dm snapshot: track snapshot reads
authorMikulas Patocka <mpatocka@redhat.com>
Mon, 21 Jul 2008 11:00:32 +0000 (12:00 +0100)
committerAlasdair G Kergon <agk@redhat.com>
Mon, 21 Jul 2008 11:00:32 +0000 (12:00 +0100)
Whenever a snapshot read gets mapped through to the origin, track it in
a per-snapshot hash table indexed by chunk number, using memory allocated
from a new per-snapshot mempool.

We need to track these reads to avoid race conditions which will be fixed
by patches that follow.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
drivers/md/dm-snap.c
drivers/md/dm-snap.h

index 1ba8a47d61b116646c6ac8ef533cd5b8f64767a7..de302702ab3e23b9dc95b0dee56b786d74a3365e 100644 (file)
  */
 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
 
+/*
+ * The size of the mempool used to track chunks in use.
+ */
+#define MIN_IOS 256
+
 static struct workqueue_struct *ksnapd;
 static void flush_queued_bios(struct work_struct *work);
 
@@ -93,6 +98,42 @@ static struct kmem_cache *exception_cache;
 static struct kmem_cache *pending_cache;
 static mempool_t *pending_pool;
 
+struct dm_snap_tracked_chunk {
+       struct hlist_node node;
+       chunk_t chunk;
+};
+
+static struct kmem_cache *tracked_chunk_cache;
+
+static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
+                                                chunk_t chunk)
+{
+       struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
+                                                       GFP_NOIO);
+       unsigned long flags;
+
+       c->chunk = chunk;
+
+       spin_lock_irqsave(&s->tracked_chunk_lock, flags);
+       hlist_add_head(&c->node,
+                      &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
+       spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
+
+       return c;
+}
+
+static void stop_tracking_chunk(struct dm_snapshot *s,
+                               struct dm_snap_tracked_chunk *c)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&s->tracked_chunk_lock, flags);
+       hlist_del(&c->node);
+       spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
+
+       mempool_free(c, s->tracked_chunk_pool);
+}
+
 /*
  * One of these per registered origin, held in the snapshot_origins hash
  */
@@ -482,6 +523,7 @@ static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg,
 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 {
        struct dm_snapshot *s;
+       int i;
        int r = -EINVAL;
        char persistent;
        char *origin_path;
@@ -564,11 +606,24 @@ static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
                goto bad5;
        }
 
+       s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
+                                                        tracked_chunk_cache);
+       if (!s->tracked_chunk_pool) {
+               ti->error = "Could not allocate tracked_chunk mempool for "
+                           "tracking reads";
+               goto bad6;
+       }
+
+       for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
+               INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
+
+       spin_lock_init(&s->tracked_chunk_lock);
+
        /* Metadata must only be loaded into one table at once */
        r = s->store.read_metadata(&s->store);
        if (r < 0) {
                ti->error = "Failed to read snapshot metadata";
-               goto bad6;
+               goto bad_load_and_register;
        } else if (r > 0) {
                s->valid = 0;
                DMWARN("Snapshot is marked invalid.");
@@ -582,7 +637,7 @@ static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
        if (register_snapshot(s)) {
                r = -EINVAL;
                ti->error = "Cannot register snapshot origin";
-               goto bad6;
+               goto bad_load_and_register;
        }
 
        ti->private = s;
@@ -590,6 +645,9 @@ static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 
        return 0;
 
+ bad_load_and_register:
+       mempool_destroy(s->tracked_chunk_pool);
+
  bad6:
        dm_kcopyd_client_destroy(s->kcopyd_client);
 
@@ -624,6 +682,9 @@ static void __free_exceptions(struct dm_snapshot *s)
 
 static void snapshot_dtr(struct dm_target *ti)
 {
+#ifdef CONFIG_DM_DEBUG
+       int i;
+#endif
        struct dm_snapshot *s = ti->private;
 
        flush_workqueue(ksnapd);
@@ -632,6 +693,13 @@ static void snapshot_dtr(struct dm_target *ti)
        /* After this returns there can be no new kcopyd jobs. */
        unregister_snapshot(s);
 
+#ifdef CONFIG_DM_DEBUG
+       for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
+               BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
+#endif
+
+       mempool_destroy(s->tracked_chunk_pool);
+
        __free_exceptions(s);
 
        dm_put_device(ti, s->origin);
@@ -974,14 +1042,10 @@ static int snapshot_map(struct dm_target *ti, struct bio *bio,
                        start_copy(pe);
                        goto out;
                }
-       } else
-               /*
-                * FIXME: this read path scares me because we
-                * always use the origin when we have a pending
-                * exception.  However I can't think of a
-                * situation where this is wrong - ejt.
-                */
+       } else {
                bio->bi_bdev = s->origin->bdev;
+               map_context->ptr = track_chunk(s, chunk);
+       }
 
  out_unlock:
        up_write(&s->lock);
@@ -989,6 +1053,18 @@ static int snapshot_map(struct dm_target *ti, struct bio *bio,
        return r;
 }
 
+static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
+                          int error, union map_info *map_context)
+{
+       struct dm_snapshot *s = ti->private;
+       struct dm_snap_tracked_chunk *c = map_context->ptr;
+
+       if (c)
+               stop_tracking_chunk(s, c);
+
+       return 0;
+}
+
 static void snapshot_resume(struct dm_target *ti)
 {
        struct dm_snapshot *s = ti->private;
@@ -1266,6 +1342,7 @@ static struct target_type snapshot_target = {
        .ctr     = snapshot_ctr,
        .dtr     = snapshot_dtr,
        .map     = snapshot_map,
+       .end_io  = snapshot_end_io,
        .resume  = snapshot_resume,
        .status  = snapshot_status,
 };
@@ -1306,11 +1383,18 @@ static int __init dm_snapshot_init(void)
                goto bad4;
        }
 
+       tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
+       if (!tracked_chunk_cache) {
+               DMERR("Couldn't create cache to track chunks in use.");
+               r = -ENOMEM;
+               goto bad5;
+       }
+
        pending_pool = mempool_create_slab_pool(128, pending_cache);
        if (!pending_pool) {
                DMERR("Couldn't create pending pool.");
                r = -ENOMEM;
-               goto bad5;
+               goto bad_pending_pool;
        }
 
        ksnapd = create_singlethread_workqueue("ksnapd");
@@ -1324,6 +1408,8 @@ static int __init dm_snapshot_init(void)
 
       bad6:
        mempool_destroy(pending_pool);
+      bad_pending_pool:
+       kmem_cache_destroy(tracked_chunk_cache);
       bad5:
        kmem_cache_destroy(pending_cache);
       bad4:
@@ -1355,6 +1441,7 @@ static void __exit dm_snapshot_exit(void)
        mempool_destroy(pending_pool);
        kmem_cache_destroy(pending_cache);
        kmem_cache_destroy(exception_cache);
+       kmem_cache_destroy(tracked_chunk_cache);
 }
 
 /* Module hooks */
index 24f9fb73b982d0cd148ef15d56428ba6c7ce6877..70dc961f40d8e7efede21d934b42888cc5762a7c 100644 (file)
@@ -130,6 +130,10 @@ struct exception_store {
        void *context;
 };
 
+#define DM_TRACKED_CHUNK_HASH_SIZE     16
+#define DM_TRACKED_CHUNK_HASH(x)       ((unsigned long)(x) & \
+                                        (DM_TRACKED_CHUNK_HASH_SIZE - 1))
+
 struct dm_snapshot {
        struct rw_semaphore lock;
        struct dm_target *ti;
@@ -174,6 +178,11 @@ struct dm_snapshot {
        /* Queue of snapshot writes for ksnapd to flush */
        struct bio_list queued_bios;
        struct work_struct queued_bios_work;
+
+       /* Chunks with outstanding reads */
+       mempool_t *tracked_chunk_pool;
+       spinlock_t tracked_chunk_lock;
+       struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
 };
 
 /*