[PATCH] Uprobes: Fix kernel oops with delayed_uprobe_remove()

[PATCH] Uprobes: Fix kernel oops with delayed_uprobe_remove()

[Ravi Bangoria]

syzbot reported a kernel crash with delayed_uprobe_remove():
  https://lkml.org/lkml/2018/11/1/1244

Backtrace mentioned in the link points to a race between process
exit and uprobe_unregister(). Fix it by locking delayed_uprobe_lock
before calling delayed_uprobe_remove() from put_uprobe().

Reported-by: syzbot+cb1fb754b771caca0a88@syzkaller.appspotmail.com
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
---
 kernel/events/uprobes.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index 96fb51f3994f..e527c4753d4f 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -572,7 +572,9 @@ static void put_uprobe(struct uprobe *uprobe)
 		 * gets called, we don't get a chance to remove uprobe from
 		 * delayed_uprobe_list from remove_breakpoint(). Do it here.
 		 */
+		mutex_lock(&delayed_uprobe_lock);
 		delayed_uprobe_remove(uprobe, NULL);
+		mutex_unlock(&delayed_uprobe_lock);
 		kfree(uprobe);
 	}
 }
-- 
2.19.1

Re: [PATCH] Uprobes: Fix kernel oops with delayed_uprobe_remove()

[Oleg Nesterov]

On 11/14, Ravi Bangoria wrote:
>
> syzbot reported a kernel crash with delayed_uprobe_remove():
>   https://lkml.org/lkml/2018/11/1/1244
>
> Backtrace mentioned in the link points to a race between process
> exit and uprobe_unregister(). Fix it by locking delayed_uprobe_lock
> before calling delayed_uprobe_remove() from put_uprobe().

The patch looks good to me, but could you update the changelog?

Please explain that the exiting task calls uprobe_clear_state() which
can race with delayed_uprobe_remove(). IIUC this is the only problem
solved by this patch, right?

Oleg.

Re: [PATCH] Uprobes: Fix kernel oops with delayed_uprobe_remove()

[Ravi Bangoria]

Hi Oleg,

On 11/14/18 9:36 PM, Oleg Nesterov wrote:
> On 11/14, Ravi Bangoria wrote:
>>
>> syzbot reported a kernel crash with delayed_uprobe_remove():
>>   https://lkml.org/lkml/2018/11/1/1244
>>
>> Backtrace mentioned in the link points to a race between process
>> exit and uprobe_unregister(). Fix it by locking delayed_uprobe_lock
>> before calling delayed_uprobe_remove() from put_uprobe().
> 
> The patch looks good to me, but could you update the changelog?
> 
> Please explain that the exiting task calls uprobe_clear_state() which
> can race with delayed_uprobe_remove(). IIUC this is the only problem
> solved by this patch, right?

Right. Is this better:

There could be a race between task exit and probe unregister:

  exit_mm()
  mmput()
  __mmput()                     uprobe_unregister()
  uprobe_clear_state()          put_uprobe()
  delayed_uprobe_remove()       delayed_uprobe_remove()

put_uprobe() is calling delayed_uprobe_remove() without taking
delayed_uprobe_lock and thus the race sometimes results in a
kernel crash. Fix this by taking delayed_uprobe_lock before
calling delayed_uprobe_remove() from put_uprobe().

Detailed crash log can be found at:
  https://lkml.org/lkml/2018/11/1/1244

Re: [PATCH] Uprobes: Fix kernel oops with delayed_uprobe_remove()

[Oleg Nesterov]

On 11/15, Ravi Bangoria wrote:
>
> There could be a race between task exit and probe unregister:
>
>   exit_mm()
>   mmput()
>   __mmput()                     uprobe_unregister()
>   uprobe_clear_state()          put_uprobe()
>   delayed_uprobe_remove()       delayed_uprobe_remove()
>
> put_uprobe() is calling delayed_uprobe_remove() without taking
> delayed_uprobe_lock and thus the race sometimes results in a
> kernel crash. Fix this by taking delayed_uprobe_lock before
> calling delayed_uprobe_remove() from put_uprobe().
>
> Detailed crash log can be found at:
>   https://lkml.org/lkml/2018/11/1/1244

Thanks, looks good,

Oleg.