在 2023/2/2 下午6:31, Michael S. Tsirkin 写道:
On Thu, Feb 02, 2023 at 05:07:04PM +0800, Heng Qi wrote:
在 2023/2/2 下午4:16, Michael S. Tsirkin 写道:
On Thu, Feb 02, 2023 at 04:14:51PM +0800, Heng Qi wrote:
在 2023/2/2 下午4:07, Michael S. Tsirkin 写道:
On Tue, Jan 31, 2023 at 04:50:04PM +0800, Heng Qi wrote:
When the single-buffer xdp is loaded and after xdp_linearize_page()
is called, *num_buf becomes 0 and (*num_buf - 1) may overflow into
a large integer in virtnet_build_xdp_buff_mrg(), resulting in
unexpected packet dropping.
Fixes: ef75cb51f139 ("virtio-net: build xdp_buff with multi buffers")
Signed-off-by: Heng Qi <hengqi@xxxxxxxxxxxxxxxxx>
Reviewed-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
Signed-off-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
---
v1->v2:
- Change the type of num_buf from unsigned int to int. @Michael S . Tsirkin
- Some cleaner codes. @Michael S . Tsirkin
drivers/net/virtio_net.c | 15 +++++++++------
1 file changed, 9 insertions(+), 6 deletions(-)
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index aaa6fe9b214a..8102861785a2 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -716,7 +716,7 @@ static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
* have enough headroom.
*/
static struct page *xdp_linearize_page(struct receive_queue *rq,
- u16 *num_buf,
+ int *num_buf,
struct page *p,
int offset,
int page_off,
@@ -816,7 +816,7 @@ static struct sk_buff *receive_small(struct net_device *dev,
if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
int offset = buf - page_address(page) + header_offset;
unsigned int tlen = len + vi->hdr_len;
- u16 num_buf = 1;
+ int num_buf = 1;
xdp_headroom = virtnet_get_headroom(vi);
header_offset = VIRTNET_RX_PAD + xdp_headroom;
@@ -989,7 +989,7 @@ static int virtnet_build_xdp_buff_mrg(struct net_device *dev,
void *buf,
unsigned int len,
unsigned int frame_sz,
- u16 *num_buf,
+ int *num_buf,
unsigned int *xdp_frags_truesize,
struct virtnet_rq_stats *stats)
{
@@ -1007,6 +1007,9 @@ static int virtnet_build_xdp_buff_mrg(struct net_device *dev,
xdp_prepare_buff(xdp, buf - VIRTIO_XDP_HEADROOM,
VIRTIO_XDP_HEADROOM + vi->hdr_len, len - vi->hdr_len, true);
+ if (!*num_buf)
+ return 0;
+
if (*num_buf > 1) {
/* If we want to build multi-buffer xdp, we need
* to specify that the flags of xdp_buff have the
Ouch. Why is this here? Merged so pls remove by a follow up patch, the
rest of the code handles 0 fine. I'm not sure this introduces a bug by
we don't want spaghetti code.
Yes it would work without this, but I was keeping this because I wanted it
to handle 0 early and exit early.
Do you want to remove this?
Thanks.
why do you want to exit early?
If num_buf is 0, we don't need to judge the subsequent process, because the
latter process
is used to build multi-buffer xdp, but this fix solves the possible problems
of single-buffer xdp.
Thanks.
An optimization then? As any optimization I'd like to see some numbers.
Should have been documented as such not as part of a bugfix.
Sure, I'm going to remove this with a follow-up patch.😁
Thanks.
@@ -1020,10 +1023,10 @@ static int virtnet_build_xdp_buff_mrg(struct net_device *dev,
shinfo->xdp_frags_size = 0;
}
- if ((*num_buf - 1) > MAX_SKB_FRAGS)
+ if (*num_buf > MAX_SKB_FRAGS + 1)
return -EINVAL;
- while ((--*num_buf) >= 1) {
+ while (--*num_buf > 0) {
buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx);
if (unlikely(!buf)) {
pr_debug("%s: rx error: %d buffers out of %d missing\n",
@@ -1076,7 +1079,7 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
struct virtnet_rq_stats *stats)
{
struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
- u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
+ int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
struct page *page = virt_to_head_page(buf);
int offset = buf - page_address(page);
struct sk_buff *head_skb, *curr_skb;
--
2.19.1.6.gb485710b
