Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * ROUTE - implementation of the IP router.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14 *
15 * Fixes:
16 * Alan Cox : Verify area fixes.
17 * Alan Cox : cli() protects routing changes
18 * Rui Oliveira : ICMP routing table updates
19 * (rco@di.uminho.pt) Routing table insertion and update
20 * Linus Torvalds : Rewrote bits to be sensible
21 * Alan Cox : Added BSD route gw semantics
22 * Alan Cox : Super /proc >4K
23 * Alan Cox : MTU in route table
24 * Alan Cox : MSS actually. Also added the window
25 * clamper.
26 * Sam Lantinga : Fixed route matching in rt_del()
27 * Alan Cox : Routing cache support.
28 * Alan Cox : Removed compatibility cruft.
29 * Alan Cox : RTF_REJECT support.
30 * Alan Cox : TCP irtt support.
31 * Jonathan Naylor : Added Metric support.
32 * Miquel van Smoorenburg : BSD API fixes.
33 * Miquel van Smoorenburg : Metrics.
34 * Alan Cox : Use __u32 properly
35 * Alan Cox : Aligned routing errors more closely with BSD
36 * our system is still very different.
37 * Alan Cox : Faster /proc handling
38 * Alexey Kuznetsov : Massive rework to support tree based routing,
39 * routing caches and better behaviour.
40 *
41 * Olaf Erb : irtt wasn't being copied right.
42 * Bjorn Ekwall : Kerneld route support.
43 * Alan Cox : Multicast fixed (I hope)
44 * Pavel Krauz : Limited broadcast fixed
45 * Mike McLagan : Routing by source
46 * Alexey Kuznetsov : End of old history. Split to fib.c and
47 * route.c and rewritten from scratch.
48 * Andi Kleen : Load-limit warning messages.
49 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
50 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
51 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
52 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
53 * Marc Boucher : routing by fwmark
54 * Robert Olsson : Added rt_cache statistics
55 * Arnaldo C. Melo : Convert proc stuff to seq_file
56 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
57 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
58 * Ilia Sotnikov : Removed TOS from hash calculations
59 */
60
61#define pr_fmt(fmt) "IPv4: " fmt
62
63#include <linux/module.h>
64#include <linux/bitops.h>
65#include <linux/kernel.h>
66#include <linux/mm.h>
67#include <linux/memblock.h>
68#include <linux/socket.h>
69#include <linux/errno.h>
70#include <linux/in.h>
71#include <linux/inet.h>
72#include <linux/netdevice.h>
73#include <linux/proc_fs.h>
74#include <linux/init.h>
75#include <linux/skbuff.h>
76#include <linux/inetdevice.h>
77#include <linux/igmp.h>
78#include <linux/pkt_sched.h>
79#include <linux/mroute.h>
80#include <linux/netfilter_ipv4.h>
81#include <linux/random.h>
82#include <linux/rcupdate.h>
83#include <linux/slab.h>
84#include <linux/jhash.h>
85#include <net/dst.h>
86#include <net/dst_metadata.h>
87#include <net/inet_dscp.h>
88#include <net/net_namespace.h>
89#include <net/ip.h>
90#include <net/route.h>
91#include <net/inetpeer.h>
92#include <net/sock.h>
93#include <net/ip_fib.h>
94#include <net/nexthop.h>
95#include <net/tcp.h>
96#include <net/icmp.h>
97#include <net/xfrm.h>
98#include <net/lwtunnel.h>
99#include <net/netevent.h>
100#include <net/rtnetlink.h>
101#ifdef CONFIG_SYSCTL
102#include <linux/sysctl.h>
103#endif
104#include <net/secure_seq.h>
105#include <net/ip_tunnels.h>
106
107#include "fib_lookup.h"
108
109#define RT_GC_TIMEOUT (300*HZ)
110
111#define DEFAULT_MIN_PMTU (512 + 20 + 20)
112#define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
113#define DEFAULT_MIN_ADVMSS 256
114static int ip_rt_max_size;
115static int ip_rt_redirect_number __read_mostly = 9;
116static int ip_rt_redirect_load __read_mostly = HZ / 50;
117static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
118static int ip_rt_error_cost __read_mostly = HZ;
119static int ip_rt_error_burst __read_mostly = 5 * HZ;
120
121static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
122
123/*
124 * Interface to generic destination cache.
125 */
126
127INDIRECT_CALLABLE_SCOPE
128struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
129static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
130INDIRECT_CALLABLE_SCOPE
131unsigned int ipv4_mtu(const struct dst_entry *dst);
132static void ipv4_negative_advice(struct sock *sk,
133 struct dst_entry *dst);
134static void ipv4_link_failure(struct sk_buff *skb);
135static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
136 struct sk_buff *skb, u32 mtu,
137 bool confirm_neigh);
138static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
139 struct sk_buff *skb);
140static void ipv4_dst_destroy(struct dst_entry *dst);
141
142static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
143{
144 WARN_ON(1);
145 return NULL;
146}
147
148static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
149 struct sk_buff *skb,
150 const void *daddr);
151static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
152
153static struct dst_ops ipv4_dst_ops = {
154 .family = AF_INET,
155 .check = ipv4_dst_check,
156 .default_advmss = ipv4_default_advmss,
157 .mtu = ipv4_mtu,
158 .cow_metrics = ipv4_cow_metrics,
159 .destroy = ipv4_dst_destroy,
160 .negative_advice = ipv4_negative_advice,
161 .link_failure = ipv4_link_failure,
162 .update_pmtu = ip_rt_update_pmtu,
163 .redirect = ip_do_redirect,
164 .local_out = __ip_local_out,
165 .neigh_lookup = ipv4_neigh_lookup,
166 .confirm_neigh = ipv4_confirm_neigh,
167};
168
169#define ECN_OR_COST(class) TC_PRIO_##class
170
171const __u8 ip_tos2prio[16] = {
172 TC_PRIO_BESTEFFORT,
173 ECN_OR_COST(BESTEFFORT),
174 TC_PRIO_BESTEFFORT,
175 ECN_OR_COST(BESTEFFORT),
176 TC_PRIO_BULK,
177 ECN_OR_COST(BULK),
178 TC_PRIO_BULK,
179 ECN_OR_COST(BULK),
180 TC_PRIO_INTERACTIVE,
181 ECN_OR_COST(INTERACTIVE),
182 TC_PRIO_INTERACTIVE,
183 ECN_OR_COST(INTERACTIVE),
184 TC_PRIO_INTERACTIVE_BULK,
185 ECN_OR_COST(INTERACTIVE_BULK),
186 TC_PRIO_INTERACTIVE_BULK,
187 ECN_OR_COST(INTERACTIVE_BULK)
188};
189EXPORT_SYMBOL(ip_tos2prio);
190
191static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
192#define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
193
194#ifdef CONFIG_PROC_FS
195static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
196{
197 if (*pos)
198 return NULL;
199 return SEQ_START_TOKEN;
200}
201
202static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
203{
204 ++*pos;
205 return NULL;
206}
207
208static void rt_cache_seq_stop(struct seq_file *seq, void *v)
209{
210}
211
212static int rt_cache_seq_show(struct seq_file *seq, void *v)
213{
214 if (v == SEQ_START_TOKEN)
215 seq_printf(seq, "%-127s\n",
216 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
217 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
218 "HHUptod\tSpecDst");
219 return 0;
220}
221
222static const struct seq_operations rt_cache_seq_ops = {
223 .start = rt_cache_seq_start,
224 .next = rt_cache_seq_next,
225 .stop = rt_cache_seq_stop,
226 .show = rt_cache_seq_show,
227};
228
229static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
230{
231 int cpu;
232
233 if (*pos == 0)
234 return SEQ_START_TOKEN;
235
236 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
237 if (!cpu_possible(cpu))
238 continue;
239 *pos = cpu+1;
240 return &per_cpu(rt_cache_stat, cpu);
241 }
242 return NULL;
243}
244
245static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
246{
247 int cpu;
248
249 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
250 if (!cpu_possible(cpu))
251 continue;
252 *pos = cpu+1;
253 return &per_cpu(rt_cache_stat, cpu);
254 }
255 (*pos)++;
256 return NULL;
257
258}
259
260static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
261{
262
263}
264
265static int rt_cpu_seq_show(struct seq_file *seq, void *v)
266{
267 struct rt_cache_stat *st = v;
268
269 if (v == SEQ_START_TOKEN) {
270 seq_puts(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");
271 return 0;
272 }
273
274 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x "
275 "%08x %08x %08x %08x %08x %08x "
276 "%08x %08x %08x %08x\n",
277 dst_entries_get_slow(&ipv4_dst_ops),
278 0, /* st->in_hit */
279 st->in_slow_tot,
280 st->in_slow_mc,
281 st->in_no_route,
282 st->in_brd,
283 st->in_martian_dst,
284 st->in_martian_src,
285
286 0, /* st->out_hit */
287 st->out_slow_tot,
288 st->out_slow_mc,
289
290 0, /* st->gc_total */
291 0, /* st->gc_ignored */
292 0, /* st->gc_goal_miss */
293 0, /* st->gc_dst_overflow */
294 0, /* st->in_hlist_search */
295 0 /* st->out_hlist_search */
296 );
297 return 0;
298}
299
300static const struct seq_operations rt_cpu_seq_ops = {
301 .start = rt_cpu_seq_start,
302 .next = rt_cpu_seq_next,
303 .stop = rt_cpu_seq_stop,
304 .show = rt_cpu_seq_show,
305};
306
307#ifdef CONFIG_IP_ROUTE_CLASSID
308static int rt_acct_proc_show(struct seq_file *m, void *v)
309{
310 struct ip_rt_acct *dst, *src;
311 unsigned int i, j;
312
313 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
314 if (!dst)
315 return -ENOMEM;
316
317 for_each_possible_cpu(i) {
318 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
319 for (j = 0; j < 256; j++) {
320 dst[j].o_bytes += src[j].o_bytes;
321 dst[j].o_packets += src[j].o_packets;
322 dst[j].i_bytes += src[j].i_bytes;
323 dst[j].i_packets += src[j].i_packets;
324 }
325 }
326
327 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
328 kfree(dst);
329 return 0;
330}
331#endif
332
333static int __net_init ip_rt_do_proc_init(struct net *net)
334{
335 struct proc_dir_entry *pde;
336
337 pde = proc_create_seq("rt_cache", 0444, net->proc_net,
338 &rt_cache_seq_ops);
339 if (!pde)
340 goto err1;
341
342 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
343 &rt_cpu_seq_ops);
344 if (!pde)
345 goto err2;
346
347#ifdef CONFIG_IP_ROUTE_CLASSID
348 pde = proc_create_single("rt_acct", 0, net->proc_net,
349 rt_acct_proc_show);
350 if (!pde)
351 goto err3;
352#endif
353 return 0;
354
355#ifdef CONFIG_IP_ROUTE_CLASSID
356err3:
357 remove_proc_entry("rt_cache", net->proc_net_stat);
358#endif
359err2:
360 remove_proc_entry("rt_cache", net->proc_net);
361err1:
362 return -ENOMEM;
363}
364
365static void __net_exit ip_rt_do_proc_exit(struct net *net)
366{
367 remove_proc_entry("rt_cache", net->proc_net_stat);
368 remove_proc_entry("rt_cache", net->proc_net);
369#ifdef CONFIG_IP_ROUTE_CLASSID
370 remove_proc_entry("rt_acct", net->proc_net);
371#endif
372}
373
374static struct pernet_operations ip_rt_proc_ops __net_initdata = {
375 .init = ip_rt_do_proc_init,
376 .exit = ip_rt_do_proc_exit,
377};
378
379static int __init ip_rt_proc_init(void)
380{
381 return register_pernet_subsys(&ip_rt_proc_ops);
382}
383
384#else
385static inline int ip_rt_proc_init(void)
386{
387 return 0;
388}
389#endif /* CONFIG_PROC_FS */
390
391static inline bool rt_is_expired(const struct rtable *rth)
392{
393 bool res;
394
395 rcu_read_lock();
396 res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev));
397 rcu_read_unlock();
398
399 return res;
400}
401
402void rt_cache_flush(struct net *net)
403{
404 rt_genid_bump_ipv4(net);
405}
406
407static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
408 struct sk_buff *skb,
409 const void *daddr)
410{
411 const struct rtable *rt = container_of(dst, struct rtable, dst);
412 struct net_device *dev = dst->dev;
413 struct neighbour *n;
414
415 rcu_read_lock();
416
417 if (likely(rt->rt_gw_family == AF_INET)) {
418 n = ip_neigh_gw4(dev, rt->rt_gw4);
419 } else if (rt->rt_gw_family == AF_INET6) {
420 n = ip_neigh_gw6(dev, &rt->rt_gw6);
421 } else {
422 __be32 pkey;
423
424 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
425 n = ip_neigh_gw4(dev, pkey);
426 }
427
428 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
429 n = NULL;
430
431 rcu_read_unlock();
432
433 return n;
434}
435
436static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
437{
438 const struct rtable *rt = container_of(dst, struct rtable, dst);
439 struct net_device *dev = dst->dev;
440 const __be32 *pkey = daddr;
441
442 if (rt->rt_gw_family == AF_INET) {
443 pkey = (const __be32 *)&rt->rt_gw4;
444 } else if (rt->rt_gw_family == AF_INET6) {
445 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
446 } else if (!daddr ||
447 (rt->rt_flags &
448 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
449 return;
450 }
451 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
452}
453
454/* Hash tables of size 2048..262144 depending on RAM size.
455 * Each bucket uses 8 bytes.
456 */
457static u32 ip_idents_mask __read_mostly;
458static atomic_t *ip_idents __read_mostly;
459static u32 *ip_tstamps __read_mostly;
460
461/* In order to protect privacy, we add a perturbation to identifiers
462 * if one generator is seldom used. This makes hard for an attacker
463 * to infer how many packets were sent between two points in time.
464 */
465static u32 ip_idents_reserve(u32 hash, int segs)
466{
467 u32 bucket, old, now = (u32)jiffies;
468 atomic_t *p_id;
469 u32 *p_tstamp;
470 u32 delta = 0;
471
472 bucket = hash & ip_idents_mask;
473 p_tstamp = ip_tstamps + bucket;
474 p_id = ip_idents + bucket;
475 old = READ_ONCE(*p_tstamp);
476
477 if (old != now && cmpxchg(p_tstamp, old, now) == old)
478 delta = get_random_u32_below(now - old);
479
480 /* If UBSAN reports an error there, please make sure your compiler
481 * supports -fno-strict-overflow before reporting it that was a bug
482 * in UBSAN, and it has been fixed in GCC-8.
483 */
484 return atomic_add_return(segs + delta, p_id) - segs;
485}
486
487void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
488{
489 u32 hash, id;
490
491 /* Note the following code is not safe, but this is okay. */
492 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
493 get_random_bytes(&net->ipv4.ip_id_key,
494 sizeof(net->ipv4.ip_id_key));
495
496 hash = siphash_3u32((__force u32)iph->daddr,
497 (__force u32)iph->saddr,
498 iph->protocol,
499 &net->ipv4.ip_id_key);
500 id = ip_idents_reserve(hash, segs);
501 iph->id = htons(id);
502}
503EXPORT_SYMBOL(__ip_select_ident);
504
505static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
506 const struct sock *sk, const struct iphdr *iph,
507 int oif, __u8 tos, u8 prot, u32 mark,
508 int flow_flags)
509{
510 __u8 scope = RT_SCOPE_UNIVERSE;
511
512 if (sk) {
513 oif = sk->sk_bound_dev_if;
514 mark = READ_ONCE(sk->sk_mark);
515 tos = ip_sock_rt_tos(sk);
516 scope = ip_sock_rt_scope(sk);
517 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
518 sk->sk_protocol;
519 }
520
521 flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
522 prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
523 sock_net_uid(net, sk));
524}
525
526static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
527 const struct sock *sk)
528{
529 const struct net *net = dev_net(skb->dev);
530 const struct iphdr *iph = ip_hdr(skb);
531 int oif = skb->dev->ifindex;
532 u8 prot = iph->protocol;
533 u32 mark = skb->mark;
534 __u8 tos = iph->tos;
535
536 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
537}
538
539static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
540{
541 const struct inet_sock *inet = inet_sk(sk);
542 const struct ip_options_rcu *inet_opt;
543 __be32 daddr = inet->inet_daddr;
544
545 rcu_read_lock();
546 inet_opt = rcu_dereference(inet->inet_opt);
547 if (inet_opt && inet_opt->opt.srr)
548 daddr = inet_opt->opt.faddr;
549 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
550 ip_sock_rt_tos(sk),
551 ip_sock_rt_scope(sk),
552 inet_test_bit(HDRINCL, sk) ?
553 IPPROTO_RAW : sk->sk_protocol,
554 inet_sk_flowi_flags(sk),
555 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
556 rcu_read_unlock();
557}
558
559static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
560 const struct sk_buff *skb)
561{
562 if (skb)
563 build_skb_flow_key(fl4, skb, sk);
564 else
565 build_sk_flow_key(fl4, sk);
566}
567
568static DEFINE_SPINLOCK(fnhe_lock);
569
570static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
571{
572 struct rtable *rt;
573
574 rt = rcu_dereference(fnhe->fnhe_rth_input);
575 if (rt) {
576 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
577 dst_dev_put(&rt->dst);
578 dst_release(&rt->dst);
579 }
580 rt = rcu_dereference(fnhe->fnhe_rth_output);
581 if (rt) {
582 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
583 dst_dev_put(&rt->dst);
584 dst_release(&rt->dst);
585 }
586}
587
588static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
589{
590 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
591 struct fib_nh_exception *fnhe, *oldest = NULL;
592
593 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
594 fnhe = rcu_dereference_protected(*fnhe_p,
595 lockdep_is_held(&fnhe_lock));
596 if (!fnhe)
597 break;
598 if (!oldest ||
599 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
600 oldest = fnhe;
601 oldest_p = fnhe_p;
602 }
603 }
604 fnhe_flush_routes(oldest);
605 *oldest_p = oldest->fnhe_next;
606 kfree_rcu(oldest, rcu);
607}
608
609static u32 fnhe_hashfun(__be32 daddr)
610{
611 static siphash_aligned_key_t fnhe_hash_key;
612 u64 hval;
613
614 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
615 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
616 return hash_64(hval, FNHE_HASH_SHIFT);
617}
618
619static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
620{
621 rt->rt_pmtu = fnhe->fnhe_pmtu;
622 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
623 rt->dst.expires = fnhe->fnhe_expires;
624
625 if (fnhe->fnhe_gw) {
626 rt->rt_flags |= RTCF_REDIRECTED;
627 rt->rt_uses_gateway = 1;
628 rt->rt_gw_family = AF_INET;
629 rt->rt_gw4 = fnhe->fnhe_gw;
630 }
631}
632
633static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
634 __be32 gw, u32 pmtu, bool lock,
635 unsigned long expires)
636{
637 struct fnhe_hash_bucket *hash;
638 struct fib_nh_exception *fnhe;
639 struct rtable *rt;
640 u32 genid, hval;
641 unsigned int i;
642 int depth;
643
644 genid = fnhe_genid(dev_net(nhc->nhc_dev));
645 hval = fnhe_hashfun(daddr);
646
647 spin_lock_bh(&fnhe_lock);
648
649 hash = rcu_dereference(nhc->nhc_exceptions);
650 if (!hash) {
651 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
652 if (!hash)
653 goto out_unlock;
654 rcu_assign_pointer(nhc->nhc_exceptions, hash);
655 }
656
657 hash += hval;
658
659 depth = 0;
660 for (fnhe = rcu_dereference(hash->chain); fnhe;
661 fnhe = rcu_dereference(fnhe->fnhe_next)) {
662 if (fnhe->fnhe_daddr == daddr)
663 break;
664 depth++;
665 }
666
667 if (fnhe) {
668 if (fnhe->fnhe_genid != genid)
669 fnhe->fnhe_genid = genid;
670 if (gw)
671 fnhe->fnhe_gw = gw;
672 if (pmtu) {
673 fnhe->fnhe_pmtu = pmtu;
674 fnhe->fnhe_mtu_locked = lock;
675 }
676 fnhe->fnhe_expires = max(1UL, expires);
677 /* Update all cached dsts too */
678 rt = rcu_dereference(fnhe->fnhe_rth_input);
679 if (rt)
680 fill_route_from_fnhe(rt, fnhe);
681 rt = rcu_dereference(fnhe->fnhe_rth_output);
682 if (rt)
683 fill_route_from_fnhe(rt, fnhe);
684 } else {
685 /* Randomize max depth to avoid some side channels attacks. */
686 int max_depth = FNHE_RECLAIM_DEPTH +
687 get_random_u32_below(FNHE_RECLAIM_DEPTH);
688
689 while (depth > max_depth) {
690 fnhe_remove_oldest(hash);
691 depth--;
692 }
693
694 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
695 if (!fnhe)
696 goto out_unlock;
697
698 fnhe->fnhe_next = hash->chain;
699
700 fnhe->fnhe_genid = genid;
701 fnhe->fnhe_daddr = daddr;
702 fnhe->fnhe_gw = gw;
703 fnhe->fnhe_pmtu = pmtu;
704 fnhe->fnhe_mtu_locked = lock;
705 fnhe->fnhe_expires = max(1UL, expires);
706
707 rcu_assign_pointer(hash->chain, fnhe);
708
709 /* Exception created; mark the cached routes for the nexthop
710 * stale, so anyone caching it rechecks if this exception
711 * applies to them.
712 */
713 rt = rcu_dereference(nhc->nhc_rth_input);
714 if (rt)
715 rt->dst.obsolete = DST_OBSOLETE_KILL;
716
717 for_each_possible_cpu(i) {
718 struct rtable __rcu **prt;
719
720 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
721 rt = rcu_dereference(*prt);
722 if (rt)
723 rt->dst.obsolete = DST_OBSOLETE_KILL;
724 }
725 }
726
727 fnhe->fnhe_stamp = jiffies;
728
729out_unlock:
730 spin_unlock_bh(&fnhe_lock);
731}
732
733static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
734 bool kill_route)
735{
736 __be32 new_gw = icmp_hdr(skb)->un.gateway;
737 __be32 old_gw = ip_hdr(skb)->saddr;
738 struct net_device *dev = skb->dev;
739 struct in_device *in_dev;
740 struct fib_result res;
741 struct neighbour *n;
742 struct net *net;
743
744 switch (icmp_hdr(skb)->code & 7) {
745 case ICMP_REDIR_NET:
746 case ICMP_REDIR_NETTOS:
747 case ICMP_REDIR_HOST:
748 case ICMP_REDIR_HOSTTOS:
749 break;
750
751 default:
752 return;
753 }
754
755 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
756 return;
757
758 in_dev = __in_dev_get_rcu(dev);
759 if (!in_dev)
760 return;
761
762 net = dev_net(dev);
763 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
764 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
765 ipv4_is_zeronet(new_gw))
766 goto reject_redirect;
767
768 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
769 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
770 goto reject_redirect;
771 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
772 goto reject_redirect;
773 } else {
774 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
775 goto reject_redirect;
776 }
777
778 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
779 if (!n)
780 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
781 if (!IS_ERR(n)) {
782 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
783 neigh_event_send(n, NULL);
784 } else {
785 if (fib_lookup(net, fl4, &res, 0) == 0) {
786 struct fib_nh_common *nhc;
787
788 fib_select_path(net, &res, fl4, skb);
789 nhc = FIB_RES_NHC(res);
790 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
791 0, false,
792 jiffies + ip_rt_gc_timeout);
793 }
794 if (kill_route)
795 rt->dst.obsolete = DST_OBSOLETE_KILL;
796 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
797 }
798 neigh_release(n);
799 }
800 return;
801
802reject_redirect:
803#ifdef CONFIG_IP_ROUTE_VERBOSE
804 if (IN_DEV_LOG_MARTIANS(in_dev)) {
805 const struct iphdr *iph = (const struct iphdr *) skb->data;
806 __be32 daddr = iph->daddr;
807 __be32 saddr = iph->saddr;
808
809 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
810 " Advised path = %pI4 -> %pI4\n",
811 &old_gw, dev->name, &new_gw,
812 &saddr, &daddr);
813 }
814#endif
815 ;
816}
817
818static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
819{
820 struct rtable *rt;
821 struct flowi4 fl4;
822 const struct iphdr *iph = (const struct iphdr *) skb->data;
823 struct net *net = dev_net(skb->dev);
824 int oif = skb->dev->ifindex;
825 u8 prot = iph->protocol;
826 u32 mark = skb->mark;
827 __u8 tos = iph->tos;
828
829 rt = dst_rtable(dst);
830
831 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
832 __ip_do_redirect(rt, skb, &fl4, true);
833}
834
835static void ipv4_negative_advice(struct sock *sk,
836 struct dst_entry *dst)
837{
838 struct rtable *rt = dst_rtable(dst);
839
840 if ((dst->obsolete > 0) ||
841 (rt->rt_flags & RTCF_REDIRECTED) ||
842 rt->dst.expires)
843 sk_dst_reset(sk);
844}
845
846/*
847 * Algorithm:
848 * 1. The first ip_rt_redirect_number redirects are sent
849 * with exponential backoff, then we stop sending them at all,
850 * assuming that the host ignores our redirects.
851 * 2. If we did not see packets requiring redirects
852 * during ip_rt_redirect_silence, we assume that the host
853 * forgot redirected route and start to send redirects again.
854 *
855 * This algorithm is much cheaper and more intelligent than dumb load limiting
856 * in icmp.c.
857 *
858 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
859 * and "frag. need" (breaks PMTU discovery) in icmp.c.
860 */
861
862void ip_rt_send_redirect(struct sk_buff *skb)
863{
864 struct rtable *rt = skb_rtable(skb);
865 struct in_device *in_dev;
866 struct inet_peer *peer;
867 struct net *net;
868 int log_martians;
869 int vif;
870
871 rcu_read_lock();
872 in_dev = __in_dev_get_rcu(rt->dst.dev);
873 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
874 rcu_read_unlock();
875 return;
876 }
877 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
878 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
879
880 net = dev_net(rt->dst.dev);
881 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
882 if (!peer) {
883 rcu_read_unlock();
884 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
885 rt_nexthop(rt, ip_hdr(skb)->daddr));
886 return;
887 }
888
889 /* No redirected packets during ip_rt_redirect_silence;
890 * reset the algorithm.
891 */
892 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
893 peer->rate_tokens = 0;
894 peer->n_redirects = 0;
895 }
896
897 /* Too many ignored redirects; do not send anything
898 * set dst.rate_last to the last seen redirected packet.
899 */
900 if (peer->n_redirects >= ip_rt_redirect_number) {
901 peer->rate_last = jiffies;
902 goto out_unlock;
903 }
904
905 /* Check for load limit; set rate_last to the latest sent
906 * redirect.
907 */
908 if (peer->n_redirects == 0 ||
909 time_after(jiffies,
910 (peer->rate_last +
911 (ip_rt_redirect_load << peer->n_redirects)))) {
912 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
913
914 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
915 peer->rate_last = jiffies;
916 ++peer->n_redirects;
917 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
918 peer->n_redirects == ip_rt_redirect_number)
919 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
920 &ip_hdr(skb)->saddr, inet_iif(skb),
921 &ip_hdr(skb)->daddr, &gw);
922 }
923out_unlock:
924 rcu_read_unlock();
925}
926
927static int ip_error(struct sk_buff *skb)
928{
929 struct rtable *rt = skb_rtable(skb);
930 struct net_device *dev = skb->dev;
931 struct in_device *in_dev;
932 struct inet_peer *peer;
933 unsigned long now;
934 struct net *net;
935 SKB_DR(reason);
936 bool send;
937 int code;
938
939 if (netif_is_l3_master(skb->dev)) {
940 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
941 if (!dev)
942 goto out;
943 }
944
945 in_dev = __in_dev_get_rcu(dev);
946
947 /* IP on this device is disabled. */
948 if (!in_dev)
949 goto out;
950
951 net = dev_net(rt->dst.dev);
952 if (!IN_DEV_FORWARD(in_dev)) {
953 switch (rt->dst.error) {
954 case EHOSTUNREACH:
955 SKB_DR_SET(reason, IP_INADDRERRORS);
956 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
957 break;
958
959 case ENETUNREACH:
960 SKB_DR_SET(reason, IP_INNOROUTES);
961 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
962 break;
963 }
964 goto out;
965 }
966
967 switch (rt->dst.error) {
968 case EINVAL:
969 default:
970 goto out;
971 case EHOSTUNREACH:
972 code = ICMP_HOST_UNREACH;
973 break;
974 case ENETUNREACH:
975 code = ICMP_NET_UNREACH;
976 SKB_DR_SET(reason, IP_INNOROUTES);
977 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
978 break;
979 case EACCES:
980 code = ICMP_PKT_FILTERED;
981 break;
982 }
983
984 rcu_read_lock();
985 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
986 l3mdev_master_ifindex_rcu(skb->dev));
987 send = true;
988 if (peer) {
989 now = jiffies;
990 peer->rate_tokens += now - peer->rate_last;
991 if (peer->rate_tokens > ip_rt_error_burst)
992 peer->rate_tokens = ip_rt_error_burst;
993 peer->rate_last = now;
994 if (peer->rate_tokens >= ip_rt_error_cost)
995 peer->rate_tokens -= ip_rt_error_cost;
996 else
997 send = false;
998 }
999 rcu_read_unlock();
1000
1001 if (send)
1002 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1003
1004out: kfree_skb_reason(skb, reason);
1005 return 0;
1006}
1007
1008static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1009{
1010 struct dst_entry *dst = &rt->dst;
1011 struct fib_result res;
1012 bool lock = false;
1013 struct net *net;
1014 u32 old_mtu;
1015
1016 if (ip_mtu_locked(dst))
1017 return;
1018
1019 old_mtu = ipv4_mtu(dst);
1020 if (old_mtu < mtu)
1021 return;
1022
1023 rcu_read_lock();
1024 net = dev_net_rcu(dst->dev);
1025 if (mtu < net->ipv4.ip_rt_min_pmtu) {
1026 lock = true;
1027 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1028 }
1029
1030 if (rt->rt_pmtu == mtu && !lock &&
1031 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1032 goto out;
1033
1034 if (fib_lookup(net, fl4, &res, 0) == 0) {
1035 struct fib_nh_common *nhc;
1036
1037 fib_select_path(net, &res, fl4, NULL);
1038#ifdef CONFIG_IP_ROUTE_MULTIPATH
1039 if (fib_info_num_path(res.fi) > 1) {
1040 int nhsel;
1041
1042 for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1043 nhc = fib_info_nhc(res.fi, nhsel);
1044 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1045 jiffies + net->ipv4.ip_rt_mtu_expires);
1046 }
1047 goto out;
1048 }
1049#endif /* CONFIG_IP_ROUTE_MULTIPATH */
1050 nhc = FIB_RES_NHC(res);
1051 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1052 jiffies + net->ipv4.ip_rt_mtu_expires);
1053 }
1054out:
1055 rcu_read_unlock();
1056}
1057
1058static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1059 struct sk_buff *skb, u32 mtu,
1060 bool confirm_neigh)
1061{
1062 struct rtable *rt = dst_rtable(dst);
1063 struct flowi4 fl4;
1064
1065 ip_rt_build_flow_key(&fl4, sk, skb);
1066
1067 /* Don't make lookup fail for bridged encapsulations */
1068 if (skb && netif_is_any_bridge_port(skb->dev))
1069 fl4.flowi4_oif = 0;
1070
1071 __ip_rt_update_pmtu(rt, &fl4, mtu);
1072}
1073
1074void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1075 int oif, u8 protocol)
1076{
1077 const struct iphdr *iph = (const struct iphdr *)skb->data;
1078 struct flowi4 fl4;
1079 struct rtable *rt;
1080 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1081
1082 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1083 0);
1084 rt = __ip_route_output_key(net, &fl4);
1085 if (!IS_ERR(rt)) {
1086 __ip_rt_update_pmtu(rt, &fl4, mtu);
1087 ip_rt_put(rt);
1088 }
1089}
1090EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1091
1092static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1093{
1094 const struct iphdr *iph = (const struct iphdr *)skb->data;
1095 struct flowi4 fl4;
1096 struct rtable *rt;
1097
1098 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1099
1100 if (!fl4.flowi4_mark)
1101 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1102
1103 rt = __ip_route_output_key(sock_net(sk), &fl4);
1104 if (!IS_ERR(rt)) {
1105 __ip_rt_update_pmtu(rt, &fl4, mtu);
1106 ip_rt_put(rt);
1107 }
1108}
1109
1110void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1111{
1112 const struct iphdr *iph = (const struct iphdr *)skb->data;
1113 struct flowi4 fl4;
1114 struct rtable *rt;
1115 struct dst_entry *odst = NULL;
1116 bool new = false;
1117 struct net *net = sock_net(sk);
1118
1119 bh_lock_sock(sk);
1120
1121 if (!ip_sk_accept_pmtu(sk))
1122 goto out;
1123
1124 odst = sk_dst_get(sk);
1125
1126 if (sock_owned_by_user(sk) || !odst) {
1127 __ipv4_sk_update_pmtu(skb, sk, mtu);
1128 goto out;
1129 }
1130
1131 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1132
1133 rt = dst_rtable(odst);
1134 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1135 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1136 if (IS_ERR(rt))
1137 goto out;
1138
1139 new = true;
1140 }
1141
1142 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1143
1144 if (!dst_check(&rt->dst, 0)) {
1145 if (new)
1146 dst_release(&rt->dst);
1147
1148 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1149 if (IS_ERR(rt))
1150 goto out;
1151
1152 new = true;
1153 }
1154
1155 if (new)
1156 sk_dst_set(sk, &rt->dst);
1157
1158out:
1159 bh_unlock_sock(sk);
1160 dst_release(odst);
1161}
1162EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1163
1164void ipv4_redirect(struct sk_buff *skb, struct net *net,
1165 int oif, u8 protocol)
1166{
1167 const struct iphdr *iph = (const struct iphdr *)skb->data;
1168 struct flowi4 fl4;
1169 struct rtable *rt;
1170
1171 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1172 rt = __ip_route_output_key(net, &fl4);
1173 if (!IS_ERR(rt)) {
1174 __ip_do_redirect(rt, skb, &fl4, false);
1175 ip_rt_put(rt);
1176 }
1177}
1178EXPORT_SYMBOL_GPL(ipv4_redirect);
1179
1180void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1181{
1182 const struct iphdr *iph = (const struct iphdr *)skb->data;
1183 struct flowi4 fl4;
1184 struct rtable *rt;
1185 struct net *net = sock_net(sk);
1186
1187 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1188 rt = __ip_route_output_key(net, &fl4);
1189 if (!IS_ERR(rt)) {
1190 __ip_do_redirect(rt, skb, &fl4, false);
1191 ip_rt_put(rt);
1192 }
1193}
1194EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1195
1196INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1197 u32 cookie)
1198{
1199 struct rtable *rt = dst_rtable(dst);
1200
1201 /* All IPV4 dsts are created with ->obsolete set to the value
1202 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1203 * into this function always.
1204 *
1205 * When a PMTU/redirect information update invalidates a route,
1206 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1207 * DST_OBSOLETE_DEAD.
1208 */
1209 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1210 return NULL;
1211 return dst;
1212}
1213EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1214
1215static void ipv4_send_dest_unreach(struct sk_buff *skb)
1216{
1217 struct net_device *dev;
1218 struct ip_options opt;
1219 int res;
1220
1221 /* Recompile ip options since IPCB may not be valid anymore.
1222 * Also check we have a reasonable ipv4 header.
1223 */
1224 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1225 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1226 return;
1227
1228 memset(&opt, 0, sizeof(opt));
1229 if (ip_hdr(skb)->ihl > 5) {
1230 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1231 return;
1232 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1233
1234 rcu_read_lock();
1235 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1236 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1237 rcu_read_unlock();
1238
1239 if (res)
1240 return;
1241 }
1242 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1243}
1244
1245static void ipv4_link_failure(struct sk_buff *skb)
1246{
1247 struct rtable *rt;
1248
1249 ipv4_send_dest_unreach(skb);
1250
1251 rt = skb_rtable(skb);
1252 if (rt)
1253 dst_set_expires(&rt->dst, 0);
1254}
1255
1256static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1257{
1258 pr_debug("%s: %pI4 -> %pI4, %s\n",
1259 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1260 skb->dev ? skb->dev->name : "?");
1261 kfree_skb(skb);
1262 WARN_ON(1);
1263 return 0;
1264}
1265
1266/*
1267 * We do not cache source address of outgoing interface,
1268 * because it is used only by IP RR, TS and SRR options,
1269 * so that it out of fast path.
1270 *
1271 * BTW remember: "addr" is allowed to be not aligned
1272 * in IP options!
1273 */
1274
1275void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1276{
1277 __be32 src;
1278
1279 if (rt_is_output_route(rt))
1280 src = ip_hdr(skb)->saddr;
1281 else {
1282 struct fib_result res;
1283 struct iphdr *iph = ip_hdr(skb);
1284 struct flowi4 fl4 = {
1285 .daddr = iph->daddr,
1286 .saddr = iph->saddr,
1287 .flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(iph)),
1288 .flowi4_oif = rt->dst.dev->ifindex,
1289 .flowi4_iif = skb->dev->ifindex,
1290 .flowi4_mark = skb->mark,
1291 };
1292
1293 rcu_read_lock();
1294 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1295 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1296 else
1297 src = inet_select_addr(rt->dst.dev,
1298 rt_nexthop(rt, iph->daddr),
1299 RT_SCOPE_UNIVERSE);
1300 rcu_read_unlock();
1301 }
1302 memcpy(addr, &src, 4);
1303}
1304
1305#ifdef CONFIG_IP_ROUTE_CLASSID
1306static void set_class_tag(struct rtable *rt, u32 tag)
1307{
1308 if (!(rt->dst.tclassid & 0xFFFF))
1309 rt->dst.tclassid |= tag & 0xFFFF;
1310 if (!(rt->dst.tclassid & 0xFFFF0000))
1311 rt->dst.tclassid |= tag & 0xFFFF0000;
1312}
1313#endif
1314
1315static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1316{
1317 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1318 unsigned int advmss;
1319 struct net *net;
1320
1321 rcu_read_lock();
1322 net = dev_net_rcu(dst->dev);
1323 advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1324 net->ipv4.ip_rt_min_advmss);
1325 rcu_read_unlock();
1326
1327 return min(advmss, IPV4_MAX_PMTU - header_size);
1328}
1329
1330INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1331{
1332 return ip_dst_mtu_maybe_forward(dst, false);
1333}
1334EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1335
1336static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1337{
1338 struct fnhe_hash_bucket *hash;
1339 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1340 u32 hval = fnhe_hashfun(daddr);
1341
1342 spin_lock_bh(&fnhe_lock);
1343
1344 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1345 lockdep_is_held(&fnhe_lock));
1346 hash += hval;
1347
1348 fnhe_p = &hash->chain;
1349 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1350 while (fnhe) {
1351 if (fnhe->fnhe_daddr == daddr) {
1352 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1353 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1354 /* set fnhe_daddr to 0 to ensure it won't bind with
1355 * new dsts in rt_bind_exception().
1356 */
1357 fnhe->fnhe_daddr = 0;
1358 fnhe_flush_routes(fnhe);
1359 kfree_rcu(fnhe, rcu);
1360 break;
1361 }
1362 fnhe_p = &fnhe->fnhe_next;
1363 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1364 lockdep_is_held(&fnhe_lock));
1365 }
1366
1367 spin_unlock_bh(&fnhe_lock);
1368}
1369
1370static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1371 __be32 daddr)
1372{
1373 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1374 struct fib_nh_exception *fnhe;
1375 u32 hval;
1376
1377 if (!hash)
1378 return NULL;
1379
1380 hval = fnhe_hashfun(daddr);
1381
1382 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1383 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1384 if (fnhe->fnhe_daddr == daddr) {
1385 if (fnhe->fnhe_expires &&
1386 time_after(jiffies, fnhe->fnhe_expires)) {
1387 ip_del_fnhe(nhc, daddr);
1388 break;
1389 }
1390 return fnhe;
1391 }
1392 }
1393 return NULL;
1394}
1395
1396/* MTU selection:
1397 * 1. mtu on route is locked - use it
1398 * 2. mtu from nexthop exception
1399 * 3. mtu from egress device
1400 */
1401
1402u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1403{
1404 struct fib_nh_common *nhc = res->nhc;
1405 struct net_device *dev = nhc->nhc_dev;
1406 struct fib_info *fi = res->fi;
1407 u32 mtu = 0;
1408
1409 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1410 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1411 mtu = fi->fib_mtu;
1412
1413 if (likely(!mtu)) {
1414 struct fib_nh_exception *fnhe;
1415
1416 fnhe = find_exception(nhc, daddr);
1417 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1418 mtu = fnhe->fnhe_pmtu;
1419 }
1420
1421 if (likely(!mtu))
1422 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1423
1424 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1425}
1426
1427static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1428 __be32 daddr, const bool do_cache)
1429{
1430 bool ret = false;
1431
1432 spin_lock_bh(&fnhe_lock);
1433
1434 if (daddr == fnhe->fnhe_daddr) {
1435 struct rtable __rcu **porig;
1436 struct rtable *orig;
1437 int genid = fnhe_genid(dev_net(rt->dst.dev));
1438
1439 if (rt_is_input_route(rt))
1440 porig = &fnhe->fnhe_rth_input;
1441 else
1442 porig = &fnhe->fnhe_rth_output;
1443 orig = rcu_dereference(*porig);
1444
1445 if (fnhe->fnhe_genid != genid) {
1446 fnhe->fnhe_genid = genid;
1447 fnhe->fnhe_gw = 0;
1448 fnhe->fnhe_pmtu = 0;
1449 fnhe->fnhe_expires = 0;
1450 fnhe->fnhe_mtu_locked = false;
1451 fnhe_flush_routes(fnhe);
1452 orig = NULL;
1453 }
1454 fill_route_from_fnhe(rt, fnhe);
1455 if (!rt->rt_gw4) {
1456 rt->rt_gw4 = daddr;
1457 rt->rt_gw_family = AF_INET;
1458 }
1459
1460 if (do_cache) {
1461 dst_hold(&rt->dst);
1462 rcu_assign_pointer(*porig, rt);
1463 if (orig) {
1464 dst_dev_put(&orig->dst);
1465 dst_release(&orig->dst);
1466 }
1467 ret = true;
1468 }
1469
1470 fnhe->fnhe_stamp = jiffies;
1471 }
1472 spin_unlock_bh(&fnhe_lock);
1473
1474 return ret;
1475}
1476
1477static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1478{
1479 struct rtable *orig, *prev, **p;
1480 bool ret = true;
1481
1482 if (rt_is_input_route(rt)) {
1483 p = (struct rtable **)&nhc->nhc_rth_input;
1484 } else {
1485 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1486 }
1487 orig = *p;
1488
1489 /* hold dst before doing cmpxchg() to avoid race condition
1490 * on this dst
1491 */
1492 dst_hold(&rt->dst);
1493 prev = cmpxchg(p, orig, rt);
1494 if (prev == orig) {
1495 if (orig) {
1496 rt_add_uncached_list(orig);
1497 dst_release(&orig->dst);
1498 }
1499 } else {
1500 dst_release(&rt->dst);
1501 ret = false;
1502 }
1503
1504 return ret;
1505}
1506
1507struct uncached_list {
1508 spinlock_t lock;
1509 struct list_head head;
1510};
1511
1512static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1513
1514void rt_add_uncached_list(struct rtable *rt)
1515{
1516 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1517
1518 rt->dst.rt_uncached_list = ul;
1519
1520 spin_lock_bh(&ul->lock);
1521 list_add_tail(&rt->dst.rt_uncached, &ul->head);
1522 spin_unlock_bh(&ul->lock);
1523}
1524
1525void rt_del_uncached_list(struct rtable *rt)
1526{
1527 if (!list_empty(&rt->dst.rt_uncached)) {
1528 struct uncached_list *ul = rt->dst.rt_uncached_list;
1529
1530 spin_lock_bh(&ul->lock);
1531 list_del_init(&rt->dst.rt_uncached);
1532 spin_unlock_bh(&ul->lock);
1533 }
1534}
1535
1536static void ipv4_dst_destroy(struct dst_entry *dst)
1537{
1538 ip_dst_metrics_put(dst);
1539 rt_del_uncached_list(dst_rtable(dst));
1540}
1541
1542void rt_flush_dev(struct net_device *dev)
1543{
1544 struct rtable *rt, *safe;
1545 int cpu;
1546
1547 for_each_possible_cpu(cpu) {
1548 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1549
1550 if (list_empty(&ul->head))
1551 continue;
1552
1553 spin_lock_bh(&ul->lock);
1554 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1555 if (rt->dst.dev != dev)
1556 continue;
1557 rt->dst.dev = blackhole_netdev;
1558 netdev_ref_replace(dev, blackhole_netdev,
1559 &rt->dst.dev_tracker, GFP_ATOMIC);
1560 list_del_init(&rt->dst.rt_uncached);
1561 }
1562 spin_unlock_bh(&ul->lock);
1563 }
1564}
1565
1566static bool rt_cache_valid(const struct rtable *rt)
1567{
1568 return rt &&
1569 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1570 !rt_is_expired(rt);
1571}
1572
1573static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1574 const struct fib_result *res,
1575 struct fib_nh_exception *fnhe,
1576 struct fib_info *fi, u16 type, u32 itag,
1577 const bool do_cache)
1578{
1579 bool cached = false;
1580
1581 if (fi) {
1582 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1583
1584 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1585 rt->rt_uses_gateway = 1;
1586 rt->rt_gw_family = nhc->nhc_gw_family;
1587 /* only INET and INET6 are supported */
1588 if (likely(nhc->nhc_gw_family == AF_INET))
1589 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1590 else
1591 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1592 }
1593
1594 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1595
1596#ifdef CONFIG_IP_ROUTE_CLASSID
1597 if (nhc->nhc_family == AF_INET) {
1598 struct fib_nh *nh;
1599
1600 nh = container_of(nhc, struct fib_nh, nh_common);
1601 rt->dst.tclassid = nh->nh_tclassid;
1602 }
1603#endif
1604 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1605 if (unlikely(fnhe))
1606 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1607 else if (do_cache)
1608 cached = rt_cache_route(nhc, rt);
1609 if (unlikely(!cached)) {
1610 /* Routes we intend to cache in nexthop exception or
1611 * FIB nexthop have the DST_NOCACHE bit clear.
1612 * However, if we are unsuccessful at storing this
1613 * route into the cache we really need to set it.
1614 */
1615 if (!rt->rt_gw4) {
1616 rt->rt_gw_family = AF_INET;
1617 rt->rt_gw4 = daddr;
1618 }
1619 rt_add_uncached_list(rt);
1620 }
1621 } else
1622 rt_add_uncached_list(rt);
1623
1624#ifdef CONFIG_IP_ROUTE_CLASSID
1625#ifdef CONFIG_IP_MULTIPLE_TABLES
1626 set_class_tag(rt, res->tclassid);
1627#endif
1628 set_class_tag(rt, itag);
1629#endif
1630}
1631
1632struct rtable *rt_dst_alloc(struct net_device *dev,
1633 unsigned int flags, u16 type,
1634 bool noxfrm)
1635{
1636 struct rtable *rt;
1637
1638 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1639 (noxfrm ? DST_NOXFRM : 0));
1640
1641 if (rt) {
1642 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1643 rt->rt_flags = flags;
1644 rt->rt_type = type;
1645 rt->rt_is_input = 0;
1646 rt->rt_iif = 0;
1647 rt->rt_pmtu = 0;
1648 rt->rt_mtu_locked = 0;
1649 rt->rt_uses_gateway = 0;
1650 rt->rt_gw_family = 0;
1651 rt->rt_gw4 = 0;
1652
1653 rt->dst.output = ip_output;
1654 if (flags & RTCF_LOCAL)
1655 rt->dst.input = ip_local_deliver;
1656 }
1657
1658 return rt;
1659}
1660EXPORT_SYMBOL(rt_dst_alloc);
1661
1662struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1663{
1664 struct rtable *new_rt;
1665
1666 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1667 rt->dst.flags);
1668
1669 if (new_rt) {
1670 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1671 new_rt->rt_flags = rt->rt_flags;
1672 new_rt->rt_type = rt->rt_type;
1673 new_rt->rt_is_input = rt->rt_is_input;
1674 new_rt->rt_iif = rt->rt_iif;
1675 new_rt->rt_pmtu = rt->rt_pmtu;
1676 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1677 new_rt->rt_gw_family = rt->rt_gw_family;
1678 if (rt->rt_gw_family == AF_INET)
1679 new_rt->rt_gw4 = rt->rt_gw4;
1680 else if (rt->rt_gw_family == AF_INET6)
1681 new_rt->rt_gw6 = rt->rt_gw6;
1682
1683 new_rt->dst.input = rt->dst.input;
1684 new_rt->dst.output = rt->dst.output;
1685 new_rt->dst.error = rt->dst.error;
1686 new_rt->dst.lastuse = jiffies;
1687 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1688 }
1689 return new_rt;
1690}
1691EXPORT_SYMBOL(rt_dst_clone);
1692
1693/* called in rcu_read_lock() section */
1694enum skb_drop_reason
1695ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1696 dscp_t dscp, struct net_device *dev,
1697 struct in_device *in_dev, u32 *itag)
1698{
1699 enum skb_drop_reason reason;
1700
1701 /* Primary sanity checks. */
1702 if (!in_dev)
1703 return SKB_DROP_REASON_NOT_SPECIFIED;
1704
1705 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1706 return SKB_DROP_REASON_IP_INVALID_SOURCE;
1707
1708 if (skb->protocol != htons(ETH_P_IP))
1709 return SKB_DROP_REASON_INVALID_PROTO;
1710
1711 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1712 return SKB_DROP_REASON_IP_LOCALNET;
1713
1714 if (ipv4_is_zeronet(saddr)) {
1715 if (!ipv4_is_local_multicast(daddr) &&
1716 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1717 return SKB_DROP_REASON_IP_INVALID_SOURCE;
1718 } else {
1719 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1720 dev, in_dev, itag);
1721 if (reason)
1722 return reason;
1723 }
1724 return SKB_NOT_DROPPED_YET;
1725}
1726
1727/* called in rcu_read_lock() section */
1728static enum skb_drop_reason
1729ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1730 dscp_t dscp, struct net_device *dev, int our)
1731{
1732 struct in_device *in_dev = __in_dev_get_rcu(dev);
1733 unsigned int flags = RTCF_MULTICAST;
1734 enum skb_drop_reason reason;
1735 struct rtable *rth;
1736 u32 itag = 0;
1737
1738 reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1739 &itag);
1740 if (reason)
1741 return reason;
1742
1743 if (our)
1744 flags |= RTCF_LOCAL;
1745
1746 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1747 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1748
1749 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1750 false);
1751 if (!rth)
1752 return SKB_DROP_REASON_NOMEM;
1753
1754#ifdef CONFIG_IP_ROUTE_CLASSID
1755 rth->dst.tclassid = itag;
1756#endif
1757 rth->dst.output = ip_rt_bug;
1758 rth->rt_is_input= 1;
1759
1760#ifdef CONFIG_IP_MROUTE
1761 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1762 rth->dst.input = ip_mr_input;
1763#endif
1764 RT_CACHE_STAT_INC(in_slow_mc);
1765
1766 skb_dst_drop(skb);
1767 skb_dst_set(skb, &rth->dst);
1768 return SKB_NOT_DROPPED_YET;
1769}
1770
1771
1772static void ip_handle_martian_source(struct net_device *dev,
1773 struct in_device *in_dev,
1774 struct sk_buff *skb,
1775 __be32 daddr,
1776 __be32 saddr)
1777{
1778 RT_CACHE_STAT_INC(in_martian_src);
1779#ifdef CONFIG_IP_ROUTE_VERBOSE
1780 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1781 /*
1782 * RFC1812 recommendation, if source is martian,
1783 * the only hint is MAC header.
1784 */
1785 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1786 &daddr, &saddr, dev->name);
1787 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1788 print_hex_dump(KERN_WARNING, "ll header: ",
1789 DUMP_PREFIX_OFFSET, 16, 1,
1790 skb_mac_header(skb),
1791 dev->hard_header_len, false);
1792 }
1793 }
1794#endif
1795}
1796
1797/* called in rcu_read_lock() section */
1798static enum skb_drop_reason
1799__mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1800 struct in_device *in_dev, __be32 daddr,
1801 __be32 saddr, dscp_t dscp)
1802{
1803 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1804 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1805 struct net_device *dev = nhc->nhc_dev;
1806 struct fib_nh_exception *fnhe;
1807 struct rtable *rth;
1808 int err;
1809 struct in_device *out_dev;
1810 bool do_cache;
1811 u32 itag = 0;
1812
1813 /* get a working reference to the output device */
1814 out_dev = __in_dev_get_rcu(dev);
1815 if (!out_dev) {
1816 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1817 return reason;
1818 }
1819
1820 err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1821 in_dev->dev, in_dev, &itag);
1822 if (err < 0) {
1823 reason = -err;
1824 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1825 saddr);
1826
1827 goto cleanup;
1828 }
1829
1830 do_cache = res->fi && !itag;
1831 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1832 skb->protocol == htons(ETH_P_IP)) {
1833 __be32 gw;
1834
1835 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1836 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1837 inet_addr_onlink(out_dev, saddr, gw))
1838 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1839 }
1840
1841 if (skb->protocol != htons(ETH_P_IP)) {
1842 /* Not IP (i.e. ARP). Do not create route, if it is
1843 * invalid for proxy arp. DNAT routes are always valid.
1844 *
1845 * Proxy arp feature have been extended to allow, ARP
1846 * replies back to the same interface, to support
1847 * Private VLAN switch technologies. See arp.c.
1848 */
1849 if (out_dev == in_dev &&
1850 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1851 reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1852 goto cleanup;
1853 }
1854 }
1855
1856 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1857 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1858
1859 fnhe = find_exception(nhc, daddr);
1860 if (do_cache) {
1861 if (fnhe)
1862 rth = rcu_dereference(fnhe->fnhe_rth_input);
1863 else
1864 rth = rcu_dereference(nhc->nhc_rth_input);
1865 if (rt_cache_valid(rth)) {
1866 skb_dst_set_noref(skb, &rth->dst);
1867 goto out;
1868 }
1869 }
1870
1871 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1872 IN_DEV_ORCONF(out_dev, NOXFRM));
1873 if (!rth) {
1874 reason = SKB_DROP_REASON_NOMEM;
1875 goto cleanup;
1876 }
1877
1878 rth->rt_is_input = 1;
1879 RT_CACHE_STAT_INC(in_slow_tot);
1880
1881 rth->dst.input = ip_forward;
1882
1883 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1884 do_cache);
1885 lwtunnel_set_redirect(&rth->dst);
1886 skb_dst_set(skb, &rth->dst);
1887out:
1888 reason = SKB_NOT_DROPPED_YET;
1889cleanup:
1890 return reason;
1891}
1892
1893#ifdef CONFIG_IP_ROUTE_MULTIPATH
1894/* To make ICMP packets follow the right flow, the multipath hash is
1895 * calculated from the inner IP addresses.
1896 */
1897static void ip_multipath_l3_keys(const struct sk_buff *skb,
1898 struct flow_keys *hash_keys)
1899{
1900 const struct iphdr *outer_iph = ip_hdr(skb);
1901 const struct iphdr *key_iph = outer_iph;
1902 const struct iphdr *inner_iph;
1903 const struct icmphdr *icmph;
1904 struct iphdr _inner_iph;
1905 struct icmphdr _icmph;
1906
1907 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1908 goto out;
1909
1910 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1911 goto out;
1912
1913 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1914 &_icmph);
1915 if (!icmph)
1916 goto out;
1917
1918 if (!icmp_is_err(icmph->type))
1919 goto out;
1920
1921 inner_iph = skb_header_pointer(skb,
1922 outer_iph->ihl * 4 + sizeof(_icmph),
1923 sizeof(_inner_iph), &_inner_iph);
1924 if (!inner_iph)
1925 goto out;
1926
1927 key_iph = inner_iph;
1928out:
1929 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1930 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1931}
1932
1933static u32 fib_multipath_custom_hash_outer(const struct net *net,
1934 const struct sk_buff *skb,
1935 bool *p_has_inner)
1936{
1937 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1938 struct flow_keys keys, hash_keys;
1939
1940 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1941 return 0;
1942
1943 memset(&hash_keys, 0, sizeof(hash_keys));
1944 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1945
1946 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1947 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1948 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1949 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1950 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1951 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1952 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1953 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1954 hash_keys.ports.src = keys.ports.src;
1955 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1956 hash_keys.ports.dst = keys.ports.dst;
1957
1958 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1959 return fib_multipath_hash_from_keys(net, &hash_keys);
1960}
1961
1962static u32 fib_multipath_custom_hash_inner(const struct net *net,
1963 const struct sk_buff *skb,
1964 bool has_inner)
1965{
1966 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1967 struct flow_keys keys, hash_keys;
1968
1969 /* We assume the packet carries an encapsulation, but if none was
1970 * encountered during dissection of the outer flow, then there is no
1971 * point in calling the flow dissector again.
1972 */
1973 if (!has_inner)
1974 return 0;
1975
1976 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1977 return 0;
1978
1979 memset(&hash_keys, 0, sizeof(hash_keys));
1980 skb_flow_dissect_flow_keys(skb, &keys, 0);
1981
1982 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1983 return 0;
1984
1985 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1986 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1987 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1988 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1989 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1990 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1991 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1992 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1993 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1994 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1995 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1996 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1997 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1998 hash_keys.tags.flow_label = keys.tags.flow_label;
1999 }
2000
2001 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2002 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2003 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2004 hash_keys.ports.src = keys.ports.src;
2005 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2006 hash_keys.ports.dst = keys.ports.dst;
2007
2008 return fib_multipath_hash_from_keys(net, &hash_keys);
2009}
2010
2011static u32 fib_multipath_custom_hash_skb(const struct net *net,
2012 const struct sk_buff *skb)
2013{
2014 u32 mhash, mhash_inner;
2015 bool has_inner = true;
2016
2017 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2018 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2019
2020 return jhash_2words(mhash, mhash_inner, 0);
2021}
2022
2023static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2024 const struct flowi4 *fl4)
2025{
2026 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2027 struct flow_keys hash_keys;
2028
2029 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2030 return 0;
2031
2032 memset(&hash_keys, 0, sizeof(hash_keys));
2033 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2034 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2035 hash_keys.addrs.v4addrs.src = fl4->saddr;
2036 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2037 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2038 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2039 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2040 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2041 hash_keys.ports.src = fl4->fl4_sport;
2042 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2043 hash_keys.ports.dst = fl4->fl4_dport;
2044
2045 return fib_multipath_hash_from_keys(net, &hash_keys);
2046}
2047
2048/* if skb is set it will be used and fl4 can be NULL */
2049int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2050 const struct sk_buff *skb, struct flow_keys *flkeys)
2051{
2052 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2053 struct flow_keys hash_keys;
2054 u32 mhash = 0;
2055
2056 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2057 case 0:
2058 memset(&hash_keys, 0, sizeof(hash_keys));
2059 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2060 if (skb) {
2061 ip_multipath_l3_keys(skb, &hash_keys);
2062 } else {
2063 hash_keys.addrs.v4addrs.src = fl4->saddr;
2064 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2065 }
2066 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2067 break;
2068 case 1:
2069 /* skb is currently provided only when forwarding */
2070 if (skb) {
2071 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2072 struct flow_keys keys;
2073
2074 /* short-circuit if we already have L4 hash present */
2075 if (skb->l4_hash)
2076 return skb_get_hash_raw(skb) >> 1;
2077
2078 memset(&hash_keys, 0, sizeof(hash_keys));
2079
2080 if (!flkeys) {
2081 skb_flow_dissect_flow_keys(skb, &keys, flag);
2082 flkeys = &keys;
2083 }
2084
2085 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2086 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2087 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2088 hash_keys.ports.src = flkeys->ports.src;
2089 hash_keys.ports.dst = flkeys->ports.dst;
2090 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2091 } else {
2092 memset(&hash_keys, 0, sizeof(hash_keys));
2093 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2094 hash_keys.addrs.v4addrs.src = fl4->saddr;
2095 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2096 hash_keys.ports.src = fl4->fl4_sport;
2097 hash_keys.ports.dst = fl4->fl4_dport;
2098 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2099 }
2100 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2101 break;
2102 case 2:
2103 memset(&hash_keys, 0, sizeof(hash_keys));
2104 /* skb is currently provided only when forwarding */
2105 if (skb) {
2106 struct flow_keys keys;
2107
2108 skb_flow_dissect_flow_keys(skb, &keys, 0);
2109 /* Inner can be v4 or v6 */
2110 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2111 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2112 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2113 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2114 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2115 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2116 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2117 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2118 hash_keys.tags.flow_label = keys.tags.flow_label;
2119 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2120 } else {
2121 /* Same as case 0 */
2122 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2123 ip_multipath_l3_keys(skb, &hash_keys);
2124 }
2125 } else {
2126 /* Same as case 0 */
2127 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2128 hash_keys.addrs.v4addrs.src = fl4->saddr;
2129 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2130 }
2131 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2132 break;
2133 case 3:
2134 if (skb)
2135 mhash = fib_multipath_custom_hash_skb(net, skb);
2136 else
2137 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2138 break;
2139 }
2140
2141 if (multipath_hash)
2142 mhash = jhash_2words(mhash, multipath_hash, 0);
2143
2144 return mhash >> 1;
2145}
2146#endif /* CONFIG_IP_ROUTE_MULTIPATH */
2147
2148static enum skb_drop_reason
2149ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2150 struct in_device *in_dev, __be32 daddr,
2151 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2152{
2153#ifdef CONFIG_IP_ROUTE_MULTIPATH
2154 if (res->fi && fib_info_num_path(res->fi) > 1) {
2155 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2156
2157 fib_select_multipath(res, h);
2158 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2159 }
2160#endif
2161
2162 /* create a routing cache entry */
2163 return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2164}
2165
2166/* Implements all the saddr-related checks as ip_route_input_slow(),
2167 * assuming daddr is valid and the destination is not a local broadcast one.
2168 * Uses the provided hint instead of performing a route lookup.
2169 */
2170enum skb_drop_reason
2171ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2172 dscp_t dscp, struct net_device *dev,
2173 const struct sk_buff *hint)
2174{
2175 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2176 struct in_device *in_dev = __in_dev_get_rcu(dev);
2177 struct rtable *rt = skb_rtable(hint);
2178 struct net *net = dev_net(dev);
2179 u32 tag = 0;
2180
2181 if (!in_dev)
2182 return reason;
2183
2184 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2185 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2186 goto martian_source;
2187 }
2188
2189 if (ipv4_is_zeronet(saddr)) {
2190 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2191 goto martian_source;
2192 }
2193
2194 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2195 reason = SKB_DROP_REASON_IP_LOCALNET;
2196 goto martian_source;
2197 }
2198
2199 if (rt->rt_type != RTN_LOCAL)
2200 goto skip_validate_source;
2201
2202 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2203 in_dev, &tag);
2204 if (reason)
2205 goto martian_source;
2206
2207skip_validate_source:
2208 skb_dst_copy(skb, hint);
2209 return SKB_NOT_DROPPED_YET;
2210
2211martian_source:
2212 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2213 return reason;
2214}
2215
2216/* get device for dst_alloc with local routes */
2217static struct net_device *ip_rt_get_dev(struct net *net,
2218 const struct fib_result *res)
2219{
2220 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2221 struct net_device *dev = NULL;
2222
2223 if (nhc)
2224 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2225
2226 return dev ? : net->loopback_dev;
2227}
2228
2229/*
2230 * NOTE. We drop all the packets that has local source
2231 * addresses, because every properly looped back packet
2232 * must have correct destination already attached by output routine.
2233 * Changes in the enforced policies must be applied also to
2234 * ip_route_use_hint().
2235 *
2236 * Such approach solves two big problems:
2237 * 1. Not simplex devices are handled properly.
2238 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2239 * called with rcu_read_lock()
2240 */
2241
2242static enum skb_drop_reason
2243ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2244 dscp_t dscp, struct net_device *dev,
2245 struct fib_result *res)
2246{
2247 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2248 struct in_device *in_dev = __in_dev_get_rcu(dev);
2249 struct flow_keys *flkeys = NULL, _flkeys;
2250 struct net *net = dev_net(dev);
2251 struct ip_tunnel_info *tun_info;
2252 int err = -EINVAL;
2253 unsigned int flags = 0;
2254 u32 itag = 0;
2255 struct rtable *rth;
2256 struct flowi4 fl4;
2257 bool do_cache = true;
2258
2259 /* IP on this device is disabled. */
2260
2261 if (!in_dev)
2262 goto out;
2263
2264 /* Check for the most weird martians, which can be not detected
2265 * by fib_lookup.
2266 */
2267
2268 tun_info = skb_tunnel_info(skb);
2269 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2270 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2271 else
2272 fl4.flowi4_tun_key.tun_id = 0;
2273 skb_dst_drop(skb);
2274
2275 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2276 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2277 goto martian_source;
2278 }
2279
2280 res->fi = NULL;
2281 res->table = NULL;
2282 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2283 goto brd_input;
2284
2285 /* Accept zero addresses only to limited broadcast;
2286 * I even do not know to fix it or not. Waiting for complains :-)
2287 */
2288 if (ipv4_is_zeronet(saddr)) {
2289 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2290 goto martian_source;
2291 }
2292
2293 if (ipv4_is_zeronet(daddr)) {
2294 reason = SKB_DROP_REASON_IP_INVALID_DEST;
2295 goto martian_destination;
2296 }
2297
2298 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2299 * and call it once if daddr or/and saddr are loopback addresses
2300 */
2301 if (ipv4_is_loopback(daddr)) {
2302 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2303 reason = SKB_DROP_REASON_IP_LOCALNET;
2304 goto martian_destination;
2305 }
2306 } else if (ipv4_is_loopback(saddr)) {
2307 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2308 reason = SKB_DROP_REASON_IP_LOCALNET;
2309 goto martian_source;
2310 }
2311 }
2312
2313 /*
2314 * Now we are ready to route packet.
2315 */
2316 fl4.flowi4_l3mdev = 0;
2317 fl4.flowi4_oif = 0;
2318 fl4.flowi4_iif = dev->ifindex;
2319 fl4.flowi4_mark = skb->mark;
2320 fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
2321 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2322 fl4.flowi4_flags = 0;
2323 fl4.daddr = daddr;
2324 fl4.saddr = saddr;
2325 fl4.flowi4_uid = sock_net_uid(net, NULL);
2326 fl4.flowi4_multipath_hash = 0;
2327
2328 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2329 flkeys = &_flkeys;
2330 } else {
2331 fl4.flowi4_proto = 0;
2332 fl4.fl4_sport = 0;
2333 fl4.fl4_dport = 0;
2334 }
2335
2336 err = fib_lookup(net, &fl4, res, 0);
2337 if (err != 0) {
2338 if (!IN_DEV_FORWARD(in_dev))
2339 err = -EHOSTUNREACH;
2340 goto no_route;
2341 }
2342
2343 if (res->type == RTN_BROADCAST) {
2344 if (IN_DEV_BFORWARD(in_dev))
2345 goto make_route;
2346 /* not do cache if bc_forwarding is enabled */
2347 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2348 do_cache = false;
2349 goto brd_input;
2350 }
2351
2352 err = -EINVAL;
2353 if (res->type == RTN_LOCAL) {
2354 reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2355 0, dev, in_dev, &itag);
2356 if (reason)
2357 goto martian_source;
2358 goto local_input;
2359 }
2360
2361 if (!IN_DEV_FORWARD(in_dev)) {
2362 err = -EHOSTUNREACH;
2363 goto no_route;
2364 }
2365 if (res->type != RTN_UNICAST) {
2366 reason = SKB_DROP_REASON_IP_INVALID_DEST;
2367 goto martian_destination;
2368 }
2369
2370make_route:
2371 reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2372 flkeys);
2373
2374out:
2375 return reason;
2376
2377brd_input:
2378 if (skb->protocol != htons(ETH_P_IP)) {
2379 reason = SKB_DROP_REASON_INVALID_PROTO;
2380 goto out;
2381 }
2382
2383 if (!ipv4_is_zeronet(saddr)) {
2384 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2385 dev, in_dev, &itag);
2386 if (reason)
2387 goto martian_source;
2388 }
2389 flags |= RTCF_BROADCAST;
2390 res->type = RTN_BROADCAST;
2391 RT_CACHE_STAT_INC(in_brd);
2392
2393local_input:
2394 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2395 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2396
2397 do_cache &= res->fi && !itag;
2398 if (do_cache) {
2399 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2400
2401 rth = rcu_dereference(nhc->nhc_rth_input);
2402 if (rt_cache_valid(rth)) {
2403 skb_dst_set_noref(skb, &rth->dst);
2404 reason = SKB_NOT_DROPPED_YET;
2405 goto out;
2406 }
2407 }
2408
2409 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2410 flags | RTCF_LOCAL, res->type, false);
2411 if (!rth)
2412 goto e_nobufs;
2413
2414 rth->dst.output= ip_rt_bug;
2415#ifdef CONFIG_IP_ROUTE_CLASSID
2416 rth->dst.tclassid = itag;
2417#endif
2418 rth->rt_is_input = 1;
2419
2420 RT_CACHE_STAT_INC(in_slow_tot);
2421 if (res->type == RTN_UNREACHABLE) {
2422 rth->dst.input= ip_error;
2423 rth->dst.error= -err;
2424 rth->rt_flags &= ~RTCF_LOCAL;
2425 }
2426
2427 if (do_cache) {
2428 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2429
2430 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2431 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2432 WARN_ON(rth->dst.input == lwtunnel_input);
2433 rth->dst.lwtstate->orig_input = rth->dst.input;
2434 rth->dst.input = lwtunnel_input;
2435 }
2436
2437 if (unlikely(!rt_cache_route(nhc, rth)))
2438 rt_add_uncached_list(rth);
2439 }
2440 skb_dst_set(skb, &rth->dst);
2441 reason = SKB_NOT_DROPPED_YET;
2442 goto out;
2443
2444no_route:
2445 RT_CACHE_STAT_INC(in_no_route);
2446 res->type = RTN_UNREACHABLE;
2447 res->fi = NULL;
2448 res->table = NULL;
2449 goto local_input;
2450
2451 /*
2452 * Do not cache martian addresses: they should be logged (RFC1812)
2453 */
2454martian_destination:
2455 RT_CACHE_STAT_INC(in_martian_dst);
2456#ifdef CONFIG_IP_ROUTE_VERBOSE
2457 if (IN_DEV_LOG_MARTIANS(in_dev))
2458 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2459 &daddr, &saddr, dev->name);
2460#endif
2461 goto out;
2462
2463e_nobufs:
2464 reason = SKB_DROP_REASON_NOMEM;
2465 goto out;
2466
2467martian_source:
2468 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2469 goto out;
2470}
2471
2472/* called with rcu_read_lock held */
2473static enum skb_drop_reason
2474ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2475 dscp_t dscp, struct net_device *dev,
2476 struct fib_result *res)
2477{
2478 /* Multicast recognition logic is moved from route cache to here.
2479 * The problem was that too many Ethernet cards have broken/missing
2480 * hardware multicast filters :-( As result the host on multicasting
2481 * network acquires a lot of useless route cache entries, sort of
2482 * SDR messages from all the world. Now we try to get rid of them.
2483 * Really, provided software IP multicast filter is organized
2484 * reasonably (at least, hashed), it does not result in a slowdown
2485 * comparing with route cache reject entries.
2486 * Note, that multicast routers are not affected, because
2487 * route cache entry is created eventually.
2488 */
2489 if (ipv4_is_multicast(daddr)) {
2490 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2491 struct in_device *in_dev = __in_dev_get_rcu(dev);
2492 int our = 0;
2493
2494 if (!in_dev)
2495 return reason;
2496
2497 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2498 ip_hdr(skb)->protocol);
2499
2500 /* check l3 master if no match yet */
2501 if (!our && netif_is_l3_slave(dev)) {
2502 struct in_device *l3_in_dev;
2503
2504 l3_in_dev = __in_dev_get_rcu(skb->dev);
2505 if (l3_in_dev)
2506 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2507 ip_hdr(skb)->protocol);
2508 }
2509
2510 if (our
2511#ifdef CONFIG_IP_MROUTE
2512 ||
2513 (!ipv4_is_local_multicast(daddr) &&
2514 IN_DEV_MFORWARD(in_dev))
2515#endif
2516 ) {
2517 reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2518 dev, our);
2519 }
2520 return reason;
2521 }
2522
2523 return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2524}
2525
2526enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2527 __be32 saddr, dscp_t dscp,
2528 struct net_device *dev)
2529{
2530 enum skb_drop_reason reason;
2531 struct fib_result res;
2532
2533 rcu_read_lock();
2534 reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2535 rcu_read_unlock();
2536
2537 return reason;
2538}
2539EXPORT_SYMBOL(ip_route_input_noref);
2540
2541/* called with rcu_read_lock() */
2542static struct rtable *__mkroute_output(const struct fib_result *res,
2543 const struct flowi4 *fl4, int orig_oif,
2544 struct net_device *dev_out,
2545 unsigned int flags)
2546{
2547 struct fib_info *fi = res->fi;
2548 struct fib_nh_exception *fnhe;
2549 struct in_device *in_dev;
2550 u16 type = res->type;
2551 struct rtable *rth;
2552 bool do_cache;
2553
2554 in_dev = __in_dev_get_rcu(dev_out);
2555 if (!in_dev)
2556 return ERR_PTR(-EINVAL);
2557
2558 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2559 if (ipv4_is_loopback(fl4->saddr) &&
2560 !(dev_out->flags & IFF_LOOPBACK) &&
2561 !netif_is_l3_master(dev_out))
2562 return ERR_PTR(-EINVAL);
2563
2564 if (ipv4_is_lbcast(fl4->daddr))
2565 type = RTN_BROADCAST;
2566 else if (ipv4_is_multicast(fl4->daddr))
2567 type = RTN_MULTICAST;
2568 else if (ipv4_is_zeronet(fl4->daddr))
2569 return ERR_PTR(-EINVAL);
2570
2571 if (dev_out->flags & IFF_LOOPBACK)
2572 flags |= RTCF_LOCAL;
2573
2574 do_cache = true;
2575 if (type == RTN_BROADCAST) {
2576 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2577 fi = NULL;
2578 } else if (type == RTN_MULTICAST) {
2579 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2580 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2581 fl4->flowi4_proto))
2582 flags &= ~RTCF_LOCAL;
2583 else
2584 do_cache = false;
2585 /* If multicast route do not exist use
2586 * default one, but do not gateway in this case.
2587 * Yes, it is hack.
2588 */
2589 if (fi && res->prefixlen < 4)
2590 fi = NULL;
2591 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2592 (orig_oif != dev_out->ifindex)) {
2593 /* For local routes that require a particular output interface
2594 * we do not want to cache the result. Caching the result
2595 * causes incorrect behaviour when there are multiple source
2596 * addresses on the interface, the end result being that if the
2597 * intended recipient is waiting on that interface for the
2598 * packet he won't receive it because it will be delivered on
2599 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2600 * be set to the loopback interface as well.
2601 */
2602 do_cache = false;
2603 }
2604
2605 fnhe = NULL;
2606 do_cache &= fi != NULL;
2607 if (fi) {
2608 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2609 struct rtable __rcu **prth;
2610
2611 fnhe = find_exception(nhc, fl4->daddr);
2612 if (!do_cache)
2613 goto add;
2614 if (fnhe) {
2615 prth = &fnhe->fnhe_rth_output;
2616 } else {
2617 if (unlikely(fl4->flowi4_flags &
2618 FLOWI_FLAG_KNOWN_NH &&
2619 !(nhc->nhc_gw_family &&
2620 nhc->nhc_scope == RT_SCOPE_LINK))) {
2621 do_cache = false;
2622 goto add;
2623 }
2624 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2625 }
2626 rth = rcu_dereference(*prth);
2627 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2628 return rth;
2629 }
2630
2631add:
2632 rth = rt_dst_alloc(dev_out, flags, type,
2633 IN_DEV_ORCONF(in_dev, NOXFRM));
2634 if (!rth)
2635 return ERR_PTR(-ENOBUFS);
2636
2637 rth->rt_iif = orig_oif;
2638
2639 RT_CACHE_STAT_INC(out_slow_tot);
2640
2641 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2642 if (flags & RTCF_LOCAL &&
2643 !(dev_out->flags & IFF_LOOPBACK)) {
2644 rth->dst.output = ip_mc_output;
2645 RT_CACHE_STAT_INC(out_slow_mc);
2646 }
2647#ifdef CONFIG_IP_MROUTE
2648 if (type == RTN_MULTICAST) {
2649 if (IN_DEV_MFORWARD(in_dev) &&
2650 !ipv4_is_local_multicast(fl4->daddr)) {
2651 rth->dst.input = ip_mr_input;
2652 rth->dst.output = ip_mc_output;
2653 }
2654 }
2655#endif
2656 }
2657
2658 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2659 lwtunnel_set_redirect(&rth->dst);
2660
2661 return rth;
2662}
2663
2664/*
2665 * Major route resolver routine.
2666 */
2667
2668struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2669 const struct sk_buff *skb)
2670{
2671 struct fib_result res = {
2672 .type = RTN_UNSPEC,
2673 .fi = NULL,
2674 .table = NULL,
2675 .tclassid = 0,
2676 };
2677 struct rtable *rth;
2678
2679 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2680 fl4->flowi4_tos &= INET_DSCP_MASK;
2681
2682 rcu_read_lock();
2683 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2684 rcu_read_unlock();
2685
2686 return rth;
2687}
2688EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2689
2690struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2691 struct fib_result *res,
2692 const struct sk_buff *skb)
2693{
2694 struct net_device *dev_out = NULL;
2695 int orig_oif = fl4->flowi4_oif;
2696 unsigned int flags = 0;
2697 struct rtable *rth;
2698 int err;
2699
2700 if (fl4->saddr) {
2701 if (ipv4_is_multicast(fl4->saddr) ||
2702 ipv4_is_lbcast(fl4->saddr) ||
2703 ipv4_is_zeronet(fl4->saddr)) {
2704 rth = ERR_PTR(-EINVAL);
2705 goto out;
2706 }
2707
2708 rth = ERR_PTR(-ENETUNREACH);
2709
2710 /* I removed check for oif == dev_out->oif here.
2711 * It was wrong for two reasons:
2712 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2713 * is assigned to multiple interfaces.
2714 * 2. Moreover, we are allowed to send packets with saddr
2715 * of another iface. --ANK
2716 */
2717
2718 if (fl4->flowi4_oif == 0 &&
2719 (ipv4_is_multicast(fl4->daddr) ||
2720 ipv4_is_lbcast(fl4->daddr))) {
2721 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2722 dev_out = __ip_dev_find(net, fl4->saddr, false);
2723 if (!dev_out)
2724 goto out;
2725
2726 /* Special hack: user can direct multicasts
2727 * and limited broadcast via necessary interface
2728 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2729 * This hack is not just for fun, it allows
2730 * vic,vat and friends to work.
2731 * They bind socket to loopback, set ttl to zero
2732 * and expect that it will work.
2733 * From the viewpoint of routing cache they are broken,
2734 * because we are not allowed to build multicast path
2735 * with loopback source addr (look, routing cache
2736 * cannot know, that ttl is zero, so that packet
2737 * will not leave this host and route is valid).
2738 * Luckily, this hack is good workaround.
2739 */
2740
2741 fl4->flowi4_oif = dev_out->ifindex;
2742 goto make_route;
2743 }
2744
2745 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2746 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2747 if (!__ip_dev_find(net, fl4->saddr, false))
2748 goto out;
2749 }
2750 }
2751
2752
2753 if (fl4->flowi4_oif) {
2754 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2755 rth = ERR_PTR(-ENODEV);
2756 if (!dev_out)
2757 goto out;
2758
2759 /* RACE: Check return value of inet_select_addr instead. */
2760 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2761 rth = ERR_PTR(-ENETUNREACH);
2762 goto out;
2763 }
2764 if (ipv4_is_local_multicast(fl4->daddr) ||
2765 ipv4_is_lbcast(fl4->daddr) ||
2766 fl4->flowi4_proto == IPPROTO_IGMP) {
2767 if (!fl4->saddr)
2768 fl4->saddr = inet_select_addr(dev_out, 0,
2769 RT_SCOPE_LINK);
2770 goto make_route;
2771 }
2772 if (!fl4->saddr) {
2773 if (ipv4_is_multicast(fl4->daddr))
2774 fl4->saddr = inet_select_addr(dev_out, 0,
2775 fl4->flowi4_scope);
2776 else if (!fl4->daddr)
2777 fl4->saddr = inet_select_addr(dev_out, 0,
2778 RT_SCOPE_HOST);
2779 }
2780 }
2781
2782 if (!fl4->daddr) {
2783 fl4->daddr = fl4->saddr;
2784 if (!fl4->daddr)
2785 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2786 dev_out = net->loopback_dev;
2787 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2788 res->type = RTN_LOCAL;
2789 flags |= RTCF_LOCAL;
2790 goto make_route;
2791 }
2792
2793 err = fib_lookup(net, fl4, res, 0);
2794 if (err) {
2795 res->fi = NULL;
2796 res->table = NULL;
2797 if (fl4->flowi4_oif &&
2798 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2799 /* Apparently, routing tables are wrong. Assume,
2800 * that the destination is on link.
2801 *
2802 * WHY? DW.
2803 * Because we are allowed to send to iface
2804 * even if it has NO routes and NO assigned
2805 * addresses. When oif is specified, routing
2806 * tables are looked up with only one purpose:
2807 * to catch if destination is gatewayed, rather than
2808 * direct. Moreover, if MSG_DONTROUTE is set,
2809 * we send packet, ignoring both routing tables
2810 * and ifaddr state. --ANK
2811 *
2812 *
2813 * We could make it even if oif is unknown,
2814 * likely IPv6, but we do not.
2815 */
2816
2817 if (fl4->saddr == 0)
2818 fl4->saddr = inet_select_addr(dev_out, 0,
2819 RT_SCOPE_LINK);
2820 res->type = RTN_UNICAST;
2821 goto make_route;
2822 }
2823 rth = ERR_PTR(err);
2824 goto out;
2825 }
2826
2827 if (res->type == RTN_LOCAL) {
2828 if (!fl4->saddr) {
2829 if (res->fi->fib_prefsrc)
2830 fl4->saddr = res->fi->fib_prefsrc;
2831 else
2832 fl4->saddr = fl4->daddr;
2833 }
2834
2835 /* L3 master device is the loopback for that domain */
2836 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2837 net->loopback_dev;
2838
2839 /* make sure orig_oif points to fib result device even
2840 * though packet rx/tx happens over loopback or l3mdev
2841 */
2842 orig_oif = FIB_RES_OIF(*res);
2843
2844 fl4->flowi4_oif = dev_out->ifindex;
2845 flags |= RTCF_LOCAL;
2846 goto make_route;
2847 }
2848
2849 fib_select_path(net, res, fl4, skb);
2850
2851 dev_out = FIB_RES_DEV(*res);
2852
2853make_route:
2854 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2855
2856out:
2857 return rth;
2858}
2859
2860static struct dst_ops ipv4_dst_blackhole_ops = {
2861 .family = AF_INET,
2862 .default_advmss = ipv4_default_advmss,
2863 .neigh_lookup = ipv4_neigh_lookup,
2864 .check = dst_blackhole_check,
2865 .cow_metrics = dst_blackhole_cow_metrics,
2866 .update_pmtu = dst_blackhole_update_pmtu,
2867 .redirect = dst_blackhole_redirect,
2868 .mtu = dst_blackhole_mtu,
2869};
2870
2871struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2872{
2873 struct rtable *ort = dst_rtable(dst_orig);
2874 struct rtable *rt;
2875
2876 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2877 if (rt) {
2878 struct dst_entry *new = &rt->dst;
2879
2880 new->__use = 1;
2881 new->input = dst_discard;
2882 new->output = dst_discard_out;
2883
2884 new->dev = net->loopback_dev;
2885 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2886
2887 rt->rt_is_input = ort->rt_is_input;
2888 rt->rt_iif = ort->rt_iif;
2889 rt->rt_pmtu = ort->rt_pmtu;
2890 rt->rt_mtu_locked = ort->rt_mtu_locked;
2891
2892 rt->rt_genid = rt_genid_ipv4(net);
2893 rt->rt_flags = ort->rt_flags;
2894 rt->rt_type = ort->rt_type;
2895 rt->rt_uses_gateway = ort->rt_uses_gateway;
2896 rt->rt_gw_family = ort->rt_gw_family;
2897 if (rt->rt_gw_family == AF_INET)
2898 rt->rt_gw4 = ort->rt_gw4;
2899 else if (rt->rt_gw_family == AF_INET6)
2900 rt->rt_gw6 = ort->rt_gw6;
2901 }
2902
2903 dst_release(dst_orig);
2904
2905 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2906}
2907
2908struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2909 const struct sock *sk)
2910{
2911 struct rtable *rt = __ip_route_output_key(net, flp4);
2912
2913 if (IS_ERR(rt))
2914 return rt;
2915
2916 if (flp4->flowi4_proto) {
2917 flp4->flowi4_oif = rt->dst.dev->ifindex;
2918 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2919 flowi4_to_flowi(flp4),
2920 sk, 0));
2921 }
2922
2923 return rt;
2924}
2925EXPORT_SYMBOL_GPL(ip_route_output_flow);
2926
2927/* called with rcu_read_lock held */
2928static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2929 struct rtable *rt, u32 table_id, dscp_t dscp,
2930 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2931 u32 seq, unsigned int flags)
2932{
2933 struct rtmsg *r;
2934 struct nlmsghdr *nlh;
2935 unsigned long expires = 0;
2936 u32 error;
2937 u32 metrics[RTAX_MAX];
2938
2939 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2940 if (!nlh)
2941 return -EMSGSIZE;
2942
2943 r = nlmsg_data(nlh);
2944 r->rtm_family = AF_INET;
2945 r->rtm_dst_len = 32;
2946 r->rtm_src_len = 0;
2947 r->rtm_tos = inet_dscp_to_dsfield(dscp);
2948 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2949 if (nla_put_u32(skb, RTA_TABLE, table_id))
2950 goto nla_put_failure;
2951 r->rtm_type = rt->rt_type;
2952 r->rtm_scope = RT_SCOPE_UNIVERSE;
2953 r->rtm_protocol = RTPROT_UNSPEC;
2954 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2955 if (rt->rt_flags & RTCF_NOTIFY)
2956 r->rtm_flags |= RTM_F_NOTIFY;
2957 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2958 r->rtm_flags |= RTCF_DOREDIRECT;
2959
2960 if (nla_put_in_addr(skb, RTA_DST, dst))
2961 goto nla_put_failure;
2962 if (src) {
2963 r->rtm_src_len = 32;
2964 if (nla_put_in_addr(skb, RTA_SRC, src))
2965 goto nla_put_failure;
2966 }
2967 if (rt->dst.dev &&
2968 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2969 goto nla_put_failure;
2970 if (rt->dst.lwtstate &&
2971 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2972 goto nla_put_failure;
2973#ifdef CONFIG_IP_ROUTE_CLASSID
2974 if (rt->dst.tclassid &&
2975 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2976 goto nla_put_failure;
2977#endif
2978 if (fl4 && !rt_is_input_route(rt) &&
2979 fl4->saddr != src) {
2980 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2981 goto nla_put_failure;
2982 }
2983 if (rt->rt_uses_gateway) {
2984 if (rt->rt_gw_family == AF_INET &&
2985 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2986 goto nla_put_failure;
2987 } else if (rt->rt_gw_family == AF_INET6) {
2988 int alen = sizeof(struct in6_addr);
2989 struct nlattr *nla;
2990 struct rtvia *via;
2991
2992 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2993 if (!nla)
2994 goto nla_put_failure;
2995
2996 via = nla_data(nla);
2997 via->rtvia_family = AF_INET6;
2998 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2999 }
3000 }
3001
3002 expires = rt->dst.expires;
3003 if (expires) {
3004 unsigned long now = jiffies;
3005
3006 if (time_before(now, expires))
3007 expires -= now;
3008 else
3009 expires = 0;
3010 }
3011
3012 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3013 if (rt->rt_pmtu && expires)
3014 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3015 if (rt->rt_mtu_locked && expires)
3016 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3017 if (rtnetlink_put_metrics(skb, metrics) < 0)
3018 goto nla_put_failure;
3019
3020 if (fl4) {
3021 if (fl4->flowi4_mark &&
3022 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3023 goto nla_put_failure;
3024
3025 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3026 nla_put_u32(skb, RTA_UID,
3027 from_kuid_munged(current_user_ns(),
3028 fl4->flowi4_uid)))
3029 goto nla_put_failure;
3030
3031 if (rt_is_input_route(rt)) {
3032#ifdef CONFIG_IP_MROUTE
3033 if (ipv4_is_multicast(dst) &&
3034 !ipv4_is_local_multicast(dst) &&
3035 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3036 int err = ipmr_get_route(net, skb,
3037 fl4->saddr, fl4->daddr,
3038 r, portid);
3039
3040 if (err <= 0) {
3041 if (err == 0)
3042 return 0;
3043 goto nla_put_failure;
3044 }
3045 } else
3046#endif
3047 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3048 goto nla_put_failure;
3049 }
3050 }
3051
3052 error = rt->dst.error;
3053
3054 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3055 goto nla_put_failure;
3056
3057 nlmsg_end(skb, nlh);
3058 return 0;
3059
3060nla_put_failure:
3061 nlmsg_cancel(skb, nlh);
3062 return -EMSGSIZE;
3063}
3064
3065static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3066 struct netlink_callback *cb, u32 table_id,
3067 struct fnhe_hash_bucket *bucket, int genid,
3068 int *fa_index, int fa_start, unsigned int flags)
3069{
3070 int i;
3071
3072 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3073 struct fib_nh_exception *fnhe;
3074
3075 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3076 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3077 struct rtable *rt;
3078 int err;
3079
3080 if (*fa_index < fa_start)
3081 goto next;
3082
3083 if (fnhe->fnhe_genid != genid)
3084 goto next;
3085
3086 if (fnhe->fnhe_expires &&
3087 time_after(jiffies, fnhe->fnhe_expires))
3088 goto next;
3089
3090 rt = rcu_dereference(fnhe->fnhe_rth_input);
3091 if (!rt)
3092 rt = rcu_dereference(fnhe->fnhe_rth_output);
3093 if (!rt)
3094 goto next;
3095
3096 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3097 table_id, 0, NULL, skb,
3098 NETLINK_CB(cb->skb).portid,
3099 cb->nlh->nlmsg_seq, flags);
3100 if (err)
3101 return err;
3102next:
3103 (*fa_index)++;
3104 }
3105 }
3106
3107 return 0;
3108}
3109
3110int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3111 u32 table_id, struct fib_info *fi,
3112 int *fa_index, int fa_start, unsigned int flags)
3113{
3114 struct net *net = sock_net(cb->skb->sk);
3115 int nhsel, genid = fnhe_genid(net);
3116
3117 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3118 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3119 struct fnhe_hash_bucket *bucket;
3120 int err;
3121
3122 if (nhc->nhc_flags & RTNH_F_DEAD)
3123 continue;
3124
3125 rcu_read_lock();
3126 bucket = rcu_dereference(nhc->nhc_exceptions);
3127 err = 0;
3128 if (bucket)
3129 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3130 genid, fa_index, fa_start,
3131 flags);
3132 rcu_read_unlock();
3133 if (err)
3134 return err;
3135 }
3136
3137 return 0;
3138}
3139
3140static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3141 u8 ip_proto, __be16 sport,
3142 __be16 dport)
3143{
3144 struct sk_buff *skb;
3145 struct iphdr *iph;
3146
3147 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3148 if (!skb)
3149 return NULL;
3150
3151 /* Reserve room for dummy headers, this skb can pass
3152 * through good chunk of routing engine.
3153 */
3154 skb_reset_mac_header(skb);
3155 skb_reset_network_header(skb);
3156 skb->protocol = htons(ETH_P_IP);
3157 iph = skb_put(skb, sizeof(struct iphdr));
3158 iph->protocol = ip_proto;
3159 iph->saddr = src;
3160 iph->daddr = dst;
3161 iph->version = 0x4;
3162 iph->frag_off = 0;
3163 iph->ihl = 0x5;
3164 skb_set_transport_header(skb, skb->len);
3165
3166 switch (iph->protocol) {
3167 case IPPROTO_UDP: {
3168 struct udphdr *udph;
3169
3170 udph = skb_put_zero(skb, sizeof(struct udphdr));
3171 udph->source = sport;
3172 udph->dest = dport;
3173 udph->len = htons(sizeof(struct udphdr));
3174 udph->check = 0;
3175 break;
3176 }
3177 case IPPROTO_TCP: {
3178 struct tcphdr *tcph;
3179
3180 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3181 tcph->source = sport;
3182 tcph->dest = dport;
3183 tcph->doff = sizeof(struct tcphdr) / 4;
3184 tcph->rst = 1;
3185 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3186 src, dst, 0);
3187 break;
3188 }
3189 case IPPROTO_ICMP: {
3190 struct icmphdr *icmph;
3191
3192 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3193 icmph->type = ICMP_ECHO;
3194 icmph->code = 0;
3195 }
3196 }
3197
3198 return skb;
3199}
3200
3201static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3202 const struct nlmsghdr *nlh,
3203 struct nlattr **tb,
3204 struct netlink_ext_ack *extack)
3205{
3206 struct rtmsg *rtm;
3207 int i, err;
3208
3209 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3210 NL_SET_ERR_MSG(extack,
3211 "ipv4: Invalid header for route get request");
3212 return -EINVAL;
3213 }
3214
3215 if (!netlink_strict_get_check(skb))
3216 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3217 rtm_ipv4_policy, extack);
3218
3219 rtm = nlmsg_data(nlh);
3220 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3221 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3222 rtm->rtm_table || rtm->rtm_protocol ||
3223 rtm->rtm_scope || rtm->rtm_type) {
3224 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3225 return -EINVAL;
3226 }
3227
3228 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3229 RTM_F_LOOKUP_TABLE |
3230 RTM_F_FIB_MATCH)) {
3231 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3232 return -EINVAL;
3233 }
3234
3235 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3236 rtm_ipv4_policy, extack);
3237 if (err)
3238 return err;
3239
3240 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3241 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3242 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3243 return -EINVAL;
3244 }
3245
3246 for (i = 0; i <= RTA_MAX; i++) {
3247 if (!tb[i])
3248 continue;
3249
3250 switch (i) {
3251 case RTA_IIF:
3252 case RTA_OIF:
3253 case RTA_SRC:
3254 case RTA_DST:
3255 case RTA_IP_PROTO:
3256 case RTA_SPORT:
3257 case RTA_DPORT:
3258 case RTA_MARK:
3259 case RTA_UID:
3260 break;
3261 default:
3262 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3263 return -EINVAL;
3264 }
3265 }
3266
3267 return 0;
3268}
3269
3270static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3271 struct netlink_ext_ack *extack)
3272{
3273 struct net *net = sock_net(in_skb->sk);
3274 struct nlattr *tb[RTA_MAX+1];
3275 u32 table_id = RT_TABLE_MAIN;
3276 __be16 sport = 0, dport = 0;
3277 struct fib_result res = {};
3278 u8 ip_proto = IPPROTO_UDP;
3279 struct rtable *rt = NULL;
3280 struct sk_buff *skb;
3281 struct rtmsg *rtm;
3282 struct flowi4 fl4 = {};
3283 __be32 dst = 0;
3284 __be32 src = 0;
3285 kuid_t uid;
3286 u32 iif;
3287 int err;
3288 int mark;
3289
3290 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3291 if (err < 0)
3292 return err;
3293
3294 rtm = nlmsg_data(nlh);
3295 src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3296 dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3297 iif = nla_get_u32_default(tb[RTA_IIF], 0);
3298 mark = nla_get_u32_default(tb[RTA_MARK], 0);
3299 if (tb[RTA_UID])
3300 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3301 else
3302 uid = (iif ? INVALID_UID : current_uid());
3303
3304 if (tb[RTA_IP_PROTO]) {
3305 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3306 &ip_proto, AF_INET, extack);
3307 if (err)
3308 return err;
3309 }
3310
3311 if (tb[RTA_SPORT])
3312 sport = nla_get_be16(tb[RTA_SPORT]);
3313
3314 if (tb[RTA_DPORT])
3315 dport = nla_get_be16(tb[RTA_DPORT]);
3316
3317 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3318 if (!skb)
3319 return -ENOBUFS;
3320
3321 fl4.daddr = dst;
3322 fl4.saddr = src;
3323 fl4.flowi4_tos = rtm->rtm_tos & INET_DSCP_MASK;
3324 fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3325 fl4.flowi4_mark = mark;
3326 fl4.flowi4_uid = uid;
3327 if (sport)
3328 fl4.fl4_sport = sport;
3329 if (dport)
3330 fl4.fl4_dport = dport;
3331 fl4.flowi4_proto = ip_proto;
3332
3333 rcu_read_lock();
3334
3335 if (iif) {
3336 struct net_device *dev;
3337
3338 dev = dev_get_by_index_rcu(net, iif);
3339 if (!dev) {
3340 err = -ENODEV;
3341 goto errout_rcu;
3342 }
3343
3344 fl4.flowi4_iif = iif; /* for rt_fill_info */
3345 skb->dev = dev;
3346 skb->mark = mark;
3347 err = ip_route_input_rcu(skb, dst, src,
3348 inet_dsfield_to_dscp(rtm->rtm_tos),
3349 dev, &res) ? -EINVAL : 0;
3350
3351 rt = skb_rtable(skb);
3352 if (err == 0 && rt->dst.error)
3353 err = -rt->dst.error;
3354 } else {
3355 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3356 skb->dev = net->loopback_dev;
3357 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3358 err = 0;
3359 if (IS_ERR(rt))
3360 err = PTR_ERR(rt);
3361 else
3362 skb_dst_set(skb, &rt->dst);
3363 }
3364
3365 if (err)
3366 goto errout_rcu;
3367
3368 if (rtm->rtm_flags & RTM_F_NOTIFY)
3369 rt->rt_flags |= RTCF_NOTIFY;
3370
3371 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3372 table_id = res.table ? res.table->tb_id : 0;
3373
3374 /* reset skb for netlink reply msg */
3375 skb_trim(skb, 0);
3376 skb_reset_network_header(skb);
3377 skb_reset_transport_header(skb);
3378 skb_reset_mac_header(skb);
3379
3380 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3381 struct fib_rt_info fri;
3382
3383 if (!res.fi) {
3384 err = fib_props[res.type].error;
3385 if (!err)
3386 err = -EHOSTUNREACH;
3387 goto errout_rcu;
3388 }
3389 fri.fi = res.fi;
3390 fri.tb_id = table_id;
3391 fri.dst = res.prefix;
3392 fri.dst_len = res.prefixlen;
3393 fri.dscp = res.dscp;
3394 fri.type = rt->rt_type;
3395 fri.offload = 0;
3396 fri.trap = 0;
3397 fri.offload_failed = 0;
3398 if (res.fa_head) {
3399 struct fib_alias *fa;
3400
3401 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3402 u8 slen = 32 - fri.dst_len;
3403
3404 if (fa->fa_slen == slen &&
3405 fa->tb_id == fri.tb_id &&
3406 fa->fa_dscp == fri.dscp &&
3407 fa->fa_info == res.fi &&
3408 fa->fa_type == fri.type) {
3409 fri.offload = READ_ONCE(fa->offload);
3410 fri.trap = READ_ONCE(fa->trap);
3411 fri.offload_failed =
3412 READ_ONCE(fa->offload_failed);
3413 break;
3414 }
3415 }
3416 }
3417 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3418 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3419 } else {
3420 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3421 skb, NETLINK_CB(in_skb).portid,
3422 nlh->nlmsg_seq, 0);
3423 }
3424 if (err < 0)
3425 goto errout_rcu;
3426
3427 rcu_read_unlock();
3428
3429 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3430
3431errout_free:
3432 return err;
3433errout_rcu:
3434 rcu_read_unlock();
3435 kfree_skb(skb);
3436 goto errout_free;
3437}
3438
3439void ip_rt_multicast_event(struct in_device *in_dev)
3440{
3441 rt_cache_flush(dev_net(in_dev->dev));
3442}
3443
3444#ifdef CONFIG_SYSCTL
3445static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3446static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3447static int ip_rt_gc_elasticity __read_mostly = 8;
3448static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3449
3450static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3451 void *buffer, size_t *lenp, loff_t *ppos)
3452{
3453 struct net *net = (struct net *)__ctl->extra1;
3454
3455 if (write) {
3456 rt_cache_flush(net);
3457 fnhe_genid_bump(net);
3458 return 0;
3459 }
3460
3461 return -EINVAL;
3462}
3463
3464static struct ctl_table ipv4_route_table[] = {
3465 {
3466 .procname = "gc_thresh",
3467 .data = &ipv4_dst_ops.gc_thresh,
3468 .maxlen = sizeof(int),
3469 .mode = 0644,
3470 .proc_handler = proc_dointvec,
3471 },
3472 {
3473 .procname = "max_size",
3474 .data = &ip_rt_max_size,
3475 .maxlen = sizeof(int),
3476 .mode = 0644,
3477 .proc_handler = proc_dointvec,
3478 },
3479 {
3480 /* Deprecated. Use gc_min_interval_ms */
3481
3482 .procname = "gc_min_interval",
3483 .data = &ip_rt_gc_min_interval,
3484 .maxlen = sizeof(int),
3485 .mode = 0644,
3486 .proc_handler = proc_dointvec_jiffies,
3487 },
3488 {
3489 .procname = "gc_min_interval_ms",
3490 .data = &ip_rt_gc_min_interval,
3491 .maxlen = sizeof(int),
3492 .mode = 0644,
3493 .proc_handler = proc_dointvec_ms_jiffies,
3494 },
3495 {
3496 .procname = "gc_timeout",
3497 .data = &ip_rt_gc_timeout,
3498 .maxlen = sizeof(int),
3499 .mode = 0644,
3500 .proc_handler = proc_dointvec_jiffies,
3501 },
3502 {
3503 .procname = "gc_interval",
3504 .data = &ip_rt_gc_interval,
3505 .maxlen = sizeof(int),
3506 .mode = 0644,
3507 .proc_handler = proc_dointvec_jiffies,
3508 },
3509 {
3510 .procname = "redirect_load",
3511 .data = &ip_rt_redirect_load,
3512 .maxlen = sizeof(int),
3513 .mode = 0644,
3514 .proc_handler = proc_dointvec,
3515 },
3516 {
3517 .procname = "redirect_number",
3518 .data = &ip_rt_redirect_number,
3519 .maxlen = sizeof(int),
3520 .mode = 0644,
3521 .proc_handler = proc_dointvec,
3522 },
3523 {
3524 .procname = "redirect_silence",
3525 .data = &ip_rt_redirect_silence,
3526 .maxlen = sizeof(int),
3527 .mode = 0644,
3528 .proc_handler = proc_dointvec,
3529 },
3530 {
3531 .procname = "error_cost",
3532 .data = &ip_rt_error_cost,
3533 .maxlen = sizeof(int),
3534 .mode = 0644,
3535 .proc_handler = proc_dointvec,
3536 },
3537 {
3538 .procname = "error_burst",
3539 .data = &ip_rt_error_burst,
3540 .maxlen = sizeof(int),
3541 .mode = 0644,
3542 .proc_handler = proc_dointvec,
3543 },
3544 {
3545 .procname = "gc_elasticity",
3546 .data = &ip_rt_gc_elasticity,
3547 .maxlen = sizeof(int),
3548 .mode = 0644,
3549 .proc_handler = proc_dointvec,
3550 },
3551};
3552
3553static const char ipv4_route_flush_procname[] = "flush";
3554
3555static struct ctl_table ipv4_route_netns_table[] = {
3556 {
3557 .procname = ipv4_route_flush_procname,
3558 .maxlen = sizeof(int),
3559 .mode = 0200,
3560 .proc_handler = ipv4_sysctl_rtcache_flush,
3561 },
3562 {
3563 .procname = "min_pmtu",
3564 .data = &init_net.ipv4.ip_rt_min_pmtu,
3565 .maxlen = sizeof(int),
3566 .mode = 0644,
3567 .proc_handler = proc_dointvec_minmax,
3568 .extra1 = &ip_min_valid_pmtu,
3569 },
3570 {
3571 .procname = "mtu_expires",
3572 .data = &init_net.ipv4.ip_rt_mtu_expires,
3573 .maxlen = sizeof(int),
3574 .mode = 0644,
3575 .proc_handler = proc_dointvec_jiffies,
3576 },
3577 {
3578 .procname = "min_adv_mss",
3579 .data = &init_net.ipv4.ip_rt_min_advmss,
3580 .maxlen = sizeof(int),
3581 .mode = 0644,
3582 .proc_handler = proc_dointvec,
3583 },
3584};
3585
3586static __net_init int sysctl_route_net_init(struct net *net)
3587{
3588 struct ctl_table *tbl;
3589 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3590
3591 tbl = ipv4_route_netns_table;
3592 if (!net_eq(net, &init_net)) {
3593 int i;
3594
3595 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3596 if (!tbl)
3597 goto err_dup;
3598
3599 /* Don't export non-whitelisted sysctls to unprivileged users */
3600 if (net->user_ns != &init_user_ns) {
3601 if (tbl[0].procname != ipv4_route_flush_procname)
3602 table_size = 0;
3603 }
3604
3605 /* Update the variables to point into the current struct net
3606 * except for the first element flush
3607 */
3608 for (i = 1; i < table_size; i++)
3609 tbl[i].data += (void *)net - (void *)&init_net;
3610 }
3611 tbl[0].extra1 = net;
3612
3613 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3614 tbl, table_size);
3615 if (!net->ipv4.route_hdr)
3616 goto err_reg;
3617 return 0;
3618
3619err_reg:
3620 if (tbl != ipv4_route_netns_table)
3621 kfree(tbl);
3622err_dup:
3623 return -ENOMEM;
3624}
3625
3626static __net_exit void sysctl_route_net_exit(struct net *net)
3627{
3628 const struct ctl_table *tbl;
3629
3630 tbl = net->ipv4.route_hdr->ctl_table_arg;
3631 unregister_net_sysctl_table(net->ipv4.route_hdr);
3632 BUG_ON(tbl == ipv4_route_netns_table);
3633 kfree(tbl);
3634}
3635
3636static __net_initdata struct pernet_operations sysctl_route_ops = {
3637 .init = sysctl_route_net_init,
3638 .exit = sysctl_route_net_exit,
3639};
3640#endif
3641
3642static __net_init int netns_ip_rt_init(struct net *net)
3643{
3644 /* Set default value for namespaceified sysctls */
3645 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3646 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3647 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3648 return 0;
3649}
3650
3651static struct pernet_operations __net_initdata ip_rt_ops = {
3652 .init = netns_ip_rt_init,
3653};
3654
3655static __net_init int rt_genid_init(struct net *net)
3656{
3657 atomic_set(&net->ipv4.rt_genid, 0);
3658 atomic_set(&net->fnhe_genid, 0);
3659 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3660 return 0;
3661}
3662
3663static __net_initdata struct pernet_operations rt_genid_ops = {
3664 .init = rt_genid_init,
3665};
3666
3667static int __net_init ipv4_inetpeer_init(struct net *net)
3668{
3669 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3670
3671 if (!bp)
3672 return -ENOMEM;
3673 inet_peer_base_init(bp);
3674 net->ipv4.peers = bp;
3675 return 0;
3676}
3677
3678static void __net_exit ipv4_inetpeer_exit(struct net *net)
3679{
3680 struct inet_peer_base *bp = net->ipv4.peers;
3681
3682 net->ipv4.peers = NULL;
3683 inetpeer_invalidate_tree(bp);
3684 kfree(bp);
3685}
3686
3687static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3688 .init = ipv4_inetpeer_init,
3689 .exit = ipv4_inetpeer_exit,
3690};
3691
3692#ifdef CONFIG_IP_ROUTE_CLASSID
3693struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3694#endif /* CONFIG_IP_ROUTE_CLASSID */
3695
3696static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3697 {.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3698 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3699};
3700
3701int __init ip_rt_init(void)
3702{
3703 void *idents_hash;
3704 int cpu;
3705
3706 /* For modern hosts, this will use 2 MB of memory */
3707 idents_hash = alloc_large_system_hash("IP idents",
3708 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3709 0,
3710 16, /* one bucket per 64 KB */
3711 HASH_ZERO,
3712 NULL,
3713 &ip_idents_mask,
3714 2048,
3715 256*1024);
3716
3717 ip_idents = idents_hash;
3718
3719 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3720
3721 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3722
3723 for_each_possible_cpu(cpu) {
3724 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3725
3726 INIT_LIST_HEAD(&ul->head);
3727 spin_lock_init(&ul->lock);
3728 }
3729#ifdef CONFIG_IP_ROUTE_CLASSID
3730 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3731 if (!ip_rt_acct)
3732 panic("IP: failed to allocate ip_rt_acct\n");
3733#endif
3734
3735 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3736 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3737
3738 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3739
3740 if (dst_entries_init(&ipv4_dst_ops) < 0)
3741 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3742
3743 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3744 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3745
3746 ipv4_dst_ops.gc_thresh = ~0;
3747 ip_rt_max_size = INT_MAX;
3748
3749 devinet_init();
3750 ip_fib_init();
3751
3752 if (ip_rt_proc_init())
3753 pr_err("Unable to create route proc files\n");
3754#ifdef CONFIG_XFRM
3755 xfrm_init();
3756 xfrm4_init();
3757#endif
3758 rtnl_register_many(ip_rt_rtnl_msg_handlers);
3759
3760#ifdef CONFIG_SYSCTL
3761 register_pernet_subsys(&sysctl_route_ops);
3762#endif
3763 register_pernet_subsys(&ip_rt_ops);
3764 register_pernet_subsys(&rt_genid_ops);
3765 register_pernet_subsys(&ipv4_inetpeer_ops);
3766 return 0;
3767}
3768
3769#ifdef CONFIG_SYSCTL
3770/*
3771 * We really need to sanitize the damn ipv4 init order, then all
3772 * this nonsense will go away.
3773 */
3774void __init ip_static_sysctl_init(void)
3775{
3776 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3777}
3778#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