ext4: implement cgroup writeback support

Hello,

This patchset implements cgroup writeback support for ext4, which is enabled for writeback and ordered data modes. This patchset contains the following two patches.

0001-ext4-replace-ext4_io_submit-io_op-with-io_wbc.patch 0002-ext4-implement-cgroup-writeback-support.patch

0001 is a prep patch which replaces io_submit->io_op w/ ->io_wbc. 0002 implements cgroup writeback support.

This patchset is on top of

[1] block/for-4.2/writeback
+[2] [PATCHSET] cgroup, writeback: misc updates for cgroup writeback support

and available on in the following git branch.

git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git review-cgroup-writeback-ext4

diffstat follows. Thanks.

 fs/ext4/ext4.h    |    2 +-
 fs/ext4/page-io.c |    8 ++++++--
 fs/ext4/super.c   |    2 ++
 3 files changed, 9 insertions(+), 3 deletions(-)

--
tejun

[1] git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux-block.git for-4.2/writeback
[2] http://lkml.kernel.org/g/1434146254-26220-1-git-send-email-tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org

ext4_io_submit_init() takes the pointer to writeback_control to test its sync_mode and determine between WRITE and WRITE_SYNC and records the result in ->io_op. This patch makes it record the pointer directly and moves the test to ext4_io_submit().

This doesn't cause any noticeable differences now but having writeback_control available throughout IO submission path will be depended upon by the planned cgroup writeback support.

---
 fs/ext4/ext4.h    | 2 +-
 fs/ext4/page-io.c | 6 ++++--
 2 files changed, 5 insertions(+), 3 deletions(-)

index 009a059..74a4923 100644
@@ -191,7 +191,7 @@ typedef struct ext4_io_end {
 } ext4_io_end_t;
 
 struct ext4_io_submit {
-	int			io_op;
+	struct writeback_control *io_wbc;
 	struct bio		*io_bio;
 	ext4_io_end_t		*io_end;
 	sector_t		io_next_block;

index c5d81e8..3f80cb2 100644
@@ -357,8 +357,10 @@ void ext4_io_submit(struct ext4_io_submit *io)
 	struct bio *bio = io->io_bio;
 
 	if (bio) {
+		int io_op = io->io_wbc->sync_mode == WB_SYNC_ALL ?
+			    WRITE_SYNC : WRITE;
 		bio_get(io->io_bio);
-		submit_bio(io->io_op, io->io_bio);
+		submit_bio(io_op, io->io_bio);
 		bio_put(io->io_bio);
 	}
 	io->io_bio = NULL;

@@ -367,7 +369,7 @@ void ext4_io_submit(struct ext4_io_submit *io)
 void ext4_io_submit_init(struct ext4_io_submit *io,
 			 struct writeback_control *wbc)
 {
-	io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
+	io->io_wbc = wbc;
 	io->io_bio = NULL;
 	io->io_end = NULL;
 }

2.4.2

For ordered and writeback data modes, all data IOs go through ext4_io_submit. This patch adds cgroup writeback support by invoking wbc_init_bio() from io_submit_init_bio() and wbc_account_io() in io_submit_add_bh(). Journal data which is written by jbd2 worker is left alone by this patch and will always be written out from the root cgroup.

ext4_fill_super() is updated to set MS_CGROUPWB when data mode is either ordered or writeback. In journaled data mode, most IOs become synchronous through the journal and enabling cgroup writeback support doesn't make much sense or difference. Journaled data mode is left alone.

Lightly tested with sequential data write workload. Behaves as expected.

---
 fs/ext4/page-io.c | 2 ++
 fs/ext4/super.c   | 2 ++
 2 files changed, 4 insertions(+)

index 3f80cb2..c56ba7b 100644
@@ -383,6 +383,7 @@ static int io_submit_init_bio(struct ext4_io_submit *io,
 	bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
 	if (!bio)
 		return -ENOMEM;
+	wbc_init_bio(io->io_wbc, bio);
 	bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
 	bio->bi_bdev = bh->b_bdev;
 	bio->bi_end_io = ext4_end_bio;

@@ -411,6 +412,7 @@ static int io_submit_add_bh(struct ext4_io_submit *io,
 	ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
 	if (ret != bh->b_size)
 		goto submit_and_retry;
+	wbc_account_io(io->io_wbc, page, bh->b_size);
 	io->io_next_block++;
 	return 0;
 }

index 56b8bb7..a9a3de3 100644
@@ -3623,6 +3623,8 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 		}
 		if (test_opt(sb, DELALLOC))
 			clear_opt(sb, DELALLOC);
+	} else {
+		sb->s_flags |= MS_CGROUPWB;
 	}
 
 	sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |

2.4.2