If md_write_start() finds that ->writes_pending is non-zero, it
should be able to avoid most of the other checks.
To ensure that a non-zero ->writes_pending does mean that other checks
have completed, move it down until after ->in_sync is known to be
clear. To avoid races with places like array_state_store() which
possible sets ->in_sync, we need to increment ->write_pending inside
the locked region. As ->writes_pending is now incremented *after*
->in_sync is tested, we must always take the spin_lock, but only if
->writes_pending was found to be zero.
If ->writes_pending is found to be non-zero, we still need to wait it
MD_CHANGE_PENDING is true.
In the common case, md_write_start() will now only
- check if data_dir is WRITE
- increment ->writes_pending
- check MD_CHANGE_PENDING is cleared.
Signed-off-by: NeilBrown <neilb@xxxxxxxx>
---
drivers/md/md.c | 32 ++++++++++++++++----------------
1 file changed, 16 insertions(+), 16 deletions(-)
diff --git a/drivers/md/md.c b/drivers/md/md.c
index 1f1c7f007b68..2f21f6c7156f 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -7686,20 +7686,18 @@ void md_write_start(struct mddev *mddev, struct bio *bi)
int did_change = 0;
if (bio_data_dir(bi) != WRITE)
return;
-
- BUG_ON(mddev->ro == 1);
- if (mddev->ro == 2) {
- /* need to switch to read/write */
- mddev->ro = 0;
- set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
- md_wakeup_thread(mddev->thread);
- md_wakeup_thread(mddev->sync_thread);
- did_change = 1;
- }
- atomic_inc(&mddev->writes_pending);
- if (mddev->safemode == 1)
- mddev->safemode = 0;
- if (mddev->in_sync) {
+ if (!atomic_inc_not_zero(&mddev->writes_pending)) {
+ BUG_ON(mddev->ro == 1);
+ if (mddev->ro == 2) {
+ /* need to switch to read/write */
+ mddev->ro = 0;
+ set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
+ md_wakeup_thread(mddev->thread);
+ md_wakeup_thread(mddev->sync_thread);
+ did_change = 1;
+ }
+ if (mddev->safemode == 1)
+ mddev->safemode = 0;
spin_lock(&mddev->lock);
if (mddev->in_sync) {
mddev->in_sync = 0;
@@ -7708,10 +7706,12 @@ void md_write_start(struct mddev *mddev, struct bio *bi)
md_wakeup_thread(mddev->thread);
did_change = 1;
}
+ atomic_inc(&mddev->writes_pending);
spin_unlock(&mddev->lock);
+
+ if (did_change)
+ sysfs_notify_dirent_safe(mddev->sysfs_state);
}
- if (did_change)
- sysfs_notify_dirent_safe(mddev->sysfs_state);
wait_event(mddev->sb_wait,
!test_bit(MD_CHANGE_PENDING, &mddev->flags));
}
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
