sys/boot/zfs/zfsimpl.c

   1 /*-
   2  * Copyright (c) 2007 Doug Rabson
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  */
  26
  27 #include <sys/cdefs.h>
  28 __FBSDID("$FreeBSD$");
  29
  30 /*
  31  *      Stand-alone ZFS file reader.
  32  */
  33
  34 #include <sys/stat.h>
  35 #include <sys/stdint.h>
  36
  37 #include "zfsimpl.h"
  38 #include "zfssubr.c"
  39
  40
  41 struct zfsmount {
  42         const spa_t     *spa;
  43         objset_phys_t   objset;
  44         uint64_t        rootobj;
  45 };
  46
  47 /*
  48  * List of all vdevs, chained through v_alllink.
  49  */
  50 static vdev_list_t zfs_vdevs;
  51
  52  /*
  53  * List of ZFS features supported for read
  54  */
  55 static const char *features_for_read[] = {
  56         "org.illumos:lz4_compress",
  57         "com.delphix:hole_birth",
  58         "com.delphix:extensible_dataset",
  59         "com.delphix:embedded_data",
  60         "org.open-zfs:large_blocks",
  61         NULL
  62 };
  63
  64 /*
  65  * List of all pools, chained through spa_link.
  66  */
  67 static spa_list_t zfs_pools;
  68
  69 static const dnode_phys_t *dnode_cache_obj = 0;
  70 static uint64_t dnode_cache_bn;
  71 static char *dnode_cache_buf;
  72 static char *zap_scratch;
  73 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
  74
  75 #define TEMP_SIZE       (1024 * 1024)
  76
  77 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
  78 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
  79 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
  80
  81 static void
  82 zfs_init(void)
  83 {
  84         STAILQ_INIT(&zfs_vdevs);
  85         STAILQ_INIT(&zfs_pools);
  86
  87         zfs_temp_buf = malloc(TEMP_SIZE);
  88         zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
  89         zfs_temp_ptr = zfs_temp_buf;
  90         dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
  91         zap_scratch = malloc(SPA_MAXBLOCKSIZE);
  92
  93         zfs_init_crc();
  94 }
  95
  96 static void *
  97 zfs_alloc(size_t size)
  98 {
  99         char *ptr;
 100
 101         if (zfs_temp_ptr + size > zfs_temp_end) {
 102                 printf("ZFS: out of temporary buffer space\n");
 103                 for (;;) ;
 104         }
 105         ptr = zfs_temp_ptr;
 106         zfs_temp_ptr += size;
 107
 108         return (ptr);
 109 }
 110
 111 static void
 112 zfs_free(void *ptr, size_t size)
 113 {
 114
 115         zfs_temp_ptr -= size;
 116         if (zfs_temp_ptr != ptr) {
 117                 printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
 118                 for (;;) ;
 119         }
 120 }
 121
 122 static int
 123 xdr_int(const unsigned char **xdr, int *ip)
 124 {
 125         *ip = ((*xdr)[0] << 24)
 126                 | ((*xdr)[1] << 16)
 127                 | ((*xdr)[2] << 8)
 128                 | ((*xdr)[3] << 0);
 129         (*xdr) += 4;
 130         return (0);
 131 }
 132
 133 static int
 134 xdr_u_int(const unsigned char **xdr, u_int *ip)
 135 {
 136         *ip = ((*xdr)[0] << 24)
 137                 | ((*xdr)[1] << 16)
 138                 | ((*xdr)[2] << 8)
 139                 | ((*xdr)[3] << 0);
 140         (*xdr) += 4;
 141         return (0);
 142 }
 143
 144 static int
 145 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
 146 {
 147         u_int hi, lo;
 148
 149         xdr_u_int(xdr, &hi);
 150         xdr_u_int(xdr, &lo);
 151         *lp = (((uint64_t) hi) << 32) | lo;
 152         return (0);
 153 }
 154
 155 static int
 156 nvlist_find(const unsigned char *nvlist, const char *name, int type,
 157             int* elementsp, void *valuep)
 158 {
 159         const unsigned char *p, *pair;
 160         int junk;
 161         int encoded_size, decoded_size;
 162
 163         p = nvlist;
 164         xdr_int(&p, &junk);
 165         xdr_int(&p, &junk);
 166
 167         pair = p;
 168         xdr_int(&p, &encoded_size);
 169         xdr_int(&p, &decoded_size);
 170         while (encoded_size && decoded_size) {
 171                 int namelen, pairtype, elements;
 172                 const char *pairname;
 173
 174                 xdr_int(&p, &namelen);
 175                 pairname = (const char*) p;
 176                 p += roundup(namelen, 4);
 177                 xdr_int(&p, &pairtype);
 178
 179                 if (!memcmp(name, pairname, namelen) && type == pairtype) {
 180                         xdr_int(&p, &elements);
 181                         if (elementsp)
 182                                 *elementsp = elements;
 183                         if (type == DATA_TYPE_UINT64) {
 184                                 xdr_uint64_t(&p, (uint64_t *) valuep);
 185                                 return (0);
 186                         } else if (type == DATA_TYPE_STRING) {
 187                                 int len;
 188                                 xdr_int(&p, &len);
 189                                 (*(const char**) valuep) = (const char*) p;
 190                                 return (0);
 191                         } else if (type == DATA_TYPE_NVLIST
 192                                    || type == DATA_TYPE_NVLIST_ARRAY) {
 193                                 (*(const unsigned char**) valuep) =
 194                                          (const unsigned char*) p;
 195                                 return (0);
 196                         } else {
 197                                 return (EIO);
 198                         }
 199                 } else {
 200                         /*
 201                          * Not the pair we are looking for, skip to the next one.
 202                          */
 203                         p = pair + encoded_size;
 204                 }
 205
 206                 pair = p;
 207                 xdr_int(&p, &encoded_size);
 208                 xdr_int(&p, &decoded_size);
 209         }
 210
 211         return (EIO);
 212 }
 213
 214 static int
 215 nvlist_check_features_for_read(const unsigned char *nvlist)
 216 {
 217         const unsigned char *p, *pair;
 218         int junk;
 219         int encoded_size, decoded_size;
 220         int rc;
 221
 222         rc = 0;
 223
 224         p = nvlist;
 225         xdr_int(&p, &junk);
 226         xdr_int(&p, &junk);
 227
 228         pair = p;
 229         xdr_int(&p, &encoded_size);
 230         xdr_int(&p, &decoded_size);
 231         while (encoded_size && decoded_size) {
 232                 int namelen, pairtype;
 233                 const char *pairname;
 234                 int i, found;
 235
 236                 found = 0;
 237
 238                 xdr_int(&p, &namelen);
 239                 pairname = (const char*) p;
 240                 p += roundup(namelen, 4);
 241                 xdr_int(&p, &pairtype);
 242
 243                 for (i = 0; features_for_read[i] != NULL; i++) {
 244                         if (!memcmp(pairname, features_for_read[i], namelen)) {
 245                                 found = 1;
 246                                 break;
 247                         }
 248                 }
 249
 250                 if (!found) {
 251                         printf("ZFS: unsupported feature: %s\n", pairname);
 252                         rc = EIO;
 253                 }
 254
 255                 p = pair + encoded_size;
 256
 257                 pair = p;
 258                 xdr_int(&p, &encoded_size);
 259                 xdr_int(&p, &decoded_size);
 260         }
 261
 262         return (rc);
 263 }
 264
 265 /*
 266  * Return the next nvlist in an nvlist array.
 267  */
 268 static const unsigned char *
 269 nvlist_next(const unsigned char *nvlist)
 270 {
 271         const unsigned char *p, *pair;
 272         int junk;
 273         int encoded_size, decoded_size;
 274
 275         p = nvlist;
 276         xdr_int(&p, &junk);
 277         xdr_int(&p, &junk);
 278
 279         pair = p;
 280         xdr_int(&p, &encoded_size);
 281         xdr_int(&p, &decoded_size);
 282         while (encoded_size && decoded_size) {
 283                 p = pair + encoded_size;
 284
 285                 pair = p;
 286                 xdr_int(&p, &encoded_size);
 287                 xdr_int(&p, &decoded_size);
 288         }
 289
 290         return p;
 291 }
 292
 293 #ifdef TEST
 294
 295 static const unsigned char *
 296 nvlist_print(const unsigned char *nvlist, unsigned int indent)
 297 {
 298         static const char* typenames[] = {
 299                 "DATA_TYPE_UNKNOWN",
 300                 "DATA_TYPE_BOOLEAN",
 301                 "DATA_TYPE_BYTE",
 302                 "DATA_TYPE_INT16",
 303                 "DATA_TYPE_UINT16",
 304                 "DATA_TYPE_INT32",
 305                 "DATA_TYPE_UINT32",
 306                 "DATA_TYPE_INT64",
 307                 "DATA_TYPE_UINT64",
 308                 "DATA_TYPE_STRING",
 309                 "DATA_TYPE_BYTE_ARRAY",
 310                 "DATA_TYPE_INT16_ARRAY",
 311                 "DATA_TYPE_UINT16_ARRAY",
 312                 "DATA_TYPE_INT32_ARRAY",
 313                 "DATA_TYPE_UINT32_ARRAY",
 314                 "DATA_TYPE_INT64_ARRAY",
 315                 "DATA_TYPE_UINT64_ARRAY",
 316                 "DATA_TYPE_STRING_ARRAY",
 317                 "DATA_TYPE_HRTIME",
 318                 "DATA_TYPE_NVLIST",
 319                 "DATA_TYPE_NVLIST_ARRAY",
 320                 "DATA_TYPE_BOOLEAN_VALUE",
 321                 "DATA_TYPE_INT8",
 322                 "DATA_TYPE_UINT8",
 323                 "DATA_TYPE_BOOLEAN_ARRAY",
 324                 "DATA_TYPE_INT8_ARRAY",
 325                 "DATA_TYPE_UINT8_ARRAY"
 326         };
 327
 328         unsigned int i, j;
 329         const unsigned char *p, *pair;
 330         int junk;
 331         int encoded_size, decoded_size;
 332
 333         p = nvlist;
 334         xdr_int(&p, &junk);
 335         xdr_int(&p, &junk);
 336
 337         pair = p;
 338         xdr_int(&p, &encoded_size);
 339         xdr_int(&p, &decoded_size);
 340         while (encoded_size && decoded_size) {
 341                 int namelen, pairtype, elements;
 342                 const char *pairname;
 343
 344                 xdr_int(&p, &namelen);
 345                 pairname = (const char*) p;
 346                 p += roundup(namelen, 4);
 347                 xdr_int(&p, &pairtype);
 348
 349                 for (i = 0; i < indent; i++)
 350                         printf(" ");
 351                 printf("%s %s", typenames[pairtype], pairname);
 352
 353                 xdr_int(&p, &elements);
 354                 switch (pairtype) {
 355                 case DATA_TYPE_UINT64: {
 356                         uint64_t val;
 357                         xdr_uint64_t(&p, &val);
 358                         printf(" = 0x%jx\n", (uintmax_t)val);
 359                         break;
 360                 }
 361
 362                 case DATA_TYPE_STRING: {
 363                         int len;
 364                         xdr_int(&p, &len);
 365                         printf(" = \"%s\"\n", p);
 366                         break;
 367                 }
 368
 369                 case DATA_TYPE_NVLIST:
 370                         printf("\n");
 371                         nvlist_print(p, indent + 1);
 372                         break;
 373
 374                 case DATA_TYPE_NVLIST_ARRAY:
 375                         for (j = 0; j < elements; j++) {
 376                                 printf("[%d]\n", j);
 377                                 p = nvlist_print(p, indent + 1);
 378                                 if (j != elements - 1) {
 379                                         for (i = 0; i < indent; i++)
 380                                                 printf(" ");
 381                                         printf("%s %s", typenames[pairtype], pairname);
 382                                 }
 383                         }
 384                         break;
 385
 386                 default:
 387                         printf("\n");
 388                 }
 389
 390                 p = pair + encoded_size;
 391
 392                 pair = p;
 393                 xdr_int(&p, &encoded_size);
 394                 xdr_int(&p, &decoded_size);
 395         }
 396
 397         return p;
 398 }
 399
 400 #endif
 401
 402 static int
 403 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
 404     off_t offset, size_t size)
 405 {
 406         size_t psize;
 407         int rc;
 408
 409         if (!vdev->v_phys_read)
 410                 return (EIO);
 411
 412         if (bp) {
 413                 psize = BP_GET_PSIZE(bp);
 414         } else {
 415                 psize = size;
 416         }
 417
 418         /*printf("ZFS: reading %d bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
 419         rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
 420         if (rc)
 421                 return (rc);
 422         if (bp && zio_checksum_verify(bp, buf))
 423                 return (EIO);
 424
 425         return (0);
 426 }
 427
 428 static int
 429 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 430     off_t offset, size_t bytes)
 431 {
 432
 433         return (vdev_read_phys(vdev, bp, buf,
 434                 offset + VDEV_LABEL_START_SIZE, bytes));
 435 }
 436
 437
 438 static int
 439 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 440     off_t offset, size_t bytes)
 441 {
 442         vdev_t *kid;
 443         int rc;
 444
 445         rc = EIO;
 446         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 447                 if (kid->v_state != VDEV_STATE_HEALTHY)
 448                         continue;
 449                 rc = kid->v_read(kid, bp, buf, offset, bytes);
 450                 if (!rc)
 451                         return (0);
 452         }
 453
 454         return (rc);
 455 }
 456
 457 static int
 458 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 459     off_t offset, size_t bytes)
 460 {
 461         vdev_t *kid;
 462
 463         /*
 464          * Here we should have two kids:
 465          * First one which is the one we are replacing and we can trust
 466          * only this one to have valid data, but it might not be present.
 467          * Second one is that one we are replacing with. It is most likely
 468          * healthy, but we can't trust it has needed data, so we won't use it.
 469          */
 470         kid = STAILQ_FIRST(&vdev->v_children);
 471         if (kid == NULL)
 472                 return (EIO);
 473         if (kid->v_state != VDEV_STATE_HEALTHY)
 474                 return (EIO);
 475         return (kid->v_read(kid, bp, buf, offset, bytes));
 476 }
 477
 478 static vdev_t *
 479 vdev_find(uint64_t guid)
 480 {
 481         vdev_t *vdev;
 482
 483         STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
 484                 if (vdev->v_guid == guid)
 485                         return (vdev);
 486
 487         return (0);
 488 }
 489
 490 static vdev_t *
 491 vdev_create(uint64_t guid, vdev_read_t *_read)
 492 {
 493         vdev_t *vdev;
 494
 495         vdev = malloc(sizeof(vdev_t));
 496         memset(vdev, 0, sizeof(vdev_t));
 497         STAILQ_INIT(&vdev->v_children);
 498         vdev->v_guid = guid;
 499         vdev->v_state = VDEV_STATE_OFFLINE;
 500         vdev->v_read = _read;
 501         vdev->v_phys_read = 0;
 502         vdev->v_read_priv = 0;
 503         STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
 504
 505         return (vdev);
 506 }
 507
 508 static int
 509 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
 510     vdev_t **vdevp, int is_newer)
 511 {
 512         int rc;
 513         uint64_t guid, id, ashift, nparity;
 514         const char *type;
 515         const char *path;
 516         vdev_t *vdev, *kid;
 517         const unsigned char *kids;
 518         int nkids, i, is_new;
 519         uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
 520
 521         if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID,
 522                         DATA_TYPE_UINT64, 0, &guid)
 523             || nvlist_find(nvlist, ZPOOL_CONFIG_ID,
 524                            DATA_TYPE_UINT64, 0, &id)
 525             || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE,
 526                            DATA_TYPE_STRING, 0, &type)) {
 527                 printf("ZFS: can't find vdev details\n");
 528                 return (ENOENT);
 529         }
 530
 531         if (strcmp(type, VDEV_TYPE_MIRROR)
 532             && strcmp(type, VDEV_TYPE_DISK)
 533 #ifdef ZFS_TEST
 534             && strcmp(type, VDEV_TYPE_FILE)
 535 #endif
 536             && strcmp(type, VDEV_TYPE_RAIDZ)
 537             && strcmp(type, VDEV_TYPE_REPLACING)) {
 538                 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 539                 return (EIO);
 540         }
 541
 542         is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
 543
 544         nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, 0,
 545                         &is_offline);
 546         nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, 0,
 547                         &is_removed);
 548         nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, 0,
 549                         &is_faulted);
 550         nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, 0,
 551                         &is_degraded);
 552         nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, 0,
 553                         &isnt_present);
 554
 555         vdev = vdev_find(guid);
 556         if (!vdev) {
 557                 is_new = 1;
 558
 559                 if (!strcmp(type, VDEV_TYPE_MIRROR))
 560                         vdev = vdev_create(guid, vdev_mirror_read);
 561                 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
 562                         vdev = vdev_create(guid, vdev_raidz_read);
 563                 else if (!strcmp(type, VDEV_TYPE_REPLACING))
 564                         vdev = vdev_create(guid, vdev_replacing_read);
 565                 else
 566                         vdev = vdev_create(guid, vdev_disk_read);
 567
 568                 vdev->v_id = id;
 569                 vdev->v_top = pvdev != NULL ? pvdev : vdev;
 570                 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
 571                         DATA_TYPE_UINT64, 0, &ashift) == 0)
 572                         vdev->v_ashift = ashift;
 573                 else
 574                         vdev->v_ashift = 0;
 575                 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
 576                         DATA_TYPE_UINT64, 0, &nparity) == 0)
 577                         vdev->v_nparity = nparity;
 578                 else
 579                         vdev->v_nparity = 0;
 580                 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
 581                                 DATA_TYPE_STRING, 0, &path) == 0) {
 582                         if (strncmp(path, "/dev/", 5) == 0)
 583                                 path += 5;
 584                         vdev->v_name = strdup(path);
 585                 } else {
 586                         if (!strcmp(type, "raidz")) {
 587                                 if (vdev->v_nparity == 1)
 588                                         vdev->v_name = "raidz1";
 589                                 else if (vdev->v_nparity == 2)
 590                                         vdev->v_name = "raidz2";
 591                                 else if (vdev->v_nparity == 3)
 592                                         vdev->v_name = "raidz3";
 593                                 else {
 594                                         printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 595                                         return (EIO);
 596                                 }
 597                         } else {
 598                                 vdev->v_name = strdup(type);
 599                         }
 600                 }
 601         } else {
 602                 is_new = 0;
 603         }
 604
 605         if (is_new || is_newer) {
 606                 /*
 607                  * This is either new vdev or we've already seen this vdev,
 608                  * but from an older vdev label, so let's refresh its state
 609                  * from the newer label.
 610                  */
 611                 if (is_offline)
 612                         vdev->v_state = VDEV_STATE_OFFLINE;
 613                 else if (is_removed)
 614                         vdev->v_state = VDEV_STATE_REMOVED;
 615                 else if (is_faulted)
 616                         vdev->v_state = VDEV_STATE_FAULTED;
 617                 else if (is_degraded)
 618                         vdev->v_state = VDEV_STATE_DEGRADED;
 619                 else if (isnt_present)
 620                         vdev->v_state = VDEV_STATE_CANT_OPEN;
 621         }
 622
 623         rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN,
 624                          DATA_TYPE_NVLIST_ARRAY, &nkids, &kids);
 625         /*
 626          * Its ok if we don't have any kids.
 627          */
 628         if (rc == 0) {
 629                 vdev->v_nchildren = nkids;
 630                 for (i = 0; i < nkids; i++) {
 631                         rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
 632                         if (rc)
 633                                 return (rc);
 634                         if (is_new)
 635                                 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
 636                                                    v_childlink);
 637                         kids = nvlist_next(kids);
 638                 }
 639         } else {
 640                 vdev->v_nchildren = 0;
 641         }
 642
 643         if (vdevp)
 644                 *vdevp = vdev;
 645         return (0);
 646 }
 647
 648 static void
 649 vdev_set_state(vdev_t *vdev)
 650 {
 651         vdev_t *kid;
 652         int good_kids;
 653         int bad_kids;
 654
 655         /*
 656          * A mirror or raidz is healthy if all its kids are healthy. A
 657          * mirror is degraded if any of its kids is healthy; a raidz
 658          * is degraded if at most nparity kids are offline.
 659          */
 660         if (STAILQ_FIRST(&vdev->v_children)) {
 661                 good_kids = 0;
 662                 bad_kids = 0;
 663                 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 664                         if (kid->v_state == VDEV_STATE_HEALTHY)
 665                                 good_kids++;
 666                         else
 667                                 bad_kids++;
 668                 }
 669                 if (bad_kids == 0) {
 670                         vdev->v_state = VDEV_STATE_HEALTHY;
 671                 } else {
 672                         if (vdev->v_read == vdev_mirror_read) {
 673                                 if (good_kids) {
 674                                         vdev->v_state = VDEV_STATE_DEGRADED;
 675                                 } else {
 676                                         vdev->v_state = VDEV_STATE_OFFLINE;
 677                                 }
 678                         } else if (vdev->v_read == vdev_raidz_read) {
 679                                 if (bad_kids > vdev->v_nparity) {
 680                                         vdev->v_state = VDEV_STATE_OFFLINE;
 681                                 } else {
 682                                         vdev->v_state = VDEV_STATE_DEGRADED;
 683                                 }
 684                         }
 685                 }
 686         }
 687 }
 688
 689 static spa_t *
 690 spa_find_by_guid(uint64_t guid)
 691 {
 692         spa_t *spa;
 693
 694         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 695                 if (spa->spa_guid == guid)
 696                         return (spa);
 697
 698         return (0);
 699 }
 700
 701 static spa_t *
 702 spa_find_by_name(const char *name)
 703 {
 704         spa_t *spa;
 705
 706         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 707                 if (!strcmp(spa->spa_name, name))
 708                         return (spa);
 709
 710         return (0);
 711 }
 712
 713 #ifdef BOOT2
 714 static spa_t *
 715 spa_get_primary(void)
 716 {
 717
 718         return (STAILQ_FIRST(&zfs_pools));
 719 }
 720
 721 static vdev_t *
 722 spa_get_primary_vdev(const spa_t *spa)
 723 {
 724         vdev_t *vdev;
 725         vdev_t *kid;
 726
 727         if (spa == NULL)
 728                 spa = spa_get_primary();
 729         if (spa == NULL)
 730                 return (NULL);
 731         vdev = STAILQ_FIRST(&spa->spa_vdevs);
 732         if (vdev == NULL)
 733                 return (NULL);
 734         for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
 735              kid = STAILQ_FIRST(&vdev->v_children))
 736                 vdev = kid;
 737         return (vdev);
 738 }
 739 #endif
 740
 741 static spa_t *
 742 spa_create(uint64_t guid)
 743 {
 744         spa_t *spa;
 745
 746         spa = malloc(sizeof(spa_t));
 747         memset(spa, 0, sizeof(spa_t));
 748         STAILQ_INIT(&spa->spa_vdevs);
 749         spa->spa_guid = guid;
 750         STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
 751
 752         return (spa);
 753 }
 754
 755 static const char *
 756 state_name(vdev_state_t state)
 757 {
 758         static const char* names[] = {
 759                 "UNKNOWN",
 760                 "CLOSED",
 761                 "OFFLINE",
 762                 "REMOVED",
 763                 "CANT_OPEN",
 764                 "FAULTED",
 765                 "DEGRADED",
 766                 "ONLINE"
 767         };
 768         return names[state];
 769 }
 770
 771 #ifdef BOOT2
 772
 773 #define pager_printf printf
 774
 775 #else
 776
 777 static void
 778 pager_printf(const char *fmt, ...)
 779 {
 780         char line[80];
 781         va_list args;
 782
 783         va_start(args, fmt);
 784         vsprintf(line, fmt, args);
 785         va_end(args);
 786         pager_output(line);
 787 }
 788
 789 #endif
 790
 791 #define STATUS_FORMAT   "        %s %s\n"
 792
 793 static void
 794 print_state(int indent, const char *name, vdev_state_t state)
 795 {
 796         int i;
 797         char buf[512];
 798
 799         buf[0] = 0;
 800         for (i = 0; i < indent; i++)
 801                 strcat(buf, "  ");
 802         strcat(buf, name);
 803         pager_printf(STATUS_FORMAT, buf, state_name(state));
 804
 805 }
 806
 807 static void
 808 vdev_status(vdev_t *vdev, int indent)
 809 {
 810         vdev_t *kid;
 811         print_state(indent, vdev->v_name, vdev->v_state);
 812
 813         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 814                 vdev_status(kid, indent + 1);
 815         }
 816 }
 817
 818 static void
 819 spa_status(spa_t *spa)
 820 {
 821         static char bootfs[ZFS_MAXNAMELEN];
 822         uint64_t rootid;
 823         vdev_t *vdev;
 824         int good_kids, bad_kids, degraded_kids;
 825         vdev_state_t state;
 826
 827         pager_printf("  pool: %s\n", spa->spa_name);
 828         if (zfs_get_root(spa, &rootid) == 0 &&
 829             zfs_rlookup(spa, rootid, bootfs) == 0) {
 830                 if (bootfs[0] == '\0')
 831                         pager_printf("bootfs: %s\n", spa->spa_name);
 832                 else
 833                         pager_printf("bootfs: %s/%s\n", spa->spa_name, bootfs);
 834         }
 835         pager_printf("config:\n\n");
 836         pager_printf(STATUS_FORMAT, "NAME", "STATE");
 837
 838         good_kids = 0;
 839         degraded_kids = 0;
 840         bad_kids = 0;
 841         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 842                 if (vdev->v_state == VDEV_STATE_HEALTHY)
 843                         good_kids++;
 844                 else if (vdev->v_state == VDEV_STATE_DEGRADED)
 845                         degraded_kids++;
 846                 else
 847                         bad_kids++;
 848         }
 849
 850         state = VDEV_STATE_CLOSED;
 851         if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
 852                 state = VDEV_STATE_HEALTHY;
 853         else if ((good_kids + degraded_kids) > 0)
 854                 state = VDEV_STATE_DEGRADED;
 855
 856         print_state(0, spa->spa_name, state);
 857         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 858                 vdev_status(vdev, 1);
 859         }
 860 }
 861
 862 static void
 863 spa_all_status(void)
 864 {
 865         spa_t *spa;
 866         int first = 1;
 867
 868         STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
 869                 if (!first)
 870                         pager_printf("\n");
 871                 first = 0;
 872                 spa_status(spa);
 873         }
 874 }
 875
 876 static int
 877 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
 878 {
 879         vdev_t vtmp;
 880         vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
 881         spa_t *spa;
 882         vdev_t *vdev, *top_vdev, *pool_vdev;
 883         off_t off;
 884         blkptr_t bp;
 885         const unsigned char *nvlist;
 886         uint64_t val;
 887         uint64_t guid;
 888         uint64_t pool_txg, pool_guid;
 889         uint64_t is_log;
 890         const char *pool_name;
 891         const unsigned char *vdevs;
 892         const unsigned char *features;
 893         int i, rc, is_newer;
 894         char *upbuf;
 895         const struct uberblock *up;
 896
 897         /*
 898          * Load the vdev label and figure out which
 899          * uberblock is most current.
 900          */
 901         memset(&vtmp, 0, sizeof(vtmp));
 902         vtmp.v_phys_read = _read;
 903         vtmp.v_read_priv = read_priv;
 904         off = offsetof(vdev_label_t, vl_vdev_phys);
 905         BP_ZERO(&bp);
 906         BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
 907         BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
 908         BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
 909         BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
 910         DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
 911         ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
 912         if (vdev_read_phys(&vtmp, &bp, vdev_label, off, 0))
 913                 return (EIO);
 914
 915         if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR) {
 916                 return (EIO);
 917         }
 918
 919         nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
 920
 921         if (nvlist_find(nvlist,
 922                         ZPOOL_CONFIG_VERSION,
 923                         DATA_TYPE_UINT64, 0, &val)) {
 924                 return (EIO);
 925         }
 926
 927         if (!SPA_VERSION_IS_SUPPORTED(val)) {
 928                 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
 929                     (unsigned) val, (unsigned) SPA_VERSION);
 930                 return (EIO);
 931         }
 932
 933         /* Check ZFS features for read */
 934         if (nvlist_find(nvlist,
 935                         ZPOOL_CONFIG_FEATURES_FOR_READ,
 936                         DATA_TYPE_NVLIST, 0, &features) == 0
 937             && nvlist_check_features_for_read(features) != 0)
 938                 return (EIO);
 939
 940         if (nvlist_find(nvlist,
 941                         ZPOOL_CONFIG_POOL_STATE,
 942                         DATA_TYPE_UINT64, 0, &val)) {
 943                 return (EIO);
 944         }
 945
 946         if (val == POOL_STATE_DESTROYED) {
 947                 /* We don't boot only from destroyed pools. */
 948                 return (EIO);
 949         }
 950
 951         if (nvlist_find(nvlist,
 952                         ZPOOL_CONFIG_POOL_TXG,
 953                         DATA_TYPE_UINT64, 0, &pool_txg)
 954             || nvlist_find(nvlist,
 955                            ZPOOL_CONFIG_POOL_GUID,
 956                            DATA_TYPE_UINT64, 0, &pool_guid)
 957             || nvlist_find(nvlist,
 958                            ZPOOL_CONFIG_POOL_NAME,
 959                            DATA_TYPE_STRING, 0, &pool_name)) {
 960                 /*
 961                  * Cache and spare devices end up here - just ignore
 962                  * them.
 963                  */
 964                 /*printf("ZFS: can't find pool details\n");*/
 965                 return (EIO);
 966         }
 967
 968         is_log = 0;
 969         (void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, 0,
 970             &is_log);
 971         if (is_log)
 972                 return (EIO);
 973
 974         /*
 975          * Create the pool if this is the first time we've seen it.
 976          */
 977         spa = spa_find_by_guid(pool_guid);
 978         if (!spa) {
 979                 spa = spa_create(pool_guid);
 980                 spa->spa_name = strdup(pool_name);
 981         }
 982         if (pool_txg > spa->spa_txg) {
 983                 spa->spa_txg = pool_txg;
 984                 is_newer = 1;
 985         } else
 986                 is_newer = 0;
 987
 988         /*
 989          * Get the vdev tree and create our in-core copy of it.
 990          * If we already have a vdev with this guid, this must
 991          * be some kind of alias (overlapping slices, dangerously dedicated
 992          * disks etc).
 993          */
 994         if (nvlist_find(nvlist,
 995                         ZPOOL_CONFIG_GUID,
 996                         DATA_TYPE_UINT64, 0, &guid)) {
 997                 return (EIO);
 998         }
 999         vdev = vdev_find(guid);
1000         if (vdev && vdev->v_phys_read)  /* Has this vdev already been inited? */
1001                 return (EIO);
1002
1003         if (nvlist_find(nvlist,
1004                         ZPOOL_CONFIG_VDEV_TREE,
1005                         DATA_TYPE_NVLIST, 0, &vdevs)) {
1006                 return (EIO);
1007         }
1008
1009         rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1010         if (rc)
1011                 return (rc);
1012
1013         /*
1014          * Add the toplevel vdev to the pool if its not already there.
1015          */
1016         STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1017                 if (top_vdev == pool_vdev)
1018                         break;
1019         if (!pool_vdev && top_vdev)
1020                 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1021
1022         /*
1023          * We should already have created an incomplete vdev for this
1024          * vdev. Find it and initialise it with our read proc.
1025          */
1026         vdev = vdev_find(guid);
1027         if (vdev) {
1028                 vdev->v_phys_read = _read;
1029                 vdev->v_read_priv = read_priv;
1030                 vdev->v_state = VDEV_STATE_HEALTHY;
1031         } else {
1032                 printf("ZFS: inconsistent nvlist contents\n");
1033                 return (EIO);
1034         }
1035
1036         /*
1037          * Re-evaluate top-level vdev state.
1038          */
1039         vdev_set_state(top_vdev);
1040
1041         /*
1042          * Ok, we are happy with the pool so far. Lets find
1043          * the best uberblock and then we can actually access
1044          * the contents of the pool.
1045          */
1046         upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
1047         up = (const struct uberblock *)upbuf;
1048         for (i = 0;
1049              i < VDEV_UBERBLOCK_COUNT(vdev);
1050              i++) {
1051                 off = VDEV_UBERBLOCK_OFFSET(vdev, i);
1052                 BP_ZERO(&bp);
1053                 DVA_SET_OFFSET(&bp.blk_dva[0], off);
1054                 BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1055                 BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1056                 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1057                 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1058                 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1059
1060                 if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
1061                         continue;
1062
1063                 if (up->ub_magic != UBERBLOCK_MAGIC)
1064                         continue;
1065                 if (up->ub_txg < spa->spa_txg)
1066                         continue;
1067                 if (up->ub_txg > spa->spa_uberblock.ub_txg) {
1068                         spa->spa_uberblock = *up;
1069                 } else if (up->ub_txg == spa->spa_uberblock.ub_txg) {
1070                         if (up->ub_timestamp > spa->spa_uberblock.ub_timestamp)
1071                                 spa->spa_uberblock = *up;
1072                 }
1073         }
1074         zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
1075
1076         if (spap)
1077                 *spap = spa;
1078         return (0);
1079 }
1080
1081 static int
1082 ilog2(int n)
1083 {
1084         int v;
1085
1086         for (v = 0; v < 32; v++)
1087                 if (n == (1 << v))
1088                         return v;
1089         return -1;
1090 }
1091
1092 static int
1093 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1094 {
1095         blkptr_t gbh_bp;
1096         zio_gbh_phys_t zio_gb;
1097         char *pbuf;
1098         int i;
1099
1100         /* Artificial BP for gang block header. */
1101         gbh_bp = *bp;
1102         BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1103         BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1104         BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1105         BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1106         for (i = 0; i < SPA_DVAS_PER_BP; i++)
1107                 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1108
1109         /* Read gang header block using the artificial BP. */
1110         if (zio_read(spa, &gbh_bp, &zio_gb))
1111                 return (EIO);
1112
1113         pbuf = buf;
1114         for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1115                 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1116
1117                 if (BP_IS_HOLE(gbp))
1118                         continue;
1119                 if (zio_read(spa, gbp, pbuf))
1120                         return (EIO);
1121                 pbuf += BP_GET_PSIZE(gbp);
1122         }
1123
1124         if (zio_checksum_verify(bp, buf))
1125                 return (EIO);
1126         return (0);
1127 }
1128
1129 static int
1130 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1131 {
1132         int cpfunc = BP_GET_COMPRESS(bp);
1133         uint64_t align, size;
1134         void *pbuf;
1135         int i, error;
1136
1137         /*
1138          * Process data embedded in block pointer
1139          */
1140         if (BP_IS_EMBEDDED(bp)) {
1141                 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1142
1143                 size = BPE_GET_PSIZE(bp);
1144                 ASSERT(size <= BPE_PAYLOAD_SIZE);
1145
1146                 if (cpfunc != ZIO_COMPRESS_OFF)
1147                         pbuf = zfs_alloc(size);
1148                 else
1149                         pbuf = buf;
1150
1151                 decode_embedded_bp_compressed(bp, pbuf);
1152                 error = 0;
1153
1154                 if (cpfunc != ZIO_COMPRESS_OFF) {
1155                         error = zio_decompress_data(cpfunc, pbuf,
1156                             size, buf, BP_GET_LSIZE(bp));
1157                         zfs_free(pbuf, size);
1158                 }
1159                 if (error != 0)
1160                         printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
1161                             error);
1162                 return (error);
1163         }
1164
1165         error = EIO;
1166
1167         for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1168                 const dva_t *dva = &bp->blk_dva[i];
1169                 vdev_t *vdev;
1170                 int vdevid;
1171                 off_t offset;
1172
1173                 if (!dva->dva_word[0] && !dva->dva_word[1])
1174                         continue;
1175
1176                 vdevid = DVA_GET_VDEV(dva);
1177                 offset = DVA_GET_OFFSET(dva);
1178                 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1179                         if (vdev->v_id == vdevid)
1180                                 break;
1181                 }
1182                 if (!vdev || !vdev->v_read)
1183                         continue;
1184
1185                 size = BP_GET_PSIZE(bp);
1186                 if (vdev->v_read == vdev_raidz_read) {
1187                         align = 1ULL << vdev->v_top->v_ashift;
1188                         if (P2PHASE(size, align) != 0)
1189                                 size = P2ROUNDUP(size, align);
1190                 }
1191                 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1192                         pbuf = zfs_alloc(size);
1193                 else
1194                         pbuf = buf;
1195
1196                 if (DVA_GET_GANG(dva))
1197                         error = zio_read_gang(spa, bp, pbuf);
1198                 else
1199                         error = vdev->v_read(vdev, bp, pbuf, offset, size);
1200                 if (error == 0) {
1201                         if (cpfunc != ZIO_COMPRESS_OFF)
1202                                 error = zio_decompress_data(cpfunc, pbuf,
1203                                     BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1204                         else if (size != BP_GET_PSIZE(bp))
1205                                 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1206                 }
1207                 if (buf != pbuf)
1208                         zfs_free(pbuf, size);
1209                 if (error == 0)
1210                         break;
1211         }
1212         if (error != 0)
1213                 printf("ZFS: i/o error - all block copies unavailable\n");
1214         return (error);
1215 }
1216
1217 static int
1218 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1219 {
1220         int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1221         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1222         int nlevels = dnode->dn_nlevels;
1223         int i, rc;
1224
1225         if (bsize > SPA_MAXBLOCKSIZE) {
1226                 printf("ZFS: I/O error - blocks larger than 128K are not supported\n");
1227                 return (EIO);
1228         }
1229
1230         /*
1231          * Note: bsize may not be a power of two here so we need to do an
1232          * actual divide rather than a bitshift.
1233          */
1234         while (buflen > 0) {
1235                 uint64_t bn = offset / bsize;
1236                 int boff = offset % bsize;
1237                 int ibn;
1238                 const blkptr_t *indbp;
1239                 blkptr_t bp;
1240
1241                 if (bn > dnode->dn_maxblkid)
1242                         return (EIO);
1243
1244                 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1245                         goto cached;
1246
1247                 indbp = dnode->dn_blkptr;
1248                 for (i = 0; i < nlevels; i++) {
1249                         /*
1250                          * Copy the bp from the indirect array so that
1251                          * we can re-use the scratch buffer for multi-level
1252                          * objects.
1253                          */
1254                         ibn = bn >> ((nlevels - i - 1) * ibshift);
1255                         ibn &= ((1 << ibshift) - 1);
1256                         bp = indbp[ibn];
1257                         if (BP_IS_HOLE(&bp)) {
1258                                 memset(dnode_cache_buf, 0, bsize);
1259                                 break;
1260                         }
1261                         rc = zio_read(spa, &bp, dnode_cache_buf);
1262                         if (rc)
1263                                 return (rc);
1264                         indbp = (const blkptr_t *) dnode_cache_buf;
1265                 }
1266                 dnode_cache_obj = dnode;
1267                 dnode_cache_bn = bn;
1268         cached:
1269
1270                 /*
1271                  * The buffer contains our data block. Copy what we
1272                  * need from it and loop.
1273                  */
1274                 i = bsize - boff;
1275                 if (i > buflen) i = buflen;
1276                 memcpy(buf, &dnode_cache_buf[boff], i);
1277                 buf = ((char*) buf) + i;
1278                 offset += i;
1279                 buflen -= i;
1280         }
1281
1282         return (0);
1283 }
1284
1285 /*
1286  * Lookup a value in a microzap directory. Assumes that the zap
1287  * scratch buffer contains the directory contents.
1288  */
1289 static int
1290 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1291 {
1292         const mzap_phys_t *mz;
1293         const mzap_ent_phys_t *mze;
1294         size_t size;
1295         int chunks, i;
1296
1297         /*
1298          * Microzap objects use exactly one block. Read the whole
1299          * thing.
1300          */
1301         size = dnode->dn_datablkszsec * 512;
1302
1303         mz = (const mzap_phys_t *) zap_scratch;
1304         chunks = size / MZAP_ENT_LEN - 1;
1305
1306         for (i = 0; i < chunks; i++) {
1307                 mze = &mz->mz_chunk[i];
1308                 if (!strcmp(mze->mze_name, name)) {
1309                         *value = mze->mze_value;
1310                         return (0);
1311                 }
1312         }
1313
1314         return (ENOENT);
1315 }
1316
1317 /*
1318  * Compare a name with a zap leaf entry. Return non-zero if the name
1319  * matches.
1320  */
1321 static int
1322 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1323 {
1324         size_t namelen;
1325         const zap_leaf_chunk_t *nc;
1326         const char *p;
1327
1328         namelen = zc->l_entry.le_name_numints;
1329
1330         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1331         p = name;
1332         while (namelen > 0) {
1333                 size_t len;
1334                 len = namelen;
1335                 if (len > ZAP_LEAF_ARRAY_BYTES)
1336                         len = ZAP_LEAF_ARRAY_BYTES;
1337                 if (memcmp(p, nc->l_array.la_array, len))
1338                         return (0);
1339                 p += len;
1340                 namelen -= len;
1341                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1342         }
1343
1344         return 1;
1345 }
1346
1347 /*
1348  * Extract a uint64_t value from a zap leaf entry.
1349  */
1350 static uint64_t
1351 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1352 {
1353         const zap_leaf_chunk_t *vc;
1354         int i;
1355         uint64_t value;
1356         const uint8_t *p;
1357
1358         vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1359         for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1360                 value = (value << 8) | p[i];
1361         }
1362
1363         return value;
1364 }
1365
1366 /*
1367  * Lookup a value in a fatzap directory. Assumes that the zap scratch
1368  * buffer contains the directory header.
1369  */
1370 static int
1371 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1372 {
1373         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1374         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1375         fat_zap_t z;
1376         uint64_t *ptrtbl;
1377         uint64_t hash;
1378         int rc;
1379
1380         if (zh.zap_magic != ZAP_MAGIC)
1381                 return (EIO);
1382
1383         z.zap_block_shift = ilog2(bsize);
1384         z.zap_phys = (zap_phys_t *) zap_scratch;
1385
1386         /*
1387          * Figure out where the pointer table is and read it in if necessary.
1388          */
1389         if (zh.zap_ptrtbl.zt_blk) {
1390                 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1391                                zap_scratch, bsize);
1392                 if (rc)
1393                         return (rc);
1394                 ptrtbl = (uint64_t *) zap_scratch;
1395         } else {
1396                 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1397         }
1398
1399         hash = zap_hash(zh.zap_salt, name);
1400
1401         zap_leaf_t zl;
1402         zl.l_bs = z.zap_block_shift;
1403
1404         off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1405         zap_leaf_chunk_t *zc;
1406
1407         rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1408         if (rc)
1409                 return (rc);
1410
1411         zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1412
1413         /*
1414          * Make sure this chunk matches our hash.
1415          */
1416         if (zl.l_phys->l_hdr.lh_prefix_len > 0
1417             && zl.l_phys->l_hdr.lh_prefix
1418             != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1419                 return (ENOENT);
1420
1421         /*
1422          * Hash within the chunk to find our entry.
1423          */
1424         int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1425         int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1426         h = zl.l_phys->l_hash[h];
1427         if (h == 0xffff)
1428                 return (ENOENT);
1429         zc = &ZAP_LEAF_CHUNK(&zl, h);
1430         while (zc->l_entry.le_hash != hash) {
1431                 if (zc->l_entry.le_next == 0xffff) {
1432                         zc = 0;
1433                         break;
1434                 }
1435                 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1436         }
1437         if (fzap_name_equal(&zl, zc, name)) {
1438                 if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints > 8)
1439                         return (E2BIG);
1440                 *value = fzap_leaf_value(&zl, zc);
1441                 return (0);
1442         }
1443
1444         return (ENOENT);
1445 }
1446
1447 /*
1448  * Lookup a name in a zap object and return its value as a uint64_t.
1449  */
1450 static int
1451 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1452 {
1453         int rc;
1454         uint64_t zap_type;
1455         size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1456
1457         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1458         if (rc)
1459                 return (rc);
1460
1461         zap_type = *(uint64_t *) zap_scratch;
1462         if (zap_type == ZBT_MICRO)
1463                 return mzap_lookup(dnode, name, value);
1464         else if (zap_type == ZBT_HEADER)
1465                 return fzap_lookup(spa, dnode, name, value);
1466         printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
1467         return (EIO);
1468 }
1469
1470 /*
1471  * List a microzap directory. Assumes that the zap scratch buffer contains
1472  * the directory contents.
1473  */
1474 static int
1475 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *))
1476 {
1477         const mzap_phys_t *mz;
1478         const mzap_ent_phys_t *mze;
1479         size_t size;
1480         int chunks, i;
1481
1482         /*
1483          * Microzap objects use exactly one block. Read the whole
1484          * thing.
1485          */
1486         size = dnode->dn_datablkszsec * 512;
1487         mz = (const mzap_phys_t *) zap_scratch;
1488         chunks = size / MZAP_ENT_LEN - 1;
1489
1490         for (i = 0; i < chunks; i++) {
1491                 mze = &mz->mz_chunk[i];
1492                 if (mze->mze_name[0])
1493                         //printf("%-32s 0x%jx\n", mze->mze_name, (uintmax_t)mze->mze_value);
1494                         callback(mze->mze_name);
1495         }
1496
1497         return (0);
1498 }
1499
1500 /*
1501  * List a fatzap directory. Assumes that the zap scratch buffer contains
1502  * the directory header.
1503  */
1504 static int
1505 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *))
1506 {
1507         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1508         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1509         fat_zap_t z;
1510         int i, j;
1511
1512         if (zh.zap_magic != ZAP_MAGIC)
1513                 return (EIO);
1514
1515         z.zap_block_shift = ilog2(bsize);
1516         z.zap_phys = (zap_phys_t *) zap_scratch;
1517
1518         /*
1519          * This assumes that the leaf blocks start at block 1. The
1520          * documentation isn't exactly clear on this.
1521          */
1522         zap_leaf_t zl;
1523         zl.l_bs = z.zap_block_shift;
1524         for (i = 0; i < zh.zap_num_leafs; i++) {
1525                 off_t off = (i + 1) << zl.l_bs;
1526                 char name[256], *p;
1527                 uint64_t value;
1528
1529                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1530                         return (EIO);
1531
1532                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1533
1534                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1535                         zap_leaf_chunk_t *zc, *nc;
1536                         int namelen;
1537
1538                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1539                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1540                                 continue;
1541                         namelen = zc->l_entry.le_name_numints;
1542                         if (namelen > sizeof(name))
1543                                 namelen = sizeof(name);
1544
1545                         /*
1546                          * Paste the name back together.
1547                          */
1548                         nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1549                         p = name;
1550                         while (namelen > 0) {
1551                                 int len;
1552                                 len = namelen;
1553                                 if (len > ZAP_LEAF_ARRAY_BYTES)
1554                                         len = ZAP_LEAF_ARRAY_BYTES;
1555                                 memcpy(p, nc->l_array.la_array, len);
1556                                 p += len;
1557                                 namelen -= len;
1558                                 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1559                         }
1560
1561                         /*
1562                          * Assume the first eight bytes of the value are
1563                          * a uint64_t.
1564                          */
1565                         value = fzap_leaf_value(&zl, zc);
1566
1567                         //printf("%s 0x%jx\n", name, (uintmax_t)value);
1568                         callback((const char *)name);
1569                 }
1570         }
1571
1572         return (0);
1573 }
1574
1575 static int zfs_printf(const char *name)
1576 {
1577
1578         printf("%s\n", name);
1579
1580         return (0);
1581 }
1582
1583 /*
1584  * List a zap directory.
1585  */
1586 static int
1587 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
1588 {
1589         uint64_t zap_type;
1590         size_t size = dnode->dn_datablkszsec * 512;
1591
1592         if (dnode_read(spa, dnode, 0, zap_scratch, size))
1593                 return (EIO);
1594
1595         zap_type = *(uint64_t *) zap_scratch;
1596         if (zap_type == ZBT_MICRO)
1597                 return mzap_list(dnode, zfs_printf);
1598         else
1599                 return fzap_list(spa, dnode, zfs_printf);
1600 }
1601
1602 static int
1603 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1604 {
1605         off_t offset;
1606
1607         offset = objnum * sizeof(dnode_phys_t);
1608         return dnode_read(spa, &os->os_meta_dnode, offset,
1609                 dnode, sizeof(dnode_phys_t));
1610 }
1611
1612 static int
1613 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1614 {
1615         const mzap_phys_t *mz;
1616         const mzap_ent_phys_t *mze;
1617         size_t size;
1618         int chunks, i;
1619
1620         /*
1621          * Microzap objects use exactly one block. Read the whole
1622          * thing.
1623          */
1624         size = dnode->dn_datablkszsec * 512;
1625
1626         mz = (const mzap_phys_t *) zap_scratch;
1627         chunks = size / MZAP_ENT_LEN - 1;
1628
1629         for (i = 0; i < chunks; i++) {
1630                 mze = &mz->mz_chunk[i];
1631                 if (value == mze->mze_value) {
1632                         strcpy(name, mze->mze_name);
1633                         return (0);
1634                 }
1635         }
1636
1637         return (ENOENT);
1638 }
1639
1640 static void
1641 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
1642 {
1643         size_t namelen;
1644         const zap_leaf_chunk_t *nc;
1645         char *p;
1646
1647         namelen = zc->l_entry.le_name_numints;
1648
1649         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1650         p = name;
1651         while (namelen > 0) {
1652                 size_t len;
1653                 len = namelen;
1654                 if (len > ZAP_LEAF_ARRAY_BYTES)
1655                         len = ZAP_LEAF_ARRAY_BYTES;
1656                 memcpy(p, nc->l_array.la_array, len);
1657                 p += len;
1658                 namelen -= len;
1659                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1660         }
1661
1662         *p = '\0';
1663 }
1664
1665 static int
1666 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1667 {
1668         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1669         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1670         fat_zap_t z;
1671         int i, j;
1672
1673         if (zh.zap_magic != ZAP_MAGIC)
1674                 return (EIO);
1675
1676         z.zap_block_shift = ilog2(bsize);
1677         z.zap_phys = (zap_phys_t *) zap_scratch;
1678
1679         /*
1680          * This assumes that the leaf blocks start at block 1. The
1681          * documentation isn't exactly clear on this.
1682          */
1683         zap_leaf_t zl;
1684         zl.l_bs = z.zap_block_shift;
1685         for (i = 0; i < zh.zap_num_leafs; i++) {
1686                 off_t off = (i + 1) << zl.l_bs;
1687
1688                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1689                         return (EIO);
1690
1691                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1692
1693                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1694                         zap_leaf_chunk_t *zc;
1695
1696                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1697                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1698                                 continue;
1699                         if (zc->l_entry.le_value_intlen != 8 ||
1700                             zc->l_entry.le_value_numints != 1)
1701                                 continue;
1702
1703                         if (fzap_leaf_value(&zl, zc) == value) {
1704                                 fzap_name_copy(&zl, zc, name);
1705                                 return (0);
1706                         }
1707                 }
1708         }
1709
1710         return (ENOENT);
1711 }
1712
1713 static int
1714 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1715 {
1716         int rc;
1717         uint64_t zap_type;
1718         size_t size = dnode->dn_datablkszsec * 512;
1719
1720         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1721         if (rc)
1722                 return (rc);
1723
1724         zap_type = *(uint64_t *) zap_scratch;
1725         if (zap_type == ZBT_MICRO)
1726                 return mzap_rlookup(spa, dnode, name, value);
1727         else
1728                 return fzap_rlookup(spa, dnode, name, value);
1729 }
1730
1731 static int
1732 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
1733 {
1734         char name[256];
1735         char component[256];
1736         uint64_t dir_obj, parent_obj, child_dir_zapobj;
1737         dnode_phys_t child_dir_zap, dataset, dir, parent;
1738         dsl_dir_phys_t *dd;
1739         dsl_dataset_phys_t *ds;
1740         char *p;
1741         int len;
1742
1743         p = &name[sizeof(name) - 1];
1744         *p = '\0';
1745
1746         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1747                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1748                 return (EIO);
1749         }
1750         ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
1751         dir_obj = ds->ds_dir_obj;
1752
1753         for (;;) {
1754                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
1755                         return (EIO);
1756                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1757
1758                 /* Actual loop condition. */
1759                 parent_obj  = dd->dd_parent_obj;
1760                 if (parent_obj == 0)
1761                         break;
1762
1763                 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
1764                         return (EIO);
1765                 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
1766                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1767                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1768                         return (EIO);
1769                 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
1770                         return (EIO);
1771
1772                 len = strlen(component);
1773                 p -= len;
1774                 memcpy(p, component, len);
1775                 --p;
1776                 *p = '/';
1777
1778                 /* Actual loop iteration. */
1779                 dir_obj = parent_obj;
1780         }
1781
1782         if (*p != '\0')
1783                 ++p;
1784         strcpy(result, p);
1785
1786         return (0);
1787 }
1788
1789 static int
1790 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
1791 {
1792         char element[256];
1793         uint64_t dir_obj, child_dir_zapobj;
1794         dnode_phys_t child_dir_zap, dir;
1795         dsl_dir_phys_t *dd;
1796         const char *p, *q;
1797
1798         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
1799                 return (EIO);
1800         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &dir_obj))
1801                 return (EIO);
1802
1803         p = name;
1804         for (;;) {
1805                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
1806                         return (EIO);
1807                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1808
1809                 while (*p == '/')
1810                         p++;
1811                 /* Actual loop condition #1. */
1812                 if (*p == '\0')
1813                         break;
1814
1815                 q = strchr(p, '/');
1816                 if (q) {
1817                         memcpy(element, p, q - p);
1818                         element[q - p] = '\0';
1819                         p = q + 1;
1820                 } else {
1821                         strcpy(element, p);
1822                         p += strlen(p);
1823                 }
1824
1825                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1826                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1827                         return (EIO);
1828
1829                 /* Actual loop condition #2. */
1830                 if (zap_lookup(spa, &child_dir_zap, element, &dir_obj) != 0)
1831                         return (ENOENT);
1832         }
1833
1834         *objnum = dd->dd_head_dataset_obj;
1835         return (0);
1836 }
1837
1838 #ifndef BOOT2
1839 static int
1840 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
1841 {
1842         uint64_t dir_obj, child_dir_zapobj;
1843         dnode_phys_t child_dir_zap, dir, dataset;
1844         dsl_dataset_phys_t *ds;
1845         dsl_dir_phys_t *dd;
1846
1847         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1848                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1849                 return (EIO);
1850         }
1851         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1852         dir_obj = ds->ds_dir_obj;
1853
1854         if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
1855                 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
1856                 return (EIO);
1857         }
1858         dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1859
1860         child_dir_zapobj = dd->dd_child_dir_zapobj;
1861         if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
1862                 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
1863                 return (EIO);
1864         }
1865
1866         return (zap_list(spa, &child_dir_zap) != 0);
1867 }
1868
1869 int
1870 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *name))
1871 {
1872         uint64_t dir_obj, child_dir_zapobj, zap_type;
1873         dnode_phys_t child_dir_zap, dir, dataset;
1874         dsl_dataset_phys_t *ds;
1875         dsl_dir_phys_t *dd;
1876         int err;
1877
1878         err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
1879         if (err != 0) {
1880                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1881                 return (err);
1882         }
1883         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1884         dir_obj = ds->ds_dir_obj;
1885
1886         err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
1887         if (err != 0) {
1888                 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
1889                 return (err);
1890         }
1891         dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1892
1893         child_dir_zapobj = dd->dd_child_dir_zapobj;
1894         err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
1895         if (err != 0) {
1896                 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
1897                 return (err);
1898         }
1899
1900         err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
1901         if (err != 0)
1902                 return (err);
1903
1904         zap_type = *(uint64_t *) zap_scratch;
1905         if (zap_type == ZBT_MICRO)
1906                 return mzap_list(&child_dir_zap, callback);
1907         else
1908                 return fzap_list(spa, &child_dir_zap, callback);
1909 }
1910 #endif
1911
1912 /*
1913  * Find the object set given the object number of its dataset object
1914  * and return its details in *objset
1915  */
1916 static int
1917 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
1918 {
1919         dnode_phys_t dataset;
1920         dsl_dataset_phys_t *ds;
1921
1922         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1923                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1924                 return (EIO);
1925         }
1926
1927         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1928         if (zio_read(spa, &ds->ds_bp, objset)) {
1929                 printf("ZFS: can't read object set for dataset %ju\n",
1930                     (uintmax_t)objnum);
1931                 return (EIO);
1932         }
1933
1934         return (0);
1935 }
1936
1937 /*
1938  * Find the object set pointed to by the BOOTFS property or the root
1939  * dataset if there is none and return its details in *objset
1940  */
1941 static int
1942 zfs_get_root(const spa_t *spa, uint64_t *objid)
1943 {
1944         dnode_phys_t dir, propdir;
1945         uint64_t props, bootfs, root;
1946
1947         *objid = 0;
1948
1949         /*
1950          * Start with the MOS directory object.
1951          */
1952         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
1953                 printf("ZFS: can't read MOS object directory\n");
1954                 return (EIO);
1955         }
1956
1957         /*
1958          * Lookup the pool_props and see if we can find a bootfs.
1959          */
1960         if (zap_lookup(spa, &dir, DMU_POOL_PROPS, &props) == 0
1961              && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
1962              && zap_lookup(spa, &propdir, "bootfs", &bootfs) == 0
1963              && bootfs != 0)
1964         {
1965                 *objid = bootfs;
1966                 return (0);
1967         }
1968         /*
1969          * Lookup the root dataset directory
1970          */
1971         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &root)
1972             || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
1973                 printf("ZFS: can't find root dsl_dir\n");
1974                 return (EIO);
1975         }
1976
1977         /*
1978          * Use the information from the dataset directory's bonus buffer
1979          * to find the dataset object and from that the object set itself.
1980          */
1981         dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
1982         *objid = dd->dd_head_dataset_obj;
1983         return (0);
1984 }
1985
1986 static int
1987 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
1988 {
1989
1990         mount->spa = spa;
1991
1992         /*
1993          * Find the root object set if not explicitly provided
1994          */
1995         if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
1996                 printf("ZFS: can't find root filesystem\n");
1997                 return (EIO);
1998         }
1999
2000         if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
2001                 printf("ZFS: can't open root filesystem\n");
2002                 return (EIO);
2003         }
2004
2005         mount->rootobj = rootobj;
2006
2007         return (0);
2008 }
2009
2010 static int
2011 zfs_spa_init(spa_t *spa)
2012 {
2013
2014         if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
2015                 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
2016                 return (EIO);
2017         }
2018         if (spa->spa_mos.os_type != DMU_OST_META) {
2019                 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
2020                 return (EIO);
2021         }
2022         return (0);
2023 }
2024
2025 static int
2026 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
2027 {
2028
2029         if (dn->dn_bonustype != DMU_OT_SA) {
2030                 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
2031
2032                 sb->st_mode = zp->zp_mode;
2033                 sb->st_uid = zp->zp_uid;
2034                 sb->st_gid = zp->zp_gid;
2035                 sb->st_size = zp->zp_size;
2036         } else {
2037                 sa_hdr_phys_t *sahdrp;
2038                 int hdrsize;
2039                 size_t size = 0;
2040                 void *buf = NULL;
2041
2042                 if (dn->dn_bonuslen != 0)
2043                         sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2044                 else {
2045                         if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
2046                                 blkptr_t *bp = &dn->dn_spill;
2047                                 int error;
2048
2049                                 size = BP_GET_LSIZE(bp);
2050                                 buf = zfs_alloc(size);
2051                                 error = zio_read(spa, bp, buf);
2052                                 if (error != 0) {
2053                                         zfs_free(buf, size);
2054                                         return (error);
2055                                 }
2056                                 sahdrp = buf;
2057                         } else {
2058                                 return (EIO);
2059                         }
2060                 }
2061                 hdrsize = SA_HDR_SIZE(sahdrp);
2062                 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
2063                     SA_MODE_OFFSET);
2064                 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
2065                     SA_UID_OFFSET);
2066                 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
2067                     SA_GID_OFFSET);
2068                 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
2069                     SA_SIZE_OFFSET);
2070                 if (buf != NULL)
2071                         zfs_free(buf, size);
2072         }
2073
2074         return (0);
2075 }
2076
2077 /*
2078  * Lookup a file and return its dnode.
2079  */
2080 static int
2081 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
2082 {
2083         int rc;
2084         uint64_t objnum, rootnum, parentnum;
2085         const spa_t *spa;
2086         dnode_phys_t dn;
2087         const char *p, *q;
2088         char element[256];
2089         char path[1024];
2090         int symlinks_followed = 0;
2091         struct stat sb;
2092
2093         spa = mount->spa;
2094         if (mount->objset.os_type != DMU_OST_ZFS) {
2095                 printf("ZFS: unexpected object set type %ju\n",
2096                     (uintmax_t)mount->objset.os_type);
2097                 return (EIO);
2098         }
2099
2100         /*
2101          * Get the root directory dnode.
2102          */
2103         rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
2104         if (rc)
2105                 return (rc);
2106
2107         rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, &rootnum);
2108         if (rc)
2109                 return (rc);
2110
2111         rc = objset_get_dnode(spa, &mount->objset, rootnum, &dn);
2112         if (rc)
2113                 return (rc);
2114
2115         objnum = rootnum;
2116         p = upath;
2117         while (p && *p) {
2118                 while (*p == '/')
2119                         p++;
2120                 if (!*p)
2121                         break;
2122                 q = strchr(p, '/');
2123                 if (q) {
2124                         memcpy(element, p, q - p);
2125                         element[q - p] = 0;
2126                         p = q;
2127                 } else {
2128                         strcpy(element, p);
2129                         p = 0;
2130                 }
2131
2132                 rc = zfs_dnode_stat(spa, &dn, &sb);
2133                 if (rc)
2134                         return (rc);
2135                 if (!S_ISDIR(sb.st_mode))
2136                         return (ENOTDIR);
2137
2138                 parentnum = objnum;
2139                 rc = zap_lookup(spa, &dn, element, &objnum);
2140                 if (rc)
2141                         return (rc);
2142                 objnum = ZFS_DIRENT_OBJ(objnum);
2143
2144                 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
2145                 if (rc)
2146                         return (rc);
2147
2148                 /*
2149                  * Check for symlink.
2150                  */
2151                 rc = zfs_dnode_stat(spa, &dn, &sb);
2152                 if (rc)
2153                         return (rc);
2154                 if (S_ISLNK(sb.st_mode)) {
2155                         if (symlinks_followed > 10)
2156                                 return (EMLINK);
2157                         symlinks_followed++;
2158
2159                         /*
2160                          * Read the link value and copy the tail of our
2161                          * current path onto the end.
2162                          */
2163                         if (p)
2164                                 strcpy(&path[sb.st_size], p);
2165                         else
2166                                 path[sb.st_size] = 0;
2167                         /*
2168                          * Second test is purely to silence bogus compiler
2169                          * warning about accessing past the end of dn_bonus.
2170                          */
2171                         if (sb.st_size + sizeof(znode_phys_t) <=
2172                             dn.dn_bonuslen && sizeof(znode_phys_t) <=
2173                             sizeof(dn.dn_bonus)) {
2174                                 memcpy(path, &dn.dn_bonus[sizeof(znode_phys_t)],
2175                                         sb.st_size);
2176                         } else {
2177                                 rc = dnode_read(spa, &dn, 0, path, sb.st_size);
2178                                 if (rc)
2179                                         return (rc);
2180                         }
2181
2182                         /*
2183                          * Restart with the new path, starting either at
2184                          * the root or at the parent depending whether or
2185                          * not the link is relative.
2186                          */
2187                         p = path;
2188                         if (*p == '/')
2189                                 objnum = rootnum;
2190                         else
2191                                 objnum = parentnum;
2192                         objset_get_dnode(spa, &mount->objset, objnum, &dn);
2193                 }
2194         }
2195
2196         *dnode = dn;
2197         return (0);
2198 }