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 "btree_update.h"
   5 #include "dirent.h"
   6 #include "error.h"
   7 #include "fs-common.h"
   8 #include "fsck.h"
   9 #include "inode.h"
  10 #include "keylist.h"
  11 #include "super.h"
  12 #include "xattr.h"
  13
  14 #include <linux/dcache.h> /* struct qstr */
  15 #include <linux/generic-radix-tree.h>
  16
  17 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  18
  19 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum)
  20 {
  21         struct btree_iter *iter;
  22         struct bkey_s_c k;
  23         u64 sectors = 0;
  24         int ret;
  25
  26         for_each_btree_key(trans, iter, BTREE_ID_EXTENTS,
  27                            POS(inum, 0), 0, k, ret) {
  28                 if (k.k->p.inode != inum)
  29                         break;
  30
  31                 if (bkey_extent_is_allocation(k.k))
  32                         sectors += k.k->size;
  33         }
  34
  35         bch2_trans_iter_free(trans, iter);
  36
  37         return ret ?: sectors;
  38 }
  39
  40 static int remove_dirent(struct btree_trans *trans,
  41                          struct bkey_s_c_dirent dirent)
  42 {
  43         struct bch_fs *c = trans->c;
  44         struct qstr name;
  45         struct bch_inode_unpacked dir_inode;
  46         struct bch_hash_info dir_hash_info;
  47         u64 dir_inum = dirent.k->p.inode;
  48         int ret;
  49         char *buf;
  50
  51         name.len = bch2_dirent_name_bytes(dirent);
  52         buf = kmalloc(name.len + 1, GFP_KERNEL);
  53         if (!buf)
  54                 return -ENOMEM;
  55
  56         memcpy(buf, dirent.v->d_name, name.len);
  57         buf[name.len] = '\0';
  58         name.name = buf;
  59
  60         /* Unlock so we don't deadlock, after copying name: */
  61         bch2_trans_unlock(trans);
  62
  63         ret = bch2_inode_find_by_inum(c, dir_inum, &dir_inode);
  64         if (ret) {
  65                 bch_err(c, "remove_dirent: err %i looking up directory inode", ret);
  66                 goto err;
  67         }
  68
  69         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
  70
  71         ret = bch2_dirent_delete(c, dir_inum, &dir_hash_info, &name, NULL);
  72         if (ret)
  73                 bch_err(c, "remove_dirent: err %i deleting dirent", ret);
  74 err:
  75         kfree(buf);
  76         return ret;
  77 }
  78
  79 static int reattach_inode(struct bch_fs *c,
  80                           struct bch_inode_unpacked *lostfound_inode,
  81                           u64 inum)
  82 {
  83         struct bch_inode_unpacked inode_u;
  84         char name_buf[20];
  85         struct qstr name;
  86         int ret;
  87
  88         snprintf(name_buf, sizeof(name_buf), "%llu", inum);
  89         name = (struct qstr) QSTR(name_buf);
  90
  91         ret = bch2_trans_do(c, NULL,
  92                             BTREE_INSERT_ATOMIC|
  93                             BTREE_INSERT_LAZY_RW,
  94                 bch2_link_trans(&trans, lostfound_inode->bi_inum,
  95                                 inum, &inode_u, &name));
  96         if (ret)
  97                 bch_err(c, "error %i reattaching inode %llu", ret, inum);
  98
  99         return ret;
 100 }
 101
 102 struct inode_walker {
 103         bool                    first_this_inode;
 104         bool                    have_inode;
 105         u64                     cur_inum;
 106         struct bch_inode_unpacked inode;
 107 };
 108
 109 static struct inode_walker inode_walker_init(void)
 110 {
 111         return (struct inode_walker) {
 112                 .cur_inum       = -1,
 113                 .have_inode     = false,
 114         };
 115 }
 116
 117 static int walk_inode(struct btree_trans *trans,
 118                       struct inode_walker *w, u64 inum)
 119 {
 120         if (inum != w->cur_inum) {
 121                 int ret = bch2_inode_find_by_inum_trans(trans, inum,
 122                                                         &w->inode);
 123
 124                 if (ret && ret != -ENOENT)
 125                         return ret;
 126
 127                 w->have_inode   = !ret;
 128                 w->cur_inum     = inum;
 129                 w->first_this_inode = true;
 130         } else {
 131                 w->first_this_inode = false;
 132         }
 133
 134         return 0;
 135 }
 136
 137 struct hash_check {
 138         struct bch_hash_info    info;
 139
 140         /* start of current chain of hash collisions: */
 141         struct btree_iter       *chain;
 142
 143         /* next offset in current chain of hash collisions: */
 144         u64                     chain_end;
 145 };
 146
 147 static void hash_check_init(struct hash_check *h)
 148 {
 149         h->chain = NULL;
 150         h->chain_end = 0;
 151 }
 152
 153 static void hash_stop_chain(struct btree_trans *trans,
 154                             struct hash_check *h)
 155 {
 156         if (h->chain)
 157                 bch2_trans_iter_free(trans, h->chain);
 158         h->chain = NULL;
 159 }
 160
 161 static void hash_check_set_inode(struct btree_trans *trans,
 162                                  struct hash_check *h,
 163                                  const struct bch_inode_unpacked *bi)
 164 {
 165         h->info = bch2_hash_info_init(trans->c, bi);
 166         hash_stop_chain(trans, h);
 167 }
 168
 169 static int hash_redo_key(const struct bch_hash_desc desc,
 170                          struct btree_trans *trans, struct hash_check *h,
 171                          struct btree_iter *k_iter, struct bkey_s_c k,
 172                          u64 hashed)
 173 {
 174         struct bkey_i *tmp;
 175         int ret = 0;
 176
 177         tmp = kmalloc(bkey_bytes(k.k), GFP_KERNEL);
 178         if (!tmp)
 179                 return -ENOMEM;
 180
 181         bkey_reassemble(tmp, k);
 182
 183         ret = bch2_btree_delete_at(trans, k_iter, 0);
 184         if (ret)
 185                 goto err;
 186
 187         bch2_hash_set(trans, desc, &h->info, k_iter->pos.inode,
 188                       tmp, BCH_HASH_SET_MUST_CREATE);
 189         ret = bch2_trans_commit(trans, NULL, NULL,
 190                                 BTREE_INSERT_NOFAIL|
 191                                 BTREE_INSERT_LAZY_RW);
 192 err:
 193         kfree(tmp);
 194         return ret;
 195 }
 196
 197 static int fsck_hash_delete_at(struct btree_trans *trans,
 198                                const struct bch_hash_desc desc,
 199                                struct bch_hash_info *info,
 200                                struct btree_iter *iter)
 201 {
 202         int ret;
 203 retry:
 204         ret   = bch2_hash_delete_at(trans, desc, info, iter) ?:
 205                 bch2_trans_commit(trans, NULL, NULL,
 206                                   BTREE_INSERT_ATOMIC|
 207                                   BTREE_INSERT_NOFAIL|
 208                                   BTREE_INSERT_LAZY_RW);
 209         if (ret == -EINTR) {
 210                 ret = bch2_btree_iter_traverse(iter);
 211                 if (!ret)
 212                         goto retry;
 213         }
 214
 215         return ret;
 216 }
 217
 218 static int hash_check_duplicates(struct btree_trans *trans,
 219                         const struct bch_hash_desc desc, struct hash_check *h,
 220                         struct btree_iter *k_iter, struct bkey_s_c k)
 221 {
 222         struct bch_fs *c = trans->c;
 223         struct btree_iter *iter;
 224         struct bkey_s_c k2;
 225         char buf[200];
 226         int ret = 0;
 227
 228         if (!bkey_cmp(h->chain->pos, k_iter->pos))
 229                 return 0;
 230
 231         iter = bch2_trans_copy_iter(trans, h->chain);
 232         BUG_ON(IS_ERR(iter));
 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_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                         BUG_ON(IS_ERR(h->chain));
 270                 }
 271
 272                 h->chain_end = k.k->p.offset;
 273         }
 274 }
 275
 276 static bool key_has_correct_hash(struct btree_trans *trans,
 277                         const struct bch_hash_desc desc,
 278                         struct hash_check *h,
 279                         struct btree_iter *k_iter, struct bkey_s_c k)
 280 {
 281         u64 hash;
 282
 283         hash_set_chain_start(trans, desc, h, k_iter, k);
 284
 285         if (k.k->type != desc.key_type)
 286                 return true;
 287
 288         hash = desc.hash_bkey(&h->info, k);
 289
 290         return hash >= h->chain->pos.offset &&
 291                 hash <= k.k->p.offset;
 292 }
 293
 294 static int hash_check_key(struct btree_trans *trans,
 295                         const struct bch_hash_desc desc, struct hash_check *h,
 296                         struct btree_iter *k_iter, struct bkey_s_c k)
 297 {
 298         struct bch_fs *c = trans->c;
 299         char buf[200];
 300         u64 hashed;
 301         int ret = 0;
 302
 303         hash_set_chain_start(trans, desc, h, k_iter, k);
 304
 305         if (k.k->type != desc.key_type)
 306                 return 0;
 307
 308         hashed = desc.hash_bkey(&h->info, k);
 309
 310         if (fsck_err_on(hashed < h->chain->pos.offset ||
 311                         hashed > k.k->p.offset, c,
 312                         "hash table key at wrong offset: btree %u, %llu, "
 313                         "hashed to %llu chain starts at %llu\n%s",
 314                         desc.btree_id, k.k->p.offset,
 315                         hashed, h->chain->pos.offset,
 316                         (bch2_bkey_val_to_text(&PBUF(buf), c,
 317                                                k), buf))) {
 318                 ret = hash_redo_key(desc, trans, h, k_iter, k, hashed);
 319                 if (ret) {
 320                         bch_err(c, "hash_redo_key err %i", ret);
 321                         return ret;
 322                 }
 323                 return 1;
 324         }
 325
 326         ret = hash_check_duplicates(trans, desc, h, k_iter, k);
 327 fsck_err:
 328         return ret;
 329 }
 330
 331 static int check_dirent_hash(struct btree_trans *trans, struct hash_check *h,
 332                              struct btree_iter *iter, struct bkey_s_c *k)
 333 {
 334         struct bch_fs *c = trans->c;
 335         struct bkey_i_dirent *d = NULL;
 336         int ret = -EINVAL;
 337         char buf[200];
 338         unsigned len;
 339         u64 hash;
 340
 341         if (key_has_correct_hash(trans, bch2_dirent_hash_desc, h, iter, *k))
 342                 return 0;
 343
 344         len = bch2_dirent_name_bytes(bkey_s_c_to_dirent(*k));
 345         BUG_ON(!len);
 346
 347         memcpy(buf, bkey_s_c_to_dirent(*k).v->d_name, len);
 348         buf[len] = '\0';
 349
 350         d = kmalloc(bkey_bytes(k->k), GFP_KERNEL);
 351         if (!d) {
 352                 bch_err(c, "memory allocation failure");
 353                 return -ENOMEM;
 354         }
 355
 356         bkey_reassemble(&d->k_i, *k);
 357
 358         do {
 359                 --len;
 360                 if (!len)
 361                         goto err_redo;
 362
 363                 d->k.u64s = BKEY_U64s + dirent_val_u64s(len);
 364
 365                 BUG_ON(bkey_val_bytes(&d->k) <
 366                        offsetof(struct bch_dirent, d_name) + len);
 367
 368                 memset(d->v.d_name + len, 0,
 369                        bkey_val_bytes(&d->k) -
 370                        offsetof(struct bch_dirent, d_name) - len);
 371
 372                 hash = bch2_dirent_hash_desc.hash_bkey(&h->info,
 373                                                 bkey_i_to_s_c(&d->k_i));
 374         } while (hash < h->chain->pos.offset ||
 375                  hash > k->k->p.offset);
 376
 377         if (fsck_err(c, "dirent with junk at end, was %s (%zu) now %s (%u)",
 378                      buf, strlen(buf), d->v.d_name, len)) {
 379                 bch2_trans_update(trans, iter, &d->k_i);
 380
 381                 ret = bch2_trans_commit(trans, NULL, NULL,
 382                                         BTREE_INSERT_NOFAIL|
 383                                         BTREE_INSERT_LAZY_RW);
 384                 if (ret)
 385                         goto err;
 386
 387                 *k = bch2_btree_iter_peek(iter);
 388
 389                 BUG_ON(k->k->type != KEY_TYPE_dirent);
 390         }
 391 err:
 392 fsck_err:
 393         kfree(d);
 394         return ret;
 395 err_redo:
 396         hash = bch2_dirent_hash_desc.hash_bkey(&h->info, *k);
 397
 398         if (fsck_err(c, "cannot fix dirent by removing trailing garbage %s (%zu)\n"
 399                      "hash table key at wrong offset: btree %u, offset %llu, "
 400                      "hashed to %llu chain starts at %llu\n%s",
 401                      buf, strlen(buf), BTREE_ID_DIRENTS,
 402                      k->k->p.offset, hash, h->chain->pos.offset,
 403                      (bch2_bkey_val_to_text(&PBUF(buf), c,
 404                                             *k), buf))) {
 405                 ret = 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 /*
 424  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
 425  * that i_size an i_sectors are consistent
 426  */
 427 noinline_for_stack
 428 static int check_extents(struct bch_fs *c)
 429 {
 430         struct inode_walker w = inode_walker_init();
 431         struct btree_trans trans;
 432         struct btree_iter *iter;
 433         struct bkey_s_c k;
 434         u64 i_sectors;
 435         int ret = 0;
 436
 437         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 438
 439         bch_verbose(c, "checking extents");
 440
 441         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 442                                    POS(BCACHEFS_ROOT_INO, 0), 0);
 443 retry:
 444         for_each_btree_key_continue(iter, 0, k, ret) {
 445                 ret = walk_inode(&trans, &w, k.k->p.inode);
 446                 if (ret)
 447                         break;
 448
 449                 if (fsck_err_on(!w.have_inode, c,
 450                         "extent type %u for missing inode %llu",
 451                         k.k->type, k.k->p.inode) ||
 452                     fsck_err_on(w.have_inode &&
 453                         !S_ISREG(w.inode.bi_mode) && !S_ISLNK(w.inode.bi_mode), c,
 454                         "extent type %u for non regular file, inode %llu mode %o",
 455                         k.k->type, k.k->p.inode, w.inode.bi_mode)) {
 456                         bch2_trans_unlock(&trans);
 457
 458                         ret = bch2_inode_truncate(c, k.k->p.inode, 0);
 459                         if (ret)
 460                                 goto err;
 461                         continue;
 462                 }
 463
 464                 if (fsck_err_on(w.first_this_inode &&
 465                         w.have_inode &&
 466                         !(w.inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY) &&
 467                         w.inode.bi_sectors !=
 468                         (i_sectors = bch2_count_inode_sectors(&trans, w.cur_inum)),
 469                         c, "i_sectors wrong: got %llu, should be %llu",
 470                         w.inode.bi_sectors, i_sectors)) {
 471                         struct bkey_inode_buf p;
 472
 473                         w.inode.bi_sectors = i_sectors;
 474
 475                         bch2_trans_unlock(&trans);
 476
 477                         bch2_inode_pack(&p, &w.inode);
 478
 479                         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 480                                                 &p.inode.k_i, NULL, NULL,
 481                                                 BTREE_INSERT_NOFAIL|
 482                                                 BTREE_INSERT_LAZY_RW);
 483                         if (ret) {
 484                                 bch_err(c, "error in fsck: error %i updating inode", ret);
 485                                 goto err;
 486                         }
 487
 488                         /* revalidate iterator: */
 489                         k = bch2_btree_iter_peek(iter);
 490                 }
 491
 492                 if (fsck_err_on(w.have_inode &&
 493                         !(w.inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
 494                         k.k->type != KEY_TYPE_reservation &&
 495                         k.k->p.offset > round_up(w.inode.bi_size, block_bytes(c)) >> 9, c,
 496                         "extent type %u offset %llu past end of inode %llu, i_size %llu",
 497                         k.k->type, k.k->p.offset, k.k->p.inode, w.inode.bi_size)) {
 498                         bch2_trans_unlock(&trans);
 499
 500                         ret = bch2_inode_truncate(c, k.k->p.inode,
 501                                                   w.inode.bi_size);
 502                         if (ret)
 503                                 goto err;
 504                         continue;
 505                 }
 506         }
 507 err:
 508 fsck_err:
 509         if (ret == -EINTR)
 510                 goto retry;
 511         return bch2_trans_exit(&trans) ?: ret;
 512 }
 513
 514 /*
 515  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
 516  * validate d_type
 517  */
 518 noinline_for_stack
 519 static int check_dirents(struct bch_fs *c)
 520 {
 521         struct inode_walker w = inode_walker_init();
 522         struct hash_check h;
 523         struct btree_trans trans;
 524         struct btree_iter *iter;
 525         struct bkey_s_c k;
 526         unsigned name_len;
 527         char buf[200];
 528         int ret = 0;
 529
 530         bch_verbose(c, "checking dirents");
 531
 532         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 533
 534         hash_check_init(&h);
 535
 536         iter = bch2_trans_get_iter(&trans, BTREE_ID_DIRENTS,
 537                                    POS(BCACHEFS_ROOT_INO, 0), 0);
 538 retry:
 539         for_each_btree_key_continue(iter, 0, k, ret) {
 540                 struct bkey_s_c_dirent d;
 541                 struct bch_inode_unpacked target;
 542                 bool have_target;
 543                 u64 d_inum;
 544
 545                 ret = walk_inode(&trans, &w, k.k->p.inode);
 546                 if (ret)
 547                         break;
 548
 549                 if (fsck_err_on(!w.have_inode, c,
 550                                 "dirent in nonexisting directory:\n%s",
 551                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 552                                                        k), buf)) ||
 553                     fsck_err_on(!S_ISDIR(w.inode.bi_mode), c,
 554                                 "dirent in non directory inode type %u:\n%s",
 555                                 mode_to_type(w.inode.bi_mode),
 556                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 557                                                        k), buf))) {
 558                         ret = bch2_btree_delete_at(&trans, iter, 0);
 559                         if (ret)
 560                                 goto err;
 561                         continue;
 562                 }
 563
 564                 if (w.first_this_inode && w.have_inode)
 565                         hash_check_set_inode(&trans, &h, &w.inode);
 566
 567                 ret = check_dirent_hash(&trans, &h, iter, &k);
 568                 if (ret > 0) {
 569                         ret = 0;
 570                         continue;
 571                 }
 572                 if (ret)
 573                         goto fsck_err;
 574
 575                 if (ret)
 576                         goto fsck_err;
 577
 578                 if (k.k->type != KEY_TYPE_dirent)
 579                         continue;
 580
 581                 d = bkey_s_c_to_dirent(k);
 582                 d_inum = le64_to_cpu(d.v->d_inum);
 583
 584                 name_len = bch2_dirent_name_bytes(d);
 585
 586                 if (fsck_err_on(!name_len, c, "empty dirent") ||
 587                     fsck_err_on(name_len == 1 &&
 588                                 !memcmp(d.v->d_name, ".", 1), c,
 589                                 ". dirent") ||
 590                     fsck_err_on(name_len == 2 &&
 591                                 !memcmp(d.v->d_name, "..", 2), c,
 592                                 ".. dirent") ||
 593                     fsck_err_on(name_len == 2 &&
 594                                 !memcmp(d.v->d_name, "..", 2), c,
 595                                 ".. dirent") ||
 596                     fsck_err_on(memchr(d.v->d_name, '/', name_len), c,
 597                                 "dirent name has invalid chars")) {
 598                         ret = remove_dirent(&trans, d);
 599                         if (ret)
 600                                 goto err;
 601                         continue;
 602                 }
 603
 604                 if (fsck_err_on(d_inum == d.k->p.inode, c,
 605                                 "dirent points to own directory:\n%s",
 606                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 607                                                        k), buf))) {
 608                         ret = remove_dirent(&trans, d);
 609                         if (ret)
 610                                 goto err;
 611                         continue;
 612                 }
 613
 614                 ret = bch2_inode_find_by_inum_trans(&trans, d_inum, &target);
 615                 if (ret && ret != -ENOENT)
 616                         break;
 617
 618                 have_target = !ret;
 619                 ret = 0;
 620
 621                 if (fsck_err_on(!have_target, c,
 622                                 "dirent points to missing inode:\n%s",
 623                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 624                                                        k), buf))) {
 625                         ret = remove_dirent(&trans, d);
 626                         if (ret)
 627                                 goto err;
 628                         continue;
 629                 }
 630
 631                 if (fsck_err_on(have_target &&
 632                                 d.v->d_type !=
 633                                 mode_to_type(target.bi_mode), c,
 634                                 "incorrect d_type: should be %u:\n%s",
 635                                 mode_to_type(target.bi_mode),
 636                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 637                                                        k), buf))) {
 638                         struct bkey_i_dirent *n;
 639
 640                         n = kmalloc(bkey_bytes(d.k), GFP_KERNEL);
 641                         if (!n) {
 642                                 ret = -ENOMEM;
 643                                 goto err;
 644                         }
 645
 646                         bkey_reassemble(&n->k_i, d.s_c);
 647                         n->v.d_type = mode_to_type(target.bi_mode);
 648
 649                         bch2_trans_update(&trans, iter, &n->k_i);
 650
 651                         ret = bch2_trans_commit(&trans, NULL, NULL,
 652                                                 BTREE_INSERT_NOFAIL|
 653                                                 BTREE_INSERT_LAZY_RW);
 654                         kfree(n);
 655                         if (ret)
 656                                 goto err;
 657
 658                 }
 659         }
 660
 661         hash_stop_chain(&trans, &h);
 662 err:
 663 fsck_err:
 664         if (ret == -EINTR)
 665                 goto retry;
 666
 667         return bch2_trans_exit(&trans) ?: ret;
 668 }
 669
 670 /*
 671  * Walk xattrs: verify that they all have a corresponding inode
 672  */
 673 noinline_for_stack
 674 static int check_xattrs(struct bch_fs *c)
 675 {
 676         struct inode_walker w = inode_walker_init();
 677         struct hash_check h;
 678         struct btree_trans trans;
 679         struct btree_iter *iter;
 680         struct bkey_s_c k;
 681         int ret = 0;
 682
 683         bch_verbose(c, "checking xattrs");
 684
 685         hash_check_init(&h);
 686
 687         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 688
 689         iter = bch2_trans_get_iter(&trans, BTREE_ID_XATTRS,
 690                                    POS(BCACHEFS_ROOT_INO, 0), 0);
 691 retry:
 692         for_each_btree_key_continue(iter, 0, k, ret) {
 693                 ret = walk_inode(&trans, &w, k.k->p.inode);
 694                 if (ret)
 695                         break;
 696
 697                 if (fsck_err_on(!w.have_inode, c,
 698                                 "xattr for missing inode %llu",
 699                                 k.k->p.inode)) {
 700                         ret = bch2_btree_delete_at(&trans, iter, 0);
 701                         if (ret)
 702                                 goto err;
 703                         continue;
 704                 }
 705
 706                 if (w.first_this_inode && w.have_inode)
 707                         hash_check_set_inode(&trans, &h, &w.inode);
 708
 709                 ret = hash_check_key(&trans, bch2_xattr_hash_desc,
 710                                      &h, iter, k);
 711                 if (ret)
 712                         goto fsck_err;
 713         }
 714 err:
 715 fsck_err:
 716         if (ret == -EINTR)
 717                 goto retry;
 718         return bch2_trans_exit(&trans) ?: ret;
 719 }
 720
 721 /* Get root directory, create if it doesn't exist: */
 722 static int check_root(struct bch_fs *c, struct bch_inode_unpacked *root_inode)
 723 {
 724         struct bkey_inode_buf packed;
 725         int ret;
 726
 727         bch_verbose(c, "checking root directory");
 728
 729         ret = bch2_inode_find_by_inum(c, BCACHEFS_ROOT_INO, root_inode);
 730         if (ret && ret != -ENOENT)
 731                 return ret;
 732
 733         if (fsck_err_on(ret, c, "root directory missing"))
 734                 goto create_root;
 735
 736         if (fsck_err_on(!S_ISDIR(root_inode->bi_mode), c,
 737                         "root inode not a directory"))
 738                 goto create_root;
 739
 740         return 0;
 741 fsck_err:
 742         return ret;
 743 create_root:
 744         bch2_inode_init(c, root_inode, 0, 0, S_IFDIR|0755,
 745                         0, NULL);
 746         root_inode->bi_inum = BCACHEFS_ROOT_INO;
 747
 748         bch2_inode_pack(&packed, root_inode);
 749
 750         return bch2_btree_insert(c, BTREE_ID_INODES, &packed.inode.k_i,
 751                                  NULL, NULL,
 752                                  BTREE_INSERT_NOFAIL|
 753                                  BTREE_INSERT_LAZY_RW);
 754 }
 755
 756 /* Get lost+found, create if it doesn't exist: */
 757 static int check_lostfound(struct bch_fs *c,
 758                            struct bch_inode_unpacked *root_inode,
 759                            struct bch_inode_unpacked *lostfound_inode)
 760 {
 761         struct qstr lostfound = QSTR("lost+found");
 762         struct bch_hash_info root_hash_info =
 763                 bch2_hash_info_init(c, root_inode);
 764         u64 inum;
 765         int ret;
 766
 767         bch_verbose(c, "checking lost+found");
 768
 769         inum = bch2_dirent_lookup(c, BCACHEFS_ROOT_INO, &root_hash_info,
 770                                  &lostfound);
 771         if (!inum) {
 772                 bch_notice(c, "creating lost+found");
 773                 goto create_lostfound;
 774         }
 775
 776         ret = bch2_inode_find_by_inum(c, inum, lostfound_inode);
 777         if (ret && ret != -ENOENT)
 778                 return ret;
 779
 780         if (fsck_err_on(ret, c, "lost+found missing"))
 781                 goto create_lostfound;
 782
 783         if (fsck_err_on(!S_ISDIR(lostfound_inode->bi_mode), c,
 784                         "lost+found inode not a directory"))
 785                 goto create_lostfound;
 786
 787         return 0;
 788 fsck_err:
 789         return ret;
 790 create_lostfound:
 791         bch2_inode_init_early(c, lostfound_inode);
 792
 793         ret = bch2_trans_do(c, NULL,
 794                             BTREE_INSERT_ATOMIC|
 795                             BTREE_INSERT_NOFAIL|
 796                             BTREE_INSERT_LAZY_RW,
 797                 bch2_create_trans(&trans,
 798                                   BCACHEFS_ROOT_INO, root_inode,
 799                                   lostfound_inode, &lostfound,
 800                                   0, 0, S_IFDIR|0755, 0, NULL, NULL));
 801         if (ret)
 802                 bch_err(c, "error creating lost+found: %i", ret);
 803
 804         return ret;
 805 }
 806
 807 struct inode_bitmap {
 808         unsigned long   *bits;
 809         size_t          size;
 810 };
 811
 812 static inline bool inode_bitmap_test(struct inode_bitmap *b, size_t nr)
 813 {
 814         return nr < b->size ? test_bit(nr, b->bits) : false;
 815 }
 816
 817 static inline int inode_bitmap_set(struct inode_bitmap *b, size_t nr)
 818 {
 819         if (nr >= b->size) {
 820                 size_t new_size = max_t(size_t, max_t(size_t,
 821                                         PAGE_SIZE * 8,
 822                                         b->size * 2),
 823                                         nr + 1);
 824                 void *n;
 825
 826                 new_size = roundup_pow_of_two(new_size);
 827                 n = krealloc(b->bits, new_size / 8, GFP_KERNEL|__GFP_ZERO);
 828                 if (!n) {
 829                         return -ENOMEM;
 830                 }
 831
 832                 b->bits = n;
 833                 b->size = new_size;
 834         }
 835
 836         __set_bit(nr, b->bits);
 837         return 0;
 838 }
 839
 840 struct pathbuf {
 841         size_t          nr;
 842         size_t          size;
 843
 844         struct pathbuf_entry {
 845                 u64     inum;
 846                 u64     offset;
 847         }               *entries;
 848 };
 849
 850 static int path_down(struct pathbuf *p, u64 inum)
 851 {
 852         if (p->nr == p->size) {
 853                 size_t new_size = max_t(size_t, 256UL, p->size * 2);
 854                 void *n = krealloc(p->entries,
 855                                    new_size * sizeof(p->entries[0]),
 856                                    GFP_KERNEL);
 857                 if (!n)
 858                         return -ENOMEM;
 859
 860                 p->entries = n;
 861                 p->size = new_size;
 862         };
 863
 864         p->entries[p->nr++] = (struct pathbuf_entry) {
 865                 .inum = inum,
 866                 .offset = 0,
 867         };
 868         return 0;
 869 }
 870
 871 noinline_for_stack
 872 static int check_directory_structure(struct bch_fs *c,
 873                                      struct bch_inode_unpacked *lostfound_inode)
 874 {
 875         struct inode_bitmap dirs_done = { NULL, 0 };
 876         struct pathbuf path = { 0, 0, NULL };
 877         struct pathbuf_entry *e;
 878         struct btree_trans trans;
 879         struct btree_iter *iter;
 880         struct bkey_s_c k;
 881         struct bkey_s_c_dirent dirent;
 882         bool had_unreachable;
 883         u64 d_inum;
 884         int ret = 0;
 885
 886         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 887
 888         bch_verbose(c, "checking directory structure");
 889
 890         /* DFS: */
 891 restart_dfs:
 892         had_unreachable = false;
 893
 894         ret = inode_bitmap_set(&dirs_done, BCACHEFS_ROOT_INO);
 895         if (ret) {
 896                 bch_err(c, "memory allocation failure in inode_bitmap_set()");
 897                 goto err;
 898         }
 899
 900         ret = path_down(&path, BCACHEFS_ROOT_INO);
 901         if (ret)
 902                 goto err;
 903
 904         while (path.nr) {
 905 next:
 906                 e = &path.entries[path.nr - 1];
 907
 908                 if (e->offset == U64_MAX)
 909                         goto up;
 910
 911                 for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS,
 912                                    POS(e->inum, e->offset + 1), 0, k, ret) {
 913                         if (k.k->p.inode != e->inum)
 914                                 break;
 915
 916                         e->offset = k.k->p.offset;
 917
 918                         if (k.k->type != KEY_TYPE_dirent)
 919                                 continue;
 920
 921                         dirent = bkey_s_c_to_dirent(k);
 922
 923                         if (dirent.v->d_type != DT_DIR)
 924                                 continue;
 925
 926                         d_inum = le64_to_cpu(dirent.v->d_inum);
 927
 928                         if (fsck_err_on(inode_bitmap_test(&dirs_done, d_inum), c,
 929                                         "directory %llu has multiple hardlinks",
 930                                         d_inum)) {
 931                                 ret = remove_dirent(&trans, dirent);
 932                                 if (ret)
 933                                         goto err;
 934                                 continue;
 935                         }
 936
 937                         ret = inode_bitmap_set(&dirs_done, d_inum);
 938                         if (ret) {
 939                                 bch_err(c, "memory allocation failure in inode_bitmap_set()");
 940                                 goto err;
 941                         }
 942
 943                         ret = path_down(&path, d_inum);
 944                         if (ret) {
 945                                 goto err;
 946                         }
 947
 948                         ret = bch2_trans_iter_free(&trans, iter);
 949                         if (ret) {
 950                                 bch_err(c, "btree error %i in fsck", ret);
 951                                 goto err;
 952                         }
 953                         goto next;
 954                 }
 955                 ret = bch2_trans_iter_free(&trans, iter) ?: ret;
 956                 if (ret) {
 957                         bch_err(c, "btree error %i in fsck", ret);
 958                         goto err;
 959                 }
 960 up:
 961                 path.nr--;
 962         }
 963
 964         iter = bch2_trans_get_iter(&trans, BTREE_ID_INODES, POS_MIN, 0);
 965 retry:
 966         for_each_btree_key_continue(iter, 0, k, ret) {
 967                 if (k.k->type != KEY_TYPE_inode)
 968                         continue;
 969
 970                 if (!S_ISDIR(le16_to_cpu(bkey_s_c_to_inode(k).v->bi_mode)))
 971                         continue;
 972
 973                 ret = bch2_empty_dir_trans(&trans, k.k->p.inode);
 974                 if (ret == -EINTR)
 975                         goto retry;
 976                 if (!ret)
 977                         continue;
 978
 979                 if (fsck_err_on(!inode_bitmap_test(&dirs_done, k.k->p.inode), c,
 980                                 "unreachable directory found (inum %llu)",
 981                                 k.k->p.inode)) {
 982                         bch2_trans_unlock(&trans);
 983
 984                         ret = reattach_inode(c, lostfound_inode, k.k->p.inode);
 985                         if (ret) {
 986                                 goto err;
 987                         }
 988
 989                         had_unreachable = true;
 990                 }
 991         }
 992         bch2_trans_iter_free(&trans, iter);
 993         if (ret)
 994                 goto err;
 995
 996         if (had_unreachable) {
 997                 bch_info(c, "reattached unreachable directories, restarting pass to check for loops");
 998                 kfree(dirs_done.bits);
 999                 kfree(path.entries);
1000                 memset(&dirs_done, 0, sizeof(dirs_done));
1001                 memset(&path, 0, sizeof(path));
1002                 goto restart_dfs;
1003         }
1004 err:
1005 fsck_err:
1006         ret = bch2_trans_exit(&trans) ?: ret;
1007         kfree(dirs_done.bits);
1008         kfree(path.entries);
1009         return ret;
1010 }
1011
1012 struct nlink {
1013         u32     count;
1014         u32     dir_count;
1015 };
1016
1017 typedef GENRADIX(struct nlink) nlink_table;
1018
1019 static void inc_link(struct bch_fs *c, nlink_table *links,
1020                      u64 range_start, u64 *range_end,
1021                      u64 inum, bool dir)
1022 {
1023         struct nlink *link;
1024
1025         if (inum < range_start || inum >= *range_end)
1026                 return;
1027
1028         link = genradix_ptr_alloc(links, inum - range_start, GFP_KERNEL);
1029         if (!link) {
1030                 bch_verbose(c, "allocation failed during fsck - will need another pass");
1031                 *range_end = inum;
1032                 return;
1033         }
1034
1035         if (dir)
1036                 link->dir_count++;
1037         else
1038                 link->count++;
1039 }
1040
1041 noinline_for_stack
1042 static int bch2_gc_walk_dirents(struct bch_fs *c, nlink_table *links,
1043                                u64 range_start, u64 *range_end)
1044 {
1045         struct btree_trans trans;
1046         struct btree_iter *iter;
1047         struct bkey_s_c k;
1048         struct bkey_s_c_dirent d;
1049         u64 d_inum;
1050         int ret;
1051
1052         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1053
1054         inc_link(c, links, range_start, range_end, BCACHEFS_ROOT_INO, false);
1055
1056         for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS, POS_MIN, 0, k, ret) {
1057                 switch (k.k->type) {
1058                 case KEY_TYPE_dirent:
1059                         d = bkey_s_c_to_dirent(k);
1060                         d_inum = le64_to_cpu(d.v->d_inum);
1061
1062                         if (d.v->d_type == DT_DIR)
1063                                 inc_link(c, links, range_start, range_end,
1064                                          d.k->p.inode, true);
1065
1066                         inc_link(c, links, range_start, range_end,
1067                                  d_inum, false);
1068
1069                         break;
1070                 }
1071
1072                 bch2_trans_cond_resched(&trans);
1073         }
1074         ret = bch2_trans_exit(&trans) ?: ret;
1075         if (ret)
1076                 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
1077
1078         return ret;
1079 }
1080
1081 static int check_inode_nlink(struct bch_fs *c,
1082                              struct bch_inode_unpacked *lostfound_inode,
1083                              struct bch_inode_unpacked *u,
1084                              struct nlink *link,
1085                              bool *do_update)
1086 {
1087         u32 i_nlink = bch2_inode_nlink_get(u);
1088         u32 real_i_nlink =
1089                 link->count * nlink_bias(u->bi_mode) +
1090                 link->dir_count;
1091         int ret = 0;
1092
1093         /*
1094          * These should have been caught/fixed by earlier passes, we don't
1095          * repair them here:
1096          */
1097         if (S_ISDIR(u->bi_mode) && link->count > 1) {
1098                 need_fsck_err(c, "directory %llu with multiple hardlinks: %u",
1099                               u->bi_inum, link->count);
1100                 return 0;
1101         }
1102
1103         if (S_ISDIR(u->bi_mode) && !link->count) {
1104                 need_fsck_err(c, "unreachable directory found (inum %llu)",
1105                               u->bi_inum);
1106                 return 0;
1107         }
1108
1109         if (!S_ISDIR(u->bi_mode) && link->dir_count) {
1110                 need_fsck_err(c, "non directory with subdirectories",
1111                               u->bi_inum);
1112                 return 0;
1113         }
1114
1115         if (!link->count &&
1116             !(u->bi_flags & BCH_INODE_UNLINKED) &&
1117             (c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
1118                 if (fsck_err(c, "unreachable inode %llu not marked as unlinked (type %u)",
1119                              u->bi_inum, mode_to_type(u->bi_mode)) ==
1120                     FSCK_ERR_IGNORE)
1121                         return 0;
1122
1123                 ret = reattach_inode(c, lostfound_inode, u->bi_inum);
1124                 if (ret)
1125                         return ret;
1126
1127                 link->count = 1;
1128                 real_i_nlink = nlink_bias(u->bi_mode) + link->dir_count;
1129                 goto set_i_nlink;
1130         }
1131
1132         if (i_nlink < link->count) {
1133                 if (fsck_err(c, "inode %llu i_link too small (%u < %u, type %i)",
1134                              u->bi_inum, i_nlink, link->count,
1135                              mode_to_type(u->bi_mode)) == FSCK_ERR_IGNORE)
1136                         return 0;
1137                 goto set_i_nlink;
1138         }
1139
1140         if (i_nlink != real_i_nlink &&
1141             c->sb.clean) {
1142                 if (fsck_err(c, "filesystem marked clean, "
1143                              "but inode %llu has wrong i_nlink "
1144                              "(type %u i_nlink %u, should be %u)",
1145                              u->bi_inum, mode_to_type(u->bi_mode),
1146                              i_nlink, real_i_nlink) == FSCK_ERR_IGNORE)
1147                         return 0;
1148                 goto set_i_nlink;
1149         }
1150
1151         if (i_nlink != real_i_nlink &&
1152             (c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
1153                 if (fsck_err(c, "inode %llu has wrong i_nlink "
1154                              "(type %u i_nlink %u, should be %u)",
1155                              u->bi_inum, mode_to_type(u->bi_mode),
1156                              i_nlink, real_i_nlink) == FSCK_ERR_IGNORE)
1157                         return 0;
1158                 goto set_i_nlink;
1159         }
1160
1161         if (real_i_nlink && i_nlink != real_i_nlink)
1162                 bch_verbose(c, "setting inode %llu nlink from %u to %u",
1163                             u->bi_inum, i_nlink, real_i_nlink);
1164 set_i_nlink:
1165         if (i_nlink != real_i_nlink) {
1166                 bch2_inode_nlink_set(u, real_i_nlink);
1167                 *do_update = true;
1168         }
1169 fsck_err:
1170         return ret;
1171 }
1172
1173 static int check_inode(struct btree_trans *trans,
1174                        struct bch_inode_unpacked *lostfound_inode,
1175                        struct btree_iter *iter,
1176                        struct bkey_s_c_inode inode,
1177                        struct nlink *link)
1178 {
1179         struct bch_fs *c = trans->c;
1180         struct bch_inode_unpacked u;
1181         bool do_update = false;
1182         int ret = 0;
1183
1184         ret = bch2_inode_unpack(inode, &u);
1185
1186         bch2_trans_unlock(trans);
1187
1188         if (bch2_fs_inconsistent_on(ret, c,
1189                          "error unpacking inode %llu in fsck",
1190                          inode.k->p.inode))
1191                 return ret;
1192
1193         if (link) {
1194                 ret = check_inode_nlink(c, lostfound_inode, &u, link,
1195                                         &do_update);
1196                 if (ret)
1197                         return ret;
1198         }
1199
1200         if (u.bi_flags & BCH_INODE_UNLINKED &&
1201             (!c->sb.clean ||
1202              fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
1203                       u.bi_inum))) {
1204                 bch_verbose(c, "deleting inode %llu", u.bi_inum);
1205
1206                 ret = bch2_inode_rm(c, u.bi_inum);
1207                 if (ret)
1208                         bch_err(c, "error in fsck: error %i while deleting inode", ret);
1209                 return ret;
1210         }
1211
1212         if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
1213             (!c->sb.clean ||
1214              fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
1215                       u.bi_inum))) {
1216                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
1217
1218                 /*
1219                  * XXX: need to truncate partial blocks too here - or ideally
1220                  * just switch units to bytes and that issue goes away
1221                  */
1222
1223                 ret = bch2_inode_truncate(c, u.bi_inum, u.bi_size);
1224                 if (ret) {
1225                         bch_err(c, "error in fsck: error %i truncating inode", ret);
1226                         return ret;
1227                 }
1228
1229                 /*
1230                  * We truncated without our normal sector accounting hook, just
1231                  * make sure we recalculate it:
1232                  */
1233                 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
1234
1235                 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
1236                 do_update = true;
1237         }
1238
1239         if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
1240             (!c->sb.clean ||
1241              fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
1242                       u.bi_inum))) {
1243                 s64 sectors;
1244
1245                 bch_verbose(c, "recounting sectors for inode %llu",
1246                             u.bi_inum);
1247
1248                 sectors = bch2_count_inode_sectors(trans, u.bi_inum);
1249                 if (sectors < 0) {
1250                         bch_err(c, "error in fsck: error %i recounting inode sectors",
1251                                 (int) sectors);
1252                         return sectors;
1253                 }
1254
1255                 u.bi_sectors = sectors;
1256                 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
1257                 do_update = true;
1258         }
1259
1260         if (do_update) {
1261                 struct bkey_inode_buf p;
1262
1263                 bch2_inode_pack(&p, &u);
1264                 bch2_trans_update(trans, iter, &p.inode.k_i);
1265
1266                 ret = bch2_trans_commit(trans, NULL, NULL,
1267                                         BTREE_INSERT_NOFAIL|
1268                                         BTREE_INSERT_LAZY_RW);
1269                 if (ret && ret != -EINTR)
1270                         bch_err(c, "error in fsck: error %i "
1271                                 "updating inode", ret);
1272         }
1273 fsck_err:
1274         return ret;
1275 }
1276
1277 noinline_for_stack
1278 static int bch2_gc_walk_inodes(struct bch_fs *c,
1279                                struct bch_inode_unpacked *lostfound_inode,
1280                                nlink_table *links,
1281                                u64 range_start, u64 range_end)
1282 {
1283         struct btree_trans trans;
1284         struct btree_iter *iter;
1285         struct bkey_s_c k;
1286         struct nlink *link, zero_links = { 0, 0 };
1287         struct genradix_iter nlinks_iter;
1288         int ret = 0, ret2 = 0;
1289         u64 nlinks_pos;
1290
1291         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1292
1293         iter = bch2_trans_get_iter(&trans, BTREE_ID_INODES,
1294                                    POS(range_start, 0), 0);
1295         nlinks_iter = genradix_iter_init(links, 0);
1296
1297         while ((k = bch2_btree_iter_peek(iter)).k &&
1298                !(ret2 = bkey_err(k))) {
1299 peek_nlinks:    link = genradix_iter_peek(&nlinks_iter, links);
1300
1301                 if (!link && (!k.k || iter->pos.inode >= range_end))
1302                         break;
1303
1304                 nlinks_pos = range_start + nlinks_iter.pos;
1305                 if (iter->pos.inode > nlinks_pos) {
1306                         /* Should have been caught by dirents pass: */
1307                         need_fsck_err_on(link && link->count, c,
1308                                 "missing inode %llu (nlink %u)",
1309                                 nlinks_pos, link->count);
1310                         genradix_iter_advance(&nlinks_iter, links);
1311                         goto peek_nlinks;
1312                 }
1313
1314                 if (iter->pos.inode < nlinks_pos || !link)
1315                         link = &zero_links;
1316
1317                 if (k.k && k.k->type == KEY_TYPE_inode) {
1318                         ret = check_inode(&trans, lostfound_inode, iter,
1319                                           bkey_s_c_to_inode(k), link);
1320                         BUG_ON(ret == -EINTR);
1321                         if (ret)
1322                                 break;
1323                 } else {
1324                         /* Should have been caught by dirents pass: */
1325                         need_fsck_err_on(link->count, c,
1326                                 "missing inode %llu (nlink %u)",
1327                                 nlinks_pos, link->count);
1328                 }
1329
1330                 if (nlinks_pos == iter->pos.inode)
1331                         genradix_iter_advance(&nlinks_iter, links);
1332
1333                 bch2_btree_iter_next(iter);
1334                 bch2_trans_cond_resched(&trans);
1335         }
1336 fsck_err:
1337         bch2_trans_exit(&trans);
1338
1339         if (ret2)
1340                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret2);
1341
1342         return ret ?: ret2;
1343 }
1344
1345 noinline_for_stack
1346 static int check_inode_nlinks(struct bch_fs *c,
1347                               struct bch_inode_unpacked *lostfound_inode)
1348 {
1349         nlink_table links;
1350         u64 this_iter_range_start, next_iter_range_start = 0;
1351         int ret = 0;
1352
1353         bch_verbose(c, "checking inode nlinks");
1354
1355         genradix_init(&links);
1356
1357         do {
1358                 this_iter_range_start = next_iter_range_start;
1359                 next_iter_range_start = U64_MAX;
1360
1361                 ret = bch2_gc_walk_dirents(c, &links,
1362                                           this_iter_range_start,
1363                                           &next_iter_range_start);
1364                 if (ret)
1365                         break;
1366
1367                 ret = bch2_gc_walk_inodes(c, lostfound_inode, &links,
1368                                          this_iter_range_start,
1369                                          next_iter_range_start);
1370                 if (ret)
1371                         break;
1372
1373                 genradix_free(&links);
1374         } while (next_iter_range_start != U64_MAX);
1375
1376         genradix_free(&links);
1377
1378         return ret;
1379 }
1380
1381 /*
1382  * Checks for inconsistencies that shouldn't happen, unless we have a bug.
1383  * Doesn't fix them yet, mainly because they haven't yet been observed:
1384  */
1385 int bch2_fsck_full(struct bch_fs *c)
1386 {
1387         struct bch_inode_unpacked root_inode, lostfound_inode;
1388
1389         return  check_extents(c) ?:
1390                 check_dirents(c) ?:
1391                 check_xattrs(c) ?:
1392                 check_root(c, &root_inode) ?:
1393                 check_lostfound(c, &root_inode, &lostfound_inode) ?:
1394                 check_directory_structure(c, &lostfound_inode) ?:
1395                 check_inode_nlinks(c, &lostfound_inode);
1396 }
1397
1398 int bch2_fsck_inode_nlink(struct bch_fs *c)
1399 {
1400         struct bch_inode_unpacked root_inode, lostfound_inode;
1401
1402         return  check_root(c, &root_inode) ?:
1403                 check_lostfound(c, &root_inode, &lostfound_inode) ?:
1404                 check_inode_nlinks(c, &lostfound_inode);
1405 }
1406
1407 int bch2_fsck_walk_inodes_only(struct bch_fs *c)
1408 {
1409         struct btree_trans trans;
1410         struct btree_iter *iter;
1411         struct bkey_s_c k;
1412         struct bkey_s_c_inode inode;
1413         int ret;
1414
1415         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1416
1417         for_each_btree_key(&trans, iter, BTREE_ID_INODES, POS_MIN, 0, k, ret) {
1418                 if (k.k->type != KEY_TYPE_inode)
1419                         continue;
1420
1421                 inode = bkey_s_c_to_inode(k);
1422
1423                 if (inode.v->bi_flags &
1424                     (BCH_INODE_I_SIZE_DIRTY|
1425                      BCH_INODE_I_SECTORS_DIRTY|
1426                      BCH_INODE_UNLINKED)) {
1427                         ret = check_inode(&trans, NULL, iter, inode, NULL);
1428                         BUG_ON(ret == -EINTR);
1429                         if (ret)
1430                                 break;
1431                 }
1432         }
1433         BUG_ON(ret == -EINTR);
1434
1435         return bch2_trans_exit(&trans) ?: ret;
1436 }