xref: /linux/fs/bcachefs/error.h (revision a3a02a52bcfcbcc4a637d4b68bf1bc391c9fad02)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_ERROR_H
3 #define _BCACHEFS_ERROR_H
4 
5 #include <linux/list.h>
6 #include <linux/printk.h>
7 #include "sb-errors.h"
8 
9 struct bch_dev;
10 struct bch_fs;
11 struct work_struct;
12 
13 /*
14  * XXX: separate out errors that indicate on disk data is inconsistent, and flag
15  * superblock as such
16  */
17 
18 /* Error messages: */
19 
20 /*
21  * Inconsistency errors: The on disk data is inconsistent. If these occur during
22  * initial recovery, they don't indicate a bug in the running code - we walk all
23  * the metadata before modifying anything. If they occur at runtime, they
24  * indicate either a bug in the running code or (less likely) data is being
25  * silently corrupted under us.
26  *
27  * XXX: audit all inconsistent errors and make sure they're all recoverable, in
28  * BCH_ON_ERROR_CONTINUE mode
29  */
30 
31 bool bch2_inconsistent_error(struct bch_fs *);
32 
33 int bch2_topology_error(struct bch_fs *);
34 
35 #define bch2_fs_topology_error(c, ...)					\
36 ({									\
37 	bch_err(c, "btree topology error: " __VA_ARGS__);		\
38 	bch2_topology_error(c);						\
39 })
40 
41 #define bch2_fs_inconsistent(c, ...)					\
42 ({									\
43 	bch_err(c, __VA_ARGS__);					\
44 	bch2_inconsistent_error(c);					\
45 })
46 
47 #define bch2_fs_inconsistent_on(cond, c, ...)				\
48 ({									\
49 	bool _ret = unlikely(!!(cond));					\
50 									\
51 	if (_ret)							\
52 		bch2_fs_inconsistent(c, __VA_ARGS__);			\
53 	_ret;								\
54 })
55 
56 /*
57  * Later we might want to mark only the particular device inconsistent, not the
58  * entire filesystem:
59  */
60 
61 #define bch2_dev_inconsistent(ca, ...)					\
62 do {									\
63 	bch_err(ca, __VA_ARGS__);					\
64 	bch2_inconsistent_error((ca)->fs);				\
65 } while (0)
66 
67 #define bch2_dev_inconsistent_on(cond, ca, ...)				\
68 ({									\
69 	bool _ret = unlikely(!!(cond));					\
70 									\
71 	if (_ret)							\
72 		bch2_dev_inconsistent(ca, __VA_ARGS__);			\
73 	_ret;								\
74 })
75 
76 /*
77  * When a transaction update discovers or is causing a fs inconsistency, it's
78  * helpful to also dump the pending updates:
79  */
80 #define bch2_trans_inconsistent(trans, ...)				\
81 ({									\
82 	bch_err(trans->c, __VA_ARGS__);					\
83 	bch2_dump_trans_updates(trans);					\
84 	bch2_inconsistent_error(trans->c);				\
85 })
86 
87 #define bch2_trans_inconsistent_on(cond, trans, ...)			\
88 ({									\
89 	bool _ret = unlikely(!!(cond));					\
90 									\
91 	if (_ret)							\
92 		bch2_trans_inconsistent(trans, __VA_ARGS__);		\
93 	_ret;								\
94 })
95 
96 /*
97  * Fsck errors: inconsistency errors we detect at mount time, and should ideally
98  * be able to repair:
99  */
100 
101 struct fsck_err_state {
102 	struct list_head	list;
103 	const char		*fmt;
104 	u64			nr;
105 	bool			ratelimited;
106 	int			ret;
107 	int			fix;
108 	char			*last_msg;
109 };
110 
111 #define fsck_err_count(_c, _err)	bch2_sb_err_count(_c, BCH_FSCK_ERR_##_err)
112 
113 __printf(5, 6) __cold
114 int __bch2_fsck_err(struct bch_fs *, struct btree_trans *,
115 		  enum bch_fsck_flags,
116 		  enum bch_sb_error_id,
117 		  const char *, ...);
118 #define bch2_fsck_err(c, _flags, _err_type, ...)				\
119 	__bch2_fsck_err(type_is(c, struct bch_fs *) ? (struct bch_fs *) c : NULL,\
120 			type_is(c, struct btree_trans *) ? (struct btree_trans *) c : NULL,\
121 			_flags, BCH_FSCK_ERR_##_err_type, __VA_ARGS__)
122 
123 void bch2_flush_fsck_errs(struct bch_fs *);
124 
125 #define __fsck_err(c, _flags, _err_type, ...)				\
126 ({									\
127 	int _ret = bch2_fsck_err(c, _flags, _err_type, __VA_ARGS__);	\
128 	if (_ret != -BCH_ERR_fsck_fix &&				\
129 	    _ret != -BCH_ERR_fsck_ignore) {				\
130 		ret = _ret;						\
131 		goto fsck_err;						\
132 	}								\
133 									\
134 	_ret == -BCH_ERR_fsck_fix;					\
135 })
136 
137 /* These macros return true if error should be fixed: */
138 
139 /* XXX: mark in superblock that filesystem contains errors, if we ignore: */
140 
141 #define __fsck_err_on(cond, c, _flags, _err_type, ...)			\
142 ({									\
143 	might_sleep();							\
144 									\
145 	if (type_is(c, struct bch_fs *))				\
146 		WARN_ON(bch2_current_has_btree_trans((struct bch_fs *) c));\
147 									\
148 	(unlikely(cond) ? __fsck_err(c, _flags, _err_type, __VA_ARGS__) : false);\
149 })
150 
151 #define need_fsck_err_on(cond, c, _err_type, ...)				\
152 	__fsck_err_on(cond, c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, _err_type, __VA_ARGS__)
153 
154 #define need_fsck_err(c, _err_type, ...)				\
155 	__fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, _err_type, __VA_ARGS__)
156 
157 #define mustfix_fsck_err(c, _err_type, ...)				\
158 	__fsck_err(c, FSCK_CAN_FIX, _err_type, __VA_ARGS__)
159 
160 #define mustfix_fsck_err_on(cond, c, _err_type, ...)			\
161 	__fsck_err_on(cond, c, FSCK_CAN_FIX, _err_type, __VA_ARGS__)
162 
163 #define fsck_err(c, _err_type, ...)					\
164 	__fsck_err(c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, _err_type, __VA_ARGS__)
165 
166 #define fsck_err_on(cond, c, _err_type, ...)				\
167 	__fsck_err_on(cond, c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, _err_type, __VA_ARGS__)
168 
169 __printf(4, 0)
170 static inline void bch2_bkey_fsck_err(struct bch_fs *c,
171 				     struct printbuf *err_msg,
172 				     enum bch_sb_error_id err_type,
173 				     const char *fmt, ...)
174 {
175 	va_list args;
176 
177 	va_start(args, fmt);
178 	prt_vprintf(err_msg, fmt, args);
179 	va_end(args);
180 }
181 
182 #define bkey_fsck_err(c, _err_msg, _err_type, ...)			\
183 do {									\
184 	prt_printf(_err_msg, __VA_ARGS__);				\
185 	bch2_sb_error_count(c, BCH_FSCK_ERR_##_err_type);		\
186 	ret = -BCH_ERR_invalid_bkey;					\
187 	goto fsck_err;							\
188 } while (0)
189 
190 #define bkey_fsck_err_on(cond, ...)					\
191 do {									\
192 	if (unlikely(cond))						\
193 		bkey_fsck_err(__VA_ARGS__);				\
194 } while (0)
195 
196 /*
197  * Fatal errors: these don't indicate a bug, but we can't continue running in RW
198  * mode - pretty much just due to metadata IO errors:
199  */
200 
201 void bch2_fatal_error(struct bch_fs *);
202 
203 #define bch2_fs_fatal_error(c, _msg, ...)				\
204 do {									\
205 	bch_err(c, "%s(): fatal error " _msg, __func__, ##__VA_ARGS__);	\
206 	bch2_fatal_error(c);						\
207 } while (0)
208 
209 #define bch2_fs_fatal_err_on(cond, c, ...)				\
210 ({									\
211 	bool _ret = unlikely(!!(cond));					\
212 									\
213 	if (_ret)							\
214 		bch2_fs_fatal_error(c, __VA_ARGS__);			\
215 	_ret;								\
216 })
217 
218 /*
219  * IO errors: either recoverable metadata IO (because we have replicas), or data
220  * IO - we need to log it and print out a message, but we don't (necessarily)
221  * want to shut down the fs:
222  */
223 
224 void bch2_io_error_work(struct work_struct *);
225 
226 /* Does the error handling without logging a message */
227 void bch2_io_error(struct bch_dev *, enum bch_member_error_type);
228 
229 #define bch2_dev_io_err_on(cond, ca, _type, ...)			\
230 ({									\
231 	bool _ret = (cond);						\
232 									\
233 	if (_ret) {							\
234 		bch_err_dev_ratelimited(ca, __VA_ARGS__);		\
235 		bch2_io_error(ca, _type);				\
236 	}								\
237 	_ret;								\
238 })
239 
240 #define bch2_dev_inum_io_err_on(cond, ca, _type, ...)			\
241 ({									\
242 	bool _ret = (cond);						\
243 									\
244 	if (_ret) {							\
245 		bch_err_inum_offset_ratelimited(ca, __VA_ARGS__);	\
246 		bch2_io_error(ca, _type);				\
247 	}								\
248 	_ret;								\
249 })
250 
251 #endif /* _BCACHEFS_ERROR_H */
252