xref: /linux/fs/xfs/scrub/health.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2019-2023 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_btree.h"
13 #include "xfs_ag.h"
14 #include "xfs_health.h"
15 #include "xfs_rtgroup.h"
16 #include "scrub/scrub.h"
17 #include "scrub/health.h"
18 #include "scrub/common.h"
19 
20 /*
21  * Scrub and In-Core Filesystem Health Assessments
22  * ===============================================
23  *
24  * Online scrub and repair have the time and the ability to perform stronger
25  * checks than we can do from the metadata verifiers, because they can
26  * cross-reference records between data structures.  Therefore, scrub is in a
27  * good position to update the online filesystem health assessments to reflect
28  * the good/bad state of the data structure.
29  *
30  * We therefore extend scrub in the following ways to achieve this:
31  *
32  * 1. Create a "sick_mask" field in the scrub context.  When we're setting up a
33  * scrub call, set this to the default XFS_SICK_* flag(s) for the selected
34  * scrub type (call it A).  Scrub and repair functions can override the default
35  * sick_mask value if they choose.
36  *
37  * 2. If the scrubber returns a runtime error code, we exit making no changes
38  * to the incore sick state.
39  *
40  * 3. If the scrubber finds that A is clean, use sick_mask to clear the incore
41  * sick flags before exiting.
42  *
43  * 4. If the scrubber finds that A is corrupt, use sick_mask to set the incore
44  * sick flags.  If the user didn't want to repair then we exit, leaving the
45  * metadata structure unfixed and the sick flag set.
46  *
47  * 5. Now we know that A is corrupt and the user wants to repair, so run the
48  * repairer.  If the repairer returns an error code, we exit with that error
49  * code, having made no further changes to the incore sick state.
50  *
51  * 6. If repair rebuilds A correctly and the subsequent re-scrub of A is clean,
52  * use sick_mask to clear the incore sick flags.  This should have the effect
53  * that A is no longer marked sick.
54  *
55  * 7. If repair rebuilds A incorrectly, the re-scrub will find it corrupt and
56  * use sick_mask to set the incore sick flags.  This should have no externally
57  * visible effect since we already set them in step (4).
58  *
59  * There are some complications to this story, however.  For certain types of
60  * complementary metadata indices (e.g. inobt/finobt), it is easier to rebuild
61  * both structures at the same time.  The following principles apply to this
62  * type of repair strategy:
63  *
64  * 8. Any repair function that rebuilds multiple structures should update
65  * sick_mask_visible to reflect whatever other structures are rebuilt, and
66  * verify that all the rebuilt structures can pass a scrub check.  The outcomes
67  * of 5-7 still apply, but with a sick_mask that covers everything being
68  * rebuilt.
69  */
70 
71 /* Map our scrub type to a sick mask and a set of health update functions. */
72 
73 enum xchk_health_group {
74 	XHG_FS = 1,
75 	XHG_AG,
76 	XHG_INO,
77 	XHG_RTGROUP,
78 };
79 
80 struct xchk_health_map {
81 	enum xchk_health_group	group;
82 	unsigned int		sick_mask;
83 };
84 
85 static const struct xchk_health_map type_to_health_flag[XFS_SCRUB_TYPE_NR] = {
86 	[XFS_SCRUB_TYPE_SB]		= { XHG_AG,  XFS_SICK_AG_SB },
87 	[XFS_SCRUB_TYPE_AGF]		= { XHG_AG,  XFS_SICK_AG_AGF },
88 	[XFS_SCRUB_TYPE_AGFL]		= { XHG_AG,  XFS_SICK_AG_AGFL },
89 	[XFS_SCRUB_TYPE_AGI]		= { XHG_AG,  XFS_SICK_AG_AGI },
90 	[XFS_SCRUB_TYPE_BNOBT]		= { XHG_AG,  XFS_SICK_AG_BNOBT },
91 	[XFS_SCRUB_TYPE_CNTBT]		= { XHG_AG,  XFS_SICK_AG_CNTBT },
92 	[XFS_SCRUB_TYPE_INOBT]		= { XHG_AG,  XFS_SICK_AG_INOBT },
93 	[XFS_SCRUB_TYPE_FINOBT]		= { XHG_AG,  XFS_SICK_AG_FINOBT },
94 	[XFS_SCRUB_TYPE_RMAPBT]		= { XHG_AG,  XFS_SICK_AG_RMAPBT },
95 	[XFS_SCRUB_TYPE_REFCNTBT]	= { XHG_AG,  XFS_SICK_AG_REFCNTBT },
96 	[XFS_SCRUB_TYPE_INODE]		= { XHG_INO, XFS_SICK_INO_CORE },
97 	[XFS_SCRUB_TYPE_BMBTD]		= { XHG_INO, XFS_SICK_INO_BMBTD },
98 	[XFS_SCRUB_TYPE_BMBTA]		= { XHG_INO, XFS_SICK_INO_BMBTA },
99 	[XFS_SCRUB_TYPE_BMBTC]		= { XHG_INO, XFS_SICK_INO_BMBTC },
100 	[XFS_SCRUB_TYPE_DIR]		= { XHG_INO, XFS_SICK_INO_DIR },
101 	[XFS_SCRUB_TYPE_XATTR]		= { XHG_INO, XFS_SICK_INO_XATTR },
102 	[XFS_SCRUB_TYPE_SYMLINK]	= { XHG_INO, XFS_SICK_INO_SYMLINK },
103 	[XFS_SCRUB_TYPE_PARENT]		= { XHG_INO, XFS_SICK_INO_PARENT },
104 	[XFS_SCRUB_TYPE_RTBITMAP]	= { XHG_RTGROUP, XFS_SICK_RG_BITMAP },
105 	[XFS_SCRUB_TYPE_RTSUM]		= { XHG_RTGROUP, XFS_SICK_RG_SUMMARY },
106 	[XFS_SCRUB_TYPE_UQUOTA]		= { XHG_FS,  XFS_SICK_FS_UQUOTA },
107 	[XFS_SCRUB_TYPE_GQUOTA]		= { XHG_FS,  XFS_SICK_FS_GQUOTA },
108 	[XFS_SCRUB_TYPE_PQUOTA]		= { XHG_FS,  XFS_SICK_FS_PQUOTA },
109 	[XFS_SCRUB_TYPE_FSCOUNTERS]	= { XHG_FS,  XFS_SICK_FS_COUNTERS },
110 	[XFS_SCRUB_TYPE_QUOTACHECK]	= { XHG_FS,  XFS_SICK_FS_QUOTACHECK },
111 	[XFS_SCRUB_TYPE_NLINKS]		= { XHG_FS,  XFS_SICK_FS_NLINKS },
112 	[XFS_SCRUB_TYPE_DIRTREE]	= { XHG_INO, XFS_SICK_INO_DIRTREE },
113 	[XFS_SCRUB_TYPE_METAPATH]	= { XHG_FS,  XFS_SICK_FS_METAPATH },
114 	[XFS_SCRUB_TYPE_RGSUPER]	= { XHG_RTGROUP, XFS_SICK_RG_SUPER },
115 };
116 
117 /* Return the health status mask for this scrub type. */
118 unsigned int
119 xchk_health_mask_for_scrub_type(
120 	__u32			scrub_type)
121 {
122 	return type_to_health_flag[scrub_type].sick_mask;
123 }
124 
125 /*
126  * If the scrub state is clean, add @mask to the scrub sick mask to clear
127  * additional sick flags from the metadata object's sick state.
128  */
129 void
130 xchk_mark_healthy_if_clean(
131 	struct xfs_scrub	*sc,
132 	unsigned int		mask)
133 {
134 	if (!(sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
135 				  XFS_SCRUB_OFLAG_XCORRUPT)))
136 		sc->sick_mask |= mask;
137 }
138 
139 /*
140  * If we're scrubbing a piece of file metadata for the first time, does it look
141  * like it has been zapped?  Skip the check if we just repaired the metadata
142  * and are revalidating it.
143  */
144 bool
145 xchk_file_looks_zapped(
146 	struct xfs_scrub	*sc,
147 	unsigned int		mask)
148 {
149 	ASSERT((mask & ~XFS_SICK_INO_ZAPPED) == 0);
150 
151 	if (sc->flags & XREP_ALREADY_FIXED)
152 		return false;
153 
154 	return xfs_inode_has_sickness(sc->ip, mask);
155 }
156 
157 /*
158  * Scrub gave the filesystem a clean bill of health, so clear all the indirect
159  * markers of past problems (at least for the fs and ags) so that we can be
160  * healthy again.
161  */
162 STATIC void
163 xchk_mark_all_healthy(
164 	struct xfs_mount	*mp)
165 {
166 	struct xfs_perag	*pag = NULL;
167 	struct xfs_rtgroup	*rtg = NULL;
168 
169 	xfs_fs_mark_healthy(mp, XFS_SICK_FS_INDIRECT);
170 	while ((pag = xfs_perag_next(mp, pag)))
171 		xfs_group_mark_healthy(pag_group(pag), XFS_SICK_AG_INDIRECT);
172 	while ((rtg = xfs_rtgroup_next(mp, rtg)))
173 		xfs_group_mark_healthy(rtg_group(rtg), XFS_SICK_RG_INDIRECT);
174 }
175 
176 /*
177  * Update filesystem health assessments based on what we found and did.
178  *
179  * If the scrubber finds errors, we mark sick whatever's mentioned in
180  * sick_mask, no matter whether this is a first scan or an
181  * evaluation of repair effectiveness.
182  *
183  * Otherwise, no direct corruption was found, so mark whatever's in
184  * sick_mask as healthy.
185  */
186 void
187 xchk_update_health(
188 	struct xfs_scrub	*sc)
189 {
190 	struct xfs_perag	*pag;
191 	struct xfs_rtgroup	*rtg;
192 	bool			bad;
193 
194 	/*
195 	 * The HEALTHY scrub type is a request from userspace to clear all the
196 	 * indirect flags after a clean scan of the entire filesystem.  As such
197 	 * there's no sick flag defined for it, so we branch here ahead of the
198 	 * mask check.
199 	 */
200 	if (sc->sm->sm_type == XFS_SCRUB_TYPE_HEALTHY &&
201 	    !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
202 		xchk_mark_all_healthy(sc->mp);
203 		return;
204 	}
205 
206 	if (!sc->sick_mask)
207 		return;
208 
209 	bad = (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
210 				   XFS_SCRUB_OFLAG_XCORRUPT));
211 	switch (type_to_health_flag[sc->sm->sm_type].group) {
212 	case XHG_AG:
213 		pag = xfs_perag_get(sc->mp, sc->sm->sm_agno);
214 		if (bad)
215 			xfs_group_mark_corrupt(pag_group(pag), sc->sick_mask);
216 		else
217 			xfs_group_mark_healthy(pag_group(pag), sc->sick_mask);
218 		xfs_perag_put(pag);
219 		break;
220 	case XHG_INO:
221 		if (!sc->ip)
222 			return;
223 		if (bad) {
224 			unsigned int	mask = sc->sick_mask;
225 
226 			/*
227 			 * If we're coming in for repairs then we don't want
228 			 * sickness flags to propagate to the incore health
229 			 * status if the inode gets inactivated before we can
230 			 * fix it.
231 			 */
232 			if (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)
233 				mask |= XFS_SICK_INO_FORGET;
234 			xfs_inode_mark_corrupt(sc->ip, mask);
235 		} else
236 			xfs_inode_mark_healthy(sc->ip, sc->sick_mask);
237 		break;
238 	case XHG_FS:
239 		if (bad)
240 			xfs_fs_mark_corrupt(sc->mp, sc->sick_mask);
241 		else
242 			xfs_fs_mark_healthy(sc->mp, sc->sick_mask);
243 		break;
244 	case XHG_RTGROUP:
245 		rtg = xfs_rtgroup_get(sc->mp, sc->sm->sm_agno);
246 		if (bad)
247 			xfs_group_mark_corrupt(rtg_group(rtg), sc->sick_mask);
248 		else
249 			xfs_group_mark_healthy(rtg_group(rtg), sc->sick_mask);
250 		xfs_rtgroup_put(rtg);
251 		break;
252 	default:
253 		ASSERT(0);
254 		break;
255 	}
256 }
257 
258 /* Is the given per-AG btree healthy enough for scanning? */
259 void
260 xchk_ag_btree_del_cursor_if_sick(
261 	struct xfs_scrub	*sc,
262 	struct xfs_btree_cur	**curp,
263 	unsigned int		sm_type)
264 {
265 	unsigned int		mask = (*curp)->bc_ops->sick_mask;
266 
267 	/*
268 	 * We always want the cursor if it's the same type as whatever we're
269 	 * scrubbing, even if we already know the structure is corrupt.
270 	 *
271 	 * Otherwise, we're only interested in the btree for cross-referencing.
272 	 * If we know the btree is bad then don't bother, just set XFAIL.
273 	 */
274 	if (sc->sm->sm_type == sm_type)
275 		return;
276 
277 	/*
278 	 * If we just repaired some AG metadata, sc->sick_mask will reflect all
279 	 * the per-AG metadata types that were repaired.  Exclude these from
280 	 * the filesystem health query because we have not yet updated the
281 	 * health status and we want everything to be scanned.
282 	 */
283 	if ((sc->flags & XREP_ALREADY_FIXED) &&
284 	    type_to_health_flag[sc->sm->sm_type].group == XHG_AG)
285 		mask &= ~sc->sick_mask;
286 
287 	if (xfs_group_has_sickness((*curp)->bc_group, mask)) {
288 		sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
289 		xfs_btree_del_cursor(*curp, XFS_BTREE_NOERROR);
290 		*curp = NULL;
291 	}
292 }
293 
294 /*
295  * Quick scan to double-check that there isn't any evidence of lingering
296  * primary health problems.  If we're still clear, then the health update will
297  * take care of clearing the indirect evidence.
298  */
299 int
300 xchk_health_record(
301 	struct xfs_scrub	*sc)
302 {
303 	struct xfs_mount	*mp = sc->mp;
304 	struct xfs_perag	*pag = NULL;
305 	struct xfs_rtgroup	*rtg = NULL;
306 	unsigned int		sick;
307 	unsigned int		checked;
308 
309 	xfs_fs_measure_sickness(mp, &sick, &checked);
310 	if (sick & XFS_SICK_FS_PRIMARY)
311 		xchk_set_corrupt(sc);
312 
313 	while ((pag = xfs_perag_next(mp, pag))) {
314 		xfs_group_measure_sickness(pag_group(pag), &sick, &checked);
315 		if (sick & XFS_SICK_AG_PRIMARY)
316 			xchk_set_corrupt(sc);
317 	}
318 
319 	while ((rtg = xfs_rtgroup_next(mp, rtg))) {
320 		xfs_group_measure_sickness(rtg_group(rtg), &sick, &checked);
321 		if (sick & XFS_SICK_RG_PRIMARY)
322 			xchk_set_corrupt(sc);
323 	}
324 
325 	return 0;
326 }
327