2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * ROUTE - implementation of the IP router.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
65 #define pr_fmt(fmt) "IPv4: " fmt
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/bootmem.h>
74 #include <linux/string.h>
75 #include <linux/socket.h>
76 #include <linux/sockios.h>
77 #include <linux/errno.h>
79 #include <linux/inet.h>
80 #include <linux/netdevice.h>
81 #include <linux/proc_fs.h>
82 #include <linux/init.h>
83 #include <linux/workqueue.h>
84 #include <linux/skbuff.h>
85 #include <linux/inetdevice.h>
86 #include <linux/igmp.h>
87 #include <linux/pkt_sched.h>
88 #include <linux/mroute.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/jhash.h>
92 #include <linux/rcupdate.h>
93 #include <linux/times.h>
94 #include <linux/slab.h>
95 #include <linux/prefetch.h>
97 #include <net/net_namespace.h>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/inetpeer.h>
102 #include <net/sock.h>
103 #include <net/ip_fib.h>
106 #include <net/icmp.h>
107 #include <net/xfrm.h>
108 #include <net/netevent.h>
109 #include <net/rtnetlink.h>
111 #include <linux/sysctl.h>
112 #include <linux/kmemleak.h>
114 #include <net/secure_seq.h>
116 #define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
119 #define IP_MAX_MTU 0xFFF0
121 #define RT_GC_TIMEOUT (300*HZ)
123 static int ip_rt_max_size;
124 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
126 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
127 static int ip_rt_redirect_number __read_mostly = 9;
128 static int ip_rt_redirect_load __read_mostly = HZ / 50;
129 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
130 static int ip_rt_error_cost __read_mostly = HZ;
131 static int ip_rt_error_burst __read_mostly = 5 * HZ;
132 static int ip_rt_gc_elasticity __read_mostly = 8;
133 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
134 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
135 static int ip_rt_min_advmss __read_mostly = 256;
136 static int rt_chain_length_max __read_mostly = 20;
138 static struct delayed_work expires_work;
139 static unsigned long expires_ljiffies;
142 * Interface to generic destination cache.
145 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
146 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
147 static unsigned int ipv4_mtu(const struct dst_entry *dst);
148 static void ipv4_dst_destroy(struct dst_entry *dst);
149 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
150 static void ipv4_link_failure(struct sk_buff *skb);
151 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
152 static int rt_garbage_collect(struct dst_ops *ops);
154 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
159 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
161 struct rtable *rt = (struct rtable *) dst;
162 struct inet_peer *peer;
165 peer = rt_get_peer_create(rt, rt->rt_dst);
167 u32 *old_p = __DST_METRICS_PTR(old);
168 unsigned long prev, new;
171 if (inet_metrics_new(peer))
172 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
174 new = (unsigned long) p;
175 prev = cmpxchg(&dst->_metrics, old, new);
178 p = __DST_METRICS_PTR(prev);
179 if (prev & DST_METRICS_READ_ONLY)
183 fib_info_put(rt->fi);
191 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
195 static struct dst_ops ipv4_dst_ops = {
197 .protocol = cpu_to_be16(ETH_P_IP),
198 .gc = rt_garbage_collect,
199 .check = ipv4_dst_check,
200 .default_advmss = ipv4_default_advmss,
202 .cow_metrics = ipv4_cow_metrics,
203 .destroy = ipv4_dst_destroy,
204 .ifdown = ipv4_dst_ifdown,
205 .negative_advice = ipv4_negative_advice,
206 .link_failure = ipv4_link_failure,
207 .update_pmtu = ip_rt_update_pmtu,
208 .local_out = __ip_local_out,
209 .neigh_lookup = ipv4_neigh_lookup,
212 #define ECN_OR_COST(class) TC_PRIO_##class
214 const __u8 ip_tos2prio[16] = {
216 ECN_OR_COST(BESTEFFORT),
218 ECN_OR_COST(BESTEFFORT),
224 ECN_OR_COST(INTERACTIVE),
226 ECN_OR_COST(INTERACTIVE),
227 TC_PRIO_INTERACTIVE_BULK,
228 ECN_OR_COST(INTERACTIVE_BULK),
229 TC_PRIO_INTERACTIVE_BULK,
230 ECN_OR_COST(INTERACTIVE_BULK)
232 EXPORT_SYMBOL(ip_tos2prio);
238 /* The locking scheme is rather straight forward:
240 * 1) Read-Copy Update protects the buckets of the central route hash.
241 * 2) Only writers remove entries, and they hold the lock
242 * as they look at rtable reference counts.
243 * 3) Only readers acquire references to rtable entries,
244 * they do so with atomic increments and with the
248 struct rt_hash_bucket {
249 struct rtable __rcu *chain;
252 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
253 defined(CONFIG_PROVE_LOCKING)
255 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
256 * The size of this table is a power of two and depends on the number of CPUS.
257 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
259 #ifdef CONFIG_LOCKDEP
260 # define RT_HASH_LOCK_SZ 256
263 # define RT_HASH_LOCK_SZ 4096
265 # define RT_HASH_LOCK_SZ 2048
267 # define RT_HASH_LOCK_SZ 1024
269 # define RT_HASH_LOCK_SZ 512
271 # define RT_HASH_LOCK_SZ 256
275 static spinlock_t *rt_hash_locks;
276 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
278 static __init void rt_hash_lock_init(void)
282 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
285 panic("IP: failed to allocate rt_hash_locks\n");
287 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
288 spin_lock_init(&rt_hash_locks[i]);
291 # define rt_hash_lock_addr(slot) NULL
293 static inline void rt_hash_lock_init(void)
298 static struct rt_hash_bucket *rt_hash_table __read_mostly;
299 static unsigned int rt_hash_mask __read_mostly;
300 static unsigned int rt_hash_log __read_mostly;
302 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
303 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
305 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
308 return jhash_3words((__force u32)daddr, (__force u32)saddr,
313 static inline int rt_genid(struct net *net)
315 return atomic_read(&net->ipv4.rt_genid);
318 #ifdef CONFIG_PROC_FS
319 struct rt_cache_iter_state {
320 struct seq_net_private p;
325 static struct rtable *rt_cache_get_first(struct seq_file *seq)
327 struct rt_cache_iter_state *st = seq->private;
328 struct rtable *r = NULL;
330 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
331 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
334 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
336 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
337 r->rt_genid == st->genid)
339 r = rcu_dereference_bh(r->dst.rt_next);
341 rcu_read_unlock_bh();
346 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
349 struct rt_cache_iter_state *st = seq->private;
351 r = rcu_dereference_bh(r->dst.rt_next);
353 rcu_read_unlock_bh();
355 if (--st->bucket < 0)
357 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
359 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
364 static struct rtable *rt_cache_get_next(struct seq_file *seq,
367 struct rt_cache_iter_state *st = seq->private;
368 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
369 if (dev_net(r->dst.dev) != seq_file_net(seq))
371 if (r->rt_genid == st->genid)
377 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
379 struct rtable *r = rt_cache_get_first(seq);
382 while (pos && (r = rt_cache_get_next(seq, r)))
384 return pos ? NULL : r;
387 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
389 struct rt_cache_iter_state *st = seq->private;
391 return rt_cache_get_idx(seq, *pos - 1);
392 st->genid = rt_genid(seq_file_net(seq));
393 return SEQ_START_TOKEN;
396 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
400 if (v == SEQ_START_TOKEN)
401 r = rt_cache_get_first(seq);
403 r = rt_cache_get_next(seq, v);
408 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
410 if (v && v != SEQ_START_TOKEN)
411 rcu_read_unlock_bh();
414 static int rt_cache_seq_show(struct seq_file *seq, void *v)
416 if (v == SEQ_START_TOKEN)
417 seq_printf(seq, "%-127s\n",
418 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
419 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
422 struct rtable *r = v;
425 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
426 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
427 r->dst.dev ? r->dst.dev->name : "*",
428 (__force u32)r->rt_dst,
429 (__force u32)r->rt_gateway,
430 r->rt_flags, atomic_read(&r->dst.__refcnt),
431 r->dst.__use, 0, (__force u32)r->rt_src,
432 dst_metric_advmss(&r->dst) + 40,
433 dst_metric(&r->dst, RTAX_WINDOW), 0,
437 seq_printf(seq, "%*s\n", 127 - len, "");
442 static const struct seq_operations rt_cache_seq_ops = {
443 .start = rt_cache_seq_start,
444 .next = rt_cache_seq_next,
445 .stop = rt_cache_seq_stop,
446 .show = rt_cache_seq_show,
449 static int rt_cache_seq_open(struct inode *inode, struct file *file)
451 return seq_open_net(inode, file, &rt_cache_seq_ops,
452 sizeof(struct rt_cache_iter_state));
455 static const struct file_operations rt_cache_seq_fops = {
456 .owner = THIS_MODULE,
457 .open = rt_cache_seq_open,
460 .release = seq_release_net,
464 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
469 return SEQ_START_TOKEN;
471 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
472 if (!cpu_possible(cpu))
475 return &per_cpu(rt_cache_stat, cpu);
480 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
484 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
485 if (!cpu_possible(cpu))
488 return &per_cpu(rt_cache_stat, cpu);
494 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
499 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
501 struct rt_cache_stat *st = v;
503 if (v == SEQ_START_TOKEN) {
504 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");
508 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
509 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
510 dst_entries_get_slow(&ipv4_dst_ops),
533 static const struct seq_operations rt_cpu_seq_ops = {
534 .start = rt_cpu_seq_start,
535 .next = rt_cpu_seq_next,
536 .stop = rt_cpu_seq_stop,
537 .show = rt_cpu_seq_show,
541 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
543 return seq_open(file, &rt_cpu_seq_ops);
546 static const struct file_operations rt_cpu_seq_fops = {
547 .owner = THIS_MODULE,
548 .open = rt_cpu_seq_open,
551 .release = seq_release,
554 #ifdef CONFIG_IP_ROUTE_CLASSID
555 static int rt_acct_proc_show(struct seq_file *m, void *v)
557 struct ip_rt_acct *dst, *src;
560 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
564 for_each_possible_cpu(i) {
565 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
566 for (j = 0; j < 256; j++) {
567 dst[j].o_bytes += src[j].o_bytes;
568 dst[j].o_packets += src[j].o_packets;
569 dst[j].i_bytes += src[j].i_bytes;
570 dst[j].i_packets += src[j].i_packets;
574 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
579 static int rt_acct_proc_open(struct inode *inode, struct file *file)
581 return single_open(file, rt_acct_proc_show, NULL);
584 static const struct file_operations rt_acct_proc_fops = {
585 .owner = THIS_MODULE,
586 .open = rt_acct_proc_open,
589 .release = single_release,
593 static int __net_init ip_rt_do_proc_init(struct net *net)
595 struct proc_dir_entry *pde;
597 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
602 pde = proc_create("rt_cache", S_IRUGO,
603 net->proc_net_stat, &rt_cpu_seq_fops);
607 #ifdef CONFIG_IP_ROUTE_CLASSID
608 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
614 #ifdef CONFIG_IP_ROUTE_CLASSID
616 remove_proc_entry("rt_cache", net->proc_net_stat);
619 remove_proc_entry("rt_cache", net->proc_net);
624 static void __net_exit ip_rt_do_proc_exit(struct net *net)
626 remove_proc_entry("rt_cache", net->proc_net_stat);
627 remove_proc_entry("rt_cache", net->proc_net);
628 #ifdef CONFIG_IP_ROUTE_CLASSID
629 remove_proc_entry("rt_acct", net->proc_net);
633 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
634 .init = ip_rt_do_proc_init,
635 .exit = ip_rt_do_proc_exit,
638 static int __init ip_rt_proc_init(void)
640 return register_pernet_subsys(&ip_rt_proc_ops);
644 static inline int ip_rt_proc_init(void)
648 #endif /* CONFIG_PROC_FS */
650 static inline void rt_free(struct rtable *rt)
652 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
655 static inline void rt_drop(struct rtable *rt)
658 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
661 static inline int rt_fast_clean(struct rtable *rth)
663 /* Kill broadcast/multicast entries very aggresively, if they
664 collide in hash table with more useful entries */
665 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
666 rt_is_input_route(rth) && rth->dst.rt_next;
669 static inline int rt_valuable(struct rtable *rth)
671 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
672 (rt_has_peer(rth) && rt_peer_ptr(rth)->pmtu_expires);
675 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
680 if (atomic_read(&rth->dst.__refcnt))
683 age = jiffies - rth->dst.lastuse;
684 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
685 (age <= tmo2 && rt_valuable(rth)))
691 /* Bits of score are:
693 * 30: not quite useless
694 * 29..0: usage counter
696 static inline u32 rt_score(struct rtable *rt)
698 u32 score = jiffies - rt->dst.lastuse;
700 score = ~score & ~(3<<30);
705 if (rt_is_output_route(rt) ||
706 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
712 static inline bool rt_caching(const struct net *net)
714 return net->ipv4.current_rt_cache_rebuild_count <=
715 net->ipv4.sysctl_rt_cache_rebuild_count;
718 static inline bool compare_hash_inputs(const struct rtable *rt1,
719 const struct rtable *rt2)
721 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
722 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
723 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
726 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
728 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
729 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
730 (rt1->rt_mark ^ rt2->rt_mark) |
731 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
732 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
733 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
736 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
738 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
741 static inline int rt_is_expired(struct rtable *rth)
743 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
747 * Perform a full scan of hash table and free all entries.
748 * Can be called by a softirq or a process.
749 * In the later case, we want to be reschedule if necessary
751 static void rt_do_flush(struct net *net, int process_context)
754 struct rtable *rth, *next;
756 for (i = 0; i <= rt_hash_mask; i++) {
757 struct rtable __rcu **pprev;
760 if (process_context && need_resched())
762 rth = rcu_access_pointer(rt_hash_table[i].chain);
766 spin_lock_bh(rt_hash_lock_addr(i));
769 pprev = &rt_hash_table[i].chain;
770 rth = rcu_dereference_protected(*pprev,
771 lockdep_is_held(rt_hash_lock_addr(i)));
774 next = rcu_dereference_protected(rth->dst.rt_next,
775 lockdep_is_held(rt_hash_lock_addr(i)));
778 net_eq(dev_net(rth->dst.dev), net)) {
779 rcu_assign_pointer(*pprev, next);
780 rcu_assign_pointer(rth->dst.rt_next, list);
783 pprev = &rth->dst.rt_next;
788 spin_unlock_bh(rt_hash_lock_addr(i));
790 for (; list; list = next) {
791 next = rcu_dereference_protected(list->dst.rt_next, 1);
798 * While freeing expired entries, we compute average chain length
799 * and standard deviation, using fixed-point arithmetic.
800 * This to have an estimation of rt_chain_length_max
801 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
802 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
806 #define ONE (1UL << FRACT_BITS)
809 * Given a hash chain and an item in this hash chain,
810 * find if a previous entry has the same hash_inputs
811 * (but differs on tos, mark or oif)
812 * Returns 0 if an alias is found.
813 * Returns ONE if rth has no alias before itself.
815 static int has_noalias(const struct rtable *head, const struct rtable *rth)
817 const struct rtable *aux = head;
820 if (compare_hash_inputs(aux, rth))
822 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
827 static void rt_check_expire(void)
829 static unsigned int rover;
830 unsigned int i = rover, goal;
832 struct rtable __rcu **rthp;
833 unsigned long samples = 0;
834 unsigned long sum = 0, sum2 = 0;
838 delta = jiffies - expires_ljiffies;
839 expires_ljiffies = jiffies;
840 mult = ((u64)delta) << rt_hash_log;
841 if (ip_rt_gc_timeout > 1)
842 do_div(mult, ip_rt_gc_timeout);
843 goal = (unsigned int)mult;
844 if (goal > rt_hash_mask)
845 goal = rt_hash_mask + 1;
846 for (; goal > 0; goal--) {
847 unsigned long tmo = ip_rt_gc_timeout;
848 unsigned long length;
850 i = (i + 1) & rt_hash_mask;
851 rthp = &rt_hash_table[i].chain;
858 if (rcu_dereference_raw(*rthp) == NULL)
861 spin_lock_bh(rt_hash_lock_addr(i));
862 while ((rth = rcu_dereference_protected(*rthp,
863 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
864 prefetch(rth->dst.rt_next);
865 if (rt_is_expired(rth) ||
866 rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
867 *rthp = rth->dst.rt_next;
872 /* We only count entries on a chain with equal
873 * hash inputs once so that entries for
874 * different QOS levels, and other non-hash
875 * input attributes don't unfairly skew the
879 rthp = &rth->dst.rt_next;
880 length += has_noalias(rt_hash_table[i].chain, rth);
882 spin_unlock_bh(rt_hash_lock_addr(i));
884 sum2 += length*length;
887 unsigned long avg = sum / samples;
888 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
889 rt_chain_length_max = max_t(unsigned long,
891 (avg + 4*sd) >> FRACT_BITS);
897 * rt_worker_func() is run in process context.
898 * we call rt_check_expire() to scan part of the hash table
900 static void rt_worker_func(struct work_struct *work)
903 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
907 * Perturbation of rt_genid by a small quantity [1..256]
908 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
909 * many times (2^24) without giving recent rt_genid.
910 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
912 static void rt_cache_invalidate(struct net *net)
914 unsigned char shuffle;
916 get_random_bytes(&shuffle, sizeof(shuffle));
917 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
918 inetpeer_invalidate_family(AF_INET);
922 * delay < 0 : invalidate cache (fast : entries will be deleted later)
923 * delay >= 0 : invalidate & flush cache (can be long)
925 void rt_cache_flush(struct net *net, int delay)
927 rt_cache_invalidate(net);
929 rt_do_flush(net, !in_softirq());
932 /* Flush previous cache invalidated entries from the cache */
933 void rt_cache_flush_batch(struct net *net)
935 rt_do_flush(net, !in_softirq());
938 static void rt_emergency_hash_rebuild(struct net *net)
940 net_warn_ratelimited("Route hash chain too long!\n");
941 rt_cache_invalidate(net);
945 Short description of GC goals.
947 We want to build algorithm, which will keep routing cache
948 at some equilibrium point, when number of aged off entries
949 is kept approximately equal to newly generated ones.
951 Current expiration strength is variable "expire".
952 We try to adjust it dynamically, so that if networking
953 is idle expires is large enough to keep enough of warm entries,
954 and when load increases it reduces to limit cache size.
957 static int rt_garbage_collect(struct dst_ops *ops)
959 static unsigned long expire = RT_GC_TIMEOUT;
960 static unsigned long last_gc;
962 static int equilibrium;
964 struct rtable __rcu **rthp;
965 unsigned long now = jiffies;
967 int entries = dst_entries_get_fast(&ipv4_dst_ops);
970 * Garbage collection is pretty expensive,
971 * do not make it too frequently.
974 RT_CACHE_STAT_INC(gc_total);
976 if (now - last_gc < ip_rt_gc_min_interval &&
977 entries < ip_rt_max_size) {
978 RT_CACHE_STAT_INC(gc_ignored);
982 entries = dst_entries_get_slow(&ipv4_dst_ops);
983 /* Calculate number of entries, which we want to expire now. */
984 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
986 if (equilibrium < ipv4_dst_ops.gc_thresh)
987 equilibrium = ipv4_dst_ops.gc_thresh;
988 goal = entries - equilibrium;
990 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
991 goal = entries - equilibrium;
994 /* We are in dangerous area. Try to reduce cache really
997 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
998 equilibrium = entries - goal;
1001 if (now - last_gc >= ip_rt_gc_min_interval)
1005 equilibrium += goal;
1012 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1013 unsigned long tmo = expire;
1015 k = (k + 1) & rt_hash_mask;
1016 rthp = &rt_hash_table[k].chain;
1017 spin_lock_bh(rt_hash_lock_addr(k));
1018 while ((rth = rcu_dereference_protected(*rthp,
1019 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1020 if (!rt_is_expired(rth) &&
1021 !rt_may_expire(rth, tmo, expire)) {
1023 rthp = &rth->dst.rt_next;
1026 *rthp = rth->dst.rt_next;
1030 spin_unlock_bh(rt_hash_lock_addr(k));
1039 /* Goal is not achieved. We stop process if:
1041 - if expire reduced to zero. Otherwise, expire is halfed.
1042 - if table is not full.
1043 - if we are called from interrupt.
1044 - jiffies check is just fallback/debug loop breaker.
1045 We will not spin here for long time in any case.
1048 RT_CACHE_STAT_INC(gc_goal_miss);
1055 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1057 } while (!in_softirq() && time_before_eq(jiffies, now));
1059 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1061 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1063 net_warn_ratelimited("dst cache overflow\n");
1064 RT_CACHE_STAT_INC(gc_dst_overflow);
1068 expire += ip_rt_gc_min_interval;
1069 if (expire > ip_rt_gc_timeout ||
1070 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1071 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1072 expire = ip_rt_gc_timeout;
1077 * Returns number of entries in a hash chain that have different hash_inputs
1079 static int slow_chain_length(const struct rtable *head)
1082 const struct rtable *rth = head;
1085 length += has_noalias(head, rth);
1086 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1088 return length >> FRACT_BITS;
1091 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
1092 struct sk_buff *skb,
1095 struct net_device *dev = dst->dev;
1096 const __be32 *pkey = daddr;
1097 const struct rtable *rt;
1098 struct neighbour *n;
1100 rt = (const struct rtable *) dst;
1102 pkey = (const __be32 *) &rt->rt_gateway;
1104 pkey = &ip_hdr(skb)->daddr;
1106 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1109 return neigh_create(&arp_tbl, pkey, dev);
1112 static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1113 struct sk_buff *skb, int ifindex)
1115 struct rtable *rth, *cand;
1116 struct rtable __rcu **rthp, **candp;
1123 min_score = ~(u32)0;
1128 if (!rt_caching(dev_net(rt->dst.dev)) || (rt->dst.flags & DST_NOCACHE)) {
1130 * If we're not caching, just tell the caller we
1131 * were successful and don't touch the route. The
1132 * caller hold the sole reference to the cache entry, and
1133 * it will be released when the caller is done with it.
1134 * If we drop it here, the callers have no way to resolve routes
1135 * when we're not caching. Instead, just point *rp at rt, so
1136 * the caller gets a single use out of the route
1137 * Note that we do rt_free on this new route entry, so that
1138 * once its refcount hits zero, we are still able to reap it
1140 * Note: To avoid expensive rcu stuff for this uncached dst,
1141 * we set DST_NOCACHE so that dst_release() can free dst without
1142 * waiting a grace period.
1145 rt->dst.flags |= DST_NOCACHE;
1149 rthp = &rt_hash_table[hash].chain;
1151 spin_lock_bh(rt_hash_lock_addr(hash));
1152 while ((rth = rcu_dereference_protected(*rthp,
1153 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1154 if (rt_is_expired(rth)) {
1155 *rthp = rth->dst.rt_next;
1159 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1161 *rthp = rth->dst.rt_next;
1163 * Since lookup is lockfree, the deletion
1164 * must be visible to another weakly ordered CPU before
1165 * the insertion at the start of the hash chain.
1167 rcu_assign_pointer(rth->dst.rt_next,
1168 rt_hash_table[hash].chain);
1170 * Since lookup is lockfree, the update writes
1171 * must be ordered for consistency on SMP.
1173 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1175 dst_use(&rth->dst, now);
1176 spin_unlock_bh(rt_hash_lock_addr(hash));
1180 skb_dst_set(skb, &rth->dst);
1184 if (!atomic_read(&rth->dst.__refcnt)) {
1185 u32 score = rt_score(rth);
1187 if (score <= min_score) {
1196 rthp = &rth->dst.rt_next;
1200 /* ip_rt_gc_elasticity used to be average length of chain
1201 * length, when exceeded gc becomes really aggressive.
1203 * The second limit is less certain. At the moment it allows
1204 * only 2 entries per bucket. We will see.
1206 if (chain_length > ip_rt_gc_elasticity) {
1207 *candp = cand->dst.rt_next;
1211 if (chain_length > rt_chain_length_max &&
1212 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1213 struct net *net = dev_net(rt->dst.dev);
1214 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1215 if (!rt_caching(net)) {
1216 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1217 rt->dst.dev->name, num);
1219 rt_emergency_hash_rebuild(net);
1220 spin_unlock_bh(rt_hash_lock_addr(hash));
1222 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1223 ifindex, rt_genid(net));
1228 rt->dst.rt_next = rt_hash_table[hash].chain;
1231 * Since lookup is lockfree, we must make sure
1232 * previous writes to rt are committed to memory
1233 * before making rt visible to other CPUS.
1235 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1237 spin_unlock_bh(rt_hash_lock_addr(hash));
1241 skb_dst_set(skb, &rt->dst);
1245 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1247 static u32 rt_peer_genid(void)
1249 return atomic_read(&__rt_peer_genid);
1252 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1254 struct inet_peer_base *base;
1255 struct inet_peer *peer;
1257 base = inetpeer_base_ptr(rt->_peer);
1261 peer = inet_getpeer_v4(base, daddr, create);
1263 if (!rt_set_peer(rt, peer))
1266 rt->rt_peer_genid = rt_peer_genid();
1271 * Peer allocation may fail only in serious out-of-memory conditions. However
1272 * we still can generate some output.
1273 * Random ID selection looks a bit dangerous because we have no chances to
1274 * select ID being unique in a reasonable period of time.
1275 * But broken packet identifier may be better than no packet at all.
1277 static void ip_select_fb_ident(struct iphdr *iph)
1279 static DEFINE_SPINLOCK(ip_fb_id_lock);
1280 static u32 ip_fallback_id;
1283 spin_lock_bh(&ip_fb_id_lock);
1284 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1285 iph->id = htons(salt & 0xFFFF);
1286 ip_fallback_id = salt;
1287 spin_unlock_bh(&ip_fb_id_lock);
1290 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1292 struct rtable *rt = (struct rtable *) dst;
1294 if (rt && !(rt->dst.flags & DST_NOPEER)) {
1295 struct inet_peer *peer = rt_get_peer_create(rt, rt->rt_dst);
1297 /* If peer is attached to destination, it is never detached,
1298 so that we need not to grab a lock to dereference it.
1301 iph->id = htons(inet_getid(peer, more));
1305 pr_debug("rt_bind_peer(0) @%p\n", __builtin_return_address(0));
1307 ip_select_fb_ident(iph);
1309 EXPORT_SYMBOL(__ip_select_ident);
1311 static void rt_del(unsigned int hash, struct rtable *rt)
1313 struct rtable __rcu **rthp;
1316 rthp = &rt_hash_table[hash].chain;
1317 spin_lock_bh(rt_hash_lock_addr(hash));
1319 while ((aux = rcu_dereference_protected(*rthp,
1320 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1321 if (aux == rt || rt_is_expired(aux)) {
1322 *rthp = aux->dst.rt_next;
1326 rthp = &aux->dst.rt_next;
1328 spin_unlock_bh(rt_hash_lock_addr(hash));
1331 static void check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1333 struct rtable *rt = (struct rtable *) dst;
1334 __be32 orig_gw = rt->rt_gateway;
1335 struct neighbour *n;
1337 dst_confirm(&rt->dst);
1339 rt->rt_gateway = peer->redirect_learned.a4;
1341 n = ipv4_neigh_lookup(&rt->dst, NULL, &rt->rt_gateway);
1343 rt->rt_gateway = orig_gw;
1346 if (!(n->nud_state & NUD_VALID)) {
1347 neigh_event_send(n, NULL);
1349 rt->rt_flags |= RTCF_REDIRECTED;
1350 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1355 /* called in rcu_read_lock() section */
1356 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1357 __be32 saddr, struct net_device *dev)
1360 struct in_device *in_dev = __in_dev_get_rcu(dev);
1361 __be32 skeys[2] = { saddr, 0 };
1362 int ikeys[2] = { dev->ifindex, 0 };
1363 struct inet_peer *peer;
1370 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1371 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1372 ipv4_is_zeronet(new_gw))
1373 goto reject_redirect;
1375 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1376 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1377 goto reject_redirect;
1378 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1379 goto reject_redirect;
1381 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1382 goto reject_redirect;
1385 for (s = 0; s < 2; s++) {
1386 for (i = 0; i < 2; i++) {
1388 struct rtable __rcu **rthp;
1391 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1393 rthp = &rt_hash_table[hash].chain;
1395 while ((rt = rcu_dereference(*rthp)) != NULL) {
1396 rthp = &rt->dst.rt_next;
1398 if (rt->rt_key_dst != daddr ||
1399 rt->rt_key_src != skeys[s] ||
1400 rt->rt_oif != ikeys[i] ||
1401 rt_is_input_route(rt) ||
1402 rt_is_expired(rt) ||
1403 !net_eq(dev_net(rt->dst.dev), net) ||
1405 rt->dst.dev != dev ||
1406 rt->rt_gateway != old_gw)
1409 peer = rt_get_peer_create(rt, rt->rt_dst);
1411 if (peer->redirect_learned.a4 != new_gw) {
1412 peer->redirect_learned.a4 = new_gw;
1413 atomic_inc(&__rt_peer_genid);
1415 check_peer_redir(&rt->dst, peer);
1423 #ifdef CONFIG_IP_ROUTE_VERBOSE
1424 if (IN_DEV_LOG_MARTIANS(in_dev))
1425 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1426 " Advised path = %pI4 -> %pI4\n",
1427 &old_gw, dev->name, &new_gw,
1433 static bool peer_pmtu_expired(struct inet_peer *peer)
1435 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1438 time_after_eq(jiffies, orig) &&
1439 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1442 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1444 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1447 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1450 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1452 struct rtable *rt = (struct rtable *)dst;
1453 struct dst_entry *ret = dst;
1456 if (dst->obsolete > 0) {
1459 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1460 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1462 rt_genid(dev_net(dst->dev)));
1465 } else if (rt_has_peer(rt)) {
1466 struct inet_peer *peer = rt_peer_ptr(rt);
1467 if (peer_pmtu_expired(peer))
1468 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1476 * 1. The first ip_rt_redirect_number redirects are sent
1477 * with exponential backoff, then we stop sending them at all,
1478 * assuming that the host ignores our redirects.
1479 * 2. If we did not see packets requiring redirects
1480 * during ip_rt_redirect_silence, we assume that the host
1481 * forgot redirected route and start to send redirects again.
1483 * This algorithm is much cheaper and more intelligent than dumb load limiting
1486 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1487 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1490 void ip_rt_send_redirect(struct sk_buff *skb)
1492 struct rtable *rt = skb_rtable(skb);
1493 struct in_device *in_dev;
1494 struct inet_peer *peer;
1498 in_dev = __in_dev_get_rcu(rt->dst.dev);
1499 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1503 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1506 peer = rt_get_peer_create(rt, rt->rt_dst);
1508 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1512 /* No redirected packets during ip_rt_redirect_silence;
1513 * reset the algorithm.
1515 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1516 peer->rate_tokens = 0;
1518 /* Too many ignored redirects; do not send anything
1519 * set dst.rate_last to the last seen redirected packet.
1521 if (peer->rate_tokens >= ip_rt_redirect_number) {
1522 peer->rate_last = jiffies;
1526 /* Check for load limit; set rate_last to the latest sent
1529 if (peer->rate_tokens == 0 ||
1532 (ip_rt_redirect_load << peer->rate_tokens)))) {
1533 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1534 peer->rate_last = jiffies;
1535 ++peer->rate_tokens;
1536 #ifdef CONFIG_IP_ROUTE_VERBOSE
1538 peer->rate_tokens == ip_rt_redirect_number)
1539 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1540 &ip_hdr(skb)->saddr, rt->rt_iif,
1541 &rt->rt_dst, &rt->rt_gateway);
1546 static int ip_error(struct sk_buff *skb)
1548 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1549 struct rtable *rt = skb_rtable(skb);
1550 struct inet_peer *peer;
1556 net = dev_net(rt->dst.dev);
1557 if (!IN_DEV_FORWARD(in_dev)) {
1558 switch (rt->dst.error) {
1560 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
1564 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1570 switch (rt->dst.error) {
1575 code = ICMP_HOST_UNREACH;
1578 code = ICMP_NET_UNREACH;
1579 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
1582 code = ICMP_PKT_FILTERED;
1586 peer = rt_get_peer_create(rt, rt->rt_dst);
1591 peer->rate_tokens += now - peer->rate_last;
1592 if (peer->rate_tokens > ip_rt_error_burst)
1593 peer->rate_tokens = ip_rt_error_burst;
1594 peer->rate_last = now;
1595 if (peer->rate_tokens >= ip_rt_error_cost)
1596 peer->rate_tokens -= ip_rt_error_cost;
1601 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1603 out: kfree_skb(skb);
1607 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1609 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1613 if (time_before(jiffies, expires)) {
1614 u32 orig_dst_mtu = dst_mtu(dst);
1615 if (peer->pmtu_learned < orig_dst_mtu) {
1616 if (!peer->pmtu_orig)
1617 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1618 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1620 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1621 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1624 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1626 struct rtable *rt = (struct rtable *) dst;
1627 struct inet_peer *peer;
1631 peer = rt_get_peer_create(rt, rt->rt_dst);
1633 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1635 if (mtu < ip_rt_min_pmtu)
1636 mtu = ip_rt_min_pmtu;
1637 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1639 pmtu_expires = jiffies + ip_rt_mtu_expires;
1643 peer->pmtu_learned = mtu;
1644 peer->pmtu_expires = pmtu_expires;
1646 atomic_inc(&__rt_peer_genid);
1647 rt->rt_peer_genid = rt_peer_genid();
1649 check_peer_pmtu(dst, peer);
1653 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1654 int oif, u32 mark, u8 protocol, int flow_flags)
1656 const struct iphdr *iph = (const struct iphdr *)skb->data;
1660 flowi4_init_output(&fl4, oif, mark, RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
1661 protocol, flow_flags | FLOWI_FLAG_PRECOW_METRICS,
1662 iph->daddr, iph->saddr, 0, 0);
1663 rt = __ip_route_output_key(net, &fl4);
1665 ip_rt_update_pmtu(&rt->dst, mtu);
1669 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1671 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1673 const struct inet_sock *inet = inet_sk(sk);
1675 return ipv4_update_pmtu(skb, sock_net(sk), mtu,
1676 sk->sk_bound_dev_if, sk->sk_mark,
1677 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
1678 inet_sk_flowi_flags(sk));
1680 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1682 static void ipv4_validate_peer(struct rtable *rt)
1684 if (rt->rt_peer_genid != rt_peer_genid()) {
1685 struct inet_peer *peer = rt_get_peer(rt, rt->rt_dst);
1688 check_peer_pmtu(&rt->dst, peer);
1690 if (peer->redirect_learned.a4 &&
1691 peer->redirect_learned.a4 != rt->rt_gateway)
1692 check_peer_redir(&rt->dst, peer);
1695 rt->rt_peer_genid = rt_peer_genid();
1699 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1701 struct rtable *rt = (struct rtable *) dst;
1703 if (rt_is_expired(rt))
1705 ipv4_validate_peer(rt);
1709 static void ipv4_dst_destroy(struct dst_entry *dst)
1711 struct rtable *rt = (struct rtable *) dst;
1714 fib_info_put(rt->fi);
1717 if (rt_has_peer(rt)) {
1718 struct inet_peer *peer = rt_peer_ptr(rt);
1724 static void ipv4_link_failure(struct sk_buff *skb)
1728 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1730 rt = skb_rtable(skb);
1731 if (rt && rt_has_peer(rt)) {
1732 struct inet_peer *peer = rt_peer_ptr(rt);
1733 if (peer_pmtu_cleaned(peer))
1734 dst_metric_set(&rt->dst, RTAX_MTU, peer->pmtu_orig);
1738 static int ip_rt_bug(struct sk_buff *skb)
1740 pr_debug("%s: %pI4 -> %pI4, %s\n",
1741 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1742 skb->dev ? skb->dev->name : "?");
1749 We do not cache source address of outgoing interface,
1750 because it is used only by IP RR, TS and SRR options,
1751 so that it out of fast path.
1753 BTW remember: "addr" is allowed to be not aligned
1757 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1761 if (rt_is_output_route(rt))
1762 src = ip_hdr(skb)->saddr;
1764 struct fib_result res;
1770 memset(&fl4, 0, sizeof(fl4));
1771 fl4.daddr = iph->daddr;
1772 fl4.saddr = iph->saddr;
1773 fl4.flowi4_tos = RT_TOS(iph->tos);
1774 fl4.flowi4_oif = rt->dst.dev->ifindex;
1775 fl4.flowi4_iif = skb->dev->ifindex;
1776 fl4.flowi4_mark = skb->mark;
1779 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1780 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1782 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1786 memcpy(addr, &src, 4);
1789 #ifdef CONFIG_IP_ROUTE_CLASSID
1790 static void set_class_tag(struct rtable *rt, u32 tag)
1792 if (!(rt->dst.tclassid & 0xFFFF))
1793 rt->dst.tclassid |= tag & 0xFFFF;
1794 if (!(rt->dst.tclassid & 0xFFFF0000))
1795 rt->dst.tclassid |= tag & 0xFFFF0000;
1799 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1801 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1804 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1806 if (advmss > 65535 - 40)
1807 advmss = 65535 - 40;
1812 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1814 const struct rtable *rt = (const struct rtable *) dst;
1815 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1817 if (mtu && rt_is_output_route(rt))
1820 mtu = dst->dev->mtu;
1822 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1824 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1828 if (mtu > IP_MAX_MTU)
1834 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1835 struct fib_info *fi)
1837 struct inet_peer_base *base;
1838 struct inet_peer *peer;
1841 /* If a peer entry exists for this destination, we must hook
1842 * it up in order to get at cached metrics.
1844 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1847 base = inetpeer_base_ptr(rt->_peer);
1850 peer = inet_getpeer_v4(base, rt->rt_dst, create);
1852 __rt_set_peer(rt, peer);
1853 rt->rt_peer_genid = rt_peer_genid();
1854 if (inet_metrics_new(peer))
1855 memcpy(peer->metrics, fi->fib_metrics,
1856 sizeof(u32) * RTAX_MAX);
1857 dst_init_metrics(&rt->dst, peer->metrics, false);
1859 check_peer_pmtu(&rt->dst, peer);
1861 if (peer->redirect_learned.a4 &&
1862 peer->redirect_learned.a4 != rt->rt_gateway) {
1863 rt->rt_gateway = peer->redirect_learned.a4;
1864 rt->rt_flags |= RTCF_REDIRECTED;
1867 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1869 atomic_inc(&fi->fib_clntref);
1871 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1875 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1876 const struct fib_result *res,
1877 struct fib_info *fi, u16 type, u32 itag)
1879 struct dst_entry *dst = &rt->dst;
1882 if (FIB_RES_GW(*res) &&
1883 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1884 rt->rt_gateway = FIB_RES_GW(*res);
1885 rt_init_metrics(rt, fl4, fi);
1886 #ifdef CONFIG_IP_ROUTE_CLASSID
1887 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1891 if (dst_mtu(dst) > IP_MAX_MTU)
1892 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1894 #ifdef CONFIG_IP_ROUTE_CLASSID
1895 #ifdef CONFIG_IP_MULTIPLE_TABLES
1896 set_class_tag(rt, fib_rules_tclass(res));
1898 set_class_tag(rt, itag);
1902 static struct rtable *rt_dst_alloc(struct net_device *dev,
1903 bool nopolicy, bool noxfrm)
1905 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1907 (nopolicy ? DST_NOPOLICY : 0) |
1908 (noxfrm ? DST_NOXFRM : 0));
1911 /* called in rcu_read_lock() section */
1912 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1913 u8 tos, struct net_device *dev, int our)
1917 struct in_device *in_dev = __in_dev_get_rcu(dev);
1921 /* Primary sanity checks. */
1926 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1927 skb->protocol != htons(ETH_P_IP))
1930 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1931 if (ipv4_is_loopback(saddr))
1934 if (ipv4_is_zeronet(saddr)) {
1935 if (!ipv4_is_local_multicast(daddr))
1938 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1943 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1944 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1948 #ifdef CONFIG_IP_ROUTE_CLASSID
1949 rth->dst.tclassid = itag;
1951 rth->dst.output = ip_rt_bug;
1953 rth->rt_key_dst = daddr;
1954 rth->rt_key_src = saddr;
1955 rth->rt_genid = rt_genid(dev_net(dev));
1956 rth->rt_flags = RTCF_MULTICAST;
1957 rth->rt_type = RTN_MULTICAST;
1958 rth->rt_key_tos = tos;
1959 rth->rt_dst = daddr;
1960 rth->rt_src = saddr;
1961 rth->rt_route_iif = dev->ifindex;
1962 rth->rt_iif = dev->ifindex;
1964 rth->rt_mark = skb->mark;
1965 rth->rt_gateway = daddr;
1966 rth->rt_peer_genid = 0;
1967 rt_init_peer(rth, dev_net(dev)->ipv4.peers);
1970 rth->dst.input= ip_local_deliver;
1971 rth->rt_flags |= RTCF_LOCAL;
1974 #ifdef CONFIG_IP_MROUTE
1975 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1976 rth->dst.input = ip_mr_input;
1978 RT_CACHE_STAT_INC(in_slow_mc);
1980 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1981 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1982 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1993 static void ip_handle_martian_source(struct net_device *dev,
1994 struct in_device *in_dev,
1995 struct sk_buff *skb,
1999 RT_CACHE_STAT_INC(in_martian_src);
2000 #ifdef CONFIG_IP_ROUTE_VERBOSE
2001 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2003 * RFC1812 recommendation, if source is martian,
2004 * the only hint is MAC header.
2006 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
2007 &daddr, &saddr, dev->name);
2008 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2009 print_hex_dump(KERN_WARNING, "ll header: ",
2010 DUMP_PREFIX_OFFSET, 16, 1,
2011 skb_mac_header(skb),
2012 dev->hard_header_len, true);
2018 /* called in rcu_read_lock() section */
2019 static int __mkroute_input(struct sk_buff *skb,
2020 const struct fib_result *res,
2021 struct in_device *in_dev,
2022 __be32 daddr, __be32 saddr, u32 tos,
2023 struct rtable **result)
2027 struct in_device *out_dev;
2028 unsigned int flags = 0;
2031 /* get a working reference to the output device */
2032 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2033 if (out_dev == NULL) {
2034 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
2039 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2040 in_dev->dev, in_dev, &itag);
2042 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2049 flags |= RTCF_DIRECTSRC;
2051 if (out_dev == in_dev && err &&
2052 (IN_DEV_SHARED_MEDIA(out_dev) ||
2053 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2054 flags |= RTCF_DOREDIRECT;
2056 if (skb->protocol != htons(ETH_P_IP)) {
2057 /* Not IP (i.e. ARP). Do not create route, if it is
2058 * invalid for proxy arp. DNAT routes are always valid.
2060 * Proxy arp feature have been extended to allow, ARP
2061 * replies back to the same interface, to support
2062 * Private VLAN switch technologies. See arp.c.
2064 if (out_dev == in_dev &&
2065 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2071 rth = rt_dst_alloc(out_dev->dev,
2072 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2073 IN_DEV_CONF_GET(out_dev, NOXFRM));
2079 rth->rt_key_dst = daddr;
2080 rth->rt_key_src = saddr;
2081 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2082 rth->rt_flags = flags;
2083 rth->rt_type = res->type;
2084 rth->rt_key_tos = tos;
2085 rth->rt_dst = daddr;
2086 rth->rt_src = saddr;
2087 rth->rt_route_iif = in_dev->dev->ifindex;
2088 rth->rt_iif = in_dev->dev->ifindex;
2090 rth->rt_mark = skb->mark;
2091 rth->rt_gateway = daddr;
2092 rth->rt_peer_genid = 0;
2093 rt_init_peer(rth, &res->table->tb_peers);
2096 rth->dst.input = ip_forward;
2097 rth->dst.output = ip_output;
2099 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2107 static int ip_mkroute_input(struct sk_buff *skb,
2108 struct fib_result *res,
2109 const struct flowi4 *fl4,
2110 struct in_device *in_dev,
2111 __be32 daddr, __be32 saddr, u32 tos)
2113 struct rtable *rth = NULL;
2117 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2118 if (res->fi && res->fi->fib_nhs > 1)
2119 fib_select_multipath(res);
2122 /* create a routing cache entry */
2123 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2127 /* put it into the cache */
2128 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2129 rt_genid(dev_net(rth->dst.dev)));
2130 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2132 return PTR_ERR(rth);
2137 * NOTE. We drop all the packets that has local source
2138 * addresses, because every properly looped back packet
2139 * must have correct destination already attached by output routine.
2141 * Such approach solves two big problems:
2142 * 1. Not simplex devices are handled properly.
2143 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2144 * called with rcu_read_lock()
2147 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2148 u8 tos, struct net_device *dev)
2150 struct fib_result res;
2151 struct in_device *in_dev = __in_dev_get_rcu(dev);
2153 unsigned int flags = 0;
2158 struct net *net = dev_net(dev);
2160 /* IP on this device is disabled. */
2165 /* Check for the most weird martians, which can be not detected
2169 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2170 goto martian_source;
2172 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2175 /* Accept zero addresses only to limited broadcast;
2176 * I even do not know to fix it or not. Waiting for complains :-)
2178 if (ipv4_is_zeronet(saddr))
2179 goto martian_source;
2181 if (ipv4_is_zeronet(daddr))
2182 goto martian_destination;
2184 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
2185 if (ipv4_is_loopback(daddr))
2186 goto martian_destination;
2188 if (ipv4_is_loopback(saddr))
2189 goto martian_source;
2193 * Now we are ready to route packet.
2196 fl4.flowi4_iif = dev->ifindex;
2197 fl4.flowi4_mark = skb->mark;
2198 fl4.flowi4_tos = tos;
2199 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2202 err = fib_lookup(net, &fl4, &res);
2206 RT_CACHE_STAT_INC(in_slow_tot);
2208 if (res.type == RTN_BROADCAST)
2211 if (res.type == RTN_LOCAL) {
2212 err = fib_validate_source(skb, saddr, daddr, tos,
2213 net->loopback_dev->ifindex,
2214 dev, in_dev, &itag);
2216 goto martian_source_keep_err;
2218 flags |= RTCF_DIRECTSRC;
2222 if (!IN_DEV_FORWARD(in_dev))
2224 if (res.type != RTN_UNICAST)
2225 goto martian_destination;
2227 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2231 if (skb->protocol != htons(ETH_P_IP))
2234 if (!ipv4_is_zeronet(saddr)) {
2235 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2238 goto martian_source_keep_err;
2240 flags |= RTCF_DIRECTSRC;
2242 flags |= RTCF_BROADCAST;
2243 res.type = RTN_BROADCAST;
2244 RT_CACHE_STAT_INC(in_brd);
2247 rth = rt_dst_alloc(net->loopback_dev,
2248 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2252 rth->dst.input= ip_local_deliver;
2253 rth->dst.output= ip_rt_bug;
2254 #ifdef CONFIG_IP_ROUTE_CLASSID
2255 rth->dst.tclassid = itag;
2258 rth->rt_key_dst = daddr;
2259 rth->rt_key_src = saddr;
2260 rth->rt_genid = rt_genid(net);
2261 rth->rt_flags = flags|RTCF_LOCAL;
2262 rth->rt_type = res.type;
2263 rth->rt_key_tos = tos;
2264 rth->rt_dst = daddr;
2265 rth->rt_src = saddr;
2266 rth->rt_route_iif = dev->ifindex;
2267 rth->rt_iif = dev->ifindex;
2269 rth->rt_mark = skb->mark;
2270 rth->rt_gateway = daddr;
2271 rth->rt_peer_genid = 0;
2272 rt_init_peer(rth, net->ipv4.peers);
2274 if (res.type == RTN_UNREACHABLE) {
2275 rth->dst.input= ip_error;
2276 rth->dst.error= -err;
2277 rth->rt_flags &= ~RTCF_LOCAL;
2279 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2280 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2287 RT_CACHE_STAT_INC(in_no_route);
2288 res.type = RTN_UNREACHABLE;
2294 * Do not cache martian addresses: they should be logged (RFC1812)
2296 martian_destination:
2297 RT_CACHE_STAT_INC(in_martian_dst);
2298 #ifdef CONFIG_IP_ROUTE_VERBOSE
2299 if (IN_DEV_LOG_MARTIANS(in_dev))
2300 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2301 &daddr, &saddr, dev->name);
2314 martian_source_keep_err:
2315 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2319 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2320 u8 tos, struct net_device *dev, bool noref)
2324 int iif = dev->ifindex;
2332 if (!rt_caching(net))
2335 tos &= IPTOS_RT_MASK;
2336 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2338 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2339 rth = rcu_dereference(rth->dst.rt_next)) {
2340 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2341 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2342 (rth->rt_route_iif ^ iif) |
2343 (rth->rt_key_tos ^ tos)) == 0 &&
2344 rth->rt_mark == skb->mark &&
2345 net_eq(dev_net(rth->dst.dev), net) &&
2346 !rt_is_expired(rth)) {
2347 ipv4_validate_peer(rth);
2349 dst_use_noref(&rth->dst, jiffies);
2350 skb_dst_set_noref(skb, &rth->dst);
2352 dst_use(&rth->dst, jiffies);
2353 skb_dst_set(skb, &rth->dst);
2355 RT_CACHE_STAT_INC(in_hit);
2359 RT_CACHE_STAT_INC(in_hlist_search);
2363 /* Multicast recognition logic is moved from route cache to here.
2364 The problem was that too many Ethernet cards have broken/missing
2365 hardware multicast filters :-( As result the host on multicasting
2366 network acquires a lot of useless route cache entries, sort of
2367 SDR messages from all the world. Now we try to get rid of them.
2368 Really, provided software IP multicast filter is organized
2369 reasonably (at least, hashed), it does not result in a slowdown
2370 comparing with route cache reject entries.
2371 Note, that multicast routers are not affected, because
2372 route cache entry is created eventually.
2374 if (ipv4_is_multicast(daddr)) {
2375 struct in_device *in_dev = __in_dev_get_rcu(dev);
2378 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2379 ip_hdr(skb)->protocol);
2381 #ifdef CONFIG_IP_MROUTE
2383 (!ipv4_is_local_multicast(daddr) &&
2384 IN_DEV_MFORWARD(in_dev))
2387 int res = ip_route_input_mc(skb, daddr, saddr,
2396 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2400 EXPORT_SYMBOL(ip_route_input_common);
2402 /* called with rcu_read_lock() */
2403 static struct rtable *__mkroute_output(const struct fib_result *res,
2404 const struct flowi4 *fl4,
2405 __be32 orig_daddr, __be32 orig_saddr,
2406 int orig_oif, __u8 orig_rtos,
2407 struct net_device *dev_out,
2410 struct fib_info *fi = res->fi;
2411 struct in_device *in_dev;
2412 u16 type = res->type;
2415 in_dev = __in_dev_get_rcu(dev_out);
2417 return ERR_PTR(-EINVAL);
2419 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2420 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2421 return ERR_PTR(-EINVAL);
2423 if (ipv4_is_lbcast(fl4->daddr))
2424 type = RTN_BROADCAST;
2425 else if (ipv4_is_multicast(fl4->daddr))
2426 type = RTN_MULTICAST;
2427 else if (ipv4_is_zeronet(fl4->daddr))
2428 return ERR_PTR(-EINVAL);
2430 if (dev_out->flags & IFF_LOOPBACK)
2431 flags |= RTCF_LOCAL;
2433 if (type == RTN_BROADCAST) {
2434 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2436 } else if (type == RTN_MULTICAST) {
2437 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2438 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2440 flags &= ~RTCF_LOCAL;
2441 /* If multicast route do not exist use
2442 * default one, but do not gateway in this case.
2445 if (fi && res->prefixlen < 4)
2449 rth = rt_dst_alloc(dev_out,
2450 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2451 IN_DEV_CONF_GET(in_dev, NOXFRM));
2453 return ERR_PTR(-ENOBUFS);
2455 rth->dst.output = ip_output;
2457 rth->rt_key_dst = orig_daddr;
2458 rth->rt_key_src = orig_saddr;
2459 rth->rt_genid = rt_genid(dev_net(dev_out));
2460 rth->rt_flags = flags;
2461 rth->rt_type = type;
2462 rth->rt_key_tos = orig_rtos;
2463 rth->rt_dst = fl4->daddr;
2464 rth->rt_src = fl4->saddr;
2465 rth->rt_route_iif = 0;
2466 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2467 rth->rt_oif = orig_oif;
2468 rth->rt_mark = fl4->flowi4_mark;
2469 rth->rt_gateway = fl4->daddr;
2470 rth->rt_peer_genid = 0;
2471 rt_init_peer(rth, (res->table ?
2472 &res->table->tb_peers :
2473 dev_net(dev_out)->ipv4.peers));
2476 RT_CACHE_STAT_INC(out_slow_tot);
2478 if (flags & RTCF_LOCAL)
2479 rth->dst.input = ip_local_deliver;
2480 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2481 if (flags & RTCF_LOCAL &&
2482 !(dev_out->flags & IFF_LOOPBACK)) {
2483 rth->dst.output = ip_mc_output;
2484 RT_CACHE_STAT_INC(out_slow_mc);
2486 #ifdef CONFIG_IP_MROUTE
2487 if (type == RTN_MULTICAST) {
2488 if (IN_DEV_MFORWARD(in_dev) &&
2489 !ipv4_is_local_multicast(fl4->daddr)) {
2490 rth->dst.input = ip_mr_input;
2491 rth->dst.output = ip_mc_output;
2497 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2499 if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
2500 rth->dst.flags |= DST_NOCACHE;
2506 * Major route resolver routine.
2507 * called with rcu_read_lock();
2510 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2512 struct net_device *dev_out = NULL;
2513 __u8 tos = RT_FL_TOS(fl4);
2514 unsigned int flags = 0;
2515 struct fib_result res;
2523 #ifdef CONFIG_IP_MULTIPLE_TABLES
2527 orig_daddr = fl4->daddr;
2528 orig_saddr = fl4->saddr;
2529 orig_oif = fl4->flowi4_oif;
2531 fl4->flowi4_iif = net->loopback_dev->ifindex;
2532 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2533 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2534 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2538 rth = ERR_PTR(-EINVAL);
2539 if (ipv4_is_multicast(fl4->saddr) ||
2540 ipv4_is_lbcast(fl4->saddr) ||
2541 ipv4_is_zeronet(fl4->saddr))
2544 /* I removed check for oif == dev_out->oif here.
2545 It was wrong for two reasons:
2546 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2547 is assigned to multiple interfaces.
2548 2. Moreover, we are allowed to send packets with saddr
2549 of another iface. --ANK
2552 if (fl4->flowi4_oif == 0 &&
2553 (ipv4_is_multicast(fl4->daddr) ||
2554 ipv4_is_lbcast(fl4->daddr))) {
2555 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2556 dev_out = __ip_dev_find(net, fl4->saddr, false);
2557 if (dev_out == NULL)
2560 /* Special hack: user can direct multicasts
2561 and limited broadcast via necessary interface
2562 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2563 This hack is not just for fun, it allows
2564 vic,vat and friends to work.
2565 They bind socket to loopback, set ttl to zero
2566 and expect that it will work.
2567 From the viewpoint of routing cache they are broken,
2568 because we are not allowed to build multicast path
2569 with loopback source addr (look, routing cache
2570 cannot know, that ttl is zero, so that packet
2571 will not leave this host and route is valid).
2572 Luckily, this hack is good workaround.
2575 fl4->flowi4_oif = dev_out->ifindex;
2579 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2580 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2581 if (!__ip_dev_find(net, fl4->saddr, false))
2587 if (fl4->flowi4_oif) {
2588 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2589 rth = ERR_PTR(-ENODEV);
2590 if (dev_out == NULL)
2593 /* RACE: Check return value of inet_select_addr instead. */
2594 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2595 rth = ERR_PTR(-ENETUNREACH);
2598 if (ipv4_is_local_multicast(fl4->daddr) ||
2599 ipv4_is_lbcast(fl4->daddr)) {
2601 fl4->saddr = inet_select_addr(dev_out, 0,
2606 if (ipv4_is_multicast(fl4->daddr))
2607 fl4->saddr = inet_select_addr(dev_out, 0,
2609 else if (!fl4->daddr)
2610 fl4->saddr = inet_select_addr(dev_out, 0,
2616 fl4->daddr = fl4->saddr;
2618 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2619 dev_out = net->loopback_dev;
2620 fl4->flowi4_oif = net->loopback_dev->ifindex;
2621 res.type = RTN_LOCAL;
2622 flags |= RTCF_LOCAL;
2626 if (fib_lookup(net, fl4, &res)) {
2629 if (fl4->flowi4_oif) {
2630 /* Apparently, routing tables are wrong. Assume,
2631 that the destination is on link.
2634 Because we are allowed to send to iface
2635 even if it has NO routes and NO assigned
2636 addresses. When oif is specified, routing
2637 tables are looked up with only one purpose:
2638 to catch if destination is gatewayed, rather than
2639 direct. Moreover, if MSG_DONTROUTE is set,
2640 we send packet, ignoring both routing tables
2641 and ifaddr state. --ANK
2644 We could make it even if oif is unknown,
2645 likely IPv6, but we do not.
2648 if (fl4->saddr == 0)
2649 fl4->saddr = inet_select_addr(dev_out, 0,
2651 res.type = RTN_UNICAST;
2654 rth = ERR_PTR(-ENETUNREACH);
2658 if (res.type == RTN_LOCAL) {
2660 if (res.fi->fib_prefsrc)
2661 fl4->saddr = res.fi->fib_prefsrc;
2663 fl4->saddr = fl4->daddr;
2665 dev_out = net->loopback_dev;
2666 fl4->flowi4_oif = dev_out->ifindex;
2668 flags |= RTCF_LOCAL;
2672 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2673 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2674 fib_select_multipath(&res);
2677 if (!res.prefixlen &&
2678 res.table->tb_num_default > 1 &&
2679 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2680 fib_select_default(&res);
2683 fl4->saddr = FIB_RES_PREFSRC(net, res);
2685 dev_out = FIB_RES_DEV(res);
2686 fl4->flowi4_oif = dev_out->ifindex;
2690 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2691 tos, dev_out, flags);
2695 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2696 rt_genid(dev_net(dev_out)));
2697 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2705 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2710 if (!rt_caching(net))
2713 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2716 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2717 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2718 if (rth->rt_key_dst == flp4->daddr &&
2719 rth->rt_key_src == flp4->saddr &&
2720 rt_is_output_route(rth) &&
2721 rth->rt_oif == flp4->flowi4_oif &&
2722 rth->rt_mark == flp4->flowi4_mark &&
2723 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2724 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2725 net_eq(dev_net(rth->dst.dev), net) &&
2726 !rt_is_expired(rth)) {
2727 ipv4_validate_peer(rth);
2728 dst_use(&rth->dst, jiffies);
2729 RT_CACHE_STAT_INC(out_hit);
2730 rcu_read_unlock_bh();
2732 flp4->saddr = rth->rt_src;
2734 flp4->daddr = rth->rt_dst;
2737 RT_CACHE_STAT_INC(out_hlist_search);
2739 rcu_read_unlock_bh();
2742 return ip_route_output_slow(net, flp4);
2744 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2746 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2751 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2753 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2755 return mtu ? : dst->dev->mtu;
2758 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2762 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2768 static struct dst_ops ipv4_dst_blackhole_ops = {
2770 .protocol = cpu_to_be16(ETH_P_IP),
2771 .destroy = ipv4_dst_destroy,
2772 .check = ipv4_blackhole_dst_check,
2773 .mtu = ipv4_blackhole_mtu,
2774 .default_advmss = ipv4_default_advmss,
2775 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2776 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2777 .neigh_lookup = ipv4_neigh_lookup,
2780 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2782 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2783 struct rtable *ort = (struct rtable *) dst_orig;
2786 struct dst_entry *new = &rt->dst;
2789 new->input = dst_discard;
2790 new->output = dst_discard;
2791 dst_copy_metrics(new, &ort->dst);
2793 new->dev = ort->dst.dev;
2797 rt->rt_key_dst = ort->rt_key_dst;
2798 rt->rt_key_src = ort->rt_key_src;
2799 rt->rt_key_tos = ort->rt_key_tos;
2800 rt->rt_route_iif = ort->rt_route_iif;
2801 rt->rt_iif = ort->rt_iif;
2802 rt->rt_oif = ort->rt_oif;
2803 rt->rt_mark = ort->rt_mark;
2805 rt->rt_genid = rt_genid(net);
2806 rt->rt_flags = ort->rt_flags;
2807 rt->rt_type = ort->rt_type;
2808 rt->rt_dst = ort->rt_dst;
2809 rt->rt_src = ort->rt_src;
2810 rt->rt_gateway = ort->rt_gateway;
2811 rt_transfer_peer(rt, ort);
2814 atomic_inc(&rt->fi->fib_clntref);
2819 dst_release(dst_orig);
2821 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2824 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2827 struct rtable *rt = __ip_route_output_key(net, flp4);
2832 if (flp4->flowi4_proto)
2833 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2834 flowi4_to_flowi(flp4),
2839 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2841 static int rt_fill_info(struct net *net,
2842 struct sk_buff *skb, u32 pid, u32 seq, int event,
2843 int nowait, unsigned int flags)
2845 struct rtable *rt = skb_rtable(skb);
2847 struct nlmsghdr *nlh;
2848 unsigned long expires = 0;
2849 u32 id = 0, ts = 0, tsage = 0, error;
2851 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2855 r = nlmsg_data(nlh);
2856 r->rtm_family = AF_INET;
2857 r->rtm_dst_len = 32;
2859 r->rtm_tos = rt->rt_key_tos;
2860 r->rtm_table = RT_TABLE_MAIN;
2861 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2862 goto nla_put_failure;
2863 r->rtm_type = rt->rt_type;
2864 r->rtm_scope = RT_SCOPE_UNIVERSE;
2865 r->rtm_protocol = RTPROT_UNSPEC;
2866 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2867 if (rt->rt_flags & RTCF_NOTIFY)
2868 r->rtm_flags |= RTM_F_NOTIFY;
2870 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2871 goto nla_put_failure;
2872 if (rt->rt_key_src) {
2873 r->rtm_src_len = 32;
2874 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2875 goto nla_put_failure;
2878 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2879 goto nla_put_failure;
2880 #ifdef CONFIG_IP_ROUTE_CLASSID
2881 if (rt->dst.tclassid &&
2882 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2883 goto nla_put_failure;
2885 if (!rt_is_input_route(rt) &&
2886 rt->rt_src != rt->rt_key_src) {
2887 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2888 goto nla_put_failure;
2890 if (rt->rt_dst != rt->rt_gateway &&
2891 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2892 goto nla_put_failure;
2894 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2895 goto nla_put_failure;
2898 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2899 goto nla_put_failure;
2901 error = rt->dst.error;
2902 if (rt_has_peer(rt)) {
2903 const struct inet_peer *peer = rt_peer_ptr(rt);
2904 inet_peer_refcheck(peer);
2905 id = atomic_read(&peer->ip_id_count) & 0xffff;
2906 if (peer->tcp_ts_stamp) {
2908 tsage = get_seconds() - peer->tcp_ts_stamp;
2910 expires = ACCESS_ONCE(peer->pmtu_expires);
2912 if (time_before(jiffies, expires))
2919 if (rt_is_input_route(rt)) {
2920 #ifdef CONFIG_IP_MROUTE
2921 __be32 dst = rt->rt_dst;
2923 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2924 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2925 int err = ipmr_get_route(net, skb,
2926 rt->rt_src, rt->rt_dst,
2932 goto nla_put_failure;
2934 if (err == -EMSGSIZE)
2935 goto nla_put_failure;
2941 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2942 goto nla_put_failure;
2945 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2946 expires, error) < 0)
2947 goto nla_put_failure;
2949 return nlmsg_end(skb, nlh);
2952 nlmsg_cancel(skb, nlh);
2956 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2958 struct net *net = sock_net(in_skb->sk);
2960 struct nlattr *tb[RTA_MAX+1];
2961 struct rtable *rt = NULL;
2967 struct sk_buff *skb;
2969 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2973 rtm = nlmsg_data(nlh);
2975 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2981 /* Reserve room for dummy headers, this skb can pass
2982 through good chunk of routing engine.
2984 skb_reset_mac_header(skb);
2985 skb_reset_network_header(skb);
2987 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2988 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2989 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2991 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2992 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2993 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2994 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2997 struct net_device *dev;
2999 dev = __dev_get_by_index(net, iif);
3005 skb->protocol = htons(ETH_P_IP);
3009 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3012 rt = skb_rtable(skb);
3013 if (err == 0 && rt->dst.error)
3014 err = -rt->dst.error;
3016 struct flowi4 fl4 = {
3019 .flowi4_tos = rtm->rtm_tos,
3020 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3021 .flowi4_mark = mark,
3023 rt = ip_route_output_key(net, &fl4);
3033 skb_dst_set(skb, &rt->dst);
3034 if (rtm->rtm_flags & RTM_F_NOTIFY)
3035 rt->rt_flags |= RTCF_NOTIFY;
3037 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3038 RTM_NEWROUTE, 0, 0);
3042 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3051 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3058 net = sock_net(skb->sk);
3063 s_idx = idx = cb->args[1];
3064 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3065 if (!rt_hash_table[h].chain)
3068 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3069 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3070 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3072 if (rt_is_expired(rt))
3074 skb_dst_set_noref(skb, &rt->dst);
3075 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3076 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3077 1, NLM_F_MULTI) <= 0) {
3079 rcu_read_unlock_bh();
3084 rcu_read_unlock_bh();
3093 void ip_rt_multicast_event(struct in_device *in_dev)
3095 rt_cache_flush(dev_net(in_dev->dev), 0);
3098 #ifdef CONFIG_SYSCTL
3099 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3100 void __user *buffer,
3101 size_t *lenp, loff_t *ppos)
3108 memcpy(&ctl, __ctl, sizeof(ctl));
3109 ctl.data = &flush_delay;
3110 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3112 net = (struct net *)__ctl->extra1;
3113 rt_cache_flush(net, flush_delay);
3120 static ctl_table ipv4_route_table[] = {
3122 .procname = "gc_thresh",
3123 .data = &ipv4_dst_ops.gc_thresh,
3124 .maxlen = sizeof(int),
3126 .proc_handler = proc_dointvec,
3129 .procname = "max_size",
3130 .data = &ip_rt_max_size,
3131 .maxlen = sizeof(int),
3133 .proc_handler = proc_dointvec,
3136 /* Deprecated. Use gc_min_interval_ms */
3138 .procname = "gc_min_interval",
3139 .data = &ip_rt_gc_min_interval,
3140 .maxlen = sizeof(int),
3142 .proc_handler = proc_dointvec_jiffies,
3145 .procname = "gc_min_interval_ms",
3146 .data = &ip_rt_gc_min_interval,
3147 .maxlen = sizeof(int),
3149 .proc_handler = proc_dointvec_ms_jiffies,
3152 .procname = "gc_timeout",
3153 .data = &ip_rt_gc_timeout,
3154 .maxlen = sizeof(int),
3156 .proc_handler = proc_dointvec_jiffies,
3159 .procname = "gc_interval",
3160 .data = &ip_rt_gc_interval,
3161 .maxlen = sizeof(int),
3163 .proc_handler = proc_dointvec_jiffies,
3166 .procname = "redirect_load",
3167 .data = &ip_rt_redirect_load,
3168 .maxlen = sizeof(int),
3170 .proc_handler = proc_dointvec,
3173 .procname = "redirect_number",
3174 .data = &ip_rt_redirect_number,
3175 .maxlen = sizeof(int),
3177 .proc_handler = proc_dointvec,
3180 .procname = "redirect_silence",
3181 .data = &ip_rt_redirect_silence,
3182 .maxlen = sizeof(int),
3184 .proc_handler = proc_dointvec,
3187 .procname = "error_cost",
3188 .data = &ip_rt_error_cost,
3189 .maxlen = sizeof(int),
3191 .proc_handler = proc_dointvec,
3194 .procname = "error_burst",
3195 .data = &ip_rt_error_burst,
3196 .maxlen = sizeof(int),
3198 .proc_handler = proc_dointvec,
3201 .procname = "gc_elasticity",
3202 .data = &ip_rt_gc_elasticity,
3203 .maxlen = sizeof(int),
3205 .proc_handler = proc_dointvec,
3208 .procname = "mtu_expires",
3209 .data = &ip_rt_mtu_expires,
3210 .maxlen = sizeof(int),
3212 .proc_handler = proc_dointvec_jiffies,
3215 .procname = "min_pmtu",
3216 .data = &ip_rt_min_pmtu,
3217 .maxlen = sizeof(int),
3219 .proc_handler = proc_dointvec,
3222 .procname = "min_adv_mss",
3223 .data = &ip_rt_min_advmss,
3224 .maxlen = sizeof(int),
3226 .proc_handler = proc_dointvec,
3231 static struct ctl_table ipv4_route_flush_table[] = {
3233 .procname = "flush",
3234 .maxlen = sizeof(int),
3236 .proc_handler = ipv4_sysctl_rtcache_flush,
3241 static __net_init int sysctl_route_net_init(struct net *net)
3243 struct ctl_table *tbl;
3245 tbl = ipv4_route_flush_table;
3246 if (!net_eq(net, &init_net)) {
3247 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3251 tbl[0].extra1 = net;
3253 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3254 if (net->ipv4.route_hdr == NULL)
3259 if (tbl != ipv4_route_flush_table)
3265 static __net_exit void sysctl_route_net_exit(struct net *net)
3267 struct ctl_table *tbl;
3269 tbl = net->ipv4.route_hdr->ctl_table_arg;
3270 unregister_net_sysctl_table(net->ipv4.route_hdr);
3271 BUG_ON(tbl == ipv4_route_flush_table);
3275 static __net_initdata struct pernet_operations sysctl_route_ops = {
3276 .init = sysctl_route_net_init,
3277 .exit = sysctl_route_net_exit,
3281 static __net_init int rt_genid_init(struct net *net)
3283 get_random_bytes(&net->ipv4.rt_genid,
3284 sizeof(net->ipv4.rt_genid));
3285 get_random_bytes(&net->ipv4.dev_addr_genid,
3286 sizeof(net->ipv4.dev_addr_genid));
3290 static __net_initdata struct pernet_operations rt_genid_ops = {
3291 .init = rt_genid_init,
3294 static int __net_init ipv4_inetpeer_init(struct net *net)
3296 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3300 inet_peer_base_init(bp);
3301 net->ipv4.peers = bp;
3305 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3307 struct inet_peer_base *bp = net->ipv4.peers;
3309 net->ipv4.peers = NULL;
3310 inetpeer_invalidate_tree(bp);
3314 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3315 .init = ipv4_inetpeer_init,
3316 .exit = ipv4_inetpeer_exit,
3319 #ifdef CONFIG_IP_ROUTE_CLASSID
3320 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3321 #endif /* CONFIG_IP_ROUTE_CLASSID */
3323 static __initdata unsigned long rhash_entries;
3324 static int __init set_rhash_entries(char *str)
3331 ret = kstrtoul(str, 0, &rhash_entries);
3337 __setup("rhash_entries=", set_rhash_entries);
3339 int __init ip_rt_init(void)
3343 #ifdef CONFIG_IP_ROUTE_CLASSID
3344 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3346 panic("IP: failed to allocate ip_rt_acct\n");
3349 ipv4_dst_ops.kmem_cachep =
3350 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3351 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3353 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3355 if (dst_entries_init(&ipv4_dst_ops) < 0)
3356 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3358 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3359 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3361 rt_hash_table = (struct rt_hash_bucket *)
3362 alloc_large_system_hash("IP route cache",
3363 sizeof(struct rt_hash_bucket),
3365 (totalram_pages >= 128 * 1024) ?
3371 rhash_entries ? 0 : 512 * 1024);
3372 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3373 rt_hash_lock_init();
3375 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3376 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3381 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3382 expires_ljiffies = jiffies;
3383 schedule_delayed_work(&expires_work,
3384 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3386 if (ip_rt_proc_init())
3387 pr_err("Unable to create route proc files\n");
3390 xfrm4_init(ip_rt_max_size);
3392 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3394 #ifdef CONFIG_SYSCTL
3395 register_pernet_subsys(&sysctl_route_ops);
3397 register_pernet_subsys(&rt_genid_ops);
3398 register_pernet_subsys(&ipv4_inetpeer_ops);
3402 #ifdef CONFIG_SYSCTL
3404 * We really need to sanitize the damn ipv4 init order, then all
3405 * this nonsense will go away.
3407 void __init ip_static_sysctl_init(void)
3409 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);