[PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[Oleg Nesterov]

Andrew Tridgell reports that aio_read(SIGEV_SIGNAL) can fail if the
the notification from the helper thread races with setresuid(), see
http://samba.org/~tridge/junkcode/aio_uid.c

This happens because check_kill_permission() doesn't allow to send
a signal to the task with the different cred->xids. But there is no
any security reason to check ->cred's when the task sends a signal
(private or group-wide) to its sub-thread. Whatever we do, any thread
can bypass all security checks and send SIGKILL to all threads, or
it can block a signal SIG and do kill(gettid(), SIG) to deliver this
signal to another sub-thread. Not to mention that CLONE_THREAD implies
CLONE_VM.

Change check_kill_permission() to avoid the credentials check when
the sender and the target are from the same thread group.

Also, move "cred = current_cred()" down to avoid calling get_current()
twice.

Note: David Howells pointed out we could relax this even more, the
CLONE_SIGHAND (without CLONE_THREAD) case probably does not need
these checks too.

Reported-by: Andrew Tridgell <tridge@samba.org>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---

 kernel/signal.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

--- 34-rc1/kernel/signal.c~CKP_CK_STG	2010-05-09 21:09:31.000000000 +0200
+++ 34-rc1/kernel/signal.c	2010-05-17 17:02:09.000000000 +0200
@@ -642,7 +642,7 @@ static inline bool si_fromuser(const str
 static int check_kill_permission(int sig, struct siginfo *info,
 				 struct task_struct *t)
 {
-	const struct cred *cred = current_cred(), *tcred;
+	const struct cred *cred, *tcred;
 	struct pid *sid;
 	int error;
 
@@ -656,8 +656,10 @@ static int check_kill_permission(int sig
 	if (error)
 		return error;
 
+	cred = current_cred();
 	tcred = __task_cred(t);
-	if ((cred->euid ^ tcred->suid) &&
+	if (!same_thread_group(current, t) &&
+	    (cred->euid ^ tcred->suid) &&
 	    (cred->euid ^ tcred->uid) &&
 	    (cred->uid  ^ tcred->suid) &&
 	    (cred->uid  ^ tcred->uid) &&

Re: [PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[Roland McGrath]

Acked-by: Roland McGrath <roland@redhat.com>

I concur that there is no real security benefit from checking uids for any
CLONE_VM case.  But there is also no reason to change the old behavior in
this corner for any case but CLONE_THREAD.


Thanks,
Roland

Re: [PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[David Howells]

Oleg Nesterov <oleg@redhat.com> wrote:

> Also, move "cred = current_cred()" down to avoid calling get_current()
> twice.

I don't see what you mean by this.  same_thread_group() doesn't call
current_cred(), so why this change?

David

Re: [PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[Oleg Nesterov]

On 05/18, David Howells wrote:
>
> Oleg Nesterov <oleg@redhat.com> wrote:
>
> > Also, move "cred = current_cred()" down to avoid calling get_current()
> > twice.
>
> I don't see what you mean by this.  same_thread_group() doesn't call
> current_cred(), so why this change?

Yes, but both current_cred() and same_thread_group(current, t) call
get_current(), and gcc doesn't cache the result because we call
audit_signal_info() in between.

In fact, initially I was going to send the patch below, but then
decided to make a more simple change.

Oleg.

--- x/kernel/signal.c
+++ x/kernel/signal.c
@@ -642,8 +642,6 @@ static inline bool si_fromuser(const str
 static int check_kill_permission(int sig, struct siginfo *info,
 				 struct task_struct *t)
 {
-	const struct cred *cred = current_cred(), *tcred;
-	struct pid *sid;
 	int error;
 
 	if (!valid_signal(sig))
@@ -656,23 +654,29 @@ static int check_kill_permission(int sig
 	if (error)
 		return error;
 
-	tcred = __task_cred(t);
-	if ((cred->euid ^ tcred->suid) &&
-	    (cred->euid ^ tcred->uid) &&
-	    (cred->uid  ^ tcred->suid) &&
-	    (cred->uid  ^ tcred->uid) &&
-	    !capable(CAP_KILL)) {
-		switch (sig) {
-		case SIGCONT:
-			sid = task_session(t);
-			/*
-			 * We don't return the error if sid == NULL. The
-			 * task was unhashed, the caller must notice this.
-			 */
-			if (!sid || sid == task_session(current))
-				break;
-		default:
-			return -EPERM;
+	if (!same_thread_group(current, t)) {
+		const struct cred *cred = current_cred();
+		const struct cred *tcred = __task_cred(t);
+		struct pid *sid;
+
+		if ((cred->euid ^ tcred->suid) &&
+		    (cred->euid ^ tcred->uid) &&
+		    (cred->uid  ^ tcred->suid) &&
+		    (cred->uid  ^ tcred->uid) &&
+		    !capable(CAP_KILL)) {
+			switch (sig) {
+			case SIGCONT:
+				sid = task_session(t);
+				/*
+				 * We don't return the error if sid == NULL.
+				 * The task was unhashed, the caller must
+				 * notice this.
+				 */
+				if (!sid || sid == task_session(current))
+					break;
+			default:
+				return -EPERM;
+			}
 		}
 	}
 

Re: [PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[David Howells]

Oleg Nesterov <oleg@redhat.com> wrote:

> Yes, but both current_cred() and same_thread_group(current, t) call
> get_current(), and gcc doesn't cache the result because we call
> audit_signal_info() in between.

Sorry, yes.  I was reading get_current() as current_cred() for some reason.

However, you are _still_ calling get_current() twice...  So that bit of your
changelog isn't really correct.

In fact, get_current() should be __attribute_const__ since it can't change
whilst you're looking at it, except within switch_to(), probably in a piece of
assembly code, so gcc should be free to cache it as long as it likes.

David

Re: [PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[Oleg Nesterov]

On 05/18, David Howells wrote:
>
> Oleg Nesterov <oleg@redhat.com> wrote:
>
> > Yes, but both current_cred() and same_thread_group(current, t) call
> > get_current(), and gcc doesn't cache the result because we call
> > audit_signal_info() in between.
>
> Sorry, yes.  I was reading get_current() as current_cred() for some reason.
>
> However, you are _still_ calling get_current() twice...  So that bit of your
> changelog isn't really correct.

If I read kernel/signal.s correctly - no.

Well, yes, get_current() is still called twice inside check_kill_permission().
But this is because we have audit_signal_info()->audit_dummy_context() which
uses current too.

But "cred = current_cred()" and same_thread_group(current, t) read
gs:current_task only once, so this change really helps (although the
optimization is very minor, of course).

> In fact, get_current() should be __attribute_const__ since it can't change
> whilst you're looking at it, except within switch_to(), probably in a piece of
> assembly code, so gcc should be free to cache it as long as it likes.

Agreed! I thought about this many times.

Oleg.

Re: [PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[Andrew Morton]

On Mon, 17 May 2010 21:54:14 +0200
Oleg Nesterov <oleg@redhat.com> wrote:

> Andrew Tridgell reports that aio_read(SIGEV_SIGNAL) can fail if the
> the notification from the helper thread races with setresuid(), see
> http://samba.org/~tridge/junkcode/aio_uid.c
> 
> This happens because check_kill_permission() doesn't allow to send
> a signal to the task with the different cred->xids. But there is no
> any security reason to check ->cred's when the task sends a signal
> (private or group-wide) to its sub-thread. Whatever we do, any thread
> can bypass all security checks and send SIGKILL to all threads, or
> it can block a signal SIG and do kill(gettid(), SIG) to deliver this
> signal to another sub-thread. Not to mention that CLONE_THREAD implies
> CLONE_VM.
> 
> Change check_kill_permission() to avoid the credentials check when
> the sender and the target are from the same thread group.
> 
> Also, move "cred = current_cred()" down to avoid calling get_current()
> twice.
> 
> Note: David Howells pointed out we could relax this even more, the
> CLONE_SIGHAND (without CLONE_THREAD) case probably does not need
> these checks too.

So... which kernel(s) do we think this fix should be merged into?

Re: [PATCH] signals: check_kill_permission: don't check creds if same_thread_group()

[Roland McGrath]

> So... which kernel(s) do we think this fix should be merged into?

I'd say all.  The glibc (libpthread) that does set*id across threads has
been in use for a while (2.3.4?), probably in distro's using kernels as old
or older than any active -stable streams.  In the race in question, this
kernel bug is breaking valid POSIX application expectations.


Thanks,
Roland