Re: [patch] clone.2: Add EACCES with CLONE_INTO_CGROUP + clone3 to ERRORS

From: Christian Brauner
Date: Mon Aug 30 2021 - 03:38:41 EST


On Sun, Aug 29, 2021 at 03:57:06PM -0400, Andrew Wock wrote:
> Resending because it's my first time mailing the lkml and I used html.
> Reattempting w/ gmail's plaintext mode. I apologise if this is
> reaching you twice.
>
> I noticed that clone3 can send the EACCES errno after I wrote a
> program that used clone3 with the CLONE_INTO_CGROUP flag. To me, it's
> important to know what kind of failure occurred if the clone3 fails,
> so I was glad that a unique errno is set for this case, but it wasn't
> documented on the clone man page.

In essence, any error that could occur during regular fs-based migration
at write-time can also occur during CLONE_INTO_CGROUP. The clone3()
manpage just has the inverse of that above statement:

"Note that all of the usual restrictions (described in cgroups(7)) on
placing a process into a version 2 cgroup apply."

>
> I've attached a patch and a test program.
>
> Test program is attached as clone3_doc.c. Create
> /sys/fs/cgroup/not-allowed as root, then run the program. It should
> set errno to EACCES.

This is a manpage update, right? In that case it's not necessarily
needed to Cc lkml aka linux-kernel@... itself.

For the content:
Acked-by: Christian Brauner <christian.brauner@xxxxxxxxxx>

(I have no idea what patch format Michael will accept so I can't really
ack that. :))

>
> Thanks,
> Andrew Wock

> #include <stdio.h>
> #include <errno.h>
> #include <stdlib.h>
> #include <string.h>
> #include <signal.h>
> #include <fcntl.h>
>
> #include <linux/sched.h> /* Definition of struct clone_args */
> #include <sched.h> /* Definition of CLONE_* constants */
> #include <sys/syscall.h> /* Definition of SYS_* constants */
> #include <unistd.h>
>
> /*
> * Preconditions:
> * - /sys/fs/cgroup/not-allowed is a real cgroup.
> * - You are not root and do not have write permissions to
> * /sys/fs/cgroup/not-allowed/cgroup.procs
> */
> int main() {
> pid_t pid;
> int fd;
> struct clone_args cl_args = {0};
> char *cgPath = "/sys/fs/cgroup/not-allowed";
>
> fd = open(cgPath, O_RDONLY);
> if (fd == -1) {
> fprintf(stderr, "Could not open cgroup %s: %s\n", cgPath, strerror(errno));
> exit(1);
> }
>
> cl_args.exit_signal = SIGCHLD;
> cl_args.flags = CLONE_INTO_CGROUP;
> cl_args.cgroup = fd;
> pid = syscall(SYS_clone3, &cl_args, sizeof(cl_args));
> if (pid == -1) {
> if (errno == EACCES) {
> printf("EACCES detected\n");
> exit(0);
> }
> fprintf(stderr, "Could not clone into cgroup: %s\n", strerror(errno));
> } else if (pid == 0) {
> fprintf(stderr, "Are you root, or do you have write access to %s?\n", cgPath);
> }
> exit(1);
> }