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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "btree_update.h"
   6 #include "dirent.h"
   7 #include "error.h"
   8 #include "fs-common.h"
   9 #include "fsck.h"
  10 #include "inode.h"
  11 #include "keylist.h"
  12 #include "super.h"
  13 #include "xattr.h"
  14
  15 #include <linux/dcache.h> /* struct qstr */
  16 #include <linux/generic-radix-tree.h>
  17
  18 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  19
  20 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum)
  21 {
  22         struct btree_iter *iter;
  23         struct bkey_s_c k;
  24         u64 sectors = 0;
  25         int ret;
  26
  27         for_each_btree_key(trans, iter, BTREE_ID_extents,
  28                            POS(inum, 0), 0, k, ret) {
  29                 if (k.k->p.inode != inum)
  30                         break;
  31
  32                 if (bkey_extent_is_allocation(k.k))
  33                         sectors += k.k->size;
  34         }
  35
  36         bch2_trans_iter_free(trans, iter);
  37
  38         return ret ?: sectors;
  39 }
  40
  41 static int __remove_dirent(struct btree_trans *trans,
  42                            struct bkey_s_c_dirent dirent)
  43 {
  44         struct bch_fs *c = trans->c;
  45         struct qstr name;
  46         struct bch_inode_unpacked dir_inode;
  47         struct bch_hash_info dir_hash_info;
  48         u64 dir_inum = dirent.k->p.inode;
  49         int ret;
  50         char *buf;
  51
  52         name.len = bch2_dirent_name_bytes(dirent);
  53         buf = bch2_trans_kmalloc(trans, name.len + 1);
  54         if (IS_ERR(buf))
  55                 return PTR_ERR(buf);
  56
  57         memcpy(buf, dirent.v->d_name, name.len);
  58         buf[name.len] = '\0';
  59         name.name = buf;
  60
  61         ret = __bch2_inode_find_by_inum_trans(trans, dir_inum, &dir_inode, 0);
  62         if (ret && ret != -EINTR)
  63                 bch_err(c, "remove_dirent: err %i looking up directory inode", ret);
  64         if (ret)
  65                 return ret;
  66
  67         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
  68
  69         ret = bch2_hash_delete(trans, bch2_dirent_hash_desc,
  70                                &dir_hash_info, dir_inum, &name);
  71         if (ret && ret != -EINTR)
  72                 bch_err(c, "remove_dirent: err %i deleting dirent", ret);
  73         if (ret)
  74                 return ret;
  75
  76         return 0;
  77 }
  78
  79 static int remove_dirent(struct btree_trans *trans,
  80                          struct bkey_s_c_dirent dirent)
  81 {
  82         return __bch2_trans_do(trans, NULL, NULL,
  83                                BTREE_INSERT_NOFAIL|
  84                                BTREE_INSERT_LAZY_RW,
  85                                __remove_dirent(trans, dirent));
  86 }
  87
  88 static int reattach_inode(struct bch_fs *c,
  89                           struct bch_inode_unpacked *lostfound_inode,
  90                           u64 inum)
  91 {
  92         struct bch_inode_unpacked dir_u, inode_u;
  93         char name_buf[20];
  94         struct qstr name;
  95         int ret;
  96
  97         snprintf(name_buf, sizeof(name_buf), "%llu", inum);
  98         name = (struct qstr) QSTR(name_buf);
  99
 100         ret = bch2_trans_do(c, NULL, NULL,
 101                             BTREE_INSERT_LAZY_RW,
 102                 bch2_link_trans(&trans, lostfound_inode->bi_inum,
 103                                 inum, &dir_u, &inode_u, &name));
 104         if (ret)
 105                 bch_err(c, "error %i reattaching inode %llu", ret, inum);
 106
 107         return ret;
 108 }
 109
 110 struct inode_walker {
 111         bool                    first_this_inode;
 112         bool                    have_inode;
 113         u64                     cur_inum;
 114         struct bch_inode_unpacked inode;
 115 };
 116
 117 static struct inode_walker inode_walker_init(void)
 118 {
 119         return (struct inode_walker) {
 120                 .cur_inum       = -1,
 121                 .have_inode     = false,
 122         };
 123 }
 124
 125 static int walk_inode(struct btree_trans *trans,
 126                       struct inode_walker *w, u64 inum)
 127 {
 128         if (inum != w->cur_inum) {
 129                 int ret = __bch2_inode_find_by_inum_trans(trans, inum,
 130                                                           &w->inode, 0);
 131
 132                 if (ret && ret != -ENOENT)
 133                         return ret;
 134
 135                 w->have_inode   = !ret;
 136                 w->cur_inum     = inum;
 137                 w->first_this_inode = true;
 138         } else {
 139                 w->first_this_inode = false;
 140         }
 141
 142         return 0;
 143 }
 144
 145 struct hash_check {
 146         struct bch_hash_info    info;
 147
 148         /* start of current chain of hash collisions: */
 149         struct btree_iter       *chain;
 150
 151         /* next offset in current chain of hash collisions: */
 152         u64                     chain_end;
 153 };
 154
 155 static void hash_check_init(struct hash_check *h)
 156 {
 157         h->chain = NULL;
 158         h->chain_end = 0;
 159 }
 160
 161 static void hash_stop_chain(struct btree_trans *trans,
 162                             struct hash_check *h)
 163 {
 164         if (h->chain)
 165                 bch2_trans_iter_free(trans, h->chain);
 166         h->chain = NULL;
 167 }
 168
 169 static void hash_check_set_inode(struct btree_trans *trans,
 170                                  struct hash_check *h,
 171                                  const struct bch_inode_unpacked *bi)
 172 {
 173         h->info = bch2_hash_info_init(trans->c, bi);
 174         hash_stop_chain(trans, h);
 175 }
 176
 177 static int hash_redo_key(const struct bch_hash_desc desc,
 178                          struct btree_trans *trans, struct hash_check *h,
 179                          struct btree_iter *k_iter, struct bkey_s_c k,
 180                          u64 hashed)
 181 {
 182         struct bkey_i delete;
 183         struct bkey_i *tmp;
 184
 185         tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
 186         if (IS_ERR(tmp))
 187                 return PTR_ERR(tmp);
 188
 189         bkey_reassemble(tmp, k);
 190
 191         bkey_init(&delete.k);
 192         delete.k.p = k_iter->pos;
 193         bch2_trans_update(trans, k_iter, &delete, 0);
 194
 195         return bch2_hash_set(trans, desc, &h->info, k_iter->pos.inode,
 196                              tmp, 0);
 197 }
 198
 199 static int fsck_hash_delete_at(struct btree_trans *trans,
 200                                const struct bch_hash_desc desc,
 201                                struct bch_hash_info *info,
 202                                struct btree_iter *iter)
 203 {
 204         int ret;
 205 retry:
 206         ret   = bch2_hash_delete_at(trans, desc, info, iter) ?:
 207                 bch2_trans_commit(trans, NULL, NULL,
 208                                   BTREE_INSERT_NOFAIL|
 209                                   BTREE_INSERT_LAZY_RW);
 210         if (ret == -EINTR) {
 211                 ret = bch2_btree_iter_traverse(iter);
 212                 if (!ret)
 213                         goto retry;
 214         }
 215
 216         return ret;
 217 }
 218
 219 static int hash_check_duplicates(struct btree_trans *trans,
 220                         const struct bch_hash_desc desc, struct hash_check *h,
 221                         struct btree_iter *k_iter, struct bkey_s_c k)
 222 {
 223         struct bch_fs *c = trans->c;
 224         struct btree_iter *iter;
 225         struct bkey_s_c k2;
 226         char buf[200];
 227         int ret = 0;
 228
 229         if (!bkey_cmp(h->chain->pos, k_iter->pos))
 230                 return 0;
 231
 232         iter = bch2_trans_copy_iter(trans, h->chain);
 233
 234         for_each_btree_key_continue(iter, 0, k2, ret) {
 235                 if (bkey_cmp(k2.k->p, k.k->p) >= 0)
 236                         break;
 237
 238                 if (fsck_err_on(k2.k->type == desc.key_type &&
 239                                 !desc.cmp_bkey(k, k2), c,
 240                                 "duplicate hash table keys:\n%s",
 241                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 242                                                        k), buf))) {
 243                         ret = fsck_hash_delete_at(trans, desc, &h->info, k_iter);
 244                         if (ret)
 245                                 return ret;
 246                         ret = 1;
 247                         break;
 248                 }
 249         }
 250 fsck_err:
 251         bch2_trans_iter_free(trans, iter);
 252         return ret;
 253 }
 254
 255 static void hash_set_chain_start(struct btree_trans *trans,
 256                         const struct bch_hash_desc desc,
 257                         struct hash_check *h,
 258                         struct btree_iter *k_iter, struct bkey_s_c k)
 259 {
 260         bool hole = (k.k->type != KEY_TYPE_hash_whiteout &&
 261                      k.k->type != desc.key_type);
 262
 263         if (hole || k.k->p.offset > h->chain_end + 1)
 264                 hash_stop_chain(trans, h);
 265
 266         if (!hole) {
 267                 if (!h->chain)
 268                         h->chain = bch2_trans_copy_iter(trans, k_iter);
 269
 270                 h->chain_end = k.k->p.offset;
 271         }
 272 }
 273
 274 static bool key_has_correct_hash(struct btree_trans *trans,
 275                         const struct bch_hash_desc desc,
 276                         struct hash_check *h,
 277                         struct btree_iter *k_iter, struct bkey_s_c k)
 278 {
 279         u64 hash;
 280
 281         hash_set_chain_start(trans, desc, h, k_iter, k);
 282
 283         if (k.k->type != desc.key_type)
 284                 return true;
 285
 286         hash = desc.hash_bkey(&h->info, k);
 287
 288         return hash >= h->chain->pos.offset &&
 289                 hash <= k.k->p.offset;
 290 }
 291
 292 static int hash_check_key(struct btree_trans *trans,
 293                         const struct bch_hash_desc desc, struct hash_check *h,
 294                         struct btree_iter *k_iter, struct bkey_s_c k)
 295 {
 296         struct bch_fs *c = trans->c;
 297         char buf[200];
 298         u64 hashed;
 299         int ret = 0;
 300
 301         hash_set_chain_start(trans, desc, h, k_iter, k);
 302
 303         if (k.k->type != desc.key_type)
 304                 return 0;
 305
 306         hashed = desc.hash_bkey(&h->info, k);
 307
 308         if (fsck_err_on(hashed < h->chain->pos.offset ||
 309                         hashed > k.k->p.offset, c,
 310                         "hash table key at wrong offset: btree %u, %llu, "
 311                         "hashed to %llu chain starts at %llu\n%s",
 312                         desc.btree_id, k.k->p.offset,
 313                         hashed, h->chain->pos.offset,
 314                         (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf))) {
 315                 ret = __bch2_trans_do(trans, NULL, NULL,
 316                                       BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
 317                         hash_redo_key(desc, trans, h, k_iter, k, hashed));
 318                 if (ret) {
 319                         bch_err(c, "hash_redo_key err %i", ret);
 320                         return ret;
 321                 }
 322                 return 1;
 323         }
 324
 325         ret = hash_check_duplicates(trans, desc, h, k_iter, k);
 326 fsck_err:
 327         return ret;
 328 }
 329
 330 static int check_dirent_hash(struct btree_trans *trans, struct hash_check *h,
 331                              struct btree_iter *iter, struct bkey_s_c *k)
 332 {
 333         struct bch_fs *c = trans->c;
 334         struct bkey_i_dirent *d = NULL;
 335         int ret = -EINVAL;
 336         char buf[200];
 337         unsigned len;
 338         u64 hash;
 339
 340         if (key_has_correct_hash(trans, bch2_dirent_hash_desc, h, iter, *k))
 341                 return 0;
 342
 343         len = bch2_dirent_name_bytes(bkey_s_c_to_dirent(*k));
 344         BUG_ON(!len);
 345
 346         memcpy(buf, bkey_s_c_to_dirent(*k).v->d_name, len);
 347         buf[len] = '\0';
 348
 349         d = kmalloc(bkey_bytes(k->k), GFP_KERNEL);
 350         if (!d) {
 351                 bch_err(c, "memory allocation failure");
 352                 return -ENOMEM;
 353         }
 354
 355         bkey_reassemble(&d->k_i, *k);
 356
 357         do {
 358                 --len;
 359                 if (!len)
 360                         goto err_redo;
 361
 362                 d->k.u64s = BKEY_U64s + dirent_val_u64s(len);
 363
 364                 BUG_ON(bkey_val_bytes(&d->k) <
 365                        offsetof(struct bch_dirent, d_name) + len);
 366
 367                 memset(d->v.d_name + len, 0,
 368                        bkey_val_bytes(&d->k) -
 369                        offsetof(struct bch_dirent, d_name) - len);
 370
 371                 hash = bch2_dirent_hash_desc.hash_bkey(&h->info,
 372                                                 bkey_i_to_s_c(&d->k_i));
 373         } while (hash < h->chain->pos.offset ||
 374                  hash > k->k->p.offset);
 375
 376         if (fsck_err(c, "dirent with junk at end, was %s (%zu) now %s (%u)",
 377                      buf, strlen(buf), d->v.d_name, len)) {
 378                 ret = __bch2_trans_do(trans, NULL, NULL,
 379                                       BTREE_INSERT_NOFAIL|
 380                                       BTREE_INSERT_LAZY_RW,
 381                         (bch2_trans_update(trans, iter, &d->k_i, 0), 0));
 382                 if (ret)
 383                         goto err;
 384
 385                 *k = bch2_btree_iter_peek(iter);
 386
 387                 BUG_ON(k->k->type != KEY_TYPE_dirent);
 388         }
 389 err:
 390 fsck_err:
 391         kfree(d);
 392         return ret;
 393 err_redo:
 394         hash = bch2_dirent_hash_desc.hash_bkey(&h->info, *k);
 395
 396         if (fsck_err(c, "cannot fix dirent by removing trailing garbage %s (%zu)\n"
 397                      "hash table key at wrong offset: btree %u, offset %llu, "
 398                      "hashed to %llu chain starts at %llu\n%s",
 399                      buf, strlen(buf), BTREE_ID_dirents,
 400                      k->k->p.offset, hash, h->chain->pos.offset,
 401                      (bch2_bkey_val_to_text(&PBUF(buf), c,
 402                                             *k), buf))) {
 403                 ret = __bch2_trans_do(trans, NULL, NULL,
 404                                       BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
 405                         hash_redo_key(bch2_dirent_hash_desc, trans,
 406                                       h, iter, *k, hash));
 407                 if (ret)
 408                         bch_err(c, "hash_redo_key err %i", ret);
 409                 else
 410                         ret = 1;
 411         }
 412
 413         goto err;
 414 }
 415
 416 static int bch2_inode_truncate(struct bch_fs *c, u64 inode_nr, u64 new_size)
 417 {
 418         return bch2_btree_delete_range(c, BTREE_ID_extents,
 419                         POS(inode_nr, round_up(new_size, block_bytes(c)) >> 9),
 420                         POS(inode_nr + 1, 0), NULL);
 421 }
 422
 423 static int bch2_fix_overlapping_extent(struct btree_trans *trans,
 424                                        struct btree_iter *iter,
 425                                        struct bkey_s_c k, struct bpos cut_at)
 426 {
 427         struct btree_iter *u_iter;
 428         struct bkey_i *u;
 429         int ret;
 430
 431         u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
 432         ret = PTR_ERR_OR_ZERO(u);
 433         if (ret)
 434                 return ret;
 435
 436         bkey_reassemble(u, k);
 437         bch2_cut_front(cut_at, u);
 438
 439         u_iter = bch2_trans_copy_iter(trans, iter);
 440
 441         /*
 442          * We don't want to go through the
 443          * extent_handle_overwrites path:
 444          */
 445         u_iter->flags &= ~BTREE_ITER_IS_EXTENTS;
 446         bch2_btree_iter_set_pos(u_iter, u->k.p);
 447
 448         /*
 449          * XXX: this is going to leave disk space
 450          * accounting slightly wrong
 451          */
 452         ret = bch2_trans_update(trans, u_iter, u, 0);
 453         bch2_trans_iter_put(trans, u_iter);
 454         return ret;
 455 }
 456
 457 /*
 458  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
 459  * that i_size an i_sectors are consistent
 460  */
 461 noinline_for_stack
 462 static int check_extents(struct bch_fs *c)
 463 {
 464         struct inode_walker w = inode_walker_init();
 465         struct btree_trans trans;
 466         struct btree_iter *iter;
 467         struct bkey_s_c k;
 468         struct bkey_buf prev;
 469         u64 i_sectors;
 470         int ret = 0;
 471
 472         bch2_bkey_buf_init(&prev);
 473         prev.k->k = KEY(0, 0, 0);
 474         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 475
 476         bch_verbose(c, "checking extents");
 477
 478         iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
 479                                    POS(BCACHEFS_ROOT_INO, 0),
 480                                    BTREE_ITER_INTENT);
 481 retry:
 482         for_each_btree_key_continue(iter, 0, k, ret) {
 483                 /*
 484                  * due to retry errors we might see the same extent twice:
 485                  */
 486                 if (bkey_cmp(prev.k->k.p, k.k->p) &&
 487                     bkey_cmp(prev.k->k.p, bkey_start_pos(k.k)) > 0) {
 488                         char buf1[200];
 489                         char buf2[200];
 490
 491                         bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
 492                         bch2_bkey_val_to_text(&PBUF(buf2), c, k);
 493
 494                         if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2)) {
 495                                 ret = __bch2_trans_do(&trans, NULL, NULL,
 496                                                       BTREE_INSERT_NOFAIL|
 497                                                       BTREE_INSERT_LAZY_RW,
 498                                                 bch2_fix_overlapping_extent(&trans,
 499                                                                 iter, k, prev.k->k.p));
 500                                 if (ret)
 501                                         goto err;
 502                         }
 503                 }
 504                 bch2_bkey_buf_reassemble(&prev, c, k);
 505
 506                 ret = walk_inode(&trans, &w, k.k->p.inode);
 507                 if (ret)
 508                         break;
 509
 510                 if (fsck_err_on(!w.have_inode, c,
 511                         "extent type %u for missing inode %llu",
 512                         k.k->type, k.k->p.inode) ||
 513                     fsck_err_on(w.have_inode &&
 514                         !S_ISREG(w.inode.bi_mode) && !S_ISLNK(w.inode.bi_mode), c,
 515                         "extent type %u for non regular file, inode %llu mode %o",
 516                         k.k->type, k.k->p.inode, w.inode.bi_mode)) {
 517                         bch2_trans_unlock(&trans);
 518
 519                         ret = bch2_inode_truncate(c, k.k->p.inode, 0);
 520                         if (ret)
 521                                 goto err;
 522                         continue;
 523                 }
 524
 525                 if (fsck_err_on(w.first_this_inode &&
 526                         w.have_inode &&
 527                         !(w.inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY) &&
 528                         w.inode.bi_sectors !=
 529                         (i_sectors = bch2_count_inode_sectors(&trans, w.cur_inum)),
 530                         c, "inode %llu has incorrect i_sectors: got %llu, should be %llu",
 531                         w.inode.bi_inum,
 532                         w.inode.bi_sectors, i_sectors)) {
 533                         struct bkey_inode_buf p;
 534
 535                         w.inode.bi_sectors = i_sectors;
 536
 537                         bch2_trans_unlock(&trans);
 538
 539                         bch2_inode_pack(c, &p, &w.inode);
 540
 541                         ret = bch2_btree_insert(c, BTREE_ID_inodes,
 542                                                 &p.inode.k_i, NULL, NULL,
 543                                                 BTREE_INSERT_NOFAIL|
 544                                                 BTREE_INSERT_LAZY_RW);
 545                         if (ret) {
 546                                 bch_err(c, "error in fsck: error %i updating inode", ret);
 547                                 goto err;
 548                         }
 549
 550                         /* revalidate iterator: */
 551                         k = bch2_btree_iter_peek(iter);
 552                 }
 553
 554                 if (fsck_err_on(w.have_inode &&
 555                         !(w.inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
 556                         k.k->type != KEY_TYPE_reservation &&
 557                         k.k->p.offset > round_up(w.inode.bi_size, block_bytes(c)) >> 9, c,
 558                         "extent type %u offset %llu past end of inode %llu, i_size %llu",
 559                         k.k->type, k.k->p.offset, k.k->p.inode, w.inode.bi_size)) {
 560                         bch2_trans_unlock(&trans);
 561
 562                         ret = bch2_inode_truncate(c, k.k->p.inode,
 563                                                   w.inode.bi_size);
 564                         if (ret)
 565                                 goto err;
 566                         continue;
 567                 }
 568         }
 569 err:
 570 fsck_err:
 571         if (ret == -EINTR)
 572                 goto retry;
 573         bch2_bkey_buf_exit(&prev, c);
 574         return bch2_trans_exit(&trans) ?: ret;
 575 }
 576
 577 /*
 578  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
 579  * validate d_type
 580  */
 581 noinline_for_stack
 582 static int check_dirents(struct bch_fs *c)
 583 {
 584         struct inode_walker w = inode_walker_init();
 585         struct hash_check h;
 586         struct btree_trans trans;
 587         struct btree_iter *iter;
 588         struct bkey_s_c k;
 589         unsigned name_len;
 590         char buf[200];
 591         int ret = 0;
 592
 593         bch_verbose(c, "checking dirents");
 594
 595         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 596
 597         hash_check_init(&h);
 598
 599         iter = bch2_trans_get_iter(&trans, BTREE_ID_dirents,
 600                                    POS(BCACHEFS_ROOT_INO, 0), 0);
 601 retry:
 602         for_each_btree_key_continue(iter, 0, k, ret) {
 603                 struct bkey_s_c_dirent d;
 604                 struct bch_inode_unpacked target;
 605                 bool have_target;
 606                 u64 d_inum;
 607
 608                 ret = walk_inode(&trans, &w, k.k->p.inode);
 609                 if (ret)
 610                         break;
 611
 612                 if (fsck_err_on(!w.have_inode, c,
 613                                 "dirent in nonexisting directory:\n%s",
 614                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 615                                                        k), buf)) ||
 616                     fsck_err_on(!S_ISDIR(w.inode.bi_mode), c,
 617                                 "dirent in non directory inode type %u:\n%s",
 618                                 mode_to_type(w.inode.bi_mode),
 619                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 620                                                        k), buf))) {
 621                         ret = bch2_btree_delete_at(&trans, iter, 0);
 622                         if (ret)
 623                                 goto err;
 624                         continue;
 625                 }
 626
 627                 if (w.first_this_inode && w.have_inode)
 628                         hash_check_set_inode(&trans, &h, &w.inode);
 629
 630                 ret = check_dirent_hash(&trans, &h, iter, &k);
 631                 if (ret > 0) {
 632                         ret = 0;
 633                         continue;
 634                 }
 635                 if (ret)
 636                         goto fsck_err;
 637
 638                 if (ret)
 639                         goto fsck_err;
 640
 641                 if (k.k->type != KEY_TYPE_dirent)
 642                         continue;
 643
 644                 d = bkey_s_c_to_dirent(k);
 645                 d_inum = le64_to_cpu(d.v->d_inum);
 646
 647                 name_len = bch2_dirent_name_bytes(d);
 648
 649                 if (fsck_err_on(!name_len, c, "empty dirent") ||
 650                     fsck_err_on(name_len == 1 &&
 651                                 !memcmp(d.v->d_name, ".", 1), c,
 652                                 ". dirent") ||
 653                     fsck_err_on(name_len == 2 &&
 654                                 !memcmp(d.v->d_name, "..", 2), c,
 655                                 ".. dirent") ||
 656                     fsck_err_on(name_len == 2 &&
 657                                 !memcmp(d.v->d_name, "..", 2), c,
 658                                 ".. dirent") ||
 659                     fsck_err_on(memchr(d.v->d_name, '/', name_len), c,
 660                                 "dirent name has invalid chars")) {
 661                         ret = remove_dirent(&trans, d);
 662                         if (ret)
 663                                 goto err;
 664                         continue;
 665                 }
 666
 667                 if (fsck_err_on(d_inum == d.k->p.inode, c,
 668                                 "dirent points to own directory:\n%s",
 669                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 670                                                        k), buf))) {
 671                         ret = remove_dirent(&trans, d);
 672                         if (ret)
 673                                 goto err;
 674                         continue;
 675                 }
 676
 677                 ret = __bch2_inode_find_by_inum_trans(&trans, d_inum, &target, 0);
 678                 if (ret && ret != -ENOENT)
 679                         break;
 680
 681                 have_target = !ret;
 682                 ret = 0;
 683
 684                 if (fsck_err_on(!have_target, c,
 685                                 "dirent points to missing inode:\n%s",
 686                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 687                                                        k), buf))) {
 688                         ret = remove_dirent(&trans, d);
 689                         if (ret)
 690                                 goto err;
 691                         continue;
 692                 }
 693
 694                 if (fsck_err_on(have_target &&
 695                                 d.v->d_type !=
 696                                 mode_to_type(target.bi_mode), c,
 697                                 "incorrect d_type: should be %u:\n%s",
 698                                 mode_to_type(target.bi_mode),
 699                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 700                                                        k), buf))) {
 701                         struct bkey_i_dirent *n;
 702
 703                         n = kmalloc(bkey_bytes(d.k), GFP_KERNEL);
 704                         if (!n) {
 705                                 ret = -ENOMEM;
 706                                 goto err;
 707                         }
 708
 709                         bkey_reassemble(&n->k_i, d.s_c);
 710                         n->v.d_type = mode_to_type(target.bi_mode);
 711
 712                         ret = __bch2_trans_do(&trans, NULL, NULL,
 713                                               BTREE_INSERT_NOFAIL|
 714                                               BTREE_INSERT_LAZY_RW,
 715                                 (bch2_trans_update(&trans, iter, &n->k_i, 0), 0));
 716                         kfree(n);
 717                         if (ret)
 718                                 goto err;
 719
 720                 }
 721         }
 722
 723         hash_stop_chain(&trans, &h);
 724 err:
 725 fsck_err:
 726         if (ret == -EINTR)
 727                 goto retry;
 728
 729         return bch2_trans_exit(&trans) ?: ret;
 730 }
 731
 732 /*
 733  * Walk xattrs: verify that they all have a corresponding inode
 734  */
 735 noinline_for_stack
 736 static int check_xattrs(struct bch_fs *c)
 737 {
 738         struct inode_walker w = inode_walker_init();
 739         struct hash_check h;
 740         struct btree_trans trans;
 741         struct btree_iter *iter;
 742         struct bkey_s_c k;
 743         int ret = 0;
 744
 745         bch_verbose(c, "checking xattrs");
 746
 747         hash_check_init(&h);
 748
 749         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 750
 751         iter = bch2_trans_get_iter(&trans, BTREE_ID_xattrs,
 752                                    POS(BCACHEFS_ROOT_INO, 0), 0);
 753 retry:
 754         for_each_btree_key_continue(iter, 0, k, ret) {
 755                 ret = walk_inode(&trans, &w, k.k->p.inode);
 756                 if (ret)
 757                         break;
 758
 759                 if (fsck_err_on(!w.have_inode, c,
 760                                 "xattr for missing inode %llu",
 761                                 k.k->p.inode)) {
 762                         ret = bch2_btree_delete_at(&trans, iter, 0);
 763                         if (ret)
 764                                 goto err;
 765                         continue;
 766                 }
 767
 768                 if (w.first_this_inode && w.have_inode)
 769                         hash_check_set_inode(&trans, &h, &w.inode);
 770
 771                 ret = hash_check_key(&trans, bch2_xattr_hash_desc,
 772                                      &h, iter, k);
 773                 if (ret)
 774                         goto fsck_err;
 775         }
 776 err:
 777 fsck_err:
 778         if (ret == -EINTR)
 779                 goto retry;
 780         return bch2_trans_exit(&trans) ?: ret;
 781 }
 782
 783 /* Get root directory, create if it doesn't exist: */
 784 static int check_root(struct bch_fs *c, struct bch_inode_unpacked *root_inode)
 785 {
 786         struct bkey_inode_buf packed;
 787         int ret;
 788
 789         bch_verbose(c, "checking root directory");
 790
 791         ret = bch2_trans_do(c, NULL, NULL, 0,
 792                 __bch2_inode_find_by_inum_trans(&trans, BCACHEFS_ROOT_INO,
 793                                                 root_inode, 0));
 794         if (ret && ret != -ENOENT)
 795                 return ret;
 796
 797         if (fsck_err_on(ret, c, "root directory missing"))
 798                 goto create_root;
 799
 800         if (fsck_err_on(!S_ISDIR(root_inode->bi_mode), c,
 801                         "root inode not a directory"))
 802                 goto create_root;
 803
 804         return 0;
 805 fsck_err:
 806         return ret;
 807 create_root:
 808         bch2_inode_init(c, root_inode, 0, 0, S_IFDIR|0755,
 809                         0, NULL);
 810         root_inode->bi_inum = BCACHEFS_ROOT_INO;
 811
 812         bch2_inode_pack(c, &packed, root_inode);
 813
 814         return bch2_btree_insert(c, BTREE_ID_inodes, &packed.inode.k_i,
 815                                  NULL, NULL,
 816                                  BTREE_INSERT_NOFAIL|
 817                                  BTREE_INSERT_LAZY_RW);
 818 }
 819
 820 /* Get lost+found, create if it doesn't exist: */
 821 static int check_lostfound(struct bch_fs *c,
 822                            struct bch_inode_unpacked *root_inode,
 823                            struct bch_inode_unpacked *lostfound_inode)
 824 {
 825         struct qstr lostfound = QSTR("lost+found");
 826         struct bch_hash_info root_hash_info =
 827                 bch2_hash_info_init(c, root_inode);
 828         u64 inum;
 829         int ret;
 830
 831         bch_verbose(c, "checking lost+found");
 832
 833         inum = bch2_dirent_lookup(c, BCACHEFS_ROOT_INO, &root_hash_info,
 834                                  &lostfound);
 835         if (!inum) {
 836                 bch_notice(c, "creating lost+found");
 837                 goto create_lostfound;
 838         }
 839
 840         ret = bch2_trans_do(c, NULL, NULL, 0,
 841                 __bch2_inode_find_by_inum_trans(&trans, inum, lostfound_inode, 0));
 842         if (ret && ret != -ENOENT)
 843                 return ret;
 844
 845         if (fsck_err_on(ret, c, "lost+found missing"))
 846                 goto create_lostfound;
 847
 848         if (fsck_err_on(!S_ISDIR(lostfound_inode->bi_mode), c,
 849                         "lost+found inode not a directory"))
 850                 goto create_lostfound;
 851
 852         return 0;
 853 fsck_err:
 854         return ret;
 855 create_lostfound:
 856         bch2_inode_init_early(c, lostfound_inode);
 857
 858         ret = bch2_trans_do(c, NULL, NULL,
 859                             BTREE_INSERT_NOFAIL|
 860                             BTREE_INSERT_LAZY_RW,
 861                 bch2_create_trans(&trans,
 862                                   BCACHEFS_ROOT_INO, root_inode,
 863                                   lostfound_inode, &lostfound,
 864                                   0, 0, S_IFDIR|0700, 0, NULL, NULL));
 865         if (ret)
 866                 bch_err(c, "error creating lost+found: %i", ret);
 867
 868         return ret;
 869 }
 870
 871 typedef GENRADIX(unsigned long) inode_bitmap;
 872
 873 static inline bool inode_bitmap_test(inode_bitmap *b, size_t nr)
 874 {
 875         unsigned long *w = genradix_ptr(b, nr / BITS_PER_LONG);
 876         return w ? test_bit(nr & (BITS_PER_LONG - 1), w) : false;
 877 }
 878
 879 static inline int inode_bitmap_set(inode_bitmap *b, size_t nr)
 880 {
 881         unsigned long *w = genradix_ptr_alloc(b, nr / BITS_PER_LONG, GFP_KERNEL);
 882
 883         if (!w)
 884                 return -ENOMEM;
 885
 886         *w |= 1UL << (nr & (BITS_PER_LONG - 1));
 887         return 0;
 888 }
 889
 890 struct pathbuf {
 891         size_t          nr;
 892         size_t          size;
 893
 894         struct pathbuf_entry {
 895                 u64     inum;
 896                 u64     offset;
 897         }               *entries;
 898 };
 899
 900 static int path_down(struct pathbuf *p, u64 inum)
 901 {
 902         if (p->nr == p->size) {
 903                 size_t new_size = max_t(size_t, 256UL, p->size * 2);
 904                 void *n = krealloc(p->entries,
 905                                    new_size * sizeof(p->entries[0]),
 906                                    GFP_KERNEL);
 907                 if (!n)
 908                         return -ENOMEM;
 909
 910                 p->entries = n;
 911                 p->size = new_size;
 912         };
 913
 914         p->entries[p->nr++] = (struct pathbuf_entry) {
 915                 .inum = inum,
 916                 .offset = 0,
 917         };
 918         return 0;
 919 }
 920
 921 noinline_for_stack
 922 static int check_directory_structure(struct bch_fs *c,
 923                                      struct bch_inode_unpacked *lostfound_inode)
 924 {
 925         inode_bitmap dirs_done;
 926         struct pathbuf path = { 0, 0, NULL };
 927         struct pathbuf_entry *e;
 928         struct btree_trans trans;
 929         struct btree_iter *iter;
 930         struct bkey_s_c k;
 931         struct bkey_s_c_dirent dirent;
 932         bool had_unreachable;
 933         u64 d_inum;
 934         int ret = 0;
 935
 936         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 937
 938         bch_verbose(c, "checking directory structure");
 939
 940         /* DFS: */
 941 restart_dfs:
 942         genradix_init(&dirs_done);
 943         had_unreachable = false;
 944
 945         ret = inode_bitmap_set(&dirs_done, BCACHEFS_ROOT_INO);
 946         if (ret) {
 947                 bch_err(c, "memory allocation failure in inode_bitmap_set()");
 948                 goto err;
 949         }
 950
 951         ret = path_down(&path, BCACHEFS_ROOT_INO);
 952         if (ret)
 953                 goto err;
 954
 955         while (path.nr) {
 956 next:
 957                 e = &path.entries[path.nr - 1];
 958
 959                 if (e->offset == U64_MAX)
 960                         goto up;
 961
 962                 for_each_btree_key(&trans, iter, BTREE_ID_dirents,
 963                                    POS(e->inum, e->offset + 1), 0, k, ret) {
 964                         if (k.k->p.inode != e->inum)
 965                                 break;
 966
 967                         e->offset = k.k->p.offset;
 968
 969                         if (k.k->type != KEY_TYPE_dirent)
 970                                 continue;
 971
 972                         dirent = bkey_s_c_to_dirent(k);
 973
 974                         if (dirent.v->d_type != DT_DIR)
 975                                 continue;
 976
 977                         d_inum = le64_to_cpu(dirent.v->d_inum);
 978
 979                         if (fsck_err_on(inode_bitmap_test(&dirs_done, d_inum), c,
 980                                         "directory %llu has multiple hardlinks",
 981                                         d_inum)) {
 982                                 ret = remove_dirent(&trans, dirent);
 983                                 if (ret)
 984                                         goto err;
 985                                 continue;
 986                         }
 987
 988                         ret = inode_bitmap_set(&dirs_done, d_inum);
 989                         if (ret) {
 990                                 bch_err(c, "memory allocation failure in inode_bitmap_set()");
 991                                 goto err;
 992                         }
 993
 994                         ret = path_down(&path, d_inum);
 995                         if (ret) {
 996                                 goto err;
 997                         }
 998
 999                         ret = bch2_trans_iter_free(&trans, iter);
1000                         if (ret) {
1001                                 bch_err(c, "btree error %i in fsck", ret);
1002                                 goto err;
1003                         }
1004                         goto next;
1005                 }
1006                 ret = bch2_trans_iter_free(&trans, iter) ?: ret;
1007                 if (ret) {
1008                         bch_err(c, "btree error %i in fsck", ret);
1009                         goto err;
1010                 }
1011 up:
1012                 path.nr--;
1013         }
1014
1015         iter = bch2_trans_get_iter(&trans, BTREE_ID_inodes, POS_MIN, 0);
1016 retry:
1017         for_each_btree_key_continue(iter, 0, k, ret) {
1018                 if (k.k->type != KEY_TYPE_inode)
1019                         continue;
1020
1021                 if (!S_ISDIR(le16_to_cpu(bkey_s_c_to_inode(k).v->bi_mode)))
1022                         continue;
1023
1024                 ret = bch2_empty_dir_trans(&trans, k.k->p.inode);
1025                 if (ret == -EINTR)
1026                         goto retry;
1027                 if (!ret)
1028                         continue;
1029
1030                 if (fsck_err_on(!inode_bitmap_test(&dirs_done, k.k->p.offset), c,
1031                                 "unreachable directory found (inum %llu)",
1032                                 k.k->p.offset)) {
1033                         bch2_trans_unlock(&trans);
1034
1035                         ret = reattach_inode(c, lostfound_inode, k.k->p.offset);
1036                         if (ret) {
1037                                 goto err;
1038                         }
1039
1040                         had_unreachable = true;
1041                 }
1042         }
1043         bch2_trans_iter_free(&trans, iter);
1044         if (ret)
1045                 goto err;
1046
1047         if (had_unreachable) {
1048                 bch_info(c, "reattached unreachable directories, restarting pass to check for loops");
1049                 genradix_free(&dirs_done);
1050                 kfree(path.entries);
1051                 memset(&dirs_done, 0, sizeof(dirs_done));
1052                 memset(&path, 0, sizeof(path));
1053                 goto restart_dfs;
1054         }
1055 err:
1056 fsck_err:
1057         ret = bch2_trans_exit(&trans) ?: ret;
1058         genradix_free(&dirs_done);
1059         kfree(path.entries);
1060         return ret;
1061 }
1062
1063 struct nlink {
1064         u32     count;
1065         u32     dir_count;
1066 };
1067
1068 typedef GENRADIX(struct nlink) nlink_table;
1069
1070 static void inc_link(struct bch_fs *c, nlink_table *links,
1071                      u64 range_start, u64 *range_end,
1072                      u64 inum, bool dir)
1073 {
1074         struct nlink *link;
1075
1076         if (inum < range_start || inum >= *range_end)
1077                 return;
1078
1079         if (inum - range_start >= SIZE_MAX / sizeof(struct nlink)) {
1080                 *range_end = inum;
1081                 return;
1082         }
1083
1084         link = genradix_ptr_alloc(links, inum - range_start, GFP_KERNEL);
1085         if (!link) {
1086                 bch_verbose(c, "allocation failed during fsck - will need another pass");
1087                 *range_end = inum;
1088                 return;
1089         }
1090
1091         if (dir)
1092                 link->dir_count++;
1093         else
1094                 link->count++;
1095 }
1096
1097 noinline_for_stack
1098 static int bch2_gc_walk_dirents(struct bch_fs *c, nlink_table *links,
1099                                u64 range_start, u64 *range_end)
1100 {
1101         struct btree_trans trans;
1102         struct btree_iter *iter;
1103         struct bkey_s_c k;
1104         struct bkey_s_c_dirent d;
1105         u64 d_inum;
1106         int ret;
1107
1108         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1109
1110         inc_link(c, links, range_start, range_end, BCACHEFS_ROOT_INO, false);
1111
1112         for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN, 0, k, ret) {
1113                 switch (k.k->type) {
1114                 case KEY_TYPE_dirent:
1115                         d = bkey_s_c_to_dirent(k);
1116                         d_inum = le64_to_cpu(d.v->d_inum);
1117
1118                         if (d.v->d_type == DT_DIR)
1119                                 inc_link(c, links, range_start, range_end,
1120                                          d.k->p.inode, true);
1121
1122                         inc_link(c, links, range_start, range_end,
1123                                  d_inum, false);
1124
1125                         break;
1126                 }
1127
1128                 bch2_trans_cond_resched(&trans);
1129         }
1130         ret = bch2_trans_exit(&trans) ?: ret;
1131         if (ret)
1132                 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
1133
1134         return ret;
1135 }
1136
1137 static int check_inode_nlink(struct bch_fs *c,
1138                              struct bch_inode_unpacked *lostfound_inode,
1139                              struct bch_inode_unpacked *u,
1140                              struct nlink *link,
1141                              bool *do_update)
1142 {
1143         u32 i_nlink = bch2_inode_nlink_get(u);
1144         u32 real_i_nlink =
1145                 link->count * nlink_bias(u->bi_mode) +
1146                 link->dir_count;
1147         int ret = 0;
1148
1149         /*
1150          * These should have been caught/fixed by earlier passes, we don't
1151          * repair them here:
1152          */
1153         if (S_ISDIR(u->bi_mode) && link->count > 1) {
1154                 need_fsck_err(c, "directory %llu with multiple hardlinks: %u",
1155                               u->bi_inum, link->count);
1156                 return 0;
1157         }
1158
1159         if (S_ISDIR(u->bi_mode) && !link->count) {
1160                 need_fsck_err(c, "unreachable directory found (inum %llu)",
1161                               u->bi_inum);
1162                 return 0;
1163         }
1164
1165         if (!S_ISDIR(u->bi_mode) && link->dir_count) {
1166                 need_fsck_err(c, "non directory with subdirectories (inum %llu)",
1167                               u->bi_inum);
1168                 return 0;
1169         }
1170
1171         if (!link->count &&
1172             !(u->bi_flags & BCH_INODE_UNLINKED) &&
1173             (c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1174                 if (fsck_err(c, "unreachable inode %llu not marked as unlinked (type %u)",
1175                              u->bi_inum, mode_to_type(u->bi_mode)) ==
1176                     FSCK_ERR_IGNORE)
1177                         return 0;
1178
1179                 ret = reattach_inode(c, lostfound_inode, u->bi_inum);
1180                 if (ret)
1181                         return ret;
1182
1183                 link->count = 1;
1184                 real_i_nlink = nlink_bias(u->bi_mode) + link->dir_count;
1185                 goto set_i_nlink;
1186         }
1187
1188         if (i_nlink < link->count) {
1189                 if (fsck_err(c, "inode %llu i_link too small (%u < %u, type %i)",
1190                              u->bi_inum, i_nlink, link->count,
1191                              mode_to_type(u->bi_mode)) == FSCK_ERR_IGNORE)
1192                         return 0;
1193                 goto set_i_nlink;
1194         }
1195
1196         if (i_nlink != real_i_nlink &&
1197             c->sb.clean) {
1198                 if (fsck_err(c, "filesystem marked clean, "
1199                              "but inode %llu has wrong i_nlink "
1200                              "(type %u i_nlink %u, should be %u)",
1201                              u->bi_inum, mode_to_type(u->bi_mode),
1202                              i_nlink, real_i_nlink) == FSCK_ERR_IGNORE)
1203                         return 0;
1204                 goto set_i_nlink;
1205         }
1206
1207         if (i_nlink != real_i_nlink &&
1208             (c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1209                 if (fsck_err(c, "inode %llu has wrong i_nlink "
1210                              "(type %u i_nlink %u, should be %u)",
1211                              u->bi_inum, mode_to_type(u->bi_mode),
1212                              i_nlink, real_i_nlink) == FSCK_ERR_IGNORE)
1213                         return 0;
1214                 goto set_i_nlink;
1215         }
1216
1217         if (real_i_nlink && i_nlink != real_i_nlink)
1218                 bch_verbose(c, "setting inode %llu nlink from %u to %u",
1219                             u->bi_inum, i_nlink, real_i_nlink);
1220 set_i_nlink:
1221         if (i_nlink != real_i_nlink) {
1222                 bch2_inode_nlink_set(u, real_i_nlink);
1223                 *do_update = true;
1224         }
1225 fsck_err:
1226         return ret;
1227 }
1228
1229 static int check_inode(struct btree_trans *trans,
1230                        struct bch_inode_unpacked *lostfound_inode,
1231                        struct btree_iter *iter,
1232                        struct bkey_s_c_inode inode,
1233                        struct nlink *link)
1234 {
1235         struct bch_fs *c = trans->c;
1236         struct bch_inode_unpacked u;
1237         bool do_update = false;
1238         int ret = 0;
1239
1240         ret = bch2_inode_unpack(inode, &u);
1241
1242         bch2_trans_unlock(trans);
1243
1244         if (bch2_fs_inconsistent_on(ret, c,
1245                          "error unpacking inode %llu in fsck",
1246                          inode.k->p.inode))
1247                 return ret;
1248
1249         if (link) {
1250                 ret = check_inode_nlink(c, lostfound_inode, &u, link,
1251                                         &do_update);
1252                 if (ret)
1253                         return ret;
1254         }
1255
1256         if (u.bi_flags & BCH_INODE_UNLINKED &&
1257             (!c->sb.clean ||
1258              fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
1259                       u.bi_inum))) {
1260                 bch_verbose(c, "deleting inode %llu", u.bi_inum);
1261
1262                 bch2_fs_lazy_rw(c);
1263
1264                 ret = bch2_inode_rm(c, u.bi_inum, false);
1265                 if (ret)
1266                         bch_err(c, "error in fsck: error %i while deleting inode", ret);
1267                 return ret;
1268         }
1269
1270         if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
1271             (!c->sb.clean ||
1272              fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
1273                       u.bi_inum))) {
1274                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
1275
1276                 bch2_fs_lazy_rw(c);
1277
1278                 /*
1279                  * XXX: need to truncate partial blocks too here - or ideally
1280                  * just switch units to bytes and that issue goes away
1281                  */
1282
1283                 ret = bch2_inode_truncate(c, u.bi_inum, u.bi_size);
1284                 if (ret) {
1285                         bch_err(c, "error in fsck: error %i truncating inode", ret);
1286                         return ret;
1287                 }
1288
1289                 /*
1290                  * We truncated without our normal sector accounting hook, just
1291                  * make sure we recalculate it:
1292                  */
1293                 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
1294
1295                 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
1296                 do_update = true;
1297         }
1298
1299         if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
1300             (!c->sb.clean ||
1301              fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
1302                       u.bi_inum))) {
1303                 s64 sectors;
1304
1305                 bch_verbose(c, "recounting sectors for inode %llu",
1306                             u.bi_inum);
1307
1308                 sectors = bch2_count_inode_sectors(trans, u.bi_inum);
1309                 if (sectors < 0) {
1310                         bch_err(c, "error in fsck: error %i recounting inode sectors",
1311                                 (int) sectors);
1312                         return sectors;
1313                 }
1314
1315                 u.bi_sectors = sectors;
1316                 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
1317                 do_update = true;
1318         }
1319
1320         if (do_update) {
1321                 struct bkey_inode_buf p;
1322
1323                 bch2_inode_pack(c, &p, &u);
1324
1325                 ret = __bch2_trans_do(trans, NULL, NULL,
1326                                       BTREE_INSERT_NOFAIL|
1327                                       BTREE_INSERT_LAZY_RW,
1328                         (bch2_trans_update(trans, iter, &p.inode.k_i, 0), 0));
1329                 if (ret)
1330                         bch_err(c, "error in fsck: error %i "
1331                                 "updating inode", ret);
1332         }
1333 fsck_err:
1334         return ret;
1335 }
1336
1337 noinline_for_stack
1338 static int bch2_gc_walk_inodes(struct bch_fs *c,
1339                                struct bch_inode_unpacked *lostfound_inode,
1340                                nlink_table *links,
1341                                u64 range_start, u64 range_end)
1342 {
1343         struct btree_trans trans;
1344         struct btree_iter *iter;
1345         struct bkey_s_c k;
1346         struct nlink *link, zero_links = { 0, 0 };
1347         struct genradix_iter nlinks_iter;
1348         int ret = 0, ret2 = 0;
1349         u64 nlinks_pos;
1350
1351         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1352
1353         iter = bch2_trans_get_iter(&trans, BTREE_ID_inodes,
1354                                    POS(0, range_start), 0);
1355         nlinks_iter = genradix_iter_init(links, 0);
1356
1357         while ((k = bch2_btree_iter_peek(iter)).k &&
1358                !(ret2 = bkey_err(k)) &&
1359                iter->pos.offset < range_end) {
1360 peek_nlinks:    link = genradix_iter_peek(&nlinks_iter, links);
1361
1362                 if (!link && (!k.k || iter->pos.offset >= range_end))
1363                         break;
1364
1365                 nlinks_pos = range_start + nlinks_iter.pos;
1366
1367                 if (link && nlinks_pos < iter->pos.offset) {
1368                         /* Should have been caught by dirents pass: */
1369                         need_fsck_err_on(link->count, c,
1370                                 "missing inode %llu (nlink %u)",
1371                                 nlinks_pos, link->count);
1372                         genradix_iter_advance(&nlinks_iter, links);
1373                         goto peek_nlinks;
1374                 }
1375
1376                 if (!link || nlinks_pos > iter->pos.offset)
1377                         link = &zero_links;
1378
1379                 if (k.k && k.k->type == KEY_TYPE_inode) {
1380                         ret = check_inode(&trans, lostfound_inode, iter,
1381                                           bkey_s_c_to_inode(k), link);
1382                         BUG_ON(ret == -EINTR);
1383                         if (ret)
1384                                 break;
1385                 } else {
1386                         /* Should have been caught by dirents pass: */
1387                         need_fsck_err_on(link->count, c,
1388                                 "missing inode %llu (nlink %u)",
1389                                 nlinks_pos, link->count);
1390                 }
1391
1392                 if (nlinks_pos == iter->pos.offset)
1393                         genradix_iter_advance(&nlinks_iter, links);
1394
1395                 bch2_btree_iter_next(iter);
1396                 bch2_trans_cond_resched(&trans);
1397         }
1398 fsck_err:
1399         bch2_trans_exit(&trans);
1400
1401         if (ret2)
1402                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret2);
1403
1404         return ret ?: ret2;
1405 }
1406
1407 noinline_for_stack
1408 static int check_inode_nlinks(struct bch_fs *c,
1409                               struct bch_inode_unpacked *lostfound_inode)
1410 {
1411         nlink_table links;
1412         u64 this_iter_range_start, next_iter_range_start = 0;
1413         int ret = 0;
1414
1415         bch_verbose(c, "checking inode nlinks");
1416
1417         genradix_init(&links);
1418
1419         do {
1420                 this_iter_range_start = next_iter_range_start;
1421                 next_iter_range_start = U64_MAX;
1422
1423                 ret = bch2_gc_walk_dirents(c, &links,
1424                                           this_iter_range_start,
1425                                           &next_iter_range_start);
1426                 if (ret)
1427                         break;
1428
1429                 ret = bch2_gc_walk_inodes(c, lostfound_inode, &links,
1430                                          this_iter_range_start,
1431                                          next_iter_range_start);
1432                 if (ret)
1433                         break;
1434
1435                 genradix_free(&links);
1436         } while (next_iter_range_start != U64_MAX);
1437
1438         genradix_free(&links);
1439
1440         return ret;
1441 }
1442
1443 /*
1444  * Checks for inconsistencies that shouldn't happen, unless we have a bug.
1445  * Doesn't fix them yet, mainly because they haven't yet been observed:
1446  */
1447 int bch2_fsck_full(struct bch_fs *c)
1448 {
1449         struct bch_inode_unpacked root_inode, lostfound_inode;
1450
1451         return  check_extents(c) ?:
1452                 check_dirents(c) ?:
1453                 check_xattrs(c) ?:
1454                 check_root(c, &root_inode) ?:
1455                 check_lostfound(c, &root_inode, &lostfound_inode) ?:
1456                 check_directory_structure(c, &lostfound_inode) ?:
1457                 check_inode_nlinks(c, &lostfound_inode);
1458 }
1459
1460 int bch2_fsck_inode_nlink(struct bch_fs *c)
1461 {
1462         struct bch_inode_unpacked root_inode, lostfound_inode;
1463
1464         return  check_root(c, &root_inode) ?:
1465                 check_lostfound(c, &root_inode, &lostfound_inode) ?:
1466                 check_inode_nlinks(c, &lostfound_inode);
1467 }
1468
1469 int bch2_fsck_walk_inodes_only(struct bch_fs *c)
1470 {
1471         struct btree_trans trans;
1472         struct btree_iter *iter;
1473         struct bkey_s_c k;
1474         struct bkey_s_c_inode inode;
1475         int ret;
1476
1477         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1478
1479         for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN, 0, k, ret) {
1480                 if (k.k->type != KEY_TYPE_inode)
1481                         continue;
1482
1483                 inode = bkey_s_c_to_inode(k);
1484
1485                 if (inode.v->bi_flags &
1486                     (BCH_INODE_I_SIZE_DIRTY|
1487                      BCH_INODE_I_SECTORS_DIRTY|
1488                      BCH_INODE_UNLINKED)) {
1489                         ret = check_inode(&trans, NULL, iter, inode, NULL);
1490                         BUG_ON(ret == -EINTR);
1491                         if (ret)
1492                                 break;
1493                 }
1494         }
1495         BUG_ON(ret == -EINTR);
1496
1497         return bch2_trans_exit(&trans) ?: ret;
1498 }