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 #ifdef __KERNEL__
  69 #define ask_yn()        false
  70 #else
  71 #include "tools-util.h"
  72 #endif
  73
  74 static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
  75 {
  76         struct fsck_err_state *s;
  77
  78         if (test_bit(BCH_FS_FSCK_DONE, &c->flags))
  79                 return NULL;
  80
  81         list_for_each_entry(s, &c->fsck_errors, list)
  82                 if (s->fmt == fmt) {
  83                         /*
  84                          * move it to the head of the list: repeated fsck errors
  85                          * are common
  86                          */
  87                         list_move(&s->list, &c->fsck_errors);
  88                         return s;
  89                 }
  90
  91         s = kzalloc(sizeof(*s), GFP_NOFS);
  92         if (!s) {
  93                 if (!c->fsck_alloc_err)
  94                         bch_err(c, "kmalloc err, cannot ratelimit fsck errs");
  95                 c->fsck_alloc_err = true;
  96                 return NULL;
  97         }
  98
  99         INIT_LIST_HEAD(&s->list);
 100         s->fmt = fmt;
 101         s->buf = PRINTBUF;
 102         list_add(&s->list, &c->fsck_errors);
 103         return s;
 104 }
 105
 106 int bch2_fsck_err(struct bch_fs *c, unsigned flags, const char *fmt, ...)
 107 {
 108         struct fsck_err_state *s = NULL;
 109         va_list args;
 110         bool print = true, suppressing = false, inconsistent = false;
 111         struct printbuf buf = PRINTBUF, *out = &buf;
 112         int ret = -BCH_ERR_fsck_ignore;
 113
 114         mutex_lock(&c->fsck_error_lock);
 115         s = fsck_err_get(c, fmt);
 116         if (s) {
 117                 if (c->opts.ratelimit_errors &&
 118                     !(flags & FSCK_NO_RATELIMIT) &&
 119                     s->nr >= FSCK_ERR_RATELIMIT_NR) {
 120                         if (s->nr == FSCK_ERR_RATELIMIT_NR)
 121                                 suppressing = true;
 122                         else
 123                                 print = false;
 124                 }
 125
 126                 printbuf_reset(&s->buf);
 127                 out = &s->buf;
 128                 s->nr++;
 129         }
 130
 131 #ifdef BCACHEFS_LOG_PREFIX
 132         if (!strncmp(fmt, "bcachefs:", 9))
 133                 prt_printf(out, bch2_log_msg(c, ""));
 134 #endif
 135
 136         va_start(args, fmt);
 137         prt_vprintf(out, fmt, args);
 138         va_end(args);
 139
 140         if (test_bit(BCH_FS_FSCK_DONE, &c->flags)) {
 141                 if (c->opts.errors != BCH_ON_ERROR_continue ||
 142                     !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
 143                         prt_str(out, ", shutting down");
 144                         inconsistent = true;
 145                         ret = -BCH_ERR_fsck_errors_not_fixed;
 146                 } else if (flags & FSCK_CAN_FIX) {
 147                         prt_str(out, ", fixing");
 148                         ret = -BCH_ERR_fsck_fix;
 149                 } else {
 150                         prt_str(out, ", continuing");
 151                         ret = -BCH_ERR_fsck_ignore;
 152                 }
 153         } else if (c->opts.fix_errors == FSCK_OPT_EXIT) {
 154                 prt_str(out, ", exiting");
 155                 ret = -BCH_ERR_fsck_errors_not_fixed;
 156         } else if (flags & FSCK_CAN_FIX) {
 157                 if (c->opts.fix_errors == FSCK_OPT_ASK) {
 158                         prt_str(out, ": fix?");
 159                         bch2_print_string_as_lines(KERN_ERR, out->buf);
 160                         print = false;
 161                         ret = ask_yn()
 162                                 ? -BCH_ERR_fsck_fix
 163                                 : -BCH_ERR_fsck_ignore;
 164                 } else if (c->opts.fix_errors == FSCK_OPT_YES ||
 165                            (c->opts.nochanges &&
 166                             !(flags & FSCK_CAN_IGNORE))) {
 167                         prt_str(out, ", fixing");
 168                         ret = -BCH_ERR_fsck_fix;
 169                 } else {
 170                         prt_str(out, ", not fixing");
 171                 }
 172         } else if (flags & FSCK_NEED_FSCK) {
 173                 prt_str(out, " (run fsck to correct)");
 174         } else {
 175                 prt_str(out, " (repair unimplemented)");
 176         }
 177
 178         if (ret == -BCH_ERR_fsck_ignore &&
 179             (c->opts.fix_errors == FSCK_OPT_EXIT ||
 180              !(flags & FSCK_CAN_IGNORE)))
 181                 ret = -BCH_ERR_fsck_errors_not_fixed;
 182
 183         if (print)
 184                 bch2_print_string_as_lines(KERN_ERR, out->buf);
 185
 186         if (!test_bit(BCH_FS_FSCK_DONE, &c->flags) &&
 187             (ret != -BCH_ERR_fsck_fix &&
 188              ret != -BCH_ERR_fsck_ignore))
 189                 bch_err(c, "Unable to continue, halting");
 190         else if (suppressing)
 191                 bch_err(c, "Ratelimiting new instances of previous error");
 192
 193         mutex_unlock(&c->fsck_error_lock);
 194
 195         printbuf_exit(&buf);
 196
 197         if (inconsistent)
 198                 bch2_inconsistent_error(c);
 199
 200         if (ret == -BCH_ERR_fsck_fix) {
 201                 set_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 202         } else {
 203                 set_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags);
 204                 set_bit(BCH_FS_ERROR, &c->flags);
 205         }
 206
 207         return ret;
 208 }
 209
 210 void bch2_flush_fsck_errs(struct bch_fs *c)
 211 {
 212         struct fsck_err_state *s, *n;
 213
 214         mutex_lock(&c->fsck_error_lock);
 215
 216         list_for_each_entry_safe(s, n, &c->fsck_errors, list) {
 217                 if (s->ratelimited)
 218                         bch_err(c, "Saw %llu errors like:\n    %s", s->nr, s->buf.buf);
 219
 220                 list_del(&s->list);
 221                 printbuf_exit(&s->buf);
 222                 kfree(s);
 223         }
 224
 225         mutex_unlock(&c->fsck_error_lock);
 226 }