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 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_btree.h"
11 #include "xfs_ag.h"
12 #include "xfs_health.h"
13 #include "scrub/scrub.h"
14 #include "scrub/health.h"
15
16 /*
17 * Scrub and In-Core Filesystem Health Assessments
18 * ===============================================
19 *
20 * Online scrub and repair have the time and the ability to perform stronger
21 * checks than we can do from the metadata verifiers, because they can
22 * cross-reference records between data structures. Therefore, scrub is in a
23 * good position to update the online filesystem health assessments to reflect
24 * the good/bad state of the data structure.
25 *
26 * We therefore extend scrub in the following ways to achieve this:
27 *
28 * 1. Create a "sick_mask" field in the scrub context. When we're setting up a
29 * scrub call, set this to the default XFS_SICK_* flag(s) for the selected
30 * scrub type (call it A). Scrub and repair functions can override the default
31 * sick_mask value if they choose.
32 *
33 * 2. If the scrubber returns a runtime error code, we exit making no changes
34 * to the incore sick state.
35 *
36 * 3. If the scrubber finds that A is clean, use sick_mask to clear the incore
37 * sick flags before exiting.
38 *
39 * 4. If the scrubber finds that A is corrupt, use sick_mask to set the incore
40 * sick flags. If the user didn't want to repair then we exit, leaving the
41 * metadata structure unfixed and the sick flag set.
42 *
43 * 5. Now we know that A is corrupt and the user wants to repair, so run the
44 * repairer. If the repairer returns an error code, we exit with that error
45 * code, having made no further changes to the incore sick state.
46 *
47 * 6. If repair rebuilds A correctly and the subsequent re-scrub of A is clean,
48 * use sick_mask to clear the incore sick flags. This should have the effect
49 * that A is no longer marked sick.
50 *
51 * 7. If repair rebuilds A incorrectly, the re-scrub will find it corrupt and
52 * use sick_mask to set the incore sick flags. This should have no externally
53 * visible effect since we already set them in step (4).
54 *
55 * There are some complications to this story, however. For certain types of
56 * complementary metadata indices (e.g. inobt/finobt), it is easier to rebuild
57 * both structures at the same time. The following principles apply to this
58 * type of repair strategy:
59 *
60 * 8. Any repair function that rebuilds multiple structures should update
61 * sick_mask_visible to reflect whatever other structures are rebuilt, and
62 * verify that all the rebuilt structures can pass a scrub check. The outcomes
63 * of 5-7 still apply, but with a sick_mask that covers everything being
64 * rebuilt.
65 */
66
67 /* Map our scrub type to a sick mask and a set of health update functions. */
68
69 enum xchk_health_group {
70 XHG_FS = 1,
71 XHG_RT,
72 XHG_AG,
73 XHG_INO,
74 };
75
76 struct xchk_health_map {
77 enum xchk_health_group group;
78 unsigned int sick_mask;
79 };
80
81 static const struct xchk_health_map type_to_health_flag[XFS_SCRUB_TYPE_NR] = {
82 [XFS_SCRUB_TYPE_SB] = { XHG_AG, XFS_SICK_AG_SB },
83 [XFS_SCRUB_TYPE_AGF] = { XHG_AG, XFS_SICK_AG_AGF },
84 [XFS_SCRUB_TYPE_AGFL] = { XHG_AG, XFS_SICK_AG_AGFL },
85 [XFS_SCRUB_TYPE_AGI] = { XHG_AG, XFS_SICK_AG_AGI },
86 [XFS_SCRUB_TYPE_BNOBT] = { XHG_AG, XFS_SICK_AG_BNOBT },
87 [XFS_SCRUB_TYPE_CNTBT] = { XHG_AG, XFS_SICK_AG_CNTBT },
88 [XFS_SCRUB_TYPE_INOBT] = { XHG_AG, XFS_SICK_AG_INOBT },
89 [XFS_SCRUB_TYPE_FINOBT] = { XHG_AG, XFS_SICK_AG_FINOBT },
90 [XFS_SCRUB_TYPE_RMAPBT] = { XHG_AG, XFS_SICK_AG_RMAPBT },
91 [XFS_SCRUB_TYPE_REFCNTBT] = { XHG_AG, XFS_SICK_AG_REFCNTBT },
92 [XFS_SCRUB_TYPE_INODE] = { XHG_INO, XFS_SICK_INO_CORE },
93 [XFS_SCRUB_TYPE_BMBTD] = { XHG_INO, XFS_SICK_INO_BMBTD },
94 [XFS_SCRUB_TYPE_BMBTA] = { XHG_INO, XFS_SICK_INO_BMBTA },
95 [XFS_SCRUB_TYPE_BMBTC] = { XHG_INO, XFS_SICK_INO_BMBTC },
96 [XFS_SCRUB_TYPE_DIR] = { XHG_INO, XFS_SICK_INO_DIR },
97 [XFS_SCRUB_TYPE_XATTR] = { XHG_INO, XFS_SICK_INO_XATTR },
98 [XFS_SCRUB_TYPE_SYMLINK] = { XHG_INO, XFS_SICK_INO_SYMLINK },
99 [XFS_SCRUB_TYPE_PARENT] = { XHG_INO, XFS_SICK_INO_PARENT },
100 [XFS_SCRUB_TYPE_RTBITMAP] = { XHG_RT, XFS_SICK_RT_BITMAP },
101 [XFS_SCRUB_TYPE_RTSUM] = { XHG_RT, XFS_SICK_RT_SUMMARY },
102 [XFS_SCRUB_TYPE_UQUOTA] = { XHG_FS, XFS_SICK_FS_UQUOTA },
103 [XFS_SCRUB_TYPE_GQUOTA] = { XHG_FS, XFS_SICK_FS_GQUOTA },
104 [XFS_SCRUB_TYPE_PQUOTA] = { XHG_FS, XFS_SICK_FS_PQUOTA },
105 [XFS_SCRUB_TYPE_FSCOUNTERS] = { XHG_FS, XFS_SICK_FS_COUNTERS },
106 };
107
108 /* Return the health status mask for this scrub type. */
109 unsigned int
xchk_health_mask_for_scrub_type(__u32 scrub_type)110 xchk_health_mask_for_scrub_type(
111 __u32 scrub_type)
112 {
113 return type_to_health_flag[scrub_type].sick_mask;
114 }
115
116 /*
117 * Update filesystem health assessments based on what we found and did.
118 *
119 * If the scrubber finds errors, we mark sick whatever's mentioned in
120 * sick_mask, no matter whether this is a first scan or an
121 * evaluation of repair effectiveness.
122 *
123 * Otherwise, no direct corruption was found, so mark whatever's in
124 * sick_mask as healthy.
125 */
126 void
xchk_update_health(struct xfs_scrub * sc)127 xchk_update_health(
128 struct xfs_scrub *sc)
129 {
130 struct xfs_perag *pag;
131 bool bad;
132
133 if (!sc->sick_mask)
134 return;
135
136 bad = (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
137 XFS_SCRUB_OFLAG_XCORRUPT));
138 switch (type_to_health_flag[sc->sm->sm_type].group) {
139 case XHG_AG:
140 pag = xfs_perag_get(sc->mp, sc->sm->sm_agno);
141 if (bad)
142 xfs_ag_mark_sick(pag, sc->sick_mask);
143 else
144 xfs_ag_mark_healthy(pag, sc->sick_mask);
145 xfs_perag_put(pag);
146 break;
147 case XHG_INO:
148 if (!sc->ip)
149 return;
150 if (bad)
151 xfs_inode_mark_sick(sc->ip, sc->sick_mask);
152 else
153 xfs_inode_mark_healthy(sc->ip, sc->sick_mask);
154 break;
155 case XHG_FS:
156 if (bad)
157 xfs_fs_mark_sick(sc->mp, sc->sick_mask);
158 else
159 xfs_fs_mark_healthy(sc->mp, sc->sick_mask);
160 break;
161 case XHG_RT:
162 if (bad)
163 xfs_rt_mark_sick(sc->mp, sc->sick_mask);
164 else
165 xfs_rt_mark_healthy(sc->mp, sc->sick_mask);
166 break;
167 default:
168 ASSERT(0);
169 break;
170 }
171 }
172
173 /* Is the given per-AG btree healthy enough for scanning? */
174 bool
xchk_ag_btree_healthy_enough(struct xfs_scrub * sc,struct xfs_perag * pag,xfs_btnum_t btnum)175 xchk_ag_btree_healthy_enough(
176 struct xfs_scrub *sc,
177 struct xfs_perag *pag,
178 xfs_btnum_t btnum)
179 {
180 unsigned int mask = 0;
181
182 /*
183 * We always want the cursor if it's the same type as whatever we're
184 * scrubbing, even if we already know the structure is corrupt.
185 *
186 * Otherwise, we're only interested in the btree for cross-referencing.
187 * If we know the btree is bad then don't bother, just set XFAIL.
188 */
189 switch (btnum) {
190 case XFS_BTNUM_BNO:
191 if (sc->sm->sm_type == XFS_SCRUB_TYPE_BNOBT)
192 return true;
193 mask = XFS_SICK_AG_BNOBT;
194 break;
195 case XFS_BTNUM_CNT:
196 if (sc->sm->sm_type == XFS_SCRUB_TYPE_CNTBT)
197 return true;
198 mask = XFS_SICK_AG_CNTBT;
199 break;
200 case XFS_BTNUM_INO:
201 if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
202 return true;
203 mask = XFS_SICK_AG_INOBT;
204 break;
205 case XFS_BTNUM_FINO:
206 if (sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT)
207 return true;
208 mask = XFS_SICK_AG_FINOBT;
209 break;
210 case XFS_BTNUM_RMAP:
211 if (sc->sm->sm_type == XFS_SCRUB_TYPE_RMAPBT)
212 return true;
213 mask = XFS_SICK_AG_RMAPBT;
214 break;
215 case XFS_BTNUM_REFC:
216 if (sc->sm->sm_type == XFS_SCRUB_TYPE_REFCNTBT)
217 return true;
218 mask = XFS_SICK_AG_REFCNTBT;
219 break;
220 default:
221 ASSERT(0);
222 return true;
223 }
224
225 if (xfs_ag_has_sickness(pag, mask)) {
226 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
227 return false;
228 }
229
230 return true;
231 }
232