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
diff --git a/drivers/md/dm-cache-policy-cleaner.c b/drivers/md/dm-cache-policy-cleaner.c
index b04d1f9..004e463 100644
--- a/drivers/md/dm-cache-policy-cleaner.c
+++ b/drivers/md/dm-cache-policy-cleaner.c
@@ -171,7 +171,8 @@
/* 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 2256a1f..c198e6d 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 3ddd116..515d44b 100644
--- a/drivers/md/dm-cache-policy-mq.c
+++ b/drivers/md/dm-cache-policy-mq.c
@@ -693,9 +693,10 @@
* - 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 @@
*/
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 @@
* 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 @@
/* 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_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 @@
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_cache(struct mq_policy *mq, dm_oblock_t oblock,
+ struct policy_locker *locker,
struct policy_result *result)
{
int r;
@@ -879,7 +891,7 @@
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 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 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 @@
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 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 @@
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 f50fe36..5524e21 100644
--- a/drivers/md/dm-cache-policy.h
+++ b/drivers/md/dm-cache-policy.h
@@ -70,6 +70,18 @@
};
/*
+ * 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 @@
*/
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 41b2594..d598248 100644
--- a/drivers/md/dm-cache-target.c
+++ b/drivers/md/dm-cache-target.c
@@ -1439,16 +1439,43 @@
&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 @@
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 @@
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 @@
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 @@
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;