block: loop: support DIO & AIO

There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:

1) double cache can be avoided, then memory usage gets
decreased a lot

2) not like user space direct I/O, there isn't cost of
pinning pages

3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime

xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.

Follows the fio test result for performance purpose:
	4 jobs fio test inside ext4 file system over loop block

1) How to run
	- KVM: 4 VCPUs, 2G RAM
	- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
	- the loop block is over one image on SSD.
	- linux psync, 4 jobs, size 1500M, ext4 over loop block
	- test result: IOPS from fio output

2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
        -------------------------------------------------------------
        test cases          |randread   |read   |randwrite  |write  |
        -------------------------------------------------------------
        base                |8015       |113811 |67442      |106978
        -------------------------------------------------------------
        base+loop aio       |8136       |125040 |67811      |111376
        -------------------------------------------------------------

- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O

3) context switch
        - context switch decreased by ~50% with loop direct I/O(aio)
	compared with loop buffered I/O(4.2-rc6-next-20150814)

4) memory usage from /proc/meminfo
        -------------------------------------------------------------
                                   | Buffers       | Cached
        -------------------------------------------------------------
        base                       | > 760MB       | ~950MB
        -------------------------------------------------------------
        base+loop direct I/O(aio)  | < 5MB         | ~1.6GB
        -------------------------------------------------------------

- so there are much more page caches available for application with
direct I/O

[1] https://lwn.net/Articles/612483/

Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>

2 files changed, 97 insertions, 3 deletions

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 75db3b98ec2b..23376084a5cb 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -445,6 +445,90 @@ static int lo_req_flush(struct loop_device *lo, struct request *rq)
         return ret;
 }
 
+static inline void handle_partial_read(struct loop_cmd *cmd, long bytes)
+{
+        if (bytes < 0 || (cmd->rq->cmd_flags & REQ_WRITE))
+                return;
+
+        if (unlikely(bytes < blk_rq_bytes(cmd->rq))) {
+                struct bio *bio = cmd->rq->bio;
+
+                bio_advance(bio, bytes);
+                zero_fill_bio(bio);
+        }
+}
+
+static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
+{
+        struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
+        struct request *rq = cmd->rq;
+
+        handle_partial_read(cmd, ret);
+
+        if (ret > 0)
+                ret = 0;
+        else if (ret < 0)
+                ret = -EIO;
+
+        rq->errors = ret;
+        blk_mq_complete_request(rq);
+}
+
+static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
+                     loff_t pos, bool rw)
+{
+        struct iov_iter iter;
+        struct bio_vec *bvec;
+        struct bio *bio = cmd->rq->bio;
+        struct file *file = lo->lo_backing_file;
+        int ret;
+
+        /* nomerge for loop request queue */
+        WARN_ON(cmd->rq->bio != cmd->rq->biotail);
+
+        bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
+        iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
+                      bio_segments(bio), blk_rq_bytes(cmd->rq));
+
+        cmd->iocb.ki_pos = pos;
+        cmd->iocb.ki_filp = file;
+        cmd->iocb.ki_complete = lo_rw_aio_complete;
+        cmd->iocb.ki_flags = IOCB_DIRECT;
+
+        if (rw == WRITE)
+                ret = file->f_op->write_iter(&cmd->iocb, &iter);
+        else
+                ret = file->f_op->read_iter(&cmd->iocb, &iter);
+
+        if (ret != -EIOCBQUEUED)
+                cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
+        return 0;
+}
+
+
+static inline int lo_rw_simple(struct loop_device *lo,
+                struct request *rq, loff_t pos, bool rw)
+{
+        struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
+
+        if (cmd->use_aio)
+                return lo_rw_aio(lo, cmd, pos, rw);
+
+        /*
+         * lo_write_simple and lo_read_simple should have been covered
+         * by io submit style function like lo_rw_aio(), one blocker
+         * is that lo_read_simple() need to call flush_dcache_page after
+         * the page is written from kernel, and it isn't easy to handle
+         * this in io submit style function which submits all segments
+         * of the req at one time. And direct read IO doesn't need to
+         * run flush_dcache_page().
+         */
+        if (rw == WRITE)
+                return lo_write_simple(lo, rq, pos);
+        else
+                return lo_read_simple(lo, rq, pos);
+}
+
 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
 {
         loff_t pos;
@@ -460,13 +544,13 @@ static int do_req_filebacked(struct loop_device *lo, struct request *rq)
                 else if (lo->transfer)
                         ret = lo_write_transfer(lo, rq, pos);
                 else
-                        ret = lo_write_simple(lo, rq, pos);
+                        ret = lo_rw_simple(lo, rq, pos, WRITE);
 
         } else {
                 if (lo->transfer)
                         ret = lo_read_transfer(lo, rq, pos);
                 else
-                        ret = lo_read_simple(lo, rq, pos);
+                        ret = lo_rw_simple(lo, rq, pos, READ);
         }
 
         return ret;
@@ -1570,6 +1654,12 @@ static int loop_queue_rq(struct blk_mq_hw_ctx *hctx,
         if (lo->lo_state != Lo_bound)
                 return -EIO;
 
+        if (lo->use_dio && !(cmd->rq->cmd_flags & (REQ_FLUSH |
+                                        REQ_DISCARD)))
+                cmd->use_aio = true;
+        else
+                cmd->use_aio = false;
+
         queue_kthread_work(&lo->worker, &cmd->work);
 
         return BLK_MQ_RQ_QUEUE_OK;
@@ -1589,7 +1679,9 @@ static void loop_handle_cmd(struct loop_cmd *cmd)
  failed:
         if (ret)
                 cmd->rq->errors = -EIO;
-        blk_mq_complete_request(cmd->rq);
+        /* complete non-aio request */
+        if (!cmd->use_aio || ret)
+                blk_mq_complete_request(cmd->rq);
 }
 
 static void loop_queue_work(struct kthread_work *work)
diff --git a/drivers/block/loop.h b/drivers/block/loop.h
index d1de2217c09a..fb2237c73e61 100644
--- a/drivers/block/loop.h
+++ b/drivers/block/loop.h
@@ -69,6 +69,8 @@ struct loop_cmd {
         struct kthread_work work;
         struct request *rq;
         struct list_head list;
+        bool use_aio;           /* use AIO interface to handle I/O */
+        struct kiocb iocb;
 };
 
 /* Support for loadable transfer modules */