stable.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] memcg: don't initialize kmem-cache destroying work for root caches
@ 2013-05-22  8:09 Andrey Vagin
  0 siblings, 0 replies; 6+ messages in thread
From: Andrey Vagin @ 2013-05-22  8:09 UTC (permalink / raw)
  To: Andrew Morton
  Cc: linux-kernel, linux-mm, cgroups, Andrey Vagin, stable,
	Konstantin Khlebnikov, Glauber Costa, Johannes Weiner,
	Balbir Singh, KAMEZAWA Hiroyuki

struct memcg_cache_params has a union. Different parts of this union are
used for root and non-root caches. A part with destroying work is used only
for non-root caches.

[  115.096202] BUG: unable to handle kernel paging request at 0000000fffffffe0
[  115.096785] IP: [<ffffffff8116b641>] kmem_cache_alloc+0x41/0x1f0
[  115.097024] PGD 7ace1067 PUD 0
[  115.097024] Oops: 0000 [#4] SMP
[  115.097024] Modules linked in: netlink_diag af_packet_diag udp_diag tcp_diag inet_diag unix_diag ip6table_filter ip6_tables i2c_piix4 virtio_net virtio_balloon microcode i2c_core pcspkr floppy
[  115.097024] CPU: 0 PID: 1929 Comm: lt-vzctl Tainted: G      D      3.10.0-rc1+ #2
[  115.097024] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[  115.097024] task: ffff88007b5aaee0 ti: ffff88007bf0c000 task.ti: ffff88007bf0c000
[  115.097024] RIP: 0010<ffffffff8116b641>]  [<ffffffff8116b641>] kmem_cache_alloc+0x41/0x1f0
[  115.097024] RSP: 0018:ffff88007bf0de68  EFLAGS: 00010202
[  115.097024] RAX: 0000000fffffffe0 RBX: 00007fff4014f200 RCX: 0000000000000300
[  115.097024] RDX: 0000000000000005 RSI: 00000000000000d0 RDI: ffff88007d001300
[  115.097024] RBP: ffff88007bf0dea8 R08: 00007f849c3141b7 R09: ffffffff8118e100
[  115.097024] R10: 0000000000000001 R11: 0000000000000246 R12: 00000000000000d0
[  115.097024] R13: 0000000fffffffe0 R14: ffff88007d001300 R15: 0000000000001000
[  115.097024] FS:  00007f849cbb8b40(0000) GS:ffff88007fc00000(0000) knlGS:0000000000000000
[  115.097024] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  115.097024] CR2: 0000000fffffffe0 CR3: 000000007bc38000 CR4: 00000000000006f0
[  115.097024] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  115.097024] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[  115.097024] Stack:
[  115.097024]  ffffffff8118e100 ffffffff81149ea1 0000000000000008 00007fff4014f200
[  115.097024]  00007fff4014f200 0000000000000000 0000000000000000 0000000000001000
[  115.097024]  ffff88007bf0dee8 ffffffff8118e100 ffff880037598e00 00007fff4014f200
[  115.097024] Call Trace:
[  115.097024]  [<ffffffff8118e100>] ? getname_flags.part.34+0x30/0x140
[  115.097024]  [<ffffffff81149ea1>] ? vma_rb_erase+0x121/0x210
[  115.097024]  [<ffffffff8118e100>] getname_flags.part.34+0x30/0x140
[  115.097024]  [<ffffffff8118e248>] getname+0x38/0x60
[  115.097024]  [<ffffffff81181d55>] do_sys_open+0xc5/0x1e0
[  115.097024]  [<ffffffff81181e92>] SyS_open+0x22/0x30
[  115.097024]  [<ffffffff8161cb82>] system_call_fastpath+0x16/0x1b
[  115.097024] Code: f4 53 48 83 ec 18 8b 05 8e 53 b7 00 4c 8b 4d 08 21 f0 a8 10 74 0d 4c 89 4d c0 e8 1b 76 4a 00 4c 8b 4d c0 e9 92 00 00 00 4d 89 f5 <4d> 8b 45 00 65 4c 03 04 25 48 cd 00 00 49 8b 50 08 4d 8b 38 49
[  115.097024] RIP  [<ffffffff8116b641>] kmem_cache_alloc+0x41/0x1f0
[  115.097024]  RSP <ffff88007bf0de68>
[  115.097024] CR2: 0000000fffffffe0
[  115.121352] ---[ end trace 16bb8e8408b97d0e ]---

Cc: stable@vger.kernel.org # 3.9
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Konstantin Khlebnikov <khlebnikov@openvz.org>
Cc: Glauber Costa <glommer@parallels.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrey Vagin <avagin@openvz.org>
---
 mm/memcontrol.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index cb1c9de..764b9e4 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -3141,8 +3141,6 @@ int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
 			return -ENOMEM;
 		}
 
-		INIT_WORK(&s->memcg_params->destroy,
-				kmem_cache_destroy_work_func);
 		s->memcg_params->is_root_cache = true;
 
 		/*
-- 
1.8.1.4


^ permalink raw reply related	[flat|nested] 6+ messages in thread

* [PATCH] memcg: don't initialize kmem-cache destroying work for root caches
@ 2013-08-05 16:09 Andrey Vagin
  2013-08-05 20:05 ` Andrew Morton
  0 siblings, 1 reply; 6+ messages in thread
From: Andrey Vagin @ 2013-08-05 16:09 UTC (permalink / raw)
  To: linux-mm
  Cc: cgroups, linux-kernel, Andrey Vagin, Glauber Costa,
	Johannes Weiner, Michal Hocko, Balbir Singh, KAMEZAWA Hiroyuki,
	Andrew Morton, Konstantin Khlebnikov, stable

struct memcg_cache_params has a union. Different parts of this union
are used for root and non-root caches. A part with destroying work is
used only for non-root caches.

I fixed the same problem in another place v3.9-rc1-16204-gf101a94, but
didn't notice this one.

Cc: Glauber Costa <glommer@openvz.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Konstantin Khlebnikov <khlebnikov@openvz.org>
Cc: <stable@vger.kernel.org>    [3.9.x]
Signed-off-by: Andrey Vagin <avagin@openvz.org>
---
 mm/memcontrol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index c290a1c..c5792a5 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -3195,11 +3195,11 @@ int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
 	if (!s->memcg_params)
 		return -ENOMEM;
 
-	INIT_WORK(&s->memcg_params->destroy,
-			kmem_cache_destroy_work_func);
 	if (memcg) {
 		s->memcg_params->memcg = memcg;
 		s->memcg_params->root_cache = root_cache;
+		INIT_WORK(&s->memcg_params->destroy,
+				kmem_cache_destroy_work_func);
 	} else
 		s->memcg_params->is_root_cache = true;
 
-- 
1.8.3.1


^ permalink raw reply related	[flat|nested] 6+ messages in thread

* Re: [PATCH] memcg: don't initialize kmem-cache destroying work for root caches
  2013-08-05 16:09 Andrey Vagin
@ 2013-08-05 20:05 ` Andrew Morton
  2013-08-05 21:01   ` Andrew Vagin
  0 siblings, 1 reply; 6+ messages in thread
From: Andrew Morton @ 2013-08-05 20:05 UTC (permalink / raw)
  To: Andrey Vagin
  Cc: linux-mm, cgroups, linux-kernel, Glauber Costa, Johannes Weiner,
	Michal Hocko, Balbir Singh, KAMEZAWA Hiroyuki,
	Konstantin Khlebnikov, stable

On Mon,  5 Aug 2013 20:09:40 +0400 Andrey Vagin <avagin@openvz.org> wrote:

> struct memcg_cache_params has a union. Different parts of this union
> are used for root and non-root caches. A part with destroying work is
> used only for non-root caches.
> 
> I fixed the same problem in another place v3.9-rc1-16204-gf101a94, but
> didn't notice this one.
> 
> Cc: <stable@vger.kernel.org>    [3.9.x]

hm, why the cc:stable?

> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -3195,11 +3195,11 @@ int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
>  	if (!s->memcg_params)
>  		return -ENOMEM;
>  
> -	INIT_WORK(&s->memcg_params->destroy,
> -			kmem_cache_destroy_work_func);
>  	if (memcg) {
>  		s->memcg_params->memcg = memcg;
>  		s->memcg_params->root_cache = root_cache;
> +		INIT_WORK(&s->memcg_params->destroy,
> +				kmem_cache_destroy_work_func);
>  	} else
>  		s->memcg_params->is_root_cache = true;

So the bug here is that we'll scribble on some entries in
memcg_caches[].  Those scribbles may or may not be within the part of
that array which is actually used.  If there's code which expects
memcg_caches[] entries to be zeroed at initialisation then yes, we have
a problem.

But I rather doubt whether this bug was causing runtime problems?


Presently memcg_register_cache() allocates too much memory for the
memcg_caches[] array.  If that was fixed then this INIT_WORK() might
scribble into unknown memory, which is of course serious.

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] memcg: don't initialize kmem-cache destroying work for root caches
  2013-08-05 20:05 ` Andrew Morton
@ 2013-08-05 21:01   ` Andrew Vagin
  2013-08-05 21:16     ` Andrew Morton
  0 siblings, 1 reply; 6+ messages in thread
From: Andrew Vagin @ 2013-08-05 21:01 UTC (permalink / raw)
  To: Andrew Morton
  Cc: Andrey Vagin, linux-mm, cgroups, linux-kernel, Glauber Costa,
	Johannes Weiner, Michal Hocko, Balbir Singh, KAMEZAWA Hiroyuki,
	Konstantin Khlebnikov, stable

On Mon, Aug 05, 2013 at 01:05:30PM -0700, Andrew Morton wrote:
> On Mon,  5 Aug 2013 20:09:40 +0400 Andrey Vagin <avagin@openvz.org> wrote:
> 
> > struct memcg_cache_params has a union. Different parts of this union
> > are used for root and non-root caches. A part with destroying work is
> > used only for non-root caches.
> > 
> > I fixed the same problem in another place v3.9-rc1-16204-gf101a94, but
> > didn't notice this one.
> > 
> > Cc: <stable@vger.kernel.org>    [3.9.x]
> 
> hm, why the cc:stable?

Because this patch fixes the kernel panic:

[   46.848187] BUG: unable to handle kernel paging request at 000000fffffffeb8
[   46.849026] IP: [<ffffffff811a484c>] kmem_cache_destroy_memcg_children+0x6c/0xc0
[   46.849092] PGD 0
[   46.849092] Oops: 0000 [#1] SMP
[   46.849092] Modules linked in: vzethdev vznetdev pio_direct pfmt_raw pfmt_ploop1 ploop simfs ipt_MASQUERADE nf_conntrack_netbios_ns nf_conntrack_broadcast ip6table_mangle ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 iptable_nat nf_nat_ipv4 nf_nat iptable_mangle nf_conntrack_ipv4 nf_defrag_ipv4 xt_conntrack nf_conntrack ebtable_filter ebtables ip6table_filter ip6_tables vzevent microcode joydev pcspkr virtio_balloon i2c_piix4 i2c_core virtio_net virtio_blk floppy
[   46.849092] CPU 0
[   46.849092] Pid: 6, comm: kworker/u:0 ve: 0 Not tainted 3.9.4+ #42 ovz.2.4 Red Hat KVM
[   46.849092] RIP: 0010:[<ffffffff811a484c>]  [<ffffffff811a484c>] kmem_cache_destroy_memcg_children+0x6c/0xc0
[   46.849092] RSP: 0018:ffff88007c7dfcd8  EFLAGS: 00010206
[   46.849092] RAX: ffff88007b65f180 RBX: 000000fffffffe00 RCX: 0000000000000004
[   46.849092] RDX: 0000000000000005 RSI: ffff88007fc17cc8 RDI: ffffffff81c5e5a0
[   46.849092] RBP: ffff88007c7dfcf8 R08: ffffea0001ee1b20 R09: 0000000000000000
[   46.849092] R10: ffff88007fbe5fe0 R11: 0000000000000000 R12: 0000000000000005
[   46.849092] R13: ffff88007b8c2400 R14: 0000000000000000 R15: ffff88007c008005
[   46.849092] FS:  0000000000000000(0000) GS:ffff88007fc00000(0000) knlGS:0000000000000000
[   46.849092] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[   46.849092] CR2: 000000fffffffeb8 CR3: 00000000375a3000 CR4: 00000000000006f0
[   46.849092] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   46.849092] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[   46.849092] Process kworker/u:0 (pid: 6, ve: 0, threadinfo ffff88007c7de000, task ffff88007c7e8000)
[   46.849092] Stack:
[   46.849092]  ffff88007bc90870 ffff88007b8c2400 ffff88007bc90000 ffff88007bc90870
[   46.849092]  ffff88007c7dfd18 ffffffff81166a14 0000000080000003 ffff88007bc90000
[   46.849092]  ffff88007c7dfd48 ffffffffa007b3c8 ffff88007c7dfd48 ffff88007bc90000
[   46.849092] Call Trace:
[   46.849092]  [<ffffffff81166a14>] kmem_cache_destroy+0x14/0xf0
[   46.849092]  [<ffffffffa007b3c8>] nf_conntrack_cleanup_net+0xf8/0x120 [nf_conntrack]
[   46.849092]  [<ffffffffa007d221>] nf_conntrack_pernet_exit+0x41/0x50 [nf_conntrack]
[   46.849092]  [<ffffffff81536779>] ops_exit_list+0x39/0x60
[   46.849092]  [<ffffffff81536cfb>] cleanup_net+0xfb/0x200
[   46.849092]  [<ffffffff8107f43b>] process_one_work+0x17b/0x3d0
[   46.849092]  [<ffffffff81082639>] worker_thread+0x119/0x380
[   46.849092]  [<ffffffff81082520>] ? manage_workers+0x350/0x350
[   46.849092]  [<ffffffff8108796e>] kthread+0xce/0xe0
[   46.849092]  [<ffffffff810878a0>] ? kthread_freezable_should_stop+0x70/0x70
[   46.849092]  [<ffffffff8163faec>] ret_from_fork+0x7c/0xb0
[   46.849092]  [<ffffffff810878a0>] ? kthread_freezable_should_stop+0x70/0x70
[   46.849092] Code: fb 48 00 8b 1d 6e f0 1d 01 85 db 7e 4e 45 31 e4 0f 1f 80 00 00 00 00 49 8b 85 b8 00 00 00 49 63 d4 48 8b 5c d0 08 48 85 db 74 23 <48> 8b 83 b8 00 00 00 c6 40 28 00 48 8b bb b8 00 00 00 48 83 c7

This bug was added by v3.9-rc1-221-g15cf17d

> 
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -3195,11 +3195,11 @@ int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
> >  	if (!s->memcg_params)
> >  		return -ENOMEM;
> >  
> > -	INIT_WORK(&s->memcg_params->destroy,
> > -			kmem_cache_destroy_work_func);
> >  	if (memcg) {
> >  		s->memcg_params->memcg = memcg;
> >  		s->memcg_params->root_cache = root_cache;
> > +		INIT_WORK(&s->memcg_params->destroy,
> > +				kmem_cache_destroy_work_func);
> >  	} else
> >  		s->memcg_params->is_root_cache = true;
> 
> So the bug here is that we'll scribble on some entries in
> memcg_caches[].  Those scribbles may or may not be within the part of
> that array which is actually used.  If there's code which expects
> memcg_caches[] entries to be zeroed at initialisation then yes, we have
> a problem.

INIT_WORK() sets s->memcg_params->memcg_caches[5] to 0xfffffffe00.

Look at kmem_cache_destroy_memcg_children()

        for (i = 0; i < memcg_limited_groups_array_size; i++) {
                c = s->memcg_params->memcg_caches[i];
                if (!c)
                        continue;
...
		c->memcg_params->dead = false;

This code tries dereference 0xfffffffe00->dead and the kernel panics

> 
> But I rather doubt whether this bug was causing runtime problems?
> 
> 
> Presently memcg_register_cache() allocates too much memory for the
> memcg_caches[] array.  If that was fixed then this INIT_WORK() might
> scribble into unknown memory, which is of course serious.

Looks like you find another bug:

struct memcg_cache_params {
        bool is_root_cache;
        union {
                struct kmem_cache *memcg_caches[0];
                struct {
                        struct mem_cgroup *memcg;
                        struct list_head list;
                        struct kmem_cache *root_cache;
                        bool dead;
                        atomic_t nr_pages;
                        struct work_struct destroy;
                };
        };
};

The size of this strcture is 80 bytes, then look at memcg_register_cache()

size_t size = sizeof(struct memcg_cache_params);

if (!memcg_kmem_enabled())
        return 0;

if (!memcg)
        size += memcg_limited_groups_array_size * sizeof(void *);

s->memcg_params = kzalloc(size, GFP_KERNEL);

Actually it allocates too much memory. It allocates memory as if struct
memcg_cache_params would have been written without union.

Actually you already suggested to rework this code
https://lkml.org/lkml/2013/5/28/585

Thanks.

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] memcg: don't initialize kmem-cache destroying work for root caches
  2013-08-05 21:01   ` Andrew Vagin
@ 2013-08-05 21:16     ` Andrew Morton
  2013-08-05 21:41       ` Andrew Vagin
  0 siblings, 1 reply; 6+ messages in thread
From: Andrew Morton @ 2013-08-05 21:16 UTC (permalink / raw)
  To: Andrew Vagin
  Cc: Andrey Vagin, linux-mm, cgroups, linux-kernel, Glauber Costa,
	Johannes Weiner, Michal Hocko, Balbir Singh, KAMEZAWA Hiroyuki,
	Konstantin Khlebnikov, stable

On Tue, 6 Aug 2013 01:01:28 +0400 Andrew Vagin <avagin@parallels.com> wrote:

> On Mon, Aug 05, 2013 at 01:05:30PM -0700, Andrew Morton wrote:
> > On Mon,  5 Aug 2013 20:09:40 +0400 Andrey Vagin <avagin@openvz.org> wrote:
> > 
> > > struct memcg_cache_params has a union. Different parts of this union
> > > are used for root and non-root caches. A part with destroying work is
> > > used only for non-root caches.
> > > 
> > > I fixed the same problem in another place v3.9-rc1-16204-gf101a94, but
> > > didn't notice this one.
> > > 
> > > Cc: <stable@vger.kernel.org>    [3.9.x]
> > 
> > hm, why the cc:stable?
> 
> Because this patch fixes the kernel panic:
> 
> [   46.848187] BUG: unable to handle kernel paging request at 000000fffffffeb8
> [   46.849026] IP: [<ffffffff811a484c>] kmem_cache_destroy_memcg_children+0x6c/0xc0
> [   46.849092] PGD 0
> [   46.849092] Oops: 0000 [#1] SMP

OK, pretty soon we'll have a changelog!

What does one do to trigger this oops?  The bug has been there since
3.9, so the means-of-triggering must be quite special?


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] memcg: don't initialize kmem-cache destroying work for root caches
  2013-08-05 21:16     ` Andrew Morton
@ 2013-08-05 21:41       ` Andrew Vagin
  0 siblings, 0 replies; 6+ messages in thread
From: Andrew Vagin @ 2013-08-05 21:41 UTC (permalink / raw)
  To: Andrew Morton
  Cc: Andrey Vagin, linux-mm, cgroups, linux-kernel, Glauber Costa,
	Johannes Weiner, Michal Hocko, Balbir Singh, KAMEZAWA Hiroyuki,
	Konstantin Khlebnikov, stable

On Mon, Aug 05, 2013 at 02:16:09PM -0700, Andrew Morton wrote:
> On Tue, 6 Aug 2013 01:01:28 +0400 Andrew Vagin <avagin@parallels.com> wrote:
> 
> > On Mon, Aug 05, 2013 at 01:05:30PM -0700, Andrew Morton wrote:
> > > On Mon,  5 Aug 2013 20:09:40 +0400 Andrey Vagin <avagin@openvz.org> wrote:
> > > 
> > > > struct memcg_cache_params has a union. Different parts of this union
> > > > are used for root and non-root caches. A part with destroying work is
> > > > used only for non-root caches.
> > > > 
> > > > I fixed the same problem in another place v3.9-rc1-16204-gf101a94, but
> > > > didn't notice this one.
> > > > 
> > > > Cc: <stable@vger.kernel.org>    [3.9.x]
> > > 
> > > hm, why the cc:stable?
> > 
> > Because this patch fixes the kernel panic:
> > 
> > [   46.848187] BUG: unable to handle kernel paging request at 000000fffffffeb8
> > [   46.849026] IP: [<ffffffff811a484c>] kmem_cache_destroy_memcg_children+0x6c/0xc0
> > [   46.849092] PGD 0
> > [   46.849092] Oops: 0000 [#1] SMP
> 
> OK, pretty soon we'll have a changelog!

Sorry, probably I had to write all these in the initial commit message. I
just thought that this patch is an additional part of v3.9-rc1-16204-gf101a94.

> 
> What does one do to trigger this oops?  The bug has been there since
> 3.9, so the means-of-triggering must be quite special?

I don't think that so many people use cgroups with limits of the kernel memory.

I use the vzctl utility to operate with containers. vzctl limits the
kernel memory of containers by default. A container should be started
and stoped a few times (five or four) to reproduce the bug.

And one more thing is that nf_conntrack should be loaded. It creates
a new kmem_cache for each network namespace.

Thanks

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2013-08-05 21:41 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-05-22  8:09 [PATCH] memcg: don't initialize kmem-cache destroying work for root caches Andrey Vagin
  -- strict thread matches above, loose matches on Subject: below --
2013-08-05 16:09 Andrey Vagin
2013-08-05 20:05 ` Andrew Morton
2013-08-05 21:01   ` Andrew Vagin
2013-08-05 21:16     ` Andrew Morton
2013-08-05 21:41       ` Andrew Vagin

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).