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