Re: [PATCH] Fix a race in pid generation that causes pids to be reused immediately.

[Michel Lespinasse <walken@google.com>]

On Tue, Jun 08, 2010 at 11:24:38PM -0700, Salman wrote:
> A program that repeatedly forks and waits is susceptible to having the
> same pid repeated, especially when it competes with another instance of the
> same program.  This is really bad for bash implementation.  Furthermore, many shell
> scripts assume that pid numbers will not be used for some length of time.
> 
> Thanks to Ted Tso for the key ideas of this implementation.
> 
> Signed-off-by: Salman Qazi <sqazi@google.com>
> ---
>  kernel/pid.c |   11 ++++++++++-
>  1 files changed, 10 insertions(+), 1 deletions(-)
> 
> diff --git a/kernel/pid.c b/kernel/pid.c
> index e9fd8c1..8cedeab 100644
> --- a/kernel/pid.c
> +++ b/kernel/pid.c
> @@ -153,8 +153,17 @@ static int alloc_pidmap(struct pid_namespace *pid_ns)
>  		if (likely(atomic_read(&map->nr_free))) {
>  			do {
>  				if (!test_and_set_bit(offset, map->page)) {
> +					int prev;
>  					atomic_dec(&map->nr_free);
> -					pid_ns->last_pid = pid;
> +
> +					do {
> +						prev = last;
> +						last = cmpxchg(&pid_ns->last_pid,
> +							       prev, pid);
> +						if (last >= pid)
> +							break;

You should make sure to handle pid wrap-around for this last/pid comparison.
I think proper way to do that would be:

	/* last is the pid we started scanning at
         * last_read is the last observed value of pid_ns->last_pid
         */
        last_read = last;
        do {
                prev = last_read;
                last_read = cmpxchg(&pid_ns->last_pid, prev, pid);
        /* Exit if one of these conditions is true:
         * - cmpxchg succeeded
         * - last <= pid <= last_read  (other thread already bumped last_pid)
         * - last_read <= last <= pid  (same with wraparound)
         * - pid <= last_read <= last  (same with different wraparound)
         */
        } while (last_read != prev &&
                 (last > pid || pid > last_read) &&
                 (last_read > last || last > pid) &&
                 (pid > last_read || last_read > last));

The last_read == pid case is also interesting - it means another thread found
the same pid, forked a child with that pid, and the child exited already
(since the bitmap was cleared). However we don't need to handle that case -
first, that race is much less likely to happen, and second, the duplicate
pid would be returned in two separate tasks - so this would not cause problems
in bash as in your example.

-- 
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.