Re: [PATCH] Introduce ActivePid: in /proc/self/status (v2, was Vpid:)

[Bryan Donlan]

On Wed, Jun 22, 2011 at 11:00, Greg Kurz <gkurz@xxxxxxxxxx> wrote:
> On Mon, 2011-06-20 at 13:37 -0400, Bryan Donlan wrote:
>> On Mon, Jun 20, 2011 at 07:45, Greg Kurz <gkurz@xxxxxxxxxx> wrote:
>> > On Thu, 2011-06-16 at 13:54 -0400, Bryan Donlan wrote:
>>
>> >> Although getting the in-namespace PID is a useful thing, wouldn't a
>> >> truly race-free API be preferable? Any access by PID has the race
>> >> condition in which the target process could die, and its PID get
>> >> recycled between retrieving the PID and doing something with it.
>> >
>> > Well the PID is a racy construct when used by another task than the
>> > parent... fortunately, most userland code can cope with it ! :)
>>
>> That doesn't mean we shouldn't try to fix the race! :)
>>
>> >> Perhaps a file-descriptor API would be better, such as something like
>> >> this:
>> >>
>> >> int openpid(int id, int flags);
>> >> int rt_sigqueueinfo_fd(int process_fd, int sig, siginfo_t *info);
>> >> int sigqueue_fd(int process_fd, int sig, const union sigval value); //
>> >> glibc wrapper
>> >>
>> >
>> > The race still exists: openpid() is being passed a PID... Only the
>> > parent can legitimately know that this PID identifies a specific
>> > unwaited child.
>>
>> Yes, the idea would be either the parent process, or the target
>> process itself would open the PID, then pass the resulting file
>> descriptor to whatever process is actually doing the killing.
>
> Agreed. Such an API would be useful in a scenario where the task to be
> killed and the killing task can share a file descriptor: same thread
> group or inherited with clone() or connected with an AF_UNIX socket.
> My point was just that the racy pid based API will still be needed to
> handle all the other scenarios. But maybe it's fine to have two sets of
> process handling calls.

Actually, with a /proc/self/sigqueue file as Eric suggested, , it
would be possible to mitigate some races even if you're not passed a
fd from the target or its parent. Consider:

1) Read somedaemon.pid
2) Open the /proc/$pid directory
3) Use openat to open /proc/$pid/status; check that Name: matches the
daemon's name; if not go to 1
4) Use openat to open /proc/$pid/sigqueue; send a signal

It's not foolproof, in that you might hit a different process with the
same name+uid+gid+groups+cmdline+environment+etc, but any such
cross-process operation is racy to the extent that, at some level, you
need to have criteria for selecting your target, and such criteria
might allow for false positives.
--
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/