sys/geom/raid/md_promise.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
   5  * Copyright (c) 2000 - 2008 Søren Schmidt <sos@FreeBSD.org>
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  */
  29
  30 #include <sys/cdefs.h>
  31 __FBSDID("$FreeBSD$");
  32
  33 #include <sys/param.h>
  34 #include <sys/bio.h>
  35 #include <sys/endian.h>
  36 #include <sys/kernel.h>
  37 #include <sys/kobj.h>
  38 #include <sys/limits.h>
  39 #include <sys/lock.h>
  40 #include <sys/malloc.h>
  41 #include <sys/mutex.h>
  42 #include <sys/systm.h>
  43 #include <geom/geom.h>
  44 #include <geom/geom_dbg.h>
  45 #include "geom/raid/g_raid.h"
  46 #include "g_raid_md_if.h"
  47
  48 static MALLOC_DEFINE(M_MD_PROMISE, "md_promise_data", "GEOM_RAID Promise metadata");
  49
  50 #define PROMISE_MAX_DISKS       8
  51 #define PROMISE_MAX_SUBDISKS    2
  52 #define PROMISE_META_OFFSET     14
  53
  54 struct promise_raid_disk {
  55         uint8_t         flags;                  /* Subdisk status. */
  56 #define PROMISE_F_VALID         0x01
  57 #define PROMISE_F_ONLINE        0x02
  58 #define PROMISE_F_ASSIGNED      0x04
  59 #define PROMISE_F_SPARE         0x08
  60 #define PROMISE_F_DUPLICATE     0x10
  61 #define PROMISE_F_REDIR         0x20
  62 #define PROMISE_F_DOWN          0x40
  63 #define PROMISE_F_READY         0x80
  64
  65         uint8_t         number;                 /* Position in a volume. */
  66         uint8_t         channel;                /* ATA channel number. */
  67         uint8_t         device;                 /* ATA device number. */
  68         uint64_t        id __packed;            /* Subdisk ID. */
  69 } __packed;
  70
  71 struct promise_raid_conf {
  72         char            promise_id[24];
  73 #define PROMISE_MAGIC           "Promise Technology, Inc."
  74 #define FREEBSD_MAGIC           "FreeBSD ATA driver RAID "
  75
  76         uint32_t        dummy_0;
  77         uint64_t        magic_0;
  78 #define PROMISE_MAGIC0(x)       (((uint64_t)(x.channel) << 48) | \
  79                                 ((uint64_t)(x.device != 0) << 56))
  80         uint16_t        magic_1;
  81         uint32_t        magic_2;
  82         uint8_t         filler1[470];
  83
  84         uint32_t        integrity;
  85 #define PROMISE_I_VALID         0x00000080
  86
  87         struct promise_raid_disk        disk;   /* This subdisk info. */
  88         uint32_t        disk_offset;            /* Subdisk offset. */
  89         uint32_t        disk_sectors;           /* Subdisk size */
  90         uint32_t        disk_rebuild;           /* Rebuild position. */
  91         uint16_t        generation;             /* Generation number. */
  92         uint8_t         status;                 /* Volume status. */
  93 #define PROMISE_S_VALID         0x01
  94 #define PROMISE_S_ONLINE        0x02
  95 #define PROMISE_S_INITED        0x04
  96 #define PROMISE_S_READY         0x08
  97 #define PROMISE_S_DEGRADED      0x10
  98 #define PROMISE_S_MARKED        0x20
  99 #define PROMISE_S_MIGRATING     0x40
 100 #define PROMISE_S_FUNCTIONAL    0x80
 101
 102         uint8_t         type;                   /* Voluem type. */
 103 #define PROMISE_T_RAID0         0x00
 104 #define PROMISE_T_RAID1         0x01
 105 #define PROMISE_T_RAID3         0x02
 106 #define PROMISE_T_RAID5         0x04
 107 #define PROMISE_T_SPAN          0x08
 108 #define PROMISE_T_JBOD          0x10
 109
 110         uint8_t         total_disks;            /* Disks in this volume. */
 111         uint8_t         stripe_shift;           /* Strip size. */
 112         uint8_t         array_width;            /* Number of RAID0 stripes. */
 113         uint8_t         array_number;           /* Global volume number. */
 114         uint32_t        total_sectors;          /* Volume size. */
 115         uint16_t        cylinders;              /* Volume geometry: C. */
 116         uint8_t         heads;                  /* Volume geometry: H. */
 117         uint8_t         sectors;                /* Volume geometry: S. */
 118         uint64_t        volume_id __packed;     /* Volume ID, */
 119         struct promise_raid_disk        disks[PROMISE_MAX_DISKS];
 120                                                 /* Subdisks in this volume. */
 121         char            name[32];               /* Volume label. */
 122
 123         uint32_t        filler2[8];
 124         uint32_t        magic_3;        /* Something related to rebuild. */
 125         uint64_t        rebuild_lba64;  /* Per-volume rebuild position. */
 126         uint32_t        magic_4;
 127         uint32_t        magic_5;
 128         uint32_t        total_sectors_high;
 129         uint8_t         magic_6;
 130         uint8_t         sector_size;
 131         uint16_t        magic_7;
 132         uint32_t        magic_8[31];
 133         uint32_t        backup_time;
 134         uint16_t        magic_9;
 135         uint32_t        disk_offset_high;
 136         uint32_t        disk_sectors_high;
 137         uint32_t        disk_rebuild_high;
 138         uint16_t        magic_10;
 139         uint32_t        magic_11[3];
 140         uint32_t        filler3[284];
 141         uint32_t        checksum;
 142 } __packed;
 143
 144 struct g_raid_md_promise_perdisk {
 145         int              pd_updated;
 146         int              pd_subdisks;
 147         struct promise_raid_conf        *pd_meta[PROMISE_MAX_SUBDISKS];
 148 };
 149
 150 struct g_raid_md_promise_pervolume {
 151         struct promise_raid_conf        *pv_meta;
 152         uint64_t                         pv_id;
 153         uint16_t                         pv_generation;
 154         int                              pv_disks_present;
 155         int                              pv_started;
 156         struct callout                   pv_start_co;   /* STARTING state timer. */
 157 };
 158
 159 static g_raid_md_create_t g_raid_md_create_promise;
 160 static g_raid_md_taste_t g_raid_md_taste_promise;
 161 static g_raid_md_event_t g_raid_md_event_promise;
 162 static g_raid_md_volume_event_t g_raid_md_volume_event_promise;
 163 static g_raid_md_ctl_t g_raid_md_ctl_promise;
 164 static g_raid_md_write_t g_raid_md_write_promise;
 165 static g_raid_md_fail_disk_t g_raid_md_fail_disk_promise;
 166 static g_raid_md_free_disk_t g_raid_md_free_disk_promise;
 167 static g_raid_md_free_volume_t g_raid_md_free_volume_promise;
 168 static g_raid_md_free_t g_raid_md_free_promise;
 169
 170 static kobj_method_t g_raid_md_promise_methods[] = {
 171         KOBJMETHOD(g_raid_md_create,    g_raid_md_create_promise),
 172         KOBJMETHOD(g_raid_md_taste,     g_raid_md_taste_promise),
 173         KOBJMETHOD(g_raid_md_event,     g_raid_md_event_promise),
 174         KOBJMETHOD(g_raid_md_volume_event,      g_raid_md_volume_event_promise),
 175         KOBJMETHOD(g_raid_md_ctl,       g_raid_md_ctl_promise),
 176         KOBJMETHOD(g_raid_md_write,     g_raid_md_write_promise),
 177         KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_promise),
 178         KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_promise),
 179         KOBJMETHOD(g_raid_md_free_volume,       g_raid_md_free_volume_promise),
 180         KOBJMETHOD(g_raid_md_free,      g_raid_md_free_promise),
 181         { 0, 0 }
 182 };
 183
 184 static struct g_raid_md_class g_raid_md_promise_class = {
 185         "Promise",
 186         g_raid_md_promise_methods,
 187         sizeof(struct g_raid_md_object),
 188         .mdc_enable = 1,
 189         .mdc_priority = 100
 190 };
 191
 192 static void
 193 g_raid_md_promise_print(struct promise_raid_conf *meta)
 194 {
 195         int i;
 196
 197         if (g_raid_debug < 1)
 198                 return;
 199
 200         printf("********* ATA Promise Metadata *********\n");
 201         printf("promise_id          <%.24s>\n", meta->promise_id);
 202         printf("disk                %02x %02x %02x %02x %016jx\n",
 203             meta->disk.flags, meta->disk.number, meta->disk.channel,
 204             meta->disk.device, meta->disk.id);
 205         printf("disk_offset         %u\n", meta->disk_offset);
 206         printf("disk_sectors        %u\n", meta->disk_sectors);
 207         printf("disk_rebuild        %u\n", meta->disk_rebuild);
 208         printf("generation          %u\n", meta->generation);
 209         printf("status              0x%02x\n", meta->status);
 210         printf("type                %u\n", meta->type);
 211         printf("total_disks         %u\n", meta->total_disks);
 212         printf("stripe_shift        %u\n", meta->stripe_shift);
 213         printf("array_width         %u\n", meta->array_width);
 214         printf("array_number        %u\n", meta->array_number);
 215         printf("total_sectors       %u\n", meta->total_sectors);
 216         printf("cylinders           %u\n", meta->cylinders);
 217         printf("heads               %u\n", meta->heads);
 218         printf("sectors             %u\n", meta->sectors);
 219         printf("volume_id           0x%016jx\n", meta->volume_id);
 220         printf("disks:\n");
 221         for (i = 0; i < PROMISE_MAX_DISKS; i++ ) {
 222                 printf("                    %02x %02x %02x %02x %016jx\n",
 223                     meta->disks[i].flags, meta->disks[i].number,
 224                     meta->disks[i].channel, meta->disks[i].device,
 225                     meta->disks[i].id);
 226         }
 227         printf("name                <%.32s>\n", meta->name);
 228         printf("magic_3             0x%08x\n", meta->magic_3);
 229         printf("rebuild_lba64       %ju\n", meta->rebuild_lba64);
 230         printf("magic_4             0x%08x\n", meta->magic_4);
 231         printf("magic_5             0x%08x\n", meta->magic_5);
 232         printf("total_sectors_high  0x%08x\n", meta->total_sectors_high);
 233         printf("sector_size         %u\n", meta->sector_size);
 234         printf("backup_time         %d\n", meta->backup_time);
 235         printf("disk_offset_high    0x%08x\n", meta->disk_offset_high);
 236         printf("disk_sectors_high   0x%08x\n", meta->disk_sectors_high);
 237         printf("disk_rebuild_high   0x%08x\n", meta->disk_rebuild_high);
 238         printf("=================================================\n");
 239 }
 240
 241 static struct promise_raid_conf *
 242 promise_meta_copy(struct promise_raid_conf *meta)
 243 {
 244         struct promise_raid_conf *nmeta;
 245
 246         nmeta = malloc(sizeof(*nmeta), M_MD_PROMISE, M_WAITOK);
 247         memcpy(nmeta, meta, sizeof(*nmeta));
 248         return (nmeta);
 249 }
 250
 251 static int
 252 promise_meta_find_disk(struct promise_raid_conf *meta, uint64_t id)
 253 {
 254         int pos;
 255
 256         for (pos = 0; pos < meta->total_disks; pos++) {
 257                 if (meta->disks[pos].id == id)
 258                         return (pos);
 259         }
 260         return (-1);
 261 }
 262
 263 static int
 264 promise_meta_unused_range(struct promise_raid_conf **metaarr, int nsd,
 265     off_t sectors, off_t *off, off_t *size)
 266 {
 267         off_t coff, csize, tmp;
 268         int i, j;
 269
 270         sectors -= 131072;
 271         *off = 0;
 272         *size = 0;
 273         coff = 0;
 274         csize = sectors;
 275         i = 0;
 276         while (1) {
 277                 for (j = 0; j < nsd; j++) {
 278                         tmp = ((off_t)metaarr[j]->disk_offset_high << 32) +
 279                             metaarr[j]->disk_offset;
 280                         if (tmp >= coff)
 281                                 csize = MIN(csize, tmp - coff);
 282                 }
 283                 if (csize > *size) {
 284                         *off = coff;
 285                         *size = csize;
 286                 }
 287                 if (i >= nsd)
 288                         break;
 289                 coff = ((off_t)metaarr[i]->disk_offset_high << 32) +
 290                      metaarr[i]->disk_offset +
 291                     ((off_t)metaarr[i]->disk_sectors_high << 32) +
 292                      metaarr[i]->disk_sectors;
 293                 csize = sectors - coff;
 294                 i++;
 295         }
 296         return ((*size > 0) ? 1 : 0);
 297 }
 298
 299 static int
 300 promise_meta_translate_disk(struct g_raid_volume *vol, int md_disk_pos)
 301 {
 302         int disk_pos, width;
 303
 304         if (md_disk_pos >= 0 && vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
 305                 width = vol->v_disks_count / 2;
 306                 disk_pos = (md_disk_pos / width) +
 307                     (md_disk_pos % width) * width;
 308         } else
 309                 disk_pos = md_disk_pos;
 310         return (disk_pos);
 311 }
 312
 313 static void
 314 promise_meta_get_name(struct promise_raid_conf *meta, char *buf)
 315 {
 316         int i;
 317
 318         strncpy(buf, meta->name, 32);
 319         buf[32] = 0;
 320         for (i = 31; i >= 0; i--) {
 321                 if (buf[i] > 0x20)
 322                         break;
 323                 buf[i] = 0;
 324         }
 325 }
 326
 327 static void
 328 promise_meta_put_name(struct promise_raid_conf *meta, char *buf)
 329 {
 330
 331         memset(meta->name, 0x20, 32);
 332         memcpy(meta->name, buf, MIN(strlen(buf), 32));
 333 }
 334
 335 static int
 336 promise_meta_read(struct g_consumer *cp, struct promise_raid_conf **metaarr)
 337 {
 338         struct g_provider *pp;
 339         struct promise_raid_conf *meta;
 340         char *buf;
 341         int error, i, subdisks;
 342         uint32_t checksum, *ptr;
 343
 344         pp = cp->provider;
 345         subdisks = 0;
 346
 347         if (pp->sectorsize * 4 < sizeof(*meta))
 348                 return (subdisks);
 349         if (pp->sectorsize * 4 > maxphys) {
 350                 G_RAID_DEBUG(1, "%s: Blocksize is too big.", pp->name);
 351                 return (subdisks);
 352         }
 353 next:
 354         /* Read metadata block. */
 355         buf = g_read_data(cp, pp->mediasize - pp->sectorsize *
 356             (63 - subdisks * PROMISE_META_OFFSET),
 357             pp->sectorsize * 4, &error);
 358         if (buf == NULL) {
 359                 G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
 360                     pp->name, error);
 361                 return (subdisks);
 362         }
 363         meta = (struct promise_raid_conf *)buf;
 364
 365         /* Check if this is an Promise RAID struct */
 366         if (strncmp(meta->promise_id, PROMISE_MAGIC, strlen(PROMISE_MAGIC)) &&
 367             strncmp(meta->promise_id, FREEBSD_MAGIC, strlen(FREEBSD_MAGIC))) {
 368                 if (subdisks == 0)
 369                         G_RAID_DEBUG(1,
 370                             "Promise signature check failed on %s", pp->name);
 371                 g_free(buf);
 372                 return (subdisks);
 373         }
 374         meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK);
 375         memcpy(meta, buf, MIN(sizeof(*meta), pp->sectorsize * 4));
 376         g_free(buf);
 377
 378         /* Check metadata checksum. */
 379         for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
 380                 checksum += *ptr++;
 381         if (checksum != meta->checksum) {
 382                 G_RAID_DEBUG(1, "Promise checksum check failed on %s", pp->name);
 383                 free(meta, M_MD_PROMISE);
 384                 return (subdisks);
 385         }
 386
 387         if ((meta->integrity & PROMISE_I_VALID) == 0) {
 388                 G_RAID_DEBUG(1, "Promise metadata is invalid on %s", pp->name);
 389                 free(meta, M_MD_PROMISE);
 390                 return (subdisks);
 391         }
 392
 393         if (meta->total_disks > PROMISE_MAX_DISKS) {
 394                 G_RAID_DEBUG(1, "Wrong number of disks on %s (%d)",
 395                     pp->name, meta->total_disks);
 396                 free(meta, M_MD_PROMISE);
 397                 return (subdisks);
 398         }
 399
 400         /* Remove filler garbage from fields used in newer metadata. */
 401         if (meta->disk_offset_high == 0x8b8c8d8e &&
 402             meta->disk_sectors_high == 0x8788898a &&
 403             meta->disk_rebuild_high == 0x83848586) {
 404                 meta->disk_offset_high = 0;
 405                 meta->disk_sectors_high = 0;
 406                 if (meta->disk_rebuild == UINT32_MAX)
 407                         meta->disk_rebuild_high = UINT32_MAX;
 408                 else
 409                         meta->disk_rebuild_high = 0;
 410                 if (meta->total_sectors_high == 0x15161718) {
 411                         meta->total_sectors_high = 0;
 412                         meta->backup_time = 0;
 413                         if (meta->rebuild_lba64 == 0x2122232425262728)
 414                                 meta->rebuild_lba64 = UINT64_MAX;
 415                 }
 416         }
 417         if (meta->sector_size < 1 || meta->sector_size > 8)
 418                 meta->sector_size = 1;
 419
 420         /* Save this part and look for next. */
 421         *metaarr = meta;
 422         metaarr++;
 423         subdisks++;
 424         if (subdisks < PROMISE_MAX_SUBDISKS)
 425                 goto next;
 426
 427         return (subdisks);
 428 }
 429
 430 static int
 431 promise_meta_write(struct g_consumer *cp,
 432     struct promise_raid_conf **metaarr, int nsd)
 433 {
 434         struct g_provider *pp;
 435         struct promise_raid_conf *meta;
 436         char *buf;
 437         off_t off, size;
 438         int error, i, subdisk, fake;
 439         uint32_t checksum, *ptr;
 440
 441         pp = cp->provider;
 442         subdisk = 0;
 443         fake = 0;
 444 next:
 445         buf = malloc(pp->sectorsize * 4, M_MD_PROMISE, M_WAITOK | M_ZERO);
 446         meta = NULL;
 447         if (subdisk < nsd) {
 448                 meta = metaarr[subdisk];
 449         } else if (!fake && promise_meta_unused_range(metaarr, nsd,
 450             cp->provider->mediasize / cp->provider->sectorsize,
 451             &off, &size)) {
 452                 /* Optionally add record for unused space. */
 453                 meta = (struct promise_raid_conf *)buf;
 454                 memcpy(&meta->promise_id[0], PROMISE_MAGIC,
 455                     sizeof(PROMISE_MAGIC) - 1);
 456                 meta->dummy_0 = 0x00020000;
 457                 meta->integrity = PROMISE_I_VALID;
 458                 meta->disk.flags = PROMISE_F_ONLINE | PROMISE_F_VALID;
 459                 meta->disk.number = 0xff;
 460                 arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
 461                 meta->disk_offset_high = off >> 32;
 462                 meta->disk_offset = (uint32_t)off;
 463                 meta->disk_sectors_high = size >> 32;
 464                 meta->disk_sectors = (uint32_t)size;
 465                 meta->disk_rebuild_high = UINT32_MAX;
 466                 meta->disk_rebuild = UINT32_MAX;
 467                 fake = 1;
 468         }
 469         if (meta != NULL) {
 470                 /* Recalculate checksum for case if metadata were changed. */
 471                 meta->checksum = 0;
 472                 for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
 473                         checksum += *ptr++;
 474                 meta->checksum = checksum;
 475                 memcpy(buf, meta, MIN(pp->sectorsize * 4, sizeof(*meta)));
 476         }
 477         error = g_write_data(cp, pp->mediasize - pp->sectorsize *
 478             (63 - subdisk * PROMISE_META_OFFSET),
 479             buf, pp->sectorsize * 4);
 480         if (error != 0) {
 481                 G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
 482                     pp->name, error);
 483         }
 484         free(buf, M_MD_PROMISE);
 485
 486         subdisk++;
 487         if (subdisk < PROMISE_MAX_SUBDISKS)
 488                 goto next;
 489
 490         return (error);
 491 }
 492
 493 static int
 494 promise_meta_erase(struct g_consumer *cp)
 495 {
 496         struct g_provider *pp;
 497         char *buf;
 498         int error, subdisk;
 499
 500         pp = cp->provider;
 501         buf = malloc(4 * pp->sectorsize, M_MD_PROMISE, M_WAITOK | M_ZERO);
 502         for (subdisk = 0; subdisk < PROMISE_MAX_SUBDISKS; subdisk++) {
 503                 error = g_write_data(cp, pp->mediasize - pp->sectorsize *
 504                     (63 - subdisk * PROMISE_META_OFFSET),
 505                     buf, 4 * pp->sectorsize);
 506                 if (error != 0) {
 507                         G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
 508                             pp->name, error);
 509                 }
 510         }
 511         free(buf, M_MD_PROMISE);
 512         return (error);
 513 }
 514
 515 static int
 516 promise_meta_write_spare(struct g_consumer *cp)
 517 {
 518         struct promise_raid_conf *meta;
 519         off_t tmp;
 520         int error;
 521
 522         meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
 523         memcpy(&meta->promise_id[0], PROMISE_MAGIC, sizeof(PROMISE_MAGIC) - 1);
 524         meta->dummy_0 = 0x00020000;
 525         meta->integrity = PROMISE_I_VALID;
 526         meta->disk.flags = PROMISE_F_SPARE | PROMISE_F_ONLINE | PROMISE_F_VALID;
 527         meta->disk.number = 0xff;
 528         arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
 529         tmp = cp->provider->mediasize / cp->provider->sectorsize - 131072;
 530         meta->disk_sectors_high = tmp >> 32;
 531         meta->disk_sectors = (uint32_t)tmp;
 532         meta->disk_rebuild_high = UINT32_MAX;
 533         meta->disk_rebuild = UINT32_MAX;
 534         error = promise_meta_write(cp, &meta, 1);
 535         free(meta, M_MD_PROMISE);
 536         return (error);
 537 }
 538
 539 static struct g_raid_volume *
 540 g_raid_md_promise_get_volume(struct g_raid_softc *sc, uint64_t id)
 541 {
 542         struct g_raid_volume    *vol;
 543         struct g_raid_md_promise_pervolume *pv;
 544
 545         TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
 546                 pv = vol->v_md_data;
 547                 if (pv->pv_id == id)
 548                         break;
 549         }
 550         return (vol);
 551 }
 552
 553 static int
 554 g_raid_md_promise_purge_volumes(struct g_raid_softc *sc)
 555 {
 556         struct g_raid_volume    *vol, *tvol;
 557         struct g_raid_md_promise_pervolume *pv;
 558         int i, res;
 559
 560         res = 0;
 561         TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tvol) {
 562                 pv = vol->v_md_data;
 563                 if (!pv->pv_started || vol->v_stopping)
 564                         continue;
 565                 for (i = 0; i < vol->v_disks_count; i++) {
 566                         if (vol->v_subdisks[i].sd_state != G_RAID_SUBDISK_S_NONE)
 567                                 break;
 568                 }
 569                 if (i >= vol->v_disks_count) {
 570                         g_raid_destroy_volume(vol);
 571                         res = 1;
 572                 }
 573         }
 574         return (res);
 575 }
 576
 577 static int
 578 g_raid_md_promise_purge_disks(struct g_raid_softc *sc)
 579 {
 580         struct g_raid_disk      *disk, *tdisk;
 581         struct g_raid_volume    *vol;
 582         struct g_raid_md_promise_perdisk *pd;
 583         int i, j, res;
 584
 585         res = 0;
 586         TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) {
 587                 if (disk->d_state == G_RAID_DISK_S_SPARE)
 588                         continue;
 589                 pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
 590
 591                 /* Scan for deleted volumes. */
 592                 for (i = 0; i < pd->pd_subdisks; ) {
 593                         vol = g_raid_md_promise_get_volume(sc,
 594                             pd->pd_meta[i]->volume_id);
 595                         if (vol != NULL && !vol->v_stopping) {
 596                                 i++;
 597                                 continue;
 598                         }
 599                         free(pd->pd_meta[i], M_MD_PROMISE);
 600                         for (j = i; j < pd->pd_subdisks - 1; j++)
 601                                 pd->pd_meta[j] = pd->pd_meta[j + 1];
 602                         pd->pd_meta[pd->pd_subdisks - 1] = NULL;
 603                         pd->pd_subdisks--;
 604                         pd->pd_updated = 1;
 605                 }
 606
 607                 /* If there is no metadata left - erase and delete disk. */
 608                 if (pd->pd_subdisks == 0) {
 609                         promise_meta_erase(disk->d_consumer);
 610                         g_raid_destroy_disk(disk);
 611                         res = 1;
 612                 }
 613         }
 614         return (res);
 615 }
 616
 617 static int
 618 g_raid_md_promise_supported(int level, int qual, int disks, int force)
 619 {
 620
 621         if (disks > PROMISE_MAX_DISKS)
 622                 return (0);
 623         switch (level) {
 624         case G_RAID_VOLUME_RL_RAID0:
 625                 if (disks < 1)
 626                         return (0);
 627                 if (!force && disks < 2)
 628                         return (0);
 629                 break;
 630         case G_RAID_VOLUME_RL_RAID1:
 631                 if (disks < 1)
 632                         return (0);
 633                 if (!force && (disks != 2))
 634                         return (0);
 635                 break;
 636         case G_RAID_VOLUME_RL_RAID1E:
 637                 if (disks < 2)
 638                         return (0);
 639                 if (disks % 2 != 0)
 640                         return (0);
 641                 if (!force && (disks != 4))
 642                         return (0);
 643                 break;
 644         case G_RAID_VOLUME_RL_SINGLE:
 645                 if (disks != 1)
 646                         return (0);
 647                 break;
 648         case G_RAID_VOLUME_RL_CONCAT:
 649                 if (disks < 2)
 650                         return (0);
 651                 break;
 652         case G_RAID_VOLUME_RL_RAID5:
 653                 if (disks < 3)
 654                         return (0);
 655                 if (qual != G_RAID_VOLUME_RLQ_R5LA)
 656                         return (0);
 657                 break;
 658         default:
 659                 return (0);
 660         }
 661         if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
 662                 return (0);
 663         return (1);
 664 }
 665
 666 static int
 667 g_raid_md_promise_start_disk(struct g_raid_disk *disk, int sdn,
 668     struct g_raid_volume *vol)
 669 {
 670         struct g_raid_softc *sc;
 671         struct g_raid_subdisk *sd;
 672         struct g_raid_md_promise_perdisk *pd;
 673         struct g_raid_md_promise_pervolume *pv;
 674         struct promise_raid_conf *meta;
 675         off_t eoff, esize, size;
 676         int disk_pos, md_disk_pos, i, resurrection = 0;
 677
 678         sc = disk->d_softc;
 679         pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
 680
 681         pv = vol->v_md_data;
 682         meta = pv->pv_meta;
 683
 684         if (sdn >= 0) {
 685                 /* Find disk position in metadata by its serial. */
 686                 md_disk_pos = promise_meta_find_disk(meta, pd->pd_meta[sdn]->disk.id);
 687                 /* For RAID0+1 we need to translate order. */
 688                 disk_pos = promise_meta_translate_disk(vol, md_disk_pos);
 689         } else {
 690                 md_disk_pos = -1;
 691                 disk_pos = -1;
 692         }
 693         if (disk_pos < 0) {
 694                 G_RAID_DEBUG1(1, sc, "Disk %s is not part of the volume %s",
 695                     g_raid_get_diskname(disk), vol->v_name);
 696                 /* Failed stale disk is useless for us. */
 697                 if (sdn >= 0 &&
 698                     pd->pd_meta[sdn]->disk.flags & PROMISE_F_DOWN) {
 699                         g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED);
 700                         return (0);
 701                 }
 702                 /* If we were given specific metadata subdisk - erase it. */
 703                 if (sdn >= 0) {
 704                         free(pd->pd_meta[sdn], M_MD_PROMISE);
 705                         for (i = sdn; i < pd->pd_subdisks - 1; i++)
 706                                 pd->pd_meta[i] = pd->pd_meta[i + 1];
 707                         pd->pd_meta[pd->pd_subdisks - 1] = NULL;
 708                         pd->pd_subdisks--;
 709                 }
 710                 /* If we are in the start process, that's all for now. */
 711                 if (!pv->pv_started)
 712                         goto nofit;
 713                 /*
 714                  * If we have already started - try to get use of the disk.
 715                  * Try to replace OFFLINE disks first, then FAILED.
 716                  */
 717                 promise_meta_unused_range(pd->pd_meta, pd->pd_subdisks,
 718                     disk->d_consumer->provider->mediasize /
 719                     disk->d_consumer->provider->sectorsize,
 720                     &eoff, &esize);
 721                 if (esize == 0) {
 722                         G_RAID_DEBUG1(1, sc, "No free space on disk %s",
 723                             g_raid_get_diskname(disk));
 724                         goto nofit;
 725                 }
 726                 size = INT64_MAX;
 727                 for (i = 0; i < vol->v_disks_count; i++) {
 728                         sd = &vol->v_subdisks[i];
 729                         if (sd->sd_state != G_RAID_SUBDISK_S_NONE)
 730                                 size = sd->sd_size;
 731                         if (sd->sd_state <= G_RAID_SUBDISK_S_FAILED &&
 732                             (disk_pos < 0 ||
 733                              vol->v_subdisks[i].sd_state < sd->sd_state))
 734                                 disk_pos = i;
 735                 }
 736                 if (disk_pos >= 0 &&
 737                     vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT &&
 738                     (off_t)esize * 512 < size) {
 739                         G_RAID_DEBUG1(1, sc, "Disk %s free space "
 740                             "is too small (%ju < %ju)",
 741                             g_raid_get_diskname(disk),
 742                             (off_t)esize * 512, size);
 743                         disk_pos = -1;
 744                 }
 745                 if (disk_pos >= 0) {
 746                         if (vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT)
 747                                 esize = size / 512;
 748                         /* For RAID0+1 we need to translate order. */
 749                         md_disk_pos = promise_meta_translate_disk(vol, disk_pos);
 750                 } else {
 751 nofit:
 752                         if (pd->pd_subdisks == 0) {
 753                                 g_raid_change_disk_state(disk,
 754                                     G_RAID_DISK_S_SPARE);
 755                         }
 756                         return (0);
 757                 }
 758                 G_RAID_DEBUG1(1, sc, "Disk %s takes pos %d in the volume %s",
 759                     g_raid_get_diskname(disk), disk_pos, vol->v_name);
 760                 resurrection = 1;
 761         }
 762
 763         sd = &vol->v_subdisks[disk_pos];
 764
 765         if (resurrection && sd->sd_disk != NULL) {
 766                 g_raid_change_disk_state(sd->sd_disk,
 767                     G_RAID_DISK_S_STALE_FAILED);
 768                 TAILQ_REMOVE(&sd->sd_disk->d_subdisks,
 769                     sd, sd_next);
 770         }
 771         vol->v_subdisks[disk_pos].sd_disk = disk;
 772         TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
 773
 774         /* Welcome the new disk. */
 775         if (resurrection)
 776                 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
 777         else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN)
 778                 g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
 779         else
 780                 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
 781
 782         if (resurrection) {
 783                 sd->sd_offset = (off_t)eoff * 512;
 784                 sd->sd_size = (off_t)esize * 512;
 785         } else {
 786                 sd->sd_offset = (((off_t)pd->pd_meta[sdn]->disk_offset_high
 787                     << 32) + pd->pd_meta[sdn]->disk_offset) * 512;
 788                 sd->sd_size = (((off_t)pd->pd_meta[sdn]->disk_sectors_high
 789                     << 32) + pd->pd_meta[sdn]->disk_sectors) * 512;
 790         }
 791
 792         if (resurrection) {
 793                 /* Stale disk, almost same as new. */
 794                 g_raid_change_subdisk_state(sd,
 795                     G_RAID_SUBDISK_S_NEW);
 796         } else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN) {
 797                 /* Failed disk. */
 798                 g_raid_change_subdisk_state(sd,
 799                     G_RAID_SUBDISK_S_FAILED);
 800         } else if (meta->disks[md_disk_pos].flags & PROMISE_F_REDIR) {
 801                 /* Rebuilding disk. */
 802                 g_raid_change_subdisk_state(sd,
 803                     G_RAID_SUBDISK_S_REBUILD);
 804                 if (pd->pd_meta[sdn]->generation != meta->generation)
 805                         sd->sd_rebuild_pos = 0;
 806                 else {
 807                         sd->sd_rebuild_pos =
 808                             (((off_t)pd->pd_meta[sdn]->disk_rebuild_high << 32) +
 809                              pd->pd_meta[sdn]->disk_rebuild) * 512;
 810                 }
 811         } else if (!(meta->disks[md_disk_pos].flags & PROMISE_F_ONLINE)) {
 812                 /* Rebuilding disk. */
 813                 g_raid_change_subdisk_state(sd,
 814                     G_RAID_SUBDISK_S_NEW);
 815         } else if (pd->pd_meta[sdn]->generation != meta->generation ||
 816             (meta->status & PROMISE_S_MARKED)) {
 817                 /* Stale disk or dirty volume (unclean shutdown). */
 818                 g_raid_change_subdisk_state(sd,
 819                     G_RAID_SUBDISK_S_STALE);
 820         } else {
 821                 /* Up to date disk. */
 822                 g_raid_change_subdisk_state(sd,
 823                     G_RAID_SUBDISK_S_ACTIVE);
 824         }
 825         g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
 826             G_RAID_EVENT_SUBDISK);
 827
 828         return (resurrection);
 829 }
 830
 831 static void
 832 g_raid_md_promise_refill(struct g_raid_softc *sc)
 833 {
 834         struct g_raid_volume *vol;
 835         struct g_raid_subdisk *sd;
 836         struct g_raid_disk *disk;
 837         struct g_raid_md_object *md;
 838         struct g_raid_md_promise_perdisk *pd;
 839         struct g_raid_md_promise_pervolume *pv;
 840         int update, updated, i, bad;
 841
 842         md = sc->sc_md;
 843 restart:
 844         updated = 0;
 845         TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
 846                 pv = vol->v_md_data;
 847                 if (!pv->pv_started || vol->v_stopping)
 848                         continue;
 849
 850                 /* Search for subdisk that needs replacement. */
 851                 bad = 0;
 852                 for (i = 0; i < vol->v_disks_count; i++) {
 853                         sd = &vol->v_subdisks[i];
 854                         if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
 855                             sd->sd_state == G_RAID_SUBDISK_S_FAILED)
 856                                 bad = 1;
 857                 }
 858                 if (!bad)
 859                         continue;
 860
 861                 G_RAID_DEBUG1(1, sc, "Volume %s is not complete, "
 862                     "trying to refill.", vol->v_name);
 863
 864                 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
 865                         /* Skip failed. */
 866                         if (disk->d_state < G_RAID_DISK_S_SPARE)
 867                                 continue;
 868                         /* Skip already used by this volume. */
 869                         for (i = 0; i < vol->v_disks_count; i++) {
 870                                 sd = &vol->v_subdisks[i];
 871                                 if (sd->sd_disk == disk)
 872                                         break;
 873                         }
 874                         if (i < vol->v_disks_count)
 875                                 continue;
 876
 877                         /* Try to use disk if it has empty extents. */
 878                         pd = disk->d_md_data;
 879                         if (pd->pd_subdisks < PROMISE_MAX_SUBDISKS) {
 880                                 update =
 881                                     g_raid_md_promise_start_disk(disk, -1, vol);
 882                         } else
 883                                 update = 0;
 884                         if (update) {
 885                                 updated = 1;
 886                                 g_raid_md_write_promise(md, vol, NULL, disk);
 887                                 break;
 888                         }
 889                 }
 890         }
 891         if (updated)
 892                 goto restart;
 893 }
 894
 895 static void
 896 g_raid_md_promise_start(struct g_raid_volume *vol)
 897 {
 898         struct g_raid_softc *sc;
 899         struct g_raid_subdisk *sd;
 900         struct g_raid_disk *disk;
 901         struct g_raid_md_object *md;
 902         struct g_raid_md_promise_perdisk *pd;
 903         struct g_raid_md_promise_pervolume *pv;
 904         struct promise_raid_conf *meta;
 905         u_int i;
 906
 907         sc = vol->v_softc;
 908         md = sc->sc_md;
 909         pv = vol->v_md_data;
 910         meta = pv->pv_meta;
 911
 912         vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
 913         if (meta->type == PROMISE_T_RAID0)
 914                 vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
 915         else if (meta->type == PROMISE_T_RAID1) {
 916                 if (meta->array_width == 1)
 917                         vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
 918                 else
 919                         vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
 920         } else if (meta->type == PROMISE_T_RAID3)
 921                 vol->v_raid_level = G_RAID_VOLUME_RL_RAID3;
 922         else if (meta->type == PROMISE_T_RAID5) {
 923                 vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
 924                 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
 925         } else if (meta->type == PROMISE_T_SPAN)
 926                 vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
 927         else if (meta->type == PROMISE_T_JBOD)
 928                 vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
 929         else
 930                 vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
 931         vol->v_strip_size = 512 << meta->stripe_shift; //ZZZ
 932         vol->v_disks_count = meta->total_disks;
 933         vol->v_mediasize = (off_t)meta->total_sectors * 512; //ZZZ
 934         if (meta->total_sectors_high < 256) /* If value looks sane. */
 935                 vol->v_mediasize +=
 936                     ((off_t)meta->total_sectors_high << 32) * 512; //ZZZ
 937         vol->v_sectorsize = 512 * meta->sector_size;
 938         for (i = 0; i < vol->v_disks_count; i++) {
 939                 sd = &vol->v_subdisks[i];
 940                 sd->sd_offset = (((off_t)meta->disk_offset_high << 32) +
 941                     meta->disk_offset) * 512;
 942                 sd->sd_size = (((off_t)meta->disk_sectors_high << 32) +
 943                     meta->disk_sectors) * 512;
 944         }
 945         g_raid_start_volume(vol);
 946
 947         /* Make all disks found till the moment take their places. */
 948         TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
 949                 pd = disk->d_md_data;
 950                 for (i = 0; i < pd->pd_subdisks; i++) {
 951                         if (pd->pd_meta[i]->volume_id == meta->volume_id)
 952                                 g_raid_md_promise_start_disk(disk, i, vol);
 953                 }
 954         }
 955
 956         pv->pv_started = 1;
 957         callout_stop(&pv->pv_start_co);
 958         G_RAID_DEBUG1(0, sc, "Volume started.");
 959         g_raid_md_write_promise(md, vol, NULL, NULL);
 960
 961         /* Pickup any STALE/SPARE disks to refill array if needed. */
 962         g_raid_md_promise_refill(sc);
 963
 964         g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
 965 }
 966
 967 static void
 968 g_raid_promise_go(void *arg)
 969 {
 970         struct g_raid_volume *vol;
 971         struct g_raid_softc *sc;
 972         struct g_raid_md_promise_pervolume *pv;
 973
 974         vol = arg;
 975         pv = vol->v_md_data;
 976         sc = vol->v_softc;
 977         if (!pv->pv_started) {
 978                 G_RAID_DEBUG1(0, sc, "Force volume start due to timeout.");
 979                 g_raid_event_send(vol, G_RAID_VOLUME_E_STARTMD,
 980                     G_RAID_EVENT_VOLUME);
 981         }
 982 }
 983
 984 static void
 985 g_raid_md_promise_new_disk(struct g_raid_disk *disk)
 986 {
 987         struct g_raid_softc *sc;
 988         struct g_raid_md_object *md;
 989         struct promise_raid_conf *pdmeta;
 990         struct g_raid_md_promise_perdisk *pd;
 991         struct g_raid_md_promise_pervolume *pv;
 992         struct g_raid_volume *vol;
 993         int i;
 994         char buf[33];
 995
 996         sc = disk->d_softc;
 997         md = sc->sc_md;
 998         pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
 999
1000         if (pd->pd_subdisks == 0) {
1001                 g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
1002                 g_raid_md_promise_refill(sc);
1003                 return;
1004         }
1005
1006         for (i = 0; i < pd->pd_subdisks; i++) {
1007                 pdmeta = pd->pd_meta[i];
1008
1009                 /* Look for volume with matching ID. */
1010                 vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
1011                 if (vol == NULL) {
1012                         promise_meta_get_name(pdmeta, buf);
1013                         vol = g_raid_create_volume(sc, buf, pdmeta->array_number);
1014                         pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1015                         pv->pv_id = pdmeta->volume_id;
1016                         vol->v_md_data = pv;
1017                         callout_init(&pv->pv_start_co, 1);
1018                         callout_reset(&pv->pv_start_co,
1019                             g_raid_start_timeout * hz,
1020                             g_raid_promise_go, vol);
1021                 } else
1022                         pv = vol->v_md_data;
1023
1024                 /* If we haven't started yet - check metadata freshness. */
1025                 if (pv->pv_meta == NULL || !pv->pv_started) {
1026                         if (pv->pv_meta == NULL ||
1027                             ((int16_t)(pdmeta->generation - pv->pv_generation)) > 0) {
1028                                 G_RAID_DEBUG1(1, sc, "Newer disk");
1029                                 if (pv->pv_meta != NULL)
1030                                         free(pv->pv_meta, M_MD_PROMISE);
1031                                 pv->pv_meta = promise_meta_copy(pdmeta);
1032                                 pv->pv_generation = pv->pv_meta->generation;
1033                                 pv->pv_disks_present = 1;
1034                         } else if (pdmeta->generation == pv->pv_generation) {
1035                                 pv->pv_disks_present++;
1036                                 G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
1037                                     pv->pv_disks_present,
1038                                     pv->pv_meta->total_disks);
1039                         } else {
1040                                 G_RAID_DEBUG1(1, sc, "Older disk");
1041                         }
1042                 }
1043         }
1044
1045         for (i = 0; i < pd->pd_subdisks; i++) {
1046                 pdmeta = pd->pd_meta[i];
1047
1048                 /* Look for volume with matching ID. */
1049                 vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
1050                 if (vol == NULL)
1051                         continue;
1052                 pv = vol->v_md_data;
1053
1054                 if (pv->pv_started) {
1055                         if (g_raid_md_promise_start_disk(disk, i, vol))
1056                                 g_raid_md_write_promise(md, vol, NULL, NULL);
1057                 } else {
1058                         /* If we collected all needed disks - start array. */
1059                         if (pv->pv_disks_present == pv->pv_meta->total_disks)
1060                                 g_raid_md_promise_start(vol);
1061                 }
1062         }
1063 }
1064
1065 static int
1066 g_raid_md_create_promise(struct g_raid_md_object *md, struct g_class *mp,
1067     struct g_geom **gp)
1068 {
1069         struct g_geom *geom;
1070         struct g_raid_softc *sc;
1071
1072         /* Search for existing node. */
1073         LIST_FOREACH(geom, &mp->geom, geom) {
1074                 sc = geom->softc;
1075                 if (sc == NULL)
1076                         continue;
1077                 if (sc->sc_stopping != 0)
1078                         continue;
1079                 if (sc->sc_md->mdo_class != md->mdo_class)
1080                         continue;
1081                 break;
1082         }
1083         if (geom != NULL) {
1084                 *gp = geom;
1085                 return (G_RAID_MD_TASTE_EXISTING);
1086         }
1087
1088         /* Create new one if not found. */
1089         sc = g_raid_create_node(mp, "Promise", md);
1090         if (sc == NULL)
1091                 return (G_RAID_MD_TASTE_FAIL);
1092         md->mdo_softc = sc;
1093         *gp = sc->sc_geom;
1094         return (G_RAID_MD_TASTE_NEW);
1095 }
1096
1097 static int
1098 g_raid_md_taste_promise(struct g_raid_md_object *md, struct g_class *mp,
1099                               struct g_consumer *cp, struct g_geom **gp)
1100 {
1101         struct g_consumer *rcp;
1102         struct g_provider *pp;
1103         struct g_raid_softc *sc;
1104         struct g_raid_disk *disk;
1105         struct promise_raid_conf *metaarr[4];
1106         struct g_raid_md_promise_perdisk *pd;
1107         struct g_geom *geom;
1108         int i, j, result, len, subdisks;
1109         char name[16];
1110         uint16_t vendor;
1111
1112         G_RAID_DEBUG(1, "Tasting Promise on %s", cp->provider->name);
1113         pp = cp->provider;
1114
1115         /* Read metadata from device. */
1116         g_topology_unlock();
1117         vendor = 0xffff;
1118         len = sizeof(vendor);
1119         if (pp->geom->rank == 1)
1120                 g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
1121         subdisks = promise_meta_read(cp, metaarr);
1122         g_topology_lock();
1123         if (subdisks == 0) {
1124                 if (g_raid_aggressive_spare) {
1125                         if (vendor == 0x105a || vendor == 0x1002) {
1126                                 G_RAID_DEBUG(1,
1127                                     "No Promise metadata, forcing spare.");
1128                                 goto search;
1129                         } else {
1130                                 G_RAID_DEBUG(1,
1131                                     "Promise/ATI vendor mismatch "
1132                                     "0x%04x != 0x105a/0x1002",
1133                                     vendor);
1134                         }
1135                 }
1136                 return (G_RAID_MD_TASTE_FAIL);
1137         }
1138
1139         /* Metadata valid. Print it. */
1140         for (i = 0; i < subdisks; i++)
1141                 g_raid_md_promise_print(metaarr[i]);
1142
1143         /* Purge meaningless (empty/spare) records. */
1144         for (i = 0; i < subdisks; ) {
1145                 if (metaarr[i]->disk.flags & PROMISE_F_ASSIGNED) {
1146                         i++;
1147                         continue;
1148                 }
1149                 free(metaarr[i], M_MD_PROMISE);
1150                 for (j = i; j < subdisks - 1; j++)
1151                         metaarr[i] = metaarr[j + 1];
1152                 metaarr[subdisks - 1] = NULL;
1153                 subdisks--;
1154         }
1155
1156 search:
1157         /* Search for matching node. */
1158         sc = NULL;
1159         LIST_FOREACH(geom, &mp->geom, geom) {
1160                 sc = geom->softc;
1161                 if (sc == NULL)
1162                         continue;
1163                 if (sc->sc_stopping != 0)
1164                         continue;
1165                 if (sc->sc_md->mdo_class != md->mdo_class)
1166                         continue;
1167                 break;
1168         }
1169
1170         /* Found matching node. */
1171         if (geom != NULL) {
1172                 G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
1173                 result = G_RAID_MD_TASTE_EXISTING;
1174
1175         } else { /* Not found matching node -- create one. */
1176                 result = G_RAID_MD_TASTE_NEW;
1177                 snprintf(name, sizeof(name), "Promise");
1178                 sc = g_raid_create_node(mp, name, md);
1179                 md->mdo_softc = sc;
1180                 geom = sc->sc_geom;
1181         }
1182
1183         /* There is no return after this point, so we close passed consumer. */
1184         g_access(cp, -1, 0, 0);
1185
1186         rcp = g_new_consumer(geom);
1187         rcp->flags |= G_CF_DIRECT_RECEIVE;
1188         g_attach(rcp, pp);
1189         if (g_access(rcp, 1, 1, 1) != 0)
1190                 ; //goto fail1;
1191
1192         g_topology_unlock();
1193         sx_xlock(&sc->sc_lock);
1194
1195         pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1196         pd->pd_subdisks = subdisks;
1197         for (i = 0; i < subdisks; i++)
1198                 pd->pd_meta[i] = metaarr[i];
1199         disk = g_raid_create_disk(sc);
1200         disk->d_md_data = (void *)pd;
1201         disk->d_consumer = rcp;
1202         rcp->private = disk;
1203
1204         g_raid_get_disk_info(disk);
1205
1206         g_raid_md_promise_new_disk(disk);
1207
1208         sx_xunlock(&sc->sc_lock);
1209         g_topology_lock();
1210         *gp = geom;
1211         return (result);
1212 }
1213
1214 static int
1215 g_raid_md_event_promise(struct g_raid_md_object *md,
1216     struct g_raid_disk *disk, u_int event)
1217 {
1218         struct g_raid_softc *sc;
1219
1220         sc = md->mdo_softc;
1221         if (disk == NULL)
1222                 return (-1);
1223         switch (event) {
1224         case G_RAID_DISK_E_DISCONNECTED:
1225                 /* Delete disk. */
1226                 g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1227                 g_raid_destroy_disk(disk);
1228                 g_raid_md_promise_purge_volumes(sc);
1229
1230                 /* Write updated metadata to all disks. */
1231                 g_raid_md_write_promise(md, NULL, NULL, NULL);
1232
1233                 /* Check if anything left. */
1234                 if (g_raid_ndisks(sc, -1) == 0)
1235                         g_raid_destroy_node(sc, 0);
1236                 else
1237                         g_raid_md_promise_refill(sc);
1238                 return (0);
1239         }
1240         return (-2);
1241 }
1242
1243 static int
1244 g_raid_md_volume_event_promise(struct g_raid_md_object *md,
1245     struct g_raid_volume *vol, u_int event)
1246 {
1247         struct g_raid_md_promise_pervolume *pv;
1248
1249         pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1250         switch (event) {
1251         case G_RAID_VOLUME_E_STARTMD:
1252                 if (!pv->pv_started)
1253                         g_raid_md_promise_start(vol);
1254                 return (0);
1255         }
1256         return (-2);
1257 }
1258
1259 static int
1260 g_raid_md_ctl_promise(struct g_raid_md_object *md,
1261     struct gctl_req *req)
1262 {
1263         struct g_raid_softc *sc;
1264         struct g_raid_volume *vol, *vol1;
1265         struct g_raid_subdisk *sd;
1266         struct g_raid_disk *disk, *disks[PROMISE_MAX_DISKS];
1267         struct g_raid_md_promise_perdisk *pd;
1268         struct g_raid_md_promise_pervolume *pv;
1269         struct g_consumer *cp;
1270         struct g_provider *pp;
1271         char arg[16];
1272         const char *nodename, *verb, *volname, *levelname, *diskname;
1273         char *tmp;
1274         int *nargs, *force;
1275         off_t esize, offs[PROMISE_MAX_DISKS], size, sectorsize, strip;
1276         intmax_t *sizearg, *striparg;
1277         int numdisks, i, len, level, qual;
1278         int error;
1279
1280         sc = md->mdo_softc;
1281         verb = gctl_get_param(req, "verb", NULL);
1282         nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1283         error = 0;
1284         if (strcmp(verb, "label") == 0) {
1285                 if (*nargs < 4) {
1286                         gctl_error(req, "Invalid number of arguments.");
1287                         return (-1);
1288                 }
1289                 volname = gctl_get_asciiparam(req, "arg1");
1290                 if (volname == NULL) {
1291                         gctl_error(req, "No volume name.");
1292                         return (-2);
1293                 }
1294                 levelname = gctl_get_asciiparam(req, "arg2");
1295                 if (levelname == NULL) {
1296                         gctl_error(req, "No RAID level.");
1297                         return (-3);
1298                 }
1299                 if (strcasecmp(levelname, "RAID5") == 0)
1300                         levelname = "RAID5-LA";
1301                 if (g_raid_volume_str2level(levelname, &level, &qual)) {
1302                         gctl_error(req, "Unknown RAID level '%s'.", levelname);
1303                         return (-4);
1304                 }
1305                 numdisks = *nargs - 3;
1306                 force = gctl_get_paraml(req, "force", sizeof(*force));
1307                 if (!g_raid_md_promise_supported(level, qual, numdisks,
1308                     force ? *force : 0)) {
1309                         gctl_error(req, "Unsupported RAID level "
1310                             "(0x%02x/0x%02x), or number of disks (%d).",
1311                             level, qual, numdisks);
1312                         return (-5);
1313                 }
1314
1315                 /* Search for disks, connect them and probe. */
1316                 size = INT64_MAX;
1317                 sectorsize = 0;
1318                 bzero(disks, sizeof(disks));
1319                 bzero(offs, sizeof(offs));
1320                 for (i = 0; i < numdisks; i++) {
1321                         snprintf(arg, sizeof(arg), "arg%d", i + 3);
1322                         diskname = gctl_get_asciiparam(req, arg);
1323                         if (diskname == NULL) {
1324                                 gctl_error(req, "No disk name (%s).", arg);
1325                                 error = -6;
1326                                 break;
1327                         }
1328                         if (strcmp(diskname, "NONE") == 0)
1329                                 continue;
1330
1331                         TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1332                                 if (disk->d_consumer != NULL &&
1333                                     disk->d_consumer->provider != NULL &&
1334                                     strcmp(disk->d_consumer->provider->name,
1335                                      diskname) == 0)
1336                                         break;
1337                         }
1338                         if (disk != NULL) {
1339                                 if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
1340                                         gctl_error(req, "Disk '%s' is in a "
1341                                             "wrong state (%s).", diskname,
1342                                             g_raid_disk_state2str(disk->d_state));
1343                                         error = -7;
1344                                         break;
1345                                 }
1346                                 pd = disk->d_md_data;
1347                                 if (pd->pd_subdisks >= PROMISE_MAX_SUBDISKS) {
1348                                         gctl_error(req, "Disk '%s' already "
1349                                             "used by %d volumes.",
1350                                             diskname, pd->pd_subdisks);
1351                                         error = -7;
1352                                         break;
1353                                 }
1354                                 pp = disk->d_consumer->provider;
1355                                 disks[i] = disk;
1356                                 promise_meta_unused_range(pd->pd_meta,
1357                                     pd->pd_subdisks,
1358                                     pp->mediasize / pp->sectorsize,
1359                                     &offs[i], &esize);
1360                                 size = MIN(size, (off_t)esize * pp->sectorsize);
1361                                 sectorsize = MAX(sectorsize, pp->sectorsize);
1362                                 continue;
1363                         }
1364
1365                         g_topology_lock();
1366                         cp = g_raid_open_consumer(sc, diskname);
1367                         if (cp == NULL) {
1368                                 gctl_error(req, "Can't open disk '%s'.",
1369                                     diskname);
1370                                 g_topology_unlock();
1371                                 error = -8;
1372                                 break;
1373                         }
1374                         pp = cp->provider;
1375                         pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1376                         disk = g_raid_create_disk(sc);
1377                         disk->d_md_data = (void *)pd;
1378                         disk->d_consumer = cp;
1379                         disks[i] = disk;
1380                         cp->private = disk;
1381                         g_topology_unlock();
1382
1383                         g_raid_get_disk_info(disk);
1384
1385                         /* Reserve some space for metadata. */
1386                         size = MIN(size, pp->mediasize - 131072llu * pp->sectorsize);
1387                         sectorsize = MAX(sectorsize, pp->sectorsize);
1388                 }
1389                 if (error != 0) {
1390                         for (i = 0; i < numdisks; i++) {
1391                                 if (disks[i] != NULL &&
1392                                     disks[i]->d_state == G_RAID_DISK_S_NONE)
1393                                         g_raid_destroy_disk(disks[i]);
1394                         }
1395                         return (error);
1396                 }
1397
1398                 if (sectorsize <= 0) {
1399                         gctl_error(req, "Can't get sector size.");
1400                         return (-8);
1401                 }
1402
1403                 /* Handle size argument. */
1404                 len = sizeof(*sizearg);
1405                 sizearg = gctl_get_param(req, "size", &len);
1406                 if (sizearg != NULL && len == sizeof(*sizearg) &&
1407                     *sizearg > 0) {
1408                         if (*sizearg > size) {
1409                                 gctl_error(req, "Size too big %lld > %lld.",
1410                                     (long long)*sizearg, (long long)size);
1411                                 return (-9);
1412                         }
1413                         size = *sizearg;
1414                 }
1415
1416                 /* Handle strip argument. */
1417                 strip = 131072;
1418                 len = sizeof(*striparg);
1419                 striparg = gctl_get_param(req, "strip", &len);
1420                 if (striparg != NULL && len == sizeof(*striparg) &&
1421                     *striparg > 0) {
1422                         if (*striparg < sectorsize) {
1423                                 gctl_error(req, "Strip size too small.");
1424                                 return (-10);
1425                         }
1426                         if (*striparg % sectorsize != 0) {
1427                                 gctl_error(req, "Incorrect strip size.");
1428                                 return (-11);
1429                         }
1430                         strip = *striparg;
1431                 }
1432
1433                 /* Round size down to strip or sector. */
1434                 if (level == G_RAID_VOLUME_RL_RAID1 ||
1435                     level == G_RAID_VOLUME_RL_SINGLE ||
1436                     level == G_RAID_VOLUME_RL_CONCAT)
1437                         size -= (size % sectorsize);
1438                 else if (level == G_RAID_VOLUME_RL_RAID1E &&
1439                     (numdisks & 1) != 0)
1440                         size -= (size % (2 * strip));
1441                 else
1442                         size -= (size % strip);
1443                 if (size <= 0) {
1444                         gctl_error(req, "Size too small.");
1445                         return (-13);
1446                 }
1447
1448                 /* We have all we need, create things: volume, ... */
1449                 pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1450                 arc4rand(&pv->pv_id, sizeof(pv->pv_id), 0);
1451                 pv->pv_generation = 0;
1452                 pv->pv_started = 1;
1453                 vol = g_raid_create_volume(sc, volname, -1);
1454                 vol->v_md_data = pv;
1455                 vol->v_raid_level = level;
1456                 vol->v_raid_level_qualifier = qual;
1457                 vol->v_strip_size = strip;
1458                 vol->v_disks_count = numdisks;
1459                 if (level == G_RAID_VOLUME_RL_RAID0 ||
1460                     level == G_RAID_VOLUME_RL_CONCAT ||
1461                     level == G_RAID_VOLUME_RL_SINGLE)
1462                         vol->v_mediasize = size * numdisks;
1463                 else if (level == G_RAID_VOLUME_RL_RAID1)
1464                         vol->v_mediasize = size;
1465                 else if (level == G_RAID_VOLUME_RL_RAID3 ||
1466                     level == G_RAID_VOLUME_RL_RAID5)
1467                         vol->v_mediasize = size * (numdisks - 1);
1468                 else { /* RAID1E */
1469                         vol->v_mediasize = ((size * numdisks) / strip / 2) *
1470                             strip;
1471                 }
1472                 vol->v_sectorsize = sectorsize;
1473                 g_raid_start_volume(vol);
1474
1475                 /* , and subdisks. */
1476                 for (i = 0; i < numdisks; i++) {
1477                         disk = disks[i];
1478                         sd = &vol->v_subdisks[i];
1479                         sd->sd_disk = disk;
1480                         sd->sd_offset = (off_t)offs[i] * 512;
1481                         sd->sd_size = size;
1482                         if (disk == NULL)
1483                                 continue;
1484                         TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1485                         g_raid_change_disk_state(disk,
1486                             G_RAID_DISK_S_ACTIVE);
1487                         g_raid_change_subdisk_state(sd,
1488                             G_RAID_SUBDISK_S_ACTIVE);
1489                         g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1490                             G_RAID_EVENT_SUBDISK);
1491                 }
1492
1493                 /* Write metadata based on created entities. */
1494                 G_RAID_DEBUG1(0, sc, "Array started.");
1495                 g_raid_md_write_promise(md, vol, NULL, NULL);
1496
1497                 /* Pickup any STALE/SPARE disks to refill array if needed. */
1498                 g_raid_md_promise_refill(sc);
1499
1500                 g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1501                     G_RAID_EVENT_VOLUME);
1502                 return (0);
1503         }
1504         if (strcmp(verb, "add") == 0) {
1505                 gctl_error(req, "`add` command is not applicable, "
1506                     "use `label` instead.");
1507                 return (-99);
1508         }
1509         if (strcmp(verb, "delete") == 0) {
1510                 nodename = gctl_get_asciiparam(req, "arg0");
1511                 if (nodename != NULL && strcasecmp(sc->sc_name, nodename) != 0)
1512                         nodename = NULL;
1513
1514                 /* Full node destruction. */
1515                 if (*nargs == 1 && nodename != NULL) {
1516                         /* Check if some volume is still open. */
1517                         force = gctl_get_paraml(req, "force", sizeof(*force));
1518                         if (force != NULL && *force == 0 &&
1519                             g_raid_nopens(sc) != 0) {
1520                                 gctl_error(req, "Some volume is still open.");
1521                                 return (-4);
1522                         }
1523
1524                         TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1525                                 if (disk->d_consumer)
1526                                         promise_meta_erase(disk->d_consumer);
1527                         }
1528                         g_raid_destroy_node(sc, 0);
1529                         return (0);
1530                 }
1531
1532                 /* Destroy specified volume. If it was last - all node. */
1533                 if (*nargs > 2) {
1534                         gctl_error(req, "Invalid number of arguments.");
1535                         return (-1);
1536                 }
1537                 volname = gctl_get_asciiparam(req,
1538                     nodename != NULL ? "arg1" : "arg0");
1539                 if (volname == NULL) {
1540                         gctl_error(req, "No volume name.");
1541                         return (-2);
1542                 }
1543
1544                 /* Search for volume. */
1545                 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1546                         if (strcmp(vol->v_name, volname) == 0)
1547                                 break;
1548                         pp = vol->v_provider;
1549                         if (pp == NULL)
1550                                 continue;
1551                         if (strcmp(pp->name, volname) == 0)
1552                                 break;
1553                         if (strncmp(pp->name, "raid/", 5) == 0 &&
1554                             strcmp(pp->name + 5, volname) == 0)
1555                                 break;
1556                 }
1557                 if (vol == NULL) {
1558                         i = strtol(volname, &tmp, 10);
1559                         if (verb != volname && tmp[0] == 0) {
1560                                 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1561                                         if (vol->v_global_id == i)
1562                                                 break;
1563                                 }
1564                         }
1565                 }
1566                 if (vol == NULL) {
1567                         gctl_error(req, "Volume '%s' not found.", volname);
1568                         return (-3);
1569                 }
1570
1571                 /* Check if volume is still open. */
1572                 force = gctl_get_paraml(req, "force", sizeof(*force));
1573                 if (force != NULL && *force == 0 &&
1574                     vol->v_provider_open != 0) {
1575                         gctl_error(req, "Volume is still open.");
1576                         return (-4);
1577                 }
1578
1579                 /* Destroy volume and potentially node. */
1580                 i = 0;
1581                 TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next)
1582                         i++;
1583                 if (i >= 2) {
1584                         g_raid_destroy_volume(vol);
1585                         g_raid_md_promise_purge_disks(sc);
1586                         g_raid_md_write_promise(md, NULL, NULL, NULL);
1587                 } else {
1588                         TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1589                                 if (disk->d_consumer)
1590                                         promise_meta_erase(disk->d_consumer);
1591                         }
1592                         g_raid_destroy_node(sc, 0);
1593                 }
1594                 return (0);
1595         }
1596         if (strcmp(verb, "remove") == 0 ||
1597             strcmp(verb, "fail") == 0) {
1598                 if (*nargs < 2) {
1599                         gctl_error(req, "Invalid number of arguments.");
1600                         return (-1);
1601                 }
1602                 for (i = 1; i < *nargs; i++) {
1603                         snprintf(arg, sizeof(arg), "arg%d", i);
1604                         diskname = gctl_get_asciiparam(req, arg);
1605                         if (diskname == NULL) {
1606                                 gctl_error(req, "No disk name (%s).", arg);
1607                                 error = -2;
1608                                 break;
1609                         }
1610                         if (strncmp(diskname, _PATH_DEV, 5) == 0)
1611                                 diskname += 5;
1612
1613                         TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1614                                 if (disk->d_consumer != NULL &&
1615                                     disk->d_consumer->provider != NULL &&
1616                                     strcmp(disk->d_consumer->provider->name,
1617                                      diskname) == 0)
1618                                         break;
1619                         }
1620                         if (disk == NULL) {
1621                                 gctl_error(req, "Disk '%s' not found.",
1622                                     diskname);
1623                                 error = -3;
1624                                 break;
1625                         }
1626
1627                         if (strcmp(verb, "fail") == 0) {
1628                                 g_raid_md_fail_disk_promise(md, NULL, disk);
1629                                 continue;
1630                         }
1631
1632                         /* Erase metadata on deleting disk and destroy it. */
1633                         promise_meta_erase(disk->d_consumer);
1634                         g_raid_destroy_disk(disk);
1635                 }
1636                 g_raid_md_promise_purge_volumes(sc);
1637
1638                 /* Write updated metadata to remaining disks. */
1639                 g_raid_md_write_promise(md, NULL, NULL, NULL);
1640
1641                 /* Check if anything left. */
1642                 if (g_raid_ndisks(sc, -1) == 0)
1643                         g_raid_destroy_node(sc, 0);
1644                 else
1645                         g_raid_md_promise_refill(sc);
1646                 return (error);
1647         }
1648         if (strcmp(verb, "insert") == 0) {
1649                 if (*nargs < 2) {
1650                         gctl_error(req, "Invalid number of arguments.");
1651                         return (-1);
1652                 }
1653                 for (i = 1; i < *nargs; i++) {
1654                         /* Get disk name. */
1655                         snprintf(arg, sizeof(arg), "arg%d", i);
1656                         diskname = gctl_get_asciiparam(req, arg);
1657                         if (diskname == NULL) {
1658                                 gctl_error(req, "No disk name (%s).", arg);
1659                                 error = -3;
1660                                 break;
1661                         }
1662
1663                         /* Try to find provider with specified name. */
1664                         g_topology_lock();
1665                         cp = g_raid_open_consumer(sc, diskname);
1666                         if (cp == NULL) {
1667                                 gctl_error(req, "Can't open disk '%s'.",
1668                                     diskname);
1669                                 g_topology_unlock();
1670                                 error = -4;
1671                                 break;
1672                         }
1673                         pp = cp->provider;
1674                         g_topology_unlock();
1675
1676                         pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1677
1678                         disk = g_raid_create_disk(sc);
1679                         disk->d_consumer = cp;
1680                         disk->d_md_data = (void *)pd;
1681                         cp->private = disk;
1682
1683                         g_raid_get_disk_info(disk);
1684
1685                         /* Welcome the "new" disk. */
1686                         g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
1687                         promise_meta_write_spare(cp);
1688                         g_raid_md_promise_refill(sc);
1689                 }
1690                 return (error);
1691         }
1692         return (-100);
1693 }
1694
1695 static int
1696 g_raid_md_write_promise(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1697     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1698 {
1699         struct g_raid_softc *sc;
1700         struct g_raid_volume *vol;
1701         struct g_raid_subdisk *sd;
1702         struct g_raid_disk *disk;
1703         struct g_raid_md_promise_perdisk *pd;
1704         struct g_raid_md_promise_pervolume *pv;
1705         struct promise_raid_conf *meta;
1706         off_t rebuild_lba64;
1707         int i, j, pos, rebuild;
1708
1709         sc = md->mdo_softc;
1710
1711         if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1712                 return (0);
1713
1714         /* Generate new per-volume metadata for affected volumes. */
1715         TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1716                 if (vol->v_stopping)
1717                         continue;
1718
1719                 /* Skip volumes not related to specified targets. */
1720                 if (tvol != NULL && vol != tvol)
1721                         continue;
1722                 if (tsd != NULL && vol != tsd->sd_volume)
1723                         continue;
1724                 if (tdisk != NULL) {
1725                         for (i = 0; i < vol->v_disks_count; i++) {
1726                                 if (vol->v_subdisks[i].sd_disk == tdisk)
1727                                         break;
1728                         }
1729                         if (i >= vol->v_disks_count)
1730                                 continue;
1731                 }
1732
1733                 pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1734                 pv->pv_generation++;
1735
1736                 meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
1737                 if (pv->pv_meta != NULL)
1738                         memcpy(meta, pv->pv_meta, sizeof(*meta));
1739                 memcpy(meta->promise_id, PROMISE_MAGIC,
1740                     sizeof(PROMISE_MAGIC) - 1);
1741                 meta->dummy_0 = 0x00020000;
1742                 meta->integrity = PROMISE_I_VALID;
1743
1744                 meta->generation = pv->pv_generation;
1745                 meta->status = PROMISE_S_VALID | PROMISE_S_ONLINE |
1746                     PROMISE_S_INITED | PROMISE_S_READY;
1747                 if (vol->v_state <= G_RAID_VOLUME_S_DEGRADED)
1748                         meta->status |= PROMISE_S_DEGRADED;
1749                 if (vol->v_dirty)
1750                         meta->status |= PROMISE_S_MARKED; /* XXX: INVENTED! */
1751                 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0 ||
1752                     vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1753                         meta->type = PROMISE_T_RAID0;
1754                 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1755                     vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1756                         meta->type = PROMISE_T_RAID1;
1757                 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3)
1758                         meta->type = PROMISE_T_RAID3;
1759                 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1760                         meta->type = PROMISE_T_RAID5;
1761                 else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT)
1762                         meta->type = PROMISE_T_SPAN;
1763                 else
1764                         meta->type = PROMISE_T_JBOD;
1765                 meta->total_disks = vol->v_disks_count;
1766                 meta->stripe_shift = ffs(vol->v_strip_size / 1024);
1767                 meta->array_width = vol->v_disks_count;
1768                 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1769                     vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1770                         meta->array_width /= 2;
1771                 meta->array_number = vol->v_global_id;
1772                 meta->total_sectors = vol->v_mediasize / 512;
1773                 meta->total_sectors_high = (vol->v_mediasize / 512) >> 32;
1774                 meta->sector_size = vol->v_sectorsize / 512;
1775                 meta->cylinders = meta->total_sectors / (255 * 63) - 1;
1776                 meta->heads = 254;
1777                 meta->sectors = 63;
1778                 meta->volume_id = pv->pv_id;
1779                 rebuild_lba64 = UINT64_MAX;
1780                 rebuild = 0;
1781                 for (i = 0; i < vol->v_disks_count; i++) {
1782                         sd = &vol->v_subdisks[i];
1783                         /* For RAID0+1 we need to translate order. */
1784                         pos = promise_meta_translate_disk(vol, i);
1785                         meta->disks[pos].flags = PROMISE_F_VALID |
1786                             PROMISE_F_ASSIGNED;
1787                         if (sd->sd_state == G_RAID_SUBDISK_S_NONE) {
1788                                 meta->disks[pos].flags |= 0;
1789                         } else if (sd->sd_state == G_RAID_SUBDISK_S_FAILED) {
1790                                 meta->disks[pos].flags |=
1791                                     PROMISE_F_DOWN | PROMISE_F_REDIR;
1792                         } else if (sd->sd_state <= G_RAID_SUBDISK_S_REBUILD) {
1793                                 meta->disks[pos].flags |=
1794                                     PROMISE_F_ONLINE | PROMISE_F_REDIR;
1795                                 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1796                                         rebuild_lba64 = MIN(rebuild_lba64,
1797                                             sd->sd_rebuild_pos / 512);
1798                                 } else
1799                                         rebuild_lba64 = 0;
1800                                 rebuild = 1;
1801                         } else {
1802                                 meta->disks[pos].flags |= PROMISE_F_ONLINE;
1803                                 if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE) {
1804                                         meta->status |= PROMISE_S_MARKED;
1805                                         if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
1806                                                 rebuild_lba64 = MIN(rebuild_lba64,
1807                                                     sd->sd_rebuild_pos / 512);
1808                                         } else
1809                                                 rebuild_lba64 = 0;
1810                                 }
1811                         }
1812                         if (pv->pv_meta != NULL) {
1813                                 meta->disks[pos].id = pv->pv_meta->disks[pos].id;
1814                         } else {
1815                                 meta->disks[pos].number = i * 2;
1816                                 arc4rand(&meta->disks[pos].id,
1817                                     sizeof(meta->disks[pos].id), 0);
1818                         }
1819                 }
1820                 promise_meta_put_name(meta, vol->v_name);
1821
1822                 /* Try to mimic AMD BIOS rebuild/resync behavior. */
1823                 if (rebuild_lba64 != UINT64_MAX) {
1824                         if (rebuild)
1825                                 meta->magic_3 = 0x03040010UL; /* Rebuild? */
1826                         else
1827                                 meta->magic_3 = 0x03040008UL; /* Resync? */
1828                         /* Translate from per-disk to per-volume LBA. */
1829                         if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1830                             vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
1831                                 rebuild_lba64 *= meta->array_width;
1832                         } else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1833                             vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) {
1834                                 rebuild_lba64 *= meta->array_width - 1;
1835                         } else
1836                                 rebuild_lba64 = 0;
1837                 } else
1838                         meta->magic_3 = 0x03000000UL;
1839                 meta->rebuild_lba64 = rebuild_lba64;
1840                 meta->magic_4 = 0x04010101UL;
1841
1842                 /* Replace per-volume metadata with new. */
1843                 if (pv->pv_meta != NULL)
1844                         free(pv->pv_meta, M_MD_PROMISE);
1845                 pv->pv_meta = meta;
1846
1847                 /* Copy new metadata to the disks, adding or replacing old. */
1848                 for (i = 0; i < vol->v_disks_count; i++) {
1849                         sd = &vol->v_subdisks[i];
1850                         disk = sd->sd_disk;
1851                         if (disk == NULL)
1852                                 continue;
1853                         /* For RAID0+1 we need to translate order. */
1854                         pos = promise_meta_translate_disk(vol, i);
1855                         pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1856                         for (j = 0; j < pd->pd_subdisks; j++) {
1857                                 if (pd->pd_meta[j]->volume_id == meta->volume_id)
1858                                         break;
1859                         }
1860                         if (j == pd->pd_subdisks)
1861                                 pd->pd_subdisks++;
1862                         if (pd->pd_meta[j] != NULL)
1863                                 free(pd->pd_meta[j], M_MD_PROMISE);
1864                         pd->pd_meta[j] = promise_meta_copy(meta);
1865                         pd->pd_meta[j]->disk = meta->disks[pos];
1866                         pd->pd_meta[j]->disk.number = pos;
1867                         pd->pd_meta[j]->disk_offset_high =
1868                             (sd->sd_offset / 512) >> 32;
1869                         pd->pd_meta[j]->disk_offset = sd->sd_offset / 512;
1870                         pd->pd_meta[j]->disk_sectors_high =
1871                             (sd->sd_size / 512) >> 32;
1872                         pd->pd_meta[j]->disk_sectors = sd->sd_size / 512;
1873                         if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1874                                 pd->pd_meta[j]->disk_rebuild_high =
1875                                     (sd->sd_rebuild_pos / 512) >> 32;
1876                                 pd->pd_meta[j]->disk_rebuild =
1877                                     sd->sd_rebuild_pos / 512;
1878                         } else if (sd->sd_state < G_RAID_SUBDISK_S_REBUILD) {
1879                                 pd->pd_meta[j]->disk_rebuild_high = 0;
1880                                 pd->pd_meta[j]->disk_rebuild = 0;
1881                         } else {
1882                                 pd->pd_meta[j]->disk_rebuild_high = UINT32_MAX;
1883                                 pd->pd_meta[j]->disk_rebuild = UINT32_MAX;
1884                         }
1885                         pd->pd_updated = 1;
1886                 }
1887         }
1888
1889         TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1890                 pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1891                 if (disk->d_state != G_RAID_DISK_S_ACTIVE)
1892                         continue;
1893                 if (!pd->pd_updated)
1894                         continue;
1895                 G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1896                     g_raid_get_diskname(disk));
1897                 for (i = 0; i < pd->pd_subdisks; i++)
1898                         g_raid_md_promise_print(pd->pd_meta[i]);
1899                 promise_meta_write(disk->d_consumer,
1900                     pd->pd_meta, pd->pd_subdisks);
1901                 pd->pd_updated = 0;
1902         }
1903
1904         return (0);
1905 }
1906
1907 static int
1908 g_raid_md_fail_disk_promise(struct g_raid_md_object *md,
1909     struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1910 {
1911         struct g_raid_softc *sc;
1912         struct g_raid_md_promise_perdisk *pd;
1913         struct g_raid_subdisk *sd;
1914         int i, pos;
1915
1916         sc = md->mdo_softc;
1917         pd = (struct g_raid_md_promise_perdisk *)tdisk->d_md_data;
1918
1919         /* We can't fail disk that is not a part of array now. */
1920         if (tdisk->d_state != G_RAID_DISK_S_ACTIVE)
1921                 return (-1);
1922
1923         /*
1924          * Mark disk as failed in metadata and try to write that metadata
1925          * to the disk itself to prevent it's later resurrection as STALE.
1926          */
1927         if (pd->pd_subdisks > 0 && tdisk->d_consumer != NULL)
1928                 G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1929                     g_raid_get_diskname(tdisk));
1930         for (i = 0; i < pd->pd_subdisks; i++) {
1931                 pd->pd_meta[i]->disk.flags |=
1932                     PROMISE_F_DOWN | PROMISE_F_REDIR;
1933                 pos = pd->pd_meta[i]->disk.number;
1934                 if (pos >= 0 && pos < PROMISE_MAX_DISKS) {
1935                         pd->pd_meta[i]->disks[pos].flags |=
1936                             PROMISE_F_DOWN | PROMISE_F_REDIR;
1937                 }
1938                 g_raid_md_promise_print(pd->pd_meta[i]);
1939         }
1940         if (tdisk->d_consumer != NULL)
1941                 promise_meta_write(tdisk->d_consumer,
1942                     pd->pd_meta, pd->pd_subdisks);
1943
1944         /* Change states. */
1945         g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1946         TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1947                 g_raid_change_subdisk_state(sd,
1948                     G_RAID_SUBDISK_S_FAILED);
1949                 g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1950                     G_RAID_EVENT_SUBDISK);
1951         }
1952
1953         /* Write updated metadata to remaining disks. */
1954         g_raid_md_write_promise(md, NULL, NULL, tdisk);
1955
1956         g_raid_md_promise_refill(sc);
1957         return (0);
1958 }
1959
1960 static int
1961 g_raid_md_free_disk_promise(struct g_raid_md_object *md,
1962     struct g_raid_disk *disk)
1963 {
1964         struct g_raid_md_promise_perdisk *pd;
1965         int i;
1966
1967         pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1968         for (i = 0; i < pd->pd_subdisks; i++) {
1969                 if (pd->pd_meta[i] != NULL) {
1970                         free(pd->pd_meta[i], M_MD_PROMISE);
1971                         pd->pd_meta[i] = NULL;
1972                 }
1973         }
1974         free(pd, M_MD_PROMISE);
1975         disk->d_md_data = NULL;
1976         return (0);
1977 }
1978
1979 static int
1980 g_raid_md_free_volume_promise(struct g_raid_md_object *md,
1981     struct g_raid_volume *vol)
1982 {
1983         struct g_raid_md_promise_pervolume *pv;
1984
1985         pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1986         if (pv && pv->pv_meta != NULL) {
1987                 free(pv->pv_meta, M_MD_PROMISE);
1988                 pv->pv_meta = NULL;
1989         }
1990         if (pv && !pv->pv_started) {
1991                 pv->pv_started = 1;
1992                 callout_stop(&pv->pv_start_co);
1993         }
1994         free(pv, M_MD_PROMISE);
1995         vol->v_md_data = NULL;
1996         return (0);
1997 }
1998
1999 static int
2000 g_raid_md_free_promise(struct g_raid_md_object *md)
2001 {
2002
2003         return (0);
2004 }
2005
2006 G_RAID_MD_DECLARE(promise, "Promise");