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

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include "bcachefs.h"
   3 #include "error.h"
   4 #include "super.h"
   5 #include "thread_with_file.h"
   6
   7 #define FSCK_ERR_RATELIMIT_NR   10
   8
   9 bool bch2_inconsistent_error(struct bch_fs *c)
  10 {
  11         set_bit(BCH_FS_error, &c->flags);
  12
  13         switch (c->opts.errors) {
  14         case BCH_ON_ERROR_continue:
  15                 return false;
  16         case BCH_ON_ERROR_ro:
  17                 if (bch2_fs_emergency_read_only(c))
  18                         bch_err(c, "inconsistency detected - emergency read only");
  19                 return true;
  20         case BCH_ON_ERROR_panic:
  21                 panic(bch2_fmt(c, "panic after error"));
  22                 return true;
  23         default:
  24                 BUG();
  25         }
  26 }
  27
  28 void bch2_topology_error(struct bch_fs *c)
  29 {
  30         set_bit(BCH_FS_topology_error, &c->flags);
  31         if (!test_bit(BCH_FS_fsck_running, &c->flags))
  32                 bch2_inconsistent_error(c);
  33 }
  34
  35 void bch2_fatal_error(struct bch_fs *c)
  36 {
  37         if (bch2_fs_emergency_read_only(c))
  38                 bch_err(c, "fatal error - emergency read only");
  39 }
  40
  41 void bch2_io_error_work(struct work_struct *work)
  42 {
  43         struct bch_dev *ca = container_of(work, struct bch_dev, io_error_work);
  44         struct bch_fs *c = ca->fs;
  45         bool dev;
  46
  47         down_write(&c->state_lock);
  48         dev = bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_ro,
  49                                     BCH_FORCE_IF_DEGRADED);
  50         if (dev
  51             ? __bch2_dev_set_state(c, ca, BCH_MEMBER_STATE_ro,
  52                                   BCH_FORCE_IF_DEGRADED)
  53             : bch2_fs_emergency_read_only(c))
  54                 bch_err(ca,
  55                         "too many IO errors, setting %s RO",
  56                         dev ? "device" : "filesystem");
  57         up_write(&c->state_lock);
  58 }
  59
  60 void bch2_io_error(struct bch_dev *ca, enum bch_member_error_type type)
  61 {
  62         atomic64_inc(&ca->errors[type]);
  63         //queue_work(system_long_wq, &ca->io_error_work);
  64 }
  65
  66 enum ask_yn {
  67         YN_NO,
  68         YN_YES,
  69         YN_ALLNO,
  70         YN_ALLYES,
  71 };
  72
  73 static enum ask_yn parse_yn_response(char *buf)
  74 {
  75         buf = strim(buf);
  76
  77         if (strlen(buf) == 1)
  78                 switch (buf[0]) {
  79                 case 'n':
  80                         return YN_NO;
  81                 case 'y':
  82                         return YN_YES;
  83                 case 'N':
  84                         return YN_ALLNO;
  85                 case 'Y':
  86                         return YN_ALLYES;
  87                 }
  88         return -1;
  89 }
  90
  91 #ifdef __KERNEL__
  92 static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c)
  93 {
  94         struct stdio_redirect *stdio = c->stdio;
  95
  96         if (c->stdio_filter && c->stdio_filter != current)
  97                 stdio = NULL;
  98
  99         if (!stdio)
 100                 return YN_NO;
 101
 102         char buf[100];
 103         int ret;
 104
 105         do {
 106                 bch2_print(c, " (y,n, or Y,N for all errors of this type) ");
 107
 108                 int r = bch2_stdio_redirect_readline(stdio, buf, sizeof(buf) - 1);
 109                 if (r < 0)
 110                         return YN_NO;
 111                 buf[r] = '\0';
 112         } while ((ret = parse_yn_response(buf)) < 0);
 113
 114         return ret;
 115 }
 116 #else
 117
 118 #include "tools-util.h"
 119
 120 static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c)
 121 {
 122         char *buf = NULL;
 123         size_t buflen = 0;
 124         int ret;
 125
 126         do {
 127                 fputs(" (y,n, or Y,N for all errors of this type) ", stdout);
 128                 fflush(stdout);
 129
 130                 if (getline(&buf, &buflen, stdin) < 0)
 131                         die("error reading from standard input");
 132         } while ((ret = parse_yn_response(buf)) < 0);
 133
 134         free(buf);
 135         return ret;
 136 }
 137
 138 #endif
 139
 140 static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
 141 {
 142         struct fsck_err_state *s;
 143
 144         if (!test_bit(BCH_FS_fsck_running, &c->flags))
 145                 return NULL;
 146
 147         list_for_each_entry(s, &c->fsck_error_msgs, list)
 148                 if (s->fmt == fmt) {
 149                         /*
 150                          * move it to the head of the list: repeated fsck errors
 151                          * are common
 152                          */
 153                         list_move(&s->list, &c->fsck_error_msgs);
 154                         return s;
 155                 }
 156
 157         s = kzalloc(sizeof(*s), GFP_NOFS);
 158         if (!s) {
 159                 if (!c->fsck_alloc_msgs_err)
 160                         bch_err(c, "kmalloc err, cannot ratelimit fsck errs");
 161                 c->fsck_alloc_msgs_err = true;
 162                 return NULL;
 163         }
 164
 165         INIT_LIST_HEAD(&s->list);
 166         s->fmt = fmt;
 167         list_add(&s->list, &c->fsck_error_msgs);
 168         return s;
 169 }
 170
 171 int bch2_fsck_err(struct bch_fs *c,
 172                   enum bch_fsck_flags flags,
 173                   enum bch_sb_error_id err,
 174                   const char *fmt, ...)
 175 {
 176         struct fsck_err_state *s = NULL;
 177         va_list args;
 178         bool print = true, suppressing = false, inconsistent = false;
 179         struct printbuf buf = PRINTBUF, *out = &buf;
 180         int ret = -BCH_ERR_fsck_ignore;
 181
 182         if ((flags & FSCK_CAN_FIX) &&
 183             test_bit(err, c->sb.errors_silent))
 184                 return -BCH_ERR_fsck_fix;
 185
 186         bch2_sb_error_count(c, err);
 187
 188         va_start(args, fmt);
 189         prt_vprintf(out, fmt, args);
 190         va_end(args);
 191
 192         mutex_lock(&c->fsck_error_msgs_lock);
 193         s = fsck_err_get(c, fmt);
 194         if (s) {
 195                 /*
 196                  * We may be called multiple times for the same error on
 197                  * transaction restart - this memoizes instead of asking the user
 198                  * multiple times for the same error:
 199                  */
 200                 if (s->last_msg && !strcmp(buf.buf, s->last_msg)) {
 201                         ret = s->ret;
 202                         mutex_unlock(&c->fsck_error_msgs_lock);
 203                         printbuf_exit(&buf);
 204                         return ret;
 205                 }
 206
 207                 kfree(s->last_msg);
 208                 s->last_msg = kstrdup(buf.buf, GFP_KERNEL);
 209
 210                 if (c->opts.ratelimit_errors &&
 211                     !(flags & FSCK_NO_RATELIMIT) &&
 212                     s->nr >= FSCK_ERR_RATELIMIT_NR) {
 213                         if (s->nr == FSCK_ERR_RATELIMIT_NR)
 214                                 suppressing = true;
 215                         else
 216                                 print = false;
 217                 }
 218
 219                 s->nr++;
 220         }
 221
 222 #ifdef BCACHEFS_LOG_PREFIX
 223         if (!strncmp(fmt, "bcachefs:", 9))
 224                 prt_printf(out, bch2_log_msg(c, ""));
 225 #endif
 226
 227         if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
 228                 if (c->opts.errors != BCH_ON_ERROR_continue ||
 229                     !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
 230                         prt_str(out, ", shutting down");
 231                         inconsistent = true;
 232                         ret = -BCH_ERR_fsck_errors_not_fixed;
 233                 } else if (flags & FSCK_CAN_FIX) {
 234                         prt_str(out, ", fixing");
 235                         ret = -BCH_ERR_fsck_fix;
 236                 } else {
 237                         prt_str(out, ", continuing");
 238                         ret = -BCH_ERR_fsck_ignore;
 239                 }
 240         } else if (c->opts.fix_errors == FSCK_FIX_exit) {
 241                 prt_str(out, ", exiting");
 242                 ret = -BCH_ERR_fsck_errors_not_fixed;
 243         } else if (flags & FSCK_CAN_FIX) {
 244                 int fix = s && s->fix
 245                         ? s->fix
 246                         : c->opts.fix_errors;
 247
 248                 if (fix == FSCK_FIX_ask) {
 249                         int ask;
 250
 251                         prt_str(out, ": fix?");
 252                         if (bch2_fs_stdio_redirect(c))
 253                                 bch2_print(c, "%s", out->buf);
 254                         else
 255                                 bch2_print_string_as_lines(KERN_ERR, out->buf);
 256                         print = false;
 257
 258                         ask = bch2_fsck_ask_yn(c);
 259
 260                         if (ask >= YN_ALLNO && s)
 261                                 s->fix = ask == YN_ALLNO
 262                                         ? FSCK_FIX_no
 263                                         : FSCK_FIX_yes;
 264
 265                         ret = ask & 1
 266                                 ? -BCH_ERR_fsck_fix
 267                                 : -BCH_ERR_fsck_ignore;
 268                 } else if (fix == FSCK_FIX_yes ||
 269                            (c->opts.nochanges &&
 270                             !(flags & FSCK_CAN_IGNORE))) {
 271                         prt_str(out, ", fixing");
 272                         ret = -BCH_ERR_fsck_fix;
 273                 } else {
 274                         prt_str(out, ", not fixing");
 275                 }
 276         } else if (flags & FSCK_NEED_FSCK) {
 277                 prt_str(out, " (run fsck to correct)");
 278         } else {
 279                 prt_str(out, " (repair unimplemented)");
 280         }
 281
 282         if (ret == -BCH_ERR_fsck_ignore &&
 283             (c->opts.fix_errors == FSCK_FIX_exit ||
 284              !(flags & FSCK_CAN_IGNORE)))
 285                 ret = -BCH_ERR_fsck_errors_not_fixed;
 286
 287         if (print) {
 288                 if (bch2_fs_stdio_redirect(c))
 289                         bch2_print(c, "%s\n", out->buf);
 290                 else
 291                         bch2_print_string_as_lines(KERN_ERR, out->buf);
 292         }
 293
 294         if (test_bit(BCH_FS_fsck_running, &c->flags) &&
 295             (ret != -BCH_ERR_fsck_fix &&
 296              ret != -BCH_ERR_fsck_ignore))
 297                 bch_err(c, "Unable to continue, halting");
 298         else if (suppressing)
 299                 bch_err(c, "Ratelimiting new instances of previous error");
 300
 301         if (s)
 302                 s->ret = ret;
 303
 304         mutex_unlock(&c->fsck_error_msgs_lock);
 305
 306         printbuf_exit(&buf);
 307
 308         if (inconsistent)
 309                 bch2_inconsistent_error(c);
 310
 311         if (ret == -BCH_ERR_fsck_fix) {
 312                 set_bit(BCH_FS_errors_fixed, &c->flags);
 313         } else {
 314                 set_bit(BCH_FS_errors_not_fixed, &c->flags);
 315                 set_bit(BCH_FS_error, &c->flags);
 316         }
 317
 318         return ret;
 319 }
 320
 321 void bch2_flush_fsck_errs(struct bch_fs *c)
 322 {
 323         struct fsck_err_state *s, *n;
 324
 325         mutex_lock(&c->fsck_error_msgs_lock);
 326
 327         list_for_each_entry_safe(s, n, &c->fsck_error_msgs, list) {
 328                 if (s->ratelimited && s->last_msg)
 329                         bch_err(c, "Saw %llu errors like:\n    %s", s->nr, s->last_msg);
 330
 331                 list_del(&s->list);
 332                 kfree(s->last_msg);
 333                 kfree(s);
 334         }
 335
 336         mutex_unlock(&c->fsck_error_msgs_lock);
 337 }