xref: /linux/fs/xfs/scrub/nlinks_repair.c (revision d85fe250f2eb61e19029e9e0d30095c5f646e2f2)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (c) 2021-2024 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <djwong@kernel.org>
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_trans_resv.h"
11 #include "xfs_mount.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans.h"
14 #include "xfs_inode.h"
15 #include "xfs_icache.h"
16 #include "xfs_bmap_util.h"
17 #include "xfs_iwalk.h"
18 #include "xfs_ialloc.h"
19 #include "xfs_sb.h"
20 #include "xfs_ag.h"
21 #include "scrub/scrub.h"
22 #include "scrub/common.h"
23 #include "scrub/repair.h"
24 #include "scrub/xfile.h"
25 #include "scrub/xfarray.h"
26 #include "scrub/iscan.h"
27 #include "scrub/orphanage.h"
28 #include "scrub/nlinks.h"
29 #include "scrub/trace.h"
30 #include "scrub/tempfile.h"
31 
32 /*
33  * Live Inode Link Count Repair
34  * ============================
35  *
36  * Use the live inode link count information that we collected to replace the
37  * nlink values of the incore inodes.  A scrub->repair cycle should have left
38  * the live data and hooks active, so this is safe so long as we make sure the
39  * inode is locked.
40  */
41 
42 /* Set up to repair inode link counts. */
43 int
44 xrep_setup_nlinks(
45 	struct xfs_scrub	*sc)
46 {
47 	return xrep_orphanage_try_create(sc);
48 }
49 
50 /*
51  * Inodes that aren't the root directory or the orphanage, have a nonzero link
52  * count, and no observed parents should be moved to the orphanage.
53  */
54 static inline bool
55 xrep_nlinks_is_orphaned(
56 	struct xfs_scrub	*sc,
57 	struct xfs_inode	*ip,
58 	unsigned int		actual_nlink,
59 	const struct xchk_nlink	*obs)
60 {
61 	struct xfs_mount	*mp = ip->i_mount;
62 
63 	if (obs->parents != 0)
64 		return false;
65 	if (ip == mp->m_rootip || ip == sc->orphanage)
66 		return false;
67 	return actual_nlink != 0;
68 }
69 
70 /* Remove an inode from the unlinked list. */
71 STATIC int
72 xrep_nlinks_iunlink_remove(
73 	struct xfs_scrub	*sc)
74 {
75 	struct xfs_perag	*pag;
76 	int			error;
77 
78 	pag = xfs_perag_get(sc->mp, XFS_INO_TO_AGNO(sc->mp, sc->ip->i_ino));
79 	error = xfs_iunlink_remove(sc->tp, pag, sc->ip);
80 	xfs_perag_put(pag);
81 	return error;
82 }
83 
84 /*
85  * Correct the link count of the given inode.  Because we have to grab locks
86  * and resources in a certain order, it's possible that this will be a no-op.
87  */
88 STATIC int
89 xrep_nlinks_repair_inode(
90 	struct xchk_nlink_ctrs	*xnc)
91 {
92 	struct xchk_nlink	obs;
93 	struct xfs_scrub	*sc = xnc->sc;
94 	struct xfs_mount	*mp = sc->mp;
95 	struct xfs_inode	*ip = sc->ip;
96 	uint64_t		total_links;
97 	uint64_t		actual_nlink;
98 	bool			orphanage_available = false;
99 	bool			dirty = false;
100 	int			error;
101 
102 	/*
103 	 * Ignore temporary files being used to stage repairs, since we assume
104 	 * they're correct for non-directories, and the directory repair code
105 	 * doesn't bump the link counts for the children.
106 	 */
107 	if (xrep_is_tempfile(ip))
108 		return 0;
109 
110 	/*
111 	 * If the filesystem has an orphanage attached to the scrub context,
112 	 * prepare for a link count repair that could involve @ip being adopted
113 	 * by the lost+found.
114 	 */
115 	if (xrep_orphanage_can_adopt(sc)) {
116 		error = xrep_orphanage_iolock_two(sc);
117 		if (error)
118 			return error;
119 
120 		error = xrep_adoption_trans_alloc(sc, &xnc->adoption);
121 		if (error) {
122 			xchk_iunlock(sc, XFS_IOLOCK_EXCL);
123 			xrep_orphanage_iunlock(sc, XFS_IOLOCK_EXCL);
124 		} else {
125 			orphanage_available = true;
126 		}
127 	}
128 
129 	/*
130 	 * Either there is no orphanage or we couldn't allocate resources for
131 	 * that kind of update.  Let's try again with only the resources we
132 	 * need for a simple link count update, since that's much more common.
133 	 */
134 	if (!orphanage_available) {
135 		xchk_ilock(sc, XFS_IOLOCK_EXCL);
136 
137 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0,
138 				&sc->tp);
139 		if (error)
140 			return error;
141 
142 		xchk_ilock(sc, XFS_ILOCK_EXCL);
143 		xfs_trans_ijoin(sc->tp, ip, 0);
144 	}
145 
146 	mutex_lock(&xnc->lock);
147 
148 	if (xchk_iscan_aborted(&xnc->collect_iscan)) {
149 		error = -ECANCELED;
150 		goto out_scanlock;
151 	}
152 
153 	error = xfarray_load_sparse(xnc->nlinks, ip->i_ino, &obs);
154 	if (error)
155 		goto out_scanlock;
156 
157 	/*
158 	 * We're done accessing the shared scan data, so we can drop the lock.
159 	 * We still hold @ip's ILOCK, so its link count cannot change.
160 	 */
161 	mutex_unlock(&xnc->lock);
162 
163 	total_links = xchk_nlink_total(ip, &obs);
164 	actual_nlink = VFS_I(ip)->i_nlink;
165 
166 	/*
167 	 * Non-directories cannot have directories pointing up to them.
168 	 *
169 	 * We previously set error to zero, but set it again because one static
170 	 * checker author fears that programmers will fail to maintain this
171 	 * invariant and built their tool to flag this as a security risk.  A
172 	 * different tool author made their bot complain about the redundant
173 	 * store.  This is a never-ending and stupid battle; both tools missed
174 	 * *actual bugs* elsewhere; and I no longer care.
175 	 */
176 	if (!S_ISDIR(VFS_I(ip)->i_mode) && obs.children != 0) {
177 		trace_xrep_nlinks_unfixable_inode(mp, ip, &obs);
178 		error = 0;
179 		goto out_trans;
180 	}
181 
182 	/*
183 	 * Decide if we're going to move this file to the orphanage, and fix
184 	 * up the incore link counts if we are.
185 	 */
186 	if (orphanage_available &&
187 	    xrep_nlinks_is_orphaned(sc, ip, actual_nlink, &obs)) {
188 		/* Figure out what name we're going to use here. */
189 		error = xrep_adoption_compute_name(&xnc->adoption, &xnc->xname);
190 		if (error)
191 			goto out_trans;
192 
193 		/*
194 		 * Reattach this file to the directory tree by moving it to
195 		 * the orphanage per the adoption parameters that we already
196 		 * computed.
197 		 */
198 		error = xrep_adoption_move(&xnc->adoption);
199 		if (error)
200 			goto out_trans;
201 
202 		/*
203 		 * Re-read the link counts since the reparenting will have
204 		 * updated our scan info.
205 		 */
206 		mutex_lock(&xnc->lock);
207 		error = xfarray_load_sparse(xnc->nlinks, ip->i_ino, &obs);
208 		mutex_unlock(&xnc->lock);
209 		if (error)
210 			goto out_trans;
211 
212 		total_links = xchk_nlink_total(ip, &obs);
213 		actual_nlink = VFS_I(ip)->i_nlink;
214 		dirty = true;
215 	}
216 
217 	/*
218 	 * If this inode is linked from the directory tree and on the unlinked
219 	 * list, remove it from the unlinked list.
220 	 */
221 	if (total_links > 0 && xfs_inode_on_unlinked_list(ip)) {
222 		error = xrep_nlinks_iunlink_remove(sc);
223 		if (error)
224 			goto out_trans;
225 		dirty = true;
226 	}
227 
228 	/*
229 	 * If this inode is not linked from the directory tree yet not on the
230 	 * unlinked list, put it on the unlinked list.
231 	 */
232 	if (total_links == 0 && !xfs_inode_on_unlinked_list(ip)) {
233 		error = xfs_iunlink(sc->tp, ip);
234 		if (error)
235 			goto out_trans;
236 		dirty = true;
237 	}
238 
239 	/* Commit the new link count if it changed. */
240 	if (total_links != actual_nlink) {
241 		trace_xrep_nlinks_update_inode(mp, ip, &obs);
242 
243 		set_nlink(VFS_I(ip), min_t(unsigned long long, total_links,
244 					   XFS_NLINK_PINNED));
245 		dirty = true;
246 	}
247 
248 	if (!dirty) {
249 		error = 0;
250 		goto out_trans;
251 	}
252 
253 	xfs_trans_log_inode(sc->tp, ip, XFS_ILOG_CORE);
254 
255 	error = xrep_trans_commit(sc);
256 	goto out_unlock;
257 
258 out_scanlock:
259 	mutex_unlock(&xnc->lock);
260 out_trans:
261 	xchk_trans_cancel(sc);
262 out_unlock:
263 	xchk_iunlock(sc, XFS_ILOCK_EXCL);
264 	if (orphanage_available) {
265 		xrep_orphanage_iunlock(sc, XFS_ILOCK_EXCL);
266 		xrep_orphanage_iunlock(sc, XFS_IOLOCK_EXCL);
267 	}
268 	xchk_iunlock(sc, XFS_IOLOCK_EXCL);
269 	return error;
270 }
271 
272 /*
273  * Try to visit every inode in the filesystem for repairs.  Move on if we can't
274  * grab an inode, since we're still making forward progress.
275  */
276 static int
277 xrep_nlinks_iter(
278 	struct xchk_nlink_ctrs	*xnc,
279 	struct xfs_inode	**ipp)
280 {
281 	int			error;
282 
283 	do {
284 		error = xchk_iscan_iter(&xnc->compare_iscan, ipp);
285 	} while (error == -EBUSY);
286 
287 	return error;
288 }
289 
290 /* Commit the new inode link counters. */
291 int
292 xrep_nlinks(
293 	struct xfs_scrub	*sc)
294 {
295 	struct xchk_nlink_ctrs	*xnc = sc->buf;
296 	int			error;
297 
298 	/*
299 	 * We need ftype for an accurate count of the number of child
300 	 * subdirectory links.  Child subdirectories with a back link (dotdot
301 	 * entry) but no forward link are moved to the orphanage, so we cannot
302 	 * repair the link count of the parent directory based on the back link
303 	 * count alone.  Filesystems without ftype support are rare (old V4) so
304 	 * we just skip out here.
305 	 */
306 	if (!xfs_has_ftype(sc->mp))
307 		return -EOPNOTSUPP;
308 
309 	/*
310 	 * Use the inobt to walk all allocated inodes to compare and fix the
311 	 * link counts.  Retry iget every tenth of a second for up to 30
312 	 * seconds -- even if repair misses a few inodes, we still try to fix
313 	 * as many of them as we can.
314 	 */
315 	xchk_iscan_start(sc, 30000, 100, &xnc->compare_iscan);
316 	ASSERT(sc->ip == NULL);
317 
318 	while ((error = xrep_nlinks_iter(xnc, &sc->ip)) == 1) {
319 		/*
320 		 * Commit the scrub transaction so that we can create repair
321 		 * transactions with the correct reservations.
322 		 */
323 		xchk_trans_cancel(sc);
324 
325 		error = xrep_nlinks_repair_inode(xnc);
326 		xchk_iscan_mark_visited(&xnc->compare_iscan, sc->ip);
327 		xchk_irele(sc, sc->ip);
328 		sc->ip = NULL;
329 		if (error)
330 			break;
331 
332 		if (xchk_should_terminate(sc, &error))
333 			break;
334 
335 		/*
336 		 * Create a new empty transaction so that we can advance the
337 		 * iscan cursor without deadlocking if the inobt has a cycle.
338 		 * We can only push the inactivation workqueues with an empty
339 		 * transaction.
340 		 */
341 		error = xchk_trans_alloc_empty(sc);
342 		if (error)
343 			break;
344 	}
345 	xchk_iscan_iter_finish(&xnc->compare_iscan);
346 	xchk_iscan_teardown(&xnc->compare_iscan);
347 
348 	return error;
349 }
350