Merge tag 'bcachefs-2024-04-01' of https://evilpiepirate.org/git/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "Lots of fixes for situations with extreme filesystem damage.

  One fix ("Fix journal pins in btree write buffer") applicable to
  normal usage; also a dio performance fix.

  New repair/construction code is in the final stages, should be ready
  in about a week. Anyone that lost btree interior nodes (or a variety
  of other damage) as a result of the splitbrain bug will be able to
  repair then"

* tag 'bcachefs-2024-04-01' of https://evilpiepirate.org/git/bcachefs: (32 commits)
  bcachefs: On emergency shutdown, print out current journal sequence number
  bcachefs: Fix overlapping extent repair
  bcachefs: Fix remove_dirent()
  bcachefs: Logged op errors should be ignored
  bcachefs: Improve -o norecovery; opts.recovery_pass_limit
  bcachefs: bch2_run_explicit_recovery_pass_persistent()
  bcachefs: Ensure bch_sb_field_ext always exists
  bcachefs: Flush journal immediately after replay if we did early repair
  bcachefs: Resume logged ops after fsck
  bcachefs: Add error messages to logged ops fns
  bcachefs: Split out recovery_passes.c
  bcachefs: fix backpointer for missing alloc key msg
  bcachefs: Fix bch2_btree_increase_depth()
  bcachefs: Kill bch2_bkey_ptr_data_type()
  bcachefs: Fix use after free in check_root_trans()
  bcachefs: Fix repair path for missing indirect extents
  bcachefs: Fix use after free in bch2_check_fix_ptrs()
  bcachefs: Fix btree node keys accounting in topology repair path
  bcachefs: Check btree ptr min_key in .invalid
  bcachefs: add REQ_SYNC and REQ_IDLE in write dio
  ...

48 files changed, 814 insertions, 646 deletions

diff --git a/fs/bcachefs/Makefile b/fs/bcachefs/Makefile
index b02796c8a59533..f6d86eb4b97652 100644
--- a/fs/bcachefs/Makefile
+++ b/fs/bcachefs/Makefile
@@ -67,6 +67,7 @@ bcachefs-y		:=	\
 	quota.o			\
 	rebalance.o		\
 	recovery.o		\
+	recovery_passes.o	\
 	reflink.o		\
 	replicas.o		\
 	sb-clean.o		\
diff --git a/fs/bcachefs/backpointers.c b/fs/bcachefs/backpointers.c
index 8cb35ea572cb95..762c8ddfc5e73d 100644
--- a/fs/bcachefs/backpointers.c
+++ b/fs/bcachefs/backpointers.c
@@ -29,8 +29,7 @@ static bool extent_matches_bp(struct bch_fs *c,
 		if (p.ptr.cached)
 			continue;
 
-		bch2_extent_ptr_to_bp(c, btree_id, level, k, p,
-				      &bucket2, &bp2);
+		bch2_extent_ptr_to_bp(c, btree_id, level, k, p, entry, &bucket2, &bp2);
 		if (bpos_eq(bucket, bucket2) &&
 		    !memcmp(&bp, &bp2, sizeof(bp)))
 			return true;
@@ -44,6 +43,11 @@ int bch2_backpointer_invalid(struct bch_fs *c, struct bkey_s_c k,
 			     struct printbuf *err)
 {
 	struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
+
+	/* these will be caught by fsck */
+	if (!bch2_dev_exists2(c, bp.k->p.inode))
+		return 0;
+
 	struct bpos bucket = bp_pos_to_bucket(c, bp.k->p);
 	int ret = 0;
 
@@ -378,7 +382,7 @@ static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_
 			backpointer_to_missing_alloc,
 			"backpointer for nonexistent alloc key: %llu:%llu:0\n%s",
 			alloc_iter.pos.inode, alloc_iter.pos.offset,
-			(bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
+			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 		ret = bch2_btree_delete_at(trans, bp_iter, 0);
 		goto out;
 	}
@@ -502,8 +506,7 @@ static int check_extent_to_backpointers(struct btree_trans *trans,
 		if (p.ptr.cached)
 			continue;
 
-		bch2_extent_ptr_to_bp(c, btree, level,
-				      k, p, &bucket_pos, &bp);
+		bch2_extent_ptr_to_bp(c, btree, level, k, p, entry, &bucket_pos, &bp);
 
 		ret = check_bp_exists(trans, s, bucket_pos, bp, k);
 		if (ret)
diff --git a/fs/bcachefs/backpointers.h b/fs/bcachefs/backpointers.h
index 327365a9feac4e..da012ca7daee55 100644
--- a/fs/bcachefs/backpointers.h
+++ b/fs/bcachefs/backpointers.h
@@ -90,20 +90,40 @@ static inline int bch2_bucket_backpointer_mod(struct btree_trans *trans,
 	return bch2_trans_update_buffered(trans, BTREE_ID_backpointers, &bp_k.k_i);
 }
 
-static inline enum bch_data_type bkey_ptr_data_type(enum btree_id btree_id, unsigned level,
-						    struct bkey_s_c k, struct extent_ptr_decoded p)
+static inline enum bch_data_type bch2_bkey_ptr_data_type(struct bkey_s_c k,
+							 struct extent_ptr_decoded p,
+							 const union bch_extent_entry *entry)
 {
-	return  level		? BCH_DATA_btree :
-		p.has_ec	? BCH_DATA_stripe :
-				  BCH_DATA_user;
+	switch (k.k->type) {
+	case KEY_TYPE_btree_ptr:
+	case KEY_TYPE_btree_ptr_v2:
+		return BCH_DATA_btree;
+	case KEY_TYPE_extent:
+	case KEY_TYPE_reflink_v:
+		return p.has_ec ? BCH_DATA_stripe : BCH_DATA_user;
+	case KEY_TYPE_stripe: {
+		const struct bch_extent_ptr *ptr = &entry->ptr;
+		struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
+
+		BUG_ON(ptr < s.v->ptrs ||
+		       ptr >= s.v->ptrs + s.v->nr_blocks);
+
+		return ptr >= s.v->ptrs + s.v->nr_blocks - s.v->nr_redundant
+			? BCH_DATA_parity
+			: BCH_DATA_user;
+	}
+	default:
+		BUG();
+	}
 }
 
 static inline void bch2_extent_ptr_to_bp(struct bch_fs *c,
 			   enum btree_id btree_id, unsigned level,
 			   struct bkey_s_c k, struct extent_ptr_decoded p,
+			   const union bch_extent_entry *entry,
 			   struct bpos *bucket_pos, struct bch_backpointer *bp)
 {
-	enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
+	enum bch_data_type data_type = bch2_bkey_ptr_data_type(k, p, entry);
 	s64 sectors = level ? btree_sectors(c) : k.k->size;
 	u32 bucket_offset;
 
diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index 799aa32b6b4d99..963162a627cd6a 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -209,7 +209,7 @@
 #include "fifo.h"
 #include "nocow_locking_types.h"
 #include "opts.h"
-#include "recovery_types.h"
+#include "recovery_passes_types.h"
 #include "sb-errors_types.h"
 #include "seqmutex.h"
 #include "time_stats.h"
@@ -810,7 +810,6 @@ struct bch_fs {
 
 	/* snapshot.c: */
 	struct snapshot_table __rcu *snapshots;
-	size_t			snapshot_table_size;
 	struct mutex		snapshot_table_lock;
 	struct rw_semaphore	snapshot_create_lock;
 
diff --git a/fs/bcachefs/bset.c b/fs/bcachefs/bset.c
index 3fd1085b6c61ee..3bb477840eab6b 100644
--- a/fs/bcachefs/bset.c
+++ b/fs/bcachefs/bset.c
@@ -134,18 +134,24 @@ void bch2_dump_btree_node_iter(struct btree *b,
 	printbuf_exit(&buf);
 }
 
-#ifdef CONFIG_BCACHEFS_DEBUG
-
-void __bch2_verify_btree_nr_keys(struct btree *b)
+struct btree_nr_keys bch2_btree_node_count_keys(struct btree *b)
 {
 	struct bset_tree *t;
 	struct bkey_packed *k;
-	struct btree_nr_keys nr = { 0 };
+	struct btree_nr_keys nr = {};
 
 	for_each_bset(b, t)
 		bset_tree_for_each_key(b, t, k)
 			if (!bkey_deleted(k))
 				btree_keys_account_key_add(&nr, t - b->set, k);
+	return nr;
+}
+
+#ifdef CONFIG_BCACHEFS_DEBUG
+
+void __bch2_verify_btree_nr_keys(struct btree *b)
+{
+	struct btree_nr_keys nr = bch2_btree_node_count_keys(b);
 
 	BUG_ON(memcmp(&nr, &b->nr, sizeof(nr)));
 }
diff --git a/fs/bcachefs/bset.h b/fs/bcachefs/bset.h
index 79c77baaa38386..120a79fd456bd5 100644
--- a/fs/bcachefs/bset.h
+++ b/fs/bcachefs/bset.h
@@ -458,6 +458,8 @@ struct bkey_s_c bch2_btree_node_iter_peek_unpack(struct btree_node_iter *,
 
 /* Accounting: */
 
+struct btree_nr_keys bch2_btree_node_count_keys(struct btree *);
+
 static inline void btree_keys_account_key(struct btree_nr_keys *n,
 					  unsigned bset,
 					  struct bkey_packed *k,
diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c
index 562561a9a510e8..84474324dba9b5 100644
--- a/fs/bcachefs/btree_cache.c
+++ b/fs/bcachefs/btree_cache.c
@@ -808,7 +808,8 @@ static noinline void btree_bad_header(struct bch_fs *c, struct btree *b)
 	prt_printf(&buf, "\nmax ");
 	bch2_bpos_to_text(&buf, b->data->max_key);
 
-	bch2_fs_inconsistent(c, "%s", buf.buf);
+	bch2_fs_topology_error(c, "%s", buf.buf);
+
 	printbuf_exit(&buf);
 }
 
@@ -1134,6 +1135,8 @@ void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
 	b = btree_cache_find(bc, k);
 	if (!b)
 		return;
+
+	BUG_ON(b == btree_node_root(trans->c, b));
 wait_on_io:
 	/* not allowed to wait on io with btree locks held: */
 
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index bdaed29f084a4d..e5d2c6daa663de 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -7,6 +7,7 @@
 #include "bcachefs.h"
 #include "alloc_background.h"
 #include "alloc_foreground.h"
+#include "backpointers.h"
 #include "bkey_methods.h"
 #include "bkey_buf.h"
 #include "btree_journal_iter.h"
@@ -24,7 +25,7 @@
 #include "journal.h"
 #include "keylist.h"
 #include "move.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "reflink.h"
 #include "replicas.h"
 #include "super-io.h"
@@ -70,90 +71,6 @@ static inline void gc_pos_set(struct bch_fs *c, struct gc_pos new_pos)
 	__gc_pos_set(c, new_pos);
 }
 
-/*
- * Missing: if an interior btree node is empty, we need to do something -
- * perhaps just kill it
- */
-static int bch2_gc_check_topology(struct bch_fs *c,
-				  struct btree *b,
-				  struct bkey_buf *prev,
-				  struct bkey_buf cur,
-				  bool is_last)
-{
-	struct bpos node_start	= b->data->min_key;
-	struct bpos node_end	= b->data->max_key;
-	struct bpos expected_start = bkey_deleted(&prev->k->k)
-		? node_start
-		: bpos_successor(prev->k->k.p);
-	struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
-	int ret = 0;
-
-	if (cur.k->k.type == KEY_TYPE_btree_ptr_v2) {
-		struct bkey_i_btree_ptr_v2 *bp = bkey_i_to_btree_ptr_v2(cur.k);
-
-		if (!bpos_eq(expected_start, bp->v.min_key)) {
-			bch2_topology_error(c);
-
-			if (bkey_deleted(&prev->k->k)) {
-				prt_printf(&buf1, "start of node: ");
-				bch2_bpos_to_text(&buf1, node_start);
-			} else {
-				bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(prev->k));
-			}
-			bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(cur.k));
-
-			if (__fsck_err(c,
-				       FSCK_CAN_FIX|
-				       FSCK_CAN_IGNORE|
-				       FSCK_NO_RATELIMIT,
-				       btree_node_topology_bad_min_key,
-				       "btree node with incorrect min_key at btree %s level %u:\n"
-				       "  prev %s\n"
-				       "  cur %s",
-				       bch2_btree_id_str(b->c.btree_id), b->c.level,
-				       buf1.buf, buf2.buf) && should_restart_for_topology_repair(c)) {
-				bch_info(c, "Halting mark and sweep to start topology repair pass");
-				ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology);
-				goto err;
-			} else {
-				set_bit(BCH_FS_initial_gc_unfixed, &c->flags);
-			}
-		}
-	}
-
-	if (is_last && !bpos_eq(cur.k->k.p, node_end)) {
-		bch2_topology_error(c);
-
-		printbuf_reset(&buf1);
-		printbuf_reset(&buf2);
-
-		bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(cur.k));
-		bch2_bpos_to_text(&buf2, node_end);
-
-		if (__fsck_err(c, FSCK_CAN_FIX|FSCK_CAN_IGNORE|FSCK_NO_RATELIMIT,
-			  btree_node_topology_bad_max_key,
-			  "btree node with incorrect max_key at btree %s level %u:\n"
-			  "  %s\n"
-			  "  expected %s",
-			  bch2_btree_id_str(b->c.btree_id), b->c.level,
-			  buf1.buf, buf2.buf) &&
-		    should_restart_for_topology_repair(c)) {
-			bch_info(c, "Halting mark and sweep to start topology repair pass");
-			ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology);
-			goto err;
-		} else {
-			set_bit(BCH_FS_initial_gc_unfixed, &c->flags);
-		}
-	}
-
-	bch2_bkey_buf_copy(prev, c, cur.k);
-err:
-fsck_err:
-	printbuf_exit(&buf2);
-	printbuf_exit(&buf1);
-	return ret;
-}
-
 static void btree_ptr_to_v2(struct btree *b, struct bkey_i_btree_ptr_v2 *dst)
 {
 	switch (b->key.k.type) {
@@ -445,6 +362,7 @@ again:
 		prev = NULL;
 
 		if (ret == DROP_PREV_NODE) {
+			bch_info(c, "dropped prev node");
 			bch2_btree_node_evict(trans, prev_k.k);
 			ret = bch2_journal_key_delete(c, b->c.btree_id,
 						      b->c.level, prev_k.k->k.p);
@@ -591,7 +509,7 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 	bkey_for_each_ptr_decode(k->k, ptrs_c, p, entry_c) {
 		struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
 		struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
-		enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, &entry_c->ptr);
+		enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, p, entry_c);
 
 		if (fsck_err_on(!g->gen_valid,
 				c, ptr_to_missing_alloc_key,
@@ -657,7 +575,8 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 			continue;
 
 		if (fsck_err_on(bucket_data_type(g->data_type) &&
-				bucket_data_type(g->data_type) != data_type, c,
+				bucket_data_type(g->data_type) !=
+				bucket_data_type(data_type), c,
 				ptr_bucket_data_type_mismatch,
 				"bucket %u:%zu different types of data in same bucket: %s, %s\n"
 				"while marking %s",
@@ -698,18 +617,13 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 	}
 
 	if (do_update) {
-		struct bkey_ptrs ptrs;
-		union bch_extent_entry *entry;
-		struct bch_extent_ptr *ptr;
-		struct bkey_i *new;
-
 		if (is_root) {
 			bch_err(c, "cannot update btree roots yet");
 			ret = -EINVAL;
 			goto err;
 		}
 
-		new = kmalloc(bkey_bytes(k->k), GFP_KERNEL);
+		struct bkey_i *new = kmalloc(bkey_bytes(k->k), GFP_KERNEL);
 		if (!new) {
 			ret = -BCH_ERR_ENOMEM_gc_repair_key;
 			bch_err_msg(c, ret, "allocating new key");
@@ -724,7 +638,7 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 			 * btree node isn't there anymore, the read path will
 			 * sort it out:
 			 */
-			ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
+			struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
 			bkey_for_each_ptr(ptrs, ptr) {
 				struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 				struct bucket *g = PTR_GC_BUCKET(ca, ptr);
@@ -732,19 +646,26 @@ static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id
 				ptr->gen = g->gen;
 			}
 		} else {
-			bch2_bkey_drop_ptrs(bkey_i_to_s(new), ptr, ({
-				struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
-				struct bucket *g = PTR_GC_BUCKET(ca, ptr);
-				enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, ptr);
-
-				(ptr->cached &&
-				 (!g->gen_valid || gen_cmp(ptr->gen, g->gen) > 0)) ||
-				(!ptr->cached &&
-				 gen_cmp(ptr->gen, g->gen) < 0) ||
-				gen_cmp(g->gen, ptr->gen) > BUCKET_GC_GEN_MAX ||
-				(g->data_type &&
-				 g->data_type != data_type);
-			}));
+			struct bkey_ptrs ptrs;
+			union bch_extent_entry *entry;
+restart_drop_ptrs:
+			ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
+			bkey_for_each_ptr_decode(bkey_i_to_s(new).k, ptrs, p, entry) {
+				struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
+				struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
+				enum bch_data_type data_type = bch2_bkey_ptr_data_type(bkey_i_to_s_c(new), p, entry);
+
+				if ((p.ptr.cached &&
+				     (!g->gen_valid || gen_cmp(p.ptr.gen, g->gen) > 0)) ||
+				    (!p.ptr.cached &&
+				     gen_cmp(p.ptr.gen, g->gen) < 0) ||
+				    gen_cmp(g->gen, p.ptr.gen) > BUCKET_GC_GEN_MAX ||
+				    (g->data_type &&
+				     g->data_type != data_type)) {
+					bch2_bkey_drop_ptr(bkey_i_to_s(new), &entry->ptr);
+					goto restart_drop_ptrs;
+				}
+			}
 again:
 			ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
 			bkey_extent_entry_for_each(ptrs, entry) {
@@ -774,12 +695,6 @@ found:
 			}
 		}
 
-		ret = bch2_journal_key_insert_take(c, btree_id, level, new);
-		if (ret) {
-			kfree(new);
-			goto err;
-		}
-
 		if (level)
 			bch2_btree_node_update_key_early(trans, btree_id, level - 1, *k, new);
 
@@ -793,6 +708,12 @@ found:
 			bch_info(c, "new key %s", buf.buf);
 		}
 
+		ret = bch2_journal_key_insert_take(c, btree_id, level, new);
+		if (ret) {
+			kfree(new);
+			goto err;
+		}
+
 		*k = bkey_i_to_s_c(new);
 	}
 err:
@@ -819,10 +740,6 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 		BUG_ON(bch2_journal_seq_verify &&
 		       k->k->version.lo > atomic64_read(&c->journal.seq));
 
-		ret = bch2_check_fix_ptrs(trans, btree_id, level, is_root, k);
-		if (ret)
-			goto err;
-
 		if (fsck_err_on(k->k->version.lo > atomic64_read(&c->key_version), c,
 				bkey_version_in_future,
 				"key version number higher than recorded: %llu > %llu",
@@ -831,8 +748,13 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 			atomic64_set(&c->key_version, k->k->version.lo);
 	}
 
+	ret = bch2_check_fix_ptrs(trans, btree_id, level, is_root, k);
+	if (ret)
+		goto err;
+
 	ret = commit_do(trans, NULL, NULL, 0,
-			bch2_key_trigger(trans, btree_id, level, old, unsafe_bkey_s_c_to_s(*k), BTREE_TRIGGER_GC));
+			bch2_key_trigger(trans, btree_id, level, old,
+					 unsafe_bkey_s_c_to_s(*k), BTREE_TRIGGER_GC));
 fsck_err:
 err:
 	bch_err_fn(c, ret);
@@ -841,42 +763,30 @@ err:
 
 static int btree_gc_mark_node(struct btree_trans *trans, struct btree *b, bool initial)
 {
-	struct bch_fs *c = trans->c;
 	struct btree_node_iter iter;
 	struct bkey unpacked;
 	struct bkey_s_c k;
-	struct bkey_buf prev, cur;
 	int ret = 0;
 
+	ret = bch2_btree_node_check_topology(trans, b);
+	if (ret)
+		return ret;
+
 	if (!btree_node_type_needs_gc(btree_node_type(b)))
 		return 0;
 
 	bch2_btree_node_iter_init_from_start(&iter, b);
-	bch2_bkey_buf_init(&prev);
-	bch2_bkey_buf_init(&cur);
-	bkey_init(&prev.k->k);
 
 	while ((k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked)).k) {
 		ret = bch2_gc_mark_key(trans, b->c.btree_id, b->c.level, false,
 				       &k, initial);
 		if (ret)
-			break;
+			return ret;
 
 		bch2_btree_node_iter_advance(&iter, b);
-
-		if (b->c.level) {
-			bch2_bkey_buf_reassemble(&cur, c, k);
-
-			ret = bch2_gc_check_topology(c, b, &prev, cur,
-					bch2_btree_node_iter_end(&iter));
-			if (ret)
-				break;
-		}
 	}
 
-	bch2_bkey_buf_exit(&cur, c);
-	bch2_bkey_buf_exit(&prev, c);
-	return ret;
+	return 0;
 }
 
 static int bch2_gc_btree(struct btree_trans *trans, enum btree_id btree_id,
@@ -925,14 +835,16 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b
 	struct bch_fs *c = trans->c;
 	struct btree_and_journal_iter iter;
 	struct bkey_s_c k;
-	struct bkey_buf cur, prev;
+	struct bkey_buf cur;
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
+	ret = bch2_btree_node_check_topology(trans, b);
+	if (ret)
+		return ret;
+
 	bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b);
-	bch2_bkey_buf_init(&prev);
 	bch2_bkey_buf_init(&cur);
-	bkey_init(&prev.k->k);
 
 	while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 		BUG_ON(bpos_lt(k.k->p, b->data->min_key));
@@ -943,20 +855,7 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b
 		if (ret)
 			goto fsck_err;
 
-		if (b->c.level) {
-			bch2_bkey_buf_reassemble(&cur, c, k);
-			k = bkey_i_to_s_c(cur.k);
-
-			bch2_btree_and_journal_iter_advance(&iter);
-
-			ret = bch2_gc_check_topology(c, b,
-					&prev, cur,
-					!bch2_btree_and_journal_iter_peek(&iter).k);
-			if (ret)
-				goto fsck_err;
-		} else {
-			bch2_btree_and_journal_iter_advance(&iter);
-		}
+		bch2_btree_and_journal_iter_advance(&iter);
 	}
 
 	if (b->c.level > target_depth) {
@@ -1015,7 +914,6 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b
 	}
 fsck_err:
 	bch2_bkey_buf_exit(&cur, c);
-	bch2_bkey_buf_exit(&prev, c);
 	bch2_btree_and_journal_iter_exit(&iter);
 	printbuf_exit(&buf);
 	return ret;
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index 34df8ccc5fecc2..9c71e6fb9c41b6 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -654,6 +654,7 @@ void bch2_btree_node_drop_keys_outside_node(struct btree *b)
 	 */
 	bch2_bset_set_no_aux_tree(b, b->set);
 	bch2_btree_build_aux_trees(b);
+	b->nr = bch2_btree_node_count_keys(b);
 
 	struct bkey_s_c k;
 	struct bkey unpacked;
@@ -1657,7 +1658,7 @@ void bch2_btree_node_read(struct btree_trans *trans, struct btree *b,
 
 		prt_str(&buf, "btree node read error: no device to read from\n at ");
 		bch2_btree_pos_to_text(&buf, c, b);
-		bch_err(c, "%s", buf.buf);
+		bch_err_ratelimited(c, "%s", buf.buf);
 
 		if (c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_check_topology) &&
 		    c->curr_recovery_pass > BCH_RECOVERY_PASS_check_topology)
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index 51bcdc6c6d1cda..2a211a4bebd153 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -927,8 +927,22 @@ static __always_inline int btree_path_down(struct btree_trans *trans,
 		if (ret)
 			goto err;
 	} else {
-		bch2_bkey_buf_unpack(&tmp, c, l->b,
-				 bch2_btree_node_iter_peek(&l->iter, l->b));
+		struct bkey_packed *k = bch2_btree_node_iter_peek(&l->iter, l->b);
+		if (!k) {
+			struct printbuf buf = PRINTBUF;
+
+			prt_str(&buf, "node not found at pos ");
+			bch2_bpos_to_text(&buf, path->pos);
+			prt_str(&buf, " within parent node ");
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&l->b->key));
+
+			bch2_fs_fatal_error(c, "%s", buf.buf);
+			printbuf_exit(&buf);
+			ret = -BCH_ERR_btree_need_topology_repair;
+			goto err;
+		}
+
+		bch2_bkey_buf_unpack(&tmp, c, l->b, k);
 
 		if ((flags & BTREE_ITER_PREFETCH) &&
 		    c->opts.btree_node_prefetch) {
@@ -962,7 +976,6 @@ err:
 	return ret;
 }
 
-
 static int bch2_btree_path_traverse_all(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
@@ -2790,6 +2803,31 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 	struct btree_transaction_stats *s = btree_trans_stats(trans);
 	s->max_mem = max(s->max_mem, new_bytes);
 
+	if (trans->used_mempool) {
+		if (trans->mem_bytes >= new_bytes)
+			goto out_change_top;
+
+		/* No more space from mempool item, need malloc new one */
+		new_mem = kmalloc(new_bytes, GFP_NOWAIT|__GFP_NOWARN);
+		if (unlikely(!new_mem)) {
+			bch2_trans_unlock(trans);
+
+			new_mem = kmalloc(new_bytes, GFP_KERNEL);
+			if (!new_mem)
+				return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
+
+			ret = bch2_trans_relock(trans);
+			if (ret) {
+				kfree(new_mem);
+				return ERR_PTR(ret);
+			}
+		}
+		memcpy(new_mem, trans->mem, trans->mem_top);
+		trans->used_mempool = false;
+		mempool_free(trans->mem, &c->btree_trans_mem_pool);
+		goto out_new_mem;
+	}
+
 	new_mem = krealloc(trans->mem, new_bytes, GFP_NOWAIT|__GFP_NOWARN);
 	if (unlikely(!new_mem)) {
 		bch2_trans_unlock(trans);
@@ -2798,6 +2836,8 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 		if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) {
 			new_mem = mempool_alloc(&c->btree_trans_mem_pool, GFP_KERNEL);
 			new_bytes = BTREE_TRANS_MEM_MAX;
+			memcpy(new_mem, trans->mem, trans->mem_top);
+			trans->used_mempool = true;
 			kfree(trans->mem);
 		}
 
@@ -2811,7 +2851,7 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 		if (ret)
 			return ERR_PTR(ret);
 	}
-
+out_new_mem:
 	trans->mem = new_mem;
 	trans->mem_bytes = new_bytes;
 
@@ -2819,7 +2859,7 @@ void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 		trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes);
 		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced));
 	}
-
+out_change_top:
 	p = trans->mem + trans->mem_top;
 	trans->mem_top += size;
 	memset(p, 0, size);
@@ -3093,7 +3133,7 @@ void bch2_trans_put(struct btree_trans *trans)
 	if (paths_allocated != trans->_paths_allocated)
 		kvfree_rcu_mightsleep(paths_allocated);
 
-	if (trans->mem_bytes == BTREE_TRANS_MEM_MAX)
+	if (trans->used_mempool)
 		mempool_free(trans->mem, &c->btree_trans_mem_pool);
 	else
 		kfree(trans->mem);
diff --git a/fs/bcachefs/btree_journal_iter.c b/fs/bcachefs/btree_journal_iter.c
index 50e04356d72c8e..1f588264575db2 100644
--- a/fs/bcachefs/btree_journal_iter.c
+++ b/fs/bcachefs/btree_journal_iter.c
@@ -261,6 +261,22 @@ int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 	return bch2_journal_key_insert(c, id, level, &whiteout);
 }
 
+bool bch2_key_deleted_in_journal(struct btree_trans *trans, enum btree_id btree,
+				 unsigned level, struct bpos pos)
+{
+	struct journal_keys *keys = &trans->c->journal_keys;
+	size_t idx = bch2_journal_key_search(keys, btree, level, pos);
+
+	if (!trans->journal_replay_not_finished)
+		return false;
+
+	return (idx < keys->size &&
+		keys->data[idx].btree_id	== btree &&
+		keys->data[idx].level		== level &&
+		bpos_eq(keys->data[idx].k->k.p, pos) &&
+		bkey_deleted(&keys->data[idx].k->k));
+}
+
 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 				  unsigned level, struct bpos pos)
 {
@@ -363,7 +379,7 @@ static void btree_and_journal_iter_prefetch(struct btree_and_journal_iter *_iter
 
 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 {
-	struct bkey_s_c btree_k, journal_k, ret;
+	struct bkey_s_c btree_k, journal_k = bkey_s_c_null, ret;
 
 	if (iter->prefetch && iter->journal.level)
 		btree_and_journal_iter_prefetch(iter);
@@ -375,9 +391,10 @@ again:
 	       bpos_lt(btree_k.k->p, iter->pos))
 		bch2_journal_iter_advance_btree(iter);
 
-	while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
-	       bpos_lt(journal_k.k->p, iter->pos))
-		bch2_journal_iter_advance(&iter->journal);
+	if (iter->trans->journal_replay_not_finished)
+		while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
+		       bpos_lt(journal_k.k->p, iter->pos))
+			bch2_journal_iter_advance(&iter->journal);
 
 	ret = journal_k.k &&
 		(!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p))
@@ -435,7 +452,9 @@ void bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
 
 	bch2_btree_node_iter_init_from_start(&node_iter, b);
 	__bch2_btree_and_journal_iter_init_node_iter(trans, iter, b, node_iter, b->data->min_key);
-	list_add(&iter->journal.list, &trans->c->journal_iters);
+	if (trans->journal_replay_not_finished &&
+	    !test_bit(BCH_FS_may_go_rw, &trans->c->flags))
+		list_add(&iter->journal.list, &trans->c->journal_iters);
 }
 
 /* sort and dedup all keys in the journal: */
diff --git a/fs/bcachefs/btree_journal_iter.h b/fs/bcachefs/btree_journal_iter.h
index c9d19da3ea0480..09cd53091a05b4 100644
--- a/fs/bcachefs/btree_journal_iter.h
+++ b/fs/bcachefs/btree_journal_iter.h
@@ -40,8 +40,8 @@ int bch2_journal_key_insert(struct bch_fs *, enum btree_id,
 			    unsigned, struct bkey_i *);
 int bch2_journal_key_delete(struct bch_fs *, enum btree_id,
 			    unsigned, struct bpos);
-void bch2_journal_key_overwritten(struct bch_fs *, enum btree_id,
-				  unsigned, struct bpos);
+bool bch2_key_deleted_in_journal(struct btree_trans *, enum btree_id, unsigned, struct bpos);
+void bch2_journal_key_overwritten(struct bch_fs *, enum btree_id, unsigned, struct bpos);
 
 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *);
 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *);
diff --git a/fs/bcachefs/btree_trans_commit.c b/fs/bcachefs/btree_trans_commit.c
index 30d69a6d133eec..96669fede7d344 100644
--- a/fs/bcachefs/btree_trans_commit.c
+++ b/fs/bcachefs/btree_trans_commit.c
@@ -318,7 +318,7 @@ static inline void btree_insert_entry_checks(struct btree_trans *trans,
 		!(i->flags & BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) &&
 		test_bit(JOURNAL_REPLAY_DONE, &trans->c->journal.flags) &&
 		i->k->k.p.snapshot &&
-		bch2_snapshot_is_internal_node(trans->c, i->k->k.p.snapshot));
+		bch2_snapshot_is_internal_node(trans->c, i->k->k.p.snapshot) > 0);
 }
 
 static __always_inline int bch2_trans_journal_res_get(struct btree_trans *trans,
diff --git a/fs/bcachefs/btree_update.c b/fs/bcachefs/btree_update.c
index a4b40c1656a54b..8e47e260eba59b 100644
--- a/fs/bcachefs/btree_update.c
+++ b/fs/bcachefs/btree_update.c
@@ -38,6 +38,9 @@ static noinline int extent_front_merge(struct btree_trans *trans,
 	struct bkey_i *update;
 	int ret;
 
+	if (unlikely(trans->journal_replay_not_finished))
+		return 0;
+
 	update = bch2_bkey_make_mut_noupdate(trans, k);
 	ret = PTR_ERR_OR_ZERO(update);
 	if (ret)
@@ -69,6 +72,9 @@ static noinline int extent_back_merge(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	int ret;
 
+	if (unlikely(trans->journal_replay_not_finished))
+		return 0;
+
 	ret =   bch2_key_has_snapshot_overwrites(trans, iter->btree_id, insert->k.p) ?:
 		bch2_key_has_snapshot_overwrites(trans, iter->btree_id, k.k->p);
 	if (ret < 0)
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index b2f5f2e50f7e19..983bb27298cc24 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -2,6 +2,7 @@
 
 #include "bcachefs.h"
 #include "alloc_foreground.h"
+#include "bkey_buf.h"
 #include "bkey_methods.h"
 #include "btree_cache.h"
 #include "btree_gc.h"
@@ -18,6 +19,7 @@
 #include "journal.h"
 #include "journal_reclaim.h"
 #include "keylist.h"
+#include "recovery_passes.h"
 #include "replicas.h"
 #include "super-io.h"
 #include "trace.h"
@@ -44,56 +46,103 @@ static btree_path_idx_t get_unlocked_mut_path(struct btree_trans *trans,
 	return path_idx;
 }
 
-/* Debug code: */
-
 /*
  * Verify that child nodes correctly span parent node's range:
  */
-static void btree_node_interior_verify(struct bch_fs *c, struct btree *b)
+int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 {
-#ifdef CONFIG_BCACHEFS_DEBUG
-	struct bpos next_node = b->data->min_key;
-	struct btree_node_iter iter;
+	struct bch_fs *c = trans->c;
+	struct bpos node_min = b->key.k.type == KEY_TYPE_btree_ptr_v2
+		? bkey_i_to_btree_ptr_v2(&b->key)->v.min_key
+		: b->data->min_key;
+	struct btree_and_journal_iter iter;
 	struct bkey_s_c k;
-	struct bkey_s_c_btree_ptr_v2 bp;
-	struct bkey unpacked;
-	struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
+	struct printbuf buf = PRINTBUF;
+	struct bkey_buf prev;
+	int ret = 0;
 
-	BUG_ON(!b->c.level);
+	BUG_ON(b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
+	       !bpos_eq(bkey_i_to_btree_ptr_v2(&b->key)->v.min_key,
+			b->data->min_key));
 
-	if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
-		return;
+	if (!b->c.level)
+		return 0;
 
-	bch2_btree_node_iter_init_from_start(&iter, b);
+	bch2_bkey_buf_init(&prev);
+	bkey_init(&prev.k->k);
+	bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b);
 
-	while (1) {
-		k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked);
+	while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 		if (k.k->type != KEY_TYPE_btree_ptr_v2)
-			break;
-		bp = bkey_s_c_to_btree_ptr_v2(k);
+			goto out;
 
-		if (!bpos_eq(next_node, bp.v->min_key)) {
-			bch2_dump_btree_node(c, b);
-			bch2_bpos_to_text(&buf1, next_node);
-			bch2_bpos_to_text(&buf2, bp.v->min_key);
-			panic("expected next min_key %s got %s\n", buf1.buf, buf2.buf);
-		}
+		struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 
-		bch2_btree_node_iter_advance(&iter, b);
+		struct bpos expected_min = bkey_deleted(&prev.k->k)
+			? node_min
+			: bpos_successor(prev.k->k.p);
 
-		if (bch2_btree_node_iter_end(&iter)) {
-			if (!bpos_eq(k.k->p, b->key.k.p)) {
-				bch2_dump_btree_node(c, b);
-				bch2_bpos_to_text(&buf1, b->key.k.p);
-				bch2_bpos_to_text(&buf2, k.k->p);
-				panic("expected end %s got %s\n", buf1.buf, buf2.buf);
-			}
-			break;
+		if (!bpos_eq(expected_min, bp.v->min_key)) {
+			bch2_topology_error(c);
+
+			printbuf_reset(&buf);
+			prt_str(&buf, "end of prev node doesn't match start of next node\n"),
+			prt_printf(&buf, "  in btree %s level %u node ",
+				   bch2_btree_id_str(b->c.btree_id), b->c.level);
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+			prt_str(&buf, "\n  prev ");
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
+			prt_str(&buf, "\n  next ");
+			bch2_bkey_val_to_text(&buf, c, k);
+
+			need_fsck_err(c, btree_node_topology_bad_min_key, "%s", buf.buf);
+			goto topology_repair;
 		}
 
-		next_node = bpos_successor(k.k->p);
+		bch2_bkey_buf_reassemble(&prev, c, k);
+		bch2_btree_and_journal_iter_advance(&iter);
 	}
-#endif
+
+	if (bkey_deleted(&prev.k->k)) {
+		bch2_topology_error(c);
+
+		printbuf_reset(&buf);
+		prt_str(&buf, "empty interior node\n");
+		prt_printf(&buf, "  in btree %s level %u node ",
+			   bch2_btree_id_str(b->c.btree_id), b->c.level);
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+
+		need_fsck_err(c, btree_node_topology_empty_interior_node, "%s", buf.buf);
+		goto topology_repair;
+	} else if (!bpos_eq(prev.k->k.p, b->key.k.p)) {
+		bch2_topology_error(c);
+
+		printbuf_reset(&buf);
+		prt_str(&buf, "last child node doesn't end at end of parent node\n");
+		prt_printf(&buf, "  in btree %s level %u node ",
+			   bch2_btree_id_str(b->c.btree_id), b->c.level);
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+		prt_str(&buf, "\n  last key ");
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
+
+		need_fsck_err(c, btree_node_topology_bad_max_key, "%s", buf.buf);
+		goto topology_repair;
+	}
+out:
+fsck_err:
+	bch2_btree_and_journal_iter_exit(&iter);
+	bch2_bkey_buf_exit(&prev, c);
+	printbuf_exit(&buf);
+	return ret;
+topology_repair:
+	if ((c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_check_topology)) &&
+	    c->curr_recovery_pass > BCH_RECOVERY_PASS_check_topology) {
+		bch2_inconsistent_error(c);
+		ret = -BCH_ERR_btree_need_topology_repair;
+	} else {
+		ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology);
+	}
+	goto out;
 }
 
 /* Calculate ideal packed bkey format for new btree nodes: */
@@ -1380,9 +1429,16 @@ static void __btree_split_node(struct btree_update *as,
 		if (bkey_deleted(k))
 			continue;
 
+		uk = bkey_unpack_key(b, k);
+
+		if (b->c.level &&
+		    u64s < n1_u64s &&
+		    u64s + k->u64s >= n1_u64s &&
+		    bch2_key_deleted_in_journal(trans, b->c.btree_id, b->c.level, uk.p))
+			n1_u64s += k->u64s;
+
 		i = u64s >= n1_u64s;
 		u64s += k->u64s;
-		uk = bkey_unpack_key(b, k);
 		if (!i)
 			n1_pos = uk.p;
 		bch2_bkey_format_add_key(&format[i], &uk);
@@ -1441,8 +1497,7 @@ static void __btree_split_node(struct btree_update *as,
 
 		bch2_verify_btree_nr_keys(n[i]);
 
-		if (b->c.level)
-			btree_node_interior_verify(as->c, n[i]);
+		BUG_ON(bch2_btree_node_check_topology(trans, n[i]));
 	}
 }
 
@@ -1473,7 +1528,7 @@ static void btree_split_insert_keys(struct btree_update *as,
 
 		__bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys);
 
-		btree_node_interior_verify(as->c, b);
+		BUG_ON(bch2_btree_node_check_topology(trans, b));
 	}
 }
 
@@ -1488,9 +1543,14 @@ static int btree_split(struct btree_update *as, struct btree_trans *trans,
 	u64 start_time = local_clock();
 	int ret = 0;
 
+	bch2_verify_btree_nr_keys(b);
 	BUG_ON(!parent && (b != btree_node_root(c, b)));
 	BUG_ON(parent && !btree_node_intent_locked(trans->paths + path, b->c.level + 1));
 
+	ret = bch2_btree_node_check_topology(trans, b);
+	if (ret)
+		return ret;
+
 	bch2_btree_interior_update_will_free_node(as, b);
 
 	if (b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c)) {
@@ -1710,7 +1770,11 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
 		goto split;
 	}
 
-	btree_node_interior_verify(c, b);
+	ret = bch2_btree_node_check_topology(trans, b);
+	if (ret) {
+		bch2_btree_node_unlock_write(trans, path, b);
+		return ret;
+	}
 
 	bch2_btree_insert_keys_interior(as, trans, path, b, keys);
 
@@ -1728,7 +1792,7 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
 
 	bch2_btree_node_unlock_write(trans, path, b);
 
-	btree_node_interior_verify(c, b);
+	BUG_ON(bch2_btree_node_check_topology(trans, b));
 	return 0;
 split:
 	/*
@@ -1818,9 +1882,12 @@ int bch2_btree_increase_depth(struct btree_trans *trans, btree_path_idx_t path,
 {
 	struct bch_fs *c = trans->c;
 	struct btree *b = bch2_btree_id_root(c, trans->paths[path].btree_id)->b;
+
+	if (btree_node_fake(b))
+		return bch2_btree_split_leaf(trans, path, flags);
+
 	struct btree_update *as =
-		bch2_btree_update_start(trans, trans->paths + path,
-					b->c.level, true, flags);
+		bch2_btree_update_start(trans, trans->paths + path, b->c.level, true, flags);
 	if (IS_ERR(as))
 		return PTR_ERR(as);
 
diff --git a/fs/bcachefs/btree_update_interior.h b/fs/bcachefs/btree_update_interior.h
index f651dd48aaa049..7ac3e17b84fde4 100644
--- a/fs/bcachefs/btree_update_interior.h
+++ b/fs/bcachefs/btree_update_interior.h
@@ -10,6 +10,8 @@
 
 #define BTREE_UPDATE_JOURNAL_RES	(BTREE_UPDATE_NODES_MAX * (BKEY_BTREE_PTR_U64s_MAX + 1))
 
+int bch2_btree_node_check_topology(struct btree_trans *, struct btree *);
+
 /*
  * Tracks an in progress split/rewrite of a btree node and the update to the
  * parent node:
diff --git a/fs/bcachefs/btree_write_buffer.c b/fs/bcachefs/btree_write_buffer.c
index 5cbad8445782c4..baf63e2fddb64c 100644
--- a/fs/bcachefs/btree_write_buffer.c
+++ b/fs/bcachefs/btree_write_buffer.c
@@ -11,6 +11,7 @@
 #include "journal_reclaim.h"
 
 #include <linux/prefetch.h>
+#include <linux/sort.h>
 
 static int bch2_btree_write_buffer_journal_flush(struct journal *,
 				struct journal_entry_pin *, u64);
@@ -46,6 +47,14 @@ static inline bool wb_key_ref_cmp(const struct wb_key_ref *l, const struct wb_ke
 #endif
 }
 
+static int wb_key_seq_cmp(const void *_l, const void *_r)
+{
+	const struct btree_write_buffered_key *l = _l;
+	const struct btree_write_buffered_key *r = _r;
+
+	return cmp_int(l->journal_seq, r->journal_seq);
+}
+
 /* Compare excluding idx, the low 24 bits: */
 static inline bool wb_key_eq(const void *_l, const void *_r)
 {
@@ -357,6 +366,11 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 		 */
 		trace_and_count(c, write_buffer_flush_slowpath, trans, slowpath, wb->flushing.keys.nr);
 
+		sort(wb->flushing.keys.data,
+		     wb->flushing.keys.nr,
+		     sizeof(wb->flushing.keys.data[0]),
+		     wb_key_seq_cmp, NULL);
+
 		darray_for_each(wb->flushing.keys, i) {
 			if (!i->journal_seq)
 				continue;
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index 96edf2c34d433d..941401a210f569 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -525,6 +525,7 @@ int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
 			"different types of data in same bucket: %s, %s",
 			bch2_data_type_str(g->data_type),
 			bch2_data_type_str(data_type))) {
+		BUG();
 		ret = -EIO;
 		goto err;
 	}
@@ -628,6 +629,7 @@ int bch2_check_bucket_ref(struct btree_trans *trans,
 			bch2_data_type_str(ptr_data_type),
 			(printbuf_reset(&buf),
 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
+		BUG();
 		ret = -EIO;
 		goto err;
 	}
@@ -815,14 +817,14 @@ static int __mark_pointer(struct btree_trans *trans,
 static int bch2_trigger_pointer(struct btree_trans *trans,
 			enum btree_id btree_id, unsigned level,
 			struct bkey_s_c k, struct extent_ptr_decoded p,
-			s64 *sectors,
-			unsigned flags)
+			const union bch_extent_entry *entry,
+			s64 *sectors, unsigned flags)
 {
 	bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
 	struct bpos bucket;
 	struct bch_backpointer bp;
 
-	bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
+	bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, entry, &bucket, &bp);
 	*sectors = insert ? bp.bucket_len : -((s64) bp.bucket_len);
 
 	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
@@ -851,7 +853,7 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 	if (flags & BTREE_TRIGGER_GC) {
 		struct bch_fs *c = trans->c;
 		struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
-		enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
+		enum bch_data_type data_type = bch2_bkey_ptr_data_type(k, p, entry);
 
 		percpu_down_read(&c->mark_lock);
 		struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
@@ -979,7 +981,7 @@ static int __trigger_extent(struct btree_trans *trans,
 
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 		s64 disk_sectors;
-		ret = bch2_trigger_pointer(trans, btree_id, level, k, p, &disk_sectors, flags);
+		ret = bch2_trigger_pointer(trans, btree_id, level, k, p, entry, &disk_sectors, flags);
 		if (ret < 0)
 			return ret;
 
diff --git a/fs/bcachefs/chardev.c b/fs/bcachefs/chardev.c
index 38defa19d52d70..cbfa6459bdbcee 100644
--- a/fs/bcachefs/chardev.c
+++ b/fs/bcachefs/chardev.c
@@ -7,7 +7,7 @@
 #include "chardev.h"
 #include "journal.h"
 #include "move.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "replicas.h"
 #include "super.h"
 #include "super-io.h"
diff --git a/fs/bcachefs/data_update.c b/fs/bcachefs/data_update.c
index 4150feca42a2e6..b564404dad711b 100644
--- a/fs/bcachefs/data_update.c
+++ b/fs/bcachefs/data_update.c
@@ -14,6 +14,7 @@
 #include "move.h"
 #include "nocow_locking.h"
 #include "rebalance.h"
+#include "snapshot.h"
 #include "subvolume.h"
 #include "trace.h"
 
@@ -509,6 +510,14 @@ int bch2_data_update_init(struct btree_trans *trans,
 	unsigned ptrs_locked = 0;
 	int ret = 0;
 
+	/*
+	 * fs is corrupt  we have a key for a snapshot node that doesn't exist,
+	 * and we have to check for this because we go rw before repairing the
+	 * snapshots table - just skip it, we can move it later.
+	 */
+	if (unlikely(k.k->p.snapshot && !bch2_snapshot_equiv(c, k.k->p.snapshot)))
+		return -BCH_ERR_data_update_done;
+
 	bch2_bkey_buf_init(&m->k);
 	bch2_bkey_buf_reassemble(&m->k, c, k);
 	m->btree_id	= btree_id;
diff --git a/fs/bcachefs/errcode.h b/fs/bcachefs/errcode.h
index af25d8ec60f221..01a79fa3eacb21 100644
--- a/fs/bcachefs/errcode.h
+++ b/fs/bcachefs/errcode.h
@@ -252,7 +252,8 @@
 	x(BCH_ERR_nopromote,		nopromote_in_flight)			\
 	x(BCH_ERR_nopromote,		nopromote_no_writes)			\
 	x(BCH_ERR_nopromote,		nopromote_enomem)			\
-	x(0,				need_inode_lock)
+	x(0,				need_inode_lock)			\
+	x(0,				invalid_snapshot_node)
 
 enum bch_errcode {
 	BCH_ERR_START		= 2048,
diff --git a/fs/bcachefs/error.c b/fs/bcachefs/error.c
index 043431206799d8..82a6656c941c5f 100644
--- a/fs/bcachefs/error.c
+++ b/fs/bcachefs/error.c
@@ -1,7 +1,8 @@
 // SPDX-License-Identifier: GPL-2.0
 #include "bcachefs.h"
 #include "error.h"
-#include "recovery.h"
+#include "journal.h"
+#include "recovery_passes.h"
 #include "super.h"
 #include "thread_with_file.h"
 
@@ -16,7 +17,8 @@ bool bch2_inconsistent_error(struct bch_fs *c)
 		return false;
 	case BCH_ON_ERROR_ro:
 		if (bch2_fs_emergency_read_only(c))
-			bch_err(c, "inconsistency detected - emergency read only");
+			bch_err(c, "inconsistency detected - emergency read only at journal seq %llu",
+				journal_cur_seq(&c->journal));
 		return true;
 	case BCH_ON_ERROR_panic:
 		panic(bch2_fmt(c, "panic after error"));
diff --git a/fs/bcachefs/error.h b/fs/bcachefs/error.h
index ae1d6674c512d4..36caedf72d89ab 100644
--- a/fs/bcachefs/error.h
+++ b/fs/bcachefs/error.h
@@ -32,6 +32,12 @@ bool bch2_inconsistent_error(struct bch_fs *);
 
 int bch2_topology_error(struct bch_fs *);
 
+#define bch2_fs_topology_error(c, ...)					\
+({									\
+	bch_err(c, "btree topology error: " __VA_ARGS__);		\
+	bch2_topology_error(c);						\
+})
+
 #define bch2_fs_inconsistent(c, ...)					\
 ({									\
 	bch_err(c, __VA_ARGS__);					\
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 61395b113df9bd..b2432d88cda64c 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -189,13 +189,18 @@ int bch2_btree_ptr_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
 			      enum bkey_invalid_flags flags,
 			      struct printbuf *err)
 {
+	struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 	int ret = 0;
 
-	bkey_fsck_err_on(bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX, c, err,
-			 btree_ptr_v2_val_too_big,
+	bkey_fsck_err_on(bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX,
+			 c, err, btree_ptr_v2_val_too_big,
 			 "value too big (%zu > %zu)",
 			 bkey_val_u64s(k.k), BKEY_BTREE_PTR_VAL_U64s_MAX);
 
+	bkey_fsck_err_on(bpos_ge(bp.v->min_key, bp.k->p),
+			 c, err, btree_ptr_v2_min_key_bad,
+			 "min_key > key");
+
 	ret = bch2_bkey_ptrs_invalid(c, k, flags, err);
 fsck_err:
 	return ret;
diff --git a/fs/bcachefs/extents.h b/fs/bcachefs/extents.h
index fd2669cdd76f3b..3fd0169b98c182 100644
--- a/fs/bcachefs/extents.h
+++ b/fs/bcachefs/extents.h
@@ -596,30 +596,6 @@ static inline struct bch_devs_list bch2_bkey_cached_devs(struct bkey_s_c k)
 	return ret;
 }
 
-static inline unsigned bch2_bkey_ptr_data_type(struct bkey_s_c k, const struct bch_extent_ptr *ptr)
-{
-	switch (k.k->type) {
-	case KEY_TYPE_btree_ptr:
-	case KEY_TYPE_btree_ptr_v2:
-		return BCH_DATA_btree;
-	case KEY_TYPE_extent:
-	case KEY_TYPE_reflink_v:
-		return BCH_DATA_user;
-	case KEY_TYPE_stripe: {
-		struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
-
-		BUG_ON(ptr < s.v->ptrs ||
-		       ptr >= s.v->ptrs + s.v->nr_blocks);
-
-		return ptr >= s.v->ptrs + s.v->nr_blocks - s.v->nr_redundant
-			? BCH_DATA_parity
-			: BCH_DATA_user;
-	}
-	default:
-		BUG();
-	}
-}
-
 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c);
 unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c);
 unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c);
diff --git a/fs/bcachefs/fs-io-direct.c b/fs/bcachefs/fs-io-direct.c
index 33cb6da3a5ad28..f49e6c0f0f6835 100644
--- a/fs/bcachefs/fs-io-direct.c
+++ b/fs/bcachefs/fs-io-direct.c
@@ -536,7 +536,7 @@ static __always_inline long bch2_dio_write_loop(struct dio_write *dio)
 		if (likely(!dio->iter.count) || dio->op.error)
 			break;
 
-		bio_reset(bio, NULL, REQ_OP_WRITE);
+		bio_reset(bio, NULL, REQ_OP_WRITE | REQ_SYNC | REQ_IDLE);
 	}
 out:
 	return bch2_dio_write_done(dio);
@@ -618,7 +618,7 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter)
 
 	bio = bio_alloc_bioset(NULL,
 			       bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS),
-			       REQ_OP_WRITE,
+			       REQ_OP_WRITE | REQ_SYNC | REQ_IDLE,
 			       GFP_KERNEL,
 			       &c->dio_write_bioset);
 	dio = container_of(bio, struct dio_write, op.wbio.bio);
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 0ccee05f6887b3..b5ea9fa1259d14 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -1997,6 +1997,7 @@ out:
 	return dget(sb->s_root);
 
 err_put_super:
+	__bch2_fs_stop(c);
 	deactivate_locked_super(sb);
 	return ERR_PTR(bch2_err_class(ret));
 }
diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index 47d4eefaba7ba0..cbb8b43e419fec 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -12,7 +12,7 @@
 #include "fsck.h"
 #include "inode.h"
 #include "keylist.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "snapshot.h"
 #include "super.h"
 #include "xattr.h"
@@ -158,9 +158,10 @@ static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 
-	ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
-				  &dir_hash_info, &iter,
-				  BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+	ret =   bch2_btree_iter_traverse(&iter) ?:
+		bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
+				    &dir_hash_info, &iter,
+				    BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 	bch2_trans_iter_exit(trans, &iter);
 err:
 	bch_err_fn(c, ret);
@@ -1371,10 +1372,6 @@ static int check_overlapping_extents(struct btree_trans *trans,
 			goto err;
 	}
 
-	ret = extent_ends_at(c, extent_ends, seen, k);
-	if (ret)
-		goto err;
-
 	extent_ends->last_pos = k.k->p;
 err:
 	return ret;
@@ -1504,6 +1501,12 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
 
 		i->seen_this_pos = true;
 	}
+
+	if (k.k->type != KEY_TYPE_whiteout) {
+		ret = extent_ends_at(c, extent_ends, s, k);
+		if (ret)
+			goto err;
+	}
 out:
 err:
 fsck_err:
@@ -2098,17 +2101,21 @@ static int check_root_trans(struct btree_trans *trans)
 
 	if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
 				"root subvol missing")) {
-		struct bkey_i_subvolume root_subvol;
+		struct bkey_i_subvolume *root_subvol =
+			bch2_trans_kmalloc(trans, sizeof(*root_subvol));
+		ret = PTR_ERR_OR_ZERO(root_subvol);
+		if (ret)
+			goto err;
 
 		snapshot	= U32_MAX;
 		inum		= BCACHEFS_ROOT_INO;
 
-		bkey_subvolume_init(&root_subvol.k_i);
-		root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
-		root_subvol.v.flags	= 0;
-		root_subvol.v.snapshot	= cpu_to_le32(snapshot);
-		root_subvol.v.inode	= cpu_to_le64(inum);
-		ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol.k_i, 0);
+		bkey_subvolume_init(&root_subvol->k_i);
+		root_subvol->k.p.offset = BCACHEFS_ROOT_SUBVOL;
+		root_subvol->v.flags	= 0;
+		root_subvol->v.snapshot	= cpu_to_le32(snapshot);
+		root_subvol->v.inode	= cpu_to_le64(inum);
+		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;
diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index 2b5e06770ab39e..ca4a066e9a5428 100644
--- a/fs/bcachefs/inode.c
+++ b/fs/bcachefs/inode.c
@@ -552,8 +552,8 @@ static void __bch2_inode_unpacked_to_text(struct printbuf *out,
 	prt_printf(out, "bi_sectors=%llu", inode->bi_sectors);
 	prt_newline(out);
 
-	prt_newline(out);
 	prt_printf(out, "bi_version=%llu", inode->bi_version);
+	prt_newline(out);
 
 #define x(_name, _bits)						\
 	prt_printf(out, #_name "=%llu", (u64) inode->_name);	\
diff --git a/fs/bcachefs/io_misc.c b/fs/bcachefs/io_misc.c
index 1baf78594ccaf8..82f9170dab3fdc 100644
--- a/fs/bcachefs/io_misc.c
+++ b/fs/bcachefs/io_misc.c
@@ -264,6 +264,7 @@ static int __bch2_resume_logged_op_truncate(struct btree_trans *trans,
 		ret = 0;
 err:
 	bch2_logged_op_finish(trans, op_k);
+	bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -476,6 +477,7 @@ case LOGGED_OP_FINSERT_finish:
 	break;
 	}
 err:
+	bch_err_fn(c, ret);
 	bch2_logged_op_finish(trans, op_k);
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
diff --git a/fs/bcachefs/logged_ops.c b/fs/bcachefs/logged_ops.c
index 9fac838d123e8e..b82f8209041ffb 100644
--- a/fs/bcachefs/logged_ops.c
+++ b/fs/bcachefs/logged_ops.c
@@ -37,7 +37,6 @@ static int resume_logged_op(struct btree_trans *trans, struct btree_iter *iter,
 	const struct bch_logged_op_fn *fn = logged_op_fn(k.k->type);
 	struct bkey_buf sk;
 	u32 restart_count = trans->restart_count;
-	int ret;
 
 	if (!fn)
 		return 0;
@@ -45,11 +44,11 @@ static int resume_logged_op(struct btree_trans *trans, struct btree_iter *iter,
 	bch2_bkey_buf_init(&sk);
 	bch2_bkey_buf_reassemble(&sk, c, k);
 
-	ret =   drop_locks_do(trans, (bch2_fs_lazy_rw(c), 0)) ?:
-		fn->resume(trans, sk.k) ?: trans_was_restarted(trans, restart_count);
+	fn->resume(trans, sk.k);
 
 	bch2_bkey_buf_exit(&sk, c);
-	return ret;
+
+	return trans_was_restarted(trans, restart_count);
 }
 
 int bch2_resume_logged_ops(struct bch_fs *c)
diff --git a/fs/bcachefs/opts.c b/fs/bcachefs/opts.c
index 08ea0cfc4aef08..e1800c4119b5fb 100644
--- a/fs/bcachefs/opts.c
+++ b/fs/bcachefs/opts.c
@@ -7,6 +7,7 @@
 #include "disk_groups.h"
 #include "error.h"
 #include "opts.h"
+#include "recovery_passes.h"
 #include "super-io.h"
 #include "util.h"
 
@@ -205,6 +206,9 @@ const struct bch_option bch2_opt_table[] = {
 #define OPT_STR(_choices)	.type = BCH_OPT_STR,			\
 				.min = 0, .max = ARRAY_SIZE(_choices),	\
 				.choices = _choices
+#define OPT_STR_NOLIMIT(_choices)	.type = BCH_OPT_STR,		\
+				.min = 0, .max = U64_MAX,		\
+				.choices = _choices
 #define OPT_FN(_fn)		.type = BCH_OPT_FN, .fn	= _fn
 
 #define x(_name, _bits, _flags, _type, _sb_opt, _default, _hint, _help)	\
diff --git a/fs/bcachefs/opts.h b/fs/bcachefs/opts.h
index 136083c11f3a3a..084247c13bd8fd 100644
--- a/fs/bcachefs/opts.h
+++ b/fs/bcachefs/opts.h
@@ -362,7 +362,12 @@ enum fsck_err_opts {
 	  OPT_FS|OPT_MOUNT,						\
 	  OPT_BOOL(),							\
 	  BCH2_NO_SB_OPT,		false,				\
-	  NULL,		"Don't replay the journal")			\
+	  NULL,		"Exit recovery immediately prior to journal replay")\
+	x(recovery_pass_last,		u8,				\
+	  OPT_FS|OPT_MOUNT,						\
+	  OPT_STR_NOLIMIT(bch2_recovery_passes),			\
+	  BCH2_NO_SB_OPT,		0,				\
+	  NULL,		"Exit recovery after specified pass")		\
 	x(keep_journal,			u8,				\
 	  0,								\
 	  OPT_BOOL(),							\
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index 03f9d6afe46788..e8b43400929362 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -1,35 +1,30 @@
 // SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
-#include "backpointers.h"
-#include "bkey_buf.h"
 #include "alloc_background.h"
-#include "btree_gc.h"
+#include "bkey_buf.h"
 #include "btree_journal_iter.h"
 #include "btree_update.h"
 #include "btree_update_interior.h"
 #include "btree_io.h"
 #include "buckets.h"
 #include "dirent.h"
-#include "ec.h"
 #include "errcode.h"
 #include "error.h"
 #include "fs-common.h"
-#include "fsck.h"
 #include "journal_io.h"
 #include "journal_reclaim.h"
 #include "journal_seq_blacklist.h"
-#include "lru.h"
 #include "logged_ops.h"
 #include "move.h"
 #include "quota.h"
 #include "rebalance.h"
 #include "recovery.h"
+#include "recovery_passes.h"
 #include "replicas.h"
 #include "sb-clean.h"
 #include "sb-downgrade.h"
 #include "snapshot.h"
-#include "subvolume.h"
 #include "super-io.h"
 
 #include <linux/sort.h>
@@ -186,7 +181,7 @@ static int journal_sort_seq_cmp(const void *_l, const void *_r)
 	return cmp_int(l->journal_seq, r->journal_seq);
 }
 
-static int bch2_journal_replay(struct bch_fs *c)
+int bch2_journal_replay(struct bch_fs *c)
 {
 	struct journal_keys *keys = &c->journal_keys;
 	DARRAY(struct journal_key *) keys_sorted = { 0 };
@@ -194,6 +189,7 @@ static int bch2_journal_replay(struct bch_fs *c)
 	u64 start_seq	= c->journal_replay_seq_start;
 	u64 end_seq	= c->journal_replay_seq_start;
 	struct btree_trans *trans = bch2_trans_get(c);
+	bool immediate_flush = false;
 	int ret = 0;
 
 	if (keys->nr) {
@@ -215,6 +211,13 @@ static int bch2_journal_replay(struct bch_fs *c)
 	darray_for_each(*keys, k) {
 		cond_resched();
 
+		/*
+		 * k->allocated means the key wasn't read in from the journal,
+		 * rather it was from early repair code
+		 */
+		if (k->allocated)
+			immediate_flush = true;
+
 		/* Skip fastpath if we're low on space in the journal */
 		ret = c->journal.watermark ? -1 :
 			commit_do(trans, NULL, NULL,
@@ -266,7 +269,8 @@ static int bch2_journal_replay(struct bch_fs *c)
 	bch2_trans_put(trans);
 	trans = NULL;
 
-	if (!c->opts.keep_journal)
+	if (!c->opts.keep_journal &&
+	    c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay)
 		bch2_journal_keys_put_initial(c);
 
 	replay_now_at(j, j->replay_journal_seq_end);
@@ -274,6 +278,12 @@ static int bch2_journal_replay(struct bch_fs *c)
 
 	bch2_journal_set_replay_done(j);
 
+	/* if we did any repair, flush it immediately */
+	if (immediate_flush) {
+		bch2_journal_flush_all_pins(&c->journal);
+		ret = bch2_journal_meta(&c->journal);
+	}
+
 	if (keys->nr)
 		bch2_journal_log_msg(c, "journal replay finished");
 err:
@@ -471,150 +481,6 @@ fsck_err:
 	return ret;
 }
 
-static int bch2_initialize_subvolumes(struct bch_fs *c)
-{
-	struct bkey_i_snapshot_tree	root_tree;
-	struct bkey_i_snapshot		root_snapshot;
-	struct bkey_i_subvolume		root_volume;
-	int ret;
-
-	bkey_snapshot_tree_init(&root_tree.k_i);
-	root_tree.k.p.offset		= 1;
-	root_tree.v.master_subvol	= cpu_to_le32(1);
-	root_tree.v.root_snapshot	= cpu_to_le32(U32_MAX);
-
-	bkey_snapshot_init(&root_snapshot.k_i);
-	root_snapshot.k.p.offset = U32_MAX;
-	root_snapshot.v.flags	= 0;
-	root_snapshot.v.parent	= 0;
-	root_snapshot.v.subvol	= cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
-	root_snapshot.v.tree	= cpu_to_le32(1);
-	SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
-
-	bkey_subvolume_init(&root_volume.k_i);
-	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
-	root_volume.v.flags	= 0;
-	root_volume.v.snapshot	= cpu_to_le32(U32_MAX);
-	root_volume.v.inode	= cpu_to_le64(BCACHEFS_ROOT_INO);
-
-	ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,	&root_tree.k_i, NULL, 0) ?:
-		bch2_btree_insert(c, BTREE_ID_snapshots,	&root_snapshot.k_i, NULL, 0) ?:
-		bch2_btree_insert(c, BTREE_ID_subvolumes,	&root_volume.k_i, NULL, 0);
-	bch_err_fn(c, ret);
-	return ret;
-}
-
-static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
-{
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	struct bch_inode_unpacked inode;
-	int ret;
-
-	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
-			       SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
-	ret = bkey_err(k);
-	if (ret)
-		return ret;
-
-	if (!bkey_is_inode(k.k)) {
-		bch_err(trans->c, "root inode not found");
-		ret = -BCH_ERR_ENOENT_inode;
-		goto err;
-	}
-
-	ret = bch2_inode_unpack(k, &inode);
-	BUG_ON(ret);
-
-	inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
-
-	ret = bch2_inode_write(trans, &iter, &inode);
-err:
-	bch2_trans_iter_exit(trans, &iter);
-	return ret;
-}
-
-/* set bi_subvol on root inode */
-noinline_for_stack
-static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
-{
-	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
-				__bch2_fs_upgrade_for_subvolumes(trans));
-	bch_err_fn(c, ret);
-	return ret;
-}
-
-const char * const bch2_recovery_passes[] = {
-#define x(_fn, ...)	#_fn,
-	BCH_RECOVERY_PASSES()
-#undef x
-	NULL
-};
-
-static int bch2_check_allocations(struct bch_fs *c)
-{
-	return bch2_gc(c, true, c->opts.norecovery);
-}
-
-static int bch2_set_may_go_rw(struct bch_fs *c)
-{
-	struct journal_keys *keys = &c->journal_keys;
-
-	/*
-	 * After we go RW, the journal keys buffer can't be modified (except for
-	 * setting journal_key->overwritten: it will be accessed by multiple
-	 * threads
-	 */
-	move_gap(keys, keys->nr);
-
-	set_bit(BCH_FS_may_go_rw, &c->flags);
-
-	if (keys->nr || c->opts.fsck || !c->sb.clean)
-		return bch2_fs_read_write_early(c);
-	return 0;
-}
-
-struct recovery_pass_fn {
-	int		(*fn)(struct bch_fs *);
-	unsigned	when;
-};
-
-static struct recovery_pass_fn recovery_pass_fns[] = {
-#define x(_fn, _id, _when)	{ .fn = bch2_##_fn, .when = _when },
-	BCH_RECOVERY_PASSES()
-#undef x
-};
-
-u64 bch2_recovery_passes_to_stable(u64 v)
-{
-	static const u8 map[] = {
-#define x(n, id, ...)	[BCH_RECOVERY_PASS_##n] = BCH_RECOVERY_PASS_STABLE_##n,
-	BCH_RECOVERY_PASSES()
-#undef x
-	};
-
-	u64 ret = 0;
-	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
-		if (v & BIT_ULL(i))
-			ret |= BIT_ULL(map[i]);
-	return ret;
-}
-
-u64 bch2_recovery_passes_from_stable(u64 v)
-{
-	static const u8 map[] = {
-#define x(n, id, ...)	[BCH_RECOVERY_PASS_STABLE_##n] = BCH_RECOVERY_PASS_##n,
-	BCH_RECOVERY_PASSES()
-#undef x
-	};
-
-	u64 ret = 0;
-	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
-		if (v & BIT_ULL(i))
-			ret |= BIT_ULL(map[i]);
-	return ret;
-}
-
 static bool check_version_upgrade(struct bch_fs *c)
 {
 	unsigned latest_version	= bcachefs_metadata_version_current;
@@ -687,96 +553,6 @@ static bool check_version_upgrade(struct bch_fs *c)
 	return false;
 }
 
-u64 bch2_fsck_recovery_passes(void)
-{
-	u64 ret = 0;
-
-	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
-		if (recovery_pass_fns[i].when & PASS_FSCK)
-			ret |= BIT_ULL(i);
-	return ret;
-}
-
-static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
-{
-	struct recovery_pass_fn *p = recovery_pass_fns + pass;
-
-	if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
-		return false;
-	if (c->recovery_passes_explicit & BIT_ULL(pass))
-		return true;
-	if ((p->when & PASS_FSCK) && c->opts.fsck)
-		return true;
-	if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
-		return true;
-	if (p->when & PASS_ALWAYS)
-		return true;
-	return false;
-}
-
-static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
-{
-	struct recovery_pass_fn *p = recovery_pass_fns + pass;
-	int ret;
-
-	if (!(p->when & PASS_SILENT))
-		bch2_print(c, KERN_INFO bch2_log_msg(c, "%s..."),
-			   bch2_recovery_passes[pass]);
-	ret = p->fn(c);
-	if (ret)
-		return ret;
-	if (!(p->when & PASS_SILENT))
-		bch2_print(c, KERN_CONT " done\n");
-
-	return 0;
-}
-
-static int bch2_run_recovery_passes(struct bch_fs *c)
-{
-	int ret = 0;
-
-	while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
-		if (should_run_recovery_pass(c, c->curr_recovery_pass)) {
-			unsigned pass = c->curr_recovery_pass;
-
-			ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
-			if (bch2_err_matches(ret, BCH_ERR_restart_recovery) ||
-			    (ret && c->curr_recovery_pass < pass))
-				continue;
-			if (ret)
-				break;
-
-			c->recovery_passes_complete |= BIT_ULL(c->curr_recovery_pass);
-		}
-		c->curr_recovery_pass++;
-		c->recovery_pass_done = max(c->recovery_pass_done, c->curr_recovery_pass);
-	}
-
-	return ret;
-}
-
-int bch2_run_online_recovery_passes(struct bch_fs *c)
-{
-	int ret = 0;
-
-	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++) {
-		struct recovery_pass_fn *p = recovery_pass_fns + i;
-
-		if (!(p->when & PASS_ONLINE))
-			continue;
-
-		ret = bch2_run_recovery_pass(c, i);
-		if (bch2_err_matches(ret, BCH_ERR_restart_recovery)) {
-			i = c->curr_recovery_pass;
-			continue;
-		}
-		if (ret)
-			break;
-	}
-
-	return ret;
-}
-
 int bch2_fs_recovery(struct bch_fs *c)
 {
 	struct bch_sb_field_clean *clean = NULL;
@@ -809,24 +585,14 @@ int bch2_fs_recovery(struct bch_fs *c)
 		goto err;
 	}
 
-	if (c->opts.fsck && c->opts.norecovery) {
-		bch_err(c, "cannot select both norecovery and fsck");
-		ret = -EINVAL;
-		goto err;
-	}
+	if (c->opts.norecovery)
+		c->opts.recovery_pass_last = BCH_RECOVERY_PASS_journal_replay - 1;
 
 	if (!c->opts.nochanges) {
 		mutex_lock(&c->sb_lock);
+		struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 		bool write_sb = false;
 
-		struct bch_sb_field_ext *ext =
-			bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
-		if (!ext) {
-			ret = -BCH_ERR_ENOSPC_sb;
-			mutex_unlock(&c->sb_lock);
-			goto err;
-		}
-
 		if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)) {
 			ext->recovery_passes_required[0] |=
 				cpu_to_le64(bch2_recovery_passes_to_stable(BIT_ULL(BCH_RECOVERY_PASS_check_topology)));
@@ -1017,6 +783,12 @@ use_clean:
 
 	clear_bit(BCH_FS_fsck_running, &c->flags);
 
+	/* fsync if we fixed errors */
+	if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
+		bch2_journal_flush_all_pins(&c->journal);
+		bch2_journal_meta(&c->journal);
+	}
+
 	/* If we fixed errors, verify that fs is actually clean now: */
 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 	    test_bit(BCH_FS_errors_fixed, &c->flags) &&
@@ -1051,6 +823,7 @@ use_clean:
 	}
 
 	mutex_lock(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 	bool write_sb = false;
 
 	if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != le16_to_cpu(c->disk_sb.sb->version)) {
@@ -1064,15 +837,12 @@ use_clean:
 		write_sb = true;
 	}
 
-	if (!test_bit(BCH_FS_error, &c->flags)) {
-		struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
-		if (ext &&
-		    (!bch2_is_zero(ext->recovery_passes_required, sizeof(ext->recovery_passes_required)) ||
-		     !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent)))) {
-			memset(ext->recovery_passes_required, 0, sizeof(ext->recovery_passes_required));
-			memset(ext->errors_silent, 0, sizeof(ext->errors_silent));
-			write_sb = true;
-		}
+	if (!test_bit(BCH_FS_error, &c->flags) &&
+	    (!bch2_is_zero(ext->recovery_passes_required, sizeof(ext->recovery_passes_required)) ||
+	     !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent)))) {
+		memset(ext->recovery_passes_required, 0, sizeof(ext->recovery_passes_required));
+		memset(ext->errors_silent, 0, sizeof(ext->errors_silent));
+		write_sb = true;
 	}
 
 	if (c->opts.fsck &&
@@ -1155,7 +925,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	}
 	mutex_unlock(&c->sb_lock);
 
-	c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
+	c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
 	set_bit(BCH_FS_may_go_rw, &c->flags);
 
 	for (unsigned i = 0; i < BTREE_ID_NR; i++)
@@ -1230,7 +1000,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	if (ret)
 		goto err;
 
-	c->recovery_pass_done = ARRAY_SIZE(recovery_pass_fns) - 1;
+	c->recovery_pass_done = BCH_RECOVERY_PASS_NR - 1;
 
 	if (enabled_qtypes(c)) {
 		ret = bch2_fs_quota_read(c);
diff --git a/fs/bcachefs/recovery.h b/fs/bcachefs/recovery.h
index 4e9d24719b2e85..3962fd87b50d8f 100644
--- a/fs/bcachefs/recovery.h
+++ b/fs/bcachefs/recovery.h
@@ -2,37 +2,7 @@
 #ifndef _BCACHEFS_RECOVERY_H
 #define _BCACHEFS_RECOVERY_H
 
-extern const char * const bch2_recovery_passes[];
-
-u64 bch2_recovery_passes_to_stable(u64 v);
-u64 bch2_recovery_passes_from_stable(u64 v);
-
-/*
- * For when we need to rewind recovery passes and run a pass we skipped:
- */
-static inline int bch2_run_explicit_recovery_pass(struct bch_fs *c,
-						  enum bch_recovery_pass pass)
-{
-	if (c->recovery_passes_explicit & BIT_ULL(pass))
-		return 0;
-
-	bch_info(c, "running explicit recovery pass %s (%u), currently at %s (%u)",
-		 bch2_recovery_passes[pass], pass,
-		 bch2_recovery_passes[c->curr_recovery_pass], c->curr_recovery_pass);
-
-	c->recovery_passes_explicit |= BIT_ULL(pass);
-
-	if (c->curr_recovery_pass >= pass) {
-		c->curr_recovery_pass = pass;
-		c->recovery_passes_complete &= (1ULL << pass) >> 1;
-		return -BCH_ERR_restart_recovery;
-	} else {
-		return 0;
-	}
-}
-
-int bch2_run_online_recovery_passes(struct bch_fs *);
-u64 bch2_fsck_recovery_passes(void);
+int bch2_journal_replay(struct bch_fs *);
 
 int bch2_fs_recovery(struct bch_fs *);
 int bch2_fs_initialize(struct bch_fs *);
diff --git a/fs/bcachefs/recovery_passes.c b/fs/bcachefs/recovery_passes.c
new file mode 100644
index 00000000000000..0089d9456b1006
--- /dev/null
+++ b/fs/bcachefs/recovery_passes.c
@@ -0,0 +1,225 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "bcachefs.h"
+#include "alloc_background.h"
+#include "backpointers.h"
+#include "btree_gc.h"
+#include "ec.h"
+#include "fsck.h"
+#include "inode.h"
+#include "journal.h"
+#include "lru.h"
+#include "logged_ops.h"
+#include "rebalance.h"
+#include "recovery.h"
+#include "recovery_passes.h"
+#include "snapshot.h"
+#include "subvolume.h"
+#include "super.h"
+#include "super-io.h"
+
+const char * const bch2_recovery_passes[] = {
+#define x(_fn, ...)	#_fn,
+	BCH_RECOVERY_PASSES()
+#undef x
+	NULL
+};
+
+static int bch2_check_allocations(struct bch_fs *c)
+{
+	return bch2_gc(c, true, false);
+}
+
+static int bch2_set_may_go_rw(struct bch_fs *c)
+{
+	struct journal_keys *keys = &c->journal_keys;
+
+	/*
+	 * After we go RW, the journal keys buffer can't be modified (except for
+	 * setting journal_key->overwritten: it will be accessed by multiple
+	 * threads
+	 */
+	move_gap(keys, keys->nr);
+
+	set_bit(BCH_FS_may_go_rw, &c->flags);
+
+	if (keys->nr || c->opts.fsck || !c->sb.clean)
+		return bch2_fs_read_write_early(c);
+	return 0;
+}
+
+struct recovery_pass_fn {
+	int		(*fn)(struct bch_fs *);
+	unsigned	when;
+};
+
+static struct recovery_pass_fn recovery_pass_fns[] = {
+#define x(_fn, _id, _when)	{ .fn = bch2_##_fn, .when = _when },
+	BCH_RECOVERY_PASSES()
+#undef x
+};
+
+u64 bch2_recovery_passes_to_stable(u64 v)
+{
+	static const u8 map[] = {
+#define x(n, id, ...)	[BCH_RECOVERY_PASS_##n] = BCH_RECOVERY_PASS_STABLE_##n,
+	BCH_RECOVERY_PASSES()
+#undef x
+	};
+
+	u64 ret = 0;
+	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
+		if (v & BIT_ULL(i))
+			ret |= BIT_ULL(map[i]);
+	return ret;
+}
+
+u64 bch2_recovery_passes_from_stable(u64 v)
+{
+	static const u8 map[] = {
+#define x(n, id, ...)	[BCH_RECOVERY_PASS_STABLE_##n] = BCH_RECOVERY_PASS_##n,
+	BCH_RECOVERY_PASSES()
+#undef x
+	};
+
+	u64 ret = 0;
+	for (unsigned i = 0; i < ARRAY_SIZE(map); i++)
+		if (v & BIT_ULL(i))
+			ret |= BIT_ULL(map[i]);
+	return ret;
+}
+
+/*
+ * For when we need to rewind recovery passes and run a pass we skipped:
+ */
+int bch2_run_explicit_recovery_pass(struct bch_fs *c,
+				    enum bch_recovery_pass pass)
+{
+	if (c->recovery_passes_explicit & BIT_ULL(pass))
+		return 0;
+
+	bch_info(c, "running explicit recovery pass %s (%u), currently at %s (%u)",
+		 bch2_recovery_passes[pass], pass,
+		 bch2_recovery_passes[c->curr_recovery_pass], c->curr_recovery_pass);
+
+	c->recovery_passes_explicit |= BIT_ULL(pass);
+
+	if (c->curr_recovery_pass >= pass) {
+		c->curr_recovery_pass = pass;
+		c->recovery_passes_complete &= (1ULL << pass) >> 1;
+		return -BCH_ERR_restart_recovery;
+	} else {
+		return 0;
+	}
+}
+
+int bch2_run_explicit_recovery_pass_persistent(struct bch_fs *c,
+					       enum bch_recovery_pass pass)
+{
+	__le64 s = cpu_to_le64(bch2_recovery_passes_to_stable(BIT_ULL(pass)));
+
+	mutex_lock(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
+
+	if (!(ext->recovery_passes_required[0] & s)) {
+		ext->recovery_passes_required[0] |= s;
+		bch2_write_super(c);
+	}
+	mutex_unlock(&c->sb_lock);
+
+	return bch2_run_explicit_recovery_pass(c, pass);
+}
+
+u64 bch2_fsck_recovery_passes(void)
+{
+	u64 ret = 0;
+
+	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
+		if (recovery_pass_fns[i].when & PASS_FSCK)
+			ret |= BIT_ULL(i);
+	return ret;
+}
+
+static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
+{
+	struct recovery_pass_fn *p = recovery_pass_fns + pass;
+
+	if (c->recovery_passes_explicit & BIT_ULL(pass))
+		return true;
+	if ((p->when & PASS_FSCK) && c->opts.fsck)
+		return true;
+	if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
+		return true;
+	if (p->when & PASS_ALWAYS)
+		return true;
+	return false;
+}
+
+static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
+{
+	struct recovery_pass_fn *p = recovery_pass_fns + pass;
+	int ret;
+
+	if (!(p->when & PASS_SILENT))
+		bch2_print(c, KERN_INFO bch2_log_msg(c, "%s..."),
+			   bch2_recovery_passes[pass]);
+	ret = p->fn(c);
+	if (ret)
+		return ret;
+	if (!(p->when & PASS_SILENT))
+		bch2_print(c, KERN_CONT " done\n");
+
+	return 0;
+}
+
+int bch2_run_online_recovery_passes(struct bch_fs *c)
+{
+	int ret = 0;
+
+	for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++) {
+		struct recovery_pass_fn *p = recovery_pass_fns + i;
+
+		if (!(p->when & PASS_ONLINE))
+			continue;
+
+		ret = bch2_run_recovery_pass(c, i);
+		if (bch2_err_matches(ret, BCH_ERR_restart_recovery)) {
+			i = c->curr_recovery_pass;
+			continue;
+		}
+		if (ret)
+			break;
+	}
+
+	return ret;
+}
+
+int bch2_run_recovery_passes(struct bch_fs *c)
+{
+	int ret = 0;
+
+	while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
+		if (c->opts.recovery_pass_last &&
+		    c->curr_recovery_pass > c->opts.recovery_pass_last)
+			break;
+
+		if (should_run_recovery_pass(c, c->curr_recovery_pass)) {
+			unsigned pass = c->curr_recovery_pass;
+
+			ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
+			if (bch2_err_matches(ret, BCH_ERR_restart_recovery) ||
+			    (ret && c->curr_recovery_pass < pass))
+				continue;
+			if (ret)
+				break;
+
+			c->recovery_passes_complete |= BIT_ULL(c->curr_recovery_pass);
+		}
+
+		c->recovery_pass_done = max(c->recovery_pass_done, c->curr_recovery_pass);
+
+		c->curr_recovery_pass++;
+	}
+
+	return ret;
+}
diff --git a/fs/bcachefs/recovery_passes.h b/fs/bcachefs/recovery_passes.h
new file mode 100644
index 00000000000000..99b464e127b80b
--- /dev/null
+++ b/fs/bcachefs/recovery_passes.h
@@ -0,0 +1,17 @@
+#ifndef _BCACHEFS_RECOVERY_PASSES_H
+#define _BCACHEFS_RECOVERY_PASSES_H
+
+extern const char * const bch2_recovery_passes[];
+
+u64 bch2_recovery_passes_to_stable(u64 v);
+u64 bch2_recovery_passes_from_stable(u64 v);
+
+u64 bch2_fsck_recovery_passes(void);
+
+int bch2_run_explicit_recovery_pass(struct bch_fs *, enum bch_recovery_pass);
+int bch2_run_explicit_recovery_pass_persistent(struct bch_fs *, enum bch_recovery_pass);
+
+int bch2_run_online_recovery_passes(struct bch_fs *);
+int bch2_run_recovery_passes(struct bch_fs *);
+
+#endif /* _BCACHEFS_RECOVERY_PASSES_H */
diff --git a/fs/bcachefs/recovery_types.h b/fs/bcachefs/recovery_passes_types.h
index 4959e95e7c7465..f3052128570683 100644
--- a/fs/bcachefs/recovery_types.h
+++ b/fs/bcachefs/recovery_passes_types.h
@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _BCACHEFS_RECOVERY_TYPES_H
-#define _BCACHEFS_RECOVERY_TYPES_H
+#ifndef _BCACHEFS_RECOVERY_PASSES_TYPES_H
+#define _BCACHEFS_RECOVERY_PASSES_TYPES_H
 
 #define PASS_SILENT		BIT(0)
 #define PASS_FSCK		BIT(1)
@@ -37,7 +37,6 @@
 	x(check_subvol_children,		35, PASS_ONLINE|PASS_FSCK)	\
 	x(delete_dead_snapshots,		21, PASS_ONLINE|PASS_FSCK)	\
 	x(fs_upgrade_for_subvolumes,		22, 0)				\
-	x(resume_logged_ops,			23, PASS_ALWAYS)		\
 	x(check_inodes,				24, PASS_FSCK)			\
 	x(check_extents,			25, PASS_FSCK)			\
 	x(check_indirect_extents,		26, PASS_FSCK)			\
@@ -47,6 +46,7 @@
 	x(check_subvolume_structure,		36, PASS_ONLINE|PASS_FSCK)	\
 	x(check_directory_structure,		30, PASS_ONLINE|PASS_FSCK)	\
 	x(check_nlinks,				31, PASS_FSCK)			\
+	x(resume_logged_ops,			23, PASS_ALWAYS)		\
 	x(delete_dead_inodes,			32, PASS_FSCK|PASS_UNCLEAN)	\
 	x(fix_reflink_p,			33, 0)				\
 	x(set_fs_needs_rebalance,		34, 0)				\
@@ -56,6 +56,7 @@ enum bch_recovery_pass {
 #define x(n, id, when)	BCH_RECOVERY_PASS_##n,
 	BCH_RECOVERY_PASSES()
 #undef x
+	BCH_RECOVERY_PASS_NR
 };
 
 /* But we also need stable identifiers that can be used in the superblock */
@@ -65,4 +66,4 @@ enum bch_recovery_pass_stable {
 #undef x
 };
 
-#endif /* _BCACHEFS_RECOVERY_TYPES_H */
+#endif /* _BCACHEFS_RECOVERY_PASSES_TYPES_H */
diff --git a/fs/bcachefs/reflink.c b/fs/bcachefs/reflink.c
index c47c66c2b394dc..ff7864731a073d 100644
--- a/fs/bcachefs/reflink.c
+++ b/fs/bcachefs/reflink.c
@@ -185,8 +185,7 @@ not_found:
 		} else {
 			bkey_error_init(update);
 			update->k.p		= p.k->p;
-			update->k.p.offset	= next_idx;
-			update->k.size		= next_idx - *idx;
+			update->k.size		= p.k->size;
 			set_bkey_val_u64s(&update->k, 0);
 		}
 
diff --git a/fs/bcachefs/sb-downgrade.c b/fs/bcachefs/sb-downgrade.c
index e4396cb0bacb03..d6f81179c3a29b 100644
--- a/fs/bcachefs/sb-downgrade.c
+++ b/fs/bcachefs/sb-downgrade.c
@@ -7,7 +7,7 @@
 
 #include "bcachefs.h"
 #include "darray.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "sb-downgrade.h"
 #include "sb-errors.h"
 #include "super-io.h"
diff --git a/fs/bcachefs/sb-errors_types.h b/fs/bcachefs/sb-errors_types.h
index 5178bf579f7c53..73e9634df8ffb0 100644
--- a/fs/bcachefs/sb-errors_types.h
+++ b/fs/bcachefs/sb-errors_types.h
@@ -265,7 +265,9 @@
 	x(subvol_children_bad,					257)	\
 	x(subvol_loop,						258)	\
 	x(subvol_unreachable,					259)	\
-	x(btree_node_bkey_bad_u64s,				260)
+	x(btree_node_bkey_bad_u64s,				260)	\
+	x(btree_node_topology_empty_interior_node,		261)	\
+	x(btree_ptr_v2_min_key_bad,				262)
 
 enum bch_sb_error_id {
 #define x(t, n) BCH_FSCK_ERR_##t = n,
diff --git a/fs/bcachefs/snapshot.c b/fs/bcachefs/snapshot.c
index 39debe814bf392..4577ee7939a2a5 100644
--- a/fs/bcachefs/snapshot.c
+++ b/fs/bcachefs/snapshot.c
@@ -93,8 +93,10 @@ static int bch2_snapshot_tree_create(struct btree_trans *trans,
 
 static bool __bch2_snapshot_is_ancestor_early(struct snapshot_table *t, u32 id, u32 ancestor)
 {
-	while (id && id < ancestor)
-		id = __snapshot_t(t, id)->parent;
+	while (id && id < ancestor) {
+		const struct snapshot_t *s = __snapshot_t(t, id);
+		id = s ? s->parent : 0;
+	}
 	return id == ancestor;
 }
 
@@ -110,6 +112,8 @@ static bool bch2_snapshot_is_ancestor_early(struct bch_fs *c, u32 id, u32 ancest
 static inline u32 get_ancestor_below(struct snapshot_table *t, u32 id, u32 ancestor)
 {
 	const struct snapshot_t *s = __snapshot_t(t, id);
+	if (!s)
+		return 0;
 
 	if (s->skip[2] <= ancestor)
 		return s->skip[2];
@@ -127,7 +131,7 @@ bool __bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ancestor)
 	rcu_read_lock();
 	struct snapshot_table *t = rcu_dereference(c->snapshots);
 
-	if (unlikely(c->recovery_pass_done <= BCH_RECOVERY_PASS_check_snapshots)) {
+	if (unlikely(c->recovery_pass_done < BCH_RECOVERY_PASS_check_snapshots)) {
 		ret = __bch2_snapshot_is_ancestor_early(t, id, ancestor);
 		goto out;
 	}
@@ -151,36 +155,39 @@ out:
 static noinline struct snapshot_t *__snapshot_t_mut(struct bch_fs *c, u32 id)
 {
 	size_t idx = U32_MAX - id;
-	size_t new_size;
 	struct snapshot_table *new, *old;
 
-	new_size = max(16UL, roundup_pow_of_two(idx + 1));
+	size_t new_bytes = kmalloc_size_roundup(struct_size(new, s, idx + 1));
+	size_t new_size = (new_bytes - sizeof(*new)) / sizeof(new->s[0]);
 
-	new = kvzalloc(struct_size(new, s, new_size), GFP_KERNEL);
+	new = kvzalloc(new_bytes, GFP_KERNEL);
 	if (!new)
 		return NULL;
 
+	new->nr = new_size;
+
 	old = rcu_dereference_protected(c->snapshots, true);
 	if (old)
-		memcpy(new->s,
-		       rcu_dereference_protected(c->snapshots, true)->s,
-		       sizeof(new->s[0]) * c->snapshot_table_size);
+		memcpy(new->s, old->s, sizeof(old->s[0]) * old->nr);
 
 	rcu_assign_pointer(c->snapshots, new);
-	c->snapshot_table_size = new_size;
-	kvfree_rcu_mightsleep(old);
+	kvfree_rcu(old, rcu);
 
-	return &rcu_dereference_protected(c->snapshots, true)->s[idx];
+	return &rcu_dereference_protected(c->snapshots,
+				lockdep_is_held(&c->snapshot_table_lock))->s[idx];
 }
 
 static inline struct snapshot_t *snapshot_t_mut(struct bch_fs *c, u32 id)
 {
 	size_t idx = U32_MAX - id;
+	struct snapshot_table *table =
+		rcu_dereference_protected(c->snapshots,
+				lockdep_is_held(&c->snapshot_table_lock));
 
 	lockdep_assert_held(&c->snapshot_table_lock);
 
-	if (likely(idx < c->snapshot_table_size))
-		return &rcu_dereference_protected(c->snapshots, true)->s[idx];
+	if (likely(table && idx < table->nr))
+		return &table->s[idx];
 
 	return __snapshot_t_mut(c, id);
 }
diff --git a/fs/bcachefs/snapshot.h b/fs/bcachefs/snapshot.h
index 7c66ffc06385dd..331f20fd8d03d1 100644
--- a/fs/bcachefs/snapshot.h
+++ b/fs/bcachefs/snapshot.h
@@ -33,7 +33,11 @@ int bch2_mark_snapshot(struct btree_trans *, enum btree_id, unsigned,
 
 static inline struct snapshot_t *__snapshot_t(struct snapshot_table *t, u32 id)
 {
-	return &t->s[U32_MAX - id];
+	u32 idx = U32_MAX - id;
+
+	return likely(t && idx < t->nr)
+		? &t->s[idx]
+		: NULL;
 }
 
 static inline const struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
@@ -44,7 +48,8 @@ static inline const struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
 static inline u32 bch2_snapshot_tree(struct bch_fs *c, u32 id)
 {
 	rcu_read_lock();
-	id = snapshot_t(c, id)->tree;
+	const struct snapshot_t *s = snapshot_t(c, id);
+	id = s ? s->tree : 0;
 	rcu_read_unlock();
 
 	return id;
@@ -52,7 +57,8 @@ static inline u32 bch2_snapshot_tree(struct bch_fs *c, u32 id)
 
 static inline u32 __bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
 {
-	return snapshot_t(c, id)->parent;
+	const struct snapshot_t *s = snapshot_t(c, id);
+	return s ? s->parent : 0;
 }
 
 static inline u32 bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
@@ -66,19 +72,19 @@ static inline u32 bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
 
 static inline u32 __bch2_snapshot_parent(struct bch_fs *c, u32 id)
 {
-#ifdef CONFIG_BCACHEFS_DEBUG
-	u32 parent = snapshot_t(c, id)->parent;
+	const struct snapshot_t *s = snapshot_t(c, id);
+	if (!s)
+		return 0;
 
-	if (parent &&
-	    snapshot_t(c, id)->depth != snapshot_t(c, parent)->depth + 1)
+	u32 parent = s->parent;
+	if (IS_ENABLED(CONFIG_BCACHEFS_DEBU) &&
+	    parent &&
+	    s->depth != snapshot_t(c, parent)->depth + 1)
 		panic("id %u depth=%u parent %u depth=%u\n",
 		      id, snapshot_t(c, id)->depth,
 		      parent, snapshot_t(c, parent)->depth);
 
 	return parent;
-#else
-	return snapshot_t(c, id)->parent;
-#endif
 }
 
 static inline u32 bch2_snapshot_parent(struct bch_fs *c, u32 id)
@@ -116,7 +122,8 @@ static inline u32 bch2_snapshot_root(struct bch_fs *c, u32 id)
 
 static inline u32 __bch2_snapshot_equiv(struct bch_fs *c, u32 id)
 {
-	return snapshot_t(c, id)->equiv;
+	const struct snapshot_t *s = snapshot_t(c, id);
+	return s ? s->equiv : 0;
 }
 
 static inline u32 bch2_snapshot_equiv(struct bch_fs *c, u32 id)
@@ -133,38 +140,22 @@ static inline bool bch2_snapshot_is_equiv(struct bch_fs *c, u32 id)
 	return id == bch2_snapshot_equiv(c, id);
 }
 
-static inline bool bch2_snapshot_is_internal_node(struct bch_fs *c, u32 id)
+static inline int bch2_snapshot_is_internal_node(struct bch_fs *c, u32 id)
 {
-	const struct snapshot_t *s;
-	bool ret;
-
 	rcu_read_lock();
-	s = snapshot_t(c, id);
-	ret = s->children[0];
+	const struct snapshot_t *s = snapshot_t(c, id);
+	int ret = s ? s->children[0] : -BCH_ERR_invalid_snapshot_node;
 	rcu_read_unlock();
 
 	return ret;
 }
 
-static inline u32 bch2_snapshot_is_leaf(struct bch_fs *c, u32 id)
-{
-	return !bch2_snapshot_is_internal_node(c, id);
-}
-
-static inline u32 bch2_snapshot_sibling(struct bch_fs *c, u32 id)
+static inline int bch2_snapshot_is_leaf(struct bch_fs *c, u32 id)
 {
-	const struct snapshot_t *s;
-	u32 parent = __bch2_snapshot_parent(c, id);
-
-	if (!parent)
-		return 0;
-
-	s = snapshot_t(c, __bch2_snapshot_parent(c, id));
-	if (id == s->children[0])
-		return s->children[1];
-	if (id == s->children[1])
-		return s->children[0];
-	return 0;
+	int ret = bch2_snapshot_is_internal_node(c, id);
+	if (ret < 0)
+		return ret;
+	return !ret;
 }
 
 static inline u32 bch2_snapshot_depth(struct bch_fs *c, u32 parent)
@@ -249,7 +240,7 @@ static inline int bch2_key_has_snapshot_overwrites(struct btree_trans *trans,
 					  struct bpos pos)
 {
 	if (!btree_type_has_snapshots(id) ||
-	    bch2_snapshot_is_leaf(trans->c, pos.snapshot))
+	    bch2_snapshot_is_leaf(trans->c, pos.snapshot) > 0)
 		return 0;
 
 	return __bch2_key_has_snapshot_overwrites(trans, id, pos);
diff --git a/fs/bcachefs/subvolume.c b/fs/bcachefs/subvolume.c
index ce7aed12194238..88a79c82327687 100644
--- a/fs/bcachefs/subvolume.c
+++ b/fs/bcachefs/subvolume.c
@@ -595,6 +595,78 @@ err:
 	return ret;
 }
 
+int bch2_initialize_subvolumes(struct bch_fs *c)
+{
+	struct bkey_i_snapshot_tree	root_tree;
+	struct bkey_i_snapshot		root_snapshot;
+	struct bkey_i_subvolume		root_volume;
+	int ret;
+
+	bkey_snapshot_tree_init(&root_tree.k_i);
+	root_tree.k.p.offset		= 1;
+	root_tree.v.master_subvol	= cpu_to_le32(1);
+	root_tree.v.root_snapshot	= cpu_to_le32(U32_MAX);
+
+	bkey_snapshot_init(&root_snapshot.k_i);
+	root_snapshot.k.p.offset = U32_MAX;
+	root_snapshot.v.flags	= 0;
+	root_snapshot.v.parent	= 0;
+	root_snapshot.v.subvol	= cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
+	root_snapshot.v.tree	= cpu_to_le32(1);
+	SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
+
+	bkey_subvolume_init(&root_volume.k_i);
+	root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
+	root_volume.v.flags	= 0;
+	root_volume.v.snapshot	= cpu_to_le32(U32_MAX);
+	root_volume.v.inode	= cpu_to_le64(BCACHEFS_ROOT_INO);
+
+	ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,	&root_tree.k_i, NULL, 0) ?:
+		bch2_btree_insert(c, BTREE_ID_snapshots,	&root_snapshot.k_i, NULL, 0) ?:
+		bch2_btree_insert(c, BTREE_ID_subvolumes,	&root_volume.k_i, NULL, 0);
+	bch_err_fn(c, ret);
+	return ret;
+}
+
+static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	struct bch_inode_unpacked inode;
+	int ret;
+
+	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
+			       SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
+	ret = bkey_err(k);
+	if (ret)
+		return ret;
+
+	if (!bkey_is_inode(k.k)) {
+		bch_err(trans->c, "root inode not found");
+		ret = -BCH_ERR_ENOENT_inode;
+		goto err;
+	}
+
+	ret = bch2_inode_unpack(k, &inode);
+	BUG_ON(ret);
+
+	inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
+
+	ret = bch2_inode_write(trans, &iter, &inode);
+err:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
+/* set bi_subvol on root inode */
+int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
+{
+	int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
+				__bch2_fs_upgrade_for_subvolumes(trans));
+	bch_err_fn(c, ret);
+	return ret;
+}
+
 int bch2_fs_subvolumes_init(struct bch_fs *c)
 {
 	INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
diff --git a/fs/bcachefs/subvolume.h b/fs/bcachefs/subvolume.h
index 903c05162c0688..d2015d549bd2a3 100644
--- a/fs/bcachefs/subvolume.h
+++ b/fs/bcachefs/subvolume.h
@@ -37,6 +37,9 @@ void bch2_delete_dead_snapshots_async(struct bch_fs *);
 int bch2_subvolume_unlink(struct btree_trans *, u32);
 int bch2_subvolume_create(struct btree_trans *, u64, u32, u32, u32 *, u32 *, bool);
 
+int bch2_initialize_subvolumes(struct bch_fs *);
+int bch2_fs_upgrade_for_subvolumes(struct bch_fs *);
+
 int bch2_fs_subvolumes_init(struct bch_fs *);
 
 #endif /* _BCACHEFS_SUBVOLUME_H */
diff --git a/fs/bcachefs/subvolume_types.h b/fs/bcachefs/subvolume_types.h
index ae644adfc39168..9b10c8947828e0 100644
--- a/fs/bcachefs/subvolume_types.h
+++ b/fs/bcachefs/subvolume_types.h
@@ -20,6 +20,8 @@ struct snapshot_t {
 };
 
 struct snapshot_table {
+	struct rcu_head		rcu;
+	size_t			nr;
 #ifndef RUST_BINDGEN
 	DECLARE_FLEX_ARRAY(struct snapshot_t, s);
 #else
diff --git a/fs/bcachefs/super-io.c b/fs/bcachefs/super-io.c
index ad28e370b6404c..e15f8b1f30c2ac 100644
--- a/fs/bcachefs/super-io.c
+++ b/fs/bcachefs/super-io.c
@@ -8,7 +8,7 @@
 #include "journal.h"
 #include "journal_sb.h"
 #include "journal_seq_blacklist.h"
-#include "recovery.h"
+#include "recovery_passes.h"
 #include "replicas.h"
 #include "quota.h"
 #include "sb-clean.h"
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 1ad6e5cd9476c8..bc026a77eb99d4 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -365,7 +365,7 @@ void bch2_fs_read_only(struct bch_fs *c)
 	    !test_bit(BCH_FS_emergency_ro, &c->flags) &&
 	    test_bit(BCH_FS_started, &c->flags) &&
 	    test_bit(BCH_FS_clean_shutdown, &c->flags) &&
-	    !c->opts.norecovery) {
+	    c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) {
 		BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
 		BUG_ON(atomic_read(&c->btree_cache.dirty));
 		BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
@@ -510,7 +510,8 @@ err:
 
 int bch2_fs_read_write(struct bch_fs *c)
 {
-	if (c->opts.norecovery)
+	if (c->opts.recovery_pass_last &&
+	    c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay)
 		return -BCH_ERR_erofs_norecovery;
 
 	if (c->opts.nochanges)
@@ -1015,8 +1016,16 @@ int bch2_fs_start(struct bch_fs *c)
 	for_each_online_member(c, ca)
 		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
 
+	struct bch_sb_field_ext *ext =
+		bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
 	mutex_unlock(&c->sb_lock);
 
+	if (!ext) {
+		bch_err(c, "insufficient space in superblock for sb_field_ext");
+		ret = -BCH_ERR_ENOSPC_sb;
+		goto err;
+	}
+
 	for_each_rw_member(c, ca)
 		bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);