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 {
  27         return xchk_setup_inode_contents(sc, 0);
  28 }
  29
  30 /* Directories */
  31
  32 /* Scrub a directory entry. */
  33
  34 struct xchk_dir_ctx {
  35         /* VFS fill-directory iterator */
  36         struct dir_context      dir_iter;
  37
  38         struct xfs_scrub        *sc;
  39 };
  40
  41 /* Check that an inode's mode matches a given DT_ type. */
  42 STATIC int
  43 xchk_dir_check_ftype(
  44         struct xchk_dir_ctx     *sdc,
  45         xfs_fileoff_t           offset,
  46         xfs_ino_t               inum,
  47         int                     dtype)
  48 {
  49         struct xfs_mount        *mp = sdc->sc->mp;
  50         struct xfs_inode        *ip;
  51         int                     ino_dtype;
  52         int                     error = 0;
  53
  54         if (!xfs_sb_version_hasftype(&mp->m_sb)) {
  55                 if (dtype != DT_UNKNOWN && dtype != DT_DIR)
  56                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
  57                                         offset);
  58                 goto out;
  59         }
  60
  61         /*
  62          * Grab the inode pointed to by the dirent.  We release the
  63          * inode before we cancel the scrub transaction.  Since we're
  64          * don't know a priori that releasing the inode won't trigger
  65          * eofblocks cleanup (which allocates what would be a nested
  66          * transaction), we can't use DONTCACHE here because DONTCACHE
  67          * inodes can trigger immediate inactive cleanup of the inode.
  68          *
  69          * If _iget returns -EINVAL or -ENOENT then the child inode number is
  70          * garbage and the directory is corrupt.  If the _iget returns
  71          * -EFSCORRUPTED or -EFSBADCRC then the child is corrupt which is a
  72          *  cross referencing error.  Any other error is an operational error.
  73          */
  74         error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
  75         if (error == -EINVAL || error == -ENOENT) {
  76                 error = -EFSCORRUPTED;
  77                 xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, 0, &error);
  78                 goto out;
  79         }
  80         if (!xchk_fblock_xref_process_error(sdc->sc, XFS_DATA_FORK, offset,
  81                         &error))
  82                 goto out;
  83
  84         /* Convert mode to the DT_* values that dir_emit uses. */
  85         ino_dtype = xfs_dir3_get_dtype(mp,
  86                         xfs_mode_to_ftype(VFS_I(ip)->i_mode));
  87         if (ino_dtype != dtype)
  88                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
  89         xfs_irele(ip);
  90 out:
  91         return error;
  92 }
  93
  94 /*
  95  * Scrub a single directory entry.
  96  *
  97  * We use the VFS directory iterator (i.e. readdir) to call this
  98  * function for every directory entry in a directory.  Once we're here,
  99  * we check the inode number to make sure it's sane, then we check that
 100  * we can look up this filename.  Finally, we check the ftype.
 101  */
 102 STATIC int
 103 xchk_dir_actor(
 104         struct dir_context      *dir_iter,
 105         const char              *name,
 106         int                     namelen,
 107         loff_t                  pos,
 108         u64                     ino,
 109         unsigned                type)
 110 {
 111         struct xfs_mount        *mp;
 112         struct xfs_inode        *ip;
 113         struct xchk_dir_ctx     *sdc;
 114         struct xfs_name         xname;
 115         xfs_ino_t               lookup_ino;
 116         xfs_dablk_t             offset;
 117         bool                    checked_ftype = false;
 118         int                     error = 0;
 119
 120         sdc = container_of(dir_iter, struct xchk_dir_ctx, dir_iter);
 121         ip = sdc->sc->ip;
 122         mp = ip->i_mount;
 123         offset = xfs_dir2_db_to_da(mp->m_dir_geo,
 124                         xfs_dir2_dataptr_to_db(mp->m_dir_geo, pos));
 125
 126         if (xchk_should_terminate(sdc->sc, &error))
 127                 return error;
 128
 129         /* Does this inode number make sense? */
 130         if (!xfs_verify_dir_ino(mp, ino)) {
 131                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 132                 goto out;
 133         }
 134
 135         /* Does this name make sense? */
 136         if (!xfs_dir2_namecheck(name, namelen)) {
 137                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 138                 goto out;
 139         }
 140
 141         if (!strncmp(".", name, namelen)) {
 142                 /* If this is "." then check that the inum matches the dir. */
 143                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 144                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 145                                         offset);
 146                 checked_ftype = true;
 147                 if (ino != ip->i_ino)
 148                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 149                                         offset);
 150         } else if (!strncmp("..", name, namelen)) {
 151                 /*
 152                  * If this is ".." in the root inode, check that the inum
 153                  * matches this dir.
 154                  */
 155                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 156                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 157                                         offset);
 158                 checked_ftype = true;
 159                 if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
 160                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 161                                         offset);
 162         }
 163
 164         /* Verify that we can look up this name by hash. */
 165         xname.name = name;
 166         xname.len = namelen;
 167         xname.type = XFS_DIR3_FT_UNKNOWN;
 168
 169         error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
 170         /* ENOENT means the hash lookup failed and the dir is corrupt */
 171         if (error == -ENOENT)
 172                 error = -EFSCORRUPTED;
 173         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 174                         &error))
 175                 goto out;
 176         if (lookup_ino != ino) {
 177                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 178                 goto out;
 179         }
 180
 181         /* Verify the file type.  This function absorbs error codes. */
 182         if (!checked_ftype) {
 183                 error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 184                 if (error)
 185                         goto out;
 186         }
 187 out:
 188         /*
 189          * A negative error code returned here is supposed to cause the
 190          * dir_emit caller (xfs_readdir) to abort the directory iteration
 191          * and return zero to xchk_directory.
 192          */
 193         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 194                 return -EFSCORRUPTED;
 195         return error;
 196 }
 197
 198 /* Scrub a directory btree record. */
 199 STATIC int
 200 xchk_dir_rec(
 201         struct xchk_da_btree            *ds,
 202         int                             level)
 203 {
 204         struct xfs_da_state_blk         *blk = &ds->state->path.blk[level];
 205         struct xfs_mount                *mp = ds->state->mp;
 206         struct xfs_inode                *dp = ds->dargs.dp;
 207         struct xfs_da_geometry          *geo = mp->m_dir_geo;
 208         struct xfs_dir2_data_entry      *dent;
 209         struct xfs_buf                  *bp;
 210         struct xfs_dir2_leaf_entry      *ent;
 211         unsigned int                    end;
 212         unsigned int                    iter_off;
 213         xfs_ino_t                       ino;
 214         xfs_dablk_t                     rec_bno;
 215         xfs_dir2_db_t                   db;
 216         xfs_dir2_data_aoff_t            off;
 217         xfs_dir2_dataptr_t              ptr;
 218         xfs_dahash_t                    calc_hash;
 219         xfs_dahash_t                    hash;
 220         struct xfs_dir3_icleaf_hdr      hdr;
 221         unsigned int                    tag;
 222         int                             error;
 223
 224         ASSERT(blk->magic == XFS_DIR2_LEAF1_MAGIC ||
 225                blk->magic == XFS_DIR2_LEAFN_MAGIC);
 226
 227         xfs_dir2_leaf_hdr_from_disk(mp, &hdr, blk->bp->b_addr);
 228         ent = hdr.ents + blk->index;
 229
 230         /* Check the hash of the entry. */
 231         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 232         if (error)
 233                 goto out;
 234
 235         /* Valid hash pointer? */
 236         ptr = be32_to_cpu(ent->address);
 237         if (ptr == 0)
 238                 return 0;
 239
 240         /* Find the directory entry's location. */
 241         db = xfs_dir2_dataptr_to_db(geo, ptr);
 242         off = xfs_dir2_dataptr_to_off(geo, ptr);
 243         rec_bno = xfs_dir2_db_to_da(geo, db);
 244
 245         if (rec_bno >= geo->leafblk) {
 246                 xchk_da_set_corrupt(ds, level);
 247                 goto out;
 248         }
 249         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno,
 250                         XFS_DABUF_MAP_HOLE_OK, &bp);
 251         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 252                         &error))
 253                 goto out;
 254         if (!bp) {
 255                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 256                 goto out;
 257         }
 258         xchk_buffer_recheck(ds->sc, bp);
 259
 260         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 261                 goto out_relse;
 262
 263         dent = bp->b_addr + off;
 264
 265         /* Make sure we got a real directory entry. */
 266         iter_off = geo->data_entry_offset;
 267         end = xfs_dir3_data_end_offset(geo, bp->b_addr);
 268         if (!end) {
 269                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 270                 goto out_relse;
 271         }
 272         for (;;) {
 273                 struct xfs_dir2_data_entry      *dep = bp->b_addr + iter_off;
 274                 struct xfs_dir2_data_unused     *dup = bp->b_addr + iter_off;
 275
 276                 if (iter_off >= end) {
 277                         xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 278                         goto out_relse;
 279                 }
 280
 281                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 282                         iter_off += be16_to_cpu(dup->length);
 283                         continue;
 284                 }
 285                 if (dep == dent)
 286                         break;
 287                 iter_off += xfs_dir2_data_entsize(mp, dep->namelen);
 288         }
 289
 290         /* Retrieve the entry, sanity check it, and compare hashes. */
 291         ino = be64_to_cpu(dent->inumber);
 292         hash = be32_to_cpu(ent->hashval);
 293         tag = be16_to_cpup(xfs_dir2_data_entry_tag_p(mp, dent));
 294         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 295                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 296         if (dent->namelen == 0) {
 297                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 298                 goto out_relse;
 299         }
 300         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 301         if (calc_hash != hash)
 302                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 303
 304 out_relse:
 305         xfs_trans_brelse(ds->dargs.trans, bp);
 306 out:
 307         return error;
 308 }
 309
 310 /*
 311  * Is this unused entry either in the bestfree or smaller than all of
 312  * them?  We've already checked that the bestfrees are sorted longest to
 313  * shortest, and that there aren't any bogus entries.
 314  */
 315 STATIC void
 316 xchk_directory_check_free_entry(
 317         struct xfs_scrub                *sc,
 318         xfs_dablk_t                     lblk,
 319         struct xfs_dir2_data_free       *bf,
 320         struct xfs_dir2_data_unused     *dup)
 321 {
 322         struct xfs_dir2_data_free       *dfp;
 323         unsigned int                    dup_length;
 324
 325         dup_length = be16_to_cpu(dup->length);
 326
 327         /* Unused entry is shorter than any of the bestfrees */
 328         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 329                 return;
 330
 331         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 332                 if (dup_length == be16_to_cpu(dfp->length))
 333                         return;
 334
 335         /* Unused entry should be in the bestfrees but wasn't found. */
 336         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 337 }
 338
 339 /* Check free space info in a directory data block. */
 340 STATIC int
 341 xchk_directory_data_bestfree(
 342         struct xfs_scrub                *sc,
 343         xfs_dablk_t                     lblk,
 344         bool                            is_block)
 345 {
 346         struct xfs_dir2_data_unused     *dup;
 347         struct xfs_dir2_data_free       *dfp;
 348         struct xfs_buf                  *bp;
 349         struct xfs_dir2_data_free       *bf;
 350         struct xfs_mount                *mp = sc->mp;
 351         u16                             tag;
 352         unsigned int                    nr_bestfrees = 0;
 353         unsigned int                    nr_frees = 0;
 354         unsigned int                    smallest_bestfree;
 355         int                             newlen;
 356         unsigned int                    offset;
 357         unsigned int                    end;
 358         int                             error;
 359
 360         if (is_block) {
 361                 /* dir block format */
 362                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 363                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 364                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 365         } else {
 366                 /* dir data format */
 367                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, 0, &bp);
 368         }
 369         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 370                 goto out;
 371         xchk_buffer_recheck(sc, bp);
 372
 373         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 374
 375         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 376                 goto out_buf;
 377
 378         /* Do the bestfrees correspond to actual free space? */
 379         bf = xfs_dir2_data_bestfree_p(mp, bp->b_addr);
 380         smallest_bestfree = UINT_MAX;
 381         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 382                 offset = be16_to_cpu(dfp->offset);
 383                 if (offset == 0)
 384                         continue;
 385                 if (offset >= mp->m_dir_geo->blksize) {
 386                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 387                         goto out_buf;
 388                 }
 389                 dup = bp->b_addr + offset;
 390                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 391
 392                 /* bestfree doesn't match the entry it points at? */
 393                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 394                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 395                     tag != offset) {
 396                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 397                         goto out_buf;
 398                 }
 399
 400                 /* bestfree records should be ordered largest to smallest */
 401                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 402                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 403                         goto out_buf;
 404                 }
 405
 406                 smallest_bestfree = be16_to_cpu(dfp->length);
 407                 nr_bestfrees++;
 408         }
 409
 410         /* Make sure the bestfrees are actually the best free spaces. */
 411         offset = mp->m_dir_geo->data_entry_offset;
 412         end = xfs_dir3_data_end_offset(mp->m_dir_geo, bp->b_addr);
 413
 414         /* Iterate the entries, stopping when we hit or go past the end. */
 415         while (offset < end) {
 416                 dup = bp->b_addr + offset;
 417
 418                 /* Skip real entries */
 419                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 420                         struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
 421
 422                         newlen = xfs_dir2_data_entsize(mp, dep->namelen);
 423                         if (newlen <= 0) {
 424                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 425                                                 lblk);
 426                                 goto out_buf;
 427                         }
 428                         offset += newlen;
 429                         continue;
 430                 }
 431
 432                 /* Spot check this free entry */
 433                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 434                 if (tag != offset) {
 435                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 436                         goto out_buf;
 437                 }
 438
 439                 /*
 440                  * Either this entry is a bestfree or it's smaller than
 441                  * any of the bestfrees.
 442                  */
 443                 xchk_directory_check_free_entry(sc, lblk, bf, dup);
 444                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 445                         goto out_buf;
 446
 447                 /* Move on. */
 448                 newlen = be16_to_cpu(dup->length);
 449                 if (newlen <= 0) {
 450                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 451                         goto out_buf;
 452                 }
 453                 offset += newlen;
 454                 if (offset <= end)
 455                         nr_frees++;
 456         }
 457
 458         /* We're required to fill all the space. */
 459         if (offset != end)
 460                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 461
 462         /* Did we see at least as many free slots as there are bestfrees? */
 463         if (nr_frees < nr_bestfrees)
 464                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 465 out_buf:
 466         xfs_trans_brelse(sc->tp, bp);
 467 out:
 468         return error;
 469 }
 470
 471 /*
 472  * Does the free space length in the free space index block ($len) match
 473  * the longest length in the directory data block's bestfree array?
 474  * Assume that we've already checked that the data block's bestfree
 475  * array is in order.
 476  */
 477 STATIC void
 478 xchk_directory_check_freesp(
 479         struct xfs_scrub                *sc,
 480         xfs_dablk_t                     lblk,
 481         struct xfs_buf                  *dbp,
 482         unsigned int                    len)
 483 {
 484         struct xfs_dir2_data_free       *dfp;
 485
 486         dfp = xfs_dir2_data_bestfree_p(sc->mp, dbp->b_addr);
 487
 488         if (len != be16_to_cpu(dfp->length))
 489                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 490
 491         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 492                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 493 }
 494
 495 /* Check free space info in a directory leaf1 block. */
 496 STATIC int
 497 xchk_directory_leaf1_bestfree(
 498         struct xfs_scrub                *sc,
 499         struct xfs_da_args              *args,
 500         xfs_dablk_t                     lblk)
 501 {
 502         struct xfs_dir3_icleaf_hdr      leafhdr;
 503         struct xfs_dir2_leaf_tail       *ltp;
 504         struct xfs_dir2_leaf            *leaf;
 505         struct xfs_buf                  *dbp;
 506         struct xfs_buf                  *bp;
 507         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 508         __be16                          *bestp;
 509         __u16                           best;
 510         __u32                           hash;
 511         __u32                           lasthash = 0;
 512         __u32                           bestcount;
 513         unsigned int                    stale = 0;
 514         int                             i;
 515         int                             error;
 516
 517         /* Read the free space block. */
 518         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, &bp);
 519         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 520                 return error;
 521         xchk_buffer_recheck(sc, bp);
 522
 523         leaf = bp->b_addr;
 524         xfs_dir2_leaf_hdr_from_disk(sc->ip->i_mount, &leafhdr, leaf);
 525         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 526         bestcount = be32_to_cpu(ltp->bestcount);
 527         bestp = xfs_dir2_leaf_bests_p(ltp);
 528
 529         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 530                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 531
 532                 if (hdr3->pad != cpu_to_be32(0))
 533                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 534         }
 535
 536         /*
 537          * There should be as many bestfree slots as there are dir data
 538          * blocks that can fit under i_size.
 539          */
 540         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_disk_size)) {
 541                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 542                 goto out;
 543         }
 544
 545         /* Is the leaf count even remotely sane? */
 546         if (leafhdr.count > geo->leaf_max_ents) {
 547                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 548                 goto out;
 549         }
 550
 551         /* Leaves and bests don't overlap in leaf format. */
 552         if ((char *)&leafhdr.ents[leafhdr.count] > (char *)bestp) {
 553                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 554                 goto out;
 555         }
 556
 557         /* Check hash value order, count stale entries.  */
 558         for (i = 0; i < leafhdr.count; i++) {
 559                 hash = be32_to_cpu(leafhdr.ents[i].hashval);
 560                 if (i > 0 && lasthash > hash)
 561                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 562                 lasthash = hash;
 563                 if (leafhdr.ents[i].address ==
 564                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 565                         stale++;
 566         }
 567         if (leafhdr.stale != stale)
 568                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 569         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 570                 goto out;
 571
 572         /* Check all the bestfree entries. */
 573         for (i = 0; i < bestcount; i++, bestp++) {
 574                 best = be16_to_cpu(*bestp);
 575                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 576                                 xfs_dir2_db_to_da(args->geo, i),
 577                                 XFS_DABUF_MAP_HOLE_OK,
 578                                 &dbp);
 579                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 580                                 &error))
 581                         break;
 582
 583                 if (!dbp) {
 584                         if (best != NULLDATAOFF) {
 585                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 586                                                 lblk);
 587                                 break;
 588                         }
 589                         continue;
 590                 }
 591
 592                 if (best == NULLDATAOFF)
 593                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 594                 else
 595                         xchk_directory_check_freesp(sc, lblk, dbp, best);
 596                 xfs_trans_brelse(sc->tp, dbp);
 597                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 598                         break;
 599         }
 600 out:
 601         xfs_trans_brelse(sc->tp, bp);
 602         return error;
 603 }
 604
 605 /* Check free space info in a directory freespace block. */
 606 STATIC int
 607 xchk_directory_free_bestfree(
 608         struct xfs_scrub                *sc,
 609         struct xfs_da_args              *args,
 610         xfs_dablk_t                     lblk)
 611 {
 612         struct xfs_dir3_icfree_hdr      freehdr;
 613         struct xfs_buf                  *dbp;
 614         struct xfs_buf                  *bp;
 615         __u16                           best;
 616         unsigned int                    stale = 0;
 617         int                             i;
 618         int                             error;
 619
 620         /* Read the free space block */
 621         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 622         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 623                 return error;
 624         xchk_buffer_recheck(sc, bp);
 625
 626         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 627                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 628
 629                 if (hdr3->pad != cpu_to_be32(0))
 630                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 631         }
 632
 633         /* Check all the entries. */
 634         xfs_dir2_free_hdr_from_disk(sc->ip->i_mount, &freehdr, bp->b_addr);
 635         for (i = 0; i < freehdr.nvalid; i++) {
 636                 best = be16_to_cpu(freehdr.bests[i]);
 637                 if (best == NULLDATAOFF) {
 638                         stale++;
 639                         continue;
 640                 }
 641                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 642                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 643                                 0, &dbp);
 644                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 645                                 &error))
 646                         goto out;
 647                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 648                 xfs_trans_brelse(sc->tp, dbp);
 649         }
 650
 651         if (freehdr.nused + stale != freehdr.nvalid)
 652                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 653 out:
 654         xfs_trans_brelse(sc->tp, bp);
 655         return error;
 656 }
 657
 658 /* Check free space information in directories. */
 659 STATIC int
 660 xchk_directory_blocks(
 661         struct xfs_scrub        *sc)
 662 {
 663         struct xfs_bmbt_irec    got;
 664         struct xfs_da_args      args;
 665         struct xfs_ifork        *ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 666         struct xfs_mount        *mp = sc->mp;
 667         xfs_fileoff_t           leaf_lblk;
 668         xfs_fileoff_t           free_lblk;
 669         xfs_fileoff_t           lblk;
 670         struct xfs_iext_cursor  icur;
 671         xfs_dablk_t             dabno;
 672         bool                    found;
 673         int                     is_block = 0;
 674         int                     error;
 675
 676         /* Ignore local format directories. */
 677         if (ifp->if_format != XFS_DINODE_FMT_EXTENTS &&
 678             ifp->if_format != XFS_DINODE_FMT_BTREE)
 679                 return 0;
 680
 681         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 682         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 683         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 684
 685         /* Is this a block dir? */
 686         args.dp = sc->ip;
 687         args.geo = mp->m_dir_geo;
 688         args.trans = sc->tp;
 689         error = xfs_dir2_isblock(&args, &is_block);
 690         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 691                 goto out;
 692
 693         /* Iterate all the data extents in the directory... */
 694         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 695         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 696                 /* No more data blocks... */
 697                 if (got.br_startoff >= leaf_lblk)
 698                         break;
 699
 700                 /*
 701                  * Check each data block's bestfree data.
 702                  *
 703                  * Iterate all the fsbcount-aligned block offsets in
 704                  * this directory.  The directory block reading code is
 705                  * smart enough to do its own bmap lookups to handle
 706                  * discontiguous directory blocks.  When we're done
 707                  * with the extent record, re-query the bmap at the
 708                  * next fsbcount-aligned offset to avoid redundant
 709                  * block checks.
 710                  */
 711                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 712                                 args.geo->fsbcount);
 713                      lblk < got.br_startoff + got.br_blockcount;
 714                      lblk += args.geo->fsbcount) {
 715                         error = xchk_directory_data_bestfree(sc, lblk,
 716                                         is_block);
 717                         if (error)
 718                                 goto out;
 719                 }
 720                 dabno = got.br_startoff + got.br_blockcount;
 721                 lblk = roundup(dabno, args.geo->fsbcount);
 722                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 723         }
 724
 725         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 726                 goto out;
 727
 728         /* Look for a leaf1 block, which has free info. */
 729         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 730             got.br_startoff == leaf_lblk &&
 731             got.br_blockcount == args.geo->fsbcount &&
 732             !xfs_iext_next_extent(ifp, &icur, &got)) {
 733                 if (is_block) {
 734                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 735                         goto out;
 736                 }
 737                 error = xchk_directory_leaf1_bestfree(sc, &args,
 738                                 leaf_lblk);
 739                 if (error)
 740                         goto out;
 741         }
 742
 743         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 744                 goto out;
 745
 746         /* Scan for free blocks */
 747         lblk = free_lblk;
 748         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 749         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 750                 /*
 751                  * Dirs can't have blocks mapped above 2^32.
 752                  * Single-block dirs shouldn't even be here.
 753                  */
 754                 lblk = got.br_startoff;
 755                 if (lblk & ~0xFFFFFFFFULL) {
 756                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 757                         goto out;
 758                 }
 759                 if (is_block) {
 760                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 761                         goto out;
 762                 }
 763
 764                 /*
 765                  * Check each dir free block's bestfree data.
 766                  *
 767                  * Iterate all the fsbcount-aligned block offsets in
 768                  * this directory.  The directory block reading code is
 769                  * smart enough to do its own bmap lookups to handle
 770                  * discontiguous directory blocks.  When we're done
 771                  * with the extent record, re-query the bmap at the
 772                  * next fsbcount-aligned offset to avoid redundant
 773                  * block checks.
 774                  */
 775                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 776                                 args.geo->fsbcount);
 777                      lblk < got.br_startoff + got.br_blockcount;
 778                      lblk += args.geo->fsbcount) {
 779                         error = xchk_directory_free_bestfree(sc, &args,
 780                                         lblk);
 781                         if (error)
 782                                 goto out;
 783                 }
 784                 dabno = got.br_startoff + got.br_blockcount;
 785                 lblk = roundup(dabno, args.geo->fsbcount);
 786                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 787         }
 788 out:
 789         return error;
 790 }
 791
 792 /* Scrub a whole directory. */
 793 int
 794 xchk_directory(
 795         struct xfs_scrub        *sc)
 796 {
 797         struct xchk_dir_ctx     sdc = {
 798                 .dir_iter.actor = xchk_dir_actor,
 799                 .dir_iter.pos = 0,
 800                 .sc = sc,
 801         };
 802         size_t                  bufsize;
 803         loff_t                  oldpos;
 804         int                     error = 0;
 805
 806         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 807                 return -ENOENT;
 808
 809         /* Plausible size? */
 810         if (sc->ip->i_disk_size < xfs_dir2_sf_hdr_size(0)) {
 811                 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
 812                 goto out;
 813         }
 814
 815         /* Check directory tree structure */
 816         error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
 817         if (error)
 818                 return error;
 819
 820         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 821                 return error;
 822
 823         /* Check the freespace. */
 824         error = xchk_directory_blocks(sc);
 825         if (error)
 826                 return error;
 827
 828         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 829                 return error;
 830
 831         /*
 832          * Check that every dirent we see can also be looked up by hash.
 833          * Userspace usually asks for a 32k buffer, so we will too.
 834          */
 835         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 836                         sc->ip->i_disk_size);
 837
 838         /*
 839          * Look up every name in this directory by hash.
 840          *
 841          * Use the xfs_readdir function to call xchk_dir_actor on
 842          * every directory entry in this directory.  In _actor, we check
 843          * the name, inode number, and ftype (if applicable) of the
 844          * entry.  xfs_readdir uses the VFS filldir functions to provide
 845          * iteration context.
 846          *
 847          * The VFS grabs a read or write lock via i_rwsem before it reads
 848          * or writes to a directory.  If we've gotten this far we've
 849          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 850          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 851          * to drop the ILOCK here in order to reuse the _readdir and
 852          * _dir_lookup routines, which do their own ILOCK locking.
 853          */
 854         oldpos = 0;
 855         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 856         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 857         while (true) {
 858                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 859                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
 860                                 &error))
 861                         goto out;
 862                 if (oldpos == sdc.dir_iter.pos)
 863                         break;
 864                 oldpos = sdc.dir_iter.pos;
 865         }
 866
 867 out:
 868         return error;
 869 }