[PATCH] proc: drop tasklist lock in task_state()

task_state() needs tasklist_lock to protect ->parent/->real_parent.  However
task->parent points to nowhere only when the actions below happen in order

	1) release_task(task)
	2) release_task(task->parent)
	3) a grace period passed

But 3) implies that the memory ops from 1) should be finished, so pid_alive()
can't be true in such a case.

Otherwise, we don't care if ->parent/->real_parent changes under us.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

diff --git a/fs/proc/array.c b/fs/proc/array.c
index 35bd39d..25e917f 100644 (file)
--- a/fs/proc/array.c
+++ b/fs/proc/array.c
@@ -162,7 +162,7 @@ static inline char * task_state(struct task_struct *p, char *buffer)
        int g;
        struct fdtable *fdt = NULL;
 
-       read_lock(&tasklist_lock);
+       rcu_read_lock();
        buffer += sprintf(buffer,
                "State:\t%s\n"
                "SleepAVG:\t%lu%%\n"
@@ -174,14 +174,13 @@ static inline char * task_state(struct task_struct *p, char *buffer)
                "Gid:\t%d\t%d\t%d\t%d\n",
                get_task_state(p),
                (p->sleep_avg/1024)*100/(1020000000/1024),
-               p->tgid,
-               p->pid, pid_alive(p) ? p->group_leader->real_parent->tgid : 0,
-               pid_alive(p) && p->ptrace ? p->parent->pid : 0,
+               p->tgid, p->pid,
+               pid_alive(p) ? rcu_dereference(p->real_parent)->tgid : 0,
+               pid_alive(p) && p->ptrace ? rcu_dereference(p->parent)->pid : 0,
                p->uid, p->euid, p->suid, p->fsuid,
                p->gid, p->egid, p->sgid, p->fsgid);
-       read_unlock(&tasklist_lock);
+
        task_lock(p);
-       rcu_read_lock();
        if (p->files)
                fdt = files_fdtable(p->files);
        buffer += sprintf(buffer,