Skip to contentSkip to conversation

IPVS changes

PATCH Applied Julian Anastasov 2 participants 7 messages Mar 9, 2013 – Mar 18, 2013

Subsystems (5)
lore
Share & tools
lore

Julian Anastasov

author maintainer
#0 lore raw

cover IPVS changes

This patchset adds new backup_only flag and fixes
some warnings found by sparse. I hope the SCTP part is correct. In fact I see other problems with the SCTP state machine that need to be resolved.

The patchset does not overlap with the other
patchset for IPVS optimizations, so can be applied independently.

Julian Anastasov (5):
ipvs: add backup_only flag to avoid loops
ipvs: remove extra rcu lock
ipvs: fix sctp chunk length order
ipvs: fix hashing in ip_vs_svc_hashkey
ipvs: fix some sparse warnings

 Documentation/networking/ipvs-sysctl.txt |    7 +++++++
 include/net/ip_vs.h                      |   14 +++++++++++++-
 net/netfilter/ipvs/ip_vs_core.c          |   22 +++++++++-------------
 net/netfilter/ipvs/ip_vs_ctl.c           |   15 ++++++++++++---
 net/netfilter/ipvs/ip_vs_est.c           |    2 +-
 net/netfilter/ipvs/ip_vs_proto_sctp.c    |   16 +++++++++-------
 6 files changed, 51 insertions(+), 25 deletions(-)
Signature · 1 line
1.7.3.4
author maintainer
#1 lore raw

PATCH 1/5 ipvs: add backup_only flag to avoid loops

Dmitry Akindinov is reporting for a problem where
SYNs are looping between the master and backup server when the backup server is used as real server in DR mode and has IPVS rules to function as director.

Even when the backup function is enabled we
continue to forward traffic and schedule new connections when the current master is using the backup server as real server. While this is not a problem for NAT, for DR and TUN method the backup server can not determine if a request comes from client or from director.

To avoid such loops add new sysctl flag
backup_only. It can be needed for DR/TUN setups that do not need backup and director function at the same time. When the backup function is enabled we stop any forwarding and pass the traffic to the local stack (real server mode). The flag disables the director function when the backup function is enabled.

For setups that enable backup function for
some virtual services and director function for other virtual services there should be another more complex solution to support DR/TUN mode, may be to assign per-virtual service syncid value, so that we can differentiate the requests.

---
 Documentation/networking/ipvs-sysctl.txt |    7 +++++++
 include/net/ip_vs.h                      |   12 ++++++++++++
 net/netfilter/ipvs/ip_vs_core.c          |   12 ++++++++----
 net/netfilter/ipvs/ip_vs_ctl.c           |    7 +++++++
 4 files changed, 34 insertions(+), 4 deletions(-)
Documentation/networking/ipvs-sysctl.txt +7 -0
index f2a2488..9573d0c 100644
@@ -15,6 +15,13 @@ amemthresh - INTEGER
         enabled and the variable is automatically set to 2, otherwise
         the strategy is disabled and the variable is  set  to 1.
 
+backup_only - BOOLEAN
+	0 - disabled (default)
+	not 0 - enabled
+
+	If set, disable the director function while the server is
+	in backup mode to avoid packet loops for DR/TUN methods.
+
 conntrack - BOOLEAN
 	0 - disabled (default)
 	not 0 - enabled
include/net/ip_vs.h +12 -0
index d1fb021..5ed8939 100644
@@ -997,6 +997,7 @@ struct netns_ipvs {
 	int			sysctl_sync_retries;
 	int			sysctl_nat_icmp_send;
 	int			sysctl_pmtu_disc;
+	int			sysctl_backup_only;
 
 	/* ip_vs_lblc */
 	int			sysctl_lblc_expiration;
@@ -1088,6 +1089,12 @@ static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs)
 	return ipvs->sysctl_pmtu_disc;
 }
 
+static inline int sysctl_backup_only(struct netns_ipvs *ipvs)
+{
+	return ipvs->sync_state & IP_VS_STATE_BACKUP &&
+	       ipvs->sysctl_backup_only;
+}
+
 #else
 
 static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs)
@@ -1135,6 +1142,11 @@ static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs)
 	return 1;
 }
 
+static inline int sysctl_backup_only(struct netns_ipvs *ipvs)
+{
+	return 0;
+}
+
 #endif
 
 /*
net/netfilter/ipvs/ip_vs_core.c +8 -4
index 2ea2862..15a051c 100644
@@ -1579,7 +1579,8 @@ ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
 	}
 	/* ipvs enabled in this netns ? */
 	net = skb_net(skb);
-	if (!net_ipvs(net)->enable)
+	ipvs = net_ipvs(net);
+	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, &iph);
@@ -1656,7 +1657,6 @@ ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
 	}
 
 	IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
-	ipvs = net_ipvs(net);
 	/* Check the server status */
 	if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 		/* the destination server is not available */
@@ -1817,13 +1817,15 @@ ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
 {
 	int r;
 	struct net *net;
+	struct netns_ipvs *ipvs;
 
 	if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
 		return NF_ACCEPT;
 
 	/* ipvs enabled in this netns ? */
 	net = skb_net(skb);
-	if (!net_ipvs(net)->enable)
+	ipvs = net_ipvs(net);
+	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
 		return NF_ACCEPT;
 
 	return ip_vs_in_icmp(skb, &r, hooknum);
@@ -1837,6 +1839,7 @@ ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
 {
 	int r;
 	struct net *net;
+	struct netns_ipvs *ipvs;
 	struct ip_vs_iphdr iphdr;
 
 	ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
@@ -1845,7 +1848,8 @@ ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
 
 	/* ipvs enabled in this netns ? */
 	net = skb_net(skb);
-	if (!net_ipvs(net)->enable)
+	ipvs = net_ipvs(net);
+	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
 		return NF_ACCEPT;
 
 	return ip_vs_in_icmp_v6(skb, &r, hooknum, &iphdr);
net/netfilter/ipvs/ip_vs_ctl.c +7 -0
index f4b53c4..11326db 100644
@@ -1806,6 +1806,12 @@ static struct ctl_table vs_vars[] = {
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec,
 	},
+	{
+		.procname	= "backup_only",
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
 #ifdef CONFIG_IP_VS_DEBUG
 	{
 		.procname	= "debug_level",
@@ -3739,6 +3745,7 @@ static int __net_init ip_vs_control_net_init_sysctl(struct net *net)
 	tbl[idx++].data = &ipvs->sysctl_nat_icmp_send;
 	ipvs->sysctl_pmtu_disc = 1;
 	tbl[idx++].data = &ipvs->sysctl_pmtu_disc;
+	tbl[idx++].data = &ipvs->sysctl_backup_only;
 
 
 	ipvs->sysctl_hdr = register_net_sysctl(net, "net/ipv4/vs", tbl);
Signature · 1 line
1.7.3.4
author maintainer
#2 lore raw

PATCH 2/5 ipvs: remove extra rcu lock

In reply to Julian Anastasov

In 3.7 we added code that uses ipv4_update_pmtu
but after commit c5ae7d4192 (ipv4: must use rcu protection while calling fib_lookup) the RCU lock is not needed.

---
 net/netfilter/ipvs/ip_vs_core.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)
net/netfilter/ipvs/ip_vs_core.c +0 -2
index 15a051c..207c3b9 100644
@@ -1393,10 +1393,8 @@ ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
 			skb_reset_network_header(skb);
 			IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
 				&ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
-			rcu_read_lock();
 			ipv4_update_pmtu(skb, dev_net(skb->dev),
 					 mtu, 0, 0, 0, 0);
-			rcu_read_unlock();
 			/* Client uses PMTUD? */
 			if (!(cih->frag_off & htons(IP_DF)))
 				goto ignore_ipip;
Signature · 1 line
1.7.3.4
author maintainer
#3 lore raw

PATCH 3/5 ipvs: fix sctp chunk length order

In reply to Julian Anastasov

Fix wrong but non-fatal access to chunk length.
sch->length should be in network order, next chunk should be aligned to 4 bytes. Problem noticed in sparse output.

---
 net/netfilter/ipvs/ip_vs_proto_sctp.c |   16 +++++++++-------
 1 files changed, 9 insertions(+), 7 deletions(-)
net/netfilter/ipvs/ip_vs_proto_sctp.c +9 -7
index 9302448..f7190cd 100644
@@ -906,7 +906,7 @@ set_sctp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp,
 	sctp_chunkhdr_t _sctpch, *sch;
 	unsigned char chunk_type;
 	int event, next_state;
-	int ihl;
+	int ihl, cofs;
 
 #ifdef CONFIG_IP_VS_IPV6
 	ihl = cp->af == AF_INET ? ip_hdrlen(skb) : sizeof(struct ipv6hdr);
@@ -914,8 +914,8 @@ set_sctp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp,
 	ihl = ip_hdrlen(skb);
 #endif
 
-	sch = skb_header_pointer(skb, ihl + sizeof(sctp_sctphdr_t),
-				sizeof(_sctpch), &_sctpch);
+	cofs = ihl + sizeof(sctp_sctphdr_t);
+	sch = skb_header_pointer(skb, cofs, sizeof(_sctpch), &_sctpch);
 	if (sch == NULL)
 		return;
 
@@ -933,10 +933,12 @@ set_sctp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp,
 	 */
 	if ((sch->type == SCTP_CID_COOKIE_ECHO) ||
 	    (sch->type == SCTP_CID_COOKIE_ACK)) {
-		sch = skb_header_pointer(skb, (ihl + sizeof(sctp_sctphdr_t) +
-				sch->length), sizeof(_sctpch), &_sctpch);
-		if (sch) {
-			if (sch->type == SCTP_CID_ABORT)
+		int clen = ntohs(sch->length);
+
+		if (clen >= sizeof(sctp_chunkhdr_t)) {
+			sch = skb_header_pointer(skb, cofs + ALIGN(clen, 4),
+						 sizeof(_sctpch), &_sctpch);
+			if (sch && sch->type == SCTP_CID_ABORT)
 				chunk_type = sch->type;
 		}
 	}
Signature · 1 line
1.7.3.4
author maintainer
#4 lore raw

PATCH 4/5 ipvs: fix hashing in ip_vs_svc_hashkey

In reply to Julian Anastasov

net is a pointer in host order, mix it properly
with other keys in network order. Fixes sparse warning.

---
 net/netfilter/ipvs/ip_vs_ctl.c |    8 +++++---
 1 files changed, 5 insertions(+), 3 deletions(-)
net/netfilter/ipvs/ip_vs_ctl.c +5 -3
index 11326db..2c8db45 100644
@@ -271,16 +271,18 @@ ip_vs_svc_hashkey(struct net *net, int af, unsigned int proto,
 {
 	register unsigned int porth = ntohs(port);
 	__be32 addr_fold = addr->ip;
+	__u32 ahash;
 
 #ifdef CONFIG_IP_VS_IPV6
 	if (af == AF_INET6)
 		addr_fold = addr->ip6[0]^addr->ip6[1]^
 			    addr->ip6[2]^addr->ip6[3];
 #endif
-	addr_fold ^= ((size_t)net>>8);
+	ahash = ntohl(addr_fold);
+	ahash ^= ((size_t) net >> 8);
 
-	return (proto^ntohl(addr_fold)^(porth>>IP_VS_SVC_TAB_BITS)^porth)
-		& IP_VS_SVC_TAB_MASK;
+	return (proto ^ ahash ^ (porth >> IP_VS_SVC_TAB_BITS) ^ porth) &
+	       IP_VS_SVC_TAB_MASK;
 }
 
 /*
Signature · 1 line
1.7.3.4
author maintainer
#5 lore raw

PATCH 5/5 ipvs: fix some sparse warnings

In reply to Julian Anastasov

Add missing __percpu annotations and make
ip_vs_net_id static.

---
 include/net/ip_vs.h             |    2 +-
 net/netfilter/ipvs/ip_vs_core.c |    8 +-------
 net/netfilter/ipvs/ip_vs_est.c  |    2 +-
 3 files changed, 3 insertions(+), 9 deletions(-)
include/net/ip_vs.h +1 -1
index 5ed8939..43886bb 100644
@@ -474,7 +474,7 @@ struct ip_vs_estimator {
 struct ip_vs_stats {
 	struct ip_vs_stats_user	ustats;		/* statistics */
 	struct ip_vs_estimator	est;		/* estimator */
-	struct ip_vs_cpu_stats	*cpustats;	/* per cpu counters */
+	struct ip_vs_cpu_stats __percpu	*cpustats;	/* per cpu counters */
 	spinlock_t		lock;		/* spin lock */
 	struct ip_vs_stats_user	ustats0;	/* reset values */
 };
net/netfilter/ipvs/ip_vs_core.c +1 -6
index 207c3b9..4fc749c 100644
@@ -69,10 +69,7 @@ EXPORT_SYMBOL(ip_vs_conn_put);
 EXPORT_SYMBOL(ip_vs_get_debug_level);
 #endif
 
-int ip_vs_net_id __read_mostly;
-#ifdef IP_VS_GENERIC_NETNS
-EXPORT_SYMBOL(ip_vs_net_id);
-#endif
+static int ip_vs_net_id __read_mostly;
 /* netns cnt used for uniqueness */
 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
 
@@ -1180,9 +1177,6 @@ ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
 						iph.len)))) {
 #ifdef CONFIG_IP_VS_IPV6
 				if (af == AF_INET6) {
-					struct net *net =
-						dev_net(skb_dst(skb)->dev);
maintainer Applied
#6 lore raw
In reply to Julian Anastasov
Quoted 8 lines · Julian Anastasov

This patchset adds new backup_only flag and fixes
some warnings found by sparse. I hope the SCTP part is correct. In fact I see other problems with the SCTP state machine that need to be resolved.

The patchset does not overlap with the other
patchset for IPVS optimizations, so can be applied independently.

Thanks, applied.