Re: [PATCH v2 05/11] md: export helpers to stop sync_thread

From: Yu Kuai
Date: Thu Jan 25 2024 - 21:38:43 EST


Hi,

在 2024/01/25 15:57, Yu Kuai 写道:
Hi,

在 2024/01/25 15:51, Xiao Ni 写道:
On Wed, Jan 24, 2024 at 5:19 PM Yu Kuai <yukuai3@xxxxxxxxxx> wrote:

The new heleprs will be used in dm-raid in later patches to fix
regressions and prevent calling md_reap_sync_thread() directly.

Signed-off-by: Yu Kuai <yukuai3@xxxxxxxxxx>
---
  drivers/md/md.c | 41 +++++++++++++++++++++++++++++++++++++----
  drivers/md/md.h |  3 +++
  2 files changed, 40 insertions(+), 4 deletions(-)

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 6c5d0a372927..90cf31b53804 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -4915,30 +4915,63 @@ static void stop_sync_thread(struct mddev *mddev, bool locked, bool check_seq)
                 mddev_lock_nointr(mddev);
  }

-static void idle_sync_thread(struct mddev *mddev)
+void md_idle_sync_thread(struct mddev *mddev)
  {
+       lockdep_assert_held(mddev->reconfig_mutex);
+

Hi Kuai

There is a building error. It should give a pointer to
lockdep_assert_held. And same with the other two places in this patch.

Yes, I forgot that I disabled all the debug config in order to let tests
finish quickly.

I enabled these debuging conifg, and turns out this patch has some
problem, see below.

Thanks for the notince, will fix this in v3.

Thanks,
Kuai


Regards
Xiao

         mutex_lock(&mddev->sync_mutex);
         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+       stop_sync_thread(mddev, true, true);
+       mutex_unlock(&mddev->sync_mutex);
+}
+EXPORT_SYMBOL_GPL(md_idle_sync_thread);
+
+void md_frozen_sync_thread(struct mddev *mddev)
+{
+       lockdep_assert_held(mddev->reconfig_mutex);
+
+       mutex_lock(&mddev->sync_mutex);
+       set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+       stop_sync_thread(mddev, true, false);

stop_sync_thread() may release 'reconfig_mutex' and grab it again, hence
it's not right to grab 'sync_mutex' before 'reconfig_mutex'.

Since 'sync_mutex' is introduced to serialize sysfs api 'sync_action'
writers, while is not involved for dm-raid. I'll remove 'sync_mutex' for
the new helper in v3.

Thanks,
Kuai

+       mutex_unlock(&mddev->sync_mutex);
+}
+EXPORT_SYMBOL_GPL(md_frozen_sync_thread);

+void md_unfrozen_sync_thread(struct mddev *mddev)
+{
+       lockdep_assert_held(mddev->reconfig_mutex);
+
+       mutex_lock(&mddev->sync_mutex);
+       clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+       set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
+       md_wakeup_thread(mddev->thread);
+       sysfs_notify_dirent_safe(mddev->sysfs_action);
+       mutex_unlock(&mddev->sync_mutex);
+}
+EXPORT_SYMBOL_GPL(md_unfrozen_sync_thread);
+
+static void idle_sync_thread(struct mddev *mddev)
+{
         if (mddev_lock(mddev)) {
                 mutex_unlock(&mddev->sync_mutex);
                 return;
         }

+       mutex_lock(&mddev->sync_mutex);
+       clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
         stop_sync_thread(mddev, false, true);
         mutex_unlock(&mddev->sync_mutex);
  }

  static void frozen_sync_thread(struct mddev *mddev)
  {
-       mutex_lock(&mddev->sync_mutex);
-       set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
-
         if (mddev_lock(mddev)) {
                 mutex_unlock(&mddev->sync_mutex);
                 return;
         }

+       mutex_lock(&mddev->sync_mutex);
+       set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
         stop_sync_thread(mddev, false, false);
         mutex_unlock(&mddev->sync_mutex);
  }
diff --git a/drivers/md/md.h b/drivers/md/md.h
index 8d881cc59799..437ab70ce79b 100644
--- a/drivers/md/md.h
+++ b/drivers/md/md.h
@@ -781,6 +781,9 @@ extern void md_rdev_clear(struct md_rdev *rdev);
  extern void md_handle_request(struct mddev *mddev, struct bio *bio);
  extern int mddev_suspend(struct mddev *mddev, bool interruptible);
  extern void mddev_resume(struct mddev *mddev);
+extern void md_idle_sync_thread(struct mddev *mddev);
+extern void md_frozen_sync_thread(struct mddev *mddev);
+extern void md_unfrozen_sync_thread(struct mddev *mddev);

  extern void md_reload_sb(struct mddev *mddev, int raid_disk);
  extern void md_update_sb(struct mddev *mddev, int force);
--
2.39.2


.


.