[PATCH 1/4] exit: Move read_unlock() up in mm_update_next_owner()

[Kirill Tkhai]

All the places, we update task's mm, are made this
under task_lock(). These are exec_mmap(), exit_mm()
and kernel thread's use_mm() and unuse_mm().
The only exception is copy_mm(), which initializes
newborn task's mm, but we can't race with it, as
mm_update_next_owner() iterates tasks already linked
to task list.

get_task_struct() guarantees that task_struct can't free.
exec_mmap() and exit_mm() both call mm_update_next_owner():

        task_lock(current);
        current->mm = new_mm;
        task_unlock(current);
        mm_update_next_owner(mm);

and since mm_update_next_owner() checks and assigns mm under
task_lock() too, they can't miss they became a new mm owner.
So we can relax the lock and release tasklist_lock earlier.

Signed-off-by: Kirill Tkhai <ktk...@virtuozzo.com>
---
 kernel/exit.c |    9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/kernel/exit.c b/kernel/exit.c
index c3c7ac560114..9fb7b699bdeb 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -464,18 +464,15 @@ void mm_update_next_owner(struct mm_struct *mm)
        return;
 
 assign_new_owner:
-       BUG_ON(c == p);
        get_task_struct(c);
+       read_unlock(&tasklist_lock);
+       BUG_ON(c == p);
+
        /*
         * The task_lock protects c->mm from changing.
         * We always want mm->owner->mm == mm
         */
        task_lock(c);
-       /*
-        * Delay read_unlock() till we have the task_lock()
-        * to ensure that c does not slip away underneath us
-        */
-       read_unlock(&tasklist_lock);
        if (c->mm != mm) {
                task_unlock(c);
                put_task_struct(c);