aio: don't bother with unmapping when aio_free_ring() is coming from exit_aio()
authorAl Viro <viro@zeniv.linux.org.uk>
Sat, 21 Apr 2012 01:49:41 +0000 (21:49 -0400)
committerAl Viro <viro@zeniv.linux.org.uk>
Sat, 21 Apr 2012 05:58:16 +0000 (01:58 -0400)
... since exit_mmap() is coming and it will munmap() everything anyway.
In all other cases aio_free_ring() has ctx->mm == current->mm; moreover,
all other callers of vm_munmap() have mm == current->mm, so this will
allow us to get rid of mm argument of vm_munmap().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/aio.c

index 99bd790e8cd253479befd5013d6b38a5781d3927..976e33d9741303d0d1643ddaf865918ebeca27c9 100644 (file)
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -92,8 +92,10 @@ static void aio_free_ring(struct kioctx *ctx)
        for (i=0; i<info->nr_pages; i++)
                put_page(info->ring_pages[i]);
 
-       if (info->mmap_size)
+       if (info->mmap_size) {
+               BUG_ON(ctx->mm != current->mm);
                vm_munmap(ctx->mm, info->mmap_base, info->mmap_size);
+       }
 
        if (info->ring_pages && info->ring_pages != info->internal_pages)
                kfree(info->ring_pages);
@@ -386,6 +388,17 @@ void exit_aio(struct mm_struct *mm)
                                "exit_aio:ioctx still alive: %d %d %d\n",
                                atomic_read(&ctx->users), ctx->dead,
                                ctx->reqs_active);
+               /*
+                * We don't need to bother with munmap() here -
+                * exit_mmap(mm) is coming and it'll unmap everything.
+                * Since aio_free_ring() uses non-zero ->mmap_size
+                * as indicator that it needs to unmap the area,
+                * just set it to 0; aio_free_ring() is the only
+                * place that uses ->mmap_size, so it's safe.
+                * That way we get all munmap done to current->mm -
+                * all other callers have ctx->mm == current->mm.
+                */
+               ctx->ring_info.mmap_size = 0;
                put_ioctx(ctx);
        }
 }