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, const void *daddr);
193 static struct dst_ops ipv4_dst_ops = {
195 .protocol = cpu_to_be16(ETH_P_IP),
196 .gc = rt_garbage_collect,
197 .check = ipv4_dst_check,
198 .default_advmss = ipv4_default_advmss,
200 .cow_metrics = ipv4_cow_metrics,
201 .destroy = ipv4_dst_destroy,
202 .ifdown = ipv4_dst_ifdown,
203 .negative_advice = ipv4_negative_advice,
204 .link_failure = ipv4_link_failure,
205 .update_pmtu = ip_rt_update_pmtu,
206 .local_out = __ip_local_out,
207 .neigh_lookup = ipv4_neigh_lookup,
210 #define ECN_OR_COST(class) TC_PRIO_##class
212 const __u8 ip_tos2prio[16] = {
214 ECN_OR_COST(BESTEFFORT),
216 ECN_OR_COST(BESTEFFORT),
222 ECN_OR_COST(INTERACTIVE),
224 ECN_OR_COST(INTERACTIVE),
225 TC_PRIO_INTERACTIVE_BULK,
226 ECN_OR_COST(INTERACTIVE_BULK),
227 TC_PRIO_INTERACTIVE_BULK,
228 ECN_OR_COST(INTERACTIVE_BULK)
230 EXPORT_SYMBOL(ip_tos2prio);
236 /* The locking scheme is rather straight forward:
238 * 1) Read-Copy Update protects the buckets of the central route hash.
239 * 2) Only writers remove entries, and they hold the lock
240 * as they look at rtable reference counts.
241 * 3) Only readers acquire references to rtable entries,
242 * they do so with atomic increments and with the
246 struct rt_hash_bucket {
247 struct rtable __rcu *chain;
250 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
251 defined(CONFIG_PROVE_LOCKING)
253 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
254 * The size of this table is a power of two and depends on the number of CPUS.
255 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
257 #ifdef CONFIG_LOCKDEP
258 # define RT_HASH_LOCK_SZ 256
261 # define RT_HASH_LOCK_SZ 4096
263 # define RT_HASH_LOCK_SZ 2048
265 # define RT_HASH_LOCK_SZ 1024
267 # define RT_HASH_LOCK_SZ 512
269 # define RT_HASH_LOCK_SZ 256
273 static spinlock_t *rt_hash_locks;
274 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
276 static __init void rt_hash_lock_init(void)
280 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
283 panic("IP: failed to allocate rt_hash_locks\n");
285 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
286 spin_lock_init(&rt_hash_locks[i]);
289 # define rt_hash_lock_addr(slot) NULL
291 static inline void rt_hash_lock_init(void)
296 static struct rt_hash_bucket *rt_hash_table __read_mostly;
297 static unsigned int rt_hash_mask __read_mostly;
298 static unsigned int rt_hash_log __read_mostly;
300 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
301 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
303 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
306 return jhash_3words((__force u32)daddr, (__force u32)saddr,
311 static inline int rt_genid(struct net *net)
313 return atomic_read(&net->ipv4.rt_genid);
316 #ifdef CONFIG_PROC_FS
317 struct rt_cache_iter_state {
318 struct seq_net_private p;
323 static struct rtable *rt_cache_get_first(struct seq_file *seq)
325 struct rt_cache_iter_state *st = seq->private;
326 struct rtable *r = NULL;
328 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
329 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
332 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
334 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
335 r->rt_genid == st->genid)
337 r = rcu_dereference_bh(r->dst.rt_next);
339 rcu_read_unlock_bh();
344 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
347 struct rt_cache_iter_state *st = seq->private;
349 r = rcu_dereference_bh(r->dst.rt_next);
351 rcu_read_unlock_bh();
353 if (--st->bucket < 0)
355 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
357 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
362 static struct rtable *rt_cache_get_next(struct seq_file *seq,
365 struct rt_cache_iter_state *st = seq->private;
366 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
367 if (dev_net(r->dst.dev) != seq_file_net(seq))
369 if (r->rt_genid == st->genid)
375 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
377 struct rtable *r = rt_cache_get_first(seq);
380 while (pos && (r = rt_cache_get_next(seq, r)))
382 return pos ? NULL : r;
385 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
387 struct rt_cache_iter_state *st = seq->private;
389 return rt_cache_get_idx(seq, *pos - 1);
390 st->genid = rt_genid(seq_file_net(seq));
391 return SEQ_START_TOKEN;
394 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
398 if (v == SEQ_START_TOKEN)
399 r = rt_cache_get_first(seq);
401 r = rt_cache_get_next(seq, v);
406 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
408 if (v && v != SEQ_START_TOKEN)
409 rcu_read_unlock_bh();
412 static int rt_cache_seq_show(struct seq_file *seq, void *v)
414 if (v == SEQ_START_TOKEN)
415 seq_printf(seq, "%-127s\n",
416 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
417 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
420 struct rtable *r = v;
425 n = dst_get_neighbour_noref(&r->dst);
426 HHUptod = (n && (n->nud_state & NUD_CONNECTED)) ? 1 : 0;
429 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
430 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
431 r->dst.dev ? r->dst.dev->name : "*",
432 (__force u32)r->rt_dst,
433 (__force u32)r->rt_gateway,
434 r->rt_flags, atomic_read(&r->dst.__refcnt),
435 r->dst.__use, 0, (__force u32)r->rt_src,
436 dst_metric_advmss(&r->dst) + 40,
437 dst_metric(&r->dst, RTAX_WINDOW),
438 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
439 dst_metric(&r->dst, RTAX_RTTVAR)),
443 r->rt_spec_dst, &len);
445 seq_printf(seq, "%*s\n", 127 - len, "");
450 static const struct seq_operations rt_cache_seq_ops = {
451 .start = rt_cache_seq_start,
452 .next = rt_cache_seq_next,
453 .stop = rt_cache_seq_stop,
454 .show = rt_cache_seq_show,
457 static int rt_cache_seq_open(struct inode *inode, struct file *file)
459 return seq_open_net(inode, file, &rt_cache_seq_ops,
460 sizeof(struct rt_cache_iter_state));
463 static const struct file_operations rt_cache_seq_fops = {
464 .owner = THIS_MODULE,
465 .open = rt_cache_seq_open,
468 .release = seq_release_net,
472 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
477 return SEQ_START_TOKEN;
479 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
480 if (!cpu_possible(cpu))
483 return &per_cpu(rt_cache_stat, cpu);
488 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
492 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
493 if (!cpu_possible(cpu))
496 return &per_cpu(rt_cache_stat, cpu);
502 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
507 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
509 struct rt_cache_stat *st = v;
511 if (v == SEQ_START_TOKEN) {
512 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");
516 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
517 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
518 dst_entries_get_slow(&ipv4_dst_ops),
541 static const struct seq_operations rt_cpu_seq_ops = {
542 .start = rt_cpu_seq_start,
543 .next = rt_cpu_seq_next,
544 .stop = rt_cpu_seq_stop,
545 .show = rt_cpu_seq_show,
549 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
551 return seq_open(file, &rt_cpu_seq_ops);
554 static const struct file_operations rt_cpu_seq_fops = {
555 .owner = THIS_MODULE,
556 .open = rt_cpu_seq_open,
559 .release = seq_release,
562 #ifdef CONFIG_IP_ROUTE_CLASSID
563 static int rt_acct_proc_show(struct seq_file *m, void *v)
565 struct ip_rt_acct *dst, *src;
568 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
572 for_each_possible_cpu(i) {
573 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
574 for (j = 0; j < 256; j++) {
575 dst[j].o_bytes += src[j].o_bytes;
576 dst[j].o_packets += src[j].o_packets;
577 dst[j].i_bytes += src[j].i_bytes;
578 dst[j].i_packets += src[j].i_packets;
582 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
587 static int rt_acct_proc_open(struct inode *inode, struct file *file)
589 return single_open(file, rt_acct_proc_show, NULL);
592 static const struct file_operations rt_acct_proc_fops = {
593 .owner = THIS_MODULE,
594 .open = rt_acct_proc_open,
597 .release = single_release,
601 static int __net_init ip_rt_do_proc_init(struct net *net)
603 struct proc_dir_entry *pde;
605 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
610 pde = proc_create("rt_cache", S_IRUGO,
611 net->proc_net_stat, &rt_cpu_seq_fops);
615 #ifdef CONFIG_IP_ROUTE_CLASSID
616 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
622 #ifdef CONFIG_IP_ROUTE_CLASSID
624 remove_proc_entry("rt_cache", net->proc_net_stat);
627 remove_proc_entry("rt_cache", net->proc_net);
632 static void __net_exit ip_rt_do_proc_exit(struct net *net)
634 remove_proc_entry("rt_cache", net->proc_net_stat);
635 remove_proc_entry("rt_cache", net->proc_net);
636 #ifdef CONFIG_IP_ROUTE_CLASSID
637 remove_proc_entry("rt_acct", net->proc_net);
641 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
642 .init = ip_rt_do_proc_init,
643 .exit = ip_rt_do_proc_exit,
646 static int __init ip_rt_proc_init(void)
648 return register_pernet_subsys(&ip_rt_proc_ops);
652 static inline int ip_rt_proc_init(void)
656 #endif /* CONFIG_PROC_FS */
658 static inline void rt_free(struct rtable *rt)
660 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
663 static inline void rt_drop(struct rtable *rt)
666 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
669 static inline int rt_fast_clean(struct rtable *rth)
671 /* Kill broadcast/multicast entries very aggresively, if they
672 collide in hash table with more useful entries */
673 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
674 rt_is_input_route(rth) && rth->dst.rt_next;
677 static inline int rt_valuable(struct rtable *rth)
679 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
680 (rt_has_peer(rth) && rt_peer_ptr(rth)->pmtu_expires);
683 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
688 if (atomic_read(&rth->dst.__refcnt))
691 age = jiffies - rth->dst.lastuse;
692 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
693 (age <= tmo2 && rt_valuable(rth)))
699 /* Bits of score are:
701 * 30: not quite useless
702 * 29..0: usage counter
704 static inline u32 rt_score(struct rtable *rt)
706 u32 score = jiffies - rt->dst.lastuse;
708 score = ~score & ~(3<<30);
713 if (rt_is_output_route(rt) ||
714 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
720 static inline bool rt_caching(const struct net *net)
722 return net->ipv4.current_rt_cache_rebuild_count <=
723 net->ipv4.sysctl_rt_cache_rebuild_count;
726 static inline bool compare_hash_inputs(const struct rtable *rt1,
727 const struct rtable *rt2)
729 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
730 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
731 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
734 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
736 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
737 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
738 (rt1->rt_mark ^ rt2->rt_mark) |
739 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
740 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
741 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
744 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
746 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
749 static inline int rt_is_expired(struct rtable *rth)
751 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
755 * Perform a full scan of hash table and free all entries.
756 * Can be called by a softirq or a process.
757 * In the later case, we want to be reschedule if necessary
759 static void rt_do_flush(struct net *net, int process_context)
762 struct rtable *rth, *next;
764 for (i = 0; i <= rt_hash_mask; i++) {
765 struct rtable __rcu **pprev;
768 if (process_context && need_resched())
770 rth = rcu_access_pointer(rt_hash_table[i].chain);
774 spin_lock_bh(rt_hash_lock_addr(i));
777 pprev = &rt_hash_table[i].chain;
778 rth = rcu_dereference_protected(*pprev,
779 lockdep_is_held(rt_hash_lock_addr(i)));
782 next = rcu_dereference_protected(rth->dst.rt_next,
783 lockdep_is_held(rt_hash_lock_addr(i)));
786 net_eq(dev_net(rth->dst.dev), net)) {
787 rcu_assign_pointer(*pprev, next);
788 rcu_assign_pointer(rth->dst.rt_next, list);
791 pprev = &rth->dst.rt_next;
796 spin_unlock_bh(rt_hash_lock_addr(i));
798 for (; list; list = next) {
799 next = rcu_dereference_protected(list->dst.rt_next, 1);
806 * While freeing expired entries, we compute average chain length
807 * and standard deviation, using fixed-point arithmetic.
808 * This to have an estimation of rt_chain_length_max
809 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
810 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
814 #define ONE (1UL << FRACT_BITS)
817 * Given a hash chain and an item in this hash chain,
818 * find if a previous entry has the same hash_inputs
819 * (but differs on tos, mark or oif)
820 * Returns 0 if an alias is found.
821 * Returns ONE if rth has no alias before itself.
823 static int has_noalias(const struct rtable *head, const struct rtable *rth)
825 const struct rtable *aux = head;
828 if (compare_hash_inputs(aux, rth))
830 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
835 static void rt_check_expire(void)
837 static unsigned int rover;
838 unsigned int i = rover, goal;
840 struct rtable __rcu **rthp;
841 unsigned long samples = 0;
842 unsigned long sum = 0, sum2 = 0;
846 delta = jiffies - expires_ljiffies;
847 expires_ljiffies = jiffies;
848 mult = ((u64)delta) << rt_hash_log;
849 if (ip_rt_gc_timeout > 1)
850 do_div(mult, ip_rt_gc_timeout);
851 goal = (unsigned int)mult;
852 if (goal > rt_hash_mask)
853 goal = rt_hash_mask + 1;
854 for (; goal > 0; goal--) {
855 unsigned long tmo = ip_rt_gc_timeout;
856 unsigned long length;
858 i = (i + 1) & rt_hash_mask;
859 rthp = &rt_hash_table[i].chain;
866 if (rcu_dereference_raw(*rthp) == NULL)
869 spin_lock_bh(rt_hash_lock_addr(i));
870 while ((rth = rcu_dereference_protected(*rthp,
871 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
872 prefetch(rth->dst.rt_next);
873 if (rt_is_expired(rth)) {
874 *rthp = rth->dst.rt_next;
878 if (rth->dst.expires) {
879 /* Entry is expired even if it is in use */
880 if (time_before_eq(jiffies, rth->dst.expires)) {
883 rthp = &rth->dst.rt_next;
885 * We only count entries on
886 * a chain with equal hash inputs once
887 * so that entries for different QOS
888 * levels, and other non-hash input
889 * attributes don't unfairly skew
890 * the length computation
892 length += has_noalias(rt_hash_table[i].chain, rth);
895 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
898 /* Cleanup aged off entries. */
899 *rthp = rth->dst.rt_next;
902 spin_unlock_bh(rt_hash_lock_addr(i));
904 sum2 += length*length;
907 unsigned long avg = sum / samples;
908 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
909 rt_chain_length_max = max_t(unsigned long,
911 (avg + 4*sd) >> FRACT_BITS);
917 * rt_worker_func() is run in process context.
918 * we call rt_check_expire() to scan part of the hash table
920 static void rt_worker_func(struct work_struct *work)
923 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
927 * Perturbation of rt_genid by a small quantity [1..256]
928 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
929 * many times (2^24) without giving recent rt_genid.
930 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
932 static void rt_cache_invalidate(struct net *net)
934 unsigned char shuffle;
936 get_random_bytes(&shuffle, sizeof(shuffle));
937 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
938 inetpeer_invalidate_family(AF_INET);
942 * delay < 0 : invalidate cache (fast : entries will be deleted later)
943 * delay >= 0 : invalidate & flush cache (can be long)
945 void rt_cache_flush(struct net *net, int delay)
947 rt_cache_invalidate(net);
949 rt_do_flush(net, !in_softirq());
952 /* Flush previous cache invalidated entries from the cache */
953 void rt_cache_flush_batch(struct net *net)
955 rt_do_flush(net, !in_softirq());
958 static void rt_emergency_hash_rebuild(struct net *net)
960 net_warn_ratelimited("Route hash chain too long!\n");
961 rt_cache_invalidate(net);
965 Short description of GC goals.
967 We want to build algorithm, which will keep routing cache
968 at some equilibrium point, when number of aged off entries
969 is kept approximately equal to newly generated ones.
971 Current expiration strength is variable "expire".
972 We try to adjust it dynamically, so that if networking
973 is idle expires is large enough to keep enough of warm entries,
974 and when load increases it reduces to limit cache size.
977 static int rt_garbage_collect(struct dst_ops *ops)
979 static unsigned long expire = RT_GC_TIMEOUT;
980 static unsigned long last_gc;
982 static int equilibrium;
984 struct rtable __rcu **rthp;
985 unsigned long now = jiffies;
987 int entries = dst_entries_get_fast(&ipv4_dst_ops);
990 * Garbage collection is pretty expensive,
991 * do not make it too frequently.
994 RT_CACHE_STAT_INC(gc_total);
996 if (now - last_gc < ip_rt_gc_min_interval &&
997 entries < ip_rt_max_size) {
998 RT_CACHE_STAT_INC(gc_ignored);
1002 entries = dst_entries_get_slow(&ipv4_dst_ops);
1003 /* Calculate number of entries, which we want to expire now. */
1004 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
1006 if (equilibrium < ipv4_dst_ops.gc_thresh)
1007 equilibrium = ipv4_dst_ops.gc_thresh;
1008 goal = entries - equilibrium;
1010 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1011 goal = entries - equilibrium;
1014 /* We are in dangerous area. Try to reduce cache really
1017 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1018 equilibrium = entries - goal;
1021 if (now - last_gc >= ip_rt_gc_min_interval)
1025 equilibrium += goal;
1032 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1033 unsigned long tmo = expire;
1035 k = (k + 1) & rt_hash_mask;
1036 rthp = &rt_hash_table[k].chain;
1037 spin_lock_bh(rt_hash_lock_addr(k));
1038 while ((rth = rcu_dereference_protected(*rthp,
1039 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1040 if (!rt_is_expired(rth) &&
1041 !rt_may_expire(rth, tmo, expire)) {
1043 rthp = &rth->dst.rt_next;
1046 *rthp = rth->dst.rt_next;
1050 spin_unlock_bh(rt_hash_lock_addr(k));
1059 /* Goal is not achieved. We stop process if:
1061 - if expire reduced to zero. Otherwise, expire is halfed.
1062 - if table is not full.
1063 - if we are called from interrupt.
1064 - jiffies check is just fallback/debug loop breaker.
1065 We will not spin here for long time in any case.
1068 RT_CACHE_STAT_INC(gc_goal_miss);
1075 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1077 } while (!in_softirq() && time_before_eq(jiffies, now));
1079 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1081 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1083 net_warn_ratelimited("dst cache overflow\n");
1084 RT_CACHE_STAT_INC(gc_dst_overflow);
1088 expire += ip_rt_gc_min_interval;
1089 if (expire > ip_rt_gc_timeout ||
1090 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1091 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1092 expire = ip_rt_gc_timeout;
1097 * Returns number of entries in a hash chain that have different hash_inputs
1099 static int slow_chain_length(const struct rtable *head)
1102 const struct rtable *rth = head;
1105 length += has_noalias(head, rth);
1106 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1108 return length >> FRACT_BITS;
1111 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr)
1113 static const __be32 inaddr_any = 0;
1114 struct net_device *dev = dst->dev;
1115 const __be32 *pkey = daddr;
1116 const struct rtable *rt;
1117 struct neighbour *n;
1119 rt = (const struct rtable *) dst;
1121 if (dev->flags & (IFF_LOOPBACK | IFF_POINTOPOINT))
1123 else if (rt->rt_gateway)
1124 pkey = (const __be32 *) &rt->rt_gateway;
1126 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1129 return neigh_create(&arp_tbl, pkey, dev);
1132 static int rt_bind_neighbour(struct rtable *rt)
1134 struct neighbour *n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1137 dst_set_neighbour(&rt->dst, n);
1142 static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1143 struct sk_buff *skb, int ifindex)
1145 struct rtable *rth, *cand;
1146 struct rtable __rcu **rthp, **candp;
1150 int attempts = !in_softirq();
1154 min_score = ~(u32)0;
1159 if (!rt_caching(dev_net(rt->dst.dev))) {
1161 * If we're not caching, just tell the caller we
1162 * were successful and don't touch the route. The
1163 * caller hold the sole reference to the cache entry, and
1164 * it will be released when the caller is done with it.
1165 * If we drop it here, the callers have no way to resolve routes
1166 * when we're not caching. Instead, just point *rp at rt, so
1167 * the caller gets a single use out of the route
1168 * Note that we do rt_free on this new route entry, so that
1169 * once its refcount hits zero, we are still able to reap it
1171 * Note: To avoid expensive rcu stuff for this uncached dst,
1172 * we set DST_NOCACHE so that dst_release() can free dst without
1173 * waiting a grace period.
1176 rt->dst.flags |= DST_NOCACHE;
1177 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1178 int err = rt_bind_neighbour(rt);
1180 net_warn_ratelimited("Neighbour table failure & not caching routes\n");
1182 return ERR_PTR(err);
1189 rthp = &rt_hash_table[hash].chain;
1191 spin_lock_bh(rt_hash_lock_addr(hash));
1192 while ((rth = rcu_dereference_protected(*rthp,
1193 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1194 if (rt_is_expired(rth)) {
1195 *rthp = rth->dst.rt_next;
1199 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1201 *rthp = rth->dst.rt_next;
1203 * Since lookup is lockfree, the deletion
1204 * must be visible to another weakly ordered CPU before
1205 * the insertion at the start of the hash chain.
1207 rcu_assign_pointer(rth->dst.rt_next,
1208 rt_hash_table[hash].chain);
1210 * Since lookup is lockfree, the update writes
1211 * must be ordered for consistency on SMP.
1213 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1215 dst_use(&rth->dst, now);
1216 spin_unlock_bh(rt_hash_lock_addr(hash));
1220 skb_dst_set(skb, &rth->dst);
1224 if (!atomic_read(&rth->dst.__refcnt)) {
1225 u32 score = rt_score(rth);
1227 if (score <= min_score) {
1236 rthp = &rth->dst.rt_next;
1240 /* ip_rt_gc_elasticity used to be average length of chain
1241 * length, when exceeded gc becomes really aggressive.
1243 * The second limit is less certain. At the moment it allows
1244 * only 2 entries per bucket. We will see.
1246 if (chain_length > ip_rt_gc_elasticity) {
1247 *candp = cand->dst.rt_next;
1251 if (chain_length > rt_chain_length_max &&
1252 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1253 struct net *net = dev_net(rt->dst.dev);
1254 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1255 if (!rt_caching(net)) {
1256 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1257 rt->dst.dev->name, num);
1259 rt_emergency_hash_rebuild(net);
1260 spin_unlock_bh(rt_hash_lock_addr(hash));
1262 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1263 ifindex, rt_genid(net));
1268 /* Try to bind route to arp only if it is output
1269 route or unicast forwarding path.
1271 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1272 int err = rt_bind_neighbour(rt);
1274 spin_unlock_bh(rt_hash_lock_addr(hash));
1276 if (err != -ENOBUFS) {
1278 return ERR_PTR(err);
1281 /* Neighbour tables are full and nothing
1282 can be released. Try to shrink route cache,
1283 it is most likely it holds some neighbour records.
1285 if (attempts-- > 0) {
1286 int saved_elasticity = ip_rt_gc_elasticity;
1287 int saved_int = ip_rt_gc_min_interval;
1288 ip_rt_gc_elasticity = 1;
1289 ip_rt_gc_min_interval = 0;
1290 rt_garbage_collect(&ipv4_dst_ops);
1291 ip_rt_gc_min_interval = saved_int;
1292 ip_rt_gc_elasticity = saved_elasticity;
1296 net_warn_ratelimited("Neighbour table overflow\n");
1298 return ERR_PTR(-ENOBUFS);
1302 rt->dst.rt_next = rt_hash_table[hash].chain;
1305 * Since lookup is lockfree, we must make sure
1306 * previous writes to rt are committed to memory
1307 * before making rt visible to other CPUS.
1309 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1311 spin_unlock_bh(rt_hash_lock_addr(hash));
1315 skb_dst_set(skb, &rt->dst);
1319 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1321 static u32 rt_peer_genid(void)
1323 return atomic_read(&__rt_peer_genid);
1326 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1328 struct inet_peer_base *base;
1329 struct inet_peer *peer;
1331 base = inetpeer_base_ptr(rt->_peer);
1335 peer = inet_getpeer_v4(base, daddr, create);
1337 if (!rt_set_peer(rt, peer))
1340 rt->rt_peer_genid = rt_peer_genid();
1345 * Peer allocation may fail only in serious out-of-memory conditions. However
1346 * we still can generate some output.
1347 * Random ID selection looks a bit dangerous because we have no chances to
1348 * select ID being unique in a reasonable period of time.
1349 * But broken packet identifier may be better than no packet at all.
1351 static void ip_select_fb_ident(struct iphdr *iph)
1353 static DEFINE_SPINLOCK(ip_fb_id_lock);
1354 static u32 ip_fallback_id;
1357 spin_lock_bh(&ip_fb_id_lock);
1358 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1359 iph->id = htons(salt & 0xFFFF);
1360 ip_fallback_id = salt;
1361 spin_unlock_bh(&ip_fb_id_lock);
1364 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1366 struct rtable *rt = (struct rtable *) dst;
1368 if (rt && !(rt->dst.flags & DST_NOPEER)) {
1369 struct inet_peer *peer = rt_get_peer_create(rt, rt->rt_dst);
1371 /* If peer is attached to destination, it is never detached,
1372 so that we need not to grab a lock to dereference it.
1375 iph->id = htons(inet_getid(peer, more));
1379 pr_debug("rt_bind_peer(0) @%p\n", __builtin_return_address(0));
1381 ip_select_fb_ident(iph);
1383 EXPORT_SYMBOL(__ip_select_ident);
1385 static void rt_del(unsigned int hash, struct rtable *rt)
1387 struct rtable __rcu **rthp;
1390 rthp = &rt_hash_table[hash].chain;
1391 spin_lock_bh(rt_hash_lock_addr(hash));
1393 while ((aux = rcu_dereference_protected(*rthp,
1394 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1395 if (aux == rt || rt_is_expired(aux)) {
1396 *rthp = aux->dst.rt_next;
1400 rthp = &aux->dst.rt_next;
1402 spin_unlock_bh(rt_hash_lock_addr(hash));
1405 static void check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1407 struct rtable *rt = (struct rtable *) dst;
1408 __be32 orig_gw = rt->rt_gateway;
1409 struct neighbour *n, *old_n;
1411 dst_confirm(&rt->dst);
1413 rt->rt_gateway = peer->redirect_learned.a4;
1415 n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1417 rt->rt_gateway = orig_gw;
1420 old_n = xchg(&rt->dst._neighbour, n);
1422 neigh_release(old_n);
1423 if (!(n->nud_state & NUD_VALID)) {
1424 neigh_event_send(n, NULL);
1426 rt->rt_flags |= RTCF_REDIRECTED;
1427 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1431 /* called in rcu_read_lock() section */
1432 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1433 __be32 saddr, struct net_device *dev)
1436 struct in_device *in_dev = __in_dev_get_rcu(dev);
1437 __be32 skeys[2] = { saddr, 0 };
1438 int ikeys[2] = { dev->ifindex, 0 };
1439 struct inet_peer *peer;
1446 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1447 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1448 ipv4_is_zeronet(new_gw))
1449 goto reject_redirect;
1451 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1452 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1453 goto reject_redirect;
1454 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1455 goto reject_redirect;
1457 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1458 goto reject_redirect;
1461 for (s = 0; s < 2; s++) {
1462 for (i = 0; i < 2; i++) {
1464 struct rtable __rcu **rthp;
1467 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1469 rthp = &rt_hash_table[hash].chain;
1471 while ((rt = rcu_dereference(*rthp)) != NULL) {
1472 rthp = &rt->dst.rt_next;
1474 if (rt->rt_key_dst != daddr ||
1475 rt->rt_key_src != skeys[s] ||
1476 rt->rt_oif != ikeys[i] ||
1477 rt_is_input_route(rt) ||
1478 rt_is_expired(rt) ||
1479 !net_eq(dev_net(rt->dst.dev), net) ||
1481 rt->dst.dev != dev ||
1482 rt->rt_gateway != old_gw)
1485 peer = rt_get_peer_create(rt, rt->rt_dst);
1487 if (peer->redirect_learned.a4 != new_gw) {
1488 peer->redirect_learned.a4 = new_gw;
1489 atomic_inc(&__rt_peer_genid);
1491 check_peer_redir(&rt->dst, peer);
1499 #ifdef CONFIG_IP_ROUTE_VERBOSE
1500 if (IN_DEV_LOG_MARTIANS(in_dev))
1501 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1502 " Advised path = %pI4 -> %pI4\n",
1503 &old_gw, dev->name, &new_gw,
1509 static bool peer_pmtu_expired(struct inet_peer *peer)
1511 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1514 time_after_eq(jiffies, orig) &&
1515 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1518 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1520 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1523 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1526 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1528 struct rtable *rt = (struct rtable *)dst;
1529 struct dst_entry *ret = dst;
1532 if (dst->obsolete > 0) {
1535 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1536 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1538 rt_genid(dev_net(dst->dev)));
1541 } else if (rt_has_peer(rt)) {
1542 struct inet_peer *peer = rt_peer_ptr(rt);
1543 if (peer_pmtu_expired(peer))
1544 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1552 * 1. The first ip_rt_redirect_number redirects are sent
1553 * with exponential backoff, then we stop sending them at all,
1554 * assuming that the host ignores our redirects.
1555 * 2. If we did not see packets requiring redirects
1556 * during ip_rt_redirect_silence, we assume that the host
1557 * forgot redirected route and start to send redirects again.
1559 * This algorithm is much cheaper and more intelligent than dumb load limiting
1562 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1563 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1566 void ip_rt_send_redirect(struct sk_buff *skb)
1568 struct rtable *rt = skb_rtable(skb);
1569 struct in_device *in_dev;
1570 struct inet_peer *peer;
1574 in_dev = __in_dev_get_rcu(rt->dst.dev);
1575 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1579 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1582 peer = rt_get_peer_create(rt, rt->rt_dst);
1584 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1588 /* No redirected packets during ip_rt_redirect_silence;
1589 * reset the algorithm.
1591 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1592 peer->rate_tokens = 0;
1594 /* Too many ignored redirects; do not send anything
1595 * set dst.rate_last to the last seen redirected packet.
1597 if (peer->rate_tokens >= ip_rt_redirect_number) {
1598 peer->rate_last = jiffies;
1602 /* Check for load limit; set rate_last to the latest sent
1605 if (peer->rate_tokens == 0 ||
1608 (ip_rt_redirect_load << peer->rate_tokens)))) {
1609 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1610 peer->rate_last = jiffies;
1611 ++peer->rate_tokens;
1612 #ifdef CONFIG_IP_ROUTE_VERBOSE
1614 peer->rate_tokens == ip_rt_redirect_number)
1615 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1616 &ip_hdr(skb)->saddr, rt->rt_iif,
1617 &rt->rt_dst, &rt->rt_gateway);
1622 static int ip_error(struct sk_buff *skb)
1624 struct rtable *rt = skb_rtable(skb);
1625 struct inet_peer *peer;
1630 switch (rt->dst.error) {
1635 code = ICMP_HOST_UNREACH;
1638 code = ICMP_NET_UNREACH;
1639 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1640 IPSTATS_MIB_INNOROUTES);
1643 code = ICMP_PKT_FILTERED;
1647 peer = rt_get_peer_create(rt, rt->rt_dst);
1652 peer->rate_tokens += now - peer->rate_last;
1653 if (peer->rate_tokens > ip_rt_error_burst)
1654 peer->rate_tokens = ip_rt_error_burst;
1655 peer->rate_last = now;
1656 if (peer->rate_tokens >= ip_rt_error_cost)
1657 peer->rate_tokens -= ip_rt_error_cost;
1662 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1664 out: kfree_skb(skb);
1668 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1670 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1674 if (time_before(jiffies, expires)) {
1675 u32 orig_dst_mtu = dst_mtu(dst);
1676 if (peer->pmtu_learned < orig_dst_mtu) {
1677 if (!peer->pmtu_orig)
1678 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1679 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1681 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1682 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1685 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1687 struct rtable *rt = (struct rtable *) dst;
1688 struct inet_peer *peer;
1692 peer = rt_get_peer_create(rt, rt->rt_dst);
1694 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1696 if (mtu < ip_rt_min_pmtu)
1697 mtu = ip_rt_min_pmtu;
1698 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1700 pmtu_expires = jiffies + ip_rt_mtu_expires;
1704 peer->pmtu_learned = mtu;
1705 peer->pmtu_expires = pmtu_expires;
1707 atomic_inc(&__rt_peer_genid);
1708 rt->rt_peer_genid = rt_peer_genid();
1710 check_peer_pmtu(dst, peer);
1715 static void ipv4_validate_peer(struct rtable *rt)
1717 if (rt->rt_peer_genid != rt_peer_genid()) {
1718 struct inet_peer *peer = rt_get_peer(rt, rt->rt_dst);
1721 check_peer_pmtu(&rt->dst, peer);
1723 if (peer->redirect_learned.a4 &&
1724 peer->redirect_learned.a4 != rt->rt_gateway)
1725 check_peer_redir(&rt->dst, peer);
1728 rt->rt_peer_genid = rt_peer_genid();
1732 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1734 struct rtable *rt = (struct rtable *) dst;
1736 if (rt_is_expired(rt))
1738 ipv4_validate_peer(rt);
1742 static void ipv4_dst_destroy(struct dst_entry *dst)
1744 struct rtable *rt = (struct rtable *) dst;
1747 fib_info_put(rt->fi);
1750 if (rt_has_peer(rt)) {
1751 struct inet_peer *peer = rt_peer_ptr(rt);
1757 static void ipv4_link_failure(struct sk_buff *skb)
1761 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1763 rt = skb_rtable(skb);
1764 if (rt && rt_has_peer(rt)) {
1765 struct inet_peer *peer = rt_peer_ptr(rt);
1766 if (peer_pmtu_cleaned(peer))
1767 dst_metric_set(&rt->dst, RTAX_MTU, peer->pmtu_orig);
1771 static int ip_rt_bug(struct sk_buff *skb)
1773 pr_debug("%s: %pI4 -> %pI4, %s\n",
1774 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1775 skb->dev ? skb->dev->name : "?");
1782 We do not cache source address of outgoing interface,
1783 because it is used only by IP RR, TS and SRR options,
1784 so that it out of fast path.
1786 BTW remember: "addr" is allowed to be not aligned
1790 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1794 if (rt_is_output_route(rt))
1795 src = ip_hdr(skb)->saddr;
1797 struct fib_result res;
1803 memset(&fl4, 0, sizeof(fl4));
1804 fl4.daddr = iph->daddr;
1805 fl4.saddr = iph->saddr;
1806 fl4.flowi4_tos = RT_TOS(iph->tos);
1807 fl4.flowi4_oif = rt->dst.dev->ifindex;
1808 fl4.flowi4_iif = skb->dev->ifindex;
1809 fl4.flowi4_mark = skb->mark;
1812 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1813 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1815 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1819 memcpy(addr, &src, 4);
1822 #ifdef CONFIG_IP_ROUTE_CLASSID
1823 static void set_class_tag(struct rtable *rt, u32 tag)
1825 if (!(rt->dst.tclassid & 0xFFFF))
1826 rt->dst.tclassid |= tag & 0xFFFF;
1827 if (!(rt->dst.tclassid & 0xFFFF0000))
1828 rt->dst.tclassid |= tag & 0xFFFF0000;
1832 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1834 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1837 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1839 if (advmss > 65535 - 40)
1840 advmss = 65535 - 40;
1845 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1847 const struct rtable *rt = (const struct rtable *) dst;
1848 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1850 if (mtu && rt_is_output_route(rt))
1853 mtu = dst->dev->mtu;
1855 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1857 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1861 if (mtu > IP_MAX_MTU)
1867 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1868 struct fib_info *fi)
1870 struct inet_peer_base *base;
1871 struct inet_peer *peer;
1874 /* If a peer entry exists for this destination, we must hook
1875 * it up in order to get at cached metrics.
1877 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1880 base = inetpeer_base_ptr(rt->_peer);
1883 peer = inet_getpeer_v4(base, rt->rt_dst, create);
1885 __rt_set_peer(rt, peer);
1886 rt->rt_peer_genid = rt_peer_genid();
1887 if (inet_metrics_new(peer))
1888 memcpy(peer->metrics, fi->fib_metrics,
1889 sizeof(u32) * RTAX_MAX);
1890 dst_init_metrics(&rt->dst, peer->metrics, false);
1892 check_peer_pmtu(&rt->dst, peer);
1894 if (peer->redirect_learned.a4 &&
1895 peer->redirect_learned.a4 != rt->rt_gateway) {
1896 rt->rt_gateway = peer->redirect_learned.a4;
1897 rt->rt_flags |= RTCF_REDIRECTED;
1900 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1902 atomic_inc(&fi->fib_clntref);
1904 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1908 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1909 const struct fib_result *res,
1910 struct fib_info *fi, u16 type, u32 itag)
1912 struct dst_entry *dst = &rt->dst;
1915 if (FIB_RES_GW(*res) &&
1916 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1917 rt->rt_gateway = FIB_RES_GW(*res);
1918 rt_init_metrics(rt, fl4, fi);
1919 #ifdef CONFIG_IP_ROUTE_CLASSID
1920 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1924 if (dst_mtu(dst) > IP_MAX_MTU)
1925 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1926 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1927 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1929 #ifdef CONFIG_IP_ROUTE_CLASSID
1930 #ifdef CONFIG_IP_MULTIPLE_TABLES
1931 set_class_tag(rt, fib_rules_tclass(res));
1933 set_class_tag(rt, itag);
1937 static struct rtable *rt_dst_alloc(struct net_device *dev,
1938 bool nopolicy, bool noxfrm)
1940 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1942 (nopolicy ? DST_NOPOLICY : 0) |
1943 (noxfrm ? DST_NOXFRM : 0));
1946 /* called in rcu_read_lock() section */
1947 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1948 u8 tos, struct net_device *dev, int our)
1953 struct in_device *in_dev = __in_dev_get_rcu(dev);
1957 /* Primary sanity checks. */
1962 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1963 skb->protocol != htons(ETH_P_IP))
1966 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1967 if (ipv4_is_loopback(saddr))
1970 if (ipv4_is_zeronet(saddr)) {
1971 if (!ipv4_is_local_multicast(daddr))
1973 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1975 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
1980 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1981 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1985 #ifdef CONFIG_IP_ROUTE_CLASSID
1986 rth->dst.tclassid = itag;
1988 rth->dst.output = ip_rt_bug;
1990 rth->rt_key_dst = daddr;
1991 rth->rt_key_src = saddr;
1992 rth->rt_genid = rt_genid(dev_net(dev));
1993 rth->rt_flags = RTCF_MULTICAST;
1994 rth->rt_type = RTN_MULTICAST;
1995 rth->rt_key_tos = tos;
1996 rth->rt_dst = daddr;
1997 rth->rt_src = saddr;
1998 rth->rt_route_iif = dev->ifindex;
1999 rth->rt_iif = dev->ifindex;
2001 rth->rt_mark = skb->mark;
2002 rth->rt_gateway = daddr;
2003 rth->rt_spec_dst= spec_dst;
2004 rth->rt_peer_genid = 0;
2005 rt_init_peer(rth, dev_net(dev)->ipv4.peers);
2008 rth->dst.input= ip_local_deliver;
2009 rth->rt_flags |= RTCF_LOCAL;
2012 #ifdef CONFIG_IP_MROUTE
2013 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
2014 rth->dst.input = ip_mr_input;
2016 RT_CACHE_STAT_INC(in_slow_mc);
2018 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
2019 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
2020 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
2031 static void ip_handle_martian_source(struct net_device *dev,
2032 struct in_device *in_dev,
2033 struct sk_buff *skb,
2037 RT_CACHE_STAT_INC(in_martian_src);
2038 #ifdef CONFIG_IP_ROUTE_VERBOSE
2039 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2041 * RFC1812 recommendation, if source is martian,
2042 * the only hint is MAC header.
2044 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
2045 &daddr, &saddr, dev->name);
2046 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2047 print_hex_dump(KERN_WARNING, "ll header: ",
2048 DUMP_PREFIX_OFFSET, 16, 1,
2049 skb_mac_header(skb),
2050 dev->hard_header_len, true);
2056 /* called in rcu_read_lock() section */
2057 static int __mkroute_input(struct sk_buff *skb,
2058 const struct fib_result *res,
2059 struct in_device *in_dev,
2060 __be32 daddr, __be32 saddr, u32 tos,
2061 struct rtable **result)
2065 struct in_device *out_dev;
2066 unsigned int flags = 0;
2070 /* get a working reference to the output device */
2071 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2072 if (out_dev == NULL) {
2073 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
2078 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2079 in_dev->dev, &spec_dst, &itag);
2081 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2088 flags |= RTCF_DIRECTSRC;
2090 if (out_dev == in_dev && err &&
2091 (IN_DEV_SHARED_MEDIA(out_dev) ||
2092 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2093 flags |= RTCF_DOREDIRECT;
2095 if (skb->protocol != htons(ETH_P_IP)) {
2096 /* Not IP (i.e. ARP). Do not create route, if it is
2097 * invalid for proxy arp. DNAT routes are always valid.
2099 * Proxy arp feature have been extended to allow, ARP
2100 * replies back to the same interface, to support
2101 * Private VLAN switch technologies. See arp.c.
2103 if (out_dev == in_dev &&
2104 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2110 rth = rt_dst_alloc(out_dev->dev,
2111 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2112 IN_DEV_CONF_GET(out_dev, NOXFRM));
2118 rth->rt_key_dst = daddr;
2119 rth->rt_key_src = saddr;
2120 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2121 rth->rt_flags = flags;
2122 rth->rt_type = res->type;
2123 rth->rt_key_tos = tos;
2124 rth->rt_dst = daddr;
2125 rth->rt_src = saddr;
2126 rth->rt_route_iif = in_dev->dev->ifindex;
2127 rth->rt_iif = in_dev->dev->ifindex;
2129 rth->rt_mark = skb->mark;
2130 rth->rt_gateway = daddr;
2131 rth->rt_spec_dst= spec_dst;
2132 rth->rt_peer_genid = 0;
2133 rt_init_peer(rth, &res->table->tb_peers);
2136 rth->dst.input = ip_forward;
2137 rth->dst.output = ip_output;
2139 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2147 static int ip_mkroute_input(struct sk_buff *skb,
2148 struct fib_result *res,
2149 const struct flowi4 *fl4,
2150 struct in_device *in_dev,
2151 __be32 daddr, __be32 saddr, u32 tos)
2153 struct rtable *rth = NULL;
2157 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2158 if (res->fi && res->fi->fib_nhs > 1)
2159 fib_select_multipath(res);
2162 /* create a routing cache entry */
2163 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2167 /* put it into the cache */
2168 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2169 rt_genid(dev_net(rth->dst.dev)));
2170 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2172 return PTR_ERR(rth);
2177 * NOTE. We drop all the packets that has local source
2178 * addresses, because every properly looped back packet
2179 * must have correct destination already attached by output routine.
2181 * Such approach solves two big problems:
2182 * 1. Not simplex devices are handled properly.
2183 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2184 * called with rcu_read_lock()
2187 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2188 u8 tos, struct net_device *dev)
2190 struct fib_result res;
2191 struct in_device *in_dev = __in_dev_get_rcu(dev);
2193 unsigned int flags = 0;
2199 struct net *net = dev_net(dev);
2201 /* IP on this device is disabled. */
2206 /* Check for the most weird martians, which can be not detected
2210 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2211 goto martian_source;
2213 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2216 /* Accept zero addresses only to limited broadcast;
2217 * I even do not know to fix it or not. Waiting for complains :-)
2219 if (ipv4_is_zeronet(saddr))
2220 goto martian_source;
2222 if (ipv4_is_zeronet(daddr))
2223 goto martian_destination;
2225 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
2226 if (ipv4_is_loopback(daddr))
2227 goto martian_destination;
2229 if (ipv4_is_loopback(saddr))
2230 goto martian_source;
2234 * Now we are ready to route packet.
2237 fl4.flowi4_iif = dev->ifindex;
2238 fl4.flowi4_mark = skb->mark;
2239 fl4.flowi4_tos = tos;
2240 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2243 err = fib_lookup(net, &fl4, &res);
2245 if (!IN_DEV_FORWARD(in_dev))
2250 RT_CACHE_STAT_INC(in_slow_tot);
2252 if (res.type == RTN_BROADCAST)
2255 if (res.type == RTN_LOCAL) {
2256 err = fib_validate_source(skb, saddr, daddr, tos,
2257 net->loopback_dev->ifindex,
2258 dev, &spec_dst, &itag);
2260 goto martian_source_keep_err;
2262 flags |= RTCF_DIRECTSRC;
2267 if (!IN_DEV_FORWARD(in_dev))
2269 if (res.type != RTN_UNICAST)
2270 goto martian_destination;
2272 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2276 if (skb->protocol != htons(ETH_P_IP))
2279 if (ipv4_is_zeronet(saddr))
2280 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2282 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2285 goto martian_source_keep_err;
2287 flags |= RTCF_DIRECTSRC;
2289 flags |= RTCF_BROADCAST;
2290 res.type = RTN_BROADCAST;
2291 RT_CACHE_STAT_INC(in_brd);
2294 rth = rt_dst_alloc(net->loopback_dev,
2295 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2299 rth->dst.input= ip_local_deliver;
2300 rth->dst.output= ip_rt_bug;
2301 #ifdef CONFIG_IP_ROUTE_CLASSID
2302 rth->dst.tclassid = itag;
2305 rth->rt_key_dst = daddr;
2306 rth->rt_key_src = saddr;
2307 rth->rt_genid = rt_genid(net);
2308 rth->rt_flags = flags|RTCF_LOCAL;
2309 rth->rt_type = res.type;
2310 rth->rt_key_tos = tos;
2311 rth->rt_dst = daddr;
2312 rth->rt_src = saddr;
2313 #ifdef CONFIG_IP_ROUTE_CLASSID
2314 rth->dst.tclassid = itag;
2316 rth->rt_route_iif = dev->ifindex;
2317 rth->rt_iif = dev->ifindex;
2319 rth->rt_mark = skb->mark;
2320 rth->rt_gateway = daddr;
2321 rth->rt_spec_dst= spec_dst;
2322 rth->rt_peer_genid = 0;
2323 rt_init_peer(rth, net->ipv4.peers);
2325 if (res.type == RTN_UNREACHABLE) {
2326 rth->dst.input= ip_error;
2327 rth->dst.error= -err;
2328 rth->rt_flags &= ~RTCF_LOCAL;
2330 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2331 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2338 RT_CACHE_STAT_INC(in_no_route);
2339 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2340 res.type = RTN_UNREACHABLE;
2346 * Do not cache martian addresses: they should be logged (RFC1812)
2348 martian_destination:
2349 RT_CACHE_STAT_INC(in_martian_dst);
2350 #ifdef CONFIG_IP_ROUTE_VERBOSE
2351 if (IN_DEV_LOG_MARTIANS(in_dev))
2352 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2353 &daddr, &saddr, dev->name);
2357 err = -EHOSTUNREACH;
2370 martian_source_keep_err:
2371 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2375 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2376 u8 tos, struct net_device *dev, bool noref)
2380 int iif = dev->ifindex;
2388 if (!rt_caching(net))
2391 tos &= IPTOS_RT_MASK;
2392 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2394 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2395 rth = rcu_dereference(rth->dst.rt_next)) {
2396 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2397 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2398 (rth->rt_route_iif ^ iif) |
2399 (rth->rt_key_tos ^ tos)) == 0 &&
2400 rth->rt_mark == skb->mark &&
2401 net_eq(dev_net(rth->dst.dev), net) &&
2402 !rt_is_expired(rth)) {
2403 ipv4_validate_peer(rth);
2405 dst_use_noref(&rth->dst, jiffies);
2406 skb_dst_set_noref(skb, &rth->dst);
2408 dst_use(&rth->dst, jiffies);
2409 skb_dst_set(skb, &rth->dst);
2411 RT_CACHE_STAT_INC(in_hit);
2415 RT_CACHE_STAT_INC(in_hlist_search);
2419 /* Multicast recognition logic is moved from route cache to here.
2420 The problem was that too many Ethernet cards have broken/missing
2421 hardware multicast filters :-( As result the host on multicasting
2422 network acquires a lot of useless route cache entries, sort of
2423 SDR messages from all the world. Now we try to get rid of them.
2424 Really, provided software IP multicast filter is organized
2425 reasonably (at least, hashed), it does not result in a slowdown
2426 comparing with route cache reject entries.
2427 Note, that multicast routers are not affected, because
2428 route cache entry is created eventually.
2430 if (ipv4_is_multicast(daddr)) {
2431 struct in_device *in_dev = __in_dev_get_rcu(dev);
2434 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2435 ip_hdr(skb)->protocol);
2437 #ifdef CONFIG_IP_MROUTE
2439 (!ipv4_is_local_multicast(daddr) &&
2440 IN_DEV_MFORWARD(in_dev))
2443 int res = ip_route_input_mc(skb, daddr, saddr,
2452 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2456 EXPORT_SYMBOL(ip_route_input_common);
2458 /* called with rcu_read_lock() */
2459 static struct rtable *__mkroute_output(const struct fib_result *res,
2460 const struct flowi4 *fl4,
2461 __be32 orig_daddr, __be32 orig_saddr,
2462 int orig_oif, __u8 orig_rtos,
2463 struct net_device *dev_out,
2466 struct fib_info *fi = res->fi;
2467 struct in_device *in_dev;
2468 u16 type = res->type;
2471 in_dev = __in_dev_get_rcu(dev_out);
2473 return ERR_PTR(-EINVAL);
2475 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2476 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2477 return ERR_PTR(-EINVAL);
2479 if (ipv4_is_lbcast(fl4->daddr))
2480 type = RTN_BROADCAST;
2481 else if (ipv4_is_multicast(fl4->daddr))
2482 type = RTN_MULTICAST;
2483 else if (ipv4_is_zeronet(fl4->daddr))
2484 return ERR_PTR(-EINVAL);
2486 if (dev_out->flags & IFF_LOOPBACK)
2487 flags |= RTCF_LOCAL;
2489 if (type == RTN_BROADCAST) {
2490 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2492 } else if (type == RTN_MULTICAST) {
2493 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2494 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2496 flags &= ~RTCF_LOCAL;
2497 /* If multicast route do not exist use
2498 * default one, but do not gateway in this case.
2501 if (fi && res->prefixlen < 4)
2505 rth = rt_dst_alloc(dev_out,
2506 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2507 IN_DEV_CONF_GET(in_dev, NOXFRM));
2509 return ERR_PTR(-ENOBUFS);
2511 rth->dst.output = ip_output;
2513 rth->rt_key_dst = orig_daddr;
2514 rth->rt_key_src = orig_saddr;
2515 rth->rt_genid = rt_genid(dev_net(dev_out));
2516 rth->rt_flags = flags;
2517 rth->rt_type = type;
2518 rth->rt_key_tos = orig_rtos;
2519 rth->rt_dst = fl4->daddr;
2520 rth->rt_src = fl4->saddr;
2521 rth->rt_route_iif = 0;
2522 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2523 rth->rt_oif = orig_oif;
2524 rth->rt_mark = fl4->flowi4_mark;
2525 rth->rt_gateway = fl4->daddr;
2526 rth->rt_spec_dst= fl4->saddr;
2527 rth->rt_peer_genid = 0;
2528 rt_init_peer(rth, (res->table ?
2529 &res->table->tb_peers :
2530 dev_net(dev_out)->ipv4.peers));
2533 RT_CACHE_STAT_INC(out_slow_tot);
2535 if (flags & RTCF_LOCAL) {
2536 rth->dst.input = ip_local_deliver;
2537 rth->rt_spec_dst = fl4->daddr;
2539 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2540 rth->rt_spec_dst = fl4->saddr;
2541 if (flags & RTCF_LOCAL &&
2542 !(dev_out->flags & IFF_LOOPBACK)) {
2543 rth->dst.output = ip_mc_output;
2544 RT_CACHE_STAT_INC(out_slow_mc);
2546 #ifdef CONFIG_IP_MROUTE
2547 if (type == RTN_MULTICAST) {
2548 if (IN_DEV_MFORWARD(in_dev) &&
2549 !ipv4_is_local_multicast(fl4->daddr)) {
2550 rth->dst.input = ip_mr_input;
2551 rth->dst.output = ip_mc_output;
2557 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2563 * Major route resolver routine.
2564 * called with rcu_read_lock();
2567 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2569 struct net_device *dev_out = NULL;
2570 __u8 tos = RT_FL_TOS(fl4);
2571 unsigned int flags = 0;
2572 struct fib_result res;
2580 #ifdef CONFIG_IP_MULTIPLE_TABLES
2584 orig_daddr = fl4->daddr;
2585 orig_saddr = fl4->saddr;
2586 orig_oif = fl4->flowi4_oif;
2588 fl4->flowi4_iif = net->loopback_dev->ifindex;
2589 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2590 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2591 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2595 rth = ERR_PTR(-EINVAL);
2596 if (ipv4_is_multicast(fl4->saddr) ||
2597 ipv4_is_lbcast(fl4->saddr) ||
2598 ipv4_is_zeronet(fl4->saddr))
2601 /* I removed check for oif == dev_out->oif here.
2602 It was wrong for two reasons:
2603 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2604 is assigned to multiple interfaces.
2605 2. Moreover, we are allowed to send packets with saddr
2606 of another iface. --ANK
2609 if (fl4->flowi4_oif == 0 &&
2610 (ipv4_is_multicast(fl4->daddr) ||
2611 ipv4_is_lbcast(fl4->daddr))) {
2612 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2613 dev_out = __ip_dev_find(net, fl4->saddr, false);
2614 if (dev_out == NULL)
2617 /* Special hack: user can direct multicasts
2618 and limited broadcast via necessary interface
2619 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2620 This hack is not just for fun, it allows
2621 vic,vat and friends to work.
2622 They bind socket to loopback, set ttl to zero
2623 and expect that it will work.
2624 From the viewpoint of routing cache they are broken,
2625 because we are not allowed to build multicast path
2626 with loopback source addr (look, routing cache
2627 cannot know, that ttl is zero, so that packet
2628 will not leave this host and route is valid).
2629 Luckily, this hack is good workaround.
2632 fl4->flowi4_oif = dev_out->ifindex;
2636 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2637 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2638 if (!__ip_dev_find(net, fl4->saddr, false))
2644 if (fl4->flowi4_oif) {
2645 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2646 rth = ERR_PTR(-ENODEV);
2647 if (dev_out == NULL)
2650 /* RACE: Check return value of inet_select_addr instead. */
2651 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2652 rth = ERR_PTR(-ENETUNREACH);
2655 if (ipv4_is_local_multicast(fl4->daddr) ||
2656 ipv4_is_lbcast(fl4->daddr)) {
2658 fl4->saddr = inet_select_addr(dev_out, 0,
2663 if (ipv4_is_multicast(fl4->daddr))
2664 fl4->saddr = inet_select_addr(dev_out, 0,
2666 else if (!fl4->daddr)
2667 fl4->saddr = inet_select_addr(dev_out, 0,
2673 fl4->daddr = fl4->saddr;
2675 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2676 dev_out = net->loopback_dev;
2677 fl4->flowi4_oif = net->loopback_dev->ifindex;
2678 res.type = RTN_LOCAL;
2679 flags |= RTCF_LOCAL;
2683 if (fib_lookup(net, fl4, &res)) {
2686 if (fl4->flowi4_oif) {
2687 /* Apparently, routing tables are wrong. Assume,
2688 that the destination is on link.
2691 Because we are allowed to send to iface
2692 even if it has NO routes and NO assigned
2693 addresses. When oif is specified, routing
2694 tables are looked up with only one purpose:
2695 to catch if destination is gatewayed, rather than
2696 direct. Moreover, if MSG_DONTROUTE is set,
2697 we send packet, ignoring both routing tables
2698 and ifaddr state. --ANK
2701 We could make it even if oif is unknown,
2702 likely IPv6, but we do not.
2705 if (fl4->saddr == 0)
2706 fl4->saddr = inet_select_addr(dev_out, 0,
2708 res.type = RTN_UNICAST;
2711 rth = ERR_PTR(-ENETUNREACH);
2715 if (res.type == RTN_LOCAL) {
2717 if (res.fi->fib_prefsrc)
2718 fl4->saddr = res.fi->fib_prefsrc;
2720 fl4->saddr = fl4->daddr;
2722 dev_out = net->loopback_dev;
2723 fl4->flowi4_oif = dev_out->ifindex;
2725 flags |= RTCF_LOCAL;
2729 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2730 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2731 fib_select_multipath(&res);
2734 if (!res.prefixlen &&
2735 res.table->tb_num_default > 1 &&
2736 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2737 fib_select_default(&res);
2740 fl4->saddr = FIB_RES_PREFSRC(net, res);
2742 dev_out = FIB_RES_DEV(res);
2743 fl4->flowi4_oif = dev_out->ifindex;
2747 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2748 tos, dev_out, flags);
2752 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2753 rt_genid(dev_net(dev_out)));
2754 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2762 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2767 if (!rt_caching(net))
2770 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2773 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2774 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2775 if (rth->rt_key_dst == flp4->daddr &&
2776 rth->rt_key_src == flp4->saddr &&
2777 rt_is_output_route(rth) &&
2778 rth->rt_oif == flp4->flowi4_oif &&
2779 rth->rt_mark == flp4->flowi4_mark &&
2780 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2781 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2782 net_eq(dev_net(rth->dst.dev), net) &&
2783 !rt_is_expired(rth)) {
2784 ipv4_validate_peer(rth);
2785 dst_use(&rth->dst, jiffies);
2786 RT_CACHE_STAT_INC(out_hit);
2787 rcu_read_unlock_bh();
2789 flp4->saddr = rth->rt_src;
2791 flp4->daddr = rth->rt_dst;
2794 RT_CACHE_STAT_INC(out_hlist_search);
2796 rcu_read_unlock_bh();
2799 return ip_route_output_slow(net, flp4);
2801 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2803 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2808 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2810 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2812 return mtu ? : dst->dev->mtu;
2815 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2819 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2825 static struct dst_ops ipv4_dst_blackhole_ops = {
2827 .protocol = cpu_to_be16(ETH_P_IP),
2828 .destroy = ipv4_dst_destroy,
2829 .check = ipv4_blackhole_dst_check,
2830 .mtu = ipv4_blackhole_mtu,
2831 .default_advmss = ipv4_default_advmss,
2832 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2833 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2834 .neigh_lookup = ipv4_neigh_lookup,
2837 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2839 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2840 struct rtable *ort = (struct rtable *) dst_orig;
2843 struct dst_entry *new = &rt->dst;
2846 new->input = dst_discard;
2847 new->output = dst_discard;
2848 dst_copy_metrics(new, &ort->dst);
2850 new->dev = ort->dst.dev;
2854 rt->rt_key_dst = ort->rt_key_dst;
2855 rt->rt_key_src = ort->rt_key_src;
2856 rt->rt_key_tos = ort->rt_key_tos;
2857 rt->rt_route_iif = ort->rt_route_iif;
2858 rt->rt_iif = ort->rt_iif;
2859 rt->rt_oif = ort->rt_oif;
2860 rt->rt_mark = ort->rt_mark;
2862 rt->rt_genid = rt_genid(net);
2863 rt->rt_flags = ort->rt_flags;
2864 rt->rt_type = ort->rt_type;
2865 rt->rt_dst = ort->rt_dst;
2866 rt->rt_src = ort->rt_src;
2867 rt->rt_gateway = ort->rt_gateway;
2868 rt->rt_spec_dst = ort->rt_spec_dst;
2869 rt_transfer_peer(rt, ort);
2872 atomic_inc(&rt->fi->fib_clntref);
2877 dst_release(dst_orig);
2879 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2882 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2885 struct rtable *rt = __ip_route_output_key(net, flp4);
2890 if (flp4->flowi4_proto)
2891 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2892 flowi4_to_flowi(flp4),
2897 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2899 static int rt_fill_info(struct net *net,
2900 struct sk_buff *skb, u32 pid, u32 seq, int event,
2901 int nowait, unsigned int flags)
2903 struct rtable *rt = skb_rtable(skb);
2905 struct nlmsghdr *nlh;
2906 unsigned long expires = 0;
2907 u32 id = 0, ts = 0, tsage = 0, error;
2909 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2913 r = nlmsg_data(nlh);
2914 r->rtm_family = AF_INET;
2915 r->rtm_dst_len = 32;
2917 r->rtm_tos = rt->rt_key_tos;
2918 r->rtm_table = RT_TABLE_MAIN;
2919 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2920 goto nla_put_failure;
2921 r->rtm_type = rt->rt_type;
2922 r->rtm_scope = RT_SCOPE_UNIVERSE;
2923 r->rtm_protocol = RTPROT_UNSPEC;
2924 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2925 if (rt->rt_flags & RTCF_NOTIFY)
2926 r->rtm_flags |= RTM_F_NOTIFY;
2928 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2929 goto nla_put_failure;
2930 if (rt->rt_key_src) {
2931 r->rtm_src_len = 32;
2932 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2933 goto nla_put_failure;
2936 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2937 goto nla_put_failure;
2938 #ifdef CONFIG_IP_ROUTE_CLASSID
2939 if (rt->dst.tclassid &&
2940 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2941 goto nla_put_failure;
2943 if (rt_is_input_route(rt)) {
2944 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_spec_dst))
2945 goto nla_put_failure;
2946 } else if (rt->rt_src != rt->rt_key_src) {
2947 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2948 goto nla_put_failure;
2950 if (rt->rt_dst != rt->rt_gateway &&
2951 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2952 goto nla_put_failure;
2954 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2955 goto nla_put_failure;
2958 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2959 goto nla_put_failure;
2961 error = rt->dst.error;
2962 if (rt_has_peer(rt)) {
2963 const struct inet_peer *peer = rt_peer_ptr(rt);
2964 inet_peer_refcheck(peer);
2965 id = atomic_read(&peer->ip_id_count) & 0xffff;
2966 if (peer->tcp_ts_stamp) {
2968 tsage = get_seconds() - peer->tcp_ts_stamp;
2970 expires = ACCESS_ONCE(peer->pmtu_expires);
2972 if (time_before(jiffies, expires))
2979 if (rt_is_input_route(rt)) {
2980 #ifdef CONFIG_IP_MROUTE
2981 __be32 dst = rt->rt_dst;
2983 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2984 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2985 int err = ipmr_get_route(net, skb,
2986 rt->rt_src, rt->rt_dst,
2992 goto nla_put_failure;
2994 if (err == -EMSGSIZE)
2995 goto nla_put_failure;
3001 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
3002 goto nla_put_failure;
3005 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
3006 expires, error) < 0)
3007 goto nla_put_failure;
3009 return nlmsg_end(skb, nlh);
3012 nlmsg_cancel(skb, nlh);
3016 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
3018 struct net *net = sock_net(in_skb->sk);
3020 struct nlattr *tb[RTA_MAX+1];
3021 struct rtable *rt = NULL;
3027 struct sk_buff *skb;
3029 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
3033 rtm = nlmsg_data(nlh);
3035 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3041 /* Reserve room for dummy headers, this skb can pass
3042 through good chunk of routing engine.
3044 skb_reset_mac_header(skb);
3045 skb_reset_network_header(skb);
3047 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3048 ip_hdr(skb)->protocol = IPPROTO_ICMP;
3049 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
3051 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
3052 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
3053 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3054 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3057 struct net_device *dev;
3059 dev = __dev_get_by_index(net, iif);
3065 skb->protocol = htons(ETH_P_IP);
3069 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3072 rt = skb_rtable(skb);
3073 if (err == 0 && rt->dst.error)
3074 err = -rt->dst.error;
3076 struct flowi4 fl4 = {
3079 .flowi4_tos = rtm->rtm_tos,
3080 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3081 .flowi4_mark = mark,
3083 rt = ip_route_output_key(net, &fl4);
3093 skb_dst_set(skb, &rt->dst);
3094 if (rtm->rtm_flags & RTM_F_NOTIFY)
3095 rt->rt_flags |= RTCF_NOTIFY;
3097 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3098 RTM_NEWROUTE, 0, 0);
3102 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3111 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3118 net = sock_net(skb->sk);
3123 s_idx = idx = cb->args[1];
3124 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3125 if (!rt_hash_table[h].chain)
3128 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3129 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3130 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3132 if (rt_is_expired(rt))
3134 skb_dst_set_noref(skb, &rt->dst);
3135 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3136 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3137 1, NLM_F_MULTI) <= 0) {
3139 rcu_read_unlock_bh();
3144 rcu_read_unlock_bh();
3153 void ip_rt_multicast_event(struct in_device *in_dev)
3155 rt_cache_flush(dev_net(in_dev->dev), 0);
3158 #ifdef CONFIG_SYSCTL
3159 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3160 void __user *buffer,
3161 size_t *lenp, loff_t *ppos)
3168 memcpy(&ctl, __ctl, sizeof(ctl));
3169 ctl.data = &flush_delay;
3170 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3172 net = (struct net *)__ctl->extra1;
3173 rt_cache_flush(net, flush_delay);
3180 static ctl_table ipv4_route_table[] = {
3182 .procname = "gc_thresh",
3183 .data = &ipv4_dst_ops.gc_thresh,
3184 .maxlen = sizeof(int),
3186 .proc_handler = proc_dointvec,
3189 .procname = "max_size",
3190 .data = &ip_rt_max_size,
3191 .maxlen = sizeof(int),
3193 .proc_handler = proc_dointvec,
3196 /* Deprecated. Use gc_min_interval_ms */
3198 .procname = "gc_min_interval",
3199 .data = &ip_rt_gc_min_interval,
3200 .maxlen = sizeof(int),
3202 .proc_handler = proc_dointvec_jiffies,
3205 .procname = "gc_min_interval_ms",
3206 .data = &ip_rt_gc_min_interval,
3207 .maxlen = sizeof(int),
3209 .proc_handler = proc_dointvec_ms_jiffies,
3212 .procname = "gc_timeout",
3213 .data = &ip_rt_gc_timeout,
3214 .maxlen = sizeof(int),
3216 .proc_handler = proc_dointvec_jiffies,
3219 .procname = "gc_interval",
3220 .data = &ip_rt_gc_interval,
3221 .maxlen = sizeof(int),
3223 .proc_handler = proc_dointvec_jiffies,
3226 .procname = "redirect_load",
3227 .data = &ip_rt_redirect_load,
3228 .maxlen = sizeof(int),
3230 .proc_handler = proc_dointvec,
3233 .procname = "redirect_number",
3234 .data = &ip_rt_redirect_number,
3235 .maxlen = sizeof(int),
3237 .proc_handler = proc_dointvec,
3240 .procname = "redirect_silence",
3241 .data = &ip_rt_redirect_silence,
3242 .maxlen = sizeof(int),
3244 .proc_handler = proc_dointvec,
3247 .procname = "error_cost",
3248 .data = &ip_rt_error_cost,
3249 .maxlen = sizeof(int),
3251 .proc_handler = proc_dointvec,
3254 .procname = "error_burst",
3255 .data = &ip_rt_error_burst,
3256 .maxlen = sizeof(int),
3258 .proc_handler = proc_dointvec,
3261 .procname = "gc_elasticity",
3262 .data = &ip_rt_gc_elasticity,
3263 .maxlen = sizeof(int),
3265 .proc_handler = proc_dointvec,
3268 .procname = "mtu_expires",
3269 .data = &ip_rt_mtu_expires,
3270 .maxlen = sizeof(int),
3272 .proc_handler = proc_dointvec_jiffies,
3275 .procname = "min_pmtu",
3276 .data = &ip_rt_min_pmtu,
3277 .maxlen = sizeof(int),
3279 .proc_handler = proc_dointvec,
3282 .procname = "min_adv_mss",
3283 .data = &ip_rt_min_advmss,
3284 .maxlen = sizeof(int),
3286 .proc_handler = proc_dointvec,
3291 static struct ctl_table ipv4_route_flush_table[] = {
3293 .procname = "flush",
3294 .maxlen = sizeof(int),
3296 .proc_handler = ipv4_sysctl_rtcache_flush,
3301 static __net_init int sysctl_route_net_init(struct net *net)
3303 struct ctl_table *tbl;
3305 tbl = ipv4_route_flush_table;
3306 if (!net_eq(net, &init_net)) {
3307 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3311 tbl[0].extra1 = net;
3313 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3314 if (net->ipv4.route_hdr == NULL)
3319 if (tbl != ipv4_route_flush_table)
3325 static __net_exit void sysctl_route_net_exit(struct net *net)
3327 struct ctl_table *tbl;
3329 tbl = net->ipv4.route_hdr->ctl_table_arg;
3330 unregister_net_sysctl_table(net->ipv4.route_hdr);
3331 BUG_ON(tbl == ipv4_route_flush_table);
3335 static __net_initdata struct pernet_operations sysctl_route_ops = {
3336 .init = sysctl_route_net_init,
3337 .exit = sysctl_route_net_exit,
3341 static __net_init int rt_genid_init(struct net *net)
3343 get_random_bytes(&net->ipv4.rt_genid,
3344 sizeof(net->ipv4.rt_genid));
3345 get_random_bytes(&net->ipv4.dev_addr_genid,
3346 sizeof(net->ipv4.dev_addr_genid));
3350 static __net_initdata struct pernet_operations rt_genid_ops = {
3351 .init = rt_genid_init,
3354 static int __net_init ipv4_inetpeer_init(struct net *net)
3356 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3360 inet_peer_base_init(bp);
3361 net->ipv4.peers = bp;
3365 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3367 struct inet_peer_base *bp = net->ipv4.peers;
3369 net->ipv4.peers = NULL;
3370 inetpeer_invalidate_tree(bp);
3374 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3375 .init = ipv4_inetpeer_init,
3376 .exit = ipv4_inetpeer_exit,
3379 #ifdef CONFIG_IP_ROUTE_CLASSID
3380 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3381 #endif /* CONFIG_IP_ROUTE_CLASSID */
3383 static __initdata unsigned long rhash_entries;
3384 static int __init set_rhash_entries(char *str)
3391 ret = kstrtoul(str, 0, &rhash_entries);
3397 __setup("rhash_entries=", set_rhash_entries);
3399 int __init ip_rt_init(void)
3403 #ifdef CONFIG_IP_ROUTE_CLASSID
3404 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3406 panic("IP: failed to allocate ip_rt_acct\n");
3409 ipv4_dst_ops.kmem_cachep =
3410 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3411 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3413 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3415 if (dst_entries_init(&ipv4_dst_ops) < 0)
3416 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3418 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3419 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3421 rt_hash_table = (struct rt_hash_bucket *)
3422 alloc_large_system_hash("IP route cache",
3423 sizeof(struct rt_hash_bucket),
3425 (totalram_pages >= 128 * 1024) ?
3431 rhash_entries ? 0 : 512 * 1024);
3432 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3433 rt_hash_lock_init();
3435 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3436 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3441 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3442 expires_ljiffies = jiffies;
3443 schedule_delayed_work(&expires_work,
3444 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3446 if (ip_rt_proc_init())
3447 pr_err("Unable to create route proc files\n");
3450 xfrm4_init(ip_rt_max_size);
3452 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3454 #ifdef CONFIG_SYSCTL
3455 register_pernet_subsys(&sysctl_route_ops);
3457 register_pernet_subsys(&rt_genid_ops);
3458 register_pernet_subsys(&ipv4_inetpeer_ops);
3462 #ifdef CONFIG_SYSCTL
3464 * We really need to sanitize the damn ipv4 init order, then all
3465 * this nonsense will go away.
3467 void __init ip_static_sysctl_init(void)
3469 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);