fs/xfs/scrub/dir.c

   1 /*
   2  * Copyright (C) 2017 Oracle.  All Rights Reserved.
   3  *
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version 2
   9  * of the License, or (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it would be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write the Free Software Foundation,
  18  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
  19  */
  20 #include "xfs.h"
  21 #include "xfs_fs.h"
  22 #include "xfs_shared.h"
  23 #include "xfs_format.h"
  24 #include "xfs_trans_resv.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_defer.h"
  27 #include "xfs_btree.h"
  28 #include "xfs_bit.h"
  29 #include "xfs_log_format.h"
  30 #include "xfs_trans.h"
  31 #include "xfs_sb.h"
  32 #include "xfs_inode.h"
  33 #include "xfs_icache.h"
  34 #include "xfs_itable.h"
  35 #include "xfs_da_format.h"
  36 #include "xfs_da_btree.h"
  37 #include "xfs_dir2.h"
  38 #include "xfs_dir2_priv.h"
  39 #include "xfs_ialloc.h"
  40 #include "scrub/xfs_scrub.h"
  41 #include "scrub/scrub.h"
  42 #include "scrub/common.h"
  43 #include "scrub/trace.h"
  44 #include "scrub/dabtree.h"
  45
  46 /* Set us up to scrub directories. */
  47 int
  48 xfs_scrub_setup_directory(
  49         struct xfs_scrub_context        *sc,
  50         struct xfs_inode                *ip)
  51 {
  52         return xfs_scrub_setup_inode_contents(sc, ip, 0);
  53 }
  54
  55 /* Directories */
  56
  57 /* Scrub a directory entry. */
  58
  59 struct xfs_scrub_dir_ctx {
  60         /* VFS fill-directory iterator */
  61         struct dir_context              dir_iter;
  62
  63         struct xfs_scrub_context        *sc;
  64 };
  65
  66 /* Check that an inode's mode matches a given DT_ type. */
  67 STATIC int
  68 xfs_scrub_dir_check_ftype(
  69         struct xfs_scrub_dir_ctx        *sdc,
  70         xfs_fileoff_t                   offset,
  71         xfs_ino_t                       inum,
  72         int                             dtype)
  73 {
  74         struct xfs_mount                *mp = sdc->sc->mp;
  75         struct xfs_inode                *ip;
  76         int                             ino_dtype;
  77         int                             error = 0;
  78
  79         if (!xfs_sb_version_hasftype(&mp->m_sb)) {
  80                 if (dtype != DT_UNKNOWN && dtype != DT_DIR)
  81                         xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
  82                                         offset);
  83                 goto out;
  84         }
  85
  86         /*
  87          * Grab the inode pointed to by the dirent.  We release the
  88          * inode before we cancel the scrub transaction.  Since we're
  89          * don't know a priori that releasing the inode won't trigger
  90          * eofblocks cleanup (which allocates what would be a nested
  91          * transaction), we can't use DONTCACHE here because DONTCACHE
  92          * inodes can trigger immediate inactive cleanup of the inode.
  93          */
  94         error = xfs_iget(mp, sdc->sc->tp, inum, 0, 0, &ip);
  95         if (!xfs_scrub_fblock_xref_process_error(sdc->sc, XFS_DATA_FORK, offset,
  96                         &error))
  97                 goto out;
  98
  99         /* Convert mode to the DT_* values that dir_emit uses. */
 100         ino_dtype = xfs_dir3_get_dtype(mp,
 101                         xfs_mode_to_ftype(VFS_I(ip)->i_mode));
 102         if (ino_dtype != dtype)
 103                 xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 104         iput(VFS_I(ip));
 105 out:
 106         return error;
 107 }
 108
 109 /*
 110  * Scrub a single directory entry.
 111  *
 112  * We use the VFS directory iterator (i.e. readdir) to call this
 113  * function for every directory entry in a directory.  Once we're here,
 114  * we check the inode number to make sure it's sane, then we check that
 115  * we can look up this filename.  Finally, we check the ftype.
 116  */
 117 STATIC int
 118 xfs_scrub_dir_actor(
 119         struct dir_context              *dir_iter,
 120         const char                      *name,
 121         int                             namelen,
 122         loff_t                          pos,
 123         u64                             ino,
 124         unsigned                        type)
 125 {
 126         struct xfs_mount                *mp;
 127         struct xfs_inode                *ip;
 128         struct xfs_scrub_dir_ctx        *sdc;
 129         struct xfs_name                 xname;
 130         xfs_ino_t                       lookup_ino;
 131         xfs_dablk_t                     offset;
 132         int                             error = 0;
 133
 134         sdc = container_of(dir_iter, struct xfs_scrub_dir_ctx, dir_iter);
 135         ip = sdc->sc->ip;
 136         mp = ip->i_mount;
 137         offset = xfs_dir2_db_to_da(mp->m_dir_geo,
 138                         xfs_dir2_dataptr_to_db(mp->m_dir_geo, pos));
 139
 140         /* Does this inode number make sense? */
 141         if (!xfs_verify_dir_ino(mp, ino)) {
 142                 xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK, offset);
 143                 goto out;
 144         }
 145
 146         if (!strncmp(".", name, namelen)) {
 147                 /* If this is "." then check that the inum matches the dir. */
 148                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 149                         xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 150                                         offset);
 151                 if (ino != ip->i_ino)
 152                         xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 153                                         offset);
 154         } else if (!strncmp("..", name, namelen)) {
 155                 /*
 156                  * If this is ".." in the root inode, check that the inum
 157                  * matches this dir.
 158                  */
 159                 if (xfs_sb_version_hasftype(&mp->m_sb) && type != DT_DIR)
 160                         xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 161                                         offset);
 162                 if (ip->i_ino == mp->m_sb.sb_rootino && ino != ip->i_ino)
 163                         xfs_scrub_fblock_set_corrupt(sdc->sc, XFS_DATA_FORK,
 164                                         offset);
 165         }
 166
 167         /* Verify that we can look up this name by hash. */
 168         xname.name = name;
 169         xname.len = namelen;
 170         xname.type = XFS_DIR3_FT_UNKNOWN;
 171
 172         error = xfs_dir_lookup(sdc->sc->tp, ip, &xname, &lookup_ino, NULL);
 173         if (!xfs_scrub_fblock_process_error(sdc->sc, XFS_DATA_FORK, offset,
 174                         &error))
 175                 goto fail_xref;
 176         if (lookup_ino != ino) {
 177                 xfs_scrub_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         error = xfs_scrub_dir_check_ftype(sdc, offset, lookup_ino, type);
 183         if (error)
 184                 goto out;
 185 out:
 186         return error;
 187 fail_xref:
 188         return error;
 189 }
 190
 191 /* Scrub a directory btree record. */
 192 STATIC int
 193 xfs_scrub_dir_rec(
 194         struct xfs_scrub_da_btree       *ds,
 195         int                             level,
 196         void                            *rec)
 197 {
 198         struct xfs_mount                *mp = ds->state->mp;
 199         struct xfs_dir2_leaf_entry      *ent = rec;
 200         struct xfs_inode                *dp = ds->dargs.dp;
 201         struct xfs_dir2_data_entry      *dent;
 202         struct xfs_buf                  *bp;
 203         char                            *p, *endp;
 204         xfs_ino_t                       ino;
 205         xfs_dablk_t                     rec_bno;
 206         xfs_dir2_db_t                   db;
 207         xfs_dir2_data_aoff_t            off;
 208         xfs_dir2_dataptr_t              ptr;
 209         xfs_dahash_t                    calc_hash;
 210         xfs_dahash_t                    hash;
 211         unsigned int                    tag;
 212         int                             error;
 213
 214         /* Check the hash of the entry. */
 215         error = xfs_scrub_da_btree_hash(ds, level, &ent->hashval);
 216         if (error)
 217                 goto out;
 218
 219         /* Valid hash pointer? */
 220         ptr = be32_to_cpu(ent->address);
 221         if (ptr == 0)
 222                 return 0;
 223
 224         /* Find the directory entry's location. */
 225         db = xfs_dir2_dataptr_to_db(mp->m_dir_geo, ptr);
 226         off = xfs_dir2_dataptr_to_off(mp->m_dir_geo, ptr);
 227         rec_bno = xfs_dir2_db_to_da(mp->m_dir_geo, db);
 228
 229         if (rec_bno >= mp->m_dir_geo->leafblk) {
 230                 xfs_scrub_da_set_corrupt(ds, level);
 231                 goto out;
 232         }
 233         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
 234         if (!xfs_scrub_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 235                         &error))
 236                 goto out;
 237         if (!bp) {
 238                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 239                 goto out;
 240         }
 241         xfs_scrub_buffer_recheck(ds->sc, bp);
 242
 243         dent = (struct xfs_dir2_data_entry *)(((char *)bp->b_addr) + off);
 244
 245         /* Make sure we got a real directory entry. */
 246         p = (char *)mp->m_dir_inode_ops->data_entry_p(bp->b_addr);
 247         endp = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 248         if (!endp) {
 249                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 250                 goto out_relse;
 251         }
 252         while (p < endp) {
 253                 struct xfs_dir2_data_entry      *dep;
 254                 struct xfs_dir2_data_unused     *dup;
 255
 256                 dup = (struct xfs_dir2_data_unused *)p;
 257                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 258                         p += be16_to_cpu(dup->length);
 259                         continue;
 260                 }
 261                 dep = (struct xfs_dir2_data_entry *)p;
 262                 if (dep == dent)
 263                         break;
 264                 p += mp->m_dir_inode_ops->data_entsize(dep->namelen);
 265         }
 266         if (p >= endp) {
 267                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 268                 goto out_relse;
 269         }
 270
 271         /* Retrieve the entry, sanity check it, and compare hashes. */
 272         ino = be64_to_cpu(dent->inumber);
 273         hash = be32_to_cpu(ent->hashval);
 274         tag = be16_to_cpup(dp->d_ops->data_entry_tag_p(dent));
 275         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 276                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 277         if (dent->namelen == 0) {
 278                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 279                 goto out_relse;
 280         }
 281         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 282         if (calc_hash != hash)
 283                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 284
 285 out_relse:
 286         xfs_trans_brelse(ds->dargs.trans, bp);
 287 out:
 288         return error;
 289 }
 290
 291 /*
 292  * Is this unused entry either in the bestfree or smaller than all of
 293  * them?  We've already checked that the bestfrees are sorted longest to
 294  * shortest, and that there aren't any bogus entries.
 295  */
 296 STATIC void
 297 xfs_scrub_directory_check_free_entry(
 298         struct xfs_scrub_context        *sc,
 299         xfs_dablk_t                     lblk,
 300         struct xfs_dir2_data_free       *bf,
 301         struct xfs_dir2_data_unused     *dup)
 302 {
 303         struct xfs_dir2_data_free       *dfp;
 304         unsigned int                    dup_length;
 305
 306         dup_length = be16_to_cpu(dup->length);
 307
 308         /* Unused entry is shorter than any of the bestfrees */
 309         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 310                 return;
 311
 312         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 313                 if (dup_length == be16_to_cpu(dfp->length))
 314                         return;
 315
 316         /* Unused entry should be in the bestfrees but wasn't found. */
 317         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 318 }
 319
 320 /* Check free space info in a directory data block. */
 321 STATIC int
 322 xfs_scrub_directory_data_bestfree(
 323         struct xfs_scrub_context        *sc,
 324         xfs_dablk_t                     lblk,
 325         bool                            is_block)
 326 {
 327         struct xfs_dir2_data_unused     *dup;
 328         struct xfs_dir2_data_free       *dfp;
 329         struct xfs_buf                  *bp;
 330         struct xfs_dir2_data_free       *bf;
 331         struct xfs_mount                *mp = sc->mp;
 332         const struct xfs_dir_ops        *d_ops;
 333         char                            *ptr;
 334         char                            *endptr;
 335         u16                             tag;
 336         unsigned int                    nr_bestfrees = 0;
 337         unsigned int                    nr_frees = 0;
 338         unsigned int                    smallest_bestfree;
 339         int                             newlen;
 340         int                             offset;
 341         int                             error;
 342
 343         d_ops = sc->ip->d_ops;
 344
 345         if (is_block) {
 346                 /* dir block format */
 347                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 348                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 349                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 350         } else {
 351                 /* dir data format */
 352                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, -1, &bp);
 353         }
 354         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 355                 goto out;
 356         xfs_scrub_buffer_recheck(sc, bp);
 357
 358         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 359
 360         /* Do the bestfrees correspond to actual free space? */
 361         bf = d_ops->data_bestfree_p(bp->b_addr);
 362         smallest_bestfree = UINT_MAX;
 363         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 364                 offset = be16_to_cpu(dfp->offset);
 365                 if (offset == 0)
 366                         continue;
 367                 if (offset >= mp->m_dir_geo->blksize) {
 368                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 369                         goto out_buf;
 370                 }
 371                 dup = (struct xfs_dir2_data_unused *)(bp->b_addr + offset);
 372                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 373
 374                 /* bestfree doesn't match the entry it points at? */
 375                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 376                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 377                     tag != ((char *)dup - (char *)bp->b_addr)) {
 378                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 379                         goto out_buf;
 380                 }
 381
 382                 /* bestfree records should be ordered largest to smallest */
 383                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 384                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 385                         goto out_buf;
 386                 }
 387
 388                 smallest_bestfree = be16_to_cpu(dfp->length);
 389                 nr_bestfrees++;
 390         }
 391
 392         /* Make sure the bestfrees are actually the best free spaces. */
 393         ptr = (char *)d_ops->data_entry_p(bp->b_addr);
 394         endptr = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 395
 396         /* Iterate the entries, stopping when we hit or go past the end. */
 397         while (ptr < endptr) {
 398                 dup = (struct xfs_dir2_data_unused *)ptr;
 399                 /* Skip real entries */
 400                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 401                         struct xfs_dir2_data_entry      *dep;
 402
 403                         dep = (struct xfs_dir2_data_entry *)ptr;
 404                         newlen = d_ops->data_entsize(dep->namelen);
 405                         if (newlen <= 0) {
 406                                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK,
 407                                                 lblk);
 408                                 goto out_buf;
 409                         }
 410                         ptr += newlen;
 411                         continue;
 412                 }
 413
 414                 /* Spot check this free entry */
 415                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 416                 if (tag != ((char *)dup - (char *)bp->b_addr))
 417                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 418
 419                 /*
 420                  * Either this entry is a bestfree or it's smaller than
 421                  * any of the bestfrees.
 422                  */
 423                 xfs_scrub_directory_check_free_entry(sc, lblk, bf, dup);
 424
 425                 /* Move on. */
 426                 newlen = be16_to_cpu(dup->length);
 427                 if (newlen <= 0) {
 428                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 429                         goto out_buf;
 430                 }
 431                 ptr += newlen;
 432                 if (ptr <= endptr)
 433                         nr_frees++;
 434         }
 435
 436         /* We're required to fill all the space. */
 437         if (ptr != endptr)
 438                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 439
 440         /* Did we see at least as many free slots as there are bestfrees? */
 441         if (nr_frees < nr_bestfrees)
 442                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 443 out_buf:
 444         xfs_trans_brelse(sc->tp, bp);
 445 out:
 446         return error;
 447 }
 448
 449 /*
 450  * Does the free space length in the free space index block ($len) match
 451  * the longest length in the directory data block's bestfree array?
 452  * Assume that we've already checked that the data block's bestfree
 453  * array is in order.
 454  */
 455 STATIC void
 456 xfs_scrub_directory_check_freesp(
 457         struct xfs_scrub_context        *sc,
 458         xfs_dablk_t                     lblk,
 459         struct xfs_buf                  *dbp,
 460         unsigned int                    len)
 461 {
 462         struct xfs_dir2_data_free       *dfp;
 463
 464         dfp = sc->ip->d_ops->data_bestfree_p(dbp->b_addr);
 465
 466         if (len != be16_to_cpu(dfp->length))
 467                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 468
 469         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 470                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 471 }
 472
 473 /* Check free space info in a directory leaf1 block. */
 474 STATIC int
 475 xfs_scrub_directory_leaf1_bestfree(
 476         struct xfs_scrub_context        *sc,
 477         struct xfs_da_args              *args,
 478         xfs_dablk_t                     lblk)
 479 {
 480         struct xfs_dir3_icleaf_hdr      leafhdr;
 481         struct xfs_dir2_leaf_entry      *ents;
 482         struct xfs_dir2_leaf_tail       *ltp;
 483         struct xfs_dir2_leaf            *leaf;
 484         struct xfs_buf                  *dbp;
 485         struct xfs_buf                  *bp;
 486         const struct xfs_dir_ops        *d_ops = sc->ip->d_ops;
 487         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 488         __be16                          *bestp;
 489         __u16                           best;
 490         __u32                           hash;
 491         __u32                           lasthash = 0;
 492         __u32                           bestcount;
 493         unsigned int                    stale = 0;
 494         int                             i;
 495         int                             error;
 496
 497         /* Read the free space block. */
 498         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, -1, &bp);
 499         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 500                 goto out;
 501         xfs_scrub_buffer_recheck(sc, bp);
 502
 503         leaf = bp->b_addr;
 504         d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 505         ents = d_ops->leaf_ents_p(leaf);
 506         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 507         bestcount = be32_to_cpu(ltp->bestcount);
 508         bestp = xfs_dir2_leaf_bests_p(ltp);
 509
 510         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 511                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 512
 513                 if (hdr3->pad != cpu_to_be32(0))
 514                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 515         }
 516
 517         /*
 518          * There should be as many bestfree slots as there are dir data
 519          * blocks that can fit under i_size.
 520          */
 521         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 522                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 523                 goto out;
 524         }
 525
 526         /* Is the leaf count even remotely sane? */
 527         if (leafhdr.count > d_ops->leaf_max_ents(geo)) {
 528                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 529                 goto out;
 530         }
 531
 532         /* Leaves and bests don't overlap in leaf format. */
 533         if ((char *)&ents[leafhdr.count] > (char *)bestp) {
 534                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 535                 goto out;
 536         }
 537
 538         /* Check hash value order, count stale entries.  */
 539         for (i = 0; i < leafhdr.count; i++) {
 540                 hash = be32_to_cpu(ents[i].hashval);
 541                 if (i > 0 && lasthash > hash)
 542                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 543                 lasthash = hash;
 544                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 545                         stale++;
 546         }
 547         if (leafhdr.stale != stale)
 548                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 549
 550         /* Check all the bestfree entries. */
 551         for (i = 0; i < bestcount; i++, bestp++) {
 552                 best = be16_to_cpu(*bestp);
 553                 if (best == NULLDATAOFF)
 554                         continue;
 555                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 556                                 i * args->geo->fsbcount, -1, &dbp);
 557                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 558                                 &error))
 559                         continue;
 560                 xfs_scrub_directory_check_freesp(sc, lblk, dbp, best);
 561                 xfs_trans_brelse(sc->tp, dbp);
 562         }
 563 out:
 564         return error;
 565 }
 566
 567 /* Check free space info in a directory freespace block. */
 568 STATIC int
 569 xfs_scrub_directory_free_bestfree(
 570         struct xfs_scrub_context        *sc,
 571         struct xfs_da_args              *args,
 572         xfs_dablk_t                     lblk)
 573 {
 574         struct xfs_dir3_icfree_hdr      freehdr;
 575         struct xfs_buf                  *dbp;
 576         struct xfs_buf                  *bp;
 577         __be16                          *bestp;
 578         __u16                           best;
 579         unsigned int                    stale = 0;
 580         int                             i;
 581         int                             error;
 582
 583         /* Read the free space block */
 584         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 585         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 586                 goto out;
 587         xfs_scrub_buffer_recheck(sc, bp);
 588
 589         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 590                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 591
 592                 if (hdr3->pad != cpu_to_be32(0))
 593                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 594         }
 595
 596         /* Check all the entries. */
 597         sc->ip->d_ops->free_hdr_from_disk(&freehdr, bp->b_addr);
 598         bestp = sc->ip->d_ops->free_bests_p(bp->b_addr);
 599         for (i = 0; i < freehdr.nvalid; i++, bestp++) {
 600                 best = be16_to_cpu(*bestp);
 601                 if (best == NULLDATAOFF) {
 602                         stale++;
 603                         continue;
 604                 }
 605                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 606                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 607                                 -1, &dbp);
 608                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 609                                 &error))
 610                         continue;
 611                 xfs_scrub_directory_check_freesp(sc, lblk, dbp, best);
 612                 xfs_trans_brelse(sc->tp, dbp);
 613         }
 614
 615         if (freehdr.nused + stale != freehdr.nvalid)
 616                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 617 out:
 618         return error;
 619 }
 620
 621 /* Check free space information in directories. */
 622 STATIC int
 623 xfs_scrub_directory_blocks(
 624         struct xfs_scrub_context        *sc)
 625 {
 626         struct xfs_bmbt_irec            got;
 627         struct xfs_da_args              args;
 628         struct xfs_ifork                *ifp;
 629         struct xfs_mount                *mp = sc->mp;
 630         xfs_fileoff_t                   leaf_lblk;
 631         xfs_fileoff_t                   free_lblk;
 632         xfs_fileoff_t                   lblk;
 633         struct xfs_iext_cursor          icur;
 634         xfs_dablk_t                     dabno;
 635         bool                            found;
 636         int                             is_block = 0;
 637         int                             error;
 638
 639         /* Ignore local format directories. */
 640         if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 641             sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
 642                 return 0;
 643
 644         ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 645         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 646         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 647         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 648
 649         /* Is this a block dir? */
 650         args.dp = sc->ip;
 651         args.geo = mp->m_dir_geo;
 652         args.trans = sc->tp;
 653         error = xfs_dir2_isblock(&args, &is_block);
 654         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 655                 goto out;
 656
 657         /* Iterate all the data extents in the directory... */
 658         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 659         while (found) {
 660                 /* Block directories only have a single block at offset 0. */
 661                 if (is_block &&
 662                     (got.br_startoff > 0 ||
 663                      got.br_blockcount != args.geo->fsbcount)) {
 664                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK,
 665                                         got.br_startoff);
 666                         break;
 667                 }
 668
 669                 /* No more data blocks... */
 670                 if (got.br_startoff >= leaf_lblk)
 671                         break;
 672
 673                 /*
 674                  * Check each data block's bestfree data.
 675                  *
 676                  * Iterate all the fsbcount-aligned block offsets in
 677                  * this directory.  The directory block reading code is
 678                  * smart enough to do its own bmap lookups to handle
 679                  * discontiguous directory blocks.  When we're done
 680                  * with the extent record, re-query the bmap at the
 681                  * next fsbcount-aligned offset to avoid redundant
 682                  * block checks.
 683                  */
 684                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 685                                 args.geo->fsbcount);
 686                      lblk < got.br_startoff + got.br_blockcount;
 687                      lblk += args.geo->fsbcount) {
 688                         error = xfs_scrub_directory_data_bestfree(sc, lblk,
 689                                         is_block);
 690                         if (error)
 691                                 goto out;
 692                 }
 693                 dabno = got.br_startoff + got.br_blockcount;
 694                 lblk = roundup(dabno, args.geo->fsbcount);
 695                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 696         }
 697
 698         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 699                 goto out;
 700
 701         /* Look for a leaf1 block, which has free info. */
 702         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 703             got.br_startoff == leaf_lblk &&
 704             got.br_blockcount == args.geo->fsbcount &&
 705             !xfs_iext_next_extent(ifp, &icur, &got)) {
 706                 if (is_block) {
 707                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 708                         goto out;
 709                 }
 710                 error = xfs_scrub_directory_leaf1_bestfree(sc, &args,
 711                                 leaf_lblk);
 712                 if (error)
 713                         goto out;
 714         }
 715
 716         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 717                 goto out;
 718
 719         /* Scan for free blocks */
 720         lblk = free_lblk;
 721         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 722         while (found) {
 723                 /*
 724                  * Dirs can't have blocks mapped above 2^32.
 725                  * Single-block dirs shouldn't even be here.
 726                  */
 727                 lblk = got.br_startoff;
 728                 if (lblk & ~0xFFFFFFFFULL) {
 729                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 730                         goto out;
 731                 }
 732                 if (is_block) {
 733                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 734                         goto out;
 735                 }
 736
 737                 /*
 738                  * Check each dir free block's bestfree data.
 739                  *
 740                  * Iterate all the fsbcount-aligned block offsets in
 741                  * this directory.  The directory block reading code is
 742                  * smart enough to do its own bmap lookups to handle
 743                  * discontiguous directory blocks.  When we're done
 744                  * with the extent record, re-query the bmap at the
 745                  * next fsbcount-aligned offset to avoid redundant
 746                  * block checks.
 747                  */
 748                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 749                                 args.geo->fsbcount);
 750                      lblk < got.br_startoff + got.br_blockcount;
 751                      lblk += args.geo->fsbcount) {
 752                         error = xfs_scrub_directory_free_bestfree(sc, &args,
 753                                         lblk);
 754                         if (error)
 755                                 goto out;
 756                 }
 757                 dabno = got.br_startoff + got.br_blockcount;
 758                 lblk = roundup(dabno, args.geo->fsbcount);
 759                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 760         }
 761 out:
 762         return error;
 763 }
 764
 765 /* Scrub a whole directory. */
 766 int
 767 xfs_scrub_directory(
 768         struct xfs_scrub_context        *sc)
 769 {
 770         struct xfs_scrub_dir_ctx        sdc = {
 771                 .dir_iter.actor = xfs_scrub_dir_actor,
 772                 .dir_iter.pos = 0,
 773                 .sc = sc,
 774         };
 775         size_t                          bufsize;
 776         loff_t                          oldpos;
 777         int                             error = 0;
 778
 779         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 780                 return -ENOENT;
 781
 782         /* Plausible size? */
 783         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 784                 xfs_scrub_ino_set_corrupt(sc, sc->ip->i_ino, NULL);
 785                 goto out;
 786         }
 787
 788         /* Check directory tree structure */
 789         error = xfs_scrub_da_btree(sc, XFS_DATA_FORK, xfs_scrub_dir_rec, NULL);
 790         if (error)
 791                 return error;
 792
 793         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 794                 return error;
 795
 796         /* Check the freespace. */
 797         error = xfs_scrub_directory_blocks(sc);
 798         if (error)
 799                 return error;
 800
 801         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 802                 return error;
 803
 804         /*
 805          * Check that every dirent we see can also be looked up by hash.
 806          * Userspace usually asks for a 32k buffer, so we will too.
 807          */
 808         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 809                         sc->ip->i_d.di_size);
 810
 811         /*
 812          * Look up every name in this directory by hash.
 813          *
 814          * Use the xfs_readdir function to call xfs_scrub_dir_actor on
 815          * every directory entry in this directory.  In _actor, we check
 816          * the name, inode number, and ftype (if applicable) of the
 817          * entry.  xfs_readdir uses the VFS filldir functions to provide
 818          * iteration context.
 819          *
 820          * The VFS grabs a read or write lock via i_rwsem before it reads
 821          * or writes to a directory.  If we've gotten this far we've
 822          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 823          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 824          * to drop the ILOCK here in order to reuse the _readdir and
 825          * _dir_lookup routines, which do their own ILOCK locking.
 826          */
 827         oldpos = 0;
 828         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 829         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 830         while (true) {
 831                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 832                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0,
 833                                 &error))
 834                         goto out;
 835                 if (oldpos == sdc.dir_iter.pos)
 836                         break;
 837                 oldpos = sdc.dir_iter.pos;
 838         }
 839
 840 out:
 841         return error;
 842 }