[PATCH v1] vhost: release memory objects in error path

[PATCH v1] vhost: release memory objects in error path

[P J P]

From: Prasad Pandit <pjp@fedoraproject.org>

vhost_dev_start function does not release memory objects in case
of an error. This may crash the guest with:

  stack trace of thread 125653:
  Program terminated with signal SIGSEGV, Segmentation fault
  #0  memory_listener_register (qemu-kvm + 0x6cda0f)
  #1  vhost_dev_start (qemu-kvm + 0x699301)
  #2  vhost_net_start (qemu-kvm + 0x45b03f)
  #3  virtio_net_set_status (qemu-kvm + 0x665672)
  #4  qmp_set_link (qemu-kvm + 0x548fd5)
  #5  net_vhost_user_event (qemu-kvm + 0x552c45)
  #6  tcp_chr_connect (qemu-kvm + 0x88d473)
  #7  tcp_chr_new_client (qemu-kvm + 0x88cf83)
  #8  tcp_chr_accept (qemu-kvm + 0x88b429)
  #9  qio_net_listener_channel_func (qemu-kvm + 0x7ac07c)
  #10 g_main_context_dispatch (libglib-2.0.so.0 + 0x54e2f)

Release memory_listener and virtqueue objects in the error path.

Signed-off-by: Prasad Pandit <pjp@fedoraproject.org>
---
 hw/virtio/vhost.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

v1: log debug message when event_notifier_init fails

diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
index 23da579ce2..1de3029ae7 100644
--- a/hw/virtio/vhost.c
+++ b/hw/virtio/vhost.c
@@ -1942,7 +1942,8 @@ int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev, bool vrings)
     r = event_notifier_init(
         &hdev->vqs[VHOST_QUEUE_NUM_CONFIG_INR].masked_config_notifier, 0);
     if (r < 0) {
-        return r;
+        VHOST_OPS_DEBUG(r, "event_notifier_init failed");
+        goto fail_vq;
     }
     event_notifier_test_and_clear(
         &hdev->vqs[VHOST_QUEUE_NUM_CONFIG_INR].masked_config_notifier);
@@ -2004,6 +2005,9 @@ fail_vq:
     }
 
 fail_mem:
+    if (vhost_dev_has_iommu(hdev)) {
+        memory_listener_unregister(&hdev->iommu_listener);
+    }
 fail_features:
     vdev->vhost_started = false;
     hdev->started = false;
-- 
2.40.1

Re: [PATCH v1] vhost: release memory objects in error path

[Peter Xu]

On Fri, May 26, 2023 at 01:10:51PM +0530, P J P wrote:
> From: Prasad Pandit <pjp@fedoraproject.org>
> 
> vhost_dev_start function does not release memory objects in case
> of an error. This may crash the guest with:
> 
>   stack trace of thread 125653:
>   Program terminated with signal SIGSEGV, Segmentation fault
>   #0  memory_listener_register (qemu-kvm + 0x6cda0f)
>   #1  vhost_dev_start (qemu-kvm + 0x699301)
>   #2  vhost_net_start (qemu-kvm + 0x45b03f)
>   #3  virtio_net_set_status (qemu-kvm + 0x665672)
>   #4  qmp_set_link (qemu-kvm + 0x548fd5)
>   #5  net_vhost_user_event (qemu-kvm + 0x552c45)
>   #6  tcp_chr_connect (qemu-kvm + 0x88d473)
>   #7  tcp_chr_new_client (qemu-kvm + 0x88cf83)
>   #8  tcp_chr_accept (qemu-kvm + 0x88b429)
>   #9  qio_net_listener_channel_func (qemu-kvm + 0x7ac07c)
>   #10 g_main_context_dispatch (libglib-2.0.so.0 + 0x54e2f)
> 
> Release memory_listener and virtqueue objects in the error path.

IMHO it'll still good to mention that it's not event_notifier_init() that
failed but vhost_set_mem_table(), since that is probably not crystal clear.
The event_notifier_init() change is just something to fix altogether (or
with a separate patch, but this is fine by me).

Please also consider adding a "Bugzilla:" link to the original bug report.

> 
> Signed-off-by: Prasad Pandit <pjp@fedoraproject.org>

Reviewed-by: Peter Xu <peterx@redhat.com>

Thanks!

-- 
Peter Xu