git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/recovery.c

   1
   2 #include "bcachefs.h"
   3 #include "alloc_background.h"
   4 #include "btree_gc.h"
   5 #include "btree_update.h"
   6 #include "btree_update_interior.h"
   7 #include "btree_io.h"
   8 #include "buckets.h"
   9 #include "dirent.h"
  10 #include "ec.h"
  11 #include "error.h"
  12 #include "fsck.h"
  13 #include "journal_io.h"
  14 #include "quota.h"
  15 #include "recovery.h"
  16 #include "replicas.h"
  17 #include "super-io.h"
  18
  19 #include <linux/stat.h>
  20
  21 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  22
  23 static struct bkey_i *btree_root_find(struct bch_fs *c,
  24                                       struct bch_sb_field_clean *clean,
  25                                       struct jset *j,
  26                                       enum btree_id id, unsigned *level)
  27 {
  28         struct bkey_i *k;
  29         struct jset_entry *entry, *start, *end;
  30
  31         if (clean) {
  32                 start = clean->start;
  33                 end = vstruct_end(&clean->field);
  34         } else {
  35                 start = j->start;
  36                 end = vstruct_last(j);
  37         }
  38
  39         for (entry = start; entry < end; entry = vstruct_next(entry))
  40                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
  41                     entry->btree_id == id)
  42                         goto found;
  43
  44         return NULL;
  45 found:
  46         if (!entry->u64s)
  47                 return ERR_PTR(-EINVAL);
  48
  49         k = entry->start;
  50         *level = entry->level;
  51         return k;
  52 }
  53
  54 static int journal_replay_entry_early(struct bch_fs *c,
  55                                       struct jset_entry *entry)
  56 {
  57         int ret = 0;
  58
  59         switch (entry->type) {
  60         case BCH_JSET_ENTRY_btree_root: {
  61                 struct btree_root *r = &c->btree_roots[entry->btree_id];
  62
  63                 if (entry->u64s) {
  64                         r->level = entry->level;
  65                         bkey_copy(&r->key, &entry->start[0]);
  66                         r->error = 0;
  67                 } else {
  68                         r->error = -EIO;
  69                 }
  70                 r->alive = true;
  71                 break;
  72         }
  73         case BCH_JSET_ENTRY_usage: {
  74                 struct jset_entry_usage *u =
  75                         container_of(entry, struct jset_entry_usage, entry);
  76
  77                 switch (entry->btree_id) {
  78                 case FS_USAGE_RESERVED:
  79                         if (entry->level < BCH_REPLICAS_MAX)
  80                                 percpu_u64_set(&c->usage[0]->
  81                                                persistent_reserved[entry->level],
  82                                                le64_to_cpu(u->v));
  83                         break;
  84                 case FS_USAGE_INODES:
  85                         percpu_u64_set(&c->usage[0]->nr_inodes,
  86                                        le64_to_cpu(u->v));
  87                         break;
  88                 case FS_USAGE_KEY_VERSION:
  89                         atomic64_set(&c->key_version,
  90                                      le64_to_cpu(u->v));
  91                         break;
  92                 }
  93
  94                 break;
  95         }
  96         case BCH_JSET_ENTRY_data_usage: {
  97                 struct jset_entry_data_usage *u =
  98                         container_of(entry, struct jset_entry_data_usage, entry);
  99                 ret = bch2_replicas_set_usage(c, &u->r,
 100                                               le64_to_cpu(u->v));
 101                 break;
 102         }
 103         }
 104
 105         return ret;
 106 }
 107
 108 static int verify_superblock_clean(struct bch_fs *c,
 109                                    struct bch_sb_field_clean **cleanp,
 110                                    struct jset *j)
 111 {
 112         unsigned i;
 113         struct bch_sb_field_clean *clean = *cleanp;
 114         int ret = 0;
 115
 116         if (!clean || !j)
 117                 return 0;
 118
 119         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 120                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 121                         le64_to_cpu(clean->journal_seq),
 122                         le64_to_cpu(j->seq))) {
 123                 kfree(clean);
 124                 *cleanp = NULL;
 125                 return 0;
 126         }
 127
 128         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 129                         "superblock read clock doesn't match journal after clean shutdown");
 130         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 131                         "superblock read clock doesn't match journal after clean shutdown");
 132
 133         for (i = 0; i < BTREE_ID_NR; i++) {
 134                 struct bkey_i *k1, *k2;
 135                 unsigned l1 = 0, l2 = 0;
 136
 137                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 138                 k2 = btree_root_find(c, NULL, j, i, &l2);
 139
 140                 if (!k1 && !k2)
 141                         continue;
 142
 143                 mustfix_fsck_err_on(!k1 || !k2 ||
 144                                     IS_ERR(k1) ||
 145                                     IS_ERR(k2) ||
 146                                     k1->k.u64s != k2->k.u64s ||
 147                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 148                                     l1 != l2, c,
 149                         "superblock btree root doesn't match journal after clean shutdown");
 150         }
 151 fsck_err:
 152         return ret;
 153 }
 154
 155 static bool journal_empty(struct list_head *journal)
 156 {
 157         struct journal_replay *i;
 158         struct jset_entry *entry;
 159
 160         if (list_empty(journal))
 161                 return true;
 162
 163         i = list_last_entry(journal, struct journal_replay, list);
 164
 165         if (i->j.last_seq != i->j.seq)
 166                 return false;
 167
 168         list_for_each_entry(i, journal, list) {
 169                 vstruct_for_each(&i->j, entry) {
 170                         if (entry->type == BCH_JSET_ENTRY_btree_root ||
 171                             entry->type == BCH_JSET_ENTRY_usage ||
 172                             entry->type == BCH_JSET_ENTRY_data_usage)
 173                                 continue;
 174
 175                         if (entry->type == BCH_JSET_ENTRY_btree_keys &&
 176                             !entry->u64s)
 177                                 continue;
 178                         return false;
 179                 }
 180         }
 181
 182         return true;
 183 }
 184
 185 int bch2_fs_recovery(struct bch_fs *c)
 186 {
 187         const char *err = "cannot allocate memory";
 188         struct bch_sb_field_clean *clean = NULL, *sb_clean = NULL;
 189         struct jset_entry *entry;
 190         LIST_HEAD(journal);
 191         struct jset *j = NULL;
 192         unsigned i;
 193         bool run_gc = c->opts.fsck ||
 194                 !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO));
 195         int ret;
 196
 197         mutex_lock(&c->sb_lock);
 198         if (!c->replicas.entries) {
 199                 bch_info(c, "building replicas info");
 200                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 201         }
 202
 203         if (c->sb.clean)
 204                 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 205         if (sb_clean) {
 206                 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 207                                 GFP_KERNEL);
 208                 if (!clean) {
 209                         ret = -ENOMEM;
 210                         mutex_unlock(&c->sb_lock);
 211                         goto err;
 212                 }
 213
 214                 if (le16_to_cpu(c->disk_sb.sb->version) <
 215                     bcachefs_metadata_version_bkey_renumber)
 216                         bch2_sb_clean_renumber(clean, READ);
 217         }
 218         mutex_unlock(&c->sb_lock);
 219
 220         if (clean)
 221                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 222                          le64_to_cpu(clean->journal_seq));
 223
 224         if (!clean || c->opts.fsck) {
 225                 ret = bch2_journal_read(c, &journal);
 226                 if (ret)
 227                         goto err;
 228
 229                 j = &list_entry(journal.prev, struct journal_replay, list)->j;
 230         } else {
 231                 ret = bch2_journal_set_seq(c,
 232                                            le64_to_cpu(clean->journal_seq),
 233                                            le64_to_cpu(clean->journal_seq));
 234                 BUG_ON(ret);
 235         }
 236
 237         ret = verify_superblock_clean(c, &clean, j);
 238         if (ret)
 239                 goto err;
 240
 241         fsck_err_on(clean && !journal_empty(&journal), c,
 242                     "filesystem marked clean but journal not empty");
 243
 244         err = "insufficient memory";
 245         if (clean) {
 246                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 247                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 248
 249                 for (entry = clean->start;
 250                      entry != vstruct_end(&clean->field);
 251                      entry = vstruct_next(entry)) {
 252                         ret = journal_replay_entry_early(c, entry);
 253                         if (ret)
 254                                 goto err;
 255                 }
 256         } else {
 257                 struct journal_replay *i;
 258
 259                 c->bucket_clock[READ].hand = le16_to_cpu(j->read_clock);
 260                 c->bucket_clock[WRITE].hand = le16_to_cpu(j->write_clock);
 261
 262                 list_for_each_entry(i, &journal, list)
 263                         vstruct_for_each(&i->j, entry) {
 264                                 ret = journal_replay_entry_early(c, entry);
 265                                 if (ret)
 266                                         goto err;
 267                         }
 268         }
 269
 270         bch2_fs_usage_initialize(c);
 271
 272         for (i = 0; i < BTREE_ID_NR; i++) {
 273                 struct btree_root *r = &c->btree_roots[i];
 274
 275                 if (!r->alive)
 276                         continue;
 277
 278                 err = "invalid btree root pointer";
 279                 ret = -1;
 280                 if (r->error)
 281                         goto err;
 282
 283                 if (i == BTREE_ID_ALLOC &&
 284                     test_reconstruct_alloc(c))
 285                         continue;
 286
 287                 err = "error reading btree root";
 288                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 289                 if (ret) {
 290                         if (i != BTREE_ID_ALLOC)
 291                                 goto err;
 292
 293                         mustfix_fsck_err(c, "error reading btree root");
 294                         run_gc = true;
 295                 }
 296         }
 297
 298         for (i = 0; i < BTREE_ID_NR; i++)
 299                 if (!c->btree_roots[i].b)
 300                         bch2_btree_root_alloc(c, i);
 301
 302         err = "error reading allocation information";
 303         ret = bch2_alloc_read(c, &journal);
 304         if (ret)
 305                 goto err;
 306
 307         bch_verbose(c, "starting stripes_read");
 308         ret = bch2_stripes_read(c, &journal);
 309         if (ret)
 310                 goto err;
 311         bch_verbose(c, "stripes_read done");
 312
 313         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 314
 315         if (run_gc) {
 316                 bch_verbose(c, "starting mark and sweep:");
 317                 err = "error in recovery";
 318                 ret = bch2_gc(c, &journal, true);
 319                 if (ret)
 320                         goto err;
 321                 bch_verbose(c, "mark and sweep done");
 322         }
 323
 324         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 325         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 326
 327         /*
 328          * Skip past versions that might have possibly been used (as nonces),
 329          * but hadn't had their pointers written:
 330          */
 331         if (c->sb.encryption_type && !c->sb.clean)
 332                 atomic64_add(1 << 16, &c->key_version);
 333
 334         if (c->opts.noreplay)
 335                 goto out;
 336
 337         /*
 338          * bch2_fs_journal_start() can't happen sooner, or btree_gc_finish()
 339          * will give spurious errors about oldest_gen > bucket_gen -
 340          * this is a hack but oh well.
 341          */
 342         bch2_fs_journal_start(&c->journal);
 343
 344         bch_verbose(c, "starting journal replay:");
 345         err = "journal replay failed";
 346         ret = bch2_journal_replay(c, &journal);
 347         if (ret)
 348                 goto err;
 349         bch_verbose(c, "journal replay done");
 350
 351         if (c->opts.norecovery)
 352                 goto out;
 353
 354         err = "error in fsck";
 355         ret = bch2_fsck(c);
 356         if (ret)
 357                 goto err;
 358
 359         mutex_lock(&c->sb_lock);
 360         if (c->opts.version_upgrade) {
 361                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 362                         c->disk_sb.sb->version_min =
 363                                 le16_to_cpu(bcachefs_metadata_version_min);
 364                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 365         }
 366
 367         if (c->opts.fsck &&
 368             !test_bit(BCH_FS_ERROR, &c->flags)) {
 369                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 370                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 371         }
 372         mutex_unlock(&c->sb_lock);
 373
 374         if (enabled_qtypes(c)) {
 375                 bch_verbose(c, "reading quotas:");
 376                 ret = bch2_fs_quota_read(c);
 377                 if (ret)
 378                         goto err;
 379                 bch_verbose(c, "quotas done");
 380         }
 381
 382 out:
 383         bch2_journal_entries_free(&journal);
 384         kfree(clean);
 385         return ret;
 386 err:
 387 fsck_err:
 388         pr_err("Error in recovery: %s (%i)", err, ret);
 389         goto out;
 390 }
 391
 392 int bch2_fs_initialize(struct bch_fs *c)
 393 {
 394         struct bch_inode_unpacked root_inode, lostfound_inode;
 395         struct bkey_inode_buf packed_inode;
 396         struct bch_hash_info root_hash_info;
 397         struct qstr lostfound = QSTR("lost+found");
 398         const char *err = "cannot allocate memory";
 399         struct bch_dev *ca;
 400         LIST_HEAD(journal);
 401         unsigned i;
 402         int ret;
 403
 404         bch_notice(c, "initializing new filesystem");
 405
 406         mutex_lock(&c->sb_lock);
 407         for_each_online_member(ca, c, i)
 408                 bch2_mark_dev_superblock(c, ca, 0);
 409         mutex_unlock(&c->sb_lock);
 410
 411         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 412         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 413
 414         for (i = 0; i < BTREE_ID_NR; i++)
 415                 bch2_btree_root_alloc(c, i);
 416
 417         err = "unable to allocate journal buckets";
 418         for_each_online_member(ca, c, i) {
 419                 ret = bch2_dev_journal_alloc(ca);
 420                 if (ret) {
 421                         percpu_ref_put(&ca->io_ref);
 422                         goto err;
 423                 }
 424         }
 425
 426         /*
 427          * journal_res_get() will crash if called before this has
 428          * set up the journal.pin FIFO and journal.cur pointer:
 429          */
 430         bch2_fs_journal_start(&c->journal);
 431         bch2_journal_set_replay_done(&c->journal);
 432
 433         err = "error going read write";
 434         ret = __bch2_fs_read_write(c, true);
 435         if (ret)
 436                 goto err;
 437
 438         bch2_inode_init(c, &root_inode, 0, 0,
 439                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
 440         root_inode.bi_inum = BCACHEFS_ROOT_INO;
 441         root_inode.bi_nlink++; /* lost+found */
 442         bch2_inode_pack(&packed_inode, &root_inode);
 443
 444         err = "error creating root directory";
 445         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 446                                 &packed_inode.inode.k_i,
 447                                 NULL, NULL, 0);
 448         if (ret)
 449                 goto err;
 450
 451         bch2_inode_init(c, &lostfound_inode, 0, 0,
 452                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0,
 453                         &root_inode);
 454         lostfound_inode.bi_inum = BCACHEFS_ROOT_INO + 1;
 455         bch2_inode_pack(&packed_inode, &lostfound_inode);
 456
 457         err = "error creating lost+found";
 458         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 459                                 &packed_inode.inode.k_i,
 460                                 NULL, NULL, 0);
 461         if (ret)
 462                 goto err;
 463
 464         root_hash_info = bch2_hash_info_init(c, &root_inode);
 465
 466         ret = bch2_dirent_create(c, BCACHEFS_ROOT_INO, &root_hash_info, DT_DIR,
 467                                  &lostfound, lostfound_inode.bi_inum, NULL,
 468                                  BTREE_INSERT_NOFAIL);
 469         if (ret)
 470                 goto err;
 471
 472         if (enabled_qtypes(c)) {
 473                 ret = bch2_fs_quota_read(c);
 474                 if (ret)
 475                         goto err;
 476         }
 477
 478         err = "error writing first journal entry";
 479         ret = bch2_journal_meta(&c->journal);
 480         if (ret)
 481                 goto err;
 482
 483         mutex_lock(&c->sb_lock);
 484         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
 485                 le16_to_cpu(bcachefs_metadata_version_current);
 486         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 487
 488         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
 489         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 490
 491         bch2_write_super(c);
 492         mutex_unlock(&c->sb_lock);
 493
 494         return 0;
 495 err:
 496         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
 497         return ret;
 498 }