root/fs/xfs/scrub/dir.c

DEFINITIONS

1. xchk_setup_directory
2. xchk_dir_check_ftype
3. xchk_dir_actor
4. xchk_dir_rec
5. xchk_directory_check_free_entry
6. xchk_directory_data_bestfree
7. xchk_directory_check_freesp
8. xchk_directory_leaf1_bestfree
9. xchk_directory_free_bestfree
10. xchk_directory_blocks
11. xchk_directory

   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         /* Does this inode number make sense? */
 117         if (!xfs_verify_dir_ino(mp, ino)) {
 118                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 119                 goto out;
 120         }
 121 
 122         /* Does this name make sense? */
 123         if (!xfs_dir2_namecheck(name, namelen)) {
 124                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 125                 goto out;
 126         }
 127 
 128         if (!strncmp(".", name, namelen)) {
 129                 /* If this is "." then check that the inum matches the dir. */
 130                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 131                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 132                                         offset);
 133                 if (ino != ip->i_ino)
 134                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 135                                         offset);
 136         } else if (!strncmp("..", name, namelen)) {
 137                 /*
 138                  * If this is ".." in the root inode, check that the inum
 139                  * matches this dir.
 140                  */
 141                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 142                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 143                                         offset);
 144                 if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
 145                         xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 146                                         offset);
 147         }
 148 
 149         /* Verify that we can look up this name by hash. */
 150         xname.name = name;
 151         xname.len = namelen;
 152         xname.type = XFS_DIR3_FT_UNKNOWN;
 153 
 154         error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
 155         if (!xchk_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 156                         &error))
 157                 goto out;
 158         if (lookup_ino != ino) {
 159                 xchk_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 160                 goto out;
 161         }
 162 
 163         /* Verify the file type.  This function absorbs error codes. */
 164         error = xchk_dir_check_ftype(sdc, offset, lookup_ino, type);
 165         if (error)
 166                 goto out;
 167 out:
 168         /*
 169          * A negative error code returned here is supposed to cause the
 170          * dir_emit caller (xfs_readdir) to abort the directory iteration
 171          * and return zero to xchk_directory.
 172          */
 173         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 174                 return -EFSCORRUPTED;
 175         return error;
 176 }
 177 
 178 /* Scrub a directory btree record. */
 179 STATIC int
 180 xchk_dir_rec(
 181         struct xchk_da_btree            *ds,
 182         int                             level,
 183         void                            *rec)
 184 {
 185         struct xfs_mount                *mp = ds->state->mp;
 186         struct xfs_dir2_leaf_entry      *ent = rec;
 187         struct xfs_inode                *dp = ds->dargs.dp;
 188         struct xfs_dir2_data_entry      *dent;
 189         struct xfs_buf                  *bp;
 190         char                            *p, *endp;
 191         xfs_ino_t                       ino;
 192         xfs_dablk_t                     rec_bno;
 193         xfs_dir2_db_t                   db;
 194         xfs_dir2_data_aoff_t            off;
 195         xfs_dir2_dataptr_t              ptr;
 196         xfs_dahash_t                    calc_hash;
 197         xfs_dahash_t                    hash;
 198         unsigned int                    tag;
 199         int                             error;
 200 
 201         /* Check the hash of the entry. */
 202         error = xchk_da_btree_hash(ds, level, &ent->hashval);
 203         if (error)
 204                 goto out;
 205 
 206         /* Valid hash pointer? */
 207         ptr = be32_to_cpu(ent->address);
 208         if (ptr == 0)
 209                 return 0;
 210 
 211         /* Find the directory entry's location. */
 212         db = xfs_dir2_dataptr_to_db(mp->m_dir_geo, ptr);
 213         off = xfs_dir2_dataptr_to_off(mp->m_dir_geo, ptr);
 214         rec_bno = xfs_dir2_db_to_da(mp->m_dir_geo, db);
 215 
 216         if (rec_bno >= mp->m_dir_geo->leafblk) {
 217                 xchk_da_set_corrupt(ds, level);
 218                 goto out;
 219         }
 220         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
 221         if (!xchk_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 222                         &error))
 223                 goto out;
 224         if (!bp) {
 225                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 226                 goto out;
 227         }
 228         xchk_buffer_recheck(ds->sc, bp);
 229 
 230         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 231                 goto out_relse;
 232 
 233         dent = (struct xfs_dir2_data_entry *)(((char *)bp->b_addr) + off);
 234 
 235         /* Make sure we got a real directory entry. */
 236         p = (char *)mp->m_dir_inode_ops->data_entry_p(bp->b_addr);
 237         endp = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 238         if (!endp) {
 239                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 240                 goto out_relse;
 241         }
 242         while (p < endp) {
 243                 struct xfs_dir2_data_entry      *dep;
 244                 struct xfs_dir2_data_unused     *dup;
 245 
 246                 dup = (struct xfs_dir2_data_unused *)p;
 247                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 248                         p += be16_to_cpu(dup->length);
 249                         continue;
 250                 }
 251                 dep = (struct xfs_dir2_data_entry *)p;
 252                 if (dep == dent)
 253                         break;
 254                 p += mp->m_dir_inode_ops->data_entsize(dep->namelen);
 255         }
 256         if (p >= endp) {
 257                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 258                 goto out_relse;
 259         }
 260 
 261         /* Retrieve the entry, sanity check it, and compare hashes. */
 262         ino = be64_to_cpu(dent->inumber);
 263         hash = be32_to_cpu(ent->hashval);
 264         tag = be16_to_cpup(dp->d_ops->data_entry_tag_p(dent));
 265         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 266                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 267         if (dent->namelen == 0) {
 268                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 269                 goto out_relse;
 270         }
 271         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 272         if (calc_hash != hash)
 273                 xchk_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 274 
 275 out_relse:
 276         xfs_trans_brelse(ds->dargs.trans, bp);
 277 out:
 278         return error;
 279 }
 280 
 281 /*
 282  * Is this unused entry either in the bestfree or smaller than all of
 283  * them?  We've already checked that the bestfrees are sorted longest to
 284  * shortest, and that there aren't any bogus entries.
 285  */
 286 STATIC void
 287 xchk_directory_check_free_entry(
 288         struct xfs_scrub                *sc,
 289         xfs_dablk_t                     lblk,
 290         struct xfs_dir2_data_free       *bf,
 291         struct xfs_dir2_data_unused     *dup)
 292 {
 293         struct xfs_dir2_data_free       *dfp;
 294         unsigned int                    dup_length;
 295 
 296         dup_length = be16_to_cpu(dup->length);
 297 
 298         /* Unused entry is shorter than any of the bestfrees */
 299         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 300                 return;
 301 
 302         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 303                 if (dup_length == be16_to_cpu(dfp->length))
 304                         return;
 305 
 306         /* Unused entry should be in the bestfrees but wasn't found. */
 307         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 308 }
 309 
 310 /* Check free space info in a directory data block. */
 311 STATIC int
 312 xchk_directory_data_bestfree(
 313         struct xfs_scrub                *sc,
 314         xfs_dablk_t                     lblk,
 315         bool                            is_block)
 316 {
 317         struct xfs_dir2_data_unused     *dup;
 318         struct xfs_dir2_data_free       *dfp;
 319         struct xfs_buf                  *bp;
 320         struct xfs_dir2_data_free       *bf;
 321         struct xfs_mount                *mp = sc->mp;
 322         const struct xfs_dir_ops        *d_ops;
 323         char                            *ptr;
 324         char                            *endptr;
 325         u16                             tag;
 326         unsigned int                    nr_bestfrees = 0;
 327         unsigned int                    nr_frees = 0;
 328         unsigned int                    smallest_bestfree;
 329         int                             newlen;
 330         int                             offset;
 331         int                             error;
 332 
 333         d_ops = sc->ip->d_ops;
 334 
 335         if (is_block) {
 336                 /* dir block format */
 337                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 338                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 339                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 340         } else {
 341                 /* dir data format */
 342                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, -1, &bp);
 343         }
 344         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 345                 goto out;
 346         xchk_buffer_recheck(sc, bp);
 347 
 348         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 349 
 350         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 351                 goto out_buf;
 352 
 353         /* Do the bestfrees correspond to actual free space? */
 354         bf = d_ops->data_bestfree_p(bp->b_addr);
 355         smallest_bestfree = UINT_MAX;
 356         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 357                 offset = be16_to_cpu(dfp->offset);
 358                 if (offset == 0)
 359                         continue;
 360                 if (offset >= mp->m_dir_geo->blksize) {
 361                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 362                         goto out_buf;
 363                 }
 364                 dup = (struct xfs_dir2_data_unused *)(bp->b_addr + offset);
 365                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 366 
 367                 /* bestfree doesn't match the entry it points at? */
 368                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 369                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 370                     tag != ((char *)dup - (char *)bp->b_addr)) {
 371                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 372                         goto out_buf;
 373                 }
 374 
 375                 /* bestfree records should be ordered largest to smallest */
 376                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 377                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 378                         goto out_buf;
 379                 }
 380 
 381                 smallest_bestfree = be16_to_cpu(dfp->length);
 382                 nr_bestfrees++;
 383         }
 384 
 385         /* Make sure the bestfrees are actually the best free spaces. */
 386         ptr = (char *)d_ops->data_entry_p(bp->b_addr);
 387         endptr = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 388 
 389         /* Iterate the entries, stopping when we hit or go past the end. */
 390         while (ptr < endptr) {
 391                 dup = (struct xfs_dir2_data_unused *)ptr;
 392                 /* Skip real entries */
 393                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 394                         struct xfs_dir2_data_entry      *dep;
 395 
 396                         dep = (struct xfs_dir2_data_entry *)ptr;
 397                         newlen = d_ops->data_entsize(dep->namelen);
 398                         if (newlen <= 0) {
 399                                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 400                                                 lblk);
 401                                 goto out_buf;
 402                         }
 403                         ptr += newlen;
 404                         continue;
 405                 }
 406 
 407                 /* Spot check this free entry */
 408                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 409                 if (tag != ((char *)dup - (char *)bp->b_addr)) {
 410                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 411                         goto out_buf;
 412                 }
 413 
 414                 /*
 415                  * Either this entry is a bestfree or it's smaller than
 416                  * any of the bestfrees.
 417                  */
 418                 xchk_directory_check_free_entry(sc, lblk, bf, dup);
 419                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 420                         goto out_buf;
 421 
 422                 /* Move on. */
 423                 newlen = be16_to_cpu(dup->length);
 424                 if (newlen <= 0) {
 425                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 426                         goto out_buf;
 427                 }
 428                 ptr += newlen;
 429                 if (ptr <= endptr)
 430                         nr_frees++;
 431         }
 432 
 433         /* We're required to fill all the space. */
 434         if (ptr != endptr)
 435                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 436 
 437         /* Did we see at least as many free slots as there are bestfrees? */
 438         if (nr_frees < nr_bestfrees)
 439                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 440 out_buf:
 441         xfs_trans_brelse(sc->tp, bp);
 442 out:
 443         return error;
 444 }
 445 
 446 /*
 447  * Does the free space length in the free space index block ($len) match
 448  * the longest length in the directory data block's bestfree array?
 449  * Assume that we've already checked that the data block's bestfree
 450  * array is in order.
 451  */
 452 STATIC void
 453 xchk_directory_check_freesp(
 454         struct xfs_scrub                *sc,
 455         xfs_dablk_t                     lblk,
 456         struct xfs_buf                  *dbp,
 457         unsigned int                    len)
 458 {
 459         struct xfs_dir2_data_free       *dfp;
 460 
 461         dfp = sc->ip->d_ops->data_bestfree_p(dbp->b_addr);
 462 
 463         if (len != be16_to_cpu(dfp->length))
 464                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 465 
 466         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 467                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 468 }
 469 
 470 /* Check free space info in a directory leaf1 block. */
 471 STATIC int
 472 xchk_directory_leaf1_bestfree(
 473         struct xfs_scrub                *sc,
 474         struct xfs_da_args              *args,
 475         xfs_dablk_t                     lblk)
 476 {
 477         struct xfs_dir3_icleaf_hdr      leafhdr;
 478         struct xfs_dir2_leaf_entry      *ents;
 479         struct xfs_dir2_leaf_tail       *ltp;
 480         struct xfs_dir2_leaf            *leaf;
 481         struct xfs_buf                  *dbp;
 482         struct xfs_buf                  *bp;
 483         const struct xfs_dir_ops        *d_ops = sc->ip->d_ops;
 484         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 485         __be16                          *bestp;
 486         __u16                           best;
 487         __u32                           hash;
 488         __u32                           lasthash = 0;
 489         __u32                           bestcount;
 490         unsigned int                    stale = 0;
 491         int                             i;
 492         int                             error;
 493 
 494         /* Read the free space block. */
 495         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, -1, &bp);
 496         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 497                 goto out;
 498         xchk_buffer_recheck(sc, bp);
 499 
 500         leaf = bp->b_addr;
 501         d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 502         ents = d_ops->leaf_ents_p(leaf);
 503         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 504         bestcount = be32_to_cpu(ltp->bestcount);
 505         bestp = xfs_dir2_leaf_bests_p(ltp);
 506 
 507         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 508                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 509 
 510                 if (hdr3->pad != cpu_to_be32(0))
 511                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 512         }
 513 
 514         /*
 515          * There should be as many bestfree slots as there are dir data
 516          * blocks that can fit under i_size.
 517          */
 518         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 519                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 520                 goto out;
 521         }
 522 
 523         /* Is the leaf count even remotely sane? */
 524         if (leafhdr.count > d_ops->leaf_max_ents(geo)) {
 525                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 526                 goto out;
 527         }
 528 
 529         /* Leaves and bests don't overlap in leaf format. */
 530         if ((char *)&ents[leafhdr.count] > (char *)bestp) {
 531                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 532                 goto out;
 533         }
 534 
 535         /* Check hash value order, count stale entries.  */
 536         for (i = 0; i < leafhdr.count; i++) {
 537                 hash = be32_to_cpu(ents[i].hashval);
 538                 if (i > 0 && lasthash > hash)
 539                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 540                 lasthash = hash;
 541                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 542                         stale++;
 543         }
 544         if (leafhdr.stale != stale)
 545                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 546         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 547                 goto out;
 548 
 549         /* Check all the bestfree entries. */
 550         for (i = 0; i < bestcount; i++, bestp++) {
 551                 best = be16_to_cpu(*bestp);
 552                 if (best == NULLDATAOFF)
 553                         continue;
 554                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 555                                 i * args->geo->fsbcount, -1, &dbp);
 556                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 557                                 &error))
 558                         break;
 559                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 560                 xfs_trans_brelse(sc->tp, dbp);
 561                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 562                         goto out;
 563         }
 564 out:
 565         return error;
 566 }
 567 
 568 /* Check free space info in a directory freespace block. */
 569 STATIC int
 570 xchk_directory_free_bestfree(
 571         struct xfs_scrub                *sc,
 572         struct xfs_da_args              *args,
 573         xfs_dablk_t                     lblk)
 574 {
 575         struct xfs_dir3_icfree_hdr      freehdr;
 576         struct xfs_buf                  *dbp;
 577         struct xfs_buf                  *bp;
 578         __be16                          *bestp;
 579         __u16                           best;
 580         unsigned int                    stale = 0;
 581         int                             i;
 582         int                             error;
 583 
 584         /* Read the free space block */
 585         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 586         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 587                 goto out;
 588         xchk_buffer_recheck(sc, bp);
 589 
 590         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 591                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 592 
 593                 if (hdr3->pad != cpu_to_be32(0))
 594                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 595         }
 596 
 597         /* Check all the entries. */
 598         sc->ip->d_ops->free_hdr_from_disk(&freehdr, bp->b_addr);
 599         bestp = sc->ip->d_ops->free_bests_p(bp->b_addr);
 600         for (i = 0; i < freehdr.nvalid; i++, bestp++) {
 601                 best = be16_to_cpu(*bestp);
 602                 if (best == NULLDATAOFF) {
 603                         stale++;
 604                         continue;
 605                 }
 606                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 607                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 608                                 -1, &dbp);
 609                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 610                                 &error))
 611                         break;
 612                 xchk_directory_check_freesp(sc, lblk, dbp, best);
 613                 xfs_trans_brelse(sc->tp, dbp);
 614         }
 615 
 616         if (freehdr.nused + stale != freehdr.nvalid)
 617                 xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 618 out:
 619         return error;
 620 }
 621 
 622 /* Check free space information in directories. */
 623 STATIC int
 624 xchk_directory_blocks(
 625         struct xfs_scrub        *sc)
 626 {
 627         struct xfs_bmbt_irec    got;
 628         struct xfs_da_args      args;
 629         struct xfs_ifork        *ifp;
 630         struct xfs_mount        *mp = sc->mp;
 631         xfs_fileoff_t           leaf_lblk;
 632         xfs_fileoff_t           free_lblk;
 633         xfs_fileoff_t           lblk;
 634         struct xfs_iext_cursor  icur;
 635         xfs_dablk_t             dabno;
 636         bool                    found;
 637         int                     is_block = 0;
 638         int                     error;
 639 
 640         /* Ignore local format directories. */
 641         if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 642             sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
 643                 return 0;
 644 
 645         ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 646         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 647         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 648         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 649 
 650         /* Is this a block dir? */
 651         args.dp = sc->ip;
 652         args.geo = mp->m_dir_geo;
 653         args.trans = sc->tp;
 654         error = xfs_dir2_isblock(&args, &is_block);
 655         if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 656                 goto out;
 657 
 658         /* Iterate all the data extents in the directory... */
 659         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 660         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 661                 /* Block directories only have a single block at offset 0. */
 662                 if (is_block &&
 663                     (got.br_startoff > 0 ||
 664                      got.br_blockcount != args.geo->fsbcount)) {
 665                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK,
 666                                         got.br_startoff);
 667                         break;
 668                 }
 669 
 670                 /* No more data blocks... */
 671                 if (got.br_startoff >= leaf_lblk)
 672                         break;
 673 
 674                 /*
 675                  * Check each data block's bestfree data.
 676                  *
 677                  * Iterate all the fsbcount-aligned block offsets in
 678                  * this directory.  The directory block reading code is
 679                  * smart enough to do its own bmap lookups to handle
 680                  * discontiguous directory blocks.  When we're done
 681                  * with the extent record, re-query the bmap at the
 682                  * next fsbcount-aligned offset to avoid redundant
 683                  * block checks.
 684                  */
 685                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 686                                 args.geo->fsbcount);
 687                      lblk < got.br_startoff + got.br_blockcount;
 688                      lblk += args.geo->fsbcount) {
 689                         error = xchk_directory_data_bestfree(sc, lblk,
 690                                         is_block);
 691                         if (error)
 692                                 goto out;
 693                 }
 694                 dabno = got.br_startoff + got.br_blockcount;
 695                 lblk = roundup(dabno, args.geo->fsbcount);
 696                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 697         }
 698 
 699         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 700                 goto out;
 701 
 702         /* Look for a leaf1 block, which has free info. */
 703         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 704             got.br_startoff == leaf_lblk &&
 705             got.br_blockcount == args.geo->fsbcount &&
 706             !xfs_iext_next_extent(ifp, &icur, &got)) {
 707                 if (is_block) {
 708                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 709                         goto out;
 710                 }
 711                 error = xchk_directory_leaf1_bestfree(sc, &args,
 712                                 leaf_lblk);
 713                 if (error)
 714                         goto out;
 715         }
 716 
 717         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 718                 goto out;
 719 
 720         /* Scan for free blocks */
 721         lblk = free_lblk;
 722         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 723         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 724                 /*
 725                  * Dirs can't have blocks mapped above 2^32.
 726                  * Single-block dirs shouldn't even be here.
 727                  */
 728                 lblk = got.br_startoff;
 729                 if (lblk & ~0xFFFFFFFFULL) {
 730                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 731                         goto out;
 732                 }
 733                 if (is_block) {
 734                         xchk_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 735                         goto out;
 736                 }
 737 
 738                 /*
 739                  * Check each dir free block's bestfree data.
 740                  *
 741                  * Iterate all the fsbcount-aligned block offsets in
 742                  * this directory.  The directory block reading code is
 743                  * smart enough to do its own bmap lookups to handle
 744                  * discontiguous directory blocks.  When we're done
 745                  * with the extent record, re-query the bmap at the
 746                  * next fsbcount-aligned offset to avoid redundant
 747                  * block checks.
 748                  */
 749                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 750                                 args.geo->fsbcount);
 751                      lblk < got.br_startoff + got.br_blockcount;
 752                      lblk += args.geo->fsbcount) {
 753                         error = xchk_directory_free_bestfree(sc, &args,
 754                                         lblk);
 755                         if (error)
 756                                 goto out;
 757                 }
 758                 dabno = got.br_startoff + got.br_blockcount;
 759                 lblk = roundup(dabno, args.geo->fsbcount);
 760                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 761         }
 762 out:
 763         return error;
 764 }
 765 
 766 /* Scrub a whole directory. */
 767 int
 768 xchk_directory(
 769         struct xfs_scrub        *sc)
 770 {
 771         struct xchk_dir_ctx     sdc = {
 772                 .dir_iter.actor = xchk_dir_actor,
 773                 .dir_iter.pos = 0,
 774                 .sc = sc,
 775         };
 776         size_t                  bufsize;
 777         loff_t                  oldpos;
 778         int                     error = 0;
 779 
 780         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 781                 return -ENOENT;
 782 
 783         /* Plausible size? */
 784         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 785                 xchk_ino_set_corrupt(sc, sc->ip->i_ino);
 786                 goto out;
 787         }
 788 
 789         /* Check directory tree structure */
 790         error = xchk_da_btree(sc, XFS_DATA_FORK, xchk_dir_rec, NULL);
 791         if (error)
 792                 return error;
 793 
 794         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 795                 return error;
 796 
 797         /* Check the freespace. */
 798         error = xchk_directory_blocks(sc);
 799         if (error)
 800                 return error;
 801 
 802         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 803                 return error;
 804 
 805         /*
 806          * Check that every dirent we see can also be looked up by hash.
 807          * Userspace usually asks for a 32k buffer, so we will too.
 808          */
 809         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 810                         sc->ip->i_d.di_size);
 811 
 812         /*
 813          * Look up every name in this directory by hash.
 814          *
 815          * Use the xfs_readdir function to call xchk_dir_actor on
 816          * every directory entry in this directory.  In _actor, we check
 817          * the name, inode number, and ftype (if applicable) of the
 818          * entry.  xfs_readdir uses the VFS filldir functions to provide
 819          * iteration context.
 820          *
 821          * The VFS grabs a read or write lock via i_rwsem before it reads
 822          * or writes to a directory.  If we've gotten this far we've
 823          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 824          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 825          * to drop the ILOCK here in order to reuse the _readdir and
 826          * _dir_lookup routines, which do their own ILOCK locking.
 827          */
 828         oldpos = 0;
 829         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 830         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 831         while (true) {
 832                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 833                 if (!xchk_fblock_process_error(sc, XFS_DATA_FORK, 0,
 834                                 &error))
 835                         goto out;
 836                 if (oldpos == sdc.dir_iter.pos)
 837                         break;
 838                 oldpos = sdc.dir_iter.pos;
 839         }
 840 
 841 out:
 842         return error;
 843 }