Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * ROUTE - implementation of the IP router.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14 *
15 * Fixes:
16 * Alan Cox : Verify area fixes.
17 * Alan Cox : cli() protects routing changes
18 * Rui Oliveira : ICMP routing table updates
19 * (rco@di.uminho.pt) Routing table insertion and update
20 * Linus Torvalds : Rewrote bits to be sensible
21 * Alan Cox : Added BSD route gw semantics
22 * Alan Cox : Super /proc >4K
23 * Alan Cox : MTU in route table
24 * Alan Cox : MSS actually. Also added the window
25 * clamper.
26 * Sam Lantinga : Fixed route matching in rt_del()
27 * Alan Cox : Routing cache support.
28 * Alan Cox : Removed compatibility cruft.
29 * Alan Cox : RTF_REJECT support.
30 * Alan Cox : TCP irtt support.
31 * Jonathan Naylor : Added Metric support.
32 * Miquel van Smoorenburg : BSD API fixes.
33 * Miquel van Smoorenburg : Metrics.
34 * Alan Cox : Use __u32 properly
35 * Alan Cox : Aligned routing errors more closely with BSD
36 * our system is still very different.
37 * Alan Cox : Faster /proc handling
38 * Alexey Kuznetsov : Massive rework to support tree based routing,
39 * routing caches and better behaviour.
40 *
41 * Olaf Erb : irtt wasn't being copied right.
42 * Bjorn Ekwall : Kerneld route support.
43 * Alan Cox : Multicast fixed (I hope)
44 * Pavel Krauz : Limited broadcast fixed
45 * Mike McLagan : Routing by source
46 * Alexey Kuznetsov : End of old history. Split to fib.c and
47 * route.c and rewritten from scratch.
48 * Andi Kleen : Load-limit warning messages.
49 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
50 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
51 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
52 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
53 * Marc Boucher : routing by fwmark
54 * Robert Olsson : Added rt_cache statistics
55 * Arnaldo C. Melo : Convert proc stuff to seq_file
56 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
57 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
58 * Ilia Sotnikov : Removed TOS from hash calculations
59 */
60
61#define pr_fmt(fmt) "IPv4: " fmt
62
63#include <linux/module.h>
64#include <linux/uaccess.h>
65#include <linux/bitops.h>
66#include <linux/types.h>
67#include <linux/kernel.h>
68#include <linux/mm.h>
69#include <linux/memblock.h>
70#include <linux/string.h>
71#include <linux/socket.h>
72#include <linux/sockios.h>
73#include <linux/errno.h>
74#include <linux/in.h>
75#include <linux/inet.h>
76#include <linux/netdevice.h>
77#include <linux/proc_fs.h>
78#include <linux/init.h>
79#include <linux/skbuff.h>
80#include <linux/inetdevice.h>
81#include <linux/igmp.h>
82#include <linux/pkt_sched.h>
83#include <linux/mroute.h>
84#include <linux/netfilter_ipv4.h>
85#include <linux/random.h>
86#include <linux/rcupdate.h>
87#include <linux/times.h>
88#include <linux/slab.h>
89#include <linux/jhash.h>
90#include <net/dst.h>
91#include <net/dst_metadata.h>
92#include <net/net_namespace.h>
93#include <net/protocol.h>
94#include <net/ip.h>
95#include <net/route.h>
96#include <net/inetpeer.h>
97#include <net/sock.h>
98#include <net/ip_fib.h>
99#include <net/nexthop.h>
100#include <net/arp.h>
101#include <net/tcp.h>
102#include <net/icmp.h>
103#include <net/xfrm.h>
104#include <net/lwtunnel.h>
105#include <net/netevent.h>
106#include <net/rtnetlink.h>
107#ifdef CONFIG_SYSCTL
108#include <linux/sysctl.h>
109#endif
110#include <net/secure_seq.h>
111#include <net/ip_tunnels.h>
112#include <net/l3mdev.h>
113
114#include "fib_lookup.h"
115
116#define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
118
119#define RT_GC_TIMEOUT (300*HZ)
120
121static int ip_rt_max_size;
122static int ip_rt_redirect_number __read_mostly = 9;
123static int ip_rt_redirect_load __read_mostly = HZ / 50;
124static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
125static int ip_rt_error_cost __read_mostly = HZ;
126static int ip_rt_error_burst __read_mostly = 5 * HZ;
127static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
128static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
129static int ip_rt_min_advmss __read_mostly = 256;
130
131static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
132
133/*
134 * Interface to generic destination cache.
135 */
136
137INDIRECT_CALLABLE_SCOPE
138struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
139static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
140INDIRECT_CALLABLE_SCOPE
141unsigned int ipv4_mtu(const struct dst_entry *dst);
142static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143static void ipv4_link_failure(struct sk_buff *skb);
144static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
145 struct sk_buff *skb, u32 mtu,
146 bool confirm_neigh);
147static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
148 struct sk_buff *skb);
149static void ipv4_dst_destroy(struct dst_entry *dst);
150
151static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
152{
153 WARN_ON(1);
154 return NULL;
155}
156
157static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
158 struct sk_buff *skb,
159 const void *daddr);
160static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
161
162static struct dst_ops ipv4_dst_ops = {
163 .family = AF_INET,
164 .check = ipv4_dst_check,
165 .default_advmss = ipv4_default_advmss,
166 .mtu = ipv4_mtu,
167 .cow_metrics = ipv4_cow_metrics,
168 .destroy = ipv4_dst_destroy,
169 .negative_advice = ipv4_negative_advice,
170 .link_failure = ipv4_link_failure,
171 .update_pmtu = ip_rt_update_pmtu,
172 .redirect = ip_do_redirect,
173 .local_out = __ip_local_out,
174 .neigh_lookup = ipv4_neigh_lookup,
175 .confirm_neigh = ipv4_confirm_neigh,
176};
177
178#define ECN_OR_COST(class) TC_PRIO_##class
179
180const __u8 ip_tos2prio[16] = {
181 TC_PRIO_BESTEFFORT,
182 ECN_OR_COST(BESTEFFORT),
183 TC_PRIO_BESTEFFORT,
184 ECN_OR_COST(BESTEFFORT),
185 TC_PRIO_BULK,
186 ECN_OR_COST(BULK),
187 TC_PRIO_BULK,
188 ECN_OR_COST(BULK),
189 TC_PRIO_INTERACTIVE,
190 ECN_OR_COST(INTERACTIVE),
191 TC_PRIO_INTERACTIVE,
192 ECN_OR_COST(INTERACTIVE),
193 TC_PRIO_INTERACTIVE_BULK,
194 ECN_OR_COST(INTERACTIVE_BULK),
195 TC_PRIO_INTERACTIVE_BULK,
196 ECN_OR_COST(INTERACTIVE_BULK)
197};
198EXPORT_SYMBOL(ip_tos2prio);
199
200static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
201#define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
202
203#ifdef CONFIG_PROC_FS
204static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
205{
206 if (*pos)
207 return NULL;
208 return SEQ_START_TOKEN;
209}
210
211static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
212{
213 ++*pos;
214 return NULL;
215}
216
217static void rt_cache_seq_stop(struct seq_file *seq, void *v)
218{
219}
220
221static int rt_cache_seq_show(struct seq_file *seq, void *v)
222{
223 if (v == SEQ_START_TOKEN)
224 seq_printf(seq, "%-127s\n",
225 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
226 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
227 "HHUptod\tSpecDst");
228 return 0;
229}
230
231static const struct seq_operations rt_cache_seq_ops = {
232 .start = rt_cache_seq_start,
233 .next = rt_cache_seq_next,
234 .stop = rt_cache_seq_stop,
235 .show = rt_cache_seq_show,
236};
237
238static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
239{
240 int cpu;
241
242 if (*pos == 0)
243 return SEQ_START_TOKEN;
244
245 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
246 if (!cpu_possible(cpu))
247 continue;
248 *pos = cpu+1;
249 return &per_cpu(rt_cache_stat, cpu);
250 }
251 return NULL;
252}
253
254static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
255{
256 int cpu;
257
258 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
259 if (!cpu_possible(cpu))
260 continue;
261 *pos = cpu+1;
262 return &per_cpu(rt_cache_stat, cpu);
263 }
264 (*pos)++;
265 return NULL;
266
267}
268
269static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
270{
271
272}
273
274static int rt_cpu_seq_show(struct seq_file *seq, void *v)
275{
276 struct rt_cache_stat *st = v;
277
278 if (v == SEQ_START_TOKEN) {
279 seq_printf(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
280 return 0;
281 }
282
283 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
284 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
285 dst_entries_get_slow(&ipv4_dst_ops),
286 0, /* st->in_hit */
287 st->in_slow_tot,
288 st->in_slow_mc,
289 st->in_no_route,
290 st->in_brd,
291 st->in_martian_dst,
292 st->in_martian_src,
293
294 0, /* st->out_hit */
295 st->out_slow_tot,
296 st->out_slow_mc,
297
298 0, /* st->gc_total */
299 0, /* st->gc_ignored */
300 0, /* st->gc_goal_miss */
301 0, /* st->gc_dst_overflow */
302 0, /* st->in_hlist_search */
303 0 /* st->out_hlist_search */
304 );
305 return 0;
306}
307
308static const struct seq_operations rt_cpu_seq_ops = {
309 .start = rt_cpu_seq_start,
310 .next = rt_cpu_seq_next,
311 .stop = rt_cpu_seq_stop,
312 .show = rt_cpu_seq_show,
313};
314
315#ifdef CONFIG_IP_ROUTE_CLASSID
316static int rt_acct_proc_show(struct seq_file *m, void *v)
317{
318 struct ip_rt_acct *dst, *src;
319 unsigned int i, j;
320
321 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
322 if (!dst)
323 return -ENOMEM;
324
325 for_each_possible_cpu(i) {
326 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
327 for (j = 0; j < 256; j++) {
328 dst[j].o_bytes += src[j].o_bytes;
329 dst[j].o_packets += src[j].o_packets;
330 dst[j].i_bytes += src[j].i_bytes;
331 dst[j].i_packets += src[j].i_packets;
332 }
333 }
334
335 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
336 kfree(dst);
337 return 0;
338}
339#endif
340
341static int __net_init ip_rt_do_proc_init(struct net *net)
342{
343 struct proc_dir_entry *pde;
344
345 pde = proc_create_seq("rt_cache", 0444, net->proc_net,
346 &rt_cache_seq_ops);
347 if (!pde)
348 goto err1;
349
350 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
351 &rt_cpu_seq_ops);
352 if (!pde)
353 goto err2;
354
355#ifdef CONFIG_IP_ROUTE_CLASSID
356 pde = proc_create_single("rt_acct", 0, net->proc_net,
357 rt_acct_proc_show);
358 if (!pde)
359 goto err3;
360#endif
361 return 0;
362
363#ifdef CONFIG_IP_ROUTE_CLASSID
364err3:
365 remove_proc_entry("rt_cache", net->proc_net_stat);
366#endif
367err2:
368 remove_proc_entry("rt_cache", net->proc_net);
369err1:
370 return -ENOMEM;
371}
372
373static void __net_exit ip_rt_do_proc_exit(struct net *net)
374{
375 remove_proc_entry("rt_cache", net->proc_net_stat);
376 remove_proc_entry("rt_cache", net->proc_net);
377#ifdef CONFIG_IP_ROUTE_CLASSID
378 remove_proc_entry("rt_acct", net->proc_net);
379#endif
380}
381
382static struct pernet_operations ip_rt_proc_ops __net_initdata = {
383 .init = ip_rt_do_proc_init,
384 .exit = ip_rt_do_proc_exit,
385};
386
387static int __init ip_rt_proc_init(void)
388{
389 return register_pernet_subsys(&ip_rt_proc_ops);
390}
391
392#else
393static inline int ip_rt_proc_init(void)
394{
395 return 0;
396}
397#endif /* CONFIG_PROC_FS */
398
399static inline bool rt_is_expired(const struct rtable *rth)
400{
401 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
402}
403
404void rt_cache_flush(struct net *net)
405{
406 rt_genid_bump_ipv4(net);
407}
408
409static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
410 struct sk_buff *skb,
411 const void *daddr)
412{
413 const struct rtable *rt = container_of(dst, struct rtable, dst);
414 struct net_device *dev = dst->dev;
415 struct neighbour *n;
416
417 rcu_read_lock_bh();
418
419 if (likely(rt->rt_gw_family == AF_INET)) {
420 n = ip_neigh_gw4(dev, rt->rt_gw4);
421 } else if (rt->rt_gw_family == AF_INET6) {
422 n = ip_neigh_gw6(dev, &rt->rt_gw6);
423 } else {
424 __be32 pkey;
425
426 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
427 n = ip_neigh_gw4(dev, pkey);
428 }
429
430 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
431 n = NULL;
432
433 rcu_read_unlock_bh();
434
435 return n;
436}
437
438static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
439{
440 const struct rtable *rt = container_of(dst, struct rtable, dst);
441 struct net_device *dev = dst->dev;
442 const __be32 *pkey = daddr;
443
444 if (rt->rt_gw_family == AF_INET) {
445 pkey = (const __be32 *)&rt->rt_gw4;
446 } else if (rt->rt_gw_family == AF_INET6) {
447 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
448 } else if (!daddr ||
449 (rt->rt_flags &
450 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
451 return;
452 }
453 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
454}
455
456/* Hash tables of size 2048..262144 depending on RAM size.
457 * Each bucket uses 8 bytes.
458 */
459static u32 ip_idents_mask __read_mostly;
460static atomic_t *ip_idents __read_mostly;
461static u32 *ip_tstamps __read_mostly;
462
463/* In order to protect privacy, we add a perturbation to identifiers
464 * if one generator is seldom used. This makes hard for an attacker
465 * to infer how many packets were sent between two points in time.
466 */
467u32 ip_idents_reserve(u32 hash, int segs)
468{
469 u32 bucket, old, now = (u32)jiffies;
470 atomic_t *p_id;
471 u32 *p_tstamp;
472 u32 delta = 0;
473
474 bucket = hash & ip_idents_mask;
475 p_tstamp = ip_tstamps + bucket;
476 p_id = ip_idents + bucket;
477 old = READ_ONCE(*p_tstamp);
478
479 if (old != now && cmpxchg(p_tstamp, old, now) == old)
480 delta = prandom_u32_max(now - old);
481
482 /* If UBSAN reports an error there, please make sure your compiler
483 * supports -fno-strict-overflow before reporting it that was a bug
484 * in UBSAN, and it has been fixed in GCC-8.
485 */
486 return atomic_add_return(segs + delta, p_id) - segs;
487}
488EXPORT_SYMBOL(ip_idents_reserve);
489
490void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
491{
492 u32 hash, id;
493
494 /* Note the following code is not safe, but this is okay. */
495 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
496 get_random_bytes(&net->ipv4.ip_id_key,
497 sizeof(net->ipv4.ip_id_key));
498
499 hash = siphash_3u32((__force u32)iph->daddr,
500 (__force u32)iph->saddr,
501 iph->protocol,
502 &net->ipv4.ip_id_key);
503 id = ip_idents_reserve(hash, segs);
504 iph->id = htons(id);
505}
506EXPORT_SYMBOL(__ip_select_ident);
507
508static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
509 const struct sock *sk,
510 const struct iphdr *iph,
511 int oif, u8 tos,
512 u8 prot, u32 mark, int flow_flags)
513{
514 if (sk) {
515 const struct inet_sock *inet = inet_sk(sk);
516
517 oif = sk->sk_bound_dev_if;
518 mark = sk->sk_mark;
519 tos = RT_CONN_FLAGS(sk);
520 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
521 }
522 flowi4_init_output(fl4, oif, mark, tos,
523 RT_SCOPE_UNIVERSE, prot,
524 flow_flags,
525 iph->daddr, iph->saddr, 0, 0,
526 sock_net_uid(net, sk));
527}
528
529static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
530 const struct sock *sk)
531{
532 const struct net *net = dev_net(skb->dev);
533 const struct iphdr *iph = ip_hdr(skb);
534 int oif = skb->dev->ifindex;
535 u8 tos = RT_TOS(iph->tos);
536 u8 prot = iph->protocol;
537 u32 mark = skb->mark;
538
539 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
540}
541
542static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
543{
544 const struct inet_sock *inet = inet_sk(sk);
545 const struct ip_options_rcu *inet_opt;
546 __be32 daddr = inet->inet_daddr;
547
548 rcu_read_lock();
549 inet_opt = rcu_dereference(inet->inet_opt);
550 if (inet_opt && inet_opt->opt.srr)
551 daddr = inet_opt->opt.faddr;
552 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
553 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
554 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
555 inet_sk_flowi_flags(sk),
556 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
557 rcu_read_unlock();
558}
559
560static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
561 const struct sk_buff *skb)
562{
563 if (skb)
564 build_skb_flow_key(fl4, skb, sk);
565 else
566 build_sk_flow_key(fl4, sk);
567}
568
569static DEFINE_SPINLOCK(fnhe_lock);
570
571static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
572{
573 struct rtable *rt;
574
575 rt = rcu_dereference(fnhe->fnhe_rth_input);
576 if (rt) {
577 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
578 dst_dev_put(&rt->dst);
579 dst_release(&rt->dst);
580 }
581 rt = rcu_dereference(fnhe->fnhe_rth_output);
582 if (rt) {
583 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
584 dst_dev_put(&rt->dst);
585 dst_release(&rt->dst);
586 }
587}
588
589static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
590{
591 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
592 struct fib_nh_exception *fnhe, *oldest = NULL;
593
594 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
595 fnhe = rcu_dereference_protected(*fnhe_p,
596 lockdep_is_held(&fnhe_lock));
597 if (!fnhe)
598 break;
599 if (!oldest ||
600 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
601 oldest = fnhe;
602 oldest_p = fnhe_p;
603 }
604 }
605 fnhe_flush_routes(oldest);
606 *oldest_p = oldest->fnhe_next;
607 kfree_rcu(oldest, rcu);
608}
609
610static u32 fnhe_hashfun(__be32 daddr)
611{
612 static siphash_key_t fnhe_hash_key __read_mostly;
613 u64 hval;
614
615 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
616 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
617 return hash_64(hval, FNHE_HASH_SHIFT);
618}
619
620static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
621{
622 rt->rt_pmtu = fnhe->fnhe_pmtu;
623 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
624 rt->dst.expires = fnhe->fnhe_expires;
625
626 if (fnhe->fnhe_gw) {
627 rt->rt_flags |= RTCF_REDIRECTED;
628 rt->rt_uses_gateway = 1;
629 rt->rt_gw_family = AF_INET;
630 rt->rt_gw4 = fnhe->fnhe_gw;
631 }
632}
633
634static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
635 __be32 gw, u32 pmtu, bool lock,
636 unsigned long expires)
637{
638 struct fnhe_hash_bucket *hash;
639 struct fib_nh_exception *fnhe;
640 struct rtable *rt;
641 u32 genid, hval;
642 unsigned int i;
643 int depth;
644
645 genid = fnhe_genid(dev_net(nhc->nhc_dev));
646 hval = fnhe_hashfun(daddr);
647
648 spin_lock_bh(&fnhe_lock);
649
650 hash = rcu_dereference(nhc->nhc_exceptions);
651 if (!hash) {
652 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
653 if (!hash)
654 goto out_unlock;
655 rcu_assign_pointer(nhc->nhc_exceptions, hash);
656 }
657
658 hash += hval;
659
660 depth = 0;
661 for (fnhe = rcu_dereference(hash->chain); fnhe;
662 fnhe = rcu_dereference(fnhe->fnhe_next)) {
663 if (fnhe->fnhe_daddr == daddr)
664 break;
665 depth++;
666 }
667
668 if (fnhe) {
669 if (fnhe->fnhe_genid != genid)
670 fnhe->fnhe_genid = genid;
671 if (gw)
672 fnhe->fnhe_gw = gw;
673 if (pmtu) {
674 fnhe->fnhe_pmtu = pmtu;
675 fnhe->fnhe_mtu_locked = lock;
676 }
677 fnhe->fnhe_expires = max(1UL, expires);
678 /* Update all cached dsts too */
679 rt = rcu_dereference(fnhe->fnhe_rth_input);
680 if (rt)
681 fill_route_from_fnhe(rt, fnhe);
682 rt = rcu_dereference(fnhe->fnhe_rth_output);
683 if (rt)
684 fill_route_from_fnhe(rt, fnhe);
685 } else {
686 /* Randomize max depth to avoid some side channels attacks. */
687 int max_depth = FNHE_RECLAIM_DEPTH +
688 prandom_u32_max(FNHE_RECLAIM_DEPTH);
689
690 while (depth > max_depth) {
691 fnhe_remove_oldest(hash);
692 depth--;
693 }
694
695 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
696 if (!fnhe)
697 goto out_unlock;
698
699 fnhe->fnhe_next = hash->chain;
700
701 fnhe->fnhe_genid = genid;
702 fnhe->fnhe_daddr = daddr;
703 fnhe->fnhe_gw = gw;
704 fnhe->fnhe_pmtu = pmtu;
705 fnhe->fnhe_mtu_locked = lock;
706 fnhe->fnhe_expires = max(1UL, expires);
707
708 rcu_assign_pointer(hash->chain, fnhe);
709
710 /* Exception created; mark the cached routes for the nexthop
711 * stale, so anyone caching it rechecks if this exception
712 * applies to them.
713 */
714 rt = rcu_dereference(nhc->nhc_rth_input);
715 if (rt)
716 rt->dst.obsolete = DST_OBSOLETE_KILL;
717
718 for_each_possible_cpu(i) {
719 struct rtable __rcu **prt;
720
721 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
722 rt = rcu_dereference(*prt);
723 if (rt)
724 rt->dst.obsolete = DST_OBSOLETE_KILL;
725 }
726 }
727
728 fnhe->fnhe_stamp = jiffies;
729
730out_unlock:
731 spin_unlock_bh(&fnhe_lock);
732}
733
734static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
735 bool kill_route)
736{
737 __be32 new_gw = icmp_hdr(skb)->un.gateway;
738 __be32 old_gw = ip_hdr(skb)->saddr;
739 struct net_device *dev = skb->dev;
740 struct in_device *in_dev;
741 struct fib_result res;
742 struct neighbour *n;
743 struct net *net;
744
745 switch (icmp_hdr(skb)->code & 7) {
746 case ICMP_REDIR_NET:
747 case ICMP_REDIR_NETTOS:
748 case ICMP_REDIR_HOST:
749 case ICMP_REDIR_HOSTTOS:
750 break;
751
752 default:
753 return;
754 }
755
756 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
757 return;
758
759 in_dev = __in_dev_get_rcu(dev);
760 if (!in_dev)
761 return;
762
763 net = dev_net(dev);
764 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
765 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
766 ipv4_is_zeronet(new_gw))
767 goto reject_redirect;
768
769 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
770 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
771 goto reject_redirect;
772 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
773 goto reject_redirect;
774 } else {
775 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
776 goto reject_redirect;
777 }
778
779 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
780 if (!n)
781 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
782 if (!IS_ERR(n)) {
783 if (!(n->nud_state & NUD_VALID)) {
784 neigh_event_send(n, NULL);
785 } else {
786 if (fib_lookup(net, fl4, &res, 0) == 0) {
787 struct fib_nh_common *nhc;
788
789 fib_select_path(net, &res, fl4, skb);
790 nhc = FIB_RES_NHC(res);
791 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
792 0, false,
793 jiffies + ip_rt_gc_timeout);
794 }
795 if (kill_route)
796 rt->dst.obsolete = DST_OBSOLETE_KILL;
797 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
798 }
799 neigh_release(n);
800 }
801 return;
802
803reject_redirect:
804#ifdef CONFIG_IP_ROUTE_VERBOSE
805 if (IN_DEV_LOG_MARTIANS(in_dev)) {
806 const struct iphdr *iph = (const struct iphdr *) skb->data;
807 __be32 daddr = iph->daddr;
808 __be32 saddr = iph->saddr;
809
810 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
811 " Advised path = %pI4 -> %pI4\n",
812 &old_gw, dev->name, &new_gw,
813 &saddr, &daddr);
814 }
815#endif
816 ;
817}
818
819static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
820{
821 struct rtable *rt;
822 struct flowi4 fl4;
823 const struct iphdr *iph = (const struct iphdr *) skb->data;
824 struct net *net = dev_net(skb->dev);
825 int oif = skb->dev->ifindex;
826 u8 tos = RT_TOS(iph->tos);
827 u8 prot = iph->protocol;
828 u32 mark = skb->mark;
829
830 rt = (struct rtable *) dst;
831
832 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
833 __ip_do_redirect(rt, skb, &fl4, true);
834}
835
836static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
837{
838 struct rtable *rt = (struct rtable *)dst;
839 struct dst_entry *ret = dst;
840
841 if (rt) {
842 if (dst->obsolete > 0) {
843 ip_rt_put(rt);
844 ret = NULL;
845 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
846 rt->dst.expires) {
847 ip_rt_put(rt);
848 ret = NULL;
849 }
850 }
851 return ret;
852}
853
854/*
855 * Algorithm:
856 * 1. The first ip_rt_redirect_number redirects are sent
857 * with exponential backoff, then we stop sending them at all,
858 * assuming that the host ignores our redirects.
859 * 2. If we did not see packets requiring redirects
860 * during ip_rt_redirect_silence, we assume that the host
861 * forgot redirected route and start to send redirects again.
862 *
863 * This algorithm is much cheaper and more intelligent than dumb load limiting
864 * in icmp.c.
865 *
866 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
867 * and "frag. need" (breaks PMTU discovery) in icmp.c.
868 */
869
870void ip_rt_send_redirect(struct sk_buff *skb)
871{
872 struct rtable *rt = skb_rtable(skb);
873 struct in_device *in_dev;
874 struct inet_peer *peer;
875 struct net *net;
876 int log_martians;
877 int vif;
878
879 rcu_read_lock();
880 in_dev = __in_dev_get_rcu(rt->dst.dev);
881 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
882 rcu_read_unlock();
883 return;
884 }
885 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
886 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
887 rcu_read_unlock();
888
889 net = dev_net(rt->dst.dev);
890 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
891 if (!peer) {
892 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
893 rt_nexthop(rt, ip_hdr(skb)->daddr));
894 return;
895 }
896
897 /* No redirected packets during ip_rt_redirect_silence;
898 * reset the algorithm.
899 */
900 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
901 peer->rate_tokens = 0;
902 peer->n_redirects = 0;
903 }
904
905 /* Too many ignored redirects; do not send anything
906 * set dst.rate_last to the last seen redirected packet.
907 */
908 if (peer->n_redirects >= ip_rt_redirect_number) {
909 peer->rate_last = jiffies;
910 goto out_put_peer;
911 }
912
913 /* Check for load limit; set rate_last to the latest sent
914 * redirect.
915 */
916 if (peer->n_redirects == 0 ||
917 time_after(jiffies,
918 (peer->rate_last +
919 (ip_rt_redirect_load << peer->n_redirects)))) {
920 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
921
922 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
923 peer->rate_last = jiffies;
924 ++peer->n_redirects;
925#ifdef CONFIG_IP_ROUTE_VERBOSE
926 if (log_martians &&
927 peer->n_redirects == ip_rt_redirect_number)
928 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
929 &ip_hdr(skb)->saddr, inet_iif(skb),
930 &ip_hdr(skb)->daddr, &gw);
931#endif
932 }
933out_put_peer:
934 inet_putpeer(peer);
935}
936
937static int ip_error(struct sk_buff *skb)
938{
939 struct rtable *rt = skb_rtable(skb);
940 struct net_device *dev = skb->dev;
941 struct in_device *in_dev;
942 struct inet_peer *peer;
943 unsigned long now;
944 struct net *net;
945 bool send;
946 int code;
947
948 if (netif_is_l3_master(skb->dev)) {
949 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
950 if (!dev)
951 goto out;
952 }
953
954 in_dev = __in_dev_get_rcu(dev);
955
956 /* IP on this device is disabled. */
957 if (!in_dev)
958 goto out;
959
960 net = dev_net(rt->dst.dev);
961 if (!IN_DEV_FORWARD(in_dev)) {
962 switch (rt->dst.error) {
963 case EHOSTUNREACH:
964 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
965 break;
966
967 case ENETUNREACH:
968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
969 break;
970 }
971 goto out;
972 }
973
974 switch (rt->dst.error) {
975 case EINVAL:
976 default:
977 goto out;
978 case EHOSTUNREACH:
979 code = ICMP_HOST_UNREACH;
980 break;
981 case ENETUNREACH:
982 code = ICMP_NET_UNREACH;
983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
984 break;
985 case EACCES:
986 code = ICMP_PKT_FILTERED;
987 break;
988 }
989
990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
991 l3mdev_master_ifindex(skb->dev), 1);
992
993 send = true;
994 if (peer) {
995 now = jiffies;
996 peer->rate_tokens += now - peer->rate_last;
997 if (peer->rate_tokens > ip_rt_error_burst)
998 peer->rate_tokens = ip_rt_error_burst;
999 peer->rate_last = now;
1000 if (peer->rate_tokens >= ip_rt_error_cost)
1001 peer->rate_tokens -= ip_rt_error_cost;
1002 else
1003 send = false;
1004 inet_putpeer(peer);
1005 }
1006 if (send)
1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1008
1009out: kfree_skb(skb);
1010 return 0;
1011}
1012
1013static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1014{
1015 struct dst_entry *dst = &rt->dst;
1016 struct net *net = dev_net(dst->dev);
1017 struct fib_result res;
1018 bool lock = false;
1019 u32 old_mtu;
1020
1021 if (ip_mtu_locked(dst))
1022 return;
1023
1024 old_mtu = ipv4_mtu(dst);
1025 if (old_mtu < mtu)
1026 return;
1027
1028 if (mtu < ip_rt_min_pmtu) {
1029 lock = true;
1030 mtu = min(old_mtu, ip_rt_min_pmtu);
1031 }
1032
1033 if (rt->rt_pmtu == mtu && !lock &&
1034 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1035 return;
1036
1037 rcu_read_lock();
1038 if (fib_lookup(net, fl4, &res, 0) == 0) {
1039 struct fib_nh_common *nhc;
1040
1041 fib_select_path(net, &res, fl4, NULL);
1042 nhc = FIB_RES_NHC(res);
1043 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1044 jiffies + ip_rt_mtu_expires);
1045 }
1046 rcu_read_unlock();
1047}
1048
1049static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1050 struct sk_buff *skb, u32 mtu,
1051 bool confirm_neigh)
1052{
1053 struct rtable *rt = (struct rtable *) dst;
1054 struct flowi4 fl4;
1055
1056 ip_rt_build_flow_key(&fl4, sk, skb);
1057
1058 /* Don't make lookup fail for bridged encapsulations */
1059 if (skb && netif_is_any_bridge_port(skb->dev))
1060 fl4.flowi4_oif = 0;
1061
1062 __ip_rt_update_pmtu(rt, &fl4, mtu);
1063}
1064
1065void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1066 int oif, u8 protocol)
1067{
1068 const struct iphdr *iph = (const struct iphdr *)skb->data;
1069 struct flowi4 fl4;
1070 struct rtable *rt;
1071 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1072
1073 __build_flow_key(net, &fl4, NULL, iph, oif,
1074 RT_TOS(iph->tos), protocol, mark, 0);
1075 rt = __ip_route_output_key(net, &fl4);
1076 if (!IS_ERR(rt)) {
1077 __ip_rt_update_pmtu(rt, &fl4, mtu);
1078 ip_rt_put(rt);
1079 }
1080}
1081EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1082
1083static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1084{
1085 const struct iphdr *iph = (const struct iphdr *)skb->data;
1086 struct flowi4 fl4;
1087 struct rtable *rt;
1088
1089 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1090
1091 if (!fl4.flowi4_mark)
1092 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1093
1094 rt = __ip_route_output_key(sock_net(sk), &fl4);
1095 if (!IS_ERR(rt)) {
1096 __ip_rt_update_pmtu(rt, &fl4, mtu);
1097 ip_rt_put(rt);
1098 }
1099}
1100
1101void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1102{
1103 const struct iphdr *iph = (const struct iphdr *)skb->data;
1104 struct flowi4 fl4;
1105 struct rtable *rt;
1106 struct dst_entry *odst = NULL;
1107 bool new = false;
1108 struct net *net = sock_net(sk);
1109
1110 bh_lock_sock(sk);
1111
1112 if (!ip_sk_accept_pmtu(sk))
1113 goto out;
1114
1115 odst = sk_dst_get(sk);
1116
1117 if (sock_owned_by_user(sk) || !odst) {
1118 __ipv4_sk_update_pmtu(skb, sk, mtu);
1119 goto out;
1120 }
1121
1122 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1123
1124 rt = (struct rtable *)odst;
1125 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1126 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1127 if (IS_ERR(rt))
1128 goto out;
1129
1130 new = true;
1131 }
1132
1133 __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1134
1135 if (!dst_check(&rt->dst, 0)) {
1136 if (new)
1137 dst_release(&rt->dst);
1138
1139 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1140 if (IS_ERR(rt))
1141 goto out;
1142
1143 new = true;
1144 }
1145
1146 if (new)
1147 sk_dst_set(sk, &rt->dst);
1148
1149out:
1150 bh_unlock_sock(sk);
1151 dst_release(odst);
1152}
1153EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1154
1155void ipv4_redirect(struct sk_buff *skb, struct net *net,
1156 int oif, u8 protocol)
1157{
1158 const struct iphdr *iph = (const struct iphdr *)skb->data;
1159 struct flowi4 fl4;
1160 struct rtable *rt;
1161
1162 __build_flow_key(net, &fl4, NULL, iph, oif,
1163 RT_TOS(iph->tos), protocol, 0, 0);
1164 rt = __ip_route_output_key(net, &fl4);
1165 if (!IS_ERR(rt)) {
1166 __ip_do_redirect(rt, skb, &fl4, false);
1167 ip_rt_put(rt);
1168 }
1169}
1170EXPORT_SYMBOL_GPL(ipv4_redirect);
1171
1172void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1173{
1174 const struct iphdr *iph = (const struct iphdr *)skb->data;
1175 struct flowi4 fl4;
1176 struct rtable *rt;
1177 struct net *net = sock_net(sk);
1178
1179 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1180 rt = __ip_route_output_key(net, &fl4);
1181 if (!IS_ERR(rt)) {
1182 __ip_do_redirect(rt, skb, &fl4, false);
1183 ip_rt_put(rt);
1184 }
1185}
1186EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1187
1188INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1189 u32 cookie)
1190{
1191 struct rtable *rt = (struct rtable *) dst;
1192
1193 /* All IPV4 dsts are created with ->obsolete set to the value
1194 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1195 * into this function always.
1196 *
1197 * When a PMTU/redirect information update invalidates a route,
1198 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1199 * DST_OBSOLETE_DEAD.
1200 */
1201 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1202 return NULL;
1203 return dst;
1204}
1205EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1206
1207static void ipv4_send_dest_unreach(struct sk_buff *skb)
1208{
1209 struct ip_options opt;
1210 int res;
1211
1212 /* Recompile ip options since IPCB may not be valid anymore.
1213 * Also check we have a reasonable ipv4 header.
1214 */
1215 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1216 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1217 return;
1218
1219 memset(&opt, 0, sizeof(opt));
1220 if (ip_hdr(skb)->ihl > 5) {
1221 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1222 return;
1223 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1224
1225 rcu_read_lock();
1226 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1227 rcu_read_unlock();
1228
1229 if (res)
1230 return;
1231 }
1232 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1233}
1234
1235static void ipv4_link_failure(struct sk_buff *skb)
1236{
1237 struct rtable *rt;
1238
1239 ipv4_send_dest_unreach(skb);
1240
1241 rt = skb_rtable(skb);
1242 if (rt)
1243 dst_set_expires(&rt->dst, 0);
1244}
1245
1246static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1247{
1248 pr_debug("%s: %pI4 -> %pI4, %s\n",
1249 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1250 skb->dev ? skb->dev->name : "?");
1251 kfree_skb(skb);
1252 WARN_ON(1);
1253 return 0;
1254}
1255
1256/*
1257 * We do not cache source address of outgoing interface,
1258 * because it is used only by IP RR, TS and SRR options,
1259 * so that it out of fast path.
1260 *
1261 * BTW remember: "addr" is allowed to be not aligned
1262 * in IP options!
1263 */
1264
1265void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1266{
1267 __be32 src;
1268
1269 if (rt_is_output_route(rt))
1270 src = ip_hdr(skb)->saddr;
1271 else {
1272 struct fib_result res;
1273 struct iphdr *iph = ip_hdr(skb);
1274 struct flowi4 fl4 = {
1275 .daddr = iph->daddr,
1276 .saddr = iph->saddr,
1277 .flowi4_tos = RT_TOS(iph->tos),
1278 .flowi4_oif = rt->dst.dev->ifindex,
1279 .flowi4_iif = skb->dev->ifindex,
1280 .flowi4_mark = skb->mark,
1281 };
1282
1283 rcu_read_lock();
1284 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1285 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1286 else
1287 src = inet_select_addr(rt->dst.dev,
1288 rt_nexthop(rt, iph->daddr),
1289 RT_SCOPE_UNIVERSE);
1290 rcu_read_unlock();
1291 }
1292 memcpy(addr, &src, 4);
1293}
1294
1295#ifdef CONFIG_IP_ROUTE_CLASSID
1296static void set_class_tag(struct rtable *rt, u32 tag)
1297{
1298 if (!(rt->dst.tclassid & 0xFFFF))
1299 rt->dst.tclassid |= tag & 0xFFFF;
1300 if (!(rt->dst.tclassid & 0xFFFF0000))
1301 rt->dst.tclassid |= tag & 0xFFFF0000;
1302}
1303#endif
1304
1305static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1306{
1307 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1308 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1309 ip_rt_min_advmss);
1310
1311 return min(advmss, IPV4_MAX_PMTU - header_size);
1312}
1313
1314INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1315{
1316 const struct rtable *rt = (const struct rtable *)dst;
1317 unsigned int mtu = rt->rt_pmtu;
1318
1319 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1320 mtu = dst_metric_raw(dst, RTAX_MTU);
1321
1322 if (mtu)
1323 goto out;
1324
1325 mtu = READ_ONCE(dst->dev->mtu);
1326
1327 if (unlikely(ip_mtu_locked(dst))) {
1328 if (rt->rt_uses_gateway && mtu > 576)
1329 mtu = 576;
1330 }
1331
1332out:
1333 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1334
1335 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1336}
1337EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1338
1339static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1340{
1341 struct fnhe_hash_bucket *hash;
1342 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1343 u32 hval = fnhe_hashfun(daddr);
1344
1345 spin_lock_bh(&fnhe_lock);
1346
1347 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1348 lockdep_is_held(&fnhe_lock));
1349 hash += hval;
1350
1351 fnhe_p = &hash->chain;
1352 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1353 while (fnhe) {
1354 if (fnhe->fnhe_daddr == daddr) {
1355 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1356 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1357 /* set fnhe_daddr to 0 to ensure it won't bind with
1358 * new dsts in rt_bind_exception().
1359 */
1360 fnhe->fnhe_daddr = 0;
1361 fnhe_flush_routes(fnhe);
1362 kfree_rcu(fnhe, rcu);
1363 break;
1364 }
1365 fnhe_p = &fnhe->fnhe_next;
1366 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1367 lockdep_is_held(&fnhe_lock));
1368 }
1369
1370 spin_unlock_bh(&fnhe_lock);
1371}
1372
1373static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1374 __be32 daddr)
1375{
1376 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1377 struct fib_nh_exception *fnhe;
1378 u32 hval;
1379
1380 if (!hash)
1381 return NULL;
1382
1383 hval = fnhe_hashfun(daddr);
1384
1385 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1386 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1387 if (fnhe->fnhe_daddr == daddr) {
1388 if (fnhe->fnhe_expires &&
1389 time_after(jiffies, fnhe->fnhe_expires)) {
1390 ip_del_fnhe(nhc, daddr);
1391 break;
1392 }
1393 return fnhe;
1394 }
1395 }
1396 return NULL;
1397}
1398
1399/* MTU selection:
1400 * 1. mtu on route is locked - use it
1401 * 2. mtu from nexthop exception
1402 * 3. mtu from egress device
1403 */
1404
1405u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1406{
1407 struct fib_nh_common *nhc = res->nhc;
1408 struct net_device *dev = nhc->nhc_dev;
1409 struct fib_info *fi = res->fi;
1410 u32 mtu = 0;
1411
1412 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1413 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1414 mtu = fi->fib_mtu;
1415
1416 if (likely(!mtu)) {
1417 struct fib_nh_exception *fnhe;
1418
1419 fnhe = find_exception(nhc, daddr);
1420 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1421 mtu = fnhe->fnhe_pmtu;
1422 }
1423
1424 if (likely(!mtu))
1425 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1426
1427 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1428}
1429
1430static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1431 __be32 daddr, const bool do_cache)
1432{
1433 bool ret = false;
1434
1435 spin_lock_bh(&fnhe_lock);
1436
1437 if (daddr == fnhe->fnhe_daddr) {
1438 struct rtable __rcu **porig;
1439 struct rtable *orig;
1440 int genid = fnhe_genid(dev_net(rt->dst.dev));
1441
1442 if (rt_is_input_route(rt))
1443 porig = &fnhe->fnhe_rth_input;
1444 else
1445 porig = &fnhe->fnhe_rth_output;
1446 orig = rcu_dereference(*porig);
1447
1448 if (fnhe->fnhe_genid != genid) {
1449 fnhe->fnhe_genid = genid;
1450 fnhe->fnhe_gw = 0;
1451 fnhe->fnhe_pmtu = 0;
1452 fnhe->fnhe_expires = 0;
1453 fnhe->fnhe_mtu_locked = false;
1454 fnhe_flush_routes(fnhe);
1455 orig = NULL;
1456 }
1457 fill_route_from_fnhe(rt, fnhe);
1458 if (!rt->rt_gw4) {
1459 rt->rt_gw4 = daddr;
1460 rt->rt_gw_family = AF_INET;
1461 }
1462
1463 if (do_cache) {
1464 dst_hold(&rt->dst);
1465 rcu_assign_pointer(*porig, rt);
1466 if (orig) {
1467 dst_dev_put(&orig->dst);
1468 dst_release(&orig->dst);
1469 }
1470 ret = true;
1471 }
1472
1473 fnhe->fnhe_stamp = jiffies;
1474 }
1475 spin_unlock_bh(&fnhe_lock);
1476
1477 return ret;
1478}
1479
1480static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1481{
1482 struct rtable *orig, *prev, **p;
1483 bool ret = true;
1484
1485 if (rt_is_input_route(rt)) {
1486 p = (struct rtable **)&nhc->nhc_rth_input;
1487 } else {
1488 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1489 }
1490 orig = *p;
1491
1492 /* hold dst before doing cmpxchg() to avoid race condition
1493 * on this dst
1494 */
1495 dst_hold(&rt->dst);
1496 prev = cmpxchg(p, orig, rt);
1497 if (prev == orig) {
1498 if (orig) {
1499 rt_add_uncached_list(orig);
1500 dst_release(&orig->dst);
1501 }
1502 } else {
1503 dst_release(&rt->dst);
1504 ret = false;
1505 }
1506
1507 return ret;
1508}
1509
1510struct uncached_list {
1511 spinlock_t lock;
1512 struct list_head head;
1513};
1514
1515static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1516
1517void rt_add_uncached_list(struct rtable *rt)
1518{
1519 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1520
1521 rt->rt_uncached_list = ul;
1522
1523 spin_lock_bh(&ul->lock);
1524 list_add_tail(&rt->rt_uncached, &ul->head);
1525 spin_unlock_bh(&ul->lock);
1526}
1527
1528void rt_del_uncached_list(struct rtable *rt)
1529{
1530 if (!list_empty(&rt->rt_uncached)) {
1531 struct uncached_list *ul = rt->rt_uncached_list;
1532
1533 spin_lock_bh(&ul->lock);
1534 list_del(&rt->rt_uncached);
1535 spin_unlock_bh(&ul->lock);
1536 }
1537}
1538
1539static void ipv4_dst_destroy(struct dst_entry *dst)
1540{
1541 struct rtable *rt = (struct rtable *)dst;
1542
1543 ip_dst_metrics_put(dst);
1544 rt_del_uncached_list(rt);
1545}
1546
1547void rt_flush_dev(struct net_device *dev)
1548{
1549 struct rtable *rt;
1550 int cpu;
1551
1552 for_each_possible_cpu(cpu) {
1553 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1554
1555 spin_lock_bh(&ul->lock);
1556 list_for_each_entry(rt, &ul->head, rt_uncached) {
1557 if (rt->dst.dev != dev)
1558 continue;
1559 rt->dst.dev = blackhole_netdev;
1560 dev_hold(rt->dst.dev);
1561 dev_put(dev);
1562 }
1563 spin_unlock_bh(&ul->lock);
1564 }
1565}
1566
1567static bool rt_cache_valid(const struct rtable *rt)
1568{
1569 return rt &&
1570 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1571 !rt_is_expired(rt);
1572}
1573
1574static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1575 const struct fib_result *res,
1576 struct fib_nh_exception *fnhe,
1577 struct fib_info *fi, u16 type, u32 itag,
1578 const bool do_cache)
1579{
1580 bool cached = false;
1581
1582 if (fi) {
1583 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1584
1585 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1586 rt->rt_uses_gateway = 1;
1587 rt->rt_gw_family = nhc->nhc_gw_family;
1588 /* only INET and INET6 are supported */
1589 if (likely(nhc->nhc_gw_family == AF_INET))
1590 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1591 else
1592 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1593 }
1594
1595 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1596
1597#ifdef CONFIG_IP_ROUTE_CLASSID
1598 if (nhc->nhc_family == AF_INET) {
1599 struct fib_nh *nh;
1600
1601 nh = container_of(nhc, struct fib_nh, nh_common);
1602 rt->dst.tclassid = nh->nh_tclassid;
1603 }
1604#endif
1605 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1606 if (unlikely(fnhe))
1607 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1608 else if (do_cache)
1609 cached = rt_cache_route(nhc, rt);
1610 if (unlikely(!cached)) {
1611 /* Routes we intend to cache in nexthop exception or
1612 * FIB nexthop have the DST_NOCACHE bit clear.
1613 * However, if we are unsuccessful at storing this
1614 * route into the cache we really need to set it.
1615 */
1616 if (!rt->rt_gw4) {
1617 rt->rt_gw_family = AF_INET;
1618 rt->rt_gw4 = daddr;
1619 }
1620 rt_add_uncached_list(rt);
1621 }
1622 } else
1623 rt_add_uncached_list(rt);
1624
1625#ifdef CONFIG_IP_ROUTE_CLASSID
1626#ifdef CONFIG_IP_MULTIPLE_TABLES
1627 set_class_tag(rt, res->tclassid);
1628#endif
1629 set_class_tag(rt, itag);
1630#endif
1631}
1632
1633struct rtable *rt_dst_alloc(struct net_device *dev,
1634 unsigned int flags, u16 type,
1635 bool nopolicy, bool noxfrm)
1636{
1637 struct rtable *rt;
1638
1639 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1640 (nopolicy ? DST_NOPOLICY : 0) |
1641 (noxfrm ? DST_NOXFRM : 0));
1642
1643 if (rt) {
1644 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1645 rt->rt_flags = flags;
1646 rt->rt_type = type;
1647 rt->rt_is_input = 0;
1648 rt->rt_iif = 0;
1649 rt->rt_pmtu = 0;
1650 rt->rt_mtu_locked = 0;
1651 rt->rt_uses_gateway = 0;
1652 rt->rt_gw_family = 0;
1653 rt->rt_gw4 = 0;
1654 INIT_LIST_HEAD(&rt->rt_uncached);
1655
1656 rt->dst.output = ip_output;
1657 if (flags & RTCF_LOCAL)
1658 rt->dst.input = ip_local_deliver;
1659 }
1660
1661 return rt;
1662}
1663EXPORT_SYMBOL(rt_dst_alloc);
1664
1665struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1666{
1667 struct rtable *new_rt;
1668
1669 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1670 rt->dst.flags);
1671
1672 if (new_rt) {
1673 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1674 new_rt->rt_flags = rt->rt_flags;
1675 new_rt->rt_type = rt->rt_type;
1676 new_rt->rt_is_input = rt->rt_is_input;
1677 new_rt->rt_iif = rt->rt_iif;
1678 new_rt->rt_pmtu = rt->rt_pmtu;
1679 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1680 new_rt->rt_gw_family = rt->rt_gw_family;
1681 if (rt->rt_gw_family == AF_INET)
1682 new_rt->rt_gw4 = rt->rt_gw4;
1683 else if (rt->rt_gw_family == AF_INET6)
1684 new_rt->rt_gw6 = rt->rt_gw6;
1685 INIT_LIST_HEAD(&new_rt->rt_uncached);
1686
1687 new_rt->dst.input = rt->dst.input;
1688 new_rt->dst.output = rt->dst.output;
1689 new_rt->dst.error = rt->dst.error;
1690 new_rt->dst.lastuse = jiffies;
1691 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1692 }
1693 return new_rt;
1694}
1695EXPORT_SYMBOL(rt_dst_clone);
1696
1697/* called in rcu_read_lock() section */
1698int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1699 u8 tos, struct net_device *dev,
1700 struct in_device *in_dev, u32 *itag)
1701{
1702 int err;
1703
1704 /* Primary sanity checks. */
1705 if (!in_dev)
1706 return -EINVAL;
1707
1708 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1709 skb->protocol != htons(ETH_P_IP))
1710 return -EINVAL;
1711
1712 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1713 return -EINVAL;
1714
1715 if (ipv4_is_zeronet(saddr)) {
1716 if (!ipv4_is_local_multicast(daddr) &&
1717 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1718 return -EINVAL;
1719 } else {
1720 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1721 in_dev, itag);
1722 if (err < 0)
1723 return err;
1724 }
1725 return 0;
1726}
1727
1728/* called in rcu_read_lock() section */
1729static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1730 u8 tos, struct net_device *dev, int our)
1731{
1732 struct in_device *in_dev = __in_dev_get_rcu(dev);
1733 unsigned int flags = RTCF_MULTICAST;
1734 struct rtable *rth;
1735 u32 itag = 0;
1736 int err;
1737
1738 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1739 if (err)
1740 return err;
1741
1742 if (our)
1743 flags |= RTCF_LOCAL;
1744
1745 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1746 IN_DEV_ORCONF(in_dev, NOPOLICY), false);
1747 if (!rth)
1748 return -ENOBUFS;
1749
1750#ifdef CONFIG_IP_ROUTE_CLASSID
1751 rth->dst.tclassid = itag;
1752#endif
1753 rth->dst.output = ip_rt_bug;
1754 rth->rt_is_input= 1;
1755
1756#ifdef CONFIG_IP_MROUTE
1757 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1758 rth->dst.input = ip_mr_input;
1759#endif
1760 RT_CACHE_STAT_INC(in_slow_mc);
1761
1762 skb_dst_set(skb, &rth->dst);
1763 return 0;
1764}
1765
1766
1767static void ip_handle_martian_source(struct net_device *dev,
1768 struct in_device *in_dev,
1769 struct sk_buff *skb,
1770 __be32 daddr,
1771 __be32 saddr)
1772{
1773 RT_CACHE_STAT_INC(in_martian_src);
1774#ifdef CONFIG_IP_ROUTE_VERBOSE
1775 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1776 /*
1777 * RFC1812 recommendation, if source is martian,
1778 * the only hint is MAC header.
1779 */
1780 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1781 &daddr, &saddr, dev->name);
1782 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1783 print_hex_dump(KERN_WARNING, "ll header: ",
1784 DUMP_PREFIX_OFFSET, 16, 1,
1785 skb_mac_header(skb),
1786 dev->hard_header_len, false);
1787 }
1788 }
1789#endif
1790}
1791
1792/* called in rcu_read_lock() section */
1793static int __mkroute_input(struct sk_buff *skb,
1794 const struct fib_result *res,
1795 struct in_device *in_dev,
1796 __be32 daddr, __be32 saddr, u32 tos)
1797{
1798 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1799 struct net_device *dev = nhc->nhc_dev;
1800 struct fib_nh_exception *fnhe;
1801 struct rtable *rth;
1802 int err;
1803 struct in_device *out_dev;
1804 bool do_cache;
1805 u32 itag = 0;
1806
1807 /* get a working reference to the output device */
1808 out_dev = __in_dev_get_rcu(dev);
1809 if (!out_dev) {
1810 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1811 return -EINVAL;
1812 }
1813
1814 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1815 in_dev->dev, in_dev, &itag);
1816 if (err < 0) {
1817 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1818 saddr);
1819
1820 goto cleanup;
1821 }
1822
1823 do_cache = res->fi && !itag;
1824 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1825 skb->protocol == htons(ETH_P_IP)) {
1826 __be32 gw;
1827
1828 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1829 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1830 inet_addr_onlink(out_dev, saddr, gw))
1831 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1832 }
1833
1834 if (skb->protocol != htons(ETH_P_IP)) {
1835 /* Not IP (i.e. ARP). Do not create route, if it is
1836 * invalid for proxy arp. DNAT routes are always valid.
1837 *
1838 * Proxy arp feature have been extended to allow, ARP
1839 * replies back to the same interface, to support
1840 * Private VLAN switch technologies. See arp.c.
1841 */
1842 if (out_dev == in_dev &&
1843 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1844 err = -EINVAL;
1845 goto cleanup;
1846 }
1847 }
1848
1849 fnhe = find_exception(nhc, daddr);
1850 if (do_cache) {
1851 if (fnhe)
1852 rth = rcu_dereference(fnhe->fnhe_rth_input);
1853 else
1854 rth = rcu_dereference(nhc->nhc_rth_input);
1855 if (rt_cache_valid(rth)) {
1856 skb_dst_set_noref(skb, &rth->dst);
1857 goto out;
1858 }
1859 }
1860
1861 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1862 IN_DEV_ORCONF(in_dev, NOPOLICY),
1863 IN_DEV_ORCONF(out_dev, NOXFRM));
1864 if (!rth) {
1865 err = -ENOBUFS;
1866 goto cleanup;
1867 }
1868
1869 rth->rt_is_input = 1;
1870 RT_CACHE_STAT_INC(in_slow_tot);
1871
1872 rth->dst.input = ip_forward;
1873
1874 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1875 do_cache);
1876 lwtunnel_set_redirect(&rth->dst);
1877 skb_dst_set(skb, &rth->dst);
1878out:
1879 err = 0;
1880 cleanup:
1881 return err;
1882}
1883
1884#ifdef CONFIG_IP_ROUTE_MULTIPATH
1885/* To make ICMP packets follow the right flow, the multipath hash is
1886 * calculated from the inner IP addresses.
1887 */
1888static void ip_multipath_l3_keys(const struct sk_buff *skb,
1889 struct flow_keys *hash_keys)
1890{
1891 const struct iphdr *outer_iph = ip_hdr(skb);
1892 const struct iphdr *key_iph = outer_iph;
1893 const struct iphdr *inner_iph;
1894 const struct icmphdr *icmph;
1895 struct iphdr _inner_iph;
1896 struct icmphdr _icmph;
1897
1898 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1899 goto out;
1900
1901 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1902 goto out;
1903
1904 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1905 &_icmph);
1906 if (!icmph)
1907 goto out;
1908
1909 if (!icmp_is_err(icmph->type))
1910 goto out;
1911
1912 inner_iph = skb_header_pointer(skb,
1913 outer_iph->ihl * 4 + sizeof(_icmph),
1914 sizeof(_inner_iph), &_inner_iph);
1915 if (!inner_iph)
1916 goto out;
1917
1918 key_iph = inner_iph;
1919out:
1920 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1921 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1922}
1923
1924static u32 fib_multipath_custom_hash_outer(const struct net *net,
1925 const struct sk_buff *skb,
1926 bool *p_has_inner)
1927{
1928 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1929 struct flow_keys keys, hash_keys;
1930
1931 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1932 return 0;
1933
1934 memset(&hash_keys, 0, sizeof(hash_keys));
1935 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1936
1937 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1938 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1939 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1940 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1941 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1942 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1943 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1944 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1945 hash_keys.ports.src = keys.ports.src;
1946 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1947 hash_keys.ports.dst = keys.ports.dst;
1948
1949 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1950 return flow_hash_from_keys(&hash_keys);
1951}
1952
1953static u32 fib_multipath_custom_hash_inner(const struct net *net,
1954 const struct sk_buff *skb,
1955 bool has_inner)
1956{
1957 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1958 struct flow_keys keys, hash_keys;
1959
1960 /* We assume the packet carries an encapsulation, but if none was
1961 * encountered during dissection of the outer flow, then there is no
1962 * point in calling the flow dissector again.
1963 */
1964 if (!has_inner)
1965 return 0;
1966
1967 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1968 return 0;
1969
1970 memset(&hash_keys, 0, sizeof(hash_keys));
1971 skb_flow_dissect_flow_keys(skb, &keys, 0);
1972
1973 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1974 return 0;
1975
1976 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1977 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1978 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1979 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1980 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1981 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1982 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1983 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1984 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1985 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1986 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1987 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1988 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1989 hash_keys.tags.flow_label = keys.tags.flow_label;
1990 }
1991
1992 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1993 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1994 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1995 hash_keys.ports.src = keys.ports.src;
1996 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1997 hash_keys.ports.dst = keys.ports.dst;
1998
1999 return flow_hash_from_keys(&hash_keys);
2000}
2001
2002static u32 fib_multipath_custom_hash_skb(const struct net *net,
2003 const struct sk_buff *skb)
2004{
2005 u32 mhash, mhash_inner;
2006 bool has_inner = true;
2007
2008 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2009 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2010
2011 return jhash_2words(mhash, mhash_inner, 0);
2012}
2013
2014static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2015 const struct flowi4 *fl4)
2016{
2017 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
2018 struct flow_keys hash_keys;
2019
2020 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2021 return 0;
2022
2023 memset(&hash_keys, 0, sizeof(hash_keys));
2024 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2025 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2026 hash_keys.addrs.v4addrs.src = fl4->saddr;
2027 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2028 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2029 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2030 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2031 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2032 hash_keys.ports.src = fl4->fl4_sport;
2033 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2034 hash_keys.ports.dst = fl4->fl4_dport;
2035
2036 return flow_hash_from_keys(&hash_keys);
2037}
2038
2039/* if skb is set it will be used and fl4 can be NULL */
2040int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2041 const struct sk_buff *skb, struct flow_keys *flkeys)
2042{
2043 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2044 struct flow_keys hash_keys;
2045 u32 mhash = 0;
2046
2047 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
2048 case 0:
2049 memset(&hash_keys, 0, sizeof(hash_keys));
2050 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2051 if (skb) {
2052 ip_multipath_l3_keys(skb, &hash_keys);
2053 } else {
2054 hash_keys.addrs.v4addrs.src = fl4->saddr;
2055 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2056 }
2057 mhash = flow_hash_from_keys(&hash_keys);
2058 break;
2059 case 1:
2060 /* skb is currently provided only when forwarding */
2061 if (skb) {
2062 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2063 struct flow_keys keys;
2064
2065 /* short-circuit if we already have L4 hash present */
2066 if (skb->l4_hash)
2067 return skb_get_hash_raw(skb) >> 1;
2068
2069 memset(&hash_keys, 0, sizeof(hash_keys));
2070
2071 if (!flkeys) {
2072 skb_flow_dissect_flow_keys(skb, &keys, flag);
2073 flkeys = &keys;
2074 }
2075
2076 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2077 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2078 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2079 hash_keys.ports.src = flkeys->ports.src;
2080 hash_keys.ports.dst = flkeys->ports.dst;
2081 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2082 } else {
2083 memset(&hash_keys, 0, sizeof(hash_keys));
2084 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2085 hash_keys.addrs.v4addrs.src = fl4->saddr;
2086 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2087 hash_keys.ports.src = fl4->fl4_sport;
2088 hash_keys.ports.dst = fl4->fl4_dport;
2089 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2090 }
2091 mhash = flow_hash_from_keys(&hash_keys);
2092 break;
2093 case 2:
2094 memset(&hash_keys, 0, sizeof(hash_keys));
2095 /* skb is currently provided only when forwarding */
2096 if (skb) {
2097 struct flow_keys keys;
2098
2099 skb_flow_dissect_flow_keys(skb, &keys, 0);
2100 /* Inner can be v4 or v6 */
2101 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2102 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2103 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2104 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2105 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2106 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2107 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2108 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2109 hash_keys.tags.flow_label = keys.tags.flow_label;
2110 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2111 } else {
2112 /* Same as case 0 */
2113 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2114 ip_multipath_l3_keys(skb, &hash_keys);
2115 }
2116 } else {
2117 /* Same as case 0 */
2118 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2119 hash_keys.addrs.v4addrs.src = fl4->saddr;
2120 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2121 }
2122 mhash = flow_hash_from_keys(&hash_keys);
2123 break;
2124 case 3:
2125 if (skb)
2126 mhash = fib_multipath_custom_hash_skb(net, skb);
2127 else
2128 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2129 break;
2130 }
2131
2132 if (multipath_hash)
2133 mhash = jhash_2words(mhash, multipath_hash, 0);
2134
2135 return mhash >> 1;
2136}
2137#endif /* CONFIG_IP_ROUTE_MULTIPATH */
2138
2139static int ip_mkroute_input(struct sk_buff *skb,
2140 struct fib_result *res,
2141 struct in_device *in_dev,
2142 __be32 daddr, __be32 saddr, u32 tos,
2143 struct flow_keys *hkeys)
2144{
2145#ifdef CONFIG_IP_ROUTE_MULTIPATH
2146 if (res->fi && fib_info_num_path(res->fi) > 1) {
2147 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2148
2149 fib_select_multipath(res, h);
2150 }
2151#endif
2152
2153 /* create a routing cache entry */
2154 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2155}
2156
2157/* Implements all the saddr-related checks as ip_route_input_slow(),
2158 * assuming daddr is valid and the destination is not a local broadcast one.
2159 * Uses the provided hint instead of performing a route lookup.
2160 */
2161int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2162 u8 tos, struct net_device *dev,
2163 const struct sk_buff *hint)
2164{
2165 struct in_device *in_dev = __in_dev_get_rcu(dev);
2166 struct rtable *rt = skb_rtable(hint);
2167 struct net *net = dev_net(dev);
2168 int err = -EINVAL;
2169 u32 tag = 0;
2170
2171 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2172 goto martian_source;
2173
2174 if (ipv4_is_zeronet(saddr))
2175 goto martian_source;
2176
2177 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2178 goto martian_source;
2179
2180 if (rt->rt_type != RTN_LOCAL)
2181 goto skip_validate_source;
2182
2183 tos &= IPTOS_RT_MASK;
2184 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2185 if (err < 0)
2186 goto martian_source;
2187
2188skip_validate_source:
2189 skb_dst_copy(skb, hint);
2190 return 0;
2191
2192martian_source:
2193 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2194 return err;
2195}
2196
2197/* get device for dst_alloc with local routes */
2198static struct net_device *ip_rt_get_dev(struct net *net,
2199 const struct fib_result *res)
2200{
2201 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2202 struct net_device *dev = NULL;
2203
2204 if (nhc)
2205 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2206
2207 return dev ? : net->loopback_dev;
2208}
2209
2210/*
2211 * NOTE. We drop all the packets that has local source
2212 * addresses, because every properly looped back packet
2213 * must have correct destination already attached by output routine.
2214 * Changes in the enforced policies must be applied also to
2215 * ip_route_use_hint().
2216 *
2217 * Such approach solves two big problems:
2218 * 1. Not simplex devices are handled properly.
2219 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2220 * called with rcu_read_lock()
2221 */
2222
2223static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2224 u8 tos, struct net_device *dev,
2225 struct fib_result *res)
2226{
2227 struct in_device *in_dev = __in_dev_get_rcu(dev);
2228 struct flow_keys *flkeys = NULL, _flkeys;
2229 struct net *net = dev_net(dev);
2230 struct ip_tunnel_info *tun_info;
2231 int err = -EINVAL;
2232 unsigned int flags = 0;
2233 u32 itag = 0;
2234 struct rtable *rth;
2235 struct flowi4 fl4;
2236 bool do_cache = true;
2237
2238 /* IP on this device is disabled. */
2239
2240 if (!in_dev)
2241 goto out;
2242
2243 /* Check for the most weird martians, which can be not detected
2244 * by fib_lookup.
2245 */
2246
2247 tun_info = skb_tunnel_info(skb);
2248 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2249 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2250 else
2251 fl4.flowi4_tun_key.tun_id = 0;
2252 skb_dst_drop(skb);
2253
2254 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2255 goto martian_source;
2256
2257 res->fi = NULL;
2258 res->table = NULL;
2259 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2260 goto brd_input;
2261
2262 /* Accept zero addresses only to limited broadcast;
2263 * I even do not know to fix it or not. Waiting for complains :-)
2264 */
2265 if (ipv4_is_zeronet(saddr))
2266 goto martian_source;
2267
2268 if (ipv4_is_zeronet(daddr))
2269 goto martian_destination;
2270
2271 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2272 * and call it once if daddr or/and saddr are loopback addresses
2273 */
2274 if (ipv4_is_loopback(daddr)) {
2275 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2276 goto martian_destination;
2277 } else if (ipv4_is_loopback(saddr)) {
2278 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2279 goto martian_source;
2280 }
2281
2282 /*
2283 * Now we are ready to route packet.
2284 */
2285 fl4.flowi4_oif = 0;
2286 fl4.flowi4_iif = dev->ifindex;
2287 fl4.flowi4_mark = skb->mark;
2288 fl4.flowi4_tos = tos;
2289 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2290 fl4.flowi4_flags = 0;
2291 fl4.daddr = daddr;
2292 fl4.saddr = saddr;
2293 fl4.flowi4_uid = sock_net_uid(net, NULL);
2294 fl4.flowi4_multipath_hash = 0;
2295
2296 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2297 flkeys = &_flkeys;
2298 } else {
2299 fl4.flowi4_proto = 0;
2300 fl4.fl4_sport = 0;
2301 fl4.fl4_dport = 0;
2302 }
2303
2304 err = fib_lookup(net, &fl4, res, 0);
2305 if (err != 0) {
2306 if (!IN_DEV_FORWARD(in_dev))
2307 err = -EHOSTUNREACH;
2308 goto no_route;
2309 }
2310
2311 if (res->type == RTN_BROADCAST) {
2312 if (IN_DEV_BFORWARD(in_dev))
2313 goto make_route;
2314 /* not do cache if bc_forwarding is enabled */
2315 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2316 do_cache = false;
2317 goto brd_input;
2318 }
2319
2320 if (res->type == RTN_LOCAL) {
2321 err = fib_validate_source(skb, saddr, daddr, tos,
2322 0, dev, in_dev, &itag);
2323 if (err < 0)
2324 goto martian_source;
2325 goto local_input;
2326 }
2327
2328 if (!IN_DEV_FORWARD(in_dev)) {
2329 err = -EHOSTUNREACH;
2330 goto no_route;
2331 }
2332 if (res->type != RTN_UNICAST)
2333 goto martian_destination;
2334
2335make_route:
2336 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2337out: return err;
2338
2339brd_input:
2340 if (skb->protocol != htons(ETH_P_IP))
2341 goto e_inval;
2342
2343 if (!ipv4_is_zeronet(saddr)) {
2344 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2345 in_dev, &itag);
2346 if (err < 0)
2347 goto martian_source;
2348 }
2349 flags |= RTCF_BROADCAST;
2350 res->type = RTN_BROADCAST;
2351 RT_CACHE_STAT_INC(in_brd);
2352
2353local_input:
2354 do_cache &= res->fi && !itag;
2355 if (do_cache) {
2356 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2357
2358 rth = rcu_dereference(nhc->nhc_rth_input);
2359 if (rt_cache_valid(rth)) {
2360 skb_dst_set_noref(skb, &rth->dst);
2361 err = 0;
2362 goto out;
2363 }
2364 }
2365
2366 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2367 flags | RTCF_LOCAL, res->type,
2368 IN_DEV_ORCONF(in_dev, NOPOLICY), false);
2369 if (!rth)
2370 goto e_nobufs;
2371
2372 rth->dst.output= ip_rt_bug;
2373#ifdef CONFIG_IP_ROUTE_CLASSID
2374 rth->dst.tclassid = itag;
2375#endif
2376 rth->rt_is_input = 1;
2377
2378 RT_CACHE_STAT_INC(in_slow_tot);
2379 if (res->type == RTN_UNREACHABLE) {
2380 rth->dst.input= ip_error;
2381 rth->dst.error= -err;
2382 rth->rt_flags &= ~RTCF_LOCAL;
2383 }
2384
2385 if (do_cache) {
2386 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2387
2388 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2389 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2390 WARN_ON(rth->dst.input == lwtunnel_input);
2391 rth->dst.lwtstate->orig_input = rth->dst.input;
2392 rth->dst.input = lwtunnel_input;
2393 }
2394
2395 if (unlikely(!rt_cache_route(nhc, rth)))
2396 rt_add_uncached_list(rth);
2397 }
2398 skb_dst_set(skb, &rth->dst);
2399 err = 0;
2400 goto out;
2401
2402no_route:
2403 RT_CACHE_STAT_INC(in_no_route);
2404 res->type = RTN_UNREACHABLE;
2405 res->fi = NULL;
2406 res->table = NULL;
2407 goto local_input;
2408
2409 /*
2410 * Do not cache martian addresses: they should be logged (RFC1812)
2411 */
2412martian_destination:
2413 RT_CACHE_STAT_INC(in_martian_dst);
2414#ifdef CONFIG_IP_ROUTE_VERBOSE
2415 if (IN_DEV_LOG_MARTIANS(in_dev))
2416 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2417 &daddr, &saddr, dev->name);
2418#endif
2419
2420e_inval:
2421 err = -EINVAL;
2422 goto out;
2423
2424e_nobufs:
2425 err = -ENOBUFS;
2426 goto out;
2427
2428martian_source:
2429 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2430 goto out;
2431}
2432
2433int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2434 u8 tos, struct net_device *dev)
2435{
2436 struct fib_result res;
2437 int err;
2438
2439 tos &= IPTOS_RT_MASK;
2440 rcu_read_lock();
2441 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2442 rcu_read_unlock();
2443
2444 return err;
2445}
2446EXPORT_SYMBOL(ip_route_input_noref);
2447
2448/* called with rcu_read_lock held */
2449int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2450 u8 tos, struct net_device *dev, struct fib_result *res)
2451{
2452 /* Multicast recognition logic is moved from route cache to here.
2453 * The problem was that too many Ethernet cards have broken/missing
2454 * hardware multicast filters :-( As result the host on multicasting
2455 * network acquires a lot of useless route cache entries, sort of
2456 * SDR messages from all the world. Now we try to get rid of them.
2457 * Really, provided software IP multicast filter is organized
2458 * reasonably (at least, hashed), it does not result in a slowdown
2459 * comparing with route cache reject entries.
2460 * Note, that multicast routers are not affected, because
2461 * route cache entry is created eventually.
2462 */
2463 if (ipv4_is_multicast(daddr)) {
2464 struct in_device *in_dev = __in_dev_get_rcu(dev);
2465 int our = 0;
2466 int err = -EINVAL;
2467
2468 if (!in_dev)
2469 return err;
2470 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2471 ip_hdr(skb)->protocol);
2472
2473 /* check l3 master if no match yet */
2474 if (!our && netif_is_l3_slave(dev)) {
2475 struct in_device *l3_in_dev;
2476
2477 l3_in_dev = __in_dev_get_rcu(skb->dev);
2478 if (l3_in_dev)
2479 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2480 ip_hdr(skb)->protocol);
2481 }
2482
2483 if (our
2484#ifdef CONFIG_IP_MROUTE
2485 ||
2486 (!ipv4_is_local_multicast(daddr) &&
2487 IN_DEV_MFORWARD(in_dev))
2488#endif
2489 ) {
2490 err = ip_route_input_mc(skb, daddr, saddr,
2491 tos, dev, our);
2492 }
2493 return err;
2494 }
2495
2496 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2497}
2498
2499/* called with rcu_read_lock() */
2500static struct rtable *__mkroute_output(const struct fib_result *res,
2501 const struct flowi4 *fl4, int orig_oif,
2502 struct net_device *dev_out,
2503 unsigned int flags)
2504{
2505 struct fib_info *fi = res->fi;
2506 struct fib_nh_exception *fnhe;
2507 struct in_device *in_dev;
2508 u16 type = res->type;
2509 struct rtable *rth;
2510 bool do_cache;
2511
2512 in_dev = __in_dev_get_rcu(dev_out);
2513 if (!in_dev)
2514 return ERR_PTR(-EINVAL);
2515
2516 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2517 if (ipv4_is_loopback(fl4->saddr) &&
2518 !(dev_out->flags & IFF_LOOPBACK) &&
2519 !netif_is_l3_master(dev_out))
2520 return ERR_PTR(-EINVAL);
2521
2522 if (ipv4_is_lbcast(fl4->daddr))
2523 type = RTN_BROADCAST;
2524 else if (ipv4_is_multicast(fl4->daddr))
2525 type = RTN_MULTICAST;
2526 else if (ipv4_is_zeronet(fl4->daddr))
2527 return ERR_PTR(-EINVAL);
2528
2529 if (dev_out->flags & IFF_LOOPBACK)
2530 flags |= RTCF_LOCAL;
2531
2532 do_cache = true;
2533 if (type == RTN_BROADCAST) {
2534 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2535 fi = NULL;
2536 } else if (type == RTN_MULTICAST) {
2537 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2538 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2539 fl4->flowi4_proto))
2540 flags &= ~RTCF_LOCAL;
2541 else
2542 do_cache = false;
2543 /* If multicast route do not exist use
2544 * default one, but do not gateway in this case.
2545 * Yes, it is hack.
2546 */
2547 if (fi && res->prefixlen < 4)
2548 fi = NULL;
2549 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2550 (orig_oif != dev_out->ifindex)) {
2551 /* For local routes that require a particular output interface
2552 * we do not want to cache the result. Caching the result
2553 * causes incorrect behaviour when there are multiple source
2554 * addresses on the interface, the end result being that if the
2555 * intended recipient is waiting on that interface for the
2556 * packet he won't receive it because it will be delivered on
2557 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2558 * be set to the loopback interface as well.
2559 */
2560 do_cache = false;
2561 }
2562
2563 fnhe = NULL;
2564 do_cache &= fi != NULL;
2565 if (fi) {
2566 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2567 struct rtable __rcu **prth;
2568
2569 fnhe = find_exception(nhc, fl4->daddr);
2570 if (!do_cache)
2571 goto add;
2572 if (fnhe) {
2573 prth = &fnhe->fnhe_rth_output;
2574 } else {
2575 if (unlikely(fl4->flowi4_flags &
2576 FLOWI_FLAG_KNOWN_NH &&
2577 !(nhc->nhc_gw_family &&
2578 nhc->nhc_scope == RT_SCOPE_LINK))) {
2579 do_cache = false;
2580 goto add;
2581 }
2582 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2583 }
2584 rth = rcu_dereference(*prth);
2585 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2586 return rth;
2587 }
2588
2589add:
2590 rth = rt_dst_alloc(dev_out, flags, type,
2591 IN_DEV_ORCONF(in_dev, NOPOLICY),
2592 IN_DEV_ORCONF(in_dev, NOXFRM));
2593 if (!rth)
2594 return ERR_PTR(-ENOBUFS);
2595
2596 rth->rt_iif = orig_oif;
2597
2598 RT_CACHE_STAT_INC(out_slow_tot);
2599
2600 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2601 if (flags & RTCF_LOCAL &&
2602 !(dev_out->flags & IFF_LOOPBACK)) {
2603 rth->dst.output = ip_mc_output;
2604 RT_CACHE_STAT_INC(out_slow_mc);
2605 }
2606#ifdef CONFIG_IP_MROUTE
2607 if (type == RTN_MULTICAST) {
2608 if (IN_DEV_MFORWARD(in_dev) &&
2609 !ipv4_is_local_multicast(fl4->daddr)) {
2610 rth->dst.input = ip_mr_input;
2611 rth->dst.output = ip_mc_output;
2612 }
2613 }
2614#endif
2615 }
2616
2617 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2618 lwtunnel_set_redirect(&rth->dst);
2619
2620 return rth;
2621}
2622
2623/*
2624 * Major route resolver routine.
2625 */
2626
2627struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2628 const struct sk_buff *skb)
2629{
2630 __u8 tos = RT_FL_TOS(fl4);
2631 struct fib_result res = {
2632 .type = RTN_UNSPEC,
2633 .fi = NULL,
2634 .table = NULL,
2635 .tclassid = 0,
2636 };
2637 struct rtable *rth;
2638
2639 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2640 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2641 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2642 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2643
2644 rcu_read_lock();
2645 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2646 rcu_read_unlock();
2647
2648 return rth;
2649}
2650EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2651
2652struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2653 struct fib_result *res,
2654 const struct sk_buff *skb)
2655{
2656 struct net_device *dev_out = NULL;
2657 int orig_oif = fl4->flowi4_oif;
2658 unsigned int flags = 0;
2659 struct rtable *rth;
2660 int err;
2661
2662 if (fl4->saddr) {
2663 if (ipv4_is_multicast(fl4->saddr) ||
2664 ipv4_is_lbcast(fl4->saddr) ||
2665 ipv4_is_zeronet(fl4->saddr)) {
2666 rth = ERR_PTR(-EINVAL);
2667 goto out;
2668 }
2669
2670 rth = ERR_PTR(-ENETUNREACH);
2671
2672 /* I removed check for oif == dev_out->oif here.
2673 * It was wrong for two reasons:
2674 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2675 * is assigned to multiple interfaces.
2676 * 2. Moreover, we are allowed to send packets with saddr
2677 * of another iface. --ANK
2678 */
2679
2680 if (fl4->flowi4_oif == 0 &&
2681 (ipv4_is_multicast(fl4->daddr) ||
2682 ipv4_is_lbcast(fl4->daddr))) {
2683 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2684 dev_out = __ip_dev_find(net, fl4->saddr, false);
2685 if (!dev_out)
2686 goto out;
2687
2688 /* Special hack: user can direct multicasts
2689 * and limited broadcast via necessary interface
2690 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2691 * This hack is not just for fun, it allows
2692 * vic,vat and friends to work.
2693 * They bind socket to loopback, set ttl to zero
2694 * and expect that it will work.
2695 * From the viewpoint of routing cache they are broken,
2696 * because we are not allowed to build multicast path
2697 * with loopback source addr (look, routing cache
2698 * cannot know, that ttl is zero, so that packet
2699 * will not leave this host and route is valid).
2700 * Luckily, this hack is good workaround.
2701 */
2702
2703 fl4->flowi4_oif = dev_out->ifindex;
2704 goto make_route;
2705 }
2706
2707 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2708 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2709 if (!__ip_dev_find(net, fl4->saddr, false))
2710 goto out;
2711 }
2712 }
2713
2714
2715 if (fl4->flowi4_oif) {
2716 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2717 rth = ERR_PTR(-ENODEV);
2718 if (!dev_out)
2719 goto out;
2720
2721 /* RACE: Check return value of inet_select_addr instead. */
2722 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2723 rth = ERR_PTR(-ENETUNREACH);
2724 goto out;
2725 }
2726 if (ipv4_is_local_multicast(fl4->daddr) ||
2727 ipv4_is_lbcast(fl4->daddr) ||
2728 fl4->flowi4_proto == IPPROTO_IGMP) {
2729 if (!fl4->saddr)
2730 fl4->saddr = inet_select_addr(dev_out, 0,
2731 RT_SCOPE_LINK);
2732 goto make_route;
2733 }
2734 if (!fl4->saddr) {
2735 if (ipv4_is_multicast(fl4->daddr))
2736 fl4->saddr = inet_select_addr(dev_out, 0,
2737 fl4->flowi4_scope);
2738 else if (!fl4->daddr)
2739 fl4->saddr = inet_select_addr(dev_out, 0,
2740 RT_SCOPE_HOST);
2741 }
2742 }
2743
2744 if (!fl4->daddr) {
2745 fl4->daddr = fl4->saddr;
2746 if (!fl4->daddr)
2747 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2748 dev_out = net->loopback_dev;
2749 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2750 res->type = RTN_LOCAL;
2751 flags |= RTCF_LOCAL;
2752 goto make_route;
2753 }
2754
2755 err = fib_lookup(net, fl4, res, 0);
2756 if (err) {
2757 res->fi = NULL;
2758 res->table = NULL;
2759 if (fl4->flowi4_oif &&
2760 (ipv4_is_multicast(fl4->daddr) ||
2761 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2762 /* Apparently, routing tables are wrong. Assume,
2763 * that the destination is on link.
2764 *
2765 * WHY? DW.
2766 * Because we are allowed to send to iface
2767 * even if it has NO routes and NO assigned
2768 * addresses. When oif is specified, routing
2769 * tables are looked up with only one purpose:
2770 * to catch if destination is gatewayed, rather than
2771 * direct. Moreover, if MSG_DONTROUTE is set,
2772 * we send packet, ignoring both routing tables
2773 * and ifaddr state. --ANK
2774 *
2775 *
2776 * We could make it even if oif is unknown,
2777 * likely IPv6, but we do not.
2778 */
2779
2780 if (fl4->saddr == 0)
2781 fl4->saddr = inet_select_addr(dev_out, 0,
2782 RT_SCOPE_LINK);
2783 res->type = RTN_UNICAST;
2784 goto make_route;
2785 }
2786 rth = ERR_PTR(err);
2787 goto out;
2788 }
2789
2790 if (res->type == RTN_LOCAL) {
2791 if (!fl4->saddr) {
2792 if (res->fi->fib_prefsrc)
2793 fl4->saddr = res->fi->fib_prefsrc;
2794 else
2795 fl4->saddr = fl4->daddr;
2796 }
2797
2798 /* L3 master device is the loopback for that domain */
2799 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2800 net->loopback_dev;
2801
2802 /* make sure orig_oif points to fib result device even
2803 * though packet rx/tx happens over loopback or l3mdev
2804 */
2805 orig_oif = FIB_RES_OIF(*res);
2806
2807 fl4->flowi4_oif = dev_out->ifindex;
2808 flags |= RTCF_LOCAL;
2809 goto make_route;
2810 }
2811
2812 fib_select_path(net, res, fl4, skb);
2813
2814 dev_out = FIB_RES_DEV(*res);
2815
2816make_route:
2817 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2818
2819out:
2820 return rth;
2821}
2822
2823static struct dst_ops ipv4_dst_blackhole_ops = {
2824 .family = AF_INET,
2825 .default_advmss = ipv4_default_advmss,
2826 .neigh_lookup = ipv4_neigh_lookup,
2827 .check = dst_blackhole_check,
2828 .cow_metrics = dst_blackhole_cow_metrics,
2829 .update_pmtu = dst_blackhole_update_pmtu,
2830 .redirect = dst_blackhole_redirect,
2831 .mtu = dst_blackhole_mtu,
2832};
2833
2834struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2835{
2836 struct rtable *ort = (struct rtable *) dst_orig;
2837 struct rtable *rt;
2838
2839 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2840 if (rt) {
2841 struct dst_entry *new = &rt->dst;
2842
2843 new->__use = 1;
2844 new->input = dst_discard;
2845 new->output = dst_discard_out;
2846
2847 new->dev = net->loopback_dev;
2848 if (new->dev)
2849 dev_hold(new->dev);
2850
2851 rt->rt_is_input = ort->rt_is_input;
2852 rt->rt_iif = ort->rt_iif;
2853 rt->rt_pmtu = ort->rt_pmtu;
2854 rt->rt_mtu_locked = ort->rt_mtu_locked;
2855
2856 rt->rt_genid = rt_genid_ipv4(net);
2857 rt->rt_flags = ort->rt_flags;
2858 rt->rt_type = ort->rt_type;
2859 rt->rt_uses_gateway = ort->rt_uses_gateway;
2860 rt->rt_gw_family = ort->rt_gw_family;
2861 if (rt->rt_gw_family == AF_INET)
2862 rt->rt_gw4 = ort->rt_gw4;
2863 else if (rt->rt_gw_family == AF_INET6)
2864 rt->rt_gw6 = ort->rt_gw6;
2865
2866 INIT_LIST_HEAD(&rt->rt_uncached);
2867 }
2868
2869 dst_release(dst_orig);
2870
2871 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2872}
2873
2874struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2875 const struct sock *sk)
2876{
2877 struct rtable *rt = __ip_route_output_key(net, flp4);
2878
2879 if (IS_ERR(rt))
2880 return rt;
2881
2882 if (flp4->flowi4_proto) {
2883 flp4->flowi4_oif = rt->dst.dev->ifindex;
2884 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2885 flowi4_to_flowi(flp4),
2886 sk, 0);
2887 }
2888
2889 return rt;
2890}
2891EXPORT_SYMBOL_GPL(ip_route_output_flow);
2892
2893struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2894 struct net_device *dev,
2895 struct net *net, __be32 *saddr,
2896 const struct ip_tunnel_info *info,
2897 u8 protocol, bool use_cache)
2898{
2899#ifdef CONFIG_DST_CACHE
2900 struct dst_cache *dst_cache;
2901#endif
2902 struct rtable *rt = NULL;
2903 struct flowi4 fl4;
2904 __u8 tos;
2905
2906#ifdef CONFIG_DST_CACHE
2907 dst_cache = (struct dst_cache *)&info->dst_cache;
2908 if (use_cache) {
2909 rt = dst_cache_get_ip4(dst_cache, saddr);
2910 if (rt)
2911 return rt;
2912 }
2913#endif
2914 memset(&fl4, 0, sizeof(fl4));
2915 fl4.flowi4_mark = skb->mark;
2916 fl4.flowi4_proto = protocol;
2917 fl4.daddr = info->key.u.ipv4.dst;
2918 fl4.saddr = info->key.u.ipv4.src;
2919 tos = info->key.tos;
2920 fl4.flowi4_tos = RT_TOS(tos);
2921
2922 rt = ip_route_output_key(net, &fl4);
2923 if (IS_ERR(rt)) {
2924 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2925 return ERR_PTR(-ENETUNREACH);
2926 }
2927 if (rt->dst.dev == dev) { /* is this necessary? */
2928 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2929 ip_rt_put(rt);
2930 return ERR_PTR(-ELOOP);
2931 }
2932#ifdef CONFIG_DST_CACHE
2933 if (use_cache)
2934 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2935#endif
2936 *saddr = fl4.saddr;
2937 return rt;
2938}
2939EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2940
2941/* called with rcu_read_lock held */
2942static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2943 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2944 struct sk_buff *skb, u32 portid, u32 seq,
2945 unsigned int flags)
2946{
2947 struct rtmsg *r;
2948 struct nlmsghdr *nlh;
2949 unsigned long expires = 0;
2950 u32 error;
2951 u32 metrics[RTAX_MAX];
2952
2953 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2954 if (!nlh)
2955 return -EMSGSIZE;
2956
2957 r = nlmsg_data(nlh);
2958 r->rtm_family = AF_INET;
2959 r->rtm_dst_len = 32;
2960 r->rtm_src_len = 0;
2961 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2962 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2963 if (nla_put_u32(skb, RTA_TABLE, table_id))
2964 goto nla_put_failure;
2965 r->rtm_type = rt->rt_type;
2966 r->rtm_scope = RT_SCOPE_UNIVERSE;
2967 r->rtm_protocol = RTPROT_UNSPEC;
2968 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2969 if (rt->rt_flags & RTCF_NOTIFY)
2970 r->rtm_flags |= RTM_F_NOTIFY;
2971 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2972 r->rtm_flags |= RTCF_DOREDIRECT;
2973
2974 if (nla_put_in_addr(skb, RTA_DST, dst))
2975 goto nla_put_failure;
2976 if (src) {
2977 r->rtm_src_len = 32;
2978 if (nla_put_in_addr(skb, RTA_SRC, src))
2979 goto nla_put_failure;
2980 }
2981 if (rt->dst.dev &&
2982 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2983 goto nla_put_failure;
2984 if (rt->dst.lwtstate &&
2985 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2986 goto nla_put_failure;
2987#ifdef CONFIG_IP_ROUTE_CLASSID
2988 if (rt->dst.tclassid &&
2989 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2990 goto nla_put_failure;
2991#endif
2992 if (fl4 && !rt_is_input_route(rt) &&
2993 fl4->saddr != src) {
2994 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2995 goto nla_put_failure;
2996 }
2997 if (rt->rt_uses_gateway) {
2998 if (rt->rt_gw_family == AF_INET &&
2999 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3000 goto nla_put_failure;
3001 } else if (rt->rt_gw_family == AF_INET6) {
3002 int alen = sizeof(struct in6_addr);
3003 struct nlattr *nla;
3004 struct rtvia *via;
3005
3006 nla = nla_reserve(skb, RTA_VIA, alen + 2);
3007 if (!nla)
3008 goto nla_put_failure;
3009
3010 via = nla_data(nla);
3011 via->rtvia_family = AF_INET6;
3012 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3013 }
3014 }
3015
3016 expires = rt->dst.expires;
3017 if (expires) {
3018 unsigned long now = jiffies;
3019
3020 if (time_before(now, expires))
3021 expires -= now;
3022 else
3023 expires = 0;
3024 }
3025
3026 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3027 if (rt->rt_pmtu && expires)
3028 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3029 if (rt->rt_mtu_locked && expires)
3030 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3031 if (rtnetlink_put_metrics(skb, metrics) < 0)
3032 goto nla_put_failure;
3033
3034 if (fl4) {
3035 if (fl4->flowi4_mark &&
3036 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3037 goto nla_put_failure;
3038
3039 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3040 nla_put_u32(skb, RTA_UID,
3041 from_kuid_munged(current_user_ns(),
3042 fl4->flowi4_uid)))
3043 goto nla_put_failure;
3044
3045 if (rt_is_input_route(rt)) {
3046#ifdef CONFIG_IP_MROUTE
3047 if (ipv4_is_multicast(dst) &&
3048 !ipv4_is_local_multicast(dst) &&
3049 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3050 int err = ipmr_get_route(net, skb,
3051 fl4->saddr, fl4->daddr,
3052 r, portid);
3053
3054 if (err <= 0) {
3055 if (err == 0)
3056 return 0;
3057 goto nla_put_failure;
3058 }
3059 } else
3060#endif
3061 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3062 goto nla_put_failure;
3063 }
3064 }
3065
3066 error = rt->dst.error;
3067
3068 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3069 goto nla_put_failure;
3070
3071 nlmsg_end(skb, nlh);
3072 return 0;
3073
3074nla_put_failure:
3075 nlmsg_cancel(skb, nlh);
3076 return -EMSGSIZE;
3077}
3078
3079static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3080 struct netlink_callback *cb, u32 table_id,
3081 struct fnhe_hash_bucket *bucket, int genid,
3082 int *fa_index, int fa_start, unsigned int flags)
3083{
3084 int i;
3085
3086 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3087 struct fib_nh_exception *fnhe;
3088
3089 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3090 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3091 struct rtable *rt;
3092 int err;
3093
3094 if (*fa_index < fa_start)
3095 goto next;
3096
3097 if (fnhe->fnhe_genid != genid)
3098 goto next;
3099
3100 if (fnhe->fnhe_expires &&
3101 time_after(jiffies, fnhe->fnhe_expires))
3102 goto next;
3103
3104 rt = rcu_dereference(fnhe->fnhe_rth_input);
3105 if (!rt)
3106 rt = rcu_dereference(fnhe->fnhe_rth_output);
3107 if (!rt)
3108 goto next;
3109
3110 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3111 table_id, NULL, skb,
3112 NETLINK_CB(cb->skb).portid,
3113 cb->nlh->nlmsg_seq, flags);
3114 if (err)
3115 return err;
3116next:
3117 (*fa_index)++;
3118 }
3119 }
3120
3121 return 0;
3122}
3123
3124int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3125 u32 table_id, struct fib_info *fi,
3126 int *fa_index, int fa_start, unsigned int flags)
3127{
3128 struct net *net = sock_net(cb->skb->sk);
3129 int nhsel, genid = fnhe_genid(net);
3130
3131 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3132 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3133 struct fnhe_hash_bucket *bucket;
3134 int err;
3135
3136 if (nhc->nhc_flags & RTNH_F_DEAD)
3137 continue;
3138
3139 rcu_read_lock();
3140 bucket = rcu_dereference(nhc->nhc_exceptions);
3141 err = 0;
3142 if (bucket)
3143 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3144 genid, fa_index, fa_start,
3145 flags);
3146 rcu_read_unlock();
3147 if (err)
3148 return err;
3149 }
3150
3151 return 0;
3152}
3153
3154static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3155 u8 ip_proto, __be16 sport,
3156 __be16 dport)
3157{
3158 struct sk_buff *skb;
3159 struct iphdr *iph;
3160
3161 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3162 if (!skb)
3163 return NULL;
3164
3165 /* Reserve room for dummy headers, this skb can pass
3166 * through good chunk of routing engine.
3167 */
3168 skb_reset_mac_header(skb);
3169 skb_reset_network_header(skb);
3170 skb->protocol = htons(ETH_P_IP);
3171 iph = skb_put(skb, sizeof(struct iphdr));
3172 iph->protocol = ip_proto;
3173 iph->saddr = src;
3174 iph->daddr = dst;
3175 iph->version = 0x4;
3176 iph->frag_off = 0;
3177 iph->ihl = 0x5;
3178 skb_set_transport_header(skb, skb->len);
3179
3180 switch (iph->protocol) {
3181 case IPPROTO_UDP: {
3182 struct udphdr *udph;
3183
3184 udph = skb_put_zero(skb, sizeof(struct udphdr));
3185 udph->source = sport;
3186 udph->dest = dport;
3187 udph->len = htons(sizeof(struct udphdr));
3188 udph->check = 0;
3189 break;
3190 }
3191 case IPPROTO_TCP: {
3192 struct tcphdr *tcph;
3193
3194 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3195 tcph->source = sport;
3196 tcph->dest = dport;
3197 tcph->doff = sizeof(struct tcphdr) / 4;
3198 tcph->rst = 1;
3199 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3200 src, dst, 0);
3201 break;
3202 }
3203 case IPPROTO_ICMP: {
3204 struct icmphdr *icmph;
3205
3206 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3207 icmph->type = ICMP_ECHO;
3208 icmph->code = 0;
3209 }
3210 }
3211
3212 return skb;
3213}
3214
3215static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3216 const struct nlmsghdr *nlh,
3217 struct nlattr **tb,
3218 struct netlink_ext_ack *extack)
3219{
3220 struct rtmsg *rtm;
3221 int i, err;
3222
3223 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3224 NL_SET_ERR_MSG(extack,
3225 "ipv4: Invalid header for route get request");
3226 return -EINVAL;
3227 }
3228
3229 if (!netlink_strict_get_check(skb))
3230 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3231 rtm_ipv4_policy, extack);
3232
3233 rtm = nlmsg_data(nlh);
3234 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3235 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3236 rtm->rtm_table || rtm->rtm_protocol ||
3237 rtm->rtm_scope || rtm->rtm_type) {
3238 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3239 return -EINVAL;
3240 }
3241
3242 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3243 RTM_F_LOOKUP_TABLE |
3244 RTM_F_FIB_MATCH)) {
3245 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3246 return -EINVAL;
3247 }
3248
3249 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3250 rtm_ipv4_policy, extack);
3251 if (err)
3252 return err;
3253
3254 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3255 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3256 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3257 return -EINVAL;
3258 }
3259
3260 for (i = 0; i <= RTA_MAX; i++) {
3261 if (!tb[i])
3262 continue;
3263
3264 switch (i) {
3265 case RTA_IIF:
3266 case RTA_OIF:
3267 case RTA_SRC:
3268 case RTA_DST:
3269 case RTA_IP_PROTO:
3270 case RTA_SPORT:
3271 case RTA_DPORT:
3272 case RTA_MARK:
3273 case RTA_UID:
3274 break;
3275 default:
3276 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3277 return -EINVAL;
3278 }
3279 }
3280
3281 return 0;
3282}
3283
3284static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3285 struct netlink_ext_ack *extack)
3286{
3287 struct net *net = sock_net(in_skb->sk);
3288 struct nlattr *tb[RTA_MAX+1];
3289 u32 table_id = RT_TABLE_MAIN;
3290 __be16 sport = 0, dport = 0;
3291 struct fib_result res = {};
3292 u8 ip_proto = IPPROTO_UDP;
3293 struct rtable *rt = NULL;
3294 struct sk_buff *skb;
3295 struct rtmsg *rtm;
3296 struct flowi4 fl4 = {};
3297 __be32 dst = 0;
3298 __be32 src = 0;
3299 kuid_t uid;
3300 u32 iif;
3301 int err;
3302 int mark;
3303
3304 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3305 if (err < 0)
3306 return err;
3307
3308 rtm = nlmsg_data(nlh);
3309 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3310 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3311 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3312 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3313 if (tb[RTA_UID])
3314 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3315 else
3316 uid = (iif ? INVALID_UID : current_uid());
3317
3318 if (tb[RTA_IP_PROTO]) {
3319 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3320 &ip_proto, AF_INET, extack);
3321 if (err)
3322 return err;
3323 }
3324
3325 if (tb[RTA_SPORT])
3326 sport = nla_get_be16(tb[RTA_SPORT]);
3327
3328 if (tb[RTA_DPORT])
3329 dport = nla_get_be16(tb[RTA_DPORT]);
3330
3331 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3332 if (!skb)
3333 return -ENOBUFS;
3334
3335 fl4.daddr = dst;
3336 fl4.saddr = src;
3337 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3338 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3339 fl4.flowi4_mark = mark;
3340 fl4.flowi4_uid = uid;
3341 if (sport)
3342 fl4.fl4_sport = sport;
3343 if (dport)
3344 fl4.fl4_dport = dport;
3345 fl4.flowi4_proto = ip_proto;
3346
3347 rcu_read_lock();
3348
3349 if (iif) {
3350 struct net_device *dev;
3351
3352 dev = dev_get_by_index_rcu(net, iif);
3353 if (!dev) {
3354 err = -ENODEV;
3355 goto errout_rcu;
3356 }
3357
3358 fl4.flowi4_iif = iif; /* for rt_fill_info */
3359 skb->dev = dev;
3360 skb->mark = mark;
3361 err = ip_route_input_rcu(skb, dst, src,
3362 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3363 &res);
3364
3365 rt = skb_rtable(skb);
3366 if (err == 0 && rt->dst.error)
3367 err = -rt->dst.error;
3368 } else {
3369 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3370 skb->dev = net->loopback_dev;
3371 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3372 err = 0;
3373 if (IS_ERR(rt))
3374 err = PTR_ERR(rt);
3375 else
3376 skb_dst_set(skb, &rt->dst);
3377 }
3378
3379 if (err)
3380 goto errout_rcu;
3381
3382 if (rtm->rtm_flags & RTM_F_NOTIFY)
3383 rt->rt_flags |= RTCF_NOTIFY;
3384
3385 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3386 table_id = res.table ? res.table->tb_id : 0;
3387
3388 /* reset skb for netlink reply msg */
3389 skb_trim(skb, 0);
3390 skb_reset_network_header(skb);
3391 skb_reset_transport_header(skb);
3392 skb_reset_mac_header(skb);
3393
3394 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3395 struct fib_rt_info fri;
3396
3397 if (!res.fi) {
3398 err = fib_props[res.type].error;
3399 if (!err)
3400 err = -EHOSTUNREACH;
3401 goto errout_rcu;
3402 }
3403 fri.fi = res.fi;
3404 fri.tb_id = table_id;
3405 fri.dst = res.prefix;
3406 fri.dst_len = res.prefixlen;
3407 fri.tos = fl4.flowi4_tos;
3408 fri.type = rt->rt_type;
3409 fri.offload = 0;
3410 fri.trap = 0;
3411 fri.offload_failed = 0;
3412 if (res.fa_head) {
3413 struct fib_alias *fa;
3414
3415 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3416 u8 slen = 32 - fri.dst_len;
3417
3418 if (fa->fa_slen == slen &&
3419 fa->tb_id == fri.tb_id &&
3420 fa->fa_tos == fri.tos &&
3421 fa->fa_info == res.fi &&
3422 fa->fa_type == fri.type) {
3423 fri.offload = fa->offload;
3424 fri.trap = fa->trap;
3425 break;
3426 }
3427 }
3428 }
3429 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3430 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3431 } else {
3432 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3433 NETLINK_CB(in_skb).portid,
3434 nlh->nlmsg_seq, 0);
3435 }
3436 if (err < 0)
3437 goto errout_rcu;
3438
3439 rcu_read_unlock();
3440
3441 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3442
3443errout_free:
3444 return err;
3445errout_rcu:
3446 rcu_read_unlock();
3447 kfree_skb(skb);
3448 goto errout_free;
3449}
3450
3451void ip_rt_multicast_event(struct in_device *in_dev)
3452{
3453 rt_cache_flush(dev_net(in_dev->dev));
3454}
3455
3456#ifdef CONFIG_SYSCTL
3457static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3458static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3459static int ip_rt_gc_elasticity __read_mostly = 8;
3460static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3461
3462static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3463 void *buffer, size_t *lenp, loff_t *ppos)
3464{
3465 struct net *net = (struct net *)__ctl->extra1;
3466
3467 if (write) {
3468 rt_cache_flush(net);
3469 fnhe_genid_bump(net);
3470 return 0;
3471 }
3472
3473 return -EINVAL;
3474}
3475
3476static struct ctl_table ipv4_route_table[] = {
3477 {
3478 .procname = "gc_thresh",
3479 .data = &ipv4_dst_ops.gc_thresh,
3480 .maxlen = sizeof(int),
3481 .mode = 0644,
3482 .proc_handler = proc_dointvec,
3483 },
3484 {
3485 .procname = "max_size",
3486 .data = &ip_rt_max_size,
3487 .maxlen = sizeof(int),
3488 .mode = 0644,
3489 .proc_handler = proc_dointvec,
3490 },
3491 {
3492 /* Deprecated. Use gc_min_interval_ms */
3493
3494 .procname = "gc_min_interval",
3495 .data = &ip_rt_gc_min_interval,
3496 .maxlen = sizeof(int),
3497 .mode = 0644,
3498 .proc_handler = proc_dointvec_jiffies,
3499 },
3500 {
3501 .procname = "gc_min_interval_ms",
3502 .data = &ip_rt_gc_min_interval,
3503 .maxlen = sizeof(int),
3504 .mode = 0644,
3505 .proc_handler = proc_dointvec_ms_jiffies,
3506 },
3507 {
3508 .procname = "gc_timeout",
3509 .data = &ip_rt_gc_timeout,
3510 .maxlen = sizeof(int),
3511 .mode = 0644,
3512 .proc_handler = proc_dointvec_jiffies,
3513 },
3514 {
3515 .procname = "gc_interval",
3516 .data = &ip_rt_gc_interval,
3517 .maxlen = sizeof(int),
3518 .mode = 0644,
3519 .proc_handler = proc_dointvec_jiffies,
3520 },
3521 {
3522 .procname = "redirect_load",
3523 .data = &ip_rt_redirect_load,
3524 .maxlen = sizeof(int),
3525 .mode = 0644,
3526 .proc_handler = proc_dointvec,
3527 },
3528 {
3529 .procname = "redirect_number",
3530 .data = &ip_rt_redirect_number,
3531 .maxlen = sizeof(int),
3532 .mode = 0644,
3533 .proc_handler = proc_dointvec,
3534 },
3535 {
3536 .procname = "redirect_silence",
3537 .data = &ip_rt_redirect_silence,
3538 .maxlen = sizeof(int),
3539 .mode = 0644,
3540 .proc_handler = proc_dointvec,
3541 },
3542 {
3543 .procname = "error_cost",
3544 .data = &ip_rt_error_cost,
3545 .maxlen = sizeof(int),
3546 .mode = 0644,
3547 .proc_handler = proc_dointvec,
3548 },
3549 {
3550 .procname = "error_burst",
3551 .data = &ip_rt_error_burst,
3552 .maxlen = sizeof(int),
3553 .mode = 0644,
3554 .proc_handler = proc_dointvec,
3555 },
3556 {
3557 .procname = "gc_elasticity",
3558 .data = &ip_rt_gc_elasticity,
3559 .maxlen = sizeof(int),
3560 .mode = 0644,
3561 .proc_handler = proc_dointvec,
3562 },
3563 {
3564 .procname = "mtu_expires",
3565 .data = &ip_rt_mtu_expires,
3566 .maxlen = sizeof(int),
3567 .mode = 0644,
3568 .proc_handler = proc_dointvec_jiffies,
3569 },
3570 {
3571 .procname = "min_pmtu",
3572 .data = &ip_rt_min_pmtu,
3573 .maxlen = sizeof(int),
3574 .mode = 0644,
3575 .proc_handler = proc_dointvec_minmax,
3576 .extra1 = &ip_min_valid_pmtu,
3577 },
3578 {
3579 .procname = "min_adv_mss",
3580 .data = &ip_rt_min_advmss,
3581 .maxlen = sizeof(int),
3582 .mode = 0644,
3583 .proc_handler = proc_dointvec,
3584 },
3585 { }
3586};
3587
3588static const char ipv4_route_flush_procname[] = "flush";
3589
3590static struct ctl_table ipv4_route_flush_table[] = {
3591 {
3592 .procname = ipv4_route_flush_procname,
3593 .maxlen = sizeof(int),
3594 .mode = 0200,
3595 .proc_handler = ipv4_sysctl_rtcache_flush,
3596 },
3597 { },
3598};
3599
3600static __net_init int sysctl_route_net_init(struct net *net)
3601{
3602 struct ctl_table *tbl;
3603
3604 tbl = ipv4_route_flush_table;
3605 if (!net_eq(net, &init_net)) {
3606 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3607 if (!tbl)
3608 goto err_dup;
3609
3610 /* Don't export non-whitelisted sysctls to unprivileged users */
3611 if (net->user_ns != &init_user_ns) {
3612 if (tbl[0].procname != ipv4_route_flush_procname)
3613 tbl[0].procname = NULL;
3614 }
3615 }
3616 tbl[0].extra1 = net;
3617
3618 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3619 if (!net->ipv4.route_hdr)
3620 goto err_reg;
3621 return 0;
3622
3623err_reg:
3624 if (tbl != ipv4_route_flush_table)
3625 kfree(tbl);
3626err_dup:
3627 return -ENOMEM;
3628}
3629
3630static __net_exit void sysctl_route_net_exit(struct net *net)
3631{
3632 struct ctl_table *tbl;
3633
3634 tbl = net->ipv4.route_hdr->ctl_table_arg;
3635 unregister_net_sysctl_table(net->ipv4.route_hdr);
3636 BUG_ON(tbl == ipv4_route_flush_table);
3637 kfree(tbl);
3638}
3639
3640static __net_initdata struct pernet_operations sysctl_route_ops = {
3641 .init = sysctl_route_net_init,
3642 .exit = sysctl_route_net_exit,
3643};
3644#endif
3645
3646static __net_init int rt_genid_init(struct net *net)
3647{
3648 atomic_set(&net->ipv4.rt_genid, 0);
3649 atomic_set(&net->fnhe_genid, 0);
3650 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3651 return 0;
3652}
3653
3654static __net_initdata struct pernet_operations rt_genid_ops = {
3655 .init = rt_genid_init,
3656};
3657
3658static int __net_init ipv4_inetpeer_init(struct net *net)
3659{
3660 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3661
3662 if (!bp)
3663 return -ENOMEM;
3664 inet_peer_base_init(bp);
3665 net->ipv4.peers = bp;
3666 return 0;
3667}
3668
3669static void __net_exit ipv4_inetpeer_exit(struct net *net)
3670{
3671 struct inet_peer_base *bp = net->ipv4.peers;
3672
3673 net->ipv4.peers = NULL;
3674 inetpeer_invalidate_tree(bp);
3675 kfree(bp);
3676}
3677
3678static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3679 .init = ipv4_inetpeer_init,
3680 .exit = ipv4_inetpeer_exit,
3681};
3682
3683#ifdef CONFIG_IP_ROUTE_CLASSID
3684struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3685#endif /* CONFIG_IP_ROUTE_CLASSID */
3686
3687int __init ip_rt_init(void)
3688{
3689 void *idents_hash;
3690 int cpu;
3691
3692 /* For modern hosts, this will use 2 MB of memory */
3693 idents_hash = alloc_large_system_hash("IP idents",
3694 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3695 0,
3696 16, /* one bucket per 64 KB */
3697 HASH_ZERO,
3698 NULL,
3699 &ip_idents_mask,
3700 2048,
3701 256*1024);
3702
3703 ip_idents = idents_hash;
3704
3705 prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3706
3707 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3708
3709 for_each_possible_cpu(cpu) {
3710 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3711
3712 INIT_LIST_HEAD(&ul->head);
3713 spin_lock_init(&ul->lock);
3714 }
3715#ifdef CONFIG_IP_ROUTE_CLASSID
3716 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3717 if (!ip_rt_acct)
3718 panic("IP: failed to allocate ip_rt_acct\n");
3719#endif
3720
3721 ipv4_dst_ops.kmem_cachep =
3722 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3723 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3724
3725 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3726
3727 if (dst_entries_init(&ipv4_dst_ops) < 0)
3728 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3729
3730 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3731 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3732
3733 ipv4_dst_ops.gc_thresh = ~0;
3734 ip_rt_max_size = INT_MAX;
3735
3736 devinet_init();
3737 ip_fib_init();
3738
3739 if (ip_rt_proc_init())
3740 pr_err("Unable to create route proc files\n");
3741#ifdef CONFIG_XFRM
3742 xfrm_init();
3743 xfrm4_init();
3744#endif
3745 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3746 RTNL_FLAG_DOIT_UNLOCKED);
3747
3748#ifdef CONFIG_SYSCTL
3749 register_pernet_subsys(&sysctl_route_ops);
3750#endif
3751 register_pernet_subsys(&rt_genid_ops);
3752 register_pernet_subsys(&ipv4_inetpeer_ops);
3753 return 0;
3754}
3755
3756#ifdef CONFIG_SYSCTL
3757/*
3758 * We really need to sanitize the damn ipv4 init order, then all
3759 * this nonsense will go away.
3760 */
3761void __init ip_static_sysctl_init(void)
3762{
3763 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3764}
3765#endif
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_FL_TOS(oldflp4) \
110 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
111
112#define RT_GC_TIMEOUT (300*HZ)
113
114#define DEFAULT_MIN_PMTU (512 + 20 + 20)
115#define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
116#define DEFAULT_MIN_ADVMSS 256
117static int ip_rt_max_size;
118static int ip_rt_redirect_number __read_mostly = 9;
119static int ip_rt_redirect_load __read_mostly = HZ / 50;
120static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
121static int ip_rt_error_cost __read_mostly = HZ;
122static int ip_rt_error_burst __read_mostly = 5 * HZ;
123
124static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125
126/*
127 * Interface to generic destination cache.
128 */
129
130INDIRECT_CALLABLE_SCOPE
131struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
132static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
133INDIRECT_CALLABLE_SCOPE
134unsigned int ipv4_mtu(const struct dst_entry *dst);
135static void ipv4_negative_advice(struct sock *sk,
136 struct dst_entry *dst);
137static void ipv4_link_failure(struct sk_buff *skb);
138static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
139 struct sk_buff *skb, u32 mtu,
140 bool confirm_neigh);
141static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
142 struct sk_buff *skb);
143static void ipv4_dst_destroy(struct dst_entry *dst);
144
145static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
146{
147 WARN_ON(1);
148 return NULL;
149}
150
151static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
152 struct sk_buff *skb,
153 const void *daddr);
154static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
155
156static struct dst_ops ipv4_dst_ops = {
157 .family = AF_INET,
158 .check = ipv4_dst_check,
159 .default_advmss = ipv4_default_advmss,
160 .mtu = ipv4_mtu,
161 .cow_metrics = ipv4_cow_metrics,
162 .destroy = ipv4_dst_destroy,
163 .negative_advice = ipv4_negative_advice,
164 .link_failure = ipv4_link_failure,
165 .update_pmtu = ip_rt_update_pmtu,
166 .redirect = ip_do_redirect,
167 .local_out = __ip_local_out,
168 .neigh_lookup = ipv4_neigh_lookup,
169 .confirm_neigh = ipv4_confirm_neigh,
170};
171
172#define ECN_OR_COST(class) TC_PRIO_##class
173
174const __u8 ip_tos2prio[16] = {
175 TC_PRIO_BESTEFFORT,
176 ECN_OR_COST(BESTEFFORT),
177 TC_PRIO_BESTEFFORT,
178 ECN_OR_COST(BESTEFFORT),
179 TC_PRIO_BULK,
180 ECN_OR_COST(BULK),
181 TC_PRIO_BULK,
182 ECN_OR_COST(BULK),
183 TC_PRIO_INTERACTIVE,
184 ECN_OR_COST(INTERACTIVE),
185 TC_PRIO_INTERACTIVE,
186 ECN_OR_COST(INTERACTIVE),
187 TC_PRIO_INTERACTIVE_BULK,
188 ECN_OR_COST(INTERACTIVE_BULK),
189 TC_PRIO_INTERACTIVE_BULK,
190 ECN_OR_COST(INTERACTIVE_BULK)
191};
192EXPORT_SYMBOL(ip_tos2prio);
193
194static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
195#define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
196
197#ifdef CONFIG_PROC_FS
198static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
199{
200 if (*pos)
201 return NULL;
202 return SEQ_START_TOKEN;
203}
204
205static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
206{
207 ++*pos;
208 return NULL;
209}
210
211static void rt_cache_seq_stop(struct seq_file *seq, void *v)
212{
213}
214
215static int rt_cache_seq_show(struct seq_file *seq, void *v)
216{
217 if (v == SEQ_START_TOKEN)
218 seq_printf(seq, "%-127s\n",
219 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
220 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
221 "HHUptod\tSpecDst");
222 return 0;
223}
224
225static const struct seq_operations rt_cache_seq_ops = {
226 .start = rt_cache_seq_start,
227 .next = rt_cache_seq_next,
228 .stop = rt_cache_seq_stop,
229 .show = rt_cache_seq_show,
230};
231
232static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
233{
234 int cpu;
235
236 if (*pos == 0)
237 return SEQ_START_TOKEN;
238
239 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
240 if (!cpu_possible(cpu))
241 continue;
242 *pos = cpu+1;
243 return &per_cpu(rt_cache_stat, cpu);
244 }
245 return NULL;
246}
247
248static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
249{
250 int cpu;
251
252 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
253 if (!cpu_possible(cpu))
254 continue;
255 *pos = cpu+1;
256 return &per_cpu(rt_cache_stat, cpu);
257 }
258 (*pos)++;
259 return NULL;
260
261}
262
263static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
264{
265
266}
267
268static int rt_cpu_seq_show(struct seq_file *seq, void *v)
269{
270 struct rt_cache_stat *st = v;
271
272 if (v == SEQ_START_TOKEN) {
273 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");
274 return 0;
275 }
276
277 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x "
278 "%08x %08x %08x %08x %08x %08x "
279 "%08x %08x %08x %08x\n",
280 dst_entries_get_slow(&ipv4_dst_ops),
281 0, /* st->in_hit */
282 st->in_slow_tot,
283 st->in_slow_mc,
284 st->in_no_route,
285 st->in_brd,
286 st->in_martian_dst,
287 st->in_martian_src,
288
289 0, /* st->out_hit */
290 st->out_slow_tot,
291 st->out_slow_mc,
292
293 0, /* st->gc_total */
294 0, /* st->gc_ignored */
295 0, /* st->gc_goal_miss */
296 0, /* st->gc_dst_overflow */
297 0, /* st->in_hlist_search */
298 0 /* st->out_hlist_search */
299 );
300 return 0;
301}
302
303static const struct seq_operations rt_cpu_seq_ops = {
304 .start = rt_cpu_seq_start,
305 .next = rt_cpu_seq_next,
306 .stop = rt_cpu_seq_stop,
307 .show = rt_cpu_seq_show,
308};
309
310#ifdef CONFIG_IP_ROUTE_CLASSID
311static int rt_acct_proc_show(struct seq_file *m, void *v)
312{
313 struct ip_rt_acct *dst, *src;
314 unsigned int i, j;
315
316 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
317 if (!dst)
318 return -ENOMEM;
319
320 for_each_possible_cpu(i) {
321 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
322 for (j = 0; j < 256; j++) {
323 dst[j].o_bytes += src[j].o_bytes;
324 dst[j].o_packets += src[j].o_packets;
325 dst[j].i_bytes += src[j].i_bytes;
326 dst[j].i_packets += src[j].i_packets;
327 }
328 }
329
330 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
331 kfree(dst);
332 return 0;
333}
334#endif
335
336static int __net_init ip_rt_do_proc_init(struct net *net)
337{
338 struct proc_dir_entry *pde;
339
340 pde = proc_create_seq("rt_cache", 0444, net->proc_net,
341 &rt_cache_seq_ops);
342 if (!pde)
343 goto err1;
344
345 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
346 &rt_cpu_seq_ops);
347 if (!pde)
348 goto err2;
349
350#ifdef CONFIG_IP_ROUTE_CLASSID
351 pde = proc_create_single("rt_acct", 0, net->proc_net,
352 rt_acct_proc_show);
353 if (!pde)
354 goto err3;
355#endif
356 return 0;
357
358#ifdef CONFIG_IP_ROUTE_CLASSID
359err3:
360 remove_proc_entry("rt_cache", net->proc_net_stat);
361#endif
362err2:
363 remove_proc_entry("rt_cache", net->proc_net);
364err1:
365 return -ENOMEM;
366}
367
368static void __net_exit ip_rt_do_proc_exit(struct net *net)
369{
370 remove_proc_entry("rt_cache", net->proc_net_stat);
371 remove_proc_entry("rt_cache", net->proc_net);
372#ifdef CONFIG_IP_ROUTE_CLASSID
373 remove_proc_entry("rt_acct", net->proc_net);
374#endif
375}
376
377static struct pernet_operations ip_rt_proc_ops __net_initdata = {
378 .init = ip_rt_do_proc_init,
379 .exit = ip_rt_do_proc_exit,
380};
381
382static int __init ip_rt_proc_init(void)
383{
384 return register_pernet_subsys(&ip_rt_proc_ops);
385}
386
387#else
388static inline int ip_rt_proc_init(void)
389{
390 return 0;
391}
392#endif /* CONFIG_PROC_FS */
393
394static inline bool rt_is_expired(const struct rtable *rth)
395{
396 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
397}
398
399void rt_cache_flush(struct net *net)
400{
401 rt_genid_bump_ipv4(net);
402}
403
404static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
405 struct sk_buff *skb,
406 const void *daddr)
407{
408 const struct rtable *rt = container_of(dst, struct rtable, dst);
409 struct net_device *dev = dst->dev;
410 struct neighbour *n;
411
412 rcu_read_lock();
413
414 if (likely(rt->rt_gw_family == AF_INET)) {
415 n = ip_neigh_gw4(dev, rt->rt_gw4);
416 } else if (rt->rt_gw_family == AF_INET6) {
417 n = ip_neigh_gw6(dev, &rt->rt_gw6);
418 } else {
419 __be32 pkey;
420
421 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
422 n = ip_neigh_gw4(dev, pkey);
423 }
424
425 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
426 n = NULL;
427
428 rcu_read_unlock();
429
430 return n;
431}
432
433static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
434{
435 const struct rtable *rt = container_of(dst, struct rtable, dst);
436 struct net_device *dev = dst->dev;
437 const __be32 *pkey = daddr;
438
439 if (rt->rt_gw_family == AF_INET) {
440 pkey = (const __be32 *)&rt->rt_gw4;
441 } else if (rt->rt_gw_family == AF_INET6) {
442 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
443 } else if (!daddr ||
444 (rt->rt_flags &
445 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
446 return;
447 }
448 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
449}
450
451/* Hash tables of size 2048..262144 depending on RAM size.
452 * Each bucket uses 8 bytes.
453 */
454static u32 ip_idents_mask __read_mostly;
455static atomic_t *ip_idents __read_mostly;
456static u32 *ip_tstamps __read_mostly;
457
458/* In order to protect privacy, we add a perturbation to identifiers
459 * if one generator is seldom used. This makes hard for an attacker
460 * to infer how many packets were sent between two points in time.
461 */
462static u32 ip_idents_reserve(u32 hash, int segs)
463{
464 u32 bucket, old, now = (u32)jiffies;
465 atomic_t *p_id;
466 u32 *p_tstamp;
467 u32 delta = 0;
468
469 bucket = hash & ip_idents_mask;
470 p_tstamp = ip_tstamps + bucket;
471 p_id = ip_idents + bucket;
472 old = READ_ONCE(*p_tstamp);
473
474 if (old != now && cmpxchg(p_tstamp, old, now) == old)
475 delta = get_random_u32_below(now - old);
476
477 /* If UBSAN reports an error there, please make sure your compiler
478 * supports -fno-strict-overflow before reporting it that was a bug
479 * in UBSAN, and it has been fixed in GCC-8.
480 */
481 return atomic_add_return(segs + delta, p_id) - segs;
482}
483
484void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
485{
486 u32 hash, id;
487
488 /* Note the following code is not safe, but this is okay. */
489 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
490 get_random_bytes(&net->ipv4.ip_id_key,
491 sizeof(net->ipv4.ip_id_key));
492
493 hash = siphash_3u32((__force u32)iph->daddr,
494 (__force u32)iph->saddr,
495 iph->protocol,
496 &net->ipv4.ip_id_key);
497 id = ip_idents_reserve(hash, segs);
498 iph->id = htons(id);
499}
500EXPORT_SYMBOL(__ip_select_ident);
501
502static void ip_rt_fix_tos(struct flowi4 *fl4)
503{
504 __u8 tos = RT_FL_TOS(fl4);
505
506 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
507 if (tos & RTO_ONLINK)
508 fl4->flowi4_scope = RT_SCOPE_LINK;
509}
510
511static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
512 const struct sock *sk, const struct iphdr *iph,
513 int oif, __u8 tos, u8 prot, u32 mark,
514 int flow_flags)
515{
516 __u8 scope = RT_SCOPE_UNIVERSE;
517
518 if (sk) {
519 oif = sk->sk_bound_dev_if;
520 mark = READ_ONCE(sk->sk_mark);
521 tos = ip_sock_rt_tos(sk);
522 scope = ip_sock_rt_scope(sk);
523 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
524 sk->sk_protocol;
525 }
526
527 flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
528 prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
529 sock_net_uid(net, sk));
530}
531
532static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
533 const struct sock *sk)
534{
535 const struct net *net = dev_net(skb->dev);
536 const struct iphdr *iph = ip_hdr(skb);
537 int oif = skb->dev->ifindex;
538 u8 prot = iph->protocol;
539 u32 mark = skb->mark;
540 __u8 tos = iph->tos;
541
542 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
543}
544
545static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
546{
547 const struct inet_sock *inet = inet_sk(sk);
548 const struct ip_options_rcu *inet_opt;
549 __be32 daddr = inet->inet_daddr;
550
551 rcu_read_lock();
552 inet_opt = rcu_dereference(inet->inet_opt);
553 if (inet_opt && inet_opt->opt.srr)
554 daddr = inet_opt->opt.faddr;
555 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
556 ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
557 ip_sock_rt_scope(sk),
558 inet_test_bit(HDRINCL, sk) ?
559 IPPROTO_RAW : sk->sk_protocol,
560 inet_sk_flowi_flags(sk),
561 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
562 rcu_read_unlock();
563}
564
565static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
566 const struct sk_buff *skb)
567{
568 if (skb)
569 build_skb_flow_key(fl4, skb, sk);
570 else
571 build_sk_flow_key(fl4, sk);
572}
573
574static DEFINE_SPINLOCK(fnhe_lock);
575
576static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
577{
578 struct rtable *rt;
579
580 rt = rcu_dereference(fnhe->fnhe_rth_input);
581 if (rt) {
582 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
583 dst_dev_put(&rt->dst);
584 dst_release(&rt->dst);
585 }
586 rt = rcu_dereference(fnhe->fnhe_rth_output);
587 if (rt) {
588 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
589 dst_dev_put(&rt->dst);
590 dst_release(&rt->dst);
591 }
592}
593
594static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
595{
596 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
597 struct fib_nh_exception *fnhe, *oldest = NULL;
598
599 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
600 fnhe = rcu_dereference_protected(*fnhe_p,
601 lockdep_is_held(&fnhe_lock));
602 if (!fnhe)
603 break;
604 if (!oldest ||
605 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
606 oldest = fnhe;
607 oldest_p = fnhe_p;
608 }
609 }
610 fnhe_flush_routes(oldest);
611 *oldest_p = oldest->fnhe_next;
612 kfree_rcu(oldest, rcu);
613}
614
615static u32 fnhe_hashfun(__be32 daddr)
616{
617 static siphash_aligned_key_t fnhe_hash_key;
618 u64 hval;
619
620 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
621 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
622 return hash_64(hval, FNHE_HASH_SHIFT);
623}
624
625static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
626{
627 rt->rt_pmtu = fnhe->fnhe_pmtu;
628 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
629 rt->dst.expires = fnhe->fnhe_expires;
630
631 if (fnhe->fnhe_gw) {
632 rt->rt_flags |= RTCF_REDIRECTED;
633 rt->rt_uses_gateway = 1;
634 rt->rt_gw_family = AF_INET;
635 rt->rt_gw4 = fnhe->fnhe_gw;
636 }
637}
638
639static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
640 __be32 gw, u32 pmtu, bool lock,
641 unsigned long expires)
642{
643 struct fnhe_hash_bucket *hash;
644 struct fib_nh_exception *fnhe;
645 struct rtable *rt;
646 u32 genid, hval;
647 unsigned int i;
648 int depth;
649
650 genid = fnhe_genid(dev_net(nhc->nhc_dev));
651 hval = fnhe_hashfun(daddr);
652
653 spin_lock_bh(&fnhe_lock);
654
655 hash = rcu_dereference(nhc->nhc_exceptions);
656 if (!hash) {
657 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
658 if (!hash)
659 goto out_unlock;
660 rcu_assign_pointer(nhc->nhc_exceptions, hash);
661 }
662
663 hash += hval;
664
665 depth = 0;
666 for (fnhe = rcu_dereference(hash->chain); fnhe;
667 fnhe = rcu_dereference(fnhe->fnhe_next)) {
668 if (fnhe->fnhe_daddr == daddr)
669 break;
670 depth++;
671 }
672
673 if (fnhe) {
674 if (fnhe->fnhe_genid != genid)
675 fnhe->fnhe_genid = genid;
676 if (gw)
677 fnhe->fnhe_gw = gw;
678 if (pmtu) {
679 fnhe->fnhe_pmtu = pmtu;
680 fnhe->fnhe_mtu_locked = lock;
681 }
682 fnhe->fnhe_expires = max(1UL, expires);
683 /* Update all cached dsts too */
684 rt = rcu_dereference(fnhe->fnhe_rth_input);
685 if (rt)
686 fill_route_from_fnhe(rt, fnhe);
687 rt = rcu_dereference(fnhe->fnhe_rth_output);
688 if (rt)
689 fill_route_from_fnhe(rt, fnhe);
690 } else {
691 /* Randomize max depth to avoid some side channels attacks. */
692 int max_depth = FNHE_RECLAIM_DEPTH +
693 get_random_u32_below(FNHE_RECLAIM_DEPTH);
694
695 while (depth > max_depth) {
696 fnhe_remove_oldest(hash);
697 depth--;
698 }
699
700 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
701 if (!fnhe)
702 goto out_unlock;
703
704 fnhe->fnhe_next = hash->chain;
705
706 fnhe->fnhe_genid = genid;
707 fnhe->fnhe_daddr = daddr;
708 fnhe->fnhe_gw = gw;
709 fnhe->fnhe_pmtu = pmtu;
710 fnhe->fnhe_mtu_locked = lock;
711 fnhe->fnhe_expires = max(1UL, expires);
712
713 rcu_assign_pointer(hash->chain, fnhe);
714
715 /* Exception created; mark the cached routes for the nexthop
716 * stale, so anyone caching it rechecks if this exception
717 * applies to them.
718 */
719 rt = rcu_dereference(nhc->nhc_rth_input);
720 if (rt)
721 rt->dst.obsolete = DST_OBSOLETE_KILL;
722
723 for_each_possible_cpu(i) {
724 struct rtable __rcu **prt;
725
726 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
727 rt = rcu_dereference(*prt);
728 if (rt)
729 rt->dst.obsolete = DST_OBSOLETE_KILL;
730 }
731 }
732
733 fnhe->fnhe_stamp = jiffies;
734
735out_unlock:
736 spin_unlock_bh(&fnhe_lock);
737}
738
739static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
740 bool kill_route)
741{
742 __be32 new_gw = icmp_hdr(skb)->un.gateway;
743 __be32 old_gw = ip_hdr(skb)->saddr;
744 struct net_device *dev = skb->dev;
745 struct in_device *in_dev;
746 struct fib_result res;
747 struct neighbour *n;
748 struct net *net;
749
750 switch (icmp_hdr(skb)->code & 7) {
751 case ICMP_REDIR_NET:
752 case ICMP_REDIR_NETTOS:
753 case ICMP_REDIR_HOST:
754 case ICMP_REDIR_HOSTTOS:
755 break;
756
757 default:
758 return;
759 }
760
761 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
762 return;
763
764 in_dev = __in_dev_get_rcu(dev);
765 if (!in_dev)
766 return;
767
768 net = dev_net(dev);
769 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
770 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
771 ipv4_is_zeronet(new_gw))
772 goto reject_redirect;
773
774 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
775 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
776 goto reject_redirect;
777 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
778 goto reject_redirect;
779 } else {
780 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
781 goto reject_redirect;
782 }
783
784 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
785 if (!n)
786 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
787 if (!IS_ERR(n)) {
788 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
789 neigh_event_send(n, NULL);
790 } else {
791 if (fib_lookup(net, fl4, &res, 0) == 0) {
792 struct fib_nh_common *nhc;
793
794 fib_select_path(net, &res, fl4, skb);
795 nhc = FIB_RES_NHC(res);
796 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
797 0, false,
798 jiffies + ip_rt_gc_timeout);
799 }
800 if (kill_route)
801 rt->dst.obsolete = DST_OBSOLETE_KILL;
802 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
803 }
804 neigh_release(n);
805 }
806 return;
807
808reject_redirect:
809#ifdef CONFIG_IP_ROUTE_VERBOSE
810 if (IN_DEV_LOG_MARTIANS(in_dev)) {
811 const struct iphdr *iph = (const struct iphdr *) skb->data;
812 __be32 daddr = iph->daddr;
813 __be32 saddr = iph->saddr;
814
815 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
816 " Advised path = %pI4 -> %pI4\n",
817 &old_gw, dev->name, &new_gw,
818 &saddr, &daddr);
819 }
820#endif
821 ;
822}
823
824static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
825{
826 struct rtable *rt;
827 struct flowi4 fl4;
828 const struct iphdr *iph = (const struct iphdr *) skb->data;
829 struct net *net = dev_net(skb->dev);
830 int oif = skb->dev->ifindex;
831 u8 prot = iph->protocol;
832 u32 mark = skb->mark;
833 __u8 tos = iph->tos;
834
835 rt = dst_rtable(dst);
836
837 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
838 __ip_do_redirect(rt, skb, &fl4, true);
839}
840
841static void ipv4_negative_advice(struct sock *sk,
842 struct dst_entry *dst)
843{
844 struct rtable *rt = dst_rtable(dst);
845
846 if ((dst->obsolete > 0) ||
847 (rt->rt_flags & RTCF_REDIRECTED) ||
848 rt->dst.expires)
849 sk_dst_reset(sk);
850}
851
852/*
853 * Algorithm:
854 * 1. The first ip_rt_redirect_number redirects are sent
855 * with exponential backoff, then we stop sending them at all,
856 * assuming that the host ignores our redirects.
857 * 2. If we did not see packets requiring redirects
858 * during ip_rt_redirect_silence, we assume that the host
859 * forgot redirected route and start to send redirects again.
860 *
861 * This algorithm is much cheaper and more intelligent than dumb load limiting
862 * in icmp.c.
863 *
864 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
865 * and "frag. need" (breaks PMTU discovery) in icmp.c.
866 */
867
868void ip_rt_send_redirect(struct sk_buff *skb)
869{
870 struct rtable *rt = skb_rtable(skb);
871 struct in_device *in_dev;
872 struct inet_peer *peer;
873 struct net *net;
874 int log_martians;
875 int vif;
876
877 rcu_read_lock();
878 in_dev = __in_dev_get_rcu(rt->dst.dev);
879 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
880 rcu_read_unlock();
881 return;
882 }
883 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
884 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
885 rcu_read_unlock();
886
887 net = dev_net(rt->dst.dev);
888 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
889 if (!peer) {
890 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
891 rt_nexthop(rt, ip_hdr(skb)->daddr));
892 return;
893 }
894
895 /* No redirected packets during ip_rt_redirect_silence;
896 * reset the algorithm.
897 */
898 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
899 peer->rate_tokens = 0;
900 peer->n_redirects = 0;
901 }
902
903 /* Too many ignored redirects; do not send anything
904 * set dst.rate_last to the last seen redirected packet.
905 */
906 if (peer->n_redirects >= ip_rt_redirect_number) {
907 peer->rate_last = jiffies;
908 goto out_put_peer;
909 }
910
911 /* Check for load limit; set rate_last to the latest sent
912 * redirect.
913 */
914 if (peer->n_redirects == 0 ||
915 time_after(jiffies,
916 (peer->rate_last +
917 (ip_rt_redirect_load << peer->n_redirects)))) {
918 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
919
920 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
921 peer->rate_last = jiffies;
922 ++peer->n_redirects;
923 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
924 peer->n_redirects == ip_rt_redirect_number)
925 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
926 &ip_hdr(skb)->saddr, inet_iif(skb),
927 &ip_hdr(skb)->daddr, &gw);
928 }
929out_put_peer:
930 inet_putpeer(peer);
931}
932
933static int ip_error(struct sk_buff *skb)
934{
935 struct rtable *rt = skb_rtable(skb);
936 struct net_device *dev = skb->dev;
937 struct in_device *in_dev;
938 struct inet_peer *peer;
939 unsigned long now;
940 struct net *net;
941 SKB_DR(reason);
942 bool send;
943 int code;
944
945 if (netif_is_l3_master(skb->dev)) {
946 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
947 if (!dev)
948 goto out;
949 }
950
951 in_dev = __in_dev_get_rcu(dev);
952
953 /* IP on this device is disabled. */
954 if (!in_dev)
955 goto out;
956
957 net = dev_net(rt->dst.dev);
958 if (!IN_DEV_FORWARD(in_dev)) {
959 switch (rt->dst.error) {
960 case EHOSTUNREACH:
961 SKB_DR_SET(reason, IP_INADDRERRORS);
962 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
963 break;
964
965 case ENETUNREACH:
966 SKB_DR_SET(reason, IP_INNOROUTES);
967 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
968 break;
969 }
970 goto out;
971 }
972
973 switch (rt->dst.error) {
974 case EINVAL:
975 default:
976 goto out;
977 case EHOSTUNREACH:
978 code = ICMP_HOST_UNREACH;
979 break;
980 case ENETUNREACH:
981 code = ICMP_NET_UNREACH;
982 SKB_DR_SET(reason, IP_INNOROUTES);
983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
984 break;
985 case EACCES:
986 code = ICMP_PKT_FILTERED;
987 break;
988 }
989
990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
991 l3mdev_master_ifindex(skb->dev), 1);
992
993 send = true;
994 if (peer) {
995 now = jiffies;
996 peer->rate_tokens += now - peer->rate_last;
997 if (peer->rate_tokens > ip_rt_error_burst)
998 peer->rate_tokens = ip_rt_error_burst;
999 peer->rate_last = now;
1000 if (peer->rate_tokens >= ip_rt_error_cost)
1001 peer->rate_tokens -= ip_rt_error_cost;
1002 else
1003 send = false;
1004 inet_putpeer(peer);
1005 }
1006 if (send)
1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1008
1009out: kfree_skb_reason(skb, reason);
1010 return 0;
1011}
1012
1013static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1014{
1015 struct dst_entry *dst = &rt->dst;
1016 struct net *net = dev_net(dst->dev);
1017 struct fib_result res;
1018 bool lock = false;
1019 u32 old_mtu;
1020
1021 if (ip_mtu_locked(dst))
1022 return;
1023
1024 old_mtu = ipv4_mtu(dst);
1025 if (old_mtu < mtu)
1026 return;
1027
1028 if (mtu < net->ipv4.ip_rt_min_pmtu) {
1029 lock = true;
1030 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1031 }
1032
1033 if (rt->rt_pmtu == mtu && !lock &&
1034 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1035 return;
1036
1037 rcu_read_lock();
1038 if (fib_lookup(net, fl4, &res, 0) == 0) {
1039 struct fib_nh_common *nhc;
1040
1041 fib_select_path(net, &res, fl4, NULL);
1042 nhc = FIB_RES_NHC(res);
1043 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1044 jiffies + net->ipv4.ip_rt_mtu_expires);
1045 }
1046 rcu_read_unlock();
1047}
1048
1049static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1050 struct sk_buff *skb, u32 mtu,
1051 bool confirm_neigh)
1052{
1053 struct rtable *rt = dst_rtable(dst);
1054 struct flowi4 fl4;
1055
1056 ip_rt_build_flow_key(&fl4, sk, skb);
1057
1058 /* Don't make lookup fail for bridged encapsulations */
1059 if (skb && netif_is_any_bridge_port(skb->dev))
1060 fl4.flowi4_oif = 0;
1061
1062 __ip_rt_update_pmtu(rt, &fl4, mtu);
1063}
1064
1065void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1066 int oif, u8 protocol)
1067{
1068 const struct iphdr *iph = (const struct iphdr *)skb->data;
1069 struct flowi4 fl4;
1070 struct rtable *rt;
1071 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1072
1073 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1074 0);
1075 rt = __ip_route_output_key(net, &fl4);
1076 if (!IS_ERR(rt)) {
1077 __ip_rt_update_pmtu(rt, &fl4, mtu);
1078 ip_rt_put(rt);
1079 }
1080}
1081EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1082
1083static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1084{
1085 const struct iphdr *iph = (const struct iphdr *)skb->data;
1086 struct flowi4 fl4;
1087 struct rtable *rt;
1088
1089 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1090
1091 if (!fl4.flowi4_mark)
1092 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1093
1094 rt = __ip_route_output_key(sock_net(sk), &fl4);
1095 if (!IS_ERR(rt)) {
1096 __ip_rt_update_pmtu(rt, &fl4, mtu);
1097 ip_rt_put(rt);
1098 }
1099}
1100
1101void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1102{
1103 const struct iphdr *iph = (const struct iphdr *)skb->data;
1104 struct flowi4 fl4;
1105 struct rtable *rt;
1106 struct dst_entry *odst = NULL;
1107 bool new = false;
1108 struct net *net = sock_net(sk);
1109
1110 bh_lock_sock(sk);
1111
1112 if (!ip_sk_accept_pmtu(sk))
1113 goto out;
1114
1115 odst = sk_dst_get(sk);
1116
1117 if (sock_owned_by_user(sk) || !odst) {
1118 __ipv4_sk_update_pmtu(skb, sk, mtu);
1119 goto out;
1120 }
1121
1122 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1123
1124 rt = dst_rtable(odst);
1125 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1126 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1127 if (IS_ERR(rt))
1128 goto out;
1129
1130 new = true;
1131 }
1132
1133 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1134
1135 if (!dst_check(&rt->dst, 0)) {
1136 if (new)
1137 dst_release(&rt->dst);
1138
1139 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1140 if (IS_ERR(rt))
1141 goto out;
1142
1143 new = true;
1144 }
1145
1146 if (new)
1147 sk_dst_set(sk, &rt->dst);
1148
1149out:
1150 bh_unlock_sock(sk);
1151 dst_release(odst);
1152}
1153EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1154
1155void ipv4_redirect(struct sk_buff *skb, struct net *net,
1156 int oif, u8 protocol)
1157{
1158 const struct iphdr *iph = (const struct iphdr *)skb->data;
1159 struct flowi4 fl4;
1160 struct rtable *rt;
1161
1162 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1163 rt = __ip_route_output_key(net, &fl4);
1164 if (!IS_ERR(rt)) {
1165 __ip_do_redirect(rt, skb, &fl4, false);
1166 ip_rt_put(rt);
1167 }
1168}
1169EXPORT_SYMBOL_GPL(ipv4_redirect);
1170
1171void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1172{
1173 const struct iphdr *iph = (const struct iphdr *)skb->data;
1174 struct flowi4 fl4;
1175 struct rtable *rt;
1176 struct net *net = sock_net(sk);
1177
1178 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1179 rt = __ip_route_output_key(net, &fl4);
1180 if (!IS_ERR(rt)) {
1181 __ip_do_redirect(rt, skb, &fl4, false);
1182 ip_rt_put(rt);
1183 }
1184}
1185EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1186
1187INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1188 u32 cookie)
1189{
1190 struct rtable *rt = dst_rtable(dst);
1191
1192 /* All IPV4 dsts are created with ->obsolete set to the value
1193 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1194 * into this function always.
1195 *
1196 * When a PMTU/redirect information update invalidates a route,
1197 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1198 * DST_OBSOLETE_DEAD.
1199 */
1200 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1201 return NULL;
1202 return dst;
1203}
1204EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1205
1206static void ipv4_send_dest_unreach(struct sk_buff *skb)
1207{
1208 struct net_device *dev;
1209 struct ip_options opt;
1210 int res;
1211
1212 /* Recompile ip options since IPCB may not be valid anymore.
1213 * Also check we have a reasonable ipv4 header.
1214 */
1215 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1216 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1217 return;
1218
1219 memset(&opt, 0, sizeof(opt));
1220 if (ip_hdr(skb)->ihl > 5) {
1221 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1222 return;
1223 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1224
1225 rcu_read_lock();
1226 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1227 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1228 rcu_read_unlock();
1229
1230 if (res)
1231 return;
1232 }
1233 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1234}
1235
1236static void ipv4_link_failure(struct sk_buff *skb)
1237{
1238 struct rtable *rt;
1239
1240 ipv4_send_dest_unreach(skb);
1241
1242 rt = skb_rtable(skb);
1243 if (rt)
1244 dst_set_expires(&rt->dst, 0);
1245}
1246
1247static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1248{
1249 pr_debug("%s: %pI4 -> %pI4, %s\n",
1250 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1251 skb->dev ? skb->dev->name : "?");
1252 kfree_skb(skb);
1253 WARN_ON(1);
1254 return 0;
1255}
1256
1257/*
1258 * We do not cache source address of outgoing interface,
1259 * because it is used only by IP RR, TS and SRR options,
1260 * so that it out of fast path.
1261 *
1262 * BTW remember: "addr" is allowed to be not aligned
1263 * in IP options!
1264 */
1265
1266void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1267{
1268 __be32 src;
1269
1270 if (rt_is_output_route(rt))
1271 src = ip_hdr(skb)->saddr;
1272 else {
1273 struct fib_result res;
1274 struct iphdr *iph = ip_hdr(skb);
1275 struct flowi4 fl4 = {
1276 .daddr = iph->daddr,
1277 .saddr = iph->saddr,
1278 .flowi4_tos = RT_TOS(iph->tos),
1279 .flowi4_oif = rt->dst.dev->ifindex,
1280 .flowi4_iif = skb->dev->ifindex,
1281 .flowi4_mark = skb->mark,
1282 };
1283
1284 rcu_read_lock();
1285 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1286 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1287 else
1288 src = inet_select_addr(rt->dst.dev,
1289 rt_nexthop(rt, iph->daddr),
1290 RT_SCOPE_UNIVERSE);
1291 rcu_read_unlock();
1292 }
1293 memcpy(addr, &src, 4);
1294}
1295
1296#ifdef CONFIG_IP_ROUTE_CLASSID
1297static void set_class_tag(struct rtable *rt, u32 tag)
1298{
1299 if (!(rt->dst.tclassid & 0xFFFF))
1300 rt->dst.tclassid |= tag & 0xFFFF;
1301 if (!(rt->dst.tclassid & 0xFFFF0000))
1302 rt->dst.tclassid |= tag & 0xFFFF0000;
1303}
1304#endif
1305
1306static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1307{
1308 struct net *net = dev_net(dst->dev);
1309 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1310 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1311 net->ipv4.ip_rt_min_advmss);
1312
1313 return min(advmss, IPV4_MAX_PMTU - header_size);
1314}
1315
1316INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1317{
1318 return ip_dst_mtu_maybe_forward(dst, false);
1319}
1320EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1321
1322static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1323{
1324 struct fnhe_hash_bucket *hash;
1325 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1326 u32 hval = fnhe_hashfun(daddr);
1327
1328 spin_lock_bh(&fnhe_lock);
1329
1330 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1331 lockdep_is_held(&fnhe_lock));
1332 hash += hval;
1333
1334 fnhe_p = &hash->chain;
1335 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1336 while (fnhe) {
1337 if (fnhe->fnhe_daddr == daddr) {
1338 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1339 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1340 /* set fnhe_daddr to 0 to ensure it won't bind with
1341 * new dsts in rt_bind_exception().
1342 */
1343 fnhe->fnhe_daddr = 0;
1344 fnhe_flush_routes(fnhe);
1345 kfree_rcu(fnhe, rcu);
1346 break;
1347 }
1348 fnhe_p = &fnhe->fnhe_next;
1349 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1350 lockdep_is_held(&fnhe_lock));
1351 }
1352
1353 spin_unlock_bh(&fnhe_lock);
1354}
1355
1356static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1357 __be32 daddr)
1358{
1359 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1360 struct fib_nh_exception *fnhe;
1361 u32 hval;
1362
1363 if (!hash)
1364 return NULL;
1365
1366 hval = fnhe_hashfun(daddr);
1367
1368 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1369 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1370 if (fnhe->fnhe_daddr == daddr) {
1371 if (fnhe->fnhe_expires &&
1372 time_after(jiffies, fnhe->fnhe_expires)) {
1373 ip_del_fnhe(nhc, daddr);
1374 break;
1375 }
1376 return fnhe;
1377 }
1378 }
1379 return NULL;
1380}
1381
1382/* MTU selection:
1383 * 1. mtu on route is locked - use it
1384 * 2. mtu from nexthop exception
1385 * 3. mtu from egress device
1386 */
1387
1388u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1389{
1390 struct fib_nh_common *nhc = res->nhc;
1391 struct net_device *dev = nhc->nhc_dev;
1392 struct fib_info *fi = res->fi;
1393 u32 mtu = 0;
1394
1395 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1396 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1397 mtu = fi->fib_mtu;
1398
1399 if (likely(!mtu)) {
1400 struct fib_nh_exception *fnhe;
1401
1402 fnhe = find_exception(nhc, daddr);
1403 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1404 mtu = fnhe->fnhe_pmtu;
1405 }
1406
1407 if (likely(!mtu))
1408 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1409
1410 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1411}
1412
1413static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1414 __be32 daddr, const bool do_cache)
1415{
1416 bool ret = false;
1417
1418 spin_lock_bh(&fnhe_lock);
1419
1420 if (daddr == fnhe->fnhe_daddr) {
1421 struct rtable __rcu **porig;
1422 struct rtable *orig;
1423 int genid = fnhe_genid(dev_net(rt->dst.dev));
1424
1425 if (rt_is_input_route(rt))
1426 porig = &fnhe->fnhe_rth_input;
1427 else
1428 porig = &fnhe->fnhe_rth_output;
1429 orig = rcu_dereference(*porig);
1430
1431 if (fnhe->fnhe_genid != genid) {
1432 fnhe->fnhe_genid = genid;
1433 fnhe->fnhe_gw = 0;
1434 fnhe->fnhe_pmtu = 0;
1435 fnhe->fnhe_expires = 0;
1436 fnhe->fnhe_mtu_locked = false;
1437 fnhe_flush_routes(fnhe);
1438 orig = NULL;
1439 }
1440 fill_route_from_fnhe(rt, fnhe);
1441 if (!rt->rt_gw4) {
1442 rt->rt_gw4 = daddr;
1443 rt->rt_gw_family = AF_INET;
1444 }
1445
1446 if (do_cache) {
1447 dst_hold(&rt->dst);
1448 rcu_assign_pointer(*porig, rt);
1449 if (orig) {
1450 dst_dev_put(&orig->dst);
1451 dst_release(&orig->dst);
1452 }
1453 ret = true;
1454 }
1455
1456 fnhe->fnhe_stamp = jiffies;
1457 }
1458 spin_unlock_bh(&fnhe_lock);
1459
1460 return ret;
1461}
1462
1463static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1464{
1465 struct rtable *orig, *prev, **p;
1466 bool ret = true;
1467
1468 if (rt_is_input_route(rt)) {
1469 p = (struct rtable **)&nhc->nhc_rth_input;
1470 } else {
1471 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1472 }
1473 orig = *p;
1474
1475 /* hold dst before doing cmpxchg() to avoid race condition
1476 * on this dst
1477 */
1478 dst_hold(&rt->dst);
1479 prev = cmpxchg(p, orig, rt);
1480 if (prev == orig) {
1481 if (orig) {
1482 rt_add_uncached_list(orig);
1483 dst_release(&orig->dst);
1484 }
1485 } else {
1486 dst_release(&rt->dst);
1487 ret = false;
1488 }
1489
1490 return ret;
1491}
1492
1493struct uncached_list {
1494 spinlock_t lock;
1495 struct list_head head;
1496 struct list_head quarantine;
1497};
1498
1499static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1500
1501void rt_add_uncached_list(struct rtable *rt)
1502{
1503 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1504
1505 rt->dst.rt_uncached_list = ul;
1506
1507 spin_lock_bh(&ul->lock);
1508 list_add_tail(&rt->dst.rt_uncached, &ul->head);
1509 spin_unlock_bh(&ul->lock);
1510}
1511
1512void rt_del_uncached_list(struct rtable *rt)
1513{
1514 if (!list_empty(&rt->dst.rt_uncached)) {
1515 struct uncached_list *ul = rt->dst.rt_uncached_list;
1516
1517 spin_lock_bh(&ul->lock);
1518 list_del_init(&rt->dst.rt_uncached);
1519 spin_unlock_bh(&ul->lock);
1520 }
1521}
1522
1523static void ipv4_dst_destroy(struct dst_entry *dst)
1524{
1525 ip_dst_metrics_put(dst);
1526 rt_del_uncached_list(dst_rtable(dst));
1527}
1528
1529void rt_flush_dev(struct net_device *dev)
1530{
1531 struct rtable *rt, *safe;
1532 int cpu;
1533
1534 for_each_possible_cpu(cpu) {
1535 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1536
1537 if (list_empty(&ul->head))
1538 continue;
1539
1540 spin_lock_bh(&ul->lock);
1541 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1542 if (rt->dst.dev != dev)
1543 continue;
1544 rt->dst.dev = blackhole_netdev;
1545 netdev_ref_replace(dev, blackhole_netdev,
1546 &rt->dst.dev_tracker, GFP_ATOMIC);
1547 list_move(&rt->dst.rt_uncached, &ul->quarantine);
1548 }
1549 spin_unlock_bh(&ul->lock);
1550 }
1551}
1552
1553static bool rt_cache_valid(const struct rtable *rt)
1554{
1555 return rt &&
1556 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1557 !rt_is_expired(rt);
1558}
1559
1560static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1561 const struct fib_result *res,
1562 struct fib_nh_exception *fnhe,
1563 struct fib_info *fi, u16 type, u32 itag,
1564 const bool do_cache)
1565{
1566 bool cached = false;
1567
1568 if (fi) {
1569 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1570
1571 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1572 rt->rt_uses_gateway = 1;
1573 rt->rt_gw_family = nhc->nhc_gw_family;
1574 /* only INET and INET6 are supported */
1575 if (likely(nhc->nhc_gw_family == AF_INET))
1576 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1577 else
1578 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1579 }
1580
1581 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1582
1583#ifdef CONFIG_IP_ROUTE_CLASSID
1584 if (nhc->nhc_family == AF_INET) {
1585 struct fib_nh *nh;
1586
1587 nh = container_of(nhc, struct fib_nh, nh_common);
1588 rt->dst.tclassid = nh->nh_tclassid;
1589 }
1590#endif
1591 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1592 if (unlikely(fnhe))
1593 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1594 else if (do_cache)
1595 cached = rt_cache_route(nhc, rt);
1596 if (unlikely(!cached)) {
1597 /* Routes we intend to cache in nexthop exception or
1598 * FIB nexthop have the DST_NOCACHE bit clear.
1599 * However, if we are unsuccessful at storing this
1600 * route into the cache we really need to set it.
1601 */
1602 if (!rt->rt_gw4) {
1603 rt->rt_gw_family = AF_INET;
1604 rt->rt_gw4 = daddr;
1605 }
1606 rt_add_uncached_list(rt);
1607 }
1608 } else
1609 rt_add_uncached_list(rt);
1610
1611#ifdef CONFIG_IP_ROUTE_CLASSID
1612#ifdef CONFIG_IP_MULTIPLE_TABLES
1613 set_class_tag(rt, res->tclassid);
1614#endif
1615 set_class_tag(rt, itag);
1616#endif
1617}
1618
1619struct rtable *rt_dst_alloc(struct net_device *dev,
1620 unsigned int flags, u16 type,
1621 bool noxfrm)
1622{
1623 struct rtable *rt;
1624
1625 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1626 (noxfrm ? DST_NOXFRM : 0));
1627
1628 if (rt) {
1629 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1630 rt->rt_flags = flags;
1631 rt->rt_type = type;
1632 rt->rt_is_input = 0;
1633 rt->rt_iif = 0;
1634 rt->rt_pmtu = 0;
1635 rt->rt_mtu_locked = 0;
1636 rt->rt_uses_gateway = 0;
1637 rt->rt_gw_family = 0;
1638 rt->rt_gw4 = 0;
1639
1640 rt->dst.output = ip_output;
1641 if (flags & RTCF_LOCAL)
1642 rt->dst.input = ip_local_deliver;
1643 }
1644
1645 return rt;
1646}
1647EXPORT_SYMBOL(rt_dst_alloc);
1648
1649struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1650{
1651 struct rtable *new_rt;
1652
1653 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1654 rt->dst.flags);
1655
1656 if (new_rt) {
1657 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1658 new_rt->rt_flags = rt->rt_flags;
1659 new_rt->rt_type = rt->rt_type;
1660 new_rt->rt_is_input = rt->rt_is_input;
1661 new_rt->rt_iif = rt->rt_iif;
1662 new_rt->rt_pmtu = rt->rt_pmtu;
1663 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1664 new_rt->rt_gw_family = rt->rt_gw_family;
1665 if (rt->rt_gw_family == AF_INET)
1666 new_rt->rt_gw4 = rt->rt_gw4;
1667 else if (rt->rt_gw_family == AF_INET6)
1668 new_rt->rt_gw6 = rt->rt_gw6;
1669
1670 new_rt->dst.input = rt->dst.input;
1671 new_rt->dst.output = rt->dst.output;
1672 new_rt->dst.error = rt->dst.error;
1673 new_rt->dst.lastuse = jiffies;
1674 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1675 }
1676 return new_rt;
1677}
1678EXPORT_SYMBOL(rt_dst_clone);
1679
1680/* called in rcu_read_lock() section */
1681int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1682 u8 tos, struct net_device *dev,
1683 struct in_device *in_dev, u32 *itag)
1684{
1685 int err;
1686
1687 /* Primary sanity checks. */
1688 if (!in_dev)
1689 return -EINVAL;
1690
1691 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1692 skb->protocol != htons(ETH_P_IP))
1693 return -EINVAL;
1694
1695 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1696 return -EINVAL;
1697
1698 if (ipv4_is_zeronet(saddr)) {
1699 if (!ipv4_is_local_multicast(daddr) &&
1700 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1701 return -EINVAL;
1702 } else {
1703 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1704 in_dev, itag);
1705 if (err < 0)
1706 return err;
1707 }
1708 return 0;
1709}
1710
1711/* called in rcu_read_lock() section */
1712static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1713 u8 tos, struct net_device *dev, int our)
1714{
1715 struct in_device *in_dev = __in_dev_get_rcu(dev);
1716 unsigned int flags = RTCF_MULTICAST;
1717 struct rtable *rth;
1718 u32 itag = 0;
1719 int err;
1720
1721 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1722 if (err)
1723 return err;
1724
1725 if (our)
1726 flags |= RTCF_LOCAL;
1727
1728 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1729 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1730
1731 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1732 false);
1733 if (!rth)
1734 return -ENOBUFS;
1735
1736#ifdef CONFIG_IP_ROUTE_CLASSID
1737 rth->dst.tclassid = itag;
1738#endif
1739 rth->dst.output = ip_rt_bug;
1740 rth->rt_is_input= 1;
1741
1742#ifdef CONFIG_IP_MROUTE
1743 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1744 rth->dst.input = ip_mr_input;
1745#endif
1746 RT_CACHE_STAT_INC(in_slow_mc);
1747
1748 skb_dst_drop(skb);
1749 skb_dst_set(skb, &rth->dst);
1750 return 0;
1751}
1752
1753
1754static void ip_handle_martian_source(struct net_device *dev,
1755 struct in_device *in_dev,
1756 struct sk_buff *skb,
1757 __be32 daddr,
1758 __be32 saddr)
1759{
1760 RT_CACHE_STAT_INC(in_martian_src);
1761#ifdef CONFIG_IP_ROUTE_VERBOSE
1762 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1763 /*
1764 * RFC1812 recommendation, if source is martian,
1765 * the only hint is MAC header.
1766 */
1767 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1768 &daddr, &saddr, dev->name);
1769 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1770 print_hex_dump(KERN_WARNING, "ll header: ",
1771 DUMP_PREFIX_OFFSET, 16, 1,
1772 skb_mac_header(skb),
1773 dev->hard_header_len, false);
1774 }
1775 }
1776#endif
1777}
1778
1779/* called in rcu_read_lock() section */
1780static int __mkroute_input(struct sk_buff *skb,
1781 const struct fib_result *res,
1782 struct in_device *in_dev,
1783 __be32 daddr, __be32 saddr, u32 tos)
1784{
1785 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1786 struct net_device *dev = nhc->nhc_dev;
1787 struct fib_nh_exception *fnhe;
1788 struct rtable *rth;
1789 int err;
1790 struct in_device *out_dev;
1791 bool do_cache;
1792 u32 itag = 0;
1793
1794 /* get a working reference to the output device */
1795 out_dev = __in_dev_get_rcu(dev);
1796 if (!out_dev) {
1797 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1798 return -EINVAL;
1799 }
1800
1801 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1802 in_dev->dev, in_dev, &itag);
1803 if (err < 0) {
1804 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1805 saddr);
1806
1807 goto cleanup;
1808 }
1809
1810 do_cache = res->fi && !itag;
1811 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1812 skb->protocol == htons(ETH_P_IP)) {
1813 __be32 gw;
1814
1815 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1816 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1817 inet_addr_onlink(out_dev, saddr, gw))
1818 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1819 }
1820
1821 if (skb->protocol != htons(ETH_P_IP)) {
1822 /* Not IP (i.e. ARP). Do not create route, if it is
1823 * invalid for proxy arp. DNAT routes are always valid.
1824 *
1825 * Proxy arp feature have been extended to allow, ARP
1826 * replies back to the same interface, to support
1827 * Private VLAN switch technologies. See arp.c.
1828 */
1829 if (out_dev == in_dev &&
1830 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1831 err = -EINVAL;
1832 goto cleanup;
1833 }
1834 }
1835
1836 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1837 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1838
1839 fnhe = find_exception(nhc, daddr);
1840 if (do_cache) {
1841 if (fnhe)
1842 rth = rcu_dereference(fnhe->fnhe_rth_input);
1843 else
1844 rth = rcu_dereference(nhc->nhc_rth_input);
1845 if (rt_cache_valid(rth)) {
1846 skb_dst_set_noref(skb, &rth->dst);
1847 goto out;
1848 }
1849 }
1850
1851 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1852 IN_DEV_ORCONF(out_dev, NOXFRM));
1853 if (!rth) {
1854 err = -ENOBUFS;
1855 goto cleanup;
1856 }
1857
1858 rth->rt_is_input = 1;
1859 RT_CACHE_STAT_INC(in_slow_tot);
1860
1861 rth->dst.input = ip_forward;
1862
1863 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1864 do_cache);
1865 lwtunnel_set_redirect(&rth->dst);
1866 skb_dst_set(skb, &rth->dst);
1867out:
1868 err = 0;
1869 cleanup:
1870 return err;
1871}
1872
1873#ifdef CONFIG_IP_ROUTE_MULTIPATH
1874/* To make ICMP packets follow the right flow, the multipath hash is
1875 * calculated from the inner IP addresses.
1876 */
1877static void ip_multipath_l3_keys(const struct sk_buff *skb,
1878 struct flow_keys *hash_keys)
1879{
1880 const struct iphdr *outer_iph = ip_hdr(skb);
1881 const struct iphdr *key_iph = outer_iph;
1882 const struct iphdr *inner_iph;
1883 const struct icmphdr *icmph;
1884 struct iphdr _inner_iph;
1885 struct icmphdr _icmph;
1886
1887 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1888 goto out;
1889
1890 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1891 goto out;
1892
1893 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1894 &_icmph);
1895 if (!icmph)
1896 goto out;
1897
1898 if (!icmp_is_err(icmph->type))
1899 goto out;
1900
1901 inner_iph = skb_header_pointer(skb,
1902 outer_iph->ihl * 4 + sizeof(_icmph),
1903 sizeof(_inner_iph), &_inner_iph);
1904 if (!inner_iph)
1905 goto out;
1906
1907 key_iph = inner_iph;
1908out:
1909 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1910 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1911}
1912
1913static u32 fib_multipath_custom_hash_outer(const struct net *net,
1914 const struct sk_buff *skb,
1915 bool *p_has_inner)
1916{
1917 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1918 struct flow_keys keys, hash_keys;
1919
1920 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1921 return 0;
1922
1923 memset(&hash_keys, 0, sizeof(hash_keys));
1924 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1925
1926 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1927 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1928 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1929 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1930 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1931 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1932 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1933 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1934 hash_keys.ports.src = keys.ports.src;
1935 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1936 hash_keys.ports.dst = keys.ports.dst;
1937
1938 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1939 return flow_hash_from_keys(&hash_keys);
1940}
1941
1942static u32 fib_multipath_custom_hash_inner(const struct net *net,
1943 const struct sk_buff *skb,
1944 bool has_inner)
1945{
1946 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1947 struct flow_keys keys, hash_keys;
1948
1949 /* We assume the packet carries an encapsulation, but if none was
1950 * encountered during dissection of the outer flow, then there is no
1951 * point in calling the flow dissector again.
1952 */
1953 if (!has_inner)
1954 return 0;
1955
1956 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1957 return 0;
1958
1959 memset(&hash_keys, 0, sizeof(hash_keys));
1960 skb_flow_dissect_flow_keys(skb, &keys, 0);
1961
1962 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1963 return 0;
1964
1965 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1966 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1967 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1968 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1969 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1970 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1971 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1972 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1973 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1974 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1975 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1976 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1977 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1978 hash_keys.tags.flow_label = keys.tags.flow_label;
1979 }
1980
1981 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1982 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1983 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1984 hash_keys.ports.src = keys.ports.src;
1985 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1986 hash_keys.ports.dst = keys.ports.dst;
1987
1988 return flow_hash_from_keys(&hash_keys);
1989}
1990
1991static u32 fib_multipath_custom_hash_skb(const struct net *net,
1992 const struct sk_buff *skb)
1993{
1994 u32 mhash, mhash_inner;
1995 bool has_inner = true;
1996
1997 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1998 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1999
2000 return jhash_2words(mhash, mhash_inner, 0);
2001}
2002
2003static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2004 const struct flowi4 *fl4)
2005{
2006 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2007 struct flow_keys hash_keys;
2008
2009 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2010 return 0;
2011
2012 memset(&hash_keys, 0, sizeof(hash_keys));
2013 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2014 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2015 hash_keys.addrs.v4addrs.src = fl4->saddr;
2016 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2017 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2018 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2019 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2020 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2021 hash_keys.ports.src = fl4->fl4_sport;
2022 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2023 hash_keys.ports.dst = fl4->fl4_dport;
2024
2025 return flow_hash_from_keys(&hash_keys);
2026}
2027
2028/* if skb is set it will be used and fl4 can be NULL */
2029int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2030 const struct sk_buff *skb, struct flow_keys *flkeys)
2031{
2032 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2033 struct flow_keys hash_keys;
2034 u32 mhash = 0;
2035
2036 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2037 case 0:
2038 memset(&hash_keys, 0, sizeof(hash_keys));
2039 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2040 if (skb) {
2041 ip_multipath_l3_keys(skb, &hash_keys);
2042 } else {
2043 hash_keys.addrs.v4addrs.src = fl4->saddr;
2044 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2045 }
2046 mhash = flow_hash_from_keys(&hash_keys);
2047 break;
2048 case 1:
2049 /* skb is currently provided only when forwarding */
2050 if (skb) {
2051 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2052 struct flow_keys keys;
2053
2054 /* short-circuit if we already have L4 hash present */
2055 if (skb->l4_hash)
2056 return skb_get_hash_raw(skb) >> 1;
2057
2058 memset(&hash_keys, 0, sizeof(hash_keys));
2059
2060 if (!flkeys) {
2061 skb_flow_dissect_flow_keys(skb, &keys, flag);
2062 flkeys = &keys;
2063 }
2064
2065 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2066 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2067 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2068 hash_keys.ports.src = flkeys->ports.src;
2069 hash_keys.ports.dst = flkeys->ports.dst;
2070 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2071 } else {
2072 memset(&hash_keys, 0, sizeof(hash_keys));
2073 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2074 hash_keys.addrs.v4addrs.src = fl4->saddr;
2075 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2076 hash_keys.ports.src = fl4->fl4_sport;
2077 hash_keys.ports.dst = fl4->fl4_dport;
2078 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2079 }
2080 mhash = flow_hash_from_keys(&hash_keys);
2081 break;
2082 case 2:
2083 memset(&hash_keys, 0, sizeof(hash_keys));
2084 /* skb is currently provided only when forwarding */
2085 if (skb) {
2086 struct flow_keys keys;
2087
2088 skb_flow_dissect_flow_keys(skb, &keys, 0);
2089 /* Inner can be v4 or v6 */
2090 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2091 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2092 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2093 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2094 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2095 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2096 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2097 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2098 hash_keys.tags.flow_label = keys.tags.flow_label;
2099 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2100 } else {
2101 /* Same as case 0 */
2102 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2103 ip_multipath_l3_keys(skb, &hash_keys);
2104 }
2105 } else {
2106 /* Same as case 0 */
2107 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2108 hash_keys.addrs.v4addrs.src = fl4->saddr;
2109 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2110 }
2111 mhash = flow_hash_from_keys(&hash_keys);
2112 break;
2113 case 3:
2114 if (skb)
2115 mhash = fib_multipath_custom_hash_skb(net, skb);
2116 else
2117 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2118 break;
2119 }
2120
2121 if (multipath_hash)
2122 mhash = jhash_2words(mhash, multipath_hash, 0);
2123
2124 return mhash >> 1;
2125}
2126#endif /* CONFIG_IP_ROUTE_MULTIPATH */
2127
2128static int ip_mkroute_input(struct sk_buff *skb,
2129 struct fib_result *res,
2130 struct in_device *in_dev,
2131 __be32 daddr, __be32 saddr, u32 tos,
2132 struct flow_keys *hkeys)
2133{
2134#ifdef CONFIG_IP_ROUTE_MULTIPATH
2135 if (res->fi && fib_info_num_path(res->fi) > 1) {
2136 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2137
2138 fib_select_multipath(res, h);
2139 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2140 }
2141#endif
2142
2143 /* create a routing cache entry */
2144 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2145}
2146
2147/* Implements all the saddr-related checks as ip_route_input_slow(),
2148 * assuming daddr is valid and the destination is not a local broadcast one.
2149 * Uses the provided hint instead of performing a route lookup.
2150 */
2151int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2152 u8 tos, struct net_device *dev,
2153 const struct sk_buff *hint)
2154{
2155 struct in_device *in_dev = __in_dev_get_rcu(dev);
2156 struct rtable *rt = skb_rtable(hint);
2157 struct net *net = dev_net(dev);
2158 int err = -EINVAL;
2159 u32 tag = 0;
2160
2161 if (!in_dev)
2162 return -EINVAL;
2163
2164 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2165 goto martian_source;
2166
2167 if (ipv4_is_zeronet(saddr))
2168 goto martian_source;
2169
2170 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2171 goto martian_source;
2172
2173 if (rt->rt_type != RTN_LOCAL)
2174 goto skip_validate_source;
2175
2176 tos &= IPTOS_RT_MASK;
2177 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2178 if (err < 0)
2179 goto martian_source;
2180
2181skip_validate_source:
2182 skb_dst_copy(skb, hint);
2183 return 0;
2184
2185martian_source:
2186 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2187 return err;
2188}
2189
2190/* get device for dst_alloc with local routes */
2191static struct net_device *ip_rt_get_dev(struct net *net,
2192 const struct fib_result *res)
2193{
2194 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2195 struct net_device *dev = NULL;
2196
2197 if (nhc)
2198 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2199
2200 return dev ? : net->loopback_dev;
2201}
2202
2203/*
2204 * NOTE. We drop all the packets that has local source
2205 * addresses, because every properly looped back packet
2206 * must have correct destination already attached by output routine.
2207 * Changes in the enforced policies must be applied also to
2208 * ip_route_use_hint().
2209 *
2210 * Such approach solves two big problems:
2211 * 1. Not simplex devices are handled properly.
2212 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2213 * called with rcu_read_lock()
2214 */
2215
2216static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2217 u8 tos, struct net_device *dev,
2218 struct fib_result *res)
2219{
2220 struct in_device *in_dev = __in_dev_get_rcu(dev);
2221 struct flow_keys *flkeys = NULL, _flkeys;
2222 struct net *net = dev_net(dev);
2223 struct ip_tunnel_info *tun_info;
2224 int err = -EINVAL;
2225 unsigned int flags = 0;
2226 u32 itag = 0;
2227 struct rtable *rth;
2228 struct flowi4 fl4;
2229 bool do_cache = true;
2230
2231 /* IP on this device is disabled. */
2232
2233 if (!in_dev)
2234 goto out;
2235
2236 /* Check for the most weird martians, which can be not detected
2237 * by fib_lookup.
2238 */
2239
2240 tun_info = skb_tunnel_info(skb);
2241 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2242 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2243 else
2244 fl4.flowi4_tun_key.tun_id = 0;
2245 skb_dst_drop(skb);
2246
2247 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2248 goto martian_source;
2249
2250 res->fi = NULL;
2251 res->table = NULL;
2252 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2253 goto brd_input;
2254
2255 /* Accept zero addresses only to limited broadcast;
2256 * I even do not know to fix it or not. Waiting for complains :-)
2257 */
2258 if (ipv4_is_zeronet(saddr))
2259 goto martian_source;
2260
2261 if (ipv4_is_zeronet(daddr))
2262 goto martian_destination;
2263
2264 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2265 * and call it once if daddr or/and saddr are loopback addresses
2266 */
2267 if (ipv4_is_loopback(daddr)) {
2268 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2269 goto martian_destination;
2270 } else if (ipv4_is_loopback(saddr)) {
2271 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2272 goto martian_source;
2273 }
2274
2275 /*
2276 * Now we are ready to route packet.
2277 */
2278 fl4.flowi4_l3mdev = 0;
2279 fl4.flowi4_oif = 0;
2280 fl4.flowi4_iif = dev->ifindex;
2281 fl4.flowi4_mark = skb->mark;
2282 fl4.flowi4_tos = tos;
2283 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2284 fl4.flowi4_flags = 0;
2285 fl4.daddr = daddr;
2286 fl4.saddr = saddr;
2287 fl4.flowi4_uid = sock_net_uid(net, NULL);
2288 fl4.flowi4_multipath_hash = 0;
2289
2290 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2291 flkeys = &_flkeys;
2292 } else {
2293 fl4.flowi4_proto = 0;
2294 fl4.fl4_sport = 0;
2295 fl4.fl4_dport = 0;
2296 }
2297
2298 err = fib_lookup(net, &fl4, res, 0);
2299 if (err != 0) {
2300 if (!IN_DEV_FORWARD(in_dev))
2301 err = -EHOSTUNREACH;
2302 goto no_route;
2303 }
2304
2305 if (res->type == RTN_BROADCAST) {
2306 if (IN_DEV_BFORWARD(in_dev))
2307 goto make_route;
2308 /* not do cache if bc_forwarding is enabled */
2309 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2310 do_cache = false;
2311 goto brd_input;
2312 }
2313
2314 if (res->type == RTN_LOCAL) {
2315 err = fib_validate_source(skb, saddr, daddr, tos,
2316 0, dev, in_dev, &itag);
2317 if (err < 0)
2318 goto martian_source;
2319 goto local_input;
2320 }
2321
2322 if (!IN_DEV_FORWARD(in_dev)) {
2323 err = -EHOSTUNREACH;
2324 goto no_route;
2325 }
2326 if (res->type != RTN_UNICAST)
2327 goto martian_destination;
2328
2329make_route:
2330 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2331out: return err;
2332
2333brd_input:
2334 if (skb->protocol != htons(ETH_P_IP))
2335 goto e_inval;
2336
2337 if (!ipv4_is_zeronet(saddr)) {
2338 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2339 in_dev, &itag);
2340 if (err < 0)
2341 goto martian_source;
2342 }
2343 flags |= RTCF_BROADCAST;
2344 res->type = RTN_BROADCAST;
2345 RT_CACHE_STAT_INC(in_brd);
2346
2347local_input:
2348 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2349 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2350
2351 do_cache &= res->fi && !itag;
2352 if (do_cache) {
2353 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2354
2355 rth = rcu_dereference(nhc->nhc_rth_input);
2356 if (rt_cache_valid(rth)) {
2357 skb_dst_set_noref(skb, &rth->dst);
2358 err = 0;
2359 goto out;
2360 }
2361 }
2362
2363 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2364 flags | RTCF_LOCAL, res->type, false);
2365 if (!rth)
2366 goto e_nobufs;
2367
2368 rth->dst.output= ip_rt_bug;
2369#ifdef CONFIG_IP_ROUTE_CLASSID
2370 rth->dst.tclassid = itag;
2371#endif
2372 rth->rt_is_input = 1;
2373
2374 RT_CACHE_STAT_INC(in_slow_tot);
2375 if (res->type == RTN_UNREACHABLE) {
2376 rth->dst.input= ip_error;
2377 rth->dst.error= -err;
2378 rth->rt_flags &= ~RTCF_LOCAL;
2379 }
2380
2381 if (do_cache) {
2382 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2383
2384 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2385 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2386 WARN_ON(rth->dst.input == lwtunnel_input);
2387 rth->dst.lwtstate->orig_input = rth->dst.input;
2388 rth->dst.input = lwtunnel_input;
2389 }
2390
2391 if (unlikely(!rt_cache_route(nhc, rth)))
2392 rt_add_uncached_list(rth);
2393 }
2394 skb_dst_set(skb, &rth->dst);
2395 err = 0;
2396 goto out;
2397
2398no_route:
2399 RT_CACHE_STAT_INC(in_no_route);
2400 res->type = RTN_UNREACHABLE;
2401 res->fi = NULL;
2402 res->table = NULL;
2403 goto local_input;
2404
2405 /*
2406 * Do not cache martian addresses: they should be logged (RFC1812)
2407 */
2408martian_destination:
2409 RT_CACHE_STAT_INC(in_martian_dst);
2410#ifdef CONFIG_IP_ROUTE_VERBOSE
2411 if (IN_DEV_LOG_MARTIANS(in_dev))
2412 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2413 &daddr, &saddr, dev->name);
2414#endif
2415
2416e_inval:
2417 err = -EINVAL;
2418 goto out;
2419
2420e_nobufs:
2421 err = -ENOBUFS;
2422 goto out;
2423
2424martian_source:
2425 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2426 goto out;
2427}
2428
2429/* called with rcu_read_lock held */
2430static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2431 u8 tos, struct net_device *dev, struct fib_result *res)
2432{
2433 /* Multicast recognition logic is moved from route cache to here.
2434 * The problem was that too many Ethernet cards have broken/missing
2435 * hardware multicast filters :-( As result the host on multicasting
2436 * network acquires a lot of useless route cache entries, sort of
2437 * SDR messages from all the world. Now we try to get rid of them.
2438 * Really, provided software IP multicast filter is organized
2439 * reasonably (at least, hashed), it does not result in a slowdown
2440 * comparing with route cache reject entries.
2441 * Note, that multicast routers are not affected, because
2442 * route cache entry is created eventually.
2443 */
2444 if (ipv4_is_multicast(daddr)) {
2445 struct in_device *in_dev = __in_dev_get_rcu(dev);
2446 int our = 0;
2447 int err = -EINVAL;
2448
2449 if (!in_dev)
2450 return err;
2451 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2452 ip_hdr(skb)->protocol);
2453
2454 /* check l3 master if no match yet */
2455 if (!our && netif_is_l3_slave(dev)) {
2456 struct in_device *l3_in_dev;
2457
2458 l3_in_dev = __in_dev_get_rcu(skb->dev);
2459 if (l3_in_dev)
2460 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2461 ip_hdr(skb)->protocol);
2462 }
2463
2464 if (our
2465#ifdef CONFIG_IP_MROUTE
2466 ||
2467 (!ipv4_is_local_multicast(daddr) &&
2468 IN_DEV_MFORWARD(in_dev))
2469#endif
2470 ) {
2471 err = ip_route_input_mc(skb, daddr, saddr,
2472 tos, dev, our);
2473 }
2474 return err;
2475 }
2476
2477 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2478}
2479
2480int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2481 u8 tos, struct net_device *dev)
2482{
2483 struct fib_result res;
2484 int err;
2485
2486 tos &= IPTOS_RT_MASK;
2487 rcu_read_lock();
2488 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2489 rcu_read_unlock();
2490
2491 return err;
2492}
2493EXPORT_SYMBOL(ip_route_input_noref);
2494
2495/* called with rcu_read_lock() */
2496static struct rtable *__mkroute_output(const struct fib_result *res,
2497 const struct flowi4 *fl4, int orig_oif,
2498 struct net_device *dev_out,
2499 unsigned int flags)
2500{
2501 struct fib_info *fi = res->fi;
2502 struct fib_nh_exception *fnhe;
2503 struct in_device *in_dev;
2504 u16 type = res->type;
2505 struct rtable *rth;
2506 bool do_cache;
2507
2508 in_dev = __in_dev_get_rcu(dev_out);
2509 if (!in_dev)
2510 return ERR_PTR(-EINVAL);
2511
2512 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2513 if (ipv4_is_loopback(fl4->saddr) &&
2514 !(dev_out->flags & IFF_LOOPBACK) &&
2515 !netif_is_l3_master(dev_out))
2516 return ERR_PTR(-EINVAL);
2517
2518 if (ipv4_is_lbcast(fl4->daddr))
2519 type = RTN_BROADCAST;
2520 else if (ipv4_is_multicast(fl4->daddr))
2521 type = RTN_MULTICAST;
2522 else if (ipv4_is_zeronet(fl4->daddr))
2523 return ERR_PTR(-EINVAL);
2524
2525 if (dev_out->flags & IFF_LOOPBACK)
2526 flags |= RTCF_LOCAL;
2527
2528 do_cache = true;
2529 if (type == RTN_BROADCAST) {
2530 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2531 fi = NULL;
2532 } else if (type == RTN_MULTICAST) {
2533 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2534 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2535 fl4->flowi4_proto))
2536 flags &= ~RTCF_LOCAL;
2537 else
2538 do_cache = false;
2539 /* If multicast route do not exist use
2540 * default one, but do not gateway in this case.
2541 * Yes, it is hack.
2542 */
2543 if (fi && res->prefixlen < 4)
2544 fi = NULL;
2545 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2546 (orig_oif != dev_out->ifindex)) {
2547 /* For local routes that require a particular output interface
2548 * we do not want to cache the result. Caching the result
2549 * causes incorrect behaviour when there are multiple source
2550 * addresses on the interface, the end result being that if the
2551 * intended recipient is waiting on that interface for the
2552 * packet he won't receive it because it will be delivered on
2553 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2554 * be set to the loopback interface as well.
2555 */
2556 do_cache = false;
2557 }
2558
2559 fnhe = NULL;
2560 do_cache &= fi != NULL;
2561 if (fi) {
2562 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2563 struct rtable __rcu **prth;
2564
2565 fnhe = find_exception(nhc, fl4->daddr);
2566 if (!do_cache)
2567 goto add;
2568 if (fnhe) {
2569 prth = &fnhe->fnhe_rth_output;
2570 } else {
2571 if (unlikely(fl4->flowi4_flags &
2572 FLOWI_FLAG_KNOWN_NH &&
2573 !(nhc->nhc_gw_family &&
2574 nhc->nhc_scope == RT_SCOPE_LINK))) {
2575 do_cache = false;
2576 goto add;
2577 }
2578 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2579 }
2580 rth = rcu_dereference(*prth);
2581 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2582 return rth;
2583 }
2584
2585add:
2586 rth = rt_dst_alloc(dev_out, flags, type,
2587 IN_DEV_ORCONF(in_dev, NOXFRM));
2588 if (!rth)
2589 return ERR_PTR(-ENOBUFS);
2590
2591 rth->rt_iif = orig_oif;
2592
2593 RT_CACHE_STAT_INC(out_slow_tot);
2594
2595 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2596 if (flags & RTCF_LOCAL &&
2597 !(dev_out->flags & IFF_LOOPBACK)) {
2598 rth->dst.output = ip_mc_output;
2599 RT_CACHE_STAT_INC(out_slow_mc);
2600 }
2601#ifdef CONFIG_IP_MROUTE
2602 if (type == RTN_MULTICAST) {
2603 if (IN_DEV_MFORWARD(in_dev) &&
2604 !ipv4_is_local_multicast(fl4->daddr)) {
2605 rth->dst.input = ip_mr_input;
2606 rth->dst.output = ip_mc_output;
2607 }
2608 }
2609#endif
2610 }
2611
2612 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2613 lwtunnel_set_redirect(&rth->dst);
2614
2615 return rth;
2616}
2617
2618/*
2619 * Major route resolver routine.
2620 */
2621
2622struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2623 const struct sk_buff *skb)
2624{
2625 struct fib_result res = {
2626 .type = RTN_UNSPEC,
2627 .fi = NULL,
2628 .table = NULL,
2629 .tclassid = 0,
2630 };
2631 struct rtable *rth;
2632
2633 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2634 ip_rt_fix_tos(fl4);
2635
2636 rcu_read_lock();
2637 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2638 rcu_read_unlock();
2639
2640 return rth;
2641}
2642EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2643
2644struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2645 struct fib_result *res,
2646 const struct sk_buff *skb)
2647{
2648 struct net_device *dev_out = NULL;
2649 int orig_oif = fl4->flowi4_oif;
2650 unsigned int flags = 0;
2651 struct rtable *rth;
2652 int err;
2653
2654 if (fl4->saddr) {
2655 if (ipv4_is_multicast(fl4->saddr) ||
2656 ipv4_is_lbcast(fl4->saddr) ||
2657 ipv4_is_zeronet(fl4->saddr)) {
2658 rth = ERR_PTR(-EINVAL);
2659 goto out;
2660 }
2661
2662 rth = ERR_PTR(-ENETUNREACH);
2663
2664 /* I removed check for oif == dev_out->oif here.
2665 * It was wrong for two reasons:
2666 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2667 * is assigned to multiple interfaces.
2668 * 2. Moreover, we are allowed to send packets with saddr
2669 * of another iface. --ANK
2670 */
2671
2672 if (fl4->flowi4_oif == 0 &&
2673 (ipv4_is_multicast(fl4->daddr) ||
2674 ipv4_is_lbcast(fl4->daddr))) {
2675 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2676 dev_out = __ip_dev_find(net, fl4->saddr, false);
2677 if (!dev_out)
2678 goto out;
2679
2680 /* Special hack: user can direct multicasts
2681 * and limited broadcast via necessary interface
2682 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2683 * This hack is not just for fun, it allows
2684 * vic,vat and friends to work.
2685 * They bind socket to loopback, set ttl to zero
2686 * and expect that it will work.
2687 * From the viewpoint of routing cache they are broken,
2688 * because we are not allowed to build multicast path
2689 * with loopback source addr (look, routing cache
2690 * cannot know, that ttl is zero, so that packet
2691 * will not leave this host and route is valid).
2692 * Luckily, this hack is good workaround.
2693 */
2694
2695 fl4->flowi4_oif = dev_out->ifindex;
2696 goto make_route;
2697 }
2698
2699 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2700 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2701 if (!__ip_dev_find(net, fl4->saddr, false))
2702 goto out;
2703 }
2704 }
2705
2706
2707 if (fl4->flowi4_oif) {
2708 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2709 rth = ERR_PTR(-ENODEV);
2710 if (!dev_out)
2711 goto out;
2712
2713 /* RACE: Check return value of inet_select_addr instead. */
2714 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2715 rth = ERR_PTR(-ENETUNREACH);
2716 goto out;
2717 }
2718 if (ipv4_is_local_multicast(fl4->daddr) ||
2719 ipv4_is_lbcast(fl4->daddr) ||
2720 fl4->flowi4_proto == IPPROTO_IGMP) {
2721 if (!fl4->saddr)
2722 fl4->saddr = inet_select_addr(dev_out, 0,
2723 RT_SCOPE_LINK);
2724 goto make_route;
2725 }
2726 if (!fl4->saddr) {
2727 if (ipv4_is_multicast(fl4->daddr))
2728 fl4->saddr = inet_select_addr(dev_out, 0,
2729 fl4->flowi4_scope);
2730 else if (!fl4->daddr)
2731 fl4->saddr = inet_select_addr(dev_out, 0,
2732 RT_SCOPE_HOST);
2733 }
2734 }
2735
2736 if (!fl4->daddr) {
2737 fl4->daddr = fl4->saddr;
2738 if (!fl4->daddr)
2739 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2740 dev_out = net->loopback_dev;
2741 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2742 res->type = RTN_LOCAL;
2743 flags |= RTCF_LOCAL;
2744 goto make_route;
2745 }
2746
2747 err = fib_lookup(net, fl4, res, 0);
2748 if (err) {
2749 res->fi = NULL;
2750 res->table = NULL;
2751 if (fl4->flowi4_oif &&
2752 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2753 /* Apparently, routing tables are wrong. Assume,
2754 * that the destination is on link.
2755 *
2756 * WHY? DW.
2757 * Because we are allowed to send to iface
2758 * even if it has NO routes and NO assigned
2759 * addresses. When oif is specified, routing
2760 * tables are looked up with only one purpose:
2761 * to catch if destination is gatewayed, rather than
2762 * direct. Moreover, if MSG_DONTROUTE is set,
2763 * we send packet, ignoring both routing tables
2764 * and ifaddr state. --ANK
2765 *
2766 *
2767 * We could make it even if oif is unknown,
2768 * likely IPv6, but we do not.
2769 */
2770
2771 if (fl4->saddr == 0)
2772 fl4->saddr = inet_select_addr(dev_out, 0,
2773 RT_SCOPE_LINK);
2774 res->type = RTN_UNICAST;
2775 goto make_route;
2776 }
2777 rth = ERR_PTR(err);
2778 goto out;
2779 }
2780
2781 if (res->type == RTN_LOCAL) {
2782 if (!fl4->saddr) {
2783 if (res->fi->fib_prefsrc)
2784 fl4->saddr = res->fi->fib_prefsrc;
2785 else
2786 fl4->saddr = fl4->daddr;
2787 }
2788
2789 /* L3 master device is the loopback for that domain */
2790 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2791 net->loopback_dev;
2792
2793 /* make sure orig_oif points to fib result device even
2794 * though packet rx/tx happens over loopback or l3mdev
2795 */
2796 orig_oif = FIB_RES_OIF(*res);
2797
2798 fl4->flowi4_oif = dev_out->ifindex;
2799 flags |= RTCF_LOCAL;
2800 goto make_route;
2801 }
2802
2803 fib_select_path(net, res, fl4, skb);
2804
2805 dev_out = FIB_RES_DEV(*res);
2806
2807make_route:
2808 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2809
2810out:
2811 return rth;
2812}
2813
2814static struct dst_ops ipv4_dst_blackhole_ops = {
2815 .family = AF_INET,
2816 .default_advmss = ipv4_default_advmss,
2817 .neigh_lookup = ipv4_neigh_lookup,
2818 .check = dst_blackhole_check,
2819 .cow_metrics = dst_blackhole_cow_metrics,
2820 .update_pmtu = dst_blackhole_update_pmtu,
2821 .redirect = dst_blackhole_redirect,
2822 .mtu = dst_blackhole_mtu,
2823};
2824
2825struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2826{
2827 struct rtable *ort = dst_rtable(dst_orig);
2828 struct rtable *rt;
2829
2830 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2831 if (rt) {
2832 struct dst_entry *new = &rt->dst;
2833
2834 new->__use = 1;
2835 new->input = dst_discard;
2836 new->output = dst_discard_out;
2837
2838 new->dev = net->loopback_dev;
2839 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2840
2841 rt->rt_is_input = ort->rt_is_input;
2842 rt->rt_iif = ort->rt_iif;
2843 rt->rt_pmtu = ort->rt_pmtu;
2844 rt->rt_mtu_locked = ort->rt_mtu_locked;
2845
2846 rt->rt_genid = rt_genid_ipv4(net);
2847 rt->rt_flags = ort->rt_flags;
2848 rt->rt_type = ort->rt_type;
2849 rt->rt_uses_gateway = ort->rt_uses_gateway;
2850 rt->rt_gw_family = ort->rt_gw_family;
2851 if (rt->rt_gw_family == AF_INET)
2852 rt->rt_gw4 = ort->rt_gw4;
2853 else if (rt->rt_gw_family == AF_INET6)
2854 rt->rt_gw6 = ort->rt_gw6;
2855 }
2856
2857 dst_release(dst_orig);
2858
2859 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2860}
2861
2862struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2863 const struct sock *sk)
2864{
2865 struct rtable *rt = __ip_route_output_key(net, flp4);
2866
2867 if (IS_ERR(rt))
2868 return rt;
2869
2870 if (flp4->flowi4_proto) {
2871 flp4->flowi4_oif = rt->dst.dev->ifindex;
2872 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2873 flowi4_to_flowi(flp4),
2874 sk, 0));
2875 }
2876
2877 return rt;
2878}
2879EXPORT_SYMBOL_GPL(ip_route_output_flow);
2880
2881/* called with rcu_read_lock held */
2882static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2883 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2884 struct sk_buff *skb, u32 portid, u32 seq,
2885 unsigned int flags)
2886{
2887 struct rtmsg *r;
2888 struct nlmsghdr *nlh;
2889 unsigned long expires = 0;
2890 u32 error;
2891 u32 metrics[RTAX_MAX];
2892
2893 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2894 if (!nlh)
2895 return -EMSGSIZE;
2896
2897 r = nlmsg_data(nlh);
2898 r->rtm_family = AF_INET;
2899 r->rtm_dst_len = 32;
2900 r->rtm_src_len = 0;
2901 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2902 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2903 if (nla_put_u32(skb, RTA_TABLE, table_id))
2904 goto nla_put_failure;
2905 r->rtm_type = rt->rt_type;
2906 r->rtm_scope = RT_SCOPE_UNIVERSE;
2907 r->rtm_protocol = RTPROT_UNSPEC;
2908 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2909 if (rt->rt_flags & RTCF_NOTIFY)
2910 r->rtm_flags |= RTM_F_NOTIFY;
2911 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2912 r->rtm_flags |= RTCF_DOREDIRECT;
2913
2914 if (nla_put_in_addr(skb, RTA_DST, dst))
2915 goto nla_put_failure;
2916 if (src) {
2917 r->rtm_src_len = 32;
2918 if (nla_put_in_addr(skb, RTA_SRC, src))
2919 goto nla_put_failure;
2920 }
2921 if (rt->dst.dev &&
2922 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2923 goto nla_put_failure;
2924 if (rt->dst.lwtstate &&
2925 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2926 goto nla_put_failure;
2927#ifdef CONFIG_IP_ROUTE_CLASSID
2928 if (rt->dst.tclassid &&
2929 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2930 goto nla_put_failure;
2931#endif
2932 if (fl4 && !rt_is_input_route(rt) &&
2933 fl4->saddr != src) {
2934 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2935 goto nla_put_failure;
2936 }
2937 if (rt->rt_uses_gateway) {
2938 if (rt->rt_gw_family == AF_INET &&
2939 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2940 goto nla_put_failure;
2941 } else if (rt->rt_gw_family == AF_INET6) {
2942 int alen = sizeof(struct in6_addr);
2943 struct nlattr *nla;
2944 struct rtvia *via;
2945
2946 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2947 if (!nla)
2948 goto nla_put_failure;
2949
2950 via = nla_data(nla);
2951 via->rtvia_family = AF_INET6;
2952 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2953 }
2954 }
2955
2956 expires = rt->dst.expires;
2957 if (expires) {
2958 unsigned long now = jiffies;
2959
2960 if (time_before(now, expires))
2961 expires -= now;
2962 else
2963 expires = 0;
2964 }
2965
2966 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2967 if (rt->rt_pmtu && expires)
2968 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2969 if (rt->rt_mtu_locked && expires)
2970 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2971 if (rtnetlink_put_metrics(skb, metrics) < 0)
2972 goto nla_put_failure;
2973
2974 if (fl4) {
2975 if (fl4->flowi4_mark &&
2976 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2977 goto nla_put_failure;
2978
2979 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2980 nla_put_u32(skb, RTA_UID,
2981 from_kuid_munged(current_user_ns(),
2982 fl4->flowi4_uid)))
2983 goto nla_put_failure;
2984
2985 if (rt_is_input_route(rt)) {
2986#ifdef CONFIG_IP_MROUTE
2987 if (ipv4_is_multicast(dst) &&
2988 !ipv4_is_local_multicast(dst) &&
2989 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
2990 int err = ipmr_get_route(net, skb,
2991 fl4->saddr, fl4->daddr,
2992 r, portid);
2993
2994 if (err <= 0) {
2995 if (err == 0)
2996 return 0;
2997 goto nla_put_failure;
2998 }
2999 } else
3000#endif
3001 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3002 goto nla_put_failure;
3003 }
3004 }
3005
3006 error = rt->dst.error;
3007
3008 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3009 goto nla_put_failure;
3010
3011 nlmsg_end(skb, nlh);
3012 return 0;
3013
3014nla_put_failure:
3015 nlmsg_cancel(skb, nlh);
3016 return -EMSGSIZE;
3017}
3018
3019static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3020 struct netlink_callback *cb, u32 table_id,
3021 struct fnhe_hash_bucket *bucket, int genid,
3022 int *fa_index, int fa_start, unsigned int flags)
3023{
3024 int i;
3025
3026 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3027 struct fib_nh_exception *fnhe;
3028
3029 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3030 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3031 struct rtable *rt;
3032 int err;
3033
3034 if (*fa_index < fa_start)
3035 goto next;
3036
3037 if (fnhe->fnhe_genid != genid)
3038 goto next;
3039
3040 if (fnhe->fnhe_expires &&
3041 time_after(jiffies, fnhe->fnhe_expires))
3042 goto next;
3043
3044 rt = rcu_dereference(fnhe->fnhe_rth_input);
3045 if (!rt)
3046 rt = rcu_dereference(fnhe->fnhe_rth_output);
3047 if (!rt)
3048 goto next;
3049
3050 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3051 table_id, NULL, skb,
3052 NETLINK_CB(cb->skb).portid,
3053 cb->nlh->nlmsg_seq, flags);
3054 if (err)
3055 return err;
3056next:
3057 (*fa_index)++;
3058 }
3059 }
3060
3061 return 0;
3062}
3063
3064int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3065 u32 table_id, struct fib_info *fi,
3066 int *fa_index, int fa_start, unsigned int flags)
3067{
3068 struct net *net = sock_net(cb->skb->sk);
3069 int nhsel, genid = fnhe_genid(net);
3070
3071 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3072 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3073 struct fnhe_hash_bucket *bucket;
3074 int err;
3075
3076 if (nhc->nhc_flags & RTNH_F_DEAD)
3077 continue;
3078
3079 rcu_read_lock();
3080 bucket = rcu_dereference(nhc->nhc_exceptions);
3081 err = 0;
3082 if (bucket)
3083 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3084 genid, fa_index, fa_start,
3085 flags);
3086 rcu_read_unlock();
3087 if (err)
3088 return err;
3089 }
3090
3091 return 0;
3092}
3093
3094static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3095 u8 ip_proto, __be16 sport,
3096 __be16 dport)
3097{
3098 struct sk_buff *skb;
3099 struct iphdr *iph;
3100
3101 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3102 if (!skb)
3103 return NULL;
3104
3105 /* Reserve room for dummy headers, this skb can pass
3106 * through good chunk of routing engine.
3107 */
3108 skb_reset_mac_header(skb);
3109 skb_reset_network_header(skb);
3110 skb->protocol = htons(ETH_P_IP);
3111 iph = skb_put(skb, sizeof(struct iphdr));
3112 iph->protocol = ip_proto;
3113 iph->saddr = src;
3114 iph->daddr = dst;
3115 iph->version = 0x4;
3116 iph->frag_off = 0;
3117 iph->ihl = 0x5;
3118 skb_set_transport_header(skb, skb->len);
3119
3120 switch (iph->protocol) {
3121 case IPPROTO_UDP: {
3122 struct udphdr *udph;
3123
3124 udph = skb_put_zero(skb, sizeof(struct udphdr));
3125 udph->source = sport;
3126 udph->dest = dport;
3127 udph->len = htons(sizeof(struct udphdr));
3128 udph->check = 0;
3129 break;
3130 }
3131 case IPPROTO_TCP: {
3132 struct tcphdr *tcph;
3133
3134 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3135 tcph->source = sport;
3136 tcph->dest = dport;
3137 tcph->doff = sizeof(struct tcphdr) / 4;
3138 tcph->rst = 1;
3139 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3140 src, dst, 0);
3141 break;
3142 }
3143 case IPPROTO_ICMP: {
3144 struct icmphdr *icmph;
3145
3146 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3147 icmph->type = ICMP_ECHO;
3148 icmph->code = 0;
3149 }
3150 }
3151
3152 return skb;
3153}
3154
3155static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3156 const struct nlmsghdr *nlh,
3157 struct nlattr **tb,
3158 struct netlink_ext_ack *extack)
3159{
3160 struct rtmsg *rtm;
3161 int i, err;
3162
3163 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3164 NL_SET_ERR_MSG(extack,
3165 "ipv4: Invalid header for route get request");
3166 return -EINVAL;
3167 }
3168
3169 if (!netlink_strict_get_check(skb))
3170 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3171 rtm_ipv4_policy, extack);
3172
3173 rtm = nlmsg_data(nlh);
3174 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3175 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3176 rtm->rtm_table || rtm->rtm_protocol ||
3177 rtm->rtm_scope || rtm->rtm_type) {
3178 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3179 return -EINVAL;
3180 }
3181
3182 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3183 RTM_F_LOOKUP_TABLE |
3184 RTM_F_FIB_MATCH)) {
3185 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3186 return -EINVAL;
3187 }
3188
3189 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3190 rtm_ipv4_policy, extack);
3191 if (err)
3192 return err;
3193
3194 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3195 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3196 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3197 return -EINVAL;
3198 }
3199
3200 for (i = 0; i <= RTA_MAX; i++) {
3201 if (!tb[i])
3202 continue;
3203
3204 switch (i) {
3205 case RTA_IIF:
3206 case RTA_OIF:
3207 case RTA_SRC:
3208 case RTA_DST:
3209 case RTA_IP_PROTO:
3210 case RTA_SPORT:
3211 case RTA_DPORT:
3212 case RTA_MARK:
3213 case RTA_UID:
3214 break;
3215 default:
3216 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3217 return -EINVAL;
3218 }
3219 }
3220
3221 return 0;
3222}
3223
3224static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3225 struct netlink_ext_ack *extack)
3226{
3227 struct net *net = sock_net(in_skb->sk);
3228 struct nlattr *tb[RTA_MAX+1];
3229 u32 table_id = RT_TABLE_MAIN;
3230 __be16 sport = 0, dport = 0;
3231 struct fib_result res = {};
3232 u8 ip_proto = IPPROTO_UDP;
3233 struct rtable *rt = NULL;
3234 struct sk_buff *skb;
3235 struct rtmsg *rtm;
3236 struct flowi4 fl4 = {};
3237 __be32 dst = 0;
3238 __be32 src = 0;
3239 kuid_t uid;
3240 u32 iif;
3241 int err;
3242 int mark;
3243
3244 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3245 if (err < 0)
3246 return err;
3247
3248 rtm = nlmsg_data(nlh);
3249 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3250 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3251 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3252 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3253 if (tb[RTA_UID])
3254 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3255 else
3256 uid = (iif ? INVALID_UID : current_uid());
3257
3258 if (tb[RTA_IP_PROTO]) {
3259 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3260 &ip_proto, AF_INET, extack);
3261 if (err)
3262 return err;
3263 }
3264
3265 if (tb[RTA_SPORT])
3266 sport = nla_get_be16(tb[RTA_SPORT]);
3267
3268 if (tb[RTA_DPORT])
3269 dport = nla_get_be16(tb[RTA_DPORT]);
3270
3271 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3272 if (!skb)
3273 return -ENOBUFS;
3274
3275 fl4.daddr = dst;
3276 fl4.saddr = src;
3277 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3278 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3279 fl4.flowi4_mark = mark;
3280 fl4.flowi4_uid = uid;
3281 if (sport)
3282 fl4.fl4_sport = sport;
3283 if (dport)
3284 fl4.fl4_dport = dport;
3285 fl4.flowi4_proto = ip_proto;
3286
3287 rcu_read_lock();
3288
3289 if (iif) {
3290 struct net_device *dev;
3291
3292 dev = dev_get_by_index_rcu(net, iif);
3293 if (!dev) {
3294 err = -ENODEV;
3295 goto errout_rcu;
3296 }
3297
3298 fl4.flowi4_iif = iif; /* for rt_fill_info */
3299 skb->dev = dev;
3300 skb->mark = mark;
3301 err = ip_route_input_rcu(skb, dst, src,
3302 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3303 &res);
3304
3305 rt = skb_rtable(skb);
3306 if (err == 0 && rt->dst.error)
3307 err = -rt->dst.error;
3308 } else {
3309 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3310 skb->dev = net->loopback_dev;
3311 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3312 err = 0;
3313 if (IS_ERR(rt))
3314 err = PTR_ERR(rt);
3315 else
3316 skb_dst_set(skb, &rt->dst);
3317 }
3318
3319 if (err)
3320 goto errout_rcu;
3321
3322 if (rtm->rtm_flags & RTM_F_NOTIFY)
3323 rt->rt_flags |= RTCF_NOTIFY;
3324
3325 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3326 table_id = res.table ? res.table->tb_id : 0;
3327
3328 /* reset skb for netlink reply msg */
3329 skb_trim(skb, 0);
3330 skb_reset_network_header(skb);
3331 skb_reset_transport_header(skb);
3332 skb_reset_mac_header(skb);
3333
3334 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3335 struct fib_rt_info fri;
3336
3337 if (!res.fi) {
3338 err = fib_props[res.type].error;
3339 if (!err)
3340 err = -EHOSTUNREACH;
3341 goto errout_rcu;
3342 }
3343 fri.fi = res.fi;
3344 fri.tb_id = table_id;
3345 fri.dst = res.prefix;
3346 fri.dst_len = res.prefixlen;
3347 fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3348 fri.type = rt->rt_type;
3349 fri.offload = 0;
3350 fri.trap = 0;
3351 fri.offload_failed = 0;
3352 if (res.fa_head) {
3353 struct fib_alias *fa;
3354
3355 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3356 u8 slen = 32 - fri.dst_len;
3357
3358 if (fa->fa_slen == slen &&
3359 fa->tb_id == fri.tb_id &&
3360 fa->fa_dscp == fri.dscp &&
3361 fa->fa_info == res.fi &&
3362 fa->fa_type == fri.type) {
3363 fri.offload = READ_ONCE(fa->offload);
3364 fri.trap = READ_ONCE(fa->trap);
3365 fri.offload_failed =
3366 READ_ONCE(fa->offload_failed);
3367 break;
3368 }
3369 }
3370 }
3371 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3372 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3373 } else {
3374 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3375 NETLINK_CB(in_skb).portid,
3376 nlh->nlmsg_seq, 0);
3377 }
3378 if (err < 0)
3379 goto errout_rcu;
3380
3381 rcu_read_unlock();
3382
3383 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3384
3385errout_free:
3386 return err;
3387errout_rcu:
3388 rcu_read_unlock();
3389 kfree_skb(skb);
3390 goto errout_free;
3391}
3392
3393void ip_rt_multicast_event(struct in_device *in_dev)
3394{
3395 rt_cache_flush(dev_net(in_dev->dev));
3396}
3397
3398#ifdef CONFIG_SYSCTL
3399static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3400static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3401static int ip_rt_gc_elasticity __read_mostly = 8;
3402static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3403
3404static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3405 void *buffer, size_t *lenp, loff_t *ppos)
3406{
3407 struct net *net = (struct net *)__ctl->extra1;
3408
3409 if (write) {
3410 rt_cache_flush(net);
3411 fnhe_genid_bump(net);
3412 return 0;
3413 }
3414
3415 return -EINVAL;
3416}
3417
3418static struct ctl_table ipv4_route_table[] = {
3419 {
3420 .procname = "gc_thresh",
3421 .data = &ipv4_dst_ops.gc_thresh,
3422 .maxlen = sizeof(int),
3423 .mode = 0644,
3424 .proc_handler = proc_dointvec,
3425 },
3426 {
3427 .procname = "max_size",
3428 .data = &ip_rt_max_size,
3429 .maxlen = sizeof(int),
3430 .mode = 0644,
3431 .proc_handler = proc_dointvec,
3432 },
3433 {
3434 /* Deprecated. Use gc_min_interval_ms */
3435
3436 .procname = "gc_min_interval",
3437 .data = &ip_rt_gc_min_interval,
3438 .maxlen = sizeof(int),
3439 .mode = 0644,
3440 .proc_handler = proc_dointvec_jiffies,
3441 },
3442 {
3443 .procname = "gc_min_interval_ms",
3444 .data = &ip_rt_gc_min_interval,
3445 .maxlen = sizeof(int),
3446 .mode = 0644,
3447 .proc_handler = proc_dointvec_ms_jiffies,
3448 },
3449 {
3450 .procname = "gc_timeout",
3451 .data = &ip_rt_gc_timeout,
3452 .maxlen = sizeof(int),
3453 .mode = 0644,
3454 .proc_handler = proc_dointvec_jiffies,
3455 },
3456 {
3457 .procname = "gc_interval",
3458 .data = &ip_rt_gc_interval,
3459 .maxlen = sizeof(int),
3460 .mode = 0644,
3461 .proc_handler = proc_dointvec_jiffies,
3462 },
3463 {
3464 .procname = "redirect_load",
3465 .data = &ip_rt_redirect_load,
3466 .maxlen = sizeof(int),
3467 .mode = 0644,
3468 .proc_handler = proc_dointvec,
3469 },
3470 {
3471 .procname = "redirect_number",
3472 .data = &ip_rt_redirect_number,
3473 .maxlen = sizeof(int),
3474 .mode = 0644,
3475 .proc_handler = proc_dointvec,
3476 },
3477 {
3478 .procname = "redirect_silence",
3479 .data = &ip_rt_redirect_silence,
3480 .maxlen = sizeof(int),
3481 .mode = 0644,
3482 .proc_handler = proc_dointvec,
3483 },
3484 {
3485 .procname = "error_cost",
3486 .data = &ip_rt_error_cost,
3487 .maxlen = sizeof(int),
3488 .mode = 0644,
3489 .proc_handler = proc_dointvec,
3490 },
3491 {
3492 .procname = "error_burst",
3493 .data = &ip_rt_error_burst,
3494 .maxlen = sizeof(int),
3495 .mode = 0644,
3496 .proc_handler = proc_dointvec,
3497 },
3498 {
3499 .procname = "gc_elasticity",
3500 .data = &ip_rt_gc_elasticity,
3501 .maxlen = sizeof(int),
3502 .mode = 0644,
3503 .proc_handler = proc_dointvec,
3504 },
3505 { }
3506};
3507
3508static const char ipv4_route_flush_procname[] = "flush";
3509
3510static struct ctl_table ipv4_route_netns_table[] = {
3511 {
3512 .procname = ipv4_route_flush_procname,
3513 .maxlen = sizeof(int),
3514 .mode = 0200,
3515 .proc_handler = ipv4_sysctl_rtcache_flush,
3516 },
3517 {
3518 .procname = "min_pmtu",
3519 .data = &init_net.ipv4.ip_rt_min_pmtu,
3520 .maxlen = sizeof(int),
3521 .mode = 0644,
3522 .proc_handler = proc_dointvec_minmax,
3523 .extra1 = &ip_min_valid_pmtu,
3524 },
3525 {
3526 .procname = "mtu_expires",
3527 .data = &init_net.ipv4.ip_rt_mtu_expires,
3528 .maxlen = sizeof(int),
3529 .mode = 0644,
3530 .proc_handler = proc_dointvec_jiffies,
3531 },
3532 {
3533 .procname = "min_adv_mss",
3534 .data = &init_net.ipv4.ip_rt_min_advmss,
3535 .maxlen = sizeof(int),
3536 .mode = 0644,
3537 .proc_handler = proc_dointvec,
3538 },
3539 { },
3540};
3541
3542static __net_init int sysctl_route_net_init(struct net *net)
3543{
3544 struct ctl_table *tbl;
3545 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3546
3547 tbl = ipv4_route_netns_table;
3548 if (!net_eq(net, &init_net)) {
3549 int i;
3550
3551 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3552 if (!tbl)
3553 goto err_dup;
3554
3555 /* Don't export non-whitelisted sysctls to unprivileged users */
3556 if (net->user_ns != &init_user_ns) {
3557 if (tbl[0].procname != ipv4_route_flush_procname) {
3558 tbl[0].procname = NULL;
3559 table_size = 0;
3560 }
3561 }
3562
3563 /* Update the variables to point into the current struct net
3564 * except for the first element flush
3565 */
3566 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3567 tbl[i].data += (void *)net - (void *)&init_net;
3568 }
3569 tbl[0].extra1 = net;
3570
3571 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3572 tbl, table_size);
3573 if (!net->ipv4.route_hdr)
3574 goto err_reg;
3575 return 0;
3576
3577err_reg:
3578 if (tbl != ipv4_route_netns_table)
3579 kfree(tbl);
3580err_dup:
3581 return -ENOMEM;
3582}
3583
3584static __net_exit void sysctl_route_net_exit(struct net *net)
3585{
3586 struct ctl_table *tbl;
3587
3588 tbl = net->ipv4.route_hdr->ctl_table_arg;
3589 unregister_net_sysctl_table(net->ipv4.route_hdr);
3590 BUG_ON(tbl == ipv4_route_netns_table);
3591 kfree(tbl);
3592}
3593
3594static __net_initdata struct pernet_operations sysctl_route_ops = {
3595 .init = sysctl_route_net_init,
3596 .exit = sysctl_route_net_exit,
3597};
3598#endif
3599
3600static __net_init int netns_ip_rt_init(struct net *net)
3601{
3602 /* Set default value for namespaceified sysctls */
3603 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3604 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3605 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3606 return 0;
3607}
3608
3609static struct pernet_operations __net_initdata ip_rt_ops = {
3610 .init = netns_ip_rt_init,
3611};
3612
3613static __net_init int rt_genid_init(struct net *net)
3614{
3615 atomic_set(&net->ipv4.rt_genid, 0);
3616 atomic_set(&net->fnhe_genid, 0);
3617 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3618 return 0;
3619}
3620
3621static __net_initdata struct pernet_operations rt_genid_ops = {
3622 .init = rt_genid_init,
3623};
3624
3625static int __net_init ipv4_inetpeer_init(struct net *net)
3626{
3627 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3628
3629 if (!bp)
3630 return -ENOMEM;
3631 inet_peer_base_init(bp);
3632 net->ipv4.peers = bp;
3633 return 0;
3634}
3635
3636static void __net_exit ipv4_inetpeer_exit(struct net *net)
3637{
3638 struct inet_peer_base *bp = net->ipv4.peers;
3639
3640 net->ipv4.peers = NULL;
3641 inetpeer_invalidate_tree(bp);
3642 kfree(bp);
3643}
3644
3645static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3646 .init = ipv4_inetpeer_init,
3647 .exit = ipv4_inetpeer_exit,
3648};
3649
3650#ifdef CONFIG_IP_ROUTE_CLASSID
3651struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3652#endif /* CONFIG_IP_ROUTE_CLASSID */
3653
3654int __init ip_rt_init(void)
3655{
3656 void *idents_hash;
3657 int cpu;
3658
3659 /* For modern hosts, this will use 2 MB of memory */
3660 idents_hash = alloc_large_system_hash("IP idents",
3661 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3662 0,
3663 16, /* one bucket per 64 KB */
3664 HASH_ZERO,
3665 NULL,
3666 &ip_idents_mask,
3667 2048,
3668 256*1024);
3669
3670 ip_idents = idents_hash;
3671
3672 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3673
3674 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3675
3676 for_each_possible_cpu(cpu) {
3677 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3678
3679 INIT_LIST_HEAD(&ul->head);
3680 INIT_LIST_HEAD(&ul->quarantine);
3681 spin_lock_init(&ul->lock);
3682 }
3683#ifdef CONFIG_IP_ROUTE_CLASSID
3684 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3685 if (!ip_rt_acct)
3686 panic("IP: failed to allocate ip_rt_acct\n");
3687#endif
3688
3689 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3690 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3691
3692 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3693
3694 if (dst_entries_init(&ipv4_dst_ops) < 0)
3695 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3696
3697 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3698 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3699
3700 ipv4_dst_ops.gc_thresh = ~0;
3701 ip_rt_max_size = INT_MAX;
3702
3703 devinet_init();
3704 ip_fib_init();
3705
3706 if (ip_rt_proc_init())
3707 pr_err("Unable to create route proc files\n");
3708#ifdef CONFIG_XFRM
3709 xfrm_init();
3710 xfrm4_init();
3711#endif
3712 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3713 RTNL_FLAG_DOIT_UNLOCKED);
3714
3715#ifdef CONFIG_SYSCTL
3716 register_pernet_subsys(&sysctl_route_ops);
3717#endif
3718 register_pernet_subsys(&ip_rt_ops);
3719 register_pernet_subsys(&rt_genid_ops);
3720 register_pernet_subsys(&ipv4_inetpeer_ops);
3721 return 0;
3722}
3723
3724#ifdef CONFIG_SYSCTL
3725/*
3726 * We really need to sanitize the damn ipv4 init order, then all
3727 * this nonsense will go away.
3728 */
3729void __init ip_static_sysctl_init(void)
3730{
3731 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3732}
3733#endif