Update bcachefs sources to c1ec91699f60 workqueue: Basic memory allocation profiling supportv1.25.3

34 files changed, 305 insertions, 196 deletions

diff --git a/.bcachefs_revision b/.bcachefs_revision
index 7abb0304..fad9022e 100644
--- a/.bcachefs_revision
+++ b/.bcachefs_revision
@@ -1 +1 @@
-deeffbdc52f1092dadb3d523c4429e002c7fc485
+c1ec91699f6089754548cfc01391fff17a633516
diff --git a/libbcachefs/acl.c b/libbcachefs/acl.c
index 307824d6..8f970dc1 100644
--- a/libbcachefs/acl.c
+++ b/libbcachefs/acl.c
@@ -138,8 +138,8 @@ static struct posix_acl *bch2_acl_from_disk(struct btree_trans *trans,
 
 	acl = allocate_dropping_locks(trans, ret,
 			posix_acl_alloc(count, _gfp));
-	if (!acl)
-		return ERR_PTR(-ENOMEM);
+	if (!acl && !ret)
+		ret = bch_err_throw(trans->c, ENOMEM_acl);
 	if (ret) {
 		kfree(acl);
 		return ERR_PTR(ret);
diff --git a/libbcachefs/alloc_background.c b/libbcachefs/alloc_background.c
index 4c1604fd..f1d35b7f 100644
--- a/libbcachefs/alloc_background.c
+++ b/libbcachefs/alloc_background.c
@@ -337,9 +337,10 @@ void bch2_alloc_v4_swab(struct bkey_s k)
 }
 
 static inline void __bch2_alloc_v4_to_text(struct printbuf *out, struct bch_fs *c,
-					   unsigned dev, const struct bch_alloc_v4 *a)
+					   struct bkey_s_c k,
+					   const struct bch_alloc_v4 *a)
 {
-	struct bch_dev *ca = c ? bch2_dev_tryget_noerror(c, dev) : NULL;
+	struct bch_dev *ca = c ? bch2_dev_tryget_noerror(c, k.k->p.inode) : NULL;
 
 	prt_newline(out);
 	printbuf_indent_add(out, 2);
@@ -348,11 +349,14 @@ static inline void __bch2_alloc_v4_to_text(struct printbuf *out, struct bch_fs *
 	bch2_prt_data_type(out, a->data_type);
 	prt_newline(out);
 	prt_printf(out, "journal_seq_nonempty %llu\n",	a->journal_seq_nonempty);
-	prt_printf(out, "journal_seq_empty    %llu\n",	a->journal_seq_empty);
+	if (bkey_val_bytes(k.k) > offsetof(struct bch_alloc_v4, journal_seq_empty))
+		prt_printf(out, "journal_seq_empty    %llu\n",	a->journal_seq_empty);
+
 	prt_printf(out, "need_discard         %llu\n",	BCH_ALLOC_V4_NEED_DISCARD(a));
 	prt_printf(out, "need_inc_gen         %llu\n",	BCH_ALLOC_V4_NEED_INC_GEN(a));
 	prt_printf(out, "dirty_sectors        %u\n",	a->dirty_sectors);
-	prt_printf(out, "stripe_sectors       %u\n",	a->stripe_sectors);
+	if (bkey_val_bytes(k.k) > offsetof(struct bch_alloc_v4, stripe_sectors))
+		prt_printf(out, "stripe_sectors       %u\n",	a->stripe_sectors);
 	prt_printf(out, "cached_sectors       %u\n",	a->cached_sectors);
 	prt_printf(out, "stripe               %u\n",	a->stripe);
 	prt_printf(out, "stripe_redundancy    %u\n",	a->stripe_redundancy);
@@ -372,12 +376,12 @@ void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c
 	struct bch_alloc_v4 _a;
 	const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
 
-	__bch2_alloc_v4_to_text(out, c, k.k->p.inode, a);
+	__bch2_alloc_v4_to_text(out, c, k, a);
 }
 
 void bch2_alloc_v4_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 {
-	__bch2_alloc_v4_to_text(out, c, k.k->p.inode, bkey_s_c_to_alloc_v4(k).v);
+	__bch2_alloc_v4_to_text(out, c, k, bkey_s_c_to_alloc_v4(k).v);
 }
 
 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
diff --git a/libbcachefs/btree_gc.c b/libbcachefs/btree_gc.c
index 34cb8a43..e95bb684 100644
--- a/libbcachefs/btree_gc.c
+++ b/libbcachefs/btree_gc.c
@@ -44,10 +44,6 @@
 #include <linux/rcupdate.h>
 #include <linux/sched/task.h>
 
-#define DROP_THIS_NODE		10
-#define DROP_PREV_NODE		11
-#define DID_FILL_FROM_SCAN	12
-
 /*
  * Returns true if it's a btree we can easily reconstruct, or otherwise won't
  * cause data loss if it's missing:
@@ -252,7 +248,7 @@ static int btree_check_node_boundaries(struct btree_trans *trans, struct btree *
 				return ret;
 
 			*pulled_from_scan = cur->data->min_key;
-			ret = DID_FILL_FROM_SCAN;
+			ret = bch_err_throw(c, topology_repair_did_fill_from_scan);
 		} else {
 			if (mustfix_fsck_err(trans, btree_node_topology_bad_min_key,
 					     "btree node with incorrect min_key%s", buf.buf))
@@ -263,7 +259,7 @@ static int btree_check_node_boundaries(struct btree_trans *trans, struct btree *
 			if (bpos_ge(prev->data->min_key, cur->data->min_key)) {		/* fully? */
 				if (mustfix_fsck_err(trans, btree_node_topology_overwritten_by_next_node,
 						     "btree node overwritten by next node%s", buf.buf))
-					ret = DROP_PREV_NODE;
+					ret = bch_err_throw(c, topology_repair_drop_prev_node);
 			} else {
 				if (mustfix_fsck_err(trans, btree_node_topology_bad_max_key,
 						     "btree node with incorrect max_key%s", buf.buf))
@@ -274,7 +270,7 @@ static int btree_check_node_boundaries(struct btree_trans *trans, struct btree *
 			if (bpos_ge(expected_start, cur->data->max_key)) {		/* fully? */
 				if (mustfix_fsck_err(trans, btree_node_topology_overwritten_by_prev_node,
 						     "btree node overwritten by prev node%s", buf.buf))
-					ret = DROP_THIS_NODE;
+					ret = bch_err_throw(c, topology_repair_drop_this_node);
 			} else {
 				if (mustfix_fsck_err(trans, btree_node_topology_bad_min_key,
 						     "btree node with incorrect min_key%s", buf.buf))
@@ -314,7 +310,7 @@ static int btree_repair_node_end(struct btree_trans *trans, struct btree *b,
 				return ret;
 
 			*pulled_from_scan = b->key.k.p;
-			ret = DID_FILL_FROM_SCAN;
+			ret = bch_err_throw(c, topology_repair_did_fill_from_scan);
 		} else {
 			ret = set_node_max(c, child, b->key.k.p);
 		}
@@ -391,15 +387,15 @@ again:
 
 		ret = lockrestart_do(trans,
 			btree_check_node_boundaries(trans, b, prev, cur, pulled_from_scan));
-		if (ret < 0)
+		if (ret && !bch2_err_matches(ret, BCH_ERR_topology_repair))
 			goto err;
 
-		if (ret == DID_FILL_FROM_SCAN) {
+		if (bch2_err_matches(ret, BCH_ERR_topology_repair_did_fill_from_scan)) {
 			new_pass = true;
 			ret = 0;
 		}
 
-		if (ret == DROP_THIS_NODE) {
+		if (bch2_err_matches(ret, BCH_ERR_topology_repair_drop_this_node)) {
 			six_unlock_read(&cur->c.lock);
 			bch2_btree_node_evict(trans, cur_k.k);
 			ret = bch2_journal_key_delete(c, b->c.btree_id,
@@ -414,7 +410,7 @@ again:
 			six_unlock_read(&prev->c.lock);
 		prev = NULL;
 
-		if (ret == DROP_PREV_NODE) {
+		if (bch2_err_matches(ret, BCH_ERR_topology_repair_drop_prev_node)) {
 			bch_info(c, "dropped prev node");
 			bch2_btree_node_evict(trans, prev_k.k);
 			ret = bch2_journal_key_delete(c, b->c.btree_id,
@@ -436,7 +432,7 @@ again:
 		BUG_ON(cur);
 		ret = lockrestart_do(trans,
 			btree_repair_node_end(trans, b, prev, pulled_from_scan));
-		if (ret == DID_FILL_FROM_SCAN) {
+		if (bch2_err_matches(ret, BCH_ERR_topology_repair_did_fill_from_scan)) {
 			new_pass = true;
 			ret = 0;
 		}
@@ -477,7 +473,7 @@ again:
 		six_unlock_read(&cur->c.lock);
 		cur = NULL;
 
-		if (ret == DROP_THIS_NODE) {
+		if (bch2_err_matches(ret, BCH_ERR_topology_repair_drop_this_node)) {
 			bch2_btree_node_evict(trans, cur_k.k);
 			ret = bch2_journal_key_delete(c, b->c.btree_id,
 						      b->c.level, cur_k.k->k.p);
@@ -504,7 +500,7 @@ again:
 	if (mustfix_fsck_err_on(!have_child,
 			c, btree_node_topology_interior_node_empty,
 			"empty interior btree node at %s", buf.buf))
-		ret = DROP_THIS_NODE;
+		ret = bch_err_throw(c, topology_repair_drop_this_node);
 err:
 fsck_err:
 	if (!IS_ERR_OR_NULL(prev))
@@ -521,7 +517,8 @@ fsck_err:
 
 	bch2_bkey_buf_exit(&prev_k, c);
 	bch2_bkey_buf_exit(&cur_k, c);
-	bch_err_fn(c, ret);
+	if (!bch2_err_matches(ret, BCH_ERR_topology_repair))
+		bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -592,7 +589,7 @@ recover:
 		ret = bch2_btree_repair_topology_recurse(trans, b, &pulled_from_scan);
 		six_unlock_read(&b->c.lock);
 
-		if (ret == DROP_THIS_NODE) {
+		if (bch2_err_matches(ret, BCH_ERR_topology_repair_drop_this_node)) {
 			scoped_guard(mutex, &c->btree_cache.lock)
 				bch2_btree_node_hash_remove(&c->btree_cache, b);
 
diff --git a/libbcachefs/btree_iter.c b/libbcachefs/btree_iter.c
index cc771aff..a282c388 100644
--- a/libbcachefs/btree_iter.c
+++ b/libbcachefs/btree_iter.c
@@ -2860,8 +2860,9 @@ struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_trans *trans, struct btre
 		    !bkey_deleted(k.k) &&
 		    (k2 = btree_trans_peek_key_cache(trans, iter, iter->pos)).k) {
 			k = k2;
-			if (!bkey_err(k))
-				iter->k = *k.k;
+			if (bkey_err(k))
+				goto out;
+			iter->k = *k.k;
 		}
 
 		if (unlikely(k.k->type == KEY_TYPE_whiteout &&
diff --git a/libbcachefs/btree_key_cache.c b/libbcachefs/btree_key_cache.c
index ebba14da..d61b7820 100644
--- a/libbcachefs/btree_key_cache.c
+++ b/libbcachefs/btree_key_cache.c
@@ -254,11 +254,13 @@ static int btree_key_cache_create(struct btree_trans *trans,
 
 		struct bkey_i *new_k = allocate_dropping_locks(trans, ret,
 				kmalloc(key_u64s * sizeof(u64), _gfp));
-		if (unlikely(!new_k)) {
+		if (unlikely(!new_k && !ret)) {
 			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_throw(c, ENOMEM_btree_key_cache_fill);
-		} else if (ret) {
+		}
+
+		if (unlikely(ret)) {
 			kfree(new_k);
 			goto err;
 		}
@@ -407,7 +409,7 @@ int bch2_btree_path_traverse_cached(struct btree_trans *trans,
 			btree_node_unlock(trans, path, 0);
 			path->l[0].b = ERR_PTR(ret);
 		}
-	} else {
+	} else if (!(flags & BTREE_ITER_cached_nofill)) {
 		BUG_ON(path->uptodate);
 		BUG_ON(!path->nodes_locked);
 	}
diff --git a/libbcachefs/btree_node_scan.c b/libbcachefs/btree_node_scan.c
index d997e381..f4f958f4 100644
--- a/libbcachefs/btree_node_scan.c
+++ b/libbcachefs/btree_node_scan.c
@@ -270,6 +270,9 @@ static int read_btree_nodes(struct find_btree_nodes *f)
 	int ret = 0;
 
 	closure_init_stack(&cl);
+	CLASS(printbuf, buf)();
+
+	prt_printf(&buf, "scanning for btree nodes on");
 
 	for_each_online_member(c, ca, BCH_DEV_READ_REF_btree_node_scan) {
 		if (!(ca->mi.data_allowed & BIT(BCH_DATA_btree)))
@@ -295,10 +298,14 @@ static int read_btree_nodes(struct find_btree_nodes *f)
 			break;
 		}
 
+		prt_printf(&buf, " %s", ca->name);
+
 		closure_get(&cl);
 		enumerated_ref_get(&ca->io_ref[READ], BCH_DEV_READ_REF_btree_node_scan);
 		wake_up_process(t);
 	}
+
+	bch_notice(c, "%s", buf.buf);
 err:
 	while (closure_sync_timeout(&cl, sysctl_hung_task_timeout_secs * HZ / 2))
 		;
diff --git a/libbcachefs/btree_trans_commit.c b/libbcachefs/btree_trans_commit.c
index 1f9965ae..58590ccc 100644
--- a/libbcachefs/btree_trans_commit.c
+++ b/libbcachefs/btree_trans_commit.c
@@ -772,12 +772,13 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 		trans->journal_res.offset	+= trans->journal_entries.u64s;
 		trans->journal_res.u64s		-= trans->journal_entries.u64s;
 
-		memcpy_u64s_small(bch2_journal_add_entry(j, &trans->journal_res,
-						BCH_JSET_ENTRY_write_buffer_keys,
-						BTREE_ID_accounting, 0,
-						trans->accounting.u64s)->_data,
-				  btree_trans_subbuf_base(trans, &trans->accounting),
-				  trans->accounting.u64s);
+		if (trans->accounting.u64s)
+			memcpy_u64s_small(bch2_journal_add_entry(j, &trans->journal_res,
+							BCH_JSET_ENTRY_write_buffer_keys,
+							BTREE_ID_accounting, 0,
+							trans->accounting.u64s)->_data,
+					  btree_trans_subbuf_base(trans, &trans->accounting),
+					  trans->accounting.u64s);
 
 		if (trans->journal_seq)
 			*trans->journal_seq = trans->journal_res.seq;
@@ -1065,11 +1066,15 @@ int __bch2_trans_commit(struct btree_trans *trans, enum bch_trans_commit_flags f
 
 	EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags));
 
-	journal_u64s = jset_u64s(trans->accounting.u64s);
+	journal_u64s = 0;
+
 	trans->journal_transaction_names = READ_ONCE(c->opts.journal_transaction_names);
 	if (trans->journal_transaction_names)
 		journal_u64s += jset_u64s(JSET_ENTRY_LOG_U64s);
 
+	if (trans->accounting.u64s)
+		journal_u64s += jset_u64s(trans->accounting.u64s);
+
 	trans_for_each_update(trans, i) {
 		struct btree_path *path = trans->paths + i->path;
 
diff --git a/libbcachefs/btree_update_interior.c b/libbcachefs/btree_update_interior.c
index 312ef203..e4aa4fa7 100644
--- a/libbcachefs/btree_update_interior.c
+++ b/libbcachefs/btree_update_interior.c
@@ -14,6 +14,7 @@
 #include "btree_locking.h"
 #include "buckets.h"
 #include "clock.h"
+#include "disk_groups.h"
 #include "enumerated_ref.h"
 #include "error.h"
 #include "extents.h"
@@ -277,6 +278,36 @@ static void bch2_btree_node_free_never_used(struct btree_update *as,
 	bch2_trans_node_drop(trans, b);
 }
 
+static bool can_use_btree_node(struct bch_fs *c,
+			       struct disk_reservation *res,
+			       unsigned target,
+			       struct bkey_s_c k)
+{
+	if (!bch2_bkey_devs_rw(c, k))
+		return false;
+
+	if (target && !bch2_bkey_in_target(c, k, target))
+		return false;
+
+	unsigned durability = bch2_bkey_durability(c, k);
+
+	if (durability >= res->nr_replicas)
+		return true;
+
+	struct bch_devs_mask devs = target_rw_devs(c, BCH_DATA_btree, target);
+
+	guard(rcu)();
+
+	unsigned durability_available = 0, i;
+	for_each_set_bit(i, devs.d, BCH_SB_MEMBERS_MAX) {
+		struct bch_dev *ca = bch2_dev_rcu_noerror(c, i);
+		if (ca)
+			durability_available += ca->mi.durability;
+	}
+
+	return durability >= durability_available;
+}
+
 static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans,
 					     struct disk_reservation *res,
 					     struct closure *cl,
@@ -303,10 +334,14 @@ static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans,
 	mutex_lock(&c->btree_reserve_cache_lock);
 	if (c->btree_reserve_cache_nr > nr_reserve) {
 		for (struct btree_alloc *a = c->btree_reserve_cache;
-		     a < c->btree_reserve_cache + c->btree_reserve_cache_nr;
-		     a++) {
-			if (target && !bch2_bkey_in_target(c, bkey_i_to_s_c(&a->k), target))
+		     a < c->btree_reserve_cache + c->btree_reserve_cache_nr;) {
+			/* check if it has sufficient durability */
+
+			if (!can_use_btree_node(c, res, target, bkey_i_to_s_c(&a->k))) {
+				bch2_open_buckets_put(c, &a->ob);
+				*a = c->btree_reserve_cache[--c->btree_reserve_cache_nr];
 				continue;
+			}
 
 			bkey_copy(&b->key, &a->k);
 			b->ob = a->ob;
diff --git a/libbcachefs/clock.c b/libbcachefs/clock.c
index 5185794f..1c6d0cdc 100644
--- a/libbcachefs/clock.c
+++ b/libbcachefs/clock.c
@@ -21,7 +21,7 @@ static const struct min_heap_callbacks callbacks = {
 
 void bch2_io_timer_add(struct io_clock *clock, struct io_timer *timer)
 {
-	guard(spinlock)(&clock->timer_lock);
+	spin_lock(&clock->timer_lock);
 
 	if (time_after_eq64((u64) atomic64_read(&clock->now), timer->expire)) {
 		spin_unlock(&clock->timer_lock);
@@ -31,9 +31,11 @@ void bch2_io_timer_add(struct io_clock *clock, struct io_timer *timer)
 
 	for (size_t i = 0; i < clock->timers.nr; i++)
 		if (clock->timers.data[i] == timer)
-			return;
+			goto out;
 
 	BUG_ON(!min_heap_push(&clock->timers, &timer, &callbacks, NULL));
+out:
+	spin_unlock(&clock->timer_lock);
 }
 
 void bch2_io_timer_del(struct io_clock *clock, struct io_timer *timer)
diff --git a/libbcachefs/debug.c b/libbcachefs/debug.c
index 97d7655a..33cb94f7 100644
--- a/libbcachefs/debug.c
+++ b/libbcachefs/debug.c
@@ -462,7 +462,7 @@ static void bch2_cached_btree_node_to_text(struct printbuf *out, struct bch_fs *
 					   struct btree *b)
 {
 	if (!out->nr_tabstops)
-		printbuf_tabstop_push(out, 32);
+		printbuf_tabstop_push(out, 36);
 
 	prt_printf(out, "%px ", b);
 	bch2_btree_id_level_to_text(out, b->c.btree_id, b->c.level);
diff --git a/libbcachefs/dirent.h b/libbcachefs/dirent.h
index 373d382b..efb58d2d 100644
--- a/libbcachefs/dirent.h
+++ b/libbcachefs/dirent.h
@@ -26,6 +26,13 @@ struct bch_inode_info;
 #if IS_ENABLED(CONFIG_UNICODE)
 int bch2_casefold(struct btree_trans *, const struct bch_hash_info *,
 		  const struct qstr *, struct qstr *);
+#else
+static inline int bch2_casefold(struct btree_trans *trans, const struct bch_hash_info *info,
+				const struct qstr *str, struct qstr *out_cf)
+{
+	return bch_err_throw(trans->c, no_casefolding_without_utf8);
+}
+#endif
 
 static inline int bch2_maybe_casefold(struct btree_trans *trans,
 				      const struct bch_hash_info *info,
@@ -38,14 +45,6 @@ static inline int bch2_maybe_casefold(struct btree_trans *trans,
 		return bch2_casefold(trans, info, str, out_cf);
 	}
 }
-#else
-static inline int bch2_maybe_casefold(struct btree_trans *trans,
-				      const struct bch_hash_info *info,
-				      const struct qstr *str, struct qstr *out_cf)
-{
-	return bch_err_throw(trans->c, no_casefolding_without_utf8);
-}
-#endif
 
 struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent);
 
diff --git a/libbcachefs/ec.c b/libbcachefs/ec.c
index 62dda821..bea14f02 100644
--- a/libbcachefs/ec.c
+++ b/libbcachefs/ec.c
@@ -2060,6 +2060,9 @@ allocated:
 	BUG_ON(trans->restarted);
 	return h;
 err:
+	if (waiting &&
+	    !bch2_err_matches(ret, BCH_ERR_operation_blocked))
+		closure_wake_up(&c->freelist_wait);
 	bch2_ec_stripe_head_put(c, h);
 	return ERR_PTR(ret);
 }
diff --git a/libbcachefs/errcode.h b/libbcachefs/errcode.h
index 2de0dc91..cec8b0f4 100644
--- a/libbcachefs/errcode.h
+++ b/libbcachefs/errcode.h
@@ -90,6 +90,8 @@
 	x(ENOMEM,			ENOMEM_disk_accounting)			\
 	x(ENOMEM,			ENOMEM_stripe_head_alloc)		\
 	x(ENOMEM,                       ENOMEM_journal_read_bucket)             \
+	x(ENOMEM,                       ENOMEM_acl)				\
+	x(ENOMEM,                       ENOMEM_move_extent)			\
 	x(ENOSPC,			ENOSPC_disk_reservation)		\
 	x(ENOSPC,			ENOSPC_bucket_alloc)			\
 	x(ENOSPC,			ENOSPC_disk_label_add)			\
@@ -216,9 +218,13 @@
 	x(EINVAL,			varint_decode_error)			\
 	x(EINVAL,			erasure_coding_found_btree_node)	\
 	x(EINVAL,			option_negative)			\
+	x(EINVAL,			topology_repair)			\
+	x(BCH_ERR_topology_repair,	topology_repair_drop_this_node)		\
+	x(BCH_ERR_topology_repair,	topology_repair_drop_prev_node)		\
+	x(BCH_ERR_topology_repair,	topology_repair_did_fill_from_scan)	\
 	x(EOPNOTSUPP,			may_not_use_incompat_feature)		\
 	x(EOPNOTSUPP,			no_casefolding_without_utf8)		\
-	x(EOPNOTSUPP,			casefolding_disabled)		\
+	x(EOPNOTSUPP,			casefolding_disabled)			\
 	x(EOPNOTSUPP,			casefold_opt_is_dir_only)		\
 	x(EOPNOTSUPP,			unsupported_fsx_flag)			\
 	x(EOPNOTSUPP,			unsupported_fa_flag)			\
diff --git a/libbcachefs/extents.c b/libbcachefs/extents.c
index b36ecfc0..8152ef1c 100644
--- a/libbcachefs/extents.c
+++ b/libbcachefs/extents.c
@@ -1006,6 +1006,20 @@ const struct bch_extent_ptr *bch2_bkey_has_device_c(struct bkey_s_c k, unsigned
 	return NULL;
 }
 
+bool bch2_bkey_devs_rw(struct bch_fs *c, struct bkey_s_c k)
+{
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+
+	guard(rcu)();
+	bkey_for_each_ptr(ptrs, ptr) {
+		CLASS(bch2_dev_tryget, ca)(c, ptr->dev);
+		if (!ca || ca->mi.state != BCH_MEMBER_STATE_rw)
+			return false;
+	}
+
+	return true;
+}
+
 bool bch2_bkey_has_target(struct bch_fs *c, struct bkey_s_c k, unsigned target)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
diff --git a/libbcachefs/extents.h b/libbcachefs/extents.h
index f212f91c..35ee03cd 100644
--- a/libbcachefs/extents.h
+++ b/libbcachefs/extents.h
@@ -614,6 +614,8 @@ static inline struct bch_extent_ptr *bch2_bkey_has_device(struct bkey_s k, unsig
 	return (void *) bch2_bkey_has_device_c(k.s_c, dev);
 }
 
+bool bch2_bkey_devs_rw(struct bch_fs *, struct bkey_s_c);
+
 bool bch2_bkey_has_target(struct bch_fs *, struct bkey_s_c, unsigned);
 bool bch2_bkey_in_target(struct bch_fs *, struct bkey_s_c, unsigned);
 
diff --git a/libbcachefs/fs-io-direct.c b/libbcachefs/fs-io-direct.c
index 73d44875..e53fee05 100644
--- a/libbcachefs/fs-io-direct.c
+++ b/libbcachefs/fs-io-direct.c
@@ -127,7 +127,7 @@ static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter)
 	 * the dirtying of requests that are internal from the kernel (i.e. from
 	 * loopback), because we'll deadlock on page_lock.
 	 */
-	dio->should_dirty = iter_is_iovec(iter);
+	dio->should_dirty = user_backed_iter(iter);
 
 	blk_start_plug(&plug);
 
diff --git a/libbcachefs/fs-io.c b/libbcachefs/fs-io.c
index cc203752..93ad33f0 100644
--- a/libbcachefs/fs-io.c
+++ b/libbcachefs/fs-io.c
@@ -223,9 +223,8 @@ static int bch2_flush_inode(struct bch_fs *c,
 	if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_fsync))
 		return -EROFS;
 
-	CLASS(btree_trans, trans)(c);
 	u64 seq;
-	int ret = commit_do(trans, NULL, NULL, 0,
+	int ret = bch2_trans_commit_do(c, NULL, NULL, 0,
 			    bch2_get_inode_journal_seq_trans(trans, inode_inum(inode), &seq)) ?:
 		  bch2_journal_flush_seq(&c->journal, seq, TASK_INTERRUPTIBLE) ?:
 		  bch2_inode_flush_nocow_writes(c, inode);
diff --git a/libbcachefs/fs.c b/libbcachefs/fs.c
index 2789b30a..56b7126b 100644
--- a/libbcachefs/fs.c
+++ b/libbcachefs/fs.c
@@ -1295,8 +1295,14 @@ static int bch2_fill_extent(struct bch_fs *c,
 					       flags|
 					       FIEMAP_EXTENT_DELALLOC|
 					       FIEMAP_EXTENT_UNWRITTEN);
+	} else if (k.k->type == KEY_TYPE_error) {
+		return 0;
 	} else {
-		BUG();
+		WARN_ONCE(1, "unhandled key type %s",
+			  k.k->type < KEY_TYPE_MAX
+			  ? bch2_bkey_types[k.k->type]
+			  : "(unknown)");
+		return 0;
 	}
 }
 
diff --git a/libbcachefs/fsck.c b/libbcachefs/fsck.c
index df0aa252..183b88bb 100644
--- a/libbcachefs/fsck.c
+++ b/libbcachefs/fsck.c
@@ -1975,6 +1975,10 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
 		}
 	}
 
+	ret = check_extent_overbig(trans, iter, k);
+	if (ret)
+		goto err;
+
 	ret = bch2_trans_commit(trans, res, NULL, BCH_TRANS_COMMIT_no_enospc);
 	if (ret)
 		goto err;
@@ -2021,8 +2025,7 @@ int bch2_check_extents(struct bch_fs *c)
 				POS(BCACHEFS_ROOT_INO, 0),
 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k, ({
 			bch2_disk_reservation_put(c, &res);
-			check_extent(trans, &iter, k, &w, &s, &extent_ends, &res) ?:
-			check_extent_overbig(trans, &iter, k);
+			check_extent(trans, &iter, k, &w, &s, &extent_ends, &res);
 		})) ?:
 		check_i_sectors_notnested(trans, &w);
 
diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c
index 3ba1f9fd..97760e89 100644
--- a/libbcachefs/journal.c
+++ b/libbcachefs/journal.c
@@ -182,6 +182,8 @@ journal_error_check_stuck(struct journal *j, int error, unsigned flags)
 
 void bch2_journal_do_writes(struct journal *j)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+
 	for (u64 seq = journal_last_unwritten_seq(j);
 	     seq <= journal_cur_seq(j);
 	     seq++) {
@@ -196,6 +198,7 @@ void bch2_journal_do_writes(struct journal *j)
 		if (!journal_state_seq_count(j, j->reservations, seq)) {
 			j->seq_write_started = seq;
 			w->write_started = true;
+			closure_get(&c->cl);
 			closure_call(&w->io, bch2_journal_write, j->wq, NULL);
 		}
 
@@ -1063,6 +1066,8 @@ static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct jou
 			ret = journal_state_count(s, idx & JOURNAL_STATE_BUF_MASK) > open
 				? ERR_PTR(-EAGAIN)
 				: buf;
+			if (!ret)
+				smp_mb();
 			break;
 		}
 	}
diff --git a/libbcachefs/journal.h b/libbcachefs/journal.h
index b46b9718..c05aa942 100644
--- a/libbcachefs/journal.h
+++ b/libbcachefs/journal.h
@@ -267,7 +267,7 @@ static inline union journal_res_state journal_state_buf_put(struct journal *j, u
 {
 	union journal_res_state s;
 
-	s.v = atomic64_sub_return(((union journal_res_state) {
+	s.v = atomic64_sub_return_release(((union journal_res_state) {
 				    .buf0_count = idx == 0,
 				    .buf1_count = idx == 1,
 				    .buf2_count = idx == 2,
diff --git a/libbcachefs/journal_io.c b/libbcachefs/journal_io.c
index 2835250a..47224666 100644
--- a/libbcachefs/journal_io.c
+++ b/libbcachefs/journal_io.c
@@ -1820,6 +1820,8 @@ static CLOSURE_CALLBACK(journal_write_done)
 
 	if (do_discards)
 		bch2_do_discards(c);
+
+	closure_put(&c->cl);
 }
 
 static void journal_write_endio(struct bio *bio)
diff --git a/libbcachefs/journal_reclaim.c b/libbcachefs/journal_reclaim.c
index be50455c..f23e5ee9 100644
--- a/libbcachefs/journal_reclaim.c
+++ b/libbcachefs/journal_reclaim.c
@@ -874,7 +874,34 @@ static int journal_flush_done(struct journal *j, u64 seq_to_flush,
 	     --type)
 		if (journal_flush_pins_or_still_flushing(j, seq_to_flush, BIT(type))) {
 			*did_work = true;
-			return ret;
+
+			/*
+			 * Question from Dan Carpenter, on the early return:
+			 *
+			 * If journal_flush_pins_or_still_flushing() returns
+			 * true, then the flush hasn't complete and we must
+			 * return 0; we want the outer closure_wait_event() in
+			 * journal_flush_pins() to continue.
+			 *
+			 * The early return is there because we don't want to
+			 * call journal_entry_close() until we've finished
+			 * flushing all outstanding journal pins - otherwise
+			 * seq_to_flush can be U64_MAX, and we'll close a bunch
+			 * of journal entries and write tiny ones completely
+			 * unnecessarily.
+			 *
+			 * Having the early return be in the loop where we loop
+			 * over types is important, because flushing one journal
+			 * pin can cause new journal pins to be added (even of
+			 * the same type, btree node writes may generate more
+			 * btree node writes, when updating the parent pointer
+			 * has a full node and has to trigger a split/compact).
+			 *
+			 * This is part of our shutdown sequence, where order of
+			 * flushing is important in order to make sure that it
+			 * terminates...
+			 */
+			return 0;
 		}
 
 	if (seq_to_flush > journal_cur_seq(j))
diff --git a/libbcachefs/move.c b/libbcachefs/move.c
index 3f44bb54..84a228c4 100644
--- a/libbcachefs/move.c
+++ b/libbcachefs/move.c
@@ -330,7 +330,7 @@ int bch2_move_extent(struct moving_context *ctxt,
 {
 	struct btree_trans *trans = ctxt->trans;
 	struct bch_fs *c = trans->c;
-	int ret = -ENOMEM;
+	int ret = 0;
 
 	if (trace_io_move_enabled())
 		trace_io_move2(c, k, &io_opts, &data_opts);
@@ -351,11 +351,10 @@ int bch2_move_extent(struct moving_context *ctxt,
 
 	struct moving_io *io = allocate_dropping_locks(trans, ret,
 				kzalloc(sizeof(struct moving_io), _gfp));
-	if (!io)
-		goto err;
-
+	if (!io && !ret)
+		ret = bch_err_throw(c, ENOMEM_move_extent);
 	if (ret)
-		goto err_free;
+		goto err;
 
 	INIT_LIST_HEAD(&io->io_list);
 	io->write.ctxt		= ctxt;
@@ -366,7 +365,7 @@ int bch2_move_extent(struct moving_context *ctxt,
 		ret = bch2_data_update_init(trans, iter, ctxt, &io->write, ctxt->wp,
 					    &io_opts, data_opts, iter->btree_id, k);
 		if (ret)
-			goto err_free;
+			goto err;
 
 		io->write.op.end_io	= move_write_done;
 	} else {
@@ -380,7 +379,7 @@ int bch2_move_extent(struct moving_context *ctxt,
 
 		ret = bch2_data_update_bios_init(&io->write, c, &io_opts);
 		if (ret)
-			goto err_free;
+			goto err;
 	}
 
 	io->write.rbio.bio.bi_end_io = move_read_endio;
@@ -423,9 +422,8 @@ int bch2_move_extent(struct moving_context *ctxt,
 			   BCH_READ_last_fragment,
 			   data_opts.scrub ?  data_opts.read_dev : -1);
 	return 0;
-err_free:
-	kfree(io);
 err:
+	kfree(io);
 	if (bch2_err_matches(ret, EROFS) ||
 	    bch2_err_matches(ret, BCH_ERR_transaction_restart))
 		return ret;
@@ -795,50 +793,50 @@ out:
 	return ret;
 }
 
-int __bch2_move_data(struct moving_context *ctxt,
-		     struct bbpos start,
-		     struct bbpos end,
-		     move_pred_fn pred, void *arg)
+static int bch2_move_data(struct bch_fs *c,
+			  struct bbpos start,
+			  struct bbpos end,
+			  unsigned min_depth,
+			  struct bch_ratelimit *rate,
+			  struct bch_move_stats *stats,
+			  struct write_point_specifier wp,
+			  bool wait_on_copygc,
+			  move_pred_fn pred, void *arg)
 {
-	struct bch_fs *c = ctxt->trans->c;
-	enum btree_id id;
 	int ret = 0;
 
-	for (id = start.btree;
+	struct moving_context ctxt;
+	bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
+
+	for (enum btree_id id = start.btree;
 	     id <= min_t(unsigned, end.btree, btree_id_nr_alive(c) - 1);
 	     id++) {
-		ctxt->stats->pos = BBPOS(id, POS_MIN);
+		ctxt.stats->pos = BBPOS(id, POS_MIN);
 
-		if (!btree_type_has_ptrs(id) ||
-		    !bch2_btree_id_root(c, id)->b)
+		if (!bch2_btree_id_root(c, id)->b)
 			continue;
 
-		ret = bch2_move_data_btree(ctxt,
-				       id == start.btree ? start.pos : POS_MIN,
-				       id == end.btree   ? end.pos   : POS_MAX,
-				       pred, arg, id, 0);
+		unsigned min_depth_this_btree = min_depth;
+
+		if (!btree_type_has_ptrs(id))
+			min_depth_this_btree = max(min_depth_this_btree, 1);
+
+		for (unsigned level = min_depth_this_btree;
+		     level < BTREE_MAX_DEPTH;
+		     level++) {
+			ret = bch2_move_data_btree(&ctxt,
+						   id == start.btree ? start.pos : POS_MIN,
+						   id == end.btree   ? end.pos   : POS_MAX,
+						   pred, arg, id, level);
+			if (ret)
+				break;
+		}
+
 		if (ret)
 			break;
 	}
 
-	return ret;
-}
-
-int bch2_move_data(struct bch_fs *c,
-		   struct bbpos start,
-		   struct bbpos end,
-		   struct bch_ratelimit *rate,
-		   struct bch_move_stats *stats,
-		   struct write_point_specifier wp,
-		   bool wait_on_copygc,
-		   move_pred_fn pred, void *arg)
-{
-	struct moving_context ctxt;
-
-	bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
-	int ret = __bch2_move_data(&ctxt, start, end, pred, arg);
 	bch2_moving_ctxt_exit(&ctxt);
-
 	return ret;
 }
 
@@ -1206,14 +1204,6 @@ static bool migrate_pred(struct bch_fs *c, void *arg,
 	return data_opts->rewrite_ptrs != 0;
 }
 
-static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
-				   struct btree *b,
-				   struct bch_io_opts *io_opts,
-				   struct data_update_opts *data_opts)
-{
-	return rereplicate_pred(c, arg, b->c.btree_id, bkey_i_to_s_c(&b->key), io_opts, data_opts);
-}
-
 /*
  * Ancient versions of bcachefs produced packed formats which could represent
  * keys that the in memory format cannot represent; this checks for those
@@ -1293,15 +1283,6 @@ static bool drop_extra_replicas_pred(struct bch_fs *c, void *arg,
 	return data_opts->kill_ptrs != 0;
 }
 
-static bool drop_extra_replicas_btree_pred(struct bch_fs *c, void *arg,
-				   struct btree *b,
-				   struct bch_io_opts *io_opts,
-				   struct data_update_opts *data_opts)
-{
-	return drop_extra_replicas_pred(c, arg, b->c.btree_id, bkey_i_to_s_c(&b->key),
-					io_opts, data_opts);
-}
-
 static bool scrub_pred(struct bch_fs *c, void *_arg,
 		       enum btree_id btree, struct bkey_s_c k,
 		       struct bch_io_opts *io_opts,
@@ -1359,14 +1340,11 @@ int bch2_data_job(struct bch_fs *c,
 	case BCH_DATA_OP_rereplicate:
 		stats->data_type = BCH_DATA_journal;
 		ret = bch2_journal_flush_device_pins(&c->journal, -1);
-		ret = bch2_move_btree(c, start, end,
-				      rereplicate_btree_pred, c, stats) ?: ret;
-		ret = bch2_move_data(c, start, end,
-				     NULL,
-				     stats,
+		ret = bch2_move_data(c, start, end, 0, NULL, stats,
 				     writepoint_hashed((unsigned long) current),
 				     true,
 				     rereplicate_pred, c) ?: ret;
+		bch2_btree_interior_updates_flush(c);
 		ret = bch2_replicas_gc2(c) ?: ret;
 		break;
 	case BCH_DATA_OP_migrate:
@@ -1389,12 +1367,10 @@ int bch2_data_job(struct bch_fs *c,
 		ret = bch2_scan_old_btree_nodes(c, stats);
 		break;
 	case BCH_DATA_OP_drop_extra_replicas:
-		ret = bch2_move_btree(c, start, end,
-				drop_extra_replicas_btree_pred, c, stats) ?: ret;
-		ret = bch2_move_data(c, start, end, NULL, stats,
-				writepoint_hashed((unsigned long) current),
-				true,
-				drop_extra_replicas_pred, c) ?: ret;
+		ret = bch2_move_data(c, start, end, 0, NULL, stats,
+				     writepoint_hashed((unsigned long) current),
+				     true,
+				     drop_extra_replicas_pred, c) ?: ret;
 		ret = bch2_replicas_gc2(c) ?: ret;
 		break;
 	default:
diff --git a/libbcachefs/move.h b/libbcachefs/move.h
index fe92ca6d..481026ff 100644
--- a/libbcachefs/move.h
+++ b/libbcachefs/move.h
@@ -128,18 +128,6 @@ struct bch_io_opts *bch2_move_get_io_opts(struct btree_trans *,
 
 int bch2_move_data_btree(struct moving_context *, struct bpos, struct bpos,
 			 move_pred_fn, void *, enum btree_id, unsigned);
-int __bch2_move_data(struct moving_context *,
-		     struct bbpos,
-		     struct bbpos,
-		     move_pred_fn, void *);
-int bch2_move_data(struct bch_fs *,
-		   struct bbpos start,
-		   struct bbpos end,
-		   struct bch_ratelimit *,
-		   struct bch_move_stats *,
-		   struct write_point_specifier,
-		   bool,
-		   move_pred_fn, void *);
 
 int bch2_move_data_phys(struct bch_fs *, unsigned, u64, u64, unsigned,
 			struct bch_ratelimit *, struct bch_move_stats *,
diff --git a/libbcachefs/recovery.c b/libbcachefs/recovery.c
index a8eea478..58c159e5 100644
--- a/libbcachefs/recovery.c
+++ b/libbcachefs/recovery.c
@@ -37,78 +37,79 @@ int bch2_btree_lost_data(struct bch_fs *c,
 			 struct printbuf *msg,
 			 enum btree_id btree)
 {
-	u64 b = BIT_ULL(btree);
 	int ret = 0;
 
 	guard(mutex)(&c->sb_lock);
+	bool write_sb = false;
 	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 
-	if (!(c->sb.btrees_lost_data & b)) {
+	if (!(c->sb.btrees_lost_data & BIT_ULL(btree))) {
 		prt_printf(msg, "flagging btree ");
 		bch2_btree_id_to_text(msg, btree);
 		prt_printf(msg, " lost data\n");
 
-		ext->btrees_lost_data |= cpu_to_le64(b);
+		write_sb |= !__test_and_set_bit_le64(btree, &ext->btrees_lost_data);
 	}
 
 	/* Once we have runtime self healing for topology errors we won't need this: */
-	ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_topology, 0) ?: ret;
+	ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_topology, 0, &write_sb) ?: ret;
 
 	/* Btree node accounting will be off: */
-	__set_bit_le64(BCH_FSCK_ERR_accounting_mismatch, ext->errors_silent);
-	ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_allocations, 0) ?: ret;
+	write_sb |= !__test_and_set_bit_le64(BCH_FSCK_ERR_accounting_mismatch, ext->errors_silent);
+	ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_allocations, 0, &write_sb) ?: ret;
 
 #ifdef CONFIG_BCACHEFS_DEBUG
 	/*
 	 * These are much more minor, and don't need to be corrected right away,
 	 * but in debug mode we want the next fsck run to be clean:
 	 */
-	ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_lrus, 0) ?: ret;
-	ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_backpointers_to_extents, 0) ?: ret;
+	ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_lrus, 0, &write_sb) ?: ret;
+	ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_backpointers_to_extents, 0, &write_sb) ?: ret;
 #endif
 
 	switch (btree) {
 	case BTREE_ID_alloc:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0) ?: ret;
-
-		__set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent);
-		__set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent);
-		__set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
-		__set_bit_le64(BCH_FSCK_ERR_alloc_key_cached_sectors_wrong, ext->errors_silent);
-		__set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_wrong, ext->errors_silent);
-		__set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_redundancy_wrong, ext->errors_silent);
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0, &write_sb) ?: ret;
+
+		write_sb |= !__test_and_set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent);
+		write_sb |= !__test_and_set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent);
+		write_sb |= !__test_and_set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
+		write_sb |= !__test_and_set_bit_le64(BCH_FSCK_ERR_alloc_key_cached_sectors_wrong, ext->errors_silent);
+		write_sb |= !__test_and_set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_wrong, ext->errors_silent);
+		write_sb |= !__test_and_set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_redundancy_wrong, ext->errors_silent);
 		goto out;
 	case BTREE_ID_backpointers:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_btree_backpointers, 0) ?: ret;
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_extents_to_backpointers, 0) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_btree_backpointers, 0, &write_sb) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_extents_to_backpointers, 0, &write_sb) ?: ret;
 		goto out;
 	case BTREE_ID_need_discard:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0, &write_sb) ?: ret;
 		goto out;
 	case BTREE_ID_freespace:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0, &write_sb) ?: ret;
 		goto out;
 	case BTREE_ID_bucket_gens:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0, &write_sb) ?: ret;
 		goto out;
 	case BTREE_ID_lru:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_alloc_info, 0, &write_sb) ?: ret;
 		goto out;
 	case BTREE_ID_accounting:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_allocations, 0) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_allocations, 0, &write_sb) ?: ret;
 		goto out;
 	case BTREE_ID_snapshots:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_reconstruct_snapshots, 0) ?: ret;
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_topology, 0) ?: ret;
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_scan_for_btree_nodes, 0) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_reconstruct_snapshots, 0, &write_sb) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_topology, 0, &write_sb) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_scan_for_btree_nodes, 0, &write_sb) ?: ret;
 		goto out;
 	default:
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_topology, 0) ?: ret;
-		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_scan_for_btree_nodes, 0) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_check_topology, 0, &write_sb) ?: ret;
+		ret = __bch2_run_explicit_recovery_pass(c, msg, BCH_RECOVERY_PASS_scan_for_btree_nodes, 0, &write_sb) ?: ret;
 		goto out;
 	}
 out:
-	bch2_write_super(c);
+	if (write_sb)
+		bch2_write_super(c);
 	return ret;
 }
 
diff --git a/libbcachefs/recovery_passes.c b/libbcachefs/recovery_passes.c
index f9d1c492..b2cdd111 100644
--- a/libbcachefs/recovery_passes.c
+++ b/libbcachefs/recovery_passes.c
@@ -340,7 +340,8 @@ static bool recovery_pass_needs_set(struct bch_fs *c,
 int __bch2_run_explicit_recovery_pass(struct bch_fs *c,
 				      struct printbuf *out,
 				      enum bch_recovery_pass pass,
-				      enum bch_run_recovery_pass_flags flags)
+				      enum bch_run_recovery_pass_flags flags,
+				      bool *write_sb)
 {
 	struct bch_fs_recovery *r = &c->recovery;
 	int ret = 0;
@@ -362,7 +363,8 @@ int __bch2_run_explicit_recovery_pass(struct bch_fs *c,
 
 	if (!(flags & RUN_RECOVERY_PASS_nopersistent)) {
 		struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
-		__set_bit_le64(bch2_recovery_pass_to_stable(pass), ext->recovery_passes_required);
+		*write_sb |= !__test_and_set_bit_le64(bch2_recovery_pass_to_stable(pass),
+						     ext->recovery_passes_required);
 	}
 
 	if (pass < BCH_RECOVERY_PASS_set_may_go_rw &&
@@ -408,14 +410,19 @@ int bch2_run_explicit_recovery_pass(struct bch_fs *c,
 				    enum bch_recovery_pass pass,
 				    enum bch_run_recovery_pass_flags flags)
 {
-	int ret = 0;
+	/*
+	 * With RUN_RECOVERY_PASS_ratelimit, recovery_pass_needs_set needs
+	 * sb_lock
+	 */
+	if (!(flags & RUN_RECOVERY_PASS_ratelimit) &&
+	    !recovery_pass_needs_set(c, pass, &flags))
+		return 0;
 
-	if (recovery_pass_needs_set(c, pass, &flags)) {
-		guard(mutex)(&c->sb_lock);
-		ret = __bch2_run_explicit_recovery_pass(c, out, pass, flags);
+	guard(mutex)(&c->sb_lock);
+	bool write_sb = false;
+	int ret = __bch2_run_explicit_recovery_pass(c, out, pass, flags, &write_sb);
+	if (write_sb)
 		bch2_write_super(c);
-	}
-
 	return ret;
 }
 
@@ -438,14 +445,13 @@ int bch2_require_recovery_pass(struct bch_fs *c,
 		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);
+	bool write_sb = false;
+	int ret = __bch2_run_explicit_recovery_pass(c, out, pass, flags, &write_sb) ?:
+		bch_err_throw(c, recovery_pass_will_run);
+	if (write_sb)
 		bch2_write_super(c);
-	}
-
-	return ret ?: bch_err_throw(c, recovery_pass_will_run);
+	return ret;
 }
 
 int bch2_run_print_explicit_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
@@ -459,8 +465,10 @@ int bch2_run_print_explicit_recovery_pass(struct bch_fs *c, enum bch_recovery_pa
 	bch2_log_msg_start(c, &buf);
 
 	guard(mutex)(&c->sb_lock);
+	bool write_sb = false;
 	int ret = __bch2_run_explicit_recovery_pass(c, &buf, pass,
-						RUN_RECOVERY_PASS_nopersistent);
+						RUN_RECOVERY_PASS_nopersistent,
+						&write_sb);
 
 	bch2_print_str(c, KERN_NOTICE, buf.buf);
 	return ret;
diff --git a/libbcachefs/recovery_passes.h b/libbcachefs/recovery_passes.h
index 2117f0ce..4f2c2f81 100644
--- a/libbcachefs/recovery_passes.h
+++ b/libbcachefs/recovery_passes.h
@@ -30,7 +30,8 @@ int bch2_run_print_explicit_recovery_pass(struct bch_fs *, enum bch_recovery_pas
 
 int __bch2_run_explicit_recovery_pass(struct bch_fs *, struct printbuf *,
 				      enum bch_recovery_pass,
-				      enum bch_run_recovery_pass_flags);
+				      enum bch_run_recovery_pass_flags,
+				      bool *);
 int bch2_run_explicit_recovery_pass(struct bch_fs *, struct printbuf *,
 				    enum bch_recovery_pass,
 				    enum bch_run_recovery_pass_flags);
diff --git a/libbcachefs/str_hash.c b/libbcachefs/str_hash.c
index dfe4b6ae..3e08e55d 100644
--- a/libbcachefs/str_hash.c
+++ b/libbcachefs/str_hash.c
@@ -329,7 +329,6 @@ duplicate_entries:
 out:
 fsck_err:
 	bch2_trans_iter_exit(trans, dup_iter);
-	printbuf_exit(&buf);
 	if (free_snapshots_seen)
 		darray_exit(&s->ids);
 	return ret;
diff --git a/libbcachefs/super.c b/libbcachefs/super.c
index 4e038f65..0fc0b222 100644
--- a/libbcachefs/super.c
+++ b/libbcachefs/super.c
@@ -514,6 +514,10 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 	if (ret)
 		return ret;
 
+	ret = bch2_fs_mark_dirty(c);
+	if (ret)
+		return ret;
+
 	clear_bit(BCH_FS_clean_shutdown, &c->flags);
 
 	scoped_guard(rcu)
@@ -537,10 +541,6 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 		bch2_journal_space_available(&c->journal);
 	}
 
-	ret = bch2_fs_mark_dirty(c);
-	if (ret)
-		return ret;
-
 	/*
 	 * Don't jump to our error path, and call bch2_fs_read_only(), unless we
 	 * successfully marked the filesystem dirty
diff --git a/libbcachefs/sysfs.c b/libbcachefs/sysfs.c
index 158f526e..bd3fa9c3 100644
--- a/libbcachefs/sysfs.c
+++ b/libbcachefs/sysfs.c
@@ -18,6 +18,7 @@
 #include "btree_key_cache.h"
 #include "btree_update.h"
 #include "btree_update_interior.h"
+#include "btree_write_buffer.h"
 #include "btree_gc.h"
 #include "buckets.h"
 #include "clock.h"
@@ -150,6 +151,7 @@ write_attribute(trigger_journal_flush);
 write_attribute(trigger_journal_writes);
 write_attribute(trigger_btree_cache_shrink);
 write_attribute(trigger_btree_key_cache_shrink);
+write_attribute(trigger_btree_write_buffer_flush);
 write_attribute(trigger_btree_updates);
 write_attribute(trigger_freelist_wakeup);
 write_attribute(trigger_recalc_capacity);
@@ -539,6 +541,11 @@ STORE(bch2_fs)
 		c->btree_key_cache.shrink->scan_objects(c->btree_key_cache.shrink, &sc);
 	}
 
+	if (attr == &sysfs_trigger_btree_write_buffer_flush)
+		bch2_trans_do(c,
+			      (bch2_btree_write_buffer_flush_sync(trans),
+			       bch2_trans_begin(trans)));
+
 	if (attr == &sysfs_trigger_gc)
 		bch2_gc_gens(c);
 
@@ -709,6 +716,7 @@ struct attribute *bch2_fs_internal_files[] = {
 	&sysfs_trigger_journal_writes,
 	&sysfs_trigger_btree_cache_shrink,
 	&sysfs_trigger_btree_key_cache_shrink,
+	&sysfs_trigger_btree_write_buffer_flush,
 	&sysfs_trigger_btree_updates,
 	&sysfs_trigger_freelist_wakeup,
 	&sysfs_trigger_recalc_capacity,
diff --git a/libbcachefs/util.h b/libbcachefs/util.h
index 768528c2..52ac8230 100644
--- a/libbcachefs/util.h
+++ b/libbcachefs/util.h
@@ -733,6 +733,13 @@ static inline bool test_bit_le64(size_t bit, __le64 *addr)
 	return (addr[bit / 64] & cpu_to_le64(BIT_ULL(bit % 64))) != 0;
 }
 
+static inline bool __test_and_set_bit_le64(size_t bit, __le64 *addr)
+{
+	bool ret = test_bit_le64(bit, addr);
+	__set_bit_le64(bit, addr);
+	return ret;
+}
+
 static inline void memcpy_swab(void *_dst, void *_src, size_t len)
 {
 	u8 *dst = _dst + len;
diff --git a/linux/closure.c b/linux/closure.c
index 2bfe7d2a..4fb78d18 100644
--- a/linux/closure.c
+++ b/linux/closure.c
@@ -13,23 +13,25 @@
 #include <linux/seq_file.h>
 #include <linux/sched/debug.h>
 
-static inline void closure_put_after_sub_checks(int flags)
+static inline void closure_put_after_sub_checks(struct closure *cl, int flags)
 {
 	int r = flags & CLOSURE_REMAINING_MASK;
 
 	if (WARN(flags & CLOSURE_GUARD_MASK,
-		 "closure has guard bits set: %x (%u)",
+		 "closure %ps has guard bits set: %x (%u)",
+		 cl->fn,
 		 flags & CLOSURE_GUARD_MASK, (unsigned) __fls(r)))
 		r &= ~CLOSURE_GUARD_MASK;
 
 	WARN(!r && (flags & ~CLOSURE_DESTRUCTOR),
-	     "closure ref hit 0 with incorrect flags set: %x (%u)",
+	     "closure %ps ref hit 0 with incorrect flags set: %x (%u)",
+	     cl->fn,
 	     flags & ~CLOSURE_DESTRUCTOR, (unsigned) __fls(flags));
 }
 
 static inline void closure_put_after_sub(struct closure *cl, int flags)
 {
-	closure_put_after_sub_checks(flags);
+	closure_put_after_sub_checks(cl, flags);
 
 	if (!(flags & CLOSURE_REMAINING_MASK)) {
 		smp_acquire__after_ctrl_dep();
@@ -167,7 +169,7 @@ void __sched closure_return_sync(struct closure *cl)
 	unsigned flags = atomic_sub_return_release(1 + CLOSURE_RUNNING - CLOSURE_DESTRUCTOR,
 						   &cl->remaining);
 
-	closure_put_after_sub_checks(flags);
+	closure_put_after_sub_checks(cl, flags);
 
 	if (unlikely(flags & CLOSURE_REMAINING_MASK)) {
 		while (1) {