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)
165 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
169 static struct dst_ops ipv4_dst_ops = {
171 .protocol = cpu_to_be16(ETH_P_IP),
172 .gc = rt_garbage_collect,
173 .check = ipv4_dst_check,
174 .default_advmss = ipv4_default_advmss,
176 .cow_metrics = ipv4_cow_metrics,
177 .destroy = ipv4_dst_destroy,
178 .ifdown = ipv4_dst_ifdown,
179 .negative_advice = ipv4_negative_advice,
180 .link_failure = ipv4_link_failure,
181 .update_pmtu = ip_rt_update_pmtu,
182 .local_out = __ip_local_out,
183 .neigh_lookup = ipv4_neigh_lookup,
186 #define ECN_OR_COST(class) TC_PRIO_##class
188 const __u8 ip_tos2prio[16] = {
190 ECN_OR_COST(BESTEFFORT),
192 ECN_OR_COST(BESTEFFORT),
198 ECN_OR_COST(INTERACTIVE),
200 ECN_OR_COST(INTERACTIVE),
201 TC_PRIO_INTERACTIVE_BULK,
202 ECN_OR_COST(INTERACTIVE_BULK),
203 TC_PRIO_INTERACTIVE_BULK,
204 ECN_OR_COST(INTERACTIVE_BULK)
206 EXPORT_SYMBOL(ip_tos2prio);
212 /* The locking scheme is rather straight forward:
214 * 1) Read-Copy Update protects the buckets of the central route hash.
215 * 2) Only writers remove entries, and they hold the lock
216 * as they look at rtable reference counts.
217 * 3) Only readers acquire references to rtable entries,
218 * they do so with atomic increments and with the
222 struct rt_hash_bucket {
223 struct rtable __rcu *chain;
226 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
227 defined(CONFIG_PROVE_LOCKING)
229 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
230 * The size of this table is a power of two and depends on the number of CPUS.
231 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
233 #ifdef CONFIG_LOCKDEP
234 # define RT_HASH_LOCK_SZ 256
237 # define RT_HASH_LOCK_SZ 4096
239 # define RT_HASH_LOCK_SZ 2048
241 # define RT_HASH_LOCK_SZ 1024
243 # define RT_HASH_LOCK_SZ 512
245 # define RT_HASH_LOCK_SZ 256
249 static spinlock_t *rt_hash_locks;
250 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
252 static __init void rt_hash_lock_init(void)
256 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
259 panic("IP: failed to allocate rt_hash_locks\n");
261 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
262 spin_lock_init(&rt_hash_locks[i]);
265 # define rt_hash_lock_addr(slot) NULL
267 static inline void rt_hash_lock_init(void)
272 static struct rt_hash_bucket *rt_hash_table __read_mostly;
273 static unsigned int rt_hash_mask __read_mostly;
274 static unsigned int rt_hash_log __read_mostly;
276 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
277 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
279 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
282 return jhash_3words((__force u32)daddr, (__force u32)saddr,
287 static inline int rt_genid(struct net *net)
289 return atomic_read(&net->ipv4.rt_genid);
292 #ifdef CONFIG_PROC_FS
293 struct rt_cache_iter_state {
294 struct seq_net_private p;
299 static struct rtable *rt_cache_get_first(struct seq_file *seq)
301 struct rt_cache_iter_state *st = seq->private;
302 struct rtable *r = NULL;
304 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
305 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
308 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
310 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
311 r->rt_genid == st->genid)
313 r = rcu_dereference_bh(r->dst.rt_next);
315 rcu_read_unlock_bh();
320 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
323 struct rt_cache_iter_state *st = seq->private;
325 r = rcu_dereference_bh(r->dst.rt_next);
327 rcu_read_unlock_bh();
329 if (--st->bucket < 0)
331 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
333 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
338 static struct rtable *rt_cache_get_next(struct seq_file *seq,
341 struct rt_cache_iter_state *st = seq->private;
342 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
343 if (dev_net(r->dst.dev) != seq_file_net(seq))
345 if (r->rt_genid == st->genid)
351 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
353 struct rtable *r = rt_cache_get_first(seq);
356 while (pos && (r = rt_cache_get_next(seq, r)))
358 return pos ? NULL : r;
361 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
363 struct rt_cache_iter_state *st = seq->private;
365 return rt_cache_get_idx(seq, *pos - 1);
366 st->genid = rt_genid(seq_file_net(seq));
367 return SEQ_START_TOKEN;
370 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
374 if (v == SEQ_START_TOKEN)
375 r = rt_cache_get_first(seq);
377 r = rt_cache_get_next(seq, v);
382 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
384 if (v && v != SEQ_START_TOKEN)
385 rcu_read_unlock_bh();
388 static int rt_cache_seq_show(struct seq_file *seq, void *v)
390 if (v == SEQ_START_TOKEN)
391 seq_printf(seq, "%-127s\n",
392 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
393 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
396 struct rtable *r = v;
399 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
400 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
401 r->dst.dev ? r->dst.dev->name : "*",
402 (__force u32)r->rt_dst,
403 (__force u32)r->rt_gateway,
404 r->rt_flags, atomic_read(&r->dst.__refcnt),
405 r->dst.__use, 0, (__force u32)r->rt_src,
406 dst_metric_advmss(&r->dst) + 40,
407 dst_metric(&r->dst, RTAX_WINDOW), 0,
411 seq_printf(seq, "%*s\n", 127 - len, "");
416 static const struct seq_operations rt_cache_seq_ops = {
417 .start = rt_cache_seq_start,
418 .next = rt_cache_seq_next,
419 .stop = rt_cache_seq_stop,
420 .show = rt_cache_seq_show,
423 static int rt_cache_seq_open(struct inode *inode, struct file *file)
425 return seq_open_net(inode, file, &rt_cache_seq_ops,
426 sizeof(struct rt_cache_iter_state));
429 static const struct file_operations rt_cache_seq_fops = {
430 .owner = THIS_MODULE,
431 .open = rt_cache_seq_open,
434 .release = seq_release_net,
438 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
443 return SEQ_START_TOKEN;
445 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
446 if (!cpu_possible(cpu))
449 return &per_cpu(rt_cache_stat, cpu);
454 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
458 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
459 if (!cpu_possible(cpu))
462 return &per_cpu(rt_cache_stat, cpu);
468 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
473 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
475 struct rt_cache_stat *st = v;
477 if (v == SEQ_START_TOKEN) {
478 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");
482 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
483 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
484 dst_entries_get_slow(&ipv4_dst_ops),
507 static const struct seq_operations rt_cpu_seq_ops = {
508 .start = rt_cpu_seq_start,
509 .next = rt_cpu_seq_next,
510 .stop = rt_cpu_seq_stop,
511 .show = rt_cpu_seq_show,
515 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
517 return seq_open(file, &rt_cpu_seq_ops);
520 static const struct file_operations rt_cpu_seq_fops = {
521 .owner = THIS_MODULE,
522 .open = rt_cpu_seq_open,
525 .release = seq_release,
528 #ifdef CONFIG_IP_ROUTE_CLASSID
529 static int rt_acct_proc_show(struct seq_file *m, void *v)
531 struct ip_rt_acct *dst, *src;
534 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
538 for_each_possible_cpu(i) {
539 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
540 for (j = 0; j < 256; j++) {
541 dst[j].o_bytes += src[j].o_bytes;
542 dst[j].o_packets += src[j].o_packets;
543 dst[j].i_bytes += src[j].i_bytes;
544 dst[j].i_packets += src[j].i_packets;
548 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
553 static int rt_acct_proc_open(struct inode *inode, struct file *file)
555 return single_open(file, rt_acct_proc_show, NULL);
558 static const struct file_operations rt_acct_proc_fops = {
559 .owner = THIS_MODULE,
560 .open = rt_acct_proc_open,
563 .release = single_release,
567 static int __net_init ip_rt_do_proc_init(struct net *net)
569 struct proc_dir_entry *pde;
571 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
576 pde = proc_create("rt_cache", S_IRUGO,
577 net->proc_net_stat, &rt_cpu_seq_fops);
581 #ifdef CONFIG_IP_ROUTE_CLASSID
582 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
588 #ifdef CONFIG_IP_ROUTE_CLASSID
590 remove_proc_entry("rt_cache", net->proc_net_stat);
593 remove_proc_entry("rt_cache", net->proc_net);
598 static void __net_exit ip_rt_do_proc_exit(struct net *net)
600 remove_proc_entry("rt_cache", net->proc_net_stat);
601 remove_proc_entry("rt_cache", net->proc_net);
602 #ifdef CONFIG_IP_ROUTE_CLASSID
603 remove_proc_entry("rt_acct", net->proc_net);
607 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
608 .init = ip_rt_do_proc_init,
609 .exit = ip_rt_do_proc_exit,
612 static int __init ip_rt_proc_init(void)
614 return register_pernet_subsys(&ip_rt_proc_ops);
618 static inline int ip_rt_proc_init(void)
622 #endif /* CONFIG_PROC_FS */
624 static inline void rt_free(struct rtable *rt)
626 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
629 static inline void rt_drop(struct rtable *rt)
632 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
635 static inline int rt_fast_clean(struct rtable *rth)
637 /* Kill broadcast/multicast entries very aggresively, if they
638 collide in hash table with more useful entries */
639 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
640 rt_is_input_route(rth) && rth->dst.rt_next;
643 static inline int rt_valuable(struct rtable *rth)
645 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
649 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
654 if (atomic_read(&rth->dst.__refcnt))
657 age = jiffies - rth->dst.lastuse;
658 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
659 (age <= tmo2 && rt_valuable(rth)))
665 /* Bits of score are:
667 * 30: not quite useless
668 * 29..0: usage counter
670 static inline u32 rt_score(struct rtable *rt)
672 u32 score = jiffies - rt->dst.lastuse;
674 score = ~score & ~(3<<30);
679 if (rt_is_output_route(rt) ||
680 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
686 static inline bool rt_caching(const struct net *net)
688 return net->ipv4.current_rt_cache_rebuild_count <=
689 net->ipv4.sysctl_rt_cache_rebuild_count;
692 static inline bool compare_hash_inputs(const struct rtable *rt1,
693 const struct rtable *rt2)
695 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
696 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
697 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
700 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
702 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
703 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
704 (rt1->rt_mark ^ rt2->rt_mark) |
705 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
706 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
707 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
710 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
712 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
715 static inline int rt_is_expired(struct rtable *rth)
717 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
721 * Perform a full scan of hash table and free all entries.
722 * Can be called by a softirq or a process.
723 * In the later case, we want to be reschedule if necessary
725 static void rt_do_flush(struct net *net, int process_context)
728 struct rtable *rth, *next;
730 for (i = 0; i <= rt_hash_mask; i++) {
731 struct rtable __rcu **pprev;
734 if (process_context && need_resched())
736 rth = rcu_access_pointer(rt_hash_table[i].chain);
740 spin_lock_bh(rt_hash_lock_addr(i));
743 pprev = &rt_hash_table[i].chain;
744 rth = rcu_dereference_protected(*pprev,
745 lockdep_is_held(rt_hash_lock_addr(i)));
748 next = rcu_dereference_protected(rth->dst.rt_next,
749 lockdep_is_held(rt_hash_lock_addr(i)));
752 net_eq(dev_net(rth->dst.dev), net)) {
753 rcu_assign_pointer(*pprev, next);
754 rcu_assign_pointer(rth->dst.rt_next, list);
757 pprev = &rth->dst.rt_next;
762 spin_unlock_bh(rt_hash_lock_addr(i));
764 for (; list; list = next) {
765 next = rcu_dereference_protected(list->dst.rt_next, 1);
772 * While freeing expired entries, we compute average chain length
773 * and standard deviation, using fixed-point arithmetic.
774 * This to have an estimation of rt_chain_length_max
775 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
776 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
780 #define ONE (1UL << FRACT_BITS)
783 * Given a hash chain and an item in this hash chain,
784 * find if a previous entry has the same hash_inputs
785 * (but differs on tos, mark or oif)
786 * Returns 0 if an alias is found.
787 * Returns ONE if rth has no alias before itself.
789 static int has_noalias(const struct rtable *head, const struct rtable *rth)
791 const struct rtable *aux = head;
794 if (compare_hash_inputs(aux, rth))
796 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
801 static void rt_check_expire(void)
803 static unsigned int rover;
804 unsigned int i = rover, goal;
806 struct rtable __rcu **rthp;
807 unsigned long samples = 0;
808 unsigned long sum = 0, sum2 = 0;
812 delta = jiffies - expires_ljiffies;
813 expires_ljiffies = jiffies;
814 mult = ((u64)delta) << rt_hash_log;
815 if (ip_rt_gc_timeout > 1)
816 do_div(mult, ip_rt_gc_timeout);
817 goal = (unsigned int)mult;
818 if (goal > rt_hash_mask)
819 goal = rt_hash_mask + 1;
820 for (; goal > 0; goal--) {
821 unsigned long tmo = ip_rt_gc_timeout;
822 unsigned long length;
824 i = (i + 1) & rt_hash_mask;
825 rthp = &rt_hash_table[i].chain;
832 if (rcu_dereference_raw(*rthp) == NULL)
835 spin_lock_bh(rt_hash_lock_addr(i));
836 while ((rth = rcu_dereference_protected(*rthp,
837 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
838 prefetch(rth->dst.rt_next);
839 if (rt_is_expired(rth) ||
840 rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
841 *rthp = rth->dst.rt_next;
846 /* We only count entries on a chain with equal
847 * hash inputs once so that entries for
848 * different QOS levels, and other non-hash
849 * input attributes don't unfairly skew the
853 rthp = &rth->dst.rt_next;
854 length += has_noalias(rt_hash_table[i].chain, rth);
856 spin_unlock_bh(rt_hash_lock_addr(i));
858 sum2 += length*length;
861 unsigned long avg = sum / samples;
862 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
863 rt_chain_length_max = max_t(unsigned long,
865 (avg + 4*sd) >> FRACT_BITS);
871 * rt_worker_func() is run in process context.
872 * we call rt_check_expire() to scan part of the hash table
874 static void rt_worker_func(struct work_struct *work)
877 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
881 * Perturbation of rt_genid by a small quantity [1..256]
882 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
883 * many times (2^24) without giving recent rt_genid.
884 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
886 static void rt_cache_invalidate(struct net *net)
888 unsigned char shuffle;
890 get_random_bytes(&shuffle, sizeof(shuffle));
891 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
895 * delay < 0 : invalidate cache (fast : entries will be deleted later)
896 * delay >= 0 : invalidate & flush cache (can be long)
898 void rt_cache_flush(struct net *net, int delay)
900 rt_cache_invalidate(net);
902 rt_do_flush(net, !in_softirq());
905 /* Flush previous cache invalidated entries from the cache */
906 void rt_cache_flush_batch(struct net *net)
908 rt_do_flush(net, !in_softirq());
911 static void rt_emergency_hash_rebuild(struct net *net)
913 net_warn_ratelimited("Route hash chain too long!\n");
914 rt_cache_invalidate(net);
918 Short description of GC goals.
920 We want to build algorithm, which will keep routing cache
921 at some equilibrium point, when number of aged off entries
922 is kept approximately equal to newly generated ones.
924 Current expiration strength is variable "expire".
925 We try to adjust it dynamically, so that if networking
926 is idle expires is large enough to keep enough of warm entries,
927 and when load increases it reduces to limit cache size.
930 static int rt_garbage_collect(struct dst_ops *ops)
932 static unsigned long expire = RT_GC_TIMEOUT;
933 static unsigned long last_gc;
935 static int equilibrium;
937 struct rtable __rcu **rthp;
938 unsigned long now = jiffies;
940 int entries = dst_entries_get_fast(&ipv4_dst_ops);
943 * Garbage collection is pretty expensive,
944 * do not make it too frequently.
947 RT_CACHE_STAT_INC(gc_total);
949 if (now - last_gc < ip_rt_gc_min_interval &&
950 entries < ip_rt_max_size) {
951 RT_CACHE_STAT_INC(gc_ignored);
955 entries = dst_entries_get_slow(&ipv4_dst_ops);
956 /* Calculate number of entries, which we want to expire now. */
957 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
959 if (equilibrium < ipv4_dst_ops.gc_thresh)
960 equilibrium = ipv4_dst_ops.gc_thresh;
961 goal = entries - equilibrium;
963 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
964 goal = entries - equilibrium;
967 /* We are in dangerous area. Try to reduce cache really
970 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
971 equilibrium = entries - goal;
974 if (now - last_gc >= ip_rt_gc_min_interval)
985 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
986 unsigned long tmo = expire;
988 k = (k + 1) & rt_hash_mask;
989 rthp = &rt_hash_table[k].chain;
990 spin_lock_bh(rt_hash_lock_addr(k));
991 while ((rth = rcu_dereference_protected(*rthp,
992 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
993 if (!rt_is_expired(rth) &&
994 !rt_may_expire(rth, tmo, expire)) {
996 rthp = &rth->dst.rt_next;
999 *rthp = rth->dst.rt_next;
1003 spin_unlock_bh(rt_hash_lock_addr(k));
1012 /* Goal is not achieved. We stop process if:
1014 - if expire reduced to zero. Otherwise, expire is halfed.
1015 - if table is not full.
1016 - if we are called from interrupt.
1017 - jiffies check is just fallback/debug loop breaker.
1018 We will not spin here for long time in any case.
1021 RT_CACHE_STAT_INC(gc_goal_miss);
1028 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1030 } while (!in_softirq() && time_before_eq(jiffies, now));
1032 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1034 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1036 net_warn_ratelimited("dst cache overflow\n");
1037 RT_CACHE_STAT_INC(gc_dst_overflow);
1041 expire += ip_rt_gc_min_interval;
1042 if (expire > ip_rt_gc_timeout ||
1043 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1044 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1045 expire = ip_rt_gc_timeout;
1050 * Returns number of entries in a hash chain that have different hash_inputs
1052 static int slow_chain_length(const struct rtable *head)
1055 const struct rtable *rth = head;
1058 length += has_noalias(head, rth);
1059 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1061 return length >> FRACT_BITS;
1064 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
1065 struct sk_buff *skb,
1068 struct net_device *dev = dst->dev;
1069 const __be32 *pkey = daddr;
1070 const struct rtable *rt;
1071 struct neighbour *n;
1073 rt = (const struct rtable *) dst;
1075 pkey = (const __be32 *) &rt->rt_gateway;
1077 pkey = &ip_hdr(skb)->daddr;
1079 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1082 return neigh_create(&arp_tbl, pkey, dev);
1085 static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1086 struct sk_buff *skb, int ifindex)
1088 struct rtable *rth, *cand;
1089 struct rtable __rcu **rthp, **candp;
1096 min_score = ~(u32)0;
1101 if (!rt_caching(dev_net(rt->dst.dev)) || (rt->dst.flags & DST_NOCACHE)) {
1103 * If we're not caching, just tell the caller we
1104 * were successful and don't touch the route. The
1105 * caller hold the sole reference to the cache entry, and
1106 * it will be released when the caller is done with it.
1107 * If we drop it here, the callers have no way to resolve routes
1108 * when we're not caching. Instead, just point *rp at rt, so
1109 * the caller gets a single use out of the route
1110 * Note that we do rt_free on this new route entry, so that
1111 * once its refcount hits zero, we are still able to reap it
1113 * Note: To avoid expensive rcu stuff for this uncached dst,
1114 * we set DST_NOCACHE so that dst_release() can free dst without
1115 * waiting a grace period.
1118 rt->dst.flags |= DST_NOCACHE;
1122 rthp = &rt_hash_table[hash].chain;
1124 spin_lock_bh(rt_hash_lock_addr(hash));
1125 while ((rth = rcu_dereference_protected(*rthp,
1126 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1127 if (rt_is_expired(rth)) {
1128 *rthp = rth->dst.rt_next;
1132 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1134 *rthp = rth->dst.rt_next;
1136 * Since lookup is lockfree, the deletion
1137 * must be visible to another weakly ordered CPU before
1138 * the insertion at the start of the hash chain.
1140 rcu_assign_pointer(rth->dst.rt_next,
1141 rt_hash_table[hash].chain);
1143 * Since lookup is lockfree, the update writes
1144 * must be ordered for consistency on SMP.
1146 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1148 dst_use(&rth->dst, now);
1149 spin_unlock_bh(rt_hash_lock_addr(hash));
1153 skb_dst_set(skb, &rth->dst);
1157 if (!atomic_read(&rth->dst.__refcnt)) {
1158 u32 score = rt_score(rth);
1160 if (score <= min_score) {
1169 rthp = &rth->dst.rt_next;
1173 /* ip_rt_gc_elasticity used to be average length of chain
1174 * length, when exceeded gc becomes really aggressive.
1176 * The second limit is less certain. At the moment it allows
1177 * only 2 entries per bucket. We will see.
1179 if (chain_length > ip_rt_gc_elasticity) {
1180 *candp = cand->dst.rt_next;
1184 if (chain_length > rt_chain_length_max &&
1185 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1186 struct net *net = dev_net(rt->dst.dev);
1187 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1188 if (!rt_caching(net)) {
1189 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1190 rt->dst.dev->name, num);
1192 rt_emergency_hash_rebuild(net);
1193 spin_unlock_bh(rt_hash_lock_addr(hash));
1195 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1196 ifindex, rt_genid(net));
1201 rt->dst.rt_next = rt_hash_table[hash].chain;
1204 * Since lookup is lockfree, we must make sure
1205 * previous writes to rt are committed to memory
1206 * before making rt visible to other CPUS.
1208 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1210 spin_unlock_bh(rt_hash_lock_addr(hash));
1214 skb_dst_set(skb, &rt->dst);
1219 * Peer allocation may fail only in serious out-of-memory conditions. However
1220 * we still can generate some output.
1221 * Random ID selection looks a bit dangerous because we have no chances to
1222 * select ID being unique in a reasonable period of time.
1223 * But broken packet identifier may be better than no packet at all.
1225 static void ip_select_fb_ident(struct iphdr *iph)
1227 static DEFINE_SPINLOCK(ip_fb_id_lock);
1228 static u32 ip_fallback_id;
1231 spin_lock_bh(&ip_fb_id_lock);
1232 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1233 iph->id = htons(salt & 0xFFFF);
1234 ip_fallback_id = salt;
1235 spin_unlock_bh(&ip_fb_id_lock);
1238 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1240 struct net *net = dev_net(dst->dev);
1241 struct inet_peer *peer;
1243 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
1245 iph->id = htons(inet_getid(peer, more));
1250 ip_select_fb_ident(iph);
1252 EXPORT_SYMBOL(__ip_select_ident);
1254 static void rt_del(unsigned int hash, struct rtable *rt)
1256 struct rtable __rcu **rthp;
1259 rthp = &rt_hash_table[hash].chain;
1260 spin_lock_bh(rt_hash_lock_addr(hash));
1262 while ((aux = rcu_dereference_protected(*rthp,
1263 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1264 if (aux == rt || rt_is_expired(aux)) {
1265 *rthp = aux->dst.rt_next;
1269 rthp = &aux->dst.rt_next;
1271 spin_unlock_bh(rt_hash_lock_addr(hash));
1274 static void ip_do_redirect(struct rtable *rt, __be32 old_gw, __be32 new_gw)
1276 struct neighbour *n;
1278 if (rt->rt_gateway != old_gw)
1281 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
1283 if (!(n->nud_state & NUD_VALID)) {
1284 neigh_event_send(n, NULL);
1286 rt->rt_gateway = new_gw;
1287 rt->rt_flags |= RTCF_REDIRECTED;
1288 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1294 /* called in rcu_read_lock() section */
1295 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1296 __be32 saddr, struct net_device *dev)
1299 struct in_device *in_dev = __in_dev_get_rcu(dev);
1300 __be32 skeys[2] = { saddr, 0 };
1301 int ikeys[2] = { dev->ifindex, 0 };
1308 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1309 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1310 ipv4_is_zeronet(new_gw))
1311 goto reject_redirect;
1313 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1314 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1315 goto reject_redirect;
1316 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1317 goto reject_redirect;
1319 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1320 goto reject_redirect;
1323 for (s = 0; s < 2; s++) {
1324 for (i = 0; i < 2; i++) {
1326 struct rtable __rcu **rthp;
1329 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1331 rthp = &rt_hash_table[hash].chain;
1333 while ((rt = rcu_dereference(*rthp)) != NULL) {
1334 rthp = &rt->dst.rt_next;
1336 if (rt->rt_key_dst != daddr ||
1337 rt->rt_key_src != skeys[s] ||
1338 rt->rt_oif != ikeys[i] ||
1339 rt_is_input_route(rt) ||
1340 rt_is_expired(rt) ||
1341 !net_eq(dev_net(rt->dst.dev), net) ||
1346 ip_do_redirect(rt, old_gw, new_gw);
1353 #ifdef CONFIG_IP_ROUTE_VERBOSE
1354 if (IN_DEV_LOG_MARTIANS(in_dev))
1355 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1356 " Advised path = %pI4 -> %pI4\n",
1357 &old_gw, dev->name, &new_gw,
1363 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1365 struct rtable *rt = (struct rtable *)dst;
1366 struct dst_entry *ret = dst;
1369 if (dst->obsolete > 0) {
1372 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1374 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1376 rt_genid(dev_net(dst->dev)));
1386 * 1. The first ip_rt_redirect_number redirects are sent
1387 * with exponential backoff, then we stop sending them at all,
1388 * assuming that the host ignores our redirects.
1389 * 2. If we did not see packets requiring redirects
1390 * during ip_rt_redirect_silence, we assume that the host
1391 * forgot redirected route and start to send redirects again.
1393 * This algorithm is much cheaper and more intelligent than dumb load limiting
1396 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1397 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1400 void ip_rt_send_redirect(struct sk_buff *skb)
1402 struct rtable *rt = skb_rtable(skb);
1403 struct in_device *in_dev;
1404 struct inet_peer *peer;
1409 in_dev = __in_dev_get_rcu(rt->dst.dev);
1410 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1414 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1417 net = dev_net(rt->dst.dev);
1418 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
1420 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1424 /* No redirected packets during ip_rt_redirect_silence;
1425 * reset the algorithm.
1427 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1428 peer->rate_tokens = 0;
1430 /* Too many ignored redirects; do not send anything
1431 * set dst.rate_last to the last seen redirected packet.
1433 if (peer->rate_tokens >= ip_rt_redirect_number) {
1434 peer->rate_last = jiffies;
1438 /* Check for load limit; set rate_last to the latest sent
1441 if (peer->rate_tokens == 0 ||
1444 (ip_rt_redirect_load << peer->rate_tokens)))) {
1445 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1446 peer->rate_last = jiffies;
1447 ++peer->rate_tokens;
1448 #ifdef CONFIG_IP_ROUTE_VERBOSE
1450 peer->rate_tokens == ip_rt_redirect_number)
1451 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1452 &ip_hdr(skb)->saddr, rt->rt_iif,
1453 &rt->rt_dst, &rt->rt_gateway);
1460 static int ip_error(struct sk_buff *skb)
1462 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1463 struct rtable *rt = skb_rtable(skb);
1464 struct inet_peer *peer;
1470 net = dev_net(rt->dst.dev);
1471 if (!IN_DEV_FORWARD(in_dev)) {
1472 switch (rt->dst.error) {
1474 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
1478 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1484 switch (rt->dst.error) {
1489 code = ICMP_HOST_UNREACH;
1492 code = ICMP_NET_UNREACH;
1493 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1496 code = ICMP_PKT_FILTERED;
1500 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
1505 peer->rate_tokens += now - peer->rate_last;
1506 if (peer->rate_tokens > ip_rt_error_burst)
1507 peer->rate_tokens = ip_rt_error_burst;
1508 peer->rate_last = now;
1509 if (peer->rate_tokens >= ip_rt_error_cost)
1510 peer->rate_tokens -= ip_rt_error_cost;
1516 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1518 out: kfree_skb(skb);
1522 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1524 struct rtable *rt = (struct rtable *) dst;
1528 if (mtu < ip_rt_min_pmtu)
1529 mtu = ip_rt_min_pmtu;
1532 dst_set_expires(&rt->dst, ip_rt_mtu_expires);
1535 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1536 int oif, u32 mark, u8 protocol, int flow_flags)
1538 const struct iphdr *iph = (const struct iphdr *)skb->data;
1542 flowi4_init_output(&fl4, oif, mark, RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
1543 protocol, flow_flags,
1544 iph->daddr, iph->saddr, 0, 0);
1545 rt = __ip_route_output_key(net, &fl4);
1547 ip_rt_update_pmtu(&rt->dst, mtu);
1551 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1553 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1555 const struct inet_sock *inet = inet_sk(sk);
1557 return ipv4_update_pmtu(skb, sock_net(sk), mtu,
1558 sk->sk_bound_dev_if, sk->sk_mark,
1559 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
1560 inet_sk_flowi_flags(sk));
1562 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1564 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1566 struct rtable *rt = (struct rtable *) dst;
1568 if (rt_is_expired(rt))
1573 static void ipv4_dst_destroy(struct dst_entry *dst)
1575 struct rtable *rt = (struct rtable *) dst;
1578 fib_info_put(rt->fi);
1584 static void ipv4_link_failure(struct sk_buff *skb)
1588 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1590 rt = skb_rtable(skb);
1592 dst_set_expires(&rt->dst, 0);
1595 static int ip_rt_bug(struct sk_buff *skb)
1597 pr_debug("%s: %pI4 -> %pI4, %s\n",
1598 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1599 skb->dev ? skb->dev->name : "?");
1606 We do not cache source address of outgoing interface,
1607 because it is used only by IP RR, TS and SRR options,
1608 so that it out of fast path.
1610 BTW remember: "addr" is allowed to be not aligned
1614 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1618 if (rt_is_output_route(rt))
1619 src = ip_hdr(skb)->saddr;
1621 struct fib_result res;
1627 memset(&fl4, 0, sizeof(fl4));
1628 fl4.daddr = iph->daddr;
1629 fl4.saddr = iph->saddr;
1630 fl4.flowi4_tos = RT_TOS(iph->tos);
1631 fl4.flowi4_oif = rt->dst.dev->ifindex;
1632 fl4.flowi4_iif = skb->dev->ifindex;
1633 fl4.flowi4_mark = skb->mark;
1636 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1637 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1639 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1643 memcpy(addr, &src, 4);
1646 #ifdef CONFIG_IP_ROUTE_CLASSID
1647 static void set_class_tag(struct rtable *rt, u32 tag)
1649 if (!(rt->dst.tclassid & 0xFFFF))
1650 rt->dst.tclassid |= tag & 0xFFFF;
1651 if (!(rt->dst.tclassid & 0xFFFF0000))
1652 rt->dst.tclassid |= tag & 0xFFFF0000;
1656 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1658 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1661 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1663 if (advmss > 65535 - 40)
1664 advmss = 65535 - 40;
1669 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1671 const struct rtable *rt = (const struct rtable *) dst;
1672 unsigned int mtu = rt->rt_pmtu;
1674 if (mtu && time_after_eq(jiffies, rt->dst.expires))
1678 mtu = dst_metric_raw(dst, RTAX_MTU);
1680 if (mtu && rt_is_output_route(rt))
1683 mtu = dst->dev->mtu;
1685 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1687 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1691 if (mtu > IP_MAX_MTU)
1697 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1698 struct fib_info *fi)
1700 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1702 atomic_inc(&fi->fib_clntref);
1704 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1707 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1708 const struct fib_result *res,
1709 struct fib_info *fi, u16 type, u32 itag)
1712 if (FIB_RES_GW(*res) &&
1713 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1714 rt->rt_gateway = FIB_RES_GW(*res);
1715 rt_init_metrics(rt, fl4, fi);
1716 #ifdef CONFIG_IP_ROUTE_CLASSID
1717 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1721 #ifdef CONFIG_IP_ROUTE_CLASSID
1722 #ifdef CONFIG_IP_MULTIPLE_TABLES
1723 set_class_tag(rt, fib_rules_tclass(res));
1725 set_class_tag(rt, itag);
1729 static struct rtable *rt_dst_alloc(struct net_device *dev,
1730 bool nopolicy, bool noxfrm)
1732 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1734 (nopolicy ? DST_NOPOLICY : 0) |
1735 (noxfrm ? DST_NOXFRM : 0));
1738 /* called in rcu_read_lock() section */
1739 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1740 u8 tos, struct net_device *dev, int our)
1744 struct in_device *in_dev = __in_dev_get_rcu(dev);
1748 /* Primary sanity checks. */
1753 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1754 skb->protocol != htons(ETH_P_IP))
1757 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1758 if (ipv4_is_loopback(saddr))
1761 if (ipv4_is_zeronet(saddr)) {
1762 if (!ipv4_is_local_multicast(daddr))
1765 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1770 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1771 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1775 #ifdef CONFIG_IP_ROUTE_CLASSID
1776 rth->dst.tclassid = itag;
1778 rth->dst.output = ip_rt_bug;
1780 rth->rt_key_dst = daddr;
1781 rth->rt_key_src = saddr;
1782 rth->rt_genid = rt_genid(dev_net(dev));
1783 rth->rt_flags = RTCF_MULTICAST;
1784 rth->rt_type = RTN_MULTICAST;
1785 rth->rt_key_tos = tos;
1786 rth->rt_dst = daddr;
1787 rth->rt_src = saddr;
1788 rth->rt_route_iif = dev->ifindex;
1789 rth->rt_iif = dev->ifindex;
1791 rth->rt_mark = skb->mark;
1793 rth->rt_gateway = daddr;
1796 rth->dst.input= ip_local_deliver;
1797 rth->rt_flags |= RTCF_LOCAL;
1800 #ifdef CONFIG_IP_MROUTE
1801 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1802 rth->dst.input = ip_mr_input;
1804 RT_CACHE_STAT_INC(in_slow_mc);
1806 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1807 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1808 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1819 static void ip_handle_martian_source(struct net_device *dev,
1820 struct in_device *in_dev,
1821 struct sk_buff *skb,
1825 RT_CACHE_STAT_INC(in_martian_src);
1826 #ifdef CONFIG_IP_ROUTE_VERBOSE
1827 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1829 * RFC1812 recommendation, if source is martian,
1830 * the only hint is MAC header.
1832 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1833 &daddr, &saddr, dev->name);
1834 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1835 print_hex_dump(KERN_WARNING, "ll header: ",
1836 DUMP_PREFIX_OFFSET, 16, 1,
1837 skb_mac_header(skb),
1838 dev->hard_header_len, true);
1844 /* called in rcu_read_lock() section */
1845 static int __mkroute_input(struct sk_buff *skb,
1846 const struct fib_result *res,
1847 struct in_device *in_dev,
1848 __be32 daddr, __be32 saddr, u32 tos,
1849 struct rtable **result)
1853 struct in_device *out_dev;
1854 unsigned int flags = 0;
1857 /* get a working reference to the output device */
1858 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1859 if (out_dev == NULL) {
1860 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1865 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1866 in_dev->dev, in_dev, &itag);
1868 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1875 flags |= RTCF_DIRECTSRC;
1877 if (out_dev == in_dev && err &&
1878 (IN_DEV_SHARED_MEDIA(out_dev) ||
1879 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1880 flags |= RTCF_DOREDIRECT;
1882 if (skb->protocol != htons(ETH_P_IP)) {
1883 /* Not IP (i.e. ARP). Do not create route, if it is
1884 * invalid for proxy arp. DNAT routes are always valid.
1886 * Proxy arp feature have been extended to allow, ARP
1887 * replies back to the same interface, to support
1888 * Private VLAN switch technologies. See arp.c.
1890 if (out_dev == in_dev &&
1891 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1897 rth = rt_dst_alloc(out_dev->dev,
1898 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1899 IN_DEV_CONF_GET(out_dev, NOXFRM));
1905 rth->rt_key_dst = daddr;
1906 rth->rt_key_src = saddr;
1907 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1908 rth->rt_flags = flags;
1909 rth->rt_type = res->type;
1910 rth->rt_key_tos = tos;
1911 rth->rt_dst = daddr;
1912 rth->rt_src = saddr;
1913 rth->rt_route_iif = in_dev->dev->ifindex;
1914 rth->rt_iif = in_dev->dev->ifindex;
1916 rth->rt_mark = skb->mark;
1918 rth->rt_gateway = daddr;
1921 rth->dst.input = ip_forward;
1922 rth->dst.output = ip_output;
1924 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
1932 static int ip_mkroute_input(struct sk_buff *skb,
1933 struct fib_result *res,
1934 const struct flowi4 *fl4,
1935 struct in_device *in_dev,
1936 __be32 daddr, __be32 saddr, u32 tos)
1938 struct rtable *rth = NULL;
1942 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1943 if (res->fi && res->fi->fib_nhs > 1)
1944 fib_select_multipath(res);
1947 /* create a routing cache entry */
1948 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1952 /* put it into the cache */
1953 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
1954 rt_genid(dev_net(rth->dst.dev)));
1955 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
1957 return PTR_ERR(rth);
1962 * NOTE. We drop all the packets that has local source
1963 * addresses, because every properly looped back packet
1964 * must have correct destination already attached by output routine.
1966 * Such approach solves two big problems:
1967 * 1. Not simplex devices are handled properly.
1968 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1969 * called with rcu_read_lock()
1972 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1973 u8 tos, struct net_device *dev)
1975 struct fib_result res;
1976 struct in_device *in_dev = __in_dev_get_rcu(dev);
1978 unsigned int flags = 0;
1983 struct net *net = dev_net(dev);
1985 /* IP on this device is disabled. */
1990 /* Check for the most weird martians, which can be not detected
1994 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1995 goto martian_source;
1997 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2000 /* Accept zero addresses only to limited broadcast;
2001 * I even do not know to fix it or not. Waiting for complains :-)
2003 if (ipv4_is_zeronet(saddr))
2004 goto martian_source;
2006 if (ipv4_is_zeronet(daddr))
2007 goto martian_destination;
2009 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
2010 if (ipv4_is_loopback(daddr))
2011 goto martian_destination;
2013 if (ipv4_is_loopback(saddr))
2014 goto martian_source;
2018 * Now we are ready to route packet.
2021 fl4.flowi4_iif = dev->ifindex;
2022 fl4.flowi4_mark = skb->mark;
2023 fl4.flowi4_tos = tos;
2024 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2027 err = fib_lookup(net, &fl4, &res);
2031 RT_CACHE_STAT_INC(in_slow_tot);
2033 if (res.type == RTN_BROADCAST)
2036 if (res.type == RTN_LOCAL) {
2037 err = fib_validate_source(skb, saddr, daddr, tos,
2038 net->loopback_dev->ifindex,
2039 dev, in_dev, &itag);
2041 goto martian_source_keep_err;
2043 flags |= RTCF_DIRECTSRC;
2047 if (!IN_DEV_FORWARD(in_dev))
2049 if (res.type != RTN_UNICAST)
2050 goto martian_destination;
2052 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2056 if (skb->protocol != htons(ETH_P_IP))
2059 if (!ipv4_is_zeronet(saddr)) {
2060 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2063 goto martian_source_keep_err;
2065 flags |= RTCF_DIRECTSRC;
2067 flags |= RTCF_BROADCAST;
2068 res.type = RTN_BROADCAST;
2069 RT_CACHE_STAT_INC(in_brd);
2072 rth = rt_dst_alloc(net->loopback_dev,
2073 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2077 rth->dst.input= ip_local_deliver;
2078 rth->dst.output= ip_rt_bug;
2079 #ifdef CONFIG_IP_ROUTE_CLASSID
2080 rth->dst.tclassid = itag;
2083 rth->rt_key_dst = daddr;
2084 rth->rt_key_src = saddr;
2085 rth->rt_genid = rt_genid(net);
2086 rth->rt_flags = flags|RTCF_LOCAL;
2087 rth->rt_type = res.type;
2088 rth->rt_key_tos = tos;
2089 rth->rt_dst = daddr;
2090 rth->rt_src = saddr;
2091 rth->rt_route_iif = dev->ifindex;
2092 rth->rt_iif = dev->ifindex;
2094 rth->rt_mark = skb->mark;
2096 rth->rt_gateway = daddr;
2098 if (res.type == RTN_UNREACHABLE) {
2099 rth->dst.input= ip_error;
2100 rth->dst.error= -err;
2101 rth->rt_flags &= ~RTCF_LOCAL;
2103 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2104 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2111 RT_CACHE_STAT_INC(in_no_route);
2112 res.type = RTN_UNREACHABLE;
2118 * Do not cache martian addresses: they should be logged (RFC1812)
2120 martian_destination:
2121 RT_CACHE_STAT_INC(in_martian_dst);
2122 #ifdef CONFIG_IP_ROUTE_VERBOSE
2123 if (IN_DEV_LOG_MARTIANS(in_dev))
2124 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2125 &daddr, &saddr, dev->name);
2138 martian_source_keep_err:
2139 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2143 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2144 u8 tos, struct net_device *dev, bool noref)
2148 int iif = dev->ifindex;
2156 if (!rt_caching(net))
2159 tos &= IPTOS_RT_MASK;
2160 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2162 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2163 rth = rcu_dereference(rth->dst.rt_next)) {
2164 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2165 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2166 (rth->rt_route_iif ^ iif) |
2167 (rth->rt_key_tos ^ tos)) == 0 &&
2168 rth->rt_mark == skb->mark &&
2169 net_eq(dev_net(rth->dst.dev), net) &&
2170 !rt_is_expired(rth)) {
2172 dst_use_noref(&rth->dst, jiffies);
2173 skb_dst_set_noref(skb, &rth->dst);
2175 dst_use(&rth->dst, jiffies);
2176 skb_dst_set(skb, &rth->dst);
2178 RT_CACHE_STAT_INC(in_hit);
2182 RT_CACHE_STAT_INC(in_hlist_search);
2186 /* Multicast recognition logic is moved from route cache to here.
2187 The problem was that too many Ethernet cards have broken/missing
2188 hardware multicast filters :-( As result the host on multicasting
2189 network acquires a lot of useless route cache entries, sort of
2190 SDR messages from all the world. Now we try to get rid of them.
2191 Really, provided software IP multicast filter is organized
2192 reasonably (at least, hashed), it does not result in a slowdown
2193 comparing with route cache reject entries.
2194 Note, that multicast routers are not affected, because
2195 route cache entry is created eventually.
2197 if (ipv4_is_multicast(daddr)) {
2198 struct in_device *in_dev = __in_dev_get_rcu(dev);
2201 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2202 ip_hdr(skb)->protocol);
2204 #ifdef CONFIG_IP_MROUTE
2206 (!ipv4_is_local_multicast(daddr) &&
2207 IN_DEV_MFORWARD(in_dev))
2210 int res = ip_route_input_mc(skb, daddr, saddr,
2219 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2223 EXPORT_SYMBOL(ip_route_input_common);
2225 /* called with rcu_read_lock() */
2226 static struct rtable *__mkroute_output(const struct fib_result *res,
2227 const struct flowi4 *fl4,
2228 __be32 orig_daddr, __be32 orig_saddr,
2229 int orig_oif, __u8 orig_rtos,
2230 struct net_device *dev_out,
2233 struct fib_info *fi = res->fi;
2234 struct in_device *in_dev;
2235 u16 type = res->type;
2238 in_dev = __in_dev_get_rcu(dev_out);
2240 return ERR_PTR(-EINVAL);
2242 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2243 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2244 return ERR_PTR(-EINVAL);
2246 if (ipv4_is_lbcast(fl4->daddr))
2247 type = RTN_BROADCAST;
2248 else if (ipv4_is_multicast(fl4->daddr))
2249 type = RTN_MULTICAST;
2250 else if (ipv4_is_zeronet(fl4->daddr))
2251 return ERR_PTR(-EINVAL);
2253 if (dev_out->flags & IFF_LOOPBACK)
2254 flags |= RTCF_LOCAL;
2256 if (type == RTN_BROADCAST) {
2257 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2259 } else if (type == RTN_MULTICAST) {
2260 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2261 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2263 flags &= ~RTCF_LOCAL;
2264 /* If multicast route do not exist use
2265 * default one, but do not gateway in this case.
2268 if (fi && res->prefixlen < 4)
2272 rth = rt_dst_alloc(dev_out,
2273 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2274 IN_DEV_CONF_GET(in_dev, NOXFRM));
2276 return ERR_PTR(-ENOBUFS);
2278 rth->dst.output = ip_output;
2280 rth->rt_key_dst = orig_daddr;
2281 rth->rt_key_src = orig_saddr;
2282 rth->rt_genid = rt_genid(dev_net(dev_out));
2283 rth->rt_flags = flags;
2284 rth->rt_type = type;
2285 rth->rt_key_tos = orig_rtos;
2286 rth->rt_dst = fl4->daddr;
2287 rth->rt_src = fl4->saddr;
2288 rth->rt_route_iif = 0;
2289 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2290 rth->rt_oif = orig_oif;
2291 rth->rt_mark = fl4->flowi4_mark;
2293 rth->rt_gateway = fl4->daddr;
2296 RT_CACHE_STAT_INC(out_slow_tot);
2298 if (flags & RTCF_LOCAL)
2299 rth->dst.input = ip_local_deliver;
2300 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2301 if (flags & RTCF_LOCAL &&
2302 !(dev_out->flags & IFF_LOOPBACK)) {
2303 rth->dst.output = ip_mc_output;
2304 RT_CACHE_STAT_INC(out_slow_mc);
2306 #ifdef CONFIG_IP_MROUTE
2307 if (type == RTN_MULTICAST) {
2308 if (IN_DEV_MFORWARD(in_dev) &&
2309 !ipv4_is_local_multicast(fl4->daddr)) {
2310 rth->dst.input = ip_mr_input;
2311 rth->dst.output = ip_mc_output;
2317 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2319 if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
2320 rth->dst.flags |= DST_NOCACHE;
2326 * Major route resolver routine.
2327 * called with rcu_read_lock();
2330 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2332 struct net_device *dev_out = NULL;
2333 __u8 tos = RT_FL_TOS(fl4);
2334 unsigned int flags = 0;
2335 struct fib_result res;
2343 #ifdef CONFIG_IP_MULTIPLE_TABLES
2347 orig_daddr = fl4->daddr;
2348 orig_saddr = fl4->saddr;
2349 orig_oif = fl4->flowi4_oif;
2351 fl4->flowi4_iif = net->loopback_dev->ifindex;
2352 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2353 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2354 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2358 rth = ERR_PTR(-EINVAL);
2359 if (ipv4_is_multicast(fl4->saddr) ||
2360 ipv4_is_lbcast(fl4->saddr) ||
2361 ipv4_is_zeronet(fl4->saddr))
2364 /* I removed check for oif == dev_out->oif here.
2365 It was wrong for two reasons:
2366 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2367 is assigned to multiple interfaces.
2368 2. Moreover, we are allowed to send packets with saddr
2369 of another iface. --ANK
2372 if (fl4->flowi4_oif == 0 &&
2373 (ipv4_is_multicast(fl4->daddr) ||
2374 ipv4_is_lbcast(fl4->daddr))) {
2375 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2376 dev_out = __ip_dev_find(net, fl4->saddr, false);
2377 if (dev_out == NULL)
2380 /* Special hack: user can direct multicasts
2381 and limited broadcast via necessary interface
2382 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2383 This hack is not just for fun, it allows
2384 vic,vat and friends to work.
2385 They bind socket to loopback, set ttl to zero
2386 and expect that it will work.
2387 From the viewpoint of routing cache they are broken,
2388 because we are not allowed to build multicast path
2389 with loopback source addr (look, routing cache
2390 cannot know, that ttl is zero, so that packet
2391 will not leave this host and route is valid).
2392 Luckily, this hack is good workaround.
2395 fl4->flowi4_oif = dev_out->ifindex;
2399 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2400 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2401 if (!__ip_dev_find(net, fl4->saddr, false))
2407 if (fl4->flowi4_oif) {
2408 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2409 rth = ERR_PTR(-ENODEV);
2410 if (dev_out == NULL)
2413 /* RACE: Check return value of inet_select_addr instead. */
2414 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2415 rth = ERR_PTR(-ENETUNREACH);
2418 if (ipv4_is_local_multicast(fl4->daddr) ||
2419 ipv4_is_lbcast(fl4->daddr)) {
2421 fl4->saddr = inet_select_addr(dev_out, 0,
2426 if (ipv4_is_multicast(fl4->daddr))
2427 fl4->saddr = inet_select_addr(dev_out, 0,
2429 else if (!fl4->daddr)
2430 fl4->saddr = inet_select_addr(dev_out, 0,
2436 fl4->daddr = fl4->saddr;
2438 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2439 dev_out = net->loopback_dev;
2440 fl4->flowi4_oif = net->loopback_dev->ifindex;
2441 res.type = RTN_LOCAL;
2442 flags |= RTCF_LOCAL;
2446 if (fib_lookup(net, fl4, &res)) {
2449 if (fl4->flowi4_oif) {
2450 /* Apparently, routing tables are wrong. Assume,
2451 that the destination is on link.
2454 Because we are allowed to send to iface
2455 even if it has NO routes and NO assigned
2456 addresses. When oif is specified, routing
2457 tables are looked up with only one purpose:
2458 to catch if destination is gatewayed, rather than
2459 direct. Moreover, if MSG_DONTROUTE is set,
2460 we send packet, ignoring both routing tables
2461 and ifaddr state. --ANK
2464 We could make it even if oif is unknown,
2465 likely IPv6, but we do not.
2468 if (fl4->saddr == 0)
2469 fl4->saddr = inet_select_addr(dev_out, 0,
2471 res.type = RTN_UNICAST;
2474 rth = ERR_PTR(-ENETUNREACH);
2478 if (res.type == RTN_LOCAL) {
2480 if (res.fi->fib_prefsrc)
2481 fl4->saddr = res.fi->fib_prefsrc;
2483 fl4->saddr = fl4->daddr;
2485 dev_out = net->loopback_dev;
2486 fl4->flowi4_oif = dev_out->ifindex;
2488 flags |= RTCF_LOCAL;
2492 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2493 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2494 fib_select_multipath(&res);
2497 if (!res.prefixlen &&
2498 res.table->tb_num_default > 1 &&
2499 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2500 fib_select_default(&res);
2503 fl4->saddr = FIB_RES_PREFSRC(net, res);
2505 dev_out = FIB_RES_DEV(res);
2506 fl4->flowi4_oif = dev_out->ifindex;
2510 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2511 tos, dev_out, flags);
2515 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2516 rt_genid(dev_net(dev_out)));
2517 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2525 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2530 if (!rt_caching(net))
2533 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2536 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2537 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2538 if (rth->rt_key_dst == flp4->daddr &&
2539 rth->rt_key_src == flp4->saddr &&
2540 rt_is_output_route(rth) &&
2541 rth->rt_oif == flp4->flowi4_oif &&
2542 rth->rt_mark == flp4->flowi4_mark &&
2543 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2544 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2545 net_eq(dev_net(rth->dst.dev), net) &&
2546 !rt_is_expired(rth)) {
2547 dst_use(&rth->dst, jiffies);
2548 RT_CACHE_STAT_INC(out_hit);
2549 rcu_read_unlock_bh();
2551 flp4->saddr = rth->rt_src;
2553 flp4->daddr = rth->rt_dst;
2556 RT_CACHE_STAT_INC(out_hlist_search);
2558 rcu_read_unlock_bh();
2561 return ip_route_output_slow(net, flp4);
2563 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2565 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2570 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2572 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2574 return mtu ? : dst->dev->mtu;
2577 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2581 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2587 static struct dst_ops ipv4_dst_blackhole_ops = {
2589 .protocol = cpu_to_be16(ETH_P_IP),
2590 .destroy = ipv4_dst_destroy,
2591 .check = ipv4_blackhole_dst_check,
2592 .mtu = ipv4_blackhole_mtu,
2593 .default_advmss = ipv4_default_advmss,
2594 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2595 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2596 .neigh_lookup = ipv4_neigh_lookup,
2599 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2601 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2602 struct rtable *ort = (struct rtable *) dst_orig;
2605 struct dst_entry *new = &rt->dst;
2608 new->input = dst_discard;
2609 new->output = dst_discard;
2611 new->dev = ort->dst.dev;
2615 rt->rt_key_dst = ort->rt_key_dst;
2616 rt->rt_key_src = ort->rt_key_src;
2617 rt->rt_key_tos = ort->rt_key_tos;
2618 rt->rt_route_iif = ort->rt_route_iif;
2619 rt->rt_iif = ort->rt_iif;
2620 rt->rt_oif = ort->rt_oif;
2621 rt->rt_mark = ort->rt_mark;
2622 rt->rt_pmtu = ort->rt_pmtu;
2624 rt->rt_genid = rt_genid(net);
2625 rt->rt_flags = ort->rt_flags;
2626 rt->rt_type = ort->rt_type;
2627 rt->rt_dst = ort->rt_dst;
2628 rt->rt_src = ort->rt_src;
2629 rt->rt_gateway = ort->rt_gateway;
2632 atomic_inc(&rt->fi->fib_clntref);
2637 dst_release(dst_orig);
2639 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2642 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2645 struct rtable *rt = __ip_route_output_key(net, flp4);
2650 if (flp4->flowi4_proto)
2651 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2652 flowi4_to_flowi(flp4),
2657 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2659 static int rt_fill_info(struct net *net,
2660 struct sk_buff *skb, u32 pid, u32 seq, int event,
2661 int nowait, unsigned int flags)
2663 struct rtable *rt = skb_rtable(skb);
2665 struct nlmsghdr *nlh;
2666 unsigned long expires = 0;
2669 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2673 r = nlmsg_data(nlh);
2674 r->rtm_family = AF_INET;
2675 r->rtm_dst_len = 32;
2677 r->rtm_tos = rt->rt_key_tos;
2678 r->rtm_table = RT_TABLE_MAIN;
2679 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2680 goto nla_put_failure;
2681 r->rtm_type = rt->rt_type;
2682 r->rtm_scope = RT_SCOPE_UNIVERSE;
2683 r->rtm_protocol = RTPROT_UNSPEC;
2684 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2685 if (rt->rt_flags & RTCF_NOTIFY)
2686 r->rtm_flags |= RTM_F_NOTIFY;
2688 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2689 goto nla_put_failure;
2690 if (rt->rt_key_src) {
2691 r->rtm_src_len = 32;
2692 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2693 goto nla_put_failure;
2696 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2697 goto nla_put_failure;
2698 #ifdef CONFIG_IP_ROUTE_CLASSID
2699 if (rt->dst.tclassid &&
2700 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2701 goto nla_put_failure;
2703 if (!rt_is_input_route(rt) &&
2704 rt->rt_src != rt->rt_key_src) {
2705 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2706 goto nla_put_failure;
2708 if (rt->rt_dst != rt->rt_gateway &&
2709 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2710 goto nla_put_failure;
2712 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2713 goto nla_put_failure;
2716 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2717 goto nla_put_failure;
2719 error = rt->dst.error;
2720 expires = rt->dst.expires;
2722 if (time_before(jiffies, expires))
2728 if (rt_is_input_route(rt)) {
2729 #ifdef CONFIG_IP_MROUTE
2730 __be32 dst = rt->rt_dst;
2732 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2733 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2734 int err = ipmr_get_route(net, skb,
2735 rt->rt_src, rt->rt_dst,
2741 goto nla_put_failure;
2743 if (err == -EMSGSIZE)
2744 goto nla_put_failure;
2750 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2751 goto nla_put_failure;
2754 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2755 goto nla_put_failure;
2757 return nlmsg_end(skb, nlh);
2760 nlmsg_cancel(skb, nlh);
2764 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2766 struct net *net = sock_net(in_skb->sk);
2768 struct nlattr *tb[RTA_MAX+1];
2769 struct rtable *rt = NULL;
2775 struct sk_buff *skb;
2777 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2781 rtm = nlmsg_data(nlh);
2783 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2789 /* Reserve room for dummy headers, this skb can pass
2790 through good chunk of routing engine.
2792 skb_reset_mac_header(skb);
2793 skb_reset_network_header(skb);
2795 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2796 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2797 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2799 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2800 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2801 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2802 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2805 struct net_device *dev;
2807 dev = __dev_get_by_index(net, iif);
2813 skb->protocol = htons(ETH_P_IP);
2817 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2820 rt = skb_rtable(skb);
2821 if (err == 0 && rt->dst.error)
2822 err = -rt->dst.error;
2824 struct flowi4 fl4 = {
2827 .flowi4_tos = rtm->rtm_tos,
2828 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2829 .flowi4_mark = mark,
2831 rt = ip_route_output_key(net, &fl4);
2841 skb_dst_set(skb, &rt->dst);
2842 if (rtm->rtm_flags & RTM_F_NOTIFY)
2843 rt->rt_flags |= RTCF_NOTIFY;
2845 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2846 RTM_NEWROUTE, 0, 0);
2850 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2859 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2866 net = sock_net(skb->sk);
2871 s_idx = idx = cb->args[1];
2872 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2873 if (!rt_hash_table[h].chain)
2876 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
2877 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
2878 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
2880 if (rt_is_expired(rt))
2882 skb_dst_set_noref(skb, &rt->dst);
2883 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
2884 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2885 1, NLM_F_MULTI) <= 0) {
2887 rcu_read_unlock_bh();
2892 rcu_read_unlock_bh();
2901 void ip_rt_multicast_event(struct in_device *in_dev)
2903 rt_cache_flush(dev_net(in_dev->dev), 0);
2906 #ifdef CONFIG_SYSCTL
2907 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2908 void __user *buffer,
2909 size_t *lenp, loff_t *ppos)
2916 memcpy(&ctl, __ctl, sizeof(ctl));
2917 ctl.data = &flush_delay;
2918 proc_dointvec(&ctl, write, buffer, lenp, ppos);
2920 net = (struct net *)__ctl->extra1;
2921 rt_cache_flush(net, flush_delay);
2928 static ctl_table ipv4_route_table[] = {
2930 .procname = "gc_thresh",
2931 .data = &ipv4_dst_ops.gc_thresh,
2932 .maxlen = sizeof(int),
2934 .proc_handler = proc_dointvec,
2937 .procname = "max_size",
2938 .data = &ip_rt_max_size,
2939 .maxlen = sizeof(int),
2941 .proc_handler = proc_dointvec,
2944 /* Deprecated. Use gc_min_interval_ms */
2946 .procname = "gc_min_interval",
2947 .data = &ip_rt_gc_min_interval,
2948 .maxlen = sizeof(int),
2950 .proc_handler = proc_dointvec_jiffies,
2953 .procname = "gc_min_interval_ms",
2954 .data = &ip_rt_gc_min_interval,
2955 .maxlen = sizeof(int),
2957 .proc_handler = proc_dointvec_ms_jiffies,
2960 .procname = "gc_timeout",
2961 .data = &ip_rt_gc_timeout,
2962 .maxlen = sizeof(int),
2964 .proc_handler = proc_dointvec_jiffies,
2967 .procname = "gc_interval",
2968 .data = &ip_rt_gc_interval,
2969 .maxlen = sizeof(int),
2971 .proc_handler = proc_dointvec_jiffies,
2974 .procname = "redirect_load",
2975 .data = &ip_rt_redirect_load,
2976 .maxlen = sizeof(int),
2978 .proc_handler = proc_dointvec,
2981 .procname = "redirect_number",
2982 .data = &ip_rt_redirect_number,
2983 .maxlen = sizeof(int),
2985 .proc_handler = proc_dointvec,
2988 .procname = "redirect_silence",
2989 .data = &ip_rt_redirect_silence,
2990 .maxlen = sizeof(int),
2992 .proc_handler = proc_dointvec,
2995 .procname = "error_cost",
2996 .data = &ip_rt_error_cost,
2997 .maxlen = sizeof(int),
2999 .proc_handler = proc_dointvec,
3002 .procname = "error_burst",
3003 .data = &ip_rt_error_burst,
3004 .maxlen = sizeof(int),
3006 .proc_handler = proc_dointvec,
3009 .procname = "gc_elasticity",
3010 .data = &ip_rt_gc_elasticity,
3011 .maxlen = sizeof(int),
3013 .proc_handler = proc_dointvec,
3016 .procname = "mtu_expires",
3017 .data = &ip_rt_mtu_expires,
3018 .maxlen = sizeof(int),
3020 .proc_handler = proc_dointvec_jiffies,
3023 .procname = "min_pmtu",
3024 .data = &ip_rt_min_pmtu,
3025 .maxlen = sizeof(int),
3027 .proc_handler = proc_dointvec,
3030 .procname = "min_adv_mss",
3031 .data = &ip_rt_min_advmss,
3032 .maxlen = sizeof(int),
3034 .proc_handler = proc_dointvec,
3039 static struct ctl_table ipv4_route_flush_table[] = {
3041 .procname = "flush",
3042 .maxlen = sizeof(int),
3044 .proc_handler = ipv4_sysctl_rtcache_flush,
3049 static __net_init int sysctl_route_net_init(struct net *net)
3051 struct ctl_table *tbl;
3053 tbl = ipv4_route_flush_table;
3054 if (!net_eq(net, &init_net)) {
3055 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3059 tbl[0].extra1 = net;
3061 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3062 if (net->ipv4.route_hdr == NULL)
3067 if (tbl != ipv4_route_flush_table)
3073 static __net_exit void sysctl_route_net_exit(struct net *net)
3075 struct ctl_table *tbl;
3077 tbl = net->ipv4.route_hdr->ctl_table_arg;
3078 unregister_net_sysctl_table(net->ipv4.route_hdr);
3079 BUG_ON(tbl == ipv4_route_flush_table);
3083 static __net_initdata struct pernet_operations sysctl_route_ops = {
3084 .init = sysctl_route_net_init,
3085 .exit = sysctl_route_net_exit,
3089 static __net_init int rt_genid_init(struct net *net)
3091 get_random_bytes(&net->ipv4.rt_genid,
3092 sizeof(net->ipv4.rt_genid));
3093 get_random_bytes(&net->ipv4.dev_addr_genid,
3094 sizeof(net->ipv4.dev_addr_genid));
3098 static __net_initdata struct pernet_operations rt_genid_ops = {
3099 .init = rt_genid_init,
3102 static int __net_init ipv4_inetpeer_init(struct net *net)
3104 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3108 inet_peer_base_init(bp);
3109 net->ipv4.peers = bp;
3113 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3115 struct inet_peer_base *bp = net->ipv4.peers;
3117 net->ipv4.peers = NULL;
3118 inetpeer_invalidate_tree(bp);
3122 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3123 .init = ipv4_inetpeer_init,
3124 .exit = ipv4_inetpeer_exit,
3127 #ifdef CONFIG_IP_ROUTE_CLASSID
3128 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3129 #endif /* CONFIG_IP_ROUTE_CLASSID */
3131 static __initdata unsigned long rhash_entries;
3132 static int __init set_rhash_entries(char *str)
3139 ret = kstrtoul(str, 0, &rhash_entries);
3145 __setup("rhash_entries=", set_rhash_entries);
3147 int __init ip_rt_init(void)
3151 #ifdef CONFIG_IP_ROUTE_CLASSID
3152 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3154 panic("IP: failed to allocate ip_rt_acct\n");
3157 ipv4_dst_ops.kmem_cachep =
3158 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3159 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3161 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3163 if (dst_entries_init(&ipv4_dst_ops) < 0)
3164 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3166 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3167 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3169 rt_hash_table = (struct rt_hash_bucket *)
3170 alloc_large_system_hash("IP route cache",
3171 sizeof(struct rt_hash_bucket),
3173 (totalram_pages >= 128 * 1024) ?
3179 rhash_entries ? 0 : 512 * 1024);
3180 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3181 rt_hash_lock_init();
3183 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3184 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3189 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3190 expires_ljiffies = jiffies;
3191 schedule_delayed_work(&expires_work,
3192 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3194 if (ip_rt_proc_init())
3195 pr_err("Unable to create route proc files\n");
3198 xfrm4_init(ip_rt_max_size);
3200 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3202 #ifdef CONFIG_SYSCTL
3203 register_pernet_subsys(&sysctl_route_ops);
3205 register_pernet_subsys(&rt_genid_ops);
3206 register_pernet_subsys(&ipv4_inetpeer_ops);
3210 #ifdef CONFIG_SYSCTL
3212 * We really need to sanitize the damn ipv4 init order, then all
3213 * this nonsense will go away.
3215 void __init ip_static_sysctl_init(void)
3217 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);