signals: do_tkill: don't use tasklist_lock

Convert do_tkill() to use rcu_read_lock() + lock_task_sighand() to avoid
taking tasklist lock.

Note that we don't return an error if lock_task_sighand() fails, we pretend
the task dies after receiving the signal.  Otherwise, we should fight with the
nasty races with mt-exec without having any advantage.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 9 insertions, 5 deletions

diff --git a/kernel/signal.c b/kernel/signal.c
index b3dedf1..13371d1 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2219,6 +2219,7 @@ static int do_tkill(int tgid, int pid, int sig)
 	int error;
 	struct siginfo info;
 	struct task_struct *p;
+	unsigned long flags;
 
 	error = -ESRCH;
 	info.si_signo = sig;
@@ -2227,21 +2228,24 @@ static int do_tkill(int tgid, int pid, int sig)
 	info.si_pid = task_tgid_vnr(current);
 	info.si_uid = current->uid;
 
-	read_lock(&tasklist_lock);
+	rcu_read_lock();
 	p = find_task_by_vpid(pid);
 	if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
 		error = check_kill_permission(sig, &info, p);
 		/*
 		 * The null signal is a permissions and process existence
 		 * probe.  No signal is actually delivered.
+		 *
+		 * If lock_task_sighand() fails we pretend the task dies
+		 * after receiving the signal. The window is tiny, and the
+		 * signal is private anyway.
 		 */
-		if (!error && sig && p->sighand) {
-			spin_lock_irq(&p->sighand->siglock);
+		if (!error && sig && lock_task_sighand(p, &flags)) {
 			error = specific_send_sig_info(sig, &info, p);
-			spin_unlock_irq(&p->sighand->siglock);
+			unlock_task_sighand(p, &flags);
 		}
 	}
-	read_unlock(&tasklist_lock);
+	rcu_read_unlock();
 
 	return error;
 }