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 #pragma ident "%Z%%M% %I% %E% SMI" 24 25 /* 26 * The zfs plug-in routines for GRUB are: 27 * 28 * zfs_mount() - locates a valid uberblock of the root pool and reads 29 * in its MOS at the memory address MOS. 30 * 31 * zfs_open() - locates a plain file object by following the MOS 32 * and places its dnode at the memory address DNODE. 33 * 34 * zfs_read() - read in the data blocks pointed by the DNODE. 35 * 36 * ZFS_SCRATCH is used as a working area. 37 * 38 * (memory addr) MOS DNODE ZFS_SCRATCH 39 * | | | 40 * +-------V---------V----------V---------------+ 41 * memory | | dnode | dnode | scratch | 42 * | | 512B | 512B | area | 43 * +--------------------------------------------+ 44 */ 45 46 #ifdef FSYS_ZFS 47 48 #include "shared.h" 49 #include "filesys.h" 50 #include "fsys_zfs.h" 51 52 /* cache for a file block of the currently zfs_open()-ed file */ 53 static void *file_buf = NULL; 54 static uint64_t file_start = 0; 55 static uint64_t file_end = 0; 56 57 /* cache for a dnode block */ 58 static dnode_phys_t *dnode_buf = NULL; 59 static dnode_phys_t *dnode_mdn = NULL; 60 static uint64_t dnode_start = 0; 61 static uint64_t dnode_end = 0; 62 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 /* 75 * Our own version of bcmp(). 76 */ 77 static int 78 zfs_bcmp(const void *s1, const void *s2, size_t n) 79 { 80 const uchar_t *ps1 = s1; 81 const uchar_t *ps2 = s2; 82 83 if (s1 != s2 && n != 0) { 84 do { 85 if (*ps1++ != *ps2++) 86 return (1); 87 } while (--n != 0); 88 } 89 90 return (0); 91 } 92 93 /* 94 * Our own version of log2(). Same thing as highbit()-1. 95 */ 96 static int 97 zfs_log2(uint64_t num) 98 { 99 int i = 0; 100 101 while (num > 1) { 102 i++; 103 num = num >> 1; 104 } 105 106 return (i); 107 } 108 109 /* Checksum Functions */ 110 static void 111 zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp) 112 { 113 ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0); 114 } 115 116 /* Checksum Table and Values */ 117 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = { 118 NULL, NULL, 0, 0, "inherit", 119 NULL, NULL, 0, 0, "on", 120 zio_checksum_off, zio_checksum_off, 0, 0, "off", 121 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "label", 122 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "gang_header", 123 fletcher_2_native, fletcher_2_byteswap, 0, 1, "zilog", 124 fletcher_2_native, fletcher_2_byteswap, 0, 0, "fletcher2", 125 fletcher_4_native, fletcher_4_byteswap, 1, 0, "fletcher4", 126 zio_checksum_SHA256, zio_checksum_SHA256, 1, 0, "SHA256", 127 }; 128 129 /* 130 * zio_checksum_verify: Provides support for checksum verification. 131 * 132 * Fletcher2, Fletcher4, and SHA256 are supported. 133 * 134 * Return: 135 * -1 = Failure 136 * 0 = Success 137 */ 138 static int 139 zio_checksum_verify(blkptr_t *bp, char *data, int size) 140 { 141 zio_cksum_t zc = bp->blk_cksum; 142 uint32_t checksum = BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : 143 BP_GET_CHECKSUM(bp); 144 int byteswap = BP_SHOULD_BYTESWAP(bp); 145 zio_block_tail_t *zbt = (zio_block_tail_t *)(data + size) - 1; 146 zio_checksum_info_t *ci = &zio_checksum_table[checksum]; 147 zio_cksum_t actual_cksum, expected_cksum; 148 149 /* byteswap is not supported */ 150 if (byteswap) 151 return (-1); 152 153 if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL) 154 return (-1); 155 156 if (ci->ci_zbt) { 157 if (checksum == ZIO_CHECKSUM_GANG_HEADER) { 158 /* 159 * 'gang blocks' is not supported. 160 */ 161 return (-1); 162 } 163 164 if (zbt->zbt_magic == BSWAP_64(ZBT_MAGIC)) { 165 /* byte swapping is not supported */ 166 return (-1); 167 } else { 168 expected_cksum = zbt->zbt_cksum; 169 zbt->zbt_cksum = zc; 170 ci->ci_func[0](data, size, &actual_cksum); 171 zbt->zbt_cksum = expected_cksum; 172 } 173 zc = expected_cksum; 174 175 } else { 176 if (BP_IS_GANG(bp)) 177 return (-1); 178 ci->ci_func[byteswap](data, size, &actual_cksum); 179 } 180 181 if ((actual_cksum.zc_word[0] - zc.zc_word[0]) | 182 (actual_cksum.zc_word[1] - zc.zc_word[1]) | 183 (actual_cksum.zc_word[2] - zc.zc_word[2]) | 184 (actual_cksum.zc_word[3] - zc.zc_word[3])) 185 return (-1); 186 187 return (0); 188 } 189 190 /* 191 * vdev_label_offset takes "offset" (the offset within a vdev_label) and 192 * returns its physical disk offset (starting from the beginning of the vdev). 193 * 194 * Input: 195 * psize : Physical size of this vdev 196 * l : Label Number (0-3) 197 * offset : The offset with a vdev_label in which we want the physical 198 * address 199 * Return: 200 * Success : physical disk offset 201 * Failure : errnum = ERR_BAD_ARGUMENT, return value is meaningless 202 */ 203 static uint64_t 204 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 205 { 206 /* XXX Need to add back label support! */ 207 if (l >= VDEV_LABELS/2 || offset > sizeof (vdev_label_t)) { 208 errnum = ERR_BAD_ARGUMENT; 209 return (0); 210 } 211 212 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 213 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 214 215 } 216 217 /* 218 * vdev_uberblock_compare takes two uberblock structures and returns an integer 219 * indicating the more recent of the two. 220 * Return Value = 1 if ub2 is more recent 221 * Return Value = -1 if ub1 is more recent 222 * The most recent uberblock is determined using its transaction number and 223 * timestamp. The uberblock with the highest transaction number is 224 * considered "newer". If the transaction numbers of the two blocks match, the 225 * timestamps are compared to determine the "newer" of the two. 226 */ 227 static int 228 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 229 { 230 if (ub1->ub_txg < ub2->ub_txg) 231 return (-1); 232 if (ub1->ub_txg > ub2->ub_txg) 233 return (1); 234 235 if (ub1->ub_timestamp < ub2->ub_timestamp) 236 return (-1); 237 if (ub1->ub_timestamp > ub2->ub_timestamp) 238 return (1); 239 240 return (0); 241 } 242 243 /* 244 * Three pieces of information are needed to verify an uberblock: the magic 245 * number, the version number, and the checksum. 246 * 247 * Currently Implemented: version number, magic number 248 * Need to Implement: checksum 249 * 250 * Return: 251 * 0 - Success 252 * -1 - Failure 253 */ 254 static int 255 uberblock_verify(uberblock_phys_t *ub, int offset) 256 { 257 258 uberblock_t *uber = &ub->ubp_uberblock; 259 blkptr_t bp; 260 261 BP_ZERO(&bp); 262 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL); 263 BP_SET_BYTEORDER(&bp, ZFS_HOST_BYTEORDER); 264 ZIO_SET_CHECKSUM(&bp.blk_cksum, offset, 0, 0, 0); 265 266 if (zio_checksum_verify(&bp, (char *)ub, UBERBLOCK_SIZE) != 0) 267 return (-1); 268 269 if (uber->ub_magic == UBERBLOCK_MAGIC && 270 uber->ub_version >= SPA_VERSION_1 && 271 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 /* pick a good dva from the block pointer */ 331 for (i = 0; i < SPA_DVAS_PER_BP; i++) { 332 333 if (bp->blk_dva[i].dva_word[0] == 0 && 334 bp->blk_dva[i].dva_word[1] == 0) 335 continue; 336 337 /* read in a block */ 338 offset = DVA_GET_OFFSET(&bp->blk_dva[i]); 339 sector = DVA_OFFSET_TO_PHYS_SECTOR(offset); 340 341 if (comp != ZIO_COMPRESS_OFF) { 342 343 if (devread(sector, 0, psize, stack) == 0) 344 continue; 345 if (zio_checksum_verify(bp, stack, psize) != 0) 346 continue; 347 decomp_table[comp].decomp_func(stack, buf, psize, 348 lsize); 349 } else { 350 if (devread(sector, 0, psize, buf) == 0) 351 continue; 352 if (zio_checksum_verify(bp, buf, psize) != 0) 353 continue; 354 } 355 return (0); 356 } 357 358 return (ERR_FSYS_CORRUPT); 359 } 360 361 /* 362 * Get the block from a block id. 363 * push the block onto the stack. 364 * 365 * Return: 366 * 0 - success 367 * errnum - failure 368 */ 369 static int 370 dmu_read(dnode_phys_t *dn, uint64_t blkid, void *buf, char *stack) 371 { 372 int idx, level; 373 blkptr_t *bp_array = dn->dn_blkptr; 374 int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 375 blkptr_t *bp, *tmpbuf; 376 377 bp = (blkptr_t *)stack; 378 stack += sizeof (blkptr_t); 379 380 tmpbuf = (blkptr_t *)stack; 381 stack += 1<<dn->dn_indblkshift; 382 383 for (level = dn->dn_nlevels - 1; level >= 0; level--) { 384 idx = (blkid >> (epbs * level)) & ((1<<epbs)-1); 385 *bp = bp_array[idx]; 386 if (level == 0) 387 tmpbuf = buf; 388 if (BP_IS_HOLE(bp)) { 389 grub_memset(buf, 0, 390 dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); 391 break; 392 } else if (errnum = zio_read(bp, tmpbuf, stack)) { 393 return (errnum); 394 } 395 396 bp_array = tmpbuf; 397 } 398 399 return (0); 400 } 401 402 /* 403 * mzap_lookup: Looks up property described by "name" and returns the value 404 * in "value". 405 * 406 * Return: 407 * 0 - success 408 * errnum - failure 409 */ 410 static int 411 mzap_lookup(mzap_phys_t *zapobj, int objsize, char *name, 412 uint64_t *value) 413 { 414 int i, chunks; 415 mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk; 416 417 chunks = objsize/MZAP_ENT_LEN - 1; 418 for (i = 0; i < chunks; i++) { 419 if (grub_strcmp(mzap_ent[i].mze_name, name) == 0) { 420 *value = mzap_ent[i].mze_value; 421 return (0); 422 } 423 } 424 425 return (ERR_FSYS_CORRUPT); 426 } 427 428 static uint64_t 429 zap_hash(uint64_t salt, const char *name) 430 { 431 static uint64_t table[256]; 432 const uint8_t *cp; 433 uint8_t c; 434 uint64_t crc = salt; 435 436 if (table[128] == 0) { 437 uint64_t *ct; 438 int i, j; 439 for (i = 0; i < 256; i++) { 440 for (ct = table + i, *ct = i, j = 8; j > 0; j--) 441 *ct = (*ct >> 1) ^ (-(*ct & 1) & 442 ZFS_CRC64_POLY); 443 } 444 } 445 446 if (crc == 0 || table[128] != ZFS_CRC64_POLY) { 447 errnum = ERR_FSYS_CORRUPT; 448 return (0); 449 } 450 451 for (cp = (const uint8_t *)name; (c = *cp) != '\0'; cp++) 452 crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF]; 453 454 /* 455 * Only use 28 bits, since we need 4 bits in the cookie for the 456 * collision differentiator. We MUST use the high bits, since 457 * those are the onces that we first pay attention to when 458 * chosing the bucket. 459 */ 460 crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1); 461 462 return (crc); 463 } 464 465 /* 466 * Only to be used on 8-bit arrays. 467 * array_len is actual len in bytes (not encoded le_value_length). 468 * buf is null-terminated. 469 */ 470 static int 471 zap_leaf_array_equal(zap_leaf_phys_t *l, int blksft, int chunk, 472 int array_len, const char *buf) 473 { 474 int bseen = 0; 475 476 while (bseen < array_len) { 477 struct zap_leaf_array *la = 478 &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array; 479 int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES); 480 481 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 482 return (0); 483 484 if (zfs_bcmp(la->la_array, buf + bseen, toread) != 0) 485 break; 486 chunk = la->la_next; 487 bseen += toread; 488 } 489 return (bseen == array_len); 490 } 491 492 /* 493 * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the 494 * value for the property "name". 495 * 496 * Return: 497 * 0 - success 498 * errnum - failure 499 */ 500 static int 501 zap_leaf_lookup(zap_leaf_phys_t *l, int blksft, uint64_t h, 502 const char *name, uint64_t *value) 503 { 504 uint16_t chunk; 505 struct zap_leaf_entry *le; 506 507 /* Verify if this is a valid leaf block */ 508 if (l->l_hdr.lh_block_type != ZBT_LEAF) 509 return (ERR_FSYS_CORRUPT); 510 if (l->l_hdr.lh_magic != ZAP_LEAF_MAGIC) 511 return (ERR_FSYS_CORRUPT); 512 513 for (chunk = l->l_hash[LEAF_HASH(blksft, h)]; 514 chunk != CHAIN_END; chunk = le->le_next) { 515 516 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 517 return (ERR_FSYS_CORRUPT); 518 519 le = ZAP_LEAF_ENTRY(l, blksft, chunk); 520 521 /* Verify the chunk entry */ 522 if (le->le_type != ZAP_CHUNK_ENTRY) 523 return (ERR_FSYS_CORRUPT); 524 525 if (le->le_hash != h) 526 continue; 527 528 if (zap_leaf_array_equal(l, blksft, le->le_name_chunk, 529 le->le_name_length, name)) { 530 531 struct zap_leaf_array *la; 532 uint8_t *ip; 533 534 if (le->le_int_size != 8 || le->le_value_length != 1) 535 return (ERR_FSYS_CORRUPT); 536 537 /* get the uint64_t property value */ 538 la = &ZAP_LEAF_CHUNK(l, blksft, 539 le->le_value_chunk).l_array; 540 ip = la->la_array; 541 542 *value = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 | 543 (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 | 544 (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 | 545 (uint64_t)ip[6] << 8 | (uint64_t)ip[7]; 546 547 return (0); 548 } 549 } 550 551 return (ERR_FSYS_CORRUPT); 552 } 553 554 /* 555 * Fat ZAP lookup 556 * 557 * Return: 558 * 0 - success 559 * errnum - failure 560 */ 561 static int 562 fzap_lookup(dnode_phys_t *zap_dnode, zap_phys_t *zap, 563 char *name, uint64_t *value, char *stack) 564 { 565 zap_leaf_phys_t *l; 566 uint64_t hash, idx, blkid; 567 int blksft = zfs_log2(zap_dnode->dn_datablkszsec << DNODE_SHIFT); 568 569 /* Verify if this is a fat zap header block */ 570 if (zap->zap_magic != (uint64_t)ZAP_MAGIC) 571 return (ERR_FSYS_CORRUPT); 572 573 hash = zap_hash(zap->zap_salt, name); 574 if (errnum) 575 return (errnum); 576 577 /* get block id from index */ 578 if (zap->zap_ptrtbl.zt_numblks != 0) { 579 /* external pointer tables not supported */ 580 return (ERR_FSYS_CORRUPT); 581 } 582 idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift); 583 blkid = ((uint64_t *)zap)[idx + (1<<(blksft-3-1))]; 584 585 /* Get the leaf block */ 586 l = (zap_leaf_phys_t *)stack; 587 stack += 1<<blksft; 588 if (errnum = dmu_read(zap_dnode, blkid, l, stack)) 589 return (errnum); 590 591 return (zap_leaf_lookup(l, blksft, hash, name, value)); 592 } 593 594 /* 595 * Read in the data of a zap object and find the value for a matching 596 * property name. 597 * 598 * Return: 599 * 0 - success 600 * errnum - failure 601 */ 602 static int 603 zap_lookup(dnode_phys_t *zap_dnode, char *name, uint64_t *val, char *stack) 604 { 605 uint64_t block_type; 606 int size; 607 void *zapbuf; 608 609 /* Read in the first block of the zap object data. */ 610 zapbuf = stack; 611 size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 612 stack += size; 613 if (errnum = dmu_read(zap_dnode, 0, zapbuf, stack)) 614 return (errnum); 615 616 block_type = *((uint64_t *)zapbuf); 617 618 if (block_type == ZBT_MICRO) { 619 return (mzap_lookup(zapbuf, size, name, val)); 620 } else if (block_type == ZBT_HEADER) { 621 /* this is a fat zap */ 622 return (fzap_lookup(zap_dnode, zapbuf, name, 623 val, stack)); 624 } 625 626 return (ERR_FSYS_CORRUPT); 627 } 628 629 /* 630 * Get the dnode of an object number from the metadnode of an object set. 631 * 632 * Input 633 * mdn - metadnode to get the object dnode 634 * objnum - object number for the object dnode 635 * buf - data buffer that holds the returning dnode 636 * stack - scratch area 637 * 638 * Return: 639 * 0 - success 640 * errnum - failure 641 */ 642 static int 643 dnode_get(dnode_phys_t *mdn, uint64_t objnum, uint8_t type, dnode_phys_t *buf, 644 char *stack) 645 { 646 uint64_t blkid, blksz; /* the block id this object dnode is in */ 647 int epbs; /* shift of number of dnodes in a block */ 648 int idx; /* index within a block */ 649 dnode_phys_t *dnbuf; 650 651 blksz = mdn->dn_datablkszsec << SPA_MINBLOCKSHIFT; 652 epbs = zfs_log2(blksz) - DNODE_SHIFT; 653 blkid = objnum >> epbs; 654 idx = objnum & ((1<<epbs)-1); 655 656 if (dnode_buf != NULL && dnode_mdn == mdn && 657 objnum >= dnode_start && objnum < dnode_end) { 658 grub_memmove(buf, &dnode_buf[idx], DNODE_SIZE); 659 VERIFY_DN_TYPE(buf, type); 660 return (0); 661 } 662 663 if (dnode_buf && blksz == 1<<DNODE_BLOCK_SHIFT) { 664 dnbuf = dnode_buf; 665 dnode_mdn = mdn; 666 dnode_start = blkid << epbs; 667 dnode_end = (blkid + 1) << epbs; 668 } else { 669 dnbuf = (dnode_phys_t *)stack; 670 stack += blksz; 671 } 672 673 if (errnum = dmu_read(mdn, blkid, (char *)dnbuf, stack)) 674 return (errnum); 675 676 grub_memmove(buf, &dnbuf[idx], DNODE_SIZE); 677 VERIFY_DN_TYPE(buf, type); 678 679 return (0); 680 } 681 682 /* 683 * Check if this is a special file that resides at the top 684 * dataset of the pool. Currently this is the GRUB menu, 685 * boot signature and boot signature backup. 686 * str starts with '/'. 687 */ 688 static int 689 is_top_dataset_file(char *str) 690 { 691 char *tptr; 692 693 if ((tptr = grub_strstr(str, "menu.lst")) && 694 (tptr[8] == '\0' || tptr[8] == ' ') && 695 *(tptr-1) == '/') 696 return (1); 697 698 if (grub_strncmp(str, BOOTSIGN_DIR"/", 699 grub_strlen(BOOTSIGN_DIR) + 1) == 0) 700 return (1); 701 702 if (grub_strcmp(str, BOOTSIGN_BACKUP) == 0) 703 return (1); 704 705 return (0); 706 } 707 708 /* 709 * Get the file dnode for a given file name where mdn is the meta dnode 710 * for this ZFS object set. When found, place the file dnode in dn. 711 * The 'path' argument will be mangled. 712 * 713 * Return: 714 * 0 - success 715 * errnum - failure 716 */ 717 static int 718 dnode_get_path(dnode_phys_t *mdn, char *path, dnode_phys_t *dn, 719 char *stack) 720 { 721 uint64_t objnum, version; 722 char *cname, ch; 723 724 if (errnum = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE, 725 dn, stack)) 726 return (errnum); 727 728 if (errnum = zap_lookup(dn, ZPL_VERSION_STR, &version, stack)) 729 return (errnum); 730 if (version > ZPL_VERSION) 731 return (-1); 732 733 if (errnum = zap_lookup(dn, ZFS_ROOT_OBJ, &objnum, stack)) 734 return (errnum); 735 736 if (errnum = dnode_get(mdn, objnum, DMU_OT_DIRECTORY_CONTENTS, 737 dn, stack)) 738 return (errnum); 739 740 /* skip leading slashes */ 741 while (*path == '/') 742 path++; 743 744 while (*path && !isspace(*path)) { 745 746 /* get the next component name */ 747 cname = path; 748 while (*path && !isspace(*path) && *path != '/') 749 path++; 750 ch = *path; 751 *path = 0; /* ensure null termination */ 752 753 if (errnum = zap_lookup(dn, cname, &objnum, stack)) 754 return (errnum); 755 756 objnum = ZFS_DIRENT_OBJ(objnum); 757 if (errnum = dnode_get(mdn, objnum, 0, dn, stack)) 758 return (errnum); 759 760 *path = ch; 761 while (*path == '/') 762 path++; 763 } 764 765 /* We found the dnode for this file. Verify if it is a plain file. */ 766 VERIFY_DN_TYPE(dn, DMU_OT_PLAIN_FILE_CONTENTS); 767 768 return (0); 769 } 770 771 /* 772 * Get the default 'bootfs' property value from the rootpool. 773 * 774 * Return: 775 * 0 - success 776 * errnum -failure 777 */ 778 static int 779 get_default_bootfsobj(dnode_phys_t *mosmdn, uint64_t *obj, char *stack) 780 { 781 uint64_t objnum = 0; 782 dnode_phys_t *dn = (dnode_phys_t *)stack; 783 stack += DNODE_SIZE; 784 785 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 786 DMU_OT_OBJECT_DIRECTORY, dn, stack)) 787 return (errnum); 788 789 /* 790 * find the object number for 'pool_props', and get the dnode 791 * of the 'pool_props'. 792 */ 793 if (zap_lookup(dn, DMU_POOL_PROPS, &objnum, stack)) 794 return (ERR_FILESYSTEM_NOT_FOUND); 795 796 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_POOL_PROPS, dn, stack)) 797 return (errnum); 798 799 if (zap_lookup(dn, ZPOOL_PROP_BOOTFS, &objnum, stack)) 800 return (ERR_FILESYSTEM_NOT_FOUND); 801 802 if (!objnum) 803 return (ERR_FILESYSTEM_NOT_FOUND); 804 805 *obj = objnum; 806 return (0); 807 } 808 809 /* 810 * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname), 811 * e.g. pool/rootfs, or a given object number (obj), e.g. the object number 812 * of pool/rootfs. 813 * 814 * If no fsname and no obj are given, return the DSL_DIR metadnode. 815 * If fsname is given, return its metadnode and its matching object number. 816 * If only obj is given, return the metadnode for this object number. 817 * 818 * Return: 819 * 0 - success 820 * errnum - failure 821 */ 822 static int 823 get_objset_mdn(dnode_phys_t *mosmdn, char *fsname, uint64_t *obj, 824 dnode_phys_t *mdn, char *stack) 825 { 826 uint64_t objnum, headobj; 827 char *cname, ch; 828 blkptr_t *bp; 829 objset_phys_t *osp; 830 831 if (fsname == NULL && obj) { 832 headobj = *obj; 833 goto skip; 834 } 835 836 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 837 DMU_OT_OBJECT_DIRECTORY, mdn, stack)) 838 return (errnum); 839 840 if (errnum = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum, 841 stack)) 842 return (errnum); 843 844 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, stack)) 845 return (errnum); 846 847 if (fsname == NULL) { 848 headobj = 849 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 850 goto skip; 851 } 852 853 /* take out the pool name */ 854 while (*fsname && !isspace(*fsname) && *fsname != '/') 855 fsname++; 856 857 while (*fsname && !isspace(*fsname)) { 858 uint64_t childobj; 859 860 while (*fsname == '/') 861 fsname++; 862 863 cname = fsname; 864 while (*fsname && !isspace(*fsname) && *fsname != '/') 865 fsname++; 866 ch = *fsname; 867 *fsname = 0; 868 869 childobj = 870 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_child_dir_zapobj; 871 if (errnum = dnode_get(mosmdn, childobj, 872 DMU_OT_DSL_DIR_CHILD_MAP, mdn, stack)) 873 return (errnum); 874 875 if (zap_lookup(mdn, cname, &objnum, stack)) 876 return (ERR_FILESYSTEM_NOT_FOUND); 877 878 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, 879 mdn, stack)) 880 return (errnum); 881 882 *fsname = ch; 883 } 884 headobj = ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 885 if (obj) 886 *obj = headobj; 887 888 skip: 889 if (errnum = dnode_get(mosmdn, headobj, DMU_OT_DSL_DATASET, mdn, stack)) 890 return (errnum); 891 892 /* TODO: Add snapshot support here - for fsname=snapshot-name */ 893 894 bp = &((dsl_dataset_phys_t *)DN_BONUS(mdn))->ds_bp; 895 osp = (objset_phys_t *)stack; 896 stack += sizeof (objset_phys_t); 897 if (errnum = zio_read(bp, osp, stack)) 898 return (errnum); 899 900 grub_memmove((char *)mdn, (char *)&osp->os_meta_dnode, DNODE_SIZE); 901 902 return (0); 903 } 904 905 /* 906 * For a given XDR packed nvlist, verify the first 4 bytes and move on. 907 * 908 * An XDR packed nvlist is encoded as (comments from nvs_xdr_create) : 909 * 910 * encoding method/host endian (4 bytes) 911 * nvl_version (4 bytes) 912 * nvl_nvflag (4 bytes) 913 * encoded nvpairs: 914 * encoded size of the nvpair (4 bytes) 915 * decoded size of the nvpair (4 bytes) 916 * name string size (4 bytes) 917 * name string data (sizeof(NV_ALIGN4(string)) 918 * data type (4 bytes) 919 * # of elements in the nvpair (4 bytes) 920 * data 921 * 2 zero's for the last nvpair 922 * (end of the entire list) (8 bytes) 923 * 924 * Return: 925 * 0 - success 926 * 1 - failure 927 */ 928 static int 929 nvlist_unpack(char *nvlist, char **out) 930 { 931 /* Verify if the 1st and 2nd byte in the nvlist are valid. */ 932 if (nvlist[0] != NV_ENCODE_XDR || nvlist[1] != HOST_ENDIAN) 933 return (1); 934 935 nvlist += 4; 936 *out = nvlist; 937 return (0); 938 } 939 940 static char * 941 nvlist_array(char *nvlist, int index) 942 { 943 int i, encode_size; 944 945 for (i = 0; i < index; i++) { 946 /* skip the header, nvl_version, and nvl_nvflag */ 947 nvlist = nvlist + 4 * 2; 948 949 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) 950 nvlist += encode_size; /* goto the next nvpair */ 951 952 nvlist = nvlist + 4 * 2; /* skip the ending 2 zeros - 8 bytes */ 953 } 954 955 return (nvlist); 956 } 957 958 static int 959 nvlist_lookup_value(char *nvlist, char *name, void *val, int valtype, 960 int *nelmp) 961 { 962 int name_len, type, slen, encode_size; 963 char *nvpair, *nvp_name, *strval = val; 964 uint64_t *intval = val; 965 966 /* skip the header, nvl_version, and nvl_nvflag */ 967 nvlist = nvlist + 4 * 2; 968 969 /* 970 * Loop thru the nvpair list 971 * The XDR representation of an integer is in big-endian byte order. 972 */ 973 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) { 974 975 nvpair = nvlist + 4 * 2; /* skip the encode/decode size */ 976 977 name_len = BSWAP_32(*(uint32_t *)nvpair); 978 nvpair += 4; 979 980 nvp_name = nvpair; 981 nvpair = nvpair + ((name_len + 3) & ~3); /* align */ 982 983 type = BSWAP_32(*(uint32_t *)nvpair); 984 nvpair += 4; 985 986 if ((grub_strncmp(nvp_name, name, name_len) == 0) && 987 type == valtype) { 988 int nelm; 989 990 if ((nelm = BSWAP_32(*(uint32_t *)nvpair)) < 1) 991 return (1); 992 nvpair += 4; 993 994 switch (valtype) { 995 case DATA_TYPE_STRING: 996 slen = BSWAP_32(*(uint32_t *)nvpair); 997 nvpair += 4; 998 grub_memmove(strval, nvpair, slen); 999 strval[slen] = '\0'; 1000 return (0); 1001 1002 case DATA_TYPE_UINT64: 1003 *intval = BSWAP_64(*(uint64_t *)nvpair); 1004 return (0); 1005 1006 case DATA_TYPE_NVLIST: 1007 *(void **)val = (void *)nvpair; 1008 return (0); 1009 1010 case DATA_TYPE_NVLIST_ARRAY: 1011 *(void **)val = (void *)nvpair; 1012 if (nelmp) 1013 *nelmp = nelm; 1014 return (0); 1015 } 1016 } 1017 1018 nvlist += encode_size; /* goto the next nvpair */ 1019 } 1020 1021 return (1); 1022 } 1023 1024 /* 1025 * Check if this vdev is online and is in a good state. 1026 */ 1027 static int 1028 vdev_validate(char *nv) 1029 { 1030 uint64_t ival; 1031 1032 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_OFFLINE, &ival, 1033 DATA_TYPE_UINT64, NULL) == 0 || 1034 nvlist_lookup_value(nv, ZPOOL_CONFIG_FAULTED, &ival, 1035 DATA_TYPE_UINT64, NULL) == 0 || 1036 nvlist_lookup_value(nv, ZPOOL_CONFIG_DEGRADED, &ival, 1037 DATA_TYPE_UINT64, NULL) == 0 || 1038 nvlist_lookup_value(nv, ZPOOL_CONFIG_REMOVED, &ival, 1039 DATA_TYPE_UINT64, NULL) == 0) 1040 return (ERR_DEV_VALUES); 1041 1042 return (0); 1043 } 1044 1045 /* 1046 * Get a list of valid vdev pathname from the boot device. 1047 * The caller should already allocate MAXNAMELEN memory for bootpath. 1048 */ 1049 static int 1050 vdev_get_bootpath(char *nv, char *bootpath) 1051 { 1052 char type[16]; 1053 1054 bootpath[0] = '\0'; 1055 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_TYPE, &type, DATA_TYPE_STRING, 1056 NULL)) 1057 return (ERR_FSYS_CORRUPT); 1058 1059 if (strcmp(type, VDEV_TYPE_DISK) == 0) { 1060 if (vdev_validate(nv) != 0 || 1061 nvlist_lookup_value(nv, ZPOOL_CONFIG_PHYS_PATH, bootpath, 1062 DATA_TYPE_STRING, NULL) != 0) 1063 return (ERR_NO_BOOTPATH); 1064 1065 } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0) { 1066 int nelm, i; 1067 char *child; 1068 1069 if (nvlist_lookup_value(nv, ZPOOL_CONFIG_CHILDREN, &child, 1070 DATA_TYPE_NVLIST_ARRAY, &nelm)) 1071 return (ERR_FSYS_CORRUPT); 1072 1073 for (i = 0; i < nelm; i++) { 1074 char tmp_path[MAXNAMELEN]; 1075 char *child_i; 1076 1077 child_i = nvlist_array(child, i); 1078 if (vdev_validate(child_i) != 0) 1079 continue; 1080 1081 if (nvlist_lookup_value(child_i, ZPOOL_CONFIG_PHYS_PATH, 1082 tmp_path, DATA_TYPE_STRING, NULL) != 0) 1083 return (ERR_NO_BOOTPATH); 1084 1085 if ((strlen(bootpath) + strlen(tmp_path)) > MAXNAMELEN) 1086 return (ERR_WONT_FIT); 1087 1088 if (strlen(bootpath) == 0) 1089 sprintf(bootpath, "%s", tmp_path); 1090 else 1091 sprintf(bootpath, "%s %s", bootpath, tmp_path); 1092 } 1093 } 1094 1095 return (strlen(bootpath) > 0 ? 0 : ERR_NO_BOOTPATH); 1096 } 1097 1098 /* 1099 * Check the disk label information and retrieve needed vdev name-value pairs. 1100 * 1101 * Return: 1102 * 0 - success 1103 * ERR_* - failure 1104 */ 1105 static int 1106 check_pool_label(int label, char *stack) 1107 { 1108 vdev_phys_t *vdev; 1109 uint64_t sector, pool_state, txg = 0; 1110 char *nvlist, *nv; 1111 1112 sector = (label * sizeof (vdev_label_t) + VDEV_SKIP_SIZE + 1113 VDEV_BOOT_HEADER_SIZE) >> SPA_MINBLOCKSHIFT; 1114 1115 /* Read in the vdev name-value pair list (112K). */ 1116 if (devread(sector, 0, VDEV_PHYS_SIZE, stack) == 0) 1117 return (ERR_READ); 1118 1119 vdev = (vdev_phys_t *)stack; 1120 1121 if (nvlist_unpack(vdev->vp_nvlist, &nvlist)) 1122 return (ERR_FSYS_CORRUPT); 1123 1124 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_STATE, &pool_state, 1125 DATA_TYPE_UINT64, NULL)) 1126 return (ERR_FSYS_CORRUPT); 1127 1128 if (pool_state == POOL_STATE_DESTROYED) 1129 return (ERR_FILESYSTEM_NOT_FOUND); 1130 1131 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_NAME, 1132 current_rootpool, DATA_TYPE_STRING, NULL)) 1133 return (ERR_FSYS_CORRUPT); 1134 1135 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_TXG, &txg, 1136 DATA_TYPE_UINT64, NULL)) 1137 return (ERR_FSYS_CORRUPT); 1138 1139 /* not an active device */ 1140 if (txg == 0) 1141 return (ERR_NO_BOOTPATH); 1142 1143 if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VDEV_TREE, &nv, 1144 DATA_TYPE_NVLIST, NULL)) 1145 return (ERR_FSYS_CORRUPT); 1146 1147 if (vdev_get_bootpath(nv, current_bootpath)) 1148 return (ERR_NO_BOOTPATH); 1149 1150 return (0); 1151 } 1152 1153 /* 1154 * zfs_mount() locates a valid uberblock of the root pool and read in its MOS 1155 * to the memory address MOS. 1156 * 1157 * Return: 1158 * 1 - success 1159 * 0 - failure 1160 */ 1161 int 1162 zfs_mount(void) 1163 { 1164 char *stack; 1165 int label = 0; 1166 uberblock_phys_t *ub_array, *ubbest = NULL; 1167 vdev_boot_header_t *bh; 1168 objset_phys_t *osp; 1169 1170 stackbase = ZFS_SCRATCH; 1171 stack = stackbase; 1172 ub_array = (uberblock_phys_t *)stack; 1173 stack += VDEV_UBERBLOCK_RING; 1174 1175 bh = (vdev_boot_header_t *)stack; 1176 stack += VDEV_BOOT_HEADER_SIZE; 1177 1178 osp = (objset_phys_t *)stack; 1179 stack += sizeof (objset_phys_t); 1180 1181 /* XXX add back labels support? */ 1182 for (label = 0; ubbest == NULL && label < (VDEV_LABELS/2); label++) { 1183 uint64_t sector = (label * sizeof (vdev_label_t) + 1184 VDEV_SKIP_SIZE) >> SPA_MINBLOCKSHIFT; 1185 if (devread(sector, 0, VDEV_BOOT_HEADER_SIZE, 1186 (char *)bh) == 0) 1187 continue; 1188 if ((bh->vb_magic != VDEV_BOOT_MAGIC) || 1189 (bh->vb_version != VDEV_BOOT_VERSION)) { 1190 continue; 1191 } 1192 sector += (VDEV_BOOT_HEADER_SIZE + 1193 VDEV_PHYS_SIZE) >> SPA_MINBLOCKSHIFT; 1194 1195 /* Read in the uberblock ring (128K). */ 1196 if (devread(sector, 0, VDEV_UBERBLOCK_RING, 1197 (char *)ub_array) == 0) 1198 continue; 1199 1200 if ((ubbest = find_bestub(ub_array, label)) != NULL && 1201 zio_read(&ubbest->ubp_uberblock.ub_rootbp, osp, stack) 1202 == 0) { 1203 1204 VERIFY_OS_TYPE(osp, DMU_OST_META); 1205 1206 /* Got the MOS. Save it at the memory addr MOS. */ 1207 grub_memmove(MOS, &osp->os_meta_dnode, DNODE_SIZE); 1208 1209 if (check_pool_label(label, stack)) 1210 return (0); 1211 1212 is_zfs_mount = 1; 1213 return (1); 1214 } 1215 } 1216 1217 return (0); 1218 } 1219 1220 /* 1221 * zfs_open() locates a file in the rootpool by following the 1222 * MOS and places the dnode of the file in the memory address DNODE. 1223 * 1224 * Return: 1225 * 1 - success 1226 * 0 - failure 1227 */ 1228 int 1229 zfs_open(char *filename) 1230 { 1231 char *stack; 1232 dnode_phys_t *mdn; 1233 1234 file_buf = NULL; 1235 stackbase = ZFS_SCRATCH; 1236 stack = stackbase; 1237 1238 mdn = (dnode_phys_t *)stack; 1239 stack += sizeof (dnode_phys_t); 1240 1241 dnode_mdn = NULL; 1242 dnode_buf = (dnode_phys_t *)stack; 1243 stack += 1<<DNODE_BLOCK_SHIFT; 1244 1245 /* 1246 * menu.lst is placed at the root pool filesystem level, 1247 * do not goto 'current_bootfs'. 1248 */ 1249 if (is_top_dataset_file(filename)) { 1250 if (errnum = get_objset_mdn(MOS, NULL, NULL, mdn, stack)) 1251 return (0); 1252 1253 current_bootfs_obj = 0; 1254 } else { 1255 if (current_bootfs[0] == '\0') { 1256 /* Get the default root filesystem object number */ 1257 if (errnum = get_default_bootfsobj(MOS, 1258 ¤t_bootfs_obj, stack)) 1259 return (0); 1260 1261 if (errnum = get_objset_mdn(MOS, NULL, 1262 ¤t_bootfs_obj, mdn, stack)) 1263 return (0); 1264 } else { 1265 if (errnum = get_objset_mdn(MOS, current_bootfs, 1266 ¤t_bootfs_obj, mdn, stack)) { 1267 memset(current_bootfs, 0, MAXNAMELEN); 1268 return (0); 1269 } 1270 } 1271 } 1272 1273 if (dnode_get_path(mdn, filename, DNODE, stack)) { 1274 errnum = ERR_FILE_NOT_FOUND; 1275 return (0); 1276 } 1277 1278 /* get the file size and set the file position to 0 */ 1279 filemax = ((znode_phys_t *)DN_BONUS(DNODE))->zp_size; 1280 filepos = 0; 1281 1282 dnode_buf = NULL; 1283 return (1); 1284 } 1285 1286 /* 1287 * zfs_read reads in the data blocks pointed by the DNODE. 1288 * 1289 * Return: 1290 * len - the length successfully read in to the buffer 1291 * 0 - failure 1292 */ 1293 int 1294 zfs_read(char *buf, int len) 1295 { 1296 char *stack; 1297 char *tmpbuf; 1298 int blksz, length, movesize; 1299 1300 if (file_buf == NULL) { 1301 file_buf = stackbase; 1302 stackbase += SPA_MAXBLOCKSIZE; 1303 file_start = file_end = 0; 1304 } 1305 stack = stackbase; 1306 1307 /* 1308 * If offset is in memory, move it into the buffer provided and return. 1309 */ 1310 if (filepos >= file_start && filepos+len <= file_end) { 1311 grub_memmove(buf, file_buf + filepos - file_start, len); 1312 filepos += len; 1313 return (len); 1314 } 1315 1316 blksz = DNODE->dn_datablkszsec << SPA_MINBLOCKSHIFT; 1317 1318 /* 1319 * Entire Dnode is too big to fit into the space available. We 1320 * will need to read it in chunks. This could be optimized to 1321 * read in as large a chunk as there is space available, but for 1322 * now, this only reads in one data block at a time. 1323 */ 1324 length = len; 1325 while (length) { 1326 /* 1327 * Find requested blkid and the offset within that block. 1328 */ 1329 uint64_t blkid = filepos / blksz; 1330 1331 if (errnum = dmu_read(DNODE, blkid, file_buf, stack)) 1332 return (0); 1333 1334 file_start = blkid * blksz; 1335 file_end = file_start + blksz; 1336 1337 movesize = MIN(length, file_end - filepos); 1338 1339 grub_memmove(buf, file_buf + filepos - file_start, 1340 movesize); 1341 buf += movesize; 1342 length -= movesize; 1343 filepos += movesize; 1344 } 1345 1346 return (len); 1347 } 1348 1349 /* 1350 * No-Op 1351 */ 1352 int 1353 zfs_embed(int *start_sector, int needed_sectors) 1354 { 1355 return (1); 1356 } 1357 1358 #endif /* FSYS_ZFS */ 1359