Loading...
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#ifndef __XFS_HEALTH_H__
7#define __XFS_HEALTH_H__
8
9/*
10 * In-Core Filesystem Health Assessments
11 * =====================================
12 *
13 * We'd like to be able to summarize the current health status of the
14 * filesystem so that the administrator knows when it's necessary to schedule
15 * some downtime for repairs. Until then, we would also like to avoid abrupt
16 * shutdowns due to corrupt metadata.
17 *
18 * The online scrub feature evaluates the health of all filesystem metadata.
19 * When scrub detects corruption in a piece of metadata it will set the
20 * corresponding sickness flag, and repair will clear it if successful. If
21 * problems remain at unmount time, we can also request manual intervention by
22 * logging a notice to run xfs_repair.
23 *
24 * Each health tracking group uses a pair of fields for reporting. The
25 * "checked" field tell us if a given piece of metadata has ever been examined,
26 * and the "sick" field tells us if that piece was found to need repairs.
27 * Therefore we can conclude that for a given sick flag value:
28 *
29 * - checked && sick => metadata needs repair
30 * - checked && !sick => metadata is ok
31 * - !checked => has not been examined since mount
32 */
33
34struct xfs_mount;
35struct xfs_perag;
36struct xfs_inode;
37struct xfs_fsop_geom;
38
39/* Observable health issues for metadata spanning the entire filesystem. */
40#define XFS_SICK_FS_COUNTERS (1 << 0) /* summary counters */
41#define XFS_SICK_FS_UQUOTA (1 << 1) /* user quota */
42#define XFS_SICK_FS_GQUOTA (1 << 2) /* group quota */
43#define XFS_SICK_FS_PQUOTA (1 << 3) /* project quota */
44
45/* Observable health issues for realtime volume metadata. */
46#define XFS_SICK_RT_BITMAP (1 << 0) /* realtime bitmap */
47#define XFS_SICK_RT_SUMMARY (1 << 1) /* realtime summary */
48
49/* Observable health issues for AG metadata. */
50#define XFS_SICK_AG_SB (1 << 0) /* superblock */
51#define XFS_SICK_AG_AGF (1 << 1) /* AGF header */
52#define XFS_SICK_AG_AGFL (1 << 2) /* AGFL header */
53#define XFS_SICK_AG_AGI (1 << 3) /* AGI header */
54#define XFS_SICK_AG_BNOBT (1 << 4) /* free space by block */
55#define XFS_SICK_AG_CNTBT (1 << 5) /* free space by length */
56#define XFS_SICK_AG_INOBT (1 << 6) /* inode index */
57#define XFS_SICK_AG_FINOBT (1 << 7) /* free inode index */
58#define XFS_SICK_AG_RMAPBT (1 << 8) /* reverse mappings */
59#define XFS_SICK_AG_REFCNTBT (1 << 9) /* reference counts */
60
61/* Observable health issues for inode metadata. */
62#define XFS_SICK_INO_CORE (1 << 0) /* inode core */
63#define XFS_SICK_INO_BMBTD (1 << 1) /* data fork */
64#define XFS_SICK_INO_BMBTA (1 << 2) /* attr fork */
65#define XFS_SICK_INO_BMBTC (1 << 3) /* cow fork */
66#define XFS_SICK_INO_DIR (1 << 4) /* directory */
67#define XFS_SICK_INO_XATTR (1 << 5) /* extended attributes */
68#define XFS_SICK_INO_SYMLINK (1 << 6) /* symbolic link remote target */
69#define XFS_SICK_INO_PARENT (1 << 7) /* parent pointers */
70
71/* Primary evidence of health problems in a given group. */
72#define XFS_SICK_FS_PRIMARY (XFS_SICK_FS_COUNTERS | \
73 XFS_SICK_FS_UQUOTA | \
74 XFS_SICK_FS_GQUOTA | \
75 XFS_SICK_FS_PQUOTA)
76
77#define XFS_SICK_RT_PRIMARY (XFS_SICK_RT_BITMAP | \
78 XFS_SICK_RT_SUMMARY)
79
80#define XFS_SICK_AG_PRIMARY (XFS_SICK_AG_SB | \
81 XFS_SICK_AG_AGF | \
82 XFS_SICK_AG_AGFL | \
83 XFS_SICK_AG_AGI | \
84 XFS_SICK_AG_BNOBT | \
85 XFS_SICK_AG_CNTBT | \
86 XFS_SICK_AG_INOBT | \
87 XFS_SICK_AG_FINOBT | \
88 XFS_SICK_AG_RMAPBT | \
89 XFS_SICK_AG_REFCNTBT)
90
91#define XFS_SICK_INO_PRIMARY (XFS_SICK_INO_CORE | \
92 XFS_SICK_INO_BMBTD | \
93 XFS_SICK_INO_BMBTA | \
94 XFS_SICK_INO_BMBTC | \
95 XFS_SICK_INO_DIR | \
96 XFS_SICK_INO_XATTR | \
97 XFS_SICK_INO_SYMLINK | \
98 XFS_SICK_INO_PARENT)
99
100/* These functions must be provided by the xfs implementation. */
101
102void xfs_fs_mark_sick(struct xfs_mount *mp, unsigned int mask);
103void xfs_fs_mark_healthy(struct xfs_mount *mp, unsigned int mask);
104void xfs_fs_measure_sickness(struct xfs_mount *mp, unsigned int *sick,
105 unsigned int *checked);
106
107void xfs_rt_mark_sick(struct xfs_mount *mp, unsigned int mask);
108void xfs_rt_mark_healthy(struct xfs_mount *mp, unsigned int mask);
109void xfs_rt_measure_sickness(struct xfs_mount *mp, unsigned int *sick,
110 unsigned int *checked);
111
112void xfs_ag_mark_sick(struct xfs_perag *pag, unsigned int mask);
113void xfs_ag_mark_healthy(struct xfs_perag *pag, unsigned int mask);
114void xfs_ag_measure_sickness(struct xfs_perag *pag, unsigned int *sick,
115 unsigned int *checked);
116
117void xfs_inode_mark_sick(struct xfs_inode *ip, unsigned int mask);
118void xfs_inode_mark_healthy(struct xfs_inode *ip, unsigned int mask);
119void xfs_inode_measure_sickness(struct xfs_inode *ip, unsigned int *sick,
120 unsigned int *checked);
121
122void xfs_health_unmount(struct xfs_mount *mp);
123
124/* Now some helpers. */
125
126static inline bool
127xfs_fs_has_sickness(struct xfs_mount *mp, unsigned int mask)
128{
129 unsigned int sick, checked;
130
131 xfs_fs_measure_sickness(mp, &sick, &checked);
132 return sick & mask;
133}
134
135static inline bool
136xfs_rt_has_sickness(struct xfs_mount *mp, unsigned int mask)
137{
138 unsigned int sick, checked;
139
140 xfs_rt_measure_sickness(mp, &sick, &checked);
141 return sick & mask;
142}
143
144static inline bool
145xfs_ag_has_sickness(struct xfs_perag *pag, unsigned int mask)
146{
147 unsigned int sick, checked;
148
149 xfs_ag_measure_sickness(pag, &sick, &checked);
150 return sick & mask;
151}
152
153static inline bool
154xfs_inode_has_sickness(struct xfs_inode *ip, unsigned int mask)
155{
156 unsigned int sick, checked;
157
158 xfs_inode_measure_sickness(ip, &sick, &checked);
159 return sick & mask;
160}
161
162static inline bool
163xfs_fs_is_healthy(struct xfs_mount *mp)
164{
165 return !xfs_fs_has_sickness(mp, -1U);
166}
167
168static inline bool
169xfs_rt_is_healthy(struct xfs_mount *mp)
170{
171 return !xfs_rt_has_sickness(mp, -1U);
172}
173
174static inline bool
175xfs_ag_is_healthy(struct xfs_perag *pag)
176{
177 return !xfs_ag_has_sickness(pag, -1U);
178}
179
180static inline bool
181xfs_inode_is_healthy(struct xfs_inode *ip)
182{
183 return !xfs_inode_has_sickness(ip, -1U);
184}
185
186void xfs_fsop_geom_health(struct xfs_mount *mp, struct xfs_fsop_geom *geo);
187void xfs_ag_geom_health(struct xfs_perag *pag, struct xfs_ag_geometry *ageo);
188void xfs_bulkstat_health(struct xfs_inode *ip, struct xfs_bulkstat *bs);
189
190#endif /* __XFS_HEALTH_H__ */
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#ifndef __XFS_HEALTH_H__
7#define __XFS_HEALTH_H__
8
9/*
10 * In-Core Filesystem Health Assessments
11 * =====================================
12 *
13 * We'd like to be able to summarize the current health status of the
14 * filesystem so that the administrator knows when it's necessary to schedule
15 * some downtime for repairs. Until then, we would also like to avoid abrupt
16 * shutdowns due to corrupt metadata.
17 *
18 * The online scrub feature evaluates the health of all filesystem metadata.
19 * When scrub detects corruption in a piece of metadata it will set the
20 * corresponding sickness flag, and repair will clear it if successful. If
21 * problems remain at unmount time, we can also request manual intervention by
22 * logging a notice to run xfs_repair.
23 *
24 * Each health tracking group uses a pair of fields for reporting. The
25 * "checked" field tell us if a given piece of metadata has ever been examined,
26 * and the "sick" field tells us if that piece was found to need repairs.
27 * Therefore we can conclude that for a given sick flag value:
28 *
29 * - checked && sick => metadata needs repair
30 * - checked && !sick => metadata is ok
31 * - !checked && sick => errors have been observed during normal operation,
32 * but the metadata has not been checked thoroughly
33 * - !checked && !sick => has not been examined since mount
34 *
35 * Evidence of health problems can be sorted into three basic categories:
36 *
37 * a) Primary evidence, which signals that something is defective within the
38 * general grouping of metadata.
39 *
40 * b) Secondary evidence, which are side effects of primary problem but are
41 * not themselves problems. These can be forgotten when the primary
42 * health problems are addressed.
43 *
44 * c) Indirect evidence, which points to something being wrong in another
45 * group, but we had to release resources and this is all that's left of
46 * that state.
47 */
48
49struct xfs_mount;
50struct xfs_perag;
51struct xfs_inode;
52struct xfs_fsop_geom;
53struct xfs_btree_cur;
54struct xfs_da_args;
55
56/* Observable health issues for metadata spanning the entire filesystem. */
57#define XFS_SICK_FS_COUNTERS (1 << 0) /* summary counters */
58#define XFS_SICK_FS_UQUOTA (1 << 1) /* user quota */
59#define XFS_SICK_FS_GQUOTA (1 << 2) /* group quota */
60#define XFS_SICK_FS_PQUOTA (1 << 3) /* project quota */
61#define XFS_SICK_FS_QUOTACHECK (1 << 4) /* quota counts */
62#define XFS_SICK_FS_NLINKS (1 << 5) /* inode link counts */
63
64/* Observable health issues for realtime volume metadata. */
65#define XFS_SICK_RT_BITMAP (1 << 0) /* realtime bitmap */
66#define XFS_SICK_RT_SUMMARY (1 << 1) /* realtime summary */
67
68/* Observable health issues for AG metadata. */
69#define XFS_SICK_AG_SB (1 << 0) /* superblock */
70#define XFS_SICK_AG_AGF (1 << 1) /* AGF header */
71#define XFS_SICK_AG_AGFL (1 << 2) /* AGFL header */
72#define XFS_SICK_AG_AGI (1 << 3) /* AGI header */
73#define XFS_SICK_AG_BNOBT (1 << 4) /* free space by block */
74#define XFS_SICK_AG_CNTBT (1 << 5) /* free space by length */
75#define XFS_SICK_AG_INOBT (1 << 6) /* inode index */
76#define XFS_SICK_AG_FINOBT (1 << 7) /* free inode index */
77#define XFS_SICK_AG_RMAPBT (1 << 8) /* reverse mappings */
78#define XFS_SICK_AG_REFCNTBT (1 << 9) /* reference counts */
79#define XFS_SICK_AG_INODES (1 << 10) /* inactivated bad inodes */
80
81/* Observable health issues for inode metadata. */
82#define XFS_SICK_INO_CORE (1 << 0) /* inode core */
83#define XFS_SICK_INO_BMBTD (1 << 1) /* data fork */
84#define XFS_SICK_INO_BMBTA (1 << 2) /* attr fork */
85#define XFS_SICK_INO_BMBTC (1 << 3) /* cow fork */
86#define XFS_SICK_INO_DIR (1 << 4) /* directory */
87#define XFS_SICK_INO_XATTR (1 << 5) /* extended attributes */
88#define XFS_SICK_INO_SYMLINK (1 << 6) /* symbolic link remote target */
89#define XFS_SICK_INO_PARENT (1 << 7) /* parent pointers */
90
91#define XFS_SICK_INO_BMBTD_ZAPPED (1 << 8) /* data fork erased */
92#define XFS_SICK_INO_BMBTA_ZAPPED (1 << 9) /* attr fork erased */
93#define XFS_SICK_INO_DIR_ZAPPED (1 << 10) /* directory erased */
94#define XFS_SICK_INO_SYMLINK_ZAPPED (1 << 11) /* symlink erased */
95
96/* Don't propagate sick status to ag health summary during inactivation */
97#define XFS_SICK_INO_FORGET (1 << 12)
98
99/* Primary evidence of health problems in a given group. */
100#define XFS_SICK_FS_PRIMARY (XFS_SICK_FS_COUNTERS | \
101 XFS_SICK_FS_UQUOTA | \
102 XFS_SICK_FS_GQUOTA | \
103 XFS_SICK_FS_PQUOTA | \
104 XFS_SICK_FS_QUOTACHECK | \
105 XFS_SICK_FS_NLINKS)
106
107#define XFS_SICK_RT_PRIMARY (XFS_SICK_RT_BITMAP | \
108 XFS_SICK_RT_SUMMARY)
109
110#define XFS_SICK_AG_PRIMARY (XFS_SICK_AG_SB | \
111 XFS_SICK_AG_AGF | \
112 XFS_SICK_AG_AGFL | \
113 XFS_SICK_AG_AGI | \
114 XFS_SICK_AG_BNOBT | \
115 XFS_SICK_AG_CNTBT | \
116 XFS_SICK_AG_INOBT | \
117 XFS_SICK_AG_FINOBT | \
118 XFS_SICK_AG_RMAPBT | \
119 XFS_SICK_AG_REFCNTBT)
120
121#define XFS_SICK_INO_PRIMARY (XFS_SICK_INO_CORE | \
122 XFS_SICK_INO_BMBTD | \
123 XFS_SICK_INO_BMBTA | \
124 XFS_SICK_INO_BMBTC | \
125 XFS_SICK_INO_DIR | \
126 XFS_SICK_INO_XATTR | \
127 XFS_SICK_INO_SYMLINK | \
128 XFS_SICK_INO_PARENT)
129
130#define XFS_SICK_INO_ZAPPED (XFS_SICK_INO_BMBTD_ZAPPED | \
131 XFS_SICK_INO_BMBTA_ZAPPED | \
132 XFS_SICK_INO_DIR_ZAPPED | \
133 XFS_SICK_INO_SYMLINK_ZAPPED)
134
135/* Secondary state related to (but not primary evidence of) health problems. */
136#define XFS_SICK_FS_SECONDARY (0)
137#define XFS_SICK_RT_SECONDARY (0)
138#define XFS_SICK_AG_SECONDARY (0)
139#define XFS_SICK_INO_SECONDARY (XFS_SICK_INO_FORGET)
140
141/* Evidence of health problems elsewhere. */
142#define XFS_SICK_FS_INDIRECT (0)
143#define XFS_SICK_RT_INDIRECT (0)
144#define XFS_SICK_AG_INDIRECT (XFS_SICK_AG_INODES)
145#define XFS_SICK_INO_INDIRECT (0)
146
147/* All health masks. */
148#define XFS_SICK_FS_ALL (XFS_SICK_FS_PRIMARY | \
149 XFS_SICK_FS_SECONDARY | \
150 XFS_SICK_FS_INDIRECT)
151
152#define XFS_SICK_RT_ALL (XFS_SICK_RT_PRIMARY | \
153 XFS_SICK_RT_SECONDARY | \
154 XFS_SICK_RT_INDIRECT)
155
156#define XFS_SICK_AG_ALL (XFS_SICK_AG_PRIMARY | \
157 XFS_SICK_AG_SECONDARY | \
158 XFS_SICK_AG_INDIRECT)
159
160#define XFS_SICK_INO_ALL (XFS_SICK_INO_PRIMARY | \
161 XFS_SICK_INO_SECONDARY | \
162 XFS_SICK_INO_INDIRECT | \
163 XFS_SICK_INO_ZAPPED)
164
165/*
166 * These functions must be provided by the xfs implementation. Function
167 * behavior with respect to the first argument should be as follows:
168 *
169 * xfs_*_mark_sick: Set the sick flags and do not set checked flags.
170 * Runtime code should call this upon encountering
171 * a corruption.
172 *
173 * xfs_*_mark_corrupt: Set the sick and checked flags simultaneously.
174 * Fsck tools should call this when corruption is
175 * found.
176 *
177 * xfs_*_mark_healthy: Clear the sick flags and set the checked flags.
178 * Fsck tools should call this after correcting errors.
179 *
180 * xfs_*_measure_sickness: Return the sick and check status in the provided
181 * out parameters.
182 */
183
184void xfs_fs_mark_sick(struct xfs_mount *mp, unsigned int mask);
185void xfs_fs_mark_corrupt(struct xfs_mount *mp, unsigned int mask);
186void xfs_fs_mark_healthy(struct xfs_mount *mp, unsigned int mask);
187void xfs_fs_measure_sickness(struct xfs_mount *mp, unsigned int *sick,
188 unsigned int *checked);
189
190void xfs_rt_mark_sick(struct xfs_mount *mp, unsigned int mask);
191void xfs_rt_mark_corrupt(struct xfs_mount *mp, unsigned int mask);
192void xfs_rt_mark_healthy(struct xfs_mount *mp, unsigned int mask);
193void xfs_rt_measure_sickness(struct xfs_mount *mp, unsigned int *sick,
194 unsigned int *checked);
195
196void xfs_agno_mark_sick(struct xfs_mount *mp, xfs_agnumber_t agno,
197 unsigned int mask);
198void xfs_ag_mark_sick(struct xfs_perag *pag, unsigned int mask);
199void xfs_ag_mark_corrupt(struct xfs_perag *pag, unsigned int mask);
200void xfs_ag_mark_healthy(struct xfs_perag *pag, unsigned int mask);
201void xfs_ag_measure_sickness(struct xfs_perag *pag, unsigned int *sick,
202 unsigned int *checked);
203
204void xfs_inode_mark_sick(struct xfs_inode *ip, unsigned int mask);
205void xfs_inode_mark_corrupt(struct xfs_inode *ip, unsigned int mask);
206void xfs_inode_mark_healthy(struct xfs_inode *ip, unsigned int mask);
207void xfs_inode_measure_sickness(struct xfs_inode *ip, unsigned int *sick,
208 unsigned int *checked);
209
210void xfs_health_unmount(struct xfs_mount *mp);
211void xfs_bmap_mark_sick(struct xfs_inode *ip, int whichfork);
212void xfs_btree_mark_sick(struct xfs_btree_cur *cur);
213void xfs_dirattr_mark_sick(struct xfs_inode *ip, int whichfork);
214void xfs_da_mark_sick(struct xfs_da_args *args);
215
216/* Now some helpers. */
217
218static inline bool
219xfs_fs_has_sickness(struct xfs_mount *mp, unsigned int mask)
220{
221 unsigned int sick, checked;
222
223 xfs_fs_measure_sickness(mp, &sick, &checked);
224 return sick & mask;
225}
226
227static inline bool
228xfs_rt_has_sickness(struct xfs_mount *mp, unsigned int mask)
229{
230 unsigned int sick, checked;
231
232 xfs_rt_measure_sickness(mp, &sick, &checked);
233 return sick & mask;
234}
235
236static inline bool
237xfs_ag_has_sickness(struct xfs_perag *pag, unsigned int mask)
238{
239 unsigned int sick, checked;
240
241 xfs_ag_measure_sickness(pag, &sick, &checked);
242 return sick & mask;
243}
244
245static inline bool
246xfs_inode_has_sickness(struct xfs_inode *ip, unsigned int mask)
247{
248 unsigned int sick, checked;
249
250 xfs_inode_measure_sickness(ip, &sick, &checked);
251 return sick & mask;
252}
253
254static inline bool
255xfs_fs_is_healthy(struct xfs_mount *mp)
256{
257 return !xfs_fs_has_sickness(mp, -1U);
258}
259
260static inline bool
261xfs_rt_is_healthy(struct xfs_mount *mp)
262{
263 return !xfs_rt_has_sickness(mp, -1U);
264}
265
266static inline bool
267xfs_ag_is_healthy(struct xfs_perag *pag)
268{
269 return !xfs_ag_has_sickness(pag, -1U);
270}
271
272static inline bool
273xfs_inode_is_healthy(struct xfs_inode *ip)
274{
275 return !xfs_inode_has_sickness(ip, -1U);
276}
277
278void xfs_fsop_geom_health(struct xfs_mount *mp, struct xfs_fsop_geom *geo);
279void xfs_ag_geom_health(struct xfs_perag *pag, struct xfs_ag_geometry *ageo);
280void xfs_bulkstat_health(struct xfs_inode *ip, struct xfs_bulkstat *bs);
281
282#define xfs_metadata_is_sick(error) \
283 (unlikely((error) == -EFSCORRUPTED || (error) == -EFSBADCRC))
284
285#endif /* __XFS_HEALTH_H__ */