fs/xfs/scrub/dir.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_trans_resv.h"
  11 #include "xfs_mount.h"
  12 #include "xfs_log_format.h"
  13 #include "xfs_trans.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_icache.h"
  16 #include "xfs_dir2.h"
  17 #include "xfs_dir2_priv.h"
  18 #include "scrub/scrub.h"
  19 #include "scrub/common.h"
  20 #include "scrub/dabtree.h"
  21
  22 /* Set us up to scrub directories. */
  23 int
  24 xchk_setup_directory(
  25         struct xfs_scrub        *sc)
  26 {
  27         return xchk_setup_inode_contents(sc, 0);
  28 }
  29
  30 /* Directories */
  31
  32 /* Scrub a directory entry. */
  33
  34 struct xchk_dir_ctx {
  35         /* VFS fill-directory iterator */
  36         struct dir_context      dir_iter;
  37
  38         struct xfs_scrub        *sc;
  39 };
  40
  41 /* Check that an inode's mode matches a given DT_ type. */
  42 STATIC int
  43 xchk_dir_check_ftype(
  44         struct xchk_dir_ctx     *sdc,
  45         xfs_fileoff_t           offset,
  46         xfs_ino_t               inum,
  47         int                     dtype)
  48 {
  49         struct xfs_mount        *mp = sdc->sc->mp;
  50         struct xfs_inode        *ip;
  51         int                     ino_dtype;
  52         int                     error = 0;
  53
  54         if (!xfs_has_ftype(mp)) {
  55                 if (dtype != DT_UNKNOWN && dtype != DT_DIR)
  56                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
  57                                         offset);
  58                 goto out;
  59         }
  60
  61         /*
  62          * Grab the inode pointed to by the dirent.  We release the
  63          * inode before we cancel the scrub transaction.  Since we're
  64          * don't know a priori that releasing the inode won't trigger
  65          * eofblocks cleanup (which allocates what would be a nested
  66          * transaction), we can't use DONTCACHE here because DONTCACHE
  67          * inodes can trigger immediate inactive cleanup of the inode.
  68          *
  69          * If _iget returns -EINVAL or -ENOENT then the child inode number is
  70          * garbage and the directory is corrupt.  If the _iget returns
  71          * -EFSCORRUPTED or -EFSBADCRC then the child is corrupt which is a
  72          *  cross referencing error.  Any other error is an operational error.
  73          */
  74         error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
  75         if (error == -EINVAL || error == -ENOENT) {
  76                 error = -EFSCORRUPTED;
  77                 xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, 0, &error);
  78                 goto out;
  79         }
  80         if (!xchk_fblock_xref_process_error(sdc->sc, XFS_DATA_FORK, offset,
  81                         &error))
  82                 goto out;
  83
  84         /* Convert mode to the DT_* values that dir_emit uses. */
  85         ino_dtype = xfs_dir3_get_dtype(mp,
  86                         xfs_mode_to_ftype(VFS_I(ip)->i_mode));
  87         if (ino_dtype != dtype)
  88                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
  89         xfs_irele(ip);
  90 out:
  91         return error;
  92 }
  93
  94 /*
  95  * Scrub a single directory entry.
  96  *
  97  * We use the VFS directory iterator (i.e. readdir) to call this
  98  * function for every directory entry in a directory.  Once we're here,
  99  * we check the inode number to make sure it's sane, then we check that
 100  * we can look up this filename.  Finally, we check the ftype.
 101  */
 102 STATIC int
 103 xchk_dir_actor(
 104         struct dir_context      *dir_iter,
 105         const char              *name,
 106         int                     namelen,
 107         loff_t                  pos,
 108         u64                     ino,
 109         unsigned                type)
 110 {
 111         struct xfs_mount        *mp;
 112         struct xfs_inode        *ip;
 113         struct xchk_dir_ctx     *sdc;
 114         struct xfs_name         xname;
 115         xfs_ino_t               lookup_ino;
 116         xfs_dablk_t             offset;
 117         bool                    checked_ftype = false;
 118         int                     error = 0;
 119
 120         sdc = container_of(dir_iter, struct xchk_dir_ctx, dir_iter);
 121         ip = sdc->sc->ip;
 122         mp = ip->i_mount;
 123         offset = xfs_dir2_db_to_da(mp->m_dir_geo,
 124                         xfs_dir2_dataptr_to_db(mp->m_dir_geo, pos));
 125
 126         if (xchk_should_terminate(sdc->sc, &error))
 127                 return error;
 128
 129         /* Does this inode number make sense? */
 130         if (!xfs_verify_dir_ino(mp, ino)) {
 131                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 132                 goto out;
 133         }
 134
 135         /* Does this name make sense? */
 136         if (!xfs_dir2_namecheck(name, namelen)) {
 137                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 138                 goto out;
 139         }
 140
 141         if (!strncmp(".", name, namelen)) {
 142                 /* If this is "." then check that the inum matches the dir. */
 143                 if (xfs_has_ftype(mp) && type != DT_DIR)
 144                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 145                                         offset);
 146                 checked_ftype = true;
 147                 if (ino != ip->i_ino)
 148                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 149                                         offset);
 150         } else if (!strncmp("..", name, namelen)) {
 151                 /*
 152                  * If this is ".." in the root inode, check that the inum
 153                  * matches this dir.
 154                  */
 155                 if (xfs_has_ftype(mp) && type != DT_DIR)
 156                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 157                                         offset);
 158                 checked_ftype = true;
 159                 if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
 160                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 161                                         offset);
 162         }
 163
 164         /* Verify that we can look up this name by hash. */
 165         xname.name = name;
 166         xname.len = namelen;
 167         xname.type = XFS_DIR3_FT_UNKNOWN;
 168
 169         error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
 170         /* ENOENT means the hash lookup failed and the dir is corrupt */
 171         if (error == -ENOENT)
 172                 error = -EFSCORRUPTED;
 173         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 174                         &error))
 175                 goto out;
 176         if (lookup_ino != ino) {
 177                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 178                 goto out;
 179         }
 180
 181         /* Verify the file type.  This function absorbs error codes. */
 182         if (!checked_ftype) {
 183                 error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 184                 if (error)
 185                         goto out;
 186         }
 187 out:
 188         /*
 189          * A negative error code returned here is supposed to cause the
 190          * dir_emit caller (xfs_readdir) to abort the directory iteration
 191          * and return zero to xchk_directory.
 192          */
 193         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 194                 return -EFSCORRUPTED;
 195         return error;
 196 }
 197
 198 /* Scrub a directory btree record. */
 199 STATIC int
 200 xchk_dir_rec(
 201         struct xchk_da_btree            *ds,
 202         int                             level)
 203 {
 204         struct xfs_da_state_blk         *blk = &ds->state->path.blk[level];
 205         struct xfs_mount                *mp = ds->state->mp;
 206         struct xfs_inode                *dp = ds->dargs.dp;
 207         struct xfs_da_geometry          *geo = mp->m_dir_geo;
 208         struct xfs_dir2_data_entry      *dent;
 209         struct xfs_buf                  *bp;
 210         struct xfs_dir2_leaf_entry      *ent;
 211         unsigned int                    end;
 212         unsigned int                    iter_off;
 213         xfs_ino_t                       ino;
 214         xfs_dablk_t                     rec_bno;
 215         xfs_dir2_db_t                   db;
 216         xfs_dir2_data_aoff_t            off;
 217         xfs_dir2_dataptr_t              ptr;
 218         xfs_dahash_t                    calc_hash;
 219         xfs_dahash_t                    hash;
 220         struct xfs_dir3_icleaf_hdr      hdr;
 221         unsigned int                    tag;
 222         int                             error;
 223
 224         ASSERT(blk->magic == XFS_DIR2_LEAF1_MAGIC ||
 225                blk->magic == XFS_DIR2_LEAFN_MAGIC);
 226
 227         xfs_dir2_leaf_hdr_from_disk(mp, &hdr, blk->bp->b_addr);
 228         ent = hdr.ents + blk->index;
 229
 230         /* Check the hash of the entry. */
 231         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 232         if (error)
 233                 goto out;
 234
 235         /* Valid hash pointer? */
 236         ptr = be32_to_cpu(ent->address);
 237         if (ptr == 0)
 238                 return 0;
 239
 240         /* Find the directory entry's location. */
 241         db = xfs_dir2_dataptr_to_db(geo, ptr);
 242         off = xfs_dir2_dataptr_to_off(geo, ptr);
 243         rec_bno = xfs_dir2_db_to_da(geo, db);
 244
 245         if (rec_bno >= geo->leafblk) {
 246                 xchk_da_set_corrupt(ds, level);
 247                 goto out;
 248         }
 249         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno,
 250                         XFS_DABUF_MAP_HOLE_OK, &bp);
 251         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 252                         &error))
 253                 goto out;
 254         if (!bp) {
 255                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 256                 goto out;
 257         }
 258         xchk_buffer_recheck(ds->sc, bp);
 259
 260         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 261                 goto out_relse;
 262
 263         dent = bp->b_addr + off;
 264
 265         /* Make sure we got a real directory entry. */
 266         iter_off = geo->data_entry_offset;
 267         end = xfs_dir3_data_end_offset(geo, bp->b_addr);
 268         if (!end) {
 269                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 270                 goto out_relse;
 271         }
 272         for (;;) {
 273                 struct xfs_dir2_data_entry      *dep = bp->b_addr + iter_off;
 274                 struct xfs_dir2_data_unused     *dup = bp->b_addr + iter_off;
 275
 276                 if (iter_off >= end) {
 277                         xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 278                         goto out_relse;
 279                 }
 280
 281                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 282                         iter_off += be16_to_cpu(dup->length);
 283                         continue;
 284                 }
 285                 if (dep == dent)
 286                         break;
 287                 iter_off += xfs_dir2_data_entsize(mp, dep->namelen);
 288         }
 289
 290         /* Retrieve the entry, sanity check it, and compare hashes. */
 291         ino = be64_to_cpu(dent->inumber);
 292         hash = be32_to_cpu(ent->hashval);
 293         tag = be16_to_cpup(xfs_dir2_data_entry_tag_p(mp, dent));
 294         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 295                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 296         if (dent->namelen == 0) {
 297                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 298                 goto out_relse;
 299         }
 300         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 301         if (calc_hash != hash)
 302                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 303
 304 out_relse:
 305         xfs_trans_brelse(ds->dargs.trans, bp);
 306 out:
 307         return error;
 308 }
 309
 310 /*
 311  * Is this unused entry either in the bestfree or smaller than all of
 312  * them?  We've already checked that the bestfrees are sorted longest to
 313  * shortest, and that there aren't any bogus entries.
 314  */
 315 STATIC void
 316 xchk_directory_check_free_entry(
 317         struct xfs_scrub                *sc,
 318         xfs_dablk_t                     lblk,
 319         struct xfs_dir2_data_free       *bf,
 320         struct xfs_dir2_data_unused     *dup)
 321 {
 322         struct xfs_dir2_data_free       *dfp;
 323         unsigned int                    dup_length;
 324
 325         dup_length = be16_to_cpu(dup->length);
 326
 327         /* Unused entry is shorter than any of the bestfrees */
 328         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 329                 return;
 330
 331         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 332                 if (dup_length == be16_to_cpu(dfp->length))
 333                         return;
 334
 335         /* Unused entry should be in the bestfrees but wasn't found. */
 336         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 337 }
 338
 339 /* Check free space info in a directory data block. */
 340 STATIC int
 341 xchk_directory_data_bestfree(
 342         struct xfs_scrub                *sc,
 343         xfs_dablk_t                     lblk,
 344         bool                            is_block)
 345 {
 346         struct xfs_dir2_data_unused     *dup;
 347         struct xfs_dir2_data_free       *dfp;
 348         struct xfs_buf                  *bp;
 349         struct xfs_dir2_data_free       *bf;
 350         struct xfs_mount                *mp = sc->mp;
 351         u16                             tag;
 352         unsigned int                    nr_bestfrees = 0;
 353         unsigned int                    nr_frees = 0;
 354         unsigned int                    smallest_bestfree;
 355         int                             newlen;
 356         unsigned int                    offset;
 357         unsigned int                    end;
 358         int                             error;
 359
 360         if (is_block) {
 361                 /* dir block format */
 362                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 363                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 364                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 365         } else {
 366                 /* dir data format */
 367                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, 0, &bp);
 368         }
 369         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 370                 goto out;
 371         xchk_buffer_recheck(sc, bp);
 372
 373         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 374
 375         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 376                 goto out_buf;
 377
 378         /* Do the bestfrees correspond to actual free space? */
 379         bf = xfs_dir2_data_bestfree_p(mp, bp->b_addr);
 380         smallest_bestfree = UINT_MAX;
 381         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 382                 offset = be16_to_cpu(dfp->offset);
 383                 if (offset == 0)
 384                         continue;
 385                 if (offset >= mp->m_dir_geo->blksize) {
 386                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 387                         goto out_buf;
 388                 }
 389                 dup = bp->b_addr + offset;
 390                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 391
 392                 /* bestfree doesn't match the entry it points at? */
 393                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 394                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 395                     tag != offset) {
 396                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 397                         goto out_buf;
 398                 }
 399
 400                 /* bestfree records should be ordered largest to smallest */
 401                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 402                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 403                         goto out_buf;
 404                 }
 405
 406                 smallest_bestfree = be16_to_cpu(dfp->length);
 407                 nr_bestfrees++;
 408         }
 409
 410         /* Make sure the bestfrees are actually the best free spaces. */
 411         offset = mp->m_dir_geo->data_entry_offset;
 412         end = xfs_dir3_data_end_offset(mp->m_dir_geo, bp->b_addr);
 413
 414         /* Iterate the entries, stopping when we hit or go past the end. */
 415         while (offset < end) {
 416                 dup = bp->b_addr + offset;
 417
 418                 /* Skip real entries */
 419                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 420                         struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
 421
 422                         newlen = xfs_dir2_data_entsize(mp, dep->namelen);
 423                         if (newlen <= 0) {
 424                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 425                                                 lblk);
 426                                 goto out_buf;
 427                         }
 428                         offset += newlen;
 429                         continue;
 430                 }
 431
 432                 /* Spot check this free entry */
 433                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 434                 if (tag != offset) {
 435                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 436                         goto out_buf;
 437                 }
 438
 439                 /*
 440                  * Either this entry is a bestfree or it's smaller than
 441                  * any of the bestfrees.
 442                  */
 443                 xchk_directory_check_free_entry(sc, lblk, bf, dup);
 444                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 445                         goto out_buf;
 446
 447                 /* Move on. */
 448                 newlen = be16_to_cpu(dup->length);
 449                 if (newlen <= 0) {
 450                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 451                         goto out_buf;
 452                 }
 453                 offset += newlen;
 454                 if (offset <= end)
 455                         nr_frees++;
 456         }
 457
 458         /* We're required to fill all the space. */
 459         if (offset != end)
 460                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 461
 462         /* Did we see at least as many free slots as there are bestfrees? */
 463         if (nr_frees < nr_bestfrees)
 464                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 465 out_buf:
 466         xfs_trans_brelse(sc->tp, bp);
 467 out:
 468         return error;
 469 }
 470
 471 /*
 472  * Does the free space length in the free space index block ($len) match
 473  * the longest length in the directory data block's bestfree array?
 474  * Assume that we've already checked that the data block's bestfree
 475  * array is in order.
 476  */
 477 STATIC void
 478 xchk_directory_check_freesp(
 479         struct xfs_scrub                *sc,
 480         xfs_dablk_t                     lblk,
 481         struct xfs_buf                  *dbp,
 482         unsigned int                    len)
 483 {
 484         struct xfs_dir2_data_free       *dfp;
 485
 486         dfp = xfs_dir2_data_bestfree_p(sc->mp, dbp->b_addr);
 487
 488         if (len != be16_to_cpu(dfp->length))
 489                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 490
 491         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 492                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 493 }
 494
 495 /* Check free space info in a directory leaf1 block. */
 496 STATIC int
 497 xchk_directory_leaf1_bestfree(
 498         struct xfs_scrub                *sc,
 499         struct xfs_da_args              *args,
 500         xfs_dir2_db_t                   last_data_db,
 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_has_crc(sc->mp)) {
 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 must be enough bestfree slots to cover all the directory data
 539          * blocks that we scanned.  It is possible for there to be a hole
 540          * between the last data block and i_disk_size.  This seems like an
 541          * oversight to the scrub author, but as we have been writing out
 542          * directories like this (and xfs_repair doesn't mind them) for years,
 543          * that's what we have to check.
 544          */
 545         if (bestcount != last_data_db + 1) {
 546                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 547                 goto out;
 548         }
 549
 550         /* Is the leaf count even remotely sane? */
 551         if (leafhdr.count > geo->leaf_max_ents) {
 552                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 553                 goto out;
 554         }
 555
 556         /* Leaves and bests don't overlap in leaf format. */
 557         if ((char *)&leafhdr.ents[leafhdr.count] > (char *)bestp) {
 558                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 559                 goto out;
 560         }
 561
 562         /* Check hash value order, count stale entries.  */
 563         for (i = 0; i < leafhdr.count; i++) {
 564                 hash = be32_to_cpu(leafhdr.ents[i].hashval);
 565                 if (i > 0 && lasthash > hash)
 566                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 567                 lasthash = hash;
 568                 if (leafhdr.ents[i].address ==
 569                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 570                         stale++;
 571         }
 572         if (leafhdr.stale != stale)
 573                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 574         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 575                 goto out;
 576
 577         /* Check all the bestfree entries. */
 578         for (i = 0; i < bestcount; i++, bestp++) {
 579                 best = be16_to_cpu(*bestp);
 580                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 581                                 xfs_dir2_db_to_da(args->geo, i),
 582                                 XFS_DABUF_MAP_HOLE_OK,
 583                                 &dbp);
 584                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 585                                 &error))
 586                         break;
 587
 588                 if (!dbp) {
 589                         if (best != NULLDATAOFF) {
 590                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 591                                                 lblk);
 592                                 break;
 593                         }
 594                         continue;
 595                 }
 596
 597                 if (best == NULLDATAOFF)
 598                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 599                 else
 600                         xchk_directory_check_freesp(sc, lblk, dbp, best);
 601                 xfs_trans_brelse(sc->tp, dbp);
 602                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 603                         break;
 604         }
 605 out:
 606         xfs_trans_brelse(sc->tp, bp);
 607         return error;
 608 }
 609
 610 /* Check free space info in a directory freespace block. */
 611 STATIC int
 612 xchk_directory_free_bestfree(
 613         struct xfs_scrub                *sc,
 614         struct xfs_da_args              *args,
 615         xfs_dablk_t                     lblk)
 616 {
 617         struct xfs_dir3_icfree_hdr      freehdr;
 618         struct xfs_buf                  *dbp;
 619         struct xfs_buf                  *bp;
 620         __u16                           best;
 621         unsigned int                    stale = 0;
 622         int                             i;
 623         int                             error;
 624
 625         /* Read the free space block */
 626         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 627         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 628                 return error;
 629         xchk_buffer_recheck(sc, bp);
 630
 631         if (xfs_has_crc(sc->mp)) {
 632                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 633
 634                 if (hdr3->pad != cpu_to_be32(0))
 635                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 636         }
 637
 638         /* Check all the entries. */
 639         xfs_dir2_free_hdr_from_disk(sc->ip->i_mount, &freehdr, bp->b_addr);
 640         for (i = 0; i < freehdr.nvalid; i++) {
 641                 best = be16_to_cpu(freehdr.bests[i]);
 642                 if (best == NULLDATAOFF) {
 643                         stale++;
 644                         continue;
 645                 }
 646                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 647                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 648                                 0, &dbp);
 649                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 650                                 &error))
 651                         goto out;
 652                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 653                 xfs_trans_brelse(sc->tp, dbp);
 654         }
 655
 656         if (freehdr.nused + stale != freehdr.nvalid)
 657                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 658 out:
 659         xfs_trans_brelse(sc->tp, bp);
 660         return error;
 661 }
 662
 663 /* Check free space information in directories. */
 664 STATIC int
 665 xchk_directory_blocks(
 666         struct xfs_scrub        *sc)
 667 {
 668         struct xfs_bmbt_irec    got;
 669         struct xfs_da_args      args;
 670         struct xfs_ifork        *ifp = xfs_ifork_ptr(sc->ip, XFS_DATA_FORK);
 671         struct xfs_mount        *mp = sc->mp;
 672         xfs_fileoff_t           leaf_lblk;
 673         xfs_fileoff_t           free_lblk;
 674         xfs_fileoff_t           lblk;
 675         struct xfs_iext_cursor  icur;
 676         xfs_dablk_t             dabno;
 677         xfs_dir2_db_t           last_data_db = 0;
 678         bool                    found;
 679         int                     is_block = 0;
 680         int                     error;
 681
 682         /* Ignore local format directories. */
 683         if (ifp->if_format != XFS_DINODE_FMT_EXTENTS &&
 684             ifp->if_format != XFS_DINODE_FMT_BTREE)
 685                 return 0;
 686
 687         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 688         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 689         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 690
 691         /* Is this a block dir? */
 692         args.dp = sc->ip;
 693         args.geo = mp->m_dir_geo;
 694         args.trans = sc->tp;
 695         error = xfs_dir2_isblock(&args, &is_block);
 696         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 697                 goto out;
 698
 699         /* Iterate all the data extents in the directory... */
 700         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 701         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 702                 /* No more data blocks... */
 703                 if (got.br_startoff >= leaf_lblk)
 704                         break;
 705
 706                 /*
 707                  * Check each data block's bestfree data.
 708                  *
 709                  * Iterate all the fsbcount-aligned block offsets in
 710                  * this directory.  The directory block reading code is
 711                  * smart enough to do its own bmap lookups to handle
 712                  * discontiguous directory blocks.  When we're done
 713                  * with the extent record, re-query the bmap at the
 714                  * next fsbcount-aligned offset to avoid redundant
 715                  * block checks.
 716                  */
 717                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 718                                 args.geo->fsbcount);
 719                      lblk < got.br_startoff + got.br_blockcount;
 720                      lblk += args.geo->fsbcount) {
 721                         last_data_db = xfs_dir2_da_to_db(args.geo, lblk);
 722                         error = xchk_directory_data_bestfree(sc, lblk,
 723                                         is_block);
 724                         if (error)
 725                                 goto out;
 726                 }
 727                 dabno = got.br_startoff + got.br_blockcount;
 728                 lblk = roundup(dabno, args.geo->fsbcount);
 729                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 730         }
 731
 732         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 733                 goto out;
 734
 735         /* Look for a leaf1 block, which has free info. */
 736         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 737             got.br_startoff == leaf_lblk &&
 738             got.br_blockcount == args.geo->fsbcount &&
 739             !xfs_iext_next_extent(ifp, &icur, &got)) {
 740                 if (is_block) {
 741                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 742                         goto out;
 743                 }
 744                 error = xchk_directory_leaf1_bestfree(sc, &args, last_data_db,
 745                                 leaf_lblk);
 746                 if (error)
 747                         goto out;
 748         }
 749
 750         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 751                 goto out;
 752
 753         /* Scan for free blocks */
 754         lblk = free_lblk;
 755         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 756         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 757                 /*
 758                  * Dirs can't have blocks mapped above 2^32.
 759                  * Single-block dirs shouldn't even be here.
 760                  */
 761                 lblk = got.br_startoff;
 762                 if (lblk & ~0xFFFFFFFFULL) {
 763                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 764                         goto out;
 765                 }
 766                 if (is_block) {
 767                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 768                         goto out;
 769                 }
 770
 771                 /*
 772                  * Check each dir free block's bestfree data.
 773                  *
 774                  * Iterate all the fsbcount-aligned block offsets in
 775                  * this directory.  The directory block reading code is
 776                  * smart enough to do its own bmap lookups to handle
 777                  * discontiguous directory blocks.  When we're done
 778                  * with the extent record, re-query the bmap at the
 779                  * next fsbcount-aligned offset to avoid redundant
 780                  * block checks.
 781                  */
 782                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 783                                 args.geo->fsbcount);
 784                      lblk < got.br_startoff + got.br_blockcount;
 785                      lblk += args.geo->fsbcount) {
 786                         error = xchk_directory_free_bestfree(sc, &args,
 787                                         lblk);
 788                         if (error)
 789                                 goto out;
 790                 }
 791                 dabno = got.br_startoff + got.br_blockcount;
 792                 lblk = roundup(dabno, args.geo->fsbcount);
 793                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 794         }
 795 out:
 796         return error;
 797 }
 798
 799 /* Scrub a whole directory. */
 800 int
 801 xchk_directory(
 802         struct xfs_scrub        *sc)
 803 {
 804         struct xchk_dir_ctx     sdc = {
 805                 .dir_iter.actor = xchk_dir_actor,
 806                 .dir_iter.pos = 0,
 807                 .sc = sc,
 808         };
 809         size_t                  bufsize;
 810         loff_t                  oldpos;
 811         int                     error = 0;
 812
 813         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 814                 return -ENOENT;
 815
 816         /* Plausible size? */
 817         if (sc->ip->i_disk_size < xfs_dir2_sf_hdr_size(0)) {
 818                 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
 819                 goto out;
 820         }
 821
 822         /* Check directory tree structure */
 823         error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
 824         if (error)
 825                 return error;
 826
 827         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 828                 return error;
 829
 830         /* Check the freespace. */
 831         error = xchk_directory_blocks(sc);
 832         if (error)
 833                 return error;
 834
 835         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 836                 return error;
 837
 838         /*
 839          * Check that every dirent we see can also be looked up by hash.
 840          * Userspace usually asks for a 32k buffer, so we will too.
 841          */
 842         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 843                         sc->ip->i_disk_size);
 844
 845         /*
 846          * Look up every name in this directory by hash.
 847          *
 848          * Use the xfs_readdir function to call xchk_dir_actor on
 849          * every directory entry in this directory.  In _actor, we check
 850          * the name, inode number, and ftype (if applicable) of the
 851          * entry.  xfs_readdir uses the VFS filldir functions to provide
 852          * iteration context.
 853          *
 854          * The VFS grabs a read or write lock via i_rwsem before it reads
 855          * or writes to a directory.  If we've gotten this far we've
 856          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 857          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 858          * to drop the ILOCK here in order to reuse the _readdir and
 859          * _dir_lookup routines, which do their own ILOCK locking.
 860          */
 861         oldpos = 0;
 862         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 863         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 864         while (true) {
 865                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 866                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
 867                                 &error))
 868                         goto out;
 869                 if (oldpos == sdc.dir_iter.pos)
 870                         break;
 871                 oldpos = sdc.dir_iter.pos;
 872         }
 873
 874 out:
 875         return error;
 876 }