git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/error.h

   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
   8 struct bch_dev;
   9 struct bch_fs;
  10 struct work_struct;
  11
  12 /*
  13  * XXX: separate out errors that indicate on disk data is inconsistent, and flag
  14  * superblock as such
  15  */
  16
  17 /* Error messages: */
  18
  19 /*
  20  * Inconsistency errors: The on disk data is inconsistent. If these occur during
  21  * initial recovery, they don't indicate a bug in the running code - we walk all
  22  * the metadata before modifying anything. If they occur at runtime, they
  23  * indicate either a bug in the running code or (less likely) data is being
  24  * silently corrupted under us.
  25  *
  26  * XXX: audit all inconsistent errors and make sure they're all recoverable, in
  27  * BCH_ON_ERROR_CONTINUE mode
  28  */
  29
  30 bool bch2_inconsistent_error(struct bch_fs *);
  31
  32 void bch2_topology_error(struct bch_fs *);
  33
  34 #define bch2_fs_inconsistent(c, ...)                                    \
  35 ({                                                                      \
  36         bch_err(c, __VA_ARGS__);                                        \
  37         bch2_inconsistent_error(c);                                     \
  38 })
  39
  40 #define bch2_fs_inconsistent_on(cond, c, ...)                           \
  41 ({                                                                      \
  42         bool _ret = unlikely(!!(cond));                                 \
  43                                                                         \
  44         if (_ret)                                                       \
  45                 bch2_fs_inconsistent(c, __VA_ARGS__);                   \
  46         _ret;                                                           \
  47 })
  48
  49 /*
  50  * Later we might want to mark only the particular device inconsistent, not the
  51  * entire filesystem:
  52  */
  53
  54 #define bch2_dev_inconsistent(ca, ...)                                  \
  55 do {                                                                    \
  56         bch_err(ca, __VA_ARGS__);                                       \
  57         bch2_inconsistent_error((ca)->fs);                              \
  58 } while (0)
  59
  60 #define bch2_dev_inconsistent_on(cond, ca, ...)                         \
  61 ({                                                                      \
  62         bool _ret = unlikely(!!(cond));                                 \
  63                                                                         \
  64         if (_ret)                                                       \
  65                 bch2_dev_inconsistent(ca, __VA_ARGS__);                 \
  66         _ret;                                                           \
  67 })
  68
  69 /*
  70  * When a transaction update discovers or is causing a fs inconsistency, it's
  71  * helpful to also dump the pending updates:
  72  */
  73 #define bch2_trans_inconsistent(trans, ...)                             \
  74 ({                                                                      \
  75         bch_err(trans->c, __VA_ARGS__);                                 \
  76         bch2_dump_trans_updates(trans);                                 \
  77         bch2_inconsistent_error(trans->c);                              \
  78 })
  79
  80 #define bch2_trans_inconsistent_on(cond, trans, ...)                    \
  81 ({                                                                      \
  82         bool _ret = unlikely(!!(cond));                                 \
  83                                                                         \
  84         if (_ret)                                                       \
  85                 bch2_trans_inconsistent(trans, __VA_ARGS__);            \
  86         _ret;                                                           \
  87 })
  88
  89 /*
  90  * Fsck errors: inconsistency errors we detect at mount time, and should ideally
  91  * be able to repair:
  92  */
  93
  94 struct fsck_err_state {
  95         struct list_head        list;
  96         const char              *fmt;
  97         u64                     nr;
  98         bool                    ratelimited;
  99         int                     ret;
 100         int                     fix;
 101         char                    *last_msg;
 102 };
 103
 104 #define FSCK_CAN_FIX            (1 << 0)
 105 #define FSCK_CAN_IGNORE         (1 << 1)
 106 #define FSCK_NEED_FSCK          (1 << 2)
 107 #define FSCK_NO_RATELIMIT       (1 << 3)
 108
 109 __printf(3, 4) __cold
 110 int bch2_fsck_err(struct bch_fs *, unsigned, const char *, ...);
 111 void bch2_flush_fsck_errs(struct bch_fs *);
 112
 113 #define __fsck_err(c, _flags, msg, ...)                                 \
 114 ({                                                                      \
 115         int _ret = bch2_fsck_err(c, _flags, msg, ##__VA_ARGS__);        \
 116                                                                         \
 117         if (_ret != -BCH_ERR_fsck_fix &&                                \
 118             _ret != -BCH_ERR_fsck_ignore) {                             \
 119                 ret = _ret;                                             \
 120                 goto fsck_err;                                          \
 121         }                                                               \
 122                                                                         \
 123         _ret == -BCH_ERR_fsck_fix;                                      \
 124 })
 125
 126 /* These macros return true if error should be fixed: */
 127
 128 /* XXX: mark in superblock that filesystem contains errors, if we ignore: */
 129
 130 #define __fsck_err_on(cond, c, _flags, ...)                             \
 131         (unlikely(cond) ? __fsck_err(c, _flags, ##__VA_ARGS__) : false)
 132
 133 #define need_fsck_err_on(cond, c, ...)                                  \
 134         __fsck_err_on(cond, c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, ##__VA_ARGS__)
 135
 136 #define need_fsck_err(c, ...)                                           \
 137         __fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, ##__VA_ARGS__)
 138
 139 #define mustfix_fsck_err(c, ...)                                        \
 140         __fsck_err(c, FSCK_CAN_FIX, ##__VA_ARGS__)
 141
 142 #define mustfix_fsck_err_on(cond, c, ...)                               \
 143         __fsck_err_on(cond, c, FSCK_CAN_FIX, ##__VA_ARGS__)
 144
 145 #define fsck_err(c, ...)                                                \
 146         __fsck_err(c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, ##__VA_ARGS__)
 147
 148 #define fsck_err_on(cond, c, ...)                                       \
 149         __fsck_err_on(cond, c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, ##__VA_ARGS__)
 150
 151 /*
 152  * Fatal errors: these don't indicate a bug, but we can't continue running in RW
 153  * mode - pretty much just due to metadata IO errors:
 154  */
 155
 156 void bch2_fatal_error(struct bch_fs *);
 157
 158 #define bch2_fs_fatal_error(c, ...)                                     \
 159 do {                                                                    \
 160         bch_err(c, __VA_ARGS__);                                        \
 161         bch2_fatal_error(c);                                            \
 162 } while (0)
 163
 164 #define bch2_fs_fatal_err_on(cond, c, ...)                              \
 165 ({                                                                      \
 166         bool _ret = unlikely(!!(cond));                                 \
 167                                                                         \
 168         if (_ret)                                                       \
 169                 bch2_fs_fatal_error(c, __VA_ARGS__);                    \
 170         _ret;                                                           \
 171 })
 172
 173 /*
 174  * IO errors: either recoverable metadata IO (because we have replicas), or data
 175  * IO - we need to log it and print out a message, but we don't (necessarily)
 176  * want to shut down the fs:
 177  */
 178
 179 void bch2_io_error_work(struct work_struct *);
 180
 181 /* Does the error handling without logging a message */
 182 void bch2_io_error(struct bch_dev *);
 183
 184 #define bch2_dev_io_err_on(cond, ca, ...)                               \
 185 ({                                                                      \
 186         bool _ret = (cond);                                             \
 187                                                                         \
 188         if (_ret) {                                                     \
 189                 bch_err_dev_ratelimited(ca, __VA_ARGS__);               \
 190                 bch2_io_error(ca);                                      \
 191         }                                                               \
 192         _ret;                                                           \
 193 })
 194
 195 #define bch2_dev_inum_io_err_on(cond, ca, ...)                          \
 196 ({                                                                      \
 197         bool _ret = (cond);                                             \
 198                                                                         \
 199         if (_ret) {                                                     \
 200                 bch_err_inum_offset_ratelimited(ca, __VA_ARGS__);       \
 201                 bch2_io_error(ca);                                      \
 202         }                                                               \
 203         _ret;                                                           \
 204 })
 205
 206 #endif /* _BCACHEFS_ERROR_H */