aboutsummaryrefslogtreecommitdiffstats
path: root/net
diff options
context:
space:
mode:
Diffstat (limited to 'net')
-rw-r--r--net/ipv4/route.c76
1 files changed, 43 insertions, 33 deletions
diff --git a/net/ipv4/route.c b/net/ipv4/route.c
index e4e37edbad60..67c3ed772c27 100644
--- a/net/ipv4/route.c
+++ b/net/ipv4/route.c
@@ -250,7 +250,6 @@ static inline void rt_hash_lock_init(void)
250static struct rt_hash_bucket *rt_hash_table __read_mostly; 250static struct rt_hash_bucket *rt_hash_table __read_mostly;
251static unsigned rt_hash_mask __read_mostly; 251static unsigned rt_hash_mask __read_mostly;
252static unsigned int rt_hash_log __read_mostly; 252static unsigned int rt_hash_log __read_mostly;
253static atomic_t rt_genid __read_mostly;
254 253
255static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 254static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
256#define RT_CACHE_STAT_INC(field) \ 255#define RT_CACHE_STAT_INC(field) \
@@ -265,6 +264,11 @@ static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
265 & rt_hash_mask; 264 & rt_hash_mask;
266} 265}
267 266
267static inline int rt_genid(struct net *net)
268{
269 return atomic_read(&net->ipv4.rt_genid);
270}
271
268#ifdef CONFIG_PROC_FS 272#ifdef CONFIG_PROC_FS
269struct rt_cache_iter_state { 273struct rt_cache_iter_state {
270 struct seq_net_private p; 274 struct seq_net_private p;
@@ -334,7 +338,7 @@ static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
334 struct rt_cache_iter_state *st = seq->private; 338 struct rt_cache_iter_state *st = seq->private;
335 if (*pos) 339 if (*pos)
336 return rt_cache_get_idx(seq, *pos - 1); 340 return rt_cache_get_idx(seq, *pos - 1);
337 st->genid = atomic_read(&rt_genid); 341 st->genid = rt_genid(seq_file_net(seq));
338 return SEQ_START_TOKEN; 342 return SEQ_START_TOKEN;
339} 343}
340 344
@@ -681,6 +685,11 @@ static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
681 return dev_net(rt1->u.dst.dev) == dev_net(rt2->u.dst.dev); 685 return dev_net(rt1->u.dst.dev) == dev_net(rt2->u.dst.dev);
682} 686}
683 687
688static inline int rt_is_expired(struct rtable *rth)
689{
690 return rth->rt_genid != rt_genid(dev_net(rth->u.dst.dev));
691}
692
684/* 693/*
685 * Perform a full scan of hash table and free all entries. 694 * Perform a full scan of hash table and free all entries.
686 * Can be called by a softirq or a process. 695 * Can be called by a softirq or a process.
@@ -736,7 +745,7 @@ static void rt_check_expire(void)
736 continue; 745 continue;
737 spin_lock_bh(rt_hash_lock_addr(i)); 746 spin_lock_bh(rt_hash_lock_addr(i));
738 while ((rth = *rthp) != NULL) { 747 while ((rth = *rthp) != NULL) {
739 if (rth->rt_genid != atomic_read(&rt_genid)) { 748 if (rt_is_expired(rth)) {
740 *rthp = rth->u.dst.rt_next; 749 *rthp = rth->u.dst.rt_next;
741 rt_free(rth); 750 rt_free(rth);
742 continue; 751 continue;
@@ -784,7 +793,7 @@ static void rt_cache_invalidate(struct net *net)
784 unsigned char shuffle; 793 unsigned char shuffle;
785 794
786 get_random_bytes(&shuffle, sizeof(shuffle)); 795 get_random_bytes(&shuffle, sizeof(shuffle));
787 atomic_add(shuffle + 1U, &rt_genid); 796 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
788} 797}
789 798
790/* 799/*
@@ -881,7 +890,7 @@ static int rt_garbage_collect(struct dst_ops *ops)
881 rthp = &rt_hash_table[k].chain; 890 rthp = &rt_hash_table[k].chain;
882 spin_lock_bh(rt_hash_lock_addr(k)); 891 spin_lock_bh(rt_hash_lock_addr(k));
883 while ((rth = *rthp) != NULL) { 892 while ((rth = *rthp) != NULL) {
884 if (rth->rt_genid == atomic_read(&rt_genid) && 893 if (!rt_is_expired(rth) &&
885 !rt_may_expire(rth, tmo, expire)) { 894 !rt_may_expire(rth, tmo, expire)) {
886 tmo >>= 1; 895 tmo >>= 1;
887 rthp = &rth->u.dst.rt_next; 896 rthp = &rth->u.dst.rt_next;
@@ -963,7 +972,7 @@ restart:
963 972
964 spin_lock_bh(rt_hash_lock_addr(hash)); 973 spin_lock_bh(rt_hash_lock_addr(hash));
965 while ((rth = *rthp) != NULL) { 974 while ((rth = *rthp) != NULL) {
966 if (rth->rt_genid != atomic_read(&rt_genid)) { 975 if (rt_is_expired(rth)) {
967 *rthp = rth->u.dst.rt_next; 976 *rthp = rth->u.dst.rt_next;
968 rt_free(rth); 977 rt_free(rth);
969 continue; 978 continue;
@@ -1139,7 +1148,7 @@ static void rt_del(unsigned hash, struct rtable *rt)
1139 spin_lock_bh(rt_hash_lock_addr(hash)); 1148 spin_lock_bh(rt_hash_lock_addr(hash));
1140 ip_rt_put(rt); 1149 ip_rt_put(rt);
1141 while ((aux = *rthp) != NULL) { 1150 while ((aux = *rthp) != NULL) {
1142 if (aux == rt || (aux->rt_genid != atomic_read(&rt_genid))) { 1151 if (aux == rt || rt_is_expired(aux)) {
1143 *rthp = aux->u.dst.rt_next; 1152 *rthp = aux->u.dst.rt_next;
1144 rt_free(aux); 1153 rt_free(aux);
1145 continue; 1154 continue;
@@ -1182,7 +1191,7 @@ void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1182 for (i = 0; i < 2; i++) { 1191 for (i = 0; i < 2; i++) {
1183 for (k = 0; k < 2; k++) { 1192 for (k = 0; k < 2; k++) {
1184 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k], 1193 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1185 atomic_read(&rt_genid)); 1194 rt_genid(net));
1186 1195
1187 rthp=&rt_hash_table[hash].chain; 1196 rthp=&rt_hash_table[hash].chain;
1188 1197
@@ -1194,7 +1203,7 @@ void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1194 rth->fl.fl4_src != skeys[i] || 1203 rth->fl.fl4_src != skeys[i] ||
1195 rth->fl.oif != ikeys[k] || 1204 rth->fl.oif != ikeys[k] ||
1196 rth->fl.iif != 0 || 1205 rth->fl.iif != 0 ||
1197 rth->rt_genid != atomic_read(&rt_genid) || 1206 rt_is_expired(rth) ||
1198 !net_eq(dev_net(rth->u.dst.dev), net)) { 1207 !net_eq(dev_net(rth->u.dst.dev), net)) {
1199 rthp = &rth->u.dst.rt_next; 1208 rthp = &rth->u.dst.rt_next;
1200 continue; 1209 continue;
@@ -1233,7 +1242,7 @@ void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1233 rt->u.dst.neighbour = NULL; 1242 rt->u.dst.neighbour = NULL;
1234 rt->u.dst.hh = NULL; 1243 rt->u.dst.hh = NULL;
1235 rt->u.dst.xfrm = NULL; 1244 rt->u.dst.xfrm = NULL;
1236 rt->rt_genid = atomic_read(&rt_genid); 1245 rt->rt_genid = rt_genid(net);
1237 rt->rt_flags |= RTCF_REDIRECTED; 1246 rt->rt_flags |= RTCF_REDIRECTED;
1238 1247
1239 /* Gateway is different ... */ 1248 /* Gateway is different ... */
@@ -1298,7 +1307,7 @@ static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1298 rt->u.dst.expires) { 1307 rt->u.dst.expires) {
1299 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src, 1308 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1300 rt->fl.oif, 1309 rt->fl.oif,
1301 atomic_read(&rt_genid)); 1310 rt_genid(dev_net(dst->dev)));
1302#if RT_CACHE_DEBUG >= 1 1311#if RT_CACHE_DEBUG >= 1
1303 printk(KERN_DEBUG "ipv4_negative_advice: redirect to " 1312 printk(KERN_DEBUG "ipv4_negative_advice: redirect to "
1304 NIPQUAD_FMT "/%02x dropped\n", 1313 NIPQUAD_FMT "/%02x dropped\n",
@@ -1448,7 +1457,7 @@ unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1448 for (k = 0; k < 2; k++) { 1457 for (k = 0; k < 2; k++) {
1449 for (i = 0; i < 2; i++) { 1458 for (i = 0; i < 2; i++) {
1450 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k], 1459 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1451 atomic_read(&rt_genid)); 1460 rt_genid(net));
1452 1461
1453 rcu_read_lock(); 1462 rcu_read_lock();
1454 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth; 1463 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
@@ -1463,7 +1472,7 @@ unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1463 rth->fl.iif != 0 || 1472 rth->fl.iif != 0 ||
1464 dst_metric_locked(&rth->u.dst, RTAX_MTU) || 1473 dst_metric_locked(&rth->u.dst, RTAX_MTU) ||
1465 !net_eq(dev_net(rth->u.dst.dev), net) || 1474 !net_eq(dev_net(rth->u.dst.dev), net) ||
1466 rth->rt_genid != atomic_read(&rt_genid)) 1475 !rt_is_expired(rth))
1467 continue; 1476 continue;
1468 1477
1469 if (new_mtu < 68 || new_mtu >= old_mtu) { 1478 if (new_mtu < 68 || new_mtu >= old_mtu) {
@@ -1698,7 +1707,7 @@ static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1698 rth->fl.oif = 0; 1707 rth->fl.oif = 0;
1699 rth->rt_gateway = daddr; 1708 rth->rt_gateway = daddr;
1700 rth->rt_spec_dst= spec_dst; 1709 rth->rt_spec_dst= spec_dst;
1701 rth->rt_genid = atomic_read(&rt_genid); 1710 rth->rt_genid = rt_genid(dev_net(dev));
1702 rth->rt_flags = RTCF_MULTICAST; 1711 rth->rt_flags = RTCF_MULTICAST;
1703 rth->rt_type = RTN_MULTICAST; 1712 rth->rt_type = RTN_MULTICAST;
1704 if (our) { 1713 if (our) {
@@ -1713,7 +1722,7 @@ static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1713 RT_CACHE_STAT_INC(in_slow_mc); 1722 RT_CACHE_STAT_INC(in_slow_mc);
1714 1723
1715 in_dev_put(in_dev); 1724 in_dev_put(in_dev);
1716 hash = rt_hash(daddr, saddr, dev->ifindex, atomic_read(&rt_genid)); 1725 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1717 return rt_intern_hash(hash, rth, &skb->rtable); 1726 return rt_intern_hash(hash, rth, &skb->rtable);
1718 1727
1719e_nobufs: 1728e_nobufs:
@@ -1839,7 +1848,7 @@ static int __mkroute_input(struct sk_buff *skb,
1839 1848
1840 rth->u.dst.input = ip_forward; 1849 rth->u.dst.input = ip_forward;
1841 rth->u.dst.output = ip_output; 1850 rth->u.dst.output = ip_output;
1842 rth->rt_genid = atomic_read(&rt_genid); 1851 rth->rt_genid = rt_genid(dev_net(rth->u.dst.dev));
1843 1852
1844 rt_set_nexthop(rth, res, itag); 1853 rt_set_nexthop(rth, res, itag);
1845 1854
@@ -1874,7 +1883,8 @@ static int ip_mkroute_input(struct sk_buff *skb,
1874 return err; 1883 return err;
1875 1884
1876 /* put it into the cache */ 1885 /* put it into the cache */
1877 hash = rt_hash(daddr, saddr, fl->iif, atomic_read(&rt_genid)); 1886 hash = rt_hash(daddr, saddr, fl->iif,
1887 rt_genid(dev_net(rth->u.dst.dev)));
1878 return rt_intern_hash(hash, rth, &skb->rtable); 1888 return rt_intern_hash(hash, rth, &skb->rtable);
1879} 1889}
1880 1890
@@ -2000,7 +2010,7 @@ local_input:
2000 goto e_nobufs; 2010 goto e_nobufs;
2001 2011
2002 rth->u.dst.output= ip_rt_bug; 2012 rth->u.dst.output= ip_rt_bug;
2003 rth->rt_genid = atomic_read(&rt_genid); 2013 rth->rt_genid = rt_genid(net);
2004 2014
2005 atomic_set(&rth->u.dst.__refcnt, 1); 2015 atomic_set(&rth->u.dst.__refcnt, 1);
2006 rth->u.dst.flags= DST_HOST; 2016 rth->u.dst.flags= DST_HOST;
@@ -2030,7 +2040,7 @@ local_input:
2030 rth->rt_flags &= ~RTCF_LOCAL; 2040 rth->rt_flags &= ~RTCF_LOCAL;
2031 } 2041 }
2032 rth->rt_type = res.type; 2042 rth->rt_type = res.type;
2033 hash = rt_hash(daddr, saddr, fl.iif, atomic_read(&rt_genid)); 2043 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2034 err = rt_intern_hash(hash, rth, &skb->rtable); 2044 err = rt_intern_hash(hash, rth, &skb->rtable);
2035 goto done; 2045 goto done;
2036 2046
@@ -2081,7 +2091,7 @@ int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2081 2091
2082 net = dev_net(dev); 2092 net = dev_net(dev);
2083 tos &= IPTOS_RT_MASK; 2093 tos &= IPTOS_RT_MASK;
2084 hash = rt_hash(daddr, saddr, iif, atomic_read(&rt_genid)); 2094 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2085 2095
2086 rcu_read_lock(); 2096 rcu_read_lock();
2087 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth; 2097 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
@@ -2093,7 +2103,7 @@ int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2093 (rth->fl.fl4_tos ^ tos)) == 0 && 2103 (rth->fl.fl4_tos ^ tos)) == 0 &&
2094 rth->fl.mark == skb->mark && 2104 rth->fl.mark == skb->mark &&
2095 net_eq(dev_net(rth->u.dst.dev), net) && 2105 net_eq(dev_net(rth->u.dst.dev), net) &&
2096 rth->rt_genid == atomic_read(&rt_genid)) { 2106 !rt_is_expired(rth)) {
2097 dst_use(&rth->u.dst, jiffies); 2107 dst_use(&rth->u.dst, jiffies);
2098 RT_CACHE_STAT_INC(in_hit); 2108 RT_CACHE_STAT_INC(in_hit);
2099 rcu_read_unlock(); 2109 rcu_read_unlock();
@@ -2221,7 +2231,7 @@ static int __mkroute_output(struct rtable **result,
2221 rth->rt_spec_dst= fl->fl4_src; 2231 rth->rt_spec_dst= fl->fl4_src;
2222 2232
2223 rth->u.dst.output=ip_output; 2233 rth->u.dst.output=ip_output;
2224 rth->rt_genid = atomic_read(&rt_genid); 2234 rth->rt_genid = rt_genid(dev_net(dev_out));
2225 2235
2226 RT_CACHE_STAT_INC(out_slow_tot); 2236 RT_CACHE_STAT_INC(out_slow_tot);
2227 2237
@@ -2271,7 +2281,7 @@ static int ip_mkroute_output(struct rtable **rp,
2271 unsigned hash; 2281 unsigned hash;
2272 if (err == 0) { 2282 if (err == 0) {
2273 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif, 2283 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2274 atomic_read(&rt_genid)); 2284 rt_genid(dev_net(dev_out)));
2275 err = rt_intern_hash(hash, rth, rp); 2285 err = rt_intern_hash(hash, rth, rp);
2276 } 2286 }
2277 2287
@@ -2483,8 +2493,7 @@ int __ip_route_output_key(struct net *net, struct rtable **rp,
2483 unsigned hash; 2493 unsigned hash;
2484 struct rtable *rth; 2494 struct rtable *rth;
2485 2495
2486 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, 2496 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2487 atomic_read(&rt_genid));
2488 2497
2489 rcu_read_lock_bh(); 2498 rcu_read_lock_bh();
2490 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth; 2499 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
@@ -2497,7 +2506,7 @@ int __ip_route_output_key(struct net *net, struct rtable **rp,
2497 !((rth->fl.fl4_tos ^ flp->fl4_tos) & 2506 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2498 (IPTOS_RT_MASK | RTO_ONLINK)) && 2507 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2499 net_eq(dev_net(rth->u.dst.dev), net) && 2508 net_eq(dev_net(rth->u.dst.dev), net) &&
2500 rth->rt_genid == atomic_read(&rt_genid)) { 2509 !rt_is_expired(rth)) {
2501 dst_use(&rth->u.dst, jiffies); 2510 dst_use(&rth->u.dst, jiffies);
2502 RT_CACHE_STAT_INC(out_hit); 2511 RT_CACHE_STAT_INC(out_hit);
2503 rcu_read_unlock_bh(); 2512 rcu_read_unlock_bh();
@@ -2528,7 +2537,7 @@ static struct dst_ops ipv4_dst_blackhole_ops = {
2528}; 2537};
2529 2538
2530 2539
2531static int ipv4_dst_blackhole(struct rtable **rp, struct flowi *flp) 2540static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2532{ 2541{
2533 struct rtable *ort = *rp; 2542 struct rtable *ort = *rp;
2534 struct rtable *rt = (struct rtable *) 2543 struct rtable *rt = (struct rtable *)
@@ -2552,7 +2561,7 @@ static int ipv4_dst_blackhole(struct rtable **rp, struct flowi *flp)
2552 rt->idev = ort->idev; 2561 rt->idev = ort->idev;
2553 if (rt->idev) 2562 if (rt->idev)
2554 in_dev_hold(rt->idev); 2563 in_dev_hold(rt->idev);
2555 rt->rt_genid = atomic_read(&rt_genid); 2564 rt->rt_genid = rt_genid(net);
2556 rt->rt_flags = ort->rt_flags; 2565 rt->rt_flags = ort->rt_flags;
2557 rt->rt_type = ort->rt_type; 2566 rt->rt_type = ort->rt_type;
2558 rt->rt_dst = ort->rt_dst; 2567 rt->rt_dst = ort->rt_dst;
@@ -2588,7 +2597,7 @@ int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2588 err = __xfrm_lookup((struct dst_entry **)rp, flp, sk, 2597 err = __xfrm_lookup((struct dst_entry **)rp, flp, sk,
2589 flags ? XFRM_LOOKUP_WAIT : 0); 2598 flags ? XFRM_LOOKUP_WAIT : 0);
2590 if (err == -EREMOTE) 2599 if (err == -EREMOTE)
2591 err = ipv4_dst_blackhole(rp, flp); 2600 err = ipv4_dst_blackhole(net, rp, flp);
2592 2601
2593 return err; 2602 return err;
2594 } 2603 }
@@ -2807,7 +2816,7 @@ int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2807 rt = rcu_dereference(rt->u.dst.rt_next), idx++) { 2816 rt = rcu_dereference(rt->u.dst.rt_next), idx++) {
2808 if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx) 2817 if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx)
2809 continue; 2818 continue;
2810 if (rt->rt_genid != atomic_read(&rt_genid)) 2819 if (rt_is_expired(rt))
2811 continue; 2820 continue;
2812 skb->dst = dst_clone(&rt->u.dst); 2821 skb->dst = dst_clone(&rt->u.dst);
2813 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid, 2822 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
@@ -3081,6 +3090,10 @@ static __net_initdata struct pernet_operations sysctl_route_ops = {
3081 3090
3082static __net_init int rt_secret_timer_init(struct net *net) 3091static __net_init int rt_secret_timer_init(struct net *net)
3083{ 3092{
3093 atomic_set(&net->ipv4.rt_genid,
3094 (int) ((num_physpages ^ (num_physpages>>8)) ^
3095 (jiffies ^ (jiffies >> 7))));
3096
3084 net->ipv4.rt_secret_timer.function = rt_secret_rebuild; 3097 net->ipv4.rt_secret_timer.function = rt_secret_rebuild;
3085 net->ipv4.rt_secret_timer.data = (unsigned long)net; 3098 net->ipv4.rt_secret_timer.data = (unsigned long)net;
3086 init_timer_deferrable(&net->ipv4.rt_secret_timer); 3099 init_timer_deferrable(&net->ipv4.rt_secret_timer);
@@ -3121,9 +3134,6 @@ int __init ip_rt_init(void)
3121{ 3134{
3122 int rc = 0; 3135 int rc = 0;
3123 3136
3124 atomic_set(&rt_genid, (int) ((num_physpages ^ (num_physpages>>8)) ^
3125 (jiffies ^ (jiffies >> 7))));
3126
3127#ifdef CONFIG_NET_CLS_ROUTE 3137#ifdef CONFIG_NET_CLS_ROUTE
3128 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct)); 3138 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct));
3129 if (!ip_rt_acct) 3139 if (!ip_rt_acct)