[PATCH] block: loop: mark bvec as ITER_BVEC_FLAG_NO_REF

[PATCH] block: loop: mark bvec as ITER_BVEC_FLAG_NO_REF

[Ming Lei]

loop is one block device, for any bio submitted to this device,
the upper layer does guarantee that pages added to loop's bio won't
go away when the bio is in-flight.

So mark loop's bvec as ITER_BVEC_FLAG_NO_REF then get_page/put_page
can be saved for serving loop's IO.

Cc: linux-fsdevel@vger.kernel.org
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 drivers/block/loop.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index bf1c61cab8eb..3d25611eebdc 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -264,12 +264,20 @@ lo_do_transfer(struct loop_device *lo, int cmd,
 	return ret;
 }
 
+static inline void loop_iov_iter_bvec(struct iov_iter *i,
+		unsigned int direction, const struct bio_vec *bvec,
+		unsigned long nr_segs, size_t count)
+{
+	iov_iter_bvec(i, direction, bvec, nr_segs, count);
+	i->type |= ITER_BVEC_FLAG_NO_REF;
+}
+
 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
 {
 	struct iov_iter i;
 	ssize_t bw;
 
-	iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
+	loop_iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
 
 	file_start_write(file);
 	bw = vfs_iter_write(file, &i, ppos, 0);
@@ -347,7 +355,7 @@ static int lo_read_simple(struct loop_device *lo, struct request *rq,
 	ssize_t len;
 
 	rq_for_each_segment(bvec, rq, iter) {
-		iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
+		loop_iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
 		len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
 		if (len < 0)
 			return len;
@@ -388,7 +396,7 @@ static int lo_read_transfer(struct loop_device *lo, struct request *rq,
 		b.bv_offset = 0;
 		b.bv_len = bvec.bv_len;
 
-		iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
+		loop_iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
 		len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
 		if (len < 0) {
 			ret = len;
@@ -555,7 +563,7 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
 	}
 	atomic_set(&cmd->ref, 2);
 
-	iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
+	loop_iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
 	iter.iov_offset = offset;
 
 	cmd->iocb.ki_pos = pos;
-- 
2.9.5

Re: [PATCH] block: loop: mark bvec as ITER_BVEC_FLAG_NO_REF

[Jens Axboe]

On 3/27/19 9:05 PM, Ming Lei wrote:
> loop is one block device, for any bio submitted to this device,
> the upper layer does guarantee that pages added to loop's bio won't
> go away when the bio is in-flight.
> 
> So mark loop's bvec as ITER_BVEC_FLAG_NO_REF then get_page/put_page
> can be saved for serving loop's IO.

Curious if this is a noticable win on loop? In any case, looks
fine to me, I'll queue it up for 5.2.

-- 
Jens Axboe

Re: [PATCH] block: loop: mark bvec as ITER_BVEC_FLAG_NO_REF

[Christoph Hellwig]

> +	iov_iter_bvec(i, direction, bvec, nr_segs, count);
> +	i->type |= ITER_BVEC_FLAG_NO_REF;
> +}

What about simply providing a iov_iter_bvec_noref in core code and
also using that in io_uring?

Otherwise this looks fine to me.

Re: [PATCH] block: loop: mark bvec as ITER_BVEC_FLAG_NO_REF

[Jens Axboe]

On 3/28/19 9:02 AM, Christoph Hellwig wrote:
>> +	iov_iter_bvec(i, direction, bvec, nr_segs, count);
>> +	i->type |= ITER_BVEC_FLAG_NO_REF;
>> +}
> 
> What about simply providing a iov_iter_bvec_noref in core code and
> also using that in io_uring?
> 
> Otherwise this looks fine to me.

I'd be fine with that helper.

-- 
Jens Axboe