This is mostly cleanup and optimization, but also fixes the bug.
proc_flush_task() checks upid->nr == 1 to detect the case when
a sub-namespace exits. However, this doesn't work in case when
a multithreaded init execs and calls release_task(old_leader),
the old leader has the same pid 1.
Move pid_ns_release_proc() to zap_pid_ns_processes(), it is called
when we know for sure that init is exiting.
Note: with or without this change this mntput() can happen before the
EXIT_DEAD tasks not visible to do_wait() have passed proc_flush_task().
We need more fixes.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
fs/proc/base.c | 4 ----
kernel/pid_namespace.c | 2 ++
2 files changed, 2 insertions(+), 4 deletions(-)
--- 35-rc3/fs/proc/base.c~PNS_5_MOVE_MNTPUT_TO_ZAP 2010-06-23 22:06:01.000000000 +0200
+++ 35-rc3/fs/proc/base.c 2010-06-23 22:10:26.000000000 +0200
@@ -2745,10 +2745,6 @@ void proc_flush_task(struct task_struct
proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
tgid->numbers[i].nr);
}
-
- upid = &pid->numbers[pid->level];
- if (upid->nr == 1)
- pid_ns_release_proc(upid->ns);
}
static struct dentry *proc_pid_instantiate(struct inode *dir,
--- 35-rc3/kernel/pid_namespace.c~PNS_5_MOVE_MNTPUT_TO_ZAP 2010-06-23 22:13:07.000000000 +0200
+++ 35-rc3/kernel/pid_namespace.c 2010-06-23 22:13:55.000000000 +0200
@@ -189,6 +189,8 @@ void zap_pid_ns_processes(struct pid_nam
} while (rc != -ECHILD);
acct_exit_ns(pid_ns);
+ pid_ns_release_proc(pid_ns);
+
return;
}
--
