From: Ilya Maximets <i.maximets@samsung.com>
To: qemu-devel@nongnu.org, "Michael S. Tsirkin" <mst@redhat.com>
Cc: Ilya Maximets <i.maximets@samsung.com>,
Jason Wang <jasowang@redhat.com>,
Dyasly Sergey <s.dyasly@samsung.com>
Subject: [Qemu-devel] [PATCH 2/4] vhost: prevent double stop of vhost_net device.
Date: Wed, 30 Mar 2016 18:14:07 +0300 [thread overview]
Message-ID: <1459350849-31989-3-git-send-email-i.maximets@samsung.com> (raw)
In-Reply-To: <1459350849-31989-1-git-send-email-i.maximets@samsung.com>
There are few control flows in vhost's implementation where
vhost_net_stop() may be re entered several times in a row. This
leads to triggering of assertion while duble freeing of same resources.
For example, if vhost-user socket disconnection occured:
[-------------------------------- cut -----------------------------------]
[guest]# echo -n '0000:00:01.0' > /sys/bus/pci/drivers/virtio-pci/unbind
qemu: Failed to read msg header. Read 0 instead of 12. Original request 11.
qemu: Failed to read msg header. Read 0 instead of 12. Original request 11.
qemu: Failed to read msg header. Read 0 instead of 12. Original request 11.
qemu: Failed to read msg header. Read 0 instead of 12. Original request 11.
qemu: /qemu/memory.c:1852: memory_region_del_eventfd:
Assertion `i != mr->ioeventfd_nb' failed.
Child terminated with signal = 0x6 (SIGABRT)
GDBserver exiting
[-------------------------------- cut -----------------------------------]
[host]# gdb
Breakpoint 1 in virtio_pci_set_host_notifier_internal()
#0 virtio_pci_set_host_notifier_internal
#1 vhost_dev_disable_notifiers
#2 vhost_net_stop_one
#3 vhost_net_stop (dev=dev@entry=0x13a45f8, ncs=0x1ce89f0, <...>)
#4 virtio_net_vhost_status
#5 virtio_net_set_status
#6 qmp_set_link (<...>, up=up@entry=false, <...>)
#7 net_vhost_user_event (<...>, event=5)
#8 qemu_chr_be_event
#9 tcp_chr_disconnect
#10 tcp_chr_sync_read
#11 qemu_chr_fe_read_all
#12 vhost_user_read
#13 vhost_user_get_vring_base
#14 vhost_virtqueue_stop
#15 vhost_dev_stop
#16 vhost_net_stop_one
#17 vhost_net_stop (dev=dev@entry=0x13a45f8, ncs=0x1ce89f0, <...>)
#18 virtio_net_vhost_status
#19 virtio_net_set_status
#20 virtio_set_status
<...>
[-------------------------------- cut -----------------------------------]
In example above assertion will fail when control will be brought back
to function at #17 and it will try to free 'eventfd' that was already
freed at call #3.
Fix that by disallowing execution of vhost_net_stop() if we're
already inside of it.
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
---
hw/net/vhost_net.c | 8 ++++++++
hw/net/virtio-net.c | 1 +
include/hw/virtio/virtio-net.h | 1 +
3 files changed, 10 insertions(+)
diff --git a/hw/net/vhost_net.c b/hw/net/vhost_net.c
index 55d5456..0996e5d 100644
--- a/hw/net/vhost_net.c
+++ b/hw/net/vhost_net.c
@@ -335,8 +335,14 @@ void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
VirtioBusState *vbus = VIRTIO_BUS(qbus);
VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
+ VirtIONet *n = VIRTIO_NET(dev);
int i, r;
+ if (n->vhost_stopping_in_progress) {
+ return;
+ }
+ n->vhost_stopping_in_progress = 1;
+
for (i = 0; i < total_queues; i++) {
vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
put_vhost_net(ncs[i].peer);
@@ -348,6 +354,8 @@ void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
fflush(stderr);
}
assert(r >= 0);
+
+ n->vhost_stopping_in_progress = 0;
}
void vhost_net_cleanup(struct vhost_net *net)
diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index e5456ef..9b1539c 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -149,6 +149,7 @@ static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
}
n->vhost_started = 1;
+ n->vhost_stopping_in_progress = 0;
r = vhost_net_start(vdev, n->nic->ncs, queues);
if (r < 0) {
error_report("unable to start vhost net: %d: "
diff --git a/include/hw/virtio/virtio-net.h b/include/hw/virtio/virtio-net.h
index 0cabdb6..fc07385 100644
--- a/include/hw/virtio/virtio-net.h
+++ b/include/hw/virtio/virtio-net.h
@@ -74,6 +74,7 @@ typedef struct VirtIONet {
uint8_t nouni;
uint8_t nobcast;
uint8_t vhost_started;
+ uint8_t vhost_stopping_in_progress;
struct {
uint32_t in_use;
uint32_t first_multi;
--
2.5.0
next prev parent reply other threads:[~2016-03-30 15:14 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-03-30 15:14 [Qemu-devel] [PATCH 0/4] Fix QEMU crash on vhost-user socket disconnect Ilya Maximets
2016-03-30 15:14 ` [Qemu-devel] [PATCH 1/4] vhost-user: fix crash on " Ilya Maximets
2016-03-30 15:14 ` Ilya Maximets [this message]
2016-03-30 15:14 ` [Qemu-devel] [PATCH 3/4] vhost: check for vhost_net device validity Ilya Maximets
2016-03-30 15:14 ` [Qemu-devel] [PATCH 4/4] net: notify about link status only if it changed Ilya Maximets
2016-03-30 17:01 ` [Qemu-devel] [PATCH 0/4] Fix QEMU crash on vhost-user socket disconnect Michael S. Tsirkin
2016-03-31 6:02 ` Ilya Maximets
2016-03-31 9:21 ` Michael S. Tsirkin
2016-03-31 10:48 ` Ilya Maximets
2016-04-05 10:46 ` Michael S. Tsirkin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1459350849-31989-3-git-send-email-i.maximets@samsung.com \
--to=i.maximets@samsung.com \
--cc=jasowang@redhat.com \
--cc=mst@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=s.dyasly@samsung.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).