xref: /openbmc/linux/fs/xfs/scrub/parent.c (revision 2359ccdd)
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 	 * 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.
175 	 */
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);
180 		goto out;
181 	}
182 	if (!xfs_scrub_fblock_xref_process_error(sc, XFS_DATA_FORK, 0, &error))
183 		goto out;
184 	if (dp == sc->ip || !S_ISDIR(VFS_I(dp)->i_mode)) {
185 		xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
186 		goto out_rele;
187 	}
188 
189 	/*
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
194 	 * the child inodes.
195 	 */
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,
199 				&error))
200 			goto out_unlock;
201 		if (nlink != expected_nlink)
202 			xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
203 		goto out_unlock;
204 	}
205 
206 	/*
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.
211 	 */
212 	xfs_iunlock(sc->ip, sc->ilock_flags);
213 	sc->ilock_flags = 0;
214 	xfs_ilock(dp, XFS_IOLOCK_SHARED);
215 
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))
219 		goto out_unlock;
220 
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);
225 
226 	/*
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.
230 	 */
231 	expected_nlink = VFS_I(sc->ip)->i_nlink == 0 ? 0 : 1;
232 
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))
236 		goto out_rele;
237 
238 	/* Drat, parent changed.  Try again! */
239 	if (dnum != dp->i_ino) {
240 		iput(VFS_I(dp));
241 		*try_again = true;
242 		return 0;
243 	}
244 	iput(VFS_I(dp));
245 
246 	/*
247 	 * '..' didn't change, so check that there was only one entry
248 	 * for us in the parent.
249 	 */
250 	if (nlink != expected_nlink)
251 		xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
252 	return error;
253 
254 out_unlock:
255 	xfs_iunlock(dp, XFS_IOLOCK_SHARED);
256 out_rele:
257 	iput(VFS_I(dp));
258 out:
259 	return error;
260 }
261 
262 /* Scrub a parent pointer. */
263 int
264 xfs_scrub_parent(
265 	struct xfs_scrub_context	*sc)
266 {
267 	struct xfs_mount		*mp = sc->mp;
268 	xfs_ino_t			dnum;
269 	bool				try_again;
270 	int				tries = 0;
271 	int				error = 0;
272 
273 	/*
274 	 * If we're a directory, check that the '..' link points up to
275 	 * a directory that has one entry pointing to us.
276 	 */
277 	if (!S_ISDIR(VFS_I(sc->ip)->i_mode))
278 		return -ENOENT;
279 
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);
283 		goto out;
284 	}
285 
286 	/*
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.
292 	 */
293 	sc->ilock_flags &= ~(XFS_ILOCK_EXCL | XFS_MMAPLOCK_EXCL);
294 	xfs_iunlock(sc->ip, XFS_ILOCK_EXCL | XFS_MMAPLOCK_EXCL);
295 
296 	/* Look up '..' */
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))
299 		goto out;
300 	if (!xfs_verify_dir_ino(mp, dnum)) {
301 		xfs_scrub_fblock_set_corrupt(sc, XFS_DATA_FORK, 0);
302 		goto out;
303 	}
304 
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);
310 		goto out;
311 	}
312 
313 	do {
314 		error = xfs_scrub_parent_validate(sc, dnum, &try_again);
315 		if (error)
316 			goto out;
317 	} while (try_again && ++tries < 20);
318 
319 	/*
320 	 * We gave it our best shot but failed, so mark this scrub
321 	 * incomplete.  Userspace can decide if it wants to try again.
322 	 */
323 	if (try_again && tries == 20)
324 		xfs_scrub_set_incomplete(sc);
325 out:
326 	return error;
327 }
328