All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH] dmeventd: remove memory leak
@ 2013-07-23  2:42 NeilBrown
  2013-07-31 20:28 ` Brassow Jonathan
  0 siblings, 1 reply; 2+ messages in thread
From: NeilBrown @ 2013-07-23  2:42 UTC (permalink / raw)
  To: lvm-devel


If e.g. a snapshot volume is repeatly added and removed, dmeventd will
leake memory.

This is because the 'timeout' thread is not detached and is never
joined.  So the resources it uses are never freed.

This patch changes _pthread_create_smallstack to set
PTHREAD_CREATE_DETACHED if the pthread pointer passed is NULL, and
changes _register_for_timeout to pass a NULL pointer rather than a
pointer to an unused variable.

Signed-off-by: NeilBrown <neilb@suse.de>

---

Customer found this memory leak.  It can be easily seen by e.g.

while true; do
   lvcreate -n snaptest -L 900m -s /dev/vgtest/lvtest
   sleep 1
   lvremove -f /dev/vgtest/snaptest
done

and watching /proc/`pid-of dmeventd`/maps

NeilBrown

 daemons/dmeventd/dmeventd.c |   12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

--- LVM2.2.02.84.orig/daemons/dmeventd/dmeventd.c
+++ LVM2.2.02.84/daemons/dmeventd/dmeventd.c
@@ -256,12 +256,20 @@ static struct dso_data *_alloc_dso_data(
 /* Create a device monitoring thread. */
 static int _pthread_create_smallstack(pthread_t *t, void *(*fun)(void *), void *arg)
 {
+	pthread_t tmp;
 	pthread_attr_t attr;
 	pthread_attr_init(&attr);
 	/*
 	 * We use a smaller stack since it gets preallocated in its entirety
 	 */
 	pthread_attr_setstacksize(&attr, THREAD_STACK_SIZE);
+	/*
+	 * If no-one will be waiting, we need to detach.
+	 */
+	if (!t) {
+		pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
+		t = &tmp;
+	}
 	return pthread_create(t, &attr, fun, arg);
 }
 
@@ -544,9 +552,7 @@ static int _register_for_timeout(struct
 	}
 
 	if (!_timeout_running) {
-		pthread_t timeout_id;
-
-		if (!(ret = -_pthread_create_smallstack(&timeout_id, _timeout_thread, NULL)))
+		if (!(ret = -_pthread_create_smallstack(NULL, _timeout_thread, NULL)))
 			_timeout_running = 1;
 	}
 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 828 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/lvm-devel/attachments/20130723/020f5a59/attachment.sig>

^ permalink raw reply	[flat|nested] 2+ messages in thread

* [PATCH] dmeventd: remove memory leak
  2013-07-23  2:42 [PATCH] dmeventd: remove memory leak NeilBrown
@ 2013-07-31 20:28 ` Brassow Jonathan
  0 siblings, 0 replies; 2+ messages in thread
From: Brassow Jonathan @ 2013-07-31 20:28 UTC (permalink / raw)
  To: lvm-devel

I've checked this in with some modification (git commit ID 5ca54c4).  However, there still seems to be a memory leak somewhere.  I've tried using valgrind to find it, but ended up chasing some non-issues.  :(

 brassow

On Jul 22, 2013, at 9:42 PM, NeilBrown wrote:

> 
> If e.g. a snapshot volume is repeatly added and removed, dmeventd will
> leake memory.
> 
> This is because the 'timeout' thread is not detached and is never
> joined.  So the resources it uses are never freed.
> 
> This patch changes _pthread_create_smallstack to set
> PTHREAD_CREATE_DETACHED if the pthread pointer passed is NULL, and
> changes _register_for_timeout to pass a NULL pointer rather than a
> pointer to an unused variable.
> 
> Signed-off-by: NeilBrown <neilb@suse.de>
> 
> ---
> 
> Customer found this memory leak.  It can be easily seen by e.g.
> 
> while true; do
>   lvcreate -n snaptest -L 900m -s /dev/vgtest/lvtest
>   sleep 1
>   lvremove -f /dev/vgtest/snaptest
> done
> 
> and watching /proc/`pid-of dmeventd`/maps
> 
> NeilBrown
> 
> daemons/dmeventd/dmeventd.c |   12 +++++++++---
> 1 file changed, 9 insertions(+), 3 deletions(-)
> 
> --- LVM2.2.02.84.orig/daemons/dmeventd/dmeventd.c
> +++ LVM2.2.02.84/daemons/dmeventd/dmeventd.c
> @@ -256,12 +256,20 @@ static struct dso_data *_alloc_dso_data(
> /* Create a device monitoring thread. */
> static int _pthread_create_smallstack(pthread_t *t, void *(*fun)(void *), void *arg)
> {
> +	pthread_t tmp;
> 	pthread_attr_t attr;
> 	pthread_attr_init(&attr);
> 	/*
> 	 * We use a smaller stack since it gets preallocated in its entirety
> 	 */
> 	pthread_attr_setstacksize(&attr, THREAD_STACK_SIZE);
> +	/*
> +	 * If no-one will be waiting, we need to detach.
> +	 */
> +	if (!t) {
> +		pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
> +		t = &tmp;
> +	}
> 	return pthread_create(t, &attr, fun, arg);
> }
> 
> @@ -544,9 +552,7 @@ static int _register_for_timeout(struct
> 	}
> 
> 	if (!_timeout_running) {
> -		pthread_t timeout_id;
> -
> -		if (!(ret = -_pthread_create_smallstack(&timeout_id, _timeout_thread, NULL)))
> +		if (!(ret = -_pthread_create_smallstack(NULL, _timeout_thread, NULL)))
> 			_timeout_running = 1;
> 	}
> 
> --
> lvm-devel mailing list
> lvm-devel at redhat.com
> https://www.redhat.com/mailman/listinfo/lvm-devel




^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2013-07-31 20:28 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-07-23  2:42 [PATCH] dmeventd: remove memory leak NeilBrown
2013-07-31 20:28 ` Brassow Jonathan

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.