sys/boot/zfs/zfsimpl.c

   1 /*-
   2  * Copyright (c) 2007 Doug Rabson
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  */
  26
  27 #include <sys/cdefs.h>
  28 __FBSDID("$FreeBSD$");
  29
  30 /*
  31  *      Stand-alone ZFS file reader.
  32  */
  33
  34 #include <sys/stat.h>
  35 #include <sys/stdint.h>
  36
  37 #include "zfsimpl.h"
  38 #include "zfssubr.c"
  39
  40
  41 struct zfsmount {
  42         const spa_t     *spa;
  43         objset_phys_t   objset;
  44         uint64_t        rootobj;
  45 };
  46
  47 /*
  48  * List of all vdevs, chained through v_alllink.
  49  */
  50 static vdev_list_t zfs_vdevs;
  51
  52 /*
  53  * List of all pools, chained through spa_link.
  54  */
  55 static spa_list_t zfs_pools;
  56
  57 static uint64_t zfs_crc64_table[256];
  58 static const dnode_phys_t *dnode_cache_obj = 0;
  59 static uint64_t dnode_cache_bn;
  60 static char *dnode_cache_buf;
  61 static char *zap_scratch;
  62 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
  63
  64 #define TEMP_SIZE       (1024 * 1024)
  65
  66 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
  67
  68 static void
  69 zfs_init(void)
  70 {
  71         STAILQ_INIT(&zfs_vdevs);
  72         STAILQ_INIT(&zfs_pools);
  73
  74         zfs_temp_buf = malloc(TEMP_SIZE);
  75         zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
  76         zfs_temp_ptr = zfs_temp_buf;
  77         dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
  78         zap_scratch = malloc(SPA_MAXBLOCKSIZE);
  79
  80         zfs_init_crc();
  81 }
  82
  83 static void *
  84 zfs_alloc(size_t size)
  85 {
  86         char *ptr;
  87
  88         if (zfs_temp_ptr + size > zfs_temp_end) {
  89                 printf("ZFS: out of temporary buffer space\n");
  90                 for (;;) ;
  91         }
  92         ptr = zfs_temp_ptr;
  93         zfs_temp_ptr += size;
  94
  95         return (ptr);
  96 }
  97
  98 static void
  99 zfs_free(void *ptr, size_t size)
 100 {
 101
 102         zfs_temp_ptr -= size;
 103         if (zfs_temp_ptr != ptr) {
 104                 printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
 105                 for (;;) ;
 106         }
 107 }
 108
 109 static int
 110 xdr_int(const unsigned char **xdr, int *ip)
 111 {
 112         *ip = ((*xdr)[0] << 24)
 113                 | ((*xdr)[1] << 16)
 114                 | ((*xdr)[2] << 8)
 115                 | ((*xdr)[3] << 0);
 116         (*xdr) += 4;
 117         return (0);
 118 }
 119
 120 static int
 121 xdr_u_int(const unsigned char **xdr, u_int *ip)
 122 {
 123         *ip = ((*xdr)[0] << 24)
 124                 | ((*xdr)[1] << 16)
 125                 | ((*xdr)[2] << 8)
 126                 | ((*xdr)[3] << 0);
 127         (*xdr) += 4;
 128         return (0);
 129 }
 130
 131 static int
 132 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
 133 {
 134         u_int hi, lo;
 135
 136         xdr_u_int(xdr, &hi);
 137         xdr_u_int(xdr, &lo);
 138         *lp = (((uint64_t) hi) << 32) | lo;
 139         return (0);
 140 }
 141
 142 static int
 143 nvlist_find(const unsigned char *nvlist, const char *name, int type,
 144             int* elementsp, void *valuep)
 145 {
 146         const unsigned char *p, *pair;
 147         int junk;
 148         int encoded_size, decoded_size;
 149
 150         p = nvlist;
 151         xdr_int(&p, &junk);
 152         xdr_int(&p, &junk);
 153
 154         pair = p;
 155         xdr_int(&p, &encoded_size);
 156         xdr_int(&p, &decoded_size);
 157         while (encoded_size && decoded_size) {
 158                 int namelen, pairtype, elements;
 159                 const char *pairname;
 160
 161                 xdr_int(&p, &namelen);
 162                 pairname = (const char*) p;
 163                 p += roundup(namelen, 4);
 164                 xdr_int(&p, &pairtype);
 165
 166                 if (!memcmp(name, pairname, namelen) && type == pairtype) {
 167                         xdr_int(&p, &elements);
 168                         if (elementsp)
 169                                 *elementsp = elements;
 170                         if (type == DATA_TYPE_UINT64) {
 171                                 xdr_uint64_t(&p, (uint64_t *) valuep);
 172                                 return (0);
 173                         } else if (type == DATA_TYPE_STRING) {
 174                                 int len;
 175                                 xdr_int(&p, &len);
 176                                 (*(const char**) valuep) = (const char*) p;
 177                                 return (0);
 178                         } else if (type == DATA_TYPE_NVLIST
 179                                    || type == DATA_TYPE_NVLIST_ARRAY) {
 180                                 (*(const unsigned char**) valuep) =
 181                                          (const unsigned char*) p;
 182                                 return (0);
 183                         } else {
 184                                 return (EIO);
 185                         }
 186                 } else {
 187                         /*
 188                          * Not the pair we are looking for, skip to the next one.
 189                          */
 190                         p = pair + encoded_size;
 191                 }
 192
 193                 pair = p;
 194                 xdr_int(&p, &encoded_size);
 195                 xdr_int(&p, &decoded_size);
 196         }
 197
 198         return (EIO);
 199 }
 200
 201 /*
 202  * Return the next nvlist in an nvlist array.
 203  */
 204 static const unsigned char *
 205 nvlist_next(const unsigned char *nvlist)
 206 {
 207         const unsigned char *p, *pair;
 208         int junk;
 209         int encoded_size, decoded_size;
 210
 211         p = nvlist;
 212         xdr_int(&p, &junk);
 213         xdr_int(&p, &junk);
 214
 215         pair = p;
 216         xdr_int(&p, &encoded_size);
 217         xdr_int(&p, &decoded_size);
 218         while (encoded_size && decoded_size) {
 219                 p = pair + encoded_size;
 220
 221                 pair = p;
 222                 xdr_int(&p, &encoded_size);
 223                 xdr_int(&p, &decoded_size);
 224         }
 225
 226         return p;
 227 }
 228
 229 #ifdef TEST
 230
 231 static const unsigned char *
 232 nvlist_print(const unsigned char *nvlist, unsigned int indent)
 233 {
 234         static const char* typenames[] = {
 235                 "DATA_TYPE_UNKNOWN",
 236                 "DATA_TYPE_BOOLEAN",
 237                 "DATA_TYPE_BYTE",
 238                 "DATA_TYPE_INT16",
 239                 "DATA_TYPE_UINT16",
 240                 "DATA_TYPE_INT32",
 241                 "DATA_TYPE_UINT32",
 242                 "DATA_TYPE_INT64",
 243                 "DATA_TYPE_UINT64",
 244                 "DATA_TYPE_STRING",
 245                 "DATA_TYPE_BYTE_ARRAY",
 246                 "DATA_TYPE_INT16_ARRAY",
 247                 "DATA_TYPE_UINT16_ARRAY",
 248                 "DATA_TYPE_INT32_ARRAY",
 249                 "DATA_TYPE_UINT32_ARRAY",
 250                 "DATA_TYPE_INT64_ARRAY",
 251                 "DATA_TYPE_UINT64_ARRAY",
 252                 "DATA_TYPE_STRING_ARRAY",
 253                 "DATA_TYPE_HRTIME",
 254                 "DATA_TYPE_NVLIST",
 255                 "DATA_TYPE_NVLIST_ARRAY",
 256                 "DATA_TYPE_BOOLEAN_VALUE",
 257                 "DATA_TYPE_INT8",
 258                 "DATA_TYPE_UINT8",
 259                 "DATA_TYPE_BOOLEAN_ARRAY",
 260                 "DATA_TYPE_INT8_ARRAY",
 261                 "DATA_TYPE_UINT8_ARRAY"
 262         };
 263
 264         unsigned int i, j;
 265         const unsigned char *p, *pair;
 266         int junk;
 267         int encoded_size, decoded_size;
 268
 269         p = nvlist;
 270         xdr_int(&p, &junk);
 271         xdr_int(&p, &junk);
 272
 273         pair = p;
 274         xdr_int(&p, &encoded_size);
 275         xdr_int(&p, &decoded_size);
 276         while (encoded_size && decoded_size) {
 277                 int namelen, pairtype, elements;
 278                 const char *pairname;
 279
 280                 xdr_int(&p, &namelen);
 281                 pairname = (const char*) p;
 282                 p += roundup(namelen, 4);
 283                 xdr_int(&p, &pairtype);
 284
 285                 for (i = 0; i < indent; i++)
 286                         printf(" ");
 287                 printf("%s %s", typenames[pairtype], pairname);
 288
 289                 xdr_int(&p, &elements);
 290                 switch (pairtype) {
 291                 case DATA_TYPE_UINT64: {
 292                         uint64_t val;
 293                         xdr_uint64_t(&p, &val);
 294                         printf(" = 0x%jx\n", (uintmax_t)val);
 295                         break;
 296                 }
 297
 298                 case DATA_TYPE_STRING: {
 299                         int len;
 300                         xdr_int(&p, &len);
 301                         printf(" = \"%s\"\n", p);
 302                         break;
 303                 }
 304
 305                 case DATA_TYPE_NVLIST:
 306                         printf("\n");
 307                         nvlist_print(p, indent + 1);
 308                         break;
 309
 310                 case DATA_TYPE_NVLIST_ARRAY:
 311                         for (j = 0; j < elements; j++) {
 312                                 printf("[%d]\n", j);
 313                                 p = nvlist_print(p, indent + 1);
 314                                 if (j != elements - 1) {
 315                                         for (i = 0; i < indent; i++)
 316                                                 printf(" ");
 317                                         printf("%s %s", typenames[pairtype], pairname);
 318                                 }
 319                         }
 320                         break;
 321
 322                 default:
 323                         printf("\n");
 324                 }
 325
 326                 p = pair + encoded_size;
 327
 328                 pair = p;
 329                 xdr_int(&p, &encoded_size);
 330                 xdr_int(&p, &decoded_size);
 331         }
 332
 333         return p;
 334 }
 335
 336 #endif
 337
 338 static int
 339 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
 340     off_t offset, size_t size)
 341 {
 342         size_t psize;
 343         int rc;
 344
 345         if (!vdev->v_phys_read)
 346                 return (EIO);
 347
 348         if (bp) {
 349                 psize = BP_GET_PSIZE(bp);
 350         } else {
 351                 psize = size;
 352         }
 353
 354         /*printf("ZFS: reading %d bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
 355         rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
 356         if (rc)
 357                 return (rc);
 358         if (bp && zio_checksum_verify(bp, buf))
 359                 return (EIO);
 360
 361         return (0);
 362 }
 363
 364 static int
 365 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 366     off_t offset, size_t bytes)
 367 {
 368
 369         return (vdev_read_phys(vdev, bp, buf,
 370                 offset + VDEV_LABEL_START_SIZE, bytes));
 371 }
 372
 373
 374 static int
 375 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 376     off_t offset, size_t bytes)
 377 {
 378         vdev_t *kid;
 379         int rc;
 380
 381         rc = EIO;
 382         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 383                 if (kid->v_state != VDEV_STATE_HEALTHY)
 384                         continue;
 385                 rc = kid->v_read(kid, bp, buf, offset, bytes);
 386                 if (!rc)
 387                         return (0);
 388         }
 389
 390         return (rc);
 391 }
 392
 393 static int
 394 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
 395     off_t offset, size_t bytes)
 396 {
 397         vdev_t *kid;
 398
 399         /*
 400          * Here we should have two kids:
 401          * First one which is the one we are replacing and we can trust
 402          * only this one to have valid data, but it might not be present.
 403          * Second one is that one we are replacing with. It is most likely
 404          * healthy, but we can't trust it has needed data, so we won't use it.
 405          */
 406         kid = STAILQ_FIRST(&vdev->v_children);
 407         if (kid == NULL)
 408                 return (EIO);
 409         if (kid->v_state != VDEV_STATE_HEALTHY)
 410                 return (EIO);
 411         return (kid->v_read(kid, bp, buf, offset, bytes));
 412 }
 413
 414 static vdev_t *
 415 vdev_find(uint64_t guid)
 416 {
 417         vdev_t *vdev;
 418
 419         STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
 420                 if (vdev->v_guid == guid)
 421                         return (vdev);
 422
 423         return (0);
 424 }
 425
 426 static vdev_t *
 427 vdev_create(uint64_t guid, vdev_read_t *read)
 428 {
 429         vdev_t *vdev;
 430
 431         vdev = malloc(sizeof(vdev_t));
 432         memset(vdev, 0, sizeof(vdev_t));
 433         STAILQ_INIT(&vdev->v_children);
 434         vdev->v_guid = guid;
 435         vdev->v_state = VDEV_STATE_OFFLINE;
 436         vdev->v_read = read;
 437         vdev->v_phys_read = 0;
 438         vdev->v_read_priv = 0;
 439         STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
 440
 441         return (vdev);
 442 }
 443
 444 static int
 445 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
 446     vdev_t **vdevp, int is_newer)
 447 {
 448         int rc;
 449         uint64_t guid, id, ashift, nparity;
 450         const char *type;
 451         const char *path;
 452         vdev_t *vdev, *kid;
 453         const unsigned char *kids;
 454         int nkids, i, is_new;
 455         uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
 456
 457         if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID,
 458                         DATA_TYPE_UINT64, 0, &guid)
 459             || nvlist_find(nvlist, ZPOOL_CONFIG_ID,
 460                            DATA_TYPE_UINT64, 0, &id)
 461             || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE,
 462                            DATA_TYPE_STRING, 0, &type)) {
 463                 printf("ZFS: can't find vdev details\n");
 464                 return (ENOENT);
 465         }
 466
 467         if (strcmp(type, VDEV_TYPE_MIRROR)
 468             && strcmp(type, VDEV_TYPE_DISK)
 469 #ifdef ZFS_TEST
 470             && strcmp(type, VDEV_TYPE_FILE)
 471 #endif
 472             && strcmp(type, VDEV_TYPE_RAIDZ)
 473             && strcmp(type, VDEV_TYPE_REPLACING)) {
 474                 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 475                 return (EIO);
 476         }
 477
 478         is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
 479
 480         nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, 0,
 481                         &is_offline);
 482         nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, 0,
 483                         &is_removed);
 484         nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, 0,
 485                         &is_faulted);
 486         nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, 0,
 487                         &is_degraded);
 488         nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, 0,
 489                         &isnt_present);
 490
 491         vdev = vdev_find(guid);
 492         if (!vdev) {
 493                 is_new = 1;
 494
 495                 if (!strcmp(type, VDEV_TYPE_MIRROR))
 496                         vdev = vdev_create(guid, vdev_mirror_read);
 497                 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
 498                         vdev = vdev_create(guid, vdev_raidz_read);
 499                 else if (!strcmp(type, VDEV_TYPE_REPLACING))
 500                         vdev = vdev_create(guid, vdev_replacing_read);
 501                 else
 502                         vdev = vdev_create(guid, vdev_disk_read);
 503
 504                 vdev->v_id = id;
 505                 vdev->v_top = pvdev != NULL ? pvdev : vdev;
 506                 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
 507                         DATA_TYPE_UINT64, 0, &ashift) == 0)
 508                         vdev->v_ashift = ashift;
 509                 else
 510                         vdev->v_ashift = 0;
 511                 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
 512                         DATA_TYPE_UINT64, 0, &nparity) == 0)
 513                         vdev->v_nparity = nparity;
 514                 else
 515                         vdev->v_nparity = 0;
 516                 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
 517                                 DATA_TYPE_STRING, 0, &path) == 0) {
 518                         if (strncmp(path, "/dev/", 5) == 0)
 519                                 path += 5;
 520                         vdev->v_name = strdup(path);
 521                 } else {
 522                         if (!strcmp(type, "raidz")) {
 523                                 if (vdev->v_nparity == 1)
 524                                         vdev->v_name = "raidz1";
 525                                 else if (vdev->v_nparity == 2)
 526                                         vdev->v_name = "raidz2";
 527                                 else if (vdev->v_nparity == 3)
 528                                         vdev->v_name = "raidz3";
 529                                 else {
 530                                         printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
 531                                         return (EIO);
 532                                 }
 533                         } else {
 534                                 vdev->v_name = strdup(type);
 535                         }
 536                 }
 537         } else {
 538                 is_new = 0;
 539         }
 540
 541         if (is_new || is_newer) {
 542                 /*
 543                  * This is either new vdev or we've already seen this vdev,
 544                  * but from an older vdev label, so let's refresh its state
 545                  * from the newer label.
 546                  */
 547                 if (is_offline)
 548                         vdev->v_state = VDEV_STATE_OFFLINE;
 549                 else if (is_removed)
 550                         vdev->v_state = VDEV_STATE_REMOVED;
 551                 else if (is_faulted)
 552                         vdev->v_state = VDEV_STATE_FAULTED;
 553                 else if (is_degraded)
 554                         vdev->v_state = VDEV_STATE_DEGRADED;
 555                 else if (isnt_present)
 556                         vdev->v_state = VDEV_STATE_CANT_OPEN;
 557         }
 558
 559         rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN,
 560                          DATA_TYPE_NVLIST_ARRAY, &nkids, &kids);
 561         /*
 562          * Its ok if we don't have any kids.
 563          */
 564         if (rc == 0) {
 565                 vdev->v_nchildren = nkids;
 566                 for (i = 0; i < nkids; i++) {
 567                         rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
 568                         if (rc)
 569                                 return (rc);
 570                         if (is_new)
 571                                 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
 572                                                    v_childlink);
 573                         kids = nvlist_next(kids);
 574                 }
 575         } else {
 576                 vdev->v_nchildren = 0;
 577         }
 578
 579         if (vdevp)
 580                 *vdevp = vdev;
 581         return (0);
 582 }
 583
 584 static void
 585 vdev_set_state(vdev_t *vdev)
 586 {
 587         vdev_t *kid;
 588         int good_kids;
 589         int bad_kids;
 590
 591         /*
 592          * A mirror or raidz is healthy if all its kids are healthy. A
 593          * mirror is degraded if any of its kids is healthy; a raidz
 594          * is degraded if at most nparity kids are offline.
 595          */
 596         if (STAILQ_FIRST(&vdev->v_children)) {
 597                 good_kids = 0;
 598                 bad_kids = 0;
 599                 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 600                         if (kid->v_state == VDEV_STATE_HEALTHY)
 601                                 good_kids++;
 602                         else
 603                                 bad_kids++;
 604                 }
 605                 if (bad_kids == 0) {
 606                         vdev->v_state = VDEV_STATE_HEALTHY;
 607                 } else {
 608                         if (vdev->v_read == vdev_mirror_read) {
 609                                 if (good_kids) {
 610                                         vdev->v_state = VDEV_STATE_DEGRADED;
 611                                 } else {
 612                                         vdev->v_state = VDEV_STATE_OFFLINE;
 613                                 }
 614                         } else if (vdev->v_read == vdev_raidz_read) {
 615                                 if (bad_kids > vdev->v_nparity) {
 616                                         vdev->v_state = VDEV_STATE_OFFLINE;
 617                                 } else {
 618                                         vdev->v_state = VDEV_STATE_DEGRADED;
 619                                 }
 620                         }
 621                 }
 622         }
 623 }
 624
 625 static spa_t *
 626 spa_find_by_guid(uint64_t guid)
 627 {
 628         spa_t *spa;
 629
 630         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 631                 if (spa->spa_guid == guid)
 632                         return (spa);
 633
 634         return (0);
 635 }
 636
 637 static spa_t *
 638 spa_find_by_name(const char *name)
 639 {
 640         spa_t *spa;
 641
 642         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
 643                 if (!strcmp(spa->spa_name, name))
 644                         return (spa);
 645
 646         return (0);
 647 }
 648
 649 static spa_t *
 650 spa_create(uint64_t guid)
 651 {
 652         spa_t *spa;
 653
 654         spa = malloc(sizeof(spa_t));
 655         memset(spa, 0, sizeof(spa_t));
 656         STAILQ_INIT(&spa->spa_vdevs);
 657         spa->spa_guid = guid;
 658         STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
 659
 660         return (spa);
 661 }
 662
 663 static const char *
 664 state_name(vdev_state_t state)
 665 {
 666         static const char* names[] = {
 667                 "UNKNOWN",
 668                 "CLOSED",
 669                 "OFFLINE",
 670                 "REMOVED",
 671                 "CANT_OPEN",
 672                 "FAULTED",
 673                 "DEGRADED",
 674                 "ONLINE"
 675         };
 676         return names[state];
 677 }
 678
 679 #ifdef BOOT2
 680
 681 #define pager_printf printf
 682
 683 #else
 684
 685 static void
 686 pager_printf(const char *fmt, ...)
 687 {
 688         char line[80];
 689         va_list args;
 690
 691         va_start(args, fmt);
 692         vsprintf(line, fmt, args);
 693         va_end(args);
 694         pager_output(line);
 695 }
 696
 697 #endif
 698
 699 #define STATUS_FORMAT   "        %s %s\n"
 700
 701 static void
 702 print_state(int indent, const char *name, vdev_state_t state)
 703 {
 704         int i;
 705         char buf[512];
 706
 707         buf[0] = 0;
 708         for (i = 0; i < indent; i++)
 709                 strcat(buf, "  ");
 710         strcat(buf, name);
 711         pager_printf(STATUS_FORMAT, buf, state_name(state));
 712
 713 }
 714
 715 static void
 716 vdev_status(vdev_t *vdev, int indent)
 717 {
 718         vdev_t *kid;
 719         print_state(indent, vdev->v_name, vdev->v_state);
 720
 721         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
 722                 vdev_status(kid, indent + 1);
 723         }
 724 }
 725
 726 static void
 727 spa_status(spa_t *spa)
 728 {
 729         vdev_t *vdev;
 730         int good_kids, bad_kids, degraded_kids;
 731         vdev_state_t state;
 732
 733         pager_printf("  pool: %s\n", spa->spa_name);
 734         pager_printf("config:\n\n");
 735         pager_printf(STATUS_FORMAT, "NAME", "STATE");
 736
 737         good_kids = 0;
 738         degraded_kids = 0;
 739         bad_kids = 0;
 740         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 741                 if (vdev->v_state == VDEV_STATE_HEALTHY)
 742                         good_kids++;
 743                 else if (vdev->v_state == VDEV_STATE_DEGRADED)
 744                         degraded_kids++;
 745                 else
 746                         bad_kids++;
 747         }
 748
 749         state = VDEV_STATE_CLOSED;
 750         if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
 751                 state = VDEV_STATE_HEALTHY;
 752         else if ((good_kids + degraded_kids) > 0)
 753                 state = VDEV_STATE_DEGRADED;
 754
 755         print_state(0, spa->spa_name, state);
 756         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
 757                 vdev_status(vdev, 1);
 758         }
 759 }
 760
 761 static void
 762 spa_all_status(void)
 763 {
 764         spa_t *spa;
 765         int first = 1;
 766
 767         STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
 768                 if (!first)
 769                         pager_printf("\n");
 770                 first = 0;
 771                 spa_status(spa);
 772         }
 773 }
 774
 775 static int
 776 vdev_probe(vdev_phys_read_t *read, void *read_priv, spa_t **spap)
 777 {
 778         vdev_t vtmp;
 779         vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
 780         spa_t *spa;
 781         vdev_t *vdev, *top_vdev, *pool_vdev;
 782         off_t off;
 783         blkptr_t bp;
 784         const unsigned char *nvlist;
 785         uint64_t val;
 786         uint64_t guid;
 787         uint64_t pool_txg, pool_guid;
 788         uint64_t is_log;
 789         const char *pool_name;
 790         const unsigned char *vdevs;
 791         int i, rc, is_newer;
 792         char *upbuf;
 793         const struct uberblock *up;
 794
 795         /*
 796          * Load the vdev label and figure out which
 797          * uberblock is most current.
 798          */
 799         memset(&vtmp, 0, sizeof(vtmp));
 800         vtmp.v_phys_read = read;
 801         vtmp.v_read_priv = read_priv;
 802         off = offsetof(vdev_label_t, vl_vdev_phys);
 803         BP_ZERO(&bp);
 804         BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
 805         BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
 806         BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
 807         BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
 808         DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
 809         ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
 810         if (vdev_read_phys(&vtmp, &bp, vdev_label, off, 0))
 811                 return (EIO);
 812
 813         if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR) {
 814                 return (EIO);
 815         }
 816
 817         nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
 818
 819         if (nvlist_find(nvlist,
 820                         ZPOOL_CONFIG_VERSION,
 821                         DATA_TYPE_UINT64, 0, &val)) {
 822                 return (EIO);
 823         }
 824
 825         if (val > SPA_VERSION) {
 826                 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
 827                     (unsigned) val, (unsigned) SPA_VERSION);
 828                 return (EIO);
 829         }
 830
 831         if (nvlist_find(nvlist,
 832                         ZPOOL_CONFIG_POOL_STATE,
 833                         DATA_TYPE_UINT64, 0, &val)) {
 834                 return (EIO);
 835         }
 836
 837         if (val == POOL_STATE_DESTROYED) {
 838                 /* We don't boot only from destroyed pools. */
 839                 return (EIO);
 840         }
 841
 842         if (nvlist_find(nvlist,
 843                         ZPOOL_CONFIG_POOL_TXG,
 844                         DATA_TYPE_UINT64, 0, &pool_txg)
 845             || nvlist_find(nvlist,
 846                            ZPOOL_CONFIG_POOL_GUID,
 847                            DATA_TYPE_UINT64, 0, &pool_guid)
 848             || nvlist_find(nvlist,
 849                            ZPOOL_CONFIG_POOL_NAME,
 850                            DATA_TYPE_STRING, 0, &pool_name)) {
 851                 /*
 852                  * Cache and spare devices end up here - just ignore
 853                  * them.
 854                  */
 855                 /*printf("ZFS: can't find pool details\n");*/
 856                 return (EIO);
 857         }
 858
 859         is_log = 0;
 860         (void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, 0,
 861             &is_log);
 862         if (is_log)
 863                 return (EIO);
 864
 865         /*
 866          * Create the pool if this is the first time we've seen it.
 867          */
 868         spa = spa_find_by_guid(pool_guid);
 869         if (!spa) {
 870                 spa = spa_create(pool_guid);
 871                 spa->spa_name = strdup(pool_name);
 872         }
 873         if (pool_txg > spa->spa_txg) {
 874                 spa->spa_txg = pool_txg;
 875                 is_newer = 1;
 876         } else
 877                 is_newer = 0;
 878
 879         /*
 880          * Get the vdev tree and create our in-core copy of it.
 881          * If we already have a vdev with this guid, this must
 882          * be some kind of alias (overlapping slices, dangerously dedicated
 883          * disks etc).
 884          */
 885         if (nvlist_find(nvlist,
 886                         ZPOOL_CONFIG_GUID,
 887                         DATA_TYPE_UINT64, 0, &guid)) {
 888                 return (EIO);
 889         }
 890         vdev = vdev_find(guid);
 891         if (vdev && vdev->v_phys_read)  /* Has this vdev already been inited? */
 892                 return (EIO);
 893
 894         if (nvlist_find(nvlist,
 895                         ZPOOL_CONFIG_VDEV_TREE,
 896                         DATA_TYPE_NVLIST, 0, &vdevs)) {
 897                 return (EIO);
 898         }
 899
 900         rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
 901         if (rc)
 902                 return (rc);
 903
 904         /*
 905          * Add the toplevel vdev to the pool if its not already there.
 906          */
 907         STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
 908                 if (top_vdev == pool_vdev)
 909                         break;
 910         if (!pool_vdev && top_vdev)
 911                 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
 912
 913         /*
 914          * We should already have created an incomplete vdev for this
 915          * vdev. Find it and initialise it with our read proc.
 916          */
 917         vdev = vdev_find(guid);
 918         if (vdev) {
 919                 vdev->v_phys_read = read;
 920                 vdev->v_read_priv = read_priv;
 921                 vdev->v_state = VDEV_STATE_HEALTHY;
 922         } else {
 923                 printf("ZFS: inconsistent nvlist contents\n");
 924                 return (EIO);
 925         }
 926
 927         /*
 928          * Re-evaluate top-level vdev state.
 929          */
 930         vdev_set_state(top_vdev);
 931
 932         /*
 933          * Ok, we are happy with the pool so far. Lets find
 934          * the best uberblock and then we can actually access
 935          * the contents of the pool.
 936          */
 937         upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
 938         up = (const struct uberblock *)upbuf;
 939         for (i = 0;
 940              i < VDEV_UBERBLOCK_COUNT(vdev);
 941              i++) {
 942                 off = VDEV_UBERBLOCK_OFFSET(vdev, i);
 943                 BP_ZERO(&bp);
 944                 DVA_SET_OFFSET(&bp.blk_dva[0], off);
 945                 BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
 946                 BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
 947                 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
 948                 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
 949                 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
 950
 951                 if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
 952                         continue;
 953
 954                 if (up->ub_magic != UBERBLOCK_MAGIC)
 955                         continue;
 956                 if (up->ub_txg < spa->spa_txg)
 957                         continue;
 958                 if (up->ub_txg > spa->spa_uberblock.ub_txg) {
 959                         spa->spa_uberblock = *up;
 960                 } else if (up->ub_txg == spa->spa_uberblock.ub_txg) {
 961                         if (up->ub_timestamp > spa->spa_uberblock.ub_timestamp)
 962                                 spa->spa_uberblock = *up;
 963                 }
 964         }
 965         zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
 966
 967         if (spap)
 968                 *spap = spa;
 969         return (0);
 970 }
 971
 972 static int
 973 ilog2(int n)
 974 {
 975         int v;
 976
 977         for (v = 0; v < 32; v++)
 978                 if (n == (1 << v))
 979                         return v;
 980         return -1;
 981 }
 982
 983 static int
 984 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
 985 {
 986         blkptr_t gbh_bp;
 987         zio_gbh_phys_t zio_gb;
 988         char *pbuf;
 989         int i;
 990
 991         /* Artificial BP for gang block header. */
 992         gbh_bp = *bp;
 993         BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
 994         BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
 995         BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
 996         BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
 997         for (i = 0; i < SPA_DVAS_PER_BP; i++)
 998                 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
 999
1000         /* Read gang header block using the artificial BP. */
1001         if (zio_read(spa, &gbh_bp, &zio_gb))
1002                 return (EIO);
1003
1004         pbuf = buf;
1005         for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1006                 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1007
1008                 if (BP_IS_HOLE(gbp))
1009                         continue;
1010                 if (zio_read(spa, gbp, pbuf))
1011                         return (EIO);
1012                 pbuf += BP_GET_PSIZE(gbp);
1013         }
1014
1015         if (zio_checksum_verify(bp, buf))
1016                 return (EIO);
1017         return (0);
1018 }
1019
1020 static int
1021 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1022 {
1023         int cpfunc = BP_GET_COMPRESS(bp);
1024         uint64_t align, size;
1025         void *pbuf;
1026         int i, error;
1027
1028         error = EIO;
1029
1030         for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1031                 const dva_t *dva = &bp->blk_dva[i];
1032                 vdev_t *vdev;
1033                 int vdevid;
1034                 off_t offset;
1035
1036                 if (!dva->dva_word[0] && !dva->dva_word[1])
1037                         continue;
1038
1039                 vdevid = DVA_GET_VDEV(dva);
1040                 offset = DVA_GET_OFFSET(dva);
1041                 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1042                         if (vdev->v_id == vdevid)
1043                                 break;
1044                 }
1045                 if (!vdev || !vdev->v_read)
1046                         continue;
1047
1048                 size = BP_GET_PSIZE(bp);
1049                 if (vdev->v_read == vdev_raidz_read) {
1050                         align = 1ULL << vdev->v_top->v_ashift;
1051                         if (P2PHASE(size, align) != 0)
1052                                 size = P2ROUNDUP(size, align);
1053                 }
1054                 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1055                         pbuf = zfs_alloc(size);
1056                 else
1057                         pbuf = buf;
1058
1059                 if (DVA_GET_GANG(dva))
1060                         error = zio_read_gang(spa, bp, pbuf);
1061                 else
1062                         error = vdev->v_read(vdev, bp, pbuf, offset, size);
1063                 if (error == 0) {
1064                         if (cpfunc != ZIO_COMPRESS_OFF)
1065                                 error = zio_decompress_data(cpfunc, pbuf,
1066                                     BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1067                         else if (size != BP_GET_PSIZE(bp))
1068                                 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1069                 }
1070                 if (buf != pbuf)
1071                         zfs_free(pbuf, size);
1072                 if (error == 0)
1073                         break;
1074         }
1075         if (error != 0)
1076                 printf("ZFS: i/o error - all block copies unavailable\n");
1077         return (error);
1078 }
1079
1080 static int
1081 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1082 {
1083         int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1084         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1085         int nlevels = dnode->dn_nlevels;
1086         int i, rc;
1087
1088         /*
1089          * Note: bsize may not be a power of two here so we need to do an
1090          * actual divide rather than a bitshift.
1091          */
1092         while (buflen > 0) {
1093                 uint64_t bn = offset / bsize;
1094                 int boff = offset % bsize;
1095                 int ibn;
1096                 const blkptr_t *indbp;
1097                 blkptr_t bp;
1098
1099                 if (bn > dnode->dn_maxblkid)
1100                         return (EIO);
1101
1102                 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1103                         goto cached;
1104
1105                 indbp = dnode->dn_blkptr;
1106                 for (i = 0; i < nlevels; i++) {
1107                         /*
1108                          * Copy the bp from the indirect array so that
1109                          * we can re-use the scratch buffer for multi-level
1110                          * objects.
1111                          */
1112                         ibn = bn >> ((nlevels - i - 1) * ibshift);
1113                         ibn &= ((1 << ibshift) - 1);
1114                         bp = indbp[ibn];
1115                         rc = zio_read(spa, &bp, dnode_cache_buf);
1116                         if (rc)
1117                                 return (rc);
1118                         indbp = (const blkptr_t *) dnode_cache_buf;
1119                 }
1120                 dnode_cache_obj = dnode;
1121                 dnode_cache_bn = bn;
1122         cached:
1123
1124                 /*
1125                  * The buffer contains our data block. Copy what we
1126                  * need from it and loop.
1127                  */
1128                 i = bsize - boff;
1129                 if (i > buflen) i = buflen;
1130                 memcpy(buf, &dnode_cache_buf[boff], i);
1131                 buf = ((char*) buf) + i;
1132                 offset += i;
1133                 buflen -= i;
1134         }
1135
1136         return (0);
1137 }
1138
1139 /*
1140  * Lookup a value in a microzap directory. Assumes that the zap
1141  * scratch buffer contains the directory contents.
1142  */
1143 static int
1144 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1145 {
1146         const mzap_phys_t *mz;
1147         const mzap_ent_phys_t *mze;
1148         size_t size;
1149         int chunks, i;
1150
1151         /*
1152          * Microzap objects use exactly one block. Read the whole
1153          * thing.
1154          */
1155         size = dnode->dn_datablkszsec * 512;
1156
1157         mz = (const mzap_phys_t *) zap_scratch;
1158         chunks = size / MZAP_ENT_LEN - 1;
1159
1160         for (i = 0; i < chunks; i++) {
1161                 mze = &mz->mz_chunk[i];
1162                 if (!strcmp(mze->mze_name, name)) {
1163                         *value = mze->mze_value;
1164                         return (0);
1165                 }
1166         }
1167
1168         return (ENOENT);
1169 }
1170
1171 /*
1172  * Compare a name with a zap leaf entry. Return non-zero if the name
1173  * matches.
1174  */
1175 static int
1176 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1177 {
1178         size_t namelen;
1179         const zap_leaf_chunk_t *nc;
1180         const char *p;
1181
1182         namelen = zc->l_entry.le_name_numints;
1183
1184         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1185         p = name;
1186         while (namelen > 0) {
1187                 size_t len;
1188                 len = namelen;
1189                 if (len > ZAP_LEAF_ARRAY_BYTES)
1190                         len = ZAP_LEAF_ARRAY_BYTES;
1191                 if (memcmp(p, nc->l_array.la_array, len))
1192                         return (0);
1193                 p += len;
1194                 namelen -= len;
1195                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1196         }
1197
1198         return 1;
1199 }
1200
1201 /*
1202  * Extract a uint64_t value from a zap leaf entry.
1203  */
1204 static uint64_t
1205 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1206 {
1207         const zap_leaf_chunk_t *vc;
1208         int i;
1209         uint64_t value;
1210         const uint8_t *p;
1211
1212         vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1213         for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1214                 value = (value << 8) | p[i];
1215         }
1216
1217         return value;
1218 }
1219
1220 /*
1221  * Lookup a value in a fatzap directory. Assumes that the zap scratch
1222  * buffer contains the directory header.
1223  */
1224 static int
1225 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1226 {
1227         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1228         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1229         fat_zap_t z;
1230         uint64_t *ptrtbl;
1231         uint64_t hash;
1232         int rc;
1233
1234         if (zh.zap_magic != ZAP_MAGIC)
1235                 return (EIO);
1236
1237         z.zap_block_shift = ilog2(bsize);
1238         z.zap_phys = (zap_phys_t *) zap_scratch;
1239
1240         /*
1241          * Figure out where the pointer table is and read it in if necessary.
1242          */
1243         if (zh.zap_ptrtbl.zt_blk) {
1244                 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1245                                zap_scratch, bsize);
1246                 if (rc)
1247                         return (rc);
1248                 ptrtbl = (uint64_t *) zap_scratch;
1249         } else {
1250                 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1251         }
1252
1253         hash = zap_hash(zh.zap_salt, name);
1254
1255         zap_leaf_t zl;
1256         zl.l_bs = z.zap_block_shift;
1257
1258         off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1259         zap_leaf_chunk_t *zc;
1260
1261         rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1262         if (rc)
1263                 return (rc);
1264
1265         zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1266
1267         /*
1268          * Make sure this chunk matches our hash.
1269          */
1270         if (zl.l_phys->l_hdr.lh_prefix_len > 0
1271             && zl.l_phys->l_hdr.lh_prefix
1272             != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1273                 return (ENOENT);
1274
1275         /*
1276          * Hash within the chunk to find our entry.
1277          */
1278         int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1279         int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1280         h = zl.l_phys->l_hash[h];
1281         if (h == 0xffff)
1282                 return (ENOENT);
1283         zc = &ZAP_LEAF_CHUNK(&zl, h);
1284         while (zc->l_entry.le_hash != hash) {
1285                 if (zc->l_entry.le_next == 0xffff) {
1286                         zc = 0;
1287                         break;
1288                 }
1289                 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1290         }
1291         if (fzap_name_equal(&zl, zc, name)) {
1292                 *value = fzap_leaf_value(&zl, zc);
1293                 return (0);
1294         }
1295
1296         return (ENOENT);
1297 }
1298
1299 /*
1300  * Lookup a name in a zap object and return its value as a uint64_t.
1301  */
1302 static int
1303 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1304 {
1305         int rc;
1306         uint64_t zap_type;
1307         size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1308
1309         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1310         if (rc)
1311                 return (rc);
1312
1313         zap_type = *(uint64_t *) zap_scratch;
1314         if (zap_type == ZBT_MICRO)
1315                 return mzap_lookup(dnode, name, value);
1316         else if (zap_type == ZBT_HEADER)
1317                 return fzap_lookup(spa, dnode, name, value);
1318         printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
1319         return (EIO);
1320 }
1321
1322 #ifdef BOOT2
1323
1324 /*
1325  * List a microzap directory. Assumes that the zap scratch buffer contains
1326  * the directory contents.
1327  */
1328 static int
1329 mzap_list(const dnode_phys_t *dnode)
1330 {
1331         const mzap_phys_t *mz;
1332         const mzap_ent_phys_t *mze;
1333         size_t size;
1334         int chunks, i;
1335
1336         /*
1337          * Microzap objects use exactly one block. Read the whole
1338          * thing.
1339          */
1340         size = dnode->dn_datablkszsec * 512;
1341         mz = (const mzap_phys_t *) zap_scratch;
1342         chunks = size / MZAP_ENT_LEN - 1;
1343
1344         for (i = 0; i < chunks; i++) {
1345                 mze = &mz->mz_chunk[i];
1346                 if (mze->mze_name[0])
1347                         //printf("%-32s 0x%jx\n", mze->mze_name, (uintmax_t)mze->mze_value);
1348                         printf("%s\n", mze->mze_name);
1349         }
1350
1351         return (0);
1352 }
1353
1354 /*
1355  * List a fatzap directory. Assumes that the zap scratch buffer contains
1356  * the directory header.
1357  */
1358 static int
1359 fzap_list(const spa_t *spa, const dnode_phys_t *dnode)
1360 {
1361         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1362         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1363         fat_zap_t z;
1364         int i, j;
1365
1366         if (zh.zap_magic != ZAP_MAGIC)
1367                 return (EIO);
1368
1369         z.zap_block_shift = ilog2(bsize);
1370         z.zap_phys = (zap_phys_t *) zap_scratch;
1371
1372         /*
1373          * This assumes that the leaf blocks start at block 1. The
1374          * documentation isn't exactly clear on this.
1375          */
1376         zap_leaf_t zl;
1377         zl.l_bs = z.zap_block_shift;
1378         for (i = 0; i < zh.zap_num_leafs; i++) {
1379                 off_t off = (i + 1) << zl.l_bs;
1380                 char name[256], *p;
1381                 uint64_t value;
1382
1383                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1384                         return (EIO);
1385
1386                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1387
1388                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1389                         zap_leaf_chunk_t *zc, *nc;
1390                         int namelen;
1391
1392                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1393                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1394                                 continue;
1395                         namelen = zc->l_entry.le_name_numints;
1396                         if (namelen > sizeof(name))
1397                                 namelen = sizeof(name);
1398
1399                         /*
1400                          * Paste the name back together.
1401                          */
1402                         nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1403                         p = name;
1404                         while (namelen > 0) {
1405                                 int len;
1406                                 len = namelen;
1407                                 if (len > ZAP_LEAF_ARRAY_BYTES)
1408                                         len = ZAP_LEAF_ARRAY_BYTES;
1409                                 memcpy(p, nc->l_array.la_array, len);
1410                                 p += len;
1411                                 namelen -= len;
1412                                 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1413                         }
1414
1415                         /*
1416                          * Assume the first eight bytes of the value are
1417                          * a uint64_t.
1418                          */
1419                         value = fzap_leaf_value(&zl, zc);
1420
1421                         //printf("%s 0x%jx\n", name, (uintmax_t)value);
1422                         printf("%s\n", name);
1423                 }
1424         }
1425
1426         return (0);
1427 }
1428
1429 /*
1430  * List a zap directory.
1431  */
1432 static int
1433 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
1434 {
1435         uint64_t zap_type;
1436         size_t size = dnode->dn_datablkszsec * 512;
1437
1438         if (dnode_read(spa, dnode, 0, zap_scratch, size))
1439                 return (EIO);
1440
1441         zap_type = *(uint64_t *) zap_scratch;
1442         if (zap_type == ZBT_MICRO)
1443                 return mzap_list(dnode);
1444         else
1445                 return fzap_list(spa, dnode);
1446 }
1447
1448 #endif
1449
1450 static int
1451 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1452 {
1453         off_t offset;
1454
1455         offset = objnum * sizeof(dnode_phys_t);
1456         return dnode_read(spa, &os->os_meta_dnode, offset,
1457                 dnode, sizeof(dnode_phys_t));
1458 }
1459
1460 static int
1461 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1462 {
1463         const mzap_phys_t *mz;
1464         const mzap_ent_phys_t *mze;
1465         size_t size;
1466         int chunks, i;
1467
1468         /*
1469          * Microzap objects use exactly one block. Read the whole
1470          * thing.
1471          */
1472         size = dnode->dn_datablkszsec * 512;
1473
1474         mz = (const mzap_phys_t *) zap_scratch;
1475         chunks = size / MZAP_ENT_LEN - 1;
1476
1477         for (i = 0; i < chunks; i++) {
1478                 mze = &mz->mz_chunk[i];
1479                 if (value == mze->mze_value) {
1480                         strcpy(name, mze->mze_name);
1481                         return (0);
1482                 }
1483         }
1484
1485         return (ENOENT);
1486 }
1487
1488 static void
1489 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
1490 {
1491         size_t namelen;
1492         const zap_leaf_chunk_t *nc;
1493         char *p;
1494
1495         namelen = zc->l_entry.le_name_numints;
1496
1497         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1498         p = name;
1499         while (namelen > 0) {
1500                 size_t len;
1501                 len = namelen;
1502                 if (len > ZAP_LEAF_ARRAY_BYTES)
1503                         len = ZAP_LEAF_ARRAY_BYTES;
1504                 memcpy(p, nc->l_array.la_array, len);
1505                 p += len;
1506                 namelen -= len;
1507                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1508         }
1509
1510         *p = '\0';
1511 }
1512
1513 static int
1514 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1515 {
1516         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1517         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1518         fat_zap_t z;
1519         int i, j;
1520
1521         if (zh.zap_magic != ZAP_MAGIC)
1522                 return (EIO);
1523
1524         z.zap_block_shift = ilog2(bsize);
1525         z.zap_phys = (zap_phys_t *) zap_scratch;
1526
1527         /*
1528          * This assumes that the leaf blocks start at block 1. The
1529          * documentation isn't exactly clear on this.
1530          */
1531         zap_leaf_t zl;
1532         zl.l_bs = z.zap_block_shift;
1533         for (i = 0; i < zh.zap_num_leafs; i++) {
1534                 off_t off = (i + 1) << zl.l_bs;
1535
1536                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1537                         return (EIO);
1538
1539                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1540
1541                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1542                         zap_leaf_chunk_t *zc;
1543
1544                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1545                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1546                                 continue;
1547                         if (zc->l_entry.le_value_intlen != 8 ||
1548                             zc->l_entry.le_value_numints != 1)
1549                                 continue;
1550
1551                         if (fzap_leaf_value(&zl, zc) == value) {
1552                                 fzap_name_copy(&zl, zc, name);
1553                                 return (0);
1554                         }
1555                 }
1556         }
1557
1558         return (ENOENT);
1559 }
1560
1561 static int
1562 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1563 {
1564         int rc;
1565         uint64_t zap_type;
1566         size_t size = dnode->dn_datablkszsec * 512;
1567
1568         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1569         if (rc)
1570                 return (rc);
1571
1572         zap_type = *(uint64_t *) zap_scratch;
1573         if (zap_type == ZBT_MICRO)
1574                 return mzap_rlookup(spa, dnode, name, value);
1575         else
1576                 return fzap_rlookup(spa, dnode, name, value);
1577 }
1578
1579 static int
1580 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
1581 {
1582         char name[256];
1583         char component[256];
1584         uint64_t dir_obj, parent_obj, child_dir_zapobj;
1585         dnode_phys_t child_dir_zap, dataset, dir, parent;
1586         dsl_dir_phys_t *dd;
1587         dsl_dataset_phys_t *ds;
1588         char *p;
1589         int len;
1590
1591         p = &name[sizeof(name) - 1];
1592         *p = '\0';
1593
1594         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1595                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1596                 return (EIO);
1597         }
1598         ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
1599         dir_obj = ds->ds_dir_obj;
1600
1601         for (;;) {
1602                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
1603                         return (EIO);
1604                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1605
1606                 /* Actual loop condition. */
1607                 parent_obj  = dd->dd_parent_obj;
1608                 if (parent_obj == 0)
1609                         break;
1610
1611                 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
1612                         return (EIO);
1613                 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
1614                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1615                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1616                         return (EIO);
1617                 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
1618                         return (EIO);
1619
1620                 len = strlen(component);
1621                 p -= len;
1622                 memcpy(p, component, len);
1623                 --p;
1624                 *p = '/';
1625
1626                 /* Actual loop iteration. */
1627                 dir_obj = parent_obj;
1628         }
1629
1630         if (*p != '\0')
1631                 ++p;
1632         strcpy(result, p);
1633
1634         return (0);
1635 }
1636
1637 static int
1638 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
1639 {
1640         char element[256];
1641         uint64_t dir_obj, child_dir_zapobj;
1642         dnode_phys_t child_dir_zap, dir;
1643         dsl_dir_phys_t *dd;
1644         const char *p, *q;
1645
1646         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
1647                 return (EIO);
1648         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &dir_obj))
1649                 return (EIO);
1650
1651         p = name;
1652         for (;;) {
1653                 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
1654                         return (EIO);
1655                 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1656
1657                 while (*p == '/')
1658                         p++;
1659                 /* Actual loop condition #1. */
1660                 if (*p == '\0')
1661                         break;
1662
1663                 q = strchr(p, '/');
1664                 if (q) {
1665                         memcpy(element, p, q - p);
1666                         element[q - p] = '\0';
1667                         p = q + 1;
1668                 } else {
1669                         strcpy(element, p);
1670                         p += strlen(p);
1671                 }
1672
1673                 child_dir_zapobj = dd->dd_child_dir_zapobj;
1674                 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1675                         return (EIO);
1676
1677                 /* Actual loop condition #2. */
1678                 if (zap_lookup(spa, &child_dir_zap, element, &dir_obj) != 0)
1679                         return (ENOENT);
1680         }
1681
1682         *objnum = dd->dd_head_dataset_obj;
1683         return (0);
1684 }
1685
1686 /*
1687  * Find the object set given the object number of its dataset object
1688  * and return its details in *objset
1689  */
1690 static int
1691 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
1692 {
1693         dnode_phys_t dataset;
1694         dsl_dataset_phys_t *ds;
1695
1696         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1697                 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1698                 return (EIO);
1699         }
1700
1701         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1702         if (zio_read(spa, &ds->ds_bp, objset)) {
1703                 printf("ZFS: can't read object set for dataset %ju\n",
1704                     (uintmax_t)objnum);
1705                 return (EIO);
1706         }
1707
1708         return (0);
1709 }
1710
1711 /*
1712  * Find the object set pointed to by the BOOTFS property or the root
1713  * dataset if there is none and return its details in *objset
1714  */
1715 static int
1716 zfs_get_root(const spa_t *spa, uint64_t *objid)
1717 {
1718         dnode_phys_t dir, propdir;
1719         uint64_t props, bootfs, root;
1720
1721         *objid = 0;
1722
1723         /*
1724          * Start with the MOS directory object.
1725          */
1726         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
1727                 printf("ZFS: can't read MOS object directory\n");
1728                 return (EIO);
1729         }
1730
1731         /*
1732          * Lookup the pool_props and see if we can find a bootfs.
1733          */
1734         if (zap_lookup(spa, &dir, DMU_POOL_PROPS, &props) == 0
1735              && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
1736              && zap_lookup(spa, &propdir, "bootfs", &bootfs) == 0
1737              && bootfs != 0)
1738         {
1739                 *objid = bootfs;
1740                 return (0);
1741         }
1742         /*
1743          * Lookup the root dataset directory
1744          */
1745         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &root)
1746             || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
1747                 printf("ZFS: can't find root dsl_dir\n");
1748                 return (EIO);
1749         }
1750
1751         /*
1752          * Use the information from the dataset directory's bonus buffer
1753          * to find the dataset object and from that the object set itself.
1754          */
1755         dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
1756         *objid = dd->dd_head_dataset_obj;
1757         return (0);
1758 }
1759
1760 static int
1761 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
1762 {
1763
1764         mount->spa = spa;
1765
1766         /*
1767          * Find the root object set if not explicitly provided
1768          */
1769         if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
1770                 printf("ZFS: can't find root filesystem\n");
1771                 return (EIO);
1772         }
1773
1774         if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
1775                 printf("ZFS: can't open root filesystem\n");
1776                 return (EIO);
1777         }
1778
1779         mount->rootobj = rootobj;
1780
1781         return (0);
1782 }
1783
1784 static int
1785 zfs_spa_init(spa_t *spa)
1786 {
1787
1788         if (spa->spa_inited)
1789                 return (0);
1790         if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
1791                 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
1792                 return (EIO);
1793         }
1794         spa->spa_inited = 1;
1795         return (0);
1796 }
1797
1798 static int
1799 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
1800 {
1801
1802         if (dn->dn_bonustype != DMU_OT_SA) {
1803                 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
1804
1805                 sb->st_mode = zp->zp_mode;
1806                 sb->st_uid = zp->zp_uid;
1807                 sb->st_gid = zp->zp_gid;
1808                 sb->st_size = zp->zp_size;
1809         } else {
1810                 sa_hdr_phys_t *sahdrp;
1811                 int hdrsize;
1812                 size_t size = 0;
1813                 void *buf = NULL;
1814
1815                 if (dn->dn_bonuslen != 0)
1816                         sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
1817                 else {
1818                         if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
1819                                 blkptr_t *bp = &dn->dn_spill;
1820                                 int error;
1821
1822                                 size = BP_GET_LSIZE(bp);
1823                                 buf = zfs_alloc(size);
1824                                 error = zio_read(spa, bp, buf);
1825                                 if (error != 0) {
1826                                         zfs_free(buf, size);
1827                                         return (error);
1828                                 }
1829                                 sahdrp = buf;
1830                         } else {
1831                                 return (EIO);
1832                         }
1833                 }
1834                 hdrsize = SA_HDR_SIZE(sahdrp);
1835                 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
1836                     SA_MODE_OFFSET);
1837                 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
1838                     SA_UID_OFFSET);
1839                 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
1840                     SA_GID_OFFSET);
1841                 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
1842                     SA_SIZE_OFFSET);
1843                 if (buf != NULL)
1844                         zfs_free(buf, size);
1845         }
1846
1847         return (0);
1848 }
1849
1850 /*
1851  * Lookup a file and return its dnode.
1852  */
1853 static int
1854 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
1855 {
1856         int rc;
1857         uint64_t objnum, rootnum, parentnum;
1858         const spa_t *spa;
1859         dnode_phys_t dn;
1860         const char *p, *q;
1861         char element[256];
1862         char path[1024];
1863         int symlinks_followed = 0;
1864         struct stat sb;
1865
1866         spa = mount->spa;
1867         if (mount->objset.os_type != DMU_OST_ZFS) {
1868                 printf("ZFS: unexpected object set type %ju\n",
1869                     (uintmax_t)mount->objset.os_type);
1870                 return (EIO);
1871         }
1872
1873         /*
1874          * Get the root directory dnode.
1875          */
1876         rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
1877         if (rc)
1878                 return (rc);
1879
1880         rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, &rootnum);
1881         if (rc)
1882                 return (rc);
1883
1884         rc = objset_get_dnode(spa, &mount->objset, rootnum, &dn);
1885         if (rc)
1886                 return (rc);
1887
1888         objnum = rootnum;
1889         p = upath;
1890         while (p && *p) {
1891                 while (*p == '/')
1892                         p++;
1893                 if (!*p)
1894                         break;
1895                 q = strchr(p, '/');
1896                 if (q) {
1897                         memcpy(element, p, q - p);
1898                         element[q - p] = 0;
1899                         p = q;
1900                 } else {
1901                         strcpy(element, p);
1902                         p = 0;
1903                 }
1904
1905                 rc = zfs_dnode_stat(spa, &dn, &sb);
1906                 if (rc)
1907                         return (rc);
1908                 if (!S_ISDIR(sb.st_mode))
1909                         return (ENOTDIR);
1910
1911                 parentnum = objnum;
1912                 rc = zap_lookup(spa, &dn, element, &objnum);
1913                 if (rc)
1914                         return (rc);
1915                 objnum = ZFS_DIRENT_OBJ(objnum);
1916
1917                 rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
1918                 if (rc)
1919                         return (rc);
1920
1921                 /*
1922                  * Check for symlink.
1923                  */
1924                 rc = zfs_dnode_stat(spa, &dn, &sb);
1925                 if (rc)
1926                         return (rc);
1927                 if (S_ISLNK(sb.st_mode)) {
1928                         if (symlinks_followed > 10)
1929                                 return (EMLINK);
1930                         symlinks_followed++;
1931
1932                         /*
1933                          * Read the link value and copy the tail of our
1934                          * current path onto the end.
1935                          */
1936                         if (p)
1937                                 strcpy(&path[sb.st_size], p);
1938                         else
1939                                 path[sb.st_size] = 0;
1940                         if (sb.st_size + sizeof(znode_phys_t) <= dn.dn_bonuslen) {
1941                                 memcpy(path, &dn.dn_bonus[sizeof(znode_phys_t)],
1942                                         sb.st_size);
1943                         } else {
1944                                 rc = dnode_read(spa, &dn, 0, path, sb.st_size);
1945                                 if (rc)
1946                                         return (rc);
1947                         }
1948
1949                         /*
1950                          * Restart with the new path, starting either at
1951                          * the root or at the parent depending whether or
1952                          * not the link is relative.
1953                          */
1954                         p = path;
1955                         if (*p == '/')
1956                                 objnum = rootnum;
1957                         else
1958                                 objnum = parentnum;
1959                         objset_get_dnode(spa, &mount->objset, objnum, &dn);
1960                 }
1961         }
1962
1963         *dnode = dn;
1964         return (0);
1965 }