1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * Copyright (C) 2018 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_log_format.h" 14 #include "xfs_trans.h" 15 #include "xfs_sb.h" 16 #include "xfs_alloc.h" 17 #include "xfs_alloc_btree.h" 18 #include "xfs_ialloc.h" 19 #include "xfs_ialloc_btree.h" 20 #include "xfs_rmap.h" 21 #include "xfs_rmap_btree.h" 22 #include "xfs_refcount_btree.h" 23 #include "xfs_ag.h" 24 #include "scrub/scrub.h" 25 #include "scrub/common.h" 26 #include "scrub/trace.h" 27 #include "scrub/repair.h" 28 #include "scrub/bitmap.h" 29 30 /* Superblock */ 31 32 /* Repair the superblock. */ 33 int 34 xrep_superblock( 35 struct xfs_scrub *sc) 36 { 37 struct xfs_mount *mp = sc->mp; 38 struct xfs_buf *bp; 39 xfs_agnumber_t agno; 40 int error; 41 42 /* Don't try to repair AG 0's sb; let xfs_repair deal with it. */ 43 agno = sc->sm->sm_agno; 44 if (agno == 0) 45 return -EOPNOTSUPP; 46 47 error = xfs_sb_get_secondary(mp, sc->tp, agno, &bp); 48 if (error) 49 return error; 50 51 /* Copy AG 0's superblock to this one. */ 52 xfs_buf_zero(bp, 0, BBTOB(bp->b_length)); 53 xfs_sb_to_disk(bp->b_addr, &mp->m_sb); 54 55 /* Write this to disk. */ 56 xfs_trans_buf_set_type(sc->tp, bp, XFS_BLFT_SB_BUF); 57 xfs_trans_log_buf(sc->tp, bp, 0, BBTOB(bp->b_length) - 1); 58 return error; 59 } 60 61 /* AGF */ 62 63 struct xrep_agf_allocbt { 64 struct xfs_scrub *sc; 65 xfs_agblock_t freeblks; 66 xfs_agblock_t longest; 67 }; 68 69 /* Record free space shape information. */ 70 STATIC int 71 xrep_agf_walk_allocbt( 72 struct xfs_btree_cur *cur, 73 const struct xfs_alloc_rec_incore *rec, 74 void *priv) 75 { 76 struct xrep_agf_allocbt *raa = priv; 77 int error = 0; 78 79 if (xchk_should_terminate(raa->sc, &error)) 80 return error; 81 82 raa->freeblks += rec->ar_blockcount; 83 if (rec->ar_blockcount > raa->longest) 84 raa->longest = rec->ar_blockcount; 85 return error; 86 } 87 88 /* Does this AGFL block look sane? */ 89 STATIC int 90 xrep_agf_check_agfl_block( 91 struct xfs_mount *mp, 92 xfs_agblock_t agbno, 93 void *priv) 94 { 95 struct xfs_scrub *sc = priv; 96 97 if (!xfs_verify_agbno(mp, sc->sa.pag->pag_agno, agbno)) 98 return -EFSCORRUPTED; 99 return 0; 100 } 101 102 /* 103 * Offset within the xrep_find_ag_btree array for each btree type. Avoid the 104 * XFS_BTNUM_ names here to avoid creating a sparse array. 105 */ 106 enum { 107 XREP_AGF_BNOBT = 0, 108 XREP_AGF_CNTBT, 109 XREP_AGF_RMAPBT, 110 XREP_AGF_REFCOUNTBT, 111 XREP_AGF_END, 112 XREP_AGF_MAX 113 }; 114 115 /* Check a btree root candidate. */ 116 static inline bool 117 xrep_check_btree_root( 118 struct xfs_scrub *sc, 119 struct xrep_find_ag_btree *fab) 120 { 121 struct xfs_mount *mp = sc->mp; 122 xfs_agnumber_t agno = sc->sm->sm_agno; 123 124 return xfs_verify_agbno(mp, agno, fab->root) && 125 fab->height <= fab->maxlevels; 126 } 127 128 /* 129 * Given the btree roots described by *fab, find the roots, check them for 130 * sanity, and pass the root data back out via *fab. 131 * 132 * This is /also/ a chicken and egg problem because we have to use the rmapbt 133 * (rooted in the AGF) to find the btrees rooted in the AGF. We also have no 134 * idea if the btrees make any sense. If we hit obvious corruptions in those 135 * btrees we'll bail out. 136 */ 137 STATIC int 138 xrep_agf_find_btrees( 139 struct xfs_scrub *sc, 140 struct xfs_buf *agf_bp, 141 struct xrep_find_ag_btree *fab, 142 struct xfs_buf *agfl_bp) 143 { 144 struct xfs_agf *old_agf = agf_bp->b_addr; 145 int error; 146 147 /* Go find the root data. */ 148 error = xrep_find_ag_btree_roots(sc, agf_bp, fab, agfl_bp); 149 if (error) 150 return error; 151 152 /* We must find the bnobt, cntbt, and rmapbt roots. */ 153 if (!xrep_check_btree_root(sc, &fab[XREP_AGF_BNOBT]) || 154 !xrep_check_btree_root(sc, &fab[XREP_AGF_CNTBT]) || 155 !xrep_check_btree_root(sc, &fab[XREP_AGF_RMAPBT])) 156 return -EFSCORRUPTED; 157 158 /* 159 * We relied on the rmapbt to reconstruct the AGF. If we get a 160 * different root then something's seriously wrong. 161 */ 162 if (fab[XREP_AGF_RMAPBT].root != 163 be32_to_cpu(old_agf->agf_roots[XFS_BTNUM_RMAPi])) 164 return -EFSCORRUPTED; 165 166 /* We must find the refcountbt root if that feature is enabled. */ 167 if (xfs_has_reflink(sc->mp) && 168 !xrep_check_btree_root(sc, &fab[XREP_AGF_REFCOUNTBT])) 169 return -EFSCORRUPTED; 170 171 return 0; 172 } 173 174 /* 175 * Reinitialize the AGF header, making an in-core copy of the old contents so 176 * that we know which in-core state needs to be reinitialized. 177 */ 178 STATIC void 179 xrep_agf_init_header( 180 struct xfs_scrub *sc, 181 struct xfs_buf *agf_bp, 182 struct xfs_agf *old_agf) 183 { 184 struct xfs_mount *mp = sc->mp; 185 struct xfs_agf *agf = agf_bp->b_addr; 186 187 memcpy(old_agf, agf, sizeof(*old_agf)); 188 memset(agf, 0, BBTOB(agf_bp->b_length)); 189 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC); 190 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION); 191 agf->agf_seqno = cpu_to_be32(sc->sa.pag->pag_agno); 192 agf->agf_length = cpu_to_be32(xfs_ag_block_count(mp, 193 sc->sa.pag->pag_agno)); 194 agf->agf_flfirst = old_agf->agf_flfirst; 195 agf->agf_fllast = old_agf->agf_fllast; 196 agf->agf_flcount = old_agf->agf_flcount; 197 if (xfs_has_crc(mp)) 198 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid); 199 200 /* Mark the incore AGF data stale until we're done fixing things. */ 201 ASSERT(sc->sa.pag->pagf_init); 202 sc->sa.pag->pagf_init = 0; 203 } 204 205 /* Set btree root information in an AGF. */ 206 STATIC void 207 xrep_agf_set_roots( 208 struct xfs_scrub *sc, 209 struct xfs_agf *agf, 210 struct xrep_find_ag_btree *fab) 211 { 212 agf->agf_roots[XFS_BTNUM_BNOi] = 213 cpu_to_be32(fab[XREP_AGF_BNOBT].root); 214 agf->agf_levels[XFS_BTNUM_BNOi] = 215 cpu_to_be32(fab[XREP_AGF_BNOBT].height); 216 217 agf->agf_roots[XFS_BTNUM_CNTi] = 218 cpu_to_be32(fab[XREP_AGF_CNTBT].root); 219 agf->agf_levels[XFS_BTNUM_CNTi] = 220 cpu_to_be32(fab[XREP_AGF_CNTBT].height); 221 222 agf->agf_roots[XFS_BTNUM_RMAPi] = 223 cpu_to_be32(fab[XREP_AGF_RMAPBT].root); 224 agf->agf_levels[XFS_BTNUM_RMAPi] = 225 cpu_to_be32(fab[XREP_AGF_RMAPBT].height); 226 227 if (xfs_has_reflink(sc->mp)) { 228 agf->agf_refcount_root = 229 cpu_to_be32(fab[XREP_AGF_REFCOUNTBT].root); 230 agf->agf_refcount_level = 231 cpu_to_be32(fab[XREP_AGF_REFCOUNTBT].height); 232 } 233 } 234 235 /* Update all AGF fields which derive from btree contents. */ 236 STATIC int 237 xrep_agf_calc_from_btrees( 238 struct xfs_scrub *sc, 239 struct xfs_buf *agf_bp) 240 { 241 struct xrep_agf_allocbt raa = { .sc = sc }; 242 struct xfs_btree_cur *cur = NULL; 243 struct xfs_agf *agf = agf_bp->b_addr; 244 struct xfs_mount *mp = sc->mp; 245 xfs_agblock_t btreeblks; 246 xfs_agblock_t blocks; 247 int error; 248 249 /* Update the AGF counters from the bnobt. */ 250 cur = xfs_allocbt_init_cursor(mp, sc->tp, agf_bp, 251 sc->sa.pag, XFS_BTNUM_BNO); 252 error = xfs_alloc_query_all(cur, xrep_agf_walk_allocbt, &raa); 253 if (error) 254 goto err; 255 error = xfs_btree_count_blocks(cur, &blocks); 256 if (error) 257 goto err; 258 xfs_btree_del_cursor(cur, error); 259 btreeblks = blocks - 1; 260 agf->agf_freeblks = cpu_to_be32(raa.freeblks); 261 agf->agf_longest = cpu_to_be32(raa.longest); 262 263 /* Update the AGF counters from the cntbt. */ 264 cur = xfs_allocbt_init_cursor(mp, sc->tp, agf_bp, 265 sc->sa.pag, XFS_BTNUM_CNT); 266 error = xfs_btree_count_blocks(cur, &blocks); 267 if (error) 268 goto err; 269 xfs_btree_del_cursor(cur, error); 270 btreeblks += blocks - 1; 271 272 /* Update the AGF counters from the rmapbt. */ 273 cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.pag); 274 error = xfs_btree_count_blocks(cur, &blocks); 275 if (error) 276 goto err; 277 xfs_btree_del_cursor(cur, error); 278 agf->agf_rmap_blocks = cpu_to_be32(blocks); 279 btreeblks += blocks - 1; 280 281 agf->agf_btreeblks = cpu_to_be32(btreeblks); 282 283 /* Update the AGF counters from the refcountbt. */ 284 if (xfs_has_reflink(mp)) { 285 cur = xfs_refcountbt_init_cursor(mp, sc->tp, agf_bp, 286 sc->sa.pag); 287 error = xfs_btree_count_blocks(cur, &blocks); 288 if (error) 289 goto err; 290 xfs_btree_del_cursor(cur, error); 291 agf->agf_refcount_blocks = cpu_to_be32(blocks); 292 } 293 294 return 0; 295 err: 296 xfs_btree_del_cursor(cur, error); 297 return error; 298 } 299 300 /* Commit the new AGF and reinitialize the incore state. */ 301 STATIC int 302 xrep_agf_commit_new( 303 struct xfs_scrub *sc, 304 struct xfs_buf *agf_bp) 305 { 306 struct xfs_perag *pag; 307 struct xfs_agf *agf = agf_bp->b_addr; 308 309 /* Trigger fdblocks recalculation */ 310 xfs_force_summary_recalc(sc->mp); 311 312 /* Write this to disk. */ 313 xfs_trans_buf_set_type(sc->tp, agf_bp, XFS_BLFT_AGF_BUF); 314 xfs_trans_log_buf(sc->tp, agf_bp, 0, BBTOB(agf_bp->b_length) - 1); 315 316 /* Now reinitialize the in-core counters we changed. */ 317 pag = sc->sa.pag; 318 pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); 319 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); 320 pag->pagf_longest = be32_to_cpu(agf->agf_longest); 321 pag->pagf_levels[XFS_BTNUM_BNOi] = 322 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); 323 pag->pagf_levels[XFS_BTNUM_CNTi] = 324 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); 325 pag->pagf_levels[XFS_BTNUM_RMAPi] = 326 be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]); 327 pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level); 328 pag->pagf_init = 1; 329 330 return 0; 331 } 332 333 /* Repair the AGF. v5 filesystems only. */ 334 int 335 xrep_agf( 336 struct xfs_scrub *sc) 337 { 338 struct xrep_find_ag_btree fab[XREP_AGF_MAX] = { 339 [XREP_AGF_BNOBT] = { 340 .rmap_owner = XFS_RMAP_OWN_AG, 341 .buf_ops = &xfs_bnobt_buf_ops, 342 .maxlevels = sc->mp->m_alloc_maxlevels, 343 }, 344 [XREP_AGF_CNTBT] = { 345 .rmap_owner = XFS_RMAP_OWN_AG, 346 .buf_ops = &xfs_cntbt_buf_ops, 347 .maxlevels = sc->mp->m_alloc_maxlevels, 348 }, 349 [XREP_AGF_RMAPBT] = { 350 .rmap_owner = XFS_RMAP_OWN_AG, 351 .buf_ops = &xfs_rmapbt_buf_ops, 352 .maxlevels = sc->mp->m_rmap_maxlevels, 353 }, 354 [XREP_AGF_REFCOUNTBT] = { 355 .rmap_owner = XFS_RMAP_OWN_REFC, 356 .buf_ops = &xfs_refcountbt_buf_ops, 357 .maxlevels = sc->mp->m_refc_maxlevels, 358 }, 359 [XREP_AGF_END] = { 360 .buf_ops = NULL, 361 }, 362 }; 363 struct xfs_agf old_agf; 364 struct xfs_mount *mp = sc->mp; 365 struct xfs_buf *agf_bp; 366 struct xfs_buf *agfl_bp; 367 struct xfs_agf *agf; 368 int error; 369 370 /* We require the rmapbt to rebuild anything. */ 371 if (!xfs_has_rmapbt(mp)) 372 return -EOPNOTSUPP; 373 374 /* 375 * Make sure we have the AGF buffer, as scrub might have decided it 376 * was corrupt after xfs_alloc_read_agf failed with -EFSCORRUPTED. 377 */ 378 error = xfs_trans_read_buf(mp, sc->tp, mp->m_ddev_targp, 379 XFS_AG_DADDR(mp, sc->sa.pag->pag_agno, 380 XFS_AGF_DADDR(mp)), 381 XFS_FSS_TO_BB(mp, 1), 0, &agf_bp, NULL); 382 if (error) 383 return error; 384 agf_bp->b_ops = &xfs_agf_buf_ops; 385 agf = agf_bp->b_addr; 386 387 /* 388 * Load the AGFL so that we can screen out OWN_AG blocks that are on 389 * the AGFL now; these blocks might have once been part of the 390 * bno/cnt/rmap btrees but are not now. This is a chicken and egg 391 * problem: the AGF is corrupt, so we have to trust the AGFL contents 392 * because we can't do any serious cross-referencing with any of the 393 * btrees rooted in the AGF. If the AGFL contents are obviously bad 394 * then we'll bail out. 395 */ 396 error = xfs_alloc_read_agfl(mp, sc->tp, sc->sa.pag->pag_agno, &agfl_bp); 397 if (error) 398 return error; 399 400 /* 401 * Spot-check the AGFL blocks; if they're obviously corrupt then 402 * there's nothing we can do but bail out. 403 */ 404 error = xfs_agfl_walk(sc->mp, agf_bp->b_addr, agfl_bp, 405 xrep_agf_check_agfl_block, sc); 406 if (error) 407 return error; 408 409 /* 410 * Find the AGF btree roots. This is also a chicken-and-egg situation; 411 * see the function for more details. 412 */ 413 error = xrep_agf_find_btrees(sc, agf_bp, fab, agfl_bp); 414 if (error) 415 return error; 416 417 /* Start rewriting the header and implant the btrees we found. */ 418 xrep_agf_init_header(sc, agf_bp, &old_agf); 419 xrep_agf_set_roots(sc, agf, fab); 420 error = xrep_agf_calc_from_btrees(sc, agf_bp); 421 if (error) 422 goto out_revert; 423 424 /* Commit the changes and reinitialize incore state. */ 425 return xrep_agf_commit_new(sc, agf_bp); 426 427 out_revert: 428 /* Mark the incore AGF state stale and revert the AGF. */ 429 sc->sa.pag->pagf_init = 0; 430 memcpy(agf, &old_agf, sizeof(old_agf)); 431 return error; 432 } 433 434 /* AGFL */ 435 436 struct xrep_agfl { 437 /* Bitmap of other OWN_AG metadata blocks. */ 438 struct xbitmap agmetablocks; 439 440 /* Bitmap of free space. */ 441 struct xbitmap *freesp; 442 443 struct xfs_scrub *sc; 444 }; 445 446 /* Record all OWN_AG (free space btree) information from the rmap data. */ 447 STATIC int 448 xrep_agfl_walk_rmap( 449 struct xfs_btree_cur *cur, 450 const struct xfs_rmap_irec *rec, 451 void *priv) 452 { 453 struct xrep_agfl *ra = priv; 454 xfs_fsblock_t fsb; 455 int error = 0; 456 457 if (xchk_should_terminate(ra->sc, &error)) 458 return error; 459 460 /* Record all the OWN_AG blocks. */ 461 if (rec->rm_owner == XFS_RMAP_OWN_AG) { 462 fsb = XFS_AGB_TO_FSB(cur->bc_mp, cur->bc_ag.pag->pag_agno, 463 rec->rm_startblock); 464 error = xbitmap_set(ra->freesp, fsb, rec->rm_blockcount); 465 if (error) 466 return error; 467 } 468 469 return xbitmap_set_btcur_path(&ra->agmetablocks, cur); 470 } 471 472 /* 473 * Map out all the non-AGFL OWN_AG space in this AG so that we can deduce 474 * which blocks belong to the AGFL. 475 * 476 * Compute the set of old AGFL blocks by subtracting from the list of OWN_AG 477 * blocks the list of blocks owned by all other OWN_AG metadata (bnobt, cntbt, 478 * rmapbt). These are the old AGFL blocks, so return that list and the number 479 * of blocks we're actually going to put back on the AGFL. 480 */ 481 STATIC int 482 xrep_agfl_collect_blocks( 483 struct xfs_scrub *sc, 484 struct xfs_buf *agf_bp, 485 struct xbitmap *agfl_extents, 486 xfs_agblock_t *flcount) 487 { 488 struct xrep_agfl ra; 489 struct xfs_mount *mp = sc->mp; 490 struct xfs_btree_cur *cur; 491 int error; 492 493 ra.sc = sc; 494 ra.freesp = agfl_extents; 495 xbitmap_init(&ra.agmetablocks); 496 497 /* Find all space used by the free space btrees & rmapbt. */ 498 cur = xfs_rmapbt_init_cursor(mp, sc->tp, agf_bp, sc->sa.pag); 499 error = xfs_rmap_query_all(cur, xrep_agfl_walk_rmap, &ra); 500 if (error) 501 goto err; 502 xfs_btree_del_cursor(cur, error); 503 504 /* Find all blocks currently being used by the bnobt. */ 505 cur = xfs_allocbt_init_cursor(mp, sc->tp, agf_bp, 506 sc->sa.pag, XFS_BTNUM_BNO); 507 error = xbitmap_set_btblocks(&ra.agmetablocks, cur); 508 if (error) 509 goto err; 510 xfs_btree_del_cursor(cur, error); 511 512 /* Find all blocks currently being used by the cntbt. */ 513 cur = xfs_allocbt_init_cursor(mp, sc->tp, agf_bp, 514 sc->sa.pag, XFS_BTNUM_CNT); 515 error = xbitmap_set_btblocks(&ra.agmetablocks, cur); 516 if (error) 517 goto err; 518 519 xfs_btree_del_cursor(cur, error); 520 521 /* 522 * Drop the freesp meta blocks that are in use by btrees. 523 * The remaining blocks /should/ be AGFL blocks. 524 */ 525 error = xbitmap_disunion(agfl_extents, &ra.agmetablocks); 526 xbitmap_destroy(&ra.agmetablocks); 527 if (error) 528 return error; 529 530 /* 531 * Calculate the new AGFL size. If we found more blocks than fit in 532 * the AGFL we'll free them later. 533 */ 534 *flcount = min_t(uint64_t, xbitmap_hweight(agfl_extents), 535 xfs_agfl_size(mp)); 536 return 0; 537 538 err: 539 xbitmap_destroy(&ra.agmetablocks); 540 xfs_btree_del_cursor(cur, error); 541 return error; 542 } 543 544 /* Update the AGF and reset the in-core state. */ 545 STATIC void 546 xrep_agfl_update_agf( 547 struct xfs_scrub *sc, 548 struct xfs_buf *agf_bp, 549 xfs_agblock_t flcount) 550 { 551 struct xfs_agf *agf = agf_bp->b_addr; 552 553 ASSERT(flcount <= xfs_agfl_size(sc->mp)); 554 555 /* Trigger fdblocks recalculation */ 556 xfs_force_summary_recalc(sc->mp); 557 558 /* Update the AGF counters. */ 559 if (sc->sa.pag->pagf_init) 560 sc->sa.pag->pagf_flcount = flcount; 561 agf->agf_flfirst = cpu_to_be32(0); 562 agf->agf_flcount = cpu_to_be32(flcount); 563 agf->agf_fllast = cpu_to_be32(flcount - 1); 564 565 xfs_alloc_log_agf(sc->tp, agf_bp, 566 XFS_AGF_FLFIRST | XFS_AGF_FLLAST | XFS_AGF_FLCOUNT); 567 } 568 569 /* Write out a totally new AGFL. */ 570 STATIC void 571 xrep_agfl_init_header( 572 struct xfs_scrub *sc, 573 struct xfs_buf *agfl_bp, 574 struct xbitmap *agfl_extents, 575 xfs_agblock_t flcount) 576 { 577 struct xfs_mount *mp = sc->mp; 578 __be32 *agfl_bno; 579 struct xbitmap_range *br; 580 struct xbitmap_range *n; 581 struct xfs_agfl *agfl; 582 xfs_agblock_t agbno; 583 unsigned int fl_off; 584 585 ASSERT(flcount <= xfs_agfl_size(mp)); 586 587 /* 588 * Start rewriting the header by setting the bno[] array to 589 * NULLAGBLOCK, then setting AGFL header fields. 590 */ 591 agfl = XFS_BUF_TO_AGFL(agfl_bp); 592 memset(agfl, 0xFF, BBTOB(agfl_bp->b_length)); 593 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC); 594 agfl->agfl_seqno = cpu_to_be32(sc->sa.pag->pag_agno); 595 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid); 596 597 /* 598 * Fill the AGFL with the remaining blocks. If agfl_extents has more 599 * blocks than fit in the AGFL, they will be freed in a subsequent 600 * step. 601 */ 602 fl_off = 0; 603 agfl_bno = xfs_buf_to_agfl_bno(agfl_bp); 604 for_each_xbitmap_extent(br, n, agfl_extents) { 605 agbno = XFS_FSB_TO_AGBNO(mp, br->start); 606 607 trace_xrep_agfl_insert(mp, sc->sa.pag->pag_agno, agbno, 608 br->len); 609 610 while (br->len > 0 && fl_off < flcount) { 611 agfl_bno[fl_off] = cpu_to_be32(agbno); 612 fl_off++; 613 agbno++; 614 615 /* 616 * We've now used br->start by putting it in the AGFL, 617 * so bump br so that we don't reap the block later. 618 */ 619 br->start++; 620 br->len--; 621 } 622 623 if (br->len) 624 break; 625 list_del(&br->list); 626 kmem_free(br); 627 } 628 629 /* Write new AGFL to disk. */ 630 xfs_trans_buf_set_type(sc->tp, agfl_bp, XFS_BLFT_AGFL_BUF); 631 xfs_trans_log_buf(sc->tp, agfl_bp, 0, BBTOB(agfl_bp->b_length) - 1); 632 } 633 634 /* Repair the AGFL. */ 635 int 636 xrep_agfl( 637 struct xfs_scrub *sc) 638 { 639 struct xbitmap agfl_extents; 640 struct xfs_mount *mp = sc->mp; 641 struct xfs_buf *agf_bp; 642 struct xfs_buf *agfl_bp; 643 xfs_agblock_t flcount; 644 int error; 645 646 /* We require the rmapbt to rebuild anything. */ 647 if (!xfs_has_rmapbt(mp)) 648 return -EOPNOTSUPP; 649 650 xbitmap_init(&agfl_extents); 651 652 /* 653 * Read the AGF so that we can query the rmapbt. We hope that there's 654 * nothing wrong with the AGF, but all the AG header repair functions 655 * have this chicken-and-egg problem. 656 */ 657 error = xfs_alloc_read_agf(mp, sc->tp, sc->sa.pag->pag_agno, 0, 658 &agf_bp); 659 if (error) 660 return error; 661 662 /* 663 * Make sure we have the AGFL buffer, as scrub might have decided it 664 * was corrupt after xfs_alloc_read_agfl failed with -EFSCORRUPTED. 665 */ 666 error = xfs_trans_read_buf(mp, sc->tp, mp->m_ddev_targp, 667 XFS_AG_DADDR(mp, sc->sa.pag->pag_agno, 668 XFS_AGFL_DADDR(mp)), 669 XFS_FSS_TO_BB(mp, 1), 0, &agfl_bp, NULL); 670 if (error) 671 return error; 672 agfl_bp->b_ops = &xfs_agfl_buf_ops; 673 674 /* Gather all the extents we're going to put on the new AGFL. */ 675 error = xrep_agfl_collect_blocks(sc, agf_bp, &agfl_extents, &flcount); 676 if (error) 677 goto err; 678 679 /* 680 * Update AGF and AGFL. We reset the global free block counter when 681 * we adjust the AGF flcount (which can fail) so avoid updating any 682 * buffers until we know that part works. 683 */ 684 xrep_agfl_update_agf(sc, agf_bp, flcount); 685 xrep_agfl_init_header(sc, agfl_bp, &agfl_extents, flcount); 686 687 /* 688 * Ok, the AGFL should be ready to go now. Roll the transaction to 689 * make the new AGFL permanent before we start using it to return 690 * freespace overflow to the freespace btrees. 691 */ 692 sc->sa.agf_bp = agf_bp; 693 sc->sa.agfl_bp = agfl_bp; 694 error = xrep_roll_ag_trans(sc); 695 if (error) 696 goto err; 697 698 /* Dump any AGFL overflow. */ 699 error = xrep_reap_extents(sc, &agfl_extents, &XFS_RMAP_OINFO_AG, 700 XFS_AG_RESV_AGFL); 701 err: 702 xbitmap_destroy(&agfl_extents); 703 return error; 704 } 705 706 /* AGI */ 707 708 /* 709 * Offset within the xrep_find_ag_btree array for each btree type. Avoid the 710 * XFS_BTNUM_ names here to avoid creating a sparse array. 711 */ 712 enum { 713 XREP_AGI_INOBT = 0, 714 XREP_AGI_FINOBT, 715 XREP_AGI_END, 716 XREP_AGI_MAX 717 }; 718 719 /* 720 * Given the inode btree roots described by *fab, find the roots, check them 721 * for sanity, and pass the root data back out via *fab. 722 */ 723 STATIC int 724 xrep_agi_find_btrees( 725 struct xfs_scrub *sc, 726 struct xrep_find_ag_btree *fab) 727 { 728 struct xfs_buf *agf_bp; 729 struct xfs_mount *mp = sc->mp; 730 int error; 731 732 /* Read the AGF. */ 733 error = xfs_alloc_read_agf(mp, sc->tp, sc->sa.pag->pag_agno, 0, 734 &agf_bp); 735 if (error) 736 return error; 737 738 /* Find the btree roots. */ 739 error = xrep_find_ag_btree_roots(sc, agf_bp, fab, NULL); 740 if (error) 741 return error; 742 743 /* We must find the inobt root. */ 744 if (!xrep_check_btree_root(sc, &fab[XREP_AGI_INOBT])) 745 return -EFSCORRUPTED; 746 747 /* We must find the finobt root if that feature is enabled. */ 748 if (xfs_has_finobt(mp) && 749 !xrep_check_btree_root(sc, &fab[XREP_AGI_FINOBT])) 750 return -EFSCORRUPTED; 751 752 return 0; 753 } 754 755 /* 756 * Reinitialize the AGI header, making an in-core copy of the old contents so 757 * that we know which in-core state needs to be reinitialized. 758 */ 759 STATIC void 760 xrep_agi_init_header( 761 struct xfs_scrub *sc, 762 struct xfs_buf *agi_bp, 763 struct xfs_agi *old_agi) 764 { 765 struct xfs_agi *agi = agi_bp->b_addr; 766 struct xfs_mount *mp = sc->mp; 767 768 memcpy(old_agi, agi, sizeof(*old_agi)); 769 memset(agi, 0, BBTOB(agi_bp->b_length)); 770 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC); 771 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION); 772 agi->agi_seqno = cpu_to_be32(sc->sa.pag->pag_agno); 773 agi->agi_length = cpu_to_be32(xfs_ag_block_count(mp, 774 sc->sa.pag->pag_agno)); 775 agi->agi_newino = cpu_to_be32(NULLAGINO); 776 agi->agi_dirino = cpu_to_be32(NULLAGINO); 777 if (xfs_has_crc(mp)) 778 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid); 779 780 /* We don't know how to fix the unlinked list yet. */ 781 memcpy(&agi->agi_unlinked, &old_agi->agi_unlinked, 782 sizeof(agi->agi_unlinked)); 783 784 /* Mark the incore AGF data stale until we're done fixing things. */ 785 ASSERT(sc->sa.pag->pagi_init); 786 sc->sa.pag->pagi_init = 0; 787 } 788 789 /* Set btree root information in an AGI. */ 790 STATIC void 791 xrep_agi_set_roots( 792 struct xfs_scrub *sc, 793 struct xfs_agi *agi, 794 struct xrep_find_ag_btree *fab) 795 { 796 agi->agi_root = cpu_to_be32(fab[XREP_AGI_INOBT].root); 797 agi->agi_level = cpu_to_be32(fab[XREP_AGI_INOBT].height); 798 799 if (xfs_has_finobt(sc->mp)) { 800 agi->agi_free_root = cpu_to_be32(fab[XREP_AGI_FINOBT].root); 801 agi->agi_free_level = cpu_to_be32(fab[XREP_AGI_FINOBT].height); 802 } 803 } 804 805 /* Update the AGI counters. */ 806 STATIC int 807 xrep_agi_calc_from_btrees( 808 struct xfs_scrub *sc, 809 struct xfs_buf *agi_bp) 810 { 811 struct xfs_btree_cur *cur; 812 struct xfs_agi *agi = agi_bp->b_addr; 813 struct xfs_mount *mp = sc->mp; 814 xfs_agino_t count; 815 xfs_agino_t freecount; 816 int error; 817 818 cur = xfs_inobt_init_cursor(mp, sc->tp, agi_bp, 819 sc->sa.pag, XFS_BTNUM_INO); 820 error = xfs_ialloc_count_inodes(cur, &count, &freecount); 821 if (error) 822 goto err; 823 if (xfs_has_inobtcounts(mp)) { 824 xfs_agblock_t blocks; 825 826 error = xfs_btree_count_blocks(cur, &blocks); 827 if (error) 828 goto err; 829 agi->agi_iblocks = cpu_to_be32(blocks); 830 } 831 xfs_btree_del_cursor(cur, error); 832 833 agi->agi_count = cpu_to_be32(count); 834 agi->agi_freecount = cpu_to_be32(freecount); 835 836 if (xfs_has_finobt(mp) && xfs_has_inobtcounts(mp)) { 837 xfs_agblock_t blocks; 838 839 cur = xfs_inobt_init_cursor(mp, sc->tp, agi_bp, 840 sc->sa.pag, XFS_BTNUM_FINO); 841 error = xfs_btree_count_blocks(cur, &blocks); 842 if (error) 843 goto err; 844 xfs_btree_del_cursor(cur, error); 845 agi->agi_fblocks = cpu_to_be32(blocks); 846 } 847 848 return 0; 849 err: 850 xfs_btree_del_cursor(cur, error); 851 return error; 852 } 853 854 /* Trigger reinitialization of the in-core data. */ 855 STATIC int 856 xrep_agi_commit_new( 857 struct xfs_scrub *sc, 858 struct xfs_buf *agi_bp) 859 { 860 struct xfs_perag *pag; 861 struct xfs_agi *agi = agi_bp->b_addr; 862 863 /* Trigger inode count recalculation */ 864 xfs_force_summary_recalc(sc->mp); 865 866 /* Write this to disk. */ 867 xfs_trans_buf_set_type(sc->tp, agi_bp, XFS_BLFT_AGI_BUF); 868 xfs_trans_log_buf(sc->tp, agi_bp, 0, BBTOB(agi_bp->b_length) - 1); 869 870 /* Now reinitialize the in-core counters if necessary. */ 871 pag = sc->sa.pag; 872 pag->pagi_count = be32_to_cpu(agi->agi_count); 873 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); 874 pag->pagi_init = 1; 875 876 return 0; 877 } 878 879 /* Repair the AGI. */ 880 int 881 xrep_agi( 882 struct xfs_scrub *sc) 883 { 884 struct xrep_find_ag_btree fab[XREP_AGI_MAX] = { 885 [XREP_AGI_INOBT] = { 886 .rmap_owner = XFS_RMAP_OWN_INOBT, 887 .buf_ops = &xfs_inobt_buf_ops, 888 .maxlevels = M_IGEO(sc->mp)->inobt_maxlevels, 889 }, 890 [XREP_AGI_FINOBT] = { 891 .rmap_owner = XFS_RMAP_OWN_INOBT, 892 .buf_ops = &xfs_finobt_buf_ops, 893 .maxlevels = M_IGEO(sc->mp)->inobt_maxlevels, 894 }, 895 [XREP_AGI_END] = { 896 .buf_ops = NULL 897 }, 898 }; 899 struct xfs_agi old_agi; 900 struct xfs_mount *mp = sc->mp; 901 struct xfs_buf *agi_bp; 902 struct xfs_agi *agi; 903 int error; 904 905 /* We require the rmapbt to rebuild anything. */ 906 if (!xfs_has_rmapbt(mp)) 907 return -EOPNOTSUPP; 908 909 /* 910 * Make sure we have the AGI buffer, as scrub might have decided it 911 * was corrupt after xfs_ialloc_read_agi failed with -EFSCORRUPTED. 912 */ 913 error = xfs_trans_read_buf(mp, sc->tp, mp->m_ddev_targp, 914 XFS_AG_DADDR(mp, sc->sa.pag->pag_agno, 915 XFS_AGI_DADDR(mp)), 916 XFS_FSS_TO_BB(mp, 1), 0, &agi_bp, NULL); 917 if (error) 918 return error; 919 agi_bp->b_ops = &xfs_agi_buf_ops; 920 agi = agi_bp->b_addr; 921 922 /* Find the AGI btree roots. */ 923 error = xrep_agi_find_btrees(sc, fab); 924 if (error) 925 return error; 926 927 /* Start rewriting the header and implant the btrees we found. */ 928 xrep_agi_init_header(sc, agi_bp, &old_agi); 929 xrep_agi_set_roots(sc, agi, fab); 930 error = xrep_agi_calc_from_btrees(sc, agi_bp); 931 if (error) 932 goto out_revert; 933 934 /* Reinitialize in-core state. */ 935 return xrep_agi_commit_new(sc, agi_bp); 936 937 out_revert: 938 /* Mark the incore AGI state stale and revert the AGI. */ 939 sc->sa.pag->pagi_init = 0; 940 memcpy(agi, &old_agi, sizeof(old_agi)); 941 return error; 942 } 943