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_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_ag.h"
16 #include "xfs_health.h"
17 #include "scrub/scrub.h"
18 #include "scrub/health.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_RT,
76 XHG_AG,
77 XHG_INO,
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_RT, XFS_SICK_RT_BITMAP },
105 [XFS_SCRUB_TYPE_RTSUM] = { XHG_RT, XFS_SICK_RT_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 };
111
112 /* Return the health status mask for this scrub type. */
113 unsigned int
xchk_health_mask_for_scrub_type(__u32 scrub_type)114 xchk_health_mask_for_scrub_type(
115 __u32 scrub_type)
116 {
117 return type_to_health_flag[scrub_type].sick_mask;
118 }
119
120 /*
121 * Update filesystem health assessments based on what we found and did.
122 *
123 * If the scrubber finds errors, we mark sick whatever's mentioned in
124 * sick_mask, no matter whether this is a first scan or an
125 * evaluation of repair effectiveness.
126 *
127 * Otherwise, no direct corruption was found, so mark whatever's in
128 * sick_mask as healthy.
129 */
130 void
xchk_update_health(struct xfs_scrub * sc)131 xchk_update_health(
132 struct xfs_scrub *sc)
133 {
134 struct xfs_perag *pag;
135 bool bad;
136
137 if (!sc->sick_mask)
138 return;
139
140 bad = (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
141 XFS_SCRUB_OFLAG_XCORRUPT));
142 switch (type_to_health_flag[sc->sm->sm_type].group) {
143 case XHG_AG:
144 pag = xfs_perag_get(sc->mp, sc->sm->sm_agno);
145 if (bad)
146 xfs_ag_mark_sick(pag, sc->sick_mask);
147 else
148 xfs_ag_mark_healthy(pag, sc->sick_mask);
149 xfs_perag_put(pag);
150 break;
151 case XHG_INO:
152 if (!sc->ip)
153 return;
154 if (bad)
155 xfs_inode_mark_sick(sc->ip, sc->sick_mask);
156 else
157 xfs_inode_mark_healthy(sc->ip, sc->sick_mask);
158 break;
159 case XHG_FS:
160 if (bad)
161 xfs_fs_mark_sick(sc->mp, sc->sick_mask);
162 else
163 xfs_fs_mark_healthy(sc->mp, sc->sick_mask);
164 break;
165 case XHG_RT:
166 if (bad)
167 xfs_rt_mark_sick(sc->mp, sc->sick_mask);
168 else
169 xfs_rt_mark_healthy(sc->mp, sc->sick_mask);
170 break;
171 default:
172 ASSERT(0);
173 break;
174 }
175 }
176
177 /* Is the given per-AG btree healthy enough for scanning? */
178 bool
xchk_ag_btree_healthy_enough(struct xfs_scrub * sc,struct xfs_perag * pag,xfs_btnum_t btnum)179 xchk_ag_btree_healthy_enough(
180 struct xfs_scrub *sc,
181 struct xfs_perag *pag,
182 xfs_btnum_t btnum)
183 {
184 unsigned int mask = 0;
185
186 /*
187 * We always want the cursor if it's the same type as whatever we're
188 * scrubbing, even if we already know the structure is corrupt.
189 *
190 * Otherwise, we're only interested in the btree for cross-referencing.
191 * If we know the btree is bad then don't bother, just set XFAIL.
192 */
193 switch (btnum) {
194 case XFS_BTNUM_BNO:
195 if (sc->sm->sm_type == XFS_SCRUB_TYPE_BNOBT)
196 return true;
197 mask = XFS_SICK_AG_BNOBT;
198 break;
199 case XFS_BTNUM_CNT:
200 if (sc->sm->sm_type == XFS_SCRUB_TYPE_CNTBT)
201 return true;
202 mask = XFS_SICK_AG_CNTBT;
203 break;
204 case XFS_BTNUM_INO:
205 if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
206 return true;
207 mask = XFS_SICK_AG_INOBT;
208 break;
209 case XFS_BTNUM_FINO:
210 if (sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT)
211 return true;
212 mask = XFS_SICK_AG_FINOBT;
213 break;
214 case XFS_BTNUM_RMAP:
215 if (sc->sm->sm_type == XFS_SCRUB_TYPE_RMAPBT)
216 return true;
217 mask = XFS_SICK_AG_RMAPBT;
218 break;
219 case XFS_BTNUM_REFC:
220 if (sc->sm->sm_type == XFS_SCRUB_TYPE_REFCNTBT)
221 return true;
222 mask = XFS_SICK_AG_REFCNTBT;
223 break;
224 default:
225 ASSERT(0);
226 return true;
227 }
228
229 /*
230 * If we just repaired some AG metadata, sc->sick_mask will reflect all
231 * the per-AG metadata types that were repaired. Exclude these from
232 * the filesystem health query because we have not yet updated the
233 * health status and we want everything to be scanned.
234 */
235 if ((sc->flags & XREP_ALREADY_FIXED) &&
236 type_to_health_flag[sc->sm->sm_type].group == XHG_AG)
237 mask &= ~sc->sick_mask;
238
239 if (xfs_ag_has_sickness(pag, mask)) {
240 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
241 return false;
242 }
243
244 return true;
245 }
246