If perf_event_open() is called with another task as target and
perf_event_attr::sigtrap is set, and the target task's user does not
match the calling user, also require the CAP_KILL capability.
Otherwise, with the CAP_PERFMON capability alone it would be possible
for a user to send SIGTRAP signals via perf events to another user's
tasks. This could potentially result in those tasks being terminated if
they cannot handle SIGTRAP signals.
Note: The check complements the existing capability check, but is not
supposed to supersede the ptrace_may_access() check. At a high level we
now have:
capable of CAP_PERFMON and (CAP_KILL if sigtrap)
OR
ptrace_may_access() // also checks for same thread-group and uid
Fixes: 97ba62b27867 ("perf: Add support for SIGTRAP on perf events")
Cc: <[email protected]> # 5.13+
Reported-by: Dmitry Vyukov <[email protected]>
Signed-off-by: Marco Elver <[email protected]>
---
v2:
* Drop kill_capable() and just check CAP_KILL (reported by Ondrej Mosnacek).
* Use ns_capable(__task_cred(task)->user_ns, CAP_KILL) to check for
capability in target task's ns (reported by Ondrej Mosnacek).
---
kernel/events/core.c | 15 ++++++++++++++-
1 file changed, 14 insertions(+), 1 deletion(-)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index fe88d6eea3c2..43c99695dc3f 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -12152,10 +12152,23 @@ SYSCALL_DEFINE5(perf_event_open,
}
if (task) {
+ bool is_capable;
+
err = down_read_interruptible(&task->signal->exec_update_lock);
if (err)
goto err_file;
+ is_capable = perfmon_capable();
+ if (attr.sigtrap) {
+ /*
+ * perf_event_attr::sigtrap sends signals to the other
+ * task. Require the current task to have CAP_KILL.
+ */
+ rcu_read_lock();
+ is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
+ rcu_read_unlock();
+ }
+
/*
* Preserve ptrace permission check for backwards compatibility.
*
@@ -12165,7 +12178,7 @@ SYSCALL_DEFINE5(perf_event_open,
* perf_event_exit_task() that could imply).
*/
err = -EACCES;
- if (!perfmon_capable() && !ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS))
+ if (!is_capable && !ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS))
goto err_cred;
}
--
2.32.0.93.g670b81a890-goog
Marco Elver <[email protected]> writes:
> If perf_event_open() is called with another task as target and
> perf_event_attr::sigtrap is set, and the target task's user does not
> match the calling user, also require the CAP_KILL capability.
>
> Otherwise, with the CAP_PERFMON capability alone it would be possible
> for a user to send SIGTRAP signals via perf events to another user's
> tasks. This could potentially result in those tasks being terminated if
> they cannot handle SIGTRAP signals.
>
> Note: The check complements the existing capability check, but is not
> supposed to supersede the ptrace_may_access() check. At a high level we
> now have:
>
> capable of CAP_PERFMON and (CAP_KILL if sigtrap)
> OR
> ptrace_may_access() // also checks for same thread-group and uid
Is there anyway we could have a comment that makes the required
capability checks clear?
Basically I see an inlined version of kill_ok_by_cred being implemented
without the comments on why the various pieces make sense.
Certainly ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS) should not
be a check to allow writing/changing a task. It needs to be
PTRACE_MODE_ATTACH_REALCREDS, like /proc/self/mem uses.
Now in practice I think your patch probably has the proper checks in
place for sending a signal but it is far from clear.
Eric
> Fixes: 97ba62b27867 ("perf: Add support for SIGTRAP on perf events")
> Cc: <[email protected]> # 5.13+
> Reported-by: Dmitry Vyukov <[email protected]>
> Signed-off-by: Marco Elver <[email protected]>
> ---
> v2:
> * Drop kill_capable() and just check CAP_KILL (reported by Ondrej Mosnacek).
> * Use ns_capable(__task_cred(task)->user_ns, CAP_KILL) to check for
> capability in target task's ns (reported by Ondrej Mosnacek).
> ---
> kernel/events/core.c | 15 ++++++++++++++-
> 1 file changed, 14 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index fe88d6eea3c2..43c99695dc3f 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -12152,10 +12152,23 @@ SYSCALL_DEFINE5(perf_event_open,
> }
>
> if (task) {
> + bool is_capable;
> +
> err = down_read_interruptible(&task->signal->exec_update_lock);
> if (err)
> goto err_file;
>
> + is_capable = perfmon_capable();
> + if (attr.sigtrap) {
> + /*
> + * perf_event_attr::sigtrap sends signals to the other
> + * task. Require the current task to have CAP_KILL.
> + */
> + rcu_read_lock();
> + is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
> + rcu_read_unlock();
> + }
> +
> /*
> * Preserve ptrace permission check for backwards compatibility.
> *
> @@ -12165,7 +12178,7 @@ SYSCALL_DEFINE5(perf_event_open,
> * perf_event_exit_task() that could imply).
> */
> err = -EACCES;
> - if (!perfmon_capable() && !ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS))
> + if (!is_capable && !ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS))
> goto err_cred;
> }
On Thu, 1 Jul 2021 at 23:41, Eric W. Biederman <[email protected]> wrote:
>
> Marco Elver <[email protected]> writes:
>
> > If perf_event_open() is called with another task as target and
> > perf_event_attr::sigtrap is set, and the target task's user does not
> > match the calling user, also require the CAP_KILL capability.
> >
> > Otherwise, with the CAP_PERFMON capability alone it would be possible
> > for a user to send SIGTRAP signals via perf events to another user's
> > tasks. This could potentially result in those tasks being terminated if
> > they cannot handle SIGTRAP signals.
> >
> > Note: The check complements the existing capability check, but is not
> > supposed to supersede the ptrace_may_access() check. At a high level we
> > now have:
> >
> > capable of CAP_PERFMON and (CAP_KILL if sigtrap)
> > OR
> > ptrace_may_access() // also checks for same thread-group and uid
>
> Is there anyway we could have a comment that makes the required
> capability checks clear?
>
> Basically I see an inlined version of kill_ok_by_cred being implemented
> without the comments on why the various pieces make sense.
I'll add more comments. It probably also makes sense to factor the
code here into its own helper.
> Certainly ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS) should not
> be a check to allow writing/changing a task. It needs to be
> PTRACE_MODE_ATTACH_REALCREDS, like /proc/self/mem uses.
So if attr.sigtrap the checked ptrace mode needs to switch to
PTRACE_MODE_ATTACH_REALCREDS. Otherwise, it is possible to send a
signal if only read-ptrace permissions are granted.
Is my assumption here correct?
> Now in practice I think your patch probably has the proper checks in
> place for sending a signal but it is far from clear.
Thanks,
-- Marco