2 * Copyright (C) 2017 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
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.
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.
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.
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"
29 #include "xfs_log_format.h"
30 #include "xfs_trans.h"
32 #include "xfs_inode.h"
33 #include "xfs_icache.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"
42 /* Set us up to scrub parents. */
44 xfs_scrub_setup_parent(
45 struct xfs_scrub_context *sc,
48 return xfs_scrub_setup_inode_contents(sc, ip, 0);
53 /* Look for an entry in a parent pointing to this inode. */
55 struct xfs_scrub_parent_ctx {
56 struct dir_context dc;
61 /* Look for a single entry in a directory pointing to an inode. */
63 xfs_scrub_parent_actor(
64 struct dir_context *dc,
71 struct xfs_scrub_parent_ctx *spc;
73 spc = container_of(dc, struct xfs_scrub_parent_ctx, dc);
79 /* Count the number of dentries in the parent dir that point to this inode. */
81 xfs_scrub_parent_count_parent_dentries(
82 struct xfs_scrub_context *sc,
83 struct xfs_inode *parent,
86 struct xfs_scrub_parent_ctx spc = {
87 .dc.actor = xfs_scrub_parent_actor,
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,
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);
111 * Iterate the parent dir to confirm that there is
112 * exactly one entry pointing back to the inode being
115 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE,
116 parent->i_d.di_size);
119 error = xfs_readdir(sc->tp, parent, &spc.dc, bufsize);
122 if (oldpos == spc.dc.pos)
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.
137 xfs_scrub_parent_validate(
138 struct xfs_scrub_context *sc,
142 struct xfs_mount *mp = sc->mp;
143 struct xfs_inode *dp = NULL;
144 xfs_nlink_t expected_nlink;
150 /* '..' must not point to ourselves. */
151 if (sc->ip->i_ino == dnum) {
152 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
157 * If we're an unlinked directory, the parent /won't/ have a link
158 * to us. Otherwise, it should have one link.
160 expected_nlink = VFS_I(sc->ip)->i_nlink == 0 ? 0 : 1;
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.
171 * If _iget returns -EINVAL then the parent inode number is garbage
172 * and the directory is corrupt. If the _iget returns -EFSCORRUPTED
173 * or -EFSBADCRC then the parent is corrupt which is a cross
174 * referencing error. Any other error is an operational error.
176 error = xfs_iget(mp, sc->tp, dnum, XFS_IGET_UNTRUSTED, 0, &dp);
177 if (error == -EINVAL) {
178 error = -EFSCORRUPTED;
179 xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0, &error);
182 if (!xfs_scrub_fblock_xref_process_error(sc, XFS_DATA_FORK, 0, &error))
184 if (dp == sc->ip || !S_ISDIR(VFS_I(dp)->i_mode)) {
185 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
190 * We prefer to keep the inode locked while we lock and search
191 * its alleged parent for a forward reference. If we can grab
192 * the iolock, validate the pointers and we're done. We must
193 * use nowait here to avoid an ABBA deadlock on the parent and
196 if (xfs_ilock_nowait(dp, XFS_IOLOCK_SHARED)) {
197 error = xfs_scrub_parent_count_parent_dentries(sc, dp, &nlink);
198 if (!xfs_scrub_fblock_xref_process_error(sc, XFS_DATA_FORK, 0,
201 if (nlink != expected_nlink)
202 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
207 * The game changes if we get here. We failed to lock the parent,
208 * so we're going to try to verify both pointers while only holding
209 * one lock so as to avoid deadlocking with something that's actually
210 * trying to traverse down the directory tree.
212 xfs_iunlock(sc->ip, sc->ilock_flags);
214 xfs_ilock(dp, XFS_IOLOCK_SHARED);
216 /* Go looking for our dentry. */
217 error = xfs_scrub_parent_count_parent_dentries(sc, dp, &nlink);
218 if (!xfs_scrub_fblock_xref_process_error(sc, XFS_DATA_FORK, 0, &error))
221 /* Drop the parent lock, relock this inode. */
222 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
223 sc->ilock_flags = XFS_IOLOCK_EXCL;
224 xfs_ilock(sc->ip, sc->ilock_flags);
227 * If we're an unlinked directory, the parent /won't/ have a link
228 * to us. Otherwise, it should have one link. We have to re-set
229 * it here because we dropped the lock on sc->ip.
231 expected_nlink = VFS_I(sc->ip)->i_nlink == 0 ? 0 : 1;
233 /* Look up '..' to see if the inode changed. */
234 error = xfs_dir_lookup(sc->tp, sc->ip, &xfs_name_dotdot, &dnum, NULL);
235 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0, &error))
238 /* Drat, parent changed. Try again! */
239 if (dnum != dp->i_ino) {
247 * '..' didn't change, so check that there was only one entry
248 * for us in the parent.
250 if (nlink != expected_nlink)
251 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
255 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
262 /* Scrub a parent pointer. */
265 struct xfs_scrub_context *sc)
267 struct xfs_mount *mp = sc->mp;
274 * If we're a directory, check that the '..' link points up to
275 * a directory that has one entry pointing to us.
277 if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
280 /* We're not a special inode, are we? */
281 if (!xfs_verify_dir_ino(mp, sc->ip->i_ino)) {
282 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
287 * The VFS grabs a read or write lock via i_rwsem before it reads
288 * or writes to a directory. If we've gotten this far we've
289 * already obtained IOLOCK_EXCL, which (since 4.10) is the same as
290 * getting a write lock on i_rwsem. Therefore, it is safe for us
291 * to drop the ILOCK here in order to do directory lookups.
293 sc->ilock_flags &= ~(XFS_ILOCK_EXCL | XFS_MMAPLOCK_EXCL);
294 xfs_iunlock(sc->ip, XFS_ILOCK_EXCL | XFS_MMAPLOCK_EXCL);
297 error = xfs_dir_lookup(sc->tp, sc->ip, &xfs_name_dotdot, &dnum, NULL);
298 if (!xfs_scrub_fblock_process_error(sc, XFS_DATA_FORK, 0, &error))
300 if (!xfs_verify_dir_ino(mp, dnum)) {
301 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
305 /* Is this the root dir? Then '..' must point to itself. */
306 if (sc->ip == mp->m_rootip) {
307 if (sc->ip->i_ino != mp->m_sb.sb_rootino ||
308 sc->ip->i_ino != dnum)
309 xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
314 error = xfs_scrub_parent_validate(sc, dnum, &try_again);
317 } while (try_again && ++tries < 20);
320 * We gave it our best shot but failed, so mark this scrub
321 * incomplete. Userspace can decide if it wants to try again.
323 if (try_again && tries == 20)
324 xfs_scrub_set_incomplete(sc);