stand/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 static struct zfsmount zfsmount __unused;
  47
  48 /*
  49  * List of all vdevs, chained through v_alllink.
  50  */
  51 static vdev_list_t zfs_vdevs;
  52
  53  /*
  54  * List of ZFS features supported for read
  55  */
  56 static const char *features_for_read[] = {
  57         "org.illumos:lz4_compress",
  58         "com.delphix:hole_birth",
  59         "com.delphix:extensible_dataset",
  60         "com.delphix:embedded_data",
  61         "org.open-zfs:large_blocks",
  62         "org.illumos:sha512",
  63         "org.illumos:skein",
  64         "org.zfsonlinux:large_dnode",
  65         NULL
  66 };
  67
  68 /*
  69  * List of all pools, chained through spa_link.
  70  */
  71 static spa_list_t zfs_pools;
  72
  73 static const dnode_phys_t *dnode_cache_obj;
  74 static uint64_t dnode_cache_bn;
  75 static char *dnode_cache_buf;
  76 static char *zap_scratch;
  77 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
  78
  79 #define TEMP_SIZE       (1024 * 1024)
  80
  81 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
  82 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
  83 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
  84 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
  85     const char *name, uint64_t integer_size, uint64_t num_integers,
  86     void *value);
  87
  88 static void
  89 zfs_init(void)
  90 {
  91         STAILQ_INIT(&zfs_vdevs);
  92         STAILQ_INIT(&zfs_pools);
  93
  94         zfs_temp_buf = malloc(TEMP_SIZE);
  95         zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
  96         zfs_temp_ptr = zfs_temp_buf;
  97         dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
  98         zap_scratch = malloc(SPA_MAXBLOCKSIZE);
  99
 100         zfs_init_crc();
 101 }
 102
 103 static void *
 104 zfs_alloc(size_t size)
 105 {
 106         char *ptr;
 107
 108         if (zfs_temp_ptr + size > zfs_temp_end) {
 109                 printf("ZFS: out of temporary buffer space\n");
 110                 for (;;) ;
 111         }
 112         ptr = zfs_temp_ptr;
 113         zfs_temp_ptr += size;
 114
 115         return (ptr);
 116 }
 117
 118 static void
 119 zfs_free(void *ptr, size_t size)
 120 {
 121
 122         zfs_temp_ptr -= size;
 123         if (zfs_temp_ptr != ptr) {
 124                 printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
 125                 for (;;) ;
 126         }
 127 }
 128
 129 static int
 130 xdr_int(const unsigned char **xdr, int *ip)
 131 {
 132         *ip = ((*xdr)[0] << 24)
 133                 | ((*xdr)[1] << 16)
 134                 | ((*xdr)[2] << 8)
 135                 | ((*xdr)[3] << 0);
 136         (*xdr) += 4;
 137         return (0);
 138 }
 139
 140 static int
 141 xdr_u_int(const unsigned char **xdr, u_int *ip)
 142 {
 143         *ip = ((*xdr)[0] << 24)
 144                 | ((*xdr)[1] << 16)
 145                 | ((*xdr)[2] << 8)
 146                 | ((*xdr)[3] << 0);
 147         (*xdr) += 4;
 148         return (0);
 149 }
 150
 151 static int
 152 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
 153 {
 154         u_int hi, lo;
 155
 156         xdr_u_int(xdr, &hi);
 157         xdr_u_int(xdr, &lo);
 158         *lp = (((uint64_t) hi) << 32) | lo;
 159         return (0);
 160 }
 161
 162 static int
 163 nvlist_find(const unsigned char *nvlist, const char *name, int type,
 164             int* elementsp, void *valuep)
 165 {
 166         const unsigned char *p, *pair;
 167         int junk;
 168         int encoded_size, decoded_size;
 169
 170         p = nvlist;
 171         xdr_int(&p, &junk);
 172         xdr_int(&p, &junk);
 173
 174         pair = p;
 175         xdr_int(&p, &encoded_size);
 176         xdr_int(&p, &decoded_size);
 177         while (encoded_size && decoded_size) {
 178                 int namelen, pairtype, elements;
 179                 const char *pairname;
 180
 181                 xdr_int(&p, &namelen);
 182                 pairname = (const char*) p;
 183                 p += roundup(namelen, 4);
 184                 xdr_int(&p, &pairtype);
 185
 186                 if (!memcmp(name, pairname, namelen) && type == pairtype) {
 187                         xdr_int(&p, &elements);
 188                         if (elementsp)
 189                                 *elementsp = elements;
 190                         if (type == DATA_TYPE_UINT64) {
 191                                 xdr_uint64_t(&p, (uint64_t *) valuep);
 192                                 return (0);
 193                         } else if (type == DATA_TYPE_STRING) {
 194                                 int len;
 195                                 xdr_int(&p, &len);
 196                                 (*(const char**) valuep) = (const char*) p;
 197                                 return (0);
 198                         } else if (type == DATA_TYPE_NVLIST
 199                                    || type == DATA_TYPE_NVLIST_ARRAY) {
 200                                 (*(const unsigned char**) valuep) =
 201                                          (const unsigned char*) p;
 202                                 return (0);
 203                         } else {
 204                                 return (EIO);
 205                         }
 206                 } else {
 207                         /*
 208                          * Not the pair we are looking for, skip to the next one.
 209                          */
 210                         p = pair + encoded_size;
 211                 }
 212
 213                 pair = p;
 214                 xdr_int(&p, &encoded_size);
 215                 xdr_int(&p, &decoded_size);
 216         }
 217
 218         return (EIO);
 219 }
 220
 221 static int
 222 nvlist_check_features_for_read(const unsigned char *nvlist)
 223 {
 224         const unsigned char *p, *pair;
 225         int junk;
 226         int encoded_size, decoded_size;
 227         int rc;
 228
 229         rc = 0;
 230
 231         p = nvlist;
 232         xdr_int(&p, &junk);
 233         xdr_int(&p, &junk);
 234
 235         pair = p;
 236         xdr_int(&p, &encoded_size);
 237         xdr_int(&p, &decoded_size);
 238         while (encoded_size && decoded_size) {
 239                 int namelen, pairtype;
 240                 const char *pairname;
 241                 int i, found;
 242
 243                 found = 0;
 244
 245                 xdr_int(&p, &namelen);
 246                 pairname = (const char*) p;
 247                 p += roundup(namelen, 4);
 248                 xdr_int(&p, &pairtype);
 249
 250                 for (i = 0; features_for_read[i] != NULL; i++) {
 251                         if (!memcmp(pairname, features_for_read[i], namelen)) {
 252                                 found = 1;
 253                                 break;
 254                         }
 255                 }
 256
 257                 if (!found) {
 258                         printf("ZFS: unsupported feature: %s\n", pairname);
 259                         rc = EIO;
 260                 }
 261
 262                 p = pair + encoded_size;
 263
 264                 pair = p;
 265                 xdr_int(&p, &encoded_size);
 266                 xdr_int(&p, &decoded_size);
 267         }
 268
 269         return (rc);
 270 }
 271
 272 /*
 273  * Return the next nvlist in an nvlist array.
 274  */
 275 static const unsigned char *
 276 nvlist_next(const unsigned char *nvlist)
 277 {
 278         const unsigned char *p, *pair;
 279         int junk;
 280         int encoded_size, decoded_size;
 281
 282         p = nvlist;
 283         xdr_int(&p, &junk);
 284         xdr_int(&p, &junk);
 285
 286         pair = p;
 287         xdr_int(&p, &encoded_size);
 288         xdr_int(&p, &decoded_size);
 289         while (encoded_size && decoded_size) {
 290                 p = pair + encoded_size;
 291
 292                 pair = p;
 293                 xdr_int(&p, &encoded_size);
 294                 xdr_int(&p, &decoded_size);
 295         }
 296
 297         return p;
 298 }
 299
 300 #ifdef TEST
 301
 302 static const unsigned char *
 303 nvlist_print(const unsigned char *nvlist, unsigned int indent)
 304 {
 305         static const char* typenames[] = {
 306                 "DATA_TYPE_UNKNOWN",
 307                 "DATA_TYPE_BOOLEAN",
 308                 "DATA_TYPE_BYTE",
 309                 "DATA_TYPE_INT16",
 310                 "DATA_TYPE_UINT16",
 311                 "DATA_TYPE_INT32",
 312                 "DATA_TYPE_UINT32",
 313                 "DATA_TYPE_INT64",
 314                 "DATA_TYPE_UINT64",
 315                 "DATA_TYPE_STRING",
 316                 "DATA_TYPE_BYTE_ARRAY",
 317                 "DATA_TYPE_INT16_ARRAY",
 318                 "DATA_TYPE_UINT16_ARRAY",
 319                 "DATA_TYPE_INT32_ARRAY",
 320                 "DATA_TYPE_UINT32_ARRAY",
 321                 "DATA_TYPE_INT64_ARRAY",
 322                 "DATA_TYPE_UINT64_ARRAY",
 323                 "DATA_TYPE_STRING_ARRAY",
 324                 "DATA_TYPE_HRTIME",
 325                 "DATA_TYPE_NVLIST",
 326                 "DATA_TYPE_NVLIST_ARRAY",
 327                 "DATA_TYPE_BOOLEAN_VALUE",
 328                 "DATA_TYPE_INT8",
 329                 "DATA_TYPE_UINT8",
 330                 "DATA_TYPE_BOOLEAN_ARRAY",
 331                 "DATA_TYPE_INT8_ARRAY",
 332                 "DATA_TYPE_UINT8_ARRAY"
 333         };
 334
 335         unsigned int i, j;
 336         const unsigned char *p, *pair;
 337         int junk;
 338         int encoded_size, decoded_size;
 339
 340         p = nvlist;
 341         xdr_int(&p, &junk);
 342         xdr_int(&p, &junk);
 343
 344         pair = p;
 345         xdr_int(&p, &encoded_size);
 346         xdr_int(&p, &decoded_size);
 347         while (encoded_size && decoded_size) {
 348                 int namelen, pairtype, elements;
 349                 const char *pairname;
 350
 351                 xdr_int(&p, &namelen);
 352                 pairname = (const char*) p;
 353                 p += roundup(namelen, 4);
 354                 xdr_int(&p, &pairtype);
 355
 356                 for (i = 0; i < indent; i++)
 357                         printf(" ");
 358                 printf("%s %s", typenames[pairtype], pairname);
 359
 360                 xdr_int(&p, &elements);
 361                 switch (pairtype) {
 362                 case DATA_TYPE_UINT64: {
 363                         uint64_t val;
 364                         xdr_uint64_t(&p, &val);
 365                         printf(" = 0x%jx\n", (uintmax_t)val);
 366                         break;
 367                 }
 368
 369                 case DATA_TYPE_STRING: {
 370                         int len;
 371                         xdr_int(&p, &len);
 372                         printf(" = \"%s\"\n", p);
 373                         break;
 374                 }
 375
 376                 case DATA_TYPE_NVLIST:
 377                         printf("\n");
 378                         nvlist_print(p, indent + 1);
 379                         break;
 380
 381                 case DATA_TYPE_NVLIST_ARRAY:
 382                         for (j = 0; j < elements; j++) {
 383                                 printf("[%d]\n", j);
 384                                 p = nvlist_print(p, indent + 1);
 385                                 if (j != elements - 1) {
 386                                         for (i = 0; i < indent; i++)
 387                                                 printf(" ");
 388                                         printf("%s %s", typenames[pairtype], pairname);
 389                                 }
 390                         }
 391                         break;
 392
 393                 default:
 394                         printf("\n");
 395                 }
 396
 397                 p = pair + encoded_size;
 398
 399                 pair = p;
 400                 xdr_int(&p, &encoded_size);
 401                 xdr_int(&p, &decoded_size);
 402         }
 403
 404         return p;
 405 }
 406
 407 #endif
 408
 409 static int
 410 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
 411     off_t offset, size_t size)
 412 {
 413         size_t psize;
 414         int rc;
 415
 416         if (!vdev->v_phys_read)
 417                 return (EIO);
 418
 419         if (bp) {
 420                 psize = BP_GET_PSIZE(bp);
 421         } else {
 422                 psize = size;
 423         }
 424
 425         /*printf("ZFS: reading %zu bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
 426         rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
 427         if (rc)
 428                 return (rc);
 429         if (bp && zio_checksum_verify(vdev->spa, bp, buf))
 430                 return (EIO);
 431
 432         return (0);
 433 }
 434
 435 static int
 436 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 437     off_t offset, size_t bytes)
 438 {
 439
 440         return (vdev_read_phys(vdev, bp, buf,
 441                 offset + VDEV_LABEL_START_SIZE, bytes));
 442 }
 443
 444
 445 static int
 446 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 447     off_t offset, size_t bytes)
 448 {
 449         vdev_t *kid;
 450         int rc;
 451
 452         rc = EIO;
 453         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 454                 if (kid->v_state != VDEV_STATE_HEALTHY)
 455                         continue;
 456                 rc = kid->v_read(kid, bp, buf, offset, bytes);
 457                 if (!rc)
 458                         return (0);
 459         }
 460
 461         return (rc);
 462 }
 463
 464 static int
 465 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 466     off_t offset, size_t bytes)
 467 {
 468         vdev_t *kid;
 469
 470         /*
 471          * Here we should have two kids:
 472          * First one which is the one we are replacing and we can trust
 473          * only this one to have valid data, but it might not be present.
 474          * Second one is that one we are replacing with. It is most likely
 475          * healthy, but we can't trust it has needed data, so we won't use it.
 476          */
 477         kid = STAILQ_FIRST(&vdev->v_children);
 478         if (kid == NULL)
 479                 return (EIO);
 480         if (kid->v_state != VDEV_STATE_HEALTHY)
 481                 return (EIO);
 482         return (kid->v_read(kid, bp, buf, offset, bytes));
 483 }
 484
 485 static vdev_t *
 486 vdev_find(uint64_t guid)
 487 {
 488         vdev_t *vdev;
 489
 490         STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
 491                 if (vdev->v_guid == guid)
 492                         return (vdev);
 493
 494         return (0);
 495 }
 496
 497 static vdev_t *
 498 vdev_create(uint64_t guid, vdev_read_t *_read)
 499 {
 500         vdev_t *vdev;
 501
 502         vdev = malloc(sizeof(vdev_t));
 503         memset(vdev, 0, sizeof(vdev_t));
 504         STAILQ_INIT(&vdev->v_children);
 505         vdev->v_guid = guid;
 506         vdev->v_state = VDEV_STATE_OFFLINE;
 507         vdev->v_read = _read;
 508         vdev->v_phys_read = 0;
 509         vdev->v_read_priv = 0;
 510         STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
 511
 512         return (vdev);
 513 }
 514
 515 static int
 516 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
 517     vdev_t **vdevp, int is_newer)
 518 {
 519         int rc;
 520         uint64_t guid, id, ashift, nparity;
 521         const char *type;
 522         const char *path;
 523         vdev_t *vdev, *kid;
 524         const unsigned char *kids;
 525         int nkids, i, is_new;
 526         uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
 527
 528         if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
 529             NULL, &guid)
 530             || nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id)
 531             || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
 532             NULL, &type)) {
 533                 printf("ZFS: can't find vdev details\n");
 534                 return (ENOENT);
 535         }
 536
 537         if (strcmp(type, VDEV_TYPE_MIRROR)
 538             && strcmp(type, VDEV_TYPE_DISK)
 539 #ifdef ZFS_TEST
 540             && strcmp(type, VDEV_TYPE_FILE)
 541 #endif
 542             && strcmp(type, VDEV_TYPE_RAIDZ)
 543             && strcmp(type, VDEV_TYPE_REPLACING)) {
 544                 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 545                 return (EIO);
 546         }
 547
 548         is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
 549
 550         nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
 551                         &is_offline);
 552         nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
 553                         &is_removed);
 554         nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
 555                         &is_faulted);
 556         nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, NULL,
 557                         &is_degraded);
 558         nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, NULL,
 559                         &isnt_present);
 560
 561         vdev = vdev_find(guid);
 562         if (!vdev) {
 563                 is_new = 1;
 564
 565                 if (!strcmp(type, VDEV_TYPE_MIRROR))
 566                         vdev = vdev_create(guid, vdev_mirror_read);
 567                 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
 568                         vdev = vdev_create(guid, vdev_raidz_read);
 569                 else if (!strcmp(type, VDEV_TYPE_REPLACING))
 570                         vdev = vdev_create(guid, vdev_replacing_read);
 571                 else
 572                         vdev = vdev_create(guid, vdev_disk_read);
 573
 574                 vdev->v_id = id;
 575                 vdev->v_top = pvdev != NULL ? pvdev : vdev;
 576                 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
 577                         DATA_TYPE_UINT64, NULL, &ashift) == 0) {
 578                         vdev->v_ashift = ashift;
 579                 } else {
 580                         vdev->v_ashift = 0;
 581                 }
 582                 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
 583                         DATA_TYPE_UINT64, NULL, &nparity) == 0) {
 584                         vdev->v_nparity = nparity;
 585                 } else {
 586                         vdev->v_nparity = 0;
 587                 }
 588                 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
 589                                 DATA_TYPE_STRING, NULL, &path) == 0) {
 590                         if (strncmp(path, "/dev/", 5) == 0)
 591                                 path += 5;
 592                         vdev->v_name = strdup(path);
 593                 } else {
 594                         if (!strcmp(type, "raidz")) {
 595                                 if (vdev->v_nparity == 1)
 596                                         vdev->v_name = "raidz1";
 597                                 else if (vdev->v_nparity == 2)
 598                                         vdev->v_name = "raidz2";
 599                                 else if (vdev->v_nparity == 3)
 600                                         vdev->v_name = "raidz3";
 601                                 else {
 602                                         printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 603                                         return (EIO);
 604                                 }
 605                         } else {
 606                                 vdev->v_name = strdup(type);
 607                         }
 608                 }
 609         } else {
 610                 is_new = 0;
 611         }
 612
 613         if (is_new || is_newer) {
 614                 /*
 615                  * This is either new vdev or we've already seen this vdev,
 616                  * but from an older vdev label, so let's refresh its state
 617                  * from the newer label.
 618                  */
 619                 if (is_offline)
 620                         vdev->v_state = VDEV_STATE_OFFLINE;
 621                 else if (is_removed)
 622                         vdev->v_state = VDEV_STATE_REMOVED;
 623                 else if (is_faulted)
 624                         vdev->v_state = VDEV_STATE_FAULTED;
 625                 else if (is_degraded)
 626                         vdev->v_state = VDEV_STATE_DEGRADED;
 627                 else if (isnt_present)
 628                         vdev->v_state = VDEV_STATE_CANT_OPEN;
 629         }
 630
 631         rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
 632             &nkids, &kids);
 633         /*
 634          * Its ok if we don't have any kids.
 635          */
 636         if (rc == 0) {
 637                 vdev->v_nchildren = nkids;
 638                 for (i = 0; i < nkids; i++) {
 639                         rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
 640                         if (rc)
 641                                 return (rc);
 642                         if (is_new)
 643                                 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
 644                                                    v_childlink);
 645                         kids = nvlist_next(kids);
 646                 }
 647         } else {
 648                 vdev->v_nchildren = 0;
 649         }
 650
 651         if (vdevp)
 652                 *vdevp = vdev;
 653         return (0);
 654 }
 655
 656 static void
 657 vdev_set_state(vdev_t *vdev)
 658 {
 659         vdev_t *kid;
 660         int good_kids;
 661         int bad_kids;
 662
 663         /*
 664          * A mirror or raidz is healthy if all its kids are healthy. A
 665          * mirror is degraded if any of its kids is healthy; a raidz
 666          * is degraded if at most nparity kids are offline.
 667          */
 668         if (STAILQ_FIRST(&vdev->v_children)) {
 669                 good_kids = 0;
 670                 bad_kids = 0;
 671                 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 672                         if (kid->v_state == VDEV_STATE_HEALTHY)
 673                                 good_kids++;
 674                         else
 675                                 bad_kids++;
 676                 }
 677                 if (bad_kids == 0) {
 678                         vdev->v_state = VDEV_STATE_HEALTHY;
 679                 } else {
 680                         if (vdev->v_read == vdev_mirror_read) {
 681                                 if (good_kids) {
 682                                         vdev->v_state = VDEV_STATE_DEGRADED;
 683                                 } else {
 684                                         vdev->v_state = VDEV_STATE_OFFLINE;
 685                                 }
 686                         } else if (vdev->v_read == vdev_raidz_read) {
 687                                 if (bad_kids > vdev->v_nparity) {
 688                                         vdev->v_state = VDEV_STATE_OFFLINE;
 689                                 } else {
 690                                         vdev->v_state = VDEV_STATE_DEGRADED;
 691                                 }
 692                         }
 693                 }
 694         }
 695 }
 696
 697 static spa_t *
 698 spa_find_by_guid(uint64_t guid)
 699 {
 700         spa_t *spa;
 701
 702         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 703                 if (spa->spa_guid == guid)
 704                         return (spa);
 705
 706         return (0);
 707 }
 708
 709 static spa_t *
 710 spa_find_by_name(const char *name)
 711 {
 712         spa_t *spa;
 713
 714         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 715                 if (!strcmp(spa->spa_name, name))
 716                         return (spa);
 717
 718         return (0);
 719 }
 720
 721 #ifdef BOOT2
 722 static spa_t *
 723 spa_get_primary(void)
 724 {
 725
 726         return (STAILQ_FIRST(&zfs_pools));
 727 }
 728
 729 static vdev_t *
 730 spa_get_primary_vdev(const spa_t *spa)
 731 {
 732         vdev_t *vdev;
 733         vdev_t *kid;
 734
 735         if (spa == NULL)
 736                 spa = spa_get_primary();
 737         if (spa == NULL)
 738                 return (NULL);
 739         vdev = STAILQ_FIRST(&spa->spa_vdevs);
 740         if (vdev == NULL)
 741                 return (NULL);
 742         for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
 743              kid = STAILQ_FIRST(&vdev->v_children))
 744                 vdev = kid;
 745         return (vdev);
 746 }
 747 #endif
 748
 749 static spa_t *
 750 spa_create(uint64_t guid, const char *name)
 751 {
 752         spa_t *spa;
 753
 754         if ((spa = malloc(sizeof(spa_t))) == NULL)
 755                 return (NULL);
 756         memset(spa, 0, sizeof(spa_t));
 757         if ((spa->spa_name = strdup(name)) == NULL) {
 758                 free(spa);
 759                 return (NULL);
 760         }
 761         STAILQ_INIT(&spa->spa_vdevs);
 762         spa->spa_guid = guid;
 763         STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
 764
 765         return (spa);
 766 }
 767
 768 static const char *
 769 state_name(vdev_state_t state)
 770 {
 771         static const char* names[] = {
 772                 "UNKNOWN",
 773                 "CLOSED",
 774                 "OFFLINE",
 775                 "REMOVED",
 776                 "CANT_OPEN",
 777                 "FAULTED",
 778                 "DEGRADED",
 779                 "ONLINE"
 780         };
 781         return names[state];
 782 }
 783
 784 #ifdef BOOT2
 785
 786 #define pager_printf printf
 787
 788 #else
 789
 790 static int
 791 pager_printf(const char *fmt, ...)
 792 {
 793         char line[80];
 794         va_list args;
 795
 796         va_start(args, fmt);
 797         vsprintf(line, fmt, args);
 798         va_end(args);
 799
 800         return (pager_output(line));
 801 }
 802
 803 #endif
 804
 805 #define STATUS_FORMAT   "        %s %s\n"
 806
 807 static int
 808 print_state(int indent, const char *name, vdev_state_t state)
 809 {
 810         char buf[512];
 811         int i;
 812
 813         buf[0] = 0;
 814         for (i = 0; i < indent; i++)
 815                 strcat(buf, "  ");
 816         strcat(buf, name);
 817
 818         return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
 819 }
 820
 821 static int
 822 vdev_status(vdev_t *vdev, int indent)
 823 {
 824         vdev_t *kid;
 825         int ret;
 826         ret = print_state(indent, vdev->v_name, vdev->v_state);
 827         if (ret != 0)
 828                 return (ret);
 829
 830         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 831                 ret = vdev_status(kid, indent + 1);
 832                 if (ret != 0)
 833                         return (ret);
 834         }
 835         return (ret);
 836 }
 837
 838 static int
 839 spa_status(spa_t *spa)
 840 {
 841         static char bootfs[ZFS_MAXNAMELEN];
 842         uint64_t rootid;
 843         vdev_t *vdev;
 844         int good_kids, bad_kids, degraded_kids, ret;
 845         vdev_state_t state;
 846
 847         ret = pager_printf("  pool: %s\n", spa->spa_name);
 848         if (ret != 0)
 849                 return (ret);
 850
 851         if (zfs_get_root(spa, &rootid) == 0 &&
 852             zfs_rlookup(spa, rootid, bootfs) == 0) {
 853                 if (bootfs[0] == '\0')
 854                         ret = pager_printf("bootfs: %s\n", spa->spa_name);
 855                 else
 856                         ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
 857                             bootfs);
 858                 if (ret != 0)
 859                         return (ret);
 860         }
 861         ret = pager_printf("config:\n\n");
 862         if (ret != 0)
 863                 return (ret);
 864         ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
 865         if (ret != 0)
 866                 return (ret);
 867
 868         good_kids = 0;
 869         degraded_kids = 0;
 870         bad_kids = 0;
 871         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 872                 if (vdev->v_state == VDEV_STATE_HEALTHY)
 873                         good_kids++;
 874                 else if (vdev->v_state == VDEV_STATE_DEGRADED)
 875                         degraded_kids++;
 876                 else
 877                         bad_kids++;
 878         }
 879
 880         state = VDEV_STATE_CLOSED;
 881         if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
 882                 state = VDEV_STATE_HEALTHY;
 883         else if ((good_kids + degraded_kids) > 0)
 884                 state = VDEV_STATE_DEGRADED;
 885
 886         ret = print_state(0, spa->spa_name, state);
 887         if (ret != 0)
 888                 return (ret);
 889         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 890                 ret = vdev_status(vdev, 1);
 891                 if (ret != 0)
 892                         return (ret);
 893         }
 894         return (ret);
 895 }
 896
 897 static int
 898 spa_all_status(void)
 899 {
 900         spa_t *spa;
 901         int first = 1, ret = 0;
 902
 903         STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
 904                 if (!first) {
 905                         ret = pager_printf("\n");
 906                         if (ret != 0)
 907                                 return (ret);
 908                 }
 909                 first = 0;
 910                 ret = spa_status(spa);
 911                 if (ret != 0)
 912                         return (ret);
 913         }
 914         return (ret);
 915 }
 916
 917 static uint64_t
 918 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
 919 {
 920         uint64_t label_offset;
 921
 922         if (l < VDEV_LABELS / 2)
 923                 label_offset = 0;
 924         else
 925                 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
 926
 927         return (offset + l * sizeof (vdev_label_t) + label_offset);
 928 }
 929
 930 static int
 931 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
 932 {
 933         vdev_t vtmp;
 934         vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
 935         vdev_phys_t *tmp_label;
 936         spa_t *spa;
 937         vdev_t *vdev, *top_vdev, *pool_vdev;
 938         off_t off;
 939         blkptr_t bp;
 940         const unsigned char *nvlist = NULL;
 941         uint64_t val;
 942         uint64_t guid;
 943         uint64_t best_txg = 0;
 944         uint64_t pool_txg, pool_guid;
 945         uint64_t psize;
 946         const char *pool_name;
 947         const unsigned char *vdevs;
 948         const unsigned char *features;
 949         int i, l, rc, is_newer;
 950         char *upbuf;
 951         const struct uberblock *up;
 952
 953         /*
 954          * Load the vdev label and figure out which
 955          * uberblock is most current.
 956          */
 957         memset(&vtmp, 0, sizeof(vtmp));
 958         vtmp.v_phys_read = _read;
 959         vtmp.v_read_priv = read_priv;
 960         psize = P2ALIGN(ldi_get_size(read_priv),
 961             (uint64_t)sizeof (vdev_label_t));
 962
 963         /* Test for minimum pool size. */
 964         if (psize < SPA_MINDEVSIZE)
 965                 return (EIO);
 966
 967         tmp_label = zfs_alloc(sizeof(vdev_phys_t));
 968
 969         for (l = 0; l < VDEV_LABELS; l++) {
 970                 off = vdev_label_offset(psize, l,
 971                     offsetof(vdev_label_t, vl_vdev_phys));
 972
 973                 BP_ZERO(&bp);
 974                 BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
 975                 BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
 976                 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
 977                 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
 978                 DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
 979                 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
 980
 981                 if (vdev_read_phys(&vtmp, &bp, tmp_label, off, 0))
 982                         continue;
 983
 984                 if (tmp_label->vp_nvlist[0] != NV_ENCODE_XDR)
 985                         continue;
 986
 987                 nvlist = (const unsigned char *) tmp_label->vp_nvlist + 4;
 988                 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
 989                     DATA_TYPE_UINT64, NULL, &pool_txg) != 0)
 990                         continue;
 991
 992                 if (best_txg <= pool_txg) {
 993                         best_txg = pool_txg;
 994                         memcpy(vdev_label, tmp_label, sizeof (vdev_phys_t));
 995                 }
 996         }
 997
 998         zfs_free(tmp_label, sizeof (vdev_phys_t));
 999
1000         if (best_txg == 0)
1001                 return (EIO);
1002
1003         if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR)
1004                 return (EIO);
1005
1006         nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
1007
1008         if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1009             NULL, &val) != 0) {
1010                 return (EIO);
1011         }
1012
1013         if (!SPA_VERSION_IS_SUPPORTED(val)) {
1014                 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1015                     (unsigned) val, (unsigned) SPA_VERSION);
1016                 return (EIO);
1017         }
1018
1019         /* Check ZFS features for read */
1020         if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1021             DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1022             nvlist_check_features_for_read(features) != 0) {
1023                 return (EIO);
1024         }
1025
1026         if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1027             NULL, &val) != 0) {
1028                 return (EIO);
1029         }
1030
1031         if (val == POOL_STATE_DESTROYED) {
1032                 /* We don't boot only from destroyed pools. */
1033                 return (EIO);
1034         }
1035
1036         if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1037             NULL, &pool_txg) != 0 ||
1038             nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1039             NULL, &pool_guid) != 0 ||
1040             nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1041             NULL, &pool_name) != 0) {
1042                 /*
1043                  * Cache and spare devices end up here - just ignore
1044                  * them.
1045                  */
1046                 /*printf("ZFS: can't find pool details\n");*/
1047                 return (EIO);
1048         }
1049
1050         if (nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64,
1051             NULL, &val) == 0 && val != 0) {
1052                 return (EIO);
1053         }
1054
1055         /*
1056          * Create the pool if this is the first time we've seen it.
1057          */
1058         spa = spa_find_by_guid(pool_guid);
1059         if (spa == NULL) {
1060                 spa = spa_create(pool_guid, pool_name);
1061                 if (spa == NULL)
1062                         return (ENOMEM);
1063         }
1064         if (pool_txg > spa->spa_txg) {
1065                 spa->spa_txg = pool_txg;
1066                 is_newer = 1;
1067         } else {
1068                 is_newer = 0;
1069         }
1070
1071         /*
1072          * Get the vdev tree and create our in-core copy of it.
1073          * If we already have a vdev with this guid, this must
1074          * be some kind of alias (overlapping slices, dangerously dedicated
1075          * disks etc).
1076          */
1077         if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1078             NULL, &guid) != 0) {
1079                 return (EIO);
1080         }
1081         vdev = vdev_find(guid);
1082         if (vdev && vdev->v_phys_read)  /* Has this vdev already been inited? */
1083                 return (EIO);
1084
1085         if (nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1086             NULL, &vdevs)) {
1087                 return (EIO);
1088         }
1089
1090         rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1091         if (rc != 0)
1092                 return (rc);
1093
1094         /*
1095          * Add the toplevel vdev to the pool if its not already there.
1096          */
1097         STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1098                 if (top_vdev == pool_vdev)
1099                         break;
1100         if (!pool_vdev && top_vdev) {
1101                 top_vdev->spa = spa;
1102                 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1103         }
1104
1105         /*
1106          * We should already have created an incomplete vdev for this
1107          * vdev. Find it and initialise it with our read proc.
1108          */
1109         vdev = vdev_find(guid);
1110         if (vdev) {
1111                 vdev->v_phys_read = _read;
1112                 vdev->v_read_priv = read_priv;
1113                 vdev->v_state = VDEV_STATE_HEALTHY;
1114         } else {
1115                 printf("ZFS: inconsistent nvlist contents\n");
1116                 return (EIO);
1117         }
1118
1119         /*
1120          * Re-evaluate top-level vdev state.
1121          */
1122         vdev_set_state(top_vdev);
1123
1124         /*
1125          * Ok, we are happy with the pool so far. Lets find
1126          * the best uberblock and then we can actually access
1127          * the contents of the pool.
1128          */
1129         upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
1130         up = (const struct uberblock *)upbuf;
1131         for (l = 0; l < VDEV_LABELS; l++) {
1132                 for (i = 0; i < VDEV_UBERBLOCK_COUNT(vdev); i++) {
1133                         off = vdev_label_offset(psize, l,
1134                             VDEV_UBERBLOCK_OFFSET(vdev, i));
1135                         BP_ZERO(&bp);
1136                         DVA_SET_OFFSET(&bp.blk_dva[0], off);
1137                         BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1138                         BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1139                         BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1140                         BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1141                         ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1142
1143                         if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
1144                                 continue;
1145
1146                         if (up->ub_magic != UBERBLOCK_MAGIC)
1147                                 continue;
1148                         if (up->ub_txg < spa->spa_txg)
1149                                 continue;
1150                         if (up->ub_txg > spa->spa_uberblock.ub_txg ||
1151                             (up->ub_txg == spa->spa_uberblock.ub_txg &&
1152                             up->ub_timestamp >
1153                             spa->spa_uberblock.ub_timestamp)) {
1154                                 spa->spa_uberblock = *up;
1155                         }
1156                 }
1157         }
1158         zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
1159
1160         vdev->spa = spa;
1161         if (spap != NULL)
1162                 *spap = spa;
1163         return (0);
1164 }
1165
1166 static int
1167 ilog2(int n)
1168 {
1169         int v;
1170
1171         for (v = 0; v < 32; v++)
1172                 if (n == (1 << v))
1173                         return v;
1174         return -1;
1175 }
1176
1177 static int
1178 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1179 {
1180         blkptr_t gbh_bp;
1181         zio_gbh_phys_t zio_gb;
1182         char *pbuf;
1183         int i;
1184
1185         /* Artificial BP for gang block header. */
1186         gbh_bp = *bp;
1187         BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1188         BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1189         BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1190         BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1191         for (i = 0; i < SPA_DVAS_PER_BP; i++)
1192                 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1193
1194         /* Read gang header block using the artificial BP. */
1195         if (zio_read(spa, &gbh_bp, &zio_gb))
1196                 return (EIO);
1197
1198         pbuf = buf;
1199         for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1200                 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1201
1202                 if (BP_IS_HOLE(gbp))
1203                         continue;
1204                 if (zio_read(spa, gbp, pbuf))
1205                         return (EIO);
1206                 pbuf += BP_GET_PSIZE(gbp);
1207         }
1208
1209         if (zio_checksum_verify(spa, bp, buf))
1210                 return (EIO);
1211         return (0);
1212 }
1213
1214 static int
1215 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1216 {
1217         int cpfunc = BP_GET_COMPRESS(bp);
1218         uint64_t align, size;
1219         void *pbuf;
1220         int i, error;
1221
1222         /*
1223          * Process data embedded in block pointer
1224          */
1225         if (BP_IS_EMBEDDED(bp)) {
1226                 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1227
1228                 size = BPE_GET_PSIZE(bp);
1229                 ASSERT(size <= BPE_PAYLOAD_SIZE);
1230
1231                 if (cpfunc != ZIO_COMPRESS_OFF)
1232                         pbuf = zfs_alloc(size);
1233                 else
1234                         pbuf = buf;
1235
1236                 decode_embedded_bp_compressed(bp, pbuf);
1237                 error = 0;
1238
1239                 if (cpfunc != ZIO_COMPRESS_OFF) {
1240                         error = zio_decompress_data(cpfunc, pbuf,
1241                             size, buf, BP_GET_LSIZE(bp));
1242                         zfs_free(pbuf, size);
1243                 }
1244                 if (error != 0)
1245                         printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
1246                             error);
1247                 return (error);
1248         }
1249
1250         error = EIO;
1251
1252         for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1253                 const dva_t *dva = &bp->blk_dva[i];
1254                 vdev_t *vdev;
1255                 int vdevid;
1256                 off_t offset;
1257
1258                 if (!dva->dva_word[0] && !dva->dva_word[1])
1259                         continue;
1260
1261                 vdevid = DVA_GET_VDEV(dva);
1262                 offset = DVA_GET_OFFSET(dva);
1263                 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1264                         if (vdev->v_id == vdevid)
1265                                 break;
1266                 }
1267                 if (!vdev || !vdev->v_read)
1268                         continue;
1269
1270                 size = BP_GET_PSIZE(bp);
1271                 if (vdev->v_read == vdev_raidz_read) {
1272                         align = 1ULL << vdev->v_top->v_ashift;
1273                         if (P2PHASE(size, align) != 0)
1274                                 size = P2ROUNDUP(size, align);
1275                 }
1276                 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1277                         pbuf = zfs_alloc(size);
1278                 else
1279                         pbuf = buf;
1280
1281                 if (DVA_GET_GANG(dva))
1282                         error = zio_read_gang(spa, bp, pbuf);
1283                 else
1284                         error = vdev->v_read(vdev, bp, pbuf, offset, size);
1285                 if (error == 0) {
1286                         if (cpfunc != ZIO_COMPRESS_OFF)
1287                                 error = zio_decompress_data(cpfunc, pbuf,
1288                                     BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1289                         else if (size != BP_GET_PSIZE(bp))
1290                                 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1291                 }
1292                 if (buf != pbuf)
1293                         zfs_free(pbuf, size);
1294                 if (error == 0)
1295                         break;
1296         }
1297         if (error != 0)
1298                 printf("ZFS: i/o error - all block copies unavailable\n");
1299         return (error);
1300 }
1301
1302 static int
1303 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1304 {
1305         int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1306         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1307         int nlevels = dnode->dn_nlevels;
1308         int i, rc;
1309
1310         if (bsize > SPA_MAXBLOCKSIZE) {
1311                 printf("ZFS: I/O error - blocks larger than %llu are not "
1312                     "supported\n", SPA_MAXBLOCKSIZE);
1313                 return (EIO);
1314         }
1315
1316         /*
1317          * Note: bsize may not be a power of two here so we need to do an
1318          * actual divide rather than a bitshift.
1319          */
1320         while (buflen > 0) {
1321                 uint64_t bn = offset / bsize;
1322                 int boff = offset % bsize;
1323                 int ibn;
1324                 const blkptr_t *indbp;
1325                 blkptr_t bp;
1326
1327                 if (bn > dnode->dn_maxblkid)
1328                         return (EIO);
1329
1330                 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1331                         goto cached;
1332
1333                 indbp = dnode->dn_blkptr;
1334                 for (i = 0; i < nlevels; i++) {
1335                         /*
1336                          * Copy the bp from the indirect array so that
1337                          * we can re-use the scratch buffer for multi-level
1338                          * objects.
1339                          */
1340                         ibn = bn >> ((nlevels - i - 1) * ibshift);
1341                         ibn &= ((1 << ibshift) - 1);
1342                         bp = indbp[ibn];
1343                         if (BP_IS_HOLE(&bp)) {
1344                                 memset(dnode_cache_buf, 0, bsize);
1345                                 break;
1346                         }
1347                         rc = zio_read(spa, &bp, dnode_cache_buf);
1348                         if (rc)
1349                                 return (rc);
1350                         indbp = (const blkptr_t *) dnode_cache_buf;
1351                 }
1352                 dnode_cache_obj = dnode;
1353                 dnode_cache_bn = bn;
1354         cached:
1355
1356                 /*
1357                  * The buffer contains our data block. Copy what we
1358                  * need from it and loop.
1359                  */
1360                 i = bsize - boff;
1361                 if (i > buflen) i = buflen;
1362                 memcpy(buf, &dnode_cache_buf[boff], i);
1363                 buf = ((char*) buf) + i;
1364                 offset += i;
1365                 buflen -= i;
1366         }
1367
1368         return (0);
1369 }
1370
1371 /*
1372  * Lookup a value in a microzap directory. Assumes that the zap
1373  * scratch buffer contains the directory contents.
1374  */
1375 static int
1376 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1377 {
1378         const mzap_phys_t *mz;
1379         const mzap_ent_phys_t *mze;
1380         size_t size;
1381         int chunks, i;
1382
1383         /*
1384          * Microzap objects use exactly one block. Read the whole
1385          * thing.
1386          */
1387         size = dnode->dn_datablkszsec * 512;
1388
1389         mz = (const mzap_phys_t *) zap_scratch;
1390         chunks = size / MZAP_ENT_LEN - 1;
1391
1392         for (i = 0; i < chunks; i++) {
1393                 mze = &mz->mz_chunk[i];
1394                 if (!strcmp(mze->mze_name, name)) {
1395                         *value = mze->mze_value;
1396                         return (0);
1397                 }
1398         }
1399
1400         return (ENOENT);
1401 }
1402
1403 /*
1404  * Compare a name with a zap leaf entry. Return non-zero if the name
1405  * matches.
1406  */
1407 static int
1408 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1409 {
1410         size_t namelen;
1411         const zap_leaf_chunk_t *nc;
1412         const char *p;
1413
1414         namelen = zc->l_entry.le_name_numints;
1415
1416         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1417         p = name;
1418         while (namelen > 0) {
1419                 size_t len;
1420                 len = namelen;
1421                 if (len > ZAP_LEAF_ARRAY_BYTES)
1422                         len = ZAP_LEAF_ARRAY_BYTES;
1423                 if (memcmp(p, nc->l_array.la_array, len))
1424                         return (0);
1425                 p += len;
1426                 namelen -= len;
1427                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1428         }
1429
1430         return 1;
1431 }
1432
1433 /*
1434  * Extract a uint64_t value from a zap leaf entry.
1435  */
1436 static uint64_t
1437 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1438 {
1439         const zap_leaf_chunk_t *vc;
1440         int i;
1441         uint64_t value;
1442         const uint8_t *p;
1443
1444         vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1445         for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1446                 value = (value << 8) | p[i];
1447         }
1448
1449         return value;
1450 }
1451
1452 static void
1453 stv(int len, void *addr, uint64_t value)
1454 {
1455         switch (len) {
1456         case 1:
1457                 *(uint8_t *)addr = value;
1458                 return;
1459         case 2:
1460                 *(uint16_t *)addr = value;
1461                 return;
1462         case 4:
1463                 *(uint32_t *)addr = value;
1464                 return;
1465         case 8:
1466                 *(uint64_t *)addr = value;
1467                 return;
1468         }
1469 }
1470
1471 /*
1472  * Extract a array from a zap leaf entry.
1473  */
1474 static void
1475 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
1476     uint64_t integer_size, uint64_t num_integers, void *buf)
1477 {
1478         uint64_t array_int_len = zc->l_entry.le_value_intlen;
1479         uint64_t value = 0;
1480         uint64_t *u64 = buf;
1481         char *p = buf;
1482         int len = MIN(zc->l_entry.le_value_numints, num_integers);
1483         int chunk = zc->l_entry.le_value_chunk;
1484         int byten = 0;
1485
1486         if (integer_size == 8 && len == 1) {
1487                 *u64 = fzap_leaf_value(zl, zc);
1488                 return;
1489         }
1490
1491         while (len > 0) {
1492                 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
1493                 int i;
1494
1495                 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
1496                 for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
1497                         value = (value << 8) | la->la_array[i];
1498                         byten++;
1499                         if (byten == array_int_len) {
1500                                 stv(integer_size, p, value);
1501                                 byten = 0;
1502                                 len--;
1503                                 if (len == 0)
1504                                         return;
1505                                 p += integer_size;
1506                         }
1507                 }
1508                 chunk = la->la_next;
1509         }
1510 }
1511
1512 /*
1513  * Lookup a value in a fatzap directory. Assumes that the zap scratch
1514  * buffer contains the directory header.
1515  */
1516 static int
1517 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
1518     uint64_t integer_size, uint64_t num_integers, void *value)
1519 {
1520         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1521         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1522         fat_zap_t z;
1523         uint64_t *ptrtbl;
1524         uint64_t hash;
1525         int rc;
1526
1527         if (zh.zap_magic != ZAP_MAGIC)
1528                 return (EIO);
1529
1530         z.zap_block_shift = ilog2(bsize);
1531         z.zap_phys = (zap_phys_t *) zap_scratch;
1532
1533         /*
1534          * Figure out where the pointer table is and read it in if necessary.
1535          */
1536         if (zh.zap_ptrtbl.zt_blk) {
1537                 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1538                                zap_scratch, bsize);
1539                 if (rc)
1540                         return (rc);
1541                 ptrtbl = (uint64_t *) zap_scratch;
1542         } else {
1543                 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1544         }
1545
1546         hash = zap_hash(zh.zap_salt, name);
1547
1548         zap_leaf_t zl;
1549         zl.l_bs = z.zap_block_shift;
1550
1551         off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1552         zap_leaf_chunk_t *zc;
1553
1554         rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1555         if (rc)
1556                 return (rc);
1557
1558         zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1559
1560         /*
1561          * Make sure this chunk matches our hash.
1562          */
1563         if (zl.l_phys->l_hdr.lh_prefix_len > 0
1564             && zl.l_phys->l_hdr.lh_prefix
1565             != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1566                 return (ENOENT);
1567
1568         /*
1569          * Hash within the chunk to find our entry.
1570          */
1571         int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1572         int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1573         h = zl.l_phys->l_hash[h];
1574         if (h == 0xffff)
1575                 return (ENOENT);
1576         zc = &ZAP_LEAF_CHUNK(&zl, h);
1577         while (zc->l_entry.le_hash != hash) {
1578                 if (zc->l_entry.le_next == 0xffff) {
1579                         zc = NULL;
1580                         break;
1581                 }
1582                 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1583         }
1584         if (fzap_name_equal(&zl, zc, name)) {
1585                 if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints >
1586                     integer_size * num_integers)
1587                         return (E2BIG);
1588                 fzap_leaf_array(&zl, zc, integer_size, num_integers, value);
1589                 return (0);
1590         }
1591
1592         return (ENOENT);
1593 }
1594
1595 /*
1596  * Lookup a name in a zap object and return its value as a uint64_t.
1597  */
1598 static int
1599 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
1600     uint64_t integer_size, uint64_t num_integers, void *value)
1601 {
1602         int rc;
1603         uint64_t zap_type;
1604         size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1605
1606         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1607         if (rc)
1608                 return (rc);
1609
1610         zap_type = *(uint64_t *) zap_scratch;
1611         if (zap_type == ZBT_MICRO)
1612                 return mzap_lookup(dnode, name, value);
1613         else if (zap_type == ZBT_HEADER) {
1614                 return fzap_lookup(spa, dnode, name, integer_size,
1615                     num_integers, value);
1616         }
1617         printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
1618         return (EIO);
1619 }
1620
1621 /*
1622  * List a microzap directory. Assumes that the zap scratch buffer contains
1623  * the directory contents.
1624  */
1625 static int
1626 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1627 {
1628         const mzap_phys_t *mz;
1629         const mzap_ent_phys_t *mze;
1630         size_t size;
1631         int chunks, i, rc;
1632
1633         /*
1634          * Microzap objects use exactly one block. Read the whole
1635          * thing.
1636          */
1637         size = dnode->dn_datablkszsec * 512;
1638         mz = (const mzap_phys_t *) zap_scratch;
1639         chunks = size / MZAP_ENT_LEN - 1;
1640
1641         for (i = 0; i < chunks; i++) {
1642                 mze = &mz->mz_chunk[i];
1643                 if (mze->mze_name[0]) {
1644                         rc = callback(mze->mze_name, mze->mze_value);
1645                         if (rc != 0)
1646                                 return (rc);
1647                 }
1648         }
1649
1650         return (0);
1651 }
1652
1653 /*
1654  * List a fatzap directory. Assumes that the zap scratch buffer contains
1655  * the directory header.
1656  */
1657 static int
1658 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1659 {
1660         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1661         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1662         fat_zap_t z;
1663         int i, j, rc;
1664
1665         if (zh.zap_magic != ZAP_MAGIC)
1666                 return (EIO);
1667
1668         z.zap_block_shift = ilog2(bsize);
1669         z.zap_phys = (zap_phys_t *) zap_scratch;
1670
1671         /*
1672          * This assumes that the leaf blocks start at block 1. The
1673          * documentation isn't exactly clear on this.
1674          */
1675         zap_leaf_t zl;
1676         zl.l_bs = z.zap_block_shift;
1677         for (i = 0; i < zh.zap_num_leafs; i++) {
1678                 off_t off = (i + 1) << zl.l_bs;
1679                 char name[256], *p;
1680                 uint64_t value;
1681
1682                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1683                         return (EIO);
1684
1685                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1686
1687                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1688                         zap_leaf_chunk_t *zc, *nc;
1689                         int namelen;
1690
1691                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1692                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1693                                 continue;
1694                         namelen = zc->l_entry.le_name_numints;
1695                         if (namelen > sizeof(name))
1696                                 namelen = sizeof(name);
1697
1698                         /*
1699                          * Paste the name back together.
1700                          */
1701                         nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1702                         p = name;
1703                         while (namelen > 0) {
1704                                 int len;
1705                                 len = namelen;
1706                                 if (len > ZAP_LEAF_ARRAY_BYTES)
1707                                         len = ZAP_LEAF_ARRAY_BYTES;
1708                                 memcpy(p, nc->l_array.la_array, len);
1709                                 p += len;
1710                                 namelen -= len;
1711                                 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1712                         }
1713
1714                         /*
1715                          * Assume the first eight bytes of the value are
1716                          * a uint64_t.
1717                          */
1718                         value = fzap_leaf_value(&zl, zc);
1719
1720                         //printf("%s 0x%jx\n", name, (uintmax_t)value);
1721                         rc = callback((const char *)name, value);
1722                         if (rc != 0)
1723                                 return (rc);
1724                 }
1725         }
1726
1727         return (0);
1728 }
1729
1730 static int zfs_printf(const char *name, uint64_t value __unused)
1731 {
1732
1733         printf("%s\n", name);
1734
1735         return (0);
1736 }
1737
1738 /*
1739  * List a zap directory.
1740  */
1741 static int
1742 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
1743 {
1744         uint64_t zap_type;
1745         size_t size = dnode->dn_datablkszsec * 512;
1746
1747         if (dnode_read(spa, dnode, 0, zap_scratch, size))
1748                 return (EIO);
1749
1750         zap_type = *(uint64_t *) zap_scratch;
1751         if (zap_type == ZBT_MICRO)
1752                 return mzap_list(dnode, zfs_printf);
1753         else
1754                 return fzap_list(spa, dnode, zfs_printf);
1755 }
1756
1757 static int
1758 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1759 {
1760         off_t offset;
1761
1762         offset = objnum * sizeof(dnode_phys_t);
1763         return dnode_read(spa, &os->os_meta_dnode, offset,
1764                 dnode, sizeof(dnode_phys_t));
1765 }
1766
1767 static int
1768 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1769 {
1770         const mzap_phys_t *mz;
1771         const mzap_ent_phys_t *mze;
1772         size_t size;
1773         int chunks, i;
1774
1775         /*
1776          * Microzap objects use exactly one block. Read the whole
1777          * thing.
1778          */
1779         size = dnode->dn_datablkszsec * 512;
1780
1781         mz = (const mzap_phys_t *) zap_scratch;
1782         chunks = size / MZAP_ENT_LEN - 1;
1783
1784         for (i = 0; i < chunks; i++) {
1785                 mze = &mz->mz_chunk[i];
1786                 if (value == mze->mze_value) {
1787                         strcpy(name, mze->mze_name);
1788                         return (0);
1789                 }
1790         }
1791
1792         return (ENOENT);
1793 }
1794
1795 static void
1796 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
1797 {
1798         size_t namelen;
1799         const zap_leaf_chunk_t *nc;
1800         char *p;
1801
1802         namelen = zc->l_entry.le_name_numints;
1803
1804         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1805         p = name;
1806         while (namelen > 0) {
1807                 size_t len;
1808                 len = namelen;
1809                 if (len > ZAP_LEAF_ARRAY_BYTES)
1810                         len = ZAP_LEAF_ARRAY_BYTES;
1811                 memcpy(p, nc->l_array.la_array, len);
1812                 p += len;
1813                 namelen -= len;
1814                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1815         }
1816
1817         *p = '\0';
1818 }
1819
1820 static int
1821 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1822 {
1823         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1824         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1825         fat_zap_t z;
1826         int i, j;
1827
1828         if (zh.zap_magic != ZAP_MAGIC)
1829                 return (EIO);
1830
1831         z.zap_block_shift = ilog2(bsize);
1832         z.zap_phys = (zap_phys_t *) zap_scratch;
1833
1834         /*
1835          * This assumes that the leaf blocks start at block 1. The
1836          * documentation isn't exactly clear on this.
1837          */
1838         zap_leaf_t zl;
1839         zl.l_bs = z.zap_block_shift;
1840         for (i = 0; i < zh.zap_num_leafs; i++) {
1841                 off_t off = (i + 1) << zl.l_bs;
1842
1843                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1844                         return (EIO);
1845
1846                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1847
1848                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1849                         zap_leaf_chunk_t *zc;
1850
1851                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1852                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1853                                 continue;
1854                         if (zc->l_entry.le_value_intlen != 8 ||
1855                             zc->l_entry.le_value_numints != 1)
1856                                 continue;
1857
1858                         if (fzap_leaf_value(&zl, zc) == value) {
1859                                 fzap_name_copy(&zl, zc, name);
1860                                 return (0);
1861                         }
1862                 }
1863         }
1864
1865         return (ENOENT);
1866 }
1867
1868 static int
1869 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1870 {
1871         int rc;
1872         uint64_t zap_type;
1873         size_t size = dnode->dn_datablkszsec * 512;
1874
1875         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1876         if (rc)
1877                 return (rc);
1878
1879         zap_type = *(uint64_t *) zap_scratch;
1880         if (zap_type == ZBT_MICRO)
1881                 return mzap_rlookup(spa, dnode, name, value);
1882         else
1883                 return fzap_rlookup(spa, dnode, name, value);
1884 }
1885
1886 static int
1887 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
1888 {
1889         char name[256];
1890         char component[256];
1891         uint64_t dir_obj, parent_obj, child_dir_zapobj;
1892         dnode_phys_t child_dir_zap, dataset, dir, parent;
1893         dsl_dir_phys_t *dd;
1894         dsl_dataset_phys_t *ds;
1895         char *p;
1896         int len;
1897
1898         p = &name[sizeof(name) - 1];
1899         *p = '\0';
1900
1901         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1902                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1903                 return (EIO);
1904         }
1905         ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
1906         dir_obj = ds->ds_dir_obj;
1907
1908         for (;;) {
1909                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
1910                         return (EIO);
1911                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1912
1913                 /* Actual loop condition. */
1914                 parent_obj  = dd->dd_parent_obj;
1915                 if (parent_obj == 0)
1916                         break;
1917
1918                 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
1919                         return (EIO);
1920                 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
1921                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1922                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1923                         return (EIO);
1924                 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
1925                         return (EIO);
1926
1927                 len = strlen(component);
1928                 p -= len;
1929                 memcpy(p, component, len);
1930                 --p;
1931                 *p = '/';
1932
1933                 /* Actual loop iteration. */
1934                 dir_obj = parent_obj;
1935         }
1936
1937         if (*p != '\0')
1938                 ++p;
1939         strcpy(result, p);
1940
1941         return (0);
1942 }
1943
1944 static int
1945 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
1946 {
1947         char element[256];
1948         uint64_t dir_obj, child_dir_zapobj;
1949         dnode_phys_t child_dir_zap, dir;
1950         dsl_dir_phys_t *dd;
1951         const char *p, *q;
1952
1953         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
1954                 return (EIO);
1955         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
1956             1, &dir_obj))
1957                 return (EIO);
1958
1959         p = name;
1960         for (;;) {
1961                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
1962                         return (EIO);
1963                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1964
1965                 while (*p == '/')
1966                         p++;
1967                 /* Actual loop condition #1. */
1968                 if (*p == '\0')
1969                         break;
1970
1971                 q = strchr(p, '/');
1972                 if (q) {
1973                         memcpy(element, p, q - p);
1974                         element[q - p] = '\0';
1975                         p = q + 1;
1976                 } else {
1977                         strcpy(element, p);
1978                         p += strlen(p);
1979                 }
1980
1981                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1982                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1983                         return (EIO);
1984
1985                 /* Actual loop condition #2. */
1986                 if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
1987                     1, &dir_obj) != 0)
1988                         return (ENOENT);
1989         }
1990
1991         *objnum = dd->dd_head_dataset_obj;
1992         return (0);
1993 }
1994
1995 #ifndef BOOT2
1996 static int
1997 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
1998 {
1999         uint64_t dir_obj, child_dir_zapobj;
2000         dnode_phys_t child_dir_zap, dir, dataset;
2001         dsl_dataset_phys_t *ds;
2002         dsl_dir_phys_t *dd;
2003
2004         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2005                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2006                 return (EIO);
2007         }
2008         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2009         dir_obj = ds->ds_dir_obj;
2010
2011         if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
2012                 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2013                 return (EIO);
2014         }
2015         dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2016
2017         child_dir_zapobj = dd->dd_child_dir_zapobj;
2018         if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
2019                 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2020                 return (EIO);
2021         }
2022
2023         return (zap_list(spa, &child_dir_zap) != 0);
2024 }
2025
2026 int
2027 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *, uint64_t))
2028 {
2029         uint64_t dir_obj, child_dir_zapobj, zap_type;
2030         dnode_phys_t child_dir_zap, dir, dataset;
2031         dsl_dataset_phys_t *ds;
2032         dsl_dir_phys_t *dd;
2033         int err;
2034
2035         err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
2036         if (err != 0) {
2037                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2038                 return (err);
2039         }
2040         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2041         dir_obj = ds->ds_dir_obj;
2042
2043         err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
2044         if (err != 0) {
2045                 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2046                 return (err);
2047         }
2048         dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2049
2050         child_dir_zapobj = dd->dd_child_dir_zapobj;
2051         err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
2052         if (err != 0) {
2053                 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2054                 return (err);
2055         }
2056
2057         err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
2058         if (err != 0)
2059                 return (err);
2060
2061         zap_type = *(uint64_t *) zap_scratch;
2062         if (zap_type == ZBT_MICRO)
2063                 return mzap_list(&child_dir_zap, callback);
2064         else
2065                 return fzap_list(spa, &child_dir_zap, callback);
2066 }
2067 #endif
2068
2069 /*
2070  * Find the object set given the object number of its dataset object
2071  * and return its details in *objset
2072  */
2073 static int
2074 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2075 {
2076         dnode_phys_t dataset;
2077         dsl_dataset_phys_t *ds;
2078
2079         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2080                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2081                 return (EIO);
2082         }
2083
2084         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2085         if (zio_read(spa, &ds->ds_bp, objset)) {
2086                 printf("ZFS: can't read object set for dataset %ju\n",
2087                     (uintmax_t)objnum);
2088                 return (EIO);
2089         }
2090
2091         return (0);
2092 }
2093
2094 /*
2095  * Find the object set pointed to by the BOOTFS property or the root
2096  * dataset if there is none and return its details in *objset
2097  */
2098 static int
2099 zfs_get_root(const spa_t *spa, uint64_t *objid)
2100 {
2101         dnode_phys_t dir, propdir;
2102         uint64_t props, bootfs, root;
2103
2104         *objid = 0;
2105
2106         /*
2107          * Start with the MOS directory object.
2108          */
2109         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
2110                 printf("ZFS: can't read MOS object directory\n");
2111                 return (EIO);
2112         }
2113
2114         /*
2115          * Lookup the pool_props and see if we can find a bootfs.
2116          */
2117         if (zap_lookup(spa, &dir, DMU_POOL_PROPS, sizeof (props), 1, &props) == 0
2118              && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
2119              && zap_lookup(spa, &propdir, "bootfs", sizeof (bootfs), 1, &bootfs) == 0
2120              && bootfs != 0)
2121         {
2122                 *objid = bootfs;
2123                 return (0);
2124         }
2125         /*
2126          * Lookup the root dataset directory
2127          */
2128         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (root), 1, &root)
2129             || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
2130                 printf("ZFS: can't find root dsl_dir\n");
2131                 return (EIO);
2132         }
2133
2134         /*
2135          * Use the information from the dataset directory's bonus buffer
2136          * to find the dataset object and from that the object set itself.
2137          */
2138         dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
2139         *objid = dd->dd_head_dataset_obj;
2140         return (0);
2141 }
2142
2143 static int
2144 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
2145 {
2146
2147         mount->spa = spa;
2148
2149         /*
2150          * Find the root object set if not explicitly provided
2151          */
2152         if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
2153                 printf("ZFS: can't find root filesystem\n");
2154                 return (EIO);
2155         }
2156
2157         if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
2158                 printf("ZFS: can't open root filesystem\n");
2159                 return (EIO);
2160         }
2161
2162         mount->rootobj = rootobj;
2163
2164         return (0);
2165 }
2166
2167 /*
2168  * callback function for feature name checks.
2169  */
2170 static int
2171 check_feature(const char *name, uint64_t value)
2172 {
2173         int i;
2174
2175         if (value == 0)
2176                 return (0);
2177         if (name[0] == '\0')
2178                 return (0);
2179
2180         for (i = 0; features_for_read[i] != NULL; i++) {
2181                 if (strcmp(name, features_for_read[i]) == 0)
2182                         return (0);
2183         }
2184         printf("ZFS: unsupported feature: %s\n", name);
2185         return (EIO);
2186 }
2187
2188 /*
2189  * Checks whether the MOS features that are active are supported.
2190  */
2191 static int
2192 check_mos_features(const spa_t *spa)
2193 {
2194         dnode_phys_t dir;
2195         uint64_t objnum, zap_type;
2196         size_t size;
2197         int rc;
2198
2199         if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
2200             &dir)) != 0)
2201                 return (rc);
2202         if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
2203             sizeof (objnum), 1, &objnum)) != 0) {
2204                 /*
2205                  * It is older pool without features. As we have already
2206                  * tested the label, just return without raising the error.
2207                  */
2208                 return (0);
2209         }
2210
2211         if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
2212                 return (rc);
2213
2214         if (dir.dn_type != DMU_OTN_ZAP_METADATA)
2215                 return (EIO);
2216
2217         size = dir.dn_datablkszsec * 512;
2218         if (dnode_read(spa, &dir, 0, zap_scratch, size))
2219                 return (EIO);
2220
2221         zap_type = *(uint64_t *) zap_scratch;
2222         if (zap_type == ZBT_MICRO)
2223                 rc = mzap_list(&dir, check_feature);
2224         else
2225                 rc = fzap_list(spa, &dir, check_feature);
2226
2227         return (rc);
2228 }
2229
2230 static int
2231 zfs_spa_init(spa_t *spa)
2232 {
2233         dnode_phys_t dir;
2234         int rc;
2235
2236         if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
2237                 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
2238                 return (EIO);
2239         }
2240         if (spa->spa_mos.os_type != DMU_OST_META) {
2241                 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
2242                 return (EIO);
2243         }
2244
2245         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT,
2246             &dir)) {
2247                 printf("ZFS: failed to read pool %s directory object\n",
2248                     spa->spa_name);
2249                 return (EIO);
2250         }
2251         /* this is allowed to fail, older pools do not have salt */
2252         rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
2253             sizeof (spa->spa_cksum_salt.zcs_bytes),
2254             spa->spa_cksum_salt.zcs_bytes);
2255
2256         rc = check_mos_features(spa);
2257         if (rc != 0) {
2258                 printf("ZFS: pool %s is not supported\n", spa->spa_name);
2259         }
2260
2261         return (rc);
2262 }
2263
2264 static int
2265 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
2266 {
2267
2268         if (dn->dn_bonustype != DMU_OT_SA) {
2269                 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
2270
2271                 sb->st_mode = zp->zp_mode;
2272                 sb->st_uid = zp->zp_uid;
2273                 sb->st_gid = zp->zp_gid;
2274                 sb->st_size = zp->zp_size;
2275         } else {
2276                 sa_hdr_phys_t *sahdrp;
2277                 int hdrsize;
2278                 size_t size = 0;
2279                 void *buf = NULL;
2280
2281                 if (dn->dn_bonuslen != 0)
2282                         sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2283                 else {
2284                         if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
2285                                 blkptr_t *bp = DN_SPILL_BLKPTR(dn);
2286                                 int error;
2287
2288                                 size = BP_GET_LSIZE(bp);
2289                                 buf = zfs_alloc(size);
2290                                 error = zio_read(spa, bp, buf);
2291                                 if (error != 0) {
2292                                         zfs_free(buf, size);
2293                                         return (error);
2294                                 }
2295                                 sahdrp = buf;
2296                         } else {
2297                                 return (EIO);
2298                         }
2299                 }
2300                 hdrsize = SA_HDR_SIZE(sahdrp);
2301                 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
2302                     SA_MODE_OFFSET);
2303                 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
2304                     SA_UID_OFFSET);
2305                 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
2306                     SA_GID_OFFSET);
2307                 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
2308                     SA_SIZE_OFFSET);
2309                 if (buf != NULL)
2310                         zfs_free(buf, size);
2311         }
2312
2313         return (0);
2314 }
2315
2316 static int
2317 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
2318 {
2319         int rc = 0;
2320
2321         if (dn->dn_bonustype == DMU_OT_SA) {
2322                 sa_hdr_phys_t *sahdrp = NULL;
2323                 size_t size = 0;
2324                 void *buf = NULL;
2325                 int hdrsize;
2326                 char *p;
2327
2328                 if (dn->dn_bonuslen != 0)
2329                         sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2330                 else {
2331                         blkptr_t *bp;
2332
2333                         if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
2334                                 return (EIO);
2335                         bp = DN_SPILL_BLKPTR(dn);
2336
2337                         size = BP_GET_LSIZE(bp);
2338                         buf = zfs_alloc(size);
2339                         rc = zio_read(spa, bp, buf);
2340                         if (rc != 0) {
2341                                 zfs_free(buf, size);
2342                                 return (rc);
2343                         }
2344                         sahdrp = buf;
2345                 }
2346                 hdrsize = SA_HDR_SIZE(sahdrp);
2347                 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
2348                 memcpy(path, p, psize);
2349                 if (buf != NULL)
2350                         zfs_free(buf, size);
2351                 return (0);
2352         }
2353         /*
2354          * Second test is purely to silence bogus compiler
2355          * warning about accessing past the end of dn_bonus.
2356          */
2357         if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
2358             sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
2359                 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
2360         } else {
2361                 rc = dnode_read(spa, dn, 0, path, psize);
2362         }
2363         return (rc);
2364 }
2365
2366 struct obj_list {
2367         uint64_t                objnum;
2368         STAILQ_ENTRY(obj_list)  entry;
2369 };
2370
2371 /*
2372  * Lookup a file and return its dnode.
2373  */
2374 static int
2375 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
2376 {
2377         int rc;
2378         uint64_t objnum;
2379         const spa_t *spa;
2380         dnode_phys_t dn;
2381         const char *p, *q;
2382         char element[256];
2383         char path[1024];
2384         int symlinks_followed = 0;
2385         struct stat sb;
2386         struct obj_list *entry, *tentry;
2387         STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
2388
2389         spa = mount->spa;
2390         if (mount->objset.os_type != DMU_OST_ZFS) {
2391                 printf("ZFS: unexpected object set type %ju\n",
2392                     (uintmax_t)mount->objset.os_type);
2393                 return (EIO);
2394         }
2395
2396         if ((entry = malloc(sizeof(struct obj_list))) == NULL)
2397                 return (ENOMEM);
2398
2399         /*
2400          * Get the root directory dnode.
2401          */
2402         rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
2403         if (rc) {
2404                 free(entry);
2405                 return (rc);
2406         }
2407
2408         rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof (objnum), 1, &objnum);
2409         if (rc) {
2410                 free(entry);
2411                 return (rc);
2412         }
2413         entry->objnum = objnum;
2414         STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2415
2416         rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
2417         if (rc != 0)
2418                 goto done;
2419
2420         p = upath;
2421         while (p && *p) {
2422                 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
2423                 if (rc != 0)
2424                         goto done;
2425
2426                 while (*p == '/')
2427                         p++;
2428                 if (*p == '\0')
2429                         break;
2430                 q = p;
2431                 while (*q != '\0' && *q != '/')
2432                         q++;
2433
2434                 /* skip dot */
2435                 if (p + 1 == q && p[0] == '.') {
2436                         p++;
2437                         continue;
2438                 }
2439                 /* double dot */
2440                 if (p + 2 == q && p[0] == '.' && p[1] == '.') {
2441                         p += 2;
2442                         if (STAILQ_FIRST(&on_cache) ==
2443                             STAILQ_LAST(&on_cache, obj_list, entry)) {
2444                                 rc = ENOENT;
2445                                 goto done;
2446                         }
2447                         entry = STAILQ_FIRST(&on_cache);
2448                         STAILQ_REMOVE_HEAD(&on_cache, entry);
2449                         free(entry);
2450                         objnum = (STAILQ_FIRST(&on_cache))->objnum;
2451                         continue;
2452                 }
2453                 if (q - p + 1 > sizeof(element)) {
2454                         rc = ENAMETOOLONG;
2455                         goto done;
2456                 }
2457                 memcpy(element, p, q - p);
2458                 element[q - p] = 0;
2459                 p = q;
2460
2461                 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
2462                         goto done;
2463                 if (!S_ISDIR(sb.st_mode)) {
2464                         rc = ENOTDIR;
2465                         goto done;
2466                 }
2467
2468                 rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
2469                 if (rc)
2470                         goto done;
2471                 objnum = ZFS_DIRENT_OBJ(objnum);
2472
2473                 if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
2474                         rc = ENOMEM;
2475                         goto done;
2476                 }
2477                 entry->objnum = objnum;
2478                 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2479                 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
2480                 if (rc)
2481                         goto done;
2482
2483                 /*
2484                  * Check for symlink.
2485                  */
2486                 rc = zfs_dnode_stat(spa, &dn, &sb);
2487                 if (rc)
2488                         goto done;
2489                 if (S_ISLNK(sb.st_mode)) {
2490                         if (symlinks_followed > 10) {
2491                                 rc = EMLINK;
2492                                 goto done;
2493                         }
2494                         symlinks_followed++;
2495
2496                         /*
2497                          * Read the link value and copy the tail of our
2498                          * current path onto the end.
2499                          */
2500                         if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
2501                                 rc = ENAMETOOLONG;
2502                                 goto done;
2503                         }
2504                         strcpy(&path[sb.st_size], p);
2505
2506                         rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
2507                         if (rc != 0)
2508                                 goto done;
2509
2510                         /*
2511                          * Restart with the new path, starting either at
2512                          * the root or at the parent depending whether or
2513                          * not the link is relative.
2514                          */
2515                         p = path;
2516                         if (*p == '/') {
2517                                 while (STAILQ_FIRST(&on_cache) !=
2518                                     STAILQ_LAST(&on_cache, obj_list, entry)) {
2519                                         entry = STAILQ_FIRST(&on_cache);
2520                                         STAILQ_REMOVE_HEAD(&on_cache, entry);
2521                                         free(entry);
2522                                 }
2523                         } else {
2524                                 entry = STAILQ_FIRST(&on_cache);
2525                                 STAILQ_REMOVE_HEAD(&on_cache, entry);
2526                                 free(entry);
2527                         }
2528                         objnum = (STAILQ_FIRST(&on_cache))->objnum;
2529                 }
2530         }
2531
2532         *dnode = dn;
2533 done:
2534         STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
2535                 free(entry);
2536         return (rc);
2537 }