1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * Copyright (C) 2017 Oracle. All Rights Reserved. 4 * Author: Darrick J. Wong <darrick.wong@oracle.com> 5 */ 6 #include "xfs.h" 7 #include "xfs_fs.h" 8 #include "xfs_shared.h" 9 #include "xfs_format.h" 10 #include "xfs_trans_resv.h" 11 #include "xfs_mount.h" 12 #include "xfs_log_format.h" 13 #include "xfs_trans.h" 14 #include "xfs_inode.h" 15 #include "xfs_dir2.h" 16 #include "xfs_dir2_priv.h" 17 #include "xfs_attr_leaf.h" 18 #include "scrub/scrub.h" 19 #include "scrub/common.h" 20 #include "scrub/trace.h" 21 #include "scrub/dabtree.h" 22 23 /* Directory/Attribute Btree */ 24 25 /* 26 * Check for da btree operation errors. See the section about handling 27 * operational errors in common.c. 28 */ 29 bool 30 xchk_da_process_error( 31 struct xchk_da_btree *ds, 32 int level, 33 int *error) 34 { 35 struct xfs_scrub *sc = ds->sc; 36 37 if (*error == 0) 38 return true; 39 40 switch (*error) { 41 case -EDEADLOCK: 42 /* Used to restart an op with deadlock avoidance. */ 43 trace_xchk_deadlock_retry(sc->ip, sc->sm, *error); 44 break; 45 case -EFSBADCRC: 46 case -EFSCORRUPTED: 47 /* Note the badness but don't abort. */ 48 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT; 49 *error = 0; 50 fallthrough; 51 default: 52 trace_xchk_file_op_error(sc, ds->dargs.whichfork, 53 xfs_dir2_da_to_db(ds->dargs.geo, 54 ds->state->path.blk[level].blkno), 55 *error, __return_address); 56 break; 57 } 58 return false; 59 } 60 61 /* 62 * Check for da btree corruption. See the section about handling 63 * operational errors in common.c. 64 */ 65 void 66 xchk_da_set_corrupt( 67 struct xchk_da_btree *ds, 68 int level) 69 { 70 struct xfs_scrub *sc = ds->sc; 71 72 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT; 73 74 trace_xchk_fblock_error(sc, ds->dargs.whichfork, 75 xfs_dir2_da_to_db(ds->dargs.geo, 76 ds->state->path.blk[level].blkno), 77 __return_address); 78 } 79 80 static struct xfs_da_node_entry * 81 xchk_da_btree_node_entry( 82 struct xchk_da_btree *ds, 83 int level) 84 { 85 struct xfs_da_state_blk *blk = &ds->state->path.blk[level]; 86 struct xfs_da3_icnode_hdr hdr; 87 88 ASSERT(blk->magic == XFS_DA_NODE_MAGIC); 89 90 xfs_da3_node_hdr_from_disk(ds->sc->mp, &hdr, blk->bp->b_addr); 91 return hdr.btree + blk->index; 92 } 93 94 /* Scrub a da btree hash (key). */ 95 int 96 xchk_da_btree_hash( 97 struct xchk_da_btree *ds, 98 int level, 99 __be32 *hashp) 100 { 101 struct xfs_da_node_entry *entry; 102 xfs_dahash_t hash; 103 xfs_dahash_t parent_hash; 104 105 /* Is this hash in order? */ 106 hash = be32_to_cpu(*hashp); 107 if (hash < ds->hashes[level]) 108 xchk_da_set_corrupt(ds, level); 109 ds->hashes[level] = hash; 110 111 if (level == 0) 112 return 0; 113 114 /* Is this hash no larger than the parent hash? */ 115 entry = xchk_da_btree_node_entry(ds, level - 1); 116 parent_hash = be32_to_cpu(entry->hashval); 117 if (parent_hash < hash) 118 xchk_da_set_corrupt(ds, level); 119 120 return 0; 121 } 122 123 /* 124 * Check a da btree pointer. Returns true if it's ok to use this 125 * pointer. 126 */ 127 STATIC bool 128 xchk_da_btree_ptr_ok( 129 struct xchk_da_btree *ds, 130 int level, 131 xfs_dablk_t blkno) 132 { 133 if (blkno < ds->lowest || (ds->highest != 0 && blkno >= ds->highest)) { 134 xchk_da_set_corrupt(ds, level); 135 return false; 136 } 137 138 return true; 139 } 140 141 /* 142 * The da btree scrubber can handle leaf1 blocks as a degenerate 143 * form of leafn blocks. Since the regular da code doesn't handle 144 * leaf1, we must multiplex the verifiers. 145 */ 146 static void 147 xchk_da_btree_read_verify( 148 struct xfs_buf *bp) 149 { 150 struct xfs_da_blkinfo *info = bp->b_addr; 151 152 switch (be16_to_cpu(info->magic)) { 153 case XFS_DIR2_LEAF1_MAGIC: 154 case XFS_DIR3_LEAF1_MAGIC: 155 bp->b_ops = &xfs_dir3_leaf1_buf_ops; 156 bp->b_ops->verify_read(bp); 157 return; 158 default: 159 /* 160 * xfs_da3_node_buf_ops already know how to handle 161 * DA*_NODE, ATTR*_LEAF, and DIR*_LEAFN blocks. 162 */ 163 bp->b_ops = &xfs_da3_node_buf_ops; 164 bp->b_ops->verify_read(bp); 165 return; 166 } 167 } 168 static void 169 xchk_da_btree_write_verify( 170 struct xfs_buf *bp) 171 { 172 struct xfs_da_blkinfo *info = bp->b_addr; 173 174 switch (be16_to_cpu(info->magic)) { 175 case XFS_DIR2_LEAF1_MAGIC: 176 case XFS_DIR3_LEAF1_MAGIC: 177 bp->b_ops = &xfs_dir3_leaf1_buf_ops; 178 bp->b_ops->verify_write(bp); 179 return; 180 default: 181 /* 182 * xfs_da3_node_buf_ops already know how to handle 183 * DA*_NODE, ATTR*_LEAF, and DIR*_LEAFN blocks. 184 */ 185 bp->b_ops = &xfs_da3_node_buf_ops; 186 bp->b_ops->verify_write(bp); 187 return; 188 } 189 } 190 static void * 191 xchk_da_btree_verify( 192 struct xfs_buf *bp) 193 { 194 struct xfs_da_blkinfo *info = bp->b_addr; 195 196 switch (be16_to_cpu(info->magic)) { 197 case XFS_DIR2_LEAF1_MAGIC: 198 case XFS_DIR3_LEAF1_MAGIC: 199 bp->b_ops = &xfs_dir3_leaf1_buf_ops; 200 return bp->b_ops->verify_struct(bp); 201 default: 202 bp->b_ops = &xfs_da3_node_buf_ops; 203 return bp->b_ops->verify_struct(bp); 204 } 205 } 206 207 static const struct xfs_buf_ops xchk_da_btree_buf_ops = { 208 .name = "xchk_da_btree", 209 .verify_read = xchk_da_btree_read_verify, 210 .verify_write = xchk_da_btree_write_verify, 211 .verify_struct = xchk_da_btree_verify, 212 }; 213 214 /* Check a block's sibling. */ 215 STATIC int 216 xchk_da_btree_block_check_sibling( 217 struct xchk_da_btree *ds, 218 int level, 219 int direction, 220 xfs_dablk_t sibling) 221 { 222 struct xfs_da_state_path *path = &ds->state->path; 223 struct xfs_da_state_path *altpath = &ds->state->altpath; 224 int retval; 225 int plevel; 226 int error; 227 228 memcpy(altpath, path, sizeof(ds->state->altpath)); 229 230 /* 231 * If the pointer is null, we shouldn't be able to move the upper 232 * level pointer anywhere. 233 */ 234 if (sibling == 0) { 235 error = xfs_da3_path_shift(ds->state, altpath, direction, 236 false, &retval); 237 if (error == 0 && retval == 0) 238 xchk_da_set_corrupt(ds, level); 239 error = 0; 240 goto out; 241 } 242 243 /* Move the alternate cursor one block in the direction given. */ 244 error = xfs_da3_path_shift(ds->state, altpath, direction, false, 245 &retval); 246 if (!xchk_da_process_error(ds, level, &error)) 247 goto out; 248 if (retval) { 249 xchk_da_set_corrupt(ds, level); 250 goto out; 251 } 252 if (altpath->blk[level].bp) 253 xchk_buffer_recheck(ds->sc, altpath->blk[level].bp); 254 255 /* Compare upper level pointer to sibling pointer. */ 256 if (altpath->blk[level].blkno != sibling) 257 xchk_da_set_corrupt(ds, level); 258 259 out: 260 /* Free all buffers in the altpath that aren't referenced from path. */ 261 for (plevel = 0; plevel < altpath->active; plevel++) { 262 if (altpath->blk[plevel].bp == NULL || 263 (plevel < path->active && 264 altpath->blk[plevel].bp == path->blk[plevel].bp)) 265 continue; 266 267 xfs_trans_brelse(ds->dargs.trans, altpath->blk[plevel].bp); 268 altpath->blk[plevel].bp = NULL; 269 } 270 271 return error; 272 } 273 274 /* Check a block's sibling pointers. */ 275 STATIC int 276 xchk_da_btree_block_check_siblings( 277 struct xchk_da_btree *ds, 278 int level, 279 struct xfs_da_blkinfo *hdr) 280 { 281 xfs_dablk_t forw; 282 xfs_dablk_t back; 283 int error = 0; 284 285 forw = be32_to_cpu(hdr->forw); 286 back = be32_to_cpu(hdr->back); 287 288 /* Top level blocks should not have sibling pointers. */ 289 if (level == 0) { 290 if (forw != 0 || back != 0) 291 xchk_da_set_corrupt(ds, level); 292 return 0; 293 } 294 295 /* 296 * Check back (left) and forw (right) pointers. These functions 297 * absorb error codes for us. 298 */ 299 error = xchk_da_btree_block_check_sibling(ds, level, 0, back); 300 if (error) 301 goto out; 302 error = xchk_da_btree_block_check_sibling(ds, level, 1, forw); 303 304 out: 305 memset(&ds->state->altpath, 0, sizeof(ds->state->altpath)); 306 return error; 307 } 308 309 /* Load a dir/attribute block from a btree. */ 310 STATIC int 311 xchk_da_btree_block( 312 struct xchk_da_btree *ds, 313 int level, 314 xfs_dablk_t blkno) 315 { 316 struct xfs_da_state_blk *blk; 317 struct xfs_da_intnode *node; 318 struct xfs_da_node_entry *btree; 319 struct xfs_da3_blkinfo *hdr3; 320 struct xfs_da_args *dargs = &ds->dargs; 321 struct xfs_inode *ip = ds->dargs.dp; 322 xfs_ino_t owner; 323 int *pmaxrecs; 324 struct xfs_da3_icnode_hdr nodehdr; 325 int error = 0; 326 327 blk = &ds->state->path.blk[level]; 328 ds->state->path.active = level + 1; 329 330 /* Release old block. */ 331 if (blk->bp) { 332 xfs_trans_brelse(dargs->trans, blk->bp); 333 blk->bp = NULL; 334 } 335 336 /* Check the pointer. */ 337 blk->blkno = blkno; 338 if (!xchk_da_btree_ptr_ok(ds, level, blkno)) 339 goto out_nobuf; 340 341 /* Read the buffer. */ 342 error = xfs_da_read_buf(dargs->trans, dargs->dp, blk->blkno, 343 XFS_DABUF_MAP_HOLE_OK, &blk->bp, dargs->whichfork, 344 &xchk_da_btree_buf_ops); 345 if (!xchk_da_process_error(ds, level, &error)) 346 goto out_nobuf; 347 if (blk->bp) 348 xchk_buffer_recheck(ds->sc, blk->bp); 349 350 /* 351 * We didn't find a dir btree root block, which means that 352 * there's no LEAF1/LEAFN tree (at least not where it's supposed 353 * to be), so jump out now. 354 */ 355 if (ds->dargs.whichfork == XFS_DATA_FORK && level == 0 && 356 blk->bp == NULL) 357 goto out_nobuf; 358 359 /* It's /not/ ok for attr trees not to have a da btree. */ 360 if (blk->bp == NULL) { 361 xchk_da_set_corrupt(ds, level); 362 goto out_nobuf; 363 } 364 365 hdr3 = blk->bp->b_addr; 366 blk->magic = be16_to_cpu(hdr3->hdr.magic); 367 pmaxrecs = &ds->maxrecs[level]; 368 369 /* We only started zeroing the header on v5 filesystems. */ 370 if (xfs_has_crc(ds->sc->mp) && hdr3->hdr.pad) 371 xchk_da_set_corrupt(ds, level); 372 373 /* Check the owner. */ 374 if (xfs_has_crc(ip->i_mount)) { 375 owner = be64_to_cpu(hdr3->owner); 376 if (owner != ip->i_ino) 377 xchk_da_set_corrupt(ds, level); 378 } 379 380 /* Check the siblings. */ 381 error = xchk_da_btree_block_check_siblings(ds, level, &hdr3->hdr); 382 if (error) 383 goto out; 384 385 /* Interpret the buffer. */ 386 switch (blk->magic) { 387 case XFS_ATTR_LEAF_MAGIC: 388 case XFS_ATTR3_LEAF_MAGIC: 389 xfs_trans_buf_set_type(dargs->trans, blk->bp, 390 XFS_BLFT_ATTR_LEAF_BUF); 391 blk->magic = XFS_ATTR_LEAF_MAGIC; 392 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, pmaxrecs); 393 if (ds->tree_level != 0) 394 xchk_da_set_corrupt(ds, level); 395 break; 396 case XFS_DIR2_LEAFN_MAGIC: 397 case XFS_DIR3_LEAFN_MAGIC: 398 xfs_trans_buf_set_type(dargs->trans, blk->bp, 399 XFS_BLFT_DIR_LEAFN_BUF); 400 blk->magic = XFS_DIR2_LEAFN_MAGIC; 401 blk->hashval = xfs_dir2_leaf_lasthash(ip, blk->bp, pmaxrecs); 402 if (ds->tree_level != 0) 403 xchk_da_set_corrupt(ds, level); 404 break; 405 case XFS_DIR2_LEAF1_MAGIC: 406 case XFS_DIR3_LEAF1_MAGIC: 407 xfs_trans_buf_set_type(dargs->trans, blk->bp, 408 XFS_BLFT_DIR_LEAF1_BUF); 409 blk->magic = XFS_DIR2_LEAF1_MAGIC; 410 blk->hashval = xfs_dir2_leaf_lasthash(ip, blk->bp, pmaxrecs); 411 if (ds->tree_level != 0) 412 xchk_da_set_corrupt(ds, level); 413 break; 414 case XFS_DA_NODE_MAGIC: 415 case XFS_DA3_NODE_MAGIC: 416 xfs_trans_buf_set_type(dargs->trans, blk->bp, 417 XFS_BLFT_DA_NODE_BUF); 418 blk->magic = XFS_DA_NODE_MAGIC; 419 node = blk->bp->b_addr; 420 xfs_da3_node_hdr_from_disk(ip->i_mount, &nodehdr, node); 421 btree = nodehdr.btree; 422 *pmaxrecs = nodehdr.count; 423 blk->hashval = be32_to_cpu(btree[*pmaxrecs - 1].hashval); 424 if (level == 0) { 425 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) { 426 xchk_da_set_corrupt(ds, level); 427 goto out_freebp; 428 } 429 ds->tree_level = nodehdr.level; 430 } else { 431 if (ds->tree_level != nodehdr.level) { 432 xchk_da_set_corrupt(ds, level); 433 goto out_freebp; 434 } 435 } 436 437 /* XXX: Check hdr3.pad32 once we know how to fix it. */ 438 break; 439 default: 440 xchk_da_set_corrupt(ds, level); 441 goto out_freebp; 442 } 443 444 /* 445 * If we've been handed a block that is below the dabtree root, does 446 * its hashval match what the parent block expected to see? 447 */ 448 if (level > 0) { 449 struct xfs_da_node_entry *key; 450 451 key = xchk_da_btree_node_entry(ds, level - 1); 452 if (be32_to_cpu(key->hashval) != blk->hashval) { 453 xchk_da_set_corrupt(ds, level); 454 goto out_freebp; 455 } 456 } 457 458 out: 459 return error; 460 out_freebp: 461 xfs_trans_brelse(dargs->trans, blk->bp); 462 blk->bp = NULL; 463 out_nobuf: 464 blk->blkno = 0; 465 return error; 466 } 467 468 /* Visit all nodes and leaves of a da btree. */ 469 int 470 xchk_da_btree( 471 struct xfs_scrub *sc, 472 int whichfork, 473 xchk_da_btree_rec_fn scrub_fn, 474 void *private) 475 { 476 struct xchk_da_btree *ds; 477 struct xfs_mount *mp = sc->mp; 478 struct xfs_da_state_blk *blks; 479 struct xfs_da_node_entry *key; 480 xfs_dablk_t blkno; 481 int level; 482 int error; 483 484 /* Skip short format data structures; no btree to scan. */ 485 if (!xfs_ifork_has_extents(xfs_ifork_ptr(sc->ip, whichfork))) 486 return 0; 487 488 /* Set up initial da state. */ 489 ds = kmem_zalloc(sizeof(struct xchk_da_btree), KM_NOFS | KM_MAYFAIL); 490 if (!ds) 491 return -ENOMEM; 492 ds->dargs.dp = sc->ip; 493 ds->dargs.whichfork = whichfork; 494 ds->dargs.trans = sc->tp; 495 ds->dargs.op_flags = XFS_DA_OP_OKNOENT; 496 ds->state = xfs_da_state_alloc(&ds->dargs); 497 ds->sc = sc; 498 ds->private = private; 499 if (whichfork == XFS_ATTR_FORK) { 500 ds->dargs.geo = mp->m_attr_geo; 501 ds->lowest = 0; 502 ds->highest = 0; 503 } else { 504 ds->dargs.geo = mp->m_dir_geo; 505 ds->lowest = ds->dargs.geo->leafblk; 506 ds->highest = ds->dargs.geo->freeblk; 507 } 508 blkno = ds->lowest; 509 level = 0; 510 511 /* Find the root of the da tree, if present. */ 512 blks = ds->state->path.blk; 513 error = xchk_da_btree_block(ds, level, blkno); 514 if (error) 515 goto out_state; 516 /* 517 * We didn't find a block at ds->lowest, which means that there's 518 * no LEAF1/LEAFN tree (at least not where it's supposed to be), 519 * so jump out now. 520 */ 521 if (blks[level].bp == NULL) 522 goto out_state; 523 524 blks[level].index = 0; 525 while (level >= 0 && level < XFS_DA_NODE_MAXDEPTH) { 526 /* Handle leaf block. */ 527 if (blks[level].magic != XFS_DA_NODE_MAGIC) { 528 /* End of leaf, pop back towards the root. */ 529 if (blks[level].index >= ds->maxrecs[level]) { 530 if (level > 0) 531 blks[level - 1].index++; 532 ds->tree_level++; 533 level--; 534 continue; 535 } 536 537 /* Dispatch record scrubbing. */ 538 error = scrub_fn(ds, level); 539 if (error) 540 break; 541 if (xchk_should_terminate(sc, &error) || 542 (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) 543 break; 544 545 blks[level].index++; 546 continue; 547 } 548 549 550 /* End of node, pop back towards the root. */ 551 if (blks[level].index >= ds->maxrecs[level]) { 552 if (level > 0) 553 blks[level - 1].index++; 554 ds->tree_level++; 555 level--; 556 continue; 557 } 558 559 /* Hashes in order for scrub? */ 560 key = xchk_da_btree_node_entry(ds, level); 561 error = xchk_da_btree_hash(ds, level, &key->hashval); 562 if (error) 563 goto out; 564 565 /* Drill another level deeper. */ 566 blkno = be32_to_cpu(key->before); 567 level++; 568 if (level >= XFS_DA_NODE_MAXDEPTH) { 569 /* Too deep! */ 570 xchk_da_set_corrupt(ds, level - 1); 571 break; 572 } 573 ds->tree_level--; 574 error = xchk_da_btree_block(ds, level, blkno); 575 if (error) 576 goto out; 577 if (blks[level].bp == NULL) 578 goto out; 579 580 blks[level].index = 0; 581 } 582 583 out: 584 /* Release all the buffers we're tracking. */ 585 for (level = 0; level < XFS_DA_NODE_MAXDEPTH; level++) { 586 if (blks[level].bp == NULL) 587 continue; 588 xfs_trans_brelse(sc->tp, blks[level].bp); 589 blks[level].bp = NULL; 590 } 591 592 out_state: 593 xfs_da_state_free(ds->state); 594 kmem_free(ds); 595 return error; 596 } 597