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 2009 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 uberblock_t current_uberblock; 63 static char *stackbase; 64 65 decomp_entry_t decomp_table[ZIO_COMPRESS_FUNCTIONS] = 66 { 67 {"inherit", 0}, /* ZIO_COMPRESS_INHERIT */ 68 {"on", lzjb_decompress}, /* ZIO_COMPRESS_ON */ 69 {"off", 0}, /* ZIO_COMPRESS_OFF */ 70 {"lzjb", lzjb_decompress}, /* ZIO_COMPRESS_LZJB */ 71 {"empty", 0} /* ZIO_COMPRESS_EMPTY */ 72 }; 73 74 static int zio_read_data(blkptr_t *bp, void *buf, char *stack); 75 76 /* 77 * Our own version of bcmp(). 78 */ 79 static int 80 zfs_bcmp(const void *s1, const void *s2, size_t n) 81 { 82 const uchar_t *ps1 = s1; 83 const uchar_t *ps2 = s2; 84 85 if (s1 != s2 && n != 0) { 86 do { 87 if (*ps1++ != *ps2++) 88 return (1); 89 } while (--n != 0); 90 } 91 92 return (0); 93 } 94 95 /* 96 * Our own version of log2(). Same thing as highbit()-1. 97 */ 98 static int 99 zfs_log2(uint64_t num) 100 { 101 int i = 0; 102 103 while (num > 1) { 104 i++; 105 num = num >> 1; 106 } 107 108 return (i); 109 } 110 111 /* Checksum Functions */ 112 static void 113 zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp) 114 { 115 ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0); 116 } 117 118 /* Checksum Table and Values */ 119 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = { 120 NULL, NULL, 0, 0, "inherit", 121 NULL, NULL, 0, 0, "on", 122 zio_checksum_off, zio_checksum_off, 0, 0, "off", 123 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "label", 124 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "gang_header", 125 fletcher_2_native, fletcher_2_byteswap, 0, 1, "zilog", 126 fletcher_2_native, fletcher_2_byteswap, 0, 0, "fletcher2", 127 fletcher_4_native, fletcher_4_byteswap, 1, 0, "fletcher4", 128 zio_checksum_SHA256, zio_checksum_SHA256, 1, 0, "SHA256", 129 }; 130 131 /* 132 * zio_checksum_verify: Provides support for checksum verification. 133 * 134 * Fletcher2, Fletcher4, and SHA256 are supported. 135 * 136 * Return: 137 * -1 = Failure 138 * 0 = Success 139 */ 140 static int 141 zio_checksum_verify(blkptr_t *bp, char *data, int size) 142 { 143 zio_cksum_t zc = bp->blk_cksum; 144 uint32_t checksum = BP_GET_CHECKSUM(bp); 145 int byteswap = BP_SHOULD_BYTESWAP(bp); 146 zio_block_tail_t *zbt = (zio_block_tail_t *)(data + size) - 1; 147 zio_checksum_info_t *ci = &zio_checksum_table[checksum]; 148 zio_cksum_t actual_cksum, expected_cksum; 149 150 /* byteswap is not supported */ 151 if (byteswap) 152 return (-1); 153 154 if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL) 155 return (-1); 156 157 if (ci->ci_zbt) { 158 expected_cksum = zbt->zbt_cksum; 159 zbt->zbt_cksum = zc; 160 ci->ci_func[0](data, size, &actual_cksum); 161 zbt->zbt_cksum = expected_cksum; 162 zc = expected_cksum; 163 164 } else { 165 ci->ci_func[byteswap](data, size, &actual_cksum); 166 } 167 168 if ((actual_cksum.zc_word[0] - zc.zc_word[0]) | 169 (actual_cksum.zc_word[1] - zc.zc_word[1]) | 170 (actual_cksum.zc_word[2] - zc.zc_word[2]) | 171 (actual_cksum.zc_word[3] - zc.zc_word[3])) 172 return (-1); 173 174 return (0); 175 } 176 177 /* 178 * vdev_label_offset takes "offset" (the offset within a vdev_label) and 179 * returns its physical disk offset (starting from the beginning of the vdev). 180 * 181 * Input: 182 * psize : Physical size of this vdev 183 * l : Label Number (0-3) 184 * offset : The offset with a vdev_label in which we want the physical 185 * address 186 * Return: 187 * Success : physical disk offset 188 * Failure : errnum = ERR_BAD_ARGUMENT, return value is meaningless 189 */ 190 static uint64_t 191 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 192 { 193 /* XXX Need to add back label support! */ 194 if (l >= VDEV_LABELS/2 || offset > sizeof (vdev_label_t)) { 195 errnum = ERR_BAD_ARGUMENT; 196 return (0); 197 } 198 199 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 200 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 201 202 } 203 204 /* 205 * vdev_uberblock_compare takes two uberblock structures and returns an integer 206 * indicating the more recent of the two. 207 * Return Value = 1 if ub2 is more recent 208 * Return Value = -1 if ub1 is more recent 209 * The most recent uberblock is determined using its transaction number and 210 * timestamp. The uberblock with the highest transaction number is 211 * considered "newer". If the transaction numbers of the two blocks match, the 212 * timestamps are compared to determine the "newer" of the two. 213 */ 214 static int 215 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 216 { 217 if (ub1->ub_txg < ub2->ub_txg) 218 return (-1); 219 if (ub1->ub_txg > ub2->ub_txg) 220 return (1); 221 222 if (ub1->ub_timestamp < ub2->ub_timestamp) 223 return (-1); 224 if (ub1->ub_timestamp > ub2->ub_timestamp) 225 return (1); 226 227 return (0); 228 } 229 230 /* 231 * Three pieces of information are needed to verify an uberblock: the magic 232 * number, the version number, and the checksum. 233 * 234 * Currently Implemented: version number, magic number 235 * Need to Implement: checksum 236 * 237 * Return: 238 * 0 - Success 239 * -1 - Failure 240 */ 241 static int 242 uberblock_verify(uberblock_phys_t *ub, int offset) 243 { 244 245 uberblock_t *uber = &ub->ubp_uberblock; 246 blkptr_t bp; 247 248 BP_ZERO(&bp); 249 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL); 250 BP_SET_BYTEORDER(&bp, ZFS_HOST_BYTEORDER); 251 ZIO_SET_CHECKSUM(&bp.blk_cksum, offset, 0, 0, 0); 252 253 if (zio_checksum_verify(&bp, (char *)ub, UBERBLOCK_SIZE) != 0) 254 return (-1); 255 256 if (uber->ub_magic == UBERBLOCK_MAGIC && 257 uber->ub_version > 0 && uber->ub_version <= SPA_VERSION) 258 return (0); 259 260 return (-1); 261 } 262 263 /* 264 * Find the best uberblock. 265 * Return: 266 * Success - Pointer to the best uberblock. 267 * Failure - NULL 268 */ 269 static uberblock_phys_t * 270 find_bestub(uberblock_phys_t *ub_array, int label) 271 { 272 uberblock_phys_t *ubbest = NULL; 273 int i, offset; 274 275 for (i = 0; i < (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT); i++) { 276 offset = vdev_label_offset(0, label, VDEV_UBERBLOCK_OFFSET(i)); 277 if (errnum == ERR_BAD_ARGUMENT) 278 return (NULL); 279 if (uberblock_verify(&ub_array[i], offset) == 0) { 280 if (ubbest == NULL) { 281 ubbest = &ub_array[i]; 282 } else if (vdev_uberblock_compare( 283 &(ub_array[i].ubp_uberblock), 284 &(ubbest->ubp_uberblock)) > 0) { 285 ubbest = &ub_array[i]; 286 } 287 } 288 } 289 290 return (ubbest); 291 } 292 293 /* 294 * Read a block of data based on the gang block address dva, 295 * and put its data in buf. 296 * 297 * Return: 298 * 0 - success 299 * 1 - failure 300 */ 301 static int 302 zio_read_gang(blkptr_t *bp, dva_t *dva, void *buf, char *stack) 303 { 304 zio_gbh_phys_t *zio_gb; 305 uint64_t offset, sector; 306 blkptr_t tmpbp; 307 int i; 308 309 zio_gb = (zio_gbh_phys_t *)stack; 310 stack += SPA_GANGBLOCKSIZE; 311 offset = DVA_GET_OFFSET(dva); 312 sector = DVA_OFFSET_TO_PHYS_SECTOR(offset); 313 314 /* read in the gang block header */ 315 if (devread(sector, 0, SPA_GANGBLOCKSIZE, (char *)zio_gb) == 0) { 316 grub_printf("failed to read in a gang block header\n"); 317 return (1); 318 } 319 320 /* self checksuming the gang block header */ 321 BP_ZERO(&tmpbp); 322 BP_SET_CHECKSUM(&tmpbp, ZIO_CHECKSUM_GANG_HEADER); 323 BP_SET_BYTEORDER(&tmpbp, ZFS_HOST_BYTEORDER); 324 ZIO_SET_CHECKSUM(&tmpbp.blk_cksum, DVA_GET_VDEV(dva), 325 DVA_GET_OFFSET(dva), bp->blk_birth, 0); 326 if (zio_checksum_verify(&tmpbp, (char *)zio_gb, SPA_GANGBLOCKSIZE)) { 327 grub_printf("failed to checksum a gang block header\n"); 328 return (1); 329 } 330 331 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) { 332 if (zio_gb->zg_blkptr[i].blk_birth == 0) 333 continue; 334 335 if (zio_read_data(&zio_gb->zg_blkptr[i], buf, stack)) 336 return (1); 337 buf += BP_GET_PSIZE(&zio_gb->zg_blkptr[i]); 338 } 339 340 return (0); 341 } 342 343 /* 344 * Read in a block of raw data to buf. 345 * 346 * Return: 347 * 0 - success 348 * 1 - failure 349 */ 350 static int 351 zio_read_data(blkptr_t *bp, void *buf, char *stack) 352 { 353 int i, psize; 354 355 psize = BP_GET_PSIZE(bp); 356 357 /* pick a good dva from the block pointer */ 358 for (i = 0; i < SPA_DVAS_PER_BP; i++) { 359 uint64_t offset, sector; 360 361 if (bp->blk_dva[i].dva_word[0] == 0 && 362 bp->blk_dva[i].dva_word[1] == 0) 363 continue; 364 365 if (DVA_GET_GANG(&bp->blk_dva[i])) { 366 if (zio_read_gang(bp, &bp->blk_dva[i], buf, stack) == 0) 367 return (0); 368 } else { 369 /* read in a data block */ 370 offset = DVA_GET_OFFSET(&bp->blk_dva[i]); 371 sector = DVA_OFFSET_TO_PHYS_SECTOR(offset); 372 if (devread(sector, 0, psize, buf)) 373 return (0); 374 } 375 } 376 377 return (1); 378 } 379 380 /* 381 * Read in a block of data, verify its checksum, decompress if needed, 382 * and put the uncompressed data in buf. 383 * 384 * Return: 385 * 0 - success 386 * errnum - failure 387 */ 388 static int 389 zio_read(blkptr_t *bp, void *buf, char *stack) 390 { 391 int lsize, psize, comp; 392 char *retbuf; 393 394 comp = BP_GET_COMPRESS(bp); 395 lsize = BP_GET_LSIZE(bp); 396 psize = BP_GET_PSIZE(bp); 397 398 if ((unsigned int)comp >= ZIO_COMPRESS_FUNCTIONS || 399 (comp != ZIO_COMPRESS_OFF && 400 decomp_table[comp].decomp_func == NULL)) { 401 grub_printf("compression algorithm not supported\n"); 402 return (ERR_FSYS_CORRUPT); 403 } 404 405 if ((char *)buf < stack && ((char *)buf) + lsize > stack) { 406 grub_printf("not enough memory allocated\n"); 407 return (ERR_WONT_FIT); 408 } 409 410 retbuf = buf; 411 if (comp != ZIO_COMPRESS_OFF) { 412 buf = stack; 413 stack += psize; 414 } 415 416 if (zio_read_data(bp, buf, stack)) { 417 grub_printf("zio_read_data failed\n"); 418 return (ERR_FSYS_CORRUPT); 419 } 420 421 if (zio_checksum_verify(bp, buf, psize) != 0) { 422 grub_printf("checksum verification failed\n"); 423 return (ERR_FSYS_CORRUPT); 424 } 425 426 if (comp != ZIO_COMPRESS_OFF) 427 decomp_table[comp].decomp_func(buf, retbuf, psize, lsize); 428 429 return (0); 430 } 431 432 /* 433 * Get the block from a block id. 434 * push the block onto the stack. 435 * 436 * Return: 437 * 0 - success 438 * errnum - failure 439 */ 440 static int 441 dmu_read(dnode_phys_t *dn, uint64_t blkid, void *buf, char *stack) 442 { 443 int idx, level; 444 blkptr_t *bp_array = dn->dn_blkptr; 445 int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 446 blkptr_t *bp, *tmpbuf; 447 448 bp = (blkptr_t *)stack; 449 stack += sizeof (blkptr_t); 450 451 tmpbuf = (blkptr_t *)stack; 452 stack += 1<<dn->dn_indblkshift; 453 454 for (level = dn->dn_nlevels - 1; level >= 0; level--) { 455 idx = (blkid >> (epbs * level)) & ((1<<epbs)-1); 456 *bp = bp_array[idx]; 457 if (level == 0) 458 tmpbuf = buf; 459 if (BP_IS_HOLE(bp)) { 460 grub_memset(buf, 0, 461 dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); 462 break; 463 } else if (errnum = zio_read(bp, tmpbuf, stack)) { 464 return (errnum); 465 } 466 467 bp_array = tmpbuf; 468 } 469 470 return (0); 471 } 472 473 /* 474 * mzap_lookup: Looks up property described by "name" and returns the value 475 * in "value". 476 * 477 * Return: 478 * 0 - success 479 * errnum - failure 480 */ 481 static int 482 mzap_lookup(mzap_phys_t *zapobj, int objsize, char *name, 483 uint64_t *value) 484 { 485 int i, chunks; 486 mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk; 487 488 chunks = objsize/MZAP_ENT_LEN - 1; 489 for (i = 0; i < chunks; i++) { 490 if (grub_strcmp(mzap_ent[i].mze_name, name) == 0) { 491 *value = mzap_ent[i].mze_value; 492 return (0); 493 } 494 } 495 496 return (ERR_FSYS_CORRUPT); 497 } 498 499 static uint64_t 500 zap_hash(uint64_t salt, const char *name) 501 { 502 static uint64_t table[256]; 503 const uint8_t *cp; 504 uint8_t c; 505 uint64_t crc = salt; 506 507 if (table[128] == 0) { 508 uint64_t *ct; 509 int i, j; 510 for (i = 0; i < 256; i++) { 511 for (ct = table + i, *ct = i, j = 8; j > 0; j--) 512 *ct = (*ct >> 1) ^ (-(*ct & 1) & 513 ZFS_CRC64_POLY); 514 } 515 } 516 517 if (crc == 0 || table[128] != ZFS_CRC64_POLY) { 518 errnum = ERR_FSYS_CORRUPT; 519 return (0); 520 } 521 522 for (cp = (const uint8_t *)name; (c = *cp) != '\0'; cp++) 523 crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF]; 524 525 /* 526 * Only use 28 bits, since we need 4 bits in the cookie for the 527 * collision differentiator. We MUST use the high bits, since 528 * those are the onces that we first pay attention to when 529 * chosing the bucket. 530 */ 531 crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1); 532 533 return (crc); 534 } 535 536 /* 537 * Only to be used on 8-bit arrays. 538 * array_len is actual len in bytes (not encoded le_value_length). 539 * buf is null-terminated. 540 */ 541 static int 542 zap_leaf_array_equal(zap_leaf_phys_t *l, int blksft, int chunk, 543 int array_len, const char *buf) 544 { 545 int bseen = 0; 546 547 while (bseen < array_len) { 548 struct zap_leaf_array *la = 549 &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array; 550 int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES); 551 552 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 553 return (0); 554 555 if (zfs_bcmp(la->la_array, buf + bseen, toread) != 0) 556 break; 557 chunk = la->la_next; 558 bseen += toread; 559 } 560 return (bseen == array_len); 561 } 562 563 /* 564 * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the 565 * value for the property "name". 566 * 567 * Return: 568 * 0 - success 569 * errnum - failure 570 */ 571 static int 572 zap_leaf_lookup(zap_leaf_phys_t *l, int blksft, uint64_t h, 573 const char *name, uint64_t *value) 574 { 575 uint16_t chunk; 576 struct zap_leaf_entry *le; 577 578 /* Verify if this is a valid leaf block */ 579 if (l->l_hdr.lh_block_type != ZBT_LEAF) 580 return (ERR_FSYS_CORRUPT); 581 if (l->l_hdr.lh_magic != ZAP_LEAF_MAGIC) 582 return (ERR_FSYS_CORRUPT); 583 584 for (chunk = l->l_hash[LEAF_HASH(blksft, h)]; 585 chunk != CHAIN_END; chunk = le->le_next) { 586 587 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 588 return (ERR_FSYS_CORRUPT); 589 590 le = ZAP_LEAF_ENTRY(l, blksft, chunk); 591 592 /* Verify the chunk entry */ 593 if (le->le_type != ZAP_CHUNK_ENTRY) 594 return (ERR_FSYS_CORRUPT); 595 596 if (le->le_hash != h) 597 continue; 598 599 if (zap_leaf_array_equal(l, blksft, le->le_name_chunk, 600 le->le_name_length, name)) { 601 602 struct zap_leaf_array *la; 603 uint8_t *ip; 604 605 if (le->le_int_size != 8 || le->le_value_length != 1) 606 return (ERR_FSYS_CORRUPT); 607 608 /* get the uint64_t property value */ 609 la = &ZAP_LEAF_CHUNK(l, blksft, 610 le->le_value_chunk).l_array; 611 ip = la->la_array; 612 613 *value = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 | 614 (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 | 615 (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 | 616 (uint64_t)ip[6] << 8 | (uint64_t)ip[7]; 617 618 return (0); 619 } 620 } 621 622 return (ERR_FSYS_CORRUPT); 623 } 624 625 /* 626 * Fat ZAP lookup 627 * 628 * Return: 629 * 0 - success 630 * errnum - failure 631 */ 632 static int 633 fzap_lookup(dnode_phys_t *zap_dnode, zap_phys_t *zap, 634 char *name, uint64_t *value, char *stack) 635 { 636 zap_leaf_phys_t *l; 637 uint64_t hash, idx, blkid; 638 int blksft = zfs_log2(zap_dnode->dn_datablkszsec << DNODE_SHIFT); 639 640 /* Verify if this is a fat zap header block */ 641 if (zap->zap_magic != (uint64_t)ZAP_MAGIC) 642 return (ERR_FSYS_CORRUPT); 643 644 hash = zap_hash(zap->zap_salt, name); 645 if (errnum) 646 return (errnum); 647 648 /* get block id from index */ 649 if (zap->zap_ptrtbl.zt_numblks != 0) { 650 /* external pointer tables not supported */ 651 return (ERR_FSYS_CORRUPT); 652 } 653 idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift); 654 blkid = ((uint64_t *)zap)[idx + (1<<(blksft-3-1))]; 655 656 /* Get the leaf block */ 657 l = (zap_leaf_phys_t *)stack; 658 stack += 1<<blksft; 659 if ((1<<blksft) < sizeof (zap_leaf_phys_t)) 660 return (ERR_FSYS_CORRUPT); 661 if (errnum = dmu_read(zap_dnode, blkid, l, stack)) 662 return (errnum); 663 664 return (zap_leaf_lookup(l, blksft, hash, name, value)); 665 } 666 667 /* 668 * Read in the data of a zap object and find the value for a matching 669 * property name. 670 * 671 * Return: 672 * 0 - success 673 * errnum - failure 674 */ 675 static int 676 zap_lookup(dnode_phys_t *zap_dnode, char *name, uint64_t *val, char *stack) 677 { 678 uint64_t block_type; 679 int size; 680 void *zapbuf; 681 682 /* Read in the first block of the zap object data. */ 683 zapbuf = stack; 684 size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 685 stack += size; 686 if (errnum = dmu_read(zap_dnode, 0, zapbuf, stack)) 687 return (errnum); 688 689 block_type = *((uint64_t *)zapbuf); 690 691 if (block_type == ZBT_MICRO) { 692 return (mzap_lookup(zapbuf, size, name, val)); 693 } else if (block_type == ZBT_HEADER) { 694 /* this is a fat zap */ 695 return (fzap_lookup(zap_dnode, zapbuf, name, 696 val, stack)); 697 } 698 699 return (ERR_FSYS_CORRUPT); 700 } 701 702 /* 703 * Get the dnode of an object number from the metadnode of an object set. 704 * 705 * Input 706 * mdn - metadnode to get the object dnode 707 * objnum - object number for the object dnode 708 * buf - data buffer that holds the returning dnode 709 * stack - scratch area 710 * 711 * Return: 712 * 0 - success 713 * errnum - failure 714 */ 715 static int 716 dnode_get(dnode_phys_t *mdn, uint64_t objnum, uint8_t type, dnode_phys_t *buf, 717 char *stack) 718 { 719 uint64_t blkid, blksz; /* the block id this object dnode is in */ 720 int epbs; /* shift of number of dnodes in a block */ 721 int idx; /* index within a block */ 722 dnode_phys_t *dnbuf; 723 724 blksz = mdn->dn_datablkszsec << SPA_MINBLOCKSHIFT; 725 epbs = zfs_log2(blksz) - DNODE_SHIFT; 726 blkid = objnum >> epbs; 727 idx = objnum & ((1<<epbs)-1); 728 729 if (dnode_buf != NULL && dnode_mdn == mdn && 730 objnum >= dnode_start && objnum < dnode_end) { 731 grub_memmove(buf, &dnode_buf[idx], DNODE_SIZE); 732 VERIFY_DN_TYPE(buf, type); 733 return (0); 734 } 735 736 if (dnode_buf && blksz == 1<<DNODE_BLOCK_SHIFT) { 737 dnbuf = dnode_buf; 738 dnode_mdn = mdn; 739 dnode_start = blkid << epbs; 740 dnode_end = (blkid + 1) << epbs; 741 } else { 742 dnbuf = (dnode_phys_t *)stack; 743 stack += blksz; 744 } 745 746 if (errnum = dmu_read(mdn, blkid, (char *)dnbuf, stack)) 747 return (errnum); 748 749 grub_memmove(buf, &dnbuf[idx], DNODE_SIZE); 750 VERIFY_DN_TYPE(buf, type); 751 752 return (0); 753 } 754 755 /* 756 * Check if this is a special file that resides at the top 757 * dataset of the pool. Currently this is the GRUB menu, 758 * boot signature and boot signature backup. 759 * str starts with '/'. 760 */ 761 static int 762 is_top_dataset_file(char *str) 763 { 764 char *tptr; 765 766 if ((tptr = grub_strstr(str, "menu.lst")) && 767 (tptr[8] == '\0' || tptr[8] == ' ') && 768 *(tptr-1) == '/') 769 return (1); 770 771 if (grub_strncmp(str, BOOTSIGN_DIR"/", 772 grub_strlen(BOOTSIGN_DIR) + 1) == 0) 773 return (1); 774 775 if (grub_strcmp(str, BOOTSIGN_BACKUP) == 0) 776 return (1); 777 778 return (0); 779 } 780 781 /* 782 * Get the file dnode for a given file name where mdn is the meta dnode 783 * for this ZFS object set. When found, place the file dnode in dn. 784 * The 'path' argument will be mangled. 785 * 786 * Return: 787 * 0 - success 788 * errnum - failure 789 */ 790 static int 791 dnode_get_path(dnode_phys_t *mdn, char *path, dnode_phys_t *dn, 792 char *stack) 793 { 794 uint64_t objnum, version; 795 char *cname, ch; 796 797 if (errnum = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE, 798 dn, stack)) 799 return (errnum); 800 801 if (errnum = zap_lookup(dn, ZPL_VERSION_STR, &version, stack)) 802 return (errnum); 803 if (version > ZPL_VERSION) 804 return (-1); 805 806 if (errnum = zap_lookup(dn, ZFS_ROOT_OBJ, &objnum, stack)) 807 return (errnum); 808 809 if (errnum = dnode_get(mdn, objnum, DMU_OT_DIRECTORY_CONTENTS, 810 dn, stack)) 811 return (errnum); 812 813 /* skip leading slashes */ 814 while (*path == '/') 815 path++; 816 817 while (*path && !isspace(*path)) { 818 819 /* get the next component name */ 820 cname = path; 821 while (*path && !isspace(*path) && *path != '/') 822 path++; 823 ch = *path; 824 *path = 0; /* ensure null termination */ 825 826 if (errnum = zap_lookup(dn, cname, &objnum, stack)) 827 return (errnum); 828 829 objnum = ZFS_DIRENT_OBJ(objnum); 830 if (errnum = dnode_get(mdn, objnum, 0, dn, stack)) 831 return (errnum); 832 833 *path = ch; 834 while (*path == '/') 835 path++; 836 } 837 838 /* We found the dnode for this file. Verify if it is a plain file. */ 839 VERIFY_DN_TYPE(dn, DMU_OT_PLAIN_FILE_CONTENTS); 840 841 return (0); 842 } 843 844 /* 845 * Get the default 'bootfs' property value from the rootpool. 846 * 847 * Return: 848 * 0 - success 849 * errnum -failure 850 */ 851 static int 852 get_default_bootfsobj(dnode_phys_t *mosmdn, uint64_t *obj, char *stack) 853 { 854 uint64_t objnum = 0; 855 dnode_phys_t *dn = (dnode_phys_t *)stack; 856 stack += DNODE_SIZE; 857 858 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 859 DMU_OT_OBJECT_DIRECTORY, dn, stack)) 860 return (errnum); 861 862 /* 863 * find the object number for 'pool_props', and get the dnode 864 * of the 'pool_props'. 865 */ 866 if (zap_lookup(dn, DMU_POOL_PROPS, &objnum, stack)) 867 return (ERR_FILESYSTEM_NOT_FOUND); 868 869 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_POOL_PROPS, dn, stack)) 870 return (errnum); 871 872 if (zap_lookup(dn, ZPOOL_PROP_BOOTFS, &objnum, stack)) 873 return (ERR_FILESYSTEM_NOT_FOUND); 874 875 if (!objnum) 876 return (ERR_FILESYSTEM_NOT_FOUND); 877 878 *obj = objnum; 879 return (0); 880 } 881 882 /* 883 * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname), 884 * e.g. pool/rootfs, or a given object number (obj), e.g. the object number 885 * of pool/rootfs. 886 * 887 * If no fsname and no obj are given, return the DSL_DIR metadnode. 888 * If fsname is given, return its metadnode and its matching object number. 889 * If only obj is given, return the metadnode for this object number. 890 * 891 * Return: 892 * 0 - success 893 * errnum - failure 894 */ 895 static int 896 get_objset_mdn(dnode_phys_t *mosmdn, char *fsname, uint64_t *obj, 897 dnode_phys_t *mdn, char *stack) 898 { 899 uint64_t objnum, headobj; 900 char *cname, ch; 901 blkptr_t *bp; 902 objset_phys_t *osp; 903 int issnapshot = 0; 904 char *snapname; 905 906 if (fsname == NULL && obj) { 907 headobj = *obj; 908 goto skip; 909 } 910 911 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 912 DMU_OT_OBJECT_DIRECTORY, mdn, stack)) 913 return (errnum); 914 915 if (errnum = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum, 916 stack)) 917 return (errnum); 918 919 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, stack)) 920 return (errnum); 921 922 if (fsname == NULL) { 923 headobj = 924 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 925 goto skip; 926 } 927 928 /* take out the pool name */ 929 while (*fsname && !isspace(*fsname) && *fsname != '/') 930 fsname++; 931 932 while (*fsname && !isspace(*fsname)) { 933 uint64_t childobj; 934 935 while (*fsname == '/') 936 fsname++; 937 938 cname = fsname; 939 while (*fsname && !isspace(*fsname) && *fsname != '/') 940 fsname++; 941 ch = *fsname; 942 *fsname = 0; 943 944 snapname = cname; 945 while (*snapname && !isspace(*snapname) && *snapname != '@') 946 snapname++; 947 if (*snapname == '@') { 948 issnapshot = 1; 949 *snapname = 0; 950 } 951 childobj = 952 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_child_dir_zapobj; 953 if (errnum = dnode_get(mosmdn, childobj, 954 DMU_OT_DSL_DIR_CHILD_MAP, mdn, stack)) 955 return (errnum); 956 957 if (zap_lookup(mdn, cname, &objnum, stack)) 958 return (ERR_FILESYSTEM_NOT_FOUND); 959 960 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, 961 mdn, stack)) 962 return (errnum); 963 964 *fsname = ch; 965 if (issnapshot) 966 *snapname = '@'; 967 } 968 headobj = ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 969 if (obj) 970 *obj = headobj; 971 972 skip: 973 if (errnum = dnode_get(mosmdn, headobj, DMU_OT_DSL_DATASET, mdn, stack)) 974 return (errnum); 975 if (issnapshot) { 976 uint64_t snapobj; 977 978 snapobj = ((dsl_dataset_phys_t *)DN_BONUS(mdn))-> 979 ds_snapnames_zapobj; 980 981 if (errnum = dnode_get(mosmdn, snapobj, 982 DMU_OT_DSL_DS_SNAP_MAP, mdn, stack)) 983 return (errnum); 984 if (zap_lookup(mdn, snapname + 1, &headobj, stack)) 985 return (ERR_FILESYSTEM_NOT_FOUND); 986 if (errnum = dnode_get(mosmdn, headobj, 987 DMU_OT_DSL_DATASET, mdn, stack)) 988 return (errnum); 989 if (obj) 990 *obj = headobj; 991 } 992 993 bp = &((dsl_dataset_phys_t *)DN_BONUS(mdn))->ds_bp; 994 osp = (objset_phys_t *)stack; 995 stack += sizeof (objset_phys_t); 996 if (errnum = zio_read(bp, osp, stack)) 997 return (errnum); 998 999 grub_memmove((char *)mdn, (char *)&osp->os_meta_dnode, DNODE_SIZE); 1000 1001 return (0); 1002 } 1003 1004 /* 1005 * For a given XDR packed nvlist, verify the first 4 bytes and move on. 1006 * 1007 * An XDR packed nvlist is encoded as (comments from nvs_xdr_create) : 1008 * 1009 * encoding method/host endian (4 bytes) 1010 * nvl_version (4 bytes) 1011 * nvl_nvflag (4 bytes) 1012 * encoded nvpairs: 1013 * encoded size of the nvpair (4 bytes) 1014 * decoded size of the nvpair (4 bytes) 1015 * name string size (4 bytes) 1016 * name string data (sizeof(NV_ALIGN4(string)) 1017 * data type (4 bytes) 1018 * # of elements in the nvpair (4 bytes) 1019 * data 1020 * 2 zero's for the last nvpair 1021 * (end of the entire list) (8 bytes) 1022 * 1023 * Return: 1024 * 0 - success 1025 * 1 - failure 1026 */ 1027 static int 1028 nvlist_unpack(char *nvlist, char **out) 1029 { 1030 /* Verify if the 1st and 2nd byte in the nvlist are valid. */ 1031 if (nvlist[0] != NV_ENCODE_XDR || nvlist[1] != HOST_ENDIAN) 1032 return (1); 1033 1034 nvlist += 4; 1035 *out = nvlist; 1036 return (0); 1037 } 1038 1039 static char * 1040 nvlist_array(char *nvlist, int index) 1041 { 1042 int i, encode_size; 1043 1044 for (i = 0; i < index; i++) { 1045 /* skip the header, nvl_version, and nvl_nvflag */ 1046 nvlist = nvlist + 4 * 2; 1047 1048 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) 1049 nvlist += encode_size; /* goto the next nvpair */ 1050 1051 nvlist = nvlist + 4 * 2; /* skip the ending 2 zeros - 8 bytes */ 1052 } 1053 1054 return (nvlist); 1055 } 1056 1057 static int 1058 nvlist_lookup_value(char *nvlist, char *name, void *val, int valtype, 1059 int *nelmp) 1060 { 1061 int name_len, type, slen, encode_size; 1062 char *nvpair, *nvp_name, *strval = val; 1063 uint64_t *intval = val; 1064 1065 /* skip the header, nvl_version, and nvl_nvflag */ 1066 nvlist = nvlist + 4 * 2; 1067 1068 /* 1069 * Loop thru the nvpair list 1070 * The XDR representation of an integer is in big-endian byte order. 1071 */ 1072 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) { 1073 1074 nvpair = nvlist + 4 * 2; /* skip the encode/decode size */ 1075 1076 name_len = BSWAP_32(*(uint32_t *)nvpair); 1077 nvpair += 4; 1078 1079 nvp_name = nvpair; 1080 nvpair = nvpair + ((name_len + 3) & ~3); /* align */ 1081 1082 type = BSWAP_32(*(uint32_t *)nvpair); 1083 nvpair += 4; 1084 1085 if ((grub_strncmp(nvp_name, name, name_len) == 0) && 1086 type == valtype) { 1087 int nelm; 1088 1089 if ((nelm = BSWAP_32(*(uint32_t *)nvpair)) < 1) 1090 return (1); 1091 nvpair += 4; 1092 1093 switch (valtype) { 1094 case DATA_TYPE_STRING: 1095 slen = BSWAP_32(*(uint32_t *)nvpair); 1096 nvpair += 4; 1097 grub_memmove(strval, nvpair, slen); 1098 strval[slen] = '\0'; 1099 return (0); 1100 1101 case DATA_TYPE_UINT64: 1102 *intval = BSWAP_64(*(uint64_t *)nvpair); 1103 return (0); 1104 1105 case DATA_TYPE_NVLIST: 1106 *(void **)val = (void *)nvpair; 1107 return (0); 1108 1109 case DATA_TYPE_NVLIST_ARRAY: 1110 *(void **)val = (void *)nvpair; 1111 if (nelmp) 1112 *nelmp = nelm; 1113 return (0); 1114 } 1115 } 1116 1117 nvlist += encode_size; /* goto the next nvpair */ 1118 } 1119 1120 return (1); 1121 } 1122 1123 /* 1124 * Check if this vdev is online and is in a good state. 1125 */ 1126 static int 1127 vdev_validate(char *nv) 1128 { 1129 uint64_t ival; 1130 1131 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_OFFLINE, &ival, 1132 DATA_TYPE_UINT64, NULL) == 0 || 1133 nvlist_lookup_value(nv, ZPOOL_CONFIG_FAULTED, &ival, 1134 DATA_TYPE_UINT64, NULL) == 0 || 1135 nvlist_lookup_value(nv, ZPOOL_CONFIG_REMOVED, &ival, 1136 DATA_TYPE_UINT64, NULL) == 0) 1137 return (ERR_DEV_VALUES); 1138 1139 return (0); 1140 } 1141 1142 /* 1143 * Get a valid vdev pathname/devid from the boot device. 1144 * The caller should already allocate MAXPATHLEN memory for bootpath and devid. 1145 */ 1146 static int 1147 vdev_get_bootpath(char *nv, uint64_t inguid, char *devid, char *bootpath, 1148 int is_spare) 1149 { 1150 char type[16]; 1151 1152 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_TYPE, &type, DATA_TYPE_STRING, 1153 NULL)) 1154 return (ERR_FSYS_CORRUPT); 1155 1156 if (strcmp(type, VDEV_TYPE_DISK) == 0) { 1157 uint64_t guid; 1158 1159 if (vdev_validate(nv) != 0) 1160 return (ERR_NO_BOOTPATH); 1161 1162 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_GUID, 1163 &guid, DATA_TYPE_UINT64, NULL) != 0) 1164 return (ERR_NO_BOOTPATH); 1165 1166 if (guid != inguid) 1167 return (ERR_NO_BOOTPATH); 1168 1169 /* for a spare vdev, pick the disk labeled with "is_spare" */ 1170 if (is_spare) { 1171 uint64_t spare = 0; 1172 (void) nvlist_lookup_value(nv, ZPOOL_CONFIG_IS_SPARE, 1173 &spare, DATA_TYPE_UINT64, NULL); 1174 if (!spare) 1175 return (ERR_NO_BOOTPATH); 1176 } 1177 1178 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_PHYS_PATH, 1179 bootpath, DATA_TYPE_STRING, NULL) != 0) 1180 bootpath[0] = '\0'; 1181 1182 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_DEVID, 1183 devid, DATA_TYPE_STRING, NULL) != 0) 1184 devid[0] = '\0'; 1185 1186 if (strlen(bootpath) >= MAXPATHLEN || 1187 strlen(devid) >= MAXPATHLEN) 1188 return (ERR_WONT_FIT); 1189 1190 return (0); 1191 1192 } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 || 1193 strcmp(type, VDEV_TYPE_REPLACING) == 0 || 1194 (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) { 1195 int nelm, i; 1196 char *child; 1197 1198 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_CHILDREN, &child, 1199 DATA_TYPE_NVLIST_ARRAY, &nelm)) 1200 return (ERR_FSYS_CORRUPT); 1201 1202 for (i = 0; i < nelm; i++) { 1203 char *child_i; 1204 1205 child_i = nvlist_array(child, i); 1206 if (vdev_get_bootpath(child_i, inguid, devid, 1207 bootpath, is_spare) == 0) 1208 return (0); 1209 } 1210 } 1211 1212 return (ERR_NO_BOOTPATH); 1213 } 1214 1215 /* 1216 * Check the disk label information and retrieve needed vdev name-value pairs. 1217 * 1218 * Return: 1219 * 0 - success 1220 * ERR_* - failure 1221 */ 1222 int 1223 check_pool_label(int label, char *stack, char *outdevid, char *outpath) 1224 { 1225 vdev_phys_t *vdev; 1226 uint64_t sector, pool_state, txg = 0; 1227 char *nvlist, *nv; 1228 uint64_t diskguid; 1229 uint64_t version; 1230 1231 sector = (label * sizeof (vdev_label_t) + VDEV_SKIP_SIZE) 1232 >> SPA_MINBLOCKSHIFT; 1233 1234 /* Read in the vdev name-value pair list (112K). */ 1235 if (devread(sector, 0, VDEV_PHYS_SIZE, stack) == 0) 1236 return (ERR_READ); 1237 1238 vdev = (vdev_phys_t *)stack; 1239 stack += sizeof (vdev_phys_t); 1240 1241 if (nvlist_unpack(vdev->vp_nvlist, &nvlist)) 1242 return (ERR_FSYS_CORRUPT); 1243 1244 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_STATE, &pool_state, 1245 DATA_TYPE_UINT64, NULL)) 1246 return (ERR_FSYS_CORRUPT); 1247 1248 if (pool_state == POOL_STATE_DESTROYED) 1249 return (ERR_FILESYSTEM_NOT_FOUND); 1250 1251 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_NAME, 1252 current_rootpool, DATA_TYPE_STRING, NULL)) 1253 return (ERR_FSYS_CORRUPT); 1254 1255 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_TXG, &txg, 1256 DATA_TYPE_UINT64, NULL)) 1257 return (ERR_FSYS_CORRUPT); 1258 1259 /* not an active device */ 1260 if (txg == 0) 1261 return (ERR_NO_BOOTPATH); 1262 1263 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VERSION, &version, 1264 DATA_TYPE_UINT64, NULL)) 1265 return (ERR_FSYS_CORRUPT); 1266 if (version > SPA_VERSION) 1267 return (ERR_NEWER_VERSION); 1268 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VDEV_TREE, &nv, 1269 DATA_TYPE_NVLIST, NULL)) 1270 return (ERR_FSYS_CORRUPT); 1271 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_GUID, &diskguid, 1272 DATA_TYPE_UINT64, NULL)) 1273 return (ERR_FSYS_CORRUPT); 1274 if (vdev_get_bootpath(nv, diskguid, outdevid, outpath, 0)) 1275 return (ERR_NO_BOOTPATH); 1276 return (0); 1277 } 1278 1279 /* 1280 * zfs_mount() locates a valid uberblock of the root pool and read in its MOS 1281 * to the memory address MOS. 1282 * 1283 * Return: 1284 * 1 - success 1285 * 0 - failure 1286 */ 1287 int 1288 zfs_mount(void) 1289 { 1290 char *stack; 1291 int label = 0; 1292 uberblock_phys_t *ub_array, *ubbest = NULL; 1293 objset_phys_t *osp; 1294 char tmp_bootpath[MAXNAMELEN]; 1295 char tmp_devid[MAXNAMELEN]; 1296 1297 /* if it's our first time here, zero the best uberblock out */ 1298 if (best_drive == 0 && best_part == 0 && find_best_root) 1299 grub_memset(¤t_uberblock, 0, sizeof (uberblock_t)); 1300 1301 stackbase = ZFS_SCRATCH; 1302 stack = stackbase; 1303 ub_array = (uberblock_phys_t *)stack; 1304 stack += VDEV_UBERBLOCK_RING; 1305 1306 osp = (objset_phys_t *)stack; 1307 stack += sizeof (objset_phys_t); 1308 1309 /* XXX add back labels support? */ 1310 for (label = 0; ubbest == NULL && label < (VDEV_LABELS/2); label++) { 1311 uint64_t sector = (label * sizeof (vdev_label_t) + 1312 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE) >> SPA_MINBLOCKSHIFT; 1313 1314 /* Read in the uberblock ring (128K). */ 1315 if (devread(sector, 0, VDEV_UBERBLOCK_RING, 1316 (char *)ub_array) == 0) 1317 continue; 1318 1319 if ((ubbest = find_bestub(ub_array, label)) != NULL && 1320 zio_read(&ubbest->ubp_uberblock.ub_rootbp, osp, stack) 1321 == 0) { 1322 1323 VERIFY_OS_TYPE(osp, DMU_OST_META); 1324 1325 if (check_pool_label(label, stack, tmp_devid, 1326 tmp_bootpath)) 1327 return (0); 1328 1329 if (find_best_root && 1330 vdev_uberblock_compare(&ubbest->ubp_uberblock, 1331 &(current_uberblock)) <= 0) 1332 continue; 1333 1334 /* Got the MOS. Save it at the memory addr MOS. */ 1335 grub_memmove(MOS, &osp->os_meta_dnode, DNODE_SIZE); 1336 grub_memmove(¤t_uberblock, 1337 &ubbest->ubp_uberblock, sizeof (uberblock_t)); 1338 grub_memmove(current_bootpath, tmp_bootpath, 1339 MAXNAMELEN); 1340 grub_memmove(current_devid, tmp_devid, 1341 grub_strlen(tmp_devid)); 1342 is_zfs_mount = 1; 1343 return (1); 1344 } 1345 } 1346 1347 return (0); 1348 } 1349 1350 /* 1351 * zfs_open() locates a file in the rootpool by following the 1352 * MOS and places the dnode of the file in the memory address DNODE. 1353 * 1354 * Return: 1355 * 1 - success 1356 * 0 - failure 1357 */ 1358 int 1359 zfs_open(char *filename) 1360 { 1361 char *stack; 1362 dnode_phys_t *mdn; 1363 1364 file_buf = NULL; 1365 stackbase = ZFS_SCRATCH; 1366 stack = stackbase; 1367 1368 mdn = (dnode_phys_t *)stack; 1369 stack += sizeof (dnode_phys_t); 1370 1371 dnode_mdn = NULL; 1372 dnode_buf = (dnode_phys_t *)stack; 1373 stack += 1<<DNODE_BLOCK_SHIFT; 1374 1375 /* 1376 * menu.lst is placed at the root pool filesystem level, 1377 * do not goto 'current_bootfs'. 1378 */ 1379 if (is_top_dataset_file(filename)) { 1380 if (errnum = get_objset_mdn(MOS, NULL, NULL, mdn, stack)) 1381 return (0); 1382 1383 current_bootfs_obj = 0; 1384 } else { 1385 if (current_bootfs[0] == '\0') { 1386 /* Get the default root filesystem object number */ 1387 if (errnum = get_default_bootfsobj(MOS, 1388 ¤t_bootfs_obj, stack)) 1389 return (0); 1390 1391 if (errnum = get_objset_mdn(MOS, NULL, 1392 ¤t_bootfs_obj, mdn, stack)) 1393 return (0); 1394 } else { 1395 if (errnum = get_objset_mdn(MOS, current_bootfs, 1396 ¤t_bootfs_obj, mdn, stack)) { 1397 grub_memset(current_bootfs, 0, MAXNAMELEN); 1398 return (0); 1399 } 1400 } 1401 } 1402 1403 if (dnode_get_path(mdn, filename, DNODE, stack)) { 1404 errnum = ERR_FILE_NOT_FOUND; 1405 return (0); 1406 } 1407 1408 /* get the file size and set the file position to 0 */ 1409 filemax = ((znode_phys_t *)DN_BONUS(DNODE))->zp_size; 1410 filepos = 0; 1411 1412 dnode_buf = NULL; 1413 return (1); 1414 } 1415 1416 /* 1417 * zfs_read reads in the data blocks pointed by the DNODE. 1418 * 1419 * Return: 1420 * len - the length successfully read in to the buffer 1421 * 0 - failure 1422 */ 1423 int 1424 zfs_read(char *buf, int len) 1425 { 1426 char *stack; 1427 char *tmpbuf; 1428 int blksz, length, movesize; 1429 1430 if (file_buf == NULL) { 1431 file_buf = stackbase; 1432 stackbase += SPA_MAXBLOCKSIZE; 1433 file_start = file_end = 0; 1434 } 1435 stack = stackbase; 1436 1437 /* 1438 * If offset is in memory, move it into the buffer provided and return. 1439 */ 1440 if (filepos >= file_start && filepos+len <= file_end) { 1441 grub_memmove(buf, file_buf + filepos - file_start, len); 1442 filepos += len; 1443 return (len); 1444 } 1445 1446 blksz = DNODE->dn_datablkszsec << SPA_MINBLOCKSHIFT; 1447 1448 /* 1449 * Entire Dnode is too big to fit into the space available. We 1450 * will need to read it in chunks. This could be optimized to 1451 * read in as large a chunk as there is space available, but for 1452 * now, this only reads in one data block at a time. 1453 */ 1454 length = len; 1455 while (length) { 1456 /* 1457 * Find requested blkid and the offset within that block. 1458 */ 1459 uint64_t blkid = filepos / blksz; 1460 1461 if (errnum = dmu_read(DNODE, blkid, file_buf, stack)) 1462 return (0); 1463 1464 file_start = blkid * blksz; 1465 file_end = file_start + blksz; 1466 1467 movesize = MIN(length, file_end - filepos); 1468 1469 grub_memmove(buf, file_buf + filepos - file_start, 1470 movesize); 1471 buf += movesize; 1472 length -= movesize; 1473 filepos += movesize; 1474 } 1475 1476 return (len); 1477 } 1478 1479 /* 1480 * No-Op 1481 */ 1482 int 1483 zfs_embed(int *start_sector, int needed_sectors) 1484 { 1485 return (1); 1486 } 1487 1488 #endif /* FSYS_ZFS */ 1489