[PATCH 8/8] vfork: do not block SIG_DFL/SIG_IGN signals issingle-threaded

From: Oleg Nesterov
Date: Wed Jul 27 2011 - 12:37:45 EST


vfork() blocks all signals except SIGKILL and SIGSTOP. This means
it doesn't react to ^Z or other fatal signals.

We can unblock all signals which doesn't have a handler and solve
this. Unfortunately, without the really ugly hacks we can not do
this in the multithreaded case, we can not trust sighand->action[]
otherwise.

Let's do this in the single-threaded case at least. Anyway, I do
not think that vfork() from the multithreaded application is sane.
And even in this case other threads can handle the blocked signals
unless they exit after clone_vfork_prepare().

Note: "sighand->count == 1" doesn't handle the dead-leader case,
this is easy to fix.

Signed-off-by: Oleg Nesterov <oleg@xxxxxxxxxx>
---

kernel/fork.c | 18 +++++++++++++++++-
1 file changed, 17 insertions(+), 1 deletion(-)

--- 3.1/kernel/fork.c~8_more_signals 2011-07-27 17:48:53.000000000 +0200
+++ 3.1/kernel/fork.c 2011-07-27 17:51:54.000000000 +0200
@@ -1506,10 +1506,26 @@ static long clone_vfork_restart(struct r

static void clone_vfork_prepare(void)
{
+ struct sighand_struct *sigh = current->sighand;
sigset_t vfork_mask;

- current->saved_sigmask = current->blocked;
siginitsetinv(&vfork_mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
+ if (atomic_read(&sigh->count) == 1) {
+ __sighandler_t h;
+ int signr;
+ /*
+ * Nobody can play with ->action[], we can unblock all
+ * signals which do not have a handler, they can not
+ * trigger return-to-user-mode.
+ */
+ for (signr = 1; signr <= _NSIG; ++signr) {
+ h = sigh->action[signr-1].sa.sa_handler;
+ if (h == SIG_DFL || h == SIG_IGN)
+ sigdelset(&vfork_mask, signr);
+ }
+ }
+
+ current->saved_sigmask = current->blocked;
set_current_blocked(&vfork_mask);
}


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/