b.liu | e958203 | 2025-04-17 19:18:16 +0800 | [diff] [blame^] | 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_btree.h" |
| 13 | #include "xfs_bit.h" |
| 14 | #include "xfs_log_format.h" |
| 15 | #include "xfs_trans.h" |
| 16 | #include "xfs_inode.h" |
| 17 | #include "xfs_alloc.h" |
| 18 | #include "xfs_bmap.h" |
| 19 | #include "xfs_bmap_btree.h" |
| 20 | #include "xfs_rmap.h" |
| 21 | #include "xfs_rmap_btree.h" |
| 22 | #include "scrub/scrub.h" |
| 23 | #include "scrub/common.h" |
| 24 | #include "scrub/btree.h" |
| 25 | |
| 26 | /* Set us up with an inode's bmap. */ |
| 27 | int |
| 28 | xchk_setup_inode_bmap( |
| 29 | struct xfs_scrub *sc, |
| 30 | struct xfs_inode *ip) |
| 31 | { |
| 32 | int error; |
| 33 | |
| 34 | error = xchk_get_inode(sc, ip); |
| 35 | if (error) |
| 36 | goto out; |
| 37 | |
| 38 | sc->ilock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; |
| 39 | xfs_ilock(sc->ip, sc->ilock_flags); |
| 40 | |
| 41 | /* |
| 42 | * We don't want any ephemeral data fork updates sitting around |
| 43 | * while we inspect block mappings, so wait for directio to finish |
| 44 | * and flush dirty data if we have delalloc reservations. |
| 45 | */ |
| 46 | if (S_ISREG(VFS_I(sc->ip)->i_mode) && |
| 47 | sc->sm->sm_type == XFS_SCRUB_TYPE_BMBTD) { |
| 48 | struct address_space *mapping = VFS_I(sc->ip)->i_mapping; |
| 49 | |
| 50 | inode_dio_wait(VFS_I(sc->ip)); |
| 51 | |
| 52 | /* |
| 53 | * Try to flush all incore state to disk before we examine the |
| 54 | * space mappings for the data fork. Leave accumulated errors |
| 55 | * in the mapping for the writer threads to consume. |
| 56 | * |
| 57 | * On ENOSPC or EIO writeback errors, we continue into the |
| 58 | * extent mapping checks because write failures do not |
| 59 | * necessarily imply anything about the correctness of the file |
| 60 | * metadata. The metadata and the file data could be on |
| 61 | * completely separate devices; a media failure might only |
| 62 | * affect a subset of the disk, etc. We can handle delalloc |
| 63 | * extents in the scrubber, so leaving them in memory is fine. |
| 64 | */ |
| 65 | error = filemap_fdatawrite(mapping); |
| 66 | if (!error) |
| 67 | error = filemap_fdatawait_keep_errors(mapping); |
| 68 | if (error && (error != -ENOSPC && error != -EIO)) |
| 69 | goto out; |
| 70 | } |
| 71 | |
| 72 | /* Got the inode, lock it and we're ready to go. */ |
| 73 | error = xchk_trans_alloc(sc, 0); |
| 74 | if (error) |
| 75 | goto out; |
| 76 | sc->ilock_flags |= XFS_ILOCK_EXCL; |
| 77 | xfs_ilock(sc->ip, XFS_ILOCK_EXCL); |
| 78 | |
| 79 | out: |
| 80 | /* scrub teardown will unlock and release the inode */ |
| 81 | return error; |
| 82 | } |
| 83 | |
| 84 | /* |
| 85 | * Inode fork block mapping (BMBT) scrubber. |
| 86 | * More complex than the others because we have to scrub |
| 87 | * all the extents regardless of whether or not the fork |
| 88 | * is in btree format. |
| 89 | */ |
| 90 | |
| 91 | struct xchk_bmap_info { |
| 92 | struct xfs_scrub *sc; |
| 93 | xfs_fileoff_t lastoff; |
| 94 | bool is_rt; |
| 95 | bool is_shared; |
| 96 | bool was_loaded; |
| 97 | int whichfork; |
| 98 | }; |
| 99 | |
| 100 | /* Look for a corresponding rmap for this irec. */ |
| 101 | static inline bool |
| 102 | xchk_bmap_get_rmap( |
| 103 | struct xchk_bmap_info *info, |
| 104 | struct xfs_bmbt_irec *irec, |
| 105 | xfs_agblock_t agbno, |
| 106 | uint64_t owner, |
| 107 | struct xfs_rmap_irec *rmap) |
| 108 | { |
| 109 | xfs_fileoff_t offset; |
| 110 | unsigned int rflags = 0; |
| 111 | int has_rmap; |
| 112 | int error; |
| 113 | |
| 114 | if (info->whichfork == XFS_ATTR_FORK) |
| 115 | rflags |= XFS_RMAP_ATTR_FORK; |
| 116 | if (irec->br_state == XFS_EXT_UNWRITTEN) |
| 117 | rflags |= XFS_RMAP_UNWRITTEN; |
| 118 | |
| 119 | /* |
| 120 | * CoW staging extents are owned (on disk) by the refcountbt, so |
| 121 | * their rmaps do not have offsets. |
| 122 | */ |
| 123 | if (info->whichfork == XFS_COW_FORK) |
| 124 | offset = 0; |
| 125 | else |
| 126 | offset = irec->br_startoff; |
| 127 | |
| 128 | /* |
| 129 | * If the caller thinks this could be a shared bmbt extent (IOWs, |
| 130 | * any data fork extent of a reflink inode) then we have to use the |
| 131 | * range rmap lookup to make sure we get the correct owner/offset. |
| 132 | */ |
| 133 | if (info->is_shared) { |
| 134 | error = xfs_rmap_lookup_le_range(info->sc->sa.rmap_cur, agbno, |
| 135 | owner, offset, rflags, rmap, &has_rmap); |
| 136 | if (!xchk_should_check_xref(info->sc, &error, |
| 137 | &info->sc->sa.rmap_cur)) |
| 138 | return false; |
| 139 | goto out; |
| 140 | } |
| 141 | |
| 142 | /* |
| 143 | * Otherwise, use the (faster) regular lookup. |
| 144 | */ |
| 145 | error = xfs_rmap_lookup_le(info->sc->sa.rmap_cur, agbno, 0, owner, |
| 146 | offset, rflags, &has_rmap); |
| 147 | if (!xchk_should_check_xref(info->sc, &error, |
| 148 | &info->sc->sa.rmap_cur)) |
| 149 | return false; |
| 150 | if (!has_rmap) |
| 151 | goto out; |
| 152 | |
| 153 | error = xfs_rmap_get_rec(info->sc->sa.rmap_cur, rmap, &has_rmap); |
| 154 | if (!xchk_should_check_xref(info->sc, &error, |
| 155 | &info->sc->sa.rmap_cur)) |
| 156 | return false; |
| 157 | |
| 158 | out: |
| 159 | if (!has_rmap) |
| 160 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
| 161 | irec->br_startoff); |
| 162 | return has_rmap; |
| 163 | } |
| 164 | |
| 165 | /* Make sure that we have rmapbt records for this extent. */ |
| 166 | STATIC void |
| 167 | xchk_bmap_xref_rmap( |
| 168 | struct xchk_bmap_info *info, |
| 169 | struct xfs_bmbt_irec *irec, |
| 170 | xfs_agblock_t agbno) |
| 171 | { |
| 172 | struct xfs_rmap_irec rmap; |
| 173 | unsigned long long rmap_end; |
| 174 | uint64_t owner; |
| 175 | |
| 176 | if (!info->sc->sa.rmap_cur || xchk_skip_xref(info->sc->sm)) |
| 177 | return; |
| 178 | |
| 179 | if (info->whichfork == XFS_COW_FORK) |
| 180 | owner = XFS_RMAP_OWN_COW; |
| 181 | else |
| 182 | owner = info->sc->ip->i_ino; |
| 183 | |
| 184 | /* Find the rmap record for this irec. */ |
| 185 | if (!xchk_bmap_get_rmap(info, irec, agbno, owner, &rmap)) |
| 186 | return; |
| 187 | |
| 188 | /* Check the rmap. */ |
| 189 | rmap_end = (unsigned long long)rmap.rm_startblock + rmap.rm_blockcount; |
| 190 | if (rmap.rm_startblock > agbno || |
| 191 | agbno + irec->br_blockcount > rmap_end) |
| 192 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
| 193 | irec->br_startoff); |
| 194 | |
| 195 | /* |
| 196 | * Check the logical offsets if applicable. CoW staging extents |
| 197 | * don't track logical offsets since the mappings only exist in |
| 198 | * memory. |
| 199 | */ |
| 200 | if (info->whichfork != XFS_COW_FORK) { |
| 201 | rmap_end = (unsigned long long)rmap.rm_offset + |
| 202 | rmap.rm_blockcount; |
| 203 | if (rmap.rm_offset > irec->br_startoff || |
| 204 | irec->br_startoff + irec->br_blockcount > rmap_end) |
| 205 | xchk_fblock_xref_set_corrupt(info->sc, |
| 206 | info->whichfork, irec->br_startoff); |
| 207 | } |
| 208 | |
| 209 | if (rmap.rm_owner != owner) |
| 210 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
| 211 | irec->br_startoff); |
| 212 | |
| 213 | /* |
| 214 | * Check for discrepancies between the unwritten flag in the irec and |
| 215 | * the rmap. Note that the (in-memory) CoW fork distinguishes between |
| 216 | * unwritten and written extents, but we don't track that in the rmap |
| 217 | * records because the blocks are owned (on-disk) by the refcountbt, |
| 218 | * which doesn't track unwritten state. |
| 219 | */ |
| 220 | if (owner != XFS_RMAP_OWN_COW && |
| 221 | !!(irec->br_state == XFS_EXT_UNWRITTEN) != |
| 222 | !!(rmap.rm_flags & XFS_RMAP_UNWRITTEN)) |
| 223 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
| 224 | irec->br_startoff); |
| 225 | |
| 226 | if (!!(info->whichfork == XFS_ATTR_FORK) != |
| 227 | !!(rmap.rm_flags & XFS_RMAP_ATTR_FORK)) |
| 228 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
| 229 | irec->br_startoff); |
| 230 | if (rmap.rm_flags & XFS_RMAP_BMBT_BLOCK) |
| 231 | xchk_fblock_xref_set_corrupt(info->sc, info->whichfork, |
| 232 | irec->br_startoff); |
| 233 | } |
| 234 | |
| 235 | /* Cross-reference a single rtdev extent record. */ |
| 236 | STATIC void |
| 237 | xchk_bmap_rt_iextent_xref( |
| 238 | struct xfs_inode *ip, |
| 239 | struct xchk_bmap_info *info, |
| 240 | struct xfs_bmbt_irec *irec) |
| 241 | { |
| 242 | xchk_xref_is_used_rt_space(info->sc, irec->br_startblock, |
| 243 | irec->br_blockcount); |
| 244 | } |
| 245 | |
| 246 | /* Cross-reference a single datadev extent record. */ |
| 247 | STATIC void |
| 248 | xchk_bmap_iextent_xref( |
| 249 | struct xfs_inode *ip, |
| 250 | struct xchk_bmap_info *info, |
| 251 | struct xfs_bmbt_irec *irec) |
| 252 | { |
| 253 | struct xfs_mount *mp = info->sc->mp; |
| 254 | xfs_agnumber_t agno; |
| 255 | xfs_agblock_t agbno; |
| 256 | xfs_extlen_t len; |
| 257 | int error; |
| 258 | |
| 259 | agno = XFS_FSB_TO_AGNO(mp, irec->br_startblock); |
| 260 | agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock); |
| 261 | len = irec->br_blockcount; |
| 262 | |
| 263 | error = xchk_ag_init(info->sc, agno, &info->sc->sa); |
| 264 | if (!xchk_fblock_process_error(info->sc, info->whichfork, |
| 265 | irec->br_startoff, &error)) |
| 266 | return; |
| 267 | |
| 268 | xchk_xref_is_used_space(info->sc, agbno, len); |
| 269 | xchk_xref_is_not_inode_chunk(info->sc, agbno, len); |
| 270 | xchk_bmap_xref_rmap(info, irec, agbno); |
| 271 | switch (info->whichfork) { |
| 272 | case XFS_DATA_FORK: |
| 273 | if (xfs_is_reflink_inode(info->sc->ip)) |
| 274 | break; |
| 275 | /* fall through */ |
| 276 | case XFS_ATTR_FORK: |
| 277 | xchk_xref_is_not_shared(info->sc, agbno, |
| 278 | irec->br_blockcount); |
| 279 | break; |
| 280 | case XFS_COW_FORK: |
| 281 | xchk_xref_is_cow_staging(info->sc, agbno, |
| 282 | irec->br_blockcount); |
| 283 | break; |
| 284 | } |
| 285 | |
| 286 | xchk_ag_free(info->sc, &info->sc->sa); |
| 287 | } |
| 288 | |
| 289 | /* |
| 290 | * Directories and attr forks should never have blocks that can't be addressed |
| 291 | * by a xfs_dablk_t. |
| 292 | */ |
| 293 | STATIC void |
| 294 | xchk_bmap_dirattr_extent( |
| 295 | struct xfs_inode *ip, |
| 296 | struct xchk_bmap_info *info, |
| 297 | struct xfs_bmbt_irec *irec) |
| 298 | { |
| 299 | struct xfs_mount *mp = ip->i_mount; |
| 300 | xfs_fileoff_t off; |
| 301 | |
| 302 | if (!S_ISDIR(VFS_I(ip)->i_mode) && info->whichfork != XFS_ATTR_FORK) |
| 303 | return; |
| 304 | |
| 305 | if (!xfs_verify_dablk(mp, irec->br_startoff)) |
| 306 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 307 | irec->br_startoff); |
| 308 | |
| 309 | off = irec->br_startoff + irec->br_blockcount - 1; |
| 310 | if (!xfs_verify_dablk(mp, off)) |
| 311 | xchk_fblock_set_corrupt(info->sc, info->whichfork, off); |
| 312 | } |
| 313 | |
| 314 | /* Scrub a single extent record. */ |
| 315 | STATIC int |
| 316 | xchk_bmap_iextent( |
| 317 | struct xfs_inode *ip, |
| 318 | struct xchk_bmap_info *info, |
| 319 | struct xfs_bmbt_irec *irec) |
| 320 | { |
| 321 | struct xfs_mount *mp = info->sc->mp; |
| 322 | xfs_filblks_t end; |
| 323 | int error = 0; |
| 324 | |
| 325 | /* |
| 326 | * Check for out-of-order extents. This record could have come |
| 327 | * from the incore list, for which there is no ordering check. |
| 328 | */ |
| 329 | if (irec->br_startoff < info->lastoff) |
| 330 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 331 | irec->br_startoff); |
| 332 | |
| 333 | xchk_bmap_dirattr_extent(ip, info, irec); |
| 334 | |
| 335 | /* There should never be a "hole" extent in either extent list. */ |
| 336 | if (irec->br_startblock == HOLESTARTBLOCK) |
| 337 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 338 | irec->br_startoff); |
| 339 | |
| 340 | /* |
| 341 | * Check for delalloc extents. We never iterate the ones in the |
| 342 | * in-core extent scan, and we should never see these in the bmbt. |
| 343 | */ |
| 344 | if (isnullstartblock(irec->br_startblock)) |
| 345 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 346 | irec->br_startoff); |
| 347 | |
| 348 | /* Make sure the extent points to a valid place. */ |
| 349 | if (irec->br_blockcount > MAXEXTLEN) |
| 350 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 351 | irec->br_startoff); |
| 352 | if (irec->br_startblock + irec->br_blockcount <= irec->br_startblock) |
| 353 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 354 | irec->br_startoff); |
| 355 | end = irec->br_startblock + irec->br_blockcount - 1; |
| 356 | if (info->is_rt && |
| 357 | (!xfs_verify_rtbno(mp, irec->br_startblock) || |
| 358 | !xfs_verify_rtbno(mp, end))) |
| 359 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 360 | irec->br_startoff); |
| 361 | if (!info->is_rt && |
| 362 | (!xfs_verify_fsbno(mp, irec->br_startblock) || |
| 363 | !xfs_verify_fsbno(mp, end) || |
| 364 | XFS_FSB_TO_AGNO(mp, irec->br_startblock) != |
| 365 | XFS_FSB_TO_AGNO(mp, end))) |
| 366 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 367 | irec->br_startoff); |
| 368 | |
| 369 | /* We don't allow unwritten extents on attr forks. */ |
| 370 | if (irec->br_state == XFS_EXT_UNWRITTEN && |
| 371 | info->whichfork == XFS_ATTR_FORK) |
| 372 | xchk_fblock_set_corrupt(info->sc, info->whichfork, |
| 373 | irec->br_startoff); |
| 374 | |
| 375 | if (info->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) |
| 376 | return 0; |
| 377 | |
| 378 | if (info->is_rt) |
| 379 | xchk_bmap_rt_iextent_xref(ip, info, irec); |
| 380 | else |
| 381 | xchk_bmap_iextent_xref(ip, info, irec); |
| 382 | |
| 383 | info->lastoff = irec->br_startoff + irec->br_blockcount; |
| 384 | return error; |
| 385 | } |
| 386 | |
| 387 | /* Scrub a bmbt record. */ |
| 388 | STATIC int |
| 389 | xchk_bmapbt_rec( |
| 390 | struct xchk_btree *bs, |
| 391 | union xfs_btree_rec *rec) |
| 392 | { |
| 393 | struct xfs_bmbt_irec irec; |
| 394 | struct xfs_bmbt_irec iext_irec; |
| 395 | struct xfs_iext_cursor icur; |
| 396 | struct xchk_bmap_info *info = bs->private; |
| 397 | struct xfs_inode *ip = bs->cur->bc_private.b.ip; |
| 398 | struct xfs_buf *bp = NULL; |
| 399 | struct xfs_btree_block *block; |
| 400 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, info->whichfork); |
| 401 | uint64_t owner; |
| 402 | int i; |
| 403 | |
| 404 | /* |
| 405 | * Check the owners of the btree blocks up to the level below |
| 406 | * the root since the verifiers don't do that. |
| 407 | */ |
| 408 | if (xfs_sb_version_hascrc(&bs->cur->bc_mp->m_sb) && |
| 409 | bs->cur->bc_ptrs[0] == 1) { |
| 410 | for (i = 0; i < bs->cur->bc_nlevels - 1; i++) { |
| 411 | block = xfs_btree_get_block(bs->cur, i, &bp); |
| 412 | owner = be64_to_cpu(block->bb_u.l.bb_owner); |
| 413 | if (owner != ip->i_ino) |
| 414 | xchk_fblock_set_corrupt(bs->sc, |
| 415 | info->whichfork, 0); |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | /* |
| 420 | * Check that the incore extent tree contains an extent that matches |
| 421 | * this one exactly. We validate those cached bmaps later, so we don't |
| 422 | * need to check them here. If the incore extent tree was just loaded |
| 423 | * from disk by the scrubber, we assume that its contents match what's |
| 424 | * on disk (we still hold the ILOCK) and skip the equivalence check. |
| 425 | */ |
| 426 | if (!info->was_loaded) |
| 427 | return 0; |
| 428 | |
| 429 | xfs_bmbt_disk_get_all(&rec->bmbt, &irec); |
| 430 | if (!xfs_iext_lookup_extent(ip, ifp, irec.br_startoff, &icur, |
| 431 | &iext_irec) || |
| 432 | irec.br_startoff != iext_irec.br_startoff || |
| 433 | irec.br_startblock != iext_irec.br_startblock || |
| 434 | irec.br_blockcount != iext_irec.br_blockcount || |
| 435 | irec.br_state != iext_irec.br_state) |
| 436 | xchk_fblock_set_corrupt(bs->sc, info->whichfork, |
| 437 | irec.br_startoff); |
| 438 | return 0; |
| 439 | } |
| 440 | |
| 441 | /* Scan the btree records. */ |
| 442 | STATIC int |
| 443 | xchk_bmap_btree( |
| 444 | struct xfs_scrub *sc, |
| 445 | int whichfork, |
| 446 | struct xchk_bmap_info *info) |
| 447 | { |
| 448 | struct xfs_owner_info oinfo; |
| 449 | struct xfs_ifork *ifp = XFS_IFORK_PTR(sc->ip, whichfork); |
| 450 | struct xfs_mount *mp = sc->mp; |
| 451 | struct xfs_inode *ip = sc->ip; |
| 452 | struct xfs_btree_cur *cur; |
| 453 | int error; |
| 454 | |
| 455 | /* Load the incore bmap cache if it's not loaded. */ |
| 456 | info->was_loaded = ifp->if_flags & XFS_IFEXTENTS; |
| 457 | if (!info->was_loaded) { |
| 458 | error = xfs_iread_extents(sc->tp, ip, whichfork); |
| 459 | if (!xchk_fblock_process_error(sc, whichfork, 0, &error)) |
| 460 | goto out; |
| 461 | } |
| 462 | |
| 463 | /* Check the btree structure. */ |
| 464 | cur = xfs_bmbt_init_cursor(mp, sc->tp, ip, whichfork); |
| 465 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); |
| 466 | error = xchk_btree(sc, cur, xchk_bmapbt_rec, &oinfo, info); |
| 467 | xfs_btree_del_cursor(cur, error); |
| 468 | out: |
| 469 | return error; |
| 470 | } |
| 471 | |
| 472 | struct xchk_bmap_check_rmap_info { |
| 473 | struct xfs_scrub *sc; |
| 474 | int whichfork; |
| 475 | struct xfs_iext_cursor icur; |
| 476 | }; |
| 477 | |
| 478 | /* Can we find bmaps that fit this rmap? */ |
| 479 | STATIC int |
| 480 | xchk_bmap_check_rmap( |
| 481 | struct xfs_btree_cur *cur, |
| 482 | struct xfs_rmap_irec *rec, |
| 483 | void *priv) |
| 484 | { |
| 485 | struct xfs_bmbt_irec irec; |
| 486 | struct xchk_bmap_check_rmap_info *sbcri = priv; |
| 487 | struct xfs_ifork *ifp; |
| 488 | struct xfs_scrub *sc = sbcri->sc; |
| 489 | bool have_map; |
| 490 | |
| 491 | /* Is this even the right fork? */ |
| 492 | if (rec->rm_owner != sc->ip->i_ino) |
| 493 | return 0; |
| 494 | if ((sbcri->whichfork == XFS_ATTR_FORK) ^ |
| 495 | !!(rec->rm_flags & XFS_RMAP_ATTR_FORK)) |
| 496 | return 0; |
| 497 | if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK) |
| 498 | return 0; |
| 499 | |
| 500 | /* Now look up the bmbt record. */ |
| 501 | ifp = XFS_IFORK_PTR(sc->ip, sbcri->whichfork); |
| 502 | if (!ifp) { |
| 503 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
| 504 | rec->rm_offset); |
| 505 | goto out; |
| 506 | } |
| 507 | have_map = xfs_iext_lookup_extent(sc->ip, ifp, rec->rm_offset, |
| 508 | &sbcri->icur, &irec); |
| 509 | if (!have_map) |
| 510 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
| 511 | rec->rm_offset); |
| 512 | /* |
| 513 | * bmap extent record lengths are constrained to 2^21 blocks in length |
| 514 | * because of space constraints in the on-disk metadata structure. |
| 515 | * However, rmap extent record lengths are constrained only by AG |
| 516 | * length, so we have to loop through the bmbt to make sure that the |
| 517 | * entire rmap is covered by bmbt records. |
| 518 | */ |
| 519 | while (have_map) { |
| 520 | if (irec.br_startoff != rec->rm_offset) |
| 521 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
| 522 | rec->rm_offset); |
| 523 | if (irec.br_startblock != XFS_AGB_TO_FSB(sc->mp, |
| 524 | cur->bc_private.a.agno, rec->rm_startblock)) |
| 525 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
| 526 | rec->rm_offset); |
| 527 | if (irec.br_blockcount > rec->rm_blockcount) |
| 528 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
| 529 | rec->rm_offset); |
| 530 | if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) |
| 531 | break; |
| 532 | rec->rm_startblock += irec.br_blockcount; |
| 533 | rec->rm_offset += irec.br_blockcount; |
| 534 | rec->rm_blockcount -= irec.br_blockcount; |
| 535 | if (rec->rm_blockcount == 0) |
| 536 | break; |
| 537 | have_map = xfs_iext_next_extent(ifp, &sbcri->icur, &irec); |
| 538 | if (!have_map) |
| 539 | xchk_fblock_set_corrupt(sc, sbcri->whichfork, |
| 540 | rec->rm_offset); |
| 541 | } |
| 542 | |
| 543 | out: |
| 544 | if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) |
| 545 | return -ECANCELED; |
| 546 | return 0; |
| 547 | } |
| 548 | |
| 549 | /* Make sure each rmap has a corresponding bmbt entry. */ |
| 550 | STATIC int |
| 551 | xchk_bmap_check_ag_rmaps( |
| 552 | struct xfs_scrub *sc, |
| 553 | int whichfork, |
| 554 | xfs_agnumber_t agno) |
| 555 | { |
| 556 | struct xchk_bmap_check_rmap_info sbcri; |
| 557 | struct xfs_btree_cur *cur; |
| 558 | struct xfs_buf *agf; |
| 559 | int error; |
| 560 | |
| 561 | error = xfs_alloc_read_agf(sc->mp, sc->tp, agno, 0, &agf); |
| 562 | if (error) |
| 563 | return error; |
| 564 | |
| 565 | cur = xfs_rmapbt_init_cursor(sc->mp, sc->tp, agf, agno); |
| 566 | if (!cur) { |
| 567 | error = -ENOMEM; |
| 568 | goto out_agf; |
| 569 | } |
| 570 | |
| 571 | sbcri.sc = sc; |
| 572 | sbcri.whichfork = whichfork; |
| 573 | error = xfs_rmap_query_all(cur, xchk_bmap_check_rmap, &sbcri); |
| 574 | if (error == -ECANCELED) |
| 575 | error = 0; |
| 576 | |
| 577 | xfs_btree_del_cursor(cur, error); |
| 578 | out_agf: |
| 579 | xfs_trans_brelse(sc->tp, agf); |
| 580 | return error; |
| 581 | } |
| 582 | |
| 583 | /* Make sure each rmap has a corresponding bmbt entry. */ |
| 584 | STATIC int |
| 585 | xchk_bmap_check_rmaps( |
| 586 | struct xfs_scrub *sc, |
| 587 | int whichfork) |
| 588 | { |
| 589 | loff_t size; |
| 590 | xfs_agnumber_t agno; |
| 591 | int error; |
| 592 | |
| 593 | if (!xfs_sb_version_hasrmapbt(&sc->mp->m_sb) || |
| 594 | whichfork == XFS_COW_FORK || |
| 595 | (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) |
| 596 | return 0; |
| 597 | |
| 598 | /* Don't support realtime rmap checks yet. */ |
| 599 | if (XFS_IS_REALTIME_INODE(sc->ip) && whichfork == XFS_DATA_FORK) |
| 600 | return 0; |
| 601 | |
| 602 | /* |
| 603 | * Only do this for complex maps that are in btree format, or for |
| 604 | * situations where we would seem to have a size but zero extents. |
| 605 | * The inode repair code can zap broken iforks, which means we have |
| 606 | * to flag this bmap as corrupt if there are rmaps that need to be |
| 607 | * reattached. |
| 608 | */ |
| 609 | switch (whichfork) { |
| 610 | case XFS_DATA_FORK: |
| 611 | size = i_size_read(VFS_I(sc->ip)); |
| 612 | break; |
| 613 | case XFS_ATTR_FORK: |
| 614 | size = XFS_IFORK_Q(sc->ip); |
| 615 | break; |
| 616 | default: |
| 617 | size = 0; |
| 618 | break; |
| 619 | } |
| 620 | if (XFS_IFORK_FORMAT(sc->ip, whichfork) != XFS_DINODE_FMT_BTREE && |
| 621 | (size == 0 || XFS_IFORK_NEXTENTS(sc->ip, whichfork) > 0)) |
| 622 | return 0; |
| 623 | |
| 624 | for (agno = 0; agno < sc->mp->m_sb.sb_agcount; agno++) { |
| 625 | error = xchk_bmap_check_ag_rmaps(sc, whichfork, agno); |
| 626 | if (error) |
| 627 | return error; |
| 628 | if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) |
| 629 | break; |
| 630 | } |
| 631 | |
| 632 | return 0; |
| 633 | } |
| 634 | |
| 635 | /* |
| 636 | * Scrub an inode fork's block mappings. |
| 637 | * |
| 638 | * First we scan every record in every btree block, if applicable. |
| 639 | * Then we unconditionally scan the incore extent cache. |
| 640 | */ |
| 641 | STATIC int |
| 642 | xchk_bmap( |
| 643 | struct xfs_scrub *sc, |
| 644 | int whichfork) |
| 645 | { |
| 646 | struct xfs_bmbt_irec irec; |
| 647 | struct xchk_bmap_info info = { NULL }; |
| 648 | struct xfs_mount *mp = sc->mp; |
| 649 | struct xfs_inode *ip = sc->ip; |
| 650 | struct xfs_ifork *ifp; |
| 651 | xfs_fileoff_t endoff; |
| 652 | struct xfs_iext_cursor icur; |
| 653 | int error = 0; |
| 654 | |
| 655 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 656 | |
| 657 | info.is_rt = whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip); |
| 658 | info.whichfork = whichfork; |
| 659 | info.is_shared = whichfork == XFS_DATA_FORK && xfs_is_reflink_inode(ip); |
| 660 | info.sc = sc; |
| 661 | |
| 662 | switch (whichfork) { |
| 663 | case XFS_COW_FORK: |
| 664 | /* Non-existent CoW forks are ignorable. */ |
| 665 | if (!ifp) |
| 666 | goto out; |
| 667 | /* No CoW forks on non-reflink inodes/filesystems. */ |
| 668 | if (!xfs_is_reflink_inode(ip)) { |
| 669 | xchk_ino_set_corrupt(sc, sc->ip->i_ino); |
| 670 | goto out; |
| 671 | } |
| 672 | break; |
| 673 | case XFS_ATTR_FORK: |
| 674 | if (!ifp) |
| 675 | goto out_check_rmap; |
| 676 | if (!xfs_sb_version_hasattr(&mp->m_sb) && |
| 677 | !xfs_sb_version_hasattr2(&mp->m_sb)) |
| 678 | xchk_ino_set_corrupt(sc, sc->ip->i_ino); |
| 679 | break; |
| 680 | default: |
| 681 | ASSERT(whichfork == XFS_DATA_FORK); |
| 682 | break; |
| 683 | } |
| 684 | |
| 685 | /* Check the fork values */ |
| 686 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { |
| 687 | case XFS_DINODE_FMT_UUID: |
| 688 | case XFS_DINODE_FMT_DEV: |
| 689 | case XFS_DINODE_FMT_LOCAL: |
| 690 | /* No mappings to check. */ |
| 691 | goto out; |
| 692 | case XFS_DINODE_FMT_EXTENTS: |
| 693 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
| 694 | xchk_fblock_set_corrupt(sc, whichfork, 0); |
| 695 | goto out; |
| 696 | } |
| 697 | break; |
| 698 | case XFS_DINODE_FMT_BTREE: |
| 699 | if (whichfork == XFS_COW_FORK) { |
| 700 | xchk_fblock_set_corrupt(sc, whichfork, 0); |
| 701 | goto out; |
| 702 | } |
| 703 | |
| 704 | error = xchk_bmap_btree(sc, whichfork, &info); |
| 705 | if (error) |
| 706 | goto out; |
| 707 | break; |
| 708 | default: |
| 709 | xchk_fblock_set_corrupt(sc, whichfork, 0); |
| 710 | goto out; |
| 711 | } |
| 712 | |
| 713 | if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) |
| 714 | goto out; |
| 715 | |
| 716 | /* Find the offset of the last extent in the mapping. */ |
| 717 | error = xfs_bmap_last_offset(ip, &endoff, whichfork); |
| 718 | if (!xchk_fblock_process_error(sc, whichfork, 0, &error)) |
| 719 | goto out; |
| 720 | |
| 721 | /* Scrub extent records. */ |
| 722 | info.lastoff = 0; |
| 723 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 724 | for_each_xfs_iext(ifp, &icur, &irec) { |
| 725 | if (xchk_should_terminate(sc, &error) || |
| 726 | (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) |
| 727 | goto out; |
| 728 | if (isnullstartblock(irec.br_startblock)) |
| 729 | continue; |
| 730 | if (irec.br_startoff >= endoff) { |
| 731 | xchk_fblock_set_corrupt(sc, whichfork, |
| 732 | irec.br_startoff); |
| 733 | goto out; |
| 734 | } |
| 735 | error = xchk_bmap_iextent(ip, &info, &irec); |
| 736 | if (error) |
| 737 | goto out; |
| 738 | } |
| 739 | |
| 740 | out_check_rmap: |
| 741 | error = xchk_bmap_check_rmaps(sc, whichfork); |
| 742 | if (!xchk_fblock_xref_process_error(sc, whichfork, 0, &error)) |
| 743 | goto out; |
| 744 | out: |
| 745 | return error; |
| 746 | } |
| 747 | |
| 748 | /* Scrub an inode's data fork. */ |
| 749 | int |
| 750 | xchk_bmap_data( |
| 751 | struct xfs_scrub *sc) |
| 752 | { |
| 753 | return xchk_bmap(sc, XFS_DATA_FORK); |
| 754 | } |
| 755 | |
| 756 | /* Scrub an inode's attr fork. */ |
| 757 | int |
| 758 | xchk_bmap_attr( |
| 759 | struct xfs_scrub *sc) |
| 760 | { |
| 761 | return xchk_bmap(sc, XFS_ATTR_FORK); |
| 762 | } |
| 763 | |
| 764 | /* Scrub an inode's CoW fork. */ |
| 765 | int |
| 766 | xchk_bmap_cow( |
| 767 | struct xfs_scrub *sc) |
| 768 | { |
| 769 | if (!xfs_is_reflink_inode(sc->ip)) |
| 770 | return -ENOENT; |
| 771 | |
| 772 | return xchk_bmap(sc, XFS_COW_FORK); |
| 773 | } |