ipv4: Cache output routes in fib_info nexthops.
authorDavid S. Miller <davem@davemloft.net>
Tue, 17 Jul 2012 19:20:47 +0000 (12:20 -0700)
committerDavid S. Miller <davem@davemloft.net>
Fri, 20 Jul 2012 20:36:16 +0000 (13:36 -0700)
If we have an output route that lacks nexthop exceptions, we can cache
it in the FIB info nexthop.

Such routes will have DST_HOST cleared because such routes refer to a
family of destinations, rather than just one.

The sequence of the handling of exceptions during route lookup is
adjusted to make the logic work properly.

Before we allocate the route, we lookup the exception.

Then we know if we will cache this route or not, and therefore whether
DST_HOST should be set on the allocated route.

Then we use DST_HOST to key off whether we should store the resulting
route, during rt_set_nexthop(), in the FIB nexthop cache.

With help from Eric Dumazet.

Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/ip_fib.h
net/ipv4/fib_semantics.c
net/ipv4/route.c

index 2daf096dfc6021d9e5860a8e0e30464080c5179a..fb62c590360ed289a4534f02fe01c6962308168f 100644 (file)
@@ -46,6 +46,7 @@ struct fib_config {
  };
 
 struct fib_info;
+struct rtable;
 
 struct fib_nh_exception {
        struct fib_nh_exception __rcu   *fnhe_next;
@@ -80,6 +81,7 @@ struct fib_nh {
        __be32                  nh_gw;
        __be32                  nh_saddr;
        int                     nh_saddr_genid;
+       struct rtable           *nh_rth_output;
        struct fnhe_hash_bucket *nh_exceptions;
 };
 
index 2b57d768240d9499db471e7d889f422652b84245..83d0f42b619a439bacd59b6173a86e100524a457 100644 (file)
@@ -171,6 +171,8 @@ static void free_fib_info_rcu(struct rcu_head *head)
                        dev_put(nexthop_nh->nh_dev);
                if (nexthop_nh->nh_exceptions)
                        free_nh_exceptions(nexthop_nh);
+               if (nexthop_nh->nh_rth_output)
+                       dst_release(&nexthop_nh->nh_rth_output->dst);
        } endfor_nexthops(fi);
 
        release_net(fi->fib_net);
index d52f7699c2fa803dd4ce8d24ff5d64b48c834a08..8a0260010ea1c30a212af73d43780a3deebcd4f1 100644 (file)
@@ -1158,8 +1158,7 @@ static unsigned int ipv4_mtu(const struct dst_entry *dst)
        return mtu;
 }
 
-static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
-                           struct fib_info *fi)
+static void rt_init_metrics(struct rtable *rt, struct fib_info *fi)
 {
        if (fi->fib_metrics != (u32 *) dst_default_metrics) {
                rt->fi = fi;
@@ -1168,50 +1167,83 @@ static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
        dst_init_metrics(&rt->dst, fi->fib_metrics, true);
 }
 
-static void rt_bind_exception(struct rtable *rt, struct fib_nh *nh, __be32 daddr)
+static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
 {
        struct fnhe_hash_bucket *hash = nh->nh_exceptions;
        struct fib_nh_exception *fnhe;
        u32 hval;
 
+       if (!hash)
+               return NULL;
+
        hval = fnhe_hashfun(daddr);
 
-restart:
        for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
             fnhe = rcu_dereference(fnhe->fnhe_next)) {
-               __be32 fnhe_daddr, gw;
-               unsigned long expires;
-               unsigned int seq;
-               u32 pmtu;
-
-               seq = read_seqbegin(&fnhe_seqlock);
-               fnhe_daddr = fnhe->fnhe_daddr;
-               gw = fnhe->fnhe_gw;
-               pmtu = fnhe->fnhe_pmtu;
-               expires = fnhe->fnhe_expires;
-               if (read_seqretry(&fnhe_seqlock, seq))
-                       goto restart;
-               if (daddr != fnhe_daddr)
-                       continue;
-               if (pmtu) {
-                       unsigned long diff = expires - jiffies;
+               if (fnhe->fnhe_daddr == daddr)
+                       return fnhe;
+       }
+       return NULL;
+}
 
-                       if (time_before(jiffies, expires)) {
-                               rt->rt_pmtu = pmtu;
-                               dst_set_expires(&rt->dst, diff);
-                       }
-               }
-               if (gw) {
-                       rt->rt_flags |= RTCF_REDIRECTED;
-                       rt->rt_gateway = gw;
+static void rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
+                             __be32 daddr)
+{
+       __be32 fnhe_daddr, gw;
+       unsigned long expires;
+       unsigned int seq;
+       u32 pmtu;
+
+restart:
+       seq = read_seqbegin(&fnhe_seqlock);
+       fnhe_daddr = fnhe->fnhe_daddr;
+       gw = fnhe->fnhe_gw;
+       pmtu = fnhe->fnhe_pmtu;
+       expires = fnhe->fnhe_expires;
+       if (read_seqretry(&fnhe_seqlock, seq))
+               goto restart;
+
+       if (daddr != fnhe_daddr)
+               return;
+
+       if (pmtu) {
+               unsigned long diff = expires - jiffies;
+
+               if (time_before(jiffies, expires)) {
+                       rt->rt_pmtu = pmtu;
+                       dst_set_expires(&rt->dst, diff);
                }
-               fnhe->fnhe_stamp = jiffies;
-               break;
+       }
+       if (gw) {
+               rt->rt_flags |= RTCF_REDIRECTED;
+               rt->rt_gateway = gw;
+       }
+       fnhe->fnhe_stamp = jiffies;
+}
+
+static inline void rt_release_rcu(struct rcu_head *head)
+{
+       struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
+       dst_release(dst);
+}
+
+static void rt_cache_route(struct fib_nh *nh, struct rtable *rt)
+{
+       struct rtable *orig, *prev, **p = &nh->nh_rth_output;
+
+       orig = *p;
+
+       prev = cmpxchg(p, orig, rt);
+       if (prev == orig) {
+               dst_clone(&rt->dst);
+               if (orig)
+                       call_rcu_bh(&orig->dst.rcu_head, rt_release_rcu);
        }
 }
 
-static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
+static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
                           const struct fib_result *res,
+                          struct fib_nh_exception *fnhe,
                           struct fib_info *fi, u16 type, u32 itag)
 {
        if (fi) {
@@ -1219,12 +1251,15 @@ static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
 
                if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
                        rt->rt_gateway = nh->nh_gw;
-               if (unlikely(nh->nh_exceptions))
-                       rt_bind_exception(rt, nh, fl4->daddr);
-               rt_init_metrics(rt, fl4, fi);
+               if (unlikely(fnhe))
+                       rt_bind_exception(rt, fnhe, daddr);
+               rt_init_metrics(rt, fi);
 #ifdef CONFIG_IP_ROUTE_CLASSID
-               rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
+               rt->dst.tclassid = nh->nh_tclassid;
 #endif
+               if (!(rt->dst.flags & DST_HOST) &&
+                   rt_is_output_route(rt))
+                       rt_cache_route(nh, rt);
        }
 
 #ifdef CONFIG_IP_ROUTE_CLASSID
@@ -1236,10 +1271,10 @@ static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
 }
 
 static struct rtable *rt_dst_alloc(struct net_device *dev,
-                                  bool nopolicy, bool noxfrm)
+                                  bool nopolicy, bool noxfrm, bool will_cache)
 {
        return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
-                        DST_HOST | DST_NOCACHE |
+                        (will_cache ? 0 : DST_HOST) | DST_NOCACHE |
                         (nopolicy ? DST_NOPOLICY : 0) |
                         (noxfrm ? DST_NOXFRM : 0));
 }
@@ -1276,7 +1311,7 @@ static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
                        goto e_err;
        }
        rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
-                          IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
+                          IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
        if (!rth)
                goto e_nobufs;
 
@@ -1349,6 +1384,7 @@ static int __mkroute_input(struct sk_buff *skb,
                           __be32 daddr, __be32 saddr, u32 tos,
                           struct rtable **result)
 {
+       struct fib_nh_exception *fnhe;
        struct rtable *rth;
        int err;
        struct in_device *out_dev;
@@ -1395,9 +1431,13 @@ static int __mkroute_input(struct sk_buff *skb,
                }
        }
 
+       fnhe = NULL;
+       if (res->fi)
+               fnhe = find_exception(&FIB_RES_NH(*res), daddr);
+
        rth = rt_dst_alloc(out_dev->dev,
                           IN_DEV_CONF_GET(in_dev, NOPOLICY),
-                          IN_DEV_CONF_GET(out_dev, NOXFRM));
+                          IN_DEV_CONF_GET(out_dev, NOXFRM), false);
        if (!rth) {
                err = -ENOBUFS;
                goto cleanup;
@@ -1416,7 +1456,7 @@ static int __mkroute_input(struct sk_buff *skb,
        rth->dst.input = ip_forward;
        rth->dst.output = ip_output;
 
-       rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
+       rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
 
        *result = rth;
        err = 0;
@@ -1558,7 +1598,7 @@ brd_input:
 
 local_input:
        rth = rt_dst_alloc(net->loopback_dev,
-                          IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
+                          IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
        if (!rth)
                goto e_nobufs;
 
@@ -1672,6 +1712,7 @@ static struct rtable *__mkroute_output(const struct fib_result *res,
                                       unsigned int flags)
 {
        struct fib_info *fi = res->fi;
+       struct fib_nh_exception *fnhe;
        struct in_device *in_dev;
        u16 type = res->type;
        struct rtable *rth;
@@ -1710,9 +1751,22 @@ static struct rtable *__mkroute_output(const struct fib_result *res,
                        fi = NULL;
        }
 
+       fnhe = NULL;
+       if (fi) {
+               fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr);
+               if (!fnhe) {
+                       rth = FIB_RES_NH(*res).nh_rth_output;
+                       if (rth &&
+                           rth->dst.obsolete == DST_OBSOLETE_FORCE_CHK) {
+                               dst_use(&rth->dst, jiffies);
+                               return rth;
+                       }
+               }
+       }
        rth = rt_dst_alloc(dev_out,
                           IN_DEV_CONF_GET(in_dev, NOPOLICY),
-                          IN_DEV_CONF_GET(in_dev, NOXFRM));
+                          IN_DEV_CONF_GET(in_dev, NOXFRM),
+                          fi && !fnhe);
        if (!rth)
                return ERR_PTR(-ENOBUFS);
 
@@ -1749,7 +1803,7 @@ static struct rtable *__mkroute_output(const struct fib_result *res,
 #endif
        }
 
-       rt_set_nexthop(rth, fl4, res, fi, type, 0);
+       rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
 
        if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
                rth->dst.flags |= DST_NOCACHE;