fs/xfs/scrub/agheader_repair.c

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