[PATCH v3] fork: check exit_signal passed in clone3() call

[PATCH v3] fork: check exit_signal passed in clone3() call

[Eugene Syromiatnikov]

Hello.

As was agreed[1][2], clone3 should fail if the provided exit_signal
value fails valid_signal() check, hence the new version.

Changees since v2[3][4]:
 - Rewrite the check to check exit_signal against valid_signal().

Changes since v1[5]:
 - Check changed to comparison against negated CSIGNAL to address
   the bug reported by Oleg[6].
 - Added a comment to _do_fork that exit_signal has to be checked
   by the caller.

[1] https://lkml.org/lkml/2019/9/11/503
[2] https://lkml.org/lkml/2019/9/11/518
[3] https://lkml.org/lkml/2019/9/10/764
[4] https://lkml.org/lkml/2019/9/10/765
[5] https://lkml.org/lkml/2019/9/10/411
[6] https://lkml.org/lkml/2019/9/10/467


Eugene Syromiatnikov (1):
  fork: check exit_signal passed in clone3() call

 kernel/fork.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

-- 
2.1.4

[PATCH v3] fork: check exit_signal passed in clone3() call

[Eugene Syromiatnikov]

Previously, higher 32 bits of exit_signal fields were lost when
copied to the kernel args structure (that uses int as a type for the
respective field).  Moreover, as Oleg has noted[1], exit_signal is used
unchecked, so it has to be checked for sanity before use; for the legacy
syscalls, applying CSIGNAL mask guarantees that it is at least non-negative;
however, there's no such thing is done in clone3() code path, and that can
break at least thread_group_leader.

Adding checks that user-passed exit_signal fits into int and passes
valid_signal() check solves both of these problems.

[1] https://lkml.org/lkml/2019/9/10/467

* kernel/fork.c (copy_clone_args_from_user): Fail with -EINVAL if
args.exit_signal is greater than UINT_MAX or is not a valid signal.
(_do_fork): Note that exit_signal is expected to be checked for the
sanity by the caller.

Fixes: 7f192e3cd316 ("fork: add clone3")
Reported-by: Oleg Nesterov <oleg@redhat.com>
Co-authored-by: Oleg Nesterov <oleg@redhat.com>
Co-authored-by: Dmitry V. Levin <ldv@altlinux.org>
Signed-off-by: Eugene Syromiatnikov <esyr@redhat.com>
---
 kernel/fork.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/kernel/fork.c b/kernel/fork.c
index 2852d0e..f98314b 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -2338,6 +2338,8 @@ struct mm_struct *copy_init_mm(void)
  *
  * It copies the process, and if successful kick-starts
  * it and waits for it to finish using the VM if required.
+ *
+ * args->exit_signal is expected to be checked for sanity by the caller.
  */
 long _do_fork(struct kernel_clone_args *args)
 {
@@ -2562,6 +2564,15 @@ noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
 	if (copy_from_user(&args, uargs, size))
 		return -EFAULT;
 
+	/*
+	 * Two separate checks are needed, as valid_signal() takes unsigned long
+	 * as an argument, and struct kernel_clone_args uses int type
+	 * for the exit_signal field.
+	 */
+	if (unlikely((args.exit_signal > UINT_MAX) ||
+		     !valid_signal(args.exit_signal)))
+		return -EINVAL;
+
 	*kargs = (struct kernel_clone_args){
 		.flags		= args.flags,
 		.pidfd		= u64_to_user_ptr(args.pidfd),
-- 
2.1.4

Re: [PATCH v3] fork: check exit_signal passed in clone3() call

[Oleg Nesterov]

On 09/11, Eugene Syromiatnikov wrote:
>
> @@ -2562,6 +2564,15 @@ noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
>  	if (copy_from_user(&args, uargs, size))
>  		return -EFAULT;
>  
> +	/*
> +	 * Two separate checks are needed, as valid_signal() takes unsigned long
> +	 * as an argument, and struct kernel_clone_args uses int type
> +	 * for the exit_signal field.
> +	 */
> +	if (unlikely((args.exit_signal > UINT_MAX) ||
> +		     !valid_signal(args.exit_signal)))
> +		return -EINVAL;

OK, I equally agree with this version. Although I'd simply do

	if (args.exit_signal > _NSIG)
		return -EINVAL;

but this is cosmetic.

Acked-by: Oleg Nesterov <oleg@redhat.com>

Re: [PATCH v3] fork: check exit_signal passed in clone3() call

[Christian Brauner]

On Wed, Sep 11, 2019 at 06:45:40PM +0100, Eugene Syromiatnikov wrote:
> Previously, higher 32 bits of exit_signal fields were lost when
> copied to the kernel args structure (that uses int as a type for the
> respective field).  Moreover, as Oleg has noted[1], exit_signal is used
> unchecked, so it has to be checked for sanity before use; for the legacy
> syscalls, applying CSIGNAL mask guarantees that it is at least non-negative;
> however, there's no such thing is done in clone3() code path, and that can
> break at least thread_group_leader.
> 
> Adding checks that user-passed exit_signal fits into int and passes
> valid_signal() check solves both of these problems.
> 
> [1] https://lkml.org/lkml/2019/9/10/467
> 
> * kernel/fork.c (copy_clone_args_from_user): Fail with -EINVAL if
> args.exit_signal is greater than UINT_MAX or is not a valid signal.
> (_do_fork): Note that exit_signal is expected to be checked for the
> sanity by the caller.
> 
> Fixes: 7f192e3cd316 ("fork: add clone3")
> Reported-by: Oleg Nesterov <oleg@redhat.com>
> Co-authored-by: Oleg Nesterov <oleg@redhat.com>
> Co-authored-by: Dmitry V. Levin <ldv@altlinux.org>
> Signed-off-by: Eugene Syromiatnikov <esyr@redhat.com>

For the sake of posterity I appended a reproducer to the patch (cf. [1])
that I pushed to mainline which illustrates how without this patch you
can cause a crash. I'll also explain how I think this happens here.

By passing in a negative signal you can cause a segfault in
proc_flush_task(). The reason is that at process creation time

static inline bool thread_group_leader(struct task_struct *p)
{
	return p->exit_signal >= 0;
}

will return false even though it should return true because exit_signal
overflowed in copy_clone_args_from_user. This means, the kernel
thinks this is a thread and doesn't make the process a thread-group
leader. In proc_flush_task() the kernel will then try to retrieve the
thread group leader but there'll be none ultimately causing a segfault.
Specifically:

void proc_flush_task(struct task_struct *task)
{
	int i;
	struct pid *pid, *tgid;
	struct upid *upid;

	pid = task_pid(task);
	tgid = task_tgid(task); #### tgid is NULL ####

	for (i = 0; i <= pid->level; i++) {
		upid = &pid->numbers[i];
		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
					tgid->numbers[i].nr); #### NULL pointer deref ####
	}
}

[1]:
  #define _GNU_SOURCE
 #include <linux/sched.h>
 #include <linux/types.h>
 #include <sched.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <sys/syscall.h>
 #include <sys/wait.h>
 #include <unistd.h>

 int main(int argc, char *argv[])
 {
        pid_t pid = -1;
        struct clone_args args = {0};
        args.exit_signal = -1;

        pid = syscall(__NR_clone3, &args, sizeof(struct clone_args));
        if (pid < 0)
                exit(EXIT_FAILURE);

        if (pid == 0)
                exit(EXIT_SUCCESS);

        wait(NULL);

        exit(EXIT_SUCCESS);
 }

 Christian