| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | |
| 3 | #include "fs.h" |
| 4 | #include "messages.h" |
| 5 | #include "discard.h" |
| 6 | #include "super.h" |
| 7 | |
| 8 | #ifdef CONFIG_PRINTK |
| 9 | |
| 10 | #define STATE_STRING_PREFACE " state " |
| 11 | #define STATE_STRING_BUF_LEN (sizeof(STATE_STRING_PREFACE) + BTRFS_FS_STATE_COUNT + 1) |
| 12 | |
| 13 | /* |
| 14 | * Characters to print to indicate error conditions or uncommon filesystem state. |
| 15 | * RO is not an error. |
| 16 | */ |
| 17 | static const char fs_state_chars[] = { |
| 18 | [BTRFS_FS_STATE_REMOUNTING] = 'M', |
| 19 | [BTRFS_FS_STATE_RO] = 0, |
| 20 | [BTRFS_FS_STATE_TRANS_ABORTED] = 'A', |
| 21 | [BTRFS_FS_STATE_DEV_REPLACING] = 'R', |
| 22 | [BTRFS_FS_STATE_DUMMY_FS_INFO] = 0, |
| 23 | [BTRFS_FS_STATE_NO_DATA_CSUMS] = 'C', |
| 24 | [BTRFS_FS_STATE_SKIP_META_CSUMS] = 'S', |
| 25 | [BTRFS_FS_STATE_LOG_CLEANUP_ERROR] = 'L', |
| 26 | }; |
| 27 | |
| 28 | static void btrfs_state_to_string(const struct btrfs_fs_info *info, char *buf) |
| 29 | { |
| 30 | unsigned int bit; |
| 31 | bool states_printed = false; |
| 32 | unsigned long fs_state = READ_ONCE(info->fs_state); |
| 33 | char *curr = buf; |
| 34 | |
| 35 | memcpy(curr, STATE_STRING_PREFACE, sizeof(STATE_STRING_PREFACE)); |
| 36 | curr += sizeof(STATE_STRING_PREFACE) - 1; |
| 37 | |
| 38 | if (BTRFS_FS_ERROR(info)) { |
| 39 | *curr++ = 'E'; |
| 40 | states_printed = true; |
| 41 | } |
| 42 | |
| 43 | for_each_set_bit(bit, &fs_state, sizeof(fs_state)) { |
| 44 | WARN_ON_ONCE(bit >= BTRFS_FS_STATE_COUNT); |
| 45 | if ((bit < BTRFS_FS_STATE_COUNT) && fs_state_chars[bit]) { |
| 46 | *curr++ = fs_state_chars[bit]; |
| 47 | states_printed = true; |
| 48 | } |
| 49 | } |
| 50 | |
| 51 | /* If no states were printed, reset the buffer */ |
| 52 | if (!states_printed) |
| 53 | curr = buf; |
| 54 | |
| 55 | *curr++ = 0; |
| 56 | } |
| 57 | #endif |
| 58 | |
| 59 | /* |
| 60 | * Generally the error codes correspond to their respective errors, but there |
| 61 | * are a few special cases. |
| 62 | * |
| 63 | * EUCLEAN: Any sort of corruption that we encounter. The tree-checker for |
| 64 | * instance will return EUCLEAN if any of the blocks are corrupted in |
| 65 | * a way that is problematic. We want to reserve EUCLEAN for these |
| 66 | * sort of corruptions. |
| 67 | * |
| 68 | * EROFS: If we check BTRFS_FS_STATE_ERROR and fail out with a return error, we |
| 69 | * need to use EROFS for this case. We will have no idea of the |
| 70 | * original failure, that will have been reported at the time we tripped |
| 71 | * over the error. Each subsequent error that doesn't have any context |
| 72 | * of the original error should use EROFS when handling BTRFS_FS_STATE_ERROR. |
| 73 | */ |
| 74 | const char * __attribute_const__ btrfs_decode_error(int error) |
| 75 | { |
| 76 | char *errstr = "unknown"; |
| 77 | |
| 78 | switch (error) { |
| 79 | case -ENOENT: /* -2 */ |
| 80 | errstr = "No such entry"; |
| 81 | break; |
| 82 | case -EIO: /* -5 */ |
| 83 | errstr = "IO failure"; |
| 84 | break; |
| 85 | case -ENOMEM: /* -12*/ |
| 86 | errstr = "Out of memory"; |
| 87 | break; |
| 88 | case -EEXIST: /* -17 */ |
| 89 | errstr = "Object already exists"; |
| 90 | break; |
| 91 | case -ENOSPC: /* -28 */ |
| 92 | errstr = "No space left"; |
| 93 | break; |
| 94 | case -EROFS: /* -30 */ |
| 95 | errstr = "Readonly filesystem"; |
| 96 | break; |
| 97 | case -EOPNOTSUPP: /* -95 */ |
| 98 | errstr = "Operation not supported"; |
| 99 | break; |
| 100 | case -EUCLEAN: /* -117 */ |
| 101 | errstr = "Filesystem corrupted"; |
| 102 | break; |
| 103 | case -EDQUOT: /* -122 */ |
| 104 | errstr = "Quota exceeded"; |
| 105 | break; |
| 106 | } |
| 107 | |
| 108 | return errstr; |
| 109 | } |
| 110 | |
| 111 | /* |
| 112 | * Decodes expected errors from the caller and invokes the appropriate error |
| 113 | * response. |
| 114 | */ |
| 115 | __cold |
| 116 | void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function, |
| 117 | unsigned int line, int error, const char *fmt, ...) |
| 118 | { |
| 119 | struct super_block *sb = fs_info->sb; |
| 120 | #ifdef CONFIG_PRINTK |
| 121 | char statestr[STATE_STRING_BUF_LEN]; |
| 122 | const char *errstr; |
| 123 | #endif |
| 124 | |
| 125 | #ifdef CONFIG_PRINTK_INDEX |
| 126 | printk_index_subsys_emit( |
| 127 | "BTRFS: error (device %s%s) in %s:%d: errno=%d %s", KERN_CRIT, fmt); |
| 128 | #endif |
| 129 | |
| 130 | /* |
| 131 | * Special case: if the error is EROFS, and we're already under |
| 132 | * SB_RDONLY, then it is safe here. |
| 133 | */ |
| 134 | if (error == -EROFS && sb_rdonly(sb)) |
| 135 | return; |
| 136 | |
| 137 | #ifdef CONFIG_PRINTK |
| 138 | errstr = btrfs_decode_error(error); |
| 139 | btrfs_state_to_string(fs_info, statestr); |
| 140 | if (fmt) { |
| 141 | struct va_format vaf; |
| 142 | va_list args; |
| 143 | |
| 144 | va_start(args, fmt); |
| 145 | vaf.fmt = fmt; |
| 146 | vaf.va = &args; |
| 147 | |
| 148 | pr_crit("BTRFS: error (device %s%s) in %s:%d: errno=%d %s (%pV)\n", |
| 149 | sb->s_id, statestr, function, line, error, errstr, &vaf); |
| 150 | va_end(args); |
| 151 | } else { |
| 152 | pr_crit("BTRFS: error (device %s%s) in %s:%d: errno=%d %s\n", |
| 153 | sb->s_id, statestr, function, line, error, errstr); |
| 154 | } |
| 155 | #endif |
| 156 | |
| 157 | /* |
| 158 | * Today we only save the error info to memory. Long term we'll also |
| 159 | * send it down to the disk. |
| 160 | */ |
| 161 | WRITE_ONCE(fs_info->fs_error, error); |
| 162 | |
| 163 | /* Don't go through full error handling during mount. */ |
| 164 | if (!(sb->s_flags & SB_BORN)) |
| 165 | return; |
| 166 | |
| 167 | if (sb_rdonly(sb)) |
| 168 | return; |
| 169 | |
| 170 | btrfs_discard_stop(fs_info); |
| 171 | |
| 172 | /* Handle error by forcing the filesystem readonly. */ |
| 173 | btrfs_set_sb_rdonly(sb); |
| 174 | btrfs_info(fs_info, "forced readonly"); |
| 175 | /* |
| 176 | * Note that a running device replace operation is not canceled here |
| 177 | * although there is no way to update the progress. It would add the |
| 178 | * risk of a deadlock, therefore the canceling is omitted. The only |
| 179 | * penalty is that some I/O remains active until the procedure |
| 180 | * completes. The next time when the filesystem is mounted writable |
| 181 | * again, the device replace operation continues. |
| 182 | */ |
| 183 | } |
| 184 | |
| 185 | #ifdef CONFIG_PRINTK |
| 186 | static const char * const logtypes[] = { |
| 187 | "emergency", |
| 188 | "alert", |
| 189 | "critical", |
| 190 | "error", |
| 191 | "warning", |
| 192 | "notice", |
| 193 | "info", |
| 194 | "debug", |
| 195 | }; |
| 196 | |
| 197 | /* |
| 198 | * Use one ratelimit state per log level so that a flood of less important |
| 199 | * messages doesn't cause more important ones to be dropped. |
| 200 | */ |
| 201 | static struct ratelimit_state printk_limits[] = { |
| 202 | RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100), |
| 203 | RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100), |
| 204 | RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100), |
| 205 | RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100), |
| 206 | RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100), |
| 207 | RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100), |
| 208 | RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100), |
| 209 | RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100), |
| 210 | }; |
| 211 | |
| 212 | void __cold _btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...) |
| 213 | { |
| 214 | char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0"; |
| 215 | struct va_format vaf; |
| 216 | va_list args; |
| 217 | int kern_level; |
| 218 | const char *type = logtypes[4]; |
| 219 | struct ratelimit_state *ratelimit = &printk_limits[4]; |
| 220 | |
| 221 | #ifdef CONFIG_PRINTK_INDEX |
| 222 | printk_index_subsys_emit("%sBTRFS %s (device %s): ", NULL, fmt); |
| 223 | #endif |
| 224 | |
| 225 | va_start(args, fmt); |
| 226 | |
| 227 | while ((kern_level = printk_get_level(fmt)) != 0) { |
| 228 | size_t size = printk_skip_level(fmt) - fmt; |
| 229 | |
| 230 | if (kern_level >= '0' && kern_level <= '7') { |
| 231 | memcpy(lvl, fmt, size); |
| 232 | lvl[size] = '\0'; |
| 233 | type = logtypes[kern_level - '0']; |
| 234 | ratelimit = &printk_limits[kern_level - '0']; |
| 235 | } |
| 236 | fmt += size; |
| 237 | } |
| 238 | |
| 239 | vaf.fmt = fmt; |
| 240 | vaf.va = &args; |
| 241 | |
| 242 | /* Do not ratelimit if CONFIG_BTRFS_DEBUG is enabled. */ |
| 243 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG) || __ratelimit(ratelimit)) { |
| 244 | if (fs_info) { |
| 245 | char statestr[STATE_STRING_BUF_LEN]; |
| 246 | |
| 247 | btrfs_state_to_string(fs_info, statestr); |
| 248 | _printk("%sBTRFS %s (device %s%s): %pV\n", lvl, type, |
| 249 | fs_info->sb->s_id, statestr, &vaf); |
| 250 | } else { |
| 251 | _printk("%sBTRFS %s: %pV\n", lvl, type, &vaf); |
| 252 | } |
| 253 | } |
| 254 | |
| 255 | va_end(args); |
| 256 | } |
| 257 | #endif |
| 258 | |
| 259 | #if BITS_PER_LONG == 32 |
| 260 | void __cold btrfs_warn_32bit_limit(struct btrfs_fs_info *fs_info) |
| 261 | { |
| 262 | if (!test_and_set_bit(BTRFS_FS_32BIT_WARN, &fs_info->flags)) { |
| 263 | btrfs_warn(fs_info, "reaching 32bit limit for logical addresses"); |
| 264 | btrfs_warn(fs_info, |
| 265 | "due to page cache limit on 32bit systems, btrfs can't access metadata at or beyond %lluT", |
| 266 | BTRFS_32BIT_MAX_FILE_SIZE >> 40); |
| 267 | btrfs_warn(fs_info, |
| 268 | "please consider upgrading to 64bit kernel/hardware"); |
| 269 | } |
| 270 | } |
| 271 | |
| 272 | void __cold btrfs_err_32bit_limit(struct btrfs_fs_info *fs_info) |
| 273 | { |
| 274 | if (!test_and_set_bit(BTRFS_FS_32BIT_ERROR, &fs_info->flags)) { |
| 275 | btrfs_err(fs_info, "reached 32bit limit for logical addresses"); |
| 276 | btrfs_err(fs_info, |
| 277 | "due to page cache limit on 32bit systems, metadata beyond %lluT can't be accessed", |
| 278 | BTRFS_32BIT_MAX_FILE_SIZE >> 40); |
| 279 | btrfs_err(fs_info, |
| 280 | "please consider upgrading to 64bit kernel/hardware"); |
| 281 | } |
| 282 | } |
| 283 | #endif |
| 284 | |
| 285 | /* |
| 286 | * Decode unexpected, fatal errors from the caller, issue an alert, and either |
| 287 | * panic or BUGs, depending on mount options. |
| 288 | */ |
| 289 | __cold |
| 290 | void __btrfs_panic(const struct btrfs_fs_info *fs_info, const char *function, |
| 291 | unsigned int line, int error, const char *fmt, ...) |
| 292 | { |
| 293 | char *s_id = "<unknown>"; |
| 294 | const char *errstr; |
| 295 | struct va_format vaf = { .fmt = fmt }; |
| 296 | va_list args; |
| 297 | |
| 298 | if (fs_info) |
| 299 | s_id = fs_info->sb->s_id; |
| 300 | |
| 301 | va_start(args, fmt); |
| 302 | vaf.va = &args; |
| 303 | |
| 304 | errstr = btrfs_decode_error(error); |
| 305 | if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR))) |
| 306 | panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n", |
| 307 | s_id, function, line, &vaf, error, errstr); |
| 308 | |
| 309 | btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)", |
| 310 | function, line, &vaf, error, errstr); |
| 311 | va_end(args); |
| 312 | /* Caller calls BUG() */ |
| 313 | } |