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
  79 #include "tools-util.h"
  80
  81 enum ask_yn bch2_fsck_ask_yn(void)
  82 {
  83         char *buf = NULL;
  84         size_t buflen = 0;
  85         bool ret;
  86
  87         while (true) {
  88                 fputs(" (y,n, or Y,N for all errors of this type) ", stdout);
  89                 fflush(stdout);
  90
  91                 if (getline(&buf, &buflen, stdin) < 0)
  92                         die("error reading from standard input");
  93
  94                 strim(buf);
  95                 if (strlen(buf) != 1)
  96                         continue;
  97
  98                 switch (buf[0]) {
  99                 case 'n':
 100                         return YN_NO;
 101                 case 'y':
 102                         return YN_YES;
 103                 case 'N':
 104                         return YN_ALLNO;
 105                 case 'Y':
 106                         return YN_ALLYES;
 107                 }
 108         }
 109
 110         free(buf);
 111         return ret;
 112 }
 113
 114 #endif
 115
 116 static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
 117 {
 118         struct fsck_err_state *s;
 119
 120         if (test_bit(BCH_FS_FSCK_DONE, &c->flags))
 121                 return NULL;
 122
 123         list_for_each_entry(s, &c->fsck_errors, list)
 124                 if (s->fmt == fmt) {
 125                         /*
 126                          * move it to the head of the list: repeated fsck errors
 127                          * are common
 128                          */
 129                         list_move(&s->list, &c->fsck_errors);
 130                         return s;
 131                 }
 132
 133         s = kzalloc(sizeof(*s), GFP_NOFS);
 134         if (!s) {
 135                 if (!c->fsck_alloc_err)
 136                         bch_err(c, "kmalloc err, cannot ratelimit fsck errs");
 137                 c->fsck_alloc_err = true;
 138                 return NULL;
 139         }
 140
 141         INIT_LIST_HEAD(&s->list);
 142         s->fmt = fmt;
 143         list_add(&s->list, &c->fsck_errors);
 144         return s;
 145 }
 146
 147 int bch2_fsck_err(struct bch_fs *c, unsigned flags, 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         va_start(args, fmt);
 156         prt_vprintf(out, fmt, args);
 157         va_end(args);
 158
 159         mutex_lock(&c->fsck_error_lock);
 160         s = fsck_err_get(c, fmt);
 161         if (s) {
 162                 /*
 163                  * We may be called multiple times for the same error on
 164                  * transaction restart - this memoizes instead of asking the user
 165                  * multiple times for the same error:
 166                  */
 167                 if (s->last_msg && !strcmp(buf.buf, s->last_msg)) {
 168                         ret = s->ret;
 169                         mutex_unlock(&c->fsck_error_lock);
 170                         printbuf_exit(&buf);
 171                         return ret;
 172                 }
 173
 174                 kfree(s->last_msg);
 175                 s->last_msg = kstrdup(buf.buf, GFP_KERNEL);
 176
 177                 if (c->opts.ratelimit_errors &&
 178                     !(flags & FSCK_NO_RATELIMIT) &&
 179                     s->nr >= FSCK_ERR_RATELIMIT_NR) {
 180                         if (s->nr == FSCK_ERR_RATELIMIT_NR)
 181                                 suppressing = true;
 182                         else
 183                                 print = false;
 184                 }
 185
 186                 s->nr++;
 187         }
 188
 189 #ifdef BCACHEFS_LOG_PREFIX
 190         if (!strncmp(fmt, "bcachefs:", 9))
 191                 prt_printf(out, bch2_log_msg(c, ""));
 192 #endif
 193
 194         if (test_bit(BCH_FS_FSCK_DONE, &c->flags)) {
 195                 if (c->opts.errors != BCH_ON_ERROR_continue ||
 196                     !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
 197                         prt_str(out, ", shutting down");
 198                         inconsistent = true;
 199                         ret = -BCH_ERR_fsck_errors_not_fixed;
 200                 } else if (flags & FSCK_CAN_FIX) {
 201                         prt_str(out, ", fixing");
 202                         ret = -BCH_ERR_fsck_fix;
 203                 } else {
 204                         prt_str(out, ", continuing");
 205                         ret = -BCH_ERR_fsck_ignore;
 206                 }
 207         } else if (c->opts.fix_errors == FSCK_OPT_EXIT) {
 208                 prt_str(out, ", exiting");
 209                 ret = -BCH_ERR_fsck_errors_not_fixed;
 210         } else if (flags & FSCK_CAN_FIX) {
 211                 int fix = s && s->fix
 212                         ? s->fix
 213                         : c->opts.fix_errors;
 214
 215                 if (fix == FSCK_OPT_ASK) {
 216                         int ask;
 217
 218                         prt_str(out, ": fix?");
 219                         bch2_print_string_as_lines(KERN_ERR, out->buf);
 220                         print = false;
 221
 222                         ask = bch2_fsck_ask_yn();
 223
 224                         if (ask >= YN_ALLNO && s)
 225                                 s->fix = ask == YN_ALLNO
 226                                         ? FSCK_OPT_NO
 227                                         : FSCK_OPT_YES;
 228
 229                         ret = ask & 1
 230                                 ? -BCH_ERR_fsck_fix
 231                                 : -BCH_ERR_fsck_ignore;
 232                 } else if (fix == FSCK_OPT_YES ||
 233                            (c->opts.nochanges &&
 234                             !(flags & FSCK_CAN_IGNORE))) {
 235                         prt_str(out, ", fixing");
 236                         ret = -BCH_ERR_fsck_fix;
 237                 } else {
 238                         prt_str(out, ", not fixing");
 239                 }
 240         } else if (flags & FSCK_NEED_FSCK) {
 241                 prt_str(out, " (run fsck to correct)");
 242         } else {
 243                 prt_str(out, " (repair unimplemented)");
 244         }
 245
 246         if (ret == -BCH_ERR_fsck_ignore &&
 247             (c->opts.fix_errors == FSCK_OPT_EXIT ||
 248              !(flags & FSCK_CAN_IGNORE)))
 249                 ret = -BCH_ERR_fsck_errors_not_fixed;
 250
 251         if (print)
 252                 bch2_print_string_as_lines(KERN_ERR, out->buf);
 253
 254         if (!test_bit(BCH_FS_FSCK_DONE, &c->flags) &&
 255             (ret != -BCH_ERR_fsck_fix &&
 256              ret != -BCH_ERR_fsck_ignore))
 257                 bch_err(c, "Unable to continue, halting");
 258         else if (suppressing)
 259                 bch_err(c, "Ratelimiting new instances of previous error");
 260
 261         if (s)
 262                 s->ret = ret;
 263
 264         mutex_unlock(&c->fsck_error_lock);
 265
 266         printbuf_exit(&buf);
 267
 268         if (inconsistent)
 269                 bch2_inconsistent_error(c);
 270
 271         if (ret == -BCH_ERR_fsck_fix) {
 272                 set_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 273         } else {
 274                 set_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags);
 275                 set_bit(BCH_FS_ERROR, &c->flags);
 276         }
 277
 278         return ret;
 279 }
 280
 281 void bch2_flush_fsck_errs(struct bch_fs *c)
 282 {
 283         struct fsck_err_state *s, *n;
 284
 285         mutex_lock(&c->fsck_error_lock);
 286
 287         list_for_each_entry_safe(s, n, &c->fsck_errors, list) {
 288                 if (s->ratelimited && s->last_msg)
 289                         bch_err(c, "Saw %llu errors like:\n    %s", s->nr, s->last_msg);
 290
 291                 list_del(&s->list);
 292                 kfree(s->last_msg);
 293                 kfree(s);
 294         }
 295
 296         mutex_unlock(&c->fsck_error_lock);
 297 }