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 out;
 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         /*
 187          * A negative error code returned here is supposed to cause the
 188          * dir_emit caller (xfs_readdir) to abort the directory iteration
 189          * and return zero to xfs_scrub_directory.
 190          */
 191         if (error == 0 && sdc->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 192                 return -EFSCORRUPTED;
 193         return error;
 194 }
 195
 196 /* Scrub a directory btree record. */
 197 STATIC int
 198 xfs_scrub_dir_rec(
 199         struct xfs_scrub_da_btree       *ds,
 200         int                             level,
 201         void                            *rec)
 202 {
 203         struct xfs_mount                *mp = ds->state->mp;
 204         struct xfs_dir2_leaf_entry      *ent = rec;
 205         struct xfs_inode                *dp = ds->dargs.dp;
 206         struct xfs_dir2_data_entry      *dent;
 207         struct xfs_buf                  *bp;
 208         char                            *p, *endp;
 209         xfs_ino_t                       ino;
 210         xfs_dablk_t                     rec_bno;
 211         xfs_dir2_db_t                   db;
 212         xfs_dir2_data_aoff_t            off;
 213         xfs_dir2_dataptr_t              ptr;
 214         xfs_dahash_t                    calc_hash;
 215         xfs_dahash_t                    hash;
 216         unsigned int                    tag;
 217         int                             error;
 218
 219         /* Check the hash of the entry. */
 220         error = xfs_scrub_da_btree_hash(ds, level, &ent->hashval);
 221         if (error)
 222                 goto out;
 223
 224         /* Valid hash pointer? */
 225         ptr = be32_to_cpu(ent->address);
 226         if (ptr == 0)
 227                 return 0;
 228
 229         /* Find the directory entry's location. */
 230         db = xfs_dir2_dataptr_to_db(mp->m_dir_geo, ptr);
 231         off = xfs_dir2_dataptr_to_off(mp->m_dir_geo, ptr);
 232         rec_bno = xfs_dir2_db_to_da(mp->m_dir_geo, db);
 233
 234         if (rec_bno >= mp->m_dir_geo->leafblk) {
 235                 xfs_scrub_da_set_corrupt(ds, level);
 236                 goto out;
 237         }
 238         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
 239         if (!xfs_scrub_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 240                         &error))
 241                 goto out;
 242         if (!bp) {
 243                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 244                 goto out;
 245         }
 246         xfs_scrub_buffer_recheck(ds->sc, bp);
 247
 248         if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 249                 goto out_relse;
 250
 251         dent = (struct xfs_dir2_data_entry *)(((char *)bp->b_addr) + off);
 252
 253         /* Make sure we got a real directory entry. */
 254         p = (char *)mp->m_dir_inode_ops->data_entry_p(bp->b_addr);
 255         endp = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 256         if (!endp) {
 257                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 258                 goto out_relse;
 259         }
 260         while (p < endp) {
 261                 struct xfs_dir2_data_entry      *dep;
 262                 struct xfs_dir2_data_unused     *dup;
 263
 264                 dup = (struct xfs_dir2_data_unused *)p;
 265                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 266                         p += be16_to_cpu(dup->length);
 267                         continue;
 268                 }
 269                 dep = (struct xfs_dir2_data_entry *)p;
 270                 if (dep == dent)
 271                         break;
 272                 p += mp->m_dir_inode_ops->data_entsize(dep->namelen);
 273         }
 274         if (p >= endp) {
 275                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 276                 goto out_relse;
 277         }
 278
 279         /* Retrieve the entry, sanity check it, and compare hashes. */
 280         ino = be64_to_cpu(dent->inumber);
 281         hash = be32_to_cpu(ent->hashval);
 282         tag = be16_to_cpup(dp->d_ops->data_entry_tag_p(dent));
 283         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 284                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 285         if (dent->namelen == 0) {
 286                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 287                 goto out_relse;
 288         }
 289         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 290         if (calc_hash != hash)
 291                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 292
 293 out_relse:
 294         xfs_trans_brelse(ds->dargs.trans, bp);
 295 out:
 296         return error;
 297 }
 298
 299 /*
 300  * Is this unused entry either in the bestfree or smaller than all of
 301  * them?  We've already checked that the bestfrees are sorted longest to
 302  * shortest, and that there aren't any bogus entries.
 303  */
 304 STATIC void
 305 xfs_scrub_directory_check_free_entry(
 306         struct xfs_scrub_context        *sc,
 307         xfs_dablk_t                     lblk,
 308         struct xfs_dir2_data_free       *bf,
 309         struct xfs_dir2_data_unused     *dup)
 310 {
 311         struct xfs_dir2_data_free       *dfp;
 312         unsigned int                    dup_length;
 313
 314         dup_length = be16_to_cpu(dup->length);
 315
 316         /* Unused entry is shorter than any of the bestfrees */
 317         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 318                 return;
 319
 320         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 321                 if (dup_length == be16_to_cpu(dfp->length))
 322                         return;
 323
 324         /* Unused entry should be in the bestfrees but wasn't found. */
 325         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 326 }
 327
 328 /* Check free space info in a directory data block. */
 329 STATIC int
 330 xfs_scrub_directory_data_bestfree(
 331         struct xfs_scrub_context        *sc,
 332         xfs_dablk_t                     lblk,
 333         bool                            is_block)
 334 {
 335         struct xfs_dir2_data_unused     *dup;
 336         struct xfs_dir2_data_free       *dfp;
 337         struct xfs_buf                  *bp;
 338         struct xfs_dir2_data_free       *bf;
 339         struct xfs_mount                *mp = sc->mp;
 340         const struct xfs_dir_ops        *d_ops;
 341         char                            *ptr;
 342         char                            *endptr;
 343         u16                             tag;
 344         unsigned int                    nr_bestfrees = 0;
 345         unsigned int                    nr_frees = 0;
 346         unsigned int                    smallest_bestfree;
 347         int                             newlen;
 348         int                             offset;
 349         int                             error;
 350
 351         d_ops = sc->ip->d_ops;
 352
 353         if (is_block) {
 354                 /* dir block format */
 355                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 356                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 357                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 358         } else {
 359                 /* dir data format */
 360                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, -1, &bp);
 361         }
 362         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 363                 goto out;
 364         xfs_scrub_buffer_recheck(sc, bp);
 365
 366         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 367
 368         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 369                 goto out_buf;
 370
 371         /* Do the bestfrees correspond to actual free space? */
 372         bf = d_ops->data_bestfree_p(bp->b_addr);
 373         smallest_bestfree = UINT_MAX;
 374         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 375                 offset = be16_to_cpu(dfp->offset);
 376                 if (offset == 0)
 377                         continue;
 378                 if (offset >= mp->m_dir_geo->blksize) {
 379                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 380                         goto out_buf;
 381                 }
 382                 dup = (struct xfs_dir2_data_unused *)(bp->b_addr + offset);
 383                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 384
 385                 /* bestfree doesn't match the entry it points at? */
 386                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 387                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 388                     tag != ((char *)dup - (char *)bp->b_addr)) {
 389                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 390                         goto out_buf;
 391                 }
 392
 393                 /* bestfree records should be ordered largest to smallest */
 394                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 395                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 396                         goto out_buf;
 397                 }
 398
 399                 smallest_bestfree = be16_to_cpu(dfp->length);
 400                 nr_bestfrees++;
 401         }
 402
 403         /* Make sure the bestfrees are actually the best free spaces. */
 404         ptr = (char *)d_ops->data_entry_p(bp->b_addr);
 405         endptr = xfs_dir3_data_endp(mp->m_dir_geo, bp->b_addr);
 406
 407         /* Iterate the entries, stopping when we hit or go past the end. */
 408         while (ptr < endptr) {
 409                 dup = (struct xfs_dir2_data_unused *)ptr;
 410                 /* Skip real entries */
 411                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 412                         struct xfs_dir2_data_entry      *dep;
 413
 414                         dep = (struct xfs_dir2_data_entry *)ptr;
 415                         newlen = d_ops->data_entsize(dep->namelen);
 416                         if (newlen <= 0) {
 417                                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK,
 418                                                 lblk);
 419                                 goto out_buf;
 420                         }
 421                         ptr += newlen;
 422                         continue;
 423                 }
 424
 425                 /* Spot check this free entry */
 426                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 427                 if (tag != ((char *)dup - (char *)bp->b_addr)) {
 428                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 429                         goto out_buf;
 430                 }
 431
 432                 /*
 433                  * Either this entry is a bestfree or it's smaller than
 434                  * any of the bestfrees.
 435                  */
 436                 xfs_scrub_directory_check_free_entry(sc, lblk, bf, dup);
 437                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 438                         goto out_buf;
 439
 440                 /* Move on. */
 441                 newlen = be16_to_cpu(dup->length);
 442                 if (newlen <= 0) {
 443                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 444                         goto out_buf;
 445                 }
 446                 ptr += newlen;
 447                 if (ptr <= endptr)
 448                         nr_frees++;
 449         }
 450
 451         /* We're required to fill all the space. */
 452         if (ptr != endptr)
 453                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 454
 455         /* Did we see at least as many free slots as there are bestfrees? */
 456         if (nr_frees < nr_bestfrees)
 457                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 458 out_buf:
 459         xfs_trans_brelse(sc->tp, bp);
 460 out:
 461         return error;
 462 }
 463
 464 /*
 465  * Does the free space length in the free space index block ($len) match
 466  * the longest length in the directory data block's bestfree array?
 467  * Assume that we've already checked that the data block's bestfree
 468  * array is in order.
 469  */
 470 STATIC void
 471 xfs_scrub_directory_check_freesp(
 472         struct xfs_scrub_context        *sc,
 473         xfs_dablk_t                     lblk,
 474         struct xfs_buf                  *dbp,
 475         unsigned int                    len)
 476 {
 477         struct xfs_dir2_data_free       *dfp;
 478
 479         dfp = sc->ip->d_ops->data_bestfree_p(dbp->b_addr);
 480
 481         if (len != be16_to_cpu(dfp->length))
 482                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 483
 484         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 485                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 486 }
 487
 488 /* Check free space info in a directory leaf1 block. */
 489 STATIC int
 490 xfs_scrub_directory_leaf1_bestfree(
 491         struct xfs_scrub_context        *sc,
 492         struct xfs_da_args              *args,
 493         xfs_dablk_t                     lblk)
 494 {
 495         struct xfs_dir3_icleaf_hdr      leafhdr;
 496         struct xfs_dir2_leaf_entry      *ents;
 497         struct xfs_dir2_leaf_tail       *ltp;
 498         struct xfs_dir2_leaf            *leaf;
 499         struct xfs_buf                  *dbp;
 500         struct xfs_buf                  *bp;
 501         const struct xfs_dir_ops        *d_ops = sc->ip->d_ops;
 502         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 503         __be16                          *bestp;
 504         __u16                           best;
 505         __u32                           hash;
 506         __u32                           lasthash = 0;
 507         __u32                           bestcount;
 508         unsigned int                    stale = 0;
 509         int                             i;
 510         int                             error;
 511
 512         /* Read the free space block. */
 513         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, -1, &bp);
 514         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 515                 goto out;
 516         xfs_scrub_buffer_recheck(sc, bp);
 517
 518         leaf = bp->b_addr;
 519         d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 520         ents = d_ops->leaf_ents_p(leaf);
 521         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 522         bestcount = be32_to_cpu(ltp->bestcount);
 523         bestp = xfs_dir2_leaf_bests_p(ltp);
 524
 525         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 526                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 527
 528                 if (hdr3->pad != cpu_to_be32(0))
 529                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 530         }
 531
 532         /*
 533          * There should be as many bestfree slots as there are dir data
 534          * blocks that can fit under i_size.
 535          */
 536         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 537                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 538                 goto out;
 539         }
 540
 541         /* Is the leaf count even remotely sane? */
 542         if (leafhdr.count > d_ops->leaf_max_ents(geo)) {
 543                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 544                 goto out;
 545         }
 546
 547         /* Leaves and bests don't overlap in leaf format. */
 548         if ((char *)&ents[leafhdr.count] > (char *)bestp) {
 549                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 550                 goto out;
 551         }
 552
 553         /* Check hash value order, count stale entries.  */
 554         for (i = 0; i < leafhdr.count; i++) {
 555                 hash = be32_to_cpu(ents[i].hashval);
 556                 if (i > 0 && lasthash > hash)
 557                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 558                 lasthash = hash;
 559                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 560                         stale++;
 561         }
 562         if (leafhdr.stale != stale)
 563                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 564         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 565                 goto out;
 566
 567         /* Check all the bestfree entries. */
 568         for (i = 0; i < bestcount; i++, bestp++) {
 569                 best = be16_to_cpu(*bestp);
 570                 if (best == NULLDATAOFF)
 571                         continue;
 572                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 573                                 i * args->geo->fsbcount, -1, &dbp);
 574                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 575                                 &error))
 576                         break;
 577                 xfs_scrub_directory_check_freesp(sc, lblk, dbp, best);
 578                 xfs_trans_brelse(sc->tp, dbp);
 579                 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 580                         goto out;
 581         }
 582 out:
 583         return error;
 584 }
 585
 586 /* Check free space info in a directory freespace block. */
 587 STATIC int
 588 xfs_scrub_directory_free_bestfree(
 589         struct xfs_scrub_context        *sc,
 590         struct xfs_da_args              *args,
 591         xfs_dablk_t                     lblk)
 592 {
 593         struct xfs_dir3_icfree_hdr      freehdr;
 594         struct xfs_buf                  *dbp;
 595         struct xfs_buf                  *bp;
 596         __be16                          *bestp;
 597         __u16                           best;
 598         unsigned int                    stale = 0;
 599         int                             i;
 600         int                             error;
 601
 602         /* Read the free space block */
 603         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 604         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 605                 goto out;
 606         xfs_scrub_buffer_recheck(sc, bp);
 607
 608         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 609                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 610
 611                 if (hdr3->pad != cpu_to_be32(0))
 612                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 613         }
 614
 615         /* Check all the entries. */
 616         sc->ip->d_ops->free_hdr_from_disk(&freehdr, bp->b_addr);
 617         bestp = sc->ip->d_ops->free_bests_p(bp->b_addr);
 618         for (i = 0; i < freehdr.nvalid; i++, bestp++) {
 619                 best = be16_to_cpu(*bestp);
 620                 if (best == NULLDATAOFF) {
 621                         stale++;
 622                         continue;
 623                 }
 624                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 625                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 626                                 -1, &dbp);
 627                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 628                                 &error))
 629                         break;
 630                 xfs_scrub_directory_check_freesp(sc, lblk, dbp, best);
 631                 xfs_trans_brelse(sc->tp, dbp);
 632         }
 633
 634         if (freehdr.nused + stale != freehdr.nvalid)
 635                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 636 out:
 637         return error;
 638 }
 639
 640 /* Check free space information in directories. */
 641 STATIC int
 642 xfs_scrub_directory_blocks(
 643         struct xfs_scrub_context        *sc)
 644 {
 645         struct xfs_bmbt_irec            got;
 646         struct xfs_da_args              args;
 647         struct xfs_ifork                *ifp;
 648         struct xfs_mount                *mp = sc->mp;
 649         xfs_fileoff_t                   leaf_lblk;
 650         xfs_fileoff_t                   free_lblk;
 651         xfs_fileoff_t                   lblk;
 652         struct xfs_iext_cursor          icur;
 653         xfs_dablk_t                     dabno;
 654         bool                            found;
 655         int                             is_block = 0;
 656         int                             error;
 657
 658         /* Ignore local format directories. */
 659         if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 660             sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
 661                 return 0;
 662
 663         ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 664         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 665         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 666         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 667
 668         /* Is this a block dir? */
 669         args.dp = sc->ip;
 670         args.geo = mp->m_dir_geo;
 671         args.trans = sc->tp;
 672         error = xfs_dir2_isblock(&args, &is_block);
 673         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 674                 goto out;
 675
 676         /* Iterate all the data extents in the directory... */
 677         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 678         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 679                 /* Block directories only have a single block at offset 0. */
 680                 if (is_block &&
 681                     (got.br_startoff > 0 ||
 682                      got.br_blockcount != args.geo->fsbcount)) {
 683                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK,
 684                                         got.br_startoff);
 685                         break;
 686                 }
 687
 688                 /* No more data blocks... */
 689                 if (got.br_startoff >= leaf_lblk)
 690                         break;
 691
 692                 /*
 693                  * Check each data block's bestfree data.
 694                  *
 695                  * Iterate all the fsbcount-aligned block offsets in
 696                  * this directory.  The directory block reading code is
 697                  * smart enough to do its own bmap lookups to handle
 698                  * discontiguous directory blocks.  When we're done
 699                  * with the extent record, re-query the bmap at the
 700                  * next fsbcount-aligned offset to avoid redundant
 701                  * block checks.
 702                  */
 703                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 704                                 args.geo->fsbcount);
 705                      lblk < got.br_startoff + got.br_blockcount;
 706                      lblk += args.geo->fsbcount) {
 707                         error = xfs_scrub_directory_data_bestfree(sc, lblk,
 708                                         is_block);
 709                         if (error)
 710                                 goto out;
 711                 }
 712                 dabno = got.br_startoff + got.br_blockcount;
 713                 lblk = roundup(dabno, args.geo->fsbcount);
 714                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 715         }
 716
 717         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 718                 goto out;
 719
 720         /* Look for a leaf1 block, which has free info. */
 721         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 722             got.br_startoff == leaf_lblk &&
 723             got.br_blockcount == args.geo->fsbcount &&
 724             !xfs_iext_next_extent(ifp, &icur, &got)) {
 725                 if (is_block) {
 726                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 727                         goto out;
 728                 }
 729                 error = xfs_scrub_directory_leaf1_bestfree(sc, &args,
 730                                 leaf_lblk);
 731                 if (error)
 732                         goto out;
 733         }
 734
 735         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 736                 goto out;
 737
 738         /* Scan for free blocks */
 739         lblk = free_lblk;
 740         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 741         while (found && !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
 742                 /*
 743                  * Dirs can't have blocks mapped above 2^32.
 744                  * Single-block dirs shouldn't even be here.
 745                  */
 746                 lblk = got.br_startoff;
 747                 if (lblk & ~0xFFFFFFFFULL) {
 748                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 749                         goto out;
 750                 }
 751                 if (is_block) {
 752                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 753                         goto out;
 754                 }
 755
 756                 /*
 757                  * Check each dir free block's bestfree data.
 758                  *
 759                  * Iterate all the fsbcount-aligned block offsets in
 760                  * this directory.  The directory block reading code is
 761                  * smart enough to do its own bmap lookups to handle
 762                  * discontiguous directory blocks.  When we're done
 763                  * with the extent record, re-query the bmap at the
 764                  * next fsbcount-aligned offset to avoid redundant
 765                  * block checks.
 766                  */
 767                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 768                                 args.geo->fsbcount);
 769                      lblk < got.br_startoff + got.br_blockcount;
 770                      lblk += args.geo->fsbcount) {
 771                         error = xfs_scrub_directory_free_bestfree(sc, &args,
 772                                         lblk);
 773                         if (error)
 774                                 goto out;
 775                 }
 776                 dabno = got.br_startoff + got.br_blockcount;
 777                 lblk = roundup(dabno, args.geo->fsbcount);
 778                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 779         }
 780 out:
 781         return error;
 782 }
 783
 784 /* Scrub a whole directory. */
 785 int
 786 xfs_scrub_directory(
 787         struct xfs_scrub_context        *sc)
 788 {
 789         struct xfs_scrub_dir_ctx        sdc = {
 790                 .dir_iter.actor = xfs_scrub_dir_actor,
 791                 .dir_iter.pos = 0,
 792                 .sc = sc,
 793         };
 794         size_t                          bufsize;
 795         loff_t                          oldpos;
 796         int                             error = 0;
 797
 798         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 799                 return -ENOENT;
 800
 801         /* Plausible size? */
 802         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 803                 xfs_scrub_ino_set_corrupt(sc, sc->ip->i_ino);
 804                 goto out;
 805         }
 806
 807         /* Check directory tree structure */
 808         error = xfs_scrub_da_btree(sc, XFS_DATA_FORK, xfs_scrub_dir_rec, NULL);
 809         if (error)
 810                 return error;
 811
 812         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 813                 return error;
 814
 815         /* Check the freespace. */
 816         error = xfs_scrub_directory_blocks(sc);
 817         if (error)
 818                 return error;
 819
 820         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 821                 return error;
 822
 823         /*
 824          * Check that every dirent we see can also be looked up by hash.
 825          * Userspace usually asks for a 32k buffer, so we will too.
 826          */
 827         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 828                         sc->ip->i_d.di_size);
 829
 830         /*
 831          * Look up every name in this directory by hash.
 832          *
 833          * Use the xfs_readdir function to call xfs_scrub_dir_actor on
 834          * every directory entry in this directory.  In _actor, we check
 835          * the name, inode number, and ftype (if applicable) of the
 836          * entry.  xfs_readdir uses the VFS filldir functions to provide
 837          * iteration context.
 838          *
 839          * The VFS grabs a read or write lock via i_rwsem before it reads
 840          * or writes to a directory.  If we've gotten this far we've
 841          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 842          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 843          * to drop the ILOCK here in order to reuse the _readdir and
 844          * _dir_lookup routines, which do their own ILOCK locking.
 845          */
 846         oldpos = 0;
 847         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 848         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 849         while (true) {
 850                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 851                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0,
 852                                 &error))
 853                         goto out;
 854                 if (oldpos == sdc.dir_iter.pos)
 855                         break;
 856                 oldpos = sdc.dir_iter.pos;
 857         }
 858
 859 out:
 860         return error;
 861 }