| Message ID | 20241030121334.134794-1-yuri.benditovich@daynix.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | virtio-net: Fix missing hash type and hash value | expand |
On 2024/10/30 21:13, Yuri Benditovich wrote: > virtio_net_receive_rcu does not populate hash value and > hash type in case it needs to change queue number from the > initially defined by tap/tun to one calculated according > to the packet hash. Earlier commit a4c960e places the hash > value/type into local extra_hdr structure but the extra_hdr > is not filled in recursive call to virtio_net_receive_rcu. > Current commit discards the recursive call and just continues > execution with changed net client state and uses correct > virtio-net queue for the packet. > > Fixes: a4c960e ("virtio-net: Do not write hashes to peer buffer") > > Jira link: https://issues.redhat.com/browse/RHEL-59572 > > Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com> Hi, I have submitted a similar patch: https://patchew.org/QEMU/20240915-queue-v1-0-b49bd49b926d@daynix.com/20240915-queue-v1-4-b49bd49b926d@daynix.com/ Jason, please pick the buglink, which was missing in my patch. Regards, Akihiko Odaki > --- > hw/net/virtio-net.c | 9 +++++---- > 1 file changed, 5 insertions(+), 4 deletions(-) > > diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c > index f2104ed364..ed4e33b48c 100644 > --- a/hw/net/virtio-net.c > +++ b/hw/net/virtio-net.c > @@ -1898,7 +1898,7 @@ static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, > size_t size, bool no_rss) > { > VirtIONet *n = qemu_get_nic_opaque(nc); > - VirtIONetQueue *q = virtio_net_get_subqueue(nc); > + VirtIONetQueue *q; > VirtIODevice *vdev = VIRTIO_DEVICE(n); > VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE]; > size_t lens[VIRTQUEUE_MAX_SIZE]; > @@ -1915,12 +1915,13 @@ static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, > if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) { > int index = virtio_net_process_rss(nc, buf, size, &extra_hdr); > if (index >= 0) { > - NetClientState *nc2 = > - qemu_get_subqueue(n->nic, index % n->curr_queue_pairs); > - return virtio_net_receive_rcu(nc2, buf, size, true); > + /* change nc to proper one */ > + nc = qemu_get_subqueue(n->nic, index % n->curr_queue_pairs); > } > } > > + q = virtio_net_get_subqueue(nc); > + > /* hdr_len refers to the header we supply to the guest */ > if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { > return 0;
diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c index f2104ed364..ed4e33b48c 100644 --- a/hw/net/virtio-net.c +++ b/hw/net/virtio-net.c @@ -1898,7 +1898,7 @@ static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, size_t size, bool no_rss) { VirtIONet *n = qemu_get_nic_opaque(nc); - VirtIONetQueue *q = virtio_net_get_subqueue(nc); + VirtIONetQueue *q; VirtIODevice *vdev = VIRTIO_DEVICE(n); VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE]; size_t lens[VIRTQUEUE_MAX_SIZE]; @@ -1915,12 +1915,13 @@ static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) { int index = virtio_net_process_rss(nc, buf, size, &extra_hdr); if (index >= 0) { - NetClientState *nc2 = - qemu_get_subqueue(n->nic, index % n->curr_queue_pairs); - return virtio_net_receive_rcu(nc2, buf, size, true); + /* change nc to proper one */ + nc = qemu_get_subqueue(n->nic, index % n->curr_queue_pairs); } } + q = virtio_net_get_subqueue(nc); + /* hdr_len refers to the header we supply to the guest */ if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { return 0;
virtio_net_receive_rcu does not populate hash value and hash type in case it needs to change queue number from the initially defined by tap/tun to one calculated according to the packet hash. Earlier commit a4c960e places the hash value/type into local extra_hdr structure but the extra_hdr is not filled in recursive call to virtio_net_receive_rcu. Current commit discards the recursive call and just continues execution with changed net client state and uses correct virtio-net queue for the packet. Fixes: a4c960e ("virtio-net: Do not write hashes to peer buffer") Jira link: https://issues.redhat.com/browse/RHEL-59572 Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com> --- hw/net/virtio-net.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-)