Re: [PATCH v4 2/3] nsproxy: attach to namespaces via pidfds

Hi.

On Tue, May 05, 2020 at 04:04:31PM +0200, Christian Brauner [off-list ref] wrote:

-SYSCALL_DEFINE2(setns, int, fd, int, nstype)
+SYSCALL_DEFINE2(setns, int, fd, int, flags)
[...]
-	file = proc_ns_fget(fd);
-	if (IS_ERR(file))
-		return PTR_ERR(file);
+	int err = 0;
 
-	err = -EINVAL;
-	ns = get_proc_ns(file_inode(file));
-	if (nstype && (ns->ops->type != nstype))
+	file = fget(fd);
+	if (!file)
+		return -EBADF;
+
+	if (proc_ns_file(file)) {
+		ns = get_proc_ns(file_inode(file));
+		if (flags && (ns->ops->type != flags))
+			err = -EINVAL;
+		flags = ns->ops->type;
+	} else if (pidfd_pid(file)) {
+		err = check_setns_flags(flags);
+	} else {
+		err = -EBADF;
+	}
+	if (err)
 		goto out;
 
-	err = prepare_nsset(ns->ops->type, &nsset);
+	err = prepare_nsset(flags, &nsset);
 	if (err)
 		goto out;

This modification changed the returned error when a valid file
descriptor is passed but it doesn't represent a namespace (nor pidfd).
The error is now EBADF although originally and per man page it
was/should be EINVAL.

A change like below would restore it, however, I see it may be less
consistent with other pidfd calls(?), then I'd suggest updating the
manpage to capture this.

--- a/kernel/nsproxy.c
+++ b/kernel/nsproxy.c

@@ -531,7 +531,7 @@ SYSCALL_DEFINE2(setns, int, fd, int, flags)
        } else if (!IS_ERR(pidfd_pid(file))) {
                err = check_setns_flags(flags);
        } else {
-               err = -EBADF;
+               err = -EINVAL;
        }
        if (err)
                goto out;

I noticed this breaks systemd self tests [1].

Regards,
Michal


[1] https://github.com/systemd/systemd/blob/a1ba8c5b71164665ccb53c9cec384e5eef7d3689/src/test/test-seccomp.c#L246