1 files changed, 221 insertions, 95 deletions

diff --git a/fs/bcachefs/journal_reclaim.c b/fs/bcachefs/journal_reclaim.c
index 820d25e19e5f..cd6201741c59 100644
--- a/fs/bcachefs/journal_reclaim.c
+++ b/fs/bcachefs/journal_reclaim.c
@@ -17,6 +17,8 @@
 #include <linux/kthread.h>
 #include <linux/sched/mm.h>
 
+static bool __should_discard_bucket(struct journal *, struct journal_device *);
+
 /* Free space calculations: */
 
 static unsigned journal_space_from(struct journal_device *ja,
@@ -38,6 +40,9 @@ unsigned bch2_journal_dev_buckets_available(struct journal *j,
 					    struct journal_device *ja,
 					    enum journal_space_from from)
 {
+	if (!ja->nr)
+		return 0;
+
 	unsigned available = (journal_space_from(ja, from) -
 			      ja->cur_idx - 1 + ja->nr) % ja->nr;
 
@@ -62,14 +67,13 @@ void bch2_journal_set_watermark(struct journal *j)
 		? BCH_WATERMARK_reclaim
 		: BCH_WATERMARK_stripe;
 
-	if (track_event_change(&c->times[BCH_TIME_blocked_journal_low_on_space],
-			       &j->low_on_space_start, low_on_space) ||
-	    track_event_change(&c->times[BCH_TIME_blocked_journal_low_on_pin],
-			       &j->low_on_pin_start, low_on_pin) ||
-	    track_event_change(&c->times[BCH_TIME_blocked_write_buffer_full],
-			       &j->write_buffer_full_start, low_on_wb))
+	if (track_event_change(&c->times[BCH_TIME_blocked_journal_low_on_space], low_on_space) ||
+	    track_event_change(&c->times[BCH_TIME_blocked_journal_low_on_pin], low_on_pin) ||
+	    track_event_change(&c->times[BCH_TIME_blocked_write_buffer_full], low_on_wb))
 		trace_and_count(c, journal_full, c);
 
+	mod_bit(JOURNAL_space_low, &j->flags, low_on_space || low_on_pin);
+
 	swap(watermark, j->watermark);
 	if (watermark > j->watermark)
 		journal_wake(j);
@@ -79,18 +83,20 @@ static struct journal_space
 journal_dev_space_available(struct journal *j, struct bch_dev *ca,
 			    enum journal_space_from from)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_device *ja = &ca->journal;
 	unsigned sectors, buckets, unwritten;
+	unsigned bucket_size_aligned = round_down(ca->mi.bucket_size, block_sectors(c));
 	u64 seq;
 
 	if (from == journal_space_total)
 		return (struct journal_space) {
-			.next_entry	= ca->mi.bucket_size,
-			.total		= ca->mi.bucket_size * ja->nr,
+			.next_entry	= bucket_size_aligned,
+			.total		= bucket_size_aligned * ja->nr,
 		};
 
 	buckets = bch2_journal_dev_buckets_available(j, ja, from);
-	sectors = ja->sectors_free;
+	sectors = round_down(ja->sectors_free, block_sectors(c));
 
 	/*
 	 * We that we don't allocate the space for a journal entry
@@ -105,7 +111,7 @@ journal_dev_space_available(struct journal *j, struct bch_dev *ca,
 			continue;
 
 		/* entry won't fit on this device, skip: */
-		if (unwritten > ca->mi.bucket_size)
+		if (unwritten > bucket_size_aligned)
 			continue;
 
 		if (unwritten >= sectors) {
@@ -115,7 +121,7 @@ journal_dev_space_available(struct journal *j, struct bch_dev *ca,
 			}
 
 			buckets--;
-			sectors = ca->mi.bucket_size;
+			sectors = bucket_size_aligned;
 		}
 
 		sectors -= unwritten;
@@ -123,12 +129,12 @@ journal_dev_space_available(struct journal *j, struct bch_dev *ca,
 
 	if (sectors < ca->mi.bucket_size && buckets) {
 		buckets--;
-		sectors = ca->mi.bucket_size;
+		sectors = bucket_size_aligned;
 	}
 
 	return (struct journal_space) {
 		.next_entry	= sectors,
-		.total		= sectors + buckets * ca->mi.bucket_size,
+		.total		= sectors + buckets * bucket_size_aligned,
 	};
 }
 
@@ -138,14 +144,17 @@ static struct journal_space __journal_space_available(struct journal *j, unsigne
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	unsigned pos, nr_devs = 0;
 	struct journal_space space, dev_space[BCH_SB_MEMBERS_MAX];
+	unsigned min_bucket_size = U32_MAX;
 
 	BUG_ON(nr_devs_want > ARRAY_SIZE(dev_space));
 
-	rcu_read_lock();
 	for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) {
-		if (!ca->journal.nr)
+		if (!ca->journal.nr ||
+		    !ca->mi.durability)
 			continue;
 
+		min_bucket_size = min(min_bucket_size, ca->mi.bucket_size);
+
 		space = journal_dev_space_available(j, ca, from);
 		if (!space.next_entry)
 			continue;
@@ -156,7 +165,6 @@ static struct journal_space __journal_space_available(struct journal *j, unsigne
 
 		array_insert_item(dev_space, nr_devs, pos, space);
 	}
-	rcu_read_unlock();
 
 	if (nr_devs < nr_devs_want)
 		return (struct journal_space) { 0, 0 };
@@ -165,7 +173,9 @@ static struct journal_space __journal_space_available(struct journal *j, unsigne
 	 * We sorted largest to smallest, and we want the smallest out of the
 	 * @nr_devs_want largest devices:
 	 */
-	return dev_space[nr_devs_want - 1];
+	space = dev_space[nr_devs_want - 1];
+	space.next_entry = min(space.next_entry, min_bucket_size);
+	return space;
 }
 
 void bch2_journal_space_available(struct journal *j)
@@ -179,8 +189,8 @@ void bch2_journal_space_available(struct journal *j)
 	int ret = 0;
 
 	lockdep_assert_held(&j->lock);
+	guard(rcu)();
 
-	rcu_read_lock();
 	for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) {
 		struct journal_device *ja = &ca->journal;
 
@@ -195,18 +205,28 @@ void bch2_journal_space_available(struct journal *j)
 		       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;
+		can_discard |= __should_discard_bucket(j, ja);
 
 		max_entry_size = min_t(unsigned, max_entry_size, ca->mi.bucket_size);
 		nr_online++;
 	}
-	rcu_read_unlock();
 
 	j->can_discard = can_discard;
 
-	if (nr_online < c->opts.metadata_replicas_required) {
-		ret = JOURNAL_ERR_insufficient_devices;
+	if (nr_online < metadata_replicas_required(c)) {
+		if (!(c->sb.features & BIT_ULL(BCH_FEATURE_small_image))) {
+			struct printbuf buf = PRINTBUF;
+			buf.atomic++;
+			prt_printf(&buf, "insufficient writeable journal devices available: have %u, need %u\n"
+				   "rw journal devs:", nr_online, metadata_replicas_required(c));
+
+			for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal])
+				prt_printf(&buf, " %s", ca->name);
+
+			bch_err(c, "%s", buf.buf);
+			printbuf_exit(&buf);
+		}
+		ret = bch_err_throw(c, insufficient_journal_devices);
 		goto out;
 	}
 
@@ -220,19 +240,21 @@ void bch2_journal_space_available(struct journal *j)
 	total		= j->space[journal_space_total].total;
 
 	if (!j->space[journal_space_discarded].next_entry)
-		ret = JOURNAL_ERR_journal_full;
+		ret = bch_err_throw(c, journal_full);
 
 	if ((j->space[journal_space_clean_ondisk].next_entry <
 	     j->space[journal_space_clean_ondisk].total) &&
 	    (clean - clean_ondisk <= total / 8) &&
 	    (clean_ondisk * 2 > clean))
-		set_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags);
+		set_bit(JOURNAL_may_skip_flush, &j->flags);
 	else
-		clear_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags);
+		clear_bit(JOURNAL_may_skip_flush, &j->flags);
 
 	bch2_journal_set_watermark(j);
 out:
-	j->cur_entry_sectors	= !ret ? j->space[journal_space_discarded].next_entry : 0;
+	j->cur_entry_sectors	= !ret
+		? j->space[journal_space_discarded].next_entry
+		: 0;
 	j->cur_entry_error	= ret;
 
 	if (!ret)
@@ -241,12 +263,19 @@ out:
 
 /* Discards - last part of journal reclaim: */
 
-static bool should_discard_bucket(struct journal *j, struct journal_device *ja)
+static bool __should_discard_bucket(struct journal *j, struct journal_device *ja)
 {
-	bool ret;
+	unsigned min_free = max(4, ja->nr / 8);
+
+	return bch2_journal_dev_buckets_available(j, ja, journal_space_discarded) <
+		min_free &&
+		ja->discard_idx != ja->dirty_idx_ondisk;
+}
 
+static bool should_discard_bucket(struct journal *j, struct journal_device *ja)
+{
 	spin_lock(&j->lock);
-	ret = ja->discard_idx != ja->dirty_idx_ondisk;
+	bool ret = __should_discard_bucket(j, ja);
 	spin_unlock(&j->lock);
 
 	return ret;
@@ -262,12 +291,12 @@ void bch2_journal_do_discards(struct journal *j)
 
 	mutex_lock(&j->discard_lock);
 
-	for_each_rw_member(c, ca) {
+	for_each_rw_member(c, ca, BCH_DEV_WRITE_REF_journal_do_discards) {
 		struct journal_device *ja = &ca->journal;
 
 		while (should_discard_bucket(j, ja)) {
 			if (!c->opts.nochanges &&
-			    ca->mi.discard &&
+			    bch2_discard_opt_enabled(c, ca) &&
 			    bdev_max_discard_sectors(ca->disk_sb.bdev))
 				blkdev_issue_discard(ca->disk_sb.bdev,
 					bucket_to_sector(ca,
@@ -307,8 +336,10 @@ void bch2_journal_reclaim_fast(struct journal *j)
 		popped = true;
 	}
 
-	if (popped)
+	if (popped) {
 		bch2_journal_space_available(j);
+		__closure_wake_up(&j->reclaim_flush_wait);
+	}
 }
 
 bool __bch2_journal_pin_put(struct journal *j, u64 seq)
@@ -342,6 +373,9 @@ static inline bool __journal_pin_drop(struct journal *j,
 	pin->seq = 0;
 	list_del_init(&pin->list);
 
+	if (j->reclaim_flush_wait.list.first)
+		__closure_wake_up(&j->reclaim_flush_wait);
+
 	/*
 	 * Unpinning a journal entry may make journal_next_bucket() succeed, if
 	 * writing a new last_seq will now make another bucket available:
@@ -359,15 +393,19 @@ void bch2_journal_pin_drop(struct journal *j,
 	spin_unlock(&j->lock);
 }
 
-static enum journal_pin_type journal_pin_type(journal_pin_flush_fn fn)
+static enum journal_pin_type journal_pin_type(struct journal_entry_pin *pin,
+					      journal_pin_flush_fn fn)
 {
 	if (fn == bch2_btree_node_flush0 ||
-	    fn == bch2_btree_node_flush1)
-		return JOURNAL_PIN_btree;
-	else if (fn == bch2_btree_key_cache_journal_flush)
-		return JOURNAL_PIN_key_cache;
+	    fn == bch2_btree_node_flush1) {
+		unsigned idx = fn == bch2_btree_node_flush1;
+		struct btree *b = container_of(pin, struct btree, writes[idx].journal);
+
+		return JOURNAL_PIN_TYPE_btree0 - b->c.level;
+	} else if (fn == bch2_btree_key_cache_journal_flush)
+		return JOURNAL_PIN_TYPE_key_cache;
 	else
-		return JOURNAL_PIN_other;
+		return JOURNAL_PIN_TYPE_other;
 }
 
 static inline void bch2_journal_pin_set_locked(struct journal *j, u64 seq,
@@ -386,7 +424,12 @@ static inline void bch2_journal_pin_set_locked(struct journal *j, u64 seq,
 	atomic_inc(&pin_list->count);
 	pin->seq	= seq;
 	pin->flush	= flush_fn;
-	list_add(&pin->list, &pin_list->list[type]);
+
+	if (list_empty(&pin_list->unflushed[type]) &&
+	    j->reclaim_flush_wait.list.first)
+		__closure_wake_up(&j->reclaim_flush_wait);
+
+	list_add(&pin->list, &pin_list->unflushed[type]);
 }
 
 void bch2_journal_pin_copy(struct journal *j,
@@ -394,8 +437,6 @@ void bch2_journal_pin_copy(struct journal *j,
 			   struct journal_entry_pin *src,
 			   journal_pin_flush_fn flush_fn)
 {
-	bool reclaim;
-
 	spin_lock(&j->lock);
 
 	u64 seq = READ_ONCE(src->seq);
@@ -411,44 +452,44 @@ void bch2_journal_pin_copy(struct journal *j,
 		return;
 	}
 
-	reclaim = __journal_pin_drop(j, dst);
+	bool reclaim = __journal_pin_drop(j, dst);
 
-	bch2_journal_pin_set_locked(j, seq, dst, flush_fn, journal_pin_type(flush_fn));
+	bch2_journal_pin_set_locked(j, seq, dst, flush_fn, journal_pin_type(dst, flush_fn));
 
 	if (reclaim)
 		bch2_journal_reclaim_fast(j);
-	spin_unlock(&j->lock);
 
 	/*
 	 * If the journal is currently full,  we might want to call flush_fn
 	 * immediately:
 	 */
-	journal_wake(j);
+	if (seq == journal_last_seq(j))
+		journal_wake(j);
+	spin_unlock(&j->lock);
 }
 
 void bch2_journal_pin_set(struct journal *j, u64 seq,
 			  struct journal_entry_pin *pin,
 			  journal_pin_flush_fn flush_fn)
 {
-	bool reclaim;
-
 	spin_lock(&j->lock);
 
 	BUG_ON(seq < journal_last_seq(j));
 
-	reclaim = __journal_pin_drop(j, pin);
+	bool reclaim = __journal_pin_drop(j, pin);
 
-	bch2_journal_pin_set_locked(j, seq, pin, flush_fn, journal_pin_type(flush_fn));
+	bch2_journal_pin_set_locked(j, seq, pin, flush_fn, journal_pin_type(pin, flush_fn));
 
 	if (reclaim)
 		bch2_journal_reclaim_fast(j);
-	spin_unlock(&j->lock);
-
 	/*
 	 * If the journal is currently full,  we might want to call flush_fn
 	 * immediately:
 	 */
-	journal_wake(j);
+	if (seq == journal_last_seq(j))
+		journal_wake(j);
+
+	spin_unlock(&j->lock);
 }
 
 /**
@@ -481,16 +522,15 @@ journal_get_next_pin(struct journal *j,
 {
 	struct journal_entry_pin_list *pin_list;
 	struct journal_entry_pin *ret = NULL;
-	unsigned i;
 
 	fifo_for_each_entry_ptr(pin_list, &j->pin, *seq) {
 		if (*seq > seq_to_flush && !allowed_above_seq)
 			break;
 
-		for (i = 0; i < JOURNAL_PIN_NR; i++)
-			if ((((1U << i) & allowed_below_seq) && *seq <= seq_to_flush) ||
-			    ((1U << i) & allowed_above_seq)) {
-				ret = list_first_entry_or_null(&pin_list->list[i],
+		for (unsigned i = 0; i < JOURNAL_PIN_TYPE_NR; i++)
+			if (((BIT(i) & allowed_below_seq) && *seq <= seq_to_flush) ||
+			    (BIT(i) & allowed_above_seq)) {
+				ret = list_first_entry_or_null(&pin_list->unflushed[i],
 					struct journal_entry_pin, list);
 				if (ret)
 					return ret;
@@ -526,8 +566,8 @@ static size_t journal_flush_pins(struct journal *j,
 		}
 
 		if (min_key_cache) {
-			allowed_above |= 1U << JOURNAL_PIN_key_cache;
-			allowed_below |= 1U << JOURNAL_PIN_key_cache;
+			allowed_above |= BIT(JOURNAL_PIN_TYPE_key_cache);
+			allowed_below |= BIT(JOURNAL_PIN_TYPE_key_cache);
 		}
 
 		cond_resched();
@@ -535,7 +575,9 @@ static size_t journal_flush_pins(struct journal *j,
 		j->last_flushed = jiffies;
 
 		spin_lock(&j->lock);
-		pin = journal_get_next_pin(j, seq_to_flush, allowed_below, allowed_above, &seq);
+		pin = journal_get_next_pin(j, seq_to_flush,
+					   allowed_below,
+					   allowed_above, &seq);
 		if (pin) {
 			BUG_ON(j->flush_in_progress);
 			j->flush_in_progress = pin;
@@ -558,7 +600,7 @@ static size_t journal_flush_pins(struct journal *j,
 		spin_lock(&j->lock);
 		/* Pin might have been dropped or rearmed: */
 		if (likely(!err && !j->flush_in_progress_dropped))
-			list_move(&pin->list, &journal_seq_pin(j, seq)->flushed);
+			list_move(&pin->list, &journal_seq_pin(j, seq)->flushed[journal_pin_type(pin, flush_fn)]);
 		j->flush_in_progress = NULL;
 		j->flush_in_progress_dropped = false;
 		spin_unlock(&j->lock);
@@ -579,9 +621,10 @@ static u64 journal_seq_to_flush(struct journal *j)
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	u64 seq_to_flush = 0;
 
-	spin_lock(&j->lock);
+	guard(spinlock)(&j->lock);
+	guard(rcu)();
 
-	for_each_rw_member(c, ca) {
+	for_each_rw_member_rcu(c, ca) {
 		struct journal_device *ja = &ca->journal;
 		unsigned nr_buckets, bucket_to_flush;
 
@@ -591,20 +634,15 @@ static u64 journal_seq_to_flush(struct journal *j)
 		/* Try to keep the journal at most half full: */
 		nr_buckets = ja->nr / 2;
 
-		nr_buckets = min(nr_buckets, ja->nr);
-
 		bucket_to_flush = (ja->cur_idx + nr_buckets) % ja->nr;
 		seq_to_flush = max(seq_to_flush,
 				   ja->bucket_seq[bucket_to_flush]);
 	}
 
 	/* Also flush if the pin fifo is more than half full */
-	seq_to_flush = max_t(s64, seq_to_flush,
-			     (s64) journal_cur_seq(j) -
-			     (j->pin.size >> 1));
-	spin_unlock(&j->lock);
-
-	return seq_to_flush;
+	return max_t(s64, seq_to_flush,
+		     (s64) journal_cur_seq(j) -
+		     (j->pin.size >> 1));
 }
 
 /**
@@ -612,7 +650,6 @@ static u64 journal_seq_to_flush(struct journal *j)
  * @j:		journal object
  * @direct:	direct or background reclaim?
  * @kicked:	requested to run since we last ran?
- * Returns:	0 on success, or -EIO if the journal has been shutdown
  *
  * Background journal reclaim writes out btree nodes. It should be run
  * early enough so that we never completely run out of journal buckets.
@@ -632,6 +669,7 @@ static u64 journal_seq_to_flush(struct journal *j)
 static int __bch2_journal_reclaim(struct journal *j, bool direct, bool kicked)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	struct btree_cache *bc = &c->btree_cache;
 	bool kthread = (current->flags & PF_KTHREAD) != 0;
 	u64 seq_to_flush;
 	size_t min_nr, min_key_cache, nr_flushed;
@@ -651,11 +689,11 @@ static int __bch2_journal_reclaim(struct journal *j, bool direct, bool kicked)
 		if (kthread && kthread_should_stop())
 			break;
 
-		if (bch2_journal_error(j)) {
-			ret = -EIO;
+		ret = bch2_journal_error(j);
+		if (ret)
 			break;
-		}
 
+		/* XXX shove journal discards off to another thread */
 		bch2_journal_do_discards(j);
 
 		seq_to_flush = journal_seq_to_flush(j);
@@ -672,7 +710,8 @@ static int __bch2_journal_reclaim(struct journal *j, bool direct, bool kicked)
 		if (j->watermark != BCH_WATERMARK_stripe)
 			min_nr = 1;
 
-		if (atomic_read(&c->btree_cache.dirty) * 2 > c->btree_cache.used)
+		size_t btree_cache_live = bc->live[0].nr + bc->live[1].nr;
+		if (atomic_long_read(&bc->nr_dirty) * 2 > btree_cache_live)
 			min_nr = 1;
 
 		min_key_cache = min(bch2_nr_btree_keys_need_flush(c), (size_t) 128);
@@ -680,8 +719,7 @@ static int __bch2_journal_reclaim(struct journal *j, bool direct, bool kicked)
 		trace_and_count(c, journal_reclaim_start, c,
 				direct, kicked,
 				min_nr, min_key_cache,
-				atomic_read(&c->btree_cache.dirty),
-				c->btree_cache.used,
+				atomic_long_read(&bc->nr_dirty), btree_cache_live,
 				atomic_long_read(&c->btree_key_cache.nr_dirty),
 				atomic_long_read(&c->btree_key_cache.nr_keys));
 
@@ -748,10 +786,12 @@ static int bch2_journal_reclaim_thread(void *arg)
 			journal_empty = fifo_empty(&j->pin);
 			spin_unlock(&j->lock);
 
+			long timeout = j->next_reclaim - jiffies;
+
 			if (journal_empty)
 				schedule();
-			else if (time_after(j->next_reclaim, jiffies))
-				schedule_timeout(j->next_reclaim - jiffies);
+			else if (timeout > 0)
+				schedule_timeout(timeout);
 			else
 				break;
 		}
@@ -795,10 +835,41 @@ int bch2_journal_reclaim_start(struct journal *j)
 	return 0;
 }
 
+static bool journal_pins_still_flushing(struct journal *j, u64 seq_to_flush,
+					unsigned types)
+{
+	struct journal_entry_pin_list *pin_list;
+	u64 seq;
+
+	spin_lock(&j->lock);
+	fifo_for_each_entry_ptr(pin_list, &j->pin, seq) {
+		if (seq > seq_to_flush)
+			break;
+
+		for (unsigned i = 0; i < JOURNAL_PIN_TYPE_NR; i++)
+			if ((BIT(i) & types) &&
+			    (!list_empty(&pin_list->unflushed[i]) ||
+			     !list_empty(&pin_list->flushed[i]))) {
+				spin_unlock(&j->lock);
+				return true;
+			}
+	}
+	spin_unlock(&j->lock);
+
+	return false;
+}
+
+static bool journal_flush_pins_or_still_flushing(struct journal *j, u64 seq_to_flush,
+						 unsigned types)
+{
+	return  journal_flush_pins(j, seq_to_flush, types, 0, 0, 0) ||
+		journal_pins_still_flushing(j, seq_to_flush, types);
+}
+
 static int journal_flush_done(struct journal *j, u64 seq_to_flush,
 			      bool *did_work)
 {
-	int ret;
+	int ret = 0;
 
 	ret = bch2_journal_error(j);
 	if (ret)
@@ -806,12 +877,13 @@ static int journal_flush_done(struct journal *j, u64 seq_to_flush,
 
 	mutex_lock(&j->reclaim_lock);
 
-	if (journal_flush_pins(j, seq_to_flush,
-			       (1U << JOURNAL_PIN_key_cache)|
-			       (1U << JOURNAL_PIN_other), 0, 0, 0) ||
-	    journal_flush_pins(j, seq_to_flush,
-			       (1U << JOURNAL_PIN_btree), 0, 0, 0))
-		*did_work = true;
+	for (int type = JOURNAL_PIN_TYPE_NR - 1;
+	     type >= 0;
+	     --type)
+		if (journal_flush_pins_or_still_flushing(j, seq_to_flush, BIT(type))) {
+			*did_work = true;
+			goto unlock;
+		}
 
 	if (seq_to_flush > journal_cur_seq(j))
 		bch2_journal_entry_close(j);
@@ -821,11 +893,12 @@ static int journal_flush_done(struct journal *j, u64 seq_to_flush,
 	 * If journal replay hasn't completed, the unreplayed journal entries
 	 * hold refs on their corresponding sequence numbers
 	 */
-	ret = !test_bit(JOURNAL_REPLAY_DONE, &j->flags) ||
+	ret = !test_bit(JOURNAL_replay_done, &j->flags) ||
 		journal_last_seq(j) > seq_to_flush ||
 		!fifo_used(&j->pin);
 
 	spin_unlock(&j->lock);
+unlock:
 	mutex_unlock(&j->reclaim_lock);
 
 	return ret;
@@ -836,10 +909,10 @@ bool bch2_journal_flush_pins(struct journal *j, u64 seq_to_flush)
 	/* time_stats this */
 	bool did_work = false;
 
-	if (!test_bit(JOURNAL_STARTED, &j->flags))
+	if (!test_bit(JOURNAL_running, &j->flags))
 		return false;
 
-	closure_wait_event(&j->async_wait,
+	closure_wait_event(&j->reclaim_flush_wait,
 		journal_flush_done(j, seq_to_flush, &did_work));
 
 	return did_work;
@@ -883,7 +956,7 @@ int bch2_journal_flush_device_pins(struct journal *j, int dev_idx)
 	seq = 0;
 	spin_lock(&j->lock);
 	while (!ret) {
-		struct bch_replicas_padded replicas;
+		union bch_replicas_padded replicas;
 
 		seq = max(seq, journal_last_seq(j));
 		if (seq >= j->pin.back)
@@ -892,9 +965,11 @@ int bch2_journal_flush_device_pins(struct journal *j, int dev_idx)
 					 journal_seq_pin(j, seq)->devs);
 		seq++;
 
-		spin_unlock(&j->lock);
-		ret = bch2_mark_replicas(c, &replicas.e);
-		spin_lock(&j->lock);
+		if (replicas.e.nr_devs) {
+			spin_unlock(&j->lock);
+			ret = bch2_mark_replicas(c, &replicas.e);
+			spin_lock(&j->lock);
+		}
 	}
 	spin_unlock(&j->lock);
 err:
@@ -903,3 +978,54 @@ err:
 
 	return ret;
 }
+
+bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
+{
+	struct journal_entry_pin_list *pin_list;
+	struct journal_entry_pin *pin;
+
+	spin_lock(&j->lock);
+	if (!test_bit(JOURNAL_running, &j->flags)) {
+		spin_unlock(&j->lock);
+		return true;
+	}
+
+	*seq = max(*seq, j->pin.front);
+
+	if (*seq >= j->pin.back) {
+		spin_unlock(&j->lock);
+		return true;
+	}
+
+	out->atomic++;
+
+	pin_list = journal_seq_pin(j, *seq);
+
+	prt_printf(out, "%llu: count %u\n", *seq, atomic_read(&pin_list->count));
+	printbuf_indent_add(out, 2);
+
+	prt_printf(out, "unflushed:\n");
+	for (unsigned i = 0; i < ARRAY_SIZE(pin_list->unflushed); i++)
+		list_for_each_entry(pin, &pin_list->unflushed[i], list)
+			prt_printf(out, "\t%px %ps\n", pin, pin->flush);
+
+	prt_printf(out, "flushed:\n");
+	for (unsigned i = 0; i < ARRAY_SIZE(pin_list->flushed); i++)
+		list_for_each_entry(pin, &pin_list->flushed[i], list)
+			prt_printf(out, "\t%px %ps\n", pin, pin->flush);
+
+	printbuf_indent_sub(out, 2);
+
+	--out->atomic;
+	spin_unlock(&j->lock);
+
+	return false;
+}
+
+void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
+{
+	u64 seq = 0;
+
+	while (!bch2_journal_seq_pins_to_text(out, j, &seq))
+		seq++;
+}