xref: /linux/fs/xfs/scrub/health.c (revision d9041681dd2f5334529a68868c9266631c384de4)
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
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  * If the scrub state is clean, add @mask to the scrub sick mask to clear
122  * additional sick flags from the metadata object's sick state.
123  */
124 void
125 xchk_mark_healthy_if_clean(
126 	struct xfs_scrub	*sc,
127 	unsigned int		mask)
128 {
129 	if (!(sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
130 				  XFS_SCRUB_OFLAG_XCORRUPT)))
131 		sc->sick_mask |= mask;
132 }
133 
134 /*
135  * If we're scrubbing a piece of file metadata for the first time, does it look
136  * like it has been zapped?  Skip the check if we just repaired the metadata
137  * and are revalidating it.
138  */
139 bool
140 xchk_file_looks_zapped(
141 	struct xfs_scrub	*sc,
142 	unsigned int		mask)
143 {
144 	ASSERT((mask & ~XFS_SICK_INO_ZAPPED) == 0);
145 
146 	if (sc->flags & XREP_ALREADY_FIXED)
147 		return false;
148 
149 	return xfs_inode_has_sickness(sc->ip, mask);
150 }
151 
152 /*
153  * Update filesystem health assessments based on what we found and did.
154  *
155  * If the scrubber finds errors, we mark sick whatever's mentioned in
156  * sick_mask, no matter whether this is a first scan or an
157  * evaluation of repair effectiveness.
158  *
159  * Otherwise, no direct corruption was found, so mark whatever's in
160  * sick_mask as healthy.
161  */
162 void
163 xchk_update_health(
164 	struct xfs_scrub	*sc)
165 {
166 	struct xfs_perag	*pag;
167 	bool			bad;
168 
169 	if (!sc->sick_mask)
170 		return;
171 
172 	bad = (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
173 				   XFS_SCRUB_OFLAG_XCORRUPT));
174 	switch (type_to_health_flag[sc->sm->sm_type].group) {
175 	case XHG_AG:
176 		pag = xfs_perag_get(sc->mp, sc->sm->sm_agno);
177 		if (bad)
178 			xfs_ag_mark_sick(pag, sc->sick_mask);
179 		else
180 			xfs_ag_mark_healthy(pag, sc->sick_mask);
181 		xfs_perag_put(pag);
182 		break;
183 	case XHG_INO:
184 		if (!sc->ip)
185 			return;
186 		if (bad)
187 			xfs_inode_mark_sick(sc->ip, sc->sick_mask);
188 		else
189 			xfs_inode_mark_healthy(sc->ip, sc->sick_mask);
190 		break;
191 	case XHG_FS:
192 		if (bad)
193 			xfs_fs_mark_sick(sc->mp, sc->sick_mask);
194 		else
195 			xfs_fs_mark_healthy(sc->mp, sc->sick_mask);
196 		break;
197 	case XHG_RT:
198 		if (bad)
199 			xfs_rt_mark_sick(sc->mp, sc->sick_mask);
200 		else
201 			xfs_rt_mark_healthy(sc->mp, sc->sick_mask);
202 		break;
203 	default:
204 		ASSERT(0);
205 		break;
206 	}
207 }
208 
209 /* Is the given per-AG btree healthy enough for scanning? */
210 bool
211 xchk_ag_btree_healthy_enough(
212 	struct xfs_scrub	*sc,
213 	struct xfs_perag	*pag,
214 	xfs_btnum_t		btnum)
215 {
216 	unsigned int		mask = 0;
217 
218 	/*
219 	 * We always want the cursor if it's the same type as whatever we're
220 	 * scrubbing, even if we already know the structure is corrupt.
221 	 *
222 	 * Otherwise, we're only interested in the btree for cross-referencing.
223 	 * If we know the btree is bad then don't bother, just set XFAIL.
224 	 */
225 	switch (btnum) {
226 	case XFS_BTNUM_BNO:
227 		if (sc->sm->sm_type == XFS_SCRUB_TYPE_BNOBT)
228 			return true;
229 		mask = XFS_SICK_AG_BNOBT;
230 		break;
231 	case XFS_BTNUM_CNT:
232 		if (sc->sm->sm_type == XFS_SCRUB_TYPE_CNTBT)
233 			return true;
234 		mask = XFS_SICK_AG_CNTBT;
235 		break;
236 	case XFS_BTNUM_INO:
237 		if (sc->sm->sm_type == XFS_SCRUB_TYPE_INOBT)
238 			return true;
239 		mask = XFS_SICK_AG_INOBT;
240 		break;
241 	case XFS_BTNUM_FINO:
242 		if (sc->sm->sm_type == XFS_SCRUB_TYPE_FINOBT)
243 			return true;
244 		mask = XFS_SICK_AG_FINOBT;
245 		break;
246 	case XFS_BTNUM_RMAP:
247 		if (sc->sm->sm_type == XFS_SCRUB_TYPE_RMAPBT)
248 			return true;
249 		mask = XFS_SICK_AG_RMAPBT;
250 		break;
251 	case XFS_BTNUM_REFC:
252 		if (sc->sm->sm_type == XFS_SCRUB_TYPE_REFCNTBT)
253 			return true;
254 		mask = XFS_SICK_AG_REFCNTBT;
255 		break;
256 	default:
257 		ASSERT(0);
258 		return true;
259 	}
260 
261 	/*
262 	 * If we just repaired some AG metadata, sc->sick_mask will reflect all
263 	 * the per-AG metadata types that were repaired.  Exclude these from
264 	 * the filesystem health query because we have not yet updated the
265 	 * health status and we want everything to be scanned.
266 	 */
267 	if ((sc->flags & XREP_ALREADY_FIXED) &&
268 	    type_to_health_flag[sc->sm->sm_type].group == XHG_AG)
269 		mask &= ~sc->sick_mask;
270 
271 	if (xfs_ag_has_sickness(pag, mask)) {
272 		sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
273 		return false;
274 	}
275 
276 	return true;
277 }
278