2008-01-14 21:03:34

by Oleg Nesterov

Subject: [PATCH] fix the "remove task_ppid_nr_ns" commit

The commit 84427eaef1fb91704c7112bdb598c810003b99f3 (remove task_ppid_nr_ns)
moves the task_tgid_nr_ns(task->real_parent) outside of lock_task_sighand().
This is wrong, ->real_parent could be freed/reused.

Both ->parent/real_parent point to nothnig after __exit_signal() because we
remove the child from ->children list, and thus the child can't be reparented
when its parent exits.

rcu_read_lock() protects ->parent/real_parent, but _only_ if we know it was
valid before we take rcu lock.

Revert this part of the patch.

Signed-off-by: Oleg Nesterov <[email protected]>

--- t/fs/proc/array.c~ppid_fix 2008-01-14 23:45:35.000000000 +0300
+++ t/fs/proc/array.c 2008-01-14 23:59:27.000000000 +0300
@@ -426,7 +426,6 @@ static int do_task_stat(struct task_stru
cgtime = gtime = cputime_zero;

rcu_read_lock();
- ppid = task_tgid_nr_ns(task->real_parent, ns);
if (lock_task_sighand(task, &flags)) {
struct signal_struct *sig = task->signal;

@@ -465,6 +464,7 @@ static int do_task_stat(struct task_stru
}

sid = task_session_nr_ns(task, ns);
+ ppid = task_tgid_nr_ns(task->real_parent, ns);
pgid = task_pgrp_nr_ns(task, ns);

unlock_task_sighand(task, &flags);

2008-01-14 21:22:57

by Roland McGrath

Subject: Re: [PATCH] fix the "remove task_ppid_nr_ns" commit

This preserves the parts of my change that mattered, is correctly more
conservative than my change was, and is fine with me.

Thanks,
Roland