On 01/03/2018 03:03 PM, Coly Li wrote:
> Delayed worker dc->writeback_rate_update and kernel thread
> dc->writeback_thread reference cache set data structure in their routine,
> Therefor, before they are stopped, cache set should not be release. Other-
> wise, NULL pointer deference will be triggered.
>
> Currenly delayed worker dc->writeback_rate_update and kernel thread
> dc->writeback_thread are stopped in cached_dev_free(). When cache set is
> retiring by too many I/O errors, cached_dev_free() is called when refcount
> of bcache device's closure (disk.cl) reaches 0. In most of cases, last
> refcount of disk.cl is dropped in last line of cached_dev_detach_finish().
> But in cached_dev_detach_finish() before calling closure_put(&dc->disk.cl),
> bcache_device_detach() is called, and inside bcache_device_detach()
> refcount of cache_set->caching is dropped by closure_put(&d->c->caching).
>
> It is very probably this is the last refcount of this closure, so routine
> cache_set_flush() will be called (it is set in __cache_set_unregister()),
> and its parent closure cache_set->cl may also drop its last refcount and
> cache_set_free() is called too. In cache_set_free() the last refcount of
> cache_set->kobj is dropped and then bch_cache_set_release() is called. Now
> in bch_cache_set_release(), the memory of struct cache_set is freeed.
>
> bch_cache_set_release() is called before cached_dev_free(), then there is a
> time window after cache set memory freed and before dc->writeback_thread
> and dc->writeback_rate_update stopped, if one of them is scheduled to run,
> a NULL pointer deference will be triggered.
>
> This patch fixes the above problem by stopping dc->writeback_thread and
> dc->writeback_rate_update earlier in bcache_device_detach() before calling
> closure_put(&d->c->caching). Because cancel_delayed_work_sync() and
> kthread_stop() are synchronized operations, we can make sure cache set
> is available when the delayed work and kthread are stopping.
>
> Because cached_dev_free() can also be called by writing 1 to sysfs file
> /sys/block/bcache<N>/bcache/stop, this code path may not call
> bcache_device_detach() if d-c is NULL. So stopping dc->writeback_thread
> and dc->writeback_rate_update in cached_dev_free() is still necessary. In
> order to avoid stop them twice, dc->rate_update_canceled is added to
> indicate dc->writeback_rate_update is canceled, and dc->writeback_thread
> is set to NULL to indicate it is stopped.
>
> Signed-off-by: Coly Li <colyli@xxxxxxx>
> ---
> drivers/md/bcache/bcache.h | 1 +
> drivers/md/bcache/super.c | 21 +++++++++++++++++++--
> drivers/md/bcache/writeback.c | 1 +
> 3 files changed, 21 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h
> index 83c569942bd0..395b87942a2f 100644
> --- a/drivers/md/bcache/bcache.h
> +++ b/drivers/md/bcache/bcache.h
> @@ -322,6 +322,7 @@ struct cached_dev {
>
> struct bch_ratelimit writeback_rate;
> struct delayed_work writeback_rate_update;
> + bool rate_update_canceled;
>
> /*
> * Internal to the writeback code, so read_dirty() can keep track of
> diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
> index 5401d2356aa3..8912be4165c5 100644
> --- a/drivers/md/bcache/super.c
> +++ b/drivers/md/bcache/super.c
> @@ -696,8 +696,20 @@ static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
>
> static void bcache_device_detach(struct bcache_device *d)
> {
> + struct cached_dev *dc;
> +
> lockdep_assert_held(&bch_register_lock);
>
> + dc = container_of(d, struct cached_dev, disk);
> + if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
> + kthread_stop(dc->writeback_thread);
> + dc->writeback_thread = NULL;
> + }
> + if (!dc->rate_update_canceled) {
> + cancel_delayed_work_sync(&dc->writeback_rate_update);
> + dc->rate_update_canceled = true;
> + }
> +
> if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
> struct uuid_entry *u = d->c->uuids + d->id;
>
> @@ -1071,9 +1083,14 @@ static void cached_dev_free(struct closure *cl)
> {
> struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
>
> - cancel_delayed_work_sync(&dc->writeback_rate_update);
> - if (!IS_ERR_OR_NULL(dc->writeback_thread))
> + if (!dc->rate_update_canceled) {
> + cancel_delayed_work_sync(&dc->writeback_rate_update);
> + dc->rate_update_canceled = true;
> + }
> + if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
> kthread_stop(dc->writeback_thread);
> + dc->writeback_thread = NULL;
> + }
> if (dc->writeback_write_wq)
> destroy_workqueue(dc->writeback_write_wq);
>
> diff --git a/drivers/md/bcache/writeback.c b/drivers/md/bcache/writeback.c
> index 745d9b2a326f..ab2ac3d72393 100644
> --- a/drivers/md/bcache/writeback.c
> +++ b/drivers/md/bcache/writeback.c
> @@ -548,6 +548,7 @@ void bch_cached_dev_writeback_init(struct cached_dev *dc)
> dc->writeback_rate_i_term_inverse = 10000;
>
> INIT_DELAYED_WORK(&dc->writeback_rate_update, update_writeback_rate);
> + dc->rate_update_canceled = false;
> }
>
> int bch_cached_dev_writeback_start(struct cached_dev *dc)
>
Hehe. Just as I said in the comment to the previous patch.
I would suggest merge this and the previous patch :-)
But in general, I don't think you need 'rate_update_canceled'.
cancel_delayed_work_sync() will be a no-op if no work item has been
scheduled.
Cheers,
Hannes
--
Dr. Hannes Reinecke Teamlead Storage & Networking
hare@xxxxxxx +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: F. Imendörffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG Nürnberg)
