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