pin iocb through aio.

aio_poll() is not the only case that needs file pinned; worse, while
aio_read()/aio_write() can live without pinning iocb itself, the
proof is rather brittle and can easily break on later changes.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

diff --git a/fs/aio.c b/fs/aio.c
index 38b741a..07083fc 100644 (file)
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -1022,6 +1022,9 @@ static bool get_reqs_available(struct kioctx *ctx)
 /* aio_get_req
  *     Allocate a slot for an aio request.
  * Returns NULL if no requests are free.
+ *
+ * The refcount is initialized to 2 - one for the async op completion,
+ * one for the synchronous code that does this.
  */
 static inline struct aio_kiocb *aio_get_req(struct kioctx *ctx)
 {
@@ -1034,7 +1037,7 @@ static inline struct aio_kiocb *aio_get_req(struct kioctx *ctx)
        percpu_ref_get(&ctx->reqs);
        req->ki_ctx = ctx;
        INIT_LIST_HEAD(&req->ki_list);
-       refcount_set(&req->ki_refcnt, 0);
+       refcount_set(&req->ki_refcnt, 2);
        req->ki_eventfd = NULL;
        return req;
 }
@@ -1067,15 +1070,18 @@ out:
        return ret;
 }
 
+static inline void iocb_destroy(struct aio_kiocb *iocb)
+{
+       if (iocb->ki_filp)
+               fput(iocb->ki_filp);
+       percpu_ref_put(&iocb->ki_ctx->reqs);
+       kmem_cache_free(kiocb_cachep, iocb);
+}
+
 static inline void iocb_put(struct aio_kiocb *iocb)
 {
-       if (refcount_read(&iocb->ki_refcnt) == 0 ||
-           refcount_dec_and_test(&iocb->ki_refcnt)) {
-               if (iocb->ki_filp)
-                       fput(iocb->ki_filp);
-               percpu_ref_put(&iocb->ki_ctx->reqs);
-               kmem_cache_free(kiocb_cachep, iocb);
-       }
+       if (refcount_dec_and_test(&iocb->ki_refcnt))
+               iocb_destroy(iocb);
 }
 
 static void aio_fill_event(struct io_event *ev, struct aio_kiocb *iocb,
@@ -1749,9 +1755,6 @@ static ssize_t aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
        INIT_LIST_HEAD(&req->wait.entry);
        init_waitqueue_func_entry(&req->wait, aio_poll_wake);
 
-       /* one for removal from waitqueue, one for this function */
-       refcount_set(&aiocb->ki_refcnt, 2);
-
        mask = vfs_poll(req->file, &apt.pt) & req->events;
        if (unlikely(!req->head)) {
                /* we did not manage to set up a waitqueue, done */
@@ -1782,7 +1785,6 @@ out:
 
        if (mask)
                aio_poll_complete(aiocb, mask);
-       iocb_put(aiocb);
        return 0;
 }
 
@@ -1873,18 +1875,21 @@ static int __io_submit_one(struct kioctx *ctx, const struct iocb *iocb,
                break;
        }
 
+       /* Done with the synchronous reference */
+       iocb_put(req);
+
        /*
         * If ret is 0, we'd either done aio_complete() ourselves or have
         * arranged for that to be done asynchronously.  Anything non-zero
         * means that we need to destroy req ourselves.
         */
-       if (ret)
-               goto out_put_req;
-       return 0;
+       if (!ret)
+               return 0;
+
 out_put_req:
        if (req->ki_eventfd)
                eventfd_ctx_put(req->ki_eventfd);
-       iocb_put(req);
+       iocb_destroy(req);
 out_put_reqs_available:
        put_reqs_available(ctx, 1);
        return ret;