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 <[email protected]>
   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         /* ENOENT means the hash lookup failed and the dir is corrupt */
 159         if (error == -ENOENT)
 160                 error = -EFSCORRUPTED;
 161         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 162                         &error))
 163                 goto out;
 164         if (lookup_ino != ino) {
 165                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 166                 goto out;
 167         }
 168
 169         /* Verify the file type.  This function absorbs error codes. */
 170         error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 171         if (error)
 172                 goto out;
 173 out:
 174         /*
 175          * A negative error code returned here is supposed to cause the
 176          * dir_emit caller (xfs_readdir) to abort the directory iteration
 177          * and return zero to xchk_directory.
 178          */
 179         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 180                 return -EFSCORRUPTED;
 181         return error;
 182 }
 183
 184 /* Scrub a directory btree record. */
 185 STATIC int
 186 xchk_dir_rec(
 187         struct xchk_da_btree            *ds,
 188         int                             level)
 189 {
 190         struct xfs_da_state_blk         *blk = &ds->state->path.blk[level];
 191         struct xfs_mount                *mp = ds->state->mp;
 192         struct xfs_inode                *dp = ds->dargs.dp;
 193         struct xfs_da_geometry          *geo = mp->m_dir_geo;
 194         struct xfs_dir2_data_entry      *dent;
 195         struct xfs_buf                  *bp;
 196         struct xfs_dir2_leaf_entry      *ent;
 197         unsigned int                    end;
 198         unsigned int                    iter_off;
 199         xfs_ino_t                       ino;
 200         xfs_dablk_t                     rec_bno;
 201         xfs_dir2_db_t                   db;
 202         xfs_dir2_data_aoff_t            off;
 203         xfs_dir2_dataptr_t              ptr;
 204         xfs_dahash_t                    calc_hash;
 205         xfs_dahash_t                    hash;
 206         struct xfs_dir3_icleaf_hdr      hdr;
 207         unsigned int                    tag;
 208         int                             error;
 209
 210         ASSERT(blk->magic == XFS_DIR2_LEAF1_MAGIC ||
 211                blk->magic == XFS_DIR2_LEAFN_MAGIC);
 212
 213         xfs_dir2_leaf_hdr_from_disk(mp, &hdr, blk->bp->b_addr);
 214         ent = hdr.ents + blk->index;
 215
 216         /* Check the hash of the entry. */
 217         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 218         if (error)
 219                 goto out;
 220
 221         /* Valid hash pointer? */
 222         ptr = be32_to_cpu(ent->address);
 223         if (ptr == 0)
 224                 return 0;
 225
 226         /* Find the directory entry's location. */
 227         db = xfs_dir2_dataptr_to_db(geo, ptr);
 228         off = xfs_dir2_dataptr_to_off(geo, ptr);
 229         rec_bno = xfs_dir2_db_to_da(geo, db);
 230
 231         if (rec_bno >= geo->leafblk) {
 232                 xchk_da_set_corrupt(ds, level);
 233                 goto out;
 234         }
 235         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno,
 236                         XFS_DABUF_MAP_HOLE_OK, &bp);
 237         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 238                         &error))
 239                 goto out;
 240         if (!bp) {
 241                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 242                 goto out;
 243         }
 244         xchk_buffer_recheck(ds->sc, bp);
 245
 246         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 247                 goto out_relse;
 248
 249         dent = bp->b_addr + off;
 250
 251         /* Make sure we got a real directory entry. */
 252         iter_off = geo->data_entry_offset;
 253         end = xfs_dir3_data_end_offset(geo, bp->b_addr);
 254         if (!end) {
 255                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 256                 goto out_relse;
 257         }
 258         for (;;) {
 259                 struct xfs_dir2_data_entry      *dep = bp->b_addr + iter_off;
 260                 struct xfs_dir2_data_unused     *dup = bp->b_addr + iter_off;
 261
 262                 if (iter_off >= end) {
 263                         xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 264                         goto out_relse;
 265                 }
 266
 267                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 268                         iter_off += be16_to_cpu(dup->length);
 269                         continue;
 270                 }
 271                 if (dep == dent)
 272                         break;
 273                 iter_off += xfs_dir2_data_entsize(mp, dep->namelen);
 274         }
 275
 276         /* Retrieve the entry, sanity check it, and compare hashes. */
 277         ino = be64_to_cpu(dent->inumber);
 278         hash = be32_to_cpu(ent->hashval);
 279         tag = be16_to_cpup(xfs_dir2_data_entry_tag_p(mp, dent));
 280         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 281                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 282         if (dent->namelen == 0) {
 283                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 284                 goto out_relse;
 285         }
 286         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 287         if (calc_hash != hash)
 288                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 289
 290 out_relse:
 291         xfs_trans_brelse(ds->dargs.trans, bp);
 292 out:
 293         return error;
 294 }
 295
 296 /*
 297  * Is this unused entry either in the bestfree or smaller than all of
 298  * them?  We've already checked that the bestfrees are sorted longest to
 299  * shortest, and that there aren't any bogus entries.
 300  */
 301 STATIC void
 302 xchk_directory_check_free_entry(
 303         struct xfs_scrub                *sc,
 304         xfs_dablk_t                     lblk,
 305         struct xfs_dir2_data_free       *bf,
 306         struct xfs_dir2_data_unused     *dup)
 307 {
 308         struct xfs_dir2_data_free       *dfp;
 309         unsigned int                    dup_length;
 310
 311         dup_length = be16_to_cpu(dup->length);
 312
 313         /* Unused entry is shorter than any of the bestfrees */
 314         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 315                 return;
 316
 317         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 318                 if (dup_length == be16_to_cpu(dfp->length))
 319                         return;
 320
 321         /* Unused entry should be in the bestfrees but wasn't found. */
 322         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 323 }
 324
 325 /* Check free space info in a directory data block. */
 326 STATIC int
 327 xchk_directory_data_bestfree(
 328         struct xfs_scrub                *sc,
 329         xfs_dablk_t                     lblk,
 330         bool                            is_block)
 331 {
 332         struct xfs_dir2_data_unused     *dup;
 333         struct xfs_dir2_data_free       *dfp;
 334         struct xfs_buf                  *bp;
 335         struct xfs_dir2_data_free       *bf;
 336         struct xfs_mount                *mp = sc->mp;
 337         u16                             tag;
 338         unsigned int                    nr_bestfrees = 0;
 339         unsigned int                    nr_frees = 0;
 340         unsigned int                    smallest_bestfree;
 341         int                             newlen;
 342         unsigned int                    offset;
 343         unsigned int                    end;
 344         int                             error;
 345
 346         if (is_block) {
 347                 /* dir block format */
 348                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 349                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 350                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 351         } else {
 352                 /* dir data format */
 353                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, 0, &bp);
 354         }
 355         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 356                 goto out;
 357         xchk_buffer_recheck(sc, bp);
 358
 359         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 360
 361         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 362                 goto out_buf;
 363
 364         /* Do the bestfrees correspond to actual free space? */
 365         bf = xfs_dir2_data_bestfree_p(mp, bp->b_addr);
 366         smallest_bestfree = UINT_MAX;
 367         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 368                 offset = be16_to_cpu(dfp->offset);
 369                 if (offset == 0)
 370                         continue;
 371                 if (offset >= mp->m_dir_geo->blksize) {
 372                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 373                         goto out_buf;
 374                 }
 375                 dup = bp->b_addr + offset;
 376                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 377
 378                 /* bestfree doesn't match the entry it points at? */
 379                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 380                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 381                     tag != offset) {
 382                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 383                         goto out_buf;
 384                 }
 385
 386                 /* bestfree records should be ordered largest to smallest */
 387                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 388                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 389                         goto out_buf;
 390                 }
 391
 392                 smallest_bestfree = be16_to_cpu(dfp->length);
 393                 nr_bestfrees++;
 394         }
 395
 396         /* Make sure the bestfrees are actually the best free spaces. */
 397         offset = mp->m_dir_geo->data_entry_offset;
 398         end = xfs_dir3_data_end_offset(mp->m_dir_geo, bp->b_addr);
 399
 400         /* Iterate the entries, stopping when we hit or go past the end. */
 401         while (offset < end) {
 402                 dup = bp->b_addr + offset;
 403
 404                 /* Skip real entries */
 405                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 406                         struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
 407
 408                         newlen = xfs_dir2_data_entsize(mp, dep->namelen);
 409                         if (newlen <= 0) {
 410                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 411                                                 lblk);
 412                                 goto out_buf;
 413                         }
 414                         offset += newlen;
 415                         continue;
 416                 }
 417
 418                 /* Spot check this free entry */
 419                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 420                 if (tag != offset) {
 421                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 422                         goto out_buf;
 423                 }
 424
 425                 /*
 426                  * Either this entry is a bestfree or it's smaller than
 427                  * any of the bestfrees.
 428                  */
 429                 xchk_directory_check_free_entry(sc, lblk, bf, dup);
 430                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 431                         goto out_buf;
 432
 433                 /* Move on. */
 434                 newlen = be16_to_cpu(dup->length);
 435                 if (newlen <= 0) {
 436                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 437                         goto out_buf;
 438                 }
 439                 offset += newlen;
 440                 if (offset <= end)
 441                         nr_frees++;
 442         }
 443
 444         /* We're required to fill all the space. */
 445         if (offset != end)
 446                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 447
 448         /* Did we see at least as many free slots as there are bestfrees? */
 449         if (nr_frees < nr_bestfrees)
 450                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 451 out_buf:
 452         xfs_trans_brelse(sc->tp, bp);
 453 out:
 454         return error;
 455 }
 456
 457 /*
 458  * Does the free space length in the free space index block ($len) match
 459  * the longest length in the directory data block's bestfree array?
 460  * Assume that we've already checked that the data block's bestfree
 461  * array is in order.
 462  */
 463 STATIC void
 464 xchk_directory_check_freesp(
 465         struct xfs_scrub                *sc,
 466         xfs_dablk_t                     lblk,
 467         struct xfs_buf                  *dbp,
 468         unsigned int                    len)
 469 {
 470         struct xfs_dir2_data_free       *dfp;
 471
 472         dfp = xfs_dir2_data_bestfree_p(sc->mp, dbp->b_addr);
 473
 474         if (len != be16_to_cpu(dfp->length))
 475                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 476
 477         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 478                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 479 }
 480
 481 /* Check free space info in a directory leaf1 block. */
 482 STATIC int
 483 xchk_directory_leaf1_bestfree(
 484         struct xfs_scrub                *sc,
 485         struct xfs_da_args              *args,
 486         xfs_dablk_t                     lblk)
 487 {
 488         struct xfs_dir3_icleaf_hdr      leafhdr;
 489         struct xfs_dir2_leaf_tail       *ltp;
 490         struct xfs_dir2_leaf            *leaf;
 491         struct xfs_buf                  *dbp;
 492         struct xfs_buf                  *bp;
 493         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 494         __be16                          *bestp;
 495         __u16                           best;
 496         __u32                           hash;
 497         __u32                           lasthash = 0;
 498         __u32                           bestcount;
 499         unsigned int                    stale = 0;
 500         int                             i;
 501         int                             error;
 502
 503         /* Read the free space block. */
 504         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, &bp);
 505         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 506                 return error;
 507         xchk_buffer_recheck(sc, bp);
 508
 509         leaf = bp->b_addr;
 510         xfs_dir2_leaf_hdr_from_disk(sc->ip->i_mount, &leafhdr, leaf);
 511         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 512         bestcount = be32_to_cpu(ltp->bestcount);
 513         bestp = xfs_dir2_leaf_bests_p(ltp);
 514
 515         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 516                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 517
 518                 if (hdr3->pad != cpu_to_be32(0))
 519                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 520         }
 521
 522         /*
 523          * There should be as many bestfree slots as there are dir data
 524          * blocks that can fit under i_size.
 525          */
 526         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 527                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 528                 goto out;
 529         }
 530
 531         /* Is the leaf count even remotely sane? */
 532         if (leafhdr.count > geo->leaf_max_ents) {
 533                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 534                 goto out;
 535         }
 536
 537         /* Leaves and bests don't overlap in leaf format. */
 538         if ((char *)&leafhdr.ents[leafhdr.count] > (char *)bestp) {
 539                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 540                 goto out;
 541         }
 542
 543         /* Check hash value order, count stale entries.  */
 544         for (i = 0; i < leafhdr.count; i++) {
 545                 hash = be32_to_cpu(leafhdr.ents[i].hashval);
 546                 if (i > 0 && lasthash > hash)
 547                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 548                 lasthash = hash;
 549                 if (leafhdr.ents[i].address ==
 550                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 551                         stale++;
 552         }
 553         if (leafhdr.stale != stale)
 554                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 555         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 556                 goto out;
 557
 558         /* Check all the bestfree entries. */
 559         for (i = 0; i < bestcount; i++, bestp++) {
 560                 best = be16_to_cpu(*bestp);
 561                 if (best == NULLDATAOFF)
 562                         continue;
 563                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 564                                 i * args->geo->fsbcount, 0, &dbp);
 565                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 566                                 &error))
 567                         break;
 568                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 569                 xfs_trans_brelse(sc->tp, dbp);
 570                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 571                         break;
 572         }
 573 out:
 574         xfs_trans_brelse(sc->tp, bp);
 575         return error;
 576 }
 577
 578 /* Check free space info in a directory freespace block. */
 579 STATIC int
 580 xchk_directory_free_bestfree(
 581         struct xfs_scrub                *sc,
 582         struct xfs_da_args              *args,
 583         xfs_dablk_t                     lblk)
 584 {
 585         struct xfs_dir3_icfree_hdr      freehdr;
 586         struct xfs_buf                  *dbp;
 587         struct xfs_buf                  *bp;
 588         __u16                           best;
 589         unsigned int                    stale = 0;
 590         int                             i;
 591         int                             error;
 592
 593         /* Read the free space block */
 594         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 595         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 596                 return error;
 597         xchk_buffer_recheck(sc, bp);
 598
 599         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 600                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 601
 602                 if (hdr3->pad != cpu_to_be32(0))
 603                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 604         }
 605
 606         /* Check all the entries. */
 607         xfs_dir2_free_hdr_from_disk(sc->ip->i_mount, &freehdr, bp->b_addr);
 608         for (i = 0; i < freehdr.nvalid; i++) {
 609                 best = be16_to_cpu(freehdr.bests[i]);
 610                 if (best == NULLDATAOFF) {
 611                         stale++;
 612                         continue;
 613                 }
 614                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 615                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 616                                 0, &dbp);
 617                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 618                                 &error))
 619                         goto out;
 620                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 621                 xfs_trans_brelse(sc->tp, dbp);
 622         }
 623
 624         if (freehdr.nused + stale != freehdr.nvalid)
 625                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 626 out:
 627         xfs_trans_brelse(sc->tp, bp);
 628         return error;
 629 }
 630
 631 /* Check free space information in directories. */
 632 STATIC int
 633 xchk_directory_blocks(
 634         struct xfs_scrub        *sc)
 635 {
 636         struct xfs_bmbt_irec    got;
 637         struct xfs_da_args      args;
 638         struct xfs_ifork        *ifp;
 639         struct xfs_mount        *mp = sc->mp;
 640         xfs_fileoff_t           leaf_lblk;
 641         xfs_fileoff_t           free_lblk;
 642         xfs_fileoff_t           lblk;
 643         struct xfs_iext_cursor  icur;
 644         xfs_dablk_t             dabno;
 645         bool                    found;
 646         int                     is_block = 0;
 647         int                     error;
 648
 649         /* Ignore local format directories. */
 650         if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 651             sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
 652                 return 0;
 653
 654         ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 655         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 656         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 657         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 658
 659         /* Is this a block dir? */
 660         args.dp = sc->ip;
 661         args.geo = mp->m_dir_geo;
 662         args.trans = sc->tp;
 663         error = xfs_dir2_isblock(&args, &is_block);
 664         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 665                 goto out;
 666
 667         /* Iterate all the data extents in the directory... */
 668         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 669         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 670                 /* Block directories only have a single block at offset 0. */
 671                 if (is_block &&
 672                     (got.br_startoff > 0 ||
 673                      got.br_blockcount != args.geo->fsbcount)) {
 674                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 675                                         got.br_startoff);
 676                         break;
 677                 }
 678
 679                 /* No more data blocks... */
 680                 if (got.br_startoff >= leaf_lblk)
 681                         break;
 682
 683                 /*
 684                  * Check each data block's bestfree data.
 685                  *
 686                  * Iterate all the fsbcount-aligned block offsets in
 687                  * this directory.  The directory block reading code is
 688                  * smart enough to do its own bmap lookups to handle
 689                  * discontiguous directory blocks.  When we're done
 690                  * with the extent record, re-query the bmap at the
 691                  * next fsbcount-aligned offset to avoid redundant
 692                  * block checks.
 693                  */
 694                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 695                                 args.geo->fsbcount);
 696                      lblk < got.br_startoff + got.br_blockcount;
 697                      lblk += args.geo->fsbcount) {
 698                         error = xchk_directory_data_bestfree(sc, lblk,
 699                                         is_block);
 700                         if (error)
 701                                 goto out;
 702                 }
 703                 dabno = got.br_startoff + got.br_blockcount;
 704                 lblk = roundup(dabno, args.geo->fsbcount);
 705                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 706         }
 707
 708         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 709                 goto out;
 710
 711         /* Look for a leaf1 block, which has free info. */
 712         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 713             got.br_startoff == leaf_lblk &&
 714             got.br_blockcount == args.geo->fsbcount &&
 715             !xfs_iext_next_extent(ifp, &icur, &got)) {
 716                 if (is_block) {
 717                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 718                         goto out;
 719                 }
 720                 error = xchk_directory_leaf1_bestfree(sc, &args,
 721                                 leaf_lblk);
 722                 if (error)
 723                         goto out;
 724         }
 725
 726         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 727                 goto out;
 728
 729         /* Scan for free blocks */
 730         lblk = free_lblk;
 731         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 732         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 733                 /*
 734                  * Dirs can't have blocks mapped above 2^32.
 735                  * Single-block dirs shouldn't even be here.
 736                  */
 737                 lblk = got.br_startoff;
 738                 if (lblk & ~0xFFFFFFFFULL) {
 739                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 740                         goto out;
 741                 }
 742                 if (is_block) {
 743                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 744                         goto out;
 745                 }
 746
 747                 /*
 748                  * Check each dir free block's bestfree data.
 749                  *
 750                  * Iterate all the fsbcount-aligned block offsets in
 751                  * this directory.  The directory block reading code is
 752                  * smart enough to do its own bmap lookups to handle
 753                  * discontiguous directory blocks.  When we're done
 754                  * with the extent record, re-query the bmap at the
 755                  * next fsbcount-aligned offset to avoid redundant
 756                  * block checks.
 757                  */
 758                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 759                                 args.geo->fsbcount);
 760                      lblk < got.br_startoff + got.br_blockcount;
 761                      lblk += args.geo->fsbcount) {
 762                         error = xchk_directory_free_bestfree(sc, &args,
 763                                         lblk);
 764                         if (error)
 765                                 goto out;
 766                 }
 767                 dabno = got.br_startoff + got.br_blockcount;
 768                 lblk = roundup(dabno, args.geo->fsbcount);
 769                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 770         }
 771 out:
 772         return error;
 773 }
 774
 775 /* Scrub a whole directory. */
 776 int
 777 xchk_directory(
 778         struct xfs_scrub        *sc)
 779 {
 780         struct xchk_dir_ctx     sdc = {
 781                 .dir_iter.actor = xchk_dir_actor,
 782                 .dir_iter.pos = 0,
 783                 .sc = sc,
 784         };
 785         size_t                  bufsize;
 786         loff_t                  oldpos;
 787         int                     error = 0;
 788
 789         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 790                 return -ENOENT;
 791
 792         /* Plausible size? */
 793         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 794                 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
 795                 goto out;
 796         }
 797
 798         /* Check directory tree structure */
 799         error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
 800         if (error)
 801                 return error;
 802
 803         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 804                 return error;
 805
 806         /* Check the freespace. */
 807         error = xchk_directory_blocks(sc);
 808         if (error)
 809                 return error;
 810
 811         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 812                 return error;
 813
 814         /*
 815          * Check that every dirent we see can also be looked up by hash.
 816          * Userspace usually asks for a 32k buffer, so we will too.
 817          */
 818         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 819                         sc->ip->i_d.di_size);
 820
 821         /*
 822          * Look up every name in this directory by hash.
 823          *
 824          * Use the xfs_readdir function to call xchk_dir_actor on
 825          * every directory entry in this directory.  In _actor, we check
 826          * the name, inode number, and ftype (if applicable) of the
 827          * entry.  xfs_readdir uses the VFS filldir functions to provide
 828          * iteration context.
 829          *
 830          * The VFS grabs a read or write lock via i_rwsem before it reads
 831          * or writes to a directory.  If we've gotten this far we've
 832          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 833          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 834          * to drop the ILOCK here in order to reuse the _readdir and
 835          * _dir_lookup routines, which do their own ILOCK locking.
 836          */
 837         oldpos = 0;
 838         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 839         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 840         while (true) {
 841                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 842                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
 843                                 &error))
 844                         goto out;
 845                 if (oldpos == sdc.dir_iter.pos)
 846                         break;
 847                 oldpos = sdc.dir_iter.pos;
 848         }
 849
 850 out:
 851         return error;
 852 }