is pid_namespace leak in v3.10?

DISCUSSION Xishi Qiu 2 participants 2 messages Jul 5, 2016 Memory management

Share & tools

Xishi Qiu

author

2016-07-05 11:41 UTC #0 lore raw

I find pid_namespace leak by "cat /proc/slabinfo | grep pid_namespace". The kernel version is RHEL 7.1 (kernel v3.10 stable).
The following is the test case, after several times, the count of pid_namespace become very large, is it correct?

I also test mainline, and the count will increase too, but it seems stably later.

BTW, this patch doesn't help. 24c037ebf5723d4d9ab0996433cee4f96c292a4d
exit: pidns: alloc_pid() leaks pid_namespace if child_reaper is exiting

Thanks,
Xishi Qiu

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <signal.h>

#ifndef CLONE_NEWPID
#define CLONE_NEWPID 0x20000000
#endif

void test(void)

{
        printf("clone child\n");
        exit(0);
}

int main()

{
        pid_t pid, child_pid;
        int  i, status;
        void *stack;

        for (i = 0; i < 100; i++) {
                stack = malloc(8192);
                pid = clone(&test, (char *)stack + 8192, CLONE_NEWPID|SIGCHLD, 0);
        }

        sleep(5);

        return 0;
}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Oleg Nesterov

2016-07-05 14:34 UTC #1 lore raw

Quoted 5 lines · Xishi Qiu

I find pid_namespace leak by "cat /proc/slabinfo | grep pid_namespace". The kernel version is RHEL 7.1 (kernel v3.10 stable).
The following is the test case, after several times, the count of pid_namespace become very large, is it correct?

Apparently not,

Quoted 1 line

I also test mainline, and the count will increase too, but it seems stably later.

And I can't reproduce the problem with the latest rhel7 kernel.

And just in case, I have no idea what actually slub reports as "active_objs" but certainly this is not the number of allocated "in use" objects, so it is fine if this counter doesn't go to zero when your test-case exits. But it should not grow "too much".

Quoted 3 lines

BTW, this patch doesn't help. 24c037ebf5723d4d9ab0996433cee4f96c292a4d
exit: pidns: alloc_pid() leaks pid_namespace if child_reaper is exiting

Sure, it can't help, your test-case doesn't fork other processes which could race with the exiting sub-namespace init.

Quoted 12 lines

int main()

{
        pid_t pid, child_pid;
        int  i, status;
        void *stack;

        for (i = 0; i < 100; i++) {
                stack = malloc(8192);
                pid = clone(&test, (char *)stack + 8192, CLONE_NEWPID|SIGCHLD, 0);
        }

        sleep(5);

is this sleep() really needed to trigger the problem?

Oleg.