bcache: use group_for_each_cache_rcu() in journal code

This way we only look at active devices after they have completed
bch_cache_read_write(). Previously, we would iterate over all
cache devices, skipping inactive devices or those not in tier 0.
However, newly-added devices are initialized after they are already
active, which would cause a race in journal_reclaim().

Also, move group_for_each_cache_rcu() from alloc.c to super.h.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

3 files changed, 21 insertions, 29 deletions

diff --git a/drivers/md/bcache/alloc.c b/drivers/md/bcache/alloc.c
index 207e7495d97c..9b8b068e3e56 100644
--- a/drivers/md/bcache/alloc.c
+++ b/drivers/md/bcache/alloc.c
@@ -97,21 +97,6 @@ void bch_cache_group_add_cache(struct cache_group *grp, struct cache *ca)
 	write_seqcount_end(&grp->lock);
 }
 
-static inline struct cache *cache_group_next(struct cache_group *devs,
-					     unsigned *iter)
-{
-	struct cache *ret = NULL;
-
-	while (*iter < devs->nr_devices &&
-	       !(ret = rcu_dereference(devs->devices[*iter])))
-		(*iter)++;
-
-	return ret;
-}
-
-#define group_for_each_cache_rcu(ca, devs, iter)			\
-	for ((iter) = 0; ((ca) = cache_group_next((devs), &(iter))); (iter)++)
-
 /* Ratelimiting/PD controllers */
 
 static void pd_controllers_update(struct work_struct *work)
diff --git a/drivers/md/bcache/journal.c b/drivers/md/bcache/journal.c
index 3ee0d45619e0..e9a39c71cef1 100644
--- a/drivers/md/bcache/journal.c
+++ b/drivers/md/bcache/journal.c
@@ -773,15 +773,11 @@ static void journal_reclaim_work(struct work_struct *work)
 
 	rcu_read_lock();
 
-	for_each_cache_rcu(ca, c, iter) {
+	group_for_each_cache_rcu(ca, &c->cache_tiers[0], iter) {
 		struct journal_device *ja = &ca->journal;
 		unsigned nr = bch_nr_journal_buckets(&ca->sb);
 		unsigned next = (ja->cur_idx + (nr >> 1)) % nr;
 
-		if ((CACHE_TIER(&ca->mi) != 0)
-		    || (CACHE_STATE(&ca->mi) != CACHE_ACTIVE))
-			continue;
-
 		/*
 		 * Write out enough btree nodes to free up 50% journal
 		 * buckets
@@ -853,14 +849,10 @@ static bool journal_reclaim_fast(struct cache_set *c)
 	 * Advance last_idx to point to the oldest journal entry containing
 	 * btree node updates that have not yet been written out
 	 */
-	for_each_cache_rcu(ca, c, iter) {
+	group_for_each_cache_rcu(ca, &c->cache_tiers[0], iter) {
 		struct journal_device *ja = &ca->journal;
 		unsigned nr = bch_nr_journal_buckets(&ca->sb);
 
-		if ((CACHE_TIER(&ca->mi) != 0)
-		    || (CACHE_STATE(&ca->mi) != CACHE_ACTIVE))
-			continue;
-
 		while (ja->last_idx != ja->cur_idx &&
 		       ja->seq[ja->last_idx] < last_seq)
 			ja->last_idx = (ja->last_idx + 1) % nr;
@@ -919,7 +911,7 @@ static void journal_next_bucket(struct cache_set *c)
 	 * Determine location of the next journal write:
 	 * XXX: sort caches by free journal space
 	 */
-	for_each_cache_rcu(ca, c, iter) {
+	group_for_each_cache_rcu(ca, &c->cache_tiers[0], iter) {
 		struct journal_device *ja = &ca->journal;
 		unsigned next = (ja->cur_idx + 1) %
 			bch_nr_journal_buckets(&ca->sb);
@@ -931,9 +923,7 @@ static void journal_next_bucket(struct cache_set *c)
 		 * Check that we can use this device, and aren't already using
 		 * it:
 		 */
-		if ((CACHE_TIER(&ca->mi) != 0) ||
-		    (CACHE_STATE(&ca->mi) != CACHE_ACTIVE) ||
-		    bch_extent_has_device(extent_s_to_s_c(e),
+		if (bch_extent_has_device(extent_s_to_s_c(e),
 					  ca->sb.nr_this_dev))
 			continue;
 
diff --git a/drivers/md/bcache/super.h b/drivers/md/bcache/super.h
index 317f0f557f9f..9c46b724cff9 100644
--- a/drivers/md/bcache/super.h
+++ b/drivers/md/bcache/super.h
@@ -59,6 +59,23 @@ static inline struct cache *bch_get_next_cache(struct cache_set *c,
 	     (ca = bch_get_next_cache(c, &(iter)));			\
 	     percpu_ref_put(&ca->ref), (iter)++)
 
+static inline struct cache *cache_group_next(struct cache_group *devs,
+					     unsigned *iter)
+{
+	struct cache *ret = NULL;
+
+	while (*iter < devs->nr_devices &&
+	       !(ret = rcu_dereference(devs->devices[*iter])))
+		(*iter)++;
+
+	return ret;
+}
+
+#define group_for_each_cache_rcu(ca, devs, iter)			\
+	for ((iter) = 0;						\
+	     ((ca) = cache_group_next((devs), &(iter)));		\
+	     (iter)++)
+
 u64 bch_checksum_update(unsigned, u64, const void *, size_t);
 u64 bch_checksum(unsigned, const void *, size_t);