xref: /linux/fs/xfs/xfs_health.c (revision d4b996f9ef1fe83d9ce9ad5c1ca0bd8231638ce5)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright (C) 2019 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trace.h"
15 #include "xfs_health.h"
16 #include "xfs_ag.h"
17 
18 /*
19  * Warn about metadata corruption that we detected but haven't fixed, and
20  * make sure we're not sitting on anything that would get in the way of
21  * recovery.
22  */
23 void
24 xfs_health_unmount(
25 	struct xfs_mount	*mp)
26 {
27 	struct xfs_perag	*pag;
28 	xfs_agnumber_t		agno;
29 	unsigned int		sick = 0;
30 	unsigned int		checked = 0;
31 	bool			warn = false;
32 
33 	if (XFS_FORCED_SHUTDOWN(mp))
34 		return;
35 
36 	/* Measure AG corruption levels. */
37 	for_each_perag(mp, agno, pag) {
38 		xfs_ag_measure_sickness(pag, &sick, &checked);
39 		if (sick) {
40 			trace_xfs_ag_unfixed_corruption(mp, agno, sick);
41 			warn = true;
42 		}
43 	}
44 
45 	/* Measure realtime volume corruption levels. */
46 	xfs_rt_measure_sickness(mp, &sick, &checked);
47 	if (sick) {
48 		trace_xfs_rt_unfixed_corruption(mp, sick);
49 		warn = true;
50 	}
51 
52 	/*
53 	 * Measure fs corruption and keep the sample around for the warning.
54 	 * See the note below for why we exempt FS_COUNTERS.
55 	 */
56 	xfs_fs_measure_sickness(mp, &sick, &checked);
57 	if (sick & ~XFS_SICK_FS_COUNTERS) {
58 		trace_xfs_fs_unfixed_corruption(mp, sick);
59 		warn = true;
60 	}
61 
62 	if (warn) {
63 		xfs_warn(mp,
64 "Uncorrected metadata errors detected; please run xfs_repair.");
65 
66 		/*
67 		 * We discovered uncorrected metadata problems at some point
68 		 * during this filesystem mount and have advised the
69 		 * administrator to run repair once the unmount completes.
70 		 *
71 		 * However, we must be careful -- when FSCOUNTERS are flagged
72 		 * unhealthy, the unmount procedure omits writing the clean
73 		 * unmount record to the log so that the next mount will run
74 		 * recovery and recompute the summary counters.  In other
75 		 * words, we leave a dirty log to get the counters fixed.
76 		 *
77 		 * Unfortunately, xfs_repair cannot recover dirty logs, so if
78 		 * there were filesystem problems, FSCOUNTERS was flagged, and
79 		 * the administrator takes our advice to run xfs_repair,
80 		 * they'll have to zap the log before repairing structures.
81 		 * We don't really want to encourage this, so we mark the
82 		 * FSCOUNTERS healthy so that a subsequent repair run won't see
83 		 * a dirty log.
84 		 */
85 		if (sick & XFS_SICK_FS_COUNTERS)
86 			xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS);
87 	}
88 }
89 
90 /* Mark unhealthy per-fs metadata. */
91 void
92 xfs_fs_mark_sick(
93 	struct xfs_mount	*mp,
94 	unsigned int		mask)
95 {
96 	ASSERT(!(mask & ~XFS_SICK_FS_PRIMARY));
97 	trace_xfs_fs_mark_sick(mp, mask);
98 
99 	spin_lock(&mp->m_sb_lock);
100 	mp->m_fs_sick |= mask;
101 	mp->m_fs_checked |= mask;
102 	spin_unlock(&mp->m_sb_lock);
103 }
104 
105 /* Mark a per-fs metadata healed. */
106 void
107 xfs_fs_mark_healthy(
108 	struct xfs_mount	*mp,
109 	unsigned int		mask)
110 {
111 	ASSERT(!(mask & ~XFS_SICK_FS_PRIMARY));
112 	trace_xfs_fs_mark_healthy(mp, mask);
113 
114 	spin_lock(&mp->m_sb_lock);
115 	mp->m_fs_sick &= ~mask;
116 	mp->m_fs_checked |= mask;
117 	spin_unlock(&mp->m_sb_lock);
118 }
119 
120 /* Sample which per-fs metadata are unhealthy. */
121 void
122 xfs_fs_measure_sickness(
123 	struct xfs_mount	*mp,
124 	unsigned int		*sick,
125 	unsigned int		*checked)
126 {
127 	spin_lock(&mp->m_sb_lock);
128 	*sick = mp->m_fs_sick;
129 	*checked = mp->m_fs_checked;
130 	spin_unlock(&mp->m_sb_lock);
131 }
132 
133 /* Mark unhealthy realtime metadata. */
134 void
135 xfs_rt_mark_sick(
136 	struct xfs_mount	*mp,
137 	unsigned int		mask)
138 {
139 	ASSERT(!(mask & ~XFS_SICK_RT_PRIMARY));
140 	trace_xfs_rt_mark_sick(mp, mask);
141 
142 	spin_lock(&mp->m_sb_lock);
143 	mp->m_rt_sick |= mask;
144 	mp->m_rt_checked |= mask;
145 	spin_unlock(&mp->m_sb_lock);
146 }
147 
148 /* Mark a realtime metadata healed. */
149 void
150 xfs_rt_mark_healthy(
151 	struct xfs_mount	*mp,
152 	unsigned int		mask)
153 {
154 	ASSERT(!(mask & ~XFS_SICK_RT_PRIMARY));
155 	trace_xfs_rt_mark_healthy(mp, mask);
156 
157 	spin_lock(&mp->m_sb_lock);
158 	mp->m_rt_sick &= ~mask;
159 	mp->m_rt_checked |= mask;
160 	spin_unlock(&mp->m_sb_lock);
161 }
162 
163 /* Sample which realtime metadata are unhealthy. */
164 void
165 xfs_rt_measure_sickness(
166 	struct xfs_mount	*mp,
167 	unsigned int		*sick,
168 	unsigned int		*checked)
169 {
170 	spin_lock(&mp->m_sb_lock);
171 	*sick = mp->m_rt_sick;
172 	*checked = mp->m_rt_checked;
173 	spin_unlock(&mp->m_sb_lock);
174 }
175 
176 /* Mark unhealthy per-ag metadata. */
177 void
178 xfs_ag_mark_sick(
179 	struct xfs_perag	*pag,
180 	unsigned int		mask)
181 {
182 	ASSERT(!(mask & ~XFS_SICK_AG_PRIMARY));
183 	trace_xfs_ag_mark_sick(pag->pag_mount, pag->pag_agno, mask);
184 
185 	spin_lock(&pag->pag_state_lock);
186 	pag->pag_sick |= mask;
187 	pag->pag_checked |= mask;
188 	spin_unlock(&pag->pag_state_lock);
189 }
190 
191 /* Mark per-ag metadata ok. */
192 void
193 xfs_ag_mark_healthy(
194 	struct xfs_perag	*pag,
195 	unsigned int		mask)
196 {
197 	ASSERT(!(mask & ~XFS_SICK_AG_PRIMARY));
198 	trace_xfs_ag_mark_healthy(pag->pag_mount, pag->pag_agno, mask);
199 
200 	spin_lock(&pag->pag_state_lock);
201 	pag->pag_sick &= ~mask;
202 	pag->pag_checked |= mask;
203 	spin_unlock(&pag->pag_state_lock);
204 }
205 
206 /* Sample which per-ag metadata are unhealthy. */
207 void
208 xfs_ag_measure_sickness(
209 	struct xfs_perag	*pag,
210 	unsigned int		*sick,
211 	unsigned int		*checked)
212 {
213 	spin_lock(&pag->pag_state_lock);
214 	*sick = pag->pag_sick;
215 	*checked = pag->pag_checked;
216 	spin_unlock(&pag->pag_state_lock);
217 }
218 
219 /* Mark the unhealthy parts of an inode. */
220 void
221 xfs_inode_mark_sick(
222 	struct xfs_inode	*ip,
223 	unsigned int		mask)
224 {
225 	ASSERT(!(mask & ~XFS_SICK_INO_PRIMARY));
226 	trace_xfs_inode_mark_sick(ip, mask);
227 
228 	spin_lock(&ip->i_flags_lock);
229 	ip->i_sick |= mask;
230 	ip->i_checked |= mask;
231 	spin_unlock(&ip->i_flags_lock);
232 
233 	/*
234 	 * Keep this inode around so we don't lose the sickness report.  Scrub
235 	 * grabs inodes with DONTCACHE assuming that most inode are ok, which
236 	 * is not the case here.
237 	 */
238 	spin_lock(&VFS_I(ip)->i_lock);
239 	VFS_I(ip)->i_state &= ~I_DONTCACHE;
240 	spin_unlock(&VFS_I(ip)->i_lock);
241 }
242 
243 /* Mark parts of an inode healed. */
244 void
245 xfs_inode_mark_healthy(
246 	struct xfs_inode	*ip,
247 	unsigned int		mask)
248 {
249 	ASSERT(!(mask & ~XFS_SICK_INO_PRIMARY));
250 	trace_xfs_inode_mark_healthy(ip, mask);
251 
252 	spin_lock(&ip->i_flags_lock);
253 	ip->i_sick &= ~mask;
254 	ip->i_checked |= mask;
255 	spin_unlock(&ip->i_flags_lock);
256 }
257 
258 /* Sample which parts of an inode are unhealthy. */
259 void
260 xfs_inode_measure_sickness(
261 	struct xfs_inode	*ip,
262 	unsigned int		*sick,
263 	unsigned int		*checked)
264 {
265 	spin_lock(&ip->i_flags_lock);
266 	*sick = ip->i_sick;
267 	*checked = ip->i_checked;
268 	spin_unlock(&ip->i_flags_lock);
269 }
270 
271 /* Mappings between internal sick masks and ioctl sick masks. */
272 
273 struct ioctl_sick_map {
274 	unsigned int		sick_mask;
275 	unsigned int		ioctl_mask;
276 };
277 
278 static const struct ioctl_sick_map fs_map[] = {
279 	{ XFS_SICK_FS_COUNTERS,	XFS_FSOP_GEOM_SICK_COUNTERS},
280 	{ XFS_SICK_FS_UQUOTA,	XFS_FSOP_GEOM_SICK_UQUOTA },
281 	{ XFS_SICK_FS_GQUOTA,	XFS_FSOP_GEOM_SICK_GQUOTA },
282 	{ XFS_SICK_FS_PQUOTA,	XFS_FSOP_GEOM_SICK_PQUOTA },
283 	{ 0, 0 },
284 };
285 
286 static const struct ioctl_sick_map rt_map[] = {
287 	{ XFS_SICK_RT_BITMAP,	XFS_FSOP_GEOM_SICK_RT_BITMAP },
288 	{ XFS_SICK_RT_SUMMARY,	XFS_FSOP_GEOM_SICK_RT_SUMMARY },
289 	{ 0, 0 },
290 };
291 
292 static inline void
293 xfgeo_health_tick(
294 	struct xfs_fsop_geom		*geo,
295 	unsigned int			sick,
296 	unsigned int			checked,
297 	const struct ioctl_sick_map	*m)
298 {
299 	if (checked & m->sick_mask)
300 		geo->checked |= m->ioctl_mask;
301 	if (sick & m->sick_mask)
302 		geo->sick |= m->ioctl_mask;
303 }
304 
305 /* Fill out fs geometry health info. */
306 void
307 xfs_fsop_geom_health(
308 	struct xfs_mount		*mp,
309 	struct xfs_fsop_geom		*geo)
310 {
311 	const struct ioctl_sick_map	*m;
312 	unsigned int			sick;
313 	unsigned int			checked;
314 
315 	geo->sick = 0;
316 	geo->checked = 0;
317 
318 	xfs_fs_measure_sickness(mp, &sick, &checked);
319 	for (m = fs_map; m->sick_mask; m++)
320 		xfgeo_health_tick(geo, sick, checked, m);
321 
322 	xfs_rt_measure_sickness(mp, &sick, &checked);
323 	for (m = rt_map; m->sick_mask; m++)
324 		xfgeo_health_tick(geo, sick, checked, m);
325 }
326 
327 static const struct ioctl_sick_map ag_map[] = {
328 	{ XFS_SICK_AG_SB,	XFS_AG_GEOM_SICK_SB },
329 	{ XFS_SICK_AG_AGF,	XFS_AG_GEOM_SICK_AGF },
330 	{ XFS_SICK_AG_AGFL,	XFS_AG_GEOM_SICK_AGFL },
331 	{ XFS_SICK_AG_AGI,	XFS_AG_GEOM_SICK_AGI },
332 	{ XFS_SICK_AG_BNOBT,	XFS_AG_GEOM_SICK_BNOBT },
333 	{ XFS_SICK_AG_CNTBT,	XFS_AG_GEOM_SICK_CNTBT },
334 	{ XFS_SICK_AG_INOBT,	XFS_AG_GEOM_SICK_INOBT },
335 	{ XFS_SICK_AG_FINOBT,	XFS_AG_GEOM_SICK_FINOBT },
336 	{ XFS_SICK_AG_RMAPBT,	XFS_AG_GEOM_SICK_RMAPBT },
337 	{ XFS_SICK_AG_REFCNTBT,	XFS_AG_GEOM_SICK_REFCNTBT },
338 	{ 0, 0 },
339 };
340 
341 /* Fill out ag geometry health info. */
342 void
343 xfs_ag_geom_health(
344 	struct xfs_perag		*pag,
345 	struct xfs_ag_geometry		*ageo)
346 {
347 	const struct ioctl_sick_map	*m;
348 	unsigned int			sick;
349 	unsigned int			checked;
350 
351 	ageo->ag_sick = 0;
352 	ageo->ag_checked = 0;
353 
354 	xfs_ag_measure_sickness(pag, &sick, &checked);
355 	for (m = ag_map; m->sick_mask; m++) {
356 		if (checked & m->sick_mask)
357 			ageo->ag_checked |= m->ioctl_mask;
358 		if (sick & m->sick_mask)
359 			ageo->ag_sick |= m->ioctl_mask;
360 	}
361 }
362 
363 static const struct ioctl_sick_map ino_map[] = {
364 	{ XFS_SICK_INO_CORE,	XFS_BS_SICK_INODE },
365 	{ XFS_SICK_INO_BMBTD,	XFS_BS_SICK_BMBTD },
366 	{ XFS_SICK_INO_BMBTA,	XFS_BS_SICK_BMBTA },
367 	{ XFS_SICK_INO_BMBTC,	XFS_BS_SICK_BMBTC },
368 	{ XFS_SICK_INO_DIR,	XFS_BS_SICK_DIR },
369 	{ XFS_SICK_INO_XATTR,	XFS_BS_SICK_XATTR },
370 	{ XFS_SICK_INO_SYMLINK,	XFS_BS_SICK_SYMLINK },
371 	{ XFS_SICK_INO_PARENT,	XFS_BS_SICK_PARENT },
372 	{ 0, 0 },
373 };
374 
375 /* Fill out bulkstat health info. */
376 void
377 xfs_bulkstat_health(
378 	struct xfs_inode		*ip,
379 	struct xfs_bulkstat		*bs)
380 {
381 	const struct ioctl_sick_map	*m;
382 	unsigned int			sick;
383 	unsigned int			checked;
384 
385 	bs->bs_sick = 0;
386 	bs->bs_checked = 0;
387 
388 	xfs_inode_measure_sickness(ip, &sick, &checked);
389 	for (m = ino_map; m->sick_mask; m++) {
390 		if (checked & m->sick_mask)
391 			bs->bs_checked |= m->ioctl_mask;
392 		if (sick & m->sick_mask)
393 			bs->bs_sick |= m->ioctl_mask;
394 	}
395 }
396