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_defer.h"
  13 #include "xfs_btree.h"
  14 #include "xfs_bit.h"
  15 #include "xfs_log_format.h"
  16 #include "xfs_trans.h"
  17 #include "xfs_sb.h"
  18 #include "xfs_inode.h"
  19 #include "xfs_icache.h"
  20 #include "xfs_itable.h"
  21 #include "xfs_da_format.h"
  22 #include "xfs_da_btree.h"
  23 #include "xfs_dir2.h"
  24 #include "xfs_dir2_priv.h"
  25 #include "xfs_ialloc.h"
  26 #include "scrub/xfs_scrub.h"
  27 #include "scrub/scrub.h"
  28 #include "scrub/common.h"
  29 #include "scrub/trace.h"
  30 #include "scrub/dabtree.h"
  31
  32 /* Set us up to scrub directories. */
  33 int
  34 xchk_setup_directory(
  35         struct xfs_scrub        *sc,
  36         struct xfs_inode        *ip)
  37 {
  38         return xchk_setup_inode_contents(sc, ip, 0);
  39 }
  40
  41 /* Directories */
  42
  43 /* Scrub a directory entry. */
  44
  45 struct xchk_dir_ctx {
  46         /* VFS fill-directory iterator */
  47         struct dir_context      dir_iter;
  48
  49         struct xfs_scrub        *sc;
  50 };
  51
  52 /* Check that an inode's mode matches a given DT_ type. */
  53 STATIC int
  54 xchk_dir_check_ftype(
  55         struct xchk_dir_ctx     *sdc,
  56         xfs_fileoff_t           offset,
  57         xfs_ino_t               inum,
  58         int                     dtype)
  59 {
  60         struct xfs_mount        *mp = sdc->sc->mp;
  61         struct xfs_inode        *ip;
  62         int                     ino_dtype;
  63         int                     error = 0;
  64
  65         if (!xfs_sb_version_hasftype(&mp->m_sb)) {
  66                 if (dtype != DT_UNKNOWN && dtype != DT_DIR)
  67                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
  68                                         offset);
  69                 goto out;
  70         }
  71
  72         /*
  73          * Grab the inode pointed to by the dirent.  We release the
  74          * inode before we cancel the scrub transaction.  Since we're
  75          * don't know a priori that releasing the inode won't trigger
  76          * eofblocks cleanup (which allocates what would be a nested
  77          * transaction), we can't use DONTCACHE here because DONTCACHE
  78          * inodes can trigger immediate inactive cleanup of the inode.
  79          */
  80         error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
  81         if (!xchk_fblock_xref_process_error(sdc->sc, XFS_DATA_FORK, offset,
  82                         &error))
  83                 goto out;
  84
  85         /* Convert mode to the DT_* values that dir_emit uses. */
  86         ino_dtype = xfs_dir3_get_dtype(mp,
  87                         xfs_mode_to_ftype(VFS_I(ip)->i_mode));
  88         if (ino_dtype != dtype)
  89                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
  90         xfs_irele(ip);
  91 out:
  92         return error;
  93 }
  94
  95 /*
  96  * Scrub a single directory entry.
  97  *
  98  * We use the VFS directory iterator (i.e. readdir) to call this
  99  * function for every directory entry in a directory.  Once we're here,
 100  * we check the inode number to make sure it's sane, then we check that
 101  * we can look up this filename.  Finally, we check the ftype.
 102  */
 103 STATIC int
 104 xchk_dir_actor(
 105         struct dir_context      *dir_iter,
 106         const char              *name,
 107         int                     namelen,
 108         loff_t                  pos,
 109         u64                     ino,
 110         unsigned                type)
 111 {
 112         struct xfs_mount        *mp;
 113         struct xfs_inode        *ip;
 114         struct xchk_dir_ctx     *sdc;
 115         struct xfs_name         xname;
 116         xfs_ino_t               lookup_ino;
 117         xfs_dablk_t             offset;
 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         /* Does this inode number make sense? */
 127         if (!xfs_verify_dir_ino(mp, ino)) {
 128                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 129                 goto out;
 130         }
 131
 132         if (!strncmp(".", name, namelen)) {
 133                 /* If this is "." then check that the inum matches the dir. */
 134                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 135                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 136                                         offset);
 137                 if (ino != ip->i_ino)
 138                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 139                                         offset);
 140         } else if (!strncmp("..", name, namelen)) {
 141                 /*
 142                  * If this is ".." in the root inode, check that the inum
 143                  * matches this dir.
 144                  */
 145                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 146                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 147                                         offset);
 148                 if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
 149                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 150                                         offset);
 151         }
 152
 153         /* Verify that we can look up this name by hash. */
 154         xname.name = name;
 155         xname.len = namelen;
 156         xname.type = XFS_DIR3_FT_UNKNOWN;
 157
 158         error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
 159         /* ENOENT means the hash lookup failed and the dir is corrupt */
 160         if (error == -ENOENT)
 161                 error = -EFSCORRUPTED;
 162         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 163                         &error))
 164                 goto out;
 165         if (lookup_ino != ino) {
 166                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 167                 goto out;
 168         }
 169
 170         /* Verify the file type.  This function absorbs error codes. */
 171         error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 172         if (error)
 173                 goto out;
 174 out:
 175         /*
 176          * A negative error code returned here is supposed to cause the
 177          * dir_emit caller (xfs_readdir) to abort the directory iteration
 178          * and return zero to xchk_directory.
 179          */
 180         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 181                 return -EFSCORRUPTED;
 182         return error;
 183 }
 184
 185 /* Scrub a directory btree record. */
 186 STATIC int
 187 xchk_dir_rec(
 188         struct xchk_da_btree            *ds,
 189         int                             level,
 190         void                            *rec)
 191 {
 192         struct xfs_mount                *mp = ds->state->mp;
 193         struct xfs_dir2_leaf_entry      *ent = rec;
 194         struct xfs_inode                *dp = ds->dargs.dp;
 195         struct xfs_dir2_data_entry      *dent;
 196         struct xfs_buf                  *bp;
 197         char                            *p, *endp;
 198         xfs_ino_t                       ino;
 199         xfs_dablk_t                     rec_bno;
 200         xfs_dir2_db_t                   db;
 201         xfs_dir2_data_aoff_t            off;
 202         xfs_dir2_dataptr_t              ptr;
 203         xfs_dahash_t                    calc_hash;
 204         xfs_dahash_t                    hash;
 205         unsigned int                    tag;
 206         int                             error;
 207
 208         /* Check the hash of the entry. */
 209         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 210         if (error)
 211                 goto out;
 212
 213         /* Valid hash pointer? */
 214         ptr = be32_to_cpu(ent->address);
 215         if (ptr == 0)
 216                 return 0;
 217
 218         /* Find the directory entry's location. */
 219         db = xfs_dir2_dataptr_to_db(mp->m_dir_geo, ptr);
 220         off = xfs_dir2_dataptr_to_off(mp->m_dir_geo, ptr);
 221         rec_bno = xfs_dir2_db_to_da(mp->m_dir_geo, db);
 222
 223         if (rec_bno >= mp->m_dir_geo->leafblk) {
 224                 xchk_da_set_corrupt(ds, level);
 225                 goto out;
 226         }
 227         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
 228         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 229                         &error))
 230                 goto out;
 231         if (!bp) {
 232                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 233                 goto out;
 234         }
 235         xchk_buffer_recheck(ds->sc, bp);
 236
 237         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 238                 goto out_relse;
 239
 240         dent = (struct xfs_dir2_data_entry *)(((char *)bp->b_addr) + off);
 241
 242         /* Make sure we got a real directory entry. */
 243         p = (char *)mp->m_dir_inode_ops->data_entry_p(bp->b_addr);
 244         endp = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 245         if (!endp) {
 246                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 247                 goto out_relse;
 248         }
 249         while (p < endp) {
 250                 struct xfs_dir2_data_entry      *dep;
 251                 struct xfs_dir2_data_unused     *dup;
 252
 253                 dup = (struct xfs_dir2_data_unused *)p;
 254                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 255                         p += be16_to_cpu(dup->length);
 256                         continue;
 257                 }
 258                 dep = (struct xfs_dir2_data_entry *)p;
 259                 if (dep == dent)
 260                         break;
 261                 p += mp->m_dir_inode_ops->data_entsize(dep->namelen);
 262         }
 263         if (p >= endp) {
 264                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 265                 goto out_relse;
 266         }
 267
 268         /* Retrieve the entry, sanity check it, and compare hashes. */
 269         ino = be64_to_cpu(dent->inumber);
 270         hash = be32_to_cpu(ent->hashval);
 271         tag = be16_to_cpup(dp->d_ops->data_entry_tag_p(dent));
 272         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 273                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 274         if (dent->namelen == 0) {
 275                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 276                 goto out_relse;
 277         }
 278         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 279         if (calc_hash != hash)
 280                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 281
 282 out_relse:
 283         xfs_trans_brelse(ds->dargs.trans, bp);
 284 out:
 285         return error;
 286 }
 287
 288 /*
 289  * Is this unused entry either in the bestfree or smaller than all of
 290  * them?  We've already checked that the bestfrees are sorted longest to
 291  * shortest, and that there aren't any bogus entries.
 292  */
 293 STATIC void
 294 xchk_directory_check_free_entry(
 295         struct xfs_scrub                *sc,
 296         xfs_dablk_t                     lblk,
 297         struct xfs_dir2_data_free       *bf,
 298         struct xfs_dir2_data_unused     *dup)
 299 {
 300         struct xfs_dir2_data_free       *dfp;
 301         unsigned int                    dup_length;
 302
 303         dup_length = be16_to_cpu(dup->length);
 304
 305         /* Unused entry is shorter than any of the bestfrees */
 306         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 307                 return;
 308
 309         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 310                 if (dup_length == be16_to_cpu(dfp->length))
 311                         return;
 312
 313         /* Unused entry should be in the bestfrees but wasn't found. */
 314         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 315 }
 316
 317 /* Check free space info in a directory data block. */
 318 STATIC int
 319 xchk_directory_data_bestfree(
 320         struct xfs_scrub                *sc,
 321         xfs_dablk_t                     lblk,
 322         bool                            is_block)
 323 {
 324         struct xfs_dir2_data_unused     *dup;
 325         struct xfs_dir2_data_free       *dfp;
 326         struct xfs_buf                  *bp;
 327         struct xfs_dir2_data_free       *bf;
 328         struct xfs_mount                *mp = sc->mp;
 329         const struct xfs_dir_ops        *d_ops;
 330         char                            *ptr;
 331         char                            *endptr;
 332         u16                             tag;
 333         unsigned int                    nr_bestfrees = 0;
 334         unsigned int                    nr_frees = 0;
 335         unsigned int                    smallest_bestfree;
 336         int                             newlen;
 337         int                             offset;
 338         int                             error;
 339
 340         d_ops = sc->ip->d_ops;
 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 = d_ops->data_bestfree_p(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 = (struct xfs_dir2_data_unused *)(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 != ((char *)dup - (char *)bp->b_addr)) {
 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         ptr = (char *)d_ops->data_entry_p(bp->b_addr);
 394         endptr = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 395
 396         /* Iterate the entries, stopping when we hit or go past the end. */
 397         while (ptr < endptr) {
 398                 dup = (struct xfs_dir2_data_unused *)ptr;
 399                 /* Skip real entries */
 400                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 401                         struct xfs_dir2_data_entry      *dep;
 402
 403                         dep = (struct xfs_dir2_data_entry *)ptr;
 404                         newlen = d_ops->data_entsize(dep->namelen);
 405                         if (newlen <= 0) {
 406                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 407                                                 lblk);
 408                                 goto out_buf;
 409                         }
 410                         ptr += 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 != ((char *)dup - (char *)bp->b_addr)) {
 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                 ptr += newlen;
 436                 if (ptr <= endptr)
 437                         nr_frees++;
 438         }
 439
 440         /* We're required to fill all the space. */
 441         if (ptr != endptr)
 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 = sc->ip->d_ops->data_bestfree_p(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_entry      *ents;
 486         struct xfs_dir2_leaf_tail       *ltp;
 487         struct xfs_dir2_leaf            *leaf;
 488         struct xfs_buf                  *dbp;
 489         struct xfs_buf                  *bp;
 490         const struct xfs_dir_ops        *d_ops = sc->ip->d_ops;
 491         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 492         __be16                          *bestp;
 493         __u16                           best;
 494         __u32                           hash;
 495         __u32                           lasthash = 0;
 496         __u32                           bestcount;
 497         unsigned int                    stale = 0;
 498         int                             i;
 499         int                             error;
 500
 501         /* Read the free space block. */
 502         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, -1, &bp);
 503         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 504                 goto out;
 505         xchk_buffer_recheck(sc, bp);
 506
 507         leaf = bp->b_addr;
 508         d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 509         ents = d_ops->leaf_ents_p(leaf);
 510         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 511         bestcount = be32_to_cpu(ltp->bestcount);
 512         bestp = xfs_dir2_leaf_bests_p(ltp);
 513
 514         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 515                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 516
 517                 if (hdr3->pad != cpu_to_be32(0))
 518                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 519         }
 520
 521         /*
 522          * There should be as many bestfree slots as there are dir data
 523          * blocks that can fit under i_size.
 524          */
 525         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 526                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 527                 goto out;
 528         }
 529
 530         /* Is the leaf count even remotely sane? */
 531         if (leafhdr.count > d_ops->leaf_max_ents(geo)) {
 532                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 533                 goto out;
 534         }
 535
 536         /* Leaves and bests don't overlap in leaf format. */
 537         if ((char *)&ents[leafhdr.count] > (char *)bestp) {
 538                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 539                 goto out;
 540         }
 541
 542         /* Check hash value order, count stale entries.  */
 543         for (i = 0; i < leafhdr.count; i++) {
 544                 hash = be32_to_cpu(ents[i].hashval);
 545                 if (i > 0 && lasthash > hash)
 546                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 547                 lasthash = hash;
 548                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 549                         stale++;
 550         }
 551         if (leafhdr.stale != stale)
 552                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 553         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 554                 goto out;
 555
 556         /* Check all the bestfree entries. */
 557         for (i = 0; i < bestcount; i++, bestp++) {
 558                 best = be16_to_cpu(*bestp);
 559                 if (best == NULLDATAOFF)
 560                         continue;
 561                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 562                                 i * args->geo->fsbcount, -1, &dbp);
 563                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 564                                 &error))
 565                         break;
 566                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 567                 xfs_trans_brelse(sc->tp, dbp);
 568                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 569                         goto out;
 570         }
 571 out:
 572         return error;
 573 }
 574
 575 /* Check free space info in a directory freespace block. */
 576 STATIC int
 577 xchk_directory_free_bestfree(
 578         struct xfs_scrub                *sc,
 579         struct xfs_da_args              *args,
 580         xfs_dablk_t                     lblk)
 581 {
 582         struct xfs_dir3_icfree_hdr      freehdr;
 583         struct xfs_buf                  *dbp;
 584         struct xfs_buf                  *bp;
 585         __be16                          *bestp;
 586         __u16                           best;
 587         unsigned int                    stale = 0;
 588         int                             i;
 589         int                             error;
 590
 591         /* Read the free space block */
 592         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 593         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 594                 goto out;
 595         xchk_buffer_recheck(sc, bp);
 596
 597         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 598                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 599
 600                 if (hdr3->pad != cpu_to_be32(0))
 601                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 602         }
 603
 604         /* Check all the entries. */
 605         sc->ip->d_ops->free_hdr_from_disk(&freehdr, bp->b_addr);
 606         bestp = sc->ip->d_ops->free_bests_p(bp->b_addr);
 607         for (i = 0; i < freehdr.nvalid; i++, bestp++) {
 608                 best = be16_to_cpu(*bestp);
 609                 if (best == NULLDATAOFF) {
 610                         stale++;
 611                         continue;
 612                 }
 613                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 614                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 615                                 -1, &dbp);
 616                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 617                                 &error))
 618                         break;
 619                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 620                 xfs_trans_brelse(sc->tp, dbp);
 621         }
 622
 623         if (freehdr.nused + stale != freehdr.nvalid)
 624                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 625 out:
 626         return error;
 627 }
 628
 629 /* Check free space information in directories. */
 630 STATIC int
 631 xchk_directory_blocks(
 632         struct xfs_scrub        *sc)
 633 {
 634         struct xfs_bmbt_irec    got;
 635         struct xfs_da_args      args;
 636         struct xfs_ifork        *ifp;
 637         struct xfs_mount        *mp = sc->mp;
 638         xfs_fileoff_t           leaf_lblk;
 639         xfs_fileoff_t           free_lblk;
 640         xfs_fileoff_t           lblk;
 641         struct xfs_iext_cursor  icur;
 642         xfs_dablk_t             dabno;
 643         bool                    found;
 644         int                     is_block = 0;
 645         int                     error;
 646
 647         /* Ignore local format directories. */
 648         if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 649             sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
 650                 return 0;
 651
 652         ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 653         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 654         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 655         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 656
 657         /* Is this a block dir? */
 658         args.dp = sc->ip;
 659         args.geo = mp->m_dir_geo;
 660         args.trans = sc->tp;
 661         error = xfs_dir2_isblock(&args, &is_block);
 662         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 663                 goto out;
 664
 665         /* Iterate all the data extents in the directory... */
 666         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 667         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 668                 /* Block directories only have a single block at offset 0. */
 669                 if (is_block &&
 670                     (got.br_startoff > 0 ||
 671                      got.br_blockcount != args.geo->fsbcount)) {
 672                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 673                                         got.br_startoff);
 674                         break;
 675                 }
 676
 677                 /* No more data blocks... */
 678                 if (got.br_startoff >= leaf_lblk)
 679                         break;
 680
 681                 /*
 682                  * Check each data block's bestfree data.
 683                  *
 684                  * Iterate all the fsbcount-aligned block offsets in
 685                  * this directory.  The directory block reading code is
 686                  * smart enough to do its own bmap lookups to handle
 687                  * discontiguous directory blocks.  When we're done
 688                  * with the extent record, re-query the bmap at the
 689                  * next fsbcount-aligned offset to avoid redundant
 690                  * block checks.
 691                  */
 692                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 693                                 args.geo->fsbcount);
 694                      lblk < got.br_startoff + got.br_blockcount;
 695                      lblk += args.geo->fsbcount) {
 696                         error = xchk_directory_data_bestfree(sc, lblk,
 697                                         is_block);
 698                         if (error)
 699                                 goto out;
 700                 }
 701                 dabno = got.br_startoff + got.br_blockcount;
 702                 lblk = roundup(dabno, args.geo->fsbcount);
 703                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 704         }
 705
 706         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 707                 goto out;
 708
 709         /* Look for a leaf1 block, which has free info. */
 710         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 711             got.br_startoff == leaf_lblk &&
 712             got.br_blockcount == args.geo->fsbcount &&
 713             !xfs_iext_next_extent(ifp, &icur, &got)) {
 714                 if (is_block) {
 715                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 716                         goto out;
 717                 }
 718                 error = xchk_directory_leaf1_bestfree(sc, &args,
 719                                 leaf_lblk);
 720                 if (error)
 721                         goto out;
 722         }
 723
 724         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 725                 goto out;
 726
 727         /* Scan for free blocks */
 728         lblk = free_lblk;
 729         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 730         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 731                 /*
 732                  * Dirs can't have blocks mapped above 2^32.
 733                  * Single-block dirs shouldn't even be here.
 734                  */
 735                 lblk = got.br_startoff;
 736                 if (lblk & ~0xFFFFFFFFULL) {
 737                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 738                         goto out;
 739                 }
 740                 if (is_block) {
 741                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 742                         goto out;
 743                 }
 744
 745                 /*
 746                  * Check each dir free block's bestfree data.
 747                  *
 748                  * Iterate all the fsbcount-aligned block offsets in
 749                  * this directory.  The directory block reading code is
 750                  * smart enough to do its own bmap lookups to handle
 751                  * discontiguous directory blocks.  When we're done
 752                  * with the extent record, re-query the bmap at the
 753                  * next fsbcount-aligned offset to avoid redundant
 754                  * block checks.
 755                  */
 756                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 757                                 args.geo->fsbcount);
 758                      lblk < got.br_startoff + got.br_blockcount;
 759                      lblk += args.geo->fsbcount) {
 760                         error = xchk_directory_free_bestfree(sc, &args,
 761                                         lblk);
 762                         if (error)
 763                                 goto out;
 764                 }
 765                 dabno = got.br_startoff + got.br_blockcount;
 766                 lblk = roundup(dabno, args.geo->fsbcount);
 767                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 768         }
 769 out:
 770         return error;
 771 }
 772
 773 /* Scrub a whole directory. */
 774 int
 775 xchk_directory(
 776         struct xfs_scrub        *sc)
 777 {
 778         struct xchk_dir_ctx     sdc = {
 779                 .dir_iter.actor = xchk_dir_actor,
 780                 .dir_iter.pos = 0,
 781                 .sc = sc,
 782         };
 783         size_t                  bufsize;
 784         loff_t                  oldpos;
 785         int                     error = 0;
 786
 787         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 788                 return -ENOENT;
 789
 790         /* Plausible size? */
 791         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 792                 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
 793                 goto out;
 794         }
 795
 796         /* Check directory tree structure */
 797         error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
 798         if (error)
 799                 return error;
 800
 801         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 802                 return error;
 803
 804         /* Check the freespace. */
 805         error = xchk_directory_blocks(sc);
 806         if (error)
 807                 return error;
 808
 809         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 810                 return error;
 811
 812         /*
 813          * Check that every dirent we see can also be looked up by hash.
 814          * Userspace usually asks for a 32k buffer, so we will too.
 815          */
 816         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 817                         sc->ip->i_d.di_size);
 818
 819         /*
 820          * Look up every name in this directory by hash.
 821          *
 822          * Use the xfs_readdir function to call xchk_dir_actor on
 823          * every directory entry in this directory.  In _actor, we check
 824          * the name, inode number, and ftype (if applicable) of the
 825          * entry.  xfs_readdir uses the VFS filldir functions to provide
 826          * iteration context.
 827          *
 828          * The VFS grabs a read or write lock via i_rwsem before it reads
 829          * or writes to a directory.  If we've gotten this far we've
 830          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 831          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 832          * to drop the ILOCK here in order to reuse the _readdir and
 833          * _dir_lookup routines, which do their own ILOCK locking.
 834          */
 835         oldpos = 0;
 836         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 837         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 838         while (true) {
 839                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 840                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
 841                                 &error))
 842                         goto out;
 843                 if (oldpos == sdc.dir_iter.pos)
 844                         break;
 845                 oldpos = sdc.dir_iter.pos;
 846         }
 847
 848 out:
 849         return error;
 850 }