[xfrm_user] BUG: sleeping function called from invalid context

[xfrm_user] BUG: sleeping function called from invalid context

[Luca Tettamanti]

Hello,
I just noticed this message in my log (I'm not sure what charon was
doing at the time...):

BUG: sleeping function called from invalid context at /home/kronos/src/linux-2.6.git/mm/slub.c:1701
in_atomic(): 1, irqs_disabled(): 0, pid: 10411, name: charon
Pid: 10411, comm: charon Not tainted 2.6.35 #271
Call Trace:
 [<ffffffff810275d7>] __might_sleep+0xf8/0xfa
 [<ffffffff810a99be>] kmem_cache_alloc+0x35/0x9d
 [<ffffffff81265c8e>] xfrm_policy_alloc+0x20/0xa1
 [<ffffffffa0372d73>] xfrm_compile_policy+0x112/0x16f [xfrm_user]
 [<ffffffff812686e4>] xfrm_user_policy+0xb8/0x131
 [<ffffffff81233b5b>] do_ip_setsockopt+0xa83/0xb2e
 [<ffffffff8102c0ce>] ? get_parent_ip+0x11/0x41
 [<ffffffff81080838>] ? unlock_page+0x22/0x27
 [<ffffffff810935c3>] ? __do_fault+0x4d4/0x50c
 [<ffffffff8109497d>] ? handle_mm_fault+0x45d/0x8a4
 [<ffffffff810afbef>] ? get_empty_filp+0x79/0x15a
 [<ffffffff8102c0e5>] ? get_parent_ip+0x28/0x41
 [<ffffffff8102c0e5>] ? get_parent_ip+0x28/0x41
 [<ffffffff81283385>] ? sub_preempt_count+0x92/0xa6
 [<ffffffff812832b6>] ? do_page_fault+0x3d5/0x412
 [<ffffffff8125522e>] ? inet_bind+0x1cc/0x1dc
 [<ffffffff81233cc2>] ip_setsockopt+0x23/0x82
 [<ffffffff810acfbf>] ? fd_install+0x54/0x5d
 [<ffffffff8124d175>] udp_setsockopt+0x24/0x26
 [<ffffffff811fd5c5>] sock_common_setsockopt+0xf/0x11
 [<ffffffff811fb0a8>] sys_setsockopt+0x81/0xa2
 [<ffffffff8100296b>] system_call_fastpath+0x16/0x1b


xfrm_user_policy takes read_lock(&xfrm_km_lock) before calling
xfrm_compile_policy (via km->compile_policy), which in turn calls
xfrm_policy_alloc with GFP_KERNEL.

Luca

Re: [xfrm_user] BUG: sleeping function called from invalid context

[Alex Badea]

Hi,

On 08/12/2010 03:50 PM, Luca Tettamanti wrote:
>  [<ffffffff8124d175>] udp_setsockopt+0x24/0x26
>  [<ffffffff811fd5c5>] sock_common_setsockopt+0xf/0x11
>  [<ffffffff811fb0a8>] sys_setsockopt+0x81/0xa2
>  [<ffffffff8100296b>] system_call_fastpath+0x16/0x1b
> 
> 
> xfrm_user_policy takes read_lock(&xfrm_km_lock) before calling
> xfrm_compile_policy (via km->compile_policy), which in turn calls
> xfrm_policy_alloc with GFP_KERNEL.

Since it's a setsockopt(), I suspect it was setting up per-socket bypass
policies.

Regards,
Alex

Re: [xfrm_user] BUG: sleeping function called from invalid context

[Herbert Xu]

Luca Tettamanti <kronos.it@gmail.com> wrote:
>
> xfrm_user_policy takes read_lock(&xfrm_km_lock) before calling
> xfrm_compile_policy (via km->compile_policy), which in turn calls
> xfrm_policy_alloc with GFP_KERNEL.

Thanks for discovering this bug, it only took 8 years :)

xfrm: Use GFP_ATOMIC in xfrm_compile_policy

As xfrm_compile_policy runs within a read_lock, we cannot use
GFP_KERNEL for memory allocations.

Reported-by: Luca Tettamanti <kronos.it@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

diff --git a/net/xfrm/xfrm_user.c b/net/xfrm/xfrm_user.c
index ba59983..b14ed4b 100644
--- a/net/xfrm/xfrm_user.c
+++ b/net/xfrm/xfrm_user.c
@@ -2504,7 +2504,7 @@ static struct xfrm_policy *xfrm_compile_policy(struct sock *sk, int opt,
 	if (p->dir > XFRM_POLICY_OUT)
 		return NULL;
 
-	xp = xfrm_policy_alloc(net, GFP_KERNEL);
+	xp = xfrm_policy_alloc(net, GFP_ATOMIC);
 	if (xp == NULL) {
 		*dir = -ENOBUFS;
 		return NULL;

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [xfrm_user] BUG: sleeping function called from invalid context

[David Miller]

From: Herbert Xu <herbert@gondor.apana.org.au>
Date: Fri, 13 Aug 2010 11:36:58 -0400

> Luca Tettamanti <kronos.it@gmail.com> wrote:
>>
>> xfrm_user_policy takes read_lock(&xfrm_km_lock) before calling
>> xfrm_compile_policy (via km->compile_policy), which in turn calls
>> xfrm_policy_alloc with GFP_KERNEL.
> 
> Thanks for discovering this bug, it only took 8 years :)

We stumble over one of these every so often don't we? :)

> 
> xfrm: Use GFP_ATOMIC in xfrm_compile_policy
> 
> As xfrm_compile_policy runs within a read_lock, we cannot use
> GFP_KERNEL for memory allocations.
> 
> Reported-by: Luca Tettamanti <kronos.it@gmail.com>
> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Applied, thanks Herbert.