super-ddf.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2006-2014 Neil Brown <neilb@suse.de>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neil@brown.name>
  23  *
  24  * Specifications for DDF taken from Common RAID DDF Specification Revision 1.2
  25  * (July 28 2006).  Reused by permission of SNIA.
  26  */
  27
  28 #define HAVE_STDINT_H 1
  29 #include "mdadm.h"
  30 #include "mdmon.h"
  31 #include "sha1.h"
  32 #include <values.h>
  33 #include <stddef.h>
  34
  35 /* a non-official T10 name for creation GUIDs */
  36 static char T10[] = "Linux-MD";
  37
  38 /* DDF timestamps are 1980 based, so we need to add
  39  * second-in-decade-of-seventies to convert to linux timestamps.
  40  * 10 years with 2 leap years.
  41  */
  42 #define DECADE (3600*24*(365*10+2))
  43 unsigned long crc32(
  44         unsigned long crc,
  45         const unsigned char *buf,
  46         unsigned len);
  47
  48 #define DDF_NOTFOUND (~0U)
  49 #define DDF_CONTAINER (DDF_NOTFOUND-1)
  50
  51 /* Default for safe_mode_delay. Same value as for IMSM.
  52  */
  53 static const int DDF_SAFE_MODE_DELAY = 4000;
  54
  55 /* The DDF metadata handling.
  56  * DDF metadata lives at the end of the device.
  57  * The last 512 byte block provides an 'anchor' which is used to locate
  58  * the rest of the metadata which usually lives immediately behind the anchor.
  59  *
  60  * Note:
  61  *  - all multibyte numeric fields are bigendian.
  62  *  - all strings are space padded.
  63  *
  64  */
  65
  66 typedef struct __be16 {
  67         __u16 _v16;
  68 } be16;
  69 #define be16_eq(x, y) ((x)._v16 == (y)._v16)
  70 #define be16_and(x, y) ((x)._v16 & (y)._v16)
  71 #define be16_or(x, y) ((x)._v16 | (y)._v16)
  72 #define be16_clear(x, y) ((x)._v16 &= ~(y)._v16)
  73 #define be16_set(x, y) ((x)._v16 |= (y)._v16)
  74
  75 typedef struct __be32 {
  76         __u32 _v32;
  77 } be32;
  78 #define be32_eq(x, y) ((x)._v32 == (y)._v32)
  79
  80 typedef struct __be64 {
  81         __u64 _v64;
  82 } be64;
  83 #define be64_eq(x, y) ((x)._v64 == (y)._v64)
  84
  85 #define be16_to_cpu(be) __be16_to_cpu((be)._v16)
  86 static inline be16 cpu_to_be16(__u16 x)
  87 {
  88         be16 be = { ._v16 = __cpu_to_be16(x) };
  89         return be;
  90 }
  91
  92 #define be32_to_cpu(be) __be32_to_cpu((be)._v32)
  93 static inline be32 cpu_to_be32(__u32 x)
  94 {
  95         be32 be = { ._v32 = __cpu_to_be32(x) };
  96         return be;
  97 }
  98
  99 #define be64_to_cpu(be) __be64_to_cpu((be)._v64)
 100 static inline be64 cpu_to_be64(__u64 x)
 101 {
 102         be64 be = { ._v64 = __cpu_to_be64(x) };
 103         return be;
 104 }
 105
 106 /* Primary Raid Level (PRL) */
 107 #define DDF_RAID0       0x00
 108 #define DDF_RAID1       0x01
 109 #define DDF_RAID3       0x03
 110 #define DDF_RAID4       0x04
 111 #define DDF_RAID5       0x05
 112 #define DDF_RAID1E      0x11
 113 #define DDF_JBOD        0x0f
 114 #define DDF_CONCAT      0x1f
 115 #define DDF_RAID5E      0x15
 116 #define DDF_RAID5EE     0x25
 117 #define DDF_RAID6       0x06
 118
 119 /* Raid Level Qualifier (RLQ) */
 120 #define DDF_RAID0_SIMPLE        0x00
 121 #define DDF_RAID1_SIMPLE        0x00 /* just 2 devices in this plex */
 122 #define DDF_RAID1_MULTI         0x01 /* exactly 3 devices in this plex */
 123 #define DDF_RAID3_0             0x00 /* parity in first extent */
 124 #define DDF_RAID3_N             0x01 /* parity in last extent */
 125 #define DDF_RAID4_0             0x00 /* parity in first extent */
 126 #define DDF_RAID4_N             0x01 /* parity in last extent */
 127 /* these apply to raid5e and raid5ee as well */
 128 #define DDF_RAID5_0_RESTART     0x00 /* same as 'right asymmetric' - layout 1 */
 129 #define DDF_RAID6_0_RESTART     0x01 /* raid6 different from raid5 here!!! */
 130 #define DDF_RAID5_N_RESTART     0x02 /* same as 'left asymmetric' - layout 0 */
 131 #define DDF_RAID5_N_CONTINUE    0x03 /* same as 'left symmetric' - layout 2 */
 132
 133 #define DDF_RAID1E_ADJACENT     0x00 /* raid10 nearcopies==2 */
 134 #define DDF_RAID1E_OFFSET       0x01 /* raid10 offsetcopies==2 */
 135
 136 /* Secondary RAID Level (SRL) */
 137 #define DDF_2STRIPED    0x00    /* This is weirder than RAID0 !! */
 138 #define DDF_2MIRRORED   0x01
 139 #define DDF_2CONCAT     0x02
 140 #define DDF_2SPANNED    0x03    /* This is also weird - be careful */
 141
 142 /* Magic numbers */
 143 #define DDF_HEADER_MAGIC        cpu_to_be32(0xDE11DE11)
 144 #define DDF_CONTROLLER_MAGIC    cpu_to_be32(0xAD111111)
 145 #define DDF_PHYS_RECORDS_MAGIC  cpu_to_be32(0x22222222)
 146 #define DDF_PHYS_DATA_MAGIC     cpu_to_be32(0x33333333)
 147 #define DDF_VIRT_RECORDS_MAGIC  cpu_to_be32(0xDDDDDDDD)
 148 #define DDF_VD_CONF_MAGIC       cpu_to_be32(0xEEEEEEEE)
 149 #define DDF_SPARE_ASSIGN_MAGIC  cpu_to_be32(0x55555555)
 150 #define DDF_VU_CONF_MAGIC       cpu_to_be32(0x88888888)
 151 #define DDF_VENDOR_LOG_MAGIC    cpu_to_be32(0x01dBEEF0)
 152 #define DDF_BBM_LOG_MAGIC       cpu_to_be32(0xABADB10C)
 153
 154 #define DDF_GUID_LEN    24
 155 #define DDF_REVISION_0  "01.00.00"
 156 #define DDF_REVISION_2  "01.02.00"
 157
 158 struct ddf_header {
 159         be32    magic;          /* DDF_HEADER_MAGIC */
 160         be32    crc;
 161         char    guid[DDF_GUID_LEN];
 162         char    revision[8];    /* 01.02.00 */
 163         be32    seq;            /* starts at '1' */
 164         be32    timestamp;
 165         __u8    openflag;
 166         __u8    foreignflag;
 167         __u8    enforcegroups;
 168         __u8    pad0;           /* 0xff */
 169         __u8    pad1[12];       /* 12 * 0xff */
 170         /* 64 bytes so far */
 171         __u8    header_ext[32]; /* reserved: fill with 0xff */
 172         be64    primary_lba;
 173         be64    secondary_lba;
 174         __u8    type;
 175         __u8    pad2[3];        /* 0xff */
 176         be32    workspace_len;  /* sectors for vendor space -
 177                                  * at least 32768(sectors) */
 178         be64    workspace_lba;
 179         be16    max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
 180         be16    max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
 181         be16    max_partitions; /* i.e. max num of configuration
 182                                    record entries per disk */
 183         be16    config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
 184                                                  *12/512) */
 185         be16    max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
 186         __u8    pad3[54];       /* 0xff */
 187         /* 192 bytes so far */
 188         be32    controller_section_offset;
 189         be32    controller_section_length;
 190         be32    phys_section_offset;
 191         be32    phys_section_length;
 192         be32    virt_section_offset;
 193         be32    virt_section_length;
 194         be32    config_section_offset;
 195         be32    config_section_length;
 196         be32    data_section_offset;
 197         be32    data_section_length;
 198         be32    bbm_section_offset;
 199         be32    bbm_section_length;
 200         be32    diag_space_offset;
 201         be32    diag_space_length;
 202         be32    vendor_offset;
 203         be32    vendor_length;
 204         /* 256 bytes so far */
 205         __u8    pad4[256];      /* 0xff */
 206 };
 207
 208 /* type field */
 209 #define DDF_HEADER_ANCHOR       0x00
 210 #define DDF_HEADER_PRIMARY      0x01
 211 #define DDF_HEADER_SECONDARY    0x02
 212
 213 /* The content of the 'controller section' - global scope */
 214 struct ddf_controller_data {
 215         be32    magic;                  /* DDF_CONTROLLER_MAGIC */
 216         be32    crc;
 217         char    guid[DDF_GUID_LEN];
 218         struct controller_type {
 219                 be16 vendor_id;
 220                 be16 device_id;
 221                 be16 sub_vendor_id;
 222                 be16 sub_device_id;
 223         } type;
 224         char    product_id[16];
 225         __u8    pad[8]; /* 0xff */
 226         __u8    vendor_data[448];
 227 };
 228
 229 /* The content of phys_section - global scope */
 230 struct phys_disk {
 231         be32    magic;          /* DDF_PHYS_RECORDS_MAGIC */
 232         be32    crc;
 233         be16    used_pdes;      /* This is a counter, not a max - the list
 234                                  * of used entries may not be dense */
 235         be16    max_pdes;
 236         __u8    pad[52];
 237         struct phys_disk_entry {
 238                 char    guid[DDF_GUID_LEN];
 239                 be32    refnum;
 240                 be16    type;
 241                 be16    state;
 242                 be64    config_size;    /* DDF structures must be after here */
 243                 char    path[18];       /* Another horrible structure really
 244                                          * but is "used for information
 245                                          * purposes only" */
 246                 __u8    pad[6];
 247         } entries[0];
 248 };
 249
 250 /* phys_disk_entry.type is a bitmap - bigendian remember */
 251 #define DDF_Forced_PD_GUID              1
 252 #define DDF_Active_in_VD                2
 253 #define DDF_Global_Spare                4 /* VD_CONF records are ignored */
 254 #define DDF_Spare                       8 /* overrides Global_spare */
 255 #define DDF_Foreign                     16
 256 #define DDF_Legacy                      32 /* no DDF on this device */
 257
 258 #define DDF_Interface_mask              0xf00
 259 #define DDF_Interface_SCSI              0x100
 260 #define DDF_Interface_SAS               0x200
 261 #define DDF_Interface_SATA              0x300
 262 #define DDF_Interface_FC                0x400
 263
 264 /* phys_disk_entry.state is a bigendian bitmap */
 265 #define DDF_Online                      1
 266 #define DDF_Failed                      2 /* overrides  1,4,8 */
 267 #define DDF_Rebuilding                  4
 268 #define DDF_Transition                  8
 269 #define DDF_SMART                       16
 270 #define DDF_ReadErrors                  32
 271 #define DDF_Missing                     64
 272
 273 /* The content of the virt_section global scope */
 274 struct virtual_disk {
 275         be32    magic;          /* DDF_VIRT_RECORDS_MAGIC */
 276         be32    crc;
 277         be16    populated_vdes;
 278         be16    max_vdes;
 279         __u8    pad[52];
 280         struct virtual_entry {
 281                 char    guid[DDF_GUID_LEN];
 282                 be16    unit;
 283                 __u16   pad0;   /* 0xffff */
 284                 be16    guid_crc;
 285                 be16    type;
 286                 __u8    state;
 287                 __u8    init_state;
 288                 __u8    pad1[14];
 289                 char    name[16];
 290         } entries[0];
 291 };
 292
 293 /* virtual_entry.type is a bitmap - bigendian */
 294 #define DDF_Shared              1
 295 #define DDF_Enforce_Groups      2
 296 #define DDF_Unicode             4
 297 #define DDF_Owner_Valid         8
 298
 299 /* virtual_entry.state is a bigendian bitmap */
 300 #define DDF_state_mask          0x7
 301 #define DDF_state_optimal       0x0
 302 #define DDF_state_degraded      0x1
 303 #define DDF_state_deleted       0x2
 304 #define DDF_state_missing       0x3
 305 #define DDF_state_failed        0x4
 306 #define DDF_state_part_optimal  0x5
 307
 308 #define DDF_state_morphing      0x8
 309 #define DDF_state_inconsistent  0x10
 310
 311 /* virtual_entry.init_state is a bigendian bitmap */
 312 #define DDF_initstate_mask      0x03
 313 #define DDF_init_not            0x00
 314 #define DDF_init_quick          0x01 /* initialisation is progress.
 315                                       * i.e. 'state_inconsistent' */
 316 #define DDF_init_full           0x02
 317
 318 #define DDF_access_mask         0xc0
 319 #define DDF_access_rw           0x00
 320 #define DDF_access_ro           0x80
 321 #define DDF_access_blocked      0xc0
 322
 323 /* The content of the config_section - local scope
 324  * It has multiple records each config_record_len sectors
 325  * They can be vd_config or spare_assign
 326  */
 327
 328 struct vd_config {
 329         be32    magic;          /* DDF_VD_CONF_MAGIC */
 330         be32    crc;
 331         char    guid[DDF_GUID_LEN];
 332         be32    timestamp;
 333         be32    seqnum;
 334         __u8    pad0[24];
 335         be16    prim_elmnt_count;
 336         __u8    chunk_shift;    /* 0 == 512, 1==1024 etc */
 337         __u8    prl;
 338         __u8    rlq;
 339         __u8    sec_elmnt_count;
 340         __u8    sec_elmnt_seq;
 341         __u8    srl;
 342         be64    blocks;         /* blocks per component could be different
 343                                  * on different component devices...(only
 344                                  * for concat I hope) */
 345         be64    array_blocks;   /* blocks in array */
 346         __u8    pad1[8];
 347         be32    spare_refs[8];  /* This is used to detect missing spares.
 348                                  * As we don't have an interface for that
 349                                  * the values are ignored.
 350                                  */
 351         __u8    cache_pol[8];
 352         __u8    bg_rate;
 353         __u8    pad2[3];
 354         __u8    pad3[52];
 355         __u8    pad4[192];
 356         __u8    v0[32]; /* reserved- 0xff */
 357         __u8    v1[32]; /* reserved- 0xff */
 358         __u8    v2[16]; /* reserved- 0xff */
 359         __u8    v3[16]; /* reserved- 0xff */
 360         __u8    vendor[32];
 361         be32    phys_refnum[0]; /* refnum of each disk in sequence */
 362       /*__u64   lba_offset[0];  LBA offset in each phys.  Note extents in a
 363                                 bvd are always the same size */
 364 };
 365 #define LBA_OFFSET(ddf, vd) ((be64 *) &(vd)->phys_refnum[(ddf)->mppe])
 366
 367 /* vd_config.cache_pol[7] is a bitmap */
 368 #define DDF_cache_writeback     1       /* else writethrough */
 369 #define DDF_cache_wadaptive     2       /* only applies if writeback */
 370 #define DDF_cache_readahead     4
 371 #define DDF_cache_radaptive     8       /* only if doing read-ahead */
 372 #define DDF_cache_ifnobatt      16      /* even to write cache if battery is poor */
 373 #define DDF_cache_wallowed      32      /* enable write caching */
 374 #define DDF_cache_rallowed      64      /* enable read caching */
 375
 376 struct spare_assign {
 377         be32    magic;          /* DDF_SPARE_ASSIGN_MAGIC */
 378         be32    crc;
 379         be32    timestamp;
 380         __u8    reserved[7];
 381         __u8    type;
 382         be16    populated;      /* SAEs used */
 383         be16    max;            /* max SAEs */
 384         __u8    pad[8];
 385         struct spare_assign_entry {
 386                 char    guid[DDF_GUID_LEN];
 387                 be16    secondary_element;
 388                 __u8    pad[6];
 389         } spare_ents[0];
 390 };
 391 /* spare_assign.type is a bitmap */
 392 #define DDF_spare_dedicated     0x1     /* else global */
 393 #define DDF_spare_revertible    0x2     /* else committable */
 394 #define DDF_spare_active        0x4     /* else not active */
 395 #define DDF_spare_affinity      0x8     /* enclosure affinity */
 396
 397 /* The data_section contents - local scope */
 398 struct disk_data {
 399         be32    magic;          /* DDF_PHYS_DATA_MAGIC */
 400         be32    crc;
 401         char    guid[DDF_GUID_LEN];
 402         be32    refnum;         /* crc of some magic drive data ... */
 403         __u8    forced_ref;     /* set when above was not result of magic */
 404         __u8    forced_guid;    /* set if guid was forced rather than magic */
 405         __u8    vendor[32];
 406         __u8    pad[442];
 407 };
 408
 409 /* bbm_section content */
 410 struct bad_block_log {
 411         be32    magic;
 412         be32    crc;
 413         be16    entry_count;
 414         be32    spare_count;
 415         __u8    pad[10];
 416         be64    first_spare;
 417         struct mapped_block {
 418                 be64    defective_start;
 419                 be32    replacement_start;
 420                 be16    remap_count;
 421                 __u8    pad[2];
 422         } entries[0];
 423 };
 424
 425 /* Struct for internally holding ddf structures */
 426 /* The DDF structure stored on each device is potentially
 427  * quite different, as some data is global and some is local.
 428  * The global data is:
 429  *   - ddf header
 430  *   - controller_data
 431  *   - Physical disk records
 432  *   - Virtual disk records
 433  * The local data is:
 434  *   - Configuration records
 435  *   - Physical Disk data section
 436  *  (  and Bad block and vendor which I don't care about yet).
 437  *
 438  * The local data is parsed into separate lists as it is read
 439  * and reconstructed for writing.  This means that we only need
 440  * to make config changes once and they are automatically
 441  * propagated to all devices.
 442  * The global (config and disk data) records are each in a list
 443  * of separate data structures.  When writing we find the entry
 444  * or entries applicable to the particular device.
 445  */
 446 struct ddf_super {
 447         struct ddf_header       anchor, primary, secondary;
 448         struct ddf_controller_data controller;
 449         struct ddf_header       *active;
 450         struct phys_disk        *phys;
 451         struct virtual_disk     *virt;
 452         char                    *conf;
 453         int                     pdsize, vdsize;
 454         unsigned int            max_part, mppe, conf_rec_len;
 455         int                     currentdev;
 456         int                     updates_pending;
 457         struct vcl {
 458                 union {
 459                         char space[512];
 460                         struct {
 461                                 struct vcl      *next;
 462                                 unsigned int    vcnum; /* index into ->virt */
 463                                 /* For an array with a secondary level there are
 464                                  * multiple vd_config structures, all with the same
 465                                  * guid but with different sec_elmnt_seq.
 466                                  * One of these structures is in 'conf' below.
 467                                  * The others are in other_bvds, not in any
 468                                  * particular order.
 469                                  */
 470                                 struct vd_config **other_bvds;
 471                                 __u64           *block_sizes; /* NULL if all the same */
 472                         };
 473                 };
 474                 struct vd_config conf;
 475         } *conflist, *currentconf;
 476         struct dl {
 477                 union {
 478                         char space[512];
 479                         struct {
 480                                 struct dl       *next;
 481                                 int major, minor;
 482                                 char *devname;
 483                                 int fd;
 484                                 unsigned long long size; /* sectors */
 485                                 be64 primary_lba; /* sectors */
 486                                 be64 secondary_lba; /* sectors */
 487                                 be64 workspace_lba; /* sectors */
 488                                 int pdnum;      /* index in ->phys */
 489                                 struct spare_assign *spare;
 490                                 void *mdupdate; /* hold metadata update */
 491
 492                                 /* These fields used by auto-layout */
 493                                 int raiddisk; /* slot to fill in autolayout */
 494                                 __u64 esize;
 495                                 int displayed;
 496                         };
 497                 };
 498                 struct disk_data disk;
 499                 struct vcl *vlist[0]; /* max_part in size */
 500         } *dlist, *add_list;
 501 };
 502
 503 #ifndef MDASSEMBLE
 504 static int load_super_ddf_all(struct supertype *st, int fd,
 505                               void **sbp, char *devname);
 506 static int get_svd_state(const struct ddf_super *, const struct vcl *);
 507 static int
 508 validate_geometry_ddf_container(struct supertype *st,
 509                                 int level, int layout, int raiddisks,
 510                                 int chunk, unsigned long long size,
 511                                 unsigned long long data_offset,
 512                                 char *dev, unsigned long long *freesize,
 513                                 int verbose);
 514
 515 static int validate_geometry_ddf_bvd(struct supertype *st,
 516                                      int level, int layout, int raiddisks,
 517                                      int *chunk, unsigned long long size,
 518                                      unsigned long long data_offset,
 519                                      char *dev, unsigned long long *freesize,
 520                                      int verbose);
 521 #endif
 522
 523 static void free_super_ddf(struct supertype *st);
 524 static int all_ff(const char *guid);
 525 static unsigned int get_pd_index_from_refnum(const struct vcl *vc,
 526                                              be32 refnum, unsigned int nmax,
 527                                              const struct vd_config **bvd,
 528                                              unsigned int *idx);
 529 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map);
 530 static void uuid_from_ddf_guid(const char *guid, int uuid[4]);
 531 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
 532 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i);
 533 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map);
 534 static int init_super_ddf_bvd(struct supertype *st,
 535                               mdu_array_info_t *info,
 536                               unsigned long long size,
 537                               char *name, char *homehost,
 538                               int *uuid, unsigned long long data_offset);
 539
 540 #if DEBUG
 541 static void pr_state(struct ddf_super *ddf, const char *msg)
 542 {
 543         unsigned int i;
 544         dprintf("%s: ", msg);
 545         for (i = 0; i < be16_to_cpu(ddf->active->max_vd_entries); i++) {
 546                 if (all_ff(ddf->virt->entries[i].guid))
 547                         continue;
 548                 dprintf_cont("%u(s=%02x i=%02x) ", i,
 549                         ddf->virt->entries[i].state,
 550                         ddf->virt->entries[i].init_state);
 551         }
 552         dprintf_cont("\n");
 553 }
 554 #else
 555 static void pr_state(const struct ddf_super *ddf, const char *msg) {}
 556 #endif
 557
 558 static void _ddf_set_updates_pending(struct ddf_super *ddf, struct vd_config *vc,
 559                                      const char *func)
 560 {
 561         if (vc) {
 562                 vc->timestamp = cpu_to_be32(time(0)-DECADE);
 563                 vc->seqnum = cpu_to_be32(be32_to_cpu(vc->seqnum) + 1);
 564         }
 565         if (ddf->updates_pending)
 566                 return;
 567         ddf->updates_pending = 1;
 568         ddf->active->seq = cpu_to_be32((be32_to_cpu(ddf->active->seq)+1));
 569         pr_state(ddf, func);
 570 }
 571
 572 #define ddf_set_updates_pending(x,v) _ddf_set_updates_pending((x), (v), __func__)
 573
 574 static be32 calc_crc(void *buf, int len)
 575 {
 576         /* crcs are always at the same place as in the ddf_header */
 577         struct ddf_header *ddf = buf;
 578         be32 oldcrc = ddf->crc;
 579         __u32 newcrc;
 580         ddf->crc = cpu_to_be32(0xffffffff);
 581
 582         newcrc = crc32(0, buf, len);
 583         ddf->crc = oldcrc;
 584         /* The crc is stored (like everything) bigendian, so convert
 585          * here for simplicity
 586          */
 587         return cpu_to_be32(newcrc);
 588 }
 589
 590 #define DDF_INVALID_LEVEL 0xff
 591 #define DDF_NO_SECONDARY 0xff
 592 static int err_bad_md_layout(const mdu_array_info_t *array)
 593 {
 594         pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
 595                array->level, array->layout, array->raid_disks);
 596         return -1;
 597 }
 598
 599 static int layout_md2ddf(const mdu_array_info_t *array,
 600                          struct vd_config *conf)
 601 {
 602         be16 prim_elmnt_count = cpu_to_be16(array->raid_disks);
 603         __u8 prl = DDF_INVALID_LEVEL, rlq = 0;
 604         __u8 sec_elmnt_count = 1;
 605         __u8 srl = DDF_NO_SECONDARY;
 606
 607         switch (array->level) {
 608         case LEVEL_LINEAR:
 609                 prl = DDF_CONCAT;
 610                 break;
 611         case 0:
 612                 rlq = DDF_RAID0_SIMPLE;
 613                 prl = DDF_RAID0;
 614                 break;
 615         case 1:
 616                 switch (array->raid_disks) {
 617                 case 2:
 618                         rlq = DDF_RAID1_SIMPLE;
 619                         break;
 620                 case 3:
 621                         rlq = DDF_RAID1_MULTI;
 622                         break;
 623                 default:
 624                         return err_bad_md_layout(array);
 625                 }
 626                 prl = DDF_RAID1;
 627                 break;
 628         case 4:
 629                 if (array->layout != 0)
 630                         return err_bad_md_layout(array);
 631                 rlq = DDF_RAID4_N;
 632                 prl = DDF_RAID4;
 633                 break;
 634         case 5:
 635                 switch (array->layout) {
 636                 case ALGORITHM_LEFT_ASYMMETRIC:
 637                         rlq = DDF_RAID5_N_RESTART;
 638                         break;
 639                 case ALGORITHM_RIGHT_ASYMMETRIC:
 640                         rlq = DDF_RAID5_0_RESTART;
 641                         break;
 642                 case ALGORITHM_LEFT_SYMMETRIC:
 643                         rlq = DDF_RAID5_N_CONTINUE;
 644                         break;
 645                 case ALGORITHM_RIGHT_SYMMETRIC:
 646                         /* not mentioned in standard */
 647                 default:
 648                         return err_bad_md_layout(array);
 649                 }
 650                 prl = DDF_RAID5;
 651                 break;
 652         case 6:
 653                 switch (array->layout) {
 654                 case ALGORITHM_ROTATING_N_RESTART:
 655                         rlq = DDF_RAID5_N_RESTART;
 656                         break;
 657                 case ALGORITHM_ROTATING_ZERO_RESTART:
 658                         rlq = DDF_RAID6_0_RESTART;
 659                         break;
 660                 case ALGORITHM_ROTATING_N_CONTINUE:
 661                         rlq = DDF_RAID5_N_CONTINUE;
 662                         break;
 663                 default:
 664                         return err_bad_md_layout(array);
 665                 }
 666                 prl = DDF_RAID6;
 667                 break;
 668         case 10:
 669                 if (array->raid_disks % 2 == 0 && array->layout == 0x102) {
 670                         rlq = DDF_RAID1_SIMPLE;
 671                         prim_elmnt_count =  cpu_to_be16(2);
 672                         sec_elmnt_count = array->raid_disks / 2;
 673                         srl = DDF_2SPANNED;
 674                         prl = DDF_RAID1;
 675                 } else if (array->raid_disks % 3 == 0
 676                            && array->layout == 0x103) {
 677                         rlq = DDF_RAID1_MULTI;
 678                         prim_elmnt_count =  cpu_to_be16(3);
 679                         sec_elmnt_count = array->raid_disks / 3;
 680                         srl = DDF_2SPANNED;
 681                         prl = DDF_RAID1;
 682                 } else if (array->layout == 0x201) {
 683                         prl = DDF_RAID1E;
 684                         rlq = DDF_RAID1E_OFFSET;
 685                 } else if (array->layout == 0x102) {
 686                         prl = DDF_RAID1E;
 687                         rlq = DDF_RAID1E_ADJACENT;
 688                 } else
 689                         return err_bad_md_layout(array);
 690                 break;
 691         default:
 692                 return err_bad_md_layout(array);
 693         }
 694         conf->prl = prl;
 695         conf->prim_elmnt_count = prim_elmnt_count;
 696         conf->rlq = rlq;
 697         conf->srl = srl;
 698         conf->sec_elmnt_count = sec_elmnt_count;
 699         return 0;
 700 }
 701
 702 static int err_bad_ddf_layout(const struct vd_config *conf)
 703 {
 704         pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
 705                conf->prl, conf->rlq, be16_to_cpu(conf->prim_elmnt_count));
 706         return -1;
 707 }
 708
 709 static int layout_ddf2md(const struct vd_config *conf,
 710                          mdu_array_info_t *array)
 711 {
 712         int level = LEVEL_UNSUPPORTED;
 713         int layout = 0;
 714         int raiddisks = be16_to_cpu(conf->prim_elmnt_count);
 715
 716         if (conf->sec_elmnt_count > 1) {
 717                 /* see also check_secondary() */
 718                 if (conf->prl != DDF_RAID1 ||
 719                     (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED)) {
 720                         pr_err("Unsupported secondary RAID level %u/%u\n",
 721                                conf->prl, conf->srl);
 722                         return -1;
 723                 }
 724                 if (raiddisks == 2 && conf->rlq == DDF_RAID1_SIMPLE)
 725                         layout = 0x102;
 726                 else if  (raiddisks == 3 && conf->rlq == DDF_RAID1_MULTI)
 727                         layout = 0x103;
 728                 else
 729                         return err_bad_ddf_layout(conf);
 730                 raiddisks *= conf->sec_elmnt_count;
 731                 level = 10;
 732                 goto good;
 733         }
 734
 735         switch (conf->prl) {
 736         case DDF_CONCAT:
 737                 level = LEVEL_LINEAR;
 738                 break;
 739         case DDF_RAID0:
 740                 if (conf->rlq != DDF_RAID0_SIMPLE)
 741                         return err_bad_ddf_layout(conf);
 742                 level = 0;
 743                 break;
 744         case DDF_RAID1:
 745                 if (!((conf->rlq == DDF_RAID1_SIMPLE && raiddisks == 2) ||
 746                       (conf->rlq == DDF_RAID1_MULTI && raiddisks == 3)))
 747                         return err_bad_ddf_layout(conf);
 748                 level = 1;
 749                 break;
 750         case DDF_RAID1E:
 751                 if (conf->rlq == DDF_RAID1E_ADJACENT)
 752                         layout = 0x102;
 753                 else if (conf->rlq == DDF_RAID1E_OFFSET)
 754                         layout = 0x201;
 755                 else
 756                         return err_bad_ddf_layout(conf);
 757                 level = 10;
 758                 break;
 759         case DDF_RAID4:
 760                 if (conf->rlq != DDF_RAID4_N)
 761                         return err_bad_ddf_layout(conf);
 762                 level = 4;
 763                 break;
 764         case DDF_RAID5:
 765                 switch (conf->rlq) {
 766                 case DDF_RAID5_N_RESTART:
 767                         layout = ALGORITHM_LEFT_ASYMMETRIC;
 768                         break;
 769                 case DDF_RAID5_0_RESTART:
 770                         layout = ALGORITHM_RIGHT_ASYMMETRIC;
 771                         break;
 772                 case DDF_RAID5_N_CONTINUE:
 773                         layout = ALGORITHM_LEFT_SYMMETRIC;
 774                         break;
 775                 default:
 776                         return err_bad_ddf_layout(conf);
 777                 }
 778                 level = 5;
 779                 break;
 780         case DDF_RAID6:
 781                 switch (conf->rlq) {
 782                 case DDF_RAID5_N_RESTART:
 783                         layout = ALGORITHM_ROTATING_N_RESTART;
 784                         break;
 785                 case DDF_RAID6_0_RESTART:
 786                         layout = ALGORITHM_ROTATING_ZERO_RESTART;
 787                         break;
 788                 case DDF_RAID5_N_CONTINUE:
 789                         layout = ALGORITHM_ROTATING_N_CONTINUE;
 790                         break;
 791                 default:
 792                         return err_bad_ddf_layout(conf);
 793                 }
 794                 level = 6;
 795                 break;
 796         default:
 797                 return err_bad_ddf_layout(conf);
 798         };
 799
 800 good:
 801         array->level = level;
 802         array->layout = layout;
 803         array->raid_disks = raiddisks;
 804         return 0;
 805 }
 806
 807 static int load_ddf_header(int fd, unsigned long long lba,
 808                            unsigned long long size,
 809                            int type,
 810                            struct ddf_header *hdr, struct ddf_header *anchor)
 811 {
 812         /* read a ddf header (primary or secondary) from fd/lba
 813          * and check that it is consistent with anchor
 814          * Need to check:
 815          *   magic, crc, guid, rev, and LBA's header_type, and
 816          *  everything after header_type must be the same
 817          */
 818         if (lba >= size-1)
 819                 return 0;
 820
 821         if (lseek64(fd, lba<<9, 0) < 0)
 822                 return 0;
 823
 824         if (read(fd, hdr, 512) != 512)
 825                 return 0;
 826
 827         if (!be32_eq(hdr->magic, DDF_HEADER_MAGIC)) {
 828                 pr_err("bad header magic\n");
 829                 return 0;
 830         }
 831         if (!be32_eq(calc_crc(hdr, 512), hdr->crc)) {
 832                 pr_err("bad CRC\n");
 833                 return 0;
 834         }
 835         if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
 836             memcmp(anchor->revision, hdr->revision, 8) != 0 ||
 837             !be64_eq(anchor->primary_lba, hdr->primary_lba) ||
 838             !be64_eq(anchor->secondary_lba, hdr->secondary_lba) ||
 839             hdr->type != type ||
 840             memcmp(anchor->pad2, hdr->pad2, 512 -
 841                    offsetof(struct ddf_header, pad2)) != 0) {
 842                 pr_err("header mismatch\n");
 843                 return 0;
 844         }
 845
 846         /* Looks good enough to me... */
 847         return 1;
 848 }
 849
 850 static void *load_section(int fd, struct ddf_super *super, void *buf,
 851                           be32 offset_be, be32 len_be, int check)
 852 {
 853         unsigned long long offset = be32_to_cpu(offset_be);
 854         unsigned long long len = be32_to_cpu(len_be);
 855         int dofree = (buf == NULL);
 856
 857         if (check)
 858                 if (len != 2 && len != 8 && len != 32
 859                     && len != 128 && len != 512)
 860                         return NULL;
 861
 862         if (len > 1024)
 863                 return NULL;
 864         if (!buf && posix_memalign(&buf, 512, len<<9) != 0)
 865                 buf = NULL;
 866
 867         if (!buf)
 868                 return NULL;
 869
 870         if (super->active->type == 1)
 871                 offset += be64_to_cpu(super->active->primary_lba);
 872         else
 873                 offset += be64_to_cpu(super->active->secondary_lba);
 874
 875         if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) {
 876                 if (dofree)
 877                         free(buf);
 878                 return NULL;
 879         }
 880         if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) {
 881                 if (dofree)
 882                         free(buf);
 883                 return NULL;
 884         }
 885         return buf;
 886 }
 887
 888 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
 889 {
 890         unsigned long long dsize;
 891
 892         get_dev_size(fd, NULL, &dsize);
 893
 894         if (lseek64(fd, dsize-512, 0) < 0) {
 895                 if (devname)
 896                         pr_err("Cannot seek to anchor block on %s: %s\n",
 897                                devname, strerror(errno));
 898                 return 1;
 899         }
 900         if (read(fd, &super->anchor, 512) != 512) {
 901                 if (devname)
 902                         pr_err("Cannot read anchor block on %s: %s\n",
 903                                devname, strerror(errno));
 904                 return 1;
 905         }
 906         if (!be32_eq(super->anchor.magic, DDF_HEADER_MAGIC)) {
 907                 if (devname)
 908                         pr_err("no DDF anchor found on %s\n",
 909                                 devname);
 910                 return 2;
 911         }
 912         if (!be32_eq(calc_crc(&super->anchor, 512), super->anchor.crc)) {
 913                 if (devname)
 914                         pr_err("bad CRC on anchor on %s\n",
 915                                 devname);
 916                 return 2;
 917         }
 918         if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
 919             memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
 920                 if (devname)
 921                         pr_err("can only support super revision %.8s and earlier, not %.8s on %s\n",
 922                                 DDF_REVISION_2, super->anchor.revision,devname);
 923                 return 2;
 924         }
 925         super->active = NULL;
 926         if (load_ddf_header(fd, be64_to_cpu(super->anchor.primary_lba),
 927                             dsize >> 9,  1,
 928                             &super->primary, &super->anchor) == 0) {
 929                 if (devname)
 930                         pr_err("Failed to load primary DDF header on %s\n", devname);
 931         } else
 932                 super->active = &super->primary;
 933
 934         if (load_ddf_header(fd, be64_to_cpu(super->anchor.secondary_lba),
 935                             dsize >> 9,  2,
 936                             &super->secondary, &super->anchor)) {
 937                 if (super->active == NULL
 938                     || (be32_to_cpu(super->primary.seq)
 939                         < be32_to_cpu(super->secondary.seq) &&
 940                         !super->secondary.openflag)
 941                     || (be32_to_cpu(super->primary.seq)
 942                         == be32_to_cpu(super->secondary.seq) &&
 943                         super->primary.openflag && !super->secondary.openflag)
 944                         )
 945                         super->active = &super->secondary;
 946         } else if (devname &&
 947                    be64_to_cpu(super->anchor.secondary_lba) != ~(__u64)0)
 948                 pr_err("Failed to load secondary DDF header on %s\n",
 949                        devname);
 950         if (super->active == NULL)
 951                 return 2;
 952         return 0;
 953 }
 954
 955 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
 956 {
 957         void *ok;
 958         ok = load_section(fd, super, &super->controller,
 959                           super->active->controller_section_offset,
 960                           super->active->controller_section_length,
 961                           0);
 962         super->phys = load_section(fd, super, NULL,
 963                                    super->active->phys_section_offset,
 964                                    super->active->phys_section_length,
 965                                    1);
 966         super->pdsize = be32_to_cpu(super->active->phys_section_length) * 512;
 967
 968         super->virt = load_section(fd, super, NULL,
 969                                    super->active->virt_section_offset,
 970                                    super->active->virt_section_length,
 971                                    1);
 972         super->vdsize = be32_to_cpu(super->active->virt_section_length) * 512;
 973         if (!ok ||
 974             !super->phys ||
 975             !super->virt) {
 976                 free(super->phys);
 977                 free(super->virt);
 978                 super->phys = NULL;
 979                 super->virt = NULL;
 980                 return 2;
 981         }
 982         super->conflist = NULL;
 983         super->dlist = NULL;
 984
 985         super->max_part = be16_to_cpu(super->active->max_partitions);
 986         super->mppe = be16_to_cpu(super->active->max_primary_element_entries);
 987         super->conf_rec_len = be16_to_cpu(super->active->config_record_len);
 988         return 0;
 989 }
 990
 991 #define DDF_UNUSED_BVD 0xff
 992 static int alloc_other_bvds(const struct ddf_super *ddf, struct vcl *vcl)
 993 {
 994         unsigned int n_vds = vcl->conf.sec_elmnt_count - 1;
 995         unsigned int i, vdsize;
 996         void *p;
 997         if (n_vds == 0) {
 998                 vcl->other_bvds = NULL;
 999                 return 0;
1000         }
1001         vdsize = ddf->conf_rec_len * 512;
1002         if (posix_memalign(&p, 512, n_vds *
1003                            (vdsize +  sizeof(struct vd_config *))) != 0)
1004                 return -1;
1005         vcl->other_bvds = (struct vd_config **) (p + n_vds * vdsize);
1006         for (i = 0; i < n_vds; i++) {
1007                 vcl->other_bvds[i] = p + i * vdsize;
1008                 memset(vcl->other_bvds[i], 0, vdsize);
1009                 vcl->other_bvds[i]->sec_elmnt_seq = DDF_UNUSED_BVD;
1010         }
1011         return 0;
1012 }
1013
1014 static void add_other_bvd(struct vcl *vcl, struct vd_config *vd,
1015                           unsigned int len)
1016 {
1017         int i;
1018         for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++)
1019                 if (vcl->other_bvds[i]->sec_elmnt_seq == vd->sec_elmnt_seq)
1020                         break;
1021
1022         if (i < vcl->conf.sec_elmnt_count-1) {
1023                 if (be32_to_cpu(vd->seqnum) <=
1024                     be32_to_cpu(vcl->other_bvds[i]->seqnum))
1025                         return;
1026         } else {
1027                 for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++)
1028                         if (vcl->other_bvds[i]->sec_elmnt_seq == DDF_UNUSED_BVD)
1029                                 break;
1030                 if (i == vcl->conf.sec_elmnt_count-1) {
1031                         pr_err("no space for sec level config %u, count is %u\n",
1032                                vd->sec_elmnt_seq, vcl->conf.sec_elmnt_count);
1033                         return;
1034                 }
1035         }
1036         memcpy(vcl->other_bvds[i], vd, len);
1037 }
1038
1039 static int load_ddf_local(int fd, struct ddf_super *super,
1040                           char *devname, int keep)
1041 {
1042         struct dl *dl;
1043         struct stat stb;
1044         char *conf;
1045         unsigned int i;
1046         unsigned int confsec;
1047         int vnum;
1048         unsigned int max_virt_disks =
1049                 be16_to_cpu(super->active->max_vd_entries);
1050         unsigned long long dsize;
1051
1052         /* First the local disk info */
1053         if (posix_memalign((void**)&dl, 512,
1054                            sizeof(*dl) +
1055                            (super->max_part) * sizeof(dl->vlist[0])) != 0) {
1056                 pr_err("could not allocate disk info buffer\n");
1057                 return 1;
1058         }
1059
1060         load_section(fd, super, &dl->disk,
1061                      super->active->data_section_offset,
1062                      super->active->data_section_length,
1063                      0);
1064         dl->devname = devname ? xstrdup(devname) : NULL;
1065
1066         fstat(fd, &stb);
1067         dl->major = major(stb.st_rdev);
1068         dl->minor = minor(stb.st_rdev);
1069         dl->next = super->dlist;
1070         dl->fd = keep ? fd : -1;
1071
1072         dl->size = 0;
1073         if (get_dev_size(fd, devname, &dsize))
1074                 dl->size = dsize >> 9;
1075         /* If the disks have different sizes, the LBAs will differ
1076          * between phys disks.
1077          * At this point here, the values in super->active must be valid
1078          * for this phys disk. */
1079         dl->primary_lba = super->active->primary_lba;
1080         dl->secondary_lba = super->active->secondary_lba;
1081         dl->workspace_lba = super->active->workspace_lba;
1082         dl->spare = NULL;
1083         for (i = 0 ; i < super->max_part ; i++)
1084                 dl->vlist[i] = NULL;
1085         super->dlist = dl;
1086         dl->pdnum = -1;
1087         for (i = 0; i < be16_to_cpu(super->active->max_pd_entries); i++)
1088                 if (memcmp(super->phys->entries[i].guid,
1089                            dl->disk.guid, DDF_GUID_LEN) == 0)
1090                         dl->pdnum = i;
1091
1092         /* Now the config list. */
1093         /* 'conf' is an array of config entries, some of which are
1094          * probably invalid.  Those which are good need to be copied into
1095          * the conflist
1096          */
1097
1098         conf = load_section(fd, super, super->conf,
1099                             super->active->config_section_offset,
1100                             super->active->config_section_length,
1101                             0);
1102         super->conf = conf;
1103         vnum = 0;
1104         for (confsec = 0;
1105              confsec < be32_to_cpu(super->active->config_section_length);
1106              confsec += super->conf_rec_len) {
1107                 struct vd_config *vd =
1108                         (struct vd_config *)((char*)conf + confsec*512);
1109                 struct vcl *vcl;
1110
1111                 if (be32_eq(vd->magic, DDF_SPARE_ASSIGN_MAGIC)) {
1112                         if (dl->spare)
1113                                 continue;
1114                         if (posix_memalign((void**)&dl->spare, 512,
1115                                            super->conf_rec_len*512) != 0) {
1116                                 pr_err("could not allocate spare info buf\n");
1117                                 return 1;
1118                         }
1119
1120                         memcpy(dl->spare, vd, super->conf_rec_len*512);
1121                         continue;
1122                 }
1123                 if (!be32_eq(vd->magic, DDF_VD_CONF_MAGIC))
1124                         /* Must be vendor-unique - I cannot handle those */
1125                         continue;
1126
1127                 for (vcl = super->conflist; vcl; vcl = vcl->next) {
1128                         if (memcmp(vcl->conf.guid,
1129                                    vd->guid, DDF_GUID_LEN) == 0)
1130                                 break;
1131                 }
1132
1133                 if (vcl) {
1134                         dl->vlist[vnum++] = vcl;
1135                         if (vcl->other_bvds != NULL &&
1136                             vcl->conf.sec_elmnt_seq != vd->sec_elmnt_seq) {
1137                                 add_other_bvd(vcl, vd, super->conf_rec_len*512);
1138                                 continue;
1139                         }
1140                         if (be32_to_cpu(vd->seqnum) <=
1141                             be32_to_cpu(vcl->conf.seqnum))
1142                                 continue;
1143                 } else {
1144                         if (posix_memalign((void**)&vcl, 512,
1145                                            (super->conf_rec_len*512 +
1146                                             offsetof(struct vcl, conf))) != 0) {
1147                                 pr_err("could not allocate vcl buf\n");
1148                                 return 1;
1149                         }
1150                         vcl->next = super->conflist;
1151                         vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1152                         vcl->conf.sec_elmnt_count = vd->sec_elmnt_count;
1153                         if (alloc_other_bvds(super, vcl) != 0) {
1154                                 pr_err("could not allocate other bvds\n");
1155                                 free(vcl);
1156                                 return 1;
1157                         };
1158                         super->conflist = vcl;
1159                         dl->vlist[vnum++] = vcl;
1160                 }
1161                 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
1162                 for (i=0; i < max_virt_disks ; i++)
1163                         if (memcmp(super->virt->entries[i].guid,
1164                                    vcl->conf.guid, DDF_GUID_LEN)==0)
1165                                 break;
1166                 if (i < max_virt_disks)
1167                         vcl->vcnum = i;
1168         }
1169
1170         return 0;
1171 }
1172
1173 static int load_super_ddf(struct supertype *st, int fd,
1174                           char *devname)
1175 {
1176         unsigned long long dsize;
1177         struct ddf_super *super;
1178         int rv;
1179
1180         if (get_dev_size(fd, devname, &dsize) == 0)
1181                 return 1;
1182
1183         if (test_partition(fd))
1184                 /* DDF is not allowed on partitions */
1185                 return 1;
1186
1187         /* 32M is a lower bound */
1188         if (dsize <= 32*1024*1024) {
1189                 if (devname)
1190                         pr_err("%s is too small for ddf: size is %llu sectors.\n",
1191                                devname, dsize>>9);
1192                 return 1;
1193         }
1194         if (dsize & 511) {
1195                 if (devname)
1196                         pr_err("%s is an odd size for ddf: size is %llu bytes.\n",
1197                                devname, dsize);
1198                 return 1;
1199         }
1200
1201         free_super_ddf(st);
1202
1203         if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
1204                 pr_err("malloc of %zu failed.\n",
1205                         sizeof(*super));
1206                 return 1;
1207         }
1208         memset(super, 0, sizeof(*super));
1209
1210         rv = load_ddf_headers(fd, super, devname);
1211         if (rv) {
1212                 free(super);
1213                 return rv;
1214         }
1215
1216         /* Have valid headers and have chosen the best. Let's read in the rest*/
1217
1218         rv = load_ddf_global(fd, super, devname);
1219
1220         if (rv) {
1221                 if (devname)
1222                         pr_err("Failed to load all information sections on %s\n", devname);
1223                 free(super);
1224                 return rv;
1225         }
1226
1227         rv = load_ddf_local(fd, super, devname, 0);
1228
1229         if (rv) {
1230                 if (devname)
1231                         pr_err("Failed to load all information sections on %s\n", devname);
1232                 free(super);
1233                 return rv;
1234         }
1235
1236         /* Should possibly check the sections .... */
1237
1238         st->sb = super;
1239         if (st->ss == NULL) {
1240                 st->ss = &super_ddf;
1241                 st->minor_version = 0;
1242                 st->max_devs = 512;
1243         }
1244         return 0;
1245
1246 }
1247
1248 static void free_super_ddf(struct supertype *st)
1249 {
1250         struct ddf_super *ddf = st->sb;
1251         if (ddf == NULL)
1252                 return;
1253         free(ddf->phys);
1254         free(ddf->virt);
1255         free(ddf->conf);
1256         while (ddf->conflist) {
1257                 struct vcl *v = ddf->conflist;
1258                 ddf->conflist = v->next;
1259                 if (v->block_sizes)
1260                         free(v->block_sizes);
1261                 if (v->other_bvds)
1262                         /*
1263                            v->other_bvds[0] points to beginning of buffer,
1264                            see alloc_other_bvds()
1265                         */
1266                         free(v->other_bvds[0]);
1267                 free(v);
1268         }
1269         while (ddf->dlist) {
1270                 struct dl *d = ddf->dlist;
1271                 ddf->dlist = d->next;
1272                 if (d->fd >= 0)
1273                         close(d->fd);
1274                 if (d->spare)
1275                         free(d->spare);
1276                 free(d);
1277         }
1278         while (ddf->add_list) {
1279                 struct dl *d = ddf->add_list;
1280                 ddf->add_list = d->next;
1281                 if (d->fd >= 0)
1282                         close(d->fd);
1283                 if (d->spare)
1284                         free(d->spare);
1285                 free(d);
1286         }
1287         free(ddf);
1288         st->sb = NULL;
1289 }
1290
1291 static struct supertype *match_metadata_desc_ddf(char *arg)
1292 {
1293         /* 'ddf' only supports containers */
1294         struct supertype *st;
1295         if (strcmp(arg, "ddf") != 0 &&
1296             strcmp(arg, "default") != 0
1297                 )
1298                 return NULL;
1299
1300         st = xcalloc(1, sizeof(*st));
1301         st->ss = &super_ddf;
1302         st->max_devs = 512;
1303         st->minor_version = 0;
1304         st->sb = NULL;
1305         return st;
1306 }
1307
1308 #ifndef MDASSEMBLE
1309
1310 static mapping_t ddf_state[] = {
1311         { "Optimal", 0},
1312         { "Degraded", 1},
1313         { "Deleted", 2},
1314         { "Missing", 3},
1315         { "Failed", 4},
1316         { "Partially Optimal", 5},
1317         { "-reserved-", 6},
1318         { "-reserved-", 7},
1319         { NULL, 0}
1320 };
1321
1322 static mapping_t ddf_init_state[] = {
1323         { "Not Initialised", 0},
1324         { "QuickInit in Progress", 1},
1325         { "Fully Initialised", 2},
1326         { "*UNKNOWN*", 3},
1327         { NULL, 0}
1328 };
1329 static mapping_t ddf_access[] = {
1330         { "Read/Write", 0},
1331         { "Reserved", 1},
1332         { "Read Only", 2},
1333         { "Blocked (no access)", 3},
1334         { NULL ,0}
1335 };
1336
1337 static mapping_t ddf_level[] = {
1338         { "RAID0", DDF_RAID0},
1339         { "RAID1", DDF_RAID1},
1340         { "RAID3", DDF_RAID3},
1341         { "RAID4", DDF_RAID4},
1342         { "RAID5", DDF_RAID5},
1343         { "RAID1E",DDF_RAID1E},
1344         { "JBOD",  DDF_JBOD},
1345         { "CONCAT",DDF_CONCAT},
1346         { "RAID5E",DDF_RAID5E},
1347         { "RAID5EE",DDF_RAID5EE},
1348         { "RAID6", DDF_RAID6},
1349         { NULL, 0}
1350 };
1351 static mapping_t ddf_sec_level[] = {
1352         { "Striped", DDF_2STRIPED},
1353         { "Mirrored", DDF_2MIRRORED},
1354         { "Concat", DDF_2CONCAT},
1355         { "Spanned", DDF_2SPANNED},
1356         { NULL, 0}
1357 };
1358 #endif
1359
1360 static int all_ff(const char *guid)
1361 {
1362         int i;
1363         for (i = 0; i < DDF_GUID_LEN; i++)
1364                 if (guid[i] != (char)0xff)
1365                         return 0;
1366         return 1;
1367 }
1368
1369 static const char *guid_str(const char *guid)
1370 {
1371         static char buf[DDF_GUID_LEN*2+1];
1372         int i;
1373         char *p = buf;
1374         for (i = 0; i < DDF_GUID_LEN; i++) {
1375                 unsigned char c = guid[i];
1376                 if (c >= 32 && c < 127)
1377                         p += sprintf(p, "%c", c);
1378                 else
1379                         p += sprintf(p, "%02x", c);
1380         }
1381         *p = '\0';
1382         return (const char *) buf;
1383 }
1384
1385 #ifndef MDASSEMBLE
1386 static void print_guid(char *guid, int tstamp)
1387 {
1388         /* A GUIDs are part (or all) ASCII and part binary.
1389          * They tend to be space padded.
1390          * We print the GUID in HEX, then in parentheses add
1391          * any initial ASCII sequence, and a possible
1392          * time stamp from bytes 16-19
1393          */
1394         int l = DDF_GUID_LEN;
1395         int i;
1396
1397         for (i=0 ; i<DDF_GUID_LEN ; i++) {
1398                 if ((i&3)==0 && i != 0) printf(":");
1399                 printf("%02X", guid[i]&255);
1400         }
1401
1402         printf("\n                  (");
1403         while (l && guid[l-1] == ' ')
1404                 l--;
1405         for (i=0 ; i<l ; i++) {
1406                 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1407                         fputc(guid[i], stdout);
1408                 else
1409                         break;
1410         }
1411         if (tstamp) {
1412                 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1413                 char tbuf[100];
1414                 struct tm *tm;
1415                 tm = localtime(&then);
1416                 strftime(tbuf, 100, " %D %T",tm);
1417                 fputs(tbuf, stdout);
1418         }
1419         printf(")");
1420 }
1421
1422 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1423 {
1424         int crl = sb->conf_rec_len;
1425         struct vcl *vcl;
1426
1427         for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1428                 unsigned int i;
1429                 struct vd_config *vc = &vcl->conf;
1430
1431                 if (!be32_eq(calc_crc(vc, crl*512), vc->crc))
1432                         continue;
1433                 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1434                         continue;
1435
1436                 /* Ok, we know about this VD, let's give more details */
1437                 printf(" Raid Devices[%d] : %d (", n,
1438                        be16_to_cpu(vc->prim_elmnt_count));
1439                 for (i = 0; i < be16_to_cpu(vc->prim_elmnt_count); i++) {
1440                         int j;
1441                         int cnt = be16_to_cpu(sb->phys->max_pdes);
1442                         for (j=0; j<cnt; j++)
1443                                 if (be32_eq(vc->phys_refnum[i],
1444                                             sb->phys->entries[j].refnum))
1445                                         break;
1446                         if (i) printf(" ");
1447                         if (j < cnt)
1448                                 printf("%d", j);
1449                         else
1450                                 printf("--");
1451                         printf("@%lluK", (unsigned long long) be64_to_cpu(LBA_OFFSET(sb, vc)[i])/2);
1452                 }
1453                 printf(")\n");
1454                 if (vc->chunk_shift != 255)
1455                         printf("   Chunk Size[%d] : %d sectors\n", n,
1456                                1 << vc->chunk_shift);
1457                 printf("   Raid Level[%d] : %s\n", n,
1458                        map_num(ddf_level, vc->prl)?:"-unknown-");
1459                 if (vc->sec_elmnt_count != 1) {
1460                         printf("  Secondary Position[%d] : %d of %d\n", n,
1461                                vc->sec_elmnt_seq, vc->sec_elmnt_count);
1462                         printf("  Secondary Level[%d] : %s\n", n,
1463                                map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1464                 }
1465                 printf("  Device Size[%d] : %llu\n", n,
1466                        be64_to_cpu(vc->blocks)/2);
1467                 printf("   Array Size[%d] : %llu\n", n,
1468                        be64_to_cpu(vc->array_blocks)/2);
1469         }
1470 }
1471
1472 static void examine_vds(struct ddf_super *sb)
1473 {
1474         int cnt = be16_to_cpu(sb->virt->populated_vdes);
1475         unsigned int i;
1476         printf("  Virtual Disks : %d\n", cnt);
1477
1478         for (i = 0; i < be16_to_cpu(sb->virt->max_vdes); i++) {
1479                 struct virtual_entry *ve = &sb->virt->entries[i];
1480                 if (all_ff(ve->guid))
1481                         continue;
1482                 printf("\n");
1483                 printf("      VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1484                 printf("\n");
1485                 printf("         unit[%d] : %d\n", i, be16_to_cpu(ve->unit));
1486                 printf("        state[%d] : %s, %s%s\n", i,
1487                        map_num(ddf_state, ve->state & 7),
1488                        (ve->state & DDF_state_morphing) ? "Morphing, ": "",
1489                        (ve->state & DDF_state_inconsistent)? "Not Consistent" : "Consistent");
1490                 printf("   init state[%d] : %s\n", i,
1491                        map_num(ddf_init_state, ve->init_state&DDF_initstate_mask));
1492                 printf("       access[%d] : %s\n", i,
1493                        map_num(ddf_access, (ve->init_state & DDF_access_mask) >> 6));
1494                 printf("         Name[%d] : %.16s\n", i, ve->name);
1495                 examine_vd(i, sb, ve->guid);
1496         }
1497         if (cnt) printf("\n");
1498 }
1499
1500 static void examine_pds(struct ddf_super *sb)
1501 {
1502         int cnt = be16_to_cpu(sb->phys->max_pdes);
1503         int i;
1504         struct dl *dl;
1505         int unlisted = 0;
1506         printf(" Physical Disks : %d\n", cnt);
1507         printf("      Number    RefNo      Size       Device      Type/State\n");
1508
1509         for (dl = sb->dlist; dl; dl = dl->next)
1510                 dl->displayed = 0;
1511
1512         for (i=0 ; i<cnt ; i++) {
1513                 struct phys_disk_entry *pd = &sb->phys->entries[i];
1514                 int type = be16_to_cpu(pd->type);
1515                 int state = be16_to_cpu(pd->state);
1516
1517                 if (be32_to_cpu(pd->refnum) == 0xffffffff)
1518                         /* Not in use */
1519                         continue;
1520                 //printf("      PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1521                 //printf("\n");
1522                 printf("       %3d    %08x  ", i,
1523                        be32_to_cpu(pd->refnum));
1524                 printf("%8lluK ",
1525                        be64_to_cpu(pd->config_size)>>1);
1526                 for (dl = sb->dlist; dl ; dl = dl->next) {
1527                         if (be32_eq(dl->disk.refnum, pd->refnum)) {
1528                                 char *dv = map_dev(dl->major, dl->minor, 0);
1529                                 if (dv) {
1530                                         printf("%-15s", dv);
1531                                         break;
1532                                 }
1533                         }
1534                 }
1535                 if (!dl)
1536                         printf("%15s","");
1537                 else
1538                         dl->displayed = 1;
1539                 printf(" %s%s%s%s%s",
1540                        (type&2) ? "active":"",
1541                        (type&4) ? "Global-Spare":"",
1542                        (type&8) ? "spare" : "",
1543                        (type&16)? ", foreign" : "",
1544                        (type&32)? "pass-through" : "");
1545                 if (state & DDF_Failed)
1546                         /* This over-rides these three */
1547                         state &= ~(DDF_Online|DDF_Rebuilding|DDF_Transition);
1548                 printf("/%s%s%s%s%s%s%s",
1549                        (state&1)? "Online": "Offline",
1550                        (state&2)? ", Failed": "",
1551                        (state&4)? ", Rebuilding": "",
1552                        (state&8)? ", in-transition": "",
1553                        (state&16)? ", SMART-errors": "",
1554                        (state&32)? ", Unrecovered-Read-Errors": "",
1555                        (state&64)? ", Missing" : "");
1556                 printf("\n");
1557         }
1558         for (dl = sb->dlist; dl; dl = dl->next) {
1559                 char *dv;
1560                 if (dl->displayed)
1561                         continue;
1562                 if (!unlisted)
1563                         printf(" Physical disks not in metadata!:\n");
1564                 unlisted = 1;
1565                 dv = map_dev(dl->major, dl->minor, 0);
1566                 printf("   %08x %s\n", be32_to_cpu(dl->disk.refnum),
1567                        dv ? dv : "-unknown-");
1568         }
1569         if (unlisted)
1570                 printf("\n");
1571 }
1572
1573 static void examine_super_ddf(struct supertype *st, char *homehost)
1574 {
1575         struct ddf_super *sb = st->sb;
1576
1577         printf("          Magic : %08x\n", be32_to_cpu(sb->anchor.magic));
1578         printf("        Version : %.8s\n", sb->anchor.revision);
1579         printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1580         printf("\n");
1581         printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1582         printf("\n");
1583         printf("            Seq : %08x\n", be32_to_cpu(sb->active->seq));
1584         printf("  Redundant hdr : %s\n", (be32_eq(sb->secondary.magic,
1585                                                  DDF_HEADER_MAGIC)
1586                                           ?"yes" : "no"));
1587         examine_vds(sb);
1588         examine_pds(sb);
1589 }
1590
1591 static unsigned int get_vd_num_of_subarray(struct supertype *st)
1592 {
1593         /*
1594          * Figure out the VD number for this supertype.
1595          * Returns DDF_CONTAINER for the container itself,
1596          * and DDF_NOTFOUND on error.
1597          */
1598         struct ddf_super *ddf = st->sb;
1599         struct mdinfo *sra;
1600         char *sub, *end;
1601         unsigned int vcnum;
1602
1603         if (*st->container_devnm == '\0')
1604                 return DDF_CONTAINER;
1605
1606         sra = sysfs_read(-1, st->devnm, GET_VERSION);
1607         if (!sra || sra->array.major_version != -1 ||
1608             sra->array.minor_version != -2 ||
1609             !is_subarray(sra->text_version))
1610                 return DDF_NOTFOUND;
1611
1612         sub = strchr(sra->text_version + 1, '/');
1613         if (sub != NULL)
1614                 vcnum = strtoul(sub + 1, &end, 10);
1615         if (sub == NULL || *sub == '\0' || *end != '\0' ||
1616             vcnum >= be16_to_cpu(ddf->active->max_vd_entries))
1617                 return DDF_NOTFOUND;
1618
1619         return vcnum;
1620 }
1621
1622 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1623 {
1624         /* We just write a generic DDF ARRAY entry
1625          */
1626         struct mdinfo info;
1627         char nbuf[64];
1628         getinfo_super_ddf(st, &info, NULL);
1629         fname_from_uuid(st, &info, nbuf, ':');
1630
1631         printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1632 }
1633
1634 static void brief_examine_subarrays_ddf(struct supertype *st, int verbose)
1635 {
1636         /* We write a DDF ARRAY member entry for each vd, identifying container
1637          * by uuid and member by unit number and uuid.
1638          */
1639         struct ddf_super *ddf = st->sb;
1640         struct mdinfo info;
1641         unsigned int i;
1642         char nbuf[64];
1643         getinfo_super_ddf(st, &info, NULL);
1644         fname_from_uuid(st, &info, nbuf, ':');
1645
1646         for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
1647                 struct virtual_entry *ve = &ddf->virt->entries[i];
1648                 struct vcl vcl;
1649                 char nbuf1[64];
1650                 char namebuf[17];
1651                 if (all_ff(ve->guid))
1652                         continue;
1653                 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1654                 ddf->currentconf =&vcl;
1655                 vcl.vcnum = i;
1656                 uuid_from_super_ddf(st, info.uuid);
1657                 fname_from_uuid(st, &info, nbuf1, ':');
1658                 _ddf_array_name(namebuf, ddf, i);
1659                 printf("ARRAY%s%s container=%s member=%d UUID=%s\n",
1660                        namebuf[0] == '\0' ? "" : " /dev/md/", namebuf,
1661                        nbuf+5, i, nbuf1+5);
1662         }
1663 }
1664
1665 static void export_examine_super_ddf(struct supertype *st)
1666 {
1667         struct mdinfo info;
1668         char nbuf[64];
1669         getinfo_super_ddf(st, &info, NULL);
1670         fname_from_uuid(st, &info, nbuf, ':');
1671         printf("MD_METADATA=ddf\n");
1672         printf("MD_LEVEL=container\n");
1673         printf("MD_UUID=%s\n", nbuf+5);
1674         printf("MD_DEVICES=%u\n",
1675                 be16_to_cpu(((struct ddf_super *)st->sb)->phys->used_pdes));
1676 }
1677
1678 static int copy_metadata_ddf(struct supertype *st, int from, int to)
1679 {
1680         void *buf;
1681         unsigned long long dsize, offset;
1682         int bytes;
1683         struct ddf_header *ddf;
1684         int written = 0;
1685
1686         /* The meta consists of an anchor, a primary, and a secondary.
1687          * This all lives at the end of the device.
1688          * So it is easiest to find the earliest of primary and
1689          * secondary, and copy everything from there.
1690          *
1691          * Anchor is 512 from end.  It contains primary_lba and secondary_lba
1692          * we choose one of those
1693          */
1694
1695         if (posix_memalign(&buf, 4096, 4096) != 0)
1696                 return 1;
1697
1698         if (!get_dev_size(from, NULL, &dsize))
1699                 goto err;
1700
1701         if (lseek64(from, dsize-512, 0) < 0)
1702                 goto err;
1703         if (read(from, buf, 512) != 512)
1704                 goto err;
1705         ddf = buf;
1706         if (!be32_eq(ddf->magic, DDF_HEADER_MAGIC) ||
1707             !be32_eq(calc_crc(ddf, 512), ddf->crc) ||
1708             (memcmp(ddf->revision, DDF_REVISION_0, 8) != 0 &&
1709              memcmp(ddf->revision, DDF_REVISION_2, 8) != 0))
1710                 goto err;
1711
1712         offset = dsize - 512;
1713         if ((be64_to_cpu(ddf->primary_lba) << 9) < offset)
1714                 offset = be64_to_cpu(ddf->primary_lba) << 9;
1715         if ((be64_to_cpu(ddf->secondary_lba) << 9) < offset)
1716                 offset = be64_to_cpu(ddf->secondary_lba) << 9;
1717
1718         bytes = dsize - offset;
1719
1720         if (lseek64(from, offset, 0) < 0 ||
1721             lseek64(to, offset, 0) < 0)
1722                 goto err;
1723         while (written < bytes) {
1724                 int n = bytes - written;
1725                 if (n > 4096)
1726                         n = 4096;
1727                 if (read(from, buf, n) != n)
1728                         goto err;
1729                 if (write(to, buf, n) != n)
1730                         goto err;
1731                 written += n;
1732         }
1733         free(buf);
1734         return 0;
1735 err:
1736         free(buf);
1737         return 1;
1738 }
1739
1740 static void detail_super_ddf(struct supertype *st, char *homehost)
1741 {
1742         struct ddf_super *sb = st->sb;
1743         int cnt = be16_to_cpu(sb->virt->populated_vdes);
1744
1745         printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1746         printf("\n");
1747         printf("            Seq : %08x\n", be32_to_cpu(sb->active->seq));
1748         printf("  Virtual Disks : %d\n", cnt);
1749         printf("\n");
1750 }
1751 #endif
1752
1753 static const char *vendors_with_variable_volume_UUID[] = {
1754         "LSI      ",
1755 };
1756
1757 static int volume_id_is_reliable(const struct ddf_super *ddf)
1758 {
1759         int n = ARRAY_SIZE(vendors_with_variable_volume_UUID);
1760         int i;
1761         for (i = 0; i < n; i++)
1762                 if (!memcmp(ddf->controller.guid,
1763                         vendors_with_variable_volume_UUID[i], 8))
1764                 return 0;
1765         return 1;
1766 }
1767
1768 static void uuid_of_ddf_subarray(const struct ddf_super *ddf,
1769                                  unsigned int vcnum, int uuid[4])
1770 {
1771         char buf[DDF_GUID_LEN+18], sha[20], *p;
1772         struct sha1_ctx ctx;
1773         if (volume_id_is_reliable(ddf)) {
1774                 uuid_from_ddf_guid(ddf->virt->entries[vcnum].guid, uuid);
1775                 return;
1776         }
1777         /*
1778          * Some fake RAID BIOSes (in particular, LSI ones) change the
1779          * VD GUID at every boot. These GUIDs are not suitable for
1780          * identifying an array. Luckily the header GUID appears to
1781          * remain constant.
1782          * We construct a pseudo-UUID from the header GUID and those
1783          * properties of the subarray that we expect to remain constant.
1784          */
1785         memset(buf, 0, sizeof(buf));
1786         p = buf;
1787         memcpy(p, ddf->anchor.guid, DDF_GUID_LEN);
1788         p += DDF_GUID_LEN;
1789         memcpy(p, ddf->virt->entries[vcnum].name, 16);
1790         p += 16;
1791         *((__u16 *) p) = vcnum;
1792         sha1_init_ctx(&ctx);
1793         sha1_process_bytes(buf, sizeof(buf), &ctx);
1794         sha1_finish_ctx(&ctx, sha);
1795         memcpy(uuid, sha, 4*4);
1796 }
1797
1798 #ifndef MDASSEMBLE
1799 static void brief_detail_super_ddf(struct supertype *st)
1800 {
1801         struct mdinfo info;
1802         char nbuf[64];
1803         struct ddf_super *ddf = st->sb;
1804         unsigned int vcnum = get_vd_num_of_subarray(st);
1805         if (vcnum == DDF_CONTAINER)
1806                 uuid_from_super_ddf(st, info.uuid);
1807         else if (vcnum == DDF_NOTFOUND)
1808                 return;
1809         else
1810                 uuid_of_ddf_subarray(ddf, vcnum, info.uuid);
1811         fname_from_uuid(st, &info, nbuf,':');
1812         printf(" UUID=%s", nbuf + 5);
1813 }
1814 #endif
1815
1816 static int match_home_ddf(struct supertype *st, char *homehost)
1817 {
1818         /* It matches 'this' host if the controller is a
1819          * Linux-MD controller with vendor_data matching
1820          * the hostname.  It would be nice if we could
1821          * test against controller found in /sys or somewhere...
1822          */
1823         struct ddf_super *ddf = st->sb;
1824         unsigned int len;
1825
1826         if (!homehost)
1827                 return 0;
1828         len = strlen(homehost);
1829
1830         return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1831                 len < sizeof(ddf->controller.vendor_data) &&
1832                 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1833                 ddf->controller.vendor_data[len] == 0);
1834 }
1835
1836 #ifndef MDASSEMBLE
1837 static int find_index_in_bvd(const struct ddf_super *ddf,
1838                              const struct vd_config *conf, unsigned int n,
1839                              unsigned int *n_bvd)
1840 {
1841         /*
1842          * Find the index of the n-th valid physical disk in this BVD.
1843          * Unused entries can be sprinkled in with the used entries,
1844          * but don't count.
1845          */
1846         unsigned int i, j;
1847         for (i = 0, j = 0;
1848              i < ddf->mppe && j < be16_to_cpu(conf->prim_elmnt_count);
1849              i++) {
1850                 if (be32_to_cpu(conf->phys_refnum[i]) != 0xffffffff) {
1851                         if (n == j) {
1852                                 *n_bvd = i;
1853                                 return 1;
1854                         }
1855                         j++;
1856                 }
1857         }
1858         dprintf("couldn't find BVD member %u (total %u)\n",
1859                 n, be16_to_cpu(conf->prim_elmnt_count));
1860         return 0;
1861 }
1862
1863 /* Given a member array instance number, and a raid disk within that instance,
1864  * find the vd_config structure.  The offset of the given disk in the phys_refnum
1865  * table is returned in n_bvd.
1866  * For two-level members with a secondary raid level the vd_config for
1867  * the appropriate BVD is returned.
1868  * The return value is always &vlc->conf, where vlc is returned in last pointer.
1869  */
1870 static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst,
1871                                    unsigned int n,
1872                                    unsigned int *n_bvd, struct vcl **vcl)
1873 {
1874         struct vcl *v;
1875
1876         for (v = ddf->conflist; v; v = v->next) {
1877                 unsigned int nsec, ibvd = 0;
1878                 struct vd_config *conf;
1879                 if (inst != v->vcnum)
1880                         continue;
1881                 conf = &v->conf;
1882                 if (conf->sec_elmnt_count == 1) {
1883                         if (find_index_in_bvd(ddf, conf, n, n_bvd)) {
1884                                 *vcl = v;
1885                                 return conf;
1886                         } else
1887                                 goto bad;
1888                 }
1889                 if (v->other_bvds == NULL) {
1890                         pr_err("BUG: other_bvds is NULL, nsec=%u\n",
1891                                conf->sec_elmnt_count);
1892                         goto bad;
1893                 }
1894                 nsec = n / be16_to_cpu(conf->prim_elmnt_count);
1895                 if (conf->sec_elmnt_seq != nsec) {
1896                         for (ibvd = 1; ibvd < conf->sec_elmnt_count; ibvd++) {
1897                                 if (v->other_bvds[ibvd-1]->sec_elmnt_seq
1898                                     == nsec)
1899                                         break;
1900                         }
1901                         if (ibvd == conf->sec_elmnt_count)
1902                                 goto bad;
1903                         conf = v->other_bvds[ibvd-1];
1904                 }
1905                 if (!find_index_in_bvd(ddf, conf,
1906                                        n - nsec*conf->sec_elmnt_count, n_bvd))
1907                         goto bad;
1908                 dprintf("found disk %u as member %u in bvd %d of array %u\n",
1909                         n, *n_bvd, ibvd, inst);
1910                 *vcl = v;
1911                 return conf;
1912         }
1913 bad:
1914         pr_err("Could't find disk %d in array %u\n", n, inst);
1915         return NULL;
1916 }
1917 #endif
1918
1919 static int find_phys(const struct ddf_super *ddf, be32 phys_refnum)
1920 {
1921         /* Find the entry in phys_disk which has the given refnum
1922          * and return it's index
1923          */
1924         unsigned int i;
1925         for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++)
1926                 if (be32_eq(ddf->phys->entries[i].refnum, phys_refnum))
1927                         return i;
1928         return -1;
1929 }
1930
1931 static void uuid_from_ddf_guid(const char *guid, int uuid[4])
1932 {
1933         char buf[20];
1934         struct sha1_ctx ctx;
1935         sha1_init_ctx(&ctx);
1936         sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1937         sha1_finish_ctx(&ctx, buf);
1938         memcpy(uuid, buf, 4*4);
1939 }
1940
1941 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1942 {
1943         /* The uuid returned here is used for:
1944          *  uuid to put into bitmap file (Create, Grow)
1945          *  uuid for backup header when saving critical section (Grow)
1946          *  comparing uuids when re-adding a device into an array
1947          *    In these cases the uuid required is that of the data-array,
1948          *    not the device-set.
1949          *  uuid to recognise same set when adding a missing device back
1950          *    to an array.   This is a uuid for the device-set.
1951          *
1952          * For each of these we can make do with a truncated
1953          * or hashed uuid rather than the original, as long as
1954          * everyone agrees.
1955          * In the case of SVD we assume the BVD is of interest,
1956          * though that might be the case if a bitmap were made for
1957          * a mirrored SVD - worry about that later.
1958          * So we need to find the VD configuration record for the
1959          * relevant BVD and extract the GUID and Secondary_Element_Seq.
1960          * The first 16 bytes of the sha1 of these is used.
1961          */
1962         struct ddf_super *ddf = st->sb;
1963         struct vcl *vcl = ddf->currentconf;
1964
1965         if (vcl)
1966                 uuid_of_ddf_subarray(ddf, vcl->vcnum, uuid);
1967         else
1968                 uuid_from_ddf_guid(ddf->anchor.guid, uuid);
1969 }
1970
1971 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map)
1972 {
1973         struct ddf_super *ddf = st->sb;
1974         int map_disks = info->array.raid_disks;
1975         __u32 *cptr;
1976
1977         if (ddf->currentconf) {
1978                 getinfo_super_ddf_bvd(st, info, map);
1979                 return;
1980         }
1981         memset(info, 0, sizeof(*info));
1982
1983         info->array.raid_disks    = be16_to_cpu(ddf->phys->used_pdes);
1984         info->array.level         = LEVEL_CONTAINER;
1985         info->array.layout        = 0;
1986         info->array.md_minor      = -1;
1987         cptr = (__u32 *)(ddf->anchor.guid + 16);
1988         info->array.ctime         = DECADE + __be32_to_cpu(*cptr);
1989
1990         info->array.chunk_size    = 0;
1991         info->container_enough    = 1;
1992
1993         info->disk.major          = 0;
1994         info->disk.minor          = 0;
1995         if (ddf->dlist) {
1996                 struct phys_disk_entry *pde = NULL;
1997                 info->disk.number = be32_to_cpu(ddf->dlist->disk.refnum);
1998                 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1999
2000                 info->data_offset = be64_to_cpu(ddf->phys->
2001                                                   entries[info->disk.raid_disk].
2002                                                   config_size);
2003                 info->component_size = ddf->dlist->size - info->data_offset;
2004                 if (info->disk.raid_disk >= 0)
2005                         pde = ddf->phys->entries + info->disk.raid_disk;
2006                 if (pde &&
2007                     !(be16_to_cpu(pde->state) & DDF_Failed) &&
2008                     !(be16_to_cpu(pde->state) & DDF_Missing))
2009                         info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2010                 else
2011                         info->disk.state = 1 << MD_DISK_FAULTY;
2012
2013         } else {
2014                 /* There should always be a dlist, but just in case...*/
2015                 info->disk.number = -1;
2016                 info->disk.raid_disk = -1;
2017                 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2018         }
2019         info->events = be32_to_cpu(ddf->active->seq);
2020         info->array.utime = DECADE + be32_to_cpu(ddf->active->timestamp);
2021
2022         info->recovery_start = MaxSector;
2023         info->reshape_active = 0;
2024         info->recovery_blocked = 0;
2025         info->name[0] = 0;
2026
2027         info->array.major_version = -1;
2028         info->array.minor_version = -2;
2029         strcpy(info->text_version, "ddf");
2030         info->safe_mode_delay = 0;
2031
2032         uuid_from_super_ddf(st, info->uuid);
2033
2034         if (map) {
2035                 int i, e = 0;
2036                 int max = be16_to_cpu(ddf->phys->max_pdes);
2037                 for (i = e = 0 ; i < map_disks ; i++, e++) {
2038                         while (e < max &&
2039                                be32_to_cpu(ddf->phys->entries[e].refnum) == 0xffffffff)
2040                                 e++;
2041                         if (i < info->array.raid_disks && e < max &&
2042                             !(be16_to_cpu(ddf->phys->entries[e].state)
2043                               & DDF_Failed))
2044                                 map[i] = 1;
2045                         else
2046                                 map[i] = 0;
2047                 }
2048         }
2049 }
2050
2051 /* size of name must be at least 17 bytes! */
2052 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i)
2053 {
2054         int j;
2055         memcpy(name, ddf->virt->entries[i].name, 16);
2056         name[16] = 0;
2057         for(j = 0; j < 16; j++)
2058                 if (name[j] == ' ')
2059                         name[j] = 0;
2060 }
2061
2062 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map)
2063 {
2064         struct ddf_super *ddf = st->sb;
2065         struct vcl *vc = ddf->currentconf;
2066         int cd = ddf->currentdev;
2067         int n_prim;
2068         int j;
2069         struct dl *dl = NULL;
2070         int map_disks = info->array.raid_disks;
2071         __u32 *cptr;
2072         struct vd_config *conf;
2073
2074         memset(info, 0, sizeof(*info));
2075         if (layout_ddf2md(&vc->conf, &info->array) == -1)
2076                 return;
2077         info->array.md_minor      = -1;
2078         cptr = (__u32 *)(vc->conf.guid + 16);
2079         info->array.ctime         = DECADE + __be32_to_cpu(*cptr);
2080         info->array.utime         = DECADE + be32_to_cpu(vc->conf.timestamp);
2081         info->array.chunk_size    = 512 << vc->conf.chunk_shift;
2082         info->custom_array_size   = be64_to_cpu(vc->conf.array_blocks);
2083
2084         conf = &vc->conf;
2085         n_prim = be16_to_cpu(conf->prim_elmnt_count);
2086         if (conf->sec_elmnt_count > 1 && cd >= n_prim) {
2087                 int ibvd = cd / n_prim - 1;
2088                 cd %= n_prim;
2089                 conf = vc->other_bvds[ibvd];
2090         }
2091
2092         if (cd >= 0 && (unsigned)cd < ddf->mppe) {
2093                 info->data_offset =
2094                         be64_to_cpu(LBA_OFFSET(ddf, conf)[cd]);
2095                 if (vc->block_sizes)
2096                         info->component_size = vc->block_sizes[cd];
2097                 else
2098                         info->component_size = be64_to_cpu(conf->blocks);
2099
2100                 for (dl = ddf->dlist; dl ; dl = dl->next)
2101                         if (be32_eq(dl->disk.refnum, conf->phys_refnum[cd]))
2102                                 break;
2103         }
2104
2105         info->disk.major = 0;
2106         info->disk.minor = 0;
2107         info->disk.state = 0;
2108         if (dl && dl->pdnum >= 0) {
2109                 info->disk.major = dl->major;
2110                 info->disk.minor = dl->minor;
2111                 info->disk.raid_disk = cd + conf->sec_elmnt_seq
2112                         * be16_to_cpu(conf->prim_elmnt_count);
2113                 info->disk.number = dl->pdnum;
2114                 info->disk.state = 0;
2115                 if (info->disk.number >= 0 &&
2116                     (be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Online) &&
2117                     !(be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Failed))
2118                         info->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2119                 info->events = be32_to_cpu(ddf->active->seq);
2120         }
2121
2122         info->container_member = ddf->currentconf->vcnum;
2123
2124         info->recovery_start = MaxSector;
2125         info->resync_start = 0;
2126         info->reshape_active = 0;
2127         info->recovery_blocked = 0;
2128         if (!(ddf->virt->entries[info->container_member].state
2129               & DDF_state_inconsistent)  &&
2130             (ddf->virt->entries[info->container_member].init_state
2131              & DDF_initstate_mask)
2132             == DDF_init_full)
2133                 info->resync_start = MaxSector;
2134
2135         uuid_from_super_ddf(st, info->uuid);
2136
2137         info->array.major_version = -1;
2138         info->array.minor_version = -2;
2139         sprintf(info->text_version, "/%s/%d",
2140                 st->container_devnm,
2141                 info->container_member);
2142         info->safe_mode_delay = DDF_SAFE_MODE_DELAY;
2143
2144         _ddf_array_name(info->name, ddf, info->container_member);
2145
2146         if (map)
2147                 for (j = 0; j < map_disks; j++) {
2148                         map[j] = 0;
2149                         if (j <  info->array.raid_disks) {
2150                                 int i = find_phys(ddf, vc->conf.phys_refnum[j]);
2151                                 if (i >= 0 &&
2152                                     (be16_to_cpu(ddf->phys->entries[i].state)
2153                                      & DDF_Online) &&
2154                                     !(be16_to_cpu(ddf->phys->entries[i].state)
2155                                       & DDF_Failed))
2156                                         map[i] = 1;
2157                         }
2158                 }
2159 }
2160
2161 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
2162                             char *update,
2163                             char *devname, int verbose,
2164                             int uuid_set, char *homehost)
2165 {
2166         /* For 'assemble' and 'force' we need to return non-zero if any
2167          * change was made.  For others, the return value is ignored.
2168          * Update options are:
2169          *  force-one : This device looks a bit old but needs to be included,
2170          *        update age info appropriately.
2171          *  assemble: clear any 'faulty' flag to allow this device to
2172          *              be assembled.
2173          *  force-array: Array is degraded but being forced, mark it clean
2174          *         if that will be needed to assemble it.
2175          *
2176          *  newdev:  not used ????
2177          *  grow:  Array has gained a new device - this is currently for
2178          *              linear only
2179          *  resync: mark as dirty so a resync will happen.
2180          *  uuid:  Change the uuid of the array to match what is given
2181          *  homehost:  update the recorded homehost
2182          *  name:  update the name - preserving the homehost
2183          *  _reshape_progress: record new reshape_progress position.
2184          *
2185          * Following are not relevant for this version:
2186          *  sparc2.2 : update from old dodgey metadata
2187          *  super-minor: change the preferred_minor number
2188          *  summaries:  update redundant counters.
2189          */
2190         int rv = 0;
2191 //      struct ddf_super *ddf = st->sb;
2192 //      struct vd_config *vd = find_vdcr(ddf, info->container_member);
2193 //      struct virtual_entry *ve = find_ve(ddf);
2194
2195         /* we don't need to handle "force-*" or "assemble" as
2196          * there is no need to 'trick' the kernel.  When the metadata is
2197          * first updated to activate the array, all the implied modifications
2198          * will just happen.
2199          */
2200
2201         if (strcmp(update, "grow") == 0) {
2202                 /* FIXME */
2203         } else if (strcmp(update, "resync") == 0) {
2204 //              info->resync_checkpoint = 0;
2205         } else if (strcmp(update, "homehost") == 0) {
2206                 /* homehost is stored in controller->vendor_data,
2207                  * or it is when we are the vendor
2208                  */
2209 //              if (info->vendor_is_local)
2210 //                      strcpy(ddf->controller.vendor_data, homehost);
2211                 rv = -1;
2212         } else if (strcmp(update, "name") == 0) {
2213                 /* name is stored in virtual_entry->name */
2214 //              memset(ve->name, ' ', 16);
2215 //              strncpy(ve->name, info->name, 16);
2216                 rv = -1;
2217         } else if (strcmp(update, "_reshape_progress") == 0) {
2218                 /* We don't support reshape yet */
2219         } else if (strcmp(update, "assemble") == 0 ) {
2220                 /* Do nothing, just succeed */
2221                 rv = 0;
2222         } else
2223                 rv = -1;
2224
2225 //      update_all_csum(ddf);
2226
2227         return rv;
2228 }
2229
2230 static void make_header_guid(char *guid)
2231 {
2232         be32 stamp;
2233         /* Create a DDF Header of Virtual Disk GUID */
2234
2235         /* 24 bytes of fiction required.
2236          * first 8 are a 'vendor-id'  - "Linux-MD"
2237          * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2238          * Remaining 8 random number plus timestamp
2239          */
2240         memcpy(guid, T10, sizeof(T10));
2241         stamp = cpu_to_be32(0xdeadbeef);
2242         memcpy(guid+8, &stamp, 4);
2243         stamp = cpu_to_be32(0);
2244         memcpy(guid+12, &stamp, 4);
2245         stamp = cpu_to_be32(time(0) - DECADE);
2246         memcpy(guid+16, &stamp, 4);
2247         stamp._v32 = random32();
2248         memcpy(guid+20, &stamp, 4);
2249 }
2250
2251 static unsigned int find_unused_vde(const struct ddf_super *ddf)
2252 {
2253         unsigned int i;
2254         for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2255                 if (all_ff(ddf->virt->entries[i].guid))
2256                         return i;
2257         }
2258         return DDF_NOTFOUND;
2259 }
2260
2261 static unsigned int find_vde_by_name(const struct ddf_super *ddf,
2262                                      const char *name)
2263 {
2264         unsigned int i;
2265         if (name == NULL)
2266                 return DDF_NOTFOUND;
2267         for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2268                 if (all_ff(ddf->virt->entries[i].guid))
2269                         continue;
2270                 if (!strncmp(name, ddf->virt->entries[i].name,
2271                              sizeof(ddf->virt->entries[i].name)))
2272                         return i;
2273         }
2274         return DDF_NOTFOUND;
2275 }
2276
2277 #ifndef MDASSEMBLE
2278 static unsigned int find_vde_by_guid(const struct ddf_super *ddf,
2279                                      const char *guid)
2280 {
2281         unsigned int i;
2282         if (guid == NULL || all_ff(guid))
2283                 return DDF_NOTFOUND;
2284         for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++)
2285                 if (!memcmp(ddf->virt->entries[i].guid, guid, DDF_GUID_LEN))
2286                         return i;
2287         return DDF_NOTFOUND;
2288 }
2289 #endif
2290
2291 static int init_super_ddf(struct supertype *st,
2292                           mdu_array_info_t *info,
2293                           unsigned long long size, char *name, char *homehost,
2294                           int *uuid, unsigned long long data_offset)
2295 {
2296         /* This is primarily called by Create when creating a new array.
2297          * We will then get add_to_super called for each component, and then
2298          * write_init_super called to write it out to each device.
2299          * For DDF, Create can create on fresh devices or on a pre-existing
2300          * array.
2301          * To create on a pre-existing array a different method will be called.
2302          * This one is just for fresh drives.
2303          *
2304          * We need to create the entire 'ddf' structure which includes:
2305          *  DDF headers - these are easy.
2306          *  Controller data - a Sector describing this controller .. not that
2307          *                    this is a controller exactly.
2308          *  Physical Disk Record - one entry per device, so
2309          *                         leave plenty of space.
2310          *  Virtual Disk Records - again, just leave plenty of space.
2311          *                         This just lists VDs, doesn't give details.
2312          *  Config records - describe the VDs that use this disk
2313          *  DiskData  - describes 'this' device.
2314          *  BadBlockManagement - empty
2315          *  Diag Space - empty
2316          *  Vendor Logs - Could we put bitmaps here?
2317          *
2318          */
2319         struct ddf_super *ddf;
2320         char hostname[17];
2321         int hostlen;
2322         int max_phys_disks, max_virt_disks;
2323         unsigned long long sector;
2324         int clen;
2325         int i;
2326         int pdsize, vdsize;
2327         struct phys_disk *pd;
2328         struct virtual_disk *vd;
2329
2330         if (st->sb)
2331                 return init_super_ddf_bvd(st, info, size, name, homehost, uuid,
2332                                           data_offset);
2333
2334         if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
2335                 pr_err("could not allocate superblock\n");
2336                 return 0;
2337         }
2338         memset(ddf, 0, sizeof(*ddf));
2339         st->sb = ddf;
2340
2341         if (info == NULL) {
2342                 /* zeroing superblock */
2343                 return 0;
2344         }
2345
2346         /* At least 32MB *must* be reserved for the ddf.  So let's just
2347          * start 32MB from the end, and put the primary header there.
2348          * Don't do secondary for now.
2349          * We don't know exactly where that will be yet as it could be
2350          * different on each device.  So just set up the lengths.
2351          */
2352
2353         ddf->anchor.magic = DDF_HEADER_MAGIC;
2354         make_header_guid(ddf->anchor.guid);
2355
2356         memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
2357         ddf->anchor.seq = cpu_to_be32(1);
2358         ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
2359         ddf->anchor.openflag = 0xFF;
2360         ddf->anchor.foreignflag = 0;
2361         ddf->anchor.enforcegroups = 0; /* Is this best?? */
2362         ddf->anchor.pad0 = 0xff;
2363         memset(ddf->anchor.pad1, 0xff, 12);
2364         memset(ddf->anchor.header_ext, 0xff, 32);
2365         ddf->anchor.primary_lba = cpu_to_be64(~(__u64)0);
2366         ddf->anchor.secondary_lba = cpu_to_be64(~(__u64)0);
2367         ddf->anchor.type = DDF_HEADER_ANCHOR;
2368         memset(ddf->anchor.pad2, 0xff, 3);
2369         ddf->anchor.workspace_len = cpu_to_be32(32768); /* Must be reserved */
2370         /* Put this at bottom of 32M reserved.. */
2371         ddf->anchor.workspace_lba = cpu_to_be64(~(__u64)0);
2372         max_phys_disks = 1023;   /* Should be enough, 4095 is also allowed */
2373         ddf->anchor.max_pd_entries = cpu_to_be16(max_phys_disks);
2374         max_virt_disks = 255; /* 15, 63, 255, 1024, 4095 are all allowed */
2375         ddf->anchor.max_vd_entries = cpu_to_be16(max_virt_disks);
2376         ddf->max_part = 64;
2377         ddf->anchor.max_partitions = cpu_to_be16(ddf->max_part);
2378         ddf->mppe = 256; /* 16, 64, 256, 1024, 4096 are all allowed */
2379         ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
2380         ddf->anchor.config_record_len = cpu_to_be16(ddf->conf_rec_len);
2381         ddf->anchor.max_primary_element_entries = cpu_to_be16(ddf->mppe);
2382         memset(ddf->anchor.pad3, 0xff, 54);
2383         /* Controller section is one sector long immediately
2384          * after the ddf header */
2385         sector = 1;
2386         ddf->anchor.controller_section_offset = cpu_to_be32(sector);
2387         ddf->anchor.controller_section_length = cpu_to_be32(1);
2388         sector += 1;
2389
2390         /* phys is 8 sectors after that */
2391         pdsize = ROUND_UP(sizeof(struct phys_disk) +
2392                           sizeof(struct phys_disk_entry)*max_phys_disks,
2393                           512);
2394         switch(pdsize/512) {
2395         case 2: case 8: case 32: case 128: case 512: break;
2396         default: abort();
2397         }
2398         ddf->anchor.phys_section_offset = cpu_to_be32(sector);
2399         ddf->anchor.phys_section_length =
2400                 cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
2401         sector += pdsize/512;
2402
2403         /* virt is another 32 sectors */
2404         vdsize = ROUND_UP(sizeof(struct virtual_disk) +
2405                           sizeof(struct virtual_entry) * max_virt_disks,
2406                           512);
2407         switch(vdsize/512) {
2408         case 2: case 8: case 32: case 128: case 512: break;
2409         default: abort();
2410         }
2411         ddf->anchor.virt_section_offset = cpu_to_be32(sector);
2412         ddf->anchor.virt_section_length =
2413                 cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
2414         sector += vdsize/512;
2415
2416         clen = ddf->conf_rec_len * (ddf->max_part+1);
2417         ddf->anchor.config_section_offset = cpu_to_be32(sector);
2418         ddf->anchor.config_section_length = cpu_to_be32(clen);
2419         sector += clen;
2420
2421         ddf->anchor.data_section_offset = cpu_to_be32(sector);
2422         ddf->anchor.data_section_length = cpu_to_be32(1);
2423         sector += 1;
2424
2425         ddf->anchor.bbm_section_length = cpu_to_be32(0);
2426         ddf->anchor.bbm_section_offset = cpu_to_be32(0xFFFFFFFF);
2427         ddf->anchor.diag_space_length = cpu_to_be32(0);
2428         ddf->anchor.diag_space_offset = cpu_to_be32(0xFFFFFFFF);
2429         ddf->anchor.vendor_length = cpu_to_be32(0);
2430         ddf->anchor.vendor_offset = cpu_to_be32(0xFFFFFFFF);
2431
2432         memset(ddf->anchor.pad4, 0xff, 256);
2433
2434         memcpy(&ddf->primary, &ddf->anchor, 512);
2435         memcpy(&ddf->secondary, &ddf->anchor, 512);
2436
2437         ddf->primary.openflag = 1; /* I guess.. */
2438         ddf->primary.type = DDF_HEADER_PRIMARY;
2439
2440         ddf->secondary.openflag = 1; /* I guess.. */
2441         ddf->secondary.type = DDF_HEADER_SECONDARY;
2442
2443         ddf->active = &ddf->primary;
2444
2445         ddf->controller.magic = DDF_CONTROLLER_MAGIC;
2446
2447         /* 24 more bytes of fiction required.
2448          * first 8 are a 'vendor-id'  - "Linux-MD"
2449          * Remaining 16 are serial number.... maybe a hostname would do?
2450          */
2451         memcpy(ddf->controller.guid, T10, sizeof(T10));
2452         gethostname(hostname, sizeof(hostname));
2453         hostname[sizeof(hostname) - 1] = 0;
2454         hostlen = strlen(hostname);
2455         memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
2456         for (i = strlen(T10) ; i+hostlen < 24; i++)
2457                 ddf->controller.guid[i] = ' ';
2458
2459         ddf->controller.type.vendor_id = cpu_to_be16(0xDEAD);
2460         ddf->controller.type.device_id = cpu_to_be16(0xBEEF);
2461         ddf->controller.type.sub_vendor_id = cpu_to_be16(0);
2462         ddf->controller.type.sub_device_id = cpu_to_be16(0);
2463         memcpy(ddf->controller.product_id, "What Is My PID??", 16);
2464         memset(ddf->controller.pad, 0xff, 8);
2465         memset(ddf->controller.vendor_data, 0xff, 448);
2466         if (homehost && strlen(homehost) < 440)
2467                 strcpy((char*)ddf->controller.vendor_data, homehost);
2468
2469         if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
2470                 pr_err("could not allocate pd\n");
2471                 return 0;
2472         }
2473         ddf->phys = pd;
2474         ddf->pdsize = pdsize;
2475
2476         memset(pd, 0xff, pdsize);
2477         memset(pd, 0, sizeof(*pd));
2478         pd->magic = DDF_PHYS_RECORDS_MAGIC;
2479         pd->used_pdes = cpu_to_be16(0);
2480         pd->max_pdes = cpu_to_be16(max_phys_disks);
2481         memset(pd->pad, 0xff, 52);
2482         for (i = 0; i < max_phys_disks; i++)
2483                 memset(pd->entries[i].guid, 0xff, DDF_GUID_LEN);
2484
2485         if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
2486                 pr_err("could not allocate vd\n");
2487                 return 0;
2488         }
2489         ddf->virt = vd;
2490         ddf->vdsize = vdsize;
2491         memset(vd, 0, vdsize);
2492         vd->magic = DDF_VIRT_RECORDS_MAGIC;
2493         vd->populated_vdes = cpu_to_be16(0);
2494         vd->max_vdes = cpu_to_be16(max_virt_disks);
2495         memset(vd->pad, 0xff, 52);
2496
2497         for (i=0; i<max_virt_disks; i++)
2498                 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
2499
2500         st->sb = ddf;
2501         ddf_set_updates_pending(ddf, NULL);
2502         return 1;
2503 }
2504
2505 static int chunk_to_shift(int chunksize)
2506 {
2507         return ffs(chunksize/512)-1;
2508 }
2509
2510 #ifndef MDASSEMBLE
2511 struct extent {
2512         unsigned long long start, size;
2513 };
2514 static int cmp_extent(const void *av, const void *bv)
2515 {
2516         const struct extent *a = av;
2517         const struct extent *b = bv;
2518         if (a->start < b->start)
2519                 return -1;
2520         if (a->start > b->start)
2521                 return 1;
2522         return 0;
2523 }
2524
2525 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
2526 {
2527         /* Find a list of used extents on the given physical device
2528          * (dnum) of the given ddf.
2529          * Return a malloced array of 'struct extent'
2530          */
2531         struct extent *rv;
2532         int n = 0;
2533         unsigned int i;
2534         __u16 state;
2535
2536         if (dl->pdnum < 0)
2537                 return NULL;
2538         state = be16_to_cpu(ddf->phys->entries[dl->pdnum].state);
2539
2540         if ((state & (DDF_Online|DDF_Failed|DDF_Missing)) != DDF_Online)
2541                 return NULL;
2542
2543         rv = xmalloc(sizeof(struct extent) * (ddf->max_part + 2));
2544
2545         for (i = 0; i < ddf->max_part; i++) {
2546                 const struct vd_config *bvd;
2547                 unsigned int ibvd;
2548                 struct vcl *v = dl->vlist[i];
2549                 if (v == NULL ||
2550                     get_pd_index_from_refnum(v, dl->disk.refnum, ddf->mppe,
2551                                              &bvd, &ibvd) == DDF_NOTFOUND)
2552                         continue;
2553                 rv[n].start = be64_to_cpu(LBA_OFFSET(ddf, bvd)[ibvd]);
2554                 rv[n].size = be64_to_cpu(bvd->blocks);
2555                 n++;
2556         }
2557         qsort(rv, n, sizeof(*rv), cmp_extent);
2558
2559         rv[n].start = be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
2560         rv[n].size = 0;
2561         return rv;
2562 }
2563
2564 static unsigned long long find_space(
2565         struct ddf_super *ddf, struct dl *dl,
2566         unsigned long long data_offset,
2567         unsigned long long *size)
2568 {
2569         /* Find if the requested amount of space is available.
2570          * If it is, return start.
2571          * If not, set *size to largest space.
2572          * If data_offset != INVALID_SECTORS, then the space must start
2573          * at this location.
2574          */
2575         struct extent *e = get_extents(ddf, dl);
2576         int i = 0;
2577         unsigned long long pos = 0;
2578         unsigned long long max_size = 0;
2579
2580         if (!e) {
2581                 *size = 0;
2582                 return INVALID_SECTORS;
2583         }
2584         do {
2585                 unsigned long long esize = e[i].start - pos;
2586                 if (data_offset != INVALID_SECTORS &&
2587                     pos <= data_offset &&
2588                     e[i].start > data_offset) {
2589                         pos = data_offset;
2590                         esize = e[i].start - pos;
2591                 }
2592                 if (data_offset != INVALID_SECTORS &&
2593                     pos != data_offset) {
2594                         i++;
2595                         continue;
2596                 }
2597                 if (esize >= *size) {
2598                         /* Found! */
2599                         free(e);
2600                         return pos;
2601                 }
2602                 if (esize > max_size)
2603                         max_size = esize;
2604                 pos = e[i].start + e[i].size;
2605                 i++;
2606         } while (e[i-1].size);
2607         *size = max_size;
2608         free(e);
2609         return INVALID_SECTORS;
2610 }
2611 #endif
2612
2613 static int init_super_ddf_bvd(struct supertype *st,
2614                               mdu_array_info_t *info,
2615                               unsigned long long size,
2616                               char *name, char *homehost,
2617                               int *uuid, unsigned long long data_offset)
2618 {
2619         /* We are creating a BVD inside a pre-existing container.
2620          * so st->sb is already set.
2621          * We need to create a new vd_config and a new virtual_entry
2622          */
2623         struct ddf_super *ddf = st->sb;
2624         unsigned int venum, i;
2625         struct virtual_entry *ve;
2626         struct vcl *vcl;
2627         struct vd_config *vc;
2628
2629         if (find_vde_by_name(ddf, name) != DDF_NOTFOUND) {
2630                 pr_err("This ddf already has an array called %s\n", name);
2631                 return 0;
2632         }
2633         venum = find_unused_vde(ddf);
2634         if (venum == DDF_NOTFOUND) {
2635                 pr_err("Cannot find spare slot for virtual disk\n");
2636                 return 0;
2637         }
2638         ve = &ddf->virt->entries[venum];
2639
2640         /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2641          * timestamp, random number
2642          */
2643         make_header_guid(ve->guid);
2644         ve->unit = cpu_to_be16(info->md_minor);
2645         ve->pad0 = 0xFFFF;
2646         ve->guid_crc._v16 = crc32(0, (unsigned char *)ddf->anchor.guid,
2647                                   DDF_GUID_LEN);
2648         ve->type = cpu_to_be16(0);
2649         ve->state = DDF_state_degraded; /* Will be modified as devices are added */
2650         if (info->state & 1) /* clean */
2651                 ve->init_state = DDF_init_full;
2652         else
2653                 ve->init_state = DDF_init_not;
2654
2655         memset(ve->pad1, 0xff, 14);
2656         memset(ve->name, ' ', 16);
2657         if (name)
2658                 strncpy(ve->name, name, 16);
2659         ddf->virt->populated_vdes =
2660                 cpu_to_be16(be16_to_cpu(ddf->virt->populated_vdes)+1);
2661
2662         /* Now create a new vd_config */
2663         if (posix_memalign((void**)&vcl, 512,
2664                            (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
2665                 pr_err("could not allocate vd_config\n");
2666                 return 0;
2667         }
2668         vcl->vcnum = venum;
2669         vcl->block_sizes = NULL; /* FIXME not for CONCAT */
2670         vc = &vcl->conf;
2671
2672         vc->magic = DDF_VD_CONF_MAGIC;
2673         memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
2674         vc->timestamp = cpu_to_be32(time(0)-DECADE);
2675         vc->seqnum = cpu_to_be32(1);
2676         memset(vc->pad0, 0xff, 24);
2677         vc->chunk_shift = chunk_to_shift(info->chunk_size);
2678         if (layout_md2ddf(info, vc) == -1 ||
2679                 be16_to_cpu(vc->prim_elmnt_count) > ddf->mppe) {
2680                 pr_err("unsupported RAID level/layout %d/%d with %d disks\n",
2681                        info->level, info->layout, info->raid_disks);
2682                 free(vcl);
2683                 return 0;
2684         }
2685         vc->sec_elmnt_seq = 0;
2686         if (alloc_other_bvds(ddf, vcl) != 0) {
2687                 pr_err("could not allocate other bvds\n");
2688                 free(vcl);
2689                 return 0;
2690         }
2691         vc->blocks = cpu_to_be64(size * 2);
2692         vc->array_blocks = cpu_to_be64(
2693                 calc_array_size(info->level, info->raid_disks, info->layout,
2694                                 info->chunk_size, size * 2));
2695         memset(vc->pad1, 0xff, 8);
2696         vc->spare_refs[0] = cpu_to_be32(0xffffffff);
2697         vc->spare_refs[1] = cpu_to_be32(0xffffffff);
2698         vc->spare_refs[2] = cpu_to_be32(0xffffffff);
2699         vc->spare_refs[3] = cpu_to_be32(0xffffffff);
2700         vc->spare_refs[4] = cpu_to_be32(0xffffffff);
2701         vc->spare_refs[5] = cpu_to_be32(0xffffffff);
2702         vc->spare_refs[6] = cpu_to_be32(0xffffffff);
2703         vc->spare_refs[7] = cpu_to_be32(0xffffffff);
2704         memset(vc->cache_pol, 0, 8);
2705         vc->bg_rate = 0x80;
2706         memset(vc->pad2, 0xff, 3);
2707         memset(vc->pad3, 0xff, 52);
2708         memset(vc->pad4, 0xff, 192);
2709         memset(vc->v0, 0xff, 32);
2710         memset(vc->v1, 0xff, 32);
2711         memset(vc->v2, 0xff, 16);
2712         memset(vc->v3, 0xff, 16);
2713         memset(vc->vendor, 0xff, 32);
2714
2715         memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2716         memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2717
2718         for (i = 1; i < vc->sec_elmnt_count; i++) {
2719                 memcpy(vcl->other_bvds[i-1], vc, ddf->conf_rec_len * 512);
2720                 vcl->other_bvds[i-1]->sec_elmnt_seq = i;
2721         }
2722
2723         vcl->next = ddf->conflist;
2724         ddf->conflist = vcl;
2725         ddf->currentconf = vcl;
2726         ddf_set_updates_pending(ddf, NULL);
2727         return 1;
2728 }
2729
2730 #ifndef MDASSEMBLE
2731 static void add_to_super_ddf_bvd(struct supertype *st,
2732                                  mdu_disk_info_t *dk, int fd, char *devname,
2733                                  unsigned long long data_offset)
2734 {
2735         /* fd and devname identify a device within the ddf container (st).
2736          * dk identifies a location in the new BVD.
2737          * We need to find suitable free space in that device and update
2738          * the phys_refnum and lba_offset for the newly created vd_config.
2739          * We might also want to update the type in the phys_disk
2740          * section.
2741          *
2742          * Alternately: fd == -1 and we have already chosen which device to
2743          * use and recorded in dlist->raid_disk;
2744          */
2745         struct dl *dl;
2746         struct ddf_super *ddf = st->sb;
2747         struct vd_config *vc;
2748         unsigned int i;
2749         unsigned long long blocks, pos;
2750         unsigned int raid_disk = dk->raid_disk;
2751
2752         if (fd == -1) {
2753                 for (dl = ddf->dlist; dl ; dl = dl->next)
2754                         if (dl->raiddisk == dk->raid_disk)
2755                                 break;
2756         } else {
2757                 for (dl = ddf->dlist; dl ; dl = dl->next)
2758                         if (dl->major == dk->major &&
2759                             dl->minor == dk->minor)
2760                                 break;
2761         }
2762         if (!dl || dl->pdnum < 0 || ! (dk->state & (1<<MD_DISK_SYNC)))
2763                 return;
2764
2765         vc = &ddf->currentconf->conf;
2766         if (vc->sec_elmnt_count > 1) {
2767                 unsigned int n = be16_to_cpu(vc->prim_elmnt_count);
2768                 if (raid_disk >= n)
2769                         vc = ddf->currentconf->other_bvds[raid_disk / n - 1];
2770                 raid_disk %= n;
2771         }
2772
2773         blocks = be64_to_cpu(vc->blocks);
2774         if (ddf->currentconf->block_sizes)
2775                 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2776
2777         pos = find_space(ddf, dl, data_offset, &blocks);
2778         if (pos == INVALID_SECTORS)
2779                 return;
2780
2781         ddf->currentdev = dk->raid_disk;
2782         vc->phys_refnum[raid_disk] = dl->disk.refnum;
2783         LBA_OFFSET(ddf, vc)[raid_disk] = cpu_to_be64(pos);
2784
2785         for (i = 0; i < ddf->max_part ; i++)
2786                 if (dl->vlist[i] == NULL)
2787                         break;
2788         if (i == ddf->max_part)
2789                 return;
2790         dl->vlist[i] = ddf->currentconf;
2791
2792         if (fd >= 0)
2793                 dl->fd = fd;
2794         if (devname)
2795                 dl->devname = devname;
2796
2797         /* Check if we can mark array as optimal yet */
2798         i = ddf->currentconf->vcnum;
2799         ddf->virt->entries[i].state =
2800                 (ddf->virt->entries[i].state & ~DDF_state_mask)
2801                 | get_svd_state(ddf, ddf->currentconf);
2802         be16_clear(ddf->phys->entries[dl->pdnum].type,
2803                    cpu_to_be16(DDF_Global_Spare));
2804         be16_set(ddf->phys->entries[dl->pdnum].type,
2805                  cpu_to_be16(DDF_Active_in_VD));
2806         dprintf("added disk %d/%08x to VD %d/%s as disk %d\n",
2807                 dl->pdnum, be32_to_cpu(dl->disk.refnum),
2808                 ddf->currentconf->vcnum, guid_str(vc->guid),
2809                 dk->raid_disk);
2810         ddf_set_updates_pending(ddf, vc);
2811 }
2812
2813 static unsigned int find_unused_pde(const struct ddf_super *ddf)
2814 {
2815         unsigned int i;
2816         for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) {
2817                 if (all_ff(ddf->phys->entries[i].guid))
2818                         return i;
2819         }
2820         return DDF_NOTFOUND;
2821 }
2822
2823 static void _set_config_size(struct phys_disk_entry *pde, const struct dl *dl)
2824 {
2825         __u64 cfs, t;
2826         cfs = min(dl->size - 32*1024*2ULL, be64_to_cpu(dl->primary_lba));
2827         t = be64_to_cpu(dl->secondary_lba);
2828         if (t != ~(__u64)0)
2829                 cfs = min(cfs, t);
2830         /*
2831          * Some vendor DDF structures interpret workspace_lba
2832          * very differently than we do: Make a sanity check on the value.
2833          */
2834         t = be64_to_cpu(dl->workspace_lba);
2835         if (t < cfs) {
2836                 __u64 wsp = cfs - t;
2837                 if (wsp > 1024*1024*2ULL && wsp > dl->size / 16) {
2838                         pr_err("%x:%x: workspace size 0x%llx too big, ignoring\n",
2839                                dl->major, dl->minor, (unsigned long long)wsp);
2840                 } else
2841                         cfs = t;
2842         }
2843         pde->config_size = cpu_to_be64(cfs);
2844         dprintf("%x:%x config_size %llx, DDF structure is %llx blocks\n",
2845                 dl->major, dl->minor,
2846                 (unsigned long long)cfs, (unsigned long long)(dl->size-cfs));
2847 }
2848
2849 /* Add a device to a container, either while creating it or while
2850  * expanding a pre-existing container
2851  */
2852 static int add_to_super_ddf(struct supertype *st,
2853                             mdu_disk_info_t *dk, int fd, char *devname,
2854                             unsigned long long data_offset)
2855 {
2856         struct ddf_super *ddf = st->sb;
2857         struct dl *dd;
2858         time_t now;
2859         struct tm *tm;
2860         unsigned long long size;
2861         struct phys_disk_entry *pde;
2862         unsigned int n, i;
2863         struct stat stb;
2864         __u32 *tptr;
2865
2866         if (ddf->currentconf) {
2867                 add_to_super_ddf_bvd(st, dk, fd, devname, data_offset);
2868                 return 0;
2869         }
2870
2871         /* This is device numbered dk->number.  We need to create
2872          * a phys_disk entry and a more detailed disk_data entry.
2873          */
2874         fstat(fd, &stb);
2875         n = find_unused_pde(ddf);
2876         if (n == DDF_NOTFOUND) {
2877                 pr_err("No free slot in array, cannot add disk\n");
2878                 return 1;
2879         }
2880         pde = &ddf->phys->entries[n];
2881         get_dev_size(fd, NULL, &size);
2882         if (size <= 32*1024*1024) {
2883                 pr_err("device size must be at least 32MB\n");
2884                 return 1;
2885         }
2886         size >>= 9;
2887
2888         if (posix_memalign((void**)&dd, 512,
2889                            sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2890                 pr_err("could allocate buffer for new disk, aborting\n");
2891                 return 1;
2892         }
2893         dd->major = major(stb.st_rdev);
2894         dd->minor = minor(stb.st_rdev);
2895         dd->devname = devname;
2896         dd->fd = fd;
2897         dd->spare = NULL;
2898
2899         dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2900         now = time(0);
2901         tm = localtime(&now);
2902         sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2903                 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2904         tptr = (__u32 *)(dd->disk.guid + 16);
2905         *tptr++ = random32();
2906         *tptr = random32();
2907
2908         do {
2909                 /* Cannot be bothered finding a CRC of some irrelevant details*/
2910                 dd->disk.refnum._v32 = random32();
2911                 for (i = be16_to_cpu(ddf->active->max_pd_entries);
2912                      i > 0; i--)
2913                         if (be32_eq(ddf->phys->entries[i-1].refnum,
2914                                     dd->disk.refnum))
2915                                 break;
2916         } while (i > 0);
2917
2918         dd->disk.forced_ref = 1;
2919         dd->disk.forced_guid = 1;
2920         memset(dd->disk.vendor, ' ', 32);
2921         memcpy(dd->disk.vendor, "Linux", 5);
2922         memset(dd->disk.pad, 0xff, 442);
2923         for (i = 0; i < ddf->max_part ; i++)
2924                 dd->vlist[i] = NULL;
2925
2926         dd->pdnum = n;
2927
2928         if (st->update_tail) {
2929                 int len = (sizeof(struct phys_disk) +
2930                            sizeof(struct phys_disk_entry));
2931                 struct phys_disk *pd;
2932
2933                 pd = xmalloc(len);
2934                 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2935                 pd->used_pdes = cpu_to_be16(n);
2936                 pde = &pd->entries[0];
2937                 dd->mdupdate = pd;
2938         } else
2939                 ddf->phys->used_pdes = cpu_to_be16(
2940                         1 + be16_to_cpu(ddf->phys->used_pdes));
2941
2942         memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2943         pde->refnum = dd->disk.refnum;
2944         pde->type = cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2945         pde->state = cpu_to_be16(DDF_Online);
2946         dd->size = size;
2947         /*
2948          * If there is already a device in dlist, try to reserve the same
2949          * amount of workspace. Otherwise, use 32MB.
2950          * We checked disk size above already.
2951          */
2952 #define __calc_lba(new, old, lba, mb) do { \
2953                 unsigned long long dif; \
2954                 if ((old) != NULL) \
2955                         dif = (old)->size - be64_to_cpu((old)->lba); \
2956                 else \
2957                         dif = (new)->size; \
2958                 if ((new)->size > dif) \
2959                         (new)->lba = cpu_to_be64((new)->size - dif); \
2960                 else \
2961                         (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2962         } while (0)
2963         __calc_lba(dd, ddf->dlist, workspace_lba, 32);
2964         __calc_lba(dd, ddf->dlist, primary_lba, 16);
2965         if (ddf->dlist == NULL ||
2966             be64_to_cpu(ddf->dlist->secondary_lba) != ~(__u64)0)
2967                 __calc_lba(dd, ddf->dlist, secondary_lba, 32);
2968         _set_config_size(pde, dd);
2969
2970         sprintf(pde->path, "%17.17s","Information: nil") ;
2971         memset(pde->pad, 0xff, 6);
2972
2973         if (st->update_tail) {
2974                 dd->next = ddf->add_list;
2975                 ddf->add_list = dd;
2976         } else {
2977                 dd->next = ddf->dlist;
2978                 ddf->dlist = dd;
2979                 ddf_set_updates_pending(ddf, NULL);
2980         }
2981
2982         return 0;
2983 }
2984
2985 static int remove_from_super_ddf(struct supertype *st, mdu_disk_info_t *dk)
2986 {
2987         struct ddf_super *ddf = st->sb;
2988         struct dl *dl;
2989
2990         /* mdmon has noticed that this disk (dk->major/dk->minor) has
2991          * disappeared from the container.
2992          * We need to arrange that it disappears from the metadata and
2993          * internal data structures too.
2994          * Most of the work is done by ddf_process_update which edits
2995          * the metadata and closes the file handle and attaches the memory
2996          * where free_updates will free it.
2997          */
2998         for (dl = ddf->dlist; dl ; dl = dl->next)
2999                 if (dl->major == dk->major &&
3000                     dl->minor == dk->minor)
3001                         break;
3002         if (!dl || dl->pdnum < 0)
3003                 return -1;
3004
3005         if (st->update_tail) {
3006                 int len = (sizeof(struct phys_disk) +
3007                            sizeof(struct phys_disk_entry));
3008                 struct phys_disk *pd;
3009
3010                 pd = xmalloc(len);
3011                 pd->magic = DDF_PHYS_RECORDS_MAGIC;
3012                 pd->used_pdes = cpu_to_be16(dl->pdnum);
3013                 pd->entries[0].state = cpu_to_be16(DDF_Missing);
3014                 append_metadata_update(st, pd, len);
3015         }
3016         return 0;
3017 }
3018 #endif
3019
3020 /*
3021  * This is the write_init_super method for a ddf container.  It is
3022  * called when creating a container or adding another device to a
3023  * container.
3024  */
3025
3026 static int __write_ddf_structure(struct dl *d, struct ddf_super *ddf, __u8 type)
3027 {
3028         unsigned long long sector;
3029         struct ddf_header *header;
3030         int fd, i, n_config, conf_size, buf_size;
3031         int ret = 0;
3032         char *conf;
3033
3034         fd = d->fd;
3035
3036         switch (type) {
3037         case DDF_HEADER_PRIMARY:
3038                 header = &ddf->primary;
3039                 sector = be64_to_cpu(header->primary_lba);
3040                 break;
3041         case DDF_HEADER_SECONDARY:
3042                 header = &ddf->secondary;
3043                 sector = be64_to_cpu(header->secondary_lba);
3044                 break;
3045         default:
3046                 return 0;
3047         }
3048         if (sector == ~(__u64)0)
3049                 return 0;
3050
3051         header->type = type;
3052         header->openflag = 1;
3053         header->crc = calc_crc(header, 512);
3054
3055         lseek64(fd, sector<<9, 0);
3056         if (write(fd, header, 512) < 0)
3057                 goto out;
3058
3059         ddf->controller.crc = calc_crc(&ddf->controller, 512);
3060         if (write(fd, &ddf->controller, 512) < 0)
3061                 goto out;
3062
3063         ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
3064         if (write(fd, ddf->phys, ddf->pdsize) < 0)
3065                 goto out;
3066         ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
3067         if (write(fd, ddf->virt, ddf->vdsize) < 0)
3068                 goto out;
3069
3070         /* Now write lots of config records. */
3071         n_config = ddf->max_part;
3072         conf_size = ddf->conf_rec_len * 512;
3073         conf = ddf->conf;
3074         buf_size = conf_size * (n_config + 1);
3075         if (!conf) {
3076                 if (posix_memalign((void**)&conf, 512, buf_size) != 0)
3077                         goto out;
3078                 ddf->conf = conf;
3079         }
3080         for (i = 0 ; i <= n_config ; i++) {
3081                 struct vcl *c;
3082                 struct vd_config *vdc = NULL;
3083                 if (i == n_config) {
3084                         c = (struct vcl *)d->spare;
3085                         if (c)
3086                                 vdc = &c->conf;
3087                 } else {
3088                         unsigned int dummy;
3089                         c = d->vlist[i];
3090                         if (c)
3091                                 get_pd_index_from_refnum(
3092                                         c, d->disk.refnum,
3093                                         ddf->mppe,
3094                                         (const struct vd_config **)&vdc,
3095                                         &dummy);
3096                 }
3097                 if (vdc) {
3098                         dprintf("writing conf record %i on disk %08x for %s/%u\n",
3099                                 i, be32_to_cpu(d->disk.refnum),
3100                                 guid_str(vdc->guid),
3101                                 vdc->sec_elmnt_seq);
3102                         vdc->crc = calc_crc(vdc, conf_size);
3103                         memcpy(conf + i*conf_size, vdc, conf_size);
3104                 } else
3105                         memset(conf + i*conf_size, 0xff, conf_size);
3106         }
3107         if (write(fd, conf, buf_size) != buf_size)
3108                 goto out;
3109
3110         d->disk.crc = calc_crc(&d->disk, 512);
3111         if (write(fd, &d->disk, 512) < 0)
3112                 goto out;
3113
3114         ret = 1;
3115 out:
3116         header->openflag = 0;
3117         header->crc = calc_crc(header, 512);
3118
3119         lseek64(fd, sector<<9, 0);
3120         if (write(fd, header, 512) < 0)
3121                 ret = 0;
3122
3123         return ret;
3124 }
3125
3126 static int _write_super_to_disk(struct ddf_super *ddf, struct dl *d)
3127 {
3128         unsigned long long size;
3129         int fd = d->fd;
3130         if (fd < 0)
3131                 return 0;
3132
3133         /* We need to fill in the primary, (secondary) and workspace
3134          * lba's in the headers, set their checksums,
3135          * Also checksum phys, virt....
3136          *
3137          * Then write everything out, finally the anchor is written.
3138          */
3139         get_dev_size(fd, NULL, &size);
3140         size /= 512;
3141         memcpy(&ddf->anchor, ddf->active, 512);
3142         if (be64_to_cpu(d->workspace_lba) != 0ULL)
3143                 ddf->anchor.workspace_lba = d->workspace_lba;
3144         else
3145                 ddf->anchor.workspace_lba =
3146                         cpu_to_be64(size - 32*1024*2);
3147         if (be64_to_cpu(d->primary_lba) != 0ULL)
3148                 ddf->anchor.primary_lba = d->primary_lba;
3149         else
3150                 ddf->anchor.primary_lba =
3151                         cpu_to_be64(size - 16*1024*2);
3152         if (be64_to_cpu(d->secondary_lba) != 0ULL)
3153                 ddf->anchor.secondary_lba = d->secondary_lba;
3154         else
3155                 ddf->anchor.secondary_lba =
3156                         cpu_to_be64(size - 32*1024*2);
3157         ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
3158         memcpy(&ddf->primary, &ddf->anchor, 512);
3159         memcpy(&ddf->secondary, &ddf->anchor, 512);
3160
3161         ddf->anchor.type = DDF_HEADER_ANCHOR;
3162         ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
3163         ddf->anchor.seq = cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */
3164         ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
3165
3166         if (!__write_ddf_structure(d, ddf, DDF_HEADER_PRIMARY))
3167                 return 0;
3168
3169         if (!__write_ddf_structure(d, ddf, DDF_HEADER_SECONDARY))
3170                 return 0;
3171
3172         lseek64(fd, (size-1)*512, SEEK_SET);
3173         if (write(fd, &ddf->anchor, 512) < 0)
3174                 return 0;
3175
3176         return 1;
3177 }
3178
3179 #ifndef MDASSEMBLE
3180 static int __write_init_super_ddf(struct supertype *st)
3181 {
3182         struct ddf_super *ddf = st->sb;
3183         struct dl *d;
3184         int attempts = 0;
3185         int successes = 0;
3186
3187         pr_state(ddf, __func__);
3188
3189         /* try to write updated metadata,
3190          * if we catch a failure move on to the next disk
3191          */
3192         for (d = ddf->dlist; d; d=d->next) {
3193                 attempts++;
3194                 successes += _write_super_to_disk(ddf, d);
3195         }
3196
3197         return attempts != successes;
3198 }
3199
3200 static int write_init_super_ddf(struct supertype *st)
3201 {
3202         struct ddf_super *ddf = st->sb;
3203         struct vcl *currentconf = ddf->currentconf;
3204
3205         /* We are done with currentconf - reset it so st refers to the container */
3206         ddf->currentconf = NULL;
3207
3208         if (st->update_tail) {
3209                 /* queue the virtual_disk and vd_config as metadata updates */
3210                 struct virtual_disk *vd;
3211                 struct vd_config *vc;
3212                 int len, tlen;
3213                 unsigned int i;
3214
3215                 if (!currentconf) {
3216                         /* Must be adding a physical disk to the container */
3217                         int len = (sizeof(struct phys_disk) +
3218                                    sizeof(struct phys_disk_entry));
3219
3220                         /* adding a disk to the container. */
3221                         if (!ddf->add_list)
3222                                 return 0;
3223
3224                         append_metadata_update(st, ddf->add_list->mdupdate, len);
3225                         ddf->add_list->mdupdate = NULL;
3226                         return 0;
3227                 }
3228
3229                 /* Newly created VD */
3230
3231                 /* First the virtual disk.  We have a slightly fake header */
3232                 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
3233                 vd = xmalloc(len);
3234                 *vd = *ddf->virt;
3235                 vd->entries[0] = ddf->virt->entries[currentconf->vcnum];
3236                 vd->populated_vdes = cpu_to_be16(currentconf->vcnum);
3237                 append_metadata_update(st, vd, len);
3238
3239                 /* Then the vd_config */
3240                 len = ddf->conf_rec_len * 512;
3241                 tlen = len * currentconf->conf.sec_elmnt_count;
3242                 vc = xmalloc(tlen);
3243                 memcpy(vc, &currentconf->conf, len);
3244                 for (i = 1; i < currentconf->conf.sec_elmnt_count; i++)
3245                         memcpy((char *)vc + i*len, currentconf->other_bvds[i-1],
3246                                len);
3247                 append_metadata_update(st, vc, tlen);
3248
3249                 return 0;
3250         } else {
3251                 struct dl *d;
3252                 if (!currentconf)
3253                         for (d = ddf->dlist; d; d=d->next)
3254                                 while (Kill(d->devname, NULL, 0, -1, 1) == 0);
3255                 /* Note: we don't close the fd's now, but a subsequent
3256                  * ->free_super() will
3257                  */
3258                 return __write_init_super_ddf(st);
3259         }
3260 }
3261
3262 #endif
3263
3264 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize,
3265                             unsigned long long data_offset)
3266 {
3267         /* We must reserve the last 32Meg */
3268         if (devsize <= 32*1024*2)
3269                 return 0;
3270         return devsize - 32*1024*2;
3271 }
3272
3273 #ifndef MDASSEMBLE
3274
3275 static int reserve_space(struct supertype *st, int raiddisks,
3276                          unsigned long long size, int chunk,
3277                          unsigned long long data_offset,
3278                          unsigned long long *freesize)
3279 {
3280         /* Find 'raiddisks' spare extents at least 'size' big (but
3281          * only caring about multiples of 'chunk') and remember
3282          * them.   If size==0, find the largest size possible.
3283          * Report available size in *freesize
3284          * If space cannot be found, fail.
3285          */
3286         struct dl *dl;
3287         struct ddf_super *ddf = st->sb;
3288         int cnt = 0;
3289
3290         for (dl = ddf->dlist; dl ; dl=dl->next) {
3291                 dl->raiddisk = -1;
3292                 dl->esize = 0;
3293         }
3294         /* Now find largest extent on each device */
3295         for (dl = ddf->dlist ; dl ; dl=dl->next) {
3296                 unsigned long long minsize = ULLONG_MAX;
3297
3298                 find_space(ddf, dl, data_offset, &minsize);
3299                 if (minsize >= size && minsize >= (unsigned)chunk) {
3300                         cnt++;
3301                         dl->esize = minsize;
3302                 }
3303         }
3304         if (cnt < raiddisks) {
3305                 pr_err("not enough devices with space to create array.\n");
3306                 return 0; /* No enough free spaces large enough */
3307         }
3308         if (size == 0) {
3309                 /* choose the largest size of which there are at least 'raiddisk' */
3310                 for (dl = ddf->dlist ; dl ; dl=dl->next) {
3311                         struct dl *dl2;
3312                         if (dl->esize <= size)
3313                                 continue;
3314                         /* This is bigger than 'size', see if there are enough */
3315                         cnt = 0;
3316                         for (dl2 = ddf->dlist; dl2 ; dl2=dl2->next)
3317                                 if (dl2->esize >= dl->esize)
3318                                         cnt++;
3319                         if (cnt >= raiddisks)
3320                                 size = dl->esize;
3321                 }
3322                 if (chunk) {
3323                         size = size / chunk;
3324                         size *= chunk;
3325                 }
3326                 *freesize = size;
3327                 if (size < 32) {
3328                         pr_err("not enough spare devices to create array.\n");
3329                         return 0;
3330                 }
3331         }
3332         /* We have a 'size' of which there are enough spaces.
3333          * We simply do a first-fit */
3334         cnt = 0;
3335         for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
3336                 if (dl->esize < size)
3337                         continue;
3338
3339                 dl->raiddisk = cnt;
3340                 cnt++;
3341         }
3342         return 1;
3343 }
3344
3345 static int validate_geometry_ddf(struct supertype *st,
3346                                  int level, int layout, int raiddisks,
3347                                  int *chunk, unsigned long long size,
3348                                  unsigned long long data_offset,
3349                                  char *dev, unsigned long long *freesize,
3350                                  int verbose)
3351 {
3352         int fd;
3353         struct mdinfo *sra;
3354         int cfd;
3355
3356         /* ddf potentially supports lots of things, but it depends on
3357          * what devices are offered (and maybe kernel version?)
3358          * If given unused devices, we will make a container.
3359          * If given devices in a container, we will make a BVD.
3360          * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3361          */
3362
3363         if (*chunk == UnSet)
3364                 *chunk = DEFAULT_CHUNK;
3365
3366         if (level == LEVEL_NONE)
3367                 level = LEVEL_CONTAINER;
3368         if (level == LEVEL_CONTAINER) {
3369                 /* Must be a fresh device to add to a container */
3370                 return validate_geometry_ddf_container(st, level, layout,
3371                                                        raiddisks, *chunk,
3372                                                        size, data_offset, dev,
3373                                                        freesize,
3374                                                        verbose);
3375         }
3376
3377         if (!dev) {
3378                 mdu_array_info_t array = {
3379                         .level = level,
3380                         .layout = layout,
3381                         .raid_disks = raiddisks
3382                 };
3383                 struct vd_config conf;
3384                 if (layout_md2ddf(&array, &conf) == -1) {
3385                         if (verbose)
3386                                 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3387                                        level, layout, raiddisks);
3388                         return 0;
3389                 }
3390                 /* Should check layout? etc */
3391
3392                 if (st->sb && freesize) {
3393                         /* --create was given a container to create in.
3394                          * So we need to check that there are enough
3395                          * free spaces and return the amount of space.
3396                          * We may as well remember which drives were
3397                          * chosen so that add_to_super/getinfo_super
3398                          * can return them.
3399                          */
3400                         return reserve_space(st, raiddisks, size, *chunk,
3401                                              data_offset, freesize);
3402                 }
3403                 return 1;
3404         }
3405
3406         if (st->sb) {
3407                 /* A container has already been opened, so we are
3408                  * creating in there.  Maybe a BVD, maybe an SVD.
3409                  * Should make a distinction one day.
3410                  */
3411                 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
3412                                                  chunk, size, data_offset, dev,
3413                                                  freesize,
3414                                                  verbose);
3415         }
3416         /* This is the first device for the array.
3417          * If it is a container, we read it in and do automagic allocations,
3418          * no other devices should be given.
3419          * Otherwise it must be a member device of a container, and we
3420          * do manual allocation.
3421          * Later we should check for a BVD and make an SVD.
3422          */
3423         fd = open(dev, O_RDONLY|O_EXCL, 0);
3424         if (fd >= 0) {
3425                 close(fd);
3426                 /* Just a bare device, no good to us */
3427                 if (verbose)
3428                         pr_err("ddf: Cannot create this array on device %s - a container is required.\n",
3429                                dev);
3430                 return 0;
3431         }
3432         if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3433                 if (verbose)
3434                         pr_err("ddf: Cannot open %s: %s\n",
3435                                dev, strerror(errno));
3436                 return 0;
3437         }
3438         /* Well, it is in use by someone, maybe a 'ddf' container. */
3439         cfd = open_container(fd);
3440         if (cfd < 0) {
3441                 close(fd);
3442                 if (verbose)
3443                         pr_err("ddf: Cannot use %s: %s\n",
3444                                dev, strerror(EBUSY));
3445                 return 0;
3446         }
3447         sra = sysfs_read(cfd, NULL, GET_VERSION);
3448         close(fd);
3449         if (sra && sra->array.major_version == -1 &&
3450             strcmp(sra->text_version, "ddf") == 0) {
3451                 /* This is a member of a ddf container.  Load the container
3452                  * and try to create a bvd
3453                  */
3454                 struct ddf_super *ddf;
3455                 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL) == 0) {
3456                         st->sb = ddf;
3457                         strcpy(st->container_devnm, fd2devnm(cfd));
3458                         close(cfd);
3459                         return validate_geometry_ddf_bvd(st, level, layout,
3460                                                          raiddisks, chunk, size,
3461                                                          data_offset,
3462                                                          dev, freesize,
3463                                                          verbose);
3464                 }
3465                 close(cfd);
3466         } else /* device may belong to a different container */
3467                 return 0;
3468
3469         return 1;
3470 }
3471
3472 static int
3473 validate_geometry_ddf_container(struct supertype *st,
3474                                 int level, int layout, int raiddisks,
3475                                 int chunk, unsigned long long size,
3476                                 unsigned long long data_offset,
3477                                 char *dev, unsigned long long *freesize,
3478                                 int verbose)
3479 {
3480         int fd;
3481         unsigned long long ldsize;
3482
3483         if (level != LEVEL_CONTAINER)
3484                 return 0;
3485         if (!dev)
3486                 return 1;
3487
3488         fd = open(dev, O_RDONLY|O_EXCL, 0);
3489         if (fd < 0) {
3490                 if (verbose)
3491                         pr_err("ddf: Cannot open %s: %s\n",
3492                                dev, strerror(errno));
3493                 return 0;
3494         }
3495         if (!get_dev_size(fd, dev, &ldsize)) {
3496                 close(fd);
3497                 return 0;
3498         }
3499         close(fd);
3500
3501         *freesize = avail_size_ddf(st, ldsize >> 9, INVALID_SECTORS);
3502         if (*freesize == 0)
3503                 return 0;
3504
3505         return 1;
3506 }
3507
3508 static int validate_geometry_ddf_bvd(struct supertype *st,
3509                                      int level, int layout, int raiddisks,
3510                                      int *chunk, unsigned long long size,
3511                                      unsigned long long data_offset,
3512                                      char *dev, unsigned long long *freesize,
3513                                      int verbose)
3514 {
3515         struct stat stb;
3516         struct ddf_super *ddf = st->sb;
3517         struct dl *dl;
3518         unsigned long long maxsize;
3519         /* ddf/bvd supports lots of things, but not containers */
3520         if (level == LEVEL_CONTAINER) {
3521                 if (verbose)
3522                         pr_err("DDF cannot create a container within an container\n");
3523                 return 0;
3524         }
3525         /* We must have the container info already read in. */
3526         if (!ddf)
3527                 return 0;
3528
3529         if (!dev) {
3530                 /* General test:  make sure there is space for
3531                  * 'raiddisks' device extents of size 'size'.
3532                  */
3533                 unsigned long long minsize = size;
3534                 int dcnt = 0;
3535                 if (minsize == 0)
3536                         minsize = 8;
3537                 for (dl = ddf->dlist; dl ; dl = dl->next) {
3538                         if (find_space(ddf, dl, data_offset, &minsize)
3539                             != INVALID_SECTORS)
3540                                 dcnt++;
3541                 }
3542                 if (dcnt < raiddisks) {
3543                         if (verbose)
3544                                 pr_err("ddf: Not enough devices with space for this array (%d < %d)\n",
3545                                        dcnt, raiddisks);
3546                         return 0;
3547                 }
3548                 return 1;
3549         }
3550         /* This device must be a member of the set */
3551         if (stat(dev, &stb) < 0)
3552                 return 0;
3553         if ((S_IFMT & stb.st_mode) != S_IFBLK)
3554                 return 0;
3555         for (dl = ddf->dlist ; dl ; dl = dl->next) {
3556                 if (dl->major == (int)major(stb.st_rdev) &&
3557                     dl->minor == (int)minor(stb.st_rdev))
3558                         break;
3559         }
3560         if (!dl) {
3561                 if (verbose)
3562                         pr_err("ddf: %s is not in the same DDF set\n",
3563                                dev);
3564                 return 0;
3565         }
3566         maxsize = ULLONG_MAX;
3567         find_space(ddf, dl, data_offset, &maxsize);
3568         *freesize = maxsize;
3569
3570         return 1;
3571 }
3572
3573 static int load_super_ddf_all(struct supertype *st, int fd,
3574                               void **sbp, char *devname)
3575 {
3576         struct mdinfo *sra;
3577         struct ddf_super *super;
3578         struct mdinfo *sd, *best = NULL;
3579         int bestseq = 0;
3580         int seq;
3581         char nm[20];
3582         int dfd;
3583
3584         sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
3585         if (!sra)
3586                 return 1;
3587         if (sra->array.major_version != -1 ||
3588             sra->array.minor_version != -2 ||
3589             strcmp(sra->text_version, "ddf") != 0)
3590                 return 1;
3591
3592         if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
3593                 return 1;
3594         memset(super, 0, sizeof(*super));
3595
3596         /* first, try each device, and choose the best ddf */
3597         for (sd = sra->devs ; sd ; sd = sd->next) {
3598                 int rv;
3599                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3600                 dfd = dev_open(nm, O_RDONLY);
3601                 if (dfd < 0)
3602                         return 2;
3603                 rv = load_ddf_headers(dfd, super, NULL);
3604                 close(dfd);
3605                 if (rv == 0) {
3606                         seq = be32_to_cpu(super->active->seq);
3607                         if (super->active->openflag)
3608                                 seq--;
3609                         if (!best || seq > bestseq) {
3610                                 bestseq = seq;
3611                                 best = sd;
3612                         }
3613                 }
3614         }
3615         if (!best)
3616                 return 1;
3617         /* OK, load this ddf */
3618         sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
3619         dfd = dev_open(nm, O_RDONLY);
3620         if (dfd < 0)
3621                 return 1;
3622         load_ddf_headers(dfd, super, NULL);
3623         load_ddf_global(dfd, super, NULL);
3624         close(dfd);
3625         /* Now we need the device-local bits */
3626         for (sd = sra->devs ; sd ; sd = sd->next) {
3627                 int rv;
3628
3629                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3630                 dfd = dev_open(nm, O_RDWR);
3631                 if (dfd < 0)
3632                         return 2;
3633                 rv = load_ddf_headers(dfd, super, NULL);
3634                 if (rv == 0)
3635                         rv = load_ddf_local(dfd, super, NULL, 1);
3636                 if (rv)
3637                         return 1;
3638         }
3639
3640         *sbp = super;
3641         if (st->ss == NULL) {
3642                 st->ss = &super_ddf;
3643                 st->minor_version = 0;
3644                 st->max_devs = 512;
3645         }
3646         strcpy(st->container_devnm, fd2devnm(fd));
3647         return 0;
3648 }
3649
3650 static int load_container_ddf(struct supertype *st, int fd,
3651                               char *devname)
3652 {
3653         return load_super_ddf_all(st, fd, &st->sb, devname);
3654 }
3655
3656 #endif /* MDASSEMBLE */
3657
3658 static int check_secondary(const struct vcl *vc)
3659 {
3660         const struct vd_config *conf = &vc->conf;
3661         int i;
3662
3663         /* The only DDF secondary RAID level md can support is
3664          * RAID 10, if the stripe sizes and Basic volume sizes
3665          * are all equal.
3666          * Other configurations could in theory be supported by exposing
3667          * the BVDs to user space and using device mapper for the secondary
3668          * mapping. So far we don't support that.
3669          */
3670
3671         __u64 sec_elements[4] = {0, 0, 0, 0};
3672 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3673 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3674
3675         if (vc->other_bvds == NULL) {
3676                 pr_err("No BVDs for secondary RAID found\n");
3677                 return -1;
3678         }
3679         if (conf->prl != DDF_RAID1) {
3680                 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3681                 return -1;
3682         }
3683         if (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED) {
3684                 pr_err("Secondary RAID level %d is unsupported\n",
3685                        conf->srl);
3686                 return -1;
3687         }
3688         __set_sec_seen(conf->sec_elmnt_seq);
3689         for (i = 0; i < conf->sec_elmnt_count-1; i++) {
3690                 const struct vd_config *bvd = vc->other_bvds[i];
3691                 if (bvd->sec_elmnt_seq == DDF_UNUSED_BVD)
3692                         continue;
3693                 if (bvd->srl != conf->srl) {
3694                         pr_err("Inconsistent secondary RAID level across BVDs\n");
3695                         return -1;
3696                 }
3697                 if (bvd->prl != conf->prl) {
3698                         pr_err("Different RAID levels for BVDs are unsupported\n");
3699                         return -1;
3700                 }
3701                 if (!be16_eq(bvd->prim_elmnt_count, conf->prim_elmnt_count)) {
3702                         pr_err("All BVDs must have the same number of primary elements\n");
3703                         return -1;
3704                 }
3705                 if (bvd->chunk_shift != conf->chunk_shift) {
3706                         pr_err("Different strip sizes for BVDs are unsupported\n");
3707                         return -1;
3708                 }
3709                 if (!be64_eq(bvd->array_blocks, conf->array_blocks)) {
3710                         pr_err("Different BVD sizes are unsupported\n");
3711                         return -1;
3712                 }
3713                 __set_sec_seen(bvd->sec_elmnt_seq);
3714         }
3715         for (i = 0; i < conf->sec_elmnt_count; i++) {
3716                 if (!__was_sec_seen(i)) {
3717                         /* pr_err("BVD %d is missing\n", i); */
3718                         return -1;
3719                 }
3720         }
3721         return 0;
3722 }
3723
3724 static unsigned int get_pd_index_from_refnum(const struct vcl *vc,
3725                                              be32 refnum, unsigned int nmax,
3726                                              const struct vd_config **bvd,
3727                                              unsigned int *idx)
3728 {
3729         unsigned int i, j, n, sec, cnt;
3730
3731         cnt = be16_to_cpu(vc->conf.prim_elmnt_count);
3732         sec = (vc->conf.sec_elmnt_count == 1 ? 0 : vc->conf.sec_elmnt_seq);
3733
3734         for (i = 0, j = 0 ; i < nmax ; i++) {
3735                 /* j counts valid entries for this BVD */
3736                 if (be32_eq(vc->conf.phys_refnum[i], refnum)) {
3737                         *bvd = &vc->conf;
3738                         *idx = i;
3739                         return sec * cnt + j;
3740                 }
3741                 if (be32_to_cpu(vc->conf.phys_refnum[i]) != 0xffffffff)
3742                         j++;
3743         }
3744         if (vc->other_bvds == NULL)
3745                 goto bad;
3746
3747         for (n = 1; n < vc->conf.sec_elmnt_count; n++) {
3748                 struct vd_config *vd = vc->other_bvds[n-1];
3749                 sec = vd->sec_elmnt_seq;
3750                 if (sec == DDF_UNUSED_BVD)
3751                         continue;
3752                 for (i = 0, j = 0 ; i < nmax ; i++) {
3753                         if (be32_eq(vd->phys_refnum[i], refnum)) {
3754                                 *bvd = vd;
3755                                 *idx = i;
3756                                 return sec * cnt + j;
3757                         }
3758                         if (be32_to_cpu(vd->phys_refnum[i]) != 0xffffffff)
3759                                 j++;
3760                 }
3761         }
3762 bad:
3763         *bvd = NULL;
3764         return DDF_NOTFOUND;
3765 }
3766
3767 static struct mdinfo *container_content_ddf(struct supertype *st, char *subarray)
3768 {
3769         /* Given a container loaded by load_super_ddf_all,
3770          * extract information about all the arrays into
3771          * an mdinfo tree.
3772          *
3773          * For each vcl in conflist: create an mdinfo, fill it in,
3774          *  then look for matching devices (phys_refnum) in dlist
3775          *  and create appropriate device mdinfo.
3776          */
3777         struct ddf_super *ddf = st->sb;
3778         struct mdinfo *rest = NULL;
3779         struct vcl *vc;
3780
3781         for (vc = ddf->conflist ; vc ; vc=vc->next) {
3782                 unsigned int i;
3783                 struct mdinfo *this;
3784                 char *ep;
3785                 __u32 *cptr;
3786                 unsigned int pd;
3787
3788                 if (subarray &&
3789                     (strtoul(subarray, &ep, 10) != vc->vcnum ||
3790                      *ep != '\0'))
3791                         continue;
3792
3793                 if (vc->conf.sec_elmnt_count > 1) {
3794                         if (check_secondary(vc) != 0)
3795                                 continue;
3796                 }
3797
3798                 this = xcalloc(1, sizeof(*this));
3799                 this->next = rest;
3800                 rest = this;
3801
3802                 if (layout_ddf2md(&vc->conf, &this->array))
3803                         continue;
3804                 this->array.md_minor      = -1;
3805                 this->array.major_version = -1;
3806                 this->array.minor_version = -2;
3807                 this->safe_mode_delay     = DDF_SAFE_MODE_DELAY;
3808                 cptr = (__u32 *)(vc->conf.guid + 16);
3809                 this->array.ctime         = DECADE + __be32_to_cpu(*cptr);
3810                 this->array.utime         = DECADE +
3811                         be32_to_cpu(vc->conf.timestamp);
3812                 this->array.chunk_size    = 512 << vc->conf.chunk_shift;
3813
3814                 i = vc->vcnum;
3815                 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
3816                     (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
3817                     DDF_init_full) {
3818                         this->array.state = 0;
3819                         this->resync_start = 0;
3820                 } else {
3821                         this->array.state = 1;
3822                         this->resync_start = MaxSector;
3823                 }
3824                 _ddf_array_name(this->name, ddf, i);
3825                 memset(this->uuid, 0, sizeof(this->uuid));
3826                 this->component_size      = be64_to_cpu(vc->conf.blocks);
3827                 this->array.size          = this->component_size / 2;
3828                 this->container_member    = i;
3829
3830                 ddf->currentconf = vc;
3831                 uuid_from_super_ddf(st, this->uuid);
3832                 if (!subarray)
3833                         ddf->currentconf = NULL;
3834
3835                 sprintf(this->text_version, "/%s/%d",
3836                         st->container_devnm, this->container_member);
3837
3838                 for (pd = 0; pd < be16_to_cpu(ddf->phys->max_pdes); pd++) {
3839                         struct mdinfo *dev;
3840                         struct dl *d;
3841                         const struct vd_config *bvd;
3842                         unsigned int iphys;
3843                         int stt;
3844
3845                         if (be32_to_cpu(ddf->phys->entries[pd].refnum)
3846                             == 0xFFFFFFFF)
3847                                 continue;
3848
3849                         stt = be16_to_cpu(ddf->phys->entries[pd].state);
3850                         if ((stt & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3851                             != DDF_Online)
3852                                 continue;
3853
3854                         i = get_pd_index_from_refnum(
3855                                 vc, ddf->phys->entries[pd].refnum,
3856                                 ddf->mppe, &bvd, &iphys);
3857                         if (i == DDF_NOTFOUND)
3858                                 continue;
3859
3860                         this->array.working_disks++;
3861
3862                         for (d = ddf->dlist; d ; d=d->next)
3863                                 if (be32_eq(d->disk.refnum,
3864                                             ddf->phys->entries[pd].refnum))
3865                                         break;
3866                         if (d == NULL)
3867                                 /* Haven't found that one yet, maybe there are others */
3868                                 continue;
3869
3870                         dev = xcalloc(1, sizeof(*dev));
3871                         dev->next        = this->devs;
3872                         this->devs       = dev;
3873
3874                         dev->disk.number = be32_to_cpu(d->disk.refnum);
3875                         dev->disk.major  = d->major;
3876                         dev->disk.minor  = d->minor;
3877                         dev->disk.raid_disk = i;
3878                         dev->disk.state  = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
3879                         dev->recovery_start = MaxSector;
3880
3881                         dev->events      = be32_to_cpu(ddf->active->seq);
3882                         dev->data_offset =
3883                                 be64_to_cpu(LBA_OFFSET(ddf, bvd)[iphys]);
3884                         dev->component_size = be64_to_cpu(bvd->blocks);
3885                         if (d->devname)
3886                                 strcpy(dev->name, d->devname);
3887                 }
3888         }
3889         return rest;
3890 }
3891
3892 static int store_super_ddf(struct supertype *st, int fd)
3893 {
3894         struct ddf_super *ddf = st->sb;
3895         unsigned long long dsize;
3896         void *buf;
3897         int rc;
3898
3899         if (!ddf)
3900                 return 1;
3901
3902         if (!get_dev_size(fd, NULL, &dsize))
3903                 return 1;
3904
3905         if (ddf->dlist || ddf->conflist) {
3906                 struct stat sta;
3907                 struct dl *dl;
3908                 int ofd, ret;
3909
3910                 if (fstat(fd, &sta) == -1 || !S_ISBLK(sta.st_mode)) {
3911                         pr_err("file descriptor for invalid device\n");
3912                         return 1;
3913                 }
3914                 for (dl = ddf->dlist; dl; dl = dl->next)
3915                         if (dl->major == (int)major(sta.st_rdev) &&
3916                             dl->minor == (int)minor(sta.st_rdev))
3917                                 break;
3918                 if (!dl) {
3919                         pr_err("couldn't find disk %d/%d\n",
3920                                (int)major(sta.st_rdev),
3921                                (int)minor(sta.st_rdev));
3922                         return 1;
3923                 }
3924                 ofd = dl->fd;
3925                 dl->fd = fd;
3926                 ret = (_write_super_to_disk(ddf, dl) != 1);
3927                 dl->fd = ofd;
3928                 return ret;
3929         }
3930
3931         if (posix_memalign(&buf, 512, 512) != 0)
3932                 return 1;
3933         memset(buf, 0, 512);
3934
3935         lseek64(fd, dsize-512, 0);
3936         rc = write(fd, buf, 512);
3937         free(buf);
3938         if (rc < 0)
3939                 return 1;
3940         return 0;
3941 }
3942
3943 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
3944 {
3945         /*
3946          * return:
3947          *  0 same, or first was empty, and second was copied
3948          *  1 second had wrong magic number - but that isn't possible
3949          *  2 wrong uuid
3950          *  3 wrong other info
3951          */
3952         struct ddf_super *first = st->sb;
3953         struct ddf_super *second = tst->sb;
3954         struct dl *dl1, *dl2;
3955         struct vcl *vl1, *vl2;
3956         unsigned int max_vds, max_pds, pd, vd;
3957
3958         if (!first) {
3959                 st->sb = tst->sb;
3960                 tst->sb = NULL;
3961                 return 0;
3962         }
3963
3964         if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3965                 return 2;
3966
3967         /* It is only OK to compare info in the anchor.  Anything else
3968          * could be changing due to a reconfig so must be ignored.
3969          * guid really should be enough anyway.
3970          */
3971
3972         if (!be32_eq(first->active->seq, second->active->seq)) {
3973                 dprintf("sequence number mismatch %u<->%u\n",
3974                         be32_to_cpu(first->active->seq),
3975                         be32_to_cpu(second->active->seq));
3976                 return 0;
3977         }
3978
3979         /*
3980          * At this point we are fairly sure that the meta data matches.
3981          * But the new disk may contain additional local data.
3982          * Add it to the super block.
3983          */
3984         max_vds = be16_to_cpu(first->active->max_vd_entries);
3985         max_pds = be16_to_cpu(first->phys->max_pdes);
3986         for (vl2 = second->conflist; vl2; vl2 = vl2->next) {
3987                 for (vl1 = first->conflist; vl1; vl1 = vl1->next)
3988                         if (!memcmp(vl1->conf.guid, vl2->conf.guid,
3989                                     DDF_GUID_LEN))
3990                                 break;
3991                 if (vl1) {
3992                         if (vl1->other_bvds != NULL &&
3993                             vl1->conf.sec_elmnt_seq !=
3994                             vl2->conf.sec_elmnt_seq) {
3995                                 dprintf("adding BVD %u\n",
3996                                         vl2->conf.sec_elmnt_seq);
3997                                 add_other_bvd(vl1, &vl2->conf,
3998                                               first->conf_rec_len*512);
3999                         }
4000                         continue;
4001                 }
4002
4003                 if (posix_memalign((void **)&vl1, 512,
4004                                    (first->conf_rec_len*512 +
4005                                     offsetof(struct vcl, conf))) != 0) {
4006                         pr_err("could not allocate vcl buf\n");
4007                         return 3;
4008                 }
4009
4010                 vl1->next = first->conflist;
4011                 vl1->block_sizes = NULL;
4012                 memcpy(&vl1->conf, &vl2->conf, first->conf_rec_len*512);
4013                 if (alloc_other_bvds(first, vl1) != 0) {
4014                         pr_err("could not allocate other bvds\n");
4015                         free(vl1);
4016                         return 3;
4017                 }
4018                 for (vd = 0; vd < max_vds; vd++)
4019                         if (!memcmp(first->virt->entries[vd].guid,
4020                                     vl1->conf.guid, DDF_GUID_LEN))
4021                                 break;
4022                 vl1->vcnum = vd;
4023                 dprintf("added config for VD %u\n", vl1->vcnum);
4024                 first->conflist = vl1;
4025         }
4026
4027         for (dl2 = second->dlist; dl2; dl2 = dl2->next) {
4028                 for (dl1 = first->dlist; dl1; dl1 = dl1->next)
4029                         if (be32_eq(dl1->disk.refnum, dl2->disk.refnum))
4030                                 break;
4031                 if (dl1)
4032                         continue;
4033
4034                 if (posix_memalign((void **)&dl1, 512,
4035                        sizeof(*dl1) + (first->max_part) * sizeof(dl1->vlist[0]))
4036                     != 0) {
4037                         pr_err("could not allocate disk info buffer\n");
4038                         return 3;
4039                 }
4040                 memcpy(dl1, dl2, sizeof(*dl1));
4041                 dl1->mdupdate = NULL;
4042                 dl1->next = first->dlist;
4043                 dl1->fd = -1;
4044                 for (pd = 0; pd < max_pds; pd++)
4045                         if (be32_eq(first->phys->entries[pd].refnum,
4046                                     dl1->disk.refnum))
4047                                 break;
4048                 dl1->pdnum = pd < max_pds ? (int)pd : -1;
4049                 if (dl2->spare) {
4050                         if (posix_memalign((void **)&dl1->spare, 512,
4051                                        first->conf_rec_len*512) != 0) {
4052                                 pr_err("could not allocate spare info buf\n");
4053                                 return 3;
4054                         }
4055                         memcpy(dl1->spare, dl2->spare, first->conf_rec_len*512);
4056                 }
4057                 for (vd = 0 ; vd < first->max_part ; vd++) {
4058                         if (!dl2->vlist[vd]) {
4059                                 dl1->vlist[vd] = NULL;
4060                                 continue;
4061                         }
4062                         for (vl1 = first->conflist; vl1; vl1 = vl1->next) {
4063                                 if (!memcmp(vl1->conf.guid,
4064                                             dl2->vlist[vd]->conf.guid,
4065                                             DDF_GUID_LEN))
4066                                         break;
4067                                 dl1->vlist[vd] = vl1;
4068                         }
4069                 }
4070                 first->dlist = dl1;
4071                 dprintf("added disk %d: %08x\n", dl1->pdnum,
4072                         be32_to_cpu(dl1->disk.refnum));
4073         }
4074
4075         return 0;
4076 }
4077
4078 #ifndef MDASSEMBLE
4079 /*
4080  * A new array 'a' has been started which claims to be instance 'inst'
4081  * within container 'c'.
4082  * We need to confirm that the array matches the metadata in 'c' so
4083  * that we don't corrupt any metadata.
4084  */
4085 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
4086 {
4087         struct ddf_super *ddf = c->sb;
4088         int n = atoi(inst);
4089         struct mdinfo *dev;
4090         struct dl *dl;
4091         static const char faulty[] = "faulty";
4092
4093         if (all_ff(ddf->virt->entries[n].guid)) {
4094                 pr_err("subarray %d doesn't exist\n", n);
4095                 return -ENODEV;
4096         }
4097         dprintf("new subarray %d, GUID: %s\n", n,
4098                 guid_str(ddf->virt->entries[n].guid));
4099         for (dev = a->info.devs; dev; dev = dev->next) {
4100                 for (dl = ddf->dlist; dl; dl = dl->next)
4101                         if (dl->major == dev->disk.major &&
4102                             dl->minor == dev->disk.minor)
4103                                 break;
4104                 if (!dl || dl->pdnum < 0) {
4105                         pr_err("device %d/%d of subarray %d not found in meta data\n",
4106                                 dev->disk.major, dev->disk.minor, n);
4107                         return -1;
4108                 }
4109                 if ((be16_to_cpu(ddf->phys->entries[dl->pdnum].state) &
4110                         (DDF_Online|DDF_Missing|DDF_Failed)) != DDF_Online) {
4111                         pr_err("new subarray %d contains broken device %d/%d (%02x)\n",
4112                                n, dl->major, dl->minor,
4113                                be16_to_cpu(ddf->phys->entries[dl->pdnum].state));
4114                         if (write(dev->state_fd, faulty, sizeof(faulty)-1) !=
4115                             sizeof(faulty) - 1)
4116                                 pr_err("Write to state_fd failed\n");
4117                         dev->curr_state = DS_FAULTY;
4118                 }
4119         }
4120         a->info.container_member = n;
4121         return 0;
4122 }
4123
4124 static void handle_missing(struct ddf_super *ddf, struct active_array *a, int inst)
4125 {
4126         /* This member array is being activated.  If any devices
4127          * are missing they must now be marked as failed.
4128          */
4129         struct vd_config *vc;
4130         unsigned int n_bvd;
4131         struct vcl *vcl;
4132         struct dl *dl;
4133         int pd;
4134         int n;
4135         int state;
4136
4137         for (n = 0; ; n++) {
4138                 vc = find_vdcr(ddf, inst, n, &n_bvd, &vcl);
4139                 if (!vc)
4140                         break;
4141                 for (dl = ddf->dlist; dl; dl = dl->next)
4142                         if (be32_eq(dl->disk.refnum, vc->phys_refnum[n_bvd]))
4143                                 break;
4144                 if (dl)
4145                         /* Found this disk, so not missing */
4146                         continue;
4147
4148                 /* Mark the device as failed/missing. */
4149                 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4150                 if (pd >= 0 && be16_and(ddf->phys->entries[pd].state,
4151                                         cpu_to_be16(DDF_Online))) {
4152                         be16_clear(ddf->phys->entries[pd].state,
4153                                    cpu_to_be16(DDF_Online));
4154                         be16_set(ddf->phys->entries[pd].state,
4155                                  cpu_to_be16(DDF_Failed|DDF_Missing));
4156                         vc->phys_refnum[n_bvd] = cpu_to_be32(0);
4157                         ddf_set_updates_pending(ddf, vc);
4158                 }
4159
4160                 /* Mark the array as Degraded */
4161                 state = get_svd_state(ddf, vcl);
4162                 if (ddf->virt->entries[inst].state !=
4163                     ((ddf->virt->entries[inst].state & ~DDF_state_mask)
4164                      | state)) {
4165                         ddf->virt->entries[inst].state =
4166                                 (ddf->virt->entries[inst].state & ~DDF_state_mask)
4167                                 | state;
4168                         a->check_degraded = 1;
4169                         ddf_set_updates_pending(ddf, vc);
4170                 }
4171         }
4172 }
4173
4174 /*
4175  * The array 'a' is to be marked clean in the metadata.
4176  * If '->resync_start' is not ~(unsigned long long)0, then the array is only
4177  * clean up to the point (in sectors).  If that cannot be recorded in the
4178  * metadata, then leave it as dirty.
4179  *
4180  * For DDF, we need to clear the DDF_state_inconsistent bit in the
4181  * !global! virtual_disk.virtual_entry structure.
4182  */
4183 static int ddf_set_array_state(struct active_array *a, int consistent)
4184 {
4185         struct ddf_super *ddf = a->container->sb;
4186         int inst = a->info.container_member;
4187         int old = ddf->virt->entries[inst].state;
4188         if (consistent == 2) {
4189                 handle_missing(ddf, a, inst);
4190                 consistent = 1;
4191                 if (!is_resync_complete(&a->info))
4192                         consistent = 0;
4193         }
4194         if (consistent)
4195                 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
4196         else
4197                 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
4198         if (old != ddf->virt->entries[inst].state)
4199                 ddf_set_updates_pending(ddf, NULL);
4200
4201         old = ddf->virt->entries[inst].init_state;
4202         ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
4203         if (is_resync_complete(&a->info))
4204                 ddf->virt->entries[inst].init_state |= DDF_init_full;
4205         else if (a->info.resync_start == 0)
4206                 ddf->virt->entries[inst].init_state |= DDF_init_not;
4207         else
4208                 ddf->virt->entries[inst].init_state |= DDF_init_quick;
4209         if (old != ddf->virt->entries[inst].init_state)
4210                 ddf_set_updates_pending(ddf, NULL);
4211
4212         dprintf("ddf mark %d/%s (%d) %s %llu\n", inst,
4213                 guid_str(ddf->virt->entries[inst].guid), a->curr_state,
4214                 consistent?"clean":"dirty",
4215                 a->info.resync_start);
4216         return consistent;
4217 }
4218
4219 static int get_bvd_state(const struct ddf_super *ddf,
4220                          const struct vd_config *vc)
4221 {
4222         unsigned int i, n_bvd, working = 0;
4223         unsigned int n_prim = be16_to_cpu(vc->prim_elmnt_count);
4224         int pd, st, state;
4225         char *avail = xcalloc(1, n_prim);
4226         mdu_array_info_t array;
4227
4228         layout_ddf2md(vc, &array);
4229
4230         for (i = 0; i < n_prim; i++) {
4231                 if (!find_index_in_bvd(ddf, vc, i, &n_bvd))
4232                         continue;
4233                 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4234                 if (pd < 0)
4235                         continue;
4236                 st = be16_to_cpu(ddf->phys->entries[pd].state);
4237                 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
4238                     == DDF_Online) {
4239                         working++;
4240                         avail[i] = 1;
4241                 }
4242         }
4243
4244         state = DDF_state_degraded;
4245         if (working == n_prim)
4246                 state = DDF_state_optimal;
4247         else
4248                 switch (vc->prl) {
4249                 case DDF_RAID0:
4250                 case DDF_CONCAT:
4251                 case DDF_JBOD:
4252                         state = DDF_state_failed;
4253                         break;
4254                 case DDF_RAID1:
4255                         if (working == 0)
4256                                 state = DDF_state_failed;
4257                         else if (working >= 2)
4258                                 state = DDF_state_part_optimal;
4259                         break;
4260                 case DDF_RAID1E:
4261                         if (!enough(10, n_prim, array.layout, 1, avail))
4262                                 state = DDF_state_failed;
4263                         break;
4264                 case DDF_RAID4:
4265                 case DDF_RAID5:
4266                         if (working < n_prim - 1)
4267                                 state = DDF_state_failed;
4268                         break;
4269                 case DDF_RAID6:
4270                         if (working < n_prim - 2)
4271                                 state = DDF_state_failed;
4272                         else if (working == n_prim - 1)
4273                                 state = DDF_state_part_optimal;
4274                         break;
4275                 }
4276         return state;
4277 }
4278
4279 static int secondary_state(int state, int other, int seclevel)
4280 {
4281         if (state == DDF_state_optimal && other == DDF_state_optimal)
4282                 return DDF_state_optimal;
4283         if (seclevel == DDF_2MIRRORED) {
4284                 if (state == DDF_state_optimal || other == DDF_state_optimal)
4285                         return DDF_state_part_optimal;
4286                 if (state == DDF_state_failed && other == DDF_state_failed)
4287                         return DDF_state_failed;
4288                 return DDF_state_degraded;
4289         } else {
4290                 if (state == DDF_state_failed || other == DDF_state_failed)
4291                         return DDF_state_failed;
4292                 if (state == DDF_state_degraded || other == DDF_state_degraded)
4293                         return DDF_state_degraded;
4294                 return DDF_state_part_optimal;
4295         }
4296 }
4297
4298 static int get_svd_state(const struct ddf_super *ddf, const struct vcl *vcl)
4299 {
4300         int state = get_bvd_state(ddf, &vcl->conf);
4301         unsigned int i;
4302         for (i = 1; i < vcl->conf.sec_elmnt_count; i++) {
4303                 state = secondary_state(
4304                         state,
4305                         get_bvd_state(ddf, vcl->other_bvds[i-1]),
4306                         vcl->conf.srl);
4307         }
4308         return state;
4309 }
4310
4311 /*
4312  * The state of each disk is stored in the global phys_disk structure
4313  * in phys_disk.entries[n].state.
4314  * This makes various combinations awkward.
4315  * - When a device fails in any array, it must be failed in all arrays
4316  *   that include a part of this device.
4317  * - When a component is rebuilding, we cannot include it officially in the
4318  *   array unless this is the only array that uses the device.
4319  *
4320  * So: when transitioning:
4321  *   Online -> failed,  just set failed flag.  monitor will propagate
4322  *   spare -> online,   the device might need to be added to the array.
4323  *   spare -> failed,   just set failed.  Don't worry if in array or not.
4324  */
4325 static void ddf_set_disk(struct active_array *a, int n, int state)
4326 {
4327         struct ddf_super *ddf = a->container->sb;
4328         unsigned int inst = a->info.container_member, n_bvd;
4329         struct vcl *vcl;
4330         struct vd_config *vc = find_vdcr(ddf, inst, (unsigned int)n,
4331                                          &n_bvd, &vcl);
4332         int pd;
4333         struct mdinfo *mdi;
4334         struct dl *dl;
4335         int update = 0;
4336
4337         dprintf("%d to %x\n", n, state);
4338         if (vc == NULL) {
4339                 dprintf("ddf: cannot find instance %d!!\n", inst);
4340                 return;
4341         }
4342         /* Find the matching slot in 'info'. */
4343         for (mdi = a->info.devs; mdi; mdi = mdi->next)
4344                 if (mdi->disk.raid_disk == n)
4345                         break;
4346         if (!mdi) {
4347                 pr_err("cannot find raid disk %d\n", n);
4348                 return;
4349         }
4350
4351         /* and find the 'dl' entry corresponding to that. */
4352         for (dl = ddf->dlist; dl; dl = dl->next)
4353                 if (mdi->state_fd >= 0 &&
4354                     mdi->disk.major == dl->major &&
4355                     mdi->disk.minor == dl->minor)
4356                         break;
4357         if (!dl) {
4358                 pr_err("cannot find raid disk %d (%d/%d)\n",
4359                        n, mdi->disk.major, mdi->disk.minor);
4360                 return;
4361         }
4362
4363         pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4364         if (pd < 0 || pd != dl->pdnum) {
4365                 /* disk doesn't currently exist or has changed.
4366                  * If it is now in_sync, insert it. */
4367                 dprintf("phys disk not found for %d: %d/%d ref %08x\n",
4368                         dl->pdnum, dl->major, dl->minor,
4369                         be32_to_cpu(dl->disk.refnum));
4370                 dprintf("array %u disk %u ref %08x pd %d\n",
4371                         inst, n_bvd,
4372                         be32_to_cpu(vc->phys_refnum[n_bvd]), pd);
4373                 if ((state & DS_INSYNC) && ! (state & DS_FAULTY) &&
4374                     dl->pdnum >= 0) {
4375                         pd = dl->pdnum;
4376                         vc->phys_refnum[n_bvd] = dl->disk.refnum;
4377                         LBA_OFFSET(ddf, vc)[n_bvd] =
4378                                 cpu_to_be64(mdi->data_offset);
4379                         be16_clear(ddf->phys->entries[pd].type,
4380                                    cpu_to_be16(DDF_Global_Spare));
4381                         be16_set(ddf->phys->entries[pd].type,
4382                                  cpu_to_be16(DDF_Active_in_VD));
4383                         update = 1;
4384                 }
4385         } else {
4386                 be16 old = ddf->phys->entries[pd].state;
4387                 if (state & DS_FAULTY)
4388                         be16_set(ddf->phys->entries[pd].state,
4389                                  cpu_to_be16(DDF_Failed));
4390                 if (state & DS_INSYNC) {
4391                         be16_set(ddf->phys->entries[pd].state,
4392                                  cpu_to_be16(DDF_Online));
4393                         be16_clear(ddf->phys->entries[pd].state,
4394                                    cpu_to_be16(DDF_Rebuilding));
4395                 }
4396                 if (!be16_eq(old, ddf->phys->entries[pd].state))
4397                         update = 1;
4398         }
4399
4400         dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n,
4401                 be32_to_cpu(dl->disk.refnum), state,
4402                 be16_to_cpu(ddf->phys->entries[pd].state));
4403
4404         /* Now we need to check the state of the array and update
4405          * virtual_disk.entries[n].state.
4406          * It needs to be one of "optimal", "degraded", "failed".
4407          * I don't understand 'deleted' or 'missing'.
4408          */
4409         state = get_svd_state(ddf, vcl);
4410
4411         if (ddf->virt->entries[inst].state !=
4412             ((ddf->virt->entries[inst].state & ~DDF_state_mask)
4413              | state)) {
4414                 ddf->virt->entries[inst].state =
4415                         (ddf->virt->entries[inst].state & ~DDF_state_mask)
4416                         | state;
4417                 update = 1;
4418         }
4419         if (update)
4420                 ddf_set_updates_pending(ddf, vc);
4421 }
4422
4423 static void ddf_sync_metadata(struct supertype *st)
4424 {
4425         /*
4426          * Write all data to all devices.
4427          * Later, we might be able to track whether only local changes
4428          * have been made, or whether any global data has been changed,
4429          * but ddf is sufficiently weird that it probably always
4430          * changes global data ....
4431          */
4432         struct ddf_super *ddf = st->sb;
4433         if (!ddf->updates_pending)
4434                 return;
4435         ddf->updates_pending = 0;
4436         __write_init_super_ddf(st);
4437         dprintf("ddf: sync_metadata\n");
4438 }
4439
4440 static int del_from_conflist(struct vcl **list, const char *guid)
4441 {
4442         struct vcl **p;
4443         int found = 0;
4444         for (p = list; p && *p; p = &((*p)->next))
4445                 if (!memcmp((*p)->conf.guid, guid, DDF_GUID_LEN)) {
4446                         found = 1;
4447                         *p = (*p)->next;
4448                 }
4449         return found;
4450 }
4451
4452 static int _kill_subarray_ddf(struct ddf_super *ddf, const char *guid)
4453 {
4454         struct dl *dl;
4455         unsigned int vdnum, i;
4456         vdnum = find_vde_by_guid(ddf, guid);
4457         if (vdnum == DDF_NOTFOUND) {
4458                 pr_err("could not find VD %s\n", guid_str(guid));
4459                 return -1;
4460         }
4461         if (del_from_conflist(&ddf->conflist, guid) == 0) {
4462                 pr_err("could not find conf %s\n", guid_str(guid));
4463                 return -1;
4464         }
4465         for (dl = ddf->dlist; dl; dl = dl->next)
4466                 for (i = 0; i < ddf->max_part; i++)
4467                         if (dl->vlist[i] != NULL &&
4468                             !memcmp(dl->vlist[i]->conf.guid, guid,
4469                                     DDF_GUID_LEN))
4470                                 dl->vlist[i] = NULL;
4471         memset(ddf->virt->entries[vdnum].guid, 0xff, DDF_GUID_LEN);
4472         dprintf("deleted %s\n", guid_str(guid));
4473         return 0;
4474 }
4475
4476 static int kill_subarray_ddf(struct supertype *st)
4477 {
4478         struct ddf_super *ddf = st->sb;
4479         /*
4480          *  currentconf is set in container_content_ddf,
4481          *  called with subarray arg
4482          */
4483         struct vcl *victim = ddf->currentconf;
4484         struct vd_config *conf;
4485         unsigned int vdnum;
4486
4487         ddf->currentconf = NULL;
4488         if (!victim) {
4489                 pr_err("nothing to kill\n");
4490                 return -1;
4491         }
4492         conf = &victim->conf;
4493         vdnum = find_vde_by_guid(ddf, conf->guid);
4494         if (vdnum == DDF_NOTFOUND) {
4495                 pr_err("could not find VD %s\n", guid_str(conf->guid));
4496                 return -1;
4497         }
4498         if (st->update_tail) {
4499                 struct virtual_disk *vd;
4500                 int len = sizeof(struct virtual_disk)
4501                         + sizeof(struct virtual_entry);
4502                 vd = xmalloc(len);
4503                 if (vd == NULL) {
4504                         pr_err("failed to allocate %d bytes\n", len);
4505                         return -1;
4506                 }
4507                 memset(vd, 0 , len);
4508                 vd->magic = DDF_VIRT_RECORDS_MAGIC;
4509                 vd->populated_vdes = cpu_to_be16(0);
4510                 memcpy(vd->entries[0].guid, conf->guid, DDF_GUID_LEN);
4511                 /* we use DDF_state_deleted as marker */
4512                 vd->entries[0].state = DDF_state_deleted;
4513                 append_metadata_update(st, vd, len);
4514         } else {
4515                 _kill_subarray_ddf(ddf, conf->guid);
4516                 ddf_set_updates_pending(ddf, NULL);
4517                 ddf_sync_metadata(st);
4518         }
4519         return 0;
4520 }
4521
4522 static void copy_matching_bvd(struct ddf_super *ddf,
4523                               struct vd_config *conf,
4524                               const struct metadata_update *update)
4525 {
4526         unsigned int mppe =
4527                 be16_to_cpu(ddf->anchor.max_primary_element_entries);
4528         unsigned int len = ddf->conf_rec_len * 512;
4529         char *p;
4530         struct vd_config *vc;
4531         for (p = update->buf; p < update->buf + update->len; p += len) {
4532                 vc = (struct vd_config *) p;
4533                 if (vc->sec_elmnt_seq == conf->sec_elmnt_seq) {
4534                         memcpy(conf->phys_refnum, vc->phys_refnum,
4535                                mppe * (sizeof(__u32) + sizeof(__u64)));
4536                         return;
4537                 }
4538         }
4539         pr_err("no match for BVD %d of %s in update\n",
4540                conf->sec_elmnt_seq, guid_str(conf->guid));
4541 }
4542
4543 static void ddf_process_phys_update(struct supertype *st,
4544                                     struct metadata_update *update)
4545 {
4546         struct ddf_super *ddf = st->sb;
4547         struct phys_disk *pd;
4548         unsigned int ent;
4549
4550         pd = (struct phys_disk*)update->buf;
4551         ent = be16_to_cpu(pd->used_pdes);
4552         if (ent >= be16_to_cpu(ddf->phys->max_pdes))
4553                 return;
4554         if (be16_and(pd->entries[0].state, cpu_to_be16(DDF_Missing))) {
4555                 struct dl **dlp;
4556                 /* removing this disk. */
4557                 be16_set(ddf->phys->entries[ent].state,
4558                          cpu_to_be16(DDF_Missing));
4559                 for (dlp = &ddf->dlist; *dlp; dlp = &(*dlp)->next) {
4560                         struct dl *dl = *dlp;
4561                         if (dl->pdnum == (signed)ent) {
4562                                 close(dl->fd);
4563                                 dl->fd = -1;
4564                                 *dlp = dl->next;
4565                                 update->space = dl->devname;
4566                                 *(void**)dl = update->space_list;
4567                                 update->space_list = (void**)dl;
4568                                 break;
4569                         }
4570                 }
4571                 ddf_set_updates_pending(ddf, NULL);
4572                 return;
4573         }
4574         if (!all_ff(ddf->phys->entries[ent].guid))
4575                 return;
4576         ddf->phys->entries[ent] = pd->entries[0];
4577         ddf->phys->used_pdes = cpu_to_be16
4578                 (1 + be16_to_cpu(ddf->phys->used_pdes));
4579         ddf_set_updates_pending(ddf, NULL);
4580         if (ddf->add_list) {
4581                 struct active_array *a;
4582                 struct dl *al = ddf->add_list;
4583                 ddf->add_list = al->next;
4584
4585                 al->next = ddf->dlist;
4586                 ddf->dlist = al;
4587
4588                 /* As a device has been added, we should check
4589                  * for any degraded devices that might make
4590                  * use of this spare */
4591                 for (a = st->arrays ; a; a=a->next)
4592                         a->check_degraded = 1;
4593         }
4594 }
4595
4596 static void ddf_process_virt_update(struct supertype *st,
4597                                     struct metadata_update *update)
4598 {
4599         struct ddf_super *ddf = st->sb;
4600         struct virtual_disk *vd;
4601         unsigned int ent;
4602
4603         vd = (struct virtual_disk*)update->buf;
4604
4605         if (vd->entries[0].state == DDF_state_deleted) {
4606                 if (_kill_subarray_ddf(ddf, vd->entries[0].guid))
4607                         return;
4608         } else {
4609                 ent = find_vde_by_guid(ddf, vd->entries[0].guid);
4610                 if (ent != DDF_NOTFOUND) {
4611                         dprintf("VD %s exists already in slot %d\n",
4612                                 guid_str(vd->entries[0].guid),
4613                                 ent);
4614                         return;
4615                 }
4616                 ent = find_unused_vde(ddf);
4617                 if (ent == DDF_NOTFOUND)
4618                         return;
4619                 ddf->virt->entries[ent] = vd->entries[0];
4620                 ddf->virt->populated_vdes =
4621                         cpu_to_be16(
4622                                 1 + be16_to_cpu(
4623                                         ddf->virt->populated_vdes));
4624                 dprintf("added VD %s in slot %d(s=%02x i=%02x)\n",
4625                         guid_str(vd->entries[0].guid), ent,
4626                         ddf->virt->entries[ent].state,
4627                         ddf->virt->entries[ent].init_state);
4628         }
4629         ddf_set_updates_pending(ddf, NULL);
4630 }
4631
4632 static void ddf_remove_failed(struct ddf_super *ddf)
4633 {
4634         /* Now remove any 'Failed' devices that are not part
4635          * of any VD.  They will have the Transition flag set.
4636          * Once done, we need to update all dl->pdnum numbers.
4637          */
4638         unsigned int pdnum;
4639         unsigned int pd2 = 0;
4640         struct dl *dl;
4641
4642         for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes);
4643              pdnum++) {
4644                 if (be32_to_cpu(ddf->phys->entries[pdnum].refnum) ==
4645                     0xFFFFFFFF)
4646                         continue;
4647                 if (be16_and(ddf->phys->entries[pdnum].state,
4648                              cpu_to_be16(DDF_Failed))
4649                     && be16_and(ddf->phys->entries[pdnum].state,
4650                                 cpu_to_be16(DDF_Transition))) {
4651                         /* skip this one unless in dlist*/
4652                         for (dl = ddf->dlist; dl; dl = dl->next)
4653                                 if (dl->pdnum == (int)pdnum)
4654                                         break;
4655                         if (!dl)
4656                                 continue;
4657                 }
4658                 if (pdnum == pd2)
4659                         pd2++;
4660                 else {
4661                         ddf->phys->entries[pd2] =
4662                                 ddf->phys->entries[pdnum];
4663                         for (dl = ddf->dlist; dl; dl = dl->next)
4664                                 if (dl->pdnum == (int)pdnum)
4665                                         dl->pdnum = pd2;
4666                         pd2++;
4667                 }
4668         }
4669         ddf->phys->used_pdes = cpu_to_be16(pd2);
4670         while (pd2 < pdnum) {
4671                 memset(ddf->phys->entries[pd2].guid, 0xff,
4672                        DDF_GUID_LEN);
4673                 pd2++;
4674         }
4675 }
4676
4677 static void ddf_update_vlist(struct ddf_super *ddf, struct dl *dl)
4678 {
4679         struct vcl *vcl;
4680         unsigned int vn = 0;
4681         int in_degraded = 0;
4682
4683         if (dl->pdnum < 0)
4684                 return;
4685         for (vcl = ddf->conflist; vcl ; vcl = vcl->next) {
4686                 unsigned int dn, ibvd;
4687                 const struct vd_config *conf;
4688                 int vstate;
4689                 dn = get_pd_index_from_refnum(vcl,
4690                                               dl->disk.refnum,
4691                                               ddf->mppe,
4692                                               &conf, &ibvd);
4693                 if (dn == DDF_NOTFOUND)
4694                         continue;
4695                 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4696                         dl->pdnum,
4697                         be32_to_cpu(dl->disk.refnum),
4698                         guid_str(conf->guid),
4699                         conf->sec_elmnt_seq, vn);
4700                 /* Clear the Transition flag */
4701                 if (be16_and
4702                     (ddf->phys->entries[dl->pdnum].state,
4703                      cpu_to_be16(DDF_Failed)))
4704                         be16_clear(ddf->phys
4705                                    ->entries[dl->pdnum].state,
4706                                    cpu_to_be16(DDF_Transition));
4707                 dl->vlist[vn++] = vcl;
4708                 vstate = ddf->virt->entries[vcl->vcnum].state
4709                         & DDF_state_mask;
4710                 if (vstate == DDF_state_degraded ||
4711                     vstate == DDF_state_part_optimal)
4712                         in_degraded = 1;
4713         }
4714         while (vn < ddf->max_part)
4715                 dl->vlist[vn++] = NULL;
4716         if (dl->vlist[0]) {
4717                 be16_clear(ddf->phys->entries[dl->pdnum].type,
4718                            cpu_to_be16(DDF_Global_Spare));
4719                 if (!be16_and(ddf->phys
4720                               ->entries[dl->pdnum].type,
4721                               cpu_to_be16(DDF_Active_in_VD))) {
4722                         be16_set(ddf->phys
4723                                  ->entries[dl->pdnum].type,
4724                                  cpu_to_be16(DDF_Active_in_VD));
4725                         if (in_degraded)
4726                                 be16_set(ddf->phys
4727                                          ->entries[dl->pdnum]
4728                                          .state,
4729                                          cpu_to_be16
4730                                          (DDF_Rebuilding));
4731                 }
4732         }
4733         if (dl->spare) {
4734                 be16_clear(ddf->phys->entries[dl->pdnum].type,
4735                            cpu_to_be16(DDF_Global_Spare));
4736                 be16_set(ddf->phys->entries[dl->pdnum].type,
4737                          cpu_to_be16(DDF_Spare));
4738         }
4739         if (!dl->vlist[0] && !dl->spare) {
4740                 be16_set(ddf->phys->entries[dl->pdnum].type,
4741                          cpu_to_be16(DDF_Global_Spare));
4742                 be16_clear(ddf->phys->entries[dl->pdnum].type,
4743                            cpu_to_be16(DDF_Spare));
4744                 be16_clear(ddf->phys->entries[dl->pdnum].type,
4745                            cpu_to_be16(DDF_Active_in_VD));
4746         }
4747 }
4748
4749 static void ddf_process_conf_update(struct supertype *st,
4750                                     struct metadata_update *update)
4751 {
4752         struct ddf_super *ddf = st->sb;
4753         struct vd_config *vc;
4754         struct vcl *vcl;
4755         struct dl *dl;
4756         unsigned int ent;
4757         unsigned int pdnum, len;
4758
4759         vc = (struct vd_config*)update->buf;
4760         len = ddf->conf_rec_len * 512;
4761         if ((unsigned int)update->len != len * vc->sec_elmnt_count) {
4762                 pr_err("%s: insufficient data (%d) for %u BVDs\n",
4763                        guid_str(vc->guid), update->len,
4764                        vc->sec_elmnt_count);
4765                 return;
4766         }
4767         for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
4768                 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
4769                         break;
4770         dprintf("conf update for %s (%s)\n",
4771                 guid_str(vc->guid), (vcl ? "old" : "new"));
4772         if (vcl) {
4773                 /* An update, just copy the phys_refnum and lba_offset
4774                  * fields
4775                  */
4776                 unsigned int i;
4777                 unsigned int k;
4778                 copy_matching_bvd(ddf, &vcl->conf, update);
4779                 for (k = 0; k < be16_to_cpu(vc->prim_elmnt_count); k++)
4780                         dprintf("BVD %u has %08x at %llu\n", 0,
4781                                 be32_to_cpu(vcl->conf.phys_refnum[k]),
4782                                 be64_to_cpu(LBA_OFFSET(ddf,
4783                                                        &vcl->conf)[k]));
4784                 for (i = 1; i < vc->sec_elmnt_count; i++) {
4785                         copy_matching_bvd(ddf, vcl->other_bvds[i-1],
4786                                           update);
4787                         for (k = 0; k < be16_to_cpu(
4788                                      vc->prim_elmnt_count); k++)
4789                                 dprintf("BVD %u has %08x at %llu\n", i,
4790                                         be32_to_cpu
4791                                         (vcl->other_bvds[i-1]->
4792                                          phys_refnum[k]),
4793                                         be64_to_cpu
4794                                         (LBA_OFFSET
4795                                          (ddf,
4796                                           vcl->other_bvds[i-1])[k]));
4797                 }
4798         } else {
4799                 /* A new VD_CONF */
4800                 unsigned int i;
4801                 if (!update->space)
4802                         return;
4803                 vcl = update->space;
4804                 update->space = NULL;
4805                 vcl->next = ddf->conflist;
4806                 memcpy(&vcl->conf, vc, len);
4807                 ent = find_vde_by_guid(ddf, vc->guid);
4808                 if (ent == DDF_NOTFOUND)
4809                         return;
4810                 vcl->vcnum = ent;
4811                 ddf->conflist = vcl;
4812                 for (i = 1; i < vc->sec_elmnt_count; i++)
4813                         memcpy(vcl->other_bvds[i-1],
4814                                update->buf + len * i, len);
4815         }
4816         /* Set DDF_Transition on all Failed devices - to help
4817          * us detect those that are no longer in use
4818          */
4819         for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes);
4820              pdnum++)
4821                 if (be16_and(ddf->phys->entries[pdnum].state,
4822                              cpu_to_be16(DDF_Failed)))
4823                         be16_set(ddf->phys->entries[pdnum].state,
4824                                  cpu_to_be16(DDF_Transition));
4825
4826         /* Now make sure vlist is correct for each dl. */
4827         for (dl = ddf->dlist; dl; dl = dl->next)
4828                 ddf_update_vlist(ddf, dl);
4829         ddf_remove_failed(ddf);
4830
4831         ddf_set_updates_pending(ddf, vc);
4832 }
4833
4834 static void ddf_process_update(struct supertype *st,
4835                                struct metadata_update *update)
4836 {
4837         /* Apply this update to the metadata.
4838          * The first 4 bytes are a DDF_*_MAGIC which guides
4839          * our actions.
4840          * Possible update are:
4841          *  DDF_PHYS_RECORDS_MAGIC
4842          *    Add a new physical device or remove an old one.
4843          *    Changes to this record only happen implicitly.
4844          *    used_pdes is the device number.
4845          *  DDF_VIRT_RECORDS_MAGIC
4846          *    Add a new VD.  Possibly also change the 'access' bits.
4847          *    populated_vdes is the entry number.
4848          *  DDF_VD_CONF_MAGIC
4849          *    New or updated VD.  the VIRT_RECORD must already
4850          *    exist.  For an update, phys_refnum and lba_offset
4851          *    (at least) are updated, and the VD_CONF must
4852          *    be written to precisely those devices listed with
4853          *    a phys_refnum.
4854          *  DDF_SPARE_ASSIGN_MAGIC
4855          *    replacement Spare Assignment Record... but for which device?
4856          *
4857          * So, e.g.:
4858          *  - to create a new array, we send a VIRT_RECORD and
4859          *    a VD_CONF.  Then assemble and start the array.
4860          *  - to activate a spare we send a VD_CONF to add the phys_refnum
4861          *    and offset.  This will also mark the spare as active with
4862          *    a spare-assignment record.
4863          */
4864         be32 *magic = (be32 *)update->buf;
4865
4866         dprintf("Process update %x\n", be32_to_cpu(*magic));
4867
4868         if (be32_eq(*magic, DDF_PHYS_RECORDS_MAGIC)) {
4869                 if (update->len == (sizeof(struct phys_disk) +
4870                                     sizeof(struct phys_disk_entry)))
4871                         ddf_process_phys_update(st, update);
4872         } else if (be32_eq(*magic, DDF_VIRT_RECORDS_MAGIC)) {
4873                 if (update->len == (sizeof(struct virtual_disk) +
4874                                     sizeof(struct virtual_entry)))
4875                         ddf_process_virt_update(st, update);
4876         } else if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) {
4877                 ddf_process_conf_update(st, update);
4878         }
4879         /* case DDF_SPARE_ASSIGN_MAGIC */
4880 }
4881
4882 static int ddf_prepare_update(struct supertype *st,
4883                               struct metadata_update *update)
4884 {
4885         /* This update arrived at managemon.
4886          * We are about to pass it to monitor.
4887          * If a malloc is needed, do it here.
4888          */
4889         struct ddf_super *ddf = st->sb;
4890         be32 *magic;
4891         if (update->len < 4)
4892                 return 0;
4893         magic = (be32 *)update->buf;
4894         if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) {
4895                 struct vcl *vcl;
4896                 struct vd_config *conf;
4897                 if (update->len < (int)sizeof(*conf))
4898                         return 0;
4899                 conf = (struct vd_config *) update->buf;
4900                 if (posix_memalign(&update->space, 512,
4901                                    offsetof(struct vcl, conf)
4902                                    + ddf->conf_rec_len * 512) != 0) {
4903                         update->space = NULL;
4904                         return 0;
4905                 }
4906                 vcl = update->space;
4907                 vcl->conf.sec_elmnt_count = conf->sec_elmnt_count;
4908                 if (alloc_other_bvds(ddf, vcl) != 0) {
4909                         free(update->space);
4910                         update->space = NULL;
4911                         return 0;
4912                 }
4913         }
4914         return 1;
4915 }
4916
4917 /*
4918  * Check degraded state of a RAID10.
4919  * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error
4920  */
4921 static int raid10_degraded(struct mdinfo *info)
4922 {
4923         int n_prim, n_bvds;
4924         int i;
4925         struct mdinfo *d;
4926         char *found;
4927         int ret = -1;
4928
4929         n_prim = info->array.layout & ~0x100;
4930         n_bvds = info->array.raid_disks / n_prim;
4931         found = xmalloc(n_bvds);
4932         if (found == NULL)
4933                 return ret;
4934         memset(found, 0, n_bvds);
4935         for (d = info->devs; d; d = d->next) {
4936                 i = d->disk.raid_disk / n_prim;
4937                 if (i >= n_bvds) {
4938                         pr_err("BUG: invalid raid disk\n");
4939                         goto out;
4940                 }
4941                 if (d->state_fd > 0)
4942                         found[i]++;
4943         }
4944         ret = 2;
4945         for (i = 0; i < n_bvds; i++)
4946                 if (!found[i]) {
4947                         dprintf("BVD %d/%d failed\n", i, n_bvds);
4948                         ret = 0;
4949                         goto out;
4950                 } else if (found[i] < n_prim) {
4951                         dprintf("BVD %d/%d degraded\n", i, n_bvds);
4952                         ret = 1;
4953                 }
4954 out:
4955         free(found);
4956         return ret;
4957 }
4958
4959 /*
4960  * Check if the array 'a' is degraded but not failed.
4961  * If it is, find as many spares as are available and needed and
4962  * arrange for their inclusion.
4963  * We only choose devices which are not already in the array,
4964  * and prefer those with a spare-assignment to this array.
4965  * Otherwise we choose global spares - assuming always that
4966  * there is enough room.
4967  * For each spare that we assign, we return an 'mdinfo' which
4968  * describes the position for the device in the array.
4969  * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4970  * the new phys_refnum and lba_offset values.
4971  *
4972  * Only worry about BVDs at the moment.
4973  */
4974 static struct mdinfo *ddf_activate_spare(struct active_array *a,
4975                                          struct metadata_update **updates)
4976 {
4977         int working = 0;
4978         struct mdinfo *d;
4979         struct ddf_super *ddf = a->container->sb;
4980         int global_ok = 0;
4981         struct mdinfo *rv = NULL;
4982         struct mdinfo *di;
4983         struct metadata_update *mu;
4984         struct dl *dl;
4985         int i;
4986         unsigned int j;
4987         struct vcl *vcl;
4988         struct vd_config *vc;
4989         unsigned int n_bvd;
4990
4991         for (d = a->info.devs ; d ; d = d->next) {
4992                 if ((d->curr_state & DS_FAULTY) &&
4993                     d->state_fd >= 0)
4994                         /* wait for Removal to happen */
4995                         return NULL;
4996                 if (d->state_fd >= 0)
4997                         working ++;
4998         }
4999
5000         dprintf("working=%d (%d) level=%d\n", working,
5001                 a->info.array.raid_disks,
5002                 a->info.array.level);
5003         if (working == a->info.array.raid_disks)
5004                 return NULL; /* array not degraded */
5005         switch (a->info.array.level) {
5006         case 1:
5007                 if (working == 0)
5008                         return NULL; /* failed */
5009                 break;
5010         case 4:
5011         case 5:
5012                 if (working < a->info.array.raid_disks - 1)
5013                         return NULL; /* failed */
5014                 break;
5015         case 6:
5016                 if (working < a->info.array.raid_disks - 2)
5017                         return NULL; /* failed */
5018                 break;
5019         case 10:
5020                 if (raid10_degraded(&a->info) < 1)
5021                         return NULL;
5022                 break;
5023         default: /* concat or stripe */
5024                 return NULL; /* failed */
5025         }
5026
5027         /* For each slot, if it is not working, find a spare */
5028         dl = ddf->dlist;
5029         for (i = 0; i < a->info.array.raid_disks; i++) {
5030                 for (d = a->info.devs ; d ; d = d->next)
5031                         if (d->disk.raid_disk == i)
5032                                 break;
5033                 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
5034                 if (d && (d->state_fd >= 0))
5035                         continue;
5036
5037                 /* OK, this device needs recovery.  Find a spare */
5038         again:
5039                 for ( ; dl ; dl = dl->next) {
5040                         unsigned long long esize;
5041                         unsigned long long pos;
5042                         struct mdinfo *d2;
5043                         int is_global = 0;
5044                         int is_dedicated = 0;
5045                         be16 state;
5046
5047                         if (dl->pdnum < 0)
5048                                 continue;
5049                         state = ddf->phys->entries[dl->pdnum].state;
5050                         if (be16_and(state,
5051                                      cpu_to_be16(DDF_Failed|DDF_Missing)) ||
5052                             !be16_and(state,
5053                                       cpu_to_be16(DDF_Online)))
5054                                 continue;
5055
5056                         /* If in this array, skip */
5057                         for (d2 = a->info.devs ; d2 ; d2 = d2->next)
5058                                 if (d2->state_fd >= 0 &&
5059                                     d2->disk.major == dl->major &&
5060                                     d2->disk.minor == dl->minor) {
5061                                         dprintf("%x:%x (%08x) already in array\n",
5062                                                 dl->major, dl->minor,
5063                                                 be32_to_cpu(dl->disk.refnum));
5064                                         break;
5065                                 }
5066                         if (d2)
5067                                 continue;
5068                         if (be16_and(ddf->phys->entries[dl->pdnum].type,
5069                                      cpu_to_be16(DDF_Spare))) {
5070                                 /* Check spare assign record */
5071                                 if (dl->spare) {
5072                                         if (dl->spare->type & DDF_spare_dedicated) {
5073                                                 /* check spare_ents for guid */
5074                                                 unsigned int j;
5075                                                 for (j = 0 ;
5076                                                      j < be16_to_cpu
5077                                                              (dl->spare
5078                                                               ->populated);
5079                                                      j++) {
5080                                                         if (memcmp(dl->spare->spare_ents[j].guid,
5081                                                                    ddf->virt->entries[a->info.container_member].guid,
5082                                                                    DDF_GUID_LEN) == 0)
5083                                                                 is_dedicated = 1;
5084                                                 }
5085                                         } else
5086                                                 is_global = 1;
5087                                 }
5088                         } else if (be16_and(ddf->phys->entries[dl->pdnum].type,
5089                                             cpu_to_be16(DDF_Global_Spare))) {
5090                                 is_global = 1;
5091                         } else if (!be16_and(ddf->phys
5092                                              ->entries[dl->pdnum].state,
5093                                              cpu_to_be16(DDF_Failed))) {
5094                                 /* we can possibly use some of this */
5095                                 is_global = 1;
5096                         }
5097                         if ( ! (is_dedicated ||
5098                                 (is_global && global_ok))) {
5099                                 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
5100                                         is_dedicated, is_global);
5101                                 continue;
5102                         }
5103
5104                         /* We are allowed to use this device - is there space?
5105                          * We need a->info.component_size sectors */
5106                         esize = a->info.component_size;
5107                         pos = find_space(ddf, dl, INVALID_SECTORS, &esize);
5108
5109                         if (esize < a->info.component_size) {
5110                                 dprintf("%x:%x has no room: %llu %llu\n",
5111                                         dl->major, dl->minor,
5112                                         esize, a->info.component_size);
5113                                 /* No room */
5114                                 continue;
5115                         }
5116
5117                         /* Cool, we have a device with some space at pos */
5118                         di = xcalloc(1, sizeof(*di));
5119                         di->disk.number = i;
5120                         di->disk.raid_disk = i;
5121                         di->disk.major = dl->major;
5122                         di->disk.minor = dl->minor;
5123                         di->disk.state = 0;
5124                         di->recovery_start = 0;
5125                         di->data_offset = pos;
5126                         di->component_size = a->info.component_size;
5127                         di->next = rv;
5128                         rv = di;
5129                         dprintf("%x:%x (%08x) to be %d at %llu\n",
5130                                 dl->major, dl->minor,
5131                                 be32_to_cpu(dl->disk.refnum), i, pos);
5132
5133                         break;
5134                 }
5135                 if (!dl && ! global_ok) {
5136                         /* not enough dedicated spares, try global */
5137                         global_ok = 1;
5138                         dl = ddf->dlist;
5139                         goto again;
5140                 }
5141         }
5142
5143         if (!rv)
5144                 /* No spares found */
5145                 return rv;
5146         /* Now 'rv' has a list of devices to return.
5147          * Create a metadata_update record to update the
5148          * phys_refnum and lba_offset values
5149          */
5150         vc = find_vdcr(ddf, a->info.container_member, rv->disk.raid_disk,
5151                        &n_bvd, &vcl);
5152         if (vc == NULL)
5153                 return NULL;
5154
5155         mu = xmalloc(sizeof(*mu));
5156         if (posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) {
5157                 free(mu);
5158                 mu = NULL;
5159         }
5160
5161         mu->len = ddf->conf_rec_len * 512 * vcl->conf.sec_elmnt_count;
5162         mu->buf = xmalloc(mu->len);
5163         mu->space = NULL;
5164         mu->space_list = NULL;
5165         mu->next = *updates;
5166         memcpy(mu->buf, &vcl->conf, ddf->conf_rec_len * 512);
5167         for (j = 1; j < vcl->conf.sec_elmnt_count; j++)
5168                 memcpy(mu->buf + j * ddf->conf_rec_len * 512,
5169                        vcl->other_bvds[j-1], ddf->conf_rec_len * 512);
5170
5171         vc = (struct vd_config*)mu->buf;
5172         for (di = rv ; di ; di = di->next) {
5173                 unsigned int i_sec, i_prim;
5174                 i_sec = di->disk.raid_disk
5175                         / be16_to_cpu(vcl->conf.prim_elmnt_count);
5176                 i_prim = di->disk.raid_disk
5177                         % be16_to_cpu(vcl->conf.prim_elmnt_count);
5178                 vc = (struct vd_config *)(mu->buf
5179                                           + i_sec * ddf->conf_rec_len * 512);
5180                 for (dl = ddf->dlist; dl; dl = dl->next)
5181                         if (dl->major == di->disk.major
5182                             && dl->minor == di->disk.minor)
5183                                 break;
5184                 if (!dl || dl->pdnum < 0) {
5185                         pr_err("BUG: can't find disk %d (%d/%d)\n",
5186                                di->disk.raid_disk,
5187                                di->disk.major, di->disk.minor);
5188                         return NULL;
5189                 }
5190                 vc->phys_refnum[i_prim] = ddf->phys->entries[dl->pdnum].refnum;
5191                 LBA_OFFSET(ddf, vc)[i_prim] = cpu_to_be64(di->data_offset);
5192                 dprintf("BVD %u gets %u: %08x at %llu\n", i_sec, i_prim,
5193                         be32_to_cpu(vc->phys_refnum[i_prim]),
5194                         be64_to_cpu(LBA_OFFSET(ddf, vc)[i_prim]));
5195         }
5196         *updates = mu;
5197         return rv;
5198 }
5199 #endif /* MDASSEMBLE */
5200
5201 static int ddf_level_to_layout(int level)
5202 {
5203         switch(level) {
5204         case 0:
5205         case 1:
5206                 return 0;
5207         case 5:
5208                 return ALGORITHM_LEFT_SYMMETRIC;
5209         case 6:
5210                 return ALGORITHM_ROTATING_N_CONTINUE;
5211         case 10:
5212                 return 0x102;
5213         default:
5214                 return UnSet;
5215         }
5216 }
5217
5218 static void default_geometry_ddf(struct supertype *st, int *level, int *layout, int *chunk)
5219 {
5220         if (level && *level == UnSet)
5221                 *level = LEVEL_CONTAINER;
5222
5223         if (level && layout && *layout == UnSet)
5224                 *layout = ddf_level_to_layout(*level);
5225 }
5226
5227 struct superswitch super_ddf = {
5228 #ifndef MDASSEMBLE
5229         .examine_super  = examine_super_ddf,
5230         .brief_examine_super = brief_examine_super_ddf,
5231         .brief_examine_subarrays = brief_examine_subarrays_ddf,
5232         .export_examine_super = export_examine_super_ddf,
5233         .detail_super   = detail_super_ddf,
5234         .brief_detail_super = brief_detail_super_ddf,
5235         .validate_geometry = validate_geometry_ddf,
5236         .write_init_super = write_init_super_ddf,
5237         .add_to_super   = add_to_super_ddf,
5238         .remove_from_super = remove_from_super_ddf,
5239         .load_container = load_container_ddf,
5240         .copy_metadata = copy_metadata_ddf,
5241         .kill_subarray  = kill_subarray_ddf,
5242 #endif
5243         .match_home     = match_home_ddf,
5244         .uuid_from_super= uuid_from_super_ddf,
5245         .getinfo_super  = getinfo_super_ddf,
5246         .update_super   = update_super_ddf,
5247
5248         .avail_size     = avail_size_ddf,
5249
5250         .compare_super  = compare_super_ddf,
5251
5252         .load_super     = load_super_ddf,
5253         .init_super     = init_super_ddf,
5254         .store_super    = store_super_ddf,
5255         .free_super     = free_super_ddf,
5256         .match_metadata_desc = match_metadata_desc_ddf,
5257         .container_content = container_content_ddf,
5258         .default_geometry = default_geometry_ddf,
5259
5260         .external       = 1,
5261
5262 #ifndef MDASSEMBLE
5263 /* for mdmon */
5264         .open_new       = ddf_open_new,
5265         .set_array_state= ddf_set_array_state,
5266         .set_disk       = ddf_set_disk,
5267         .sync_metadata  = ddf_sync_metadata,
5268         .process_update = ddf_process_update,
5269         .prepare_update = ddf_prepare_update,
5270         .activate_spare = ddf_activate_spare,
5271 #endif
5272         .name = "ddf",
5273 };