[PATCH] do_sched_setscheduler(): don't take tasklist_lock
authorOleg Nesterov <oleg@tv-sign.ru>
Fri, 29 Sep 2006 09:00:48 +0000 (02:00 -0700)
committerLinus Torvalds <torvalds@g5.osdl.org>
Fri, 29 Sep 2006 16:18:17 +0000 (09:18 -0700)
Use rcu locks instead. sched_setscheduler() now takes ->siglock
before reading ->signal->rlim[].

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
kernel/sched.c

index b946209d9c1571bfc0c443101546e92a67b4e587..f9b3c6a414f189bf443044b9947715cb94825fa5 100644 (file)
@@ -4080,6 +4080,8 @@ static void __setscheduler(struct task_struct *p, int policy, int prio)
  * @p: the task in question.
  * @policy: new policy.
  * @param: structure containing the new RT priority.
+ *
+ * NOTE: the task may be already dead
  */
 int sched_setscheduler(struct task_struct *p, int policy,
                       struct sched_param *param)
@@ -4115,19 +4117,26 @@ recheck:
         * Allow unprivileged RT tasks to decrease priority:
         */
        if (!capable(CAP_SYS_NICE)) {
+               unsigned long rlim_rtprio;
+               unsigned long flags;
+
+               if (!lock_task_sighand(p, &flags))
+                       return -ESRCH;
+               rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur;
+               unlock_task_sighand(p, &flags);
+
                /*
                 * can't change policy, except between SCHED_NORMAL
                 * and SCHED_BATCH:
                 */
                if (((policy != SCHED_NORMAL && p->policy != SCHED_BATCH) &&
                        (policy != SCHED_BATCH && p->policy != SCHED_NORMAL)) &&
-                               !p->signal->rlim[RLIMIT_RTPRIO].rlim_cur)
+                               !rlim_rtprio)
                        return -EPERM;
                /* can't increase priority */
                if ((policy != SCHED_NORMAL && policy != SCHED_BATCH) &&
                    param->sched_priority > p->rt_priority &&
-                   param->sched_priority >
-                               p->signal->rlim[RLIMIT_RTPRIO].rlim_cur)
+                   param->sched_priority > rlim_rtprio)
                        return -EPERM;
                /* can't change other user's priorities */
                if ((current->euid != p->euid) &&
@@ -4193,14 +4202,13 @@ do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
                return -EINVAL;
        if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
                return -EFAULT;
-       read_lock_irq(&tasklist_lock);
+
+       rcu_read_lock();
+       retval = -ESRCH;
        p = find_process_by_pid(pid);
-       if (!p) {
-               read_unlock_irq(&tasklist_lock);
-               return -ESRCH;
-       }
-       retval = sched_setscheduler(p, policy, &lparam);
-       read_unlock_irq(&tasklist_lock);
+       if (p != NULL)
+               retval = sched_setscheduler(p, policy, &lparam);
+       rcu_read_unlock();
 
        return retval;
 }