fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / net / ipv4 / fib_semantics.c
index 1de08c1..e63b8a9 100644 (file)
@@ -15,7 +15,6 @@
  *             2 of the License, or (at your option) any later version.
  */
 
-#include <linux/config.h>
 #include <asm/uaccess.h>
 #include <asm/system.h>
 #include <linux/bitops.h>
@@ -34,7 +33,6 @@
 #include <linux/if_arp.h>
 #include <linux/proc_fs.h>
 #include <linux/skbuff.h>
-#include <linux/netlink.h>
 #include <linux/init.h>
 
 #include <net/arp.h>
 #include <net/sock.h>
 #include <net/ip_fib.h>
 #include <net/ip_mp_alg.h>
+#include <net/netlink.h>
+#include <net/nexthop.h>
 
 #include "fib_lookup.h"
 
 #define FSprintk(a...)
 
-static DEFINE_RWLOCK(fib_info_lock);
+static DEFINE_SPINLOCK(fib_info_lock);
 static struct hlist_head *fib_info_hash;
 static struct hlist_head *fib_info_laddrhash;
 static unsigned int fib_hash_size;
@@ -160,7 +160,7 @@ void free_fib_info(struct fib_info *fi)
 
 void fib_release_info(struct fib_info *fi)
 {
-       write_lock_bh(&fib_info_lock);
+       spin_lock_bh(&fib_info_lock);
        if (fi && --fi->fib_treeref == 0) {
                hlist_del(&fi->fib_hash);
                if (fi->fib_prefsrc)
@@ -173,7 +173,7 @@ void fib_release_info(struct fib_info *fi)
                fi->fib_dead = 1;
                fib_info_put(fi);
        }
-       write_unlock_bh(&fib_info_lock);
+       spin_unlock_bh(&fib_info_lock);
 }
 
 static __inline__ int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
@@ -203,7 +203,7 @@ static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
        unsigned int val = fi->fib_nhs;
 
        val ^= fi->fib_protocol;
-       val ^= fi->fib_prefsrc;
+       val ^= (__force u32)fi->fib_prefsrc;
        val ^= fi->fib_priority;
 
        return (val ^ (val >> 7) ^ (val >> 12)) & mask;
@@ -248,14 +248,14 @@ static inline unsigned int fib_devindex_hashfn(unsigned int val)
    Used only by redirect accept routine.
  */
 
-int ip_fib_check_default(u32 gw, struct net_device *dev)
+int ip_fib_check_default(__be32 gw, struct net_device *dev)
 {
        struct hlist_head *head;
        struct hlist_node *node;
        struct fib_nh *nh;
        unsigned int hash;
 
-       read_lock(&fib_info_lock);
+       spin_lock(&fib_info_lock);
 
        hash = fib_devindex_hashfn(dev->ifindex);
        head = &fib_info_devhash[hash];
@@ -263,41 +263,65 @@ int ip_fib_check_default(u32 gw, struct net_device *dev)
                if (nh->nh_dev == dev &&
                    nh->nh_gw == gw &&
                    !(nh->nh_flags&RTNH_F_DEAD)) {
-                       read_unlock(&fib_info_lock);
+                       spin_unlock(&fib_info_lock);
                        return 0;
                }
        }
 
-       read_unlock(&fib_info_lock);
+       spin_unlock(&fib_info_lock);
 
        return -1;
 }
 
-void rtmsg_fib(int event, u32 key, struct fib_alias *fa,
-              int z, int tb_id,
-              struct nlmsghdr *n, struct netlink_skb_parms *req)
+static inline size_t fib_nlmsg_size(struct fib_info *fi)
 {
-       struct sk_buff *skb;
-       u32 pid = req ? req->pid : n->nlmsg_pid;
-       int size = NLMSG_SPACE(sizeof(struct rtmsg)+256);
+       size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
+                        + nla_total_size(4) /* RTA_TABLE */
+                        + nla_total_size(4) /* RTA_DST */
+                        + nla_total_size(4) /* RTA_PRIORITY */
+                        + nla_total_size(4); /* RTA_PREFSRC */
 
-       skb = alloc_skb(size, GFP_KERNEL);
-       if (!skb)
-               return;
+       /* space for nested metrics */
+       payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 
-       if (fib_dump_info(skb, pid, n->nlmsg_seq, event, tb_id,
-                         fa->fa_type, fa->fa_scope, &key, z,
-                         fa->fa_tos,
-                         fa->fa_info, 0) < 0) {
-               kfree_skb(skb);
-               return;
+       if (fi->fib_nhs) {
+               /* Also handles the special case fib_nhs == 1 */
+
+               /* each nexthop is packed in an attribute */
+               size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
+
+               /* may contain flow and gateway attribute */
+               nhsize += 2 * nla_total_size(4);
+
+               /* all nexthops are packed in a nested attribute */
+               payload += nla_total_size(fi->fib_nhs * nhsize);
        }
-       NETLINK_CB(skb).dst_group = RTNLGRP_IPV4_ROUTE;
-       if (n->nlmsg_flags&NLM_F_ECHO)
-               atomic_inc(&skb->users);
-       netlink_broadcast(rtnl, skb, pid, RTNLGRP_IPV4_ROUTE, GFP_KERNEL);
-       if (n->nlmsg_flags&NLM_F_ECHO)
-               netlink_unicast(rtnl, skb, pid, MSG_DONTWAIT);
+
+       return payload;
+}
+
+void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
+              int dst_len, u32 tb_id, struct nl_info *info)
+{
+       struct sk_buff *skb;
+       u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
+       int err = -ENOBUFS;
+
+       skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
+       if (skb == NULL)
+               goto errout;
+
+       err = fib_dump_info(skb, info->pid, seq, event, tb_id,
+                           fa->fa_type, fa->fa_scope, key, dst_len,
+                           fa->fa_tos, fa->fa_info, 0);
+       /* failure implies BUG in fib_nlmsg_size() */
+       BUG_ON(err < 0);
+
+       err = rtnl_notify(skb, info->pid, RTNLGRP_IPV4_ROUTE,
+                         info->nlh, GFP_KERNEL);
+errout:
+       if (err < 0)
+               rtnl_set_sk_err(RTNLGRP_IPV4_ROUTE, err);
 }
 
 /* Return the first fib alias matching TOS with
@@ -343,102 +367,100 @@ int fib_detect_death(struct fib_info *fi, int order,
 
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 
-static u32 fib_get_attr32(struct rtattr *attr, int attrlen, int type)
-{
-       while (RTA_OK(attr,attrlen)) {
-               if (attr->rta_type == type)
-                       return *(u32*)RTA_DATA(attr);
-               attr = RTA_NEXT(attr, attrlen);
-       }
-       return 0;
-}
-
-static int
-fib_count_nexthops(struct rtattr *rta)
+static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 {
        int nhs = 0;
-       struct rtnexthop *nhp = RTA_DATA(rta);
-       int nhlen = RTA_PAYLOAD(rta);
 
-       while (nhlen >= (int)sizeof(struct rtnexthop)) {
-               if ((nhlen -= nhp->rtnh_len) < 0)
-                       return 0;
+       while (rtnh_ok(rtnh, remaining)) {
                nhs++;
-               nhp = RTNH_NEXT(nhp);
-       };
-       return nhs;
+               rtnh = rtnh_next(rtnh, &remaining);
+       }
+
+       /* leftover implies invalid nexthop configuration, discard it */
+       return remaining > 0 ? 0 : nhs;
 }
 
-static int
-fib_get_nhs(struct fib_info *fi, const struct rtattr *rta, const struct rtmsg *r)
+static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
+                      int remaining, struct fib_config *cfg)
 {
-       struct rtnexthop *nhp = RTA_DATA(rta);
-       int nhlen = RTA_PAYLOAD(rta);
-
        change_nexthops(fi) {
-               int attrlen = nhlen - sizeof(struct rtnexthop);
-               if (attrlen < 0 || (nhlen -= nhp->rtnh_len) < 0)
+               int attrlen;
+
+               if (!rtnh_ok(rtnh, remaining))
                        return -EINVAL;
-               nh->nh_flags = (r->rtm_flags&~0xFF) | nhp->rtnh_flags;
-               nh->nh_oif = nhp->rtnh_ifindex;
-               nh->nh_weight = nhp->rtnh_hops + 1;
-               if (attrlen) {
-                       nh->nh_gw = fib_get_attr32(RTNH_DATA(nhp), attrlen, RTA_GATEWAY);
+
+               nh->nh_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
+               nh->nh_oif = rtnh->rtnh_ifindex;
+               nh->nh_weight = rtnh->rtnh_hops + 1;
+
+               attrlen = rtnh_attrlen(rtnh);
+               if (attrlen > 0) {
+                       struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
+
+                       nla = nla_find(attrs, attrlen, RTA_GATEWAY);
+                       nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 #ifdef CONFIG_NET_CLS_ROUTE
-                       nh->nh_tclassid = fib_get_attr32(RTNH_DATA(nhp), attrlen, RTA_FLOW);
+                       nla = nla_find(attrs, attrlen, RTA_FLOW);
+                       nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 #endif
                }
-               nhp = RTNH_NEXT(nhp);
+
+               rtnh = rtnh_next(rtnh, &remaining);
        } endfor_nexthops(fi);
+
        return 0;
 }
 
 #endif
 
-int fib_nh_match(struct rtmsg *r, struct nlmsghdr *nlh, struct kern_rta *rta,
-                struct fib_info *fi)
+int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 {
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
-       struct rtnexthop *nhp;
-       int nhlen;
+       struct rtnexthop *rtnh;
+       int remaining;
 #endif
 
-       if (rta->rta_priority &&
-           *rta->rta_priority != fi->fib_priority)
+       if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
                return 1;
 
-       if (rta->rta_oif || rta->rta_gw) {
-               if ((!rta->rta_oif || *rta->rta_oif == fi->fib_nh->nh_oif) &&
-                   (!rta->rta_gw  || memcmp(rta->rta_gw, &fi->fib_nh->nh_gw, 4) == 0))
+       if (cfg->fc_oif || cfg->fc_gw) {
+               if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
+                   (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
                        return 0;
                return 1;
        }
 
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
-       if (rta->rta_mp == NULL)
+       if (cfg->fc_mp == NULL)
                return 0;
-       nhp = RTA_DATA(rta->rta_mp);
-       nhlen = RTA_PAYLOAD(rta->rta_mp);
+
+       rtnh = cfg->fc_mp;
+       remaining = cfg->fc_mp_len;
        
        for_nexthops(fi) {
-               int attrlen = nhlen - sizeof(struct rtnexthop);
-               u32 gw;
+               int attrlen;
 
-               if (attrlen < 0 || (nhlen -= nhp->rtnh_len) < 0)
+               if (!rtnh_ok(rtnh, remaining))
                        return -EINVAL;
-               if (nhp->rtnh_ifindex && nhp->rtnh_ifindex != nh->nh_oif)
+
+               if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
                        return 1;
-               if (attrlen) {
-                       gw = fib_get_attr32(RTNH_DATA(nhp), attrlen, RTA_GATEWAY);
-                       if (gw && gw != nh->nh_gw)
+
+               attrlen = rtnh_attrlen(rtnh);
+               if (attrlen < 0) {
+                       struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
+
+                       nla = nla_find(attrs, attrlen, RTA_GATEWAY);
+                       if (nla && nla_get_be32(nla) != nh->nh_gw)
                                return 1;
 #ifdef CONFIG_NET_CLS_ROUTE
-                       gw = fib_get_attr32(RTNH_DATA(nhp), attrlen, RTA_FLOW);
-                       if (gw && gw != nh->nh_tclassid)
+                       nla = nla_find(attrs, attrlen, RTA_FLOW);
+                       if (nla && nla_get_u32(nla) != nh->nh_tclassid)
                                return 1;
 #endif
                }
-               nhp = RTNH_NEXT(nhp);
+
+               rtnh = rtnh_next(rtnh, &remaining);
        } endfor_nexthops(fi);
 #endif
        return 0;
@@ -489,7 +511,8 @@ int fib_nh_match(struct rtmsg *r, struct nlmsghdr *nlh, struct kern_rta *rta,
                                                |-> {local prefix} (terminal node)
  */
 
-static int fib_check_nh(const struct rtmsg *r, struct fib_info *fi, struct fib_nh *nh)
+static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
+                       struct fib_nh *nh)
 {
        int err;
 
@@ -503,7 +526,7 @@ static int fib_check_nh(const struct rtmsg *r, struct fib_info *fi, struct fib_n
                if (nh->nh_flags&RTNH_F_ONLINK) {
                        struct net_device *dev;
 
-                       if (r->rtm_scope >= RT_SCOPE_LINK)
+                       if (cfg->fc_scope >= RT_SCOPE_LINK)
                                return -EINVAL;
                        if (inet_addr_type(nh->nh_gw) != RTN_UNICAST)
                                return -EINVAL;
@@ -517,10 +540,15 @@ static int fib_check_nh(const struct rtmsg *r, struct fib_info *fi, struct fib_n
                        return 0;
                }
                {
-                       struct flowi fl = { .nl_u = { .ip4_u =
-                                                     { .daddr = nh->nh_gw,
-                                                       .scope = r->rtm_scope + 1 } },
-                                           .oif = nh->nh_oif };
+                       struct flowi fl = {
+                               .nl_u = {
+                                       .ip4_u = {
+                                               .daddr = nh->nh_gw,
+                                               .scope = cfg->fc_scope + 1,
+                                       },
+                               },
+                               .oif = nh->nh_oif,
+                       };
 
                        /* It is not necessary, but requires a bit of thinking */
                        if (fl.fl4_scope < RT_SCOPE_LINK)
@@ -564,11 +592,11 @@ out:
        return 0;
 }
 
-static inline unsigned int fib_laddr_hashfn(u32 val)
+static inline unsigned int fib_laddr_hashfn(__be32 val)
 {
        unsigned int mask = (fib_hash_size - 1);
 
-       return (val ^ (val >> 7) ^ (val >> 14)) & mask;
+       return ((__force u32)val ^ ((__force u32)val >> 7) ^ ((__force u32)val >> 14)) & mask;
 }
 
 static struct hlist_head *fib_hash_alloc(int bytes)
@@ -599,7 +627,7 @@ static void fib_hash_move(struct hlist_head *new_info_hash,
        unsigned int old_size = fib_hash_size;
        unsigned int i, bytes;
 
-       write_lock_bh(&fib_info_lock);
+       spin_lock_bh(&fib_info_lock);
        old_info_hash = fib_info_hash;
        old_laddrhash = fib_info_laddrhash;
        fib_hash_size = new_size;
@@ -640,46 +668,35 @@ static void fib_hash_move(struct hlist_head *new_info_hash,
        }
        fib_info_laddrhash = new_laddrhash;
 
-       write_unlock_bh(&fib_info_lock);
+       spin_unlock_bh(&fib_info_lock);
 
        bytes = old_size * sizeof(struct hlist_head *);
        fib_hash_free(old_info_hash, bytes);
        fib_hash_free(old_laddrhash, bytes);
 }
 
-struct fib_info *
-fib_create_info(const struct rtmsg *r, struct kern_rta *rta,
-               const struct nlmsghdr *nlh, int *errp)
+struct fib_info *fib_create_info(struct fib_config *cfg)
 {
        int err;
        struct fib_info *fi = NULL;
        struct fib_info *ofi;
-#ifdef CONFIG_IP_ROUTE_MULTIPATH
        int nhs = 1;
-#else
-       const int nhs = 1;
-#endif
-#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
-       u32 mp_alg = IP_MP_ALG_NONE;
-#endif
 
        /* Fast check to catch the most weird cases */
-       if (fib_props[r->rtm_type].scope > r->rtm_scope)
+       if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
                goto err_inval;
 
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
-       if (rta->rta_mp) {
-               nhs = fib_count_nexthops(rta->rta_mp);
+       if (cfg->fc_mp) {
+               nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
                if (nhs == 0)
                        goto err_inval;
        }
 #endif
 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
-       if (rta->rta_mp_alg) {
-               mp_alg = *rta->rta_mp_alg;
-
-               if (mp_alg < IP_MP_ALG_NONE ||
-                   mp_alg > IP_MP_ALG_MAX)
+       if (cfg->fc_mp_alg) {
+               if (cfg->fc_mp_alg < IP_MP_ALG_NONE ||
+                   cfg->fc_mp_alg > IP_MP_ALG_MAX)
                        goto err_inval;
        }
 #endif
@@ -710,49 +727,47 @@ fib_create_info(const struct rtmsg *r, struct kern_rta *rta,
                        goto failure;
        }
 
-       fi = kmalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
+       fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
        if (fi == NULL)
                goto failure;
        fib_info_cnt++;
-       memset(fi, 0, sizeof(*fi)+nhs*sizeof(struct fib_nh));
 
-       fi->fib_protocol = r->rtm_protocol;
+       fi->fib_protocol = cfg->fc_protocol;
+       fi->fib_flags = cfg->fc_flags;
+       fi->fib_priority = cfg->fc_priority;
+       fi->fib_prefsrc = cfg->fc_prefsrc;
 
        fi->fib_nhs = nhs;
        change_nexthops(fi) {
                nh->nh_parent = fi;
        } endfor_nexthops(fi)
 
-       fi->fib_flags = r->rtm_flags;
-       if (rta->rta_priority)
-               fi->fib_priority = *rta->rta_priority;
-       if (rta->rta_mx) {
-               int attrlen = RTA_PAYLOAD(rta->rta_mx);
-               struct rtattr *attr = RTA_DATA(rta->rta_mx);
-
-               while (RTA_OK(attr, attrlen)) {
-                       unsigned flavor = attr->rta_type;
-                       if (flavor) {
-                               if (flavor > RTAX_MAX)
+       if (cfg->fc_mx) {
+               struct nlattr *nla;
+               int remaining;
+
+               nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
+                       int type = nla->nla_type;
+
+                       if (type) {
+                               if (type > RTAX_MAX)
                                        goto err_inval;
-                               fi->fib_metrics[flavor-1] = *(unsigned*)RTA_DATA(attr);
+                               fi->fib_metrics[type - 1] = nla_get_u32(nla);
                        }
-                       attr = RTA_NEXT(attr, attrlen);
                }
        }
-       if (rta->rta_prefsrc)
-               memcpy(&fi->fib_prefsrc, rta->rta_prefsrc, 4);
 
-       if (rta->rta_mp) {
+       if (cfg->fc_mp) {
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
-               if ((err = fib_get_nhs(fi, rta->rta_mp, r)) != 0)
+               err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
+               if (err != 0)
                        goto failure;
-               if (rta->rta_oif && fi->fib_nh->nh_oif != *rta->rta_oif)
+               if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
                        goto err_inval;
-               if (rta->rta_gw && memcmp(&fi->fib_nh->nh_gw, rta->rta_gw, 4))
+               if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
                        goto err_inval;
 #ifdef CONFIG_NET_CLS_ROUTE
-               if (rta->rta_flow && memcmp(&fi->fib_nh->nh_tclassid, rta->rta_flow, 4))
+               if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
                        goto err_inval;
 #endif
 #else
@@ -760,34 +775,32 @@ fib_create_info(const struct rtmsg *r, struct kern_rta *rta,
 #endif
        } else {
                struct fib_nh *nh = fi->fib_nh;
-               if (rta->rta_oif)
-                       nh->nh_oif = *rta->rta_oif;
-               if (rta->rta_gw)
-                       memcpy(&nh->nh_gw, rta->rta_gw, 4);
+
+               nh->nh_oif = cfg->fc_oif;
+               nh->nh_gw = cfg->fc_gw;
+               nh->nh_flags = cfg->fc_flags;
 #ifdef CONFIG_NET_CLS_ROUTE
-               if (rta->rta_flow)
-                       memcpy(&nh->nh_tclassid, rta->rta_flow, 4);
+               nh->nh_tclassid = cfg->fc_flow;
 #endif
-               nh->nh_flags = r->rtm_flags;
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
                nh->nh_weight = 1;
 #endif
        }
 
 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
-       fi->fib_mp_alg = mp_alg;
+       fi->fib_mp_alg = cfg->fc_mp_alg;
 #endif
 
-       if (fib_props[r->rtm_type].error) {
-               if (rta->rta_gw || rta->rta_oif || rta->rta_mp)
+       if (fib_props[cfg->fc_type].error) {
+               if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
                        goto err_inval;
                goto link_it;
        }
 
-       if (r->rtm_scope > RT_SCOPE_HOST)
+       if (cfg->fc_scope > RT_SCOPE_HOST)
                goto err_inval;
 
-       if (r->rtm_scope == RT_SCOPE_HOST) {
+       if (cfg->fc_scope == RT_SCOPE_HOST) {
                struct fib_nh *nh = fi->fib_nh;
 
                /* Local address is added. */
@@ -800,14 +813,14 @@ fib_create_info(const struct rtmsg *r, struct kern_rta *rta,
                        goto failure;
        } else {
                change_nexthops(fi) {
-                       if ((err = fib_check_nh(r, fi, nh)) != 0)
+                       if ((err = fib_check_nh(cfg, fi, nh)) != 0)
                                goto failure;
                } endfor_nexthops(fi)
        }
 
        if (fi->fib_prefsrc) {
-               if (r->rtm_type != RTN_LOCAL || rta->rta_dst == NULL ||
-                   memcmp(&fi->fib_prefsrc, rta->rta_dst, 4))
+               if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
+                   fi->fib_prefsrc != cfg->fc_dst)
                        if (inet_addr_type(fi->fib_prefsrc) != RTN_LOCAL)
                                goto err_inval;
        }
@@ -822,7 +835,7 @@ link_it:
 
        fi->fib_treeref++;
        atomic_inc(&fi->fib_clntref);
-       write_lock_bh(&fib_info_lock);
+       spin_lock_bh(&fib_info_lock);
        hlist_add_head(&fi->fib_hash,
                       &fib_info_hash[fib_info_hashfn(fi)]);
        if (fi->fib_prefsrc) {
@@ -841,24 +854,24 @@ link_it:
                head = &fib_info_devhash[hash];
                hlist_add_head(&nh->nh_hash, head);
        } endfor_nexthops(fi)
-       write_unlock_bh(&fib_info_lock);
+       spin_unlock_bh(&fib_info_lock);
        return fi;
 
 err_inval:
        err = -EINVAL;
 
 failure:
-        *errp = err;
         if (fi) {
                fi->fib_dead = 1;
                free_fib_info(fi);
        }
-       return NULL;
+
+       return ERR_PTR(err);
 }
 
 /* Note! fib_semantic_match intentionally uses  RCU list functions. */
 int fib_semantic_match(struct list_head *head, const struct flowi *flp,
-                      struct fib_result *res, __u32 zone, __u32 mask, 
+                      struct fib_result *res, __be32 zone, __be32 mask,
                        int prefixlen)
 {
        struct fib_alias *fa;
@@ -925,8 +938,7 @@ out_fill_res:
        res->fi = fa->fa_info;
 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
        res->netmask = mask;
-       res->network = zone &
-               (0xFFFFFFFF >> (32 - prefixlen));
+       res->network = zone & inet_make_mask(prefixlen);
 #endif
        atomic_inc(&res->fi->fib_clntref);
        return 0;
@@ -934,225 +946,94 @@ out_fill_res:
 
 /* Find appropriate source address to this destination */
 
-u32 __fib_res_prefsrc(struct fib_result *res)
+__be32 __fib_res_prefsrc(struct fib_result *res)
 {
        return inet_select_addr(FIB_RES_DEV(*res), FIB_RES_GW(*res), res->scope);
 }
 
-int
-fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
-             u8 tb_id, u8 type, u8 scope, void *dst, int dst_len, u8 tos,
-             struct fib_info *fi, unsigned int flags)
+int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
+                 u32 tb_id, u8 type, u8 scope, __be32 dst, int dst_len, u8 tos,
+                 struct fib_info *fi, unsigned int flags)
 {
+       struct nlmsghdr *nlh;
        struct rtmsg *rtm;
-       struct nlmsghdr  *nlh;
-       unsigned char    *b = skb->tail;
 
-       nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*rtm), flags);
-       rtm = NLMSG_DATA(nlh);
+       nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
+       if (nlh == NULL)
+               return -ENOBUFS;
+
+       rtm = nlmsg_data(nlh);
        rtm->rtm_family = AF_INET;
        rtm->rtm_dst_len = dst_len;
        rtm->rtm_src_len = 0;
        rtm->rtm_tos = tos;
        rtm->rtm_table = tb_id;
+       NLA_PUT_U32(skb, RTA_TABLE, tb_id);
        rtm->rtm_type = type;
        rtm->rtm_flags = fi->fib_flags;
        rtm->rtm_scope = scope;
-       if (rtm->rtm_dst_len)
-               RTA_PUT(skb, RTA_DST, 4, dst);
        rtm->rtm_protocol = fi->fib_protocol;
+
+       if (rtm->rtm_dst_len)
+               NLA_PUT_BE32(skb, RTA_DST, dst);
+
        if (fi->fib_priority)
-               RTA_PUT(skb, RTA_PRIORITY, 4, &fi->fib_priority);
-#ifdef CONFIG_NET_CLS_ROUTE
-       if (fi->fib_nh[0].nh_tclassid)
-               RTA_PUT(skb, RTA_FLOW, 4, &fi->fib_nh[0].nh_tclassid);
-#endif
+               NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
+
        if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
-               goto rtattr_failure;
+               goto nla_put_failure;
+
        if (fi->fib_prefsrc)
-               RTA_PUT(skb, RTA_PREFSRC, 4, &fi->fib_prefsrc);
+               NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
+
        if (fi->fib_nhs == 1) {
                if (fi->fib_nh->nh_gw)
-                       RTA_PUT(skb, RTA_GATEWAY, 4, &fi->fib_nh->nh_gw);
+                       NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
+
                if (fi->fib_nh->nh_oif)
-                       RTA_PUT(skb, RTA_OIF, sizeof(int), &fi->fib_nh->nh_oif);
+                       NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
+#ifdef CONFIG_NET_CLS_ROUTE
+               if (fi->fib_nh[0].nh_tclassid)
+                       NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
+#endif
        }
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
        if (fi->fib_nhs > 1) {
-               struct rtnexthop *nhp;
-               struct rtattr *mp_head;
-               if (skb_tailroom(skb) <= RTA_SPACE(0))
-                       goto rtattr_failure;
-               mp_head = (struct rtattr*)skb_put(skb, RTA_SPACE(0));
+               struct rtnexthop *rtnh;
+               struct nlattr *mp;
 
-               for_nexthops(fi) {
-                       if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
-                               goto rtattr_failure;
-                       nhp = (struct rtnexthop*)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
-                       nhp->rtnh_flags = nh->nh_flags & 0xFF;
-                       nhp->rtnh_hops = nh->nh_weight-1;
-                       nhp->rtnh_ifindex = nh->nh_oif;
-                       if (nh->nh_gw)
-                               RTA_PUT(skb, RTA_GATEWAY, 4, &nh->nh_gw);
-                       nhp->rtnh_len = skb->tail - (unsigned char*)nhp;
-               } endfor_nexthops(fi);
-               mp_head->rta_type = RTA_MULTIPATH;
-               mp_head->rta_len = skb->tail - (u8*)mp_head;
-       }
-#endif
-       nlh->nlmsg_len = skb->tail - b;
-       return skb->len;
+               mp = nla_nest_start(skb, RTA_MULTIPATH);
+               if (mp == NULL)
+                       goto nla_put_failure;
 
-nlmsg_failure:
-rtattr_failure:
-       skb_trim(skb, b - skb->data);
-       return -1;
-}
-
-#ifndef CONFIG_IP_NOSIOCRT
-
-int
-fib_convert_rtentry(int cmd, struct nlmsghdr *nl, struct rtmsg *rtm,
-                   struct kern_rta *rta, struct rtentry *r)
-{
-       int    plen;
-       u32    *ptr;
-
-       memset(rtm, 0, sizeof(*rtm));
-       memset(rta, 0, sizeof(*rta));
-
-       if (r->rt_dst.sa_family != AF_INET)
-               return -EAFNOSUPPORT;
-
-       /* Check mask for validity:
-          a) it must be contiguous.
-          b) destination must have all host bits clear.
-          c) if application forgot to set correct family (AF_INET),
-             reject request unless it is absolutely clear i.e.
-             both family and mask are zero.
-        */
-       plen = 32;
-       ptr = &((struct sockaddr_in*)&r->rt_dst)->sin_addr.s_addr;
-       if (!(r->rt_flags&RTF_HOST)) {
-               u32 mask = ((struct sockaddr_in*)&r->rt_genmask)->sin_addr.s_addr;
-               if (r->rt_genmask.sa_family != AF_INET) {
-                       if (mask || r->rt_genmask.sa_family)
-                               return -EAFNOSUPPORT;
-               }
-               if (bad_mask(mask, *ptr))
-                       return -EINVAL;
-               plen = inet_mask_len(mask);
-       }
-
-       nl->nlmsg_flags = NLM_F_REQUEST;
-       nl->nlmsg_pid = 0;
-       nl->nlmsg_seq = 0;
-       nl->nlmsg_len = NLMSG_LENGTH(sizeof(*rtm));
-       if (cmd == SIOCDELRT) {
-               nl->nlmsg_type = RTM_DELROUTE;
-               nl->nlmsg_flags = 0;
-       } else {
-               nl->nlmsg_type = RTM_NEWROUTE;
-               nl->nlmsg_flags = NLM_F_REQUEST|NLM_F_CREATE;
-               rtm->rtm_protocol = RTPROT_BOOT;
-       }
+               for_nexthops(fi) {
+                       rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
+                       if (rtnh == NULL)
+                               goto nla_put_failure;
 
-       rtm->rtm_dst_len = plen;
-       rta->rta_dst = ptr;
+                       rtnh->rtnh_flags = nh->nh_flags & 0xFF;
+                       rtnh->rtnh_hops = nh->nh_weight - 1;
+                       rtnh->rtnh_ifindex = nh->nh_oif;
 
-       if (r->rt_metric) {
-               *(u32*)&r->rt_pad3 = r->rt_metric - 1;
-               rta->rta_priority = (u32*)&r->rt_pad3;
-       }
-       if (r->rt_flags&RTF_REJECT) {
-               rtm->rtm_scope = RT_SCOPE_HOST;
-               rtm->rtm_type = RTN_UNREACHABLE;
-               return 0;
-       }
-       rtm->rtm_scope = RT_SCOPE_NOWHERE;
-       rtm->rtm_type = RTN_UNICAST;
-
-       if (r->rt_dev) {
-               char *colon;
-               struct net_device *dev;
-               char   devname[IFNAMSIZ];
-
-               if (copy_from_user(devname, r->rt_dev, IFNAMSIZ-1))
-                       return -EFAULT;
-               devname[IFNAMSIZ-1] = 0;
-               colon = strchr(devname, ':');
-               if (colon)
-                       *colon = 0;
-               dev = __dev_get_by_name(devname);
-               if (!dev)
-                       return -ENODEV;
-               rta->rta_oif = &dev->ifindex;
-               if (colon) {
-                       struct in_ifaddr *ifa;
-                       struct in_device *in_dev = __in_dev_get_rtnl(dev);
-                       if (!in_dev)
-                               return -ENODEV;
-                       *colon = ':';
-                       for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
-                               if (strcmp(ifa->ifa_label, devname) == 0)
-                                       break;
-                       if (ifa == NULL)
-                               return -ENODEV;
-                       rta->rta_prefsrc = &ifa->ifa_local;
-               }
-       }
+                       if (nh->nh_gw)
+                               NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
+#ifdef CONFIG_NET_CLS_ROUTE
+                       if (nh->nh_tclassid)
+                               NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
+#endif
+                       /* length of rtnetlink header + attributes */
+                       rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
+               } endfor_nexthops(fi);
 
-       ptr = &((struct sockaddr_in*)&r->rt_gateway)->sin_addr.s_addr;
-       if (r->rt_gateway.sa_family == AF_INET && *ptr) {
-               rta->rta_gw = ptr;
-               if (r->rt_flags&RTF_GATEWAY && inet_addr_type(*ptr) == RTN_UNICAST)
-                       rtm->rtm_scope = RT_SCOPE_UNIVERSE;
+               nla_nest_end(skb, mp);
        }
+#endif
+       return nlmsg_end(skb, nlh);
 
-       if (cmd == SIOCDELRT)
-               return 0;
-
-       if (r->rt_flags&RTF_GATEWAY && rta->rta_gw == NULL)
-               return -EINVAL;
-
-       if (rtm->rtm_scope == RT_SCOPE_NOWHERE)
-               rtm->rtm_scope = RT_SCOPE_LINK;
-
-       if (r->rt_flags&(RTF_MTU|RTF_WINDOW|RTF_IRTT)) {
-               struct rtattr *rec;
-               struct rtattr *mx = kmalloc(RTA_LENGTH(3*RTA_LENGTH(4)), GFP_KERNEL);
-               if (mx == NULL)
-                       return -ENOMEM;
-               rta->rta_mx = mx;
-               mx->rta_type = RTA_METRICS;
-               mx->rta_len  = RTA_LENGTH(0);
-               if (r->rt_flags&RTF_MTU) {
-                       rec = (void*)((char*)mx + RTA_ALIGN(mx->rta_len));
-                       rec->rta_type = RTAX_ADVMSS;
-                       rec->rta_len = RTA_LENGTH(4);
-                       mx->rta_len += RTA_LENGTH(4);
-                       *(u32*)RTA_DATA(rec) = r->rt_mtu - 40;
-               }
-               if (r->rt_flags&RTF_WINDOW) {
-                       rec = (void*)((char*)mx + RTA_ALIGN(mx->rta_len));
-                       rec->rta_type = RTAX_WINDOW;
-                       rec->rta_len = RTA_LENGTH(4);
-                       mx->rta_len += RTA_LENGTH(4);
-                       *(u32*)RTA_DATA(rec) = r->rt_window;
-               }
-               if (r->rt_flags&RTF_IRTT) {
-                       rec = (void*)((char*)mx + RTA_ALIGN(mx->rta_len));
-                       rec->rta_type = RTAX_RTT;
-                       rec->rta_len = RTA_LENGTH(4);
-                       mx->rta_len += RTA_LENGTH(4);
-                       *(u32*)RTA_DATA(rec) = r->rt_irtt<<3;
-               }
-       }
-       return 0;
+nla_put_failure:
+       return nlmsg_cancel(skb, nlh);
 }
 
-#endif
-
 /*
    Update FIB if:
    - local address disappeared -> we must delete all the entries
@@ -1160,7 +1041,7 @@ fib_convert_rtentry(int cmd, struct nlmsghdr *nl, struct rtmsg *rtm,
    - device went down -> we must shutdown all nexthops going via it.
  */
 
-int fib_sync_down(u32 local, struct net_device *dev, int force)
+int fib_sync_down(__be32 local, struct net_device *dev, int force)
 {
        int ret = 0;
        int scope = RT_SCOPE_NOWHERE;