From: Eric Caruso ejcaruso@google.com
commit 2895a5e5b3ae78d9923a91fce405d4a2f32c4309 upstream.
timerfd gives processes a way to set wake alarms, but unlike timers made using timer_create, timerfds don't check whether the process has CAP_WAKE_ALARM before setting alarm-time timers. CAP_WAKE_ALARM is supposed to gate this behavior and so it makes sense that we should deny permission to create such timerfds if the process doesn't have this capability.
Signed-off-by: Eric Caruso ejcaruso@google.com Cc: Todd Poynor toddpoynor@google.com Link: http://lkml.kernel.org/r/1465427339-96209-1-git-send-email-ejcaruso@chromium... Signed-off-by: Thomas Gleixner tglx@linutronix.de Cc: Kasper Zwijsen Kasper.Zwijsen@UGent.be Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org --- fs/timerfd.c | 10 ++++++++++ 1 file changed, 10 insertions(+)
--- a/fs/timerfd.c +++ b/fs/timerfd.c @@ -400,6 +400,11 @@ SYSCALL_DEFINE2(timerfd_create, int, clo clockid != CLOCK_BOOTTIME_ALARM)) return -EINVAL;
+ if (!capable(CAP_WAKE_ALARM) && + (clockid == CLOCK_REALTIME_ALARM || + clockid == CLOCK_BOOTTIME_ALARM)) + return -EPERM; + ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); if (!ctx) return -ENOMEM; @@ -444,6 +449,11 @@ static int do_timerfd_settime(int ufd, i return ret; ctx = f.file->private_data;
+ if (!capable(CAP_WAKE_ALARM) && isalarm(ctx)) { + fdput(f); + return -EPERM; + } + timerfd_setup_cancel(ctx, flags);
/*