fs/bcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_background.h"
   5 #include "btree_gc.h"
   6 #include "btree_update.h"
   7 #include "btree_update_interior.h"
   8 #include "btree_io.h"
   9 #include "buckets.h"
  10 #include "dirent.h"
  11 #include "ec.h"
  12 #include "error.h"
  13 #include "fsck.h"
  14 #include "journal_io.h"
  15 #include "quota.h"
  16 #include "recovery.h"
  17 #include "replicas.h"
  18 #include "super-io.h"
  19
  20 #include <linux/stat.h>
  21
  22 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  23
  24 static struct bkey_i *btree_root_find(struct bch_fs *c,
  25                                       struct bch_sb_field_clean *clean,
  26                                       struct jset *j,
  27                                       enum btree_id id, unsigned *level)
  28 {
  29         struct bkey_i *k;
  30         struct jset_entry *entry, *start, *end;
  31
  32         if (clean) {
  33                 start = clean->start;
  34                 end = vstruct_end(&clean->field);
  35         } else {
  36                 start = j->start;
  37                 end = vstruct_last(j);
  38         }
  39
  40         for (entry = start; entry < end; entry = vstruct_next(entry))
  41                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
  42                     entry->btree_id == id)
  43                         goto found;
  44
  45         return NULL;
  46 found:
  47         if (!entry->u64s)
  48                 return ERR_PTR(-EINVAL);
  49
  50         k = entry->start;
  51         *level = entry->level;
  52         return k;
  53 }
  54
  55 static int journal_replay_entry_early(struct bch_fs *c,
  56                                       struct jset_entry *entry)
  57 {
  58         int ret = 0;
  59
  60         switch (entry->type) {
  61         case BCH_JSET_ENTRY_btree_root: {
  62                 struct btree_root *r = &c->btree_roots[entry->btree_id];
  63
  64                 if (entry->u64s) {
  65                         r->level = entry->level;
  66                         bkey_copy(&r->key, &entry->start[0]);
  67                         r->error = 0;
  68                 } else {
  69                         r->error = -EIO;
  70                 }
  71                 r->alive = true;
  72                 break;
  73         }
  74         case BCH_JSET_ENTRY_usage: {
  75                 struct jset_entry_usage *u =
  76                         container_of(entry, struct jset_entry_usage, entry);
  77
  78                 switch (entry->btree_id) {
  79                 case FS_USAGE_RESERVED:
  80                         if (entry->level < BCH_REPLICAS_MAX)
  81                                 percpu_u64_set(&c->usage[0]->
  82                                                persistent_reserved[entry->level],
  83                                                le64_to_cpu(u->v));
  84                         break;
  85                 case FS_USAGE_INODES:
  86                         percpu_u64_set(&c->usage[0]->s.nr_inodes,
  87                                        le64_to_cpu(u->v));
  88                         break;
  89                 case FS_USAGE_KEY_VERSION:
  90                         atomic64_set(&c->key_version,
  91                                      le64_to_cpu(u->v));
  92                         break;
  93                 }
  94
  95                 break;
  96         }
  97         case BCH_JSET_ENTRY_data_usage: {
  98                 struct jset_entry_data_usage *u =
  99                         container_of(entry, struct jset_entry_data_usage, entry);
 100                 ret = bch2_replicas_set_usage(c, &u->r,
 101                                               le64_to_cpu(u->v));
 102                 break;
 103         }
 104         }
 105
 106         return ret;
 107 }
 108
 109 static int verify_superblock_clean(struct bch_fs *c,
 110                                    struct bch_sb_field_clean *clean,
 111                                    struct jset *j)
 112 {
 113         unsigned i;
 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                 bch2_fs_mark_clean(c, false);
 124
 125         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 126                         "superblock read clock doesn't match journal after clean shutdown");
 127         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 128                         "superblock read clock doesn't match journal after clean shutdown");
 129
 130         for (i = 0; i < BTREE_ID_NR; i++) {
 131                 struct bkey_i *k1, *k2;
 132                 unsigned l1 = 0, l2 = 0;
 133
 134                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 135                 k2 = btree_root_find(c, NULL, j, i, &l2);
 136
 137                 if (!k1 && !k2)
 138                         continue;
 139
 140                 mustfix_fsck_err_on(!k1 || !k2 ||
 141                                     IS_ERR(k1) ||
 142                                     IS_ERR(k2) ||
 143                                     k1->k.u64s != k2->k.u64s ||
 144                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 145                                     l1 != l2, c,
 146                         "superblock btree root doesn't match journal after clean shutdown");
 147         }
 148 fsck_err:
 149         return ret;
 150 }
 151
 152 static bool journal_empty(struct list_head *journal)
 153 {
 154         struct journal_replay *i;
 155         struct jset_entry *entry;
 156
 157         if (list_empty(journal))
 158                 return true;
 159
 160         i = list_last_entry(journal, struct journal_replay, list);
 161
 162         if (i->j.last_seq != i->j.seq)
 163                 return false;
 164
 165         list_for_each_entry(i, journal, list) {
 166                 vstruct_for_each(&i->j, entry) {
 167                         if (entry->type == BCH_JSET_ENTRY_btree_root ||
 168                             entry->type == BCH_JSET_ENTRY_usage ||
 169                             entry->type == BCH_JSET_ENTRY_data_usage)
 170                                 continue;
 171
 172                         if (entry->type == BCH_JSET_ENTRY_btree_keys &&
 173                             !entry->u64s)
 174                                 continue;
 175                         return false;
 176                 }
 177         }
 178
 179         return true;
 180 }
 181
 182 int bch2_fs_recovery(struct bch_fs *c)
 183 {
 184         const char *err = "cannot allocate memory";
 185         struct bch_sb_field_clean *clean = NULL, *sb_clean = NULL;
 186         struct jset_entry *entry;
 187         LIST_HEAD(journal);
 188         struct jset *j = NULL;
 189         unsigned i;
 190         int ret;
 191
 192         mutex_lock(&c->sb_lock);
 193         if (!c->replicas.entries) {
 194                 bch_info(c, "building replicas info");
 195                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 196         }
 197
 198         if (c->sb.clean)
 199                 sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 200         if (sb_clean) {
 201                 clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 202                                 GFP_KERNEL);
 203                 if (!clean) {
 204                         ret = -ENOMEM;
 205                         mutex_unlock(&c->sb_lock);
 206                         goto err;
 207                 }
 208
 209                 if (le16_to_cpu(c->disk_sb.sb->version) <
 210                     bcachefs_metadata_version_bkey_renumber)
 211                         bch2_sb_clean_renumber(clean, READ);
 212         }
 213         mutex_unlock(&c->sb_lock);
 214
 215         if (clean)
 216                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 217                          le64_to_cpu(clean->journal_seq));
 218
 219         if (!clean || c->opts.fsck) {
 220                 ret = bch2_journal_read(c, &journal);
 221                 if (ret)
 222                         goto err;
 223
 224                 j = &list_entry(journal.prev, struct journal_replay, list)->j;
 225         } else {
 226                 ret = bch2_journal_set_seq(c,
 227                                            le64_to_cpu(clean->journal_seq),
 228                                            le64_to_cpu(clean->journal_seq));
 229                 BUG_ON(ret);
 230         }
 231
 232         ret = verify_superblock_clean(c, clean, j);
 233         if (ret)
 234                 goto err;
 235
 236         fsck_err_on(clean && !journal_empty(&journal), c,
 237                     "filesystem marked clean but journal not empty");
 238
 239         err = "insufficient memory";
 240         if (clean) {
 241                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 242                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 243
 244                 for (entry = clean->start;
 245                      entry != vstruct_end(&clean->field);
 246                      entry = vstruct_next(entry)) {
 247                         ret = journal_replay_entry_early(c, entry);
 248                         if (ret)
 249                                 goto err;
 250                 }
 251         } else {
 252                 struct journal_replay *i;
 253
 254                 c->bucket_clock[READ].hand = le16_to_cpu(j->read_clock);
 255                 c->bucket_clock[WRITE].hand = le16_to_cpu(j->write_clock);
 256
 257                 list_for_each_entry(i, &journal, list)
 258                         vstruct_for_each(&i->j, entry) {
 259                                 ret = journal_replay_entry_early(c, entry);
 260                                 if (ret)
 261                                         goto err;
 262                         }
 263         }
 264
 265         bch2_fs_usage_initialize(c);
 266
 267         for (i = 0; i < BTREE_ID_NR; i++) {
 268                 struct btree_root *r = &c->btree_roots[i];
 269
 270                 if (!r->alive)
 271                         continue;
 272
 273                 err = "invalid btree root pointer";
 274                 if (r->error)
 275                         goto err;
 276
 277                 err = "error reading btree root";
 278                 if (bch2_btree_root_read(c, i, &r->key, r->level)) {
 279                         if (i != BTREE_ID_ALLOC)
 280                                 goto err;
 281
 282                         mustfix_fsck_err(c, "error reading btree root");
 283                 }
 284         }
 285
 286         for (i = 0; i < BTREE_ID_NR; i++)
 287                 if (!c->btree_roots[i].b)
 288                         bch2_btree_root_alloc(c, i);
 289
 290         err = "error reading allocation information";
 291         ret = bch2_alloc_read(c, &journal);
 292         if (ret)
 293                 goto err;
 294
 295         bch_verbose(c, "starting stripes_read");
 296         ret = bch2_stripes_read(c, &journal);
 297         if (ret)
 298                 goto err;
 299         bch_verbose(c, "stripes_read done");
 300
 301         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 302
 303         bch_verbose(c, "starting mark and sweep:");
 304         err = "error in recovery";
 305         ret = bch2_gc(c, &journal, true);
 306         if (ret)
 307                 goto err;
 308         bch_verbose(c, "mark and sweep done");
 309
 310         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 311
 312         /*
 313          * Skip past versions that might have possibly been used (as nonces),
 314          * but hadn't had their pointers written:
 315          */
 316         if (c->sb.encryption_type && !c->sb.clean)
 317                 atomic64_add(1 << 16, &c->key_version);
 318
 319         if (c->opts.noreplay)
 320                 goto out;
 321
 322         /*
 323          * Mark dirty before journal replay, fsck:
 324          * XXX: after a clean shutdown, this could be done lazily only when fsck
 325          * finds an error
 326          */
 327         bch2_fs_mark_clean(c, false);
 328
 329         /*
 330          * bch2_fs_journal_start() can't happen sooner, or btree_gc_finish()
 331          * will give spurious errors about oldest_gen > bucket_gen -
 332          * this is a hack but oh well.
 333          */
 334         bch2_fs_journal_start(&c->journal);
 335
 336         err = "error starting allocator";
 337         ret = bch2_fs_allocator_start(c);
 338         if (ret)
 339                 goto err;
 340
 341         bch_verbose(c, "starting journal replay:");
 342         err = "journal replay failed";
 343         ret = bch2_journal_replay(c, &journal);
 344         if (ret)
 345                 goto err;
 346         bch_verbose(c, "journal replay done");
 347
 348         if (c->opts.norecovery)
 349                 goto out;
 350
 351         err = "error in fsck";
 352         ret = bch2_fsck(c);
 353         if (ret)
 354                 goto err;
 355
 356         mutex_lock(&c->sb_lock);
 357         if (c->opts.version_upgrade) {
 358                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 359                         c->disk_sb.sb->version_min =
 360                                 le16_to_cpu(bcachefs_metadata_version_min);
 361                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 362         }
 363
 364         if (!test_bit(BCH_FS_FSCK_UNFIXED_ERRORS, &c->flags))
 365                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 366         mutex_unlock(&c->sb_lock);
 367
 368         if (enabled_qtypes(c)) {
 369                 bch_verbose(c, "reading quotas:");
 370                 ret = bch2_fs_quota_read(c);
 371                 if (ret)
 372                         goto err;
 373                 bch_verbose(c, "quotas done");
 374         }
 375
 376 out:
 377         bch2_journal_entries_free(&journal);
 378         kfree(clean);
 379         return ret;
 380 err:
 381 fsck_err:
 382         pr_err("Error in recovery: %s (%i)", err, ret);
 383         goto out;
 384 }
 385
 386 int bch2_fs_initialize(struct bch_fs *c)
 387 {
 388         struct bch_inode_unpacked root_inode, lostfound_inode;
 389         struct bkey_inode_buf packed_inode;
 390         struct bch_hash_info root_hash_info;
 391         struct qstr lostfound = QSTR("lost+found");
 392         const char *err = "cannot allocate memory";
 393         struct bch_dev *ca;
 394         LIST_HEAD(journal);
 395         unsigned i;
 396         int ret;
 397
 398         bch_notice(c, "initializing new filesystem");
 399
 400         mutex_lock(&c->sb_lock);
 401         for_each_online_member(ca, c, i)
 402                 bch2_mark_dev_superblock(c, ca, 0);
 403         mutex_unlock(&c->sb_lock);
 404
 405         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 406
 407         for (i = 0; i < BTREE_ID_NR; i++)
 408                 bch2_btree_root_alloc(c, i);
 409
 410         ret = bch2_gc(c, &journal, true);
 411         if (ret)
 412                 goto err;
 413
 414         err = "unable to allocate journal buckets";
 415         for_each_online_member(ca, c, i)
 416                 if (bch2_dev_journal_alloc(ca)) {
 417                         percpu_ref_put(&ca->io_ref);
 418                         goto err;
 419                 }
 420
 421         /*
 422          * journal_res_get() will crash if called before this has
 423          * set up the journal.pin FIFO and journal.cur pointer:
 424          */
 425         bch2_fs_journal_start(&c->journal);
 426         bch2_journal_set_replay_done(&c->journal);
 427
 428         err = "error starting allocator";
 429         ret = bch2_fs_allocator_start(c);
 430         if (ret)
 431                 goto err;
 432
 433         bch2_inode_init(c, &root_inode, 0, 0,
 434                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
 435         root_inode.bi_inum = BCACHEFS_ROOT_INO;
 436         root_inode.bi_nlink++; /* lost+found */
 437         bch2_inode_pack(&packed_inode, &root_inode);
 438
 439         err = "error creating root directory";
 440         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 441                                 &packed_inode.inode.k_i,
 442                                 NULL, NULL, 0);
 443         if (ret)
 444                 goto err;
 445
 446         bch2_inode_init(c, &lostfound_inode, 0, 0,
 447                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0,
 448                         &root_inode);
 449         lostfound_inode.bi_inum = BCACHEFS_ROOT_INO + 1;
 450         bch2_inode_pack(&packed_inode, &lostfound_inode);
 451
 452         err = "error creating lost+found";
 453         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 454                                 &packed_inode.inode.k_i,
 455                                 NULL, NULL, 0);
 456         if (ret)
 457                 goto err;
 458
 459         root_hash_info = bch2_hash_info_init(c, &root_inode);
 460
 461         ret = bch2_dirent_create(c, BCACHEFS_ROOT_INO, &root_hash_info, DT_DIR,
 462                                  &lostfound, lostfound_inode.bi_inum, NULL,
 463                                  BTREE_INSERT_NOFAIL);
 464         if (ret)
 465                 goto err;
 466
 467         if (enabled_qtypes(c)) {
 468                 ret = bch2_fs_quota_read(c);
 469                 if (ret)
 470                         goto err;
 471         }
 472
 473         err = "error writing first journal entry";
 474         ret = bch2_journal_meta(&c->journal);
 475         if (ret)
 476                 goto err;
 477
 478         mutex_lock(&c->sb_lock);
 479         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
 480                 le16_to_cpu(bcachefs_metadata_version_current);
 481         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 482
 483         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
 484         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 485
 486         bch2_write_super(c);
 487         mutex_unlock(&c->sb_lock);
 488
 489         return 0;
 490 err:
 491         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
 492         return ret;
 493 }