421
1422         mpb->check_sum = __gen_imsm_checksum(mpb);
1423 }
1424
1425 /*******************************************************************************
1426  * function: imsm_check_attributes
1427  * Description: Function checks if features represented by attributes flags
1428  *              are supported by mdadm.
1429  * Parameters:
1430  *              attributes - Attributes read from metadata
1431  * Returns:
1432  *              0 - passed attributes contains unsupported features flags
1433  *              1 - all features are supported
1434  ******************************************************************************/
1435 static int imsm_check_attributes(__u32 attributes)
1436 {
1437         int ret_val = 1;
1438         __u32 not_supported = MPB_ATTRIB_SUPPORTED^0xffffffff;
1439
1440         not_supported &= ~MPB_ATTRIB_IGNORED;
1441
1442         not_supported &= attributes;
1443         if (not_supported) {
1444                 pr_err("(IMSM): Unsupported attributes : %x\n",
1445                         (unsigned)__le32_to_cpu(not_supported));
1446                 if (not_supported & MPB_ATTRIB_CHECKSUM_VERIFY) {
1447                         dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1448                         not_supported ^= MPB_ATTRIB_CHECKSUM_VERIFY;
1449                 }
1450                 if (not_supported & MPB_ATTRIB_2TB) {
1451                         dprintf("\t\tMPB_ATTRIB_2TB\n");
1452                         not_supported ^= MPB_ATTRIB_2TB;
1453                 }
1454                 if (not_supported & MPB_ATTRIB_RAID0) {
1455                         dprintf("\t\tMPB_ATTRIB_RAID0\n");
1456                         not_supported ^= MPB_ATTRIB_RAID0;
1457                 }
1458                 if (not_supported & MPB_ATTRIB_RAID1) {
1459                         dprintf("\t\tMPB_ATTRIB_RAID1\n");
1460                         not_supported ^= MPB_ATTRIB_RAID1;
1461                 }
1462                 if (not_supported & MPB_ATTRIB_RAID10) {
1463                         dprintf("\t\tMPB_ATTRIB_RAID10\n");
1464                         not_supported ^= MPB_ATTRIB_RAID10;
1465                 }
1466                 if (not_supported & MPB_ATTRIB_RAID1E) {
1467                         dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1468                         not_supported ^= MPB_ATTRIB_RAID1E;
1469                 }
1470                 if (not_supported & MPB_ATTRIB_RAID5) {
1471                 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1472                         not_supported ^= MPB_ATTRIB_RAID5;
1473                 }
1474                 if (not_supported & MPB_ATTRIB_RAIDCNG) {
1475                         dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1476                         not_supported ^= MPB_ATTRIB_RAIDCNG;
1477                 }
1478                 if (not_supported & MPB_ATTRIB_BBM) {
1479                         dprintf("\t\tMPB_ATTRIB_BBM\n");
1480                 not_supported ^= MPB_ATTRIB_BBM;
1481                 }
1482                 if (not_supported & MPB_ATTRIB_CHECKSUM_VERIFY) {
1483                         dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1484                         not_supported ^= MPB_ATTRIB_CHECKSUM_VERIFY;
1485                 }
1486                 if (not_supported & MPB_ATTRIB_EXP_STRIPE_SIZE) {
1487                         dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1488                         not_supported ^= MPB_ATTRIB_EXP_STRIPE_SIZE;
1489                 }
1490                 if (not_supported & MPB_ATTRIB_2TB_DISK) {
1491                         dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1492                         not_supported ^= MPB_ATTRIB_2TB_DISK;
1493                 }
1494                 if (not_supported & MPB_ATTRIB_NEVER_USE2) {
1495                         dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1496                         not_supported ^= MPB_ATTRIB_NEVER_USE2;
1497                 }
1498                 if (not_supported & MPB_ATTRIB_NEVER_USE) {
1499                         dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1500                         not_supported ^= MPB_ATTRIB_NEVER_USE;
1501                 }
1502
1503                 if (not_supported)
1504                         dprintf("(IMSM): Unknown attributes : %x\n", not_supported);
1505
1506                 ret_val = 0;
1507         }
1508
1509         return ret_val;
1510 }
1511
1512 #ifndef MDASSEMBLE
1513 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map);
1514
1515 static void examine_super_imsm(struct supertype *st, char *homehost)
1516 {
1517         struct intel_super *super = st->sb;
1518         struct imsm_super *mpb = super->anchor;
1519         char str[MAX_SIGNATURE_LENGTH];
1520         int i;
1521         struct mdinfo info;
1522         char nbuf[64];
1523         __u32 sum;
1524         __u32 reserved = imsm_reserved_sectors(super, super->disks);
1525         struct dl *dl;
1526
1527         snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
1528         printf("          Magic : %s\n", str);
1529         snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
1530         printf("        Version : %s\n", get_imsm_version(mpb));
1531         printf("    Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
1532         printf("         Family : %08x\n", __le32_to_cpu(mpb->family_num));
1533         printf("     Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
1534         printf("     Attributes : ");
1535         if (imsm_check_attributes(mpb->attributes))
1536                 printf("All supported\n");
1537         else
1538                 printf("not supported\n");
1539         getinfo_super_imsm(st, &info, NULL);
1540         fname_from_uuid(st, &info, nbuf, ':');
1541         printf("           UUID : %s\n", nbuf + 5);
1542         sum = __le32_to_cpu(mpb->check_sum);
1543         printf("       Checksum : %08x %s\n", sum,
1544                 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
1545         printf("    MPB Sectors : %d\n", mpb_sectors(mpb, super->sector_size));
1546         printf("          Disks : %d\n", mpb->num_disks);
1547         printf("   RAID Devices : %d\n", mpb->num_raid_devs);
1548         print_imsm_disk(__get_imsm_disk(mpb, super->disks->index), super->disks->index, reserved);
1549         if (super->bbm_log) {
1550                 struct bbm_log *log = super->bbm_log;
1551
1552                 printf("\n");
1553                 printf("Bad Block Management Log:\n");
1554                 printf("       Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
1555                 printf("      Signature : %x\n", __le32_to_cpu(log->signature));
1556                 printf("    Entry Count : %d\n", __le32_to_cpu(log->entry_count));
1557                 printf("   Spare Blocks : %d\n",  __le32_to_cpu(log->reserved_spare_block_count));
1558                 printf("    First Spare : %llx\n",
1559                        (unsigned long long) __le64_to_cpu(log->first_spare_lba));
1560         }
1561         for (i = 0; i < mpb->num_raid_devs; i++) {
1562                 struct mdinfo info;
1563                 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
1564
1565                 super->current_vol = i;
1566                 getinfo_super_imsm(st, &info, NULL);
1567                 fname_from_uuid(st, &info, nbuf, ':');
1568                 print_imsm_dev(super, dev, nbuf + 5, super->disks->index);
1569         }
1570         for (i = 0; i < mpb->num_disks; i++) {
1571                 if (i == super->disks->index)
1572                         continue;
1573                 print_imsm_disk(__get_imsm_disk(mpb, i), i, reserved);
1574         }
1575
1576         for (dl = super->disks; dl; dl = dl->next)
1577                 if (dl->index == -1)
1578                         print_imsm_disk(&dl->disk, -1, reserved);
1579
1580         examine_migr_rec_imsm(super);
1581 }
1582
1583 static void brief_examine_super_imsm(struct supertype *st, int verbose)
1584 {
1585         /* We just write a generic IMSM ARRAY entry */
1586         struct mdinfo info;
1587         char nbuf[64];
1588         struct intel_super *super = st->sb;
1589
1590         if (!super->anchor->num_raid_devs) {
1591                 printf("ARRAY metadata=imsm\n");
1592                 return;
1593         }
1594
1595         getinfo_super_imsm(st, &info, NULL);
1596         fname_from_uuid(st, &info, nbuf, ':');
1597         printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
1598 }
1599
1600 static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
1601 {
1602         /* We just write a generic IMSM ARRAY entry */
1603         struct mdinfo info;
1604         char nbuf[64];
1605         char nbuf1[64];
1606         struct intel_super *super = st->sb;
1607         int i;
1608
1609         if (!super->anchor->num_raid_devs)
1610                 return;
1611
1612         getinfo_super_imsm(st, &info, NULL);
1613         fname_from_uuid(st, &info, nbuf, ':');
1614         for (i = 0; i < super->anchor->num_raid_devs; i++) {
1615                 struct imsm_dev *dev = get_imsm_dev(super, i);
1616
1617                 super->current_vol = i;
1618                 getinfo_super_imsm(st, &info, NULL);
1619                 fname_from_uuid(st, &info, nbuf1, ':');
1620                 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1621                        dev->volume, nbuf + 5, i, nbuf1 + 5);
1622         }
1623 }
1624
1625 static void export_examine_super_imsm(struct supertype *st)
1626 {
1627         struct intel_super *super = st->sb;
1628         struct imsm_super *mpb = super->anchor;
1629         struct mdinfo info;
1630         char nbuf[64];
1631
1632         getinfo_super_imsm(st, &info, NULL);
1633         fname_from_uuid(st, &info, nbuf, ':');
1634         printf("MD_METADATA=imsm\n");
1635         printf("MD_LEVEL=container\n");
1636         printf("MD_UUID=%s\n", nbuf+5);
1637         printf("MD_DEVICES=%u\n", mpb->num_disks);
1638 }
1639
1640 static int copy_metadata_imsm(struct supertype *st, int from, int to)
1641 {
1642         /* The second last sector of the device contains
1643          * the "struct imsm_super" metadata.
1644          * This contains mpb_size which is the size in bytes of the
1645          * extended metadata.  This is located immediately before
1646          * the imsm_super.
1647          * We want to read all that, plus the last sector which
1648          * may contain a migration record, and write it all
1649          * to the target.
1650          */
1651         void *buf;
1652         unsigned long long dsize, offset;
1653         int sectors;
1654         struct imsm_super *sb;
1655         struct intel_super *super = st->sb;
1656         unsigned int sector_size = super->sector_size;
1657         unsigned int written = 0;
1658
1659         if (posix_memalign(&buf, MAX_SECTOR_SIZE, MAX_SECTOR_SIZE) != 0)
1660                 return 1;
1661
1662         if (!get_dev_size(from, NULL, &dsize))
1663                 goto err;
1664
1665         if (lseek64(from, dsize-(2*sector_size), 0) < 0)
1666                 goto err;
1667         if (read(from, buf, sector_size) != sector_size)
1668                 goto err;
1669         sb = buf;
1670         if (strncmp((char*)sb->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0)
1671                 goto err;
1672
1673         sectors = mpb_sectors(sb, sector_size) + 2;
1674         offset = dsize - sectors * sector_size;
1675         if (lseek64(from, offset, 0) < 0 ||
1676             lseek64(to, offset, 0) < 0)
1677                 goto err;
1678         while (written < sectors * sector_size) {
1679                 int n = sectors*sector_size - written;
1680                 if (n > 4096)
1681                         n = 4096;
1682                 if (read(from, buf, n) != n)
1683                         goto err;
1684                 if (write(to, buf, n) != n)
1685                         goto err;
1686                 written += n;
1687         }
1688         free(buf);
1689         return 0;
1690 err:
1691         free(buf);
1692         return 1;
1693 }
1694
1695 static void detail_super_imsm(struct supertype *st, char *homehost)
1696 {
1697         struct mdinfo info;
1698         char nbuf[64];
1699
1700         getinfo_super_imsm(st, &info, NULL);
1701         fname_from_uuid(st, &info, nbuf, ':');
1702         printf("\n           UUID : %s\n", nbuf + 5);
1703 }
1704
1705 static void brief_detail_super_imsm(struct supertype *st)
1706 {
1707         struct mdinfo info;
1708         char nbuf[64];
1709         getinfo_super_imsm(st, &info, NULL);
1710         fname_from_uuid(st, &info, nbuf, ':');
1711         printf(" UUID=%s", nbuf + 5);
1712 }
1713
1714 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
1715 static void fd2devname(int fd, char *name);
1716
1717 static int ahci_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
1718 {
1719         /* dump an unsorted list of devices attached to AHCI Intel storage
1720          * controller, as well as non-connected ports
1721          */
1722         int hba_len = strlen(hba_path) + 1;
1723         struct dirent *ent;
1724         DIR *dir;
1725         char *path = NULL;
1726         int err = 0;
1727         unsigned long port_mask = (1 << port_count) - 1;
1728
1729         if (port_count > (int)sizeof(port_mask) * 8) {
1730                 if (verbose > 0)
1731                         pr_err("port_count %d out of range\n", port_count);
1732                 return 2;
1733         }
1734
1735         /* scroll through /sys/dev/block looking for devices attached to
1736          * this hba
1737          */
1738         dir = opendir("/sys/dev/block");
1739         if (!dir)
1740                 return 1;
1741
1742         for (ent = readdir(dir); ent; ent = readdir(dir)) {
1743                 int fd;
1744                 char model[64];
1745                 char vendor[64];
1746                 char buf[1024];
1747                 int major, minor;
1748                 char *device;
1749                 char *c;
1750                 int port;
1751                 int type;
1752
1753                 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
1754                         continue;
1755                 path = devt_to_devpath(makedev(major, minor));
1756                 if (!path)
1757                         continue;
1758                 if (!path_attached_to_hba(path, hba_path)) {
1759                         free(path);
1760                         path = NULL;
1761                         continue;
1762                 }
1763
1764                 /* retrieve the scsi device type */
1765                 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
1766                         if (verbose > 0)
1767                                 pr_err("failed to allocate 'device'\n");
1768                         err = 2;
1769                         break;
1770                 }
1771                 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
1772                 if (load_sys(device, buf, sizeof(buf)) != 0) {
1773                         if (verbose > 0)
1774                                 pr_err("failed to read device type for %s\n",
1775                                         path);
1776                         err = 2;
1777                         free(device);
1778                         break;
1779                 }
1780                 type = strtoul(buf, NULL, 10);
1781
1782                 /* if it's not a disk print the vendor and model */
1783                 if (!(type == 0 || type == 7 || type == 14)) {
1784                         vendor[0] = '\0';
1785                         model[0] = '\0';
1786                         sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
1787                         if (load_sys(device, buf, sizeof(buf)) == 0) {
1788                                 strncpy(vendor, buf, sizeof(vendor));
1789                                 vendor[sizeof(vendor) - 1] = '\0';
1790                                 c = (char *) &vendor[sizeof(vendor) - 1];
1791                                 while (isspace(*c) || *c == '\0')
1792                                         *c-- = '\0';
1793
1794                         }
1795                         sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
1796                         if (load_sys(device, buf, sizeof(buf)) == 0) {
1797                                 strncpy(model, buf, sizeof(model));
1798                                 model[sizeof(model) - 1] = '\0';
1799                                 c = (char *) &model[sizeof(model) - 1];
1800                                 while (isspace(*c) || *c == '\0')
1801                                         *c-- = '\0';
1802                         }
1803
1804                         if (vendor[0] && model[0])
1805                                 sprintf(buf, "%.64s %.64s", vendor, model);
1806                         else
1807                                 switch (type) { /* numbers from hald/linux/device.c */
1808                                 case 1: sprintf(buf, "tape"); break;
1809                                 case 2: sprintf(buf, "printer"); break;
1810                                 case 3: sprintf(buf, "processor"); break;
1811                                 case 4:
1812                                 case 5: sprintf(buf, "cdrom"); break;
1813                                 case 6: sprintf(buf, "scanner"); break;
1814                                 case 8: sprintf(buf, "media_changer"); break;
1815                                 case 9: sprintf(buf, "comm"); break;
1816                                 case 12: sprintf(buf, "raid"); break;
1817                                 default: sprintf(buf, "unknown");
1818                                 }
1819                 } else
1820                         buf[0] = '\0';
1821                 free(device);
1822
1823                 /* chop device path to 'host%d' and calculate the port number */
1824                 c = strchr(&path[hba_len], '/');
1825                 if (!c) {
1826                         if (verbose > 0)
1827                                 pr_err("%s - invalid path name\n", path + hba_len);
1828                         err = 2;
1829                         break;
1830                 }
1831                 *c = '\0';
1832                 if ((sscanf(&path[hba_len], "ata%d", &port) == 1) ||
1833                    ((sscanf(&path[hba_len], "host%d", &port) == 1)))
1834                         port -= host_base;
1835                 else {
1836                         if (verbose > 0) {
1837                                 *c = '/'; /* repair the full string */
1838                                 pr_err("failed to determine port number for %s\n",
1839                                         path);
1840                         }
1841                         err = 2;
1842                         break;
1843                 }
1844
1845                 /* mark this port as used */
1846                 port_mask &= ~(1 << port);
1847
1848                 /* print out the device information */
1849                 if (buf[0]) {
1850                         printf("          Port%d : - non-disk device (%s) -\n", port, buf);
1851                         continue;
1852                 }
1853
1854                 fd = dev_open(ent->d_name, O_RDONLY);
1855                 if (fd < 0)
1856                         printf("          Port%d : - disk info unavailable -\n", port);
1857                 else {
1858                         fd2devname(fd, buf);
1859                         printf("          Port%d : %s", port, buf);
1860                         if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
1861                                 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN, buf);
1862                         else
1863                                 printf(" ()\n");
1864                         close(fd);
1865                 }
1866                 free(path);
1867                 path = NULL;
1868         }
1869         if (path)
1870                 free(path);
1871         if (dir)
1872                 closedir(dir);
1873         if (err == 0) {
1874                 int i;
1875
1876                 for (i = 0; i < port_count; i++)
1877                         if (port_mask & (1 << i))
1878                                 printf("          Port%d : - no device attached -\n", i);
1879         }
1880
1881         return err;
1882 }
1883
1884 static int print_vmd_attached_devs(struct sys_dev *hba)
1885 {
1886         struct dirent *ent;
1887         DIR *dir;
1888         char path[292];
1889         char link[256];
1890         char *c, *rp;
1891
1892         if (hba->type != SYS_DEV_VMD)
1893                 return 1;
1894
1895         /* scroll through /sys/dev/block looking for devices attached to
1896          * this hba
1897          */
1898         dir = opendir("/sys/bus/pci/drivers/nvme");
1899         if (!dir)
1900                 return 1;
1901
1902         for (ent = readdir(dir); ent; ent = readdir(dir)) {
1903                 int n;
1904
1905                 /* is 'ent' a device? check that the 'subsystem' link exists and
1906                  * that its target matches 'bus'
1907                  */
1908                 sprintf(path, "/sys/bus/pci/drivers/nvme/%s/subsystem",
1909                         ent->d_name);
1910                 n = readlink(path, link, sizeof(link));
1911                 if (n < 0 || n >= (int)sizeof(link))
1912                         continue;
1913                 link[n] = '\0';
1914                 c = strrchr(link, '/');
1915                 if (!c)
1916                         continue;
1917                 if (strncmp("pci", c+1, strlen("pci")) != 0)
1918                         continue;
1919
1920                 sprintf(path, "/sys/bus/pci/drivers/nvme/%s", ent->d_name);
1921                 /* if not a intel NVMe - skip it*/
1922                 if (devpath_to_vendor(path) != 0x8086)
1923                         continue;
1924
1925                 rp = realpath(path, NULL);
1926                 if (!rp)
1927                         continue;
1928
1929                 if (path_attached_to_hba(rp, hba->path)) {
1930                         printf(" NVMe under VMD : %s\n", rp);
1931                 }
1932                 free(rp);
1933         }
1934
1935         closedir(dir);
1936         return 0;
1937 }
1938
1939 static void print_found_intel_controllers(struct sys_dev *elem)
1940 {
1941         for (; elem; elem = elem->next) {
1942                 pr_err("found Intel(R) ");
1943                 if (elem->type == SYS_DEV_SATA)
1944                         fprintf(stderr, "SATA ");
1945                 else if (elem->type == SYS_DEV_SAS)
1946                         fprintf(stderr, "SAS ");
1947                 else if (elem->type == SYS_DEV_NVME)
1948                         fprintf(stderr, "NVMe ");
1949
1950                 if (elem->type == SYS_DEV_VMD)
1951                         fprintf(stderr, "VMD domain");
1952                 else
1953                         fprintf(stderr, "RAID controller");
1954
1955                 if (elem->pci_id)
1956                         fprintf(stderr, " at %s", elem->pci_id);
1957                 fprintf(stderr, ".\n");
1958         }
1959         fflush(stderr);
1960 }
1961
1962 static int ahci_get_port_count(const char *hba_path, int *port_count)
1963 {
1964         struct dirent *ent;
1965         DIR *dir;
1966         int host_base = -1;
1967
1968         *port_count = 0;
1969         if ((dir = opendir(hba_path)) == NULL)
1970                 return -1;
1971
1972         for (ent = readdir(dir); ent; ent = readdir(dir)) {
1973                 int host;
1974
1975                 if ((sscanf(ent->d_name, "ata%d", &host) != 1) &&
1976                    ((sscanf(ent->d_name, "host%d", &host) != 1)))
1977                         continue;
1978                 if (*port_count == 0)
1979                         host_base = host;
1980                 else if (host < host_base)
1981                         host_base = host;
1982
1983                 if (host + 1 > *port_count + host_base)
1984                         *port_count = host + 1 - host_base;
1985         }
1986         closedir(dir);
1987         return host_base;
1988 }
1989
1990 static void print_imsm_capability(const struct imsm_orom *orom)
1991 {
1992         printf("       Platform : Intel(R) ");
1993         if (orom->capabilities == 0 && orom->driver_features == 0)
1994                 printf("Matrix Storage Manager\n");
1995         else
1996                 printf("Rapid Storage Technology%s\n",
1997                         imsm_orom_is_enterprise(orom) ? " enterprise" : "");
1998         if (orom->major_ver || orom->minor_ver || orom->hotfix_ver || orom->build)
1999                 printf("        Version : %d.%d.%d.%d\n", orom->major_ver,
2000                                 orom->minor_ver, orom->hotfix_ver, orom->build);
2001         printf("    RAID Levels :%s%s%s%s%s\n",
2002                imsm_orom_has_raid0(orom) ? " raid0" : "",
2003                imsm_orom_has_raid1(orom) ? " raid1" : "",
2004                imsm_orom_has_raid1e(orom) ? " raid1e" : "",
2005                imsm_orom_has_raid10(orom) ? " raid10" : "",
2006                imsm_orom_has_raid5(orom) ? " raid5" : "");
2007         printf("    Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2008                imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
2009                imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
2010                imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
2011                imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
2012                imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
2013                imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
2014                imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
2015                imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
2016                imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
2017                imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
2018                imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
2019                imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
2020                imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
2021                imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
2022                imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
2023                imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
2024         printf("    2TB volumes :%s supported\n",
2025                (orom->attr & IMSM_OROM_ATTR_2TB)?"":" not");
2026         printf("      2TB disks :%s supported\n",
2027                (orom->attr & IMSM_OROM_ATTR_2TB_DISK)?"":" not");
2028         printf("      Max Disks : %d\n", orom->tds);
2029         printf("    Max Volumes : %d per array, %d per %s\n",
2030                orom->vpa, orom->vphba,
2031                imsm_orom_is_nvme(orom) ? "platform" : "controller");
2032         return;
2033 }
2034
2035 static void print_imsm_capability_export(const struct imsm_orom *orom)
2036 {
2037         printf("MD_FIRMWARE_TYPE=imsm\n");
2038         if (orom->major_ver || orom->minor_ver || orom->hotfix_ver || orom->build)
2039                 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
2040                                 orom->hotfix_ver, orom->build);
2041         printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2042                         imsm_orom_has_raid0(orom) ? "raid0 " : "",
2043                         imsm_orom_has_raid1(orom) ? "raid1 " : "",
2044                         imsm_orom_has_raid1e(orom) ? "raid1e " : "",
2045                         imsm_orom_has_raid5(orom) ? "raid10 " : "",
2046                         imsm_orom_has_raid10(orom) ? "raid5 " : "");
2047         printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2048                         imsm_orom_has_chunk(orom, 2) ? "2k " : "",
2049                         imsm_orom_has_chunk(orom, 4) ? "4k " : "",
2050                         imsm_orom_has_chunk(orom, 8) ? "8k " : "",
2051                         imsm_orom_has_chunk(orom, 16) ? "16k " : "",
2052                         imsm_orom_has_chunk(orom, 32) ? "32k " : "",
2053                         imsm_orom_has_chunk(orom, 64) ? "64k " : "",
2054                         imsm_orom_has_chunk(orom, 128) ? "128k " : "",
2055                         imsm_orom_has_chunk(orom, 256) ? "256k " : "",
2056                         imsm_orom_has_chunk(orom, 512) ? "512k " : "",
2057                         imsm_orom_has_chunk(orom, 1024*1) ? "1M " : "",
2058                         imsm_orom_has_chunk(orom, 1024*2) ? "2M " : "",
2059                         imsm_orom_has_chunk(orom, 1024*4) ? "4M " : "",
2060                         imsm_orom_has_chunk(orom, 1024*8) ? "8M " : "",
2061                         imsm_orom_has_chunk(orom, 1024*16) ? "16M " : "",
2062                         imsm_orom_has_chunk(orom, 1024*32) ? "32M " : "",
2063                         imsm_orom_has_chunk(orom, 1024*64) ? "64M " : "");
2064         printf("IMSM_2TB_VOLUMES=%s\n",(orom->attr & IMSM_OROM_ATTR_2TB) ? "yes" : "no");
2065         printf("IMSM_2TB_DISKS=%s\n",(orom->attr & IMSM_OROM_ATTR_2TB_DISK) ? "yes" : "no");
2066         printf("IMSM_MAX_DISKS=%d\n",orom->tds);
2067         printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom->vpa);
2068         printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom->vphba);
2069 }
2070
2071 static int detail_platform_imsm(int verbose, int enumerate_only, char *controller_path)
2072 {
2073         /* There are two components to imsm platform support, the ahci SATA
2074          * controller and the option-rom.  To find the SATA controller we
2075          * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2076          * controller with the Intel vendor id is present.  This approach
2077          * allows mdadm to leverage the kernel's ahci detection logic, with the
2078          * caveat that if ahci.ko is not loaded mdadm will not be able to
2079          * detect platform raid capabilities.  The option-rom resides in a
2080          * platform "Adapter ROM".  We scan for its signature to retrieve the
2081          * platform capabilities.  If raid support is disabled in the BIOS the
2082          * option-rom capability structure will not be available.
2083          */
2084         struct sys_dev *list, *hba;
2085         int host_base = 0;
2086         int port_count = 0;
2087         int result=1;
2088
2089         if (enumerate_only) {
2090                 if (check_env("IMSM_NO_PLATFORM"))
2091                         return 0;
2092                 list = find_intel_devices();
2093                 if (!list)
2094                         return 2;
2095                 for (hba = list; hba; hba = hba->next) {
2096                         if (find_imsm_capability(hba)) {
2097                                 result = 0;
2098                                 break;
2099                         }
2100                         else
2101                                 result = 2;
2102                 }
2103                 return result;
2104         }
2105
2106         list = find_intel_devices();
2107         if (!list) {
2108                 if (verbose > 0)
2109                         pr_err("no active Intel(R) RAID controller found.\n");
2110                 return 2;
2111         } else if (verbose > 0)
2112                 print_found_intel_controllers(list);
2113
2114         for (hba = list; hba; hba = hba->next) {
2115                 if (controller_path && (compare_paths(hba->path, controller_path) != 0))
2116                         continue;
2117                 if (!find_imsm_capability(hba)) {
2118                         char buf[PATH_MAX];
2119                         pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2120                                   hba->type == SYS_DEV_VMD ? vmd_domain_to_controller(hba, buf) : hba->path,
2121                                   get_sys_dev_type(hba->type));
2122                         continue;
2123                 }
2124                 result = 0;
2125         }
2126
2127         if (controller_path && result == 1) {
2128                 pr_err("no active Intel(R) RAID controller found under %s\n",
2129                                 controller_path);
2130                 return result;
2131         }
2132
2133         const struct orom_entry *entry;
2134
2135         for (entry = orom_entries; entry; entry = entry->next) {
2136                 if (entry->type == SYS_DEV_VMD) {
2137                         print_imsm_capability(&entry->orom);
2138                         for (hba = list; hba; hba = hba->next) {
2139                                 if (hba->type == SYS_DEV_VMD) {
2140                                         char buf[PATH_MAX];
2141                                         printf(" I/O Controller : %s (%s)\n",
2142                                                 vmd_domain_to_controller(hba, buf), get_sys_dev_type(hba->type));
2143                                         if (print_vmd_attached_devs(hba)) {
2144                                                 if (verbose > 0)
2145                                                         pr_err("failed to get devices attached to VMD domain.\n");
2146                                                 result |= 2;
2147                                         }
2148                                 }
2149                         }
2150                         printf("\n");
2151                         continue;
2152                 }
2153
2154                 print_imsm_capability(&entry->orom);
2155                 if (entry->type == SYS_DEV_NVME) {
2156                         for (hba = list; hba; hba = hba->next) {
2157                                 if (hba->type == SYS_DEV_NVME)
2158                                         printf("    NVMe Device : %s\n", hba->path);
2159                         }
2160                         printf("\n");
2161                         continue;
2162                 }
2163
2164                 struct devid_list *devid;
2165                 for (devid = entry->devid_list; devid; devid = devid->next) {
2166                         hba = device_by_id(devid->devid);
2167                         if (!hba)
2168                                 continue;
2169
2170                         printf(" I/O Controller : %s (%s)\n",
2171                                 hba->path, get_sys_dev_type(hba->type));
2172                         if (hba->type == SYS_DEV_SATA) {
2173                                 host_base = ahci_get_port_count(hba->path, &port_count);
2174                                 if (ahci_enumerate_ports(hba->path, port_count, host_base, verbose)) {
2175                                         if (verbose > 0)
2176                                                 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba->pci_id);
2177                                         result |= 2;
2178                                 }
2179                         }
2180                 }
2181                 printf("\n");
2182         }
2183
2184         return result;
2185 }
2186
2187 static int export_detail_platform_imsm(int verbose, char *controller_path)
2188 {
2189         struct sys_dev *list, *hba;
2190         int result=1;
2191
2192         list = find_intel_devices();
2193         if (!list) {
2194                 if (verbose > 0)
2195                         pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2196                 result = 2;
2197                 return result;
2198         }
2199
2200         for (hba = list; hba; hba = hba->next) {
2201                 if (controller_path && (compare_paths(hba->path,controller_path) != 0))
2202                         continue;
2203                 if (!find_imsm_capability(hba) && verbose > 0) {
2204                         char buf[PATH_MAX];
2205                         pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2206                         hba->type == SYS_DEV_VMD ? vmd_domain_to_controller(hba, buf) : hba->path);
2207                 }
2208                 else
2209                         result = 0;
2210         }
2211
2212         const struct orom_entry *entry;
2213
2214         for (entry = orom_entries; entry; entry = entry->next) {
2215                 if (entry->type == SYS_DEV_VMD) {
2216                         for (hba = list; hba; hba = hba->next)
2217                                 print_imsm_capability_export(&entry->orom);
2218                         continue;
2219                 }
2220                 print_imsm_capability_export(&entry->orom);
2221         }
2222
2223         return result;
2224 }
2225
2226 #endif
2227
2228 static int match_home_imsm(struct supertype *st, char *homehost)
2229 {
2230         /* the imsm metadata format does not specify any host
2231          * identification information.  We return -1 since we can never
2232          * confirm nor deny whether a given array is "meant" for this
2233          * host.  We rely on compare_super and the 'family_num' fields to
2234          * exclude member disks that do not belong, and we rely on
2235          * mdadm.conf to specify the arrays that should be assembled.
2236          * Auto-assembly may still pick up "foreign" arrays.
2237          */
2238
2239         return -1;
2240 }
2241
2242 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
2243 {
2244         /* The uuid returned here is used for:
2245          *  uuid to put into bitmap file (Create, Grow)
2246          *  uuid for backup header when saving critical section (Grow)
2247          *  comparing uuids when re-adding a device into an array
2248          *    In these cases the uuid required is that of the data-array,
2249          *    not the device-set.
2250          *  uuid to recognise same set when adding a missing device back
2251          *    to an array.   This is a uuid for the device-set.
2252          *
2253          * For each of these we can make do with a truncated
2254          * or hashed uuid rather than the original, as long as
2255          * everyone agrees.
2256          * In each case the uuid required is that of the data-array,
2257          * not the device-set.
2258          */
2259         /* imsm does not track uuid's so we synthesis one using sha1 on
2260          * - The signature (Which is constant for all imsm array, but no matter)
2261          * - the orig_family_num of the container
2262          * - the index number of the volume
2263          * - the 'serial' number of the volume.
2264          * Hopefully these are all constant.
2265          */
2266         struct intel_super *super = st->sb;
2267
2268         char buf[20];
2269         struct sha1_ctx ctx;
2270         struct imsm_dev *dev = NULL;
2271         __u32 family_num;
2272
2273         /* some mdadm versions failed to set ->orig_family_num, in which
2274          * case fall back to ->family_num.  orig_family_num will be
2275          * fixed up with the first metadata update.
2276          */
2277         family_num = super->anchor->orig_family_num;
2278         if (family_num == 0)
2279                 family_num = super->anchor->family_num;
2280         sha1_init_ctx(&ctx);
2281         sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
2282         sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
2283         if (super->current_vol >= 0)
2284                 dev = get_imsm_dev(super, super->current_vol);
2285         if (dev) {
2286                 __u32 vol = super->current_vol;
2287                 sha1_process_bytes(&vol, sizeof(vol), &ctx);
2288                 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
2289         }
2290         sha1_finish_ctx(&ctx, buf);
2291         memcpy(uuid, buf, 4*4);
2292 }
2293
2294 #if 0
2295 static void
2296 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
2297 {
2298         __u8 *v = get_imsm_version(mpb);
2299         __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
2300         char major[] = { 0, 0, 0 };
2301         char minor[] = { 0 ,0, 0 };
2302         char patch[] = { 0, 0, 0 };
2303         char *ver_parse[] = { major, minor, patch };
2304         int i, j;
2305
2306         i = j = 0;
2307         while (*v != '\0' && v < end) {
2308                 if (*v != '.' && j < 2)
2309                         ver_parse[i][j++] = *v;
2310                 else {
2311                         i++;
2312                         j = 0;
2313                 }
2314                 v++;
2315         }
2316
2317         *m = strtol(minor, NULL, 0);
2318         *p = strtol(patch, NULL, 0);
2319 }
2320 #endif
2321
2322 static __u32 migr_strip_blocks_resync(struct imsm_dev *dev)
2323 {
2324         /* migr_strip_size when repairing or initializing parity */
2325         struct imsm_map *map = get_imsm_map(dev, MAP_0);
2326         __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
2327
2328         switch (get_imsm_raid_level(map)) {
2329         case 5:
2330         case 10:
2331                 return chunk;
2332         default:
2333                 return 128*1024 >> 9;
2334         }
2335 }
2336
2337 static __u32 migr_strip_blocks_rebuild(struct imsm_dev *dev)
2338 {
2339         /* migr_strip_size when rebuilding a degraded disk, no idea why
2340          * this is different than migr_strip_size_resync(), but it's good
2341          * to be compatible
2342          */
2343         struct imsm_map *map = get_imsm_map(dev, MAP_1);
2344         __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
2345
2346         switch (get_imsm_raid_level(map)) {
2347         case 1:
2348         case 10:
2349                 if (map->num_members % map->num_domains == 0)
2350                         return 128*1024 >> 9;
2351                 else
2352                         return chunk;
2353         case 5:
2354                 return max((__u32) 64*1024 >> 9, chunk);
2355         default:
2356                 return 128*1024 >> 9;
2357         }
2358 }
2359
2360 static __u32 num_stripes_per_unit_resync(struct imsm_dev *dev)
2361 {
2362         struct imsm_map *lo = get_imsm_map(dev, MAP_0);
2363         struct imsm_map *hi = get_imsm_map(dev, MAP_1);
2364         __u32 lo_chunk = __le32_to_cpu(lo->blocks_per_strip);
2365         __u32 hi_chunk = __le32_to_cpu(hi->blocks_per_strip);
2366
2367         return max((__u32) 1, hi_chunk / lo_chunk);
2368 }
2369
2370 static __u32 num_stripes_per_unit_rebuild(struct imsm_dev *dev)
2371 {
2372         struct imsm_map *lo = get_imsm_map(dev, MAP_0);
2373         int level = get_imsm_raid_level(lo);
2374
2375         if (level == 1 || level == 10) {
2376                 struct imsm_map *hi = get_imsm_map(dev, MAP_1);
2377
2378                 return hi->num_domains;
2379         } else
2380                 return num_stripes_per_unit_resync(dev);
2381 }
2382
2383 static __u8 imsm_num_data_members(struct imsm_dev *dev, int second_map)
2384 {
2385         /* named 'imsm_' because raid0, raid1 and raid10
2386          * counter-intuitively have the same number of data disks
2387          */
2388         struct imsm_map *map = get_imsm_map(dev, second_map);
2389
2390         switch (get_imsm_raid_level(map)) {
2391         case 0:
2392                 return map->num_members;
2393                 break;
2394         case 1:
2395         case 10:
2396                 return map->num_members/2;
2397         case 5:
2398                 return map->num_members - 1;
2399         default:
2400                 dprintf("unsupported raid level\n");
2401                 return 0;
2402         }
2403 }
2404
2405 static __u32 parity_segment_depth(struct imsm_dev *dev)
2406 {
2407         struct imsm_map *map = get_imsm_map(dev, MAP_0);
2408         __u32 chunk =  __le32_to_cpu(map->blocks_per_strip);
2409
2410         switch(get_imsm_raid_level(map)) {
2411         case 1:
2412         case 10:
2413                 return chunk * map->num_domains;
2414         case 5:
2415                 return chunk * map->num_members;
2416         default:
2417                 return chunk;
2418         }
2419 }
2420
2421 static __u32 map_migr_block(struct imsm_dev *dev, __u32 block)
2422 {
2423         struct imsm_map *map = get_imsm_map(dev, MAP_1);
2424         __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
2425         __u32 strip = block / chunk;
2426
2427         switch (get_imsm_raid_level(map)) {
2428         case 1:
2429         case 10: {
2430                 __u32 vol_strip = (strip * map->num_domains) + 1;
2431                 __u32 vol_stripe = vol_strip / map->num_members;
2432
2433                 return vol_stripe * chunk + block % chunk;
2434         } case 5: {
2435                 __u32 stripe = strip / (map->num_members - 1);
2436
2437                 return stripe * chunk + block % chunk;
2438         }
2439         default:
2440                 return 0;
2441         }
2442 }
2443
2444 static __u64 blocks_per_migr_unit(struct intel_super *super,
2445                                   struct imsm_dev *dev)
2446 {
2447         /* calculate the conversion factor between per member 'blocks'
2448          * (md/{resync,rebuild}_start) and imsm migration units, return
2449          * 0 for the 'not migrating' and 'unsupported migration' cases
2450          */
2451         if (!dev->vol.migr_state)
2452                 return 0;
2453
2454         switch (migr_type(dev)) {
2455         case MIGR_GEN_MIGR: {
2456                 struct migr_record *migr_rec = super->migr_rec;
2457                 return __le32_to_cpu(migr_rec->blocks_per_unit);
2458         }
2459         case MIGR_VERIFY:
2460         case MIGR_REPAIR:
2461         case MIGR_INIT: {
2462                 struct imsm_map *map = get_imsm_map(dev, MAP_0);
2463                 __u32 stripes_per_unit;
2464                 __u32 blocks_per_unit;
2465                 __u32 parity_depth;
2466                 __u32 migr_chunk;
2467                 __u32 block_map;
2468                 __u32 block_rel;
2469                 __u32 segment;
2470                 __u32 stripe;
2471                 __u8  disks;
2472
2473                 /* yes, this is really the translation of migr_units to
2474                  * per-member blocks in the 'resync' case
2475                  */
2476                 stripes_per_unit = num_stripes_per_unit_resync(dev);
2477                 migr_chunk = migr_strip_blocks_resync(dev);
2478                 disks = imsm_num_data_members(dev, MAP_0);
2479                 blocks_per_unit = stripes_per_unit * migr_chunk * disks;
2480                 stripe = __le16_to_cpu(map->blocks_per_strip) * disks;
2481                 segment = blocks_per_unit / stripe;
2482                 block_rel = blocks_per_unit - segment * stripe;
2483                 parity_depth = parity_segment_depth(dev);
2484                 block_map = map_migr_block(dev, block_rel);
2485                 return block_map + parity_depth * segment;
2486         }
2487         case MIGR_REBUILD: {
2488                 __u32 stripes_per_unit;
2489                 __u32 migr_chunk;
2490
2491                 stripes_per_unit = num_stripes_per_unit_rebuild(dev);
2492                 migr_chunk = migr_strip_blocks_rebuild(dev);
2493                 return migr_chunk * stripes_per_unit;
2494         }
2495         case MIGR_STATE_CHANGE:
2496         default:
2497                 return 0;
2498         }
2499 }
2500
2501 static int imsm_level_to_layout(int level)
2502 {
2503         switch (level) {
2504         case 0:
2505         case 1:
2506                 return 0;
2507         case 5:
2508         case 6:
2509                 return ALGORITHM_LEFT_ASYMMETRIC;
2510         case 10:
2511                 return 0x102;
2512         }
2513         return UnSet;
2514 }
2515
2516 /*******************************************************************************
2517  * Function:    read_imsm_migr_rec
2518  * Description: Function reads imsm migration record from last sector of disk
2519  * Parameters:
2520  *      fd      : disk descriptor
2521  *      super   : metadata info
2522  * Returns:
2523  *       0 : success,
2524  *      -1 : fail
2525  ******************************************************************************/
2526 static int read_imsm_migr_rec(int fd, struct intel_super *super)
2527 {
2528         int ret_val = -1;
2529         unsigned int sector_size = super->sector_size;
2530         unsigned long long dsize;
2531
2532         get_dev_size(fd, NULL, &dsize);
2533         if (lseek64(fd, dsize - (sector_size*MIGR_REC_SECTOR_POSITION),
2534                    SEEK_SET) < 0) {
2535                 pr_err("Cannot seek to anchor block: %s\n",
2536                        strerror(errno));
2537                 goto out;
2538         }
2539         if (read(fd, super->migr_rec_buf,
2540             MIGR_REC_BUF_SECTORS*sector_size) !=
2541             MIGR_REC_BUF_SECTORS*sector_size) {
2542                 pr_err("Cannot read migr record block: %s\n",
2543                        strerror(errno));
2544                 goto out;
2545         }
2546         ret_val = 0;
2547         if (sector_size == 4096)
2548                 convert_from_4k_imsm_migr_rec(super);
2549
2550 out:
2551         return ret_val;
2552 }
2553
2554 static struct imsm_dev *imsm_get_device_during_migration(
2555         struct intel_super *super)
2556 {
2557
2558         struct intel_dev *dv;
2559
2560         for (dv = super->devlist; dv; dv = dv->next) {
2561                 if (is_gen_migration(dv->dev))
2562                         return dv->dev;
2563         }
2564         return NULL;
2565 }
2566
2567 /*******************************************************************************
2568  * Function:    load_imsm_migr_rec
2569  * Description: Function reads imsm migration record (it is stored at the last
2570  *              sector of disk)
2571  * Parameters:
2572  *      super   : imsm internal array info
2573  *      info    : general array info
2574  * Returns:
2575  *       0 : success
2576  *      -1 : fail
2577  *      -2 : no migration in progress
2578  ******************************************************************************/
2579 static int load_imsm_migr_rec(struct intel_super *super, struct mdinfo *info)
2580 {
2581         struct mdinfo *sd;
2582         struct dl *dl;
2583         char nm[30];
2584         int retval = -1;
2585         int fd = -1;
2586         struct imsm_dev *dev;
2587         struct imsm_map *map;
2588         int slot = -1;
2589
2590         /* find map under migration */
2591         dev = imsm_get_device_during_migration(super);
2592         /* nothing to load,no migration in progress?
2593         */
2594         if (dev == NULL)
2595                 return -2;
2596
2597         if (info) {
2598                 for (sd = info->devs ; sd ; sd = sd->next) {
2599                         /* read only from one of the first two slots */
2600                         if ((sd->disk.raid_disk < 0) ||
2601                             (sd->disk.raid_disk > 1))
2602                                 continue;
2603
2604                         sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2605                         fd = dev_open(nm, O_RDONLY);
2606                         if (fd >= 0)
2607                                 break;
2608                 }
2609         }
2610         if (fd < 0) {
2611                 map = get_imsm_map(dev, MAP_0);
2612                 for (dl = super->disks; dl; dl = dl->next) {
2613                         /* skip spare and failed disks
2614                         */
2615                         if (dl->index < 0)
2616                                 continue;
2617                         /* read only from one of the first two slots */
2618                         if (map)
2619                                 slot = get_imsm_disk_slot(map, dl->index);
2620                         if (map == NULL || slot > 1 || slot < 0)
2621                                 continue;
2622                         sprintf(nm, "%d:%d", dl->major, dl->minor);
2623                         fd = dev_open(nm, O_RDONLY);
2624                         if (fd >= 0)
2625                                 break;
2626                 }
2627         }
2628         if (fd < 0)
2629                 goto out;
2630         retval = read_imsm_migr_rec(fd, super);
2631
2632 out:
2633         if (fd >= 0)
2634                 close(fd);
2635         return retval;
2636 }
2637
2638 #ifndef MDASSEMBLE
2639 /*******************************************************************************
2640  * function: imsm_create_metadata_checkpoint_update
2641  * Description: It creates update for checkpoint change.
2642  * Parameters:
2643  *      super   : imsm internal array info
2644  *      u       : pointer to prepared update
2645  * Returns:
2646  *      Uptate length.
2647  *      If length is equal to 0, input pointer u contains no update
2648  ******************************************************************************/
2649 static int imsm_create_metadata_checkpoint_update(
2650         struct intel_super *super,
2651         struct imsm_update_general_migration_checkpoint **u)
2652 {
2653
2654         int update_memory_size = 0;
2655
2656         dprintf("(enter)\n");
2657
2658         if (u == NULL)
2659                 return 0;
2660         *u = NULL;
2661
2662         /* size of all update data without anchor */
2663         update_memory_size =
2664                 sizeof(struct imsm_update_general_migration_checkpoint);
2665
2666         *u = xcalloc(1, update_memory_size);
2667         if (*u == NULL) {
2668                 dprintf("error: cannot get memory\n");
2669                 return 0;
2670         }
2671         (*u)->type = update_general_migration_checkpoint;
2672         (*u)->curr_migr_unit = __le32_to_cpu(super->migr_rec->curr_migr_unit);
2673         dprintf("prepared for %u\n", (*u)->curr_migr_unit);
2674
2675         return update_memory_size;
2676 }
2677
2678 static void imsm_update_metadata_locally(struct supertype *st,
2679                                          void *buf, int len);
2680
2681 /*******************************************************************************
2682  * Function:    write_imsm_migr_rec
2683  * Description: Function writes imsm migration record
2684  *              (at the last sector of disk)
2685  * Parameters:
2686  *      super   : imsm internal array info
2687  * Returns:
2688  *       0 : success
2689  *      -1 : if fail
2690  ******************************************************************************/
2691 static int write_imsm_migr_rec(struct supertype *st)
2692 {
2693         struct intel_super *super = st->sb;
2694         unsigned int sector_size = super->sector_size;
2695         unsigned long long dsize;
2696         char nm[30];
2697         int fd = -1;
2698         int retval = -1;
2699         struct dl *sd;
2700         int len;
2701         struct imsm_update_general_migration_checkpoint *u;
2702         struct imsm_dev *dev;
2703         struct imsm_map *map;
2704
2705         /* find map under migration */
2706         dev = imsm_get_device_during_migration(super);
2707         /* if no migration, write buffer anyway to clear migr_record
2708          * on disk based on first available device
2709         */
2710         if (dev == NULL)
2711                 dev = get_imsm_dev(super, super->current_vol < 0 ? 0 :
2712                                           super->current_vol);
2713
2714         map = get_imsm_map(dev, MAP_0);
2715
2716         if (sector_size == 4096)
2717                 convert_to_4k_imsm_migr_rec(super);
2718         for (sd = super->disks ; sd ; sd = sd->next) {
2719                 int slot = -1;
2720
2721                 /* skip failed and spare devices */
2722                 if (sd->index < 0)
2723                         continue;
2724                 /* write to 2 first slots only */
2725                 if (map)
2726                         slot = get_imsm_disk_slot(map, sd->index);
2727                 if (map == NULL || slot > 1 || slot < 0)
2728                         continue;
2729
2730                 sprintf(nm, "%d:%d", sd->major, sd->minor);
2731                 fd = dev_open(nm, O_RDWR);
2732                 if (fd < 0)
2733                         continue;
2734                 get_dev_size(fd, NULL, &dsize);
2735                 if (lseek64(fd, dsize - (MIGR_REC_SECTOR_POSITION*sector_size),
2736                     SEEK_SET) < 0) {
2737                         pr_err("Cannot seek to anchor block: %s\n",
2738                                strerror(errno));
2739                         goto out;
2740                 }
2741                 if (write(fd, super->migr_rec_buf,
2742                     MIGR_REC_BUF_SECTORS*sector_size) !=
2743                     MIGR_REC_BUF_SECTORS*sector_size) {
2744                         pr_err("Cannot write migr record block: %s\n",
2745                                strerror(errno));
2746                         goto out;
2747                 }
2748                 close(fd);
2749                 fd = -1;
2750         }
2751         if (sector_size == 4096)
2752                 convert_from_4k_imsm_migr_rec(super);
2753         /* update checkpoint information in metadata */
2754         len = imsm_create_metadata_checkpoint_update(super, &u);
2755         if (len <= 0) {
2756                 dprintf("imsm: Cannot prepare update\n");
2757                 goto out;
2758         }
2759         /* update metadata locally */
2760         imsm_update_metadata_locally(st, u, len);
2761         /* and possibly remotely */
2762         if (st->update_tail) {
2763                 append_metadata_update(st, u, len);
2764                 /* during reshape we do all work inside metadata handler
2765                  * manage_reshape(), so metadata update has to be triggered
2766                  * insida it
2767                  */
2768                 flush_metadata_updates(st);
2769                 st->update_tail = &st->updates;
2770         } else
2771                 free(u);
2772
2773         retval = 0;
2774  out:
2775         if (fd >= 0)
2776                 close(fd);
2777         return retval;
2778 }
2779 #endif /* MDASSEMBLE */
2780
2781 /* spare/missing disks activations are not allowe when
2782  * array/container performs reshape operation, because
2783  * all arrays in container works on the same disks set
2784  */
2785 int imsm_reshape_blocks_arrays_changes(struct intel_super *super)
2786 {
2787         int rv = 0;
2788         struct intel_dev *i_dev;
2789         struct imsm_dev *dev;
2790
2791         /* check whole container
2792          */
2793         for (i_dev = super->devlist; i_dev; i_dev = i_dev->next) {
2794                 dev = i_dev->dev;
2795                 if (is_gen_migration(dev)) {
2796                         /* No repair during any migration in container
2797                          */
2798                         rv = 1;
2799                         break;
2800                 }
2801         }
2802         return rv;
2803 }
2804 static unsigned long long imsm_component_size_aligment_check(int level,
2805                                               int chunk_size,
2806                                               unsigned int sector_size,
2807                                               unsigned long long component_size)
2808 {
2809         unsigned int component_size_alligment;
2810
2811         /* check component size aligment
2812         */
2813         component_size_alligment = component_size % (chunk_size/sector_size);
2814
2815         dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
2816                 level, chunk_size, component_size,
2817                 component_size_alligment);
2818
2819         if (component_size_alligment && (level != 1) && (level != UnSet)) {
2820                 dprintf("imsm: reported component size alligned from %llu ",
2821                         component_size);
2822                 component_size -= component_size_alligment;
2823                 dprintf_cont("to %llu (%i).\n",
2824                         component_size, component_size_alligment);
2825         }
2826
2827         return component_size;
2828 }
2829
2830 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info, char *dmap)
2831 {
2832         struct intel_super *super = st->sb;
2833         struct migr_record *migr_rec = super->migr_rec;
2834         struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
2835         struct imsm_map *map = get_imsm_map(dev, MAP_0);
2836         struct imsm_map *prev_map = get_imsm_map(dev, MAP_1);
2837         struct imsm_map *map_to_analyse = map;
2838         struct dl *dl;
2839         int map_disks = info->array.raid_disks;
2840
2841         memset(info, 0, sizeof(*info));
2842         if (prev_map)
2843                 map_to_analyse = prev_map;
2844
2845         dl = super->current_disk;
2846
2847         info->container_member    = super->current_vol;
2848         info->array.raid_disks    = map->num_members;
2849         info->array.level         = get_imsm_raid_level(map_to_analyse);
2850         info->array.layout        = imsm_level_to_layout(info->array.level);
2851         info->array.md_minor      = -1;
2852         info->array.ctime         = 0;
2853         info->array.utime         = 0;
2854         info->array.chunk_size    =
2855                 __le16_to_cpu(map_to_analyse->blocks_per_strip) << 9;
2856         info->array.state         = !dev->vol.dirty;
2857         info->custom_array_size   = __le32_to_cpu(dev->size_high);
2858         info->custom_array_size   <<= 32;
2859         info->custom_array_size   |= __le32_to_cpu(dev->size_low);
2860         info->recovery_blocked = imsm_reshape_blocks_arrays_changes(st->sb);
2861
2862         if (is_gen_migration(dev)) {
2863                 info->reshape_active = 1;
2864                 info->new_level = get_imsm_raid_level(map);
2865                 info->new_layout = imsm_level_to_layout(info->new_level);
2866                 info->new_chunk = __le16_to_cpu(map->blocks_per_strip) << 9;
2867                 info->delta_disks = map->num_members - prev_map->num_members;
2868                 if (info->delta_disks) {
2869                         /* this needs to be applied to every array
2870                          * in the container.
2871                          */
2872                         info->reshape_active = CONTAINER_RESHAPE;
2873                 }
2874                 /* We shape information that we give to md might have to be
2875                  * modify to cope with md's requirement for reshaping arrays.
2876                  * For example, when reshaping a RAID0, md requires it to be
2877                  * presented as a degraded RAID4.
2878                  * Also if a RAID0 is migrating to a RAID5 we need to specify
2879                  * the array as already being RAID5, but the 'before' layout
2880                  * is a RAID4-like layout.
2881                  */
2882                 switch (info->array.level) {
2883                 case 0:
2884                         switch(info->new_level) {
2885                         case 0:
2886                                 /* conversion is happening as RAID4 */
2887                                 info->array.level = 4;
2888                                 info->array.raid_disks += 1;
2889                                 break;
2890                         case 5:
2891                                 /* conversion is happening as RAID5 */
2892                                 info->array.level = 5;
2893                                 info->array.layout = ALGORITHM_PARITY_N;
2894                                 info->delta_disks -= 1;
2895                                 break;
2896                         default:
2897                                 /* FIXME error message */
2898                                 info->array.level = UnSet;
2899                                 break;
2900                         }
2901                         break;
2902                 }
2903         } else {
2904                 info->new_level = UnSet;
2905                 info->new_layout = UnSet;
2906                 info->new_chunk = info->array.chunk_size;
2907                 info->delta_disks = 0;
2908         }
2909
2910         if (dl) {
2911                 info->disk.major = dl->major;
2912                 info->disk.minor = dl->minor;
2913                 info->disk.number = dl->index;
2914                 info->disk.raid_disk = get_imsm_disk_slot(map_to_analyse,
2915                                                           dl->index);
2916         }
2917
2918         info->data_offset         = pba_of_lba0(map_to_analyse);
2919         info->component_size      = blocks_per_member(map_to_analyse);
2920
2921         info->component_size = imsm_component_size_aligment_check(
2922                                                         info->array.level,
2923                                                         info->array.chunk_size,
2924                                                         super->sector_size,
2925                                                         info->component_size);
2926
2927         memset(info->uuid, 0, sizeof(info->uuid));
2928         info->recovery_start = MaxSector;
2929
2930         info->reshape_progress = 0;
2931         info->resync_start = MaxSector;
2932         if ((map_to_analyse->map_state == IMSM_T_STATE_UNINITIALIZED ||
2933             dev->vol.dirty) &&
2934             imsm_reshape_blocks_arrays_changes(super) == 0) {
2935                 info->resync_start = 0;
2936         }
2937         if (dev->vol.migr_state) {
2938                 switch (migr_type(dev)) {
2939                 case MIGR_REPAIR:
2940                 case MIGR_INIT: {
2941                         __u64 blocks_per_unit = blocks_per_migr_unit(super,
2942                                                                      dev);
2943                         __u64 units = __le32_to_cpu(dev->vol.curr_migr_unit);
2944
2945                         info->resync_start = blocks_per_unit * units;
2946                         break;
2947                 }
2948                 case MIGR_GEN_MIGR: {
2949                         __u64 blocks_per_unit = blocks_per_migr_unit(super,
2950                                                                      dev);
2951                         __u64 units = __le32_to_cpu(migr_rec->curr_migr_unit);
2952                         unsigned long long array_blocks;
2953                         int used_disks;
2954
2955                         if (__le32_to_cpu(migr_rec->ascending_migr) &&
2956                             (units <
2957                                 (__le32_to_cpu(migr_rec->num_migr_units)-1)) &&
2958                             (super->migr_rec->rec_status ==
2959                                         __cpu_to_le32(UNIT_SRC_IN_CP_AREA)))
2960                                 units++;
2961
2962                         info->reshape_progress = blocks_per_unit * units;
2963
2964                         dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
2965                                 (unsigned long long)units,
2966                                 (unsigned long long)blocks_per_unit,
2967                                 info->reshape_progress);
2968
2969                         used_disks = imsm_num_data_members(dev, MAP_1);
2970                         if (used_disks > 0) {
2971                                 array_blocks = blocks_per_member(map) *
2972                                         used_disks;
2973                                 /* round array size down to closest MB
2974                                  */
2975                                 info->custom_array_size = (array_blocks
2976                                                 >> SECT_PER_MB_SHIFT)
2977                                                 << SECT_PER_MB_SHIFT;
2978                         }
2979                 }
2980                 case MIGR_VERIFY:
2981                         /* we could emulate the checkpointing of
2982                          * 'sync_action=check' migrations, but for now
2983                          * we just immediately complete them
2984                          */
2985                 case MIGR_REBUILD:
2986                         /* this is handled by container_content_imsm() */
2987                 case MIGR_STATE_CHANGE:
2988                         /* FIXME handle other migrations */
2989                 default:
2990                         /* we are not dirty, so... */
2991                         info->resync_start = MaxSector;
2992                 }
2993         }
2994
2995         strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
2996         info->name[MAX_RAID_SERIAL_LEN] = 0;
2997
2998         info->array.major_version = -1;
2999         info->array.minor_version = -2;
3000         sprintf(info->text_version, "/%s/%d", st->container_devnm, info->container_member);
3001         info->safe_mode_delay = 4000;  /* 4 secs like the Matrix driver */
3002         uuid_from_super_imsm(st, info->uuid);
3003
3004         if (dmap) {
3005                 int i, j;
3006                 for (i=0; i<map_disks; i++) {
3007                         dmap[i] = 0;
3008                         if (i < info->array.raid_disks) {
3009                                 struct imsm_disk *dsk;
3010                                 j = get_imsm_disk_idx(dev, i, MAP_X);
3011                                 dsk = get_imsm_disk(super, j);
3012                                 if (dsk && (dsk->status & CONFIGURED_DISK))
3013                                         dmap[i] = 1;
3014                         }
3015                 }
3016         }
3017 }
3018
3019 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev,
3020                                 int failed, int look_in_map);
3021
3022 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev,
3023                              int look_in_map);
3024
3025 #ifndef MDASSEMBLE
3026 static void manage_second_map(struct intel_super *super, struct imsm_dev *dev)
3027 {
3028         if (is_gen_migration(dev)) {
3029                 int failed;
3030                 __u8 map_state;
3031                 struct imsm_map *map2 = get_imsm_map(dev, MAP_1);
3032
3033                 failed = imsm_count_failed(super, dev, MAP_1);
3034                 map_state = imsm_check_degraded(super, dev, failed, MAP_1);
3035                 if (map2->map_state != map_state) {
3036                         map2->map_state = map_state;
3037                         super->updates_pending++;
3038                 }
3039         }
3040 }
3041 #endif
3042
3043 static struct imsm_disk *get_imsm_missing(struct intel_super *super, __u8 index)
3044 {
3045         struct dl *d;
3046
3047         for (d = super->missing; d; d = d->next)
3048                 if (d->index == index)
3049                         return &d->disk;
3050         return NULL;
3051 }
3052
3053 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map)
3054 {
3055         struct intel_super *super = st->sb;
3056         struct imsm_disk *disk;
3057         int map_disks = info->array.raid_disks;
3058         int max_enough = -1;
3059         int i;
3060         struct imsm_super *mpb;
3061
3062         if (super->current_vol >= 0) {
3063                 getinfo_super_imsm_volume(st, info, map);
3064                 return;
3065         }
3066         memset(info, 0, sizeof(*info));
3067
3068         /* Set raid_disks to zero so that Assemble will always pull in valid
3069          * spares
3070          */
3071         info->array.raid_disks    = 0;
3072         info->array.level         = LEVEL_CONTAINER;
3073         info->array.layout        = 0;
3074         info->array.md_minor      = -1;
3075         info->array.ctime         = 0; /* N/A for imsm */
3076         info->array.utime         = 0;
3077         info->array.chunk_size    = 0;
3078
3079         info->disk.major = 0;
3080         info->disk.minor = 0;
3081         info->disk.raid_disk = -1;
3082         info->reshape_active = 0;
3083         info->array.major_version = -1;
3084         info->array.minor_version = -2;
3085         strcpy(info->text_version, "imsm");
3086         info->safe_mode_delay = 0;
3087         info->disk.number = -1;
3088         info->disk.state = 0;
3089         info->name[0] = 0;
3090         info->recovery_start = MaxSector;
3091         info->recovery_blocked = imsm_reshape_blocks_arrays_changes(st->sb);
3092
3093         /* do we have the all the insync disks that we expect? */
3094         mpb = super->anchor;
3095
3096         for (i = 0; i < mpb->num_raid_devs; i++) {
3097                 struct imsm_dev *dev = get_imsm_dev(super, i);
3098                 int failed, enough, j, missing = 0;
3099                 struct imsm_map *map;
3100                 __u8 state;
3101
3102                 failed = imsm_count_failed(super, dev, MAP_0);
3103                 state = imsm_check_degraded(super, dev, failed, MAP_0);
3104                 map = get_imsm_map(dev, MAP_0);
3105
3106                 /* any newly missing disks?
3107                  * (catches single-degraded vs double-degraded)
3108                  */
3109                 for (j = 0; j < map->num_members; j++) {
3110                         __u32 ord = get_imsm_ord_tbl_ent(dev, j, MAP_0);
3111                         __u32 idx = ord_to_idx(ord);
3112
3113                         if (!(ord & IMSM_ORD_REBUILD) &&
3114                             get_imsm_missing(super, idx)) {
3115                                 missing = 1;
3116                                 break;
3117                         }
3118                 }
3119
3120                 if (state == IMSM_T_STATE_FAILED)
3121                         enough = -1;
3122                 else if (state == IMSM_T_STATE_DEGRADED &&
3123                          (state != map->map_state || missing))
3124                         enough = 0;
3125                 else /* we're normal, or already degraded */
3126                         enough = 1;
3127                 if (is_gen_migration(dev) && missing) {
3128                         /* during general migration we need all disks
3129                          * that process is running on.
3130                          * No new missing disk is allowed.
3131                          */
3132                         max_enough = -1;
3133                         enough = -1;
3134                         /* no more checks necessary
3135                          */
3136                         break;
3137                 }
3138                 /* in the missing/failed disk case check to see
3139                  * if at least one array is runnable
3140                  */
3141                 max_enough = max(max_enough, enough);
3142         }
3143         dprintf("enough: %d\n", max_enough);
3144         info->container_enough = max_enough;
3145
3146         if (super->disks) {
3147                 __u32 reserved = imsm_reserved_sectors(super, super->disks);
3148
3149                 disk = &super->disks->disk;
3150                 info->data_offset = total_blocks(&super->disks->disk) - reserved;
3151                 info->component_size = reserved;
3152                 info->disk.state  = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
3153                 /* we don't change info->disk.raid_disk here because
3154                  * this state will be finalized in mdmon after we have
3155                  * found the 'most fresh' version of the metadata
3156                  */
3157                 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
3158                 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
3159         }
3160
3161         /* only call uuid_from_super_imsm when this disk is part of a populated container,
3162          * ->compare_super may have updated the 'num_raid_devs' field for spares
3163          */
3164         if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
3165                 uuid_from_super_imsm(st, info->uuid);
3166         else
3167                 memcpy(info->uuid, uuid_zero, sizeof(uuid_zero));
3168
3169         /* I don't know how to compute 'map' on imsm, so use safe default */
3170         if (map) {
3171                 int i;
3172                 for (i = 0; i < map_disks; i++)
3173                         map[i] = 1;
3174         }
3175
3176 }
3177
3178 /* allocates memory and fills disk in mdinfo structure
3179  * for each disk in array */
3180 struct mdinfo *getinfo_super_disks_imsm(struct supertype *st)
3181 {
3182         struct mdinfo *mddev;
3183         struct intel_super *super = st->sb;
3184         struct imsm_disk *disk;
3185         int count = 0;
3186         struct dl *dl;
3187         if (!super || !super->disks)
3188                 return NULL;
3189         dl = super->disks;
3190         mddev = xcalloc(1, sizeof(*mddev));
3191         while (dl) {
3192                 struct mdinfo *tmp;
3193                 disk = &dl->disk;
3194                 tmp = xcalloc(1, sizeof(*tmp));
3195                 if (mddev->devs)
3196                         tmp->next = mddev->devs;
3197                 mddev->devs = tmp;
3198                 tmp->disk.number = count++;
3199                 tmp->disk.major = dl->major;
3200                 tmp->disk.minor = dl->minor;
3201                 tmp->disk.state = is_configured(disk) ?
3202                                   (1 << MD_DISK_ACTIVE) : 0;
3203                 tmp->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
3204                 tmp->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
3205                 tmp->disk.raid_disk = -1;
3206                 dl = dl->next;
3207         }
3208         return mddev;
3209 }
3210
3211 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
3212                              char *update, char *devname, int verbose,
3213                              int uuid_set, char *homehost)
3214 {
3215         /* For 'assemble' and 'force' we need to return non-zero if any
3216          * change was made.  For others, the return value is ignored.
3217          * Update options are:
3218          *  force-one : This device looks a bit old but needs to be included,
3219          *        update age info appropriately.
3220          *  assemble: clear any 'faulty' flag to allow this device to
3221          *              be assembled.
3222          *  force-array: Array is degraded but being forced, mark it clean
3223          *         if that will be needed to assemble it.
3224          *
3225          *  newdev:  not used ????
3226          *  grow:  Array has gained a new device - this is currently for
3227          *              linear only
3228          *  resync: mark as dirty so a resync will happen.
3229          *  name:  update the name - preserving the homehost
3230          *  uuid:  Change the uuid of the array to match watch is given
3231          *
3232          * Following are not relevant for this imsm:
3233          *  sparc2.2 : update from old dodgey metadata
3234          *  super-minor: change the preferred_minor number
3235          *  summaries:  update redundant counters.
3236          *  homehost:  update the recorded homehost
3237          *  _reshape_progress: record new reshape_progress position.
3238          */
3239         int rv = 1;
3240         struct intel_super *super = st->sb;
3241         struct imsm_super *mpb;
3242
3243         /* we can only update container info */
3244         if (!super || super->current_vol >= 0 || !super->anchor)
3245                 return 1;
3246
3247         mpb = super->anchor;
3248
3249         if (strcmp(update, "uuid") == 0) {
3250                 /* We take this to mean that the family_num should be updated.
3251                  * However that is much smaller than the uuid so we cannot really
3252                  * allow an explicit uuid to be given.  And it is hard to reliably
3253                  * know if one was.
3254                  * So if !uuid_set we know the current uuid is random and just used
3255                  * the first 'int' and copy it to the other 3 positions.
3256                  * Otherwise we require the 4 'int's to be the same as would be the
3257                  * case if we are using a random uuid.  So an explicit uuid will be
3258                  * accepted as long as all for ints are the same... which shouldn't hurt
3259                  */
3260                 if (!uuid_set) {
3261                         info->uuid[1] = info->uuid[2] = info->uuid[3] = info->uuid[0];
3262                         rv = 0;
3263                 } else {
3264                         if (info->uuid[0] != info->uuid[1] ||
3265                             info->uuid[1] != info->uuid[2] ||
3266                             info->uuid[2] != info->uuid[3])
3267                                 rv = -1;
3268                         else
3269                                 rv = 0;
3270                 }
3271                 if (rv == 0)
3272                         mpb->orig_family_num = info->uuid[0];
3273         } else if (strcmp(update, "assemble") == 0)
3274                 rv = 0;
3275         else
3276                 rv = -1;
3277
3278         /* successful update? recompute checksum */
3279         if (rv == 0)
3280                 mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb));
3281
3282         return rv;
3283 }
3284
3285 static size_t disks_to_mpb_size(int disks)
3286 {
3287         size_t size;
3288
3289         size = sizeof(struct imsm_super);
3290         size += (disks - 1) * sizeof(struct imsm_disk);
3291         size += 2 * sizeof(struct imsm_dev);
3292         /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3293         size += (4 - 2) * sizeof(struct imsm_map);
3294         /* 4 possible disk_ord_tbl's */
3295         size += 4 * (disks - 1) * sizeof(__u32);
3296
3297         return size;
3298 }
3299
3300 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize,
3301                              unsigned long long data_offset)
3302 {
3303         if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
3304                 return 0;
3305
3306         return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
3307 }
3308
3309 static void free_devlist(struct intel_super *super)
3310 {
3311         struct intel_dev *dv;
3312
3313         while (super->devlist) {
3314                 dv = super->devlist->next;
3315                 free(super->devlist->dev);
3316                 free(super->devlist);
3317                 super->devlist = dv;
3318         }
3319 }
3320
3321 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
3322 {
3323         memcpy(dest, src, sizeof_imsm_dev(src, 0));
3324 }
3325
3326 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
3327 {
3328         /*
3329          * return:
3330          *  0 same, or first was empty, and second was copied
3331          *  1 second had wrong number
3332          *  2 wrong uuid
3333          *  3 wrong other info
3334          */
3335         struct intel_super *first = st->sb;
3336         struct intel_super *sec = tst->sb;
3337
3338         if (!first) {
3339                 st->sb = tst->sb;
3340                 tst->sb = NULL;
3341                 return 0;
3342         }
3343         /* in platform dependent environment test if the disks
3344          * use the same Intel hba
3345          * If not on Intel hba at all, allow anything.
3346          */
3347         if (!check_env("IMSM_NO_PLATFORM") && first->hba && sec->hba) {
3348                 if (first->hba->type != sec->hba->type) {
3349                         fprintf(stderr,
3350                                 "HBAs of devices do not match %s != %s\n",
3351                                 get_sys_dev_type(first->hba->type),
3352                                 get_sys_dev_type(sec->hba->type));
3353                         return 3;
3354                 }
3355                 if (first->orom != sec->orom) {
3356                         fprintf(stderr,
3357                                 "HBAs of devices do not match %s != %s\n",
3358                                 first->hba->pci_id, sec->hba->pci_id);
3359                         return 3;
3360                 }
3361         }
3362
3363         /* if an anchor does not have num_raid_devs set then it is a free
3364          * floating spare
3365          */
3366         if (first->anchor->num_raid_devs > 0 &&
3367             sec->anchor->num_raid_devs > 0) {
3368                 /* Determine if these disks might ever have been
3369                  * related.  Further disambiguation can only take place
3370                  * in load_super_imsm_all
3371                  */
3372                 __u32 first_family = first->anchor->orig_family_num;
3373                 __u32 sec_family = sec->anchor->orig_family_num;
3374
3375                 if (memcmp(first->anchor->sig, sec->anchor->sig,
3376                            MAX_SIGNATURE_LENGTH) != 0)
3377                         return 3;
3378
3379                 if (first_family == 0)
3380                         first_family = first->anchor->family_num;
3381                 if (sec_family == 0)
3382                         sec_family = sec->anchor->family_num;
3383
3384                 if (first_family != sec_family)
3385                         return 3;
3386
3387         }
3388
3389         /* if 'first' is a spare promote it to a populated mpb with sec's
3390          * family number
3391          */
3392         if (first->anchor->num_raid_devs == 0 &&
3393             sec->anchor->num_raid_devs > 0) {
3394                 int i;
3395                 struct intel_dev *dv;
3396                 struct imsm_dev *dev;
3397
3398                 /* we need to copy raid device info from sec if an allocation
3399                  * fails here we don't associate the spare
3400                  */
3401                 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
3402                         dv = xmalloc(sizeof(*dv));
3403                         dev = xmalloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
3404                         dv->dev = dev;
3405                         dv->index = i;
3406                         dv->next = first->devlist;
3407                         first->devlist = dv;
3408                 }
3409                 if (i < sec->anchor->num_raid_devs) {
3410                         /* allocation failure */
3411                         free_devlist(first);
3412                         pr_err("imsm: failed to associate spare\n");
3413                         return 3;
3414                 }
3415                 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
3416                 first->anchor->orig_family_num = sec->anchor->orig_family_num;
3417                 first->anchor->family_num = sec->anchor->family_num;
3418                 memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH);
3419                 for (i = 0; i < sec->anchor->num_raid_devs; i++)
3420                         imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
3421         }
3422
3423         return 0;
3424 }
3425
3426 static void fd2devname(int fd, char *name)
3427 {
3428         struct stat st;
3429         char path[256];
3430         char dname[PATH_MAX];
3431         char *nm;
3432         int rv;
3433
3434         name[0] = '\0';
3435         if (fstat(fd, &st) != 0)
3436                 return;
3437         sprintf(path, "/sys/dev/block/%d:%d",
3438                 major(st.st_rdev), minor(st.st_rdev));
3439
3440         rv = readlink(path, dname, sizeof(dname)-1);
3441         if (rv <= 0)
3442                 return;
3443
3444         dname[rv] = '\0';
3445         nm = strrchr(dname, '/');
3446         if (nm) {
3447                 nm++;
3448                 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
3449         }
3450 }
3451
3452 static int nvme_get_serial(int fd, void *buf, size_t buf_len)
3453 {
3454         char path[60];
3455         char *name = fd2kname(fd);
3456
3457         if (!name)
3458                 return 1;
3459
3460         if (strncmp(name, "nvme", 4) != 0)
3461                 return 1;
3462
3463         snprintf(path, sizeof(path) - 1, "/sys/block/%s/device/serial", name);
3464
3465         return load_sys(path, buf, buf_len);
3466 }
3467
3468 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
3469
3470 static int imsm_read_serial(int fd, char *devname,
3471                             __u8 serial[MAX_RAID_SERIAL_LEN])
3472 {
3473         char buf[50];
3474         int rv;
3475         int len;
3476         char *dest;
3477         char *src;
3478         unsigned int i;
3479
3480         memset(buf, 0, sizeof(buf));
3481
3482         rv = nvme_get_serial(fd, buf, sizeof(buf));
3483
3484         if (rv)
3485                 rv = scsi_get_serial(fd, buf, sizeof(buf));
3486
3487         if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
3488                 memset(serial, 0, MAX_RAID_SERIAL_LEN);
3489                 fd2devname(fd, (char *) serial);
3490                 return 0;
3491         }
3492
3493         if (rv != 0) {
3494                 if (devname)
3495                         pr_err("Failed to retrieve serial for %s\n",
3496                                devname);
3497                 return rv;
3498         }
3499
3500         /* trim all whitespace and non-printable characters and convert
3501          * ':' to ';'
3502          */
3503         for (i = 0, dest = buf; i < sizeof(buf) && buf[i]; i++) {
3504                 src = &buf[i];
3505                 if (*src > 0x20) {
3506                         /* ':' is reserved for use in placeholder serial
3507                          * numbers for missing disks
3508                          */
3509                         if (*src == ':')
3510                                 *dest++ = ';';
3511                         else
3512                                 *dest++ = *src;
3513                 }
3514         }
3515         len = dest - buf;
3516         dest = buf;
3517
3518         /* truncate leading characters */
3519         if (len > MAX_RAID_SERIAL_LEN) {
3520                 dest += len - MAX_RAID_SERIAL_LEN;
3521                 len = MAX_RAID_SERIAL_LEN;
3522         }
3523
3524         memset(serial, 0, MAX_RAID_SERIAL_LEN);
3525         memcpy(serial, dest, len);
3526
3527         return 0;
3528 }
3529
3530 static int serialcmp(__u8 *s1, __u8 *s2)
3531 {
3532         return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
3533 }
3534
3535 static void serialcpy(__u8 *dest, __u8 *src)
3536 {
3537         strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
3538 }
3539
3540 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
3541 {
3542         struct dl *dl;
3543
3544         for (dl = super->disks; dl; dl = dl->next)
3545                 if (serialcmp(dl->serial, serial) == 0)
3546                         break;
3547
3548         return dl;
3549 }
3550
3551 static struct imsm_disk *
3552 __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
3553 {
3554         int i;
3555
3556         for (i = 0; i < mpb->num_disks; i++) {
3557                 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
3558
3559                 if (serialcmp(disk->serial, serial) == 0) {
3560                         if (idx)
3561                                 *idx = i;
3562                         return disk;
3563                 }
3564         }
3565
3566         return NULL;
3567 }
3568
3569 static int
3570 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
3571 {
3572         struct imsm_disk *disk;
3573         struct dl *dl;
3574         struct stat stb;
3575         int rv;
3576         char name[40];
3577         __u8 serial[MAX_RAID_SERIAL_LEN];
3578
3579         rv = imsm_read_serial(fd, devname, serial);
3580
3581         if (rv != 0)
3582                 return 2;
3583
3584         dl = xcalloc(1, sizeof(*dl));
3585
3586         fstat(fd, &stb);
3587         dl->major = major(stb.st_rdev);
3588         dl->minor = minor(stb.st_rdev);
3589         dl->next = super->disks;
3590         dl->fd = keep_fd ? fd : -1;
3591         assert(super->disks == NULL);
3592         super->disks = dl;
3593         serialcpy(dl->serial, serial);
3594         dl->index = -2;
3595         dl->e = NULL;
3596         fd2devname(fd, name);
3597         if (devname)
3598                 dl->devname = xstrdup(devname);
3599         else
3600                 dl->devname = xstrdup(name);
3601
3602         /* look up this disk's index in the current anchor */
3603         disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
3604         if (disk) {
3605                 dl->disk = *disk;
3606                 /* only set index on disks that are a member of a
3607                  * populated contianer, i.e. one with raid_devs
3608                  */
3609                 if (is_failed(&dl->disk))
3610                         dl->index = -2;
3611                 else if (is_spare(&dl->disk))
3612                         dl->index = -1;
3613         }
3614
3615         return 0;
3616 }
3617
3618 #ifndef MDASSEMBLE
3619 /* When migrating map0 contains the 'destination' state while map1
3620  * contains the current state.  When not migrating map0 contains the
3621  * current state.  This routine assumes that map[0].map_state is set to
3622  * the current array state before being called.
3623  *
3624  * Migration is indicated by one of the following states
3625  * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3626  * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3627  *    map1state=unitialized)
3628  * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR  map0state=normal
3629  *    map1state=normal)
3630  * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3631  *    map1state=degraded)
3632  * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3633  *    map1state=normal)
3634  */
3635 static void migrate(struct imsm_dev *dev, struct intel_super *super,
3636                     __u8 to_state, int migr_type)
3637 {
3638         struct imsm_map *dest;
3639         struct imsm_map *src = get_imsm_map(dev, MAP_0);
3640
3641         dev->vol.migr_state = 1;
3642         set_migr_type(dev, migr_type);
3643         dev->vol.curr_migr_unit = 0;
3644         dest = get_imsm_map(dev, MAP_1);
3645
3646         /* duplicate and then set the target end state in map[0] */
3647         memcpy(dest, src, sizeof_imsm_map(src));
3648         if (migr_type == MIGR_REBUILD || migr_type ==  MIGR_GEN_MIGR) {
3649                 __u32 ord;
3650                 int i;
3651
3652                 for (i = 0; i < src->num_members; i++) {
3653                         ord = __le32_to_cpu(src->disk_ord_tbl[i]);
3654                         set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
3655                 }
3656         }
3657
3658         if (migr_type == MIGR_GEN_MIGR)
3659                 /* Clear migration record */
3660                 memset(super->migr_rec, 0, sizeof(struct migr_record));
3661
3662         src->map_state = to_state;
3663 }
3664
3665 static void end_migration(struct imsm_dev *dev, struct intel_super *super,
3666                           __u8 map_state)
3667 {
3668         struct imsm_map *map = get_imsm_map(dev, MAP_0);
3669         struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state == 0 ?
3670                                                     MAP_0 : MAP_1);
3671         int i, j;
3672
3673         /* merge any IMSM_ORD_REBUILD bits that were not successfully
3674          * completed in the last migration.
3675          *
3676          * FIXME add support for raid-level-migration
3677          */
3678         if (map_state != map->map_state && (is_gen_migration(dev) == 0) &&
3679             prev->map_state != IMSM_T_STATE_UNINITIALIZED) {
3680                 /* when final map state is other than expected
3681                  * merge maps (not for migration)
3682                  */
3683                 int failed;
3684
3685                 for (i = 0; i < prev->num_members; i++)
3686                         for (j = 0; j < map->num_members; j++)
3687                                 /* during online capacity expansion
3688                                  * disks position can be changed
3689                                  * if takeover is used
3690                                  */
3691                                 if (ord_to_idx(map->disk_ord_tbl[j]) ==
3692                                     ord_to_idx(prev->disk_ord_tbl[i])) {
3693                                         map->disk_ord_tbl[j] |=
3694                                                 prev->disk_ord_tbl[i];
3695                                         break;
3696                                 }
3697                 failed = imsm_count_failed(super, dev, MAP_0);
3698                 map_state = imsm_check_degraded(super, dev, failed, MAP_0);
3699         }
3700
3701         dev->vol.migr_state = 0;
3702         set_migr_type(dev, 0);
3703         dev->vol.curr_migr_unit = 0;
3704         map->map_state = map_state;
3705 }
3706 #endif
3707
3708 static int parse_raid_devices(struct intel_super *super)
3709 {
3710         int i;
3711         struct imsm_dev *dev_new;
3712         size_t len, len_migr;
3713         size_t max_len = 0;
3714         size_t space_needed = 0;
3715         struct imsm_super *mpb = super->anchor;
3716
3717         for (i = 0; i < super->anchor->num_raid_devs; i++) {
3718                 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
3719                 struct intel_dev *dv;
3720
3721                 len = sizeof_imsm_dev(dev_iter, 0);
3722                 len_migr = sizeof_imsm_dev(dev_iter, 1);
3723                 if (len_migr > len)
3724                         space_needed += len_migr - len;
3725
3726                 dv = xmalloc(sizeof(*dv));
3727                 if (max_len < len_migr)
3728                         max_len = len_migr;
3729                 if (max_len > len_migr)
3730                         space_needed += max_len - len_migr;
3731                 dev_new = xmalloc(max_len);
3732                 imsm_copy_dev(dev_new, dev_iter);
3733                 dv->dev = dev_new;
3734                 dv->index = i;
3735                 dv->next = super->devlist;
3736                 super->devlist = dv;
3737         }
3738
3739         /* ensure that super->buf is large enough when all raid devices
3740          * are migrating
3741          */
3742         if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
3743                 void *buf;
3744
3745                 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed,
3746                               super->sector_size);
3747                 if (posix_memalign(&buf, MAX_SECTOR_SIZE, len) != 0)
3748                         return 1;
3749
3750                 memcpy(buf, super->buf, super->len);
3751                 memset(buf + super->len, 0, len - super->len);
3752                 free(super->buf);
3753                 super->buf = buf;
3754                 super->len = len;
3755         }
3756
3757         return 0;
3758 }
3759
3760 /* retrieve a pointer to the bbm log which starts after all raid devices */
3761 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
3762 {
3763         void *ptr = NULL;
3764
3765         if (__le32_to_cpu(mpb->bbm_log_size)) {
3766                 ptr = mpb;
3767                 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
3768         }
3769
3770         return ptr;
3771 }
3772
3773 /*******************************************************************************
3774  * Function:    check_mpb_migr_compatibility
3775  * Description: Function checks for unsupported migration features:
3776  *              - migration optimization area (pba_of_lba0)
3777  *              - descending reshape (ascending_migr)
3778  * Parameters:
3779  *      super   : imsm metadata information
3780  * Returns:
3781  *       0 : migration is compatible
3782  *      -1 : migration is not compatible
3783  ******************************************************************************/
3784 int check_mpb_migr_compatibility(struct intel_super *super)
3785 {
3786         struct imsm_map *map0, *map1;
3787         struct migr_record *migr_rec = super->migr_rec;
3788         int i;
3789
3790         for (i = 0; i < super->anchor->num_raid_devs; i++) {
3791                 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
3792
3793                 if (dev_iter &&
3794                     dev_iter->vol.migr_state == 1 &&
3795                     dev_iter->vol.migr_type == MIGR_GEN_MIGR) {
3796                         /* This device is migrating */
3797                         map0 = get_imsm_map(dev_iter, MAP_0);
3798                         map1 = get_imsm_map(dev_iter, MAP_1);
3799                         if (pba_of_lba0(map0) != pba_of_lba0(map1))
3800                                 /* migration optimization area was used */
3801                                 return -1;
3802                         if (migr_rec->ascending_migr == 0
3803                                 && migr_rec->dest_depth_per_unit > 0)
3804                                 /* descending reshape not supported yet */
3805                                 return -1;
3806                 }
3807         }
3808         return 0;
3809 }
3810
3811 static void __free_imsm(struct intel_super *super, int free_disks);
3812
3813 /* load_imsm_mpb - read matrix metadata
3814  * allocates super->mpb to be freed by free_imsm
3815  */
3816 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
3817 {
3818         unsigned long long dsize;
3819         unsigned long long sectors;
3820         unsigned int sector_size = super->sector_size;
3821         struct stat;
3822         struct imsm_super *anchor;
3823         __u32 check_sum;
3824
3825         get_dev_size(fd, NULL, &dsize);
3826         if (dsize < 2*sector_size) {
3827                 if (devname)
3828                         pr_err("%s: device to small for imsm\n",
3829                                devname);
3830                 return 1;
3831         }
3832
3833         if (lseek64(fd, dsize - (sector_size * 2), SEEK_SET) < 0) {
3834                 if (devname)
3835                         pr_err("Cannot seek to anchor block on %s: %s\n",
3836                                devname, strerror(errno));
3837                 return 1;
3838         }
3839
3840         if (posix_memalign((void **)&anchor, sector_size, sector_size) != 0) {
3841                 if (devname)
3842                         pr_err("Failed to allocate imsm anchor buffer on %s\n", devname);
3843                 return 1;
3844         }
3845         if (read(fd, anchor, sector_size) != sector_size) {
3846                 if (devname)
3847                         pr_err("Cannot read anchor block on %s: %s\n",
3848                                devname, strerror(errno));
3849                 free(anchor);
3850                 return 1;
3851         }
3852
3853         if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
3854                 if (devname)
3855                         pr_err("no IMSM anchor on %s\n", devname);
3856                 free(anchor);
3857                 return 2;
3858         }
3859
3860         __free_imsm(super, 0);
3861         /*  reload capability and hba */
3862
3863         /* capability and hba must be updated with new super allocation */
3864         find_intel_hba_capability(fd, super, devname);
3865         super->len = ROUND_UP(anchor->mpb_size, sector_size);
3866         if (posix_memalign(&super->buf, MAX_SECTOR_SIZE, super->len) != 0) {
3867                 if (devname)
3868                         pr_err("unable to allocate %zu byte mpb buffer\n",
3869                                super->len);
3870                 free(anchor);
3871                 return 2;
3872         }
3873         memcpy(super->buf, anchor, sector_size);
3874
3875         sectors = mpb_sectors(anchor, sector_size) - 1;
3876         free(anchor);
3877
3878         if (posix_memalign(&super->migr_rec_buf, sector_size,
3879             MIGR_REC_BUF_SECTORS*sector_size) != 0) {
3880                 pr_err("could not allocate migr_rec buffer\n");
3881                 free(super->buf);
3882                 return 2;
3883         }
3884         super->clean_migration_record_by_mdmon = 0;
3885
3886         if (!sectors) {
3887                 check_sum = __gen_imsm_checksum(super->anchor);
3888                 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
3889                         if (devname)
3890                                 pr_err("IMSM checksum %x != %x on %s\n",
3891                                        check_sum,
3892                                        __le32_to_cpu(super->anchor->check_sum),
3893                                        devname);
3894                         return 2;
3895                 }
3896
3897                 return 0;
3898         }
3899
3900         /* read the extended mpb */
3901         if (lseek64(fd, dsize - (sector_size * (2 + sectors)), SEEK_SET) < 0) {
3902                 if (devname)
3903                         pr_err("Cannot seek to extended mpb on %s: %s\n",
3904                                devname, strerror(errno));
3905                 return 1;
3906         }
3907
3908         if ((unsigned int)read(fd, super->buf + sector_size,
3909                     super->len - sector_size) != super->len - sector_size) {
3910                 if (devname)
3911                         pr_err("Cannot read extended mpb on %s: %s\n",
3912                                devname, strerror(errno));
3913                 return 2;
3914         }
3915
3916         check_sum = __gen_imsm_checksum(super->anchor);
3917         if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
3918                 if (devname)
3919                         pr_err("IMSM checksum %x != %x on %s\n",
3920                                check_sum, __le32_to_cpu(super->anchor->check_sum),
3921                                devname);
3922                 return 3;
3923         }
3924
3925         /* FIXME the BBM log is disk specific so we cannot use this global
3926          * buffer for all disks.  Ok for now since we only look at the global
3927          * bbm_log_size parameter to gate assembly
3928          */
3929         super->bbm_log = __get_imsm_bbm_log(super->anchor);
3930
3931         return 0;
3932 }
3933
3934 static int read_imsm_migr_rec(int fd, struct intel_super *super);
3935
3936 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
3937 static void clear_hi(struct intel_super *super)
3938 {
3939         struct imsm_super *mpb = super->anchor;
3940         int i, n;
3941         if (mpb->attributes & MPB_ATTRIB_2TB_DISK)
3942                 return;
3943         for (i = 0; i < mpb->num_disks; ++i) {
3944                 struct imsm_disk *disk = &mpb->disk[i];
3945                 disk->total_blocks_hi = 0;
3946         }
3947         for (i = 0; i < mpb->num_raid_devs; ++i) {
3948                 struct imsm_dev *dev = get_imsm_dev(super, i);
3949                 if (!dev)
3950                         return;
3951                 for (n = 0; n < 2; ++n) {
3952                         struct imsm_map *map = get_imsm_map(dev, n);
3953                         if (!map)
3954                                 continue;
3955                         map->pba_of_lba0_hi = 0;
3956                         map->blocks_per_member_hi = 0;
3957                         map->num_data_stripes_hi = 0;
3958                 }
3959         }
3960 }
3961
3962 static int
3963 load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
3964 {
3965         int err;
3966
3967         err = load_imsm_mpb(fd, super, devname);
3968         if (err)
3969                 return err;
3970         if (super->sector_size == 4096)
3971                 convert_from_4k(super);
3972         err = load_imsm_disk(fd, super, devname, keep_fd);
3973         if (err)
3974                 return err;
3975         err = parse_raid_devices(super);
3976         clear_hi(super);
3977         return err;
3978 }
3979
3980 static void __free_imsm_disk(struct dl *d)
3981 {
3982         if (d->fd >= 0)
3983                 close(d->fd);
3984         if (d->devname)
3985                 free(d->devname);
3986         if (d->e)
3987                 free(d->e);
3988         free(d);
3989
3990 }
3991
3992 static void free_imsm_disks(struct intel_super *super)
3993 {
3994         struct dl *d;
3995
3996         while (super->disks) {
3997                 d = super->disks;
3998                 super->disks = d->next;
3999                 __free_imsm_disk(d);
4000         }
4001         while (super->disk_mgmt_list) {
4002                 d = super->disk_mgmt_list;
4003                 super->disk_mgmt_list = d->next;
4004                 __free_imsm_disk(d);
4005         }
4006         while (super->missing) {
4007                 d = super->missing;
4008                 super->missing = d->next;
4009                 __free_imsm_disk(d);
4010         }
4011
4012 }
4013
4014 /* free all the pieces hanging off of a super pointer */
4015 static void __free_imsm(struct intel_super *super, int free_disks)
4016 {
4017         struct intel_hba *elem, *next;
4018
4019         if (super->buf) {
4020                 free(super->buf);
4021                 super->buf = NULL;
4022         }
4023         /* unlink capability description */
4024         super->orom = NULL;
4025         if (super->migr_rec_buf) {
4026                 free(super->migr_rec_buf);
4027                 super->migr_rec_buf = NULL;
4028         }
4029         if (free_disks)
4030                 free_imsm_disks(super);
4031         free_devlist(super);
4032         elem = super->hba;
4033         while (elem) {
4034                 if (elem->path)
4035                         free((void *)elem->path);
4036                 next = elem->next;
4037                 free(elem);
4038                 elem = next;
4039         }
4040         super->hba = NULL;
4041 }
4042
4043 static void free_imsm(struct intel_super *super)
4044 {
4045         __free_imsm(super, 1);
4046         free(super);
4047 }
4048
4049 static void free_super_imsm(struct supertype *st)
4050 {
4051         struct intel_super *super = st->sb;
4052
4053         if (!super)
4054                 return;
4055
4056         free_imsm(super);
4057         st->sb = NULL;
4058 }
4059
4060 static struct intel_super *alloc_super(void)
4061 {
4062         struct intel_super *super = xcalloc(1, sizeof(*super));
4063
4064         super->current_vol = -1;
4065         super->create_offset = ~((unsigned long long) 0);
4066         return super;
4067 }
4068
4069 /*
4070  * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4071  */
4072 static int find_intel_hba_capability(int fd, struct intel_super *super, char *devname)
4073 {
4074         struct sys_dev *hba_name;
4075         int rv = 0;
4076
4077         if (fd < 0 || check_env("IMSM_NO_PLATFORM")) {
4078                 super->orom = NULL;
4079                 super->hba = NULL;
4080                 return 0;
4081         }
4082         hba_name = find_disk_attached_hba(fd, NULL);
4083         if (!hba_name) {
4084                 if (devname)
4085                         pr_err("%s is not attached to Intel(R) RAID controller.\n",
4086                                devname);
4087                 return 1;
4088         }
4089         rv = attach_hba_to_super(super, hba_name);
4090         if (rv == 2) {
4091                 if (devname) {
4092                         struct intel_hba *hba = super->hba;
4093
4094                         pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4095                                 "    but the container is assigned to Intel(R) %s %s (",
4096                                 devname,
4097                                 get_sys_dev_type(hba_name->type),
4098                                 hba_name->type == SYS_DEV_VMD ? "domain" : "RAID controller",
4099                                 hba_name->pci_id ? : "Err!",
4100                                 get_sys_dev_type(super->hba->type),
4101                                 hba->type == SYS_DEV_VMD ? "domain" : "RAID controller");
4102
4103                         while (hba) {
4104                                 fprintf(stderr, "%s", hba->pci_id ? : "Err!");
4105                                 if (hba->next)
4106                                         fprintf(stderr, ", ");
4107                                 hba = hba->next;
4108                         }
4109                         fprintf(stderr, ").\n"
4110                                 "    Mixing devices attached to different %s is not allowed.\n",
4111                                 hba_name->type == SYS_DEV_VMD ? "VMD domains" : "controllers");
4112                 }
4113                 return 2;
4114         }
4115         super->orom = find_imsm_capability(hba_name);
4116         if (!super->orom)
4117                 return 3;
4118
4119         return 0;
4120 }
4121
4122 /* find_missing - helper routine for load_super_imsm_all that identifies
4123  * disks that have disappeared from the system.  This routine relies on
4124  * the mpb being uptodate, which it is at load time.
4125  */
4126 static int find_missing(struct intel_super *super)
4127 {
4128         int i;
4129         struct imsm_super *mpb = super->anchor;
4130         struct dl *dl;
4131         struct imsm_disk *disk;
4132
4133         for (i = 0; i < mpb->num_disks; i++) {
4134                 disk = __get_imsm_disk(mpb, i);
4135                 dl = serial_to_dl(disk->serial, super);
4136                 if (dl)
4137                         continue;
4138
4139                 dl = xmalloc(sizeof(*dl));
4140                 dl->major = 0;
4141                 dl->minor = 0;
4142                 dl->fd = -1;
4143                 dl->devname = xstrdup("missing");
4144                 dl->index = i;
4145                 serialcpy(dl->serial, disk->serial);
4146                 dl->disk = *disk;
4147                 dl->e = NULL;
4148                 dl->next = super->missing;
4149                 super->missing = dl;
4150         }
4151
4152         return 0;
4153 }
4154
4155 #ifndef MDASSEMBLE
4156 static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
4157 {
4158         struct intel_disk *idisk = disk_list;
4159
4160         while (idisk) {
4161                 if (serialcmp(idisk->disk.serial, serial) == 0)
4162                         break;
4163                 idisk = idisk->next;
4164         }
4165
4166         return idisk;
4167 }
4168
4169 static int __prep_thunderdome(struct intel_super **table, int tbl_size,
4170                               struct intel_super *super,
4171                               struct intel_disk **disk_list)
4172 {
4173         struct imsm_disk *d = &super->disks->disk;
4174         struct imsm_super *mpb = super->anchor;
4175         int i, j;
4176
4177         for (i = 0; i < tbl_size; i++) {
4178                 struct imsm_super *tbl_mpb = table[i]->anchor;
4179                 struct imsm_disk *tbl_d = &table[i]->disks->disk;
4180
4181                 if (tbl_mpb->family_num == mpb->family_num) {
4182                         if (tbl_mpb->check_sum == mpb->check_sum) {
4183                                 dprintf("mpb from %d:%d matches %d:%d\n",
4184                                         super->disks->major,
4185                                         super->disks->minor,
4186                                         table[i]->disks->major,
4187                                         table[i]->disks->minor);
4188                                 break;
4189                         }
4190
4191                         if (((is_configured(d) && !is_configured(tbl_d)) ||
4192                              is_configured(d) == is_configured(tbl_d)) &&
4193                             tbl_mpb->generation_num < mpb->generation_num) {
4194                                 /* current version of the mpb is a
4195                                  * better candidate than the one in
4196                                  * super_table, but copy over "cross
4197                                  * generational" status
4198                                  */
4199                                 struct intel_disk *idisk;
4200
4201                                 dprintf("mpb from %d:%d replaces %d:%d\n",
4202                                         super->disks->major,
4203                                         super->disks->minor,
4204                                         table[i]->disks->major,
4205                                         table[i]->disks->minor);
4206
4207                                 idisk = disk_list_get(tbl_d->serial, *disk_list);
4208                                 if (idisk && is_failed(&idisk->disk))
4209                                         tbl_d->status |= FAILED_DISK;
4210                                 break;
4211                         } else {
4212                                 struct intel_disk *idisk;
4213                                 struct imsm_disk *disk;
4214
4215                                 /* tbl_mpb is more up to date, but copy
4216                                  * over cross generational status before
4217                                  * returning
4218                                  */
4219                                 disk = __serial_to_disk(d->serial, mpb, NULL);
4220                                 if (disk && is_failed(disk))
4221                                         d->status |= FAILED_DISK;
4222
4223                                 idisk = disk_list_get(d->serial, *disk_list);
4224                                 if (idisk) {
4225                                         idisk->owner = i;
4226                                         if (disk && is_configured(disk))
4227                                                 idisk->disk.status |= CONFIGURED_DISK;
4228                                 }
4229
4230                                 dprintf("mpb from %d:%d prefer %d:%d\n",
4231                                         super->disks->major,
4232                                         super->disks->minor,
4233                                         table[i]->disks->major,
4234                                         table[i]->disks->minor);
4235
4236                                 return tbl_size;
4237                         }
4238                 }
4239         }
4240
4241         if (i >= tbl_size)
4242                 table[tbl_size++] = super;
4243         else
4244                 table[i] = super;
4245
4246         /* update/extend the merged list of imsm_disk records */
4247         for (j = 0; j < mpb->num_disks; j++) {
4248                 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
4249                 struct intel_disk *idisk;
4250
4251                 idisk = disk_list_get(disk->serial, *disk_list);
4252                 if (idisk) {
4253                         idisk->disk.status |= disk->status;
4254                         if (is_configured(&idisk->disk) ||
4255                             is_failed(&idisk->disk))
4256                                 idisk->disk.status &= ~(SPARE_DISK);
4257                 } else {
4258                         idisk = xcalloc(1, sizeof(*idisk));
4259                         idisk->owner = IMSM_UNKNOWN_OWNER;
4260                         idisk->disk = *disk;
4261                         idisk->next = *disk_list;
4262                         *disk_list = idisk;
4263                 }
4264
4265                 if (serialcmp(idisk->disk.serial, d->serial) == 0)
4266                         idisk->owner = i;
4267         }
4268
4269         return tbl_size;
4270 }
4271
4272 static struct intel_super *
4273 validate_members(struct intel_super *super, struct intel_disk *disk_list,
4274                  const int owner)
4275 {
4276         struct imsm_super *mpb = super->anchor;
4277         int ok_count = 0;
4278         int i;
4279
4280         for (i = 0; i < mpb->num_disks; i++) {
4281                 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
4282                 struct intel_disk *idisk;
4283
4284                 idisk = disk_list_get(disk->serial, disk_list);
4285                 if (idisk) {
4286                         if (idisk->owner == owner ||
4287                             idisk->owner == IMSM_UNKNOWN_OWNER)
4288                                 ok_count++;
4289                         else
4290                                 dprintf("'%.16s' owner %d != %d\n",
4291                                         disk->serial, idisk->owner,
4292                                         owner);
4293                 } else {
4294                         dprintf("unknown disk %x [%d]: %.16s\n",
4295                                 __le32_to_cpu(mpb->family_num), i,
4296                                 disk->serial);
4297                         break;
4298                 }
4299         }
4300
4301         if (ok_count == mpb->num_disks)
4302                 return super;
4303         return NULL;
4304 }
4305
4306 static void show_conflicts(__u32 family_num, struct intel_super *super_list)
4307 {
4308         struct intel_super *s;
4309
4310         for (s = super_list; s; s = s->next) {
4311                 if (family_num != s->anchor->family_num)
4312                         continue;
4313                 pr_err("Conflict, offlining family %#x on '%s'\n",
4314                         __le32_to_cpu(family_num), s->disks->devname);
4315         }
4316 }
4317
4318 static struct intel_super *
4319 imsm_thunderdome(struct intel_super **super_list, int len)
4320 {
4321         struct intel_super *super_table[len];
4322         struct intel_disk *disk_list = NULL;
4323         struct intel_super *champion, *spare;
4324         struct intel_super *s, **del;
4325         int tbl_size = 0;
4326         int conflict;
4327         int i;
4328
4329         memset(super_table, 0, sizeof(super_table));
4330         for (s = *super_list; s; s = s->next)
4331                 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
4332
4333         for (i = 0; i < tbl_size; i++) {
4334                 struct imsm_disk *d;
4335                 struct intel_disk *idisk;
4336                 struct imsm_super *mpb = super_table[i]->anchor;
4337
4338                 s = super_table[i];
4339                 d = &s->disks->disk;
4340
4341                 /* 'd' must appear in merged disk list for its
4342                  * configuration to be valid
4343                  */
4344                 idisk = disk_list_get(d->serial, disk_list);
4345                 if (idisk && idisk->owner == i)
4346                         s = validate_members(s, disk_list, i);
4347                 else
4348                         s = NULL;
4349
4350                 if (!s)
4351                         dprintf("marking family: %#x from %d:%d offline\n",
4352                                 mpb->family_num,
4353                                 super_table[i]->disks->major,
4354                                 super_table[i]->disks->minor);
4355                 super_table[i] = s;
4356         }
4357
4358         /* This is where the mdadm implementation differs from the Windows
4359          * driver which has no strict concept of a container.  We can only
4360          * assemble one family from a container, so when returning a prodigal
4361          * array member to this system the code will not be able to disambiguate
4362          * the container contents that should be assembled ("foreign" versus
4363          * "local").  It requires user intervention to set the orig_family_num
4364          * to a new value to establish a new container.  The Windows driver in
4365          * this situation fixes up the volume name in place and manages the
4366          * foreign array as an independent entity.
4367          */
4368         s = NULL;
4369         spare = NULL;
4370         conflict = 0;
4371         for (i = 0; i < tbl_size; i++) {
4372                 struct intel_super *tbl_ent = super_table[i];
4373                 int is_spare = 0;
4374
4375                 if (!tbl_ent)
4376                         continue;
4377
4378                 if (tbl_ent->anchor->num_raid_devs == 0) {
4379                         spare = tbl_ent;
4380                         is_spare = 1;
4381                 }
4382
4383                 if (s && !is_spare) {
4384                         show_conflicts(tbl_ent->anchor->family_num, *super_list);
4385                         conflict++;
4386                 } else if (!s && !is_spare)
4387                         s = tbl_ent;
4388         }
4389
4390         if (!s)
4391                 s = spare;
4392         if (!s) {
4393                 champion = NULL;
4394                 goto out;
4395         }
4396         champion = s;
4397
4398         if (conflict)
4399                 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4400                         __le32_to_cpu(s->anchor->family_num), s->disks->devname);
4401
4402         /* collect all dl's onto 'champion', and update them to
4403          * champion's version of the status
4404          */
4405         for (s = *super_list; s; s = s->next) {
4406                 struct imsm_super *mpb = champion->anchor;
4407                 struct dl *dl = s->disks;
4408
4409                 if (s == champion)
4410                         continue;
4411
4412                 mpb->attributes |= s->anchor->attributes & MPB_ATTRIB_2TB_DISK;
4413
4414                 for (i = 0; i < mpb->num_disks; i++) {
4415                         struct imsm_disk *disk;
4416
4417                         disk = __serial_to_disk(dl->serial, mpb, &dl->index);
4418                         if (disk) {
4419                                 dl->disk = *disk;
4420                                 /* only set index on disks that are a member of
4421                                  * a populated contianer, i.e. one with
4422                                  * raid_devs
4423                                  */
4424                                 if (is_failed(&dl->disk))
4425                                         dl->index = -2;
4426                                 else if (is_spare(&dl->disk))
4427                                         dl->index = -1;
4428                                 break;
4429                         }
4430                 }
4431
4432                 if (i >= mpb->num_disks) {
4433                         struct intel_disk *idisk;
4434
4435                         idisk = disk_list_get(dl->serial, disk_list);
4436                         if (idisk && is_spare(&idisk->disk) &&
4437                             !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
4438                                 dl->index = -1;
4439                         else {
4440                                 dl->index = -2;
4441                                 continue;
4442                         }
4443                 }
4444
4445                 dl->next = champion->disks;
4446                 champion->disks = dl;
4447                 s->disks = NULL;
4448         }
4449
4450         /* delete 'champion' from super_list */
4451         for (del = super_list; *del; ) {
4452                 if (*del == champion) {
4453                         *del = (*del)->next;
4454                         break;
4455                 } else
4456                         del = &(*del)->next;
4457         }
4458         champion->next = NULL;
4459
4460  out:
4461         while (disk_list) {
4462                 struct intel_disk *idisk = disk_list;
4463
4464                 disk_list = disk_list->next;
4465                 free(idisk);
4466         }
4467
4468         return champion;
4469 }
4470
4471 static int
4472 get_sra_super_block(int fd, struct intel_super **super_list, char *devname, int *max, int keep_fd);
4473 static int get_super_block(struct intel_super **super_list, char *devnm, char *devname,
4474                            int major, int minor, int keep_fd);
4475 static int
4476 get_devlist_super_block(struct md_list *devlist, struct intel_super **super_list,
4477                         int *max, int keep_fd);
4478
4479 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
4480                                char *devname, struct md_list *devlist,
4481                                int keep_fd)
4482 {
4483         struct intel_super *super_list = NULL;
4484         struct intel_super *super = NULL;
4485         int err = 0;
4486         int i = 0;
4487
4488         if (fd >= 0)
4489                 /* 'fd' is an opened container */
4490                 err = get_sra_super_block(fd, &super_list, devname, &i, keep_fd);
4491         else
4492                 /* get super block from devlist devices */
4493                 err = get_devlist_super_block(devlist, &super_list, &i, keep_fd);
4494         if (err)
4495                 goto error;
4496         /* all mpbs enter, maybe one leaves */
4497         super = imsm_thunderdome(&super_list, i);
4498         if (!super) {
4499                 err = 1;
4500                 goto error;
4501         }
4502
4503         if (find_missing(super) != 0) {
4504                 free_imsm(super);
4505                 err = 2;
4506                 goto error;
4507         }
4508
4509         /* load migration record */
4510         err = load_imsm_migr_rec(super, NULL);
4511         if (err == -1) {
4512                 /* migration is in progress,
4513                  * but migr_rec cannot be loaded,
4514                  */
4515                 err = 4;
4516                 goto error;
4517         }
4518
4519         /* Check migration compatibility */
4520         if (err == 0 && check_mpb_migr_compatibility(super) != 0) {
4521                 pr_err("Unsupported migration detected");
4522                 if (devname)
4523                         fprintf(stderr, " on %s\n", devname);
4524                 else
4525                         fprintf(stderr, " (IMSM).\n");
4526
4527                 err = 5;
4528                 goto error;
4529         }
4530
4531         err = 0;
4532
4533  error:
4534         while (super_list) {
4535                 struct intel_super *s = super_list;
4536
4537                 super_list = super_list->next;
4538                 free_imsm(s);
4539         }
4540
4541         if (err)
4542                 return err;
4543
4544         *sbp = super;
4545         if (fd >= 0)
4546                 strcpy(st->container_devnm, fd2devnm(fd));
4547         else
4548                 st->container_devnm[0] = 0;
4549         if (err == 0 && st->ss == NULL) {
4550                 st->ss = &super_imsm;
4551                 st->minor_version = 0;
4552                 st->max_devs = IMSM_MAX_DEVICES;
4553         }
4554         return 0;
4555 }
4556
4557 static int
4558 get_devlist_super_block(struct md_list *devlist, struct intel_super **super_list,
4559                         int *max, int keep_fd)
4560 {
4561         struct md_list *tmpdev;
4562         int err = 0;
4563         int i = 0;
4564
4565         for (i = 0, tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
4566                 if (tmpdev->used != 1)
4567                         continue;
4568                 if (tmpdev->container == 1) {
4569                         int lmax = 0;
4570                         int fd = dev_open(tmpdev->devname, O_RDONLY|O_EXCL);
4571                         if (fd < 0) {
4572                                 pr_err("cannot open device %s: %s\n",
4573                                         tmpdev->devname, strerror(errno));
4574                                 err = 8;
4575                                 goto error;
4576                         }
4577                         err = get_sra_super_block(fd, super_list,
4578                                                   tmpdev->devname, &lmax,
4579                                                   keep_fd);
4580                         i += lmax;
4581                         close(fd);
4582                         if (err) {
4583                                 err = 7;
4584                                 goto error;
4585                         }
4586                 } else {
4587                         int major = major(tmpdev->st_rdev);
4588                         int minor = minor(tmpdev->st_rdev);
4589                         err = get_super_block(super_list,
4590                                               NULL,
4591                                               tmpdev->devname,
4592                                               major, minor,
4593                                               keep_fd);
4594                         i++;
4595                         if (err) {
4596                                 err = 6;
4597                                 goto error;
4598                         }
4599                 }
4600         }
4601  error:
4602         *max = i;
4603         return err;
4604 }
4605
4606 static int get_super_block(struct intel_super **super_list, char *devnm, char *devname,
4607                            int major, int minor, int keep_fd)
4608 {
4609         struct intel_super *s;
4610         char nm[32];
4611         int dfd = -1;
4612         int err = 0;
4613         int retry;
4614
4615         s = alloc_super();
4616         if (!s) {
4617                 err = 1;
4618                 goto error;
4619         }
4620
4621         sprintf(nm, "%d:%d", major, minor);
4622         dfd = dev_open(nm, O_RDWR);
4623         if (dfd < 0) {
4624                 err = 2;
4625                 goto error;
4626         }
4627
4628         get_dev_sector_size(dfd, NULL, &s->sector_size);
4629         find_intel_hba_capability(dfd, s, devname);
4630         err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
4631
4632         /* retry the load if we might have raced against mdmon */
4633         if (err == 3 && devnm && mdmon_running(devnm))
4634                 for (retry = 0; retry < 3; retry++) {
4635                         usleep(3000);
4636                         err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
4637                         if (err != 3)
4638                                 break;
4639                 }
4640  error:
4641         if (!err) {
4642                 s->next = *super_list;
4643                 *super_list = s;
4644         } else {
4645                 if (s)
4646                         free(s);
4647                 if (dfd >= 0)
4648                         close(dfd);
4649         }
4650         if (dfd >= 0 && !keep_fd)
4651                 close(dfd);
4652         return err;
4653
4654 }
4655
4656 static int
4657 get_sra_super_block(int fd, struct intel_super **super_list, char *devname, int *max, int keep_fd)
4658 {
4659         struct mdinfo *sra;
4660         char *devnm;
4661         struct mdinfo *sd;
4662         int err = 0;
4663         int i = 0;
4664         sra = sysfs_read(fd, NULL, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
4665         if (!sra)
4666                 return 1;
4667
4668         if (sra->array.major_version != -1 ||
4669             sra->array.minor_version != -2 ||
4670             strcmp(sra->text_version, "imsm") != 0) {
4671                 err = 1;
4672                 goto error;
4673         }
4674         /* load all mpbs */
4675         devnm = fd2devnm(fd);
4676         for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
4677                 if (get_super_block(super_list, devnm, devname,
4678                                     sd->disk.major, sd->disk.minor, keep_fd) != 0) {
4679                         err = 7;
4680                         goto error;
4681                 }
4682         }
4683  error:
4684         sysfs_free(sra);
4685         *max = i;
4686         return err;
4687 }
4688
4689 static int load_container_imsm(struct supertype *st, int fd, char *devname)
4690 {
4691         return load_super_imsm_all(st, fd, &st->sb, devname, NULL, 1);
4692 }
4693 #endif
4694
4695 static int load_super_imsm(struct supertype *st, int fd, char *devname)
4696 {
4697         struct intel_super *super;
4698         int rv;
4699         int retry;
4700
4701         if (test_partition(fd))
4702                 /* IMSM not allowed on partitions */
4703                 return 1;
4704
4705         free_super_imsm(st);
4706
4707         super = alloc_super();
4708         get_dev_sector_size(fd, NULL, &super->sector_size);
4709         /* Load hba and capabilities if they exist.
4710          * But do not preclude loading metadata in case capabilities or hba are
4711          * non-compliant and ignore_hw_compat is set.
4712          */
4713         rv = find_intel_hba_capability(fd, super, devname);
4714         /* no orom/efi or non-intel hba of the disk */
4715         if (rv != 0 && st->ignore_hw_compat == 0) {
4716                 if (devname)
4717                         pr_err("No OROM/EFI properties for %s\n", devname);
4718                 free_imsm(super);
4719                 return 2;
4720         }
4721         rv = load_and_parse_mpb(fd, super, devname, 0);
4722
4723         /* retry the load if we might have raced against mdmon */
4724         if (rv == 3) {
4725                 struct mdstat_ent *mdstat = NULL;
4726                 char *name = fd2kname(fd);
4727
4728                 if (name)
4729                         mdstat = mdstat_by_component(name);
4730
4731                 if (mdstat && mdmon_running(mdstat->devnm) && getpid() != mdmon_pid(mdstat->devnm)) {
4732                         for (retry = 0; retry < 3; retry++) {
4733                                 usleep(3000);
4734                                 rv = load_and_parse_mpb(fd, super, devname, 0);
4735                                 if (rv != 3)
4736                                         break;
4737                         }
4738                 }
4739
4740                 free_mdstat(mdstat);
4741         }
4742
4743         if (rv) {
4744                 if (devname)
4745                         pr_err("Failed to load all information sections on %s\n", devname);
4746                 free_imsm(super);
4747                 return rv;
4748         }
4749
4750         st->sb = super;
4751         if (st->ss == NULL) {
4752                 st->ss = &super_imsm;
4753                 st->minor_version = 0;
4754                 st->max_devs = IMSM_MAX_DEVICES;
4755         }
4756
4757         /* load migration record */
4758         if (load_imsm_migr_rec(super, NULL) == 0) {
4759                 /* Check for unsupported migration features */
4760                 if (check_mpb_migr_compatibility(super) != 0) {
4761                         pr_err("Unsupported migration detected");
4762                         if (devname)
4763                                 fprintf(stderr, " on %s\n", devname);
4764                         else
4765                                 fprintf(stderr, " (IMSM).\n");
4766                         return 3;
4767                 }
4768         }
4769
4770         return 0;
4771 }
4772
4773 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
4774 {
4775         if (info->level == 1)
4776                 return 128;
4777         return info->chunk_size >> 9;
4778 }
4779
4780 static unsigned long long info_to_blocks_per_member(mdu_array_info_t *info,
4781                                                     unsigned long long size)
4782 {
4783         if (info->level == 1)
4784                 return size * 2;
4785         else
4786                 return (size * 2) & ~(info_to_blocks_per_strip(info) - 1);
4787 }
4788
4789 static void imsm_update_version_info(struct intel_super *super)
4790 {
4791         /* update the version and attributes */
4792         struct imsm_super *mpb = super->anchor;
4793         char *version;
4794         struct imsm_dev *dev;
4795         struct imsm_map *map;
4796         int i;
4797
4798         for (i = 0; i < mpb->num_raid_devs; i++) {
4799                 dev = get_imsm_dev(super, i);
4800                 map = get_imsm_map(dev, MAP_0);
4801                 if (__le32_to_cpu(dev->size_high) > 0)
4802                         mpb->attributes |= MPB_ATTRIB_2TB;
4803
4804                 /* FIXME detect when an array spans a port multiplier */
4805                 #if 0
4806                 mpb->attributes |= MPB_ATTRIB_PM;
4807                 #endif
4808
4809                 if (mpb->num_raid_devs > 1 ||
4810                     mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
4811                         version = MPB_VERSION_ATTRIBS;
4812                         switch (get_imsm_raid_level(map)) {
4813                         case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
4814                         case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
4815                         case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
4816                         case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
4817                         }
4818                 } else {
4819                         if (map->num_members >= 5)
4820                                 version = MPB_VERSION_5OR6_DISK_ARRAY;
4821                         else if (dev->status == DEV_CLONE_N_GO)
4822                                 version = MPB_VERSION_CNG;
4823                         else if (get_imsm_raid_level(map) == 5)
4824                                 version = MPB_VERSION_RAID5;
4825                         else if (map->num_members >= 3)
4826                                 version = MPB_VERSION_3OR4_DISK_ARRAY;
4827                         else if (get_imsm_raid_level(map) == 1)
4828                                 version = MPB_VERSION_RAID1;
4829                         else
4830                                 version = MPB_VERSION_RAID0;
4831                 }
4832                 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
4833         }
4834 }
4835
4836 static int check_name(struct intel_super *super, char *name, int quiet)
4837 {
4838         struct imsm_super *mpb = super->anchor;
4839         char *reason = NULL;
4840         int i;
4841
4842         if (strlen(name) > MAX_RAID_SERIAL_LEN)
4843                 reason = "must be 16 characters or less";
4844
4845         for (i = 0; i < mpb->num_raid_devs; i++) {
4846                 struct imsm_dev *dev = get_imsm_dev(super, i);
4847
4848                 if (strncmp((char *) dev->volume, name, MAX_RAID_SERIAL_LEN) == 0) {
4849                         reason = "already exists";
4850                         break;
4851                 }
4852         }
4853
4854         if (reason && !quiet)
4855                 pr_err("imsm volume name %s\n", reason);
4856
4857         return !reason;
4858 }
4859
4860 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
4861                                   unsigned long long size, char *name,
4862                                   char *homehost, int *uuid,
4863                                   long long data_offset)
4864 {
4865         /* We are creating a volume inside a pre-existing container.
4866          * so st->sb is already set.
4867          */
4868         struct intel_super *super = st->sb;
4869         unsigned int sector_size = super->sector_size;
4870         struct imsm_super *mpb = super->anchor;
4871         struct intel_dev *dv;
4872         struct imsm_dev *dev;
4873         struct imsm_vol *vol;
4874         struct imsm_map *map;
4875         int idx = mpb->num_raid_devs;
4876         int i;
4877         unsigned long long array_blocks;
4878         size_t size_old, size_new;
4879         unsigned long long num_data_stripes;
4880
4881         if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
4882                 pr_err("This imsm-container already has the maximum of %d volumes\n", super->orom->vpa);
4883                 return 0;
4884         }
4885
4886         /* ensure the mpb is large enough for the new data */
4887         size_old = __le32_to_cpu(mpb->mpb_size);
4888         size_new = disks_to_mpb_size(info->nr_disks);
4889         if (size_new > size_old) {
4890                 void *mpb_new;
4891                 size_t size_round = ROUND_UP(size_new, sector_size);
4892
4893                 if (posix_memalign(&mpb_new, sector_size, size_round) != 0) {
4894                         pr_err("could not allocate new mpb\n");
4895                         return 0;
4896                 }
4897                 if (posix_memalign(&super->migr_rec_buf, sector_size,
4898                                    MIGR_REC_BUF_SECTORS*sector_size) != 0) {
4899                         pr_err("could not allocate migr_rec buffer\n");
4900                         free(super->buf);
4901                         free(super);
4902                         free(mpb_new);
4903                         return 0;
4904                 }
4905                 memcpy(mpb_new, mpb, size_old);
4906                 free(mpb);
4907                 mpb = mpb_new;
4908                 super->anchor = mpb_new;
4909                 mpb->mpb_size = __cpu_to_le32(size_new);
4910                 memset(mpb_new + size_old, 0, size_round - size_old);
4911         }
4912         super->current_vol = idx;
4913
4914         /* handle 'failed_disks' by either:
4915          * a) create dummy disk entries in the table if this the first
4916          *    volume in the array.  We add them here as this is the only
4917          *    opportunity to add them. add_to_super_imsm_volume()
4918          *    handles the non-failed disks and continues incrementing
4919          *    mpb->num_disks.
4920          * b) validate that 'failed_disks' matches the current number
4921          *    of missing disks if the container is populated
4922          */
4923         if (super->current_vol == 0) {
4924                 mpb->num_disks = 0;
4925                 for (i = 0; i < info->failed_disks; i++) {
4926                         struct imsm_disk *disk;
4927
4928                         mpb->num_disks++;
4929                         disk = __get_imsm_disk(mpb, i);
4930                         disk->status = CONFIGURED_DISK | FAILED_DISK;
4931                         disk->scsi_id = __cpu_to_le32(~(__u32)0);
4932                         snprintf((char *) disk->serial, MAX_RAID_SERIAL_LEN,
4933                                  "missing:%d", i);
4934                 }
4935                 find_missing(super);
4936         } else {
4937                 int missing = 0;
4938                 struct dl *d;
4939
4940                 for (d = super->missing; d; d = d->next)
4941                         missing++;
4942                 if (info->failed_disks > missing) {
4943                         pr_err("unable to add 'missing' disk to container\n");
4944                         return 0;
4945                 }
4946         }
4947
4948         if (!check_name(super, name, 0))
4949                 return 0;
4950         dv = xmalloc(sizeof(*dv));
4951         dev = xcalloc(1, sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
4952         strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
4953         array_blocks = calc_array_size(info->level, info->raid_disks,
4954                                                info->layout, info->chunk_size,
4955                                                size * 2);
4956         /* round array size down to closest MB */
4957         array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
4958
4959         dev->size_low = __cpu_to_le32((__u32) array_blocks);
4960         dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
4961         dev->status = (DEV_READ_COALESCING | DEV_WRITE_COALESCING);
4962         vol = &dev->vol;
4963         vol->migr_state = 0;
4964         set_migr_type(dev, MIGR_INIT);
4965         vol->dirty = !info->state;
4966         vol->curr_migr_unit = 0;
4967         map = get_imsm_map(dev, MAP_0);
4968         set_pba_of_lba0(map, super->create_offset);
4969         set_blocks_per_member(map, info_to_blocks_per_member(info, size));
4970         map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
4971         map->failed_disk_num = ~0;
4972         if (info->level > 0)
4973                 map->map_state = (info->state ? IMSM_T_STATE_NORMAL
4974                                   : IMSM_T_STATE_UNINITIALIZED);
4975         else
4976                 map->map_state = info->failed_disks ? IMSM_T_STATE_FAILED :
4977                                                       IMSM_T_STATE_NORMAL;
4978         map->ddf = 1;
4979
4980         if (info->level == 1 && info->raid_disks > 2) {
4981                 free(dev);
4982                 free(dv);
4983                 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
4984                 return 0;
4985         }
4986
4987         map->raid_level = info->level;
4988         if (info->level == 10) {
4989                 map->raid_level = 1;
4990                 map->num_domains = info->raid_disks / 2;
4991         } else if (info->level == 1)
4992                 map->num_domains = info->raid_disks;
4993         else
4994                 map->num_domains = 1;
4995
4996         /* info->size is only int so use the 'size' parameter instead */
4997         num_data_stripes = (size * 2) / info_to_blocks_per_strip(info);
4998         num_data_stripes /= map->num_domains;
4999         set_num_data_stripes(map, num_data_stripes);
5000
5001         map->num_members = info->raid_disks;
5002         for (i = 0; i < map->num_members; i++) {
5003                 /* initialized in add_to_super */
5004                 set_imsm_ord_tbl_ent(map, i, IMSM_ORD_REBUILD);
5005         }
5006         mpb->num_raid_devs++;
5007
5008         dv->dev = dev;
5009         dv->index = super->current_vol;
5010         dv->next = super->devlist;
5011         super->devlist = dv;
5012
5013         imsm_update_version_info(super);
5014
5015         return 1;
5016 }
5017
5018 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
5019                            unsigned long long size, char *name,
5020                            char *homehost, int *uuid,
5021                            unsigned long long data_offset)
5022 {
5023         /* This is primarily called by Create when creating a new array.
5024          * We will then get add_to_super called for each component, and then
5025          * write_init_super called to write it out to each device.
5026          * For IMSM, Create can create on fresh devices or on a pre-existing
5027          * array.
5028          * To create on a pre-existing array a different method will be called.
5029          * This one is just for fresh drives.
5030          */
5031         struct intel_super *super;
5032         struct imsm_super *mpb;
5033         size_t mpb_size;
5034         char *version;
5035
5036         if (data_offset != INVALID_SECTORS) {
5037                 pr_err("data-offset not supported by imsm\n");
5038                 return 0;
5039         }
5040
5041         if (st->sb)
5042                 return init_super_imsm_volume(st, info, size, name, homehost, uuid,
5043                                               data_offset);
5044
5045         if (info)
5046                 mpb_size = disks_to_mpb_size(info->nr_disks);
5047         else
5048                 mpb_size = MAX_SECTOR_SIZE;
5049
5050         super = alloc_super();
5051         if (super &&
5052             posix_memalign(&super->buf, MAX_SECTOR_SIZE, mpb_size) != 0) {
5053                 free(super);
5054                 super = NULL;
5055         }
5056         if (!super) {
5057                 pr_err("could not allocate superblock\n");
5058                 return 0;
5059         }
5060         if (posix_memalign(&super->migr_rec_buf, MAX_SECTOR_SIZE,
5061             MIGR_REC_BUF_SECTORS*MAX_SECTOR_SIZE) != 0) {
5062                 pr_err("could not allocate migr_rec buffer\n");
5063                 free(super->buf);
5064                 free(super);
5065                 return 0;
5066         }
5067         memset(super->buf, 0, mpb_size);
5068         mpb = super->buf;
5069         mpb->mpb_size = __cpu_to_le32(mpb_size);
5070         st->sb = super;
5071
5072         if (info == NULL) {
5073                 /* zeroing superblock */
5074                 return 0;
5075         }
5076
5077         mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
5078
5079         version = (char *) mpb->sig;
5080         strcpy(version, MPB_SIGNATURE);
5081         version += strlen(MPB_SIGNATURE);
5082         strcpy(version, MPB_VERSION_RAID0);
5083
5084         return 1;
5085 }
5086
5087 #ifndef MDASSEMBLE
5088 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
5089                                      int fd, char *devname)
5090 {
5091         struct intel_super *super = st->sb;
5092         struct imsm_super *mpb = super->anchor;
5093         struct imsm_disk *_disk;
5094         struct imsm_dev *dev;
5095         struct imsm_map *map;
5096         struct dl *dl, *df;
5097         int slot;
5098
5099         dev = get_imsm_dev(super, super->current_vol);
5100         map = get_imsm_map(dev, MAP_0);
5101
5102         if (! (dk->state & (1<<MD_DISK_SYNC))) {
5103                 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5104                         devname);
5105                 return 1;
5106         }
5107
5108         if (fd == -1) {
5109                 /* we're doing autolayout so grab the pre-marked (in
5110                  * validate_geometry) raid_disk
5111                  */
5112                 for (dl = super->disks; dl; dl = dl->next)
5113                         if (dl->raiddisk == dk->raid_disk)
5114                                 break;
5115         } else {
5116                 for (dl = super->disks; dl ; dl = dl->next)
5117                         if (dl->major == dk->major &&
5118                             dl->minor == dk->minor)
5119                                 break;
5120         }
5121
5122         if (!dl) {
5123                 pr_err("%s is not a member of the same container\n", devname);
5124                 return 1;
5125         }
5126
5127         /* add a pristine spare to the metadata */
5128         if (dl->index < 0) {
5129                 dl->index = super->anchor->num_disks;
5130                 super->anchor->num_disks++;
5131         }
5132         /* Check the device has not already been added */
5133         slot = get_imsm_disk_slot(map, dl->index);
5134         if (slot >= 0 &&
5135             (get_imsm_ord_tbl_ent(dev, slot, MAP_X) & IMSM_ORD_REBUILD) == 0) {
5136                 pr_err("%s has been included in this array twice\n",
5137                         devname);
5138                 return 1;
5139         }
5140         set_imsm_ord_tbl_ent(map, dk->raid_disk, dl->index);
5141         dl->disk.status = CONFIGURED_DISK;
5142
5143         /* update size of 'missing' disks to be at least as large as the
5144          * largest acitve member (we only have dummy missing disks when
5145          * creating the first volume)
5146          */
5147         if (super->current_vol == 0) {
5148                 for (df = super->missing; df; df = df->next) {
5149                         if (total_blocks(&dl->disk) > total_blocks(&df->disk))
5150                                 set_total_blocks(&df->disk, total_blocks(&dl->disk));
5151                         _disk = __get_imsm_disk(mpb, df->index);
5152                         *_disk = df->disk;
5153                 }
5154         }
5155
5156         /* refresh unset/failed slots to point to valid 'missing' entries */
5157         for (df = super->missing; df; df = df->next)
5158                 for (slot = 0; slot < mpb->num_disks; slot++) {
5159                         __u32 ord = get_imsm_ord_tbl_ent(dev, slot, MAP_X);
5160
5161                         if ((ord & IMSM_ORD_REBUILD) == 0)
5162                                 continue;
5163                         set_imsm_ord_tbl_ent(map, slot, df->index | IMSM_ORD_REBUILD);
5164                         if (is_gen_migration(dev)) {
5165                                 struct imsm_map *map2 = get_imsm_map(dev,
5166                                                                      MAP_1);
5167                                 int slot2 = get_imsm_disk_slot(map2, df->index);
5168                                 if (slot2 < map2->num_members && slot2 >= 0) {
5169                                         __u32 ord2 = get_imsm_ord_tbl_ent(dev,
5170                                                                          slot2,
5171                                                                          MAP_1);
5172                                         if ((unsigned)df->index ==
5173                                                                ord_to_idx(ord2))
5174                                                 set_imsm_ord_tbl_ent(map2,
5175                                                         slot2,
5176                                                         df->index |
5177                                                         IMSM_ORD_REBUILD);
5178                                 }
5179                         }
5180                         dprintf("set slot:%d to missing disk:%d\n", slot, df->index);
5181                         break;
5182                 }
5183
5184         /* if we are creating the first raid device update the family number */
5185         if (super->current_vol == 0) {
5186                 __u32 sum;
5187                 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
5188
5189                 _disk = __get_imsm_disk(mpb, dl->index);
5190                 if (!_dev || !_disk) {
5191                         pr_err("BUG mpb setup error\n");
5192                         return 1;
5193                 }
5194                 *_dev = *dev;
5195                 *_disk = dl->disk;
5196                 sum = random32();
5197                 sum += __gen_imsm_checksum(mpb);
5198                 mpb->family_num = __cpu_to_le32(sum);
5199                 mpb->orig_family_num = mpb->family_num;
5200         }
5201         super->current_disk = dl;
5202         return 0;
5203 }
5204
5205 /* mark_spare()
5206  *   Function marks disk as spare and restores disk serial
5207  *   in case it was previously marked as failed by takeover operation
5208  * reruns:
5209  *   -1 : critical error
5210  *    0 : disk is marked as spare but serial is not set
5211  *    1 : success
5212  */
5213 int mark_spare(struct dl *disk)
5214 {
5215         __u8 serial[MAX_RAID_SERIAL_LEN];
5216         int ret_val = -1;
5217
5218         if (!disk)
5219                 return ret_val;
5220
5221         ret_val = 0;
5222         if (!imsm_read_serial(disk->fd, NULL, serial)) {
5223                 /* Restore disk serial number, because takeover marks disk
5224                  * as failed and adds to serial ':0' before it becomes
5225                  * a spare disk.
5226                  */
5227                 serialcpy(disk->serial, serial);
5228                 serialcpy(disk->disk.serial, serial);
5229                 ret_val = 1;
5230         }
5231         disk->disk.status = SPARE_DISK;
5232         disk->index = -1;
5233
5234         return ret_val;
5235 }
5236
5237 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
5238                              int fd, char *devname,
5239                              unsigned long long data_offset)
5240 {
5241         struct intel_super *super = st->sb;
5242         struct dl *dd;
5243         unsigned long long size;
5244         unsigned int member_sector_size;
5245         __u32 id;
5246         int rv;
5247         struct stat stb;
5248
5249         /* If we are on an RAID enabled platform check that the disk is
5250          * attached to the raid controller.
5251          * We do not need to test disks attachment for container based additions,
5252          * they shall be already tested when container was created/assembled.
5253          */
5254         rv = find_intel_hba_capability(fd, super, devname);
5255         /* no orom/efi or non-intel hba of the disk */
5256         if (rv != 0) {
5257                 dprintf("capability: %p fd: %d ret: %d\n",
5258                         super->orom, fd, rv);
5259                 return 1;
5260         }
5261
5262         if (super->current_vol >= 0)
5263                 return add_to_super_imsm_volume(st, dk, fd, devname);
5264
5265         fstat(fd, &stb);
5266         dd = xcalloc(sizeof(*dd), 1);
5267         dd->major = major(stb.st_rdev);
5268         dd->minor = minor(stb.st_rdev);
5269         dd->devname = devname ? xstrdup(devname) : NULL;
5270         dd->fd = fd;
5271         dd->e = NULL;
5272         dd->action = DISK_ADD;
5273         rv = imsm_read_serial(fd, devname, dd->serial);
5274         if (rv) {
5275                 pr_err("failed to retrieve scsi serial, aborting\n");
5276                 if (dd->devname)
5277                         free(dd->devname);
5278                 free(dd);
5279                 abort();
5280         }
5281         if (super->hba && ((super->hba->type == SYS_DEV_NVME) ||
5282            (super->hba->type == SYS_DEV_VMD))) {
5283                 int i;
5284                 char *devpath = diskfd_to_devpath(fd);
5285                 char controller_path[PATH_MAX];
5286
5287                 if (!devpath) {
5288                         pr_err("failed to get devpath, aborting\n");
5289                         if (dd->devname)
5290                                 free(dd->devname);
5291                         free(dd);
5292                         return 1;
5293                 }
5294
5295                 snprintf(controller_path, PATH_MAX-1, "%s/device", devpath);
5296                 free(devpath);
5297
5298                 if (devpath_to_vendor(controller_path) == 0x8086) {
5299                         /*
5300                          * If Intel's NVMe drive has serial ended with
5301                          * "-A","-B","-1" or "-2" it means that this is "x8"
5302                          * device (double drive on single PCIe card).
5303                          * User should be warned about potential data loss.
5304                          */
5305                         for (i = MAX_RAID_SERIAL_LEN-1; i > 0; i--) {
5306                                 /* Skip empty character at the end */
5307                                 if (dd->serial[i] == 0)
5308                                         continue;
5309
5310                                 if (((dd->serial[i] == 'A') ||
5311                                    (dd->serial[i] == 'B') ||
5312                                    (dd->serial[i] == '1') ||
5313                                    (dd->serial[i] == '2')) &&
5314                                    (dd->serial[i-1] == '-'))
5315                                         pr_err("\tThe action you are about to take may put your data at risk.\n"
5316                                                 "\tPlease note that x8 devices may consist of two separate x4 devices "
5317                                                 "located on a single PCIe port.\n"
5318                                                 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5319                                 break;
5320                         }
5321                 }
5322         }
5323
5324         get_dev_size(fd, NULL, &size);
5325         get_dev_sector_size(fd, NULL, &member_sector_size);
5326
5327         if (super->sector_size == 0) {
5328                 /* this a first device, so sector_size is not set yet */
5329                 super->sector_size = member_sector_size;
5330         } else if (member_sector_size != super->sector_size) {
5331                 pr_err("Mixing between different sector size is forbidden, aborting...\n");
5332                 if (dd->devname)
5333                         free(dd->devname);
5334                 free(dd);
5335                 return 1;
5336         }
5337
5338         /* clear migr_rec when adding disk to container */
5339         memset(super->migr_rec_buf, 0, MIGR_REC_BUF_SECTORS*super->sector_size);
5340         if (lseek64(fd, size - MIGR_REC_SECTOR_POSITION*super->sector_size,
5341             SEEK_SET) >= 0) {
5342                 if (write(fd, super->migr_rec_buf,
5343                     MIGR_REC_BUF_SECTORS*super->sector_size) !=
5344                     MIGR_REC_BUF_SECTORS*super->sector_size)
5345                         perror("Write migr_rec failed");
5346         }
5347
5348         size /= 512;
5349         serialcpy(dd->disk.serial, dd->serial);
5350         set_total_blocks(&dd->disk, size);
5351         if (__le32_to_cpu(dd->disk.total_blocks_hi) > 0) {
5352                 struct imsm_super *mpb = super->anchor;
5353                 mpb->attributes |= MPB_ATTRIB_2TB_DISK;
5354         }
5355         mark_spare(dd);
5356         if (sysfs_disk_to_scsi_id(fd, &id) == 0)
5357                 dd->disk.scsi_id = __cpu_to_le32(id);
5358         else
5359                 dd->disk.scsi_id = __cpu_to_le32(0);
5360
5361         if (st->update_tail) {
5362                 dd->next = super->disk_mgmt_list;
5363                 super->disk_mgmt_list = dd;
5364         } else {
5365                 dd->next = super->disks;
5366                 super->disks = dd;
5367                 super->updates_pending++;
5368         }
5369
5370         return 0;
5371 }
5372
5373 static int remove_from_super_imsm(struct supertype *st, mdu_disk_info_t *dk)
5374 {
5375         struct intel_super *super = st->sb;
5376         struct dl *dd;
5377
5378         /* remove from super works only in mdmon - for communication
5379          * manager - monitor. Check if communication memory buffer
5380          * is prepared.
5381          */
5382         if (!st->update_tail) {
5383                 pr_err("shall be used in mdmon context only\n");
5384                 return 1;
5385         }
5386         dd = xcalloc(1, sizeof(*dd));
5387         dd->major = dk->major;
5388         dd->minor = dk->minor;
5389         dd->fd = -1;
5390         mark_spare(dd);
5391         dd->action = DISK_REMOVE;
5392
5393         dd->next = super->disk_mgmt_list;
5394         super->disk_mgmt_list = dd;
5395
5396         return 0;
5397 }
5398
5399 static int store_imsm_mpb(int fd, struct imsm_super *mpb);
5400
5401 static union {
5402         char buf[MAX_SECTOR_SIZE];
5403         struct imsm_super anchor;
5404 } spare_record __attribute__ ((aligned(MAX_SECTOR_SIZE)));
5405
5406 /* spare records have their own family number and do not have any defined raid
5407  * devices
5408  */
5409 static int write_super_imsm_spares(struct intel_super *super, int doclose)
5410 {
5411         struct imsm_super *mpb = super->anchor;
5412         struct imsm_super *spare = &spare_record.anchor;
5413         __u32 sum;
5414         struct dl *d;
5415
5416         spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super));
5417         spare->generation_num = __cpu_to_le32(1UL);
5418         spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
5419         spare->num_disks = 1;
5420         spare->num_raid_devs = 0;
5421         spare->cache_size = mpb->cache_size;
5422         spare->pwr_cycle_count = __cpu_to_le32(1);
5423
5424         snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
5425                  MPB_SIGNATURE MPB_VERSION_RAID0);
5426
5427         for (d = super->disks; d; d = d->next) {
5428                 if (d->index != -1)
5429                         continue;
5430
5431                 spare->disk[0] = d->disk;
5432                 if (__le32_to_cpu(d->disk.total_blocks_hi) > 0)
5433                         spare->attributes |= MPB_ATTRIB_2TB_DISK;
5434
5435                 if (super->sector_size == 4096)
5436                         convert_to_4k_imsm_disk(&spare->disk[0]);
5437
5438                 sum = __gen_imsm_checksum(spare);
5439                 spare->family_num = __cpu_to_le32(sum);
5440                 spare->orig_family_num = 0;
5441                 sum = __gen_imsm_checksum(spare);
5442                 spare->check_sum = __cpu_to_le32(sum);
5443
5444                 if (store_imsm_mpb(d->fd, spare)) {
5445                         pr_err("failed for device %d:%d %s\n",
5446                                 d->major, d->minor, strerror(errno));
5447                         return 1;
5448                 }
5449                 if (doclose) {
5450                         close(d->fd);
5451                         d->fd = -1;
5452                 }
5453         }
5454
5455         return 0;
5456 }
5457
5458 static int write_super_imsm(struct supertype *st, int doclose)
5459 {
5460         struct intel_super *super = st->sb;
5461         unsigned int sector_size = super->sector_size;
5462         struct imsm_super *mpb = super->anchor;
5463         struct dl *d;
5464         __u32 generation;
5465         __u32 sum;
5466         int spares = 0;
5467         int i;
5468         __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
5469         int num_disks = 0;
5470         int clear_migration_record = 1;
5471
5472         /* 'generation' is incremented everytime the metadata is written */
5473         generation = __le32_to_cpu(mpb->generation_num);
5474         generation++;
5475         mpb->generation_num = __cpu_to_le32(generation);
5476
5477         /* fix up cases where previous mdadm releases failed to set
5478          * orig_family_num
5479          */
5480         if (mpb->orig_family_num == 0)
5481                 mpb->orig_family_num = mpb->family_num;
5482
5483         for (d = super->disks; d; d = d->next) {
5484                 if (d->index == -1)
5485                         spares++;
5486                 else {
5487                         mpb->disk[d->index] = d->disk;
5488                         num_disks++;
5489                 }
5490         }
5491         for (d = super->missing; d; d = d->next) {
5492                 mpb->disk[d->index] = d->disk;
5493                 num_disks++;
5494         }
5495         mpb->num_disks = num_disks;
5496         mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
5497
5498         for (i = 0; i < mpb->num_raid_devs; i++) {
5499                 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
5500                 struct imsm_dev *dev2 = get_imsm_dev(super, i);
5501                 if (dev && dev2) {
5502                         imsm_copy_dev(dev, dev2);
5503                         mpb_size += sizeof_imsm_dev(dev, 0);
5504                 }
5505                 if (is_gen_migration(dev2))
5506                         clear_migration_record = 0;
5507         }
5508         mpb_size += __le32_to_cpu(mpb->bbm_log_size);
5509         mpb->mpb_size = __cpu_to_le32(mpb_size);
5510
5511         /* recalculate checksum */
5512         sum = __gen_imsm_checksum(mpb);
5513         mpb->check_sum = __cpu_to_le32(sum);
5514
5515         if (super->clean_migration_record_by_mdmon) {
5516                 clear_migration_record = 1;
5517                 super->clean_migration_record_by_mdmon = 0;
5518         }
5519         if (clear_migration_record)
5520                 memset(super->migr_rec_buf, 0,
5521                     MIGR_REC_BUF_SECTORS*sector_size);
5522
5523         if (sector_size == 4096)
5524                 convert_to_4k(super);
5525
5526         /* write the mpb for disks that compose raid devices */
5527         for (d = super->disks; d ; d = d->next) {
5528                 if (d->index < 0 || is_failed(&d->disk))
5529                         continue;
5530
5531                 if (clear_migration_record) {
5532                         unsigned long long dsize;
5533
5534                         get_dev_size(d->fd, NULL, &dsize);
5535                         if (lseek64(d->fd, dsize - sector_size,
5536                             SEEK_SET) >= 0) {
5537                                 if (write(d->fd, super->migr_rec_buf,
5538                                     MIGR_REC_BUF_SECTORS*sector_size) !=
5539                                     MIGR_REC_BUF_SECTORS*sector_size)
5540                                         perror("Write migr_rec failed");
5541                         }
5542                 }
5543
5544                 if (store_imsm_mpb(d->fd, mpb))
5545                         fprintf(stderr,
5546                                 "failed for device %d:%d (fd: %d)%s\n",
5547                                 d->major, d->minor,
5548                                 d->fd, strerror(errno));
5549
5550                 if (doclose) {
5551                         close(d->fd);
5552                         d->fd = -1;
5553                 }
5554         }
5555
5556         if (spares)
5557                 return write_super_imsm_spares(super, doclose);
5558
5559         return 0;
5560 }
5561
5562 static int create_array(struct supertype *st, int dev_idx)
5563 {
5564         size_t len;
5565         struct imsm_update_create_array *u;
5566         struct intel_super *super = st->sb;
5567         struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
5568         struct imsm_map *map = get_imsm_map(dev, MAP_0);
5569         struct disk_info *inf;
5570         struct imsm_disk *disk;
5571         int i;
5572
5573         len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
5574               sizeof(*inf) * map->num_members;
5575         u = xmalloc(len);
5576         u->type = update_create_array;
5577         u->dev_idx = dev_idx;
5578         imsm_copy_dev(&u->dev, dev);
5579         inf = get_disk_info(u);
5580         for (i = 0; i < map->num_members; i++) {
5581                 int idx = get_imsm_disk_idx(dev, i, MAP_X);
5582
5583                 disk = get_imsm_disk(super, idx);
5584                 if (!disk)
5585                         disk = get_imsm_missing(super, idx);
5586                 serialcpy(inf[i].serial, disk->serial);
5587         }
5588         append_metadata_update(st, u, len);
5589
5590         return 0;
5591 }
5592
5593 static int mgmt_disk(struct supertype *st)
5594 {
5595         struct intel_super *super = st->sb;
5596         size_t len;
5597         struct imsm_update_add_remove_disk *u;
5598
5599         if (!super->disk_mgmt_list)
5600                 return 0;
5601
5602         len = sizeof(*u);
5603         u = xmalloc(len);
5604         u->type = update_add_remove_disk;
5605         append_metadata_update(st, u, len);
5606
5607         return 0;
5608 }
5609
5610 static int write_init_super_imsm(struct supertype *st)
5611 {
5612         struct intel_super *super = st->sb;
5613         int current_vol = super->current_vol;
5614
5615         /* we are done with current_vol reset it to point st at the container */
5616         super->current_vol = -1;
5617
5618         if (st->update_tail) {
5619                 /* queue the recently created array / added disk
5620                  * as a metadata update */
5621                 int rv;
5622
5623                 /* determine if we are creating a volume or adding a disk */
5624                 if (current_vol < 0) {
5625                         /* in the mgmt (add/remove) disk case we are running
5626                          * in mdmon context, so don't close fd's
5627                          */
5628                         return mgmt_disk(st);
5629                 } else
5630                         rv = create_array(st, current_vol);
5631
5632                 return rv;
5633         } else {
5634                 struct dl *d;
5635                 for (d = super->disks; d; d = d->next)
5636                         Kill(d->devname, NULL, 0, -1, 1);
5637                 return write_super_imsm(st, 1);
5638         }
5639 }
5640 #endif
5641
5642 static int store_super_imsm(struct supertype *st, int fd)
5643 {
5644         struct intel_super *super = st->sb;
5645         struct imsm_super *mpb = super ? super->anchor : NULL;
5646
5647         if (!mpb)
5648                 return 1;
5649
5650 #ifndef MDASSEMBLE
5651         if (super->sector_size == 4096)
5652                 convert_to_4k(super);
5653         return store_imsm_mpb(fd, mpb);
5654 #else
5655         return 1;
5656 #endif
5657 }
5658
5659 static int imsm_bbm_log_size(struct imsm_super *mpb)
5660 {
5661         return __le32_to_cpu(mpb->bbm_log_size);
5662 }
5663
5664 #ifndef MDASSEMBLE
5665 static int validate_geometry_imsm_container(struct supertype *st, int level,
5666                                             int layout, int raiddisks, int chunk,
5667                                             unsigned long long size,
5668                                             unsigned long long data_offset,
5669                                             char *dev,
5670                                             unsigned long long *freesize,
5671                                             int verbose)
5672 {
5673         int fd;
5674         unsigned long long ldsize;
5675         struct intel_super *super;
5676         int rv = 0;
5677
5678         if (level != LEVEL_CONTAINER)
5679                 return 0;
5680         if (!dev)
5681                 return 1;
5682
5683         fd = open(dev, O_RDONLY|O_EXCL, 0);
5684         if (fd < 0) {
5685                 if (verbose > 0)
5686                         pr_err("imsm: Cannot open %s: %s\n",
5687                                 dev, strerror(errno));
5688                 return 0;
5689         }
5690         if (!get_dev_size(fd, dev, &ldsize)) {
5691                 close(fd);
5692                 return 0;
5693         }
5694
5695         /* capabilities retrieve could be possible
5696          * note that there is no fd for the disks in array.
5697          */
5698         super = alloc_super();
5699         if (!get_dev_sector_size(fd, NULL, &super->sector_size)) {
5700                 close(fd);
5701                 free_imsm(super);
5702                 return 0;
5703         }
5704
5705         rv = find_intel_hba_capability(fd, super, verbose > 0 ? dev : NULL);
5706         if (rv != 0) {
5707 #if DEBUG
5708                 char str[256];
5709                 fd2devname(fd, str);
5710                 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
5711                         fd, str, super->orom, rv, raiddisks);
5712 #endif
5713                 /* no orom/efi or non-intel hba of the disk */
5714                 close(fd);
5715                 free_imsm(super);
5716                 return 0;
5717         }
5718         close(fd);
5719         if (super->orom) {
5720                 if (raiddisks > super->orom->tds) {
5721                         if (verbose)
5722                                 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
5723                                         raiddisks, super->orom->tds);
5724                         free_imsm(super);
5725                         return 0;
5726                 }
5727                 if ((super->orom->attr & IMSM_OROM_ATTR_2TB_DISK) == 0 &&
5728                     (ldsize >> 9) >> 32 > 0) {
5729                         if (verbose)
5730                                 pr_err("%s exceeds maximum platform supported size\n", dev);
5731                         free_imsm(super);
5732                         return 0;
5733                 }
5734         }
5735
5736         *freesize = avail_size_imsm(st, ldsize >> 9, data_offset);
5737         free_imsm(super);
5738
5739         return 1;
5740 }
5741
5742 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
5743 {
5744         const unsigned long long base_start = e[*idx].start;
5745         unsigned long long end = base_start + e[*idx].size;
5746         int i;
5747
5748         if (base_start == end)
5749                 return 0;
5750
5751         *idx = *idx + 1;
5752         for (i = *idx; i < num_extents; i++) {
5753                 /* extend overlapping extents */
5754                 if (e[i].start >= base_start &&
5755                     e[i].start <= end) {
5756                         if (e[i].size == 0)
5757                                 return 0;
5758                         if (e[i].start + e[i].size > end)
5759                                 end = e[i].start + e[i].size;
5760                 } else if (e[i].start > end) {
5761                         *idx = i;
5762                         break;
5763                 }
5764         }
5765
5766         return end - base_start;
5767 }
5768
5769 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
5770 {
5771         /* build a composite disk with all known extents and generate a new
5772          * 'maxsize' given the "all disks in an array must share a common start
5773          * offset" constraint
5774          */
5775         struct extent *e = xcalloc(sum_extents, sizeof(*e));
5776         struct dl *dl;
5777         int i, j;
5778         int start_extent;
5779         unsigned long long pos;
5780         unsigned long long start = 0;
5781         unsigned long long maxsize;
5782         unsigned long reserve;
5783
5784         /* coalesce and sort all extents. also, check to see if we need to
5785          * reserve space between member arrays
5786          */
5787         j = 0;
5788         for (dl = super->disks; dl; dl = dl->next) {
5789                 if (!dl->e)
5790                         continue;
5791                 for (i = 0; i < dl->extent_cnt; i++)
5792                         e[j++] = dl->e[i];
5793         }
5794         qsort(e, sum_extents, sizeof(*e), cmp_extent);
5795
5796         /* merge extents */
5797         i = 0;
5798         j = 0;
5799         while (i < sum_extents) {
5800                 e[j].start = e[i].start;
5801                 e[j].size = find_size(e, &i, sum_extents);
5802                 j++;
5803                 if (e[j-1].size == 0)
5804                         break;
5805         }
5806
5807         pos = 0;
5808         maxsize = 0;
5809         start_extent = 0;
5810         i = 0;
5811         do {
5812                 unsigned long long esize;
5813
5814                 esize = e[i].start - pos;
5815                 if (esize >= maxsize) {
5816                         maxsize = esize;
5817                         start = pos;
5818                         start_extent = i;
5819                 }
5820                 pos = e[i].start + e[i].size;
5821                 i++;
5822         } while (e[i-1].size);
5823         free(e);
5824
5825         if (maxsize == 0)
5826                 return 0;
5827
5828         /* FIXME assumes volume at offset 0 is the first volume in a
5829          * container
5830          */
5831         if (start_extent > 0)
5832                 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
5833         else
5834                 reserve = 0;
5835
5836         if (maxsize < reserve)
5837                 return 0;
5838
5839         super->create_offset = ~((unsigned long long) 0);
5840         if (start + reserve > super->create_offset)
5841                 return 0; /* start overflows create_offset */
5842         super->create_offset = start + reserve;
5843
5844         return maxsize - reserve;
5845 }
5846
5847 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
5848 {
5849         if (level < 0 || level == 6 || level == 4)
5850                 return 0;
5851
5852         /* if we have an orom prevent invalid raid levels */
5853         if (orom)
5854                 switch (level) {
5855                 case 0: return imsm_orom_has_raid0(orom);
5856                 case 1:
5857                         if (raiddisks > 2)
5858                                 return imsm_orom_has_raid1e(orom);
5859                         return imsm_orom_has_raid1(orom) && raiddisks == 2;
5860                 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
5861                 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
5862                 }
5863         else
5864                 return 1; /* not on an Intel RAID platform so anything goes */
5865
5866         return 0;
5867 }
5868
5869 static int
5870 active_arrays_by_format(char *name, char* hba, struct md_list **devlist,
5871                         int dpa, int verbose)
5872 {
5873         struct mdstat_ent *mdstat = mdstat_read(0, 0);
5874         struct mdstat_ent *memb;
5875         int count = 0;
5876         int num = 0;
5877         struct md_list *dv;
5878         int found;
5879
5880         for (memb = mdstat ; memb ; memb = memb->next) {
5881                 if (memb->metadata_version &&
5882                     (strncmp(memb->metadata_version, "external:", 9) == 0)  &&
5883                     (strcmp(&memb->metadata_version[9], name) == 0) &&
5884                     !is_subarray(memb->metadata_version+9) &&
5885                     memb->members) {
5886                         struct dev_member *dev = memb->members;
5887                         int fd = -1;
5888                         while(dev && (fd < 0)) {
5889                                 char *path = xmalloc(strlen(dev->name) + strlen("/dev/") + 1);
5890                                 num = sprintf(path, "%s%s", "/dev/", dev->name);
5891                                 if (num > 0)
5892                                         fd = open(path, O_RDONLY, 0);
5893                                 if (num <= 0 || fd < 0) {
5894                                         pr_vrb("Cannot open %s: %s\n",
5895                                                dev->name, strerror(errno));
5896                                 }
5897                                 free(path);
5898                                 dev = dev->next;
5899                         }
5900                         found = 0;
5901                         if (fd >= 0 && disk_attached_to_hba(fd, hba)) {
5902                                 struct mdstat_ent *vol;
5903                                 for (vol = mdstat ; vol ; vol = vol->next) {
5904                                         if (vol->active > 0 &&
5905                                             vol->metadata_version &&
5906                                             is_container_member(vol, memb->devnm)) {
5907                                                 found++;
5908                                                 count++;
5909                                         }
5910                                 }
5911                                 if (*devlist && (found < dpa)) {
5912                                         dv = xcalloc(1, sizeof(*dv));
5913                                         dv->devname = xmalloc(strlen(memb->devnm) + strlen("/dev/") + 1);
5914                                         sprintf(dv->devname, "%s%s", "/dev/", memb->devnm);
5915                                         dv->found = found;
5916                                         dv->used = 0;
5917                                         dv->next = *devlist;
5918                                         *devlist = dv;
5919                                 }
5920                         }
5921                         if (fd >= 0)
5922                                 close(fd);
5923                 }
5924         }
5925         free_mdstat(mdstat);
5926         return count;
5927 }
5928
5929 #ifdef DEBUG_LOOP
5930 static struct md_list*
5931 get_loop_devices(void)
5932 {
5933         int i;
5934         struct md_list *devlist = NULL;
5935         struct md_list *dv;
5936
5937         for(i = 0; i < 12; i++) {
5938                 dv = xcalloc(1, sizeof(*dv));
5939                 dv->devname = xmalloc(40);
5940                 sprintf(dv->devname, "/dev/loop%d", i);
5941                 dv->next = devlist;
5942                 devlist = dv;
5943         }
5944         return devlist;
5945 }
5946 #endif
5947
5948 static struct md_list*
5949 get_devices(const char *hba_path)
5950 {
5951         struct md_list *devlist = NULL;
5952         struct md_list *dv;
5953         struct dirent *ent;
5954         DIR *dir;
5955         int err = 0;
5956
5957 #if DEBUG_LOOP
5958         devlist = get_loop_devices();
5959         return devlist;
5960 #endif
5961         /* scroll through /sys/dev/block looking for devices attached to
5962          * this hba
5963          */
5964         dir = opendir("/sys/dev/block");
5965         for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
5966                 int fd;
5967                 char buf[1024];
5968                 int major, minor;
5969                 char *path = NULL;
5970                 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
5971                         continue;
5972                 path = devt_to_devpath(makedev(major, minor));
5973                 if (!path)
5974                         continue;
5975                 if (!path_attached_to_hba(path, hba_path)) {
5976                         free(path);
5977                         path = NULL;
5978                         continue;
5979                 }
5980                 free(path);
5981                 path = NULL;
5982                 fd = dev_open(ent->d_name, O_RDONLY);
5983                 if (fd >= 0) {
5984                         fd2devname(fd, buf);
5985                         close(fd);
5986                 } else {
5987                         pr_err("cannot open device: %s\n",
5988                                 ent->d_name);
5989                         continue;
5990                 }
5991
5992                 dv = xcalloc(1, sizeof(*dv));
5993                 dv->devname = xstrdup(buf);
5994                 dv->next = devlist;
5995                 devlist = dv;
5996         }
5997         if (err) {
5998                 while(devlist) {
5999                         dv = devlist;
6000                         devlist = devlist->next;
6001                         free(dv->devname);
6002                         free(dv);
6003                 }
6004         }
6005         closedir(dir);
6006         return devlist;
6007 }
6008
6009 static int
6010 count_volumes_list(struct md_list *devlist, char *homehost,
6011                    int verbose, int *found)
6012 {
6013         struct md_list *tmpdev;
6014         int count = 0;
6015         struct supertype *st;
6016
6017         /* first walk the list of devices to find a consistent set
6018          * that match the criterea, if that is possible.
6019          * We flag the ones we like with 'used'.
6020          */
6021         *found = 0;
6022         st = match_metadata_desc_imsm("imsm");
6023         if (st == NULL) {
6024                 pr_vrb("cannot allocate memory for imsm supertype\n");
6025                 return 0;
6026         }
6027
6028         for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
6029                 char *devname = tmpdev->devname;
6030                 struct stat stb;
6031                 struct supertype *tst;
6032                 int dfd;
6033                 if (tmpdev->used > 1)
6034                         continue;
6035                 tst = dup_super(st);
6036                 if (tst == NULL) {
6037                         pr_vrb("cannot allocate memory for imsm supertype\n");
6038                         goto err_1;
6039                 }
6040                 tmpdev->container = 0;
6041                 dfd = dev_open(devname, O_RDONLY|O_EXCL);
6042                 if (dfd < 0) {
6043                         dprintf("cannot open device %s: %s\n",
6044                                 devname, strerror(errno));
6045                         tmpdev->used = 2;
6046                 } else if (fstat(dfd, &stb)< 0) {
6047                         /* Impossible! */
6048                         dprintf("fstat failed for %s: %s\n",
6049                                 devname, strerror(errno));
6050                         tmpdev->used = 2;
6051                 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
6052                         dprintf("%s is not a block device.\n",
6053                                 devname);
6054                         tmpdev->used = 2;
6055                 } else if (must_be_container(dfd)) {
6056                         struct supertype *cst;
6057                         cst = super_by_fd(dfd, NULL);
6058                         if (cst == NULL) {
6059                                 dprintf("cannot recognize container type %s\n",
6060                                         devname);
6061                                 tmpdev->used = 2;
6062                         } else if (tst->ss != st->ss) {
6063                                 dprintf("non-imsm container - ignore it: %s\n",
6064                                         devname);
6065                                 tmpdev->used = 2;
6066                         } else if (!tst->ss->load_container ||
6067                                    tst->ss->load_container(tst, dfd, NULL))
6068                                 tmpdev->used = 2;
6069                         else {
6070                                 tmpdev->container = 1;
6071                         }
6072                         if (cst)
6073                                 cst->ss->free_super(cst);
6074                 } else {
6075                         tmpdev->st_rdev = stb.st_rdev;
6076                         if (tst->ss->load_super(tst,dfd, NULL)) {
6077                                 dprintf("no RAID superblock on %s\n",
6078                                         devname);
6079                                 tmpdev->used = 2;
6080                         } else if (tst->ss->compare_super == NULL) {
6081                                 dprintf("Cannot assemble %s metadata on %s\n",
6082                                         tst->ss->name, devname);
6083                                 tmpdev->used = 2;
6084                         }
6085                 }
6086                 if (dfd >= 0)
6087                         close(dfd);
6088                 if (tmpdev->used == 2 || tmpdev->used == 4) {
6089                         /* Ignore unrecognised devices during auto-assembly */
6090                         goto loop;
6091                 }
6092                 else {
6093                         struct mdinfo info;
6094                         tst->ss->getinfo_super(tst, &info, NULL);
6095
6096                         if (st->minor_version == -1)
6097                                 st->minor_version = tst->minor_version;
6098
6099                         if (memcmp(info.uuid, uuid_zero,
6100                                    sizeof(int[4])) == 0) {
6101                                 /* this is a floating spare.  It cannot define
6102                                  * an array unless there are no more arrays of
6103                                  * this type to be found.  It can be included
6104                                  * in an array of this type though.
6105                                  */
6106                                 tmpdev->used = 3;
6107                                 goto loop;
6108                         }
6109
6110                         if (st->ss != tst->ss ||
6111                             st->minor_version != tst->minor_version ||
6112                             st->ss->compare_super(st, tst) != 0) {
6113                                 /* Some mismatch. If exactly one array matches this host,
6114                                  * we can resolve on that one.
6115                                  * Or, if we are auto assembling, we just ignore the second
6116                                  * for now.
6117                                  */
6118                                 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6119                                         devname);
6120                                 goto loop;
6121                         }
6122                         tmpdev->used = 1;
6123                         *found = 1;
6124                         dprintf("found: devname: %s\n", devname);
6125                 }
6126         loop:
6127                 if (tst)
6128                         tst->ss->free_super(tst);
6129         }
6130         if (*found != 0) {
6131                 int err;
6132                 if ((err = load_super_imsm_all(st, -1, &st->sb, NULL, devlist, 0)) == 0) {
6133                         struct mdinfo *iter, *head = st->ss->container_content(st, NULL);
6134                         for (iter = head; iter; iter = iter->next) {
6135                                 dprintf("content->text_version: %s vol\n",
6136                                         iter->text_version);
6137                                 if (iter->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
6138                                         /* do not assemble arrays with unsupported
6139                                            configurations */
6140                                         dprintf("Cannot activate member %s.\n",
6141                                                 iter->text_version);
6142                                 } else
6143                                         count++;
6144                         }
6145                         sysfs_free(head);
6146
6147                 } else {
6148                         dprintf("No valid super block on device list: err: %d %p\n",
6149                                 err, st->sb);
6150                 }
6151         } else {
6152                 dprintf("no more devices to examine\n");
6153         }
6154
6155         for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
6156                 if (tmpdev->used == 1 && tmpdev->found) {
6157                         if (count) {
6158                                 if (count < tmpdev->found)
6159                                         count = 0;
6160                                 else
6161                                         count -= tmpdev->found;
6162                         }
6163                 }
6164                 if (tmpdev->used == 1)
6165                         tmpdev->used = 4;
6166         }
6167         err_1:
6168         if (st)
6169                 st->ss->free_super(st);
6170         return count;
6171 }
6172
6173 static int
6174 count_volumes(struct intel_hba *hba, int dpa, int verbose)
6175 {
6176         struct sys_dev *idev, *intel_devices = find_intel_devices();
6177         int count = 0;
6178         const struct orom_entry *entry;
6179         struct devid_list *dv, *devid_list;
6180
6181         if (!hba || !hba->path)
6182                 return 0;
6183
6184         for (idev = intel_devices; idev; idev = idev->next) {
6185                 if (strstr(idev->path, hba->path))
6186                                 break;
6187         }
6188
6189         if (!idev || !idev->dev_id)
6190                 return 0;
6191
6192         entry = get_orom_entry_by_device_id(idev->dev_id);
6193
6194         if (!entry || !entry->devid_list)
6195                 return 0;
6196
6197         devid_list = entry->devid_list;
6198         for (dv = devid_list; dv; dv = dv->next) {
6199                 struct md_list *devlist;
6200                 struct sys_dev *device = device_by_id(dv->devid);
6201                 char *hba_path;
6202                 int found = 0;
6203
6204                 if (device)
6205                         hba_path = device->path;
6206                 else
6207                         return 0;
6208
6209                 devlist = get_devices(hba_path);
6210                 /* if no intel devices return zero volumes */
6211                 if (devlist == NULL)
6212                         return 0;
6213
6214                 count += active_arrays_by_format("imsm", hba_path, &devlist, dpa, verbose);
6215                 dprintf("path: %s active arrays: %d\n", hba_path, count);
6216                 if (devlist == NULL)
6217                         return 0;
6218                 do  {
6219                         found = 0;
6220                         count += count_volumes_list(devlist,
6221                                                         NULL,
6222                                                         verbose,
6223                                                         &found);
6224                         dprintf("found %d count: %d\n", found, count);
6225                 } while (found);
6226
6227                 dprintf("path: %s total number of volumes: %d\n", hba_path, count);
6228
6229                 while (devlist) {
6230                         struct md_list *dv = devlist;
6231                         devlist = devlist->next;
6232                         free(dv->devname);
6233                         free(dv);
6234                 }
6235         }
6236         return count;
6237 }
6238
6239 static int imsm_default_chunk(const struct imsm_orom *orom)
6240 {
6241         /* up to 512 if the plaform supports it, otherwise the platform max.
6242          * 128 if no platform detected
6243          */
6244         int fs = max(7, orom ? fls(orom->sss) : 0);
6245
6246         return min(512, (1 << fs));
6247 }
6248
6249 static int
6250 validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
6251                             int raiddisks, int *chunk, unsigned long long size, int verbose)
6252 {
6253         /* check/set platform and metadata limits/defaults */
6254         if (super->orom && raiddisks > super->orom->dpa) {
6255                 pr_vrb("platform supports a maximum of %d disks per array\n",
6256                        super->orom->dpa);
6257                 return 0;
6258         }
6259
6260         /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6261         if (!is_raid_level_supported(super->orom, level, raiddisks)) {
6262                 pr_vrb("platform does not support raid%d with %d disk%s\n",
6263                         level, raiddisks, raiddisks > 1 ? "s" : "");
6264                 return 0;
6265         }
6266
6267         if (*chunk == 0 || *chunk == UnSet)
6268                 *chunk = imsm_default_chunk(super->orom);
6269
6270         if (super->orom && !imsm_orom_has_chunk(super->orom, *chunk)) {
6271                 pr_vrb("platform does not support a chunk size of: %d\n", *chunk);
6272                 return 0;
6273         }
6274
6275         if (layout != imsm_level_to_layout(level)) {
6276                 if (level == 5)
6277                         pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
6278                 else if (level == 10)
6279                         pr_vrb("imsm raid 10 only supports the n2 layout\n");
6280                 else
6281                         pr_vrb("imsm unknown layout %#x for this raid level %d\n",
6282                                 layout, level);
6283                 return 0;
6284         }
6285
6286         if (super->orom && (super->orom->attr & IMSM_OROM_ATTR_2TB) == 0 &&
6287                         (calc_array_size(level, raiddisks, layout, *chunk, size) >> 32) > 0) {
6288                 pr_vrb("platform does not support a volume size over 2TB\n");
6289                 return 0;
6290         }
6291
6292         return 1;
6293 }
6294
6295 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6296  * FIX ME add ahci details
6297  */
6298 static int validate_geometry_imsm_volume(struct supertype *st, int level,
6299                                          int layout, int raiddisks, int *chunk,
6300                                          unsigned long long size,
6301                                          unsigned long long data_offset,
6302                                          char *dev,
6303                                          unsigned long long *freesize,
6304                                          int verbose)
6305 {
6306         struct stat stb;
6307         struct intel_super *super = st->sb;
6308         struct imsm_super *mpb;
6309         struct dl *dl;
6310         unsigned long long pos = 0;
6311         unsigned long long maxsize;
6312         struct extent *e;
6313         int i;
6314
6315         /* We must have the container info already read in. */
6316         if (!super)
6317                 return 0;
6318
6319         mpb = super->anchor;
6320
6321         if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, size, verbose)) {
6322                 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6323                 return 0;
6324         }
6325         if (!dev) {
6326                 /* General test:  make sure there is space for
6327                  * 'raiddisks' device extents of size 'size' at a given
6328                  * offset
6329                  */
6330                 unsigned long long minsize = size;
6331                 unsigned long long start_offset = MaxSector;
6332                 int dcnt = 0;
6333                 if (minsize == 0)
6334                         minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
6335                 for (dl = super->disks; dl ; dl = dl->next) {
6336                         int found = 0;
6337
6338                         pos = 0;
6339                         i = 0;
6340                         e = get_extents(super, dl);
6341                         if (!e) continue;
6342                         do {
6343                                 unsigned long long esize;
6344                                 esize = e[i].start - pos;
6345                                 if (esize >= minsize)
6346                                         found = 1;
6347                                 if (found && start_offset == MaxSector) {
6348                                         start_offset = pos;
6349                                         break;
6350                                 } else if (found && pos != start_offset) {
6351                                         found = 0;
6352                                         break;
6353                                 }
6354                                 pos = e[i].start + e[i].size;
6355                                 i++;
6356                         } while (e[i-1].size);
6357                         if (found)
6358                                 dcnt++;
6359                         free(e);
6360                 }
6361                 if (dcnt < raiddisks) {
6362                         if (verbose)
6363                                 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6364                                         dcnt, raiddisks);
6365                         return 0;
6366                 }
6367                 return 1;
6368         }
6369
6370         /* This device must be a member of the set */
6371         if (stat(dev, &stb) < 0)
6372                 return 0;
6373         if ((S_IFMT & stb.st_mode) != S_IFBLK)
6374                 return 0;
6375         for (dl = super->disks ; dl ; dl = dl->next) {
6376                 if (dl->major == (int)major(stb.st_rdev) &&
6377                     dl->minor == (int)minor(stb.st_rdev))
6378                         break;
6379         }
6380         if (!dl) {
6381                 if (verbose)
6382                         pr_err("%s is not in the same imsm set\n", dev);
6383                 return 0;
6384         } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
6385                 /* If a volume is present then the current creation attempt
6386                  * cannot incorporate new spares because the orom may not
6387                  * understand this configuration (all member disks must be
6388                  * members of each array in the container).
6389                  */
6390                 pr_err("%s is a spare and a volume is already defined for this container\n", dev);
6391                 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6392                 return 0;
6393         } else if (super->orom && mpb->num_raid_devs > 0 &&
6394                    mpb->num_disks != raiddisks) {
6395                 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6396                 return 0;
6397         }
6398
6399         /* retrieve the largest free space block */
6400         e = get_extents(super, dl);
6401         maxsize = 0;
6402         i = 0;
6403         if (e) {
6404                 do {
6405                         unsigned long long esize;
6406
6407                         esize = e[i].start - pos;
6408                         if (esize >= maxsize)
6409                                 maxsize = esize;
6410                         pos = e[i].start + e[i].size;
6411                         i++;
6412                 } while (e[i-1].size);
6413                 dl->e = e;
6414                 dl->extent_cnt = i;
6415         } else {
6416                 if (verbose)
6417                         pr_err("unable to determine free space for: %s\n",
6418                                 dev);
6419                 return 0;
6420         }
6421         if (maxsize < size) {
6422                 if (verbose)
6423                         pr_err("%s not enough space (%llu < %llu)\n",
6424                                 dev, maxsize, size);
6425                 return 0;
6426         }
6427
6428         /* count total number of extents for merge */
6429         i = 0;
6430         for (dl = super->disks; dl; dl = dl->next)
6431                 if (dl->e)
6432                         i += dl->extent_cnt;
6433
6434         maxsize = merge_extents(super, i);
6435
6436         if (!check_env("IMSM_NO_PLATFORM") &&
6437             mpb->num_raid_devs > 0 && size && size != maxsize) {
6438                 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6439                 return 0;
6440         }
6441
6442         if (maxsize < size || maxsize == 0) {
6443                 if (verbose) {
6444                         if (maxsize == 0)
6445                                 pr_err("no free space left on device. Aborting...\n");
6446                         else
6447                                 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
6448                                                 maxsize, size);
6449                 }
6450                 return 0;
6451         }
6452
6453         *freesize = maxsize;
6454
6455         if (super->orom) {
6456                 int count = count_volumes(super->hba,
6457                                       super->orom->dpa, verbose);
6458                 if (super->orom->vphba <= count) {
6459                         pr_vrb("platform does not support more than %d raid volumes.\n",
6460                                super->orom->vphba);
6461                         return 0;
6462                 }
6463         }
6464         return 1;
6465 }
6466
6467 static int imsm_get_free_size(struct supertype *st, int raiddisks,
6468                          unsigned long long size, int chunk,
6469                          unsigned long long *freesize)
6470 {
6471         struct intel_super *super = st->sb;
6472         struct imsm_super *mpb = super->anchor;
6473         struct dl *dl;
6474         int i;
6475         int extent_cnt;
6476         struct extent *e;
6477         unsigned long long maxsize;
6478         unsigned long long minsize;
6479         int cnt;
6480         int used;
6481
6482         /* find the largest common start free region of the possible disks */
6483         used = 0;
6484         extent_cnt = 0;
6485         cnt = 0;
6486         for (dl = super->disks; dl; dl = dl->next) {
6487                 dl->raiddisk = -1;
6488
6489                 if (dl->index >= 0)
6490                         used++;
6491
6492                 /* don't activate new spares if we are orom constrained
6493                  * and there is already a volume active in the container
6494                  */
6495                 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
6496                         continue;
6497
6498                 e = get_extents(super, dl);
6499                 if (!e)
6500                         continue;
6501                 for (i = 1; e[i-1].size; i++)
6502                         ;
6503                 dl->e = e;
6504                 dl->extent_cnt = i;
6505                 extent_cnt += i;
6506                 cnt++;
6507         }
6508
6509         maxsize = merge_extents(super, extent_cnt);
6510         minsize = size;
6511         if (size == 0)
6512                 /* chunk is in K */
6513                 minsize = chunk * 2;
6514
6515         if (cnt < raiddisks ||
6516             (super->orom && used && used != raiddisks) ||
6517             maxsize < minsize ||
6518             maxsize == 0) {
6519                 pr_err("not enough devices with space to create array.\n");
6520                 return 0; /* No enough free spaces large enough */
6521         }
6522
6523         if (size == 0) {
6524                 size = maxsize;
6525                 if (chunk) {
6526                         size /= 2 * chunk;
6527                         size *= 2 * chunk;
6528                 }
6529                 maxsize = size;
6530         }
6531         if (!check_env("IMSM_NO_PLATFORM") &&
6532             mpb->num_raid_devs > 0 && size && size != maxsize) {
6533                 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
6534                 return 0;
6535         }
6536         cnt = 0;
6537         for (dl = super->disks; dl; dl = dl->next)
6538                 if (dl->e)
6539                         dl->raiddisk = cnt++;
6540
6541         *freesize = size;
6542
6543         dprintf("imsm: imsm_get_free_size() returns : %llu\n", size);
6544
6545         return 1;
6546 }
6547
6548 static int reserve_space(struct supertype *st, int raiddisks,
6549                          unsigned long long size, int chunk,
6550                          unsigned long long *freesize)
6551 {
6552         struct intel_super *super = st->sb;
6553         struct dl *dl;
6554         int cnt;
6555         int rv = 0;
6556
6557         rv = imsm_get_free_size(st, raiddisks, size, chunk, freesize);
6558         if (rv) {
6559                 cnt = 0;
6560                 for (dl = super->disks; dl; dl = dl->next)
6561                         if (dl->e)
6562                                 dl->raiddisk = cnt++;
6563                 rv = 1;
6564         }
6565
6566         return rv;
6567 }
6568
6569 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
6570                                   int raiddisks, int *chunk, unsigned long long size,
6571                                   unsigned long long data_offset,
6572                                   char *dev, unsigned long long *freesize,
6573                                   int verbose)
6574 {
6575         int fd, cfd;
6576         struct mdinfo *sra;
6577         int is_member = 0;
6578
6579         /* load capability
6580          * if given unused devices create a container
6581          * if given given devices in a container create a member volume
6582          */
6583         if (level == LEVEL_CONTAINER) {
6584                 /* Must be a fresh device to add to a container */
6585                 return validate_geometry_imsm_container(st, level, layout,
6586                                                         raiddisks,
6587                                                         *chunk,
6588                                                         size, data_offset,
6589                                                         dev, freesize,
6590                                                         verbose);
6591         }
6592
6593         if (!dev) {
6594                 if (st->sb) {
6595                         struct intel_super *super = st->sb;
6596                         if (!validate_geometry_imsm_orom(st->sb, level, layout,
6597                                                          raiddisks, chunk, size,
6598                                                          verbose))
6599                                 return 0;
6600                         /* we are being asked to automatically layout a
6601                          * new volume based on the current contents of
6602                          * the container.  If the the parameters can be
6603                          * satisfied reserve_space will record the disks,
6604                          * start offset, and size of the volume to be
6605                          * created.  add_to_super and getinfo_super
6606                          * detect when autolayout is in progress.
6607                          */
6608                         /* assuming that freesize is always given when array is
6609                            created */
6610                         if (super->orom && freesize) {
6611                                 int count;
6612                                 count = count_volumes(super->hba,
6613                                                       super->orom->dpa, verbose);
6614                                 if (super->orom->vphba <= count) {
6615                                         pr_vrb("platform does not support more than %d raid volumes.\n",
6616                                                super->orom->vphba);
6617                                         return 0;
6618                                 }
6619                         }
6620                         if (freesize)
6621                                 return reserve_space(st, raiddisks, size,
6622                                                      *chunk, freesize);
6623                 }
6624                 return 1;
6625         }
6626         if (st->sb) {
6627                 /* creating in a given container */
6628                 return validate_geometry_imsm_volume(st, level, layout,
6629                                                      raiddisks, chunk, size,
6630                                                      data_offset,
6631                                                      dev, freesize, verbose);
6632         }
6633
6634         /* This device needs to be a device in an 'imsm' container */
6635         fd = open(dev, O_RDONLY|O_EXCL, 0);
6636         if (fd >= 0) {
6637                 if (verbose)
6638                         pr_err("Cannot create this array on device %s\n",
6639                                dev);
6640                 close(fd);
6641                 return 0;
6642         }
6643         if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
6644                 if (verbose)
6645                         pr_err("Cannot open %s: %s\n",
6646                                 dev, strerror(errno));
6647                 return 0;
6648         }
6649         /* Well, it is in use by someone, maybe an 'imsm' container. */
6650         cfd = open_container(fd);
6651         close(fd);
6652         if (cfd < 0) {
6653                 if (verbose)
6654                         pr_err("Cannot use %s: It is busy\n",
6655                                 dev);
6656                 return 0;
6657         }
6658         sra = sysfs_read(cfd, NULL, GET_VERSION);
6659         if (sra && sra->array.major_version == -1 &&
6660             strcmp(sra->text_version, "imsm") == 0)
6661                 is_member = 1;
6662         sysfs_free(sra);
6663         if (is_member) {
6664                 /* This is a member of a imsm container.  Load the container
6665                  * and try to create a volume
6666                  */
6667                 struct intel_super *super;
6668
6669                 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, NULL, 1) == 0) {
6670                         st->sb = super;
6671                         strcpy(st->container_devnm, fd2devnm(cfd));
6672                         close(cfd);
6673                         return validate_geometry_imsm_volume(st, level, layout,
6674                                                              raiddisks, chunk,
6675                                                              size, data_offset, dev,
6676                                                              freesize, 1)
6677                                 ? 1 : -1;
6678                 }
6679         }
6680
6681         if (verbose)
6682                 pr_err("failed container membership check\n");
6683
6684         close(cfd);
6685         return 0;
6686 }
6687
6688 static void default_geometry_imsm(struct supertype *st, int *level, int *layout, int *chunk)
6689 {
6690         struct intel_super *super = st->sb;
6691
6692         if (level && *level == UnSet)
6693                 *level = LEVEL_CONTAINER;
6694
6695         if (level && layout && *layout == UnSet)
6696                 *layout = imsm_level_to_layout(*level);
6697
6698         if (chunk && (*chunk == UnSet || *chunk == 0))
6699                 *chunk = imsm_default_chunk(super->orom);
6700 }
6701
6702 static void handle_missing(struct intel_super *super, struct imsm_dev *dev);
6703
6704 static int kill_subarray_imsm(struct supertype *st)
6705 {
6706         /* remove the subarray currently referenced by ->current_vol */
6707         __u8 i;
6708         struct intel_dev **dp;
6709         struct intel_super *super = st->sb;
6710         __u8 current_vol = super->current_vol;
6711         struct imsm_super *mpb = super->anchor;
6712
6713         if (super->current_vol < 0)
6714                 return 2;
6715         super->current_vol = -1; /* invalidate subarray cursor */
6716
6717         /* block deletions that would change the uuid of active subarrays
6718          *
6719          * FIXME when immutable ids are available, but note that we'll
6720          * also need to fixup the invalidated/active subarray indexes in
6721          * mdstat
6722          */
6723         for (i = 0; i < mpb->num_raid_devs; i++) {
6724                 char subarray[4];
6725
6726                 if (i < current_vol)
6727                         continue;
6728                 sprintf(subarray, "%u", i);
6729                 if (is_subarray_active(subarray, st->devnm)) {
6730                         pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
6731                                current_vol, i);
6732
6733                         return 2;
6734                 }
6735         }
6736
6737         if (st->update_tail) {
6738                 struct imsm_update_kill_array *u = xmalloc(sizeof(*u));
6739
6740                 u->type = update_kill_array;
6741                 u->dev_idx = current_vol;
6742                 append_metadata_update(st, u, sizeof(*u));
6743
6744                 return 0;
6745         }
6746
6747         for (dp = &super->devlist; *dp;)
6748                 if ((*dp)->index == current_vol) {
6749                         *dp = (*dp)->next;
6750                 } else {
6751                         handle_missing(super, (*dp)->dev);
6752                         if ((*dp)->index > current_vol)
6753                                 (*dp)->index--;
6754                         dp = &(*dp)->next;
6755                 }
6756
6757         /* no more raid devices, all active components are now spares,
6758          * but of course failed are still failed
6759          */
6760         if (--mpb->num_raid_devs == 0) {
6761                 struct dl *d;
6762
6763                 for (d = super->disks; d; d = d->next)
6764                         if (d->index > -2)
6765                                 mark_spare(d);
6766         }
6767
6768         super->updates_pending++;
6769
6770         return 0;
6771 }
6772
6773 static int update_subarray_imsm(struct supertype *st, char *subarray,
6774                                 char *update, struct mddev_ident *ident)
6775 {
6776         /* update the subarray currently referenced by ->current_vol */
6777         struct intel_super *super = st->sb;
6778         struct imsm_super *mpb = super->anchor;
6779
6780         if (strcmp(update, "name") == 0) {
6781                 char *name = ident->name;
6782                 char *ep;
6783                 int vol;
6784
6785                 if (is_subarray_active(subarray, st->devnm)) {
6786                         pr_err("Unable to update name of active subarray\n");
6787                         return 2;
6788                 }
6789
6790                 if (!check_name(super, name, 0))
6791                         return 2;
6792
6793                 vol = strtoul(subarray, &ep, 10);
6794                 if (*ep != '\0' || vol >= super->anchor->num_raid_devs)
6795                         return 2;
6796
6797                 if (st->update_tail) {
6798                         struct imsm_update_rename_array *u = xmalloc(sizeof(*u));
6799
6800                         u->type = update_rename_array;
6801                         u->dev_idx = vol;
6802                         snprintf((char *) u->name, MAX_RAID_SERIAL_LEN, "%s", name);
6803                         append_metadata_update(st, u, sizeof(*u));
6804                 } else {
6805                         struct imsm_dev *dev;
6806                         int i;
6807
6808                         dev = get_imsm_dev(super, vol);
6809                         snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
6810                         for (i = 0; i < mpb->num_raid_devs; i++) {
6811                                 dev = get_imsm_dev(super, i);
6812                                 handle_missing(super, dev);
6813                         }
6814                         super->updates_pending++;
6815                 }
6816         } else
6817                 return 2;
6818
6819         return 0;
6820 }
6821 #endif /* MDASSEMBLE */
6822
6823 static int is_gen_migration(struct imsm_dev *dev)
6824 {
6825         if (dev == NULL)
6826                 return 0;
6827
6828         if (!dev->vol.migr_state)
6829                 return 0;
6830
6831         if (migr_type(dev) == MIGR_GEN_MIGR)
6832                 return 1;
6833
6834         return 0;
6835 }
6836
6837 static int is_rebuilding(struct imsm_dev *dev)
6838 {
6839         struct imsm_map *migr_map;
6840
6841         if (!dev->vol.migr_state)
6842                 return 0;
6843
6844         if (migr_type(dev) != MIGR_REBUILD)
6845                 return 0;
6846
6847         migr_map = get_imsm_map(dev, MAP_1);
6848
6849         if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
6850                 return 1;
6851         else
6852                 return 0;
6853 }
6854
6855 #ifndef MDASSEMBLE
6856 static int is_initializing(struct imsm_dev *dev)
6857 {
6858         struct imsm_map *migr_map;
6859
6860         if (!dev->vol.migr_state)
6861                 return 0;
6862
6863         if (migr_type(dev) != MIGR_INIT)
6864                 return 0;
6865
6866         migr_map = get_imsm_map(dev, MAP_1);
6867
6868         if (migr_map->map_state == IMSM_T_STATE_UNINITIALIZED)
6869                 return 1;
6870
6871         return 0;
6872 }
6873 #endif
6874
6875 static void update_recovery_start(struct intel_super *super,
6876                                         struct imsm_dev *dev,
6877                                         struct mdinfo *array)
6878 {
6879         struct mdinfo *rebuild = NULL;
6880         struct mdinfo *d;
6881         __u32 units;
6882
6883         if (!is_rebuilding(dev))
6884                 return;
6885
6886         /* Find the rebuild target, but punt on the dual rebuild case */
6887         for (d = array->devs; d; d = d->next)
6888                 if (d->recovery_start == 0) {
6889                         if (rebuild)
6890                                 return;
6891                         rebuild = d;
6892                 }
6893
6894         if (!rebuild) {
6895                 /* (?) none of the disks are marked with
6896                  * IMSM_ORD_REBUILD, so assume they are missing and the
6897                  * disk_ord_tbl was not correctly updated
6898                  */
6899                 dprintf("failed to locate out-of-sync disk\n");
6900                 return;
6901         }
6902
6903         units = __le32_to_cpu(dev->vol.curr_migr_unit);
6904         rebuild->recovery_start = units * blocks_per_migr_unit(super, dev);
6905 }
6906
6907 #ifndef MDASSEMBLE
6908 static int recover_backup_imsm(struct supertype *st, struct mdinfo *info);
6909 #endif
6910
6911 static struct mdinfo *container_content_imsm(struct supertype *st, char *subarray)
6912 {
6913         /* Given a container loaded by load_super_imsm_all,
6914          * extract information about all the arrays into
6915          * an mdinfo tree.
6916          * If 'subarray' is given, just extract info about that array.
6917          *
6918          * For each imsm_dev create an mdinfo, fill it in,
6919          *  then look for matching devices in super->disks
6920          *  and create appropriate device mdinfo.
6921          */
6922         struct intel_super *super = st->sb;
6923         struct imsm_super *mpb = super->anchor;
6924         struct mdinfo *rest = NULL;
6925         unsigned int i;
6926         int sb_errors = 0;
6927         struct dl *d;
6928         int spare_disks = 0;
6929
6930         /* do not assemble arrays when not all attributes are supported */
6931         if (imsm_check_attributes(mpb->attributes) == 0) {
6932                 sb_errors = 1;
6933                 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
6934         }
6935
6936         /* check for bad blocks */
6937         if (imsm_bbm_log_size(super->anchor)) {
6938                 pr_err("BBM log found in IMSM metadata.Arrays activation is blocked.\n");
6939                 sb_errors = 1;
6940         }
6941
6942         /* count spare devices, not used in maps
6943          */
6944         for (d = super->disks; d; d = d->next)
6945                 if (d->index == -1)
6946                         spare_disks++;
6947
6948         for (i = 0; i < mpb->num_raid_devs; i++) {
6949                 struct imsm_dev *dev;
6950                 struct imsm_map *map;
6951                 struct imsm_map *map2;
6952                 struct mdinfo *this;
6953                 int slot;
6954 #ifndef MDASSEMBLE
6955                 int chunk;
6956 #endif
6957                 char *ep;
6958
6959                 if (subarray &&
6960                     (i != strtoul(subarray, &ep, 10) || *ep != '\0'))
6961                         continue;
6962
6963                 dev = get_imsm_dev(super, i);
6964                 map = get_imsm_map(dev, MAP_0);
6965                 map2 = get_imsm_map(dev, MAP_1);
6966
6967                 /* do not publish arrays that are in the middle of an
6968                  * unsupported migration
6969                  */
6970                 if (dev->vol.migr_state &&
6971                     (migr_type(dev) == MIGR_STATE_CHANGE)) {
6972                         pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
6973                                 dev->volume);
6974                         continue;
6975                 }
6976                 /* do not publish arrays that are not support by controller's
6977                  * OROM/EFI
6978                  */
6979
6980                 this = xmalloc(sizeof(*this));
6981
6982                 super->current_vol = i;
6983                 getinfo_super_imsm_volume(st, this, NULL);
6984                 this->next = rest;
6985 #ifndef MDASSEMBLE
6986                 chunk = __le16_to_cpu(map->blocks_per_strip) >> 1;
6987                 /* mdadm does not support all metadata features- set the bit in all arrays state */
6988                 if (!validate_geometry_imsm_orom(super,
6989                                                  get_imsm_raid_level(map), /* RAID level */
6990                                                  imsm_level_to_layout(get_imsm_raid_level(map)),
6991                                                  map->num_members, /* raid disks */
6992                                                  &chunk, join_u32(dev->size_low, dev->size_high),
6993                                                  1 /* verbose */)) {
6994                         pr_err("IMSM RAID geometry validation failed.  Array %s activation is blocked.\n",
6995                                 dev->volume);
6996                         this->array.state |=
6997                           (1<<MD_SB_BLOCK_CONTAINER_RESHAPE) |
6998                           (1<<MD_SB_BLOCK_VOLUME);
6999                 }
7000 #endif
7001
7002                 /* if array has bad blocks, set suitable bit in all arrays state */
7003                 if (sb_errors)
7004                         this->array.state |=
7005                           (1<<MD_SB_BLOCK_CONTAINER_RESHAPE) |
7006                           (1<<MD_SB_BLOCK_VOLUME);
7007
7008                 for (slot = 0 ; slot <  map->num_members; slot++) {
7009                         unsigned long long recovery_start;
7010                         struct mdinfo *info_d;
7011                         struct dl *d;
7012                         int idx;
7013                         int skip;
7014                         __u32 ord;
7015
7016                         skip = 0;
7017                         idx = get_imsm_disk_idx(dev, slot, MAP_0);
7018                         ord = get_imsm_ord_tbl_ent(dev, slot, MAP_X);
7019                         for (d = super->disks; d ; d = d->next)
7020                                 if (d->index == idx)
7021                                         break;
7022
7023                         recovery_start = MaxSector;
7024                         if (d == NULL)
7025                                 skip = 1;
7026                         if (d && is_failed(&d->disk))
7027                                 skip = 1;
7028                         if (ord & IMSM_ORD_REBUILD)
7029                                 recovery_start = 0;
7030
7031                         /*
7032                          * if we skip some disks the array will be assmebled degraded;
7033                          * reset resync start to avoid a dirty-degraded
7034                          * situation when performing the intial sync
7035                          *
7036                          * FIXME handle dirty degraded
7037                          */
7038                         if ((skip || recovery_start == 0) && !dev->vol.dirty)
7039                                 this->resync_start = MaxSector;
7040                         if (skip)
7041                                 continue;
7042
7043                         info_d = xcalloc(1, sizeof(*info_d));
7044                         info_d->next = this->devs;
7045                         this->devs = info_d;
7046
7047                         info_d->disk.number = d->index;
7048                         info_d->disk.major = d->major;
7049                         info_d->disk.minor = d->minor;
7050                         info_d->disk.raid_disk = slot;
7051                         info_d->recovery_start = recovery_start;
7052                         if (map2) {
7053                                 if (slot < map2->num_members)
7054                                         info_d->disk.state = (1 << MD_DISK_ACTIVE);
7055                                 else
7056                                         this->array.spare_disks++;
7057                         } else {
7058                                 if (slot < map->num_members)
7059                                         info_d->disk.state = (1 << MD_DISK_ACTIVE);
7060                                 else
7061                                         this->array.spare_disks++;
7062                         }
7063                         if (info_d->recovery_start == MaxSector)
7064                                 this->array.working_disks++;
7065
7066                         info_d->events = __le32_to_cpu(mpb->generation_num);
7067                         info_d->data_offset = pba_of_lba0(map);
7068                         info_d->component_size = blocks_per_member(map);
7069                 }
7070                 /* now that the disk list is up-to-date fixup recovery_start */
7071                 update_recovery_start(super, dev, this);
7072                 this->array.spare_disks += spare_disks;
7073
7074 #ifndef MDASSEMBLE
7075                 /* check for reshape */
7076                 if (this->reshape_active == 1)
7077                         recover_backup_imsm(st, this);
7078 #endif
7079                 rest = this;
7080         }
7081
7082         return rest;
7083 }
7084
7085 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev,
7086                                 int failed, int look_in_map)
7087 {
7088         struct imsm_map *map;
7089
7090         map = get_imsm_map(dev, look_in_map);
7091
7092         if (!failed)
7093                 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
7094                         IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
7095
7096         switch (get_imsm_raid_level(map)) {
7097         case 0:
7098                 return IMSM_T_STATE_FAILED;
7099                 break;
7100         case 1:
7101                 if (failed < map->num_members)
7102                         return IMSM_T_STATE_DEGRADED;
7103                 else
7104                         return IMSM_T_STATE_FAILED;
7105                 break;
7106         case 10:
7107         {
7108                 /**
7109                  * check to see if any mirrors have failed, otherwise we
7110                  * are degraded.  Even numbered slots are mirrored on
7111                  * slot+1
7112                  */
7113                 int i;
7114                 /* gcc -Os complains that this is unused */
7115                 int insync = insync;
7116
7117                 for (i = 0; i < map->num_members; i++) {
7118                         __u32 ord = get_imsm_ord_tbl_ent(dev, i, MAP_X);
7119                         int idx = ord_to_idx(ord);
7120                         struct imsm_disk *disk;
7121
7122                         /* reset the potential in-sync count on even-numbered
7123                          * slots.  num_copies is always 2 for imsm raid10
7124                          */
7125                         if ((i & 1) == 0)
7126                                 insync = 2;
7127
7128                         disk = get_imsm_disk(super, idx);
7129                         if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
7130                                 insync--;
7131
7132                         /* no in-sync disks left in this mirror the
7133                          * array has failed
7134                          */
7135                         if (insync == 0)
7136                                 return IMSM_T_STATE_FAILED;
7137                 }
7138
7139                 return IMSM_T_STATE_DEGRADED;
7140         }
7141         case 5:
7142                 if (failed < 2)
7143                         return IMSM_T_STATE_DEGRADED;
7144                 else
7145                         return IMSM_T_STATE_FAILED;
7146                 break;
7147         default:
7148                 break;
7149         }
7150
7151         return map->map_state;
7152 }
7153
7154 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev,
7155                              int look_in_map)
7156 {
7157         int i;
7158         int failed = 0;
7159         struct imsm_disk *disk;
7160         struct imsm_map *map = get_imsm_map(dev, MAP_0);
7161         struct imsm_map *prev = get_imsm_map(dev, MAP_1);
7162         struct imsm_map *map_for_loop;
7163         __u32 ord;
7164         int idx;
7165         int idx_1;
7166
7167         /* at the beginning of migration we set IMSM_ORD_REBUILD on
7168          * disks that are being rebuilt.  New failures are recorded to
7169          * map[0].  So we look through all the disks we started with and
7170          * see if any failures are still present, or if any new ones
7171          * have arrived
7172          */
7173         map_for_loop = map;
7174         if (prev && (map->num_members < prev->num_members))
7175                 map_for_loop = prev;
7176
7177         for (i = 0; i < map_for_loop->num_members; i++) {
7178                 idx_1 = -255;
7179                 /* when MAP_X is passed both maps failures are counted
7180                  */
7181                 if (prev &&
7182                     (look_in_map == MAP_1 || look_in_map == MAP_X) &&
7183                     i < prev->num_members) {
7184                         ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
7185                         idx_1 = ord_to_idx(ord);
7186
7187                         disk = get_imsm_disk(super, idx_1);
7188                         if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
7189                                 failed++;
7190                 }
7191                 if ((look_in_map == MAP_0 || look_in_map == MAP_X) &&
7192                     i < map->num_members) {
7193                         ord = __le32_to_cpu(map->disk_ord_tbl[i]);
7194                         idx = ord_to_idx(ord);
7195
7196                         if (idx != idx_1) {
7197                                 disk = get_imsm_disk(super, idx);
7198                                 if (!disk || is_failed(disk) ||
7199                                     ord & IMSM_ORD_REBUILD)
7200                                         failed++;
7201                         }
7202                 }
7203         }
7204
7205         return failed;
7206 }
7207
7208 #ifndef MDASSEMBLE
7209 static int imsm_open_new(struct supertype *c, struct active_array *a,
7210                          char *inst)
7211 {
7212         struct intel_super *super = c->sb;
7213         struct imsm_super *mpb = super->anchor;
7214
7215         if (atoi(inst) >= mpb->num_raid_devs) {
7216                 pr_err("subarry index %d, out of range\n", atoi(inst));
7217                 return -ENODEV;
7218         }
7219
7220         dprintf("imsm: open_new %s\n", inst);
7221         a->info.container_member = atoi(inst);
7222         return 0;
7223 }
7224
7225 static int is_resyncing(struct imsm_dev *dev)
7226 {
7227         struct imsm_map *migr_map;
7228
7229         if (!dev->vol.migr_state)
7230                 return 0;
7231
7232         if (migr_type(dev) == MIGR_INIT ||
7233             migr_type(dev) == MIGR_REPAIR)
7234                 return 1;
7235
7236         if (migr_type(dev) == MIGR_GEN_MIGR)
7237                 return 0;
7238
7239         migr_map = get_imsm_map(dev, MAP_1);
7240
7241         if (migr_map->map_state == IMSM_T_STATE_NORMAL &&
7242             dev->vol.migr_type != MIGR_GEN_MIGR)
7243                 return 1;
7244         else
7245                 return 0;
7246 }
7247
7248 /* return true if we recorded new information */
7249 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
7250 {
7251         __u32 ord;
7252         int slot;
7253         struct imsm_map *map;
7254         char buf[MAX_RAID_SERIAL_LEN+3];
7255         unsigned int len, shift = 0;
7256
7257         /* new failures are always set in map[0] */
7258         map = get_imsm_map(dev, MAP_0);
7259
7260         slot = get_imsm_disk_slot(map, idx);
7261         if (slot < 0)
7262                 return 0;
7263
7264         ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
7265         if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
7266                 return 0;
7267
7268         memcpy(buf, disk->serial, MAX_RAID_SERIAL_LEN);
7269         buf[MAX_RAID_SERIAL_LEN] = '\000';
7270         strcat(buf, ":0");
7271         if ((len = strlen(buf)) >= MAX_RAID_SERIAL_LEN)
7272                 shift = len - MAX_RAID_SERIAL_LEN + 1;
7273         strncpy((char *)disk->serial, &buf[shift], MAX_RAID_SERIAL_LEN);
7274
7275         disk->status |= FAILED_DISK;
7276         set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
7277         /* mark failures in second map if second map exists and this disk
7278          * in this slot.
7279          * This is valid for migration, initialization and rebuild
7280          */
7281         if (dev->vol.migr_state) {
7282                 struct imsm_map *map2 = get_imsm_map(dev, MAP_1);
7283                 int slot2 = get_imsm_disk_slot(map2, idx);
7284
7285                 if (slot2 < map2->num_members && slot2 >= 0)
7286                         set_imsm_ord_tbl_ent(map2, slot2,
7287                                              idx | IMSM_ORD_REBUILD);
7288         }
7289         if (map->failed_disk_num == 0xff)
7290                 map->failed_disk_num = slot;
7291         return 1;
7292 }
7293
7294 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
7295 {
7296         mark_failure(dev, disk, idx);
7297
7298         if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
7299                 return;
7300
7301         disk->scsi_id = __cpu_to_le32(~(__u32)0);
7302         memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
7303 }
7304
7305 static void handle_missing(struct intel_super *super, struct imsm_dev *dev)
7306 {
7307         struct dl *dl;
7308
7309         if (!super->missing)
7310                 return;
7311
7312         /* When orom adds replacement for missing disk it does
7313          * not remove entry of missing disk, but just updates map with
7314          * new added disk. So it is not enough just to test if there is
7315          * any missing disk, we have to look if there are any failed disks
7316          * in map to stop migration */
7317
7318         dprintf("imsm: mark missing\n");
7319         /* end process for initialization and rebuild only
7320          */
7321         if (is_gen_migration(dev) == 0) {
7322                 __u8 map_state;
7323                 int failed;
7324
7325                 failed = imsm_count_failed(super, dev, MAP_0);
7326                 map_state = imsm_check_degraded(super, dev, failed, MAP_0);
7327
7328                 if (failed)
7329                         end_migration(dev, super, map_state);
7330         }
7331         for (dl = super->missing; dl; dl = dl->next)
7332                 mark_missing(dev, &dl->disk, dl->index);
7333         super->updates_pending++;
7334 }
7335
7336 static unsigned long long imsm_set_array_size(struct imsm_dev *dev,
7337                                               long long new_size)
7338 {
7339         int used_disks = imsm_num_data_members(dev, MAP_0);
7340         unsigned long long array_blocks;
7341         struct imsm_map *map;
7342
7343         if (used_disks == 0) {
7344                 /* when problems occures
7345                  * return current array_blocks value
7346                  */
7347                 array_blocks = __le32_to_cpu(dev->size_high);
7348                 array_blocks = array_blocks << 32;
7349                 array_blocks += __le32_to_cpu(dev->size_low);
7350
7351                 return array_blocks;
7352         }
7353
7354         /* set array size in metadata
7355          */
7356         if (new_size <= 0) {
7357                 /* OLCE size change is caused by added disks
7358                  */
7359                 map = get_imsm_map(dev, MAP_0);
7360                 array_blocks = blocks_per_member(map) * used_disks;
7361         } else {
7362                 /* Online Volume Size Change
7363                  * Using  available free space
7364                  */
7365                 array_blocks = new_size;
7366         }
7367
7368         /* round array size down to closest MB
7369          */
7370         array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
7371         dev->size_low = __cpu_to_le32((__u32)array_blocks);
7372         dev->size_high = __cpu_to_le32((__u32)(array_blocks >> 32));
7373
7374         return array_blocks;
7375 }
7376
7377 static void imsm_set_disk(struct active_array *a, int n, int state);
7378
7379 static void imsm_progress_container_reshape(struct intel_super *super)
7380 {
7381         /* if no device has a migr_state, but some device has a
7382          * different number of members than the previous device, start
7383          * changing the number of devices in this device to match
7384          * previous.
7385          */
7386         struct imsm_super *mpb = super->anchor;
7387         int prev_disks = -1;
7388         int i;
7389         int copy_map_size;
7390
7391         for (i = 0; i < mpb->num_raid_devs; i++) {
7392                 struct imsm_dev *dev = get_imsm_dev(super, i);
7393                 struct imsm_map *map = get_imsm_map(dev, MAP_0);
7394                 struct imsm_map *map2;
7395                 int prev_num_members;
7396
7397                 if (dev->vol.migr_state)
7398                         return;
7399
7400                 if (prev_disks == -1)
7401                         prev_disks = map->num_members;
7402                 if (prev_disks == map->num_members)
7403                         continue;
7404
7405                 /* OK, this array needs to enter reshape mode.
7406                  * i.e it needs a migr_state
7407                  */
7408
7409                 copy_map_size = sizeof_imsm_map(map);
7410                 prev_num_members = map->num_members;
7411                 map->num_members = prev_disks;
7412                 dev->vol.migr_state = 1;
7413                 dev->vol.curr_migr_unit = 0;
7414                 set_migr_type(dev, MIGR_GEN_MIGR);
7415                 for (i = prev_num_members;
7416                      i < map->num_members; i++)
7417                         set_imsm_ord_tbl_ent(map, i, i);
7418                 map2 = get_imsm_map(dev, MAP_1);
7419                 /* Copy the current map */
7420                 memcpy(map2, map, copy_map_size);
7421                 map2->num_members = prev_num_members;
7422
7423                 imsm_set_array_size(dev, -1);
7424                 super->clean_migration_record_by_mdmon = 1;
7425                 super->updates_pending++;
7426         }
7427 }
7428
7429 /* Handle dirty -> clean transititions, resync and reshape.  Degraded and rebuild
7430  * states are handled in imsm_set_disk() with one exception, when a
7431  * resync is stopped due to a new failure this routine will set the
7432  * 'degraded' state for the array.
7433  */
7434 static int imsm_set_array_state(struct active_array *a, int consistent)
7435 {
7436         int inst = a->info.container_member;
7437         struct intel_super *super = a->container->sb;
7438         struct imsm_dev *dev = get_imsm_dev(super, inst);
7439         struct imsm_map *map = get_imsm_map(dev, MAP_0);
7440         int failed = imsm_count_failed(super, dev, MAP_0);
7441         __u8 map_state = imsm_check_degraded(super, dev, failed, MAP_0);
7442         __u32 blocks_per_unit;
7443
7444         if (dev->vol.migr_state &&
7445             dev->vol.migr_type  == MIGR_GEN_MIGR) {
7446                 /* array state change is blocked due to reshape action
7447                  * We might need to
7448                  * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
7449                  * - finish the reshape (if last_checkpoint is big and action != reshape)
7450                  * - update curr_migr_unit
7451                  */
7452                 if (a->curr_action == reshape) {
7453                         /* still reshaping, maybe update curr_migr_unit */
7454                         goto mark_checkpoint;
7455                 } else {
7456                         if (a->last_checkpoint == 0 && a->prev_action == reshape) {
7457                                 /* for some reason we aborted the reshape.
7458                                  *
7459                                  * disable automatic metadata rollback
7460                                  * user action is required to recover process
7461                                  */
7462                                 if (0) {
7463                                         struct imsm_map *map2 =
7464                                                 get_imsm_map(dev, MAP_1);
7465                                         dev->vol.migr_state = 0;
7466                                         set_migr_type(dev, 0);
7467                                         dev->vol.curr_migr_unit = 0;
7468                                         memcpy(map, map2,
7469                                                sizeof_imsm_map(map2));
7470                                         super->updates_pending++;
7471                                 }
7472                         }
7473                         if (a->last_checkpoint >= a->info.component_size) {
7474                                 unsigned long long array_blocks;
7475                                 int used_disks;
7476                                 struct mdinfo *mdi;
7477
7478                                 used_disks = imsm_num_data_members(dev, MAP_0);
7479                                 if (used_disks > 0) {
7480                                         array_blocks =
7481                                                 blocks_per_member(map) *
7482                                                 used_disks;
7483                                         /* round array size down to closest MB
7484                                          */
7485                                         array_blocks = (array_blocks
7486                                                         >> SECT_PER_MB_SHIFT)
7487                                                 << SECT_PER_MB_SHIFT;
7488                                         a->info.custom_array_size = array_blocks;
7489                                         /* encourage manager to update array
7490                                          * size
7491                                          */
7492
7493                                         a->check_reshape = 1;
7494                                 }
7495                                 /* finalize online capacity expansion/reshape */
7496                                 for (mdi = a->info.devs; mdi; mdi = mdi->next)
7497                                         imsm_set_disk(a,
7498                                                       mdi->disk.raid_disk,
7499                                                       mdi->curr_state);
7500
7501                                 imsm_progress_container_reshape(super);
7502                         }
7503                 }
7504         }
7505
7506         /* before we activate this array handle any missing disks */
7507         if (consistent == 2)
7508                 handle_missing(super, dev);
7509
7510         if (consistent == 2 &&
7511             (!is_resync_complete(&a->info) ||
7512              map_state != IMSM_T_STATE_NORMAL ||
7513              dev->vol.migr_state))
7514                 consistent = 0;
7515
7516         if (is_resync_complete(&a->info)) {
7517                 /* complete intialization / resync,
7518                  * recovery and interrupted recovery is completed in
7519                  * ->set_disk
7520                  */
7521                 if (is_resyncing(dev)) {
7522                         dprintf("imsm: mark resync done\n");
7523                         end_migration(dev, super, map_state);
7524                         super->updates_pending++;
7525                         a->last_checkpoint = 0;
7526                 }
7527         } else if ((!is_resyncing(dev) && !failed) &&
7528                    (imsm_reshape_blocks_arrays_changes(super) == 0)) {
7529                 /* mark the start of the init process if nothing is failed */
7530                 dprintf("imsm: mark resync start\n");
7531                 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
7532                         migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_INIT);
7533                 else
7534                         migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
7535                 super->updates_pending++;
7536         }
7537
7538 mark_checkpoint:
7539         /* skip checkpointing for general migration,
7540          * it is controlled in mdadm
7541          */
7542         if (is_gen_migration(dev))
7543                 goto skip_mark_checkpoint;
7544
7545         /* check if we can update curr_migr_unit from resync_start, recovery_start */
7546         blocks_per_unit = blocks_per_migr_unit(super, dev);
7547         if (blocks_per_unit) {
7548                 __u32 units32;
7549                 __u64 units;
7550
7551                 units = a->last_checkpoint / blocks_per_unit;
7552                 units32 = units;
7553
7554                 /* check that we did not overflow 32-bits, and that
7555                  * curr_migr_unit needs updating
7556                  */
7557                 if (units32 == units &&
7558                     units32 != 0 &&
7559                     __le32_to_cpu(dev->vol.curr_migr_unit) != units32) {
7560                         dprintf("imsm: mark checkpoint (%u)\n", units32);
7561                         dev->vol.curr_migr_unit = __cpu_to_le32(units32);
7562                         super->updates_pending++;
7563                 }
7564         }
7565
7566 skip_mark_checkpoint:
7567         /* mark dirty / clean */
7568         if (dev->vol.dirty != !consistent) {
7569                 dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty");
7570                 if (consistent)
7571                         dev->vol.dirty = 0;
7572                 else
7573                         dev->vol.dirty = 1;
7574                 super->updates_pending++;
7575         }
7576
7577         return consistent;
7578 }
7579
7580 static void imsm_set_disk(struct active_array *a, int n, int state)
7581 {
7582         int inst = a->info.container_member;
7583         struct intel_super *super = a->container->sb;
7584         struct imsm_dev *dev = get_imsm_dev(super, inst);
7585         struct imsm_map *map = get_imsm_map(dev, MAP_0);
7586         struct imsm_disk *disk;
7587         struct mdinfo *mdi;
7588         int recovery_not_finished = 0;
7589         int failed;
7590         __u32 ord;
7591         __u8 map_state;
7592
7593         if (n > map->num_members)
7594                 pr_err("imsm: set_disk %d out of range 0..%d\n",
7595                         n, map->num_members - 1);
7596
7597         if (n < 0)
7598                 return;
7599
7600         dprintf("imsm: set_disk %d:%x\n", n, state);
7601
7602         ord = get_imsm_ord_tbl_ent(dev, n, MAP_0);
7603         disk = get_imsm_disk(super, ord_to_idx(ord));
7604
7605         /* check for new failures */
7606         if (state & DS_FAULTY) {
7607                 if (mark_failure(dev, disk, ord_to_idx(ord)))
7608                         super->updates_pending++;
7609         }
7610
7611         /* check if in_sync */
7612         if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
7613                 struct imsm_map *migr_map = get_imsm_map(dev, MAP_1);
7614
7615                 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
7616                 super->updates_pending++;
7617         }
7618
7619         failed = imsm_count_failed(super, dev, MAP_0);
7620         map_state = imsm_check_degraded(super, dev, failed, MAP_0);
7621
7622         /* check if recovery complete, newly degraded, or failed */
7623         dprintf("imsm: Detected transition to state ");
7624         switch (map_state) {
7625         case IMSM_T_STATE_NORMAL: /* transition to normal state */
7626                 dprintf("normal: ");
7627                 if (is_rebuilding(dev)) {
7628                         dprintf_cont("while rebuilding");
7629                         /* check if recovery is really finished */
7630                         for (mdi = a->info.devs; mdi ; mdi = mdi->next)
7631                                 if (mdi->recovery_start != MaxSector) {
7632                                         recovery_not_finished = 1;
7633                                         break;
7634                                 }
7635                         if (recovery_not_finished) {
7636                                 dprintf_cont("\n");
7637                                 dprintf("Rebuild has not finished yet, state not changed");
7638                                 if (a->last_checkpoint < mdi->recovery_start) {
7639                                         a->last_checkpoint = mdi->recovery_start;
7640                                         super->updates_pending++;
7641                                 }
7642                                 break;
7643                         }
7644                         end_migration(dev, super, map_state);
7645                         map = get_imsm_map(dev, MAP_0);
7646                         map->failed_disk_num = ~0;
7647                         super->updates_pending++;
7648                         a->last_checkpoint = 0;
7649                         break;
7650                 }
7651                 if (is_gen_migration(dev)) {
7652                         dprintf_cont("while general migration");
7653                         if (a->last_checkpoint >= a->info.component_size)
7654                                 end_migration(dev, super, map_state);
7655                         else
7656                                 map->map_state = map_state;
7657                         map = get_imsm_map(dev, MAP_0);
7658                         map->failed_disk_num = ~0;
7659                         super->updates_pending++;
7660                         break;
7661                 }
7662         break;
7663         case IMSM_T_STATE_DEGRADED: /* transition to degraded state */
7664                 dprintf_cont("degraded: ");
7665                 if (map->map_state != map_state && !dev->vol.migr_state) {
7666                         dprintf_cont("mark degraded");
7667                         map->map_state = map_state;
7668                         super->updates_pending++;
7669                         a->last_checkpoint = 0;
7670                         break;
7671                 }
7672                 if (is_rebuilding(dev)) {
7673                         dprintf_cont("while rebuilding.");
7674                         if (map->map_state != map_state)  {
7675                                 dprintf_cont(" Map state change");
7676                                 end_migration(dev, super, map_state);
7677                                 super->updates_pending++;
7678                         }
7679                         break;
7680                 }
7681                 if (is_gen_migration(dev)) {
7682                         dprintf_cont("while general migration");
7683                         if (a->last_checkpoint >= a->info.component_size)
7684                                 end_migration(dev, super, map_state);
7685                         else {
7686                                 map->map_state = map_state;
7687                                 manage_second_map(super, dev);
7688                         }
7689                         super->updates_pending++;
7690                         break;
7691                 }
7692                 if (is_initializing(dev)) {
7693                         dprintf_cont("while initialization.");
7694                         map->map_state = map_state;
7695                         super->updates_pending++;
7696                         break;
7697                 }
7698         break;
7699         case IMSM_T_STATE_FAILED: /* transition to failed state */
7700                 dprintf_cont("failed: ");
7701                 if (is_gen_migration(dev)) {
7702                         dprintf_cont("while general migration");
7703                         map->map_state = map_state;
7704                         super->updates_pending++;
7705                         break;
7706                 }
7707                 if (map->map_state != map_state) {
7708                         dprintf_cont("mark failed");
7709                         end_migration(dev, super, map_state);
7710                         super->updates_pending++;
7711                         a->last_checkpoint = 0;
7712                         break;
7713                 }
7714         break;
7715         default:
7716                 dprintf_cont("state %i\n", map_state);
7717         }
7718         dprintf_cont("\n");
7719 }
7720
7721 static int store_imsm_mpb(int fd, struct imsm_super *mpb)
7722 {
7723         void *buf = mpb;
7724         __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
7725         unsigned long long dsize;
7726         unsigned long long sectors;
7727         unsigned int sector_size;
7728
7729         get_dev_sector_size(fd, NULL, &sector_size);
7730         get_dev_size(fd, NULL, &dsize);
7731
7732         if (mpb_size > sector_size) {
7733                 /* -1 to account for anchor */
7734                 sectors = mpb_sectors(mpb, sector_size) - 1;
7735
7736                 /* write the extended mpb to the sectors preceeding the anchor */
7737                 if (lseek64(fd, dsize - (sector_size * (2 + sectors)),
7738                    SEEK_SET) < 0)
7739                         return 1;
7740
7741                 if ((unsigned long long)write(fd, buf + sector_size,
7742                    sector_size * sectors) != sector_size * sectors)
7743                         return 1;
7744         }
7745
7746         /* first block is stored on second to last sector of the disk */
7747         if (lseek64(fd, dsize - (sector_size * 2), SEEK_SET) < 0)
7748                 return 1;
7749
7750         if (write(fd, buf, sector_size) != sector_size)
7751                 return 1;
7752
7753         return 0;
7754 }
7755
7756 static void imsm_sync_metadata(struct supertype *container)
7757 {
7758         struct intel_super *super = container->sb;
7759
7760         dprintf("sync metadata: %d\n", super->updates_pending);
7761         if (!super->updates_pending)
7762                 return;
7763
7764         write_super_imsm(container, 0);
7765
7766         super->updates_pending = 0;
7767 }
7768
7769 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
7770 {
7771         struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
7772         int i = get_imsm_disk_idx(dev, idx, MAP_X);
7773         struct dl *dl;
7774
7775         for (dl = super->disks; dl; dl = dl->next)
7776                 if (dl->index == i)
7777                         break;
7778
7779         if (dl && is_failed(&dl->disk))
7780                 dl = NULL;
7781
7782         if (dl)
7783                 dprintf("found %x:%x\n", dl->major, dl->minor);
7784
7785         return dl;
7786 }
7787
7788 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
7789                                  struct active_array *a, int activate_new,
7790                                  struct mdinfo *additional_test_list)
7791 {
7792         struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
7793         int idx = get_imsm_disk_idx(dev, slot, MAP_X);
7794         struct imsm_super *mpb = super->anchor;
7795         struct imsm_map *map;
7796         unsigned long long pos;
7797         struct mdinfo *d;
7798         struct extent *ex;
7799         int i, j;
7800         int found;
7801         __u32 array_start = 0;
7802         __u32 array_end = 0;
7803         struct dl *dl;
7804         struct mdinfo *test_list;
7805
7806         for (dl = super->disks; dl; dl = dl->next) {
7807                 /* If in this array, skip */
7808                 for (d = a->info.devs ; d ; d = d->next)
7809                         if (d->state_fd >= 0 &&
7810                             d->disk.major == dl->major &&
7811                             d->disk.minor == dl->minor) {
7812                                 dprintf("%x:%x already in array\n",
7813                                         dl->major, dl->minor);
7814                                 break;
7815                         }
7816                 if (d)
7817                         continue;
7818                 test_list = additional_test_list;
7819                 while (test_list) {
7820                         if (test_list->disk.major == dl->major &&
7821                             test_list->disk.minor == dl->minor) {
7822                                 dprintf("%x:%x already in additional test list\n",
7823                                         dl->major, dl->minor);
7824                                 break;
7825                         }
7826                         test_list = test_list->next;
7827                 }
7828                 if (test_list)
7829                         continue;
7830
7831                 /* skip in use or failed drives */
7832                 if (is_failed(&dl->disk) || idx == dl->index ||
7833                     dl->index == -2) {
7834                         dprintf("%x:%x status (failed: %d index: %d)\n",
7835                                 dl->major, dl->minor, is_failed(&dl->disk), idx);
7836                         continue;
7837                 }
7838
7839                 /* skip pure spares when we are looking for partially
7840                  * assimilated drives
7841                  */
7842                 if (dl->index == -1 && !activate_new)
7843                         continue;
7844
7845                 /* Does this unused device have the requisite free space?
7846                  * It needs to be able to cover all member volumes
7847                  */
7848                 ex = get_extents(super, dl);
7849                 if (!ex) {
7850                         dprintf("cannot get extents\n");
7851                         continue;
7852                 }
7853                 for (i = 0; i < mpb->num_raid_devs; i++) {
7854                         dev = get_imsm_dev(super, i);
7855                         map = get_imsm_map(dev, MAP_0);
7856
7857                         /* check if this disk is already a member of
7858                          * this array
7859                          */
7860                         if (get_imsm_disk_slot(map, dl->index) >= 0)
7861                                 continue;
7862
7863                         found = 0;
7864                         j = 0;
7865                         pos = 0;
7866                         array_start = pba_of_lba0(map);
7867                         array_end = array_start +
7868                                     blocks_per_member(map) - 1;
7869
7870                         do {
7871                                 /* check that we can start at pba_of_lba0 with
7872                                  * blocks_per_member of space
7873                                  */
7874                                 if (array_start >= pos && array_end < ex[j].start) {
7875                                         found = 1;
7876                                         break;
7877                                 }
7878                                 pos = ex[j].start + ex[j].size;
7879                                 j++;
7880                         } while (ex[j-1].size);
7881
7882                         if (!found)
7883                                 break;
7884                 }
7885
7886                 free(ex);
7887                 if (i < mpb->num_raid_devs) {
7888                         dprintf("%x:%x does not have %u to %u available\n",
7889                                 dl->major, dl->minor, array_start, array_end);
7890                         /* No room */
7891                         continue;
7892                 }
7893                 return dl;
7894         }
7895
7896         return dl;
7897 }
7898
7899 static int imsm_rebuild_allowed(struct supertype *cont, int dev_idx, int failed)
7900 {
7901         struct imsm_dev *dev2;
7902         struct imsm_map *map;
7903         struct dl *idisk;
7904         int slot;
7905         int idx;
7906         __u8 state;
7907
7908         dev2 = get_imsm_dev(cont->sb, dev_idx);
7909         if (dev2) {
7910                 state = imsm_check_degraded(cont->sb, dev2, failed, MAP_0);
7911                 if (state == IMSM_T_STATE_FAILED) {
7912                         map = get_imsm_map(dev2, MAP_0);
7913                         if (!map)
7914                                 return 1;
7915                         for (slot = 0; slot < map->num_members; slot++) {
7916                                 /*
7917                                  * Check if failed disks are deleted from intel
7918                                  * disk list or are marked to be deleted
7919                                  */
7920                                 idx = get_imsm_disk_idx(dev2, slot, MAP_X);
7921                                 idisk = get_imsm_dl_disk(cont->sb, idx);
7922                                 /*
7923                                  * Do not rebuild the array if failed disks
7924                                  * from failed sub-array are not removed from
7925                                  * container.
7926                                  */
7927                                 if (idisk &&
7928                                     is_failed(&idisk->disk) &&
7929                                     (idisk->action != DISK_REMOVE))
7930                                         return 0;
7931                         }
7932                 }
7933         }
7934         return 1;
7935 }
7936
7937 static struct mdinfo *imsm_activate_spare(struct active_array *a,
7938                                           struct metadata_update **updates)
7939 {
7940         /**
7941          * Find a device with unused free space and use it to replace a
7942          * failed/vacant region in an array.  We replace failed regions one a
7943          * array at a time.  The result is that a new spare disk will be added
7944          * to the first failed array and after the monitor has finished
7945          * propagating failures the remainder will be consumed.
7946          *
7947          * FIXME add a capability for mdmon to request spares from another
7948          * container.
7949          */
7950
7951         struct intel_super *super = a->container->sb;
7952         int inst = a->info.container_member;
7953         struct imsm_dev *dev = get_imsm_dev(super, inst);
7954         struct imsm_map *map = get_imsm_map(dev, MAP_0);
7955         int failed = a->info.array.raid_disks;
7956         struct mdinfo *rv = NULL;
7957         struct mdinfo *d;
7958         struct mdinfo *di;
7959         struct metadata_update *mu;
7960         struct dl *dl;
7961         struct imsm_update_activate_spare *u;
7962         int num_spares = 0;
7963         int i;
7964         int allowed;
7965
7966         for (d = a->info.devs ; d ; d = d->next) {
7967                 if ((d->curr_state & DS_FAULTY) &&
7968                         d->state_fd >= 0)
7969                         /* wait for Removal to happen */
7970                         return NULL;
7971                 if (d->state_fd >= 0)
7972                         failed--;
7973         }
7974
7975         dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
7976                 inst, failed, a->info.array.raid_disks, a->info.array.level);
7977
7978         if (imsm_reshape_blocks_arrays_changes(super))
7979                         return NULL;
7980
7981         /* Cannot activate another spare if rebuild is in progress already
7982          */
7983         if (is_rebuilding(dev)) {
7984                 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
7985                 return NULL;
7986         }
7987
7988         if (a->info.array.level == 4)
7989                 /* No repair for takeovered array
7990                  * imsm doesn't support raid4
7991                  */
7992                 return NULL;
7993
7994         if (imsm_check_degraded(super, dev, failed, MAP_0) !=
7995                         IMSM_T_STATE_DEGRADED)
7996                 return NULL;
7997
7998         if (get_imsm_map(dev, MAP_0)->map_state == IMSM_T_STATE_UNINITIALIZED) {
7999                 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8000                 return NULL;
8001         }
8002
8003         /*
8004          * If there are any failed disks check state of the other volume.
8005          * Block rebuild if the another one is failed until failed disks
8006          * are removed from container.
8007          */
8008         if (failed) {
8009                 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8010                         MAX_RAID_SERIAL_LEN, dev->volume);
8011                 /* check if states of the other volumes allow for rebuild */
8012                 for (i = 0; i <  super->anchor->num_raid_devs; i++) {
8013                         if (i != inst) {
8014                                 allowed = imsm_rebuild_allowed(a->container,
8015                                                                i, failed);
8016                                 if (!allowed)
8017                                         return NULL;
8018                         }
8019                 }
8020         }
8021
8022         /* For each slot, if it is not working, find a spare */
8023         for (i = 0; i < a->info.array.raid_disks; i++) {
8024                 for (d = a->info.devs ; d ; d = d->next)
8025                         if (d->disk.raid_disk == i)
8026                                 break;
8027                 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
8028                 if (d && (d->state_fd >= 0))
8029                         continue;
8030
8031                 /*
8032                  * OK, this device needs recovery.  Try to re-add the
8033                  * previous occupant of this slot, if this fails see if
8034                  * we can continue the assimilation of a spare that was
8035                  * partially assimilated, finally try to activate a new
8036                  * spare.
8037                  */
8038                 dl = imsm_readd(super, i, a);
8039                 if (!dl)
8040                         dl = imsm_add_spare(super, i, a, 0, rv);
8041                 if (!dl)
8042                         dl = imsm_add_spare(super, i, a, 1, rv);
8043                 if (!dl)
8044                         continue;
8045
8046                 /* found a usable disk with enough space */
8047                 di = xcalloc(1, sizeof(*di));
8048
8049                 /* dl->index will be -1 in the case we are activating a
8050                  * pristine spare.  imsm_process_update() will create a
8051                  * new index in this case.  Once a disk is found to be
8052                  * failed in all member arrays it is kicked from the
8053                  * metadata
8054                  */
8055                 di->disk.number = dl->index;
8056
8057                 /* (ab)use di->devs to store a pointer to the device
8058                  * we chose
8059                  */
8060                 di->devs = (struct mdinfo *) dl;
8061
8062                 di->disk.raid_disk = i;
8063                 di->disk.major = dl->major;
8064                 di->disk.minor = dl->minor;
8065                 di->disk.state = 0;
8066                 di->recovery_start = 0;
8067                 di->data_offset = pba_of_lba0(map);
8068                 di->component_size = a->info.component_size;
8069                 di->container_member = inst;
8070                 super->random = random32();
8071                 di->next = rv;
8072                 rv = di;
8073                 num_spares++;
8074                 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
8075                         i, di->data_offset);
8076         }
8077
8078         if (!rv)
8079                 /* No spares found */
8080                 return rv;
8081         /* Now 'rv' has a list of devices to return.
8082          * Create a metadata_update record to update the
8083          * disk_ord_tbl for the array
8084          */
8085         mu = xmalloc(sizeof(*mu));
8086         mu->buf = xcalloc(num_spares,
8087                           sizeof(struct imsm_update_activate_spare));
8088         mu->space = NULL;
8089         mu->space_list = NULL;
8090         mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
8091         mu->next = *updates;
8092         u = (struct imsm_update_activate_spare *) mu->buf;
8093
8094         for (di = rv ; di ; di = di->next) {
8095                 u->type = update_activate_spare;
8096                 u->dl = (struct dl *) di->devs;
8097                 di->devs = NULL;
8098                 u->slot = di->disk.raid_disk;
8099                 u->array = inst;
8100                 u->next = u + 1;
8101                 u++;
8102         }
8103         (u-1)->next = NULL;
8104         *updates = mu;
8105
8106         return rv;
8107 }
8108
8109 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
8110 {
8111         struct imsm_dev *dev = get_imsm_dev(super, idx);
8112         struct imsm_map *map = get_imsm_map(dev, MAP_0);
8113         struct imsm_map *new_map = get_imsm_map(&u->dev, MAP_0);
8114         struct disk_info *inf = get_disk_info(u);
8115         struct imsm_disk *disk;
8116         int i;
8117         int j;
8118
8119         for (i = 0; i < map->num_members; i++) {
8120                 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i, MAP_X));
8121                 for (j = 0; j < new_map->num_members; j++)
8122                         if (serialcmp(disk->serial, inf[j].serial) == 0)
8123                                 return 1;
8124         }
8125
8126         return 0;
8127 }
8128
8129 static struct dl *get_disk_super(struct intel_super *super, int major, int minor)
8130 {
8131         struct dl *dl;
8132
8133         for (dl = super->disks; dl; dl = dl->next)
8134                 if (dl->major == major &&  dl->minor == minor)
8135                         return dl;
8136         return NULL;
8137 }
8138
8139 static int remove_disk_super(struct intel_super *super, int major, int minor)
8140 {
8141         struct dl *prev;
8142         struct dl *dl;
8143
8144         prev = NULL;
8145         for (dl = super->disks; dl; dl = dl->next) {
8146                 if (dl->major == major && dl->minor == minor) {
8147                         /* remove */
8148                         if (prev)
8149                                 prev->next = dl->next;
8150                         else
8151                                 super->disks = dl->next;
8152                         dl->next = NULL;
8153                         __free_imsm_disk(dl);
8154                         dprintf("removed %x:%x\n", major, minor);
8155                         break;
8156                 }
8157                 prev = dl;
8158         }
8159         return 0;
8160 }
8161
8162 static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index);
8163
8164 static int add_remove_disk_update(struct intel_super *super)
8165 {
8166         int check_degraded = 0;
8167         struct dl *disk;
8168
8169         /* add/remove some spares to/from the metadata/contrainer */
8170         while (super->disk_mgmt_list) {
8171                 struct dl *disk_cfg;
8172
8173                 disk_cfg = super->disk_mgmt_list;
8174                 super->disk_mgmt_list = disk_cfg->next;
8175                 disk_cfg->next = NULL;
8176
8177                 if (disk_cfg->action == DISK_ADD) {
8178                         disk_cfg->next = super->disks;
8179                         super->disks = disk_cfg;
8180                         check_degraded = 1;
8181                         dprintf("added %x:%x\n",
8182                                 disk_cfg->major, disk_cfg->minor);
8183                 } else if (disk_cfg->action == DISK_REMOVE) {
8184                         dprintf("Disk remove action processed: %x.%x\n",
8185                                 disk_cfg->major, disk_cfg->minor);
8186                         disk = get_disk_super(super,
8187                                               disk_cfg->major,
8188                                               disk_cfg->minor);
8189                         if (disk) {
8190                                 /* store action status */
8191                                 disk->action = DISK_REMOVE;
8192                                 /* remove spare disks only */
8193                                 if (disk->index == -1) {
8194                                         remove_disk_super(super,
8195                                                           disk_cfg->major,
8196                                                           disk_cfg->minor);
8197                                 }
8198                         }
8199                         /* release allocate disk structure */
8200                         __free_imsm_disk(disk_cfg);
8201                 }
8202         }
8203         return check_degraded;
8204 }
8205
8206 static int apply_reshape_migration_update(struct imsm_update_reshape_migration *u,
8207                                                 struct intel_super *super,
8208                                                 void ***space_list)
8209 {
8210         struct intel_dev *id;
8211         void **tofree = NULL;
8212         int ret_val = 0;
8213
8214         dprintf("(enter)\n");
8215         if (u->subdev < 0 || u->subdev > 1) {
8216                 dprintf("imsm: Error: Wrong subdev: %i\n", u->subdev);
8217                 return ret_val;
8218         }
8219         if (space_list == NULL || *space_list == NULL) {
8220                 dprintf("imsm: Error: Memory is not allocated\n");
8221                 return ret_val;
8222         }
8223
8224         for (id = super->devlist ; id; id = id->next) {
8225                 if (id->index == (unsigned)u->subdev) {
8226                         struct imsm_dev *dev = get_imsm_dev(super, u->subdev);
8227                         struct imsm_map *map;
8228                         struct imsm_dev *new_dev =
8229                                 (struct imsm_dev *)*space_list;
8230                         struct imsm_map *migr_map = get_imsm_map(dev, MAP_1);
8231                         int to_state;
8232                         struct dl *new_disk;
8233
8234                         if (new_dev == NULL)
8235                                 return ret_val;
8236                         *space_list = **space_list;
8237                         memcpy(new_dev, dev, sizeof_imsm_dev(dev, 0));
8238                         map = get_imsm_map(new_dev, MAP_0);
8239                         if (migr_map) {
8240                                 dprintf("imsm: Error: migration in progress");
8241                                 return ret_val;
8242                         }
8243
8244                         to_state = map->map_state;
8245                         if ((u->new_level == 5) && (map->raid_level == 0)) {
8246                                 map->num_members++;
8247                                 /* this should not happen */
8248                                 if (u->new_disks[0] < 0) {
8249                                         map->failed_disk_num =
8250                                                 map->num_members - 1;
8251                                         to_state = IMSM_T_STATE_DEGRADED;
8252                                 } else
8253                                         to_state = IMSM_T_STATE_NORMAL;
8254                         }
8255                         migrate(new_dev, super, to_state, MIGR_GEN_MIGR);
8256                         if (u->new_level > -1)
8257                                 map->raid_level = u->new_level;
8258                         migr_map = get_imsm_map(new_dev, MAP_1);
8259                         if ((u->new_level == 5) &&
8260                             (migr_map->raid_level == 0)) {
8261                                 int ord = map->num_members - 1;
8262                                 migr_map->num_members--;
8263                                 if (u->new_disks[0] < 0)
8264                                         ord |= IMSM_ORD_REBUILD;
8265                                 set_imsm_ord_tbl_ent(map,
8266                                                      map->num_members - 1,
8267                                                      ord);
8268                         }
8269                         id->dev = new_dev;
8270                         tofree = (void **)dev;
8271
8272                         /* update chunk size
8273                          */
8274                         if (u->new_chunksize > 0)
8275                                 map->blocks_per_strip =
8276                                         __cpu_to_le16(u->new_chunksize * 2);
8277
8278                         /* add disk
8279                          */
8280                         if (u->new_level != 5 || migr_map->raid_level != 0 ||
8281                             migr_map->raid_level == map->raid_level)
8282                                 goto skip_disk_add;
8283
8284                         if (u->new_disks[0] >= 0) {
8285                                 /* use passes spare
8286                                  */
8287                                 new_disk = get_disk_super(super,
8288                                                         major(u->new_disks[0]),
8289                                                         minor(u->new_disks[0]));
8290                                 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8291                                         major(u->new_disks[0]),
8292                                         minor(u->new_disks[0]),
8293                                         new_disk, new_disk->index);
8294                                 if (new_disk == NULL)
8295                                         goto error_disk_add;
8296
8297                                 new_disk->index = map->num_members - 1;
8298                                 /* slot to fill in autolayout
8299                                  */
8300                                 new_disk->raiddisk = new_disk->index;
8301                                 new_disk->disk.status |= CONFIGURED_DISK;
8302                                 new_disk->disk.status &= ~SPARE_DISK;
8303                         } else
8304                                 goto error_disk_add;
8305
8306 skip_disk_add:
8307                         *tofree = *space_list;
8308                         /* calculate new size
8309                          */
8310                         imsm_set_array_size(new_dev, -1);
8311
8312                         ret_val = 1;
8313                 }
8314         }
8315
8316         if (tofree)
8317                 *space_list = tofree;
8318         return ret_val;
8319
8320 error_disk_add:
8321         dprintf("Error: imsm: Cannot find disk.\n");
8322         return ret_val;
8323 }
8324
8325 static int apply_size_change_update(struct imsm_update_size_change *u,
8326                 struct intel_super *super)
8327 {
8328         struct intel_dev *id;
8329         int ret_val = 0;
8330
8331         dprintf("(enter)\n");
8332         if (u->subdev < 0 || u->subdev > 1) {
8333                 dprintf("imsm: Error: Wrong subdev: %i\n", u->subdev);
8334                 return ret_val;
8335         }
8336
8337         for (id = super->devlist ; id; id = id->next) {
8338                 if (id->index == (unsigned)u->subdev) {
8339                         struct imsm_dev *dev = get_imsm_dev(super, u->subdev);
8340                         struct imsm_map *map = get_imsm_map(dev, MAP_0);
8341                         int used_disks = imsm_num_data_members(dev, MAP_0);
8342                         unsigned long long blocks_per_member;
8343
8344                         /* calculate new size
8345                          */
8346                         blocks_per_member = u->new_size / used_disks;
8347                         dprintf("(size: %llu, blocks per member: %llu)\n",
8348                                 u->new_size, blocks_per_member);
8349                         set_blocks_per_member(map, blocks_per_member);
8350                         imsm_set_array_size(dev, u->new_size);
8351
8352                         ret_val = 1;
8353                         break;
8354                 }
8355         }
8356
8357         return ret_val;
8358 }
8359
8360 static int apply_update_activate_spare(struct imsm_update_activate_spare *u,
8361                                        struct intel_super *super,
8362                                        struct active_array *active_array)
8363 {
8364         struct imsm_super *mpb = super->anchor;
8365         struct imsm_dev *dev = get_imsm_dev(super, u->array);
8366         struct imsm_map *map = get_imsm_map(dev, MAP_0);
8367         struct imsm_map *migr_map;
8368         struct active_array *a;
8369         struct imsm_disk *disk;
8370         __u8 to_state;
8371         struct dl *dl;
8372         unsigned int found;
8373         int failed;
8374         int victim;
8375         int i;
8376         int second_map_created = 0;
8377
8378         for (; u; u = u->next) {
8379                 victim = get_imsm_disk_idx(dev, u->slot, MAP_X);
8380
8381                 if (victim < 0)
8382                         return 0;
8383
8384                 for (dl = super->disks; dl; dl = dl->next)
8385                         if (dl == u->dl)
8386                                 break;
8387
8388                 if (!dl) {
8389                         pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
8390                                 u->dl->index);
8391                         return 0;
8392                 }
8393
8394                 /* count failures (excluding rebuilds and the victim)
8395                  * to determine map[0] state
8396                  */
8397                 failed = 0;
8398                 for (i = 0; i < map->num_members; i++) {
8399                         if (i == u->slot)
8400                                 continue;
8401                         disk = get_imsm_disk(super,
8402                                              get_imsm_disk_idx(dev, i, MAP_X));
8403                         if (!disk || is_failed(disk))
8404                                 failed++;
8405                 }
8406
8407                 /* adding a pristine spare, assign a new index */
8408                 if (dl->index < 0) {
8409                         dl->index = super->anchor->num_disks;
8410                         super->anchor->num_disks++;
8411                 }
8412                 disk = &dl->disk;
8413                 disk->status |= CONFIGURED_DISK;
8414                 disk->status &= ~SPARE_DISK;
8415
8416                 /* mark rebuild */
8417                 to_state = imsm_check_degraded(super, dev, failed, MAP_0);
8418                 if (!second_map_created) {
8419                         second_map_created = 1;
8420                         map->map_state = IMSM_T_STATE_DEGRADED;
8421                         migrate(dev, super, to_state, MIGR_REBUILD);
8422                 } else
8423                         map->map_state = to_state;
8424                 migr_map = get_imsm_map(dev, MAP_1);
8425                 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
8426                 set_imsm_ord_tbl_ent(migr_map, u->slot,
8427                                      dl->index | IMSM_ORD_REBUILD);
8428
8429                 /* update the family_num to mark a new container
8430                  * generation, being careful to record the existing
8431                  * family_num in orig_family_num to clean up after
8432                  * earlier mdadm versions that neglected to set it.
8433                  */
8434                 if (mpb->orig_family_num == 0)
8435                         mpb->orig_family_num = mpb->family_num;
8436                 mpb->family_num += super->random;
8437
8438                 /* count arrays using the victim in the metadata */
8439                 found = 0;
8440                 for (a = active_array; a ; a = a->next) {
8441                         dev = get_imsm_dev(super, a->info.container_member);
8442                         map = get_imsm_map(dev, MAP_0);
8443
8444                         if (get_imsm_disk_slot(map, victim) >= 0)
8445                                 found++;
8446                 }
8447
8448                 /* delete the victim if it is no longer being
8449                  * utilized anywhere
8450                  */
8451                 if (!found) {
8452                         struct dl **dlp;
8453
8454                         /* We know that 'manager' isn't touching anything,
8455                          * so it is safe to delete
8456                          */
8457                         for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
8458                                 if ((*dlp)->index == victim)
8459                                         break;
8460
8461                         /* victim may be on the missing list */
8462                         if (!*dlp)
8463                                 for (dlp = &super->missing; *dlp;
8464                                      dlp = &(*dlp)->next)
8465                                         if ((*dlp)->index == victim)
8466                                                 break;
8467                         imsm_delete(super, dlp, victim);
8468                 }
8469         }
8470
8471         return 1;
8472 }
8473
8474 static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
8475                                                 struct intel_super *super,
8476                                                 void ***space_list)
8477 {
8478         struct dl *new_disk;
8479         struct intel_dev *id;
8480         int i;
8481         int delta_disks = u->new_raid_disks - u->old_raid_disks;
8482         int disk_count = u->old_raid_disks;
8483         void **tofree = NULL;
8484         int devices_to_reshape = 1;
8485         struct imsm_super *mpb = super->anchor;
8486         int ret_val = 0;
8487         unsigned int dev_id;
8488
8489         dprintf("(enter)\n");
8490
8491         /* enable spares to use in array */
8492         for (i = 0; i < delta_disks; i++) {
8493                 new_disk = get_disk_super(super,
8494                                           major(u->new_disks[i]),
8495                                           minor(u->new_disks[i]));
8496                 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8497                         major(u->new_disks[i]), minor(u->new_disks[i]),
8498                         new_disk, new_disk->index);
8499                 if (new_disk == NULL ||
8500                     (new_disk->index >= 0 &&
8501                      new_disk->index < u->old_raid_disks))
8502                         goto update_reshape_exit;
8503                 new_disk->index = disk_count++;
8504                 /* slot to fill in autolayout
8505                  */
8506                 new_disk->raiddisk = new_disk->index;
8507                 new_disk->disk.status |=
8508                         CONFIGURED_DISK;
8509                 new_disk->disk.status &= ~SPARE_DISK;
8510         }
8511
8512         dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
8513                 mpb->num_raid_devs);
8514         /* manage changes in volume
8515          */
8516         for (dev_id = 0; dev_id < mpb->num_raid_devs; dev_id++) {
8517                 void **sp = *space_list;
8518                 struct imsm_dev *newdev;
8519                 struct imsm_map *newmap, *oldmap;
8520
8521                 for (id = super->devlist ; id; id = id->next) {
8522                         if (id->index == dev_id)
8523                                 break;
8524                 }
8525                 if (id == NULL)
8526                         break;
8527                 if (!sp)
8528                         continue;
8529                 *space_list = *sp;
8530                 newdev = (void*)sp;
8531                 /* Copy the dev, but not (all of) the map */
8532                 memcpy(newdev, id->dev, sizeof(*newdev));
8533                 oldmap = get_imsm_map(id->dev, MAP_0);
8534                 newmap = get_imsm_map(newdev, MAP_0);
8535                 /* Copy the current map */
8536                 memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
8537                 /* update one device only
8538                  */
8539                 if (devices_to_reshape) {
8540                         dprintf("imsm: modifying subdev: %i\n",
8541                                 id->index);
8542                         devices_to_reshape--;
8543                         newdev->vol.migr_state = 1;
8544                         newdev->vol.curr_migr_unit = 0;
8545                         set_migr_type(newdev, MIGR_GEN_MIGR);
8546                         newmap->num_members = u->new_raid_disks;
8547                         for (i = 0; i < delta_disks; i++) {
8548                                 set_imsm_ord_tbl_ent(newmap,
8549                                                      u->old_raid_disks + i,
8550                                                      u->old_raid_disks + i);
8551                         }
8552                         /* New map is correct, now need to save old map
8553                          */
8554                         newmap = get_imsm_map(newdev, MAP_1);
8555                         memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
8556
8557                         imsm_set_array_size(newdev, -1);
8558                 }
8559
8560                 sp = (void **)id->dev;
8561                 id->dev = newdev;
8562                 *sp = tofree;
8563                 tofree = sp;
8564
8565                 /* Clear migration record */
8566                 memset(super->migr_rec, 0, sizeof(struct migr_record));
8567         }
8568         if (tofree)
8569                 *space_list = tofree;
8570         ret_val = 1;
8571
8572 update_reshape_exit:
8573
8574         return ret_val;
8575 }
8576
8577 static int apply_takeover_update(struct imsm_update_takeover *u,
8578                                  struct intel_super *super,
8579                                  void ***space_list)
8580 {
8581         struct imsm_dev *dev = NULL;
8582         struct intel_dev *dv;
8583         struct imsm_dev *dev_new;
8584         struct imsm_map *map;
8585         struct dl *dm, *du;
8586         int i;
8587
8588         for (dv = super->devlist; dv; dv = dv->next)
8589                 if (dv->index == (unsigned int)u->subarray) {
8590                         dev = dv->dev;
8591                         break;
8592                 }
8593
8594         if (dev == NULL)
8595                 return 0;
8596
8597         map = get_imsm_map(dev, MAP_0);
8598
8599         if (u->direction == R10_TO_R0) {
8600                 /* Number of failed disks must be half of initial disk number */
8601                 if (imsm_count_failed(super, dev, MAP_0) !=
8602                                 (map->num_members / 2))
8603                         return 0;
8604
8605                 /* iterate through devices to mark removed disks as spare */
8606                 for (dm = super->disks; dm; dm = dm->next) {
8607                         if (dm->disk.status & FAILED_DISK) {
8608                                 int idx = dm->index;
8609                                 /* update indexes on the disk list */
8610 /* FIXME this loop-with-the-loop looks wrong,  I'm not convinced
8611    the index values will end up being correct.... NB */
8612                                 for (du = super->disks; du; du = du->next)
8613                                         if (du->index > idx)
8614                                                 du->index--;
8615                                 /* mark as spare disk */
8616                                 mark_spare(dm);
8617                         }
8618                 }
8619                 /* update map */
8620                 map->num_members = map->num_members / 2;
8621                 map->map_state = IMSM_T_STATE_NORMAL;
8622                 map->num_domains = 1;
8623                 map->raid_level = 0;
8624                 map->failed_disk_num = -1;
8625         }
8626
8627         if (u->direction == R0_TO_R10) {
8628                 void **space;
8629                 /* update slots in current disk list */
8630                 for (dm = super->disks; dm; dm = dm->next) {
8631                         if (dm->index >= 0)
8632                                 dm->index *= 2;
8633                 }
8634                 /* create new *missing* disks */
8635                 for (i = 0; i < map->num_members; i++) {
8636                         space = *space_list;
8637                         if (!space)
8638                                 continue;
8639                         *space_list = *space;
8640                         du = (void *)space;
8641                         memcpy(du, super->disks, sizeof(*du));
8642                         du->fd = -1;
8643                         du->minor = 0;
8644                         du->major = 0;
8645                         du->index = (i * 2) + 1;
8646                         sprintf((char *)du->disk.serial,
8647                                 " MISSING_%d", du->index);
8648                         sprintf((char *)du->serial,
8649                                 "MISSING_%d", du->index);
8650                         du->next = super->missing;
8651                         super->missing = du;
8652                 }
8653                 /* create new dev and map */
8654                 space = *space_list;
8655                 if (!space)
8656                         return 0;
8657                 *space_list = *space;
8658                 dev_new = (void *)space;
8659                 memcpy(dev_new, dev, sizeof(*dev));
8660                 /* update new map */
8661                 map = get_imsm_map(dev_new, MAP_0);
8662                 map->num_members = map->num_members * 2;
8663                 map->map_state = IMSM_T_STATE_DEGRADED;
8664                 map->num_domains = 2;
8665                 map->raid_level = 1;
8666                 /* replace dev<->dev_new */
8667                 dv->dev = dev_new;
8668         }
8669         /* update disk order table */
8670         for (du = super->disks; du; du = du->next)
8671                 if (du->index >= 0)
8672                         set_imsm_ord_tbl_ent(map, du->index, du->index);
8673         for (du = super->missing; du; du = du->next)
8674                 if (du->index >= 0) {
8675                         set_imsm_ord_tbl_ent(map, du->index, du->index);
8676                         mark_missing(dv->dev, &du->disk, du->index);
8677                 }
8678
8679         return 1;
8680 }
8681
8682 static void imsm_process_update(struct supertype *st,
8683                                 struct metadata_update *update)
8684 {
8685         /**
8686          * crack open the metadata_update envelope to find the update record
8687          * update can be one of:
8688          *    update_reshape_container_disks - all the arrays in the container
8689          *      are being reshaped to have more devices.  We need to mark
8690          *      the arrays for general migration and convert selected spares
8691          *      into active devices.
8692          *    update_activate_spare - a spare device has replaced a failed
8693          *      device in an array, update the disk_ord_tbl.  If this disk is
8694          *      present in all member arrays then also clear the SPARE_DISK
8695          *      flag
8696          *    update_create_array
8697          *    update_kill_array
8698          *    update_rename_array
8699          *    update_add_remove_disk
8700          */
8701         struct intel_super *super = st->sb;
8702         struct imsm_super *mpb;
8703         enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
8704
8705         /* update requires a larger buf but the allocation failed */
8706         if (super->next_len && !super->next_buf) {
8707                 super->next_len = 0;
8708                 return;
8709         }
8710
8711         if (super->next_buf) {
8712                 memcpy(super->next_buf, super->buf, super->len);
8713                 free(super->buf);
8714                 super->len = super->next_len;
8715                 super->buf = super->next_buf;
8716
8717                 super->next_len = 0;
8718                 super->next_buf = NULL;
8719         }
8720
8721         mpb = super->anchor;
8722
8723         switch (type) {
8724         case update_general_migration_checkpoint: {
8725                 struct intel_dev *id;
8726                 struct imsm_update_general_migration_checkpoint *u =
8727                                                         (void *)update->buf;
8728
8729                 dprintf("called for update_general_migration_checkpoint\n");
8730
8731                 /* find device under general migration */
8732                 for (id = super->devlist ; id; id = id->next) {
8733                         if (is_gen_migration(id->dev)) {
8734                                 id->dev->vol.curr_migr_unit =
8735                                         __cpu_to_le32(u->curr_migr_unit);
8736                                 super->updates_pending++;
8737                         }
8738                 }
8739                 break;
8740         }
8741         case update_takeover: {
8742                 struct imsm_update_takeover *u = (void *)update->buf;
8743                 if (apply_takeover_update(u, super, &update->space_list)) {
8744                         imsm_update_version_info(super);
8745                         super->updates_pending++;
8746                 }
8747                 break;
8748         }
8749
8750         case update_reshape_container_disks: {
8751                 struct imsm_update_reshape *u = (void *)update->buf;
8752                 if (apply_reshape_container_disks_update(
8753                             u, super, &update->space_list))
8754                         super->updates_pending++;
8755                 break;
8756         }
8757         case update_reshape_migration: {
8758                 struct imsm_update_reshape_migration *u = (void *)update->buf;
8759                 if (apply_reshape_migration_update(
8760                             u, super, &update->space_list))
8761                         super->updates_pending++;
8762                 break;
8763         }
8764         case update_size_change: {
8765                 struct imsm_update_size_change *u = (void *)update->buf;
8766                 if (apply_size_change_update(u, super))
8767                         super->updates_pending++;
8768                 break;
8769         }
8770         case update_activate_spare: {
8771                 struct imsm_update_activate_spare *u = (void *) update->buf;
8772                 if (apply_update_activate_spare(u, super, st->arrays))
8773                         super->updates_pending++;
8774                 break;
8775         }
8776         case update_create_array: {
8777                 /* someone wants to create a new array, we need to be aware of
8778                  * a few races/collisions:
8779                  * 1/ 'Create' called by two separate instances of mdadm
8780                  * 2/ 'Create' versus 'activate_spare': mdadm has chosen
8781                  *     devices that have since been assimilated via
8782                  *     activate_spare.
8783                  * In the event this update can not be carried out mdadm will
8784                  * (FIX ME) notice that its update did not take hold.
8785                  */
8786                 struct imsm_update_create_array *u = (void *) update->buf;
8787                 struct intel_dev *dv;
8788                 struct imsm_dev *dev;
8789                 struct imsm_map *map, *new_map;
8790                 unsigned long long start, end;
8791                 unsigned long long new_start, new_end;
8792                 int i;
8793                 struct disk_info *inf;
8794                 struct dl *dl;
8795
8796                 /* handle racing creates: first come first serve */
8797                 if (u->dev_idx < mpb->num_raid_devs) {
8798                         dprintf("subarray %d already defined\n", u->dev_idx);
8799                         goto create_error;
8800                 }
8801
8802                 /* check update is next in sequence */
8803                 if (u->dev_idx != mpb->num_raid_devs) {
8804                         dprintf("can not create array %d expected index %d\n",
8805                                 u->dev_idx, mpb->num_raid_devs);
8806                         goto create_error;
8807                 }
8808
8809                 new_map = get_imsm_map(&u->dev, MAP_0);
8810                 new_start = pba_of_lba0(new_map);
8811                 new_end = new_start + blocks_per_member(new_map);
8812                 inf = get_disk_info(u);
8813
8814                 /* handle activate_spare versus create race:
8815                  * check to make sure that overlapping arrays do not include
8816                  * overalpping disks
8817                  */
8818                 for (i = 0; i < mpb->num_raid_devs; i++) {
8819                         dev = get_imsm_dev(super, i);
8820                         map = get_imsm_map(dev, MAP_0);
8821                         start = pba_of_lba0(map);
8822                         end = start + blocks_per_member(map);
8823                         if ((new_start >= start && new_start <= end) ||
8824                             (start >= new_start && start <= new_end))
8825                                 /* overlap */;
8826                         else
8827                                 continue;
8828
8829                         if (disks_overlap(super, i, u)) {
8830                                 dprintf("arrays overlap\n");
8831                                 goto create_error;
8832                         }
8833                 }
8834
8835                 /* check that prepare update was successful */
8836                 if (!update->space) {
8837                         dprintf("prepare update failed\n");
8838                         goto create_error;
8839                 }
8840
8841                 /* check that all disks are still active before committing
8842                  * changes.  FIXME: could we instead handle this by creating a
8843                  * degraded array?  That's probably not what the user expects,
8844                  * so better to drop this update on the floor.
8845                  */
8846                 for (i = 0; i < new_map->num_members; i++) {
8847                         dl = serial_to_dl(inf[i].serial, super);
8848                         if (!dl) {
8849                                 dprintf("disk disappeared\n");
8850                                 goto create_error;
8851                         }
8852                 }
8853
8854                 super->updates_pending++;
8855
8856                 /* convert spares to members and fixup ord_tbl */
8857                 for (i = 0; i < new_map->num_members; i++) {
8858                         dl = serial_to_dl(inf[i].serial, super);
8859                         if (dl->index == -1) {
8860                                 dl->index = mpb->num_disks;
8861                                 mpb->num_disks++;
8862                                 dl->disk.status |= CONFIGURED_DISK;
8863                                 dl->disk.status &= ~SPARE_DISK;
8864                         }
8865                         set_imsm_ord_tbl_ent(new_map, i, dl->index);
8866                 }
8867
8868                 dv = update->space;
8869                 dev = dv->dev;
8870                 update->space = NULL;
8871                 imsm_copy_dev(dev, &u->dev);
8872                 dv->index = u->dev_idx;
8873                 dv->next = super->devlist;
8874                 super->devlist = dv;
8875                 mpb->num_raid_devs++;
8876
8877                 imsm_update_version_info(super);
8878                 break;
8879  create_error:
8880                 /* mdmon knows how to release update->space, but not
8881                  * ((struct intel_dev *) update->space)->dev
8882                  */
8883                 if (update->space) {
8884                         dv = update->space;
8885                         free(dv->dev);
8886                 }
8887                 break;
8888         }
8889         case update_kill_array: {
8890                 struct imsm_update_kill_array *u = (void *) update->buf;
8891                 int victim = u->dev_idx;
8892                 struct active_array *a;
8893                 struct intel_dev **dp;
8894                 struct imsm_dev *dev;
8895
8896                 /* sanity check that we are not affecting the uuid of
8897                  * active arrays, or deleting an active array
8898                  *
8899                  * FIXME when immutable ids are available, but note that
8900                  * we'll also need to fixup the invalidated/active
8901                  * subarray indexes in mdstat
8902                  */
8903                 for (a = st->arrays; a; a = a->next)
8904                         if (a->info.container_member >= victim)
8905                                 break;
8906                 /* by definition if mdmon is running at least one array
8907                  * is active in the container, so checking
8908                  * mpb->num_raid_devs is just extra paranoia
8909                  */
8910                 dev = get_imsm_dev(super, victim);
8911                 if (a || !dev || mpb->num_raid_devs == 1) {
8912                         dprintf("failed to delete subarray-%d\n", victim);
8913                         break;
8914                 }
8915
8916                 for (dp = &super->devlist; *dp;)
8917                         if ((*dp)->index == (unsigned)super->current_vol) {
8918                                 *dp = (*dp)->next;
8919                         } else {
8920                                 if ((*dp)->index > (unsigned)victim)
8921                                         (*dp)->index--;
8922                                 dp = &(*dp)->next;
8923                         }
8924                 mpb->num_raid_devs--;
8925                 super->updates_pending++;
8926                 break;
8927         }
8928         case update_rename_array: {
8929                 struct imsm_update_rename_array *u = (void *) update->buf;
8930                 char name[MAX_RAID_SERIAL_LEN+1];
8931                 int target = u->dev_idx;
8932                 struct active_array *a;
8933                 struct imsm_dev *dev;
8934
8935                 /* sanity check that we are not affecting the uuid of
8936                  * an active array
8937                  */
8938                 snprintf(name, MAX_RAID_SERIAL_LEN, "%s", (char *) u->name);
8939                 name[MAX_RAID_SERIAL_LEN] = '\0';
8940                 for (a = st->arrays; a; a = a->next)
8941                         if (a->info.container_member == target)
8942                                 break;
8943                 dev = get_imsm_dev(super, u->dev_idx);
8944                 if (a || !dev || !check_name(super, name, 1)) {
8945                         dprintf("failed to rename subarray-%d\n", target);
8946                         break;
8947                 }
8948
8949                 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
8950                 super->updates_pending++;
8951                 break;
8952         }
8953         case update_add_remove_disk: {
8954                 /* we may be able to repair some arrays if disks are
8955                  * being added, check the status of add_remove_disk
8956                  * if discs has been added.
8957                  */
8958                 if (add_remove_disk_update(super)) {
8959                         struct active_array *a;
8960
8961                         super->updates_pending++;
8962                         for (a = st->arrays; a; a = a->next)
8963                                 a->check_degraded = 1;
8964                 }
8965                 break;
8966         }
8967         default:
8968                 pr_err("error: unsuported process update type:(type: %d)\n",    type);
8969         }
8970 }
8971
8972 static struct mdinfo *get_spares_for_grow(struct supertype *st);
8973
8974 static int imsm_prepare_update(struct supertype *st,
8975                                struct metadata_update *update)
8976 {
8977         /**
8978          * Allocate space to hold new disk entries, raid-device entries or a new
8979          * mpb if necessary.  The manager synchronously waits for updates to
8980          * complete in the monitor, so new mpb buffers allocated here can be
8981          * integrated by the monitor thread without worrying about live pointers
8982          * in the manager thread.
8983          */
8984         enum imsm_update_type type;
8985         struct intel_super *super = st->sb;
8986         unsigned int sector_size = super->sector_size;
8987         struct imsm_super *mpb = super->anchor;
8988         size_t buf_len;
8989         size_t len = 0;
8990
8991         if (update->len < (int)sizeof(type))
8992                 return 0;
8993
8994         type = *(enum imsm_update_type *) update->buf;
8995
8996         switch (type) {
8997         case update_general_migration_checkpoint:
8998                 if (update->len < (int)sizeof(struct imsm_update_general_migration_checkpoint))
8999                         return 0;
9000                 dprintf("called for update_general_migration_checkpoint\n");
9001                 break;
9002         case update_takeover: {
9003                 struct imsm_update_takeover *u = (void *)update->buf;
9004                 if (update->len < (int)sizeof(*u))
9005                         return 0;
9006                 if (u->direction == R0_TO_R10) {
9007                         void **tail = (void **)&update->space_list;
9008                         struct imsm_dev *dev = get_imsm_dev(super, u->subarray);
9009                         struct imsm_map *map = get_imsm_map(dev, MAP_0);
9010                         int num_members = map->num_members;
9011                         void *space;
9012                         int size, i;
9013                         /* allocate memory for added disks */
9014                         for (i = 0; i < num_members; i++) {
9015                                 size = sizeof(struct dl);
9016                                 space = xmalloc(size);
9017                                 *tail = space;
9018                                 tail = space;
9019                                 *tail = NULL;
9020                         }
9021                         /* allocate memory for new device */
9022                         size = sizeof_imsm_dev(super->devlist->dev, 0) +
9023                                 (num_members * sizeof(__u32));
9024                         space = xmalloc(size);
9025                         *tail = space;
9026                         tail = space;
9027                         *tail = NULL;
9028                         len = disks_to_mpb_size(num_members * 2);
9029                 }
9030
9031                 break;
9032         }
9033         case update_reshape_container_disks: {
9034                 /* Every raid device in the container is about to
9035                  * gain some more devices, and we will enter a
9036                  * reconfiguration.
9037                  * So each 'imsm_map' will be bigger, and the imsm_vol
9038                  * will now hold 2 of them.
9039                  * Thus we need new 'struct imsm_dev' allocations sized
9040                  * as sizeof_imsm_dev but with more devices in both maps.
9041                  */
9042                 struct imsm_update_reshape *u = (void *)update->buf;
9043                 struct intel_dev *dl;
9044                 void **space_tail = (void**)&update->space_list;
9045
9046                 if (update->len < (int)sizeof(*u))
9047                         return 0;
9048
9049                 dprintf("for update_reshape\n");
9050
9051                 for (dl = super->devlist; dl; dl = dl->next) {
9052                         int size = sizeof_imsm_dev(dl->dev, 1);
9053                         void *s;
9054                         if (u->new_raid_disks > u->old_raid_disks)
9055                                 size += sizeof(__u32)*2*
9056                                         (u->new_raid_disks - u->old_raid_disks);
9057                         s = xmalloc(size);
9058                         *space_tail = s;
9059                         space_tail = s;
9060                         *space_tail = NULL;
9061                 }
9062
9063                 len = disks_to_mpb_size(u->new_raid_disks);
9064                 dprintf("New anchor length is %llu\n", (unsigned long long)len);
9065                 break;
9066         }
9067         case update_reshape_migration: {
9068                 /* for migration level 0->5 we need to add disks
9069                  * so the same as for container operation we will copy
9070                  * device to the bigger location.
9071                  * in memory prepared device and new disk area are prepared
9072                  * for usage in process update
9073                  */
9074                 struct imsm_update_reshape_migration *u = (void *)update->buf;
9075                 struct intel_dev *id;
9076                 void **space_tail = (void **)&update->space_list;
9077                 int size;
9078                 void *s;
9079                 int current_level = -1;
9080
9081                 if (update->len < (int)sizeof(*u))
9082                         return 0;
9083
9084                 dprintf("for update_reshape\n");
9085
9086                 /* add space for bigger array in update
9087                  */
9088                 for (id = super->devlist; id; id = id->next) {
9089                         if (id->index == (unsigned)u->subdev) {
9090                                 size = sizeof_imsm_dev(id->dev, 1);
9091                                 if (u->new_raid_disks > u->old_raid_disks)
9092                                         size += sizeof(__u32)*2*
9093                                         (u->new_raid_disks - u->old_raid_disks);
9094                                 s = xmalloc(size);
9095                                 *space_tail = s;
9096                                 space_tail = s;
9097                                 *space_tail = NULL;
9098                                 break;
9099                         }
9100                 }
9101                 if (update->space_list == NULL)
9102                         break;
9103
9104                 /* add space for disk in update
9105                  */
9106                 size = sizeof(struct dl);
9107                 s = xmalloc(size);
9108                 *space_tail = s;
9109                 space_tail = s;
9110                 *space_tail = NULL;
9111
9112                 /* add spare device to update
9113                  */
9114                 for (id = super->devlist ; id; id = id->next)
9115                         if (id->index == (unsigned)u->subdev) {
9116                                 struct imsm_dev *dev;
9117                                 struct imsm_map *map;
9118
9119                                 dev = get_imsm_dev(super, u->subdev);
9120                                 map = get_imsm_map(dev, MAP_0);
9121                                 current_level = map->raid_level;
9122                                 break;
9123                         }
9124                 if (u->new_level == 5 && u->new_level != current_level) {
9125                         struct mdinfo *spares;
9126
9127                         spares = get_spares_for_grow(st);
9128                         if (spares) {
9129                                 struct dl *dl;
9130                                 struct mdinfo *dev;
9131
9132                                 dev = spares->devs;
9133                                 if (dev) {
9134                                         u->new_disks[0] =
9135                                                 makedev(dev->disk.major,
9136                                                         dev->disk.minor);
9137                                         dl = get_disk_super(super,
9138                                                             dev->disk.major,
9139                                                             dev->disk.minor);
9140                                         dl->index = u->old_raid_disks;
9141                                         dev = dev->next;
9142                                 }
9143                                 sysfs_free(spares);
9144                         }
9145                 }
9146                 len = disks_to_mpb_size(u->new_raid_disks);
9147                 dprintf("New anchor length is %llu\n", (unsigned long long)len);
9148                 break;
9149         }
9150         case update_size_change: {
9151                 if (update->len < (int)sizeof(struct imsm_update_size_change))
9152                         return 0;
9153                 break;
9154         }
9155         case update_activate_spare: {
9156                 if (update->len < (int)sizeof(struct imsm_update_activate_spare))
9157                         return 0;
9158                 break;
9159         }
9160         case update_create_array: {
9161                 struct imsm_update_create_array *u = (void *) update->buf;
9162                 struct intel_dev *dv;
9163                 struct imsm_dev *dev = &u->dev;
9164                 struct imsm_map *map = get_imsm_map(dev, MAP_0);
9165                 struct dl *dl;
9166                 struct disk_info *inf;
9167                 int i;
9168                 int activate = 0;
9169
9170                 if (update->len < (int)sizeof(*u))
9171                         return 0;
9172
9173                 inf = get_disk_info(u);
9174                 len = sizeof_imsm_dev(dev, 1);
9175                 /* allocate a new super->devlist entry */
9176                 dv = xmalloc(sizeof(*dv));
9177                 dv->dev = xmalloc(len);
9178                 update->space = dv;
9179
9180                 /* count how many spares will be converted to members */
9181                 for (i = 0; i < map->num_members; i++) {
9182                         dl = serial_to_dl(inf[i].serial, super);
9183                         if (!dl) {
9184                                 /* hmm maybe it failed?, nothing we can do about
9185                                  * it here
9186                                  */
9187                                 continue;
9188                         }
9189                         if (count_memberships(dl, super) == 0)
9190                                 activate++;
9191                 }
9192                 len += activate * sizeof(struct imsm_disk);
9193                 break;
9194         }
9195         case update_kill_array: {
9196                 if (update->len < (int)sizeof(struct imsm_update_kill_array))
9197                         return 0;
9198                 break;
9199         }
9200         case update_rename_array: {
9201                 if (update->len < (int)sizeof(struct imsm_update_rename_array))
9202                         return 0;
9203                 break;
9204         }
9205         case update_add_remove_disk:
9206                 /* no update->len needed */
9207                 break;
9208         default:
9209                 return 0;
9210         }
9211
9212         /* check if we need a larger metadata buffer */
9213         if (super->next_buf)
9214                 buf_len = super->next_len;
9215         else
9216                 buf_len = super->len;
9217
9218         if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
9219                 /* ok we need a larger buf than what is currently allocated
9220                  * if this allocation fails process_update will notice that
9221                  * ->next_len is set and ->next_buf is NULL
9222                  */
9223                 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len,
9224                                   sector_size);
9225                 if (super->next_buf)
9226                         free(super->next_buf);
9227
9228                 super->next_len = buf_len;
9229                 if (posix_memalign(&super->next_buf, sector_size, buf_len) == 0)
9230                         memset(super->next_buf, 0, buf_len);
9231                 else
9232                         super->next_buf = NULL;
9233         }
9234         return 1;
9235 }
9236
9237 /* must be called while manager is quiesced */
9238 static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index)
9239 {
9240         struct imsm_super *mpb = super->anchor;
9241         struct dl *iter;
9242         struct imsm_dev *dev;
9243         struct imsm_map *map;
9244         int i, j, num_members;
9245         __u32 ord;
9246
9247         dprintf("deleting device[%d] from imsm_super\n", index);
9248
9249         /* shift all indexes down one */
9250         for (iter = super->disks; iter; iter = iter->next)
9251                 if (iter->index > (int)index)
9252                         iter->index--;
9253         for (iter = super->missing; iter; iter = iter->next)
9254                 if (iter->index > (int)index)
9255                         iter->index--;
9256
9257         for (i = 0; i < mpb->num_raid_devs; i++) {
9258                 dev = get_imsm_dev(super, i);
9259                 map = get_imsm_map(dev, MAP_0);
9260                 num_members = map->num_members;
9261                 for (j = 0; j < num_members; j++) {
9262                         /* update ord entries being careful not to propagate
9263                          * ord-flags to the first map
9264                          */
9265                         ord = get_imsm_ord_tbl_ent(dev, j, MAP_X);
9266
9267                         if (ord_to_idx(ord) <= index)
9268                                 continue;
9269
9270                         map = get_imsm_map(dev, MAP_0);
9271                         set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
9272                         map = get_imsm_map(dev, MAP_1);
9273                         if (map)
9274                                 set_imsm_ord_tbl_ent(map, j, ord - 1);
9275                 }
9276         }
9277
9278         mpb->num_disks--;
9279         super->updates_pending++;
9280         if (*dlp) {
9281                 struct dl *dl = *dlp;
9282
9283                 *dlp = (*dlp)->next;
9284                 __free_imsm_disk(dl);
9285         }
9286 }
9287 #endif /* MDASSEMBLE */
9288
9289 static void close_targets(int *targets, int new_disks)
9290 {
9291         int i;
9292
9293         if (!targets)
9294                 return;
9295
9296         for (i = 0; i < new_disks; i++) {
9297                 if (targets[i] >= 0) {
9298                         close(targets[i]);
9299                         targets[i] = -1;
9300                 }
9301         }
9302 }
9303
9304 static int imsm_get_allowed_degradation(int level, int raid_disks,
9305                                         struct intel_super *super,
9306                                         struct imsm_dev *dev)
9307 {
9308         switch (level) {
9309         case 1:
9310         case 10:{
9311                 int ret_val = 0;
9312                 struct imsm_map *map;
9313                 int i;
9314
9315                 ret_val = raid_disks/2;
9316                 /* check map if all disks pairs not failed
9317                  * in both maps
9318                  */
9319                 map = get_imsm_map(dev, MAP_0);
9320                 for (i = 0; i < ret_val; i++) {
9321                         int degradation = 0;
9322                         if (get_imsm_disk(super, i) == NULL)
9323                                 degradation++;
9324                         if (get_imsm_disk(super, i + 1) == NULL)
9325                                 degradation++;
9326                         if (degradation == 2)
9327                                 return 0;
9328                 }
9329                 map = get_imsm_map(dev, MAP_1);
9330                 /* if there is no second map
9331                  * result can be returned
9332                  */
9333                 if (map == NULL)
9334                         return ret_val;
9335                 /* check degradation in second map
9336                  */
9337                 for (i = 0; i < ret_val; i++) {
9338                         int degradation = 0;
9339                 if (get_imsm_disk(super, i) == NULL)
9340                                 degradation++;
9341                         if (get_imsm_disk(super, i + 1) == NULL)
9342                                 degradation++;
9343                         if (degradation == 2)
9344                                 return 0;
9345                 }
9346                 return ret_val;
9347         }
9348         case 5:
9349                 return 1;
9350         case 6:
9351                 return 2;
9352         default:
9353                 return 0;
9354         }
9355 }
9356
9357 /*******************************************************************************
9358  * Function:    open_backup_targets
9359  * Description: Function opens file descriptors for all devices given in
9360  *              info->devs
9361  * Parameters:
9362  *      info            : general array info
9363  *      raid_disks      : number of disks
9364  *      raid_fds        : table of device's file descriptors
9365  *      super           : intel super for raid10 degradation check
9366  *      dev             : intel device for raid10 degradation check
9367  * Returns:
9368  *       0 : success
9369  *      -1 : fail
9370  ******************************************************************************/
9371 int open_backup_targets(struct mdinfo *info, int raid_disks, int *raid_fds,
9372                         struct intel_super *super, struct imsm_dev *dev)
9373 {
9374         struct mdinfo *sd;
9375         int i;
9376         int opened = 0;
9377
9378         for (i = 0; i < raid_disks; i++)
9379                 raid_fds[i] = -1;
9380
9381         for (sd = info->devs ; sd ; sd = sd->next) {
9382                 char *dn;
9383
9384                 if (sd->disk.state & (1<<MD_DISK_FAULTY)) {
9385                         dprintf("disk is faulty!!\n");
9386                         continue;
9387                 }
9388
9389                 if (sd->disk.raid_disk >= raid_disks || sd->disk.raid_disk < 0)
9390                         continue;
9391
9392                 dn = map_dev(sd->disk.major,
9393                              sd->disk.minor, 1);
9394                 raid_fds[sd->disk.raid_disk] = dev_open(dn, O_RDWR);
9395                 if (raid_fds[sd->disk.raid_disk] < 0) {
9396                         pr_err("cannot open component\n");
9397                         continue;
9398                 }
9399                 opened++;
9400         }
9401         /* check if maximum array degradation level is not exceeded
9402         */
9403         if ((raid_disks - opened) >
9404             imsm_get_allowed_degradation(info->new_level, raid_disks,
9405                                          super, dev)) {
9406                 pr_err("Not enough disks can be opened.\n");
9407                 close_targets(raid_fds, raid_disks);
9408                 return -2;
9409         }
9410         return 0;
9411 }
9412
9413 /*******************************************************************************
9414  * Function:    validate_container_imsm
9415  * Description: This routine validates container after assemble,
9416  *              eg. if devices in container are under the same controller.
9417  *
9418  * Parameters:
9419  *      info    : linked list with info about devices used in array
9420  * Returns:
9421  *      1 : HBA mismatch
9422  *      0 : Success
9423  ******************************************************************************/
9424 int validate_container_imsm(struct mdinfo *info)
9425 {
9426         if (check_env("IMSM_NO_PLATFORM"))
9427                 return 0;
9428
9429         struct sys_dev *idev;
9430         struct sys_dev *hba = NULL;
9431         struct sys_dev *intel_devices = find_intel_devices();
9432         char *dev_path = devt_to_devpath(makedev(info->disk.major,
9433                                                                         info->disk.minor));
9434
9435         for (idev = intel_devices; idev; idev = idev->next) {
9436                 if (dev_path && strstr(dev_path, idev->path)) {
9437                         hba = idev;
9438                         break;
9439                 }
9440         }
9441         if (dev_path)
9442                 free(dev_path);
9443
9444         if (!hba) {
9445                 pr_err("WARNING - Cannot detect HBA for device %s!\n",
9446                                 devid2kname(makedev(info->disk.major, info->disk.minor)));
9447                 return 1;
9448         }
9449
9450         const struct imsm_orom *orom = get_orom_by_device_id(hba->dev_id);
9451         struct mdinfo *dev;
9452
9453         for (dev = info->next; dev; dev = dev->next) {
9454                 dev_path = devt_to_devpath(makedev(dev->disk.major, dev->disk.minor));
9455
9456                 struct sys_dev *hba2 = NULL;
9457                 for (idev = intel_devices; idev; idev = idev->next) {
9458                         if (dev_path && strstr(dev_path, idev->path)) {
9459                                 hba2 = idev;
9460                                 break;
9461                         }
9462                 }
9463                 if (dev_path)
9464                         free(dev_path);
9465
9466                 const struct imsm_orom *orom2 = hba2 == NULL ? NULL :
9467                                 get_orom_by_device_id(hba2->dev_id);
9468
9469                 if (hba2 && hba->type != hba2->type) {
9470                         pr_err("WARNING - HBAs of devices do not match %s != %s\n",
9471                                 get_sys_dev_type(hba->type), get_sys_dev_type(hba2->type));
9472                         return 1;
9473                 }
9474
9475                 if (orom != orom2) {
9476                         pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
9477                                 "       This operation is not supported and can lead to data loss.\n");
9478                         return 1;
9479                 }
9480
9481                 if (!orom) {
9482                         pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
9483                                 "       This operation is not supported and can lead to data loss.\n");
9484                         return 1;
9485                 }
9486         }
9487
9488         return 0;
9489 }
9490 #ifndef MDASSEMBLE
9491 /*******************************************************************************
9492  * Function:    init_migr_record_imsm
9493  * Description: Function inits imsm migration record
9494  * Parameters:
9495  *      super   : imsm internal array info
9496  *      dev     : device under migration
9497  *      info    : general array info to find the smallest device
9498  * Returns:
9499  *      none
9500  ******************************************************************************/
9501 void init_migr_record_imsm(struct supertype *st, struct imsm_dev *dev,
9502                            struct mdinfo *info)
9503 {
9504         struct intel_super *super = st->sb;
9505         struct migr_record *migr_rec = super->migr_rec;
9506         int new_data_disks;
9507         unsigned long long dsize, dev_sectors;
9508         long long unsigned min_dev_sectors = -1LLU;
9509         struct mdinfo *sd;
9510         char nm[30];
9511         int fd;
9512         struct imsm_map *map_dest = get_imsm_map(dev, MAP_0);
9513         struct imsm_map *map_src = get_imsm_map(dev, MAP_1);
9514         unsigned long long num_migr_units;
9515         unsigned long long array_blocks;
9516
9517         memset(migr_rec, 0, sizeof(struct migr_record));
9518         migr_rec->family_num = __cpu_to_le32(super->anchor->family_num);
9519
9520         /* only ascending reshape supported now */
9521         migr_rec->ascending_migr = __cpu_to_le32(1);
9522
9523         migr_rec->dest_depth_per_unit = GEN_MIGR_AREA_SIZE /
9524                 max(map_dest->blocks_per_strip, map_src->blocks_per_strip);
9525         migr_rec->dest_depth_per_unit *=
9526                 max(map_dest->blocks_per_strip, map_src->blocks_per_strip);
9527         new_data_disks = imsm_num_data_members(dev, MAP_0);
9528         migr_rec->blocks_per_unit =
9529                 __cpu_to_le32(migr_rec->dest_depth_per_unit * new_data_disks);
9530         migr_rec->dest_depth_per_unit =
9531                 __cpu_to_le32(migr_rec->dest_depth_per_unit);
9532         array_blocks = info->component_size * new_data_disks;
9533         num_migr_units =
9534                 array_blocks / __le32_to_cpu(migr_rec->blocks_per_unit);
9535
9536         if (array_blocks % __le32_to_cpu(migr_rec->blocks_per_unit))
9537                 num_migr_units++;
9538         migr_rec->num_migr_units = __cpu_to_le32(num_migr_units);
9539
9540         migr_rec->post_migr_vol_cap =  dev->size_low;
9541         migr_rec->post_migr_vol_cap_hi = dev->size_high;
9542
9543         /* Find the smallest dev */
9544         for (sd = info->devs ; sd ; sd = sd->next) {
9545                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
9546                 fd = dev_open(nm, O_RDONLY);
9547                 if (fd < 0)
9548                         continue;
9549                 get_dev_size(fd, NULL, &dsize);
9550                 dev_sectors = dsize / 512;
9551                 if (dev_sectors < min_dev_sectors)
9552                         min_dev_sectors = dev_sectors;
9553                 close(fd);
9554         }
9555         migr_rec->ckpt_area_pba = __cpu_to_le32(min_dev_sectors -
9556                                         RAID_DISK_RESERVED_BLOCKS_IMSM_HI);
9557
9558         write_imsm_migr_rec(st);
9559
9560         return;
9561 }
9562
9563 /*******************************************************************************
9564  * Function:    save_backup_imsm
9565  * Description: Function saves critical data stripes to Migration Copy Area
9566  *              and updates the current migration unit status.
9567  *              Use restore_stripes() to form a destination stripe,
9568  *              and to write it to the Copy Area.
9569  * Parameters:
9570  *      st              : supertype information
9571  *      dev             : imsm device that backup is saved for
9572  *      info            : general array info
9573  *      buf             : input buffer
9574  *      length          : length of data to backup (blocks_per_unit)
9575  * Returns:
9576  *       0 : success
9577  *,     -1 : fail
9578  ******************************************************************************/
9579 int save_backup_imsm(struct supertype *st,
9580                      struct imsm_dev *dev,
9581                      struct mdinfo *info,
9582                      void *buf,
9583                      int length)
9584 {
9585         int rv = -1;
9586         struct intel_super *super = st->sb;
9587         unsigned long long *target_offsets;
9588         int *targets;
9589         int i;
9590         struct imsm_map *map_dest = get_imsm_map(dev, MAP_0);
9591         int new_disks = map_dest->num_members;
9592         int dest_layout = 0;
9593         int dest_chunk;
9594         unsigned long long start;
9595         int data_disks = imsm_num_data_members(dev, MAP_0);
9596
9597         targets = xmalloc(new_disks * sizeof(int));
9598
9599         for (i = 0; i < new_disks; i++)
9600                 targets[i] = -1;
9601
9602         target_offsets = xcalloc(new_disks, sizeof(unsigned long long));
9603
9604         start = info->reshape_progress * 512;
9605         for (i = 0; i < new_disks; i++) {
9606                 target_offsets[i] = (unsigned long long)
9607                   __le32_to_cpu(super->migr_rec->ckpt_area_pba) * 512;
9608                 /* move back copy area adderss, it will be moved forward
9609                  * in restore_stripes() using start input variable
9610                  */
9611                 target_offsets[i] -= start/data_disks;
9612         }
9613
9614         if (open_backup_targets(info, new_disks, targets,
9615                                 super, dev))
9616                 goto abort;
9617
9618         dest_layout = imsm_level_to_layout(map_dest->raid_level);
9619         dest_chunk = __le16_to_cpu(map_dest->blocks_per_strip) * 512;
9620
9621         if (restore_stripes(targets, /* list of dest devices */
9622                             target_offsets, /* migration record offsets */
9623                             new_disks,
9624                             dest_chunk,
9625                             map_dest->raid_level,
9626                             dest_layout,
9627                             -1,    /* source backup file descriptor */
9628                             0,     /* input buf offset
9629                                     * always 0 buf is already offseted */
9630                             start,
9631                             length,
9632                             buf) != 0) {
9633                 pr_err("Error restoring stripes\n");
9634                 goto abort;
9635         }
9636
9637         rv = 0;
9638
9639 abort:
9640         if (targets) {
9641                 close_targets(targets, new_disks);
9642                 free(targets);
9643         }
9644         free(target_offsets);
9645
9646         return rv;
9647 }
9648
9649 /*******************************************************************************
9650  * Function:    save_checkpoint_imsm
9651  * Description: Function called for current unit status update
9652  *              in the migration record. It writes it to disk.
9653  * Parameters:
9654  *      super   : imsm internal array info
9655  *      info    : general array info
9656  * Returns:
9657  *      0: success
9658  *      1: failure
9659  *      2: failure, means no valid migration record
9660  *                 / no general migration in progress /
9661  ******************************************************************************/
9662 int save_checkpoint_imsm(struct supertype *st, struct mdinfo *info, int state)
9663 {
9664         struct intel_super *super = st->sb;
9665         unsigned long long blocks_per_unit;
9666         unsigned long long curr_migr_unit;
9667
9668         if (load_imsm_migr_rec(super, info) != 0) {
9669                 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
9670                 return 1;
9671         }
9672
9673         blocks_per_unit = __le32_to_cpu(super->migr_rec->blocks_per_unit);
9674         if (blocks_per_unit == 0) {
9675                 dprintf("imsm: no migration in progress.\n");
9676                 return 2;
9677         }
9678         curr_migr_unit = info->reshape_progress / blocks_per_unit;
9679         /* check if array is alligned to copy area
9680          * if it is not alligned, add one to current migration unit value
9681          * this can happend on array reshape finish only
9682          */
9683         if (info->reshape_progress % blocks_per_unit)
9684                 curr_migr_unit++;
9685
9686         super->migr_rec->curr_migr_unit =
9687                 __cpu_to_le32(curr_migr_unit);
9688         super->migr_rec->rec_status = __cpu_to_le32(state);
9689         super->migr_rec->dest_1st_member_lba =
9690                 __cpu_to_le32(curr_migr_unit *
9691                               __le32_to_cpu(super->migr_rec->dest_depth_per_unit));
9692         if (write_imsm_migr_rec(st) < 0) {
9693                 dprintf("imsm: Cannot write migration record outside backup area\n");
9694                 return 1;
9695         }
9696
9697         return 0;
9698 }
9699
9700 /*******************************************************************************
9701  * Function:    recover_backup_imsm
9702  * Description: Function recovers critical data from the Migration Copy Area
9703  *              while assembling an array.
9704  * Parameters:
9705  *      super   : imsm internal array info
9706  *      info    : general array info
9707  * Returns:
9708  *      0 : success (or there is no data to recover)
9709  *      1 : fail
9710  ******************************************************************************/
9711 int recover_backup_imsm(struct supertype *st, struct mdinfo *info)
9712 {
9713         struct intel_super *super = st->sb;
9714         struct migr_record *migr_rec = super->migr_rec;
9715         struct imsm_map *map_dest;
9716         struct intel_dev *id = NULL;
9717         unsigned long long read_offset;
9718         unsigned long long write_offset;
9719         unsigned unit_len;
9720         int *targets = NULL;
9721         int new_disks, i, err;
9722         char *buf = NULL;
9723         int retval = 1;
9724         unsigned int sector_size = super->sector_size;
9725         unsigned long curr_migr_unit = __le32_to_cpu(migr_rec->curr_migr_unit);
9726         unsigned long num_migr_units = __le32_to_cpu(migr_rec->num_migr_units);
9727         char buffer[20];
9728         int skipped_disks = 0;
9729
9730         err = sysfs_get_str(info, NULL, "array_state", (char *)buffer, 20);
9731         if (err < 1)
9732                 return 1;
9733
9734         /* recover data only during assemblation */
9735         if (strncmp(buffer, "inactive", 8) != 0)
9736                 return 0;
9737         /* no data to recover */
9738         if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL)
9739                 return 0;
9740         if (curr_migr_unit >= num_migr_units)
9741                 return 1;
9742
9743         /* find device during reshape */
9744         for (id = super->devlist; id; id = id->next)
9745                 if (is_gen_migration(id->dev))
9746                         break;
9747         if (id == NULL)
9748                 return 1;
9749
9750         map_dest = get_imsm_map(id->dev, MAP_0);
9751         new_disks = map_dest->num_members;
9752
9753         read_offset = (unsigned long long)
9754                         __le32_to_cpu(migr_rec->ckpt_area_pba) * 512;
9755
9756         write_offset = ((unsigned long long)
9757                         __le32_to_cpu(migr_rec->dest_1st_member_lba) +
9758                         pba_of_lba0(map_dest)) * 512;
9759
9760         unit_len = __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512;
9761         if (posix_memalign((void **)&buf, sector_size, unit_len) != 0)
9762                 goto abort;
9763         targets = xcalloc(new_disks, sizeof(int));
9764
9765         if (open_backup_targets(info, new_disks, targets, super, id->dev)) {
9766                 pr_err("Cannot open some devices belonging to array.\n");
9767                 goto abort;
9768         }
9769
9770         for (i = 0; i < new_disks; i++) {
9771                 if (targets[i] < 0) {
9772                         skipped_disks++;
9773                         continue;
9774                 }
9775                 if (lseek64(targets[i], read_offset, SEEK_SET) < 0) {
9776                         pr_err("Cannot seek to block: %s\n",
9777                                strerror(errno));
9778                         skipped_disks++;
9779                         continue;
9780                 }
9781                 if ((unsigned)read(targets[i], buf, unit_len) != unit_len) {
9782                         pr_err("Cannot read copy area block: %s\n",
9783                                strerror(errno));
9784                         skipped_disks++;
9785                         continue;
9786                 }
9787                 if (lseek64(targets[i], write_offset, SEEK_SET) < 0) {
9788                         pr_err("Cannot seek to block: %s\n",
9789                                strerror(errno));
9790                         skipped_disks++;
9791                         continue;
9792                 }
9793                 if ((unsigned)write(targets[i], buf, unit_len) != unit_len) {
9794                         pr_err("Cannot restore block: %s\n",
9795                                strerror(errno));
9796                         skipped_disks++;
9797                         continue;
9798                 }
9799         }
9800
9801         if (skipped_disks > imsm_get_allowed_degradation(info->new_level,
9802                                                          new_disks,
9803                                                          super,
9804                                                          id->dev)) {
9805                 pr_err("Cannot restore data from backup. Too many failed disks\n");
9806                 goto abort;
9807         }
9808
9809         if (save_checkpoint_imsm(st, info, UNIT_SRC_NORMAL)) {
9810                 /* ignore error == 2, this can mean end of reshape here
9811                  */
9812                 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
9813         } else
9814                 retval = 0;
9815
9816 abort:
9817         if (targets) {
9818                 for (i = 0; i < new_disks; i++)
9819                         if (targets[i])
9820                                 close(targets[i]);
9821                 free(targets);
9822         }
9823         free(buf);
9824         return retval;
9825 }
9826
9827 static char disk_by_path[] = "/dev/disk/by-path/";
9828
9829 static const char *imsm_get_disk_controller_domain(const char *path)
9830 {
9831         char disk_path[PATH_MAX];
9832         char *drv=NULL;
9833         struct stat st;
9834
9835         strcpy(disk_path, disk_by_path);
9836         strncat(disk_path, path, PATH_MAX - strlen(disk_path) - 1);
9837         if (stat(disk_path, &st) == 0) {
9838                 struct sys_dev* hba;
9839                 char *path;
9840
9841                 path = devt_to_devpath(st.st_rdev);
9842                 if (path == NULL)
9843                         return "unknown";
9844                 hba = find_disk_attached_hba(-1, path);
9845                 if (hba && hba->type == SYS_DEV_SAS)
9846                         drv = "isci";
9847                 else if (hba && hba->type == SYS_DEV_SATA)
9848                         drv = "ahci";
9849                 else
9850                         drv = "unknown";
9851                 dprintf("path: %s hba: %s attached: %s\n",
9852                         path, (hba) ? hba->path : "NULL", drv);
9853                 free(path);
9854         }
9855         return drv;
9856 }
9857
9858 static char *imsm_find_array_devnm_by_subdev(int subdev, char *container)
9859 {
9860         static char devnm[32];
9861         char subdev_name[20];
9862         struct mdstat_ent *mdstat;
9863
9864         sprintf(subdev_name, "%d", subdev);
9865         mdstat = mdstat_by_subdev(subdev_name, container);
9866         if (!mdstat)
9867                 return NULL;
9868
9869         strcpy(devnm, mdstat->devnm);
9870         free_mdstat(mdstat);
9871         return devnm;
9872 }
9873
9874 static int imsm_reshape_is_allowed_on_container(struct supertype *st,
9875                                                 struct geo_params *geo,
9876                                                 int *old_raid_disks,
9877                                                 int direction)
9878 {
9879         /* currently we only support increasing the number of devices
9880          * for a container.  This increases the number of device for each
9881          * member array.  They must all be RAID0 or RAID5.
9882          */
9883         int ret_val = 0;
9884         struct mdinfo *info, *member;
9885         int devices_that_can_grow = 0;
9886
9887         dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st->devnm);
9888
9889         if (geo->size > 0 ||
9890             geo->level != UnSet ||
9891             geo->layout != UnSet ||
9892             geo->chunksize != 0 ||
9893             geo->raid_disks == UnSet) {
9894                 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
9895                 return ret_val;
9896         }
9897
9898         if (direction == ROLLBACK_METADATA_CHANGES) {
9899                 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
9900                 return ret_val;
9901         }
9902
9903         info = container_content_imsm(st, NULL);
9904         for (member = info; member; member = member->next) {
9905                 char *result;
9906
9907                 dprintf("imsm: checking device_num: %i\n",
9908                         member->container_member);
9909
9910                 if (geo->raid_disks <= member->array.raid_disks) {
9911                         /* we work on container for Online Capacity Expansion
9912                          * only so raid_disks has to grow
9913                          */
9914                         dprintf("imsm: for container operation raid disks increase is required\n");
9915                         break;
9916                 }
9917
9918                 if (info->array.level != 0 && info->array.level != 5) {
9919                         /* we cannot use this container with other raid level
9920                          */
9921                         dprintf("imsm: for container operation wrong raid level (%i) detected\n",
9922                                 info->array.level);
9923                         break;
9924                 } else {
9925                         /* check for platform support
9926                          * for this raid level configuration
9927                          */
9928                         struct intel_super *super = st->sb;
9929                         if (!is_raid_level_supported(super->orom,
9930                                                      member->array.level,
9931                                                      geo->raid_disks)) {
9932                                 dprintf("platform does not support raid%d with %d disk%s\n",
9933                                          info->array.level,
9934                                          geo->raid_disks,
9935                                          geo->raid_disks > 1 ? "s" : "");
9936                                 break;
9937                         }
9938                         /* check if component size is aligned to chunk size
9939                          */
9940                         if (info->component_size %
9941                             (info->array.chunk_size/512)) {
9942                                 dprintf("Component size is not aligned to chunk size\n");
9943                                 break;
9944                         }
9945                 }
9946
9947                 if (*old_raid_disks &&
9948                     info->array.raid_disks != *old_raid_disks)
9949                         break;
9950                 *old_raid_disks = info->array.raid_disks;
9951
9952                 /* All raid5 and raid0 volumes in container
9953                  * have to be ready for Online Capacity Expansion
9954                  * so they need to be assembled.  We have already
9955                  * checked that no recovery etc is happening.
9956                  */
9957                 result = imsm_find_array_devnm_by_subdev(member->container_member,
9958                                                          st->container_devnm);
9959                 if (result == NULL) {
9960                         dprintf("imsm: cannot find array\n");
9961                         break;
9962                 }
9963                 devices_that_can_grow++;
9964         }
9965         sysfs_free(info);
9966         if (!member && devices_that_can_grow)
9967                 ret_val = 1;
9968
9969         if (ret_val)
9970                 dprintf("Container operation allowed\n");
9971         else
9972                 dprintf("Error: %i\n", ret_val);
9973
9974         return ret_val;
9975 }
9976
9977 /* Function: get_spares_for_grow
9978  * Description: Allocates memory and creates list of spare devices
9979  *              avaliable in container. Checks if spare drive size is acceptable.
9980  * Parameters: Pointer to the supertype structure
9981  * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
9982  *              NULL if fail
9983  */
9984 static struct mdinfo *get_spares_for_grow(struct supertype *st)
9985 {
9986         unsigned long long min_size = min_acceptable_spare_size_imsm(st);
9987         return container_choose_spares(st, min_size, NULL, NULL, NULL, 0);
9988 }
9989
9990 /******************************************************************************
9991  * function: imsm_create_metadata_update_for_reshape
9992  * Function creates update for whole IMSM container.
9993  *
9994  ******************************************************************************/
9995 static int imsm_create_metadata_update_for_reshape(
9996         struct supertype *st,
9997         struct geo_params *geo,
9998         int old_raid_disks,
9999         struct imsm_update_reshape **updatep)
10000 {
10001         struct intel_super *super = st->sb;
10002         struct imsm_super *mpb = super->anchor;
10003         int update_memory_size;
10004         struct imsm_update_reshape *u;
10005         struct mdinfo *spares;
10006         int i;
10007         int delta_disks;
10008         struct mdinfo *dev;
10009
10010         dprintf("(enter) raid_disks = %i\n", geo->raid_disks);
10011
10012         delta_disks = geo->raid_disks - old_raid_disks;
10013
10014         /* size of all update data without anchor */
10015         update_memory_size = sizeof(struct imsm_update_reshape);
10016
10017         /* now add space for spare disks that we need to add. */
10018         update_memory_size += sizeof(u->new_disks[0]) * (delta_disks - 1);
10019
10020         u = xcalloc(1, update_memory_size);
10021         u->type = update_reshape_container_disks;
10022         u->old_raid_disks = old_raid_disks;
10023         u->new_raid_disks = geo->raid_disks;
10024
10025         /* now get spare disks list
10026          */
10027         spares = get_spares_for_grow(st);
10028
10029         if (spares == NULL
10030             || delta_disks > spares->array.spare_disks) {
10031                 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo->dev_name);
10032                 i = -1;
10033                 goto abort;
10034         }
10035
10036         /* we have got spares
10037          * update disk list in imsm_disk list table in anchor
10038          */
10039         dprintf("imsm: %i spares are available.\n\n",
10040                 spares->array.spare_disks);
10041
10042         dev = spares->devs;
10043         for (i = 0; i < delta_disks; i++) {
10044                 struct dl *dl;
10045
10046                 if (dev == NULL)
10047                         break;
10048                 u->new_disks[i] = makedev(dev->disk.major,
10049                                           dev->disk.minor);
10050                 dl = get_disk_super(super, dev->disk.major, dev->disk.minor);
10051                 dl->index = mpb->num_disks;
10052                 mpb->num_disks++;
10053                 dev = dev->next;
10054         }
10055
10056 abort:
10057         /* free spares
10058          */
10059         sysfs_free(spares);
10060
10061         dprintf("imsm: reshape update preparation :");
10062         if (i == delta_disks) {
10063                 dprintf_cont(" OK\n");
10064                 *updatep = u;
10065                 return update_memory_size;
10066         }
10067         free(u);
10068         dprintf_cont(" Error\n");
10069
10070         return 0;
10071 }
10072
10073 /******************************************************************************
10074  * function: imsm_create_metadata_update_for_size_change()
10075  *           Creates update for IMSM array for array size change.
10076  *
10077  ******************************************************************************/
10078 static int imsm_create_metadata_update_for_size_change(
10079                                 struct supertype *st,
10080                                 struct geo_params *geo,
10081                                 struct imsm_update_size_change **updatep)
10082 {
10083         struct intel_super *super = st->sb;
10084         int update_memory_size;
10085         struct imsm_update_size_change *u;
10086
10087         dprintf("(enter) New size = %llu\n", geo->size);
10088
10089         /* size of all update data without anchor */
10090         update_memory_size = sizeof(struct imsm_update_size_change);
10091
10092         u = xcalloc(1, update_memory_size);
10093         u->type = update_size_change;
10094         u->subdev = super->current_vol;
10095         u->new_size = geo->size;
10096
10097         dprintf("imsm: reshape update preparation : OK\n");
10098         *updatep = u;
10099
10100         return update_memory_size;
10101 }
10102
10103 /******************************************************************************
10104  * function: imsm_create_metadata_update_for_migration()
10105  *           Creates update for IMSM array.
10106  *
10107  ******************************************************************************/
10108 static int imsm_create_metadata_update_for_migration(
10109                                         struct supertype *st,
10110                                         struct geo_params *geo,
10111                                         struct imsm_update_reshape_migration **updatep)
10112 {
10113         struct intel_super *super = st->sb;
10114         int update_memory_size;
10115         struct imsm_update_reshape_migration *u;
10116         struct imsm_dev *dev;
10117         int previous_level = -1;
10118
10119         dprintf("(enter) New Level = %i\n", geo->level);
10120
10121         /* size of all update data without anchor */
10122         update_memory_size = sizeof(struct imsm_update_reshape_migration);
10123
10124         u = xcalloc(1, update_memory_size);
10125         u->type = update_reshape_migration;
10126         u->subdev = super->current_vol;
10127         u->new_level = geo->level;
10128         u->new_layout = geo->layout;
10129         u->new_raid_disks = u->old_raid_disks = geo->raid_disks;
10130         u->new_disks[0] = -1;
10131         u->new_chunksize = -1;
10132
10133         dev = get_imsm_dev(super, u->subdev);
10134         if (dev) {
10135                 struct imsm_map *map;
10136
10137                 map = get_imsm_map(dev, MAP_0);
10138                 if (map) {
10139                         int current_chunk_size =
10140                                 __le16_to_cpu(map->blocks_per_strip) / 2;
10141
10142                         if (geo->chunksize != current_chunk_size) {
10143                                 u->new_chunksize = geo->chunksize / 1024;
10144                                 dprintf("imsm: chunk size change from %i to %i\n",
10145                                         current_chunk_size, u->new_chunksize);
10146                         }
10147                         previous_level = map->raid_level;
10148                 }
10149         }
10150         if (geo->level == 5 && previous_level == 0) {
10151                 struct mdinfo *spares = NULL;
10152
10153                 u->new_raid_disks++;
10154                 spares = get_spares_for_grow(st);
10155                 if (spares == NULL || spares->array.spare_disks < 1) {
10156                         free(u);
10157                         sysfs_free(spares);
10158                         update_memory_size = 0;
10159                         dprintf("error: cannot get spare device for requested migration");
10160                         return 0;
10161                 }
10162                 sysfs_free(spares);
10163         }
10164         dprintf("imsm: reshape update preparation : OK\n");
10165         *updatep = u;
10166
10167         return update_memory_size;
10168 }
10169
10170 static void imsm_update_metadata_locally(struct supertype *st,
10171                                          void *buf, int len)
10172 {
10173         struct metadata_update mu;
10174
10175         mu.buf = buf;
10176         mu.len = len;
10177         mu.space = NULL;
10178         mu.space_list = NULL;
10179         mu.next = NULL;
10180         if (imsm_prepare_update(st, &mu))
10181                 imsm_process_update(st, &mu);
10182
10183         while (mu.space_list) {
10184                 void **space = mu.space_list;
10185                 mu.space_list = *space;
10186                 free(space);
10187         }
10188 }
10189
10190 /***************************************************************************
10191 * Function:     imsm_analyze_change
10192 * Description:  Function analyze change for single volume
10193 *               and validate if transition is supported
10194 * Parameters:   Geometry parameters, supertype structure,
10195 *               metadata change direction (apply/rollback)
10196 * Returns:      Operation type code on success, -1 if fail
10197 ****************************************************************************/
10198 enum imsm_reshape_type imsm_analyze_change(struct supertype *st,
10199                                            struct geo_params *geo,
10200                                            int direction)
10201 {
10202         struct mdinfo info;
10203         int change = -1;
10204         int check_devs = 0;
10205         int chunk;
10206         /* number of added/removed disks in operation result */
10207         int devNumChange = 0;
10208         /* imsm compatible layout value for array geometry verification */
10209         int imsm_layout = -1;
10210         int data_disks;
10211         struct imsm_dev *dev;
10212         struct intel_super *super;
10213         unsigned long long current_size;
10214         unsigned long long free_size;
10215         unsigned long long max_size;
10216         int rv;
10217
10218         getinfo_super_imsm_volume(st, &info, NULL);
10219         if (geo->level != info.array.level && geo->level >= 0 &&
10220             geo->level != UnSet) {
10221                 switch (info.array.level) {
10222                 case 0:
10223                         if (geo->level == 5) {
10224                                 change = CH_MIGRATION;
10225                                 if (geo->layout != ALGORITHM_LEFT_ASYMMETRIC) {
10226                                         pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
10227                                         change = -1;
10228                                         goto analyse_change_exit;
10229                                 }
10230                                 imsm_layout =  geo->layout;
10231                                 check_devs = 1;
10232                                 devNumChange = 1; /* parity disk added */
10233                         } else if (geo->level == 10) {
10234                                 change = CH_TAKEOVER;
10235                                 check_devs = 1;
10236                                 devNumChange = 2; /* two mirrors added */
10237                                 imsm_layout = 0x102; /* imsm supported layout */
10238                         }
10239                         break;
10240                 case 1:
10241                 case 10:
10242                         if (geo->level == 0) {
10243                                 change = CH_TAKEOVER;
10244                                 check_devs = 1;
10245                                 devNumChange = -(geo->raid_disks/2);
10246                                 imsm_layout = 0; /* imsm raid0 layout */
10247                         }
10248                         break;
10249                 }
10250                 if (change == -1) {
10251                         pr_err("Error. Level Migration from %d to %d not supported!\n",
10252                                info.array.level, geo->level);
10253                         goto analyse_change_exit;
10254                 }
10255         } else
10256                 geo->level = info.array.level;
10257
10258         if (geo->layout != info.array.layout &&
10259             (geo->layout != UnSet && geo->layout != -1)) {
10260                 change = CH_MIGRATION;
10261                 if (info.array.layout == 0 && info.array.level == 5 &&
10262                     geo->layout == 5) {
10263                         /* reshape 5 -> 4 */
10264                 } else if (info.array.layout == 5 && info.array.level == 5 &&
10265                            geo->layout == 0) {
10266                         /* reshape 4 -> 5 */
10267                         geo->layout = 0;
10268                         geo->level = 5;
10269                 } else {
10270                         pr_err("Error. Layout Migration from %d to %d not supported!\n",
10271                                info.array.layout, geo->layout);
10272                         change = -1;
10273                         goto analyse_change_exit;
10274                 }
10275         } else {
10276                 geo->layout = info.array.layout;
10277                 if (imsm_layout == -1)
10278                         imsm_layout = info.array.layout;
10279         }
10280
10281         if (geo->chunksize > 0 && geo->chunksize != UnSet &&
10282             geo->chunksize != info.array.chunk_size) {
10283                 if (info.array.level == 10) {
10284                         pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
10285                         change = -1;
10286                         goto analyse_change_exit;
10287                 }
10288                 change = CH_MIGRATION;
10289         } else {
10290                 geo->chunksize = info.array.chunk_size;
10291         }
10292
10293         chunk = geo->chunksize / 1024;
10294
10295         super = st->sb;
10296         dev = get_imsm_dev(super, super->current_vol);
10297         data_disks = imsm_num_data_members(dev , MAP_0);
10298         /* compute current size per disk member
10299          */
10300         current_size = info.custom_array_size / data_disks;
10301
10302         if (geo->size > 0 && geo->size != MAX_SIZE) {
10303                 /* align component size
10304                  */
10305                 geo->size = imsm_component_size_aligment_check(
10306                                     get_imsm_raid_level(dev->vol.map),
10307                                     chunk * 1024, super->sector_size,
10308                                     geo->size * 2);
10309                 if (geo->size == 0) {
10310                         pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
10311                                    current_size);
10312                         goto analyse_change_exit;
10313                 }
10314         }
10315
10316         if (current_size != geo->size && geo->size > 0) {
10317                 if (change != -1) {
10318                         pr_err("Error. Size change should be the only one at a time.\n");
10319                         change = -1;
10320                         goto analyse_change_exit;
10321                 }
10322                 if ((super->current_vol + 1) != super->anchor->num_raid_devs) {
10323                         pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
10324                                super->current_vol, st->devnm);
10325                         goto analyse_change_exit;
10326                 }
10327                 /* check the maximum available size
10328                  */
10329                 rv =  imsm_get_free_size(st, dev->vol.map->num_members,
10330                                          0, chunk, &free_size);
10331                 if (rv == 0)
10332                         /* Cannot find maximum available space
10333                          */
10334                         max_size = 0;
10335                 else {
10336                         max_size = free_size + current_size;
10337                         /* align component size
10338                          */
10339                         max_size = imsm_component_size_aligment_check(
10340                                         get_imsm_raid_level(dev->vol.map),
10341                                         chunk * 1024, super->sector_size,
10342                                         max_size);
10343                 }
10344                 if (geo->size == MAX_SIZE) {
10345                         /* requested size change to the maximum available size
10346                          */
10347                         if (max_size == 0) {
10348                                 pr_err("Error. Cannot find maximum available space.\n");
10349                                 change = -1;
10350                                 goto analyse_change_exit;
10351                         } else
10352                                 geo->size = max_size;
10353                 }
10354
10355                 if (direction == ROLLBACK_METADATA_CHANGES) {
10356                         /* accept size for rollback only
10357                         */
10358                 } else {
10359                         /* round size due to metadata compatibility
10360                         */
10361                         geo->size = (geo->size >> SECT_PER_MB_SHIFT)
10362                                     << SECT_PER_MB_SHIFT;
10363                         dprintf("Prepare update for size change to %llu\n",
10364                                 geo->size );
10365                         if (current_size >= geo->size) {
10366                                 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
10367                                        current_size, geo->size);
10368                                 goto analyse_change_exit;
10369                         }
10370                         if (max_size && geo->size > max_size) {
10371                                 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
10372                                        max_size, geo->size);
10373                                 goto analyse_change_exit;
10374                         }
10375                 }
10376                 geo->size *= data_disks;
10377                 geo->raid_disks = dev->vol.map->num_members;
10378                 change = CH_ARRAY_SIZE;
10379         }
10380         if (!validate_geometry_imsm(st,
10381                                     geo->level,
10382                                     imsm_layout,
10383                                     geo->raid_disks + devNumChange,
10384                                     &chunk,
10385                                     geo->size, INVALID_SECTORS,
10386                                     0, 0, 1))
10387                 change = -1;
10388
10389         if (check_devs) {
10390                 struct intel_super *super = st->sb;
10391                 struct imsm_super *mpb = super->anchor;
10392
10393                 if (mpb->num_raid_devs > 1) {
10394                         pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
10395                                geo->dev_name);
10396                         change = -1;
10397                 }
10398         }
10399
10400 analyse_change_exit:
10401         if (direction == ROLLBACK_METADATA_CHANGES &&
10402             (change == CH_MIGRATION || change == CH_TAKEOVER)) {
10403                 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
10404                 change = -1;
10405         }
10406         return change;
10407 }
10408
10409 int imsm_takeover(struct supertype *st, struct geo_params *geo)
10410 {
10411         struct intel_super *super = st->sb;
10412         struct imsm_update_takeover *u;
10413
10414         u = xmalloc(sizeof(struct imsm_update_takeover));
10415
10416         u->type = update_takeover;
10417         u->subarray = super->current_vol;
10418
10419         /* 10->0 transition */
10420         if (geo->level == 0)
10421                 u->direction = R10_TO_R0;
10422
10423         /* 0->10 transition */
10424         if (geo->level == 10)
10425                 u->direction = R0_TO_R10;
10426
10427         /* update metadata locally */
10428         imsm_update_metadata_locally(st, u,
10429                                         sizeof(struct imsm_update_takeover));
10430         /* and possibly remotely */
10431         if (st->update_tail)
10432                 append_metadata_update(st, u,
10433                                         sizeof(struct imsm_update_takeover));
10434         else
10435                 free(u);
10436
10437         return 0;
10438 }
10439
10440 static int imsm_reshape_super(struct supertype *st, unsigned long long size,
10441                               int level,
10442                               int layout, int chunksize, int raid_disks,
10443                               int delta_disks, char *backup, char *dev,
10444                               int direction, int verbose)
10445 {
10446         int ret_val = 1;
10447         struct geo_params geo;
10448
10449         dprintf("(enter)\n");
10450
10451         memset(&geo, 0, sizeof(struct geo_params));
10452
10453         geo.dev_name = dev;
10454         strcpy(geo.devnm, st->devnm);
10455         geo.size = size;
10456         geo.level = level;
10457         geo.layout = layout;
10458         geo.chunksize = chunksize;
10459         geo.raid_disks = raid_disks;
10460         if (delta_disks != UnSet)
10461                 geo.raid_disks += delta_disks;
10462
10463         dprintf("for level      : %i\n", geo.level);
10464         dprintf("for raid_disks : %i\n", geo.raid_disks);
10465
10466         if (experimental() == 0)
10467                 return ret_val;
10468
10469         if (strcmp(st->container_devnm, st->devnm) == 0) {
10470                 /* On container level we can only increase number of devices. */
10471                 dprintf("imsm: info: Container operation\n");
10472                 int old_raid_disks = 0;
10473
10474                 if (imsm_reshape_is_allowed_on_container(
10475                             st, &geo, &old_raid_disks, direction)) {
10476                         struct imsm_update_reshape *u = NULL;
10477                         int len;
10478
10479                         len = imsm_create_metadata_update_for_reshape(
10480                                 st, &geo, old_raid_disks, &u);
10481
10482                         if (len <= 0) {
10483                                 dprintf("imsm: Cannot prepare update\n");
10484                                 goto exit_imsm_reshape_super;
10485                         }
10486
10487                         ret_val = 0;
10488                         /* update metadata locally */
10489                         imsm_update_metadata_locally(st, u, len);
10490                         /* and possibly remotely */
10491                         if (st->update_tail)
10492                                 append_metadata_update(st, u, len);
10493                         else
10494                                 free(u);
10495
10496                 } else {
10497                         pr_err("(imsm) Operation is not allowed on this container\n");
10498                 }
10499         } else {
10500                 /* On volume level we support following operations
10501                  * - takeover: raid10 -> raid0; raid0 -> raid10
10502                  * - chunk size migration
10503                  * - migration: raid5 -> raid0; raid0 -> raid5
10504                  */
10505                 struct intel_super *super = st->sb;
10506                 struct intel_dev *dev = super->devlist;
10507                 int change;
10508                 dprintf("imsm: info: Volume operation\n");
10509                 /* find requested device */
10510                 while (dev) {
10511                         char *devnm =
10512                                 imsm_find_array_devnm_by_subdev(
10513                                         dev->index, st->container_devnm);
10514                         if (devnm && strcmp(devnm, geo.devnm) == 0)
10515                                 break;
10516                         dev = dev->next;
10517                 }
10518                 if (dev == NULL) {
10519                         pr_err("Cannot find %s (%s) subarray\n",
10520                                 geo.dev_name, geo.devnm);
10521                         goto exit_imsm_reshape_super;
10522                 }
10523                 super->current_vol = dev->index;
10524                 change = imsm_analyze_change(st, &geo, direction);
10525                 switch (change) {
10526                 case CH_TAKEOVER:
10527                         ret_val = imsm_takeover(st, &geo);
10528                         break;
10529                 case CH_MIGRATION: {
10530                         struct imsm_update_reshape_migration *u = NULL;
10531                         int len =
10532                                 imsm_create_metadata_update_for_migration(
10533                                         st, &geo, &u);
10534                         if (len < 1) {
10535                                 dprintf("imsm: Cannot prepare update\n");
10536                                 break;
10537                         }
10538                         ret_val = 0;
10539                         /* update metadata locally */
10540                         imsm_update_metadata_locally(st, u, len);
10541                         /* and possibly remotely */
10542                         if (st->update_tail)
10543                                 append_metadata_update(st, u, len);
10544                         else
10545                                 free(u);
10546                 }
10547                 break;
10548                 case CH_ARRAY_SIZE: {
10549                         struct imsm_update_size_change *u = NULL;
10550                         int len =
10551                                 imsm_create_metadata_update_for_size_change(
10552                                         st, &geo, &u);
10553                         if (len < 1) {
10554                                 dprintf("imsm: Cannot prepare update\n");
10555                                 break;
10556                         }
10557                         ret_val = 0;
10558                         /* update metadata locally */
10559                         imsm_update_metadata_locally(st, u, len);
10560                         /* and possibly remotely */
10561                         if (st->update_tail)
10562                                 append_metadata_update(st, u, len);
10563                         else
10564                                 free(u);
10565                 }
10566                 break;
10567                 default:
10568                         ret_val = 1;
10569                 }
10570         }
10571
10572 exit_imsm_reshape_super:
10573         dprintf("imsm: reshape_super Exit code = %i\n", ret_val);
10574         return ret_val;
10575 }
10576
10577 #define COMPLETED_OK            0
10578 #define COMPLETED_NONE          1
10579 #define COMPLETED_DELAYED       2
10580
10581 static int read_completed(int fd, unsigned long long *val)
10582 {
10583         int ret;
10584         char buf[50];
10585
10586         ret = sysfs_fd_get_str(fd, buf, 50);
10587         if (ret < 0)
10588                 return ret;
10589
10590         ret = COMPLETED_OK;
10591         if (strncmp(buf, "none", 4) == 0) {
10592                 ret = COMPLETED_NONE;
10593         } else if (strncmp(buf, "delayed", 7) == 0) {
10594                 ret = COMPLETED_DELAYED;
10595         } else {
10596                 char *ep;
10597                 *val = strtoull(buf, &ep, 0);
10598                 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
10599                         ret = -1;
10600         }
10601         return ret;
10602 }
10603
10604 /*******************************************************************************
10605  * Function:    wait_for_reshape_imsm
10606  * Description: Function writes new sync_max value and waits until
10607  *              reshape process reach new position
10608  * Parameters:
10609  *      sra             : general array info
10610  *      ndata           : number of disks in new array's layout
10611  * Returns:
10612  *       0 : success,
10613  *       1 : there is no reshape in progress,
10614  *      -1 : fail
10615  ******************************************************************************/
10616 int wait_for_reshape_imsm(struct mdinfo *sra, int ndata)
10617 {
10618         int fd = sysfs_get_fd(sra, NULL, "sync_completed");
10619         int retry = 3;
10620         unsigned long long completed;
10621         /* to_complete : new sync_max position */
10622         unsigned long long to_complete = sra->reshape_progress;
10623         unsigned long long position_to_set = to_complete / ndata;
10624
10625         if (fd < 0) {
10626                 dprintf("cannot open reshape_position\n");
10627                 return 1;
10628         }
10629
10630         do {
10631                 if (sysfs_fd_get_ll(fd, &completed) < 0) {
10632                         if (!retry) {
10633                                 dprintf("cannot read reshape_position (no reshape in progres)\n");
10634                                 close(fd);
10635                                 return 1;
10636                         }
10637                         usleep(30000);
10638                 } else
10639                         break;
10640         } while (retry--);
10641
10642         if (completed > position_to_set) {
10643                 dprintf("wrong next position to set %llu (%llu)\n",
10644                         to_complete, position_to_set);
10645                 close(fd);
10646                 return -1;
10647         }
10648         dprintf("Position set: %llu\n", position_to_set);
10649         if (sysfs_set_num(sra, NULL, "sync_max",
10650                           position_to_set) != 0) {
10651                 dprintf("cannot set reshape position to %llu\n",
10652                         position_to_set);
10653                 close(fd);
10654                 return -1;
10655         }
10656
10657         do {
10658                 int rc;
10659                 char action[20];
10660                 int timeout = 3000;
10661
10662                 sysfs_wait(fd, &timeout);
10663                 if (sysfs_get_str(sra, NULL, "sync_action",
10664                                   action, 20) > 0 &&
10665                                 strncmp(action, "reshape", 7) != 0) {
10666                         if (strncmp(action, "idle", 4) == 0)
10667                                 break;
10668                         close(fd);
10669                         return -1;
10670                 }
10671
10672                 rc = read_completed(fd, &completed);
10673                 if (rc < 0) {
10674                         dprintf("cannot read reshape_position (in loop)\n");
10675                         close(fd);
10676                         return 1;
10677                 } else if (rc == COMPLETED_NONE)
10678                         break;
10679         } while (completed < position_to_set);
10680
10681         close(fd);
10682         return 0;
10683 }
10684
10685 /*******************************************************************************
10686  * Function:    check_degradation_change
10687  * Description: Check that array hasn't become failed.
10688  * Parameters:
10689  *      info    : for sysfs access
10690  *      sources : source disks descriptors
10691  *      degraded: previous degradation level
10692  * Returns:
10693  *      degradation level
10694  ******************************************************************************/
10695 int check_degradation_change(struct mdinfo *info,
10696                              int *sources,
10697                              int degraded)
10698 {
10699         unsigned long long new_degraded;
10700         int rv;
10701
10702         rv = sysfs_get_ll(info, NULL, "degraded", &new_degraded);
10703         if (rv == -1 || (new_degraded != (unsigned long long)degraded)) {
10704                 /* check each device to ensure it is still working */
10705                 struct mdinfo *sd;
10706                 new_degraded = 0;
10707                 for (sd = info->devs ; sd ; sd = sd->next) {
10708                         if (sd->disk.state & (1<<MD_DISK_FAULTY))
10709                                 continue;
10710                         if (sd->disk.state & (1<<MD_DISK_SYNC)) {
10711                                 char sbuf[100];
10712
10713                                 if (sysfs_get_str(info,
10714                                         sd, "state", sbuf, sizeof(sbuf)) < 0 ||
10715                                         strstr(sbuf, "faulty") ||
10716                                         strstr(sbuf, "in_sync") == NULL) {
10717                                         /* this device is dead */
10718                                         sd->disk.state = (1<<MD_DISK_FAULTY);
10719                                         if (sd->disk.raid_disk >= 0 &&
10720                                             sources[sd->disk.raid_disk] >= 0) {
10721                                                 close(sources[
10722                                                         sd->disk.raid_disk]);
10723                                                 sources[sd->disk.raid_disk] =
10724                                                         -1;
10725                                         }
10726                                         new_degraded++;
10727                                 }
10728                         }
10729                 }
10730         }
10731
10732         return new_degraded;
10733 }
10734
10735 /*******************************************************************************
10736  * Function:    imsm_manage_reshape
10737  * Description: Function finds array under reshape and it manages reshape
10738  *              process. It creates stripes backups (if required) and sets
10739  *              checkpoints.
10740  * Parameters:
10741  *      afd             : Backup handle (nattive) - not used
10742  *      sra             : general array info
10743  *      reshape         : reshape parameters - not used
10744  *      st              : supertype structure
10745  *      blocks          : size of critical section [blocks]
10746  *      fds             : table of source device descriptor
10747  *      offsets         : start of array (offest per devices)
10748  *      dests           : not used
10749  *      destfd          : table of destination device descriptor
10750  *      destoffsets     : table of destination offsets (per device)
10751  * Returns:
10752  *      1 : success, reshape is done
10753  *      0 : fail
10754  ******************************************************************************/
10755 static int imsm_manage_reshape(
10756         int afd, struct mdinfo *sra, struct reshape *reshape,
10757         struct supertype *st, unsigned long backup_blocks,
10758         int *fds, unsigned long long *offsets,
10759         int dests, int *destfd, unsigned long long *destoffsets)
10760 {
10761         int ret_val = 0;
10762         struct intel_super *super = st->sb;
10763         struct intel_dev *dv;
10764         unsigned int sector_size = super->sector_size;
10765         struct imsm_dev *dev = NULL;
10766         struct imsm_map *map_src;
10767         int migr_vol_qan = 0;
10768         int ndata, odata; /* [bytes] */
10769         int chunk; /* [bytes] */
10770         struct migr_record *migr_rec;
10771         char *buf = NULL;
10772         unsigned int buf_size; /* [bytes] */
10773         unsigned long long max_position; /* array size [bytes] */
10774         unsigned long long next_step; /* [blocks]/[bytes] */
10775         unsigned long long old_data_stripe_length;
10776         unsigned long long start_src; /* [bytes] */
10777         unsigned long long start; /* [bytes] */
10778         unsigned long long start_buf_shift; /* [bytes] */
10779         int degraded = 0;
10780         int source_layout = 0;
10781
10782         if (!sra)
10783                 return ret_val;
10784
10785         if (!fds || !offsets)
10786                 goto abort;
10787
10788         /* Find volume during the reshape */
10789         for (dv = super->devlist; dv; dv = dv->next) {
10790                 if (dv->dev->vol.migr_type == MIGR_GEN_MIGR
10791                     && dv->dev->vol.migr_state == 1) {
10792                         dev = dv->dev;
10793                         migr_vol_qan++;
10794                 }
10795         }
10796         /* Only one volume can migrate at the same time */
10797         if (migr_vol_qan != 1) {
10798                 pr_err("%s", migr_vol_qan ?
10799                         "Number of migrating volumes greater than 1\n" :
10800                         "There is no volume during migrationg\n");
10801                 goto abort;
10802         }
10803
10804         map_src = get_imsm_map(dev, MAP_1);
10805         if (map_src == NULL)
10806                 goto abort;
10807
10808         ndata = imsm_num_data_members(dev, MAP_0);
10809         odata = imsm_num_data_members(dev, MAP_1);
10810
10811         chunk = __le16_to_cpu(map_src->blocks_per_strip) * 512;
10812         old_data_stripe_length = odata * chunk;
10813
10814         migr_rec = super->migr_rec;
10815
10816         /* initialize migration record for start condition */
10817         if (sra->reshape_progress == 0)
10818                 init_migr_record_imsm(st, dev, sra);
10819         else {
10820                 if (__le32_to_cpu(migr_rec->rec_status) != UNIT_SRC_NORMAL) {
10821                         dprintf("imsm: cannot restart migration when data are present in copy area.\n");
10822                         goto abort;
10823                 }
10824                 /* Save checkpoint to update migration record for current
10825                  * reshape position (in md). It can be farther than current
10826                  * reshape position in metadata.
10827                  */
10828                 if (save_checkpoint_imsm(st, sra, UNIT_SRC_NORMAL) == 1) {
10829                         /* ignore error == 2, this can mean end of reshape here
10830                          */
10831                         dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
10832                         goto abort;
10833                 }
10834         }
10835
10836         /* size for data */
10837         buf_size = __le32_to_cpu(migr_rec->blocks_per_unit) * 512;
10838         /* extend  buffer size for parity disk */
10839         buf_size += __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512;
10840         /* add space for stripe aligment */
10841         buf_size += old_data_stripe_length;
10842         if (posix_memalign((void **)&buf, MAX_SECTOR_SIZE, buf_size)) {
10843                 dprintf("imsm: Cannot allocate checkpoint buffer\n");
10844                 goto abort;
10845         }
10846
10847         max_position = sra->component_size * ndata;
10848         source_layout = imsm_level_to_layout(map_src->raid_level);
10849
10850         while (__le32_to_cpu(migr_rec->curr_migr_unit) <
10851                __le32_to_cpu(migr_rec->num_migr_units)) {
10852                 /* current reshape position [blocks] */
10853                 unsigned long long current_position =
10854                         __le32_to_cpu(migr_rec->blocks_per_unit)
10855                         * __le32_to_cpu(migr_rec->curr_migr_unit);
10856                 unsigned long long border;
10857
10858                 /* Check that array hasn't become failed.
10859                  */
10860                 degraded = check_degradation_change(sra, fds, degraded);
10861                 if (degraded > 1) {
10862                         dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded);
10863                         goto abort;
10864                 }
10865
10866                 next_step = __le32_to_cpu(migr_rec->blocks_per_unit);
10867
10868                 if ((current_position + next_step) > max_position)
10869                         next_step = max_position - current_position;
10870
10871                 start = current_position * 512;
10872
10873                 /* align reading start to old geometry */
10874                 start_buf_shift = start % old_data_stripe_length;
10875                 start_src = start - start_buf_shift;
10876
10877                 border = (start_src / odata) - (start / ndata);
10878                 border /= 512;
10879                 if (border <= __le32_to_cpu(migr_rec->dest_depth_per_unit)) {
10880                         /* save critical stripes to buf
10881                          * start     - start address of current unit
10882                          *             to backup [bytes]
10883                          * start_src - start address of current unit
10884                          *             to backup alligned to source array
10885                          *             [bytes]
10886                          */
10887                         unsigned long long next_step_filler;
10888                         unsigned long long copy_length = next_step * 512;
10889
10890                         /* allign copy area length to stripe in old geometry */
10891                         next_step_filler = ((copy_length + start_buf_shift)
10892                                             % old_data_stripe_length);
10893                         if (next_step_filler)
10894                                 next_step_filler = (old_data_stripe_length
10895                                                     - next_step_filler);
10896                         dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
10897                                 start, start_src, copy_length,
10898                                 start_buf_shift, next_step_filler);
10899
10900                         if (save_stripes(fds, offsets, map_src->num_members,
10901                                          chunk, map_src->raid_level,
10902                                          source_layout, 0, NULL, start_src,
10903                                          copy_length +
10904                                          next_step_filler + start_buf_shift,
10905                                          buf)) {
10906                                 dprintf("imsm: Cannot save stripes to buffer\n");
10907                                 goto abort;
10908                         }
10909                         /* Convert data to destination format and store it
10910                          * in backup general migration area
10911                          */
10912                         if (save_backup_imsm(st, dev, sra,
10913                                 buf + start_buf_shift, copy_length)) {
10914                                 dprintf("imsm: Cannot save stripes to target devices\n");
10915                                 goto abort;
10916                         }
10917                         if (save_checkpoint_imsm(st, sra,
10918                                                  UNIT_SRC_IN_CP_AREA)) {
10919                                 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
10920                                 goto abort;
10921                         }
10922                 } else {
10923                         /* set next step to use whole border area */
10924                         border /= next_step;
10925                         if (border > 1)
10926                                 next_step *= border;
10927                 }
10928                 /* When data backed up, checkpoint stored,
10929                  * kick the kernel to reshape unit of data
10930                  */
10931                 next_step = next_step + sra->reshape_progress;
10932                 /* limit next step to array max position */
10933                 if (next_step > max_position)
10934                         next_step = max_position;
10935                 sysfs_set_num(sra, NULL, "suspend_lo", sra->reshape_progress);
10936                 sysfs_set_num(sra, NULL, "suspend_hi", next_step);
10937                 sra->reshape_progress = next_step;
10938
10939                 /* wait until reshape finish */
10940                 if (wait_for_reshape_imsm(sra, ndata)) {
10941                         dprintf("wait_for_reshape_imsm returned error!\n");
10942                         goto abort;
10943                 }
10944                 if (sigterm)
10945                         goto abort;
10946
10947                 if (save_checkpoint_imsm(st, sra, UNIT_SRC_NORMAL) == 1) {
10948                         /* ignore error == 2, this can mean end of reshape here
10949                          */
10950                         dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
10951                         goto abort;
10952                 }
10953
10954         }
10955
10956         /* clear migr_rec on disks after successful migration */
10957         struct dl *d;
10958
10959         memset(super->migr_rec_buf, 0, MIGR_REC_BUF_SECTORS*sector_size);
10960         for (d = super->disks; d; d = d->next) {
10961                 if (d->index < 0 || is_failed(&d->disk))
10962                         continue;
10963                 unsigned long long dsize;
10964
10965                 get_dev_size(d->fd, NULL, &dsize);
10966                 if (lseek64(d->fd, dsize - MIGR_REC_SECTOR_POSITION*sector_size,
10967                             SEEK_SET) >= 0) {
10968                         if (write(d->fd, super->migr_rec_buf,
10969                             MIGR_REC_BUF_SECTORS*sector_size) !=
10970                             MIGR_REC_BUF_SECTORS*sector_size)
10971                                 perror("Write migr_rec failed");
10972                 }
10973         }
10974
10975         /* return '1' if done */
10976         ret_val = 1;
10977 abort:
10978         free(buf);
10979         /* See Grow.c: abort_reshape() for further explanation */
10980         sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
10981         sysfs_set_num(sra, NULL, "suspend_hi", 0);
10982         sysfs_set_num(sra, NULL, "suspend_lo", 0);
10983
10984         return ret_val;
10985 }
10986
10987 #endif /* MDASSEMBLE */
10988
10989 struct superswitch super_imsm = {
10990 #ifndef MDASSEMBLE
10991         .examine_super  = examine_super_imsm,
10992         .brief_examine_super = brief_examine_super_imsm,
10993         .brief_examine_subarrays = brief_examine_subarrays_imsm,
10994         .export_examine_super = export_examine_super_imsm,
10995         .detail_super   = detail_super_imsm,
10996         .brief_detail_super = brief_detail_super_imsm,
10997         .write_init_super = write_init_super_imsm,
10998         .validate_geometry = validate_geometry_imsm,
10999         .add_to_super   = add_to_super_imsm,
11000         .remove_from_super = remove_from_super_imsm,
11001         .detail_platform = detail_platform_imsm,
11002         .export_detail_platform = export_detail_platform_imsm,
11003         .kill_subarray = kill_subarray_imsm,
11004         .update_subarray = update_subarray_imsm,
11005         .load_container = load_container_imsm,
11006         .default_geometry = default_geometry_imsm,
11007         .get_disk_controller_domain = imsm_get_disk_controller_domain,
11008         .reshape_super  = imsm_reshape_super,
11009         .manage_reshape = imsm_manage_reshape,
11010         .recover_backup = recover_backup_imsm,
11011         .copy_metadata = copy_metadata_imsm,
11012 #endif
11013         .match_home     = match_home_imsm,
11014         .uuid_from_super= uuid_from_super_imsm,
11015         .getinfo_super  = getinfo_super_imsm,
11016         .getinfo_super_disks = getinfo_super_disks_imsm,
11017         .update_super   = update_super_imsm,
11018
11019         .avail_size     = avail_size_imsm,
11020         .min_acceptable_spare_size = min_acceptable_spare_size_imsm,
11021
11022         .compare_super  = compare_super_imsm,
11023
11024         .load_super     = load_super_imsm,
11025         .init_super     = init_super_imsm,
11026         .store_super    = store_super_imsm,
11027         .free_super     = free_super_imsm,
11028         .match_metadata_desc = match_metadata_desc_imsm,
11029         .container_content = container_content_imsm,
11030         .validate_container = validate_container_imsm,
11031
11032         .external       = 1,
11033         .name = "imsm",
11034
11035 #ifndef MDASSEMBLE
11036 /* for mdmon */
11037         .open_new       = imsm_open_new,
11038         .set_array_state= imsm_set_array_state,
11039         .set_disk       = imsm_set_disk,
11040         .sync_metadata  = imsm_sync_metadata,
11041         .activate_spare = imsm_activate_spare,
11042         .process_update = imsm_process_update,
11043         .prepare_update = imsm_prepare_update,
11044 #endif /* MDASSEMBLE */
11045 };