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 2007 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 {"noop", 0}, 68 {"on", lzjb_decompress}, /* ZIO_COMPRESS_ON */ 69 {"off", 0}, 70 {"lzjb", lzjb_decompress} /* ZIO_COMPRESS_LZJB */ 71 }; 72 73 /* 74 * Our own version of bcmp(). 75 */ 76 static int 77 zfs_bcmp(const void *s1, const void *s2, size_t n) 78 { 79 const uchar_t *ps1 = s1; 80 const uchar_t *ps2 = s2; 81 82 if (s1 != s2 && n != 0) { 83 do { 84 if (*ps1++ != *ps2++) 85 return (1); 86 } while (--n != 0); 87 } 88 89 return (0); 90 } 91 92 /* 93 * Our own version of log2(). Same thing as highbit()-1. 94 */ 95 static int 96 zfs_log2(uint64_t num) 97 { 98 int i = 0; 99 100 while (num > 1) { 101 i++; 102 num = num >> 1; 103 } 104 105 return (i); 106 } 107 108 /* Checksum Functions */ 109 static void 110 zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp) 111 { 112 ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0); 113 } 114 115 /* Checksum Table and Values */ 116 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = { 117 NULL, NULL, 0, 0, "inherit", 118 NULL, NULL, 0, 0, "on", 119 zio_checksum_off, zio_checksum_off, 0, 0, "off", 120 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "label", 121 zio_checksum_SHA256, zio_checksum_SHA256, 1, 1, "gang_header", 122 fletcher_2_native, fletcher_2_byteswap, 0, 1, "zilog", 123 fletcher_2_native, fletcher_2_byteswap, 0, 0, "fletcher2", 124 fletcher_4_native, fletcher_4_byteswap, 1, 0, "fletcher4", 125 zio_checksum_SHA256, zio_checksum_SHA256, 1, 0, "SHA256", 126 }; 127 128 /* 129 * zio_checksum_verify: Provides support for checksum verification. 130 * 131 * Fletcher2, Fletcher4, and SHA256 are supported. 132 * 133 * Return: 134 * -1 = Failure 135 * 0 = Success 136 */ 137 static int 138 zio_checksum_verify(blkptr_t *bp, char *data, int size) 139 { 140 zio_cksum_t zc = bp->blk_cksum; 141 uint32_t checksum = BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : 142 BP_GET_CHECKSUM(bp); 143 int byteswap = BP_SHOULD_BYTESWAP(bp); 144 zio_block_tail_t *zbt = (zio_block_tail_t *)(data + size) - 1; 145 zio_checksum_info_t *ci = &zio_checksum_table[checksum]; 146 zio_cksum_t actual_cksum, expected_cksum; 147 148 /* byteswap is not supported */ 149 if (byteswap) 150 return (-1); 151 152 if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL) 153 return (-1); 154 155 if (ci->ci_zbt) { 156 if (checksum == ZIO_CHECKSUM_GANG_HEADER) { 157 /* 158 * 'gang blocks' is not supported. 159 */ 160 return (-1); 161 } 162 163 if (zbt->zbt_magic == BSWAP_64(ZBT_MAGIC)) { 164 /* byte swapping is not supported */ 165 return (-1); 166 } else { 167 expected_cksum = zbt->zbt_cksum; 168 zbt->zbt_cksum = zc; 169 ci->ci_func[0](data, size, &actual_cksum); 170 zbt->zbt_cksum = expected_cksum; 171 } 172 zc = expected_cksum; 173 174 } else { 175 if (BP_IS_GANG(bp)) 176 return (-1); 177 ci->ci_func[byteswap](data, size, &actual_cksum); 178 } 179 180 if ((actual_cksum.zc_word[0] - zc.zc_word[0]) | 181 (actual_cksum.zc_word[1] - zc.zc_word[1]) | 182 (actual_cksum.zc_word[2] - zc.zc_word[2]) | 183 (actual_cksum.zc_word[3] - zc.zc_word[3])) 184 return (-1); 185 186 return (0); 187 } 188 189 /* 190 * vdev_label_offset takes "offset" (the offset within a vdev_label) and 191 * returns its physical disk offset (starting from the beginning of the vdev). 192 * 193 * Input: 194 * psize : Physical size of this vdev 195 * l : Label Number (0-3) 196 * offset : The offset with a vdev_label in which we want the physical 197 * address 198 * Return: 199 * Success : physical disk offset 200 * Failure : errnum = ERR_BAD_ARGUMENT, return value is meaningless 201 */ 202 uint64_t 203 vdev_label_offset(uint64_t psize, int l, uint64_t offset) 204 { 205 /* XXX Need to add back label support! */ 206 if (l >= VDEV_LABELS/2 || offset > sizeof (vdev_label_t)) { 207 errnum = ERR_BAD_ARGUMENT; 208 return (0); 209 } 210 211 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 212 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 213 214 } 215 216 /* 217 * vdev_uberblock_compare takes two uberblock structures and returns an integer 218 * indicating the more recent of the two. 219 * Return Value = 1 if ub2 is more recent 220 * Return Value = -1 if ub1 is more recent 221 * The most recent uberblock is determined using its transaction number and 222 * timestamp. The uberblock with the highest transaction number is 223 * considered "newer". If the transaction numbers of the two blocks match, the 224 * timestamps are compared to determine the "newer" of the two. 225 */ 226 static int 227 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 228 { 229 if (ub1->ub_txg < ub2->ub_txg) 230 return (-1); 231 if (ub1->ub_txg > ub2->ub_txg) 232 return (1); 233 234 if (ub1->ub_timestamp < ub2->ub_timestamp) 235 return (-1); 236 if (ub1->ub_timestamp > ub2->ub_timestamp) 237 return (1); 238 239 return (0); 240 } 241 242 /* 243 * Three pieces of information are needed to verify an uberblock: the magic 244 * number, the version number, and the checksum. 245 * 246 * Currently Implemented: version number, magic number 247 * Need to Implement: checksum 248 * 249 * Return: 250 * 0 - Success 251 * -1 - Failure 252 */ 253 static int 254 uberblock_verify(uberblock_phys_t *ub, int offset) 255 { 256 257 uberblock_t *uber = &ub->ubp_uberblock; 258 blkptr_t bp; 259 260 BP_ZERO(&bp); 261 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL); 262 BP_SET_BYTEORDER(&bp, ZFS_HOST_BYTEORDER); 263 ZIO_SET_CHECKSUM(&bp.blk_cksum, offset, 0, 0, 0); 264 265 if (zio_checksum_verify(&bp, (char *)ub, UBERBLOCK_SIZE) != 0) 266 return (-1); 267 268 if (uber->ub_magic == UBERBLOCK_MAGIC && 269 uber->ub_version >= SPA_VERSION_1 && 270 uber->ub_version <= SPA_VERSION) 271 return (0); 272 273 return (-1); 274 } 275 276 /* 277 * Find the best uberblock. 278 * Return: 279 * Success - Pointer to the best uberblock. 280 * Failure - NULL 281 */ 282 static uberblock_phys_t * 283 find_bestub(uberblock_phys_t *ub_array, int label) 284 { 285 uberblock_phys_t *ubbest = NULL; 286 int i, offset; 287 288 for (i = 0; i < (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT); i++) { 289 offset = vdev_label_offset(0, label, VDEV_UBERBLOCK_OFFSET(i)); 290 if (errnum == ERR_BAD_ARGUMENT) 291 return (NULL); 292 if (uberblock_verify(&ub_array[i], offset) == 0) { 293 if (ubbest == NULL) { 294 ubbest = &ub_array[i]; 295 } else if (vdev_uberblock_compare( 296 &(ub_array[i].ubp_uberblock), 297 &(ubbest->ubp_uberblock)) > 0) { 298 ubbest = &ub_array[i]; 299 } 300 } 301 } 302 303 return (ubbest); 304 } 305 306 /* 307 * Read in a block and put its uncompressed data in buf. 308 * 309 * Return: 310 * 0 - success 311 * errnum - failure 312 */ 313 static int 314 zio_read(blkptr_t *bp, void *buf, char *stack) 315 { 316 uint64_t offset, sector; 317 int psize, lsize; 318 int i, comp, cksum; 319 320 psize = BP_GET_PSIZE(bp); 321 lsize = BP_GET_LSIZE(bp); 322 comp = BP_GET_COMPRESS(bp); 323 cksum = BP_GET_CHECKSUM(bp); 324 325 /* pick a good dva from the block pointer */ 326 for (i = 0; i < SPA_DVAS_PER_BP; i++) { 327 328 if (bp->blk_dva[i].dva_word[0] == 0 && 329 bp->blk_dva[i].dva_word[1] == 0) 330 continue; 331 332 /* read in a block */ 333 offset = DVA_GET_OFFSET(&bp->blk_dva[i]); 334 sector = DVA_OFFSET_TO_PHYS_SECTOR(offset); 335 336 if (comp != ZIO_COMPRESS_OFF) { 337 338 if (devread(sector, 0, psize, stack) == 0) 339 continue; 340 if (zio_checksum_verify(bp, stack, psize) != 0) 341 continue; 342 decomp_table[comp].decomp_func(stack, buf, psize, 343 lsize); 344 } else { 345 if (devread(sector, 0, psize, buf) == 0) 346 continue; 347 if (zio_checksum_verify(bp, buf, psize) != 0) 348 continue; 349 } 350 return (0); 351 } 352 353 return (ERR_FSYS_CORRUPT); 354 } 355 356 /* 357 * Get the block from a block id. 358 * push the block onto the stack. 359 * 360 * Return: 361 * 0 - success 362 * errnum - failure 363 */ 364 static int 365 dmu_read(dnode_phys_t *dn, uint64_t blkid, void *buf, char *stack) 366 { 367 int idx, level; 368 blkptr_t *bp_array = dn->dn_blkptr; 369 int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 370 blkptr_t *bp, *tmpbuf; 371 372 bp = (blkptr_t *)stack; 373 stack += sizeof (blkptr_t); 374 375 tmpbuf = (blkptr_t *)stack; 376 stack += 1<<dn->dn_indblkshift; 377 378 for (level = dn->dn_nlevels - 1; level >= 0; level--) { 379 idx = (blkid >> (epbs * level)) & ((1<<epbs)-1); 380 *bp = bp_array[idx]; 381 if (level == 0) 382 tmpbuf = buf; 383 if (errnum = zio_read(bp, tmpbuf, stack)) 384 return (errnum); 385 386 bp_array = tmpbuf; 387 } 388 389 return (0); 390 } 391 392 /* 393 * mzap_lookup: Looks up property described by "name" and returns the value 394 * in "value". 395 * 396 * Return: 397 * 0 - success 398 * errnum - failure 399 */ 400 static int 401 mzap_lookup(mzap_phys_t *zapobj, int objsize, char *name, 402 uint64_t *value) 403 { 404 int i, chunks; 405 mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk; 406 407 chunks = objsize/MZAP_ENT_LEN - 1; 408 for (i = 0; i < chunks; i++) { 409 if (grub_strcmp(mzap_ent[i].mze_name, name) == 0) { 410 *value = mzap_ent[i].mze_value; 411 return (0); 412 } 413 } 414 415 return (ERR_FSYS_CORRUPT); 416 } 417 418 static uint64_t 419 zap_hash(uint64_t salt, const char *name) 420 { 421 static uint64_t table[256]; 422 const uint8_t *cp; 423 uint8_t c; 424 uint64_t crc = salt; 425 426 if (table[128] == 0) { 427 uint64_t *ct; 428 int i, j; 429 for (i = 0; i < 256; i++) { 430 for (ct = table + i, *ct = i, j = 8; j > 0; j--) 431 *ct = (*ct >> 1) ^ (-(*ct & 1) & 432 ZFS_CRC64_POLY); 433 } 434 } 435 436 if (crc == 0 || table[128] != ZFS_CRC64_POLY) { 437 errnum = ERR_FSYS_CORRUPT; 438 return (0); 439 } 440 441 for (cp = (const uint8_t *)name; (c = *cp) != '\0'; cp++) 442 crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF]; 443 444 /* 445 * Only use 28 bits, since we need 4 bits in the cookie for the 446 * collision differentiator. We MUST use the high bits, since 447 * those are the onces that we first pay attention to when 448 * chosing the bucket. 449 */ 450 crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1); 451 452 return (crc); 453 } 454 455 /* 456 * Only to be used on 8-bit arrays. 457 * array_len is actual len in bytes (not encoded le_value_length). 458 * buf is null-terminated. 459 */ 460 static int 461 zap_leaf_array_equal(zap_leaf_phys_t *l, int blksft, int chunk, 462 int array_len, const char *buf) 463 { 464 int bseen = 0; 465 466 while (bseen < array_len) { 467 struct zap_leaf_array *la = 468 &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array; 469 int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES); 470 471 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 472 return (0); 473 474 if (zfs_bcmp(la->la_array, buf + bseen, toread) != 0) 475 break; 476 chunk = la->la_next; 477 bseen += toread; 478 } 479 return (bseen == array_len); 480 } 481 482 /* 483 * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the 484 * value for the property "name". 485 * 486 * Return: 487 * 0 - success 488 * errnum - failure 489 */ 490 int 491 zap_leaf_lookup(zap_leaf_phys_t *l, int blksft, uint64_t h, 492 const char *name, uint64_t *value) 493 { 494 uint16_t chunk; 495 struct zap_leaf_entry *le; 496 497 /* Verify if this is a valid leaf block */ 498 if (l->l_hdr.lh_block_type != ZBT_LEAF) 499 return (ERR_FSYS_CORRUPT); 500 if (l->l_hdr.lh_magic != ZAP_LEAF_MAGIC) 501 return (ERR_FSYS_CORRUPT); 502 503 for (chunk = l->l_hash[LEAF_HASH(blksft, h)]; 504 chunk != CHAIN_END; chunk = le->le_next) { 505 506 if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) 507 return (ERR_FSYS_CORRUPT); 508 509 le = ZAP_LEAF_ENTRY(l, blksft, chunk); 510 511 /* Verify the chunk entry */ 512 if (le->le_type != ZAP_CHUNK_ENTRY) 513 return (ERR_FSYS_CORRUPT); 514 515 if (le->le_hash != h) 516 continue; 517 518 if (zap_leaf_array_equal(l, blksft, le->le_name_chunk, 519 le->le_name_length, name)) { 520 521 struct zap_leaf_array *la; 522 uint8_t *ip; 523 524 if (le->le_int_size != 8 || le->le_value_length != 1) 525 return (ERR_FSYS_CORRUPT); 526 527 /* get the uint64_t property value */ 528 la = &ZAP_LEAF_CHUNK(l, blksft, 529 le->le_value_chunk).l_array; 530 ip = la->la_array; 531 532 *value = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 | 533 (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 | 534 (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 | 535 (uint64_t)ip[6] << 8 | (uint64_t)ip[7]; 536 537 return (0); 538 } 539 } 540 541 return (ERR_FSYS_CORRUPT); 542 } 543 544 /* 545 * Fat ZAP lookup 546 * 547 * Return: 548 * 0 - success 549 * errnum - failure 550 */ 551 int 552 fzap_lookup(dnode_phys_t *zap_dnode, zap_phys_t *zap, 553 char *name, uint64_t *value, char *stack) 554 { 555 zap_leaf_phys_t *l; 556 uint64_t hash, idx, blkid; 557 int blksft = zfs_log2(zap_dnode->dn_datablkszsec << DNODE_SHIFT); 558 559 /* Verify if this is a fat zap header block */ 560 if (zap->zap_magic != (uint64_t)ZAP_MAGIC) 561 return (ERR_FSYS_CORRUPT); 562 563 hash = zap_hash(zap->zap_salt, name); 564 if (errnum) 565 return (errnum); 566 567 /* get block id from index */ 568 if (zap->zap_ptrtbl.zt_numblks != 0) { 569 /* external pointer tables not supported */ 570 return (ERR_FSYS_CORRUPT); 571 } 572 idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift); 573 blkid = ((uint64_t *)zap)[idx + (1<<(blksft-3-1))]; 574 575 /* Get the leaf block */ 576 l = (zap_leaf_phys_t *)stack; 577 stack += 1<<blksft; 578 if (errnum = dmu_read(zap_dnode, blkid, l, stack)) 579 return (errnum); 580 581 return (zap_leaf_lookup(l, blksft, hash, name, value)); 582 } 583 584 /* 585 * Read in the data of a zap object and find the value for a matching 586 * property name. 587 * 588 * Return: 589 * 0 - success 590 * errnum - failure 591 */ 592 static int 593 zap_lookup(dnode_phys_t *zap_dnode, char *name, uint64_t *val, char *stack) 594 { 595 uint64_t block_type; 596 int size; 597 void *zapbuf; 598 599 /* Read in the first block of the zap object data. */ 600 zapbuf = stack; 601 size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT; 602 stack += size; 603 if (errnum = dmu_read(zap_dnode, 0, zapbuf, stack)) 604 return (errnum); 605 606 block_type = *((uint64_t *)zapbuf); 607 608 if (block_type == ZBT_MICRO) { 609 return (mzap_lookup(zapbuf, size, name, val)); 610 } else if (block_type == ZBT_HEADER) { 611 /* this is a fat zap */ 612 return (fzap_lookup(zap_dnode, zapbuf, name, 613 val, stack)); 614 } 615 616 return (ERR_FSYS_CORRUPT); 617 } 618 619 /* 620 * Get the dnode of an object number from the metadnode of an object set. 621 * 622 * Input 623 * mdn - metadnode to get the object dnode 624 * objnum - object number for the object dnode 625 * buf - data buffer that holds the returning dnode 626 * stack - scratch area 627 * 628 * Return: 629 * 0 - success 630 * errnum - failure 631 */ 632 static int 633 dnode_get(dnode_phys_t *mdn, uint64_t objnum, uint8_t type, dnode_phys_t *buf, 634 char *stack) 635 { 636 uint64_t blkid, blksz; /* the block id this object dnode is in */ 637 int epbs; /* shift of number of dnodes in a block */ 638 int idx; /* index within a block */ 639 dnode_phys_t *dnbuf; 640 641 blksz = mdn->dn_datablkszsec << SPA_MINBLOCKSHIFT; 642 epbs = zfs_log2(blksz) - DNODE_SHIFT; 643 blkid = objnum >> epbs; 644 idx = objnum & ((1<<epbs)-1); 645 646 if (dnode_buf != NULL && dnode_mdn == mdn && 647 objnum >= dnode_start && objnum < dnode_end) { 648 grub_memmove(buf, &dnode_buf[idx], DNODE_SIZE); 649 VERIFY_DN_TYPE(buf, type); 650 return (0); 651 } 652 653 if (dnode_buf && blksz == 1<<DNODE_BLOCK_SHIFT) { 654 dnbuf = dnode_buf; 655 dnode_mdn = mdn; 656 dnode_start = blkid << epbs; 657 dnode_end = (blkid + 1) << epbs; 658 } else { 659 dnbuf = (dnode_phys_t *)stack; 660 stack += blksz; 661 } 662 663 if (errnum = dmu_read(mdn, blkid, (char *)dnbuf, stack)) 664 return (errnum); 665 666 grub_memmove(buf, &dnbuf[idx], DNODE_SIZE); 667 VERIFY_DN_TYPE(buf, type); 668 669 return (0); 670 } 671 672 /* 673 * Check if this is the "menu.lst" file. 674 * str starts with '/'. 675 */ 676 static int 677 is_menu_lst(char *str) 678 { 679 char *tptr; 680 681 if ((tptr = grub_strstr(str, "menu.lst")) && 682 (tptr[8] == '\0' || tptr[8] == ' ') && 683 *(tptr-1) == '/') 684 return (1); 685 686 return (0); 687 } 688 689 /* 690 * Get the file dnode for a given file name where mdn is the meta dnode 691 * for this ZFS object set. When found, place the file dnode in dn. 692 * The 'path' argument will be mangled. 693 * 694 * Return: 695 * 0 - success 696 * errnum - failure 697 */ 698 static int 699 dnode_get_path(dnode_phys_t *mdn, char *path, dnode_phys_t *dn, 700 char *stack) 701 { 702 uint64_t objnum, version; 703 char *cname, ch; 704 705 if (errnum = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE, 706 dn, stack)) 707 return (errnum); 708 709 if (errnum = zap_lookup(dn, ZPL_VERSION_STR, &version, stack)) 710 return (errnum); 711 if (version > ZPL_VERSION) 712 return (-1); 713 714 if (errnum = zap_lookup(dn, ZFS_ROOT_OBJ, &objnum, stack)) 715 return (errnum); 716 717 if (errnum = dnode_get(mdn, objnum, DMU_OT_DIRECTORY_CONTENTS, 718 dn, stack)) 719 return (errnum); 720 721 /* skip leading slashes */ 722 while (*path == '/') 723 path++; 724 725 while (*path && !isspace(*path)) { 726 727 /* get the next component name */ 728 cname = path; 729 while (*path && !isspace(*path) && *path != '/') 730 path++; 731 ch = *path; 732 *path = 0; /* ensure null termination */ 733 734 if (errnum = zap_lookup(dn, cname, &objnum, stack)) 735 return (errnum); 736 737 objnum = ZFS_DIRENT_OBJ(objnum); 738 if (errnum = dnode_get(mdn, objnum, 0, dn, stack)) 739 return (errnum); 740 741 *path = ch; 742 while (*path == '/') 743 path++; 744 } 745 746 /* We found the dnode for this file. Verify if it is a plain file. */ 747 VERIFY_DN_TYPE(dn, DMU_OT_PLAIN_FILE_CONTENTS); 748 749 return (0); 750 } 751 752 /* 753 * Get the default 'bootfs' property value from the rootpool. 754 * 755 * Return: 756 * 0 - success 757 * errnum -failure 758 */ 759 static int 760 get_default_bootfsobj(dnode_phys_t *mosmdn, uint64_t *obj, char *stack) 761 { 762 uint64_t objnum = 0; 763 dnode_phys_t *dn = (dnode_phys_t *)stack; 764 stack += DNODE_SIZE; 765 766 if (dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 767 DMU_OT_OBJECT_DIRECTORY, dn, stack)) 768 return (ERR_FILESYSTEM_NOT_FOUND); 769 770 /* 771 * find the object number for 'pool_props', and get the dnode 772 * of the 'pool_props'. 773 */ 774 if (zap_lookup(dn, DMU_POOL_PROPS, &objnum, stack)) 775 return (ERR_FILESYSTEM_NOT_FOUND); 776 777 if (dnode_get(mosmdn, objnum, DMU_OT_POOL_PROPS, dn, stack)) 778 return (ERR_FILESYSTEM_NOT_FOUND); 779 780 if (zap_lookup(dn, ZPOOL_PROP_BOOTFS, &objnum, stack)) 781 return (ERR_FILESYSTEM_NOT_FOUND); 782 783 if (!objnum) 784 return (ERR_FILESYSTEM_NOT_FOUND); 785 786 *obj = objnum; 787 return (0); 788 } 789 790 /* 791 * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname), 792 * e.g. pool/rootfs, or a given object number (obj), e.g. the object number 793 * of pool/rootfs. 794 * 795 * If no fsname and no obj are given, return the DSL_DIR metadnode. 796 * If fsname is given, return its metadnode and its matching object number. 797 * If only obj is given, return the metadnode for this object number. 798 * 799 * Return: 800 * 0 - success 801 * errnum - failure 802 */ 803 static int 804 get_objset_mdn(dnode_phys_t *mosmdn, char *fsname, uint64_t *obj, 805 dnode_phys_t *mdn, char *stack) 806 { 807 uint64_t objnum, headobj; 808 char *cname, ch; 809 blkptr_t *bp; 810 objset_phys_t *osp; 811 812 if (fsname == NULL && obj) { 813 headobj = *obj; 814 goto skip; 815 } 816 817 if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT, 818 DMU_OT_OBJECT_DIRECTORY, mdn, stack)) 819 return (errnum); 820 821 if (errnum = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum, 822 stack)) 823 return (errnum); 824 825 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, stack)) 826 return (errnum); 827 828 if (fsname == NULL) { 829 headobj = 830 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 831 goto skip; 832 } 833 834 /* take out the pool name */ 835 while (*fsname && !isspace(*fsname) && *fsname != '/') 836 fsname++; 837 838 while (*fsname && !isspace(*fsname)) { 839 uint64_t childobj; 840 841 while (*fsname == '/') 842 fsname++; 843 844 cname = fsname; 845 while (*fsname && !isspace(*fsname) && *fsname != '/') 846 fsname++; 847 ch = *fsname; 848 *fsname = 0; 849 850 childobj = 851 ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_child_dir_zapobj; 852 if (errnum = dnode_get(mosmdn, childobj, 853 DMU_OT_DSL_DIR_CHILD_MAP, mdn, stack)) 854 return (errnum); 855 856 if (errnum = zap_lookup(mdn, cname, &objnum, stack)) 857 return (errnum); 858 859 if (errnum = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, 860 mdn, stack)) 861 return (errnum); 862 863 *fsname = ch; 864 } 865 headobj = ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj; 866 if (obj) 867 *obj = headobj; 868 869 skip: 870 if (errnum = dnode_get(mosmdn, headobj, DMU_OT_DSL_DATASET, mdn, stack)) 871 return (errnum); 872 873 /* TODO: Add snapshot support here - for fsname=snapshot-name */ 874 875 bp = &((dsl_dataset_phys_t *)DN_BONUS(mdn))->ds_bp; 876 osp = (objset_phys_t *)stack; 877 stack += sizeof (objset_phys_t); 878 if (errnum = zio_read(bp, osp, stack)) 879 return (errnum); 880 881 grub_memmove((char *)mdn, (char *)&osp->os_meta_dnode, DNODE_SIZE); 882 883 return (0); 884 } 885 886 /* 887 * Parse the packed nvlist and search for the string value of a given name. 888 * 889 * An XDR packed nvlist is encoded as (from nvs_xdr_create) : 890 * 891 * encoding method/host endian (4 bytes) 892 * nvl_version (4 bytes) 893 * nvl_nvflag (4 bytes) 894 * encoded nvpairs: 895 * encoded size of the nvpair (4 bytes) 896 * decoded size of the nvpair (4 bytes) 897 * name string size (4 bytes) 898 * name string data (sizeof(NV_ALIGN4(string)) 899 * data type (4 bytes) 900 * # of elements in the nvpair (4 bytes) 901 * data 902 * 2 zero's for the last nvpair 903 * (end of the entire list) (8 bytes) 904 * 905 * Return: 906 * 0 - success 907 * 1 - failure 908 */ 909 int 910 nvlist_lookup_value(char *nvlist, char *name, void *val, int valtype) 911 { 912 int name_len, type, nelm, slen, encode_size; 913 char *nvpair, *nvp_name, *strval = val; 914 uint64_t *intval = val; 915 916 /* Verify if the 1st and 2nd byte in the nvlist are valid. */ 917 if (nvlist[0] != NV_ENCODE_XDR || nvlist[1] != HOST_ENDIAN) 918 return (1); 919 920 /* skip the header, nvl_version, and nvl_nvflag */ 921 nvlist = nvlist + 4 * 3; 922 923 /* 924 * Loop thru the nvpair list 925 * The XDR representation of an integer is in big-endian byte order. 926 */ 927 while (encode_size = BSWAP_32(*(uint32_t *)nvlist)) { 928 929 nvpair = nvlist + 4 * 2; /* skip the encode/decode size */ 930 931 name_len = BSWAP_32(*(uint32_t *)nvpair); 932 nvpair += 4; 933 934 nvp_name = nvpair; 935 nvpair = nvpair + ((name_len + 3) & ~3); /* align */ 936 937 type = BSWAP_32(*(uint32_t *)nvpair); 938 nvpair += 4; 939 940 if ((grub_strncmp(nvp_name, name, name_len) == 0) && 941 type == valtype) { 942 943 if ((nelm = BSWAP_32(*(uint32_t *)nvpair)) != 1) 944 return (1); 945 nvpair += 4; 946 947 switch (valtype) { 948 case DATA_TYPE_STRING: 949 slen = BSWAP_32(*(uint32_t *)nvpair); 950 nvpair += 4; 951 grub_memmove(strval, nvpair, slen); 952 strval[slen] = '\0'; 953 return (0); 954 955 case DATA_TYPE_UINT64: 956 *intval = BSWAP_64(*(uint64_t *)nvpair); 957 return (0); 958 } 959 } 960 961 nvlist += encode_size; /* goto the next nvpair */ 962 } 963 964 return (1); 965 } 966 967 /* 968 * Get the pool name of the root pool from the vdev nvpair list of the label. 969 * 970 * Return: 971 * 0 - success 972 * errnum - failure 973 */ 974 int 975 get_pool_name_value(int label, char *name, void *value, int valtype, 976 char *stack) 977 { 978 vdev_phys_t *vdev; 979 uint64_t sector; 980 981 sector = (label * sizeof (vdev_label_t) + VDEV_SKIP_SIZE + 982 VDEV_BOOT_HEADER_SIZE) >> SPA_MINBLOCKSHIFT; 983 984 /* Read in the vdev name-value pair list (112K). */ 985 if (devread(sector, 0, VDEV_PHYS_SIZE, stack) == 0) 986 return (ERR_READ); 987 988 vdev = (vdev_phys_t *)stack; 989 990 if (nvlist_lookup_value(vdev->vp_nvlist, name, value, valtype)) 991 return (ERR_FSYS_CORRUPT); 992 else 993 return (0); 994 } 995 996 /* 997 * zfs_mount() locates a valid uberblock of the root pool and read in its MOS 998 * to the memory address MOS. 999 * 1000 * Return: 1001 * 1 - success 1002 * 0 - failure 1003 */ 1004 int 1005 zfs_mount(void) 1006 { 1007 char *stack; 1008 int label = 0; 1009 uberblock_phys_t *ub_array, *ubbest = NULL; 1010 objset_phys_t *osp; 1011 1012 /* if zfs is already mounted, don't do it again */ 1013 if (is_zfs_mount == 1) 1014 return (1); 1015 1016 stackbase = ZFS_SCRATCH; 1017 stack = stackbase; 1018 ub_array = (uberblock_phys_t *)stack; 1019 stack += VDEV_UBERBLOCK_RING; 1020 1021 osp = (objset_phys_t *)stack; 1022 stack += sizeof (objset_phys_t); 1023 1024 /* XXX add back labels support? */ 1025 for (label = 0; ubbest == NULL && label < (VDEV_LABELS/2); label++) { 1026 uint64_t sector = (label * sizeof (vdev_label_t) + 1027 VDEV_SKIP_SIZE + VDEV_BOOT_HEADER_SIZE + 1028 VDEV_PHYS_SIZE) >> SPA_MINBLOCKSHIFT; 1029 1030 /* Read in the uberblock ring (128K). */ 1031 if (devread(sector, 0, VDEV_UBERBLOCK_RING, 1032 (char *)ub_array) == 0) 1033 continue; 1034 1035 if ((ubbest = find_bestub(ub_array, label)) != NULL && 1036 zio_read(&ubbest->ubp_uberblock.ub_rootbp, osp, stack) 1037 == 0) { 1038 uint64_t pool_state; 1039 1040 VERIFY_OS_TYPE(osp, DMU_OST_META); 1041 1042 /* Got the MOS. Save it at the memory addr MOS. */ 1043 grub_memmove(MOS, &osp->os_meta_dnode, DNODE_SIZE); 1044 1045 if (get_pool_name_value(label, ZPOOL_CONFIG_POOL_STATE, 1046 &pool_state, DATA_TYPE_UINT64, stack)) 1047 return (0); 1048 1049 if (pool_state == POOL_STATE_DESTROYED) 1050 return (0); 1051 1052 if (get_pool_name_value(label, ZPOOL_CONFIG_POOL_NAME, 1053 current_rootpool, DATA_TYPE_STRING, stack)) 1054 return (0); 1055 1056 is_zfs_mount = 1; 1057 return (1); 1058 } 1059 } 1060 1061 return (0); 1062 } 1063 1064 /* 1065 * zfs_open() locates a file in the rootpool by following the 1066 * MOS and places the dnode of the file in the memory address DNODE. 1067 * 1068 * Return: 1069 * 1 - success 1070 * 0 - failure 1071 */ 1072 int 1073 zfs_open(char *filename) 1074 { 1075 char *stack; 1076 dnode_phys_t *mdn; 1077 1078 file_buf = NULL; 1079 stackbase = ZFS_SCRATCH; 1080 stack = stackbase; 1081 1082 mdn = (dnode_phys_t *)stack; 1083 stack += sizeof (dnode_phys_t); 1084 1085 dnode_mdn = NULL; 1086 dnode_buf = (dnode_phys_t *)stack; 1087 stack += 1<<DNODE_BLOCK_SHIFT; 1088 1089 /* 1090 * menu.lst is placed at the root pool filesystem level, 1091 * do not goto 'current_bootfs'. 1092 */ 1093 if (is_menu_lst(filename)) { 1094 if (errnum = get_objset_mdn(MOS, NULL, NULL, mdn, stack)) 1095 return (0); 1096 1097 current_bootfs_obj = 0; 1098 } else { 1099 if (current_bootfs[0] == '\0') { 1100 /* Get the default root filesystem object number */ 1101 if (get_default_bootfsobj(MOS, 1102 ¤t_bootfs_obj, stack)) { 1103 errnum = ERR_FILESYSTEM_NOT_FOUND; 1104 return (0); 1105 } 1106 1107 if (errnum = get_objset_mdn(MOS, NULL, 1108 ¤t_bootfs_obj, mdn, stack)) 1109 return (0); 1110 } else { 1111 if (errnum = get_objset_mdn(MOS, 1112 current_bootfs, ¤t_bootfs_obj, mdn, stack)) 1113 return (0); 1114 } 1115 } 1116 1117 if (dnode_get_path(mdn, filename, DNODE, stack)) { 1118 errnum = ERR_FILE_NOT_FOUND; 1119 return (0); 1120 } 1121 1122 /* get the file size and set the file position to 0 */ 1123 filemax = ((znode_phys_t *)DN_BONUS(DNODE))->zp_size; 1124 filepos = 0; 1125 1126 dnode_buf = NULL; 1127 return (1); 1128 } 1129 1130 /* 1131 * zfs_read reads in the data blocks pointed by the DNODE. 1132 * 1133 * Return: 1134 * len - the length successfully read in to the buffer 1135 * 0 - failure 1136 */ 1137 int 1138 zfs_read(char *buf, int len) 1139 { 1140 char *stack; 1141 char *tmpbuf; 1142 int blksz, length, movesize; 1143 1144 if (file_buf == NULL) { 1145 file_buf = stackbase; 1146 stackbase += SPA_MAXBLOCKSIZE; 1147 file_start = file_end = 0; 1148 } 1149 stack = stackbase; 1150 1151 /* 1152 * If offset is in memory, move it into the buffer provided and return. 1153 */ 1154 if (filepos >= file_start && filepos+len <= file_end) { 1155 grub_memmove(buf, file_buf + filepos - file_start, len); 1156 filepos += len; 1157 return (len); 1158 } 1159 1160 blksz = DNODE->dn_datablkszsec << SPA_MINBLOCKSHIFT; 1161 1162 /* 1163 * Entire Dnode is too big to fit into the space available. We 1164 * will need to read it in chunks. This could be optimized to 1165 * read in as large a chunk as there is space available, but for 1166 * now, this only reads in one data block at a time. 1167 */ 1168 length = len; 1169 while (length) { 1170 /* 1171 * Find requested blkid and the offset within that block. 1172 */ 1173 uint64_t blkid = filepos / blksz; 1174 1175 if (errnum = dmu_read(DNODE, blkid, file_buf, stack)) 1176 return (0); 1177 1178 file_start = blkid * blksz; 1179 file_end = file_start + blksz; 1180 1181 movesize = MIN(length, file_end - filepos); 1182 1183 grub_memmove(buf, file_buf + filepos - file_start, 1184 movesize); 1185 buf += movesize; 1186 length -= movesize; 1187 filepos += movesize; 1188 } 1189 1190 return (len); 1191 } 1192 1193 /* 1194 * No-Op 1195 */ 1196 int 1197 zfs_embed(int *start_sector, int needed_sectors) 1198 { 1199 return (1); 1200 } 1201 1202 #endif /* FSYS_ZFS */ 1203