[NET]: Detect hardware rx checksum faults correctly

Here is the patch that introduces the generic skb_checksum_complete which also checks for hardware RX checksum faults. If that happens, it'll call netdev_rx_csum_fault which currently prints out a stack trace with the device name. In future it can turn off RX checksum. I've converted every spot under net/ that does RX checksum checks to use skb_checksum_complete or __skb_checksum_complete with the exceptions of: * Those places where checksums are done bit by bit. These will call netdev_rx_csum_fault directly. * The following have not been completely checked/converted: ipmr ip_vs netfilter dccp This patch is based on patches and suggestions from Stephen Hemminger and David S. Miller. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
starfive-tech · Nov 10, 2005 · fb286bb · fb286bb
1 parent 1064e94
commit fb286bb
Show file tree

Hide file tree

Showing 19 changed files with 173 additions and 134 deletions.
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
@@ -927,6 +927,13 @@ extern int		netdev_max_backlog;
 extern int		weight_p;
 extern int		netdev_set_master(struct net_device *dev, struct net_device *master);
 extern int skb_checksum_help(struct sk_buff *skb, int inward);
+#ifdef CONFIG_BUG
+extern void netdev_rx_csum_fault(struct net_device *dev);
+#else
+static inline void netdev_rx_csum_fault(struct net_device *dev)
+{
+}
+#endif
 /* rx skb timestamps */
 extern void		net_enable_timestamp(void);
 extern void		net_disable_timestamp(void);

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
@@ -1236,8 +1236,7 @@ extern unsigned int    datagram_poll(struct file *file, struct socket *sock,
 extern int	       skb_copy_datagram_iovec(const struct sk_buff *from,
 					       int offset, struct iovec *to,
 					       int size);
-extern int	       skb_copy_and_csum_datagram_iovec(const
-							struct sk_buff *skb,
+extern int	       skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
 							int hlen,
 							struct iovec *iov);
 extern void	       skb_free_datagram(struct sock *sk, struct sk_buff *skb);
@@ -1305,6 +1304,30 @@ static inline void skb_set_timestamp(struct sk_buff *skb, const struct timeval *
 
 extern void __net_timestamp(struct sk_buff *skb);
 
+extern unsigned int __skb_checksum_complete(struct sk_buff *skb);
+
+/**
+ *	skb_checksum_complete - Calculate checksum of an entire packet
+ *	@skb: packet to process
+ *
+ *	This function calculates the checksum over the entire packet plus
+ *	the value of skb->csum.  The latter can be used to supply the
+ *	checksum of a pseudo header as used by TCP/UDP.  It returns the
+ *	checksum.
+ *
+ *	For protocols that contain complete checksums such as ICMP/TCP/UDP,
+ *	this function can be used to verify that checksum on received
+ *	packets.  In that case the function should return zero if the
+ *	checksum is correct.  In particular, this function will return zero
+ *	if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
+ *	hardware has already verified the correctness of the checksum.
+ */
+static inline unsigned int skb_checksum_complete(struct sk_buff *skb)
+{
+	return skb->ip_summed != CHECKSUM_UNNECESSARY &&
+		__skb_checksum_complete(skb);
+}
+
 #ifdef CONFIG_NETFILTER
 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
 {

diff --git a/include/net/tcp.h b/include/net/tcp.h
@@ -27,6 +27,7 @@
 #include <linux/slab.h>
 #include <linux/cache.h>
 #include <linux/percpu.h>
+#include <linux/skbuff.h>
 
 #include <net/inet_connection_sock.h>
 #include <net/inet_timewait_sock.h>
@@ -852,7 +853,7 @@ static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
 
 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
 {
-	return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
+	return __skb_checksum_complete(skb);
 }
 
 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)

diff --git a/net/core/datagram.c b/net/core/datagram.c
@@ -350,6 +350,20 @@ static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
 	return -EFAULT;
 }
 
+unsigned int __skb_checksum_complete(struct sk_buff *skb)
+{
+	unsigned int sum;
+
+	sum = (u16)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
+	if (likely(!sum)) {
+		if (unlikely(skb->ip_summed == CHECKSUM_HW))
+			netdev_rx_csum_fault(skb->dev);
+		skb->ip_summed = CHECKSUM_UNNECESSARY;
+	}
+	return sum;
+}
+EXPORT_SYMBOL(__skb_checksum_complete);
+
 /**
  *	skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
  *	@skb: skbuff
@@ -363,7 +377,7 @@ static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
  *		 -EFAULT - fault during copy. Beware, in this case iovec
  *			   can be modified!
  */
-int skb_copy_and_csum_datagram_iovec(const struct sk_buff *skb,
+int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
 				     int hlen, struct iovec *iov)
 {
 	unsigned int csum;
@@ -376,8 +390,7 @@ int skb_copy_and_csum_datagram_iovec(const struct sk_buff *skb,
 		iov++;
 
 	if (iov->iov_len < chunk) {
-		if ((unsigned short)csum_fold(skb_checksum(skb, 0, chunk + hlen,
-							   skb->csum)))
+		if (__skb_checksum_complete(skb))
 			goto csum_error;
 		if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
 			goto fault;
@@ -388,6 +401,8 @@ int skb_copy_and_csum_datagram_iovec(const struct sk_buff *skb,
 			goto fault;
 		if ((unsigned short)csum_fold(csum))
 			goto csum_error;
+		if (unlikely(skb->ip_summed == CHECKSUM_HW))
+			netdev_rx_csum_fault(skb->dev);
 		iov->iov_len -= chunk;
 		iov->iov_base += chunk;
 	}

diff --git a/net/core/dev.c b/net/core/dev.c
@@ -1108,6 +1108,18 @@ int skb_checksum_help(struct sk_buff *skb, int inward)
 	return ret;
 }
 
+/* Take action when hardware reception checksum errors are detected. */
+#ifdef CONFIG_BUG
+void netdev_rx_csum_fault(struct net_device *dev)
+{
+	if (net_ratelimit()) {
+		printk(KERN_ERR "%s: hw csum failure.\n", dev->name);
+		dump_stack();
+	}
+}
+EXPORT_SYMBOL(netdev_rx_csum_fault);
+#endif
+
 #ifdef CONFIG_HIGHMEM
 /* Actually, we should eliminate this check as soon as we know, that:
  * 1. IOMMU is present and allows to map all the memory.

diff --git a/net/core/netpoll.c b/net/core/netpoll.c
@@ -101,16 +101,20 @@ void netpoll_queue(struct sk_buff *skb)
 static int checksum_udp(struct sk_buff *skb, struct udphdr *uh,
 			     unsigned short ulen, u32 saddr, u32 daddr)
 {
-	if (uh->check == 0)
+	unsigned int psum;
+
+	if (uh->check == 0 || skb->ip_summed == CHECKSUM_UNNECESSARY)
 		return 0;
 
-	if (skb->ip_summed == CHECKSUM_HW)
-		return csum_tcpudp_magic(
-			saddr, daddr, ulen, IPPROTO_UDP, skb->csum);
+	psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
+
+	if (skb->ip_summed == CHECKSUM_HW &&
+	    !(u16)csum_fold(csum_add(psum, skb->csum)))
+		return 0;
 
-	skb->csum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
+	skb->csum = psum;
 
-	return csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
+	return __skb_checksum_complete(skb);
 }
 
 /*
@@ -489,7 +493,7 @@ int __netpoll_rx(struct sk_buff *skb)
 
 	if (ulen != len)
 		goto out;
-	if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr) < 0)
+	if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 		goto out;
 	if (np->local_ip && np->local_ip != ntohl(iph->daddr))
 		goto out;

diff --git a/net/ipv4/icmp.c b/net/ipv4/icmp.c
@@ -934,11 +934,11 @@ int icmp_rcv(struct sk_buff *skb)
 	case CHECKSUM_HW:
 		if (!(u16)csum_fold(skb->csum))
 			break;
-		LIMIT_NETDEBUG(KERN_DEBUG "icmp v4 hw csum failure\n");
+		/* fall through */
 	case CHECKSUM_NONE:
-		if ((u16)csum_fold(skb_checksum(skb, 0, skb->len, 0)))
+		skb->csum = 0;
+		if (__skb_checksum_complete(skb))
 			goto error;
-	default:;
 	}
 
 	if (!pskb_pull(skb, sizeof(struct icmphdr)))

diff --git a/net/ipv4/igmp.c b/net/ipv4/igmp.c
@@ -872,11 +872,18 @@ int igmp_rcv(struct sk_buff *skb)
 		return 0;
 	}
 
-	if (!pskb_may_pull(skb, sizeof(struct igmphdr)) || 
-	    (u16)csum_fold(skb_checksum(skb, 0, len, 0))) {
-		in_dev_put(in_dev);
-		kfree_skb(skb);
-		return 0;
+	if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
+		goto drop;
+
+	switch (skb->ip_summed) {
+	case CHECKSUM_HW:
+		if (!(u16)csum_fold(skb->csum))
+			break;
+		/* fall through */
+	case CHECKSUM_NONE:
+		skb->csum = 0;
+		if (__skb_checksum_complete(skb))
+			goto drop;
 	}
 
 	ih = skb->h.igmph;
@@ -906,6 +913,8 @@ int igmp_rcv(struct sk_buff *skb)
 	default:
 		NETDEBUG(KERN_DEBUG "New IGMP type=%d, why we do not know about it?\n", ih->type);
 	}
+
+drop:
 	in_dev_put(in_dev);
 	kfree_skb(skb);
 	return 0;

diff --git a/net/ipv4/ip_gre.c b/net/ipv4/ip_gre.c
@@ -577,15 +577,16 @@ static int ipgre_rcv(struct sk_buff *skb)
 			goto drop_nolock;
 
 		if (flags&GRE_CSUM) {
-			if (skb->ip_summed == CHECKSUM_HW) {
+			switch (skb->ip_summed) {
+			case CHECKSUM_HW:
 				csum = (u16)csum_fold(skb->csum);
-				if (csum)
-					skb->ip_summed = CHECKSUM_NONE;
-			}
-			if (skb->ip_summed == CHECKSUM_NONE) {
-				skb->csum = skb_checksum(skb, 0, skb->len, 0);
+				if (!csum)
+					break;
+				/* fall through */
+			case CHECKSUM_NONE:
+				skb->csum = 0;
+				csum = __skb_checksum_complete(skb);
 				skb->ip_summed = CHECKSUM_HW;
-				csum = (u16)csum_fold(skb->csum);
 			}
 			offset += 4;
 		}

diff --git a/net/ipv4/netfilter/ip_conntrack_proto_icmp.c b/net/ipv4/netfilter/ip_conntrack_proto_icmp.c
@@ -13,6 +13,7 @@
 #include <linux/in.h>
 #include <linux/icmp.h>
 #include <linux/seq_file.h>
+#include <linux/skbuff.h>
 #include <net/ip.h>
 #include <net/checksum.h>
 #include <linux/netfilter.h>
@@ -230,19 +231,15 @@ icmp_error(struct sk_buff *skb, enum ip_conntrack_info *ctinfo,
 	case CHECKSUM_HW:
 		if (!(u16)csum_fold(skb->csum)) 
 			break;
-		if (LOG_INVALID(IPPROTO_ICMP))
-			nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
-				      "ip_ct_icmp: bad HW ICMP checksum ");
-		return -NF_ACCEPT;
+		/* fall through */
 	case CHECKSUM_NONE:
-		if ((u16)csum_fold(skb_checksum(skb, 0, skb->len, 0))) {
+		skb->csum = 0;
+		if (__skb_checksum_complete(skb)) {
 			if (LOG_INVALID(IPPROTO_ICMP))
 				nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 					      "ip_ct_icmp: bad ICMP checksum ");
 			return -NF_ACCEPT;
 		}
-	default:
-		break;
 	}
 
 checksum_skipped:

diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
@@ -1110,24 +1110,18 @@ static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 static int tcp_v4_checksum_init(struct sk_buff *skb)
 {
 	if (skb->ip_summed == CHECKSUM_HW) {
-		skb->ip_summed = CHECKSUM_UNNECESSARY;
 		if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr,
-				  skb->nh.iph->daddr, skb->csum))
+				  skb->nh.iph->daddr, skb->csum)) {
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
 			return 0;
-
-		LIMIT_NETDEBUG(KERN_DEBUG "hw tcp v4 csum failed\n");
-		skb->ip_summed = CHECKSUM_NONE;
+		}
 	}
+
+	skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr, skb->nh.iph->daddr,
+				       skb->len, IPPROTO_TCP, 0);
+
 	if (skb->len <= 76) {
-		if (tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr,
-				 skb->nh.iph->daddr,
-				 skb_checksum(skb, 0, skb->len, 0)))
-			return -1;
-		skb->ip_summed = CHECKSUM_UNNECESSARY;
-	} else {
-		skb->csum = ~tcp_v4_check(skb->h.th, skb->len,
-					  skb->nh.iph->saddr,
-					  skb->nh.iph->daddr, 0);
+		return __skb_checksum_complete(skb);
 	}
 	return 0;
 }
@@ -1219,7 +1213,7 @@ int tcp_v4_rcv(struct sk_buff *skb)
 	 * provided case of th->doff==0 is elimineted.
 	 * So, we defer the checks. */
 	if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
-	     tcp_v4_checksum_init(skb) < 0))
+	     tcp_v4_checksum_init(skb)))
 		goto bad_packet;
 
 	th = skb->h.th;

diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
@@ -761,7 +761,7 @@ int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
 
 static __inline__ int __udp_checksum_complete(struct sk_buff *skb)
 {
-	return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
+	return __skb_checksum_complete(skb);
 }
 
 static __inline__ int udp_checksum_complete(struct sk_buff *skb)
@@ -1100,11 +1100,8 @@ static int udp_checksum_init(struct sk_buff *skb, struct udphdr *uh,
 	if (uh->check == 0) {
 		skb->ip_summed = CHECKSUM_UNNECESSARY;
 	} else if (skb->ip_summed == CHECKSUM_HW) {
-		skb->ip_summed = CHECKSUM_UNNECESSARY;
 		if (!udp_check(uh, ulen, saddr, daddr, skb->csum))
-			return 0;
-		LIMIT_NETDEBUG(KERN_DEBUG "udp v4 hw csum failure.\n");
-		skb->ip_summed = CHECKSUM_NONE;
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
 	}
 	if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 		skb->csum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);

diff --git a/net/ipv6/icmp.c b/net/ipv6/icmp.c
@@ -585,17 +585,16 @@ static int icmpv6_rcv(struct sk_buff **pskb, unsigned int *nhoffp)
 	daddr = &skb->nh.ipv6h->daddr;
 
 	/* Perform checksum. */
-	if (skb->ip_summed == CHECKSUM_HW) {
-		skb->ip_summed = CHECKSUM_UNNECESSARY;
-		if (csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_ICMPV6,
-				    skb->csum)) {
-			LIMIT_NETDEBUG(KERN_DEBUG "ICMPv6 hw checksum failed\n");
-			skb->ip_summed = CHECKSUM_NONE;
-		}
-	}
-	if (skb->ip_summed == CHECKSUM_NONE) {
-		if (csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_ICMPV6,
-				    skb_checksum(skb, 0, skb->len, 0))) {
+	switch (skb->ip_summed) {
+	case CHECKSUM_HW:
+		if (!csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_ICMPV6,
+				     skb->csum))
+			break;
+		/* fall through */
+	case CHECKSUM_NONE:
+		skb->csum = ~csum_ipv6_magic(saddr, daddr, skb->len,
+					     IPPROTO_ICMPV6, 0);
+		if (__skb_checksum_complete(skb)) {
 			LIMIT_NETDEBUG(KERN_DEBUG "ICMPv6 checksum failed [%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x > %04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x]\n",
 				       NIP6(*saddr), NIP6(*daddr));
 			goto discard_it;