]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/fsck.c
Move c_src dirs back to toplevel
[bcachefs-tools-debian] / libbcachefs / fsck.c
index 4e71005777348667bd2875031b07f279e3a618eb..4f0ecd60567570b7364cef517225ea0e3dfa5575 100644 (file)
@@ -2,6 +2,7 @@
 
 #include "bcachefs.h"
 #include "bkey_buf.h"
+#include "btree_cache.h"
 #include "btree_update.h"
 #include "buckets.h"
 #include "darray.h"
 #include "fsck.h"
 #include "inode.h"
 #include "keylist.h"
-#include "subvolume.h"
+#include "recovery.h"
+#include "snapshot.h"
 #include "super.h"
 #include "xattr.h"
 
 #include <linux/bsearch.h>
 #include <linux/dcache.h> /* struct qstr */
 
-#define QSTR(n) { { { .len = strlen(n) } }, .name = n }
-
 /*
  * XXX: this is handling transaction restarts without returning
  * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
                                    u32 snapshot)
 {
-       struct btree_iter iter;
-       struct bkey_s_c k;
        u64 sectors = 0;
-       int ret;
 
-       for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
+       int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
                                SPOS(inum, 0, snapshot),
                                POS(inum, U64_MAX),
-                               0, k, ret)
+                               0, k, ({
                if (bkey_extent_is_allocation(k.k))
                        sectors += k.k->size;
-
-       bch2_trans_iter_exit(trans, &iter);
+               0;
+       }));
 
        return ret ?: sectors;
 }
@@ -47,57 +44,23 @@ static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
                                    u32 snapshot)
 {
-       struct btree_iter iter;
-       struct bkey_s_c k;
-       struct bkey_s_c_dirent d;
        u64 subdirs = 0;
-       int ret;
 
-       for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
-                               SPOS(inum, 0, snapshot),
-                               POS(inum, U64_MAX),
-                               0, k, ret) {
-               if (k.k->type != KEY_TYPE_dirent)
-                       continue;
-
-               d = bkey_s_c_to_dirent(k);
-               if (d.v->d_type == DT_DIR)
+       int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
+                                   SPOS(inum, 0, snapshot),
+                                   POS(inum, U64_MAX),
+                                   0, k, ({
+               if (k.k->type == KEY_TYPE_dirent &&
+                   bkey_s_c_to_dirent(k).v->d_type == DT_DIR)
                        subdirs++;
-       }
-       bch2_trans_iter_exit(trans, &iter);
+               0;
+       }));
 
        return ret ?: subdirs;
 }
 
-static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
-                                   u32 *subvol)
-{
-       struct btree_iter iter;
-       struct bkey_s_c k;
-       int ret;
-
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
-                            POS(0, snapshot), 0);
-       k = bch2_btree_iter_peek_slot(&iter);
-       ret = bkey_err(k);
-       if (ret)
-               goto err;
-
-       if (k.k->type != KEY_TYPE_snapshot) {
-               bch_err(trans->c, "snapshot %u not fonud", snapshot);
-               ret = -ENOENT;
-               goto err;
-       }
-
-       *subvol = le32_to_cpu(bkey_s_c_to_snapshot(k).v->subvol);
-err:
-       bch2_trans_iter_exit(trans, &iter);
-       return ret;
-
-}
-
-static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
-                          u32 *snapshot, u64 *inum)
+static int subvol_lookup(struct btree_trans *trans, u32 subvol,
+                        u32 *snapshot, u64 *inum)
 {
        struct bch_subvolume s;
        int ret;
@@ -109,12 +72,6 @@ static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
        return ret;
 }
 
-static int subvol_lookup(struct btree_trans *trans, u32 subvol,
-                        u32 *snapshot, u64 *inum)
-{
-       return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
-}
-
 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
                              struct bch_inode_unpacked *inode)
 {
@@ -131,20 +88,18 @@ static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
                goto err;
 
        if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
-               ret = -ENOENT;
+               ret = -BCH_ERR_ENOENT_inode;
                goto err;
        }
 
        ret = bch2_inode_unpack(k, inode);
 err:
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(trans->c, "error fetching inode %llu: %s",
-                       inode_nr, bch2_err_str(ret));
+       bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
        bch2_trans_iter_exit(trans, &iter);
        return ret;
 }
 
-static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
+static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
                          struct bch_inode_unpacked *inode,
                          u32 *snapshot)
 {
@@ -152,33 +107,23 @@ static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
        struct bkey_s_c k;
        int ret;
 
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
-                            SPOS(0, inode_nr, *snapshot), 0);
-       k = bch2_btree_iter_peek_slot(&iter);
+       k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
+                              SPOS(0, inode_nr, *snapshot), 0);
        ret = bkey_err(k);
        if (ret)
                goto err;
 
        ret = bkey_is_inode(k.k)
                ? bch2_inode_unpack(k, inode)
-               : -ENOENT;
+               : -BCH_ERR_ENOENT_inode;
        if (!ret)
                *snapshot = iter.pos.snapshot;
 err:
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(trans->c, "error fetching inode %llu:%u: %s",
-                       inode_nr, *snapshot, bch2_err_str(ret));
+       bch_err_msg(trans->c, ret, "fetching inode %llu:%u", inode_nr, *snapshot);
        bch2_trans_iter_exit(trans, &iter);
        return ret;
 }
 
-static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
-                       struct bch_inode_unpacked *inode,
-                       u32 *snapshot)
-{
-       return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
-}
-
 static int __lookup_dirent(struct btree_trans *trans,
                           struct bch_hash_info hash_info,
                           subvol_inum dir, struct qstr *name,
@@ -204,96 +149,28 @@ static int __write_inode(struct btree_trans *trans,
                         struct bch_inode_unpacked *inode,
                         u32 snapshot)
 {
-       struct btree_iter iter;
-       int ret;
+       struct bkey_inode_buf *inode_p =
+               bch2_trans_kmalloc(trans, sizeof(*inode_p));
 
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
-                           SPOS(0, inode->bi_inum, snapshot),
-                           BTREE_ITER_INTENT);
+       if (IS_ERR(inode_p))
+               return PTR_ERR(inode_p);
 
-       ret   = bch2_btree_iter_traverse(&iter) ?:
-               bch2_inode_write(trans, &iter, inode);
-       bch2_trans_iter_exit(trans, &iter);
-       return ret;
-}
+       bch2_inode_pack(inode_p, inode);
+       inode_p->inode.k.p.snapshot = snapshot;
 
-static int write_inode(struct btree_trans *trans,
-                      struct bch_inode_unpacked *inode,
-                      u32 snapshot)
-{
-       int ret = commit_do(trans, NULL, NULL,
-                                 BTREE_INSERT_NOFAIL|
-                                 BTREE_INSERT_LAZY_RW,
-                                 __write_inode(trans, inode, snapshot));
-       if (ret)
-               bch_err(trans->c, "error in fsck: error updating inode: %s",
-                       bch2_err_str(ret));
-       return ret;
+       return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
+                               &inode_p->inode.k_i,
+                               BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 }
 
-static int fsck_inode_rm(struct btree_trans *trans, u64 inum, u32 snapshot)
+static int fsck_write_inode(struct btree_trans *trans,
+                           struct bch_inode_unpacked *inode,
+                           u32 snapshot)
 {
-       struct bch_fs *c = trans->c;
-       struct btree_iter iter = { NULL };
-       struct bkey_i_inode_generation delete;
-       struct bch_inode_unpacked inode_u;
-       struct bkey_s_c k;
-       int ret;
-
-       do {
-               ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
-                                                     SPOS(inum, 0, snapshot),
-                                                     SPOS(inum, U64_MAX, snapshot),
-                                                     0, NULL) ?:
-                       bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
-                                                     SPOS(inum, 0, snapshot),
-                                                     SPOS(inum, U64_MAX, snapshot),
-                                                     0, NULL) ?:
-                       bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
-                                                     SPOS(inum, 0, snapshot),
-                                                     SPOS(inum, U64_MAX, snapshot),
-                                                     0, NULL);
-       } while (ret == -BCH_ERR_transaction_restart_nested);
-       if (ret)
-               goto err;
-retry:
-       bch2_trans_begin(trans);
-
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
-                            SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
-       k = bch2_btree_iter_peek_slot(&iter);
-
-       ret = bkey_err(k);
-       if (ret)
-               goto err;
-
-       if (!bkey_is_inode(k.k)) {
-               bch2_fs_inconsistent(c,
-                                    "inode %llu:%u not found when deleting",
-                                    inum, snapshot);
-               ret = -EIO;
-               goto err;
-       }
-
-       bch2_inode_unpack(k, &inode_u);
-
-       /* Subvolume root? */
-       if (inode_u.bi_subvol)
-               bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
-
-       bkey_inode_generation_init(&delete.k_i);
-       delete.k.p = iter.pos;
-       delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
-
-       ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
-               bch2_trans_commit(trans, NULL, NULL,
-                               BTREE_INSERT_NOFAIL);
-err:
-       bch2_trans_iter_exit(trans, &iter);
-       if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               goto retry;
-
-       return ret ?: -BCH_ERR_transaction_restart_nested;
+       int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+                           __write_inode(trans, inode, snapshot));
+       bch_err_fn(trans->c, ret);
+       return ret;
 }
 
 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
@@ -317,71 +194,106 @@ static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
                                  BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
        bch2_trans_iter_exit(trans, &iter);
 err:
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
 }
 
 /* Get lost+found, create if it doesn't exist: */
-static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
+static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
                            struct bch_inode_unpacked *lostfound)
 {
        struct bch_fs *c = trans->c;
-       struct bch_inode_unpacked root;
-       struct bch_hash_info root_hash_info;
        struct qstr lostfound_str = QSTR("lost+found");
-       subvol_inum root_inum = { .subvol = subvol };
        u64 inum = 0;
        unsigned d_type = 0;
-       u32 snapshot;
        int ret;
 
-       ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
+       struct bch_snapshot_tree st;
+       ret = bch2_snapshot_tree_lookup(trans,
+                       bch2_snapshot_tree(c, snapshot), &st);
        if (ret)
                return ret;
 
-       ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
+       subvol_inum root_inum = { .subvol = le32_to_cpu(st.master_subvol) };
+       u32 subvol_snapshot;
+
+       ret = subvol_lookup(trans, le32_to_cpu(st.master_subvol),
+                           &subvol_snapshot, &root_inum.inum);
+       bch_err_msg(c, ret, "looking up root subvol");
+       if (ret)
+               return ret;
+
+       struct bch_inode_unpacked root_inode;
+       struct bch_hash_info root_hash_info;
+       ret = lookup_inode(trans, root_inum.inum, &root_inode, &snapshot);
+       bch_err_msg(c, ret, "looking up root inode");
        if (ret)
                return ret;
 
-       root_hash_info = bch2_hash_info_init(c, &root);
+       root_hash_info = bch2_hash_info_init(c, &root_inode);
 
        ret = __lookup_dirent(trans, root_hash_info, root_inum,
-                           &lostfound_str, &inum, &d_type);
-       if (ret == -ENOENT) {
-               bch_notice(c, "creating lost+found");
+                             &lostfound_str, &inum, &d_type);
+       if (bch2_err_matches(ret, ENOENT))
                goto create_lostfound;
-       }
 
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(c, "error looking up lost+found: %s", bch2_err_str(ret));
+       bch_err_fn(c, ret);
        if (ret)
                return ret;
 
        if (d_type != DT_DIR) {
                bch_err(c, "error looking up lost+found: not a directory");
-               return ret;
+               return -BCH_ERR_ENOENT_not_directory;
        }
 
        /*
-        * The check_dirents pass has already run, dangling dirents
+        * The bch2_check_dirents pass has already run, dangling dirents
         * shouldn't exist here:
         */
-       return __lookup_inode(trans, inum, lostfound, &snapshot);
+       return lookup_inode(trans, inum, lostfound, &snapshot);
 
 create_lostfound:
+       /*
+        * XXX: we could have a nicer log message here  if we had a nice way to
+        * walk backpointers to print a path
+        */
+       bch_notice(c, "creating lost+found in snapshot %u", le32_to_cpu(st.root_snapshot));
+
+       u64 now = bch2_current_time(c);
+       struct btree_iter lostfound_iter = { NULL };
+       u64 cpu = raw_smp_processor_id();
+
        bch2_inode_init_early(c, lostfound);
+       bch2_inode_init_late(lostfound, now, 0, 0, S_IFDIR|0700, 0, &root_inode);
+       lostfound->bi_dir = root_inode.bi_inum;
+
+       root_inode.bi_nlink++;
+
+       ret = bch2_inode_create(trans, &lostfound_iter, lostfound, snapshot, cpu);
+       if (ret)
+               goto err;
 
-       ret = bch2_create_trans(trans, root_inum, &root,
-                               lostfound, &lostfound_str,
-                               0, 0, S_IFDIR|0700, 0, NULL, NULL,
-                               (subvol_inum) { }, 0);
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(c, "error creating lost+found: %s", bch2_err_str(ret));
+       bch2_btree_iter_set_snapshot(&lostfound_iter, snapshot);
+       ret = bch2_btree_iter_traverse(&lostfound_iter);
+       if (ret)
+               goto err;
+
+       ret =   bch2_dirent_create_snapshot(trans,
+                               root_inode.bi_inum, snapshot, &root_hash_info,
+                               mode_to_type(lostfound->bi_mode),
+                               &lostfound_str,
+                               lostfound->bi_inum,
+                               &lostfound->bi_dir_offset,
+                               BCH_HASH_SET_MUST_CREATE) ?:
+               bch2_inode_write_flags(trans, &lostfound_iter, lostfound,
+                                      BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+err:
+       bch_err_msg(c, ret, "creating lost+found");
+       bch2_trans_iter_exit(trans, &lostfound_iter);
        return ret;
 }
 
-static int __reattach_inode(struct btree_trans *trans,
+static int reattach_inode(struct btree_trans *trans,
                          struct bch_inode_unpacked *inode,
                          u32 inode_snapshot)
 {
@@ -390,14 +302,9 @@ static int __reattach_inode(struct btree_trans *trans,
        char name_buf[20];
        struct qstr name;
        u64 dir_offset = 0;
-       u32 subvol;
        int ret;
 
-       ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
-       if (ret)
-               return ret;
-
-       ret = lookup_lostfound(trans, subvol, &lostfound);
+       ret = lookup_lostfound(trans, inode_snapshot, &lostfound);
        if (ret)
                return ret;
 
@@ -414,15 +321,12 @@ static int __reattach_inode(struct btree_trans *trans,
        snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
        name = (struct qstr) QSTR(name_buf);
 
-       ret = bch2_dirent_create(trans,
-                                (subvol_inum) {
-                                       .subvol = subvol,
-                                       .inum = lostfound.bi_inum,
-                                },
-                                &dir_hash,
-                                inode_d_type(inode),
-                                &name, inode->bi_inum, &dir_offset,
-                                BCH_HASH_SET_MUST_CREATE);
+       ret = bch2_dirent_create_snapshot(trans,
+                               lostfound.bi_inum, inode_snapshot,
+                               &dir_hash,
+                               inode_d_type(inode),
+                               &name, inode->bi_inum, &dir_offset,
+                               BCH_HASH_SET_MUST_CREATE);
        if (ret)
                return ret;
 
@@ -432,43 +336,18 @@ static int __reattach_inode(struct btree_trans *trans,
        return __write_inode(trans, inode, inode_snapshot);
 }
 
-static int reattach_inode(struct btree_trans *trans,
-                         struct bch_inode_unpacked *inode,
-                         u32 inode_snapshot)
-{
-       int ret = commit_do(trans, NULL, NULL,
-                                 BTREE_INSERT_LAZY_RW|
-                                 BTREE_INSERT_NOFAIL,
-                       __reattach_inode(trans, inode, inode_snapshot));
-       if (ret) {
-               bch_err(trans->c, "error reattaching inode %llu: %s",
-                       inode->bi_inum, bch2_err_str(ret));
-               return ret;
-       }
-
-       return ret;
-}
-
 static int remove_backpointer(struct btree_trans *trans,
                              struct bch_inode_unpacked *inode)
 {
        struct btree_iter iter;
-       struct bkey_s_c k;
+       struct bkey_s_c_dirent d;
        int ret;
 
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents,
-                            POS(inode->bi_dir, inode->bi_dir_offset), 0);
-       k = bch2_btree_iter_peek_slot(&iter);
-       ret = bkey_err(k);
-       if (ret)
-               goto out;
-       if (k.k->type != KEY_TYPE_dirent) {
-               ret = -ENOENT;
-               goto out;
-       }
-
-       ret = __remove_dirent(trans, k.k->p);
-out:
+       d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
+                                    POS(inode->bi_dir, inode->bi_dir_offset), 0,
+                                    dirent);
+       ret =   bkey_err(d) ?:
+               __remove_dirent(trans, d.k->p);
        bch2_trans_iter_exit(trans, &iter);
        return ret;
 }
@@ -493,19 +372,19 @@ static inline void snapshots_seen_init(struct snapshots_seen *s)
        memset(s, 0, sizeof(*s));
 }
 
-static int snapshots_seen_add(struct bch_fs *c, struct snapshots_seen *s, u32 id)
+static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 {
-       struct snapshots_seen_entry *i, n = { id, id };
-       int ret;
+       struct snapshots_seen_entry *i, n = {
+               .id     = id,
+               .equiv  = bch2_snapshot_equiv(c, id),
+       };
+       int ret = 0;
 
-       darray_for_each(s->ids, i) {
-               if (n.equiv < i->equiv)
+       __darray_for_each(s->ids, i) {
+               if (i->id == id)
+                       return 0;
+               if (i->id > id)
                        break;
-
-               if (i->equiv == n.equiv) {
-                       bch_err(c, "%s(): adding duplicate snapshot", __func__);
-                       return -EINVAL;
-               }
        }
 
        ret = darray_insert_item(&s->ids, i - s->ids.data, n);
@@ -518,7 +397,7 @@ static int snapshots_seen_add(struct bch_fs *c, struct snapshots_seen *s, u32 id
 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
                                 enum btree_id btree_id, struct bpos pos)
 {
-       struct snapshots_seen_entry *i, n = {
+       struct snapshots_seen_entry n = {
                .id     = pos.snapshot,
                .equiv  = bch2_snapshot_equiv(c, pos.snapshot),
        };
@@ -527,27 +406,32 @@ static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
        if (!bkey_eq(s->pos, pos))
                s->ids.nr = 0;
 
-       pos.snapshot = n.equiv;
        s->pos = pos;
+       s->pos.snapshot = n.equiv;
 
-       darray_for_each(s->ids, i)
-               if (i->equiv == n.equiv) {
-                       if (fsck_err_on(i->id != n.id, c,
-                                       "snapshot deletion did not run correctly:\n"
-                                       "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
-                                       bch2_btree_ids[btree_id],
-                                       pos.inode, pos.offset,
-                                       i->id, n.id, n.equiv))
-                               return -BCH_ERR_need_snapshot_cleanup;
-
+       darray_for_each(s->ids, i) {
+               if (i->id == n.id)
                        return 0;
+
+               /*
+                * We currently don't rigorously track for snapshot cleanup
+                * needing to be run, so it shouldn't be a fsck error yet:
+                */
+               if (i->equiv == n.equiv) {
+                       bch_err(c, "snapshot deletion did not finish:\n"
+                               "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
+                               bch2_btree_id_str(btree_id),
+                               pos.inode, pos.offset,
+                               i->id, n.id, n.equiv);
+                       set_bit(BCH_FS_need_delete_dead_snapshots, &c->flags);
+                       return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
                }
+       }
 
        ret = darray_push(&s->ids, n);
        if (ret)
                bch_err(c, "error reallocating snapshots_seen table (size %zu)",
                        s->ids.size);
-fsck_err:
        return ret;
 }
 
@@ -555,21 +439,25 @@ fsck_err:
  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
  * and @ancestor hasn't been overwritten in @seen
  *
- * That is, returns whether key in @ancestor snapshot is visible in @id snapshot
+ * @c:         filesystem handle
+ * @seen:      list of snapshot ids already seen at current position
+ * @id:                descendent snapshot id
+ * @ancestor:  ancestor snapshot id
+ *
+ * Returns:    whether key in @ancestor snapshot is visible in @id snapshot
  */
 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
                                    u32 id, u32 ancestor)
 {
        ssize_t i;
-       u32 top = seen->ids.nr ? seen->ids.data[seen->ids.nr - 1].equiv : 0;
 
-       BUG_ON(id > ancestor);
-       BUG_ON(!bch2_snapshot_is_equiv(c, id));
-       BUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
+       EBUG_ON(id > ancestor);
+       EBUG_ON(!bch2_snapshot_is_equiv(c, id));
+       EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
 
        /* @ancestor should be the snapshot most recently added to @seen */
-       BUG_ON(ancestor != seen->pos.snapshot);
-       BUG_ON(ancestor != top);
+       EBUG_ON(ancestor != seen->pos.snapshot);
+       EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
 
        if (id == ancestor)
                return true;
@@ -577,11 +465,20 @@ static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *see
        if (!bch2_snapshot_is_ancestor(c, id, ancestor))
                return false;
 
+       /*
+        * We know that @id is a descendant of @ancestor, we're checking if
+        * we've seen a key that overwrote @ancestor - i.e. also a descendent of
+        * @ascestor and with @id as a descendent.
+        *
+        * But we already know that we're scanning IDs between @id and @ancestor
+        * numerically, since snapshot ID lists are kept sorted, so if we find
+        * an id that's an ancestor of @id we're done:
+        */
+
        for (i = seen->ids.nr - 2;
             i >= 0 && seen->ids.data[i].equiv >= id;
             --i)
-               if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv) &&
-                   bch2_snapshot_is_ancestor(c, seen->ids.data[i].equiv, ancestor))
+               if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
                        return false;
 
        return true;
@@ -592,14 +489,16 @@ static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *see
  * snapshot id @dst, test whether there is some snapshot in which @dst is
  * visible.
  *
- * This assumes we're visiting @src keys in natural key order.
+ * @c:         filesystem handle
+ * @s:         list of snapshot IDs already seen at @src
+ * @src:       snapshot ID of src key
+ * @dst:       snapshot ID of dst key
+ * Returns:    true if there is some snapshot in which @dst is visible
  *
- * @s  - list of snapshot IDs already seen at @src
- * @src        - snapshot ID of src key
- * @dst        - snapshot ID of dst key
+ * Assumes we're visiting @src keys in natural key order
  */
-static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
-                      u32 src, u32 dst)
+static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
+                       u32 src, u32 dst)
 {
        return dst <= src
                ? key_visible_in_snapshot(c, s, dst, src)
@@ -628,12 +527,14 @@ static int ref_visible2(struct bch_fs *c,
 struct inode_walker_entry {
        struct bch_inode_unpacked inode;
        u32                     snapshot;
+       bool                    seen_this_pos;
        u64                     count;
 };
 
 struct inode_walker {
        bool                            first_this_inode;
-       u64                             cur_inum;
+       bool                            recalculate_sums;
+       struct bpos                     last_pos;
 
        DARRAY(struct inode_walker_entry) inodes;
 };
@@ -661,26 +562,20 @@ static int add_inode(struct bch_fs *c, struct inode_walker *w,
        }));
 }
 
-static int __walk_inode(struct btree_trans *trans,
-                       struct inode_walker *w, struct bpos pos)
+static int get_inodes_all_snapshots(struct btree_trans *trans,
+                                   struct inode_walker *w, u64 inum)
 {
        struct bch_fs *c = trans->c;
        struct btree_iter iter;
        struct bkey_s_c k;
-       u32 restart_count = trans->restart_count;
-       unsigned i;
        int ret;
 
-       pos.snapshot = bch2_snapshot_equiv(c, pos.snapshot);
-
-       if (pos.inode == w->cur_inum)
-               goto lookup_snapshot;
-
+       w->recalculate_sums = false;
        w->inodes.nr = 0;
 
-       for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, pos.inode),
-                          BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-               if (k.k->p.offset != pos.inode)
+       for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
+                                    BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
+               if (k.k->p.offset != inum)
                        break;
 
                if (bkey_is_inode(k.k))
@@ -691,40 +586,69 @@ static int __walk_inode(struct btree_trans *trans,
        if (ret)
                return ret;
 
-       w->cur_inum             = pos.inode;
-       w->first_this_inode     = true;
+       w->first_this_inode = true;
+       return 0;
+}
 
-       if (trans_was_restarted(trans, restart_count))
-               return -BCH_ERR_transaction_restart_nested;
+static struct inode_walker_entry *
+lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
+                         u32 snapshot, bool is_whiteout)
+{
+       struct inode_walker_entry *i;
+
+       snapshot = bch2_snapshot_equiv(c, snapshot);
 
-lookup_snapshot:
-       for (i = 0; i < w->inodes.nr; i++)
-               if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->inodes.data[i].snapshot))
+       __darray_for_each(w->inodes, i)
+               if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
                        goto found;
-       return INT_MAX;
+
+       return NULL;
 found:
-       BUG_ON(pos.snapshot > w->inodes.data[i].snapshot);
+       BUG_ON(snapshot > i->snapshot);
 
-       if (pos.snapshot != w->inodes.data[i].snapshot) {
-               struct inode_walker_entry e = w->inodes.data[i];
+       if (snapshot != i->snapshot && !is_whiteout) {
+               struct inode_walker_entry new = *i;
+               size_t pos;
+               int ret;
 
-               e.snapshot = pos.snapshot;
-               e.count = 0;
+               new.snapshot = snapshot;
+               new.count = 0;
 
                bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
-                        pos.inode, pos.snapshot, w->inodes.data[i].snapshot);
+                        w->last_pos.inode, snapshot, i->snapshot);
 
-               while (i && w->inodes.data[i - 1].snapshot > pos.snapshot)
+               while (i > w->inodes.data && i[-1].snapshot > snapshot)
                        --i;
 
-               ret = darray_insert_item(&w->inodes, i, e);
+               pos = i - w->inodes.data;
+               ret = darray_insert_item(&w->inodes, pos, new);
                if (ret)
-                       return ret;
+                       return ERR_PTR(ret);
+
+               i = w->inodes.data + pos;
        }
 
        return i;
 }
 
+static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
+                                            struct inode_walker *w, struct bpos pos,
+                                            bool is_whiteout)
+{
+       if (w->last_pos.inode != pos.inode) {
+               int ret = get_inodes_all_snapshots(trans, w, pos.inode);
+               if (ret)
+                       return ERR_PTR(ret);
+       } else if (bkey_cmp(w->last_pos, pos)) {
+               darray_for_each(w->inodes, i)
+                       i->seen_this_pos = false;
+       }
+
+       w->last_pos = pos;
+
+       return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
+}
+
 static int __get_visible_inodes(struct btree_trans *trans,
                                struct inode_walker *w,
                                struct snapshots_seen *s,
@@ -767,8 +691,9 @@ static int check_key_has_snapshot(struct btree_trans *trans,
        int ret = 0;
 
        if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
-                       "key in missing snapshot: %s",
-                       (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+                               bkey_in_missing_snapshot,
+                               "key in missing snapshot: %s",
+                               (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
                ret = bch2_btree_delete_at(trans, iter,
                                            BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 fsck_err:
@@ -788,7 +713,7 @@ static int hash_redo_key(struct btree_trans *trans,
        if (IS_ERR(delete))
                return PTR_ERR(delete);
 
-       tmp = bch2_bkey_make_mut(trans, k);
+       tmp = bch2_bkey_make_mut_noupdate(trans, k);
        if (IS_ERR(tmp))
                return PTR_ERR(tmp);
 
@@ -801,9 +726,7 @@ static int hash_redo_key(struct btree_trans *trans,
                                       k.k->p.snapshot, tmp,
                                       BCH_HASH_SET_MUST_CREATE,
                                       BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
-               bch2_trans_commit(trans, NULL, NULL,
-                                 BTREE_INSERT_NOFAIL|
-                                 BTREE_INSERT_LAZY_RW);
+               bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 }
 
 static int hash_check_key(struct btree_trans *trans,
@@ -837,6 +760,7 @@ static int hash_check_key(struct btree_trans *trans,
 
                if (fsck_err_on(k.k->type == desc.key_type &&
                                !desc.cmp_bkey(k, hash_k), c,
+                               hash_table_key_duplicate,
                                "duplicate hash table keys:\n%s",
                                (printbuf_reset(&buf),
                                 bch2_bkey_val_to_text(&buf, c, hash_k),
@@ -855,13 +779,13 @@ out:
        printbuf_exit(&buf);
        return ret;
 bad_hash:
-       if (fsck_err(c, "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
-                    bch2_btree_ids[desc.btree_id], hash_k.k->p.inode, hash_k.k->p.offset, hash,
+       if (fsck_err(c, hash_table_key_wrong_offset,
+                    "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
+                    bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
                     (printbuf_reset(&buf),
                      bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
                ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
-               if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-                       bch_err(c, "hash_redo_key err %s", bch2_err_str(ret));
+               bch_err_fn(c, ret);
                if (ret)
                        return ret;
                ret = -BCH_ERR_transaction_restart_nested;
@@ -870,6 +794,18 @@ fsck_err:
        goto out;
 }
 
+static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
+{
+       struct btree_iter iter;
+       struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
+       int ret = bkey_err(k);
+       if (ret)
+               return ret;
+
+       bch2_trans_iter_exit(trans, &iter);
+       return k.k->type == KEY_TYPE_set;
+}
+
 static int check_inode(struct btree_trans *trans,
                       struct btree_iter *iter,
                       struct bkey_s_c k,
@@ -892,58 +828,76 @@ static int check_inode(struct btree_trans *trans,
        if (ret)
                goto err;
 
-       /*
-        * if snapshot id isn't a leaf node, skip it - deletion in
-        * particular is not atomic, so on the internal snapshot nodes
-        * we can see inodes marked for deletion after a clean shutdown
-        */
-       if (bch2_snapshot_internal_node(c, k.k->p.snapshot))
-               return 0;
-
        if (!bkey_is_inode(k.k))
                return 0;
 
        BUG_ON(bch2_inode_unpack(k, &u));
 
        if (!full &&
-           !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
-                           BCH_INODE_I_SECTORS_DIRTY|
-                           BCH_INODE_UNLINKED)))
+           !(u.bi_flags & (BCH_INODE_i_size_dirty|
+                           BCH_INODE_i_sectors_dirty|
+                           BCH_INODE_unlinked)))
                return 0;
 
        if (prev->bi_inum != u.bi_inum)
                *prev = u;
 
        if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
-                       inode_d_type(prev)      != inode_d_type(&u), c,
+                       inode_d_type(prev)      != inode_d_type(&u),
+                       c, inode_snapshot_mismatch,
                        "inodes in different snapshots don't match")) {
                bch_err(c, "repair not implemented yet");
-               return -EINVAL;
+               return -BCH_ERR_fsck_repair_unimplemented;
        }
 
-       if (u.bi_flags & BCH_INODE_UNLINKED &&
+       if ((u.bi_flags & (BCH_INODE_i_size_dirty|BCH_INODE_unlinked)) &&
+           bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
+               struct bpos new_min_pos;
+
+               ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
+               if (ret)
+                       goto err;
+
+               u.bi_flags &= ~BCH_INODE_i_size_dirty|BCH_INODE_unlinked;
+
+               ret = __write_inode(trans, &u, iter->pos.snapshot);
+               bch_err_msg(c, ret, "in fsck updating inode");
+               if (ret)
+                       return ret;
+
+               if (!bpos_eq(new_min_pos, POS_MIN))
+                       bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
+               return 0;
+       }
+
+       if (u.bi_flags & BCH_INODE_unlinked) {
+               ret = check_inode_deleted_list(trans, k.k->p);
+               if (ret < 0)
+                       return ret;
+
+               fsck_err_on(ret, c, unlinked_inode_not_on_deleted_list,
+                           "inode %llu:%u unlinked, but not on deleted list",
+                           u.bi_inum, k.k->p.snapshot);
+               ret = 0;
+       }
+
+       if (u.bi_flags & BCH_INODE_unlinked &&
            (!c->sb.clean ||
-            fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
+            fsck_err(c, inode_unlinked_but_clean,
+                     "filesystem marked clean, but inode %llu unlinked",
                      u.bi_inum))) {
-               bch2_trans_unlock(trans);
-               bch2_fs_lazy_rw(c);
-
-               ret = fsck_inode_rm(trans, u.bi_inum, iter->pos.snapshot);
-               if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-                       bch_err(c, "error in fsck: error while deleting inode: %s",
-                               bch2_err_str(ret));
+               ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
+               bch_err_msg(c, ret, "in fsck deleting inode");
                return ret;
        }
 
-       if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
+       if (u.bi_flags & BCH_INODE_i_size_dirty &&
            (!c->sb.clean ||
-            fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
+            fsck_err(c, inode_i_size_dirty_but_clean,
+                     "filesystem marked clean, but inode %llu has i_size dirty",
                      u.bi_inum))) {
                bch_verbose(c, "truncating inode %llu", u.bi_inum);
 
-               bch2_trans_unlock(trans);
-               bch2_fs_lazy_rw(c);
-
                /*
                 * XXX: need to truncate partial blocks too here - or ideally
                 * just switch units to bytes and that issue goes away
@@ -953,9 +907,7 @@ static int check_inode(struct btree_trans *trans,
                                     iter->pos.snapshot),
                                POS(u.bi_inum, U64_MAX),
                                0, NULL);
-               if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-                       bch_err(c, "error in fsck: error truncating inode: %s",
-                               bch2_err_str(ret));
+               bch_err_msg(c, ret, "in fsck truncating inode");
                if (ret)
                        return ret;
 
@@ -963,15 +915,16 @@ static int check_inode(struct btree_trans *trans,
                 * We truncated without our normal sector accounting hook, just
                 * make sure we recalculate it:
                 */
-               u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
+               u.bi_flags |= BCH_INODE_i_sectors_dirty;
 
-               u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
+               u.bi_flags &= ~BCH_INODE_i_size_dirty;
                do_update = true;
        }
 
-       if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
+       if (u.bi_flags & BCH_INODE_i_sectors_dirty &&
            (!c->sb.clean ||
-            fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
+            fsck_err(c, inode_i_sectors_dirty_but_clean,
+                     "filesystem marked clean, but inode %llu has i_sectors dirty",
                      u.bi_inum))) {
                s64 sectors;
 
@@ -980,121 +933,59 @@ static int check_inode(struct btree_trans *trans,
 
                sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
                if (sectors < 0) {
-                       bch_err(c, "error in fsck: error recounting inode sectors: %s",
-                               bch2_err_str(sectors));
+                       bch_err_msg(c, sectors, "in fsck recounting inode sectors");
                        return sectors;
                }
 
                u.bi_sectors = sectors;
-               u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
+               u.bi_flags &= ~BCH_INODE_i_sectors_dirty;
                do_update = true;
        }
 
-       if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
+       if (u.bi_flags & BCH_INODE_backptr_untrusted) {
                u.bi_dir = 0;
                u.bi_dir_offset = 0;
-               u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
+               u.bi_flags &= ~BCH_INODE_backptr_untrusted;
                do_update = true;
        }
 
        if (do_update) {
                ret = __write_inode(trans, &u, iter->pos.snapshot);
+               bch_err_msg(c, ret, "in fsck updating inode");
                if (ret)
-                       bch_err(c, "error in fsck: error updating inode: %s",
-                               bch2_err_str(ret));
+                       return ret;
        }
 err:
 fsck_err:
-       if (ret)
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
 }
 
-noinline_for_stack
-static int check_inodes(struct bch_fs *c, bool full)
+int bch2_check_inodes(struct bch_fs *c)
 {
-       struct btree_trans trans;
-       struct btree_iter iter;
+       bool full = c->opts.fsck;
        struct bch_inode_unpacked prev = { 0 };
        struct snapshots_seen s;
-       struct bkey_s_c k;
-       int ret;
 
        snapshots_seen_init(&s);
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 
-       ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes,
-                       POS_MIN,
-                       BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
-                       NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
-               check_inode(&trans, &iter, k, &prev, &s, full));
+       int ret = bch2_trans_run(c,
+               for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
+                               POS_MIN,
+                               BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
+                               NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+                       check_inode(trans, &iter, k, &prev, &s, full)));
 
-       bch2_trans_exit(&trans);
        snapshots_seen_exit(&s);
-       if (ret)
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
 }
 
-/*
- * Checking for overlapping extents needs to be reimplemented
- */
-#if 0
-static int fix_overlapping_extent(struct btree_trans *trans,
-                                      struct bkey_s_c k, struct bpos cut_at)
-{
-       struct btree_iter iter;
-       struct bkey_i *u;
-       int ret;
-
-       u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
-       ret = PTR_ERR_OR_ZERO(u);
-       if (ret)
-               return ret;
-
-       bkey_reassemble(u, k);
-       bch2_cut_front(cut_at, u);
-
-
-       /*
-        * We don't want to go through the extent_handle_overwrites path:
-        *
-        * XXX: this is going to screw up disk accounting, extent triggers
-        * assume things about extent overwrites - we should be running the
-        * triggers manually here
-        */
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, u->k.p,
-                            BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
-
-       BUG_ON(iter.flags & BTREE_ITER_IS_EXTENTS);
-       ret   = bch2_btree_iter_traverse(&iter) ?:
-               bch2_trans_update(trans, &iter, u, BTREE_TRIGGER_NORUN) ?:
-               bch2_trans_commit(trans, NULL, NULL,
-                                 BTREE_INSERT_NOFAIL|
-                                 BTREE_INSERT_LAZY_RW);
-       bch2_trans_iter_exit(trans, &iter);
-       return ret;
-}
-#endif
-
 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
                                                struct btree_iter *iter,
                                                struct bpos pos)
 {
-       struct bkey_s_c k;
-       int ret;
-
-       bch2_trans_iter_init(trans, iter, BTREE_ID_dirents, pos, 0);
-       k = bch2_btree_iter_peek_slot(iter);
-       ret = bkey_err(k);
-       if (!ret && k.k->type != KEY_TYPE_dirent)
-               ret = -ENOENT;
-       if (ret) {
-               bch2_trans_iter_exit(trans, iter);
-               return (struct bkey_s_c_dirent) { .k = ERR_PTR(ret) };
-       }
-
-       return bkey_s_c_to_dirent(k);
+       return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
 }
 
 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
@@ -1112,29 +1003,9 @@ static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
                : le64_to_cpu(d.v->d_inum)              == inode->bi_inum;
 }
 
-static int inode_backpointer_exists(struct btree_trans *trans,
-                                   struct bch_inode_unpacked *inode,
-                                   u32 snapshot)
-{
-       struct btree_iter iter;
-       struct bkey_s_c_dirent d;
-       int ret;
-
-       d = dirent_get_by_pos(trans, &iter,
-                       SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
-       ret = bkey_err(d.s_c);
-       if (ret)
-               return ret == -ENOENT ? 0 : ret;
-
-       ret = dirent_points_to_inode(d, inode);
-       bch2_trans_iter_exit(trans, &iter);
-       return ret;
-}
-
 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
 {
        struct bch_fs *c = trans->c;
-       struct inode_walker_entry *i;
        u32 restart_count = trans->restart_count;
        int ret = 0;
        s64 count2;
@@ -1143,32 +1014,31 @@ static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
                if (i->inode.bi_sectors == i->count)
                        continue;
 
-               count2 = bch2_count_inode_sectors(trans, w->cur_inum, i->snapshot);
+               count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
 
-               if (i->count != count2) {
-                       bch_err(c, "fsck counted i_sectors wrong: got %llu should be %llu",
-                               i->count, count2);
+               if (w->recalculate_sums)
                        i->count = count2;
-                       if (i->inode.bi_sectors == i->count)
-                               continue;
+
+               if (i->count != count2) {
+                       bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
+                               w->last_pos.inode, i->snapshot, i->count, count2);
+                       return -BCH_ERR_internal_fsck_err;
                }
 
-               if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
-                           "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
-                           w->cur_inum, i->snapshot,
-                           i->inode.bi_sectors, i->count)) {
+               if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
+                               c, inode_i_sectors_wrong,
+                               "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
+                               w->last_pos.inode, i->snapshot,
+                               i->inode.bi_sectors, i->count)) {
                        i->inode.bi_sectors = i->count;
-                       ret = write_inode(trans, &i->inode, i->snapshot);
+                       ret = fsck_write_inode(trans, &i->inode, i->snapshot);
                        if (ret)
                                break;
                }
        }
 fsck_err:
-       if (ret)
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
-       if (!ret && trans_was_restarted(trans, restart_count))
-               ret = -BCH_ERR_transaction_restart_nested;
-       return ret;
+       bch_err_fn(c, ret);
+       return ret ?: trans_was_restarted(trans, restart_count);
 }
 
 struct extent_end {
@@ -1177,23 +1047,196 @@ struct extent_end {
        struct snapshots_seen   seen;
 };
 
-typedef DARRAY(struct extent_end) extent_ends;
+struct extent_ends {
+       struct bpos                     last_pos;
+       DARRAY(struct extent_end)       e;
+};
+
+static void extent_ends_reset(struct extent_ends *extent_ends)
+{
+       darray_for_each(extent_ends->e, i)
+               snapshots_seen_exit(&i->seen);
+       extent_ends->e.nr = 0;
+}
+
+static void extent_ends_exit(struct extent_ends *extent_ends)
+{
+       extent_ends_reset(extent_ends);
+       darray_exit(&extent_ends->e);
+}
+
+static void extent_ends_init(struct extent_ends *extent_ends)
+{
+       memset(extent_ends, 0, sizeof(*extent_ends));
+}
+
+static int extent_ends_at(struct bch_fs *c,
+                         struct extent_ends *extent_ends,
+                         struct snapshots_seen *seen,
+                         struct bkey_s_c k)
+{
+       struct extent_end *i, n = (struct extent_end) {
+               .offset         = k.k->p.offset,
+               .snapshot       = k.k->p.snapshot,
+               .seen           = *seen,
+       };
+
+       n.seen.ids.data = kmemdup(seen->ids.data,
+                             sizeof(seen->ids.data[0]) * seen->ids.size,
+                             GFP_KERNEL);
+       if (!n.seen.ids.data)
+               return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
+
+       __darray_for_each(extent_ends->e, i) {
+               if (i->snapshot == k.k->p.snapshot) {
+                       snapshots_seen_exit(&i->seen);
+                       *i = n;
+                       return 0;
+               }
+
+               if (i->snapshot >= k.k->p.snapshot)
+                       break;
+       }
+
+       return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
+}
+
+static int overlapping_extents_found(struct btree_trans *trans,
+                                    enum btree_id btree,
+                                    struct bpos pos1, struct snapshots_seen *pos1_seen,
+                                    struct bkey pos2,
+                                    bool *fixed,
+                                    struct extent_end *extent_end)
+{
+       struct bch_fs *c = trans->c;
+       struct printbuf buf = PRINTBUF;
+       struct btree_iter iter1, iter2 = { NULL };
+       struct bkey_s_c k1, k2;
+       int ret;
+
+       BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
+
+       bch2_trans_iter_init(trans, &iter1, btree, pos1,
+                            BTREE_ITER_ALL_SNAPSHOTS|
+                            BTREE_ITER_NOT_EXTENTS);
+       k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
+       ret = bkey_err(k1);
+       if (ret)
+               goto err;
+
+       prt_str(&buf, "\n  ");
+       bch2_bkey_val_to_text(&buf, c, k1);
+
+       if (!bpos_eq(pos1, k1.k->p)) {
+               prt_str(&buf, "\n  wanted\n  ");
+               bch2_bpos_to_text(&buf, pos1);
+               prt_str(&buf, "\n  ");
+               bch2_bkey_to_text(&buf, &pos2);
+
+               bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
+                       __func__, buf.buf);
+               ret = -BCH_ERR_internal_fsck_err;
+               goto err;
+       }
+
+       bch2_trans_copy_iter(&iter2, &iter1);
+
+       while (1) {
+               bch2_btree_iter_advance(&iter2);
+
+               k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
+               ret = bkey_err(k2);
+               if (ret)
+                       goto err;
+
+               if (bpos_ge(k2.k->p, pos2.p))
+                       break;
+       }
+
+       prt_str(&buf, "\n  ");
+       bch2_bkey_val_to_text(&buf, c, k2);
+
+       if (bpos_gt(k2.k->p, pos2.p) ||
+           pos2.size != k2.k->size) {
+               bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
+                       __func__, buf.buf);
+               ret = -BCH_ERR_internal_fsck_err;
+               goto err;
+       }
+
+       prt_printf(&buf, "\n  overwriting %s extent",
+                  pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
+
+       if (fsck_err(c, extent_overlapping,
+                    "overlapping extents%s", buf.buf)) {
+               struct btree_iter *old_iter = &iter1;
+               struct disk_reservation res = { 0 };
+
+               if (pos1.snapshot < pos2.p.snapshot) {
+                       old_iter = &iter2;
+                       swap(k1, k2);
+               }
+
+               trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
+
+               ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
+                               BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
+                               k1, k2) ?:
+                       bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
+               bch2_disk_reservation_put(c, &res);
+
+               if (ret)
+                       goto err;
+
+               *fixed = true;
+
+               if (pos1.snapshot == pos2.p.snapshot) {
+                       /*
+                        * We overwrote the first extent, and did the overwrite
+                        * in the same snapshot:
+                        */
+                       extent_end->offset = bkey_start_offset(&pos2);
+               } else if (pos1.snapshot > pos2.p.snapshot) {
+                       /*
+                        * We overwrote the first extent in pos2's snapshot:
+                        */
+                       ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
+               } else {
+                       /*
+                        * We overwrote the second extent - restart
+                        * check_extent() from the top:
+                        */
+                       ret = -BCH_ERR_transaction_restart_nested;
+               }
+       }
+fsck_err:
+err:
+       bch2_trans_iter_exit(trans, &iter2);
+       bch2_trans_iter_exit(trans, &iter1);
+       printbuf_exit(&buf);
+       return ret;
+}
 
 static int check_overlapping_extents(struct btree_trans *trans,
                              struct snapshots_seen *seen,
-                             extent_ends *extent_ends,
+                             struct extent_ends *extent_ends,
                              struct bkey_s_c k,
-                             struct btree_iter *iter)
+                             u32 equiv,
+                             struct btree_iter *iter,
+                             bool *fixed)
 {
        struct bch_fs *c = trans->c;
-       struct extent_end *i;
-       struct printbuf buf = PRINTBUF;
        int ret = 0;
 
-       darray_for_each(*extent_ends, i) {
-               /* duplicate, due to transaction restart: */
-               if (i->offset   == k.k->p.offset &&
-                   i->snapshot == k.k->p.snapshot)
+       /* transaction restart, running again */
+       if (bpos_eq(extent_ends->last_pos, k.k->p))
+               return 0;
+
+       if (extent_ends->last_pos.inode != k.k->p.inode)
+               extent_ends_reset(extent_ends);
+
+       darray_for_each(extent_ends->e, i) {
+               if (i->offset <= bkey_start_offset(k.k))
                        continue;
 
                if (!ref_visible2(c,
@@ -1201,244 +1244,217 @@ static int check_overlapping_extents(struct btree_trans *trans,
                                  i->snapshot, &i->seen))
                        continue;
 
-               if (fsck_err_on(i->offset > bkey_start_offset(k.k), c,
-                               "overlapping extents: extent in snapshot %u ends at %llu overlaps with\n%s",
-                               i->snapshot,
-                               i->offset,
-                               (printbuf_reset(&buf),
-                                bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-                       struct bkey_i *update = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
-                       if ((ret = PTR_ERR_OR_ZERO(update)))
-                               goto err;
-                       bkey_reassemble(update, k);
-                       ret = bch2_trans_update_extent(trans, iter, update, 0);
-                       if (!ret)
-                               goto err;
-               }
+               ret = overlapping_extents_found(trans, iter->btree_id,
+                                               SPOS(iter->pos.inode,
+                                                    i->offset,
+                                                    i->snapshot),
+                                               &i->seen,
+                                               *k.k, fixed, i);
+               if (ret)
+                       goto err;
        }
+
+       ret = extent_ends_at(c, extent_ends, seen, k);
+       if (ret)
+               goto err;
+
+       extent_ends->last_pos = k.k->p;
 err:
-fsck_err:
-       printbuf_exit(&buf);
        return ret;
 }
 
-static int extent_ends_at(extent_ends *extent_ends,
-                         struct snapshots_seen *seen,
-                         struct bkey_s_c k)
+static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
+                               struct bkey_s_c k)
 {
-       struct extent_end *i, n = (struct extent_end) {
-               .snapshot       = k.k->p.snapshot,
-               .offset         = k.k->p.offset,
-               .seen           = *seen,
-       };
-
-       n.seen.ids.data = kmemdup(seen->ids.data,
-                             sizeof(seen->ids.data[0]) * seen->ids.size,
-                             GFP_KERNEL);
-       if (!n.seen.ids.data)
-               return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
-
-       darray_for_each(*extent_ends, i) {
-               if (i->snapshot == k.k->p.snapshot) {
-                       snapshots_seen_exit(&i->seen);
-                       *i = n;
-                       return 0;
+       struct bch_fs *c = trans->c;
+       struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+       struct bch_extent_crc_unpacked crc;
+       const union bch_extent_entry *i;
+       unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
+
+       bkey_for_each_crc(k.k, ptrs, crc, i)
+               if (crc_is_encoded(crc) &&
+                   crc.uncompressed_size > encoded_extent_max_sectors) {
+                       struct printbuf buf = PRINTBUF;
+
+                       bch2_bkey_val_to_text(&buf, c, k);
+                       bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
+                       printbuf_exit(&buf);
                }
 
-               if (i->snapshot >= k.k->p.snapshot)
-                       break;
-       }
-
-       return darray_insert_item(extent_ends, i - extent_ends->data, n);
-}
-
-static void extent_ends_reset(extent_ends *extent_ends)
-{
-       struct extent_end *i;
-
-       darray_for_each(*extent_ends, i)
-               snapshots_seen_exit(&i->seen);
-
-       extent_ends->nr = 0;
+       return 0;
 }
 
 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
                        struct bkey_s_c k,
                        struct inode_walker *inode,
                        struct snapshots_seen *s,
-                       extent_ends *extent_ends)
+                       struct extent_ends *extent_ends)
 {
        struct bch_fs *c = trans->c;
        struct inode_walker_entry *i;
        struct printbuf buf = PRINTBUF;
-       struct bpos equiv;
+       struct bpos equiv = k.k->p;
        int ret = 0;
 
+       equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
+
        ret = check_key_has_snapshot(trans, iter, k);
        if (ret) {
                ret = ret < 0 ? ret : 0;
                goto out;
        }
 
-       equiv = k.k->p;
-       equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
-
-       ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
-       if (ret)
-               goto err;
-
-       if (k.k->type == KEY_TYPE_whiteout)
-               goto out;
-
-       if (inode->cur_inum != k.k->p.inode) {
+       if (inode->last_pos.inode != k.k->p.inode) {
                ret = check_i_sectors(trans, inode);
                if (ret)
                        goto err;
-
-               extent_ends_reset(extent_ends);
        }
 
-       BUG_ON(!iter->path->should_be_locked);
-
-       ret = check_overlapping_extents(trans, s, extent_ends, k, iter);
+       i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
+       ret = PTR_ERR_OR_ZERO(i);
        if (ret)
                goto err;
 
-       ret = extent_ends_at(extent_ends, s, k);
+       ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
        if (ret)
                goto err;
 
-       ret = __walk_inode(trans, inode, equiv);
-       if (ret < 0)
-               goto err;
-
-       if (fsck_err_on(ret == INT_MAX, c,
-                       "extent in missing inode:\n  %s",
-                       (printbuf_reset(&buf),
-                        bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-               ret = bch2_btree_delete_at(trans, iter,
-                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
-               goto out;
-       }
-
-       if (ret == INT_MAX) {
-               ret = 0;
-               goto out;
-       }
-
-       i = inode->inodes.data + ret;
-       ret = 0;
+       if (k.k->type != KEY_TYPE_whiteout) {
+               if (fsck_err_on(!i, c, extent_in_missing_inode,
+                               "extent in missing inode:\n  %s",
+                               (printbuf_reset(&buf),
+                                bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+                       goto delete;
+
+               if (fsck_err_on(i &&
+                               !S_ISREG(i->inode.bi_mode) &&
+                               !S_ISLNK(i->inode.bi_mode),
+                               c, extent_in_non_reg_inode,
+                               "extent in non regular inode mode %o:\n  %s",
+                               i->inode.bi_mode,
+                               (printbuf_reset(&buf),
+                                bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+                       goto delete;
 
-       if (fsck_err_on(!S_ISREG(i->inode.bi_mode) &&
-                       !S_ISLNK(i->inode.bi_mode), c,
-                       "extent in non regular inode mode %o:\n  %s",
-                       i->inode.bi_mode,
-                       (printbuf_reset(&buf),
-                        bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-               ret = bch2_btree_delete_at(trans, iter,
-                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
-               goto out;
+               ret = check_overlapping_extents(trans, s, extent_ends, k,
+                                               equiv.snapshot, iter,
+                                               &inode->recalculate_sums);
+               if (ret)
+                       goto err;
        }
 
        /*
-        * Check inodes in reverse order, from oldest snapshots to newest, so
-        * that we emit the fewest number of whiteouts necessary:
+        * Check inodes in reverse order, from oldest snapshots to newest,
+        * starting from the inode that matches this extent's snapshot. If we
+        * didn't have one, iterate over all inodes:
         */
-       for (i = inode->inodes.data + inode->inodes.nr - 1;
-            i >= inode->inodes.data;
+       if (!i)
+               i = inode->inodes.data + inode->inodes.nr - 1;
+
+       for (;
+            inode->inodes.data && i >= inode->inodes.data;
             --i) {
                if (i->snapshot > equiv.snapshot ||
                    !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
                        continue;
 
-               if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
-                               k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
-                               !bkey_extent_is_reservation(k), c,
-                               "extent type past end of inode %llu:%u, i_size %llu\n  %s",
-                               i->inode.bi_inum, i->snapshot, i->inode.bi_size,
-                               (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-                       struct btree_iter iter2;
-
-                       bch2_trans_copy_iter(&iter2, iter);
-                       bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
-                       ret =   bch2_btree_iter_traverse(&iter2) ?:
-                               bch2_btree_delete_at(trans, &iter2,
-                                       BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
-                       bch2_trans_iter_exit(trans, &iter2);
-                       if (ret)
-                               goto err;
-
-                       if (i->snapshot != equiv.snapshot) {
-                               ret = snapshots_seen_add(c, s, i->snapshot);
+               if (k.k->type != KEY_TYPE_whiteout) {
+                       if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_size_dirty) &&
+                                       k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
+                                       !bkey_extent_is_reservation(k),
+                                       c, extent_past_end_of_inode,
+                                       "extent type past end of inode %llu:%u, i_size %llu\n  %s",
+                                       i->inode.bi_inum, i->snapshot, i->inode.bi_size,
+                                       (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+                               struct btree_iter iter2;
+
+                               bch2_trans_copy_iter(&iter2, iter);
+                               bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
+                               ret =   bch2_btree_iter_traverse(&iter2) ?:
+                                       bch2_btree_delete_at(trans, &iter2,
+                                               BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+                               bch2_trans_iter_exit(trans, &iter2);
                                if (ret)
                                        goto err;
+
+                               iter->k.type = KEY_TYPE_whiteout;
                        }
-               }
-       }
 
-       if (bkey_extent_is_allocation(k.k))
-               for_each_visible_inode(c, s, inode, equiv.snapshot, i)
-                       i->count += k.k->size;
-#if 0
-       bch2_bkey_buf_reassemble(&prev, c, k);
-#endif
+                       if (bkey_extent_is_allocation(k.k))
+                               i->count += k.k->size;
+               }
 
+               i->seen_this_pos = true;
+       }
 out:
 err:
 fsck_err:
        printbuf_exit(&buf);
-
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
+delete:
+       ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+       goto out;
 }
 
 /*
  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
  * that i_size an i_sectors are consistent
  */
-noinline_for_stack
-static int check_extents(struct bch_fs *c)
+int bch2_check_extents(struct bch_fs *c)
 {
        struct inode_walker w = inode_walker_init();
        struct snapshots_seen s;
-       struct btree_trans trans;
-       struct btree_iter iter;
-       struct bkey_s_c k;
-       extent_ends extent_ends = { 0 };
+       struct extent_ends extent_ends;
        struct disk_reservation res = { 0 };
-       int ret = 0;
 
        snapshots_seen_init(&s);
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
-       bch_verbose(c, "checking extents");
-
-       ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_extents,
-                       POS(BCACHEFS_ROOT_INO, 0),
-                       BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
-                       &res, NULL,
-                       BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({
-               bch2_disk_reservation_put(c, &res);
-               check_extent(&trans, &iter, k, &w, &s, &extent_ends);
-       }));
+       extent_ends_init(&extent_ends);
+
+       int ret = bch2_trans_run(c,
+               for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
+                               POS(BCACHEFS_ROOT_INO, 0),
+                               BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
+                               &res, NULL,
+                               BCH_TRANS_COMMIT_no_enospc, ({
+                       bch2_disk_reservation_put(c, &res);
+                       check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
+                       check_extent_overbig(trans, &iter, k);
+               })) ?:
+               check_i_sectors(trans, &w));
 
        bch2_disk_reservation_put(c, &res);
-       extent_ends_reset(&extent_ends);
-       darray_exit(&extent_ends);
+       extent_ends_exit(&extent_ends);
        inode_walker_exit(&w);
-       bch2_trans_exit(&trans);
        snapshots_seen_exit(&s);
 
-       if (ret)
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
+       return ret;
+}
+
+int bch2_check_indirect_extents(struct bch_fs *c)
+{
+       struct disk_reservation res = { 0 };
+
+       int ret = bch2_trans_run(c,
+               for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
+                               POS_MIN,
+                               BTREE_ITER_PREFETCH, k,
+                               &res, NULL,
+                               BCH_TRANS_COMMIT_no_enospc, ({
+                       bch2_disk_reservation_put(c, &res);
+                       check_extent_overbig(trans, &iter, k);
+               })));
+
+       bch2_disk_reservation_put(c, &res);
+       bch_err_fn(c, ret);
        return ret;
 }
 
 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
 {
        struct bch_fs *c = trans->c;
-       struct inode_walker_entry *i;
        u32 restart_count = trans->restart_count;
        int ret = 0;
        s64 count2;
@@ -1447,7 +1463,7 @@ static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
                if (i->inode.bi_nlink == i->count)
                        continue;
 
-               count2 = bch2_count_subdirs(trans, w->cur_inum, i->snapshot);
+               count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
                if (count2 < 0)
                        return count2;
 
@@ -1459,21 +1475,19 @@ static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
                                continue;
                }
 
-               if (fsck_err_on(i->inode.bi_nlink != i->count, c,
+               if (fsck_err_on(i->inode.bi_nlink != i->count,
+                               c, inode_dir_wrong_nlink,
                                "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
-                               w->cur_inum, i->snapshot, i->inode.bi_nlink, i->count)) {
+                               w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
                        i->inode.bi_nlink = i->count;
-                       ret = write_inode(trans, &i->inode, i->snapshot);
+                       ret = fsck_write_inode(trans, &i->inode, i->snapshot);
                        if (ret)
                                break;
                }
        }
 fsck_err:
-       if (ret)
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
-       if (!ret && trans_was_restarted(trans, restart_count))
-               ret = -BCH_ERR_transaction_restart_nested;
-       return ret;
+       bch_err_fn(c, ret);
+       return ret ?: trans_was_restarted(trans, restart_count);
 }
 
 static int check_dirent_target(struct btree_trans *trans,
@@ -1484,8 +1498,8 @@ static int check_dirent_target(struct btree_trans *trans,
 {
        struct bch_fs *c = trans->c;
        struct bkey_i_dirent *n;
-       bool backpointer_exists = true;
        struct printbuf buf = PRINTBUF;
+       struct btree_iter bp_iter = { NULL };
        int ret = 0;
 
        if (!target->bi_dir &&
@@ -1499,34 +1513,47 @@ static int check_dirent_target(struct btree_trans *trans,
        }
 
        if (!inode_points_to_dirent(target, d)) {
-               ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
-               if (ret < 0)
+               struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
+                                     SPOS(target->bi_dir, target->bi_dir_offset, target_snapshot));
+               ret = bkey_err(bp_dirent);
+               if (ret && !bch2_err_matches(ret, ENOENT))
                        goto err;
 
-               backpointer_exists = ret;
+               bool backpointer_exists = !ret;
                ret = 0;
 
-               if (fsck_err_on(S_ISDIR(target->bi_mode) &&
-                               backpointer_exists, c,
-                               "directory %llu with multiple links",
-                               target->bi_inum)) {
+               bch2_bkey_val_to_text(&buf, c, d.s_c);
+               prt_newline(&buf);
+               if (backpointer_exists)
+                       bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
+
+               if (fsck_err_on(S_ISDIR(target->bi_mode) && backpointer_exists,
+                               c, inode_dir_multiple_links,
+                               "directory %llu:%u with multiple links\n%s",
+                               target->bi_inum, target_snapshot, buf.buf)) {
                        ret = __remove_dirent(trans, d.k->p);
                        goto out;
                }
 
-               if (fsck_err_on(backpointer_exists &&
-                               !target->bi_nlink, c,
-                               "inode %llu type %s has multiple links but i_nlink 0",
-                               target->bi_inum, bch2_d_types[d.v->d_type])) {
+               /*
+                * hardlinked file with nlink 0:
+                * We're just adjusting nlink here so check_nlinks() will pick
+                * it up, it ignores inodes with nlink 0
+                */
+               if (fsck_err_on(backpointer_exists && !target->bi_nlink,
+                               c, inode_multiple_links_but_nlink_0,
+                               "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
+                               target->bi_inum, target_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
                        target->bi_nlink++;
-                       target->bi_flags &= ~BCH_INODE_UNLINKED;
+                       target->bi_flags &= ~BCH_INODE_unlinked;
 
                        ret = __write_inode(trans, target, target_snapshot);
                        if (ret)
                                goto err;
                }
 
-               if (fsck_err_on(!backpointer_exists, c,
+               if (fsck_err_on(!backpointer_exists,
+                               c, inode_wrong_backpointer,
                                "inode %llu:%u has wrong backpointer:\n"
                                "got       %llu:%llu\n"
                                "should be %llu:%llu",
@@ -1544,7 +1571,8 @@ static int check_dirent_target(struct btree_trans *trans,
                }
        }
 
-       if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
+       if (fsck_err_on(d.v->d_type != inode_d_type(target),
+                       c, dirent_d_type_wrong,
                        "incorrect d_type: got %s, should be %s:\n%s",
                        bch2_d_type_str(d.v->d_type),
                        bch2_d_type_str(inode_d_type(target)),
@@ -1565,12 +1593,12 @@ static int check_dirent_target(struct btree_trans *trans,
                d = dirent_i_to_s_c(n);
        }
 
-       if (d.v->d_type == DT_SUBVOL &&
-           target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
-           (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
-            fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
-                     le32_to_cpu(d.v->d_parent_subvol),
-                     target->bi_parent_subvol))) {
+       if (fsck_err_on(d.v->d_type == DT_SUBVOL &&
+                       target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol),
+                       c, dirent_d_parent_subvol_wrong,
+                       "dirent has wrong d_parent_subvol field: got %u, should be %u",
+                       le32_to_cpu(d.v->d_parent_subvol),
+                       target->bi_parent_subvol)) {
                n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
                ret = PTR_ERR_OR_ZERO(n);
                if (ret)
@@ -1588,10 +1616,9 @@ static int check_dirent_target(struct btree_trans *trans,
 out:
 err:
 fsck_err:
+       bch2_trans_iter_exit(trans, &bp_iter);
        printbuf_exit(&buf);
-
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
 }
 
@@ -1625,23 +1652,24 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
        if (k.k->type == KEY_TYPE_whiteout)
                goto out;
 
-       if (dir->cur_inum != k.k->p.inode) {
+       if (dir->last_pos.inode != k.k->p.inode) {
                ret = check_subdir_count(trans, dir);
                if (ret)
                        goto err;
        }
 
-       BUG_ON(!iter->path->should_be_locked);
+       BUG_ON(!btree_iter_path(trans, iter)->should_be_locked);
 
-       ret = __walk_inode(trans, dir, equiv);
+       i = walk_inode(trans, dir, equiv, k.k->type == KEY_TYPE_whiteout);
+       ret = PTR_ERR_OR_ZERO(i);
        if (ret < 0)
                goto err;
 
-       if (dir->first_this_inode)
+       if (dir->first_this_inode && dir->inodes.nr)
                *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
        dir->first_this_inode = false;
 
-       if (fsck_err_on(ret == INT_MAX, c,
+       if (fsck_err_on(!i, c, dirent_in_missing_dir_inode,
                        "dirent in nonexisting directory:\n%s",
                        (printbuf_reset(&buf),
                         bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
@@ -1650,15 +1678,11 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
                goto out;
        }
 
-       if (ret == INT_MAX) {
-               ret = 0;
+       if (!i)
                goto out;
-       }
-
-       i = dir->inodes.data + ret;
-       ret = 0;
 
-       if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
+       if (fsck_err_on(!S_ISDIR(i->inode.bi_mode),
+                       c, dirent_in_non_dir_inode,
                        "dirent in non directory inode type %s:\n%s",
                        bch2_d_type_str(inode_d_type(&i->inode)),
                        (printbuf_reset(&buf),
@@ -1687,24 +1711,24 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
                u32 target_snapshot;
                u64 target_inum;
 
-               ret = __subvol_lookup(trans, target_subvol,
+               ret = subvol_lookup(trans, target_subvol,
                                      &target_snapshot, &target_inum);
-               if (ret && ret != -ENOENT)
+               if (ret && !bch2_err_matches(ret, ENOENT))
                        goto err;
 
-               if (fsck_err_on(ret, c,
-                               "dirent points to missing subvolume %llu",
-                               le64_to_cpu(d.v->d_child_subvol))) {
+               if (fsck_err_on(ret, c, dirent_to_missing_subvol,
+                               "dirent points to missing subvolume %u",
+                               le32_to_cpu(d.v->d_child_subvol))) {
                        ret = __remove_dirent(trans, d.k->p);
                        goto err;
                }
 
-               ret = __lookup_inode(trans, target_inum,
+               ret = lookup_inode(trans, target_inum,
                                   &subvol_root, &target_snapshot);
-               if (ret && ret != -ENOENT)
+               if (ret && !bch2_err_matches(ret, ENOENT))
                        goto err;
 
-               if (fsck_err_on(ret, c,
+               if (fsck_err_on(ret, c, subvol_to_missing_root,
                                "subvolume %u points to missing subvolume root %llu",
                                target_subvol,
                                target_inum)) {
@@ -1713,7 +1737,8 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
                        goto err;
                }
 
-               if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
+               if (fsck_err_on(subvol_root.bi_subvol != target_subvol,
+                               c, subvol_root_wrong_bi_subvol,
                                "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
                                target_inum,
                                subvol_root.bi_subvol, target_subvol)) {
@@ -1732,7 +1757,8 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
                if (ret)
                        goto err;
 
-               if (fsck_err_on(!target->inodes.nr, c,
+               if (fsck_err_on(!target->inodes.nr,
+                               c, dirent_to_missing_inode,
                                "dirent points to missing inode: (equiv %u)\n%s",
                                equiv.snapshot,
                                (printbuf_reset(&buf),
@@ -1759,9 +1785,7 @@ out:
 err:
 fsck_err:
        printbuf_exit(&buf);
-
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
 }
 
@@ -1769,38 +1793,28 @@ fsck_err:
  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
  * validate d_type
  */
-noinline_for_stack
-static int check_dirents(struct bch_fs *c)
+int bch2_check_dirents(struct bch_fs *c)
 {
        struct inode_walker dir = inode_walker_init();
        struct inode_walker target = inode_walker_init();
        struct snapshots_seen s;
        struct bch_hash_info hash_info;
-       struct btree_trans trans;
-       struct btree_iter iter;
-       struct bkey_s_c k;
-       int ret = 0;
-
-       bch_verbose(c, "checking dirents");
 
        snapshots_seen_init(&s);
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 
-       ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_dirents,
-                       POS(BCACHEFS_ROOT_INO, 0),
-                       BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
-                       k,
-                       NULL, NULL,
-                       BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
-               check_dirent(&trans, &iter, k, &hash_info, &dir, &target, &s));
+       int ret = bch2_trans_run(c,
+               for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
+                               POS(BCACHEFS_ROOT_INO, 0),
+                               BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
+                               k,
+                               NULL, NULL,
+                               BCH_TRANS_COMMIT_no_enospc,
+                       check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)));
 
-       bch2_trans_exit(&trans);
        snapshots_seen_exit(&s);
        inode_walker_exit(&dir);
        inode_walker_exit(&target);
-
-       if (ret)
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
 }
 
@@ -1810,66 +1824,54 @@ static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
                       struct inode_walker *inode)
 {
        struct bch_fs *c = trans->c;
+       struct inode_walker_entry *i;
        int ret;
 
        ret = check_key_has_snapshot(trans, iter, k);
        if (ret)
                return ret;
 
-       ret = __walk_inode(trans, inode, k.k->p);
-       if (ret < 0)
+       i = walk_inode(trans, inode, k.k->p, k.k->type == KEY_TYPE_whiteout);
+       ret = PTR_ERR_OR_ZERO(i);
+       if (ret)
                return ret;
 
-       if (inode->first_this_inode)
+       if (inode->first_this_inode && inode->inodes.nr)
                *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
        inode->first_this_inode = false;
 
-       if (fsck_err_on(ret == INT_MAX, c,
+       if (fsck_err_on(!i, c, xattr_in_missing_inode,
                        "xattr for missing inode %llu",
                        k.k->p.inode))
                return bch2_btree_delete_at(trans, iter, 0);
 
-       if (ret == INT_MAX)
+       if (!i)
                return 0;
 
-       ret = 0;
-
        ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
 fsck_err:
-       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
 }
 
 /*
  * Walk xattrs: verify that they all have a corresponding inode
  */
-noinline_for_stack
-static int check_xattrs(struct bch_fs *c)
+int bch2_check_xattrs(struct bch_fs *c)
 {
        struct inode_walker inode = inode_walker_init();
        struct bch_hash_info hash_info;
-       struct btree_trans trans;
-       struct btree_iter iter;
-       struct bkey_s_c k;
        int ret = 0;
 
-       bch_verbose(c, "checking xattrs");
-
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
-       ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_xattrs,
+       ret = bch2_trans_run(c,
+               for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
                        POS(BCACHEFS_ROOT_INO, 0),
                        BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
                        k,
                        NULL, NULL,
-                       BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
-               check_xattr(&trans, &iter, k, &hash_info, &inode));
-
-       bch2_trans_exit(&trans);
-
-       if (ret)
-               bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
+                       BCH_TRANS_COMMIT_no_enospc,
+               check_xattr(trans, &iter, k, &hash_info, &inode)));
+       bch_err_fn(c, ret);
        return ret;
 }
 
@@ -1881,11 +1883,12 @@ static int check_root_trans(struct btree_trans *trans)
        u64 inum;
        int ret;
 
-       ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
-       if (ret && ret != -ENOENT)
+       ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
+       if (ret && !bch2_err_matches(ret, ENOENT))
                return ret;
 
-       if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
+       if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
+                               "root subvol missing")) {
                struct bkey_i_subvolume root_subvol;
 
                snapshot        = U32_MAX;
@@ -1896,32 +1899,27 @@ static int check_root_trans(struct btree_trans *trans)
                root_subvol.v.flags     = 0;
                root_subvol.v.snapshot  = cpu_to_le32(snapshot);
                root_subvol.v.inode     = cpu_to_le64(inum);
-               ret = commit_do(trans, NULL, NULL,
-                                     BTREE_INSERT_NOFAIL|
-                                     BTREE_INSERT_LAZY_RW,
-                       __bch2_btree_insert(trans, BTREE_ID_subvolumes,
-                                           &root_subvol.k_i, 0));
-               if (ret) {
-                       bch_err(c, "error writing root subvol: %s", bch2_err_str(ret));
+               ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol.k_i, 0);
+               bch_err_msg(c, ret, "writing root subvol");
+               if (ret)
                        goto err;
-               }
-
        }
 
-       ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
-       if (ret && ret != -ENOENT)
+       ret = lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
+       if (ret && !bch2_err_matches(ret, ENOENT))
                return ret;
 
-       if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
-           mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
+       if (mustfix_fsck_err_on(ret, c, root_dir_missing,
+                               "root directory missing") ||
+           mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
+                               c, root_inode_not_dir,
                                "root inode not a directory")) {
                bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
                                0, NULL);
                root_inode.bi_inum = inum;
 
                ret = __write_inode(trans, &root_inode, snapshot);
-               if (ret)
-                       bch_err(c, "error writing root inode: %s", bch2_err_str(ret));
+               bch_err_msg(c, ret, "writing root inode");
        }
 err:
 fsck_err:
@@ -1929,15 +1927,12 @@ fsck_err:
 }
 
 /* Get root directory, create if it doesn't exist: */
-noinline_for_stack
-static int check_root(struct bch_fs *c)
+int bch2_check_root(struct bch_fs *c)
 {
-       bch_verbose(c, "checking root directory");
-
-       return bch2_trans_do(c, NULL, NULL,
-                            BTREE_INSERT_NOFAIL|
-                            BTREE_INSERT_LAZY_RW,
-               check_root_trans(&trans));
+       int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+               check_root_trans(trans));
+       bch_err_fn(c, ret);
+       return ret;
 }
 
 struct pathbuf_entry {
@@ -1949,13 +1944,10 @@ typedef DARRAY(struct pathbuf_entry) pathbuf;
 
 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
 {
-       struct pathbuf_entry *i;
-
        darray_for_each(*p, i)
                if (i->inum     == inum &&
                    i->snapshot == snapshot)
                        return true;
-
        return false;
 }
 
@@ -2004,20 +1996,21 @@ static int check_path(struct btree_trans *trans,
                                break;
                }
 
-               ret = lockrestart_do(trans,
-                       PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
-                                         SPOS(inode->bi_dir, inode->bi_dir_offset,
-                                              parent_snapshot))).k));
-               if (ret && ret != -ENOENT)
+               d = dirent_get_by_pos(trans, &dirent_iter,
+                                     SPOS(inode->bi_dir, inode->bi_dir_offset,
+                                          parent_snapshot));
+               ret = bkey_err(d.s_c);
+               if (ret && !bch2_err_matches(ret, ENOENT))
                        break;
 
                if (!ret && !dirent_points_to_inode(d, inode)) {
                        bch2_trans_iter_exit(trans, &dirent_iter);
-                       ret = -ENOENT;
+                       ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
                }
 
-               if (ret == -ENOENT) {
-                       if (fsck_err(c,  "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
+               if (bch2_err_matches(ret, ENOENT)) {
+                       if (fsck_err(c,  inode_unreachable,
+                                    "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
                                     inode->bi_inum, snapshot,
                                     bch2_d_type_str(inode_d_type(inode)),
                                     inode->bi_nlink,
@@ -2043,13 +2036,12 @@ static int check_path(struct btree_trans *trans,
                ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
                if (ret) {
                        /* Should have been caught in dirents pass */
-                       bch_err(c, "error looking up parent directory: %i", ret);
+                       if (!bch2_err_matches(ret, BCH_ERR_transaction_restart))
+                               bch_err(c, "error looking up parent directory: %i", ret);
                        break;
                }
 
                if (path_is_dup(p, inode->bi_inum, snapshot)) {
-                       struct pathbuf_entry *i;
-
                        /* XXX print path */
                        bch_err(c, "directory structure loop");
 
@@ -2057,70 +2049,56 @@ static int check_path(struct btree_trans *trans,
                                pr_err("%llu:%u", i->inum, i->snapshot);
                        pr_err("%llu:%u", inode->bi_inum, snapshot);
 
-                       if (!fsck_err(c, "directory structure loop"))
+                       if (!fsck_err(c, dir_loop, "directory structure loop"))
                                return 0;
 
-                       ret = commit_do(trans, NULL, NULL,
-                                             BTREE_INSERT_NOFAIL|
-                                             BTREE_INSERT_LAZY_RW,
-                                       remove_backpointer(trans, inode));
-                       if (ret) {
-                               bch_err(c, "error removing dirent: %i", ret);
+                       ret = remove_backpointer(trans, inode);
+                       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
+                               bch_err_msg(c, ret, "removing dirent");
+                       if (ret)
                                break;
-                       }
 
                        ret = reattach_inode(trans, inode, snapshot);
+                       if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
+                               bch_err_msg(c, ret, "reattaching inode %llu", inode->bi_inum);
+                       break;
                }
        }
 fsck_err:
-       if (ret)
-               bch_err(c, "%s: err %s", __func__, bch2_err_str(ret));
+       bch_err_fn(c, ret);
        return ret;
 }
 
 /*
  * Check for unreachable inodes, as well as loops in the directory structure:
- * After check_dirents(), if an inode backpointer doesn't exist that means it's
+ * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
  * unreachable:
  */
-noinline_for_stack
-static int check_directory_structure(struct bch_fs *c)
+int bch2_check_directory_structure(struct bch_fs *c)
 {
-       struct btree_trans trans;
-       struct btree_iter iter;
-       struct bkey_s_c k;
        struct bch_inode_unpacked u;
        pathbuf path = { 0, };
        int ret;
 
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
-       for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
-                          BTREE_ITER_INTENT|
-                          BTREE_ITER_PREFETCH|
-                          BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-               if (!bkey_is_inode(k.k))
-                       continue;
-
-               ret = bch2_inode_unpack(k, &u);
-               if (ret) {
-                       /* Should have been caught earlier in fsck: */
-                       bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
-                       break;
-               }
+       ret = bch2_trans_run(c,
+               for_each_btree_key_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
+                                         BTREE_ITER_INTENT|
+                                         BTREE_ITER_PREFETCH|
+                                         BTREE_ITER_ALL_SNAPSHOTS, k,
+                                         NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
+                       if (!bkey_is_inode(k.k))
+                               continue;
 
-               if (u.bi_flags & BCH_INODE_UNLINKED)
-                       continue;
+                       BUG_ON(bch2_inode_unpack(k, &u));
 
-               ret = check_path(&trans, &path, &u, iter.pos.snapshot);
-               if (ret)
-                       break;
-       }
-       bch2_trans_iter_exit(&trans, &iter);
+                       if (u.bi_flags & BCH_INODE_unlinked)
+                               continue;
 
+                       check_path(trans, &path, &u, iter.pos.snapshot);
+               })));
        darray_exit(&path);
 
-       bch2_trans_exit(&trans);
+       bch_err_fn(c, ret);
        return ret;
 }
 
@@ -2170,7 +2148,7 @@ static int nlink_cmp(const void *_l, const void *_r)
        const struct nlink *l = _l;
        const struct nlink *r = _r;
 
-       return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
+       return cmp_int(l->inum, r->inum);
 }
 
 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
@@ -2205,49 +2183,39 @@ static int check_nlinks_find_hardlinks(struct bch_fs *c,
                                       struct nlink_table *t,
                                       u64 start, u64 *end)
 {
-       struct btree_trans trans;
-       struct btree_iter iter;
-       struct bkey_s_c k;
-       struct bch_inode_unpacked u;
-       int ret = 0;
-
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
-       for_each_btree_key(&trans, iter, BTREE_ID_inodes,
-                          POS(0, start),
-                          BTREE_ITER_INTENT|
-                          BTREE_ITER_PREFETCH|
-                          BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-               if (!bkey_is_inode(k.k))
-                       continue;
-
-               /* Should never fail, checked by bch2_inode_invalid: */
-               BUG_ON(bch2_inode_unpack(k, &u));
-
-               /*
-                * Backpointer and directory structure checks are sufficient for
-                * directories, since they can't have hardlinks:
-                */
-               if (S_ISDIR(le16_to_cpu(u.bi_mode)))
-                       continue;
+       int ret = bch2_trans_run(c,
+               for_each_btree_key(trans, iter, BTREE_ID_inodes,
+                                  POS(0, start),
+                                  BTREE_ITER_INTENT|
+                                  BTREE_ITER_PREFETCH|
+                                  BTREE_ITER_ALL_SNAPSHOTS, k, ({
+                       if (!bkey_is_inode(k.k))
+                               continue;
 
-               if (!u.bi_nlink)
-                       continue;
+                       /* Should never fail, checked by bch2_inode_invalid: */
+                       struct bch_inode_unpacked u;
+                       BUG_ON(bch2_inode_unpack(k, &u));
 
-               ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
-               if (ret) {
-                       *end = k.k->p.offset;
-                       ret = 0;
-                       break;
-               }
+                       /*
+                        * Backpointer and directory structure checks are sufficient for
+                        * directories, since they can't have hardlinks:
+                        */
+                       if (S_ISDIR(u.bi_mode))
+                               continue;
 
-       }
-       bch2_trans_iter_exit(&trans, &iter);
-       bch2_trans_exit(&trans);
+                       if (!u.bi_nlink)
+                               continue;
 
-       if (ret)
-               bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
+                       ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
+                       if (ret) {
+                               *end = k.k->p.offset;
+                               ret = 0;
+                               break;
+                       }
+                       0;
+               })));
 
+       bch_err_fn(c, ret);
        return ret;
 }
 
@@ -2255,44 +2223,34 @@ noinline_for_stack
 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
                                     u64 range_start, u64 range_end)
 {
-       struct btree_trans trans;
        struct snapshots_seen s;
-       struct btree_iter iter;
-       struct bkey_s_c k;
-       struct bkey_s_c_dirent d;
-       int ret;
 
        snapshots_seen_init(&s);
 
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
-       for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN,
-                          BTREE_ITER_INTENT|
-                          BTREE_ITER_PREFETCH|
-                          BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-               ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
-               if (ret)
-                       break;
-
-               switch (k.k->type) {
-               case KEY_TYPE_dirent:
-                       d = bkey_s_c_to_dirent(k);
+       int ret = bch2_trans_run(c,
+               for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
+                                  BTREE_ITER_INTENT|
+                                  BTREE_ITER_PREFETCH|
+                                  BTREE_ITER_ALL_SNAPSHOTS, k, ({
+                       ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
+                       if (ret)
+                               break;
 
-                       if (d.v->d_type != DT_DIR &&
-                           d.v->d_type != DT_SUBVOL)
-                               inc_link(c, &s, links, range_start, range_end,
-                                        le64_to_cpu(d.v->d_inum),
-                                        bch2_snapshot_equiv(c, d.k->p.snapshot));
-                       break;
-               }
-       }
-       bch2_trans_iter_exit(&trans, &iter);
+                       if (k.k->type == KEY_TYPE_dirent) {
+                               struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 
-       if (ret)
-               bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
+                               if (d.v->d_type != DT_DIR &&
+                                   d.v->d_type != DT_SUBVOL)
+                                       inc_link(c, &s, links, range_start, range_end,
+                                                le64_to_cpu(d.v->d_inum),
+                                                bch2_snapshot_equiv(c, d.k->p.snapshot));
+                       }
+                       0;
+               })));
 
-       bch2_trans_exit(&trans);
        snapshots_seen_exit(&s);
+
+       bch_err_fn(c, ret);
        return ret;
 }
 
@@ -2314,7 +2272,7 @@ static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_ite
 
        BUG_ON(bch2_inode_unpack(k, &u));
 
-       if (S_ISDIR(le16_to_cpu(u.bi_mode)))
+       if (S_ISDIR(u.bi_mode))
                return 0;
 
        if (!u.bi_nlink)
@@ -2326,7 +2284,8 @@ static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_ite
                link = &links->d[++*idx];
        }
 
-       if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
+       if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
+                       c, inode_wrong_nlink,
                        "inode %llu type %s has wrong i_nlink (%u, should be %u)",
                        u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
                        bch2_inode_nlink_get(&u), link->count)) {
@@ -2342,39 +2301,28 @@ static int check_nlinks_update_hardlinks(struct bch_fs *c,
                               struct nlink_table *links,
                               u64 range_start, u64 range_end)
 {
-       struct btree_trans trans;
-       struct btree_iter iter;
-       struct bkey_s_c k;
        size_t idx = 0;
-       int ret = 0;
-
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
-       ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes,
-                       POS(0, range_start),
-                       BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
-                       NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
-               check_nlinks_update_inode(&trans, &iter, k, links, &idx, range_end));
-
-       bch2_trans_exit(&trans);
 
+       int ret = bch2_trans_run(c,
+               for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
+                               POS(0, range_start),
+                               BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
+                               NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+                       check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
        if (ret < 0) {
-               bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
+               bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
                return ret;
        }
 
        return 0;
 }
 
-noinline_for_stack
-static int check_nlinks(struct bch_fs *c)
+int bch2_check_nlinks(struct bch_fs *c)
 {
        struct nlink_table links = { 0 };
        u64 this_iter_range_start, next_iter_range_start = 0;
        int ret = 0;
 
-       bch_verbose(c, "checking inode nlinks");
-
        do {
                this_iter_range_start = next_iter_range_start;
                next_iter_range_start = U64_MAX;
@@ -2399,7 +2347,7 @@ static int check_nlinks(struct bch_fs *c)
        } while (next_iter_range_start != U64_MAX);
 
        kvfree(links.d);
-
+       bch_err_fn(c, ret);
        return ret;
 }
 
@@ -2408,7 +2356,6 @@ static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
 {
        struct bkey_s_c_reflink_p p;
        struct bkey_i_reflink_p *u;
-       int ret;
 
        if (k.k->type != KEY_TYPE_reflink_p)
                return 0;
@@ -2419,7 +2366,7 @@ static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
                return 0;
 
        u = bch2_trans_kmalloc(trans, sizeof(*u));
-       ret = PTR_ERR_OR_ZERO(u);
+       int ret = PTR_ERR_OR_ZERO(u);
        if (ret)
                return ret;
 
@@ -2430,63 +2377,18 @@ static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
        return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
 }
 
-noinline_for_stack
-static int fix_reflink_p(struct bch_fs *c)
+int bch2_fix_reflink_p(struct bch_fs *c)
 {
-       struct btree_trans trans;
-       struct btree_iter iter;
-       struct bkey_s_c k;
-       int ret;
-
        if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
                return 0;
 
-       bch_verbose(c, "fixing reflink_p keys");
-
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
-       ret = for_each_btree_key_commit(&trans, iter,
-                       BTREE_ID_extents, POS_MIN,
-                       BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
-                       NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
-               fix_reflink_p_key(&trans, &iter, k));
-
-       bch2_trans_exit(&trans);
-       return ret;
-}
-
-/*
- * Checks for inconsistencies that shouldn't happen, unless we have a bug.
- * Doesn't fix them yet, mainly because they haven't yet been observed:
- */
-int bch2_fsck_full(struct bch_fs *c)
-{
-       int ret;
-again:
-       ret =   bch2_fs_check_snapshots(c) ?:
-               bch2_fs_check_subvols(c) ?:
-               bch2_delete_dead_snapshots(c) ?:
-               check_inodes(c, true) ?:
-               check_extents(c) ?:
-               check_dirents(c) ?:
-               check_xattrs(c) ?:
-               check_root(c) ?:
-               check_directory_structure(c) ?:
-               check_nlinks(c) ?:
-               fix_reflink_p(c);
-
-       if (bch2_err_matches(ret, BCH_ERR_need_snapshot_cleanup)) {
-               set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
-               goto again;
-       }
-
+       int ret = bch2_trans_run(c,
+               for_each_btree_key_commit(trans, iter,
+                               BTREE_ID_extents, POS_MIN,
+                               BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
+                               BTREE_ITER_ALL_SNAPSHOTS, k,
+                               NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+                       fix_reflink_p_key(trans, &iter, k)));
+       bch_err_fn(c, ret);
        return ret;
 }
-
-int bch2_fsck_walk_inodes_only(struct bch_fs *c)
-{
-       return  bch2_fs_check_snapshots(c) ?:
-               bch2_fs_check_subvols(c) ?:
-               bch2_delete_dead_snapshots(c) ?:
-               check_inodes(c, false);
-}