1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) 2018-2023 Oracle. All Rights Reserved. 4 * Author: Darrick J. Wong <djwong@kernel.org> 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_btree.h" 13 #include "xfs_log_format.h" 14 #include "xfs_trans.h" 15 #include "xfs_sb.h" 16 #include "xfs_inode.h" 17 #include "xfs_alloc.h" 18 #include "xfs_alloc_btree.h" 19 #include "xfs_ialloc.h" 20 #include "xfs_ialloc_btree.h" 21 #include "xfs_rmap.h" 22 #include "xfs_rmap_btree.h" 23 #include "xfs_refcount_btree.h" 24 #include "xfs_extent_busy.h" 25 #include "xfs_ag.h" 26 #include "xfs_ag_resv.h" 27 #include "xfs_quota.h" 28 #include "xfs_qm.h" 29 #include "xfs_defer.h" 30 #include "scrub/scrub.h" 31 #include "scrub/common.h" 32 #include "scrub/trace.h" 33 #include "scrub/repair.h" 34 #include "scrub/bitmap.h" 35 36 /* 37 * Attempt to repair some metadata, if the metadata is corrupt and userspace 38 * told us to fix it. This function returns -EAGAIN to mean "re-run scrub", 39 * and will set *fixed to true if it thinks it repaired anything. 40 */ 41 int 42 xrep_attempt( 43 struct xfs_scrub *sc) 44 { 45 int error = 0; 46 47 trace_xrep_attempt(XFS_I(file_inode(sc->file)), sc->sm, error); 48 49 xchk_ag_btcur_free(&sc->sa); 50 51 /* Repair whatever's broken. */ 52 ASSERT(sc->ops->repair); 53 error = sc->ops->repair(sc); 54 trace_xrep_done(XFS_I(file_inode(sc->file)), sc->sm, error); 55 switch (error) { 56 case 0: 57 /* 58 * Repair succeeded. Commit the fixes and perform a second 59 * scrub so that we can tell userspace if we fixed the problem. 60 */ 61 sc->sm->sm_flags &= ~XFS_SCRUB_FLAGS_OUT; 62 sc->flags |= XREP_ALREADY_FIXED; 63 return -EAGAIN; 64 case -ECHRNG: 65 sc->flags |= XCHK_NEED_DRAIN; 66 return -EAGAIN; 67 case -EDEADLOCK: 68 /* Tell the caller to try again having grabbed all the locks. */ 69 if (!(sc->flags & XCHK_TRY_HARDER)) { 70 sc->flags |= XCHK_TRY_HARDER; 71 return -EAGAIN; 72 } 73 /* 74 * We tried harder but still couldn't grab all the resources 75 * we needed to fix it. The corruption has not been fixed, 76 * so exit to userspace with the scan's output flags unchanged. 77 */ 78 return 0; 79 default: 80 /* 81 * EAGAIN tells the caller to re-scrub, so we cannot return 82 * that here. 83 */ 84 ASSERT(error != -EAGAIN); 85 return error; 86 } 87 } 88 89 /* 90 * Complain about unfixable problems in the filesystem. We don't log 91 * corruptions when IFLAG_REPAIR wasn't set on the assumption that the driver 92 * program is xfs_scrub, which will call back with IFLAG_REPAIR set if the 93 * administrator isn't running xfs_scrub in no-repairs mode. 94 * 95 * Use this helper function because _ratelimited silently declares a static 96 * structure to track rate limiting information. 97 */ 98 void 99 xrep_failure( 100 struct xfs_mount *mp) 101 { 102 xfs_alert_ratelimited(mp, 103 "Corruption not fixed during online repair. Unmount and run xfs_repair."); 104 } 105 106 /* 107 * Repair probe -- userspace uses this to probe if we're willing to repair a 108 * given mountpoint. 109 */ 110 int 111 xrep_probe( 112 struct xfs_scrub *sc) 113 { 114 int error = 0; 115 116 if (xchk_should_terminate(sc, &error)) 117 return error; 118 119 return 0; 120 } 121 122 /* 123 * Roll a transaction, keeping the AG headers locked and reinitializing 124 * the btree cursors. 125 */ 126 int 127 xrep_roll_ag_trans( 128 struct xfs_scrub *sc) 129 { 130 int error; 131 132 /* 133 * Keep the AG header buffers locked while we roll the transaction. 134 * Ensure that both AG buffers are dirty and held when we roll the 135 * transaction so that they move forward in the log without losing the 136 * bli (and hence the bli type) when the transaction commits. 137 * 138 * Normal code would never hold clean buffers across a roll, but repair 139 * needs both buffers to maintain a total lock on the AG. 140 */ 141 if (sc->sa.agi_bp) { 142 xfs_ialloc_log_agi(sc->tp, sc->sa.agi_bp, XFS_AGI_MAGICNUM); 143 xfs_trans_bhold(sc->tp, sc->sa.agi_bp); 144 } 145 146 if (sc->sa.agf_bp) { 147 xfs_alloc_log_agf(sc->tp, sc->sa.agf_bp, XFS_AGF_MAGICNUM); 148 xfs_trans_bhold(sc->tp, sc->sa.agf_bp); 149 } 150 151 /* 152 * Roll the transaction. We still hold the AG header buffers locked 153 * regardless of whether or not that succeeds. On failure, the buffers 154 * will be released during teardown on our way out of the kernel. If 155 * successful, join the buffers to the new transaction and move on. 156 */ 157 error = xfs_trans_roll(&sc->tp); 158 if (error) 159 return error; 160 161 /* Join the AG headers to the new transaction. */ 162 if (sc->sa.agi_bp) 163 xfs_trans_bjoin(sc->tp, sc->sa.agi_bp); 164 if (sc->sa.agf_bp) 165 xfs_trans_bjoin(sc->tp, sc->sa.agf_bp); 166 167 return 0; 168 } 169 170 /* Finish all deferred work attached to the repair transaction. */ 171 int 172 xrep_defer_finish( 173 struct xfs_scrub *sc) 174 { 175 int error; 176 177 /* 178 * Keep the AG header buffers locked while we complete deferred work 179 * items. Ensure that both AG buffers are dirty and held when we roll 180 * the transaction so that they move forward in the log without losing 181 * the bli (and hence the bli type) when the transaction commits. 182 * 183 * Normal code would never hold clean buffers across a roll, but repair 184 * needs both buffers to maintain a total lock on the AG. 185 */ 186 if (sc->sa.agi_bp) { 187 xfs_ialloc_log_agi(sc->tp, sc->sa.agi_bp, XFS_AGI_MAGICNUM); 188 xfs_trans_bhold(sc->tp, sc->sa.agi_bp); 189 } 190 191 if (sc->sa.agf_bp) { 192 xfs_alloc_log_agf(sc->tp, sc->sa.agf_bp, XFS_AGF_MAGICNUM); 193 xfs_trans_bhold(sc->tp, sc->sa.agf_bp); 194 } 195 196 /* 197 * Finish all deferred work items. We still hold the AG header buffers 198 * locked regardless of whether or not that succeeds. On failure, the 199 * buffers will be released during teardown on our way out of the 200 * kernel. If successful, join the buffers to the new transaction 201 * and move on. 202 */ 203 error = xfs_defer_finish(&sc->tp); 204 if (error) 205 return error; 206 207 /* 208 * Release the hold that we set above because defer_finish won't do 209 * that for us. The defer roll code redirties held buffers after each 210 * roll, so the AG header buffers should be ready for logging. 211 */ 212 if (sc->sa.agi_bp) 213 xfs_trans_bhold_release(sc->tp, sc->sa.agi_bp); 214 if (sc->sa.agf_bp) 215 xfs_trans_bhold_release(sc->tp, sc->sa.agf_bp); 216 217 return 0; 218 } 219 220 /* 221 * Does the given AG have enough space to rebuild a btree? Neither AG 222 * reservation can be critical, and we must have enough space (factoring 223 * in AG reservations) to construct a whole btree. 224 */ 225 bool 226 xrep_ag_has_space( 227 struct xfs_perag *pag, 228 xfs_extlen_t nr_blocks, 229 enum xfs_ag_resv_type type) 230 { 231 return !xfs_ag_resv_critical(pag, XFS_AG_RESV_RMAPBT) && 232 !xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA) && 233 pag->pagf_freeblks > xfs_ag_resv_needed(pag, type) + nr_blocks; 234 } 235 236 /* 237 * Figure out how many blocks to reserve for an AG repair. We calculate the 238 * worst case estimate for the number of blocks we'd need to rebuild one of 239 * any type of per-AG btree. 240 */ 241 xfs_extlen_t 242 xrep_calc_ag_resblks( 243 struct xfs_scrub *sc) 244 { 245 struct xfs_mount *mp = sc->mp; 246 struct xfs_scrub_metadata *sm = sc->sm; 247 struct xfs_perag *pag; 248 struct xfs_buf *bp; 249 xfs_agino_t icount = NULLAGINO; 250 xfs_extlen_t aglen = NULLAGBLOCK; 251 xfs_extlen_t usedlen; 252 xfs_extlen_t freelen; 253 xfs_extlen_t bnobt_sz; 254 xfs_extlen_t inobt_sz; 255 xfs_extlen_t rmapbt_sz; 256 xfs_extlen_t refcbt_sz; 257 int error; 258 259 if (!(sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)) 260 return 0; 261 262 pag = xfs_perag_get(mp, sm->sm_agno); 263 if (xfs_perag_initialised_agi(pag)) { 264 /* Use in-core icount if possible. */ 265 icount = pag->pagi_count; 266 } else { 267 /* Try to get the actual counters from disk. */ 268 error = xfs_ialloc_read_agi(pag, NULL, &bp); 269 if (!error) { 270 icount = pag->pagi_count; 271 xfs_buf_relse(bp); 272 } 273 } 274 275 /* Now grab the block counters from the AGF. */ 276 error = xfs_alloc_read_agf(pag, NULL, 0, &bp); 277 if (error) { 278 aglen = pag->block_count; 279 freelen = aglen; 280 usedlen = aglen; 281 } else { 282 struct xfs_agf *agf = bp->b_addr; 283 284 aglen = be32_to_cpu(agf->agf_length); 285 freelen = be32_to_cpu(agf->agf_freeblks); 286 usedlen = aglen - freelen; 287 xfs_buf_relse(bp); 288 } 289 290 /* If the icount is impossible, make some worst-case assumptions. */ 291 if (icount == NULLAGINO || 292 !xfs_verify_agino(pag, icount)) { 293 icount = pag->agino_max - pag->agino_min + 1; 294 } 295 296 /* If the block counts are impossible, make worst-case assumptions. */ 297 if (aglen == NULLAGBLOCK || 298 aglen != pag->block_count || 299 freelen >= aglen) { 300 aglen = pag->block_count; 301 freelen = aglen; 302 usedlen = aglen; 303 } 304 xfs_perag_put(pag); 305 306 trace_xrep_calc_ag_resblks(mp, sm->sm_agno, icount, aglen, 307 freelen, usedlen); 308 309 /* 310 * Figure out how many blocks we'd need worst case to rebuild 311 * each type of btree. Note that we can only rebuild the 312 * bnobt/cntbt or inobt/finobt as pairs. 313 */ 314 bnobt_sz = 2 * xfs_allocbt_calc_size(mp, freelen); 315 if (xfs_has_sparseinodes(mp)) 316 inobt_sz = xfs_iallocbt_calc_size(mp, icount / 317 XFS_INODES_PER_HOLEMASK_BIT); 318 else 319 inobt_sz = xfs_iallocbt_calc_size(mp, icount / 320 XFS_INODES_PER_CHUNK); 321 if (xfs_has_finobt(mp)) 322 inobt_sz *= 2; 323 if (xfs_has_reflink(mp)) 324 refcbt_sz = xfs_refcountbt_calc_size(mp, usedlen); 325 else 326 refcbt_sz = 0; 327 if (xfs_has_rmapbt(mp)) { 328 /* 329 * Guess how many blocks we need to rebuild the rmapbt. 330 * For non-reflink filesystems we can't have more records than 331 * used blocks. However, with reflink it's possible to have 332 * more than one rmap record per AG block. We don't know how 333 * many rmaps there could be in the AG, so we start off with 334 * what we hope is an generous over-estimation. 335 */ 336 if (xfs_has_reflink(mp)) 337 rmapbt_sz = xfs_rmapbt_calc_size(mp, 338 (unsigned long long)aglen * 2); 339 else 340 rmapbt_sz = xfs_rmapbt_calc_size(mp, usedlen); 341 } else { 342 rmapbt_sz = 0; 343 } 344 345 trace_xrep_calc_ag_resblks_btsize(mp, sm->sm_agno, bnobt_sz, 346 inobt_sz, rmapbt_sz, refcbt_sz); 347 348 return max(max(bnobt_sz, inobt_sz), max(rmapbt_sz, refcbt_sz)); 349 } 350 351 /* 352 * Reconstructing per-AG Btrees 353 * 354 * When a space btree is corrupt, we don't bother trying to fix it. Instead, 355 * we scan secondary space metadata to derive the records that should be in 356 * the damaged btree, initialize a fresh btree root, and insert the records. 357 * Note that for rebuilding the rmapbt we scan all the primary data to 358 * generate the new records. 359 * 360 * However, that leaves the matter of removing all the metadata describing the 361 * old broken structure. For primary metadata we use the rmap data to collect 362 * every extent with a matching rmap owner (bitmap); we then iterate all other 363 * metadata structures with the same rmap owner to collect the extents that 364 * cannot be removed (sublist). We then subtract sublist from bitmap to 365 * derive the blocks that were used by the old btree. These blocks can be 366 * reaped. 367 * 368 * For rmapbt reconstructions we must use different tactics for extent 369 * collection. First we iterate all primary metadata (this excludes the old 370 * rmapbt, obviously) to generate new rmap records. The gaps in the rmap 371 * records are collected as bitmap. The bnobt records are collected as 372 * sublist. As with the other btrees we subtract sublist from bitmap, and the 373 * result (since the rmapbt lives in the free space) are the blocks from the 374 * old rmapbt. 375 */ 376 377 /* Ensure the freelist is the correct size. */ 378 int 379 xrep_fix_freelist( 380 struct xfs_scrub *sc, 381 bool can_shrink) 382 { 383 struct xfs_alloc_arg args = {0}; 384 385 args.mp = sc->mp; 386 args.tp = sc->tp; 387 args.agno = sc->sa.pag->pag_agno; 388 args.alignment = 1; 389 args.pag = sc->sa.pag; 390 391 return xfs_alloc_fix_freelist(&args, 392 can_shrink ? 0 : XFS_ALLOC_FLAG_NOSHRINK); 393 } 394 395 /* 396 * Finding per-AG Btree Roots for AGF/AGI Reconstruction 397 * 398 * If the AGF or AGI become slightly corrupted, it may be necessary to rebuild 399 * the AG headers by using the rmap data to rummage through the AG looking for 400 * btree roots. This is not guaranteed to work if the AG is heavily damaged 401 * or the rmap data are corrupt. 402 * 403 * Callers of xrep_find_ag_btree_roots must lock the AGF and AGFL 404 * buffers if the AGF is being rebuilt; or the AGF and AGI buffers if the 405 * AGI is being rebuilt. It must maintain these locks until it's safe for 406 * other threads to change the btrees' shapes. The caller provides 407 * information about the btrees to look for by passing in an array of 408 * xrep_find_ag_btree with the (rmap owner, buf_ops, magic) fields set. 409 * The (root, height) fields will be set on return if anything is found. The 410 * last element of the array should have a NULL buf_ops to mark the end of the 411 * array. 412 * 413 * For every rmapbt record matching any of the rmap owners in btree_info, 414 * read each block referenced by the rmap record. If the block is a btree 415 * block from this filesystem matching any of the magic numbers and has a 416 * level higher than what we've already seen, remember the block and the 417 * height of the tree required to have such a block. When the call completes, 418 * we return the highest block we've found for each btree description; those 419 * should be the roots. 420 */ 421 422 struct xrep_findroot { 423 struct xfs_scrub *sc; 424 struct xfs_buf *agfl_bp; 425 struct xfs_agf *agf; 426 struct xrep_find_ag_btree *btree_info; 427 }; 428 429 /* See if our block is in the AGFL. */ 430 STATIC int 431 xrep_findroot_agfl_walk( 432 struct xfs_mount *mp, 433 xfs_agblock_t bno, 434 void *priv) 435 { 436 xfs_agblock_t *agbno = priv; 437 438 return (*agbno == bno) ? -ECANCELED : 0; 439 } 440 441 /* Does this block match the btree information passed in? */ 442 STATIC int 443 xrep_findroot_block( 444 struct xrep_findroot *ri, 445 struct xrep_find_ag_btree *fab, 446 uint64_t owner, 447 xfs_agblock_t agbno, 448 bool *done_with_block) 449 { 450 struct xfs_mount *mp = ri->sc->mp; 451 struct xfs_buf *bp; 452 struct xfs_btree_block *btblock; 453 xfs_daddr_t daddr; 454 int block_level; 455 int error = 0; 456 457 daddr = XFS_AGB_TO_DADDR(mp, ri->sc->sa.pag->pag_agno, agbno); 458 459 /* 460 * Blocks in the AGFL have stale contents that might just happen to 461 * have a matching magic and uuid. We don't want to pull these blocks 462 * in as part of a tree root, so we have to filter out the AGFL stuff 463 * here. If the AGFL looks insane we'll just refuse to repair. 464 */ 465 if (owner == XFS_RMAP_OWN_AG) { 466 error = xfs_agfl_walk(mp, ri->agf, ri->agfl_bp, 467 xrep_findroot_agfl_walk, &agbno); 468 if (error == -ECANCELED) 469 return 0; 470 if (error) 471 return error; 472 } 473 474 /* 475 * Read the buffer into memory so that we can see if it's a match for 476 * our btree type. We have no clue if it is beforehand, and we want to 477 * avoid xfs_trans_read_buf's behavior of dumping the DONE state (which 478 * will cause needless disk reads in subsequent calls to this function) 479 * and logging metadata verifier failures. 480 * 481 * Therefore, pass in NULL buffer ops. If the buffer was already in 482 * memory from some other caller it will already have b_ops assigned. 483 * If it was in memory from a previous unsuccessful findroot_block 484 * call, the buffer won't have b_ops but it should be clean and ready 485 * for us to try to verify if the read call succeeds. The same applies 486 * if the buffer wasn't in memory at all. 487 * 488 * Note: If we never match a btree type with this buffer, it will be 489 * left in memory with NULL b_ops. This shouldn't be a problem unless 490 * the buffer gets written. 491 */ 492 error = xfs_trans_read_buf(mp, ri->sc->tp, mp->m_ddev_targp, daddr, 493 mp->m_bsize, 0, &bp, NULL); 494 if (error) 495 return error; 496 497 /* Ensure the block magic matches the btree type we're looking for. */ 498 btblock = XFS_BUF_TO_BLOCK(bp); 499 ASSERT(fab->buf_ops->magic[1] != 0); 500 if (btblock->bb_magic != fab->buf_ops->magic[1]) 501 goto out; 502 503 /* 504 * If the buffer already has ops applied and they're not the ones for 505 * this btree type, we know this block doesn't match the btree and we 506 * can bail out. 507 * 508 * If the buffer ops match ours, someone else has already validated 509 * the block for us, so we can move on to checking if this is a root 510 * block candidate. 511 * 512 * If the buffer does not have ops, nobody has successfully validated 513 * the contents and the buffer cannot be dirty. If the magic, uuid, 514 * and structure match this btree type then we'll move on to checking 515 * if it's a root block candidate. If there is no match, bail out. 516 */ 517 if (bp->b_ops) { 518 if (bp->b_ops != fab->buf_ops) 519 goto out; 520 } else { 521 ASSERT(!xfs_trans_buf_is_dirty(bp)); 522 if (!uuid_equal(&btblock->bb_u.s.bb_uuid, 523 &mp->m_sb.sb_meta_uuid)) 524 goto out; 525 /* 526 * Read verifiers can reference b_ops, so we set the pointer 527 * here. If the verifier fails we'll reset the buffer state 528 * to what it was before we touched the buffer. 529 */ 530 bp->b_ops = fab->buf_ops; 531 fab->buf_ops->verify_read(bp); 532 if (bp->b_error) { 533 bp->b_ops = NULL; 534 bp->b_error = 0; 535 goto out; 536 } 537 538 /* 539 * Some read verifiers will (re)set b_ops, so we must be 540 * careful not to change b_ops after running the verifier. 541 */ 542 } 543 544 /* 545 * This block passes the magic/uuid and verifier tests for this btree 546 * type. We don't need the caller to try the other tree types. 547 */ 548 *done_with_block = true; 549 550 /* 551 * Compare this btree block's level to the height of the current 552 * candidate root block. 553 * 554 * If the level matches the root we found previously, throw away both 555 * blocks because there can't be two candidate roots. 556 * 557 * If level is lower in the tree than the root we found previously, 558 * ignore this block. 559 */ 560 block_level = xfs_btree_get_level(btblock); 561 if (block_level + 1 == fab->height) { 562 fab->root = NULLAGBLOCK; 563 goto out; 564 } else if (block_level < fab->height) { 565 goto out; 566 } 567 568 /* 569 * This is the highest block in the tree that we've found so far. 570 * Update the btree height to reflect what we've learned from this 571 * block. 572 */ 573 fab->height = block_level + 1; 574 575 /* 576 * If this block doesn't have sibling pointers, then it's the new root 577 * block candidate. Otherwise, the root will be found farther up the 578 * tree. 579 */ 580 if (btblock->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) && 581 btblock->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK)) 582 fab->root = agbno; 583 else 584 fab->root = NULLAGBLOCK; 585 586 trace_xrep_findroot_block(mp, ri->sc->sa.pag->pag_agno, agbno, 587 be32_to_cpu(btblock->bb_magic), fab->height - 1); 588 out: 589 xfs_trans_brelse(ri->sc->tp, bp); 590 return error; 591 } 592 593 /* 594 * Do any of the blocks in this rmap record match one of the btrees we're 595 * looking for? 596 */ 597 STATIC int 598 xrep_findroot_rmap( 599 struct xfs_btree_cur *cur, 600 const struct xfs_rmap_irec *rec, 601 void *priv) 602 { 603 struct xrep_findroot *ri = priv; 604 struct xrep_find_ag_btree *fab; 605 xfs_agblock_t b; 606 bool done; 607 int error = 0; 608 609 /* Ignore anything that isn't AG metadata. */ 610 if (!XFS_RMAP_NON_INODE_OWNER(rec->rm_owner)) 611 return 0; 612 613 /* Otherwise scan each block + btree type. */ 614 for (b = 0; b < rec->rm_blockcount; b++) { 615 done = false; 616 for (fab = ri->btree_info; fab->buf_ops; fab++) { 617 if (rec->rm_owner != fab->rmap_owner) 618 continue; 619 error = xrep_findroot_block(ri, fab, 620 rec->rm_owner, rec->rm_startblock + b, 621 &done); 622 if (error) 623 return error; 624 if (done) 625 break; 626 } 627 } 628 629 return 0; 630 } 631 632 /* Find the roots of the per-AG btrees described in btree_info. */ 633 int 634 xrep_find_ag_btree_roots( 635 struct xfs_scrub *sc, 636 struct xfs_buf *agf_bp, 637 struct xrep_find_ag_btree *btree_info, 638 struct xfs_buf *agfl_bp) 639 { 640 struct xfs_mount *mp = sc->mp; 641 struct xrep_findroot ri; 642 struct xrep_find_ag_btree *fab; 643 struct xfs_btree_cur *cur; 644 int error; 645 646 ASSERT(xfs_buf_islocked(agf_bp)); 647 ASSERT(agfl_bp == NULL || xfs_buf_islocked(agfl_bp)); 648 649 ri.sc = sc; 650 ri.btree_info = btree_info; 651 ri.agf = agf_bp->b_addr; 652 ri.agfl_bp = agfl_bp; 653 for (fab = btree_info; fab->buf_ops; fab++) { 654 ASSERT(agfl_bp || fab->rmap_owner != XFS_RMAP_OWN_AG); 655 ASSERT(XFS_RMAP_NON_INODE_OWNER(fab->rmap_owner)); 656 fab->root = NULLAGBLOCK; 657 fab->height = 0; 658 } 659 660 cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.pag); 661 error = xfs_rmap_query_all(cur, xrep_findroot_rmap, &ri); 662 xfs_btree_del_cursor(cur, error); 663 664 return error; 665 } 666 667 /* Force a quotacheck the next time we mount. */ 668 void 669 xrep_force_quotacheck( 670 struct xfs_scrub *sc, 671 xfs_dqtype_t type) 672 { 673 uint flag; 674 675 flag = xfs_quota_chkd_flag(type); 676 if (!(flag & sc->mp->m_qflags)) 677 return; 678 679 mutex_lock(&sc->mp->m_quotainfo->qi_quotaofflock); 680 sc->mp->m_qflags &= ~flag; 681 spin_lock(&sc->mp->m_sb_lock); 682 sc->mp->m_sb.sb_qflags &= ~flag; 683 spin_unlock(&sc->mp->m_sb_lock); 684 xfs_log_sb(sc->tp); 685 mutex_unlock(&sc->mp->m_quotainfo->qi_quotaofflock); 686 } 687 688 /* 689 * Attach dquots to this inode, or schedule quotacheck to fix them. 690 * 691 * This function ensures that the appropriate dquots are attached to an inode. 692 * We cannot allow the dquot code to allocate an on-disk dquot block here 693 * because we're already in transaction context with the inode locked. The 694 * on-disk dquot should already exist anyway. If the quota code signals 695 * corruption or missing quota information, schedule quotacheck, which will 696 * repair corruptions in the quota metadata. 697 */ 698 int 699 xrep_ino_dqattach( 700 struct xfs_scrub *sc) 701 { 702 int error; 703 704 error = xfs_qm_dqattach_locked(sc->ip, false); 705 switch (error) { 706 case -EFSBADCRC: 707 case -EFSCORRUPTED: 708 case -ENOENT: 709 xfs_err_ratelimited(sc->mp, 710 "inode %llu repair encountered quota error %d, quotacheck forced.", 711 (unsigned long long)sc->ip->i_ino, error); 712 if (XFS_IS_UQUOTA_ON(sc->mp) && !sc->ip->i_udquot) 713 xrep_force_quotacheck(sc, XFS_DQTYPE_USER); 714 if (XFS_IS_GQUOTA_ON(sc->mp) && !sc->ip->i_gdquot) 715 xrep_force_quotacheck(sc, XFS_DQTYPE_GROUP); 716 if (XFS_IS_PQUOTA_ON(sc->mp) && !sc->ip->i_pdquot) 717 xrep_force_quotacheck(sc, XFS_DQTYPE_PROJ); 718 fallthrough; 719 case -ESRCH: 720 error = 0; 721 break; 722 default: 723 break; 724 } 725 726 return error; 727 } 728