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 2008 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 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 /* 76 * Our own version of bcmp(). 77 */ 78 static int 79 zfs_bcmp(const void *s1, const void *s2, size_t n) 80 { 81 const uchar_t *ps1 = s1; 82 const uchar_t *ps2 = s2; 83 84 if (s1 != s2 && n != 0) { 85 do { 86 if (*ps1++ != *ps2++) 87 return (1); 88 } while (--n != 0); 89 } 90 91 return (0); 92 } 93 94 /* 95 * Our own version of log2(). Same thing as highbit()-1. 96 */ 97 static int 98 zfs_log2(uint64_t num) 99 { 100 int i = 0; 101 102 while (num > 1) { 103 i++; 104 num = num >> 1; 105 } 106 107 return (i); 108 } 109 110 /* Checksum Functions */ 111 static void 112 zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp) 113 { 114 ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0); 115 } 116 117 /* Checksum Table and Values */ 118 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = { 119 NULL, NULL, 0, 0, "inherit", 120 NULL, NULL, 0, 0, "on", 121 zio_checksum_off, zio_checksum_off, 0, 0, "off", 122 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "label", 123 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "gang_header", 124 fletcher_2_native, fletcher_2_byteswap, 0, 1, "zilog", 125 fletcher_2_native, fletcher_2_byteswap, 0, 0, "fletcher2", 126 fletcher_4_native, fletcher_4_byteswap, 1, 0, "fletcher4", 127 zio_checksum_SHA256, zio_checksum_SHA256, 1, 0, "SHA256", 128 }; 129 130 /* 131 * zio_checksum_verify: Provides support for checksum verification. 132 * 133 * Fletcher2, Fletcher4, and SHA256 are supported. 134 * 135 * Return: 136 * -1 = Failure 137 * 0 = Success 138 */ 139 static int 140 zio_checksum_verify(blkptr_t *bp, char *data, int size) 141 { 142 zio_cksum_t zc = bp->blk_cksum; 143 uint32_t checksum = BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : 144 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 if (checksum == ZIO_CHECKSUM_GANG_HEADER) { 159 /* 160 * 'gang blocks' is not supported. 161 */ 162 return (-1); 163 } 164 165 if (zbt->zbt_magic == BSWAP_64(ZBT_MAGIC)) { 166 /* byte swapping is not supported */ 167 return (-1); 168 } else { 169 expected_cksum = zbt->zbt_cksum; 170 zbt->zbt_cksum = zc; 171 ci->ci_func[0](data, size, &actual_cksum); 172 zbt->zbt_cksum = expected_cksum; 173 } 174 zc = expected_cksum; 175 176 } else { 177 if (BP_IS_GANG(bp)) 178 return (-1); 179 ci->ci_func[byteswap](data, size, &actual_cksum); 180 } 181 182 if ((actual_cksum.zc_word[0] - zc.zc_word[0]) | 183 (actual_cksum.zc_word[1] - zc.zc_word[1]) | 184 (actual_cksum.zc_word[2] - zc.zc_word[2]) | 185 (actual_cksum.zc_word[3] - zc.zc_word[3])) 186 return (-1); 187 188 return (0); 189 } 190 191 /* 192 * vdev_label_offset takes "offset" (the offset within a vdev_label) and 193 * returns its physical disk offset (starting from the beginning of the vdev). 194 * 195 * Input: 196 * psize : Physical size of this vdev 197 * l : Label Number (0-3) 198 * offset : The offset with a vdev_label in which we want the physical 199 * address 200 * Return: 201 * Success : physical disk offset 202 * Failure : errnum = ERR_BAD_ARGUMENT, return value is meaningless 203 */ 204 static uint64_t 205 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 206 { 207 /* XXX Need to add back label support! */ 208 if (l >= VDEV_LABELS/2 || offset > sizeof (vdev_label_t)) { 209 errnum = ERR_BAD_ARGUMENT; 210 return (0); 211 } 212 213 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 214 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 215 216 } 217 218 /* 219 * vdev_uberblock_compare takes two uberblock structures and returns an integer 220 * indicating the more recent of the two. 221 * Return Value = 1 if ub2 is more recent 222 * Return Value = -1 if ub1 is more recent 223 * The most recent uberblock is determined using its transaction number and 224 * timestamp. The uberblock with the highest transaction number is 225 * considered "newer". If the transaction numbers of the two blocks match, the 226 * timestamps are compared to determine the "newer" of the two. 227 */ 228 static int 229 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 230 { 231 if (ub1->ub_txg < ub2->ub_txg) 232 return (-1); 233 if (ub1->ub_txg > ub2->ub_txg) 234 return (1); 235 236 if (ub1->ub_timestamp < ub2->ub_timestamp) 237 return (-1); 238 if (ub1->ub_timestamp > ub2->ub_timestamp) 239 return (1); 240 241 return (0); 242 } 243 244 /* 245 * Three pieces of information are needed to verify an uberblock: the magic 246 * number, the version number, and the checksum. 247 * 248 * Currently Implemented: version number, magic number 249 * Need to Implement: checksum 250 * 251 * Return: 252 * 0 - Success 253 * -1 - Failure 254 */ 255 static int 256 uberblock_verify(uberblock_phys_t *ub, int offset) 257 { 258 259 uberblock_t *uber = &ub->ubp_uberblock; 260 blkptr_t bp; 261 262 BP_ZERO(&bp); 263 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL); 264 BP_SET_BYTEORDER(&bp, ZFS_HOST_BYTEORDER); 265 ZIO_SET_CHECKSUM(&bp.blk_cksum, offset, 0, 0, 0); 266 267 if (zio_checksum_verify(&bp, (char *)ub, UBERBLOCK_SIZE) != 0) 268 return (-1); 269 270 if (uber->ub_magic == UBERBLOCK_MAGIC && 271 uber->ub_version > 0 && uber->ub_version <= SPA_VERSION) 272 return (0); 273 274 return (-1); 275 } 276 277 /* 278 * Find the best uberblock. 279 * Return: 280 * Success - Pointer to the best uberblock. 281 * Failure - NULL 282 */ 283 static uberblock_phys_t * 284 find_bestub(uberblock_phys_t *ub_array, int label) 285 { 286 uberblock_phys_t *ubbest = NULL; 287 int i, offset; 288 289 for (i = 0; i < (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT); i++) { 290 offset = vdev_label_offset(0, label, VDEV_UBERBLOCK_OFFSET(i)); 291 if (errnum == ERR_BAD_ARGUMENT) 292 return (NULL); 293 if (uberblock_verify(&ub_array[i], offset) == 0) { 294 if (ubbest == NULL) { 295 ubbest = &ub_array[i]; 296 } else if (vdev_uberblock_compare( 297 &(ub_array[i].ubp_uberblock), 298 &(ubbest->ubp_uberblock)) > 0) { 299 ubbest = &ub_array[i]; 300 } 301 } 302 } 303 304 return (ubbest); 305 } 306 307 /* 308 * Read in a block and put its uncompressed data in buf. 309 * 310 * Return: 311 * 0 - success 312 * errnum - failure 313 */ 314 static int 315 zio_read(blkptr_t *bp, void *buf, char *stack) 316 { 317 uint64_t offset, sector; 318 int psize, lsize; 319 int i, comp, cksum; 320 321 psize = BP_GET_PSIZE(bp); 322 lsize = BP_GET_LSIZE(bp); 323 comp = BP_GET_COMPRESS(bp); 324 cksum = BP_GET_CHECKSUM(bp); 325 326 if ((unsigned int)comp >= ZIO_COMPRESS_FUNCTIONS || 327 comp != ZIO_COMPRESS_OFF && decomp_table[comp].decomp_func == NULL) 328 return (ERR_FSYS_CORRUPT); 329 330 if ((char *)buf < stack && ((char *)buf) + lsize > stack) 331 return (ERR_FSYS_CORRUPT); 332 /* pick a good dva from the block pointer */ 333 for (i = 0; i < SPA_DVAS_PER_BP; i++) { 334 335 if (bp->blk_dva[i].dva_word[0] == 0 && 336 bp->blk_dva[i].dva_word[1] == 0) 337 continue; 338 339 /* read in a block */ 340 offset = DVA_GET_OFFSET(&bp->blk_dva[i]); 341 sector = DVA_OFFSET_TO_PHYS_SECTOR(offset); 342 343 if (comp != ZIO_COMPRESS_OFF) { 344 345 if (devread(sector, 0, psize, stack) == 0) 346 continue; 347 if (zio_checksum_verify(bp, stack, psize) != 0) 348 continue; 349 decomp_table[comp].decomp_func(stack, buf, psize, 350 lsize); 351 } else { 352 if (devread(sector, 0, psize, buf) == 0) 353 continue; 354 if (zio_checksum_verify(bp, buf, psize) != 0) 355 continue; 356 } 357 return (0); 358 } 359 360 return (ERR_FSYS_CORRUPT); 361 } 362 363 /* 364 * Get the block from a block id. 365 * push the block onto the stack. 366 * 367 * Return: 368 * 0 - success 369 * errnum - failure 370 */ 371 static int 372 dmu_read(dnode_phys_t *dn, uint64_t blkid, void *buf, char *stack) 373 { 374 int idx, level; 375 blkptr_t *bp_array = dn->dn_blkptr; 376 int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 377 blkptr_t *bp, *tmpbuf; 378 379 bp = (blkptr_t *)stack; 380 stack += sizeof (blkptr_t); 381 382 tmpbuf = (blkptr_t *)stack; 383 stack += 1<<dn->dn_indblkshift; 384 385 for (level = dn->dn_nlevels - 1; level >= 0; level--) { 386 idx = (blkid >> (epbs * level)) & ((1<<epbs)-1); 387 *bp = bp_array[idx]; 388 if (level == 0) 389 tmpbuf = buf; 390 if (BP_IS_HOLE(bp)) { 391 grub_memset(buf, 0, 392 dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); 393 break; 394 } else if (errnum = zio_read(bp, tmpbuf, stack)) { 395 return (errnum); 396 } 397 398 bp_array = tmpbuf; 399 } 400 401 return (0); 402 } 403 404 /* 405 * mzap_lookup: Looks up property described by "name" and returns the value 406 * in "value". 407 * 408 * Return: 409 * 0 - success 410 * errnum - failure 411 */ 412 static int 413 mzap_lookup(mzap_phys_t *zapobj, int objsize, char *name, 414 uint64_t *value) 415 { 416 int i, chunks; 417 mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk; 418 419 chunks = objsize/MZAP_ENT_LEN - 1; 420 for (i = 0; i < chunks; i++) { 421 if (grub_strcmp(mzap_ent[i].mze_name, name) == 0) { 422 *value = mzap_ent[i].mze_value; 423 return (0); 424 } 425 } 426 427 return (ERR_FSYS_CORRUPT); 428 } 429 430 static uint64_t 431 zap_hash(uint64_t salt, const char *name) 432 { 433 static uint64_t table[256]; 434 const uint8_t *cp; 435 uint8_t c; 436 uint64_t crc = salt; 437 438 if (table[128] == 0) { 439 uint64_t *ct; 440 int i, j; 441 for (i = 0; i < 256; i++) { 442 for (ct = table + i, *ct = i, j = 8; j > 0; j--) 443 *ct = (*ct >> 1) ^ (-(*ct & 1) & 444 ZFS_CRC64_POLY); 445 } 446 } 447 448 if (crc == 0 || table[128] != ZFS_CRC64_POLY) { 449 errnum = ERR_FSYS_CORRUPT; 450 return (0); 451 } 452 453 for (cp = (const uint8_t *)name; (c = *cp) != '\0'; cp++) 454 crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF]; 455 456 /* 457 * Only use 28 bits, since we need 4 bits in the cookie for the 458 * collision differentiator. We MUST use the high bits, since 459 * those are the onces that we first pay attention to when 460 * chosing the bucket. 461 */ 462 crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1); 463 464 return (crc); 465 } 466 467 /* 468 * Only to be used on 8-bit arrays. 469 * array_len is actual len in bytes (not encoded le_value_length). 470 * buf is null-terminated. 471 */ 472 static int 473 zap_leaf_array_equal(zap_leaf_phys_t *l, int blksft, int chunk, 474 int array_len, const char *buf) 475 { 476 int bseen = 0; 477 478 while (bseen < array_len) { 479 struct zap_leaf_array *la = 480 &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array; 481 int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES); 482 483 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 484 return (0); 485 486 if (zfs_bcmp(la->la_array, buf + bseen, toread) != 0) 487 break; 488 chunk = la->la_next; 489 bseen += toread; 490 } 491 return (bseen == array_len); 492 } 493 494 /* 495 * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the 496 * value for the property "name". 497 * 498 * Return: 499 * 0 - success 500 * errnum - failure 501 */ 502 static int 503 zap_leaf_lookup(zap_leaf_phys_t *l, int blksft, uint64_t h, 504 const char *name, uint64_t *value) 505 { 506 uint16_t chunk; 507 struct zap_leaf_entry *le; 508 509 /* Verify if this is a valid leaf block */ 510 if (l->l_hdr.lh_block_type != ZBT_LEAF) 511 return (ERR_FSYS_CORRUPT); 512 if (l->l_hdr.lh_magic != ZAP_LEAF_MAGIC) 513 return (ERR_FSYS_CORRUPT); 514 515 for (chunk = l->l_hash[LEAF_HASH(blksft, h)]; 516 chunk != CHAIN_END; chunk = le->le_next) { 517 518 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 519 return (ERR_FSYS_CORRUPT); 520 521 le = ZAP_LEAF_ENTRY(l, blksft, chunk); 522 523 /* Verify the chunk entry */ 524 if (le->le_type != ZAP_CHUNK_ENTRY) 525 return (ERR_FSYS_CORRUPT); 526 527 if (le->le_hash != h) 528 continue; 529 530 if (zap_leaf_array_equal(l, blksft, le->le_name_chunk, 531 le->le_name_length, name)) { 532 533 struct zap_leaf_array *la; 534 uint8_t *ip; 535 536 if (le->le_int_size != 8 || le->le_value_length != 1) 537 return (ERR_FSYS_CORRUPT); 538 539 /* get the uint64_t property value */ 540 la = &ZAP_LEAF_CHUNK(l, blksft, 541 le->le_value_chunk).l_array; 542 ip = la->la_array; 543 544 *value = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 | 545 (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 | 546 (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 | 547 (uint64_t)ip[6] << 8 | (uint64_t)ip[7]; 548 549 return (0); 550 } 551 } 552 553 return (ERR_FSYS_CORRUPT); 554 } 555 556 /* 557 * Fat ZAP lookup 558 * 559 * Return: 560 * 0 - success 561 * errnum - failure 562 */ 563 static int 564 fzap_lookup(dnode_phys_t *zap_dnode, zap_phys_t *zap, 565 char *name, uint64_t *value, char *stack) 566 { 567 zap_leaf_phys_t *l; 568 uint64_t hash, idx, blkid; 569 int blksft = zfs_log2(zap_dnode->dn_datablkszsec << DNODE_SHIFT); 570 571 /* Verify if this is a fat zap header block */ 572 if (zap->zap_magic != (uint64_t)ZAP_MAGIC) 573 return (ERR_FSYS_CORRUPT); 574 575 hash = zap_hash(zap->zap_salt, name); 576 if (errnum) 577 return (errnum); 578 579 /* get block id from index */ 580 if (zap->zap_ptrtbl.zt_numblks != 0) { 581 /* external pointer tables not supported */ 582 return (ERR_FSYS_CORRUPT); 583 } 584 idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift); 585 blkid = ((uint64_t *)zap)[idx + (1<<(blksft-3-1))]; 586 587 /* Get the leaf block */ 588 l = (zap_leaf_phys_t *)stack; 589 stack += 1<<blksft; 590 if ((1<<blksft) < sizeof (zap_leaf_phys_t)) 591 return (ERR_FSYS_CORRUPT); 592 if (errnum = dmu_read(zap_dnode, blkid, l, stack)) 593 return (errnum); 594 595 return (zap_leaf_lookup(l, blksft, hash, name, value)); 596 } 597 598 /* 599 * Read in the data of a zap object and find the value for a matching 600 * property name. 601 * 602 * Return: 603 * 0 - success 604 * errnum - failure 605 */ 606 static int 607 zap_lookup(dnode_phys_t *zap_dnode, char *name, uint64_t *val, char *stack) 608 { 609 uint64_t block_type; 610 int size; 611 void *zapbuf; 612 613 /* Read in the first block of the zap object data. */ 614 zapbuf = stack; 615 size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 616 stack += size; 617 if (errnum = dmu_read(zap_dnode, 0, zapbuf, stack)) 618 return (errnum); 619 620 block_type = *((uint64_t *)zapbuf); 621 622 if (block_type == ZBT_MICRO) { 623 return (mzap_lookup(zapbuf, size, name, val)); 624 } else if (block_type == ZBT_HEADER) { 625 /* this is a fat zap */ 626 return (fzap_lookup(zap_dnode, zapbuf, name, 627 val, stack)); 628 } 629 630 return (ERR_FSYS_CORRUPT); 631 } 632 633 /* 634 * Get the dnode of an object number from the metadnode of an object set. 635 * 636 * Input 637 * mdn - metadnode to get the object dnode 638 * objnum - object number for the object dnode 639 * buf - data buffer that holds the returning dnode 640 * stack - scratch area 641 * 642 * Return: 643 * 0 - success 644 * errnum - failure 645 */ 646 static int 647 dnode_get(dnode_phys_t *mdn, uint64_t objnum, uint8_t type, dnode_phys_t *buf, 648 char *stack) 649 { 650 uint64_t blkid, blksz; /* the block id this object dnode is in */ 651 int epbs; /* shift of number of dnodes in a block */ 652 int idx; /* index within a block */ 653 dnode_phys_t *dnbuf; 654 655 blksz = mdn->dn_datablkszsec << SPA_MINBLOCKSHIFT; 656 epbs = zfs_log2(blksz) - DNODE_SHIFT; 657 blkid = objnum >> epbs; 658 idx = objnum & ((1<<epbs)-1); 659 660 if (dnode_buf != NULL && dnode_mdn == mdn && 661 objnum >= dnode_start && objnum < dnode_end) { 662 grub_memmove(buf, &dnode_buf[idx], DNODE_SIZE); 663 VERIFY_DN_TYPE(buf, type); 664 return (0); 665 } 666 667 if (dnode_buf && blksz == 1<<DNODE_BLOCK_SHIFT) { 668 dnbuf = dnode_buf; 669 dnode_mdn = mdn; 670 dnode_start = blkid << epbs; 671 dnode_end = (blkid + 1) << epbs; 672 } else { 673 dnbuf = (dnode_phys_t *)stack; 674 stack += blksz; 675 } 676 677 if (errnum = dmu_read(mdn, blkid, (char *)dnbuf, stack)) 678 return (errnum); 679 680 grub_memmove(buf, &dnbuf[idx], DNODE_SIZE); 681 VERIFY_DN_TYPE(buf, type); 682 683 return (0); 684 } 685 686 /* 687 * Check if this is a special file that resides at the top 688 * dataset of the pool. Currently this is the GRUB menu, 689 * boot signature and boot signature backup. 690 * str starts with '/'. 691 */ 692 static int 693 is_top_dataset_file(char *str) 694 { 695 char *tptr; 696 697 if ((tptr = grub_strstr(str, "menu.lst")) && 698 (tptr[8] == '\0' || tptr[8] == ' ') && 699 *(tptr-1) == '/') 700 return (1); 701 702 if (grub_strncmp(str, BOOTSIGN_DIR"/", 703 grub_strlen(BOOTSIGN_DIR) + 1) == 0) 704 return (1); 705 706 if (grub_strcmp(str, BOOTSIGN_BACKUP) == 0) 707 return (1); 708 709 return (0); 710 } 711 712 /* 713 * Get the file dnode for a given file name where mdn is the meta dnode 714 * for this ZFS object set. When found, place the file dnode in dn. 715 * The 'path' argument will be mangled. 716 * 717 * Return: 718 * 0 - success 719 * errnum - failure 720 */ 721 static int 722 dnode_get_path(dnode_phys_t *mdn, char *path, dnode_phys_t *dn, 723 char *stack) 724 { 725 uint64_t objnum, version; 726 char *cname, ch; 727 728 if (errnum = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE, 729 dn, stack)) 730 return (errnum); 731 732 if (errnum = zap_lookup(dn, ZPL_VERSION_STR, &version, stack)) 733 return (errnum); 734 if (version > ZPL_VERSION) 735 return (-1); 736 737 if (errnum = zap_lookup(dn, ZFS_ROOT_OBJ, &objnum, stack)) 738 return (errnum); 739 740 if (errnum = dnode_get(mdn, objnum, DMU_OT_DIRECTORY_CONTENTS, 741 dn, stack)) 742 return (errnum); 743 744 /* skip leading slashes */ 745 while (*path == '/') 746 path++; 747 748 while (*path && !isspace(*path)) { 749 750 /* get the next component name */ 751 cname = path; 752 while (*path && !isspace(*path) && *path != '/') 753 path++; 754 ch = *path; 755 *path = 0; /* ensure null termination */ 756 757 if (errnum = zap_lookup(dn, cname, &objnum, stack)) 758 return (errnum); 759 760 objnum = ZFS_DIRENT_OBJ(objnum); 761 if (errnum = dnode_get(mdn, objnum, 0, dn, stack)) 762 return (errnum); 763 764 *path = ch; 765 while (*path == '/') 766 path++; 767 } 768 769 /* We found the dnode for this file. Verify if it is a plain file. */ 770 VERIFY_DN_TYPE(dn, DMU_OT_PLAIN_FILE_CONTENTS); 771 772 return (0); 773 } 774 775 /* 776 * Get the default 'bootfs' property value from the rootpool. 777 * 778 * Return: 779 * 0 - success 780 * errnum -failure 781 */ 782 static int 783 get_default_bootfsobj(dnode_phys_t *mosmdn, uint64_t *obj, char *stack) 784 { 785 uint64_t objnum = 0; 786 dnode_phys_t *dn = (dnode_phys_t *)stack; 787 stack += DNODE_SIZE; 788 789 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 790 DMU_OT_OBJECT_DIRECTORY, dn, stack)) 791 return (errnum); 792 793 /* 794 * find the object number for 'pool_props', and get the dnode 795 * of the 'pool_props'. 796 */ 797 if (zap_lookup(dn, DMU_POOL_PROPS, &objnum, stack)) 798 return (ERR_FILESYSTEM_NOT_FOUND); 799 800 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_POOL_PROPS, dn, stack)) 801 return (errnum); 802 803 if (zap_lookup(dn, ZPOOL_PROP_BOOTFS, &objnum, stack)) 804 return (ERR_FILESYSTEM_NOT_FOUND); 805 806 if (!objnum) 807 return (ERR_FILESYSTEM_NOT_FOUND); 808 809 *obj = objnum; 810 return (0); 811 } 812 813 /* 814 * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname), 815 * e.g. pool/rootfs, or a given object number (obj), e.g. the object number 816 * of pool/rootfs. 817 * 818 * If no fsname and no obj are given, return the DSL_DIR metadnode. 819 * If fsname is given, return its metadnode and its matching object number. 820 * If only obj is given, return the metadnode for this object number. 821 * 822 * Return: 823 * 0 - success 824 * errnum - failure 825 */ 826 static int 827 get_objset_mdn(dnode_phys_t *mosmdn, char *fsname, uint64_t *obj, 828 dnode_phys_t *mdn, char *stack) 829 { 830 uint64_t objnum, headobj; 831 char *cname, ch; 832 blkptr_t *bp; 833 objset_phys_t *osp; 834 int issnapshot = 0; 835 char *snapname; 836 837 if (fsname == NULL && obj) { 838 headobj = *obj; 839 goto skip; 840 } 841 842 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 843 DMU_OT_OBJECT_DIRECTORY, mdn, stack)) 844 return (errnum); 845 846 if (errnum = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum, 847 stack)) 848 return (errnum); 849 850 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, stack)) 851 return (errnum); 852 853 if (fsname == NULL) { 854 headobj = 855 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 856 goto skip; 857 } 858 859 /* take out the pool name */ 860 while (*fsname && !isspace(*fsname) && *fsname != '/') 861 fsname++; 862 863 while (*fsname && !isspace(*fsname)) { 864 uint64_t childobj; 865 866 while (*fsname == '/') 867 fsname++; 868 869 cname = fsname; 870 while (*fsname && !isspace(*fsname) && *fsname != '/') 871 fsname++; 872 ch = *fsname; 873 *fsname = 0; 874 875 snapname = cname; 876 while (*snapname && !isspace(*snapname) && *snapname != '@') 877 snapname++; 878 if (*snapname == '@') { 879 issnapshot = 1; 880 *snapname = 0; 881 } 882 childobj = 883 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_child_dir_zapobj; 884 if (errnum = dnode_get(mosmdn, childobj, 885 DMU_OT_DSL_DIR_CHILD_MAP, mdn, stack)) 886 return (errnum); 887 888 if (zap_lookup(mdn, cname, &objnum, stack)) 889 return (ERR_FILESYSTEM_NOT_FOUND); 890 891 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, 892 mdn, stack)) 893 return (errnum); 894 895 *fsname = ch; 896 if (issnapshot) 897 *snapname = '@'; 898 } 899 headobj = ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 900 if (obj) 901 *obj = headobj; 902 903 skip: 904 if (errnum = dnode_get(mosmdn, headobj, DMU_OT_DSL_DATASET, mdn, stack)) 905 return (errnum); 906 if (issnapshot) { 907 uint64_t snapobj; 908 909 snapobj = ((dsl_dataset_phys_t *)DN_BONUS(mdn))-> 910 ds_snapnames_zapobj; 911 912 if (errnum = dnode_get(mosmdn, snapobj, 913 DMU_OT_DSL_DS_SNAP_MAP, mdn, stack)) 914 return (errnum); 915 if (zap_lookup(mdn, snapname + 1, &headobj, stack)) 916 return (ERR_FILESYSTEM_NOT_FOUND); 917 if (errnum = dnode_get(mosmdn, headobj, 918 DMU_OT_DSL_DATASET, mdn, stack)) 919 return (errnum); 920 if (obj) 921 *obj = headobj; 922 } 923 924 bp = &((dsl_dataset_phys_t *)DN_BONUS(mdn))->ds_bp; 925 osp = (objset_phys_t *)stack; 926 stack += sizeof (objset_phys_t); 927 if (errnum = zio_read(bp, osp, stack)) 928 return (errnum); 929 930 grub_memmove((char *)mdn, (char *)&osp->os_meta_dnode, DNODE_SIZE); 931 932 return (0); 933 } 934 935 /* 936 * For a given XDR packed nvlist, verify the first 4 bytes and move on. 937 * 938 * An XDR packed nvlist is encoded as (comments from nvs_xdr_create) : 939 * 940 * encoding method/host endian (4 bytes) 941 * nvl_version (4 bytes) 942 * nvl_nvflag (4 bytes) 943 * encoded nvpairs: 944 * encoded size of the nvpair (4 bytes) 945 * decoded size of the nvpair (4 bytes) 946 * name string size (4 bytes) 947 * name string data (sizeof(NV_ALIGN4(string)) 948 * data type (4 bytes) 949 * # of elements in the nvpair (4 bytes) 950 * data 951 * 2 zero's for the last nvpair 952 * (end of the entire list) (8 bytes) 953 * 954 * Return: 955 * 0 - success 956 * 1 - failure 957 */ 958 static int 959 nvlist_unpack(char *nvlist, char **out) 960 { 961 /* Verify if the 1st and 2nd byte in the nvlist are valid. */ 962 if (nvlist[0] != NV_ENCODE_XDR || nvlist[1] != HOST_ENDIAN) 963 return (1); 964 965 nvlist += 4; 966 *out = nvlist; 967 return (0); 968 } 969 970 static char * 971 nvlist_array(char *nvlist, int index) 972 { 973 int i, encode_size; 974 975 for (i = 0; i < index; i++) { 976 /* skip the header, nvl_version, and nvl_nvflag */ 977 nvlist = nvlist + 4 * 2; 978 979 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) 980 nvlist += encode_size; /* goto the next nvpair */ 981 982 nvlist = nvlist + 4 * 2; /* skip the ending 2 zeros - 8 bytes */ 983 } 984 985 return (nvlist); 986 } 987 988 static int 989 nvlist_lookup_value(char *nvlist, char *name, void *val, int valtype, 990 int *nelmp) 991 { 992 int name_len, type, slen, encode_size; 993 char *nvpair, *nvp_name, *strval = val; 994 uint64_t *intval = val; 995 996 /* skip the header, nvl_version, and nvl_nvflag */ 997 nvlist = nvlist + 4 * 2; 998 999 /* 1000 * Loop thru the nvpair list 1001 * The XDR representation of an integer is in big-endian byte order. 1002 */ 1003 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) { 1004 1005 nvpair = nvlist + 4 * 2; /* skip the encode/decode size */ 1006 1007 name_len = BSWAP_32(*(uint32_t *)nvpair); 1008 nvpair += 4; 1009 1010 nvp_name = nvpair; 1011 nvpair = nvpair + ((name_len + 3) & ~3); /* align */ 1012 1013 type = BSWAP_32(*(uint32_t *)nvpair); 1014 nvpair += 4; 1015 1016 if ((grub_strncmp(nvp_name, name, name_len) == 0) && 1017 type == valtype) { 1018 int nelm; 1019 1020 if ((nelm = BSWAP_32(*(uint32_t *)nvpair)) < 1) 1021 return (1); 1022 nvpair += 4; 1023 1024 switch (valtype) { 1025 case DATA_TYPE_STRING: 1026 slen = BSWAP_32(*(uint32_t *)nvpair); 1027 nvpair += 4; 1028 grub_memmove(strval, nvpair, slen); 1029 strval[slen] = '\0'; 1030 return (0); 1031 1032 case DATA_TYPE_UINT64: 1033 *intval = BSWAP_64(*(uint64_t *)nvpair); 1034 return (0); 1035 1036 case DATA_TYPE_NVLIST: 1037 *(void **)val = (void *)nvpair; 1038 return (0); 1039 1040 case DATA_TYPE_NVLIST_ARRAY: 1041 *(void **)val = (void *)nvpair; 1042 if (nelmp) 1043 *nelmp = nelm; 1044 return (0); 1045 } 1046 } 1047 1048 nvlist += encode_size; /* goto the next nvpair */ 1049 } 1050 1051 return (1); 1052 } 1053 1054 /* 1055 * Check if this vdev is online and is in a good state. 1056 */ 1057 static int 1058 vdev_validate(char *nv) 1059 { 1060 uint64_t ival; 1061 1062 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_OFFLINE, &ival, 1063 DATA_TYPE_UINT64, NULL) == 0 || 1064 nvlist_lookup_value(nv, ZPOOL_CONFIG_FAULTED, &ival, 1065 DATA_TYPE_UINT64, NULL) == 0 || 1066 nvlist_lookup_value(nv, ZPOOL_CONFIG_REMOVED, &ival, 1067 DATA_TYPE_UINT64, NULL) == 0) 1068 return (ERR_DEV_VALUES); 1069 1070 return (0); 1071 } 1072 1073 /* 1074 * Get a list of valid vdev pathname from the boot device. 1075 * The caller should already allocate MAXPATHLEN memory for bootpath and devid. 1076 */ 1077 int 1078 vdev_get_bootpath(char *nv, uint64_t inguid, char *devid, char *bootpath) 1079 { 1080 char type[16]; 1081 1082 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_TYPE, &type, DATA_TYPE_STRING, 1083 NULL)) 1084 return (ERR_FSYS_CORRUPT); 1085 1086 if (strcmp(type, VDEV_TYPE_DISK) == 0) { 1087 uint64_t guid; 1088 1089 if (vdev_validate(nv) != 0) 1090 return (ERR_NO_BOOTPATH); 1091 1092 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_GUID, 1093 &guid, DATA_TYPE_UINT64, NULL) != 0) 1094 return (ERR_NO_BOOTPATH); 1095 1096 if (guid != inguid) 1097 return (ERR_NO_BOOTPATH); 1098 1099 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_PHYS_PATH, 1100 bootpath, DATA_TYPE_STRING, NULL) != 0) 1101 bootpath[0] = '\0'; 1102 1103 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_DEVID, 1104 devid, DATA_TYPE_STRING, NULL) != 0) 1105 devid[0] = '\0'; 1106 1107 if (strlen(bootpath) >= MAXPATHLEN || 1108 strlen(devid) >= MAXPATHLEN) 1109 return (ERR_WONT_FIT); 1110 1111 return (0); 1112 1113 } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0) { 1114 int nelm, i; 1115 char *child; 1116 1117 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_CHILDREN, &child, 1118 DATA_TYPE_NVLIST_ARRAY, &nelm)) 1119 return (ERR_FSYS_CORRUPT); 1120 1121 for (i = 0; i < nelm; i++) { 1122 char *child_i; 1123 1124 child_i = nvlist_array(child, i); 1125 if (vdev_get_bootpath(child_i, inguid, devid, 1126 bootpath) == 0) 1127 return (0); 1128 } 1129 } 1130 1131 return (ERR_NO_BOOTPATH); 1132 } 1133 1134 /* 1135 * Check the disk label information and retrieve needed vdev name-value pairs. 1136 * 1137 * Return: 1138 * 0 - success 1139 * ERR_* - failure 1140 */ 1141 int 1142 check_pool_label(int label, char *stack, char *outdevid, char *outpath) 1143 { 1144 vdev_phys_t *vdev; 1145 uint64_t sector, pool_state, txg = 0; 1146 char *nvlist, *nv; 1147 uint64_t diskguid; 1148 uint64_t version; 1149 1150 sector = (label * sizeof (vdev_label_t) + VDEV_SKIP_SIZE + 1151 VDEV_BOOT_HEADER_SIZE) >> SPA_MINBLOCKSHIFT; 1152 1153 /* Read in the vdev name-value pair list (112K). */ 1154 if (devread(sector, 0, VDEV_PHYS_SIZE, stack) == 0) 1155 return (ERR_READ); 1156 1157 vdev = (vdev_phys_t *)stack; 1158 stack += sizeof (vdev_phys_t); 1159 1160 if (nvlist_unpack(vdev->vp_nvlist, &nvlist)) 1161 return (ERR_FSYS_CORRUPT); 1162 1163 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_STATE, &pool_state, 1164 DATA_TYPE_UINT64, NULL)) 1165 return (ERR_FSYS_CORRUPT); 1166 1167 if (pool_state == POOL_STATE_DESTROYED) 1168 return (ERR_FILESYSTEM_NOT_FOUND); 1169 1170 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_NAME, 1171 current_rootpool, DATA_TYPE_STRING, NULL)) 1172 return (ERR_FSYS_CORRUPT); 1173 1174 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_TXG, &txg, 1175 DATA_TYPE_UINT64, NULL)) 1176 return (ERR_FSYS_CORRUPT); 1177 1178 /* not an active device */ 1179 if (txg == 0) 1180 return (ERR_NO_BOOTPATH); 1181 1182 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VERSION, &version, 1183 DATA_TYPE_UINT64, NULL)) 1184 return (ERR_FSYS_CORRUPT); 1185 if (version > SPA_VERSION) 1186 return (ERR_NEWER_VERSION); 1187 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VDEV_TREE, &nv, 1188 DATA_TYPE_NVLIST, NULL)) 1189 return (ERR_FSYS_CORRUPT); 1190 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_GUID, &diskguid, 1191 DATA_TYPE_UINT64, NULL)) 1192 return (ERR_FSYS_CORRUPT); 1193 if (vdev_get_bootpath(nv, diskguid, outdevid, outpath)) 1194 return (ERR_NO_BOOTPATH); 1195 return (0); 1196 } 1197 1198 /* 1199 * zfs_mount() locates a valid uberblock of the root pool and read in its MOS 1200 * to the memory address MOS. 1201 * 1202 * Return: 1203 * 1 - success 1204 * 0 - failure 1205 */ 1206 int 1207 zfs_mount(void) 1208 { 1209 char *stack; 1210 int label = 0; 1211 uberblock_phys_t *ub_array, *ubbest = NULL; 1212 vdev_boot_header_t *bh; 1213 objset_phys_t *osp; 1214 char tmp_bootpath[MAXNAMELEN]; 1215 char tmp_devid[MAXNAMELEN]; 1216 1217 /* if it's our first time here, zero the best uberblock out */ 1218 if (best_drive == 0 && best_part == 0 && find_best_root) 1219 grub_memset(¤t_uberblock, 0, sizeof (uberblock_t)); 1220 1221 stackbase = ZFS_SCRATCH; 1222 stack = stackbase; 1223 ub_array = (uberblock_phys_t *)stack; 1224 stack += VDEV_UBERBLOCK_RING; 1225 1226 bh = (vdev_boot_header_t *)stack; 1227 stack += VDEV_BOOT_HEADER_SIZE; 1228 1229 osp = (objset_phys_t *)stack; 1230 stack += sizeof (objset_phys_t); 1231 1232 /* XXX add back labels support? */ 1233 for (label = 0; ubbest == NULL && label < (VDEV_LABELS/2); label++) { 1234 uint64_t sector = (label * sizeof (vdev_label_t) + 1235 VDEV_SKIP_SIZE) >> SPA_MINBLOCKSHIFT; 1236 if (devread(sector, 0, VDEV_BOOT_HEADER_SIZE, 1237 (char *)bh) == 0) 1238 continue; 1239 if ((bh->vb_magic != VDEV_BOOT_MAGIC) || 1240 (bh->vb_version != VDEV_BOOT_VERSION)) { 1241 continue; 1242 } 1243 sector += (VDEV_BOOT_HEADER_SIZE + 1244 VDEV_PHYS_SIZE) >> SPA_MINBLOCKSHIFT; 1245 1246 /* Read in the uberblock ring (128K). */ 1247 if (devread(sector, 0, VDEV_UBERBLOCK_RING, 1248 (char *)ub_array) == 0) 1249 continue; 1250 1251 if ((ubbest = find_bestub(ub_array, label)) != NULL && 1252 zio_read(&ubbest->ubp_uberblock.ub_rootbp, osp, stack) 1253 == 0) { 1254 1255 VERIFY_OS_TYPE(osp, DMU_OST_META); 1256 1257 if (check_pool_label(label, stack, tmp_devid, 1258 tmp_bootpath)) 1259 return (0); 1260 1261 if (find_best_root && 1262 vdev_uberblock_compare(&ubbest->ubp_uberblock, 1263 &(current_uberblock)) <= 0) 1264 continue; 1265 1266 /* Got the MOS. Save it at the memory addr MOS. */ 1267 grub_memmove(MOS, &osp->os_meta_dnode, DNODE_SIZE); 1268 grub_memmove(¤t_uberblock, 1269 &ubbest->ubp_uberblock, sizeof (uberblock_t)); 1270 grub_memmove(current_bootpath, tmp_bootpath, 1271 MAXNAMELEN); 1272 grub_memmove(current_devid, tmp_devid, 1273 grub_strlen(tmp_devid)); 1274 is_zfs_mount = 1; 1275 return (1); 1276 } 1277 } 1278 1279 return (0); 1280 } 1281 1282 /* 1283 * zfs_open() locates a file in the rootpool by following the 1284 * MOS and places the dnode of the file in the memory address DNODE. 1285 * 1286 * Return: 1287 * 1 - success 1288 * 0 - failure 1289 */ 1290 int 1291 zfs_open(char *filename) 1292 { 1293 char *stack; 1294 dnode_phys_t *mdn; 1295 1296 file_buf = NULL; 1297 stackbase = ZFS_SCRATCH; 1298 stack = stackbase; 1299 1300 mdn = (dnode_phys_t *)stack; 1301 stack += sizeof (dnode_phys_t); 1302 1303 dnode_mdn = NULL; 1304 dnode_buf = (dnode_phys_t *)stack; 1305 stack += 1<<DNODE_BLOCK_SHIFT; 1306 1307 /* 1308 * menu.lst is placed at the root pool filesystem level, 1309 * do not goto 'current_bootfs'. 1310 */ 1311 if (is_top_dataset_file(filename)) { 1312 if (errnum = get_objset_mdn(MOS, NULL, NULL, mdn, stack)) 1313 return (0); 1314 1315 current_bootfs_obj = 0; 1316 } else { 1317 if (current_bootfs[0] == '\0') { 1318 /* Get the default root filesystem object number */ 1319 if (errnum = get_default_bootfsobj(MOS, 1320 ¤t_bootfs_obj, stack)) 1321 return (0); 1322 1323 if (errnum = get_objset_mdn(MOS, NULL, 1324 ¤t_bootfs_obj, mdn, stack)) 1325 return (0); 1326 } else { 1327 if (errnum = get_objset_mdn(MOS, current_bootfs, 1328 ¤t_bootfs_obj, mdn, stack)) { 1329 grub_memset(current_bootfs, 0, MAXNAMELEN); 1330 return (0); 1331 } 1332 } 1333 } 1334 1335 if (dnode_get_path(mdn, filename, DNODE, stack)) { 1336 errnum = ERR_FILE_NOT_FOUND; 1337 return (0); 1338 } 1339 1340 /* get the file size and set the file position to 0 */ 1341 filemax = ((znode_phys_t *)DN_BONUS(DNODE))->zp_size; 1342 filepos = 0; 1343 1344 dnode_buf = NULL; 1345 return (1); 1346 } 1347 1348 /* 1349 * zfs_read reads in the data blocks pointed by the DNODE. 1350 * 1351 * Return: 1352 * len - the length successfully read in to the buffer 1353 * 0 - failure 1354 */ 1355 int 1356 zfs_read(char *buf, int len) 1357 { 1358 char *stack; 1359 char *tmpbuf; 1360 int blksz, length, movesize; 1361 1362 if (file_buf == NULL) { 1363 file_buf = stackbase; 1364 stackbase += SPA_MAXBLOCKSIZE; 1365 file_start = file_end = 0; 1366 } 1367 stack = stackbase; 1368 1369 /* 1370 * If offset is in memory, move it into the buffer provided and return. 1371 */ 1372 if (filepos >= file_start && filepos+len <= file_end) { 1373 grub_memmove(buf, file_buf + filepos - file_start, len); 1374 filepos += len; 1375 return (len); 1376 } 1377 1378 blksz = DNODE->dn_datablkszsec << SPA_MINBLOCKSHIFT; 1379 1380 /* 1381 * Entire Dnode is too big to fit into the space available. We 1382 * will need to read it in chunks. This could be optimized to 1383 * read in as large a chunk as there is space available, but for 1384 * now, this only reads in one data block at a time. 1385 */ 1386 length = len; 1387 while (length) { 1388 /* 1389 * Find requested blkid and the offset within that block. 1390 */ 1391 uint64_t blkid = filepos / blksz; 1392 1393 if (errnum = dmu_read(DNODE, blkid, file_buf, stack)) 1394 return (0); 1395 1396 file_start = blkid * blksz; 1397 file_end = file_start + blksz; 1398 1399 movesize = MIN(length, file_end - filepos); 1400 1401 grub_memmove(buf, file_buf + filepos - file_start, 1402 movesize); 1403 buf += movesize; 1404 length -= movesize; 1405 filepos += movesize; 1406 } 1407 1408 return (len); 1409 } 1410 1411 /* 1412 * No-Op 1413 */ 1414 int 1415 zfs_embed(int *start_sector, int needed_sectors) 1416 { 1417 return (1); 1418 } 1419 1420 #endif /* FSYS_ZFS */ 1421