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
  9struct xfs_group;
 10
 11/*
 12 * In-Core Filesystem Health Assessments
 13 * =====================================
 14 *
 15 * We'd like to be able to summarize the current health status of the
 16 * filesystem so that the administrator knows when it's necessary to schedule
 17 * some downtime for repairs.  Until then, we would also like to avoid abrupt
 18 * shutdowns due to corrupt metadata.
 19 *
 20 * The online scrub feature evaluates the health of all filesystem metadata.
 21 * When scrub detects corruption in a piece of metadata it will set the
 22 * corresponding sickness flag, and repair will clear it if successful.  If
 23 * problems remain at unmount time, we can also request manual intervention by
 24 * logging a notice to run xfs_repair.
 25 *
 26 * Each health tracking group uses a pair of fields for reporting.  The
 27 * "checked" field tell us if a given piece of metadata has ever been examined,
 28 * and the "sick" field tells us if that piece was found to need repairs.
 29 * Therefore we can conclude that for a given sick flag value:
 30 *
 31 *  - checked && sick   => metadata needs repair
 32 *  - checked && !sick  => metadata is ok
 33 *  - !checked && sick  => errors have been observed during normal operation,
 34 *                         but the metadata has not been checked thoroughly
 35 *  - !checked && !sick => has not been examined since mount
 36 *
 37 * Evidence of health problems can be sorted into three basic categories:
 38 *
 39 * a) Primary evidence, which signals that something is defective within the
 40 *    general grouping of metadata.
 41 *
 42 * b) Secondary evidence, which are side effects of primary problem but are
 43 *    not themselves problems.  These can be forgotten when the primary
 44 *    health problems are addressed.
 45 *
 46 * c) Indirect evidence, which points to something being wrong in another
 47 *    group, but we had to release resources and this is all that's left of
 48 *    that state.
 49 */
 50
 51struct xfs_mount;
 52struct xfs_perag;
 53struct xfs_inode;
 54struct xfs_fsop_geom;
 55struct xfs_btree_cur;
 56struct xfs_da_args;
 57struct xfs_rtgroup;
 58
 59/* Observable health issues for metadata spanning the entire filesystem. */
 60#define XFS_SICK_FS_COUNTERS	(1 << 0)  /* summary counters */
 61#define XFS_SICK_FS_UQUOTA	(1 << 1)  /* user quota */
 62#define XFS_SICK_FS_GQUOTA	(1 << 2)  /* group quota */
 63#define XFS_SICK_FS_PQUOTA	(1 << 3)  /* project quota */
 64#define XFS_SICK_FS_QUOTACHECK	(1 << 4)  /* quota counts */
 65#define XFS_SICK_FS_NLINKS	(1 << 5)  /* inode link counts */
 66#define XFS_SICK_FS_METADIR	(1 << 6)  /* metadata directory tree */
 67#define XFS_SICK_FS_METAPATH	(1 << 7)  /* metadata directory tree path */
 68
 69/* Observable health issues for realtime group metadata. */
 70#define XFS_SICK_RG_SUPER	(1 << 0)  /* rt group superblock */
 71#define XFS_SICK_RG_BITMAP	(1 << 1)  /* rt group bitmap */
 72#define XFS_SICK_RG_SUMMARY	(1 << 2)  /* rt groups summary */
 73#define XFS_SICK_RG_RMAPBT	(1 << 3)  /* reverse mappings */
 74#define XFS_SICK_RG_REFCNTBT	(1 << 4)  /* reference counts */
 75
 76/* Observable health issues for AG metadata. */
 77#define XFS_SICK_AG_SB		(1 << 0)  /* superblock */
 78#define XFS_SICK_AG_AGF		(1 << 1)  /* AGF header */
 79#define XFS_SICK_AG_AGFL	(1 << 2)  /* AGFL header */
 80#define XFS_SICK_AG_AGI		(1 << 3)  /* AGI header */
 81#define XFS_SICK_AG_BNOBT	(1 << 4)  /* free space by block */
 82#define XFS_SICK_AG_CNTBT	(1 << 5)  /* free space by length */
 83#define XFS_SICK_AG_INOBT	(1 << 6)  /* inode index */
 84#define XFS_SICK_AG_FINOBT	(1 << 7)  /* free inode index */
 85#define XFS_SICK_AG_RMAPBT	(1 << 8)  /* reverse mappings */
 86#define XFS_SICK_AG_REFCNTBT	(1 << 9)  /* reference counts */
 87#define XFS_SICK_AG_INODES	(1 << 10) /* inactivated bad inodes */
 88
 89/* Observable health issues for inode metadata. */
 90#define XFS_SICK_INO_CORE	(1 << 0)  /* inode core */
 91#define XFS_SICK_INO_BMBTD	(1 << 1)  /* data fork */
 92#define XFS_SICK_INO_BMBTA	(1 << 2)  /* attr fork */
 93#define XFS_SICK_INO_BMBTC	(1 << 3)  /* cow fork */
 94#define XFS_SICK_INO_DIR	(1 << 4)  /* directory */
 95#define XFS_SICK_INO_XATTR	(1 << 5)  /* extended attributes */
 96#define XFS_SICK_INO_SYMLINK	(1 << 6)  /* symbolic link remote target */
 97#define XFS_SICK_INO_PARENT	(1 << 7)  /* parent pointers */
 98
 99#define XFS_SICK_INO_BMBTD_ZAPPED	(1 << 8)  /* data fork erased */
100#define XFS_SICK_INO_BMBTA_ZAPPED	(1 << 9)  /* attr fork erased */
101#define XFS_SICK_INO_DIR_ZAPPED		(1 << 10) /* directory erased */
102#define XFS_SICK_INO_SYMLINK_ZAPPED	(1 << 11) /* symlink erased */
103
104/* Don't propagate sick status to ag health summary during inactivation */
105#define XFS_SICK_INO_FORGET	(1 << 12)
106#define XFS_SICK_INO_DIRTREE	(1 << 13)  /* directory tree structure */
107
108/* Primary evidence of health problems in a given group. */
109#define XFS_SICK_FS_PRIMARY	(XFS_SICK_FS_COUNTERS | \
110				 XFS_SICK_FS_UQUOTA | \
111				 XFS_SICK_FS_GQUOTA | \
112				 XFS_SICK_FS_PQUOTA | \
113				 XFS_SICK_FS_QUOTACHECK | \
114				 XFS_SICK_FS_NLINKS | \
115				 XFS_SICK_FS_METADIR | \
116				 XFS_SICK_FS_METAPATH)
117
118#define XFS_SICK_RG_PRIMARY	(XFS_SICK_RG_SUPER | \
119				 XFS_SICK_RG_BITMAP | \
120				 XFS_SICK_RG_SUMMARY | \
121				 XFS_SICK_RG_RMAPBT | \
122				 XFS_SICK_RG_REFCNTBT)
123
124#define XFS_SICK_AG_PRIMARY	(XFS_SICK_AG_SB | \
125				 XFS_SICK_AG_AGF | \
126				 XFS_SICK_AG_AGFL | \
127				 XFS_SICK_AG_AGI | \
128				 XFS_SICK_AG_BNOBT | \
129				 XFS_SICK_AG_CNTBT | \
130				 XFS_SICK_AG_INOBT | \
131				 XFS_SICK_AG_FINOBT | \
132				 XFS_SICK_AG_RMAPBT | \
133				 XFS_SICK_AG_REFCNTBT)
134
135#define XFS_SICK_INO_PRIMARY	(XFS_SICK_INO_CORE | \
136				 XFS_SICK_INO_BMBTD | \
137				 XFS_SICK_INO_BMBTA | \
138				 XFS_SICK_INO_BMBTC | \
139				 XFS_SICK_INO_DIR | \
140				 XFS_SICK_INO_XATTR | \
141				 XFS_SICK_INO_SYMLINK | \
142				 XFS_SICK_INO_PARENT | \
143				 XFS_SICK_INO_DIRTREE)
144
145#define XFS_SICK_INO_ZAPPED	(XFS_SICK_INO_BMBTD_ZAPPED | \
146				 XFS_SICK_INO_BMBTA_ZAPPED | \
147				 XFS_SICK_INO_DIR_ZAPPED | \
148				 XFS_SICK_INO_SYMLINK_ZAPPED)
149
150/* Secondary state related to (but not primary evidence of) health problems. */
151#define XFS_SICK_FS_SECONDARY	(0)
152#define XFS_SICK_RG_SECONDARY	(0)
153#define XFS_SICK_AG_SECONDARY	(0)
154#define XFS_SICK_INO_SECONDARY	(XFS_SICK_INO_FORGET)
155
156/* Evidence of health problems elsewhere. */
157#define XFS_SICK_FS_INDIRECT	(0)
158#define XFS_SICK_RG_INDIRECT	(0)
159#define XFS_SICK_AG_INDIRECT	(XFS_SICK_AG_INODES)
160#define XFS_SICK_INO_INDIRECT	(0)
161
162/* All health masks. */
163#define XFS_SICK_FS_ALL		(XFS_SICK_FS_PRIMARY | \
164				 XFS_SICK_FS_SECONDARY | \
165				 XFS_SICK_FS_INDIRECT)
166
167#define XFS_SICK_RG_ALL		(XFS_SICK_RG_PRIMARY | \
168				 XFS_SICK_RG_SECONDARY | \
169				 XFS_SICK_RG_INDIRECT)
170
171#define XFS_SICK_AG_ALL		(XFS_SICK_AG_PRIMARY | \
172				 XFS_SICK_AG_SECONDARY | \
173				 XFS_SICK_AG_INDIRECT)
174
175#define XFS_SICK_INO_ALL	(XFS_SICK_INO_PRIMARY | \
176				 XFS_SICK_INO_SECONDARY | \
177				 XFS_SICK_INO_INDIRECT | \
178				 XFS_SICK_INO_ZAPPED)
179
180/*
181 * These functions must be provided by the xfs implementation.  Function
182 * behavior with respect to the first argument should be as follows:
183 *
184 * xfs_*_mark_sick:        Set the sick flags and do not set checked flags.
185 *                         Runtime code should call this upon encountering
186 *                         a corruption.
187 *
188 * xfs_*_mark_corrupt:     Set the sick and checked flags simultaneously.
189 *                         Fsck tools should call this when corruption is
190 *                         found.
191 *
192 * xfs_*_mark_healthy:     Clear the sick flags and set the checked flags.
193 *                         Fsck tools should call this after correcting errors.
194 *
195 * xfs_*_measure_sickness: Return the sick and check status in the provided
196 *                         out parameters.
197 */
198
199void xfs_fs_mark_sick(struct xfs_mount *mp, unsigned int mask);
200void xfs_fs_mark_corrupt(struct xfs_mount *mp, unsigned int mask);
201void xfs_fs_mark_healthy(struct xfs_mount *mp, unsigned int mask);
202void xfs_fs_measure_sickness(struct xfs_mount *mp, unsigned int *sick,
203		unsigned int *checked);
204
205void xfs_rgno_mark_sick(struct xfs_mount *mp, xfs_rgnumber_t rgno,
206		unsigned int mask);
207
208void xfs_agno_mark_sick(struct xfs_mount *mp, xfs_agnumber_t agno,
209		unsigned int mask);
210void xfs_group_mark_sick(struct xfs_group *xg, unsigned int mask);
211#define xfs_ag_mark_sick(pag, mask) \
212	xfs_group_mark_sick(pag_group(pag), (mask))
213void xfs_group_mark_corrupt(struct xfs_group *xg, unsigned int mask);
214void xfs_group_mark_healthy(struct xfs_group *xg, unsigned int mask);
215void xfs_group_measure_sickness(struct xfs_group *xg, unsigned int *sick,
216		unsigned int *checked);
217
218void xfs_inode_mark_sick(struct xfs_inode *ip, unsigned int mask);
219void xfs_inode_mark_corrupt(struct xfs_inode *ip, unsigned int mask);
220void xfs_inode_mark_healthy(struct xfs_inode *ip, unsigned int mask);
221void xfs_inode_measure_sickness(struct xfs_inode *ip, unsigned int *sick,
222		unsigned int *checked);
223
224void xfs_health_unmount(struct xfs_mount *mp);
225void xfs_bmap_mark_sick(struct xfs_inode *ip, int whichfork);
226void xfs_btree_mark_sick(struct xfs_btree_cur *cur);
227void xfs_dirattr_mark_sick(struct xfs_inode *ip, int whichfork);
228void xfs_da_mark_sick(struct xfs_da_args *args);
229
230/* Now some helpers. */
231
232static inline bool
233xfs_fs_has_sickness(struct xfs_mount *mp, unsigned int mask)
234{
235	unsigned int	sick, checked;
236
237	xfs_fs_measure_sickness(mp, &sick, &checked);
238	return sick & mask;
239}
240
241static inline bool
242xfs_group_has_sickness(
243	struct xfs_group	*xg,
244	unsigned int		mask)
245{
246	unsigned int		sick, checked;
247
248	xfs_group_measure_sickness(xg, &sick, &checked);
249	return sick & mask;
250}
251
252#define xfs_ag_has_sickness(pag, mask) \
253	xfs_group_has_sickness(pag_group(pag), (mask))
254#define xfs_ag_is_healthy(pag) \
255	(!xfs_ag_has_sickness((pag), UINT_MAX))
256
257#define xfs_rtgroup_has_sickness(rtg, mask) \
258	xfs_group_has_sickness(rtg_group(rtg), (mask))
259#define xfs_rtgroup_is_healthy(rtg) \
260	(!xfs_rtgroup_has_sickness((rtg), UINT_MAX))
261
262static inline bool
263xfs_inode_has_sickness(struct xfs_inode *ip, unsigned int mask)
264{
265	unsigned int	sick, checked;
266
267	xfs_inode_measure_sickness(ip, &sick, &checked);
268	return sick & mask;
269}
270
271static inline bool
272xfs_fs_is_healthy(struct xfs_mount *mp)
273{
274	return !xfs_fs_has_sickness(mp, -1U);
275}
276
277static inline bool
278xfs_inode_is_healthy(struct xfs_inode *ip)
279{
280	return !xfs_inode_has_sickness(ip, -1U);
281}
282
283void xfs_fsop_geom_health(struct xfs_mount *mp, struct xfs_fsop_geom *geo);
284void xfs_ag_geom_health(struct xfs_perag *pag, struct xfs_ag_geometry *ageo);
285void xfs_rtgroup_geom_health(struct xfs_rtgroup *rtg,
286		struct xfs_rtgroup_geometry *rgeo);
287void xfs_bulkstat_health(struct xfs_inode *ip, struct xfs_bulkstat *bs);
288
289#define xfs_metadata_is_sick(error) \
290	(unlikely((error) == -EFSCORRUPTED || (error) == -EFSBADCRC))
291
292#endif	/* __XFS_HEALTH_H__ */