1 /* 2 * GRUB -- GRand Unified Bootloader 3 * Copyright (C) 1999,2000,2001,2002,2003,2004 Free Software Foundation, Inc. 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation; either version 2 of the License, or 8 * (at your option) any later version. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write to the Free Software 17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 18 */ 19 /* 20 * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 21 * Use is subject to license terms. 22 */ 23 24 /* 25 * The zfs plug-in routines for GRUB are: 26 * 27 * zfs_mount() - locates a valid uberblock of the root pool and reads 28 * in its MOS at the memory address MOS. 29 * 30 * zfs_open() - locates a plain file object by following the MOS 31 * and places its dnode at the memory address DNODE. 32 * 33 * zfs_read() - read in the data blocks pointed by the DNODE. 34 * 35 * ZFS_SCRATCH is used as a working area. 36 * 37 * (memory addr) MOS DNODE ZFS_SCRATCH 38 * | | | 39 * +-------V---------V----------V---------------+ 40 * memory | | dnode | dnode | scratch | 41 * | | 512B | 512B | area | 42 * +--------------------------------------------+ 43 */ 44 45 #ifdef FSYS_ZFS 46 47 #include "shared.h" 48 #include "filesys.h" 49 #include "fsys_zfs.h" 50 51 /* cache for a file block of the currently zfs_open()-ed file */ 52 static void *file_buf = NULL; 53 static uint64_t file_start = 0; 54 static uint64_t file_end = 0; 55 56 /* cache for a dnode block */ 57 static dnode_phys_t *dnode_buf = NULL; 58 static dnode_phys_t *dnode_mdn = NULL; 59 static uint64_t dnode_start = 0; 60 static uint64_t dnode_end = 0; 61 62 static uint64_t pool_guid = 0; 63 static uberblock_t current_uberblock; 64 static char *stackbase; 65 66 decomp_entry_t decomp_table[ZIO_COMPRESS_FUNCTIONS] = 67 { 68 {"inherit", 0}, /* ZIO_COMPRESS_INHERIT */ 69 {"on", lzjb_decompress}, /* ZIO_COMPRESS_ON */ 70 {"off", 0}, /* ZIO_COMPRESS_OFF */ 71 {"lzjb", lzjb_decompress}, /* ZIO_COMPRESS_LZJB */ 72 {"empty", 0} /* ZIO_COMPRESS_EMPTY */ 73 }; 74 75 static int zio_read_data(blkptr_t *bp, void *buf, char *stack); 76 77 /* 78 * Our own version of bcmp(). 79 */ 80 static int 81 zfs_bcmp(const void *s1, const void *s2, size_t n) 82 { 83 const uchar_t *ps1 = s1; 84 const uchar_t *ps2 = s2; 85 86 if (s1 != s2 && n != 0) { 87 do { 88 if (*ps1++ != *ps2++) 89 return (1); 90 } while (--n != 0); 91 } 92 93 return (0); 94 } 95 96 /* 97 * Our own version of log2(). Same thing as highbit()-1. 98 */ 99 static int 100 zfs_log2(uint64_t num) 101 { 102 int i = 0; 103 104 while (num > 1) { 105 i++; 106 num = num >> 1; 107 } 108 109 return (i); 110 } 111 112 /* Checksum Functions */ 113 static void 114 zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp) 115 { 116 ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0); 117 } 118 119 /* Checksum Table and Values */ 120 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = { 121 NULL, NULL, 0, 0, "inherit", 122 NULL, NULL, 0, 0, "on", 123 zio_checksum_off, zio_checksum_off, 0, 0, "off", 124 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "label", 125 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "gang_header", 126 NULL, NULL, 0, 0, "zilog", 127 fletcher_2_native, fletcher_2_byteswap, 0, 0, "fletcher2", 128 fletcher_4_native, fletcher_4_byteswap, 1, 0, "fletcher4", 129 zio_checksum_SHA256, zio_checksum_SHA256, 1, 0, "SHA256", 130 NULL, NULL, 0, 0, "zilog2", 131 }; 132 133 /* 134 * zio_checksum_verify: Provides support for checksum verification. 135 * 136 * Fletcher2, Fletcher4, and SHA256 are supported. 137 * 138 * Return: 139 * -1 = Failure 140 * 0 = Success 141 */ 142 static int 143 zio_checksum_verify(blkptr_t *bp, char *data, int size) 144 { 145 zio_cksum_t zc = bp->blk_cksum; 146 uint32_t checksum = BP_GET_CHECKSUM(bp); 147 int byteswap = BP_SHOULD_BYTESWAP(bp); 148 zio_eck_t *zec = (zio_eck_t *)(data + size) - 1; 149 zio_checksum_info_t *ci = &zio_checksum_table[checksum]; 150 zio_cksum_t actual_cksum, expected_cksum; 151 152 /* byteswap is not supported */ 153 if (byteswap) 154 return (-1); 155 156 if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL) 157 return (-1); 158 159 if (ci->ci_eck) { 160 expected_cksum = zec->zec_cksum; 161 zec->zec_cksum = zc; 162 ci->ci_func[0](data, size, &actual_cksum); 163 zec->zec_cksum = expected_cksum; 164 zc = expected_cksum; 165 166 } else { 167 ci->ci_func[byteswap](data, size, &actual_cksum); 168 } 169 170 if ((actual_cksum.zc_word[0] - zc.zc_word[0]) | 171 (actual_cksum.zc_word[1] - zc.zc_word[1]) | 172 (actual_cksum.zc_word[2] - zc.zc_word[2]) | 173 (actual_cksum.zc_word[3] - zc.zc_word[3])) 174 return (-1); 175 176 return (0); 177 } 178 179 /* 180 * vdev_label_start returns the physical disk offset (in bytes) of 181 * label "l". 182 */ 183 static uint64_t 184 vdev_label_start(uint64_t psize, int l) 185 { 186 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 187 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 188 } 189 190 /* 191 * vdev_uberblock_compare takes two uberblock structures and returns an integer 192 * indicating the more recent of the two. 193 * Return Value = 1 if ub2 is more recent 194 * Return Value = -1 if ub1 is more recent 195 * The most recent uberblock is determined using its transaction number and 196 * timestamp. The uberblock with the highest transaction number is 197 * considered "newer". If the transaction numbers of the two blocks match, the 198 * timestamps are compared to determine the "newer" of the two. 199 */ 200 static int 201 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 202 { 203 if (ub1->ub_txg < ub2->ub_txg) 204 return (-1); 205 if (ub1->ub_txg > ub2->ub_txg) 206 return (1); 207 208 if (ub1->ub_timestamp < ub2->ub_timestamp) 209 return (-1); 210 if (ub1->ub_timestamp > ub2->ub_timestamp) 211 return (1); 212 213 return (0); 214 } 215 216 /* 217 * Three pieces of information are needed to verify an uberblock: the magic 218 * number, the version number, and the checksum. 219 * 220 * Currently Implemented: version number, magic number 221 * Need to Implement: checksum 222 * 223 * Return: 224 * 0 - Success 225 * -1 - Failure 226 */ 227 static int 228 uberblock_verify(uberblock_t *uber, uint64_t ub_size, uint64_t offset) 229 { 230 blkptr_t bp; 231 232 BP_ZERO(&bp); 233 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL); 234 BP_SET_BYTEORDER(&bp, ZFS_HOST_BYTEORDER); 235 ZIO_SET_CHECKSUM(&bp.blk_cksum, offset, 0, 0, 0); 236 237 if (zio_checksum_verify(&bp, (char *)uber, ub_size) != 0) 238 return (-1); 239 240 if (uber->ub_magic == UBERBLOCK_MAGIC && 241 uber->ub_version > 0 && uber->ub_version <= SPA_VERSION) 242 return (0); 243 244 return (-1); 245 } 246 247 /* 248 * Find the best uberblock. 249 * Return: 250 * Success - Pointer to the best uberblock. 251 * Failure - NULL 252 */ 253 static uberblock_t * 254 find_bestub(char *ub_array, uint64_t ashift, uint64_t sector) 255 { 256 uberblock_t *ubbest = NULL; 257 uberblock_t *ubnext; 258 uint64_t offset, ub_size; 259 int i; 260 261 ub_size = VDEV_UBERBLOCK_SIZE(ashift); 262 263 for (i = 0; i < VDEV_UBERBLOCK_COUNT(ashift); i++) { 264 ubnext = (uberblock_t *)ub_array; 265 ub_array += ub_size; 266 offset = (sector << SPA_MINBLOCKSHIFT) + 267 VDEV_UBERBLOCK_OFFSET(ashift, i); 268 269 if (uberblock_verify(ubnext, ub_size, offset) != 0) 270 continue; 271 272 if (ubbest == NULL || 273 vdev_uberblock_compare(ubnext, ubbest) > 0) 274 ubbest = ubnext; 275 } 276 277 return (ubbest); 278 } 279 280 /* 281 * Read a block of data based on the gang block address dva, 282 * and put its data in buf. 283 * 284 * Return: 285 * 0 - success 286 * 1 - failure 287 */ 288 static int 289 zio_read_gang(blkptr_t *bp, dva_t *dva, void *buf, char *stack) 290 { 291 zio_gbh_phys_t *zio_gb; 292 uint64_t offset, sector; 293 blkptr_t tmpbp; 294 int i; 295 296 zio_gb = (zio_gbh_phys_t *)stack; 297 stack += SPA_GANGBLOCKSIZE; 298 offset = DVA_GET_OFFSET(dva); 299 sector = DVA_OFFSET_TO_PHYS_SECTOR(offset); 300 301 /* read in the gang block header */ 302 if (devread(sector, 0, SPA_GANGBLOCKSIZE, (char *)zio_gb) == 0) { 303 grub_printf("failed to read in a gang block header\n"); 304 return (1); 305 } 306 307 /* self checksuming the gang block header */ 308 BP_ZERO(&tmpbp); 309 BP_SET_CHECKSUM(&tmpbp, ZIO_CHECKSUM_GANG_HEADER); 310 BP_SET_BYTEORDER(&tmpbp, ZFS_HOST_BYTEORDER); 311 ZIO_SET_CHECKSUM(&tmpbp.blk_cksum, DVA_GET_VDEV(dva), 312 DVA_GET_OFFSET(dva), bp->blk_birth, 0); 313 if (zio_checksum_verify(&tmpbp, (char *)zio_gb, SPA_GANGBLOCKSIZE)) { 314 grub_printf("failed to checksum a gang block header\n"); 315 return (1); 316 } 317 318 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) { 319 if (zio_gb->zg_blkptr[i].blk_birth == 0) 320 continue; 321 322 if (zio_read_data(&zio_gb->zg_blkptr[i], buf, stack)) 323 return (1); 324 buf += BP_GET_PSIZE(&zio_gb->zg_blkptr[i]); 325 } 326 327 return (0); 328 } 329 330 /* 331 * Read in a block of raw data to buf. 332 * 333 * Return: 334 * 0 - success 335 * 1 - failure 336 */ 337 static int 338 zio_read_data(blkptr_t *bp, void *buf, char *stack) 339 { 340 int i, psize; 341 342 psize = BP_GET_PSIZE(bp); 343 344 /* pick a good dva from the block pointer */ 345 for (i = 0; i < SPA_DVAS_PER_BP; i++) { 346 uint64_t offset, sector; 347 348 if (bp->blk_dva[i].dva_word[0] == 0 && 349 bp->blk_dva[i].dva_word[1] == 0) 350 continue; 351 352 if (DVA_GET_GANG(&bp->blk_dva[i])) { 353 if (zio_read_gang(bp, &bp->blk_dva[i], buf, stack) == 0) 354 return (0); 355 } else { 356 /* read in a data block */ 357 offset = DVA_GET_OFFSET(&bp->blk_dva[i]); 358 sector = DVA_OFFSET_TO_PHYS_SECTOR(offset); 359 if (devread(sector, 0, psize, buf)) 360 return (0); 361 } 362 } 363 364 return (1); 365 } 366 367 /* 368 * Read in a block of data, verify its checksum, decompress if needed, 369 * and put the uncompressed data in buf. 370 * 371 * Return: 372 * 0 - success 373 * errnum - failure 374 */ 375 static int 376 zio_read(blkptr_t *bp, void *buf, char *stack) 377 { 378 int lsize, psize, comp; 379 char *retbuf; 380 381 comp = BP_GET_COMPRESS(bp); 382 lsize = BP_GET_LSIZE(bp); 383 psize = BP_GET_PSIZE(bp); 384 385 if ((unsigned int)comp >= ZIO_COMPRESS_FUNCTIONS || 386 (comp != ZIO_COMPRESS_OFF && 387 decomp_table[comp].decomp_func == NULL)) { 388 grub_printf("compression algorithm not supported\n"); 389 return (ERR_FSYS_CORRUPT); 390 } 391 392 if ((char *)buf < stack && ((char *)buf) + lsize > stack) { 393 grub_printf("not enough memory allocated\n"); 394 return (ERR_WONT_FIT); 395 } 396 397 retbuf = buf; 398 if (comp != ZIO_COMPRESS_OFF) { 399 buf = stack; 400 stack += psize; 401 } 402 403 if (zio_read_data(bp, buf, stack)) { 404 grub_printf("zio_read_data failed\n"); 405 return (ERR_FSYS_CORRUPT); 406 } 407 408 if (zio_checksum_verify(bp, buf, psize) != 0) { 409 grub_printf("checksum verification failed\n"); 410 return (ERR_FSYS_CORRUPT); 411 } 412 413 if (comp != ZIO_COMPRESS_OFF) 414 decomp_table[comp].decomp_func(buf, retbuf, psize, lsize); 415 416 return (0); 417 } 418 419 /* 420 * Get the block from a block id. 421 * push the block onto the stack. 422 * 423 * Return: 424 * 0 - success 425 * errnum - failure 426 */ 427 static int 428 dmu_read(dnode_phys_t *dn, uint64_t blkid, void *buf, char *stack) 429 { 430 int idx, level; 431 blkptr_t *bp_array = dn->dn_blkptr; 432 int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 433 blkptr_t *bp, *tmpbuf; 434 435 bp = (blkptr_t *)stack; 436 stack += sizeof (blkptr_t); 437 438 tmpbuf = (blkptr_t *)stack; 439 stack += 1<<dn->dn_indblkshift; 440 441 for (level = dn->dn_nlevels - 1; level >= 0; level--) { 442 idx = (blkid >> (epbs * level)) & ((1<<epbs)-1); 443 *bp = bp_array[idx]; 444 if (level == 0) 445 tmpbuf = buf; 446 if (BP_IS_HOLE(bp)) { 447 grub_memset(buf, 0, 448 dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); 449 break; 450 } else if (errnum = zio_read(bp, tmpbuf, stack)) { 451 return (errnum); 452 } 453 454 bp_array = tmpbuf; 455 } 456 457 return (0); 458 } 459 460 /* 461 * mzap_lookup: Looks up property described by "name" and returns the value 462 * in "value". 463 * 464 * Return: 465 * 0 - success 466 * errnum - failure 467 */ 468 static int 469 mzap_lookup(mzap_phys_t *zapobj, int objsize, char *name, 470 uint64_t *value) 471 { 472 int i, chunks; 473 mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk; 474 475 chunks = objsize/MZAP_ENT_LEN - 1; 476 for (i = 0; i < chunks; i++) { 477 if (grub_strcmp(mzap_ent[i].mze_name, name) == 0) { 478 *value = mzap_ent[i].mze_value; 479 return (0); 480 } 481 } 482 483 return (ERR_FSYS_CORRUPT); 484 } 485 486 static uint64_t 487 zap_hash(uint64_t salt, const char *name) 488 { 489 static uint64_t table[256]; 490 const uint8_t *cp; 491 uint8_t c; 492 uint64_t crc = salt; 493 494 if (table[128] == 0) { 495 uint64_t *ct; 496 int i, j; 497 for (i = 0; i < 256; i++) { 498 for (ct = table + i, *ct = i, j = 8; j > 0; j--) 499 *ct = (*ct >> 1) ^ (-(*ct & 1) & 500 ZFS_CRC64_POLY); 501 } 502 } 503 504 if (crc == 0 || table[128] != ZFS_CRC64_POLY) { 505 errnum = ERR_FSYS_CORRUPT; 506 return (0); 507 } 508 509 for (cp = (const uint8_t *)name; (c = *cp) != '\0'; cp++) 510 crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF]; 511 512 /* 513 * Only use 28 bits, since we need 4 bits in the cookie for the 514 * collision differentiator. We MUST use the high bits, since 515 * those are the onces that we first pay attention to when 516 * chosing the bucket. 517 */ 518 crc &= ~((1ULL << (64 - 28)) - 1); 519 520 return (crc); 521 } 522 523 /* 524 * Only to be used on 8-bit arrays. 525 * array_len is actual len in bytes (not encoded le_value_length). 526 * buf is null-terminated. 527 */ 528 static int 529 zap_leaf_array_equal(zap_leaf_phys_t *l, int blksft, int chunk, 530 int array_len, const char *buf) 531 { 532 int bseen = 0; 533 534 while (bseen < array_len) { 535 struct zap_leaf_array *la = 536 &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array; 537 int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES); 538 539 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 540 return (0); 541 542 if (zfs_bcmp(la->la_array, buf + bseen, toread) != 0) 543 break; 544 chunk = la->la_next; 545 bseen += toread; 546 } 547 return (bseen == array_len); 548 } 549 550 /* 551 * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the 552 * value for the property "name". 553 * 554 * Return: 555 * 0 - success 556 * errnum - failure 557 */ 558 static int 559 zap_leaf_lookup(zap_leaf_phys_t *l, int blksft, uint64_t h, 560 const char *name, uint64_t *value) 561 { 562 uint16_t chunk; 563 struct zap_leaf_entry *le; 564 565 /* Verify if this is a valid leaf block */ 566 if (l->l_hdr.lh_block_type != ZBT_LEAF) 567 return (ERR_FSYS_CORRUPT); 568 if (l->l_hdr.lh_magic != ZAP_LEAF_MAGIC) 569 return (ERR_FSYS_CORRUPT); 570 571 for (chunk = l->l_hash[LEAF_HASH(blksft, h)]; 572 chunk != CHAIN_END; chunk = le->le_next) { 573 574 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 575 return (ERR_FSYS_CORRUPT); 576 577 le = ZAP_LEAF_ENTRY(l, blksft, chunk); 578 579 /* Verify the chunk entry */ 580 if (le->le_type != ZAP_CHUNK_ENTRY) 581 return (ERR_FSYS_CORRUPT); 582 583 if (le->le_hash != h) 584 continue; 585 586 if (zap_leaf_array_equal(l, blksft, le->le_name_chunk, 587 le->le_name_length, name)) { 588 589 struct zap_leaf_array *la; 590 uint8_t *ip; 591 592 if (le->le_int_size != 8 || le->le_value_length != 1) 593 return (ERR_FSYS_CORRUPT); 594 595 /* get the uint64_t property value */ 596 la = &ZAP_LEAF_CHUNK(l, blksft, 597 le->le_value_chunk).l_array; 598 ip = la->la_array; 599 600 *value = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 | 601 (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 | 602 (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 | 603 (uint64_t)ip[6] << 8 | (uint64_t)ip[7]; 604 605 return (0); 606 } 607 } 608 609 return (ERR_FSYS_CORRUPT); 610 } 611 612 /* 613 * Fat ZAP lookup 614 * 615 * Return: 616 * 0 - success 617 * errnum - failure 618 */ 619 static int 620 fzap_lookup(dnode_phys_t *zap_dnode, zap_phys_t *zap, 621 char *name, uint64_t *value, char *stack) 622 { 623 zap_leaf_phys_t *l; 624 uint64_t hash, idx, blkid; 625 int blksft = zfs_log2(zap_dnode->dn_datablkszsec << DNODE_SHIFT); 626 627 /* Verify if this is a fat zap header block */ 628 if (zap->zap_magic != (uint64_t)ZAP_MAGIC || 629 zap->zap_flags != 0) 630 return (ERR_FSYS_CORRUPT); 631 632 hash = zap_hash(zap->zap_salt, name); 633 if (errnum) 634 return (errnum); 635 636 /* get block id from index */ 637 if (zap->zap_ptrtbl.zt_numblks != 0) { 638 /* external pointer tables not supported */ 639 return (ERR_FSYS_CORRUPT); 640 } 641 idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift); 642 blkid = ((uint64_t *)zap)[idx + (1<<(blksft-3-1))]; 643 644 /* Get the leaf block */ 645 l = (zap_leaf_phys_t *)stack; 646 stack += 1<<blksft; 647 if ((1<<blksft) < sizeof (zap_leaf_phys_t)) 648 return (ERR_FSYS_CORRUPT); 649 if (errnum = dmu_read(zap_dnode, blkid, l, stack)) 650 return (errnum); 651 652 return (zap_leaf_lookup(l, blksft, hash, name, value)); 653 } 654 655 /* 656 * Read in the data of a zap object and find the value for a matching 657 * property name. 658 * 659 * Return: 660 * 0 - success 661 * errnum - failure 662 */ 663 static int 664 zap_lookup(dnode_phys_t *zap_dnode, char *name, uint64_t *val, char *stack) 665 { 666 uint64_t block_type; 667 int size; 668 void *zapbuf; 669 670 /* Read in the first block of the zap object data. */ 671 zapbuf = stack; 672 size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 673 stack += size; 674 675 if (errnum = dmu_read(zap_dnode, 0, zapbuf, stack)) 676 return (errnum); 677 678 block_type = *((uint64_t *)zapbuf); 679 680 if (block_type == ZBT_MICRO) { 681 return (mzap_lookup(zapbuf, size, name, val)); 682 } else if (block_type == ZBT_HEADER) { 683 /* this is a fat zap */ 684 return (fzap_lookup(zap_dnode, zapbuf, name, 685 val, stack)); 686 } 687 688 return (ERR_FSYS_CORRUPT); 689 } 690 691 /* 692 * Get the dnode of an object number from the metadnode of an object set. 693 * 694 * Input 695 * mdn - metadnode to get the object dnode 696 * objnum - object number for the object dnode 697 * buf - data buffer that holds the returning dnode 698 * stack - scratch area 699 * 700 * Return: 701 * 0 - success 702 * errnum - failure 703 */ 704 static int 705 dnode_get(dnode_phys_t *mdn, uint64_t objnum, uint8_t type, dnode_phys_t *buf, 706 char *stack) 707 { 708 uint64_t blkid, blksz; /* the block id this object dnode is in */ 709 int epbs; /* shift of number of dnodes in a block */ 710 int idx; /* index within a block */ 711 dnode_phys_t *dnbuf; 712 713 blksz = mdn->dn_datablkszsec << SPA_MINBLOCKSHIFT; 714 epbs = zfs_log2(blksz) - DNODE_SHIFT; 715 blkid = objnum >> epbs; 716 idx = objnum & ((1<<epbs)-1); 717 718 if (dnode_buf != NULL && dnode_mdn == mdn && 719 objnum >= dnode_start && objnum < dnode_end) { 720 grub_memmove(buf, &dnode_buf[idx], DNODE_SIZE); 721 VERIFY_DN_TYPE(buf, type); 722 return (0); 723 } 724 725 if (dnode_buf && blksz == 1<<DNODE_BLOCK_SHIFT) { 726 dnbuf = dnode_buf; 727 dnode_mdn = mdn; 728 dnode_start = blkid << epbs; 729 dnode_end = (blkid + 1) << epbs; 730 } else { 731 dnbuf = (dnode_phys_t *)stack; 732 stack += blksz; 733 } 734 735 if (errnum = dmu_read(mdn, blkid, (char *)dnbuf, stack)) 736 return (errnum); 737 738 grub_memmove(buf, &dnbuf[idx], DNODE_SIZE); 739 VERIFY_DN_TYPE(buf, type); 740 741 return (0); 742 } 743 744 /* 745 * Check if this is a special file that resides at the top 746 * dataset of the pool. Currently this is the GRUB menu, 747 * boot signature and boot signature backup. 748 * str starts with '/'. 749 */ 750 static int 751 is_top_dataset_file(char *str) 752 { 753 char *tptr; 754 755 if ((tptr = grub_strstr(str, "menu.lst")) && 756 (tptr[8] == '\0' || tptr[8] == ' ') && 757 *(tptr-1) == '/') 758 return (1); 759 760 if (grub_strncmp(str, BOOTSIGN_DIR"/", 761 grub_strlen(BOOTSIGN_DIR) + 1) == 0) 762 return (1); 763 764 if (grub_strcmp(str, BOOTSIGN_BACKUP) == 0) 765 return (1); 766 767 return (0); 768 } 769 770 /* 771 * Get the file dnode for a given file name where mdn is the meta dnode 772 * for this ZFS object set. When found, place the file dnode in dn. 773 * The 'path' argument will be mangled. 774 * 775 * Return: 776 * 0 - success 777 * errnum - failure 778 */ 779 static int 780 dnode_get_path(dnode_phys_t *mdn, char *path, dnode_phys_t *dn, 781 char *stack) 782 { 783 uint64_t objnum, version; 784 char *cname, ch; 785 786 if (errnum = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE, 787 dn, stack)) 788 return (errnum); 789 790 if (errnum = zap_lookup(dn, ZPL_VERSION_STR, &version, stack)) 791 return (errnum); 792 if (version > ZPL_VERSION) 793 return (-1); 794 795 if (errnum = zap_lookup(dn, ZFS_ROOT_OBJ, &objnum, stack)) 796 return (errnum); 797 798 if (errnum = dnode_get(mdn, objnum, DMU_OT_DIRECTORY_CONTENTS, 799 dn, stack)) 800 return (errnum); 801 802 /* skip leading slashes */ 803 while (*path == '/') 804 path++; 805 806 while (*path && !isspace(*path)) { 807 808 /* get the next component name */ 809 cname = path; 810 while (*path && !isspace(*path) && *path != '/') 811 path++; 812 ch = *path; 813 *path = 0; /* ensure null termination */ 814 815 if (errnum = zap_lookup(dn, cname, &objnum, stack)) 816 return (errnum); 817 818 objnum = ZFS_DIRENT_OBJ(objnum); 819 if (errnum = dnode_get(mdn, objnum, 0, dn, stack)) 820 return (errnum); 821 822 *path = ch; 823 while (*path == '/') 824 path++; 825 } 826 827 /* We found the dnode for this file. Verify if it is a plain file. */ 828 VERIFY_DN_TYPE(dn, DMU_OT_PLAIN_FILE_CONTENTS); 829 830 return (0); 831 } 832 833 /* 834 * Get the default 'bootfs' property value from the rootpool. 835 * 836 * Return: 837 * 0 - success 838 * errnum -failure 839 */ 840 static int 841 get_default_bootfsobj(dnode_phys_t *mosmdn, uint64_t *obj, char *stack) 842 { 843 uint64_t objnum = 0; 844 dnode_phys_t *dn = (dnode_phys_t *)stack; 845 stack += DNODE_SIZE; 846 847 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 848 DMU_OT_OBJECT_DIRECTORY, dn, stack)) 849 return (errnum); 850 851 /* 852 * find the object number for 'pool_props', and get the dnode 853 * of the 'pool_props'. 854 */ 855 if (zap_lookup(dn, DMU_POOL_PROPS, &objnum, stack)) 856 return (ERR_FILESYSTEM_NOT_FOUND); 857 858 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_POOL_PROPS, dn, stack)) 859 return (errnum); 860 861 if (zap_lookup(dn, ZPOOL_PROP_BOOTFS, &objnum, stack)) 862 return (ERR_FILESYSTEM_NOT_FOUND); 863 864 if (!objnum) 865 return (ERR_FILESYSTEM_NOT_FOUND); 866 867 *obj = objnum; 868 return (0); 869 } 870 871 /* 872 * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname), 873 * e.g. pool/rootfs, or a given object number (obj), e.g. the object number 874 * of pool/rootfs. 875 * 876 * If no fsname and no obj are given, return the DSL_DIR metadnode. 877 * If fsname is given, return its metadnode and its matching object number. 878 * If only obj is given, return the metadnode for this object number. 879 * 880 * Return: 881 * 0 - success 882 * errnum - failure 883 */ 884 static int 885 get_objset_mdn(dnode_phys_t *mosmdn, char *fsname, uint64_t *obj, 886 dnode_phys_t *mdn, char *stack) 887 { 888 uint64_t objnum, headobj; 889 char *cname, ch; 890 blkptr_t *bp; 891 objset_phys_t *osp; 892 int issnapshot = 0; 893 char *snapname; 894 895 if (fsname == NULL && obj) { 896 headobj = *obj; 897 goto skip; 898 } 899 900 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 901 DMU_OT_OBJECT_DIRECTORY, mdn, stack)) 902 return (errnum); 903 904 if (errnum = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum, 905 stack)) 906 return (errnum); 907 908 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, stack)) 909 return (errnum); 910 911 if (fsname == NULL) { 912 headobj = 913 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 914 goto skip; 915 } 916 917 /* take out the pool name */ 918 while (*fsname && !isspace(*fsname) && *fsname != '/') 919 fsname++; 920 921 while (*fsname && !isspace(*fsname)) { 922 uint64_t childobj; 923 924 while (*fsname == '/') 925 fsname++; 926 927 cname = fsname; 928 while (*fsname && !isspace(*fsname) && *fsname != '/') 929 fsname++; 930 ch = *fsname; 931 *fsname = 0; 932 933 snapname = cname; 934 while (*snapname && !isspace(*snapname) && *snapname != '@') 935 snapname++; 936 if (*snapname == '@') { 937 issnapshot = 1; 938 *snapname = 0; 939 } 940 childobj = 941 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_child_dir_zapobj; 942 if (errnum = dnode_get(mosmdn, childobj, 943 DMU_OT_DSL_DIR_CHILD_MAP, mdn, stack)) 944 return (errnum); 945 946 if (zap_lookup(mdn, cname, &objnum, stack)) 947 return (ERR_FILESYSTEM_NOT_FOUND); 948 949 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, 950 mdn, stack)) 951 return (errnum); 952 953 *fsname = ch; 954 if (issnapshot) 955 *snapname = '@'; 956 } 957 headobj = ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 958 if (obj) 959 *obj = headobj; 960 961 skip: 962 if (errnum = dnode_get(mosmdn, headobj, DMU_OT_DSL_DATASET, mdn, stack)) 963 return (errnum); 964 if (issnapshot) { 965 uint64_t snapobj; 966 967 snapobj = ((dsl_dataset_phys_t *)DN_BONUS(mdn))-> 968 ds_snapnames_zapobj; 969 970 if (errnum = dnode_get(mosmdn, snapobj, 971 DMU_OT_DSL_DS_SNAP_MAP, mdn, stack)) 972 return (errnum); 973 if (zap_lookup(mdn, snapname + 1, &headobj, stack)) 974 return (ERR_FILESYSTEM_NOT_FOUND); 975 if (errnum = dnode_get(mosmdn, headobj, 976 DMU_OT_DSL_DATASET, mdn, stack)) 977 return (errnum); 978 if (obj) 979 *obj = headobj; 980 } 981 982 bp = &((dsl_dataset_phys_t *)DN_BONUS(mdn))->ds_bp; 983 osp = (objset_phys_t *)stack; 984 stack += sizeof (objset_phys_t); 985 if (errnum = zio_read(bp, osp, stack)) 986 return (errnum); 987 988 grub_memmove((char *)mdn, (char *)&osp->os_meta_dnode, DNODE_SIZE); 989 990 return (0); 991 } 992 993 /* 994 * For a given XDR packed nvlist, verify the first 4 bytes and move on. 995 * 996 * An XDR packed nvlist is encoded as (comments from nvs_xdr_create) : 997 * 998 * encoding method/host endian (4 bytes) 999 * nvl_version (4 bytes) 1000 * nvl_nvflag (4 bytes) 1001 * encoded nvpairs: 1002 * encoded size of the nvpair (4 bytes) 1003 * decoded size of the nvpair (4 bytes) 1004 * name string size (4 bytes) 1005 * name string data (sizeof(NV_ALIGN4(string)) 1006 * data type (4 bytes) 1007 * # of elements in the nvpair (4 bytes) 1008 * data 1009 * 2 zero's for the last nvpair 1010 * (end of the entire list) (8 bytes) 1011 * 1012 * Return: 1013 * 0 - success 1014 * 1 - failure 1015 */ 1016 static int 1017 nvlist_unpack(char *nvlist, char **out) 1018 { 1019 /* Verify if the 1st and 2nd byte in the nvlist are valid. */ 1020 if (nvlist[0] != NV_ENCODE_XDR || nvlist[1] != HOST_ENDIAN) 1021 return (1); 1022 1023 nvlist += 4; 1024 *out = nvlist; 1025 return (0); 1026 } 1027 1028 static char * 1029 nvlist_array(char *nvlist, int index) 1030 { 1031 int i, encode_size; 1032 1033 for (i = 0; i < index; i++) { 1034 /* skip the header, nvl_version, and nvl_nvflag */ 1035 nvlist = nvlist + 4 * 2; 1036 1037 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) 1038 nvlist += encode_size; /* goto the next nvpair */ 1039 1040 nvlist = nvlist + 4 * 2; /* skip the ending 2 zeros - 8 bytes */ 1041 } 1042 1043 return (nvlist); 1044 } 1045 1046 static int 1047 nvlist_lookup_value(char *nvlist, char *name, void *val, int valtype, 1048 int *nelmp) 1049 { 1050 int name_len, type, slen, encode_size; 1051 char *nvpair, *nvp_name, *strval = val; 1052 uint64_t *intval = val; 1053 1054 /* skip the header, nvl_version, and nvl_nvflag */ 1055 nvlist = nvlist + 4 * 2; 1056 1057 /* 1058 * Loop thru the nvpair list 1059 * The XDR representation of an integer is in big-endian byte order. 1060 */ 1061 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) { 1062 1063 nvpair = nvlist + 4 * 2; /* skip the encode/decode size */ 1064 1065 name_len = BSWAP_32(*(uint32_t *)nvpair); 1066 nvpair += 4; 1067 1068 nvp_name = nvpair; 1069 nvpair = nvpair + ((name_len + 3) & ~3); /* align */ 1070 1071 type = BSWAP_32(*(uint32_t *)nvpair); 1072 nvpair += 4; 1073 1074 if ((grub_strncmp(nvp_name, name, name_len) == 0) && 1075 type == valtype) { 1076 int nelm; 1077 1078 if ((nelm = BSWAP_32(*(uint32_t *)nvpair)) < 1) 1079 return (1); 1080 nvpair += 4; 1081 1082 switch (valtype) { 1083 case DATA_TYPE_STRING: 1084 slen = BSWAP_32(*(uint32_t *)nvpair); 1085 nvpair += 4; 1086 grub_memmove(strval, nvpair, slen); 1087 strval[slen] = '\0'; 1088 return (0); 1089 1090 case DATA_TYPE_UINT64: 1091 *intval = BSWAP_64(*(uint64_t *)nvpair); 1092 return (0); 1093 1094 case DATA_TYPE_NVLIST: 1095 *(void **)val = (void *)nvpair; 1096 return (0); 1097 1098 case DATA_TYPE_NVLIST_ARRAY: 1099 *(void **)val = (void *)nvpair; 1100 if (nelmp) 1101 *nelmp = nelm; 1102 return (0); 1103 } 1104 } 1105 1106 nvlist += encode_size; /* goto the next nvpair */ 1107 } 1108 1109 return (1); 1110 } 1111 1112 /* 1113 * Check if this vdev is online and is in a good state. 1114 */ 1115 static int 1116 vdev_validate(char *nv) 1117 { 1118 uint64_t ival; 1119 1120 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_OFFLINE, &ival, 1121 DATA_TYPE_UINT64, NULL) == 0 || 1122 nvlist_lookup_value(nv, ZPOOL_CONFIG_FAULTED, &ival, 1123 DATA_TYPE_UINT64, NULL) == 0 || 1124 nvlist_lookup_value(nv, ZPOOL_CONFIG_REMOVED, &ival, 1125 DATA_TYPE_UINT64, NULL) == 0) 1126 return (ERR_DEV_VALUES); 1127 1128 return (0); 1129 } 1130 1131 /* 1132 * Get a valid vdev pathname/devid from the boot device. 1133 * The caller should already allocate MAXPATHLEN memory for bootpath and devid. 1134 */ 1135 static int 1136 vdev_get_bootpath(char *nv, uint64_t inguid, char *devid, char *bootpath, 1137 int is_spare) 1138 { 1139 char type[16]; 1140 1141 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_TYPE, &type, DATA_TYPE_STRING, 1142 NULL)) 1143 return (ERR_FSYS_CORRUPT); 1144 1145 if (strcmp(type, VDEV_TYPE_DISK) == 0) { 1146 uint64_t guid; 1147 1148 if (vdev_validate(nv) != 0) 1149 return (ERR_NO_BOOTPATH); 1150 1151 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_GUID, 1152 &guid, DATA_TYPE_UINT64, NULL) != 0) 1153 return (ERR_NO_BOOTPATH); 1154 1155 if (guid != inguid) 1156 return (ERR_NO_BOOTPATH); 1157 1158 /* for a spare vdev, pick the disk labeled with "is_spare" */ 1159 if (is_spare) { 1160 uint64_t spare = 0; 1161 (void) nvlist_lookup_value(nv, ZPOOL_CONFIG_IS_SPARE, 1162 &spare, DATA_TYPE_UINT64, NULL); 1163 if (!spare) 1164 return (ERR_NO_BOOTPATH); 1165 } 1166 1167 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_PHYS_PATH, 1168 bootpath, DATA_TYPE_STRING, NULL) != 0) 1169 bootpath[0] = '\0'; 1170 1171 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_DEVID, 1172 devid, DATA_TYPE_STRING, NULL) != 0) 1173 devid[0] = '\0'; 1174 1175 if (strlen(bootpath) >= MAXPATHLEN || 1176 strlen(devid) >= MAXPATHLEN) 1177 return (ERR_WONT_FIT); 1178 1179 return (0); 1180 1181 } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 || 1182 strcmp(type, VDEV_TYPE_REPLACING) == 0 || 1183 (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) { 1184 int nelm, i; 1185 char *child; 1186 1187 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_CHILDREN, &child, 1188 DATA_TYPE_NVLIST_ARRAY, &nelm)) 1189 return (ERR_FSYS_CORRUPT); 1190 1191 for (i = 0; i < nelm; i++) { 1192 char *child_i; 1193 1194 child_i = nvlist_array(child, i); 1195 if (vdev_get_bootpath(child_i, inguid, devid, 1196 bootpath, is_spare) == 0) 1197 return (0); 1198 } 1199 } 1200 1201 return (ERR_NO_BOOTPATH); 1202 } 1203 1204 /* 1205 * Check the disk label information and retrieve needed vdev name-value pairs. 1206 * 1207 * Return: 1208 * 0 - success 1209 * ERR_* - failure 1210 */ 1211 int 1212 check_pool_label(uint64_t sector, char *stack, char *outdevid, 1213 char *outpath, uint64_t *outguid, uint64_t *outashift) 1214 { 1215 vdev_phys_t *vdev; 1216 uint64_t pool_state, txg = 0; 1217 char *nvlist, *nv; 1218 uint64_t diskguid; 1219 uint64_t version; 1220 1221 sector += (VDEV_SKIP_SIZE >> SPA_MINBLOCKSHIFT); 1222 1223 /* Read in the vdev name-value pair list (112K). */ 1224 if (devread(sector, 0, VDEV_PHYS_SIZE, stack) == 0) 1225 return (ERR_READ); 1226 1227 vdev = (vdev_phys_t *)stack; 1228 stack += sizeof (vdev_phys_t); 1229 1230 if (nvlist_unpack(vdev->vp_nvlist, &nvlist)) 1231 return (ERR_FSYS_CORRUPT); 1232 1233 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_STATE, &pool_state, 1234 DATA_TYPE_UINT64, NULL)) 1235 return (ERR_FSYS_CORRUPT); 1236 1237 if (pool_state == POOL_STATE_DESTROYED) 1238 return (ERR_FILESYSTEM_NOT_FOUND); 1239 1240 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_NAME, 1241 current_rootpool, DATA_TYPE_STRING, NULL)) 1242 return (ERR_FSYS_CORRUPT); 1243 1244 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_TXG, &txg, 1245 DATA_TYPE_UINT64, NULL)) 1246 return (ERR_FSYS_CORRUPT); 1247 1248 /* not an active device */ 1249 if (txg == 0) 1250 return (ERR_NO_BOOTPATH); 1251 1252 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VERSION, &version, 1253 DATA_TYPE_UINT64, NULL)) 1254 return (ERR_FSYS_CORRUPT); 1255 if (version > SPA_VERSION) 1256 return (ERR_NEWER_VERSION); 1257 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VDEV_TREE, &nv, 1258 DATA_TYPE_NVLIST, NULL)) 1259 return (ERR_FSYS_CORRUPT); 1260 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_GUID, &diskguid, 1261 DATA_TYPE_UINT64, NULL)) 1262 return (ERR_FSYS_CORRUPT); 1263 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_ASHIFT, outashift, 1264 DATA_TYPE_UINT64, NULL) != 0) 1265 return (ERR_FSYS_CORRUPT); 1266 if (vdev_get_bootpath(nv, diskguid, outdevid, outpath, 0)) 1267 return (ERR_NO_BOOTPATH); 1268 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_GUID, outguid, 1269 DATA_TYPE_UINT64, NULL)) 1270 return (ERR_FSYS_CORRUPT); 1271 return (0); 1272 } 1273 1274 /* 1275 * zfs_mount() locates a valid uberblock of the root pool and read in its MOS 1276 * to the memory address MOS. 1277 * 1278 * Return: 1279 * 1 - success 1280 * 0 - failure 1281 */ 1282 int 1283 zfs_mount(void) 1284 { 1285 char *stack, *ub_array; 1286 int label = 0; 1287 uberblock_t *ubbest; 1288 objset_phys_t *osp; 1289 char tmp_bootpath[MAXNAMELEN]; 1290 char tmp_devid[MAXNAMELEN]; 1291 uint64_t tmp_guid, ashift; 1292 uint64_t adjpl = (uint64_t)part_length << SPA_MINBLOCKSHIFT; 1293 int err = errnum; /* preserve previous errnum state */ 1294 1295 /* if it's our first time here, zero the best uberblock out */ 1296 if (best_drive == 0 && best_part == 0 && find_best_root) { 1297 grub_memset(¤t_uberblock, 0, sizeof (uberblock_t)); 1298 pool_guid = 0; 1299 } 1300 1301 stackbase = ZFS_SCRATCH; 1302 stack = stackbase; 1303 ub_array = stack; 1304 stack += VDEV_UBERBLOCK_RING; 1305 1306 osp = (objset_phys_t *)stack; 1307 stack += sizeof (objset_phys_t); 1308 adjpl = P2ALIGN(adjpl, (uint64_t)sizeof (vdev_label_t)); 1309 1310 for (label = 0; label < VDEV_LABELS; label++) { 1311 1312 /* 1313 * some eltorito stacks don't give us a size and 1314 * we end up setting the size to MAXUINT, further 1315 * some of these devices stop working once a single 1316 * read past the end has been issued. Checking 1317 * for a maximum part_length and skipping the backup 1318 * labels at the end of the slice/partition/device 1319 * avoids breaking down on such devices. 1320 */ 1321 if (part_length == MAXUINT && label == 2) 1322 break; 1323 1324 uint64_t sector = vdev_label_start(adjpl, 1325 label) >> SPA_MINBLOCKSHIFT; 1326 1327 /* Read in the uberblock ring (128K). */ 1328 if (devread(sector + 1329 ((VDEV_SKIP_SIZE + VDEV_PHYS_SIZE) >> SPA_MINBLOCKSHIFT), 1330 0, VDEV_UBERBLOCK_RING, ub_array) == 0) 1331 continue; 1332 1333 if (check_pool_label(sector, stack, tmp_devid, 1334 tmp_bootpath, &tmp_guid, &ashift)) 1335 continue; 1336 1337 if (pool_guid == 0) 1338 pool_guid = tmp_guid; 1339 1340 if ((ubbest = find_bestub(ub_array, ashift, sector)) == NULL || 1341 zio_read(&ubbest->ub_rootbp, osp, stack) != 0) 1342 continue; 1343 1344 VERIFY_OS_TYPE(osp, DMU_OST_META); 1345 1346 if (find_best_root && ((pool_guid != tmp_guid) || 1347 vdev_uberblock_compare(ubbest, &(current_uberblock)) <= 0)) 1348 continue; 1349 1350 /* Got the MOS. Save it at the memory addr MOS. */ 1351 grub_memmove(MOS, &osp->os_meta_dnode, DNODE_SIZE); 1352 grub_memmove(¤t_uberblock, ubbest, sizeof (uberblock_t)); 1353 grub_memmove(current_bootpath, tmp_bootpath, MAXNAMELEN); 1354 grub_memmove(current_devid, tmp_devid, grub_strlen(tmp_devid)); 1355 is_zfs_mount = 1; 1356 return (1); 1357 } 1358 1359 /* 1360 * While some fs impls. (tftp) rely on setting and keeping 1361 * global errnums set, others won't reset it and will break 1362 * when issuing rawreads. The goal here is to simply not 1363 * have zfs mount attempts impact the previous state. 1364 */ 1365 errnum = err; 1366 return (0); 1367 } 1368 1369 /* 1370 * zfs_open() locates a file in the rootpool by following the 1371 * MOS and places the dnode of the file in the memory address DNODE. 1372 * 1373 * Return: 1374 * 1 - success 1375 * 0 - failure 1376 */ 1377 int 1378 zfs_open(char *filename) 1379 { 1380 char *stack; 1381 dnode_phys_t *mdn; 1382 1383 file_buf = NULL; 1384 stackbase = ZFS_SCRATCH; 1385 stack = stackbase; 1386 1387 mdn = (dnode_phys_t *)stack; 1388 stack += sizeof (dnode_phys_t); 1389 1390 dnode_mdn = NULL; 1391 dnode_buf = (dnode_phys_t *)stack; 1392 stack += 1<<DNODE_BLOCK_SHIFT; 1393 1394 /* 1395 * menu.lst is placed at the root pool filesystem level, 1396 * do not goto 'current_bootfs'. 1397 */ 1398 if (is_top_dataset_file(filename)) { 1399 if (errnum = get_objset_mdn(MOS, NULL, NULL, mdn, stack)) 1400 return (0); 1401 1402 current_bootfs_obj = 0; 1403 } else { 1404 if (current_bootfs[0] == '\0') { 1405 /* Get the default root filesystem object number */ 1406 if (errnum = get_default_bootfsobj(MOS, 1407 ¤t_bootfs_obj, stack)) 1408 return (0); 1409 1410 if (errnum = get_objset_mdn(MOS, NULL, 1411 ¤t_bootfs_obj, mdn, stack)) 1412 return (0); 1413 } else { 1414 if (errnum = get_objset_mdn(MOS, current_bootfs, 1415 ¤t_bootfs_obj, mdn, stack)) { 1416 grub_memset(current_bootfs, 0, MAXNAMELEN); 1417 return (0); 1418 } 1419 } 1420 } 1421 1422 if (dnode_get_path(mdn, filename, DNODE, stack)) { 1423 errnum = ERR_FILE_NOT_FOUND; 1424 return (0); 1425 } 1426 1427 /* get the file size and set the file position to 0 */ 1428 1429 /* 1430 * For DMU_OT_SA we will need to locate the SIZE attribute 1431 * attribute, which could be either in the bonus buffer 1432 * or the "spill" block. 1433 */ 1434 if (DNODE->dn_bonustype == DMU_OT_SA) { 1435 sa_hdr_phys_t *sahdrp; 1436 int hdrsize; 1437 1438 if (DNODE->dn_bonuslen != 0) { 1439 sahdrp = (sa_hdr_phys_t *)DN_BONUS(DNODE); 1440 } else { 1441 if (DNODE->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { 1442 blkptr_t *bp = &DNODE->dn_spill; 1443 void *buf; 1444 1445 buf = (void *)stack; 1446 stack += BP_GET_LSIZE(bp); 1447 1448 /* reset errnum to rawread() failure */ 1449 errnum = 0; 1450 if (zio_read(bp, buf, stack) != 0) { 1451 return (0); 1452 } 1453 sahdrp = buf; 1454 } else { 1455 errnum = ERR_FSYS_CORRUPT; 1456 return (0); 1457 } 1458 } 1459 hdrsize = SA_HDR_SIZE(sahdrp); 1460 filemax = *(uint64_t *)((char *)sahdrp + hdrsize + 1461 SA_SIZE_OFFSET); 1462 } else { 1463 filemax = ((znode_phys_t *)DN_BONUS(DNODE))->zp_size; 1464 } 1465 filepos = 0; 1466 1467 dnode_buf = NULL; 1468 return (1); 1469 } 1470 1471 /* 1472 * zfs_read reads in the data blocks pointed by the DNODE. 1473 * 1474 * Return: 1475 * len - the length successfully read in to the buffer 1476 * 0 - failure 1477 */ 1478 int 1479 zfs_read(char *buf, int len) 1480 { 1481 char *stack; 1482 char *tmpbuf; 1483 int blksz, length, movesize; 1484 1485 if (file_buf == NULL) { 1486 file_buf = stackbase; 1487 stackbase += SPA_MAXBLOCKSIZE; 1488 file_start = file_end = 0; 1489 } 1490 stack = stackbase; 1491 1492 /* 1493 * If offset is in memory, move it into the buffer provided and return. 1494 */ 1495 if (filepos >= file_start && filepos+len <= file_end) { 1496 grub_memmove(buf, file_buf + filepos - file_start, len); 1497 filepos += len; 1498 return (len); 1499 } 1500 1501 blksz = DNODE->dn_datablkszsec << SPA_MINBLOCKSHIFT; 1502 1503 /* 1504 * Entire Dnode is too big to fit into the space available. We 1505 * will need to read it in chunks. This could be optimized to 1506 * read in as large a chunk as there is space available, but for 1507 * now, this only reads in one data block at a time. 1508 */ 1509 length = len; 1510 while (length) { 1511 /* 1512 * Find requested blkid and the offset within that block. 1513 */ 1514 uint64_t blkid = filepos / blksz; 1515 1516 if (errnum = dmu_read(DNODE, blkid, file_buf, stack)) 1517 return (0); 1518 1519 file_start = blkid * blksz; 1520 file_end = file_start + blksz; 1521 1522 movesize = MIN(length, file_end - filepos); 1523 1524 grub_memmove(buf, file_buf + filepos - file_start, 1525 movesize); 1526 buf += movesize; 1527 length -= movesize; 1528 filepos += movesize; 1529 } 1530 1531 return (len); 1532 } 1533 1534 /* 1535 * No-Op 1536 */ 1537 int 1538 zfs_embed(int *start_sector, int needed_sectors) 1539 { 1540 return (1); 1541 } 1542 1543 #endif /* FSYS_ZFS */ 1544