aio: block exit_aio() until all context requests are completed
authorGu Zheng <guz.fnst@cn.fujitsu.com>
Wed, 3 Sep 2014 09:45:44 +0000 (17:45 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 5 Oct 2014 21:52:24 +0000 (14:52 -0700)
commit 6098b45b32e6baeacc04790773ced9340601d511 upstream.

It seems that exit_aio() also needs to wait for all iocbs to complete (like
io_destroy), but we missed the wait step in current implemention, so fix
it in the same way as we did in io_destroy.

Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
fs/aio.c

index c4b6b82..f45ddaa 100644 (file)
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -797,6 +797,9 @@ void exit_aio(struct mm_struct *mm)
        unsigned i = 0;
 
        while (1) {
+               struct completion requests_done =
+                       COMPLETION_INITIALIZER_ONSTACK(requests_done);
+
                rcu_read_lock();
                table = rcu_dereference(mm->ioctx_table);
 
@@ -824,7 +827,10 @@ void exit_aio(struct mm_struct *mm)
                 */
                ctx->mmap_size = 0;
 
-               kill_ioctx(mm, ctx, NULL);
+               kill_ioctx(mm, ctx, &requests_done);
+
+               /* Wait until all IO for the context are done. */
+               wait_for_completion(&requests_done);
        }
 }