Update bcachefs sources to a41cc7750fb8 bcachefs: Add flags to subvolume_to_text()

67 files changed, 982 insertions, 745 deletions

diff --git a/.bcachefs_revision b/.bcachefs_revision
index 957e0f1c..d586fee7 100644
--- a/.bcachefs_revision
+++ b/.bcachefs_revision
@@ -1 +1 @@
-d316ba08b04582f7dfc7d66b5c97cb50bf863a1c
+a41cc7750fb8af26a79323542de3d8244fe90bfc
diff --git a/libbcachefs/alloc_background.c b/libbcachefs/alloc_background.c
index e1231b20..b228a5a6 100644
--- a/libbcachefs/alloc_background.c
+++ b/libbcachefs/alloc_background.c
@@ -21,7 +21,6 @@
 #include "error.h"
 #include "lru.h"
 #include "recovery.h"
-#include "trace.h"
 #include "varint.h"
 
 #include <linux/kthread.h>
@@ -866,7 +865,7 @@ int bch2_trigger_alloc(struct btree_trans *trans,
 
 	struct bch_dev *ca = bch2_dev_bucket_tryget(c, new.k->p);
 	if (!ca)
-		return -BCH_ERR_trigger_alloc;
+		return bch_err_throw(c, trigger_alloc);
 
 	struct bch_alloc_v4 old_a_convert;
 	const struct bch_alloc_v4 *old_a = bch2_alloc_to_v4(old, &old_a_convert);
@@ -1045,7 +1044,7 @@ fsck_err:
 invalid_bucket:
 	bch2_fs_inconsistent(c, "reference to invalid bucket\n%s",
 			     (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf));
-	ret = -BCH_ERR_trigger_alloc;
+	ret = bch_err_throw(c, trigger_alloc);
 	goto err;
 }
 
@@ -1459,7 +1458,7 @@ delete:
 		ret =   bch2_btree_bit_mod_iter(trans, iter, false) ?:
 			bch2_trans_commit(trans, NULL, NULL,
 				BCH_TRANS_COMMIT_no_enospc) ?:
-			-BCH_ERR_transaction_restart_commit;
+			bch_err_throw(c, transaction_restart_commit);
 		goto out;
 	} else {
 		/*
@@ -1782,13 +1781,14 @@ int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
 
 static int discard_in_flight_add(struct bch_dev *ca, u64 bucket, bool in_progress)
 {
+	struct bch_fs *c = ca->fs;
 	int ret;
 
 	mutex_lock(&ca->discard_buckets_in_flight_lock);
 	struct discard_in_flight *i =
 		darray_find_p(ca->discard_buckets_in_flight, i, i->bucket == bucket);
 	if (i) {
-		ret = -BCH_ERR_EEXIST_discard_in_flight_add;
+		ret = bch_err_throw(c, EEXIST_discard_in_flight_add);
 		goto out;
 	}
 
diff --git a/libbcachefs/alloc_foreground.c b/libbcachefs/alloc_foreground.c
index e157bc86..b375ad61 100644
--- a/libbcachefs/alloc_foreground.c
+++ b/libbcachefs/alloc_foreground.c
@@ -227,7 +227,7 @@ static struct open_bucket *__try_alloc_bucket(struct bch_fs *c,
 
 		track_event_change(&c->times[BCH_TIME_blocked_allocate_open_bucket], true);
 		spin_unlock(&c->freelist_lock);
-		return ERR_PTR(-BCH_ERR_open_buckets_empty);
+		return ERR_PTR(bch_err_throw(c, open_buckets_empty));
 	}
 
 	/* Recheck under lock: */
@@ -533,7 +533,7 @@ again:
 
 		track_event_change(&c->times[BCH_TIME_blocked_allocate], true);
 
-		ob = ERR_PTR(-BCH_ERR_freelist_empty);
+		ob = ERR_PTR(bch_err_throw(c, freelist_empty));
 		goto err;
 	}
 
@@ -558,7 +558,7 @@ alloc:
 	}
 err:
 	if (!ob)
-		ob = ERR_PTR(-BCH_ERR_no_buckets_found);
+		ob = ERR_PTR(bch_err_throw(c, no_buckets_found));
 
 	if (!IS_ERR(ob))
 		ob->data_type = req->data_type;
@@ -709,7 +709,7 @@ inline int bch2_bucket_alloc_set_trans(struct btree_trans *trans,
 				       struct closure *cl)
 {
 	struct bch_fs *c = trans->c;
-	int ret = -BCH_ERR_insufficient_devices;
+	int ret = 0;
 
 	BUG_ON(req->nr_effective >= req->nr_replicas);
 
@@ -738,13 +738,16 @@ inline int bch2_bucket_alloc_set_trans(struct btree_trans *trans,
 			continue;
 		}
 
-		if (add_new_bucket(c, req, ob)) {
-			ret = 0;
+		ret = add_new_bucket(c, req, ob);
+		if (ret)
 			break;
-		}
 	}
 
-	return ret;
+	if (ret == 1)
+		return 0;
+	if (ret)
+		return ret;
+	return bch_err_throw(c, insufficient_devices);
 }
 
 /* Allocate from stripes: */
@@ -1373,11 +1376,11 @@ err:
 		goto retry;
 
 	if (cl && bch2_err_matches(ret, BCH_ERR_open_buckets_empty))
-		ret = -BCH_ERR_bucket_alloc_blocked;
+		ret = bch_err_throw(c, bucket_alloc_blocked);
 
 	if (cl && !(flags & BCH_WRITE_alloc_nowait) &&
 	    bch2_err_matches(ret, BCH_ERR_freelist_empty))
-		ret = -BCH_ERR_bucket_alloc_blocked;
+		ret = bch_err_throw(c, bucket_alloc_blocked);
 
 	return ret;
 }
diff --git a/libbcachefs/backpointers.c b/libbcachefs/backpointers.c
index ebc8ee2c..e76809e7 100644
--- a/libbcachefs/backpointers.c
+++ b/libbcachefs/backpointers.c
@@ -142,7 +142,7 @@ static noinline int backpointer_mod_err(struct btree_trans *trans,
 	}
 
 	if (!will_check && __bch2_inconsistent_error(c, &buf))
-		ret = -BCH_ERR_erofs_unfixed_errors;
+		ret = bch_err_throw(c, erofs_unfixed_errors);
 
 	bch_err(c, "%s", buf.buf);
 	printbuf_exit(&buf);
@@ -295,7 +295,7 @@ static struct btree *__bch2_backpointer_get_node(struct btree_trans *trans,
 		return b;
 
 	if (btree_node_will_make_reachable(b)) {
-		b = ERR_PTR(-BCH_ERR_backpointer_to_overwritten_btree_node);
+		b = ERR_PTR(bch_err_throw(c, backpointer_to_overwritten_btree_node));
 	} else {
 		int ret = backpointer_target_not_found(trans, bp, bkey_i_to_s_c(&b->key),
 						       last_flushed, commit);
@@ -353,7 +353,7 @@ static struct bkey_s_c __bch2_backpointer_get_key(struct btree_trans *trans,
 		return ret ? bkey_s_c_err(ret) : bkey_s_c_null;
 	} else {
 		struct btree *b = __bch2_backpointer_get_node(trans, bp, iter, last_flushed, commit);
-		if (b == ERR_PTR(-BCH_ERR_backpointer_to_overwritten_btree_node))
+		if (b == ERR_PTR(bch_err_throw(c, backpointer_to_overwritten_btree_node)))
 			return bkey_s_c_null;
 		if (IS_ERR_OR_NULL(b))
 			return ((struct bkey_s_c) { .k = ERR_CAST(b) });
@@ -651,7 +651,7 @@ check_existing_bp:
 	prt_newline(&buf);
 	bch2_bkey_val_to_text(&buf, c, other_extent);
 	bch_err(c, "%s", buf.buf);
-	ret = -BCH_ERR_fsck_repair_unimplemented;
+	ret = bch_err_throw(c, fsck_repair_unimplemented);
 	goto err;
 missing:
 	printbuf_reset(&buf);
@@ -953,7 +953,7 @@ static int check_bucket_backpointer_mismatch(struct btree_trans *trans, struct b
 		    sectors[ALLOC_cached] > a->cached_sectors ||
 		    sectors[ALLOC_stripe] > a->stripe_sectors) {
 			ret = check_bucket_backpointers_to_extents(trans, ca, alloc_k.k->p) ?:
-				-BCH_ERR_transaction_restart_nested;
+				bch_err_throw(c, transaction_restart_nested);
 			goto err;
 		}
 
@@ -1351,7 +1351,7 @@ static int bch2_bucket_bitmap_set(struct bch_dev *ca, struct bucket_bitmap *b, u
 			b->buckets = kvcalloc(BITS_TO_LONGS(ca->mi.nbuckets),
 					      sizeof(unsigned long), GFP_KERNEL);
 			if (!b->buckets)
-				return -BCH_ERR_ENOMEM_backpointer_mismatches_bitmap;
+				return bch_err_throw(ca->fs, ENOMEM_backpointer_mismatches_bitmap);
 		}
 
 		b->nr += !__test_and_set_bit(bit, b->buckets);
@@ -1360,7 +1360,8 @@ static int bch2_bucket_bitmap_set(struct bch_dev *ca, struct bucket_bitmap *b, u
 	return 0;
 }
 
-int bch2_bucket_bitmap_resize(struct bucket_bitmap *b, u64 old_size, u64 new_size)
+int bch2_bucket_bitmap_resize(struct bch_dev *ca, struct bucket_bitmap *b,
+			      u64 old_size, u64 new_size)
 {
 	scoped_guard(mutex, &b->lock) {
 		if (!b->buckets)
@@ -1369,7 +1370,7 @@ int bch2_bucket_bitmap_resize(struct bucket_bitmap *b, u64 old_size, u64 new_siz
 		unsigned long *n = kvcalloc(BITS_TO_LONGS(new_size),
 					    sizeof(unsigned long), GFP_KERNEL);
 		if (!n)
-			return -BCH_ERR_ENOMEM_backpointer_mismatches_bitmap;
+			return bch_err_throw(ca->fs, ENOMEM_backpointer_mismatches_bitmap);
 
 		memcpy(n, b->buckets,
 		       BITS_TO_LONGS(min(old_size, new_size)) * sizeof(unsigned long));
diff --git a/libbcachefs/backpointers.h b/libbcachefs/backpointers.h
index fac05948..7e71afee 100644
--- a/libbcachefs/backpointers.h
+++ b/libbcachefs/backpointers.h
@@ -194,7 +194,7 @@ static inline bool bch2_bucket_bitmap_test(struct bucket_bitmap *b, u64 i)
 	return bitmap && test_bit(i, bitmap);
 }
 
-int bch2_bucket_bitmap_resize(struct bucket_bitmap *, u64, u64);
+int bch2_bucket_bitmap_resize(struct bch_dev *, struct bucket_bitmap *, u64, u64);
 void bch2_bucket_bitmap_free(struct bucket_bitmap *);
 
 #endif /* _BCACHEFS_BACKPOINTERS_BACKGROUND_H */
diff --git a/libbcachefs/bcachefs.h b/libbcachefs/bcachefs.h
index 7824da2a..3651a296 100644
--- a/libbcachefs/bcachefs.h
+++ b/libbcachefs/bcachefs.h
@@ -183,6 +183,16 @@
 #define pr_fmt(fmt) "%s() " fmt "\n", __func__
 #endif
 
+#ifdef CONFIG_BCACHEFS_DEBUG
+#define ENUMERATED_REF_DEBUG
+#endif
+
+#ifndef dynamic_fault
+#define dynamic_fault(...)		0
+#endif
+
+#define race_fault(...)			dynamic_fault("bcachefs:race")
+
 #include <linux/backing-dev-defs.h>
 #include <linux/bug.h>
 #include <linux/bio.h>
@@ -219,15 +229,30 @@
 #include "time_stats.h"
 #include "util.h"
 
-#ifdef CONFIG_BCACHEFS_DEBUG
-#define ENUMERATED_REF_DEBUG
-#endif
-
-#ifndef dynamic_fault
-#define dynamic_fault(...)		0
-#endif
+#include "alloc_types.h"
+#include "async_objs_types.h"
+#include "btree_gc_types.h"
+#include "btree_types.h"
+#include "btree_node_scan_types.h"
+#include "btree_write_buffer_types.h"
+#include "buckets_types.h"
+#include "buckets_waiting_for_journal_types.h"
+#include "clock_types.h"
+#include "disk_groups_types.h"
+#include "ec_types.h"
+#include "enumerated_ref_types.h"
+#include "journal_types.h"
+#include "keylist_types.h"
+#include "quota_types.h"
+#include "rebalance_types.h"
+#include "recovery_passes_types.h"
+#include "replicas_types.h"
+#include "sb-members_types.h"
+#include "subvolume_types.h"
+#include "super_types.h"
+#include "thread_with_file_types.h"
 
-#define race_fault(...)			dynamic_fault("bcachefs:race")
+#include "trace.h"
 
 #define count_event(_c, _name)	this_cpu_inc((_c)->counters[BCH_COUNTER_##_name])
 
@@ -380,6 +405,14 @@ do {									\
 		pr_info(fmt, ##__VA_ARGS__);				\
 } while (0)
 
+static inline int __bch2_err_trace(struct bch_fs *c, int err)
+{
+	trace_error_throw(c, err, _THIS_IP_);
+	return err;
+}
+
+#define bch_err_throw(_c, _err) __bch2_err_trace(_c, -BCH_ERR_##_err)
+
 /* Parameters that are useful for debugging, but should always be compiled in: */
 #define BCH_DEBUG_PARAMS_ALWAYS()					\
 	BCH_DEBUG_PARAM(key_merging_disabled,				\
@@ -486,29 +519,6 @@ enum bch_time_stats {
 	BCH_TIME_STAT_NR
 };
 
-#include "alloc_types.h"
-#include "async_objs_types.h"
-#include "btree_gc_types.h"
-#include "btree_types.h"
-#include "btree_node_scan_types.h"
-#include "btree_write_buffer_types.h"
-#include "buckets_types.h"
-#include "buckets_waiting_for_journal_types.h"
-#include "clock_types.h"
-#include "disk_groups_types.h"
-#include "ec_types.h"
-#include "enumerated_ref_types.h"
-#include "journal_types.h"
-#include "keylist_types.h"
-#include "quota_types.h"
-#include "rebalance_types.h"
-#include "recovery_passes_types.h"
-#include "replicas_types.h"
-#include "sb-members_types.h"
-#include "subvolume_types.h"
-#include "super_types.h"
-#include "thread_with_file_types.h"
-
 /* Number of nodes btree coalesce will try to coalesce at once */
 #define GC_MERGE_NODES		4U
 
diff --git a/libbcachefs/btree_cache.c b/libbcachefs/btree_cache.c
index 8557cbd3..91e0aa79 100644
--- a/libbcachefs/btree_cache.c
+++ b/libbcachefs/btree_cache.c
@@ -149,7 +149,7 @@ static int btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp)
 
 	b->data = kvmalloc(btree_buf_bytes(b), gfp);
 	if (!b->data)
-		return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
+		return bch_err_throw(c, ENOMEM_btree_node_mem_alloc);
 #ifdef __KERNEL__
 	b->aux_data = kvmalloc(btree_aux_data_bytes(b), gfp);
 #else
@@ -162,7 +162,7 @@ static int btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp)
 	if (!b->aux_data) {
 		kvfree(b->data);
 		b->data = NULL;
-		return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
+		return bch_err_throw(c, ENOMEM_btree_node_mem_alloc);
 	}
 
 	return 0;
@@ -353,21 +353,21 @@ static int __btree_node_reclaim_checks(struct bch_fs *c, struct btree *b,
 
 	if (btree_node_noevict(b)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_noevict]++;
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 	if (btree_node_write_blocked(b)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_write_blocked]++;
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 	if (btree_node_will_make_reachable(b)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_will_make_reachable]++;
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 
 	if (btree_node_dirty(b)) {
 		if (!flush) {
 			bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_dirty]++;
-			return -BCH_ERR_ENOMEM_btree_node_reclaim;
+			return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 		}
 
 		if (locked) {
@@ -393,7 +393,7 @@ static int __btree_node_reclaim_checks(struct bch_fs *c, struct btree *b,
 				bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_read_in_flight]++;
 			else if (btree_node_write_in_flight(b))
 				bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_write_in_flight]++;
-			return -BCH_ERR_ENOMEM_btree_node_reclaim;
+			return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 		}
 
 		if (locked)
@@ -424,13 +424,13 @@ retry_unlocked:
 
 	if (!six_trylock_intent(&b->c.lock)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_lock_intent]++;
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 
 	if (!six_trylock_write(&b->c.lock)) {
 		bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_lock_write]++;
 		six_unlock_intent(&b->c.lock);
-		return -BCH_ERR_ENOMEM_btree_node_reclaim;
+		return bch_err_throw(c, ENOMEM_btree_node_reclaim);
 	}
 
 	/* recheck under lock */
@@ -682,7 +682,7 @@ int bch2_fs_btree_cache_init(struct bch_fs *c)
 
 	return 0;
 err:
-	return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+	return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 }
 
 void bch2_fs_btree_cache_init_early(struct btree_cache *bc)
@@ -727,7 +727,7 @@ int bch2_btree_cache_cannibalize_lock(struct btree_trans *trans, struct closure
 
 	if (!cl) {
 		trace_and_count(c, btree_cache_cannibalize_lock_fail, trans);
-		return -BCH_ERR_ENOMEM_btree_cache_cannibalize_lock;
+		return bch_err_throw(c, ENOMEM_btree_cache_cannibalize_lock);
 	}
 
 	closure_wait(&bc->alloc_wait, cl);
@@ -741,7 +741,7 @@ int bch2_btree_cache_cannibalize_lock(struct btree_trans *trans, struct closure
 	}
 
 	trace_and_count(c, btree_cache_cannibalize_lock_fail, trans);
-	return -BCH_ERR_btree_cache_cannibalize_lock_blocked;
+	return bch_err_throw(c, btree_cache_cannibalize_lock_blocked);
 
 success:
 	trace_and_count(c, btree_cache_cannibalize_lock, trans);
diff --git a/libbcachefs/btree_gc.c b/libbcachefs/btree_gc.c
index f9548672..ed942ec5 100644
--- a/libbcachefs/btree_gc.c
+++ b/libbcachefs/btree_gc.c
@@ -150,7 +150,7 @@ static int set_node_min(struct bch_fs *c, struct btree *b, struct bpos new_min)
 
 	new = kmalloc_array(BKEY_BTREE_PTR_U64s_MAX, sizeof(u64), GFP_KERNEL);
 	if (!new)
-		return -BCH_ERR_ENOMEM_gc_repair_key;
+		return bch_err_throw(c, ENOMEM_gc_repair_key);
 
 	btree_ptr_to_v2(b, new);
 	b->data->min_key	= new_min;
@@ -190,7 +190,7 @@ static int set_node_max(struct bch_fs *c, struct btree *b, struct bpos new_max)
 
 	new = kmalloc_array(BKEY_BTREE_PTR_U64s_MAX, sizeof(u64), GFP_KERNEL);
 	if (!new)
-		return -BCH_ERR_ENOMEM_gc_repair_key;
+		return bch_err_throw(c, ENOMEM_gc_repair_key);
 
 	btree_ptr_to_v2(b, new);
 	b->data->max_key	= new_max;
@@ -476,7 +476,8 @@ again:
 		if (ret)
 			goto err;
 
-		ret = bch2_btree_repair_topology_recurse(trans, cur, pulled_from_scan);
+		ret = lockrestart_do(trans,
+				bch2_btree_repair_topology_recurse(trans, cur, pulled_from_scan));
 		six_unlock_read(&cur->c.lock);
 		cur = NULL;
 
@@ -522,46 +523,60 @@ fsck_err:
 	return ret;
 }
 
+static int bch2_check_root(struct btree_trans *trans, enum btree_id i,
+			   bool *reconstructed_root)
+{
+	struct bch_fs *c = trans->c;
+	struct btree_root *r = bch2_btree_id_root(c, i);
+	struct printbuf buf = PRINTBUF;
+	int ret = 0;
+
+	bch2_btree_id_to_text(&buf, i);
+
+	if (r->error) {
+		bch_info(c, "btree root %s unreadable, must recover from scan", buf.buf);
+
+		r->alive = false;
+		r->error = 0;
+
+		if (!bch2_btree_has_scanned_nodes(c, i)) {
+			__fsck_err(trans,
+				   FSCK_CAN_FIX|(!btree_id_important(i) ? FSCK_AUTOFIX : 0),
+				   btree_root_unreadable_and_scan_found_nothing,
+				   "no nodes found for btree %s, continue?", buf.buf);
+			bch2_btree_root_alloc_fake_trans(trans, i, 0);
+		} else {
+			bch2_btree_root_alloc_fake_trans(trans, i, 1);
+			bch2_shoot_down_journal_keys(c, i, 1, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
+			ret = bch2_get_scanned_nodes(c, i, 0, POS_MIN, SPOS_MAX);
+			if (ret)
+				goto err;
+		}
+
+		*reconstructed_root = true;
+	}
+err:
+fsck_err:
+	printbuf_exit(&buf);
+	return ret;
+}
+
 int bch2_check_topology(struct bch_fs *c)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct bpos pulled_from_scan = POS_MIN;
-	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
 	bch2_trans_srcu_unlock(trans);
 
 	for (unsigned i = 0; i < btree_id_nr_alive(c) && !ret; i++) {
-		struct btree_root *r = bch2_btree_id_root(c, i);
 		bool reconstructed_root = false;
+recover:
+		ret = lockrestart_do(trans, bch2_check_root(trans, i, &reconstructed_root));
+		if (ret)
+			break;
 
-		printbuf_reset(&buf);
-		bch2_btree_id_to_text(&buf, i);
-
-		if (r->error) {
-reconstruct_root:
-			bch_info(c, "btree root %s unreadable, must recover from scan", buf.buf);
-
-			r->alive = false;
-			r->error = 0;
-
-			if (!bch2_btree_has_scanned_nodes(c, i)) {
-				__fsck_err(trans,
-					   FSCK_CAN_FIX|(!btree_id_important(i) ? FSCK_AUTOFIX : 0),
-					   btree_root_unreadable_and_scan_found_nothing,
-					   "no nodes found for btree %s, continue?", buf.buf);
-				bch2_btree_root_alloc_fake_trans(trans, i, 0);
-			} else {
-				bch2_btree_root_alloc_fake_trans(trans, i, 1);
-				bch2_shoot_down_journal_keys(c, i, 1, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
-				ret = bch2_get_scanned_nodes(c, i, 0, POS_MIN, SPOS_MAX);
-				if (ret)
-					break;
-			}
-
-			reconstructed_root = true;
-		}
-
+		struct btree_root *r = bch2_btree_id_root(c, i);
 		struct btree *b = r->b;
 
 		btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
@@ -575,17 +590,21 @@ reconstruct_root:
 
 			r->b = NULL;
 
-			if (!reconstructed_root)
-				goto reconstruct_root;
+			if (!reconstructed_root) {
+				r->error = -EIO;
+				goto recover;
+			}
 
+			struct printbuf buf = PRINTBUF;
+			bch2_btree_id_to_text(&buf, i);
 			bch_err(c, "empty btree root %s", buf.buf);
+			printbuf_exit(&buf);
 			bch2_btree_root_alloc_fake_trans(trans, i, 0);
 			r->alive = false;
 			ret = 0;
 		}
 	}
-fsck_err:
-	printbuf_exit(&buf);
+
 	bch2_trans_put(trans);
 	return ret;
 }
@@ -935,7 +954,7 @@ static int bch2_gc_alloc_start(struct bch_fs *c)
 		ret = genradix_prealloc(&ca->buckets_gc, ca->mi.nbuckets, GFP_KERNEL);
 		if (ret) {
 			bch2_dev_put(ca);
-			ret = -BCH_ERR_ENOMEM_gc_alloc_start;
+			ret = bch_err_throw(c, ENOMEM_gc_alloc_start);
 			break;
 		}
 	}
@@ -1180,7 +1199,7 @@ int bch2_gc_gens(struct bch_fs *c)
 		ca->oldest_gen = kvmalloc(gens->nbuckets, GFP_KERNEL);
 		if (!ca->oldest_gen) {
 			bch2_dev_put(ca);
-			ret = -BCH_ERR_ENOMEM_gc_gens;
+			ret = bch_err_throw(c, ENOMEM_gc_gens);
 			goto err;
 		}
 
diff --git a/libbcachefs/btree_io.c b/libbcachefs/btree_io.c
index 2e191561..57eff301 100644
--- a/libbcachefs/btree_io.c
+++ b/libbcachefs/btree_io.c
@@ -557,7 +557,7 @@ static int __btree_err(int ret,
 		       const char *fmt, ...)
 {
 	if (c->recovery.curr_pass == BCH_RECOVERY_PASS_scan_for_btree_nodes)
-		return -BCH_ERR_fsck_fix;
+		return bch_err_throw(c, fsck_fix);
 
 	bool have_retry = false;
 	int ret2;
@@ -572,9 +572,9 @@ static int __btree_err(int ret,
 	}
 
 	if (!have_retry && ret == -BCH_ERR_btree_node_read_err_want_retry)
-		ret = -BCH_ERR_btree_node_read_err_fixable;
+		ret = bch_err_throw(c, btree_node_read_err_fixable);
 	if (!have_retry && ret == -BCH_ERR_btree_node_read_err_must_retry)
-		ret = -BCH_ERR_btree_node_read_err_bad_node;
+		ret = bch_err_throw(c, btree_node_read_err_bad_node);
 
 	bch2_sb_error_count(c, err_type);
 
@@ -609,7 +609,7 @@ static int __btree_err(int ret,
 			}
 
 			if (!have_retry)
-				ret = -BCH_ERR_fsck_fix;
+				ret = bch_err_throw(c, fsck_fix);
 			goto out;
 		case -BCH_ERR_btree_node_read_err_bad_node:
 			prt_str(&out, ", ");
@@ -638,7 +638,7 @@ static int __btree_err(int ret,
 		}
 
 		if (!have_retry)
-			ret = -BCH_ERR_fsck_fix;
+			ret = bch_err_throw(c, fsck_fix);
 		goto out;
 	case -BCH_ERR_btree_node_read_err_bad_node:
 		prt_str(&out, ", ");
@@ -1687,7 +1687,7 @@ static int btree_node_read_all_replicas(struct bch_fs *c, struct btree *b, bool
 
 	ra = kzalloc(sizeof(*ra), GFP_NOFS);
 	if (!ra)
-		return -BCH_ERR_ENOMEM_btree_node_read_all_replicas;
+		return bch_err_throw(c, ENOMEM_btree_node_read_all_replicas);
 
 	closure_init(&ra->cl, NULL);
 	ra->c	= c;
@@ -1869,7 +1869,7 @@ static int __bch2_btree_root_read(struct btree_trans *trans, enum btree_id id,
 		bch2_btree_node_hash_remove(&c->btree_cache, b);
 		mutex_unlock(&c->btree_cache.lock);
 
-		ret = -BCH_ERR_btree_node_read_error;
+		ret = bch_err_throw(c, btree_node_read_error);
 		goto err;
 	}
 
@@ -2019,7 +2019,7 @@ int bch2_btree_node_scrub(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_btree_node_scrub))
-		return -BCH_ERR_erofs_no_writes;
+		return bch_err_throw(c, erofs_no_writes);
 
 	struct extent_ptr_decoded pick;
 	int ret = bch2_bkey_pick_read_device(c, k, NULL, &pick, dev);
@@ -2029,7 +2029,7 @@ int bch2_btree_node_scrub(struct btree_trans *trans,
 	struct bch_dev *ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ,
 						BCH_DEV_READ_REF_btree_node_scrub);
 	if (!ca) {
-		ret = -BCH_ERR_device_offline;
+		ret = bch_err_throw(c, device_offline);
 		goto err;
 	}
 
@@ -2166,7 +2166,7 @@ static void btree_node_write_work(struct work_struct *work)
 		bch2_dev_list_has_dev(wbio->wbio.failed, ptr->dev));
 
 	if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(&wbio->key))) {
-		ret = -BCH_ERR_btree_node_write_all_failed;
+		ret = bch_err_throw(c, btree_node_write_all_failed);
 		goto err;
 	}
 
diff --git a/libbcachefs/btree_iter.c b/libbcachefs/btree_iter.c
index c7606e0b..b7840337 100644
--- a/libbcachefs/btree_iter.c
+++ b/libbcachefs/btree_iter.c
@@ -938,7 +938,7 @@ static noinline_for_stack int btree_node_missing_err(struct btree_trans *trans,
 
 	bch2_fs_fatal_error(c, "%s", buf.buf);
 	printbuf_exit(&buf);
-	return -BCH_ERR_btree_need_topology_repair;
+	return bch_err_throw(c, btree_need_topology_repair);
 }
 
 static __always_inline int btree_path_down(struct btree_trans *trans,
@@ -1006,7 +1006,7 @@ static int bch2_btree_path_traverse_all(struct btree_trans *trans)
 	int ret = 0;
 
 	if (trans->in_traverse_all)
-		return -BCH_ERR_transaction_restart_in_traverse_all;
+		return bch_err_throw(c, transaction_restart_in_traverse_all);
 
 	trans->in_traverse_all = true;
 retry_all:
diff --git a/libbcachefs/btree_iter.h b/libbcachefs/btree_iter.h
index 2cabb5f0..09dd3e52 100644
--- a/libbcachefs/btree_iter.h
+++ b/libbcachefs/btree_iter.h
@@ -963,16 +963,6 @@ struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_trans *,
 	_p;								\
 })
 
-#define bch2_trans_run(_c, _do)						\
-({									\
-	struct btree_trans *trans = bch2_trans_get(_c);			\
-	int _ret = (_do);						\
-	bch2_trans_put(trans);						\
-	_ret;								\
-})
-
-#define bch2_trans_do(_c, _do)	bch2_trans_run(_c, lockrestart_do(trans, _do))
-
 struct btree_trans *__bch2_trans_get(struct bch_fs *, unsigned);
 void bch2_trans_put(struct btree_trans *);
 
@@ -990,6 +980,27 @@ unsigned bch2_trans_get_fn_idx(const char *);
 	__bch2_trans_get(_c, trans_fn_idx);				\
 })
 
+/*
+ * We don't use DEFINE_CLASS() because using a function for the constructor
+ * breaks bch2_trans_get()'s use of __func__
+ */
+typedef struct btree_trans * class_btree_trans_t;
+static inline void class_btree_trans_destructor(struct btree_trans **p)
+{
+	struct btree_trans *trans = *p;
+	bch2_trans_put(trans);
+}
+
+#define class_btree_trans_constructor(_c)	bch2_trans_get(_c)
+
+#define bch2_trans_run(_c, _do)						\
+({									\
+	CLASS(btree_trans, trans)(_c);					\
+	(_do);								\
+})
+
+#define bch2_trans_do(_c, _do)	bch2_trans_run(_c, lockrestart_do(trans, _do))
+
 void bch2_btree_trans_to_text(struct printbuf *, struct btree_trans *);
 
 void bch2_fs_btree_iter_exit(struct bch_fs *);
diff --git a/libbcachefs/btree_journal_iter.c b/libbcachefs/btree_journal_iter.c
index 48863e69..cf739875 100644
--- a/libbcachefs/btree_journal_iter.c
+++ b/libbcachefs/btree_journal_iter.c
@@ -292,7 +292,7 @@ int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 		if (!new_keys.data) {
 			bch_err(c, "%s: error allocating new key array (size %zu)",
 				__func__, new_keys.size);
-			return -BCH_ERR_ENOMEM_journal_key_insert;
+			return bch_err_throw(c, ENOMEM_journal_key_insert);
 		}
 
 		/* Since @keys was full, there was no gap: */
@@ -331,7 +331,7 @@ int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 
 	n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 	if (!n)
-		return -BCH_ERR_ENOMEM_journal_key_insert;
+		return bch_err_throw(c, ENOMEM_journal_key_insert);
 
 	bkey_copy(n, k);
 	ret = bch2_journal_key_insert_take(c, id, level, n);
@@ -736,7 +736,7 @@ int bch2_journal_keys_sort(struct bch_fs *c)
 				if (keys->nr * 8 > keys->size * 7) {
 					bch_err(c, "Too many journal keys for slowpath; have %zu compacted, buf size %zu, processed %zu keys at seq %llu",
 						keys->nr, keys->size, nr_read, le64_to_cpu(i->j.seq));
-					return -BCH_ERR_ENOMEM_journal_keys_sort;
+					return bch_err_throw(c, ENOMEM_journal_keys_sort);
 				}
 
 				BUG_ON(darray_push(keys, n));
diff --git a/libbcachefs/btree_key_cache.c b/libbcachefs/btree_key_cache.c
index e954b197..d96188b9 100644
--- a/libbcachefs/btree_key_cache.c
+++ b/libbcachefs/btree_key_cache.c
@@ -238,7 +238,7 @@ static int btree_key_cache_create(struct btree_trans *trans,
 		if (unlikely(!ck)) {
 			bch_err(c, "error allocating memory for key cache item, btree %s",
 				bch2_btree_id_str(ck_path->btree_id));
-			return -BCH_ERR_ENOMEM_btree_key_cache_create;
+			return bch_err_throw(c, ENOMEM_btree_key_cache_create);
 		}
 	}
 
@@ -256,7 +256,7 @@ static int btree_key_cache_create(struct btree_trans *trans,
 		if (unlikely(!new_k)) {
 			bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
 				bch2_btree_id_str(ck->key.btree_id), key_u64s);
-			ret = -BCH_ERR_ENOMEM_btree_key_cache_fill;
+			ret = bch_err_throw(c, ENOMEM_btree_key_cache_fill);
 		} else if (ret) {
 			kfree(new_k);
 			goto err;
@@ -822,20 +822,20 @@ int bch2_fs_btree_key_cache_init(struct btree_key_cache *bc)
 
 	bc->nr_pending = alloc_percpu(size_t);
 	if (!bc->nr_pending)
-		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+		return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 
 	if (rcu_pending_init(&bc->pending[0], &c->btree_trans_barrier, __bkey_cached_free) ||
 	    rcu_pending_init(&bc->pending[1], &c->btree_trans_barrier, __bkey_cached_free))
-		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+		return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 
 	if (rhashtable_init(&bc->table, &bch2_btree_key_cache_params))
-		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+		return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 
 	bc->table_init_done = true;
 
 	shrink = shrinker_alloc(0, "%s-btree_key_cache", c->name);
 	if (!shrink)
-		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
+		return bch_err_throw(c, ENOMEM_fs_btree_cache_init);
 	bc->shrink = shrink;
 	shrink->count_objects	= bch2_btree_key_cache_count;
 	shrink->scan_objects	= bch2_btree_key_cache_scan;
diff --git a/libbcachefs/btree_node_scan.c b/libbcachefs/btree_node_scan.c
index adaa3913..5a97a6b8 100644
--- a/libbcachefs/btree_node_scan.c
+++ b/libbcachefs/btree_node_scan.c
@@ -122,55 +122,22 @@ static int found_btree_node_cmp_cookie(const void *_l, const void *_r)
  * Given two found btree nodes, if their sequence numbers are equal, take the
  * one that's readable:
  */
-static int found_btree_node_cmp_time(struct bch_fs *c,
-				     struct found_btree_node *l,
-				     struct found_btree_node *r)
+static int found_btree_node_cmp_time(const struct found_btree_node *l,
+				     const struct found_btree_node *r)
 {
-	int cmp1 = cmp_int(l->seq, r->seq);
-	int cmp2 = l->journal_seq && r->journal_seq
-		? cmp_int(l->journal_seq, r->journal_seq)
-		: 0;
-	int cmp = cmp2 ?: cmp1;
-
-	if (!cmp || (cmp1 && cmp2 && cmp1 != cmp2)) {
-		struct printbuf buf = PRINTBUF;
-		bch2_log_msg_start(c, &buf);
-
-		if (cmp)
-			prt_printf(&buf, "found btree nodes in scan where seq, journal seq disagree on node age\n");
-		else
-			prt_printf(&buf, "found btree nodes where we don't know which is newer\n");
-		found_btree_node_to_text(&buf, c, l);
-		found_btree_node_to_text(&buf, c, r);
-
-		bch2_print_str(c, KERN_ERR, buf.buf);
-		printbuf_exit(&buf);
-	}
-
-	return cmp2 ?: cmp1;
-}
-
-static int found_btree_node_cmp_time_nowarn(const struct found_btree_node *l,
-					    const struct found_btree_node *r)
-{
-	int cmp1 = cmp_int(l->seq, r->seq);
-	int cmp2 = l->journal_seq && r->journal_seq
-		? cmp_int(l->journal_seq, r->journal_seq)
-		: 0;
-	return cmp2 ?: cmp1;
-
-	return cmp2 ?: cmp1;
+	return  cmp_int(l->seq, r->seq) ?:
+		cmp_int(l->journal_seq, r->journal_seq);
 }
 
 static int found_btree_node_cmp_pos(const void *_l, const void *_r)
 {
-	const struct found_btree_node *l = (void *) _l;
-	const struct found_btree_node *r = (void *) _r;
+	const struct found_btree_node *l = _l;
+	const struct found_btree_node *r = _r;
 
 	return  cmp_int(l->btree_id,	r->btree_id) ?:
 	       -cmp_int(l->level,	r->level) ?:
 		bpos_cmp(l->min_key,	r->min_key) ?:
-	       -found_btree_node_cmp_time_nowarn(l, r);
+	       -found_btree_node_cmp_time(l, r);
 }
 
 static inline bool found_btree_node_cmp_pos_less(const void *l, const void *r, void *arg)
@@ -368,7 +335,7 @@ static int handle_overwrites(struct bch_fs *c,
 
 	while ((r = min_heap_peek(nodes_heap)) &&
 	       nodes_overlap(l, r)) {
-		int cmp = found_btree_node_cmp_time(c, l, r);
+		int cmp = found_btree_node_cmp_time(l, r);
 
 		if (cmp > 0) {
 			if (bpos_cmp(l->max_key, r->max_key) >= 0)
diff --git a/libbcachefs/btree_trans_commit.c b/libbcachefs/btree_trans_commit.c
index f2d1edc9..d9710801 100644
--- a/libbcachefs/btree_trans_commit.c
+++ b/libbcachefs/btree_trans_commit.c
@@ -376,7 +376,7 @@ static inline int btree_key_can_insert(struct btree_trans *trans,
 				       struct btree *b, unsigned u64s)
 {
 	if (!bch2_btree_node_insert_fits(b, u64s))
-		return -BCH_ERR_btree_insert_btree_node_full;
+		return bch_err_throw(trans->c, btree_insert_btree_node_full);
 
 	return 0;
 }
@@ -394,9 +394,10 @@ btree_key_can_insert_cached_slowpath(struct btree_trans *trans, unsigned flags,
 
 	new_k = kmalloc(new_u64s * sizeof(u64), GFP_KERNEL);
 	if (!new_k) {
-		bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
+		struct bch_fs *c = trans->c;
+		bch_err(c, "error allocating memory for key cache key, btree %s u64s %u",
 			bch2_btree_id_str(path->btree_id), new_u64s);
-		return -BCH_ERR_ENOMEM_btree_key_cache_insert;
+		return bch_err_throw(c, ENOMEM_btree_key_cache_insert);
 	}
 
 	ret =   bch2_trans_relock(trans) ?:
@@ -432,7 +433,7 @@ static int btree_key_can_insert_cached(struct btree_trans *trans, unsigned flags
 	if (watermark < BCH_WATERMARK_reclaim &&
 	    !test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
 	    bch2_btree_key_cache_must_wait(c))
-		return -BCH_ERR_btree_insert_need_journal_reclaim;
+		return bch_err_throw(c, btree_insert_need_journal_reclaim);
 
 	/*
 	 * bch2_varint_decode can read past the end of the buffer by at most 7
@@ -894,7 +895,7 @@ int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
 		 */
 		if ((flags & BCH_TRANS_COMMIT_journal_reclaim) &&
 		    watermark < BCH_WATERMARK_reclaim) {
-			ret = -BCH_ERR_journal_reclaim_would_deadlock;
+			ret = bch_err_throw(c, journal_reclaim_would_deadlock);
 			goto out;
 		}
 
@@ -1024,7 +1025,7 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 		if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags)))
 			ret = do_bch2_trans_commit_to_journal_replay(trans);
 		else
-			ret = -BCH_ERR_erofs_trans_commit;
+			ret = bch_err_throw(c, erofs_trans_commit);
 		goto out_reset;
 	}
 
@@ -1106,7 +1107,7 @@ err:
 	 * restart:
 	 */
 	if (flags & BCH_TRANS_COMMIT_no_journal_res) {
-		ret = -BCH_ERR_transaction_restart_nested;
+		ret = bch_err_throw(c, transaction_restart_nested);
 		goto out;
 	}
 
diff --git a/libbcachefs/btree_update.c b/libbcachefs/btree_update.c
index e04508da..e97e78c1 100644
--- a/libbcachefs/btree_update.c
+++ b/libbcachefs/btree_update.c
@@ -587,7 +587,7 @@ int bch2_bkey_get_empty_slot(struct btree_trans *trans, struct btree_iter *iter,
 	BUG_ON(k.k->type != KEY_TYPE_deleted);
 
 	if (bkey_gt(k.k->p, end)) {
-		ret = -BCH_ERR_ENOSPC_btree_slot;
+		ret = bch_err_throw(trans->c, ENOSPC_btree_slot);
 		goto err;
 	}
 
diff --git a/libbcachefs/btree_update_interior.c b/libbcachefs/btree_update_interior.c
index 647b40ef..d2ecb782 100644
--- a/libbcachefs/btree_update_interior.c
+++ b/libbcachefs/btree_update_interior.c
@@ -684,12 +684,31 @@ static void btree_update_nodes_written(struct btree_update *as)
 
 	/*
 	 * Wait for any in flight writes to finish before we free the old nodes
-	 * on disk:
+	 * on disk. But we haven't pinned those old nodes in the btree cache,
+	 * they might have already been evicted.
+	 *
+	 * The update we're completing deleted references to those nodes from the
+	 * btree, so we know if they've been evicted they can't be pulled back in.
+	 * We just have to check if the nodes we have pointers to are still those
+	 * old nodes, and haven't been reused.
+	 *
+	 * This can't be done locklessly because the data buffer might have been
+	 * vmalloc allocated, and they're not RCU freed. We also need the
+	 * __no_kmsan_checks annotation because even with the btree node read
+	 * lock, nothing tells us that the data buffer has been initialized (if
+	 * the btree node has been reused for a different node, and the data
+	 * buffer swapped for a new data buffer).
 	 */
 	for (i = 0; i < as->nr_old_nodes; i++) {
 		b = as->old_nodes[i];
 
-		if (btree_node_seq_matches(b, as->old_nodes_seq[i]))
+		bch2_trans_begin(trans);
+		btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
+		bool seq_matches = btree_node_seq_matches(b, as->old_nodes_seq[i]);
+		six_unlock_read(&b->c.lock);
+		bch2_trans_unlock_long(trans);
+
+		if (seq_matches)
 			wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight_inner,
 				       TASK_UNINTERRUPTIBLE);
 	}
@@ -1244,7 +1263,7 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 		if (bch2_err_matches(ret, ENOSPC) &&
 		    (flags & BCH_TRANS_COMMIT_journal_reclaim) &&
 		    watermark < BCH_WATERMARK_reclaim) {
-			ret = -BCH_ERR_journal_reclaim_would_deadlock;
+			ret = bch_err_throw(c, journal_reclaim_would_deadlock);
 			goto err;
 		}
 
@@ -2177,7 +2196,7 @@ static int get_iter_to_node(struct btree_trans *trans, struct btree_iter *iter,
 	if (btree_iter_path(trans, iter)->l[b->c.level].b != b) {
 		/* node has been freed: */
 		BUG_ON(!btree_node_dying(b));
-		ret = -BCH_ERR_btree_node_dying;
+		ret = bch_err_throw(trans->c, btree_node_dying);
 		goto err;
 	}
 
@@ -2791,16 +2810,16 @@ int bch2_fs_btree_interior_update_init(struct bch_fs *c)
 	c->btree_interior_update_worker =
 		alloc_workqueue("btree_update", WQ_UNBOUND|WQ_MEM_RECLAIM, 8);
 	if (!c->btree_interior_update_worker)
-		return -BCH_ERR_ENOMEM_btree_interior_update_worker_init;
+		return bch_err_throw(c, ENOMEM_btree_interior_update_worker_init);
 
 	c->btree_node_rewrite_worker =
 		alloc_ordered_workqueue("btree_node_rewrite", WQ_UNBOUND);
 	if (!c->btree_node_rewrite_worker)
-		return -BCH_ERR_ENOMEM_btree_interior_update_worker_init;
+		return bch_err_throw(c, ENOMEM_btree_interior_update_worker_init);
 
 	if (mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
 				      sizeof(struct btree_update)))
-		return -BCH_ERR_ENOMEM_btree_interior_update_pool_init;
+		return bch_err_throw(c, ENOMEM_btree_interior_update_pool_init);
 
 	return 0;
 }
diff --git a/libbcachefs/btree_write_buffer.c b/libbcachefs/btree_write_buffer.c
index efb0c64d..90b21e61 100644
--- a/libbcachefs/btree_write_buffer.c
+++ b/libbcachefs/btree_write_buffer.c
@@ -394,7 +394,7 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 		bool accounting_accumulated = false;
 		do {
 			if (race_fault()) {
-				ret = -BCH_ERR_journal_reclaim_would_deadlock;
+				ret = bch_err_throw(c, journal_reclaim_would_deadlock);
 				break;
 			}
 
@@ -633,7 +633,7 @@ int bch2_btree_write_buffer_tryflush(struct btree_trans *trans)
 	struct bch_fs *c = trans->c;
 
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_btree_write_buffer))
-		return -BCH_ERR_erofs_no_writes;
+		return bch_err_throw(c, erofs_no_writes);
 
 	int ret = bch2_btree_write_buffer_flush_nocheck_rw(trans);
 	enumerated_ref_put(&c->writes, BCH_WRITE_REF_btree_write_buffer);
@@ -676,7 +676,7 @@ int bch2_btree_write_buffer_maybe_flush(struct btree_trans *trans,
 			goto err;
 
 		bch2_bkey_buf_copy(last_flushed, c, tmp.k);
-		ret = -BCH_ERR_transaction_restart_write_buffer_flush;
+		ret = bch_err_throw(c, transaction_restart_write_buffer_flush);
 	}
 err:
 	bch2_bkey_buf_exit(&tmp, c);
diff --git a/libbcachefs/buckets.c b/libbcachefs/buckets.c
index 56bd8f66..f25903c1 100644
--- a/libbcachefs/buckets.c
+++ b/libbcachefs/buckets.c
@@ -224,7 +224,7 @@ static int bch2_check_fix_ptr(struct btree_trans *trans,
 			switch (g->data_type) {
 			case BCH_DATA_sb:
 				bch_err(c, "btree and superblock in the same bucket - cannot repair");
-				ret = -BCH_ERR_fsck_repair_unimplemented;
+				ret = bch_err_throw(c, fsck_repair_unimplemented);
 				goto out;
 			case BCH_DATA_journal:
 				ret = bch2_dev_journal_bucket_delete(ca, PTR_BUCKET_NR(ca, &p.ptr));
@@ -440,7 +440,7 @@ static int bucket_ref_update_err(struct btree_trans *trans, struct printbuf *buf
 		 * us an error code for rewinding recovery
 		 */
 		if (!ret)
-			ret = -BCH_ERR_bucket_ref_update;
+			ret = bch_err_throw(c, bucket_ref_update);
 	} else {
 		/* Always ignore overwrite errors, so that deletion works */
 		ret = 0;
@@ -632,7 +632,7 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 	struct bch_dev *ca = bch2_dev_tryget(c, p.ptr.dev);
 	if (unlikely(!ca)) {
 		if (insert && p.ptr.dev != BCH_SB_MEMBER_INVALID)
-			ret = -BCH_ERR_trigger_pointer;
+			ret = bch_err_throw(c, trigger_pointer);
 		goto err;
 	}
 
@@ -640,7 +640,7 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 	if (!bucket_valid(ca, bucket.offset)) {
 		if (insert) {
 			bch2_dev_bucket_missing(ca, bucket.offset);
-			ret = -BCH_ERR_trigger_pointer;
+			ret = bch_err_throw(c, trigger_pointer);
 		}
 		goto err;
 	}
@@ -662,7 +662,7 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 		if (bch2_fs_inconsistent_on(!g, c, "reference to invalid bucket on device %u\n  %s",
 					    p.ptr.dev,
 					    (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-			ret = -BCH_ERR_trigger_pointer;
+			ret = bch_err_throw(c, trigger_pointer);
 			goto err;
 		}
 
@@ -688,6 +688,8 @@ static int bch2_trigger_stripe_ptr(struct btree_trans *trans,
 				s64 sectors,
 				enum btree_iter_update_trigger_flags flags)
 {
+	struct bch_fs *c = trans->c;
+
 	if (flags & BTREE_TRIGGER_transactional) {
 		struct btree_iter iter;
 		struct bkey_i_stripe *s = bch2_bkey_get_mut_typed(trans, &iter,
@@ -705,7 +707,7 @@ static int bch2_trigger_stripe_ptr(struct btree_trans *trans,
 			bch2_trans_inconsistent(trans,
 				"stripe pointer doesn't match stripe %llu",
 				(u64) p.ec.idx);
-			ret = -BCH_ERR_trigger_stripe_pointer;
+			ret = bch_err_throw(c, trigger_stripe_pointer);
 			goto err;
 		}
 
@@ -725,13 +727,11 @@ err:
 	}
 
 	if (flags & BTREE_TRIGGER_gc) {
-		struct bch_fs *c = trans->c;
-
 		struct gc_stripe *m = genradix_ptr_alloc(&c->gc_stripes, p.ec.idx, GFP_KERNEL);
 		if (!m) {
 			bch_err(c, "error allocating memory for gc_stripes, idx %llu",
 				(u64) p.ec.idx);
-			return -BCH_ERR_ENOMEM_mark_stripe_ptr;
+			return bch_err_throw(c, ENOMEM_mark_stripe_ptr);
 		}
 
 		gc_stripe_lock(m);
@@ -746,7 +746,7 @@ err:
 			__bch2_inconsistent_error(c, &buf);
 			bch2_print_str(c, KERN_ERR, buf.buf);
 			printbuf_exit(&buf);
-			return -BCH_ERR_trigger_stripe_pointer;
+			return bch_err_throw(c, trigger_stripe_pointer);
 		}
 
 		m->block_sectors[p.ec.block] += sectors;
@@ -1014,7 +1014,7 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
 		bch2_print_str(c, KERN_ERR, buf.buf);
 		printbuf_exit(&buf);
 		if (!ret)
-			ret = -BCH_ERR_metadata_bucket_inconsistency;
+			ret = bch_err_throw(c, metadata_bucket_inconsistency);
 		goto err;
 	}
 
@@ -1067,7 +1067,7 @@ static int bch2_mark_metadata_bucket(struct btree_trans *trans, struct bch_dev *
 err_unlock:
 	bucket_unlock(g);
 err:
-	return -BCH_ERR_metadata_bucket_inconsistency;
+	return bch_err_throw(c, metadata_bucket_inconsistency);
 }
 
 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
@@ -1282,7 +1282,7 @@ recalculate:
 		ret = 0;
 	} else {
 		atomic64_set(&c->sectors_available, sectors_available);
-		ret = -BCH_ERR_ENOSPC_disk_reservation;
+		ret = bch_err_throw(c, ENOSPC_disk_reservation);
 	}
 
 	mutex_unlock(&c->sectors_available_lock);
@@ -1311,7 +1311,7 @@ int bch2_buckets_nouse_alloc(struct bch_fs *c)
 					    GFP_KERNEL|__GFP_ZERO);
 		if (!ca->buckets_nouse) {
 			bch2_dev_put(ca);
-			return -BCH_ERR_ENOMEM_buckets_nouse;
+			return bch_err_throw(c, ENOMEM_buckets_nouse);
 		}
 	}
 
@@ -1336,12 +1336,12 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 		lockdep_assert_held(&c->state_lock);
 
 	if (resize && ca->buckets_nouse)
-		return -BCH_ERR_no_resize_with_buckets_nouse;
+		return bch_err_throw(c, no_resize_with_buckets_nouse);
 
 	bucket_gens = bch2_kvmalloc(struct_size(bucket_gens, b, nbuckets),
 				    GFP_KERNEL|__GFP_ZERO);
 	if (!bucket_gens) {
-		ret = -BCH_ERR_ENOMEM_bucket_gens;
+		ret = bch_err_throw(c, ENOMEM_bucket_gens);
 		goto err;
 	}
 
@@ -1360,9 +1360,9 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 		       sizeof(bucket_gens->b[0]) * copy);
 	}
 
-	ret =   bch2_bucket_bitmap_resize(&ca->bucket_backpointer_mismatch,
+	ret =   bch2_bucket_bitmap_resize(ca, &ca->bucket_backpointer_mismatch,
 					  ca->mi.nbuckets, nbuckets) ?:
-		bch2_bucket_bitmap_resize(&ca->bucket_backpointer_empty,
+		bch2_bucket_bitmap_resize(ca, &ca->bucket_backpointer_empty,
 					  ca->mi.nbuckets, nbuckets);
 
 	rcu_assign_pointer(ca->bucket_gens, bucket_gens);
@@ -1389,7 +1389,7 @@ int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
 {
 	ca->usage = alloc_percpu(struct bch_dev_usage_full);
 	if (!ca->usage)
-		return -BCH_ERR_ENOMEM_usage_init;
+		return bch_err_throw(c, ENOMEM_usage_init);
 
 	return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);
 }
diff --git a/libbcachefs/buckets_waiting_for_journal.c b/libbcachefs/buckets_waiting_for_journal.c
index c8a488e6..832eff93 100644
--- a/libbcachefs/buckets_waiting_for_journal.c
+++ b/libbcachefs/buckets_waiting_for_journal.c
@@ -108,7 +108,8 @@ int bch2_set_bucket_needs_journal_commit(struct buckets_waiting_for_journal *b,
 realloc:
 	n = kvmalloc(sizeof(*n) + (sizeof(n->d[0]) << new_bits), GFP_KERNEL);
 	if (!n) {
-		ret = -BCH_ERR_ENOMEM_buckets_waiting_for_journal_set;
+		struct bch_fs *c = container_of(b, struct bch_fs, buckets_waiting_for_journal);
+		ret = bch_err_throw(c, ENOMEM_buckets_waiting_for_journal_set);
 		goto out;
 	}
 
diff --git a/libbcachefs/chardev.c b/libbcachefs/chardev.c
index 2db30eb1..2d38466e 100644
--- a/libbcachefs/chardev.c
+++ b/libbcachefs/chardev.c
@@ -618,7 +618,7 @@ static long bch2_ioctl_disk_get_idx(struct bch_fs *c,
 		if (ca->dev == dev)
 			return ca->dev_idx;
 
-	return -BCH_ERR_ENOENT_dev_idx_not_found;
+	return bch_err_throw(c, ENOENT_dev_idx_not_found);
 }
 
 static long bch2_ioctl_disk_resize(struct bch_fs *c,
diff --git a/libbcachefs/checksum.c b/libbcachefs/checksum.c
index ff5ab8ad..3b5dc123 100644
--- a/libbcachefs/checksum.c
+++ b/libbcachefs/checksum.c
@@ -173,7 +173,7 @@ int bch2_encrypt(struct bch_fs *c, unsigned type,
 
 	if (bch2_fs_inconsistent_on(!c->chacha20_key_set,
 				    c, "attempting to encrypt without encryption key"))
-		return -BCH_ERR_no_encryption_key;
+		return bch_err_throw(c, no_encryption_key);
 
 	bch2_chacha20(&c->chacha20_key, nonce, data, len);
 	return 0;
@@ -262,7 +262,7 @@ int __bch2_encrypt_bio(struct bch_fs *c, unsigned type,
 
 	if (bch2_fs_inconsistent_on(!c->chacha20_key_set,
 				    c, "attempting to encrypt without encryption key"))
-		return -BCH_ERR_no_encryption_key;
+		return bch_err_throw(c, no_encryption_key);
 
 	bch2_chacha20_init(chacha_state, &c->chacha20_key, nonce);
 
@@ -375,7 +375,7 @@ int bch2_rechecksum_bio(struct bch_fs *c, struct bio *bio,
 		prt_str(&buf, ")");
 		WARN_RATELIMIT(1, "%s", buf.buf);
 		printbuf_exit(&buf);
-		return -BCH_ERR_recompute_checksum;
+		return bch_err_throw(c, recompute_checksum);
 	}
 
 	for (i = splits; i < splits + ARRAY_SIZE(splits); i++) {
@@ -659,7 +659,7 @@ int bch2_enable_encryption(struct bch_fs *c, bool keyed)
 	crypt = bch2_sb_field_resize(&c->disk_sb, crypt,
 				     sizeof(*crypt) / sizeof(u64));
 	if (!crypt) {
-		ret = -BCH_ERR_ENOSPC_sb_crypt;
+		ret = bch_err_throw(c, ENOSPC_sb_crypt);
 		goto err;
 	}
 
diff --git a/libbcachefs/compress.c b/libbcachefs/compress.c
index 1bca61d1..b37b1f32 100644
--- a/libbcachefs/compress.c
+++ b/libbcachefs/compress.c
@@ -187,7 +187,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 			     __bch2_compression_types[crc.compression_type]))
 			ret = bch2_check_set_has_compressed_data(c, opt);
 		else
-			ret = -BCH_ERR_compression_workspace_not_initialized;
+			ret = bch_err_throw(c, compression_workspace_not_initialized);
 		if (ret)
 			goto err;
 	}
@@ -200,7 +200,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 		ret2 = LZ4_decompress_safe_partial(src_data.b, dst_data,
 						   src_len, dst_len, dst_len);
 		if (ret2 != dst_len)
-			ret = -BCH_ERR_decompress_lz4;
+			ret = bch_err_throw(c, decompress_lz4);
 		break;
 	case BCH_COMPRESSION_TYPE_gzip: {
 		z_stream strm = {
@@ -219,7 +219,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 		mempool_free(workspace, workspace_pool);
 
 		if (ret2 != Z_STREAM_END)
-			ret = -BCH_ERR_decompress_gzip;
+			ret = bch_err_throw(c, decompress_gzip);
 		break;
 	}
 	case BCH_COMPRESSION_TYPE_zstd: {
@@ -227,7 +227,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 		size_t real_src_len = le32_to_cpup(src_data.b);
 
 		if (real_src_len > src_len - 4) {
-			ret = -BCH_ERR_decompress_zstd_src_len_bad;
+			ret = bch_err_throw(c, decompress_zstd_src_len_bad);
 			goto err;
 		}
 
@@ -241,7 +241,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 		mempool_free(workspace, workspace_pool);
 
 		if (ret2 != dst_len)
-			ret = -BCH_ERR_decompress_zstd;
+			ret = bch_err_throw(c, decompress_zstd);
 		break;
 	}
 	default:
@@ -270,7 +270,7 @@ int bch2_bio_uncompress_inplace(struct bch_write_op *op,
 		bch2_write_op_error(op, op->pos.offset,
 				    "extent too big to decompress (%u > %u)",
 				    crc->uncompressed_size << 9, c->opts.encoded_extent_max);
-		return -BCH_ERR_decompress_exceeded_max_encoded_extent;
+		return bch_err_throw(c, decompress_exceeded_max_encoded_extent);
 	}
 
 	data = __bounce_alloc(c, dst_len, WRITE);
@@ -314,7 +314,7 @@ int bch2_bio_uncompress(struct bch_fs *c, struct bio *src,
 
 	if (crc.uncompressed_size << 9	> c->opts.encoded_extent_max ||
 	    crc.compressed_size << 9	> c->opts.encoded_extent_max)
-		return -BCH_ERR_decompress_exceeded_max_encoded_extent;
+		return bch_err_throw(c, decompress_exceeded_max_encoded_extent);
 
 	dst_data = dst_len == dst_iter.bi_size
 		? __bio_map_or_bounce(c, dst, dst_iter, WRITE)
@@ -656,12 +656,12 @@ static int __bch2_fs_compress_init(struct bch_fs *c, u64 features)
 	if (!mempool_initialized(&c->compression_bounce[READ]) &&
 	    mempool_init_kvmalloc_pool(&c->compression_bounce[READ],
 				       1, c->opts.encoded_extent_max))
-		return -BCH_ERR_ENOMEM_compression_bounce_read_init;
+		return bch_err_throw(c, ENOMEM_compression_bounce_read_init);
 
 	if (!mempool_initialized(&c->compression_bounce[WRITE]) &&
 	    mempool_init_kvmalloc_pool(&c->compression_bounce[WRITE],
 				       1, c->opts.encoded_extent_max))
-		return -BCH_ERR_ENOMEM_compression_bounce_write_init;
+		return bch_err_throw(c, ENOMEM_compression_bounce_write_init);
 
 	for (i = compression_types;
 	     i < compression_types + ARRAY_SIZE(compression_types);
@@ -675,7 +675,7 @@ static int __bch2_fs_compress_init(struct bch_fs *c, u64 features)
 		if (mempool_init_kvmalloc_pool(
 				&c->compress_workspace[i->type],
 				1, i->compress_workspace))
-			return -BCH_ERR_ENOMEM_compression_workspace_init;
+			return bch_err_throw(c, ENOMEM_compression_workspace_init);
 	}
 
 	return 0;
diff --git a/libbcachefs/data_update.c b/libbcachefs/data_update.c
index 5c687ed1..5f117434 100644
--- a/libbcachefs/data_update.c
+++ b/libbcachefs/data_update.c
@@ -255,7 +255,7 @@ static int data_update_invalid_bkey(struct data_update *m,
 	bch2_print_str(c, KERN_ERR, buf.buf);
 	printbuf_exit(&buf);
 
-	return -BCH_ERR_invalid_bkey;
+	return bch_err_throw(c, invalid_bkey);
 }
 
 static int __bch2_data_update_index_update(struct btree_trans *trans,
@@ -772,7 +772,7 @@ static int can_write_extent(struct bch_fs *c, struct data_update *m)
 {
 	if ((m->op.flags & BCH_WRITE_alloc_nowait) &&
 	    unlikely(c->open_buckets_nr_free <= bch2_open_buckets_reserved(m->op.watermark)))
-		return -BCH_ERR_data_update_done_would_block;
+		return bch_err_throw(c, data_update_done_would_block);
 
 	unsigned target = m->op.flags & BCH_WRITE_only_specified_devs
 		? m->op.target
@@ -802,9 +802,9 @@ static int can_write_extent(struct bch_fs *c, struct data_update *m)
 	}
 
 	if (!nr_replicas)
-		return -BCH_ERR_data_update_done_no_rw_devs;
+		return bch_err_throw(c, data_update_done_no_rw_devs);
 	if (nr_replicas < m->op.nr_replicas)
-		return -BCH_ERR_insufficient_devices;
+		return bch_err_throw(c, insufficient_devices);
 	return 0;
 }
 
@@ -822,21 +822,16 @@ int bch2_data_update_init(struct btree_trans *trans,
 	int ret = 0;
 
 	if (k.k->p.snapshot) {
-		/*
-		 * We'll go ERO if we see a key for a missing snapshot, and if
-		 * we're still in recovery we want to give that a chance to
-		 * repair:
-		 */
-		if (unlikely(test_bit(BCH_FS_in_recovery, &c->flags) &&
-			     bch2_snapshot_id_state(c, k.k->p.snapshot) == SNAPSHOT_ID_empty))
-			return -BCH_ERR_data_update_done_no_snapshot;
-
 		ret = bch2_check_key_has_snapshot(trans, iter, k);
+		if (bch2_err_matches(ret, BCH_ERR_recovery_will_run)) {
+			/* Can't repair yet, waiting on other recovery passes */
+			return bch_err_throw(c, data_update_done_no_snapshot);
+		}
 		if (ret < 0)
 			return ret;
 		if (ret) /* key was deleted */
 			return bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
-				-BCH_ERR_data_update_done_no_snapshot;
+				bch_err_throw(c, data_update_done_no_snapshot);
 		ret = 0;
 	}
 
@@ -943,7 +938,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 		if (iter)
 			ret = bch2_extent_drop_ptrs(trans, iter, k, io_opts, &m->data_opts);
 		if (!ret)
-			ret = -BCH_ERR_data_update_done_no_writes_needed;
+			ret = bch_err_throw(c, data_update_done_no_writes_needed);
 		goto out_bkey_buf_exit;
 	}
 
@@ -974,19 +969,19 @@ int bch2_data_update_init(struct btree_trans *trans,
 	}
 
 	if (!bkey_get_dev_refs(c, k)) {
-		ret = -BCH_ERR_data_update_done_no_dev_refs;
+		ret = bch_err_throw(c, data_update_done_no_dev_refs);
 		goto out_put_disk_res;
 	}
 
 	if (c->opts.nocow_enabled &&
 	    !bkey_nocow_lock(c, ctxt, ptrs)) {
-		ret = -BCH_ERR_nocow_lock_blocked;
+		ret = bch_err_throw(c, nocow_lock_blocked);
 		goto out_put_dev_refs;
 	}
 
 	if (unwritten) {
 		ret = bch2_update_unwritten_extent(trans, m) ?:
-			-BCH_ERR_data_update_done_unwritten;
+			bch_err_throw(c, data_update_done_unwritten);
 		goto out_nocow_unlock;
 	}
 
diff --git a/libbcachefs/dirent.c b/libbcachefs/dirent.c
index 06613414..1760e5d0 100644
--- a/libbcachefs/dirent.c
+++ b/libbcachefs/dirent.c
@@ -231,70 +231,60 @@ void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c
 	prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
 }
 
-static struct bkey_i_dirent *dirent_alloc_key(struct btree_trans *trans,
-				subvol_inum dir,
-				u8 type,
-				int name_len, int cf_name_len,
-				u64 dst)
+int bch2_dirent_init_name(struct bkey_i_dirent *dirent,
+			  const struct bch_hash_info *hash_info,
+			  const struct qstr *name,
+			  const struct qstr *cf_name)
 {
-	struct bkey_i_dirent *dirent;
-	unsigned u64s = BKEY_U64s + dirent_val_u64s(name_len, cf_name_len);
-
-	BUG_ON(u64s > U8_MAX);
+	EBUG_ON(hash_info->cf_encoding == NULL && cf_name);
+	int cf_len = 0;
 
-	dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
-	if (IS_ERR(dirent))
-		return dirent;
+	if (name->len > BCH_NAME_MAX)
+		return -ENAMETOOLONG;
 
-	bkey_dirent_init(&dirent->k_i);
-	dirent->k.u64s = u64s;
+	dirent->v.d_casefold = hash_info->cf_encoding != NULL;
 
-	if (type != DT_SUBVOL) {
-		dirent->v.d_inum = cpu_to_le64(dst);
+	if (!dirent->v.d_casefold) {
+		memcpy(&dirent->v.d_name[0], name->name, name->len);
+		memset(&dirent->v.d_name[name->len], 0,
+		       bkey_val_bytes(&dirent->k) -
+		       offsetof(struct bch_dirent, d_name) -
+		       name->len);
 	} else {
-		dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
-		dirent->v.d_child_subvol = cpu_to_le32(dst);
-	}
+		memcpy(&dirent->v.d_cf_name_block.d_names[0], name->name, name->len);
 
-	dirent->v.d_type = type;
-	dirent->v.d_unused = 0;
-	dirent->v.d_casefold = cf_name_len ? 1 : 0;
+		char *cf_out = &dirent->v.d_cf_name_block.d_names[name->len];
 
-	return dirent;
-}
+		if (cf_name) {
+			cf_len = cf_name->len;
 
-static void dirent_init_regular_name(struct bkey_i_dirent *dirent,
-				     const struct qstr *name)
-{
-	EBUG_ON(dirent->v.d_casefold);
+			memcpy(cf_out, cf_name->name, cf_name->len);
+		} else {
+			cf_len = utf8_casefold(hash_info->cf_encoding, name,
+					       cf_out,
+					       bkey_val_end(bkey_i_to_s(&dirent->k_i)) - (void *) cf_out);
+			if (cf_len <= 0)
+				return cf_len;
+		}
 
-	memcpy(&dirent->v.d_name[0], name->name, name->len);
-	memset(&dirent->v.d_name[name->len], 0,
-		bkey_val_bytes(&dirent->k) -
-		offsetof(struct bch_dirent, d_name) -
-		name->len);
-}
+		memset(&dirent->v.d_cf_name_block.d_names[name->len + cf_len], 0,
+		       bkey_val_bytes(&dirent->k) -
+		       offsetof(struct bch_dirent, d_cf_name_block.d_names) -
+		       name->len + cf_len);
 
-static void dirent_init_casefolded_name(struct bkey_i_dirent *dirent,
-					const struct qstr *name,
-					const struct qstr *cf_name)
-{
-	EBUG_ON(!dirent->v.d_casefold);
-	EBUG_ON(!cf_name->len);
-
-	dirent->v.d_cf_name_block.d_name_len = cpu_to_le16(name->len);
-	dirent->v.d_cf_name_block.d_cf_name_len = cpu_to_le16(cf_name->len);
-	memcpy(&dirent->v.d_cf_name_block.d_names[0], name->name, name->len);
-	memcpy(&dirent->v.d_cf_name_block.d_names[name->len], cf_name->name, cf_name->len);
-	memset(&dirent->v.d_cf_name_block.d_names[name->len + cf_name->len], 0,
-		bkey_val_bytes(&dirent->k) -
-		offsetof(struct bch_dirent, d_cf_name_block.d_names) -
-		name->len + cf_name->len);
-
-	EBUG_ON(bch2_dirent_get_casefold_name(dirent_i_to_s_c(dirent)).len != cf_name->len);
+		dirent->v.d_cf_name_block.d_name_len = cpu_to_le16(name->len);
+		dirent->v.d_cf_name_block.d_cf_name_len = cpu_to_le16(cf_len);
+
+		EBUG_ON(bch2_dirent_get_casefold_name(dirent_i_to_s_c(dirent)).len != cf_len);
+	}
+
+	unsigned u64s = dirent_val_u64s(name->len, cf_len);
+	BUG_ON(u64s > bkey_val_u64s(&dirent->k));
+	set_bkey_val_u64s(&dirent->k, u64s);
+	return 0;
 }
 
-static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
+struct bkey_i_dirent *bch2_dirent_create_key(struct btree_trans *trans,
 				const struct bch_hash_info *hash_info,
 				subvol_inum dir,
 				u8 type,
@@ -302,31 +292,28 @@ static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 				const struct qstr *cf_name,
 				u64 dst)
 {
-	struct bkey_i_dirent *dirent;
-	struct qstr _cf_name;
-
-	if (name->len > BCH_NAME_MAX)
-		return ERR_PTR(-ENAMETOOLONG);
+	struct bkey_i_dirent *dirent = bch2_trans_kmalloc(trans, BKEY_U64s_MAX * sizeof(u64));
+	if (IS_ERR(dirent))
+		return dirent;
 
-	if (hash_info->cf_encoding && !cf_name) {
-		int ret = bch2_casefold(trans, hash_info, name, &_cf_name);
-		if (ret)
-			return ERR_PTR(ret);
+	bkey_dirent_init(&dirent->k_i);
+	dirent->k.u64s = BKEY_U64s_MAX;
 
-		cf_name = &_cf_name;
+	if (type != DT_SUBVOL) {
+		dirent->v.d_inum = cpu_to_le64(dst);
+	} else {
+		dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
+		dirent->v.d_child_subvol = cpu_to_le32(dst);
 	}
 
-	dirent = dirent_alloc_key(trans, dir, type, name->len, cf_name ? cf_name->len : 0, dst);
-	if (IS_ERR(dirent))
-		return dirent;
+	dirent->v.d_type = type;
+	dirent->v.d_unused = 0;
 
-	if (cf_name)
-		dirent_init_casefolded_name(dirent, name, cf_name);
-	else
-		dirent_init_regular_name(dirent, name);
+	int ret = bch2_dirent_init_name(dirent, hash_info, name, cf_name);
+	if (ret)
+		return ERR_PTR(ret);
 
 	EBUG_ON(bch2_dirent_get_name(dirent_i_to_s_c(dirent)).len != name->len);
-
 	return dirent;
 }
 
@@ -341,7 +328,7 @@ int bch2_dirent_create_snapshot(struct btree_trans *trans,
 	struct bkey_i_dirent *dirent;
 	int ret;
 
-	dirent = dirent_create_key(trans, hash_info, dir_inum, type, name, NULL, dst_inum);
+	dirent = bch2_dirent_create_key(trans, hash_info, dir_inum, type, name, NULL, dst_inum);
 	ret = PTR_ERR_OR_ZERO(dirent);
 	if (ret)
 		return ret;
@@ -365,7 +352,7 @@ int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
 	struct bkey_i_dirent *dirent;
 	int ret;
 
-	dirent = dirent_create_key(trans, hash_info, dir, type, name, NULL, dst_inum);
+	dirent = bch2_dirent_create_key(trans, hash_info, dir, type, name, NULL, dst_inum);
 	ret = PTR_ERR_OR_ZERO(dirent);
 	if (ret)
 		return ret;
@@ -470,8 +457,8 @@ int bch2_dirent_rename(struct btree_trans *trans,
 		*src_offset = dst_iter.pos.offset;
 
 	/* Create new dst key: */
-	new_dst = dirent_create_key(trans, dst_hash, dst_dir, 0, dst_name,
-				    dst_hash->cf_encoding ? &dst_name_lookup : NULL, 0);
+	new_dst = bch2_dirent_create_key(trans, dst_hash, dst_dir, 0, dst_name,
+					 dst_hash->cf_encoding ? &dst_name_lookup : NULL, 0);
 	ret = PTR_ERR_OR_ZERO(new_dst);
 	if (ret)
 		goto out;
@@ -481,8 +468,8 @@ int bch2_dirent_rename(struct btree_trans *trans,
 
 	/* Create new src key: */
 	if (mode == BCH_RENAME_EXCHANGE) {
-		new_src = dirent_create_key(trans, src_hash, src_dir, 0, src_name,
-					    src_hash->cf_encoding ? &src_name_lookup : NULL, 0);
+		new_src = bch2_dirent_create_key(trans, src_hash, src_dir, 0, src_name,
+						 src_hash->cf_encoding ? &src_name_lookup : NULL, 0);
 		ret = PTR_ERR_OR_ZERO(new_src);
 		if (ret)
 			goto out;
@@ -648,7 +635,7 @@ int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 subvol, u32
 			struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 			if (d.v->d_type == DT_SUBVOL && le32_to_cpu(d.v->d_parent_subvol) != subvol)
 				continue;
-			ret = -BCH_ERR_ENOTEMPTY_dir_not_empty;
+			ret = bch_err_throw(trans->c, ENOTEMPTY_dir_not_empty);
 			break;
 		}
 	bch2_trans_iter_exit(trans, &iter);
@@ -705,8 +692,9 @@ int bch2_readdir(struct bch_fs *c, subvol_inum inum,
 
 			subvol_inum target;
 
+			bool need_second_pass = false;
 			int ret2 = bch2_str_hash_check_key(trans, NULL, &bch2_dirent_hash_desc,
-							   hash_info, &iter, k) ?:
+							   hash_info, &iter, k, &need_second_pass) ?:
 				bch2_dirent_read_target(trans, inum, dirent, &target);
 			if (ret2 > 0)
 				continue;
@@ -737,7 +725,7 @@ static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
 		ret = bch2_inode_unpack(k, inode);
 		goto found;
 	}
-	ret = -BCH_ERR_ENOENT_inode;
+	ret = bch_err_throw(trans->c, ENOENT_inode);
 found:
 	bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
 	bch2_trans_iter_exit(trans, &iter);
diff --git a/libbcachefs/dirent.h b/libbcachefs/dirent.h
index f94d589c..70fb0b58 100644
--- a/libbcachefs/dirent.h
+++ b/libbcachefs/dirent.h
@@ -38,7 +38,7 @@ static inline int bch2_maybe_casefold(struct btree_trans *trans,
 	}
 }
 
-struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d);
+struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent);
 
 static inline unsigned dirent_val_u64s(unsigned len, unsigned cf_len)
 {
@@ -59,6 +59,14 @@ static inline void dirent_copy_target(struct bkey_i_dirent *dst,
 	dst->v.d_type = src.v->d_type;
 }
 
+int bch2_dirent_init_name(struct bkey_i_dirent *,
+			  const struct bch_hash_info *,
+			  const struct qstr *,
+			  const struct qstr *);
+struct bkey_i_dirent *bch2_dirent_create_key(struct btree_trans *,
+				const struct bch_hash_info *, subvol_inum, u8,
+				const struct qstr *, const struct qstr *, u64);
+
 int bch2_dirent_create_snapshot(struct btree_trans *, u32, u64, u32,
 			const struct bch_hash_info *, u8,
 			const struct qstr *, u64, u64 *,
diff --git a/libbcachefs/disk_accounting.c b/libbcachefs/disk_accounting.c
index 6e4a6826..3d59a57a 100644
--- a/libbcachefs/disk_accounting.c
+++ b/libbcachefs/disk_accounting.c
@@ -390,7 +390,7 @@ static int __bch2_accounting_mem_insert(struct bch_fs *c, struct bkey_s_c_accoun
 err:
 	free_percpu(n.v[1]);
 	free_percpu(n.v[0]);
-	return -BCH_ERR_ENOMEM_disk_accounting;
+	return bch_err_throw(c, ENOMEM_disk_accounting);
 }
 
 int bch2_accounting_mem_insert(struct bch_fs *c, struct bkey_s_c_accounting a,
@@ -401,7 +401,7 @@ int bch2_accounting_mem_insert(struct bch_fs *c, struct bkey_s_c_accounting a,
 	if (mode != BCH_ACCOUNTING_read &&
 	    accounting_to_replicas(&r.e, a.k->p) &&
 	    !bch2_replicas_marked_locked(c, &r.e))
-		return -BCH_ERR_btree_insert_need_mark_replicas;
+		return bch_err_throw(c, btree_insert_need_mark_replicas);
 
 	percpu_up_read(&c->mark_lock);
 	percpu_down_write(&c->mark_lock);
@@ -419,7 +419,7 @@ int bch2_accounting_mem_insert_locked(struct bch_fs *c, struct bkey_s_c_accounti
 	if (mode != BCH_ACCOUNTING_read &&
 	    accounting_to_replicas(&r.e, a.k->p) &&
 	    !bch2_replicas_marked_locked(c, &r.e))
-		return -BCH_ERR_btree_insert_need_mark_replicas;
+		return bch_err_throw(c, btree_insert_need_mark_replicas);
 
 	return __bch2_accounting_mem_insert(c, a);
 }
@@ -559,7 +559,7 @@ int bch2_gc_accounting_start(struct bch_fs *c)
 					     sizeof(u64), GFP_KERNEL);
 		if (!e->v[1]) {
 			bch2_accounting_free_counters(acc, true);
-			ret = -BCH_ERR_ENOMEM_disk_accounting;
+			ret = bch_err_throw(c, ENOMEM_disk_accounting);
 			break;
 		}
 	}
@@ -737,7 +737,7 @@ invalid_device:
 				bch2_disk_accounting_mod(trans, acc, v, nr, false)) ?:
 			-BCH_ERR_remove_disk_accounting_entry;
 	} else {
-		ret = -BCH_ERR_remove_disk_accounting_entry;
+		ret = bch_err_throw(c, remove_disk_accounting_entry);
 	}
 	goto fsck_err;
 }
diff --git a/libbcachefs/disk_groups.c b/libbcachefs/disk_groups.c
index 9a17ef78..cde842ac 100644
--- a/libbcachefs/disk_groups.c
+++ b/libbcachefs/disk_groups.c
@@ -130,7 +130,7 @@ int bch2_sb_disk_groups_to_cpu(struct bch_fs *c)
 
 	cpu_g = kzalloc(struct_size(cpu_g, entries, nr_groups), GFP_KERNEL);
 	if (!cpu_g)
-		return -BCH_ERR_ENOMEM_disk_groups_to_cpu;
+		return bch_err_throw(c, ENOMEM_disk_groups_to_cpu);
 
 	cpu_g->nr = nr_groups;
 
diff --git a/libbcachefs/ec.c b/libbcachefs/ec.c
index 0a5f3a2e..543dbba9 100644
--- a/libbcachefs/ec.c
+++ b/libbcachefs/ec.c
@@ -213,7 +213,7 @@ static int __mark_stripe_bucket(struct btree_trans *trans,
 				a->dirty_sectors,
 				a->stripe, s.k->p.offset,
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 
@@ -224,7 +224,7 @@ static int __mark_stripe_bucket(struct btree_trans *trans,
 				a->dirty_sectors,
 				a->cached_sectors,
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 	} else {
@@ -234,7 +234,7 @@ static int __mark_stripe_bucket(struct btree_trans *trans,
 				bucket.inode, bucket.offset, a->gen,
 				a->stripe,
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 
@@ -244,7 +244,7 @@ static int __mark_stripe_bucket(struct btree_trans *trans,
 				bch2_data_type_str(a->data_type),
 				bch2_data_type_str(data_type),
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 
@@ -256,7 +256,7 @@ static int __mark_stripe_bucket(struct btree_trans *trans,
 				a->dirty_sectors,
 				a->cached_sectors,
 				(bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 	}
@@ -295,7 +295,7 @@ static int mark_stripe_bucket(struct btree_trans *trans,
 	struct bch_dev *ca = bch2_dev_tryget(c, ptr->dev);
 	if (unlikely(!ca)) {
 		if (ptr->dev != BCH_SB_MEMBER_INVALID && !(flags & BTREE_TRIGGER_overwrite))
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 		goto err;
 	}
 
@@ -325,7 +325,7 @@ static int mark_stripe_bucket(struct btree_trans *trans,
 		if (bch2_fs_inconsistent_on(!g, c, "reference to invalid bucket on device %u\n%s",
 					    ptr->dev,
 					    (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
-			ret = -BCH_ERR_mark_stripe;
+			ret = bch_err_throw(c, mark_stripe);
 			goto err;
 		}
 
@@ -428,7 +428,7 @@ int bch2_trigger_stripe(struct btree_trans *trans,
 			gc = genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL);
 			if (!gc) {
 				bch_err(c, "error allocating memory for gc_stripes, idx %llu", idx);
-				return -BCH_ERR_ENOMEM_mark_stripe;
+				return bch_err_throw(c, ENOMEM_mark_stripe);
 			}
 
 			/*
@@ -536,7 +536,8 @@ static void ec_stripe_buf_exit(struct ec_stripe_buf *buf)
 }
 
 /* XXX: this is a non-mempoolified memory allocation: */
-static int ec_stripe_buf_init(struct ec_stripe_buf *buf,
+static int ec_stripe_buf_init(struct bch_fs *c,
+			      struct ec_stripe_buf *buf,
 			      unsigned offset, unsigned size)
 {
 	struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
@@ -564,7 +565,7 @@ static int ec_stripe_buf_init(struct ec_stripe_buf *buf,
 	return 0;
 err:
 	ec_stripe_buf_exit(buf);
-	return -BCH_ERR_ENOMEM_stripe_buf;
+	return bch_err_throw(c, ENOMEM_stripe_buf);
 }
 
 /* Checksumming: */
@@ -840,7 +841,7 @@ int bch2_ec_read_extent(struct btree_trans *trans, struct bch_read_bio *rbio,
 
 	buf = kzalloc(sizeof(*buf), GFP_NOFS);
 	if (!buf)
-		return -BCH_ERR_ENOMEM_ec_read_extent;
+		return bch_err_throw(c, ENOMEM_ec_read_extent);
 
 	ret = lockrestart_do(trans, get_stripe_key_trans(trans, rbio->pick.ec.idx, buf));
 	if (ret) {
@@ -861,7 +862,7 @@ int bch2_ec_read_extent(struct btree_trans *trans, struct bch_read_bio *rbio,
 		goto err;
 	}
 
-	ret = ec_stripe_buf_init(buf, offset, bio_sectors(&rbio->bio));
+	ret = ec_stripe_buf_init(c, buf, offset, bio_sectors(&rbio->bio));
 	if (ret) {
 		msg = "-ENOMEM";
 		goto err;
@@ -894,7 +895,7 @@ err:
 	bch_err_ratelimited(c,
 			    "error doing reconstruct read: %s\n  %s", msg, msgbuf.buf);
 	printbuf_exit(&msgbuf);
-	ret = -BCH_ERR_stripe_reconstruct;
+	ret = bch_err_throw(c, stripe_reconstruct);
 	goto out;
 }
 
@@ -904,7 +905,7 @@ static int __ec_stripe_mem_alloc(struct bch_fs *c, size_t idx, gfp_t gfp)
 {
 	if (c->gc_pos.phase != GC_PHASE_not_running &&
 	    !genradix_ptr_alloc(&c->gc_stripes, idx, gfp))
-		return -BCH_ERR_ENOMEM_ec_stripe_mem_alloc;
+		return bch_err_throw(c, ENOMEM_ec_stripe_mem_alloc);
 
 	return 0;
 }
@@ -1129,7 +1130,7 @@ static int ec_stripe_update_extent(struct btree_trans *trans,
 
 		bch2_fs_inconsistent(c, "%s", buf.buf);
 		printbuf_exit(&buf);
-		return -BCH_ERR_erasure_coding_found_btree_node;
+		return bch_err_throw(c, erasure_coding_found_btree_node);
 	}
 
 	k = bch2_backpointer_get_key(trans, bp, &iter, BTREE_ITER_intent, last_flushed);
@@ -1195,7 +1196,7 @@ static int ec_stripe_update_bucket(struct btree_trans *trans, struct ec_stripe_b
 
 	struct bch_dev *ca = bch2_dev_tryget(c, ptr.dev);
 	if (!ca)
-		return -BCH_ERR_ENOENT_dev_not_found;
+		return bch_err_throw(c, ENOENT_dev_not_found);
 
 	struct bpos bucket_pos = PTR_BUCKET_POS(ca, &ptr);
 
@@ -1256,7 +1257,7 @@ static void zero_out_rest_of_ec_bucket(struct bch_fs *c,
 	struct bch_dev *ca = bch2_dev_get_ioref(c, ob->dev, WRITE,
 				BCH_DEV_WRITE_REF_ec_bucket_zero);
 	if (!ca) {
-		s->err = -BCH_ERR_erofs_no_writes;
+		s->err = bch_err_throw(c, erofs_no_writes);
 		return;
 	}
 
@@ -1320,7 +1321,7 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 
 		if (ec_do_recov(c, &s->existing_stripe)) {
 			bch_err(c, "error creating stripe: error reading existing stripe");
-			ret = -BCH_ERR_ec_block_read;
+			ret = bch_err_throw(c, ec_block_read);
 			goto err;
 		}
 
@@ -1346,7 +1347,7 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 
 	if (ec_nr_failed(&s->new_stripe)) {
 		bch_err(c, "error creating stripe: error writing redundancy buckets");
-		ret = -BCH_ERR_ec_block_write;
+		ret = bch_err_throw(c, ec_block_write);
 		goto err;
 	}
 
@@ -1865,7 +1866,7 @@ static int init_new_stripe_from_existing(struct bch_fs *c, struct ec_stripe_new
 	s->nr_data = existing_v->nr_blocks -
 		existing_v->nr_redundant;
 
-	int ret = ec_stripe_buf_init(&s->existing_stripe, 0, le16_to_cpu(existing_v->sectors));
+	int ret = ec_stripe_buf_init(c, &s->existing_stripe, 0, le16_to_cpu(existing_v->sectors));
 	if (ret) {
 		bch2_stripe_close(c, s);
 		return ret;
@@ -1925,7 +1926,7 @@ static int __bch2_ec_stripe_head_reuse(struct btree_trans *trans, struct ec_stri
 	}
 	bch2_trans_iter_exit(trans, &lru_iter);
 	if (!ret)
-		ret = -BCH_ERR_stripe_alloc_blocked;
+		ret = bch_err_throw(c, stripe_alloc_blocked);
 	if (ret == 1)
 		ret = 0;
 	if (ret)
@@ -1966,7 +1967,7 @@ static int __bch2_ec_stripe_head_reserve(struct btree_trans *trans, struct ec_st
 				continue;
 			}
 
-			ret = -BCH_ERR_ENOSPC_stripe_create;
+			ret = bch_err_throw(c, ENOSPC_stripe_create);
 			break;
 		}
 
@@ -2024,7 +2025,7 @@ struct ec_stripe_head *bch2_ec_stripe_head_get(struct btree_trans *trans,
 	if (!h->s) {
 		h->s = ec_new_stripe_alloc(c, h);
 		if (!h->s) {
-			ret = -BCH_ERR_ENOMEM_ec_new_stripe_alloc;
+			ret = bch_err_throw(c, ENOMEM_ec_new_stripe_alloc);
 			bch_err(c, "failed to allocate new stripe");
 			goto err;
 		}
@@ -2089,7 +2090,7 @@ alloc_existing:
 		goto err;
 
 allocate_buf:
-	ret = ec_stripe_buf_init(&s->new_stripe, 0, h->blocksize);
+	ret = ec_stripe_buf_init(c, &s->new_stripe, 0, h->blocksize);
 	if (ret)
 		goto err;
 
@@ -2115,6 +2116,7 @@ int bch2_invalidate_stripe_to_dev(struct btree_trans *trans,
 	if (k.k->type != KEY_TYPE_stripe)
 		return 0;
 
+	struct bch_fs *c = trans->c;
 	struct bkey_i_stripe *s =
 		bch2_bkey_make_mut_typed(trans, iter, &k, 0, stripe);
 	int ret = PTR_ERR_OR_ZERO(s);
@@ -2146,17 +2148,17 @@ int bch2_invalidate_stripe_to_dev(struct btree_trans *trans,
 			if (ptr->dev == dev_idx)
 				ptr->dev = BCH_SB_MEMBER_INVALID;
 
-			struct bch_dev *ca = bch2_dev_rcu(trans->c, ptr->dev);
+			struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
 			nr_good += ca && ca->mi.state != BCH_MEMBER_STATE_failed;
 		}
 
 	if (nr_good < s->v.nr_blocks && !(flags & BCH_FORCE_IF_DATA_DEGRADED))
-		return -BCH_ERR_remove_would_lose_data;
+		return bch_err_throw(c, remove_would_lose_data);
 
 	unsigned nr_data = s->v.nr_blocks - s->v.nr_redundant;
 
 	if (nr_good < nr_data && !(flags & BCH_FORCE_IF_DATA_LOST))
-		return -BCH_ERR_remove_would_lose_data;
+		return bch_err_throw(c, remove_would_lose_data);
 
 	sectors = -sectors;
 
@@ -2177,14 +2179,15 @@ static int bch2_invalidate_stripe_to_dev_from_alloc(struct btree_trans *trans, s
 		return 0;
 
 	if (a->stripe_sectors) {
-		bch_err(trans->c, "trying to invalidate device in stripe when bucket has stripe data");
-		return -BCH_ERR_invalidate_stripe_to_dev;
+		struct bch_fs *c = trans->c;
+		bch_err(c, "trying to invalidate device in stripe when bucket has stripe data");
+		return bch_err_throw(c, invalidate_stripe_to_dev);
 	}
 
 	struct btree_iter iter;
 	struct bkey_s_c_stripe s =
 		bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_stripes, POS(0, a->stripe),
-					BTREE_ITER_slots, stripe);
+					 BTREE_ITER_slots, stripe);
 	int ret = bkey_err(s);
 	if (ret)
 		return ret;
diff --git a/libbcachefs/errcode.h b/libbcachefs/errcode.h
index 6b0791e1..6e0bf7ff 100644
--- a/libbcachefs/errcode.h
+++ b/libbcachefs/errcode.h
@@ -182,9 +182,12 @@
 	x(BCH_ERR_fsck,			fsck_errors_not_fixed)			\
 	x(BCH_ERR_fsck,			fsck_repair_unimplemented)		\
 	x(BCH_ERR_fsck,			fsck_repair_impossible)			\
-	x(EINVAL,			restart_recovery)			\
-	x(EINVAL,			cannot_rewind_recovery)			\
+	x(EINVAL,			recovery_will_run)			\
+	x(BCH_ERR_recovery_will_run,	restart_recovery)			\
+	x(BCH_ERR_recovery_will_run,	cannot_rewind_recovery)			\
+	x(BCH_ERR_recovery_will_run,	recovery_pass_will_run)			\
 	x(0,				data_update_done)			\
+	x(0,				bkey_was_deleted)			\
 	x(BCH_ERR_data_update_done,	data_update_done_would_block)		\
 	x(BCH_ERR_data_update_done,	data_update_done_unwritten)		\
 	x(BCH_ERR_data_update_done,	data_update_done_no_writes_needed)	\
diff --git a/libbcachefs/error.c b/libbcachefs/error.c
index 0488f966..ff49cebf 100644
--- a/libbcachefs/error.c
+++ b/libbcachefs/error.c
@@ -100,10 +100,10 @@ int __bch2_topology_error(struct bch_fs *c, struct printbuf *out)
 	set_bit(BCH_FS_topology_error, &c->flags);
 	if (!test_bit(BCH_FS_in_recovery, &c->flags)) {
 		__bch2_inconsistent_error(c, out);
-		return -BCH_ERR_btree_need_topology_repair;
+		return bch_err_throw(c, btree_need_topology_repair);
 	} else {
 		return bch2_run_explicit_recovery_pass(c, out, BCH_RECOVERY_PASS_check_topology, 0) ?:
-			-BCH_ERR_btree_node_read_validate_error;
+			bch_err_throw(c, btree_node_read_validate_error);
 	}
 }
 
@@ -403,23 +403,23 @@ int bch2_fsck_err_opt(struct bch_fs *c,
 
 	if (test_bit(BCH_FS_in_fsck, &c->flags)) {
 		if (!(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE)))
-			return -BCH_ERR_fsck_repair_unimplemented;
+			return bch_err_throw(c, fsck_repair_unimplemented);
 
 		switch (c->opts.fix_errors) {
 		case FSCK_FIX_exit:
-			return -BCH_ERR_fsck_errors_not_fixed;
+			return bch_err_throw(c, fsck_errors_not_fixed);
 		case FSCK_FIX_yes:
 			if (flags & FSCK_CAN_FIX)
-				return -BCH_ERR_fsck_fix;
+				return bch_err_throw(c, fsck_fix);
 			fallthrough;
 		case FSCK_FIX_no:
 			if (flags & FSCK_CAN_IGNORE)
-				return -BCH_ERR_fsck_ignore;
-			return -BCH_ERR_fsck_errors_not_fixed;
+				return bch_err_throw(c, fsck_ignore);
+			return bch_err_throw(c, fsck_errors_not_fixed);
 		case FSCK_FIX_ask:
 			if (flags & FSCK_AUTOFIX)
-				return -BCH_ERR_fsck_fix;
-			return -BCH_ERR_fsck_ask;
+				return bch_err_throw(c, fsck_fix);
+			return bch_err_throw(c, fsck_ask);
 		default:
 			BUG();
 		}
@@ -427,12 +427,12 @@ int bch2_fsck_err_opt(struct bch_fs *c,
 		if ((flags & FSCK_AUTOFIX) &&
 		    (c->opts.errors == BCH_ON_ERROR_continue ||
 		     c->opts.errors == BCH_ON_ERROR_fix_safe))
-			return -BCH_ERR_fsck_fix;
+			return bch_err_throw(c, fsck_fix);
 
 		if (c->opts.errors == BCH_ON_ERROR_continue &&
 		    (flags & FSCK_CAN_IGNORE))
-			return -BCH_ERR_fsck_ignore;
-		return -BCH_ERR_fsck_errors_not_fixed;
+			return bch_err_throw(c, fsck_ignore);
+		return bch_err_throw(c, fsck_errors_not_fixed);
 	}
 }
 
@@ -474,8 +474,8 @@ int __bch2_fsck_err(struct bch_fs *c,
 
 	if (test_bit(err, c->sb.errors_silent))
 		return flags & FSCK_CAN_FIX
-			? -BCH_ERR_fsck_fix
-			: -BCH_ERR_fsck_ignore;
+			? bch_err_throw(c, fsck_fix)
+			: bch_err_throw(c, fsck_ignore);
 
 	printbuf_indent_add_nextline(out, 2);
 
@@ -517,10 +517,10 @@ int __bch2_fsck_err(struct bch_fs *c,
 		prt_str(out, ", ");
 		if (flags & FSCK_CAN_FIX) {
 			prt_actioning(out, action);
-			ret = -BCH_ERR_fsck_fix;
+			ret = bch_err_throw(c, fsck_fix);
 		} else {
 			prt_str(out, ", continuing");
-			ret = -BCH_ERR_fsck_ignore;
+			ret = bch_err_throw(c, fsck_ignore);
 		}
 
 		goto print;
@@ -532,18 +532,18 @@ int __bch2_fsck_err(struct bch_fs *c,
 					 "run fsck, and forward to devs so error can be marked for self-healing");
 			inconsistent = true;
 			print = true;
-			ret = -BCH_ERR_fsck_errors_not_fixed;
+			ret = bch_err_throw(c, fsck_errors_not_fixed);
 		} else if (flags & FSCK_CAN_FIX) {
 			prt_str(out, ", ");
 			prt_actioning(out, action);
-			ret = -BCH_ERR_fsck_fix;
+			ret = bch_err_throw(c, fsck_fix);
 		} else {
 			prt_str(out, ", continuing");
-			ret = -BCH_ERR_fsck_ignore;
+			ret = bch_err_throw(c, fsck_ignore);
 		}
 	} else if (c->opts.fix_errors == FSCK_FIX_exit) {
 		prt_str(out, ", exiting");
-		ret = -BCH_ERR_fsck_errors_not_fixed;
+		ret = bch_err_throw(c, fsck_errors_not_fixed);
 	} else if (flags & FSCK_CAN_FIX) {
 		int fix = s && s->fix
 			? s->fix
@@ -562,33 +562,33 @@ int __bch2_fsck_err(struct bch_fs *c,
 					: FSCK_FIX_yes;
 
 			ret = ret & 1
-				? -BCH_ERR_fsck_fix
-				: -BCH_ERR_fsck_ignore;
+				? bch_err_throw(c, fsck_fix)
+				: bch_err_throw(c, fsck_ignore);
 		} else if (fix == FSCK_FIX_yes ||
 			   (c->opts.nochanges &&
 			    !(flags & FSCK_CAN_IGNORE))) {
 			prt_str(out, ", ");
 			prt_actioning(out, action);
-			ret = -BCH_ERR_fsck_fix;
+			ret = bch_err_throw(c, fsck_fix);
 		} else {
 			prt_str(out, ", not ");
 			prt_actioning(out, action);
-			ret = -BCH_ERR_fsck_ignore;
+			ret = bch_err_throw(c, fsck_ignore);
 		}
 	} else {
 		if (flags & FSCK_CAN_IGNORE) {
 			prt_str(out, ", continuing");
-			ret = -BCH_ERR_fsck_ignore;
+			ret = bch_err_throw(c, fsck_ignore);
 		} else {
 			prt_str(out, " (repair unimplemented)");
-			ret = -BCH_ERR_fsck_repair_unimplemented;
+			ret = bch_err_throw(c, fsck_repair_unimplemented);
 		}
 	}
 
 	if (bch2_err_matches(ret, BCH_ERR_fsck_ignore) &&
 	    (c->opts.fix_errors == FSCK_FIX_exit ||
 	     !(flags & FSCK_CAN_IGNORE)))
-		ret = -BCH_ERR_fsck_errors_not_fixed;
+		ret = bch_err_throw(c, fsck_errors_not_fixed);
 
 	if (test_bit(BCH_FS_in_fsck, &c->flags) &&
 	    (!bch2_err_matches(ret, BCH_ERR_fsck_fix) &&
@@ -620,6 +620,9 @@ print:
 
 	if (s)
 		s->ret = ret;
+
+	if (trans)
+		ret = bch2_trans_log_str(trans, bch2_sb_error_strs[err]) ?: ret;
 err_unlock:
 	mutex_unlock(&c->fsck_error_msgs_lock);
 err:
@@ -657,12 +660,12 @@ int __bch2_bkey_fsck_err(struct bch_fs *c,
 			 const char *fmt, ...)
 {
 	if (from.flags & BCH_VALIDATE_silent)
-		return -BCH_ERR_fsck_delete_bkey;
+		return bch_err_throw(c, fsck_delete_bkey);
 
 	unsigned fsck_flags = 0;
 	if (!(from.flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit))) {
 		if (test_bit(err, c->sb.errors_silent))
-			return -BCH_ERR_fsck_delete_bkey;
+			return bch_err_throw(c, fsck_delete_bkey);
 
 		fsck_flags |= FSCK_AUTOFIX|FSCK_CAN_FIX;
 	}
diff --git a/libbcachefs/error.h b/libbcachefs/error.h
index 4babb0d1..0c3c3a24 100644
--- a/libbcachefs/error.h
+++ b/libbcachefs/error.h
@@ -173,7 +173,7 @@ do {									\
 	if (!bch2_err_matches(_ret, BCH_ERR_fsck_fix) &&		\
 	    !bch2_err_matches(_ret, BCH_ERR_fsck_ignore))		\
 		ret = _ret;						\
-	ret = -BCH_ERR_fsck_delete_bkey;				\
+	ret = bch_err_throw(c, fsck_delete_bkey);			\
 	goto fsck_err;							\
 } while (0)
 
diff --git a/libbcachefs/extents.c b/libbcachefs/extents.c
index 677cf453..036e4ad9 100644
--- a/libbcachefs/extents.c
+++ b/libbcachefs/extents.c
@@ -193,7 +193,7 @@ int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 	bool have_dirty_ptrs = false, have_pick = false;
 
 	if (k.k->type == KEY_TYPE_error)
-		return -BCH_ERR_key_type_error;
+		return bch_err_throw(c, key_type_error);
 
 	rcu_read_lock();
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
@@ -286,17 +286,17 @@ int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 	if (!have_dirty_ptrs)
 		return 0;
 	if (have_missing_devs)
-		return -BCH_ERR_no_device_to_read_from;
+		return bch_err_throw(c, no_device_to_read_from);
 	if (have_csum_errors)
-		return -BCH_ERR_data_read_csum_err;
+		return bch_err_throw(c, data_read_csum_err);
 	if (have_io_errors)
-		return -BCH_ERR_data_read_io_err;
+		return bch_err_throw(c, data_read_io_err);
 
 	/*
 	 * If we get here, we have pointers (bkey_ptrs_validate() ensures that),
 	 * but they don't point to valid devices:
 	 */
-	return -BCH_ERR_no_devices_valid;
+	return bch_err_throw(c, no_devices_valid);
 }
 
 /* KEY_TYPE_btree_ptr: */
@@ -1515,7 +1515,7 @@ int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
 				struct bch_compression_opt opt = __bch2_compression_decode(r->compression);
 				prt_printf(err, "invalid compression opt %u:%u",
 					   opt.type, opt.level);
-				return -BCH_ERR_invalid_bkey;
+				return bch_err_throw(c, invalid_bkey);
 			}
 #endif
 			break;
diff --git a/libbcachefs/fs-io-pagecache.c b/libbcachefs/fs-io-pagecache.c
index fbae9c1d..c2cc4058 100644
--- a/libbcachefs/fs-io-pagecache.c
+++ b/libbcachefs/fs-io-pagecache.c
@@ -447,7 +447,7 @@ static int __bch2_folio_reservation_get(struct bch_fs *c,
 
 					if (!reserved) {
 						bch2_disk_reservation_put(c, &disk_res);
-						return -BCH_ERR_ENOSPC_disk_reservation;
+						return bch_err_throw(c, ENOSPC_disk_reservation);
 					}
 					break;
 				}
diff --git a/libbcachefs/fs-ioctl.c b/libbcachefs/fs-ioctl.c
index 05361a79..4e72e654 100644
--- a/libbcachefs/fs-ioctl.c
+++ b/libbcachefs/fs-ioctl.c
@@ -268,13 +268,13 @@ static long bch2_ioctl_subvolume_create(struct bch_fs *c, struct file *filp,
 	}
 
 	if (dst_dentry->d_inode) {
-		error = -BCH_ERR_EEXIST_subvolume_create;
+		error = bch_err_throw(c, EEXIST_subvolume_create);
 		goto err3;
 	}
 
 	dir = dst_path.dentry->d_inode;
 	if (IS_DEADDIR(dir)) {
-		error = -BCH_ERR_ENOENT_directory_dead;
+		error = bch_err_throw(c, ENOENT_directory_dead);
 		goto err3;
 	}
 
diff --git a/libbcachefs/fs.c b/libbcachefs/fs.c
index ebf967e8..41008d4a 100644
--- a/libbcachefs/fs.c
+++ b/libbcachefs/fs.c
@@ -2007,14 +2007,14 @@ retry:
 			goto err;
 
 		if (k.k->type != KEY_TYPE_dirent) {
-			ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
+			ret = bch_err_throw(c, ENOENT_dirent_doesnt_match_inode);
 			goto err;
 		}
 
 		d = bkey_s_c_to_dirent(k);
 		ret = bch2_dirent_read_target(trans, inode_inum(dir), d, &target);
 		if (ret > 0)
-			ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
+			ret = bch_err_throw(c, ENOENT_dirent_doesnt_match_inode);
 		if (ret)
 			goto err;
 
diff --git a/libbcachefs/fsck.c b/libbcachefs/fsck.c
index 66b7a87a..77760aaa 100644
--- a/libbcachefs/fsck.c
+++ b/libbcachefs/fsck.c
@@ -23,14 +23,15 @@
 #include <linux/bsearch.h>
 #include <linux/dcache.h> /* struct qstr */
 
-static int dirent_points_to_inode_nowarn(struct bkey_s_c_dirent d,
+static int dirent_points_to_inode_nowarn(struct bch_fs *c,
+					 struct bkey_s_c_dirent d,
 					 struct bch_inode_unpacked *inode)
 {
 	if (d.v->d_type == DT_SUBVOL
 	    ? le32_to_cpu(d.v->d_child_subvol)	== inode->bi_subvol
 	    : le64_to_cpu(d.v->d_inum)		== inode->bi_inum)
 		return 0;
-	return -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
+	return bch_err_throw(c, ENOENT_dirent_doesnt_match_inode);
 }
 
 static void dirent_inode_mismatch_msg(struct printbuf *out,
@@ -49,7 +50,7 @@ static int dirent_points_to_inode(struct bch_fs *c,
 				  struct bkey_s_c_dirent dirent,
 				  struct bch_inode_unpacked *inode)
 {
-	int ret = dirent_points_to_inode_nowarn(dirent, inode);
+	int ret = dirent_points_to_inode_nowarn(c, dirent, inode);
 	if (ret) {
 		struct printbuf buf = PRINTBUF;
 		dirent_inode_mismatch_msg(&buf, c, dirent, inode);
@@ -152,7 +153,7 @@ static int find_snapshot_tree_subvol(struct btree_trans *trans,
 			goto found;
 		}
 	}
-	ret = -BCH_ERR_ENOENT_no_snapshot_tree_subvol;
+	ret = bch_err_throw(trans->c, ENOENT_no_snapshot_tree_subvol);
 found:
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
@@ -229,7 +230,7 @@ static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
 
 	if (d_type != DT_DIR) {
 		bch_err(c, "error looking up lost+found: not a directory");
-		return -BCH_ERR_ENOENT_not_directory;
+		return bch_err_throw(c, ENOENT_not_directory);
 	}
 
 	/*
@@ -531,7 +532,7 @@ static int reconstruct_subvol(struct btree_trans *trans, u32 snapshotid, u32 sub
 
 	if (!bch2_snapshot_is_leaf(c, snapshotid)) {
 		bch_err(c, "need to reconstruct subvol, but have interior node snapshot");
-		return -BCH_ERR_fsck_repair_unimplemented;
+		return bch_err_throw(c, fsck_repair_unimplemented);
 	}
 
 	/*
@@ -939,7 +940,7 @@ lookup_inode_for_snapshot(struct btree_trans *trans, struct inode_walker *w, str
 		if (ret)
 			goto fsck_err;
 
-		ret = -BCH_ERR_transaction_restart_nested;
+		ret = bch_err_throw(c, transaction_restart_nested);
 		goto fsck_err;
 	}
 
@@ -1041,7 +1042,7 @@ static int check_inode_dirent_inode(struct btree_trans *trans,
 	if (ret && !bch2_err_matches(ret, ENOENT))
 		return ret;
 
-	if ((ret || dirent_points_to_inode_nowarn(d, inode)) &&
+	if ((ret || dirent_points_to_inode_nowarn(c, d, inode)) &&
 	    inode->bi_subvol &&
 	    (inode->bi_flags & BCH_INODE_has_child_snapshot)) {
 		/* Older version of a renamed subvolume root: we won't have a
@@ -1062,7 +1063,7 @@ static int check_inode_dirent_inode(struct btree_trans *trans,
 			trans, inode_points_to_missing_dirent,
 			"inode points to missing dirent\n%s",
 			(bch2_inode_unpacked_to_text(&buf, inode), buf.buf)) ||
-	    fsck_err_on(!ret && dirent_points_to_inode_nowarn(d, inode),
+	    fsck_err_on(!ret && dirent_points_to_inode_nowarn(c, d, inode),
 			trans, inode_points_to_wrong_dirent,
 			"%s",
 			(printbuf_reset(&buf),
@@ -1453,7 +1454,7 @@ static int check_key_has_inode(struct btree_trans *trans,
 			goto err;
 
 		inode->last_pos.inode--;
-		ret = -BCH_ERR_transaction_restart_nested;
+		ret = bch_err_throw(c, transaction_restart_nested);
 		goto err;
 	}
 
@@ -1570,7 +1571,7 @@ static int extent_ends_at(struct bch_fs *c,
 			      sizeof(seen->ids.data[0]) * seen->ids.size,
 			      GFP_KERNEL);
 	if (!n.seen.ids.data)
-		return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
+		return bch_err_throw(c, ENOMEM_fsck_extent_ends_at);
 
 	__darray_for_each(extent_ends->e, i) {
 		if (i->snapshot == k.k->p.snapshot) {
@@ -1620,7 +1621,7 @@ static int overlapping_extents_found(struct btree_trans *trans,
 
 		bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
 			__func__, buf.buf);
-		ret = -BCH_ERR_internal_fsck_err;
+		ret = bch_err_throw(c, internal_fsck_err);
 		goto err;
 	}
 
@@ -1645,7 +1646,7 @@ static int overlapping_extents_found(struct btree_trans *trans,
 	    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;
+		ret = bch_err_throw(c, internal_fsck_err);
 		goto err;
 	}
 
@@ -1693,7 +1694,7 @@ static int overlapping_extents_found(struct btree_trans *trans,
 			 * We overwrote the second extent - restart
 			 * check_extent() from the top:
 			 */
-			ret = -BCH_ERR_transaction_restart_nested;
+			ret = bch_err_throw(c, transaction_restart_nested);
 		}
 	}
 fsck_err:
@@ -2046,7 +2047,7 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 			(bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
 		if (!new_parent_subvol) {
 			bch_err(c, "could not find a subvol for snapshot %u", d.k->p.snapshot);
-			return -BCH_ERR_fsck_repair_unimplemented;
+			return bch_err_throw(c, fsck_repair_unimplemented);
 		}
 
 		struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
@@ -2108,7 +2109,7 @@ static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *
 
 	if (ret) {
 		bch_err(c, "subvol %u points to missing inode root %llu", target_subvol, target_inum);
-		ret = -BCH_ERR_fsck_repair_unimplemented;
+		ret = bch_err_throw(c, fsck_repair_unimplemented);
 		goto err;
 	}
 
@@ -2140,7 +2141,8 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 			struct bch_hash_info *hash_info,
 			struct inode_walker *dir,
 			struct inode_walker *target,
-			struct snapshots_seen *s)
+			struct snapshots_seen *s,
+			bool *need_second_pass)
 {
 	struct bch_fs *c = trans->c;
 	struct inode_walker_entry *i;
@@ -2182,7 +2184,10 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 		*hash_info = bch2_hash_info_init(c, &i->inode);
 	dir->first_this_inode = false;
 
-	ret = bch2_str_hash_check_key(trans, s, &bch2_dirent_hash_desc, hash_info, iter, k);
+	hash_info->cf_encoding = bch2_inode_casefold(c, &i->inode) ? c->cf_encoding : NULL;
+
+	ret = bch2_str_hash_check_key(trans, s, &bch2_dirent_hash_desc, hash_info,
+				      iter, k, need_second_pass);
 	if (ret < 0)
 		goto err;
 	if (ret) {
@@ -2203,31 +2208,34 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 			(printbuf_reset(&buf),
 			 bch2_bkey_val_to_text(&buf, c, k),
 			 buf.buf))) {
-		struct qstr name = bch2_dirent_get_name(d);
-		u32 subvol = d.v->d_type == DT_SUBVOL
-			? le32_to_cpu(d.v->d_parent_subvol)
-			: 0;
+		subvol_inum dir_inum = { .subvol = d.v->d_type == DT_SUBVOL
+				? le32_to_cpu(d.v->d_parent_subvol)
+				: 0,
+		};
 		u64 target = d.v->d_type == DT_SUBVOL
 			? le32_to_cpu(d.v->d_child_subvol)
 			: le64_to_cpu(d.v->d_inum);
-		u64 dir_offset;
+		struct qstr name = bch2_dirent_get_name(d);
 
-		ret =   bch2_hash_delete_at(trans,
+		struct bkey_i_dirent *new_d =
+			bch2_dirent_create_key(trans, hash_info, dir_inum,
+					       d.v->d_type, &name, NULL, target);
+		ret = PTR_ERR_OR_ZERO(new_d);
+		if (ret)
+			goto out;
+
+		new_d->k.p.inode	= d.k->p.inode;
+		new_d->k.p.snapshot	= d.k->p.snapshot;
+
+		struct btree_iter dup_iter = {};
+		ret =	bch2_hash_delete_at(trans,
 					    bch2_dirent_hash_desc, hash_info, iter,
 					    BTREE_UPDATE_internal_snapshot_node) ?:
-			bch2_dirent_create_snapshot(trans, subvol,
-						    d.k->p.inode, d.k->p.snapshot,
-						    hash_info,
-						    d.v->d_type,
-						    &name,
-						    target,
-						    &dir_offset,
-						    BTREE_ITER_with_updates|
-						    BTREE_UPDATE_internal_snapshot_node|
-						    STR_HASH_must_create) ?:
-			bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
-
-		/* might need another check_dirents pass */
+			bch2_str_hash_repair_key(trans, s,
+						 &bch2_dirent_hash_desc, hash_info,
+						 iter, bkey_i_to_s_c(&new_d->k_i),
+						 &dup_iter, bkey_s_c_null,
+						 need_second_pass);
 		goto out;
 	}
 
@@ -2295,7 +2303,6 @@ out:
 err:
 fsck_err:
 	printbuf_exit(&buf);
-	bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -2309,17 +2316,30 @@ int bch2_check_dirents(struct bch_fs *c)
 	struct inode_walker target = inode_walker_init();
 	struct snapshots_seen s;
 	struct bch_hash_info hash_info;
+	bool need_second_pass = false, did_second_pass = false;
 
 	snapshots_seen_init(&s);
-
+again:
 	int ret = bch2_trans_run(c,
 		for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
 				POS(BCACHEFS_ROOT_INO, 0),
 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
 				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
-			check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)) ?:
+			check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s,
+				     &need_second_pass)) ?:
 		check_subdir_count_notnested(trans, &dir));
 
+	if (!ret && need_second_pass && !did_second_pass) {
+		bch_info(c, "check_dirents requires second pass");
+		swap(did_second_pass, need_second_pass);
+		goto again;
+	}
+
+	if (!ret && need_second_pass) {
+		bch_err(c, "dirents not repairing");
+		ret = -EINVAL;
+	}
+
 	snapshots_seen_exit(&s);
 	inode_walker_exit(&dir);
 	inode_walker_exit(&target);
@@ -2333,16 +2353,14 @@ 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 = bch2_check_key_has_snapshot(trans, iter, k);
+	int ret = bch2_check_key_has_snapshot(trans, iter, k);
 	if (ret < 0)
 		return ret;
 	if (ret)
 		return 0;
 
-	i = walk_inode(trans, inode, k);
+	struct inode_walker_entry *i = walk_inode(trans, inode, k);
 	ret = PTR_ERR_OR_ZERO(i);
 	if (ret)
 		return ret;
@@ -2358,9 +2376,9 @@ static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
 		*hash_info = bch2_hash_info_init(c, &i->inode);
 	inode->first_this_inode = false;
 
-	ret = bch2_str_hash_check_key(trans, NULL, &bch2_xattr_hash_desc, hash_info, iter, k);
-	bch_err_fn(c, ret);
-	return ret;
+	bool need_second_pass = false;
+	return bch2_str_hash_check_key(trans, NULL, &bch2_xattr_hash_desc, hash_info,
+				      iter, k, &need_second_pass);
 }
 
 /*
@@ -2749,7 +2767,7 @@ static int add_nlink(struct bch_fs *c, struct nlink_table *t,
 		if (!d) {
 			bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
 				new_size);
-			return -BCH_ERR_ENOMEM_fsck_add_nlink;
+			return bch_err_throw(c, ENOMEM_fsck_add_nlink);
 		}
 
 		if (t->d)
diff --git a/libbcachefs/inode.c b/libbcachefs/inode.c
index 5b35dae8..391daae5 100644
--- a/libbcachefs/inode.c
+++ b/libbcachefs/inode.c
@@ -1041,7 +1041,7 @@ again:
 		goto found_slot;
 
 	if (!ret && start == min)
-		ret = -BCH_ERR_ENOSPC_inode_create;
+		ret = bch_err_throw(trans->c, ENOSPC_inode_create);
 
 	if (ret) {
 		bch2_trans_iter_exit(trans, iter);
@@ -1161,7 +1161,7 @@ retry:
 		bch2_fs_inconsistent(c,
 				     "inode %llu:%u not found when deleting",
 				     inum.inum, snapshot);
-		ret = -BCH_ERR_ENOENT_inode;
+		ret = bch_err_throw(c, ENOENT_inode);
 		goto err;
 	}
 
@@ -1328,7 +1328,7 @@ retry:
 		bch2_fs_inconsistent(c,
 				     "inode %llu:%u not found when deleting",
 				     inum, snapshot);
-		ret = -BCH_ERR_ENOENT_inode;
+		ret = bch_err_throw(c, ENOENT_inode);
 		goto err;
 	}
 
diff --git a/libbcachefs/io_misc.c b/libbcachefs/io_misc.c
index cc07729a..bf72b1d2 100644
--- a/libbcachefs/io_misc.c
+++ b/libbcachefs/io_misc.c
@@ -91,7 +91,7 @@ int bch2_extent_fallocate(struct btree_trans *trans,
 				opts.data_replicas,
 				BCH_WATERMARK_normal, 0, &cl, &wp);
 		if (bch2_err_matches(ret, BCH_ERR_operation_blocked))
-			ret = -BCH_ERR_transaction_restart_nested;
+			ret = bch_err_throw(c, transaction_restart_nested);
 		if (ret)
 			goto err;
 
diff --git a/libbcachefs/io_read.c b/libbcachefs/io_read.c
index fb83c694..a77779af 100644
--- a/libbcachefs/io_read.c
+++ b/libbcachefs/io_read.c
@@ -137,21 +137,21 @@ static inline int should_promote(struct bch_fs *c, struct bkey_s_c k,
 		BUG_ON(!opts.promote_target);
 
 		if (!(flags & BCH_READ_may_promote))
-			return -BCH_ERR_nopromote_may_not;
+			return bch_err_throw(c, nopromote_may_not);
 
 		if (bch2_bkey_has_target(c, k, opts.promote_target))
-			return -BCH_ERR_nopromote_already_promoted;
+			return bch_err_throw(c, nopromote_already_promoted);
 
 		if (bkey_extent_is_unwritten(k))
-			return -BCH_ERR_nopromote_unwritten;
+			return bch_err_throw(c, nopromote_unwritten);
 
 		if (bch2_target_congested(c, opts.promote_target))
-			return -BCH_ERR_nopromote_congested;
+			return bch_err_throw(c, nopromote_congested);
 	}
 
 	if (rhashtable_lookup_fast(&c->promote_table, &pos,
 				   bch_promote_params))
-		return -BCH_ERR_nopromote_in_flight;
+		return bch_err_throw(c, nopromote_in_flight);
 
 	return 0;
 }
@@ -239,7 +239,7 @@ static struct bch_read_bio *__promote_alloc(struct btree_trans *trans,
 
 	struct promote_op *op = kzalloc(sizeof(*op), GFP_KERNEL);
 	if (!op) {
-		ret = -BCH_ERR_nopromote_enomem;
+		ret = bch_err_throw(c, nopromote_enomem);
 		goto err_put;
 	}
 
@@ -248,7 +248,7 @@ static struct bch_read_bio *__promote_alloc(struct btree_trans *trans,
 
 	if (rhashtable_lookup_insert_fast(&c->promote_table, &op->hash,
 					  bch_promote_params)) {
-		ret = -BCH_ERR_nopromote_in_flight;
+		ret = bch_err_throw(c, nopromote_in_flight);
 		goto err;
 	}
 
@@ -544,7 +544,7 @@ retry:
 
 	if (!bkey_and_val_eq(k, bkey_i_to_s_c(u->k.k))) {
 		/* extent we wanted to read no longer exists: */
-		rbio->ret = -BCH_ERR_data_read_key_overwritten;
+		rbio->ret = bch_err_throw(trans->c, data_read_key_overwritten);
 		goto err;
 	}
 
@@ -1035,7 +1035,7 @@ int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig,
 
 	if ((bch2_bkey_extent_flags(k) & BIT_ULL(BCH_EXTENT_FLAG_poisoned)) &&
 	    !orig->data_update)
-		return -BCH_ERR_extent_poisoned;
+		return bch_err_throw(c, extent_poisoned);
 retry_pick:
 	ret = bch2_bkey_pick_read_device(c, k, failed, &pick, dev);
 
@@ -1073,7 +1073,7 @@ retry_pick:
 
 		bch_err_ratelimited(c, "%s", buf.buf);
 		printbuf_exit(&buf);
-		ret = -BCH_ERR_data_read_no_encryption_key;
+		ret = bch_err_throw(c, data_read_no_encryption_key);
 		goto err;
 	}
 
@@ -1127,7 +1127,7 @@ retry_pick:
 			if (ca)
 				enumerated_ref_put(&ca->io_ref[READ],
 					BCH_DEV_READ_REF_io_read);
-			rbio->ret = -BCH_ERR_data_read_buffer_too_small;
+			rbio->ret = bch_err_throw(c, data_read_buffer_too_small);
 			goto out_read_done;
 		}
 
@@ -1332,7 +1332,7 @@ hole:
 	 * have to signal that:
 	 */
 	if (u)
-		orig->ret = -BCH_ERR_data_read_key_overwritten;
+		orig->ret = bch_err_throw(c, data_read_key_overwritten);
 
 	zero_fill_bio_iter(&orig->bio, iter);
 out_read_done:
@@ -1509,18 +1509,18 @@ int bch2_fs_io_read_init(struct bch_fs *c)
 					 c->opts.btree_node_size,
 					 c->opts.encoded_extent_max) /
 				   PAGE_SIZE, 0))
-		return -BCH_ERR_ENOMEM_bio_bounce_pages_init;
+		return bch_err_throw(c, ENOMEM_bio_bounce_pages_init);
 
 	if (bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio),
 			BIOSET_NEED_BVECS))
-		return -BCH_ERR_ENOMEM_bio_read_init;
+		return bch_err_throw(c, ENOMEM_bio_read_init);
 
 	if (bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio),
 			BIOSET_NEED_BVECS))
-		return -BCH_ERR_ENOMEM_bio_read_split_init;
+		return bch_err_throw(c, ENOMEM_bio_read_split_init);
 
 	if (rhashtable_init(&c->promote_table, &bch_promote_params))
-		return -BCH_ERR_ENOMEM_promote_table_init;
+		return bch_err_throw(c, ENOMEM_promote_table_init);
 
 	return 0;
 }
diff --git a/libbcachefs/io_read.h b/libbcachefs/io_read.h
index c08b9c04..45c95901 100644
--- a/libbcachefs/io_read.h
+++ b/libbcachefs/io_read.h
@@ -91,6 +91,8 @@ static inline int bch2_read_indirect_extent(struct btree_trans *trans,
 		return 0;
 
 	*data_btree = BTREE_ID_reflink;
+
+	struct bch_fs *c = trans->c;
 	struct btree_iter iter;
 	struct bkey_s_c k = bch2_lookup_indirect_extent(trans, &iter,
 						offset_into_extent,
@@ -102,10 +104,10 @@ static inline int bch2_read_indirect_extent(struct btree_trans *trans,
 
 	if (bkey_deleted(k.k)) {
 		bch2_trans_iter_exit(trans, &iter);
-		return -BCH_ERR_missing_indirect_extent;
+		return bch_err_throw(c, missing_indirect_extent);
 	}
 
-	bch2_bkey_buf_reassemble(extent, trans->c, k);
+	bch2_bkey_buf_reassemble(extent, c, k);
 	bch2_trans_iter_exit(trans, &iter);
 	return 0;
 }
diff --git a/libbcachefs/io_write.c b/libbcachefs/io_write.c
index b9d624e8..88b1eec8 100644
--- a/libbcachefs/io_write.c
+++ b/libbcachefs/io_write.c
@@ -558,6 +558,7 @@ static void bch2_write_done(struct closure *cl)
 
 static noinline int bch2_write_drop_io_error_ptrs(struct bch_write_op *op)
 {
+	struct bch_fs *c = op->c;
 	struct keylist *keys = &op->insert_keys;
 	struct bkey_i *src, *dst = keys->keys, *n;
 
@@ -569,7 +570,7 @@ static noinline int bch2_write_drop_io_error_ptrs(struct bch_write_op *op)
 					    test_bit(ptr->dev, op->failed.d));
 
 			if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(src)))
-				return -BCH_ERR_data_write_io;
+				return bch_err_throw(c, data_write_io);
 		}
 
 		if (dst != src)
@@ -976,7 +977,7 @@ csum_err:
 		op->crc.csum_type < BCH_CSUM_NR
 		? __bch2_csum_types[op->crc.csum_type]
 		: "(unknown)");
-	return -BCH_ERR_data_write_csum;
+	return bch_err_throw(c, data_write_csum);
 }
 
 static int bch2_write_extent(struct bch_write_op *op, struct write_point *wp,
@@ -1287,7 +1288,7 @@ static void bch2_nocow_write_convert_unwritten(struct bch_write_op *op)
 static void __bch2_nocow_write_done(struct bch_write_op *op)
 {
 	if (unlikely(op->flags & BCH_WRITE_io_error)) {
-		op->error = -BCH_ERR_data_write_io;
+		op->error = bch_err_throw(op->c, data_write_io);
 	} else if (unlikely(op->flags & BCH_WRITE_convert_unwritten))
 		bch2_nocow_write_convert_unwritten(op);
 }
@@ -1480,10 +1481,10 @@ err_bucket_stale:
 				    "pointer to invalid bucket in nocow path on device %llu\n  %s",
 				    stale_at->b.inode,
 				    (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-		ret = -BCH_ERR_data_write_invalid_ptr;
+		ret = bch_err_throw(c, data_write_invalid_ptr);
 	} else {
 		/* We can retry this: */
-		ret = -BCH_ERR_transaction_restart;
+		ret = bch_err_throw(c, transaction_restart);
 	}
 	printbuf_exit(&buf);
 
@@ -1690,18 +1691,18 @@ CLOSURE_CALLBACK(bch2_write)
 
 	if (unlikely(bio->bi_iter.bi_size & (c->opts.block_size - 1))) {
 		bch2_write_op_error(op, op->pos.offset, "misaligned write");
-		op->error = -BCH_ERR_data_write_misaligned;
+		op->error = bch_err_throw(c, data_write_misaligned);
 		goto err;
 	}
 
 	if (c->opts.nochanges) {
-		op->error = -BCH_ERR_erofs_no_writes;
+		op->error = bch_err_throw(c, erofs_no_writes);
 		goto err;
 	}
 
 	if (!(op->flags & BCH_WRITE_move) &&
 	    !enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_write)) {
-		op->error = -BCH_ERR_erofs_no_writes;
+		op->error = bch_err_throw(c, erofs_no_writes);
 		goto err;
 	}
 
@@ -1773,7 +1774,7 @@ int bch2_fs_io_write_init(struct bch_fs *c)
 {
 	if (bioset_init(&c->bio_write,   1, offsetof(struct bch_write_bio, bio), BIOSET_NEED_BVECS) ||
 	    bioset_init(&c->replica_set, 4, offsetof(struct bch_write_bio, bio), 0))
-		return -BCH_ERR_ENOMEM_bio_write_init;
+		return bch_err_throw(c, ENOMEM_bio_write_init);
 
 	return 0;
 }
diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c
index 0348ab32..dda802a6 100644
--- a/libbcachefs/journal.c
+++ b/libbcachefs/journal.c
@@ -397,7 +397,7 @@ static int journal_entry_open(struct journal *j)
 	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 
 	if (j->blocked)
-		return -BCH_ERR_journal_blocked;
+		return bch_err_throw(c, journal_blocked);
 
 	if (j->cur_entry_error)
 		return j->cur_entry_error;
@@ -407,23 +407,23 @@ static int journal_entry_open(struct journal *j)
 		return ret;
 
 	if (!fifo_free(&j->pin))
-		return -BCH_ERR_journal_pin_full;
+		return bch_err_throw(c, journal_pin_full);
 
 	if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
-		return -BCH_ERR_journal_max_in_flight;
+		return bch_err_throw(c, journal_max_in_flight);
 
 	if (atomic64_read(&j->seq) - j->seq_write_started == JOURNAL_STATE_BUF_NR)
-		return -BCH_ERR_journal_max_open;
+		return bch_err_throw(c, journal_max_open);
 
 	if (unlikely(journal_cur_seq(j) >= JOURNAL_SEQ_MAX)) {
 		bch_err(c, "cannot start: journal seq overflow");
 		if (bch2_fs_emergency_read_only_locked(c))
 			bch_err(c, "fatal error - emergency read only");
-		return -BCH_ERR_journal_shutdown;
+		return bch_err_throw(c, journal_shutdown);
 	}
 
 	if (!j->free_buf && !buf->data)
-		return -BCH_ERR_journal_buf_enomem; /* will retry after write completion frees up a buf */
+		return bch_err_throw(c, journal_buf_enomem); /* will retry after write completion frees up a buf */
 
 	BUG_ON(!j->cur_entry_sectors);
 
@@ -447,7 +447,7 @@ static int journal_entry_open(struct journal *j)
 	u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 
 	if (u64s <= (ssize_t) j->early_journal_entries.nr)
-		return -BCH_ERR_journal_full;
+		return bch_err_throw(c, journal_full);
 
 	if (fifo_empty(&j->pin) && j->reclaim_thread)
 		wake_up_process(j->reclaim_thread);
@@ -464,7 +464,7 @@ static int journal_entry_open(struct journal *j)
 			journal_cur_seq(j));
 		if (bch2_fs_emergency_read_only_locked(c))
 			bch_err(c, "fatal error - emergency read only");
-		return -BCH_ERR_journal_shutdown;
+		return bch_err_throw(c, journal_shutdown);
 	}
 
 	BUG_ON(j->pin.back - 1 != atomic64_read(&j->seq));
@@ -597,16 +597,16 @@ retry:
 		return ret;
 
 	if (j->blocked)
-		return -BCH_ERR_journal_blocked;
+		return bch_err_throw(c, journal_blocked);
 
 	if ((flags & BCH_WATERMARK_MASK) < j->watermark) {
-		ret = -BCH_ERR_journal_full;
+		ret = bch_err_throw(c, journal_full);
 		can_discard = j->can_discard;
 		goto out;
 	}
 
 	if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf) && !journal_entry_is_open(j)) {
-		ret = -BCH_ERR_journal_max_in_flight;
+		ret = bch_err_throw(c, journal_max_in_flight);
 		goto out;
 	}
 
@@ -647,7 +647,7 @@ out:
 		goto retry;
 
 	if (journal_error_check_stuck(j, ret, flags))
-		ret = -BCH_ERR_journal_stuck;
+		ret = bch_err_throw(c, journal_stuck);
 
 	if (ret == -BCH_ERR_journal_max_in_flight &&
 	    track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight], true) &&
@@ -812,6 +812,7 @@ out:
 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 				 struct closure *parent)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_buf *buf;
 	int ret = 0;
 
@@ -827,7 +828,7 @@ int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 
 	/* Recheck under lock: */
 	if (j->err_seq && seq >= j->err_seq) {
-		ret = -BCH_ERR_journal_flush_err;
+		ret = bch_err_throw(c, journal_flush_err);
 		goto out;
 	}
 
@@ -998,7 +999,7 @@ int bch2_journal_meta(struct journal *j)
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_journal))
-		return -BCH_ERR_erofs_no_writes;
+		return bch_err_throw(c, erofs_no_writes);
 
 	int ret = __bch2_journal_meta(j);
 	enumerated_ref_put(&c->writes, BCH_WRITE_REF_journal);
@@ -1131,7 +1132,7 @@ static int bch2_set_nr_journal_buckets_iter(struct bch_dev *ca, unsigned nr,
 	new_buckets	= kcalloc(nr, sizeof(u64), GFP_KERNEL);
 	new_bucket_seq	= kcalloc(nr, sizeof(u64), GFP_KERNEL);
 	if (!bu || !ob || !new_buckets || !new_bucket_seq) {
-		ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
+		ret = bch_err_throw(c, ENOMEM_set_nr_journal_buckets);
 		goto err_free;
 	}
 
@@ -1322,7 +1323,7 @@ int bch2_dev_journal_bucket_delete(struct bch_dev *ca, u64 b)
 
 	u64 *new_buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);;
 	if (!new_buckets)
-		return -BCH_ERR_ENOMEM_set_nr_journal_buckets;
+		return bch_err_throw(c, ENOMEM_set_nr_journal_buckets);
 
 	memcpy(new_buckets, ja->buckets, ja->nr * sizeof(u64));
 	memmove(&new_buckets[pos],
@@ -1372,14 +1373,14 @@ int bch2_dev_journal_alloc(struct bch_dev *ca, bool new_fs)
 
 	if (c->sb.features & BIT_ULL(BCH_FEATURE_small_image)) {
 		bch_err(c, "cannot allocate journal, filesystem is an unresized image file");
-		return -BCH_ERR_erofs_filesystem_full;
+		return bch_err_throw(c, erofs_filesystem_full);
 	}
 
 	unsigned nr;
 	int ret;
 
 	if (dynamic_fault("bcachefs:add:journal_alloc")) {
-		ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
+		ret = bch_err_throw(c, ENOMEM_set_nr_journal_buckets);
 		goto err;
 	}
 
@@ -1518,7 +1519,7 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
 	init_fifo(&j->pin, roundup_pow_of_two(nr), GFP_KERNEL);
 	if (!j->pin.data) {
 		bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
-		return -BCH_ERR_ENOMEM_journal_pin_fifo;
+		return bch_err_throw(c, ENOMEM_journal_pin_fifo);
 	}
 
 	j->replay_journal_seq	= last_seq;
@@ -1606,6 +1607,7 @@ void bch2_dev_journal_exit(struct bch_dev *ca)
 
 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 {
+	struct bch_fs *c = ca->fs;
 	struct journal_device *ja = &ca->journal;
 	struct bch_sb_field_journal *journal_buckets =
 		bch2_sb_field_get(sb, journal);
@@ -1625,7 +1627,7 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 
 	ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
 	if (!ja->bucket_seq)
-		return -BCH_ERR_ENOMEM_dev_journal_init;
+		return bch_err_throw(c, ENOMEM_dev_journal_init);
 
 	unsigned nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
 
@@ -1633,7 +1635,7 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 		ja->bio[i] = kzalloc(struct_size(ja->bio[i], bio.bi_inline_vecs,
 				     nr_bvecs), GFP_KERNEL);
 		if (!ja->bio[i])
-			return -BCH_ERR_ENOMEM_dev_journal_init;
+			return bch_err_throw(c, ENOMEM_dev_journal_init);
 
 		ja->bio[i]->ca = ca;
 		ja->bio[i]->buf_idx = i;
@@ -1642,7 +1644,7 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 
 	ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
 	if (!ja->buckets)
-		return -BCH_ERR_ENOMEM_dev_journal_init;
+		return bch_err_throw(c, ENOMEM_dev_journal_init);
 
 	if (journal_buckets_v2) {
 		unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
@@ -1696,10 +1698,12 @@ void bch2_fs_journal_init_early(struct journal *j)
 
 int bch2_fs_journal_init(struct journal *j)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+
 	j->free_buf_size = j->buf_size_want = JOURNAL_ENTRY_SIZE_MIN;
 	j->free_buf = kvmalloc(j->free_buf_size, GFP_KERNEL);
 	if (!j->free_buf)
-		return -BCH_ERR_ENOMEM_journal_buf;
+		return bch_err_throw(c, ENOMEM_journal_buf);
 
 	for (unsigned i = 0; i < ARRAY_SIZE(j->buf); i++)
 		j->buf[i].idx = i;
@@ -1707,7 +1711,7 @@ int bch2_fs_journal_init(struct journal *j)
 	j->wq = alloc_workqueue("bcachefs_journal",
 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512);
 	if (!j->wq)
-		return -BCH_ERR_ENOMEM_fs_other_alloc;
+		return bch_err_throw(c, ENOMEM_fs_other_alloc);
 	return 0;
 }
 
diff --git a/libbcachefs/journal_io.c b/libbcachefs/journal_io.c
index 8ce41753..d4c9980a 100644
--- a/libbcachefs/journal_io.c
+++ b/libbcachefs/journal_io.c
@@ -199,7 +199,7 @@ static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
 				journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
 				GFP_KERNEL);
 	if (!_i)
-		return -BCH_ERR_ENOMEM_journal_entry_add;
+		return bch_err_throw(c, ENOMEM_journal_entry_add);
 
 	/*
 	 * Duplicate journal entries? If so we want the one that didn't have a
@@ -242,7 +242,7 @@ static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
 replace:
 	i = kvmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
 	if (!i)
-		return -BCH_ERR_ENOMEM_journal_entry_add;
+		return bch_err_throw(c, ENOMEM_journal_entry_add);
 
 	darray_init(&i->ptrs);
 	i->csum_good		= entry_ptr.csum_good;
@@ -322,7 +322,7 @@ static void journal_entry_err_msg(struct printbuf *out,
 		bch2_sb_error_count(c, BCH_FSCK_ERR_##_err);		\
 		if (bch2_fs_inconsistent(c,				\
 				"corrupt metadata before write: %s\n", _buf.buf)) {\
-			ret = -BCH_ERR_fsck_errors_not_fixed;		\
+			ret = bch_err_throw(c, fsck_errors_not_fixed);		\
 			goto fsck_err;					\
 		}							\
 		break;							\
@@ -1020,19 +1020,19 @@ struct journal_read_buf {
 	size_t		size;
 };
 
-static int journal_read_buf_realloc(struct journal_read_buf *b,
+static int journal_read_buf_realloc(struct bch_fs *c, struct journal_read_buf *b,
 				    size_t new_size)
 {
 	void *n;
 
 	/* the bios are sized for this many pages, max: */
 	if (new_size > JOURNAL_ENTRY_SIZE_MAX)
-		return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
+		return bch_err_throw(c, ENOMEM_journal_read_buf_realloc);
 
 	new_size = roundup_pow_of_two(new_size);
 	n = kvmalloc(new_size, GFP_KERNEL);
 	if (!n)
-		return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
+		return bch_err_throw(c, ENOMEM_journal_read_buf_realloc);
 
 	kvfree(b->data);
 	b->data = n;
@@ -1067,7 +1067,7 @@ reread:
 
 			bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 			if (!bio)
-				return -BCH_ERR_ENOMEM_journal_read_bucket;
+				return bch_err_throw(c, ENOMEM_journal_read_bucket);
 			bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
 
 			bio->bi_iter.bi_sector = offset;
@@ -1078,7 +1078,7 @@ reread:
 			kfree(bio);
 
 			if (!ret && bch2_meta_read_fault("journal"))
-				ret = -BCH_ERR_EIO_fault_injected;
+				ret = bch_err_throw(c, EIO_fault_injected);
 
 			bch2_account_io_completion(ca, BCH_MEMBER_ERROR_read,
 						   submit_time, !ret);
@@ -1106,7 +1106,7 @@ reread:
 			break;
 		case JOURNAL_ENTRY_REREAD:
 			if (vstruct_bytes(j) > buf->size) {
-				ret = journal_read_buf_realloc(buf,
+				ret = journal_read_buf_realloc(c, buf,
 							vstruct_bytes(j));
 				if (ret)
 					return ret;
@@ -1206,7 +1206,7 @@ static CLOSURE_CALLBACK(bch2_journal_read_device)
 	if (!ja->nr)
 		goto out;
 
-	ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
+	ret = journal_read_buf_realloc(c, &buf, PAGE_SIZE);
 	if (ret)
 		goto err;
 
@@ -1691,7 +1691,7 @@ static CLOSURE_CALLBACK(journal_write_done)
 			       : j->noflush_write_time, j->write_start_time);
 
 	if (!w->devs_written.nr) {
-		err = -BCH_ERR_journal_write_err;
+		err = bch_err_throw(c, journal_write_err);
 	} else {
 		bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
 					 w->devs_written);
diff --git a/libbcachefs/journal_reclaim.c b/libbcachefs/journal_reclaim.c
index 67be0c33..cd620174 100644
--- a/libbcachefs/journal_reclaim.c
+++ b/libbcachefs/journal_reclaim.c
@@ -226,7 +226,7 @@ void bch2_journal_space_available(struct journal *j)
 			bch_err(c, "%s", buf.buf);
 			printbuf_exit(&buf);
 		}
-		ret = -BCH_ERR_insufficient_journal_devices;
+		ret = bch_err_throw(c, insufficient_journal_devices);
 		goto out;
 	}
 
@@ -240,7 +240,7 @@ void bch2_journal_space_available(struct journal *j)
 	total		= j->space[journal_space_total].total;
 
 	if (!j->space[journal_space_discarded].next_entry)
-		ret = -BCH_ERR_journal_full;
+		ret = bch_err_throw(c, journal_full);
 
 	if ((j->space[journal_space_clean_ondisk].next_entry <
 	     j->space[journal_space_clean_ondisk].total) &&
diff --git a/libbcachefs/journal_sb.c b/libbcachefs/journal_sb.c
index 62b910f2..0cb9b93f 100644
--- a/libbcachefs/journal_sb.c
+++ b/libbcachefs/journal_sb.c
@@ -210,7 +210,7 @@ int bch2_journal_buckets_to_sb(struct bch_fs *c, struct bch_dev *ca,
 	j = bch2_sb_field_resize(&ca->disk_sb, journal_v2,
 			 (sizeof(*j) + sizeof(j->d[0]) * nr_compacted) / sizeof(u64));
 	if (!j)
-		return -BCH_ERR_ENOSPC_sb_journal;
+		return bch_err_throw(c, ENOSPC_sb_journal);
 
 	bch2_sb_field_delete(&ca->disk_sb, BCH_SB_FIELD_journal);
 
diff --git a/libbcachefs/journal_seq_blacklist.c b/libbcachefs/journal_seq_blacklist.c
index c5a7d800..af4fe416 100644
--- a/libbcachefs/journal_seq_blacklist.c
+++ b/libbcachefs/journal_seq_blacklist.c
@@ -78,7 +78,7 @@ int bch2_journal_seq_blacklist_add(struct bch_fs *c, u64 start, u64 end)
 	bl = bch2_sb_field_resize(&c->disk_sb, journal_seq_blacklist,
 				  sb_blacklist_u64s(nr + 1));
 	if (!bl) {
-		ret = -BCH_ERR_ENOSPC_sb_journal_seq_blacklist;
+		ret = bch_err_throw(c, ENOSPC_sb_journal_seq_blacklist);
 		goto out;
 	}
 
@@ -152,7 +152,7 @@ int bch2_blacklist_table_initialize(struct bch_fs *c)
 
 	t = kzalloc(struct_size(t, entries, nr), GFP_KERNEL);
 	if (!t)
-		return -BCH_ERR_ENOMEM_blacklist_table_init;
+		return bch_err_throw(c, ENOMEM_blacklist_table_init);
 
 	t->nr = nr;
 
diff --git a/libbcachefs/migrate.c b/libbcachefs/migrate.c
index bb7a9227..f296cce9 100644
--- a/libbcachefs/migrate.c
+++ b/libbcachefs/migrate.c
@@ -35,7 +35,7 @@ static int drop_dev_ptrs(struct bch_fs *c, struct bkey_s k,
 	nr_good = bch2_bkey_durability(c, k.s_c);
 	if ((!nr_good && !(flags & lost)) ||
 	    (nr_good < replicas && !(flags & degraded)))
-		return -BCH_ERR_remove_would_lose_data;
+		return bch_err_throw(c, remove_would_lose_data);
 
 	return 0;
 }
@@ -156,7 +156,7 @@ static int bch2_dev_metadata_drop(struct bch_fs *c,
 
 	/* don't handle this yet: */
 	if (flags & BCH_FORCE_IF_METADATA_LOST)
-		return -BCH_ERR_remove_with_metadata_missing_unimplemented;
+		return bch_err_throw(c, remove_with_metadata_missing_unimplemented);
 
 	trans = bch2_trans_get(c);
 	bch2_bkey_buf_init(&k);
diff --git a/libbcachefs/move.c b/libbcachefs/move.c
index 5dafc001..eec591e9 100644
--- a/libbcachefs/move.c
+++ b/libbcachefs/move.c
@@ -971,7 +971,7 @@ static int __bch2_move_data_phys(struct moving_context *ctxt,
 		if (data_opts.scrub &&
 		    !bch2_dev_idx_is_online(c, data_opts.read_dev)) {
 			bch2_trans_iter_exit(trans, &iter);
-			ret = -BCH_ERR_device_offline;
+			ret = bch_err_throw(c, device_offline);
 			break;
 		}
 
diff --git a/libbcachefs/namei.c b/libbcachefs/namei.c
index ee6fe35c..24120037 100644
--- a/libbcachefs/namei.c
+++ b/libbcachefs/namei.c
@@ -287,7 +287,7 @@ int bch2_unlink_trans(struct btree_trans *trans,
 	}
 
 	if (deleting_subvol && !inode_u->bi_subvol) {
-		ret = -BCH_ERR_ENOENT_not_subvol;
+		ret = bch_err_throw(c, ENOENT_not_subvol);
 		goto err;
 	}
 
@@ -633,7 +633,7 @@ static int __bch2_inum_to_path(struct btree_trans *trans,
 			break;
 
 		if (!inode.bi_dir && !inode.bi_dir_offset) {
-			ret = -BCH_ERR_ENOENT_inode_no_backpointer;
+			ret = bch_err_throw(trans->c, ENOENT_inode_no_backpointer);
 			goto disconnected;
 		}
 
diff --git a/libbcachefs/quota.c b/libbcachefs/quota.c
index 3d4755d7..f241efb1 100644
--- a/libbcachefs/quota.c
+++ b/libbcachefs/quota.c
@@ -527,7 +527,7 @@ int bch2_fs_quota_read(struct bch_fs *c)
 	struct bch_sb_field_quota *sb_quota = bch2_sb_get_or_create_quota(&c->disk_sb);
 	if (!sb_quota) {
 		mutex_unlock(&c->sb_lock);
-		return -BCH_ERR_ENOSPC_sb_quota;
+		return bch_err_throw(c, ENOSPC_sb_quota);
 	}
 
 	bch2_sb_quota_read(c);
@@ -572,7 +572,7 @@ static int bch2_quota_enable(struct super_block	*sb, unsigned uflags)
 	mutex_lock(&c->sb_lock);
 	sb_quota = bch2_sb_get_or_create_quota(&c->disk_sb);
 	if (!sb_quota) {
-		ret = -BCH_ERR_ENOSPC_sb_quota;
+		ret = bch_err_throw(c, ENOSPC_sb_quota);
 		goto unlock;
 	}
 
@@ -726,7 +726,7 @@ static int bch2_quota_set_info(struct super_block *sb, int type,
 	mutex_lock(&c->sb_lock);
 	sb_quota = bch2_sb_get_or_create_quota(&c->disk_sb);
 	if (!sb_quota) {
-		ret = -BCH_ERR_ENOSPC_sb_quota;
+		ret = bch_err_throw(c, ENOSPC_sb_quota);
 		goto unlock;
 	}
 
diff --git a/libbcachefs/rebalance.c b/libbcachefs/rebalance.c
index c563797e..1c345b86 100644
--- a/libbcachefs/rebalance.c
+++ b/libbcachefs/rebalance.c
@@ -443,7 +443,7 @@ static int do_rebalance_extent(struct moving_context *ctxt,
 		if (bch2_err_matches(ret, ENOMEM)) {
 			/* memory allocation failure, wait for some IO to finish */
 			bch2_move_ctxt_wait_for_io(ctxt);
-			ret = -BCH_ERR_transaction_restart_nested;
+			ret = bch_err_throw(c, transaction_restart_nested);
 		}
 
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
@@ -795,7 +795,7 @@ static int check_rebalance_work_one(struct btree_trans *trans,
 				     BTREE_ID_extents, POS_MIN,
 				     BTREE_ITER_prefetch|
 				     BTREE_ITER_all_snapshots);
-		return -BCH_ERR_transaction_restart_nested;
+		return bch_err_throw(c, transaction_restart_nested);
 	}
 
 	if (!extent_k.k && !rebalance_k.k)
diff --git a/libbcachefs/recovery.c b/libbcachefs/recovery.c
index 4b51105b..1e68e61f 100644
--- a/libbcachefs/recovery.c
+++ b/libbcachefs/recovery.c
@@ -879,7 +879,7 @@ int bch2_fs_recovery(struct bch_fs *c)
 use_clean:
 		if (!clean) {
 			bch_err(c, "no superblock clean section found");
-			ret = -BCH_ERR_fsck_repair_impossible;
+			ret = bch_err_throw(c, fsck_repair_impossible);
 			goto err;
 
 		}
diff --git a/libbcachefs/recovery_passes.c b/libbcachefs/recovery_passes.c
index 212658cb..605588e3 100644
--- a/libbcachefs/recovery_passes.c
+++ b/libbcachefs/recovery_passes.c
@@ -103,20 +103,20 @@ static void bch2_sb_recovery_passes_to_text(struct printbuf *out,
 		prt_tab(out);
 
 		bch2_pr_time_units(out, le32_to_cpu(i->last_runtime) * NSEC_PER_SEC);
+
+		if (BCH_RECOVERY_PASS_NO_RATELIMIT(i))
+			prt_str(out, " (no ratelimit)");
+
 		prt_newline(out);
 	}
 }
 
-static void bch2_sb_recovery_pass_complete(struct bch_fs *c,
-					   enum bch_recovery_pass pass,
-					   s64 start_time)
+static struct recovery_pass_entry *bch2_sb_recovery_pass_entry(struct bch_fs *c,
+							       enum bch_recovery_pass pass)
 {
 	enum bch_recovery_pass_stable stable = bch2_recovery_pass_to_stable(pass);
-	s64 end_time = ktime_get_real_seconds();
 
-	mutex_lock(&c->sb_lock);
-	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
-	__clear_bit_le64(stable, ext->recovery_passes_required);
+	lockdep_assert_held(&c->sb_lock);
 
 	struct bch_sb_field_recovery_passes *r =
 		bch2_sb_field_get(c->disk_sb.sb, recovery_passes);
@@ -127,15 +127,43 @@ static void bch2_sb_recovery_pass_complete(struct bch_fs *c,
 		r = bch2_sb_field_resize(&c->disk_sb, recovery_passes, u64s);
 		if (!r) {
 			bch_err(c, "error creating recovery_passes sb section");
-			goto out;
+			return NULL;
 		}
 	}
 
-	r->start[stable].last_run	= cpu_to_le64(end_time);
-	r->start[stable].last_runtime	= cpu_to_le32(max(0, end_time - start_time));
-out:
+	return r->start + stable;
+}
+
+static void bch2_sb_recovery_pass_complete(struct bch_fs *c,
+					   enum bch_recovery_pass pass,
+					   s64 start_time)
+{
+	guard(mutex)(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
+	__clear_bit_le64(bch2_recovery_pass_to_stable(pass),
+			 ext->recovery_passes_required);
+
+	struct recovery_pass_entry *e = bch2_sb_recovery_pass_entry(c, pass);
+	if (e) {
+		s64 end_time	= ktime_get_real_seconds();
+		e->last_run	= cpu_to_le64(end_time);
+		e->last_runtime	= cpu_to_le32(max(0, end_time - start_time));
+		SET_BCH_RECOVERY_PASS_NO_RATELIMIT(e, false);
+	}
+
 	bch2_write_super(c);
-	mutex_unlock(&c->sb_lock);
+}
+
+void bch2_recovery_pass_set_no_ratelimit(struct bch_fs *c,
+					 enum bch_recovery_pass pass)
+{
+	guard(mutex)(&c->sb_lock);
+
+	struct recovery_pass_entry *e = bch2_sb_recovery_pass_entry(c, pass);
+	if (e && !BCH_RECOVERY_PASS_NO_RATELIMIT(e)) {
+		SET_BCH_RECOVERY_PASS_NO_RATELIMIT(e, false);
+		bch2_write_super(c);
+	}
 }
 
 static bool bch2_recovery_pass_want_ratelimit(struct bch_fs *c, enum bch_recovery_pass pass)
@@ -157,6 +185,9 @@ static bool bch2_recovery_pass_want_ratelimit(struct bch_fs *c, enum bch_recover
 		 */
 		ret = (u64) le32_to_cpu(i->last_runtime) * 100 >
 			ktime_get_real_seconds() - le64_to_cpu(i->last_run);
+
+		if (BCH_RECOVERY_PASS_NO_RATELIMIT(i))
+			ret = false;
 	}
 
 	return ret;
@@ -329,7 +360,7 @@ int __bch2_run_explicit_recovery_pass(struct bch_fs *c,
 	    (!in_recovery || r->curr_pass >= BCH_RECOVERY_PASS_set_may_go_rw)) {
 		prt_printf(out, "need recovery pass %s (%u), but already rw\n",
 			   bch2_recovery_passes[pass], pass);
-		ret = -BCH_ERR_cannot_rewind_recovery;
+		ret = bch_err_throw(c, cannot_rewind_recovery);
 		goto out;
 	}
 
@@ -349,7 +380,7 @@ int __bch2_run_explicit_recovery_pass(struct bch_fs *c,
 		if (rewind) {
 			r->next_pass = pass;
 			r->passes_complete &= (1ULL << pass) >> 1;
-			ret = -BCH_ERR_restart_recovery;
+			ret = bch_err_throw(c, restart_recovery);
 		}
 	} else {
 		prt_printf(out, "scheduling recovery pass %s (%u)%s\n",
@@ -384,6 +415,35 @@ int bch2_run_explicit_recovery_pass(struct bch_fs *c,
 	return ret;
 }
 
+/*
+ * Returns 0 if @pass has run recently, otherwise one of
+ * -BCH_ERR_restart_recovery
+ * -BCH_ERR_recovery_pass_will_run
+ */
+int bch2_require_recovery_pass(struct bch_fs *c,
+			       struct printbuf *out,
+			       enum bch_recovery_pass pass)
+{
+	if (test_bit(BCH_FS_in_recovery, &c->flags) &&
+	    c->recovery.passes_complete & BIT_ULL(pass))
+		return 0;
+
+	guard(mutex)(&c->sb_lock);
+
+	if (bch2_recovery_pass_want_ratelimit(c, pass))
+		return 0;
+
+	enum bch_run_recovery_pass_flags flags = 0;
+	int ret = 0;
+
+	if (recovery_pass_needs_set(c, pass, &flags)) {
+		ret = __bch2_run_explicit_recovery_pass(c, out, pass, flags);
+		bch2_write_super(c);
+	}
+
+	return ret ?: bch_err_throw(c, recovery_pass_will_run);
+}
+
 int bch2_run_print_explicit_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 {
 	enum bch_run_recovery_pass_flags flags = RUN_RECOVERY_PASS_nopersistent;
diff --git a/libbcachefs/recovery_passes.h b/libbcachefs/recovery_passes.h
index dc0d2014..260571c7 100644
--- a/libbcachefs/recovery_passes.h
+++ b/libbcachefs/recovery_passes.h
@@ -10,6 +10,8 @@ u64 bch2_recovery_passes_from_stable(u64 v);
 
 u64 bch2_fsck_recovery_passes(void);
 
+void bch2_recovery_pass_set_no_ratelimit(struct bch_fs *, enum bch_recovery_pass);
+
 enum bch_run_recovery_pass_flags {
 	RUN_RECOVERY_PASS_nopersistent	= BIT(0),
 	RUN_RECOVERY_PASS_ratelimit	= BIT(1),
@@ -24,6 +26,9 @@ int bch2_run_explicit_recovery_pass(struct bch_fs *, struct printbuf *,
 				    enum bch_recovery_pass,
 				    enum bch_run_recovery_pass_flags);
 
+int bch2_require_recovery_pass(struct bch_fs *, struct printbuf *,
+			       enum bch_recovery_pass);
+
 int bch2_run_online_recovery_passes(struct bch_fs *, u64);
 int bch2_run_recovery_passes(struct bch_fs *, enum bch_recovery_pass);
 
diff --git a/libbcachefs/recovery_passes_format.h b/libbcachefs/recovery_passes_format.h
index c434eafb..b63c2055 100644
--- a/libbcachefs/recovery_passes_format.h
+++ b/libbcachefs/recovery_passes_format.h
@@ -87,6 +87,8 @@ struct recovery_pass_entry {
 	__le32			flags;
 };
 
+LE32_BITMASK(BCH_RECOVERY_PASS_NO_RATELIMIT,	struct recovery_pass_entry, flags, 0, 1)
+
 struct bch_sb_field_recovery_passes {
 	struct bch_sb_field	field;
 	struct recovery_pass_entry start[];
diff --git a/libbcachefs/reflink.c b/libbcachefs/reflink.c
index 41ca86cb..a535abd4 100644
--- a/libbcachefs/reflink.c
+++ b/libbcachefs/reflink.c
@@ -312,7 +312,7 @@ static int trans_trigger_reflink_p_segment(struct btree_trans *trans,
 
 	if (!bkey_refcount_c(k)) {
 		if (!(flags & BTREE_TRIGGER_overwrite))
-			ret = -BCH_ERR_missing_indirect_extent;
+			ret = bch_err_throw(c, missing_indirect_extent);
 		goto next;
 	}
 
@@ -612,7 +612,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 	int ret = 0, ret2 = 0;
 
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_reflink))
-		return -BCH_ERR_erofs_no_writes;
+		return bch_err_throw(c, erofs_no_writes);
 
 	bch2_check_set_feature(c, BCH_FEATURE_reflink);
 
@@ -848,7 +848,7 @@ int bch2_gc_reflink_start(struct bch_fs *c)
 			struct reflink_gc *r = genradix_ptr_alloc(&c->reflink_gc_table,
 							c->reflink_gc_nr++, GFP_KERNEL);
 			if (!r) {
-				ret = -BCH_ERR_ENOMEM_gc_reflink_start;
+				ret = bch_err_throw(c, ENOMEM_gc_reflink_start);
 				break;
 			}
 
diff --git a/libbcachefs/replicas.c b/libbcachefs/replicas.c
index 4698f886..8383bd7f 100644
--- a/libbcachefs/replicas.c
+++ b/libbcachefs/replicas.c
@@ -119,7 +119,7 @@ int bch2_replicas_entry_validate(struct bch_replicas_entry_v1 *r,
 	return 0;
 bad:
 	bch2_replicas_entry_to_text(err, r);
-	return -BCH_ERR_invalid_replicas_entry;
+	return bch_err_throw(c, invalid_replicas_entry);
 }
 
 void bch2_cpu_replicas_to_text(struct printbuf *out,
@@ -311,7 +311,7 @@ static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 	    !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 		new_gc = cpu_replicas_add_entry(c, &c->replicas_gc, new_entry);
 		if (!new_gc.entries) {
-			ret = -BCH_ERR_ENOMEM_cpu_replicas;
+			ret = bch_err_throw(c, ENOMEM_cpu_replicas);
 			goto err;
 		}
 	}
@@ -319,7 +319,7 @@ static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 	if (!__replicas_has_entry(&c->replicas, new_entry)) {
 		new_r = cpu_replicas_add_entry(c, &c->replicas, new_entry);
 		if (!new_r.entries) {
-			ret = -BCH_ERR_ENOMEM_cpu_replicas;
+			ret = bch_err_throw(c, ENOMEM_cpu_replicas);
 			goto err;
 		}
 
@@ -422,7 +422,7 @@ int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 	if (!c->replicas_gc.entries) {
 		mutex_unlock(&c->sb_lock);
 		bch_err(c, "error allocating c->replicas_gc");
-		return -BCH_ERR_ENOMEM_replicas_gc;
+		return bch_err_throw(c, ENOMEM_replicas_gc);
 	}
 
 	for_each_cpu_replicas_entry(&c->replicas, e)
@@ -458,7 +458,7 @@ retry:
 	new.entries	= kcalloc(nr, new.entry_size, GFP_KERNEL);
 	if (!new.entries) {
 		bch_err(c, "error allocating c->replicas_gc");
-		return -BCH_ERR_ENOMEM_replicas_gc;
+		return bch_err_throw(c, ENOMEM_replicas_gc);
 	}
 
 	mutex_lock(&c->sb_lock);
@@ -622,7 +622,7 @@ static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 	sb_r = bch2_sb_field_resize(&c->disk_sb, replicas_v0,
 			DIV_ROUND_UP(bytes, sizeof(u64)));
 	if (!sb_r)
-		return -BCH_ERR_ENOSPC_sb_replicas;
+		return bch_err_throw(c, ENOSPC_sb_replicas);
 
 	bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 	sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas_v0);
@@ -667,7 +667,7 @@ static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 	sb_r = bch2_sb_field_resize(&c->disk_sb, replicas,
 			DIV_ROUND_UP(bytes, sizeof(u64)));
 	if (!sb_r)
-		return -BCH_ERR_ENOSPC_sb_replicas;
+		return bch_err_throw(c, ENOSPC_sb_replicas);
 
 	bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 	sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas);
diff --git a/libbcachefs/sb-downgrade.c b/libbcachefs/sb-downgrade.c
index 861fce16..b61f8845 100644
--- a/libbcachefs/sb-downgrade.c
+++ b/libbcachefs/sb-downgrade.c
@@ -417,7 +417,7 @@ int bch2_sb_downgrade_update(struct bch_fs *c)
 
 	d = bch2_sb_field_resize(&c->disk_sb, downgrade, sb_u64s);
 	if (!d) {
-		ret = -BCH_ERR_ENOSPC_sb_downgrade;
+		ret = bch_err_throw(c, ENOSPC_sb_downgrade);
 		goto out;
 	}
 
diff --git a/libbcachefs/sb-members.c b/libbcachefs/sb-members.c
index ab5673b3..363eb0c6 100644
--- a/libbcachefs/sb-members.c
+++ b/libbcachefs/sb-members.c
@@ -101,7 +101,7 @@ static int sb_members_v2_resize_entries(struct bch_fs *c)
 
 		mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
 		if (!mi)
-			return -BCH_ERR_ENOSPC_sb_members_v2;
+			return bch_err_throw(c, ENOSPC_sb_members_v2);
 
 		for (int i = c->disk_sb.sb->nr_devices - 1; i >= 0; --i) {
 			void *dst = (void *) mi->_members + (i * sizeof(struct bch_member));
diff --git a/libbcachefs/snapshot.c b/libbcachefs/snapshot.c
index 612c526a..23a332d7 100644
--- a/libbcachefs/snapshot.c
+++ b/libbcachefs/snapshot.c
@@ -54,7 +54,7 @@ int bch2_snapshot_tree_lookup(struct btree_trans *trans, u32 id,
 					  BTREE_ITER_with_updates, snapshot_tree, s);
 
 	if (bch2_err_matches(ret, ENOENT))
-		ret = -BCH_ERR_ENOENT_snapshot_tree;
+		ret = bch_err_throw(trans->c, ENOENT_snapshot_tree);
 	return ret;
 }
 
@@ -67,7 +67,7 @@ __bch2_snapshot_tree_create(struct btree_trans *trans)
 	struct bkey_i_snapshot_tree *s_t;
 
 	if (ret == -BCH_ERR_ENOSPC_btree_slot)
-		ret = -BCH_ERR_ENOSPC_snapshot_tree;
+		ret = bch_err_throw(trans->c, ENOSPC_snapshot_tree);
 	if (ret)
 		return ERR_PTR(ret);
 
@@ -285,7 +285,7 @@ static int bch2_snapshot_table_make_room(struct bch_fs *c, u32 id)
 	mutex_lock(&c->snapshot_table_lock);
 	int ret = snapshot_t_mut(c, id)
 		? 0
-		: -BCH_ERR_ENOMEM_mark_snapshot;
+		: bch_err_throw(c, ENOMEM_mark_snapshot);
 	mutex_unlock(&c->snapshot_table_lock);
 	return ret;
 }
@@ -304,7 +304,7 @@ static int __bch2_mark_snapshot(struct btree_trans *trans,
 
 	t = snapshot_t_mut(c, id);
 	if (!t) {
-		ret = -BCH_ERR_ENOMEM_mark_snapshot;
+		ret = bch_err_throw(c, ENOMEM_mark_snapshot);
 		goto err;
 	}
 
@@ -1006,7 +1006,7 @@ int bch2_reconstruct_snapshots(struct bch_fs *c)
 					"snapshot node %u from tree %s missing, recreate?", *id, buf.buf)) {
 				if (t->nr > 1) {
 					bch_err(c, "cannot reconstruct snapshot trees with multiple nodes");
-					ret = -BCH_ERR_fsck_repair_unimplemented;
+					ret = bch_err_throw(c, fsck_repair_unimplemented);
 					goto err;
 				}
 
@@ -1045,19 +1045,39 @@ int __bch2_check_key_has_snapshot(struct btree_trans *trans,
 		ret = bch2_btree_delete_at(trans, iter,
 					   BTREE_UPDATE_internal_snapshot_node) ?: 1;
 
-	/*
-	 * Snapshot missing: we should have caught this with btree_lost_data and
-	 * kicked off reconstruct_snapshots, so if we end up here we have no
-	 * idea what happened:
-	 */
-	if (fsck_err_on(state == SNAPSHOT_ID_empty,
-			trans, bkey_in_missing_snapshot,
-			"key in missing snapshot %s, delete?",
-			(bch2_btree_id_to_text(&buf, iter->btree_id),
-			 prt_char(&buf, ' '),
-			 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
-		ret = bch2_btree_delete_at(trans, iter,
-					   BTREE_UPDATE_internal_snapshot_node) ?: 1;
+	if (state == SNAPSHOT_ID_empty) {
+		/*
+		 * Snapshot missing: we should have caught this with btree_lost_data and
+		 * kicked off reconstruct_snapshots, so if we end up here we have no
+		 * idea what happened.
+		 *
+		 * Do not delete unless we know that subvolumes and snapshots
+		 * are consistent:
+		 *
+		 * XXX:
+		 *
+		 * We could be smarter here, and instead of using the generic
+		 * recovery pass ratelimiting, track if there have been any
+		 * changes to the snapshots or inodes btrees since those passes
+		 * last ran.
+		 */
+		ret = bch2_require_recovery_pass(c, &buf, BCH_RECOVERY_PASS_check_snapshots) ?: ret;
+		ret = bch2_require_recovery_pass(c, &buf, BCH_RECOVERY_PASS_check_subvols) ?: ret;
+
+		if (c->sb.btrees_lost_data & BIT_ULL(BTREE_ID_snapshots))
+			ret = bch2_require_recovery_pass(c, &buf, BCH_RECOVERY_PASS_reconstruct_snapshots) ?: ret;
+
+		unsigned repair_flags = FSCK_CAN_IGNORE | (!ret ? FSCK_CAN_FIX : 0);
+
+		if (__fsck_err(trans, repair_flags, bkey_in_missing_snapshot,
+			     "key in missing snapshot %s, delete?",
+			     (bch2_btree_id_to_text(&buf, iter->btree_id),
+			      prt_char(&buf, ' '),
+			      bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+			ret = bch2_btree_delete_at(trans, iter,
+						   BTREE_UPDATE_internal_snapshot_node) ?: 1;
+		}
+	}
 fsck_err:
 	printbuf_exit(&buf);
 	return ret;
@@ -1276,7 +1296,7 @@ static int create_snapids(struct btree_trans *trans, u32 parent, u32 tree,
 			goto err;
 
 		if (!k.k || !k.k->p.offset) {
-			ret = -BCH_ERR_ENOSPC_snapshot_create;
+			ret = bch_err_throw(c, ENOSPC_snapshot_create);
 			goto err;
 		}
 
@@ -1878,6 +1898,8 @@ err:
 	d->running = false;
 	mutex_unlock(&d->progress_lock);
 	bch2_trans_put(trans);
+
+	bch2_recovery_pass_set_no_ratelimit(c, BCH_RECOVERY_PASS_check_snapshots);
 out_unlock:
 	mutex_unlock(&d->lock);
 	if (!bch2_err_matches(ret, EROFS))
@@ -1913,7 +1935,7 @@ void bch2_delete_dead_snapshots_async(struct bch_fs *c)
 
 	BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags));
 
-	if (!queue_work(c->write_ref_wq, &c->snapshot_delete.work))
+	if (!queue_work(system_long_wq, &c->snapshot_delete.work))
 		enumerated_ref_put(&c->writes, BCH_WRITE_REF_delete_dead_snapshots);
 }
 
diff --git a/libbcachefs/str_hash.c b/libbcachefs/str_hash.c
index 7b97852d..71b735a8 100644
--- a/libbcachefs/str_hash.c
+++ b/libbcachefs/str_hash.c
@@ -31,14 +31,15 @@ static int bch2_dirent_has_target(struct btree_trans *trans, struct bkey_s_c_dir
 	}
 }
 
-static noinline int fsck_rename_dirent(struct btree_trans *trans,
-				       struct snapshots_seen *s,
-				       const struct bch_hash_desc desc,
-				       struct bch_hash_info *hash_info,
-				       struct bkey_s_c_dirent old)
+static int bch2_fsck_rename_dirent(struct btree_trans *trans,
+				   struct snapshots_seen *s,
+				   const struct bch_hash_desc desc,
+				   struct bch_hash_info *hash_info,
+				   struct bkey_s_c_dirent old,
+				   bool *updated_before_k_pos)
 {
 	struct qstr old_name = bch2_dirent_get_name(old);
-	struct bkey_i_dirent *new = bch2_trans_kmalloc(trans, bkey_bytes(old.k) + 32);
+	struct bkey_i_dirent *new = bch2_trans_kmalloc(trans, BKEY_U64s_MAX * sizeof(u64));
 	int ret = PTR_ERR_OR_ZERO(new);
 	if (ret)
 		return ret;
@@ -47,31 +48,37 @@ static noinline int fsck_rename_dirent(struct btree_trans *trans,
 	dirent_copy_target(new, old);
 	new->k.p = old.k->p;
 
+	char *renamed_buf = bch2_trans_kmalloc(trans, old_name.len + 20);
+	ret = PTR_ERR_OR_ZERO(renamed_buf);
+	if (ret)
+		return ret;
+
 	for (unsigned i = 0; i < 1000; i++) {
-		unsigned len = sprintf(new->v.d_name, "%.*s.fsck_renamed-%u",
-				       old_name.len, old_name.name, i);
-		unsigned u64s = BKEY_U64s + dirent_val_u64s(len, 0);
+		new->k.u64s = BKEY_U64s_MAX;
 
-		if (u64s > U8_MAX)
-			return -EINVAL;
+		struct qstr renamed_name = (struct qstr) QSTR_INIT(renamed_buf,
+					sprintf(renamed_buf, "%.*s.fsck_renamed-%u",
+						old_name.len, old_name.name, i));
 
-		new->k.u64s = u64s;
+		ret = bch2_dirent_init_name(new, hash_info, &renamed_name, NULL);
+		if (ret)
+			return ret;
 
 		ret = bch2_hash_set_in_snapshot(trans, bch2_dirent_hash_desc, hash_info,
 						(subvol_inum) { 0, old.k->p.inode },
 						old.k->p.snapshot, &new->k_i,
-						BTREE_UPDATE_internal_snapshot_node);
+						BTREE_UPDATE_internal_snapshot_node|
+						STR_HASH_must_create);
 		if (ret && !bch2_err_matches(ret, EEXIST))
-			goto err;
-		if (!ret)
 			break;
+		if (!ret) {
+			if (bpos_lt(new->k.p, old.k->p))
+				*updated_before_k_pos = true;
+			break;
+		}
 	}
 
-	if (ret)
-		goto err;
-
-	ret = bch2_fsck_update_backpointers(trans, s, desc, hash_info, &new->k_i);
-err:
+	ret = ret ?: bch2_fsck_update_backpointers(trans, s, desc, hash_info, &new->k_i);
 	bch_err_fn(trans->c, ret);
 	return ret;
 }
@@ -191,7 +198,7 @@ int bch2_repair_inode_hash_info(struct btree_trans *trans,
 #endif
 		bch2_print_str(c, KERN_ERR, buf.buf);
 		printbuf_exit(&buf);
-		ret = -BCH_ERR_fsck_repair_unimplemented;
+		ret = bch_err_throw(c, fsck_repair_unimplemented);
 		goto err;
 	}
 
@@ -226,53 +233,20 @@ static noinline int check_inode_hash_info_matches_root(struct btree_trans *trans
 	return ret;
 }
 
-int __bch2_str_hash_check_key(struct btree_trans *trans,
-			      struct snapshots_seen *s,
-			      const struct bch_hash_desc *desc,
-			      struct bch_hash_info *hash_info,
-			      struct btree_iter *k_iter, struct bkey_s_c hash_k)
+/* Put a str_hash key in its proper location, checking for duplicates */
+int bch2_str_hash_repair_key(struct btree_trans *trans,
+			     struct snapshots_seen *s,
+			     const struct bch_hash_desc *desc,
+			     struct bch_hash_info *hash_info,
+			     struct btree_iter *k_iter, struct bkey_s_c k,
+			     struct btree_iter *dup_iter, struct bkey_s_c dup_k,
+			     bool *updated_before_k_pos)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter iter = {};
 	struct printbuf buf = PRINTBUF;
-	struct bkey_s_c k;
 	bool free_snapshots_seen = false;
 	int ret = 0;
 
-	u64 hash = desc->hash_bkey(hash_info, hash_k);
-	if (hash_k.k->p.offset < hash)
-		goto bad_hash;
-
-	for_each_btree_key_norestart(trans, iter, desc->btree_id,
-				     SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
-				     BTREE_ITER_slots|
-				     BTREE_ITER_with_updates, k, ret) {
-		if (bkey_eq(k.k->p, hash_k.k->p))
-			break;
-
-		if (k.k->type == desc->key_type &&
-		    !desc->cmp_bkey(k, hash_k))
-			goto duplicate_entries;
-
-		if (bkey_deleted(k.k)) {
-			bch2_trans_iter_exit(trans, &iter);
-			goto bad_hash;
-		}
-	}
-out:
-	bch2_trans_iter_exit(trans, &iter);
-	printbuf_exit(&buf);
-	if (free_snapshots_seen)
-		darray_exit(&s->ids);
-	return ret;
-bad_hash:
-	/*
-	 * Before doing any repair, check hash_info itself:
-	 */
-	ret = check_inode_hash_info_matches_root(trans, hash_k.k->p.inode, hash_info);
-	if (ret)
-		goto out;
-
 	if (!s) {
 		s = bch2_trans_kmalloc(trans, sizeof(*s));
 		ret = PTR_ERR_OR_ZERO(s);
@@ -289,27 +263,27 @@ bad_hash:
 		free_snapshots_seen = true;
 	}
 
-	if (fsck_err(trans, hash_table_key_wrong_offset,
-		     "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
-		     bch2_btree_id_str(desc->btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
-		     (printbuf_reset(&buf),
-		      bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
-		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, hash_k);
-		if (IS_ERR(new))
-			return PTR_ERR(new);
-
-		k = bch2_hash_set_or_get_in_snapshot(trans, &iter, *desc, hash_info,
-				       (subvol_inum) { 0, hash_k.k->p.inode },
-				       hash_k.k->p.snapshot, new,
+	if (!dup_k.k) {
+		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
+		ret = PTR_ERR_OR_ZERO(new);
+		if (ret)
+			goto out;
+
+		dup_k = bch2_hash_set_or_get_in_snapshot(trans, dup_iter, *desc, hash_info,
+				       (subvol_inum) { 0, new->k.p.inode },
+				       new->k.p.snapshot, new,
 				       STR_HASH_must_create|
 				       BTREE_ITER_with_updates|
 				       BTREE_UPDATE_internal_snapshot_node);
-		ret = bkey_err(k);
+		ret = bkey_err(dup_k);
 		if (ret)
 			goto out;
-		if (k.k)
+		if (dup_k.k)
 			goto duplicate_entries;
 
+		if (bpos_lt(new->k.p, k.k->p))
+			*updated_before_k_pos = true;
+
 		ret =   bch2_insert_snapshot_whiteouts(trans, desc->btree_id,
 						       k_iter->pos, new->k.p) ?:
 			bch2_hash_delete_at(trans, *desc, hash_info, k_iter,
@@ -318,39 +292,108 @@ bad_hash:
 			bch2_fsck_update_backpointers(trans, s, *desc, hash_info, new) ?:
 			bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
 			-BCH_ERR_transaction_restart_commit;
-		goto out;
+	} else {
+duplicate_entries:
+		ret = hash_pick_winner(trans, *desc, hash_info, k, dup_k);
+		if (ret < 0)
+			goto out;
+
+		if (!fsck_err(trans, hash_table_key_duplicate,
+			      "duplicate hash table keys%s:\n%s",
+			      ret != 2 ? "" : ", both point to valid inodes",
+			      (printbuf_reset(&buf),
+			       bch2_bkey_val_to_text(&buf, c, k),
+			       prt_newline(&buf),
+			       bch2_bkey_val_to_text(&buf, c, dup_k),
+			       buf.buf)))
+			goto out;
+
+		switch (ret) {
+		case 0:
+			ret = bch2_hash_delete_at(trans, *desc, hash_info, k_iter, 0);
+			break;
+		case 1:
+			ret = bch2_hash_delete_at(trans, *desc, hash_info, dup_iter, 0);
+			break;
+		case 2:
+			ret = bch2_fsck_rename_dirent(trans, s, *desc, hash_info,
+						      bkey_s_c_to_dirent(k),
+						      updated_before_k_pos) ?:
+				bch2_hash_delete_at(trans, *desc, hash_info, k_iter,
+						    BTREE_ITER_with_updates);
+			goto out;
+		}
+
+		ret = bch2_trans_commit(trans, NULL, NULL, 0) ?:
+			-BCH_ERR_transaction_restart_commit;
 	}
+out:
 fsck_err:
-	goto out;
-duplicate_entries:
-	ret = hash_pick_winner(trans, *desc, hash_info, hash_k, k);
-	if (ret < 0)
-		goto out;
+	bch2_trans_iter_exit(trans, dup_iter);
+	printbuf_exit(&buf);
+	if (free_snapshots_seen)
+		darray_exit(&s->ids);
+	return ret;
+}
 
-	if (!fsck_err(trans, hash_table_key_duplicate,
-		      "duplicate hash table keys%s:\n%s",
-		      ret != 2 ? "" : ", both point to valid inodes",
-		      (printbuf_reset(&buf),
-		       bch2_bkey_val_to_text(&buf, c, hash_k),
-		       prt_newline(&buf),
-		       bch2_bkey_val_to_text(&buf, c, k),
-		       buf.buf)))
-		goto out;
+int __bch2_str_hash_check_key(struct btree_trans *trans,
+			      struct snapshots_seen *s,
+			      const struct bch_hash_desc *desc,
+			      struct bch_hash_info *hash_info,
+			      struct btree_iter *k_iter, struct bkey_s_c hash_k,
+			      bool *updated_before_k_pos)
+{
+	struct bch_fs *c = trans->c;
+	struct btree_iter iter = {};
+	struct printbuf buf = PRINTBUF;
+	struct bkey_s_c k;
+	int ret = 0;
 
-	switch (ret) {
-	case 0:
-		ret = bch2_hash_delete_at(trans, *desc, hash_info, k_iter, 0);
-		break;
-	case 1:
-		ret = bch2_hash_delete_at(trans, *desc, hash_info, &iter, 0);
-		break;
-	case 2:
-		ret = fsck_rename_dirent(trans, s, *desc, hash_info, bkey_s_c_to_dirent(hash_k)) ?:
-			bch2_hash_delete_at(trans, *desc, hash_info, k_iter, 0);
-		goto out;
+	u64 hash = desc->hash_bkey(hash_info, hash_k);
+	if (hash_k.k->p.offset < hash)
+		goto bad_hash;
+
+	for_each_btree_key_norestart(trans, iter, desc->btree_id,
+				     SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
+				     BTREE_ITER_slots|
+				     BTREE_ITER_with_updates, k, ret) {
+		if (bkey_eq(k.k->p, hash_k.k->p))
+			break;
+
+		if (k.k->type == desc->key_type &&
+		    !desc->cmp_bkey(k, hash_k)) {
+			ret =	check_inode_hash_info_matches_root(trans, hash_k.k->p.inode,
+								   hash_info) ?:
+				bch2_str_hash_repair_key(trans, s, desc, hash_info,
+							 k_iter, hash_k,
+							 &iter, k, updated_before_k_pos);
+			break;
+		}
+
+		if (bkey_deleted(k.k))
+			goto bad_hash;
 	}
+	bch2_trans_iter_exit(trans, &iter);
+out:
+fsck_err:
+	printbuf_exit(&buf);
+	return ret;
+bad_hash:
+	bch2_trans_iter_exit(trans, &iter);
+	/*
+	 * Before doing any repair, check hash_info itself:
+	 */
+	ret = check_inode_hash_info_matches_root(trans, hash_k.k->p.inode, hash_info);
+	if (ret)
+		goto out;
 
-	ret = bch2_trans_commit(trans, NULL, NULL, 0) ?:
-		-BCH_ERR_transaction_restart_commit;
+	if (fsck_err(trans, hash_table_key_wrong_offset,
+		     "hash table key at wrong offset: should be at %llu\n%s",
+		     hash,
+		     (bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf)))
+		ret = bch2_str_hash_repair_key(trans, s, desc, hash_info,
+					       k_iter, hash_k,
+					       &iter, bkey_s_c_null,
+					       updated_before_k_pos);
 	goto out;
 }
diff --git a/libbcachefs/str_hash.h b/libbcachefs/str_hash.h
index 6762b362..79d51aef 100644
--- a/libbcachefs/str_hash.h
+++ b/libbcachefs/str_hash.h
@@ -261,6 +261,7 @@ struct bkey_s_c bch2_hash_set_or_get_in_snapshot(struct btree_trans *trans,
 			   struct bkey_i *insert,
 			   enum btree_iter_update_trigger_flags flags)
 {
+	struct bch_fs *c = trans->c;
 	struct btree_iter slot = {};
 	struct bkey_s_c k;
 	bool found = false;
@@ -288,7 +289,7 @@ struct bkey_s_c bch2_hash_set_or_get_in_snapshot(struct btree_trans *trans,
 	}
 
 	if (!ret)
-		ret = -BCH_ERR_ENOSPC_str_hash_create;
+		ret = bch_err_throw(c, ENOSPC_str_hash_create);
 out:
 	bch2_trans_iter_exit(trans, &slot);
 	bch2_trans_iter_exit(trans, iter);
@@ -300,7 +301,7 @@ not_found:
 		bch2_trans_iter_exit(trans, &slot);
 		return k;
 	} else if (!found && (flags & STR_HASH_must_replace)) {
-		ret = -BCH_ERR_ENOENT_str_hash_set_must_replace;
+		ret = bch_err_throw(c, ENOENT_str_hash_set_must_replace);
 	} else {
 		if (!found && slot.path)
 			swap(*iter, slot);
@@ -328,7 +329,7 @@ int bch2_hash_set_in_snapshot(struct btree_trans *trans,
 		return ret;
 	if (k.k) {
 		bch2_trans_iter_exit(trans, &iter);
-		return -BCH_ERR_EEXIST_str_hash_set;
+		return bch_err_throw(trans->c, EEXIST_str_hash_set);
 	}
 
 	return 0;
@@ -397,17 +398,27 @@ int bch2_hash_delete(struct btree_trans *trans,
 int bch2_repair_inode_hash_info(struct btree_trans *, struct bch_inode_unpacked *);
 
 struct snapshots_seen;
+int bch2_str_hash_repair_key(struct btree_trans *,
+			     struct snapshots_seen *,
+			     const struct bch_hash_desc *,
+			     struct bch_hash_info *,
+			     struct btree_iter *, struct bkey_s_c,
+			     struct btree_iter *, struct bkey_s_c,
+			     bool *);
+
 int __bch2_str_hash_check_key(struct btree_trans *,
 			      struct snapshots_seen *,
 			      const struct bch_hash_desc *,
 			      struct bch_hash_info *,
-			      struct btree_iter *, struct bkey_s_c);
+			      struct btree_iter *, struct bkey_s_c,
+			      bool *);
 
 static inline int bch2_str_hash_check_key(struct btree_trans *trans,
 			    struct snapshots_seen *s,
 			    const struct bch_hash_desc *desc,
 			    struct bch_hash_info *hash_info,
-			    struct btree_iter *k_iter, struct bkey_s_c hash_k)
+			    struct btree_iter *k_iter, struct bkey_s_c hash_k,
+			    bool *updated_before_k_pos)
 {
 	if (hash_k.k->type != desc->key_type)
 		return 0;
@@ -415,7 +426,8 @@ static inline int bch2_str_hash_check_key(struct btree_trans *trans,
 	if (likely(desc->hash_bkey(hash_info, hash_k) == hash_k.k->p.offset))
 		return 0;
 
-	return __bch2_str_hash_check_key(trans, s, desc, hash_info, k_iter, hash_k);
+	return __bch2_str_hash_check_key(trans, s, desc, hash_info, k_iter, hash_k,
+					 updated_before_k_pos);
 }
 
 #endif /* _BCACHEFS_STR_HASH_H */
diff --git a/libbcachefs/subvolume.c b/libbcachefs/subvolume.c
index 170c5f6b..60c10365 100644
--- a/libbcachefs/subvolume.c
+++ b/libbcachefs/subvolume.c
@@ -255,6 +255,13 @@ void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
 		prt_printf(out, " creation_parent %u", le32_to_cpu(s.v->creation_parent));
 		prt_printf(out, " fs_parent %u", le32_to_cpu(s.v->fs_path_parent));
 	}
+
+	if (BCH_SUBVOLUME_RO(s.v))
+		prt_printf(out, " ro");
+	if (BCH_SUBVOLUME_SNAP(s.v))
+		prt_printf(out, " snapshot");
+	if (BCH_SUBVOLUME_UNLINKED(s.v))
+		prt_printf(out, " unlinked");
 }
 
 static int subvolume_children_mod(struct btree_trans *trans, struct bpos pos, bool set)
@@ -482,9 +489,12 @@ err:
 
 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
 {
-	return bch2_subvolumes_reparent(trans, subvolid) ?:
+	int ret = bch2_subvolumes_reparent(trans, subvolid) ?:
 		commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 			  __bch2_subvolume_delete(trans, subvolid));
+
+	bch2_recovery_pass_set_no_ratelimit(trans->c, BCH_RECOVERY_PASS_check_subvols);
+	return ret;
 }
 
 static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
@@ -593,7 +603,7 @@ int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
 	ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
 				BTREE_ID_subvolumes, POS(0, U32_MAX));
 	if (ret == -BCH_ERR_ENOSPC_btree_slot)
-		ret = -BCH_ERR_ENOSPC_subvolume_create;
+		ret = bch_err_throw(c, ENOSPC_subvolume_create);
 	if (ret)
 		return ret;
 
@@ -699,8 +709,9 @@ static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 		return ret;
 
 	if (!bkey_is_inode(k.k)) {
-		bch_err(trans->c, "root inode not found");
-		ret = -BCH_ERR_ENOENT_inode;
+		struct bch_fs *c = trans->c;
+		bch_err(c, "root inode not found");
+		ret = bch_err_throw(c, ENOENT_inode);
 		goto err;
 	}
 
diff --git a/libbcachefs/super-io.c b/libbcachefs/super-io.c
index 6687b923..6c2e1d64 100644
--- a/libbcachefs/super-io.c
+++ b/libbcachefs/super-io.c
@@ -1112,7 +1112,7 @@ int bch2_write_super(struct bch_fs *c)
 		prt_str(&buf, ")");
 		bch2_fs_fatal_error(c, ": %s", buf.buf);
 		printbuf_exit(&buf);
-		ret = -BCH_ERR_sb_not_downgraded;
+		ret = bch_err_throw(c, sb_not_downgraded);
 		goto out;
 	}
 
@@ -1142,7 +1142,7 @@ int bch2_write_super(struct bch_fs *c)
 
 			if (c->opts.errors != BCH_ON_ERROR_continue &&
 			    c->opts.errors != BCH_ON_ERROR_fix_safe) {
-				ret = -BCH_ERR_erofs_sb_err;
+				ret = bch_err_throw(c, erofs_sb_err);
 				bch2_fs_fatal_error(c, "%s", buf.buf);
 			} else {
 				bch_err(c, "%s", buf.buf);
@@ -1161,7 +1161,7 @@ int bch2_write_super(struct bch_fs *c)
 				ca->disk_sb.seq);
 			bch2_fs_fatal_error(c, "%s", buf.buf);
 			printbuf_exit(&buf);
-			ret = -BCH_ERR_erofs_sb_err;
+			ret = bch_err_throw(c, erofs_sb_err);
 		}
 	}
 
@@ -1215,7 +1215,7 @@ int bch2_write_super(struct bch_fs *c)
 				  !can_mount_with_written), c,
 		": Unable to write superblock to sufficient devices (from %ps)",
 		(void *) _RET_IP_))
-		ret = -BCH_ERR_erofs_sb_err;
+		ret = bch_err_throw(c, erofs_sb_err);
 out:
 	/* Make new options visible after they're persistent: */
 	bch2_sb_update(c);
diff --git a/libbcachefs/super.c b/libbcachefs/super.c
index 56ca2e4a..21b1b227 100644
--- a/libbcachefs/super.c
+++ b/libbcachefs/super.c
@@ -474,16 +474,16 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 	BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags));
 
 	if (WARN_ON(c->sb.features & BIT_ULL(BCH_FEATURE_no_alloc_info)))
-		return -BCH_ERR_erofs_no_alloc_info;
+		return bch_err_throw(c, erofs_no_alloc_info);
 
 	if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
 		bch_err(c, "cannot go rw, unfixed btree errors");
-		return -BCH_ERR_erofs_unfixed_errors;
+		return bch_err_throw(c, erofs_unfixed_errors);
 	}
 
 	if (c->sb.features & BIT_ULL(BCH_FEATURE_small_image)) {
 		bch_err(c, "cannot go rw, filesystem is an unresized image file");
-		return -BCH_ERR_erofs_filesystem_full;
+		return bch_err_throw(c, erofs_filesystem_full);
 	}
 
 	if (test_bit(BCH_FS_rw, &c->flags))
@@ -564,13 +564,13 @@ int bch2_fs_read_write(struct bch_fs *c)
 {
 	if (c->opts.recovery_pass_last &&
 	    c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay)
-		return -BCH_ERR_erofs_norecovery;
+		return bch_err_throw(c, erofs_norecovery);
 
 	if (c->opts.nochanges)
-		return -BCH_ERR_erofs_nochanges;
+		return bch_err_throw(c, erofs_nochanges);
 
 	if (c->sb.features & BIT_ULL(BCH_FEATURE_no_alloc_info))
-		return -BCH_ERR_erofs_no_alloc_info;
+		return bch_err_throw(c, erofs_no_alloc_info);
 
 	return __bch2_fs_read_write(c, false);
 }
@@ -755,7 +755,7 @@ static int bch2_fs_online(struct bch_fs *c)
 	if (c->sb.multi_device &&
 	    __bch2_uuid_to_fs(c->sb.uuid)) {
 		bch_err(c, "filesystem UUID already open");
-		return -BCH_ERR_filesystem_uuid_already_open;
+		return bch_err_throw(c, filesystem_uuid_already_open);
 	}
 
 	ret = bch2_fs_chardev_init(c);
@@ -814,7 +814,7 @@ static int bch2_fs_init_rw(struct bch_fs *c)
 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
 	    !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
 				WQ_FREEZABLE, 0)))
-		return -BCH_ERR_ENOMEM_fs_other_alloc;
+		return bch_err_throw(c, ENOMEM_fs_other_alloc);
 
 	int ret = bch2_fs_btree_interior_update_init(c) ?:
 		bch2_fs_btree_write_buffer_init(c) ?:
@@ -995,7 +995,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts *opts,
 	    mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
 				       c->opts.btree_node_size) ||
 	    mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048)) {
-		ret = -BCH_ERR_ENOMEM_fs_other_alloc;
+		ret = bch_err_throw(c, ENOMEM_fs_other_alloc);
 		goto err;
 	}
 
@@ -1155,7 +1155,7 @@ int bch2_fs_start(struct bch_fs *c)
 			 unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING));
 
 	if (!bch2_fs_may_start(c))
-		return -BCH_ERR_insufficient_devices_to_start;
+		return bch_err_throw(c, insufficient_devices_to_start);
 
 	down_write(&c->state_lock);
 	mutex_lock(&c->sb_lock);
@@ -1166,7 +1166,7 @@ int bch2_fs_start(struct bch_fs *c)
 			sizeof(struct bch_sb_field_ext) / sizeof(u64))) {
 		mutex_unlock(&c->sb_lock);
 		up_write(&c->state_lock);
-		ret = -BCH_ERR_ENOSPC_sb;
+		ret = bch_err_throw(c, ENOSPC_sb);
 		goto err;
 	}
 
@@ -1208,7 +1208,7 @@ int bch2_fs_start(struct bch_fs *c)
 		goto err;
 
 	if (bch2_fs_init_fault("fs_start")) {
-		ret = -BCH_ERR_injected_fs_start;
+		ret = bch_err_throw(c, injected_fs_start);
 		goto err;
 	}
 
@@ -1235,11 +1235,11 @@ static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
 
 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
-		return -BCH_ERR_mismatched_block_size;
+		return bch_err_throw(c, mismatched_block_size);
 
 	if (le16_to_cpu(m.bucket_size) <
 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
-		return -BCH_ERR_bucket_size_too_small;
+		return bch_err_throw(c, bucket_size_too_small);
 
 	return 0;
 }
@@ -1550,7 +1550,7 @@ static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
 	bch2_dev_attach(c, ca, dev_idx);
 	return 0;
 err:
-	return -BCH_ERR_ENOMEM_dev_alloc;
+	return bch_err_throw(c, ENOMEM_dev_alloc);
 }
 
 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
@@ -1560,13 +1560,13 @@ static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
 	if (bch2_dev_is_online(ca)) {
 		bch_err(ca, "already have device online in slot %u",
 			sb->sb->dev_idx);
-		return -BCH_ERR_device_already_online;
+		return bch_err_throw(ca->fs, device_already_online);
 	}
 
 	if (get_capacity(sb->bdev->bd_disk) <
 	    ca->mi.bucket_size * ca->mi.nbuckets) {
 		bch_err(ca, "cannot online: device too small");
-		return -BCH_ERR_device_size_too_small;
+		return bch_err_throw(ca->fs, device_size_too_small);
 	}
 
 	BUG_ON(!enumerated_ref_is_zero(&ca->io_ref[READ]));
@@ -1718,7 +1718,7 @@ int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
 		return 0;
 
 	if (!bch2_dev_state_allowed(c, ca, new_state, flags))
-		return -BCH_ERR_device_state_not_allowed;
+		return bch_err_throw(c, device_state_not_allowed);
 
 	if (new_state != BCH_MEMBER_STATE_rw)
 		__bch2_dev_read_only(c, ca);
@@ -1771,7 +1771,7 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 
 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
 		bch_err(ca, "Cannot remove without losing data");
-		ret = -BCH_ERR_device_state_not_allowed;
+		ret = bch_err_throw(c, device_state_not_allowed);
 		goto err;
 	}
 
@@ -1907,7 +1907,7 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	if (list_empty(&c->list)) {
 		mutex_lock(&bch_fs_list_lock);
 		if (__bch2_uuid_to_fs(c->sb.uuid))
-			ret = -BCH_ERR_filesystem_uuid_already_open;
+			ret = bch_err_throw(c, filesystem_uuid_already_open);
 		else
 			list_add(&c->list, &bch_fs_list);
 		mutex_unlock(&bch_fs_list_lock);
@@ -2094,7 +2094,7 @@ int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
 		bch_err(ca, "Cannot offline required disk");
 		up_write(&c->state_lock);
-		return -BCH_ERR_device_state_not_allowed;
+		return bch_err_throw(c, device_state_not_allowed);
 	}
 
 	__bch2_dev_offline(c, ca);
@@ -2133,7 +2133,7 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 	if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) {
 		bch_err(ca, "New device size too big (%llu greater than max %u)",
 			nbuckets, BCH_MEMBER_NBUCKETS_MAX);
-		ret = -BCH_ERR_device_size_too_big;
+		ret = bch_err_throw(c, device_size_too_big);
 		goto err;
 	}
 
@@ -2141,7 +2141,7 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 	    get_capacity(ca->disk_sb.bdev->bd_disk) <
 	    ca->mi.bucket_size * nbuckets) {
 		bch_err(ca, "New size larger than device");
-		ret = -BCH_ERR_device_size_too_small;
+		ret = bch_err_throw(c, device_size_too_small);
 		goto err;
 	}
 
@@ -2376,7 +2376,7 @@ struct bch_fs *bch2_fs_open(darray_const_str *devices,
 	}
 
 	if (opts->nochanges && !opts->read_only) {
-		ret = -BCH_ERR_erofs_nochanges;
+		ret = bch_err_throw(c, erofs_nochanges);
 		goto err_print;
 	}
 
diff --git a/libbcachefs/trace.h b/libbcachefs/trace.h
index eb3ca963..dc095327 100644
--- a/libbcachefs/trace.h
+++ b/libbcachefs/trace.h
@@ -199,6 +199,50 @@ DECLARE_EVENT_CLASS(bio,
 		  (unsigned long long)__entry->sector, __entry->nr_sector)
 );
 
+/* errors */
+
+TRACE_EVENT(error_throw,
+	TP_PROTO(struct bch_fs *c, int bch_err, unsigned long ip),
+	TP_ARGS(c, bch_err, ip),
+
+	TP_STRUCT__entry(
+		__field(dev_t,		dev			)
+		__field(int,		err			)
+		__array(char,		err_str, 32		)
+		__array(char,		ip, 32			)
+	),
+
+	TP_fast_assign(
+		__entry->dev		= c->dev;
+		__entry->err		= bch_err;
+		strscpy(__entry->err_str, bch2_err_str(bch_err), sizeof(__entry->err_str));
+		snprintf(__entry->ip, sizeof(__entry->ip), "%ps", (void *) ip);
+	),
+
+	TP_printk("%d,%d %s ret %s", MAJOR(__entry->dev), MINOR(__entry->dev),
+		  __entry->ip, __entry->err_str)
+);
+
+TRACE_EVENT(error_downcast,
+	TP_PROTO(int bch_err, int std_err, unsigned long ip),
+	TP_ARGS(bch_err, std_err, ip),
+
+	TP_STRUCT__entry(
+		__array(char,		bch_err, 32		)
+		__array(char,		std_err, 32		)
+		__array(char,		ip, 32			)
+	),
+
+	TP_fast_assign(
+		strscpy(__entry->bch_err, bch2_err_str(bch_err), sizeof(__entry->bch_err));
+		strscpy(__entry->std_err, bch2_err_str(std_err), sizeof(__entry->std_err));
+		snprintf(__entry->ip, sizeof(__entry->ip), "%ps", (void *) ip);
+	),
+
+	TP_printk("%s ret %s -> %s %s", __entry->ip,
+		  __entry->bch_err, __entry->std_err, __entry->ip)
+);
+
 /* disk_accounting.c */
 
 TRACE_EVENT(accounting_mem_insert,
@@ -1446,25 +1490,6 @@ DEFINE_EVENT(fs_str, io_move_evacuate_bucket,
 	TP_ARGS(c, str)
 );
 
-TRACE_EVENT(error_downcast,
-	TP_PROTO(int bch_err, int std_err, unsigned long ip),
-	TP_ARGS(bch_err, std_err, ip),
-
-	TP_STRUCT__entry(
-		__array(char,		bch_err, 32		)
-		__array(char,		std_err, 32		)
-		__array(char,		ip, 32			)
-	),
-
-	TP_fast_assign(
-		strscpy(__entry->bch_err, bch2_err_str(bch_err), sizeof(__entry->bch_err));
-		strscpy(__entry->std_err, bch2_err_str(std_err), sizeof(__entry->std_err));
-		snprintf(__entry->ip, sizeof(__entry->ip), "%ps", (void *) ip);
-	),
-
-	TP_printk("%s -> %s %s", __entry->bch_err, __entry->std_err, __entry->ip)
-);
-
 #ifdef CONFIG_BCACHEFS_PATH_TRACEPOINTS
 
 TRACE_EVENT(update_by_path,