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         NULL
  60 };
  61
  62 /*
  63  * List of all pools, chained through spa_link.
  64  */
  65 static spa_list_t zfs_pools;
  66
  67 static uint64_t zfs_crc64_table[256];
  68 static const dnode_phys_t *dnode_cache_obj = 0;
  69 static uint64_t dnode_cache_bn;
  70 static char *dnode_cache_buf;
  71 static char *zap_scratch;
  72 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
  73
  74 #define TEMP_SIZE       (1024 * 1024)
  75
  76 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
  77 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
  78 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
  79
  80 static void
  81 zfs_init(void)
  82 {
  83         STAILQ_INIT(&zfs_vdevs);
  84         STAILQ_INIT(&zfs_pools);
  85
  86         zfs_temp_buf = malloc(TEMP_SIZE);
  87         zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
  88         zfs_temp_ptr = zfs_temp_buf;
  89         dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
  90         zap_scratch = malloc(SPA_MAXBLOCKSIZE);
  91
  92         zfs_init_crc();
  93 }
  94
  95 static void *
  96 zfs_alloc(size_t size)
  97 {
  98         char *ptr;
  99
 100         if (zfs_temp_ptr + size > zfs_temp_end) {
 101                 printf("ZFS: out of temporary buffer space\n");
 102                 for (;;) ;
 103         }
 104         ptr = zfs_temp_ptr;
 105         zfs_temp_ptr += size;
 106
 107         return (ptr);
 108 }
 109
 110 static void
 111 zfs_free(void *ptr, size_t size)
 112 {
 113
 114         zfs_temp_ptr -= size;
 115         if (zfs_temp_ptr != ptr) {
 116                 printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
 117                 for (;;) ;
 118         }
 119 }
 120
 121 static int
 122 xdr_int(const unsigned char **xdr, int *ip)
 123 {
 124         *ip = ((*xdr)[0] << 24)
 125                 | ((*xdr)[1] << 16)
 126                 | ((*xdr)[2] << 8)
 127                 | ((*xdr)[3] << 0);
 128         (*xdr) += 4;
 129         return (0);
 130 }
 131
 132 static int
 133 xdr_u_int(const unsigned char **xdr, u_int *ip)
 134 {
 135         *ip = ((*xdr)[0] << 24)
 136                 | ((*xdr)[1] << 16)
 137                 | ((*xdr)[2] << 8)
 138                 | ((*xdr)[3] << 0);
 139         (*xdr) += 4;
 140         return (0);
 141 }
 142
 143 static int
 144 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
 145 {
 146         u_int hi, lo;
 147
 148         xdr_u_int(xdr, &hi);
 149         xdr_u_int(xdr, &lo);
 150         *lp = (((uint64_t) hi) << 32) | lo;
 151         return (0);
 152 }
 153
 154 static int
 155 nvlist_find(const unsigned char *nvlist, const char *name, int type,
 156             int* elementsp, void *valuep)
 157 {
 158         const unsigned char *p, *pair;
 159         int junk;
 160         int encoded_size, decoded_size;
 161
 162         p = nvlist;
 163         xdr_int(&p, &junk);
 164         xdr_int(&p, &junk);
 165
 166         pair = p;
 167         xdr_int(&p, &encoded_size);
 168         xdr_int(&p, &decoded_size);
 169         while (encoded_size && decoded_size) {
 170                 int namelen, pairtype, elements;
 171                 const char *pairname;
 172
 173                 xdr_int(&p, &namelen);
 174                 pairname = (const char*) p;
 175                 p += roundup(namelen, 4);
 176                 xdr_int(&p, &pairtype);
 177
 178                 if (!memcmp(name, pairname, namelen) && type == pairtype) {
 179                         xdr_int(&p, &elements);
 180                         if (elementsp)
 181                                 *elementsp = elements;
 182                         if (type == DATA_TYPE_UINT64) {
 183                                 xdr_uint64_t(&p, (uint64_t *) valuep);
 184                                 return (0);
 185                         } else if (type == DATA_TYPE_STRING) {
 186                                 int len;
 187                                 xdr_int(&p, &len);
 188                                 (*(const char**) valuep) = (const char*) p;
 189                                 return (0);
 190                         } else if (type == DATA_TYPE_NVLIST
 191                                    || type == DATA_TYPE_NVLIST_ARRAY) {
 192                                 (*(const unsigned char**) valuep) =
 193                                          (const unsigned char*) p;
 194                                 return (0);
 195                         } else {
 196                                 return (EIO);
 197                         }
 198                 } else {
 199                         /*
 200                          * Not the pair we are looking for, skip to the next one.
 201                          */
 202                         p = pair + encoded_size;
 203                 }
 204
 205                 pair = p;
 206                 xdr_int(&p, &encoded_size);
 207                 xdr_int(&p, &decoded_size);
 208         }
 209
 210         return (EIO);
 211 }
 212
 213 static int
 214 nvlist_check_features_for_read(const unsigned char *nvlist)
 215 {
 216         const unsigned char *p, *pair;
 217         int junk;
 218         int encoded_size, decoded_size;
 219         int rc;
 220
 221         rc = 0;
 222
 223         p = nvlist;
 224         xdr_int(&p, &junk);
 225         xdr_int(&p, &junk);
 226
 227         pair = p;
 228         xdr_int(&p, &encoded_size);
 229         xdr_int(&p, &decoded_size);
 230         while (encoded_size && decoded_size) {
 231                 int namelen, pairtype;
 232                 const char *pairname;
 233                 int i, found;
 234
 235                 found = 0;
 236
 237                 xdr_int(&p, &namelen);
 238                 pairname = (const char*) p;
 239                 p += roundup(namelen, 4);
 240                 xdr_int(&p, &pairtype);
 241
 242                 for (i = 0; features_for_read[i] != NULL; i++) {
 243                         if (!memcmp(pairname, features_for_read[i], namelen)) {
 244                                 found = 1;
 245                                 break;
 246                         }
 247                 }
 248
 249                 if (!found) {
 250                         printf("ZFS: unsupported feature: %s\n", pairname);
 251                         rc = EIO;
 252                 }
 253
 254                 p = pair + encoded_size;
 255
 256                 pair = p;
 257                 xdr_int(&p, &encoded_size);
 258                 xdr_int(&p, &decoded_size);
 259         }
 260
 261         return (rc);
 262 }
 263
 264 /*
 265  * Return the next nvlist in an nvlist array.
 266  */
 267 static const unsigned char *
 268 nvlist_next(const unsigned char *nvlist)
 269 {
 270         const unsigned char *p, *pair;
 271         int junk;
 272         int encoded_size, decoded_size;
 273
 274         p = nvlist;
 275         xdr_int(&p, &junk);
 276         xdr_int(&p, &junk);
 277
 278         pair = p;
 279         xdr_int(&p, &encoded_size);
 280         xdr_int(&p, &decoded_size);
 281         while (encoded_size && decoded_size) {
 282                 p = pair + encoded_size;
 283
 284                 pair = p;
 285                 xdr_int(&p, &encoded_size);
 286                 xdr_int(&p, &decoded_size);
 287         }
 288
 289         return p;
 290 }
 291
 292 #ifdef TEST
 293
 294 static const unsigned char *
 295 nvlist_print(const unsigned char *nvlist, unsigned int indent)
 296 {
 297         static const char* typenames[] = {
 298                 "DATA_TYPE_UNKNOWN",
 299                 "DATA_TYPE_BOOLEAN",
 300                 "DATA_TYPE_BYTE",
 301                 "DATA_TYPE_INT16",
 302                 "DATA_TYPE_UINT16",
 303                 "DATA_TYPE_INT32",
 304                 "DATA_TYPE_UINT32",
 305                 "DATA_TYPE_INT64",
 306                 "DATA_TYPE_UINT64",
 307                 "DATA_TYPE_STRING",
 308                 "DATA_TYPE_BYTE_ARRAY",
 309                 "DATA_TYPE_INT16_ARRAY",
 310                 "DATA_TYPE_UINT16_ARRAY",
 311                 "DATA_TYPE_INT32_ARRAY",
 312                 "DATA_TYPE_UINT32_ARRAY",
 313                 "DATA_TYPE_INT64_ARRAY",
 314                 "DATA_TYPE_UINT64_ARRAY",
 315                 "DATA_TYPE_STRING_ARRAY",
 316                 "DATA_TYPE_HRTIME",
 317                 "DATA_TYPE_NVLIST",
 318                 "DATA_TYPE_NVLIST_ARRAY",
 319                 "DATA_TYPE_BOOLEAN_VALUE",
 320                 "DATA_TYPE_INT8",
 321                 "DATA_TYPE_UINT8",
 322                 "DATA_TYPE_BOOLEAN_ARRAY",
 323                 "DATA_TYPE_INT8_ARRAY",
 324                 "DATA_TYPE_UINT8_ARRAY"
 325         };
 326
 327         unsigned int i, j;
 328         const unsigned char *p, *pair;
 329         int junk;
 330         int encoded_size, decoded_size;
 331
 332         p = nvlist;
 333         xdr_int(&p, &junk);
 334         xdr_int(&p, &junk);
 335
 336         pair = p;
 337         xdr_int(&p, &encoded_size);
 338         xdr_int(&p, &decoded_size);
 339         while (encoded_size && decoded_size) {
 340                 int namelen, pairtype, elements;
 341                 const char *pairname;
 342
 343                 xdr_int(&p, &namelen);
 344                 pairname = (const char*) p;
 345                 p += roundup(namelen, 4);
 346                 xdr_int(&p, &pairtype);
 347
 348                 for (i = 0; i < indent; i++)
 349                         printf(" ");
 350                 printf("%s %s", typenames[pairtype], pairname);
 351
 352                 xdr_int(&p, &elements);
 353                 switch (pairtype) {
 354                 case DATA_TYPE_UINT64: {
 355                         uint64_t val;
 356                         xdr_uint64_t(&p, &val);
 357                         printf(" = 0x%jx\n", (uintmax_t)val);
 358                         break;
 359                 }
 360
 361                 case DATA_TYPE_STRING: {
 362                         int len;
 363                         xdr_int(&p, &len);
 364                         printf(" = \"%s\"\n", p);
 365                         break;
 366                 }
 367
 368                 case DATA_TYPE_NVLIST:
 369                         printf("\n");
 370                         nvlist_print(p, indent + 1);
 371                         break;
 372
 373                 case DATA_TYPE_NVLIST_ARRAY:
 374                         for (j = 0; j < elements; j++) {
 375                                 printf("[%d]\n", j);
 376                                 p = nvlist_print(p, indent + 1);
 377                                 if (j != elements - 1) {
 378                                         for (i = 0; i < indent; i++)
 379                                                 printf(" ");
 380                                         printf("%s %s", typenames[pairtype], pairname);
 381                                 }
 382                         }
 383                         break;
 384
 385                 default:
 386                         printf("\n");
 387                 }
 388
 389                 p = pair + encoded_size;
 390
 391                 pair = p;
 392                 xdr_int(&p, &encoded_size);
 393                 xdr_int(&p, &decoded_size);
 394         }
 395
 396         return p;
 397 }
 398
 399 #endif
 400
 401 static int
 402 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
 403     off_t offset, size_t size)
 404 {
 405         size_t psize;
 406         int rc;
 407
 408         if (!vdev->v_phys_read)
 409                 return (EIO);
 410
 411         if (bp) {
 412                 psize = BP_GET_PSIZE(bp);
 413         } else {
 414                 psize = size;
 415         }
 416
 417         /*printf("ZFS: reading %d bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
 418         rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
 419         if (rc)
 420                 return (rc);
 421         if (bp && zio_checksum_verify(bp, buf))
 422                 return (EIO);
 423
 424         return (0);
 425 }
 426
 427 static int
 428 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 429     off_t offset, size_t bytes)
 430 {
 431
 432         return (vdev_read_phys(vdev, bp, buf,
 433                 offset + VDEV_LABEL_START_SIZE, bytes));
 434 }
 435
 436
 437 static int
 438 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 439     off_t offset, size_t bytes)
 440 {
 441         vdev_t *kid;
 442         int rc;
 443
 444         rc = EIO;
 445         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 446                 if (kid->v_state != VDEV_STATE_HEALTHY)
 447                         continue;
 448                 rc = kid->v_read(kid, bp, buf, offset, bytes);
 449                 if (!rc)
 450                         return (0);
 451         }
 452
 453         return (rc);
 454 }
 455
 456 static int
 457 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 458     off_t offset, size_t bytes)
 459 {
 460         vdev_t *kid;
 461
 462         /*
 463          * Here we should have two kids:
 464          * First one which is the one we are replacing and we can trust
 465          * only this one to have valid data, but it might not be present.
 466          * Second one is that one we are replacing with. It is most likely
 467          * healthy, but we can't trust it has needed data, so we won't use it.
 468          */
 469         kid = STAILQ_FIRST(&vdev->v_children);
 470         if (kid == NULL)
 471                 return (EIO);
 472         if (kid->v_state != VDEV_STATE_HEALTHY)
 473                 return (EIO);
 474         return (kid->v_read(kid, bp, buf, offset, bytes));
 475 }
 476
 477 static vdev_t *
 478 vdev_find(uint64_t guid)
 479 {
 480         vdev_t *vdev;
 481
 482         STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
 483                 if (vdev->v_guid == guid)
 484                         return (vdev);
 485
 486         return (0);
 487 }
 488
 489 static vdev_t *
 490 vdev_create(uint64_t guid, vdev_read_t *read)
 491 {
 492         vdev_t *vdev;
 493
 494         vdev = malloc(sizeof(vdev_t));
 495         memset(vdev, 0, sizeof(vdev_t));
 496         STAILQ_INIT(&vdev->v_children);
 497         vdev->v_guid = guid;
 498         vdev->v_state = VDEV_STATE_OFFLINE;
 499         vdev->v_read = read;
 500         vdev->v_phys_read = 0;
 501         vdev->v_read_priv = 0;
 502         STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
 503
 504         return (vdev);
 505 }
 506
 507 static int
 508 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
 509     vdev_t **vdevp, int is_newer)
 510 {
 511         int rc;
 512         uint64_t guid, id, ashift, nparity;
 513         const char *type;
 514         const char *path;
 515         vdev_t *vdev, *kid;
 516         const unsigned char *kids;
 517         int nkids, i, is_new;
 518         uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
 519
 520         if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID,
 521                         DATA_TYPE_UINT64, 0, &guid)
 522             || nvlist_find(nvlist, ZPOOL_CONFIG_ID,
 523                            DATA_TYPE_UINT64, 0, &id)
 524             || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE,
 525                            DATA_TYPE_STRING, 0, &type)) {
 526                 printf("ZFS: can't find vdev details\n");
 527                 return (ENOENT);
 528         }
 529
 530         if (strcmp(type, VDEV_TYPE_MIRROR)
 531             && strcmp(type, VDEV_TYPE_DISK)
 532 #ifdef ZFS_TEST
 533             && strcmp(type, VDEV_TYPE_FILE)
 534 #endif
 535             && strcmp(type, VDEV_TYPE_RAIDZ)
 536             && strcmp(type, VDEV_TYPE_REPLACING)) {
 537                 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 538                 return (EIO);
 539         }
 540
 541         is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
 542
 543         nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, 0,
 544                         &is_offline);
 545         nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, 0,
 546                         &is_removed);
 547         nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, 0,
 548                         &is_faulted);
 549         nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, 0,
 550                         &is_degraded);
 551         nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, 0,
 552                         &isnt_present);
 553
 554         vdev = vdev_find(guid);
 555         if (!vdev) {
 556                 is_new = 1;
 557
 558                 if (!strcmp(type, VDEV_TYPE_MIRROR))
 559                         vdev = vdev_create(guid, vdev_mirror_read);
 560                 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
 561                         vdev = vdev_create(guid, vdev_raidz_read);
 562                 else if (!strcmp(type, VDEV_TYPE_REPLACING))
 563                         vdev = vdev_create(guid, vdev_replacing_read);
 564                 else
 565                         vdev = vdev_create(guid, vdev_disk_read);
 566
 567                 vdev->v_id = id;
 568                 vdev->v_top = pvdev != NULL ? pvdev : vdev;
 569                 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
 570                         DATA_TYPE_UINT64, 0, &ashift) == 0)
 571                         vdev->v_ashift = ashift;
 572                 else
 573                         vdev->v_ashift = 0;
 574                 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
 575                         DATA_TYPE_UINT64, 0, &nparity) == 0)
 576                         vdev->v_nparity = nparity;
 577                 else
 578                         vdev->v_nparity = 0;
 579                 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
 580                                 DATA_TYPE_STRING, 0, &path) == 0) {
 581                         if (strncmp(path, "/dev/", 5) == 0)
 582                                 path += 5;
 583                         vdev->v_name = strdup(path);
 584                 } else {
 585                         if (!strcmp(type, "raidz")) {
 586                                 if (vdev->v_nparity == 1)
 587                                         vdev->v_name = "raidz1";
 588                                 else if (vdev->v_nparity == 2)
 589                                         vdev->v_name = "raidz2";
 590                                 else if (vdev->v_nparity == 3)
 591                                         vdev->v_name = "raidz3";
 592                                 else {
 593                                         printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 594                                         return (EIO);
 595                                 }
 596                         } else {
 597                                 vdev->v_name = strdup(type);
 598                         }
 599                 }
 600         } else {
 601                 is_new = 0;
 602         }
 603
 604         if (is_new || is_newer) {
 605                 /*
 606                  * This is either new vdev or we've already seen this vdev,
 607                  * but from an older vdev label, so let's refresh its state
 608                  * from the newer label.
 609                  */
 610                 if (is_offline)
 611                         vdev->v_state = VDEV_STATE_OFFLINE;
 612                 else if (is_removed)
 613                         vdev->v_state = VDEV_STATE_REMOVED;
 614                 else if (is_faulted)
 615                         vdev->v_state = VDEV_STATE_FAULTED;
 616                 else if (is_degraded)
 617                         vdev->v_state = VDEV_STATE_DEGRADED;
 618                 else if (isnt_present)
 619                         vdev->v_state = VDEV_STATE_CANT_OPEN;
 620         }
 621
 622         rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN,
 623                          DATA_TYPE_NVLIST_ARRAY, &nkids, &kids);
 624         /*
 625          * Its ok if we don't have any kids.
 626          */
 627         if (rc == 0) {
 628                 vdev->v_nchildren = nkids;
 629                 for (i = 0; i < nkids; i++) {
 630                         rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
 631                         if (rc)
 632                                 return (rc);
 633                         if (is_new)
 634                                 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
 635                                                    v_childlink);
 636                         kids = nvlist_next(kids);
 637                 }
 638         } else {
 639                 vdev->v_nchildren = 0;
 640         }
 641
 642         if (vdevp)
 643                 *vdevp = vdev;
 644         return (0);
 645 }
 646
 647 static void
 648 vdev_set_state(vdev_t *vdev)
 649 {
 650         vdev_t *kid;
 651         int good_kids;
 652         int bad_kids;
 653
 654         /*
 655          * A mirror or raidz is healthy if all its kids are healthy. A
 656          * mirror is degraded if any of its kids is healthy; a raidz
 657          * is degraded if at most nparity kids are offline.
 658          */
 659         if (STAILQ_FIRST(&vdev->v_children)) {
 660                 good_kids = 0;
 661                 bad_kids = 0;
 662                 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 663                         if (kid->v_state == VDEV_STATE_HEALTHY)
 664                                 good_kids++;
 665                         else
 666                                 bad_kids++;
 667                 }
 668                 if (bad_kids == 0) {
 669                         vdev->v_state = VDEV_STATE_HEALTHY;
 670                 } else {
 671                         if (vdev->v_read == vdev_mirror_read) {
 672                                 if (good_kids) {
 673                                         vdev->v_state = VDEV_STATE_DEGRADED;
 674                                 } else {
 675                                         vdev->v_state = VDEV_STATE_OFFLINE;
 676                                 }
 677                         } else if (vdev->v_read == vdev_raidz_read) {
 678                                 if (bad_kids > vdev->v_nparity) {
 679                                         vdev->v_state = VDEV_STATE_OFFLINE;
 680                                 } else {
 681                                         vdev->v_state = VDEV_STATE_DEGRADED;
 682                                 }
 683                         }
 684                 }
 685         }
 686 }
 687
 688 static spa_t *
 689 spa_find_by_guid(uint64_t guid)
 690 {
 691         spa_t *spa;
 692
 693         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 694                 if (spa->spa_guid == guid)
 695                         return (spa);
 696
 697         return (0);
 698 }
 699
 700 static spa_t *
 701 spa_find_by_name(const char *name)
 702 {
 703         spa_t *spa;
 704
 705         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 706                 if (!strcmp(spa->spa_name, name))
 707                         return (spa);
 708
 709         return (0);
 710 }
 711
 712 #ifdef BOOT2
 713 static spa_t *
 714 spa_get_primary(void)
 715 {
 716
 717         return (STAILQ_FIRST(&zfs_pools));
 718 }
 719
 720 static vdev_t *
 721 spa_get_primary_vdev(const spa_t *spa)
 722 {
 723         vdev_t *vdev;
 724         vdev_t *kid;
 725
 726         if (spa == NULL)
 727                 spa = spa_get_primary();
 728         if (spa == NULL)
 729                 return (NULL);
 730         vdev = STAILQ_FIRST(&spa->spa_vdevs);
 731         if (vdev == NULL)
 732                 return (NULL);
 733         for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
 734              kid = STAILQ_FIRST(&vdev->v_children))
 735                 vdev = kid;
 736         return (vdev);
 737 }
 738 #endif
 739
 740 static spa_t *
 741 spa_create(uint64_t guid)
 742 {
 743         spa_t *spa;
 744
 745         spa = malloc(sizeof(spa_t));
 746         memset(spa, 0, sizeof(spa_t));
 747         STAILQ_INIT(&spa->spa_vdevs);
 748         spa->spa_guid = guid;
 749         STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
 750
 751         return (spa);
 752 }
 753
 754 static const char *
 755 state_name(vdev_state_t state)
 756 {
 757         static const char* names[] = {
 758                 "UNKNOWN",
 759                 "CLOSED",
 760                 "OFFLINE",
 761                 "REMOVED",
 762                 "CANT_OPEN",
 763                 "FAULTED",
 764                 "DEGRADED",
 765                 "ONLINE"
 766         };
 767         return names[state];
 768 }
 769
 770 #ifdef BOOT2
 771
 772 #define pager_printf printf
 773
 774 #else
 775
 776 static void
 777 pager_printf(const char *fmt, ...)
 778 {
 779         char line[80];
 780         va_list args;
 781
 782         va_start(args, fmt);
 783         vsprintf(line, fmt, args);
 784         va_end(args);
 785         pager_output(line);
 786 }
 787
 788 #endif
 789
 790 #define STATUS_FORMAT   "        %s %s\n"
 791
 792 static void
 793 print_state(int indent, const char *name, vdev_state_t state)
 794 {
 795         int i;
 796         char buf[512];
 797
 798         buf[0] = 0;
 799         for (i = 0; i < indent; i++)
 800                 strcat(buf, "  ");
 801         strcat(buf, name);
 802         pager_printf(STATUS_FORMAT, buf, state_name(state));
 803
 804 }
 805
 806 static void
 807 vdev_status(vdev_t *vdev, int indent)
 808 {
 809         vdev_t *kid;
 810         print_state(indent, vdev->v_name, vdev->v_state);
 811
 812         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 813                 vdev_status(kid, indent + 1);
 814         }
 815 }
 816
 817 static void
 818 spa_status(spa_t *spa)
 819 {
 820         static char bootfs[ZFS_MAXNAMELEN];
 821         uint64_t rootid;
 822         vdev_t *vdev;
 823         int good_kids, bad_kids, degraded_kids;
 824         vdev_state_t state;
 825
 826         pager_printf("  pool: %s\n", spa->spa_name);
 827         if (zfs_get_root(spa, &rootid) == 0 &&
 828             zfs_rlookup(spa, rootid, bootfs) == 0) {
 829                 if (bootfs[0] == '\0')
 830                         pager_printf("bootfs: %s\n", spa->spa_name);
 831                 else
 832                         pager_printf("bootfs: %s/%s\n", spa->spa_name, bootfs);
 833         }
 834         pager_printf("config:\n\n");
 835         pager_printf(STATUS_FORMAT, "NAME", "STATE");
 836
 837         good_kids = 0;
 838         degraded_kids = 0;
 839         bad_kids = 0;
 840         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 841                 if (vdev->v_state == VDEV_STATE_HEALTHY)
 842                         good_kids++;
 843                 else if (vdev->v_state == VDEV_STATE_DEGRADED)
 844                         degraded_kids++;
 845                 else
 846                         bad_kids++;
 847         }
 848
 849         state = VDEV_STATE_CLOSED;
 850         if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
 851                 state = VDEV_STATE_HEALTHY;
 852         else if ((good_kids + degraded_kids) > 0)
 853                 state = VDEV_STATE_DEGRADED;
 854
 855         print_state(0, spa->spa_name, state);
 856         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 857                 vdev_status(vdev, 1);
 858         }
 859 }
 860
 861 static void
 862 spa_all_status(void)
 863 {
 864         spa_t *spa;
 865         int first = 1;
 866
 867         STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
 868                 if (!first)
 869                         pager_printf("\n");
 870                 first = 0;
 871                 spa_status(spa);
 872         }
 873 }
 874
 875 static int
 876 vdev_probe(vdev_phys_read_t *read, void *read_priv, spa_t **spap)
 877 {
 878         vdev_t vtmp;
 879         vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
 880         spa_t *spa;
 881         vdev_t *vdev, *top_vdev, *pool_vdev;
 882         off_t off;
 883         blkptr_t bp;
 884         const unsigned char *nvlist;
 885         uint64_t val;
 886         uint64_t guid;
 887         uint64_t pool_txg, pool_guid;
 888         uint64_t is_log;
 889         const char *pool_name;
 890         const unsigned char *vdevs;
 891         const unsigned char *features;
 892         int i, rc, is_newer;
 893         char *upbuf;
 894         const struct uberblock *up;
 895
 896         /*
 897          * Load the vdev label and figure out which
 898          * uberblock is most current.
 899          */
 900         memset(&vtmp, 0, sizeof(vtmp));
 901         vtmp.v_phys_read = read;
 902         vtmp.v_read_priv = read_priv;
 903         off = offsetof(vdev_label_t, vl_vdev_phys);
 904         BP_ZERO(&bp);
 905         BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
 906         BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
 907         BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
 908         BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
 909         DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
 910         ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
 911         if (vdev_read_phys(&vtmp, &bp, vdev_label, off, 0))
 912                 return (EIO);
 913
 914         if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR) {
 915                 return (EIO);
 916         }
 917
 918         nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
 919
 920         if (nvlist_find(nvlist,
 921                         ZPOOL_CONFIG_VERSION,
 922                         DATA_TYPE_UINT64, 0, &val)) {
 923                 return (EIO);
 924         }
 925
 926         if (!SPA_VERSION_IS_SUPPORTED(val)) {
 927                 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
 928                     (unsigned) val, (unsigned) SPA_VERSION);
 929                 return (EIO);
 930         }
 931
 932         /* Check ZFS features for read */
 933         if (nvlist_find(nvlist,
 934                         ZPOOL_CONFIG_FEATURES_FOR_READ,
 935                         DATA_TYPE_NVLIST, 0, &features) == 0
 936             && nvlist_check_features_for_read(features) != 0)
 937                 return (EIO);
 938
 939         if (nvlist_find(nvlist,
 940                         ZPOOL_CONFIG_POOL_STATE,
 941                         DATA_TYPE_UINT64, 0, &val)) {
 942                 return (EIO);
 943         }
 944
 945         if (val == POOL_STATE_DESTROYED) {
 946                 /* We don't boot only from destroyed pools. */
 947                 return (EIO);
 948         }
 949
 950         if (nvlist_find(nvlist,
 951                         ZPOOL_CONFIG_POOL_TXG,
 952                         DATA_TYPE_UINT64, 0, &pool_txg)
 953             || nvlist_find(nvlist,
 954                            ZPOOL_CONFIG_POOL_GUID,
 955                            DATA_TYPE_UINT64, 0, &pool_guid)
 956             || nvlist_find(nvlist,
 957                            ZPOOL_CONFIG_POOL_NAME,
 958                            DATA_TYPE_STRING, 0, &pool_name)) {
 959                 /*
 960                  * Cache and spare devices end up here - just ignore
 961                  * them.
 962                  */
 963                 /*printf("ZFS: can't find pool details\n");*/
 964                 return (EIO);
 965         }
 966
 967         is_log = 0;
 968         (void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, 0,
 969             &is_log);
 970         if (is_log)
 971                 return (EIO);
 972
 973         /*
 974          * Create the pool if this is the first time we've seen it.
 975          */
 976         spa = spa_find_by_guid(pool_guid);
 977         if (!spa) {
 978                 spa = spa_create(pool_guid);
 979                 spa->spa_name = strdup(pool_name);
 980         }
 981         if (pool_txg > spa->spa_txg) {
 982                 spa->spa_txg = pool_txg;
 983                 is_newer = 1;
 984         } else
 985                 is_newer = 0;
 986
 987         /*
 988          * Get the vdev tree and create our in-core copy of it.
 989          * If we already have a vdev with this guid, this must
 990          * be some kind of alias (overlapping slices, dangerously dedicated
 991          * disks etc).
 992          */
 993         if (nvlist_find(nvlist,
 994                         ZPOOL_CONFIG_GUID,
 995                         DATA_TYPE_UINT64, 0, &guid)) {
 996                 return (EIO);
 997         }
 998         vdev = vdev_find(guid);
 999         if (vdev && vdev->v_phys_read)  /* Has this vdev already been inited? */
1000                 return (EIO);
1001
1002         if (nvlist_find(nvlist,
1003                         ZPOOL_CONFIG_VDEV_TREE,
1004                         DATA_TYPE_NVLIST, 0, &vdevs)) {
1005                 return (EIO);
1006         }
1007
1008         rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1009         if (rc)
1010                 return (rc);
1011
1012         /*
1013          * Add the toplevel vdev to the pool if its not already there.
1014          */
1015         STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1016                 if (top_vdev == pool_vdev)
1017                         break;
1018         if (!pool_vdev && top_vdev)
1019                 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1020
1021         /*
1022          * We should already have created an incomplete vdev for this
1023          * vdev. Find it and initialise it with our read proc.
1024          */
1025         vdev = vdev_find(guid);
1026         if (vdev) {
1027                 vdev->v_phys_read = read;
1028                 vdev->v_read_priv = read_priv;
1029                 vdev->v_state = VDEV_STATE_HEALTHY;
1030         } else {
1031                 printf("ZFS: inconsistent nvlist contents\n");
1032                 return (EIO);
1033         }
1034
1035         /*
1036          * Re-evaluate top-level vdev state.
1037          */
1038         vdev_set_state(top_vdev);
1039
1040         /*
1041          * Ok, we are happy with the pool so far. Lets find
1042          * the best uberblock and then we can actually access
1043          * the contents of the pool.
1044          */
1045         upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
1046         up = (const struct uberblock *)upbuf;
1047         for (i = 0;
1048              i < VDEV_UBERBLOCK_COUNT(vdev);
1049              i++) {
1050                 off = VDEV_UBERBLOCK_OFFSET(vdev, i);
1051                 BP_ZERO(&bp);
1052                 DVA_SET_OFFSET(&bp.blk_dva[0], off);
1053                 BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1054                 BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1055                 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1056                 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1057                 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1058
1059                 if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
1060                         continue;
1061
1062                 if (up->ub_magic != UBERBLOCK_MAGIC)
1063                         continue;
1064                 if (up->ub_txg < spa->spa_txg)
1065                         continue;
1066                 if (up->ub_txg > spa->spa_uberblock.ub_txg) {
1067                         spa->spa_uberblock = *up;
1068                 } else if (up->ub_txg == spa->spa_uberblock.ub_txg) {
1069                         if (up->ub_timestamp > spa->spa_uberblock.ub_timestamp)
1070                                 spa->spa_uberblock = *up;
1071                 }
1072         }
1073         zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
1074
1075         if (spap)
1076                 *spap = spa;
1077         return (0);
1078 }
1079
1080 static int
1081 ilog2(int n)
1082 {
1083         int v;
1084
1085         for (v = 0; v < 32; v++)
1086                 if (n == (1 << v))
1087                         return v;
1088         return -1;
1089 }
1090
1091 static int
1092 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1093 {
1094         blkptr_t gbh_bp;
1095         zio_gbh_phys_t zio_gb;
1096         char *pbuf;
1097         int i;
1098
1099         /* Artificial BP for gang block header. */
1100         gbh_bp = *bp;
1101         BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1102         BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1103         BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1104         BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1105         for (i = 0; i < SPA_DVAS_PER_BP; i++)
1106                 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1107
1108         /* Read gang header block using the artificial BP. */
1109         if (zio_read(spa, &gbh_bp, &zio_gb))
1110                 return (EIO);
1111
1112         pbuf = buf;
1113         for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1114                 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1115
1116                 if (BP_IS_HOLE(gbp))
1117                         continue;
1118                 if (zio_read(spa, gbp, pbuf))
1119                         return (EIO);
1120                 pbuf += BP_GET_PSIZE(gbp);
1121         }
1122
1123         if (zio_checksum_verify(bp, buf))
1124                 return (EIO);
1125         return (0);
1126 }
1127
1128 static int
1129 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1130 {
1131         int cpfunc = BP_GET_COMPRESS(bp);
1132         uint64_t align, size;
1133         void *pbuf;
1134         int i, error;
1135
1136         error = EIO;
1137
1138         for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1139                 const dva_t *dva = &bp->blk_dva[i];
1140                 vdev_t *vdev;
1141                 int vdevid;
1142                 off_t offset;
1143
1144                 if (!dva->dva_word[0] && !dva->dva_word[1])
1145                         continue;
1146
1147                 vdevid = DVA_GET_VDEV(dva);
1148                 offset = DVA_GET_OFFSET(dva);
1149                 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1150                         if (vdev->v_id == vdevid)
1151                                 break;
1152                 }
1153                 if (!vdev || !vdev->v_read)
1154                         continue;
1155
1156                 size = BP_GET_PSIZE(bp);
1157                 if (vdev->v_read == vdev_raidz_read) {
1158                         align = 1ULL << vdev->v_top->v_ashift;
1159                         if (P2PHASE(size, align) != 0)
1160                                 size = P2ROUNDUP(size, align);
1161                 }
1162                 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1163                         pbuf = zfs_alloc(size);
1164                 else
1165                         pbuf = buf;
1166
1167                 if (DVA_GET_GANG(dva))
1168                         error = zio_read_gang(spa, bp, pbuf);
1169                 else
1170                         error = vdev->v_read(vdev, bp, pbuf, offset, size);
1171                 if (error == 0) {
1172                         if (cpfunc != ZIO_COMPRESS_OFF)
1173                                 error = zio_decompress_data(cpfunc, pbuf,
1174                                     BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1175                         else if (size != BP_GET_PSIZE(bp))
1176                                 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1177                 }
1178                 if (buf != pbuf)
1179                         zfs_free(pbuf, size);
1180                 if (error == 0)
1181                         break;
1182         }
1183         if (error != 0)
1184                 printf("ZFS: i/o error - all block copies unavailable\n");
1185         return (error);
1186 }
1187
1188 static int
1189 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1190 {
1191         int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1192         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1193         int nlevels = dnode->dn_nlevels;
1194         int i, rc;
1195
1196         /*
1197          * Note: bsize may not be a power of two here so we need to do an
1198          * actual divide rather than a bitshift.
1199          */
1200         while (buflen > 0) {
1201                 uint64_t bn = offset / bsize;
1202                 int boff = offset % bsize;
1203                 int ibn;
1204                 const blkptr_t *indbp;
1205                 blkptr_t bp;
1206
1207                 if (bn > dnode->dn_maxblkid)
1208                         return (EIO);
1209
1210                 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1211                         goto cached;
1212
1213                 indbp = dnode->dn_blkptr;
1214                 for (i = 0; i < nlevels; i++) {
1215                         /*
1216                          * Copy the bp from the indirect array so that
1217                          * we can re-use the scratch buffer for multi-level
1218                          * objects.
1219                          */
1220                         ibn = bn >> ((nlevels - i - 1) * ibshift);
1221                         ibn &= ((1 << ibshift) - 1);
1222                         bp = indbp[ibn];
1223                         rc = zio_read(spa, &bp, dnode_cache_buf);
1224                         if (rc)
1225                                 return (rc);
1226                         indbp = (const blkptr_t *) dnode_cache_buf;
1227                 }
1228                 dnode_cache_obj = dnode;
1229                 dnode_cache_bn = bn;
1230         cached:
1231
1232                 /*
1233                  * The buffer contains our data block. Copy what we
1234                  * need from it and loop.
1235                  */
1236                 i = bsize - boff;
1237                 if (i > buflen) i = buflen;
1238                 memcpy(buf, &dnode_cache_buf[boff], i);
1239                 buf = ((char*) buf) + i;
1240                 offset += i;
1241                 buflen -= i;
1242         }
1243
1244         return (0);
1245 }
1246
1247 /*
1248  * Lookup a value in a microzap directory. Assumes that the zap
1249  * scratch buffer contains the directory contents.
1250  */
1251 static int
1252 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1253 {
1254         const mzap_phys_t *mz;
1255         const mzap_ent_phys_t *mze;
1256         size_t size;
1257         int chunks, i;
1258
1259         /*
1260          * Microzap objects use exactly one block. Read the whole
1261          * thing.
1262          */
1263         size = dnode->dn_datablkszsec * 512;
1264
1265         mz = (const mzap_phys_t *) zap_scratch;
1266         chunks = size / MZAP_ENT_LEN - 1;
1267
1268         for (i = 0; i < chunks; i++) {
1269                 mze = &mz->mz_chunk[i];
1270                 if (!strcmp(mze->mze_name, name)) {
1271                         *value = mze->mze_value;
1272                         return (0);
1273                 }
1274         }
1275
1276         return (ENOENT);
1277 }
1278
1279 /*
1280  * Compare a name with a zap leaf entry. Return non-zero if the name
1281  * matches.
1282  */
1283 static int
1284 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1285 {
1286         size_t namelen;
1287         const zap_leaf_chunk_t *nc;
1288         const char *p;
1289
1290         namelen = zc->l_entry.le_name_numints;
1291
1292         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1293         p = name;
1294         while (namelen > 0) {
1295                 size_t len;
1296                 len = namelen;
1297                 if (len > ZAP_LEAF_ARRAY_BYTES)
1298                         len = ZAP_LEAF_ARRAY_BYTES;
1299                 if (memcmp(p, nc->l_array.la_array, len))
1300                         return (0);
1301                 p += len;
1302                 namelen -= len;
1303                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1304         }
1305
1306         return 1;
1307 }
1308
1309 /*
1310  * Extract a uint64_t value from a zap leaf entry.
1311  */
1312 static uint64_t
1313 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1314 {
1315         const zap_leaf_chunk_t *vc;
1316         int i;
1317         uint64_t value;
1318         const uint8_t *p;
1319
1320         vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1321         for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1322                 value = (value << 8) | p[i];
1323         }
1324
1325         return value;
1326 }
1327
1328 /*
1329  * Lookup a value in a fatzap directory. Assumes that the zap scratch
1330  * buffer contains the directory header.
1331  */
1332 static int
1333 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1334 {
1335         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1336         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1337         fat_zap_t z;
1338         uint64_t *ptrtbl;
1339         uint64_t hash;
1340         int rc;
1341
1342         if (zh.zap_magic != ZAP_MAGIC)
1343                 return (EIO);
1344
1345         z.zap_block_shift = ilog2(bsize);
1346         z.zap_phys = (zap_phys_t *) zap_scratch;
1347
1348         /*
1349          * Figure out where the pointer table is and read it in if necessary.
1350          */
1351         if (zh.zap_ptrtbl.zt_blk) {
1352                 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1353                                zap_scratch, bsize);
1354                 if (rc)
1355                         return (rc);
1356                 ptrtbl = (uint64_t *) zap_scratch;
1357         } else {
1358                 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1359         }
1360
1361         hash = zap_hash(zh.zap_salt, name);
1362
1363         zap_leaf_t zl;
1364         zl.l_bs = z.zap_block_shift;
1365
1366         off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1367         zap_leaf_chunk_t *zc;
1368
1369         rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1370         if (rc)
1371                 return (rc);
1372
1373         zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1374
1375         /*
1376          * Make sure this chunk matches our hash.
1377          */
1378         if (zl.l_phys->l_hdr.lh_prefix_len > 0
1379             && zl.l_phys->l_hdr.lh_prefix
1380             != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1381                 return (ENOENT);
1382
1383         /*
1384          * Hash within the chunk to find our entry.
1385          */
1386         int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1387         int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1388         h = zl.l_phys->l_hash[h];
1389         if (h == 0xffff)
1390                 return (ENOENT);
1391         zc = &ZAP_LEAF_CHUNK(&zl, h);
1392         while (zc->l_entry.le_hash != hash) {
1393                 if (zc->l_entry.le_next == 0xffff) {
1394                         zc = 0;
1395                         break;
1396                 }
1397                 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1398         }
1399         if (fzap_name_equal(&zl, zc, name)) {
1400                 if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints > 8)
1401                         return (E2BIG);
1402                 *value = fzap_leaf_value(&zl, zc);
1403                 return (0);
1404         }
1405
1406         return (ENOENT);
1407 }
1408
1409 /*
1410  * Lookup a name in a zap object and return its value as a uint64_t.
1411  */
1412 static int
1413 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1414 {
1415         int rc;
1416         uint64_t zap_type;
1417         size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1418
1419         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1420         if (rc)
1421                 return (rc);
1422
1423         zap_type = *(uint64_t *) zap_scratch;
1424         if (zap_type == ZBT_MICRO)
1425                 return mzap_lookup(dnode, name, value);
1426         else if (zap_type == ZBT_HEADER)
1427                 return fzap_lookup(spa, dnode, name, value);
1428         printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
1429         return (EIO);
1430 }
1431
1432 /*
1433  * List a microzap directory. Assumes that the zap scratch buffer contains
1434  * the directory contents.
1435  */
1436 static int
1437 mzap_list(const dnode_phys_t *dnode)
1438 {
1439         const mzap_phys_t *mz;
1440         const mzap_ent_phys_t *mze;
1441         size_t size;
1442         int chunks, i;
1443
1444         /*
1445          * Microzap objects use exactly one block. Read the whole
1446          * thing.
1447          */
1448         size = dnode->dn_datablkszsec * 512;
1449         mz = (const mzap_phys_t *) zap_scratch;
1450         chunks = size / MZAP_ENT_LEN - 1;
1451
1452         for (i = 0; i < chunks; i++) {
1453                 mze = &mz->mz_chunk[i];
1454                 if (mze->mze_name[0])
1455                         //printf("%-32s 0x%jx\n", mze->mze_name, (uintmax_t)mze->mze_value);
1456                         printf("%s\n", mze->mze_name);
1457         }
1458
1459         return (0);
1460 }
1461
1462 /*
1463  * List a fatzap directory. Assumes that the zap scratch buffer contains
1464  * the directory header.
1465  */
1466 static int
1467 fzap_list(const spa_t *spa, const dnode_phys_t *dnode)
1468 {
1469         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1470         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1471         fat_zap_t z;
1472         int i, j;
1473
1474         if (zh.zap_magic != ZAP_MAGIC)
1475                 return (EIO);
1476
1477         z.zap_block_shift = ilog2(bsize);
1478         z.zap_phys = (zap_phys_t *) zap_scratch;
1479
1480         /*
1481          * This assumes that the leaf blocks start at block 1. The
1482          * documentation isn't exactly clear on this.
1483          */
1484         zap_leaf_t zl;
1485         zl.l_bs = z.zap_block_shift;
1486         for (i = 0; i < zh.zap_num_leafs; i++) {
1487                 off_t off = (i + 1) << zl.l_bs;
1488                 char name[256], *p;
1489                 uint64_t value;
1490
1491                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1492                         return (EIO);
1493
1494                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1495
1496                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1497                         zap_leaf_chunk_t *zc, *nc;
1498                         int namelen;
1499
1500                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1501                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1502                                 continue;
1503                         namelen = zc->l_entry.le_name_numints;
1504                         if (namelen > sizeof(name))
1505                                 namelen = sizeof(name);
1506
1507                         /*
1508                          * Paste the name back together.
1509                          */
1510                         nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1511                         p = name;
1512                         while (namelen > 0) {
1513                                 int len;
1514                                 len = namelen;
1515                                 if (len > ZAP_LEAF_ARRAY_BYTES)
1516                                         len = ZAP_LEAF_ARRAY_BYTES;
1517                                 memcpy(p, nc->l_array.la_array, len);
1518                                 p += len;
1519                                 namelen -= len;
1520                                 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1521                         }
1522
1523                         /*
1524                          * Assume the first eight bytes of the value are
1525                          * a uint64_t.
1526                          */
1527                         value = fzap_leaf_value(&zl, zc);
1528
1529                         //printf("%s 0x%jx\n", name, (uintmax_t)value);
1530                         printf("%s\n", name);
1531                 }
1532         }
1533
1534         return (0);
1535 }
1536
1537 /*
1538  * List a zap directory.
1539  */
1540 static int
1541 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
1542 {
1543         uint64_t zap_type;
1544         size_t size = dnode->dn_datablkszsec * 512;
1545
1546         if (dnode_read(spa, dnode, 0, zap_scratch, size))
1547                 return (EIO);
1548
1549         zap_type = *(uint64_t *) zap_scratch;
1550         if (zap_type == ZBT_MICRO)
1551                 return mzap_list(dnode);
1552         else
1553                 return fzap_list(spa, dnode);
1554 }
1555
1556 static int
1557 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1558 {
1559         off_t offset;
1560
1561         offset = objnum * sizeof(dnode_phys_t);
1562         return dnode_read(spa, &os->os_meta_dnode, offset,
1563                 dnode, sizeof(dnode_phys_t));
1564 }
1565
1566 static int
1567 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1568 {
1569         const mzap_phys_t *mz;
1570         const mzap_ent_phys_t *mze;
1571         size_t size;
1572         int chunks, i;
1573
1574         /*
1575          * Microzap objects use exactly one block. Read the whole
1576          * thing.
1577          */
1578         size = dnode->dn_datablkszsec * 512;
1579
1580         mz = (const mzap_phys_t *) zap_scratch;
1581         chunks = size / MZAP_ENT_LEN - 1;
1582
1583         for (i = 0; i < chunks; i++) {
1584                 mze = &mz->mz_chunk[i];
1585                 if (value == mze->mze_value) {
1586                         strcpy(name, mze->mze_name);
1587                         return (0);
1588                 }
1589         }
1590
1591         return (ENOENT);
1592 }
1593
1594 static void
1595 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
1596 {
1597         size_t namelen;
1598         const zap_leaf_chunk_t *nc;
1599         char *p;
1600
1601         namelen = zc->l_entry.le_name_numints;
1602
1603         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1604         p = name;
1605         while (namelen > 0) {
1606                 size_t len;
1607                 len = namelen;
1608                 if (len > ZAP_LEAF_ARRAY_BYTES)
1609                         len = ZAP_LEAF_ARRAY_BYTES;
1610                 memcpy(p, nc->l_array.la_array, len);
1611                 p += len;
1612                 namelen -= len;
1613                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1614         }
1615
1616         *p = '\0';
1617 }
1618
1619 static int
1620 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1621 {
1622         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1623         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1624         fat_zap_t z;
1625         int i, j;
1626
1627         if (zh.zap_magic != ZAP_MAGIC)
1628                 return (EIO);
1629
1630         z.zap_block_shift = ilog2(bsize);
1631         z.zap_phys = (zap_phys_t *) zap_scratch;
1632
1633         /*
1634          * This assumes that the leaf blocks start at block 1. The
1635          * documentation isn't exactly clear on this.
1636          */
1637         zap_leaf_t zl;
1638         zl.l_bs = z.zap_block_shift;
1639         for (i = 0; i < zh.zap_num_leafs; i++) {
1640                 off_t off = (i + 1) << zl.l_bs;
1641
1642                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1643                         return (EIO);
1644
1645                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1646
1647                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1648                         zap_leaf_chunk_t *zc;
1649
1650                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1651                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1652                                 continue;
1653                         if (zc->l_entry.le_value_intlen != 8 ||
1654                             zc->l_entry.le_value_numints != 1)
1655                                 continue;
1656
1657                         if (fzap_leaf_value(&zl, zc) == value) {
1658                                 fzap_name_copy(&zl, zc, name);
1659                                 return (0);
1660                         }
1661                 }
1662         }
1663
1664         return (ENOENT);
1665 }
1666
1667 static int
1668 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1669 {
1670         int rc;
1671         uint64_t zap_type;
1672         size_t size = dnode->dn_datablkszsec * 512;
1673
1674         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1675         if (rc)
1676                 return (rc);
1677
1678         zap_type = *(uint64_t *) zap_scratch;
1679         if (zap_type == ZBT_MICRO)
1680                 return mzap_rlookup(spa, dnode, name, value);
1681         else
1682                 return fzap_rlookup(spa, dnode, name, value);
1683 }
1684
1685 static int
1686 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
1687 {
1688         char name[256];
1689         char component[256];
1690         uint64_t dir_obj, parent_obj, child_dir_zapobj;
1691         dnode_phys_t child_dir_zap, dataset, dir, parent;
1692         dsl_dir_phys_t *dd;
1693         dsl_dataset_phys_t *ds;
1694         char *p;
1695         int len;
1696
1697         p = &name[sizeof(name) - 1];
1698         *p = '\0';
1699
1700         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1701                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1702                 return (EIO);
1703         }
1704         ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
1705         dir_obj = ds->ds_dir_obj;
1706
1707         for (;;) {
1708                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
1709                         return (EIO);
1710                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1711
1712                 /* Actual loop condition. */
1713                 parent_obj  = dd->dd_parent_obj;
1714                 if (parent_obj == 0)
1715                         break;
1716
1717                 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
1718                         return (EIO);
1719                 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
1720                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1721                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1722                         return (EIO);
1723                 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
1724                         return (EIO);
1725
1726                 len = strlen(component);
1727                 p -= len;
1728                 memcpy(p, component, len);
1729                 --p;
1730                 *p = '/';
1731
1732                 /* Actual loop iteration. */
1733                 dir_obj = parent_obj;
1734         }
1735
1736         if (*p != '\0')
1737                 ++p;
1738         strcpy(result, p);
1739
1740         return (0);
1741 }
1742
1743 static int
1744 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
1745 {
1746         char element[256];
1747         uint64_t dir_obj, child_dir_zapobj;
1748         dnode_phys_t child_dir_zap, dir;
1749         dsl_dir_phys_t *dd;
1750         const char *p, *q;
1751
1752         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
1753                 return (EIO);
1754         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &dir_obj))
1755                 return (EIO);
1756
1757         p = name;
1758         for (;;) {
1759                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
1760                         return (EIO);
1761                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1762
1763                 while (*p == '/')
1764                         p++;
1765                 /* Actual loop condition #1. */
1766                 if (*p == '\0')
1767                         break;
1768
1769                 q = strchr(p, '/');
1770                 if (q) {
1771                         memcpy(element, p, q - p);
1772                         element[q - p] = '\0';
1773                         p = q + 1;
1774                 } else {
1775                         strcpy(element, p);
1776                         p += strlen(p);
1777                 }
1778
1779                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1780                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1781                         return (EIO);
1782
1783                 /* Actual loop condition #2. */
1784                 if (zap_lookup(spa, &child_dir_zap, element, &dir_obj) != 0)
1785                         return (ENOENT);
1786         }
1787
1788         *objnum = dd->dd_head_dataset_obj;
1789         return (0);
1790 }
1791
1792 #ifndef BOOT2
1793 static int
1794 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
1795 {
1796         uint64_t dir_obj, child_dir_zapobj;
1797         dnode_phys_t child_dir_zap, dir, dataset;
1798         dsl_dataset_phys_t *ds;
1799         dsl_dir_phys_t *dd;
1800
1801         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1802                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1803                 return (EIO);
1804         }
1805         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1806         dir_obj = ds->ds_dir_obj;
1807
1808         if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
1809                 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
1810                 return (EIO);
1811         }
1812         dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1813
1814         child_dir_zapobj = dd->dd_child_dir_zapobj;
1815         if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
1816                 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
1817                 return (EIO);
1818         }
1819
1820         return (zap_list(spa, &child_dir_zap) != 0);
1821 }
1822 #endif
1823
1824 /*
1825  * Find the object set given the object number of its dataset object
1826  * and return its details in *objset
1827  */
1828 static int
1829 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
1830 {
1831         dnode_phys_t dataset;
1832         dsl_dataset_phys_t *ds;
1833
1834         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1835                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1836                 return (EIO);
1837         }
1838
1839         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1840         if (zio_read(spa, &ds->ds_bp, objset)) {
1841                 printf("ZFS: can't read object set for dataset %ju\n",
1842                     (uintmax_t)objnum);
1843                 return (EIO);
1844         }
1845
1846         return (0);
1847 }
1848
1849 /*
1850  * Find the object set pointed to by the BOOTFS property or the root
1851  * dataset if there is none and return its details in *objset
1852  */
1853 static int
1854 zfs_get_root(const spa_t *spa, uint64_t *objid)
1855 {
1856         dnode_phys_t dir, propdir;
1857         uint64_t props, bootfs, root;
1858
1859         *objid = 0;
1860
1861         /*
1862          * Start with the MOS directory object.
1863          */
1864         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
1865                 printf("ZFS: can't read MOS object directory\n");
1866                 return (EIO);
1867         }
1868
1869         /*
1870          * Lookup the pool_props and see if we can find a bootfs.
1871          */
1872         if (zap_lookup(spa, &dir, DMU_POOL_PROPS, &props) == 0
1873              && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
1874              && zap_lookup(spa, &propdir, "bootfs", &bootfs) == 0
1875              && bootfs != 0)
1876         {
1877                 *objid = bootfs;
1878                 return (0);
1879         }
1880         /*
1881          * Lookup the root dataset directory
1882          */
1883         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &root)
1884             || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
1885                 printf("ZFS: can't find root dsl_dir\n");
1886                 return (EIO);
1887         }
1888
1889         /*
1890          * Use the information from the dataset directory's bonus buffer
1891          * to find the dataset object and from that the object set itself.
1892          */
1893         dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
1894         *objid = dd->dd_head_dataset_obj;
1895         return (0);
1896 }
1897
1898 static int
1899 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
1900 {
1901
1902         mount->spa = spa;
1903
1904         /*
1905          * Find the root object set if not explicitly provided
1906          */
1907         if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
1908                 printf("ZFS: can't find root filesystem\n");
1909                 return (EIO);
1910         }
1911
1912         if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
1913                 printf("ZFS: can't open root filesystem\n");
1914                 return (EIO);
1915         }
1916
1917         mount->rootobj = rootobj;
1918
1919         return (0);
1920 }
1921
1922 static int
1923 zfs_spa_init(spa_t *spa)
1924 {
1925
1926         if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
1927                 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
1928                 return (EIO);
1929         }
1930         if (spa->spa_mos.os_type != DMU_OST_META) {
1931                 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
1932                 return (EIO);
1933         }
1934         return (0);
1935 }
1936
1937 static int
1938 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
1939 {
1940
1941         if (dn->dn_bonustype != DMU_OT_SA) {
1942                 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
1943
1944                 sb->st_mode = zp->zp_mode;
1945                 sb->st_uid = zp->zp_uid;
1946                 sb->st_gid = zp->zp_gid;
1947                 sb->st_size = zp->zp_size;
1948         } else {
1949                 sa_hdr_phys_t *sahdrp;
1950                 int hdrsize;
1951                 size_t size = 0;
1952                 void *buf = NULL;
1953
1954                 if (dn->dn_bonuslen != 0)
1955                         sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
1956                 else {
1957                         if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
1958                                 blkptr_t *bp = &dn->dn_spill;
1959                                 int error;
1960
1961                                 size = BP_GET_LSIZE(bp);
1962                                 buf = zfs_alloc(size);
1963                                 error = zio_read(spa, bp, buf);
1964                                 if (error != 0) {
1965                                         zfs_free(buf, size);
1966                                         return (error);
1967                                 }
1968                                 sahdrp = buf;
1969                         } else {
1970                                 return (EIO);
1971                         }
1972                 }
1973                 hdrsize = SA_HDR_SIZE(sahdrp);
1974                 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
1975                     SA_MODE_OFFSET);
1976                 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
1977                     SA_UID_OFFSET);
1978                 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
1979                     SA_GID_OFFSET);
1980                 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
1981                     SA_SIZE_OFFSET);
1982                 if (buf != NULL)
1983                         zfs_free(buf, size);
1984         }
1985
1986         return (0);
1987 }
1988
1989 /*
1990  * Lookup a file and return its dnode.
1991  */
1992 static int
1993 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
1994 {
1995         int rc;
1996         uint64_t objnum, rootnum, parentnum;
1997         const spa_t *spa;
1998         dnode_phys_t dn;
1999         const char *p, *q;
2000         char element[256];
2001         char path[1024];
2002         int symlinks_followed = 0;
2003         struct stat sb;
2004
2005         spa = mount->spa;
2006         if (mount->objset.os_type != DMU_OST_ZFS) {
2007                 printf("ZFS: unexpected object set type %ju\n",
2008                     (uintmax_t)mount->objset.os_type);
2009                 return (EIO);
2010         }
2011
2012         /*
2013          * Get the root directory dnode.
2014          */
2015         rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
2016         if (rc)
2017                 return (rc);
2018
2019         rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, &rootnum);
2020         if (rc)
2021                 return (rc);
2022
2023         rc = objset_get_dnode(spa, &mount->objset, rootnum, &dn);
2024         if (rc)
2025                 return (rc);
2026
2027         objnum = rootnum;
2028         p = upath;
2029         while (p && *p) {
2030                 while (*p == '/')
2031                         p++;
2032                 if (!*p)
2033                         break;
2034                 q = strchr(p, '/');
2035                 if (q) {
2036                         memcpy(element, p, q - p);
2037                         element[q - p] = 0;
2038                         p = q;
2039                 } else {
2040                         strcpy(element, p);
2041                         p = 0;
2042                 }
2043
2044                 rc = zfs_dnode_stat(spa, &dn, &sb);
2045                 if (rc)
2046                         return (rc);
2047                 if (!S_ISDIR(sb.st_mode))
2048                         return (ENOTDIR);
2049
2050                 parentnum = objnum;
2051                 rc = zap_lookup(spa, &dn, element, &objnum);
2052                 if (rc)
2053                         return (rc);
2054                 objnum = ZFS_DIRENT_OBJ(objnum);
2055
2056                 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
2057                 if (rc)
2058                         return (rc);
2059
2060                 /*
2061                  * Check for symlink.
2062                  */
2063                 rc = zfs_dnode_stat(spa, &dn, &sb);
2064                 if (rc)
2065                         return (rc);
2066                 if (S_ISLNK(sb.st_mode)) {
2067                         if (symlinks_followed > 10)
2068                                 return (EMLINK);
2069                         symlinks_followed++;
2070
2071                         /*
2072                          * Read the link value and copy the tail of our
2073                          * current path onto the end.
2074                          */
2075                         if (p)
2076                                 strcpy(&path[sb.st_size], p);
2077                         else
2078                                 path[sb.st_size] = 0;
2079                         if (sb.st_size + sizeof(znode_phys_t) <= dn.dn_bonuslen) {
2080                                 memcpy(path, &dn.dn_bonus[sizeof(znode_phys_t)],
2081                                         sb.st_size);
2082                         } else {
2083                                 rc = dnode_read(spa, &dn, 0, path, sb.st_size);
2084                                 if (rc)
2085                                         return (rc);
2086                         }
2087
2088                         /*
2089                          * Restart with the new path, starting either at
2090                          * the root or at the parent depending whether or
2091                          * not the link is relative.
2092                          */
2093                         p = path;
2094                         if (*p == '/')
2095                                 objnum = rootnum;
2096                         else
2097                                 objnum = parentnum;
2098                         objset_get_dnode(spa, &mount->objset, objnum, &dn);
2099                 }
2100         }
2101
2102         *dnode = dn;
2103         return (0);
2104 }