xref: /linux/fs/bcachefs/error.h (revision 906fd46a65383cd639e5eec72a047efc33045d86)
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(4, 5) __cold
114 int bch2_fsck_err(struct bch_fs *,
115 		  enum bch_fsck_flags,
116 		  enum bch_sb_error_id,
117 		  const char *, ...);
118 void bch2_flush_fsck_errs(struct bch_fs *);
119 
120 #define __fsck_err(c, _flags, _err_type, ...)				\
121 ({									\
122 	int _ret = bch2_fsck_err(c, _flags, BCH_FSCK_ERR_##_err_type,	\
123 				 __VA_ARGS__);				\
124 									\
125 	if (_ret != -BCH_ERR_fsck_fix &&				\
126 	    _ret != -BCH_ERR_fsck_ignore) {				\
127 		ret = _ret;						\
128 		goto fsck_err;						\
129 	}								\
130 									\
131 	_ret == -BCH_ERR_fsck_fix;					\
132 })
133 
134 /* These macros return true if error should be fixed: */
135 
136 /* XXX: mark in superblock that filesystem contains errors, if we ignore: */
137 
138 #define __fsck_err_on(cond, c, _flags, _err_type, ...)			\
139 	(unlikely(cond) ? __fsck_err(c, _flags, _err_type, __VA_ARGS__) : false)
140 
141 #define need_fsck_err_on(cond, c, _err_type, ...)				\
142 	__fsck_err_on(cond, c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, _err_type, __VA_ARGS__)
143 
144 #define need_fsck_err(c, _err_type, ...)				\
145 	__fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, _err_type, __VA_ARGS__)
146 
147 #define mustfix_fsck_err(c, _err_type, ...)				\
148 	__fsck_err(c, FSCK_CAN_FIX, _err_type, __VA_ARGS__)
149 
150 #define mustfix_fsck_err_on(cond, c, _err_type, ...)			\
151 	__fsck_err_on(cond, c, FSCK_CAN_FIX, _err_type, __VA_ARGS__)
152 
153 #define fsck_err(c, _err_type, ...)					\
154 	__fsck_err(c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, _err_type, __VA_ARGS__)
155 
156 #define fsck_err_on(cond, c, _err_type, ...)				\
157 	__fsck_err_on(cond, c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, _err_type, __VA_ARGS__)
158 
159 __printf(4, 0)
160 static inline void bch2_bkey_fsck_err(struct bch_fs *c,
161 				     struct printbuf *err_msg,
162 				     enum bch_sb_error_id err_type,
163 				     const char *fmt, ...)
164 {
165 	va_list args;
166 
167 	va_start(args, fmt);
168 	prt_vprintf(err_msg, fmt, args);
169 	va_end(args);
170 }
171 
172 #define bkey_fsck_err(c, _err_msg, _err_type, ...)			\
173 do {									\
174 	prt_printf(_err_msg, __VA_ARGS__);				\
175 	bch2_sb_error_count(c, BCH_FSCK_ERR_##_err_type);		\
176 	ret = -BCH_ERR_invalid_bkey;					\
177 	goto fsck_err;							\
178 } while (0)
179 
180 #define bkey_fsck_err_on(cond, ...)					\
181 do {									\
182 	if (unlikely(cond))						\
183 		bkey_fsck_err(__VA_ARGS__);				\
184 } while (0)
185 
186 /*
187  * Fatal errors: these don't indicate a bug, but we can't continue running in RW
188  * mode - pretty much just due to metadata IO errors:
189  */
190 
191 void bch2_fatal_error(struct bch_fs *);
192 
193 #define bch2_fs_fatal_error(c, _msg, ...)				\
194 do {									\
195 	bch_err(c, "%s(): fatal error " _msg, __func__, ##__VA_ARGS__);	\
196 	bch2_fatal_error(c);						\
197 } while (0)
198 
199 #define bch2_fs_fatal_err_on(cond, c, ...)				\
200 ({									\
201 	bool _ret = unlikely(!!(cond));					\
202 									\
203 	if (_ret)							\
204 		bch2_fs_fatal_error(c, __VA_ARGS__);			\
205 	_ret;								\
206 })
207 
208 /*
209  * IO errors: either recoverable metadata IO (because we have replicas), or data
210  * IO - we need to log it and print out a message, but we don't (necessarily)
211  * want to shut down the fs:
212  */
213 
214 void bch2_io_error_work(struct work_struct *);
215 
216 /* Does the error handling without logging a message */
217 void bch2_io_error(struct bch_dev *, enum bch_member_error_type);
218 
219 #define bch2_dev_io_err_on(cond, ca, _type, ...)			\
220 ({									\
221 	bool _ret = (cond);						\
222 									\
223 	if (_ret) {							\
224 		bch_err_dev_ratelimited(ca, __VA_ARGS__);		\
225 		bch2_io_error(ca, _type);				\
226 	}								\
227 	_ret;								\
228 })
229 
230 #define bch2_dev_inum_io_err_on(cond, ca, _type, ...)			\
231 ({									\
232 	bool _ret = (cond);						\
233 									\
234 	if (_ret) {							\
235 		bch_err_inum_offset_ratelimited(ca, __VA_ARGS__);	\
236 		bch2_io_error(ca, _type);				\
237 	}								\
238 	_ret;								\
239 })
240 
241 #endif /* _BCACHEFS_ERROR_H */
242