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