xref: /linux/fs/xfs/scrub/nlinks_repair.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 "xfs_dir2.h"
22 #include "xfs_parent.h"
23 #include "scrub/scrub.h"
24 #include "scrub/common.h"
25 #include "scrub/repair.h"
26 #include "scrub/xfile.h"
27 #include "scrub/xfarray.h"
28 #include "scrub/iscan.h"
29 #include "scrub/orphanage.h"
30 #include "scrub/nlinks.h"
31 #include "scrub/trace.h"
32 #include "scrub/tempfile.h"
33 
34 /*
35  * Live Inode Link Count Repair
36  * ============================
37  *
38  * Use the live inode link count information that we collected to replace the
39  * nlink values of the incore inodes.  A scrub->repair cycle should have left
40  * the live data and hooks active, so this is safe so long as we make sure the
41  * inode is locked.
42  */
43 
44 /* Set up to repair inode link counts. */
45 int
46 xrep_setup_nlinks(
47 	struct xfs_scrub	*sc)
48 {
49 	return xrep_orphanage_try_create(sc);
50 }
51 
52 /*
53  * Inodes that aren't the root directory or the orphanage, have a nonzero link
54  * count, and no observed parents should be moved to the orphanage.
55  */
56 static inline bool
57 xrep_nlinks_is_orphaned(
58 	struct xfs_scrub	*sc,
59 	struct xfs_inode	*ip,
60 	unsigned int		actual_nlink,
61 	const struct xchk_nlink	*obs)
62 {
63 	struct xfs_mount	*mp = ip->i_mount;
64 
65 	if (obs->parents != 0)
66 		return false;
67 	if (ip == mp->m_rootip || ip == sc->orphanage)
68 		return false;
69 	return actual_nlink != 0;
70 }
71 
72 /* Remove an inode from the unlinked list. */
73 STATIC int
74 xrep_nlinks_iunlink_remove(
75 	struct xfs_scrub	*sc)
76 {
77 	struct xfs_perag	*pag;
78 	int			error;
79 
80 	pag = xfs_perag_get(sc->mp, XFS_INO_TO_AGNO(sc->mp, sc->ip->i_ino));
81 	error = xfs_iunlink_remove(sc->tp, pag, sc->ip);
82 	xfs_perag_put(pag);
83 	return error;
84 }
85 
86 /*
87  * Correct the link count of the given inode.  Because we have to grab locks
88  * and resources in a certain order, it's possible that this will be a no-op.
89  */
90 STATIC int
91 xrep_nlinks_repair_inode(
92 	struct xchk_nlink_ctrs	*xnc)
93 {
94 	struct xchk_nlink	obs;
95 	struct xfs_scrub	*sc = xnc->sc;
96 	struct xfs_mount	*mp = sc->mp;
97 	struct xfs_inode	*ip = sc->ip;
98 	uint64_t		total_links;
99 	uint64_t		actual_nlink;
100 	bool			orphanage_available = false;
101 	bool			dirty = false;
102 	int			error;
103 
104 	/*
105 	 * Ignore temporary files being used to stage repairs, since we assume
106 	 * they're correct for non-directories, and the directory repair code
107 	 * doesn't bump the link counts for the children.
108 	 */
109 	if (xrep_is_tempfile(ip))
110 		return 0;
111 
112 	/*
113 	 * If the filesystem has an orphanage attached to the scrub context,
114 	 * prepare for a link count repair that could involve @ip being adopted
115 	 * by the lost+found.
116 	 */
117 	if (xrep_orphanage_can_adopt(sc)) {
118 		error = xrep_orphanage_iolock_two(sc);
119 		if (error)
120 			return error;
121 
122 		error = xrep_adoption_trans_alloc(sc, &xnc->adoption);
123 		if (error) {
124 			xchk_iunlock(sc, XFS_IOLOCK_EXCL);
125 			xrep_orphanage_iunlock(sc, XFS_IOLOCK_EXCL);
126 		} else {
127 			orphanage_available = true;
128 		}
129 	}
130 
131 	/*
132 	 * Either there is no orphanage or we couldn't allocate resources for
133 	 * that kind of update.  Let's try again with only the resources we
134 	 * need for a simple link count update, since that's much more common.
135 	 */
136 	if (!orphanage_available) {
137 		xchk_ilock(sc, XFS_IOLOCK_EXCL);
138 
139 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0,
140 				&sc->tp);
141 		if (error) {
142 			xchk_iunlock(sc, XFS_IOLOCK_EXCL);
143 			return error;
144 		}
145 
146 		xchk_ilock(sc, XFS_ILOCK_EXCL);
147 		xfs_trans_ijoin(sc->tp, ip, 0);
148 	}
149 
150 	mutex_lock(&xnc->lock);
151 
152 	if (xchk_iscan_aborted(&xnc->collect_iscan)) {
153 		error = -ECANCELED;
154 		goto out_scanlock;
155 	}
156 
157 	error = xfarray_load_sparse(xnc->nlinks, ip->i_ino, &obs);
158 	if (error)
159 		goto out_scanlock;
160 
161 	/*
162 	 * We're done accessing the shared scan data, so we can drop the lock.
163 	 * We still hold @ip's ILOCK, so its link count cannot change.
164 	 */
165 	mutex_unlock(&xnc->lock);
166 
167 	total_links = xchk_nlink_total(ip, &obs);
168 	actual_nlink = VFS_I(ip)->i_nlink;
169 
170 	/*
171 	 * Non-directories cannot have directories pointing up to them.
172 	 *
173 	 * We previously set error to zero, but set it again because one static
174 	 * checker author fears that programmers will fail to maintain this
175 	 * invariant and built their tool to flag this as a security risk.  A
176 	 * different tool author made their bot complain about the redundant
177 	 * store.  This is a never-ending and stupid battle; both tools missed
178 	 * *actual bugs* elsewhere; and I no longer care.
179 	 */
180 	if (!S_ISDIR(VFS_I(ip)->i_mode) && obs.children != 0) {
181 		trace_xrep_nlinks_unfixable_inode(mp, ip, &obs);
182 		error = 0;
183 		goto out_trans;
184 	}
185 
186 	/*
187 	 * Decide if we're going to move this file to the orphanage, and fix
188 	 * up the incore link counts if we are.
189 	 */
190 	if (orphanage_available &&
191 	    xrep_nlinks_is_orphaned(sc, ip, actual_nlink, &obs)) {
192 		/* Figure out what name we're going to use here. */
193 		error = xrep_adoption_compute_name(&xnc->adoption, &xnc->xname);
194 		if (error)
195 			goto out_trans;
196 
197 		/*
198 		 * Reattach this file to the directory tree by moving it to
199 		 * the orphanage per the adoption parameters that we already
200 		 * computed.
201 		 */
202 		error = xrep_adoption_move(&xnc->adoption);
203 		if (error)
204 			goto out_trans;
205 
206 		/*
207 		 * Re-read the link counts since the reparenting will have
208 		 * updated our scan info.
209 		 */
210 		mutex_lock(&xnc->lock);
211 		error = xfarray_load_sparse(xnc->nlinks, ip->i_ino, &obs);
212 		mutex_unlock(&xnc->lock);
213 		if (error)
214 			goto out_trans;
215 
216 		total_links = xchk_nlink_total(ip, &obs);
217 		actual_nlink = VFS_I(ip)->i_nlink;
218 		dirty = true;
219 	}
220 
221 	/*
222 	 * If this inode is linked from the directory tree and on the unlinked
223 	 * list, remove it from the unlinked list.
224 	 */
225 	if (total_links > 0 && xfs_inode_on_unlinked_list(ip)) {
226 		error = xrep_nlinks_iunlink_remove(sc);
227 		if (error)
228 			goto out_trans;
229 		dirty = true;
230 	}
231 
232 	/*
233 	 * If this inode is not linked from the directory tree yet not on the
234 	 * unlinked list, put it on the unlinked list.
235 	 */
236 	if (total_links == 0 && !xfs_inode_on_unlinked_list(ip)) {
237 		error = xfs_iunlink(sc->tp, ip);
238 		if (error)
239 			goto out_trans;
240 		dirty = true;
241 	}
242 
243 	/* Commit the new link count if it changed. */
244 	if (total_links != actual_nlink) {
245 		trace_xrep_nlinks_update_inode(mp, ip, &obs);
246 
247 		set_nlink(VFS_I(ip), min_t(unsigned long long, total_links,
248 					   XFS_NLINK_PINNED));
249 		dirty = true;
250 	}
251 
252 	if (!dirty) {
253 		error = 0;
254 		goto out_trans;
255 	}
256 
257 	xfs_trans_log_inode(sc->tp, ip, XFS_ILOG_CORE);
258 
259 	error = xrep_trans_commit(sc);
260 	goto out_unlock;
261 
262 out_scanlock:
263 	mutex_unlock(&xnc->lock);
264 out_trans:
265 	xchk_trans_cancel(sc);
266 out_unlock:
267 	xchk_iunlock(sc, XFS_ILOCK_EXCL);
268 	if (orphanage_available) {
269 		xrep_orphanage_iunlock(sc, XFS_ILOCK_EXCL);
270 		xrep_orphanage_iunlock(sc, XFS_IOLOCK_EXCL);
271 	}
272 	xchk_iunlock(sc, XFS_IOLOCK_EXCL);
273 	return error;
274 }
275 
276 /*
277  * Try to visit every inode in the filesystem for repairs.  Move on if we can't
278  * grab an inode, since we're still making forward progress.
279  */
280 static int
281 xrep_nlinks_iter(
282 	struct xchk_nlink_ctrs	*xnc,
283 	struct xfs_inode	**ipp)
284 {
285 	int			error;
286 
287 	do {
288 		error = xchk_iscan_iter(&xnc->compare_iscan, ipp);
289 	} while (error == -EBUSY);
290 
291 	return error;
292 }
293 
294 /* Commit the new inode link counters. */
295 int
296 xrep_nlinks(
297 	struct xfs_scrub	*sc)
298 {
299 	struct xchk_nlink_ctrs	*xnc = sc->buf;
300 	int			error;
301 
302 	/*
303 	 * We need ftype for an accurate count of the number of child
304 	 * subdirectory links.  Child subdirectories with a back link (dotdot
305 	 * entry) but no forward link are moved to the orphanage, so we cannot
306 	 * repair the link count of the parent directory based on the back link
307 	 * count alone.  Filesystems without ftype support are rare (old V4) so
308 	 * we just skip out here.
309 	 */
310 	if (!xfs_has_ftype(sc->mp))
311 		return -EOPNOTSUPP;
312 
313 	/*
314 	 * Use the inobt to walk all allocated inodes to compare and fix the
315 	 * link counts.  Retry iget every tenth of a second for up to 30
316 	 * seconds -- even if repair misses a few inodes, we still try to fix
317 	 * as many of them as we can.
318 	 */
319 	xchk_iscan_start(sc, 30000, 100, &xnc->compare_iscan);
320 	ASSERT(sc->ip == NULL);
321 
322 	while ((error = xrep_nlinks_iter(xnc, &sc->ip)) == 1) {
323 		/*
324 		 * Commit the scrub transaction so that we can create repair
325 		 * transactions with the correct reservations.
326 		 */
327 		xchk_trans_cancel(sc);
328 
329 		error = xrep_nlinks_repair_inode(xnc);
330 		xchk_iscan_mark_visited(&xnc->compare_iscan, sc->ip);
331 		xchk_irele(sc, sc->ip);
332 		sc->ip = NULL;
333 		if (error)
334 			break;
335 
336 		if (xchk_should_terminate(sc, &error))
337 			break;
338 
339 		/*
340 		 * Create a new empty transaction so that we can advance the
341 		 * iscan cursor without deadlocking if the inobt has a cycle.
342 		 * We can only push the inactivation workqueues with an empty
343 		 * transaction.
344 		 */
345 		error = xchk_trans_alloc_empty(sc);
346 		if (error)
347 			break;
348 	}
349 	xchk_iscan_iter_finish(&xnc->compare_iscan);
350 	xchk_iscan_teardown(&xnc->compare_iscan);
351 
352 	return error;
353 }
354