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