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

["zhangyi (F)" <yi.zhang@xxxxxxxxxx>]

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.

Signed-off-by: zhangyi (F) <yi.zhang@xxxxxxxxxx>
---
 drivers/md/dm-log-writes.c | 56 ++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 54 insertions(+), 2 deletions(-)

diff --git a/drivers/md/dm-log-writes.c b/drivers/md/dm-log-writes.c
index 9ea2b02..37088c7 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 +116,8 @@ struct log_writes_c {
 	struct list_head logging_blocks;
 	wait_queue_head_t wait;
 	struct task_struct *log_kthread;
+	bool submitting_super;
+	wait_queue_head_t wait_super;
 };
 
 struct pending_block {
@@ -180,6 +183,34 @@ 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;
+	unsigned long flags;
+
+	spin_lock_irqsave(&lc->blocks_lock, flags);
+	if (bio->bi_status) {
+		DMERR("Error writing super block, error=%d",
+		      bio->bi_status);
+		lc->logging_enabled = false;
+	}
+
+	WARN_ON(!lc->submitting_super);
+	lc->submitting_super = false;
+	spin_unlock_irqrestore(&lc->blocks_lock, flags);
+
+	/*
+	 * Wake up log-write kthread that previous super sector has
+	 * been written to disk.
+	 */
+	if (waitqueue_active(&lc->wait_super))
+		wake_up(&lc->wait_super);
+
+	bio_free_pages(bio);
+	put_io_block(lc);
+	bio_put(bio);
+}
+
 /*
  * Meant to be called if there is an error, it will free all the pages
  * associated with the block.
@@ -215,7 +246,8 @@ 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);
 
@@ -418,7 +450,25 @@ 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)) {
+	/*
+	 * Super sector should be writen in-order, or else the
+	 * nr_entries could be small than the real submitted entries.
+	 * So wait previous super sector submitted here.
+	 */
+	if (!lc->submitting_super)
+		goto write_super;
+
+	spin_lock_irq(&lc->blocks_lock);
+	if (!lc->submitting_super) {
+		spin_unlock_irq(&lc->blocks_lock);
+		goto write_super;
+	}
+	spin_unlock_irq(&lc->blocks_lock);
+	wait_event(lc->wait_super, !lc->submitting_super);
+write_super:
+	lc->submitting_super = true;
+	if (write_metadata(lc, &super, sizeof(super), NULL, 0,
+			   WRITE_LOG_SUPER_SECTOR)) {
 		DMERR("Couldn't write super");
 		return -1;
 	}
@@ -531,6 +581,7 @@ 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_waitqueue_head(&lc->wait_super);
 	atomic_set(&lc->io_blocks, 0);
 	atomic_set(&lc->pending_blocks, 0);
 
@@ -570,6 +621,7 @@ static int log_writes_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	lc->logging_enabled = true;
 	lc->end_sector = logdev_last_sector(lc);
 	lc->device_supports_discard = true;
+	lc->submitting_super = false;
 
 	ti->num_flush_bios = 1;
 	ti->flush_supported = true;
-- 
2.7.4

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel