Re: general protection fault in oom_unkillable_task

[Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>]

On 2019/06/16 16:37, Tetsuo Handa wrote:
> On 2019/06/16 6:33, Tetsuo Handa wrote:
>> On 2019/06/16 3:50, Shakeel Butt wrote:
>>>> While dump_tasks() traverses only each thread group, mem_cgroup_scan_tasks()
>>>> traverses each thread.
>>>
>>> I think mem_cgroup_scan_tasks() traversing threads is not intentional
>>> and css_task_iter_start in it should use CSS_TASK_ITER_PROCS as the
>>> oom killer only cares about the processes or more specifically
>>> mm_struct (though two different thread groups can have same mm_struct
>>> but that is fine).
>>
>> We can't use CSS_TASK_ITER_PROCS from mem_cgroup_scan_tasks(). I've tried
>> CSS_TASK_ITER_PROCS in an attempt to evaluate only one thread from each
>> thread group, but I found that CSS_TASK_ITER_PROCS causes skipping whole
>> threads in a thread group (and trivially allowing "Out of memory and no
>> killable processes...\n" flood) if thread group leader has already exited.
> 
> Seems that CSS_TASK_ITER_PROCS from mem_cgroup_scan_tasks() is now working.


I found a reproducer and the commit.

----------------------------------------
#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <sched.h>
#include <sys/mman.h>
#include <asm/unistd.h>

static const unsigned long size = 1048576 * 200;
static int thread(void *unused)
{
        int fd = open("/dev/zero", O_RDONLY);
        char *buf = mmap(NULL, size, PROT_WRITE | PROT_READ,
                         MAP_ANONYMOUS | MAP_SHARED, EOF, 0);
        sleep(1);
        read(fd, buf, size);
        return syscall(__NR_exit, 0);
}
int main(int argc, char *argv[])
{
        FILE *fp;
        mkdir("/sys/fs/cgroup/memory/test1", 0755);
        fp = fopen("/sys/fs/cgroup/memory/test1/memory.limit_in_bytes", "w");
        fprintf(fp, "%lu\n", size);
        fclose(fp);
        fp = fopen("/sys/fs/cgroup/memory/test1/tasks", "w");
        fprintf(fp, "%u\n", getpid());
        fclose(fp);
        clone(thread, malloc(8192) + 4096, CLONE_SIGHAND | CLONE_THREAD | CLONE_VM, NULL);
        return syscall(__NR_exit, 0);
}
----------------------------------------

Here is a patch to use CSS_TASK_ITER_PROCS.

From 415e52cf55bc4ad931e4f005421b827f0b02693d Mon Sep 17 00:00:00 2001
From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Date: Mon, 17 Jun 2019 00:09:38 +0900
Subject: [PATCH] mm: memcontrol: Use CSS_TASK_ITER_PROCS at mem_cgroup_scan_tasks().

Since commit c03cd7738a83b137 ("cgroup: Include dying leaders with live
threads in PROCS iterations") corrected how CSS_TASK_ITER_PROCS works,
mem_cgroup_scan_tasks() can use CSS_TASK_ITER_PROCS in order to check
only one thread from each thread group.

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
---
 mm/memcontrol.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index ba9138a..b09ff45 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -1163,7 +1163,7 @@ int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
 		struct css_task_iter it;
 		struct task_struct *task;
 
-		css_task_iter_start(&iter->css, 0, &it);
+		css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it);
 		while (!ret && (task = css_task_iter_next(&it)))
 			ret = fn(task, arg);
 		css_task_iter_end(&it);
-- 
1.8.3.1