fs/bcachefs/error.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include "bcachefs.h"
   3 #include "error.h"
   4 #include "journal.h"
   5 #include "recovery_passes.h"
   6 #include "super.h"
   7 #include "thread_with_file.h"
   8
   9 #define FSCK_ERR_RATELIMIT_NR   10
  10
  11 bool bch2_inconsistent_error(struct bch_fs *c)
  12 {
  13         set_bit(BCH_FS_error, &c->flags);
  14
  15         switch (c->opts.errors) {
  16         case BCH_ON_ERROR_continue:
  17                 return false;
  18         case BCH_ON_ERROR_ro:
  19                 if (bch2_fs_emergency_read_only(c))
  20                         bch_err(c, "inconsistency detected - emergency read only at journal seq %llu",
  21                                 journal_cur_seq(&c->journal));
  22                 return true;
  23         case BCH_ON_ERROR_panic:
  24                 panic(bch2_fmt(c, "panic after error"));
  25                 return true;
  26         default:
  27                 BUG();
  28         }
  29 }
  30
  31 int bch2_topology_error(struct bch_fs *c)
  32 {
  33         set_bit(BCH_FS_topology_error, &c->flags);
  34         if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
  35                 bch2_inconsistent_error(c);
  36                 return -BCH_ERR_btree_need_topology_repair;
  37         } else {
  38                 return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology) ?:
  39                         -BCH_ERR_btree_node_read_validate_error;
  40         }
  41 }
  42
  43 void bch2_fatal_error(struct bch_fs *c)
  44 {
  45         if (bch2_fs_emergency_read_only(c))
  46                 bch_err(c, "fatal error - emergency read only");
  47 }
  48
  49 void bch2_io_error_work(struct work_struct *work)
  50 {
  51         struct bch_dev *ca = container_of(work, struct bch_dev, io_error_work);
  52         struct bch_fs *c = ca->fs;
  53         bool dev;
  54
  55         down_write(&c->state_lock);
  56         dev = bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_ro,
  57                                     BCH_FORCE_IF_DEGRADED);
  58         if (dev
  59             ? __bch2_dev_set_state(c, ca, BCH_MEMBER_STATE_ro,
  60                                   BCH_FORCE_IF_DEGRADED)
  61             : bch2_fs_emergency_read_only(c))
  62                 bch_err(ca,
  63                         "too many IO errors, setting %s RO",
  64                         dev ? "device" : "filesystem");
  65         up_write(&c->state_lock);
  66 }
  67
  68 void bch2_io_error(struct bch_dev *ca, enum bch_member_error_type type)
  69 {
  70         atomic64_inc(&ca->errors[type]);
  71         //queue_work(system_long_wq, &ca->io_error_work);
  72 }
  73
  74 enum ask_yn {
  75         YN_NO,
  76         YN_YES,
  77         YN_ALLNO,
  78         YN_ALLYES,
  79 };
  80
  81 static enum ask_yn parse_yn_response(char *buf)
  82 {
  83         buf = strim(buf);
  84
  85         if (strlen(buf) == 1)
  86                 switch (buf[0]) {
  87                 case 'n':
  88                         return YN_NO;
  89                 case 'y':
  90                         return YN_YES;
  91                 case 'N':
  92                         return YN_ALLNO;
  93                 case 'Y':
  94                         return YN_ALLYES;
  95                 }
  96         return -1;
  97 }
  98
  99 #ifdef __KERNEL__
 100 static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c)
 101 {
 102         struct stdio_redirect *stdio = c->stdio;
 103
 104         if (c->stdio_filter && c->stdio_filter != current)
 105                 stdio = NULL;
 106
 107         if (!stdio)
 108                 return YN_NO;
 109
 110         char buf[100];
 111         int ret;
 112
 113         do {
 114                 bch2_print(c, " (y,n, or Y,N for all errors of this type) ");
 115
 116                 int r = bch2_stdio_redirect_readline(stdio, buf, sizeof(buf) - 1);
 117                 if (r < 0)
 118                         return YN_NO;
 119                 buf[r] = '\0';
 120         } while ((ret = parse_yn_response(buf)) < 0);
 121
 122         return ret;
 123 }
 124 #else
 125
 126 #include "tools-util.h"
 127
 128 static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c)
 129 {
 130         char *buf = NULL;
 131         size_t buflen = 0;
 132         int ret;
 133
 134         do {
 135                 fputs(" (y,n, or Y,N for all errors of this type) ", stdout);
 136                 fflush(stdout);
 137
 138                 if (getline(&buf, &buflen, stdin) < 0)
 139                         die("error reading from standard input");
 140         } while ((ret = parse_yn_response(buf)) < 0);
 141
 142         free(buf);
 143         return ret;
 144 }
 145
 146 #endif
 147
 148 static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
 149 {
 150         struct fsck_err_state *s;
 151
 152         if (!test_bit(BCH_FS_fsck_running, &c->flags))
 153                 return NULL;
 154
 155         list_for_each_entry(s, &c->fsck_error_msgs, list)
 156                 if (s->fmt == fmt) {
 157                         /*
 158                          * move it to the head of the list: repeated fsck errors
 159                          * are common
 160                          */
 161                         list_move(&s->list, &c->fsck_error_msgs);
 162                         return s;
 163                 }
 164
 165         s = kzalloc(sizeof(*s), GFP_NOFS);
 166         if (!s) {
 167                 if (!c->fsck_alloc_msgs_err)
 168                         bch_err(c, "kmalloc err, cannot ratelimit fsck errs");
 169                 c->fsck_alloc_msgs_err = true;
 170                 return NULL;
 171         }
 172
 173         INIT_LIST_HEAD(&s->list);
 174         s->fmt = fmt;
 175         list_add(&s->list, &c->fsck_error_msgs);
 176         return s;
 177 }
 178
 179 int bch2_fsck_err(struct bch_fs *c,
 180                   enum bch_fsck_flags flags,
 181                   enum bch_sb_error_id err,
 182                   const char *fmt, ...)
 183 {
 184         struct fsck_err_state *s = NULL;
 185         va_list args;
 186         bool print = true, suppressing = false, inconsistent = false;
 187         struct printbuf buf = PRINTBUF, *out = &buf;
 188         int ret = -BCH_ERR_fsck_ignore;
 189
 190         if ((flags & FSCK_CAN_FIX) &&
 191             test_bit(err, c->sb.errors_silent))
 192                 return -BCH_ERR_fsck_fix;
 193
 194         bch2_sb_error_count(c, err);
 195
 196         va_start(args, fmt);
 197         prt_vprintf(out, fmt, args);
 198         va_end(args);
 199
 200         mutex_lock(&c->fsck_error_msgs_lock);
 201         s = fsck_err_get(c, fmt);
 202         if (s) {
 203                 /*
 204                  * We may be called multiple times for the same error on
 205                  * transaction restart - this memoizes instead of asking the user
 206                  * multiple times for the same error:
 207                  */
 208                 if (s->last_msg && !strcmp(buf.buf, s->last_msg)) {
 209                         ret = s->ret;
 210                         mutex_unlock(&c->fsck_error_msgs_lock);
 211                         printbuf_exit(&buf);
 212                         return ret;
 213                 }
 214
 215                 kfree(s->last_msg);
 216                 s->last_msg = kstrdup(buf.buf, GFP_KERNEL);
 217
 218                 if (c->opts.ratelimit_errors &&
 219                     !(flags & FSCK_NO_RATELIMIT) &&
 220                     s->nr >= FSCK_ERR_RATELIMIT_NR) {
 221                         if (s->nr == FSCK_ERR_RATELIMIT_NR)
 222                                 suppressing = true;
 223                         else
 224                                 print = false;
 225                 }
 226
 227                 s->nr++;
 228         }
 229
 230 #ifdef BCACHEFS_LOG_PREFIX
 231         if (!strncmp(fmt, "bcachefs:", 9))
 232                 prt_printf(out, bch2_log_msg(c, ""));
 233 #endif
 234
 235         if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
 236                 if (c->opts.errors != BCH_ON_ERROR_continue ||
 237                     !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
 238                         prt_str(out, ", shutting down");
 239                         inconsistent = true;
 240                         ret = -BCH_ERR_fsck_errors_not_fixed;
 241                 } else if (flags & FSCK_CAN_FIX) {
 242                         prt_str(out, ", fixing");
 243                         ret = -BCH_ERR_fsck_fix;
 244                 } else {
 245                         prt_str(out, ", continuing");
 246                         ret = -BCH_ERR_fsck_ignore;
 247                 }
 248         } else if (c->opts.fix_errors == FSCK_FIX_exit) {
 249                 prt_str(out, ", exiting");
 250                 ret = -BCH_ERR_fsck_errors_not_fixed;
 251         } else if (flags & FSCK_CAN_FIX) {
 252                 int fix = s && s->fix
 253                         ? s->fix
 254                         : c->opts.fix_errors;
 255
 256                 if (fix == FSCK_FIX_ask) {
 257                         int ask;
 258
 259                         prt_str(out, ": fix?");
 260                         if (bch2_fs_stdio_redirect(c))
 261                                 bch2_print(c, "%s", out->buf);
 262                         else
 263                                 bch2_print_string_as_lines(KERN_ERR, out->buf);
 264                         print = false;
 265
 266                         ask = bch2_fsck_ask_yn(c);
 267
 268                         if (ask >= YN_ALLNO && s)
 269                                 s->fix = ask == YN_ALLNO
 270                                         ? FSCK_FIX_no
 271                                         : FSCK_FIX_yes;
 272
 273                         ret = ask & 1
 274                                 ? -BCH_ERR_fsck_fix
 275                                 : -BCH_ERR_fsck_ignore;
 276                 } else if (fix == FSCK_FIX_yes ||
 277                            (c->opts.nochanges &&
 278                             !(flags & FSCK_CAN_IGNORE))) {
 279                         prt_str(out, ", fixing");
 280                         ret = -BCH_ERR_fsck_fix;
 281                 } else {
 282                         prt_str(out, ", not fixing");
 283                 }
 284         } else if (flags & FSCK_NEED_FSCK) {
 285                 prt_str(out, " (run fsck to correct)");
 286         } else {
 287                 prt_str(out, " (repair unimplemented)");
 288         }
 289
 290         if (ret == -BCH_ERR_fsck_ignore &&
 291             (c->opts.fix_errors == FSCK_FIX_exit ||
 292              !(flags & FSCK_CAN_IGNORE)))
 293                 ret = -BCH_ERR_fsck_errors_not_fixed;
 294
 295         if (print) {
 296                 if (bch2_fs_stdio_redirect(c))
 297                         bch2_print(c, "%s\n", out->buf);
 298                 else
 299                         bch2_print_string_as_lines(KERN_ERR, out->buf);
 300         }
 301
 302         if (test_bit(BCH_FS_fsck_running, &c->flags) &&
 303             (ret != -BCH_ERR_fsck_fix &&
 304              ret != -BCH_ERR_fsck_ignore))
 305                 bch_err(c, "Unable to continue, halting");
 306         else if (suppressing)
 307                 bch_err(c, "Ratelimiting new instances of previous error");
 308
 309         if (s)
 310                 s->ret = ret;
 311
 312         mutex_unlock(&c->fsck_error_msgs_lock);
 313
 314         printbuf_exit(&buf);
 315
 316         if (inconsistent)
 317                 bch2_inconsistent_error(c);
 318
 319         if (ret == -BCH_ERR_fsck_fix) {
 320                 set_bit(BCH_FS_errors_fixed, &c->flags);
 321         } else {
 322                 set_bit(BCH_FS_errors_not_fixed, &c->flags);
 323                 set_bit(BCH_FS_error, &c->flags);
 324         }
 325
 326         return ret;
 327 }
 328
 329 void bch2_flush_fsck_errs(struct bch_fs *c)
 330 {
 331         struct fsck_err_state *s, *n;
 332
 333         mutex_lock(&c->fsck_error_msgs_lock);
 334
 335         list_for_each_entry_safe(s, n, &c->fsck_error_msgs, list) {
 336                 if (s->ratelimited && s->last_msg)
 337                         bch_err(c, "Saw %llu errors like:\n    %s", s->nr, s->last_msg);
 338
 339                 list_del(&s->list);
 340                 kfree(s->last_msg);
 341                 kfree(s);
 342         }
 343
 344         mutex_unlock(&c->fsck_error_msgs_lock);
 345 }