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)),
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 rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
875 *rthp = rth->dst.rt_next;
880 /* We only count entries on a chain with equal
881 * hash inputs once so that entries for
882 * different QOS levels, and other non-hash
883 * input attributes don't unfairly skew the
887 rthp = &rth->dst.rt_next;
888 length += has_noalias(rt_hash_table[i].chain, rth);
890 spin_unlock_bh(rt_hash_lock_addr(i));
892 sum2 += length*length;
895 unsigned long avg = sum / samples;
896 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
897 rt_chain_length_max = max_t(unsigned long,
899 (avg + 4*sd) >> FRACT_BITS);
905 * rt_worker_func() is run in process context.
906 * we call rt_check_expire() to scan part of the hash table
908 static void rt_worker_func(struct work_struct *work)
911 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
915 * Perturbation of rt_genid by a small quantity [1..256]
916 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
917 * many times (2^24) without giving recent rt_genid.
918 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
920 static void rt_cache_invalidate(struct net *net)
922 unsigned char shuffle;
924 get_random_bytes(&shuffle, sizeof(shuffle));
925 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
926 inetpeer_invalidate_family(AF_INET);
930 * delay < 0 : invalidate cache (fast : entries will be deleted later)
931 * delay >= 0 : invalidate & flush cache (can be long)
933 void rt_cache_flush(struct net *net, int delay)
935 rt_cache_invalidate(net);
937 rt_do_flush(net, !in_softirq());
940 /* Flush previous cache invalidated entries from the cache */
941 void rt_cache_flush_batch(struct net *net)
943 rt_do_flush(net, !in_softirq());
946 static void rt_emergency_hash_rebuild(struct net *net)
948 net_warn_ratelimited("Route hash chain too long!\n");
949 rt_cache_invalidate(net);
953 Short description of GC goals.
955 We want to build algorithm, which will keep routing cache
956 at some equilibrium point, when number of aged off entries
957 is kept approximately equal to newly generated ones.
959 Current expiration strength is variable "expire".
960 We try to adjust it dynamically, so that if networking
961 is idle expires is large enough to keep enough of warm entries,
962 and when load increases it reduces to limit cache size.
965 static int rt_garbage_collect(struct dst_ops *ops)
967 static unsigned long expire = RT_GC_TIMEOUT;
968 static unsigned long last_gc;
970 static int equilibrium;
972 struct rtable __rcu **rthp;
973 unsigned long now = jiffies;
975 int entries = dst_entries_get_fast(&ipv4_dst_ops);
978 * Garbage collection is pretty expensive,
979 * do not make it too frequently.
982 RT_CACHE_STAT_INC(gc_total);
984 if (now - last_gc < ip_rt_gc_min_interval &&
985 entries < ip_rt_max_size) {
986 RT_CACHE_STAT_INC(gc_ignored);
990 entries = dst_entries_get_slow(&ipv4_dst_ops);
991 /* Calculate number of entries, which we want to expire now. */
992 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
994 if (equilibrium < ipv4_dst_ops.gc_thresh)
995 equilibrium = ipv4_dst_ops.gc_thresh;
996 goal = entries - equilibrium;
998 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
999 goal = entries - equilibrium;
1002 /* We are in dangerous area. Try to reduce cache really
1005 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1006 equilibrium = entries - goal;
1009 if (now - last_gc >= ip_rt_gc_min_interval)
1013 equilibrium += goal;
1020 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1021 unsigned long tmo = expire;
1023 k = (k + 1) & rt_hash_mask;
1024 rthp = &rt_hash_table[k].chain;
1025 spin_lock_bh(rt_hash_lock_addr(k));
1026 while ((rth = rcu_dereference_protected(*rthp,
1027 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1028 if (!rt_is_expired(rth) &&
1029 !rt_may_expire(rth, tmo, expire)) {
1031 rthp = &rth->dst.rt_next;
1034 *rthp = rth->dst.rt_next;
1038 spin_unlock_bh(rt_hash_lock_addr(k));
1047 /* Goal is not achieved. We stop process if:
1049 - if expire reduced to zero. Otherwise, expire is halfed.
1050 - if table is not full.
1051 - if we are called from interrupt.
1052 - jiffies check is just fallback/debug loop breaker.
1053 We will not spin here for long time in any case.
1056 RT_CACHE_STAT_INC(gc_goal_miss);
1063 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1065 } while (!in_softirq() && time_before_eq(jiffies, now));
1067 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1069 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1071 net_warn_ratelimited("dst cache overflow\n");
1072 RT_CACHE_STAT_INC(gc_dst_overflow);
1076 expire += ip_rt_gc_min_interval;
1077 if (expire > ip_rt_gc_timeout ||
1078 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1079 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1080 expire = ip_rt_gc_timeout;
1085 * Returns number of entries in a hash chain that have different hash_inputs
1087 static int slow_chain_length(const struct rtable *head)
1090 const struct rtable *rth = head;
1093 length += has_noalias(head, rth);
1094 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1096 return length >> FRACT_BITS;
1099 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr)
1101 static const __be32 inaddr_any = 0;
1102 struct net_device *dev = dst->dev;
1103 const __be32 *pkey = daddr;
1104 const struct rtable *rt;
1105 struct neighbour *n;
1107 rt = (const struct rtable *) dst;
1109 if (dev->flags & (IFF_LOOPBACK | IFF_POINTOPOINT))
1111 else if (rt->rt_gateway)
1112 pkey = (const __be32 *) &rt->rt_gateway;
1114 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1117 return neigh_create(&arp_tbl, pkey, dev);
1120 static int rt_bind_neighbour(struct rtable *rt)
1122 struct neighbour *n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1125 dst_set_neighbour(&rt->dst, n);
1130 static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1131 struct sk_buff *skb, int ifindex)
1133 struct rtable *rth, *cand;
1134 struct rtable __rcu **rthp, **candp;
1138 int attempts = !in_softirq();
1142 min_score = ~(u32)0;
1147 if (!rt_caching(dev_net(rt->dst.dev)) || (rt->dst.flags & DST_NOCACHE)) {
1149 * If we're not caching, just tell the caller we
1150 * were successful and don't touch the route. The
1151 * caller hold the sole reference to the cache entry, and
1152 * it will be released when the caller is done with it.
1153 * If we drop it here, the callers have no way to resolve routes
1154 * when we're not caching. Instead, just point *rp at rt, so
1155 * the caller gets a single use out of the route
1156 * Note that we do rt_free on this new route entry, so that
1157 * once its refcount hits zero, we are still able to reap it
1159 * Note: To avoid expensive rcu stuff for this uncached dst,
1160 * we set DST_NOCACHE so that dst_release() can free dst without
1161 * waiting a grace period.
1164 rt->dst.flags |= DST_NOCACHE;
1165 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1166 int err = rt_bind_neighbour(rt);
1168 net_warn_ratelimited("Neighbour table failure & not caching routes\n");
1170 return ERR_PTR(err);
1177 rthp = &rt_hash_table[hash].chain;
1179 spin_lock_bh(rt_hash_lock_addr(hash));
1180 while ((rth = rcu_dereference_protected(*rthp,
1181 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1182 if (rt_is_expired(rth)) {
1183 *rthp = rth->dst.rt_next;
1187 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1189 *rthp = rth->dst.rt_next;
1191 * Since lookup is lockfree, the deletion
1192 * must be visible to another weakly ordered CPU before
1193 * the insertion at the start of the hash chain.
1195 rcu_assign_pointer(rth->dst.rt_next,
1196 rt_hash_table[hash].chain);
1198 * Since lookup is lockfree, the update writes
1199 * must be ordered for consistency on SMP.
1201 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1203 dst_use(&rth->dst, now);
1204 spin_unlock_bh(rt_hash_lock_addr(hash));
1208 skb_dst_set(skb, &rth->dst);
1212 if (!atomic_read(&rth->dst.__refcnt)) {
1213 u32 score = rt_score(rth);
1215 if (score <= min_score) {
1224 rthp = &rth->dst.rt_next;
1228 /* ip_rt_gc_elasticity used to be average length of chain
1229 * length, when exceeded gc becomes really aggressive.
1231 * The second limit is less certain. At the moment it allows
1232 * only 2 entries per bucket. We will see.
1234 if (chain_length > ip_rt_gc_elasticity) {
1235 *candp = cand->dst.rt_next;
1239 if (chain_length > rt_chain_length_max &&
1240 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1241 struct net *net = dev_net(rt->dst.dev);
1242 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1243 if (!rt_caching(net)) {
1244 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1245 rt->dst.dev->name, num);
1247 rt_emergency_hash_rebuild(net);
1248 spin_unlock_bh(rt_hash_lock_addr(hash));
1250 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1251 ifindex, rt_genid(net));
1256 /* Try to bind route to arp only if it is output
1257 route or unicast forwarding path.
1259 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1260 int err = rt_bind_neighbour(rt);
1262 spin_unlock_bh(rt_hash_lock_addr(hash));
1264 if (err != -ENOBUFS) {
1266 return ERR_PTR(err);
1269 /* Neighbour tables are full and nothing
1270 can be released. Try to shrink route cache,
1271 it is most likely it holds some neighbour records.
1273 if (attempts-- > 0) {
1274 int saved_elasticity = ip_rt_gc_elasticity;
1275 int saved_int = ip_rt_gc_min_interval;
1276 ip_rt_gc_elasticity = 1;
1277 ip_rt_gc_min_interval = 0;
1278 rt_garbage_collect(&ipv4_dst_ops);
1279 ip_rt_gc_min_interval = saved_int;
1280 ip_rt_gc_elasticity = saved_elasticity;
1284 net_warn_ratelimited("Neighbour table overflow\n");
1286 return ERR_PTR(-ENOBUFS);
1290 rt->dst.rt_next = rt_hash_table[hash].chain;
1293 * Since lookup is lockfree, we must make sure
1294 * previous writes to rt are committed to memory
1295 * before making rt visible to other CPUS.
1297 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1299 spin_unlock_bh(rt_hash_lock_addr(hash));
1303 skb_dst_set(skb, &rt->dst);
1307 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1309 static u32 rt_peer_genid(void)
1311 return atomic_read(&__rt_peer_genid);
1314 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1316 struct inet_peer_base *base;
1317 struct inet_peer *peer;
1319 base = inetpeer_base_ptr(rt->_peer);
1323 peer = inet_getpeer_v4(base, daddr, create);
1325 if (!rt_set_peer(rt, peer))
1328 rt->rt_peer_genid = rt_peer_genid();
1333 * Peer allocation may fail only in serious out-of-memory conditions. However
1334 * we still can generate some output.
1335 * Random ID selection looks a bit dangerous because we have no chances to
1336 * select ID being unique in a reasonable period of time.
1337 * But broken packet identifier may be better than no packet at all.
1339 static void ip_select_fb_ident(struct iphdr *iph)
1341 static DEFINE_SPINLOCK(ip_fb_id_lock);
1342 static u32 ip_fallback_id;
1345 spin_lock_bh(&ip_fb_id_lock);
1346 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1347 iph->id = htons(salt & 0xFFFF);
1348 ip_fallback_id = salt;
1349 spin_unlock_bh(&ip_fb_id_lock);
1352 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1354 struct rtable *rt = (struct rtable *) dst;
1356 if (rt && !(rt->dst.flags & DST_NOPEER)) {
1357 struct inet_peer *peer = rt_get_peer_create(rt, rt->rt_dst);
1359 /* If peer is attached to destination, it is never detached,
1360 so that we need not to grab a lock to dereference it.
1363 iph->id = htons(inet_getid(peer, more));
1367 pr_debug("rt_bind_peer(0) @%p\n", __builtin_return_address(0));
1369 ip_select_fb_ident(iph);
1371 EXPORT_SYMBOL(__ip_select_ident);
1373 static void rt_del(unsigned int hash, struct rtable *rt)
1375 struct rtable __rcu **rthp;
1378 rthp = &rt_hash_table[hash].chain;
1379 spin_lock_bh(rt_hash_lock_addr(hash));
1381 while ((aux = rcu_dereference_protected(*rthp,
1382 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1383 if (aux == rt || rt_is_expired(aux)) {
1384 *rthp = aux->dst.rt_next;
1388 rthp = &aux->dst.rt_next;
1390 spin_unlock_bh(rt_hash_lock_addr(hash));
1393 static void check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1395 struct rtable *rt = (struct rtable *) dst;
1396 __be32 orig_gw = rt->rt_gateway;
1397 struct neighbour *n, *old_n;
1399 dst_confirm(&rt->dst);
1401 rt->rt_gateway = peer->redirect_learned.a4;
1403 n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1405 rt->rt_gateway = orig_gw;
1408 old_n = xchg(&rt->dst._neighbour, n);
1410 neigh_release(old_n);
1411 if (!(n->nud_state & NUD_VALID)) {
1412 neigh_event_send(n, NULL);
1414 rt->rt_flags |= RTCF_REDIRECTED;
1415 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1419 /* called in rcu_read_lock() section */
1420 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1421 __be32 saddr, struct net_device *dev)
1424 struct in_device *in_dev = __in_dev_get_rcu(dev);
1425 __be32 skeys[2] = { saddr, 0 };
1426 int ikeys[2] = { dev->ifindex, 0 };
1427 struct inet_peer *peer;
1434 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1435 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1436 ipv4_is_zeronet(new_gw))
1437 goto reject_redirect;
1439 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1440 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1441 goto reject_redirect;
1442 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1443 goto reject_redirect;
1445 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1446 goto reject_redirect;
1449 for (s = 0; s < 2; s++) {
1450 for (i = 0; i < 2; i++) {
1452 struct rtable __rcu **rthp;
1455 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1457 rthp = &rt_hash_table[hash].chain;
1459 while ((rt = rcu_dereference(*rthp)) != NULL) {
1460 rthp = &rt->dst.rt_next;
1462 if (rt->rt_key_dst != daddr ||
1463 rt->rt_key_src != skeys[s] ||
1464 rt->rt_oif != ikeys[i] ||
1465 rt_is_input_route(rt) ||
1466 rt_is_expired(rt) ||
1467 !net_eq(dev_net(rt->dst.dev), net) ||
1469 rt->dst.dev != dev ||
1470 rt->rt_gateway != old_gw)
1473 peer = rt_get_peer_create(rt, rt->rt_dst);
1475 if (peer->redirect_learned.a4 != new_gw) {
1476 peer->redirect_learned.a4 = new_gw;
1477 atomic_inc(&__rt_peer_genid);
1479 check_peer_redir(&rt->dst, peer);
1487 #ifdef CONFIG_IP_ROUTE_VERBOSE
1488 if (IN_DEV_LOG_MARTIANS(in_dev))
1489 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1490 " Advised path = %pI4 -> %pI4\n",
1491 &old_gw, dev->name, &new_gw,
1497 static bool peer_pmtu_expired(struct inet_peer *peer)
1499 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1502 time_after_eq(jiffies, orig) &&
1503 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1506 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1508 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1511 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1514 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1516 struct rtable *rt = (struct rtable *)dst;
1517 struct dst_entry *ret = dst;
1520 if (dst->obsolete > 0) {
1523 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1524 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1526 rt_genid(dev_net(dst->dev)));
1529 } else if (rt_has_peer(rt)) {
1530 struct inet_peer *peer = rt_peer_ptr(rt);
1531 if (peer_pmtu_expired(peer))
1532 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1540 * 1. The first ip_rt_redirect_number redirects are sent
1541 * with exponential backoff, then we stop sending them at all,
1542 * assuming that the host ignores our redirects.
1543 * 2. If we did not see packets requiring redirects
1544 * during ip_rt_redirect_silence, we assume that the host
1545 * forgot redirected route and start to send redirects again.
1547 * This algorithm is much cheaper and more intelligent than dumb load limiting
1550 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1551 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1554 void ip_rt_send_redirect(struct sk_buff *skb)
1556 struct rtable *rt = skb_rtable(skb);
1557 struct in_device *in_dev;
1558 struct inet_peer *peer;
1562 in_dev = __in_dev_get_rcu(rt->dst.dev);
1563 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1567 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1570 peer = rt_get_peer_create(rt, rt->rt_dst);
1572 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1576 /* No redirected packets during ip_rt_redirect_silence;
1577 * reset the algorithm.
1579 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1580 peer->rate_tokens = 0;
1582 /* Too many ignored redirects; do not send anything
1583 * set dst.rate_last to the last seen redirected packet.
1585 if (peer->rate_tokens >= ip_rt_redirect_number) {
1586 peer->rate_last = jiffies;
1590 /* Check for load limit; set rate_last to the latest sent
1593 if (peer->rate_tokens == 0 ||
1596 (ip_rt_redirect_load << peer->rate_tokens)))) {
1597 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1598 peer->rate_last = jiffies;
1599 ++peer->rate_tokens;
1600 #ifdef CONFIG_IP_ROUTE_VERBOSE
1602 peer->rate_tokens == ip_rt_redirect_number)
1603 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1604 &ip_hdr(skb)->saddr, rt->rt_iif,
1605 &rt->rt_dst, &rt->rt_gateway);
1610 static int ip_error(struct sk_buff *skb)
1612 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1613 struct rtable *rt = skb_rtable(skb);
1614 struct inet_peer *peer;
1620 net = dev_net(rt->dst.dev);
1621 if (!IN_DEV_FORWARD(in_dev)) {
1622 switch (rt->dst.error) {
1624 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
1628 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1634 switch (rt->dst.error) {
1639 code = ICMP_HOST_UNREACH;
1642 code = ICMP_NET_UNREACH;
1643 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1646 code = ICMP_PKT_FILTERED;
1650 peer = rt_get_peer_create(rt, rt->rt_dst);
1655 peer->rate_tokens += now - peer->rate_last;
1656 if (peer->rate_tokens > ip_rt_error_burst)
1657 peer->rate_tokens = ip_rt_error_burst;
1658 peer->rate_last = now;
1659 if (peer->rate_tokens >= ip_rt_error_cost)
1660 peer->rate_tokens -= ip_rt_error_cost;
1665 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1667 out: kfree_skb(skb);
1671 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1673 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1677 if (time_before(jiffies, expires)) {
1678 u32 orig_dst_mtu = dst_mtu(dst);
1679 if (peer->pmtu_learned < orig_dst_mtu) {
1680 if (!peer->pmtu_orig)
1681 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1682 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1684 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1685 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1688 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1690 struct rtable *rt = (struct rtable *) dst;
1691 struct inet_peer *peer;
1695 peer = rt_get_peer_create(rt, rt->rt_dst);
1697 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1699 if (mtu < ip_rt_min_pmtu)
1700 mtu = ip_rt_min_pmtu;
1701 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1703 pmtu_expires = jiffies + ip_rt_mtu_expires;
1707 peer->pmtu_learned = mtu;
1708 peer->pmtu_expires = pmtu_expires;
1710 atomic_inc(&__rt_peer_genid);
1711 rt->rt_peer_genid = rt_peer_genid();
1713 check_peer_pmtu(dst, peer);
1717 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1718 int oif, u32 mark, u8 protocol, int flow_flags)
1720 const struct iphdr *iph = (const struct iphdr *)skb->data;
1724 flowi4_init_output(&fl4, oif, mark, RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
1725 protocol, flow_flags | FLOWI_FLAG_PRECOW_METRICS,
1726 iph->daddr, iph->saddr, 0, 0);
1727 rt = __ip_route_output_key(net, &fl4);
1729 ip_rt_update_pmtu(&rt->dst, mtu);
1733 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1735 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1737 const struct inet_sock *inet = inet_sk(sk);
1739 return ipv4_update_pmtu(skb, sock_net(sk), mtu,
1740 sk->sk_bound_dev_if, sk->sk_mark,
1741 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
1742 inet_sk_flowi_flags(sk));
1744 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1746 static void ipv4_validate_peer(struct rtable *rt)
1748 if (rt->rt_peer_genid != rt_peer_genid()) {
1749 struct inet_peer *peer = rt_get_peer(rt, rt->rt_dst);
1752 check_peer_pmtu(&rt->dst, peer);
1754 if (peer->redirect_learned.a4 &&
1755 peer->redirect_learned.a4 != rt->rt_gateway)
1756 check_peer_redir(&rt->dst, peer);
1759 rt->rt_peer_genid = rt_peer_genid();
1763 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1765 struct rtable *rt = (struct rtable *) dst;
1767 if (rt_is_expired(rt))
1769 ipv4_validate_peer(rt);
1773 static void ipv4_dst_destroy(struct dst_entry *dst)
1775 struct rtable *rt = (struct rtable *) dst;
1778 fib_info_put(rt->fi);
1781 if (rt_has_peer(rt)) {
1782 struct inet_peer *peer = rt_peer_ptr(rt);
1788 static void ipv4_link_failure(struct sk_buff *skb)
1792 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1794 rt = skb_rtable(skb);
1795 if (rt && rt_has_peer(rt)) {
1796 struct inet_peer *peer = rt_peer_ptr(rt);
1797 if (peer_pmtu_cleaned(peer))
1798 dst_metric_set(&rt->dst, RTAX_MTU, peer->pmtu_orig);
1802 static int ip_rt_bug(struct sk_buff *skb)
1804 pr_debug("%s: %pI4 -> %pI4, %s\n",
1805 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1806 skb->dev ? skb->dev->name : "?");
1813 We do not cache source address of outgoing interface,
1814 because it is used only by IP RR, TS and SRR options,
1815 so that it out of fast path.
1817 BTW remember: "addr" is allowed to be not aligned
1821 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1825 if (rt_is_output_route(rt))
1826 src = ip_hdr(skb)->saddr;
1828 struct fib_result res;
1834 memset(&fl4, 0, sizeof(fl4));
1835 fl4.daddr = iph->daddr;
1836 fl4.saddr = iph->saddr;
1837 fl4.flowi4_tos = RT_TOS(iph->tos);
1838 fl4.flowi4_oif = rt->dst.dev->ifindex;
1839 fl4.flowi4_iif = skb->dev->ifindex;
1840 fl4.flowi4_mark = skb->mark;
1843 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1844 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1846 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1850 memcpy(addr, &src, 4);
1853 #ifdef CONFIG_IP_ROUTE_CLASSID
1854 static void set_class_tag(struct rtable *rt, u32 tag)
1856 if (!(rt->dst.tclassid & 0xFFFF))
1857 rt->dst.tclassid |= tag & 0xFFFF;
1858 if (!(rt->dst.tclassid & 0xFFFF0000))
1859 rt->dst.tclassid |= tag & 0xFFFF0000;
1863 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1865 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1868 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1870 if (advmss > 65535 - 40)
1871 advmss = 65535 - 40;
1876 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1878 const struct rtable *rt = (const struct rtable *) dst;
1879 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1881 if (mtu && rt_is_output_route(rt))
1884 mtu = dst->dev->mtu;
1886 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1888 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1892 if (mtu > IP_MAX_MTU)
1898 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1899 struct fib_info *fi)
1901 struct inet_peer_base *base;
1902 struct inet_peer *peer;
1905 /* If a peer entry exists for this destination, we must hook
1906 * it up in order to get at cached metrics.
1908 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1911 base = inetpeer_base_ptr(rt->_peer);
1914 peer = inet_getpeer_v4(base, rt->rt_dst, create);
1916 __rt_set_peer(rt, peer);
1917 rt->rt_peer_genid = rt_peer_genid();
1918 if (inet_metrics_new(peer))
1919 memcpy(peer->metrics, fi->fib_metrics,
1920 sizeof(u32) * RTAX_MAX);
1921 dst_init_metrics(&rt->dst, peer->metrics, false);
1923 check_peer_pmtu(&rt->dst, peer);
1925 if (peer->redirect_learned.a4 &&
1926 peer->redirect_learned.a4 != rt->rt_gateway) {
1927 rt->rt_gateway = peer->redirect_learned.a4;
1928 rt->rt_flags |= RTCF_REDIRECTED;
1931 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1933 atomic_inc(&fi->fib_clntref);
1935 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1939 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1940 const struct fib_result *res,
1941 struct fib_info *fi, u16 type, u32 itag)
1943 struct dst_entry *dst = &rt->dst;
1946 if (FIB_RES_GW(*res) &&
1947 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1948 rt->rt_gateway = FIB_RES_GW(*res);
1949 rt_init_metrics(rt, fl4, fi);
1950 #ifdef CONFIG_IP_ROUTE_CLASSID
1951 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1955 if (dst_mtu(dst) > IP_MAX_MTU)
1956 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1958 #ifdef CONFIG_IP_ROUTE_CLASSID
1959 #ifdef CONFIG_IP_MULTIPLE_TABLES
1960 set_class_tag(rt, fib_rules_tclass(res));
1962 set_class_tag(rt, itag);
1966 static struct rtable *rt_dst_alloc(struct net_device *dev,
1967 bool nopolicy, bool noxfrm)
1969 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1971 (nopolicy ? DST_NOPOLICY : 0) |
1972 (noxfrm ? DST_NOXFRM : 0));
1975 /* called in rcu_read_lock() section */
1976 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1977 u8 tos, struct net_device *dev, int our)
1981 struct in_device *in_dev = __in_dev_get_rcu(dev);
1985 /* Primary sanity checks. */
1990 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1991 skb->protocol != htons(ETH_P_IP))
1994 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1995 if (ipv4_is_loopback(saddr))
1998 if (ipv4_is_zeronet(saddr)) {
1999 if (!ipv4_is_local_multicast(daddr))
2002 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &itag);
2006 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
2007 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2011 #ifdef CONFIG_IP_ROUTE_CLASSID
2012 rth->dst.tclassid = itag;
2014 rth->dst.output = ip_rt_bug;
2016 rth->rt_key_dst = daddr;
2017 rth->rt_key_src = saddr;
2018 rth->rt_genid = rt_genid(dev_net(dev));
2019 rth->rt_flags = RTCF_MULTICAST;
2020 rth->rt_type = RTN_MULTICAST;
2021 rth->rt_key_tos = tos;
2022 rth->rt_dst = daddr;
2023 rth->rt_src = saddr;
2024 rth->rt_route_iif = dev->ifindex;
2025 rth->rt_iif = dev->ifindex;
2027 rth->rt_mark = skb->mark;
2028 rth->rt_gateway = daddr;
2029 rth->rt_peer_genid = 0;
2030 rt_init_peer(rth, dev_net(dev)->ipv4.peers);
2033 rth->dst.input= ip_local_deliver;
2034 rth->rt_flags |= RTCF_LOCAL;
2037 #ifdef CONFIG_IP_MROUTE
2038 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
2039 rth->dst.input = ip_mr_input;
2041 RT_CACHE_STAT_INC(in_slow_mc);
2043 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
2044 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
2045 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
2056 static void ip_handle_martian_source(struct net_device *dev,
2057 struct in_device *in_dev,
2058 struct sk_buff *skb,
2062 RT_CACHE_STAT_INC(in_martian_src);
2063 #ifdef CONFIG_IP_ROUTE_VERBOSE
2064 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2066 * RFC1812 recommendation, if source is martian,
2067 * the only hint is MAC header.
2069 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
2070 &daddr, &saddr, dev->name);
2071 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2072 print_hex_dump(KERN_WARNING, "ll header: ",
2073 DUMP_PREFIX_OFFSET, 16, 1,
2074 skb_mac_header(skb),
2075 dev->hard_header_len, true);
2081 /* called in rcu_read_lock() section */
2082 static int __mkroute_input(struct sk_buff *skb,
2083 const struct fib_result *res,
2084 struct in_device *in_dev,
2085 __be32 daddr, __be32 saddr, u32 tos,
2086 struct rtable **result)
2090 struct in_device *out_dev;
2091 unsigned int flags = 0;
2094 /* get a working reference to the output device */
2095 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2096 if (out_dev == NULL) {
2097 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
2102 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2103 in_dev->dev, &itag);
2105 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2112 flags |= RTCF_DIRECTSRC;
2114 if (out_dev == in_dev && err &&
2115 (IN_DEV_SHARED_MEDIA(out_dev) ||
2116 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2117 flags |= RTCF_DOREDIRECT;
2119 if (skb->protocol != htons(ETH_P_IP)) {
2120 /* Not IP (i.e. ARP). Do not create route, if it is
2121 * invalid for proxy arp. DNAT routes are always valid.
2123 * Proxy arp feature have been extended to allow, ARP
2124 * replies back to the same interface, to support
2125 * Private VLAN switch technologies. See arp.c.
2127 if (out_dev == in_dev &&
2128 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2134 rth = rt_dst_alloc(out_dev->dev,
2135 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2136 IN_DEV_CONF_GET(out_dev, NOXFRM));
2142 rth->rt_key_dst = daddr;
2143 rth->rt_key_src = saddr;
2144 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2145 rth->rt_flags = flags;
2146 rth->rt_type = res->type;
2147 rth->rt_key_tos = tos;
2148 rth->rt_dst = daddr;
2149 rth->rt_src = saddr;
2150 rth->rt_route_iif = in_dev->dev->ifindex;
2151 rth->rt_iif = in_dev->dev->ifindex;
2153 rth->rt_mark = skb->mark;
2154 rth->rt_gateway = daddr;
2155 rth->rt_peer_genid = 0;
2156 rt_init_peer(rth, &res->table->tb_peers);
2159 rth->dst.input = ip_forward;
2160 rth->dst.output = ip_output;
2162 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2170 static int ip_mkroute_input(struct sk_buff *skb,
2171 struct fib_result *res,
2172 const struct flowi4 *fl4,
2173 struct in_device *in_dev,
2174 __be32 daddr, __be32 saddr, u32 tos)
2176 struct rtable *rth = NULL;
2180 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2181 if (res->fi && res->fi->fib_nhs > 1)
2182 fib_select_multipath(res);
2185 /* create a routing cache entry */
2186 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2190 /* put it into the cache */
2191 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2192 rt_genid(dev_net(rth->dst.dev)));
2193 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2195 return PTR_ERR(rth);
2200 * NOTE. We drop all the packets that has local source
2201 * addresses, because every properly looped back packet
2202 * must have correct destination already attached by output routine.
2204 * Such approach solves two big problems:
2205 * 1. Not simplex devices are handled properly.
2206 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2207 * called with rcu_read_lock()
2210 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2211 u8 tos, struct net_device *dev)
2213 struct fib_result res;
2214 struct in_device *in_dev = __in_dev_get_rcu(dev);
2216 unsigned int flags = 0;
2221 struct net *net = dev_net(dev);
2223 /* IP on this device is disabled. */
2228 /* Check for the most weird martians, which can be not detected
2232 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2233 goto martian_source;
2235 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2238 /* Accept zero addresses only to limited broadcast;
2239 * I even do not know to fix it or not. Waiting for complains :-)
2241 if (ipv4_is_zeronet(saddr))
2242 goto martian_source;
2244 if (ipv4_is_zeronet(daddr))
2245 goto martian_destination;
2247 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
2248 if (ipv4_is_loopback(daddr))
2249 goto martian_destination;
2251 if (ipv4_is_loopback(saddr))
2252 goto martian_source;
2256 * Now we are ready to route packet.
2259 fl4.flowi4_iif = dev->ifindex;
2260 fl4.flowi4_mark = skb->mark;
2261 fl4.flowi4_tos = tos;
2262 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2265 err = fib_lookup(net, &fl4, &res);
2269 RT_CACHE_STAT_INC(in_slow_tot);
2271 if (res.type == RTN_BROADCAST)
2274 if (res.type == RTN_LOCAL) {
2275 err = fib_validate_source(skb, saddr, daddr, tos,
2276 net->loopback_dev->ifindex,
2279 goto martian_source_keep_err;
2281 flags |= RTCF_DIRECTSRC;
2285 if (!IN_DEV_FORWARD(in_dev))
2287 if (res.type != RTN_UNICAST)
2288 goto martian_destination;
2290 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2294 if (skb->protocol != htons(ETH_P_IP))
2297 if (!ipv4_is_zeronet(saddr)) {
2298 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &itag);
2300 goto martian_source_keep_err;
2302 flags |= RTCF_DIRECTSRC;
2304 flags |= RTCF_BROADCAST;
2305 res.type = RTN_BROADCAST;
2306 RT_CACHE_STAT_INC(in_brd);
2309 rth = rt_dst_alloc(net->loopback_dev,
2310 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2314 rth->dst.input= ip_local_deliver;
2315 rth->dst.output= ip_rt_bug;
2316 #ifdef CONFIG_IP_ROUTE_CLASSID
2317 rth->dst.tclassid = itag;
2320 rth->rt_key_dst = daddr;
2321 rth->rt_key_src = saddr;
2322 rth->rt_genid = rt_genid(net);
2323 rth->rt_flags = flags|RTCF_LOCAL;
2324 rth->rt_type = res.type;
2325 rth->rt_key_tos = tos;
2326 rth->rt_dst = daddr;
2327 rth->rt_src = saddr;
2328 #ifdef CONFIG_IP_ROUTE_CLASSID
2329 rth->dst.tclassid = itag;
2331 rth->rt_route_iif = dev->ifindex;
2332 rth->rt_iif = dev->ifindex;
2334 rth->rt_mark = skb->mark;
2335 rth->rt_gateway = daddr;
2336 rth->rt_peer_genid = 0;
2337 rt_init_peer(rth, net->ipv4.peers);
2339 if (res.type == RTN_UNREACHABLE) {
2340 rth->dst.input= ip_error;
2341 rth->dst.error= -err;
2342 rth->rt_flags &= ~RTCF_LOCAL;
2344 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2345 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2352 RT_CACHE_STAT_INC(in_no_route);
2353 res.type = RTN_UNREACHABLE;
2359 * Do not cache martian addresses: they should be logged (RFC1812)
2361 martian_destination:
2362 RT_CACHE_STAT_INC(in_martian_dst);
2363 #ifdef CONFIG_IP_ROUTE_VERBOSE
2364 if (IN_DEV_LOG_MARTIANS(in_dev))
2365 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2366 &daddr, &saddr, dev->name);
2379 martian_source_keep_err:
2380 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2384 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2385 u8 tos, struct net_device *dev, bool noref)
2389 int iif = dev->ifindex;
2397 if (!rt_caching(net))
2400 tos &= IPTOS_RT_MASK;
2401 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2403 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2404 rth = rcu_dereference(rth->dst.rt_next)) {
2405 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2406 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2407 (rth->rt_route_iif ^ iif) |
2408 (rth->rt_key_tos ^ tos)) == 0 &&
2409 rth->rt_mark == skb->mark &&
2410 net_eq(dev_net(rth->dst.dev), net) &&
2411 !rt_is_expired(rth)) {
2412 ipv4_validate_peer(rth);
2414 dst_use_noref(&rth->dst, jiffies);
2415 skb_dst_set_noref(skb, &rth->dst);
2417 dst_use(&rth->dst, jiffies);
2418 skb_dst_set(skb, &rth->dst);
2420 RT_CACHE_STAT_INC(in_hit);
2424 RT_CACHE_STAT_INC(in_hlist_search);
2428 /* Multicast recognition logic is moved from route cache to here.
2429 The problem was that too many Ethernet cards have broken/missing
2430 hardware multicast filters :-( As result the host on multicasting
2431 network acquires a lot of useless route cache entries, sort of
2432 SDR messages from all the world. Now we try to get rid of them.
2433 Really, provided software IP multicast filter is organized
2434 reasonably (at least, hashed), it does not result in a slowdown
2435 comparing with route cache reject entries.
2436 Note, that multicast routers are not affected, because
2437 route cache entry is created eventually.
2439 if (ipv4_is_multicast(daddr)) {
2440 struct in_device *in_dev = __in_dev_get_rcu(dev);
2443 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2444 ip_hdr(skb)->protocol);
2446 #ifdef CONFIG_IP_MROUTE
2448 (!ipv4_is_local_multicast(daddr) &&
2449 IN_DEV_MFORWARD(in_dev))
2452 int res = ip_route_input_mc(skb, daddr, saddr,
2461 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2465 EXPORT_SYMBOL(ip_route_input_common);
2467 /* called with rcu_read_lock() */
2468 static struct rtable *__mkroute_output(const struct fib_result *res,
2469 const struct flowi4 *fl4,
2470 __be32 orig_daddr, __be32 orig_saddr,
2471 int orig_oif, __u8 orig_rtos,
2472 struct net_device *dev_out,
2475 struct fib_info *fi = res->fi;
2476 struct in_device *in_dev;
2477 u16 type = res->type;
2480 in_dev = __in_dev_get_rcu(dev_out);
2482 return ERR_PTR(-EINVAL);
2484 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2485 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2486 return ERR_PTR(-EINVAL);
2488 if (ipv4_is_lbcast(fl4->daddr))
2489 type = RTN_BROADCAST;
2490 else if (ipv4_is_multicast(fl4->daddr))
2491 type = RTN_MULTICAST;
2492 else if (ipv4_is_zeronet(fl4->daddr))
2493 return ERR_PTR(-EINVAL);
2495 if (dev_out->flags & IFF_LOOPBACK)
2496 flags |= RTCF_LOCAL;
2498 if (type == RTN_BROADCAST) {
2499 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2501 } else if (type == RTN_MULTICAST) {
2502 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2503 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2505 flags &= ~RTCF_LOCAL;
2506 /* If multicast route do not exist use
2507 * default one, but do not gateway in this case.
2510 if (fi && res->prefixlen < 4)
2514 rth = rt_dst_alloc(dev_out,
2515 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2516 IN_DEV_CONF_GET(in_dev, NOXFRM));
2518 return ERR_PTR(-ENOBUFS);
2520 rth->dst.output = ip_output;
2522 rth->rt_key_dst = orig_daddr;
2523 rth->rt_key_src = orig_saddr;
2524 rth->rt_genid = rt_genid(dev_net(dev_out));
2525 rth->rt_flags = flags;
2526 rth->rt_type = type;
2527 rth->rt_key_tos = orig_rtos;
2528 rth->rt_dst = fl4->daddr;
2529 rth->rt_src = fl4->saddr;
2530 rth->rt_route_iif = 0;
2531 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2532 rth->rt_oif = orig_oif;
2533 rth->rt_mark = fl4->flowi4_mark;
2534 rth->rt_gateway = fl4->daddr;
2535 rth->rt_peer_genid = 0;
2536 rt_init_peer(rth, (res->table ?
2537 &res->table->tb_peers :
2538 dev_net(dev_out)->ipv4.peers));
2541 RT_CACHE_STAT_INC(out_slow_tot);
2543 if (flags & RTCF_LOCAL)
2544 rth->dst.input = ip_local_deliver;
2545 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2546 if (flags & RTCF_LOCAL &&
2547 !(dev_out->flags & IFF_LOOPBACK)) {
2548 rth->dst.output = ip_mc_output;
2549 RT_CACHE_STAT_INC(out_slow_mc);
2551 #ifdef CONFIG_IP_MROUTE
2552 if (type == RTN_MULTICAST) {
2553 if (IN_DEV_MFORWARD(in_dev) &&
2554 !ipv4_is_local_multicast(fl4->daddr)) {
2555 rth->dst.input = ip_mr_input;
2556 rth->dst.output = ip_mc_output;
2562 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2564 if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
2565 rth->dst.flags |= DST_NOCACHE;
2571 * Major route resolver routine.
2572 * called with rcu_read_lock();
2575 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2577 struct net_device *dev_out = NULL;
2578 __u8 tos = RT_FL_TOS(fl4);
2579 unsigned int flags = 0;
2580 struct fib_result res;
2588 #ifdef CONFIG_IP_MULTIPLE_TABLES
2592 orig_daddr = fl4->daddr;
2593 orig_saddr = fl4->saddr;
2594 orig_oif = fl4->flowi4_oif;
2596 fl4->flowi4_iif = net->loopback_dev->ifindex;
2597 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2598 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2599 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2603 rth = ERR_PTR(-EINVAL);
2604 if (ipv4_is_multicast(fl4->saddr) ||
2605 ipv4_is_lbcast(fl4->saddr) ||
2606 ipv4_is_zeronet(fl4->saddr))
2609 /* I removed check for oif == dev_out->oif here.
2610 It was wrong for two reasons:
2611 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2612 is assigned to multiple interfaces.
2613 2. Moreover, we are allowed to send packets with saddr
2614 of another iface. --ANK
2617 if (fl4->flowi4_oif == 0 &&
2618 (ipv4_is_multicast(fl4->daddr) ||
2619 ipv4_is_lbcast(fl4->daddr))) {
2620 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2621 dev_out = __ip_dev_find(net, fl4->saddr, false);
2622 if (dev_out == NULL)
2625 /* Special hack: user can direct multicasts
2626 and limited broadcast via necessary interface
2627 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2628 This hack is not just for fun, it allows
2629 vic,vat and friends to work.
2630 They bind socket to loopback, set ttl to zero
2631 and expect that it will work.
2632 From the viewpoint of routing cache they are broken,
2633 because we are not allowed to build multicast path
2634 with loopback source addr (look, routing cache
2635 cannot know, that ttl is zero, so that packet
2636 will not leave this host and route is valid).
2637 Luckily, this hack is good workaround.
2640 fl4->flowi4_oif = dev_out->ifindex;
2644 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2645 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2646 if (!__ip_dev_find(net, fl4->saddr, false))
2652 if (fl4->flowi4_oif) {
2653 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2654 rth = ERR_PTR(-ENODEV);
2655 if (dev_out == NULL)
2658 /* RACE: Check return value of inet_select_addr instead. */
2659 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2660 rth = ERR_PTR(-ENETUNREACH);
2663 if (ipv4_is_local_multicast(fl4->daddr) ||
2664 ipv4_is_lbcast(fl4->daddr)) {
2666 fl4->saddr = inet_select_addr(dev_out, 0,
2671 if (ipv4_is_multicast(fl4->daddr))
2672 fl4->saddr = inet_select_addr(dev_out, 0,
2674 else if (!fl4->daddr)
2675 fl4->saddr = inet_select_addr(dev_out, 0,
2681 fl4->daddr = fl4->saddr;
2683 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2684 dev_out = net->loopback_dev;
2685 fl4->flowi4_oif = net->loopback_dev->ifindex;
2686 res.type = RTN_LOCAL;
2687 flags |= RTCF_LOCAL;
2691 if (fib_lookup(net, fl4, &res)) {
2694 if (fl4->flowi4_oif) {
2695 /* Apparently, routing tables are wrong. Assume,
2696 that the destination is on link.
2699 Because we are allowed to send to iface
2700 even if it has NO routes and NO assigned
2701 addresses. When oif is specified, routing
2702 tables are looked up with only one purpose:
2703 to catch if destination is gatewayed, rather than
2704 direct. Moreover, if MSG_DONTROUTE is set,
2705 we send packet, ignoring both routing tables
2706 and ifaddr state. --ANK
2709 We could make it even if oif is unknown,
2710 likely IPv6, but we do not.
2713 if (fl4->saddr == 0)
2714 fl4->saddr = inet_select_addr(dev_out, 0,
2716 res.type = RTN_UNICAST;
2719 rth = ERR_PTR(-ENETUNREACH);
2723 if (res.type == RTN_LOCAL) {
2725 if (res.fi->fib_prefsrc)
2726 fl4->saddr = res.fi->fib_prefsrc;
2728 fl4->saddr = fl4->daddr;
2730 dev_out = net->loopback_dev;
2731 fl4->flowi4_oif = dev_out->ifindex;
2733 flags |= RTCF_LOCAL;
2737 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2738 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2739 fib_select_multipath(&res);
2742 if (!res.prefixlen &&
2743 res.table->tb_num_default > 1 &&
2744 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2745 fib_select_default(&res);
2748 fl4->saddr = FIB_RES_PREFSRC(net, res);
2750 dev_out = FIB_RES_DEV(res);
2751 fl4->flowi4_oif = dev_out->ifindex;
2755 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2756 tos, dev_out, flags);
2760 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2761 rt_genid(dev_net(dev_out)));
2762 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2770 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2775 if (!rt_caching(net))
2778 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2781 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2782 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2783 if (rth->rt_key_dst == flp4->daddr &&
2784 rth->rt_key_src == flp4->saddr &&
2785 rt_is_output_route(rth) &&
2786 rth->rt_oif == flp4->flowi4_oif &&
2787 rth->rt_mark == flp4->flowi4_mark &&
2788 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2789 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2790 net_eq(dev_net(rth->dst.dev), net) &&
2791 !rt_is_expired(rth)) {
2792 ipv4_validate_peer(rth);
2793 dst_use(&rth->dst, jiffies);
2794 RT_CACHE_STAT_INC(out_hit);
2795 rcu_read_unlock_bh();
2797 flp4->saddr = rth->rt_src;
2799 flp4->daddr = rth->rt_dst;
2802 RT_CACHE_STAT_INC(out_hlist_search);
2804 rcu_read_unlock_bh();
2807 return ip_route_output_slow(net, flp4);
2809 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2811 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2816 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2818 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2820 return mtu ? : dst->dev->mtu;
2823 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2827 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2833 static struct dst_ops ipv4_dst_blackhole_ops = {
2835 .protocol = cpu_to_be16(ETH_P_IP),
2836 .destroy = ipv4_dst_destroy,
2837 .check = ipv4_blackhole_dst_check,
2838 .mtu = ipv4_blackhole_mtu,
2839 .default_advmss = ipv4_default_advmss,
2840 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2841 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2842 .neigh_lookup = ipv4_neigh_lookup,
2845 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2847 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2848 struct rtable *ort = (struct rtable *) dst_orig;
2851 struct dst_entry *new = &rt->dst;
2854 new->input = dst_discard;
2855 new->output = dst_discard;
2856 dst_copy_metrics(new, &ort->dst);
2858 new->dev = ort->dst.dev;
2862 rt->rt_key_dst = ort->rt_key_dst;
2863 rt->rt_key_src = ort->rt_key_src;
2864 rt->rt_key_tos = ort->rt_key_tos;
2865 rt->rt_route_iif = ort->rt_route_iif;
2866 rt->rt_iif = ort->rt_iif;
2867 rt->rt_oif = ort->rt_oif;
2868 rt->rt_mark = ort->rt_mark;
2870 rt->rt_genid = rt_genid(net);
2871 rt->rt_flags = ort->rt_flags;
2872 rt->rt_type = ort->rt_type;
2873 rt->rt_dst = ort->rt_dst;
2874 rt->rt_src = ort->rt_src;
2875 rt->rt_gateway = ort->rt_gateway;
2876 rt_transfer_peer(rt, ort);
2879 atomic_inc(&rt->fi->fib_clntref);
2884 dst_release(dst_orig);
2886 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2889 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2892 struct rtable *rt = __ip_route_output_key(net, flp4);
2897 if (flp4->flowi4_proto)
2898 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2899 flowi4_to_flowi(flp4),
2904 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2906 static int rt_fill_info(struct net *net,
2907 struct sk_buff *skb, u32 pid, u32 seq, int event,
2908 int nowait, unsigned int flags)
2910 struct rtable *rt = skb_rtable(skb);
2912 struct nlmsghdr *nlh;
2913 unsigned long expires = 0;
2914 u32 id = 0, ts = 0, tsage = 0, error;
2916 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2920 r = nlmsg_data(nlh);
2921 r->rtm_family = AF_INET;
2922 r->rtm_dst_len = 32;
2924 r->rtm_tos = rt->rt_key_tos;
2925 r->rtm_table = RT_TABLE_MAIN;
2926 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2927 goto nla_put_failure;
2928 r->rtm_type = rt->rt_type;
2929 r->rtm_scope = RT_SCOPE_UNIVERSE;
2930 r->rtm_protocol = RTPROT_UNSPEC;
2931 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2932 if (rt->rt_flags & RTCF_NOTIFY)
2933 r->rtm_flags |= RTM_F_NOTIFY;
2935 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2936 goto nla_put_failure;
2937 if (rt->rt_key_src) {
2938 r->rtm_src_len = 32;
2939 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2940 goto nla_put_failure;
2943 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2944 goto nla_put_failure;
2945 #ifdef CONFIG_IP_ROUTE_CLASSID
2946 if (rt->dst.tclassid &&
2947 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2948 goto nla_put_failure;
2950 if (!rt_is_input_route(rt) &&
2951 rt->rt_src != rt->rt_key_src) {
2952 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2953 goto nla_put_failure;
2955 if (rt->rt_dst != rt->rt_gateway &&
2956 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2957 goto nla_put_failure;
2959 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2960 goto nla_put_failure;
2963 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2964 goto nla_put_failure;
2966 error = rt->dst.error;
2967 if (rt_has_peer(rt)) {
2968 const struct inet_peer *peer = rt_peer_ptr(rt);
2969 inet_peer_refcheck(peer);
2970 id = atomic_read(&peer->ip_id_count) & 0xffff;
2971 if (peer->tcp_ts_stamp) {
2973 tsage = get_seconds() - peer->tcp_ts_stamp;
2975 expires = ACCESS_ONCE(peer->pmtu_expires);
2977 if (time_before(jiffies, expires))
2984 if (rt_is_input_route(rt)) {
2985 #ifdef CONFIG_IP_MROUTE
2986 __be32 dst = rt->rt_dst;
2988 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2989 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2990 int err = ipmr_get_route(net, skb,
2991 rt->rt_src, rt->rt_dst,
2997 goto nla_put_failure;
2999 if (err == -EMSGSIZE)
3000 goto nla_put_failure;
3006 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
3007 goto nla_put_failure;
3010 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
3011 expires, error) < 0)
3012 goto nla_put_failure;
3014 return nlmsg_end(skb, nlh);
3017 nlmsg_cancel(skb, nlh);
3021 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
3023 struct net *net = sock_net(in_skb->sk);
3025 struct nlattr *tb[RTA_MAX+1];
3026 struct rtable *rt = NULL;
3032 struct sk_buff *skb;
3034 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
3038 rtm = nlmsg_data(nlh);
3040 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3046 /* Reserve room for dummy headers, this skb can pass
3047 through good chunk of routing engine.
3049 skb_reset_mac_header(skb);
3050 skb_reset_network_header(skb);
3052 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3053 ip_hdr(skb)->protocol = IPPROTO_ICMP;
3054 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
3056 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
3057 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
3058 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3059 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3062 struct net_device *dev;
3064 dev = __dev_get_by_index(net, iif);
3070 skb->protocol = htons(ETH_P_IP);
3074 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3077 rt = skb_rtable(skb);
3078 if (err == 0 && rt->dst.error)
3079 err = -rt->dst.error;
3081 struct flowi4 fl4 = {
3084 .flowi4_tos = rtm->rtm_tos,
3085 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3086 .flowi4_mark = mark,
3088 rt = ip_route_output_key(net, &fl4);
3098 skb_dst_set(skb, &rt->dst);
3099 if (rtm->rtm_flags & RTM_F_NOTIFY)
3100 rt->rt_flags |= RTCF_NOTIFY;
3102 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3103 RTM_NEWROUTE, 0, 0);
3107 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3116 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3123 net = sock_net(skb->sk);
3128 s_idx = idx = cb->args[1];
3129 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3130 if (!rt_hash_table[h].chain)
3133 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3134 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3135 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3137 if (rt_is_expired(rt))
3139 skb_dst_set_noref(skb, &rt->dst);
3140 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3141 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3142 1, NLM_F_MULTI) <= 0) {
3144 rcu_read_unlock_bh();
3149 rcu_read_unlock_bh();
3158 void ip_rt_multicast_event(struct in_device *in_dev)
3160 rt_cache_flush(dev_net(in_dev->dev), 0);
3163 #ifdef CONFIG_SYSCTL
3164 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3165 void __user *buffer,
3166 size_t *lenp, loff_t *ppos)
3173 memcpy(&ctl, __ctl, sizeof(ctl));
3174 ctl.data = &flush_delay;
3175 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3177 net = (struct net *)__ctl->extra1;
3178 rt_cache_flush(net, flush_delay);
3185 static ctl_table ipv4_route_table[] = {
3187 .procname = "gc_thresh",
3188 .data = &ipv4_dst_ops.gc_thresh,
3189 .maxlen = sizeof(int),
3191 .proc_handler = proc_dointvec,
3194 .procname = "max_size",
3195 .data = &ip_rt_max_size,
3196 .maxlen = sizeof(int),
3198 .proc_handler = proc_dointvec,
3201 /* Deprecated. Use gc_min_interval_ms */
3203 .procname = "gc_min_interval",
3204 .data = &ip_rt_gc_min_interval,
3205 .maxlen = sizeof(int),
3207 .proc_handler = proc_dointvec_jiffies,
3210 .procname = "gc_min_interval_ms",
3211 .data = &ip_rt_gc_min_interval,
3212 .maxlen = sizeof(int),
3214 .proc_handler = proc_dointvec_ms_jiffies,
3217 .procname = "gc_timeout",
3218 .data = &ip_rt_gc_timeout,
3219 .maxlen = sizeof(int),
3221 .proc_handler = proc_dointvec_jiffies,
3224 .procname = "gc_interval",
3225 .data = &ip_rt_gc_interval,
3226 .maxlen = sizeof(int),
3228 .proc_handler = proc_dointvec_jiffies,
3231 .procname = "redirect_load",
3232 .data = &ip_rt_redirect_load,
3233 .maxlen = sizeof(int),
3235 .proc_handler = proc_dointvec,
3238 .procname = "redirect_number",
3239 .data = &ip_rt_redirect_number,
3240 .maxlen = sizeof(int),
3242 .proc_handler = proc_dointvec,
3245 .procname = "redirect_silence",
3246 .data = &ip_rt_redirect_silence,
3247 .maxlen = sizeof(int),
3249 .proc_handler = proc_dointvec,
3252 .procname = "error_cost",
3253 .data = &ip_rt_error_cost,
3254 .maxlen = sizeof(int),
3256 .proc_handler = proc_dointvec,
3259 .procname = "error_burst",
3260 .data = &ip_rt_error_burst,
3261 .maxlen = sizeof(int),
3263 .proc_handler = proc_dointvec,
3266 .procname = "gc_elasticity",
3267 .data = &ip_rt_gc_elasticity,
3268 .maxlen = sizeof(int),
3270 .proc_handler = proc_dointvec,
3273 .procname = "mtu_expires",
3274 .data = &ip_rt_mtu_expires,
3275 .maxlen = sizeof(int),
3277 .proc_handler = proc_dointvec_jiffies,
3280 .procname = "min_pmtu",
3281 .data = &ip_rt_min_pmtu,
3282 .maxlen = sizeof(int),
3284 .proc_handler = proc_dointvec,
3287 .procname = "min_adv_mss",
3288 .data = &ip_rt_min_advmss,
3289 .maxlen = sizeof(int),
3291 .proc_handler = proc_dointvec,
3296 static struct ctl_table ipv4_route_flush_table[] = {
3298 .procname = "flush",
3299 .maxlen = sizeof(int),
3301 .proc_handler = ipv4_sysctl_rtcache_flush,
3306 static __net_init int sysctl_route_net_init(struct net *net)
3308 struct ctl_table *tbl;
3310 tbl = ipv4_route_flush_table;
3311 if (!net_eq(net, &init_net)) {
3312 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3316 tbl[0].extra1 = net;
3318 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3319 if (net->ipv4.route_hdr == NULL)
3324 if (tbl != ipv4_route_flush_table)
3330 static __net_exit void sysctl_route_net_exit(struct net *net)
3332 struct ctl_table *tbl;
3334 tbl = net->ipv4.route_hdr->ctl_table_arg;
3335 unregister_net_sysctl_table(net->ipv4.route_hdr);
3336 BUG_ON(tbl == ipv4_route_flush_table);
3340 static __net_initdata struct pernet_operations sysctl_route_ops = {
3341 .init = sysctl_route_net_init,
3342 .exit = sysctl_route_net_exit,
3346 static __net_init int rt_genid_init(struct net *net)
3348 get_random_bytes(&net->ipv4.rt_genid,
3349 sizeof(net->ipv4.rt_genid));
3350 get_random_bytes(&net->ipv4.dev_addr_genid,
3351 sizeof(net->ipv4.dev_addr_genid));
3355 static __net_initdata struct pernet_operations rt_genid_ops = {
3356 .init = rt_genid_init,
3359 static int __net_init ipv4_inetpeer_init(struct net *net)
3361 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3365 inet_peer_base_init(bp);
3366 net->ipv4.peers = bp;
3370 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3372 struct inet_peer_base *bp = net->ipv4.peers;
3374 net->ipv4.peers = NULL;
3375 inetpeer_invalidate_tree(bp);
3379 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3380 .init = ipv4_inetpeer_init,
3381 .exit = ipv4_inetpeer_exit,
3384 #ifdef CONFIG_IP_ROUTE_CLASSID
3385 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3386 #endif /* CONFIG_IP_ROUTE_CLASSID */
3388 static __initdata unsigned long rhash_entries;
3389 static int __init set_rhash_entries(char *str)
3396 ret = kstrtoul(str, 0, &rhash_entries);
3402 __setup("rhash_entries=", set_rhash_entries);
3404 int __init ip_rt_init(void)
3408 #ifdef CONFIG_IP_ROUTE_CLASSID
3409 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3411 panic("IP: failed to allocate ip_rt_acct\n");
3414 ipv4_dst_ops.kmem_cachep =
3415 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3416 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3418 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3420 if (dst_entries_init(&ipv4_dst_ops) < 0)
3421 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3423 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3424 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3426 rt_hash_table = (struct rt_hash_bucket *)
3427 alloc_large_system_hash("IP route cache",
3428 sizeof(struct rt_hash_bucket),
3430 (totalram_pages >= 128 * 1024) ?
3436 rhash_entries ? 0 : 512 * 1024);
3437 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3438 rt_hash_lock_init();
3440 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3441 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3446 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3447 expires_ljiffies = jiffies;
3448 schedule_delayed_work(&expires_work,
3449 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3451 if (ip_rt_proc_init())
3452 pr_err("Unable to create route proc files\n");
3455 xfrm4_init(ip_rt_max_size);
3457 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3459 #ifdef CONFIG_SYSCTL
3460 register_pernet_subsys(&sysctl_route_ops);
3462 register_pernet_subsys(&rt_genid_ops);
3463 register_pernet_subsys(&ipv4_inetpeer_ops);
3467 #ifdef CONFIG_SYSCTL
3469 * We really need to sanitize the damn ipv4 init order, then all
3470 * this nonsense will go away.
3472 void __init ip_static_sysctl_init(void)
3474 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);