dm snapshot: skip reading origin when overwriting complete chunk

If we write a full chunk in the snapshot, skip reading the origin device
because the whole chunk will be overwritten anyway.

This patch changes the snapshot write logic when a full chunk is written.
In this case:
  1. allocate the exception
  2. dispatch the bio (but don't report the bio completion to device mapper)
  3. write the exception record
  4. report bio completed

Callbacks must be done through the kcopyd thread, because callbacks must not
race with each other.  So we create two new functions:

  dm_kcopyd_prepare_callback: allocate a job structure and prepare the callback.
  (This function must not be called from interrupt context.)

  dm_kcopyd_do_callback: submit callback.
  (This function may be called from interrupt context.)

Performance test (on snapshots with 4k chunk size):
  without the patch:
    non-direct-io sequential write (dd):    17.7MB/s
    direct-io sequential write (dd):        20.9MB/s
    non-direct-io random write (mkfs.ext2): 0.44s

  with the patch:
    non-direct-io sequential write (dd):    26.5MB/s
    direct-io sequential write (dd):        33.2MB/s
    non-direct-io random write (mkfs.ext2): 0.27s
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

drivers/md/dm-kcopyd.c

@@ -617,6 +617,37 @@ int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
 }
 EXPORT_SYMBOL(dm_kcopyd_copy);
 
+void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
+				 dm_kcopyd_notify_fn fn, void *context)
+{
+	struct kcopyd_job *job;
+
+	job = mempool_alloc(kc->job_pool, GFP_NOIO);
+
+	memset(job, 0, sizeof(struct kcopyd_job));
+	job->kc = kc;
+	job->fn = fn;
+	job->context = context;
+
+	atomic_inc(&kc->nr_jobs);
+
+	return job;
+}
+EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
+
+void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
+{
+	struct kcopyd_job *job = j;
+	struct dm_kcopyd_client *kc = job->kc;
+
+	job->read_err = read_err;
+	job->write_err = write_err;
+
+	push(&kc->complete_jobs, job);
+	wake(kc);
+}
+EXPORT_SYMBOL(dm_kcopyd_do_callback);
+
 /*
  * Cancels a kcopyd job, eg. someone might be deactivating a
  * mirror.

drivers/md/dm-snap.c

@@ -170,6 +170,13 @@ struct dm_snap_pending_exception {
 	 * kcopyd.
 	 */
 	int started;
+
+	/*
+	 * For writing a complete chunk, bypassing the copy.
+	 */
+	struct bio *full_bio;
+	bio_end_io_t *full_bio_end_io;
+	void *full_bio_private;
 };
 
 /*
@@ -1369,6 +1376,7 @@ static void pending_complete(struct dm_snap_pending_exception *pe, int success)
 	struct dm_snapshot *s = pe->snap;
 	struct bio *origin_bios = NULL;
 	struct bio *snapshot_bios = NULL;
+	struct bio *full_bio = NULL;
 	int error = 0;
 
 	if (!success) {
@@ -1408,6 +1416,11 @@ static void pending_complete(struct dm_snap_pending_exception *pe, int success)
 	dm_remove_exception(&pe->e);
 	snapshot_bios = bio_list_get(&pe->snapshot_bios);
 	origin_bios = bio_list_get(&pe->origin_bios);
+	full_bio = pe->full_bio;
+	if (full_bio) {
+		full_bio->bi_end_io = pe->full_bio_end_io;
+		full_bio->bi_private = pe->full_bio_private;
+	}
 	free_pending_exception(pe);
 
 	increment_pending_exceptions_done_count();
@@ -1415,10 +1428,15 @@ static void pending_complete(struct dm_snap_pending_exception *pe, int success)
 	up_write(&s->lock);
 
 	/* Submit any pending write bios */
-	if (error)
+	if (error) {
+		if (full_bio)
+			bio_io_error(full_bio);
 		error_bios(snapshot_bios);
-	else
+	} else {
+		if (full_bio)
+			bio_endio(full_bio, 0);
 		flush_bios(snapshot_bios);
+	}
 
 	retry_origin_bios(s, origin_bios);
 }
@@ -1472,6 +1490,32 @@ static void start_copy(struct dm_snap_pending_exception *pe)
 	dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
 }
 
+static void full_bio_end_io(struct bio *bio, int error)
+{
+	void *callback_data = bio->bi_private;
+
+	dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
+}
+
+static void start_full_bio(struct dm_snap_pending_exception *pe,
+			   struct bio *bio)
+{
+	struct dm_snapshot *s = pe->snap;
+	void *callback_data;
+
+	pe->full_bio = bio;
+	pe->full_bio_end_io = bio->bi_end_io;
+	pe->full_bio_private = bio->bi_private;
+
+	callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
+						   copy_callback, pe);
+
+	bio->bi_end_io = full_bio_end_io;
+	bio->bi_private = callback_data;
+
+	generic_make_request(bio);
+}
+
 static struct dm_snap_pending_exception *
 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
 {
@@ -1507,6 +1551,7 @@ __find_pending_exception(struct dm_snapshot *s,
 	bio_list_init(&pe->origin_bios);
 	bio_list_init(&pe->snapshot_bios);
 	pe->started = 0;
+	pe->full_bio = NULL;
 
 	if (s->store->type->prepare_exception(s->store, &pe->e)) {
 		free_pending_exception(pe);
@@ -1600,10 +1645,19 @@ static int snapshot_map(struct dm_target *ti, struct bio *bio,
 		}
 
 		remap_exception(s, &pe->e, bio, chunk);
-		bio_list_add(&pe->snapshot_bios, bio);
 
 		r = DM_MAPIO_SUBMITTED;
 
+		if (!pe->started &&
+		    bio->bi_size == (s->store->chunk_size << SECTOR_SHIFT)) {
+			pe->started = 1;
+			up_write(&s->lock);
+			start_full_bio(pe, bio);
+			goto out;
+		}
+
+		bio_list_add(&pe->snapshot_bios, bio);
+
 		if (!pe->started) {
 			/* this is protected by snap->lock */
 			pe->started = 1;

include/linux/dm-kcopyd.h

@@ -42,5 +42,20 @@ int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
 		   unsigned num_dests, struct dm_io_region *dests,
 		   unsigned flags, dm_kcopyd_notify_fn fn, void *context);
 
+/*
+ * Prepare a callback and submit it via the kcopyd thread.
+ *
+ * dm_kcopyd_prepare_callback allocates a callback structure and returns it.
+ * It must not be called from interrupt context.
+ * The returned value should be passed into dm_kcopyd_do_callback.
+ *
+ * dm_kcopyd_do_callback submits the callback.
+ * It may be called from interrupt context.
+ * The callback is issued from the kcopyd thread.
+ */
+void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
+				 dm_kcopyd_notify_fn fn, void *context);
+void dm_kcopyd_do_callback(void *job, int read_err, unsigned long write_err);
+
 #endif	/* __KERNEL__ */
 #endif	/* _LINUX_DM_KCOPYD_H */