Loading...
1/*
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.
5 *
6 * ROUTE - implementation of the IP router.
7 *
8 * Authors: Ross Biro
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>
13 *
14 * Fixes:
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
24 * clamper.
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.
39 *
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
58 *
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.
63 */
64
65#define pr_fmt(fmt) "IPv4: " fmt
66
67#include <linux/module.h>
68#include <linux/uaccess.h>
69#include <linux/bitops.h>
70#include <linux/types.h>
71#include <linux/kernel.h>
72#include <linux/mm.h>
73#include <linux/string.h>
74#include <linux/socket.h>
75#include <linux/sockios.h>
76#include <linux/errno.h>
77#include <linux/in.h>
78#include <linux/inet.h>
79#include <linux/netdevice.h>
80#include <linux/proc_fs.h>
81#include <linux/init.h>
82#include <linux/skbuff.h>
83#include <linux/inetdevice.h>
84#include <linux/igmp.h>
85#include <linux/pkt_sched.h>
86#include <linux/mroute.h>
87#include <linux/netfilter_ipv4.h>
88#include <linux/random.h>
89#include <linux/rcupdate.h>
90#include <linux/times.h>
91#include <linux/slab.h>
92#include <linux/jhash.h>
93#include <net/dst.h>
94#include <net/dst_metadata.h>
95#include <net/net_namespace.h>
96#include <net/protocol.h>
97#include <net/ip.h>
98#include <net/route.h>
99#include <net/inetpeer.h>
100#include <net/sock.h>
101#include <net/ip_fib.h>
102#include <net/arp.h>
103#include <net/tcp.h>
104#include <net/icmp.h>
105#include <net/xfrm.h>
106#include <net/lwtunnel.h>
107#include <net/netevent.h>
108#include <net/rtnetlink.h>
109#ifdef CONFIG_SYSCTL
110#include <linux/sysctl.h>
111#endif
112#include <net/secure_seq.h>
113#include <net/ip_tunnels.h>
114#include <net/l3mdev.h>
115
116#include "fib_lookup.h"
117
118#define RT_FL_TOS(oldflp4) \
119 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
120
121#define RT_GC_TIMEOUT (300*HZ)
122
123static int ip_rt_max_size;
124static int ip_rt_redirect_number __read_mostly = 9;
125static int ip_rt_redirect_load __read_mostly = HZ / 50;
126static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
127static int ip_rt_error_cost __read_mostly = HZ;
128static int ip_rt_error_burst __read_mostly = 5 * HZ;
129static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
130static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
131static int ip_rt_min_advmss __read_mostly = 256;
132
133static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
134
135/*
136 * Interface to generic destination cache.
137 */
138
139static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
140static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
141static unsigned int ipv4_mtu(const struct dst_entry *dst);
142static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143static void ipv4_link_failure(struct sk_buff *skb);
144static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
145 struct sk_buff *skb, u32 mtu);
146static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
147 struct sk_buff *skb);
148static void ipv4_dst_destroy(struct dst_entry *dst);
149
150static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
151{
152 WARN_ON(1);
153 return NULL;
154}
155
156static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
157 struct sk_buff *skb,
158 const void *daddr);
159static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
160
161static struct dst_ops ipv4_dst_ops = {
162 .family = AF_INET,
163 .check = ipv4_dst_check,
164 .default_advmss = ipv4_default_advmss,
165 .mtu = ipv4_mtu,
166 .cow_metrics = ipv4_cow_metrics,
167 .destroy = ipv4_dst_destroy,
168 .negative_advice = ipv4_negative_advice,
169 .link_failure = ipv4_link_failure,
170 .update_pmtu = ip_rt_update_pmtu,
171 .redirect = ip_do_redirect,
172 .local_out = __ip_local_out,
173 .neigh_lookup = ipv4_neigh_lookup,
174 .confirm_neigh = ipv4_confirm_neigh,
175};
176
177#define ECN_OR_COST(class) TC_PRIO_##class
178
179const __u8 ip_tos2prio[16] = {
180 TC_PRIO_BESTEFFORT,
181 ECN_OR_COST(BESTEFFORT),
182 TC_PRIO_BESTEFFORT,
183 ECN_OR_COST(BESTEFFORT),
184 TC_PRIO_BULK,
185 ECN_OR_COST(BULK),
186 TC_PRIO_BULK,
187 ECN_OR_COST(BULK),
188 TC_PRIO_INTERACTIVE,
189 ECN_OR_COST(INTERACTIVE),
190 TC_PRIO_INTERACTIVE,
191 ECN_OR_COST(INTERACTIVE),
192 TC_PRIO_INTERACTIVE_BULK,
193 ECN_OR_COST(INTERACTIVE_BULK),
194 TC_PRIO_INTERACTIVE_BULK,
195 ECN_OR_COST(INTERACTIVE_BULK)
196};
197EXPORT_SYMBOL(ip_tos2prio);
198
199static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
200#define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
201
202#ifdef CONFIG_PROC_FS
203static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
204{
205 if (*pos)
206 return NULL;
207 return SEQ_START_TOKEN;
208}
209
210static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
211{
212 ++*pos;
213 return NULL;
214}
215
216static void rt_cache_seq_stop(struct seq_file *seq, void *v)
217{
218}
219
220static int rt_cache_seq_show(struct seq_file *seq, void *v)
221{
222 if (v == SEQ_START_TOKEN)
223 seq_printf(seq, "%-127s\n",
224 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
225 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
226 "HHUptod\tSpecDst");
227 return 0;
228}
229
230static const struct seq_operations rt_cache_seq_ops = {
231 .start = rt_cache_seq_start,
232 .next = rt_cache_seq_next,
233 .stop = rt_cache_seq_stop,
234 .show = rt_cache_seq_show,
235};
236
237static int rt_cache_seq_open(struct inode *inode, struct file *file)
238{
239 return seq_open(file, &rt_cache_seq_ops);
240}
241
242static const struct file_operations rt_cache_seq_fops = {
243 .open = rt_cache_seq_open,
244 .read = seq_read,
245 .llseek = seq_lseek,
246 .release = seq_release,
247};
248
249
250static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
251{
252 int cpu;
253
254 if (*pos == 0)
255 return SEQ_START_TOKEN;
256
257 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
258 if (!cpu_possible(cpu))
259 continue;
260 *pos = cpu+1;
261 return &per_cpu(rt_cache_stat, cpu);
262 }
263 return NULL;
264}
265
266static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
267{
268 int cpu;
269
270 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
271 if (!cpu_possible(cpu))
272 continue;
273 *pos = cpu+1;
274 return &per_cpu(rt_cache_stat, cpu);
275 }
276 return NULL;
277
278}
279
280static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
281{
282
283}
284
285static int rt_cpu_seq_show(struct seq_file *seq, void *v)
286{
287 struct rt_cache_stat *st = v;
288
289 if (v == SEQ_START_TOKEN) {
290 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");
291 return 0;
292 }
293
294 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
295 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
296 dst_entries_get_slow(&ipv4_dst_ops),
297 0, /* st->in_hit */
298 st->in_slow_tot,
299 st->in_slow_mc,
300 st->in_no_route,
301 st->in_brd,
302 st->in_martian_dst,
303 st->in_martian_src,
304
305 0, /* st->out_hit */
306 st->out_slow_tot,
307 st->out_slow_mc,
308
309 0, /* st->gc_total */
310 0, /* st->gc_ignored */
311 0, /* st->gc_goal_miss */
312 0, /* st->gc_dst_overflow */
313 0, /* st->in_hlist_search */
314 0 /* st->out_hlist_search */
315 );
316 return 0;
317}
318
319static const struct seq_operations rt_cpu_seq_ops = {
320 .start = rt_cpu_seq_start,
321 .next = rt_cpu_seq_next,
322 .stop = rt_cpu_seq_stop,
323 .show = rt_cpu_seq_show,
324};
325
326
327static int rt_cpu_seq_open(struct inode *inode, struct file *file)
328{
329 return seq_open(file, &rt_cpu_seq_ops);
330}
331
332static const struct file_operations rt_cpu_seq_fops = {
333 .open = rt_cpu_seq_open,
334 .read = seq_read,
335 .llseek = seq_lseek,
336 .release = seq_release,
337};
338
339#ifdef CONFIG_IP_ROUTE_CLASSID
340static int rt_acct_proc_show(struct seq_file *m, void *v)
341{
342 struct ip_rt_acct *dst, *src;
343 unsigned int i, j;
344
345 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
346 if (!dst)
347 return -ENOMEM;
348
349 for_each_possible_cpu(i) {
350 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
351 for (j = 0; j < 256; j++) {
352 dst[j].o_bytes += src[j].o_bytes;
353 dst[j].o_packets += src[j].o_packets;
354 dst[j].i_bytes += src[j].i_bytes;
355 dst[j].i_packets += src[j].i_packets;
356 }
357 }
358
359 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
360 kfree(dst);
361 return 0;
362}
363
364static int rt_acct_proc_open(struct inode *inode, struct file *file)
365{
366 return single_open(file, rt_acct_proc_show, NULL);
367}
368
369static const struct file_operations rt_acct_proc_fops = {
370 .open = rt_acct_proc_open,
371 .read = seq_read,
372 .llseek = seq_lseek,
373 .release = single_release,
374};
375#endif
376
377static int __net_init ip_rt_do_proc_init(struct net *net)
378{
379 struct proc_dir_entry *pde;
380
381 pde = proc_create("rt_cache", 0444, net->proc_net,
382 &rt_cache_seq_fops);
383 if (!pde)
384 goto err1;
385
386 pde = proc_create("rt_cache", 0444,
387 net->proc_net_stat, &rt_cpu_seq_fops);
388 if (!pde)
389 goto err2;
390
391#ifdef CONFIG_IP_ROUTE_CLASSID
392 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
393 if (!pde)
394 goto err3;
395#endif
396 return 0;
397
398#ifdef CONFIG_IP_ROUTE_CLASSID
399err3:
400 remove_proc_entry("rt_cache", net->proc_net_stat);
401#endif
402err2:
403 remove_proc_entry("rt_cache", net->proc_net);
404err1:
405 return -ENOMEM;
406}
407
408static void __net_exit ip_rt_do_proc_exit(struct net *net)
409{
410 remove_proc_entry("rt_cache", net->proc_net_stat);
411 remove_proc_entry("rt_cache", net->proc_net);
412#ifdef CONFIG_IP_ROUTE_CLASSID
413 remove_proc_entry("rt_acct", net->proc_net);
414#endif
415}
416
417static struct pernet_operations ip_rt_proc_ops __net_initdata = {
418 .init = ip_rt_do_proc_init,
419 .exit = ip_rt_do_proc_exit,
420};
421
422static int __init ip_rt_proc_init(void)
423{
424 return register_pernet_subsys(&ip_rt_proc_ops);
425}
426
427#else
428static inline int ip_rt_proc_init(void)
429{
430 return 0;
431}
432#endif /* CONFIG_PROC_FS */
433
434static inline bool rt_is_expired(const struct rtable *rth)
435{
436 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
437}
438
439void rt_cache_flush(struct net *net)
440{
441 rt_genid_bump_ipv4(net);
442}
443
444static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
445 struct sk_buff *skb,
446 const void *daddr)
447{
448 struct net_device *dev = dst->dev;
449 const __be32 *pkey = daddr;
450 const struct rtable *rt;
451 struct neighbour *n;
452
453 rt = (const struct rtable *) dst;
454 if (rt->rt_gateway)
455 pkey = (const __be32 *) &rt->rt_gateway;
456 else if (skb)
457 pkey = &ip_hdr(skb)->daddr;
458
459 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
460 if (n)
461 return n;
462 return neigh_create(&arp_tbl, pkey, dev);
463}
464
465static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
466{
467 struct net_device *dev = dst->dev;
468 const __be32 *pkey = daddr;
469 const struct rtable *rt;
470
471 rt = (const struct rtable *)dst;
472 if (rt->rt_gateway)
473 pkey = (const __be32 *)&rt->rt_gateway;
474 else if (!daddr ||
475 (rt->rt_flags &
476 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
477 return;
478
479 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
480}
481
482#define IP_IDENTS_SZ 2048u
483
484static atomic_t *ip_idents __read_mostly;
485static u32 *ip_tstamps __read_mostly;
486
487/* In order to protect privacy, we add a perturbation to identifiers
488 * if one generator is seldom used. This makes hard for an attacker
489 * to infer how many packets were sent between two points in time.
490 */
491u32 ip_idents_reserve(u32 hash, int segs)
492{
493 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
494 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
495 u32 old = READ_ONCE(*p_tstamp);
496 u32 now = (u32)jiffies;
497 u32 new, delta = 0;
498
499 if (old != now && cmpxchg(p_tstamp, old, now) == old)
500 delta = prandom_u32_max(now - old);
501
502 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
503 do {
504 old = (u32)atomic_read(p_id);
505 new = old + delta + segs;
506 } while (atomic_cmpxchg(p_id, old, new) != old);
507
508 return new - segs;
509}
510EXPORT_SYMBOL(ip_idents_reserve);
511
512void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
513{
514 static u32 ip_idents_hashrnd __read_mostly;
515 u32 hash, id;
516
517 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
518
519 hash = jhash_3words((__force u32)iph->daddr,
520 (__force u32)iph->saddr,
521 iph->protocol ^ net_hash_mix(net),
522 ip_idents_hashrnd);
523 id = ip_idents_reserve(hash, segs);
524 iph->id = htons(id);
525}
526EXPORT_SYMBOL(__ip_select_ident);
527
528static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
529 const struct sock *sk,
530 const struct iphdr *iph,
531 int oif, u8 tos,
532 u8 prot, u32 mark, int flow_flags)
533{
534 if (sk) {
535 const struct inet_sock *inet = inet_sk(sk);
536
537 oif = sk->sk_bound_dev_if;
538 mark = sk->sk_mark;
539 tos = RT_CONN_FLAGS(sk);
540 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
541 }
542 flowi4_init_output(fl4, oif, mark, tos,
543 RT_SCOPE_UNIVERSE, prot,
544 flow_flags,
545 iph->daddr, iph->saddr, 0, 0,
546 sock_net_uid(net, sk));
547}
548
549static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
550 const struct sock *sk)
551{
552 const struct net *net = dev_net(skb->dev);
553 const struct iphdr *iph = ip_hdr(skb);
554 int oif = skb->dev->ifindex;
555 u8 tos = RT_TOS(iph->tos);
556 u8 prot = iph->protocol;
557 u32 mark = skb->mark;
558
559 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
560}
561
562static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
563{
564 const struct inet_sock *inet = inet_sk(sk);
565 const struct ip_options_rcu *inet_opt;
566 __be32 daddr = inet->inet_daddr;
567
568 rcu_read_lock();
569 inet_opt = rcu_dereference(inet->inet_opt);
570 if (inet_opt && inet_opt->opt.srr)
571 daddr = inet_opt->opt.faddr;
572 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
573 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
574 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
575 inet_sk_flowi_flags(sk),
576 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
577 rcu_read_unlock();
578}
579
580static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
581 const struct sk_buff *skb)
582{
583 if (skb)
584 build_skb_flow_key(fl4, skb, sk);
585 else
586 build_sk_flow_key(fl4, sk);
587}
588
589static DEFINE_SPINLOCK(fnhe_lock);
590
591static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
592{
593 struct rtable *rt;
594
595 rt = rcu_dereference(fnhe->fnhe_rth_input);
596 if (rt) {
597 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
598 dst_dev_put(&rt->dst);
599 dst_release(&rt->dst);
600 }
601 rt = rcu_dereference(fnhe->fnhe_rth_output);
602 if (rt) {
603 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
604 dst_dev_put(&rt->dst);
605 dst_release(&rt->dst);
606 }
607}
608
609static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
610{
611 struct fib_nh_exception *fnhe, *oldest;
612
613 oldest = rcu_dereference(hash->chain);
614 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
615 fnhe = rcu_dereference(fnhe->fnhe_next)) {
616 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
617 oldest = fnhe;
618 }
619 fnhe_flush_routes(oldest);
620 return oldest;
621}
622
623static inline u32 fnhe_hashfun(__be32 daddr)
624{
625 static u32 fnhe_hashrnd __read_mostly;
626 u32 hval;
627
628 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
629 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
630 return hash_32(hval, FNHE_HASH_SHIFT);
631}
632
633static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
634{
635 rt->rt_pmtu = fnhe->fnhe_pmtu;
636 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
637 rt->dst.expires = fnhe->fnhe_expires;
638
639 if (fnhe->fnhe_gw) {
640 rt->rt_flags |= RTCF_REDIRECTED;
641 rt->rt_gateway = fnhe->fnhe_gw;
642 rt->rt_uses_gateway = 1;
643 }
644}
645
646static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
647 u32 pmtu, bool lock, unsigned long expires)
648{
649 struct fnhe_hash_bucket *hash;
650 struct fib_nh_exception *fnhe;
651 struct rtable *rt;
652 u32 genid, hval;
653 unsigned int i;
654 int depth;
655
656 genid = fnhe_genid(dev_net(nh->nh_dev));
657 hval = fnhe_hashfun(daddr);
658
659 spin_lock_bh(&fnhe_lock);
660
661 hash = rcu_dereference(nh->nh_exceptions);
662 if (!hash) {
663 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
664 if (!hash)
665 goto out_unlock;
666 rcu_assign_pointer(nh->nh_exceptions, hash);
667 }
668
669 hash += hval;
670
671 depth = 0;
672 for (fnhe = rcu_dereference(hash->chain); fnhe;
673 fnhe = rcu_dereference(fnhe->fnhe_next)) {
674 if (fnhe->fnhe_daddr == daddr)
675 break;
676 depth++;
677 }
678
679 if (fnhe) {
680 if (fnhe->fnhe_genid != genid)
681 fnhe->fnhe_genid = genid;
682 if (gw)
683 fnhe->fnhe_gw = gw;
684 if (pmtu) {
685 fnhe->fnhe_pmtu = pmtu;
686 fnhe->fnhe_mtu_locked = lock;
687 }
688 fnhe->fnhe_expires = max(1UL, expires);
689 /* Update all cached dsts too */
690 rt = rcu_dereference(fnhe->fnhe_rth_input);
691 if (rt)
692 fill_route_from_fnhe(rt, fnhe);
693 rt = rcu_dereference(fnhe->fnhe_rth_output);
694 if (rt)
695 fill_route_from_fnhe(rt, fnhe);
696 } else {
697 if (depth > FNHE_RECLAIM_DEPTH)
698 fnhe = fnhe_oldest(hash);
699 else {
700 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
701 if (!fnhe)
702 goto out_unlock;
703
704 fnhe->fnhe_next = hash->chain;
705 rcu_assign_pointer(hash->chain, fnhe);
706 }
707 fnhe->fnhe_genid = genid;
708 fnhe->fnhe_daddr = daddr;
709 fnhe->fnhe_gw = gw;
710 fnhe->fnhe_pmtu = pmtu;
711 fnhe->fnhe_mtu_locked = lock;
712 fnhe->fnhe_expires = max(1UL, expires);
713
714 /* Exception created; mark the cached routes for the nexthop
715 * stale, so anyone caching it rechecks if this exception
716 * applies to them.
717 */
718 rt = rcu_dereference(nh->nh_rth_input);
719 if (rt)
720 rt->dst.obsolete = DST_OBSOLETE_KILL;
721
722 for_each_possible_cpu(i) {
723 struct rtable __rcu **prt;
724 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
725 rt = rcu_dereference(*prt);
726 if (rt)
727 rt->dst.obsolete = DST_OBSOLETE_KILL;
728 }
729 }
730
731 fnhe->fnhe_stamp = jiffies;
732
733out_unlock:
734 spin_unlock_bh(&fnhe_lock);
735}
736
737static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
738 bool kill_route)
739{
740 __be32 new_gw = icmp_hdr(skb)->un.gateway;
741 __be32 old_gw = ip_hdr(skb)->saddr;
742 struct net_device *dev = skb->dev;
743 struct in_device *in_dev;
744 struct fib_result res;
745 struct neighbour *n;
746 struct net *net;
747
748 switch (icmp_hdr(skb)->code & 7) {
749 case ICMP_REDIR_NET:
750 case ICMP_REDIR_NETTOS:
751 case ICMP_REDIR_HOST:
752 case ICMP_REDIR_HOSTTOS:
753 break;
754
755 default:
756 return;
757 }
758
759 if (rt->rt_gateway != old_gw)
760 return;
761
762 in_dev = __in_dev_get_rcu(dev);
763 if (!in_dev)
764 return;
765
766 net = dev_net(dev);
767 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
768 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
769 ipv4_is_zeronet(new_gw))
770 goto reject_redirect;
771
772 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
773 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
774 goto reject_redirect;
775 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
776 goto reject_redirect;
777 } else {
778 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
779 goto reject_redirect;
780 }
781
782 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
783 if (!n)
784 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
785 if (!IS_ERR(n)) {
786 if (!(n->nud_state & NUD_VALID)) {
787 neigh_event_send(n, NULL);
788 } else {
789 if (fib_lookup(net, fl4, &res, 0) == 0) {
790 struct fib_nh *nh = &FIB_RES_NH(res);
791
792 update_or_create_fnhe(nh, fl4->daddr, new_gw,
793 0, false,
794 jiffies + ip_rt_gc_timeout);
795 }
796 if (kill_route)
797 rt->dst.obsolete = DST_OBSOLETE_KILL;
798 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
799 }
800 neigh_release(n);
801 }
802 return;
803
804reject_redirect:
805#ifdef CONFIG_IP_ROUTE_VERBOSE
806 if (IN_DEV_LOG_MARTIANS(in_dev)) {
807 const struct iphdr *iph = (const struct iphdr *) skb->data;
808 __be32 daddr = iph->daddr;
809 __be32 saddr = iph->saddr;
810
811 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
812 " Advised path = %pI4 -> %pI4\n",
813 &old_gw, dev->name, &new_gw,
814 &saddr, &daddr);
815 }
816#endif
817 ;
818}
819
820static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
821{
822 struct rtable *rt;
823 struct flowi4 fl4;
824 const struct iphdr *iph = (const struct iphdr *) skb->data;
825 struct net *net = dev_net(skb->dev);
826 int oif = skb->dev->ifindex;
827 u8 tos = RT_TOS(iph->tos);
828 u8 prot = iph->protocol;
829 u32 mark = skb->mark;
830
831 rt = (struct rtable *) dst;
832
833 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
834 __ip_do_redirect(rt, skb, &fl4, true);
835}
836
837static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
838{
839 struct rtable *rt = (struct rtable *)dst;
840 struct dst_entry *ret = dst;
841
842 if (rt) {
843 if (dst->obsolete > 0) {
844 ip_rt_put(rt);
845 ret = NULL;
846 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
847 rt->dst.expires) {
848 ip_rt_put(rt);
849 ret = NULL;
850 }
851 }
852 return ret;
853}
854
855/*
856 * Algorithm:
857 * 1. The first ip_rt_redirect_number redirects are sent
858 * with exponential backoff, then we stop sending them at all,
859 * assuming that the host ignores our redirects.
860 * 2. If we did not see packets requiring redirects
861 * during ip_rt_redirect_silence, we assume that the host
862 * forgot redirected route and start to send redirects again.
863 *
864 * This algorithm is much cheaper and more intelligent than dumb load limiting
865 * in icmp.c.
866 *
867 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
868 * and "frag. need" (breaks PMTU discovery) in icmp.c.
869 */
870
871void ip_rt_send_redirect(struct sk_buff *skb)
872{
873 struct rtable *rt = skb_rtable(skb);
874 struct in_device *in_dev;
875 struct inet_peer *peer;
876 struct net *net;
877 int log_martians;
878 int vif;
879
880 rcu_read_lock();
881 in_dev = __in_dev_get_rcu(rt->dst.dev);
882 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
883 rcu_read_unlock();
884 return;
885 }
886 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
887 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
888 rcu_read_unlock();
889
890 net = dev_net(rt->dst.dev);
891 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
892 if (!peer) {
893 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
894 rt_nexthop(rt, ip_hdr(skb)->daddr));
895 return;
896 }
897
898 /* No redirected packets during ip_rt_redirect_silence;
899 * reset the algorithm.
900 */
901 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
902 peer->rate_tokens = 0;
903
904 /* Too many ignored redirects; do not send anything
905 * set dst.rate_last to the last seen redirected packet.
906 */
907 if (peer->rate_tokens >= ip_rt_redirect_number) {
908 peer->rate_last = jiffies;
909 goto out_put_peer;
910 }
911
912 /* Check for load limit; set rate_last to the latest sent
913 * redirect.
914 */
915 if (peer->rate_tokens == 0 ||
916 time_after(jiffies,
917 (peer->rate_last +
918 (ip_rt_redirect_load << peer->rate_tokens)))) {
919 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
920
921 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
922 peer->rate_last = jiffies;
923 ++peer->rate_tokens;
924#ifdef CONFIG_IP_ROUTE_VERBOSE
925 if (log_martians &&
926 peer->rate_tokens == ip_rt_redirect_number)
927 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
928 &ip_hdr(skb)->saddr, inet_iif(skb),
929 &ip_hdr(skb)->daddr, &gw);
930#endif
931 }
932out_put_peer:
933 inet_putpeer(peer);
934}
935
936static int ip_error(struct sk_buff *skb)
937{
938 struct rtable *rt = skb_rtable(skb);
939 struct net_device *dev = skb->dev;
940 struct in_device *in_dev;
941 struct inet_peer *peer;
942 unsigned long now;
943 struct net *net;
944 bool send;
945 int code;
946
947 if (netif_is_l3_master(skb->dev)) {
948 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
949 if (!dev)
950 goto out;
951 }
952
953 in_dev = __in_dev_get_rcu(dev);
954
955 /* IP on this device is disabled. */
956 if (!in_dev)
957 goto out;
958
959 net = dev_net(rt->dst.dev);
960 if (!IN_DEV_FORWARD(in_dev)) {
961 switch (rt->dst.error) {
962 case EHOSTUNREACH:
963 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
964 break;
965
966 case ENETUNREACH:
967 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
968 break;
969 }
970 goto out;
971 }
972
973 switch (rt->dst.error) {
974 case EINVAL:
975 default:
976 goto out;
977 case EHOSTUNREACH:
978 code = ICMP_HOST_UNREACH;
979 break;
980 case ENETUNREACH:
981 code = ICMP_NET_UNREACH;
982 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
983 break;
984 case EACCES:
985 code = ICMP_PKT_FILTERED;
986 break;
987 }
988
989 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
990 l3mdev_master_ifindex(skb->dev), 1);
991
992 send = true;
993 if (peer) {
994 now = jiffies;
995 peer->rate_tokens += now - peer->rate_last;
996 if (peer->rate_tokens > ip_rt_error_burst)
997 peer->rate_tokens = ip_rt_error_burst;
998 peer->rate_last = now;
999 if (peer->rate_tokens >= ip_rt_error_cost)
1000 peer->rate_tokens -= ip_rt_error_cost;
1001 else
1002 send = false;
1003 inet_putpeer(peer);
1004 }
1005 if (send)
1006 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1007
1008out: kfree_skb(skb);
1009 return 0;
1010}
1011
1012static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1013{
1014 struct dst_entry *dst = &rt->dst;
1015 struct fib_result res;
1016 bool lock = false;
1017
1018 if (ip_mtu_locked(dst))
1019 return;
1020
1021 if (ipv4_mtu(dst) < mtu)
1022 return;
1023
1024 if (mtu < ip_rt_min_pmtu) {
1025 lock = true;
1026 mtu = ip_rt_min_pmtu;
1027 }
1028
1029 if (rt->rt_pmtu == mtu &&
1030 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1031 return;
1032
1033 rcu_read_lock();
1034 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1035 struct fib_nh *nh = &FIB_RES_NH(res);
1036
1037 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1038 jiffies + ip_rt_mtu_expires);
1039 }
1040 rcu_read_unlock();
1041}
1042
1043static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1044 struct sk_buff *skb, u32 mtu)
1045{
1046 struct rtable *rt = (struct rtable *) dst;
1047 struct flowi4 fl4;
1048
1049 ip_rt_build_flow_key(&fl4, sk, skb);
1050 __ip_rt_update_pmtu(rt, &fl4, mtu);
1051}
1052
1053void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1054 int oif, u32 mark, u8 protocol, int flow_flags)
1055{
1056 const struct iphdr *iph = (const struct iphdr *) skb->data;
1057 struct flowi4 fl4;
1058 struct rtable *rt;
1059
1060 if (!mark)
1061 mark = IP4_REPLY_MARK(net, skb->mark);
1062
1063 __build_flow_key(net, &fl4, NULL, iph, oif,
1064 RT_TOS(iph->tos), protocol, mark, flow_flags);
1065 rt = __ip_route_output_key(net, &fl4);
1066 if (!IS_ERR(rt)) {
1067 __ip_rt_update_pmtu(rt, &fl4, mtu);
1068 ip_rt_put(rt);
1069 }
1070}
1071EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1072
1073static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1074{
1075 const struct iphdr *iph = (const struct iphdr *) skb->data;
1076 struct flowi4 fl4;
1077 struct rtable *rt;
1078
1079 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1080
1081 if (!fl4.flowi4_mark)
1082 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1083
1084 rt = __ip_route_output_key(sock_net(sk), &fl4);
1085 if (!IS_ERR(rt)) {
1086 __ip_rt_update_pmtu(rt, &fl4, mtu);
1087 ip_rt_put(rt);
1088 }
1089}
1090
1091void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1092{
1093 const struct iphdr *iph = (const struct iphdr *) skb->data;
1094 struct flowi4 fl4;
1095 struct rtable *rt;
1096 struct dst_entry *odst = NULL;
1097 bool new = false;
1098 struct net *net = sock_net(sk);
1099
1100 bh_lock_sock(sk);
1101
1102 if (!ip_sk_accept_pmtu(sk))
1103 goto out;
1104
1105 odst = sk_dst_get(sk);
1106
1107 if (sock_owned_by_user(sk) || !odst) {
1108 __ipv4_sk_update_pmtu(skb, sk, mtu);
1109 goto out;
1110 }
1111
1112 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1113
1114 rt = (struct rtable *)odst;
1115 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1116 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1117 if (IS_ERR(rt))
1118 goto out;
1119
1120 new = true;
1121 }
1122
1123 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1124
1125 if (!dst_check(&rt->dst, 0)) {
1126 if (new)
1127 dst_release(&rt->dst);
1128
1129 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1130 if (IS_ERR(rt))
1131 goto out;
1132
1133 new = true;
1134 }
1135
1136 if (new)
1137 sk_dst_set(sk, &rt->dst);
1138
1139out:
1140 bh_unlock_sock(sk);
1141 dst_release(odst);
1142}
1143EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1144
1145void ipv4_redirect(struct sk_buff *skb, struct net *net,
1146 int oif, u32 mark, u8 protocol, int flow_flags)
1147{
1148 const struct iphdr *iph = (const struct iphdr *) skb->data;
1149 struct flowi4 fl4;
1150 struct rtable *rt;
1151
1152 __build_flow_key(net, &fl4, NULL, iph, oif,
1153 RT_TOS(iph->tos), protocol, mark, flow_flags);
1154 rt = __ip_route_output_key(net, &fl4);
1155 if (!IS_ERR(rt)) {
1156 __ip_do_redirect(rt, skb, &fl4, false);
1157 ip_rt_put(rt);
1158 }
1159}
1160EXPORT_SYMBOL_GPL(ipv4_redirect);
1161
1162void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1163{
1164 const struct iphdr *iph = (const struct iphdr *) skb->data;
1165 struct flowi4 fl4;
1166 struct rtable *rt;
1167 struct net *net = sock_net(sk);
1168
1169 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1170 rt = __ip_route_output_key(net, &fl4);
1171 if (!IS_ERR(rt)) {
1172 __ip_do_redirect(rt, skb, &fl4, false);
1173 ip_rt_put(rt);
1174 }
1175}
1176EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1177
1178static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1179{
1180 struct rtable *rt = (struct rtable *) dst;
1181
1182 /* All IPV4 dsts are created with ->obsolete set to the value
1183 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1184 * into this function always.
1185 *
1186 * When a PMTU/redirect information update invalidates a route,
1187 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1188 * DST_OBSOLETE_DEAD by dst_free().
1189 */
1190 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1191 return NULL;
1192 return dst;
1193}
1194
1195static void ipv4_link_failure(struct sk_buff *skb)
1196{
1197 struct rtable *rt;
1198
1199 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1200
1201 rt = skb_rtable(skb);
1202 if (rt)
1203 dst_set_expires(&rt->dst, 0);
1204}
1205
1206static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1207{
1208 pr_debug("%s: %pI4 -> %pI4, %s\n",
1209 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1210 skb->dev ? skb->dev->name : "?");
1211 kfree_skb(skb);
1212 WARN_ON(1);
1213 return 0;
1214}
1215
1216/*
1217 We do not cache source address of outgoing interface,
1218 because it is used only by IP RR, TS and SRR options,
1219 so that it out of fast path.
1220
1221 BTW remember: "addr" is allowed to be not aligned
1222 in IP options!
1223 */
1224
1225void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1226{
1227 __be32 src;
1228
1229 if (rt_is_output_route(rt))
1230 src = ip_hdr(skb)->saddr;
1231 else {
1232 struct fib_result res;
1233 struct flowi4 fl4;
1234 struct iphdr *iph;
1235
1236 iph = ip_hdr(skb);
1237
1238 memset(&fl4, 0, sizeof(fl4));
1239 fl4.daddr = iph->daddr;
1240 fl4.saddr = iph->saddr;
1241 fl4.flowi4_tos = RT_TOS(iph->tos);
1242 fl4.flowi4_oif = rt->dst.dev->ifindex;
1243 fl4.flowi4_iif = skb->dev->ifindex;
1244 fl4.flowi4_mark = skb->mark;
1245
1246 rcu_read_lock();
1247 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1248 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1249 else
1250 src = inet_select_addr(rt->dst.dev,
1251 rt_nexthop(rt, iph->daddr),
1252 RT_SCOPE_UNIVERSE);
1253 rcu_read_unlock();
1254 }
1255 memcpy(addr, &src, 4);
1256}
1257
1258#ifdef CONFIG_IP_ROUTE_CLASSID
1259static void set_class_tag(struct rtable *rt, u32 tag)
1260{
1261 if (!(rt->dst.tclassid & 0xFFFF))
1262 rt->dst.tclassid |= tag & 0xFFFF;
1263 if (!(rt->dst.tclassid & 0xFFFF0000))
1264 rt->dst.tclassid |= tag & 0xFFFF0000;
1265}
1266#endif
1267
1268static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1269{
1270 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1271 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1272 ip_rt_min_advmss);
1273
1274 return min(advmss, IPV4_MAX_PMTU - header_size);
1275}
1276
1277static unsigned int ipv4_mtu(const struct dst_entry *dst)
1278{
1279 const struct rtable *rt = (const struct rtable *) dst;
1280 unsigned int mtu = rt->rt_pmtu;
1281
1282 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1283 mtu = dst_metric_raw(dst, RTAX_MTU);
1284
1285 if (mtu)
1286 return mtu;
1287
1288 mtu = READ_ONCE(dst->dev->mtu);
1289
1290 if (unlikely(ip_mtu_locked(dst))) {
1291 if (rt->rt_uses_gateway && mtu > 576)
1292 mtu = 576;
1293 }
1294
1295 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1296
1297 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1298}
1299
1300static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1301{
1302 struct fnhe_hash_bucket *hash;
1303 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1304 u32 hval = fnhe_hashfun(daddr);
1305
1306 spin_lock_bh(&fnhe_lock);
1307
1308 hash = rcu_dereference_protected(nh->nh_exceptions,
1309 lockdep_is_held(&fnhe_lock));
1310 hash += hval;
1311
1312 fnhe_p = &hash->chain;
1313 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1314 while (fnhe) {
1315 if (fnhe->fnhe_daddr == daddr) {
1316 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1317 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1318 fnhe_flush_routes(fnhe);
1319 kfree_rcu(fnhe, rcu);
1320 break;
1321 }
1322 fnhe_p = &fnhe->fnhe_next;
1323 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1324 lockdep_is_held(&fnhe_lock));
1325 }
1326
1327 spin_unlock_bh(&fnhe_lock);
1328}
1329
1330static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1331{
1332 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1333 struct fib_nh_exception *fnhe;
1334 u32 hval;
1335
1336 if (!hash)
1337 return NULL;
1338
1339 hval = fnhe_hashfun(daddr);
1340
1341 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1342 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1343 if (fnhe->fnhe_daddr == daddr) {
1344 if (fnhe->fnhe_expires &&
1345 time_after(jiffies, fnhe->fnhe_expires)) {
1346 ip_del_fnhe(nh, daddr);
1347 break;
1348 }
1349 return fnhe;
1350 }
1351 }
1352 return NULL;
1353}
1354
1355static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1356 __be32 daddr, const bool do_cache)
1357{
1358 bool ret = false;
1359
1360 spin_lock_bh(&fnhe_lock);
1361
1362 if (daddr == fnhe->fnhe_daddr) {
1363 struct rtable __rcu **porig;
1364 struct rtable *orig;
1365 int genid = fnhe_genid(dev_net(rt->dst.dev));
1366
1367 if (rt_is_input_route(rt))
1368 porig = &fnhe->fnhe_rth_input;
1369 else
1370 porig = &fnhe->fnhe_rth_output;
1371 orig = rcu_dereference(*porig);
1372
1373 if (fnhe->fnhe_genid != genid) {
1374 fnhe->fnhe_genid = genid;
1375 fnhe->fnhe_gw = 0;
1376 fnhe->fnhe_pmtu = 0;
1377 fnhe->fnhe_expires = 0;
1378 fnhe->fnhe_mtu_locked = false;
1379 fnhe_flush_routes(fnhe);
1380 orig = NULL;
1381 }
1382 fill_route_from_fnhe(rt, fnhe);
1383 if (!rt->rt_gateway)
1384 rt->rt_gateway = daddr;
1385
1386 if (do_cache) {
1387 dst_hold(&rt->dst);
1388 rcu_assign_pointer(*porig, rt);
1389 if (orig) {
1390 dst_dev_put(&orig->dst);
1391 dst_release(&orig->dst);
1392 }
1393 ret = true;
1394 }
1395
1396 fnhe->fnhe_stamp = jiffies;
1397 }
1398 spin_unlock_bh(&fnhe_lock);
1399
1400 return ret;
1401}
1402
1403static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1404{
1405 struct rtable *orig, *prev, **p;
1406 bool ret = true;
1407
1408 if (rt_is_input_route(rt)) {
1409 p = (struct rtable **)&nh->nh_rth_input;
1410 } else {
1411 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1412 }
1413 orig = *p;
1414
1415 /* hold dst before doing cmpxchg() to avoid race condition
1416 * on this dst
1417 */
1418 dst_hold(&rt->dst);
1419 prev = cmpxchg(p, orig, rt);
1420 if (prev == orig) {
1421 if (orig) {
1422 dst_dev_put(&orig->dst);
1423 dst_release(&orig->dst);
1424 }
1425 } else {
1426 dst_release(&rt->dst);
1427 ret = false;
1428 }
1429
1430 return ret;
1431}
1432
1433struct uncached_list {
1434 spinlock_t lock;
1435 struct list_head head;
1436};
1437
1438static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1439
1440void rt_add_uncached_list(struct rtable *rt)
1441{
1442 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1443
1444 rt->rt_uncached_list = ul;
1445
1446 spin_lock_bh(&ul->lock);
1447 list_add_tail(&rt->rt_uncached, &ul->head);
1448 spin_unlock_bh(&ul->lock);
1449}
1450
1451void rt_del_uncached_list(struct rtable *rt)
1452{
1453 if (!list_empty(&rt->rt_uncached)) {
1454 struct uncached_list *ul = rt->rt_uncached_list;
1455
1456 spin_lock_bh(&ul->lock);
1457 list_del(&rt->rt_uncached);
1458 spin_unlock_bh(&ul->lock);
1459 }
1460}
1461
1462static void ipv4_dst_destroy(struct dst_entry *dst)
1463{
1464 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1465 struct rtable *rt = (struct rtable *)dst;
1466
1467 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1468 kfree(p);
1469
1470 rt_del_uncached_list(rt);
1471}
1472
1473void rt_flush_dev(struct net_device *dev)
1474{
1475 struct net *net = dev_net(dev);
1476 struct rtable *rt;
1477 int cpu;
1478
1479 for_each_possible_cpu(cpu) {
1480 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1481
1482 spin_lock_bh(&ul->lock);
1483 list_for_each_entry(rt, &ul->head, rt_uncached) {
1484 if (rt->dst.dev != dev)
1485 continue;
1486 rt->dst.dev = net->loopback_dev;
1487 dev_hold(rt->dst.dev);
1488 dev_put(dev);
1489 }
1490 spin_unlock_bh(&ul->lock);
1491 }
1492}
1493
1494static bool rt_cache_valid(const struct rtable *rt)
1495{
1496 return rt &&
1497 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1498 !rt_is_expired(rt);
1499}
1500
1501static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1502 const struct fib_result *res,
1503 struct fib_nh_exception *fnhe,
1504 struct fib_info *fi, u16 type, u32 itag,
1505 const bool do_cache)
1506{
1507 bool cached = false;
1508
1509 if (fi) {
1510 struct fib_nh *nh = &FIB_RES_NH(*res);
1511
1512 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1513 rt->rt_gateway = nh->nh_gw;
1514 rt->rt_uses_gateway = 1;
1515 }
1516 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1517 if (fi->fib_metrics != &dst_default_metrics) {
1518 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1519 refcount_inc(&fi->fib_metrics->refcnt);
1520 }
1521#ifdef CONFIG_IP_ROUTE_CLASSID
1522 rt->dst.tclassid = nh->nh_tclassid;
1523#endif
1524 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1525 if (unlikely(fnhe))
1526 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1527 else if (do_cache)
1528 cached = rt_cache_route(nh, rt);
1529 if (unlikely(!cached)) {
1530 /* Routes we intend to cache in nexthop exception or
1531 * FIB nexthop have the DST_NOCACHE bit clear.
1532 * However, if we are unsuccessful at storing this
1533 * route into the cache we really need to set it.
1534 */
1535 if (!rt->rt_gateway)
1536 rt->rt_gateway = daddr;
1537 rt_add_uncached_list(rt);
1538 }
1539 } else
1540 rt_add_uncached_list(rt);
1541
1542#ifdef CONFIG_IP_ROUTE_CLASSID
1543#ifdef CONFIG_IP_MULTIPLE_TABLES
1544 set_class_tag(rt, res->tclassid);
1545#endif
1546 set_class_tag(rt, itag);
1547#endif
1548}
1549
1550struct rtable *rt_dst_alloc(struct net_device *dev,
1551 unsigned int flags, u16 type,
1552 bool nopolicy, bool noxfrm, bool will_cache)
1553{
1554 struct rtable *rt;
1555
1556 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1557 (will_cache ? 0 : DST_HOST) |
1558 (nopolicy ? DST_NOPOLICY : 0) |
1559 (noxfrm ? DST_NOXFRM : 0));
1560
1561 if (rt) {
1562 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1563 rt->rt_flags = flags;
1564 rt->rt_type = type;
1565 rt->rt_is_input = 0;
1566 rt->rt_iif = 0;
1567 rt->rt_pmtu = 0;
1568 rt->rt_mtu_locked = 0;
1569 rt->rt_gateway = 0;
1570 rt->rt_uses_gateway = 0;
1571 INIT_LIST_HEAD(&rt->rt_uncached);
1572
1573 rt->dst.output = ip_output;
1574 if (flags & RTCF_LOCAL)
1575 rt->dst.input = ip_local_deliver;
1576 }
1577
1578 return rt;
1579}
1580EXPORT_SYMBOL(rt_dst_alloc);
1581
1582/* called in rcu_read_lock() section */
1583int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1584 u8 tos, struct net_device *dev,
1585 struct in_device *in_dev, u32 *itag)
1586{
1587 int err;
1588
1589 /* Primary sanity checks. */
1590 if (!in_dev)
1591 return -EINVAL;
1592
1593 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1594 skb->protocol != htons(ETH_P_IP))
1595 return -EINVAL;
1596
1597 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1598 return -EINVAL;
1599
1600 if (ipv4_is_zeronet(saddr)) {
1601 if (!ipv4_is_local_multicast(daddr))
1602 return -EINVAL;
1603 } else {
1604 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1605 in_dev, itag);
1606 if (err < 0)
1607 return err;
1608 }
1609 return 0;
1610}
1611
1612/* called in rcu_read_lock() section */
1613static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1614 u8 tos, struct net_device *dev, int our)
1615{
1616 struct in_device *in_dev = __in_dev_get_rcu(dev);
1617 unsigned int flags = RTCF_MULTICAST;
1618 struct rtable *rth;
1619 u32 itag = 0;
1620 int err;
1621
1622 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1623 if (err)
1624 return err;
1625
1626 if (our)
1627 flags |= RTCF_LOCAL;
1628
1629 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1630 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1631 if (!rth)
1632 return -ENOBUFS;
1633
1634#ifdef CONFIG_IP_ROUTE_CLASSID
1635 rth->dst.tclassid = itag;
1636#endif
1637 rth->dst.output = ip_rt_bug;
1638 rth->rt_is_input= 1;
1639
1640#ifdef CONFIG_IP_MROUTE
1641 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1642 rth->dst.input = ip_mr_input;
1643#endif
1644 RT_CACHE_STAT_INC(in_slow_mc);
1645
1646 skb_dst_set(skb, &rth->dst);
1647 return 0;
1648}
1649
1650
1651static void ip_handle_martian_source(struct net_device *dev,
1652 struct in_device *in_dev,
1653 struct sk_buff *skb,
1654 __be32 daddr,
1655 __be32 saddr)
1656{
1657 RT_CACHE_STAT_INC(in_martian_src);
1658#ifdef CONFIG_IP_ROUTE_VERBOSE
1659 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1660 /*
1661 * RFC1812 recommendation, if source is martian,
1662 * the only hint is MAC header.
1663 */
1664 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1665 &daddr, &saddr, dev->name);
1666 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1667 print_hex_dump(KERN_WARNING, "ll header: ",
1668 DUMP_PREFIX_OFFSET, 16, 1,
1669 skb_mac_header(skb),
1670 dev->hard_header_len, true);
1671 }
1672 }
1673#endif
1674}
1675
1676/* called in rcu_read_lock() section */
1677static int __mkroute_input(struct sk_buff *skb,
1678 const struct fib_result *res,
1679 struct in_device *in_dev,
1680 __be32 daddr, __be32 saddr, u32 tos)
1681{
1682 struct fib_nh_exception *fnhe;
1683 struct rtable *rth;
1684 int err;
1685 struct in_device *out_dev;
1686 bool do_cache;
1687 u32 itag = 0;
1688
1689 /* get a working reference to the output device */
1690 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1691 if (!out_dev) {
1692 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1693 return -EINVAL;
1694 }
1695
1696 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1697 in_dev->dev, in_dev, &itag);
1698 if (err < 0) {
1699 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1700 saddr);
1701
1702 goto cleanup;
1703 }
1704
1705 do_cache = res->fi && !itag;
1706 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1707 skb->protocol == htons(ETH_P_IP) &&
1708 (IN_DEV_SHARED_MEDIA(out_dev) ||
1709 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1710 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1711
1712 if (skb->protocol != htons(ETH_P_IP)) {
1713 /* Not IP (i.e. ARP). Do not create route, if it is
1714 * invalid for proxy arp. DNAT routes are always valid.
1715 *
1716 * Proxy arp feature have been extended to allow, ARP
1717 * replies back to the same interface, to support
1718 * Private VLAN switch technologies. See arp.c.
1719 */
1720 if (out_dev == in_dev &&
1721 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1722 err = -EINVAL;
1723 goto cleanup;
1724 }
1725 }
1726
1727 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1728 if (do_cache) {
1729 if (fnhe)
1730 rth = rcu_dereference(fnhe->fnhe_rth_input);
1731 else
1732 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1733 if (rt_cache_valid(rth)) {
1734 skb_dst_set_noref(skb, &rth->dst);
1735 goto out;
1736 }
1737 }
1738
1739 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1740 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1741 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1742 if (!rth) {
1743 err = -ENOBUFS;
1744 goto cleanup;
1745 }
1746
1747 rth->rt_is_input = 1;
1748 RT_CACHE_STAT_INC(in_slow_tot);
1749
1750 rth->dst.input = ip_forward;
1751
1752 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1753 do_cache);
1754 lwtunnel_set_redirect(&rth->dst);
1755 skb_dst_set(skb, &rth->dst);
1756out:
1757 err = 0;
1758 cleanup:
1759 return err;
1760}
1761
1762#ifdef CONFIG_IP_ROUTE_MULTIPATH
1763/* To make ICMP packets follow the right flow, the multipath hash is
1764 * calculated from the inner IP addresses.
1765 */
1766static void ip_multipath_l3_keys(const struct sk_buff *skb,
1767 struct flow_keys *hash_keys)
1768{
1769 const struct iphdr *outer_iph = ip_hdr(skb);
1770 const struct iphdr *key_iph = outer_iph;
1771 const struct iphdr *inner_iph;
1772 const struct icmphdr *icmph;
1773 struct iphdr _inner_iph;
1774 struct icmphdr _icmph;
1775
1776 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1777 goto out;
1778
1779 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1780 goto out;
1781
1782 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1783 &_icmph);
1784 if (!icmph)
1785 goto out;
1786
1787 if (icmph->type != ICMP_DEST_UNREACH &&
1788 icmph->type != ICMP_REDIRECT &&
1789 icmph->type != ICMP_TIME_EXCEEDED &&
1790 icmph->type != ICMP_PARAMETERPROB)
1791 goto out;
1792
1793 inner_iph = skb_header_pointer(skb,
1794 outer_iph->ihl * 4 + sizeof(_icmph),
1795 sizeof(_inner_iph), &_inner_iph);
1796 if (!inner_iph)
1797 goto out;
1798
1799 key_iph = inner_iph;
1800out:
1801 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1802 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1803}
1804
1805/* if skb is set it will be used and fl4 can be NULL */
1806int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1807 const struct sk_buff *skb, struct flow_keys *flkeys)
1808{
1809 struct flow_keys hash_keys;
1810 u32 mhash;
1811
1812 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1813 case 0:
1814 memset(&hash_keys, 0, sizeof(hash_keys));
1815 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1816 if (skb) {
1817 ip_multipath_l3_keys(skb, &hash_keys);
1818 } else {
1819 hash_keys.addrs.v4addrs.src = fl4->saddr;
1820 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1821 }
1822 break;
1823 case 1:
1824 /* skb is currently provided only when forwarding */
1825 if (skb) {
1826 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1827 struct flow_keys keys;
1828
1829 /* short-circuit if we already have L4 hash present */
1830 if (skb->l4_hash)
1831 return skb_get_hash_raw(skb) >> 1;
1832
1833 memset(&hash_keys, 0, sizeof(hash_keys));
1834
1835 if (!flkeys) {
1836 skb_flow_dissect_flow_keys(skb, &keys, flag);
1837 flkeys = &keys;
1838 }
1839
1840 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1841 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1842 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1843 hash_keys.ports.src = flkeys->ports.src;
1844 hash_keys.ports.dst = flkeys->ports.dst;
1845 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1846 } else {
1847 memset(&hash_keys, 0, sizeof(hash_keys));
1848 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1849 hash_keys.addrs.v4addrs.src = fl4->saddr;
1850 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1851 hash_keys.ports.src = fl4->fl4_sport;
1852 hash_keys.ports.dst = fl4->fl4_dport;
1853 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1854 }
1855 break;
1856 }
1857 mhash = flow_hash_from_keys(&hash_keys);
1858
1859 return mhash >> 1;
1860}
1861#endif /* CONFIG_IP_ROUTE_MULTIPATH */
1862
1863static int ip_mkroute_input(struct sk_buff *skb,
1864 struct fib_result *res,
1865 struct in_device *in_dev,
1866 __be32 daddr, __be32 saddr, u32 tos,
1867 struct flow_keys *hkeys)
1868{
1869#ifdef CONFIG_IP_ROUTE_MULTIPATH
1870 if (res->fi && res->fi->fib_nhs > 1) {
1871 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1872
1873 fib_select_multipath(res, h);
1874 }
1875#endif
1876
1877 /* create a routing cache entry */
1878 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1879}
1880
1881/*
1882 * NOTE. We drop all the packets that has local source
1883 * addresses, because every properly looped back packet
1884 * must have correct destination already attached by output routine.
1885 *
1886 * Such approach solves two big problems:
1887 * 1. Not simplex devices are handled properly.
1888 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1889 * called with rcu_read_lock()
1890 */
1891
1892static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1893 u8 tos, struct net_device *dev,
1894 struct fib_result *res)
1895{
1896 struct in_device *in_dev = __in_dev_get_rcu(dev);
1897 struct flow_keys *flkeys = NULL, _flkeys;
1898 struct net *net = dev_net(dev);
1899 struct ip_tunnel_info *tun_info;
1900 int err = -EINVAL;
1901 unsigned int flags = 0;
1902 u32 itag = 0;
1903 struct rtable *rth;
1904 struct flowi4 fl4;
1905 bool do_cache;
1906
1907 /* IP on this device is disabled. */
1908
1909 if (!in_dev)
1910 goto out;
1911
1912 /* Check for the most weird martians, which can be not detected
1913 by fib_lookup.
1914 */
1915
1916 tun_info = skb_tunnel_info(skb);
1917 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1918 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1919 else
1920 fl4.flowi4_tun_key.tun_id = 0;
1921 skb_dst_drop(skb);
1922
1923 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1924 goto martian_source;
1925
1926 res->fi = NULL;
1927 res->table = NULL;
1928 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1929 goto brd_input;
1930
1931 /* Accept zero addresses only to limited broadcast;
1932 * I even do not know to fix it or not. Waiting for complains :-)
1933 */
1934 if (ipv4_is_zeronet(saddr))
1935 goto martian_source;
1936
1937 if (ipv4_is_zeronet(daddr))
1938 goto martian_destination;
1939
1940 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1941 * and call it once if daddr or/and saddr are loopback addresses
1942 */
1943 if (ipv4_is_loopback(daddr)) {
1944 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1945 goto martian_destination;
1946 } else if (ipv4_is_loopback(saddr)) {
1947 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1948 goto martian_source;
1949 }
1950
1951 /*
1952 * Now we are ready to route packet.
1953 */
1954 fl4.flowi4_oif = 0;
1955 fl4.flowi4_iif = dev->ifindex;
1956 fl4.flowi4_mark = skb->mark;
1957 fl4.flowi4_tos = tos;
1958 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1959 fl4.flowi4_flags = 0;
1960 fl4.daddr = daddr;
1961 fl4.saddr = saddr;
1962 fl4.flowi4_uid = sock_net_uid(net, NULL);
1963
1964 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
1965 flkeys = &_flkeys;
1966 } else {
1967 fl4.flowi4_proto = 0;
1968 fl4.fl4_sport = 0;
1969 fl4.fl4_dport = 0;
1970 }
1971
1972 err = fib_lookup(net, &fl4, res, 0);
1973 if (err != 0) {
1974 if (!IN_DEV_FORWARD(in_dev))
1975 err = -EHOSTUNREACH;
1976 goto no_route;
1977 }
1978
1979 if (res->type == RTN_BROADCAST)
1980 goto brd_input;
1981
1982 if (res->type == RTN_LOCAL) {
1983 err = fib_validate_source(skb, saddr, daddr, tos,
1984 0, dev, in_dev, &itag);
1985 if (err < 0)
1986 goto martian_source;
1987 goto local_input;
1988 }
1989
1990 if (!IN_DEV_FORWARD(in_dev)) {
1991 err = -EHOSTUNREACH;
1992 goto no_route;
1993 }
1994 if (res->type != RTN_UNICAST)
1995 goto martian_destination;
1996
1997 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
1998out: return err;
1999
2000brd_input:
2001 if (skb->protocol != htons(ETH_P_IP))
2002 goto e_inval;
2003
2004 if (!ipv4_is_zeronet(saddr)) {
2005 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2006 in_dev, &itag);
2007 if (err < 0)
2008 goto martian_source;
2009 }
2010 flags |= RTCF_BROADCAST;
2011 res->type = RTN_BROADCAST;
2012 RT_CACHE_STAT_INC(in_brd);
2013
2014local_input:
2015 do_cache = false;
2016 if (res->fi) {
2017 if (!itag) {
2018 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2019 if (rt_cache_valid(rth)) {
2020 skb_dst_set_noref(skb, &rth->dst);
2021 err = 0;
2022 goto out;
2023 }
2024 do_cache = true;
2025 }
2026 }
2027
2028 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2029 flags | RTCF_LOCAL, res->type,
2030 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2031 if (!rth)
2032 goto e_nobufs;
2033
2034 rth->dst.output= ip_rt_bug;
2035#ifdef CONFIG_IP_ROUTE_CLASSID
2036 rth->dst.tclassid = itag;
2037#endif
2038 rth->rt_is_input = 1;
2039
2040 RT_CACHE_STAT_INC(in_slow_tot);
2041 if (res->type == RTN_UNREACHABLE) {
2042 rth->dst.input= ip_error;
2043 rth->dst.error= -err;
2044 rth->rt_flags &= ~RTCF_LOCAL;
2045 }
2046
2047 if (do_cache) {
2048 struct fib_nh *nh = &FIB_RES_NH(*res);
2049
2050 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2051 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2052 WARN_ON(rth->dst.input == lwtunnel_input);
2053 rth->dst.lwtstate->orig_input = rth->dst.input;
2054 rth->dst.input = lwtunnel_input;
2055 }
2056
2057 if (unlikely(!rt_cache_route(nh, rth)))
2058 rt_add_uncached_list(rth);
2059 }
2060 skb_dst_set(skb, &rth->dst);
2061 err = 0;
2062 goto out;
2063
2064no_route:
2065 RT_CACHE_STAT_INC(in_no_route);
2066 res->type = RTN_UNREACHABLE;
2067 res->fi = NULL;
2068 res->table = NULL;
2069 goto local_input;
2070
2071 /*
2072 * Do not cache martian addresses: they should be logged (RFC1812)
2073 */
2074martian_destination:
2075 RT_CACHE_STAT_INC(in_martian_dst);
2076#ifdef CONFIG_IP_ROUTE_VERBOSE
2077 if (IN_DEV_LOG_MARTIANS(in_dev))
2078 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2079 &daddr, &saddr, dev->name);
2080#endif
2081
2082e_inval:
2083 err = -EINVAL;
2084 goto out;
2085
2086e_nobufs:
2087 err = -ENOBUFS;
2088 goto out;
2089
2090martian_source:
2091 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2092 goto out;
2093}
2094
2095int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2096 u8 tos, struct net_device *dev)
2097{
2098 struct fib_result res;
2099 int err;
2100
2101 tos &= IPTOS_RT_MASK;
2102 rcu_read_lock();
2103 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2104 rcu_read_unlock();
2105
2106 return err;
2107}
2108EXPORT_SYMBOL(ip_route_input_noref);
2109
2110/* called with rcu_read_lock held */
2111int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2112 u8 tos, struct net_device *dev, struct fib_result *res)
2113{
2114 /* Multicast recognition logic is moved from route cache to here.
2115 The problem was that too many Ethernet cards have broken/missing
2116 hardware multicast filters :-( As result the host on multicasting
2117 network acquires a lot of useless route cache entries, sort of
2118 SDR messages from all the world. Now we try to get rid of them.
2119 Really, provided software IP multicast filter is organized
2120 reasonably (at least, hashed), it does not result in a slowdown
2121 comparing with route cache reject entries.
2122 Note, that multicast routers are not affected, because
2123 route cache entry is created eventually.
2124 */
2125 if (ipv4_is_multicast(daddr)) {
2126 struct in_device *in_dev = __in_dev_get_rcu(dev);
2127 int our = 0;
2128 int err = -EINVAL;
2129
2130 if (in_dev)
2131 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2132 ip_hdr(skb)->protocol);
2133
2134 /* check l3 master if no match yet */
2135 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2136 struct in_device *l3_in_dev;
2137
2138 l3_in_dev = __in_dev_get_rcu(skb->dev);
2139 if (l3_in_dev)
2140 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2141 ip_hdr(skb)->protocol);
2142 }
2143
2144 if (our
2145#ifdef CONFIG_IP_MROUTE
2146 ||
2147 (!ipv4_is_local_multicast(daddr) &&
2148 IN_DEV_MFORWARD(in_dev))
2149#endif
2150 ) {
2151 err = ip_route_input_mc(skb, daddr, saddr,
2152 tos, dev, our);
2153 }
2154 return err;
2155 }
2156
2157 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2158}
2159
2160/* called with rcu_read_lock() */
2161static struct rtable *__mkroute_output(const struct fib_result *res,
2162 const struct flowi4 *fl4, int orig_oif,
2163 struct net_device *dev_out,
2164 unsigned int flags)
2165{
2166 struct fib_info *fi = res->fi;
2167 struct fib_nh_exception *fnhe;
2168 struct in_device *in_dev;
2169 u16 type = res->type;
2170 struct rtable *rth;
2171 bool do_cache;
2172
2173 in_dev = __in_dev_get_rcu(dev_out);
2174 if (!in_dev)
2175 return ERR_PTR(-EINVAL);
2176
2177 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2178 if (ipv4_is_loopback(fl4->saddr) &&
2179 !(dev_out->flags & IFF_LOOPBACK) &&
2180 !netif_is_l3_master(dev_out))
2181 return ERR_PTR(-EINVAL);
2182
2183 if (ipv4_is_lbcast(fl4->daddr))
2184 type = RTN_BROADCAST;
2185 else if (ipv4_is_multicast(fl4->daddr))
2186 type = RTN_MULTICAST;
2187 else if (ipv4_is_zeronet(fl4->daddr))
2188 return ERR_PTR(-EINVAL);
2189
2190 if (dev_out->flags & IFF_LOOPBACK)
2191 flags |= RTCF_LOCAL;
2192
2193 do_cache = true;
2194 if (type == RTN_BROADCAST) {
2195 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2196 fi = NULL;
2197 } else if (type == RTN_MULTICAST) {
2198 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2199 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2200 fl4->flowi4_proto))
2201 flags &= ~RTCF_LOCAL;
2202 else
2203 do_cache = false;
2204 /* If multicast route do not exist use
2205 * default one, but do not gateway in this case.
2206 * Yes, it is hack.
2207 */
2208 if (fi && res->prefixlen < 4)
2209 fi = NULL;
2210 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2211 (orig_oif != dev_out->ifindex)) {
2212 /* For local routes that require a particular output interface
2213 * we do not want to cache the result. Caching the result
2214 * causes incorrect behaviour when there are multiple source
2215 * addresses on the interface, the end result being that if the
2216 * intended recipient is waiting on that interface for the
2217 * packet he won't receive it because it will be delivered on
2218 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2219 * be set to the loopback interface as well.
2220 */
2221 do_cache = false;
2222 }
2223
2224 fnhe = NULL;
2225 do_cache &= fi != NULL;
2226 if (fi) {
2227 struct rtable __rcu **prth;
2228 struct fib_nh *nh = &FIB_RES_NH(*res);
2229
2230 fnhe = find_exception(nh, fl4->daddr);
2231 if (!do_cache)
2232 goto add;
2233 if (fnhe) {
2234 prth = &fnhe->fnhe_rth_output;
2235 } else {
2236 if (unlikely(fl4->flowi4_flags &
2237 FLOWI_FLAG_KNOWN_NH &&
2238 !(nh->nh_gw &&
2239 nh->nh_scope == RT_SCOPE_LINK))) {
2240 do_cache = false;
2241 goto add;
2242 }
2243 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2244 }
2245 rth = rcu_dereference(*prth);
2246 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2247 return rth;
2248 }
2249
2250add:
2251 rth = rt_dst_alloc(dev_out, flags, type,
2252 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2253 IN_DEV_CONF_GET(in_dev, NOXFRM),
2254 do_cache);
2255 if (!rth)
2256 return ERR_PTR(-ENOBUFS);
2257
2258 rth->rt_iif = orig_oif;
2259
2260 RT_CACHE_STAT_INC(out_slow_tot);
2261
2262 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2263 if (flags & RTCF_LOCAL &&
2264 !(dev_out->flags & IFF_LOOPBACK)) {
2265 rth->dst.output = ip_mc_output;
2266 RT_CACHE_STAT_INC(out_slow_mc);
2267 }
2268#ifdef CONFIG_IP_MROUTE
2269 if (type == RTN_MULTICAST) {
2270 if (IN_DEV_MFORWARD(in_dev) &&
2271 !ipv4_is_local_multicast(fl4->daddr)) {
2272 rth->dst.input = ip_mr_input;
2273 rth->dst.output = ip_mc_output;
2274 }
2275 }
2276#endif
2277 }
2278
2279 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2280 lwtunnel_set_redirect(&rth->dst);
2281
2282 return rth;
2283}
2284
2285/*
2286 * Major route resolver routine.
2287 */
2288
2289struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2290 const struct sk_buff *skb)
2291{
2292 __u8 tos = RT_FL_TOS(fl4);
2293 struct fib_result res = {
2294 .type = RTN_UNSPEC,
2295 .fi = NULL,
2296 .table = NULL,
2297 .tclassid = 0,
2298 };
2299 struct rtable *rth;
2300
2301 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2302 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2303 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2304 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2305
2306 rcu_read_lock();
2307 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2308 rcu_read_unlock();
2309
2310 return rth;
2311}
2312EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2313
2314struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2315 struct fib_result *res,
2316 const struct sk_buff *skb)
2317{
2318 struct net_device *dev_out = NULL;
2319 int orig_oif = fl4->flowi4_oif;
2320 unsigned int flags = 0;
2321 struct rtable *rth;
2322 int err = -ENETUNREACH;
2323
2324 if (fl4->saddr) {
2325 rth = ERR_PTR(-EINVAL);
2326 if (ipv4_is_multicast(fl4->saddr) ||
2327 ipv4_is_lbcast(fl4->saddr) ||
2328 ipv4_is_zeronet(fl4->saddr))
2329 goto out;
2330
2331 /* I removed check for oif == dev_out->oif here.
2332 It was wrong for two reasons:
2333 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2334 is assigned to multiple interfaces.
2335 2. Moreover, we are allowed to send packets with saddr
2336 of another iface. --ANK
2337 */
2338
2339 if (fl4->flowi4_oif == 0 &&
2340 (ipv4_is_multicast(fl4->daddr) ||
2341 ipv4_is_lbcast(fl4->daddr))) {
2342 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2343 dev_out = __ip_dev_find(net, fl4->saddr, false);
2344 if (!dev_out)
2345 goto out;
2346
2347 /* Special hack: user can direct multicasts
2348 and limited broadcast via necessary interface
2349 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2350 This hack is not just for fun, it allows
2351 vic,vat and friends to work.
2352 They bind socket to loopback, set ttl to zero
2353 and expect that it will work.
2354 From the viewpoint of routing cache they are broken,
2355 because we are not allowed to build multicast path
2356 with loopback source addr (look, routing cache
2357 cannot know, that ttl is zero, so that packet
2358 will not leave this host and route is valid).
2359 Luckily, this hack is good workaround.
2360 */
2361
2362 fl4->flowi4_oif = dev_out->ifindex;
2363 goto make_route;
2364 }
2365
2366 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2367 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2368 if (!__ip_dev_find(net, fl4->saddr, false))
2369 goto out;
2370 }
2371 }
2372
2373
2374 if (fl4->flowi4_oif) {
2375 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2376 rth = ERR_PTR(-ENODEV);
2377 if (!dev_out)
2378 goto out;
2379
2380 /* RACE: Check return value of inet_select_addr instead. */
2381 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2382 rth = ERR_PTR(-ENETUNREACH);
2383 goto out;
2384 }
2385 if (ipv4_is_local_multicast(fl4->daddr) ||
2386 ipv4_is_lbcast(fl4->daddr) ||
2387 fl4->flowi4_proto == IPPROTO_IGMP) {
2388 if (!fl4->saddr)
2389 fl4->saddr = inet_select_addr(dev_out, 0,
2390 RT_SCOPE_LINK);
2391 goto make_route;
2392 }
2393 if (!fl4->saddr) {
2394 if (ipv4_is_multicast(fl4->daddr))
2395 fl4->saddr = inet_select_addr(dev_out, 0,
2396 fl4->flowi4_scope);
2397 else if (!fl4->daddr)
2398 fl4->saddr = inet_select_addr(dev_out, 0,
2399 RT_SCOPE_HOST);
2400 }
2401 }
2402
2403 if (!fl4->daddr) {
2404 fl4->daddr = fl4->saddr;
2405 if (!fl4->daddr)
2406 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2407 dev_out = net->loopback_dev;
2408 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2409 res->type = RTN_LOCAL;
2410 flags |= RTCF_LOCAL;
2411 goto make_route;
2412 }
2413
2414 err = fib_lookup(net, fl4, res, 0);
2415 if (err) {
2416 res->fi = NULL;
2417 res->table = NULL;
2418 if (fl4->flowi4_oif &&
2419 (ipv4_is_multicast(fl4->daddr) ||
2420 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2421 /* Apparently, routing tables are wrong. Assume,
2422 that the destination is on link.
2423
2424 WHY? DW.
2425 Because we are allowed to send to iface
2426 even if it has NO routes and NO assigned
2427 addresses. When oif is specified, routing
2428 tables are looked up with only one purpose:
2429 to catch if destination is gatewayed, rather than
2430 direct. Moreover, if MSG_DONTROUTE is set,
2431 we send packet, ignoring both routing tables
2432 and ifaddr state. --ANK
2433
2434
2435 We could make it even if oif is unknown,
2436 likely IPv6, but we do not.
2437 */
2438
2439 if (fl4->saddr == 0)
2440 fl4->saddr = inet_select_addr(dev_out, 0,
2441 RT_SCOPE_LINK);
2442 res->type = RTN_UNICAST;
2443 goto make_route;
2444 }
2445 rth = ERR_PTR(err);
2446 goto out;
2447 }
2448
2449 if (res->type == RTN_LOCAL) {
2450 if (!fl4->saddr) {
2451 if (res->fi->fib_prefsrc)
2452 fl4->saddr = res->fi->fib_prefsrc;
2453 else
2454 fl4->saddr = fl4->daddr;
2455 }
2456
2457 /* L3 master device is the loopback for that domain */
2458 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2459 net->loopback_dev;
2460
2461 /* make sure orig_oif points to fib result device even
2462 * though packet rx/tx happens over loopback or l3mdev
2463 */
2464 orig_oif = FIB_RES_OIF(*res);
2465
2466 fl4->flowi4_oif = dev_out->ifindex;
2467 flags |= RTCF_LOCAL;
2468 goto make_route;
2469 }
2470
2471 fib_select_path(net, res, fl4, skb);
2472
2473 dev_out = FIB_RES_DEV(*res);
2474 fl4->flowi4_oif = dev_out->ifindex;
2475
2476
2477make_route:
2478 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2479
2480out:
2481 return rth;
2482}
2483
2484static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2485{
2486 return NULL;
2487}
2488
2489static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2490{
2491 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2492
2493 return mtu ? : dst->dev->mtu;
2494}
2495
2496static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2497 struct sk_buff *skb, u32 mtu)
2498{
2499}
2500
2501static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2502 struct sk_buff *skb)
2503{
2504}
2505
2506static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2507 unsigned long old)
2508{
2509 return NULL;
2510}
2511
2512static struct dst_ops ipv4_dst_blackhole_ops = {
2513 .family = AF_INET,
2514 .check = ipv4_blackhole_dst_check,
2515 .mtu = ipv4_blackhole_mtu,
2516 .default_advmss = ipv4_default_advmss,
2517 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2518 .redirect = ipv4_rt_blackhole_redirect,
2519 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2520 .neigh_lookup = ipv4_neigh_lookup,
2521};
2522
2523struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2524{
2525 struct rtable *ort = (struct rtable *) dst_orig;
2526 struct rtable *rt;
2527
2528 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2529 if (rt) {
2530 struct dst_entry *new = &rt->dst;
2531
2532 new->__use = 1;
2533 new->input = dst_discard;
2534 new->output = dst_discard_out;
2535
2536 new->dev = net->loopback_dev;
2537 if (new->dev)
2538 dev_hold(new->dev);
2539
2540 rt->rt_is_input = ort->rt_is_input;
2541 rt->rt_iif = ort->rt_iif;
2542 rt->rt_pmtu = ort->rt_pmtu;
2543 rt->rt_mtu_locked = ort->rt_mtu_locked;
2544
2545 rt->rt_genid = rt_genid_ipv4(net);
2546 rt->rt_flags = ort->rt_flags;
2547 rt->rt_type = ort->rt_type;
2548 rt->rt_gateway = ort->rt_gateway;
2549 rt->rt_uses_gateway = ort->rt_uses_gateway;
2550
2551 INIT_LIST_HEAD(&rt->rt_uncached);
2552 }
2553
2554 dst_release(dst_orig);
2555
2556 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2557}
2558
2559struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2560 const struct sock *sk)
2561{
2562 struct rtable *rt = __ip_route_output_key(net, flp4);
2563
2564 if (IS_ERR(rt))
2565 return rt;
2566
2567 if (flp4->flowi4_proto)
2568 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2569 flowi4_to_flowi(flp4),
2570 sk, 0);
2571
2572 return rt;
2573}
2574EXPORT_SYMBOL_GPL(ip_route_output_flow);
2575
2576/* called with rcu_read_lock held */
2577static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2578 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2579 u32 seq)
2580{
2581 struct rtable *rt = skb_rtable(skb);
2582 struct rtmsg *r;
2583 struct nlmsghdr *nlh;
2584 unsigned long expires = 0;
2585 u32 error;
2586 u32 metrics[RTAX_MAX];
2587
2588 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2589 if (!nlh)
2590 return -EMSGSIZE;
2591
2592 r = nlmsg_data(nlh);
2593 r->rtm_family = AF_INET;
2594 r->rtm_dst_len = 32;
2595 r->rtm_src_len = 0;
2596 r->rtm_tos = fl4->flowi4_tos;
2597 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2598 if (nla_put_u32(skb, RTA_TABLE, table_id))
2599 goto nla_put_failure;
2600 r->rtm_type = rt->rt_type;
2601 r->rtm_scope = RT_SCOPE_UNIVERSE;
2602 r->rtm_protocol = RTPROT_UNSPEC;
2603 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2604 if (rt->rt_flags & RTCF_NOTIFY)
2605 r->rtm_flags |= RTM_F_NOTIFY;
2606 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2607 r->rtm_flags |= RTCF_DOREDIRECT;
2608
2609 if (nla_put_in_addr(skb, RTA_DST, dst))
2610 goto nla_put_failure;
2611 if (src) {
2612 r->rtm_src_len = 32;
2613 if (nla_put_in_addr(skb, RTA_SRC, src))
2614 goto nla_put_failure;
2615 }
2616 if (rt->dst.dev &&
2617 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2618 goto nla_put_failure;
2619#ifdef CONFIG_IP_ROUTE_CLASSID
2620 if (rt->dst.tclassid &&
2621 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2622 goto nla_put_failure;
2623#endif
2624 if (!rt_is_input_route(rt) &&
2625 fl4->saddr != src) {
2626 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2627 goto nla_put_failure;
2628 }
2629 if (rt->rt_uses_gateway &&
2630 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2631 goto nla_put_failure;
2632
2633 expires = rt->dst.expires;
2634 if (expires) {
2635 unsigned long now = jiffies;
2636
2637 if (time_before(now, expires))
2638 expires -= now;
2639 else
2640 expires = 0;
2641 }
2642
2643 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2644 if (rt->rt_pmtu && expires)
2645 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2646 if (rt->rt_mtu_locked && expires)
2647 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2648 if (rtnetlink_put_metrics(skb, metrics) < 0)
2649 goto nla_put_failure;
2650
2651 if (fl4->flowi4_mark &&
2652 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2653 goto nla_put_failure;
2654
2655 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2656 nla_put_u32(skb, RTA_UID,
2657 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2658 goto nla_put_failure;
2659
2660 error = rt->dst.error;
2661
2662 if (rt_is_input_route(rt)) {
2663#ifdef CONFIG_IP_MROUTE
2664 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2665 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2666 int err = ipmr_get_route(net, skb,
2667 fl4->saddr, fl4->daddr,
2668 r, portid);
2669
2670 if (err <= 0) {
2671 if (err == 0)
2672 return 0;
2673 goto nla_put_failure;
2674 }
2675 } else
2676#endif
2677 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2678 goto nla_put_failure;
2679 }
2680
2681 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2682 goto nla_put_failure;
2683
2684 nlmsg_end(skb, nlh);
2685 return 0;
2686
2687nla_put_failure:
2688 nlmsg_cancel(skb, nlh);
2689 return -EMSGSIZE;
2690}
2691
2692static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2693 struct netlink_ext_ack *extack)
2694{
2695 struct net *net = sock_net(in_skb->sk);
2696 struct rtmsg *rtm;
2697 struct nlattr *tb[RTA_MAX+1];
2698 struct fib_result res = {};
2699 struct rtable *rt = NULL;
2700 struct flowi4 fl4;
2701 __be32 dst = 0;
2702 __be32 src = 0;
2703 u32 iif;
2704 int err;
2705 int mark;
2706 struct sk_buff *skb;
2707 u32 table_id = RT_TABLE_MAIN;
2708 kuid_t uid;
2709
2710 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2711 extack);
2712 if (err < 0)
2713 goto errout;
2714
2715 rtm = nlmsg_data(nlh);
2716
2717 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2718 if (!skb) {
2719 err = -ENOBUFS;
2720 goto errout;
2721 }
2722
2723 /* Reserve room for dummy headers, this skb can pass
2724 through good chunk of routing engine.
2725 */
2726 skb_reset_mac_header(skb);
2727 skb_reset_network_header(skb);
2728
2729 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2730 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2731 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2732 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2733 if (tb[RTA_UID])
2734 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2735 else
2736 uid = (iif ? INVALID_UID : current_uid());
2737
2738 /* Bugfix: need to give ip_route_input enough of an IP header to
2739 * not gag.
2740 */
2741 ip_hdr(skb)->protocol = IPPROTO_UDP;
2742 ip_hdr(skb)->saddr = src;
2743 ip_hdr(skb)->daddr = dst;
2744
2745 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2746
2747 memset(&fl4, 0, sizeof(fl4));
2748 fl4.daddr = dst;
2749 fl4.saddr = src;
2750 fl4.flowi4_tos = rtm->rtm_tos;
2751 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2752 fl4.flowi4_mark = mark;
2753 fl4.flowi4_uid = uid;
2754
2755 rcu_read_lock();
2756
2757 if (iif) {
2758 struct net_device *dev;
2759
2760 dev = dev_get_by_index_rcu(net, iif);
2761 if (!dev) {
2762 err = -ENODEV;
2763 goto errout_free;
2764 }
2765
2766 skb->protocol = htons(ETH_P_IP);
2767 skb->dev = dev;
2768 skb->mark = mark;
2769 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2770 dev, &res);
2771
2772 rt = skb_rtable(skb);
2773 if (err == 0 && rt->dst.error)
2774 err = -rt->dst.error;
2775 } else {
2776 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2777 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2778 err = 0;
2779 if (IS_ERR(rt))
2780 err = PTR_ERR(rt);
2781 else
2782 skb_dst_set(skb, &rt->dst);
2783 }
2784
2785 if (err)
2786 goto errout_free;
2787
2788 if (rtm->rtm_flags & RTM_F_NOTIFY)
2789 rt->rt_flags |= RTCF_NOTIFY;
2790
2791 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2792 table_id = res.table ? res.table->tb_id : 0;
2793
2794 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2795 if (!res.fi) {
2796 err = fib_props[res.type].error;
2797 if (!err)
2798 err = -EHOSTUNREACH;
2799 goto errout_free;
2800 }
2801 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2802 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2803 rt->rt_type, res.prefix, res.prefixlen,
2804 fl4.flowi4_tos, res.fi, 0);
2805 } else {
2806 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2807 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2808 }
2809 if (err < 0)
2810 goto errout_free;
2811
2812 rcu_read_unlock();
2813
2814 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2815errout:
2816 return err;
2817
2818errout_free:
2819 rcu_read_unlock();
2820 kfree_skb(skb);
2821 goto errout;
2822}
2823
2824void ip_rt_multicast_event(struct in_device *in_dev)
2825{
2826 rt_cache_flush(dev_net(in_dev->dev));
2827}
2828
2829#ifdef CONFIG_SYSCTL
2830static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2831static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2832static int ip_rt_gc_elasticity __read_mostly = 8;
2833static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
2834
2835static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2836 void __user *buffer,
2837 size_t *lenp, loff_t *ppos)
2838{
2839 struct net *net = (struct net *)__ctl->extra1;
2840
2841 if (write) {
2842 rt_cache_flush(net);
2843 fnhe_genid_bump(net);
2844 return 0;
2845 }
2846
2847 return -EINVAL;
2848}
2849
2850static struct ctl_table ipv4_route_table[] = {
2851 {
2852 .procname = "gc_thresh",
2853 .data = &ipv4_dst_ops.gc_thresh,
2854 .maxlen = sizeof(int),
2855 .mode = 0644,
2856 .proc_handler = proc_dointvec,
2857 },
2858 {
2859 .procname = "max_size",
2860 .data = &ip_rt_max_size,
2861 .maxlen = sizeof(int),
2862 .mode = 0644,
2863 .proc_handler = proc_dointvec,
2864 },
2865 {
2866 /* Deprecated. Use gc_min_interval_ms */
2867
2868 .procname = "gc_min_interval",
2869 .data = &ip_rt_gc_min_interval,
2870 .maxlen = sizeof(int),
2871 .mode = 0644,
2872 .proc_handler = proc_dointvec_jiffies,
2873 },
2874 {
2875 .procname = "gc_min_interval_ms",
2876 .data = &ip_rt_gc_min_interval,
2877 .maxlen = sizeof(int),
2878 .mode = 0644,
2879 .proc_handler = proc_dointvec_ms_jiffies,
2880 },
2881 {
2882 .procname = "gc_timeout",
2883 .data = &ip_rt_gc_timeout,
2884 .maxlen = sizeof(int),
2885 .mode = 0644,
2886 .proc_handler = proc_dointvec_jiffies,
2887 },
2888 {
2889 .procname = "gc_interval",
2890 .data = &ip_rt_gc_interval,
2891 .maxlen = sizeof(int),
2892 .mode = 0644,
2893 .proc_handler = proc_dointvec_jiffies,
2894 },
2895 {
2896 .procname = "redirect_load",
2897 .data = &ip_rt_redirect_load,
2898 .maxlen = sizeof(int),
2899 .mode = 0644,
2900 .proc_handler = proc_dointvec,
2901 },
2902 {
2903 .procname = "redirect_number",
2904 .data = &ip_rt_redirect_number,
2905 .maxlen = sizeof(int),
2906 .mode = 0644,
2907 .proc_handler = proc_dointvec,
2908 },
2909 {
2910 .procname = "redirect_silence",
2911 .data = &ip_rt_redirect_silence,
2912 .maxlen = sizeof(int),
2913 .mode = 0644,
2914 .proc_handler = proc_dointvec,
2915 },
2916 {
2917 .procname = "error_cost",
2918 .data = &ip_rt_error_cost,
2919 .maxlen = sizeof(int),
2920 .mode = 0644,
2921 .proc_handler = proc_dointvec,
2922 },
2923 {
2924 .procname = "error_burst",
2925 .data = &ip_rt_error_burst,
2926 .maxlen = sizeof(int),
2927 .mode = 0644,
2928 .proc_handler = proc_dointvec,
2929 },
2930 {
2931 .procname = "gc_elasticity",
2932 .data = &ip_rt_gc_elasticity,
2933 .maxlen = sizeof(int),
2934 .mode = 0644,
2935 .proc_handler = proc_dointvec,
2936 },
2937 {
2938 .procname = "mtu_expires",
2939 .data = &ip_rt_mtu_expires,
2940 .maxlen = sizeof(int),
2941 .mode = 0644,
2942 .proc_handler = proc_dointvec_jiffies,
2943 },
2944 {
2945 .procname = "min_pmtu",
2946 .data = &ip_rt_min_pmtu,
2947 .maxlen = sizeof(int),
2948 .mode = 0644,
2949 .proc_handler = proc_dointvec_minmax,
2950 .extra1 = &ip_min_valid_pmtu,
2951 },
2952 {
2953 .procname = "min_adv_mss",
2954 .data = &ip_rt_min_advmss,
2955 .maxlen = sizeof(int),
2956 .mode = 0644,
2957 .proc_handler = proc_dointvec,
2958 },
2959 { }
2960};
2961
2962static struct ctl_table ipv4_route_flush_table[] = {
2963 {
2964 .procname = "flush",
2965 .maxlen = sizeof(int),
2966 .mode = 0200,
2967 .proc_handler = ipv4_sysctl_rtcache_flush,
2968 },
2969 { },
2970};
2971
2972static __net_init int sysctl_route_net_init(struct net *net)
2973{
2974 struct ctl_table *tbl;
2975
2976 tbl = ipv4_route_flush_table;
2977 if (!net_eq(net, &init_net)) {
2978 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2979 if (!tbl)
2980 goto err_dup;
2981
2982 /* Don't export sysctls to unprivileged users */
2983 if (net->user_ns != &init_user_ns)
2984 tbl[0].procname = NULL;
2985 }
2986 tbl[0].extra1 = net;
2987
2988 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2989 if (!net->ipv4.route_hdr)
2990 goto err_reg;
2991 return 0;
2992
2993err_reg:
2994 if (tbl != ipv4_route_flush_table)
2995 kfree(tbl);
2996err_dup:
2997 return -ENOMEM;
2998}
2999
3000static __net_exit void sysctl_route_net_exit(struct net *net)
3001{
3002 struct ctl_table *tbl;
3003
3004 tbl = net->ipv4.route_hdr->ctl_table_arg;
3005 unregister_net_sysctl_table(net->ipv4.route_hdr);
3006 BUG_ON(tbl == ipv4_route_flush_table);
3007 kfree(tbl);
3008}
3009
3010static __net_initdata struct pernet_operations sysctl_route_ops = {
3011 .init = sysctl_route_net_init,
3012 .exit = sysctl_route_net_exit,
3013};
3014#endif
3015
3016static __net_init int rt_genid_init(struct net *net)
3017{
3018 atomic_set(&net->ipv4.rt_genid, 0);
3019 atomic_set(&net->fnhe_genid, 0);
3020 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3021 return 0;
3022}
3023
3024static __net_initdata struct pernet_operations rt_genid_ops = {
3025 .init = rt_genid_init,
3026};
3027
3028static int __net_init ipv4_inetpeer_init(struct net *net)
3029{
3030 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3031
3032 if (!bp)
3033 return -ENOMEM;
3034 inet_peer_base_init(bp);
3035 net->ipv4.peers = bp;
3036 return 0;
3037}
3038
3039static void __net_exit ipv4_inetpeer_exit(struct net *net)
3040{
3041 struct inet_peer_base *bp = net->ipv4.peers;
3042
3043 net->ipv4.peers = NULL;
3044 inetpeer_invalidate_tree(bp);
3045 kfree(bp);
3046}
3047
3048static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3049 .init = ipv4_inetpeer_init,
3050 .exit = ipv4_inetpeer_exit,
3051};
3052
3053#ifdef CONFIG_IP_ROUTE_CLASSID
3054struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3055#endif /* CONFIG_IP_ROUTE_CLASSID */
3056
3057int __init ip_rt_init(void)
3058{
3059 int cpu;
3060
3061 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
3062 if (!ip_idents)
3063 panic("IP: failed to allocate ip_idents\n");
3064
3065 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3066
3067 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3068 if (!ip_tstamps)
3069 panic("IP: failed to allocate ip_tstamps\n");
3070
3071 for_each_possible_cpu(cpu) {
3072 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3073
3074 INIT_LIST_HEAD(&ul->head);
3075 spin_lock_init(&ul->lock);
3076 }
3077#ifdef CONFIG_IP_ROUTE_CLASSID
3078 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3079 if (!ip_rt_acct)
3080 panic("IP: failed to allocate ip_rt_acct\n");
3081#endif
3082
3083 ipv4_dst_ops.kmem_cachep =
3084 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3085 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3086
3087 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3088
3089 if (dst_entries_init(&ipv4_dst_ops) < 0)
3090 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3091
3092 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3093 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3094
3095 ipv4_dst_ops.gc_thresh = ~0;
3096 ip_rt_max_size = INT_MAX;
3097
3098 devinet_init();
3099 ip_fib_init();
3100
3101 if (ip_rt_proc_init())
3102 pr_err("Unable to create route proc files\n");
3103#ifdef CONFIG_XFRM
3104 xfrm_init();
3105 xfrm4_init();
3106#endif
3107 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3108 RTNL_FLAG_DOIT_UNLOCKED);
3109
3110#ifdef CONFIG_SYSCTL
3111 register_pernet_subsys(&sysctl_route_ops);
3112#endif
3113 register_pernet_subsys(&rt_genid_ops);
3114 register_pernet_subsys(&ipv4_inetpeer_ops);
3115 return 0;
3116}
3117
3118#ifdef CONFIG_SYSCTL
3119/*
3120 * We really need to sanitize the damn ipv4 init order, then all
3121 * this nonsense will go away.
3122 */
3123void __init ip_static_sysctl_init(void)
3124{
3125 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3126}
3127#endif
1/*
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.
5 *
6 * ROUTE - implementation of the IP router.
7 *
8 * Authors: Ross Biro
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>
13 *
14 * Fixes:
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
24 * clamper.
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.
39 *
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
58 *
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.
63 */
64
65#define pr_fmt(fmt) "IPv4: " fmt
66
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>
72#include <linux/mm.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>
78#include <linux/in.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>
96#include <net/dst.h>
97#include <net/net_namespace.h>
98#include <net/protocol.h>
99#include <net/ip.h>
100#include <net/route.h>
101#include <net/inetpeer.h>
102#include <net/sock.h>
103#include <net/ip_fib.h>
104#include <net/arp.h>
105#include <net/tcp.h>
106#include <net/icmp.h>
107#include <net/xfrm.h>
108#include <net/netevent.h>
109#include <net/rtnetlink.h>
110#ifdef CONFIG_SYSCTL
111#include <linux/sysctl.h>
112#include <linux/kmemleak.h>
113#endif
114#include <net/secure_seq.h>
115
116#define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
118
119#define IP_MAX_MTU 0xFFF0
120
121#define RT_GC_TIMEOUT (300*HZ)
122
123static int ip_rt_max_size;
124static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125static int ip_rt_gc_interval __read_mostly = 60 * HZ;
126static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
127static int ip_rt_redirect_number __read_mostly = 9;
128static int ip_rt_redirect_load __read_mostly = HZ / 50;
129static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
130static int ip_rt_error_cost __read_mostly = HZ;
131static int ip_rt_error_burst __read_mostly = 5 * HZ;
132static int ip_rt_gc_elasticity __read_mostly = 8;
133static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
134static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
135static int ip_rt_min_advmss __read_mostly = 256;
136static int rt_chain_length_max __read_mostly = 20;
137
138static struct delayed_work expires_work;
139static unsigned long expires_ljiffies;
140
141/*
142 * Interface to generic destination cache.
143 */
144
145static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
146static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
147static unsigned int ipv4_mtu(const struct dst_entry *dst);
148static void ipv4_dst_destroy(struct dst_entry *dst);
149static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
150static void ipv4_link_failure(struct sk_buff *skb);
151static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
152static int rt_garbage_collect(struct dst_ops *ops);
153
154static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
155 int how)
156{
157}
158
159static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
160{
161 struct rtable *rt = (struct rtable *) dst;
162 struct inet_peer *peer;
163 u32 *p = NULL;
164
165 if (!rt->peer)
166 rt_bind_peer(rt, rt->rt_dst, 1);
167
168 peer = rt->peer;
169 if (peer) {
170 u32 *old_p = __DST_METRICS_PTR(old);
171 unsigned long prev, new;
172
173 p = peer->metrics;
174 if (inet_metrics_new(peer))
175 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
176
177 new = (unsigned long) p;
178 prev = cmpxchg(&dst->_metrics, old, new);
179
180 if (prev != old) {
181 p = __DST_METRICS_PTR(prev);
182 if (prev & DST_METRICS_READ_ONLY)
183 p = NULL;
184 } else {
185 if (rt->fi) {
186 fib_info_put(rt->fi);
187 rt->fi = NULL;
188 }
189 }
190 }
191 return p;
192}
193
194static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr);
195
196static struct dst_ops ipv4_dst_ops = {
197 .family = AF_INET,
198 .protocol = cpu_to_be16(ETH_P_IP),
199 .gc = rt_garbage_collect,
200 .check = ipv4_dst_check,
201 .default_advmss = ipv4_default_advmss,
202 .mtu = ipv4_mtu,
203 .cow_metrics = ipv4_cow_metrics,
204 .destroy = ipv4_dst_destroy,
205 .ifdown = ipv4_dst_ifdown,
206 .negative_advice = ipv4_negative_advice,
207 .link_failure = ipv4_link_failure,
208 .update_pmtu = ip_rt_update_pmtu,
209 .local_out = __ip_local_out,
210 .neigh_lookup = ipv4_neigh_lookup,
211};
212
213#define ECN_OR_COST(class) TC_PRIO_##class
214
215const __u8 ip_tos2prio[16] = {
216 TC_PRIO_BESTEFFORT,
217 ECN_OR_COST(BESTEFFORT),
218 TC_PRIO_BESTEFFORT,
219 ECN_OR_COST(BESTEFFORT),
220 TC_PRIO_BULK,
221 ECN_OR_COST(BULK),
222 TC_PRIO_BULK,
223 ECN_OR_COST(BULK),
224 TC_PRIO_INTERACTIVE,
225 ECN_OR_COST(INTERACTIVE),
226 TC_PRIO_INTERACTIVE,
227 ECN_OR_COST(INTERACTIVE),
228 TC_PRIO_INTERACTIVE_BULK,
229 ECN_OR_COST(INTERACTIVE_BULK),
230 TC_PRIO_INTERACTIVE_BULK,
231 ECN_OR_COST(INTERACTIVE_BULK)
232};
233EXPORT_SYMBOL(ip_tos2prio);
234
235/*
236 * Route cache.
237 */
238
239/* The locking scheme is rather straight forward:
240 *
241 * 1) Read-Copy Update protects the buckets of the central route hash.
242 * 2) Only writers remove entries, and they hold the lock
243 * as they look at rtable reference counts.
244 * 3) Only readers acquire references to rtable entries,
245 * they do so with atomic increments and with the
246 * lock held.
247 */
248
249struct rt_hash_bucket {
250 struct rtable __rcu *chain;
251};
252
253#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
254 defined(CONFIG_PROVE_LOCKING)
255/*
256 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
257 * The size of this table is a power of two and depends on the number of CPUS.
258 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
259 */
260#ifdef CONFIG_LOCKDEP
261# define RT_HASH_LOCK_SZ 256
262#else
263# if NR_CPUS >= 32
264# define RT_HASH_LOCK_SZ 4096
265# elif NR_CPUS >= 16
266# define RT_HASH_LOCK_SZ 2048
267# elif NR_CPUS >= 8
268# define RT_HASH_LOCK_SZ 1024
269# elif NR_CPUS >= 4
270# define RT_HASH_LOCK_SZ 512
271# else
272# define RT_HASH_LOCK_SZ 256
273# endif
274#endif
275
276static spinlock_t *rt_hash_locks;
277# define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
278
279static __init void rt_hash_lock_init(void)
280{
281 int i;
282
283 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
284 GFP_KERNEL);
285 if (!rt_hash_locks)
286 panic("IP: failed to allocate rt_hash_locks\n");
287
288 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
289 spin_lock_init(&rt_hash_locks[i]);
290}
291#else
292# define rt_hash_lock_addr(slot) NULL
293
294static inline void rt_hash_lock_init(void)
295{
296}
297#endif
298
299static struct rt_hash_bucket *rt_hash_table __read_mostly;
300static unsigned int rt_hash_mask __read_mostly;
301static unsigned int rt_hash_log __read_mostly;
302
303static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
304#define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
305
306static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
307 int genid)
308{
309 return jhash_3words((__force u32)daddr, (__force u32)saddr,
310 idx, genid)
311 & rt_hash_mask;
312}
313
314static inline int rt_genid(struct net *net)
315{
316 return atomic_read(&net->ipv4.rt_genid);
317}
318
319#ifdef CONFIG_PROC_FS
320struct rt_cache_iter_state {
321 struct seq_net_private p;
322 int bucket;
323 int genid;
324};
325
326static struct rtable *rt_cache_get_first(struct seq_file *seq)
327{
328 struct rt_cache_iter_state *st = seq->private;
329 struct rtable *r = NULL;
330
331 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
332 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
333 continue;
334 rcu_read_lock_bh();
335 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
336 while (r) {
337 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
338 r->rt_genid == st->genid)
339 return r;
340 r = rcu_dereference_bh(r->dst.rt_next);
341 }
342 rcu_read_unlock_bh();
343 }
344 return r;
345}
346
347static struct rtable *__rt_cache_get_next(struct seq_file *seq,
348 struct rtable *r)
349{
350 struct rt_cache_iter_state *st = seq->private;
351
352 r = rcu_dereference_bh(r->dst.rt_next);
353 while (!r) {
354 rcu_read_unlock_bh();
355 do {
356 if (--st->bucket < 0)
357 return NULL;
358 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
359 rcu_read_lock_bh();
360 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
361 }
362 return r;
363}
364
365static struct rtable *rt_cache_get_next(struct seq_file *seq,
366 struct rtable *r)
367{
368 struct rt_cache_iter_state *st = seq->private;
369 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
370 if (dev_net(r->dst.dev) != seq_file_net(seq))
371 continue;
372 if (r->rt_genid == st->genid)
373 break;
374 }
375 return r;
376}
377
378static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
379{
380 struct rtable *r = rt_cache_get_first(seq);
381
382 if (r)
383 while (pos && (r = rt_cache_get_next(seq, r)))
384 --pos;
385 return pos ? NULL : r;
386}
387
388static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
389{
390 struct rt_cache_iter_state *st = seq->private;
391 if (*pos)
392 return rt_cache_get_idx(seq, *pos - 1);
393 st->genid = rt_genid(seq_file_net(seq));
394 return SEQ_START_TOKEN;
395}
396
397static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
398{
399 struct rtable *r;
400
401 if (v == SEQ_START_TOKEN)
402 r = rt_cache_get_first(seq);
403 else
404 r = rt_cache_get_next(seq, v);
405 ++*pos;
406 return r;
407}
408
409static void rt_cache_seq_stop(struct seq_file *seq, void *v)
410{
411 if (v && v != SEQ_START_TOKEN)
412 rcu_read_unlock_bh();
413}
414
415static int rt_cache_seq_show(struct seq_file *seq, void *v)
416{
417 if (v == SEQ_START_TOKEN)
418 seq_printf(seq, "%-127s\n",
419 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
420 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
421 "HHUptod\tSpecDst");
422 else {
423 struct rtable *r = v;
424 struct neighbour *n;
425 int len, HHUptod;
426
427 rcu_read_lock();
428 n = dst_get_neighbour_noref(&r->dst);
429 HHUptod = (n && (n->nud_state & NUD_CONNECTED)) ? 1 : 0;
430 rcu_read_unlock();
431
432 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
433 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
434 r->dst.dev ? r->dst.dev->name : "*",
435 (__force u32)r->rt_dst,
436 (__force u32)r->rt_gateway,
437 r->rt_flags, atomic_read(&r->dst.__refcnt),
438 r->dst.__use, 0, (__force u32)r->rt_src,
439 dst_metric_advmss(&r->dst) + 40,
440 dst_metric(&r->dst, RTAX_WINDOW),
441 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
442 dst_metric(&r->dst, RTAX_RTTVAR)),
443 r->rt_key_tos,
444 -1,
445 HHUptod,
446 r->rt_spec_dst, &len);
447
448 seq_printf(seq, "%*s\n", 127 - len, "");
449 }
450 return 0;
451}
452
453static const struct seq_operations rt_cache_seq_ops = {
454 .start = rt_cache_seq_start,
455 .next = rt_cache_seq_next,
456 .stop = rt_cache_seq_stop,
457 .show = rt_cache_seq_show,
458};
459
460static int rt_cache_seq_open(struct inode *inode, struct file *file)
461{
462 return seq_open_net(inode, file, &rt_cache_seq_ops,
463 sizeof(struct rt_cache_iter_state));
464}
465
466static const struct file_operations rt_cache_seq_fops = {
467 .owner = THIS_MODULE,
468 .open = rt_cache_seq_open,
469 .read = seq_read,
470 .llseek = seq_lseek,
471 .release = seq_release_net,
472};
473
474
475static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
476{
477 int cpu;
478
479 if (*pos == 0)
480 return SEQ_START_TOKEN;
481
482 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
483 if (!cpu_possible(cpu))
484 continue;
485 *pos = cpu+1;
486 return &per_cpu(rt_cache_stat, cpu);
487 }
488 return NULL;
489}
490
491static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
492{
493 int cpu;
494
495 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
496 if (!cpu_possible(cpu))
497 continue;
498 *pos = cpu+1;
499 return &per_cpu(rt_cache_stat, cpu);
500 }
501 return NULL;
502
503}
504
505static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
506{
507
508}
509
510static int rt_cpu_seq_show(struct seq_file *seq, void *v)
511{
512 struct rt_cache_stat *st = v;
513
514 if (v == SEQ_START_TOKEN) {
515 seq_printf(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
516 return 0;
517 }
518
519 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
520 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
521 dst_entries_get_slow(&ipv4_dst_ops),
522 st->in_hit,
523 st->in_slow_tot,
524 st->in_slow_mc,
525 st->in_no_route,
526 st->in_brd,
527 st->in_martian_dst,
528 st->in_martian_src,
529
530 st->out_hit,
531 st->out_slow_tot,
532 st->out_slow_mc,
533
534 st->gc_total,
535 st->gc_ignored,
536 st->gc_goal_miss,
537 st->gc_dst_overflow,
538 st->in_hlist_search,
539 st->out_hlist_search
540 );
541 return 0;
542}
543
544static const struct seq_operations rt_cpu_seq_ops = {
545 .start = rt_cpu_seq_start,
546 .next = rt_cpu_seq_next,
547 .stop = rt_cpu_seq_stop,
548 .show = rt_cpu_seq_show,
549};
550
551
552static int rt_cpu_seq_open(struct inode *inode, struct file *file)
553{
554 return seq_open(file, &rt_cpu_seq_ops);
555}
556
557static const struct file_operations rt_cpu_seq_fops = {
558 .owner = THIS_MODULE,
559 .open = rt_cpu_seq_open,
560 .read = seq_read,
561 .llseek = seq_lseek,
562 .release = seq_release,
563};
564
565#ifdef CONFIG_IP_ROUTE_CLASSID
566static int rt_acct_proc_show(struct seq_file *m, void *v)
567{
568 struct ip_rt_acct *dst, *src;
569 unsigned int i, j;
570
571 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
572 if (!dst)
573 return -ENOMEM;
574
575 for_each_possible_cpu(i) {
576 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
577 for (j = 0; j < 256; j++) {
578 dst[j].o_bytes += src[j].o_bytes;
579 dst[j].o_packets += src[j].o_packets;
580 dst[j].i_bytes += src[j].i_bytes;
581 dst[j].i_packets += src[j].i_packets;
582 }
583 }
584
585 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
586 kfree(dst);
587 return 0;
588}
589
590static int rt_acct_proc_open(struct inode *inode, struct file *file)
591{
592 return single_open(file, rt_acct_proc_show, NULL);
593}
594
595static const struct file_operations rt_acct_proc_fops = {
596 .owner = THIS_MODULE,
597 .open = rt_acct_proc_open,
598 .read = seq_read,
599 .llseek = seq_lseek,
600 .release = single_release,
601};
602#endif
603
604static int __net_init ip_rt_do_proc_init(struct net *net)
605{
606 struct proc_dir_entry *pde;
607
608 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
609 &rt_cache_seq_fops);
610 if (!pde)
611 goto err1;
612
613 pde = proc_create("rt_cache", S_IRUGO,
614 net->proc_net_stat, &rt_cpu_seq_fops);
615 if (!pde)
616 goto err2;
617
618#ifdef CONFIG_IP_ROUTE_CLASSID
619 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
620 if (!pde)
621 goto err3;
622#endif
623 return 0;
624
625#ifdef CONFIG_IP_ROUTE_CLASSID
626err3:
627 remove_proc_entry("rt_cache", net->proc_net_stat);
628#endif
629err2:
630 remove_proc_entry("rt_cache", net->proc_net);
631err1:
632 return -ENOMEM;
633}
634
635static void __net_exit ip_rt_do_proc_exit(struct net *net)
636{
637 remove_proc_entry("rt_cache", net->proc_net_stat);
638 remove_proc_entry("rt_cache", net->proc_net);
639#ifdef CONFIG_IP_ROUTE_CLASSID
640 remove_proc_entry("rt_acct", net->proc_net);
641#endif
642}
643
644static struct pernet_operations ip_rt_proc_ops __net_initdata = {
645 .init = ip_rt_do_proc_init,
646 .exit = ip_rt_do_proc_exit,
647};
648
649static int __init ip_rt_proc_init(void)
650{
651 return register_pernet_subsys(&ip_rt_proc_ops);
652}
653
654#else
655static inline int ip_rt_proc_init(void)
656{
657 return 0;
658}
659#endif /* CONFIG_PROC_FS */
660
661static inline void rt_free(struct rtable *rt)
662{
663 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
664}
665
666static inline void rt_drop(struct rtable *rt)
667{
668 ip_rt_put(rt);
669 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
670}
671
672static inline int rt_fast_clean(struct rtable *rth)
673{
674 /* Kill broadcast/multicast entries very aggresively, if they
675 collide in hash table with more useful entries */
676 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
677 rt_is_input_route(rth) && rth->dst.rt_next;
678}
679
680static inline int rt_valuable(struct rtable *rth)
681{
682 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
683 (rth->peer && rth->peer->pmtu_expires);
684}
685
686static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
687{
688 unsigned long age;
689 int ret = 0;
690
691 if (atomic_read(&rth->dst.__refcnt))
692 goto out;
693
694 age = jiffies - rth->dst.lastuse;
695 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
696 (age <= tmo2 && rt_valuable(rth)))
697 goto out;
698 ret = 1;
699out: return ret;
700}
701
702/* Bits of score are:
703 * 31: very valuable
704 * 30: not quite useless
705 * 29..0: usage counter
706 */
707static inline u32 rt_score(struct rtable *rt)
708{
709 u32 score = jiffies - rt->dst.lastuse;
710
711 score = ~score & ~(3<<30);
712
713 if (rt_valuable(rt))
714 score |= (1<<31);
715
716 if (rt_is_output_route(rt) ||
717 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
718 score |= (1<<30);
719
720 return score;
721}
722
723static inline bool rt_caching(const struct net *net)
724{
725 return net->ipv4.current_rt_cache_rebuild_count <=
726 net->ipv4.sysctl_rt_cache_rebuild_count;
727}
728
729static inline bool compare_hash_inputs(const struct rtable *rt1,
730 const struct rtable *rt2)
731{
732 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
733 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
734 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
735}
736
737static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
738{
739 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
740 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
741 (rt1->rt_mark ^ rt2->rt_mark) |
742 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
743 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
744 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
745}
746
747static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
748{
749 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
750}
751
752static inline int rt_is_expired(struct rtable *rth)
753{
754 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
755}
756
757/*
758 * Perform a full scan of hash table and free all entries.
759 * Can be called by a softirq or a process.
760 * In the later case, we want to be reschedule if necessary
761 */
762static void rt_do_flush(struct net *net, int process_context)
763{
764 unsigned int i;
765 struct rtable *rth, *next;
766
767 for (i = 0; i <= rt_hash_mask; i++) {
768 struct rtable __rcu **pprev;
769 struct rtable *list;
770
771 if (process_context && need_resched())
772 cond_resched();
773 rth = rcu_access_pointer(rt_hash_table[i].chain);
774 if (!rth)
775 continue;
776
777 spin_lock_bh(rt_hash_lock_addr(i));
778
779 list = NULL;
780 pprev = &rt_hash_table[i].chain;
781 rth = rcu_dereference_protected(*pprev,
782 lockdep_is_held(rt_hash_lock_addr(i)));
783
784 while (rth) {
785 next = rcu_dereference_protected(rth->dst.rt_next,
786 lockdep_is_held(rt_hash_lock_addr(i)));
787
788 if (!net ||
789 net_eq(dev_net(rth->dst.dev), net)) {
790 rcu_assign_pointer(*pprev, next);
791 rcu_assign_pointer(rth->dst.rt_next, list);
792 list = rth;
793 } else {
794 pprev = &rth->dst.rt_next;
795 }
796 rth = next;
797 }
798
799 spin_unlock_bh(rt_hash_lock_addr(i));
800
801 for (; list; list = next) {
802 next = rcu_dereference_protected(list->dst.rt_next, 1);
803 rt_free(list);
804 }
805 }
806}
807
808/*
809 * While freeing expired entries, we compute average chain length
810 * and standard deviation, using fixed-point arithmetic.
811 * This to have an estimation of rt_chain_length_max
812 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
813 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
814 */
815
816#define FRACT_BITS 3
817#define ONE (1UL << FRACT_BITS)
818
819/*
820 * Given a hash chain and an item in this hash chain,
821 * find if a previous entry has the same hash_inputs
822 * (but differs on tos, mark or oif)
823 * Returns 0 if an alias is found.
824 * Returns ONE if rth has no alias before itself.
825 */
826static int has_noalias(const struct rtable *head, const struct rtable *rth)
827{
828 const struct rtable *aux = head;
829
830 while (aux != rth) {
831 if (compare_hash_inputs(aux, rth))
832 return 0;
833 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
834 }
835 return ONE;
836}
837
838static void rt_check_expire(void)
839{
840 static unsigned int rover;
841 unsigned int i = rover, goal;
842 struct rtable *rth;
843 struct rtable __rcu **rthp;
844 unsigned long samples = 0;
845 unsigned long sum = 0, sum2 = 0;
846 unsigned long delta;
847 u64 mult;
848
849 delta = jiffies - expires_ljiffies;
850 expires_ljiffies = jiffies;
851 mult = ((u64)delta) << rt_hash_log;
852 if (ip_rt_gc_timeout > 1)
853 do_div(mult, ip_rt_gc_timeout);
854 goal = (unsigned int)mult;
855 if (goal > rt_hash_mask)
856 goal = rt_hash_mask + 1;
857 for (; goal > 0; goal--) {
858 unsigned long tmo = ip_rt_gc_timeout;
859 unsigned long length;
860
861 i = (i + 1) & rt_hash_mask;
862 rthp = &rt_hash_table[i].chain;
863
864 if (need_resched())
865 cond_resched();
866
867 samples++;
868
869 if (rcu_dereference_raw(*rthp) == NULL)
870 continue;
871 length = 0;
872 spin_lock_bh(rt_hash_lock_addr(i));
873 while ((rth = rcu_dereference_protected(*rthp,
874 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
875 prefetch(rth->dst.rt_next);
876 if (rt_is_expired(rth)) {
877 *rthp = rth->dst.rt_next;
878 rt_free(rth);
879 continue;
880 }
881 if (rth->dst.expires) {
882 /* Entry is expired even if it is in use */
883 if (time_before_eq(jiffies, rth->dst.expires)) {
884nofree:
885 tmo >>= 1;
886 rthp = &rth->dst.rt_next;
887 /*
888 * We only count entries on
889 * a chain with equal hash inputs once
890 * so that entries for different QOS
891 * levels, and other non-hash input
892 * attributes don't unfairly skew
893 * the length computation
894 */
895 length += has_noalias(rt_hash_table[i].chain, rth);
896 continue;
897 }
898 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
899 goto nofree;
900
901 /* Cleanup aged off entries. */
902 *rthp = rth->dst.rt_next;
903 rt_free(rth);
904 }
905 spin_unlock_bh(rt_hash_lock_addr(i));
906 sum += length;
907 sum2 += length*length;
908 }
909 if (samples) {
910 unsigned long avg = sum / samples;
911 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
912 rt_chain_length_max = max_t(unsigned long,
913 ip_rt_gc_elasticity,
914 (avg + 4*sd) >> FRACT_BITS);
915 }
916 rover = i;
917}
918
919/*
920 * rt_worker_func() is run in process context.
921 * we call rt_check_expire() to scan part of the hash table
922 */
923static void rt_worker_func(struct work_struct *work)
924{
925 rt_check_expire();
926 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
927}
928
929/*
930 * Perturbation of rt_genid by a small quantity [1..256]
931 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
932 * many times (2^24) without giving recent rt_genid.
933 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
934 */
935static void rt_cache_invalidate(struct net *net)
936{
937 unsigned char shuffle;
938
939 get_random_bytes(&shuffle, sizeof(shuffle));
940 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
941 inetpeer_invalidate_tree(AF_INET);
942}
943
944/*
945 * delay < 0 : invalidate cache (fast : entries will be deleted later)
946 * delay >= 0 : invalidate & flush cache (can be long)
947 */
948void rt_cache_flush(struct net *net, int delay)
949{
950 rt_cache_invalidate(net);
951 if (delay >= 0)
952 rt_do_flush(net, !in_softirq());
953}
954
955/* Flush previous cache invalidated entries from the cache */
956void rt_cache_flush_batch(struct net *net)
957{
958 rt_do_flush(net, !in_softirq());
959}
960
961static void rt_emergency_hash_rebuild(struct net *net)
962{
963 net_warn_ratelimited("Route hash chain too long!\n");
964 rt_cache_invalidate(net);
965}
966
967/*
968 Short description of GC goals.
969
970 We want to build algorithm, which will keep routing cache
971 at some equilibrium point, when number of aged off entries
972 is kept approximately equal to newly generated ones.
973
974 Current expiration strength is variable "expire".
975 We try to adjust it dynamically, so that if networking
976 is idle expires is large enough to keep enough of warm entries,
977 and when load increases it reduces to limit cache size.
978 */
979
980static int rt_garbage_collect(struct dst_ops *ops)
981{
982 static unsigned long expire = RT_GC_TIMEOUT;
983 static unsigned long last_gc;
984 static int rover;
985 static int equilibrium;
986 struct rtable *rth;
987 struct rtable __rcu **rthp;
988 unsigned long now = jiffies;
989 int goal;
990 int entries = dst_entries_get_fast(&ipv4_dst_ops);
991
992 /*
993 * Garbage collection is pretty expensive,
994 * do not make it too frequently.
995 */
996
997 RT_CACHE_STAT_INC(gc_total);
998
999 if (now - last_gc < ip_rt_gc_min_interval &&
1000 entries < ip_rt_max_size) {
1001 RT_CACHE_STAT_INC(gc_ignored);
1002 goto out;
1003 }
1004
1005 entries = dst_entries_get_slow(&ipv4_dst_ops);
1006 /* Calculate number of entries, which we want to expire now. */
1007 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
1008 if (goal <= 0) {
1009 if (equilibrium < ipv4_dst_ops.gc_thresh)
1010 equilibrium = ipv4_dst_ops.gc_thresh;
1011 goal = entries - equilibrium;
1012 if (goal > 0) {
1013 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1014 goal = entries - equilibrium;
1015 }
1016 } else {
1017 /* We are in dangerous area. Try to reduce cache really
1018 * aggressively.
1019 */
1020 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1021 equilibrium = entries - goal;
1022 }
1023
1024 if (now - last_gc >= ip_rt_gc_min_interval)
1025 last_gc = now;
1026
1027 if (goal <= 0) {
1028 equilibrium += goal;
1029 goto work_done;
1030 }
1031
1032 do {
1033 int i, k;
1034
1035 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1036 unsigned long tmo = expire;
1037
1038 k = (k + 1) & rt_hash_mask;
1039 rthp = &rt_hash_table[k].chain;
1040 spin_lock_bh(rt_hash_lock_addr(k));
1041 while ((rth = rcu_dereference_protected(*rthp,
1042 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1043 if (!rt_is_expired(rth) &&
1044 !rt_may_expire(rth, tmo, expire)) {
1045 tmo >>= 1;
1046 rthp = &rth->dst.rt_next;
1047 continue;
1048 }
1049 *rthp = rth->dst.rt_next;
1050 rt_free(rth);
1051 goal--;
1052 }
1053 spin_unlock_bh(rt_hash_lock_addr(k));
1054 if (goal <= 0)
1055 break;
1056 }
1057 rover = k;
1058
1059 if (goal <= 0)
1060 goto work_done;
1061
1062 /* Goal is not achieved. We stop process if:
1063
1064 - if expire reduced to zero. Otherwise, expire is halfed.
1065 - if table is not full.
1066 - if we are called from interrupt.
1067 - jiffies check is just fallback/debug loop breaker.
1068 We will not spin here for long time in any case.
1069 */
1070
1071 RT_CACHE_STAT_INC(gc_goal_miss);
1072
1073 if (expire == 0)
1074 break;
1075
1076 expire >>= 1;
1077
1078 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1079 goto out;
1080 } while (!in_softirq() && time_before_eq(jiffies, now));
1081
1082 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1083 goto out;
1084 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1085 goto out;
1086 net_warn_ratelimited("dst cache overflow\n");
1087 RT_CACHE_STAT_INC(gc_dst_overflow);
1088 return 1;
1089
1090work_done:
1091 expire += ip_rt_gc_min_interval;
1092 if (expire > ip_rt_gc_timeout ||
1093 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1094 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1095 expire = ip_rt_gc_timeout;
1096out: return 0;
1097}
1098
1099/*
1100 * Returns number of entries in a hash chain that have different hash_inputs
1101 */
1102static int slow_chain_length(const struct rtable *head)
1103{
1104 int length = 0;
1105 const struct rtable *rth = head;
1106
1107 while (rth) {
1108 length += has_noalias(head, rth);
1109 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1110 }
1111 return length >> FRACT_BITS;
1112}
1113
1114static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr)
1115{
1116 static const __be32 inaddr_any = 0;
1117 struct net_device *dev = dst->dev;
1118 const __be32 *pkey = daddr;
1119 const struct rtable *rt;
1120 struct neighbour *n;
1121
1122 rt = (const struct rtable *) dst;
1123
1124 if (dev->flags & (IFF_LOOPBACK | IFF_POINTOPOINT))
1125 pkey = &inaddr_any;
1126 else if (rt->rt_gateway)
1127 pkey = (const __be32 *) &rt->rt_gateway;
1128
1129 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1130 if (n)
1131 return n;
1132 return neigh_create(&arp_tbl, pkey, dev);
1133}
1134
1135static int rt_bind_neighbour(struct rtable *rt)
1136{
1137 struct neighbour *n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1138 if (IS_ERR(n))
1139 return PTR_ERR(n);
1140 dst_set_neighbour(&rt->dst, n);
1141
1142 return 0;
1143}
1144
1145static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1146 struct sk_buff *skb, int ifindex)
1147{
1148 struct rtable *rth, *cand;
1149 struct rtable __rcu **rthp, **candp;
1150 unsigned long now;
1151 u32 min_score;
1152 int chain_length;
1153 int attempts = !in_softirq();
1154
1155restart:
1156 chain_length = 0;
1157 min_score = ~(u32)0;
1158 cand = NULL;
1159 candp = NULL;
1160 now = jiffies;
1161
1162 if (!rt_caching(dev_net(rt->dst.dev))) {
1163 /*
1164 * If we're not caching, just tell the caller we
1165 * were successful and don't touch the route. The
1166 * caller hold the sole reference to the cache entry, and
1167 * it will be released when the caller is done with it.
1168 * If we drop it here, the callers have no way to resolve routes
1169 * when we're not caching. Instead, just point *rp at rt, so
1170 * the caller gets a single use out of the route
1171 * Note that we do rt_free on this new route entry, so that
1172 * once its refcount hits zero, we are still able to reap it
1173 * (Thanks Alexey)
1174 * Note: To avoid expensive rcu stuff for this uncached dst,
1175 * we set DST_NOCACHE so that dst_release() can free dst without
1176 * waiting a grace period.
1177 */
1178
1179 rt->dst.flags |= DST_NOCACHE;
1180 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1181 int err = rt_bind_neighbour(rt);
1182 if (err) {
1183 net_warn_ratelimited("Neighbour table failure & not caching routes\n");
1184 ip_rt_put(rt);
1185 return ERR_PTR(err);
1186 }
1187 }
1188
1189 goto skip_hashing;
1190 }
1191
1192 rthp = &rt_hash_table[hash].chain;
1193
1194 spin_lock_bh(rt_hash_lock_addr(hash));
1195 while ((rth = rcu_dereference_protected(*rthp,
1196 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1197 if (rt_is_expired(rth)) {
1198 *rthp = rth->dst.rt_next;
1199 rt_free(rth);
1200 continue;
1201 }
1202 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1203 /* Put it first */
1204 *rthp = rth->dst.rt_next;
1205 /*
1206 * Since lookup is lockfree, the deletion
1207 * must be visible to another weakly ordered CPU before
1208 * the insertion at the start of the hash chain.
1209 */
1210 rcu_assign_pointer(rth->dst.rt_next,
1211 rt_hash_table[hash].chain);
1212 /*
1213 * Since lookup is lockfree, the update writes
1214 * must be ordered for consistency on SMP.
1215 */
1216 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1217
1218 dst_use(&rth->dst, now);
1219 spin_unlock_bh(rt_hash_lock_addr(hash));
1220
1221 rt_drop(rt);
1222 if (skb)
1223 skb_dst_set(skb, &rth->dst);
1224 return rth;
1225 }
1226
1227 if (!atomic_read(&rth->dst.__refcnt)) {
1228 u32 score = rt_score(rth);
1229
1230 if (score <= min_score) {
1231 cand = rth;
1232 candp = rthp;
1233 min_score = score;
1234 }
1235 }
1236
1237 chain_length++;
1238
1239 rthp = &rth->dst.rt_next;
1240 }
1241
1242 if (cand) {
1243 /* ip_rt_gc_elasticity used to be average length of chain
1244 * length, when exceeded gc becomes really aggressive.
1245 *
1246 * The second limit is less certain. At the moment it allows
1247 * only 2 entries per bucket. We will see.
1248 */
1249 if (chain_length > ip_rt_gc_elasticity) {
1250 *candp = cand->dst.rt_next;
1251 rt_free(cand);
1252 }
1253 } else {
1254 if (chain_length > rt_chain_length_max &&
1255 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1256 struct net *net = dev_net(rt->dst.dev);
1257 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1258 if (!rt_caching(net)) {
1259 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1260 rt->dst.dev->name, num);
1261 }
1262 rt_emergency_hash_rebuild(net);
1263 spin_unlock_bh(rt_hash_lock_addr(hash));
1264
1265 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1266 ifindex, rt_genid(net));
1267 goto restart;
1268 }
1269 }
1270
1271 /* Try to bind route to arp only if it is output
1272 route or unicast forwarding path.
1273 */
1274 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1275 int err = rt_bind_neighbour(rt);
1276 if (err) {
1277 spin_unlock_bh(rt_hash_lock_addr(hash));
1278
1279 if (err != -ENOBUFS) {
1280 rt_drop(rt);
1281 return ERR_PTR(err);
1282 }
1283
1284 /* Neighbour tables are full and nothing
1285 can be released. Try to shrink route cache,
1286 it is most likely it holds some neighbour records.
1287 */
1288 if (attempts-- > 0) {
1289 int saved_elasticity = ip_rt_gc_elasticity;
1290 int saved_int = ip_rt_gc_min_interval;
1291 ip_rt_gc_elasticity = 1;
1292 ip_rt_gc_min_interval = 0;
1293 rt_garbage_collect(&ipv4_dst_ops);
1294 ip_rt_gc_min_interval = saved_int;
1295 ip_rt_gc_elasticity = saved_elasticity;
1296 goto restart;
1297 }
1298
1299 net_warn_ratelimited("Neighbour table overflow\n");
1300 rt_drop(rt);
1301 return ERR_PTR(-ENOBUFS);
1302 }
1303 }
1304
1305 rt->dst.rt_next = rt_hash_table[hash].chain;
1306
1307 /*
1308 * Since lookup is lockfree, we must make sure
1309 * previous writes to rt are committed to memory
1310 * before making rt visible to other CPUS.
1311 */
1312 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1313
1314 spin_unlock_bh(rt_hash_lock_addr(hash));
1315
1316skip_hashing:
1317 if (skb)
1318 skb_dst_set(skb, &rt->dst);
1319 return rt;
1320}
1321
1322static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1323
1324static u32 rt_peer_genid(void)
1325{
1326 return atomic_read(&__rt_peer_genid);
1327}
1328
1329void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1330{
1331 struct inet_peer *peer;
1332
1333 peer = inet_getpeer_v4(daddr, create);
1334
1335 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1336 inet_putpeer(peer);
1337 else
1338 rt->rt_peer_genid = rt_peer_genid();
1339}
1340
1341/*
1342 * Peer allocation may fail only in serious out-of-memory conditions. However
1343 * we still can generate some output.
1344 * Random ID selection looks a bit dangerous because we have no chances to
1345 * select ID being unique in a reasonable period of time.
1346 * But broken packet identifier may be better than no packet at all.
1347 */
1348static void ip_select_fb_ident(struct iphdr *iph)
1349{
1350 static DEFINE_SPINLOCK(ip_fb_id_lock);
1351 static u32 ip_fallback_id;
1352 u32 salt;
1353
1354 spin_lock_bh(&ip_fb_id_lock);
1355 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1356 iph->id = htons(salt & 0xFFFF);
1357 ip_fallback_id = salt;
1358 spin_unlock_bh(&ip_fb_id_lock);
1359}
1360
1361void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1362{
1363 struct rtable *rt = (struct rtable *) dst;
1364
1365 if (rt && !(rt->dst.flags & DST_NOPEER)) {
1366 if (rt->peer == NULL)
1367 rt_bind_peer(rt, rt->rt_dst, 1);
1368
1369 /* If peer is attached to destination, it is never detached,
1370 so that we need not to grab a lock to dereference it.
1371 */
1372 if (rt->peer) {
1373 iph->id = htons(inet_getid(rt->peer, more));
1374 return;
1375 }
1376 } else if (!rt)
1377 pr_debug("rt_bind_peer(0) @%p\n", __builtin_return_address(0));
1378
1379 ip_select_fb_ident(iph);
1380}
1381EXPORT_SYMBOL(__ip_select_ident);
1382
1383static void rt_del(unsigned int hash, struct rtable *rt)
1384{
1385 struct rtable __rcu **rthp;
1386 struct rtable *aux;
1387
1388 rthp = &rt_hash_table[hash].chain;
1389 spin_lock_bh(rt_hash_lock_addr(hash));
1390 ip_rt_put(rt);
1391 while ((aux = rcu_dereference_protected(*rthp,
1392 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1393 if (aux == rt || rt_is_expired(aux)) {
1394 *rthp = aux->dst.rt_next;
1395 rt_free(aux);
1396 continue;
1397 }
1398 rthp = &aux->dst.rt_next;
1399 }
1400 spin_unlock_bh(rt_hash_lock_addr(hash));
1401}
1402
1403static void check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1404{
1405 struct rtable *rt = (struct rtable *) dst;
1406 __be32 orig_gw = rt->rt_gateway;
1407 struct neighbour *n, *old_n;
1408
1409 dst_confirm(&rt->dst);
1410
1411 rt->rt_gateway = peer->redirect_learned.a4;
1412
1413 n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1414 if (IS_ERR(n)) {
1415 rt->rt_gateway = orig_gw;
1416 return;
1417 }
1418 old_n = xchg(&rt->dst._neighbour, n);
1419 if (old_n)
1420 neigh_release(old_n);
1421 if (!(n->nud_state & NUD_VALID)) {
1422 neigh_event_send(n, NULL);
1423 } else {
1424 rt->rt_flags |= RTCF_REDIRECTED;
1425 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1426 }
1427}
1428
1429/* called in rcu_read_lock() section */
1430void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1431 __be32 saddr, struct net_device *dev)
1432{
1433 int s, i;
1434 struct in_device *in_dev = __in_dev_get_rcu(dev);
1435 __be32 skeys[2] = { saddr, 0 };
1436 int ikeys[2] = { dev->ifindex, 0 };
1437 struct inet_peer *peer;
1438 struct net *net;
1439
1440 if (!in_dev)
1441 return;
1442
1443 net = dev_net(dev);
1444 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1445 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1446 ipv4_is_zeronet(new_gw))
1447 goto reject_redirect;
1448
1449 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1450 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1451 goto reject_redirect;
1452 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1453 goto reject_redirect;
1454 } else {
1455 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1456 goto reject_redirect;
1457 }
1458
1459 for (s = 0; s < 2; s++) {
1460 for (i = 0; i < 2; i++) {
1461 unsigned int hash;
1462 struct rtable __rcu **rthp;
1463 struct rtable *rt;
1464
1465 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1466
1467 rthp = &rt_hash_table[hash].chain;
1468
1469 while ((rt = rcu_dereference(*rthp)) != NULL) {
1470 rthp = &rt->dst.rt_next;
1471
1472 if (rt->rt_key_dst != daddr ||
1473 rt->rt_key_src != skeys[s] ||
1474 rt->rt_oif != ikeys[i] ||
1475 rt_is_input_route(rt) ||
1476 rt_is_expired(rt) ||
1477 !net_eq(dev_net(rt->dst.dev), net) ||
1478 rt->dst.error ||
1479 rt->dst.dev != dev ||
1480 rt->rt_gateway != old_gw)
1481 continue;
1482
1483 if (!rt->peer)
1484 rt_bind_peer(rt, rt->rt_dst, 1);
1485
1486 peer = rt->peer;
1487 if (peer) {
1488 if (peer->redirect_learned.a4 != new_gw) {
1489 peer->redirect_learned.a4 = new_gw;
1490 atomic_inc(&__rt_peer_genid);
1491 }
1492 check_peer_redir(&rt->dst, peer);
1493 }
1494 }
1495 }
1496 }
1497 return;
1498
1499reject_redirect:
1500#ifdef CONFIG_IP_ROUTE_VERBOSE
1501 if (IN_DEV_LOG_MARTIANS(in_dev))
1502 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1503 " Advised path = %pI4 -> %pI4\n",
1504 &old_gw, dev->name, &new_gw,
1505 &saddr, &daddr);
1506#endif
1507 ;
1508}
1509
1510static bool peer_pmtu_expired(struct inet_peer *peer)
1511{
1512 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1513
1514 return orig &&
1515 time_after_eq(jiffies, orig) &&
1516 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1517}
1518
1519static bool peer_pmtu_cleaned(struct inet_peer *peer)
1520{
1521 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1522
1523 return orig &&
1524 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1525}
1526
1527static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1528{
1529 struct rtable *rt = (struct rtable *)dst;
1530 struct dst_entry *ret = dst;
1531
1532 if (rt) {
1533 if (dst->obsolete > 0) {
1534 ip_rt_put(rt);
1535 ret = NULL;
1536 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1537 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1538 rt->rt_oif,
1539 rt_genid(dev_net(dst->dev)));
1540 rt_del(hash, rt);
1541 ret = NULL;
1542 } else if (rt->peer && peer_pmtu_expired(rt->peer)) {
1543 dst_metric_set(dst, RTAX_MTU, rt->peer->pmtu_orig);
1544 }
1545 }
1546 return ret;
1547}
1548
1549/*
1550 * Algorithm:
1551 * 1. The first ip_rt_redirect_number redirects are sent
1552 * with exponential backoff, then we stop sending them at all,
1553 * assuming that the host ignores our redirects.
1554 * 2. If we did not see packets requiring redirects
1555 * during ip_rt_redirect_silence, we assume that the host
1556 * forgot redirected route and start to send redirects again.
1557 *
1558 * This algorithm is much cheaper and more intelligent than dumb load limiting
1559 * in icmp.c.
1560 *
1561 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1562 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1563 */
1564
1565void ip_rt_send_redirect(struct sk_buff *skb)
1566{
1567 struct rtable *rt = skb_rtable(skb);
1568 struct in_device *in_dev;
1569 struct inet_peer *peer;
1570 int log_martians;
1571
1572 rcu_read_lock();
1573 in_dev = __in_dev_get_rcu(rt->dst.dev);
1574 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1575 rcu_read_unlock();
1576 return;
1577 }
1578 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1579 rcu_read_unlock();
1580
1581 if (!rt->peer)
1582 rt_bind_peer(rt, rt->rt_dst, 1);
1583 peer = rt->peer;
1584 if (!peer) {
1585 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1586 return;
1587 }
1588
1589 /* No redirected packets during ip_rt_redirect_silence;
1590 * reset the algorithm.
1591 */
1592 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1593 peer->rate_tokens = 0;
1594
1595 /* Too many ignored redirects; do not send anything
1596 * set dst.rate_last to the last seen redirected packet.
1597 */
1598 if (peer->rate_tokens >= ip_rt_redirect_number) {
1599 peer->rate_last = jiffies;
1600 return;
1601 }
1602
1603 /* Check for load limit; set rate_last to the latest sent
1604 * redirect.
1605 */
1606 if (peer->rate_tokens == 0 ||
1607 time_after(jiffies,
1608 (peer->rate_last +
1609 (ip_rt_redirect_load << peer->rate_tokens)))) {
1610 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1611 peer->rate_last = jiffies;
1612 ++peer->rate_tokens;
1613#ifdef CONFIG_IP_ROUTE_VERBOSE
1614 if (log_martians &&
1615 peer->rate_tokens == ip_rt_redirect_number)
1616 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1617 &ip_hdr(skb)->saddr, rt->rt_iif,
1618 &rt->rt_dst, &rt->rt_gateway);
1619#endif
1620 }
1621}
1622
1623static int ip_error(struct sk_buff *skb)
1624{
1625 struct rtable *rt = skb_rtable(skb);
1626 struct inet_peer *peer;
1627 unsigned long now;
1628 bool send;
1629 int code;
1630
1631 switch (rt->dst.error) {
1632 case EINVAL:
1633 default:
1634 goto out;
1635 case EHOSTUNREACH:
1636 code = ICMP_HOST_UNREACH;
1637 break;
1638 case ENETUNREACH:
1639 code = ICMP_NET_UNREACH;
1640 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1641 IPSTATS_MIB_INNOROUTES);
1642 break;
1643 case EACCES:
1644 code = ICMP_PKT_FILTERED;
1645 break;
1646 }
1647
1648 if (!rt->peer)
1649 rt_bind_peer(rt, rt->rt_dst, 1);
1650 peer = rt->peer;
1651
1652 send = true;
1653 if (peer) {
1654 now = jiffies;
1655 peer->rate_tokens += now - peer->rate_last;
1656 if (peer->rate_tokens > ip_rt_error_burst)
1657 peer->rate_tokens = ip_rt_error_burst;
1658 peer->rate_last = now;
1659 if (peer->rate_tokens >= ip_rt_error_cost)
1660 peer->rate_tokens -= ip_rt_error_cost;
1661 else
1662 send = false;
1663 }
1664 if (send)
1665 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1666
1667out: kfree_skb(skb);
1668 return 0;
1669}
1670
1671/*
1672 * The last two values are not from the RFC but
1673 * are needed for AMPRnet AX.25 paths.
1674 */
1675
1676static const unsigned short mtu_plateau[] =
1677{32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1678
1679static inline unsigned short guess_mtu(unsigned short old_mtu)
1680{
1681 int i;
1682
1683 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1684 if (old_mtu > mtu_plateau[i])
1685 return mtu_plateau[i];
1686 return 68;
1687}
1688
1689unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1690 unsigned short new_mtu,
1691 struct net_device *dev)
1692{
1693 unsigned short old_mtu = ntohs(iph->tot_len);
1694 unsigned short est_mtu = 0;
1695 struct inet_peer *peer;
1696
1697 peer = inet_getpeer_v4(iph->daddr, 1);
1698 if (peer) {
1699 unsigned short mtu = new_mtu;
1700
1701 if (new_mtu < 68 || new_mtu >= old_mtu) {
1702 /* BSD 4.2 derived systems incorrectly adjust
1703 * tot_len by the IP header length, and report
1704 * a zero MTU in the ICMP message.
1705 */
1706 if (mtu == 0 &&
1707 old_mtu >= 68 + (iph->ihl << 2))
1708 old_mtu -= iph->ihl << 2;
1709 mtu = guess_mtu(old_mtu);
1710 }
1711
1712 if (mtu < ip_rt_min_pmtu)
1713 mtu = ip_rt_min_pmtu;
1714 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1715 unsigned long pmtu_expires;
1716
1717 pmtu_expires = jiffies + ip_rt_mtu_expires;
1718 if (!pmtu_expires)
1719 pmtu_expires = 1UL;
1720
1721 est_mtu = mtu;
1722 peer->pmtu_learned = mtu;
1723 peer->pmtu_expires = pmtu_expires;
1724 atomic_inc(&__rt_peer_genid);
1725 }
1726
1727 inet_putpeer(peer);
1728 }
1729 return est_mtu ? : new_mtu;
1730}
1731
1732static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1733{
1734 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1735
1736 if (!expires)
1737 return;
1738 if (time_before(jiffies, expires)) {
1739 u32 orig_dst_mtu = dst_mtu(dst);
1740 if (peer->pmtu_learned < orig_dst_mtu) {
1741 if (!peer->pmtu_orig)
1742 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1743 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1744 }
1745 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1746 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1747}
1748
1749static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1750{
1751 struct rtable *rt = (struct rtable *) dst;
1752 struct inet_peer *peer;
1753
1754 dst_confirm(dst);
1755
1756 if (!rt->peer)
1757 rt_bind_peer(rt, rt->rt_dst, 1);
1758 peer = rt->peer;
1759 if (peer) {
1760 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1761
1762 if (mtu < ip_rt_min_pmtu)
1763 mtu = ip_rt_min_pmtu;
1764 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1765
1766 pmtu_expires = jiffies + ip_rt_mtu_expires;
1767 if (!pmtu_expires)
1768 pmtu_expires = 1UL;
1769
1770 peer->pmtu_learned = mtu;
1771 peer->pmtu_expires = pmtu_expires;
1772
1773 atomic_inc(&__rt_peer_genid);
1774 rt->rt_peer_genid = rt_peer_genid();
1775 }
1776 check_peer_pmtu(dst, peer);
1777 }
1778}
1779
1780
1781static void ipv4_validate_peer(struct rtable *rt)
1782{
1783 if (rt->rt_peer_genid != rt_peer_genid()) {
1784 struct inet_peer *peer;
1785
1786 if (!rt->peer)
1787 rt_bind_peer(rt, rt->rt_dst, 0);
1788
1789 peer = rt->peer;
1790 if (peer) {
1791 check_peer_pmtu(&rt->dst, peer);
1792
1793 if (peer->redirect_learned.a4 &&
1794 peer->redirect_learned.a4 != rt->rt_gateway)
1795 check_peer_redir(&rt->dst, peer);
1796 }
1797
1798 rt->rt_peer_genid = rt_peer_genid();
1799 }
1800}
1801
1802static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1803{
1804 struct rtable *rt = (struct rtable *) dst;
1805
1806 if (rt_is_expired(rt))
1807 return NULL;
1808 ipv4_validate_peer(rt);
1809 return dst;
1810}
1811
1812static void ipv4_dst_destroy(struct dst_entry *dst)
1813{
1814 struct rtable *rt = (struct rtable *) dst;
1815 struct inet_peer *peer = rt->peer;
1816
1817 if (rt->fi) {
1818 fib_info_put(rt->fi);
1819 rt->fi = NULL;
1820 }
1821 if (peer) {
1822 rt->peer = NULL;
1823 inet_putpeer(peer);
1824 }
1825}
1826
1827
1828static void ipv4_link_failure(struct sk_buff *skb)
1829{
1830 struct rtable *rt;
1831
1832 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1833
1834 rt = skb_rtable(skb);
1835 if (rt && rt->peer && peer_pmtu_cleaned(rt->peer))
1836 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1837}
1838
1839static int ip_rt_bug(struct sk_buff *skb)
1840{
1841 pr_debug("%s: %pI4 -> %pI4, %s\n",
1842 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1843 skb->dev ? skb->dev->name : "?");
1844 kfree_skb(skb);
1845 WARN_ON(1);
1846 return 0;
1847}
1848
1849/*
1850 We do not cache source address of outgoing interface,
1851 because it is used only by IP RR, TS and SRR options,
1852 so that it out of fast path.
1853
1854 BTW remember: "addr" is allowed to be not aligned
1855 in IP options!
1856 */
1857
1858void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1859{
1860 __be32 src;
1861
1862 if (rt_is_output_route(rt))
1863 src = ip_hdr(skb)->saddr;
1864 else {
1865 struct fib_result res;
1866 struct flowi4 fl4;
1867 struct iphdr *iph;
1868
1869 iph = ip_hdr(skb);
1870
1871 memset(&fl4, 0, sizeof(fl4));
1872 fl4.daddr = iph->daddr;
1873 fl4.saddr = iph->saddr;
1874 fl4.flowi4_tos = RT_TOS(iph->tos);
1875 fl4.flowi4_oif = rt->dst.dev->ifindex;
1876 fl4.flowi4_iif = skb->dev->ifindex;
1877 fl4.flowi4_mark = skb->mark;
1878
1879 rcu_read_lock();
1880 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1881 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1882 else
1883 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1884 RT_SCOPE_UNIVERSE);
1885 rcu_read_unlock();
1886 }
1887 memcpy(addr, &src, 4);
1888}
1889
1890#ifdef CONFIG_IP_ROUTE_CLASSID
1891static void set_class_tag(struct rtable *rt, u32 tag)
1892{
1893 if (!(rt->dst.tclassid & 0xFFFF))
1894 rt->dst.tclassid |= tag & 0xFFFF;
1895 if (!(rt->dst.tclassid & 0xFFFF0000))
1896 rt->dst.tclassid |= tag & 0xFFFF0000;
1897}
1898#endif
1899
1900static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1901{
1902 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1903
1904 if (advmss == 0) {
1905 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1906 ip_rt_min_advmss);
1907 if (advmss > 65535 - 40)
1908 advmss = 65535 - 40;
1909 }
1910 return advmss;
1911}
1912
1913static unsigned int ipv4_mtu(const struct dst_entry *dst)
1914{
1915 const struct rtable *rt = (const struct rtable *) dst;
1916 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1917
1918 if (mtu && rt_is_output_route(rt))
1919 return mtu;
1920
1921 mtu = dst->dev->mtu;
1922
1923 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1924
1925 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1926 mtu = 576;
1927 }
1928
1929 if (mtu > IP_MAX_MTU)
1930 mtu = IP_MAX_MTU;
1931
1932 return mtu;
1933}
1934
1935static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1936 struct fib_info *fi)
1937{
1938 struct inet_peer *peer;
1939 int create = 0;
1940
1941 /* If a peer entry exists for this destination, we must hook
1942 * it up in order to get at cached metrics.
1943 */
1944 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1945 create = 1;
1946
1947 rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1948 if (peer) {
1949 rt->rt_peer_genid = rt_peer_genid();
1950 if (inet_metrics_new(peer))
1951 memcpy(peer->metrics, fi->fib_metrics,
1952 sizeof(u32) * RTAX_MAX);
1953 dst_init_metrics(&rt->dst, peer->metrics, false);
1954
1955 check_peer_pmtu(&rt->dst, peer);
1956
1957 if (peer->redirect_learned.a4 &&
1958 peer->redirect_learned.a4 != rt->rt_gateway) {
1959 rt->rt_gateway = peer->redirect_learned.a4;
1960 rt->rt_flags |= RTCF_REDIRECTED;
1961 }
1962 } else {
1963 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1964 rt->fi = fi;
1965 atomic_inc(&fi->fib_clntref);
1966 }
1967 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1968 }
1969}
1970
1971static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1972 const struct fib_result *res,
1973 struct fib_info *fi, u16 type, u32 itag)
1974{
1975 struct dst_entry *dst = &rt->dst;
1976
1977 if (fi) {
1978 if (FIB_RES_GW(*res) &&
1979 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1980 rt->rt_gateway = FIB_RES_GW(*res);
1981 rt_init_metrics(rt, fl4, fi);
1982#ifdef CONFIG_IP_ROUTE_CLASSID
1983 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1984#endif
1985 }
1986
1987 if (dst_mtu(dst) > IP_MAX_MTU)
1988 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1989 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1990 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1991
1992#ifdef CONFIG_IP_ROUTE_CLASSID
1993#ifdef CONFIG_IP_MULTIPLE_TABLES
1994 set_class_tag(rt, fib_rules_tclass(res));
1995#endif
1996 set_class_tag(rt, itag);
1997#endif
1998}
1999
2000static struct rtable *rt_dst_alloc(struct net_device *dev,
2001 bool nopolicy, bool noxfrm)
2002{
2003 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
2004 DST_HOST |
2005 (nopolicy ? DST_NOPOLICY : 0) |
2006 (noxfrm ? DST_NOXFRM : 0));
2007}
2008
2009/* called in rcu_read_lock() section */
2010static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2011 u8 tos, struct net_device *dev, int our)
2012{
2013 unsigned int hash;
2014 struct rtable *rth;
2015 __be32 spec_dst;
2016 struct in_device *in_dev = __in_dev_get_rcu(dev);
2017 u32 itag = 0;
2018 int err;
2019
2020 /* Primary sanity checks. */
2021
2022 if (in_dev == NULL)
2023 return -EINVAL;
2024
2025 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2026 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
2027 goto e_inval;
2028
2029 if (ipv4_is_zeronet(saddr)) {
2030 if (!ipv4_is_local_multicast(daddr))
2031 goto e_inval;
2032 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2033 } else {
2034 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2035 &itag);
2036 if (err < 0)
2037 goto e_err;
2038 }
2039 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
2040 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2041 if (!rth)
2042 goto e_nobufs;
2043
2044#ifdef CONFIG_IP_ROUTE_CLASSID
2045 rth->dst.tclassid = itag;
2046#endif
2047 rth->dst.output = ip_rt_bug;
2048
2049 rth->rt_key_dst = daddr;
2050 rth->rt_key_src = saddr;
2051 rth->rt_genid = rt_genid(dev_net(dev));
2052 rth->rt_flags = RTCF_MULTICAST;
2053 rth->rt_type = RTN_MULTICAST;
2054 rth->rt_key_tos = tos;
2055 rth->rt_dst = daddr;
2056 rth->rt_src = saddr;
2057 rth->rt_route_iif = dev->ifindex;
2058 rth->rt_iif = dev->ifindex;
2059 rth->rt_oif = 0;
2060 rth->rt_mark = skb->mark;
2061 rth->rt_gateway = daddr;
2062 rth->rt_spec_dst= spec_dst;
2063 rth->rt_peer_genid = 0;
2064 rth->peer = NULL;
2065 rth->fi = NULL;
2066 if (our) {
2067 rth->dst.input= ip_local_deliver;
2068 rth->rt_flags |= RTCF_LOCAL;
2069 }
2070
2071#ifdef CONFIG_IP_MROUTE
2072 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
2073 rth->dst.input = ip_mr_input;
2074#endif
2075 RT_CACHE_STAT_INC(in_slow_mc);
2076
2077 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
2078 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
2079 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
2080
2081e_nobufs:
2082 return -ENOBUFS;
2083e_inval:
2084 return -EINVAL;
2085e_err:
2086 return err;
2087}
2088
2089
2090static void ip_handle_martian_source(struct net_device *dev,
2091 struct in_device *in_dev,
2092 struct sk_buff *skb,
2093 __be32 daddr,
2094 __be32 saddr)
2095{
2096 RT_CACHE_STAT_INC(in_martian_src);
2097#ifdef CONFIG_IP_ROUTE_VERBOSE
2098 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2099 /*
2100 * RFC1812 recommendation, if source is martian,
2101 * the only hint is MAC header.
2102 */
2103 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
2104 &daddr, &saddr, dev->name);
2105 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2106 print_hex_dump(KERN_WARNING, "ll header: ",
2107 DUMP_PREFIX_OFFSET, 16, 1,
2108 skb_mac_header(skb),
2109 dev->hard_header_len, true);
2110 }
2111 }
2112#endif
2113}
2114
2115/* called in rcu_read_lock() section */
2116static int __mkroute_input(struct sk_buff *skb,
2117 const struct fib_result *res,
2118 struct in_device *in_dev,
2119 __be32 daddr, __be32 saddr, u32 tos,
2120 struct rtable **result)
2121{
2122 struct rtable *rth;
2123 int err;
2124 struct in_device *out_dev;
2125 unsigned int flags = 0;
2126 __be32 spec_dst;
2127 u32 itag;
2128
2129 /* get a working reference to the output device */
2130 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2131 if (out_dev == NULL) {
2132 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
2133 return -EINVAL;
2134 }
2135
2136
2137 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2138 in_dev->dev, &spec_dst, &itag);
2139 if (err < 0) {
2140 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2141 saddr);
2142
2143 goto cleanup;
2144 }
2145
2146 if (err)
2147 flags |= RTCF_DIRECTSRC;
2148
2149 if (out_dev == in_dev && err &&
2150 (IN_DEV_SHARED_MEDIA(out_dev) ||
2151 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2152 flags |= RTCF_DOREDIRECT;
2153
2154 if (skb->protocol != htons(ETH_P_IP)) {
2155 /* Not IP (i.e. ARP). Do not create route, if it is
2156 * invalid for proxy arp. DNAT routes are always valid.
2157 *
2158 * Proxy arp feature have been extended to allow, ARP
2159 * replies back to the same interface, to support
2160 * Private VLAN switch technologies. See arp.c.
2161 */
2162 if (out_dev == in_dev &&
2163 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2164 err = -EINVAL;
2165 goto cleanup;
2166 }
2167 }
2168
2169 rth = rt_dst_alloc(out_dev->dev,
2170 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2171 IN_DEV_CONF_GET(out_dev, NOXFRM));
2172 if (!rth) {
2173 err = -ENOBUFS;
2174 goto cleanup;
2175 }
2176
2177 rth->rt_key_dst = daddr;
2178 rth->rt_key_src = saddr;
2179 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2180 rth->rt_flags = flags;
2181 rth->rt_type = res->type;
2182 rth->rt_key_tos = tos;
2183 rth->rt_dst = daddr;
2184 rth->rt_src = saddr;
2185 rth->rt_route_iif = in_dev->dev->ifindex;
2186 rth->rt_iif = in_dev->dev->ifindex;
2187 rth->rt_oif = 0;
2188 rth->rt_mark = skb->mark;
2189 rth->rt_gateway = daddr;
2190 rth->rt_spec_dst= spec_dst;
2191 rth->rt_peer_genid = 0;
2192 rth->peer = NULL;
2193 rth->fi = NULL;
2194
2195 rth->dst.input = ip_forward;
2196 rth->dst.output = ip_output;
2197
2198 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2199
2200 *result = rth;
2201 err = 0;
2202 cleanup:
2203 return err;
2204}
2205
2206static int ip_mkroute_input(struct sk_buff *skb,
2207 struct fib_result *res,
2208 const struct flowi4 *fl4,
2209 struct in_device *in_dev,
2210 __be32 daddr, __be32 saddr, u32 tos)
2211{
2212 struct rtable *rth = NULL;
2213 int err;
2214 unsigned int hash;
2215
2216#ifdef CONFIG_IP_ROUTE_MULTIPATH
2217 if (res->fi && res->fi->fib_nhs > 1)
2218 fib_select_multipath(res);
2219#endif
2220
2221 /* create a routing cache entry */
2222 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2223 if (err)
2224 return err;
2225
2226 /* put it into the cache */
2227 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2228 rt_genid(dev_net(rth->dst.dev)));
2229 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2230 if (IS_ERR(rth))
2231 return PTR_ERR(rth);
2232 return 0;
2233}
2234
2235/*
2236 * NOTE. We drop all the packets that has local source
2237 * addresses, because every properly looped back packet
2238 * must have correct destination already attached by output routine.
2239 *
2240 * Such approach solves two big problems:
2241 * 1. Not simplex devices are handled properly.
2242 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2243 * called with rcu_read_lock()
2244 */
2245
2246static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2247 u8 tos, struct net_device *dev)
2248{
2249 struct fib_result res;
2250 struct in_device *in_dev = __in_dev_get_rcu(dev);
2251 struct flowi4 fl4;
2252 unsigned int flags = 0;
2253 u32 itag = 0;
2254 struct rtable *rth;
2255 unsigned int hash;
2256 __be32 spec_dst;
2257 int err = -EINVAL;
2258 struct net *net = dev_net(dev);
2259
2260 /* IP on this device is disabled. */
2261
2262 if (!in_dev)
2263 goto out;
2264
2265 /* Check for the most weird martians, which can be not detected
2266 by fib_lookup.
2267 */
2268
2269 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2270 ipv4_is_loopback(saddr))
2271 goto martian_source;
2272
2273 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2274 goto brd_input;
2275
2276 /* Accept zero addresses only to limited broadcast;
2277 * I even do not know to fix it or not. Waiting for complains :-)
2278 */
2279 if (ipv4_is_zeronet(saddr))
2280 goto martian_source;
2281
2282 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2283 goto martian_destination;
2284
2285 /*
2286 * Now we are ready to route packet.
2287 */
2288 fl4.flowi4_oif = 0;
2289 fl4.flowi4_iif = dev->ifindex;
2290 fl4.flowi4_mark = skb->mark;
2291 fl4.flowi4_tos = tos;
2292 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2293 fl4.daddr = daddr;
2294 fl4.saddr = saddr;
2295 err = fib_lookup(net, &fl4, &res);
2296 if (err != 0) {
2297 if (!IN_DEV_FORWARD(in_dev))
2298 goto e_hostunreach;
2299 goto no_route;
2300 }
2301
2302 RT_CACHE_STAT_INC(in_slow_tot);
2303
2304 if (res.type == RTN_BROADCAST)
2305 goto brd_input;
2306
2307 if (res.type == RTN_LOCAL) {
2308 err = fib_validate_source(skb, saddr, daddr, tos,
2309 net->loopback_dev->ifindex,
2310 dev, &spec_dst, &itag);
2311 if (err < 0)
2312 goto martian_source_keep_err;
2313 if (err)
2314 flags |= RTCF_DIRECTSRC;
2315 spec_dst = daddr;
2316 goto local_input;
2317 }
2318
2319 if (!IN_DEV_FORWARD(in_dev))
2320 goto e_hostunreach;
2321 if (res.type != RTN_UNICAST)
2322 goto martian_destination;
2323
2324 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2325out: return err;
2326
2327brd_input:
2328 if (skb->protocol != htons(ETH_P_IP))
2329 goto e_inval;
2330
2331 if (ipv4_is_zeronet(saddr))
2332 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2333 else {
2334 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2335 &itag);
2336 if (err < 0)
2337 goto martian_source_keep_err;
2338 if (err)
2339 flags |= RTCF_DIRECTSRC;
2340 }
2341 flags |= RTCF_BROADCAST;
2342 res.type = RTN_BROADCAST;
2343 RT_CACHE_STAT_INC(in_brd);
2344
2345local_input:
2346 rth = rt_dst_alloc(net->loopback_dev,
2347 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2348 if (!rth)
2349 goto e_nobufs;
2350
2351 rth->dst.input= ip_local_deliver;
2352 rth->dst.output= ip_rt_bug;
2353#ifdef CONFIG_IP_ROUTE_CLASSID
2354 rth->dst.tclassid = itag;
2355#endif
2356
2357 rth->rt_key_dst = daddr;
2358 rth->rt_key_src = saddr;
2359 rth->rt_genid = rt_genid(net);
2360 rth->rt_flags = flags|RTCF_LOCAL;
2361 rth->rt_type = res.type;
2362 rth->rt_key_tos = tos;
2363 rth->rt_dst = daddr;
2364 rth->rt_src = saddr;
2365#ifdef CONFIG_IP_ROUTE_CLASSID
2366 rth->dst.tclassid = itag;
2367#endif
2368 rth->rt_route_iif = dev->ifindex;
2369 rth->rt_iif = dev->ifindex;
2370 rth->rt_oif = 0;
2371 rth->rt_mark = skb->mark;
2372 rth->rt_gateway = daddr;
2373 rth->rt_spec_dst= spec_dst;
2374 rth->rt_peer_genid = 0;
2375 rth->peer = NULL;
2376 rth->fi = NULL;
2377 if (res.type == RTN_UNREACHABLE) {
2378 rth->dst.input= ip_error;
2379 rth->dst.error= -err;
2380 rth->rt_flags &= ~RTCF_LOCAL;
2381 }
2382 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2383 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2384 err = 0;
2385 if (IS_ERR(rth))
2386 err = PTR_ERR(rth);
2387 goto out;
2388
2389no_route:
2390 RT_CACHE_STAT_INC(in_no_route);
2391 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2392 res.type = RTN_UNREACHABLE;
2393 if (err == -ESRCH)
2394 err = -ENETUNREACH;
2395 goto local_input;
2396
2397 /*
2398 * Do not cache martian addresses: they should be logged (RFC1812)
2399 */
2400martian_destination:
2401 RT_CACHE_STAT_INC(in_martian_dst);
2402#ifdef CONFIG_IP_ROUTE_VERBOSE
2403 if (IN_DEV_LOG_MARTIANS(in_dev))
2404 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2405 &daddr, &saddr, dev->name);
2406#endif
2407
2408e_hostunreach:
2409 err = -EHOSTUNREACH;
2410 goto out;
2411
2412e_inval:
2413 err = -EINVAL;
2414 goto out;
2415
2416e_nobufs:
2417 err = -ENOBUFS;
2418 goto out;
2419
2420martian_source:
2421 err = -EINVAL;
2422martian_source_keep_err:
2423 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2424 goto out;
2425}
2426
2427int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2428 u8 tos, struct net_device *dev, bool noref)
2429{
2430 struct rtable *rth;
2431 unsigned int hash;
2432 int iif = dev->ifindex;
2433 struct net *net;
2434 int res;
2435
2436 net = dev_net(dev);
2437
2438 rcu_read_lock();
2439
2440 if (!rt_caching(net))
2441 goto skip_cache;
2442
2443 tos &= IPTOS_RT_MASK;
2444 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2445
2446 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2447 rth = rcu_dereference(rth->dst.rt_next)) {
2448 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2449 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2450 (rth->rt_route_iif ^ iif) |
2451 (rth->rt_key_tos ^ tos)) == 0 &&
2452 rth->rt_mark == skb->mark &&
2453 net_eq(dev_net(rth->dst.dev), net) &&
2454 !rt_is_expired(rth)) {
2455 ipv4_validate_peer(rth);
2456 if (noref) {
2457 dst_use_noref(&rth->dst, jiffies);
2458 skb_dst_set_noref(skb, &rth->dst);
2459 } else {
2460 dst_use(&rth->dst, jiffies);
2461 skb_dst_set(skb, &rth->dst);
2462 }
2463 RT_CACHE_STAT_INC(in_hit);
2464 rcu_read_unlock();
2465 return 0;
2466 }
2467 RT_CACHE_STAT_INC(in_hlist_search);
2468 }
2469
2470skip_cache:
2471 /* Multicast recognition logic is moved from route cache to here.
2472 The problem was that too many Ethernet cards have broken/missing
2473 hardware multicast filters :-( As result the host on multicasting
2474 network acquires a lot of useless route cache entries, sort of
2475 SDR messages from all the world. Now we try to get rid of them.
2476 Really, provided software IP multicast filter is organized
2477 reasonably (at least, hashed), it does not result in a slowdown
2478 comparing with route cache reject entries.
2479 Note, that multicast routers are not affected, because
2480 route cache entry is created eventually.
2481 */
2482 if (ipv4_is_multicast(daddr)) {
2483 struct in_device *in_dev = __in_dev_get_rcu(dev);
2484
2485 if (in_dev) {
2486 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2487 ip_hdr(skb)->protocol);
2488 if (our
2489#ifdef CONFIG_IP_MROUTE
2490 ||
2491 (!ipv4_is_local_multicast(daddr) &&
2492 IN_DEV_MFORWARD(in_dev))
2493#endif
2494 ) {
2495 int res = ip_route_input_mc(skb, daddr, saddr,
2496 tos, dev, our);
2497 rcu_read_unlock();
2498 return res;
2499 }
2500 }
2501 rcu_read_unlock();
2502 return -EINVAL;
2503 }
2504 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2505 rcu_read_unlock();
2506 return res;
2507}
2508EXPORT_SYMBOL(ip_route_input_common);
2509
2510/* called with rcu_read_lock() */
2511static struct rtable *__mkroute_output(const struct fib_result *res,
2512 const struct flowi4 *fl4,
2513 __be32 orig_daddr, __be32 orig_saddr,
2514 int orig_oif, __u8 orig_rtos,
2515 struct net_device *dev_out,
2516 unsigned int flags)
2517{
2518 struct fib_info *fi = res->fi;
2519 struct in_device *in_dev;
2520 u16 type = res->type;
2521 struct rtable *rth;
2522
2523 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2524 return ERR_PTR(-EINVAL);
2525
2526 if (ipv4_is_lbcast(fl4->daddr))
2527 type = RTN_BROADCAST;
2528 else if (ipv4_is_multicast(fl4->daddr))
2529 type = RTN_MULTICAST;
2530 else if (ipv4_is_zeronet(fl4->daddr))
2531 return ERR_PTR(-EINVAL);
2532
2533 if (dev_out->flags & IFF_LOOPBACK)
2534 flags |= RTCF_LOCAL;
2535
2536 in_dev = __in_dev_get_rcu(dev_out);
2537 if (!in_dev)
2538 return ERR_PTR(-EINVAL);
2539
2540 if (type == RTN_BROADCAST) {
2541 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2542 fi = NULL;
2543 } else if (type == RTN_MULTICAST) {
2544 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2545 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2546 fl4->flowi4_proto))
2547 flags &= ~RTCF_LOCAL;
2548 /* If multicast route do not exist use
2549 * default one, but do not gateway in this case.
2550 * Yes, it is hack.
2551 */
2552 if (fi && res->prefixlen < 4)
2553 fi = NULL;
2554 }
2555
2556 rth = rt_dst_alloc(dev_out,
2557 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2558 IN_DEV_CONF_GET(in_dev, NOXFRM));
2559 if (!rth)
2560 return ERR_PTR(-ENOBUFS);
2561
2562 rth->dst.output = ip_output;
2563
2564 rth->rt_key_dst = orig_daddr;
2565 rth->rt_key_src = orig_saddr;
2566 rth->rt_genid = rt_genid(dev_net(dev_out));
2567 rth->rt_flags = flags;
2568 rth->rt_type = type;
2569 rth->rt_key_tos = orig_rtos;
2570 rth->rt_dst = fl4->daddr;
2571 rth->rt_src = fl4->saddr;
2572 rth->rt_route_iif = 0;
2573 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2574 rth->rt_oif = orig_oif;
2575 rth->rt_mark = fl4->flowi4_mark;
2576 rth->rt_gateway = fl4->daddr;
2577 rth->rt_spec_dst= fl4->saddr;
2578 rth->rt_peer_genid = 0;
2579 rth->peer = NULL;
2580 rth->fi = NULL;
2581
2582 RT_CACHE_STAT_INC(out_slow_tot);
2583
2584 if (flags & RTCF_LOCAL) {
2585 rth->dst.input = ip_local_deliver;
2586 rth->rt_spec_dst = fl4->daddr;
2587 }
2588 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2589 rth->rt_spec_dst = fl4->saddr;
2590 if (flags & RTCF_LOCAL &&
2591 !(dev_out->flags & IFF_LOOPBACK)) {
2592 rth->dst.output = ip_mc_output;
2593 RT_CACHE_STAT_INC(out_slow_mc);
2594 }
2595#ifdef CONFIG_IP_MROUTE
2596 if (type == RTN_MULTICAST) {
2597 if (IN_DEV_MFORWARD(in_dev) &&
2598 !ipv4_is_local_multicast(fl4->daddr)) {
2599 rth->dst.input = ip_mr_input;
2600 rth->dst.output = ip_mc_output;
2601 }
2602 }
2603#endif
2604 }
2605
2606 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2607
2608 return rth;
2609}
2610
2611/*
2612 * Major route resolver routine.
2613 * called with rcu_read_lock();
2614 */
2615
2616static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2617{
2618 struct net_device *dev_out = NULL;
2619 __u8 tos = RT_FL_TOS(fl4);
2620 unsigned int flags = 0;
2621 struct fib_result res;
2622 struct rtable *rth;
2623 __be32 orig_daddr;
2624 __be32 orig_saddr;
2625 int orig_oif;
2626
2627 res.fi = NULL;
2628#ifdef CONFIG_IP_MULTIPLE_TABLES
2629 res.r = NULL;
2630#endif
2631
2632 orig_daddr = fl4->daddr;
2633 orig_saddr = fl4->saddr;
2634 orig_oif = fl4->flowi4_oif;
2635
2636 fl4->flowi4_iif = net->loopback_dev->ifindex;
2637 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2638 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2639 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2640
2641 rcu_read_lock();
2642 if (fl4->saddr) {
2643 rth = ERR_PTR(-EINVAL);
2644 if (ipv4_is_multicast(fl4->saddr) ||
2645 ipv4_is_lbcast(fl4->saddr) ||
2646 ipv4_is_zeronet(fl4->saddr))
2647 goto out;
2648
2649 /* I removed check for oif == dev_out->oif here.
2650 It was wrong for two reasons:
2651 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2652 is assigned to multiple interfaces.
2653 2. Moreover, we are allowed to send packets with saddr
2654 of another iface. --ANK
2655 */
2656
2657 if (fl4->flowi4_oif == 0 &&
2658 (ipv4_is_multicast(fl4->daddr) ||
2659 ipv4_is_lbcast(fl4->daddr))) {
2660 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2661 dev_out = __ip_dev_find(net, fl4->saddr, false);
2662 if (dev_out == NULL)
2663 goto out;
2664
2665 /* Special hack: user can direct multicasts
2666 and limited broadcast via necessary interface
2667 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2668 This hack is not just for fun, it allows
2669 vic,vat and friends to work.
2670 They bind socket to loopback, set ttl to zero
2671 and expect that it will work.
2672 From the viewpoint of routing cache they are broken,
2673 because we are not allowed to build multicast path
2674 with loopback source addr (look, routing cache
2675 cannot know, that ttl is zero, so that packet
2676 will not leave this host and route is valid).
2677 Luckily, this hack is good workaround.
2678 */
2679
2680 fl4->flowi4_oif = dev_out->ifindex;
2681 goto make_route;
2682 }
2683
2684 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2685 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2686 if (!__ip_dev_find(net, fl4->saddr, false))
2687 goto out;
2688 }
2689 }
2690
2691
2692 if (fl4->flowi4_oif) {
2693 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2694 rth = ERR_PTR(-ENODEV);
2695 if (dev_out == NULL)
2696 goto out;
2697
2698 /* RACE: Check return value of inet_select_addr instead. */
2699 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2700 rth = ERR_PTR(-ENETUNREACH);
2701 goto out;
2702 }
2703 if (ipv4_is_local_multicast(fl4->daddr) ||
2704 ipv4_is_lbcast(fl4->daddr)) {
2705 if (!fl4->saddr)
2706 fl4->saddr = inet_select_addr(dev_out, 0,
2707 RT_SCOPE_LINK);
2708 goto make_route;
2709 }
2710 if (fl4->saddr) {
2711 if (ipv4_is_multicast(fl4->daddr))
2712 fl4->saddr = inet_select_addr(dev_out, 0,
2713 fl4->flowi4_scope);
2714 else if (!fl4->daddr)
2715 fl4->saddr = inet_select_addr(dev_out, 0,
2716 RT_SCOPE_HOST);
2717 }
2718 }
2719
2720 if (!fl4->daddr) {
2721 fl4->daddr = fl4->saddr;
2722 if (!fl4->daddr)
2723 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2724 dev_out = net->loopback_dev;
2725 fl4->flowi4_oif = net->loopback_dev->ifindex;
2726 res.type = RTN_LOCAL;
2727 flags |= RTCF_LOCAL;
2728 goto make_route;
2729 }
2730
2731 if (fib_lookup(net, fl4, &res)) {
2732 res.fi = NULL;
2733 if (fl4->flowi4_oif) {
2734 /* Apparently, routing tables are wrong. Assume,
2735 that the destination is on link.
2736
2737 WHY? DW.
2738 Because we are allowed to send to iface
2739 even if it has NO routes and NO assigned
2740 addresses. When oif is specified, routing
2741 tables are looked up with only one purpose:
2742 to catch if destination is gatewayed, rather than
2743 direct. Moreover, if MSG_DONTROUTE is set,
2744 we send packet, ignoring both routing tables
2745 and ifaddr state. --ANK
2746
2747
2748 We could make it even if oif is unknown,
2749 likely IPv6, but we do not.
2750 */
2751
2752 if (fl4->saddr == 0)
2753 fl4->saddr = inet_select_addr(dev_out, 0,
2754 RT_SCOPE_LINK);
2755 res.type = RTN_UNICAST;
2756 goto make_route;
2757 }
2758 rth = ERR_PTR(-ENETUNREACH);
2759 goto out;
2760 }
2761
2762 if (res.type == RTN_LOCAL) {
2763 if (!fl4->saddr) {
2764 if (res.fi->fib_prefsrc)
2765 fl4->saddr = res.fi->fib_prefsrc;
2766 else
2767 fl4->saddr = fl4->daddr;
2768 }
2769 dev_out = net->loopback_dev;
2770 fl4->flowi4_oif = dev_out->ifindex;
2771 res.fi = NULL;
2772 flags |= RTCF_LOCAL;
2773 goto make_route;
2774 }
2775
2776#ifdef CONFIG_IP_ROUTE_MULTIPATH
2777 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2778 fib_select_multipath(&res);
2779 else
2780#endif
2781 if (!res.prefixlen &&
2782 res.table->tb_num_default > 1 &&
2783 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2784 fib_select_default(&res);
2785
2786 if (!fl4->saddr)
2787 fl4->saddr = FIB_RES_PREFSRC(net, res);
2788
2789 dev_out = FIB_RES_DEV(res);
2790 fl4->flowi4_oif = dev_out->ifindex;
2791
2792
2793make_route:
2794 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2795 tos, dev_out, flags);
2796 if (!IS_ERR(rth)) {
2797 unsigned int hash;
2798
2799 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2800 rt_genid(dev_net(dev_out)));
2801 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2802 }
2803
2804out:
2805 rcu_read_unlock();
2806 return rth;
2807}
2808
2809struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2810{
2811 struct rtable *rth;
2812 unsigned int hash;
2813
2814 if (!rt_caching(net))
2815 goto slow_output;
2816
2817 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2818
2819 rcu_read_lock_bh();
2820 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2821 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2822 if (rth->rt_key_dst == flp4->daddr &&
2823 rth->rt_key_src == flp4->saddr &&
2824 rt_is_output_route(rth) &&
2825 rth->rt_oif == flp4->flowi4_oif &&
2826 rth->rt_mark == flp4->flowi4_mark &&
2827 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2828 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2829 net_eq(dev_net(rth->dst.dev), net) &&
2830 !rt_is_expired(rth)) {
2831 ipv4_validate_peer(rth);
2832 dst_use(&rth->dst, jiffies);
2833 RT_CACHE_STAT_INC(out_hit);
2834 rcu_read_unlock_bh();
2835 if (!flp4->saddr)
2836 flp4->saddr = rth->rt_src;
2837 if (!flp4->daddr)
2838 flp4->daddr = rth->rt_dst;
2839 return rth;
2840 }
2841 RT_CACHE_STAT_INC(out_hlist_search);
2842 }
2843 rcu_read_unlock_bh();
2844
2845slow_output:
2846 return ip_route_output_slow(net, flp4);
2847}
2848EXPORT_SYMBOL_GPL(__ip_route_output_key);
2849
2850static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2851{
2852 return NULL;
2853}
2854
2855static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2856{
2857 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2858
2859 return mtu ? : dst->dev->mtu;
2860}
2861
2862static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2863{
2864}
2865
2866static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2867 unsigned long old)
2868{
2869 return NULL;
2870}
2871
2872static struct dst_ops ipv4_dst_blackhole_ops = {
2873 .family = AF_INET,
2874 .protocol = cpu_to_be16(ETH_P_IP),
2875 .destroy = ipv4_dst_destroy,
2876 .check = ipv4_blackhole_dst_check,
2877 .mtu = ipv4_blackhole_mtu,
2878 .default_advmss = ipv4_default_advmss,
2879 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2880 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2881 .neigh_lookup = ipv4_neigh_lookup,
2882};
2883
2884struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2885{
2886 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2887 struct rtable *ort = (struct rtable *) dst_orig;
2888
2889 if (rt) {
2890 struct dst_entry *new = &rt->dst;
2891
2892 new->__use = 1;
2893 new->input = dst_discard;
2894 new->output = dst_discard;
2895 dst_copy_metrics(new, &ort->dst);
2896
2897 new->dev = ort->dst.dev;
2898 if (new->dev)
2899 dev_hold(new->dev);
2900
2901 rt->rt_key_dst = ort->rt_key_dst;
2902 rt->rt_key_src = ort->rt_key_src;
2903 rt->rt_key_tos = ort->rt_key_tos;
2904 rt->rt_route_iif = ort->rt_route_iif;
2905 rt->rt_iif = ort->rt_iif;
2906 rt->rt_oif = ort->rt_oif;
2907 rt->rt_mark = ort->rt_mark;
2908
2909 rt->rt_genid = rt_genid(net);
2910 rt->rt_flags = ort->rt_flags;
2911 rt->rt_type = ort->rt_type;
2912 rt->rt_dst = ort->rt_dst;
2913 rt->rt_src = ort->rt_src;
2914 rt->rt_gateway = ort->rt_gateway;
2915 rt->rt_spec_dst = ort->rt_spec_dst;
2916 rt->peer = ort->peer;
2917 if (rt->peer)
2918 atomic_inc(&rt->peer->refcnt);
2919 rt->fi = ort->fi;
2920 if (rt->fi)
2921 atomic_inc(&rt->fi->fib_clntref);
2922
2923 dst_free(new);
2924 }
2925
2926 dst_release(dst_orig);
2927
2928 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2929}
2930
2931struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2932 struct sock *sk)
2933{
2934 struct rtable *rt = __ip_route_output_key(net, flp4);
2935
2936 if (IS_ERR(rt))
2937 return rt;
2938
2939 if (flp4->flowi4_proto)
2940 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2941 flowi4_to_flowi(flp4),
2942 sk, 0);
2943
2944 return rt;
2945}
2946EXPORT_SYMBOL_GPL(ip_route_output_flow);
2947
2948static int rt_fill_info(struct net *net,
2949 struct sk_buff *skb, u32 pid, u32 seq, int event,
2950 int nowait, unsigned int flags)
2951{
2952 struct rtable *rt = skb_rtable(skb);
2953 struct rtmsg *r;
2954 struct nlmsghdr *nlh;
2955 unsigned long expires = 0;
2956 const struct inet_peer *peer = rt->peer;
2957 u32 id = 0, ts = 0, tsage = 0, error;
2958
2959 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2960 if (nlh == NULL)
2961 return -EMSGSIZE;
2962
2963 r = nlmsg_data(nlh);
2964 r->rtm_family = AF_INET;
2965 r->rtm_dst_len = 32;
2966 r->rtm_src_len = 0;
2967 r->rtm_tos = rt->rt_key_tos;
2968 r->rtm_table = RT_TABLE_MAIN;
2969 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2970 goto nla_put_failure;
2971 r->rtm_type = rt->rt_type;
2972 r->rtm_scope = RT_SCOPE_UNIVERSE;
2973 r->rtm_protocol = RTPROT_UNSPEC;
2974 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2975 if (rt->rt_flags & RTCF_NOTIFY)
2976 r->rtm_flags |= RTM_F_NOTIFY;
2977
2978 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2979 goto nla_put_failure;
2980 if (rt->rt_key_src) {
2981 r->rtm_src_len = 32;
2982 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2983 goto nla_put_failure;
2984 }
2985 if (rt->dst.dev &&
2986 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2987 goto nla_put_failure;
2988#ifdef CONFIG_IP_ROUTE_CLASSID
2989 if (rt->dst.tclassid &&
2990 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2991 goto nla_put_failure;
2992#endif
2993 if (rt_is_input_route(rt)) {
2994 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_spec_dst))
2995 goto nla_put_failure;
2996 } else if (rt->rt_src != rt->rt_key_src) {
2997 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2998 goto nla_put_failure;
2999 }
3000 if (rt->rt_dst != rt->rt_gateway &&
3001 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
3002 goto nla_put_failure;
3003
3004 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
3005 goto nla_put_failure;
3006
3007 if (rt->rt_mark &&
3008 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
3009 goto nla_put_failure;
3010
3011 error = rt->dst.error;
3012 if (peer) {
3013 inet_peer_refcheck(rt->peer);
3014 id = atomic_read(&peer->ip_id_count) & 0xffff;
3015 if (peer->tcp_ts_stamp) {
3016 ts = peer->tcp_ts;
3017 tsage = get_seconds() - peer->tcp_ts_stamp;
3018 }
3019 expires = ACCESS_ONCE(peer->pmtu_expires);
3020 if (expires) {
3021 if (time_before(jiffies, expires))
3022 expires -= jiffies;
3023 else
3024 expires = 0;
3025 }
3026 }
3027
3028 if (rt_is_input_route(rt)) {
3029#ifdef CONFIG_IP_MROUTE
3030 __be32 dst = rt->rt_dst;
3031
3032 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
3033 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3034 int err = ipmr_get_route(net, skb,
3035 rt->rt_src, rt->rt_dst,
3036 r, nowait);
3037 if (err <= 0) {
3038 if (!nowait) {
3039 if (err == 0)
3040 return 0;
3041 goto nla_put_failure;
3042 } else {
3043 if (err == -EMSGSIZE)
3044 goto nla_put_failure;
3045 error = err;
3046 }
3047 }
3048 } else
3049#endif
3050 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
3051 goto nla_put_failure;
3052 }
3053
3054 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
3055 expires, error) < 0)
3056 goto nla_put_failure;
3057
3058 return nlmsg_end(skb, nlh);
3059
3060nla_put_failure:
3061 nlmsg_cancel(skb, nlh);
3062 return -EMSGSIZE;
3063}
3064
3065static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
3066{
3067 struct net *net = sock_net(in_skb->sk);
3068 struct rtmsg *rtm;
3069 struct nlattr *tb[RTA_MAX+1];
3070 struct rtable *rt = NULL;
3071 __be32 dst = 0;
3072 __be32 src = 0;
3073 u32 iif;
3074 int err;
3075 int mark;
3076 struct sk_buff *skb;
3077
3078 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
3079 if (err < 0)
3080 goto errout;
3081
3082 rtm = nlmsg_data(nlh);
3083
3084 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3085 if (skb == NULL) {
3086 err = -ENOBUFS;
3087 goto errout;
3088 }
3089
3090 /* Reserve room for dummy headers, this skb can pass
3091 through good chunk of routing engine.
3092 */
3093 skb_reset_mac_header(skb);
3094 skb_reset_network_header(skb);
3095
3096 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3097 ip_hdr(skb)->protocol = IPPROTO_ICMP;
3098 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
3099
3100 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
3101 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
3102 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3103 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3104
3105 if (iif) {
3106 struct net_device *dev;
3107
3108 dev = __dev_get_by_index(net, iif);
3109 if (dev == NULL) {
3110 err = -ENODEV;
3111 goto errout_free;
3112 }
3113
3114 skb->protocol = htons(ETH_P_IP);
3115 skb->dev = dev;
3116 skb->mark = mark;
3117 local_bh_disable();
3118 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3119 local_bh_enable();
3120
3121 rt = skb_rtable(skb);
3122 if (err == 0 && rt->dst.error)
3123 err = -rt->dst.error;
3124 } else {
3125 struct flowi4 fl4 = {
3126 .daddr = dst,
3127 .saddr = src,
3128 .flowi4_tos = rtm->rtm_tos,
3129 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3130 .flowi4_mark = mark,
3131 };
3132 rt = ip_route_output_key(net, &fl4);
3133
3134 err = 0;
3135 if (IS_ERR(rt))
3136 err = PTR_ERR(rt);
3137 }
3138
3139 if (err)
3140 goto errout_free;
3141
3142 skb_dst_set(skb, &rt->dst);
3143 if (rtm->rtm_flags & RTM_F_NOTIFY)
3144 rt->rt_flags |= RTCF_NOTIFY;
3145
3146 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3147 RTM_NEWROUTE, 0, 0);
3148 if (err <= 0)
3149 goto errout_free;
3150
3151 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3152errout:
3153 return err;
3154
3155errout_free:
3156 kfree_skb(skb);
3157 goto errout;
3158}
3159
3160int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3161{
3162 struct rtable *rt;
3163 int h, s_h;
3164 int idx, s_idx;
3165 struct net *net;
3166
3167 net = sock_net(skb->sk);
3168
3169 s_h = cb->args[0];
3170 if (s_h < 0)
3171 s_h = 0;
3172 s_idx = idx = cb->args[1];
3173 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3174 if (!rt_hash_table[h].chain)
3175 continue;
3176 rcu_read_lock_bh();
3177 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3178 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3179 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3180 continue;
3181 if (rt_is_expired(rt))
3182 continue;
3183 skb_dst_set_noref(skb, &rt->dst);
3184 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3185 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3186 1, NLM_F_MULTI) <= 0) {
3187 skb_dst_drop(skb);
3188 rcu_read_unlock_bh();
3189 goto done;
3190 }
3191 skb_dst_drop(skb);
3192 }
3193 rcu_read_unlock_bh();
3194 }
3195
3196done:
3197 cb->args[0] = h;
3198 cb->args[1] = idx;
3199 return skb->len;
3200}
3201
3202void ip_rt_multicast_event(struct in_device *in_dev)
3203{
3204 rt_cache_flush(dev_net(in_dev->dev), 0);
3205}
3206
3207#ifdef CONFIG_SYSCTL
3208static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3209 void __user *buffer,
3210 size_t *lenp, loff_t *ppos)
3211{
3212 if (write) {
3213 int flush_delay;
3214 ctl_table ctl;
3215 struct net *net;
3216
3217 memcpy(&ctl, __ctl, sizeof(ctl));
3218 ctl.data = &flush_delay;
3219 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3220
3221 net = (struct net *)__ctl->extra1;
3222 rt_cache_flush(net, flush_delay);
3223 return 0;
3224 }
3225
3226 return -EINVAL;
3227}
3228
3229static ctl_table ipv4_route_table[] = {
3230 {
3231 .procname = "gc_thresh",
3232 .data = &ipv4_dst_ops.gc_thresh,
3233 .maxlen = sizeof(int),
3234 .mode = 0644,
3235 .proc_handler = proc_dointvec,
3236 },
3237 {
3238 .procname = "max_size",
3239 .data = &ip_rt_max_size,
3240 .maxlen = sizeof(int),
3241 .mode = 0644,
3242 .proc_handler = proc_dointvec,
3243 },
3244 {
3245 /* Deprecated. Use gc_min_interval_ms */
3246
3247 .procname = "gc_min_interval",
3248 .data = &ip_rt_gc_min_interval,
3249 .maxlen = sizeof(int),
3250 .mode = 0644,
3251 .proc_handler = proc_dointvec_jiffies,
3252 },
3253 {
3254 .procname = "gc_min_interval_ms",
3255 .data = &ip_rt_gc_min_interval,
3256 .maxlen = sizeof(int),
3257 .mode = 0644,
3258 .proc_handler = proc_dointvec_ms_jiffies,
3259 },
3260 {
3261 .procname = "gc_timeout",
3262 .data = &ip_rt_gc_timeout,
3263 .maxlen = sizeof(int),
3264 .mode = 0644,
3265 .proc_handler = proc_dointvec_jiffies,
3266 },
3267 {
3268 .procname = "gc_interval",
3269 .data = &ip_rt_gc_interval,
3270 .maxlen = sizeof(int),
3271 .mode = 0644,
3272 .proc_handler = proc_dointvec_jiffies,
3273 },
3274 {
3275 .procname = "redirect_load",
3276 .data = &ip_rt_redirect_load,
3277 .maxlen = sizeof(int),
3278 .mode = 0644,
3279 .proc_handler = proc_dointvec,
3280 },
3281 {
3282 .procname = "redirect_number",
3283 .data = &ip_rt_redirect_number,
3284 .maxlen = sizeof(int),
3285 .mode = 0644,
3286 .proc_handler = proc_dointvec,
3287 },
3288 {
3289 .procname = "redirect_silence",
3290 .data = &ip_rt_redirect_silence,
3291 .maxlen = sizeof(int),
3292 .mode = 0644,
3293 .proc_handler = proc_dointvec,
3294 },
3295 {
3296 .procname = "error_cost",
3297 .data = &ip_rt_error_cost,
3298 .maxlen = sizeof(int),
3299 .mode = 0644,
3300 .proc_handler = proc_dointvec,
3301 },
3302 {
3303 .procname = "error_burst",
3304 .data = &ip_rt_error_burst,
3305 .maxlen = sizeof(int),
3306 .mode = 0644,
3307 .proc_handler = proc_dointvec,
3308 },
3309 {
3310 .procname = "gc_elasticity",
3311 .data = &ip_rt_gc_elasticity,
3312 .maxlen = sizeof(int),
3313 .mode = 0644,
3314 .proc_handler = proc_dointvec,
3315 },
3316 {
3317 .procname = "mtu_expires",
3318 .data = &ip_rt_mtu_expires,
3319 .maxlen = sizeof(int),
3320 .mode = 0644,
3321 .proc_handler = proc_dointvec_jiffies,
3322 },
3323 {
3324 .procname = "min_pmtu",
3325 .data = &ip_rt_min_pmtu,
3326 .maxlen = sizeof(int),
3327 .mode = 0644,
3328 .proc_handler = proc_dointvec,
3329 },
3330 {
3331 .procname = "min_adv_mss",
3332 .data = &ip_rt_min_advmss,
3333 .maxlen = sizeof(int),
3334 .mode = 0644,
3335 .proc_handler = proc_dointvec,
3336 },
3337 { }
3338};
3339
3340static struct ctl_table ipv4_route_flush_table[] = {
3341 {
3342 .procname = "flush",
3343 .maxlen = sizeof(int),
3344 .mode = 0200,
3345 .proc_handler = ipv4_sysctl_rtcache_flush,
3346 },
3347 { },
3348};
3349
3350static __net_init int sysctl_route_net_init(struct net *net)
3351{
3352 struct ctl_table *tbl;
3353
3354 tbl = ipv4_route_flush_table;
3355 if (!net_eq(net, &init_net)) {
3356 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3357 if (tbl == NULL)
3358 goto err_dup;
3359 }
3360 tbl[0].extra1 = net;
3361
3362 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3363 if (net->ipv4.route_hdr == NULL)
3364 goto err_reg;
3365 return 0;
3366
3367err_reg:
3368 if (tbl != ipv4_route_flush_table)
3369 kfree(tbl);
3370err_dup:
3371 return -ENOMEM;
3372}
3373
3374static __net_exit void sysctl_route_net_exit(struct net *net)
3375{
3376 struct ctl_table *tbl;
3377
3378 tbl = net->ipv4.route_hdr->ctl_table_arg;
3379 unregister_net_sysctl_table(net->ipv4.route_hdr);
3380 BUG_ON(tbl == ipv4_route_flush_table);
3381 kfree(tbl);
3382}
3383
3384static __net_initdata struct pernet_operations sysctl_route_ops = {
3385 .init = sysctl_route_net_init,
3386 .exit = sysctl_route_net_exit,
3387};
3388#endif
3389
3390static __net_init int rt_genid_init(struct net *net)
3391{
3392 get_random_bytes(&net->ipv4.rt_genid,
3393 sizeof(net->ipv4.rt_genid));
3394 get_random_bytes(&net->ipv4.dev_addr_genid,
3395 sizeof(net->ipv4.dev_addr_genid));
3396 return 0;
3397}
3398
3399static __net_initdata struct pernet_operations rt_genid_ops = {
3400 .init = rt_genid_init,
3401};
3402
3403
3404#ifdef CONFIG_IP_ROUTE_CLASSID
3405struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3406#endif /* CONFIG_IP_ROUTE_CLASSID */
3407
3408static __initdata unsigned long rhash_entries;
3409static int __init set_rhash_entries(char *str)
3410{
3411 ssize_t ret;
3412
3413 if (!str)
3414 return 0;
3415
3416 ret = kstrtoul(str, 0, &rhash_entries);
3417 if (ret)
3418 return 0;
3419
3420 return 1;
3421}
3422__setup("rhash_entries=", set_rhash_entries);
3423
3424int __init ip_rt_init(void)
3425{
3426 int rc = 0;
3427
3428#ifdef CONFIG_IP_ROUTE_CLASSID
3429 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3430 if (!ip_rt_acct)
3431 panic("IP: failed to allocate ip_rt_acct\n");
3432#endif
3433
3434 ipv4_dst_ops.kmem_cachep =
3435 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3436 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3437
3438 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3439
3440 if (dst_entries_init(&ipv4_dst_ops) < 0)
3441 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3442
3443 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3444 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3445
3446 rt_hash_table = (struct rt_hash_bucket *)
3447 alloc_large_system_hash("IP route cache",
3448 sizeof(struct rt_hash_bucket),
3449 rhash_entries,
3450 (totalram_pages >= 128 * 1024) ?
3451 15 : 17,
3452 0,
3453 &rt_hash_log,
3454 &rt_hash_mask,
3455 0,
3456 rhash_entries ? 0 : 512 * 1024);
3457 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3458 rt_hash_lock_init();
3459
3460 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3461 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3462
3463 devinet_init();
3464 ip_fib_init();
3465
3466 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3467 expires_ljiffies = jiffies;
3468 schedule_delayed_work(&expires_work,
3469 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3470
3471 if (ip_rt_proc_init())
3472 pr_err("Unable to create route proc files\n");
3473#ifdef CONFIG_XFRM
3474 xfrm_init();
3475 xfrm4_init(ip_rt_max_size);
3476#endif
3477 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3478
3479#ifdef CONFIG_SYSCTL
3480 register_pernet_subsys(&sysctl_route_ops);
3481#endif
3482 register_pernet_subsys(&rt_genid_ops);
3483 return rc;
3484}
3485
3486#ifdef CONFIG_SYSCTL
3487/*
3488 * We really need to sanitize the damn ipv4 init order, then all
3489 * this nonsense will go away.
3490 */
3491void __init ip_static_sysctl_init(void)
3492{
3493 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3494}
3495#endif