X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Ffsck.c;h=960280b706e382c75d630fd62724e11ebb209e46;hb=54b6beabf05b6cf62092f98f0c06395e4242b064;hp=12f2ef4417cb31255a405d5517ee727922ff5623;hpb=d01f633041c50452f9e837b7cc0223e6f37d42da;p=bcachefs-tools-debian

diff --git a/libbcachefs/fsck.c b/libbcachefs/fsck.c
index 12f2ef4..960280b 100644
--- a/libbcachefs/fsck.c
+++ b/libbcachefs/fsck.c
@@ -3,6 +3,7 @@
 #include "bcachefs.h"
 #include "bkey_buf.h"
 #include "btree_update.h"
+#include "buckets.h"
 #include "darray.h"
 #include "dirent.h"
 #include "error.h"
@@ -10,7 +11,8 @@
 #include "fsck.h"
 #include "inode.h"
 #include "keylist.h"
-#include "subvolume.h"
+#include "recovery.h"
+#include "snapshot.h"
 #include "super.h"
 #include "xattr.h"
 
@@ -31,14 +33,12 @@ static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
 	u64 sectors = 0;
 	int ret;
 
-	for_each_btree_key(trans, iter, BTREE_ID_extents,
-			   SPOS(inum, 0, snapshot), 0, k, ret) {
-		if (k.k->p.inode != inum)
-			break;
-
+	for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
+				SPOS(inum, 0, snapshot),
+				POS(inum, U64_MAX),
+				0, k, ret)
 		if (bkey_extent_is_allocation(k.k))
 			sectors += k.k->size;
-	}
 
 	bch2_trans_iter_exit(trans, &iter);
 
@@ -54,11 +54,10 @@ static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
 	u64 subdirs = 0;
 	int ret;
 
-	for_each_btree_key(trans, iter, BTREE_ID_dirents,
-			   SPOS(inum, 0, snapshot), 0, k, ret) {
-		if (k.k->p.inode != inum)
-			break;
-
+	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;
 
@@ -66,7 +65,6 @@ static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
 		if (d.v->d_type == DT_DIR)
 			subdirs++;
 	}
-
 	bch2_trans_iter_exit(trans, &iter);
 
 	return ret ?: subdirs;
@@ -75,26 +73,14 @@ static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
 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) {
+	struct bch_snapshot s;
+	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots,
+					  POS(0, snapshot), 0,
+					  snapshot, &s);
+	if (!ret)
+		*subvol = le32_to_cpu(s.subvol);
+	else if (bch2_err_matches(ret, ENOENT))
 		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;
 
 }
@@ -133,8 +119,8 @@ static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
 	if (ret)
 		goto err;
 
-	if (!k.k || bkey_cmp(k.k->p, POS(0, inode_nr))) {
-		ret = -ENOENT;
+	if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
+		ret = -BCH_ERR_ENOENT_inode;
 		goto err;
 	}
 
@@ -155,16 +141,15 @@ 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:
@@ -207,17 +192,18 @@ 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;
+
+	return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
+				&inode_p->inode.k_i,
+				BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 }
 
 static int write_inode(struct btree_trans *trans,
@@ -234,71 +220,6 @@ static int write_inode(struct btree_trans *trans,
 	return ret;
 }
 
-static int fsck_inode_rm(struct btree_trans *trans, u64 inum, 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;
-}
-
 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 {
 	struct bch_fs *c = trans->c;
@@ -321,7 +242,7 @@ static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 	bch2_trans_iter_exit(trans, &iter);
 err:
 	if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-		bch_err(c, "error from __remove_dirent(): %s", bch2_err_str(ret));
+		bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -351,7 +272,7 @@ static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
 
 	ret = __lookup_dirent(trans, root_hash_info, root_inum,
 			    &lostfound_str, &inum, &d_type);
-	if (ret == -ENOENT) {
+	if (bch2_err_matches(ret, ENOENT)) {
 		bch_notice(c, "creating lost+found");
 		goto create_lostfound;
 	}
@@ -367,7 +288,7 @@ static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
 	}
 
 	/*
-	 * 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);
@@ -456,22 +377,14 @@ 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;
 }
@@ -496,19 +409,19 @@ 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)
+		if (i->id == id)
+			return 0;
+		if (i->id > id)
 			break;
-
-		if (i->equiv == n.equiv) {
-			bch_err(c, "adding duplicate snapshot in snapshots_seen_add()");
-			return -EINVAL;
-		}
 	}
 
 	ret = darray_insert_item(&s->ids, i - s->ids.data, n);
@@ -527,30 +440,34 @@ static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 	};
 	int ret = 0;
 
-	if (bkey_cmp(s->pos, pos))
+	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_ids[btree_id],
+				pos.inode, pos.offset,
+				i->id, n.id, n.equiv);
+			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;
 }
 
@@ -564,15 +481,14 @@ static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *see
 				    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;
@@ -580,11 +496,20 @@ 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;
@@ -609,6 +534,20 @@ static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 		: bch2_snapshot_is_ancestor(c, src, dst);
 }
 
+static int ref_visible2(struct bch_fs *c,
+			u32 src, struct snapshots_seen *src_seen,
+			u32 dst, struct snapshots_seen *dst_seen)
+{
+	src = bch2_snapshot_equiv(c, src);
+	dst = bch2_snapshot_equiv(c, dst);
+
+	if (dst > src) {
+		swap(dst, src);
+		swap(dst_seen, src_seen);
+	}
+	return key_visible_in_snapshot(c, src_seen, dst, src);
+}
+
 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)				\
 	for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&	\
 	     (_i)->snapshot <= (_snapshot); _i++)					\
@@ -617,12 +556,14 @@ static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 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;
 };
@@ -650,28 +591,21 @@ 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) {
-		w->first_this_inode = false;
-		goto lookup_snapshot;
-	}
-
+	w->recalculate_sums = false;
 	w->inodes.nr = 0;
 
-	for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, pos.inode),
+	for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
 			   BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-		if (k.k->p.offset != pos.inode)
+		if (k.k->p.offset != inum)
 			break;
 
 		if (bkey_is_inode(k.k))
@@ -682,40 +616,76 @@ 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;
 
 	if (trans_was_restarted(trans, restart_count))
 		return -BCH_ERR_transaction_restart_nested;
 
-lookup_snapshot:
-	for (i = 0; i < w->inodes.nr; i++)
-		if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->inodes.data[i].snapshot))
+	return 0;
+}
+
+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);
+
+	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)) {
+		struct inode_walker_entry *i;
+
+		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,
@@ -779,12 +749,10 @@ static int hash_redo_key(struct btree_trans *trans,
 	if (IS_ERR(delete))
 		return PTR_ERR(delete);
 
-	tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+	tmp = bch2_bkey_make_mut_noupdate(trans, k);
 	if (IS_ERR(tmp))
 		return PTR_ERR(tmp);
 
-	bkey_reassemble(tmp, k);
-
 	bkey_init(&delete->k);
 	delete->k.p = k_iter->pos;
 	return  bch2_btree_iter_traverse(k_iter) ?:
@@ -823,9 +791,9 @@ static int hash_check_key(struct btree_trans *trans,
 		goto bad_hash;
 
 	for_each_btree_key_norestart(trans, iter, desc.btree_id,
-				     POS(hash_k.k->p.inode, hash),
+				     SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
 				     BTREE_ITER_SLOTS, k, ret) {
-		if (!bkey_cmp(k.k->p, hash_k.k->p))
+		if (bkey_eq(k.k->p, hash_k.k->p))
 			break;
 
 		if (fsck_err_on(k.k->type == desc.key_type &&
@@ -848,16 +816,15 @@ 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",
+	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,
 		     (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) {
+		if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 			bch_err(c, "hash_redo_key err %s", bch2_err_str(ret));
+		if (ret)
 			return ret;
-		}
 		ret = -BCH_ERR_transaction_restart_nested;
 	}
 fsck_err:
@@ -891,7 +858,7 @@ static int check_inode(struct btree_trans *trans,
 	 * 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))
+	if (bch2_snapshot_is_internal_node(c, k.k->p.snapshot))
 		return 0;
 
 	if (!bkey_is_inode(k.k))
@@ -922,7 +889,7 @@ static int check_inode(struct btree_trans *trans,
 		bch2_trans_unlock(trans);
 		bch2_fs_lazy_rw(c);
 
-		ret = fsck_inode_rm(trans, u.bi_inum, iter->pos.snapshot);
+		ret = bch2_inode_rm_snapshot(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));
@@ -947,11 +914,11 @@ static int check_inode(struct btree_trans *trans,
 				     iter->pos.snapshot),
 				POS(u.bi_inum, U64_MAX),
 				0, NULL);
-		if (ret) {
+		if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 			bch_err(c, "error in fsck: error truncating inode: %s",
 				bch2_err_str(ret));
+		if (ret)
 			return ret;
-		}
 
 		/*
 		 * We truncated without our normal sector accounting hook, just
@@ -1000,13 +967,14 @@ static int check_inode(struct btree_trans *trans,
 err:
 fsck_err:
 	if (ret)
-		bch_err(c, "error from check_inode(): %s", 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)
 {
+	bool full = c->opts.fsck;
 	struct btree_trans trans;
 	struct btree_iter iter;
 	struct bch_inode_unpacked prev = { 0 };
@@ -1026,69 +994,15 @@ static int check_inodes(struct bch_fs *c, bool full)
 	bch2_trans_exit(&trans);
 	snapshots_seen_exit(&s);
 	if (ret)
-		bch_err(c, "error from check_inodes(): %s", bch2_err_str(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);
+		bch_err_fn(c, ret);
 	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,
@@ -1116,9 +1030,9 @@ static int inode_backpointer_exists(struct btree_trans *trans,
 
 	d = dirent_get_by_pos(trans, &iter,
 			SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
-	ret = bkey_err(d.s_c);
+	ret = bkey_err(d);
 	if (ret)
-		return ret == -ENOENT ? 0 : ret;
+		return bch2_err_matches(ret, ENOENT) ? 0 : ret;
 
 	ret = dirent_points_to_inode(d, inode);
 	bch2_trans_iter_exit(trans, &iter);
@@ -1137,19 +1051,20 @@ 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,
+			    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);
@@ -1159,185 +1074,388 @@ static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
 	}
 fsck_err:
 	if (ret)
-		bch_err(c, "error from check_i_sectors(): %s", bch2_err_str(ret));
+		bch_err_fn(c, ret);
 	if (!ret && trans_was_restarted(trans, restart_count))
 		ret = -BCH_ERR_transaction_restart_nested;
 	return ret;
 }
 
-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)
+struct extent_end {
+	u32			snapshot;
+	u64			offset;
+	struct snapshots_seen	seen;
+};
+
+struct extent_ends {
+	struct bpos			last_pos;
+	DARRAY(struct extent_end)	e;
+};
+
+static void extent_ends_reset(struct extent_ends *extent_ends)
 {
-	struct bch_fs *c = trans->c;
-	struct inode_walker_entry *i;
-	struct printbuf buf = PRINTBUF;
-	struct bpos equiv;
-	int ret = 0;
+	struct extent_end *i;
 
-	ret = check_key_has_snapshot(trans, iter, k);
-	if (ret) {
-		ret = ret < 0 ? ret : 0;
-		goto out;
+	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;
 	}
 
-	equiv = k.k->p;
-	equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
+	return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
+}
 
-	ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
+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;
 
-	if (k.k->type == KEY_TYPE_whiteout)
-		goto out;
+	prt_str(&buf, "\n  ");
+	bch2_bkey_val_to_text(&buf, c, k1);
 
-	if (inode->cur_inum != k.k->p.inode) {
-		ret = check_i_sectors(trans, inode);
+	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;
 	}
 
-	BUG_ON(!iter->path->should_be_locked);
-#if 0
-	if (bkey_cmp(prev.k->k.p, bkey_start_pos(k.k)) > 0) {
-		char buf1[200];
-		char buf2[200];
+	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;
+	}
 
-		bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
-		bch2_bkey_val_to_text(&PBUF(buf2), c, k);
+	prt_printf(&buf, "\n  overwriting %s extent",
+		   pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
 
-		if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2)) {
-			ret = fix_overlapping_extent(trans, k, prev.k->k.p)
-				?: -BCH_ERR_transaction_restart_nested;
-			goto out;
+	if (fsck_err(c, "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_journal_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,
+				BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL);
+		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;
 		}
 	}
-#endif
-	ret = __walk_inode(trans, inode, equiv);
-	if (ret < 0)
+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,
+			      struct extent_ends *extent_ends,
+			      struct bkey_s_c k,
+			      u32 equiv,
+			      struct btree_iter *iter,
+			      bool *fixed)
+{
+	struct bch_fs *c = trans->c;
+	struct extent_end *i;
+	int ret = 0;
+
+	/* 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,
+				  k.k->p.snapshot, seen,
+				  i->snapshot, &i->seen))
+			continue;
+
+		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;
 
-	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);
+	extent_ends->last_pos = k.k->p;
+err:
+	return ret;
+}
+
+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,
+			struct extent_ends *extent_ends)
+{
+	struct bch_fs *c = trans->c;
+	struct inode_walker_entry *i;
+	struct printbuf buf = PRINTBUF;
+	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;
 	}
 
-	if (ret == INT_MAX) {
-		ret = 0;
-		goto out;
+	if (inode->last_pos.inode != k.k->p.inode) {
+		ret = check_i_sectors(trans, inode);
+		if (ret)
+			goto err;
 	}
 
-	i = inode->inodes.data + ret;
-	ret = 0;
+	i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
+	ret = PTR_ERR_OR_ZERO(i);
+	if (ret)
+		goto err;
 
-	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 = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
+	if (ret)
+		goto err;
+
+	if (k.k->type != KEY_TYPE_whiteout) {
+		if (fsck_err_on(!i, c,
+				"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 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;
+
+		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->type != KEY_TYPE_reservation &&
-				k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9, 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 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, "error from check_extent(): %s", 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;
+	struct extent_ends extent_ends;
+	struct disk_reservation res = { 0 };
 	int ret = 0;
 
-#if 0
-	struct bkey_buf prev;
-	bch2_bkey_buf_init(&prev);
-	prev.k->k = KEY(0, 0, 0);
-#endif
 	snapshots_seen_init(&s);
-	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
-
-	bch_verbose(c, "checking extents");
+	extent_ends_init(&extent_ends);
+	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 4096);
 
 	ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_extents,
 			POS(BCACHEFS_ROOT_INO, 0),
 			BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
-			NULL, NULL,
-			BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
-		check_extent(&trans, &iter, k, &w, &s));
-#if 0
-	bch2_bkey_buf_exit(&prev, c);
-#endif
+			&res, NULL,
+			BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({
+		bch2_disk_reservation_put(c, &res);
+		check_extent(&trans, &iter, k, &w, &s, &extent_ends);
+	})) ?:
+	check_i_sectors(&trans, &w);
+
+	bch2_disk_reservation_put(c, &res);
+	extent_ends_exit(&extent_ends);
 	inode_walker_exit(&w);
 	bch2_trans_exit(&trans);
 	snapshots_seen_exit(&s);
 
 	if (ret)
-		bch_err(c, "error from check_extents(): %s", bch2_err_str(ret));
+		bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -1353,7 +1471,7 @@ 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;
 
@@ -1367,7 +1485,7 @@ static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
 
 		if (fsck_err_on(i->inode.bi_nlink != i->count, c,
 				"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);
 			if (ret)
@@ -1376,7 +1494,7 @@ static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
 	}
 fsck_err:
 	if (ret)
-		bch_err(c, "error from check_subdir_count(): %s", bch2_err_str(ret));
+		bch_err_fn(c, ret);
 	if (!ret && trans_was_restarted(trans, restart_count))
 		ret = -BCH_ERR_transaction_restart_nested;
 	return ret;
@@ -1497,7 +1615,7 @@ fsck_err:
 	printbuf_exit(&buf);
 
 	if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-		bch_err(c, "error from check_target(): %s", bch2_err_str(ret));
+		bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -1531,7 +1649,7 @@ 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;
@@ -1539,11 +1657,16 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 
 	BUG_ON(!iter->path->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 (fsck_err_on(ret == INT_MAX, c,
+	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(!i, c,
 			"dirent in nonexisting directory:\n%s",
 			(printbuf_reset(&buf),
 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
@@ -1552,13 +1675,8 @@ 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,
 			"dirent in non directory inode type %s:\n%s",
@@ -1569,11 +1687,7 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 		goto out;
 	}
 
-	if (dir->first_this_inode)
-		*hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
-
-	ret = hash_check_key(trans, bch2_dirent_hash_desc,
-			     hash_info, iter, k);
+	ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
 	if (ret < 0)
 		goto err;
 	if (ret) {
@@ -1595,19 +1709,19 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 
 		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))) {
+				"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,
 				   &subvol_root, &target_snapshot);
-		if (ret && ret != -ENOENT)
+		if (ret && !bch2_err_matches(ret, ENOENT))
 			goto err;
 
 		if (fsck_err_on(ret, c,
@@ -1667,7 +1781,7 @@ fsck_err:
 	printbuf_exit(&buf);
 
 	if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
-		bch_err(c, "error from check_dirent(): %s", bch2_err_str(ret));
+		bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -1675,8 +1789,7 @@ 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();
@@ -1687,8 +1800,6 @@ static int check_dirents(struct bch_fs *c)
 	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);
 
@@ -1706,7 +1817,7 @@ static int check_dirents(struct bch_fs *c)
 	inode_walker_exit(&target);
 
 	if (ret)
-		bch_err(c, "error from check_dirents(): %s", bch2_err_str(ret));
+		bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -1716,41 +1827,41 @@ 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 (fsck_err_on(ret == INT_MAX, c,
+	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(!i, c,
 			"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;
-
-	if (inode->first_this_inode)
-		*hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
-
 	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, "error from check_xattr(): %s", 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;
@@ -1759,8 +1870,6 @@ static int check_xattrs(struct bch_fs *c)
 	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,
@@ -1774,7 +1883,7 @@ static int check_xattrs(struct bch_fs *c)
 	bch2_trans_exit(&trans);
 
 	if (ret)
-		bch_err(c, "error from check_xattrs(): %s", bch2_err_str(ret));
+		bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -1787,7 +1896,7 @@ static int check_root_trans(struct btree_trans *trans)
 	int ret;
 
 	ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
-	if (ret && ret != -ENOENT)
+	if (ret && !bch2_err_matches(ret, ENOENT))
 		return ret;
 
 	if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
@@ -1804,7 +1913,8 @@ static int check_root_trans(struct btree_trans *trans)
 		ret = commit_do(trans, NULL, NULL,
 				      BTREE_INSERT_NOFAIL|
 				      BTREE_INSERT_LAZY_RW,
-			__bch2_btree_insert(trans, BTREE_ID_subvolumes, &root_subvol.k_i));
+			__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));
 			goto err;
@@ -1813,7 +1923,7 @@ static int check_root_trans(struct btree_trans *trans)
 	}
 
 	ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
-	if (ret && ret != -ENOENT)
+	if (ret && !bch2_err_matches(ret, ENOENT))
 		return ret;
 
 	if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
@@ -1833,15 +1943,18 @@ 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");
+	int ret;
 
-	return bch2_trans_do(c, NULL, NULL,
+	ret = bch2_trans_do(c, NULL, NULL,
 			     BTREE_INSERT_NOFAIL|
 			     BTREE_INSERT_LAZY_RW,
 		check_root_trans(&trans));
+
+	if (ret)
+		bch_err_fn(c, ret);
+	return ret;
 }
 
 struct pathbuf_entry {
@@ -1912,15 +2025,15 @@ static int check_path(struct btree_trans *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)
+		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 (bch2_err_matches(ret, ENOENT)) {
 			if (fsck_err(c,  "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
 				     inode->bi_inum, snapshot,
 				     bch2_d_type_str(inode_d_type(inode)),
@@ -1978,17 +2091,16 @@ static int check_path(struct btree_trans *trans,
 	}
 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;
@@ -2021,13 +2133,16 @@ static int check_directory_structure(struct bch_fs *c)
 			break;
 	}
 	bch2_trans_iter_exit(&trans, &iter);
-
+	bch2_trans_exit(&trans);
 	darray_exit(&path);
 
-	bch2_trans_exit(&trans);
+	if (ret)
+		bch_err_fn(c, ret);
 	return ret;
 }
 
+/* check_nlink pass: */
+
 struct nlink_table {
 	size_t		nr;
 	size_t		size;
@@ -2044,11 +2159,12 @@ static int add_nlink(struct bch_fs *c, struct nlink_table *t,
 {
 	if (t->nr == t->size) {
 		size_t new_size = max_t(size_t, 128UL, t->size * 2);
-		void *d = kvmalloc(new_size * sizeof(t->d[0]), GFP_KERNEL);
+		void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
+
 		if (!d) {
 			bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
 				new_size);
-			return -ENOMEM;
+			return -BCH_ERR_ENOMEM_fsck_add_nlink;
 		}
 
 		if (t->d)
@@ -2131,7 +2247,7 @@ static int check_nlinks_find_hardlinks(struct bch_fs *c,
 		 * Backpointer and directory structure checks are sufficient for
 		 * directories, since they can't have hardlinks:
 		 */
-		if (S_ISDIR(le16_to_cpu(u.bi_mode)))
+		if (S_ISDIR(u.bi_mode))
 			continue;
 
 		if (!u.bi_nlink)
@@ -2217,7 +2333,7 @@ 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)
@@ -2269,15 +2385,12 @@ static int check_nlinks_update_hardlinks(struct bch_fs *c,
 	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;
@@ -2303,6 +2416,8 @@ static int check_nlinks(struct bch_fs *c)
 
 	kvfree(links.d);
 
+	if (ret)
+		bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -2333,10 +2448,8 @@ 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;
@@ -2344,52 +2457,15 @@ static int fix_reflink_p(struct bch_fs *c)
 	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;
-	}
+	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, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
+			fix_reflink_p_key(&trans, &iter, k)));
 
+	if (ret)
+		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);
-}