bcachefs: bch2_journal_space_available improvements

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

fs/bcachefs/journal.c

@@ -1160,7 +1160,7 @@ ssize_t bch2_journal_print_debug(struct journal *j, char *buf)
 		       "\tdirty_idx\t\t%u (seq %llu)\n"
 		       "\tcur_idx\t\t%u (seq %llu)\n",
 		       iter, ja->nr,
-		       bch2_journal_dev_buckets_available(j, ja),
+		       bch2_journal_dev_buckets_available(j, ja, journal_space_discarded),
 		       ja->sectors_free,
 		       ja->discard_idx,
 		       ja->dirty_idx_ondisk,	ja->bucket_seq[ja->dirty_idx_ondisk],

fs/bcachefs/journal_io.c

@@ -970,7 +970,8 @@ static int journal_write_alloc(struct journal *j, struct journal_buf *w,
 
 		if (sectors > ja->sectors_free &&
 		    sectors <= ca->mi.bucket_size &&
-		    bch2_journal_dev_buckets_available(j, ja)) {
+		    bch2_journal_dev_buckets_available(j, ja,
+					journal_space_discarded)) {
 			ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
 			ja->sectors_free = ca->mi.bucket_size;
 		}

fs/bcachefs/journal_reclaim.c

@@ -9,12 +9,28 @@
 
 /* Free space calculations: */
 
+static unsigned journal_space_from(struct journal_device *ja,
+				   enum journal_space_from from)
+{
+	switch (from) {
+	case journal_space_discarded:
+		return ja->discard_idx;
+	case journal_space_clean_ondisk:
+		return ja->dirty_idx_ondisk;
+	case journal_space_clean:
+		return ja->dirty_idx;
+	default:
+		BUG();
+	}
+}
+
 unsigned bch2_journal_dev_buckets_available(struct journal *j,
-					    struct journal_device *ja)
+					    struct journal_device *ja,
+					    enum journal_space_from from)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
-	unsigned next = (ja->cur_idx + 1) % ja->nr;
-	unsigned available = (ja->discard_idx + ja->nr - next) % ja->nr;
+	unsigned available = (journal_space_from(ja, from) -
+			      ja->cur_idx - 1 + ja->nr) % ja->nr;
 
 	/*
 	 * Allocator startup needs some journal space before we can do journal
@@ -33,53 +49,22 @@ unsigned bch2_journal_dev_buckets_available(struct journal *j,
 	return available;
 }
 
-void bch2_journal_space_available(struct journal *j)
+static struct journal_space {
+	unsigned	next_entry;
+	unsigned	remaining;
+} __journal_space_available(struct journal *j, unsigned nr_devs_want,
+			    enum journal_space_from from)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct bch_dev *ca;
 	unsigned sectors_next_entry	= UINT_MAX;
 	unsigned sectors_total		= UINT_MAX;
-	unsigned max_entry_size		= min(j->buf[0].buf_size >> 9,
-					      j->buf[1].buf_size >> 9);
-	unsigned i, nr_online = 0, nr_devs = 0;
+	unsigned i, nr_devs = 0;
 	unsigned unwritten_sectors = j->reservations.prev_buf_unwritten
 		? journal_prev_buf(j)->sectors
 		: 0;
-	bool can_discard = false;
-	int ret = 0;
-
-	lockdep_assert_held(&j->lock);
 
 	rcu_read_lock();
-	for_each_member_device_rcu(ca, c, i,
-				   &c->rw_devs[BCH_DATA_JOURNAL]) {
-		struct journal_device *ja = &ca->journal;
-
-		if (!ja->nr)
-			continue;
-
-		while (ja->dirty_idx != ja->cur_idx &&
-		       ja->bucket_seq[ja->dirty_idx] < journal_last_seq(j))
-			ja->dirty_idx = (ja->dirty_idx + 1) % ja->nr;
-
-		while (ja->dirty_idx_ondisk != ja->dirty_idx &&
-		       ja->bucket_seq[ja->dirty_idx_ondisk] < j->last_seq_ondisk)
-			ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + 1) % ja->nr;
-
-		if (ja->discard_idx != ja->dirty_idx_ondisk)
-			can_discard = true;
-
-		nr_online++;
-	}
-
-	j->can_discard = can_discard;
-
-	if (nr_online < c->opts.metadata_replicas_required) {
-		ret = -EROFS;
-		sectors_next_entry = 0;
-		goto out;
-	}
-
 	for_each_member_device_rcu(ca, c, i,
 				   &c->rw_devs[BCH_DATA_JOURNAL]) {
 		struct journal_device *ja = &ca->journal;
@@ -88,7 +73,7 @@ void bch2_journal_space_available(struct journal *j)
 		if (!ja->nr)
 			continue;
 
-		buckets_this_device = bch2_journal_dev_buckets_available(j, ja);
+		buckets_this_device = bch2_journal_dev_buckets_available(j, ja, from);
 		sectors_this_device = ja->sectors_free;
 
 		/*
@@ -121,24 +106,78 @@ void bch2_journal_space_available(struct journal *j)
 			buckets_this_device * ca->mi.bucket_size +
 			sectors_this_device);
 
-		max_entry_size = min_t(unsigned, max_entry_size,
-				       ca->mi.bucket_size);
-
 		nr_devs++;
 	}
+	rcu_read_unlock();
 
-	if (!sectors_next_entry ||
-	    nr_devs < min_t(unsigned, nr_online, c->opts.metadata_replicas)) {
-		ret = -ENOSPC;
-		sectors_next_entry = 0;
-	} else if (!fifo_free(&j->pin)) {
-		ret = -ENOSPC;
-		sectors_next_entry = 0;
+	if (nr_devs < nr_devs_want)
+		return (struct journal_space) { 0, 0 };
+
+	return (struct journal_space) {
+		.next_entry	= sectors_next_entry,
+		.remaining	= max_t(int, 0, sectors_total - sectors_next_entry),
+	};
+}
+
+void bch2_journal_space_available(struct journal *j)
+{
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	struct bch_dev *ca;
+	struct journal_space discarded, clean_ondisk, clean;
+	unsigned max_entry_size		= min(j->buf[0].buf_size >> 9,
+					      j->buf[1].buf_size >> 9);
+	unsigned i, nr_online = 0, nr_devs_want;
+	bool can_discard = false;
+	int ret = 0;
+
+	lockdep_assert_held(&j->lock);
+
+	rcu_read_lock();
+	for_each_member_device_rcu(ca, c, i,
+				   &c->rw_devs[BCH_DATA_JOURNAL]) {
+		struct journal_device *ja = &ca->journal;
+
+		if (!ja->nr)
+			continue;
+
+		while (ja->dirty_idx != ja->cur_idx &&
+		       ja->bucket_seq[ja->dirty_idx] < journal_last_seq(j))
+			ja->dirty_idx = (ja->dirty_idx + 1) % ja->nr;
+
+		while (ja->dirty_idx_ondisk != ja->dirty_idx &&
+		       ja->bucket_seq[ja->dirty_idx_ondisk] < j->last_seq_ondisk)
+			ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + 1) % ja->nr;
+
+		if (ja->discard_idx != ja->dirty_idx_ondisk)
+			can_discard = true;
+
+		max_entry_size = min_t(unsigned, max_entry_size, ca->mi.bucket_size);
+		nr_online++;
 	}
-out:
 	rcu_read_unlock();
 
-	j->cur_entry_sectors	= sectors_next_entry;
+	j->can_discard = can_discard;
+
+	if (nr_online < c->opts.metadata_replicas_required) {
+		ret = -EROFS;
+		goto out;
+	}
+
+	if (!fifo_free(&j->pin)) {
+		ret = -ENOSPC;
+		goto out;
+	}
+
+	nr_devs_want = min_t(unsigned, nr_online, c->opts.metadata_replicas);
+
+	discarded	= __journal_space_available(j, nr_devs_want, journal_space_discarded);
+	clean_ondisk	= __journal_space_available(j, nr_devs_want, journal_space_clean_ondisk);
+	clean		= __journal_space_available(j, nr_devs_want, journal_space_clean);
+
+	if (!discarded.next_entry)
+		ret = -ENOSPC;
+out:
+	j->cur_entry_sectors	= !ret ? discarded.next_entry : 0;
 	j->cur_entry_error	= ret;
 
 	if (!ret)

fs/bcachefs/journal_reclaim.h

@@ -4,8 +4,15 @@
 
 #define JOURNAL_PIN	(32 * 1024)
 
+enum journal_space_from {
+	journal_space_discarded,
+	journal_space_clean_ondisk,
+	journal_space_clean,
+};
+
 unsigned bch2_journal_dev_buckets_available(struct journal *,
-					    struct journal_device *);
+					    struct journal_device *,
+					    enum journal_space_from);
 void bch2_journal_space_available(struct journal *);
 
 static inline bool journal_pin_active(struct journal_entry_pin *pin)