fs/xfs/scrub/parent.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_dir2.h"
  35 #include "xfs_dir2_priv.h"
  36 #include "xfs_ialloc.h"
  37 #include "scrub/xfs_scrub.h"
  38 #include "scrub/scrub.h"
  39 #include "scrub/common.h"
  40 #include "scrub/trace.h"
  41
  42 /* Set us up to scrub parents. */
  43 int
  44 xfs_scrub_setup_parent(
  45         struct xfs_scrub_context        *sc,
  46         struct xfs_inode                *ip)
  47 {
  48         return xfs_scrub_setup_inode_contents(sc, ip, 0);
  49 }
  50
  51 /* Parent pointers */
  52
  53 /* Look for an entry in a parent pointing to this inode. */
  54
  55 struct xfs_scrub_parent_ctx {
  56         struct dir_context              dc;
  57         xfs_ino_t                       ino;
  58         xfs_nlink_t                     nlink;
  59 };
  60
  61 /* Look for a single entry in a directory pointing to an inode. */
  62 STATIC int
  63 xfs_scrub_parent_actor(
  64         struct dir_context              *dc,
  65         const char                      *name,
  66         int                             namelen,
  67         loff_t                          pos,
  68         u64                             ino,
  69         unsigned                        type)
  70 {
  71         struct xfs_scrub_parent_ctx     *spc;
  72
  73         spc = container_of(dc, struct xfs_scrub_parent_ctx, dc);
  74         if (spc->ino == ino)
  75                 spc->nlink++;
  76         return 0;
  77 }
  78
  79 /* Count the number of dentries in the parent dir that point to this inode. */
  80 STATIC int
  81 xfs_scrub_parent_count_parent_dentries(
  82         struct xfs_scrub_context        *sc,
  83         struct xfs_inode                *parent,
  84         xfs_nlink_t                     *nlink)
  85 {
  86         struct xfs_scrub_parent_ctx     spc = {
  87                 .dc.actor = xfs_scrub_parent_actor,
  88                 .dc.pos = 0,
  89                 .ino = sc->ip->i_ino,
  90                 .nlink = 0,
  91         };
  92         size_t                          bufsize;
  93         loff_t                          oldpos;
  94         uint                            lock_mode;
  95         int                             error = 0;
  96
  97         /*
  98          * If there are any blocks, read-ahead block 0 as we're almost
  99          * certain to have the next operation be a read there.  This is
 100          * how we guarantee that the parent's extent map has been loaded,
 101          * if there is one.
 102          */
 103         lock_mode = xfs_ilock_data_map_shared(parent);
 104         if (parent->i_d.di_nextents > 0)
 105                 error = xfs_dir3_data_readahead(parent, 0, -1);
 106         xfs_iunlock(parent, lock_mode);
 107         if (error)
 108                 return error;
 109
 110         /*
 111          * Iterate the parent dir to confirm that there is
 112          * exactly one entry pointing back to the inode being
 113          * scanned.
 114          */
 115         bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
 116                         parent->i_d.di_size);
 117         oldpos = 0;
 118         while (true) {
 119                 error = xfs_readdir(sc->tp, parent, &spc.dc, bufsize);
 120                 if (error)
 121                         goto out;
 122                 if (oldpos == spc.dc.pos)
 123                         break;
 124                 oldpos = spc.dc.pos;
 125         }
 126         *nlink = spc.nlink;
 127 out:
 128         return error;
 129 }
 130
 131 /*
 132  * Given the inode number of the alleged parent of the inode being
 133  * scrubbed, try to validate that the parent has exactly one directory
 134  * entry pointing back to the inode being scrubbed.
 135  */
 136 STATIC int
 137 xfs_scrub_parent_validate(
 138         struct xfs_scrub_context        *sc,
 139         xfs_ino_t                       dnum,
 140         bool                            *try_again)
 141 {
 142         struct xfs_mount                *mp = sc->mp;
 143         struct xfs_inode                *dp = NULL;
 144         xfs_nlink_t                     expected_nlink;
 145         xfs_nlink_t                     nlink;
 146         int                             error = 0;
 147
 148         *try_again = false;
 149
 150         /* '..' must not point to ourselves. */
 151         if (sc->ip->i_ino == dnum) {
 152                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
 153                 goto out;
 154         }
 155
 156         /*
 157          * If we're an unlinked directory, the parent /won't/ have a link
 158          * to us.  Otherwise, it should have one link.
 159          */
 160         expected_nlink = VFS_I(sc->ip)->i_nlink == 0 ? 0 : 1;
 161
 162         /*
 163          * Grab this parent inode.  We release the inode before we
 164          * cancel the scrub transaction.  Since we're don't know a
 165          * priori that releasing the inode won't trigger eofblocks
 166          * cleanup (which allocates what would be a nested transaction)
 167          * if the parent pointer erroneously points to a file, we
 168          * can't use DONTCACHE here because DONTCACHE inodes can trigger
 169          * immediate inactive cleanup of the inode.
 170          */
 171         error = xfs_iget(mp, sc->tp, dnum, 0, 0, &dp);
 172         if (!xfs_scrub_fblock_xref_process_error(sc, XFS_DATA_FORK, 0, &error))
 173                 goto out;
 174         if (dp == sc->ip || !S_ISDIR(VFS_I(dp)->i_mode)) {
 175                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
 176                 goto out_rele;
 177         }
 178
 179         /*
 180          * We prefer to keep the inode locked while we lock and search
 181          * its alleged parent for a forward reference.  If we can grab
 182          * the iolock, validate the pointers and we're done.  We must
 183          * use nowait here to avoid an ABBA deadlock on the parent and
 184          * the child inodes.
 185          */
 186         if (xfs_ilock_nowait(dp, XFS_IOLOCK_SHARED)) {
 187                 error = xfs_scrub_parent_count_parent_dentries(sc, dp, &nlink);
 188                 if (!xfs_scrub_fblock_xref_process_error(sc, XFS_DATA_FORK, 0,
 189                                 &error))
 190                         goto out_unlock;
 191                 if (nlink != expected_nlink)
 192                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
 193                 goto out_unlock;
 194         }
 195
 196         /*
 197          * The game changes if we get here.  We failed to lock the parent,
 198          * so we're going to try to verify both pointers while only holding
 199          * one lock so as to avoid deadlocking with something that's actually
 200          * trying to traverse down the directory tree.
 201          */
 202         xfs_iunlock(sc->ip, sc->ilock_flags);
 203         sc->ilock_flags = 0;
 204         xfs_ilock(dp, XFS_IOLOCK_SHARED);
 205
 206         /* Go looking for our dentry. */
 207         error = xfs_scrub_parent_count_parent_dentries(sc, dp, &nlink);
 208         if (!xfs_scrub_fblock_xref_process_error(sc, XFS_DATA_FORK, 0, &error))
 209                 goto out_unlock;
 210
 211         /* Drop the parent lock, relock this inode. */
 212         xfs_iunlock(dp, XFS_IOLOCK_SHARED);
 213         sc->ilock_flags = XFS_IOLOCK_EXCL;
 214         xfs_ilock(sc->ip, sc->ilock_flags);
 215
 216         /*
 217          * If we're an unlinked directory, the parent /won't/ have a link
 218          * to us.  Otherwise, it should have one link.  We have to re-set
 219          * it here because we dropped the lock on sc->ip.
 220          */
 221         expected_nlink = VFS_I(sc->ip)->i_nlink == 0 ? 0 : 1;
 222
 223         /* Look up '..' to see if the inode changed. */
 224         error = xfs_dir_lookup(sc->tp, sc->ip, &xfs_name_dotdot, &dnum, NULL);
 225         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0, &error))
 226                 goto out_rele;
 227
 228         /* Drat, parent changed.  Try again! */
 229         if (dnum != dp->i_ino) {
 230                 iput(VFS_I(dp));
 231                 *try_again = true;
 232                 return 0;
 233         }
 234         iput(VFS_I(dp));
 235
 236         /*
 237          * '..' didn't change, so check that there was only one entry
 238          * for us in the parent.
 239          */
 240         if (nlink != expected_nlink)
 241                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
 242         return error;
 243
 244 out_unlock:
 245         xfs_iunlock(dp, XFS_IOLOCK_SHARED);
 246 out_rele:
 247         iput(VFS_I(dp));
 248 out:
 249         return error;
 250 }
 251
 252 /* Scrub a parent pointer. */
 253 int
 254 xfs_scrub_parent(
 255         struct xfs_scrub_context        *sc)
 256 {
 257         struct xfs_mount                *mp = sc->mp;
 258         xfs_ino_t                       dnum;
 259         bool                            try_again;
 260         int                             tries = 0;
 261         int                             error = 0;
 262
 263         /*
 264          * If we're a directory, check that the '..' link points up to
 265          * a directory that has one entry pointing to us.
 266          */
 267         if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
 268                 return -ENOENT;
 269
 270         /* We're not a special inode, are we? */
 271         if (!xfs_verify_dir_ino(mp, sc->ip->i_ino)) {
 272                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
 273                 goto out;
 274         }
 275
 276         /*
 277          * The VFS grabs a read or write lock via i_rwsem before it reads
 278          * or writes to a directory.  If we've gotten this far we've
 279          * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
 280          * getting a write lock on i_rwsem.  Therefore, it is safe for us
 281          * to drop the ILOCK here in order to do directory lookups.
 282          */
 283         sc->ilock_flags &= ~(XFS_ILOCK_EXCL | XFS_MMAPLOCK_EXCL);
 284         xfs_iunlock(sc->ip, XFS_ILOCK_EXCL | XFS_MMAPLOCK_EXCL);
 285
 286         /* Look up '..' */
 287         error = xfs_dir_lookup(sc->tp, sc->ip, &xfs_name_dotdot, &dnum, NULL);
 288         if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0, &error))
 289                 goto out;
 290         if (!xfs_verify_dir_ino(mp, dnum)) {
 291                 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
 292                 goto out;
 293         }
 294
 295         /* Is this the root dir?  Then '..' must point to itself. */
 296         if (sc->ip == mp->m_rootip) {
 297                 if (sc->ip->i_ino != mp->m_sb.sb_rootino ||
 298                     sc->ip->i_ino != dnum)
 299                         xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
 300                 goto out;
 301         }
 302
 303         do {
 304                 error = xfs_scrub_parent_validate(sc, dnum, &try_again);
 305                 if (error)
 306                         goto out;
 307         } while (try_again && ++tries < 20);
 308
 309         /*
 310          * We gave it our best shot but failed, so mark this scrub
 311          * incomplete.  Userspace can decide if it wants to try again.
 312          */
 313         if (try_again && tries == 20)
 314                 xfs_scrub_set_incomplete(sc);
 315 out:
 316         return error;
 317 }