libbpf-tools: add cgroup_helpers and oomkill support display cgroup

[Rtoax]

Using a simple test program (not shown) called OOM, we performed the
following two tests:

1. Allocating unlimited memory
2. Adding the process to a cgroup named oom-memcg and limiting its memory
   usage to 200MB.

When we do not print cgroup information, we can only see process information
and cannot see the difference between memcg and non-memcg.

$ sudo ./oomkill
Tracing OOM kills... Ctrl-C to stop.
14:28:23 Triggered by PID 179201 ("oom"), OOM kill of PID 179201 ("oom"), 6114610 pages, loadavg: loadavg: 0.56 0.51 0.38 2/968 179204

14:28:42 Triggered by PID 179212 ("oom"), OOM kill of PID 179212 ("oom"), 51200 pages, loadavg: loadavg: 0.40 0.47 0.37 3/968 179212

The function implemented by this patch can clearly display cgroup information.

$ sudo ./oomkill  -c
Tracing OOM kills... Ctrl-C to stop.
14:32:59 Triggered by PID 179879 ("oom"), CGROUP 8309 ("/sys/fs/cgroup/user.slice/user-1000.slice/[email protected]/session.slice/[email protected]"), OOM kill of PID 179879 ("oom"), 6114610 pages, loadavg: loadavg: 0.50 0.38 0.35 4/970 179879

14:33:14 Triggered by PID 179884 ("oom"), CGROUP 122547 ("/sys/fs/cgroup/oom-memcg"), MEMCG 122547 ("/sys/fs/cgroup/oom-memcg"), OOM kill of PID 179884 ("oom"), 51200 pages, loadavg: loadavg: 0.47 0.38 0.35 3/971 179884

and add cgroup_helpers.c.

[yonghong-song]

For this one:

Using a simple test program (not shown) called OOM, we performed the
following two tests:
Allocating unlimited memory
Adding the process to a cgroup named oom-memcg and limiting its memory
usage to 200MB.

Could you actually show your simple test? This way, people can reproduce the issue easily.

[Rtoax]

Could you actually show your simple test? This way, people can reproduce the issue easily.

Yes, thanks, the test code:

oom_minimal.c

#include <errno.h>
#include <malloc.h>
#include <unistd.h>
#include <stdio.h>

int main(void)
{
	void *mem;

	for (;;) {
		mem = malloc(getpagesize());
		if (!mem && errno == ENOMEM) {
			fprintf(stderr, "OOMing...\n");
		}
		*(int *)mem = 1;
		/* Just leak the memory */
	}
	return 0;
}

cgroup-oom.sh

#!/bin/bash
set -e

readonly pid=$$
readonly CGROUP_NAME=oom-test

[[ -z ${OOMer} ]] && OOMer=oom_minimal

cleanup() {
	printf "\n"
	sudo rmdir /sys/fs/cgroup/${CGROUP_NAME}/
}
trap cleanup EXIT

sudo mkdir -p /sys/fs/cgroup/${CGROUP_NAME}/
echo ${pid} | sudo tee /sys/fs/cgroup/${CGROUP_NAME}/cgroup.procs
echo $((1024*1024*2)) | sudo tee /sys/fs/cgroup/${CGROUP_NAME}/memory.max
echo $((1024*1024*2)) | sudo tee /sys/fs/cgroup/${CGROUP_NAME}/memory.high

eval ./${OOMer} ${@}

compile and run

gcc oom_minimal.c -o oom_minim
./cgroup-oom.sh

then

$ sudo ./oomkill -c
09:31:49 Triggered by PID 53791 ("oom_minimal"), CGROUP 45172 ("/sys/fs/cgroup/oom-test"), MEMCG 45172 ("/sys/fs/cgroup/oom-test"), OOM kill of PID 53791 ("oom_minimal"), 2097663 pages, loadavg: loadavg: 0.19 0.21 0.20 3/1098 53792

[Rtoax]

@yonghong-song Thanks a lot, i just submit all changes and add test code above, please review, :)