dm cache: fix race when issuing a POLICY_REPLACE operation

There is a race between a policy deciding to replace a cache entry,
the core target writing back any dirty data from this block, and other
IO threads doing IO to the same block.

This sort of problem is avoided most of the time by the core target
grabbing a bio prison cell before making the request to the policy.
But for a demotion the core target doesn't know which block will be
demoted, so can't do this in advance.

Fix this demotion race by introducing a callback to the policy interface
that allows the policy to grab the cell on behalf of the core target.

Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org

5 files changed, 85 insertions, 37 deletions

diff --git a/drivers/md/dm-cache-policy-cleaner.c b/drivers/md/dm-cache-policy-cleaner.c
index b04d1f904d07..004e463c9423 100644
--- a/drivers/md/dm-cache-policy-cleaner.c
+++ b/drivers/md/dm-cache-policy-cleaner.c
@@ -171,7 +171,8 @@ static void remove_cache_hash_entry(struct wb_cache_entry *e)
 /* Public interface (see dm-cache-policy.h */
 static int wb_map(struct dm_cache_policy *pe, dm_oblock_t oblock,
 		  bool can_block, bool can_migrate, bool discarded_oblock,
-		  struct bio *bio, struct policy_result *result)
+		  struct bio *bio, struct policy_locker *locker,
+		  struct policy_result *result)
 {
 	struct policy *p = to_policy(pe);
 	struct wb_cache_entry *e;
diff --git a/drivers/md/dm-cache-policy-internal.h b/drivers/md/dm-cache-policy-internal.h
index 2256a1f24f73..c198e6defb9c 100644
--- a/drivers/md/dm-cache-policy-internal.h
+++ b/drivers/md/dm-cache-policy-internal.h
@@ -16,9 +16,10 @@
  */
 static inline int policy_map(struct dm_cache_policy *p, dm_oblock_t oblock,
 			     bool can_block, bool can_migrate, bool discarded_oblock,
-			     struct bio *bio, struct policy_result *result)
+			     struct bio *bio, struct policy_locker *locker,
+			     struct policy_result *result)
 {
-	return p->map(p, oblock, can_block, can_migrate, discarded_oblock, bio, result);
+	return p->map(p, oblock, can_block, can_migrate, discarded_oblock, bio, locker, result);
 }
 
 static inline int policy_lookup(struct dm_cache_policy *p, dm_oblock_t oblock, dm_cblock_t *cblock)
diff --git a/drivers/md/dm-cache-policy-mq.c b/drivers/md/dm-cache-policy-mq.c
index 3ddd1162334d..515d44bf24d3 100644
--- a/drivers/md/dm-cache-policy-mq.c
+++ b/drivers/md/dm-cache-policy-mq.c
@@ -693,9 +693,10 @@ static void requeue(struct mq_policy *mq, struct entry *e)
  * - set the hit count to a hard coded value other than 1, eg, is it better
  *   if it goes in at level 2?
  */
-static int demote_cblock(struct mq_policy *mq, dm_oblock_t *oblock)
+static int demote_cblock(struct mq_policy *mq,
+			 struct policy_locker *locker, dm_oblock_t *oblock)
 {
-	struct entry *demoted = pop(mq, &mq->cache_clean);
+	struct entry *demoted = peek(&mq->cache_clean);
 
 	if (!demoted)
 		/*
@@ -707,6 +708,13 @@ static int demote_cblock(struct mq_policy *mq, dm_oblock_t *oblock)
 		 */
 		return -ENOSPC;
 
+	if (locker->fn(locker, demoted->oblock))
+		/*
+		 * We couldn't lock the demoted block.
+		 */
+		return -EBUSY;
+
+	del(mq, demoted);
 	*oblock = demoted->oblock;
 	free_entry(&mq->cache_pool, demoted);
 
@@ -795,6 +803,7 @@ static int cache_entry_found(struct mq_policy *mq,
  * finding which cache block to use.
  */
 static int pre_cache_to_cache(struct mq_policy *mq, struct entry *e,
+			      struct policy_locker *locker,
 			      struct policy_result *result)
 {
 	int r;
@@ -803,11 +812,12 @@ static int pre_cache_to_cache(struct mq_policy *mq, struct entry *e,
 	/* Ensure there's a free cblock in the cache */
 	if (epool_empty(&mq->cache_pool)) {
 		result->op = POLICY_REPLACE;
-		r = demote_cblock(mq, &result->old_oblock);
+		r = demote_cblock(mq, locker, &result->old_oblock);
 		if (r) {
 			result->op = POLICY_MISS;
 			return 0;
 		}
+
 	} else
 		result->op = POLICY_NEW;
 
@@ -829,7 +839,8 @@ static int pre_cache_to_cache(struct mq_policy *mq, struct entry *e,
 
 static int pre_cache_entry_found(struct mq_policy *mq, struct entry *e,
 				 bool can_migrate, bool discarded_oblock,
-				 int data_dir, struct policy_result *result)
+				 int data_dir, struct policy_locker *locker,
+				 struct policy_result *result)
 {
 	int r = 0;
 
@@ -842,7 +853,7 @@ static int pre_cache_entry_found(struct mq_policy *mq, struct entry *e,
 
 	else {
 		requeue(mq, e);
-		r = pre_cache_to_cache(mq, e, result);
+		r = pre_cache_to_cache(mq, e, locker, result);
 	}
 
 	return r;
@@ -872,6 +883,7 @@ static void insert_in_pre_cache(struct mq_policy *mq,
 }
 
 static void insert_in_cache(struct mq_policy *mq, dm_oblock_t oblock,
+			    struct policy_locker *locker,
 			    struct policy_result *result)
 {
 	int r;
@@ -879,7 +891,7 @@ static void insert_in_cache(struct mq_policy *mq, dm_oblock_t oblock,
 
 	if (epool_empty(&mq->cache_pool)) {
 		result->op = POLICY_REPLACE;
-		r = demote_cblock(mq, &result->old_oblock);
+		r = demote_cblock(mq, locker, &result->old_oblock);
 		if (unlikely(r)) {
 			result->op = POLICY_MISS;
 			insert_in_pre_cache(mq, oblock);
@@ -907,11 +919,12 @@ static void insert_in_cache(struct mq_policy *mq, dm_oblock_t oblock,
 
 static int no_entry_found(struct mq_policy *mq, dm_oblock_t oblock,
 			  bool can_migrate, bool discarded_oblock,
-			  int data_dir, struct policy_result *result)
+			  int data_dir, struct policy_locker *locker,
+			  struct policy_result *result)
 {
 	if (adjusted_promote_threshold(mq, discarded_oblock, data_dir) <= 1) {
 		if (can_migrate)
-			insert_in_cache(mq, oblock, result);
+			insert_in_cache(mq, oblock, locker, result);
 		else
 			return -EWOULDBLOCK;
 	} else {
@@ -928,7 +941,8 @@ static int no_entry_found(struct mq_policy *mq, dm_oblock_t oblock,
  */
 static int map(struct mq_policy *mq, dm_oblock_t oblock,
 	       bool can_migrate, bool discarded_oblock,
-	       int data_dir, struct policy_result *result)
+	       int data_dir, struct policy_locker *locker,
+	       struct policy_result *result)
 {
 	int r = 0;
 	struct entry *e = hash_lookup(mq, oblock);
@@ -942,11 +956,11 @@ static int map(struct mq_policy *mq, dm_oblock_t oblock,
 
 	else if (e)
 		r = pre_cache_entry_found(mq, e, can_migrate, discarded_oblock,
-					  data_dir, result);
+					  data_dir, locker, result);
 
 	else
 		r = no_entry_found(mq, oblock, can_migrate, discarded_oblock,
-				   data_dir, result);
+				   data_dir, locker, result);
 
 	if (r == -EWOULDBLOCK)
 		result->op = POLICY_MISS;
@@ -1012,7 +1026,8 @@ static void copy_tick(struct mq_policy *mq)
 
 static int mq_map(struct dm_cache_policy *p, dm_oblock_t oblock,
 		  bool can_block, bool can_migrate, bool discarded_oblock,
-		  struct bio *bio, struct policy_result *result)
+		  struct bio *bio, struct policy_locker *locker,
+		  struct policy_result *result)
 {
 	int r;
 	struct mq_policy *mq = to_mq_policy(p);
@@ -1028,7 +1043,7 @@ static int mq_map(struct dm_cache_policy *p, dm_oblock_t oblock,
 
 	iot_examine_bio(&mq->tracker, bio);
 	r = map(mq, oblock, can_migrate, discarded_oblock,
-		bio_data_dir(bio), result);
+		bio_data_dir(bio), locker, result);
 
 	mutex_unlock(&mq->lock);
 
diff --git a/drivers/md/dm-cache-policy.h b/drivers/md/dm-cache-policy.h
index f50fe360c546..5524e21e4836 100644
--- a/drivers/md/dm-cache-policy.h
+++ b/drivers/md/dm-cache-policy.h
@@ -70,6 +70,18 @@ enum policy_operation {
 };
 
 /*
+ * When issuing a POLICY_REPLACE the policy needs to make a callback to
+ * lock the block being demoted.  This doesn't need to occur during a
+ * writeback operation since the block remains in the cache.
+ */
+struct policy_locker;
+typedef int (*policy_lock_fn)(struct policy_locker *l, dm_oblock_t oblock);
+
+struct policy_locker {
+	policy_lock_fn fn;
+};
+
+/*
  * This is the instruction passed back to the core target.
  */
 struct policy_result {
@@ -122,7 +134,8 @@ struct dm_cache_policy {
 	 */
 	int (*map)(struct dm_cache_policy *p, dm_oblock_t oblock,
 		   bool can_block, bool can_migrate, bool discarded_oblock,
-		   struct bio *bio, struct policy_result *result);
+		   struct bio *bio, struct policy_locker *locker,
+		   struct policy_result *result);
 
 	/*
 	 * Sometimes we want to see if a block is in the cache, without
diff --git a/drivers/md/dm-cache-target.c b/drivers/md/dm-cache-target.c
index 41b2594a80c6..d5982480630b 100644
--- a/drivers/md/dm-cache-target.c
+++ b/drivers/md/dm-cache-target.c
@@ -1439,16 +1439,43 @@ static void inc_miss_counter(struct cache *cache, struct bio *bio)
 		   &cache->stats.read_miss : &cache->stats.write_miss);
 }
 
+/*----------------------------------------------------------------*/
+
+struct old_oblock_lock {
+	struct policy_locker locker;
+	struct cache *cache;
+	struct prealloc *structs;
+	struct dm_bio_prison_cell *cell;
+};
+
+static int null_locker(struct policy_locker *locker, dm_oblock_t b)
+{
+	/* This should never be called */
+	BUG();
+	return 0;
+}
+
+static int cell_locker(struct policy_locker *locker, dm_oblock_t b)
+{
+	struct old_oblock_lock *l = container_of(locker, struct old_oblock_lock, locker);
+	struct dm_bio_prison_cell *cell_prealloc = prealloc_get_cell(l->structs);
+
+	return bio_detain(l->cache, b, NULL, cell_prealloc,
+			  (cell_free_fn) prealloc_put_cell,
+			  l->structs, &l->cell);
+}
+
 static void process_bio(struct cache *cache, struct prealloc *structs,
 			struct bio *bio)
 {
 	int r;
 	bool release_cell = true;
 	dm_oblock_t block = get_bio_block(cache, bio);
-	struct dm_bio_prison_cell *cell_prealloc, *old_ocell, *new_ocell;
+	struct dm_bio_prison_cell *cell_prealloc, *new_ocell;
 	struct policy_result lookup_result;
 	bool passthrough = passthrough_mode(&cache->features);
 	bool discarded_block, can_migrate;
+	struct old_oblock_lock ool;
 
 	/*
 	 * Check to see if that block is currently migrating.
@@ -1463,8 +1490,12 @@ static void process_bio(struct cache *cache, struct prealloc *structs,
 	discarded_block = is_discarded_oblock(cache, block);
 	can_migrate = !passthrough && (discarded_block || spare_migration_bandwidth(cache));
 
+	ool.locker.fn = cell_locker;
+	ool.cache = cache;
+	ool.structs = structs;
+	ool.cell = NULL;
 	r = policy_map(cache->policy, block, true, can_migrate, discarded_block,
-		       bio, &lookup_result);
+		       bio, &ool.locker, &lookup_result);
 
 	if (r == -EWOULDBLOCK)
 		/* migration has been denied */
@@ -1521,27 +1552,11 @@ static void process_bio(struct cache *cache, struct prealloc *structs,
 		break;
 
 	case POLICY_REPLACE:
-		cell_prealloc = prealloc_get_cell(structs);
-		r = bio_detain(cache, lookup_result.old_oblock, bio, cell_prealloc,
-			       (cell_free_fn) prealloc_put_cell,
-			       structs, &old_ocell);
-		if (r > 0) {
-			/*
-			 * We have to be careful to avoid lock inversion of
-			 * the cells.  So we back off, and wait for the
-			 * old_ocell to become free.
-			 */
-			policy_force_mapping(cache->policy, block,
-					     lookup_result.old_oblock);
-			atomic_inc(&cache->stats.cache_cell_clash);
-			break;
-		}
 		atomic_inc(&cache->stats.demotion);
 		atomic_inc(&cache->stats.promotion);
-
 		demote_then_promote(cache, structs, lookup_result.old_oblock,
 				    block, lookup_result.cblock,
-				    old_ocell, new_ocell);
+				    ool.cell, new_ocell);
 		release_cell = false;
 		break;
 
@@ -2589,6 +2604,9 @@ static int __cache_map(struct cache *cache, struct bio *bio, struct dm_bio_priso
 	bool discarded_block;
 	struct policy_result lookup_result;
 	struct per_bio_data *pb = init_per_bio_data(bio, pb_data_size);
+	struct old_oblock_lock ool;
+
+	ool.locker.fn = null_locker;
 
 	if (unlikely(from_oblock(block) >= from_oblock(cache->origin_blocks))) {
 		/*
@@ -2627,7 +2645,7 @@ static int __cache_map(struct cache *cache, struct bio *bio, struct dm_bio_priso
 	discarded_block = is_discarded_oblock(cache, block);
 
 	r = policy_map(cache->policy, block, false, can_migrate, discarded_block,
-		       bio, &lookup_result);
+		       bio, &ool.locker, &lookup_result);
 	if (r == -EWOULDBLOCK) {
 		cell_defer(cache, *cell, true);
 		return DM_MAPIO_SUBMITTED;