dm thin: use slab mempools
Mike Snitzer [Sat, 2 Jun 2012 23:30:00 +0000 (00:30 +0100)]
Use dedicated caches prefixed with a "dm_" name rather than relying on
kmalloc mempools backed by generic slab caches so the memory usage of
thin provisioning (and any leaks) can be accounted for independently.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

drivers/md/dm-thin.c

index eb3d138..db1b041 100644 (file)
@@ -111,7 +111,7 @@ struct cell_key {
        dm_block_t block;
 };
 
-struct cell {
+struct dm_bio_prison_cell {
        struct hlist_node list;
        struct bio_prison *prison;
        struct cell_key key;
@@ -141,6 +141,8 @@ static uint32_t calc_nr_buckets(unsigned nr_cells)
        return n;
 }
 
+static struct kmem_cache *_cell_cache;
+
 /*
  * @nr_cells should be the number of cells you want in use _concurrently_.
  * Don't confuse it with the number of distinct keys.
@@ -157,8 +159,7 @@ static struct bio_prison *prison_create(unsigned nr_cells)
                return NULL;
 
        spin_lock_init(&prison->lock);
-       prison->cell_pool = mempool_create_kmalloc_pool(nr_cells,
-                                                       sizeof(struct cell));
+       prison->cell_pool = mempool_create_slab_pool(nr_cells, _cell_cache);
        if (!prison->cell_pool) {
                kfree(prison);
                return NULL;
@@ -194,10 +195,10 @@ static int keys_equal(struct cell_key *lhs, struct cell_key *rhs)
                       (lhs->block == rhs->block);
 }
 
-static struct cell *__search_bucket(struct hlist_head *bucket,
-                                   struct cell_key *key)
+static struct dm_bio_prison_cell *__search_bucket(struct hlist_head *bucket,
+                                                 struct cell_key *key)
 {
-       struct cell *cell;
+       struct dm_bio_prison_cell *cell;
        struct hlist_node *tmp;
 
        hlist_for_each_entry(cell, tmp, bucket, list)
@@ -214,12 +215,12 @@ static struct cell *__search_bucket(struct hlist_head *bucket,
  * Returns 1 if the cell was already held, 0 if @inmate is the new holder.
  */
 static int bio_detain(struct bio_prison *prison, struct cell_key *key,
-                     struct bio *inmate, struct cell **ref)
+                     struct bio *inmate, struct dm_bio_prison_cell **ref)
 {
        int r = 1;
        unsigned long flags;
        uint32_t hash = hash_key(prison, key);
-       struct cell *cell, *cell2;
+       struct dm_bio_prison_cell *cell, *cell2;
 
        BUG_ON(hash > prison->nr_buckets);
 
@@ -273,7 +274,7 @@ out:
 /*
  * @inmates must have been initialised prior to this call
  */
-static void __cell_release(struct cell *cell, struct bio_list *inmates)
+static void __cell_release(struct dm_bio_prison_cell *cell, struct bio_list *inmates)
 {
        struct bio_prison *prison = cell->prison;
 
@@ -287,7 +288,7 @@ static void __cell_release(struct cell *cell, struct bio_list *inmates)
        mempool_free(cell, prison->cell_pool);
 }
 
-static void cell_release(struct cell *cell, struct bio_list *bios)
+static void cell_release(struct dm_bio_prison_cell *cell, struct bio_list *bios)
 {
        unsigned long flags;
        struct bio_prison *prison = cell->prison;
@@ -303,7 +304,7 @@ static void cell_release(struct cell *cell, struct bio_list *bios)
  * bio may be in the cell.  This function releases the cell, and also does
  * a sanity check.
  */
-static void __cell_release_singleton(struct cell *cell, struct bio *bio)
+static void __cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio)
 {
        BUG_ON(cell->holder != bio);
        BUG_ON(!bio_list_empty(&cell->bios));
@@ -311,7 +312,7 @@ static void __cell_release_singleton(struct cell *cell, struct bio *bio)
        __cell_release(cell, NULL);
 }
 
-static void cell_release_singleton(struct cell *cell, struct bio *bio)
+static void cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio)
 {
        unsigned long flags;
        struct bio_prison *prison = cell->prison;
@@ -324,7 +325,8 @@ static void cell_release_singleton(struct cell *cell, struct bio *bio)
 /*
  * Sometimes we don't want the holder, just the additional bios.
  */
-static void __cell_release_no_holder(struct cell *cell, struct bio_list *inmates)
+static void __cell_release_no_holder(struct dm_bio_prison_cell *cell,
+                                    struct bio_list *inmates)
 {
        struct bio_prison *prison = cell->prison;
 
@@ -334,7 +336,8 @@ static void __cell_release_no_holder(struct cell *cell, struct bio_list *inmates
        mempool_free(cell, prison->cell_pool);
 }
 
-static void cell_release_no_holder(struct cell *cell, struct bio_list *inmates)
+static void cell_release_no_holder(struct dm_bio_prison_cell *cell,
+                                  struct bio_list *inmates)
 {
        unsigned long flags;
        struct bio_prison *prison = cell->prison;
@@ -344,7 +347,7 @@ static void cell_release_no_holder(struct cell *cell, struct bio_list *inmates)
        spin_unlock_irqrestore(&prison->lock, flags);
 }
 
-static void cell_error(struct cell *cell)
+static void cell_error(struct dm_bio_prison_cell *cell)
 {
        struct bio_prison *prison = cell->prison;
        struct bio_list bios;
@@ -491,7 +494,7 @@ static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
  * also provides the interface for creating and destroying internal
  * devices.
  */
-struct new_mapping;
+struct dm_thin_new_mapping;
 
 struct pool_features {
        unsigned zero_new_blocks:1;
@@ -537,7 +540,7 @@ struct pool {
        struct deferred_set shared_read_ds;
        struct deferred_set all_io_ds;
 
-       struct new_mapping *next_mapping;
+       struct dm_thin_new_mapping *next_mapping;
        mempool_t *mapping_pool;
        mempool_t *endio_hook_pool;
 };
@@ -630,11 +633,11 @@ static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev
 
 /*----------------------------------------------------------------*/
 
-struct endio_hook {
+struct dm_thin_endio_hook {
        struct thin_c *tc;
        struct deferred_entry *shared_read_entry;
        struct deferred_entry *all_io_entry;
-       struct new_mapping *overwrite_mapping;
+       struct dm_thin_new_mapping *overwrite_mapping;
 };
 
 static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
@@ -647,7 +650,8 @@ static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
        bio_list_init(master);
 
        while ((bio = bio_list_pop(&bios))) {
-               struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
+               struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
+
                if (h->tc == tc)
                        bio_endio(bio, DM_ENDIO_REQUEUE);
                else
@@ -736,7 +740,7 @@ static void wake_worker(struct pool *pool)
 /*
  * Bio endio functions.
  */
-struct new_mapping {
+struct dm_thin_new_mapping {
        struct list_head list;
 
        unsigned quiesced:1;
@@ -746,7 +750,7 @@ struct new_mapping {
        struct thin_c *tc;
        dm_block_t virt_block;
        dm_block_t data_block;
-       struct cell *cell, *cell2;
+       struct dm_bio_prison_cell *cell, *cell2;
        int err;
 
        /*
@@ -759,7 +763,7 @@ struct new_mapping {
        bio_end_io_t *saved_bi_end_io;
 };
 
-static void __maybe_add_mapping(struct new_mapping *m)
+static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
 {
        struct pool *pool = m->tc->pool;
 
@@ -772,7 +776,7 @@ static void __maybe_add_mapping(struct new_mapping *m)
 static void copy_complete(int read_err, unsigned long write_err, void *context)
 {
        unsigned long flags;
-       struct new_mapping *m = context;
+       struct dm_thin_new_mapping *m = context;
        struct pool *pool = m->tc->pool;
 
        m->err = read_err || write_err ? -EIO : 0;
@@ -786,8 +790,8 @@ static void copy_complete(int read_err, unsigned long write_err, void *context)
 static void overwrite_endio(struct bio *bio, int err)
 {
        unsigned long flags;
-       struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
-       struct new_mapping *m = h->overwrite_mapping;
+       struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
+       struct dm_thin_new_mapping *m = h->overwrite_mapping;
        struct pool *pool = m->tc->pool;
 
        m->err = err;
@@ -811,7 +815,7 @@ static void overwrite_endio(struct bio *bio, int err)
 /*
  * This sends the bios in the cell back to the deferred_bios list.
  */
-static void cell_defer(struct thin_c *tc, struct cell *cell,
+static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell,
                       dm_block_t data_block)
 {
        struct pool *pool = tc->pool;
@@ -828,7 +832,7 @@ static void cell_defer(struct thin_c *tc, struct cell *cell,
  * Same as cell_defer above, except it omits one particular detainee,
  * a write bio that covers the block and has already been processed.
  */
-static void cell_defer_except(struct thin_c *tc, struct cell *cell)
+static void cell_defer_except(struct thin_c *tc, struct dm_bio_prison_cell *cell)
 {
        struct bio_list bios;
        struct pool *pool = tc->pool;
@@ -843,7 +847,7 @@ static void cell_defer_except(struct thin_c *tc, struct cell *cell)
        wake_worker(pool);
 }
 
-static void process_prepared_mapping(struct new_mapping *m)
+static void process_prepared_mapping(struct dm_thin_new_mapping *m)
 {
        struct thin_c *tc = m->tc;
        struct bio *bio;
@@ -886,7 +890,7 @@ static void process_prepared_mapping(struct new_mapping *m)
        mempool_free(m, tc->pool->mapping_pool);
 }
 
-static void process_prepared_discard(struct new_mapping *m)
+static void process_prepared_discard(struct dm_thin_new_mapping *m)
 {
        int r;
        struct thin_c *tc = m->tc;
@@ -909,11 +913,11 @@ static void process_prepared_discard(struct new_mapping *m)
 }
 
 static void process_prepared(struct pool *pool, struct list_head *head,
-                            void (*fn)(struct new_mapping *))
+                            void (*fn)(struct dm_thin_new_mapping *))
 {
        unsigned long flags;
        struct list_head maps;
-       struct new_mapping *m, *tmp;
+       struct dm_thin_new_mapping *m, *tmp;
 
        INIT_LIST_HEAD(&maps);
        spin_lock_irqsave(&pool->lock, flags);
@@ -957,9 +961,9 @@ static int ensure_next_mapping(struct pool *pool)
        return pool->next_mapping ? 0 : -ENOMEM;
 }
 
-static struct new_mapping *get_next_mapping(struct pool *pool)
+static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
 {
-       struct new_mapping *r = pool->next_mapping;
+       struct dm_thin_new_mapping *r = pool->next_mapping;
 
        BUG_ON(!pool->next_mapping);
 
@@ -971,11 +975,11 @@ static struct new_mapping *get_next_mapping(struct pool *pool)
 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
                          struct dm_dev *origin, dm_block_t data_origin,
                          dm_block_t data_dest,
-                         struct cell *cell, struct bio *bio)
+                         struct dm_bio_prison_cell *cell, struct bio *bio)
 {
        int r;
        struct pool *pool = tc->pool;
-       struct new_mapping *m = get_next_mapping(pool);
+       struct dm_thin_new_mapping *m = get_next_mapping(pool);
 
        INIT_LIST_HEAD(&m->list);
        m->quiesced = 0;
@@ -997,7 +1001,8 @@ static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
         * bio immediately. Otherwise we use kcopyd to clone the data first.
         */
        if (io_overwrites_block(pool, bio)) {
-               struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
+               struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
+
                h->overwrite_mapping = m;
                m->bio = bio;
                save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
@@ -1025,7 +1030,7 @@ static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
 
 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
                                   dm_block_t data_origin, dm_block_t data_dest,
-                                  struct cell *cell, struct bio *bio)
+                                  struct dm_bio_prison_cell *cell, struct bio *bio)
 {
        schedule_copy(tc, virt_block, tc->pool_dev,
                      data_origin, data_dest, cell, bio);
@@ -1033,18 +1038,18 @@ static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
 
 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
                                   dm_block_t data_dest,
-                                  struct cell *cell, struct bio *bio)
+                                  struct dm_bio_prison_cell *cell, struct bio *bio)
 {
        schedule_copy(tc, virt_block, tc->origin_dev,
                      virt_block, data_dest, cell, bio);
 }
 
 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
-                         dm_block_t data_block, struct cell *cell,
+                         dm_block_t data_block, struct dm_bio_prison_cell *cell,
                          struct bio *bio)
 {
        struct pool *pool = tc->pool;
-       struct new_mapping *m = get_next_mapping(pool);
+       struct dm_thin_new_mapping *m = get_next_mapping(pool);
 
        INIT_LIST_HEAD(&m->list);
        m->quiesced = 1;
@@ -1065,12 +1070,12 @@ static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
                process_prepared_mapping(m);
 
        else if (io_overwrites_block(pool, bio)) {
-               struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
+               struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
+
                h->overwrite_mapping = m;
                m->bio = bio;
                save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
                remap_and_issue(tc, bio, data_block);
-
        } else {
                int r;
                struct dm_io_region to;
@@ -1155,7 +1160,7 @@ static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
  */
 static void retry_on_resume(struct bio *bio)
 {
-       struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
+       struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
        struct thin_c *tc = h->tc;
        struct pool *pool = tc->pool;
        unsigned long flags;
@@ -1165,7 +1170,7 @@ static void retry_on_resume(struct bio *bio)
        spin_unlock_irqrestore(&pool->lock, flags);
 }
 
-static void no_space(struct cell *cell)
+static void no_space(struct dm_bio_prison_cell *cell)
 {
        struct bio *bio;
        struct bio_list bios;
@@ -1182,11 +1187,11 @@ static void process_discard(struct thin_c *tc, struct bio *bio)
        int r;
        unsigned long flags;
        struct pool *pool = tc->pool;
-       struct cell *cell, *cell2;
+       struct dm_bio_prison_cell *cell, *cell2;
        struct cell_key key, key2;
        dm_block_t block = get_bio_block(tc, bio);
        struct dm_thin_lookup_result lookup_result;
-       struct new_mapping *m;
+       struct dm_thin_new_mapping *m;
 
        build_virtual_key(tc->td, block, &key);
        if (bio_detain(tc->pool->prison, &key, bio, &cell))
@@ -1263,7 +1268,7 @@ static void process_discard(struct thin_c *tc, struct bio *bio)
 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
                          struct cell_key *key,
                          struct dm_thin_lookup_result *lookup_result,
-                         struct cell *cell)
+                         struct dm_bio_prison_cell *cell)
 {
        int r;
        dm_block_t data_block;
@@ -1290,7 +1295,7 @@ static void process_shared_bio(struct thin_c *tc, struct bio *bio,
                               dm_block_t block,
                               struct dm_thin_lookup_result *lookup_result)
 {
-       struct cell *cell;
+       struct dm_bio_prison_cell *cell;
        struct pool *pool = tc->pool;
        struct cell_key key;
 
@@ -1305,7 +1310,7 @@ static void process_shared_bio(struct thin_c *tc, struct bio *bio,
        if (bio_data_dir(bio) == WRITE)
                break_sharing(tc, bio, block, &key, lookup_result, cell);
        else {
-               struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
+               struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
 
                h->shared_read_entry = ds_inc(&pool->shared_read_ds);
 
@@ -1315,7 +1320,7 @@ static void process_shared_bio(struct thin_c *tc, struct bio *bio,
 }
 
 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
-                           struct cell *cell)
+                           struct dm_bio_prison_cell *cell)
 {
        int r;
        dm_block_t data_block;
@@ -1363,7 +1368,7 @@ static void process_bio(struct thin_c *tc, struct bio *bio)
 {
        int r;
        dm_block_t block = get_bio_block(tc, bio);
-       struct cell *cell;
+       struct dm_bio_prison_cell *cell;
        struct cell_key key;
        struct dm_thin_lookup_result lookup_result;
 
@@ -1432,7 +1437,7 @@ static void process_deferred_bios(struct pool *pool)
        spin_unlock_irqrestore(&pool->lock, flags);
 
        while ((bio = bio_list_pop(&bios))) {
-               struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
+               struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
                struct thin_c *tc = h->tc;
 
                /*
@@ -1522,10 +1527,10 @@ static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
        wake_worker(pool);
 }
 
-static struct endio_hook *thin_hook_bio(struct thin_c *tc, struct bio *bio)
+static struct dm_thin_endio_hook *thin_hook_bio(struct thin_c *tc, struct bio *bio)
 {
        struct pool *pool = tc->pool;
-       struct endio_hook *h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO);
+       struct dm_thin_endio_hook *h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO);
 
        h->tc = tc;
        h->shared_read_entry = NULL;
@@ -1687,6 +1692,9 @@ static void __pool_destroy(struct pool *pool)
        kfree(pool);
 }
 
+static struct kmem_cache *_new_mapping_cache;
+static struct kmem_cache *_endio_hook_cache;
+
 static struct pool *pool_create(struct mapped_device *pool_md,
                                struct block_device *metadata_dev,
                                unsigned long block_size, char **error)
@@ -1755,16 +1763,16 @@ static struct pool *pool_create(struct mapped_device *pool_md,
        ds_init(&pool->all_io_ds);
 
        pool->next_mapping = NULL;
-       pool->mapping_pool =
-               mempool_create_kmalloc_pool(MAPPING_POOL_SIZE, sizeof(struct new_mapping));
+       pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
+                                                     _new_mapping_cache);
        if (!pool->mapping_pool) {
                *error = "Error creating pool's mapping mempool";
                err_p = ERR_PTR(-ENOMEM);
                goto bad_mapping_pool;
        }
 
-       pool->endio_hook_pool =
-               mempool_create_kmalloc_pool(ENDIO_HOOK_POOL_SIZE, sizeof(struct endio_hook));
+       pool->endio_hook_pool = mempool_create_slab_pool(ENDIO_HOOK_POOL_SIZE,
+                                                        _endio_hook_cache);
        if (!pool->endio_hook_pool) {
                *error = "Error creating pool's endio_hook mempool";
                err_p = ERR_PTR(-ENOMEM);
@@ -2613,9 +2621,9 @@ static int thin_endio(struct dm_target *ti,
                      union map_info *map_context)
 {
        unsigned long flags;
-       struct endio_hook *h = map_context->ptr;
+       struct dm_thin_endio_hook *h = map_context->ptr;
        struct list_head work;
-       struct new_mapping *m, *tmp;
+       struct dm_thin_new_mapping *m, *tmp;
        struct pool *pool = h->tc->pool;
 
        if (h->shared_read_entry) {
@@ -2755,7 +2763,32 @@ static int __init dm_thin_init(void)
 
        r = dm_register_target(&pool_target);
        if (r)
-               dm_unregister_target(&thin_target);
+               goto bad_pool_target;
+
+       r = -ENOMEM;
+
+       _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0);
+       if (!_cell_cache)
+               goto bad_cell_cache;
+
+       _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
+       if (!_new_mapping_cache)
+               goto bad_new_mapping_cache;
+
+       _endio_hook_cache = KMEM_CACHE(dm_thin_endio_hook, 0);
+       if (!_endio_hook_cache)
+               goto bad_endio_hook_cache;
+
+       return 0;
+
+bad_endio_hook_cache:
+       kmem_cache_destroy(_new_mapping_cache);
+bad_new_mapping_cache:
+       kmem_cache_destroy(_cell_cache);
+bad_cell_cache:
+       dm_unregister_target(&pool_target);
+bad_pool_target:
+       dm_unregister_target(&thin_target);
 
        return r;
 }
@@ -2764,6 +2797,10 @@ static void dm_thin_exit(void)
 {
        dm_unregister_target(&thin_target);
        dm_unregister_target(&pool_target);
+
+       kmem_cache_destroy(_cell_cache);
+       kmem_cache_destroy(_new_mapping_cache);
+       kmem_cache_destroy(_endio_hook_cache);
 }
 
 module_init(dm_thin_init);