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_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         xfs_ino_t                       ino;
 204         xfs_dablk_t                     rec_bno;
 205         xfs_dir2_db_t                   db;
 206         xfs_dir2_data_aoff_t            off;
 207         xfs_dir2_dataptr_t              ptr;
 208         xfs_dahash_t                    calc_hash;
 209         xfs_dahash_t                    hash;
 210         unsigned int                    tag;
 211         int                             error;
 212
 213         /* Check the hash of the entry. */
 214         error = xfs_scrub_da_btree_hash(ds, level, &ent->hashval);
 215         if (error)
 216                 goto out;
 217
 218         /* Valid hash pointer? */
 219         ptr = be32_to_cpu(ent->address);
 220         if (ptr == 0)
 221                 return 0;
 222
 223         /* Find the directory entry's location. */
 224         db = xfs_dir2_dataptr_to_db(mp->m_dir_geo, ptr);
 225         off = xfs_dir2_dataptr_to_off(mp->m_dir_geo, ptr);
 226         rec_bno = xfs_dir2_db_to_da(mp->m_dir_geo, db);
 227
 228         if (rec_bno >= mp->m_dir_geo->leafblk) {
 229                 xfs_scrub_da_set_corrupt(ds, level);
 230                 goto out;
 231         }
 232         error = xfs_dir3_data_read(ds->dargs.trans, dp, rec_bno, -2, &bp);
 233         if (!xfs_scrub_fblock_process_error(ds->sc, XFS_DATA_FORK, rec_bno,
 234                         &error))
 235                 goto out;
 236         if (!bp) {
 237                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 238                 goto out;
 239         }
 240
 241         /* Retrieve the entry, sanity check it, and compare hashes. */
 242         dent = (struct xfs_dir2_data_entry *)(((char *)bp->b_addr) + off);
 243         ino = be64_to_cpu(dent->inumber);
 244         hash = be32_to_cpu(ent->hashval);
 245         tag = be16_to_cpup(dp->d_ops->data_entry_tag_p(dent));
 246         if (!xfs_verify_dir_ino(mp, ino) || tag != off)
 247                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 248         if (dent->namelen == 0) {
 249                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 250                 goto out_relse;
 251         }
 252         calc_hash = xfs_da_hashname(dent->name, dent->namelen);
 253         if (calc_hash != hash)
 254                 xfs_scrub_fblock_set_corrupt(ds->sc, XFS_DATA_FORK, rec_bno);
 255
 256 out_relse:
 257         xfs_trans_brelse(ds->dargs.trans, bp);
 258 out:
 259         return error;
 260 }
 261
 262 /*
 263  * Is this unused entry either in the bestfree or smaller than all of
 264  * them?  We've already checked that the bestfrees are sorted longest to
 265  * shortest, and that there aren't any bogus entries.
 266  */
 267 STATIC void
 268 xfs_scrub_directory_check_free_entry(
 269         struct xfs_scrub_context        *sc,
 270         xfs_dablk_t                     lblk,
 271         struct xfs_dir2_data_free       *bf,
 272         struct xfs_dir2_data_unused     *dup)
 273 {
 274         struct xfs_dir2_data_free       *dfp;
 275         unsigned int                    dup_length;
 276
 277         dup_length = be16_to_cpu(dup->length);
 278
 279         /* Unused entry is shorter than any of the bestfrees */
 280         if (dup_length < be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 281                 return;
 282
 283         for (dfp = &bf[XFS_DIR2_DATA_FD_COUNT - 1]; dfp >= bf; dfp--)
 284                 if (dup_length == be16_to_cpu(dfp->length))
 285                         return;
 286
 287         /* Unused entry should be in the bestfrees but wasn't found. */
 288         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 289 }
 290
 291 /* Check free space info in a directory data block. */
 292 STATIC int
 293 xfs_scrub_directory_data_bestfree(
 294         struct xfs_scrub_context        *sc,
 295         xfs_dablk_t                     lblk,
 296         bool                            is_block)
 297 {
 298         struct xfs_dir2_data_unused     *dup;
 299         struct xfs_dir2_data_free       *dfp;
 300         struct xfs_buf                  *bp;
 301         struct xfs_dir2_data_free       *bf;
 302         struct xfs_mount                *mp = sc->mp;
 303         const struct xfs_dir_ops        *d_ops;
 304         char                            *ptr;
 305         char                            *endptr;
 306         u16                             tag;
 307         unsigned int                    nr_bestfrees = 0;
 308         unsigned int                    nr_frees = 0;
 309         unsigned int                    smallest_bestfree;
 310         int                             newlen;
 311         int                             offset;
 312         int                             error;
 313
 314         d_ops = sc->ip->d_ops;
 315
 316         if (is_block) {
 317                 /* dir block format */
 318                 if (lblk != XFS_B_TO_FSBT(mp, XFS_DIR2_DATA_OFFSET))
 319                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 320                 error = xfs_dir3_block_read(sc->tp, sc->ip, &bp);
 321         } else {
 322                 /* dir data format */
 323                 error = xfs_dir3_data_read(sc->tp, sc->ip, lblk, -1, &bp);
 324         }
 325         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 326                 goto out;
 327
 328         /* XXX: Check xfs_dir3_data_hdr.pad is zero once we start setting it. */
 329
 330         /* Do the bestfrees correspond to actual free space? */
 331         bf = d_ops->data_bestfree_p(bp->b_addr);
 332         smallest_bestfree = UINT_MAX;
 333         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 334                 offset = be16_to_cpu(dfp->offset);
 335                 if (offset == 0)
 336                         continue;
 337                 if (offset >= mp->m_dir_geo->blksize) {
 338                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 339                         goto out_buf;
 340                 }
 341                 dup = (struct xfs_dir2_data_unused *)(bp->b_addr + offset);
 342                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 343
 344                 /* bestfree doesn't match the entry it points at? */
 345                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG) ||
 346                     be16_to_cpu(dup->length) != be16_to_cpu(dfp->length) ||
 347                     tag != ((char *)dup - (char *)bp->b_addr)) {
 348                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 349                         goto out_buf;
 350                 }
 351
 352                 /* bestfree records should be ordered largest to smallest */
 353                 if (smallest_bestfree < be16_to_cpu(dfp->length)) {
 354                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 355                         goto out_buf;
 356                 }
 357
 358                 smallest_bestfree = be16_to_cpu(dfp->length);
 359                 nr_bestfrees++;
 360         }
 361
 362         /* Make sure the bestfrees are actually the best free spaces. */
 363         ptr = (char *)d_ops->data_entry_p(bp->b_addr);
 364         if (is_block) {
 365                 struct xfs_dir2_block_tail      *btp;
 366
 367                 btp = xfs_dir2_block_tail_p(mp->m_dir_geo, bp->b_addr);
 368                 endptr = (char *)xfs_dir2_block_leaf_p(btp);
 369         } else
 370                 endptr = (char *)bp->b_addr + BBTOB(bp->b_length);
 371
 372         /* Iterate the entries, stopping when we hit or go past the end. */
 373         while (ptr < endptr) {
 374                 dup = (struct xfs_dir2_data_unused *)ptr;
 375                 /* Skip real entries */
 376                 if (dup->freetag != cpu_to_be16(XFS_DIR2_DATA_FREE_TAG)) {
 377                         struct xfs_dir2_data_entry      *dep;
 378
 379                         dep = (struct xfs_dir2_data_entry *)ptr;
 380                         newlen = d_ops->data_entsize(dep->namelen);
 381                         if (newlen <= 0) {
 382                                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK,
 383                                                 lblk);
 384                                 goto out_buf;
 385                         }
 386                         ptr += newlen;
 387                         continue;
 388                 }
 389
 390                 /* Spot check this free entry */
 391                 tag = be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup));
 392                 if (tag != ((char *)dup - (char *)bp->b_addr))
 393                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 394
 395                 /*
 396                  * Either this entry is a bestfree or it's smaller than
 397                  * any of the bestfrees.
 398                  */
 399                 xfs_scrub_directory_check_free_entry(sc, lblk, bf, dup);
 400
 401                 /* Move on. */
 402                 newlen = be16_to_cpu(dup->length);
 403                 if (newlen <= 0) {
 404                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 405                         goto out_buf;
 406                 }
 407                 ptr += newlen;
 408                 if (ptr <= endptr)
 409                         nr_frees++;
 410         }
 411
 412         /* We're required to fill all the space. */
 413         if (ptr != endptr)
 414                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 415
 416         /* Did we see at least as many free slots as there are bestfrees? */
 417         if (nr_frees < nr_bestfrees)
 418                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 419 out_buf:
 420         xfs_trans_brelse(sc->tp, bp);
 421 out:
 422         return error;
 423 }
 424
 425 /*
 426  * Does the free space length in the free space index block ($len) match
 427  * the longest length in the directory data block's bestfree array?
 428  * Assume that we've already checked that the data block's bestfree
 429  * array is in order.
 430  */
 431 STATIC void
 432 xfs_scrub_directory_check_freesp(
 433         struct xfs_scrub_context        *sc,
 434         xfs_dablk_t                     lblk,
 435         struct xfs_buf                  *dbp,
 436         unsigned int                    len)
 437 {
 438         struct xfs_dir2_data_free       *dfp;
 439
 440         dfp = sc->ip->d_ops->data_bestfree_p(dbp->b_addr);
 441
 442         if (len != be16_to_cpu(dfp->length))
 443                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 444
 445         if (len > 0 && be16_to_cpu(dfp->offset) == 0)
 446                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 447 }
 448
 449 /* Check free space info in a directory leaf1 block. */
 450 STATIC int
 451 xfs_scrub_directory_leaf1_bestfree(
 452         struct xfs_scrub_context        *sc,
 453         struct xfs_da_args              *args,
 454         xfs_dablk_t                     lblk)
 455 {
 456         struct xfs_dir3_icleaf_hdr      leafhdr;
 457         struct xfs_dir2_leaf_entry      *ents;
 458         struct xfs_dir2_leaf_tail       *ltp;
 459         struct xfs_dir2_leaf            *leaf;
 460         struct xfs_buf                  *dbp;
 461         struct xfs_buf                  *bp;
 462         const struct xfs_dir_ops        *d_ops = sc->ip->d_ops;
 463         struct xfs_da_geometry          *geo = sc->mp->m_dir_geo;
 464         __be16                          *bestp;
 465         __u16                           best;
 466         __u32                           hash;
 467         __u32                           lasthash = 0;
 468         __u32                           bestcount;
 469         unsigned int                    stale = 0;
 470         int                             i;
 471         int                             error;
 472
 473         /* Read the free space block. */
 474         error = xfs_dir3_leaf_read(sc->tp, sc->ip, lblk, -1, &bp);
 475         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 476                 goto out;
 477
 478         leaf = bp->b_addr;
 479         d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 480         ents = d_ops->leaf_ents_p(leaf);
 481         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 482         bestcount = be32_to_cpu(ltp->bestcount);
 483         bestp = xfs_dir2_leaf_bests_p(ltp);
 484
 485         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 486                 struct xfs_dir3_leaf_hdr        *hdr3 = bp->b_addr;
 487
 488                 if (hdr3->pad != cpu_to_be32(0))
 489                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 490         }
 491
 492         /*
 493          * There should be as many bestfree slots as there are dir data
 494          * blocks that can fit under i_size.
 495          */
 496         if (bestcount != xfs_dir2_byte_to_db(geo, sc->ip->i_d.di_size)) {
 497                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 498                 goto out;
 499         }
 500
 501         /* Is the leaf count even remotely sane? */
 502         if (leafhdr.count > d_ops->leaf_max_ents(geo)) {
 503                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 504                 goto out;
 505         }
 506
 507         /* Leaves and bests don't overlap in leaf format. */
 508         if ((char *)&ents[leafhdr.count] > (char *)bestp) {
 509                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 510                 goto out;
 511         }
 512
 513         /* Check hash value order, count stale entries.  */
 514         for (i = 0; i < leafhdr.count; i++) {
 515                 hash = be32_to_cpu(ents[i].hashval);
 516                 if (i > 0 && lasthash > hash)
 517                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 518                 lasthash = hash;
 519                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 520                         stale++;
 521         }
 522         if (leafhdr.stale != stale)
 523                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 524
 525         /* Check all the bestfree entries. */
 526         for (i = 0; i < bestcount; i++, bestp++) {
 527                 best = be16_to_cpu(*bestp);
 528                 if (best == NULLDATAOFF)
 529                         continue;
 530                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 531                                 i * args->geo->fsbcount, -1, &dbp);
 532                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 533                                 &error))
 534                         continue;
 535                 xfs_scrub_directory_check_freesp(sc, lblk, dbp, best);
 536                 xfs_trans_brelse(sc->tp, dbp);
 537         }
 538 out:
 539         return error;
 540 }
 541
 542 /* Check free space info in a directory freespace block. */
 543 STATIC int
 544 xfs_scrub_directory_free_bestfree(
 545         struct xfs_scrub_context        *sc,
 546         struct xfs_da_args              *args,
 547         xfs_dablk_t                     lblk)
 548 {
 549         struct xfs_dir3_icfree_hdr      freehdr;
 550         struct xfs_buf                  *dbp;
 551         struct xfs_buf                  *bp;
 552         __be16                          *bestp;
 553         __u16                           best;
 554         unsigned int                    stale = 0;
 555         int                             i;
 556         int                             error;
 557
 558         /* Read the free space block */
 559         error = xfs_dir2_free_read(sc->tp, sc->ip, lblk, &bp);
 560         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 561                 goto out;
 562
 563         if (xfs_sb_version_hascrc(&sc->mp->m_sb)) {
 564                 struct xfs_dir3_free_hdr        *hdr3 = bp->b_addr;
 565
 566                 if (hdr3->pad != cpu_to_be32(0))
 567                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 568         }
 569
 570         /* Check all the entries. */
 571         sc->ip->d_ops->free_hdr_from_disk(&freehdr, bp->b_addr);
 572         bestp = sc->ip->d_ops->free_bests_p(bp->b_addr);
 573         for (i = 0; i < freehdr.nvalid; i++, bestp++) {
 574                 best = be16_to_cpu(*bestp);
 575                 if (best == NULLDATAOFF) {
 576                         stale++;
 577                         continue;
 578                 }
 579                 error = xfs_dir3_data_read(sc->tp, sc->ip,
 580                                 (freehdr.firstdb + i) * args->geo->fsbcount,
 581                                 -1, &dbp);
 582                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk,
 583                                 &error))
 584                         continue;
 585                 xfs_scrub_directory_check_freesp(sc, lblk, dbp, best);
 586                 xfs_trans_brelse(sc->tp, dbp);
 587         }
 588
 589         if (freehdr.nused + stale != freehdr.nvalid)
 590                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 591 out:
 592         return error;
 593 }
 594
 595 /* Check free space information in directories. */
 596 STATIC int
 597 xfs_scrub_directory_blocks(
 598         struct xfs_scrub_context        *sc)
 599 {
 600         struct xfs_bmbt_irec            got;
 601         struct xfs_da_args              args;
 602         struct xfs_ifork                *ifp;
 603         struct xfs_mount                *mp = sc->mp;
 604         xfs_fileoff_t                   leaf_lblk;
 605         xfs_fileoff_t                   free_lblk;
 606         xfs_fileoff_t                   lblk;
 607         struct xfs_iext_cursor          icur;
 608         xfs_dablk_t                     dabno;
 609         bool                            found;
 610         int                             is_block = 0;
 611         int                             error;
 612
 613         /* Ignore local format directories. */
 614         if (sc->ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
 615             sc->ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
 616                 return 0;
 617
 618         ifp = XFS_IFORK_PTR(sc->ip, XFS_DATA_FORK);
 619         lblk = XFS_B_TO_FSB(mp, XFS_DIR2_DATA_OFFSET);
 620         leaf_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_LEAF_OFFSET);
 621         free_lblk = XFS_B_TO_FSB(mp, XFS_DIR2_FREE_OFFSET);
 622
 623         /* Is this a block dir? */
 624         args.dp = sc->ip;
 625         args.geo = mp->m_dir_geo;
 626         args.trans = sc->tp;
 627         error = xfs_dir2_isblock(&args, &is_block);
 628         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, lblk, &error))
 629                 goto out;
 630
 631         /* Iterate all the data extents in the directory... */
 632         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 633         while (found) {
 634                 /* Block directories only have a single block at offset 0. */
 635                 if (is_block &&
 636                     (got.br_startoff > 0 ||
 637                      got.br_blockcount != args.geo->fsbcount)) {
 638                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK,
 639                                         got.br_startoff);
 640                         break;
 641                 }
 642
 643                 /* No more data blocks... */
 644                 if (got.br_startoff >= leaf_lblk)
 645                         break;
 646
 647                 /*
 648                  * Check each data block's bestfree data.
 649                  *
 650                  * Iterate all the fsbcount-aligned block offsets in
 651                  * this directory.  The directory block reading code is
 652                  * smart enough to do its own bmap lookups to handle
 653                  * discontiguous directory blocks.  When we're done
 654                  * with the extent record, re-query the bmap at the
 655                  * next fsbcount-aligned offset to avoid redundant
 656                  * block checks.
 657                  */
 658                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 659                                 args.geo->fsbcount);
 660                      lblk < got.br_startoff + got.br_blockcount;
 661                      lblk += args.geo->fsbcount) {
 662                         error = xfs_scrub_directory_data_bestfree(sc, lblk,
 663                                         is_block);
 664                         if (error)
 665                                 goto out;
 666                 }
 667                 dabno = got.br_startoff + got.br_blockcount;
 668                 lblk = roundup(dabno, args.geo->fsbcount);
 669                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 670         }
 671
 672         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 673                 goto out;
 674
 675         /* Look for a leaf1 block, which has free info. */
 676         if (xfs_iext_lookup_extent(sc->ip, ifp, leaf_lblk, &icur, &got) &&
 677             got.br_startoff == leaf_lblk &&
 678             got.br_blockcount == args.geo->fsbcount &&
 679             !xfs_iext_next_extent(ifp, &icur, &got)) {
 680                 if (is_block) {
 681                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 682                         goto out;
 683                 }
 684                 error = xfs_scrub_directory_leaf1_bestfree(sc, &args,
 685                                 leaf_lblk);
 686                 if (error)
 687                         goto out;
 688         }
 689
 690         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 691                 goto out;
 692
 693         /* Scan for free blocks */
 694         lblk = free_lblk;
 695         found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 696         while (found) {
 697                 /*
 698                  * Dirs can't have blocks mapped above 2^32.
 699                  * Single-block dirs shouldn't even be here.
 700                  */
 701                 lblk = got.br_startoff;
 702                 if (lblk & ~0xFFFFFFFFULL) {
 703                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 704                         goto out;
 705                 }
 706                 if (is_block) {
 707                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, lblk);
 708                         goto out;
 709                 }
 710
 711                 /*
 712                  * Check each dir free block's bestfree data.
 713                  *
 714                  * Iterate all the fsbcount-aligned block offsets in
 715                  * this directory.  The directory block reading code is
 716                  * smart enough to do its own bmap lookups to handle
 717                  * discontiguous directory blocks.  When we're done
 718                  * with the extent record, re-query the bmap at the
 719                  * next fsbcount-aligned offset to avoid redundant
 720                  * block checks.
 721                  */
 722                 for (lblk = roundup((xfs_dablk_t)got.br_startoff,
 723                                 args.geo->fsbcount);
 724                      lblk < got.br_startoff + got.br_blockcount;
 725                      lblk += args.geo->fsbcount) {
 726                         error = xfs_scrub_directory_free_bestfree(sc, &args,
 727                                         lblk);
 728                         if (error)
 729                                 goto out;
 730                 }
 731                 dabno = got.br_startoff + got.br_blockcount;
 732                 lblk = roundup(dabno, args.geo->fsbcount);
 733                 found = xfs_iext_lookup_extent(sc->ip, ifp, lblk, &icur, &got);
 734         }
 735 out:
 736         return error;
 737 }
 738
 739 /* Scrub a whole directory. */
 740 int
 741 xfs_scrub_directory(
 742         struct xfs_scrub_context        *sc)
 743 {
 744         struct xfs_scrub_dir_ctx        sdc = {
 745                 .dir_iter.actor = xfs_scrub_dir_actor,
 746                 .dir_iter.pos = 0,
 747                 .sc = sc,
 748         };
 749         size_t                          bufsize;
 750         loff_t                          oldpos;
 751         int                             error = 0;
 752
 753         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 754                 return -ENOENT;
 755
 756         /* Plausible size? */
 757         if (sc->ip->i_d.di_size < xfs_dir2_sf_hdr_size(0)) {
 758                 xfs_scrub_ino_set_corrupt(sc, sc->ip->i_ino, NULL);
 759                 goto out;
 760         }
 761
 762         /* Check directory tree structure */
 763         error = xfs_scrub_da_btree(sc, XFS_DATA_FORK, xfs_scrub_dir_rec, NULL);
 764         if (error)
 765                 return error;
 766
 767         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 768                 return error;
 769
 770         /* Check the freespace. */
 771         error = xfs_scrub_directory_blocks(sc);
 772         if (error)
 773                 return error;
 774
 775         if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
 776                 return error;
 777
 778         /*
 779          * Check that every dirent we see can also be looked up by hash.
 780          * Userspace usually asks for a 32k buffer, so we will too.
 781          */
 782         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 783                         sc->ip->i_d.di_size);
 784
 785         /*
 786          * Look up every name in this directory by hash.
 787          *
 788          * Use the xfs_readdir function to call xfs_scrub_dir_actor on
 789          * every directory entry in this directory.  In _actor, we check
 790          * the name, inode number, and ftype (if applicable) of the
 791          * entry.  xfs_readdir uses the VFS filldir functions to provide
 792          * iteration context.
 793          *
 794          * The VFS grabs a read or write lock via i_rwsem before it reads
 795          * or writes to a directory.  If we've gotten this far we've
 796          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 797          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 798          * to drop the ILOCK here in order to reuse the _readdir and
 799          * _dir_lookup routines, which do their own ILOCK locking.
 800          */
 801         oldpos = 0;
 802         sc->ilock_flags &= ~XFS_ILOCK_EXCL;
 803         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL);
 804         while (true) {
 805                 error = xfs_readdir(sc->tp, sc->ip, &sdc.dir_iter, bufsize);
 806                 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0,
 807                                 &error))
 808                         goto out;
 809                 if (oldpos == sdc.dir_iter.pos)
 810                         break;
 811                 oldpos = sdc.dir_iter.pos;
 812         }
 813
 814 out:
 815         return error;
 816 }