Update bcachefs sources to f595b42bf8ea bcachefs: set reconcile.pending on -ENOSPC

.bcachefs_revision

@@ -1 +1 @@
-b3616d33633509373c40c441e8c32ea554e2fa35
+f595b42bf8eae730a95de7636238556ef9e86cee

libbcachefs/alloc/foreground.c

@@ -1196,36 +1196,19 @@ deallocate_extra_replicas(struct bch_fs *c,
 /*
  * Get us an open_bucket we can allocate from, return with it locked:
  */
-int bch2_alloc_sectors_start_trans(struct btree_trans *trans,
+int bch2_alloc_sectors_req(struct btree_trans *trans,
-			     unsigned target,
+			   struct alloc_request *req,
-			     unsigned erasure_code,
+			   struct write_point_specifier write_point,
-			     struct write_point_specifier write_point,
+			   unsigned nr_replicas_required,
-			     struct bch_devs_list *devs_have,
+			   struct closure *cl,
-			     unsigned nr_replicas,
+			   struct write_point **wp_ret)
-			     unsigned nr_replicas_required,
-			     enum bch_watermark watermark,
-			     enum bch_write_flags flags,
-			     struct closure *cl,
-			     struct write_point **wp_ret)
 {
 	struct bch_fs *c = trans->c;
 	struct open_bucket *ob;
 	unsigned write_points_nr;
 	int i;
 
-	struct alloc_request *req = errptr_try(bch2_trans_kmalloc_nomemzero(trans, sizeof(*req)));
+	BUG_ON(!req->nr_replicas || !nr_replicas_required);
-
-	if (!IS_ENABLED(CONFIG_BCACHEFS_ERASURE_CODING))
-		erasure_code = false;
-
-	req->nr_replicas	= nr_replicas;
-	req->target		= target;
-	req->ec			= erasure_code;
-	req->watermark		= watermark;
-	req->flags		= flags;
-	req->devs_have		= devs_have;
-
-	BUG_ON(!nr_replicas || !nr_replicas_required);
 retry:
 	req->ptrs.nr		= 0;
 	req->nr_effective	= 0;
@@ -1244,7 +1227,7 @@ retry:
 	if (req->data_type != BCH_DATA_user)
 		req->have_cache = true;
 
-	if (target && !(flags & BCH_WRITE_only_specified_devs)) {
+	if (req->target && !(req->flags & BCH_WRITE_only_specified_devs)) {
 		ret = open_bucket_add_buckets(trans, req, NULL);
 		if (!ret ||
 		    bch2_err_matches(ret, BCH_ERR_transaction_restart))
@@ -1275,7 +1258,7 @@ retry:
 alloc_done:
 	BUG_ON(!ret && req->nr_effective < req->nr_replicas);
 
-	if (erasure_code && !ec_open_bucket(c, &req->ptrs))
+	if (req->ec && !ec_open_bucket(c, &req->ptrs))
 		pr_debug("failed to get ec bucket: ret %u", ret);
 
 	if (ret == -BCH_ERR_insufficient_devices &&
@@ -1341,7 +1324,7 @@ err:
 	if (cl && bch2_err_matches(ret, BCH_ERR_open_buckets_empty))
 		ret = bch_err_throw(c, bucket_alloc_blocked);
 
-	if (cl && !(flags & BCH_WRITE_alloc_nowait) &&
+	if (cl && !(req->flags & BCH_WRITE_alloc_nowait) &&
 	    bch2_err_matches(ret, BCH_ERR_freelist_empty))
 		ret = bch_err_throw(c, bucket_alloc_blocked);
 

libbcachefs/alloc/foreground.h

@@ -5,6 +5,7 @@
 #include "bcachefs.h"
 #include "alloc/buckets.h"
 #include "alloc/types.h"
+#include "btree/iter.h"
 #include "data/extents.h"
 #include "data/write_types.h"
 #include "sb/members.h"
@@ -221,15 +222,51 @@ enum bch_write_flags;
 int bch2_bucket_alloc_set_trans(struct btree_trans *, struct alloc_request *,
 				struct dev_stripe_state *, struct closure *);
 
-int bch2_alloc_sectors_start_trans(struct btree_trans *,
+int bch2_alloc_sectors_req(struct btree_trans *, struct alloc_request *,
-				   unsigned, unsigned,
+			   struct write_point_specifier, unsigned,
-				   struct write_point_specifier,
+			   struct closure *, struct write_point **);
-				   struct bch_devs_list *,
+
-				   unsigned, unsigned,
+static inline struct alloc_request *alloc_request_get(struct btree_trans *trans,
-				   enum bch_watermark,
+						      unsigned target,
-				   enum bch_write_flags,
+						      unsigned erasure_code,
-				   struct closure *,
+						      struct bch_devs_list *devs_have,
-				   struct write_point **);
+						      unsigned nr_replicas,
+						      enum bch_watermark watermark,
+						      enum bch_write_flags flags)
+{
+	struct alloc_request *req = bch2_trans_kmalloc_nomemzero(trans, sizeof(*req));
+	if (IS_ERR(req))
+		return req;
+
+	if (!IS_ENABLED(CONFIG_BCACHEFS_ERASURE_CODING))
+		erasure_code = false;
+
+	req->nr_replicas	= nr_replicas;
+	req->target		= target;
+	req->ec			= erasure_code;
+	req->watermark		= watermark;
+	req->flags		= flags;
+	req->devs_have		= devs_have;
+	return req;
+}
+
+static inline int bch2_alloc_sectors_start_trans(struct btree_trans *trans,
+			     unsigned target,
+			     unsigned erasure_code,
+			     struct write_point_specifier write_point,
+			     struct bch_devs_list *devs_have,
+			     unsigned nr_replicas,
+			     unsigned nr_replicas_required,
+			     enum bch_watermark watermark,
+			     enum bch_write_flags flags,
+			     struct closure *cl,
+			     struct write_point **wp_ret)
+{
+	struct alloc_request *req = errptr_try(alloc_request_get(trans, target, erasure_code,
+								 devs_have, nr_replicas,
+								 watermark, flags));
+	return bch2_alloc_sectors_req(trans, req, write_point, nr_replicas_required, cl, wp_ret);
+}
 
 static inline struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *c, struct open_bucket *ob)
 {

libbcachefs/btree/bkey.h

@@ -205,12 +205,21 @@ static __always_inline bool bversion_eq(struct bversion l, struct bversion r)
 
 static inline bool bkey_and_val_eq(struct bkey_s_c l, struct bkey_s_c r)
 {
-	return  l.k->u64s == r.k->u64s &&
+	if (!(l.k->u64s == r.k->u64s &&
-		l.k->type == r.k->type &&
+	      l.k->type == r.k->type &&
-		bpos_eq(l.k->p, r.k->p) &&
+	      bpos_eq(l.k->p, r.k->p) &&
-		bversion_eq(l.k->bversion, r.k->bversion) &&
+	      bversion_eq(l.k->bversion, r.k->bversion) &&
-		l.k->size == r.k->size &&
+	      l.k->size == r.k->size))
-		!memcmp(l.v, r.v, bkey_val_bytes(l.k));
+		return false;
+
+	if (l.k->type != KEY_TYPE_btree_ptr_v2) {
+		return !memcmp(l.v, r.v, bkey_val_bytes(l.k));
+	} else {
+		/* don't compare bch_btree_ptr_v2.mem_ptr */
+		return !memcmp((void *) l.v + 8,
+			       (void *) r.v + 8,
+			       bkey_val_bytes(l.k) - 8);
+	}
 }
 
 #define ZERO_VERSION	((struct bversion) { .hi = 0, .lo = 0 })

libbcachefs/btree/commit.c

@@ -67,13 +67,15 @@ static void verify_update_old_key(struct btree_trans *trans, struct btree_insert
 			k = bkey_i_to_s_c(j_k);
 	}
 
+	/* when updating btree ptrs, mem_ptr may change underneath us, unlocked */
+
 	struct bkey_s_c old = { &i->old_k, i->old_v };
 	if (unlikely(!bkey_and_val_eq(k, old))) {
 		CLASS(printbuf, buf)();
-		prt_str(&buf, "updated cached old key doesn't match\n");
+		prt_str(&buf, "updated cached old key doesn't match");
-		prt_str(&buf, "cached: ");
+		prt_str(&buf, "\ncached: ");
 		bch2_bkey_val_to_text(&buf, c, old);
-		prt_str(&buf, "real:   ");
+		prt_str(&buf, "\nreal:   ");
 		bch2_bkey_val_to_text(&buf, c, k);
 		panic("%s\n", buf.buf);
 	}

libbcachefs/btree/interior.c

@@ -321,16 +321,13 @@ static bool can_use_btree_node(struct bch_fs *c,
 
 static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans,
 					     struct disk_reservation *res,
-					     struct closure *cl,
 					     bool interior_node,
-					     unsigned target,
+					     struct alloc_request *req,
-					     enum bch_trans_commit_flags flags)
+					     struct closure *cl)
 {
 	struct bch_fs *c = trans->c;
 	struct write_point *wp;
 	struct btree *b;
-	struct bch_devs_list devs_have = (struct bch_devs_list) { 0 };
-	enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
 	int ret;
 
 	b = bch2_btree_node_mem_alloc(trans, interior_node);
@@ -339,18 +336,10 @@ static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans,
 
 	BUG_ON(b->ob.nr);
 retry:
-	ret = bch2_alloc_sectors_start_trans(trans,
+	ret = bch2_alloc_sectors_req(trans, req,
-				      target ?:
-				      c->opts.metadata_target ?:
-				      c->opts.foreground_target,
-				      0,
 				      writepoint_ptr(&c->btree_write_point),
-				      &devs_have,
-				      res->nr_replicas,
 				      min(res->nr_replicas,
 					  c->opts.metadata_replicas_required),
-				      watermark,
-				      target ? BCH_WRITE_only_specified_devs : 0,
 				      cl, &wp);
 	if (unlikely(ret))
 		goto err;
@@ -373,7 +362,9 @@ retry:
 
 		/* check if it has sufficient durability */
 
-		if (can_use_btree_node(c, res, target, bkey_i_to_s_c(&a->k))) {
+		if (can_use_btree_node(c, res,
+				       req->flags & BCH_WRITE_only_specified_devs ? req->target : 0,
+				       bkey_i_to_s_c(&a->k))) {
 			bkey_copy(&b->key, &a->k);
 			b->ob = a->ob;
 			mutex_unlock(&c->btree_reserve_cache_lock);
@@ -546,8 +537,7 @@ static void bch2_btree_reserve_put(struct btree_update *as, struct btree_trans *
 static int bch2_btree_reserve_get(struct btree_trans *trans,
 				  struct btree_update *as,
 				  unsigned nr_nodes[2],
-				  unsigned target,
+				  struct alloc_request *req,
-				  unsigned flags,
 				  struct closure *cl)
 {
 	BUG_ON(nr_nodes[0] + nr_nodes[1] > BTREE_RESERVE_MAX);
@@ -564,7 +554,7 @@ static int bch2_btree_reserve_get(struct btree_trans *trans,
 
 		while (p->nr < nr_nodes[interior]) {
 			struct btree *b = __bch2_btree_node_alloc(trans, &as->disk_res,
-							cl, interior, target, flags);
+								  interior, req, cl);
 			ret = PTR_ERR_OR_ZERO(b);
 			if (ret)
 				goto err;
@@ -1196,16 +1186,17 @@ static struct btree_update *
 bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 			unsigned level_start, bool split,
 			unsigned target,
-			enum bch_trans_commit_flags flags)
+			enum bch_trans_commit_flags commit_flags,
+			enum bch_write_flags write_flags)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_update *as;
 	u64 start_time = local_clock();
-	int disk_res_flags = (flags & BCH_TRANS_COMMIT_no_enospc)
+	int disk_res_flags = (commit_flags & BCH_TRANS_COMMIT_no_enospc)
 		? BCH_DISK_RESERVATION_NOFAIL : 0;
 	unsigned nr_nodes[2] = { 0, 0 };
 	unsigned level_end = level_start;
-	enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
+	enum bch_watermark watermark = commit_flags & BCH_WATERMARK_MASK;
 	int ret = 0;
 	u32 restart_count = trans->restart_count;
 
@@ -1216,12 +1207,12 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	if (watermark < BCH_WATERMARK_btree)
 		watermark = BCH_WATERMARK_btree;
 
-	flags &= ~BCH_WATERMARK_MASK;
+	commit_flags &= ~BCH_WATERMARK_MASK;
-	flags |= watermark;
+	commit_flags |= watermark;
 
 	if (watermark < BCH_WATERMARK_reclaim &&
 	    journal_low_on_space(&c->journal)) {
-		if (flags & BCH_TRANS_COMMIT_journal_reclaim)
+		if (commit_flags & BCH_TRANS_COMMIT_journal_reclaim)
 			return ERR_PTR(-BCH_ERR_journal_reclaim_would_deadlock);
 
 		ret = drop_locks_do(trans,
@@ -1271,7 +1262,7 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	as->start_time		= start_time;
 	as->ip_started		= _RET_IP_;
 	as->mode		= BTREE_UPDATE_none;
-	as->flags		= flags;
+	as->flags		= commit_flags;
 	as->took_gc_lock	= true;
 	as->btree_id		= path->btree_id;
 	as->update_level_start	= level_start;
@@ -1314,26 +1305,38 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	if (ret)
 		goto err;
 
-	ret = bch2_btree_reserve_get(trans, as, nr_nodes, target, flags, NULL);
+	struct bch_devs_list devs_have = (struct bch_devs_list) { 0 };
+	struct alloc_request *req = alloc_request_get(trans,
+				      target ?:
+				      c->opts.metadata_target ?:
+				      c->opts.foreground_target,
+				      false,
+				      &devs_have,
+				      as->disk_res.nr_replicas,
+				      watermark,
+				      write_flags);
+	ret = PTR_ERR_OR_ZERO(req);
+	if (ret)
+		goto err;
+
+	ret = bch2_btree_reserve_get(trans, as, nr_nodes, req, NULL);
 	if (bch2_err_matches(ret, ENOSPC) ||
 	    bch2_err_matches(ret, ENOMEM)) {
-		struct closure cl;
-
 		/*
 		 * XXX: this should probably be a separate BTREE_INSERT_NONBLOCK
 		 * flag
 		 */
 		if (bch2_err_matches(ret, ENOSPC) &&
-		    (flags & BCH_TRANS_COMMIT_journal_reclaim) &&
+		    (commit_flags & BCH_TRANS_COMMIT_journal_reclaim) &&
 		    watermark < BCH_WATERMARK_reclaim) {
 			ret = bch_err_throw(c, journal_reclaim_would_deadlock);
 			goto err;
 		}
 
-		closure_init_stack(&cl);
+		CLASS(closure_stack, cl)();
 
 		do {
-			ret = bch2_btree_reserve_get(trans, as, nr_nodes, target, flags, &cl);
+			ret = bch2_btree_reserve_get(trans, as, nr_nodes, req, &cl);
 			if (!bch2_err_matches(ret, BCH_ERR_operation_blocked))
 				break;
 			bch2_trans_unlock(trans);
@@ -1945,7 +1948,7 @@ int bch2_btree_split_leaf(struct btree_trans *trans,
 
 	as = bch2_btree_update_start(trans, trans->paths + path,
 				     trans->paths[path].level,
-				     true, 0, flags);
+				     true, 0, flags, 0);
 	if (IS_ERR(as))
 		return PTR_ERR(as);
 
@@ -2016,7 +2019,7 @@ int bch2_btree_increase_depth(struct btree_trans *trans, btree_path_idx_t path,
 
 	struct btree_update *as =
 		bch2_btree_update_start(trans, trans->paths + path, b->c.level,
-					true, 0, flags);
+					true, 0, flags, 0);
 	if (IS_ERR(as))
 		return PTR_ERR(as);
 
@@ -2144,7 +2147,7 @@ int __bch2_foreground_maybe_merge(struct btree_trans *trans,
 
 	parent = btree_node_parent(trans->paths + path, b);
 	as = bch2_btree_update_start(trans, trans->paths + path, level, false,
-				     0, BCH_TRANS_COMMIT_no_enospc|flags);
+				     0, BCH_TRANS_COMMIT_no_enospc|flags, 0);
 	ret = PTR_ERR_OR_ZERO(as);
 	if (ret)
 		goto err;
@@ -2246,11 +2249,12 @@ int bch2_btree_node_get_iter(struct btree_trans *trans, struct btree_iter *iter,
 	return 0;
 }
 
-int bch2_btree_node_rewrite(struct btree_trans *trans,
+static int bch2_btree_node_rewrite(struct btree_trans *trans,
-			    struct btree_iter *iter,
+				   struct btree_iter *iter,
-			    struct btree *b,
+				   struct btree *b,
-			    unsigned target,
+				   unsigned target,
-			    enum bch_trans_commit_flags flags)
+				   enum bch_trans_commit_flags commit_flags,
+				   enum bch_write_flags write_flags)
 {
 	BUG_ON(btree_node_fake(b));
 
@@ -2260,12 +2264,12 @@ int bch2_btree_node_rewrite(struct btree_trans *trans,
 	btree_path_idx_t new_path = 0;
 	int ret;
 
-	flags |= BCH_TRANS_COMMIT_no_enospc;
+	commit_flags |= BCH_TRANS_COMMIT_no_enospc;
 
 	struct btree_path *path = btree_iter_path(trans, iter);
 	parent = btree_node_parent(path, b);
-	as = bch2_btree_update_start(trans, path, b->c.level,
+	as = bch2_btree_update_start(trans, path, b->c.level, false, target,
-				     false, target, flags);
+				     commit_flags, write_flags);
 	ret = PTR_ERR_OR_ZERO(as);
 	if (ret)
 		goto out;
@@ -2326,7 +2330,7 @@ int bch2_btree_node_rewrite_key(struct btree_trans *trans,
 
 	bool found = b && btree_ptr_hash_val(&b->key) == btree_ptr_hash_val(k);
 	return found
-		? bch2_btree_node_rewrite(trans, &iter, b, 0, flags)
+		? bch2_btree_node_rewrite(trans, &iter, b, 0, flags, 0)
 		: -ENOENT;
 }
 
@@ -2334,7 +2338,8 @@ int bch2_btree_node_rewrite_pos(struct btree_trans *trans,
 				enum btree_id btree, unsigned level,
 				struct bpos pos,
 				unsigned target,
-				enum bch_trans_commit_flags flags)
+				enum bch_trans_commit_flags commit_flags,
+				enum bch_write_flags write_flags)
 {
 	BUG_ON(!level);
 
@@ -2342,19 +2347,7 @@ int bch2_btree_node_rewrite_pos(struct btree_trans *trans,
 	CLASS(btree_node_iter, iter)(trans, btree, pos, 0, level - 1, 0);
 	struct btree *b = errptr_try(bch2_btree_iter_peek_node(&iter));
 
-	return bch2_btree_node_rewrite(trans, &iter, b, target, flags);
+	return bch2_btree_node_rewrite(trans, &iter, b, target, commit_flags, write_flags);
-}
-
-int bch2_btree_node_rewrite_key_get_iter(struct btree_trans *trans,
-					 struct btree *b,
-					 enum bch_trans_commit_flags flags)
-{
-	CLASS(btree_iter_uninit, iter)(trans);
-	int ret = bch2_btree_node_get_iter(trans, &iter, b);
-	if (ret)
-		return ret == -BCH_ERR_btree_node_dying ? 0 : ret;
-
-	return bch2_btree_node_rewrite(trans, &iter, b, 0, flags);
 }
 
 struct async_btree_rewrite {
@@ -2476,6 +2469,11 @@ static int __bch2_btree_node_update_key(struct btree_trans *trans,
 					bool skip_triggers)
 {
 	struct bch_fs *c = trans->c;
+	unsigned level = b->c.level;
+
+	struct btree_path *path = btree_iter_path(trans, iter);
+	BUG_ON(path->l[b->c.level].b != b);
+	BUG_ON(!btree_node_intent_locked(path, b->c.level));
 
 	if (!btree_node_will_make_reachable(b)) {
 		if (!btree_node_is_root(c, b)) {
@@ -2518,6 +2516,17 @@ static int __bch2_btree_node_update_key(struct btree_trans *trans,
 		CLASS(disk_reservation, res)(c);
 		try(bch2_trans_commit(trans, &res.r, NULL, commit_flags));
 
+		struct btree *new_b = btree_iter_path(trans, iter)->l[level].b;
+		if (new_b != b) {
+			/*
+			 * We were asked to update the key for a node that was
+			 * also modified during the commit (due to triggers),
+			 * and that node was freed:
+			 */
+			BUG_ON(!btree_node_will_make_reachable(new_b));
+			return 0;
+		}
+
 		bch2_btree_node_lock_write_nofail(trans, btree_iter_path(trans, iter), &b->c);
 		bkey_copy(&b->key, new_key);
 		bch2_btree_node_unlock_write(trans, btree_iter_path(trans, iter), b);
@@ -2580,11 +2589,10 @@ void bch2_btree_set_root_for_read(struct bch_fs *c, struct btree *b)
 int bch2_btree_root_alloc_fake_trans(struct btree_trans *trans, enum btree_id id, unsigned level)
 {
 	struct bch_fs *c = trans->c;
-	struct closure cl;
 	struct btree *b;
 	int ret;
 
-	closure_init_stack(&cl);
+	CLASS(closure_stack, cl)();
 
 	do {
 		ret = bch2_btree_cache_cannibalize_lock(trans, &cl);

libbcachefs/btree/interior.h

@@ -5,6 +5,7 @@
 #include "btree/cache.h"
 #include "btree/locking.h"
 #include "btree/update.h"
+#include "data/write_types.h"
 
 #define BTREE_UPDATE_NODES_MAX		((BTREE_MAX_DEPTH - 2) * 2 + GC_MERGE_NODES)
 
@@ -167,9 +168,6 @@ static inline int bch2_foreground_maybe_merge(struct btree_trans *trans,
 
 int bch2_btree_node_get_iter(struct btree_trans *, struct btree_iter *, struct btree *);
 
-int bch2_btree_node_rewrite(struct btree_trans *, struct btree_iter *,
-			    struct btree *, unsigned,
-			    enum bch_trans_commit_flags);
 int bch2_btree_node_rewrite_key(struct btree_trans *,
 				enum btree_id, unsigned,
 				struct bkey_i *,
@@ -177,10 +175,8 @@ int bch2_btree_node_rewrite_key(struct btree_trans *,
 int bch2_btree_node_rewrite_pos(struct btree_trans *,
 				enum btree_id, unsigned,
 				struct bpos, unsigned,
-				enum bch_trans_commit_flags);
+				enum bch_trans_commit_flags,
-int bch2_btree_node_rewrite_key_get_iter(struct btree_trans *,
+				enum bch_write_flags);
-					 struct btree *,
-					 enum bch_trans_commit_flags);
 
 void bch2_btree_node_rewrite_async(struct bch_fs *, struct btree *);
 

libbcachefs/btree/iter.c

@@ -1057,9 +1057,7 @@ retry_all:
 	trans_set_locked(trans, false);
 
 	if (unlikely(trans->memory_allocation_failure)) {
-		struct closure cl;
+		CLASS(closure_stack, cl)();
-
-		closure_init_stack(&cl);
 
 		do {
 			ret = bch2_btree_cache_cannibalize_lock(trans, &cl);

libbcachefs/btree/node_scan.c

@@ -270,10 +270,9 @@ err:
 static int read_btree_nodes(struct find_btree_nodes *f)
 {
 	struct bch_fs *c = container_of(f, struct bch_fs, found_btree_nodes);
-	struct closure cl;
 	int ret = 0;
 
-	closure_init_stack(&cl);
+	CLASS(closure_stack, cl)();
 	CLASS(printbuf, buf)();
 
 	prt_printf(&buf, "scanning for btree nodes on");

libbcachefs/btree/read.c

@@ -1165,11 +1165,10 @@ static int __bch2_btree_root_read(struct btree_trans *trans, enum btree_id id,
 				  const struct bkey_i *k, unsigned level)
 {
 	struct bch_fs *c = trans->c;
-	struct closure cl;
 	struct btree *b;
 	int ret;
 
-	closure_init_stack(&cl);
+	CLASS(closure_stack, cl)();
 
 	do {
 		ret = bch2_btree_cache_cannibalize_lock(trans, &cl);

libbcachefs/data/ec.c

@@ -826,8 +826,7 @@ int bch2_ec_read_extent(struct btree_trans *trans, struct bch_read_bio *rbio,
 	if (ret)
 		return stripe_reconstruct_err(c, orig_k, "-ENOMEM");
 
-	struct closure cl;
+	CLASS(closure_stack, cl)();
-	closure_init_stack(&cl);
 
 	for (unsigned i = 0; i < v->nr_blocks; i++)
 		ec_block_io(c, buf, REQ_OP_READ, i, &cl);

libbcachefs/data/io_misc.c

@@ -48,8 +48,7 @@ int bch2_extent_fallocate(struct btree_trans *trans,
 	struct bkey_buf new __cleanup(bch2_bkey_buf_exit);
 	bch2_bkey_buf_init(&new);
 
-	struct closure cl;
+	CLASS(closure_stack, cl)();
-	closure_init_stack(&cl);
 
 	struct bkey_s_c k = bkey_try(bch2_btree_iter_peek_slot(iter));
 

libbcachefs/data/move.c

@@ -341,7 +341,8 @@ int bch2_move_extent(struct moving_context *ctxt,
 	if (!bkey_is_btree_ptr(k.k))
 		ret = __bch2_move_extent(ctxt, bucket_in_flight, iter, k, opts, data_opts);
 	else if (data_opts.type != BCH_DATA_UPDATE_scrub)
-		ret = bch2_btree_node_rewrite_pos(trans, iter->btree_id, level, k.k->p, data_opts.target, 0);
+		ret = bch2_btree_node_rewrite_pos(trans, iter->btree_id, level, k.k->p,
+						  data_opts.target, 0, data_opts.write_flags);
 	else
 		ret = bch2_btree_node_scrub(trans, iter->btree_id, level, k, data_opts.read_dev);
 

libbcachefs/data/reconcile.c

@@ -23,6 +23,7 @@
 #include "init/progress.h"
 
 #include "fs/inode.h"
+#include "fs/namei.h"
 
 #include "snapshots/subvolume.h"
 
@@ -32,6 +33,61 @@
 #include <linux/kthread.h>
 #include <linux/sched/cputime.h>
 
+#define RECONCILE_WORK_IDS()			\
+	x(none)					\
+	x(hipri)				\
+	x(normal)				\
+	x(pending)
+
+enum reconcile_work_id {
+#define x(t)	RECONCILE_WORK_##t,
+	RECONCILE_WORK_IDS()
+#undef x
+};
+
+#define x(n) #n,
+
+static const char * const reconcile_opts[] = {
+	BCH_REBALANCE_OPTS()
+	NULL
+};
+
+static const char * const reconcile_work_ids[] = {
+	RECONCILE_WORK_IDS()
+	NULL
+};
+
+static const char * const rebalance_scan_strs[] = {
+	RECONCILE_SCAN_TYPES()
+};
+
+#undef x
+
+#define RECONCILE_SCAN_COOKIE_device	32
+#define RECONCILE_SCAN_COOKIE_pending	2
+#define RECONCILE_SCAN_COOKIE_metadata	1
+#define RECONCILE_SCAN_COOKIE_fs	0
+
+static const enum btree_id reconcile_work_btree[] = {
+	[RECONCILE_WORK_hipri]		= BTREE_ID_reconcile_hipri,
+	[RECONCILE_WORK_normal]		= BTREE_ID_reconcile_work,
+	[RECONCILE_WORK_pending]	= BTREE_ID_reconcile_pending,
+};
+
+static enum reconcile_work_id btree_to_reconcile_work_id(enum btree_id btree)
+{
+	switch (btree) {
+	case BTREE_ID_reconcile_hipri:
+		return RECONCILE_WORK_hipri;
+	case BTREE_ID_reconcile_work:
+		return RECONCILE_WORK_normal;
+	case BTREE_ID_reconcile_pending:
+		return RECONCILE_WORK_pending;
+	default:
+		BUG();
+	}
+}
+
 /* bch_extent_reconcile: */
 
 int bch2_extent_reconcile_validate(struct bch_fs *c,
@@ -78,13 +134,6 @@ const struct bch_extent_reconcile *bch2_bkey_reconcile_opts(const struct bch_fs
 	return bch2_bkey_ptrs_reconcile_opts(c, bch2_bkey_ptrs_c(k));
 }
 
-static const char * const reconcile_opts[] = {
-#define x(n) #n,
-	BCH_REBALANCE_OPTS()
-#undef x
-	NULL
-};
-
 void bch2_extent_rebalance_v1_to_text(struct printbuf *out, struct bch_fs *c,
 				      const struct bch_extent_rebalance_v1 *r)
 {
@@ -191,20 +240,20 @@ void bch2_extent_reconcile_to_text(struct printbuf *out, struct bch_fs *c,
 	}
 }
 
-/*
+static enum reconcile_work_id rb_work_id(const struct bch_extent_reconcile *r)
- * XXX: check in bkey_validate that if r->hipri or r->pending are set,
+{
- * r->data_replicas are also set
+	if (!r || !r->need_rb)
- */
+		return RECONCILE_WORK_none;
+	if (r->hipri)
+		return RECONCILE_WORK_hipri;
+	if (!r->pending)
+		return RECONCILE_WORK_normal;
+	return RECONCILE_WORK_pending;
+}
 
 static enum btree_id rb_work_btree(const struct bch_extent_reconcile *r)
 {
-	if (!r || !r->need_rb)
+	return reconcile_work_btree[rb_work_id(r)];
-		return 0;
-	if (r->hipri)
-		return BTREE_ID_reconcile_hipri;
-	if (r->pending)
-		return BTREE_ID_reconcile_pending;
-	return BTREE_ID_reconcile_work;
 }
 
 static inline unsigned rb_accounting_counters(const struct bch_extent_reconcile *r)
@@ -1008,18 +1057,6 @@ int bch2_bkey_get_io_opts(struct btree_trans *trans,
 	return 0;
 }
 
-static const char * const bch2_reconcile_state_strs[] = {
-#define x(t) #t,
-	BCH_REBALANCE_STATES()
-	NULL
-#undef x
-};
-
-#define RECONCILE_SCAN_COOKIE_device	32
-#define RECONCILE_SCAN_COOKIE_pending	2
-#define RECONCILE_SCAN_COOKIE_metadata	1
-#define RECONCILE_SCAN_COOKIE_fs	0
-
 static u64 reconcile_scan_encode(struct reconcile_scan s)
 {
 	switch (s.type) {
@@ -1058,6 +1095,24 @@ static struct reconcile_scan reconcile_scan_decode(struct bch_fs *c, u64 v)
 	return (struct reconcile_scan) { .type = RECONCILE_SCAN_fs};
 }
 
+static void reconcile_scan_to_text(struct printbuf *out,
+				   struct bch_fs *c, struct reconcile_scan s)
+{
+	prt_str(out, rebalance_scan_strs[s.type]);
+	switch (s.type) {
+	case RECONCILE_SCAN_device:
+		prt_str(out, ": ");
+		bch2_prt_member_name(out, c, s.dev);
+		break;
+	case RECONCILE_SCAN_inum:
+		prt_str(out, ": ");
+		bch2_trans_do(c, bch2_inum_snapshot_to_path(trans, s.inum, 0, NULL, out));
+		break;
+	default:
+		break;
+	}
+}
+
 int bch2_set_reconcile_needs_scan_trans(struct btree_trans *trans, struct reconcile_scan s)
 {
 	CLASS(btree_iter, iter)(trans, BTREE_ID_reconcile_scan,
@@ -1324,7 +1379,6 @@ static int __do_reconcile_extent(struct moving_context *ctxt,
 	u32 restart_count = trans->restart_count;
 
 	ctxt->stats = &c->reconcile.work_stats;
-	c->reconcile.state = BCH_REBALANCE_working;
 
 	int ret = bch2_move_extent(ctxt, NULL, snapshot_io_opts,
 				   reconcile_set_data_opts, NULL,
@@ -1334,7 +1388,8 @@ static int __do_reconcile_extent(struct moving_context *ctxt,
 	if (bch2_err_matches(ret, EROFS))
 		return ret;
 	if (bch2_err_matches(ret, BCH_ERR_data_update_fail_no_rw_devs) ||
-	    bch2_err_matches(ret, BCH_ERR_insufficient_devices)) {
+	    bch2_err_matches(ret, BCH_ERR_insufficient_devices) ||
+	    bch2_err_matches(ret, ENOSPC)) {
 		if (rb_work_btree(bch2_bkey_reconcile_opts(c, k)) !=
 		    BTREE_ID_reconcile_pending)
 			try(bch2_trans_relock(trans) ?:
@@ -1343,8 +1398,7 @@ static int __do_reconcile_extent(struct moving_context *ctxt,
 		return 0;
 	}
 	if (ret) {
-		WARN_ONCE(ret != -BCH_ERR_data_update_fail_no_snapshot &&
+		WARN_ONCE(ret != -BCH_ERR_data_update_fail_no_snapshot,
-			  ret != -BCH_ERR_data_update_fail_no_rw_devs,
 			  "unhandled error from move_extent: %s", bch2_err_str(ret));
 		/* skip it and continue */
 	}
@@ -1359,11 +1413,13 @@ static int __do_reconcile_extent(struct moving_context *ctxt,
 
 static int do_reconcile_extent(struct moving_context *ctxt,
 			       struct per_snapshot_io_opts *snapshot_io_opts,
-			       struct bbpos pos)
+			       struct bpos work_pos)
 {
+	struct bbpos data_pos = rb_work_to_data_pos(work_pos);
+
 	struct btree_trans *trans = ctxt->trans;
 
-	CLASS(btree_iter, iter)(trans, pos.btree, pos.pos, BTREE_ITER_all_snapshots);
+	CLASS(btree_iter, iter)(trans, data_pos.btree, data_pos.pos, BTREE_ITER_all_snapshots);
 	struct bkey_s_c k = bkey_try(bch2_btree_iter_peek_slot(&iter));
 
 	return __do_reconcile_extent(ctxt, snapshot_io_opts, &iter, k);
@@ -1550,8 +1606,6 @@ static int do_reconcile_scan(struct moving_context *ctxt,
 	bch2_move_stats_init(&r->scan_stats, "reconcile_scan");
 	ctxt->stats = &r->scan_stats;
 
-	r->state = BCH_REBALANCE_scanning;
-
 	struct reconcile_scan s = reconcile_scan_decode(c, cookie_pos.offset);
 	if (s.type == RECONCILE_SCAN_fs) {
 		try(do_reconcile_scan_fs(ctxt, snapshot_io_opts, false));
@@ -1614,10 +1668,10 @@ static void reconcile_wait(struct bch_fs *c)
 
 	r->wait_iotime_end		= now + (min_member_capacity >> 6);
 
-	if (r->state != BCH_REBALANCE_waiting) {
+	if (r->running) {
 		r->wait_iotime_start	= now;
 		r->wait_wallclock_start	= ktime_get_real_ns();
-		r->state		= BCH_REBALANCE_waiting;
+		r->running		= false;
 	}
 
 	bch2_kthread_io_clock_wait_once(clock, r->wait_iotime_end, MAX_SCHEDULE_TIMEOUT);
@@ -1653,11 +1707,14 @@ static int do_reconcile(struct moving_context *ctxt)
 		BTREE_ID_reconcile_pending,
 	};
 	unsigned i = 0;
-	struct bpos work_pos = POS_MIN;
+
+	r->work_pos = BBPOS(scan_btrees[i], POS_MIN);
 
 	struct bkey_i_cookie pending_cookie;
 	bkey_init(&pending_cookie.k);
 
+	bch2_btree_write_buffer_flush_sync(trans);
+
 	while (!bch2_move_ratelimit(ctxt)) {
 		if (!bch2_reconcile_enabled(c)) {
 			bch2_moving_ctxt_flush_all(ctxt);
@@ -1670,13 +1727,13 @@ static int do_reconcile(struct moving_context *ctxt)
 		if (kick != r->kick) {
 			kick		= r->kick;
 			i		= 0;
-			work_pos	= POS_MIN;
+			r->work_pos	= BBPOS(scan_btrees[i], POS_MIN);
 			work.nr		= 0;
 		}
 
 		bch2_trans_begin(trans);
 
-		struct bkey_s_c k = next_reconcile_entry(trans, &work, scan_btrees[i], &work_pos);
+		struct bkey_s_c k = next_reconcile_entry(trans, &work, r->work_pos.btree, &r->work_pos.pos);
 		ret = bkey_err(k);
 		if (ret)
 			break;
@@ -1685,14 +1742,17 @@ static int do_reconcile(struct moving_context *ctxt)
 			if (++i == ARRAY_SIZE(scan_btrees))
 				break;
 
-			work_pos = POS_MIN;
+			r->work_pos = BBPOS(scan_btrees[i], POS_MIN);
 
-			if (scan_btrees[i] == BTREE_ID_reconcile_pending &&
+			if (r->work_pos.btree == BTREE_ID_reconcile_pending &&
 			    bkey_deleted(&pending_cookie.k))
 				break;
 			continue;
 		}
 
+		r->running = true;
+		r->work_pos.pos = k.k->p;
+
 		if (k.k->type == KEY_TYPE_cookie &&
 		    reconcile_scan_decode(c, k.k->p.offset).type == RECONCILE_SCAN_pending)
 			bkey_reassemble(&pending_cookie.k_i, k);
@@ -1707,8 +1767,7 @@ static int do_reconcile(struct moving_context *ctxt)
 						 bkey_s_c_to_backpointer(k));
 		else
 			ret = lockrestart_do(trans,
-				do_reconcile_extent(ctxt, &snapshot_io_opts,
+				do_reconcile_extent(ctxt, &snapshot_io_opts, k.k->p));
-						    rb_work_to_data_pos(k.k->p)));
 
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
 			ret = 0;
@@ -1718,8 +1777,8 @@ static int do_reconcile(struct moving_context *ctxt)
 		if (ret)
 			break;
 
-		if (scan_btrees[i] == BTREE_ID_reconcile_scan)
+		if (r->work_pos.btree == BTREE_ID_reconcile_scan)
-			work_pos = bpos_successor(work_pos);
+			r->work_pos.pos = bpos_successor(r->work_pos.pos);
 	}
 
 	if (!ret && !bkey_deleted(&pending_cookie.k))
@@ -1788,13 +1847,10 @@ void bch2_reconcile_status_to_text(struct printbuf *out, struct bch_fs *c)
 	}
 
 	prt_newline(out);
+	guard(printbuf_indent_nextline)(out);
 
-	prt_str(out, bch2_reconcile_state_strs[r->state]);
+	if (!r->running) {
-	prt_newline(out);
+		prt_printf(out, "waiting:\n");
-	guard(printbuf_indent)(out);
-
-	switch (r->state) {
-	case BCH_REBALANCE_waiting: {
 		u64 now = atomic64_read(&c->io_clock[WRITE].now);
 
 		prt_printf(out, "io wait duration:\t");
@@ -1808,16 +1864,28 @@ void bch2_reconcile_status_to_text(struct printbuf *out, struct bch_fs *c)
 		prt_printf(out, "duration waited:\t");
 		bch2_pr_time_units(out, ktime_get_real_ns() - r->wait_wallclock_start);
 		prt_newline(out);
-		break;
+	} else {
+		struct bbpos work_pos = r->work_pos;
+		barrier();
+
+		if (work_pos.btree	== BTREE_ID_reconcile_scan &&
+		    work_pos.pos.inode	== 0) {
+			prt_printf(out, "scanning:\n");
+			reconcile_scan_to_text(out, c,
+				reconcile_scan_decode(c, work_pos.pos.offset));
+		} else if (work_pos.btree == BTREE_ID_reconcile_scan) {
+			prt_printf(out, "processing metadata: %s %llu\n",
+				   reconcile_work_ids[work_pos.pos.inode - 1],
+				   work_pos.pos.offset);
+
+		} else {
+			prt_printf(out, "processing data: %s ",
+				   reconcile_work_ids[btree_to_reconcile_work_id(work_pos.btree)]);
+
+			bch2_bbpos_to_text(out, rb_work_to_data_pos(work_pos.pos));
+			prt_newline(out);
+		}
 	}
-	case BCH_REBALANCE_working:
-		bch2_move_stats_to_text(out, &r->work_stats);
-		break;
-	case BCH_REBALANCE_scanning:
-		bch2_move_stats_to_text(out, &r->scan_stats);
-		break;
-	}
-	prt_newline(out);
 
 	struct task_struct *t;
 	scoped_guard(rcu) {

libbcachefs/data/reconcile.h

@@ -114,13 +114,18 @@ int bch2_bkey_set_needs_reconcile(struct btree_trans *,
 				  struct per_snapshot_io_opts *, struct bch_inode_opts *,
 				  struct bkey_i *, enum set_needs_reconcile_ctx, u32);
 
+#define RECONCILE_SCAN_TYPES()		\
+	x(fs)				\
+	x(metadata)			\
+	x(pending)			\
+	x(device)			\
+	x(inum)
+
 struct reconcile_scan {
 	enum reconcile_scan_type {
-		RECONCILE_SCAN_fs,
+#define x(t)	RECONCILE_SCAN_##t,
-		RECONCILE_SCAN_metadata,
+		RECONCILE_SCAN_TYPES()
-		RECONCILE_SCAN_pending,
+#undef x
-		RECONCILE_SCAN_device,
-		RECONCILE_SCAN_inum,
 	}			type;
 
 	union {

libbcachefs/data/reconcile_types.h

@@ -5,26 +5,16 @@
 #include "btree/bbpos_types.h"
 #include "move_types.h"
 
-#define BCH_REBALANCE_STATES()		\
-	x(waiting)			\
-	x(working)			\
-	x(scanning)
-
-enum bch_reconcile_states {
-#define x(t)	BCH_REBALANCE_##t,
-	BCH_REBALANCE_STATES()
-#undef x
-};
-
 struct bch_fs_reconcile {
 	struct task_struct __rcu	*thread;
 	u32				kick;
 
-	enum bch_reconcile_states	state;
+	bool				running;
 	u64				wait_iotime_start;
 	u64				wait_iotime_end;
 	u64				wait_wallclock_start;
 
+	struct bbpos			work_pos;
 	struct bch_move_stats		work_stats;
 
 	struct bbpos			scan_start;

libbcachefs/data/update.c

@@ -470,11 +470,10 @@ int bch2_update_unwritten_extent(struct btree_trans *trans,
 	struct bch_fs *c = update->op.c;
 	struct bkey_i_extent *e;
 	struct write_point *wp;
-	struct closure cl;
 	struct bkey_s_c k;
 	int ret = 0;
 
-	closure_init_stack(&cl);
+	CLASS(closure_stack, cl)();
 	bch2_keylist_init(&update->op.insert_keys, update->op.inline_keys);
 
 	while (bpos_lt(update->op.pos, update->k.k->k.p)) {

libbcachefs/fs/namei.c

@@ -528,9 +528,27 @@ static int __bch2_inum_to_path(struct btree_trans *trans,
 	DARRAY(subvol_inum) inums = {};
 
 	if (!snapshot) {
-		ret = bch2_subvolume_get_snapshot(trans, subvol, &snapshot);
+		if (subvol) {
-		if (ret)
+			ret = bch2_subvolume_get_snapshot(trans, subvol, &snapshot);
-			goto disconnected;
+			if (ret)
+				goto disconnected;
+		} else {
+			struct bkey_s_c k;
+			for_each_btree_key_max_norestart(trans, iter,
+					BTREE_ID_inodes,
+					POS(0, inum),
+					SPOS(0, inum, U32_MAX),
+					BTREE_ITER_all_snapshots, k, ret) {
+				if (bkey_is_inode(k.k)) {
+					snapshot = k.k->p.snapshot;
+					break;
+				}
+			}
+			if (ret)
+				return ret;
+			if (!snapshot)
+				goto disconnected;
+		}
 	}
 
 	while (true) {

libbcachefs/init/fs.c

@@ -1262,7 +1262,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts *opts,
 	return c;
 }
 
-static bool bch2_fs_may_start(struct bch_fs *c, struct printbuf *err)
+static int bch2_fs_may_start(struct bch_fs *c, struct printbuf *err)
 {
 	unsigned flags = 0;
 

libbcachefs/journal/init.c

@@ -155,8 +155,7 @@ static int bch2_set_nr_journal_buckets_loop(struct bch_fs *c, struct bch_dev *ca
 	struct journal_device *ja = &ca->journal;
 	int ret = 0;
 
-	struct closure cl;
+	CLASS(closure_stack, cl)();
-	closure_init_stack(&cl);
 
 	/* don't handle reducing nr of buckets yet: */
 	if (nr < ja->nr)

libbcachefs/sb/members.h

@@ -410,4 +410,19 @@ void bch2_dev_btree_bitmap_mark(struct bch_fs *, struct bkey_s_c);
 int bch2_sb_member_alloc(struct bch_fs *);
 void bch2_sb_members_clean_deleted(struct bch_fs *);
 
+static inline void bch2_prt_member_name(struct printbuf *out, struct bch_fs *c, unsigned idx)
+{
+	if (idx == BCH_SB_MEMBER_INVALID) {
+		prt_str(out, "(none)");
+	} else {
+		guard(rcu)();
+		guard(printbuf_atomic)(out);
+		struct bch_dev *ca = c ? bch2_dev_rcu_noerror(c, idx) : NULL;
+		if (ca)
+			prt_str(out, ca->name);
+		else
+			prt_printf(out, "(invalid device %u)", idx);
+	}
+}
+
 #endif /* _BCACHEFS_SB_MEMBERS_H */

libbcachefs/util/printbuf.h

@@ -299,18 +299,12 @@ DEFINE_GUARD(printbuf_atomic, struct printbuf *,
 	     printbuf_atomic_inc(_T),
 	     printbuf_atomic_dec(_T));
 
-static inline void printbuf_indent_add_2(struct printbuf *out)
-{
-	bch2_printbuf_indent_add(out, 2);
-}
-
-static inline void printbuf_indent_sub_2(struct printbuf *out)
-{
-	bch2_printbuf_indent_sub(out, 2);
-}
-
 DEFINE_GUARD(printbuf_indent, struct printbuf *,
-	     printbuf_indent_add_2(_T),
+	     bch2_printbuf_indent_add(_T, 2),
-	     printbuf_indent_sub_2(_T));
+	     bch2_printbuf_indent_sub(_T, 2));
+
+DEFINE_GUARD(printbuf_indent_nextline, struct printbuf *,
+	     bch2_printbuf_indent_add_nextline(_T, 2),
+	     bch2_printbuf_indent_sub(_T, 2));
 
 #endif /* _BCACHEFS_PRINTBUF_H */

libbcachefs/vendor/closure.h

@@ -336,6 +336,29 @@ static inline void closure_init_stack_release(struct closure *cl)
 #endif
 }
 
+/*
+ * Open coded DEFINE_CLASS(closure_stack, ...)
+ *
+ * We can't use DEFINE_CLASS() because that defines a destructor that destructs
+ * a copy...
+ */
+typedef struct closure class_closure_stack_t;
+
+static inline void class_closure_stack_destructor(struct closure *cl)
+{
+	closure_sync(cl);
+}
+
+static inline struct closure class_closure_stack_constructor(void)
+{
+	return (struct closure) {
+	      .remaining.counter = CLOSURE_REMAINING_INITIALIZER,
+#ifdef CONFIG_DEBUG_CLOSURES
+	      .magic = CLOSURE_MAGIC_STACK,
+#endif
+	};
+}
+
 /**
  * closure_wake_up - wake up all closures on a wait list,
  *		     with memory barrier
@@ -432,9 +455,7 @@ static inline void closure_call(struct closure *cl, closure_fn fn,
 
 #define __closure_wait_event(waitlist, _cond)				\
 do {									\
-	struct closure cl;						\
+	CLASS(closure_stack, cl)();					\
-									\
-	closure_init_stack(&cl);					\
 									\
 	while (1) {							\
 		bch2_closure_wait(waitlist, &cl);			\
@@ -443,7 +464,6 @@ do {									\
 		closure_sync(&cl);					\
 	}								\
 	closure_wake_up(waitlist);					\
-	closure_sync(&cl);						\
 } while (0)
 
 #define closure_wait_event(waitlist, _cond)				\
@@ -454,11 +474,9 @@ do {									\
 
 #define __closure_wait_event_timeout(waitlist, _cond, _until)		\
 ({									\
-	struct closure cl;						\
+	CLASS(closure_stack, cl)();					\
 	long _t;							\
 									\
-	closure_init_stack(&cl);					\
-									\
 	while (1) {							\
 		bch2_closure_wait(waitlist, &cl);			\
 		if (_cond) {						\
@@ -471,7 +489,6 @@ do {									\
 		closure_sync_timeout(&cl, _t);				\
 	}								\
 	closure_wake_up(waitlist);					\
-	closure_sync(&cl);						\
 	_t;								\
 })
 

libbcachefs/vfs/buffered.c

@@ -711,8 +711,7 @@ int bch2_writepages(struct address_space *mapping, struct writeback_control *wbc
 
 	blk_start_plug(&w->plug);
 
-	struct closure cl;
+	CLASS(closure_stack, cl)();
-	closure_init_stack(&cl);
 
 	struct folio *folio = NULL;
 	int ret = 0;

libbcachefs/vfs/io.c

@@ -101,12 +101,8 @@ void bch2_inode_flush_nocow_writes_async(struct bch_fs *c,
 static int bch2_inode_flush_nocow_writes(struct bch_fs *c,
 					 struct bch_inode_info *inode)
 {
-	struct closure cl;
+	CLASS(closure_stack, cl)();
-
-	closure_init_stack(&cl);
 	bch2_inode_flush_nocow_writes_async(c, inode, &cl);
-	closure_sync(&cl);
-
 	return 0;
 }