1 // SPDX-License-Identifier: GPL-2.0-or-later !! 1 // SPDX-License-Identifier: GPL-2.0+
2 /* 2 /*
3 * Copyright (C) 2019-2023 Oracle. All Rights !! 3 * Copyright (C) 2019 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <djwong@kernel.org> !! 4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
5 */ 5 */
6 #include "xfs.h" 6 #include "xfs.h"
7 #include "xfs_fs.h" 7 #include "xfs_fs.h"
8 #include "xfs_shared.h" 8 #include "xfs_shared.h"
9 #include "xfs_format.h" 9 #include "xfs_format.h"
>> 10 #include "xfs_btree.h"
10 #include "xfs_trans_resv.h" 11 #include "xfs_trans_resv.h"
11 #include "xfs_mount.h" 12 #include "xfs_mount.h"
12 #include "xfs_btree.h" <<
13 #include "xfs_ag.h" 13 #include "xfs_ag.h"
14 #include "xfs_health.h" 14 #include "xfs_health.h"
15 #include "scrub/scrub.h" 15 #include "scrub/scrub.h"
16 #include "scrub/health.h" 16 #include "scrub/health.h"
17 #include "scrub/common.h" <<
18 17
19 /* 18 /*
20 * Scrub and In-Core Filesystem Health Assessm 19 * Scrub and In-Core Filesystem Health Assessments
21 * =========================================== 20 * ===============================================
22 * 21 *
23 * Online scrub and repair have the time and t 22 * Online scrub and repair have the time and the ability to perform stronger
24 * checks than we can do from the metadata ver 23 * checks than we can do from the metadata verifiers, because they can
25 * cross-reference records between data struct 24 * cross-reference records between data structures. Therefore, scrub is in a
26 * good position to update the online filesyst 25 * good position to update the online filesystem health assessments to reflect
27 * the good/bad state of the data structure. 26 * the good/bad state of the data structure.
28 * 27 *
29 * We therefore extend scrub in the following 28 * We therefore extend scrub in the following ways to achieve this:
30 * 29 *
31 * 1. Create a "sick_mask" field in the scrub 30 * 1. Create a "sick_mask" field in the scrub context. When we're setting up a
32 * scrub call, set this to the default XFS_SIC 31 * scrub call, set this to the default XFS_SICK_* flag(s) for the selected
33 * scrub type (call it A). Scrub and repair f 32 * scrub type (call it A). Scrub and repair functions can override the default
34 * sick_mask value if they choose. 33 * sick_mask value if they choose.
35 * 34 *
36 * 2. If the scrubber returns a runtime error 35 * 2. If the scrubber returns a runtime error code, we exit making no changes
37 * to the incore sick state. 36 * to the incore sick state.
38 * 37 *
39 * 3. If the scrubber finds that A is clean, u 38 * 3. If the scrubber finds that A is clean, use sick_mask to clear the incore
40 * sick flags before exiting. 39 * sick flags before exiting.
41 * 40 *
42 * 4. If the scrubber finds that A is corrupt, 41 * 4. If the scrubber finds that A is corrupt, use sick_mask to set the incore
43 * sick flags. If the user didn't want to rep 42 * sick flags. If the user didn't want to repair then we exit, leaving the
44 * metadata structure unfixed and the sick fla 43 * metadata structure unfixed and the sick flag set.
45 * 44 *
46 * 5. Now we know that A is corrupt and the us 45 * 5. Now we know that A is corrupt and the user wants to repair, so run the
47 * repairer. If the repairer returns an error 46 * repairer. If the repairer returns an error code, we exit with that error
48 * code, having made no further changes to the 47 * code, having made no further changes to the incore sick state.
49 * 48 *
50 * 6. If repair rebuilds A correctly and the s 49 * 6. If repair rebuilds A correctly and the subsequent re-scrub of A is clean,
51 * use sick_mask to clear the incore sick flag 50 * use sick_mask to clear the incore sick flags. This should have the effect
52 * that A is no longer marked sick. 51 * that A is no longer marked sick.
53 * 52 *
54 * 7. If repair rebuilds A incorrectly, the re 53 * 7. If repair rebuilds A incorrectly, the re-scrub will find it corrupt and
55 * use sick_mask to set the incore sick flags. 54 * use sick_mask to set the incore sick flags. This should have no externally
56 * visible effect since we already set them in 55 * visible effect since we already set them in step (4).
57 * 56 *
58 * There are some complications to this story, 57 * There are some complications to this story, however. For certain types of
59 * complementary metadata indices (e.g. inobt/ 58 * complementary metadata indices (e.g. inobt/finobt), it is easier to rebuild
60 * both structures at the same time. The foll 59 * both structures at the same time. The following principles apply to this
61 * type of repair strategy: 60 * type of repair strategy:
62 * 61 *
63 * 8. Any repair function that rebuilds multip 62 * 8. Any repair function that rebuilds multiple structures should update
64 * sick_mask_visible to reflect whatever other 63 * sick_mask_visible to reflect whatever other structures are rebuilt, and
65 * verify that all the rebuilt structures can 64 * verify that all the rebuilt structures can pass a scrub check. The outcomes
66 * of 5-7 still apply, but with a sick_mask th 65 * of 5-7 still apply, but with a sick_mask that covers everything being
67 * rebuilt. 66 * rebuilt.
68 */ 67 */
69 68
70 /* Map our scrub type to a sick mask and a set 69 /* Map our scrub type to a sick mask and a set of health update functions. */
71 70
72 enum xchk_health_group { 71 enum xchk_health_group {
73 XHG_FS = 1, 72 XHG_FS = 1,
74 XHG_RT, 73 XHG_RT,
75 XHG_AG, 74 XHG_AG,
76 XHG_INO, 75 XHG_INO,
77 }; 76 };
78 77
79 struct xchk_health_map { 78 struct xchk_health_map {
80 enum xchk_health_group group; 79 enum xchk_health_group group;
81 unsigned int sick_mask; 80 unsigned int sick_mask;
82 }; 81 };
83 82
84 static const struct xchk_health_map type_to_he 83 static const struct xchk_health_map type_to_health_flag[XFS_SCRUB_TYPE_NR] = {
85 [XFS_SCRUB_TYPE_SB] = { XH 84 [XFS_SCRUB_TYPE_SB] = { XHG_AG, XFS_SICK_AG_SB },
86 [XFS_SCRUB_TYPE_AGF] = { XH 85 [XFS_SCRUB_TYPE_AGF] = { XHG_AG, XFS_SICK_AG_AGF },
87 [XFS_SCRUB_TYPE_AGFL] = { XH 86 [XFS_SCRUB_TYPE_AGFL] = { XHG_AG, XFS_SICK_AG_AGFL },
88 [XFS_SCRUB_TYPE_AGI] = { XH 87 [XFS_SCRUB_TYPE_AGI] = { XHG_AG, XFS_SICK_AG_AGI },
89 [XFS_SCRUB_TYPE_BNOBT] = { XH 88 [XFS_SCRUB_TYPE_BNOBT] = { XHG_AG, XFS_SICK_AG_BNOBT },
90 [XFS_SCRUB_TYPE_CNTBT] = { XH 89 [XFS_SCRUB_TYPE_CNTBT] = { XHG_AG, XFS_SICK_AG_CNTBT },
91 [XFS_SCRUB_TYPE_INOBT] = { XH 90 [XFS_SCRUB_TYPE_INOBT] = { XHG_AG, XFS_SICK_AG_INOBT },
92 [XFS_SCRUB_TYPE_FINOBT] = { XH 91 [XFS_SCRUB_TYPE_FINOBT] = { XHG_AG, XFS_SICK_AG_FINOBT },
93 [XFS_SCRUB_TYPE_RMAPBT] = { XH 92 [XFS_SCRUB_TYPE_RMAPBT] = { XHG_AG, XFS_SICK_AG_RMAPBT },
94 [XFS_SCRUB_TYPE_REFCNTBT] = { XH 93 [XFS_SCRUB_TYPE_REFCNTBT] = { XHG_AG, XFS_SICK_AG_REFCNTBT },
95 [XFS_SCRUB_TYPE_INODE] = { XH 94 [XFS_SCRUB_TYPE_INODE] = { XHG_INO, XFS_SICK_INO_CORE },
96 [XFS_SCRUB_TYPE_BMBTD] = { XH 95 [XFS_SCRUB_TYPE_BMBTD] = { XHG_INO, XFS_SICK_INO_BMBTD },
97 [XFS_SCRUB_TYPE_BMBTA] = { XH 96 [XFS_SCRUB_TYPE_BMBTA] = { XHG_INO, XFS_SICK_INO_BMBTA },
98 [XFS_SCRUB_TYPE_BMBTC] = { XH 97 [XFS_SCRUB_TYPE_BMBTC] = { XHG_INO, XFS_SICK_INO_BMBTC },
99 [XFS_SCRUB_TYPE_DIR] = { XH 98 [XFS_SCRUB_TYPE_DIR] = { XHG_INO, XFS_SICK_INO_DIR },
100 [XFS_SCRUB_TYPE_XATTR] = { XH 99 [XFS_SCRUB_TYPE_XATTR] = { XHG_INO, XFS_SICK_INO_XATTR },
101 [XFS_SCRUB_TYPE_SYMLINK] = { XH 100 [XFS_SCRUB_TYPE_SYMLINK] = { XHG_INO, XFS_SICK_INO_SYMLINK },
102 [XFS_SCRUB_TYPE_PARENT] = { XH 101 [XFS_SCRUB_TYPE_PARENT] = { XHG_INO, XFS_SICK_INO_PARENT },
103 [XFS_SCRUB_TYPE_RTBITMAP] = { XH 102 [XFS_SCRUB_TYPE_RTBITMAP] = { XHG_RT, XFS_SICK_RT_BITMAP },
104 [XFS_SCRUB_TYPE_RTSUM] = { XH 103 [XFS_SCRUB_TYPE_RTSUM] = { XHG_RT, XFS_SICK_RT_SUMMARY },
105 [XFS_SCRUB_TYPE_UQUOTA] = { XH 104 [XFS_SCRUB_TYPE_UQUOTA] = { XHG_FS, XFS_SICK_FS_UQUOTA },
106 [XFS_SCRUB_TYPE_GQUOTA] = { XH 105 [XFS_SCRUB_TYPE_GQUOTA] = { XHG_FS, XFS_SICK_FS_GQUOTA },
107 [XFS_SCRUB_TYPE_PQUOTA] = { XH 106 [XFS_SCRUB_TYPE_PQUOTA] = { XHG_FS, XFS_SICK_FS_PQUOTA },
108 [XFS_SCRUB_TYPE_FSCOUNTERS] = { XH 107 [XFS_SCRUB_TYPE_FSCOUNTERS] = { XHG_FS, XFS_SICK_FS_COUNTERS },
109 [XFS_SCRUB_TYPE_QUOTACHECK] = { XH <<
110 [XFS_SCRUB_TYPE_NLINKS] = { XH <<
111 [XFS_SCRUB_TYPE_DIRTREE] = { XH <<
112 }; 108 };
113 109
114 /* Return the health status mask for this scru 110 /* Return the health status mask for this scrub type. */
115 unsigned int 111 unsigned int
116 xchk_health_mask_for_scrub_type( 112 xchk_health_mask_for_scrub_type(
117 __u32 scrub_type) 113 __u32 scrub_type)
118 { 114 {
119 return type_to_health_flag[scrub_type] 115 return type_to_health_flag[scrub_type].sick_mask;
120 } 116 }
121 117
122 /* 118 /*
123 * If the scrub state is clean, add @mask to t <<
124 * additional sick flags from the metadata obj <<
125 */ <<
126 void <<
127 xchk_mark_healthy_if_clean( <<
128 struct xfs_scrub *sc, <<
129 unsigned int mask) <<
130 { <<
131 if (!(sc->sm->sm_flags & (XFS_SCRUB_OF <<
132 XFS_SCRUB_OF <<
133 sc->sick_mask |= mask; <<
134 } <<
135 <<
136 /* <<
137 * If we're scrubbing a piece of file metadata <<
138 * like it has been zapped? Skip the check if <<
139 * and are revalidating it. <<
140 */ <<
141 bool <<
142 xchk_file_looks_zapped( <<
143 struct xfs_scrub *sc, <<
144 unsigned int mask) <<
145 { <<
146 ASSERT((mask & ~XFS_SICK_INO_ZAPPED) = <<
147 <<
148 if (sc->flags & XREP_ALREADY_FIXED) <<
149 return false; <<
150 <<
151 return xfs_inode_has_sickness(sc->ip, <<
152 } <<
153 <<
154 /* <<
155 * Scrub gave the filesystem a clean bill of h <<
156 * markers of past problems (at least for the <<
157 * healthy again. <<
158 */ <<
159 STATIC void <<
160 xchk_mark_all_healthy( <<
161 struct xfs_mount *mp) <<
162 { <<
163 struct xfs_perag *pag; <<
164 xfs_agnumber_t agno; <<
165 <<
166 xfs_fs_mark_healthy(mp, XFS_SICK_FS_IN <<
167 xfs_rt_mark_healthy(mp, XFS_SICK_RT_IN <<
168 for_each_perag(mp, agno, pag) <<
169 xfs_ag_mark_healthy(pag, XFS_S <<
170 } <<
171 <<
172 /* <<
173 * Update filesystem health assessments based 119 * Update filesystem health assessments based on what we found and did.
174 * 120 *
175 * If the scrubber finds errors, we mark sick 121 * If the scrubber finds errors, we mark sick whatever's mentioned in
176 * sick_mask, no matter whether this is a firs 122 * sick_mask, no matter whether this is a first scan or an
177 * evaluation of repair effectiveness. 123 * evaluation of repair effectiveness.
178 * 124 *
179 * Otherwise, no direct corruption was found, 125 * Otherwise, no direct corruption was found, so mark whatever's in
180 * sick_mask as healthy. 126 * sick_mask as healthy.
181 */ 127 */
182 void 128 void
183 xchk_update_health( 129 xchk_update_health(
184 struct xfs_scrub *sc) 130 struct xfs_scrub *sc)
185 { 131 {
186 struct xfs_perag *pag; 132 struct xfs_perag *pag;
187 bool bad; 133 bool bad;
188 134
189 /* <<
190 * The HEALTHY scrub type is a request <<
191 * indirect flags after a clean scan o <<
192 * there's no sick flag defined for it <<
193 * mask check. <<
194 */ <<
195 if (sc->sm->sm_type == XFS_SCRUB_TYPE_ <<
196 !(sc->sm->sm_flags & XFS_SCRUB_OFL <<
197 xchk_mark_all_healthy(sc->mp); <<
198 return; <<
199 } <<
200 <<
201 if (!sc->sick_mask) 135 if (!sc->sick_mask)
202 return; 136 return;
203 137
204 bad = (sc->sm->sm_flags & (XFS_SCRUB_O 138 bad = (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
205 XFS_SCRUB_O 139 XFS_SCRUB_OFLAG_XCORRUPT));
206 switch (type_to_health_flag[sc->sm->sm 140 switch (type_to_health_flag[sc->sm->sm_type].group) {
207 case XHG_AG: 141 case XHG_AG:
208 pag = xfs_perag_get(sc->mp, sc 142 pag = xfs_perag_get(sc->mp, sc->sm->sm_agno);
209 if (bad) 143 if (bad)
210 xfs_ag_mark_corrupt(pa !! 144 xfs_ag_mark_sick(pag, sc->sick_mask);
211 else 145 else
212 xfs_ag_mark_healthy(pa 146 xfs_ag_mark_healthy(pag, sc->sick_mask);
213 xfs_perag_put(pag); 147 xfs_perag_put(pag);
214 break; 148 break;
215 case XHG_INO: 149 case XHG_INO:
216 if (!sc->ip) 150 if (!sc->ip)
217 return; 151 return;
218 if (bad) { !! 152 if (bad)
219 unsigned int mask = !! 153 xfs_inode_mark_sick(sc->ip, sc->sick_mask);
220 !! 154 else
221 /* <<
222 * If we're coming in <<
223 * sickness flags to p <<
224 * status if the inode <<
225 * fix it. <<
226 */ <<
227 if (sc->sm->sm_flags & <<
228 mask |= XFS_SI <<
229 xfs_inode_mark_corrupt <<
230 } else <<
231 xfs_inode_mark_healthy 155 xfs_inode_mark_healthy(sc->ip, sc->sick_mask);
232 break; 156 break;
233 case XHG_FS: 157 case XHG_FS:
234 if (bad) 158 if (bad)
235 xfs_fs_mark_corrupt(sc !! 159 xfs_fs_mark_sick(sc->mp, sc->sick_mask);
236 else 160 else
237 xfs_fs_mark_healthy(sc 161 xfs_fs_mark_healthy(sc->mp, sc->sick_mask);
238 break; 162 break;
239 case XHG_RT: 163 case XHG_RT:
240 if (bad) 164 if (bad)
241 xfs_rt_mark_corrupt(sc !! 165 xfs_rt_mark_sick(sc->mp, sc->sick_mask);
242 else 166 else
243 xfs_rt_mark_healthy(sc 167 xfs_rt_mark_healthy(sc->mp, sc->sick_mask);
244 break; 168 break;
245 default: 169 default:
246 ASSERT(0); 170 ASSERT(0);
247 break; 171 break;
248 } 172 }
249 } 173 }
250 174
251 /* Is the given per-AG btree healthy enough fo 175 /* Is the given per-AG btree healthy enough for scanning? */
252 void !! 176 bool
253 xchk_ag_btree_del_cursor_if_sick( !! 177 xchk_ag_btree_healthy_enough(
254 struct xfs_scrub *sc, 178 struct xfs_scrub *sc,
255 struct xfs_btree_cur **curp, !! 179 struct xfs_perag *pag,
256 unsigned int sm_type) !! 180 xfs_btnum_t btnum)
257 { 181 {
258 unsigned int mask = (*curp) !! 182 unsigned int mask = 0;
259 183
260 /* 184 /*
261 * We always want the cursor if it's t 185 * We always want the cursor if it's the same type as whatever we're
262 * scrubbing, even if we already know 186 * scrubbing, even if we already know the structure is corrupt.
263 * 187 *
264 * Otherwise, we're only interested in 188 * Otherwise, we're only interested in the btree for cross-referencing.
265 * If we know the btree is bad then do 189 * If we know the btree is bad then don't bother, just set XFAIL.
266 */ 190 */
267 if (sc->sm->sm_type == sm_type) !! 191 switch (btnum) {
268 return; !! 192 case XFS_BTNUM_BNO:
269 !! 193 if (sc->sm->sm_type == XFS_SCRUB_TYPE_BNOBT)
270 /* !! 194 return true;
271 * If we just repaired some AG metadat !! 195 mask = XFS_SICK_AG_BNOBT;
272 * the per-AG metadata types that were !! 196 break;
273 * the filesystem health query because !! 197 case XFS_BTNUM_CNT:
274 * health status and we want everythin !! 198 if (sc->sm->sm_type == XFS_SCRUB_TYPE_CNTBT)
275 */ !! 199 return true;
276 if ((sc->flags & XREP_ALREADY_FIXED) & !! 200 mask = XFS_SICK_AG_CNTBT;
277 type_to_health_flag[sc->sm->sm_typ !! 201 break;
278 mask &= ~sc->sick_mask; !! 202 case XFS_BTNUM_INO:
279 !! 203 if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
280 if (xfs_ag_has_sickness((*curp)->bc_ag !! 204 return true;
281 sc->sm->sm_flags |= XFS_SCRUB_ !! 205 mask = XFS_SICK_AG_INOBT;
282 xfs_btree_del_cursor(*curp, XF !! 206 break;
283 *curp = NULL; !! 207 case XFS_BTNUM_FINO:
>> 208 if (sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT)
>> 209 return true;
>> 210 mask = XFS_SICK_AG_FINOBT;
>> 211 break;
>> 212 case XFS_BTNUM_RMAP:
>> 213 if (sc->sm->sm_type == XFS_SCRUB_TYPE_RMAPBT)
>> 214 return true;
>> 215 mask = XFS_SICK_AG_RMAPBT;
>> 216 break;
>> 217 case XFS_BTNUM_REFC:
>> 218 if (sc->sm->sm_type == XFS_SCRUB_TYPE_REFCNTBT)
>> 219 return true;
>> 220 mask = XFS_SICK_AG_REFCNTBT;
>> 221 break;
>> 222 default:
>> 223 ASSERT(0);
>> 224 return true;
284 } 225 }
285 } <<
286 226
287 /* !! 227 if (xfs_ag_has_sickness(pag, mask)) {
288 * Quick scan to double-check that there isn't !! 228 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
289 * primary health problems. If we're still cl !! 229 return false;
290 * take care of clearing the indirect evidence <<
291 */ <<
292 int <<
293 xchk_health_record( <<
294 struct xfs_scrub *sc) <<
295 { <<
296 struct xfs_mount *mp = sc->mp; <<
297 struct xfs_perag *pag; <<
298 xfs_agnumber_t agno; <<
299 <<
300 unsigned int sick; <<
301 unsigned int checked; <<
302 <<
303 xfs_fs_measure_sickness(mp, &sick, &ch <<
304 if (sick & XFS_SICK_FS_PRIMARY) <<
305 xchk_set_corrupt(sc); <<
306 <<
307 xfs_rt_measure_sickness(mp, &sick, &ch <<
308 if (sick & XFS_SICK_RT_PRIMARY) <<
309 xchk_set_corrupt(sc); <<
310 <<
311 for_each_perag(mp, agno, pag) { <<
312 xfs_ag_measure_sickness(pag, & <<
313 if (sick & XFS_SICK_AG_PRIMARY <<
314 xchk_set_corrupt(sc); <<
315 } 230 }
316 231
317 return 0; !! 232 return true;
318 } 233 }
319 234
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.