Blame SOURCES/0033-virtio-scsi-Reject-scsi-cd-if-data-plane-enabled-RHE.patch

9bac43
From c9c4f117d8b507c2f86035c282d537c0a327364f Mon Sep 17 00:00:00 2001
9bac43
From: Fam Zheng <famz@redhat.com>
9bac43
Date: Wed, 14 Jun 2017 15:37:01 +0200
9bac43
Subject: virtio-scsi: Reject scsi-cd if data plane enabled [RHEL only]
9bac43
9bac43
RH-Author: Fam Zheng <famz@redhat.com>
9bac43
Message-id: <20170614153701.14757-1-famz@redhat.com>
9bac43
Patchwork-id: 75613
9bac43
O-Subject: [RHV-7.4 qemu-kvm-rhev PATCH v3] virtio-scsi: Reject scsi-cd if data plane enabled [RHEL only]
9bac43
Bugzilla: 1378816
9bac43
RH-Acked-by: Paolo Bonzini <pbonzini@redhat.com>
9bac43
RH-Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
9bac43
RH-Acked-by: Max Reitz <mreitz@redhat.com>
9bac43
9bac43
We need a fix for RHEL 7.4 and 7.3.z, but unfortunately upstream isn't
9bac43
ready. If it were, the changes will be too invasive. To have an idea:
9bac43
9bac43
https://lists.gnu.org/archive/html/qemu-devel/2017-05/msg05400.html
9bac43
9bac43
is an incomplete attempt to fix part of the issue, and the remaining
9bac43
work unfortunately involve even more complex changes.
9bac43
9bac43
As a band-aid, this partially reverts the effect of ef8875b
9bac43
(virtio-scsi: Remove op blocker for dataplane, since v2.7). We cannot
9bac43
simply revert that commit as a whole because we already shipped it in
9bac43
qemu-kvm-rhev 7.3, since when, block jobs has been possible.  We should
9bac43
only block what has been broken.  Also, faithfully reverting the above
9bac43
commit means adding back the removed op blocker, but that is not enough,
9bac43
because it still crashes when inserting media into an initially empty
9bac43
scsi-cd.
9bac43
9bac43
All in all, scsi-cd on virtio-scsi-dataplane has basically been unusable
9bac43
unless the scsi-cd never enters an empty state, so, disable it
9bac43
altogether.  Otherwise it would be much more difficult to avoid
9bac43
crashing.
9bac43
9bac43
Signed-off-by: Fam Zheng <famz@redhat.com>
9bac43
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
9bac43
(cherry picked from commit b0caf00bbc35c7d89e02999bdce86e1f867728e8)
9bac43
---
9bac43
 hw/scsi/virtio-scsi.c | 9 +++++++++
9bac43
 1 file changed, 9 insertions(+)
9bac43
9bac43
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
9bac43
index eb63944..2635c38 100644
9bac43
--- a/hw/scsi/virtio-scsi.c
9bac43
+++ b/hw/scsi/virtio-scsi.c
9bac43
@@ -790,6 +790,15 @@ static void virtio_scsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev,
9bac43
     VirtIOSCSI *s = VIRTIO_SCSI(vdev);
9bac43
     SCSIDevice *sd = SCSI_DEVICE(dev);
9bac43
 
9bac43
+    /* XXX: Remove this check once block backend is capable of handling
9bac43
+     * AioContext change upon eject/insert.
9bac43
+     * s->ctx is NULL if ioeventfd is off, s->ctx is qemu_get_aio_context() if
9bac43
+     * data plane is not used, both cases are safe for scsi-cd. */
9bac43
+    if (s->ctx && s->ctx != qemu_get_aio_context() &&
9bac43
+        object_dynamic_cast(OBJECT(dev), "scsi-cd")) {
9bac43
+        error_setg(errp, "scsi-cd is not supported by data plane");
9bac43
+        return;
9bac43
+    }
9bac43
     if (s->ctx && !s->dataplane_fenced) {
9bac43
         if (blk_op_is_blocked(sd->conf.blk, BLOCK_OP_TYPE_DATAPLANE, errp)) {
9bac43
             return;
9bac43
-- 
9bac43
1.8.3.1
9bac43