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_NONE = 1,
75 XHG_FS,
76 XHG_AG,
77 XHG_INO,
78 XHG_RTGROUP,
79 };
80
81 struct xchk_health_map {
82 enum xchk_health_group group;
83 unsigned int sick_mask;
84 };
85
86 static const struct xchk_health_map type_to_health_flag[XFS_SCRUB_TYPE_NR] = {
87 [XFS_SCRUB_TYPE_PROBE] = { XHG_NONE, 0 },
88 [XFS_SCRUB_TYPE_SB] = { XHG_AG, XFS_SICK_AG_SB },
89 [XFS_SCRUB_TYPE_AGF] = { XHG_AG, XFS_SICK_AG_AGF },
90 [XFS_SCRUB_TYPE_AGFL] = { XHG_AG, XFS_SICK_AG_AGFL },
91 [XFS_SCRUB_TYPE_AGI] = { XHG_AG, XFS_SICK_AG_AGI },
92 [XFS_SCRUB_TYPE_BNOBT] = { XHG_AG, XFS_SICK_AG_BNOBT },
93 [XFS_SCRUB_TYPE_CNTBT] = { XHG_AG, XFS_SICK_AG_CNTBT },
94 [XFS_SCRUB_TYPE_INOBT] = { XHG_AG, XFS_SICK_AG_INOBT },
95 [XFS_SCRUB_TYPE_FINOBT] = { XHG_AG, XFS_SICK_AG_FINOBT },
96 [XFS_SCRUB_TYPE_RMAPBT] = { XHG_AG, XFS_SICK_AG_RMAPBT },
97 [XFS_SCRUB_TYPE_REFCNTBT] = { XHG_AG, XFS_SICK_AG_REFCNTBT },
98 [XFS_SCRUB_TYPE_INODE] = { XHG_INO, XFS_SICK_INO_CORE },
99 [XFS_SCRUB_TYPE_BMBTD] = { XHG_INO, XFS_SICK_INO_BMBTD },
100 [XFS_SCRUB_TYPE_BMBTA] = { XHG_INO, XFS_SICK_INO_BMBTA },
101 [XFS_SCRUB_TYPE_BMBTC] = { XHG_INO, XFS_SICK_INO_BMBTC },
102 [XFS_SCRUB_TYPE_DIR] = { XHG_INO, XFS_SICK_INO_DIR },
103 [XFS_SCRUB_TYPE_XATTR] = { XHG_INO, XFS_SICK_INO_XATTR },
104 [XFS_SCRUB_TYPE_SYMLINK] = { XHG_INO, XFS_SICK_INO_SYMLINK },
105 [XFS_SCRUB_TYPE_PARENT] = { XHG_INO, XFS_SICK_INO_PARENT },
106 [XFS_SCRUB_TYPE_RTBITMAP] = { XHG_RTGROUP, XFS_SICK_RG_BITMAP },
107 [XFS_SCRUB_TYPE_RTSUM] = { XHG_RTGROUP, XFS_SICK_RG_SUMMARY },
108 [XFS_SCRUB_TYPE_UQUOTA] = { XHG_FS, XFS_SICK_FS_UQUOTA },
109 [XFS_SCRUB_TYPE_GQUOTA] = { XHG_FS, XFS_SICK_FS_GQUOTA },
110 [XFS_SCRUB_TYPE_PQUOTA] = { XHG_FS, XFS_SICK_FS_PQUOTA },
111 [XFS_SCRUB_TYPE_FSCOUNTERS] = { XHG_FS, XFS_SICK_FS_COUNTERS },
112 [XFS_SCRUB_TYPE_QUOTACHECK] = { XHG_FS, XFS_SICK_FS_QUOTACHECK },
113 [XFS_SCRUB_TYPE_NLINKS] = { XHG_FS, XFS_SICK_FS_NLINKS },
114 [XFS_SCRUB_TYPE_DIRTREE] = { XHG_INO, XFS_SICK_INO_DIRTREE },
115 [XFS_SCRUB_TYPE_METAPATH] = { XHG_FS, XFS_SICK_FS_METAPATH },
116 [XFS_SCRUB_TYPE_RGSUPER] = { XHG_RTGROUP, XFS_SICK_RG_SUPER },
117 };
118
119 /* Return the health status mask for this scrub type. */
120 unsigned int
xchk_health_mask_for_scrub_type(__u32 scrub_type)121 xchk_health_mask_for_scrub_type(
122 __u32 scrub_type)
123 {
124 return type_to_health_flag[scrub_type].sick_mask;
125 }
126
127 /*
128 * If the scrub state is clean, add @mask to the scrub sick mask to clear
129 * additional sick flags from the metadata object's sick state.
130 */
131 void
xchk_mark_healthy_if_clean(struct xfs_scrub * sc,unsigned int mask)132 xchk_mark_healthy_if_clean(
133 struct xfs_scrub *sc,
134 unsigned int mask)
135 {
136 if (!(sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
137 XFS_SCRUB_OFLAG_XCORRUPT)))
138 sc->healthy_mask |= mask;
139 }
140
141 /*
142 * If we're scrubbing a piece of file metadata for the first time, does it look
143 * like it has been zapped? Skip the check if we just repaired the metadata
144 * and are revalidating it.
145 */
146 bool
xchk_file_looks_zapped(struct xfs_scrub * sc,unsigned int mask)147 xchk_file_looks_zapped(
148 struct xfs_scrub *sc,
149 unsigned int mask)
150 {
151 ASSERT((mask & ~XFS_SICK_INO_ZAPPED) == 0);
152
153 if (sc->flags & XREP_ALREADY_FIXED)
154 return false;
155
156 return xfs_inode_has_sickness(sc->ip, mask);
157 }
158
159 /*
160 * Scrub gave the filesystem a clean bill of health, so clear all the indirect
161 * markers of past problems (at least for the fs and ags) so that we can be
162 * healthy again.
163 */
164 STATIC void
xchk_mark_all_healthy(struct xfs_mount * mp)165 xchk_mark_all_healthy(
166 struct xfs_mount *mp)
167 {
168 struct xfs_perag *pag = NULL;
169 struct xfs_rtgroup *rtg = NULL;
170
171 xfs_fs_mark_healthy(mp, XFS_SICK_FS_INDIRECT);
172 while ((pag = xfs_perag_next(mp, pag)))
173 xfs_group_mark_healthy(pag_group(pag), XFS_SICK_AG_INDIRECT);
174 while ((rtg = xfs_rtgroup_next(mp, rtg)))
175 xfs_group_mark_healthy(rtg_group(rtg), XFS_SICK_RG_INDIRECT);
176 }
177
178 /*
179 * Update filesystem health assessments based on what we found and did.
180 *
181 * If the scrubber finds errors, we mark sick whatever's mentioned in
182 * sick_mask, no matter whether this is a first scan or an
183 * evaluation of repair effectiveness.
184 *
185 * Otherwise, no direct corruption was found, so mark whatever's in
186 * sick_mask as healthy.
187 */
188 void
xchk_update_health(struct xfs_scrub * sc)189 xchk_update_health(
190 struct xfs_scrub *sc)
191 {
192 struct xfs_perag *pag;
193 struct xfs_rtgroup *rtg;
194 unsigned int mask = sc->sick_mask;
195 bool bad;
196
197 /*
198 * The HEALTHY scrub type is a request from userspace to clear all the
199 * indirect flags after a clean scan of the entire filesystem. As such
200 * there's no sick flag defined for it, so we branch here ahead of the
201 * mask check.
202 */
203 if (sc->sm->sm_type == XFS_SCRUB_TYPE_HEALTHY &&
204 !(sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)) {
205 xchk_mark_all_healthy(sc->mp);
206 return;
207 }
208
209 bad = (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
210 XFS_SCRUB_OFLAG_XCORRUPT));
211 if (!bad)
212 mask |= sc->healthy_mask;
213 switch (type_to_health_flag[sc->sm->sm_type].group) {
214 case XHG_NONE:
215 break;
216 case XHG_AG:
217 if (!mask)
218 return;
219 pag = xfs_perag_get(sc->mp, sc->sm->sm_agno);
220 if (bad)
221 xfs_group_mark_corrupt(pag_group(pag), mask);
222 else
223 xfs_group_mark_healthy(pag_group(pag), mask);
224 xfs_perag_put(pag);
225 break;
226 case XHG_INO:
227 if (!sc->ip)
228 return;
229 /*
230 * If we're coming in for repairs then we don't want sickness
231 * flags to propagate to the incore health status if the inode
232 * gets inactivated before we can fix it.
233 */
234 if (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)
235 mask |= XFS_SICK_INO_FORGET;
236 if (!mask)
237 return;
238 if (bad)
239 xfs_inode_mark_corrupt(sc->ip, mask);
240 else
241 xfs_inode_mark_healthy(sc->ip, mask);
242 break;
243 case XHG_FS:
244 if (!mask)
245 return;
246 if (bad)
247 xfs_fs_mark_corrupt(sc->mp, mask);
248 else
249 xfs_fs_mark_healthy(sc->mp, mask);
250 break;
251 case XHG_RTGROUP:
252 if (!mask)
253 return;
254 rtg = xfs_rtgroup_get(sc->mp, sc->sm->sm_agno);
255 if (bad)
256 xfs_group_mark_corrupt(rtg_group(rtg), mask);
257 else
258 xfs_group_mark_healthy(rtg_group(rtg), mask);
259 xfs_rtgroup_put(rtg);
260 break;
261 default:
262 ASSERT(0);
263 break;
264 }
265 }
266
267 /* Is the given per-AG btree healthy enough for scanning? */
268 void
xchk_ag_btree_del_cursor_if_sick(struct xfs_scrub * sc,struct xfs_btree_cur ** curp,unsigned int sm_type)269 xchk_ag_btree_del_cursor_if_sick(
270 struct xfs_scrub *sc,
271 struct xfs_btree_cur **curp,
272 unsigned int sm_type)
273 {
274 unsigned int mask = (*curp)->bc_ops->sick_mask;
275
276 /*
277 * We always want the cursor if it's the same type as whatever we're
278 * scrubbing, even if we already know the structure is corrupt.
279 *
280 * Otherwise, we're only interested in the btree for cross-referencing.
281 * If we know the btree is bad then don't bother, just set XFAIL.
282 */
283 if (sc->sm->sm_type == sm_type)
284 return;
285
286 /*
287 * If we just repaired some AG metadata, sc->sick_mask will reflect all
288 * the per-AG metadata types that were repaired. Exclude these from
289 * the filesystem health query because we have not yet updated the
290 * health status and we want everything to be scanned.
291 */
292 if ((sc->flags & XREP_ALREADY_FIXED) &&
293 type_to_health_flag[sc->sm->sm_type].group == XHG_AG)
294 mask &= ~sc->sick_mask;
295
296 if (xfs_group_has_sickness((*curp)->bc_group, mask)) {
297 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
298 xfs_btree_del_cursor(*curp, XFS_BTREE_NOERROR);
299 *curp = NULL;
300 }
301 }
302
303 /*
304 * Quick scan to double-check that there isn't any evidence of lingering
305 * primary health problems. If we're still clear, then the health update will
306 * take care of clearing the indirect evidence.
307 */
308 int
xchk_health_record(struct xfs_scrub * sc)309 xchk_health_record(
310 struct xfs_scrub *sc)
311 {
312 struct xfs_mount *mp = sc->mp;
313 struct xfs_perag *pag = NULL;
314 struct xfs_rtgroup *rtg = NULL;
315 unsigned int sick;
316 unsigned int checked;
317
318 xfs_fs_measure_sickness(mp, &sick, &checked);
319 if (sick & XFS_SICK_FS_PRIMARY)
320 xchk_set_corrupt(sc);
321
322 while ((pag = xfs_perag_next(mp, pag))) {
323 xfs_group_measure_sickness(pag_group(pag), &sick, &checked);
324 if (sick & XFS_SICK_AG_PRIMARY)
325 xchk_set_corrupt(sc);
326 }
327
328 while ((rtg = xfs_rtgroup_next(mp, rtg))) {
329 xfs_group_measure_sickness(rtg_group(rtg), &sick, &checked);
330 if (sick & XFS_SICK_RG_PRIMARY)
331 xchk_set_corrupt(sc);
332 }
333
334 return 0;
335 }
336