bcachefs: Bail out early on alloc_nowait data updates

If a data update doesn't want to block on allocations (promotes, self
healing on read error) - check if the allocation would fail before
kicking off the data update and calling into the write path.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

4 files changed, 59 insertions, 18 deletions

diff --git a/fs/bcachefs/alloc_foreground.c b/fs/bcachefs/alloc_foreground.c
index 1a539e7bedc8..1759c15a7745 100644
--- a/fs/bcachefs/alloc_foreground.c
+++ b/fs/bcachefs/alloc_foreground.c
@@ -179,23 +179,6 @@ static void open_bucket_free_unused(struct bch_fs *c, struct open_bucket *ob)
 	closure_wake_up(&c->freelist_wait);
 }
 
-static inline unsigned open_buckets_reserved(enum bch_watermark watermark)
-{
-	switch (watermark) {
-	case BCH_WATERMARK_interior_updates:
-		return 0;
-	case BCH_WATERMARK_reclaim:
-		return OPEN_BUCKETS_COUNT / 6;
-	case BCH_WATERMARK_btree:
-	case BCH_WATERMARK_btree_copygc:
-		return OPEN_BUCKETS_COUNT / 4;
-	case BCH_WATERMARK_copygc:
-		return OPEN_BUCKETS_COUNT / 3;
-	default:
-		return OPEN_BUCKETS_COUNT / 2;
-	}
-}
-
 static inline bool may_alloc_bucket(struct bch_fs *c,
 				    struct bpos bucket,
 				    struct bucket_alloc_state *s)
@@ -239,7 +222,7 @@ static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *
 
 	spin_lock(&c->freelist_lock);
 
-	if (unlikely(c->open_buckets_nr_free <= open_buckets_reserved(watermark))) {
+	if (unlikely(c->open_buckets_nr_free <= bch2_open_buckets_reserved(watermark))) {
 		if (cl)
 			closure_wait(&c->open_buckets_wait, cl);
 
diff --git a/fs/bcachefs/alloc_foreground.h b/fs/bcachefs/alloc_foreground.h
index f25481a0d1a0..baf5dc163c8a 100644
--- a/fs/bcachefs/alloc_foreground.h
+++ b/fs/bcachefs/alloc_foreground.h
@@ -33,6 +33,23 @@ static inline struct bch_dev *ob_dev(struct bch_fs *c, struct open_bucket *ob)
 	return bch2_dev_have_ref(c, ob->dev);
 }
 
+static inline unsigned bch2_open_buckets_reserved(enum bch_watermark watermark)
+{
+	switch (watermark) {
+	case BCH_WATERMARK_interior_updates:
+		return 0;
+	case BCH_WATERMARK_reclaim:
+		return OPEN_BUCKETS_COUNT / 6;
+	case BCH_WATERMARK_btree:
+	case BCH_WATERMARK_btree_copygc:
+		return OPEN_BUCKETS_COUNT / 4;
+	case BCH_WATERMARK_copygc:
+		return OPEN_BUCKETS_COUNT / 3;
+	default:
+		return OPEN_BUCKETS_COUNT / 2;
+	}
+}
+
 struct open_bucket *bch2_bucket_alloc(struct bch_fs *, struct bch_dev *,
 				      enum bch_watermark, enum bch_data_type,
 				      struct closure *);
diff --git a/fs/bcachefs/data_update.c b/fs/bcachefs/data_update.c
index 3e8ad94dca59..ec63dd494c80 100644
--- a/fs/bcachefs/data_update.c
+++ b/fs/bcachefs/data_update.c
@@ -639,6 +639,40 @@ int bch2_extent_drop_ptrs(struct btree_trans *trans,
 		bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 }
 
+static bool can_allocate_without_blocking(struct bch_fs *c,
+					  struct data_update *m)
+{
+	if (unlikely(c->open_buckets_nr_free <= bch2_open_buckets_reserved(m->op.watermark)))
+		return false;
+
+	unsigned target = m->op.flags & BCH_WRITE_only_specified_devs
+		? m->op.target
+		: 0;
+	struct bch_devs_mask devs = target_rw_devs(c, BCH_DATA_user, target);
+
+	darray_for_each(m->op.devs_have, i)
+		__clear_bit(*i, devs.d);
+
+	rcu_read_lock();
+	unsigned nr_replicas = 0, i;
+	for_each_set_bit(i, devs.d, BCH_SB_MEMBERS_MAX) {
+		struct bch_dev *ca = bch2_dev_rcu(c, i);
+
+		struct bch_dev_usage usage;
+		bch2_dev_usage_read_fast(ca, &usage);
+
+		if (!dev_buckets_free(ca, usage, m->op.watermark))
+			continue;
+
+		nr_replicas += ca->mi.durability;
+		if (nr_replicas >= m->op.nr_replicas)
+			break;
+	}
+	rcu_read_unlock();
+
+	return nr_replicas >= m->op.nr_replicas;
+}
+
 int bch2_data_update_init(struct btree_trans *trans,
 			  struct btree_iter *iter,
 			  struct moving_context *ctxt,
@@ -759,6 +793,12 @@ int bch2_data_update_init(struct btree_trans *trans,
 		goto out_bkey_buf_exit;
 	}
 
+	if ((m->op.flags & BCH_WRITE_alloc_nowait) &&
+	    !can_allocate_without_blocking(c, m)) {
+		ret = -BCH_ERR_data_update_done_would_block;
+		goto out_bkey_buf_exit;
+	}
+
 	if (reserve_sectors) {
 		ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
 				m->data_opts.extra_replicas
diff --git a/fs/bcachefs/errcode.h b/fs/bcachefs/errcode.h
index 82f950ea1c26..1e8f65f95d60 100644
--- a/fs/bcachefs/errcode.h
+++ b/fs/bcachefs/errcode.h
@@ -181,6 +181,7 @@
 	x(EINVAL,			not_in_recovery)			\
 	x(EINVAL,			cannot_rewind_recovery)			\
 	x(0,				data_update_done)			\
+	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)	\
 	x(BCH_ERR_data_update_done,	data_update_done_no_snapshot)		\