[PATCH 1/2] virtio_net: fix error handling for mergeable buffers

Eric Dumazet noticed that if we encounter an error
when processing a mergeable buffer, we don't
dequeue all of the buffers from this packet,
the result is almost sure to be loss of networking.

Jason Wang noticed that we also leak a page and that we don't decrement
the rq buf count, so we won't repost buffers (a resource leak).

Fix both issues, and also make the logic a bit more
robust against device errors by not looping when e.g. because of a leak
like the one we are fixing here the number of buffers is 0.

Cc: Rusty Russell <redacted>
Cc: Michael Dalton <redacted>
Reported-by: Eric Dumazet <edumazet@google.com>
Reported-by: Jason Wang <redacted>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---

Note: this bugfix is needed on stable too, but backport
might not be trivial.
I'll send a backport for stable separately.

 drivers/net/virtio_net.c | 84 ++++++++++++++++++++++++++++++------------------
 1 file changed, 52 insertions(+), 32 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 7bab4de..0e6ea69 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -299,41 +299,53 @@ static struct sk_buff *page_to_skb(struct receive_queue *rq,
 	return skb;
 }
 
-static int receive_mergeable(struct receive_queue *rq, struct sk_buff *head_skb)
+static struct sk_buff *receive_mergeable(struct net_device *dev,
+					 struct receive_queue *rq,
+					 void *buf,
+					 unsigned int len)
 {
-	struct skb_vnet_hdr *hdr = skb_vnet_hdr(head_skb);
+	struct skb_vnet_hdr *hdr = buf;
+	int num_buf = hdr->mhdr.num_buffers;
+	struct page *page = virt_to_head_page(buf);
+	int offset = buf - page_address(page);
+	struct sk_buff *head_skb = page_to_skb(rq, page, offset, len,
+					       MERGE_BUFFER_LEN);
 	struct sk_buff *curr_skb = head_skb;
-	char *buf;
-	struct page *page;
-	int num_buf, len, offset;
 
-	num_buf = hdr->mhdr.num_buffers;
+	if (unlikely(!curr_skb))
+		goto err_skb;
+
 	while (--num_buf) {
-		int num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
+		int num_skb_frags;
+
 		buf = virtqueue_get_buf(rq->vq, &len);
 		if (unlikely(!buf)) {
-			pr_debug("%s: rx error: %d buffers missing\n",
-				 head_skb->dev->name, hdr->mhdr.num_buffers);
-			head_skb->dev->stats.rx_length_errors++;
-			return -EINVAL;
+			pr_debug("%s: rx error: %d buffers out of %d missing\n",
+				 dev->name, num_buf, hdr->mhdr.num_buffers);
+			dev->stats.rx_length_errors++;
+			goto err_buf;
 		}
 		if (unlikely(len > MERGE_BUFFER_LEN)) {
 			pr_debug("%s: rx error: merge buffer too long\n",
-				 head_skb->dev->name);
+				 dev->name);
 			len = MERGE_BUFFER_LEN;
 		}
+
+		page = virt_to_head_page(buf);
+		--rq->num;
+
+		num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
 		if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
 			struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
-			if (unlikely(!nskb)) {
-				head_skb->dev->stats.rx_dropped++;
-				return -ENOMEM;
-			}
+
+			if (unlikely(!nskb))
+				goto err_skb;
 			if (curr_skb == head_skb)
 				skb_shinfo(curr_skb)->frag_list = nskb;
 			else
 				curr_skb->next = nskb;
-			curr_skb = nskb;
 			head_skb->truesize += nskb->truesize;
+			curr_skb = nskb;
 			num_skb_frags = 0;
 		}
 		if (curr_skb != head_skb) {

@@ -341,8 +353,7 @@ static int receive_mergeable(struct receive_queue *rq, struct sk_buff *head_skb)
 			head_skb->len += len;
 			head_skb->truesize += MERGE_BUFFER_LEN;
 		}
-		page = virt_to_head_page(buf);
-		offset = buf - (char *)page_address(page);
+		offset = buf - page_address(page);
 		if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
 			put_page(page);
 			skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,

@@ -351,9 +362,28 @@ static int receive_mergeable(struct receive_queue *rq, struct sk_buff *head_skb)
 			skb_add_rx_frag(curr_skb, num_skb_frags, page,
 					offset, len, MERGE_BUFFER_LEN);
 		}
+	}
+
+	return head_skb;
+
+err_skb:
+	put_page(page);
+err_buf:
+	dev->stats.rx_dropped++;
+	dev_kfree_skb(head_skb);
+	while (--num_buf) {
+		buf = virtqueue_get_buf(rq->vq, &len);
+		if (unlikely(!buf)) {
+			pr_debug("%s: rx error: %d buffers missing\n",
+				 dev->name, num_buf);
+			dev->stats.rx_length_errors++;
+			break;
+		}
+		page = virt_to_head_page(buf);
+		put_page(page);
 		--rq->num;
 	}
-	return 0;
+	return NULL;
 }
 
 static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)

@@ -382,19 +412,9 @@ static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
 		len -= sizeof(struct virtio_net_hdr);
 		skb_trim(skb, len);
 	} else if (vi->mergeable_rx_bufs) {
-		struct page *page = virt_to_head_page(buf);
-		skb = page_to_skb(rq, page,
-				  (char *)buf - (char *)page_address(page),
-				  len, MERGE_BUFFER_LEN);
-		if (unlikely(!skb)) {
-			dev->stats.rx_dropped++;
-			put_page(page);
+		skb = receive_mergeable(dev, rq, buf, len);
+		if (unlikely(!skb))
 			return;
-		}
-		if (receive_mergeable(rq, skb)) {
-			dev_kfree_skb(skb);
-			return;
-		}
 	} else {
 		page = buf;
 		skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);

-- 
MST

receive mergeable now handles errors internally.
Do same for big and small packet paths, otherwise
the logic is too hard to follow.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---

While I can't point at a bug this fixes, I'm not sure
there's no bug in the existing logic.
So not exactly a bug fix bug I think it's justified for net.

 drivers/net/virtio_net.c | 53 +++++++++++++++++++++++++++++++++---------------
 1 file changed, 37 insertions(+), 16 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 0e6ea69..97c6212 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -299,6 +299,35 @@ static struct sk_buff *page_to_skb(struct receive_queue *rq,
 	return skb;
 }
 
+static struct sk_buff *receive_small(void *buf, unsigned int len)
+{
+	struct sk_buff * skb = buf;
+
+	len -= sizeof(struct virtio_net_hdr);
+	skb_trim(skb, len);
+
+	return skb;
+}
+
+static struct sk_buff *receive_big(struct net_device *dev,
+				   struct receive_queue *rq,
+				   void *buf,
+				   unsigned int len)
+{
+	struct page *page = buf;
+	struct sk_buff *skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);
+
+	if (unlikely(!skb))
+		goto err;
+
+	return skb;
+
+err:
+	dev->stats.rx_dropped++;
+	give_pages(rq, page);
+	return NULL;
+}
+
 static struct sk_buff *receive_mergeable(struct net_device *dev,
 					 struct receive_queue *rq,
 					 void *buf,

@@ -407,23 +436,15 @@ static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
 		return;
 	}
 
-	if (!vi->mergeable_rx_bufs && !vi->big_packets) {
-		skb = buf;
-		len -= sizeof(struct virtio_net_hdr);
-		skb_trim(skb, len);
-	} else if (vi->mergeable_rx_bufs) {
+	if (vi->mergeable_rx_bufs)
 		skb = receive_mergeable(dev, rq, buf, len);
-		if (unlikely(!skb))
-			return;
-	} else {
-		page = buf;
-		skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);
-		if (unlikely(!skb)) {
-			dev->stats.rx_dropped++;
-			give_pages(rq, page);
-			return;
-		}
-	}
+	else if (vi->big_packets)
+		skb = receive_big(dev, rq, buf, len);
+	else
+		skb = receive_small(buf, len);
+
+	if (unlikely(!skb))
+		return;
 
 	hdr = skb_vnet_hdr(skb);
 

-- 
MST

On 11/28/2013 12:31 AM, Michael S. Tsirkin wrote:

Eric Dumazet noticed that if we encounter an error
when processing a mergeable buffer, we don't
dequeue all of the buffers from this packet,
the result is almost sure to be loss of networking.

Jason Wang noticed that we also leak a page and that we don't decrement
the rq buf count, so we won't repost buffers (a resource leak).

Fix both issues, and also make the logic a bit more
robust against device errors by not looping when e.g. because of a leak
like the one we are fixing here the number of buffers is 0.

Cc: Rusty Russell <redacted>
Cc: Michael Dalton <redacted>
Reported-by: Eric Dumazet <edumazet@google.com>
Reported-by: Jason Wang <redacted>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---

Note: this bugfix is needed on stable too, but backport
might not be trivial.
I'll send a backport for stable separately.

That will be fine.

Not sure it's correct here. Since the we break immediately if buffer is
missed in err_buf and rx_length_error will be miss counted.

Maybe an ERR_PTR(-EINVAL) is better.

 		}
 		if (unlikely(len > MERGE_BUFFER_LEN)) {
 			pr_debug("%s: rx error: merge buffer too long\n",
-				 head_skb->dev->name);
+				 dev->name);
 			len = MERGE_BUFFER_LEN;
 		}
+
+		page = virt_to_head_page(buf);
+		--rq->num;
+
+		num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
 		if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
 			struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
-			if (unlikely(!nskb)) {
-				head_skb->dev->stats.rx_dropped++;
-				return -ENOMEM;
-			}
+
+			if (unlikely(!nskb))
+				goto err_skb;
 			if (curr_skb == head_skb)
 				skb_shinfo(curr_skb)->frag_list = nskb;
 			else
 				curr_skb->next = nskb;
-			curr_skb = nskb;
 			head_skb->truesize += nskb->truesize;
+			curr_skb = nskb;

This change seems unnecessary.

Other looks good.

On 11/28/2013 12:31 AM, Michael S. Tsirkin wrote:

Acked-by: Jason Wang <redacted>

On Thu, Nov 28, 2013 at 11:14:45AM +0800, Jason Wang wrote:

On 11/28/2013 12:31 AM, Michael S. Tsirkin wrote:

quoted

Eric Dumazet noticed that if we encounter an error
when processing a mergeable buffer, we don't
dequeue all of the buffers from this packet,
the result is almost sure to be loss of networking.

Jason Wang noticed that we also leak a page and that we don't decrement
the rq buf count, so we won't repost buffers (a resource leak).

Fix both issues, and also make the logic a bit more
robust against device errors by not looping when e.g. because of a leak
like the one we are fixing here the number of buffers is 0.

Cc: Rusty Russell <redacted>
Cc: Michael Dalton <redacted>
Reported-by: Eric Dumazet <edumazet@google.com>
Reported-by: Jason Wang <redacted>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---

Note: this bugfix is needed on stable too, but backport
might not be trivial.
I'll send a backport for stable separately.

That will be fine.

quoted

 drivers/net/virtio_net.c | 84 ++++++++++++++++++++++++++++++------------------
 1 file changed, 52 insertions(+), 32 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 7bab4de..0e6ea69 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -299,41 +299,53 @@ static struct sk_buff *page_to_skb(struct receive_queue *rq,
 	return skb;
 }
 
-static int receive_mergeable(struct receive_queue *rq, struct sk_buff *head_skb)
+static struct sk_buff *receive_mergeable(struct net_device *dev,
+					 struct receive_queue *rq,
+					 void *buf,
+					 unsigned int len)
 {
-	struct skb_vnet_hdr *hdr = skb_vnet_hdr(head_skb);
+	struct skb_vnet_hdr *hdr = buf;
+	int num_buf = hdr->mhdr.num_buffers;
+	struct page *page = virt_to_head_page(buf);
+	int offset = buf - page_address(page);
+	struct sk_buff *head_skb = page_to_skb(rq, page, offset, len,
+					       MERGE_BUFFER_LEN);
 	struct sk_buff *curr_skb = head_skb;
-	char *buf;
-	struct page *page;
-	int num_buf, len, offset;
 
-	num_buf = hdr->mhdr.num_buffers;
+	if (unlikely(!curr_skb))
+		goto err_skb;
+
 	while (--num_buf) {
-		int num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
+		int num_skb_frags;
+
 		buf = virtqueue_get_buf(rq->vq, &len);
 		if (unlikely(!buf)) {
-			pr_debug("%s: rx error: %d buffers missing\n",
-				 head_skb->dev->name, hdr->mhdr.num_buffers);
-			head_skb->dev->stats.rx_length_errors++;
-			return -EINVAL;
+			pr_debug("%s: rx error: %d buffers out of %d missing\n",
+				 dev->name, num_buf, hdr->mhdr.num_buffers);
+			dev->stats.rx_length_errors++;
+			goto err_buf;

Not sure it's correct here. Since the we break immediately if buffer is
missed in err_buf and rx_length_error will be miss counted.

Note that this never happens: it's an internal device error, we are
just trying to be defensive here. I would not worry about getting
error counters right in this case.

But I agree - since we know virtqueue_get_buf failed it's useless to try
and get more. So let's move the loop before the err_buf label.

Maybe an ERR_PTR(-EINVAL) is better.

This will just make callers worry about checking IS_ERR, and
caller returns void so it has nothing to do with the return code anyway.

quoted

 		}
 		if (unlikely(len > MERGE_BUFFER_LEN)) {
 			pr_debug("%s: rx error: merge buffer too long\n",
-				 head_skb->dev->name);
+				 dev->name);
 			len = MERGE_BUFFER_LEN;
 		}
+
+		page = virt_to_head_page(buf);
+		--rq->num;
+
+		num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
 		if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
 			struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
-			if (unlikely(!nskb)) {
-				head_skb->dev->stats.rx_dropped++;
-				return -ENOMEM;
-			}
+
+			if (unlikely(!nskb))
+				goto err_skb;
 			if (curr_skb == head_skb)
 				skb_shinfo(curr_skb)->frag_list = nskb;
 			else
 				curr_skb->next = nskb;
-			curr_skb = nskb;
 			head_skb->truesize += nskb->truesize;
+			curr_skb = nskb;

This change seems unnecessary.

Right, I'll drop it.

Other looks good.

quoted

 			num_skb_frags = 0;
 		}
 		if (curr_skb != head_skb) {

@@ -341,8 +353,7 @@ static int receive_mergeable(struct receive_queue *rq, struct sk_buff *head_skb)
 			head_skb->len += len;
 			head_skb->truesize += MERGE_BUFFER_LEN;
 		}
-		page = virt_to_head_page(buf);
-		offset = buf - (char *)page_address(page);
+		offset = buf - page_address(page);
 		if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
 			put_page(page);
 			skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,

@@ -351,9 +362,28 @@ static int receive_mergeable(struct receive_queue *rq, struct sk_buff *head_skb)
 			skb_add_rx_frag(curr_skb, num_skb_frags, page,
 					offset, len, MERGE_BUFFER_LEN);
 		}
+	}
+
+	return head_skb;
+
+err_skb:
+	put_page(page);
+err_buf:
+	dev->stats.rx_dropped++;
+	dev_kfree_skb(head_skb);
+	while (--num_buf) {
+		buf = virtqueue_get_buf(rq->vq, &len);
+		if (unlikely(!buf)) {
+			pr_debug("%s: rx error: %d buffers missing\n",
+				 dev->name, num_buf);
+			dev->stats.rx_length_errors++;
+			break;
+		}
+		page = virt_to_head_page(buf);
+		put_page(page);
 		--rq->num;
 	}
-	return 0;
+	return NULL;
 }
 
 static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)

@@ -382,19 +412,9 @@ static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
 		len -= sizeof(struct virtio_net_hdr);
 		skb_trim(skb, len);
 	} else if (vi->mergeable_rx_bufs) {
-		struct page *page = virt_to_head_page(buf);
-		skb = page_to_skb(rq, page,
-				  (char *)buf - (char *)page_address(page),
-				  len, MERGE_BUFFER_LEN);
-		if (unlikely(!skb)) {
-			dev->stats.rx_dropped++;
-			put_page(page);
+		skb = receive_mergeable(dev, rq, buf, len);
+		if (unlikely(!skb))
 			return;
-		}
-		if (receive_mergeable(rq, skb)) {
-			dev_kfree_skb(skb);
-			return;
-		}
 	} else {
 		page = buf;
 		skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);