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