fs/bcachefs/fsck.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bcachefs_ioctl.h"
   5 #include "bkey_buf.h"
   6 #include "btree_cache.h"
   7 #include "btree_update.h"
   8 #include "buckets.h"
   9 #include "darray.h"
  10 #include "dirent.h"
  11 #include "error.h"
  12 #include "fs.h"
  13 #include "fsck.h"
  14 #include "inode.h"
  15 #include "keylist.h"
  16 #include "namei.h"
  17 #include "recovery_passes.h"
  18 #include "snapshot.h"
  19 #include "super.h"
  20 #include "thread_with_file.h"
  21 #include "xattr.h"
  22
  23 #include <linux/bsearch.h>
  24 #include <linux/dcache.h> /* struct qstr */
  25
  26 static int dirent_points_to_inode_nowarn(struct bch_fs *c,
  27                                          struct bkey_s_c_dirent d,
  28                                          struct bch_inode_unpacked *inode)
  29 {
  30         if (d.v->d_type == DT_SUBVOL
  31             ? le32_to_cpu(d.v->d_child_subvol)  == inode->bi_subvol
  32             : le64_to_cpu(d.v->d_inum)          == inode->bi_inum)
  33                 return 0;
  34         return bch_err_throw(c, ENOENT_dirent_doesnt_match_inode);
  35 }
  36
  37 static void dirent_inode_mismatch_msg(struct printbuf *out,
  38                                       struct bch_fs *c,
  39                                       struct bkey_s_c_dirent dirent,
  40                                       struct bch_inode_unpacked *inode)
  41 {
  42         prt_str(out, "inode points to dirent that does not point back:");
  43         prt_newline(out);
  44         bch2_bkey_val_to_text(out, c, dirent.s_c);
  45         prt_newline(out);
  46         bch2_inode_unpacked_to_text(out, inode);
  47 }
  48
  49 static int dirent_points_to_inode(struct bch_fs *c,
  50                                   struct bkey_s_c_dirent dirent,
  51                                   struct bch_inode_unpacked *inode)
  52 {
  53         int ret = dirent_points_to_inode_nowarn(c, dirent, inode);
  54         if (ret) {
  55                 struct printbuf buf = PRINTBUF;
  56                 dirent_inode_mismatch_msg(&buf, c, dirent, inode);
  57                 bch_warn(c, "%s", buf.buf);
  58                 printbuf_exit(&buf);
  59         }
  60         return ret;
  61 }
  62
  63 /*
  64  * XXX: this is handling transaction restarts without returning
  65  * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
  66  */
  67 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
  68                                     u32 snapshot)
  69 {
  70         u64 sectors = 0;
  71
  72         int ret = for_each_btree_key_max(trans, iter, BTREE_ID_extents,
  73                                 SPOS(inum, 0, snapshot),
  74                                 POS(inum, U64_MAX),
  75                                 0, k, ({
  76                 if (bkey_extent_is_allocation(k.k))
  77                         sectors += k.k->size;
  78                 0;
  79         }));
  80
  81         return ret ?: sectors;
  82 }
  83
  84 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
  85                                     u32 snapshot)
  86 {
  87         u64 subdirs = 0;
  88
  89         int ret = for_each_btree_key_max(trans, iter, BTREE_ID_dirents,
  90                                     SPOS(inum, 0, snapshot),
  91                                     POS(inum, U64_MAX),
  92                                     0, k, ({
  93                 if (k.k->type == KEY_TYPE_dirent &&
  94                     bkey_s_c_to_dirent(k).v->d_type == DT_DIR)
  95                         subdirs++;
  96                 0;
  97         }));
  98
  99         return ret ?: subdirs;
 100 }
 101
 102 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
 103                          u32 *snapshot, u64 *inum)
 104 {
 105         struct bch_subvolume s;
 106         int ret = bch2_subvolume_get(trans, subvol, false, &s);
 107
 108         *snapshot = le32_to_cpu(s.snapshot);
 109         *inum = le64_to_cpu(s.inode);
 110         return ret;
 111 }
 112
 113 static int lookup_dirent_in_snapshot(struct btree_trans *trans,
 114                            struct bch_hash_info hash_info,
 115                            subvol_inum dir, struct qstr *name,
 116                            u64 *target, unsigned *type, u32 snapshot)
 117 {
 118         struct btree_iter iter;
 119         struct bkey_s_c k = bch2_hash_lookup_in_snapshot(trans, &iter, bch2_dirent_hash_desc,
 120                                                          &hash_info, dir, name, 0, snapshot);
 121         int ret = bkey_err(k);
 122         if (ret)
 123                 return ret;
 124
 125         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 126         *target = le64_to_cpu(d.v->d_inum);
 127         *type = d.v->d_type;
 128         bch2_trans_iter_exit(trans, &iter);
 129         return 0;
 130 }
 131
 132 /*
 133  * Find any subvolume associated with a tree of snapshots
 134  * We can't rely on master_subvol - it might have been deleted.
 135  */
 136 static int find_snapshot_tree_subvol(struct btree_trans *trans,
 137                                      u32 tree_id, u32 *subvol)
 138 {
 139         struct btree_iter iter;
 140         struct bkey_s_c k;
 141         int ret;
 142
 143         for_each_btree_key_norestart(trans, iter, BTREE_ID_snapshots, POS_MIN, 0, k, ret) {
 144                 if (k.k->type != KEY_TYPE_snapshot)
 145                         continue;
 146
 147                 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
 148                 if (le32_to_cpu(s.v->tree) != tree_id)
 149                         continue;
 150
 151                 if (s.v->subvol) {
 152                         *subvol = le32_to_cpu(s.v->subvol);
 153                         goto found;
 154                 }
 155         }
 156         ret = bch_err_throw(trans->c, ENOENT_no_snapshot_tree_subvol);
 157 found:
 158         bch2_trans_iter_exit(trans, &iter);
 159         return ret;
 160 }
 161
 162 /* Get lost+found, create if it doesn't exist: */
 163 static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
 164                             struct bch_inode_unpacked *lostfound,
 165                             u64 reattaching_inum)
 166 {
 167         struct bch_fs *c = trans->c;
 168         struct qstr lostfound_str = QSTR("lost+found");
 169         struct btree_iter lostfound_iter = {};
 170         u64 inum = 0;
 171         unsigned d_type = 0;
 172         int ret;
 173
 174         struct bch_snapshot_tree st;
 175         ret = bch2_snapshot_tree_lookup(trans,
 176                         bch2_snapshot_tree(c, snapshot), &st);
 177         if (ret)
 178                 return ret;
 179
 180         u32 subvolid;
 181         ret = find_snapshot_tree_subvol(trans,
 182                                 bch2_snapshot_tree(c, snapshot), &subvolid);
 183         bch_err_msg(c, ret, "finding subvol associated with snapshot tree %u",
 184                     bch2_snapshot_tree(c, snapshot));
 185         if (ret)
 186                 return ret;
 187
 188         struct bch_subvolume subvol;
 189         ret = bch2_subvolume_get(trans, subvolid, false, &subvol);
 190         bch_err_msg(c, ret, "looking up subvol %u for snapshot %u", subvolid, snapshot);
 191         if (ret)
 192                 return ret;
 193
 194         if (!subvol.inode) {
 195                 struct btree_iter iter;
 196                 struct bkey_i_subvolume *subvol = bch2_bkey_get_mut_typed(trans, &iter,
 197                                 BTREE_ID_subvolumes, POS(0, subvolid),
 198                                 0, subvolume);
 199                 ret = PTR_ERR_OR_ZERO(subvol);
 200                 if (ret)
 201                         return ret;
 202
 203                 subvol->v.inode = cpu_to_le64(reattaching_inum);
 204                 bch2_trans_iter_exit(trans, &iter);
 205         }
 206
 207         subvol_inum root_inum = {
 208                 .subvol = subvolid,
 209                 .inum = le64_to_cpu(subvol.inode)
 210         };
 211
 212         struct bch_inode_unpacked root_inode;
 213         struct bch_hash_info root_hash_info;
 214         ret = bch2_inode_find_by_inum_snapshot(trans, root_inum.inum, snapshot, &root_inode, 0);
 215         bch_err_msg(c, ret, "looking up root inode %llu for subvol %u",
 216                     root_inum.inum, subvolid);
 217         if (ret)
 218                 return ret;
 219
 220         root_hash_info = bch2_hash_info_init(c, &root_inode);
 221
 222         ret = lookup_dirent_in_snapshot(trans, root_hash_info, root_inum,
 223                               &lostfound_str, &inum, &d_type, snapshot);
 224         if (bch2_err_matches(ret, ENOENT))
 225                 goto create_lostfound;
 226
 227         bch_err_fn(c, ret);
 228         if (ret)
 229                 return ret;
 230
 231         if (d_type != DT_DIR) {
 232                 bch_err(c, "error looking up lost+found: not a directory");
 233                 return bch_err_throw(c, ENOENT_not_directory);
 234         }
 235
 236         /*
 237          * The bch2_check_dirents pass has already run, dangling dirents
 238          * shouldn't exist here:
 239          */
 240         ret = bch2_inode_find_by_inum_snapshot(trans, inum, snapshot, lostfound, 0);
 241         bch_err_msg(c, ret, "looking up lost+found %llu:%u in (root inode %llu, snapshot root %u)",
 242                     inum, snapshot, root_inum.inum, bch2_snapshot_root(c, snapshot));
 243         return ret;
 244
 245 create_lostfound:
 246         /*
 247          * we always create lost+found in the root snapshot; we don't want
 248          * different branches of the snapshot tree to have different lost+found
 249          */
 250         snapshot = le32_to_cpu(st.root_snapshot);
 251         /*
 252          * XXX: we could have a nicer log message here  if we had a nice way to
 253          * walk backpointers to print a path
 254          */
 255         struct printbuf path = PRINTBUF;
 256         ret = bch2_inum_to_path(trans, root_inum, &path);
 257         if (ret)
 258                 goto err;
 259
 260         bch_notice(c, "creating %s/lost+found in subvol %llu snapshot %u",
 261                    path.buf, root_inum.subvol, snapshot);
 262         printbuf_exit(&path);
 263
 264         u64 now = bch2_current_time(c);
 265         u64 cpu = raw_smp_processor_id();
 266
 267         bch2_inode_init_early(c, lostfound);
 268         bch2_inode_init_late(c, lostfound, now, 0, 0, S_IFDIR|0700, 0, &root_inode);
 269         lostfound->bi_dir = root_inode.bi_inum;
 270         lostfound->bi_snapshot = le32_to_cpu(st.root_snapshot);
 271
 272         root_inode.bi_nlink++;
 273
 274         ret = bch2_inode_create(trans, &lostfound_iter, lostfound, snapshot, cpu);
 275         if (ret)
 276                 goto err;
 277
 278         bch2_btree_iter_set_snapshot(trans, &lostfound_iter, snapshot);
 279         ret = bch2_btree_iter_traverse(trans, &lostfound_iter);
 280         if (ret)
 281                 goto err;
 282
 283         ret =   bch2_dirent_create_snapshot(trans,
 284                                 0, root_inode.bi_inum, snapshot, &root_hash_info,
 285                                 mode_to_type(lostfound->bi_mode),
 286                                 &lostfound_str,
 287                                 lostfound->bi_inum,
 288                                 &lostfound->bi_dir_offset,
 289                                 BTREE_UPDATE_internal_snapshot_node|
 290                                 STR_HASH_must_create) ?:
 291                 bch2_inode_write_flags(trans, &lostfound_iter, lostfound,
 292                                        BTREE_UPDATE_internal_snapshot_node);
 293 err:
 294         bch_err_msg(c, ret, "creating lost+found");
 295         bch2_trans_iter_exit(trans, &lostfound_iter);
 296         return ret;
 297 }
 298
 299 static inline bool inode_should_reattach(struct bch_inode_unpacked *inode)
 300 {
 301         if (inode->bi_inum == BCACHEFS_ROOT_INO &&
 302             inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)
 303                 return false;
 304
 305         /*
 306          * Subvolume roots are special: older versions of subvolume roots may be
 307          * disconnected, it's only the newest version that matters.
 308          *
 309          * We only keep a single dirent pointing to a subvolume root, i.e.
 310          * older versions of snapshots will not have a different dirent pointing
 311          * to the same subvolume root.
 312          *
 313          * This is because dirents that point to subvolumes are only visible in
 314          * the parent subvolume - versioning is not needed - and keeping them
 315          * around would break fsck, because when we're crossing subvolumes we
 316          * don't have a consistent snapshot ID to do check the inode <-> dirent
 317          * relationships.
 318          *
 319          * Thus, a subvolume root that's been renamed after a snapshot will have
 320          * a disconnected older version - that's expected.
 321          *
 322          * Note that taking a snapshot always updates the root inode (to update
 323          * the dirent backpointer), so a subvolume root inode with
 324          * BCH_INODE_has_child_snapshot is never visible.
 325          */
 326         if (inode->bi_subvol &&
 327             (inode->bi_flags & BCH_INODE_has_child_snapshot))
 328                 return false;
 329
 330         return !bch2_inode_has_backpointer(inode) &&
 331                 !(inode->bi_flags & BCH_INODE_unlinked);
 332 }
 333
 334 static int maybe_delete_dirent(struct btree_trans *trans, struct bpos d_pos, u32 snapshot)
 335 {
 336         struct btree_iter iter;
 337         struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_dirents,
 338                                         SPOS(d_pos.inode, d_pos.offset, snapshot),
 339                                         BTREE_ITER_intent|
 340                                         BTREE_ITER_with_updates);
 341         int ret = bkey_err(k);
 342         if (ret)
 343                 return ret;
 344
 345         if (bpos_eq(k.k->p, d_pos)) {
 346                 /*
 347                  * delet_at() doesn't work because the update path doesn't
 348                  * internally use BTREE_ITER_with_updates yet
 349                  */
 350                 struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k));
 351                 ret = PTR_ERR_OR_ZERO(k);
 352                 if (ret)
 353                         goto err;
 354
 355                 bkey_init(&k->k);
 356                 k->k.type = KEY_TYPE_whiteout;
 357                 k->k.p = iter.pos;
 358                 ret = bch2_trans_update(trans, &iter, k, BTREE_UPDATE_internal_snapshot_node);
 359         }
 360 err:
 361         bch2_trans_iter_exit(trans, &iter);
 362         return ret;
 363 }
 364
 365 static int reattach_inode(struct btree_trans *trans, struct bch_inode_unpacked *inode)
 366 {
 367         struct bch_fs *c = trans->c;
 368         struct bch_inode_unpacked lostfound;
 369         char name_buf[20];
 370         int ret;
 371
 372         u32 dirent_snapshot = inode->bi_snapshot;
 373         if (inode->bi_subvol) {
 374                 inode->bi_parent_subvol = BCACHEFS_ROOT_SUBVOL;
 375
 376                 struct btree_iter subvol_iter;
 377                 struct bkey_i_subvolume *subvol =
 378                         bch2_bkey_get_mut_typed(trans, &subvol_iter,
 379                                                 BTREE_ID_subvolumes, POS(0, inode->bi_subvol),
 380                                                 0, subvolume);
 381                 ret = PTR_ERR_OR_ZERO(subvol);
 382                 if (ret)
 383                         return ret;
 384
 385                 subvol->v.fs_path_parent = BCACHEFS_ROOT_SUBVOL;
 386                 bch2_trans_iter_exit(trans, &subvol_iter);
 387
 388                 u64 root_inum;
 389                 ret = subvol_lookup(trans, inode->bi_parent_subvol,
 390                                     &dirent_snapshot, &root_inum);
 391                 if (ret)
 392                         return ret;
 393
 394                 snprintf(name_buf, sizeof(name_buf), "subvol-%u", inode->bi_subvol);
 395         } else {
 396                 snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 397         }
 398
 399         ret = lookup_lostfound(trans, dirent_snapshot, &lostfound, inode->bi_inum);
 400         if (ret)
 401                 return ret;
 402
 403         bch_verbose(c, "got lostfound inum %llu", lostfound.bi_inum);
 404
 405         lostfound.bi_nlink += S_ISDIR(inode->bi_mode);
 406
 407         /* ensure lost+found inode is also present in inode snapshot */
 408         if (!inode->bi_subvol) {
 409                 BUG_ON(!bch2_snapshot_is_ancestor(c, inode->bi_snapshot, lostfound.bi_snapshot));
 410                 lostfound.bi_snapshot = inode->bi_snapshot;
 411         }
 412
 413         ret = __bch2_fsck_write_inode(trans, &lostfound);
 414         if (ret)
 415                 return ret;
 416
 417         struct bch_hash_info dir_hash = bch2_hash_info_init(c, &lostfound);
 418         struct qstr name = QSTR(name_buf);
 419
 420         inode->bi_dir = lostfound.bi_inum;
 421
 422         ret = bch2_dirent_create_snapshot(trans,
 423                                 inode->bi_parent_subvol, lostfound.bi_inum,
 424                                 dirent_snapshot,
 425                                 &dir_hash,
 426                                 inode_d_type(inode),
 427                                 &name,
 428                                 inode->bi_subvol ?: inode->bi_inum,
 429                                 &inode->bi_dir_offset,
 430                                 BTREE_UPDATE_internal_snapshot_node|
 431                                 STR_HASH_must_create);
 432         if (ret) {
 433                 bch_err_msg(c, ret, "error creating dirent");
 434                 return ret;
 435         }
 436
 437         ret = __bch2_fsck_write_inode(trans, inode);
 438         if (ret)
 439                 return ret;
 440
 441         {
 442                 CLASS(printbuf, buf)();
 443                 ret = bch2_inum_snapshot_to_path(trans, inode->bi_inum,
 444                                                  inode->bi_snapshot, NULL, &buf);
 445                 if (ret)
 446                         return ret;
 447
 448                 bch_info(c, "reattached at %s", buf.buf);
 449         }
 450
 451         /*
 452          * Fix up inodes in child snapshots: if they should also be reattached
 453          * update the backpointer field, if they should not be we need to emit
 454          * whiteouts for the dirent we just created.
 455          */
 456         if (!inode->bi_subvol && bch2_snapshot_is_leaf(c, inode->bi_snapshot) <= 0) {
 457                 snapshot_id_list whiteouts_done;
 458                 struct btree_iter iter;
 459                 struct bkey_s_c k;
 460
 461                 darray_init(&whiteouts_done);
 462
 463                 for_each_btree_key_reverse_norestart(trans, iter,
 464                                 BTREE_ID_inodes, SPOS(0, inode->bi_inum, inode->bi_snapshot - 1),
 465                                 BTREE_ITER_all_snapshots|BTREE_ITER_intent, k, ret) {
 466                         if (k.k->p.offset != inode->bi_inum)
 467                                 break;
 468
 469                         if (!bkey_is_inode(k.k) ||
 470                             !bch2_snapshot_is_ancestor(c, k.k->p.snapshot, inode->bi_snapshot) ||
 471                             snapshot_list_has_ancestor(c, &whiteouts_done, k.k->p.snapshot))
 472                                 continue;
 473
 474                         struct bch_inode_unpacked child_inode;
 475                         ret = bch2_inode_unpack(k, &child_inode);
 476                         if (ret)
 477                                 break;
 478
 479                         if (!inode_should_reattach(&child_inode)) {
 480                                 ret = maybe_delete_dirent(trans,
 481                                                           SPOS(lostfound.bi_inum, inode->bi_dir_offset,
 482                                                                dirent_snapshot),
 483                                                           k.k->p.snapshot);
 484                                 if (ret)
 485                                         break;
 486
 487                                 ret = snapshot_list_add(c, &whiteouts_done, k.k->p.snapshot);
 488                                 if (ret)
 489                                         break;
 490                         } else {
 491                                 iter.snapshot = k.k->p.snapshot;
 492                                 child_inode.bi_dir = inode->bi_dir;
 493                                 child_inode.bi_dir_offset = inode->bi_dir_offset;
 494
 495                                 ret = bch2_inode_write_flags(trans, &iter, &child_inode,
 496                                                              BTREE_UPDATE_internal_snapshot_node);
 497                                 if (ret)
 498                                         break;
 499                         }
 500                 }
 501                 darray_exit(&whiteouts_done);
 502                 bch2_trans_iter_exit(trans, &iter);
 503         }
 504
 505         return ret;
 506 }
 507
 508 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
 509                                                 struct btree_iter *iter,
 510                                                 struct bpos pos)
 511 {
 512         return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
 513 }
 514
 515 static int remove_backpointer(struct btree_trans *trans,
 516                               struct bch_inode_unpacked *inode)
 517 {
 518         if (!bch2_inode_has_backpointer(inode))
 519                 return 0;
 520
 521         u32 snapshot = inode->bi_snapshot;
 522
 523         if (inode->bi_parent_subvol) {
 524                 int ret = bch2_subvolume_get_snapshot(trans, inode->bi_parent_subvol, &snapshot);
 525                 if (ret)
 526                         return ret;
 527         }
 528
 529         struct bch_fs *c = trans->c;
 530         struct btree_iter iter;
 531         struct bkey_s_c_dirent d = dirent_get_by_pos(trans, &iter,
 532                                      SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
 533         int ret = bkey_err(d) ?:
 534                   dirent_points_to_inode(c, d, inode) ?:
 535                   bch2_fsck_remove_dirent(trans, d.k->p);
 536         bch2_trans_iter_exit(trans, &iter);
 537         return ret;
 538 }
 539
 540 static int reattach_subvol(struct btree_trans *trans, struct bkey_s_c_subvolume s)
 541 {
 542         struct bch_fs *c = trans->c;
 543
 544         struct bch_inode_unpacked inode;
 545         int ret = bch2_inode_find_by_inum_trans(trans,
 546                                 (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
 547                                 &inode);
 548         if (ret)
 549                 return ret;
 550
 551         ret = remove_backpointer(trans, &inode);
 552         if (!bch2_err_matches(ret, ENOENT))
 553                 bch_err_msg(c, ret, "removing dirent");
 554         if (ret)
 555                 return ret;
 556
 557         ret = reattach_inode(trans, &inode);
 558         bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
 559         return ret;
 560 }
 561
 562 static int reconstruct_subvol(struct btree_trans *trans, u32 snapshotid, u32 subvolid, u64 inum)
 563 {
 564         struct bch_fs *c = trans->c;
 565
 566         if (!bch2_snapshot_is_leaf(c, snapshotid)) {
 567                 bch_err(c, "need to reconstruct subvol, but have interior node snapshot");
 568                 return bch_err_throw(c, fsck_repair_unimplemented);
 569         }
 570
 571         /*
 572          * If inum isn't set, that means we're being called from check_dirents,
 573          * not check_inodes - the root of this subvolume doesn't exist or we
 574          * would have found it there:
 575          */
 576         if (!inum) {
 577                 struct btree_iter inode_iter = {};
 578                 struct bch_inode_unpacked new_inode;
 579                 u64 cpu = raw_smp_processor_id();
 580
 581                 bch2_inode_init_early(c, &new_inode);
 582                 bch2_inode_init_late(c, &new_inode, bch2_current_time(c), 0, 0, S_IFDIR|0755, 0, NULL);
 583
 584                 new_inode.bi_subvol = subvolid;
 585
 586                 int ret = bch2_inode_create(trans, &inode_iter, &new_inode, snapshotid, cpu) ?:
 587                           bch2_btree_iter_traverse(trans, &inode_iter) ?:
 588                           bch2_inode_write(trans, &inode_iter, &new_inode);
 589                 bch2_trans_iter_exit(trans, &inode_iter);
 590                 if (ret)
 591                         return ret;
 592
 593                 inum = new_inode.bi_inum;
 594         }
 595
 596         bch_info(c, "reconstructing subvol %u with root inode %llu", subvolid, inum);
 597
 598         struct bkey_i_subvolume *new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
 599         int ret = PTR_ERR_OR_ZERO(new_subvol);
 600         if (ret)
 601                 return ret;
 602
 603         bkey_subvolume_init(&new_subvol->k_i);
 604         new_subvol->k.p.offset  = subvolid;
 605         new_subvol->v.snapshot  = cpu_to_le32(snapshotid);
 606         new_subvol->v.inode     = cpu_to_le64(inum);
 607         ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &new_subvol->k_i, 0);
 608         if (ret)
 609                 return ret;
 610
 611         struct btree_iter iter;
 612         struct bkey_i_snapshot *s = bch2_bkey_get_mut_typed(trans, &iter,
 613                         BTREE_ID_snapshots, POS(0, snapshotid),
 614                         0, snapshot);
 615         ret = PTR_ERR_OR_ZERO(s);
 616         bch_err_msg(c, ret, "getting snapshot %u", snapshotid);
 617         if (ret)
 618                 return ret;
 619
 620         u32 snapshot_tree = le32_to_cpu(s->v.tree);
 621
 622         s->v.subvol = cpu_to_le32(subvolid);
 623         SET_BCH_SNAPSHOT_SUBVOL(&s->v, true);
 624         bch2_trans_iter_exit(trans, &iter);
 625
 626         struct bkey_i_snapshot_tree *st = bch2_bkey_get_mut_typed(trans, &iter,
 627                         BTREE_ID_snapshot_trees, POS(0, snapshot_tree),
 628                         0, snapshot_tree);
 629         ret = PTR_ERR_OR_ZERO(st);
 630         bch_err_msg(c, ret, "getting snapshot tree %u", snapshot_tree);
 631         if (ret)
 632                 return ret;
 633
 634         if (!st->v.master_subvol)
 635                 st->v.master_subvol = cpu_to_le32(subvolid);
 636
 637         bch2_trans_iter_exit(trans, &iter);
 638         return 0;
 639 }
 640
 641 static int reconstruct_inode(struct btree_trans *trans, enum btree_id btree, u32 snapshot, u64 inum)
 642 {
 643         struct bch_fs *c = trans->c;
 644         unsigned i_mode = S_IFREG;
 645         u64 i_size = 0;
 646
 647         switch (btree) {
 648         case BTREE_ID_extents: {
 649                 struct btree_iter iter = {};
 650
 651                 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, SPOS(inum, U64_MAX, snapshot), 0);
 652                 struct bkey_s_c k = bch2_btree_iter_peek_prev_min(trans, &iter, POS(inum, 0));
 653                 bch2_trans_iter_exit(trans, &iter);
 654                 int ret = bkey_err(k);
 655                 if (ret)
 656                         return ret;
 657
 658                 i_size = k.k->p.offset << 9;
 659                 break;
 660         }
 661         case BTREE_ID_dirents:
 662                 i_mode = S_IFDIR;
 663                 break;
 664         case BTREE_ID_xattrs:
 665                 break;
 666         default:
 667                 BUG();
 668         }
 669
 670         struct bch_inode_unpacked new_inode;
 671         bch2_inode_init_early(c, &new_inode);
 672         bch2_inode_init_late(c, &new_inode, bch2_current_time(c), 0, 0, i_mode|0600, 0, NULL);
 673         new_inode.bi_size = i_size;
 674         new_inode.bi_inum = inum;
 675         new_inode.bi_snapshot = snapshot;
 676
 677         return __bch2_fsck_write_inode(trans, &new_inode);
 678 }
 679
 680 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 681 {
 682         darray_exit(&s->ids);
 683 }
 684
 685 static inline void snapshots_seen_init(struct snapshots_seen *s)
 686 {
 687         memset(s, 0, sizeof(*s));
 688 }
 689
 690 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 691 {
 692         u32 *i;
 693         __darray_for_each(s->ids, i) {
 694                 if (*i == id)
 695                         return 0;
 696                 if (*i > id)
 697                         break;
 698         }
 699
 700         int ret = darray_insert_item(&s->ids, i - s->ids.data, id);
 701         if (ret)
 702                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 703                         s->ids.size);
 704         return ret;
 705 }
 706
 707 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 708                                  enum btree_id btree_id, struct bpos pos)
 709 {
 710         if (!bkey_eq(s->pos, pos))
 711                 s->ids.nr = 0;
 712         s->pos = pos;
 713
 714         return snapshot_list_add_nodup(c, &s->ids, pos.snapshot);
 715 }
 716
 717 /**
 718  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 719  * and @ancestor hasn't been overwritten in @seen
 720  *
 721  * @c:          filesystem handle
 722  * @seen:       list of snapshot ids already seen at current position
 723  * @id:         descendent snapshot id
 724  * @ancestor:   ancestor snapshot id
 725  *
 726  * Returns:     whether key in @ancestor snapshot is visible in @id snapshot
 727  */
 728 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 729                                     u32 id, u32 ancestor)
 730 {
 731         EBUG_ON(id > ancestor);
 732
 733         if (id == ancestor)
 734                 return true;
 735
 736         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 737                 return false;
 738
 739         /*
 740          * We know that @id is a descendant of @ancestor, we're checking if
 741          * we've seen a key that overwrote @ancestor - i.e. also a descendent of
 742          * @ascestor and with @id as a descendent.
 743          *
 744          * But we already know that we're scanning IDs between @id and @ancestor
 745          * numerically, since snapshot ID lists are kept sorted, so if we find
 746          * an id that's an ancestor of @id we're done:
 747          */
 748         darray_for_each_reverse(seen->ids, i)
 749                 if (*i != ancestor && bch2_snapshot_is_ancestor(c, id, *i))
 750                         return false;
 751
 752         return true;
 753 }
 754
 755 /**
 756  * ref_visible - given a key with snapshot id @src that points to a key with
 757  * snapshot id @dst, test whether there is some snapshot in which @dst is
 758  * visible.
 759  *
 760  * @c:          filesystem handle
 761  * @s:          list of snapshot IDs already seen at @src
 762  * @src:        snapshot ID of src key
 763  * @dst:        snapshot ID of dst key
 764  * Returns:     true if there is some snapshot in which @dst is visible
 765  *
 766  * Assumes we're visiting @src keys in natural key order
 767  */
 768 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 769                         u32 src, u32 dst)
 770 {
 771         return dst <= src
 772                 ? key_visible_in_snapshot(c, s, dst, src)
 773                 : bch2_snapshot_is_ancestor(c, src, dst);
 774 }
 775
 776 static int ref_visible2(struct bch_fs *c,
 777                         u32 src, struct snapshots_seen *src_seen,
 778                         u32 dst, struct snapshots_seen *dst_seen)
 779 {
 780         if (dst > src) {
 781                 swap(dst, src);
 782                 swap(dst_seen, src_seen);
 783         }
 784         return key_visible_in_snapshot(c, src_seen, dst, src);
 785 }
 786
 787 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)                               \
 788         for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&        \
 789              (_i)->inode.bi_snapshot <= (_snapshot); _i++)                              \
 790                 if (key_visible_in_snapshot(_c, _s, _i->inode.bi_snapshot, _snapshot))
 791
 792 struct inode_walker_entry {
 793         struct bch_inode_unpacked inode;
 794         bool                    whiteout;
 795         u64                     count;
 796         u64                     i_size;
 797 };
 798
 799 struct inode_walker {
 800         bool                            first_this_inode;
 801         bool                            have_inodes;
 802         bool                            recalculate_sums;
 803         struct bpos                     last_pos;
 804
 805         DARRAY(struct inode_walker_entry) inodes;
 806         snapshot_id_list                deletes;
 807 };
 808
 809 static void inode_walker_exit(struct inode_walker *w)
 810 {
 811         darray_exit(&w->inodes);
 812         darray_exit(&w->deletes);
 813 }
 814
 815 static struct inode_walker inode_walker_init(void)
 816 {
 817         return (struct inode_walker) { 0, };
 818 }
 819
 820 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 821                      struct bkey_s_c inode)
 822 {
 823         int ret = darray_push(&w->inodes, ((struct inode_walker_entry) {
 824                 .whiteout       = !bkey_is_inode(inode.k),
 825         }));
 826         if (ret)
 827                 return ret;
 828
 829         struct inode_walker_entry *n = &darray_last(w->inodes);
 830         if (!n->whiteout) {
 831                 return bch2_inode_unpack(inode, &n->inode);
 832         } else {
 833                 n->inode.bi_inum        = inode.k->p.offset;
 834                 n->inode.bi_snapshot    = inode.k->p.snapshot;
 835                 return 0;
 836         }
 837 }
 838
 839 static int get_inodes_all_snapshots(struct btree_trans *trans,
 840                                     struct inode_walker *w, u64 inum)
 841 {
 842         struct bch_fs *c = trans->c;
 843         struct btree_iter iter;
 844         struct bkey_s_c k;
 845         int ret;
 846
 847         /*
 848          * We no longer have inodes for w->last_pos; clear this to avoid
 849          * screwing up check_i_sectors/check_subdir_count if we take a
 850          * transaction restart here:
 851          */
 852         w->have_inodes = false;
 853         w->recalculate_sums = false;
 854         w->inodes.nr = 0;
 855
 856         for_each_btree_key_max_norestart(trans, iter,
 857                         BTREE_ID_inodes, POS(0, inum), SPOS(0, inum, U32_MAX),
 858                         BTREE_ITER_all_snapshots, k, ret) {
 859                 ret = add_inode(c, w, k);
 860                 if (ret)
 861                         break;
 862         }
 863         bch2_trans_iter_exit(trans, &iter);
 864
 865         if (ret)
 866                 return ret;
 867
 868         w->first_this_inode = true;
 869         w->have_inodes = true;
 870         return 0;
 871 }
 872
 873 static int get_visible_inodes(struct btree_trans *trans,
 874                               struct inode_walker *w,
 875                               struct snapshots_seen *s,
 876                               u64 inum)
 877 {
 878         struct bch_fs *c = trans->c;
 879         struct btree_iter iter;
 880         struct bkey_s_c k;
 881         int ret;
 882
 883         w->inodes.nr = 0;
 884         w->deletes.nr = 0;
 885
 886         for_each_btree_key_reverse_norestart(trans, iter, BTREE_ID_inodes, SPOS(0, inum, s->pos.snapshot),
 887                            BTREE_ITER_all_snapshots, k, ret) {
 888                 if (k.k->p.offset != inum)
 889                         break;
 890
 891                 if (!ref_visible(c, s, s->pos.snapshot, k.k->p.snapshot))
 892                         continue;
 893
 894                 if (snapshot_list_has_ancestor(c, &w->deletes, k.k->p.snapshot))
 895                         continue;
 896
 897                 ret = bkey_is_inode(k.k)
 898                         ? add_inode(c, w, k)
 899                         : snapshot_list_add(c, &w->deletes, k.k->p.snapshot);
 900                 if (ret)
 901                         break;
 902         }
 903         bch2_trans_iter_exit(trans, &iter);
 904
 905         return ret;
 906 }
 907
 908 static struct inode_walker_entry *
 909 lookup_inode_for_snapshot(struct btree_trans *trans, struct inode_walker *w, struct bkey_s_c k)
 910 {
 911         struct bch_fs *c = trans->c;
 912
 913         struct inode_walker_entry *i = darray_find_p(w->inodes, i,
 914                     bch2_snapshot_is_ancestor(c, k.k->p.snapshot, i->inode.bi_snapshot));
 915
 916         if (!i)
 917                 return NULL;
 918
 919         struct printbuf buf = PRINTBUF;
 920         int ret = 0;
 921
 922         if (fsck_err_on(k.k->p.snapshot != i->inode.bi_snapshot,
 923                         trans, snapshot_key_missing_inode_snapshot,
 924                          "have key for inode %llu:%u but have inode in ancestor snapshot %u\n"
 925                          "unexpected because we should always update the inode when we update a key in that inode\n"
 926                          "%s",
 927                          w->last_pos.inode, k.k->p.snapshot, i->inode.bi_snapshot,
 928                          (bch2_bkey_val_to_text(&buf, c, k),
 929                           buf.buf))) {
 930                 if (!i->whiteout) {
 931                         struct bch_inode_unpacked new = i->inode;
 932                         new.bi_snapshot = k.k->p.snapshot;
 933                         ret = __bch2_fsck_write_inode(trans, &new);
 934                 } else {
 935                         struct bkey_i whiteout;
 936                         bkey_init(&whiteout.k);
 937                         whiteout.k.type = KEY_TYPE_whiteout;
 938                         whiteout.k.p = SPOS(0, i->inode.bi_inum, k.k->p.snapshot);
 939                         ret = bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
 940                                                           &whiteout,
 941                                                           BTREE_UPDATE_internal_snapshot_node);
 942                 }
 943
 944                 if (ret)
 945                         goto fsck_err;
 946
 947                 ret = bch2_trans_commit(trans, NULL, NULL, 0);
 948                 if (ret)
 949                         goto fsck_err;
 950
 951                 struct inode_walker_entry new_entry = *i;
 952
 953                 new_entry.inode.bi_snapshot     = k.k->p.snapshot;
 954                 new_entry.count                 = 0;
 955                 new_entry.i_size                = 0;
 956
 957                 while (i > w->inodes.data && i[-1].inode.bi_snapshot > k.k->p.snapshot)
 958                         --i;
 959
 960                 size_t pos = i - w->inodes.data;
 961                 ret = darray_insert_item(&w->inodes, pos, new_entry);
 962                 if (ret)
 963                         goto fsck_err;
 964
 965                 ret = bch_err_throw(c, transaction_restart_nested);
 966                 goto fsck_err;
 967         }
 968
 969         printbuf_exit(&buf);
 970         return i;
 971 fsck_err:
 972         printbuf_exit(&buf);
 973         return ERR_PTR(ret);
 974 }
 975
 976 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
 977                                              struct inode_walker *w,
 978                                              struct bkey_s_c k)
 979 {
 980         if (w->last_pos.inode != k.k->p.inode) {
 981                 int ret = get_inodes_all_snapshots(trans, w, k.k->p.inode);
 982                 if (ret)
 983                         return ERR_PTR(ret);
 984         }
 985
 986         w->last_pos = k.k->p;
 987
 988         return lookup_inode_for_snapshot(trans, w, k);
 989 }
 990
 991 /*
 992  * Prefer to delete the first one, since that will be the one at the wrong
 993  * offset:
 994  * return value: 0 -> delete k1, 1 -> delete k2
 995  */
 996 int bch2_fsck_update_backpointers(struct btree_trans *trans,
 997                                   struct snapshots_seen *s,
 998                                   const struct bch_hash_desc desc,
 999                                   struct bch_hash_info *hash_info,
1000                                   struct bkey_i *new)
1001 {
1002         if (new->k.type != KEY_TYPE_dirent)
1003                 return 0;
1004
1005         struct bkey_i_dirent *d = bkey_i_to_dirent(new);
1006         struct inode_walker target = inode_walker_init();
1007         int ret = 0;
1008
1009         if (d->v.d_type == DT_SUBVOL) {
1010                 bch_err(trans->c, "%s does not support DT_SUBVOL", __func__);
1011                 ret = -BCH_ERR_fsck_repair_unimplemented;
1012         } else {
1013                 ret = get_visible_inodes(trans, &target, s, le64_to_cpu(d->v.d_inum));
1014                 if (ret)
1015                         goto err;
1016
1017                 darray_for_each(target.inodes, i) {
1018                         i->inode.bi_dir_offset = d->k.p.offset;
1019                         ret = __bch2_fsck_write_inode(trans, &i->inode);
1020                         if (ret)
1021                                 goto err;
1022                 }
1023         }
1024 err:
1025         inode_walker_exit(&target);
1026         return ret;
1027 }
1028
1029 static struct bkey_s_c_dirent inode_get_dirent(struct btree_trans *trans,
1030                                                struct btree_iter *iter,
1031                                                struct bch_inode_unpacked *inode,
1032                                                u32 *snapshot)
1033 {
1034         if (inode->bi_subvol) {
1035                 u64 inum;
1036                 int ret = subvol_lookup(trans, inode->bi_parent_subvol, snapshot, &inum);
1037                 if (ret)
1038                         return ((struct bkey_s_c_dirent) { .k = ERR_PTR(ret) });
1039         }
1040
1041         return dirent_get_by_pos(trans, iter, SPOS(inode->bi_dir, inode->bi_dir_offset, *snapshot));
1042 }
1043
1044 static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
1045 {
1046         struct btree_iter iter;
1047         struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
1048         int ret = bkey_err(k) ?: k.k->type == KEY_TYPE_set;
1049         bch2_trans_iter_exit(trans, &iter);
1050         return ret;
1051 }
1052
1053 static int check_inode_dirent_inode(struct btree_trans *trans,
1054                                     struct bch_inode_unpacked *inode,
1055                                     bool *write_inode)
1056 {
1057         struct bch_fs *c = trans->c;
1058         struct printbuf buf = PRINTBUF;
1059
1060         u32 inode_snapshot = inode->bi_snapshot;
1061         struct btree_iter dirent_iter = {};
1062         struct bkey_s_c_dirent d = inode_get_dirent(trans, &dirent_iter, inode, &inode_snapshot);
1063         int ret = bkey_err(d);
1064         if (ret && !bch2_err_matches(ret, ENOENT))
1065                 return ret;
1066
1067         if ((ret || dirent_points_to_inode_nowarn(c, d, inode)) &&
1068             inode->bi_subvol &&
1069             (inode->bi_flags & BCH_INODE_has_child_snapshot)) {
1070                 /* Older version of a renamed subvolume root: we won't have a
1071                  * correct dirent for it. That's expected, see
1072                  * inode_should_reattach().
1073                  *
1074                  * We don't clear the backpointer field when doing the rename
1075                  * because there might be arbitrarily many versions in older
1076                  * snapshots.
1077                  */
1078                 inode->bi_dir = 0;
1079                 inode->bi_dir_offset = 0;
1080                 *write_inode = true;
1081                 goto out;
1082         }
1083
1084         if (fsck_err_on(ret,
1085                         trans, inode_points_to_missing_dirent,
1086                         "inode points to missing dirent\n%s",
1087                         (bch2_inode_unpacked_to_text(&buf, inode), buf.buf)) ||
1088             fsck_err_on(!ret && dirent_points_to_inode_nowarn(c, d, inode),
1089                         trans, inode_points_to_wrong_dirent,
1090                         "%s",
1091                         (printbuf_reset(&buf),
1092                          dirent_inode_mismatch_msg(&buf, c, d, inode),
1093                          buf.buf))) {
1094                 /*
1095                  * We just clear the backpointer fields for now. If we find a
1096                  * dirent that points to this inode in check_dirents(), we'll
1097                  * update it then; then when we get to check_path() if the
1098                  * backpointer is still 0 we'll reattach it.
1099                  */
1100                 inode->bi_dir = 0;
1101                 inode->bi_dir_offset = 0;
1102                 *write_inode = true;
1103         }
1104 out:
1105         ret = 0;
1106 fsck_err:
1107         bch2_trans_iter_exit(trans, &dirent_iter);
1108         printbuf_exit(&buf);
1109         bch_err_fn(c, ret);
1110         return ret;
1111 }
1112
1113 static int check_inode(struct btree_trans *trans,
1114                        struct btree_iter *iter,
1115                        struct bkey_s_c k,
1116                        struct bch_inode_unpacked *snapshot_root,
1117                        struct snapshots_seen *s)
1118 {
1119         struct bch_fs *c = trans->c;
1120         struct printbuf buf = PRINTBUF;
1121         struct bch_inode_unpacked u;
1122         bool do_update = false;
1123         int ret;
1124
1125         ret = bch2_check_key_has_snapshot(trans, iter, k);
1126         if (ret < 0)
1127                 goto err;
1128         if (ret)
1129                 return 0;
1130
1131         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1132         if (ret)
1133                 goto err;
1134
1135         if (!bkey_is_inode(k.k))
1136                 return 0;
1137
1138         ret = bch2_inode_unpack(k, &u);
1139         if (ret)
1140                 goto err;
1141
1142         if (snapshot_root->bi_inum != u.bi_inum) {
1143                 ret = bch2_inode_find_snapshot_root(trans, u.bi_inum, snapshot_root);
1144                 if (ret)
1145                         goto err;
1146         }
1147
1148         if (u.bi_hash_seed      != snapshot_root->bi_hash_seed ||
1149             INODE_STR_HASH(&u)  != INODE_STR_HASH(snapshot_root)) {
1150                 ret = bch2_repair_inode_hash_info(trans, snapshot_root);
1151                 BUG_ON(ret == -BCH_ERR_fsck_repair_unimplemented);
1152                 if (ret)
1153                         goto err;
1154         }
1155
1156         ret = bch2_check_inode_has_case_insensitive(trans, &u, &s->ids, &do_update);
1157         if (ret)
1158                 goto err;
1159
1160         if (bch2_inode_has_backpointer(&u)) {
1161                 ret = check_inode_dirent_inode(trans, &u, &do_update);
1162                 if (ret)
1163                         goto err;
1164         }
1165
1166         if (fsck_err_on(bch2_inode_has_backpointer(&u) &&
1167                         (u.bi_flags & BCH_INODE_unlinked),
1168                         trans, inode_unlinked_but_has_dirent,
1169                         "inode unlinked but has dirent\n%s",
1170                         (printbuf_reset(&buf),
1171                          bch2_inode_unpacked_to_text(&buf, &u),
1172                          buf.buf))) {
1173                 u.bi_flags &= ~BCH_INODE_unlinked;
1174                 do_update = true;
1175         }
1176
1177         if (S_ISDIR(u.bi_mode) && (u.bi_flags & BCH_INODE_unlinked)) {
1178                 /* Check for this early so that check_unreachable_inode() will reattach it */
1179
1180                 ret = bch2_empty_dir_snapshot(trans, k.k->p.offset, 0, k.k->p.snapshot);
1181                 if (ret && ret != -BCH_ERR_ENOTEMPTY_dir_not_empty)
1182                         goto err;
1183
1184                 fsck_err_on(ret, trans, inode_dir_unlinked_but_not_empty,
1185                             "dir unlinked but not empty\n%s",
1186                             (printbuf_reset(&buf),
1187                              bch2_inode_unpacked_to_text(&buf, &u),
1188                              buf.buf));
1189                 u.bi_flags &= ~BCH_INODE_unlinked;
1190                 do_update = true;
1191                 ret = 0;
1192         }
1193
1194         if (fsck_err_on(S_ISDIR(u.bi_mode) && u.bi_size,
1195                         trans, inode_dir_has_nonzero_i_size,
1196                         "directory %llu:%u with nonzero i_size %lli",
1197                         u.bi_inum, u.bi_snapshot, u.bi_size)) {
1198                 u.bi_size = 0;
1199                 do_update = true;
1200         }
1201
1202         ret = bch2_inode_has_child_snapshots(trans, k.k->p);
1203         if (ret < 0)
1204                 goto err;
1205
1206         if (fsck_err_on(ret != !!(u.bi_flags & BCH_INODE_has_child_snapshot),
1207                         trans, inode_has_child_snapshots_wrong,
1208                         "inode has_child_snapshots flag wrong (should be %u)\n%s",
1209                         ret,
1210                         (printbuf_reset(&buf),
1211                          bch2_inode_unpacked_to_text(&buf, &u),
1212                          buf.buf))) {
1213                 if (ret)
1214                         u.bi_flags |= BCH_INODE_has_child_snapshot;
1215                 else
1216                         u.bi_flags &= ~BCH_INODE_has_child_snapshot;
1217                 do_update = true;
1218         }
1219         ret = 0;
1220
1221         if ((u.bi_flags & BCH_INODE_unlinked) &&
1222             !(u.bi_flags & BCH_INODE_has_child_snapshot)) {
1223                 if (!test_bit(BCH_FS_started, &c->flags)) {
1224                         /*
1225                          * If we're not in online fsck, don't delete unlinked
1226                          * inodes, just make sure they're on the deleted list.
1227                          *
1228                          * They might be referred to by a logged operation -
1229                          * i.e. we might have crashed in the middle of a
1230                          * truncate on an unlinked but open file - so we want to
1231                          * let the delete_dead_inodes kill it after resuming
1232                          * logged ops.
1233                          */
1234                         ret = check_inode_deleted_list(trans, k.k->p);
1235                         if (ret < 0)
1236                                 goto err_noprint;
1237
1238                         fsck_err_on(!ret,
1239                                     trans, unlinked_inode_not_on_deleted_list,
1240                                     "inode %llu:%u unlinked, but not on deleted list",
1241                                     u.bi_inum, k.k->p.snapshot);
1242
1243                         ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, k.k->p, 1);
1244                         if (ret)
1245                                 goto err;
1246                 } else {
1247                         ret = bch2_inode_or_descendents_is_open(trans, k.k->p);
1248                         if (ret < 0)
1249                                 goto err;
1250
1251                         if (fsck_err_on(!ret,
1252                                         trans, inode_unlinked_and_not_open,
1253                                       "inode %llu:%u unlinked and not open",
1254                                       u.bi_inum, u.bi_snapshot)) {
1255                                 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
1256                                 bch_err_msg(c, ret, "in fsck deleting inode");
1257                                 goto err_noprint;
1258                         }
1259                         ret = 0;
1260                 }
1261         }
1262
1263         if (fsck_err_on(u.bi_parent_subvol &&
1264                         (u.bi_subvol == 0 ||
1265                          u.bi_subvol == BCACHEFS_ROOT_SUBVOL),
1266                         trans, inode_bi_parent_nonzero,
1267                         "inode %llu:%u has subvol %u but nonzero parent subvol %u",
1268                         u.bi_inum, k.k->p.snapshot, u.bi_subvol, u.bi_parent_subvol)) {
1269                 u.bi_parent_subvol = 0;
1270                 do_update = true;
1271         }
1272
1273         if (u.bi_subvol) {
1274                 struct bch_subvolume s;
1275
1276                 ret = bch2_subvolume_get(trans, u.bi_subvol, false, &s);
1277                 if (ret && !bch2_err_matches(ret, ENOENT))
1278                         goto err;
1279
1280                 if (ret && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
1281                         ret = reconstruct_subvol(trans, k.k->p.snapshot, u.bi_subvol, u.bi_inum);
1282                         goto do_update;
1283                 }
1284
1285                 if (fsck_err_on(ret,
1286                                 trans, inode_bi_subvol_missing,
1287                                 "inode %llu:%u bi_subvol points to missing subvolume %u",
1288                                 u.bi_inum, k.k->p.snapshot, u.bi_subvol) ||
1289                     fsck_err_on(le64_to_cpu(s.inode) != u.bi_inum ||
1290                                 !bch2_snapshot_is_ancestor(c, le32_to_cpu(s.snapshot),
1291                                                            k.k->p.snapshot),
1292                                 trans, inode_bi_subvol_wrong,
1293                                 "inode %llu:%u points to subvol %u, but subvol points to %llu:%u",
1294                                 u.bi_inum, k.k->p.snapshot, u.bi_subvol,
1295                                 le64_to_cpu(s.inode),
1296                                 le32_to_cpu(s.snapshot))) {
1297                         u.bi_subvol = 0;
1298                         u.bi_parent_subvol = 0;
1299                         do_update = true;
1300                 }
1301         }
1302
1303         if (fsck_err_on(u.bi_journal_seq > journal_cur_seq(&c->journal),
1304                         trans, inode_journal_seq_in_future,
1305                         "inode journal seq in future (currently at %llu)\n%s",
1306                         journal_cur_seq(&c->journal),
1307                         (printbuf_reset(&buf),
1308                          bch2_inode_unpacked_to_text(&buf, &u),
1309                         buf.buf))) {
1310                 u.bi_journal_seq = journal_cur_seq(&c->journal);
1311                 do_update = true;
1312         }
1313 do_update:
1314         if (do_update) {
1315                 ret = __bch2_fsck_write_inode(trans, &u);
1316                 bch_err_msg(c, ret, "in fsck updating inode");
1317                 if (ret)
1318                         goto err_noprint;
1319         }
1320 err:
1321 fsck_err:
1322         bch_err_fn(c, ret);
1323 err_noprint:
1324         printbuf_exit(&buf);
1325         return ret;
1326 }
1327
1328 int bch2_check_inodes(struct bch_fs *c)
1329 {
1330         struct bch_inode_unpacked snapshot_root = {};
1331         struct snapshots_seen s;
1332
1333         snapshots_seen_init(&s);
1334
1335         int ret = bch2_trans_run(c,
1336                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1337                                 POS_MIN,
1338                                 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1339                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1340                         check_inode(trans, &iter, k, &snapshot_root, &s)));
1341
1342         snapshots_seen_exit(&s);
1343         bch_err_fn(c, ret);
1344         return ret;
1345 }
1346
1347 static int find_oldest_inode_needs_reattach(struct btree_trans *trans,
1348                                             struct bch_inode_unpacked *inode)
1349 {
1350         struct bch_fs *c = trans->c;
1351         struct btree_iter iter;
1352         struct bkey_s_c k;
1353         int ret = 0;
1354
1355         /*
1356          * We look for inodes to reattach in natural key order, leaves first,
1357          * but we should do the reattach at the oldest version that needs to be
1358          * reattached:
1359          */
1360         for_each_btree_key_norestart(trans, iter,
1361                                      BTREE_ID_inodes,
1362                                      SPOS(0, inode->bi_inum, inode->bi_snapshot + 1),
1363                                      BTREE_ITER_all_snapshots, k, ret) {
1364                 if (k.k->p.offset != inode->bi_inum)
1365                         break;
1366
1367                 if (!bch2_snapshot_is_ancestor(c, inode->bi_snapshot, k.k->p.snapshot))
1368                         continue;
1369
1370                 if (!bkey_is_inode(k.k))
1371                         break;
1372
1373                 struct bch_inode_unpacked parent_inode;
1374                 ret = bch2_inode_unpack(k, &parent_inode);
1375                 if (ret)
1376                         break;
1377
1378                 if (!inode_should_reattach(&parent_inode))
1379                         break;
1380
1381                 *inode = parent_inode;
1382         }
1383         bch2_trans_iter_exit(trans, &iter);
1384
1385         return ret;
1386 }
1387
1388 static int check_unreachable_inode(struct btree_trans *trans,
1389                                    struct btree_iter *iter,
1390                                    struct bkey_s_c k)
1391 {
1392         struct printbuf buf = PRINTBUF;
1393         int ret = 0;
1394
1395         if (!bkey_is_inode(k.k))
1396                 return 0;
1397
1398         struct bch_inode_unpacked inode;
1399         ret = bch2_inode_unpack(k, &inode);
1400         if (ret)
1401                 return ret;
1402
1403         if (!inode_should_reattach(&inode))
1404                 return 0;
1405
1406         ret = find_oldest_inode_needs_reattach(trans, &inode);
1407         if (ret)
1408                 return ret;
1409
1410         if (fsck_err(trans, inode_unreachable,
1411                      "unreachable inode:\n%s",
1412                      (bch2_inode_unpacked_to_text(&buf, &inode),
1413                       buf.buf)))
1414                 ret = reattach_inode(trans, &inode);
1415 fsck_err:
1416         printbuf_exit(&buf);
1417         return ret;
1418 }
1419
1420 /*
1421  * Reattach unreachable (but not unlinked) inodes
1422  *
1423  * Run after check_inodes() and check_dirents(), so we node that inode
1424  * backpointer fields point to valid dirents, and every inode that has a dirent
1425  * that points to it has its backpointer field set - so we're just looking for
1426  * non-unlinked inodes without backpointers:
1427  *
1428  * XXX: this is racy w.r.t. hardlink removal in online fsck
1429  */
1430 int bch2_check_unreachable_inodes(struct bch_fs *c)
1431 {
1432         int ret = bch2_trans_run(c,
1433                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1434                                 POS_MIN,
1435                                 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1436                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1437                         check_unreachable_inode(trans, &iter, k)));
1438         bch_err_fn(c, ret);
1439         return ret;
1440 }
1441
1442 static inline bool btree_matches_i_mode(enum btree_id btree, unsigned mode)
1443 {
1444         switch (btree) {
1445         case BTREE_ID_extents:
1446                 return S_ISREG(mode) || S_ISLNK(mode);
1447         case BTREE_ID_dirents:
1448                 return S_ISDIR(mode);
1449         case BTREE_ID_xattrs:
1450                 return true;
1451         default:
1452                 BUG();
1453         }
1454 }
1455
1456 static int check_key_has_inode(struct btree_trans *trans,
1457                                struct btree_iter *iter,
1458                                struct inode_walker *inode,
1459                                struct inode_walker_entry *i,
1460                                struct bkey_s_c k)
1461 {
1462         struct bch_fs *c = trans->c;
1463         struct printbuf buf = PRINTBUF;
1464         struct btree_iter iter2 = {};
1465         int ret = PTR_ERR_OR_ZERO(i);
1466         if (ret)
1467                 return ret;
1468
1469         if (k.k->type == KEY_TYPE_whiteout)
1470                 goto out;
1471
1472         bool have_inode = i && !i->whiteout;
1473
1474         if (!have_inode && (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_inodes)))
1475                 goto reconstruct;
1476
1477         if (have_inode && btree_matches_i_mode(iter->btree_id, i->inode.bi_mode))
1478                 goto out;
1479
1480         prt_printf(&buf, ", ");
1481
1482         bool have_old_inode = false;
1483         darray_for_each(inode->inodes, i2)
1484                 if (!i2->whiteout &&
1485                     bch2_snapshot_is_ancestor(c, k.k->p.snapshot, i2->inode.bi_snapshot) &&
1486                     btree_matches_i_mode(iter->btree_id, i2->inode.bi_mode)) {
1487                         prt_printf(&buf, "but found good inode in older snapshot\n");
1488                         bch2_inode_unpacked_to_text(&buf, &i2->inode);
1489                         prt_newline(&buf);
1490                         have_old_inode = true;
1491                         break;
1492                 }
1493
1494         struct bkey_s_c k2;
1495         unsigned nr_keys = 0;
1496
1497         prt_printf(&buf, "found keys:\n");
1498
1499         for_each_btree_key_max_norestart(trans, iter2, iter->btree_id,
1500                                          SPOS(k.k->p.inode, 0, k.k->p.snapshot),
1501                                          POS(k.k->p.inode, U64_MAX),
1502                                          0, k2, ret) {
1503                 nr_keys++;
1504                 if (nr_keys <= 10) {
1505                         bch2_bkey_val_to_text(&buf, c, k2);
1506                         prt_newline(&buf);
1507                 }
1508                 if (nr_keys >= 100)
1509                         break;
1510         }
1511
1512         if (ret)
1513                 goto err;
1514
1515         if (nr_keys > 100)
1516                 prt_printf(&buf, "found > %u keys for this missing inode\n", nr_keys);
1517         else if (nr_keys > 10)
1518                 prt_printf(&buf, "found %u keys for this missing inode\n", nr_keys);
1519
1520         if (!have_inode) {
1521                 if (fsck_err_on(!have_inode,
1522                                 trans, key_in_missing_inode,
1523                                 "key in missing inode%s", buf.buf)) {
1524                         /*
1525                          * Maybe a deletion that raced with data move, or something
1526                          * weird like that? But if we know the inode was deleted, or
1527                          * it's just a few keys, we can safely delete them.
1528                          *
1529                          * If it's many keys, we should probably recreate the inode
1530                          */
1531                         if (have_old_inode || nr_keys <= 2)
1532                                 goto delete;
1533                         else
1534                                 goto reconstruct;
1535                 }
1536         } else {
1537                 /*
1538                  * not autofix, this one would be a giant wtf - bit error in the
1539                  * inode corrupting i_mode?
1540                  *
1541                  * may want to try repairing inode instead of deleting
1542                  */
1543                 if (fsck_err_on(!btree_matches_i_mode(iter->btree_id, i->inode.bi_mode),
1544                                 trans, key_in_wrong_inode_type,
1545                                 "key for wrong inode mode %o%s",
1546                                 i->inode.bi_mode, buf.buf))
1547                         goto delete;
1548         }
1549 out:
1550 err:
1551 fsck_err:
1552         bch2_trans_iter_exit(trans, &iter2);
1553         printbuf_exit(&buf);
1554         bch_err_fn(c, ret);
1555         return ret;
1556 delete:
1557         /*
1558          * XXX: print out more info
1559          * count up extents for this inode, check if we have different inode in
1560          * an older snapshot version, perhaps decide if we want to reconstitute
1561          */
1562         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_internal_snapshot_node);
1563         goto out;
1564 reconstruct:
1565         ret =   reconstruct_inode(trans, iter->btree_id, k.k->p.snapshot, k.k->p.inode) ?:
1566                 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
1567         if (ret)
1568                 goto err;
1569
1570         inode->last_pos.inode--;
1571         ret = bch_err_throw(c, transaction_restart_nested);
1572         goto out;
1573 }
1574
1575 static int check_i_sectors_notnested(struct btree_trans *trans, struct inode_walker *w)
1576 {
1577         struct bch_fs *c = trans->c;
1578         int ret = 0;
1579         s64 count2;
1580
1581         darray_for_each(w->inodes, i) {
1582                 if (i->inode.bi_sectors == i->count)
1583                         continue;
1584
1585                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->inode.bi_snapshot);
1586
1587                 if (w->recalculate_sums)
1588                         i->count = count2;
1589
1590                 if (i->count != count2) {
1591                         bch_err_ratelimited(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1592                                             w->last_pos.inode, i->inode.bi_snapshot, i->count, count2);
1593                         i->count = count2;
1594                 }
1595
1596                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1597                                 trans, inode_i_sectors_wrong,
1598                                 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1599                                 w->last_pos.inode, i->inode.bi_snapshot,
1600                                 i->inode.bi_sectors, i->count)) {
1601                         i->inode.bi_sectors = i->count;
1602                         ret = bch2_fsck_write_inode(trans, &i->inode);
1603                         if (ret)
1604                                 break;
1605                 }
1606         }
1607 fsck_err:
1608         bch_err_fn(c, ret);
1609         return ret;
1610 }
1611
1612 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1613 {
1614         u32 restart_count = trans->restart_count;
1615         return check_i_sectors_notnested(trans, w) ?:
1616                 trans_was_restarted(trans, restart_count);
1617 }
1618
1619 struct extent_end {
1620         u32                     snapshot;
1621         u64                     offset;
1622         struct snapshots_seen   seen;
1623 };
1624
1625 struct extent_ends {
1626         struct bpos                     last_pos;
1627         DARRAY(struct extent_end)       e;
1628 };
1629
1630 static void extent_ends_reset(struct extent_ends *extent_ends)
1631 {
1632         darray_for_each(extent_ends->e, i)
1633                 snapshots_seen_exit(&i->seen);
1634         extent_ends->e.nr = 0;
1635 }
1636
1637 static void extent_ends_exit(struct extent_ends *extent_ends)
1638 {
1639         extent_ends_reset(extent_ends);
1640         darray_exit(&extent_ends->e);
1641 }
1642
1643 static void extent_ends_init(struct extent_ends *extent_ends)
1644 {
1645         memset(extent_ends, 0, sizeof(*extent_ends));
1646 }
1647
1648 static int extent_ends_at(struct bch_fs *c,
1649                           struct extent_ends *extent_ends,
1650                           struct snapshots_seen *seen,
1651                           struct bkey_s_c k)
1652 {
1653         struct extent_end *i, n = (struct extent_end) {
1654                 .offset         = k.k->p.offset,
1655                 .snapshot       = k.k->p.snapshot,
1656                 .seen           = *seen,
1657         };
1658
1659         n.seen.ids.data = kmemdup(seen->ids.data,
1660                               sizeof(seen->ids.data[0]) * seen->ids.size,
1661                               GFP_KERNEL);
1662         if (!n.seen.ids.data)
1663                 return bch_err_throw(c, ENOMEM_fsck_extent_ends_at);
1664
1665         __darray_for_each(extent_ends->e, i) {
1666                 if (i->snapshot == k.k->p.snapshot) {
1667                         snapshots_seen_exit(&i->seen);
1668                         *i = n;
1669                         return 0;
1670                 }
1671
1672                 if (i->snapshot >= k.k->p.snapshot)
1673                         break;
1674         }
1675
1676         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1677 }
1678
1679 static int overlapping_extents_found(struct btree_trans *trans,
1680                                      enum btree_id btree,
1681                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1682                                      struct bkey pos2,
1683                                      bool *fixed,
1684                                      struct extent_end *extent_end)
1685 {
1686         struct bch_fs *c = trans->c;
1687         struct printbuf buf = PRINTBUF;
1688         struct btree_iter iter1, iter2 = {};
1689         struct bkey_s_c k1, k2;
1690         int ret;
1691
1692         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1693
1694         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1695                              BTREE_ITER_all_snapshots|
1696                              BTREE_ITER_not_extents);
1697         k1 = bch2_btree_iter_peek_max(trans, &iter1, POS(pos1.inode, U64_MAX));
1698         ret = bkey_err(k1);
1699         if (ret)
1700                 goto err;
1701
1702         prt_newline(&buf);
1703         bch2_bkey_val_to_text(&buf, c, k1);
1704
1705         if (!bpos_eq(pos1, k1.k->p)) {
1706                 prt_str(&buf, "\nwanted\n  ");
1707                 bch2_bpos_to_text(&buf, pos1);
1708                 prt_str(&buf, "\n");
1709                 bch2_bkey_to_text(&buf, &pos2);
1710
1711                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1712                         __func__, buf.buf);
1713                 ret = bch_err_throw(c, internal_fsck_err);
1714                 goto err;
1715         }
1716
1717         bch2_trans_copy_iter(trans, &iter2, &iter1);
1718
1719         while (1) {
1720                 bch2_btree_iter_advance(trans, &iter2);
1721
1722                 k2 = bch2_btree_iter_peek_max(trans, &iter2, POS(pos1.inode, U64_MAX));
1723                 ret = bkey_err(k2);
1724                 if (ret)
1725                         goto err;
1726
1727                 if (bpos_ge(k2.k->p, pos2.p))
1728                         break;
1729         }
1730
1731         prt_newline(&buf);
1732         bch2_bkey_val_to_text(&buf, c, k2);
1733
1734         if (bpos_gt(k2.k->p, pos2.p) ||
1735             pos2.size != k2.k->size) {
1736                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1737                         __func__, buf.buf);
1738                 ret = bch_err_throw(c, internal_fsck_err);
1739                 goto err;
1740         }
1741
1742         prt_printf(&buf, "\noverwriting %s extent",
1743                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1744
1745         if (fsck_err(trans, extent_overlapping,
1746                      "overlapping extents%s", buf.buf)) {
1747                 struct btree_iter *old_iter = &iter1;
1748                 struct disk_reservation res = { 0 };
1749
1750                 if (pos1.snapshot < pos2.p.snapshot) {
1751                         old_iter = &iter2;
1752                         swap(k1, k2);
1753                 }
1754
1755                 trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
1756
1757                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1758                                 BTREE_UPDATE_internal_snapshot_node,
1759                                 k1, k2) ?:
1760                         bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1761                 bch2_disk_reservation_put(c, &res);
1762
1763                 bch_info(c, "repair ret %s", bch2_err_str(ret));
1764
1765                 if (ret)
1766                         goto err;
1767
1768                 *fixed = true;
1769
1770                 if (pos1.snapshot == pos2.p.snapshot) {
1771                         /*
1772                          * We overwrote the first extent, and did the overwrite
1773                          * in the same snapshot:
1774                          */
1775                         extent_end->offset = bkey_start_offset(&pos2);
1776                 } else if (pos1.snapshot > pos2.p.snapshot) {
1777                         /*
1778                          * We overwrote the first extent in pos2's snapshot:
1779                          */
1780                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1781                 } else {
1782                         /*
1783                          * We overwrote the second extent - restart
1784                          * check_extent() from the top:
1785                          */
1786                         ret = bch_err_throw(c, transaction_restart_nested);
1787                 }
1788         }
1789 fsck_err:
1790 err:
1791         bch2_trans_iter_exit(trans, &iter2);
1792         bch2_trans_iter_exit(trans, &iter1);
1793         printbuf_exit(&buf);
1794         return ret;
1795 }
1796
1797 static int check_overlapping_extents(struct btree_trans *trans,
1798                               struct snapshots_seen *seen,
1799                               struct extent_ends *extent_ends,
1800                               struct bkey_s_c k,
1801                               struct btree_iter *iter,
1802                               bool *fixed)
1803 {
1804         struct bch_fs *c = trans->c;
1805         int ret = 0;
1806
1807         /* transaction restart, running again */
1808         if (bpos_eq(extent_ends->last_pos, k.k->p))
1809                 return 0;
1810
1811         if (extent_ends->last_pos.inode != k.k->p.inode)
1812                 extent_ends_reset(extent_ends);
1813
1814         darray_for_each(extent_ends->e, i) {
1815                 if (i->offset <= bkey_start_offset(k.k))
1816                         continue;
1817
1818                 if (!ref_visible2(c,
1819                                   k.k->p.snapshot, seen,
1820                                   i->snapshot, &i->seen))
1821                         continue;
1822
1823                 ret = overlapping_extents_found(trans, iter->btree_id,
1824                                                 SPOS(iter->pos.inode,
1825                                                      i->offset,
1826                                                      i->snapshot),
1827                                                 &i->seen,
1828                                                 *k.k, fixed, i);
1829                 if (ret)
1830                         goto err;
1831         }
1832
1833         extent_ends->last_pos = k.k->p;
1834 err:
1835         return ret;
1836 }
1837
1838 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1839                                 struct bkey_s_c k)
1840 {
1841         struct bch_fs *c = trans->c;
1842         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1843         struct bch_extent_crc_unpacked crc;
1844         const union bch_extent_entry *i;
1845         unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1846
1847         bkey_for_each_crc(k.k, ptrs, crc, i)
1848                 if (crc_is_encoded(crc) &&
1849                     crc.uncompressed_size > encoded_extent_max_sectors) {
1850                         struct printbuf buf = PRINTBUF;
1851
1852                         bch2_bkey_val_to_text(&buf, c, k);
1853                         bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
1854                         printbuf_exit(&buf);
1855                 }
1856
1857         return 0;
1858 }
1859
1860 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1861                         struct bkey_s_c k,
1862                         struct inode_walker *inode,
1863                         struct snapshots_seen *s,
1864                         struct extent_ends *extent_ends,
1865                         struct disk_reservation *res)
1866 {
1867         struct bch_fs *c = trans->c;
1868         struct printbuf buf = PRINTBUF;
1869         int ret = 0;
1870
1871         ret = bch2_check_key_has_snapshot(trans, iter, k);
1872         if (ret) {
1873                 ret = ret < 0 ? ret : 0;
1874                 goto out;
1875         }
1876
1877         if (inode->last_pos.inode != k.k->p.inode && inode->have_inodes) {
1878                 ret = check_i_sectors(trans, inode);
1879                 if (ret)
1880                         goto err;
1881         }
1882
1883         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1884         if (ret)
1885                 goto err;
1886
1887         struct inode_walker_entry *extent_i = walk_inode(trans, inode, k);
1888         ret = PTR_ERR_OR_ZERO(extent_i);
1889         if (ret)
1890                 goto err;
1891
1892         ret = check_key_has_inode(trans, iter, inode, extent_i, k);
1893         if (ret)
1894                 goto err;
1895
1896         if (k.k->type != KEY_TYPE_whiteout) {
1897                 ret = check_overlapping_extents(trans, s, extent_ends, k, iter,
1898                                                 &inode->recalculate_sums);
1899                 if (ret)
1900                         goto err;
1901
1902                 /*
1903                  * Check inodes in reverse order, from oldest snapshots to
1904                  * newest, starting from the inode that matches this extent's
1905                  * snapshot. If we didn't have one, iterate over all inodes:
1906                  */
1907                 for (struct inode_walker_entry *i = extent_i ?: &darray_last(inode->inodes);
1908                      inode->inodes.data && i >= inode->inodes.data;
1909                      --i) {
1910                         if (i->inode.bi_snapshot > k.k->p.snapshot ||
1911                             !key_visible_in_snapshot(c, s, i->inode.bi_snapshot, k.k->p.snapshot))
1912                                 continue;
1913
1914                         u64 last_block = round_up(i->inode.bi_size, block_bytes(c)) >> 9;
1915
1916                         if (fsck_err_on(k.k->p.offset > last_block &&
1917                                         !bkey_extent_is_reservation(k),
1918                                         trans, extent_past_end_of_inode,
1919                                         "extent type past end of inode %llu:%u, i_size %llu\n%s",
1920                                         i->inode.bi_inum, i->inode.bi_snapshot, i->inode.bi_size,
1921                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1922                                 struct bkey_i *whiteout = bch2_trans_kmalloc(trans, sizeof(*whiteout));
1923                                 ret = PTR_ERR_OR_ZERO(whiteout);
1924                                 if (ret)
1925                                         goto err;
1926
1927                                 bkey_init(&whiteout->k);
1928                                 whiteout->k.p = SPOS(k.k->p.inode,
1929                                                      last_block,
1930                                                      i->inode.bi_snapshot);
1931                                 bch2_key_resize(&whiteout->k,
1932                                                 min(KEY_SIZE_MAX & (~0 << c->block_bits),
1933                                                     U64_MAX - whiteout->k.p.offset));
1934
1935
1936                                 /*
1937                                  * Need a normal (not BTREE_ITER_all_snapshots)
1938                                  * iterator, if we're deleting in a different
1939                                  * snapshot and need to emit a whiteout
1940                                  */
1941                                 struct btree_iter iter2;
1942                                 bch2_trans_iter_init(trans, &iter2, BTREE_ID_extents,
1943                                                      bkey_start_pos(&whiteout->k),
1944                                                      BTREE_ITER_intent);
1945                                 ret =   bch2_btree_iter_traverse(trans, &iter2) ?:
1946                                         bch2_trans_update(trans, &iter2, whiteout,
1947                                                 BTREE_UPDATE_internal_snapshot_node);
1948                                 bch2_trans_iter_exit(trans, &iter2);
1949                                 if (ret)
1950                                         goto err;
1951
1952                                 iter->k.type = KEY_TYPE_whiteout;
1953                                 break;
1954                         }
1955                 }
1956         }
1957
1958         ret = bch2_trans_commit(trans, res, NULL, BCH_TRANS_COMMIT_no_enospc);
1959         if (ret)
1960                 goto err;
1961
1962         if (bkey_extent_is_allocation(k.k)) {
1963                 for (struct inode_walker_entry *i = extent_i ?: &darray_last(inode->inodes);
1964                      inode->inodes.data && i >= inode->inodes.data;
1965                      --i) {
1966                         if (i->whiteout ||
1967                             i->inode.bi_snapshot > k.k->p.snapshot ||
1968                             !key_visible_in_snapshot(c, s, i->inode.bi_snapshot, k.k->p.snapshot))
1969                                 continue;
1970
1971                         i->count += k.k->size;
1972                 }
1973         }
1974
1975         if (k.k->type != KEY_TYPE_whiteout) {
1976                 ret = extent_ends_at(c, extent_ends, s, k);
1977                 if (ret)
1978                         goto err;
1979         }
1980 out:
1981 err:
1982 fsck_err:
1983         printbuf_exit(&buf);
1984         bch_err_fn(c, ret);
1985         return ret;
1986 }
1987
1988 /*
1989  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1990  * that i_size an i_sectors are consistent
1991  */
1992 int bch2_check_extents(struct bch_fs *c)
1993 {
1994         struct inode_walker w = inode_walker_init();
1995         struct snapshots_seen s;
1996         struct extent_ends extent_ends;
1997         struct disk_reservation res = { 0 };
1998
1999         snapshots_seen_init(&s);
2000         extent_ends_init(&extent_ends);
2001
2002         int ret = bch2_trans_run(c,
2003                 for_each_btree_key(trans, iter, BTREE_ID_extents,
2004                                 POS(BCACHEFS_ROOT_INO, 0),
2005                                 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k, ({
2006                         bch2_disk_reservation_put(c, &res);
2007                         check_extent(trans, &iter, k, &w, &s, &extent_ends, &res) ?:
2008                         check_extent_overbig(trans, &iter, k);
2009                 })) ?:
2010                 check_i_sectors_notnested(trans, &w));
2011
2012         bch2_disk_reservation_put(c, &res);
2013         extent_ends_exit(&extent_ends);
2014         inode_walker_exit(&w);
2015         snapshots_seen_exit(&s);
2016
2017         bch_err_fn(c, ret);
2018         return ret;
2019 }
2020
2021 int bch2_check_indirect_extents(struct bch_fs *c)
2022 {
2023         struct disk_reservation res = { 0 };
2024
2025         int ret = bch2_trans_run(c,
2026                 for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
2027                                 POS_MIN,
2028                                 BTREE_ITER_prefetch, k,
2029                                 &res, NULL,
2030                                 BCH_TRANS_COMMIT_no_enospc, ({
2031                         bch2_disk_reservation_put(c, &res);
2032                         check_extent_overbig(trans, &iter, k);
2033                 })));
2034
2035         bch2_disk_reservation_put(c, &res);
2036         bch_err_fn(c, ret);
2037         return ret;
2038 }
2039
2040 static int check_subdir_count_notnested(struct btree_trans *trans, struct inode_walker *w)
2041 {
2042         struct bch_fs *c = trans->c;
2043         int ret = 0;
2044         s64 count2;
2045
2046         darray_for_each(w->inodes, i) {
2047                 if (i->inode.bi_nlink == i->count)
2048                         continue;
2049
2050                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->inode.bi_snapshot);
2051                 if (count2 < 0)
2052                         return count2;
2053
2054                 if (i->count != count2) {
2055                         bch_err_ratelimited(c, "fsck counted subdirectories wrong for inum %llu:%u: got %llu should be %llu",
2056                                             w->last_pos.inode, i->inode.bi_snapshot, i->count, count2);
2057                         i->count = count2;
2058                         if (i->inode.bi_nlink == i->count)
2059                                 continue;
2060                 }
2061
2062                 if (i->inode.bi_nlink != i->count) {
2063                         CLASS(printbuf, buf)();
2064
2065                         lockrestart_do(trans,
2066                                        bch2_inum_snapshot_to_path(trans, w->last_pos.inode,
2067                                                                   i->inode.bi_snapshot, NULL, &buf));
2068
2069                         if (fsck_err_on(i->inode.bi_nlink != i->count,
2070                                         trans, inode_dir_wrong_nlink,
2071                                         "directory with wrong i_nlink: got %u, should be %llu\n%s",
2072                                         i->inode.bi_nlink, i->count, buf.buf)) {
2073                                 i->inode.bi_nlink = i->count;
2074                                 ret = bch2_fsck_write_inode(trans, &i->inode);
2075                                 if (ret)
2076                                         break;
2077                         }
2078                 }
2079         }
2080 fsck_err:
2081         bch_err_fn(c, ret);
2082         return ret;
2083 }
2084
2085 static int check_subdir_dirents_count(struct btree_trans *trans, struct inode_walker *w)
2086 {
2087         u32 restart_count = trans->restart_count;
2088         return check_subdir_count_notnested(trans, w) ?:
2089                 trans_was_restarted(trans, restart_count);
2090 }
2091
2092 /* find a subvolume that's a descendent of @snapshot: */
2093 static int find_snapshot_subvol(struct btree_trans *trans, u32 snapshot, u32 *subvolid)
2094 {
2095         struct btree_iter iter;
2096         struct bkey_s_c k;
2097         int ret;
2098
2099         for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN, 0, k, ret) {
2100                 if (k.k->type != KEY_TYPE_subvolume)
2101                         continue;
2102
2103                 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2104                 if (bch2_snapshot_is_ancestor(trans->c, le32_to_cpu(s.v->snapshot), snapshot)) {
2105                         bch2_trans_iter_exit(trans, &iter);
2106                         *subvolid = k.k->p.offset;
2107                         goto found;
2108                 }
2109         }
2110         if (!ret)
2111                 ret = -ENOENT;
2112 found:
2113         bch2_trans_iter_exit(trans, &iter);
2114         return ret;
2115 }
2116
2117 noinline_for_stack
2118 static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *iter,
2119                                   struct bkey_s_c_dirent d)
2120 {
2121         struct bch_fs *c = trans->c;
2122         struct btree_iter subvol_iter = {};
2123         struct bch_inode_unpacked subvol_root;
2124         u32 parent_subvol = le32_to_cpu(d.v->d_parent_subvol);
2125         u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
2126         u32 parent_snapshot;
2127         u32 new_parent_subvol = 0;
2128         u64 parent_inum;
2129         struct printbuf buf = PRINTBUF;
2130         int ret = 0;
2131
2132         ret = subvol_lookup(trans, parent_subvol, &parent_snapshot, &parent_inum);
2133         if (ret && !bch2_err_matches(ret, ENOENT))
2134                 return ret;
2135
2136         if (ret ||
2137             (!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot))) {
2138                 int ret2 = find_snapshot_subvol(trans, d.k->p.snapshot, &new_parent_subvol);
2139                 if (ret2 && !bch2_err_matches(ret, ENOENT))
2140                         return ret2;
2141         }
2142
2143         if (ret &&
2144             !new_parent_subvol &&
2145             (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_subvolumes))) {
2146                 /*
2147                  * Couldn't find a subvol for dirent's snapshot - but we lost
2148                  * subvols, so we need to reconstruct:
2149                  */
2150                 ret = reconstruct_subvol(trans, d.k->p.snapshot, parent_subvol, 0);
2151                 if (ret)
2152                         return ret;
2153
2154                 parent_snapshot = d.k->p.snapshot;
2155         }
2156
2157         if (fsck_err_on(ret,
2158                         trans, dirent_to_missing_parent_subvol,
2159                         "dirent parent_subvol points to missing subvolume\n%s",
2160                         (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)) ||
2161             fsck_err_on(!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot),
2162                         trans, dirent_not_visible_in_parent_subvol,
2163                         "dirent not visible in parent_subvol (not an ancestor of subvol snap %u)\n%s",
2164                         parent_snapshot,
2165                         (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
2166                 if (!new_parent_subvol) {
2167                         bch_err(c, "could not find a subvol for snapshot %u", d.k->p.snapshot);
2168                         return bch_err_throw(c, fsck_repair_unimplemented);
2169                 }
2170
2171                 struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
2172                 ret = PTR_ERR_OR_ZERO(new_dirent);
2173                 if (ret)
2174                         goto err;
2175
2176                 new_dirent->v.d_parent_subvol = cpu_to_le32(new_parent_subvol);
2177         }
2178
2179         struct bkey_s_c_subvolume s =
2180                 bch2_bkey_get_iter_typed(trans, &subvol_iter,
2181                                          BTREE_ID_subvolumes, POS(0, target_subvol),
2182                                          0, subvolume);
2183         ret = bkey_err(s.s_c);
2184         if (ret && !bch2_err_matches(ret, ENOENT))
2185                 goto err;
2186
2187         if (ret) {
2188                 if (fsck_err(trans, dirent_to_missing_subvol,
2189                              "dirent points to missing subvolume\n%s",
2190                              (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)))
2191                         return bch2_fsck_remove_dirent(trans, d.k->p);
2192                 ret = 0;
2193                 goto out;
2194         }
2195
2196         if (le32_to_cpu(s.v->fs_path_parent) != parent_subvol) {
2197                 printbuf_reset(&buf);
2198
2199                 prt_printf(&buf, "subvol with wrong fs_path_parent, should be be %u\n",
2200                            parent_subvol);
2201
2202                 ret = bch2_inum_to_path(trans, (subvol_inum) { s.k->p.offset,
2203                                         le64_to_cpu(s.v->inode) }, &buf);
2204                 if (ret)
2205                         goto err;
2206                 prt_newline(&buf);
2207                 bch2_bkey_val_to_text(&buf, c, s.s_c);
2208
2209                 if (fsck_err(trans, subvol_fs_path_parent_wrong, "%s", buf.buf)) {
2210                         struct bkey_i_subvolume *n =
2211                                 bch2_bkey_make_mut_typed(trans, &subvol_iter, &s.s_c, 0, subvolume);
2212                         ret = PTR_ERR_OR_ZERO(n);
2213                         if (ret)
2214                                 goto err;
2215
2216                         n->v.fs_path_parent = cpu_to_le32(parent_subvol);
2217                 }
2218         }
2219
2220         u64 target_inum = le64_to_cpu(s.v->inode);
2221         u32 target_snapshot = le32_to_cpu(s.v->snapshot);
2222
2223         ret = bch2_inode_find_by_inum_snapshot(trans, target_inum, target_snapshot,
2224                                                &subvol_root, 0);
2225         if (ret && !bch2_err_matches(ret, ENOENT))
2226                 goto err;
2227
2228         if (ret) {
2229                 bch_err(c, "subvol %u points to missing inode root %llu", target_subvol, target_inum);
2230                 ret = bch_err_throw(c, fsck_repair_unimplemented);
2231                 goto err;
2232         }
2233
2234         if (fsck_err_on(!ret && parent_subvol != subvol_root.bi_parent_subvol,
2235                         trans, inode_bi_parent_wrong,
2236                         "subvol root %llu has wrong bi_parent_subvol: got %u, should be %u",
2237                         target_inum,
2238                         subvol_root.bi_parent_subvol, parent_subvol)) {
2239                 subvol_root.bi_parent_subvol = parent_subvol;
2240                 subvol_root.bi_snapshot = le32_to_cpu(s.v->snapshot);
2241                 ret = __bch2_fsck_write_inode(trans, &subvol_root);
2242                 if (ret)
2243                         goto err;
2244         }
2245
2246         ret = bch2_check_dirent_target(trans, iter, d, &subvol_root, true);
2247         if (ret)
2248                 goto err;
2249 out:
2250 err:
2251 fsck_err:
2252         bch2_trans_iter_exit(trans, &subvol_iter);
2253         printbuf_exit(&buf);
2254         return ret;
2255 }
2256
2257 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
2258                         struct bkey_s_c k,
2259                         struct bch_hash_info *hash_info,
2260                         struct inode_walker *dir,
2261                         struct inode_walker *target,
2262                         struct snapshots_seen *s,
2263                         bool *need_second_pass)
2264 {
2265         struct bch_fs *c = trans->c;
2266         struct inode_walker_entry *i;
2267         struct printbuf buf = PRINTBUF;
2268         int ret = 0;
2269
2270         ret = bch2_check_key_has_snapshot(trans, iter, k);
2271         if (ret) {
2272                 ret = ret < 0 ? ret : 0;
2273                 goto out;
2274         }
2275
2276         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
2277         if (ret)
2278                 goto err;
2279
2280         if (k.k->type == KEY_TYPE_whiteout)
2281                 goto out;
2282
2283         if (dir->last_pos.inode != k.k->p.inode && dir->have_inodes) {
2284                 ret = check_subdir_dirents_count(trans, dir);
2285                 if (ret)
2286                         goto err;
2287         }
2288
2289         i = walk_inode(trans, dir, k);
2290         ret = PTR_ERR_OR_ZERO(i);
2291         if (ret < 0)
2292                 goto err;
2293
2294         ret = check_key_has_inode(trans, iter, dir, i, k);
2295         if (ret)
2296                 goto err;
2297
2298         if (!i || i->whiteout)
2299                 goto out;
2300
2301         if (dir->first_this_inode)
2302                 *hash_info = bch2_hash_info_init(c, &i->inode);
2303         dir->first_this_inode = false;
2304
2305         hash_info->cf_encoding = bch2_inode_casefold(c, &i->inode) ? c->cf_encoding : NULL;
2306
2307         ret = bch2_str_hash_check_key(trans, s, &bch2_dirent_hash_desc, hash_info,
2308                                       iter, k, need_second_pass);
2309         if (ret < 0)
2310                 goto err;
2311         if (ret) {
2312                 /* dirent has been deleted */
2313                 ret = 0;
2314                 goto out;
2315         }
2316
2317         if (k.k->type != KEY_TYPE_dirent)
2318                 goto out;
2319
2320         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2321
2322         /* check casefold */
2323         if (fsck_err_on(d.v->d_casefold != !!hash_info->cf_encoding,
2324                         trans, dirent_casefold_mismatch,
2325                         "dirent casefold does not match dir casefold\n%s",
2326                         (printbuf_reset(&buf),
2327                          bch2_bkey_val_to_text(&buf, c, k),
2328                          buf.buf))) {
2329                 subvol_inum dir_inum = { .subvol = d.v->d_type == DT_SUBVOL
2330                                 ? le32_to_cpu(d.v->d_parent_subvol)
2331                                 : 0,
2332                 };
2333                 u64 target = d.v->d_type == DT_SUBVOL
2334                         ? le32_to_cpu(d.v->d_child_subvol)
2335                         : le64_to_cpu(d.v->d_inum);
2336                 struct qstr name = bch2_dirent_get_name(d);
2337
2338                 struct bkey_i_dirent *new_d =
2339                         bch2_dirent_create_key(trans, hash_info, dir_inum,
2340                                                d.v->d_type, &name, NULL, target);
2341                 ret = PTR_ERR_OR_ZERO(new_d);
2342                 if (ret)
2343                         goto out;
2344
2345                 new_d->k.p.inode        = d.k->p.inode;
2346                 new_d->k.p.snapshot     = d.k->p.snapshot;
2347
2348                 struct btree_iter dup_iter = {};
2349                 ret =   bch2_hash_delete_at(trans,
2350                                             bch2_dirent_hash_desc, hash_info, iter,
2351                                             BTREE_UPDATE_internal_snapshot_node) ?:
2352                         bch2_str_hash_repair_key(trans, s,
2353                                                  &bch2_dirent_hash_desc, hash_info,
2354                                                  iter, bkey_i_to_s_c(&new_d->k_i),
2355                                                  &dup_iter, bkey_s_c_null,
2356                                                  need_second_pass);
2357                 goto out;
2358         }
2359
2360         if (d.v->d_type == DT_SUBVOL) {
2361                 ret = check_dirent_to_subvol(trans, iter, d);
2362                 if (ret)
2363                         goto err;
2364         } else {
2365                 ret = get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
2366                 if (ret)
2367                         goto err;
2368
2369                 if (fsck_err_on(!target->inodes.nr,
2370                                 trans, dirent_to_missing_inode,
2371                                 "dirent points to missing inode:\n%s",
2372                                 (printbuf_reset(&buf),
2373                                  bch2_bkey_val_to_text(&buf, c, k),
2374                                  buf.buf))) {
2375                         ret = bch2_fsck_remove_dirent(trans, d.k->p);
2376                         if (ret)
2377                                 goto err;
2378                 }
2379
2380                 darray_for_each(target->inodes, i) {
2381                         ret = bch2_check_dirent_target(trans, iter, d, &i->inode, true);
2382                         if (ret)
2383                                 goto err;
2384                 }
2385
2386                 darray_for_each(target->deletes, i)
2387                         if (fsck_err_on(!snapshot_list_has_id(&s->ids, *i),
2388                                         trans, dirent_to_overwritten_inode,
2389                                         "dirent points to inode overwritten in snapshot %u:\n%s",
2390                                         *i,
2391                                         (printbuf_reset(&buf),
2392                                          bch2_bkey_val_to_text(&buf, c, k),
2393                                          buf.buf))) {
2394                                 struct btree_iter delete_iter;
2395                                 bch2_trans_iter_init(trans, &delete_iter,
2396                                                      BTREE_ID_dirents,
2397                                                      SPOS(k.k->p.inode, k.k->p.offset, *i),
2398                                                      BTREE_ITER_intent);
2399                                 ret =   bch2_btree_iter_traverse(trans, &delete_iter) ?:
2400                                         bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
2401                                                           hash_info,
2402                                                           &delete_iter,
2403                                                           BTREE_UPDATE_internal_snapshot_node);
2404                                 bch2_trans_iter_exit(trans, &delete_iter);
2405                                 if (ret)
2406                                         goto err;
2407
2408                         }
2409         }
2410
2411         ret = bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
2412         if (ret)
2413                 goto err;
2414
2415         for_each_visible_inode(c, s, dir, d.k->p.snapshot, i) {
2416                 if (d.v->d_type == DT_DIR)
2417                         i->count++;
2418                 i->i_size += bkey_bytes(d.k);
2419         }
2420 out:
2421 err:
2422 fsck_err:
2423         printbuf_exit(&buf);
2424         return ret;
2425 }
2426
2427 /*
2428  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
2429  * validate d_type
2430  */
2431 int bch2_check_dirents(struct bch_fs *c)
2432 {
2433         struct inode_walker dir = inode_walker_init();
2434         struct inode_walker target = inode_walker_init();
2435         struct snapshots_seen s;
2436         struct bch_hash_info hash_info;
2437         bool need_second_pass = false, did_second_pass = false;
2438         int ret;
2439
2440         snapshots_seen_init(&s);
2441 again:
2442         ret = bch2_trans_run(c,
2443                 for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
2444                                 POS(BCACHEFS_ROOT_INO, 0),
2445                                 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
2446                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2447                         check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s,
2448                                      &need_second_pass)) ?:
2449                 check_subdir_count_notnested(trans, &dir));
2450
2451         if (!ret && need_second_pass && !did_second_pass) {
2452                 bch_info(c, "check_dirents requires second pass");
2453                 swap(did_second_pass, need_second_pass);
2454                 goto again;
2455         }
2456
2457         if (!ret && need_second_pass) {
2458                 bch_err(c, "dirents not repairing");
2459                 ret = -EINVAL;
2460         }
2461
2462         snapshots_seen_exit(&s);
2463         inode_walker_exit(&dir);
2464         inode_walker_exit(&target);
2465         bch_err_fn(c, ret);
2466         return ret;
2467 }
2468
2469 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
2470                        struct bkey_s_c k,
2471                        struct bch_hash_info *hash_info,
2472                        struct inode_walker *inode)
2473 {
2474         struct bch_fs *c = trans->c;
2475
2476         int ret = bch2_check_key_has_snapshot(trans, iter, k);
2477         if (ret < 0)
2478                 return ret;
2479         if (ret)
2480                 return 0;
2481
2482         struct inode_walker_entry *i = walk_inode(trans, inode, k);
2483         ret = PTR_ERR_OR_ZERO(i);
2484         if (ret)
2485                 return ret;
2486
2487         ret = check_key_has_inode(trans, iter, inode, i, k);
2488         if (ret)
2489                 return ret;
2490
2491         if (!i || i->whiteout)
2492                 return 0;
2493
2494         if (inode->first_this_inode)
2495                 *hash_info = bch2_hash_info_init(c, &i->inode);
2496         inode->first_this_inode = false;
2497
2498         bool need_second_pass = false;
2499         return bch2_str_hash_check_key(trans, NULL, &bch2_xattr_hash_desc, hash_info,
2500                                       iter, k, &need_second_pass);
2501 }
2502
2503 /*
2504  * Walk xattrs: verify that they all have a corresponding inode
2505  */
2506 int bch2_check_xattrs(struct bch_fs *c)
2507 {
2508         struct inode_walker inode = inode_walker_init();
2509         struct bch_hash_info hash_info;
2510         int ret = 0;
2511
2512         ret = bch2_trans_run(c,
2513                 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
2514                         POS(BCACHEFS_ROOT_INO, 0),
2515                         BTREE_ITER_prefetch|BTREE_ITER_all_snapshots,
2516                         k,
2517                         NULL, NULL,
2518                         BCH_TRANS_COMMIT_no_enospc,
2519                 check_xattr(trans, &iter, k, &hash_info, &inode)));
2520
2521         inode_walker_exit(&inode);
2522         bch_err_fn(c, ret);
2523         return ret;
2524 }
2525
2526 static int check_root_trans(struct btree_trans *trans)
2527 {
2528         struct bch_fs *c = trans->c;
2529         struct bch_inode_unpacked root_inode;
2530         u32 snapshot;
2531         u64 inum;
2532         int ret;
2533
2534         ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
2535         if (ret && !bch2_err_matches(ret, ENOENT))
2536                 return ret;
2537
2538         if (mustfix_fsck_err_on(ret, trans, root_subvol_missing,
2539                                 "root subvol missing")) {
2540                 struct bkey_i_subvolume *root_subvol =
2541                         bch2_trans_kmalloc(trans, sizeof(*root_subvol));
2542                 ret = PTR_ERR_OR_ZERO(root_subvol);
2543                 if (ret)
2544                         goto err;
2545
2546                 snapshot        = U32_MAX;
2547                 inum            = BCACHEFS_ROOT_INO;
2548
2549                 bkey_subvolume_init(&root_subvol->k_i);
2550                 root_subvol->k.p.offset = BCACHEFS_ROOT_SUBVOL;
2551                 root_subvol->v.flags    = 0;
2552                 root_subvol->v.snapshot = cpu_to_le32(snapshot);
2553                 root_subvol->v.inode    = cpu_to_le64(inum);
2554                 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol->k_i, 0);
2555                 bch_err_msg(c, ret, "writing root subvol");
2556                 if (ret)
2557                         goto err;
2558         }
2559
2560         ret = bch2_inode_find_by_inum_snapshot(trans, BCACHEFS_ROOT_INO, snapshot,
2561                                                &root_inode, 0);
2562         if (ret && !bch2_err_matches(ret, ENOENT))
2563                 return ret;
2564
2565         if (mustfix_fsck_err_on(ret,
2566                                 trans, root_dir_missing,
2567                                 "root directory missing") ||
2568             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
2569                                 trans, root_inode_not_dir,
2570                                 "root inode not a directory")) {
2571                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
2572                                 0, NULL);
2573                 root_inode.bi_inum = inum;
2574                 root_inode.bi_snapshot = snapshot;
2575
2576                 ret = __bch2_fsck_write_inode(trans, &root_inode);
2577                 bch_err_msg(c, ret, "writing root inode");
2578         }
2579 err:
2580 fsck_err:
2581         return ret;
2582 }
2583
2584 /* Get root directory, create if it doesn't exist: */
2585 int bch2_check_root(struct bch_fs *c)
2586 {
2587         int ret = bch2_trans_commit_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2588                 check_root_trans(trans));
2589         bch_err_fn(c, ret);
2590         return ret;
2591 }
2592
2593 static bool darray_u32_has(darray_u32 *d, u32 v)
2594 {
2595         darray_for_each(*d, i)
2596                 if (*i == v)
2597                         return true;
2598         return false;
2599 }
2600
2601 static int check_subvol_path(struct btree_trans *trans, struct btree_iter *iter, struct bkey_s_c k)
2602 {
2603         struct bch_fs *c = trans->c;
2604         struct btree_iter parent_iter = {};
2605         darray_u32 subvol_path = {};
2606         struct printbuf buf = PRINTBUF;
2607         int ret = 0;
2608
2609         if (k.k->type != KEY_TYPE_subvolume)
2610                 return 0;
2611
2612         subvol_inum start = {
2613                 .subvol = k.k->p.offset,
2614                 .inum   = le64_to_cpu(bkey_s_c_to_subvolume(k).v->inode),
2615         };
2616
2617         while (k.k->p.offset != BCACHEFS_ROOT_SUBVOL) {
2618                 ret = darray_push(&subvol_path, k.k->p.offset);
2619                 if (ret)
2620                         goto err;
2621
2622                 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2623
2624                 struct bch_inode_unpacked subvol_root;
2625                 ret = bch2_inode_find_by_inum_trans(trans,
2626                                         (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
2627                                         &subvol_root);
2628                 if (ret)
2629                         break;
2630
2631                 u32 parent = le32_to_cpu(s.v->fs_path_parent);
2632
2633                 if (darray_u32_has(&subvol_path, parent)) {
2634                         printbuf_reset(&buf);
2635                         prt_printf(&buf, "subvolume loop: ");
2636
2637                         ret = bch2_inum_to_path(trans, start, &buf);
2638                         if (ret)
2639                                 goto err;
2640
2641                         if (fsck_err(trans, subvol_loop, "%s", buf.buf))
2642                                 ret = reattach_subvol(trans, s);
2643                         break;
2644                 }
2645
2646                 bch2_trans_iter_exit(trans, &parent_iter);
2647                 bch2_trans_iter_init(trans, &parent_iter,
2648                                      BTREE_ID_subvolumes, POS(0, parent), 0);
2649                 k = bch2_btree_iter_peek_slot(trans, &parent_iter);
2650                 ret = bkey_err(k);
2651                 if (ret)
2652                         goto err;
2653
2654                 if (fsck_err_on(k.k->type != KEY_TYPE_subvolume,
2655                                 trans, subvol_unreachable,
2656                                 "unreachable subvolume %s",
2657                                 (printbuf_reset(&buf),
2658                                  bch2_bkey_val_to_text(&buf, c, s.s_c),
2659                                  buf.buf))) {
2660                         ret = reattach_subvol(trans, s);
2661                         break;
2662                 }
2663         }
2664 fsck_err:
2665 err:
2666         printbuf_exit(&buf);
2667         darray_exit(&subvol_path);
2668         bch2_trans_iter_exit(trans, &parent_iter);
2669         return ret;
2670 }
2671
2672 int bch2_check_subvolume_structure(struct bch_fs *c)
2673 {
2674         int ret = bch2_trans_run(c,
2675                 for_each_btree_key_commit(trans, iter,
2676                                 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_prefetch, k,
2677                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2678                         check_subvol_path(trans, &iter, k)));
2679         bch_err_fn(c, ret);
2680         return ret;
2681 }
2682
2683 static int bch2_bi_depth_renumber_one(struct btree_trans *trans,
2684                                       u64 inum, u32 snapshot,
2685                                       u32 new_depth)
2686 {
2687         struct btree_iter iter;
2688         struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
2689                                                SPOS(0, inum, snapshot), 0);
2690
2691         struct bch_inode_unpacked inode;
2692         int ret = bkey_err(k) ?:
2693                 !bkey_is_inode(k.k) ? -BCH_ERR_ENOENT_inode
2694                 : bch2_inode_unpack(k, &inode);
2695         if (ret)
2696                 goto err;
2697
2698         if (inode.bi_depth != new_depth) {
2699                 inode.bi_depth = new_depth;
2700                 ret = __bch2_fsck_write_inode(trans, &inode) ?:
2701                         bch2_trans_commit(trans, NULL, NULL, 0);
2702         }
2703 err:
2704         bch2_trans_iter_exit(trans, &iter);
2705         return ret;
2706 }
2707
2708 static int bch2_bi_depth_renumber(struct btree_trans *trans, darray_u64 *path,
2709                                   u32 snapshot, u32 new_bi_depth)
2710 {
2711         u32 restart_count = trans->restart_count;
2712         int ret = 0;
2713
2714         darray_for_each_reverse(*path, i) {
2715                 ret = nested_lockrestart_do(trans,
2716                                 bch2_bi_depth_renumber_one(trans, *i, snapshot, new_bi_depth));
2717                 bch_err_fn(trans->c, ret);
2718                 if (ret)
2719                         break;
2720
2721                 new_bi_depth++;
2722         }
2723
2724         return ret ?: trans_was_restarted(trans, restart_count);
2725 }
2726
2727 static int check_path_loop(struct btree_trans *trans, struct bkey_s_c inode_k)
2728 {
2729         struct bch_fs *c = trans->c;
2730         struct btree_iter inode_iter = {};
2731         darray_u64 path = {};
2732         struct printbuf buf = PRINTBUF;
2733         u32 snapshot = inode_k.k->p.snapshot;
2734         bool redo_bi_depth = false;
2735         u32 min_bi_depth = U32_MAX;
2736         int ret = 0;
2737
2738         struct bpos start = inode_k.k->p;
2739
2740         struct bch_inode_unpacked inode;
2741         ret = bch2_inode_unpack(inode_k, &inode);
2742         if (ret)
2743                 return ret;
2744
2745         /*
2746          * If we're running full fsck, check_dirents() will have already ran,
2747          * and we shouldn't see any missing backpointers here - otherwise that's
2748          * handled separately, by check_unreachable_inodes
2749          */
2750         while (!inode.bi_subvol &&
2751                bch2_inode_has_backpointer(&inode)) {
2752                 struct btree_iter dirent_iter;
2753                 struct bkey_s_c_dirent d;
2754
2755                 d = dirent_get_by_pos(trans, &dirent_iter,
2756                                       SPOS(inode.bi_dir, inode.bi_dir_offset, snapshot));
2757                 ret = bkey_err(d.s_c);
2758                 if (ret && !bch2_err_matches(ret, ENOENT))
2759                         goto out;
2760
2761                 if (!ret && (ret = dirent_points_to_inode(c, d, &inode)))
2762                         bch2_trans_iter_exit(trans, &dirent_iter);
2763
2764                 if (bch2_err_matches(ret, ENOENT)) {
2765                         printbuf_reset(&buf);
2766                         bch2_bkey_val_to_text(&buf, c, inode_k);
2767                         bch_err(c, "unreachable inode in check_directory_structure: %s\n%s",
2768                                 bch2_err_str(ret), buf.buf);
2769                         goto out;
2770                 }
2771
2772                 bch2_trans_iter_exit(trans, &dirent_iter);
2773
2774                 ret = darray_push(&path, inode.bi_inum);
2775                 if (ret)
2776                         return ret;
2777
2778                 bch2_trans_iter_exit(trans, &inode_iter);
2779                 inode_k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes,
2780                                              SPOS(0, inode.bi_dir, snapshot), 0);
2781
2782                 struct bch_inode_unpacked parent_inode;
2783                 ret = bkey_err(inode_k) ?:
2784                         !bkey_is_inode(inode_k.k) ? -BCH_ERR_ENOENT_inode
2785                         : bch2_inode_unpack(inode_k, &parent_inode);
2786                 if (ret) {
2787                         /* Should have been caught in dirents pass */
2788                         bch_err_msg(c, ret, "error looking up parent directory");
2789                         goto out;
2790                 }
2791
2792                 min_bi_depth = parent_inode.bi_depth;
2793
2794                 if (parent_inode.bi_depth < inode.bi_depth &&
2795                     min_bi_depth < U16_MAX)
2796                         break;
2797
2798                 inode = parent_inode;
2799                 redo_bi_depth = true;
2800
2801                 if (darray_find(path, inode.bi_inum)) {
2802                         printbuf_reset(&buf);
2803                         prt_printf(&buf, "directory structure loop in snapshot %u: ",
2804                                    snapshot);
2805
2806                         ret = bch2_inum_snapshot_to_path(trans, start.offset, start.snapshot, NULL, &buf);
2807                         if (ret)
2808                                 goto out;
2809
2810                         if (c->opts.verbose) {
2811                                 prt_newline(&buf);
2812                                 darray_for_each(path, i)
2813                                         prt_printf(&buf, "%llu ", *i);
2814                         }
2815
2816                         if (fsck_err(trans, dir_loop, "%s", buf.buf)) {
2817                                 ret = remove_backpointer(trans, &inode);
2818                                 bch_err_msg(c, ret, "removing dirent");
2819                                 if (ret)
2820                                         goto out;
2821
2822                                 ret = reattach_inode(trans, &inode);
2823                                 bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
2824                         }
2825
2826                         goto out;
2827                 }
2828         }
2829
2830         if (inode.bi_subvol)
2831                 min_bi_depth = 0;
2832
2833         if (redo_bi_depth)
2834                 ret = bch2_bi_depth_renumber(trans, &path, snapshot, min_bi_depth);
2835 out:
2836 fsck_err:
2837         bch2_trans_iter_exit(trans, &inode_iter);
2838         darray_exit(&path);
2839         printbuf_exit(&buf);
2840         bch_err_fn(c, ret);
2841         return ret;
2842 }
2843
2844 /*
2845  * Check for loops in the directory structure: all other connectivity issues
2846  * have been fixed by prior passes
2847  */
2848 int bch2_check_directory_structure(struct bch_fs *c)
2849 {
2850         int ret = bch2_trans_run(c,
2851                 for_each_btree_key_reverse_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
2852                                           BTREE_ITER_intent|
2853                                           BTREE_ITER_prefetch|
2854                                           BTREE_ITER_all_snapshots, k,
2855                                           NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
2856                         if (!S_ISDIR(bkey_inode_mode(k)))
2857                                 continue;
2858
2859                         if (bch2_inode_flags(k) & BCH_INODE_unlinked)
2860                                 continue;
2861
2862                         check_path_loop(trans, k);
2863                 })));
2864
2865         bch_err_fn(c, ret);
2866         return ret;
2867 }
2868
2869 struct nlink_table {
2870         size_t          nr;
2871         size_t          size;
2872
2873         struct nlink {
2874                 u64     inum;
2875                 u32     snapshot;
2876                 u32     count;
2877         }               *d;
2878 };
2879
2880 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2881                      u64 inum, u32 snapshot)
2882 {
2883         if (t->nr == t->size) {
2884                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2885                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2886
2887                 if (!d) {
2888                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2889                                 new_size);
2890                         return bch_err_throw(c, ENOMEM_fsck_add_nlink);
2891                 }
2892
2893                 if (t->d)
2894                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2895                 kvfree(t->d);
2896
2897                 t->d = d;
2898                 t->size = new_size;
2899         }
2900
2901
2902         t->d[t->nr++] = (struct nlink) {
2903                 .inum           = inum,
2904                 .snapshot       = snapshot,
2905         };
2906
2907         return 0;
2908 }
2909
2910 static int nlink_cmp(const void *_l, const void *_r)
2911 {
2912         const struct nlink *l = _l;
2913         const struct nlink *r = _r;
2914
2915         return cmp_int(l->inum, r->inum);
2916 }
2917
2918 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2919                      struct nlink_table *links,
2920                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2921 {
2922         struct nlink *link, key = {
2923                 .inum = inum, .snapshot = U32_MAX,
2924         };
2925
2926         if (inum < range_start || inum >= range_end)
2927                 return;
2928
2929         link = __inline_bsearch(&key, links->d, links->nr,
2930                                 sizeof(links->d[0]), nlink_cmp);
2931         if (!link)
2932                 return;
2933
2934         while (link > links->d && link[0].inum == link[-1].inum)
2935                 --link;
2936
2937         for (; link < links->d + links->nr && link->inum == inum; link++)
2938                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2939                         link->count++;
2940                         if (link->snapshot >= snapshot)
2941                                 break;
2942                 }
2943 }
2944
2945 noinline_for_stack
2946 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2947                                        struct nlink_table *t,
2948                                        u64 start, u64 *end)
2949 {
2950         int ret = bch2_trans_run(c,
2951                 for_each_btree_key(trans, iter, BTREE_ID_inodes,
2952                                    POS(0, start),
2953                                    BTREE_ITER_intent|
2954                                    BTREE_ITER_prefetch|
2955                                    BTREE_ITER_all_snapshots, k, ({
2956                         if (!bkey_is_inode(k.k))
2957                                 continue;
2958
2959                         /* Should never fail, checked by bch2_inode_invalid: */
2960                         struct bch_inode_unpacked u;
2961                         _ret3 = bch2_inode_unpack(k, &u);
2962                         if (_ret3)
2963                                 break;
2964
2965                         /*
2966                          * Backpointer and directory structure checks are sufficient for
2967                          * directories, since they can't have hardlinks:
2968                          */
2969                         if (S_ISDIR(u.bi_mode))
2970                                 continue;
2971
2972                         /*
2973                          * Previous passes ensured that bi_nlink is nonzero if
2974                          * it had multiple hardlinks:
2975                          */
2976                         if (!u.bi_nlink)
2977                                 continue;
2978
2979                         ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2980                         if (ret) {
2981                                 *end = k.k->p.offset;
2982                                 ret = 0;
2983                                 break;
2984                         }
2985                         0;
2986                 })));
2987
2988         bch_err_fn(c, ret);
2989         return ret;
2990 }
2991
2992 noinline_for_stack
2993 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2994                                      u64 range_start, u64 range_end)
2995 {
2996         struct snapshots_seen s;
2997
2998         snapshots_seen_init(&s);
2999
3000         int ret = bch2_trans_run(c,
3001                 for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
3002                                    BTREE_ITER_intent|
3003                                    BTREE_ITER_prefetch|
3004                                    BTREE_ITER_all_snapshots, k, ({
3005                         ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
3006                         if (ret)
3007                                 break;
3008
3009                         if (k.k->type == KEY_TYPE_dirent) {
3010                                 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
3011
3012                                 if (d.v->d_type != DT_DIR &&
3013                                     d.v->d_type != DT_SUBVOL)
3014                                         inc_link(c, &s, links, range_start, range_end,
3015                                                  le64_to_cpu(d.v->d_inum), d.k->p.snapshot);
3016                         }
3017                         0;
3018                 })));
3019
3020         snapshots_seen_exit(&s);
3021
3022         bch_err_fn(c, ret);
3023         return ret;
3024 }
3025
3026 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
3027                                      struct bkey_s_c k,
3028                                      struct nlink_table *links,
3029                                      size_t *idx, u64 range_end)
3030 {
3031         struct bch_inode_unpacked u;
3032         struct nlink *link = &links->d[*idx];
3033         int ret = 0;
3034
3035         if (k.k->p.offset >= range_end)
3036                 return 1;
3037
3038         if (!bkey_is_inode(k.k))
3039                 return 0;
3040
3041         ret = bch2_inode_unpack(k, &u);
3042         if (ret)
3043                 return ret;
3044
3045         if (S_ISDIR(u.bi_mode))
3046                 return 0;
3047
3048         if (!u.bi_nlink)
3049                 return 0;
3050
3051         while ((cmp_int(link->inum, k.k->p.offset) ?:
3052                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
3053                 BUG_ON(*idx == links->nr);
3054                 link = &links->d[++*idx];
3055         }
3056
3057         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
3058                         trans, inode_wrong_nlink,
3059                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
3060                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
3061                         bch2_inode_nlink_get(&u), link->count)) {
3062                 bch2_inode_nlink_set(&u, link->count);
3063                 ret = __bch2_fsck_write_inode(trans, &u);
3064         }
3065 fsck_err:
3066         return ret;
3067 }
3068
3069 noinline_for_stack
3070 static int check_nlinks_update_hardlinks(struct bch_fs *c,
3071                                struct nlink_table *links,
3072                                u64 range_start, u64 range_end)
3073 {
3074         size_t idx = 0;
3075
3076         int ret = bch2_trans_run(c,
3077                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
3078                                 POS(0, range_start),
3079                                 BTREE_ITER_intent|BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
3080                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
3081                         check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
3082         if (ret < 0) {
3083                 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
3084                 return ret;
3085         }
3086
3087         return 0;
3088 }
3089
3090 int bch2_check_nlinks(struct bch_fs *c)
3091 {
3092         struct nlink_table links = { 0 };
3093         u64 this_iter_range_start, next_iter_range_start = 0;
3094         int ret = 0;
3095
3096         do {
3097                 this_iter_range_start = next_iter_range_start;
3098                 next_iter_range_start = U64_MAX;
3099
3100                 ret = check_nlinks_find_hardlinks(c, &links,
3101                                                   this_iter_range_start,
3102                                                   &next_iter_range_start);
3103
3104                 ret = check_nlinks_walk_dirents(c, &links,
3105                                           this_iter_range_start,
3106                                           next_iter_range_start);
3107                 if (ret)
3108                         break;
3109
3110                 ret = check_nlinks_update_hardlinks(c, &links,
3111                                          this_iter_range_start,
3112                                          next_iter_range_start);
3113                 if (ret)
3114                         break;
3115
3116                 links.nr = 0;
3117         } while (next_iter_range_start != U64_MAX);
3118
3119         kvfree(links.d);
3120         bch_err_fn(c, ret);
3121         return ret;
3122 }
3123
3124 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
3125                              struct bkey_s_c k)
3126 {
3127         struct bkey_s_c_reflink_p p;
3128         struct bkey_i_reflink_p *u;
3129
3130         if (k.k->type != KEY_TYPE_reflink_p)
3131                 return 0;
3132
3133         p = bkey_s_c_to_reflink_p(k);
3134
3135         if (!p.v->front_pad && !p.v->back_pad)
3136                 return 0;
3137
3138         u = bch2_trans_kmalloc(trans, sizeof(*u));
3139         int ret = PTR_ERR_OR_ZERO(u);
3140         if (ret)
3141                 return ret;
3142
3143         bkey_reassemble(&u->k_i, k);
3144         u->v.front_pad  = 0;
3145         u->v.back_pad   = 0;
3146
3147         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_norun);
3148 }
3149
3150 int bch2_fix_reflink_p(struct bch_fs *c)
3151 {
3152         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
3153                 return 0;
3154
3155         int ret = bch2_trans_run(c,
3156                 for_each_btree_key_commit(trans, iter,
3157                                 BTREE_ID_extents, POS_MIN,
3158                                 BTREE_ITER_intent|BTREE_ITER_prefetch|
3159                                 BTREE_ITER_all_snapshots, k,
3160                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
3161                         fix_reflink_p_key(trans, &iter, k)));
3162         bch_err_fn(c, ret);
3163         return ret;
3164 }
3165
3166 #ifndef NO_BCACHEFS_CHARDEV
3167
3168 struct fsck_thread {
3169         struct thread_with_stdio thr;
3170         struct bch_fs           *c;
3171         struct bch_opts         opts;
3172 };
3173
3174 static void bch2_fsck_thread_exit(struct thread_with_stdio *_thr)
3175 {
3176         struct fsck_thread *thr = container_of(_thr, struct fsck_thread, thr);
3177         kfree(thr);
3178 }
3179
3180 static int bch2_fsck_offline_thread_fn(struct thread_with_stdio *stdio)
3181 {
3182         struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
3183         struct bch_fs *c = thr->c;
3184
3185         int ret = PTR_ERR_OR_ZERO(c);
3186         if (ret)
3187                 return ret;
3188
3189         ret = bch2_fs_start(thr->c);
3190         if (ret)
3191                 goto err;
3192
3193         if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
3194                 bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: errors fixed\n", c->name);
3195                 ret |= 1;
3196         }
3197         if (test_bit(BCH_FS_error, &c->flags)) {
3198                 bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: still has errors\n", c->name);
3199                 ret |= 4;
3200         }
3201 err:
3202         bch2_fs_stop(c);
3203         return ret;
3204 }
3205
3206 static const struct thread_with_stdio_ops bch2_offline_fsck_ops = {
3207         .exit           = bch2_fsck_thread_exit,
3208         .fn             = bch2_fsck_offline_thread_fn,
3209 };
3210
3211 long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_arg)
3212 {
3213         struct bch_ioctl_fsck_offline arg;
3214         struct fsck_thread *thr = NULL;
3215         darray_const_str devs = {};
3216         long ret = 0;
3217
3218         if (copy_from_user(&arg, user_arg, sizeof(arg)))
3219                 return -EFAULT;
3220
3221         if (arg.flags)
3222                 return -EINVAL;
3223
3224         if (!capable(CAP_SYS_ADMIN))
3225                 return -EPERM;
3226
3227         for (size_t i = 0; i < arg.nr_devs; i++) {
3228                 u64 dev_u64;
3229                 ret = copy_from_user_errcode(&dev_u64, &user_arg->devs[i], sizeof(u64));
3230                 if (ret)
3231                         goto err;
3232
3233                 char *dev_str = strndup_user((char __user *)(unsigned long) dev_u64, PATH_MAX);
3234                 ret = PTR_ERR_OR_ZERO(dev_str);
3235                 if (ret)
3236                         goto err;
3237
3238                 ret = darray_push(&devs, dev_str);
3239                 if (ret) {
3240                         kfree(dev_str);
3241                         goto err;
3242                 }
3243         }
3244
3245         thr = kzalloc(sizeof(*thr), GFP_KERNEL);
3246         if (!thr) {
3247                 ret = -ENOMEM;
3248                 goto err;
3249         }
3250
3251         thr->opts = bch2_opts_empty();
3252
3253         if (arg.opts) {
3254                 char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
3255                 ret =   PTR_ERR_OR_ZERO(optstr) ?:
3256                         bch2_parse_mount_opts(NULL, &thr->opts, NULL, optstr, false);
3257                 if (!IS_ERR(optstr))
3258                         kfree(optstr);
3259
3260                 if (ret)
3261                         goto err;
3262         }
3263
3264         opt_set(thr->opts, stdio, (u64)(unsigned long)&thr->thr.stdio);
3265         opt_set(thr->opts, read_only, 1);
3266         opt_set(thr->opts, ratelimit_errors, 0);
3267
3268         /* We need request_key() to be called before we punt to kthread: */
3269         opt_set(thr->opts, nostart, true);
3270
3271         bch2_thread_with_stdio_init(&thr->thr, &bch2_offline_fsck_ops);
3272
3273         thr->c = bch2_fs_open(&devs, &thr->opts);
3274
3275         if (!IS_ERR(thr->c) &&
3276             thr->c->opts.errors == BCH_ON_ERROR_panic)
3277                 thr->c->opts.errors = BCH_ON_ERROR_ro;
3278
3279         ret = __bch2_run_thread_with_stdio(&thr->thr);
3280 out:
3281         darray_for_each(devs, i)
3282                 kfree(*i);
3283         darray_exit(&devs);
3284         return ret;
3285 err:
3286         if (thr)
3287                 bch2_fsck_thread_exit(&thr->thr);
3288         pr_err("ret %s", bch2_err_str(ret));
3289         goto out;
3290 }
3291
3292 static int bch2_fsck_online_thread_fn(struct thread_with_stdio *stdio)
3293 {
3294         struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
3295         struct bch_fs *c = thr->c;
3296
3297         c->stdio_filter = current;
3298         c->stdio = &thr->thr.stdio;
3299
3300         /*
3301          * XXX: can we figure out a way to do this without mucking with c->opts?
3302          */
3303         unsigned old_fix_errors = c->opts.fix_errors;
3304         if (opt_defined(thr->opts, fix_errors))
3305                 c->opts.fix_errors = thr->opts.fix_errors;
3306         else
3307                 c->opts.fix_errors = FSCK_FIX_ask;
3308
3309         c->opts.fsck = true;
3310         set_bit(BCH_FS_in_fsck, &c->flags);
3311
3312         int ret = bch2_run_online_recovery_passes(c, ~0ULL);
3313
3314         clear_bit(BCH_FS_in_fsck, &c->flags);
3315         bch_err_fn(c, ret);
3316
3317         c->stdio = NULL;
3318         c->stdio_filter = NULL;
3319         c->opts.fix_errors = old_fix_errors;
3320
3321         up(&c->recovery.run_lock);
3322         bch2_ro_ref_put(c);
3323         return ret;
3324 }
3325
3326 static const struct thread_with_stdio_ops bch2_online_fsck_ops = {
3327         .exit           = bch2_fsck_thread_exit,
3328         .fn             = bch2_fsck_online_thread_fn,
3329 };
3330
3331 long bch2_ioctl_fsck_online(struct bch_fs *c, struct bch_ioctl_fsck_online arg)
3332 {
3333         struct fsck_thread *thr = NULL;
3334         long ret = 0;
3335
3336         if (arg.flags)
3337                 return -EINVAL;
3338
3339         if (!capable(CAP_SYS_ADMIN))
3340                 return -EPERM;
3341
3342         if (!bch2_ro_ref_tryget(c))
3343                 return -EROFS;
3344
3345         if (down_trylock(&c->recovery.run_lock)) {
3346                 bch2_ro_ref_put(c);
3347                 return -EAGAIN;
3348         }
3349
3350         thr = kzalloc(sizeof(*thr), GFP_KERNEL);
3351         if (!thr) {
3352                 ret = -ENOMEM;
3353                 goto err;
3354         }
3355
3356         thr->c = c;
3357         thr->opts = bch2_opts_empty();
3358
3359         if (arg.opts) {
3360                 char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
3361
3362                 ret =   PTR_ERR_OR_ZERO(optstr) ?:
3363                         bch2_parse_mount_opts(c, &thr->opts, NULL, optstr, false);
3364                 if (!IS_ERR(optstr))
3365                         kfree(optstr);
3366
3367                 if (ret)
3368                         goto err;
3369         }
3370
3371         ret = bch2_run_thread_with_stdio(&thr->thr, &bch2_online_fsck_ops);
3372 err:
3373         if (ret < 0) {
3374                 bch_err_fn(c, ret);
3375                 if (thr)
3376                         bch2_fsck_thread_exit(&thr->thr);
3377                 up(&c->recovery.run_lock);
3378                 bch2_ro_ref_put(c);
3379         }
3380         return ret;
3381 }
3382
3383 #endif /* NO_BCACHEFS_CHARDEV */