fs/xfs/scrub/dir.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_trans_resv.h"
  11 #include "xfs_mount.h"
  12 #include "xfs_log_format.h"
  13 #include "xfs_trans.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_icache.h"
  16 #include "xfs_dir2.h"
  17 #include "xfs_dir2_priv.h"
  18 #include "scrub/scrub.h"
  19 #include "scrub/common.h"
  20 #include "scrub/dabtree.h"
  21
  22 /* Set us up to scrub directories. */
  23 int
  24 xchk_setup_directory(
  25         struct xfs_scrub        *sc,
  26         struct xfs_inode        *ip)
  27 {
  28         return xchk_setup_inode_contents(sc, ip, 0);
  29 }
  30
  31 /* Directories */
  32
  33 /* Scrub a directory entry. */
  34
  35 struct xchk_dir_ctx {
  36         /* VFS fill-directory iterator */
  37         struct dir_context      dir_iter;
  38
  39         struct xfs_scrub        *sc;
  40 };
  41
  42 /* Check that an inode's mode matches a given DT_ type. */
  43 STATIC int
  44 xchk_dir_check_ftype(
  45         struct xchk_dir_ctx     *sdc,
  46         xfs_fileoff_t           offset,
  47         xfs_ino_t               inum,
  48         int                     dtype)
  49 {
  50         struct xfs_mount        *mp = sdc->sc->mp;
  51         struct xfs_inode        *ip;
  52         int                     ino_dtype;
  53         int                     error = 0;
  54
  55         if (!xfs_sb_version_hasftype(&mp->m_sb)) {
  56                 if (dtype != DT_UNKNOWN && dtype != DT_DIR)
  57                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
  58                                         offset);
  59                 goto out;
  60         }
  61
  62         /*
  63          * Grab the inode pointed to by the dirent.  We release the
  64          * inode before we cancel the scrub transaction.  Since we're
  65          * don't know a priori that releasing the inode won't trigger
  66          * eofblocks cleanup (which allocates what would be a nested
  67          * transaction), we can't use DONTCACHE here because DONTCACHE
  68          * inodes can trigger immediate inactive cleanup of the inode.
  69          */
  70         error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
  71         if (!xchk_fblock_xref_process_error(sdc->sc, XFS_DATA_FORK, offset,
  72                         &error))
  73                 goto out;
  74
  75         /* Convert mode to the DT_* values that dir_emit uses. */
  76         ino_dtype = xfs_dir3_get_dtype(mp,
  77                         xfs_mode_to_ftype(VFS_I(ip)->i_mode));
  78         if (ino_dtype != dtype)
  79                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
  80         xfs_irele(ip);
  81 out:
  82         return error;
  83 }
  84
  85 /*
  86  * Scrub a single directory entry.
  87  *
  88  * We use the VFS directory iterator (i.e. readdir) to call this
  89  * function for every directory entry in a directory.  Once we're here,
  90  * we check the inode number to make sure it's sane, then we check that
  91  * we can look up this filename.  Finally, we check the ftype.
  92  */
  93 STATIC int
  94 xchk_dir_actor(
  95         struct dir_context      *dir_iter,
  96         const char              *name,
  97         int                     namelen,
  98         loff_t                  pos,
  99         u64                     ino,
 100         unsigned                type)
 101 {
 102         struct xfs_mount        *mp;
 103         struct xfs_inode        *ip;
 104         struct xchk_dir_ctx     *sdc;
 105         struct xfs_name         xname;
 106         xfs_ino_t               lookup_ino;
 107         xfs_dablk_t             offset;
 108         int                     error = 0;
 109
 110         sdc = container_of(dir_iter, struct xchk_dir_ctx, dir_iter);
 111         ip = sdc->sc->ip;
 112         mp = ip->i_mount;
 113         offset = xfs_dir2_db_to_da(mp->m_dir_geo,
 114                         xfs_dir2_dataptr_to_db(mp->m_dir_geo, pos));
 115
 116         if (xchk_should_terminate(sdc->sc, &error))
 117                 return error;
 118
 119         /* Does this inode number make sense? */
 120         if (!xfs_verify_dir_ino(mp, ino)) {
 121                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 122                 goto out;
 123         }
 124
 125         /* Does this name make sense? */
 126         if (!xfs_dir2_namecheck(name, namelen)) {
 127                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 128                 goto out;
 129         }
 130
 131         if (!strncmp(".", name, namelen)) {
 132                 /* If this is "." then check that the inum matches the dir. */
 133                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 134                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 135                                         offset);
 136                 if (ino != ip->i_ino)
 137                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 138                                         offset);
 139         } else if (!strncmp("..", name, namelen)) {
 140                 /*
 141                  * If this is ".." in the root inode, check that the inum
 142                  * matches this dir.
 143                  */
 144                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 145                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 146                                         offset);
 147                 if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
 148                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 149                                         offset);
 150         }
 151
 152         /* Verify that we can look up this name by hash. */
 153         xname.name = name;
 154         xname.len = namelen;
 155         xname.type = XFS_DIR3_FT_UNKNOWN;
 156
 157         error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
 158         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 159                         &error))
 160                 goto out;
 161         if (lookup_ino != ino) {
 162                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 163                 goto out;
 164         }
 165
 166         /* Verify the file type.  This function absorbs error codes. */
 167         error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 168         if (error)
 169                 goto out;
 170 out:
 171         /*
 172          * A negative error code returned here is supposed to cause the
 173          * dir_emit caller (xfs_readdir) to abort the directory iteration
 174          * and return zero to xchk_directory.
 175          */
 176         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 177                 return -EFSCORRUPTED;
 178         return error;
 179 }
 180
 181 /* Scrub a directory btree record. */
 182 STATIC int
 183 xchk_dir_rec(
 184         struct xchk_da_btree            *ds,
 185         int                             level)
 186 {
 187         struct xfs_da_state_blk         *blk = &ds->state->path.blk[level];
 188         struct xfs_mount                *mp = ds->state->mp;
 189         struct xfs_inode                *dp = ds->dargs.dp;
 190         struct xfs_da_geometry          *geo = mp->m_dir_geo;
 191         struct xfs_dir2_data_entry      *dent;
 192         struct xfs_buf                  *bp;
 193         struct xfs_dir2_leaf_entry      *ent;
 194         unsigned int                    end;
 195         unsigned int                    iter_off;
 196         xfs_ino_t                       ino;
 197         xfs_dablk_t                     rec_bno;
 198         xfs_dir2_db_t                   db;
 199         xfs_dir2_data_aoff_t            off;
 200         xfs_dir2_dataptr_t              ptr;
 201         xfs_dahash_t                    calc_hash;
 202         xfs_dahash_t                    hash;
 203         struct xfs_dir3_icleaf_hdr      hdr;
 204         unsigned int                    tag;
 205         int                             error;
 206
 207         ASSERT(blk->magic == XFS_DIR2_LEAF1_MAGIC ||
 208                blk->magic == XFS_DIR2_LEAFN_MAGIC);
 209
 210         xfs_dir2_leaf_hdr_from_disk(mp, &hdr, blk->bp->b_addr);
 211         ent = hdr.ents + blk->index;
 212
 213         /* Check the hash of the entry. */
 214         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 215         if (error)
 216                 goto out;
 217
 218         /* Valid hash pointer? */
 219         ptr = be32_to_cpu(ent->address);
 220         if (ptr == 0)
 221                 return 0;
 222
 223         /* Find the directory entry's location. */
 224         db = xfs_dir2_dataptr_to_db(geo, ptr);
 225         off = xfs_dir2_dataptr_to_off(geo, ptr);
 226         rec_bno = xfs_dir2_db_to_da(geo, db);
 227
 228         if (rec_bno >= geo->leafblk) {
 229                 xchk_da_set_corrupt(ds, level);
 230                 goto out;
 231         }
 232         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
 233         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 234                         &error))
 235                 goto out;
 236         if (!bp) {
 237                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 238                 goto out;
 239         }
 240         xchk_buffer_recheck(ds->sc, bp);
 241
 242         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 243                 goto out_relse;
 244
 245         dent = bp->b_addr + off;
 246
 247         /* Make sure we got a real directory entry. */
 248         iter_off = geo->data_entry_offset;
 249         end = xfs_dir3_data_end_offset(geo, bp->b_addr);
 250         if (!end) {
 251                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 252                 goto out_relse;
 253         }
 254         for (;;) {
 255                 struct xfs_dir2_data_entry      *dep = bp->b_addr + iter_off;
 256                 struct xfs_dir2_data_unused     *dup = bp->b_addr + iter_off;
 257
 258                 if (iter_off >= end) {
 259                         xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 260                         goto out_relse;
 261                 }
 262
 263                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 264                         iter_off += be16_to_cpu(dup->length);
 265                         continue;
 266                 }
 267                 if (dep == dent)
 268                         break;
 269                 iter_off += xfs_dir2_data_entsize(mp, dep->namelen);
 270         }
 271
 272         /* Retrieve the entry, sanity check it, and compare hashes. */
 273         ino = be64_to_cpu(dent->inumber);
 274         hash = be32_to_cpu(ent->hashval);
 275         tag = be16_to_cpup(xfs_dir2_data_entry_tag_p(mp, dent));
 276         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 277                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 278         if (dent->namelen == 0) {
 279                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 280                 goto out_relse;
 281         }
 282         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 283         if (calc_hash != hash)
 284                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 285
 286 out_relse:
 287         xfs_trans_brelse(ds->dargs.trans, bp);
 288 out:
 289         return error;
 290 }
 291
 292 /*
 293  * Is this unused entry either in the bestfree or smaller than all of
 294  * them?  We've already checked that the bestfrees are sorted longest to
 295  * shortest, and that there aren't any bogus entries.
 296  */
 297 STATIC void
 298 xchk_directory_check_free_entry(
 299         struct xfs_scrub                *sc,
 300         xfs_dablk_t                     lblk,
 301         struct xfs_dir2_data_free       *bf,
 302         struct xfs_dir2_data_unused     *dup)
 303 {
 304         struct xfs_dir2_data_free       *dfp;
 305         unsigned int                    dup_length;
 306
 307         dup_length = be16_to_cpu(dup->length);
 308
 309         /* Unused entry is shorter than any of the bestfrees */
 310         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 311                 return;
 312
 313         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 314                 if (dup_length == be16_to_cpu(dfp->length))
 315                         return;
 316
 317         /* Unused entry should be in the bestfrees but wasn't found. */
 318         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 319 }
 320
 321 /* Check free space info in a directory data block. */
 322 STATIC int
 323 xchk_directory_data_bestfree(
 324         struct xfs_scrub                *sc,
 325         xfs_dablk_t                     lblk,
 326         bool                            is_block)
 327 {
 328         struct xfs_dir2_data_unused     *dup;
 329         struct xfs_dir2_data_free       *dfp;
 330         struct xfs_buf                  *bp;
 331         struct xfs_dir2_data_free       *bf;
 332         struct xfs_mount                *mp = sc->mp;
 333         u16                             tag;
 334         unsigned int                    nr_bestfrees = 0;
 335         unsigned int                    nr_frees = 0;
 336         unsigned int                    smallest_bestfree;
 337         int                             newlen;
 338         unsigned int                    offset;
 339         unsigned int                    end;
 340         int                             error;
 341
 342         if (is_block) {
 343                 /* dir block format */
 344                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 345                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 346                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 347         } else {
 348                 /* dir data format */
 349                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, -1, &bp);
 350         }
 351         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 352                 goto out;
 353         xchk_buffer_recheck(sc, bp);
 354
 355         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 356
 357         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 358                 goto out_buf;
 359
 360         /* Do the bestfrees correspond to actual free space? */
 361         bf = xfs_dir2_data_bestfree_p(mp, bp->b_addr);
 362         smallest_bestfree = UINT_MAX;
 363         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 364                 offset = be16_to_cpu(dfp->offset);
 365                 if (offset == 0)
 366                         continue;
 367                 if (offset >= mp->m_dir_geo->blksize) {
 368                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 369                         goto out_buf;
 370                 }
 371                 dup = bp->b_addr + offset;
 372                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 373
 374                 /* bestfree doesn't match the entry it points at? */
 375                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 376                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 377                     tag != offset) {
 378                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 379                         goto out_buf;
 380                 }
 381
 382                 /* bestfree records should be ordered largest to smallest */
 383                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 384                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 385                         goto out_buf;
 386                 }
 387
 388                 smallest_bestfree = be16_to_cpu(dfp->length);
 389                 nr_bestfrees++;
 390         }
 391
 392         /* Make sure the bestfrees are actually the best free spaces. */
 393         offset = mp->m_dir_geo->data_entry_offset;
 394         end = xfs_dir3_data_end_offset(mp->m_dir_geo, bp->b_addr);
 395
 396         /* Iterate the entries, stopping when we hit or go past the end. */
 397         while (offset < end) {
 398                 dup = bp->b_addr + offset;
 399
 400                 /* Skip real entries */
 401                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 402                         struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
 403
 404                         newlen = xfs_dir2_data_entsize(mp, dep->namelen);
 405                         if (newlen <= 0) {
 406                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 407                                                 lblk);
 408                                 goto out_buf;
 409                         }
 410                         offset += newlen;
 411                         continue;
 412                 }
 413
 414                 /* Spot check this free entry */
 415                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 416                 if (tag != offset) {
 417                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 418                         goto out_buf;
 419                 }
 420
 421                 /*
 422                  * Either this entry is a bestfree or it's smaller than
 423                  * any of the bestfrees.
 424                  */
 425                 xchk_directory_check_free_entry(sc, lblk, bf, dup);
 426                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 427                         goto out_buf;
 428
 429                 /* Move on. */
 430                 newlen = be16_to_cpu(dup->length);
 431                 if (newlen <= 0) {
 432                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 433                         goto out_buf;
 434                 }
 435                 offset += newlen;
 436                 if (offset <= end)
 437                         nr_frees++;
 438         }
 439
 440         /* We're required to fill all the space. */
 441         if (offset != end)
 442                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 443
 444         /* Did we see at least as many free slots as there are bestfrees? */
 445         if (nr_frees < nr_bestfrees)
 446                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 447 out_buf:
 448         xfs_trans_brelse(sc->tp, bp);
 449 out:
 450         return error;
 451 }
 452
 453 /*
 454  * Does the free space length in the free space index block ($len) match
 455  * the longest length in the directory data block's bestfree array?
 456  * Assume that we've already checked that the data block's bestfree
 457  * array is in order.
 458  */
 459 STATIC void
 460 xchk_directory_check_freesp(
 461         struct xfs_scrub                *sc,
 462         xfs_dablk_t                     lblk,
 463         struct xfs_buf                  *dbp,
 464         unsigned int                    len)
 465 {
 466         struct xfs_dir2_data_free       *dfp;
 467
 468         dfp = xfs_dir2_data_bestfree_p(sc->mp, dbp->b_addr);
 469
 470         if (len != be16_to_cpu(dfp->length))
 471                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 472
 473         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 474                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 475 }
 476
 477 /* Check free space info in a directory leaf1 block. */
 478 STATIC int
 479 xchk_directory_leaf1_bestfree(
 480         struct xfs_scrub                *sc,
 481         struct xfs_da_args              *args,
 482         xfs_dablk_t                     lblk)
 483 {
 484         struct xfs_dir3_icleaf_hdr      leafhdr;
 485         struct xfs_dir2_leaf_tail       *ltp;
 486         struct xfs_dir2_leaf            *leaf;
 487         struct xfs_buf                  *dbp;
 488         struct xfs_buf                  *bp;
 489         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 490         __be16                          *bestp;
 491         __u16                           best;
 492         __u32                           hash;
 493         __u32                           lasthash = 0;
 494         __u32                           bestcount;
 495         unsigned int                    stale = 0;
 496         int                             i;
 497         int                             error;
 498
 499         /* Read the free space block. */
 500         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, &bp);
 501         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 502                 goto out;
 503         xchk_buffer_recheck(sc, bp);
 504
 505         leaf = bp->b_addr;
 506         xfs_dir2_leaf_hdr_from_disk(sc->ip->i_mount, &leafhdr, leaf);
 507         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 508         bestcount = be32_to_cpu(ltp->bestcount);
 509         bestp = xfs_dir2_leaf_bests_p(ltp);
 510
 511         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 512                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 513
 514                 if (hdr3->pad != cpu_to_be32(0))
 515                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 516         }
 517
 518         /*
 519          * There should be as many bestfree slots as there are dir data
 520          * blocks that can fit under i_size.
 521          */
 522         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 523                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 524                 goto out;
 525         }
 526
 527         /* Is the leaf count even remotely sane? */
 528         if (leafhdr.count > geo->leaf_max_ents) {
 529                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 530                 goto out;
 531         }
 532
 533         /* Leaves and bests don't overlap in leaf format. */
 534         if ((char *)&leafhdr.ents[leafhdr.count] > (char *)bestp) {
 535                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 536                 goto out;
 537         }
 538
 539         /* Check hash value order, count stale entries.  */
 540         for (i = 0; i < leafhdr.count; i++) {
 541                 hash = be32_to_cpu(leafhdr.ents[i].hashval);
 542                 if (i > 0 && lasthash > hash)
 543                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 544                 lasthash = hash;
 545                 if (leafhdr.ents[i].address ==
 546                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 547                         stale++;
 548         }
 549         if (leafhdr.stale != stale)
 550                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 551         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 552                 goto out;
 553
 554         /* Check all the bestfree entries. */
 555         for (i = 0; i < bestcount; i++, bestp++) {
 556                 best = be16_to_cpu(*bestp);
 557                 if (best == NULLDATAOFF)
 558                         continue;
 559                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 560                                 i * args->geo->fsbcount, -1, &dbp);
 561                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 562                                 &error))
 563                         break;
 564                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 565                 xfs_trans_brelse(sc->tp, dbp);
 566                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 567                         goto out;
 568         }
 569 out:
 570         return error;
 571 }
 572
 573 /* Check free space info in a directory freespace block. */
 574 STATIC int
 575 xchk_directory_free_bestfree(
 576         struct xfs_scrub                *sc,
 577         struct xfs_da_args              *args,
 578         xfs_dablk_t                     lblk)
 579 {
 580         struct xfs_dir3_icfree_hdr      freehdr;
 581         struct xfs_buf                  *dbp;
 582         struct xfs_buf                  *bp;
 583         __u16                           best;
 584         unsigned int                    stale = 0;
 585         int                             i;
 586         int                             error;
 587
 588         /* Read the free space block */
 589         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 590         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 591                 goto out;
 592         xchk_buffer_recheck(sc, bp);
 593
 594         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 595                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 596
 597                 if (hdr3->pad != cpu_to_be32(0))
 598                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 599         }
 600
 601         /* Check all the entries. */
 602         xfs_dir2_free_hdr_from_disk(sc->ip->i_mount, &freehdr, bp->b_addr);
 603         for (i = 0; i < freehdr.nvalid; i++) {
 604                 best = be16_to_cpu(freehdr.bests[i]);
 605                 if (best == NULLDATAOFF) {
 606                         stale++;
 607                         continue;
 608                 }
 609                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 610                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 611                                 -1, &dbp);
 612                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 613                                 &error))
 614                         break;
 615                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 616                 xfs_trans_brelse(sc->tp, dbp);
 617         }
 618
 619         if (freehdr.nused + stale != freehdr.nvalid)
 620                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 621 out:
 622         return error;
 623 }
 624
 625 /* Check free space information in directories. */
 626 STATIC int
 627 xchk_directory_blocks(
 628         struct xfs_scrub        *sc)
 629 {
 630         struct xfs_bmbt_irec    got;
 631         struct xfs_da_args      args;
 632         struct xfs_ifork        *ifp;
 633         struct xfs_mount        *mp = sc->mp;
 634         xfs_fileoff_t           leaf_lblk;
 635         xfs_fileoff_t           free_lblk;
 636         xfs_fileoff_t           lblk;
 637         struct xfs_iext_cursor  icur;
 638         xfs_dablk_t             dabno;
 639         bool                    found;
 640         int                     is_block = 0;
 641         int                     error;
 642
 643         /* Ignore local format directories. */
 644         if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 645             sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
 646                 return 0;
 647
 648         ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 649         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 650         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 651         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 652
 653         /* Is this a block dir? */
 654         args.dp = sc->ip;
 655         args.geo = mp->m_dir_geo;
 656         args.trans = sc->tp;
 657         error = xfs_dir2_isblock(&args, &is_block);
 658         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 659                 goto out;
 660
 661         /* Iterate all the data extents in the directory... */
 662         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 663         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 664                 /* Block directories only have a single block at offset 0. */
 665                 if (is_block &&
 666                     (got.br_startoff > 0 ||
 667                      got.br_blockcount != args.geo->fsbcount)) {
 668                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 669                                         got.br_startoff);
 670                         break;
 671                 }
 672
 673                 /* No more data blocks... */
 674                 if (got.br_startoff >= leaf_lblk)
 675                         break;
 676
 677                 /*
 678                  * Check each data block's bestfree data.
 679                  *
 680                  * Iterate all the fsbcount-aligned block offsets in
 681                  * this directory.  The directory block reading code is
 682                  * smart enough to do its own bmap lookups to handle
 683                  * discontiguous directory blocks.  When we're done
 684                  * with the extent record, re-query the bmap at the
 685                  * next fsbcount-aligned offset to avoid redundant
 686                  * block checks.
 687                  */
 688                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 689                                 args.geo->fsbcount);
 690                      lblk < got.br_startoff + got.br_blockcount;
 691                      lblk += args.geo->fsbcount) {
 692                         error = xchk_directory_data_bestfree(sc, lblk,
 693                                         is_block);
 694                         if (error)
 695                                 goto out;
 696                 }
 697                 dabno = got.br_startoff + got.br_blockcount;
 698                 lblk = roundup(dabno, args.geo->fsbcount);
 699                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 700         }
 701
 702         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 703                 goto out;
 704
 705         /* Look for a leaf1 block, which has free info. */
 706         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 707             got.br_startoff == leaf_lblk &&
 708             got.br_blockcount == args.geo->fsbcount &&
 709             !xfs_iext_next_extent(ifp, &icur, &got)) {
 710                 if (is_block) {
 711                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 712                         goto out;
 713                 }
 714                 error = xchk_directory_leaf1_bestfree(sc, &args,
 715                                 leaf_lblk);
 716                 if (error)
 717                         goto out;
 718         }
 719
 720         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 721                 goto out;
 722
 723         /* Scan for free blocks */
 724         lblk = free_lblk;
 725         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 726         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 727                 /*
 728                  * Dirs can't have blocks mapped above 2^32.
 729                  * Single-block dirs shouldn't even be here.
 730                  */
 731                 lblk = got.br_startoff;
 732                 if (lblk & ~0xFFFFFFFFULL) {
 733                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 734                         goto out;
 735                 }
 736                 if (is_block) {
 737                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 738                         goto out;
 739                 }
 740
 741                 /*
 742                  * Check each dir free block's bestfree data.
 743                  *
 744                  * Iterate all the fsbcount-aligned block offsets in
 745                  * this directory.  The directory block reading code is
 746                  * smart enough to do its own bmap lookups to handle
 747                  * discontiguous directory blocks.  When we're done
 748                  * with the extent record, re-query the bmap at the
 749                  * next fsbcount-aligned offset to avoid redundant
 750                  * block checks.
 751                  */
 752                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 753                                 args.geo->fsbcount);
 754                      lblk < got.br_startoff + got.br_blockcount;
 755                      lblk += args.geo->fsbcount) {
 756                         error = xchk_directory_free_bestfree(sc, &args,
 757                                         lblk);
 758                         if (error)
 759                                 goto out;
 760                 }
 761                 dabno = got.br_startoff + got.br_blockcount;
 762                 lblk = roundup(dabno, args.geo->fsbcount);
 763                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 764         }
 765 out:
 766         return error;
 767 }
 768
 769 /* Scrub a whole directory. */
 770 int
 771 xchk_directory(
 772         struct xfs_scrub        *sc)
 773 {
 774         struct xchk_dir_ctx     sdc = {
 775                 .dir_iter.actor = xchk_dir_actor,
 776                 .dir_iter.pos = 0,
 777                 .sc = sc,
 778         };
 779         size_t                  bufsize;
 780         loff_t                  oldpos;
 781         int                     error = 0;
 782
 783         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 784                 return -ENOENT;
 785
 786         /* Plausible size? */
 787         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 788                 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
 789                 goto out;
 790         }
 791
 792         /* Check directory tree structure */
 793         error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
 794         if (error)
 795                 return error;
 796
 797         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 798                 return error;
 799
 800         /* Check the freespace. */
 801         error = xchk_directory_blocks(sc);
 802         if (error)
 803                 return error;
 804
 805         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 806                 return error;
 807
 808         /*
 809          * Check that every dirent we see can also be looked up by hash.
 810          * Userspace usually asks for a 32k buffer, so we will too.
 811          */
 812         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 813                         sc->ip->i_d.di_size);
 814
 815         /*
 816          * Look up every name in this directory by hash.
 817          *
 818          * Use the xfs_readdir function to call xchk_dir_actor on
 819          * every directory entry in this directory.  In _actor, we check
 820          * the name, inode number, and ftype (if applicable) of the
 821          * entry.  xfs_readdir uses the VFS filldir functions to provide
 822          * iteration context.
 823          *
 824          * The VFS grabs a read or write lock via i_rwsem before it reads
 825          * or writes to a directory.  If we've gotten this far we've
 826          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 827          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 828          * to drop the ILOCK here in order to reuse the _readdir and
 829          * _dir_lookup routines, which do their own ILOCK locking.
 830          */
 831         oldpos = 0;
 832         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 833         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 834         while (true) {
 835                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 836                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
 837                                 &error))
 838                         goto out;
 839                 if (oldpos == sdc.dir_iter.pos)
 840                         break;
 841                 oldpos = sdc.dir_iter.pos;
 842         }
 843
 844 out:
 845         return error;
 846 }