[dm-devel] [PATCH v2] dm log writes: make sure the log super sectors are written in order

[zhangyi (F)]

Currently, although we submit super bios in log-write thread orderly
(the super.nr_entries is incremented by each logged entry), the
submit_bio() cannot make sure that each super sector is written to log
device in order. So the submitting bio of each super sector may be
out-of-order, and then the final nr_entries maybe small than the real
entries submitted.

This problem can be reproduced by the xfstests generic/455 with ext4,
which may complained below after running the test:

  QA output created by 455
 -Silence is golden
 +mark 'end' does not exist

This patch serialize submitting super secotrs to make sure each super
sectors are written to log disk in order, so that we can prevent the
latest nr_entries be rewritten by some old super bios.

Signed-off-by: zhangyi (F) <yi.zhang at huawei.com>
Suggested-by: Josef Bacik <josef at toxicpanda.com>
---
Changes since v1:
 - Switch to use completion instead of wait_queue to synchronize
   submitting super bios.

 drivers/md/dm-log-writes.c | 25 +++++++++++++++++++++++--
 1 file changed, 23 insertions(+), 2 deletions(-)

diff --git a/drivers/md/dm-log-writes.c b/drivers/md/dm-log-writes.c
index 9ea2b02..4854246 100644
--- a/drivers/md/dm-log-writes.c
+++ b/drivers/md/dm-log-writes.c
@@ -60,6 +60,7 @@
 
 #define WRITE_LOG_VERSION 1ULL
 #define WRITE_LOG_MAGIC 0x6a736677736872ULL
+#define WRITE_LOG_SUPER_SECTOR 0
 
 /*
  * The disk format for this is braindead simple.
@@ -115,6 +116,7 @@ struct log_writes_c {
 	struct list_head logging_blocks;
 	wait_queue_head_t wait;
 	struct task_struct *log_kthread;
+	struct completion super_done;
 };
 
 struct pending_block {
@@ -180,6 +182,15 @@ static void log_end_io(struct bio *bio)
 	bio_put(bio);
 }
 
+static void log_end_super(struct bio *bio)
+{
+	struct log_writes_c *lc = bio->bi_private;
+
+	/* Wake up log-write kthread that super has been written */
+	complete(&lc->super_done);
+	log_end_io(bio);
+}
+
 /*
  * Meant to be called if there is an error, it will free all the pages
  * associated with the block.
@@ -215,7 +226,8 @@ static int write_metadata(struct log_writes_c *lc, void *entry,
 	bio->bi_iter.bi_size = 0;
 	bio->bi_iter.bi_sector = sector;
 	bio_set_dev(bio, lc->logdev->bdev);
-	bio->bi_end_io = log_end_io;
+	bio->bi_end_io = (sector == WRITE_LOG_SUPER_SECTOR) ?
+			  log_end_super : log_end_io;
 	bio->bi_private = lc;
 	bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 
@@ -418,11 +430,19 @@ static int log_super(struct log_writes_c *lc)
 	super.nr_entries = cpu_to_le64(lc->logged_entries);
 	super.sectorsize = cpu_to_le32(lc->sectorsize);
 
-	if (write_metadata(lc, &super, sizeof(super), NULL, 0, 0)) {
+	if (write_metadata(lc, &super, sizeof(super), NULL, 0,
+			   WRITE_LOG_SUPER_SECTOR)) {
 		DMERR("Couldn't write super");
 		return -1;
 	}
 
+	/*
+	 * Super sector should be writen in-order, or else the
+	 * nr_entries could be rewritten by the old bio and small
+	 * than the real submitted entries.
+	 */
+	wait_for_completion_io(&lc->super_done);
+
 	return 0;
 }
 
@@ -531,6 +551,7 @@ static int log_writes_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	INIT_LIST_HEAD(&lc->unflushed_blocks);
 	INIT_LIST_HEAD(&lc->logging_blocks);
 	init_waitqueue_head(&lc->wait);
+	init_completion(&lc->super_done);
 	atomic_set(&lc->io_blocks, 0);
 	atomic_set(&lc->pending_blocks, 0);
 
-- 
2.7.4