[IPV6]: Fix routing by fwmark
[powerpc.git] / net / ipv4 / route.c
index cc9423d..a4d4cb8 100644 (file)
@@ -64,7 +64,6 @@
  *             2 of the License, or (at your option) any later version.
  */
 
-#include <linux/config.h>
 #include <linux/module.h>
 #include <asm/uaccess.h>
 #include <asm/system.h>
 #include <net/icmp.h>
 #include <net/xfrm.h>
 #include <net/ip_mp_alg.h>
+#include <net/netevent.h>
 #ifdef CONFIG_SYSCTL
 #include <linux/sysctl.h>
 #endif
@@ -206,21 +206,27 @@ __u8 ip_tos2prio[16] = {
 struct rt_hash_bucket {
        struct rtable   *chain;
 };
-#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
+#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
+       defined(CONFIG_PROVE_LOCKING)
 /*
  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
  * The size of this table is a power of two and depends on the number of CPUS.
+ * (on lockdep we have a quite big spinlock_t, so keep the size down there)
  */
-#if NR_CPUS >= 32
-#define RT_HASH_LOCK_SZ        4096
-#elif NR_CPUS >= 16
-#define RT_HASH_LOCK_SZ        2048
-#elif NR_CPUS >= 8
-#define RT_HASH_LOCK_SZ        1024
-#elif NR_CPUS >= 4
-#define RT_HASH_LOCK_SZ        512
+#ifdef CONFIG_LOCKDEP
+# define RT_HASH_LOCK_SZ       256
 #else
-#define RT_HASH_LOCK_SZ        256
+# if NR_CPUS >= 32
+#  define RT_HASH_LOCK_SZ      4096
+# elif NR_CPUS >= 16
+#  define RT_HASH_LOCK_SZ      2048
+# elif NR_CPUS >= 8
+#  define RT_HASH_LOCK_SZ      1024
+# elif NR_CPUS >= 4
+#  define RT_HASH_LOCK_SZ      512
+# else
+#  define RT_HASH_LOCK_SZ      256
+# endif
 #endif
 
 static spinlock_t      *rt_hash_locks;
@@ -244,7 +250,7 @@ static unsigned int         rt_hash_rnd;
 
 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 #define RT_CACHE_STAT_INC(field) \
-       (per_cpu(rt_cache_stat, raw_smp_processor_id()).field++)
+       (__raw_get_cpu_var(rt_cache_stat).field++)
 
 static int rt_intern_hash(unsigned hash, struct rtable *rth,
                                struct rtable **res);
@@ -1120,6 +1126,7 @@ void ip_rt_redirect(u32 old_gw, u32 daddr, u32 new_gw,
        struct rtable *rth, **rthp;
        u32  skeys[2] = { saddr, 0 };
        int  ikeys[2] = { dev->ifindex, 0 };
+       struct netevent_redirect netevent;
 
        if (!in_dev)
                return;
@@ -1211,6 +1218,11 @@ void ip_rt_redirect(u32 old_gw, u32 daddr, u32 new_gw,
                                        rt_drop(rt);
                                        goto do_next;
                                }
+                               
+                               netevent.old = &rth->u.dst;
+                               netevent.new = &rt->u.dst;
+                               call_netevent_notifiers(NETEVENT_REDIRECT, 
+                                                       &netevent);
 
                                rt_del(hash, rth);
                                if (!rt_intern_hash(hash, rt, &rt))
@@ -1447,6 +1459,7 @@ static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
                }
                dst->metrics[RTAX_MTU-1] = mtu;
                dst_set_expires(dst, ip_rt_mtu_expires);
+               call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
        }
 }
 
@@ -2626,51 +2639,54 @@ static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 {
        struct rtable *rt = (struct rtable*)skb->dst;
        struct rtmsg *r;
-       struct nlmsghdr  *nlh;
-       unsigned char    *b = skb->tail;
+       struct nlmsghdr *nlh;
        struct rta_cacheinfo ci;
-#ifdef CONFIG_IP_MROUTE
-       struct rtattr *eptr;
-#endif
-       nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
-       r = NLMSG_DATA(nlh);
+
+       nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
+       if (nlh == NULL)
+               return -ENOBUFS;
+
+       r = nlmsg_data(nlh);
        r->rtm_family    = AF_INET;
        r->rtm_dst_len  = 32;
        r->rtm_src_len  = 0;
        r->rtm_tos      = rt->fl.fl4_tos;
        r->rtm_table    = RT_TABLE_MAIN;
+       NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
        r->rtm_type     = rt->rt_type;
        r->rtm_scope    = RT_SCOPE_UNIVERSE;
        r->rtm_protocol = RTPROT_UNSPEC;
        r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
        if (rt->rt_flags & RTCF_NOTIFY)
                r->rtm_flags |= RTM_F_NOTIFY;
-       RTA_PUT(skb, RTA_DST, 4, &rt->rt_dst);
+
+       NLA_PUT_U32(skb, RTA_DST, rt->rt_dst);
+
        if (rt->fl.fl4_src) {
                r->rtm_src_len = 32;
-               RTA_PUT(skb, RTA_SRC, 4, &rt->fl.fl4_src);
+               NLA_PUT_U32(skb, RTA_SRC, rt->fl.fl4_src);
        }
        if (rt->u.dst.dev)
-               RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
+               NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
 #ifdef CONFIG_NET_CLS_ROUTE
        if (rt->u.dst.tclassid)
-               RTA_PUT(skb, RTA_FLOW, 4, &rt->u.dst.tclassid);
+               NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
 #endif
 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
-       if (rt->rt_multipath_alg != IP_MP_ALG_NONE) {
-               __u32 alg = rt->rt_multipath_alg;
-
-               RTA_PUT(skb, RTA_MP_ALGO, 4, &alg);
-       }
+       if (rt->rt_multipath_alg != IP_MP_ALG_NONE)
+               NLA_PUT_U32(skb, RTA_MP_ALGO, rt->rt_multipath_alg);
 #endif
        if (rt->fl.iif)
-               RTA_PUT(skb, RTA_PREFSRC, 4, &rt->rt_spec_dst);
+               NLA_PUT_U32(skb, RTA_PREFSRC, rt->rt_spec_dst);
        else if (rt->rt_src != rt->fl.fl4_src)
-               RTA_PUT(skb, RTA_PREFSRC, 4, &rt->rt_src);
+               NLA_PUT_U32(skb, RTA_PREFSRC, rt->rt_src);
+
        if (rt->rt_dst != rt->rt_gateway)
-               RTA_PUT(skb, RTA_GATEWAY, 4, &rt->rt_gateway);
+               NLA_PUT_U32(skb, RTA_GATEWAY, rt->rt_gateway);
+
        if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
-               goto rtattr_failure;
+               goto nla_put_failure;
+
        ci.rta_lastuse  = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
        ci.rta_used     = rt->u.dst.__use;
        ci.rta_clntref  = atomic_read(&rt->u.dst.__refcnt);
@@ -2687,10 +2703,7 @@ static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
                        ci.rta_tsage = xtime.tv_sec - rt->peer->tcp_ts_stamp;
                }
        }
-#ifdef CONFIG_IP_MROUTE
-       eptr = (struct rtattr*)skb->tail;
-#endif
-       RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
+
        if (rt->fl.iif) {
 #ifdef CONFIG_IP_MROUTE
                u32 dst = rt->rt_dst;
@@ -2702,41 +2715,46 @@ static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
                                if (!nowait) {
                                        if (err == 0)
                                                return 0;
-                                       goto nlmsg_failure;
+                                       goto nla_put_failure;
                                } else {
                                        if (err == -EMSGSIZE)
-                                               goto nlmsg_failure;
-                                       ((struct rta_cacheinfo*)RTA_DATA(eptr))->rta_error = err;
+                                               goto nla_put_failure;
+                                       ci.rta_error = err;
                                }
                        }
                } else
 #endif
-                       RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
+                       NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
        }
 
-       nlh->nlmsg_len = skb->tail - b;
-       return skb->len;
+       NLA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
 
-nlmsg_failure:
-rtattr_failure:
-       skb_trim(skb, b - skb->data);
-       return -1;
+       return nlmsg_end(skb, nlh);
+
+nla_put_failure:
+       return nlmsg_cancel(skb, nlh);
 }
 
 int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
 {
-       struct rtattr **rta = arg;
-       struct rtmsg *rtm = NLMSG_DATA(nlh);
+       struct rtmsg *rtm;
+       struct nlattr *tb[RTA_MAX+1];
        struct rtable *rt = NULL;
-       u32 dst = 0;
-       u32 src = 0;
-       int iif = 0;
-       int err = -ENOBUFS;
+       u32 dst, src, iif;
+       int err;
        struct sk_buff *skb;
 
+       err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
+       if (err < 0)
+               goto errout;
+
+       rtm = nlmsg_data(nlh);
+
        skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
-       if (!skb)
-               goto out;
+       if (skb == NULL) {
+               err = -ENOBUFS;
+               goto errout;
+       }
 
        /* Reserve room for dummy headers, this skb can pass
           through good chunk of routing engine.
@@ -2747,62 +2765,61 @@ int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
        skb->nh.iph->protocol = IPPROTO_ICMP;
        skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
 
-       if (rta[RTA_SRC - 1])
-               memcpy(&src, RTA_DATA(rta[RTA_SRC - 1]), 4);
-       if (rta[RTA_DST - 1])
-               memcpy(&dst, RTA_DATA(rta[RTA_DST - 1]), 4);
-       if (rta[RTA_IIF - 1])
-               memcpy(&iif, RTA_DATA(rta[RTA_IIF - 1]), sizeof(int));
+       src = tb[RTA_SRC] ? nla_get_u32(tb[RTA_SRC]) : 0;
+       dst = tb[RTA_DST] ? nla_get_u32(tb[RTA_DST]) : 0;
+       iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
 
        if (iif) {
-               struct net_device *dev = __dev_get_by_index(iif);
-               err = -ENODEV;
-               if (!dev)
-                       goto out_free;
+               struct net_device *dev;
+
+               dev = __dev_get_by_index(iif);
+               if (dev == NULL) {
+                       err = -ENODEV;
+                       goto errout_free;
+               }
+
                skb->protocol   = htons(ETH_P_IP);
                skb->dev        = dev;
                local_bh_disable();
                err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
                local_bh_enable();
-               rt = (struct rtable*)skb->dst;
-               if (!err && rt->u.dst.error)
+
+               rt = (struct rtable*) skb->dst;
+               if (err == 0 && rt->u.dst.error)
                        err = -rt->u.dst.error;
        } else {
-               struct flowi fl = { .nl_u = { .ip4_u = { .daddr = dst,
-                                                        .saddr = src,
-                                                        .tos = rtm->rtm_tos } } };
-               int oif = 0;
-               if (rta[RTA_OIF - 1])
-                       memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
-               fl.oif = oif;
+               struct flowi fl = {
+                       .nl_u = {
+                               .ip4_u = {
+                                       .daddr = dst,
+                                       .saddr = src,
+                                       .tos = rtm->rtm_tos,
+                               },
+                       },
+                       .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
+               };
                err = ip_route_output_key(&rt, &fl);
        }
+
        if (err)
-               goto out_free;
+               goto errout_free;
 
        skb->dst = &rt->u.dst;
        if (rtm->rtm_flags & RTM_F_NOTIFY)
                rt->rt_flags |= RTCF_NOTIFY;
 
-       NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
-
        err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
                                RTM_NEWROUTE, 0, 0);
-       if (!err)
-               goto out_free;
-       if (err < 0) {
-               err = -EMSGSIZE;
-               goto out_free;
-       }
+       if (err <= 0)
+               goto errout_free;
 
-       err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
-       if (err > 0)
-               err = 0;
-out:   return err;
+       err = rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
+errout:
+       return err;
 
-out_free:
+errout_free:
        kfree_skb(skb);
-       goto out;
+       goto errout;
 }
 
 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
@@ -3144,7 +3161,7 @@ int __init ip_rt_init(void)
                                        rhash_entries,
                                        (num_physpages >= 128 * 1024) ?
                                        15 : 17,
-                                       HASH_HIGHMEM,
+                                       0,
                                        &rt_hash_log,
                                        &rt_hash_mask,
                                        0);