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         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 160                         &error))
 161                 goto out;
 162         if (lookup_ino != ino) {
 163                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 164                 goto out;
 165         }
 166
 167         /* Verify the file type.  This function absorbs error codes. */
 168         error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 169         if (error)
 170                 goto out;
 171 out:
 172         /*
 173          * A negative error code returned here is supposed to cause the
 174          * dir_emit caller (xfs_readdir) to abort the directory iteration
 175          * and return zero to xchk_directory.
 176          */
 177         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 178                 return -EFSCORRUPTED;
 179         return error;
 180 }
 181
 182 /* Scrub a directory btree record. */
 183 STATIC int
 184 xchk_dir_rec(
 185         struct xchk_da_btree            *ds,
 186         int                             level,
 187         void                            *rec)
 188 {
 189         struct xfs_mount                *mp = ds->state->mp;
 190         struct xfs_dir2_leaf_entry      *ent = rec;
 191         struct xfs_inode                *dp = ds->dargs.dp;
 192         struct xfs_dir2_data_entry      *dent;
 193         struct xfs_buf                  *bp;
 194         char                            *p, *endp;
 195         xfs_ino_t                       ino;
 196         xfs_dablk_t                     rec_bno;
 197         xfs_dir2_db_t                   db;
 198         xfs_dir2_data_aoff_t            off;
 199         xfs_dir2_dataptr_t              ptr;
 200         xfs_dahash_t                    calc_hash;
 201         xfs_dahash_t                    hash;
 202         unsigned int                    tag;
 203         int                             error;
 204
 205         /* Check the hash of the entry. */
 206         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 207         if (error)
 208                 goto out;
 209
 210         /* Valid hash pointer? */
 211         ptr = be32_to_cpu(ent->address);
 212         if (ptr == 0)
 213                 return 0;
 214
 215         /* Find the directory entry's location. */
 216         db = xfs_dir2_dataptr_to_db(mp->m_dir_geo, ptr);
 217         off = xfs_dir2_dataptr_to_off(mp->m_dir_geo, ptr);
 218         rec_bno = xfs_dir2_db_to_da(mp->m_dir_geo, db);
 219
 220         if (rec_bno >= mp->m_dir_geo->leafblk) {
 221                 xchk_da_set_corrupt(ds, level);
 222                 goto out;
 223         }
 224         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
 225         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 226                         &error))
 227                 goto out;
 228         if (!bp) {
 229                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 230                 goto out;
 231         }
 232         xchk_buffer_recheck(ds->sc, bp);
 233
 234         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 235                 goto out_relse;
 236
 237         dent = (struct xfs_dir2_data_entry *)(((char *)bp->b_addr) + off);
 238
 239         /* Make sure we got a real directory entry. */
 240         p = (char *)mp->m_dir_inode_ops->data_entry_p(bp->b_addr);
 241         endp = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 242         if (!endp) {
 243                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 244                 goto out_relse;
 245         }
 246         while (p < endp) {
 247                 struct xfs_dir2_data_entry      *dep;
 248                 struct xfs_dir2_data_unused     *dup;
 249
 250                 dup = (struct xfs_dir2_data_unused *)p;
 251                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 252                         p += be16_to_cpu(dup->length);
 253                         continue;
 254                 }
 255                 dep = (struct xfs_dir2_data_entry *)p;
 256                 if (dep == dent)
 257                         break;
 258                 p += mp->m_dir_inode_ops->data_entsize(dep->namelen);
 259         }
 260         if (p >= endp) {
 261                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 262                 goto out_relse;
 263         }
 264
 265         /* Retrieve the entry, sanity check it, and compare hashes. */
 266         ino = be64_to_cpu(dent->inumber);
 267         hash = be32_to_cpu(ent->hashval);
 268         tag = be16_to_cpup(dp->d_ops->data_entry_tag_p(dent));
 269         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 270                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 271         if (dent->namelen == 0) {
 272                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 273                 goto out_relse;
 274         }
 275         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 276         if (calc_hash != hash)
 277                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 278
 279 out_relse:
 280         xfs_trans_brelse(ds->dargs.trans, bp);
 281 out:
 282         return error;
 283 }
 284
 285 /*
 286  * Is this unused entry either in the bestfree or smaller than all of
 287  * them?  We've already checked that the bestfrees are sorted longest to
 288  * shortest, and that there aren't any bogus entries.
 289  */
 290 STATIC void
 291 xchk_directory_check_free_entry(
 292         struct xfs_scrub                *sc,
 293         xfs_dablk_t                     lblk,
 294         struct xfs_dir2_data_free       *bf,
 295         struct xfs_dir2_data_unused     *dup)
 296 {
 297         struct xfs_dir2_data_free       *dfp;
 298         unsigned int                    dup_length;
 299
 300         dup_length = be16_to_cpu(dup->length);
 301
 302         /* Unused entry is shorter than any of the bestfrees */
 303         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 304                 return;
 305
 306         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 307                 if (dup_length == be16_to_cpu(dfp->length))
 308                         return;
 309
 310         /* Unused entry should be in the bestfrees but wasn't found. */
 311         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 312 }
 313
 314 /* Check free space info in a directory data block. */
 315 STATIC int
 316 xchk_directory_data_bestfree(
 317         struct xfs_scrub                *sc,
 318         xfs_dablk_t                     lblk,
 319         bool                            is_block)
 320 {
 321         struct xfs_dir2_data_unused     *dup;
 322         struct xfs_dir2_data_free       *dfp;
 323         struct xfs_buf                  *bp;
 324         struct xfs_dir2_data_free       *bf;
 325         struct xfs_mount                *mp = sc->mp;
 326         const struct xfs_dir_ops        *d_ops;
 327         char                            *ptr;
 328         char                            *endptr;
 329         u16                             tag;
 330         unsigned int                    nr_bestfrees = 0;
 331         unsigned int                    nr_frees = 0;
 332         unsigned int                    smallest_bestfree;
 333         int                             newlen;
 334         int                             offset;
 335         int                             error;
 336
 337         d_ops = sc->ip->d_ops;
 338
 339         if (is_block) {
 340                 /* dir block format */
 341                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 342                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 343                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 344         } else {
 345                 /* dir data format */
 346                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, -1, &bp);
 347         }
 348         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 349                 goto out;
 350         xchk_buffer_recheck(sc, bp);
 351
 352         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 353
 354         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 355                 goto out_buf;
 356
 357         /* Do the bestfrees correspond to actual free space? */
 358         bf = d_ops->data_bestfree_p(bp->b_addr);
 359         smallest_bestfree = UINT_MAX;
 360         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 361                 offset = be16_to_cpu(dfp->offset);
 362                 if (offset == 0)
 363                         continue;
 364                 if (offset >= mp->m_dir_geo->blksize) {
 365                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 366                         goto out_buf;
 367                 }
 368                 dup = (struct xfs_dir2_data_unused *)(bp->b_addr + offset);
 369                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 370
 371                 /* bestfree doesn't match the entry it points at? */
 372                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 373                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 374                     tag != ((char *)dup - (char *)bp->b_addr)) {
 375                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 376                         goto out_buf;
 377                 }
 378
 379                 /* bestfree records should be ordered largest to smallest */
 380                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 381                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 382                         goto out_buf;
 383                 }
 384
 385                 smallest_bestfree = be16_to_cpu(dfp->length);
 386                 nr_bestfrees++;
 387         }
 388
 389         /* Make sure the bestfrees are actually the best free spaces. */
 390         ptr = (char *)d_ops->data_entry_p(bp->b_addr);
 391         endptr = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 392
 393         /* Iterate the entries, stopping when we hit or go past the end. */
 394         while (ptr < endptr) {
 395                 dup = (struct xfs_dir2_data_unused *)ptr;
 396                 /* Skip real entries */
 397                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 398                         struct xfs_dir2_data_entry      *dep;
 399
 400                         dep = (struct xfs_dir2_data_entry *)ptr;
 401                         newlen = d_ops->data_entsize(dep->namelen);
 402                         if (newlen <= 0) {
 403                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 404                                                 lblk);
 405                                 goto out_buf;
 406                         }
 407                         ptr += newlen;
 408                         continue;
 409                 }
 410
 411                 /* Spot check this free entry */
 412                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 413                 if (tag != ((char *)dup - (char *)bp->b_addr)) {
 414                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 415                         goto out_buf;
 416                 }
 417
 418                 /*
 419                  * Either this entry is a bestfree or it's smaller than
 420                  * any of the bestfrees.
 421                  */
 422                 xchk_directory_check_free_entry(sc, lblk, bf, dup);
 423                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 424                         goto out_buf;
 425
 426                 /* Move on. */
 427                 newlen = be16_to_cpu(dup->length);
 428                 if (newlen <= 0) {
 429                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 430                         goto out_buf;
 431                 }
 432                 ptr += newlen;
 433                 if (ptr <= endptr)
 434                         nr_frees++;
 435         }
 436
 437         /* We're required to fill all the space. */
 438         if (ptr != endptr)
 439                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 440
 441         /* Did we see at least as many free slots as there are bestfrees? */
 442         if (nr_frees < nr_bestfrees)
 443                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 444 out_buf:
 445         xfs_trans_brelse(sc->tp, bp);
 446 out:
 447         return error;
 448 }
 449
 450 /*
 451  * Does the free space length in the free space index block ($len) match
 452  * the longest length in the directory data block's bestfree array?
 453  * Assume that we've already checked that the data block's bestfree
 454  * array is in order.
 455  */
 456 STATIC void
 457 xchk_directory_check_freesp(
 458         struct xfs_scrub                *sc,
 459         xfs_dablk_t                     lblk,
 460         struct xfs_buf                  *dbp,
 461         unsigned int                    len)
 462 {
 463         struct xfs_dir2_data_free       *dfp;
 464
 465         dfp = sc->ip->d_ops->data_bestfree_p(dbp->b_addr);
 466
 467         if (len != be16_to_cpu(dfp->length))
 468                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 469
 470         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 471                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 472 }
 473
 474 /* Check free space info in a directory leaf1 block. */
 475 STATIC int
 476 xchk_directory_leaf1_bestfree(
 477         struct xfs_scrub                *sc,
 478         struct xfs_da_args              *args,
 479         xfs_dablk_t                     lblk)
 480 {
 481         struct xfs_dir3_icleaf_hdr      leafhdr;
 482         struct xfs_dir2_leaf_entry      *ents;
 483         struct xfs_dir2_leaf_tail       *ltp;
 484         struct xfs_dir2_leaf            *leaf;
 485         struct xfs_buf                  *dbp;
 486         struct xfs_buf                  *bp;
 487         const struct xfs_dir_ops        *d_ops = sc->ip->d_ops;
 488         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 489         __be16                          *bestp;
 490         __u16                           best;
 491         __u32                           hash;
 492         __u32                           lasthash = 0;
 493         __u32                           bestcount;
 494         unsigned int                    stale = 0;
 495         int                             i;
 496         int                             error;
 497
 498         /* Read the free space block. */
 499         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, -1, &bp);
 500         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 501                 goto out;
 502         xchk_buffer_recheck(sc, bp);
 503
 504         leaf = bp->b_addr;
 505         d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 506         ents = d_ops->leaf_ents_p(leaf);
 507         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 508         bestcount = be32_to_cpu(ltp->bestcount);
 509         bestp = xfs_dir2_leaf_bests_p(ltp);
 510
 511         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 512                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 513
 514                 if (hdr3->pad != cpu_to_be32(0))
 515                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 516         }
 517
 518         /*
 519          * There should be as many bestfree slots as there are dir data
 520          * blocks that can fit under i_size.
 521          */
 522         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 523                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 524                 goto out;
 525         }
 526
 527         /* Is the leaf count even remotely sane? */
 528         if (leafhdr.count > d_ops->leaf_max_ents(geo)) {
 529                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 530                 goto out;
 531         }
 532
 533         /* Leaves and bests don't overlap in leaf format. */
 534         if ((char *)&ents[leafhdr.count] > (char *)bestp) {
 535                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 536                 goto out;
 537         }
 538
 539         /* Check hash value order, count stale entries.  */
 540         for (i = 0; i < leafhdr.count; i++) {
 541                 hash = be32_to_cpu(ents[i].hashval);
 542                 if (i > 0 && lasthash > hash)
 543                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 544                 lasthash = hash;
 545                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 546                         stale++;
 547         }
 548         if (leafhdr.stale != stale)
 549                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 550         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 551                 goto out;
 552
 553         /* Check all the bestfree entries. */
 554         for (i = 0; i < bestcount; i++, bestp++) {
 555                 best = be16_to_cpu(*bestp);
 556                 if (best == NULLDATAOFF)
 557                         continue;
 558                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 559                                 i * args->geo->fsbcount, -1, &dbp);
 560                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 561                                 &error))
 562                         break;
 563                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 564                 xfs_trans_brelse(sc->tp, dbp);
 565                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 566                         goto out;
 567         }
 568 out:
 569         return error;
 570 }
 571
 572 /* Check free space info in a directory freespace block. */
 573 STATIC int
 574 xchk_directory_free_bestfree(
 575         struct xfs_scrub                *sc,
 576         struct xfs_da_args              *args,
 577         xfs_dablk_t                     lblk)
 578 {
 579         struct xfs_dir3_icfree_hdr      freehdr;
 580         struct xfs_buf                  *dbp;
 581         struct xfs_buf                  *bp;
 582         __be16                          *bestp;
 583         __u16                           best;
 584         unsigned int                    stale = 0;
 585         int                             i;
 586         int                             error;
 587
 588         /* Read the free space block */
 589         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 590         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 591                 goto out;
 592         xchk_buffer_recheck(sc, bp);
 593
 594         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 595                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 596
 597                 if (hdr3->pad != cpu_to_be32(0))
 598                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 599         }
 600
 601         /* Check all the entries. */
 602         sc->ip->d_ops->free_hdr_from_disk(&freehdr, bp->b_addr);
 603         bestp = sc->ip->d_ops->free_bests_p(bp->b_addr);
 604         for (i = 0; i < freehdr.nvalid; i++, bestp++) {
 605                 best = be16_to_cpu(*bestp);
 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                                 -1, &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 }