fs/xfs/scrub/dir.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_trans_resv.h"
  11 #include "xfs_mount.h"
  12 #include "xfs_log_format.h"
  13 #include "xfs_trans.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_icache.h"
  16 #include "xfs_dir2.h"
  17 #include "xfs_dir2_priv.h"
  18 #include "scrub/scrub.h"
  19 #include "scrub/common.h"
  20 #include "scrub/dabtree.h"
  21
  22 /* Set us up to scrub directories. */
  23 int
  24 xchk_setup_directory(
  25         struct xfs_scrub        *sc,
  26         struct xfs_inode        *ip)
  27 {
  28         return xchk_setup_inode_contents(sc, ip, 0);
  29 }
  30
  31 /* Directories */
  32
  33 /* Scrub a directory entry. */
  34
  35 struct xchk_dir_ctx {
  36         /* VFS fill-directory iterator */
  37         struct dir_context      dir_iter;
  38
  39         struct xfs_scrub        *sc;
  40 };
  41
  42 /* Check that an inode's mode matches a given DT_ type. */
  43 STATIC int
  44 xchk_dir_check_ftype(
  45         struct xchk_dir_ctx     *sdc,
  46         xfs_fileoff_t           offset,
  47         xfs_ino_t               inum,
  48         int                     dtype)
  49 {
  50         struct xfs_mount        *mp = sdc->sc->mp;
  51         struct xfs_inode        *ip;
  52         int                     ino_dtype;
  53         int                     error = 0;
  54
  55         if (!xfs_sb_version_hasftype(&mp->m_sb)) {
  56                 if (dtype != DT_UNKNOWN && dtype != DT_DIR)
  57                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
  58                                         offset);
  59                 goto out;
  60         }
  61
  62         /*
  63          * Grab the inode pointed to by the dirent.  We release the
  64          * inode before we cancel the scrub transaction.  Since we're
  65          * don't know a priori that releasing the inode won't trigger
  66          * eofblocks cleanup (which allocates what would be a nested
  67          * transaction), we can't use DONTCACHE here because DONTCACHE
  68          * inodes can trigger immediate inactive cleanup of the inode.
  69          *
  70          * If _iget returns -EINVAL or -ENOENT then the child inode number is
  71          * garbage and the directory is corrupt.  If the _iget returns
  72          * -EFSCORRUPTED or -EFSBADCRC then the child is corrupt which is a
  73          *  cross referencing error.  Any other error is an operational error.
  74          */
  75         error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
  76         if (error == -EINVAL || error == -ENOENT) {
  77                 error = -EFSCORRUPTED;
  78                 xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, 0, &error);
  79                 goto out;
  80         }
  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         bool                    checked_ftype = false;
 119         int                     error = 0;
 120
 121         sdc = container_of(dir_iter, struct xchk_dir_ctx, dir_iter);
 122         ip = sdc->sc->ip;
 123         mp = ip->i_mount;
 124         offset = xfs_dir2_db_to_da(mp->m_dir_geo,
 125                         xfs_dir2_dataptr_to_db(mp->m_dir_geo, pos));
 126
 127         if (xchk_should_terminate(sdc->sc, &error))
 128                 return error;
 129
 130         /* Does this inode number make sense? */
 131         if (!xfs_verify_dir_ino(mp, ino)) {
 132                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 133                 goto out;
 134         }
 135
 136         /* Does this name make sense? */
 137         if (!xfs_dir2_namecheck(name, namelen)) {
 138                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 139                 goto out;
 140         }
 141
 142         if (!strncmp(".", name, namelen)) {
 143                 /* If this is "." then check that the inum matches the dir. */
 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                 checked_ftype = true;
 148                 if (ino != ip->i_ino)
 149                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 150                                         offset);
 151         } else if (!strncmp("..", name, namelen)) {
 152                 /*
 153                  * If this is ".." in the root inode, check that the inum
 154                  * matches this dir.
 155                  */
 156                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 157                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 158                                         offset);
 159                 checked_ftype = true;
 160                 if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
 161                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 162                                         offset);
 163         }
 164
 165         /* Verify that we can look up this name by hash. */
 166         xname.name = name;
 167         xname.len = namelen;
 168         xname.type = XFS_DIR3_FT_UNKNOWN;
 169
 170         error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
 171         /* ENOENT means the hash lookup failed and the dir is corrupt */
 172         if (error == -ENOENT)
 173                 error = -EFSCORRUPTED;
 174         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 175                         &error))
 176                 goto out;
 177         if (lookup_ino != ino) {
 178                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 179                 goto out;
 180         }
 181
 182         /* Verify the file type.  This function absorbs error codes. */
 183         if (!checked_ftype) {
 184                 error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 185                 if (error)
 186                         goto out;
 187         }
 188 out:
 189         /*
 190          * A negative error code returned here is supposed to cause the
 191          * dir_emit caller (xfs_readdir) to abort the directory iteration
 192          * and return zero to xchk_directory.
 193          */
 194         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 195                 return -EFSCORRUPTED;
 196         return error;
 197 }
 198
 199 /* Scrub a directory btree record. */
 200 STATIC int
 201 xchk_dir_rec(
 202         struct xchk_da_btree            *ds,
 203         int                             level)
 204 {
 205         struct xfs_da_state_blk         *blk = &ds->state->path.blk[level];
 206         struct xfs_mount                *mp = ds->state->mp;
 207         struct xfs_inode                *dp = ds->dargs.dp;
 208         struct xfs_da_geometry          *geo = mp->m_dir_geo;
 209         struct xfs_dir2_data_entry      *dent;
 210         struct xfs_buf                  *bp;
 211         struct xfs_dir2_leaf_entry      *ent;
 212         unsigned int                    end;
 213         unsigned int                    iter_off;
 214         xfs_ino_t                       ino;
 215         xfs_dablk_t                     rec_bno;
 216         xfs_dir2_db_t                   db;
 217         xfs_dir2_data_aoff_t            off;
 218         xfs_dir2_dataptr_t              ptr;
 219         xfs_dahash_t                    calc_hash;
 220         xfs_dahash_t                    hash;
 221         struct xfs_dir3_icleaf_hdr      hdr;
 222         unsigned int                    tag;
 223         int                             error;
 224
 225         ASSERT(blk->magic == XFS_DIR2_LEAF1_MAGIC ||
 226                blk->magic == XFS_DIR2_LEAFN_MAGIC);
 227
 228         xfs_dir2_leaf_hdr_from_disk(mp, &hdr, blk->bp->b_addr);
 229         ent = hdr.ents + blk->index;
 230
 231         /* Check the hash of the entry. */
 232         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 233         if (error)
 234                 goto out;
 235
 236         /* Valid hash pointer? */
 237         ptr = be32_to_cpu(ent->address);
 238         if (ptr == 0)
 239                 return 0;
 240
 241         /* Find the directory entry's location. */
 242         db = xfs_dir2_dataptr_to_db(geo, ptr);
 243         off = xfs_dir2_dataptr_to_off(geo, ptr);
 244         rec_bno = xfs_dir2_db_to_da(geo, db);
 245
 246         if (rec_bno >= geo->leafblk) {
 247                 xchk_da_set_corrupt(ds, level);
 248                 goto out;
 249         }
 250         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno,
 251                         XFS_DABUF_MAP_HOLE_OK, &bp);
 252         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 253                         &error))
 254                 goto out;
 255         if (!bp) {
 256                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 257                 goto out;
 258         }
 259         xchk_buffer_recheck(ds->sc, bp);
 260
 261         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 262                 goto out_relse;
 263
 264         dent = bp->b_addr + off;
 265
 266         /* Make sure we got a real directory entry. */
 267         iter_off = geo->data_entry_offset;
 268         end = xfs_dir3_data_end_offset(geo, bp->b_addr);
 269         if (!end) {
 270                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 271                 goto out_relse;
 272         }
 273         for (;;) {
 274                 struct xfs_dir2_data_entry      *dep = bp->b_addr + iter_off;
 275                 struct xfs_dir2_data_unused     *dup = bp->b_addr + iter_off;
 276
 277                 if (iter_off >= end) {
 278                         xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 279                         goto out_relse;
 280                 }
 281
 282                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 283                         iter_off += be16_to_cpu(dup->length);
 284                         continue;
 285                 }
 286                 if (dep == dent)
 287                         break;
 288                 iter_off += xfs_dir2_data_entsize(mp, dep->namelen);
 289         }
 290
 291         /* Retrieve the entry, sanity check it, and compare hashes. */
 292         ino = be64_to_cpu(dent->inumber);
 293         hash = be32_to_cpu(ent->hashval);
 294         tag = be16_to_cpup(xfs_dir2_data_entry_tag_p(mp, dent));
 295         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 296                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 297         if (dent->namelen == 0) {
 298                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 299                 goto out_relse;
 300         }
 301         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 302         if (calc_hash != hash)
 303                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 304
 305 out_relse:
 306         xfs_trans_brelse(ds->dargs.trans, bp);
 307 out:
 308         return error;
 309 }
 310
 311 /*
 312  * Is this unused entry either in the bestfree or smaller than all of
 313  * them?  We've already checked that the bestfrees are sorted longest to
 314  * shortest, and that there aren't any bogus entries.
 315  */
 316 STATIC void
 317 xchk_directory_check_free_entry(
 318         struct xfs_scrub                *sc,
 319         xfs_dablk_t                     lblk,
 320         struct xfs_dir2_data_free       *bf,
 321         struct xfs_dir2_data_unused     *dup)
 322 {
 323         struct xfs_dir2_data_free       *dfp;
 324         unsigned int                    dup_length;
 325
 326         dup_length = be16_to_cpu(dup->length);
 327
 328         /* Unused entry is shorter than any of the bestfrees */
 329         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 330                 return;
 331
 332         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 333                 if (dup_length == be16_to_cpu(dfp->length))
 334                         return;
 335
 336         /* Unused entry should be in the bestfrees but wasn't found. */
 337         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 338 }
 339
 340 /* Check free space info in a directory data block. */
 341 STATIC int
 342 xchk_directory_data_bestfree(
 343         struct xfs_scrub                *sc,
 344         xfs_dablk_t                     lblk,
 345         bool                            is_block)
 346 {
 347         struct xfs_dir2_data_unused     *dup;
 348         struct xfs_dir2_data_free       *dfp;
 349         struct xfs_buf                  *bp;
 350         struct xfs_dir2_data_free       *bf;
 351         struct xfs_mount                *mp = sc->mp;
 352         u16                             tag;
 353         unsigned int                    nr_bestfrees = 0;
 354         unsigned int                    nr_frees = 0;
 355         unsigned int                    smallest_bestfree;
 356         int                             newlen;
 357         unsigned int                    offset;
 358         unsigned int                    end;
 359         int                             error;
 360
 361         if (is_block) {
 362                 /* dir block format */
 363                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 364                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 365                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 366         } else {
 367                 /* dir data format */
 368                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, 0, &bp);
 369         }
 370         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 371                 goto out;
 372         xchk_buffer_recheck(sc, bp);
 373
 374         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 375
 376         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 377                 goto out_buf;
 378
 379         /* Do the bestfrees correspond to actual free space? */
 380         bf = xfs_dir2_data_bestfree_p(mp, bp->b_addr);
 381         smallest_bestfree = UINT_MAX;
 382         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 383                 offset = be16_to_cpu(dfp->offset);
 384                 if (offset == 0)
 385                         continue;
 386                 if (offset >= mp->m_dir_geo->blksize) {
 387                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 388                         goto out_buf;
 389                 }
 390                 dup = bp->b_addr + offset;
 391                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 392
 393                 /* bestfree doesn't match the entry it points at? */
 394                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 395                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 396                     tag != offset) {
 397                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 398                         goto out_buf;
 399                 }
 400
 401                 /* bestfree records should be ordered largest to smallest */
 402                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 403                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 404                         goto out_buf;
 405                 }
 406
 407                 smallest_bestfree = be16_to_cpu(dfp->length);
 408                 nr_bestfrees++;
 409         }
 410
 411         /* Make sure the bestfrees are actually the best free spaces. */
 412         offset = mp->m_dir_geo->data_entry_offset;
 413         end = xfs_dir3_data_end_offset(mp->m_dir_geo, bp->b_addr);
 414
 415         /* Iterate the entries, stopping when we hit or go past the end. */
 416         while (offset < end) {
 417                 dup = bp->b_addr + offset;
 418
 419                 /* Skip real entries */
 420                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 421                         struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
 422
 423                         newlen = xfs_dir2_data_entsize(mp, dep->namelen);
 424                         if (newlen <= 0) {
 425                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 426                                                 lblk);
 427                                 goto out_buf;
 428                         }
 429                         offset += newlen;
 430                         continue;
 431                 }
 432
 433                 /* Spot check this free entry */
 434                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 435                 if (tag != offset) {
 436                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 437                         goto out_buf;
 438                 }
 439
 440                 /*
 441                  * Either this entry is a bestfree or it's smaller than
 442                  * any of the bestfrees.
 443                  */
 444                 xchk_directory_check_free_entry(sc, lblk, bf, dup);
 445                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 446                         goto out_buf;
 447
 448                 /* Move on. */
 449                 newlen = be16_to_cpu(dup->length);
 450                 if (newlen <= 0) {
 451                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 452                         goto out_buf;
 453                 }
 454                 offset += newlen;
 455                 if (offset <= end)
 456                         nr_frees++;
 457         }
 458
 459         /* We're required to fill all the space. */
 460         if (offset != end)
 461                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 462
 463         /* Did we see at least as many free slots as there are bestfrees? */
 464         if (nr_frees < nr_bestfrees)
 465                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 466 out_buf:
 467         xfs_trans_brelse(sc->tp, bp);
 468 out:
 469         return error;
 470 }
 471
 472 /*
 473  * Does the free space length in the free space index block ($len) match
 474  * the longest length in the directory data block's bestfree array?
 475  * Assume that we've already checked that the data block's bestfree
 476  * array is in order.
 477  */
 478 STATIC void
 479 xchk_directory_check_freesp(
 480         struct xfs_scrub                *sc,
 481         xfs_dablk_t                     lblk,
 482         struct xfs_buf                  *dbp,
 483         unsigned int                    len)
 484 {
 485         struct xfs_dir2_data_free       *dfp;
 486
 487         dfp = xfs_dir2_data_bestfree_p(sc->mp, dbp->b_addr);
 488
 489         if (len != be16_to_cpu(dfp->length))
 490                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 491
 492         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 493                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 494 }
 495
 496 /* Check free space info in a directory leaf1 block. */
 497 STATIC int
 498 xchk_directory_leaf1_bestfree(
 499         struct xfs_scrub                *sc,
 500         struct xfs_da_args              *args,
 501         xfs_dablk_t                     lblk)
 502 {
 503         struct xfs_dir3_icleaf_hdr      leafhdr;
 504         struct xfs_dir2_leaf_tail       *ltp;
 505         struct xfs_dir2_leaf            *leaf;
 506         struct xfs_buf                  *dbp;
 507         struct xfs_buf                  *bp;
 508         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 509         __be16                          *bestp;
 510         __u16                           best;
 511         __u32                           hash;
 512         __u32                           lasthash = 0;
 513         __u32                           bestcount;
 514         unsigned int                    stale = 0;
 515         int                             i;
 516         int                             error;
 517
 518         /* Read the free space block. */
 519         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, &bp);
 520         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 521                 return error;
 522         xchk_buffer_recheck(sc, bp);
 523
 524         leaf = bp->b_addr;
 525         xfs_dir2_leaf_hdr_from_disk(sc->ip->i_mount, &leafhdr, leaf);
 526         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 527         bestcount = be32_to_cpu(ltp->bestcount);
 528         bestp = xfs_dir2_leaf_bests_p(ltp);
 529
 530         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 531                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 532
 533                 if (hdr3->pad != cpu_to_be32(0))
 534                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 535         }
 536
 537         /*
 538          * There should be as many bestfree slots as there are dir data
 539          * blocks that can fit under i_size.
 540          */
 541         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 542                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 543                 goto out;
 544         }
 545
 546         /* Is the leaf count even remotely sane? */
 547         if (leafhdr.count > geo->leaf_max_ents) {
 548                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 549                 goto out;
 550         }
 551
 552         /* Leaves and bests don't overlap in leaf format. */
 553         if ((char *)&leafhdr.ents[leafhdr.count] > (char *)bestp) {
 554                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 555                 goto out;
 556         }
 557
 558         /* Check hash value order, count stale entries.  */
 559         for (i = 0; i < leafhdr.count; i++) {
 560                 hash = be32_to_cpu(leafhdr.ents[i].hashval);
 561                 if (i > 0 && lasthash > hash)
 562                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 563                 lasthash = hash;
 564                 if (leafhdr.ents[i].address ==
 565                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 566                         stale++;
 567         }
 568         if (leafhdr.stale != stale)
 569                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 570         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 571                 goto out;
 572
 573         /* Check all the bestfree entries. */
 574         for (i = 0; i < bestcount; i++, bestp++) {
 575                 best = be16_to_cpu(*bestp);
 576                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 577                                 xfs_dir2_db_to_da(args->geo, i),
 578                                 XFS_DABUF_MAP_HOLE_OK,
 579                                 &dbp);
 580                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 581                                 &error))
 582                         break;
 583
 584                 if (!dbp) {
 585                         if (best != NULLDATAOFF) {
 586                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 587                                                 lblk);
 588                                 break;
 589                         }
 590                         continue;
 591                 }
 592
 593                 if (best == NULLDATAOFF)
 594                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 595                 else
 596                         xchk_directory_check_freesp(sc, lblk, dbp, best);
 597                 xfs_trans_brelse(sc->tp, dbp);
 598                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 599                         break;
 600         }
 601 out:
 602         xfs_trans_brelse(sc->tp, bp);
 603         return error;
 604 }
 605
 606 /* Check free space info in a directory freespace block. */
 607 STATIC int
 608 xchk_directory_free_bestfree(
 609         struct xfs_scrub                *sc,
 610         struct xfs_da_args              *args,
 611         xfs_dablk_t                     lblk)
 612 {
 613         struct xfs_dir3_icfree_hdr      freehdr;
 614         struct xfs_buf                  *dbp;
 615         struct xfs_buf                  *bp;
 616         __u16                           best;
 617         unsigned int                    stale = 0;
 618         int                             i;
 619         int                             error;
 620
 621         /* Read the free space block */
 622         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 623         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 624                 return error;
 625         xchk_buffer_recheck(sc, bp);
 626
 627         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 628                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 629
 630                 if (hdr3->pad != cpu_to_be32(0))
 631                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 632         }
 633
 634         /* Check all the entries. */
 635         xfs_dir2_free_hdr_from_disk(sc->ip->i_mount, &freehdr, bp->b_addr);
 636         for (i = 0; i < freehdr.nvalid; i++) {
 637                 best = be16_to_cpu(freehdr.bests[i]);
 638                 if (best == NULLDATAOFF) {
 639                         stale++;
 640                         continue;
 641                 }
 642                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 643                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 644                                 0, &dbp);
 645                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 646                                 &error))
 647                         goto out;
 648                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 649                 xfs_trans_brelse(sc->tp, dbp);
 650         }
 651
 652         if (freehdr.nused + stale != freehdr.nvalid)
 653                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 654 out:
 655         xfs_trans_brelse(sc->tp, bp);
 656         return error;
 657 }
 658
 659 /* Check free space information in directories. */
 660 STATIC int
 661 xchk_directory_blocks(
 662         struct xfs_scrub        *sc)
 663 {
 664         struct xfs_bmbt_irec    got;
 665         struct xfs_da_args      args;
 666         struct xfs_ifork        *ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 667         struct xfs_mount        *mp = sc->mp;
 668         xfs_fileoff_t           leaf_lblk;
 669         xfs_fileoff_t           free_lblk;
 670         xfs_fileoff_t           lblk;
 671         struct xfs_iext_cursor  icur;
 672         xfs_dablk_t             dabno;
 673         bool                    found;
 674         int                     is_block = 0;
 675         int                     error;
 676
 677         /* Ignore local format directories. */
 678         if (ifp->if_format != XFS_DINODE_FMT_EXTENTS &&
 679             ifp->if_format != XFS_DINODE_FMT_BTREE)
 680                 return 0;
 681
 682         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 683         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 684         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 685
 686         /* Is this a block dir? */
 687         args.dp = sc->ip;
 688         args.geo = mp->m_dir_geo;
 689         args.trans = sc->tp;
 690         error = xfs_dir2_isblock(&args, &is_block);
 691         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 692                 goto out;
 693
 694         /* Iterate all the data extents in the directory... */
 695         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 696         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 697                 /* Block directories only have a single block at offset 0. */
 698                 if (is_block &&
 699                     (got.br_startoff > 0 ||
 700                      got.br_blockcount != args.geo->fsbcount)) {
 701                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 702                                         got.br_startoff);
 703                         break;
 704                 }
 705
 706                 /* No more data blocks... */
 707                 if (got.br_startoff >= leaf_lblk)
 708                         break;
 709
 710                 /*
 711                  * Check each data block's bestfree data.
 712                  *
 713                  * Iterate all the fsbcount-aligned block offsets in
 714                  * this directory.  The directory block reading code is
 715                  * smart enough to do its own bmap lookups to handle
 716                  * discontiguous directory blocks.  When we're done
 717                  * with the extent record, re-query the bmap at the
 718                  * next fsbcount-aligned offset to avoid redundant
 719                  * block checks.
 720                  */
 721                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 722                                 args.geo->fsbcount);
 723                      lblk < got.br_startoff + got.br_blockcount;
 724                      lblk += args.geo->fsbcount) {
 725                         error = xchk_directory_data_bestfree(sc, lblk,
 726                                         is_block);
 727                         if (error)
 728                                 goto out;
 729                 }
 730                 dabno = got.br_startoff + got.br_blockcount;
 731                 lblk = roundup(dabno, args.geo->fsbcount);
 732                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 733         }
 734
 735         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 736                 goto out;
 737
 738         /* Look for a leaf1 block, which has free info. */
 739         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 740             got.br_startoff == leaf_lblk &&
 741             got.br_blockcount == args.geo->fsbcount &&
 742             !xfs_iext_next_extent(ifp, &icur, &got)) {
 743                 if (is_block) {
 744                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 745                         goto out;
 746                 }
 747                 error = xchk_directory_leaf1_bestfree(sc, &args,
 748                                 leaf_lblk);
 749                 if (error)
 750                         goto out;
 751         }
 752
 753         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 754                 goto out;
 755
 756         /* Scan for free blocks */
 757         lblk = free_lblk;
 758         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 759         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 760                 /*
 761                  * Dirs can't have blocks mapped above 2^32.
 762                  * Single-block dirs shouldn't even be here.
 763                  */
 764                 lblk = got.br_startoff;
 765                 if (lblk & ~0xFFFFFFFFULL) {
 766                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 767                         goto out;
 768                 }
 769                 if (is_block) {
 770                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 771                         goto out;
 772                 }
 773
 774                 /*
 775                  * Check each dir free block's bestfree data.
 776                  *
 777                  * Iterate all the fsbcount-aligned block offsets in
 778                  * this directory.  The directory block reading code is
 779                  * smart enough to do its own bmap lookups to handle
 780                  * discontiguous directory blocks.  When we're done
 781                  * with the extent record, re-query the bmap at the
 782                  * next fsbcount-aligned offset to avoid redundant
 783                  * block checks.
 784                  */
 785                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 786                                 args.geo->fsbcount);
 787                      lblk < got.br_startoff + got.br_blockcount;
 788                      lblk += args.geo->fsbcount) {
 789                         error = xchk_directory_free_bestfree(sc, &args,
 790                                         lblk);
 791                         if (error)
 792                                 goto out;
 793                 }
 794                 dabno = got.br_startoff + got.br_blockcount;
 795                 lblk = roundup(dabno, args.geo->fsbcount);
 796                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 797         }
 798 out:
 799         return error;
 800 }
 801
 802 /* Scrub a whole directory. */
 803 int
 804 xchk_directory(
 805         struct xfs_scrub        *sc)
 806 {
 807         struct xchk_dir_ctx     sdc = {
 808                 .dir_iter.actor = xchk_dir_actor,
 809                 .dir_iter.pos = 0,
 810                 .sc = sc,
 811         };
 812         size_t                  bufsize;
 813         loff_t                  oldpos;
 814         int                     error = 0;
 815
 816         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 817                 return -ENOENT;
 818
 819         /* Plausible size? */
 820         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 821                 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
 822                 goto out;
 823         }
 824
 825         /* Check directory tree structure */
 826         error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
 827         if (error)
 828                 return error;
 829
 830         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 831                 return error;
 832
 833         /* Check the freespace. */
 834         error = xchk_directory_blocks(sc);
 835         if (error)
 836                 return error;
 837
 838         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 839                 return error;
 840
 841         /*
 842          * Check that every dirent we see can also be looked up by hash.
 843          * Userspace usually asks for a 32k buffer, so we will too.
 844          */
 845         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 846                         sc->ip->i_d.di_size);
 847
 848         /*
 849          * Look up every name in this directory by hash.
 850          *
 851          * Use the xfs_readdir function to call xchk_dir_actor on
 852          * every directory entry in this directory.  In _actor, we check
 853          * the name, inode number, and ftype (if applicable) of the
 854          * entry.  xfs_readdir uses the VFS filldir functions to provide
 855          * iteration context.
 856          *
 857          * The VFS grabs a read or write lock via i_rwsem before it reads
 858          * or writes to a directory.  If we've gotten this far we've
 859          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 860          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 861          * to drop the ILOCK here in order to reuse the _readdir and
 862          * _dir_lookup routines, which do their own ILOCK locking.
 863          */
 864         oldpos = 0;
 865         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 866         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 867         while (true) {
 868                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 869                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
 870                                 &error))
 871                         goto out;
 872                 if (oldpos == sdc.dir_iter.pos)
 873                         break;
 874                 oldpos = sdc.dir_iter.pos;
 875         }
 876
 877 out:
 878         return error;
 879 }