On 01/14/2014 09:49 AM, Paolo Bonzini wrote:
Il 14/01/2014 15:24, Eric Farman ha scritto:
There is still a small window that occurs when a cancel I/O affects
an asynchronous I/O operation that hasn't started. In other words,
when the residual data length equals the expected data length.
Today, the routine virtio_scsi_command_complete fails because the
VirtIOSCSIReq pointer (from the hba_private field in SCSIRequest)
was cleared earlier when virtio_scsi_complete_req was called by
the virtio_scsi_request_cancelled routine. As a result, the
virtio_scsi_command_complete routine needs to simply return when
it is processing a SCSIRequest block that was marked canceled.
Signed-off-by: Eric Farman <far...@linux.vnet.ibm.com>
---
hw/scsi/virtio-scsi.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
index a8fde04..49a9576 100644
--- a/hw/scsi/virtio-scsi.c
+++ b/hw/scsi/virtio-scsi.c
@@ -306,6 +306,10 @@ static void virtio_scsi_command_complete(SCSIRequest *r,
uint32_t status,
VirtIOSCSIReq *req = r->hba_private;
uint32_t sense_len;
+ if (r->io_canceled) {
+ return;
+ }
+
req->resp.cmd->response = VIRTIO_SCSI_S_OK;
req->resp.cmd->status = status;
if (req->resp.cmd->status == GOOD) {
Reviewed-by: Paolo Bonzini <pbonz...@redhat.com>
Can you please look more closely into whether patch 1 is really
necessary, include the small fix I posted into that review, and post v2
of the series?
Heh, that's fantastic. I pulled patch 1 out, ran the tests that hit
into these problems, didn't see anything. Pulled out patch 2, reran the
tests, hit them almost immediately. Did this a few times, with the same
results. So I'll fix up another series, and post shortly.
Thanks for the feedback and the review!
- Eric
Thanks!
Paolo
