Re: [PATCH 3/7] seccomp_filter: Enable ftrace-based system call filtering

[Will Drewry]

On Thu, Apr 28, 2011 at 11:56 AM, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> On Thu, 2011-04-28 at 10:30 -0500, Will Drewry wrote:
>
>> >> diff --git a/kernel/seccomp.c b/kernel/seccomp.c
>> >> index 57d4b13..1bee87c 100644
>> >> --- a/kernel/seccomp.c
>> >> +++ b/kernel/seccomp.c
>> >> @@ -8,10 +8,11 @@
>> >>
>> >>  #include <linux/seccomp.h>
>> >>  #include <linux/sched.h>
>> >> +#include <linux/slab.h>
>> >>  #include <linux/compat.h>
>> >>
>> >>  /* #define SECCOMP_DEBUG 1 */
>> >> -#define NR_SECCOMP_MODES 1
>> >> +#define NR_SECCOMP_MODES 2
>> >>
>> >>  /*
>> >>   * Secure computing mode 1 allows only read/write/exit/sigreturn.
>> >> @@ -32,9 +33,11 @@ static int mode1_syscalls_32[] = {
>> >>
>> >>  void __secure_computing(int this_syscall)
>> >>  {
>> >> -     int mode = current->seccomp.mode;
>> >> +     int mode = -1;
>> >>       int * syscall;
>> >> -
>> >> +     /* Do we need an RCU read lock to access current's state? */
>> >
>> > I'm actually confused to why you are using RCU. What are you protecting.
>> > Currently, I see the state is always accessed from current->seccomp. But
>> > current should not be fighting with itself.
>> >
>> > Maybe I'm missing something.
>>
>> I'm sure it's me that's missing something.  I believe the seccomp
>> pointer can be accessed from:
>> - current
>> - via /proc/<pid>/seccomp_filter (read-only)
>>
>> Given those cases, would it make sense to ditch the RCU interface for it?
>
> Looking at this in a bit more detail. I think you can ditch the
> rcu_readlocks where current accesses its own seccomp state. As current
> is the one that duplicates it (and ups the refcount) on fork, and
> current wont free it until after it performs a rcu_synchronization. No
> one else can free current's seccomp state while current has a reference
> to it.
>
> You still need the rcu_readlocks on the readers for the proc system.
> Otherwise the handle can be freed while its still in use. With the
> rcu_readlocks, these readers will always get the refcount before current
> frees it. And then the dec_and_test should work as expected when the
> readers do the put.

Great, I'll do that!
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/