mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] bpf: Call rcu_read_unlock() before copy_to_user()
@ 2015-01-22 15:57 Michael Holzheu
  0 siblings, 0 replies; 4+ messages in thread
From: Michael Holzheu @ 2015-01-22 15:57 UTC (permalink / raw)
  To: Alexei Starovoitov; +Cc: Alexei Starovoitov, Martin Schwidefsky, linux-kernel

We must not hold locks when calling copy_to_user():

BUG: sleeping function called from invalid context at mm/memory.c:3732
in_atomic(): 0, irqs_disabled(): 0, pid: 671, name: test_maps
1 lock held by test_maps/671:
 #0:  (rcu_read_lock){......}, at: [<0000000000264190>] map_lookup_elem+0xe8/0x260
Preemption disabled at:[<00000000001be3b6>] vprintk_default+0x56/0x68

CPU: 0 PID: 671 Comm: test_maps Not tainted 3.19.0-rc5-00117-g5eb11d6-dirty #424
       000000001e447bb0 000000001e447c40 0000000000000002 0000000000000000 
       000000001e447ce0 000000001e447c58 000000001e447c58 0000000000115c8a 
       0000000000000000 0000000000c08246 0000000000c27e8a 000000000000000b 
       000000001e447ca0 000000001e447c40 0000000000000000 0000000000000000 
       0000000000000000 0000000000115c8a 000000001e447c40 000000001e447ca0 
Call Trace:
([<0000000000115b7e>] show_trace+0x12e/0x150)
 [<0000000000115c40>] show_stack+0xa0/0x100
 [<00000000009b163c>] dump_stack+0x74/0xc8
 [<000000000017424a>] ___might_sleep+0x23a/0x248
 [<00000000002b58e8>] might_fault+0x70/0xe8
 [<0000000000264230>] map_lookup_elem+0x188/0x260
 [<0000000000264716>] SyS_bpf+0x20e/0x840
 [<00000000009bbe3a>] system_call+0xd6/0x24c
 [<000003fffd15f566>] 0x3fffd15f566
1 lock held by test_maps/671:
 #0:  (rcu_read_lock){......}, at: [<0000000000264190>] map_lookup_elem+0xe8/0x260

So call rcu_read_unlock() before copy_to_user(). We can
release the lock earlier because it is not needed for copy_to_user().

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
---
 kernel/bpf/syscall.c |    7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -172,17 +172,16 @@ static int map_lookup_elem(union bpf_att
 	err = -ENOENT;
 	rcu_read_lock();
 	value = map->ops->map_lookup_elem(map, key);
+	rcu_read_unlock();
 	if (!value)
-		goto err_unlock;
+		goto free_key;
 
 	err = -EFAULT;
 	if (copy_to_user(uvalue, value, map->value_size) != 0)
-		goto err_unlock;
+		goto free_key;
 
 	err = 0;
 
-err_unlock:
-	rcu_read_unlock();
 free_key:
 	kfree(key);
 err_put:


^ permalink raw reply	[flat|nested] 4+ messages in thread
* Re: [PATCH] bpf: Call rcu_read_unlock() before copy_to_user()
@ 2015-01-22 17:27 Alexei Starovoitov
  2015-01-22 17:54 ` Michael Holzheu
  0 siblings, 1 reply; 4+ messages in thread
From: Alexei Starovoitov @ 2015-01-22 17:27 UTC (permalink / raw)
  To: Michael Holzheu
  Cc: Alexei Starovoitov, Martin Schwidefsky, linux-kernel, netdev

On Thu, Jan 22, 2015 at 7:57 AM, Michael Holzheu
<holzheu@linux.vnet.ibm.com> wrote:
> We must not hold locks when calling copy_to_user():
>
> BUG: sleeping function called from invalid context at mm/memory.c:3732
> in_atomic(): 0, irqs_disabled(): 0, pid: 671, name: test_maps
> 1 lock held by test_maps/671:
>  #0:  (rcu_read_lock){......}, at: [<0000000000264190>] map_lookup_elem+0xe8/0x260
> Preemption disabled at:[<00000000001be3b6>] vprintk_default+0x56/0x68
>
> CPU: 0 PID: 671 Comm: test_maps Not tainted 3.19.0-rc5-00117-g5eb11d6-dirty #424
>        000000001e447bb0 000000001e447c40 0000000000000002 0000000000000000
>        000000001e447ce0 000000001e447c58 000000001e447c58 0000000000115c8a
>        0000000000000000 0000000000c08246 0000000000c27e8a 000000000000000b
>        000000001e447ca0 000000001e447c40 0000000000000000 0000000000000000
>        0000000000000000 0000000000115c8a 000000001e447c40 000000001e447ca0
> Call Trace:
> ([<0000000000115b7e>] show_trace+0x12e/0x150)
>  [<0000000000115c40>] show_stack+0xa0/0x100
>  [<00000000009b163c>] dump_stack+0x74/0xc8
>  [<000000000017424a>] ___might_sleep+0x23a/0x248
>  [<00000000002b58e8>] might_fault+0x70/0xe8
>  [<0000000000264230>] map_lookup_elem+0x188/0x260
>  [<0000000000264716>] SyS_bpf+0x20e/0x840
>  [<00000000009bbe3a>] system_call+0xd6/0x24c
>  [<000003fffd15f566>] 0x3fffd15f566
> 1 lock held by test_maps/671:
>  #0:  (rcu_read_lock){......}, at: [<0000000000264190>] map_lookup_elem+0xe8/0x260
>
> So call rcu_read_unlock() before copy_to_user(). We can
> release the lock earlier because it is not needed for copy_to_user().

we cannot move the rcu unlock this way, since it protects the value.
So we need to copy the value while still under rcu.
I'm puzzled how I missed this warning.
I guess you have CONFIG_PREEMPT_RCU=y ?
and if (in_atomic()) return; as part of might_fault() hid it.

^ permalink raw reply	[flat|nested] 4+ messages in thread
* Re: [PATCH] bpf: Call rcu_read_unlock() before copy_to_user()
@ 2015-01-22 18:03 Alexei Starovoitov
  0 siblings, 0 replies; 4+ messages in thread
From: Alexei Starovoitov @ 2015-01-22 18:03 UTC (permalink / raw)
  To: Michael Holzheu
  Cc: Alexei Starovoitov, Martin Schwidefsky, linux-kernel, netdev

On Thu, Jan 22, 2015 at 9:54 AM, Michael Holzheu
<holzheu@linux.vnet.ibm.com> wrote:
>> > So call rcu_read_unlock() before copy_to_user(). We can
>> > release the lock earlier because it is not needed for copy_to_user().
>>
>> we cannot move the rcu unlock this way, since it protects the value.
>> So we need to copy the value while still under rcu.
>
> Ok, right. I assume you will provide the correct fix.

sure :) will do.

^ permalink raw reply	[flat|nested] 4+ messages in thread
end of thread, other threads:[~2015-01-22 18:03 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-01-22 15:57 [PATCH] bpf: Call rcu_read_unlock() before copy_to_user() Michael Holzheu
2015-01-22 17:27 Alexei Starovoitov
2015-01-22 17:54 ` Michael Holzheu
2015-01-22 18:03 Alexei Starovoitov

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®