Skip to contentSkip to conversation

net: core: introduce netif_skb_dev_features

PATCH In review Florian Westphal 3 participants 5 messages Feb 13, 2014 – Feb 23, 2014

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/2 net: core: introduce netif_skb_dev_features

Will be used by upcoming ipv4 forward path change that needs to determine feature mask using skb->dst->dev instead of skb->dev.

---
Changes since v1:
no more 'illegal_highdma' discards 'const' qualifier from pointer target type'

 include/linux/netdevice.h |  7 ++++++-
 net/core/dev.c            | 22 ++++++++++++----------
 2 files changed, 18 insertions(+), 11 deletions(-)
include/linux/netdevice.h +6 -1
index 440a02e..21d4e6b 100644
@@ -3068,7 +3068,12 @@ void netdev_change_features(struct net_device *dev);
 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
 					struct net_device *dev);
 
-netdev_features_t netif_skb_features(struct sk_buff *skb);
+netdev_features_t netif_skb_dev_features(struct sk_buff *skb,
+					 const struct net_device *dev);
+static inline netdev_features_t netif_skb_features(struct sk_buff *skb)
+{
+	return netif_skb_dev_features(skb, skb->dev);
+}
 
 static inline bool net_gso_ok(netdev_features_t features, int gso_type)
 {
net/core/dev.c +12 -10
index 3721db7..afc4e47 100644
@@ -2420,7 +2420,7 @@ EXPORT_SYMBOL(netdev_rx_csum_fault);
  * 2. No high memory really exists on this machine.
  */
 
-static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
+static int illegal_highdma(const struct net_device *dev, struct sk_buff *skb)
 {
 #ifdef CONFIG_HIGHMEM
 	int i;
@@ -2495,34 +2495,36 @@ static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
 }
 
 static netdev_features_t harmonize_features(struct sk_buff *skb,
-	netdev_features_t features)
+					    const struct net_device *dev,
+					    netdev_features_t features)
 {
 	if (skb->ip_summed != CHECKSUM_NONE &&
 	    !can_checksum_protocol(features, skb_network_protocol(skb))) {
 		features &= ~NETIF_F_ALL_CSUM;
-	} else if (illegal_highdma(skb->dev, skb)) {
+	} else if (illegal_highdma(dev, skb)) {
 		features &= ~NETIF_F_SG;
 	}
 
 	return features;
 }
 
-netdev_features_t netif_skb_features(struct sk_buff *skb)
+netdev_features_t netif_skb_dev_features(struct sk_buff *skb,
+					 const struct net_device *dev)
 {
 	__be16 protocol = skb->protocol;
-	netdev_features_t features = skb->dev->features;
+	netdev_features_t features = dev->features;
 
-	if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
+	if (skb_shinfo(skb)->gso_segs > dev->gso_max_segs)
 		features &= ~NETIF_F_GSO_MASK;
 
 	if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
 		struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
 		protocol = veh->h_vlan_encapsulated_proto;
 	} else if (!vlan_tx_tag_present(skb)) {
-		return harmonize_features(skb, features);
+		return harmonize_features(skb, dev, features);
 	}
 
-	features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
+	features &= (dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
 					       NETIF_F_HW_VLAN_STAG_TX);
 
 	if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
@@ -2530,9 +2532,9 @@ netdev_features_t netif_skb_features(struct sk_buff *skb)
 				NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
 				NETIF_F_HW_VLAN_STAG_TX;
 
-	return harmonize_features(skb, features);
+	return harmonize_features(skb, dev, features);
 }
-EXPORT_SYMBOL(netif_skb_features);
+EXPORT_SYMBOL(netif_skb_dev_features);
 
 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
 			struct netdev_queue *txq)
Signature · 1 line
1.8.1.5
#1 lore raw

PATCH v6 2/2 net: ip, ipv6: handle gso skbs in forwarding path

Marcelo Ricardo Leitner reported problems when the forwarding link path has a lower mtu than the incoming one if the inbound interface supports GRO.

Given:
Host <mtu1500> R1 <mtu1200> R2

Host sends tcp stream which is routed via R1 and R2. R1 performs GRO.

In this case, the kernel will fail to send ICMP fragmentation needed messages (or pkt too big for ipv6), as GSO packets currently bypass dstmtu checks in forward path. Instead, Linux tries to send out packets exceeding the mtu.

When locking route MTU on Host (i.e., no ipv4 DF bit set), R1 does not fragment the packets when forwarding, and again tries to send out packets exceeding R1-R2 link mtu.

This alters the forwarding dstmtu checks to take the individual gso segment lengths into account.

For ipv6, we send out pkt too big error for gso if the individual segments are too big.

For ipv4, we either send icmp fragmentation needed, or, if the DF bit is not set, perform software segmentation and let the output path create fragments when the packet is leaving the machine.
It is not 100% correct as the error message will contain the headers of the GRO skb instead of the original/segmented one, but it seems to work fine in my (limited) tests.

Eric Dumazet suggested to simply shrink mss via ->gso_size to avoid sofware segmentation.

However it turns out that skb_segment() assumes skb nr_frags is related to mss size so we would BUG there. I don't want to mess with it considering Herbert and Eric disagree on what the correct behavior should be.

Hannes Frederic Sowa notes that when we would shrink gso_size skb_segment would then also need to deal with the case where SKB_MAX_FRAGS would be exceeded.

This uses sofware segmentation in the forward path when we hit ipv4 non-DF packets and the outgoing link mtu is too small. Its not perfect, but given the lack of bug reports wrt. GRO fwd being broken this is a rare case anyway. Also its not like this could not be improved later once the dust settles.

---
Changes since v5; propagated Herberts Ack from v5 to changelog. [ resend is due to error in earlier patch in series ]

Changes since V4:
- use new netif_skb_dev_features instead of netif_skb_features as we need dst->dev and not skb->dev (spotted by Hannes Frederic Sowa)

Changes since V3:
- use ip_dst_mtu_maybe_forward instead of dst_mtu
- add comment wrt. DF bit not being set

Changes since V2:
- make this thing apply to current -net tree
- kill unused variables in ip_forward/ip6_output

Changes since V1:
suggestions from Eric Dumazet:
- skip more expensive computation for small packets in fwd path
- use netif_skb_features() feature mask and remove GSO flags instead of using 0 feature set.

 include/linux/skbuff.h | 17 ++++++++++++
 net/ipv4/ip_forward.c  | 71 ++++++++++++++++++++++++++++++++++++++++++++++++--
 net/ipv6/ip6_output.c  | 17 ++++++++++--
 3 files changed, 101 insertions(+), 4 deletions(-)
include/linux/skbuff.h +17 -0
index f589c9a..3ebbbe7 100644
@@ -2916,5 +2916,22 @@ static inline bool skb_head_is_locked(const struct sk_buff *skb)
 {
 	return !skb->head_frag || skb_cloned(skb);
 }
+
+/**
+ * skb_gso_network_seglen - Return length of individual segments of a gso packet
+ *
+ * @skb: GSO skb
+ *
+ * skb_gso_network_seglen is used to determine the real size of the
+ * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
+ *
+ * The MAC/L2 header is not accounted for.
+ */
+static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
+{
+	unsigned int hdr_len = skb_transport_header(skb) -
+			       skb_network_header(skb);
+	return hdr_len + skb_gso_transport_seglen(skb);
+}
 #endif	/* __KERNEL__ */
 #endif	/* _LINUX_SKBUFF_H */
net/ipv4/ip_forward.c +69 -2

89 lines not on this page · raw message

net/ipv6/ip6_output.c +15 -2
index ef02b26..070a2fa 100644
@@ -342,6 +342,20 @@ static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
 	return mtu;
 }
 
+static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
+{
+	if (skb->len <= mtu || skb->local_df)
+		return false;
+
+	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
+		return true;
+
+	if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
+		return false;
+
+	return true;
+}
+
 int ip6_forward(struct sk_buff *skb)
 {
 	struct dst_entry *dst = skb_dst(skb);
@@ -466,8 +480,7 @@ int ip6_forward(struct sk_buff *skb)
 	if (mtu < IPV6_MIN_MTU)
 		mtu = IPV6_MIN_MTU;
 
-	if ((!skb->local_df && skb->len > mtu && !skb_is_gso(skb)) ||
-	    (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)) {
+	if (ip6_pkt_too_big(skb, mtu)) {
 		/* Again, force OUTPUT device used as source address */
 		skb->dev = dst->dev;
 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
Signature · 1 line
1.8.1.5
maintainer
#2 lore raw

Re: [PATCH v6 2/2] net: ip, ipv6: handle gso skbs in forwarding path

From: Florian Westphal <[email protected]>
Date: Thu, 13 Feb 2014 23:09:12 +0100

Quoted 2 lines

Marcelo Ricardo Leitner reported problems when the forwarding link path has a lower mtu than the incoming one if the inbound interface supports GRO.

...

Quoted 14 lines

However it turns out that skb_segment() assumes skb nr_frags is related to mss size so we would BUG there. I don't want to mess with it considering Herbert and Eric disagree on what the correct behavior should be.

Hannes Frederic Sowa notes that when we would shrink gso_size skb_segment would then also need to deal with the case where SKB_MAX_FRAGS would be exceeded.

This uses sofware segmentation in the forward path when we hit ipv4 non-DF packets and the outgoing link mtu is too small. Its not perfect, but given the lack of bug reports wrt. GRO fwd being broken this is a rare case anyway. Also its not like this could not be improved later once the dust settles.

Quoted 3 lines

Also applied and queued up for -stable, thanks.

#3 lore raw

Re: [PATCH v6 2/2] net: ip, ipv6: handle gso skbs in forwarding path

In reply to Florian Westphal

net/ipv4/ip_forward.c line 84

Quoted 10 lines · Florian Westphal
net/ipv4/ip_forward.c +10 -0
+/* called if GSO skb needs to be fragmented on forward */
+static int ip_forward_finish_gso(struct sk_buff *skb)
+{
+	struct dst_entry *dst = skb_dst(skb);
+	netdev_features_t features;
+	struct sk_buff *segs;
+	int ret = 0;
+
+	features = netif_skb_dev_features(skb, dst->dev);
+	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);

While reviewing the skb->encapsulation logic I noticed that some drivers (mlx4 and i40e, i40evf) already started to set skb->encapsulation = 1 in the receive path.

Maybe it would be helpful to reset skb->encapsulation to zero before segmentation happens so we don't miss to update the fragmentation ids and offsets on udp packets?

Quoted 12 lines
net/ipv4/ip_forward.c +12 -0
+	if (IS_ERR(segs)) {
+		kfree_skb(skb);
+		return -ENOMEM;
+	}
+
+	consume_skb(skb);
+
+	do {
+		struct sk_buff *nskb = segs->next;
+		int err;
+
+		segs->next = NULL;

net/ipv4/ip_forward.c line 106

Quoted 10 lines
net/ipv4/ip_forward.c +10 -0
+		err = dst_output(segs);
+
+		if (err && ret == 0)
+			ret = err;
+		segs = nskb;
+	} while (segs);
+
+	return ret;
+}
+

Bye,

Hannes