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

   1 #ifndef _BCACHE_ERROR_H
   2 #define _BCACHE_ERROR_H
   3
   4 #include <linux/printk.h>
   5
   6 struct cache;
   7 struct cache_set;
   8
   9 /*
  10  * XXX: separate out errors that indicate on disk data is inconsistent, and flag
  11  * superblock as such
  12  */
  13
  14 /* Error messages: */
  15
  16 #define __bch_dev_error(ca, fmt, ...)                                   \
  17 do {                                                                    \
  18         char _buf[BDEVNAME_SIZE];                                       \
  19         bch_err((ca)->set, "%s: " fmt,                                  \
  20                 bdevname((ca)->disk_sb.bdev, _buf), ##__VA_ARGS__);     \
  21 } while (0)
  22
  23 /*
  24  * Very fatal logic/inconsistency errors: these indicate that we've majorly
  25  * screwed up at runtime, i.e. it's not likely that it was just caused by the
  26  * data on disk being inconsistent. These BUG():
  27  *
  28  * XXX: audit and convert to inconsistent() checks
  29  */
  30
  31 #define bch_fs_bug(c, ...)                                              \
  32 do {                                                                    \
  33         bch_err(c, __VA_ARGS__);                                        \
  34         BUG();                                                          \
  35 } while (0)
  36
  37 #define bch_fs_bug_on(cond, c, ...)                                     \
  38 do {                                                                    \
  39         if (cond)                                                       \
  40                 bch_fs_bug(c, __VA_ARGS__);                             \
  41 } while (0)
  42
  43 /*
  44  * Inconsistency errors: The on disk data is inconsistent. If these occur during
  45  * initial recovery, they don't indicate a bug in the running code - we walk all
  46  * the metadata before modifying anything. If they occur at runtime, they
  47  * indicate either a bug in the running code or (less likely) data is being
  48  * silently corrupted under us.
  49  *
  50  * XXX: audit all inconsistent errors and make sure they're all recoverable, in
  51  * BCH_ON_ERROR_CONTINUE mode
  52  */
  53
  54 void bch_inconsistent_error(struct cache_set *);
  55
  56 #define bch_fs_inconsistent(c, ...)                                     \
  57 do {                                                                    \
  58         bch_err(c, __VA_ARGS__);                                        \
  59         bch_inconsistent_error(c);                                      \
  60 } while (0)
  61
  62 #define bch_fs_inconsistent_on(cond, c, ...)                            \
  63 ({                                                                      \
  64         int _ret = !!(cond);                                            \
  65                                                                         \
  66         if (_ret)                                                       \
  67                 bch_fs_inconsistent(c, __VA_ARGS__);                    \
  68         _ret;                                                           \
  69 })
  70
  71 /*
  72  * Later we might want to mark only the particular device inconsistent, not the
  73  * entire cache set:
  74  */
  75
  76 #define bch_dev_inconsistent(ca, ...)                                   \
  77 do {                                                                    \
  78         __bch_dev_error(ca, __VA_ARGS__);                               \
  79         bch_inconsistent_error((ca)->set);                              \
  80 } while (0)
  81
  82 #define bch_dev_inconsistent_on(cond, ca, ...)                          \
  83 ({                                                                      \
  84         int _ret = !!(cond);                                            \
  85                                                                         \
  86         if (_ret)                                                       \
  87                 bch_dev_inconsistent(ca, __VA_ARGS__);                  \
  88         _ret;                                                           \
  89 })
  90
  91 /*
  92  * Fsck errors: inconsistency errors we detect at mount time, and should ideally
  93  * be able to repair:
  94  */
  95
  96 enum {
  97         BCH_FSCK_OK                     = 0,
  98         BCH_FSCK_ERRORS_NOT_FIXED       = 1,
  99         BCH_FSCK_REPAIR_UNIMPLEMENTED   = 2,
 100         BCH_FSCK_REPAIR_IMPOSSIBLE      = 3,
 101         BCH_FSCK_UNKNOWN_VERSION        = 4,
 102 };
 103
 104 /* These macros return true if error should be fixed: */
 105
 106 /* XXX: mark in superblock that filesystem contains errors, if we ignore: */
 107
 108 #ifndef __fsck_err
 109 #define __fsck_err(c, _can_fix, _can_ignore, _nofix_msg, msg, ...)      \
 110 ({                                                                      \
 111         bool _fix = false;                                              \
 112                                                                         \
 113         if (_can_fix && (c)->opts.fix_errors) {                         \
 114                 bch_err(c, msg ", fixing", ##__VA_ARGS__);              \
 115                 set_bit(BCH_FS_FSCK_FIXED_ERRORS, &(c)->flags); \
 116                 _fix = true;                                            \
 117         } else if (_can_ignore &&                                       \
 118                    (c)->opts.errors == BCH_ON_ERROR_CONTINUE) {         \
 119                 bch_err(c, msg " (ignoring)", ##__VA_ARGS__);           \
 120         } else {                                                        \
 121                 bch_err(c, msg " ("_nofix_msg")", ##__VA_ARGS__);       \
 122                 ret = BCH_FSCK_ERRORS_NOT_FIXED;                        \
 123                 goto fsck_err;                                          \
 124         }                                                               \
 125                                                                         \
 126         BUG_ON(!_fix && !_can_ignore);                                  \
 127         _fix;                                                           \
 128 })
 129 #endif
 130
 131 #define __fsck_err_on(cond, c, _can_fix, _can_ignore, _nofix_msg, ...)  \
 132         ((cond) ? __fsck_err(c, _can_fix, _can_ignore,                  \
 133                              _nofix_msg, ##__VA_ARGS__) : false)
 134
 135 #define unfixable_fsck_err_on(cond, c, ...)                             \
 136         __fsck_err_on(cond, c, false, true, "repair unimplemented", ##__VA_ARGS__)
 137
 138 #define need_fsck_err_on(cond, c, ...)                                  \
 139         __fsck_err_on(cond, c, false, true, "run fsck to correct", ##__VA_ARGS__)
 140
 141 #define mustfix_fsck_err(c, ...)                                        \
 142         __fsck_err(c, true, false, "not fixing", ##__VA_ARGS__)
 143
 144 #define mustfix_fsck_err_on(cond, c, ...)                               \
 145         __fsck_err_on(cond, c, true, false, "not fixing", ##__VA_ARGS__)
 146
 147 #define fsck_err_on(cond, c, ...)                                       \
 148         __fsck_err_on(cond, c, true, true, "not fixing", ##__VA_ARGS__)
 149
 150 /*
 151  * Fatal errors: these don't indicate a bug, but we can't continue running in RW
 152  * mode - pretty much just due to metadata IO errors:
 153  */
 154
 155 void bch_fatal_error(struct cache_set *);
 156
 157 #define bch_fs_fatal_error(c, ...)                                      \
 158 do {                                                                    \
 159         bch_err(c, __VA_ARGS__);                                        \
 160         bch_fatal_error(c);                                             \
 161 } while (0)
 162
 163 #define bch_fs_fatal_err_on(cond, c, ...)                               \
 164 ({                                                                      \
 165         int _ret = !!(cond);                                            \
 166                                                                         \
 167         if (_ret)                                                       \
 168                 bch_fs_fatal_error(c, __VA_ARGS__);                     \
 169         _ret;                                                           \
 170 })
 171
 172 #define bch_dev_fatal_error(ca, ...)                                    \
 173 do {                                                                    \
 174         __bch_dev_error(ca, __VA_ARGS__);                               \
 175         bch_fatal_error(c);                                             \
 176 } while (0)
 177
 178 #define bch_dev_fatal_io_error(ca, fmt, ...)                            \
 179 do {                                                                    \
 180         char _buf[BDEVNAME_SIZE];                                       \
 181                                                                         \
 182         printk_ratelimited(KERN_ERR bch_fmt((ca)->set,                  \
 183                 "fatal IO error on %s for " fmt),                       \
 184                 bdevname((ca)->disk_sb.bdev, _buf), ##__VA_ARGS__);     \
 185         bch_fatal_error((ca)->set);                                     \
 186 } while (0)
 187
 188 #define bch_dev_fatal_io_err_on(cond, ca, ...)                          \
 189 ({                                                                      \
 190         int _ret = !!(cond);                                            \
 191                                                                         \
 192         if (_ret)                                                       \
 193                 bch_dev_fatal_io_error(ca, __VA_ARGS__);                \
 194         _ret;                                                           \
 195 })
 196
 197 /*
 198  * Nonfatal IO errors: either recoverable metadata IO (because we have
 199  * replicas), or data IO - we need to log it and print out a message, but we
 200  * don't (necessarily) want to shut down the fs:
 201  */
 202
 203 void bch_account_io_completion(struct cache *);
 204 void bch_account_io_completion_time(struct cache *, unsigned, int);
 205
 206 void bch_nonfatal_io_error_work(struct work_struct *);
 207
 208 /* Does the error handling without logging a message */
 209 void bch_nonfatal_io_error(struct cache *);
 210
 211 #if 0
 212 #define bch_fs_nonfatal_io_error(c, ...)                                \
 213 do {                                                                    \
 214         bch_err(c, __VA_ARGS__);                                        \
 215         bch_nonfatal_io_error(c);                                       \
 216 } while (0)
 217 #endif
 218
 219 /* Logs message and handles the error: */
 220 #define bch_dev_nonfatal_io_error(ca, fmt, ...)                         \
 221 do {                                                                    \
 222         char _buf[BDEVNAME_SIZE];                                       \
 223                                                                         \
 224         printk_ratelimited(KERN_ERR bch_fmt((ca)->set,                  \
 225                 "IO error on %s for " fmt),                             \
 226                 bdevname((ca)->disk_sb.bdev, _buf), ##__VA_ARGS__);     \
 227         bch_nonfatal_io_error(ca);                                      \
 228 } while (0)
 229
 230 #define bch_dev_nonfatal_io_err_on(cond, ca, ...)                       \
 231 ({                                                                      \
 232         bool _ret = (cond);                                             \
 233                                                                         \
 234         if (_ret)                                                       \
 235                 bch_dev_nonfatal_io_error(ca, __VA_ARGS__);             \
 236         _ret;                                                           \
 237 })
 238
 239 /* kill? */
 240
 241 #define __bcache_io_error(c, fmt, ...)                                  \
 242         printk_ratelimited(KERN_ERR bch_fmt(c,                          \
 243                         "IO error: " fmt), ##__VA_ARGS__)
 244
 245 #define bcache_io_error(c, bio, fmt, ...)                               \
 246 do {                                                                    \
 247         __bcache_io_error(c, fmt, ##__VA_ARGS__);                       \
 248         (bio)->bi_error = -EIO;                                         \
 249 } while (0)
 250
 251 #endif /* _BCACHE_ERROR_H */