stand/libsa/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         "com.joyent:multi_vdev_crash_dump",
  66         NULL
  67 };
  68
  69 /*
  70  * List of all pools, chained through spa_link.
  71  */
  72 static spa_list_t zfs_pools;
  73
  74 static const dnode_phys_t *dnode_cache_obj;
  75 static uint64_t dnode_cache_bn;
  76 static char *dnode_cache_buf;
  77 static char *zap_scratch;
  78 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
  79
  80 #define TEMP_SIZE       (1024 * 1024)
  81
  82 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
  83 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
  84 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
  85 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
  86     const char *name, uint64_t integer_size, uint64_t num_integers,
  87     void *value);
  88
  89 static void
  90 zfs_init(void)
  91 {
  92         STAILQ_INIT(&zfs_vdevs);
  93         STAILQ_INIT(&zfs_pools);
  94
  95         zfs_temp_buf = malloc(TEMP_SIZE);
  96         zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
  97         zfs_temp_ptr = zfs_temp_buf;
  98         dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
  99         zap_scratch = malloc(SPA_MAXBLOCKSIZE);
 100
 101         zfs_init_crc();
 102 }
 103
 104 static void *
 105 zfs_alloc(size_t size)
 106 {
 107         char *ptr;
 108
 109         if (zfs_temp_ptr + size > zfs_temp_end) {
 110                 printf("ZFS: out of temporary buffer space\n");
 111                 for (;;) ;
 112         }
 113         ptr = zfs_temp_ptr;
 114         zfs_temp_ptr += size;
 115
 116         return (ptr);
 117 }
 118
 119 static void
 120 zfs_free(void *ptr, size_t size)
 121 {
 122
 123         zfs_temp_ptr -= size;
 124         if (zfs_temp_ptr != ptr) {
 125                 printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
 126                 for (;;) ;
 127         }
 128 }
 129
 130 static int
 131 xdr_int(const unsigned char **xdr, int *ip)
 132 {
 133         *ip = ((*xdr)[0] << 24)
 134                 | ((*xdr)[1] << 16)
 135                 | ((*xdr)[2] << 8)
 136                 | ((*xdr)[3] << 0);
 137         (*xdr) += 4;
 138         return (0);
 139 }
 140
 141 static int
 142 xdr_u_int(const unsigned char **xdr, u_int *ip)
 143 {
 144         *ip = ((*xdr)[0] << 24)
 145                 | ((*xdr)[1] << 16)
 146                 | ((*xdr)[2] << 8)
 147                 | ((*xdr)[3] << 0);
 148         (*xdr) += 4;
 149         return (0);
 150 }
 151
 152 static int
 153 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
 154 {
 155         u_int hi, lo;
 156
 157         xdr_u_int(xdr, &hi);
 158         xdr_u_int(xdr, &lo);
 159         *lp = (((uint64_t) hi) << 32) | lo;
 160         return (0);
 161 }
 162
 163 static int
 164 nvlist_find(const unsigned char *nvlist, const char *name, int type,
 165             int* elementsp, void *valuep)
 166 {
 167         const unsigned char *p, *pair;
 168         int junk;
 169         int encoded_size, decoded_size;
 170
 171         p = nvlist;
 172         xdr_int(&p, &junk);
 173         xdr_int(&p, &junk);
 174
 175         pair = p;
 176         xdr_int(&p, &encoded_size);
 177         xdr_int(&p, &decoded_size);
 178         while (encoded_size && decoded_size) {
 179                 int namelen, pairtype, elements;
 180                 const char *pairname;
 181
 182                 xdr_int(&p, &namelen);
 183                 pairname = (const char*) p;
 184                 p += roundup(namelen, 4);
 185                 xdr_int(&p, &pairtype);
 186
 187                 if (!memcmp(name, pairname, namelen) && type == pairtype) {
 188                         xdr_int(&p, &elements);
 189                         if (elementsp)
 190                                 *elementsp = elements;
 191                         if (type == DATA_TYPE_UINT64) {
 192                                 xdr_uint64_t(&p, (uint64_t *) valuep);
 193                                 return (0);
 194                         } else if (type == DATA_TYPE_STRING) {
 195                                 int len;
 196                                 xdr_int(&p, &len);
 197                                 (*(const char**) valuep) = (const char*) p;
 198                                 return (0);
 199                         } else if (type == DATA_TYPE_NVLIST
 200                                    || type == DATA_TYPE_NVLIST_ARRAY) {
 201                                 (*(const unsigned char**) valuep) =
 202                                          (const unsigned char*) p;
 203                                 return (0);
 204                         } else {
 205                                 return (EIO);
 206                         }
 207                 } else {
 208                         /*
 209                          * Not the pair we are looking for, skip to the next one.
 210                          */
 211                         p = pair + encoded_size;
 212                 }
 213
 214                 pair = p;
 215                 xdr_int(&p, &encoded_size);
 216                 xdr_int(&p, &decoded_size);
 217         }
 218
 219         return (EIO);
 220 }
 221
 222 static int
 223 nvlist_check_features_for_read(const unsigned char *nvlist)
 224 {
 225         const unsigned char *p, *pair;
 226         int junk;
 227         int encoded_size, decoded_size;
 228         int rc;
 229
 230         rc = 0;
 231
 232         p = nvlist;
 233         xdr_int(&p, &junk);
 234         xdr_int(&p, &junk);
 235
 236         pair = p;
 237         xdr_int(&p, &encoded_size);
 238         xdr_int(&p, &decoded_size);
 239         while (encoded_size && decoded_size) {
 240                 int namelen, pairtype;
 241                 const char *pairname;
 242                 int i, found;
 243
 244                 found = 0;
 245
 246                 xdr_int(&p, &namelen);
 247                 pairname = (const char*) p;
 248                 p += roundup(namelen, 4);
 249                 xdr_int(&p, &pairtype);
 250
 251                 for (i = 0; features_for_read[i] != NULL; i++) {
 252                         if (!memcmp(pairname, features_for_read[i], namelen)) {
 253                                 found = 1;
 254                                 break;
 255                         }
 256                 }
 257
 258                 if (!found) {
 259                         printf("ZFS: unsupported feature: %s\n", pairname);
 260                         rc = EIO;
 261                 }
 262
 263                 p = pair + encoded_size;
 264
 265                 pair = p;
 266                 xdr_int(&p, &encoded_size);
 267                 xdr_int(&p, &decoded_size);
 268         }
 269
 270         return (rc);
 271 }
 272
 273 /*
 274  * Return the next nvlist in an nvlist array.
 275  */
 276 static const unsigned char *
 277 nvlist_next(const unsigned char *nvlist)
 278 {
 279         const unsigned char *p, *pair;
 280         int junk;
 281         int encoded_size, decoded_size;
 282
 283         p = nvlist;
 284         xdr_int(&p, &junk);
 285         xdr_int(&p, &junk);
 286
 287         pair = p;
 288         xdr_int(&p, &encoded_size);
 289         xdr_int(&p, &decoded_size);
 290         while (encoded_size && decoded_size) {
 291                 p = pair + encoded_size;
 292
 293                 pair = p;
 294                 xdr_int(&p, &encoded_size);
 295                 xdr_int(&p, &decoded_size);
 296         }
 297
 298         return p;
 299 }
 300
 301 #ifdef TEST
 302
 303 static const unsigned char *
 304 nvlist_print(const unsigned char *nvlist, unsigned int indent)
 305 {
 306         static const char* typenames[] = {
 307                 "DATA_TYPE_UNKNOWN",
 308                 "DATA_TYPE_BOOLEAN",
 309                 "DATA_TYPE_BYTE",
 310                 "DATA_TYPE_INT16",
 311                 "DATA_TYPE_UINT16",
 312                 "DATA_TYPE_INT32",
 313                 "DATA_TYPE_UINT32",
 314                 "DATA_TYPE_INT64",
 315                 "DATA_TYPE_UINT64",
 316                 "DATA_TYPE_STRING",
 317                 "DATA_TYPE_BYTE_ARRAY",
 318                 "DATA_TYPE_INT16_ARRAY",
 319                 "DATA_TYPE_UINT16_ARRAY",
 320                 "DATA_TYPE_INT32_ARRAY",
 321                 "DATA_TYPE_UINT32_ARRAY",
 322                 "DATA_TYPE_INT64_ARRAY",
 323                 "DATA_TYPE_UINT64_ARRAY",
 324                 "DATA_TYPE_STRING_ARRAY",
 325                 "DATA_TYPE_HRTIME",
 326                 "DATA_TYPE_NVLIST",
 327                 "DATA_TYPE_NVLIST_ARRAY",
 328                 "DATA_TYPE_BOOLEAN_VALUE",
 329                 "DATA_TYPE_INT8",
 330                 "DATA_TYPE_UINT8",
 331                 "DATA_TYPE_BOOLEAN_ARRAY",
 332                 "DATA_TYPE_INT8_ARRAY",
 333                 "DATA_TYPE_UINT8_ARRAY"
 334         };
 335
 336         unsigned int i, j;
 337         const unsigned char *p, *pair;
 338         int junk;
 339         int encoded_size, decoded_size;
 340
 341         p = nvlist;
 342         xdr_int(&p, &junk);
 343         xdr_int(&p, &junk);
 344
 345         pair = p;
 346         xdr_int(&p, &encoded_size);
 347         xdr_int(&p, &decoded_size);
 348         while (encoded_size && decoded_size) {
 349                 int namelen, pairtype, elements;
 350                 const char *pairname;
 351
 352                 xdr_int(&p, &namelen);
 353                 pairname = (const char*) p;
 354                 p += roundup(namelen, 4);
 355                 xdr_int(&p, &pairtype);
 356
 357                 for (i = 0; i < indent; i++)
 358                         printf(" ");
 359                 printf("%s %s", typenames[pairtype], pairname);
 360
 361                 xdr_int(&p, &elements);
 362                 switch (pairtype) {
 363                 case DATA_TYPE_UINT64: {
 364                         uint64_t val;
 365                         xdr_uint64_t(&p, &val);
 366                         printf(" = 0x%jx\n", (uintmax_t)val);
 367                         break;
 368                 }
 369
 370                 case DATA_TYPE_STRING: {
 371                         int len;
 372                         xdr_int(&p, &len);
 373                         printf(" = \"%s\"\n", p);
 374                         break;
 375                 }
 376
 377                 case DATA_TYPE_NVLIST:
 378                         printf("\n");
 379                         nvlist_print(p, indent + 1);
 380                         break;
 381
 382                 case DATA_TYPE_NVLIST_ARRAY:
 383                         for (j = 0; j < elements; j++) {
 384                                 printf("[%d]\n", j);
 385                                 p = nvlist_print(p, indent + 1);
 386                                 if (j != elements - 1) {
 387                                         for (i = 0; i < indent; i++)
 388                                                 printf(" ");
 389                                         printf("%s %s", typenames[pairtype], pairname);
 390                                 }
 391                         }
 392                         break;
 393
 394                 default:
 395                         printf("\n");
 396                 }
 397
 398                 p = pair + encoded_size;
 399
 400                 pair = p;
 401                 xdr_int(&p, &encoded_size);
 402                 xdr_int(&p, &decoded_size);
 403         }
 404
 405         return p;
 406 }
 407
 408 #endif
 409
 410 static int
 411 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
 412     off_t offset, size_t size)
 413 {
 414         size_t psize;
 415         int rc;
 416
 417         if (!vdev->v_phys_read)
 418                 return (EIO);
 419
 420         if (bp) {
 421                 psize = BP_GET_PSIZE(bp);
 422         } else {
 423                 psize = size;
 424         }
 425
 426         /*printf("ZFS: reading %zu bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
 427         rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
 428         if (rc)
 429                 return (rc);
 430         if (bp && zio_checksum_verify(vdev->spa, bp, buf))
 431                 return (EIO);
 432
 433         return (0);
 434 }
 435
 436 static int
 437 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 438     off_t offset, size_t bytes)
 439 {
 440
 441         return (vdev_read_phys(vdev, bp, buf,
 442                 offset + VDEV_LABEL_START_SIZE, bytes));
 443 }
 444
 445
 446 static int
 447 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 448     off_t offset, size_t bytes)
 449 {
 450         vdev_t *kid;
 451         int rc;
 452
 453         rc = EIO;
 454         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 455                 if (kid->v_state != VDEV_STATE_HEALTHY)
 456                         continue;
 457                 rc = kid->v_read(kid, bp, buf, offset, bytes);
 458                 if (!rc)
 459                         return (0);
 460         }
 461
 462         return (rc);
 463 }
 464
 465 static int
 466 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 467     off_t offset, size_t bytes)
 468 {
 469         vdev_t *kid;
 470
 471         /*
 472          * Here we should have two kids:
 473          * First one which is the one we are replacing and we can trust
 474          * only this one to have valid data, but it might not be present.
 475          * Second one is that one we are replacing with. It is most likely
 476          * healthy, but we can't trust it has needed data, so we won't use it.
 477          */
 478         kid = STAILQ_FIRST(&vdev->v_children);
 479         if (kid == NULL)
 480                 return (EIO);
 481         if (kid->v_state != VDEV_STATE_HEALTHY)
 482                 return (EIO);
 483         return (kid->v_read(kid, bp, buf, offset, bytes));
 484 }
 485
 486 static vdev_t *
 487 vdev_find(uint64_t guid)
 488 {
 489         vdev_t *vdev;
 490
 491         STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
 492                 if (vdev->v_guid == guid)
 493                         return (vdev);
 494
 495         return (0);
 496 }
 497
 498 static vdev_t *
 499 vdev_create(uint64_t guid, vdev_read_t *_read)
 500 {
 501         vdev_t *vdev;
 502
 503         vdev = malloc(sizeof(vdev_t));
 504         memset(vdev, 0, sizeof(vdev_t));
 505         STAILQ_INIT(&vdev->v_children);
 506         vdev->v_guid = guid;
 507         vdev->v_state = VDEV_STATE_OFFLINE;
 508         vdev->v_read = _read;
 509         vdev->v_phys_read = 0;
 510         vdev->v_read_priv = 0;
 511         STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
 512
 513         return (vdev);
 514 }
 515
 516 static int
 517 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
 518     vdev_t **vdevp, int is_newer)
 519 {
 520         int rc;
 521         uint64_t guid, id, ashift, nparity;
 522         const char *type;
 523         const char *path;
 524         vdev_t *vdev, *kid;
 525         const unsigned char *kids;
 526         int nkids, i, is_new;
 527         uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
 528
 529         if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
 530             NULL, &guid)
 531             || nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id)
 532             || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
 533             NULL, &type)) {
 534                 printf("ZFS: can't find vdev details\n");
 535                 return (ENOENT);
 536         }
 537
 538         if (strcmp(type, VDEV_TYPE_MIRROR)
 539             && strcmp(type, VDEV_TYPE_DISK)
 540 #ifdef ZFS_TEST
 541             && strcmp(type, VDEV_TYPE_FILE)
 542 #endif
 543             && strcmp(type, VDEV_TYPE_RAIDZ)
 544             && strcmp(type, VDEV_TYPE_REPLACING)) {
 545                 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 546                 return (EIO);
 547         }
 548
 549         is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
 550
 551         nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
 552                         &is_offline);
 553         nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
 554                         &is_removed);
 555         nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
 556                         &is_faulted);
 557         nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, NULL,
 558                         &is_degraded);
 559         nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, NULL,
 560                         &isnt_present);
 561
 562         vdev = vdev_find(guid);
 563         if (!vdev) {
 564                 is_new = 1;
 565
 566                 if (!strcmp(type, VDEV_TYPE_MIRROR))
 567                         vdev = vdev_create(guid, vdev_mirror_read);
 568                 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
 569                         vdev = vdev_create(guid, vdev_raidz_read);
 570                 else if (!strcmp(type, VDEV_TYPE_REPLACING))
 571                         vdev = vdev_create(guid, vdev_replacing_read);
 572                 else
 573                         vdev = vdev_create(guid, vdev_disk_read);
 574
 575                 vdev->v_id = id;
 576                 vdev->v_top = pvdev != NULL ? pvdev : vdev;
 577                 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
 578                         DATA_TYPE_UINT64, NULL, &ashift) == 0) {
 579                         vdev->v_ashift = ashift;
 580                 } else {
 581                         vdev->v_ashift = 0;
 582                 }
 583                 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
 584                         DATA_TYPE_UINT64, NULL, &nparity) == 0) {
 585                         vdev->v_nparity = nparity;
 586                 } else {
 587                         vdev->v_nparity = 0;
 588                 }
 589                 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
 590                                 DATA_TYPE_STRING, NULL, &path) == 0) {
 591                         if (strncmp(path, "/dev/", 5) == 0)
 592                                 path += 5;
 593                         vdev->v_name = strdup(path);
 594                 } else {
 595                         if (!strcmp(type, "raidz")) {
 596                                 if (vdev->v_nparity == 1)
 597                                         vdev->v_name = "raidz1";
 598                                 else if (vdev->v_nparity == 2)
 599                                         vdev->v_name = "raidz2";
 600                                 else if (vdev->v_nparity == 3)
 601                                         vdev->v_name = "raidz3";
 602                                 else {
 603                                         printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 604                                         return (EIO);
 605                                 }
 606                         } else {
 607                                 vdev->v_name = strdup(type);
 608                         }
 609                 }
 610         } else {
 611                 is_new = 0;
 612         }
 613
 614         if (is_new || is_newer) {
 615                 /*
 616                  * This is either new vdev or we've already seen this vdev,
 617                  * but from an older vdev label, so let's refresh its state
 618                  * from the newer label.
 619                  */
 620                 if (is_offline)
 621                         vdev->v_state = VDEV_STATE_OFFLINE;
 622                 else if (is_removed)
 623                         vdev->v_state = VDEV_STATE_REMOVED;
 624                 else if (is_faulted)
 625                         vdev->v_state = VDEV_STATE_FAULTED;
 626                 else if (is_degraded)
 627                         vdev->v_state = VDEV_STATE_DEGRADED;
 628                 else if (isnt_present)
 629                         vdev->v_state = VDEV_STATE_CANT_OPEN;
 630         }
 631
 632         rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
 633             &nkids, &kids);
 634         /*
 635          * Its ok if we don't have any kids.
 636          */
 637         if (rc == 0) {
 638                 vdev->v_nchildren = nkids;
 639                 for (i = 0; i < nkids; i++) {
 640                         rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
 641                         if (rc)
 642                                 return (rc);
 643                         if (is_new)
 644                                 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
 645                                                    v_childlink);
 646                         kids = nvlist_next(kids);
 647                 }
 648         } else {
 649                 vdev->v_nchildren = 0;
 650         }
 651
 652         if (vdevp)
 653                 *vdevp = vdev;
 654         return (0);
 655 }
 656
 657 static void
 658 vdev_set_state(vdev_t *vdev)
 659 {
 660         vdev_t *kid;
 661         int good_kids;
 662         int bad_kids;
 663
 664         /*
 665          * A mirror or raidz is healthy if all its kids are healthy. A
 666          * mirror is degraded if any of its kids is healthy; a raidz
 667          * is degraded if at most nparity kids are offline.
 668          */
 669         if (STAILQ_FIRST(&vdev->v_children)) {
 670                 good_kids = 0;
 671                 bad_kids = 0;
 672                 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 673                         if (kid->v_state == VDEV_STATE_HEALTHY)
 674                                 good_kids++;
 675                         else
 676                                 bad_kids++;
 677                 }
 678                 if (bad_kids == 0) {
 679                         vdev->v_state = VDEV_STATE_HEALTHY;
 680                 } else {
 681                         if (vdev->v_read == vdev_mirror_read) {
 682                                 if (good_kids) {
 683                                         vdev->v_state = VDEV_STATE_DEGRADED;
 684                                 } else {
 685                                         vdev->v_state = VDEV_STATE_OFFLINE;
 686                                 }
 687                         } else if (vdev->v_read == vdev_raidz_read) {
 688                                 if (bad_kids > vdev->v_nparity) {
 689                                         vdev->v_state = VDEV_STATE_OFFLINE;
 690                                 } else {
 691                                         vdev->v_state = VDEV_STATE_DEGRADED;
 692                                 }
 693                         }
 694                 }
 695         }
 696 }
 697
 698 static spa_t *
 699 spa_find_by_guid(uint64_t guid)
 700 {
 701         spa_t *spa;
 702
 703         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 704                 if (spa->spa_guid == guid)
 705                         return (spa);
 706
 707         return (0);
 708 }
 709
 710 static spa_t *
 711 spa_find_by_name(const char *name)
 712 {
 713         spa_t *spa;
 714
 715         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 716                 if (!strcmp(spa->spa_name, name))
 717                         return (spa);
 718
 719         return (0);
 720 }
 721
 722 #ifdef BOOT2
 723 static spa_t *
 724 spa_get_primary(void)
 725 {
 726
 727         return (STAILQ_FIRST(&zfs_pools));
 728 }
 729
 730 static vdev_t *
 731 spa_get_primary_vdev(const spa_t *spa)
 732 {
 733         vdev_t *vdev;
 734         vdev_t *kid;
 735
 736         if (spa == NULL)
 737                 spa = spa_get_primary();
 738         if (spa == NULL)
 739                 return (NULL);
 740         vdev = STAILQ_FIRST(&spa->spa_vdevs);
 741         if (vdev == NULL)
 742                 return (NULL);
 743         for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
 744              kid = STAILQ_FIRST(&vdev->v_children))
 745                 vdev = kid;
 746         return (vdev);
 747 }
 748 #endif
 749
 750 static spa_t *
 751 spa_create(uint64_t guid, const char *name)
 752 {
 753         spa_t *spa;
 754
 755         if ((spa = malloc(sizeof(spa_t))) == NULL)
 756                 return (NULL);
 757         memset(spa, 0, sizeof(spa_t));
 758         if ((spa->spa_name = strdup(name)) == NULL) {
 759                 free(spa);
 760                 return (NULL);
 761         }
 762         STAILQ_INIT(&spa->spa_vdevs);
 763         spa->spa_guid = guid;
 764         STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
 765
 766         return (spa);
 767 }
 768
 769 static const char *
 770 state_name(vdev_state_t state)
 771 {
 772         static const char* names[] = {
 773                 "UNKNOWN",
 774                 "CLOSED",
 775                 "OFFLINE",
 776                 "REMOVED",
 777                 "CANT_OPEN",
 778                 "FAULTED",
 779                 "DEGRADED",
 780                 "ONLINE"
 781         };
 782         return names[state];
 783 }
 784
 785 #ifdef BOOT2
 786
 787 #define pager_printf printf
 788
 789 #else
 790
 791 static int
 792 pager_printf(const char *fmt, ...)
 793 {
 794         char line[80];
 795         va_list args;
 796
 797         va_start(args, fmt);
 798         vsprintf(line, fmt, args);
 799         va_end(args);
 800
 801         return (pager_output(line));
 802 }
 803
 804 #endif
 805
 806 #define STATUS_FORMAT   "        %s %s\n"
 807
 808 static int
 809 print_state(int indent, const char *name, vdev_state_t state)
 810 {
 811         char buf[512];
 812         int i;
 813
 814         buf[0] = 0;
 815         for (i = 0; i < indent; i++)
 816                 strcat(buf, "  ");
 817         strcat(buf, name);
 818
 819         return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
 820 }
 821
 822 static int
 823 vdev_status(vdev_t *vdev, int indent)
 824 {
 825         vdev_t *kid;
 826         int ret;
 827         ret = print_state(indent, vdev->v_name, vdev->v_state);
 828         if (ret != 0)
 829                 return (ret);
 830
 831         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 832                 ret = vdev_status(kid, indent + 1);
 833                 if (ret != 0)
 834                         return (ret);
 835         }
 836         return (ret);
 837 }
 838
 839 static int
 840 spa_status(spa_t *spa)
 841 {
 842         static char bootfs[ZFS_MAXNAMELEN];
 843         uint64_t rootid;
 844         vdev_t *vdev;
 845         int good_kids, bad_kids, degraded_kids, ret;
 846         vdev_state_t state;
 847
 848         ret = pager_printf("  pool: %s\n", spa->spa_name);
 849         if (ret != 0)
 850                 return (ret);
 851
 852         if (zfs_get_root(spa, &rootid) == 0 &&
 853             zfs_rlookup(spa, rootid, bootfs) == 0) {
 854                 if (bootfs[0] == '\0')
 855                         ret = pager_printf("bootfs: %s\n", spa->spa_name);
 856                 else
 857                         ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
 858                             bootfs);
 859                 if (ret != 0)
 860                         return (ret);
 861         }
 862         ret = pager_printf("config:\n\n");
 863         if (ret != 0)
 864                 return (ret);
 865         ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
 866         if (ret != 0)
 867                 return (ret);
 868
 869         good_kids = 0;
 870         degraded_kids = 0;
 871         bad_kids = 0;
 872         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 873                 if (vdev->v_state == VDEV_STATE_HEALTHY)
 874                         good_kids++;
 875                 else if (vdev->v_state == VDEV_STATE_DEGRADED)
 876                         degraded_kids++;
 877                 else
 878                         bad_kids++;
 879         }
 880
 881         state = VDEV_STATE_CLOSED;
 882         if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
 883                 state = VDEV_STATE_HEALTHY;
 884         else if ((good_kids + degraded_kids) > 0)
 885                 state = VDEV_STATE_DEGRADED;
 886
 887         ret = print_state(0, spa->spa_name, state);
 888         if (ret != 0)
 889                 return (ret);
 890         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 891                 ret = vdev_status(vdev, 1);
 892                 if (ret != 0)
 893                         return (ret);
 894         }
 895         return (ret);
 896 }
 897
 898 static int
 899 spa_all_status(void)
 900 {
 901         spa_t *spa;
 902         int first = 1, ret = 0;
 903
 904         STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
 905                 if (!first) {
 906                         ret = pager_printf("\n");
 907                         if (ret != 0)
 908                                 return (ret);
 909                 }
 910                 first = 0;
 911                 ret = spa_status(spa);
 912                 if (ret != 0)
 913                         return (ret);
 914         }
 915         return (ret);
 916 }
 917
 918 static uint64_t
 919 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
 920 {
 921         uint64_t label_offset;
 922
 923         if (l < VDEV_LABELS / 2)
 924                 label_offset = 0;
 925         else
 926                 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
 927
 928         return (offset + l * sizeof (vdev_label_t) + label_offset);
 929 }
 930
 931 static int
 932 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
 933 {
 934         vdev_t vtmp;
 935         vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
 936         vdev_phys_t *tmp_label;
 937         spa_t *spa;
 938         vdev_t *vdev, *top_vdev, *pool_vdev;
 939         off_t off;
 940         blkptr_t bp;
 941         const unsigned char *nvlist = NULL;
 942         uint64_t val;
 943         uint64_t guid;
 944         uint64_t best_txg = 0;
 945         uint64_t pool_txg, pool_guid;
 946         uint64_t psize;
 947         const char *pool_name;
 948         const unsigned char *vdevs;
 949         const unsigned char *features;
 950         int i, l, rc, is_newer;
 951         char *upbuf;
 952         const struct uberblock *up;
 953
 954         /*
 955          * Load the vdev label and figure out which
 956          * uberblock is most current.
 957          */
 958         memset(&vtmp, 0, sizeof(vtmp));
 959         vtmp.v_phys_read = _read;
 960         vtmp.v_read_priv = read_priv;
 961         psize = P2ALIGN(ldi_get_size(read_priv),
 962             (uint64_t)sizeof (vdev_label_t));
 963
 964         /* Test for minimum pool size. */
 965         if (psize < SPA_MINDEVSIZE)
 966                 return (EIO);
 967
 968         tmp_label = zfs_alloc(sizeof(vdev_phys_t));
 969
 970         for (l = 0; l < VDEV_LABELS; l++) {
 971                 off = vdev_label_offset(psize, l,
 972                     offsetof(vdev_label_t, vl_vdev_phys));
 973
 974                 BP_ZERO(&bp);
 975                 BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
 976                 BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
 977                 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
 978                 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
 979                 DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
 980                 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
 981
 982                 if (vdev_read_phys(&vtmp, &bp, tmp_label, off, 0))
 983                         continue;
 984
 985                 if (tmp_label->vp_nvlist[0] != NV_ENCODE_XDR)
 986                         continue;
 987
 988                 nvlist = (const unsigned char *) tmp_label->vp_nvlist + 4;
 989                 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
 990                     DATA_TYPE_UINT64, NULL, &pool_txg) != 0)
 991                         continue;
 992
 993                 if (best_txg <= pool_txg) {
 994                         best_txg = pool_txg;
 995                         memcpy(vdev_label, tmp_label, sizeof (vdev_phys_t));
 996                 }
 997         }
 998
 999         zfs_free(tmp_label, sizeof (vdev_phys_t));
1000
1001         if (best_txg == 0)
1002                 return (EIO);
1003
1004         if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR)
1005                 return (EIO);
1006
1007         nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
1008
1009         if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1010             NULL, &val) != 0) {
1011                 return (EIO);
1012         }
1013
1014         if (!SPA_VERSION_IS_SUPPORTED(val)) {
1015                 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1016                     (unsigned) val, (unsigned) SPA_VERSION);
1017                 return (EIO);
1018         }
1019
1020         /* Check ZFS features for read */
1021         if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1022             DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1023             nvlist_check_features_for_read(features) != 0) {
1024                 return (EIO);
1025         }
1026
1027         if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1028             NULL, &val) != 0) {
1029                 return (EIO);
1030         }
1031
1032         if (val == POOL_STATE_DESTROYED) {
1033                 /* We don't boot only from destroyed pools. */
1034                 return (EIO);
1035         }
1036
1037         if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1038             NULL, &pool_txg) != 0 ||
1039             nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1040             NULL, &pool_guid) != 0 ||
1041             nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1042             NULL, &pool_name) != 0) {
1043                 /*
1044                  * Cache and spare devices end up here - just ignore
1045                  * them.
1046                  */
1047                 /*printf("ZFS: can't find pool details\n");*/
1048                 return (EIO);
1049         }
1050
1051         if (nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64,
1052             NULL, &val) == 0 && val != 0) {
1053                 return (EIO);
1054         }
1055
1056         /*
1057          * Create the pool if this is the first time we've seen it.
1058          */
1059         spa = spa_find_by_guid(pool_guid);
1060         if (spa == NULL) {
1061                 spa = spa_create(pool_guid, pool_name);
1062                 if (spa == NULL)
1063                         return (ENOMEM);
1064         }
1065         if (pool_txg > spa->spa_txg) {
1066                 spa->spa_txg = pool_txg;
1067                 is_newer = 1;
1068         } else {
1069                 is_newer = 0;
1070         }
1071
1072         /*
1073          * Get the vdev tree and create our in-core copy of it.
1074          * If we already have a vdev with this guid, this must
1075          * be some kind of alias (overlapping slices, dangerously dedicated
1076          * disks etc).
1077          */
1078         if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1079             NULL, &guid) != 0) {
1080                 return (EIO);
1081         }
1082         vdev = vdev_find(guid);
1083         if (vdev && vdev->v_phys_read)  /* Has this vdev already been inited? */
1084                 return (EIO);
1085
1086         if (nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1087             NULL, &vdevs)) {
1088                 return (EIO);
1089         }
1090
1091         rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1092         if (rc != 0)
1093                 return (rc);
1094
1095         /*
1096          * Add the toplevel vdev to the pool if its not already there.
1097          */
1098         STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1099                 if (top_vdev == pool_vdev)
1100                         break;
1101         if (!pool_vdev && top_vdev) {
1102                 top_vdev->spa = spa;
1103                 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1104         }
1105
1106         /*
1107          * We should already have created an incomplete vdev for this
1108          * vdev. Find it and initialise it with our read proc.
1109          */
1110         vdev = vdev_find(guid);
1111         if (vdev) {
1112                 vdev->v_phys_read = _read;
1113                 vdev->v_read_priv = read_priv;
1114                 vdev->v_state = VDEV_STATE_HEALTHY;
1115         } else {
1116                 printf("ZFS: inconsistent nvlist contents\n");
1117                 return (EIO);
1118         }
1119
1120         /*
1121          * Re-evaluate top-level vdev state.
1122          */
1123         vdev_set_state(top_vdev);
1124
1125         /*
1126          * Ok, we are happy with the pool so far. Lets find
1127          * the best uberblock and then we can actually access
1128          * the contents of the pool.
1129          */
1130         upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
1131         up = (const struct uberblock *)upbuf;
1132         for (l = 0; l < VDEV_LABELS; l++) {
1133                 for (i = 0; i < VDEV_UBERBLOCK_COUNT(vdev); i++) {
1134                         off = vdev_label_offset(psize, l,
1135                             VDEV_UBERBLOCK_OFFSET(vdev, i));
1136                         BP_ZERO(&bp);
1137                         DVA_SET_OFFSET(&bp.blk_dva[0], off);
1138                         BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1139                         BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1140                         BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1141                         BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1142                         ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1143
1144                         if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
1145                                 continue;
1146
1147                         if (up->ub_magic != UBERBLOCK_MAGIC)
1148                                 continue;
1149                         if (up->ub_txg < spa->spa_txg)
1150                                 continue;
1151                         if (up->ub_txg > spa->spa_uberblock.ub_txg ||
1152                             (up->ub_txg == spa->spa_uberblock.ub_txg &&
1153                             up->ub_timestamp >
1154                             spa->spa_uberblock.ub_timestamp)) {
1155                                 spa->spa_uberblock = *up;
1156                         }
1157                 }
1158         }
1159         zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
1160
1161         vdev->spa = spa;
1162         if (spap != NULL)
1163                 *spap = spa;
1164         return (0);
1165 }
1166
1167 static int
1168 ilog2(int n)
1169 {
1170         int v;
1171
1172         for (v = 0; v < 32; v++)
1173                 if (n == (1 << v))
1174                         return v;
1175         return -1;
1176 }
1177
1178 static int
1179 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1180 {
1181         blkptr_t gbh_bp;
1182         zio_gbh_phys_t zio_gb;
1183         char *pbuf;
1184         int i;
1185
1186         /* Artificial BP for gang block header. */
1187         gbh_bp = *bp;
1188         BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1189         BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1190         BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1191         BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1192         for (i = 0; i < SPA_DVAS_PER_BP; i++)
1193                 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1194
1195         /* Read gang header block using the artificial BP. */
1196         if (zio_read(spa, &gbh_bp, &zio_gb))
1197                 return (EIO);
1198
1199         pbuf = buf;
1200         for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1201                 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1202
1203                 if (BP_IS_HOLE(gbp))
1204                         continue;
1205                 if (zio_read(spa, gbp, pbuf))
1206                         return (EIO);
1207                 pbuf += BP_GET_PSIZE(gbp);
1208         }
1209
1210         if (zio_checksum_verify(spa, bp, buf))
1211                 return (EIO);
1212         return (0);
1213 }
1214
1215 static int
1216 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1217 {
1218         int cpfunc = BP_GET_COMPRESS(bp);
1219         uint64_t align, size;
1220         void *pbuf;
1221         int i, error;
1222
1223         /*
1224          * Process data embedded in block pointer
1225          */
1226         if (BP_IS_EMBEDDED(bp)) {
1227                 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1228
1229                 size = BPE_GET_PSIZE(bp);
1230                 ASSERT(size <= BPE_PAYLOAD_SIZE);
1231
1232                 if (cpfunc != ZIO_COMPRESS_OFF)
1233                         pbuf = zfs_alloc(size);
1234                 else
1235                         pbuf = buf;
1236
1237                 decode_embedded_bp_compressed(bp, pbuf);
1238                 error = 0;
1239
1240                 if (cpfunc != ZIO_COMPRESS_OFF) {
1241                         error = zio_decompress_data(cpfunc, pbuf,
1242                             size, buf, BP_GET_LSIZE(bp));
1243                         zfs_free(pbuf, size);
1244                 }
1245                 if (error != 0)
1246                         printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
1247                             error);
1248                 return (error);
1249         }
1250
1251         error = EIO;
1252
1253         for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1254                 const dva_t *dva = &bp->blk_dva[i];
1255                 vdev_t *vdev;
1256                 int vdevid;
1257                 off_t offset;
1258
1259                 if (!dva->dva_word[0] && !dva->dva_word[1])
1260                         continue;
1261
1262                 vdevid = DVA_GET_VDEV(dva);
1263                 offset = DVA_GET_OFFSET(dva);
1264                 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1265                         if (vdev->v_id == vdevid)
1266                                 break;
1267                 }
1268                 if (!vdev || !vdev->v_read)
1269                         continue;
1270
1271                 size = BP_GET_PSIZE(bp);
1272                 if (vdev->v_read == vdev_raidz_read) {
1273                         align = 1ULL << vdev->v_top->v_ashift;
1274                         if (P2PHASE(size, align) != 0)
1275                                 size = P2ROUNDUP(size, align);
1276                 }
1277                 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1278                         pbuf = zfs_alloc(size);
1279                 else
1280                         pbuf = buf;
1281
1282                 if (DVA_GET_GANG(dva))
1283                         error = zio_read_gang(spa, bp, pbuf);
1284                 else
1285                         error = vdev->v_read(vdev, bp, pbuf, offset, size);
1286                 if (error == 0) {
1287                         if (cpfunc != ZIO_COMPRESS_OFF)
1288                                 error = zio_decompress_data(cpfunc, pbuf,
1289                                     BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1290                         else if (size != BP_GET_PSIZE(bp))
1291                                 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1292                 }
1293                 if (buf != pbuf)
1294                         zfs_free(pbuf, size);
1295                 if (error == 0)
1296                         break;
1297         }
1298         if (error != 0)
1299                 printf("ZFS: i/o error - all block copies unavailable\n");
1300         return (error);
1301 }
1302
1303 static int
1304 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1305 {
1306         int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1307         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1308         int nlevels = dnode->dn_nlevels;
1309         int i, rc;
1310
1311         if (bsize > SPA_MAXBLOCKSIZE) {
1312                 printf("ZFS: I/O error - blocks larger than %llu are not "
1313                     "supported\n", SPA_MAXBLOCKSIZE);
1314                 return (EIO);
1315         }
1316
1317         /*
1318          * Note: bsize may not be a power of two here so we need to do an
1319          * actual divide rather than a bitshift.
1320          */
1321         while (buflen > 0) {
1322                 uint64_t bn = offset / bsize;
1323                 int boff = offset % bsize;
1324                 int ibn;
1325                 const blkptr_t *indbp;
1326                 blkptr_t bp;
1327
1328                 if (bn > dnode->dn_maxblkid)
1329                         return (EIO);
1330
1331                 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1332                         goto cached;
1333
1334                 indbp = dnode->dn_blkptr;
1335                 for (i = 0; i < nlevels; i++) {
1336                         /*
1337                          * Copy the bp from the indirect array so that
1338                          * we can re-use the scratch buffer for multi-level
1339                          * objects.
1340                          */
1341                         ibn = bn >> ((nlevels - i - 1) * ibshift);
1342                         ibn &= ((1 << ibshift) - 1);
1343                         bp = indbp[ibn];
1344                         if (BP_IS_HOLE(&bp)) {
1345                                 memset(dnode_cache_buf, 0, bsize);
1346                                 break;
1347                         }
1348                         rc = zio_read(spa, &bp, dnode_cache_buf);
1349                         if (rc)
1350                                 return (rc);
1351                         indbp = (const blkptr_t *) dnode_cache_buf;
1352                 }
1353                 dnode_cache_obj = dnode;
1354                 dnode_cache_bn = bn;
1355         cached:
1356
1357                 /*
1358                  * The buffer contains our data block. Copy what we
1359                  * need from it and loop.
1360                  */
1361                 i = bsize - boff;
1362                 if (i > buflen) i = buflen;
1363                 memcpy(buf, &dnode_cache_buf[boff], i);
1364                 buf = ((char*) buf) + i;
1365                 offset += i;
1366                 buflen -= i;
1367         }
1368
1369         return (0);
1370 }
1371
1372 /*
1373  * Lookup a value in a microzap directory. Assumes that the zap
1374  * scratch buffer contains the directory contents.
1375  */
1376 static int
1377 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1378 {
1379         const mzap_phys_t *mz;
1380         const mzap_ent_phys_t *mze;
1381         size_t size;
1382         int chunks, i;
1383
1384         /*
1385          * Microzap objects use exactly one block. Read the whole
1386          * thing.
1387          */
1388         size = dnode->dn_datablkszsec * 512;
1389
1390         mz = (const mzap_phys_t *) zap_scratch;
1391         chunks = size / MZAP_ENT_LEN - 1;
1392
1393         for (i = 0; i < chunks; i++) {
1394                 mze = &mz->mz_chunk[i];
1395                 if (!strcmp(mze->mze_name, name)) {
1396                         *value = mze->mze_value;
1397                         return (0);
1398                 }
1399         }
1400
1401         return (ENOENT);
1402 }
1403
1404 /*
1405  * Compare a name with a zap leaf entry. Return non-zero if the name
1406  * matches.
1407  */
1408 static int
1409 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1410 {
1411         size_t namelen;
1412         const zap_leaf_chunk_t *nc;
1413         const char *p;
1414
1415         namelen = zc->l_entry.le_name_numints;
1416
1417         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1418         p = name;
1419         while (namelen > 0) {
1420                 size_t len;
1421                 len = namelen;
1422                 if (len > ZAP_LEAF_ARRAY_BYTES)
1423                         len = ZAP_LEAF_ARRAY_BYTES;
1424                 if (memcmp(p, nc->l_array.la_array, len))
1425                         return (0);
1426                 p += len;
1427                 namelen -= len;
1428                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1429         }
1430
1431         return 1;
1432 }
1433
1434 /*
1435  * Extract a uint64_t value from a zap leaf entry.
1436  */
1437 static uint64_t
1438 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1439 {
1440         const zap_leaf_chunk_t *vc;
1441         int i;
1442         uint64_t value;
1443         const uint8_t *p;
1444
1445         vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1446         for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1447                 value = (value << 8) | p[i];
1448         }
1449
1450         return value;
1451 }
1452
1453 static void
1454 stv(int len, void *addr, uint64_t value)
1455 {
1456         switch (len) {
1457         case 1:
1458                 *(uint8_t *)addr = value;
1459                 return;
1460         case 2:
1461                 *(uint16_t *)addr = value;
1462                 return;
1463         case 4:
1464                 *(uint32_t *)addr = value;
1465                 return;
1466         case 8:
1467                 *(uint64_t *)addr = value;
1468                 return;
1469         }
1470 }
1471
1472 /*
1473  * Extract a array from a zap leaf entry.
1474  */
1475 static void
1476 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
1477     uint64_t integer_size, uint64_t num_integers, void *buf)
1478 {
1479         uint64_t array_int_len = zc->l_entry.le_value_intlen;
1480         uint64_t value = 0;
1481         uint64_t *u64 = buf;
1482         char *p = buf;
1483         int len = MIN(zc->l_entry.le_value_numints, num_integers);
1484         int chunk = zc->l_entry.le_value_chunk;
1485         int byten = 0;
1486
1487         if (integer_size == 8 && len == 1) {
1488                 *u64 = fzap_leaf_value(zl, zc);
1489                 return;
1490         }
1491
1492         while (len > 0) {
1493                 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
1494                 int i;
1495
1496                 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
1497                 for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
1498                         value = (value << 8) | la->la_array[i];
1499                         byten++;
1500                         if (byten == array_int_len) {
1501                                 stv(integer_size, p, value);
1502                                 byten = 0;
1503                                 len--;
1504                                 if (len == 0)
1505                                         return;
1506                                 p += integer_size;
1507                         }
1508                 }
1509                 chunk = la->la_next;
1510         }
1511 }
1512
1513 /*
1514  * Lookup a value in a fatzap directory. Assumes that the zap scratch
1515  * buffer contains the directory header.
1516  */
1517 static int
1518 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
1519     uint64_t integer_size, uint64_t num_integers, void *value)
1520 {
1521         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1522         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1523         fat_zap_t z;
1524         uint64_t *ptrtbl;
1525         uint64_t hash;
1526         int rc;
1527
1528         if (zh.zap_magic != ZAP_MAGIC)
1529                 return (EIO);
1530
1531         z.zap_block_shift = ilog2(bsize);
1532         z.zap_phys = (zap_phys_t *) zap_scratch;
1533
1534         /*
1535          * Figure out where the pointer table is and read it in if necessary.
1536          */
1537         if (zh.zap_ptrtbl.zt_blk) {
1538                 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1539                                zap_scratch, bsize);
1540                 if (rc)
1541                         return (rc);
1542                 ptrtbl = (uint64_t *) zap_scratch;
1543         } else {
1544                 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1545         }
1546
1547         hash = zap_hash(zh.zap_salt, name);
1548
1549         zap_leaf_t zl;
1550         zl.l_bs = z.zap_block_shift;
1551
1552         off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1553         zap_leaf_chunk_t *zc;
1554
1555         rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1556         if (rc)
1557                 return (rc);
1558
1559         zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1560
1561         /*
1562          * Make sure this chunk matches our hash.
1563          */
1564         if (zl.l_phys->l_hdr.lh_prefix_len > 0
1565             && zl.l_phys->l_hdr.lh_prefix
1566             != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1567                 return (ENOENT);
1568
1569         /*
1570          * Hash within the chunk to find our entry.
1571          */
1572         int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1573         int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1574         h = zl.l_phys->l_hash[h];
1575         if (h == 0xffff)
1576                 return (ENOENT);
1577         zc = &ZAP_LEAF_CHUNK(&zl, h);
1578         while (zc->l_entry.le_hash != hash) {
1579                 if (zc->l_entry.le_next == 0xffff) {
1580                         zc = NULL;
1581                         break;
1582                 }
1583                 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1584         }
1585         if (fzap_name_equal(&zl, zc, name)) {
1586                 if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints >
1587                     integer_size * num_integers)
1588                         return (E2BIG);
1589                 fzap_leaf_array(&zl, zc, integer_size, num_integers, value);
1590                 return (0);
1591         }
1592
1593         return (ENOENT);
1594 }
1595
1596 /*
1597  * Lookup a name in a zap object and return its value as a uint64_t.
1598  */
1599 static int
1600 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
1601     uint64_t integer_size, uint64_t num_integers, void *value)
1602 {
1603         int rc;
1604         uint64_t zap_type;
1605         size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1606
1607         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1608         if (rc)
1609                 return (rc);
1610
1611         zap_type = *(uint64_t *) zap_scratch;
1612         if (zap_type == ZBT_MICRO)
1613                 return mzap_lookup(dnode, name, value);
1614         else if (zap_type == ZBT_HEADER) {
1615                 return fzap_lookup(spa, dnode, name, integer_size,
1616                     num_integers, value);
1617         }
1618         printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
1619         return (EIO);
1620 }
1621
1622 /*
1623  * List a microzap directory. Assumes that the zap scratch buffer contains
1624  * the directory contents.
1625  */
1626 static int
1627 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1628 {
1629         const mzap_phys_t *mz;
1630         const mzap_ent_phys_t *mze;
1631         size_t size;
1632         int chunks, i, rc;
1633
1634         /*
1635          * Microzap objects use exactly one block. Read the whole
1636          * thing.
1637          */
1638         size = dnode->dn_datablkszsec * 512;
1639         mz = (const mzap_phys_t *) zap_scratch;
1640         chunks = size / MZAP_ENT_LEN - 1;
1641
1642         for (i = 0; i < chunks; i++) {
1643                 mze = &mz->mz_chunk[i];
1644                 if (mze->mze_name[0]) {
1645                         rc = callback(mze->mze_name, mze->mze_value);
1646                         if (rc != 0)
1647                                 return (rc);
1648                 }
1649         }
1650
1651         return (0);
1652 }
1653
1654 /*
1655  * List a fatzap directory. Assumes that the zap scratch buffer contains
1656  * the directory header.
1657  */
1658 static int
1659 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1660 {
1661         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1662         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1663         fat_zap_t z;
1664         int i, j, rc;
1665
1666         if (zh.zap_magic != ZAP_MAGIC)
1667                 return (EIO);
1668
1669         z.zap_block_shift = ilog2(bsize);
1670         z.zap_phys = (zap_phys_t *) zap_scratch;
1671
1672         /*
1673          * This assumes that the leaf blocks start at block 1. The
1674          * documentation isn't exactly clear on this.
1675          */
1676         zap_leaf_t zl;
1677         zl.l_bs = z.zap_block_shift;
1678         for (i = 0; i < zh.zap_num_leafs; i++) {
1679                 off_t off = (i + 1) << zl.l_bs;
1680                 char name[256], *p;
1681                 uint64_t value;
1682
1683                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1684                         return (EIO);
1685
1686                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1687
1688                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1689                         zap_leaf_chunk_t *zc, *nc;
1690                         int namelen;
1691
1692                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1693                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1694                                 continue;
1695                         namelen = zc->l_entry.le_name_numints;
1696                         if (namelen > sizeof(name))
1697                                 namelen = sizeof(name);
1698
1699                         /*
1700                          * Paste the name back together.
1701                          */
1702                         nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1703                         p = name;
1704                         while (namelen > 0) {
1705                                 int len;
1706                                 len = namelen;
1707                                 if (len > ZAP_LEAF_ARRAY_BYTES)
1708                                         len = ZAP_LEAF_ARRAY_BYTES;
1709                                 memcpy(p, nc->l_array.la_array, len);
1710                                 p += len;
1711                                 namelen -= len;
1712                                 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1713                         }
1714
1715                         /*
1716                          * Assume the first eight bytes of the value are
1717                          * a uint64_t.
1718                          */
1719                         value = fzap_leaf_value(&zl, zc);
1720
1721                         //printf("%s 0x%jx\n", name, (uintmax_t)value);
1722                         rc = callback((const char *)name, value);
1723                         if (rc != 0)
1724                                 return (rc);
1725                 }
1726         }
1727
1728         return (0);
1729 }
1730
1731 static int zfs_printf(const char *name, uint64_t value __unused)
1732 {
1733
1734         printf("%s\n", name);
1735
1736         return (0);
1737 }
1738
1739 /*
1740  * List a zap directory.
1741  */
1742 static int
1743 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
1744 {
1745         uint64_t zap_type;
1746         size_t size = dnode->dn_datablkszsec * 512;
1747
1748         if (dnode_read(spa, dnode, 0, zap_scratch, size))
1749                 return (EIO);
1750
1751         zap_type = *(uint64_t *) zap_scratch;
1752         if (zap_type == ZBT_MICRO)
1753                 return mzap_list(dnode, zfs_printf);
1754         else
1755                 return fzap_list(spa, dnode, zfs_printf);
1756 }
1757
1758 static int
1759 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1760 {
1761         off_t offset;
1762
1763         offset = objnum * sizeof(dnode_phys_t);
1764         return dnode_read(spa, &os->os_meta_dnode, offset,
1765                 dnode, sizeof(dnode_phys_t));
1766 }
1767
1768 static int
1769 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1770 {
1771         const mzap_phys_t *mz;
1772         const mzap_ent_phys_t *mze;
1773         size_t size;
1774         int chunks, i;
1775
1776         /*
1777          * Microzap objects use exactly one block. Read the whole
1778          * thing.
1779          */
1780         size = dnode->dn_datablkszsec * 512;
1781
1782         mz = (const mzap_phys_t *) zap_scratch;
1783         chunks = size / MZAP_ENT_LEN - 1;
1784
1785         for (i = 0; i < chunks; i++) {
1786                 mze = &mz->mz_chunk[i];
1787                 if (value == mze->mze_value) {
1788                         strcpy(name, mze->mze_name);
1789                         return (0);
1790                 }
1791         }
1792
1793         return (ENOENT);
1794 }
1795
1796 static void
1797 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
1798 {
1799         size_t namelen;
1800         const zap_leaf_chunk_t *nc;
1801         char *p;
1802
1803         namelen = zc->l_entry.le_name_numints;
1804
1805         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1806         p = name;
1807         while (namelen > 0) {
1808                 size_t len;
1809                 len = namelen;
1810                 if (len > ZAP_LEAF_ARRAY_BYTES)
1811                         len = ZAP_LEAF_ARRAY_BYTES;
1812                 memcpy(p, nc->l_array.la_array, len);
1813                 p += len;
1814                 namelen -= len;
1815                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1816         }
1817
1818         *p = '\0';
1819 }
1820
1821 static int
1822 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1823 {
1824         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1825         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1826         fat_zap_t z;
1827         int i, j;
1828
1829         if (zh.zap_magic != ZAP_MAGIC)
1830                 return (EIO);
1831
1832         z.zap_block_shift = ilog2(bsize);
1833         z.zap_phys = (zap_phys_t *) zap_scratch;
1834
1835         /*
1836          * This assumes that the leaf blocks start at block 1. The
1837          * documentation isn't exactly clear on this.
1838          */
1839         zap_leaf_t zl;
1840         zl.l_bs = z.zap_block_shift;
1841         for (i = 0; i < zh.zap_num_leafs; i++) {
1842                 off_t off = (i + 1) << zl.l_bs;
1843
1844                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1845                         return (EIO);
1846
1847                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1848
1849                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1850                         zap_leaf_chunk_t *zc;
1851
1852                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1853                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1854                                 continue;
1855                         if (zc->l_entry.le_value_intlen != 8 ||
1856                             zc->l_entry.le_value_numints != 1)
1857                                 continue;
1858
1859                         if (fzap_leaf_value(&zl, zc) == value) {
1860                                 fzap_name_copy(&zl, zc, name);
1861                                 return (0);
1862                         }
1863                 }
1864         }
1865
1866         return (ENOENT);
1867 }
1868
1869 static int
1870 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1871 {
1872         int rc;
1873         uint64_t zap_type;
1874         size_t size = dnode->dn_datablkszsec * 512;
1875
1876         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1877         if (rc)
1878                 return (rc);
1879
1880         zap_type = *(uint64_t *) zap_scratch;
1881         if (zap_type == ZBT_MICRO)
1882                 return mzap_rlookup(spa, dnode, name, value);
1883         else
1884                 return fzap_rlookup(spa, dnode, name, value);
1885 }
1886
1887 static int
1888 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
1889 {
1890         char name[256];
1891         char component[256];
1892         uint64_t dir_obj, parent_obj, child_dir_zapobj;
1893         dnode_phys_t child_dir_zap, dataset, dir, parent;
1894         dsl_dir_phys_t *dd;
1895         dsl_dataset_phys_t *ds;
1896         char *p;
1897         int len;
1898
1899         p = &name[sizeof(name) - 1];
1900         *p = '\0';
1901
1902         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1903                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1904                 return (EIO);
1905         }
1906         ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
1907         dir_obj = ds->ds_dir_obj;
1908
1909         for (;;) {
1910                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
1911                         return (EIO);
1912                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1913
1914                 /* Actual loop condition. */
1915                 parent_obj  = dd->dd_parent_obj;
1916                 if (parent_obj == 0)
1917                         break;
1918
1919                 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
1920                         return (EIO);
1921                 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
1922                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1923                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1924                         return (EIO);
1925                 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
1926                         return (EIO);
1927
1928                 len = strlen(component);
1929                 p -= len;
1930                 memcpy(p, component, len);
1931                 --p;
1932                 *p = '/';
1933
1934                 /* Actual loop iteration. */
1935                 dir_obj = parent_obj;
1936         }
1937
1938         if (*p != '\0')
1939                 ++p;
1940         strcpy(result, p);
1941
1942         return (0);
1943 }
1944
1945 static int
1946 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
1947 {
1948         char element[256];
1949         uint64_t dir_obj, child_dir_zapobj;
1950         dnode_phys_t child_dir_zap, dir;
1951         dsl_dir_phys_t *dd;
1952         const char *p, *q;
1953
1954         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
1955                 return (EIO);
1956         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
1957             1, &dir_obj))
1958                 return (EIO);
1959
1960         p = name;
1961         for (;;) {
1962                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
1963                         return (EIO);
1964                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1965
1966                 while (*p == '/')
1967                         p++;
1968                 /* Actual loop condition #1. */
1969                 if (*p == '\0')
1970                         break;
1971
1972                 q = strchr(p, '/');
1973                 if (q) {
1974                         memcpy(element, p, q - p);
1975                         element[q - p] = '\0';
1976                         p = q + 1;
1977                 } else {
1978                         strcpy(element, p);
1979                         p += strlen(p);
1980                 }
1981
1982                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1983                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1984                         return (EIO);
1985
1986                 /* Actual loop condition #2. */
1987                 if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
1988                     1, &dir_obj) != 0)
1989                         return (ENOENT);
1990         }
1991
1992         *objnum = dd->dd_head_dataset_obj;
1993         return (0);
1994 }
1995
1996 #ifndef BOOT2
1997 static int
1998 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
1999 {
2000         uint64_t dir_obj, child_dir_zapobj;
2001         dnode_phys_t child_dir_zap, dir, dataset;
2002         dsl_dataset_phys_t *ds;
2003         dsl_dir_phys_t *dd;
2004
2005         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2006                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2007                 return (EIO);
2008         }
2009         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2010         dir_obj = ds->ds_dir_obj;
2011
2012         if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
2013                 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2014                 return (EIO);
2015         }
2016         dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2017
2018         child_dir_zapobj = dd->dd_child_dir_zapobj;
2019         if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
2020                 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2021                 return (EIO);
2022         }
2023
2024         return (zap_list(spa, &child_dir_zap) != 0);
2025 }
2026
2027 int
2028 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *, uint64_t))
2029 {
2030         uint64_t dir_obj, child_dir_zapobj, zap_type;
2031         dnode_phys_t child_dir_zap, dir, dataset;
2032         dsl_dataset_phys_t *ds;
2033         dsl_dir_phys_t *dd;
2034         int err;
2035
2036         err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
2037         if (err != 0) {
2038                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2039                 return (err);
2040         }
2041         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2042         dir_obj = ds->ds_dir_obj;
2043
2044         err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
2045         if (err != 0) {
2046                 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2047                 return (err);
2048         }
2049         dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2050
2051         child_dir_zapobj = dd->dd_child_dir_zapobj;
2052         err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
2053         if (err != 0) {
2054                 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2055                 return (err);
2056         }
2057
2058         err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
2059         if (err != 0)
2060                 return (err);
2061
2062         zap_type = *(uint64_t *) zap_scratch;
2063         if (zap_type == ZBT_MICRO)
2064                 return mzap_list(&child_dir_zap, callback);
2065         else
2066                 return fzap_list(spa, &child_dir_zap, callback);
2067 }
2068 #endif
2069
2070 /*
2071  * Find the object set given the object number of its dataset object
2072  * and return its details in *objset
2073  */
2074 static int
2075 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2076 {
2077         dnode_phys_t dataset;
2078         dsl_dataset_phys_t *ds;
2079         int err;
2080
2081         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2082                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2083                 return (EIO);
2084         }
2085
2086         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2087         if ((err = zio_read(spa, &ds->ds_bp, objset)) != 0) {
2088                 printf("ZFS: can't read object set for dataset %ju (error %d)\n",
2089                     (uintmax_t)objnum, err);
2090                 return (EIO);
2091         }
2092
2093         return (0);
2094 }
2095
2096 /*
2097  * Find the object set pointed to by the BOOTFS property or the root
2098  * dataset if there is none and return its details in *objset
2099  */
2100 static int
2101 zfs_get_root(const spa_t *spa, uint64_t *objid)
2102 {
2103         dnode_phys_t dir, propdir;
2104         uint64_t props, bootfs, root;
2105
2106         *objid = 0;
2107
2108         /*
2109          * Start with the MOS directory object.
2110          */
2111         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
2112                 printf("ZFS: can't read MOS object directory\n");
2113                 return (EIO);
2114         }
2115
2116         /*
2117          * Lookup the pool_props and see if we can find a bootfs.
2118          */
2119         if (zap_lookup(spa, &dir, DMU_POOL_PROPS, sizeof (props), 1, &props) == 0
2120              && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
2121              && zap_lookup(spa, &propdir, "bootfs", sizeof (bootfs), 1, &bootfs) == 0
2122              && bootfs != 0)
2123         {
2124                 *objid = bootfs;
2125                 return (0);
2126         }
2127         /*
2128          * Lookup the root dataset directory
2129          */
2130         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (root), 1, &root)
2131             || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
2132                 printf("ZFS: can't find root dsl_dir\n");
2133                 return (EIO);
2134         }
2135
2136         /*
2137          * Use the information from the dataset directory's bonus buffer
2138          * to find the dataset object and from that the object set itself.
2139          */
2140         dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
2141         *objid = dd->dd_head_dataset_obj;
2142         return (0);
2143 }
2144
2145 static int
2146 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
2147 {
2148
2149         mount->spa = spa;
2150
2151         /*
2152          * Find the root object set if not explicitly provided
2153          */
2154         if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
2155                 printf("ZFS: can't find root filesystem\n");
2156                 return (EIO);
2157         }
2158
2159         if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
2160                 printf("ZFS: can't open root filesystem\n");
2161                 return (EIO);
2162         }
2163
2164         mount->rootobj = rootobj;
2165
2166         return (0);
2167 }
2168
2169 /*
2170  * callback function for feature name checks.
2171  */
2172 static int
2173 check_feature(const char *name, uint64_t value)
2174 {
2175         int i;
2176
2177         if (value == 0)
2178                 return (0);
2179         if (name[0] == '\0')
2180                 return (0);
2181
2182         for (i = 0; features_for_read[i] != NULL; i++) {
2183                 if (strcmp(name, features_for_read[i]) == 0)
2184                         return (0);
2185         }
2186         printf("ZFS: unsupported feature: %s\n", name);
2187         return (EIO);
2188 }
2189
2190 /*
2191  * Checks whether the MOS features that are active are supported.
2192  */
2193 static int
2194 check_mos_features(const spa_t *spa)
2195 {
2196         dnode_phys_t dir;
2197         uint64_t objnum, zap_type;
2198         size_t size;
2199         int rc;
2200
2201         if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
2202             &dir)) != 0)
2203                 return (rc);
2204         if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
2205             sizeof (objnum), 1, &objnum)) != 0) {
2206                 /*
2207                  * It is older pool without features. As we have already
2208                  * tested the label, just return without raising the error.
2209                  */
2210                 return (0);
2211         }
2212
2213         if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
2214                 return (rc);
2215
2216         if (dir.dn_type != DMU_OTN_ZAP_METADATA)
2217                 return (EIO);
2218
2219         size = dir.dn_datablkszsec * 512;
2220         if (dnode_read(spa, &dir, 0, zap_scratch, size))
2221                 return (EIO);
2222
2223         zap_type = *(uint64_t *) zap_scratch;
2224         if (zap_type == ZBT_MICRO)
2225                 rc = mzap_list(&dir, check_feature);
2226         else
2227                 rc = fzap_list(spa, &dir, check_feature);
2228
2229         return (rc);
2230 }
2231
2232 static int
2233 zfs_spa_init(spa_t *spa)
2234 {
2235         dnode_phys_t dir;
2236         int rc;
2237
2238         if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
2239                 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
2240                 return (EIO);
2241         }
2242         if (spa->spa_mos.os_type != DMU_OST_META) {
2243                 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
2244                 return (EIO);
2245         }
2246
2247         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT,
2248             &dir)) {
2249                 printf("ZFS: failed to read pool %s directory object\n",
2250                     spa->spa_name);
2251                 return (EIO);
2252         }
2253         /* this is allowed to fail, older pools do not have salt */
2254         rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
2255             sizeof (spa->spa_cksum_salt.zcs_bytes),
2256             spa->spa_cksum_salt.zcs_bytes);
2257
2258         rc = check_mos_features(spa);
2259         if (rc != 0) {
2260                 printf("ZFS: pool %s is not supported\n", spa->spa_name);
2261         }
2262
2263         return (rc);
2264 }
2265
2266 static int
2267 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
2268 {
2269
2270         if (dn->dn_bonustype != DMU_OT_SA) {
2271                 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
2272
2273                 sb->st_mode = zp->zp_mode;
2274                 sb->st_uid = zp->zp_uid;
2275                 sb->st_gid = zp->zp_gid;
2276                 sb->st_size = zp->zp_size;
2277         } else {
2278                 sa_hdr_phys_t *sahdrp;
2279                 int hdrsize;
2280                 size_t size = 0;
2281                 void *buf = NULL;
2282
2283                 if (dn->dn_bonuslen != 0)
2284                         sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2285                 else {
2286                         if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
2287                                 blkptr_t *bp = DN_SPILL_BLKPTR(dn);
2288                                 int error;
2289
2290                                 size = BP_GET_LSIZE(bp);
2291                                 buf = zfs_alloc(size);
2292                                 error = zio_read(spa, bp, buf);
2293                                 if (error != 0) {
2294                                         zfs_free(buf, size);
2295                                         return (error);
2296                                 }
2297                                 sahdrp = buf;
2298                         } else {
2299                                 return (EIO);
2300                         }
2301                 }
2302                 hdrsize = SA_HDR_SIZE(sahdrp);
2303                 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
2304                     SA_MODE_OFFSET);
2305                 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
2306                     SA_UID_OFFSET);
2307                 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
2308                     SA_GID_OFFSET);
2309                 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
2310                     SA_SIZE_OFFSET);
2311                 if (buf != NULL)
2312                         zfs_free(buf, size);
2313         }
2314
2315         return (0);
2316 }
2317
2318 static int
2319 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
2320 {
2321         int rc = 0;
2322
2323         if (dn->dn_bonustype == DMU_OT_SA) {
2324                 sa_hdr_phys_t *sahdrp = NULL;
2325                 size_t size = 0;
2326                 void *buf = NULL;
2327                 int hdrsize;
2328                 char *p;
2329
2330                 if (dn->dn_bonuslen != 0)
2331                         sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2332                 else {
2333                         blkptr_t *bp;
2334
2335                         if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
2336                                 return (EIO);
2337                         bp = DN_SPILL_BLKPTR(dn);
2338
2339                         size = BP_GET_LSIZE(bp);
2340                         buf = zfs_alloc(size);
2341                         rc = zio_read(spa, bp, buf);
2342                         if (rc != 0) {
2343                                 zfs_free(buf, size);
2344                                 return (rc);
2345                         }
2346                         sahdrp = buf;
2347                 }
2348                 hdrsize = SA_HDR_SIZE(sahdrp);
2349                 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
2350                 memcpy(path, p, psize);
2351                 if (buf != NULL)
2352                         zfs_free(buf, size);
2353                 return (0);
2354         }
2355         /*
2356          * Second test is purely to silence bogus compiler
2357          * warning about accessing past the end of dn_bonus.
2358          */
2359         if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
2360             sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
2361                 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
2362         } else {
2363                 rc = dnode_read(spa, dn, 0, path, psize);
2364         }
2365         return (rc);
2366 }
2367
2368 struct obj_list {
2369         uint64_t                objnum;
2370         STAILQ_ENTRY(obj_list)  entry;
2371 };
2372
2373 /*
2374  * Lookup a file and return its dnode.
2375  */
2376 static int
2377 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
2378 {
2379         int rc;
2380         uint64_t objnum;
2381         const spa_t *spa;
2382         dnode_phys_t dn;
2383         const char *p, *q;
2384         char element[256];
2385         char path[1024];
2386         int symlinks_followed = 0;
2387         struct stat sb;
2388         struct obj_list *entry, *tentry;
2389         STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
2390
2391         spa = mount->spa;
2392         if (mount->objset.os_type != DMU_OST_ZFS) {
2393                 printf("ZFS: unexpected object set type %ju\n",
2394                     (uintmax_t)mount->objset.os_type);
2395                 return (EIO);
2396         }
2397
2398         if ((entry = malloc(sizeof(struct obj_list))) == NULL)
2399                 return (ENOMEM);
2400
2401         /*
2402          * Get the root directory dnode.
2403          */
2404         rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
2405         if (rc) {
2406                 free(entry);
2407                 return (rc);
2408         }
2409
2410         rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof (objnum), 1, &objnum);
2411         if (rc) {
2412                 free(entry);
2413                 return (rc);
2414         }
2415         entry->objnum = objnum;
2416         STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2417
2418         rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
2419         if (rc != 0)
2420                 goto done;
2421
2422         p = upath;
2423         while (p && *p) {
2424                 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
2425                 if (rc != 0)
2426                         goto done;
2427
2428                 while (*p == '/')
2429                         p++;
2430                 if (*p == '\0')
2431                         break;
2432                 q = p;
2433                 while (*q != '\0' && *q != '/')
2434                         q++;
2435
2436                 /* skip dot */
2437                 if (p + 1 == q && p[0] == '.') {
2438                         p++;
2439                         continue;
2440                 }
2441                 /* double dot */
2442                 if (p + 2 == q && p[0] == '.' && p[1] == '.') {
2443                         p += 2;
2444                         if (STAILQ_FIRST(&on_cache) ==
2445                             STAILQ_LAST(&on_cache, obj_list, entry)) {
2446                                 rc = ENOENT;
2447                                 goto done;
2448                         }
2449                         entry = STAILQ_FIRST(&on_cache);
2450                         STAILQ_REMOVE_HEAD(&on_cache, entry);
2451                         free(entry);
2452                         objnum = (STAILQ_FIRST(&on_cache))->objnum;
2453                         continue;
2454                 }
2455                 if (q - p + 1 > sizeof(element)) {
2456                         rc = ENAMETOOLONG;
2457                         goto done;
2458                 }
2459                 memcpy(element, p, q - p);
2460                 element[q - p] = 0;
2461                 p = q;
2462
2463                 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
2464                         goto done;
2465                 if (!S_ISDIR(sb.st_mode)) {
2466                         rc = ENOTDIR;
2467                         goto done;
2468                 }
2469
2470                 rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
2471                 if (rc)
2472                         goto done;
2473                 objnum = ZFS_DIRENT_OBJ(objnum);
2474
2475                 if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
2476                         rc = ENOMEM;
2477                         goto done;
2478                 }
2479                 entry->objnum = objnum;
2480                 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2481                 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
2482                 if (rc)
2483                         goto done;
2484
2485                 /*
2486                  * Check for symlink.
2487                  */
2488                 rc = zfs_dnode_stat(spa, &dn, &sb);
2489                 if (rc)
2490                         goto done;
2491                 if (S_ISLNK(sb.st_mode)) {
2492                         if (symlinks_followed > 10) {
2493                                 rc = EMLINK;
2494                                 goto done;
2495                         }
2496                         symlinks_followed++;
2497
2498                         /*
2499                          * Read the link value and copy the tail of our
2500                          * current path onto the end.
2501                          */
2502                         if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
2503                                 rc = ENAMETOOLONG;
2504                                 goto done;
2505                         }
2506                         strcpy(&path[sb.st_size], p);
2507
2508                         rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
2509                         if (rc != 0)
2510                                 goto done;
2511
2512                         /*
2513                          * Restart with the new path, starting either at
2514                          * the root or at the parent depending whether or
2515                          * not the link is relative.
2516                          */
2517                         p = path;
2518                         if (*p == '/') {
2519                                 while (STAILQ_FIRST(&on_cache) !=
2520                                     STAILQ_LAST(&on_cache, obj_list, entry)) {
2521                                         entry = STAILQ_FIRST(&on_cache);
2522                                         STAILQ_REMOVE_HEAD(&on_cache, entry);
2523                                         free(entry);
2524                                 }
2525                         } else {
2526                                 entry = STAILQ_FIRST(&on_cache);
2527                                 STAILQ_REMOVE_HEAD(&on_cache, entry);
2528                                 free(entry);
2529                         }
2530                         objnum = (STAILQ_FIRST(&on_cache))->objnum;
2531                 }
2532         }
2533
2534         *dnode = dn;
2535 done:
2536         STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
2537                 free(entry);
2538         return (rc);
2539 }