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
114 static int ip_rt_max_size;
115 static int ip_rt_redirect_number __read_mostly = 9;
116 static int ip_rt_redirect_load __read_mostly = HZ / 50;
117 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
118 static int ip_rt_error_cost __read_mostly = HZ;
119 static int ip_rt_error_burst __read_mostly = 5 * HZ;
120
121 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
122
123 /*
124 * Interface to generic destination cache.
125 */
126
127 INDIRECT_CALLABLE_SCOPE
128 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
129 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
130 INDIRECT_CALLABLE_SCOPE
131 unsigned int ipv4_mtu(const struct dst_entry *dst);
132 static void ipv4_negative_advice(struct sock *sk,
133 struct dst_entry *dst);
134 static void ipv4_link_failure(struct sk_buff *skb);
135 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
136 struct sk_buff *skb, u32 mtu,
137 bool confirm_neigh);
138 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
139 struct sk_buff *skb);
140 static void ipv4_dst_destroy(struct dst_entry *dst);
141
142 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
143 {
144 WARN_ON(1);
145 return NULL;
146 }
147
148 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
149 struct sk_buff *skb,
150 const void *daddr);
151 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
152
153 static 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
171 const __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 };
189 EXPORT_SYMBOL(ip_tos2prio);
190
191 static 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
195 static 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
202 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
203 {
204 ++*pos;
205 return NULL;
206 }
207
208 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
209 {
210 }
211
212 static 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
222 static 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
229 static 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
245 static 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
260 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
261 {
262
263 }
264
265 static 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
300 static 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
308 static 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
333 static 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
356 err3:
357 remove_proc_entry("rt_cache", net->proc_net_stat);
358 #endif
359 err2:
360 remove_proc_entry("rt_cache", net->proc_net);
361 err1:
362 return -ENOMEM;
363 }
364
365 static 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
374 static 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
379 static int __init ip_rt_proc_init(void)
380 {
381 return register_pernet_subsys(&ip_rt_proc_ops);
382 }
383
384 #else
385 static inline int ip_rt_proc_init(void)
386 {
387 return 0;
388 }
389 #endif /* CONFIG_PROC_FS */
390
391 static inline bool rt_is_expired(const struct rtable *rth)
392 {
393 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
394 }
395
396 void rt_cache_flush(struct net *net)
397 {
398 rt_genid_bump_ipv4(net);
399 }
400
401 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
402 struct sk_buff *skb,
403 const void *daddr)
404 {
405 const struct rtable *rt = container_of(dst, struct rtable, dst);
406 struct net_device *dev = dst->dev;
407 struct neighbour *n;
408
409 rcu_read_lock();
410
411 if (likely(rt->rt_gw_family == AF_INET)) {
412 n = ip_neigh_gw4(dev, rt->rt_gw4);
413 } else if (rt->rt_gw_family == AF_INET6) {
414 n = ip_neigh_gw6(dev, &rt->rt_gw6);
415 } else {
416 __be32 pkey;
417
418 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
419 n = ip_neigh_gw4(dev, pkey);
420 }
421
422 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
423 n = NULL;
424
425 rcu_read_unlock();
426
427 return n;
428 }
429
430 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
431 {
432 const struct rtable *rt = container_of(dst, struct rtable, dst);
433 struct net_device *dev = dst->dev;
434 const __be32 *pkey = daddr;
435
436 if (rt->rt_gw_family == AF_INET) {
437 pkey = (const __be32 *)&rt->rt_gw4;
438 } else if (rt->rt_gw_family == AF_INET6) {
439 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
440 } else if (!daddr ||
441 (rt->rt_flags &
442 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
443 return;
444 }
445 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
446 }
447
448 /* Hash tables of size 2048..262144 depending on RAM size.
449 * Each bucket uses 8 bytes.
450 */
451 static u32 ip_idents_mask __read_mostly;
452 static atomic_t *ip_idents __read_mostly;
453 static u32 *ip_tstamps __read_mostly;
454
455 /* In order to protect privacy, we add a perturbation to identifiers
456 * if one generator is seldom used. This makes hard for an attacker
457 * to infer how many packets were sent between two points in time.
458 */
459 static u32 ip_idents_reserve(u32 hash, int segs)
460 {
461 u32 bucket, old, now = (u32)jiffies;
462 atomic_t *p_id;
463 u32 *p_tstamp;
464 u32 delta = 0;
465
466 bucket = hash & ip_idents_mask;
467 p_tstamp = ip_tstamps + bucket;
468 p_id = ip_idents + bucket;
469 old = READ_ONCE(*p_tstamp);
470
471 if (old != now && cmpxchg(p_tstamp, old, now) == old)
472 delta = get_random_u32_below(now - old);
473
474 /* If UBSAN reports an error there, please make sure your compiler
475 * supports -fno-strict-overflow before reporting it that was a bug
476 * in UBSAN, and it has been fixed in GCC-8.
477 */
478 return atomic_add_return(segs + delta, p_id) - segs;
479 }
480
481 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
482 {
483 u32 hash, id;
484
485 /* Note the following code is not safe, but this is okay. */
486 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
487 get_random_bytes(&net->ipv4.ip_id_key,
488 sizeof(net->ipv4.ip_id_key));
489
490 hash = siphash_3u32((__force u32)iph->daddr,
491 (__force u32)iph->saddr,
492 iph->protocol,
493 &net->ipv4.ip_id_key);
494 id = ip_idents_reserve(hash, segs);
495 iph->id = htons(id);
496 }
497 EXPORT_SYMBOL(__ip_select_ident);
498
499 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
500 const struct sock *sk, const struct iphdr *iph,
501 int oif, __u8 tos, u8 prot, u32 mark,
502 int flow_flags)
503 {
504 __u8 scope = RT_SCOPE_UNIVERSE;
505
506 if (sk) {
507 oif = sk->sk_bound_dev_if;
508 mark = READ_ONCE(sk->sk_mark);
509 tos = ip_sock_rt_tos(sk);
510 scope = ip_sock_rt_scope(sk);
511 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
512 sk->sk_protocol;
513 }
514
515 flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
516 prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
517 sock_net_uid(net, sk));
518 }
519
520 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
521 const struct sock *sk)
522 {
523 const struct net *net = dev_net(skb->dev);
524 const struct iphdr *iph = ip_hdr(skb);
525 int oif = skb->dev->ifindex;
526 u8 prot = iph->protocol;
527 u32 mark = skb->mark;
528 __u8 tos = iph->tos;
529
530 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
531 }
532
533 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
534 {
535 const struct inet_sock *inet = inet_sk(sk);
536 const struct ip_options_rcu *inet_opt;
537 __be32 daddr = inet->inet_daddr;
538
539 rcu_read_lock();
540 inet_opt = rcu_dereference(inet->inet_opt);
541 if (inet_opt && inet_opt->opt.srr)
542 daddr = inet_opt->opt.faddr;
543 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
544 ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
545 ip_sock_rt_scope(sk),
546 inet_test_bit(HDRINCL, sk) ?
547 IPPROTO_RAW : sk->sk_protocol,
548 inet_sk_flowi_flags(sk),
549 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
550 rcu_read_unlock();
551 }
552
553 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
554 const struct sk_buff *skb)
555 {
556 if (skb)
557 build_skb_flow_key(fl4, skb, sk);
558 else
559 build_sk_flow_key(fl4, sk);
560 }
561
562 static DEFINE_SPINLOCK(fnhe_lock);
563
564 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
565 {
566 struct rtable *rt;
567
568 rt = rcu_dereference(fnhe->fnhe_rth_input);
569 if (rt) {
570 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
571 dst_dev_put(&rt->dst);
572 dst_release(&rt->dst);
573 }
574 rt = rcu_dereference(fnhe->fnhe_rth_output);
575 if (rt) {
576 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
577 dst_dev_put(&rt->dst);
578 dst_release(&rt->dst);
579 }
580 }
581
582 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
583 {
584 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
585 struct fib_nh_exception *fnhe, *oldest = NULL;
586
587 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
588 fnhe = rcu_dereference_protected(*fnhe_p,
589 lockdep_is_held(&fnhe_lock));
590 if (!fnhe)
591 break;
592 if (!oldest ||
593 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
594 oldest = fnhe;
595 oldest_p = fnhe_p;
596 }
597 }
598 fnhe_flush_routes(oldest);
599 *oldest_p = oldest->fnhe_next;
600 kfree_rcu(oldest, rcu);
601 }
602
603 static u32 fnhe_hashfun(__be32 daddr)
604 {
605 static siphash_aligned_key_t fnhe_hash_key;
606 u64 hval;
607
608 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
609 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
610 return hash_64(hval, FNHE_HASH_SHIFT);
611 }
612
613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
614 {
615 rt->rt_pmtu = fnhe->fnhe_pmtu;
616 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
617 rt->dst.expires = fnhe->fnhe_expires;
618
619 if (fnhe->fnhe_gw) {
620 rt->rt_flags |= RTCF_REDIRECTED;
621 rt->rt_uses_gateway = 1;
622 rt->rt_gw_family = AF_INET;
623 rt->rt_gw4 = fnhe->fnhe_gw;
624 }
625 }
626
627 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
628 __be32 gw, u32 pmtu, bool lock,
629 unsigned long expires)
630 {
631 struct fnhe_hash_bucket *hash;
632 struct fib_nh_exception *fnhe;
633 struct rtable *rt;
634 u32 genid, hval;
635 unsigned int i;
636 int depth;
637
638 genid = fnhe_genid(dev_net(nhc->nhc_dev));
639 hval = fnhe_hashfun(daddr);
640
641 spin_lock_bh(&fnhe_lock);
642
643 hash = rcu_dereference(nhc->nhc_exceptions);
644 if (!hash) {
645 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
646 if (!hash)
647 goto out_unlock;
648 rcu_assign_pointer(nhc->nhc_exceptions, hash);
649 }
650
651 hash += hval;
652
653 depth = 0;
654 for (fnhe = rcu_dereference(hash->chain); fnhe;
655 fnhe = rcu_dereference(fnhe->fnhe_next)) {
656 if (fnhe->fnhe_daddr == daddr)
657 break;
658 depth++;
659 }
660
661 if (fnhe) {
662 if (fnhe->fnhe_genid != genid)
663 fnhe->fnhe_genid = genid;
664 if (gw)
665 fnhe->fnhe_gw = gw;
666 if (pmtu) {
667 fnhe->fnhe_pmtu = pmtu;
668 fnhe->fnhe_mtu_locked = lock;
669 }
670 fnhe->fnhe_expires = max(1UL, expires);
671 /* Update all cached dsts too */
672 rt = rcu_dereference(fnhe->fnhe_rth_input);
673 if (rt)
674 fill_route_from_fnhe(rt, fnhe);
675 rt = rcu_dereference(fnhe->fnhe_rth_output);
676 if (rt)
677 fill_route_from_fnhe(rt, fnhe);
678 } else {
679 /* Randomize max depth to avoid some side channels attacks. */
680 int max_depth = FNHE_RECLAIM_DEPTH +
681 get_random_u32_below(FNHE_RECLAIM_DEPTH);
682
683 while (depth > max_depth) {
684 fnhe_remove_oldest(hash);
685 depth--;
686 }
687
688 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
689 if (!fnhe)
690 goto out_unlock;
691
692 fnhe->fnhe_next = hash->chain;
693
694 fnhe->fnhe_genid = genid;
695 fnhe->fnhe_daddr = daddr;
696 fnhe->fnhe_gw = gw;
697 fnhe->fnhe_pmtu = pmtu;
698 fnhe->fnhe_mtu_locked = lock;
699 fnhe->fnhe_expires = max(1UL, expires);
700
701 rcu_assign_pointer(hash->chain, fnhe);
702
703 /* Exception created; mark the cached routes for the nexthop
704 * stale, so anyone caching it rechecks if this exception
705 * applies to them.
706 */
707 rt = rcu_dereference(nhc->nhc_rth_input);
708 if (rt)
709 rt->dst.obsolete = DST_OBSOLETE_KILL;
710
711 for_each_possible_cpu(i) {
712 struct rtable __rcu **prt;
713
714 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
715 rt = rcu_dereference(*prt);
716 if (rt)
717 rt->dst.obsolete = DST_OBSOLETE_KILL;
718 }
719 }
720
721 fnhe->fnhe_stamp = jiffies;
722
723 out_unlock:
724 spin_unlock_bh(&fnhe_lock);
725 }
726
727 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
728 bool kill_route)
729 {
730 __be32 new_gw = icmp_hdr(skb)->un.gateway;
731 __be32 old_gw = ip_hdr(skb)->saddr;
732 struct net_device *dev = skb->dev;
733 struct in_device *in_dev;
734 struct fib_result res;
735 struct neighbour *n;
736 struct net *net;
737
738 switch (icmp_hdr(skb)->code & 7) {
739 case ICMP_REDIR_NET:
740 case ICMP_REDIR_NETTOS:
741 case ICMP_REDIR_HOST:
742 case ICMP_REDIR_HOSTTOS:
743 break;
744
745 default:
746 return;
747 }
748
749 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
750 return;
751
752 in_dev = __in_dev_get_rcu(dev);
753 if (!in_dev)
754 return;
755
756 net = dev_net(dev);
757 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
758 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
759 ipv4_is_zeronet(new_gw))
760 goto reject_redirect;
761
762 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
763 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
764 goto reject_redirect;
765 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
766 goto reject_redirect;
767 } else {
768 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
769 goto reject_redirect;
770 }
771
772 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
773 if (!n)
774 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
775 if (!IS_ERR(n)) {
776 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
777 neigh_event_send(n, NULL);
778 } else {
779 if (fib_lookup(net, fl4, &res, 0) == 0) {
780 struct fib_nh_common *nhc;
781
782 fib_select_path(net, &res, fl4, skb);
783 nhc = FIB_RES_NHC(res);
784 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
785 0, false,
786 jiffies + ip_rt_gc_timeout);
787 }
788 if (kill_route)
789 rt->dst.obsolete = DST_OBSOLETE_KILL;
790 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
791 }
792 neigh_release(n);
793 }
794 return;
795
796 reject_redirect:
797 #ifdef CONFIG_IP_ROUTE_VERBOSE
798 if (IN_DEV_LOG_MARTIANS(in_dev)) {
799 const struct iphdr *iph = (const struct iphdr *) skb->data;
800 __be32 daddr = iph->daddr;
801 __be32 saddr = iph->saddr;
802
803 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
804 " Advised path = %pI4 -> %pI4\n",
805 &old_gw, dev->name, &new_gw,
806 &saddr, &daddr);
807 }
808 #endif
809 ;
810 }
811
812 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
813 {
814 struct rtable *rt;
815 struct flowi4 fl4;
816 const struct iphdr *iph = (const struct iphdr *) skb->data;
817 struct net *net = dev_net(skb->dev);
818 int oif = skb->dev->ifindex;
819 u8 prot = iph->protocol;
820 u32 mark = skb->mark;
821 __u8 tos = iph->tos;
822
823 rt = dst_rtable(dst);
824
825 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
826 __ip_do_redirect(rt, skb, &fl4, true);
827 }
828
829 static void ipv4_negative_advice(struct sock *sk,
830 struct dst_entry *dst)
831 {
832 struct rtable *rt = dst_rtable(dst);
833
834 if ((dst->obsolete > 0) ||
835 (rt->rt_flags & RTCF_REDIRECTED) ||
836 rt->dst.expires)
837 sk_dst_reset(sk);
838 }
839
840 /*
841 * Algorithm:
842 * 1. The first ip_rt_redirect_number redirects are sent
843 * with exponential backoff, then we stop sending them at all,
844 * assuming that the host ignores our redirects.
845 * 2. If we did not see packets requiring redirects
846 * during ip_rt_redirect_silence, we assume that the host
847 * forgot redirected route and start to send redirects again.
848 *
849 * This algorithm is much cheaper and more intelligent than dumb load limiting
850 * in icmp.c.
851 *
852 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
853 * and "frag. need" (breaks PMTU discovery) in icmp.c.
854 */
855
856 void ip_rt_send_redirect(struct sk_buff *skb)
857 {
858 struct rtable *rt = skb_rtable(skb);
859 struct in_device *in_dev;
860 struct inet_peer *peer;
861 struct net *net;
862 int log_martians;
863 int vif;
864
865 rcu_read_lock();
866 in_dev = __in_dev_get_rcu(rt->dst.dev);
867 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
868 rcu_read_unlock();
869 return;
870 }
871 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
872 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
873 rcu_read_unlock();
874
875 net = dev_net(rt->dst.dev);
876 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
877 if (!peer) {
878 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
879 rt_nexthop(rt, ip_hdr(skb)->daddr));
880 return;
881 }
882
883 /* No redirected packets during ip_rt_redirect_silence;
884 * reset the algorithm.
885 */
886 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
887 peer->rate_tokens = 0;
888 peer->n_redirects = 0;
889 }
890
891 /* Too many ignored redirects; do not send anything
892 * set dst.rate_last to the last seen redirected packet.
893 */
894 if (peer->n_redirects >= ip_rt_redirect_number) {
895 peer->rate_last = jiffies;
896 goto out_put_peer;
897 }
898
899 /* Check for load limit; set rate_last to the latest sent
900 * redirect.
901 */
902 if (peer->n_redirects == 0 ||
903 time_after(jiffies,
904 (peer->rate_last +
905 (ip_rt_redirect_load << peer->n_redirects)))) {
906 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
907
908 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
909 peer->rate_last = jiffies;
910 ++peer->n_redirects;
911 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
912 peer->n_redirects == ip_rt_redirect_number)
913 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
914 &ip_hdr(skb)->saddr, inet_iif(skb),
915 &ip_hdr(skb)->daddr, &gw);
916 }
917 out_put_peer:
918 inet_putpeer(peer);
919 }
920
921 static int ip_error(struct sk_buff *skb)
922 {
923 struct rtable *rt = skb_rtable(skb);
924 struct net_device *dev = skb->dev;
925 struct in_device *in_dev;
926 struct inet_peer *peer;
927 unsigned long now;
928 struct net *net;
929 SKB_DR(reason);
930 bool send;
931 int code;
932
933 if (netif_is_l3_master(skb->dev)) {
934 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
935 if (!dev)
936 goto out;
937 }
938
939 in_dev = __in_dev_get_rcu(dev);
940
941 /* IP on this device is disabled. */
942 if (!in_dev)
943 goto out;
944
945 net = dev_net(rt->dst.dev);
946 if (!IN_DEV_FORWARD(in_dev)) {
947 switch (rt->dst.error) {
948 case EHOSTUNREACH:
949 SKB_DR_SET(reason, IP_INADDRERRORS);
950 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
951 break;
952
953 case ENETUNREACH:
954 SKB_DR_SET(reason, IP_INNOROUTES);
955 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
956 break;
957 }
958 goto out;
959 }
960
961 switch (rt->dst.error) {
962 case EINVAL:
963 default:
964 goto out;
965 case EHOSTUNREACH:
966 code = ICMP_HOST_UNREACH;
967 break;
968 case ENETUNREACH:
969 code = ICMP_NET_UNREACH;
970 SKB_DR_SET(reason, IP_INNOROUTES);
971 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
972 break;
973 case EACCES:
974 code = ICMP_PKT_FILTERED;
975 break;
976 }
977
978 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
979 l3mdev_master_ifindex(skb->dev), 1);
980
981 send = true;
982 if (peer) {
983 now = jiffies;
984 peer->rate_tokens += now - peer->rate_last;
985 if (peer->rate_tokens > ip_rt_error_burst)
986 peer->rate_tokens = ip_rt_error_burst;
987 peer->rate_last = now;
988 if (peer->rate_tokens >= ip_rt_error_cost)
989 peer->rate_tokens -= ip_rt_error_cost;
990 else
991 send = false;
992 inet_putpeer(peer);
993 }
994 if (send)
995 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
996
997 out: kfree_skb_reason(skb, reason);
998 return 0;
999 }
1000
1001 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1002 {
1003 struct dst_entry *dst = &rt->dst;
1004 struct net *net = dev_net(dst->dev);
1005 struct fib_result res;
1006 bool lock = false;
1007 u32 old_mtu;
1008
1009 if (ip_mtu_locked(dst))
1010 return;
1011
1012 old_mtu = ipv4_mtu(dst);
1013 if (old_mtu < mtu)
1014 return;
1015
1016 if (mtu < net->ipv4.ip_rt_min_pmtu) {
1017 lock = true;
1018 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1019 }
1020
1021 if (rt->rt_pmtu == mtu && !lock &&
1022 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1023 return;
1024
1025 rcu_read_lock();
1026 if (fib_lookup(net, fl4, &res, 0) == 0) {
1027 struct fib_nh_common *nhc;
1028
1029 fib_select_path(net, &res, fl4, NULL);
1030 nhc = FIB_RES_NHC(res);
1031 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1032 jiffies + net->ipv4.ip_rt_mtu_expires);
1033 }
1034 rcu_read_unlock();
1035 }
1036
1037 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1038 struct sk_buff *skb, u32 mtu,
1039 bool confirm_neigh)
1040 {
1041 struct rtable *rt = dst_rtable(dst);
1042 struct flowi4 fl4;
1043
1044 ip_rt_build_flow_key(&fl4, sk, skb);
1045
1046 /* Don't make lookup fail for bridged encapsulations */
1047 if (skb && netif_is_any_bridge_port(skb->dev))
1048 fl4.flowi4_oif = 0;
1049
1050 __ip_rt_update_pmtu(rt, &fl4, mtu);
1051 }
1052
1053 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1054 int oif, u8 protocol)
1055 {
1056 const struct iphdr *iph = (const struct iphdr *)skb->data;
1057 struct flowi4 fl4;
1058 struct rtable *rt;
1059 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1060
1061 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1062 0);
1063 rt = __ip_route_output_key(net, &fl4);
1064 if (!IS_ERR(rt)) {
1065 __ip_rt_update_pmtu(rt, &fl4, mtu);
1066 ip_rt_put(rt);
1067 }
1068 }
1069 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1070
1071 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1072 {
1073 const struct iphdr *iph = (const struct iphdr *)skb->data;
1074 struct flowi4 fl4;
1075 struct rtable *rt;
1076
1077 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1078
1079 if (!fl4.flowi4_mark)
1080 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1081
1082 rt = __ip_route_output_key(sock_net(sk), &fl4);
1083 if (!IS_ERR(rt)) {
1084 __ip_rt_update_pmtu(rt, &fl4, mtu);
1085 ip_rt_put(rt);
1086 }
1087 }
1088
1089 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1090 {
1091 const struct iphdr *iph = (const struct iphdr *)skb->data;
1092 struct flowi4 fl4;
1093 struct rtable *rt;
1094 struct dst_entry *odst = NULL;
1095 bool new = false;
1096 struct net *net = sock_net(sk);
1097
1098 bh_lock_sock(sk);
1099
1100 if (!ip_sk_accept_pmtu(sk))
1101 goto out;
1102
1103 odst = sk_dst_get(sk);
1104
1105 if (sock_owned_by_user(sk) || !odst) {
1106 __ipv4_sk_update_pmtu(skb, sk, mtu);
1107 goto out;
1108 }
1109
1110 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1111
1112 rt = dst_rtable(odst);
1113 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1114 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1115 if (IS_ERR(rt))
1116 goto out;
1117
1118 new = true;
1119 }
1120
1121 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1122
1123 if (!dst_check(&rt->dst, 0)) {
1124 if (new)
1125 dst_release(&rt->dst);
1126
1127 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1128 if (IS_ERR(rt))
1129 goto out;
1130
1131 new = true;
1132 }
1133
1134 if (new)
1135 sk_dst_set(sk, &rt->dst);
1136
1137 out:
1138 bh_unlock_sock(sk);
1139 dst_release(odst);
1140 }
1141 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1142
1143 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1144 int oif, u8 protocol)
1145 {
1146 const struct iphdr *iph = (const struct iphdr *)skb->data;
1147 struct flowi4 fl4;
1148 struct rtable *rt;
1149
1150 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1151 rt = __ip_route_output_key(net, &fl4);
1152 if (!IS_ERR(rt)) {
1153 __ip_do_redirect(rt, skb, &fl4, false);
1154 ip_rt_put(rt);
1155 }
1156 }
1157 EXPORT_SYMBOL_GPL(ipv4_redirect);
1158
1159 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1160 {
1161 const struct iphdr *iph = (const struct iphdr *)skb->data;
1162 struct flowi4 fl4;
1163 struct rtable *rt;
1164 struct net *net = sock_net(sk);
1165
1166 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1167 rt = __ip_route_output_key(net, &fl4);
1168 if (!IS_ERR(rt)) {
1169 __ip_do_redirect(rt, skb, &fl4, false);
1170 ip_rt_put(rt);
1171 }
1172 }
1173 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1174
1175 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1176 u32 cookie)
1177 {
1178 struct rtable *rt = dst_rtable(dst);
1179
1180 /* All IPV4 dsts are created with ->obsolete set to the value
1181 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1182 * into this function always.
1183 *
1184 * When a PMTU/redirect information update invalidates a route,
1185 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1186 * DST_OBSOLETE_DEAD.
1187 */
1188 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1189 return NULL;
1190 return dst;
1191 }
1192 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1193
1194 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1195 {
1196 struct net_device *dev;
1197 struct ip_options opt;
1198 int res;
1199
1200 /* Recompile ip options since IPCB may not be valid anymore.
1201 * Also check we have a reasonable ipv4 header.
1202 */
1203 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1204 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1205 return;
1206
1207 memset(&opt, 0, sizeof(opt));
1208 if (ip_hdr(skb)->ihl > 5) {
1209 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1210 return;
1211 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1212
1213 rcu_read_lock();
1214 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1215 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1216 rcu_read_unlock();
1217
1218 if (res)
1219 return;
1220 }
1221 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1222 }
1223
1224 static void ipv4_link_failure(struct sk_buff *skb)
1225 {
1226 struct rtable *rt;
1227
1228 ipv4_send_dest_unreach(skb);
1229
1230 rt = skb_rtable(skb);
1231 if (rt)
1232 dst_set_expires(&rt->dst, 0);
1233 }
1234
1235 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1236 {
1237 pr_debug("%s: %pI4 -> %pI4, %s\n",
1238 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1239 skb->dev ? skb->dev->name : "?");
1240 kfree_skb(skb);
1241 WARN_ON(1);
1242 return 0;
1243 }
1244
1245 /*
1246 * We do not cache source address of outgoing interface,
1247 * because it is used only by IP RR, TS and SRR options,
1248 * so that it out of fast path.
1249 *
1250 * BTW remember: "addr" is allowed to be not aligned
1251 * in IP options!
1252 */
1253
1254 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1255 {
1256 __be32 src;
1257
1258 if (rt_is_output_route(rt))
1259 src = ip_hdr(skb)->saddr;
1260 else {
1261 struct fib_result res;
1262 struct iphdr *iph = ip_hdr(skb);
1263 struct flowi4 fl4 = {
1264 .daddr = iph->daddr,
1265 .saddr = iph->saddr,
1266 .flowi4_tos = iph->tos & IPTOS_RT_MASK,
1267 .flowi4_oif = rt->dst.dev->ifindex,
1268 .flowi4_iif = skb->dev->ifindex,
1269 .flowi4_mark = skb->mark,
1270 };
1271
1272 rcu_read_lock();
1273 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1274 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1275 else
1276 src = inet_select_addr(rt->dst.dev,
1277 rt_nexthop(rt, iph->daddr),
1278 RT_SCOPE_UNIVERSE);
1279 rcu_read_unlock();
1280 }
1281 memcpy(addr, &src, 4);
1282 }
1283
1284 #ifdef CONFIG_IP_ROUTE_CLASSID
1285 static void set_class_tag(struct rtable *rt, u32 tag)
1286 {
1287 if (!(rt->dst.tclassid & 0xFFFF))
1288 rt->dst.tclassid |= tag & 0xFFFF;
1289 if (!(rt->dst.tclassid & 0xFFFF0000))
1290 rt->dst.tclassid |= tag & 0xFFFF0000;
1291 }
1292 #endif
1293
1294 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1295 {
1296 struct net *net = dev_net(dst->dev);
1297 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1298 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1299 net->ipv4.ip_rt_min_advmss);
1300
1301 return min(advmss, IPV4_MAX_PMTU - header_size);
1302 }
1303
1304 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1305 {
1306 return ip_dst_mtu_maybe_forward(dst, false);
1307 }
1308 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1309
1310 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1311 {
1312 struct fnhe_hash_bucket *hash;
1313 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1314 u32 hval = fnhe_hashfun(daddr);
1315
1316 spin_lock_bh(&fnhe_lock);
1317
1318 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1319 lockdep_is_held(&fnhe_lock));
1320 hash += hval;
1321
1322 fnhe_p = &hash->chain;
1323 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1324 while (fnhe) {
1325 if (fnhe->fnhe_daddr == daddr) {
1326 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1327 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1328 /* set fnhe_daddr to 0 to ensure it won't bind with
1329 * new dsts in rt_bind_exception().
1330 */
1331 fnhe->fnhe_daddr = 0;
1332 fnhe_flush_routes(fnhe);
1333 kfree_rcu(fnhe, rcu);
1334 break;
1335 }
1336 fnhe_p = &fnhe->fnhe_next;
1337 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1338 lockdep_is_held(&fnhe_lock));
1339 }
1340
1341 spin_unlock_bh(&fnhe_lock);
1342 }
1343
1344 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1345 __be32 daddr)
1346 {
1347 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1348 struct fib_nh_exception *fnhe;
1349 u32 hval;
1350
1351 if (!hash)
1352 return NULL;
1353
1354 hval = fnhe_hashfun(daddr);
1355
1356 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1357 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1358 if (fnhe->fnhe_daddr == daddr) {
1359 if (fnhe->fnhe_expires &&
1360 time_after(jiffies, fnhe->fnhe_expires)) {
1361 ip_del_fnhe(nhc, daddr);
1362 break;
1363 }
1364 return fnhe;
1365 }
1366 }
1367 return NULL;
1368 }
1369
1370 /* MTU selection:
1371 * 1. mtu on route is locked - use it
1372 * 2. mtu from nexthop exception
1373 * 3. mtu from egress device
1374 */
1375
1376 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1377 {
1378 struct fib_nh_common *nhc = res->nhc;
1379 struct net_device *dev = nhc->nhc_dev;
1380 struct fib_info *fi = res->fi;
1381 u32 mtu = 0;
1382
1383 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1384 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1385 mtu = fi->fib_mtu;
1386
1387 if (likely(!mtu)) {
1388 struct fib_nh_exception *fnhe;
1389
1390 fnhe = find_exception(nhc, daddr);
1391 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1392 mtu = fnhe->fnhe_pmtu;
1393 }
1394
1395 if (likely(!mtu))
1396 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1397
1398 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1399 }
1400
1401 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1402 __be32 daddr, const bool do_cache)
1403 {
1404 bool ret = false;
1405
1406 spin_lock_bh(&fnhe_lock);
1407
1408 if (daddr == fnhe->fnhe_daddr) {
1409 struct rtable __rcu **porig;
1410 struct rtable *orig;
1411 int genid = fnhe_genid(dev_net(rt->dst.dev));
1412
1413 if (rt_is_input_route(rt))
1414 porig = &fnhe->fnhe_rth_input;
1415 else
1416 porig = &fnhe->fnhe_rth_output;
1417 orig = rcu_dereference(*porig);
1418
1419 if (fnhe->fnhe_genid != genid) {
1420 fnhe->fnhe_genid = genid;
1421 fnhe->fnhe_gw = 0;
1422 fnhe->fnhe_pmtu = 0;
1423 fnhe->fnhe_expires = 0;
1424 fnhe->fnhe_mtu_locked = false;
1425 fnhe_flush_routes(fnhe);
1426 orig = NULL;
1427 }
1428 fill_route_from_fnhe(rt, fnhe);
1429 if (!rt->rt_gw4) {
1430 rt->rt_gw4 = daddr;
1431 rt->rt_gw_family = AF_INET;
1432 }
1433
1434 if (do_cache) {
1435 dst_hold(&rt->dst);
1436 rcu_assign_pointer(*porig, rt);
1437 if (orig) {
1438 dst_dev_put(&orig->dst);
1439 dst_release(&orig->dst);
1440 }
1441 ret = true;
1442 }
1443
1444 fnhe->fnhe_stamp = jiffies;
1445 }
1446 spin_unlock_bh(&fnhe_lock);
1447
1448 return ret;
1449 }
1450
1451 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1452 {
1453 struct rtable *orig, *prev, **p;
1454 bool ret = true;
1455
1456 if (rt_is_input_route(rt)) {
1457 p = (struct rtable **)&nhc->nhc_rth_input;
1458 } else {
1459 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1460 }
1461 orig = *p;
1462
1463 /* hold dst before doing cmpxchg() to avoid race condition
1464 * on this dst
1465 */
1466 dst_hold(&rt->dst);
1467 prev = cmpxchg(p, orig, rt);
1468 if (prev == orig) {
1469 if (orig) {
1470 rt_add_uncached_list(orig);
1471 dst_release(&orig->dst);
1472 }
1473 } else {
1474 dst_release(&rt->dst);
1475 ret = false;
1476 }
1477
1478 return ret;
1479 }
1480
1481 struct uncached_list {
1482 spinlock_t lock;
1483 struct list_head head;
1484 };
1485
1486 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1487
1488 void rt_add_uncached_list(struct rtable *rt)
1489 {
1490 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1491
1492 rt->dst.rt_uncached_list = ul;
1493
1494 spin_lock_bh(&ul->lock);
1495 list_add_tail(&rt->dst.rt_uncached, &ul->head);
1496 spin_unlock_bh(&ul->lock);
1497 }
1498
1499 void rt_del_uncached_list(struct rtable *rt)
1500 {
1501 if (!list_empty(&rt->dst.rt_uncached)) {
1502 struct uncached_list *ul = rt->dst.rt_uncached_list;
1503
1504 spin_lock_bh(&ul->lock);
1505 list_del_init(&rt->dst.rt_uncached);
1506 spin_unlock_bh(&ul->lock);
1507 }
1508 }
1509
1510 static void ipv4_dst_destroy(struct dst_entry *dst)
1511 {
1512 ip_dst_metrics_put(dst);
1513 rt_del_uncached_list(dst_rtable(dst));
1514 }
1515
1516 void rt_flush_dev(struct net_device *dev)
1517 {
1518 struct rtable *rt, *safe;
1519 int cpu;
1520
1521 for_each_possible_cpu(cpu) {
1522 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1523
1524 if (list_empty(&ul->head))
1525 continue;
1526
1527 spin_lock_bh(&ul->lock);
1528 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1529 if (rt->dst.dev != dev)
1530 continue;
1531 rt->dst.dev = blackhole_netdev;
1532 netdev_ref_replace(dev, blackhole_netdev,
1533 &rt->dst.dev_tracker, GFP_ATOMIC);
1534 list_del_init(&rt->dst.rt_uncached);
1535 }
1536 spin_unlock_bh(&ul->lock);
1537 }
1538 }
1539
1540 static bool rt_cache_valid(const struct rtable *rt)
1541 {
1542 return rt &&
1543 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1544 !rt_is_expired(rt);
1545 }
1546
1547 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1548 const struct fib_result *res,
1549 struct fib_nh_exception *fnhe,
1550 struct fib_info *fi, u16 type, u32 itag,
1551 const bool do_cache)
1552 {
1553 bool cached = false;
1554
1555 if (fi) {
1556 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1557
1558 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1559 rt->rt_uses_gateway = 1;
1560 rt->rt_gw_family = nhc->nhc_gw_family;
1561 /* only INET and INET6 are supported */
1562 if (likely(nhc->nhc_gw_family == AF_INET))
1563 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1564 else
1565 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1566 }
1567
1568 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1569
1570 #ifdef CONFIG_IP_ROUTE_CLASSID
1571 if (nhc->nhc_family == AF_INET) {
1572 struct fib_nh *nh;
1573
1574 nh = container_of(nhc, struct fib_nh, nh_common);
1575 rt->dst.tclassid = nh->nh_tclassid;
1576 }
1577 #endif
1578 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1579 if (unlikely(fnhe))
1580 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1581 else if (do_cache)
1582 cached = rt_cache_route(nhc, rt);
1583 if (unlikely(!cached)) {
1584 /* Routes we intend to cache in nexthop exception or
1585 * FIB nexthop have the DST_NOCACHE bit clear.
1586 * However, if we are unsuccessful at storing this
1587 * route into the cache we really need to set it.
1588 */
1589 if (!rt->rt_gw4) {
1590 rt->rt_gw_family = AF_INET;
1591 rt->rt_gw4 = daddr;
1592 }
1593 rt_add_uncached_list(rt);
1594 }
1595 } else
1596 rt_add_uncached_list(rt);
1597
1598 #ifdef CONFIG_IP_ROUTE_CLASSID
1599 #ifdef CONFIG_IP_MULTIPLE_TABLES
1600 set_class_tag(rt, res->tclassid);
1601 #endif
1602 set_class_tag(rt, itag);
1603 #endif
1604 }
1605
1606 struct rtable *rt_dst_alloc(struct net_device *dev,
1607 unsigned int flags, u16 type,
1608 bool noxfrm)
1609 {
1610 struct rtable *rt;
1611
1612 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1613 (noxfrm ? DST_NOXFRM : 0));
1614
1615 if (rt) {
1616 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1617 rt->rt_flags = flags;
1618 rt->rt_type = type;
1619 rt->rt_is_input = 0;
1620 rt->rt_iif = 0;
1621 rt->rt_pmtu = 0;
1622 rt->rt_mtu_locked = 0;
1623 rt->rt_uses_gateway = 0;
1624 rt->rt_gw_family = 0;
1625 rt->rt_gw4 = 0;
1626
1627 rt->dst.output = ip_output;
1628 if (flags & RTCF_LOCAL)
1629 rt->dst.input = ip_local_deliver;
1630 }
1631
1632 return rt;
1633 }
1634 EXPORT_SYMBOL(rt_dst_alloc);
1635
1636 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1637 {
1638 struct rtable *new_rt;
1639
1640 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1641 rt->dst.flags);
1642
1643 if (new_rt) {
1644 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1645 new_rt->rt_flags = rt->rt_flags;
1646 new_rt->rt_type = rt->rt_type;
1647 new_rt->rt_is_input = rt->rt_is_input;
1648 new_rt->rt_iif = rt->rt_iif;
1649 new_rt->rt_pmtu = rt->rt_pmtu;
1650 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1651 new_rt->rt_gw_family = rt->rt_gw_family;
1652 if (rt->rt_gw_family == AF_INET)
1653 new_rt->rt_gw4 = rt->rt_gw4;
1654 else if (rt->rt_gw_family == AF_INET6)
1655 new_rt->rt_gw6 = rt->rt_gw6;
1656
1657 new_rt->dst.input = rt->dst.input;
1658 new_rt->dst.output = rt->dst.output;
1659 new_rt->dst.error = rt->dst.error;
1660 new_rt->dst.lastuse = jiffies;
1661 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1662 }
1663 return new_rt;
1664 }
1665 EXPORT_SYMBOL(rt_dst_clone);
1666
1667 /* called in rcu_read_lock() section */
1668 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1669 u8 tos, struct net_device *dev,
1670 struct in_device *in_dev, u32 *itag)
1671 {
1672 int err;
1673
1674 /* Primary sanity checks. */
1675 if (!in_dev)
1676 return -EINVAL;
1677
1678 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1679 skb->protocol != htons(ETH_P_IP))
1680 return -EINVAL;
1681
1682 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1683 return -EINVAL;
1684
1685 if (ipv4_is_zeronet(saddr)) {
1686 if (!ipv4_is_local_multicast(daddr) &&
1687 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1688 return -EINVAL;
1689 } else {
1690 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1691 in_dev, itag);
1692 if (err < 0)
1693 return err;
1694 }
1695 return 0;
1696 }
1697
1698 /* called in rcu_read_lock() section */
1699 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1700 u8 tos, struct net_device *dev, int our)
1701 {
1702 struct in_device *in_dev = __in_dev_get_rcu(dev);
1703 unsigned int flags = RTCF_MULTICAST;
1704 struct rtable *rth;
1705 u32 itag = 0;
1706 int err;
1707
1708 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1709 if (err)
1710 return err;
1711
1712 if (our)
1713 flags |= RTCF_LOCAL;
1714
1715 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1716 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1717
1718 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1719 false);
1720 if (!rth)
1721 return -ENOBUFS;
1722
1723 #ifdef CONFIG_IP_ROUTE_CLASSID
1724 rth->dst.tclassid = itag;
1725 #endif
1726 rth->dst.output = ip_rt_bug;
1727 rth->rt_is_input= 1;
1728
1729 #ifdef CONFIG_IP_MROUTE
1730 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1731 rth->dst.input = ip_mr_input;
1732 #endif
1733 RT_CACHE_STAT_INC(in_slow_mc);
1734
1735 skb_dst_drop(skb);
1736 skb_dst_set(skb, &rth->dst);
1737 return 0;
1738 }
1739
1740
1741 static void ip_handle_martian_source(struct net_device *dev,
1742 struct in_device *in_dev,
1743 struct sk_buff *skb,
1744 __be32 daddr,
1745 __be32 saddr)
1746 {
1747 RT_CACHE_STAT_INC(in_martian_src);
1748 #ifdef CONFIG_IP_ROUTE_VERBOSE
1749 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1750 /*
1751 * RFC1812 recommendation, if source is martian,
1752 * the only hint is MAC header.
1753 */
1754 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1755 &daddr, &saddr, dev->name);
1756 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1757 print_hex_dump(KERN_WARNING, "ll header: ",
1758 DUMP_PREFIX_OFFSET, 16, 1,
1759 skb_mac_header(skb),
1760 dev->hard_header_len, false);
1761 }
1762 }
1763 #endif
1764 }
1765
1766 /* called in rcu_read_lock() section */
1767 static int __mkroute_input(struct sk_buff *skb,
1768 const struct fib_result *res,
1769 struct in_device *in_dev,
1770 __be32 daddr, __be32 saddr, u32 tos)
1771 {
1772 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1773 struct net_device *dev = nhc->nhc_dev;
1774 struct fib_nh_exception *fnhe;
1775 struct rtable *rth;
1776 int err;
1777 struct in_device *out_dev;
1778 bool do_cache;
1779 u32 itag = 0;
1780
1781 /* get a working reference to the output device */
1782 out_dev = __in_dev_get_rcu(dev);
1783 if (!out_dev) {
1784 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1785 return -EINVAL;
1786 }
1787
1788 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1789 in_dev->dev, in_dev, &itag);
1790 if (err < 0) {
1791 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1792 saddr);
1793
1794 goto cleanup;
1795 }
1796
1797 do_cache = res->fi && !itag;
1798 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1799 skb->protocol == htons(ETH_P_IP)) {
1800 __be32 gw;
1801
1802 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1803 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1804 inet_addr_onlink(out_dev, saddr, gw))
1805 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1806 }
1807
1808 if (skb->protocol != htons(ETH_P_IP)) {
1809 /* Not IP (i.e. ARP). Do not create route, if it is
1810 * invalid for proxy arp. DNAT routes are always valid.
1811 *
1812 * Proxy arp feature have been extended to allow, ARP
1813 * replies back to the same interface, to support
1814 * Private VLAN switch technologies. See arp.c.
1815 */
1816 if (out_dev == in_dev &&
1817 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1818 err = -EINVAL;
1819 goto cleanup;
1820 }
1821 }
1822
1823 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1824 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1825
1826 fnhe = find_exception(nhc, daddr);
1827 if (do_cache) {
1828 if (fnhe)
1829 rth = rcu_dereference(fnhe->fnhe_rth_input);
1830 else
1831 rth = rcu_dereference(nhc->nhc_rth_input);
1832 if (rt_cache_valid(rth)) {
1833 skb_dst_set_noref(skb, &rth->dst);
1834 goto out;
1835 }
1836 }
1837
1838 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1839 IN_DEV_ORCONF(out_dev, NOXFRM));
1840 if (!rth) {
1841 err = -ENOBUFS;
1842 goto cleanup;
1843 }
1844
1845 rth->rt_is_input = 1;
1846 RT_CACHE_STAT_INC(in_slow_tot);
1847
1848 rth->dst.input = ip_forward;
1849
1850 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1851 do_cache);
1852 lwtunnel_set_redirect(&rth->dst);
1853 skb_dst_set(skb, &rth->dst);
1854 out:
1855 err = 0;
1856 cleanup:
1857 return err;
1858 }
1859
1860 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1861 /* To make ICMP packets follow the right flow, the multipath hash is
1862 * calculated from the inner IP addresses.
1863 */
1864 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1865 struct flow_keys *hash_keys)
1866 {
1867 const struct iphdr *outer_iph = ip_hdr(skb);
1868 const struct iphdr *key_iph = outer_iph;
1869 const struct iphdr *inner_iph;
1870 const struct icmphdr *icmph;
1871 struct iphdr _inner_iph;
1872 struct icmphdr _icmph;
1873
1874 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1875 goto out;
1876
1877 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1878 goto out;
1879
1880 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1881 &_icmph);
1882 if (!icmph)
1883 goto out;
1884
1885 if (!icmp_is_err(icmph->type))
1886 goto out;
1887
1888 inner_iph = skb_header_pointer(skb,
1889 outer_iph->ihl * 4 + sizeof(_icmph),
1890 sizeof(_inner_iph), &_inner_iph);
1891 if (!inner_iph)
1892 goto out;
1893
1894 key_iph = inner_iph;
1895 out:
1896 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1897 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1898 }
1899
1900 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1901 const struct sk_buff *skb,
1902 bool *p_has_inner)
1903 {
1904 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1905 struct flow_keys keys, hash_keys;
1906
1907 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1908 return 0;
1909
1910 memset(&hash_keys, 0, sizeof(hash_keys));
1911 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1912
1913 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1914 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1915 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1916 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1917 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1918 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1919 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1920 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1921 hash_keys.ports.src = keys.ports.src;
1922 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1923 hash_keys.ports.dst = keys.ports.dst;
1924
1925 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1926 return fib_multipath_hash_from_keys(net, &hash_keys);
1927 }
1928
1929 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1930 const struct sk_buff *skb,
1931 bool has_inner)
1932 {
1933 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1934 struct flow_keys keys, hash_keys;
1935
1936 /* We assume the packet carries an encapsulation, but if none was
1937 * encountered during dissection of the outer flow, then there is no
1938 * point in calling the flow dissector again.
1939 */
1940 if (!has_inner)
1941 return 0;
1942
1943 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1944 return 0;
1945
1946 memset(&hash_keys, 0, sizeof(hash_keys));
1947 skb_flow_dissect_flow_keys(skb, &keys, 0);
1948
1949 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1950 return 0;
1951
1952 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1953 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1954 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1955 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1956 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1957 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1958 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1959 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1960 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1961 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1962 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1963 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1964 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1965 hash_keys.tags.flow_label = keys.tags.flow_label;
1966 }
1967
1968 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1969 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1970 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1971 hash_keys.ports.src = keys.ports.src;
1972 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1973 hash_keys.ports.dst = keys.ports.dst;
1974
1975 return fib_multipath_hash_from_keys(net, &hash_keys);
1976 }
1977
1978 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1979 const struct sk_buff *skb)
1980 {
1981 u32 mhash, mhash_inner;
1982 bool has_inner = true;
1983
1984 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1985 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1986
1987 return jhash_2words(mhash, mhash_inner, 0);
1988 }
1989
1990 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
1991 const struct flowi4 *fl4)
1992 {
1993 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1994 struct flow_keys hash_keys;
1995
1996 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1997 return 0;
1998
1999 memset(&hash_keys, 0, sizeof(hash_keys));
2000 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2001 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2002 hash_keys.addrs.v4addrs.src = fl4->saddr;
2003 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2004 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2005 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2006 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2007 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2008 hash_keys.ports.src = fl4->fl4_sport;
2009 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2010 hash_keys.ports.dst = fl4->fl4_dport;
2011
2012 return fib_multipath_hash_from_keys(net, &hash_keys);
2013 }
2014
2015 /* if skb is set it will be used and fl4 can be NULL */
2016 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2017 const struct sk_buff *skb, struct flow_keys *flkeys)
2018 {
2019 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2020 struct flow_keys hash_keys;
2021 u32 mhash = 0;
2022
2023 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2024 case 0:
2025 memset(&hash_keys, 0, sizeof(hash_keys));
2026 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2027 if (skb) {
2028 ip_multipath_l3_keys(skb, &hash_keys);
2029 } else {
2030 hash_keys.addrs.v4addrs.src = fl4->saddr;
2031 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2032 }
2033 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2034 break;
2035 case 1:
2036 /* skb is currently provided only when forwarding */
2037 if (skb) {
2038 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2039 struct flow_keys keys;
2040
2041 /* short-circuit if we already have L4 hash present */
2042 if (skb->l4_hash)
2043 return skb_get_hash_raw(skb) >> 1;
2044
2045 memset(&hash_keys, 0, sizeof(hash_keys));
2046
2047 if (!flkeys) {
2048 skb_flow_dissect_flow_keys(skb, &keys, flag);
2049 flkeys = &keys;
2050 }
2051
2052 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2053 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2054 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2055 hash_keys.ports.src = flkeys->ports.src;
2056 hash_keys.ports.dst = flkeys->ports.dst;
2057 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2058 } else {
2059 memset(&hash_keys, 0, sizeof(hash_keys));
2060 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2061 hash_keys.addrs.v4addrs.src = fl4->saddr;
2062 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2063 hash_keys.ports.src = fl4->fl4_sport;
2064 hash_keys.ports.dst = fl4->fl4_dport;
2065 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2066 }
2067 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2068 break;
2069 case 2:
2070 memset(&hash_keys, 0, sizeof(hash_keys));
2071 /* skb is currently provided only when forwarding */
2072 if (skb) {
2073 struct flow_keys keys;
2074
2075 skb_flow_dissect_flow_keys(skb, &keys, 0);
2076 /* Inner can be v4 or v6 */
2077 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2078 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2079 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2080 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2081 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2082 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2083 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2084 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2085 hash_keys.tags.flow_label = keys.tags.flow_label;
2086 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2087 } else {
2088 /* Same as case 0 */
2089 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2090 ip_multipath_l3_keys(skb, &hash_keys);
2091 }
2092 } else {
2093 /* Same as case 0 */
2094 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2095 hash_keys.addrs.v4addrs.src = fl4->saddr;
2096 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2097 }
2098 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2099 break;
2100 case 3:
2101 if (skb)
2102 mhash = fib_multipath_custom_hash_skb(net, skb);
2103 else
2104 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2105 break;
2106 }
2107
2108 if (multipath_hash)
2109 mhash = jhash_2words(mhash, multipath_hash, 0);
2110
2111 return mhash >> 1;
2112 }
2113 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2114
2115 static int ip_mkroute_input(struct sk_buff *skb,
2116 struct fib_result *res,
2117 struct in_device *in_dev,
2118 __be32 daddr, __be32 saddr, u32 tos,
2119 struct flow_keys *hkeys)
2120 {
2121 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2122 if (res->fi && fib_info_num_path(res->fi) > 1) {
2123 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2124
2125 fib_select_multipath(res, h);
2126 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2127 }
2128 #endif
2129
2130 /* create a routing cache entry */
2131 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2132 }
2133
2134 /* Implements all the saddr-related checks as ip_route_input_slow(),
2135 * assuming daddr is valid and the destination is not a local broadcast one.
2136 * Uses the provided hint instead of performing a route lookup.
2137 */
2138 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2139 u8 tos, struct net_device *dev,
2140 const struct sk_buff *hint)
2141 {
2142 struct in_device *in_dev = __in_dev_get_rcu(dev);
2143 struct rtable *rt = skb_rtable(hint);
2144 struct net *net = dev_net(dev);
2145 int err = -EINVAL;
2146 u32 tag = 0;
2147
2148 if (!in_dev)
2149 return -EINVAL;
2150
2151 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2152 goto martian_source;
2153
2154 if (ipv4_is_zeronet(saddr))
2155 goto martian_source;
2156
2157 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2158 goto martian_source;
2159
2160 if (rt->rt_type != RTN_LOCAL)
2161 goto skip_validate_source;
2162
2163 tos &= IPTOS_RT_MASK;
2164 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2165 if (err < 0)
2166 goto martian_source;
2167
2168 skip_validate_source:
2169 skb_dst_copy(skb, hint);
2170 return 0;
2171
2172 martian_source:
2173 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2174 return err;
2175 }
2176
2177 /* get device for dst_alloc with local routes */
2178 static struct net_device *ip_rt_get_dev(struct net *net,
2179 const struct fib_result *res)
2180 {
2181 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2182 struct net_device *dev = NULL;
2183
2184 if (nhc)
2185 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2186
2187 return dev ? : net->loopback_dev;
2188 }
2189
2190 /*
2191 * NOTE. We drop all the packets that has local source
2192 * addresses, because every properly looped back packet
2193 * must have correct destination already attached by output routine.
2194 * Changes in the enforced policies must be applied also to
2195 * ip_route_use_hint().
2196 *
2197 * Such approach solves two big problems:
2198 * 1. Not simplex devices are handled properly.
2199 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2200 * called with rcu_read_lock()
2201 */
2202
2203 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2204 u8 tos, struct net_device *dev,
2205 struct fib_result *res)
2206 {
2207 struct in_device *in_dev = __in_dev_get_rcu(dev);
2208 struct flow_keys *flkeys = NULL, _flkeys;
2209 struct net *net = dev_net(dev);
2210 struct ip_tunnel_info *tun_info;
2211 int err = -EINVAL;
2212 unsigned int flags = 0;
2213 u32 itag = 0;
2214 struct rtable *rth;
2215 struct flowi4 fl4;
2216 bool do_cache = true;
2217
2218 /* IP on this device is disabled. */
2219
2220 if (!in_dev)
2221 goto out;
2222
2223 /* Check for the most weird martians, which can be not detected
2224 * by fib_lookup.
2225 */
2226
2227 tun_info = skb_tunnel_info(skb);
2228 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2229 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2230 else
2231 fl4.flowi4_tun_key.tun_id = 0;
2232 skb_dst_drop(skb);
2233
2234 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2235 goto martian_source;
2236
2237 res->fi = NULL;
2238 res->table = NULL;
2239 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2240 goto brd_input;
2241
2242 /* Accept zero addresses only to limited broadcast;
2243 * I even do not know to fix it or not. Waiting for complains :-)
2244 */
2245 if (ipv4_is_zeronet(saddr))
2246 goto martian_source;
2247
2248 if (ipv4_is_zeronet(daddr))
2249 goto martian_destination;
2250
2251 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2252 * and call it once if daddr or/and saddr are loopback addresses
2253 */
2254 if (ipv4_is_loopback(daddr)) {
2255 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2256 goto martian_destination;
2257 } else if (ipv4_is_loopback(saddr)) {
2258 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2259 goto martian_source;
2260 }
2261
2262 /*
2263 * Now we are ready to route packet.
2264 */
2265 fl4.flowi4_l3mdev = 0;
2266 fl4.flowi4_oif = 0;
2267 fl4.flowi4_iif = dev->ifindex;
2268 fl4.flowi4_mark = skb->mark;
2269 fl4.flowi4_tos = tos;
2270 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2271 fl4.flowi4_flags = 0;
2272 fl4.daddr = daddr;
2273 fl4.saddr = saddr;
2274 fl4.flowi4_uid = sock_net_uid(net, NULL);
2275 fl4.flowi4_multipath_hash = 0;
2276
2277 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2278 flkeys = &_flkeys;
2279 } else {
2280 fl4.flowi4_proto = 0;
2281 fl4.fl4_sport = 0;
2282 fl4.fl4_dport = 0;
2283 }
2284
2285 err = fib_lookup(net, &fl4, res, 0);
2286 if (err != 0) {
2287 if (!IN_DEV_FORWARD(in_dev))
2288 err = -EHOSTUNREACH;
2289 goto no_route;
2290 }
2291
2292 if (res->type == RTN_BROADCAST) {
2293 if (IN_DEV_BFORWARD(in_dev))
2294 goto make_route;
2295 /* not do cache if bc_forwarding is enabled */
2296 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2297 do_cache = false;
2298 goto brd_input;
2299 }
2300
2301 if (res->type == RTN_LOCAL) {
2302 err = fib_validate_source(skb, saddr, daddr, tos,
2303 0, dev, in_dev, &itag);
2304 if (err < 0)
2305 goto martian_source;
2306 goto local_input;
2307 }
2308
2309 if (!IN_DEV_FORWARD(in_dev)) {
2310 err = -EHOSTUNREACH;
2311 goto no_route;
2312 }
2313 if (res->type != RTN_UNICAST)
2314 goto martian_destination;
2315
2316 make_route:
2317 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2318 out: return err;
2319
2320 brd_input:
2321 if (skb->protocol != htons(ETH_P_IP))
2322 goto e_inval;
2323
2324 if (!ipv4_is_zeronet(saddr)) {
2325 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2326 in_dev, &itag);
2327 if (err < 0)
2328 goto martian_source;
2329 }
2330 flags |= RTCF_BROADCAST;
2331 res->type = RTN_BROADCAST;
2332 RT_CACHE_STAT_INC(in_brd);
2333
2334 local_input:
2335 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2336 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2337
2338 do_cache &= res->fi && !itag;
2339 if (do_cache) {
2340 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2341
2342 rth = rcu_dereference(nhc->nhc_rth_input);
2343 if (rt_cache_valid(rth)) {
2344 skb_dst_set_noref(skb, &rth->dst);
2345 err = 0;
2346 goto out;
2347 }
2348 }
2349
2350 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2351 flags | RTCF_LOCAL, res->type, false);
2352 if (!rth)
2353 goto e_nobufs;
2354
2355 rth->dst.output= ip_rt_bug;
2356 #ifdef CONFIG_IP_ROUTE_CLASSID
2357 rth->dst.tclassid = itag;
2358 #endif
2359 rth->rt_is_input = 1;
2360
2361 RT_CACHE_STAT_INC(in_slow_tot);
2362 if (res->type == RTN_UNREACHABLE) {
2363 rth->dst.input= ip_error;
2364 rth->dst.error= -err;
2365 rth->rt_flags &= ~RTCF_LOCAL;
2366 }
2367
2368 if (do_cache) {
2369 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2370
2371 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2372 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2373 WARN_ON(rth->dst.input == lwtunnel_input);
2374 rth->dst.lwtstate->orig_input = rth->dst.input;
2375 rth->dst.input = lwtunnel_input;
2376 }
2377
2378 if (unlikely(!rt_cache_route(nhc, rth)))
2379 rt_add_uncached_list(rth);
2380 }
2381 skb_dst_set(skb, &rth->dst);
2382 err = 0;
2383 goto out;
2384
2385 no_route:
2386 RT_CACHE_STAT_INC(in_no_route);
2387 res->type = RTN_UNREACHABLE;
2388 res->fi = NULL;
2389 res->table = NULL;
2390 goto local_input;
2391
2392 /*
2393 * Do not cache martian addresses: they should be logged (RFC1812)
2394 */
2395 martian_destination:
2396 RT_CACHE_STAT_INC(in_martian_dst);
2397 #ifdef CONFIG_IP_ROUTE_VERBOSE
2398 if (IN_DEV_LOG_MARTIANS(in_dev))
2399 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2400 &daddr, &saddr, dev->name);
2401 #endif
2402
2403 e_inval:
2404 err = -EINVAL;
2405 goto out;
2406
2407 e_nobufs:
2408 err = -ENOBUFS;
2409 goto out;
2410
2411 martian_source:
2412 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2413 goto out;
2414 }
2415
2416 /* called with rcu_read_lock held */
2417 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2418 u8 tos, struct net_device *dev, struct fib_result *res)
2419 {
2420 /* Multicast recognition logic is moved from route cache to here.
2421 * The problem was that too many Ethernet cards have broken/missing
2422 * hardware multicast filters :-( As result the host on multicasting
2423 * network acquires a lot of useless route cache entries, sort of
2424 * SDR messages from all the world. Now we try to get rid of them.
2425 * Really, provided software IP multicast filter is organized
2426 * reasonably (at least, hashed), it does not result in a slowdown
2427 * comparing with route cache reject entries.
2428 * Note, that multicast routers are not affected, because
2429 * route cache entry is created eventually.
2430 */
2431 if (ipv4_is_multicast(daddr)) {
2432 struct in_device *in_dev = __in_dev_get_rcu(dev);
2433 int our = 0;
2434 int err = -EINVAL;
2435
2436 if (!in_dev)
2437 return err;
2438 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2439 ip_hdr(skb)->protocol);
2440
2441 /* check l3 master if no match yet */
2442 if (!our && netif_is_l3_slave(dev)) {
2443 struct in_device *l3_in_dev;
2444
2445 l3_in_dev = __in_dev_get_rcu(skb->dev);
2446 if (l3_in_dev)
2447 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2448 ip_hdr(skb)->protocol);
2449 }
2450
2451 if (our
2452 #ifdef CONFIG_IP_MROUTE
2453 ||
2454 (!ipv4_is_local_multicast(daddr) &&
2455 IN_DEV_MFORWARD(in_dev))
2456 #endif
2457 ) {
2458 err = ip_route_input_mc(skb, daddr, saddr,
2459 tos, dev, our);
2460 }
2461 return err;
2462 }
2463
2464 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2465 }
2466
2467 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2468 u8 tos, struct net_device *dev)
2469 {
2470 struct fib_result res;
2471 int err;
2472
2473 tos &= IPTOS_RT_MASK;
2474 rcu_read_lock();
2475 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2476 rcu_read_unlock();
2477
2478 return err;
2479 }
2480 EXPORT_SYMBOL(ip_route_input_noref);
2481
2482 /* called with rcu_read_lock() */
2483 static struct rtable *__mkroute_output(const struct fib_result *res,
2484 const struct flowi4 *fl4, int orig_oif,
2485 struct net_device *dev_out,
2486 unsigned int flags)
2487 {
2488 struct fib_info *fi = res->fi;
2489 struct fib_nh_exception *fnhe;
2490 struct in_device *in_dev;
2491 u16 type = res->type;
2492 struct rtable *rth;
2493 bool do_cache;
2494
2495 in_dev = __in_dev_get_rcu(dev_out);
2496 if (!in_dev)
2497 return ERR_PTR(-EINVAL);
2498
2499 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2500 if (ipv4_is_loopback(fl4->saddr) &&
2501 !(dev_out->flags & IFF_LOOPBACK) &&
2502 !netif_is_l3_master(dev_out))
2503 return ERR_PTR(-EINVAL);
2504
2505 if (ipv4_is_lbcast(fl4->daddr))
2506 type = RTN_BROADCAST;
2507 else if (ipv4_is_multicast(fl4->daddr))
2508 type = RTN_MULTICAST;
2509 else if (ipv4_is_zeronet(fl4->daddr))
2510 return ERR_PTR(-EINVAL);
2511
2512 if (dev_out->flags & IFF_LOOPBACK)
2513 flags |= RTCF_LOCAL;
2514
2515 do_cache = true;
2516 if (type == RTN_BROADCAST) {
2517 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2518 fi = NULL;
2519 } else if (type == RTN_MULTICAST) {
2520 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2521 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2522 fl4->flowi4_proto))
2523 flags &= ~RTCF_LOCAL;
2524 else
2525 do_cache = false;
2526 /* If multicast route do not exist use
2527 * default one, but do not gateway in this case.
2528 * Yes, it is hack.
2529 */
2530 if (fi && res->prefixlen < 4)
2531 fi = NULL;
2532 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2533 (orig_oif != dev_out->ifindex)) {
2534 /* For local routes that require a particular output interface
2535 * we do not want to cache the result. Caching the result
2536 * causes incorrect behaviour when there are multiple source
2537 * addresses on the interface, the end result being that if the
2538 * intended recipient is waiting on that interface for the
2539 * packet he won't receive it because it will be delivered on
2540 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2541 * be set to the loopback interface as well.
2542 */
2543 do_cache = false;
2544 }
2545
2546 fnhe = NULL;
2547 do_cache &= fi != NULL;
2548 if (fi) {
2549 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2550 struct rtable __rcu **prth;
2551
2552 fnhe = find_exception(nhc, fl4->daddr);
2553 if (!do_cache)
2554 goto add;
2555 if (fnhe) {
2556 prth = &fnhe->fnhe_rth_output;
2557 } else {
2558 if (unlikely(fl4->flowi4_flags &
2559 FLOWI_FLAG_KNOWN_NH &&
2560 !(nhc->nhc_gw_family &&
2561 nhc->nhc_scope == RT_SCOPE_LINK))) {
2562 do_cache = false;
2563 goto add;
2564 }
2565 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2566 }
2567 rth = rcu_dereference(*prth);
2568 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2569 return rth;
2570 }
2571
2572 add:
2573 rth = rt_dst_alloc(dev_out, flags, type,
2574 IN_DEV_ORCONF(in_dev, NOXFRM));
2575 if (!rth)
2576 return ERR_PTR(-ENOBUFS);
2577
2578 rth->rt_iif = orig_oif;
2579
2580 RT_CACHE_STAT_INC(out_slow_tot);
2581
2582 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2583 if (flags & RTCF_LOCAL &&
2584 !(dev_out->flags & IFF_LOOPBACK)) {
2585 rth->dst.output = ip_mc_output;
2586 RT_CACHE_STAT_INC(out_slow_mc);
2587 }
2588 #ifdef CONFIG_IP_MROUTE
2589 if (type == RTN_MULTICAST) {
2590 if (IN_DEV_MFORWARD(in_dev) &&
2591 !ipv4_is_local_multicast(fl4->daddr)) {
2592 rth->dst.input = ip_mr_input;
2593 rth->dst.output = ip_mc_output;
2594 }
2595 }
2596 #endif
2597 }
2598
2599 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2600 lwtunnel_set_redirect(&rth->dst);
2601
2602 return rth;
2603 }
2604
2605 /*
2606 * Major route resolver routine.
2607 */
2608
2609 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2610 const struct sk_buff *skb)
2611 {
2612 struct fib_result res = {
2613 .type = RTN_UNSPEC,
2614 .fi = NULL,
2615 .table = NULL,
2616 .tclassid = 0,
2617 };
2618 struct rtable *rth;
2619
2620 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2621 fl4->flowi4_tos &= IPTOS_RT_MASK;
2622
2623 rcu_read_lock();
2624 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2625 rcu_read_unlock();
2626
2627 return rth;
2628 }
2629 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2630
2631 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2632 struct fib_result *res,
2633 const struct sk_buff *skb)
2634 {
2635 struct net_device *dev_out = NULL;
2636 int orig_oif = fl4->flowi4_oif;
2637 unsigned int flags = 0;
2638 struct rtable *rth;
2639 int err;
2640
2641 if (fl4->saddr) {
2642 if (ipv4_is_multicast(fl4->saddr) ||
2643 ipv4_is_lbcast(fl4->saddr) ||
2644 ipv4_is_zeronet(fl4->saddr)) {
2645 rth = ERR_PTR(-EINVAL);
2646 goto out;
2647 }
2648
2649 rth = ERR_PTR(-ENETUNREACH);
2650
2651 /* I removed check for oif == dev_out->oif here.
2652 * It was wrong for two reasons:
2653 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2654 * is assigned to multiple interfaces.
2655 * 2. Moreover, we are allowed to send packets with saddr
2656 * of another iface. --ANK
2657 */
2658
2659 if (fl4->flowi4_oif == 0 &&
2660 (ipv4_is_multicast(fl4->daddr) ||
2661 ipv4_is_lbcast(fl4->daddr))) {
2662 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2663 dev_out = __ip_dev_find(net, fl4->saddr, false);
2664 if (!dev_out)
2665 goto out;
2666
2667 /* Special hack: user can direct multicasts
2668 * and limited broadcast via necessary interface
2669 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2670 * This hack is not just for fun, it allows
2671 * vic,vat and friends to work.
2672 * They bind socket to loopback, set ttl to zero
2673 * and expect that it will work.
2674 * From the viewpoint of routing cache they are broken,
2675 * because we are not allowed to build multicast path
2676 * with loopback source addr (look, routing cache
2677 * cannot know, that ttl is zero, so that packet
2678 * will not leave this host and route is valid).
2679 * Luckily, this hack is good workaround.
2680 */
2681
2682 fl4->flowi4_oif = dev_out->ifindex;
2683 goto make_route;
2684 }
2685
2686 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2687 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2688 if (!__ip_dev_find(net, fl4->saddr, false))
2689 goto out;
2690 }
2691 }
2692
2693
2694 if (fl4->flowi4_oif) {
2695 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2696 rth = ERR_PTR(-ENODEV);
2697 if (!dev_out)
2698 goto out;
2699
2700 /* RACE: Check return value of inet_select_addr instead. */
2701 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2702 rth = ERR_PTR(-ENETUNREACH);
2703 goto out;
2704 }
2705 if (ipv4_is_local_multicast(fl4->daddr) ||
2706 ipv4_is_lbcast(fl4->daddr) ||
2707 fl4->flowi4_proto == IPPROTO_IGMP) {
2708 if (!fl4->saddr)
2709 fl4->saddr = inet_select_addr(dev_out, 0,
2710 RT_SCOPE_LINK);
2711 goto make_route;
2712 }
2713 if (!fl4->saddr) {
2714 if (ipv4_is_multicast(fl4->daddr))
2715 fl4->saddr = inet_select_addr(dev_out, 0,
2716 fl4->flowi4_scope);
2717 else if (!fl4->daddr)
2718 fl4->saddr = inet_select_addr(dev_out, 0,
2719 RT_SCOPE_HOST);
2720 }
2721 }
2722
2723 if (!fl4->daddr) {
2724 fl4->daddr = fl4->saddr;
2725 if (!fl4->daddr)
2726 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2727 dev_out = net->loopback_dev;
2728 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2729 res->type = RTN_LOCAL;
2730 flags |= RTCF_LOCAL;
2731 goto make_route;
2732 }
2733
2734 err = fib_lookup(net, fl4, res, 0);
2735 if (err) {
2736 res->fi = NULL;
2737 res->table = NULL;
2738 if (fl4->flowi4_oif &&
2739 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2740 /* Apparently, routing tables are wrong. Assume,
2741 * that the destination is on link.
2742 *
2743 * WHY? DW.
2744 * Because we are allowed to send to iface
2745 * even if it has NO routes and NO assigned
2746 * addresses. When oif is specified, routing
2747 * tables are looked up with only one purpose:
2748 * to catch if destination is gatewayed, rather than
2749 * direct. Moreover, if MSG_DONTROUTE is set,
2750 * we send packet, ignoring both routing tables
2751 * and ifaddr state. --ANK
2752 *
2753 *
2754 * We could make it even if oif is unknown,
2755 * likely IPv6, but we do not.
2756 */
2757
2758 if (fl4->saddr == 0)
2759 fl4->saddr = inet_select_addr(dev_out, 0,
2760 RT_SCOPE_LINK);
2761 res->type = RTN_UNICAST;
2762 goto make_route;
2763 }
2764 rth = ERR_PTR(err);
2765 goto out;
2766 }
2767
2768 if (res->type == RTN_LOCAL) {
2769 if (!fl4->saddr) {
2770 if (res->fi->fib_prefsrc)
2771 fl4->saddr = res->fi->fib_prefsrc;
2772 else
2773 fl4->saddr = fl4->daddr;
2774 }
2775
2776 /* L3 master device is the loopback for that domain */
2777 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2778 net->loopback_dev;
2779
2780 /* make sure orig_oif points to fib result device even
2781 * though packet rx/tx happens over loopback or l3mdev
2782 */
2783 orig_oif = FIB_RES_OIF(*res);
2784
2785 fl4->flowi4_oif = dev_out->ifindex;
2786 flags |= RTCF_LOCAL;
2787 goto make_route;
2788 }
2789
2790 fib_select_path(net, res, fl4, skb);
2791
2792 dev_out = FIB_RES_DEV(*res);
2793
2794 make_route:
2795 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2796
2797 out:
2798 return rth;
2799 }
2800
2801 static struct dst_ops ipv4_dst_blackhole_ops = {
2802 .family = AF_INET,
2803 .default_advmss = ipv4_default_advmss,
2804 .neigh_lookup = ipv4_neigh_lookup,
2805 .check = dst_blackhole_check,
2806 .cow_metrics = dst_blackhole_cow_metrics,
2807 .update_pmtu = dst_blackhole_update_pmtu,
2808 .redirect = dst_blackhole_redirect,
2809 .mtu = dst_blackhole_mtu,
2810 };
2811
2812 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2813 {
2814 struct rtable *ort = dst_rtable(dst_orig);
2815 struct rtable *rt;
2816
2817 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2818 if (rt) {
2819 struct dst_entry *new = &rt->dst;
2820
2821 new->__use = 1;
2822 new->input = dst_discard;
2823 new->output = dst_discard_out;
2824
2825 new->dev = net->loopback_dev;
2826 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2827
2828 rt->rt_is_input = ort->rt_is_input;
2829 rt->rt_iif = ort->rt_iif;
2830 rt->rt_pmtu = ort->rt_pmtu;
2831 rt->rt_mtu_locked = ort->rt_mtu_locked;
2832
2833 rt->rt_genid = rt_genid_ipv4(net);
2834 rt->rt_flags = ort->rt_flags;
2835 rt->rt_type = ort->rt_type;
2836 rt->rt_uses_gateway = ort->rt_uses_gateway;
2837 rt->rt_gw_family = ort->rt_gw_family;
2838 if (rt->rt_gw_family == AF_INET)
2839 rt->rt_gw4 = ort->rt_gw4;
2840 else if (rt->rt_gw_family == AF_INET6)
2841 rt->rt_gw6 = ort->rt_gw6;
2842 }
2843
2844 dst_release(dst_orig);
2845
2846 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2847 }
2848
2849 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2850 const struct sock *sk)
2851 {
2852 struct rtable *rt = __ip_route_output_key(net, flp4);
2853
2854 if (IS_ERR(rt))
2855 return rt;
2856
2857 if (flp4->flowi4_proto) {
2858 flp4->flowi4_oif = rt->dst.dev->ifindex;
2859 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2860 flowi4_to_flowi(flp4),
2861 sk, 0));
2862 }
2863
2864 return rt;
2865 }
2866 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2867
2868 /* called with rcu_read_lock held */
2869 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2870 struct rtable *rt, u32 table_id, dscp_t dscp,
2871 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2872 u32 seq, unsigned int flags)
2873 {
2874 struct rtmsg *r;
2875 struct nlmsghdr *nlh;
2876 unsigned long expires = 0;
2877 u32 error;
2878 u32 metrics[RTAX_MAX];
2879
2880 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2881 if (!nlh)
2882 return -EMSGSIZE;
2883
2884 r = nlmsg_data(nlh);
2885 r->rtm_family = AF_INET;
2886 r->rtm_dst_len = 32;
2887 r->rtm_src_len = 0;
2888 r->rtm_tos = inet_dscp_to_dsfield(dscp);
2889 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2890 if (nla_put_u32(skb, RTA_TABLE, table_id))
2891 goto nla_put_failure;
2892 r->rtm_type = rt->rt_type;
2893 r->rtm_scope = RT_SCOPE_UNIVERSE;
2894 r->rtm_protocol = RTPROT_UNSPEC;
2895 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2896 if (rt->rt_flags & RTCF_NOTIFY)
2897 r->rtm_flags |= RTM_F_NOTIFY;
2898 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2899 r->rtm_flags |= RTCF_DOREDIRECT;
2900
2901 if (nla_put_in_addr(skb, RTA_DST, dst))
2902 goto nla_put_failure;
2903 if (src) {
2904 r->rtm_src_len = 32;
2905 if (nla_put_in_addr(skb, RTA_SRC, src))
2906 goto nla_put_failure;
2907 }
2908 if (rt->dst.dev &&
2909 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2910 goto nla_put_failure;
2911 if (rt->dst.lwtstate &&
2912 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2913 goto nla_put_failure;
2914 #ifdef CONFIG_IP_ROUTE_CLASSID
2915 if (rt->dst.tclassid &&
2916 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2917 goto nla_put_failure;
2918 #endif
2919 if (fl4 && !rt_is_input_route(rt) &&
2920 fl4->saddr != src) {
2921 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2922 goto nla_put_failure;
2923 }
2924 if (rt->rt_uses_gateway) {
2925 if (rt->rt_gw_family == AF_INET &&
2926 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2927 goto nla_put_failure;
2928 } else if (rt->rt_gw_family == AF_INET6) {
2929 int alen = sizeof(struct in6_addr);
2930 struct nlattr *nla;
2931 struct rtvia *via;
2932
2933 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2934 if (!nla)
2935 goto nla_put_failure;
2936
2937 via = nla_data(nla);
2938 via->rtvia_family = AF_INET6;
2939 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2940 }
2941 }
2942
2943 expires = rt->dst.expires;
2944 if (expires) {
2945 unsigned long now = jiffies;
2946
2947 if (time_before(now, expires))
2948 expires -= now;
2949 else
2950 expires = 0;
2951 }
2952
2953 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2954 if (rt->rt_pmtu && expires)
2955 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2956 if (rt->rt_mtu_locked && expires)
2957 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2958 if (rtnetlink_put_metrics(skb, metrics) < 0)
2959 goto nla_put_failure;
2960
2961 if (fl4) {
2962 if (fl4->flowi4_mark &&
2963 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2964 goto nla_put_failure;
2965
2966 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2967 nla_put_u32(skb, RTA_UID,
2968 from_kuid_munged(current_user_ns(),
2969 fl4->flowi4_uid)))
2970 goto nla_put_failure;
2971
2972 if (rt_is_input_route(rt)) {
2973 #ifdef CONFIG_IP_MROUTE
2974 if (ipv4_is_multicast(dst) &&
2975 !ipv4_is_local_multicast(dst) &&
2976 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
2977 int err = ipmr_get_route(net, skb,
2978 fl4->saddr, fl4->daddr,
2979 r, portid);
2980
2981 if (err <= 0) {
2982 if (err == 0)
2983 return 0;
2984 goto nla_put_failure;
2985 }
2986 } else
2987 #endif
2988 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2989 goto nla_put_failure;
2990 }
2991 }
2992
2993 error = rt->dst.error;
2994
2995 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2996 goto nla_put_failure;
2997
2998 nlmsg_end(skb, nlh);
2999 return 0;
3000
3001 nla_put_failure:
3002 nlmsg_cancel(skb, nlh);
3003 return -EMSGSIZE;
3004 }
3005
3006 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3007 struct netlink_callback *cb, u32 table_id,
3008 struct fnhe_hash_bucket *bucket, int genid,
3009 int *fa_index, int fa_start, unsigned int flags)
3010 {
3011 int i;
3012
3013 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3014 struct fib_nh_exception *fnhe;
3015
3016 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3017 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3018 struct rtable *rt;
3019 int err;
3020
3021 if (*fa_index < fa_start)
3022 goto next;
3023
3024 if (fnhe->fnhe_genid != genid)
3025 goto next;
3026
3027 if (fnhe->fnhe_expires &&
3028 time_after(jiffies, fnhe->fnhe_expires))
3029 goto next;
3030
3031 rt = rcu_dereference(fnhe->fnhe_rth_input);
3032 if (!rt)
3033 rt = rcu_dereference(fnhe->fnhe_rth_output);
3034 if (!rt)
3035 goto next;
3036
3037 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3038 table_id, 0, NULL, skb,
3039 NETLINK_CB(cb->skb).portid,
3040 cb->nlh->nlmsg_seq, flags);
3041 if (err)
3042 return err;
3043 next:
3044 (*fa_index)++;
3045 }
3046 }
3047
3048 return 0;
3049 }
3050
3051 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3052 u32 table_id, struct fib_info *fi,
3053 int *fa_index, int fa_start, unsigned int flags)
3054 {
3055 struct net *net = sock_net(cb->skb->sk);
3056 int nhsel, genid = fnhe_genid(net);
3057
3058 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3059 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3060 struct fnhe_hash_bucket *bucket;
3061 int err;
3062
3063 if (nhc->nhc_flags & RTNH_F_DEAD)
3064 continue;
3065
3066 rcu_read_lock();
3067 bucket = rcu_dereference(nhc->nhc_exceptions);
3068 err = 0;
3069 if (bucket)
3070 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3071 genid, fa_index, fa_start,
3072 flags);
3073 rcu_read_unlock();
3074 if (err)
3075 return err;
3076 }
3077
3078 return 0;
3079 }
3080
3081 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3082 u8 ip_proto, __be16 sport,
3083 __be16 dport)
3084 {
3085 struct sk_buff *skb;
3086 struct iphdr *iph;
3087
3088 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3089 if (!skb)
3090 return NULL;
3091
3092 /* Reserve room for dummy headers, this skb can pass
3093 * through good chunk of routing engine.
3094 */
3095 skb_reset_mac_header(skb);
3096 skb_reset_network_header(skb);
3097 skb->protocol = htons(ETH_P_IP);
3098 iph = skb_put(skb, sizeof(struct iphdr));
3099 iph->protocol = ip_proto;
3100 iph->saddr = src;
3101 iph->daddr = dst;
3102 iph->version = 0x4;
3103 iph->frag_off = 0;
3104 iph->ihl = 0x5;
3105 skb_set_transport_header(skb, skb->len);
3106
3107 switch (iph->protocol) {
3108 case IPPROTO_UDP: {
3109 struct udphdr *udph;
3110
3111 udph = skb_put_zero(skb, sizeof(struct udphdr));
3112 udph->source = sport;
3113 udph->dest = dport;
3114 udph->len = htons(sizeof(struct udphdr));
3115 udph->check = 0;
3116 break;
3117 }
3118 case IPPROTO_TCP: {
3119 struct tcphdr *tcph;
3120
3121 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3122 tcph->source = sport;
3123 tcph->dest = dport;
3124 tcph->doff = sizeof(struct tcphdr) / 4;
3125 tcph->rst = 1;
3126 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3127 src, dst, 0);
3128 break;
3129 }
3130 case IPPROTO_ICMP: {
3131 struct icmphdr *icmph;
3132
3133 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3134 icmph->type = ICMP_ECHO;
3135 icmph->code = 0;
3136 }
3137 }
3138
3139 return skb;
3140 }
3141
3142 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3143 const struct nlmsghdr *nlh,
3144 struct nlattr **tb,
3145 struct netlink_ext_ack *extack)
3146 {
3147 struct rtmsg *rtm;
3148 int i, err;
3149
3150 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3151 NL_SET_ERR_MSG(extack,
3152 "ipv4: Invalid header for route get request");
3153 return -EINVAL;
3154 }
3155
3156 if (!netlink_strict_get_check(skb))
3157 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3158 rtm_ipv4_policy, extack);
3159
3160 rtm = nlmsg_data(nlh);
3161 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3162 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3163 rtm->rtm_table || rtm->rtm_protocol ||
3164 rtm->rtm_scope || rtm->rtm_type) {
3165 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3166 return -EINVAL;
3167 }
3168
3169 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3170 RTM_F_LOOKUP_TABLE |
3171 RTM_F_FIB_MATCH)) {
3172 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3173 return -EINVAL;
3174 }
3175
3176 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3177 rtm_ipv4_policy, extack);
3178 if (err)
3179 return err;
3180
3181 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3182 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3183 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3184 return -EINVAL;
3185 }
3186
3187 for (i = 0; i <= RTA_MAX; i++) {
3188 if (!tb[i])
3189 continue;
3190
3191 switch (i) {
3192 case RTA_IIF:
3193 case RTA_OIF:
3194 case RTA_SRC:
3195 case RTA_DST:
3196 case RTA_IP_PROTO:
3197 case RTA_SPORT:
3198 case RTA_DPORT:
3199 case RTA_MARK:
3200 case RTA_UID:
3201 break;
3202 default:
3203 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3204 return -EINVAL;
3205 }
3206 }
3207
3208 return 0;
3209 }
3210
3211 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3212 struct netlink_ext_ack *extack)
3213 {
3214 struct net *net = sock_net(in_skb->sk);
3215 struct nlattr *tb[RTA_MAX+1];
3216 u32 table_id = RT_TABLE_MAIN;
3217 __be16 sport = 0, dport = 0;
3218 struct fib_result res = {};
3219 u8 ip_proto = IPPROTO_UDP;
3220 struct rtable *rt = NULL;
3221 struct sk_buff *skb;
3222 struct rtmsg *rtm;
3223 struct flowi4 fl4 = {};
3224 __be32 dst = 0;
3225 __be32 src = 0;
3226 kuid_t uid;
3227 u32 iif;
3228 int err;
3229 int mark;
3230
3231 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3232 if (err < 0)
3233 return err;
3234
3235 rtm = nlmsg_data(nlh);
3236 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3237 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3238 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3239 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3240 if (tb[RTA_UID])
3241 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3242 else
3243 uid = (iif ? INVALID_UID : current_uid());
3244
3245 if (tb[RTA_IP_PROTO]) {
3246 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3247 &ip_proto, AF_INET, extack);
3248 if (err)
3249 return err;
3250 }
3251
3252 if (tb[RTA_SPORT])
3253 sport = nla_get_be16(tb[RTA_SPORT]);
3254
3255 if (tb[RTA_DPORT])
3256 dport = nla_get_be16(tb[RTA_DPORT]);
3257
3258 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3259 if (!skb)
3260 return -ENOBUFS;
3261
3262 fl4.daddr = dst;
3263 fl4.saddr = src;
3264 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3265 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3266 fl4.flowi4_mark = mark;
3267 fl4.flowi4_uid = uid;
3268 if (sport)
3269 fl4.fl4_sport = sport;
3270 if (dport)
3271 fl4.fl4_dport = dport;
3272 fl4.flowi4_proto = ip_proto;
3273
3274 rcu_read_lock();
3275
3276 if (iif) {
3277 struct net_device *dev;
3278
3279 dev = dev_get_by_index_rcu(net, iif);
3280 if (!dev) {
3281 err = -ENODEV;
3282 goto errout_rcu;
3283 }
3284
3285 fl4.flowi4_iif = iif; /* for rt_fill_info */
3286 skb->dev = dev;
3287 skb->mark = mark;
3288 err = ip_route_input_rcu(skb, dst, src,
3289 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3290 &res);
3291
3292 rt = skb_rtable(skb);
3293 if (err == 0 && rt->dst.error)
3294 err = -rt->dst.error;
3295 } else {
3296 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3297 skb->dev = net->loopback_dev;
3298 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3299 err = 0;
3300 if (IS_ERR(rt))
3301 err = PTR_ERR(rt);
3302 else
3303 skb_dst_set(skb, &rt->dst);
3304 }
3305
3306 if (err)
3307 goto errout_rcu;
3308
3309 if (rtm->rtm_flags & RTM_F_NOTIFY)
3310 rt->rt_flags |= RTCF_NOTIFY;
3311
3312 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3313 table_id = res.table ? res.table->tb_id : 0;
3314
3315 /* reset skb for netlink reply msg */
3316 skb_trim(skb, 0);
3317 skb_reset_network_header(skb);
3318 skb_reset_transport_header(skb);
3319 skb_reset_mac_header(skb);
3320
3321 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3322 struct fib_rt_info fri;
3323
3324 if (!res.fi) {
3325 err = fib_props[res.type].error;
3326 if (!err)
3327 err = -EHOSTUNREACH;
3328 goto errout_rcu;
3329 }
3330 fri.fi = res.fi;
3331 fri.tb_id = table_id;
3332 fri.dst = res.prefix;
3333 fri.dst_len = res.prefixlen;
3334 fri.dscp = res.dscp;
3335 fri.type = rt->rt_type;
3336 fri.offload = 0;
3337 fri.trap = 0;
3338 fri.offload_failed = 0;
3339 if (res.fa_head) {
3340 struct fib_alias *fa;
3341
3342 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3343 u8 slen = 32 - fri.dst_len;
3344
3345 if (fa->fa_slen == slen &&
3346 fa->tb_id == fri.tb_id &&
3347 fa->fa_dscp == fri.dscp &&
3348 fa->fa_info == res.fi &&
3349 fa->fa_type == fri.type) {
3350 fri.offload = READ_ONCE(fa->offload);
3351 fri.trap = READ_ONCE(fa->trap);
3352 fri.offload_failed =
3353 READ_ONCE(fa->offload_failed);
3354 break;
3355 }
3356 }
3357 }
3358 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3359 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3360 } else {
3361 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3362 skb, NETLINK_CB(in_skb).portid,
3363 nlh->nlmsg_seq, 0);
3364 }
3365 if (err < 0)
3366 goto errout_rcu;
3367
3368 rcu_read_unlock();
3369
3370 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3371
3372 errout_free:
3373 return err;
3374 errout_rcu:
3375 rcu_read_unlock();
3376 kfree_skb(skb);
3377 goto errout_free;
3378 }
3379
3380 void ip_rt_multicast_event(struct in_device *in_dev)
3381 {
3382 rt_cache_flush(dev_net(in_dev->dev));
3383 }
3384
3385 #ifdef CONFIG_SYSCTL
3386 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3387 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3388 static int ip_rt_gc_elasticity __read_mostly = 8;
3389 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3390
3391 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3392 void *buffer, size_t *lenp, loff_t *ppos)
3393 {
3394 struct net *net = (struct net *)__ctl->extra1;
3395
3396 if (write) {
3397 rt_cache_flush(net);
3398 fnhe_genid_bump(net);
3399 return 0;
3400 }
3401
3402 return -EINVAL;
3403 }
3404
3405 static struct ctl_table ipv4_route_table[] = {
3406 {
3407 .procname = "gc_thresh",
3408 .data = &ipv4_dst_ops.gc_thresh,
3409 .maxlen = sizeof(int),
3410 .mode = 0644,
3411 .proc_handler = proc_dointvec,
3412 },
3413 {
3414 .procname = "max_size",
3415 .data = &ip_rt_max_size,
3416 .maxlen = sizeof(int),
3417 .mode = 0644,
3418 .proc_handler = proc_dointvec,
3419 },
3420 {
3421 /* Deprecated. Use gc_min_interval_ms */
3422
3423 .procname = "gc_min_interval",
3424 .data = &ip_rt_gc_min_interval,
3425 .maxlen = sizeof(int),
3426 .mode = 0644,
3427 .proc_handler = proc_dointvec_jiffies,
3428 },
3429 {
3430 .procname = "gc_min_interval_ms",
3431 .data = &ip_rt_gc_min_interval,
3432 .maxlen = sizeof(int),
3433 .mode = 0644,
3434 .proc_handler = proc_dointvec_ms_jiffies,
3435 },
3436 {
3437 .procname = "gc_timeout",
3438 .data = &ip_rt_gc_timeout,
3439 .maxlen = sizeof(int),
3440 .mode = 0644,
3441 .proc_handler = proc_dointvec_jiffies,
3442 },
3443 {
3444 .procname = "gc_interval",
3445 .data = &ip_rt_gc_interval,
3446 .maxlen = sizeof(int),
3447 .mode = 0644,
3448 .proc_handler = proc_dointvec_jiffies,
3449 },
3450 {
3451 .procname = "redirect_load",
3452 .data = &ip_rt_redirect_load,
3453 .maxlen = sizeof(int),
3454 .mode = 0644,
3455 .proc_handler = proc_dointvec,
3456 },
3457 {
3458 .procname = "redirect_number",
3459 .data = &ip_rt_redirect_number,
3460 .maxlen = sizeof(int),
3461 .mode = 0644,
3462 .proc_handler = proc_dointvec,
3463 },
3464 {
3465 .procname = "redirect_silence",
3466 .data = &ip_rt_redirect_silence,
3467 .maxlen = sizeof(int),
3468 .mode = 0644,
3469 .proc_handler = proc_dointvec,
3470 },
3471 {
3472 .procname = "error_cost",
3473 .data = &ip_rt_error_cost,
3474 .maxlen = sizeof(int),
3475 .mode = 0644,
3476 .proc_handler = proc_dointvec,
3477 },
3478 {
3479 .procname = "error_burst",
3480 .data = &ip_rt_error_burst,
3481 .maxlen = sizeof(int),
3482 .mode = 0644,
3483 .proc_handler = proc_dointvec,
3484 },
3485 {
3486 .procname = "gc_elasticity",
3487 .data = &ip_rt_gc_elasticity,
3488 .maxlen = sizeof(int),
3489 .mode = 0644,
3490 .proc_handler = proc_dointvec,
3491 },
3492 };
3493
3494 static const char ipv4_route_flush_procname[] = "flush";
3495
3496 static struct ctl_table ipv4_route_netns_table[] = {
3497 {
3498 .procname = ipv4_route_flush_procname,
3499 .maxlen = sizeof(int),
3500 .mode = 0200,
3501 .proc_handler = ipv4_sysctl_rtcache_flush,
3502 },
3503 {
3504 .procname = "min_pmtu",
3505 .data = &init_net.ipv4.ip_rt_min_pmtu,
3506 .maxlen = sizeof(int),
3507 .mode = 0644,
3508 .proc_handler = proc_dointvec_minmax,
3509 .extra1 = &ip_min_valid_pmtu,
3510 },
3511 {
3512 .procname = "mtu_expires",
3513 .data = &init_net.ipv4.ip_rt_mtu_expires,
3514 .maxlen = sizeof(int),
3515 .mode = 0644,
3516 .proc_handler = proc_dointvec_jiffies,
3517 },
3518 {
3519 .procname = "min_adv_mss",
3520 .data = &init_net.ipv4.ip_rt_min_advmss,
3521 .maxlen = sizeof(int),
3522 .mode = 0644,
3523 .proc_handler = proc_dointvec,
3524 },
3525 };
3526
3527 static __net_init int sysctl_route_net_init(struct net *net)
3528 {
3529 struct ctl_table *tbl;
3530 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3531
3532 tbl = ipv4_route_netns_table;
3533 if (!net_eq(net, &init_net)) {
3534 int i;
3535
3536 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3537 if (!tbl)
3538 goto err_dup;
3539
3540 /* Don't export non-whitelisted sysctls to unprivileged users */
3541 if (net->user_ns != &init_user_ns) {
3542 if (tbl[0].procname != ipv4_route_flush_procname)
3543 table_size = 0;
3544 }
3545
3546 /* Update the variables to point into the current struct net
3547 * except for the first element flush
3548 */
3549 for (i = 1; i < table_size; i++)
3550 tbl[i].data += (void *)net - (void *)&init_net;
3551 }
3552 tbl[0].extra1 = net;
3553
3554 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3555 tbl, table_size);
3556 if (!net->ipv4.route_hdr)
3557 goto err_reg;
3558 return 0;
3559
3560 err_reg:
3561 if (tbl != ipv4_route_netns_table)
3562 kfree(tbl);
3563 err_dup:
3564 return -ENOMEM;
3565 }
3566
3567 static __net_exit void sysctl_route_net_exit(struct net *net)
3568 {
3569 const struct ctl_table *tbl;
3570
3571 tbl = net->ipv4.route_hdr->ctl_table_arg;
3572 unregister_net_sysctl_table(net->ipv4.route_hdr);
3573 BUG_ON(tbl == ipv4_route_netns_table);
3574 kfree(tbl);
3575 }
3576
3577 static __net_initdata struct pernet_operations sysctl_route_ops = {
3578 .init = sysctl_route_net_init,
3579 .exit = sysctl_route_net_exit,
3580 };
3581 #endif
3582
3583 static __net_init int netns_ip_rt_init(struct net *net)
3584 {
3585 /* Set default value for namespaceified sysctls */
3586 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3587 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3588 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3589 return 0;
3590 }
3591
3592 static struct pernet_operations __net_initdata ip_rt_ops = {
3593 .init = netns_ip_rt_init,
3594 };
3595
3596 static __net_init int rt_genid_init(struct net *net)
3597 {
3598 atomic_set(&net->ipv4.rt_genid, 0);
3599 atomic_set(&net->fnhe_genid, 0);
3600 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3601 return 0;
3602 }
3603
3604 static __net_initdata struct pernet_operations rt_genid_ops = {
3605 .init = rt_genid_init,
3606 };
3607
3608 static int __net_init ipv4_inetpeer_init(struct net *net)
3609 {
3610 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3611
3612 if (!bp)
3613 return -ENOMEM;
3614 inet_peer_base_init(bp);
3615 net->ipv4.peers = bp;
3616 return 0;
3617 }
3618
3619 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3620 {
3621 struct inet_peer_base *bp = net->ipv4.peers;
3622
3623 net->ipv4.peers = NULL;
3624 inetpeer_invalidate_tree(bp);
3625 kfree(bp);
3626 }
3627
3628 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3629 .init = ipv4_inetpeer_init,
3630 .exit = ipv4_inetpeer_exit,
3631 };
3632
3633 #ifdef CONFIG_IP_ROUTE_CLASSID
3634 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3635 #endif /* CONFIG_IP_ROUTE_CLASSID */
3636
3637 int __init ip_rt_init(void)
3638 {
3639 void *idents_hash;
3640 int cpu;
3641
3642 /* For modern hosts, this will use 2 MB of memory */
3643 idents_hash = alloc_large_system_hash("IP idents",
3644 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3645 0,
3646 16, /* one bucket per 64 KB */
3647 HASH_ZERO,
3648 NULL,
3649 &ip_idents_mask,
3650 2048,
3651 256*1024);
3652
3653 ip_idents = idents_hash;
3654
3655 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3656
3657 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3658
3659 for_each_possible_cpu(cpu) {
3660 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3661
3662 INIT_LIST_HEAD(&ul->head);
3663 spin_lock_init(&ul->lock);
3664 }
3665 #ifdef CONFIG_IP_ROUTE_CLASSID
3666 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3667 if (!ip_rt_acct)
3668 panic("IP: failed to allocate ip_rt_acct\n");
3669 #endif
3670
3671 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3672 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3673
3674 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3675
3676 if (dst_entries_init(&ipv4_dst_ops) < 0)
3677 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3678
3679 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3680 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3681
3682 ipv4_dst_ops.gc_thresh = ~0;
3683 ip_rt_max_size = INT_MAX;
3684
3685 devinet_init();
3686 ip_fib_init();
3687
3688 if (ip_rt_proc_init())
3689 pr_err("Unable to create route proc files\n");
3690 #ifdef CONFIG_XFRM
3691 xfrm_init();
3692 xfrm4_init();
3693 #endif
3694 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3695 RTNL_FLAG_DOIT_UNLOCKED);
3696
3697 #ifdef CONFIG_SYSCTL
3698 register_pernet_subsys(&sysctl_route_ops);
3699 #endif
3700 register_pernet_subsys(&ip_rt_ops);
3701 register_pernet_subsys(&rt_genid_ops);
3702 register_pernet_subsys(&ipv4_inetpeer_ops);
3703 return 0;
3704 }
3705
3706 #ifdef CONFIG_SYSCTL
3707 /*
3708 * We really need to sanitize the damn ipv4 init order, then all
3709 * this nonsense will go away.
3710 */
3711 void __init ip_static_sysctl_init(void)
3712 {
3713 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3714 }
3715 #endif
3716
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.