fs/xfs/scrub/dir.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_trans_resv.h"
  11 #include "xfs_mount.h"
  12 #include "xfs_log_format.h"
  13 #include "xfs_trans.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_icache.h"
  16 #include "xfs_dir2.h"
  17 #include "xfs_dir2_priv.h"
  18 #include "scrub/scrub.h"
  19 #include "scrub/common.h"
  20 #include "scrub/dabtree.h"
  21
  22 /* Set us up to scrub directories. */
  23 int
  24 xchk_setup_directory(
  25         struct xfs_scrub        *sc,
  26         struct xfs_inode        *ip)
  27 {
  28         return xchk_setup_inode_contents(sc, ip, 0);
  29 }
  30
  31 /* Directories */
  32
  33 /* Scrub a directory entry. */
  34
  35 struct xchk_dir_ctx {
  36         /* VFS fill-directory iterator */
  37         struct dir_context      dir_iter;
  38
  39         struct xfs_scrub        *sc;
  40 };
  41
  42 /* Check that an inode's mode matches a given DT_ type. */
  43 STATIC int
  44 xchk_dir_check_ftype(
  45         struct xchk_dir_ctx     *sdc,
  46         xfs_fileoff_t           offset,
  47         xfs_ino_t               inum,
  48         int                     dtype)
  49 {
  50         struct xfs_mount        *mp = sdc->sc->mp;
  51         struct xfs_inode        *ip;
  52         int                     ino_dtype;
  53         int                     error = 0;
  54
  55         if (!xfs_sb_version_hasftype(&mp->m_sb)) {
  56                 if (dtype != DT_UNKNOWN && dtype != DT_DIR)
  57                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
  58                                         offset);
  59                 goto out;
  60         }
  61
  62         /*
  63          * Grab the inode pointed to by the dirent.  We release the
  64          * inode before we cancel the scrub transaction.  Since we're
  65          * don't know a priori that releasing the inode won't trigger
  66          * eofblocks cleanup (which allocates what would be a nested
  67          * transaction), we can't use DONTCACHE here because DONTCACHE
  68          * inodes can trigger immediate inactive cleanup of the inode.
  69          */
  70         error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
  71         if (!xchk_fblock_xref_process_error(sdc->sc, XFS_DATA_FORK, offset,
  72                         &error))
  73                 goto out;
  74
  75         /* Convert mode to the DT_* values that dir_emit uses. */
  76         ino_dtype = xfs_dir3_get_dtype(mp,
  77                         xfs_mode_to_ftype(VFS_I(ip)->i_mode));
  78         if (ino_dtype != dtype)
  79                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
  80         xfs_irele(ip);
  81 out:
  82         return error;
  83 }
  84
  85 /*
  86  * Scrub a single directory entry.
  87  *
  88  * We use the VFS directory iterator (i.e. readdir) to call this
  89  * function for every directory entry in a directory.  Once we're here,
  90  * we check the inode number to make sure it's sane, then we check that
  91  * we can look up this filename.  Finally, we check the ftype.
  92  */
  93 STATIC int
  94 xchk_dir_actor(
  95         struct dir_context      *dir_iter,
  96         const char              *name,
  97         int                     namelen,
  98         loff_t                  pos,
  99         u64                     ino,
 100         unsigned                type)
 101 {
 102         struct xfs_mount        *mp;
 103         struct xfs_inode        *ip;
 104         struct xchk_dir_ctx     *sdc;
 105         struct xfs_name         xname;
 106         xfs_ino_t               lookup_ino;
 107         xfs_dablk_t             offset;
 108         int                     error = 0;
 109
 110         sdc = container_of(dir_iter, struct xchk_dir_ctx, dir_iter);
 111         ip = sdc->sc->ip;
 112         mp = ip->i_mount;
 113         offset = xfs_dir2_db_to_da(mp->m_dir_geo,
 114                         xfs_dir2_dataptr_to_db(mp->m_dir_geo, pos));
 115
 116         if (xchk_should_terminate(sdc->sc, &error))
 117                 return error;
 118
 119         /* Does this inode number make sense? */
 120         if (!xfs_verify_dir_ino(mp, ino)) {
 121                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 122                 goto out;
 123         }
 124
 125         /* Does this name make sense? */
 126         if (!xfs_dir2_namecheck(name, namelen)) {
 127                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 128                 goto out;
 129         }
 130
 131         if (!strncmp(".", name, namelen)) {
 132                 /* If this is "." then check that the inum matches the dir. */
 133                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 134                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 135                                         offset);
 136                 if (ino != ip->i_ino)
 137                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 138                                         offset);
 139         } else if (!strncmp("..", name, namelen)) {
 140                 /*
 141                  * If this is ".." in the root inode, check that the inum
 142                  * matches this dir.
 143                  */
 144                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 145                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 146                                         offset);
 147                 if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
 148                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 149                                         offset);
 150         }
 151
 152         /* Verify that we can look up this name by hash. */
 153         xname.name = name;
 154         xname.len = namelen;
 155         xname.type = XFS_DIR3_FT_UNKNOWN;
 156
 157         error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
 158         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 159                         &error))
 160                 goto out;
 161         if (lookup_ino != ino) {
 162                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 163                 goto out;
 164         }
 165
 166         /* Verify the file type.  This function absorbs error codes. */
 167         error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 168         if (error)
 169                 goto out;
 170 out:
 171         /*
 172          * A negative error code returned here is supposed to cause the
 173          * dir_emit caller (xfs_readdir) to abort the directory iteration
 174          * and return zero to xchk_directory.
 175          */
 176         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 177                 return -EFSCORRUPTED;
 178         return error;
 179 }
 180
 181 /* Scrub a directory btree record. */
 182 STATIC int
 183 xchk_dir_rec(
 184         struct xchk_da_btree            *ds,
 185         int                             level)
 186 {
 187         struct xfs_da_state_blk         *blk = &ds->state->path.blk[level];
 188         struct xfs_mount                *mp = ds->state->mp;
 189         struct xfs_inode                *dp = ds->dargs.dp;
 190         struct xfs_da_geometry          *geo = mp->m_dir_geo;
 191         struct xfs_dir2_data_entry      *dent;
 192         struct xfs_buf                  *bp;
 193         struct xfs_dir2_leaf_entry      *ent;
 194         unsigned int                    end;
 195         unsigned int                    iter_off;
 196         xfs_ino_t                       ino;
 197         xfs_dablk_t                     rec_bno;
 198         xfs_dir2_db_t                   db;
 199         xfs_dir2_data_aoff_t            off;
 200         xfs_dir2_dataptr_t              ptr;
 201         xfs_dahash_t                    calc_hash;
 202         xfs_dahash_t                    hash;
 203         struct xfs_dir3_icleaf_hdr      hdr;
 204         unsigned int                    tag;
 205         int                             error;
 206
 207         ASSERT(blk->magic == XFS_DIR2_LEAF1_MAGIC ||
 208                blk->magic == XFS_DIR2_LEAFN_MAGIC);
 209
 210         xfs_dir2_leaf_hdr_from_disk(mp, &hdr, blk->bp->b_addr);
 211         ent = hdr.ents + blk->index;
 212
 213         /* Check the hash of the entry. */
 214         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 215         if (error)
 216                 goto out;
 217
 218         /* Valid hash pointer? */
 219         ptr = be32_to_cpu(ent->address);
 220         if (ptr == 0)
 221                 return 0;
 222
 223         /* Find the directory entry's location. */
 224         db = xfs_dir2_dataptr_to_db(geo, ptr);
 225         off = xfs_dir2_dataptr_to_off(geo, ptr);
 226         rec_bno = xfs_dir2_db_to_da(geo, db);
 227
 228         if (rec_bno >= geo->leafblk) {
 229                 xchk_da_set_corrupt(ds, level);
 230                 goto out;
 231         }
 232         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno,
 233                         XFS_DABUF_MAP_HOLE_OK, &bp);
 234         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 235                         &error))
 236                 goto out;
 237         if (!bp) {
 238                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 239                 goto out;
 240         }
 241         xchk_buffer_recheck(ds->sc, bp);
 242
 243         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 244                 goto out_relse;
 245
 246         dent = bp->b_addr + off;
 247
 248         /* Make sure we got a real directory entry. */
 249         iter_off = geo->data_entry_offset;
 250         end = xfs_dir3_data_end_offset(geo, bp->b_addr);
 251         if (!end) {
 252                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 253                 goto out_relse;
 254         }
 255         for (;;) {
 256                 struct xfs_dir2_data_entry      *dep = bp->b_addr + iter_off;
 257                 struct xfs_dir2_data_unused     *dup = bp->b_addr + iter_off;
 258
 259                 if (iter_off >= end) {
 260                         xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 261                         goto out_relse;
 262                 }
 263
 264                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 265                         iter_off += be16_to_cpu(dup->length);
 266                         continue;
 267                 }
 268                 if (dep == dent)
 269                         break;
 270                 iter_off += xfs_dir2_data_entsize(mp, dep->namelen);
 271         }
 272
 273         /* Retrieve the entry, sanity check it, and compare hashes. */
 274         ino = be64_to_cpu(dent->inumber);
 275         hash = be32_to_cpu(ent->hashval);
 276         tag = be16_to_cpup(xfs_dir2_data_entry_tag_p(mp, dent));
 277         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 278                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 279         if (dent->namelen == 0) {
 280                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 281                 goto out_relse;
 282         }
 283         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 284         if (calc_hash != hash)
 285                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 286
 287 out_relse:
 288         xfs_trans_brelse(ds->dargs.trans, bp);
 289 out:
 290         return error;
 291 }
 292
 293 /*
 294  * Is this unused entry either in the bestfree or smaller than all of
 295  * them?  We've already checked that the bestfrees are sorted longest to
 296  * shortest, and that there aren't any bogus entries.
 297  */
 298 STATIC void
 299 xchk_directory_check_free_entry(
 300         struct xfs_scrub                *sc,
 301         xfs_dablk_t                     lblk,
 302         struct xfs_dir2_data_free       *bf,
 303         struct xfs_dir2_data_unused     *dup)
 304 {
 305         struct xfs_dir2_data_free       *dfp;
 306         unsigned int                    dup_length;
 307
 308         dup_length = be16_to_cpu(dup->length);
 309
 310         /* Unused entry is shorter than any of the bestfrees */
 311         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 312                 return;
 313
 314         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 315                 if (dup_length == be16_to_cpu(dfp->length))
 316                         return;
 317
 318         /* Unused entry should be in the bestfrees but wasn't found. */
 319         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 320 }
 321
 322 /* Check free space info in a directory data block. */
 323 STATIC int
 324 xchk_directory_data_bestfree(
 325         struct xfs_scrub                *sc,
 326         xfs_dablk_t                     lblk,
 327         bool                            is_block)
 328 {
 329         struct xfs_dir2_data_unused     *dup;
 330         struct xfs_dir2_data_free       *dfp;
 331         struct xfs_buf                  *bp;
 332         struct xfs_dir2_data_free       *bf;
 333         struct xfs_mount                *mp = sc->mp;
 334         u16                             tag;
 335         unsigned int                    nr_bestfrees = 0;
 336         unsigned int                    nr_frees = 0;
 337         unsigned int                    smallest_bestfree;
 338         int                             newlen;
 339         unsigned int                    offset;
 340         unsigned int                    end;
 341         int                             error;
 342
 343         if (is_block) {
 344                 /* dir block format */
 345                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 346                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 347                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 348         } else {
 349                 /* dir data format */
 350                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, 0, &bp);
 351         }
 352         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 353                 goto out;
 354         xchk_buffer_recheck(sc, bp);
 355
 356         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 357
 358         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 359                 goto out_buf;
 360
 361         /* Do the bestfrees correspond to actual free space? */
 362         bf = xfs_dir2_data_bestfree_p(mp, bp->b_addr);
 363         smallest_bestfree = UINT_MAX;
 364         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 365                 offset = be16_to_cpu(dfp->offset);
 366                 if (offset == 0)
 367                         continue;
 368                 if (offset >= mp->m_dir_geo->blksize) {
 369                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 370                         goto out_buf;
 371                 }
 372                 dup = bp->b_addr + offset;
 373                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 374
 375                 /* bestfree doesn't match the entry it points at? */
 376                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 377                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 378                     tag != offset) {
 379                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 380                         goto out_buf;
 381                 }
 382
 383                 /* bestfree records should be ordered largest to smallest */
 384                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 385                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 386                         goto out_buf;
 387                 }
 388
 389                 smallest_bestfree = be16_to_cpu(dfp->length);
 390                 nr_bestfrees++;
 391         }
 392
 393         /* Make sure the bestfrees are actually the best free spaces. */
 394         offset = mp->m_dir_geo->data_entry_offset;
 395         end = xfs_dir3_data_end_offset(mp->m_dir_geo, bp->b_addr);
 396
 397         /* Iterate the entries, stopping when we hit or go past the end. */
 398         while (offset < end) {
 399                 dup = bp->b_addr + offset;
 400
 401                 /* Skip real entries */
 402                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 403                         struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
 404
 405                         newlen = xfs_dir2_data_entsize(mp, dep->namelen);
 406                         if (newlen <= 0) {
 407                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 408                                                 lblk);
 409                                 goto out_buf;
 410                         }
 411                         offset += newlen;
 412                         continue;
 413                 }
 414
 415                 /* Spot check this free entry */
 416                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 417                 if (tag != offset) {
 418                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 419                         goto out_buf;
 420                 }
 421
 422                 /*
 423                  * Either this entry is a bestfree or it's smaller than
 424                  * any of the bestfrees.
 425                  */
 426                 xchk_directory_check_free_entry(sc, lblk, bf, dup);
 427                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 428                         goto out_buf;
 429
 430                 /* Move on. */
 431                 newlen = be16_to_cpu(dup->length);
 432                 if (newlen <= 0) {
 433                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 434                         goto out_buf;
 435                 }
 436                 offset += newlen;
 437                 if (offset <= end)
 438                         nr_frees++;
 439         }
 440
 441         /* We're required to fill all the space. */
 442         if (offset != end)
 443                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 444
 445         /* Did we see at least as many free slots as there are bestfrees? */
 446         if (nr_frees < nr_bestfrees)
 447                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 448 out_buf:
 449         xfs_trans_brelse(sc->tp, bp);
 450 out:
 451         return error;
 452 }
 453
 454 /*
 455  * Does the free space length in the free space index block ($len) match
 456  * the longest length in the directory data block's bestfree array?
 457  * Assume that we've already checked that the data block's bestfree
 458  * array is in order.
 459  */
 460 STATIC void
 461 xchk_directory_check_freesp(
 462         struct xfs_scrub                *sc,
 463         xfs_dablk_t                     lblk,
 464         struct xfs_buf                  *dbp,
 465         unsigned int                    len)
 466 {
 467         struct xfs_dir2_data_free       *dfp;
 468
 469         dfp = xfs_dir2_data_bestfree_p(sc->mp, dbp->b_addr);
 470
 471         if (len != be16_to_cpu(dfp->length))
 472                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 473
 474         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 475                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 476 }
 477
 478 /* Check free space info in a directory leaf1 block. */
 479 STATIC int
 480 xchk_directory_leaf1_bestfree(
 481         struct xfs_scrub                *sc,
 482         struct xfs_da_args              *args,
 483         xfs_dablk_t                     lblk)
 484 {
 485         struct xfs_dir3_icleaf_hdr      leafhdr;
 486         struct xfs_dir2_leaf_tail       *ltp;
 487         struct xfs_dir2_leaf            *leaf;
 488         struct xfs_buf                  *dbp;
 489         struct xfs_buf                  *bp;
 490         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 491         __be16                          *bestp;
 492         __u16                           best;
 493         __u32                           hash;
 494         __u32                           lasthash = 0;
 495         __u32                           bestcount;
 496         unsigned int                    stale = 0;
 497         int                             i;
 498         int                             error;
 499
 500         /* Read the free space block. */
 501         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, &bp);
 502         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 503                 goto out;
 504         xchk_buffer_recheck(sc, bp);
 505
 506         leaf = bp->b_addr;
 507         xfs_dir2_leaf_hdr_from_disk(sc->ip->i_mount, &leafhdr, leaf);
 508         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 509         bestcount = be32_to_cpu(ltp->bestcount);
 510         bestp = xfs_dir2_leaf_bests_p(ltp);
 511
 512         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 513                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 514
 515                 if (hdr3->pad != cpu_to_be32(0))
 516                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 517         }
 518
 519         /*
 520          * There should be as many bestfree slots as there are dir data
 521          * blocks that can fit under i_size.
 522          */
 523         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 524                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 525                 goto out;
 526         }
 527
 528         /* Is the leaf count even remotely sane? */
 529         if (leafhdr.count > geo->leaf_max_ents) {
 530                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 531                 goto out;
 532         }
 533
 534         /* Leaves and bests don't overlap in leaf format. */
 535         if ((char *)&leafhdr.ents[leafhdr.count] > (char *)bestp) {
 536                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 537                 goto out;
 538         }
 539
 540         /* Check hash value order, count stale entries.  */
 541         for (i = 0; i < leafhdr.count; i++) {
 542                 hash = be32_to_cpu(leafhdr.ents[i].hashval);
 543                 if (i > 0 && lasthash > hash)
 544                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 545                 lasthash = hash;
 546                 if (leafhdr.ents[i].address ==
 547                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 548                         stale++;
 549         }
 550         if (leafhdr.stale != stale)
 551                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 552         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 553                 goto out;
 554
 555         /* Check all the bestfree entries. */
 556         for (i = 0; i < bestcount; i++, bestp++) {
 557                 best = be16_to_cpu(*bestp);
 558                 if (best == NULLDATAOFF)
 559                         continue;
 560                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 561                                 i * args->geo->fsbcount, 0, &dbp);
 562                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 563                                 &error))
 564                         break;
 565                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 566                 xfs_trans_brelse(sc->tp, dbp);
 567                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 568                         goto out;
 569         }
 570 out:
 571         return error;
 572 }
 573
 574 /* Check free space info in a directory freespace block. */
 575 STATIC int
 576 xchk_directory_free_bestfree(
 577         struct xfs_scrub                *sc,
 578         struct xfs_da_args              *args,
 579         xfs_dablk_t                     lblk)
 580 {
 581         struct xfs_dir3_icfree_hdr      freehdr;
 582         struct xfs_buf                  *dbp;
 583         struct xfs_buf                  *bp;
 584         __u16                           best;
 585         unsigned int                    stale = 0;
 586         int                             i;
 587         int                             error;
 588
 589         /* Read the free space block */
 590         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 591         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 592                 goto out;
 593         xchk_buffer_recheck(sc, bp);
 594
 595         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 596                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 597
 598                 if (hdr3->pad != cpu_to_be32(0))
 599                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 600         }
 601
 602         /* Check all the entries. */
 603         xfs_dir2_free_hdr_from_disk(sc->ip->i_mount, &freehdr, bp->b_addr);
 604         for (i = 0; i < freehdr.nvalid; i++) {
 605                 best = be16_to_cpu(freehdr.bests[i]);
 606                 if (best == NULLDATAOFF) {
 607                         stale++;
 608                         continue;
 609                 }
 610                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 611                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 612                                 0, &dbp);
 613                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 614                                 &error))
 615                         break;
 616                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 617                 xfs_trans_brelse(sc->tp, dbp);
 618         }
 619
 620         if (freehdr.nused + stale != freehdr.nvalid)
 621                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 622 out:
 623         return error;
 624 }
 625
 626 /* Check free space information in directories. */
 627 STATIC int
 628 xchk_directory_blocks(
 629         struct xfs_scrub        *sc)
 630 {
 631         struct xfs_bmbt_irec    got;
 632         struct xfs_da_args      args;
 633         struct xfs_ifork        *ifp;
 634         struct xfs_mount        *mp = sc->mp;
 635         xfs_fileoff_t           leaf_lblk;
 636         xfs_fileoff_t           free_lblk;
 637         xfs_fileoff_t           lblk;
 638         struct xfs_iext_cursor  icur;
 639         xfs_dablk_t             dabno;
 640         bool                    found;
 641         int                     is_block = 0;
 642         int                     error;
 643
 644         /* Ignore local format directories. */
 645         if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 646             sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
 647                 return 0;
 648
 649         ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 650         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 651         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 652         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 653
 654         /* Is this a block dir? */
 655         args.dp = sc->ip;
 656         args.geo = mp->m_dir_geo;
 657         args.trans = sc->tp;
 658         error = xfs_dir2_isblock(&args, &is_block);
 659         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 660                 goto out;
 661
 662         /* Iterate all the data extents in the directory... */
 663         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 664         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 665                 /* Block directories only have a single block at offset 0. */
 666                 if (is_block &&
 667                     (got.br_startoff > 0 ||
 668                      got.br_blockcount != args.geo->fsbcount)) {
 669                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 670                                         got.br_startoff);
 671                         break;
 672                 }
 673
 674                 /* No more data blocks... */
 675                 if (got.br_startoff >= leaf_lblk)
 676                         break;
 677
 678                 /*
 679                  * Check each data block's bestfree data.
 680                  *
 681                  * Iterate all the fsbcount-aligned block offsets in
 682                  * this directory.  The directory block reading code is
 683                  * smart enough to do its own bmap lookups to handle
 684                  * discontiguous directory blocks.  When we're done
 685                  * with the extent record, re-query the bmap at the
 686                  * next fsbcount-aligned offset to avoid redundant
 687                  * block checks.
 688                  */
 689                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 690                                 args.geo->fsbcount);
 691                      lblk < got.br_startoff + got.br_blockcount;
 692                      lblk += args.geo->fsbcount) {
 693                         error = xchk_directory_data_bestfree(sc, lblk,
 694                                         is_block);
 695                         if (error)
 696                                 goto out;
 697                 }
 698                 dabno = got.br_startoff + got.br_blockcount;
 699                 lblk = roundup(dabno, args.geo->fsbcount);
 700                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 701         }
 702
 703         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 704                 goto out;
 705
 706         /* Look for a leaf1 block, which has free info. */
 707         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 708             got.br_startoff == leaf_lblk &&
 709             got.br_blockcount == args.geo->fsbcount &&
 710             !xfs_iext_next_extent(ifp, &icur, &got)) {
 711                 if (is_block) {
 712                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 713                         goto out;
 714                 }
 715                 error = xchk_directory_leaf1_bestfree(sc, &args,
 716                                 leaf_lblk);
 717                 if (error)
 718                         goto out;
 719         }
 720
 721         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 722                 goto out;
 723
 724         /* Scan for free blocks */
 725         lblk = free_lblk;
 726         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 727         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 728                 /*
 729                  * Dirs can't have blocks mapped above 2^32.
 730                  * Single-block dirs shouldn't even be here.
 731                  */
 732                 lblk = got.br_startoff;
 733                 if (lblk & ~0xFFFFFFFFULL) {
 734                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 735                         goto out;
 736                 }
 737                 if (is_block) {
 738                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 739                         goto out;
 740                 }
 741
 742                 /*
 743                  * Check each dir free block's bestfree data.
 744                  *
 745                  * Iterate all the fsbcount-aligned block offsets in
 746                  * this directory.  The directory block reading code is
 747                  * smart enough to do its own bmap lookups to handle
 748                  * discontiguous directory blocks.  When we're done
 749                  * with the extent record, re-query the bmap at the
 750                  * next fsbcount-aligned offset to avoid redundant
 751                  * block checks.
 752                  */
 753                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 754                                 args.geo->fsbcount);
 755                      lblk < got.br_startoff + got.br_blockcount;
 756                      lblk += args.geo->fsbcount) {
 757                         error = xchk_directory_free_bestfree(sc, &args,
 758                                         lblk);
 759                         if (error)
 760                                 goto out;
 761                 }
 762                 dabno = got.br_startoff + got.br_blockcount;
 763                 lblk = roundup(dabno, args.geo->fsbcount);
 764                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 765         }
 766 out:
 767         return error;
 768 }
 769
 770 /* Scrub a whole directory. */
 771 int
 772 xchk_directory(
 773         struct xfs_scrub        *sc)
 774 {
 775         struct xchk_dir_ctx     sdc = {
 776                 .dir_iter.actor = xchk_dir_actor,
 777                 .dir_iter.pos = 0,
 778                 .sc = sc,
 779         };
 780         size_t                  bufsize;
 781         loff_t                  oldpos;
 782         int                     error = 0;
 783
 784         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 785                 return -ENOENT;
 786
 787         /* Plausible size? */
 788         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 789                 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
 790                 goto out;
 791         }
 792
 793         /* Check directory tree structure */
 794         error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
 795         if (error)
 796                 return error;
 797
 798         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 799                 return error;
 800
 801         /* Check the freespace. */
 802         error = xchk_directory_blocks(sc);
 803         if (error)
 804                 return error;
 805
 806         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 807                 return error;
 808
 809         /*
 810          * Check that every dirent we see can also be looked up by hash.
 811          * Userspace usually asks for a 32k buffer, so we will too.
 812          */
 813         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 814                         sc->ip->i_d.di_size);
 815
 816         /*
 817          * Look up every name in this directory by hash.
 818          *
 819          * Use the xfs_readdir function to call xchk_dir_actor on
 820          * every directory entry in this directory.  In _actor, we check
 821          * the name, inode number, and ftype (if applicable) of the
 822          * entry.  xfs_readdir uses the VFS filldir functions to provide
 823          * iteration context.
 824          *
 825          * The VFS grabs a read or write lock via i_rwsem before it reads
 826          * or writes to a directory.  If we've gotten this far we've
 827          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 828          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 829          * to drop the ILOCK here in order to reuse the _readdir and
 830          * _dir_lookup routines, which do their own ILOCK locking.
 831          */
 832         oldpos = 0;
 833         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 834         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 835         while (true) {
 836                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 837                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
 838                                 &error))
 839                         goto out;
 840                 if (oldpos == sdc.dir_iter.pos)
 841                         break;
 842                 oldpos = sdc.dir_iter.pos;
 843         }
 844
 845 out:
 846         return error;
 847 }