]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/commitdiff
ipv6: route: dissect flow in input path if fib rules need it
authorRoopa Prabhu <roopa@cumulusnetworks.com>
Thu, 1 Mar 2018 03:42:41 +0000 (22:42 -0500)
committerDavid S. Miller <davem@davemloft.net>
Thu, 1 Mar 2018 03:44:44 +0000 (22:44 -0500)
Dissect flow in fwd path if fib rules require it. Controlled by
a flag to avoid penatly for the common case. Flag is set when fib
rules with sport, dport and proto match that require flow dissect
are installed. Also passes the dissected hash keys to the multipath
hash function when applicable to avoid dissecting the flow again.
icmp packets will continue to use inner header for hash
calculations (Thanks to Nikolay Aleksandrov for some review here).

Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/ip_fib.h
include/net/netns/ipv4.h
net/ipv4/fib_rules.c
net/ipv4/fib_semantics.c
net/ipv4/route.c

index 15e19c5c6f26fda881a38f45ecdaa18e7f5cfd76..8812582a94d5060166bf0acad942f363439ffe78 100644 (file)
@@ -293,6 +293,13 @@ static inline unsigned int fib4_rules_seq_read(struct net *net)
        return 0;
 }
 
+static inline bool fib4_rules_early_flow_dissect(struct net *net,
+                                                struct sk_buff *skb,
+                                                struct flowi4 *fl4,
+                                                struct flow_keys *flkeys)
+{
+       return false;
+}
 #else /* CONFIG_IP_MULTIPLE_TABLES */
 int __net_init fib4_rules_init(struct net *net);
 void __net_exit fib4_rules_exit(struct net *net);
@@ -341,6 +348,24 @@ bool fib4_rule_default(const struct fib_rule *rule);
 int fib4_rules_dump(struct net *net, struct notifier_block *nb);
 unsigned int fib4_rules_seq_read(struct net *net);
 
+static inline bool fib4_rules_early_flow_dissect(struct net *net,
+                                                struct sk_buff *skb,
+                                                struct flowi4 *fl4,
+                                                struct flow_keys *flkeys)
+{
+       unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
+
+       if (!net->ipv4.fib_rules_require_fldissect)
+               return false;
+
+       skb_flow_dissect_flow_keys(skb, flkeys, flag);
+       fl4->fl4_sport = flkeys->ports.src;
+       fl4->fl4_dport = flkeys->ports.dst;
+       fl4->flowi4_proto = flkeys->basic.ip_proto;
+
+       return true;
+}
+
 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 
 /* Exported by fib_frontend.c */
@@ -371,7 +396,7 @@ int fib_sync_up(struct net_device *dev, unsigned int nh_flags);
 
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
-                      const struct sk_buff *skb);
+                      const struct sk_buff *skb, struct flow_keys *flkeys);
 #endif
 void fib_select_multipath(struct fib_result *res, int hash);
 void fib_select_path(struct net *net, struct fib_result *res,
index 44668c29701a72a5116781d4b76267c0068f1143..3a970e429ab6c324c38610b435dec77c523b72ac 100644 (file)
@@ -52,6 +52,7 @@ struct netns_ipv4 {
 #ifdef CONFIG_IP_MULTIPLE_TABLES
        struct fib_rules_ops    *rules_ops;
        bool                    fib_has_custom_rules;
+       unsigned int            fib_rules_require_fldissect;
        struct fib_table __rcu  *fib_main;
        struct fib_table __rcu  *fib_default;
 #endif
index 16083b82954bff13c6c74a94df9f5b3ee68e74b4..737d11bc8838bc175df587db4acd3fce3146c60d 100644 (file)
@@ -255,6 +255,9 @@ static int fib4_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
        }
 #endif
 
+       if (fib_rule_requires_fldissect(rule))
+               net->ipv4.fib_rules_require_fldissect++;
+
        rule4->src_len = frh->src_len;
        rule4->srcmask = inet_make_mask(rule4->src_len);
        rule4->dst_len = frh->dst_len;
@@ -283,6 +286,10 @@ static int fib4_rule_delete(struct fib_rule *rule)
                net->ipv4.fib_num_tclassid_users--;
 #endif
        net->ipv4.fib_has_custom_rules = true;
+
+       if (net->ipv4.fib_rules_require_fldissect &&
+           fib_rule_requires_fldissect(rule))
+               net->ipv4.fib_rules_require_fldissect--;
 errout:
        return err;
 }
@@ -400,6 +407,7 @@ int __net_init fib4_rules_init(struct net *net)
                goto fail;
        net->ipv4.rules_ops = ops;
        net->ipv4.fib_has_custom_rules = false;
+       net->ipv4.fib_rules_require_fldissect = 0;
        return 0;
 
 fail:
index f31e6575ab91951e8ea2f4c9e4c73912ba544547..181b0d8d589c789d95c05c69487a5bbf10afa509 100644 (file)
@@ -1770,7 +1770,7 @@ void fib_select_path(struct net *net, struct fib_result *res,
 
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
        if (res->fi->fib_nhs > 1) {
-               int h = fib_multipath_hash(res->fi, fl4, skb);
+               int h = fib_multipath_hash(res->fi, fl4, skb, NULL);
 
                fib_select_multipath(res, h);
        }
index 26eefa2eaa448a00e03918f6d25f365110071bdf..3bb686dac27307fdd9567eb24fb91c2663625366 100644 (file)
@@ -1783,7 +1783,7 @@ static void ip_multipath_l3_keys(const struct sk_buff *skb,
 
 /* if skb is set it will be used and fl4 can be NULL */
 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
-                      const struct sk_buff *skb)
+                      const struct sk_buff *skb, struct flow_keys *flkeys)
 {
        struct net *net = fi->fib_net;
        struct flow_keys hash_keys;
@@ -1810,14 +1810,23 @@ int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
                        if (skb->l4_hash)
                                return skb_get_hash_raw(skb) >> 1;
                        memset(&hash_keys, 0, sizeof(hash_keys));
-                       skb_flow_dissect_flow_keys(skb, &keys, flag);
 
-                       hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
-                       hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
-                       hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
-                       hash_keys.ports.src = keys.ports.src;
-                       hash_keys.ports.dst = keys.ports.dst;
-                       hash_keys.basic.ip_proto = keys.basic.ip_proto;
+                       if (flkeys) {
+                               hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
+                               hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
+                               hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
+                               hash_keys.ports.src = flkeys->ports.src;
+                               hash_keys.ports.dst = flkeys->ports.dst;
+                               hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
+                       } else {
+                               skb_flow_dissect_flow_keys(skb, &keys, flag);
+                               hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
+                               hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
+                               hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
+                               hash_keys.ports.src = keys.ports.src;
+                               hash_keys.ports.dst = keys.ports.dst;
+                               hash_keys.basic.ip_proto = keys.basic.ip_proto;
+                       }
                } else {
                        memset(&hash_keys, 0, sizeof(hash_keys));
                        hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
@@ -1838,11 +1847,12 @@ int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
 static int ip_mkroute_input(struct sk_buff *skb,
                            struct fib_result *res,
                            struct in_device *in_dev,
-                           __be32 daddr, __be32 saddr, u32 tos)
+                           __be32 daddr, __be32 saddr, u32 tos,
+                           struct flow_keys *hkeys)
 {
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
        if (res->fi && res->fi->fib_nhs > 1) {
-               int h = fib_multipath_hash(res->fi, NULL, skb);
+               int h = fib_multipath_hash(res->fi, NULL, skb, hkeys);
 
                fib_select_multipath(res, h);
        }
@@ -1868,13 +1878,14 @@ static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
                               struct fib_result *res)
 {
        struct in_device *in_dev = __in_dev_get_rcu(dev);
+       struct flow_keys *flkeys = NULL, _flkeys;
+       struct net    *net = dev_net(dev);
        struct ip_tunnel_info *tun_info;
-       struct flowi4   fl4;
+       int             err = -EINVAL;
        unsigned int    flags = 0;
        u32             itag = 0;
        struct rtable   *rth;
-       int             err = -EINVAL;
-       struct net    *net = dev_net(dev);
+       struct flowi4   fl4;
        bool do_cache;
 
        /* IP on this device is disabled. */
@@ -1933,6 +1944,10 @@ static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
        fl4.daddr = daddr;
        fl4.saddr = saddr;
        fl4.flowi4_uid = sock_net_uid(net, NULL);
+
+       if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys))
+               flkeys = &_flkeys;
+
        err = fib_lookup(net, &fl4, res, 0);
        if (err != 0) {
                if (!IN_DEV_FORWARD(in_dev))
@@ -1958,7 +1973,7 @@ static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
        if (res->type != RTN_UNICAST)
                goto martian_destination;
 
-       err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos);
+       err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
 out:   return err;
 
 brd_input: