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, struct sock *sk,
152 struct sk_buff *skb, u32 mtu);
153 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
154 struct sk_buff *skb);
155 static int rt_garbage_collect(struct dst_ops *ops);
157 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
162 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
168 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
172 static struct dst_ops ipv4_dst_ops = {
174 .protocol = cpu_to_be16(ETH_P_IP),
175 .gc = rt_garbage_collect,
176 .check = ipv4_dst_check,
177 .default_advmss = ipv4_default_advmss,
179 .cow_metrics = ipv4_cow_metrics,
180 .destroy = ipv4_dst_destroy,
181 .ifdown = ipv4_dst_ifdown,
182 .negative_advice = ipv4_negative_advice,
183 .link_failure = ipv4_link_failure,
184 .update_pmtu = ip_rt_update_pmtu,
185 .redirect = ip_do_redirect,
186 .local_out = __ip_local_out,
187 .neigh_lookup = ipv4_neigh_lookup,
190 #define ECN_OR_COST(class) TC_PRIO_##class
192 const __u8 ip_tos2prio[16] = {
194 ECN_OR_COST(BESTEFFORT),
196 ECN_OR_COST(BESTEFFORT),
202 ECN_OR_COST(INTERACTIVE),
204 ECN_OR_COST(INTERACTIVE),
205 TC_PRIO_INTERACTIVE_BULK,
206 ECN_OR_COST(INTERACTIVE_BULK),
207 TC_PRIO_INTERACTIVE_BULK,
208 ECN_OR_COST(INTERACTIVE_BULK)
210 EXPORT_SYMBOL(ip_tos2prio);
216 /* The locking scheme is rather straight forward:
218 * 1) Read-Copy Update protects the buckets of the central route hash.
219 * 2) Only writers remove entries, and they hold the lock
220 * as they look at rtable reference counts.
221 * 3) Only readers acquire references to rtable entries,
222 * they do so with atomic increments and with the
226 struct rt_hash_bucket {
227 struct rtable __rcu *chain;
230 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
231 defined(CONFIG_PROVE_LOCKING)
233 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
234 * The size of this table is a power of two and depends on the number of CPUS.
235 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
237 #ifdef CONFIG_LOCKDEP
238 # define RT_HASH_LOCK_SZ 256
241 # define RT_HASH_LOCK_SZ 4096
243 # define RT_HASH_LOCK_SZ 2048
245 # define RT_HASH_LOCK_SZ 1024
247 # define RT_HASH_LOCK_SZ 512
249 # define RT_HASH_LOCK_SZ 256
253 static spinlock_t *rt_hash_locks;
254 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
256 static __init void rt_hash_lock_init(void)
260 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
263 panic("IP: failed to allocate rt_hash_locks\n");
265 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
266 spin_lock_init(&rt_hash_locks[i]);
269 # define rt_hash_lock_addr(slot) NULL
271 static inline void rt_hash_lock_init(void)
276 static struct rt_hash_bucket *rt_hash_table __read_mostly;
277 static unsigned int rt_hash_mask __read_mostly;
278 static unsigned int rt_hash_log __read_mostly;
280 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
281 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
283 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
286 return jhash_3words((__force u32)daddr, (__force u32)saddr,
291 static inline int rt_genid(struct net *net)
293 return atomic_read(&net->ipv4.rt_genid);
296 #ifdef CONFIG_PROC_FS
297 struct rt_cache_iter_state {
298 struct seq_net_private p;
303 static struct rtable *rt_cache_get_first(struct seq_file *seq)
305 struct rt_cache_iter_state *st = seq->private;
306 struct rtable *r = NULL;
308 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
309 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
312 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
314 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
315 r->rt_genid == st->genid)
317 r = rcu_dereference_bh(r->dst.rt_next);
319 rcu_read_unlock_bh();
324 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
327 struct rt_cache_iter_state *st = seq->private;
329 r = rcu_dereference_bh(r->dst.rt_next);
331 rcu_read_unlock_bh();
333 if (--st->bucket < 0)
335 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
337 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
342 static struct rtable *rt_cache_get_next(struct seq_file *seq,
345 struct rt_cache_iter_state *st = seq->private;
346 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
347 if (dev_net(r->dst.dev) != seq_file_net(seq))
349 if (r->rt_genid == st->genid)
355 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
357 struct rtable *r = rt_cache_get_first(seq);
360 while (pos && (r = rt_cache_get_next(seq, r)))
362 return pos ? NULL : r;
365 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
367 struct rt_cache_iter_state *st = seq->private;
369 return rt_cache_get_idx(seq, *pos - 1);
370 st->genid = rt_genid(seq_file_net(seq));
371 return SEQ_START_TOKEN;
374 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
378 if (v == SEQ_START_TOKEN)
379 r = rt_cache_get_first(seq);
381 r = rt_cache_get_next(seq, v);
386 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
388 if (v && v != SEQ_START_TOKEN)
389 rcu_read_unlock_bh();
392 static int rt_cache_seq_show(struct seq_file *seq, void *v)
394 if (v == SEQ_START_TOKEN)
395 seq_printf(seq, "%-127s\n",
396 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
397 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
400 struct rtable *r = v;
403 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
404 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
405 r->dst.dev ? r->dst.dev->name : "*",
406 (__force u32)r->rt_dst,
407 (__force u32)r->rt_gateway,
408 r->rt_flags, atomic_read(&r->dst.__refcnt),
409 r->dst.__use, 0, (__force u32)r->rt_src,
410 dst_metric_advmss(&r->dst) + 40,
411 dst_metric(&r->dst, RTAX_WINDOW), 0,
415 seq_printf(seq, "%*s\n", 127 - len, "");
420 static const struct seq_operations rt_cache_seq_ops = {
421 .start = rt_cache_seq_start,
422 .next = rt_cache_seq_next,
423 .stop = rt_cache_seq_stop,
424 .show = rt_cache_seq_show,
427 static int rt_cache_seq_open(struct inode *inode, struct file *file)
429 return seq_open_net(inode, file, &rt_cache_seq_ops,
430 sizeof(struct rt_cache_iter_state));
433 static const struct file_operations rt_cache_seq_fops = {
434 .owner = THIS_MODULE,
435 .open = rt_cache_seq_open,
438 .release = seq_release_net,
442 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
447 return SEQ_START_TOKEN;
449 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
450 if (!cpu_possible(cpu))
453 return &per_cpu(rt_cache_stat, cpu);
458 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
462 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
463 if (!cpu_possible(cpu))
466 return &per_cpu(rt_cache_stat, cpu);
472 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
477 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
479 struct rt_cache_stat *st = v;
481 if (v == SEQ_START_TOKEN) {
482 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");
486 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
487 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
488 dst_entries_get_slow(&ipv4_dst_ops),
511 static const struct seq_operations rt_cpu_seq_ops = {
512 .start = rt_cpu_seq_start,
513 .next = rt_cpu_seq_next,
514 .stop = rt_cpu_seq_stop,
515 .show = rt_cpu_seq_show,
519 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
521 return seq_open(file, &rt_cpu_seq_ops);
524 static const struct file_operations rt_cpu_seq_fops = {
525 .owner = THIS_MODULE,
526 .open = rt_cpu_seq_open,
529 .release = seq_release,
532 #ifdef CONFIG_IP_ROUTE_CLASSID
533 static int rt_acct_proc_show(struct seq_file *m, void *v)
535 struct ip_rt_acct *dst, *src;
538 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
542 for_each_possible_cpu(i) {
543 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
544 for (j = 0; j < 256; j++) {
545 dst[j].o_bytes += src[j].o_bytes;
546 dst[j].o_packets += src[j].o_packets;
547 dst[j].i_bytes += src[j].i_bytes;
548 dst[j].i_packets += src[j].i_packets;
552 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
557 static int rt_acct_proc_open(struct inode *inode, struct file *file)
559 return single_open(file, rt_acct_proc_show, NULL);
562 static const struct file_operations rt_acct_proc_fops = {
563 .owner = THIS_MODULE,
564 .open = rt_acct_proc_open,
567 .release = single_release,
571 static int __net_init ip_rt_do_proc_init(struct net *net)
573 struct proc_dir_entry *pde;
575 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
580 pde = proc_create("rt_cache", S_IRUGO,
581 net->proc_net_stat, &rt_cpu_seq_fops);
585 #ifdef CONFIG_IP_ROUTE_CLASSID
586 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
592 #ifdef CONFIG_IP_ROUTE_CLASSID
594 remove_proc_entry("rt_cache", net->proc_net_stat);
597 remove_proc_entry("rt_cache", net->proc_net);
602 static void __net_exit ip_rt_do_proc_exit(struct net *net)
604 remove_proc_entry("rt_cache", net->proc_net_stat);
605 remove_proc_entry("rt_cache", net->proc_net);
606 #ifdef CONFIG_IP_ROUTE_CLASSID
607 remove_proc_entry("rt_acct", net->proc_net);
611 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
612 .init = ip_rt_do_proc_init,
613 .exit = ip_rt_do_proc_exit,
616 static int __init ip_rt_proc_init(void)
618 return register_pernet_subsys(&ip_rt_proc_ops);
622 static inline int ip_rt_proc_init(void)
626 #endif /* CONFIG_PROC_FS */
628 static inline void rt_free(struct rtable *rt)
630 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
633 static inline void rt_drop(struct rtable *rt)
636 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
639 static inline int rt_fast_clean(struct rtable *rth)
641 /* Kill broadcast/multicast entries very aggresively, if they
642 collide in hash table with more useful entries */
643 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
644 rt_is_input_route(rth) && rth->dst.rt_next;
647 static inline int rt_valuable(struct rtable *rth)
649 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
653 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
658 if (atomic_read(&rth->dst.__refcnt))
661 age = jiffies - rth->dst.lastuse;
662 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
663 (age <= tmo2 && rt_valuable(rth)))
669 /* Bits of score are:
671 * 30: not quite useless
672 * 29..0: usage counter
674 static inline u32 rt_score(struct rtable *rt)
676 u32 score = jiffies - rt->dst.lastuse;
678 score = ~score & ~(3<<30);
683 if (rt_is_output_route(rt) ||
684 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
690 static inline bool rt_caching(const struct net *net)
692 return net->ipv4.current_rt_cache_rebuild_count <=
693 net->ipv4.sysctl_rt_cache_rebuild_count;
696 static inline bool compare_hash_inputs(const struct rtable *rt1,
697 const struct rtable *rt2)
699 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
700 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
701 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
704 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
706 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
707 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
708 (rt1->rt_mark ^ rt2->rt_mark) |
709 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
710 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
711 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
714 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
716 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
719 static inline int rt_is_expired(struct rtable *rth)
721 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
725 * Perform a full scan of hash table and free all entries.
726 * Can be called by a softirq or a process.
727 * In the later case, we want to be reschedule if necessary
729 static void rt_do_flush(struct net *net, int process_context)
732 struct rtable *rth, *next;
734 for (i = 0; i <= rt_hash_mask; i++) {
735 struct rtable __rcu **pprev;
738 if (process_context && need_resched())
740 rth = rcu_access_pointer(rt_hash_table[i].chain);
744 spin_lock_bh(rt_hash_lock_addr(i));
747 pprev = &rt_hash_table[i].chain;
748 rth = rcu_dereference_protected(*pprev,
749 lockdep_is_held(rt_hash_lock_addr(i)));
752 next = rcu_dereference_protected(rth->dst.rt_next,
753 lockdep_is_held(rt_hash_lock_addr(i)));
756 net_eq(dev_net(rth->dst.dev), net)) {
757 rcu_assign_pointer(*pprev, next);
758 rcu_assign_pointer(rth->dst.rt_next, list);
761 pprev = &rth->dst.rt_next;
766 spin_unlock_bh(rt_hash_lock_addr(i));
768 for (; list; list = next) {
769 next = rcu_dereference_protected(list->dst.rt_next, 1);
776 * While freeing expired entries, we compute average chain length
777 * and standard deviation, using fixed-point arithmetic.
778 * This to have an estimation of rt_chain_length_max
779 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
780 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
784 #define ONE (1UL << FRACT_BITS)
787 * Given a hash chain and an item in this hash chain,
788 * find if a previous entry has the same hash_inputs
789 * (but differs on tos, mark or oif)
790 * Returns 0 if an alias is found.
791 * Returns ONE if rth has no alias before itself.
793 static int has_noalias(const struct rtable *head, const struct rtable *rth)
795 const struct rtable *aux = head;
798 if (compare_hash_inputs(aux, rth))
800 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
805 static void rt_check_expire(void)
807 static unsigned int rover;
808 unsigned int i = rover, goal;
810 struct rtable __rcu **rthp;
811 unsigned long samples = 0;
812 unsigned long sum = 0, sum2 = 0;
816 delta = jiffies - expires_ljiffies;
817 expires_ljiffies = jiffies;
818 mult = ((u64)delta) << rt_hash_log;
819 if (ip_rt_gc_timeout > 1)
820 do_div(mult, ip_rt_gc_timeout);
821 goal = (unsigned int)mult;
822 if (goal > rt_hash_mask)
823 goal = rt_hash_mask + 1;
824 for (; goal > 0; goal--) {
825 unsigned long tmo = ip_rt_gc_timeout;
826 unsigned long length;
828 i = (i + 1) & rt_hash_mask;
829 rthp = &rt_hash_table[i].chain;
836 if (rcu_dereference_raw(*rthp) == NULL)
839 spin_lock_bh(rt_hash_lock_addr(i));
840 while ((rth = rcu_dereference_protected(*rthp,
841 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
842 prefetch(rth->dst.rt_next);
843 if (rt_is_expired(rth) ||
844 rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
845 *rthp = rth->dst.rt_next;
850 /* We only count entries on a chain with equal
851 * hash inputs once so that entries for
852 * different QOS levels, and other non-hash
853 * input attributes don't unfairly skew the
857 rthp = &rth->dst.rt_next;
858 length += has_noalias(rt_hash_table[i].chain, rth);
860 spin_unlock_bh(rt_hash_lock_addr(i));
862 sum2 += length*length;
865 unsigned long avg = sum / samples;
866 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
867 rt_chain_length_max = max_t(unsigned long,
869 (avg + 4*sd) >> FRACT_BITS);
875 * rt_worker_func() is run in process context.
876 * we call rt_check_expire() to scan part of the hash table
878 static void rt_worker_func(struct work_struct *work)
881 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
885 * Perturbation of rt_genid by a small quantity [1..256]
886 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
887 * many times (2^24) without giving recent rt_genid.
888 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
890 static void rt_cache_invalidate(struct net *net)
892 unsigned char shuffle;
894 get_random_bytes(&shuffle, sizeof(shuffle));
895 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
899 * delay < 0 : invalidate cache (fast : entries will be deleted later)
900 * delay >= 0 : invalidate & flush cache (can be long)
902 void rt_cache_flush(struct net *net, int delay)
904 rt_cache_invalidate(net);
906 rt_do_flush(net, !in_softirq());
909 /* Flush previous cache invalidated entries from the cache */
910 void rt_cache_flush_batch(struct net *net)
912 rt_do_flush(net, !in_softirq());
915 static void rt_emergency_hash_rebuild(struct net *net)
917 net_warn_ratelimited("Route hash chain too long!\n");
918 rt_cache_invalidate(net);
922 Short description of GC goals.
924 We want to build algorithm, which will keep routing cache
925 at some equilibrium point, when number of aged off entries
926 is kept approximately equal to newly generated ones.
928 Current expiration strength is variable "expire".
929 We try to adjust it dynamically, so that if networking
930 is idle expires is large enough to keep enough of warm entries,
931 and when load increases it reduces to limit cache size.
934 static int rt_garbage_collect(struct dst_ops *ops)
936 static unsigned long expire = RT_GC_TIMEOUT;
937 static unsigned long last_gc;
939 static int equilibrium;
941 struct rtable __rcu **rthp;
942 unsigned long now = jiffies;
944 int entries = dst_entries_get_fast(&ipv4_dst_ops);
947 * Garbage collection is pretty expensive,
948 * do not make it too frequently.
951 RT_CACHE_STAT_INC(gc_total);
953 if (now - last_gc < ip_rt_gc_min_interval &&
954 entries < ip_rt_max_size) {
955 RT_CACHE_STAT_INC(gc_ignored);
959 entries = dst_entries_get_slow(&ipv4_dst_ops);
960 /* Calculate number of entries, which we want to expire now. */
961 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
963 if (equilibrium < ipv4_dst_ops.gc_thresh)
964 equilibrium = ipv4_dst_ops.gc_thresh;
965 goal = entries - equilibrium;
967 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
968 goal = entries - equilibrium;
971 /* We are in dangerous area. Try to reduce cache really
974 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
975 equilibrium = entries - goal;
978 if (now - last_gc >= ip_rt_gc_min_interval)
989 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
990 unsigned long tmo = expire;
992 k = (k + 1) & rt_hash_mask;
993 rthp = &rt_hash_table[k].chain;
994 spin_lock_bh(rt_hash_lock_addr(k));
995 while ((rth = rcu_dereference_protected(*rthp,
996 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
997 if (!rt_is_expired(rth) &&
998 !rt_may_expire(rth, tmo, expire)) {
1000 rthp = &rth->dst.rt_next;
1003 *rthp = rth->dst.rt_next;
1007 spin_unlock_bh(rt_hash_lock_addr(k));
1016 /* Goal is not achieved. We stop process if:
1018 - if expire reduced to zero. Otherwise, expire is halfed.
1019 - if table is not full.
1020 - if we are called from interrupt.
1021 - jiffies check is just fallback/debug loop breaker.
1022 We will not spin here for long time in any case.
1025 RT_CACHE_STAT_INC(gc_goal_miss);
1032 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1034 } while (!in_softirq() && time_before_eq(jiffies, now));
1036 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1038 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1040 net_warn_ratelimited("dst cache overflow\n");
1041 RT_CACHE_STAT_INC(gc_dst_overflow);
1045 expire += ip_rt_gc_min_interval;
1046 if (expire > ip_rt_gc_timeout ||
1047 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1048 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1049 expire = ip_rt_gc_timeout;
1054 * Returns number of entries in a hash chain that have different hash_inputs
1056 static int slow_chain_length(const struct rtable *head)
1059 const struct rtable *rth = head;
1062 length += has_noalias(head, rth);
1063 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1065 return length >> FRACT_BITS;
1068 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
1069 struct sk_buff *skb,
1072 struct net_device *dev = dst->dev;
1073 const __be32 *pkey = daddr;
1074 const struct rtable *rt;
1075 struct neighbour *n;
1077 rt = (const struct rtable *) dst;
1079 pkey = (const __be32 *) &rt->rt_gateway;
1081 pkey = &ip_hdr(skb)->daddr;
1083 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1086 return neigh_create(&arp_tbl, pkey, dev);
1089 static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1090 struct sk_buff *skb, int ifindex)
1092 struct rtable *rth, *cand;
1093 struct rtable __rcu **rthp, **candp;
1100 min_score = ~(u32)0;
1105 if (!rt_caching(dev_net(rt->dst.dev)) || (rt->dst.flags & DST_NOCACHE)) {
1107 * If we're not caching, just tell the caller we
1108 * were successful and don't touch the route. The
1109 * caller hold the sole reference to the cache entry, and
1110 * it will be released when the caller is done with it.
1111 * If we drop it here, the callers have no way to resolve routes
1112 * when we're not caching. Instead, just point *rp at rt, so
1113 * the caller gets a single use out of the route
1114 * Note that we do rt_free on this new route entry, so that
1115 * once its refcount hits zero, we are still able to reap it
1117 * Note: To avoid expensive rcu stuff for this uncached dst,
1118 * we set DST_NOCACHE so that dst_release() can free dst without
1119 * waiting a grace period.
1122 rt->dst.flags |= DST_NOCACHE;
1126 rthp = &rt_hash_table[hash].chain;
1128 spin_lock_bh(rt_hash_lock_addr(hash));
1129 while ((rth = rcu_dereference_protected(*rthp,
1130 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1131 if (rt_is_expired(rth)) {
1132 *rthp = rth->dst.rt_next;
1136 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1138 *rthp = rth->dst.rt_next;
1140 * Since lookup is lockfree, the deletion
1141 * must be visible to another weakly ordered CPU before
1142 * the insertion at the start of the hash chain.
1144 rcu_assign_pointer(rth->dst.rt_next,
1145 rt_hash_table[hash].chain);
1147 * Since lookup is lockfree, the update writes
1148 * must be ordered for consistency on SMP.
1150 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1152 dst_use(&rth->dst, now);
1153 spin_unlock_bh(rt_hash_lock_addr(hash));
1157 skb_dst_set(skb, &rth->dst);
1161 if (!atomic_read(&rth->dst.__refcnt)) {
1162 u32 score = rt_score(rth);
1164 if (score <= min_score) {
1173 rthp = &rth->dst.rt_next;
1177 /* ip_rt_gc_elasticity used to be average length of chain
1178 * length, when exceeded gc becomes really aggressive.
1180 * The second limit is less certain. At the moment it allows
1181 * only 2 entries per bucket. We will see.
1183 if (chain_length > ip_rt_gc_elasticity) {
1184 *candp = cand->dst.rt_next;
1188 if (chain_length > rt_chain_length_max &&
1189 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1190 struct net *net = dev_net(rt->dst.dev);
1191 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1192 if (!rt_caching(net)) {
1193 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1194 rt->dst.dev->name, num);
1196 rt_emergency_hash_rebuild(net);
1197 spin_unlock_bh(rt_hash_lock_addr(hash));
1199 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1200 ifindex, rt_genid(net));
1205 rt->dst.rt_next = rt_hash_table[hash].chain;
1208 * Since lookup is lockfree, we must make sure
1209 * previous writes to rt are committed to memory
1210 * before making rt visible to other CPUS.
1212 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1214 spin_unlock_bh(rt_hash_lock_addr(hash));
1218 skb_dst_set(skb, &rt->dst);
1223 * Peer allocation may fail only in serious out-of-memory conditions. However
1224 * we still can generate some output.
1225 * Random ID selection looks a bit dangerous because we have no chances to
1226 * select ID being unique in a reasonable period of time.
1227 * But broken packet identifier may be better than no packet at all.
1229 static void ip_select_fb_ident(struct iphdr *iph)
1231 static DEFINE_SPINLOCK(ip_fb_id_lock);
1232 static u32 ip_fallback_id;
1235 spin_lock_bh(&ip_fb_id_lock);
1236 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1237 iph->id = htons(salt & 0xFFFF);
1238 ip_fallback_id = salt;
1239 spin_unlock_bh(&ip_fb_id_lock);
1242 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1244 struct net *net = dev_net(dst->dev);
1245 struct inet_peer *peer;
1247 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
1249 iph->id = htons(inet_getid(peer, more));
1254 ip_select_fb_ident(iph);
1256 EXPORT_SYMBOL(__ip_select_ident);
1258 static void rt_del(unsigned int hash, struct rtable *rt)
1260 struct rtable __rcu **rthp;
1263 rthp = &rt_hash_table[hash].chain;
1264 spin_lock_bh(rt_hash_lock_addr(hash));
1266 while ((aux = rcu_dereference_protected(*rthp,
1267 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1268 if (aux == rt || rt_is_expired(aux)) {
1269 *rthp = aux->dst.rt_next;
1273 rthp = &aux->dst.rt_next;
1275 spin_unlock_bh(rt_hash_lock_addr(hash));
1278 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
1279 const struct iphdr *iph,
1281 u8 prot, u32 mark, int flow_flags)
1284 const struct inet_sock *inet = inet_sk(sk);
1286 oif = sk->sk_bound_dev_if;
1288 tos = RT_CONN_FLAGS(sk);
1289 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
1291 flowi4_init_output(fl4, oif, mark, tos,
1292 RT_SCOPE_UNIVERSE, prot,
1294 iph->daddr, iph->saddr, 0, 0);
1297 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
1298 const struct sock *sk)
1300 const struct iphdr *iph = ip_hdr(skb);
1301 int oif = skb->dev->ifindex;
1302 u8 tos = RT_TOS(iph->tos);
1303 u8 prot = iph->protocol;
1304 u32 mark = skb->mark;
1306 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
1309 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
1311 const struct inet_sock *inet = inet_sk(sk);
1312 const struct ip_options_rcu *inet_opt;
1313 __be32 daddr = inet->inet_daddr;
1316 inet_opt = rcu_dereference(inet->inet_opt);
1317 if (inet_opt && inet_opt->opt.srr)
1318 daddr = inet_opt->opt.faddr;
1319 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
1320 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
1321 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
1322 inet_sk_flowi_flags(sk),
1323 daddr, inet->inet_saddr, 0, 0);
1327 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
1328 const struct sk_buff *skb)
1331 build_skb_flow_key(fl4, skb, sk);
1333 build_sk_flow_key(fl4, sk);
1336 static DEFINE_SEQLOCK(fnhe_seqlock);
1338 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
1340 struct fib_nh_exception *fnhe, *oldest;
1342 oldest = rcu_dereference(hash->chain);
1343 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
1344 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1345 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
1351 static inline u32 fnhe_hashfun(__be32 daddr)
1355 hval = (__force u32) daddr;
1356 hval ^= (hval >> 11) ^ (hval >> 22);
1358 return hval & (FNHE_HASH_SIZE - 1);
1361 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
1362 u32 pmtu, unsigned long expires)
1364 struct fnhe_hash_bucket *hash;
1365 struct fib_nh_exception *fnhe;
1367 u32 hval = fnhe_hashfun(daddr);
1369 write_seqlock_bh(&fnhe_seqlock);
1371 hash = nh->nh_exceptions;
1373 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
1376 nh->nh_exceptions = hash;
1382 for (fnhe = rcu_dereference(hash->chain); fnhe;
1383 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1384 if (fnhe->fnhe_daddr == daddr)
1393 fnhe->fnhe_pmtu = pmtu;
1394 fnhe->fnhe_expires = expires;
1397 if (depth > FNHE_RECLAIM_DEPTH)
1398 fnhe = fnhe_oldest(hash);
1400 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
1404 fnhe->fnhe_next = hash->chain;
1405 rcu_assign_pointer(hash->chain, fnhe);
1407 fnhe->fnhe_daddr = daddr;
1409 fnhe->fnhe_pmtu = pmtu;
1410 fnhe->fnhe_expires = expires;
1413 fnhe->fnhe_stamp = jiffies;
1416 write_sequnlock_bh(&fnhe_seqlock);
1420 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4)
1422 __be32 new_gw = icmp_hdr(skb)->un.gateway;
1423 __be32 old_gw = ip_hdr(skb)->saddr;
1424 struct net_device *dev = skb->dev;
1425 struct in_device *in_dev;
1426 struct fib_result res;
1427 struct neighbour *n;
1430 switch (icmp_hdr(skb)->code & 7) {
1431 case ICMP_REDIR_NET:
1432 case ICMP_REDIR_NETTOS:
1433 case ICMP_REDIR_HOST:
1434 case ICMP_REDIR_HOSTTOS:
1441 if (rt->rt_gateway != old_gw)
1444 in_dev = __in_dev_get_rcu(dev);
1449 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1450 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1451 ipv4_is_zeronet(new_gw))
1452 goto reject_redirect;
1454 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1455 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1456 goto reject_redirect;
1457 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1458 goto reject_redirect;
1460 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1461 goto reject_redirect;
1464 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
1466 if (!(n->nud_state & NUD_VALID)) {
1467 neigh_event_send(n, NULL);
1469 if (fib_lookup(net, fl4, &res) == 0) {
1470 struct fib_nh *nh = &FIB_RES_NH(res);
1472 update_or_create_fnhe(nh, fl4->daddr, new_gw,
1475 rt->rt_gateway = new_gw;
1476 rt->rt_flags |= RTCF_REDIRECTED;
1477 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1484 #ifdef CONFIG_IP_ROUTE_VERBOSE
1485 if (IN_DEV_LOG_MARTIANS(in_dev)) {
1486 const struct iphdr *iph = (const struct iphdr *) skb->data;
1487 __be32 daddr = iph->daddr;
1488 __be32 saddr = iph->saddr;
1490 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1491 " Advised path = %pI4 -> %pI4\n",
1492 &old_gw, dev->name, &new_gw,
1499 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
1504 rt = (struct rtable *) dst;
1506 ip_rt_build_flow_key(&fl4, sk, skb);
1507 __ip_do_redirect(rt, skb, &fl4);
1510 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1512 struct rtable *rt = (struct rtable *)dst;
1513 struct dst_entry *ret = dst;
1516 if (dst->obsolete > 0) {
1519 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1521 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1523 rt_genid(dev_net(dst->dev)));
1533 * 1. The first ip_rt_redirect_number redirects are sent
1534 * with exponential backoff, then we stop sending them at all,
1535 * assuming that the host ignores our redirects.
1536 * 2. If we did not see packets requiring redirects
1537 * during ip_rt_redirect_silence, we assume that the host
1538 * forgot redirected route and start to send redirects again.
1540 * This algorithm is much cheaper and more intelligent than dumb load limiting
1543 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1544 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1547 void ip_rt_send_redirect(struct sk_buff *skb)
1549 struct rtable *rt = skb_rtable(skb);
1550 struct in_device *in_dev;
1551 struct inet_peer *peer;
1556 in_dev = __in_dev_get_rcu(rt->dst.dev);
1557 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1561 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1564 net = dev_net(rt->dst.dev);
1565 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
1567 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1571 /* No redirected packets during ip_rt_redirect_silence;
1572 * reset the algorithm.
1574 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1575 peer->rate_tokens = 0;
1577 /* Too many ignored redirects; do not send anything
1578 * set dst.rate_last to the last seen redirected packet.
1580 if (peer->rate_tokens >= ip_rt_redirect_number) {
1581 peer->rate_last = jiffies;
1585 /* Check for load limit; set rate_last to the latest sent
1588 if (peer->rate_tokens == 0 ||
1591 (ip_rt_redirect_load << peer->rate_tokens)))) {
1592 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1593 peer->rate_last = jiffies;
1594 ++peer->rate_tokens;
1595 #ifdef CONFIG_IP_ROUTE_VERBOSE
1597 peer->rate_tokens == ip_rt_redirect_number)
1598 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1599 &ip_hdr(skb)->saddr, rt->rt_iif,
1600 &rt->rt_dst, &rt->rt_gateway);
1607 static int ip_error(struct sk_buff *skb)
1609 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1610 struct rtable *rt = skb_rtable(skb);
1611 struct inet_peer *peer;
1617 net = dev_net(rt->dst.dev);
1618 if (!IN_DEV_FORWARD(in_dev)) {
1619 switch (rt->dst.error) {
1621 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
1625 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1631 switch (rt->dst.error) {
1636 code = ICMP_HOST_UNREACH;
1639 code = ICMP_NET_UNREACH;
1640 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1643 code = ICMP_PKT_FILTERED;
1647 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
1652 peer->rate_tokens += now - peer->rate_last;
1653 if (peer->rate_tokens > ip_rt_error_burst)
1654 peer->rate_tokens = ip_rt_error_burst;
1655 peer->rate_last = now;
1656 if (peer->rate_tokens >= ip_rt_error_cost)
1657 peer->rate_tokens -= ip_rt_error_cost;
1663 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1665 out: kfree_skb(skb);
1669 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1671 struct fib_result res;
1673 if (mtu < ip_rt_min_pmtu)
1674 mtu = ip_rt_min_pmtu;
1676 if (fib_lookup(dev_net(rt->dst.dev), fl4, &res) == 0) {
1677 struct fib_nh *nh = &FIB_RES_NH(res);
1679 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
1680 jiffies + ip_rt_mtu_expires);
1683 dst_set_expires(&rt->dst, ip_rt_mtu_expires);
1686 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1687 struct sk_buff *skb, u32 mtu)
1689 struct rtable *rt = (struct rtable *) dst;
1692 ip_rt_build_flow_key(&fl4, sk, skb);
1693 __ip_rt_update_pmtu(rt, &fl4, mtu);
1696 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1697 int oif, u32 mark, u8 protocol, int flow_flags)
1699 const struct iphdr *iph = (const struct iphdr *) skb->data;
1703 __build_flow_key(&fl4, NULL, iph, oif,
1704 RT_TOS(iph->tos), protocol, mark, flow_flags);
1705 rt = __ip_route_output_key(net, &fl4);
1707 __ip_rt_update_pmtu(rt, &fl4, mtu);
1711 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1713 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1715 const struct iphdr *iph = (const struct iphdr *) skb->data;
1719 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1720 rt = __ip_route_output_key(sock_net(sk), &fl4);
1722 __ip_rt_update_pmtu(rt, &fl4, mtu);
1726 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1728 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1729 int oif, u32 mark, u8 protocol, int flow_flags)
1731 const struct iphdr *iph = (const struct iphdr *) skb->data;
1735 __build_flow_key(&fl4, NULL, iph, oif,
1736 RT_TOS(iph->tos), protocol, mark, flow_flags);
1737 rt = __ip_route_output_key(net, &fl4);
1739 __ip_do_redirect(rt, skb, &fl4);
1743 EXPORT_SYMBOL_GPL(ipv4_redirect);
1745 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1747 const struct iphdr *iph = (const struct iphdr *) skb->data;
1751 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1752 rt = __ip_route_output_key(sock_net(sk), &fl4);
1754 __ip_do_redirect(rt, skb, &fl4);
1758 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1760 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1762 struct rtable *rt = (struct rtable *) dst;
1764 if (rt_is_expired(rt))
1769 static void ipv4_dst_destroy(struct dst_entry *dst)
1771 struct rtable *rt = (struct rtable *) dst;
1774 fib_info_put(rt->fi);
1780 static void ipv4_link_failure(struct sk_buff *skb)
1784 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1786 rt = skb_rtable(skb);
1788 dst_set_expires(&rt->dst, 0);
1791 static int ip_rt_bug(struct sk_buff *skb)
1793 pr_debug("%s: %pI4 -> %pI4, %s\n",
1794 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1795 skb->dev ? skb->dev->name : "?");
1802 We do not cache source address of outgoing interface,
1803 because it is used only by IP RR, TS and SRR options,
1804 so that it out of fast path.
1806 BTW remember: "addr" is allowed to be not aligned
1810 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1814 if (rt_is_output_route(rt))
1815 src = ip_hdr(skb)->saddr;
1817 struct fib_result res;
1823 memset(&fl4, 0, sizeof(fl4));
1824 fl4.daddr = iph->daddr;
1825 fl4.saddr = iph->saddr;
1826 fl4.flowi4_tos = RT_TOS(iph->tos);
1827 fl4.flowi4_oif = rt->dst.dev->ifindex;
1828 fl4.flowi4_iif = skb->dev->ifindex;
1829 fl4.flowi4_mark = skb->mark;
1832 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1833 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1835 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1839 memcpy(addr, &src, 4);
1842 #ifdef CONFIG_IP_ROUTE_CLASSID
1843 static void set_class_tag(struct rtable *rt, u32 tag)
1845 if (!(rt->dst.tclassid & 0xFFFF))
1846 rt->dst.tclassid |= tag & 0xFFFF;
1847 if (!(rt->dst.tclassid & 0xFFFF0000))
1848 rt->dst.tclassid |= tag & 0xFFFF0000;
1852 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1854 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1857 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1859 if (advmss > 65535 - 40)
1860 advmss = 65535 - 40;
1865 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1867 const struct rtable *rt = (const struct rtable *) dst;
1868 unsigned int mtu = rt->rt_pmtu;
1870 if (mtu && time_after_eq(jiffies, rt->dst.expires))
1874 mtu = dst_metric_raw(dst, RTAX_MTU);
1876 if (mtu && rt_is_output_route(rt))
1879 mtu = dst->dev->mtu;
1881 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1883 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1887 if (mtu > IP_MAX_MTU)
1893 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1894 struct fib_info *fi)
1896 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1898 atomic_inc(&fi->fib_clntref);
1900 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1903 static void rt_bind_exception(struct rtable *rt, struct fib_nh *nh, __be32 daddr)
1905 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1906 struct fib_nh_exception *fnhe;
1909 hval = fnhe_hashfun(daddr);
1912 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1913 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1914 __be32 fnhe_daddr, gw;
1915 unsigned long expires;
1919 seq = read_seqbegin(&fnhe_seqlock);
1920 fnhe_daddr = fnhe->fnhe_daddr;
1922 pmtu = fnhe->fnhe_pmtu;
1923 expires = fnhe->fnhe_expires;
1924 if (read_seqretry(&fnhe_seqlock, seq))
1926 if (daddr != fnhe_daddr)
1929 unsigned long diff = jiffies - expires;
1931 if (time_before(jiffies, expires)) {
1933 dst_set_expires(&rt->dst, diff);
1937 rt->rt_gateway = gw;
1938 fnhe->fnhe_stamp = jiffies;
1943 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1944 const struct fib_result *res,
1945 struct fib_info *fi, u16 type, u32 itag)
1948 struct fib_nh *nh = &FIB_RES_NH(*res);
1950 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
1951 rt->rt_gateway = nh->nh_gw;
1952 if (unlikely(nh->nh_exceptions))
1953 rt_bind_exception(rt, nh, fl4->daddr);
1954 rt_init_metrics(rt, fl4, fi);
1955 #ifdef CONFIG_IP_ROUTE_CLASSID
1956 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1960 #ifdef CONFIG_IP_ROUTE_CLASSID
1961 #ifdef CONFIG_IP_MULTIPLE_TABLES
1962 set_class_tag(rt, res->tclassid);
1964 set_class_tag(rt, itag);
1968 static struct rtable *rt_dst_alloc(struct net_device *dev,
1969 bool nopolicy, bool noxfrm)
1971 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1973 (nopolicy ? DST_NOPOLICY : 0) |
1974 (noxfrm ? DST_NOXFRM : 0));
1977 /* called in rcu_read_lock() section */
1978 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1979 u8 tos, struct net_device *dev, int our)
1983 struct in_device *in_dev = __in_dev_get_rcu(dev);
1987 /* Primary sanity checks. */
1992 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1993 skb->protocol != htons(ETH_P_IP))
1996 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1997 if (ipv4_is_loopback(saddr))
2000 if (ipv4_is_zeronet(saddr)) {
2001 if (!ipv4_is_local_multicast(daddr))
2004 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2009 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
2010 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2014 #ifdef CONFIG_IP_ROUTE_CLASSID
2015 rth->dst.tclassid = itag;
2017 rth->dst.output = ip_rt_bug;
2019 rth->rt_key_dst = daddr;
2020 rth->rt_key_src = saddr;
2021 rth->rt_genid = rt_genid(dev_net(dev));
2022 rth->rt_flags = RTCF_MULTICAST;
2023 rth->rt_type = RTN_MULTICAST;
2024 rth->rt_key_tos = tos;
2025 rth->rt_dst = daddr;
2026 rth->rt_src = saddr;
2027 rth->rt_route_iif = dev->ifindex;
2028 rth->rt_iif = dev->ifindex;
2030 rth->rt_mark = skb->mark;
2032 rth->rt_gateway = daddr;
2035 rth->dst.input= ip_local_deliver;
2036 rth->rt_flags |= RTCF_LOCAL;
2039 #ifdef CONFIG_IP_MROUTE
2040 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
2041 rth->dst.input = ip_mr_input;
2043 RT_CACHE_STAT_INC(in_slow_mc);
2045 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
2046 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
2047 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
2058 static void ip_handle_martian_source(struct net_device *dev,
2059 struct in_device *in_dev,
2060 struct sk_buff *skb,
2064 RT_CACHE_STAT_INC(in_martian_src);
2065 #ifdef CONFIG_IP_ROUTE_VERBOSE
2066 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2068 * RFC1812 recommendation, if source is martian,
2069 * the only hint is MAC header.
2071 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
2072 &daddr, &saddr, dev->name);
2073 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2074 print_hex_dump(KERN_WARNING, "ll header: ",
2075 DUMP_PREFIX_OFFSET, 16, 1,
2076 skb_mac_header(skb),
2077 dev->hard_header_len, true);
2083 /* called in rcu_read_lock() section */
2084 static int __mkroute_input(struct sk_buff *skb,
2085 const struct fib_result *res,
2086 struct in_device *in_dev,
2087 __be32 daddr, __be32 saddr, u32 tos,
2088 struct rtable **result)
2092 struct in_device *out_dev;
2093 unsigned int flags = 0;
2096 /* get a working reference to the output device */
2097 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2098 if (out_dev == NULL) {
2099 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
2104 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2105 in_dev->dev, in_dev, &itag);
2107 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2114 flags |= RTCF_DIRECTSRC;
2116 if (out_dev == in_dev && err &&
2117 (IN_DEV_SHARED_MEDIA(out_dev) ||
2118 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2119 flags |= RTCF_DOREDIRECT;
2121 if (skb->protocol != htons(ETH_P_IP)) {
2122 /* Not IP (i.e. ARP). Do not create route, if it is
2123 * invalid for proxy arp. DNAT routes are always valid.
2125 * Proxy arp feature have been extended to allow, ARP
2126 * replies back to the same interface, to support
2127 * Private VLAN switch technologies. See arp.c.
2129 if (out_dev == in_dev &&
2130 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2136 rth = rt_dst_alloc(out_dev->dev,
2137 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2138 IN_DEV_CONF_GET(out_dev, NOXFRM));
2144 rth->rt_key_dst = daddr;
2145 rth->rt_key_src = saddr;
2146 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2147 rth->rt_flags = flags;
2148 rth->rt_type = res->type;
2149 rth->rt_key_tos = tos;
2150 rth->rt_dst = daddr;
2151 rth->rt_src = saddr;
2152 rth->rt_route_iif = in_dev->dev->ifindex;
2153 rth->rt_iif = in_dev->dev->ifindex;
2155 rth->rt_mark = skb->mark;
2157 rth->rt_gateway = daddr;
2160 rth->dst.input = ip_forward;
2161 rth->dst.output = ip_output;
2163 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2171 static int ip_mkroute_input(struct sk_buff *skb,
2172 struct fib_result *res,
2173 const struct flowi4 *fl4,
2174 struct in_device *in_dev,
2175 __be32 daddr, __be32 saddr, u32 tos)
2177 struct rtable *rth = NULL;
2181 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2182 if (res->fi && res->fi->fib_nhs > 1)
2183 fib_select_multipath(res);
2186 /* create a routing cache entry */
2187 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2191 /* put it into the cache */
2192 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2193 rt_genid(dev_net(rth->dst.dev)));
2194 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2196 return PTR_ERR(rth);
2201 * NOTE. We drop all the packets that has local source
2202 * addresses, because every properly looped back packet
2203 * must have correct destination already attached by output routine.
2205 * Such approach solves two big problems:
2206 * 1. Not simplex devices are handled properly.
2207 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2208 * called with rcu_read_lock()
2211 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2212 u8 tos, struct net_device *dev)
2214 struct fib_result res;
2215 struct in_device *in_dev = __in_dev_get_rcu(dev);
2217 unsigned int flags = 0;
2222 struct net *net = dev_net(dev);
2224 /* IP on this device is disabled. */
2229 /* Check for the most weird martians, which can be not detected
2233 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2234 goto martian_source;
2236 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2239 /* Accept zero addresses only to limited broadcast;
2240 * I even do not know to fix it or not. Waiting for complains :-)
2242 if (ipv4_is_zeronet(saddr))
2243 goto martian_source;
2245 if (ipv4_is_zeronet(daddr))
2246 goto martian_destination;
2248 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
2249 if (ipv4_is_loopback(daddr))
2250 goto martian_destination;
2252 if (ipv4_is_loopback(saddr))
2253 goto martian_source;
2257 * Now we are ready to route packet.
2260 fl4.flowi4_iif = dev->ifindex;
2261 fl4.flowi4_mark = skb->mark;
2262 fl4.flowi4_tos = tos;
2263 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2266 err = fib_lookup(net, &fl4, &res);
2270 RT_CACHE_STAT_INC(in_slow_tot);
2272 if (res.type == RTN_BROADCAST)
2275 if (res.type == RTN_LOCAL) {
2276 err = fib_validate_source(skb, saddr, daddr, tos,
2277 net->loopback_dev->ifindex,
2278 dev, in_dev, &itag);
2280 goto martian_source_keep_err;
2282 flags |= RTCF_DIRECTSRC;
2286 if (!IN_DEV_FORWARD(in_dev))
2288 if (res.type != RTN_UNICAST)
2289 goto martian_destination;
2291 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2295 if (skb->protocol != htons(ETH_P_IP))
2298 if (!ipv4_is_zeronet(saddr)) {
2299 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2302 goto martian_source_keep_err;
2304 flags |= RTCF_DIRECTSRC;
2306 flags |= RTCF_BROADCAST;
2307 res.type = RTN_BROADCAST;
2308 RT_CACHE_STAT_INC(in_brd);
2311 rth = rt_dst_alloc(net->loopback_dev,
2312 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2316 rth->dst.input= ip_local_deliver;
2317 rth->dst.output= ip_rt_bug;
2318 #ifdef CONFIG_IP_ROUTE_CLASSID
2319 rth->dst.tclassid = itag;
2322 rth->rt_key_dst = daddr;
2323 rth->rt_key_src = saddr;
2324 rth->rt_genid = rt_genid(net);
2325 rth->rt_flags = flags|RTCF_LOCAL;
2326 rth->rt_type = res.type;
2327 rth->rt_key_tos = tos;
2328 rth->rt_dst = daddr;
2329 rth->rt_src = saddr;
2330 rth->rt_route_iif = dev->ifindex;
2331 rth->rt_iif = dev->ifindex;
2333 rth->rt_mark = skb->mark;
2335 rth->rt_gateway = daddr;
2337 if (res.type == RTN_UNREACHABLE) {
2338 rth->dst.input= ip_error;
2339 rth->dst.error= -err;
2340 rth->rt_flags &= ~RTCF_LOCAL;
2342 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2343 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2350 RT_CACHE_STAT_INC(in_no_route);
2351 res.type = RTN_UNREACHABLE;
2357 * Do not cache martian addresses: they should be logged (RFC1812)
2359 martian_destination:
2360 RT_CACHE_STAT_INC(in_martian_dst);
2361 #ifdef CONFIG_IP_ROUTE_VERBOSE
2362 if (IN_DEV_LOG_MARTIANS(in_dev))
2363 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2364 &daddr, &saddr, dev->name);
2377 martian_source_keep_err:
2378 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2382 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2383 u8 tos, struct net_device *dev, bool noref)
2387 int iif = dev->ifindex;
2395 if (!rt_caching(net))
2398 tos &= IPTOS_RT_MASK;
2399 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2401 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2402 rth = rcu_dereference(rth->dst.rt_next)) {
2403 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2404 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2405 (rth->rt_route_iif ^ iif) |
2406 (rth->rt_key_tos ^ tos)) == 0 &&
2407 rth->rt_mark == skb->mark &&
2408 net_eq(dev_net(rth->dst.dev), net) &&
2409 !rt_is_expired(rth)) {
2411 dst_use_noref(&rth->dst, jiffies);
2412 skb_dst_set_noref(skb, &rth->dst);
2414 dst_use(&rth->dst, jiffies);
2415 skb_dst_set(skb, &rth->dst);
2417 RT_CACHE_STAT_INC(in_hit);
2421 RT_CACHE_STAT_INC(in_hlist_search);
2425 /* Multicast recognition logic is moved from route cache to here.
2426 The problem was that too many Ethernet cards have broken/missing
2427 hardware multicast filters :-( As result the host on multicasting
2428 network acquires a lot of useless route cache entries, sort of
2429 SDR messages from all the world. Now we try to get rid of them.
2430 Really, provided software IP multicast filter is organized
2431 reasonably (at least, hashed), it does not result in a slowdown
2432 comparing with route cache reject entries.
2433 Note, that multicast routers are not affected, because
2434 route cache entry is created eventually.
2436 if (ipv4_is_multicast(daddr)) {
2437 struct in_device *in_dev = __in_dev_get_rcu(dev);
2440 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2441 ip_hdr(skb)->protocol);
2443 #ifdef CONFIG_IP_MROUTE
2445 (!ipv4_is_local_multicast(daddr) &&
2446 IN_DEV_MFORWARD(in_dev))
2449 int res = ip_route_input_mc(skb, daddr, saddr,
2458 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2462 EXPORT_SYMBOL(ip_route_input_common);
2464 /* called with rcu_read_lock() */
2465 static struct rtable *__mkroute_output(const struct fib_result *res,
2466 const struct flowi4 *fl4,
2467 __be32 orig_daddr, __be32 orig_saddr,
2468 int orig_oif, __u8 orig_rtos,
2469 struct net_device *dev_out,
2472 struct fib_info *fi = res->fi;
2473 struct in_device *in_dev;
2474 u16 type = res->type;
2477 in_dev = __in_dev_get_rcu(dev_out);
2479 return ERR_PTR(-EINVAL);
2481 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2482 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2483 return ERR_PTR(-EINVAL);
2485 if (ipv4_is_lbcast(fl4->daddr))
2486 type = RTN_BROADCAST;
2487 else if (ipv4_is_multicast(fl4->daddr))
2488 type = RTN_MULTICAST;
2489 else if (ipv4_is_zeronet(fl4->daddr))
2490 return ERR_PTR(-EINVAL);
2492 if (dev_out->flags & IFF_LOOPBACK)
2493 flags |= RTCF_LOCAL;
2495 if (type == RTN_BROADCAST) {
2496 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2498 } else if (type == RTN_MULTICAST) {
2499 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2500 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2502 flags &= ~RTCF_LOCAL;
2503 /* If multicast route do not exist use
2504 * default one, but do not gateway in this case.
2507 if (fi && res->prefixlen < 4)
2511 rth = rt_dst_alloc(dev_out,
2512 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2513 IN_DEV_CONF_GET(in_dev, NOXFRM));
2515 return ERR_PTR(-ENOBUFS);
2517 rth->dst.output = ip_output;
2519 rth->rt_key_dst = orig_daddr;
2520 rth->rt_key_src = orig_saddr;
2521 rth->rt_genid = rt_genid(dev_net(dev_out));
2522 rth->rt_flags = flags;
2523 rth->rt_type = type;
2524 rth->rt_key_tos = orig_rtos;
2525 rth->rt_dst = fl4->daddr;
2526 rth->rt_src = fl4->saddr;
2527 rth->rt_route_iif = 0;
2528 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2529 rth->rt_oif = orig_oif;
2530 rth->rt_mark = fl4->flowi4_mark;
2532 rth->rt_gateway = fl4->daddr;
2535 RT_CACHE_STAT_INC(out_slow_tot);
2537 if (flags & RTCF_LOCAL)
2538 rth->dst.input = ip_local_deliver;
2539 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2540 if (flags & RTCF_LOCAL &&
2541 !(dev_out->flags & IFF_LOOPBACK)) {
2542 rth->dst.output = ip_mc_output;
2543 RT_CACHE_STAT_INC(out_slow_mc);
2545 #ifdef CONFIG_IP_MROUTE
2546 if (type == RTN_MULTICAST) {
2547 if (IN_DEV_MFORWARD(in_dev) &&
2548 !ipv4_is_local_multicast(fl4->daddr)) {
2549 rth->dst.input = ip_mr_input;
2550 rth->dst.output = ip_mc_output;
2556 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2558 if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
2559 rth->dst.flags |= DST_NOCACHE;
2565 * Major route resolver routine.
2566 * called with rcu_read_lock();
2569 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2571 struct net_device *dev_out = NULL;
2572 __u8 tos = RT_FL_TOS(fl4);
2573 unsigned int flags = 0;
2574 struct fib_result res;
2584 orig_daddr = fl4->daddr;
2585 orig_saddr = fl4->saddr;
2586 orig_oif = fl4->flowi4_oif;
2588 fl4->flowi4_iif = net->loopback_dev->ifindex;
2589 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2590 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2591 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2595 rth = ERR_PTR(-EINVAL);
2596 if (ipv4_is_multicast(fl4->saddr) ||
2597 ipv4_is_lbcast(fl4->saddr) ||
2598 ipv4_is_zeronet(fl4->saddr))
2601 /* I removed check for oif == dev_out->oif here.
2602 It was wrong for two reasons:
2603 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2604 is assigned to multiple interfaces.
2605 2. Moreover, we are allowed to send packets with saddr
2606 of another iface. --ANK
2609 if (fl4->flowi4_oif == 0 &&
2610 (ipv4_is_multicast(fl4->daddr) ||
2611 ipv4_is_lbcast(fl4->daddr))) {
2612 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2613 dev_out = __ip_dev_find(net, fl4->saddr, false);
2614 if (dev_out == NULL)
2617 /* Special hack: user can direct multicasts
2618 and limited broadcast via necessary interface
2619 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2620 This hack is not just for fun, it allows
2621 vic,vat and friends to work.
2622 They bind socket to loopback, set ttl to zero
2623 and expect that it will work.
2624 From the viewpoint of routing cache they are broken,
2625 because we are not allowed to build multicast path
2626 with loopback source addr (look, routing cache
2627 cannot know, that ttl is zero, so that packet
2628 will not leave this host and route is valid).
2629 Luckily, this hack is good workaround.
2632 fl4->flowi4_oif = dev_out->ifindex;
2636 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2637 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2638 if (!__ip_dev_find(net, fl4->saddr, false))
2644 if (fl4->flowi4_oif) {
2645 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2646 rth = ERR_PTR(-ENODEV);
2647 if (dev_out == NULL)
2650 /* RACE: Check return value of inet_select_addr instead. */
2651 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2652 rth = ERR_PTR(-ENETUNREACH);
2655 if (ipv4_is_local_multicast(fl4->daddr) ||
2656 ipv4_is_lbcast(fl4->daddr)) {
2658 fl4->saddr = inet_select_addr(dev_out, 0,
2663 if (ipv4_is_multicast(fl4->daddr))
2664 fl4->saddr = inet_select_addr(dev_out, 0,
2666 else if (!fl4->daddr)
2667 fl4->saddr = inet_select_addr(dev_out, 0,
2673 fl4->daddr = fl4->saddr;
2675 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2676 dev_out = net->loopback_dev;
2677 fl4->flowi4_oif = net->loopback_dev->ifindex;
2678 res.type = RTN_LOCAL;
2679 flags |= RTCF_LOCAL;
2683 if (fib_lookup(net, fl4, &res)) {
2686 if (fl4->flowi4_oif) {
2687 /* Apparently, routing tables are wrong. Assume,
2688 that the destination is on link.
2691 Because we are allowed to send to iface
2692 even if it has NO routes and NO assigned
2693 addresses. When oif is specified, routing
2694 tables are looked up with only one purpose:
2695 to catch if destination is gatewayed, rather than
2696 direct. Moreover, if MSG_DONTROUTE is set,
2697 we send packet, ignoring both routing tables
2698 and ifaddr state. --ANK
2701 We could make it even if oif is unknown,
2702 likely IPv6, but we do not.
2705 if (fl4->saddr == 0)
2706 fl4->saddr = inet_select_addr(dev_out, 0,
2708 res.type = RTN_UNICAST;
2711 rth = ERR_PTR(-ENETUNREACH);
2715 if (res.type == RTN_LOCAL) {
2717 if (res.fi->fib_prefsrc)
2718 fl4->saddr = res.fi->fib_prefsrc;
2720 fl4->saddr = fl4->daddr;
2722 dev_out = net->loopback_dev;
2723 fl4->flowi4_oif = dev_out->ifindex;
2725 flags |= RTCF_LOCAL;
2729 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2730 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2731 fib_select_multipath(&res);
2734 if (!res.prefixlen &&
2735 res.table->tb_num_default > 1 &&
2736 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2737 fib_select_default(&res);
2740 fl4->saddr = FIB_RES_PREFSRC(net, res);
2742 dev_out = FIB_RES_DEV(res);
2743 fl4->flowi4_oif = dev_out->ifindex;
2747 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2748 tos, dev_out, flags);
2752 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2753 rt_genid(dev_net(dev_out)));
2754 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2762 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2767 if (!rt_caching(net))
2770 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2773 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2774 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2775 if (rth->rt_key_dst == flp4->daddr &&
2776 rth->rt_key_src == flp4->saddr &&
2777 rt_is_output_route(rth) &&
2778 rth->rt_oif == flp4->flowi4_oif &&
2779 rth->rt_mark == flp4->flowi4_mark &&
2780 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2781 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2782 net_eq(dev_net(rth->dst.dev), net) &&
2783 !rt_is_expired(rth)) {
2784 dst_use(&rth->dst, jiffies);
2785 RT_CACHE_STAT_INC(out_hit);
2786 rcu_read_unlock_bh();
2788 flp4->saddr = rth->rt_src;
2790 flp4->daddr = rth->rt_dst;
2793 RT_CACHE_STAT_INC(out_hlist_search);
2795 rcu_read_unlock_bh();
2798 return ip_route_output_slow(net, flp4);
2800 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2802 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2807 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2809 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2811 return mtu ? : dst->dev->mtu;
2814 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2815 struct sk_buff *skb, u32 mtu)
2819 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2820 struct sk_buff *skb)
2824 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2830 static struct dst_ops ipv4_dst_blackhole_ops = {
2832 .protocol = cpu_to_be16(ETH_P_IP),
2833 .destroy = ipv4_dst_destroy,
2834 .check = ipv4_blackhole_dst_check,
2835 .mtu = ipv4_blackhole_mtu,
2836 .default_advmss = ipv4_default_advmss,
2837 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2838 .redirect = ipv4_rt_blackhole_redirect,
2839 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2840 .neigh_lookup = ipv4_neigh_lookup,
2843 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2845 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2846 struct rtable *ort = (struct rtable *) dst_orig;
2849 struct dst_entry *new = &rt->dst;
2852 new->input = dst_discard;
2853 new->output = dst_discard;
2855 new->dev = ort->dst.dev;
2859 rt->rt_key_dst = ort->rt_key_dst;
2860 rt->rt_key_src = ort->rt_key_src;
2861 rt->rt_key_tos = ort->rt_key_tos;
2862 rt->rt_route_iif = ort->rt_route_iif;
2863 rt->rt_iif = ort->rt_iif;
2864 rt->rt_oif = ort->rt_oif;
2865 rt->rt_mark = ort->rt_mark;
2866 rt->rt_pmtu = ort->rt_pmtu;
2868 rt->rt_genid = rt_genid(net);
2869 rt->rt_flags = ort->rt_flags;
2870 rt->rt_type = ort->rt_type;
2871 rt->rt_dst = ort->rt_dst;
2872 rt->rt_src = ort->rt_src;
2873 rt->rt_gateway = ort->rt_gateway;
2876 atomic_inc(&rt->fi->fib_clntref);
2881 dst_release(dst_orig);
2883 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2886 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2889 struct rtable *rt = __ip_route_output_key(net, flp4);
2894 if (flp4->flowi4_proto)
2895 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2896 flowi4_to_flowi(flp4),
2901 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2903 static int rt_fill_info(struct net *net,
2904 struct sk_buff *skb, u32 pid, u32 seq, int event,
2905 int nowait, unsigned int flags)
2907 struct rtable *rt = skb_rtable(skb);
2909 struct nlmsghdr *nlh;
2910 unsigned long expires = 0;
2913 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2917 r = nlmsg_data(nlh);
2918 r->rtm_family = AF_INET;
2919 r->rtm_dst_len = 32;
2921 r->rtm_tos = rt->rt_key_tos;
2922 r->rtm_table = RT_TABLE_MAIN;
2923 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2924 goto nla_put_failure;
2925 r->rtm_type = rt->rt_type;
2926 r->rtm_scope = RT_SCOPE_UNIVERSE;
2927 r->rtm_protocol = RTPROT_UNSPEC;
2928 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2929 if (rt->rt_flags & RTCF_NOTIFY)
2930 r->rtm_flags |= RTM_F_NOTIFY;
2932 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2933 goto nla_put_failure;
2934 if (rt->rt_key_src) {
2935 r->rtm_src_len = 32;
2936 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2937 goto nla_put_failure;
2940 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2941 goto nla_put_failure;
2942 #ifdef CONFIG_IP_ROUTE_CLASSID
2943 if (rt->dst.tclassid &&
2944 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2945 goto nla_put_failure;
2947 if (!rt_is_input_route(rt) &&
2948 rt->rt_src != rt->rt_key_src) {
2949 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2950 goto nla_put_failure;
2952 if (rt->rt_dst != rt->rt_gateway &&
2953 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2954 goto nla_put_failure;
2956 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2957 goto nla_put_failure;
2960 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2961 goto nla_put_failure;
2963 error = rt->dst.error;
2964 expires = rt->dst.expires;
2966 if (time_before(jiffies, expires))
2972 if (rt_is_input_route(rt)) {
2973 #ifdef CONFIG_IP_MROUTE
2974 __be32 dst = rt->rt_dst;
2976 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2977 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2978 int err = ipmr_get_route(net, skb,
2979 rt->rt_src, rt->rt_dst,
2985 goto nla_put_failure;
2987 if (err == -EMSGSIZE)
2988 goto nla_put_failure;
2994 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2995 goto nla_put_failure;
2998 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2999 goto nla_put_failure;
3001 return nlmsg_end(skb, nlh);
3004 nlmsg_cancel(skb, nlh);
3008 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
3010 struct net *net = sock_net(in_skb->sk);
3012 struct nlattr *tb[RTA_MAX+1];
3013 struct rtable *rt = NULL;
3019 struct sk_buff *skb;
3021 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
3025 rtm = nlmsg_data(nlh);
3027 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3033 /* Reserve room for dummy headers, this skb can pass
3034 through good chunk of routing engine.
3036 skb_reset_mac_header(skb);
3037 skb_reset_network_header(skb);
3039 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3040 ip_hdr(skb)->protocol = IPPROTO_ICMP;
3041 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
3043 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
3044 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
3045 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3046 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3049 struct net_device *dev;
3051 dev = __dev_get_by_index(net, iif);
3057 skb->protocol = htons(ETH_P_IP);
3061 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3064 rt = skb_rtable(skb);
3065 if (err == 0 && rt->dst.error)
3066 err = -rt->dst.error;
3068 struct flowi4 fl4 = {
3071 .flowi4_tos = rtm->rtm_tos,
3072 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3073 .flowi4_mark = mark,
3075 rt = ip_route_output_key(net, &fl4);
3085 skb_dst_set(skb, &rt->dst);
3086 if (rtm->rtm_flags & RTM_F_NOTIFY)
3087 rt->rt_flags |= RTCF_NOTIFY;
3089 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3090 RTM_NEWROUTE, 0, 0);
3094 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3103 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3110 net = sock_net(skb->sk);
3115 s_idx = idx = cb->args[1];
3116 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3117 if (!rt_hash_table[h].chain)
3120 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3121 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3122 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3124 if (rt_is_expired(rt))
3126 skb_dst_set_noref(skb, &rt->dst);
3127 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3128 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3129 1, NLM_F_MULTI) <= 0) {
3131 rcu_read_unlock_bh();
3136 rcu_read_unlock_bh();
3145 void ip_rt_multicast_event(struct in_device *in_dev)
3147 rt_cache_flush(dev_net(in_dev->dev), 0);
3150 #ifdef CONFIG_SYSCTL
3151 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3152 void __user *buffer,
3153 size_t *lenp, loff_t *ppos)
3160 memcpy(&ctl, __ctl, sizeof(ctl));
3161 ctl.data = &flush_delay;
3162 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3164 net = (struct net *)__ctl->extra1;
3165 rt_cache_flush(net, flush_delay);
3172 static ctl_table ipv4_route_table[] = {
3174 .procname = "gc_thresh",
3175 .data = &ipv4_dst_ops.gc_thresh,
3176 .maxlen = sizeof(int),
3178 .proc_handler = proc_dointvec,
3181 .procname = "max_size",
3182 .data = &ip_rt_max_size,
3183 .maxlen = sizeof(int),
3185 .proc_handler = proc_dointvec,
3188 /* Deprecated. Use gc_min_interval_ms */
3190 .procname = "gc_min_interval",
3191 .data = &ip_rt_gc_min_interval,
3192 .maxlen = sizeof(int),
3194 .proc_handler = proc_dointvec_jiffies,
3197 .procname = "gc_min_interval_ms",
3198 .data = &ip_rt_gc_min_interval,
3199 .maxlen = sizeof(int),
3201 .proc_handler = proc_dointvec_ms_jiffies,
3204 .procname = "gc_timeout",
3205 .data = &ip_rt_gc_timeout,
3206 .maxlen = sizeof(int),
3208 .proc_handler = proc_dointvec_jiffies,
3211 .procname = "gc_interval",
3212 .data = &ip_rt_gc_interval,
3213 .maxlen = sizeof(int),
3215 .proc_handler = proc_dointvec_jiffies,
3218 .procname = "redirect_load",
3219 .data = &ip_rt_redirect_load,
3220 .maxlen = sizeof(int),
3222 .proc_handler = proc_dointvec,
3225 .procname = "redirect_number",
3226 .data = &ip_rt_redirect_number,
3227 .maxlen = sizeof(int),
3229 .proc_handler = proc_dointvec,
3232 .procname = "redirect_silence",
3233 .data = &ip_rt_redirect_silence,
3234 .maxlen = sizeof(int),
3236 .proc_handler = proc_dointvec,
3239 .procname = "error_cost",
3240 .data = &ip_rt_error_cost,
3241 .maxlen = sizeof(int),
3243 .proc_handler = proc_dointvec,
3246 .procname = "error_burst",
3247 .data = &ip_rt_error_burst,
3248 .maxlen = sizeof(int),
3250 .proc_handler = proc_dointvec,
3253 .procname = "gc_elasticity",
3254 .data = &ip_rt_gc_elasticity,
3255 .maxlen = sizeof(int),
3257 .proc_handler = proc_dointvec,
3260 .procname = "mtu_expires",
3261 .data = &ip_rt_mtu_expires,
3262 .maxlen = sizeof(int),
3264 .proc_handler = proc_dointvec_jiffies,
3267 .procname = "min_pmtu",
3268 .data = &ip_rt_min_pmtu,
3269 .maxlen = sizeof(int),
3271 .proc_handler = proc_dointvec,
3274 .procname = "min_adv_mss",
3275 .data = &ip_rt_min_advmss,
3276 .maxlen = sizeof(int),
3278 .proc_handler = proc_dointvec,
3283 static struct ctl_table ipv4_route_flush_table[] = {
3285 .procname = "flush",
3286 .maxlen = sizeof(int),
3288 .proc_handler = ipv4_sysctl_rtcache_flush,
3293 static __net_init int sysctl_route_net_init(struct net *net)
3295 struct ctl_table *tbl;
3297 tbl = ipv4_route_flush_table;
3298 if (!net_eq(net, &init_net)) {
3299 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3303 tbl[0].extra1 = net;
3305 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3306 if (net->ipv4.route_hdr == NULL)
3311 if (tbl != ipv4_route_flush_table)
3317 static __net_exit void sysctl_route_net_exit(struct net *net)
3319 struct ctl_table *tbl;
3321 tbl = net->ipv4.route_hdr->ctl_table_arg;
3322 unregister_net_sysctl_table(net->ipv4.route_hdr);
3323 BUG_ON(tbl == ipv4_route_flush_table);
3327 static __net_initdata struct pernet_operations sysctl_route_ops = {
3328 .init = sysctl_route_net_init,
3329 .exit = sysctl_route_net_exit,
3333 static __net_init int rt_genid_init(struct net *net)
3335 get_random_bytes(&net->ipv4.rt_genid,
3336 sizeof(net->ipv4.rt_genid));
3337 get_random_bytes(&net->ipv4.dev_addr_genid,
3338 sizeof(net->ipv4.dev_addr_genid));
3342 static __net_initdata struct pernet_operations rt_genid_ops = {
3343 .init = rt_genid_init,
3346 static int __net_init ipv4_inetpeer_init(struct net *net)
3348 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3352 inet_peer_base_init(bp);
3353 net->ipv4.peers = bp;
3357 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3359 struct inet_peer_base *bp = net->ipv4.peers;
3361 net->ipv4.peers = NULL;
3362 inetpeer_invalidate_tree(bp);
3366 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3367 .init = ipv4_inetpeer_init,
3368 .exit = ipv4_inetpeer_exit,
3371 #ifdef CONFIG_IP_ROUTE_CLASSID
3372 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3373 #endif /* CONFIG_IP_ROUTE_CLASSID */
3375 static __initdata unsigned long rhash_entries;
3376 static int __init set_rhash_entries(char *str)
3383 ret = kstrtoul(str, 0, &rhash_entries);
3389 __setup("rhash_entries=", set_rhash_entries);
3391 int __init ip_rt_init(void)
3395 #ifdef CONFIG_IP_ROUTE_CLASSID
3396 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3398 panic("IP: failed to allocate ip_rt_acct\n");
3401 ipv4_dst_ops.kmem_cachep =
3402 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3403 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3405 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3407 if (dst_entries_init(&ipv4_dst_ops) < 0)
3408 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3410 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3411 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3413 rt_hash_table = (struct rt_hash_bucket *)
3414 alloc_large_system_hash("IP route cache",
3415 sizeof(struct rt_hash_bucket),
3417 (totalram_pages >= 128 * 1024) ?
3423 rhash_entries ? 0 : 512 * 1024);
3424 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3425 rt_hash_lock_init();
3427 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3428 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3433 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3434 expires_ljiffies = jiffies;
3435 schedule_delayed_work(&expires_work,
3436 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3438 if (ip_rt_proc_init())
3439 pr_err("Unable to create route proc files\n");
3442 xfrm4_init(ip_rt_max_size);
3444 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3446 #ifdef CONFIG_SYSCTL
3447 register_pernet_subsys(&sysctl_route_ops);
3449 register_pernet_subsys(&rt_genid_ops);
3450 register_pernet_subsys(&ipv4_inetpeer_ops);
3454 #ifdef CONFIG_SYSCTL
3456 * We really need to sanitize the damn ipv4 init order, then all
3457 * this nonsense will go away.
3459 void __init ip_static_sysctl_init(void)
3461 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);