[Qemu-devel] [PULL 01/15] scsi: revert change to scsi_req_cancel_async and add assertions

[Paolo Bonzini <pbonzini@redhat.com>]

Fam Zheng noticed that the change in commit 36896bf ("scsi: always call
notifier on async cancellation", 2015-12-16) could cause a leak of
the request; scsi_req_cancel_async now calls scsi_req_ref
multiple times for multiple cancellations, but there is only
one call to scsi_req_cancel_complete.

So revert the patch and instead assert that the problematic case (a call
to scsi_req_cancel_async after the aiocb has been completed) cannot
happen.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/scsi/scsi-bus.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/hw/scsi/scsi-bus.c b/hw/scsi/scsi-bus.c
index 00bddc9..378bf4d 100644
--- a/hw/scsi/scsi-bus.c
+++ b/hw/scsi/scsi-bus.c
@@ -1759,6 +1759,15 @@ void scsi_req_cancel_async(SCSIRequest *req, Notifier *notifier)
     if (notifier) {
         notifier_list_add(&req->cancel_notifiers, notifier);
     }
+    if (req->io_canceled) {
+        /* A blk_aio_cancel_async is pending; when it finishes,
+         * scsi_req_cancel_complete will be called and will
+         * call the notifier we just added.  Just wait for that.
+         */
+        assert(req->aiocb);
+        return;
+    }
+    /* Dropped in scsi_req_cancel_complete.  */
     scsi_req_ref(req);
     scsi_req_dequeue(req);
     req->io_canceled = true;
@@ -1775,6 +1784,8 @@ void scsi_req_cancel(SCSIRequest *req)
     if (!req->enqueued) {
         return;
     }
+    assert(!req->io_canceled);
+    /* Dropped in scsi_req_cancel_complete.  */
     scsi_req_ref(req);
     scsi_req_dequeue(req);
     req->io_canceled = true;
-- 
1.8.3.1