2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #define pr_fmt(fmt) "IPv4: " fmt
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/bootmem.h>
74 #include <linux/string.h>
75 #include <linux/socket.h>
76 #include <linux/sockios.h>
77 #include <linux/errno.h>
79 #include <linux/inet.h>
80 #include <linux/netdevice.h>
81 #include <linux/proc_fs.h>
82 #include <linux/init.h>
83 #include <linux/workqueue.h>
84 #include <linux/skbuff.h>
85 #include <linux/inetdevice.h>
86 #include <linux/igmp.h>
87 #include <linux/pkt_sched.h>
88 #include <linux/mroute.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/jhash.h>
92 #include <linux/rcupdate.h>
93 #include <linux/times.h>
94 #include <linux/slab.h>
95 #include <linux/prefetch.h>
97 #include <net/net_namespace.h>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/inetpeer.h>
102 #include <net/sock.h>
103 #include <net/ip_fib.h>
106 #include <net/icmp.h>
107 #include <net/xfrm.h>
108 #include <net/netevent.h>
109 #include <net/rtnetlink.h>
111 #include <linux/sysctl.h>
112 #include <linux/kmemleak.h>
114 #include <net/secure_seq.h>
116 #define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
119 #define IP_MAX_MTU 0xFFF0
121 #define RT_GC_TIMEOUT (300*HZ)
123 static int ip_rt_max_size;
124 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
126 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
127 static int ip_rt_redirect_number __read_mostly = 9;
128 static int ip_rt_redirect_load __read_mostly = HZ / 50;
129 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
130 static int ip_rt_error_cost __read_mostly = HZ;
131 static int ip_rt_error_burst __read_mostly = 5 * HZ;
132 static int ip_rt_gc_elasticity __read_mostly = 8;
133 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
134 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
135 static int ip_rt_min_advmss __read_mostly = 256;
136 static int rt_chain_length_max __read_mostly = 20;
138 static struct delayed_work expires_work;
139 static unsigned long expires_ljiffies;
142 * Interface to generic destination cache.
145 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
146 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
147 static unsigned int ipv4_mtu(const struct dst_entry *dst);
148 static void ipv4_dst_destroy(struct dst_entry *dst);
149 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
150 static void ipv4_link_failure(struct sk_buff *skb);
151 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
152 static int rt_garbage_collect(struct dst_ops *ops);
154 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
159 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
161 struct rtable *rt = (struct rtable *) dst;
162 struct inet_peer *peer;
165 peer = rt_get_peer_create(rt, rt->rt_dst);
167 u32 *old_p = __DST_METRICS_PTR(old);
168 unsigned long prev, new;
171 if (inet_metrics_new(peer))
172 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
174 new = (unsigned long) p;
175 prev = cmpxchg(&dst->_metrics, old, new);
178 p = __DST_METRICS_PTR(prev);
179 if (prev & DST_METRICS_READ_ONLY)
183 fib_info_put(rt->fi);
191 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
195 static struct dst_ops ipv4_dst_ops = {
197 .protocol = cpu_to_be16(ETH_P_IP),
198 .gc = rt_garbage_collect,
199 .check = ipv4_dst_check,
200 .default_advmss = ipv4_default_advmss,
202 .cow_metrics = ipv4_cow_metrics,
203 .destroy = ipv4_dst_destroy,
204 .ifdown = ipv4_dst_ifdown,
205 .negative_advice = ipv4_negative_advice,
206 .link_failure = ipv4_link_failure,
207 .update_pmtu = ip_rt_update_pmtu,
208 .local_out = __ip_local_out,
209 .neigh_lookup = ipv4_neigh_lookup,
212 #define ECN_OR_COST(class) TC_PRIO_##class
214 const __u8 ip_tos2prio[16] = {
216 ECN_OR_COST(BESTEFFORT),
218 ECN_OR_COST(BESTEFFORT),
224 ECN_OR_COST(INTERACTIVE),
226 ECN_OR_COST(INTERACTIVE),
227 TC_PRIO_INTERACTIVE_BULK,
228 ECN_OR_COST(INTERACTIVE_BULK),
229 TC_PRIO_INTERACTIVE_BULK,
230 ECN_OR_COST(INTERACTIVE_BULK)
232 EXPORT_SYMBOL(ip_tos2prio);
238 /* The locking scheme is rather straight forward:
240 * 1) Read-Copy Update protects the buckets of the central route hash.
241 * 2) Only writers remove entries, and they hold the lock
242 * as they look at rtable reference counts.
243 * 3) Only readers acquire references to rtable entries,
244 * they do so with atomic increments and with the
248 struct rt_hash_bucket {
249 struct rtable __rcu *chain;
252 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
253 defined(CONFIG_PROVE_LOCKING)
255 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
256 * The size of this table is a power of two and depends on the number of CPUS.
257 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
259 #ifdef CONFIG_LOCKDEP
260 # define RT_HASH_LOCK_SZ 256
263 # define RT_HASH_LOCK_SZ 4096
265 # define RT_HASH_LOCK_SZ 2048
267 # define RT_HASH_LOCK_SZ 1024
269 # define RT_HASH_LOCK_SZ 512
271 # define RT_HASH_LOCK_SZ 256
275 static spinlock_t *rt_hash_locks;
276 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
278 static __init void rt_hash_lock_init(void)
282 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
285 panic("IP: failed to allocate rt_hash_locks\n");
287 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
288 spin_lock_init(&rt_hash_locks[i]);
291 # define rt_hash_lock_addr(slot) NULL
293 static inline void rt_hash_lock_init(void)
298 static struct rt_hash_bucket *rt_hash_table __read_mostly;
299 static unsigned int rt_hash_mask __read_mostly;
300 static unsigned int rt_hash_log __read_mostly;
302 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
303 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
305 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
308 return jhash_3words((__force u32)daddr, (__force u32)saddr,
313 static inline int rt_genid(struct net *net)
315 return atomic_read(&net->ipv4.rt_genid);
318 #ifdef CONFIG_PROC_FS
319 struct rt_cache_iter_state {
320 struct seq_net_private p;
325 static struct rtable *rt_cache_get_first(struct seq_file *seq)
327 struct rt_cache_iter_state *st = seq->private;
328 struct rtable *r = NULL;
330 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
331 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
334 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
336 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
337 r->rt_genid == st->genid)
339 r = rcu_dereference_bh(r->dst.rt_next);
341 rcu_read_unlock_bh();
346 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
349 struct rt_cache_iter_state *st = seq->private;
351 r = rcu_dereference_bh(r->dst.rt_next);
353 rcu_read_unlock_bh();
355 if (--st->bucket < 0)
357 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
359 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
364 static struct rtable *rt_cache_get_next(struct seq_file *seq,
367 struct rt_cache_iter_state *st = seq->private;
368 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
369 if (dev_net(r->dst.dev) != seq_file_net(seq))
371 if (r->rt_genid == st->genid)
377 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
379 struct rtable *r = rt_cache_get_first(seq);
382 while (pos && (r = rt_cache_get_next(seq, r)))
384 return pos ? NULL : r;
387 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
389 struct rt_cache_iter_state *st = seq->private;
391 return rt_cache_get_idx(seq, *pos - 1);
392 st->genid = rt_genid(seq_file_net(seq));
393 return SEQ_START_TOKEN;
396 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
400 if (v == SEQ_START_TOKEN)
401 r = rt_cache_get_first(seq);
403 r = rt_cache_get_next(seq, v);
408 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
410 if (v && v != SEQ_START_TOKEN)
411 rcu_read_unlock_bh();
414 static int rt_cache_seq_show(struct seq_file *seq, void *v)
416 if (v == SEQ_START_TOKEN)
417 seq_printf(seq, "%-127s\n",
418 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
419 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
422 struct rtable *r = v;
425 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
426 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
427 r->dst.dev ? r->dst.dev->name : "*",
428 (__force u32)r->rt_dst,
429 (__force u32)r->rt_gateway,
430 r->rt_flags, atomic_read(&r->dst.__refcnt),
431 r->dst.__use, 0, (__force u32)r->rt_src,
432 dst_metric_advmss(&r->dst) + 40,
433 dst_metric(&r->dst, RTAX_WINDOW),
434 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
435 dst_metric(&r->dst, RTAX_RTTVAR)),
439 seq_printf(seq, "%*s\n", 127 - len, "");
444 static const struct seq_operations rt_cache_seq_ops = {
445 .start = rt_cache_seq_start,
446 .next = rt_cache_seq_next,
447 .stop = rt_cache_seq_stop,
448 .show = rt_cache_seq_show,
451 static int rt_cache_seq_open(struct inode *inode, struct file *file)
453 return seq_open_net(inode, file, &rt_cache_seq_ops,
454 sizeof(struct rt_cache_iter_state));
457 static const struct file_operations rt_cache_seq_fops = {
458 .owner = THIS_MODULE,
459 .open = rt_cache_seq_open,
462 .release = seq_release_net,
466 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
471 return SEQ_START_TOKEN;
473 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
474 if (!cpu_possible(cpu))
477 return &per_cpu(rt_cache_stat, cpu);
482 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
486 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
487 if (!cpu_possible(cpu))
490 return &per_cpu(rt_cache_stat, cpu);
496 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
501 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
503 struct rt_cache_stat *st = v;
505 if (v == SEQ_START_TOKEN) {
506 seq_printf(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
510 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
511 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
512 dst_entries_get_slow(&ipv4_dst_ops),
535 static const struct seq_operations rt_cpu_seq_ops = {
536 .start = rt_cpu_seq_start,
537 .next = rt_cpu_seq_next,
538 .stop = rt_cpu_seq_stop,
539 .show = rt_cpu_seq_show,
543 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
545 return seq_open(file, &rt_cpu_seq_ops);
548 static const struct file_operations rt_cpu_seq_fops = {
549 .owner = THIS_MODULE,
550 .open = rt_cpu_seq_open,
553 .release = seq_release,
556 #ifdef CONFIG_IP_ROUTE_CLASSID
557 static int rt_acct_proc_show(struct seq_file *m, void *v)
559 struct ip_rt_acct *dst, *src;
562 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
566 for_each_possible_cpu(i) {
567 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
568 for (j = 0; j < 256; j++) {
569 dst[j].o_bytes += src[j].o_bytes;
570 dst[j].o_packets += src[j].o_packets;
571 dst[j].i_bytes += src[j].i_bytes;
572 dst[j].i_packets += src[j].i_packets;
576 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
581 static int rt_acct_proc_open(struct inode *inode, struct file *file)
583 return single_open(file, rt_acct_proc_show, NULL);
586 static const struct file_operations rt_acct_proc_fops = {
587 .owner = THIS_MODULE,
588 .open = rt_acct_proc_open,
591 .release = single_release,
595 static int __net_init ip_rt_do_proc_init(struct net *net)
597 struct proc_dir_entry *pde;
599 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
604 pde = proc_create("rt_cache", S_IRUGO,
605 net->proc_net_stat, &rt_cpu_seq_fops);
609 #ifdef CONFIG_IP_ROUTE_CLASSID
610 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
616 #ifdef CONFIG_IP_ROUTE_CLASSID
618 remove_proc_entry("rt_cache", net->proc_net_stat);
621 remove_proc_entry("rt_cache", net->proc_net);
626 static void __net_exit ip_rt_do_proc_exit(struct net *net)
628 remove_proc_entry("rt_cache", net->proc_net_stat);
629 remove_proc_entry("rt_cache", net->proc_net);
630 #ifdef CONFIG_IP_ROUTE_CLASSID
631 remove_proc_entry("rt_acct", net->proc_net);
635 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
636 .init = ip_rt_do_proc_init,
637 .exit = ip_rt_do_proc_exit,
640 static int __init ip_rt_proc_init(void)
642 return register_pernet_subsys(&ip_rt_proc_ops);
646 static inline int ip_rt_proc_init(void)
650 #endif /* CONFIG_PROC_FS */
652 static inline void rt_free(struct rtable *rt)
654 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
657 static inline void rt_drop(struct rtable *rt)
660 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
663 static inline int rt_fast_clean(struct rtable *rth)
665 /* Kill broadcast/multicast entries very aggresively, if they
666 collide in hash table with more useful entries */
667 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
668 rt_is_input_route(rth) && rth->dst.rt_next;
671 static inline int rt_valuable(struct rtable *rth)
673 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
674 (rt_has_peer(rth) && rt_peer_ptr(rth)->pmtu_expires);
677 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
682 if (atomic_read(&rth->dst.__refcnt))
685 age = jiffies - rth->dst.lastuse;
686 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
687 (age <= tmo2 && rt_valuable(rth)))
693 /* Bits of score are:
695 * 30: not quite useless
696 * 29..0: usage counter
698 static inline u32 rt_score(struct rtable *rt)
700 u32 score = jiffies - rt->dst.lastuse;
702 score = ~score & ~(3<<30);
707 if (rt_is_output_route(rt) ||
708 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
714 static inline bool rt_caching(const struct net *net)
716 return net->ipv4.current_rt_cache_rebuild_count <=
717 net->ipv4.sysctl_rt_cache_rebuild_count;
720 static inline bool compare_hash_inputs(const struct rtable *rt1,
721 const struct rtable *rt2)
723 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
724 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
725 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
728 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
730 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
731 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
732 (rt1->rt_mark ^ rt2->rt_mark) |
733 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
734 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
735 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
738 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
740 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
743 static inline int rt_is_expired(struct rtable *rth)
745 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
749 * Perform a full scan of hash table and free all entries.
750 * Can be called by a softirq or a process.
751 * In the later case, we want to be reschedule if necessary
753 static void rt_do_flush(struct net *net, int process_context)
756 struct rtable *rth, *next;
758 for (i = 0; i <= rt_hash_mask; i++) {
759 struct rtable __rcu **pprev;
762 if (process_context && need_resched())
764 rth = rcu_access_pointer(rt_hash_table[i].chain);
768 spin_lock_bh(rt_hash_lock_addr(i));
771 pprev = &rt_hash_table[i].chain;
772 rth = rcu_dereference_protected(*pprev,
773 lockdep_is_held(rt_hash_lock_addr(i)));
776 next = rcu_dereference_protected(rth->dst.rt_next,
777 lockdep_is_held(rt_hash_lock_addr(i)));
780 net_eq(dev_net(rth->dst.dev), net)) {
781 rcu_assign_pointer(*pprev, next);
782 rcu_assign_pointer(rth->dst.rt_next, list);
785 pprev = &rth->dst.rt_next;
790 spin_unlock_bh(rt_hash_lock_addr(i));
792 for (; list; list = next) {
793 next = rcu_dereference_protected(list->dst.rt_next, 1);
800 * While freeing expired entries, we compute average chain length
801 * and standard deviation, using fixed-point arithmetic.
802 * This to have an estimation of rt_chain_length_max
803 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
804 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
808 #define ONE (1UL << FRACT_BITS)
811 * Given a hash chain and an item in this hash chain,
812 * find if a previous entry has the same hash_inputs
813 * (but differs on tos, mark or oif)
814 * Returns 0 if an alias is found.
815 * Returns ONE if rth has no alias before itself.
817 static int has_noalias(const struct rtable *head, const struct rtable *rth)
819 const struct rtable *aux = head;
822 if (compare_hash_inputs(aux, rth))
824 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
829 static void rt_check_expire(void)
831 static unsigned int rover;
832 unsigned int i = rover, goal;
834 struct rtable __rcu **rthp;
835 unsigned long samples = 0;
836 unsigned long sum = 0, sum2 = 0;
840 delta = jiffies - expires_ljiffies;
841 expires_ljiffies = jiffies;
842 mult = ((u64)delta) << rt_hash_log;
843 if (ip_rt_gc_timeout > 1)
844 do_div(mult, ip_rt_gc_timeout);
845 goal = (unsigned int)mult;
846 if (goal > rt_hash_mask)
847 goal = rt_hash_mask + 1;
848 for (; goal > 0; goal--) {
849 unsigned long tmo = ip_rt_gc_timeout;
850 unsigned long length;
852 i = (i + 1) & rt_hash_mask;
853 rthp = &rt_hash_table[i].chain;
860 if (rcu_dereference_raw(*rthp) == NULL)
863 spin_lock_bh(rt_hash_lock_addr(i));
864 while ((rth = rcu_dereference_protected(*rthp,
865 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
866 prefetch(rth->dst.rt_next);
867 if (rt_is_expired(rth) ||
868 rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
869 *rthp = rth->dst.rt_next;
874 /* We only count entries on a chain with equal
875 * hash inputs once so that entries for
876 * different QOS levels, and other non-hash
877 * input attributes don't unfairly skew the
881 rthp = &rth->dst.rt_next;
882 length += has_noalias(rt_hash_table[i].chain, rth);
884 spin_unlock_bh(rt_hash_lock_addr(i));
886 sum2 += length*length;
889 unsigned long avg = sum / samples;
890 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
891 rt_chain_length_max = max_t(unsigned long,
893 (avg + 4*sd) >> FRACT_BITS);
899 * rt_worker_func() is run in process context.
900 * we call rt_check_expire() to scan part of the hash table
902 static void rt_worker_func(struct work_struct *work)
905 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
909 * Perturbation of rt_genid by a small quantity [1..256]
910 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
911 * many times (2^24) without giving recent rt_genid.
912 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
914 static void rt_cache_invalidate(struct net *net)
916 unsigned char shuffle;
918 get_random_bytes(&shuffle, sizeof(shuffle));
919 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
920 inetpeer_invalidate_family(AF_INET);
924 * delay < 0 : invalidate cache (fast : entries will be deleted later)
925 * delay >= 0 : invalidate & flush cache (can be long)
927 void rt_cache_flush(struct net *net, int delay)
929 rt_cache_invalidate(net);
931 rt_do_flush(net, !in_softirq());
934 /* Flush previous cache invalidated entries from the cache */
935 void rt_cache_flush_batch(struct net *net)
937 rt_do_flush(net, !in_softirq());
940 static void rt_emergency_hash_rebuild(struct net *net)
942 net_warn_ratelimited("Route hash chain too long!\n");
943 rt_cache_invalidate(net);
947 Short description of GC goals.
949 We want to build algorithm, which will keep routing cache
950 at some equilibrium point, when number of aged off entries
951 is kept approximately equal to newly generated ones.
953 Current expiration strength is variable "expire".
954 We try to adjust it dynamically, so that if networking
955 is idle expires is large enough to keep enough of warm entries,
956 and when load increases it reduces to limit cache size.
959 static int rt_garbage_collect(struct dst_ops *ops)
961 static unsigned long expire = RT_GC_TIMEOUT;
962 static unsigned long last_gc;
964 static int equilibrium;
966 struct rtable __rcu **rthp;
967 unsigned long now = jiffies;
969 int entries = dst_entries_get_fast(&ipv4_dst_ops);
972 * Garbage collection is pretty expensive,
973 * do not make it too frequently.
976 RT_CACHE_STAT_INC(gc_total);
978 if (now - last_gc < ip_rt_gc_min_interval &&
979 entries < ip_rt_max_size) {
980 RT_CACHE_STAT_INC(gc_ignored);
984 entries = dst_entries_get_slow(&ipv4_dst_ops);
985 /* Calculate number of entries, which we want to expire now. */
986 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
988 if (equilibrium < ipv4_dst_ops.gc_thresh)
989 equilibrium = ipv4_dst_ops.gc_thresh;
990 goal = entries - equilibrium;
992 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
993 goal = entries - equilibrium;
996 /* We are in dangerous area. Try to reduce cache really
999 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1000 equilibrium = entries - goal;
1003 if (now - last_gc >= ip_rt_gc_min_interval)
1007 equilibrium += goal;
1014 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1015 unsigned long tmo = expire;
1017 k = (k + 1) & rt_hash_mask;
1018 rthp = &rt_hash_table[k].chain;
1019 spin_lock_bh(rt_hash_lock_addr(k));
1020 while ((rth = rcu_dereference_protected(*rthp,
1021 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1022 if (!rt_is_expired(rth) &&
1023 !rt_may_expire(rth, tmo, expire)) {
1025 rthp = &rth->dst.rt_next;
1028 *rthp = rth->dst.rt_next;
1032 spin_unlock_bh(rt_hash_lock_addr(k));
1041 /* Goal is not achieved. We stop process if:
1043 - if expire reduced to zero. Otherwise, expire is halfed.
1044 - if table is not full.
1045 - if we are called from interrupt.
1046 - jiffies check is just fallback/debug loop breaker.
1047 We will not spin here for long time in any case.
1050 RT_CACHE_STAT_INC(gc_goal_miss);
1057 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1059 } while (!in_softirq() && time_before_eq(jiffies, now));
1061 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1063 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1065 net_warn_ratelimited("dst cache overflow\n");
1066 RT_CACHE_STAT_INC(gc_dst_overflow);
1070 expire += ip_rt_gc_min_interval;
1071 if (expire > ip_rt_gc_timeout ||
1072 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1073 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1074 expire = ip_rt_gc_timeout;
1079 * Returns number of entries in a hash chain that have different hash_inputs
1081 static int slow_chain_length(const struct rtable *head)
1084 const struct rtable *rth = head;
1087 length += has_noalias(head, rth);
1088 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1090 return length >> FRACT_BITS;
1093 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
1094 struct sk_buff *skb,
1097 struct net_device *dev = dst->dev;
1098 const __be32 *pkey = daddr;
1099 const struct rtable *rt;
1100 struct neighbour *n;
1102 rt = (const struct rtable *) dst;
1104 pkey = (const __be32 *) &rt->rt_gateway;
1106 pkey = &ip_hdr(skb)->daddr;
1108 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1111 return neigh_create(&arp_tbl, pkey, dev);
1114 static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1115 struct sk_buff *skb, int ifindex)
1117 struct rtable *rth, *cand;
1118 struct rtable __rcu **rthp, **candp;
1125 min_score = ~(u32)0;
1130 if (!rt_caching(dev_net(rt->dst.dev)) || (rt->dst.flags & DST_NOCACHE)) {
1132 * If we're not caching, just tell the caller we
1133 * were successful and don't touch the route. The
1134 * caller hold the sole reference to the cache entry, and
1135 * it will be released when the caller is done with it.
1136 * If we drop it here, the callers have no way to resolve routes
1137 * when we're not caching. Instead, just point *rp at rt, so
1138 * the caller gets a single use out of the route
1139 * Note that we do rt_free on this new route entry, so that
1140 * once its refcount hits zero, we are still able to reap it
1142 * Note: To avoid expensive rcu stuff for this uncached dst,
1143 * we set DST_NOCACHE so that dst_release() can free dst without
1144 * waiting a grace period.
1147 rt->dst.flags |= DST_NOCACHE;
1151 rthp = &rt_hash_table[hash].chain;
1153 spin_lock_bh(rt_hash_lock_addr(hash));
1154 while ((rth = rcu_dereference_protected(*rthp,
1155 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1156 if (rt_is_expired(rth)) {
1157 *rthp = rth->dst.rt_next;
1161 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1163 *rthp = rth->dst.rt_next;
1165 * Since lookup is lockfree, the deletion
1166 * must be visible to another weakly ordered CPU before
1167 * the insertion at the start of the hash chain.
1169 rcu_assign_pointer(rth->dst.rt_next,
1170 rt_hash_table[hash].chain);
1172 * Since lookup is lockfree, the update writes
1173 * must be ordered for consistency on SMP.
1175 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1177 dst_use(&rth->dst, now);
1178 spin_unlock_bh(rt_hash_lock_addr(hash));
1182 skb_dst_set(skb, &rth->dst);
1186 if (!atomic_read(&rth->dst.__refcnt)) {
1187 u32 score = rt_score(rth);
1189 if (score <= min_score) {
1198 rthp = &rth->dst.rt_next;
1202 /* ip_rt_gc_elasticity used to be average length of chain
1203 * length, when exceeded gc becomes really aggressive.
1205 * The second limit is less certain. At the moment it allows
1206 * only 2 entries per bucket. We will see.
1208 if (chain_length > ip_rt_gc_elasticity) {
1209 *candp = cand->dst.rt_next;
1213 if (chain_length > rt_chain_length_max &&
1214 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1215 struct net *net = dev_net(rt->dst.dev);
1216 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1217 if (!rt_caching(net)) {
1218 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1219 rt->dst.dev->name, num);
1221 rt_emergency_hash_rebuild(net);
1222 spin_unlock_bh(rt_hash_lock_addr(hash));
1224 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1225 ifindex, rt_genid(net));
1230 rt->dst.rt_next = rt_hash_table[hash].chain;
1233 * Since lookup is lockfree, we must make sure
1234 * previous writes to rt are committed to memory
1235 * before making rt visible to other CPUS.
1237 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1239 spin_unlock_bh(rt_hash_lock_addr(hash));
1243 skb_dst_set(skb, &rt->dst);
1247 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1249 static u32 rt_peer_genid(void)
1251 return atomic_read(&__rt_peer_genid);
1254 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1256 struct inet_peer_base *base;
1257 struct inet_peer *peer;
1259 base = inetpeer_base_ptr(rt->_peer);
1263 peer = inet_getpeer_v4(base, daddr, create);
1265 if (!rt_set_peer(rt, peer))
1268 rt->rt_peer_genid = rt_peer_genid();
1273 * Peer allocation may fail only in serious out-of-memory conditions. However
1274 * we still can generate some output.
1275 * Random ID selection looks a bit dangerous because we have no chances to
1276 * select ID being unique in a reasonable period of time.
1277 * But broken packet identifier may be better than no packet at all.
1279 static void ip_select_fb_ident(struct iphdr *iph)
1281 static DEFINE_SPINLOCK(ip_fb_id_lock);
1282 static u32 ip_fallback_id;
1285 spin_lock_bh(&ip_fb_id_lock);
1286 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1287 iph->id = htons(salt & 0xFFFF);
1288 ip_fallback_id = salt;
1289 spin_unlock_bh(&ip_fb_id_lock);
1292 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1294 struct rtable *rt = (struct rtable *) dst;
1296 if (rt && !(rt->dst.flags & DST_NOPEER)) {
1297 struct inet_peer *peer = rt_get_peer_create(rt, rt->rt_dst);
1299 /* If peer is attached to destination, it is never detached,
1300 so that we need not to grab a lock to dereference it.
1303 iph->id = htons(inet_getid(peer, more));
1307 pr_debug("rt_bind_peer(0) @%p\n", __builtin_return_address(0));
1309 ip_select_fb_ident(iph);
1311 EXPORT_SYMBOL(__ip_select_ident);
1313 static void rt_del(unsigned int hash, struct rtable *rt)
1315 struct rtable __rcu **rthp;
1318 rthp = &rt_hash_table[hash].chain;
1319 spin_lock_bh(rt_hash_lock_addr(hash));
1321 while ((aux = rcu_dereference_protected(*rthp,
1322 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1323 if (aux == rt || rt_is_expired(aux)) {
1324 *rthp = aux->dst.rt_next;
1328 rthp = &aux->dst.rt_next;
1330 spin_unlock_bh(rt_hash_lock_addr(hash));
1333 static void check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1335 struct rtable *rt = (struct rtable *) dst;
1336 __be32 orig_gw = rt->rt_gateway;
1337 struct neighbour *n;
1339 dst_confirm(&rt->dst);
1341 rt->rt_gateway = peer->redirect_learned.a4;
1343 n = ipv4_neigh_lookup(&rt->dst, NULL, &rt->rt_gateway);
1345 rt->rt_gateway = orig_gw;
1348 if (!(n->nud_state & NUD_VALID)) {
1349 neigh_event_send(n, NULL);
1351 rt->rt_flags |= RTCF_REDIRECTED;
1352 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1357 /* called in rcu_read_lock() section */
1358 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1359 __be32 saddr, struct net_device *dev)
1362 struct in_device *in_dev = __in_dev_get_rcu(dev);
1363 __be32 skeys[2] = { saddr, 0 };
1364 int ikeys[2] = { dev->ifindex, 0 };
1365 struct inet_peer *peer;
1372 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1373 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1374 ipv4_is_zeronet(new_gw))
1375 goto reject_redirect;
1377 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1378 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1379 goto reject_redirect;
1380 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1381 goto reject_redirect;
1383 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1384 goto reject_redirect;
1387 for (s = 0; s < 2; s++) {
1388 for (i = 0; i < 2; i++) {
1390 struct rtable __rcu **rthp;
1393 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1395 rthp = &rt_hash_table[hash].chain;
1397 while ((rt = rcu_dereference(*rthp)) != NULL) {
1398 rthp = &rt->dst.rt_next;
1400 if (rt->rt_key_dst != daddr ||
1401 rt->rt_key_src != skeys[s] ||
1402 rt->rt_oif != ikeys[i] ||
1403 rt_is_input_route(rt) ||
1404 rt_is_expired(rt) ||
1405 !net_eq(dev_net(rt->dst.dev), net) ||
1407 rt->dst.dev != dev ||
1408 rt->rt_gateway != old_gw)
1411 peer = rt_get_peer_create(rt, rt->rt_dst);
1413 if (peer->redirect_learned.a4 != new_gw) {
1414 peer->redirect_learned.a4 = new_gw;
1415 atomic_inc(&__rt_peer_genid);
1417 check_peer_redir(&rt->dst, peer);
1425 #ifdef CONFIG_IP_ROUTE_VERBOSE
1426 if (IN_DEV_LOG_MARTIANS(in_dev))
1427 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1428 " Advised path = %pI4 -> %pI4\n",
1429 &old_gw, dev->name, &new_gw,
1435 static bool peer_pmtu_expired(struct inet_peer *peer)
1437 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1440 time_after_eq(jiffies, orig) &&
1441 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1444 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1446 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1449 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1452 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1454 struct rtable *rt = (struct rtable *)dst;
1455 struct dst_entry *ret = dst;
1458 if (dst->obsolete > 0) {
1461 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1462 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1464 rt_genid(dev_net(dst->dev)));
1467 } else if (rt_has_peer(rt)) {
1468 struct inet_peer *peer = rt_peer_ptr(rt);
1469 if (peer_pmtu_expired(peer))
1470 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1478 * 1. The first ip_rt_redirect_number redirects are sent
1479 * with exponential backoff, then we stop sending them at all,
1480 * assuming that the host ignores our redirects.
1481 * 2. If we did not see packets requiring redirects
1482 * during ip_rt_redirect_silence, we assume that the host
1483 * forgot redirected route and start to send redirects again.
1485 * This algorithm is much cheaper and more intelligent than dumb load limiting
1488 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1489 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1492 void ip_rt_send_redirect(struct sk_buff *skb)
1494 struct rtable *rt = skb_rtable(skb);
1495 struct in_device *in_dev;
1496 struct inet_peer *peer;
1500 in_dev = __in_dev_get_rcu(rt->dst.dev);
1501 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1505 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1508 peer = rt_get_peer_create(rt, rt->rt_dst);
1510 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1514 /* No redirected packets during ip_rt_redirect_silence;
1515 * reset the algorithm.
1517 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1518 peer->rate_tokens = 0;
1520 /* Too many ignored redirects; do not send anything
1521 * set dst.rate_last to the last seen redirected packet.
1523 if (peer->rate_tokens >= ip_rt_redirect_number) {
1524 peer->rate_last = jiffies;
1528 /* Check for load limit; set rate_last to the latest sent
1531 if (peer->rate_tokens == 0 ||
1534 (ip_rt_redirect_load << peer->rate_tokens)))) {
1535 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1536 peer->rate_last = jiffies;
1537 ++peer->rate_tokens;
1538 #ifdef CONFIG_IP_ROUTE_VERBOSE
1540 peer->rate_tokens == ip_rt_redirect_number)
1541 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1542 &ip_hdr(skb)->saddr, rt->rt_iif,
1543 &rt->rt_dst, &rt->rt_gateway);
1548 static int ip_error(struct sk_buff *skb)
1550 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1551 struct rtable *rt = skb_rtable(skb);
1552 struct inet_peer *peer;
1558 net = dev_net(rt->dst.dev);
1559 if (!IN_DEV_FORWARD(in_dev)) {
1560 switch (rt->dst.error) {
1562 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
1566 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1572 switch (rt->dst.error) {
1577 code = ICMP_HOST_UNREACH;
1580 code = ICMP_NET_UNREACH;
1581 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1584 code = ICMP_PKT_FILTERED;
1588 peer = rt_get_peer_create(rt, rt->rt_dst);
1593 peer->rate_tokens += now - peer->rate_last;
1594 if (peer->rate_tokens > ip_rt_error_burst)
1595 peer->rate_tokens = ip_rt_error_burst;
1596 peer->rate_last = now;
1597 if (peer->rate_tokens >= ip_rt_error_cost)
1598 peer->rate_tokens -= ip_rt_error_cost;
1603 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1605 out: kfree_skb(skb);
1609 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1611 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1615 if (time_before(jiffies, expires)) {
1616 u32 orig_dst_mtu = dst_mtu(dst);
1617 if (peer->pmtu_learned < orig_dst_mtu) {
1618 if (!peer->pmtu_orig)
1619 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1620 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1622 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1623 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1626 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1628 struct rtable *rt = (struct rtable *) dst;
1629 struct inet_peer *peer;
1633 peer = rt_get_peer_create(rt, rt->rt_dst);
1635 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1637 if (mtu < ip_rt_min_pmtu)
1638 mtu = ip_rt_min_pmtu;
1639 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1641 pmtu_expires = jiffies + ip_rt_mtu_expires;
1645 peer->pmtu_learned = mtu;
1646 peer->pmtu_expires = pmtu_expires;
1648 atomic_inc(&__rt_peer_genid);
1649 rt->rt_peer_genid = rt_peer_genid();
1651 check_peer_pmtu(dst, peer);
1655 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1656 int oif, u32 mark, u8 protocol, int flow_flags)
1658 const struct iphdr *iph = (const struct iphdr *)skb->data;
1662 flowi4_init_output(&fl4, oif, mark, RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
1663 protocol, flow_flags | FLOWI_FLAG_PRECOW_METRICS,
1664 iph->daddr, iph->saddr, 0, 0);
1665 rt = __ip_route_output_key(net, &fl4);
1667 ip_rt_update_pmtu(&rt->dst, mtu);
1671 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1673 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1675 const struct inet_sock *inet = inet_sk(sk);
1677 return ipv4_update_pmtu(skb, sock_net(sk), mtu,
1678 sk->sk_bound_dev_if, sk->sk_mark,
1679 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
1680 inet_sk_flowi_flags(sk));
1682 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1684 static void ipv4_validate_peer(struct rtable *rt)
1686 if (rt->rt_peer_genid != rt_peer_genid()) {
1687 struct inet_peer *peer = rt_get_peer(rt, rt->rt_dst);
1690 check_peer_pmtu(&rt->dst, peer);
1692 if (peer->redirect_learned.a4 &&
1693 peer->redirect_learned.a4 != rt->rt_gateway)
1694 check_peer_redir(&rt->dst, peer);
1697 rt->rt_peer_genid = rt_peer_genid();
1701 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1703 struct rtable *rt = (struct rtable *) dst;
1705 if (rt_is_expired(rt))
1707 ipv4_validate_peer(rt);
1711 static void ipv4_dst_destroy(struct dst_entry *dst)
1713 struct rtable *rt = (struct rtable *) dst;
1716 fib_info_put(rt->fi);
1719 if (rt_has_peer(rt)) {
1720 struct inet_peer *peer = rt_peer_ptr(rt);
1726 static void ipv4_link_failure(struct sk_buff *skb)
1730 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1732 rt = skb_rtable(skb);
1733 if (rt && rt_has_peer(rt)) {
1734 struct inet_peer *peer = rt_peer_ptr(rt);
1735 if (peer_pmtu_cleaned(peer))
1736 dst_metric_set(&rt->dst, RTAX_MTU, peer->pmtu_orig);
1740 static int ip_rt_bug(struct sk_buff *skb)
1742 pr_debug("%s: %pI4 -> %pI4, %s\n",
1743 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1744 skb->dev ? skb->dev->name : "?");
1751 We do not cache source address of outgoing interface,
1752 because it is used only by IP RR, TS and SRR options,
1753 so that it out of fast path.
1755 BTW remember: "addr" is allowed to be not aligned
1759 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1763 if (rt_is_output_route(rt))
1764 src = ip_hdr(skb)->saddr;
1766 struct fib_result res;
1772 memset(&fl4, 0, sizeof(fl4));
1773 fl4.daddr = iph->daddr;
1774 fl4.saddr = iph->saddr;
1775 fl4.flowi4_tos = RT_TOS(iph->tos);
1776 fl4.flowi4_oif = rt->dst.dev->ifindex;
1777 fl4.flowi4_iif = skb->dev->ifindex;
1778 fl4.flowi4_mark = skb->mark;
1781 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1782 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1784 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1788 memcpy(addr, &src, 4);
1791 #ifdef CONFIG_IP_ROUTE_CLASSID
1792 static void set_class_tag(struct rtable *rt, u32 tag)
1794 if (!(rt->dst.tclassid & 0xFFFF))
1795 rt->dst.tclassid |= tag & 0xFFFF;
1796 if (!(rt->dst.tclassid & 0xFFFF0000))
1797 rt->dst.tclassid |= tag & 0xFFFF0000;
1801 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1803 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1806 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1808 if (advmss > 65535 - 40)
1809 advmss = 65535 - 40;
1814 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1816 const struct rtable *rt = (const struct rtable *) dst;
1817 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1819 if (mtu && rt_is_output_route(rt))
1822 mtu = dst->dev->mtu;
1824 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1826 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1830 if (mtu > IP_MAX_MTU)
1836 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1837 struct fib_info *fi)
1839 struct inet_peer_base *base;
1840 struct inet_peer *peer;
1843 /* If a peer entry exists for this destination, we must hook
1844 * it up in order to get at cached metrics.
1846 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1849 base = inetpeer_base_ptr(rt->_peer);
1852 peer = inet_getpeer_v4(base, rt->rt_dst, create);
1854 __rt_set_peer(rt, peer);
1855 rt->rt_peer_genid = rt_peer_genid();
1856 if (inet_metrics_new(peer))
1857 memcpy(peer->metrics, fi->fib_metrics,
1858 sizeof(u32) * RTAX_MAX);
1859 dst_init_metrics(&rt->dst, peer->metrics, false);
1861 check_peer_pmtu(&rt->dst, peer);
1863 if (peer->redirect_learned.a4 &&
1864 peer->redirect_learned.a4 != rt->rt_gateway) {
1865 rt->rt_gateway = peer->redirect_learned.a4;
1866 rt->rt_flags |= RTCF_REDIRECTED;
1869 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1871 atomic_inc(&fi->fib_clntref);
1873 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1877 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1878 const struct fib_result *res,
1879 struct fib_info *fi, u16 type, u32 itag)
1881 struct dst_entry *dst = &rt->dst;
1884 if (FIB_RES_GW(*res) &&
1885 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1886 rt->rt_gateway = FIB_RES_GW(*res);
1887 rt_init_metrics(rt, fl4, fi);
1888 #ifdef CONFIG_IP_ROUTE_CLASSID
1889 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1893 if (dst_mtu(dst) > IP_MAX_MTU)
1894 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1896 #ifdef CONFIG_IP_ROUTE_CLASSID
1897 #ifdef CONFIG_IP_MULTIPLE_TABLES
1898 set_class_tag(rt, fib_rules_tclass(res));
1900 set_class_tag(rt, itag);
1904 static struct rtable *rt_dst_alloc(struct net_device *dev,
1905 bool nopolicy, bool noxfrm)
1907 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1909 (nopolicy ? DST_NOPOLICY : 0) |
1910 (noxfrm ? DST_NOXFRM : 0));
1913 /* called in rcu_read_lock() section */
1914 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1915 u8 tos, struct net_device *dev, int our)
1919 struct in_device *in_dev = __in_dev_get_rcu(dev);
1923 /* Primary sanity checks. */
1928 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1929 skb->protocol != htons(ETH_P_IP))
1932 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1933 if (ipv4_is_loopback(saddr))
1936 if (ipv4_is_zeronet(saddr)) {
1937 if (!ipv4_is_local_multicast(daddr))
1940 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1945 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1946 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1950 #ifdef CONFIG_IP_ROUTE_CLASSID
1951 rth->dst.tclassid = itag;
1953 rth->dst.output = ip_rt_bug;
1955 rth->rt_key_dst = daddr;
1956 rth->rt_key_src = saddr;
1957 rth->rt_genid = rt_genid(dev_net(dev));
1958 rth->rt_flags = RTCF_MULTICAST;
1959 rth->rt_type = RTN_MULTICAST;
1960 rth->rt_key_tos = tos;
1961 rth->rt_dst = daddr;
1962 rth->rt_src = saddr;
1963 rth->rt_route_iif = dev->ifindex;
1964 rth->rt_iif = dev->ifindex;
1966 rth->rt_mark = skb->mark;
1967 rth->rt_gateway = daddr;
1968 rth->rt_peer_genid = 0;
1969 rt_init_peer(rth, dev_net(dev)->ipv4.peers);
1972 rth->dst.input= ip_local_deliver;
1973 rth->rt_flags |= RTCF_LOCAL;
1976 #ifdef CONFIG_IP_MROUTE
1977 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1978 rth->dst.input = ip_mr_input;
1980 RT_CACHE_STAT_INC(in_slow_mc);
1982 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1983 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1984 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1995 static void ip_handle_martian_source(struct net_device *dev,
1996 struct in_device *in_dev,
1997 struct sk_buff *skb,
2001 RT_CACHE_STAT_INC(in_martian_src);
2002 #ifdef CONFIG_IP_ROUTE_VERBOSE
2003 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2005 * RFC1812 recommendation, if source is martian,
2006 * the only hint is MAC header.
2008 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
2009 &daddr, &saddr, dev->name);
2010 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2011 print_hex_dump(KERN_WARNING, "ll header: ",
2012 DUMP_PREFIX_OFFSET, 16, 1,
2013 skb_mac_header(skb),
2014 dev->hard_header_len, true);
2020 /* called in rcu_read_lock() section */
2021 static int __mkroute_input(struct sk_buff *skb,
2022 const struct fib_result *res,
2023 struct in_device *in_dev,
2024 __be32 daddr, __be32 saddr, u32 tos,
2025 struct rtable **result)
2029 struct in_device *out_dev;
2030 unsigned int flags = 0;
2033 /* get a working reference to the output device */
2034 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2035 if (out_dev == NULL) {
2036 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
2041 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2042 in_dev->dev, in_dev, &itag);
2044 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2051 flags |= RTCF_DIRECTSRC;
2053 if (out_dev == in_dev && err &&
2054 (IN_DEV_SHARED_MEDIA(out_dev) ||
2055 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2056 flags |= RTCF_DOREDIRECT;
2058 if (skb->protocol != htons(ETH_P_IP)) {
2059 /* Not IP (i.e. ARP). Do not create route, if it is
2060 * invalid for proxy arp. DNAT routes are always valid.
2062 * Proxy arp feature have been extended to allow, ARP
2063 * replies back to the same interface, to support
2064 * Private VLAN switch technologies. See arp.c.
2066 if (out_dev == in_dev &&
2067 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2073 rth = rt_dst_alloc(out_dev->dev,
2074 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2075 IN_DEV_CONF_GET(out_dev, NOXFRM));
2081 rth->rt_key_dst = daddr;
2082 rth->rt_key_src = saddr;
2083 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2084 rth->rt_flags = flags;
2085 rth->rt_type = res->type;
2086 rth->rt_key_tos = tos;
2087 rth->rt_dst = daddr;
2088 rth->rt_src = saddr;
2089 rth->rt_route_iif = in_dev->dev->ifindex;
2090 rth->rt_iif = in_dev->dev->ifindex;
2092 rth->rt_mark = skb->mark;
2093 rth->rt_gateway = daddr;
2094 rth->rt_peer_genid = 0;
2095 rt_init_peer(rth, &res->table->tb_peers);
2098 rth->dst.input = ip_forward;
2099 rth->dst.output = ip_output;
2101 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2109 static int ip_mkroute_input(struct sk_buff *skb,
2110 struct fib_result *res,
2111 const struct flowi4 *fl4,
2112 struct in_device *in_dev,
2113 __be32 daddr, __be32 saddr, u32 tos)
2115 struct rtable *rth = NULL;
2119 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2120 if (res->fi && res->fi->fib_nhs > 1)
2121 fib_select_multipath(res);
2124 /* create a routing cache entry */
2125 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2129 /* put it into the cache */
2130 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2131 rt_genid(dev_net(rth->dst.dev)));
2132 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2134 return PTR_ERR(rth);
2139 * NOTE. We drop all the packets that has local source
2140 * addresses, because every properly looped back packet
2141 * must have correct destination already attached by output routine.
2143 * Such approach solves two big problems:
2144 * 1. Not simplex devices are handled properly.
2145 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2146 * called with rcu_read_lock()
2149 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2150 u8 tos, struct net_device *dev)
2152 struct fib_result res;
2153 struct in_device *in_dev = __in_dev_get_rcu(dev);
2155 unsigned int flags = 0;
2160 struct net *net = dev_net(dev);
2162 /* IP on this device is disabled. */
2167 /* Check for the most weird martians, which can be not detected
2171 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2172 goto martian_source;
2174 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2177 /* Accept zero addresses only to limited broadcast;
2178 * I even do not know to fix it or not. Waiting for complains :-)
2180 if (ipv4_is_zeronet(saddr))
2181 goto martian_source;
2183 if (ipv4_is_zeronet(daddr))
2184 goto martian_destination;
2186 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
2187 if (ipv4_is_loopback(daddr))
2188 goto martian_destination;
2190 if (ipv4_is_loopback(saddr))
2191 goto martian_source;
2195 * Now we are ready to route packet.
2198 fl4.flowi4_iif = dev->ifindex;
2199 fl4.flowi4_mark = skb->mark;
2200 fl4.flowi4_tos = tos;
2201 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2204 err = fib_lookup(net, &fl4, &res);
2208 RT_CACHE_STAT_INC(in_slow_tot);
2210 if (res.type == RTN_BROADCAST)
2213 if (res.type == RTN_LOCAL) {
2214 err = fib_validate_source(skb, saddr, daddr, tos,
2215 net->loopback_dev->ifindex,
2216 dev, in_dev, &itag);
2218 goto martian_source_keep_err;
2220 flags |= RTCF_DIRECTSRC;
2224 if (!IN_DEV_FORWARD(in_dev))
2226 if (res.type != RTN_UNICAST)
2227 goto martian_destination;
2229 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2233 if (skb->protocol != htons(ETH_P_IP))
2236 if (!ipv4_is_zeronet(saddr)) {
2237 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2240 goto martian_source_keep_err;
2242 flags |= RTCF_DIRECTSRC;
2244 flags |= RTCF_BROADCAST;
2245 res.type = RTN_BROADCAST;
2246 RT_CACHE_STAT_INC(in_brd);
2249 rth = rt_dst_alloc(net->loopback_dev,
2250 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2254 rth->dst.input= ip_local_deliver;
2255 rth->dst.output= ip_rt_bug;
2256 #ifdef CONFIG_IP_ROUTE_CLASSID
2257 rth->dst.tclassid = itag;
2260 rth->rt_key_dst = daddr;
2261 rth->rt_key_src = saddr;
2262 rth->rt_genid = rt_genid(net);
2263 rth->rt_flags = flags|RTCF_LOCAL;
2264 rth->rt_type = res.type;
2265 rth->rt_key_tos = tos;
2266 rth->rt_dst = daddr;
2267 rth->rt_src = saddr;
2268 rth->rt_route_iif = dev->ifindex;
2269 rth->rt_iif = dev->ifindex;
2271 rth->rt_mark = skb->mark;
2272 rth->rt_gateway = daddr;
2273 rth->rt_peer_genid = 0;
2274 rt_init_peer(rth, net->ipv4.peers);
2276 if (res.type == RTN_UNREACHABLE) {
2277 rth->dst.input= ip_error;
2278 rth->dst.error= -err;
2279 rth->rt_flags &= ~RTCF_LOCAL;
2281 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2282 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2289 RT_CACHE_STAT_INC(in_no_route);
2290 res.type = RTN_UNREACHABLE;
2296 * Do not cache martian addresses: they should be logged (RFC1812)
2298 martian_destination:
2299 RT_CACHE_STAT_INC(in_martian_dst);
2300 #ifdef CONFIG_IP_ROUTE_VERBOSE
2301 if (IN_DEV_LOG_MARTIANS(in_dev))
2302 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2303 &daddr, &saddr, dev->name);
2316 martian_source_keep_err:
2317 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2321 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2322 u8 tos, struct net_device *dev, bool noref)
2326 int iif = dev->ifindex;
2334 if (!rt_caching(net))
2337 tos &= IPTOS_RT_MASK;
2338 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2340 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2341 rth = rcu_dereference(rth->dst.rt_next)) {
2342 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2343 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2344 (rth->rt_route_iif ^ iif) |
2345 (rth->rt_key_tos ^ tos)) == 0 &&
2346 rth->rt_mark == skb->mark &&
2347 net_eq(dev_net(rth->dst.dev), net) &&
2348 !rt_is_expired(rth)) {
2349 ipv4_validate_peer(rth);
2351 dst_use_noref(&rth->dst, jiffies);
2352 skb_dst_set_noref(skb, &rth->dst);
2354 dst_use(&rth->dst, jiffies);
2355 skb_dst_set(skb, &rth->dst);
2357 RT_CACHE_STAT_INC(in_hit);
2361 RT_CACHE_STAT_INC(in_hlist_search);
2365 /* Multicast recognition logic is moved from route cache to here.
2366 The problem was that too many Ethernet cards have broken/missing
2367 hardware multicast filters :-( As result the host on multicasting
2368 network acquires a lot of useless route cache entries, sort of
2369 SDR messages from all the world. Now we try to get rid of them.
2370 Really, provided software IP multicast filter is organized
2371 reasonably (at least, hashed), it does not result in a slowdown
2372 comparing with route cache reject entries.
2373 Note, that multicast routers are not affected, because
2374 route cache entry is created eventually.
2376 if (ipv4_is_multicast(daddr)) {
2377 struct in_device *in_dev = __in_dev_get_rcu(dev);
2380 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2381 ip_hdr(skb)->protocol);
2383 #ifdef CONFIG_IP_MROUTE
2385 (!ipv4_is_local_multicast(daddr) &&
2386 IN_DEV_MFORWARD(in_dev))
2389 int res = ip_route_input_mc(skb, daddr, saddr,
2398 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2402 EXPORT_SYMBOL(ip_route_input_common);
2404 /* called with rcu_read_lock() */
2405 static struct rtable *__mkroute_output(const struct fib_result *res,
2406 const struct flowi4 *fl4,
2407 __be32 orig_daddr, __be32 orig_saddr,
2408 int orig_oif, __u8 orig_rtos,
2409 struct net_device *dev_out,
2412 struct fib_info *fi = res->fi;
2413 struct in_device *in_dev;
2414 u16 type = res->type;
2417 in_dev = __in_dev_get_rcu(dev_out);
2419 return ERR_PTR(-EINVAL);
2421 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2422 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2423 return ERR_PTR(-EINVAL);
2425 if (ipv4_is_lbcast(fl4->daddr))
2426 type = RTN_BROADCAST;
2427 else if (ipv4_is_multicast(fl4->daddr))
2428 type = RTN_MULTICAST;
2429 else if (ipv4_is_zeronet(fl4->daddr))
2430 return ERR_PTR(-EINVAL);
2432 if (dev_out->flags & IFF_LOOPBACK)
2433 flags |= RTCF_LOCAL;
2435 if (type == RTN_BROADCAST) {
2436 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2438 } else if (type == RTN_MULTICAST) {
2439 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2440 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2442 flags &= ~RTCF_LOCAL;
2443 /* If multicast route do not exist use
2444 * default one, but do not gateway in this case.
2447 if (fi && res->prefixlen < 4)
2451 rth = rt_dst_alloc(dev_out,
2452 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2453 IN_DEV_CONF_GET(in_dev, NOXFRM));
2455 return ERR_PTR(-ENOBUFS);
2457 rth->dst.output = ip_output;
2459 rth->rt_key_dst = orig_daddr;
2460 rth->rt_key_src = orig_saddr;
2461 rth->rt_genid = rt_genid(dev_net(dev_out));
2462 rth->rt_flags = flags;
2463 rth->rt_type = type;
2464 rth->rt_key_tos = orig_rtos;
2465 rth->rt_dst = fl4->daddr;
2466 rth->rt_src = fl4->saddr;
2467 rth->rt_route_iif = 0;
2468 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2469 rth->rt_oif = orig_oif;
2470 rth->rt_mark = fl4->flowi4_mark;
2471 rth->rt_gateway = fl4->daddr;
2472 rth->rt_peer_genid = 0;
2473 rt_init_peer(rth, (res->table ?
2474 &res->table->tb_peers :
2475 dev_net(dev_out)->ipv4.peers));
2478 RT_CACHE_STAT_INC(out_slow_tot);
2480 if (flags & RTCF_LOCAL)
2481 rth->dst.input = ip_local_deliver;
2482 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2483 if (flags & RTCF_LOCAL &&
2484 !(dev_out->flags & IFF_LOOPBACK)) {
2485 rth->dst.output = ip_mc_output;
2486 RT_CACHE_STAT_INC(out_slow_mc);
2488 #ifdef CONFIG_IP_MROUTE
2489 if (type == RTN_MULTICAST) {
2490 if (IN_DEV_MFORWARD(in_dev) &&
2491 !ipv4_is_local_multicast(fl4->daddr)) {
2492 rth->dst.input = ip_mr_input;
2493 rth->dst.output = ip_mc_output;
2499 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2501 if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
2502 rth->dst.flags |= DST_NOCACHE;
2508 * Major route resolver routine.
2509 * called with rcu_read_lock();
2512 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2514 struct net_device *dev_out = NULL;
2515 __u8 tos = RT_FL_TOS(fl4);
2516 unsigned int flags = 0;
2517 struct fib_result res;
2525 #ifdef CONFIG_IP_MULTIPLE_TABLES
2529 orig_daddr = fl4->daddr;
2530 orig_saddr = fl4->saddr;
2531 orig_oif = fl4->flowi4_oif;
2533 fl4->flowi4_iif = net->loopback_dev->ifindex;
2534 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2535 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2536 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2540 rth = ERR_PTR(-EINVAL);
2541 if (ipv4_is_multicast(fl4->saddr) ||
2542 ipv4_is_lbcast(fl4->saddr) ||
2543 ipv4_is_zeronet(fl4->saddr))
2546 /* I removed check for oif == dev_out->oif here.
2547 It was wrong for two reasons:
2548 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2549 is assigned to multiple interfaces.
2550 2. Moreover, we are allowed to send packets with saddr
2551 of another iface. --ANK
2554 if (fl4->flowi4_oif == 0 &&
2555 (ipv4_is_multicast(fl4->daddr) ||
2556 ipv4_is_lbcast(fl4->daddr))) {
2557 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2558 dev_out = __ip_dev_find(net, fl4->saddr, false);
2559 if (dev_out == NULL)
2562 /* Special hack: user can direct multicasts
2563 and limited broadcast via necessary interface
2564 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2565 This hack is not just for fun, it allows
2566 vic,vat and friends to work.
2567 They bind socket to loopback, set ttl to zero
2568 and expect that it will work.
2569 From the viewpoint of routing cache they are broken,
2570 because we are not allowed to build multicast path
2571 with loopback source addr (look, routing cache
2572 cannot know, that ttl is zero, so that packet
2573 will not leave this host and route is valid).
2574 Luckily, this hack is good workaround.
2577 fl4->flowi4_oif = dev_out->ifindex;
2581 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2582 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2583 if (!__ip_dev_find(net, fl4->saddr, false))
2589 if (fl4->flowi4_oif) {
2590 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2591 rth = ERR_PTR(-ENODEV);
2592 if (dev_out == NULL)
2595 /* RACE: Check return value of inet_select_addr instead. */
2596 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2597 rth = ERR_PTR(-ENETUNREACH);
2600 if (ipv4_is_local_multicast(fl4->daddr) ||
2601 ipv4_is_lbcast(fl4->daddr)) {
2603 fl4->saddr = inet_select_addr(dev_out, 0,
2608 if (ipv4_is_multicast(fl4->daddr))
2609 fl4->saddr = inet_select_addr(dev_out, 0,
2611 else if (!fl4->daddr)
2612 fl4->saddr = inet_select_addr(dev_out, 0,
2618 fl4->daddr = fl4->saddr;
2620 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2621 dev_out = net->loopback_dev;
2622 fl4->flowi4_oif = net->loopback_dev->ifindex;
2623 res.type = RTN_LOCAL;
2624 flags |= RTCF_LOCAL;
2628 if (fib_lookup(net, fl4, &res)) {
2631 if (fl4->flowi4_oif) {
2632 /* Apparently, routing tables are wrong. Assume,
2633 that the destination is on link.
2636 Because we are allowed to send to iface
2637 even if it has NO routes and NO assigned
2638 addresses. When oif is specified, routing
2639 tables are looked up with only one purpose:
2640 to catch if destination is gatewayed, rather than
2641 direct. Moreover, if MSG_DONTROUTE is set,
2642 we send packet, ignoring both routing tables
2643 and ifaddr state. --ANK
2646 We could make it even if oif is unknown,
2647 likely IPv6, but we do not.
2650 if (fl4->saddr == 0)
2651 fl4->saddr = inet_select_addr(dev_out, 0,
2653 res.type = RTN_UNICAST;
2656 rth = ERR_PTR(-ENETUNREACH);
2660 if (res.type == RTN_LOCAL) {
2662 if (res.fi->fib_prefsrc)
2663 fl4->saddr = res.fi->fib_prefsrc;
2665 fl4->saddr = fl4->daddr;
2667 dev_out = net->loopback_dev;
2668 fl4->flowi4_oif = dev_out->ifindex;
2670 flags |= RTCF_LOCAL;
2674 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2675 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2676 fib_select_multipath(&res);
2679 if (!res.prefixlen &&
2680 res.table->tb_num_default > 1 &&
2681 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2682 fib_select_default(&res);
2685 fl4->saddr = FIB_RES_PREFSRC(net, res);
2687 dev_out = FIB_RES_DEV(res);
2688 fl4->flowi4_oif = dev_out->ifindex;
2692 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2693 tos, dev_out, flags);
2697 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2698 rt_genid(dev_net(dev_out)));
2699 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2707 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2712 if (!rt_caching(net))
2715 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2718 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2719 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2720 if (rth->rt_key_dst == flp4->daddr &&
2721 rth->rt_key_src == flp4->saddr &&
2722 rt_is_output_route(rth) &&
2723 rth->rt_oif == flp4->flowi4_oif &&
2724 rth->rt_mark == flp4->flowi4_mark &&
2725 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2726 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2727 net_eq(dev_net(rth->dst.dev), net) &&
2728 !rt_is_expired(rth)) {
2729 ipv4_validate_peer(rth);
2730 dst_use(&rth->dst, jiffies);
2731 RT_CACHE_STAT_INC(out_hit);
2732 rcu_read_unlock_bh();
2734 flp4->saddr = rth->rt_src;
2736 flp4->daddr = rth->rt_dst;
2739 RT_CACHE_STAT_INC(out_hlist_search);
2741 rcu_read_unlock_bh();
2744 return ip_route_output_slow(net, flp4);
2746 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2748 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2753 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2755 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2757 return mtu ? : dst->dev->mtu;
2760 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2764 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2770 static struct dst_ops ipv4_dst_blackhole_ops = {
2772 .protocol = cpu_to_be16(ETH_P_IP),
2773 .destroy = ipv4_dst_destroy,
2774 .check = ipv4_blackhole_dst_check,
2775 .mtu = ipv4_blackhole_mtu,
2776 .default_advmss = ipv4_default_advmss,
2777 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2778 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2779 .neigh_lookup = ipv4_neigh_lookup,
2782 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2784 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2785 struct rtable *ort = (struct rtable *) dst_orig;
2788 struct dst_entry *new = &rt->dst;
2791 new->input = dst_discard;
2792 new->output = dst_discard;
2793 dst_copy_metrics(new, &ort->dst);
2795 new->dev = ort->dst.dev;
2799 rt->rt_key_dst = ort->rt_key_dst;
2800 rt->rt_key_src = ort->rt_key_src;
2801 rt->rt_key_tos = ort->rt_key_tos;
2802 rt->rt_route_iif = ort->rt_route_iif;
2803 rt->rt_iif = ort->rt_iif;
2804 rt->rt_oif = ort->rt_oif;
2805 rt->rt_mark = ort->rt_mark;
2807 rt->rt_genid = rt_genid(net);
2808 rt->rt_flags = ort->rt_flags;
2809 rt->rt_type = ort->rt_type;
2810 rt->rt_dst = ort->rt_dst;
2811 rt->rt_src = ort->rt_src;
2812 rt->rt_gateway = ort->rt_gateway;
2813 rt_transfer_peer(rt, ort);
2816 atomic_inc(&rt->fi->fib_clntref);
2821 dst_release(dst_orig);
2823 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2826 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2829 struct rtable *rt = __ip_route_output_key(net, flp4);
2834 if (flp4->flowi4_proto)
2835 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2836 flowi4_to_flowi(flp4),
2841 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2843 static int rt_fill_info(struct net *net,
2844 struct sk_buff *skb, u32 pid, u32 seq, int event,
2845 int nowait, unsigned int flags)
2847 struct rtable *rt = skb_rtable(skb);
2849 struct nlmsghdr *nlh;
2850 unsigned long expires = 0;
2851 u32 id = 0, ts = 0, tsage = 0, error;
2853 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2857 r = nlmsg_data(nlh);
2858 r->rtm_family = AF_INET;
2859 r->rtm_dst_len = 32;
2861 r->rtm_tos = rt->rt_key_tos;
2862 r->rtm_table = RT_TABLE_MAIN;
2863 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2864 goto nla_put_failure;
2865 r->rtm_type = rt->rt_type;
2866 r->rtm_scope = RT_SCOPE_UNIVERSE;
2867 r->rtm_protocol = RTPROT_UNSPEC;
2868 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2869 if (rt->rt_flags & RTCF_NOTIFY)
2870 r->rtm_flags |= RTM_F_NOTIFY;
2872 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2873 goto nla_put_failure;
2874 if (rt->rt_key_src) {
2875 r->rtm_src_len = 32;
2876 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2877 goto nla_put_failure;
2880 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2881 goto nla_put_failure;
2882 #ifdef CONFIG_IP_ROUTE_CLASSID
2883 if (rt->dst.tclassid &&
2884 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2885 goto nla_put_failure;
2887 if (!rt_is_input_route(rt) &&
2888 rt->rt_src != rt->rt_key_src) {
2889 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2890 goto nla_put_failure;
2892 if (rt->rt_dst != rt->rt_gateway &&
2893 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2894 goto nla_put_failure;
2896 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2897 goto nla_put_failure;
2900 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2901 goto nla_put_failure;
2903 error = rt->dst.error;
2904 if (rt_has_peer(rt)) {
2905 const struct inet_peer *peer = rt_peer_ptr(rt);
2906 inet_peer_refcheck(peer);
2907 id = atomic_read(&peer->ip_id_count) & 0xffff;
2908 if (peer->tcp_ts_stamp) {
2910 tsage = get_seconds() - peer->tcp_ts_stamp;
2912 expires = ACCESS_ONCE(peer->pmtu_expires);
2914 if (time_before(jiffies, expires))
2921 if (rt_is_input_route(rt)) {
2922 #ifdef CONFIG_IP_MROUTE
2923 __be32 dst = rt->rt_dst;
2925 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2926 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2927 int err = ipmr_get_route(net, skb,
2928 rt->rt_src, rt->rt_dst,
2934 goto nla_put_failure;
2936 if (err == -EMSGSIZE)
2937 goto nla_put_failure;
2943 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2944 goto nla_put_failure;
2947 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2948 expires, error) < 0)
2949 goto nla_put_failure;
2951 return nlmsg_end(skb, nlh);
2954 nlmsg_cancel(skb, nlh);
2958 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2960 struct net *net = sock_net(in_skb->sk);
2962 struct nlattr *tb[RTA_MAX+1];
2963 struct rtable *rt = NULL;
2969 struct sk_buff *skb;
2971 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2975 rtm = nlmsg_data(nlh);
2977 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2983 /* Reserve room for dummy headers, this skb can pass
2984 through good chunk of routing engine.
2986 skb_reset_mac_header(skb);
2987 skb_reset_network_header(skb);
2989 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2990 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2991 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2993 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2994 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2995 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2996 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2999 struct net_device *dev;
3001 dev = __dev_get_by_index(net, iif);
3007 skb->protocol = htons(ETH_P_IP);
3011 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3014 rt = skb_rtable(skb);
3015 if (err == 0 && rt->dst.error)
3016 err = -rt->dst.error;
3018 struct flowi4 fl4 = {
3021 .flowi4_tos = rtm->rtm_tos,
3022 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3023 .flowi4_mark = mark,
3025 rt = ip_route_output_key(net, &fl4);
3035 skb_dst_set(skb, &rt->dst);
3036 if (rtm->rtm_flags & RTM_F_NOTIFY)
3037 rt->rt_flags |= RTCF_NOTIFY;
3039 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3040 RTM_NEWROUTE, 0, 0);
3044 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3053 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3060 net = sock_net(skb->sk);
3065 s_idx = idx = cb->args[1];
3066 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3067 if (!rt_hash_table[h].chain)
3070 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3071 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3072 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3074 if (rt_is_expired(rt))
3076 skb_dst_set_noref(skb, &rt->dst);
3077 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3078 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3079 1, NLM_F_MULTI) <= 0) {
3081 rcu_read_unlock_bh();
3086 rcu_read_unlock_bh();
3095 void ip_rt_multicast_event(struct in_device *in_dev)
3097 rt_cache_flush(dev_net(in_dev->dev), 0);
3100 #ifdef CONFIG_SYSCTL
3101 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3102 void __user *buffer,
3103 size_t *lenp, loff_t *ppos)
3110 memcpy(&ctl, __ctl, sizeof(ctl));
3111 ctl.data = &flush_delay;
3112 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3114 net = (struct net *)__ctl->extra1;
3115 rt_cache_flush(net, flush_delay);
3122 static ctl_table ipv4_route_table[] = {
3124 .procname = "gc_thresh",
3125 .data = &ipv4_dst_ops.gc_thresh,
3126 .maxlen = sizeof(int),
3128 .proc_handler = proc_dointvec,
3131 .procname = "max_size",
3132 .data = &ip_rt_max_size,
3133 .maxlen = sizeof(int),
3135 .proc_handler = proc_dointvec,
3138 /* Deprecated. Use gc_min_interval_ms */
3140 .procname = "gc_min_interval",
3141 .data = &ip_rt_gc_min_interval,
3142 .maxlen = sizeof(int),
3144 .proc_handler = proc_dointvec_jiffies,
3147 .procname = "gc_min_interval_ms",
3148 .data = &ip_rt_gc_min_interval,
3149 .maxlen = sizeof(int),
3151 .proc_handler = proc_dointvec_ms_jiffies,
3154 .procname = "gc_timeout",
3155 .data = &ip_rt_gc_timeout,
3156 .maxlen = sizeof(int),
3158 .proc_handler = proc_dointvec_jiffies,
3161 .procname = "gc_interval",
3162 .data = &ip_rt_gc_interval,
3163 .maxlen = sizeof(int),
3165 .proc_handler = proc_dointvec_jiffies,
3168 .procname = "redirect_load",
3169 .data = &ip_rt_redirect_load,
3170 .maxlen = sizeof(int),
3172 .proc_handler = proc_dointvec,
3175 .procname = "redirect_number",
3176 .data = &ip_rt_redirect_number,
3177 .maxlen = sizeof(int),
3179 .proc_handler = proc_dointvec,
3182 .procname = "redirect_silence",
3183 .data = &ip_rt_redirect_silence,
3184 .maxlen = sizeof(int),
3186 .proc_handler = proc_dointvec,
3189 .procname = "error_cost",
3190 .data = &ip_rt_error_cost,
3191 .maxlen = sizeof(int),
3193 .proc_handler = proc_dointvec,
3196 .procname = "error_burst",
3197 .data = &ip_rt_error_burst,
3198 .maxlen = sizeof(int),
3200 .proc_handler = proc_dointvec,
3203 .procname = "gc_elasticity",
3204 .data = &ip_rt_gc_elasticity,
3205 .maxlen = sizeof(int),
3207 .proc_handler = proc_dointvec,
3210 .procname = "mtu_expires",
3211 .data = &ip_rt_mtu_expires,
3212 .maxlen = sizeof(int),
3214 .proc_handler = proc_dointvec_jiffies,
3217 .procname = "min_pmtu",
3218 .data = &ip_rt_min_pmtu,
3219 .maxlen = sizeof(int),
3221 .proc_handler = proc_dointvec,
3224 .procname = "min_adv_mss",
3225 .data = &ip_rt_min_advmss,
3226 .maxlen = sizeof(int),
3228 .proc_handler = proc_dointvec,
3233 static struct ctl_table ipv4_route_flush_table[] = {
3235 .procname = "flush",
3236 .maxlen = sizeof(int),
3238 .proc_handler = ipv4_sysctl_rtcache_flush,
3243 static __net_init int sysctl_route_net_init(struct net *net)
3245 struct ctl_table *tbl;
3247 tbl = ipv4_route_flush_table;
3248 if (!net_eq(net, &init_net)) {
3249 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3253 tbl[0].extra1 = net;
3255 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3256 if (net->ipv4.route_hdr == NULL)
3261 if (tbl != ipv4_route_flush_table)
3267 static __net_exit void sysctl_route_net_exit(struct net *net)
3269 struct ctl_table *tbl;
3271 tbl = net->ipv4.route_hdr->ctl_table_arg;
3272 unregister_net_sysctl_table(net->ipv4.route_hdr);
3273 BUG_ON(tbl == ipv4_route_flush_table);
3277 static __net_initdata struct pernet_operations sysctl_route_ops = {
3278 .init = sysctl_route_net_init,
3279 .exit = sysctl_route_net_exit,
3283 static __net_init int rt_genid_init(struct net *net)
3285 get_random_bytes(&net->ipv4.rt_genid,
3286 sizeof(net->ipv4.rt_genid));
3287 get_random_bytes(&net->ipv4.dev_addr_genid,
3288 sizeof(net->ipv4.dev_addr_genid));
3292 static __net_initdata struct pernet_operations rt_genid_ops = {
3293 .init = rt_genid_init,
3296 static int __net_init ipv4_inetpeer_init(struct net *net)
3298 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3302 inet_peer_base_init(bp);
3303 net->ipv4.peers = bp;
3307 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3309 struct inet_peer_base *bp = net->ipv4.peers;
3311 net->ipv4.peers = NULL;
3312 inetpeer_invalidate_tree(bp);
3316 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3317 .init = ipv4_inetpeer_init,
3318 .exit = ipv4_inetpeer_exit,
3321 #ifdef CONFIG_IP_ROUTE_CLASSID
3322 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3323 #endif /* CONFIG_IP_ROUTE_CLASSID */
3325 static __initdata unsigned long rhash_entries;
3326 static int __init set_rhash_entries(char *str)
3333 ret = kstrtoul(str, 0, &rhash_entries);
3339 __setup("rhash_entries=", set_rhash_entries);
3341 int __init ip_rt_init(void)
3345 #ifdef CONFIG_IP_ROUTE_CLASSID
3346 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3348 panic("IP: failed to allocate ip_rt_acct\n");
3351 ipv4_dst_ops.kmem_cachep =
3352 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3353 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3355 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3357 if (dst_entries_init(&ipv4_dst_ops) < 0)
3358 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3360 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3361 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3363 rt_hash_table = (struct rt_hash_bucket *)
3364 alloc_large_system_hash("IP route cache",
3365 sizeof(struct rt_hash_bucket),
3367 (totalram_pages >= 128 * 1024) ?
3373 rhash_entries ? 0 : 512 * 1024);
3374 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3375 rt_hash_lock_init();
3377 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3378 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3383 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3384 expires_ljiffies = jiffies;
3385 schedule_delayed_work(&expires_work,
3386 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3388 if (ip_rt_proc_init())
3389 pr_err("Unable to create route proc files\n");
3392 xfrm4_init(ip_rt_max_size);
3394 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3396 #ifdef CONFIG_SYSCTL
3397 register_pernet_subsys(&sysctl_route_ops);
3399 register_pernet_subsys(&rt_genid_ops);
3400 register_pernet_subsys(&ipv4_inetpeer_ops);
3404 #ifdef CONFIG_SYSCTL
3406 * We really need to sanitize the damn ipv4 init order, then all
3407 * this nonsense will go away.
3409 void __init ip_static_sysctl_init(void)
3411 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);