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/string.h>
70#include <linux/socket.h>
71#include <linux/sockios.h>
72#include <linux/errno.h>
73#include <linux/in.h>
74#include <linux/inet.h>
75#include <linux/netdevice.h>
76#include <linux/proc_fs.h>
77#include <linux/init.h>
78#include <linux/skbuff.h>
79#include <linux/inetdevice.h>
80#include <linux/igmp.h>
81#include <linux/pkt_sched.h>
82#include <linux/mroute.h>
83#include <linux/netfilter_ipv4.h>
84#include <linux/random.h>
85#include <linux/rcupdate.h>
86#include <linux/times.h>
87#include <linux/slab.h>
88#include <linux/jhash.h>
89#include <net/dst.h>
90#include <net/dst_metadata.h>
91#include <net/net_namespace.h>
92#include <net/protocol.h>
93#include <net/ip.h>
94#include <net/route.h>
95#include <net/inetpeer.h>
96#include <net/sock.h>
97#include <net/ip_fib.h>
98#include <net/nexthop.h>
99#include <net/arp.h>
100#include <net/tcp.h>
101#include <net/icmp.h>
102#include <net/xfrm.h>
103#include <net/lwtunnel.h>
104#include <net/netevent.h>
105#include <net/rtnetlink.h>
106#ifdef CONFIG_SYSCTL
107#include <linux/sysctl.h>
108#endif
109#include <net/secure_seq.h>
110#include <net/ip_tunnels.h>
111#include <net/l3mdev.h>
112
113#include "fib_lookup.h"
114
115#define RT_FL_TOS(oldflp4) \
116 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
117
118#define RT_GC_TIMEOUT (300*HZ)
119
120static int ip_rt_max_size;
121static int ip_rt_redirect_number __read_mostly = 9;
122static int ip_rt_redirect_load __read_mostly = HZ / 50;
123static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
124static int ip_rt_error_cost __read_mostly = HZ;
125static int ip_rt_error_burst __read_mostly = 5 * HZ;
126static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
127static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
128static int ip_rt_min_advmss __read_mostly = 256;
129
130static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
131
132/*
133 * Interface to generic destination cache.
134 */
135
136static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
137static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
138static unsigned int ipv4_mtu(const struct dst_entry *dst);
139static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
140static void ipv4_link_failure(struct sk_buff *skb);
141static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
142 struct sk_buff *skb, u32 mtu);
143static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
144 struct sk_buff *skb);
145static void ipv4_dst_destroy(struct dst_entry *dst);
146
147static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
148{
149 WARN_ON(1);
150 return NULL;
151}
152
153static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
154 struct sk_buff *skb,
155 const void *daddr);
156static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
157
158static struct dst_ops ipv4_dst_ops = {
159 .family = AF_INET,
160 .check = ipv4_dst_check,
161 .default_advmss = ipv4_default_advmss,
162 .mtu = ipv4_mtu,
163 .cow_metrics = ipv4_cow_metrics,
164 .destroy = ipv4_dst_destroy,
165 .negative_advice = ipv4_negative_advice,
166 .link_failure = ipv4_link_failure,
167 .update_pmtu = ip_rt_update_pmtu,
168 .redirect = ip_do_redirect,
169 .local_out = __ip_local_out,
170 .neigh_lookup = ipv4_neigh_lookup,
171 .confirm_neigh = ipv4_confirm_neigh,
172};
173
174#define ECN_OR_COST(class) TC_PRIO_##class
175
176const __u8 ip_tos2prio[16] = {
177 TC_PRIO_BESTEFFORT,
178 ECN_OR_COST(BESTEFFORT),
179 TC_PRIO_BESTEFFORT,
180 ECN_OR_COST(BESTEFFORT),
181 TC_PRIO_BULK,
182 ECN_OR_COST(BULK),
183 TC_PRIO_BULK,
184 ECN_OR_COST(BULK),
185 TC_PRIO_INTERACTIVE,
186 ECN_OR_COST(INTERACTIVE),
187 TC_PRIO_INTERACTIVE,
188 ECN_OR_COST(INTERACTIVE),
189 TC_PRIO_INTERACTIVE_BULK,
190 ECN_OR_COST(INTERACTIVE_BULK),
191 TC_PRIO_INTERACTIVE_BULK,
192 ECN_OR_COST(INTERACTIVE_BULK)
193};
194EXPORT_SYMBOL(ip_tos2prio);
195
196static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
197#define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
198
199#ifdef CONFIG_PROC_FS
200static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
201{
202 if (*pos)
203 return NULL;
204 return SEQ_START_TOKEN;
205}
206
207static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
208{
209 ++*pos;
210 return NULL;
211}
212
213static void rt_cache_seq_stop(struct seq_file *seq, void *v)
214{
215}
216
217static int rt_cache_seq_show(struct seq_file *seq, void *v)
218{
219 if (v == SEQ_START_TOKEN)
220 seq_printf(seq, "%-127s\n",
221 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
222 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
223 "HHUptod\tSpecDst");
224 return 0;
225}
226
227static const struct seq_operations rt_cache_seq_ops = {
228 .start = rt_cache_seq_start,
229 .next = rt_cache_seq_next,
230 .stop = rt_cache_seq_stop,
231 .show = rt_cache_seq_show,
232};
233
234static int rt_cache_seq_open(struct inode *inode, struct file *file)
235{
236 return seq_open(file, &rt_cache_seq_ops);
237}
238
239static const struct file_operations rt_cache_seq_fops = {
240 .open = rt_cache_seq_open,
241 .read = seq_read,
242 .llseek = seq_lseek,
243 .release = seq_release,
244};
245
246
247static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
248{
249 int cpu;
250
251 if (*pos == 0)
252 return SEQ_START_TOKEN;
253
254 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
255 if (!cpu_possible(cpu))
256 continue;
257 *pos = cpu+1;
258 return &per_cpu(rt_cache_stat, cpu);
259 }
260 return NULL;
261}
262
263static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
264{
265 int cpu;
266
267 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
268 if (!cpu_possible(cpu))
269 continue;
270 *pos = cpu+1;
271 return &per_cpu(rt_cache_stat, cpu);
272 }
273 return NULL;
274
275}
276
277static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
278{
279
280}
281
282static int rt_cpu_seq_show(struct seq_file *seq, void *v)
283{
284 struct rt_cache_stat *st = v;
285
286 if (v == SEQ_START_TOKEN) {
287 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");
288 return 0;
289 }
290
291 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
292 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
293 dst_entries_get_slow(&ipv4_dst_ops),
294 0, /* st->in_hit */
295 st->in_slow_tot,
296 st->in_slow_mc,
297 st->in_no_route,
298 st->in_brd,
299 st->in_martian_dst,
300 st->in_martian_src,
301
302 0, /* st->out_hit */
303 st->out_slow_tot,
304 st->out_slow_mc,
305
306 0, /* st->gc_total */
307 0, /* st->gc_ignored */
308 0, /* st->gc_goal_miss */
309 0, /* st->gc_dst_overflow */
310 0, /* st->in_hlist_search */
311 0 /* st->out_hlist_search */
312 );
313 return 0;
314}
315
316static const struct seq_operations rt_cpu_seq_ops = {
317 .start = rt_cpu_seq_start,
318 .next = rt_cpu_seq_next,
319 .stop = rt_cpu_seq_stop,
320 .show = rt_cpu_seq_show,
321};
322
323
324static int rt_cpu_seq_open(struct inode *inode, struct file *file)
325{
326 return seq_open(file, &rt_cpu_seq_ops);
327}
328
329static const struct file_operations rt_cpu_seq_fops = {
330 .open = rt_cpu_seq_open,
331 .read = seq_read,
332 .llseek = seq_lseek,
333 .release = seq_release,
334};
335
336#ifdef CONFIG_IP_ROUTE_CLASSID
337static int rt_acct_proc_show(struct seq_file *m, void *v)
338{
339 struct ip_rt_acct *dst, *src;
340 unsigned int i, j;
341
342 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
343 if (!dst)
344 return -ENOMEM;
345
346 for_each_possible_cpu(i) {
347 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
348 for (j = 0; j < 256; j++) {
349 dst[j].o_bytes += src[j].o_bytes;
350 dst[j].o_packets += src[j].o_packets;
351 dst[j].i_bytes += src[j].i_bytes;
352 dst[j].i_packets += src[j].i_packets;
353 }
354 }
355
356 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
357 kfree(dst);
358 return 0;
359}
360#endif
361
362static int __net_init ip_rt_do_proc_init(struct net *net)
363{
364 struct proc_dir_entry *pde;
365
366 pde = proc_create("rt_cache", 0444, net->proc_net,
367 &rt_cache_seq_fops);
368 if (!pde)
369 goto err1;
370
371 pde = proc_create("rt_cache", 0444,
372 net->proc_net_stat, &rt_cpu_seq_fops);
373 if (!pde)
374 goto err2;
375
376#ifdef CONFIG_IP_ROUTE_CLASSID
377 pde = proc_create_single("rt_acct", 0, net->proc_net,
378 rt_acct_proc_show);
379 if (!pde)
380 goto err3;
381#endif
382 return 0;
383
384#ifdef CONFIG_IP_ROUTE_CLASSID
385err3:
386 remove_proc_entry("rt_cache", net->proc_net_stat);
387#endif
388err2:
389 remove_proc_entry("rt_cache", net->proc_net);
390err1:
391 return -ENOMEM;
392}
393
394static void __net_exit ip_rt_do_proc_exit(struct net *net)
395{
396 remove_proc_entry("rt_cache", net->proc_net_stat);
397 remove_proc_entry("rt_cache", net->proc_net);
398#ifdef CONFIG_IP_ROUTE_CLASSID
399 remove_proc_entry("rt_acct", net->proc_net);
400#endif
401}
402
403static struct pernet_operations ip_rt_proc_ops __net_initdata = {
404 .init = ip_rt_do_proc_init,
405 .exit = ip_rt_do_proc_exit,
406};
407
408static int __init ip_rt_proc_init(void)
409{
410 return register_pernet_subsys(&ip_rt_proc_ops);
411}
412
413#else
414static inline int ip_rt_proc_init(void)
415{
416 return 0;
417}
418#endif /* CONFIG_PROC_FS */
419
420static inline bool rt_is_expired(const struct rtable *rth)
421{
422 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
423}
424
425void rt_cache_flush(struct net *net)
426{
427 rt_genid_bump_ipv4(net);
428}
429
430static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
431 struct sk_buff *skb,
432 const void *daddr)
433{
434 const struct rtable *rt = container_of(dst, struct rtable, dst);
435 struct net_device *dev = dst->dev;
436 struct neighbour *n;
437
438 rcu_read_lock_bh();
439
440 if (likely(rt->rt_gw_family == AF_INET)) {
441 n = ip_neigh_gw4(dev, rt->rt_gw4);
442 } else if (rt->rt_gw_family == AF_INET6) {
443 n = ip_neigh_gw6(dev, &rt->rt_gw6);
444 } else {
445 __be32 pkey;
446
447 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
448 n = ip_neigh_gw4(dev, pkey);
449 }
450
451 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
452 n = NULL;
453
454 rcu_read_unlock_bh();
455
456 return n;
457}
458
459static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
460{
461 const struct rtable *rt = container_of(dst, struct rtable, dst);
462 struct net_device *dev = dst->dev;
463 const __be32 *pkey = daddr;
464
465 if (rt->rt_gw_family == AF_INET) {
466 pkey = (const __be32 *)&rt->rt_gw4;
467 } else if (rt->rt_gw_family == AF_INET6) {
468 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
469 } else if (!daddr ||
470 (rt->rt_flags &
471 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
472 return;
473 }
474 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
475}
476
477#define IP_IDENTS_SZ 2048u
478
479static atomic_t *ip_idents __read_mostly;
480static u32 *ip_tstamps __read_mostly;
481
482/* In order to protect privacy, we add a perturbation to identifiers
483 * if one generator is seldom used. This makes hard for an attacker
484 * to infer how many packets were sent between two points in time.
485 */
486u32 ip_idents_reserve(u32 hash, int segs)
487{
488 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
489 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
490 u32 old = READ_ONCE(*p_tstamp);
491 u32 now = (u32)jiffies;
492 u32 new, delta = 0;
493
494 if (old != now && cmpxchg(p_tstamp, old, now) == old)
495 delta = prandom_u32_max(now - old);
496
497 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
498 do {
499 old = (u32)atomic_read(p_id);
500 new = old + delta + segs;
501 } while (atomic_cmpxchg(p_id, old, new) != old);
502
503 return new - segs;
504}
505EXPORT_SYMBOL(ip_idents_reserve);
506
507void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
508{
509 u32 hash, id;
510
511 /* Note the following code is not safe, but this is okay. */
512 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
513 get_random_bytes(&net->ipv4.ip_id_key,
514 sizeof(net->ipv4.ip_id_key));
515
516 hash = siphash_3u32((__force u32)iph->daddr,
517 (__force u32)iph->saddr,
518 iph->protocol,
519 &net->ipv4.ip_id_key);
520 id = ip_idents_reserve(hash, segs);
521 iph->id = htons(id);
522}
523EXPORT_SYMBOL(__ip_select_ident);
524
525static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
526 const struct sock *sk,
527 const struct iphdr *iph,
528 int oif, u8 tos,
529 u8 prot, u32 mark, int flow_flags)
530{
531 if (sk) {
532 const struct inet_sock *inet = inet_sk(sk);
533
534 oif = sk->sk_bound_dev_if;
535 mark = sk->sk_mark;
536 tos = RT_CONN_FLAGS(sk);
537 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
538 }
539 flowi4_init_output(fl4, oif, mark, tos,
540 RT_SCOPE_UNIVERSE, prot,
541 flow_flags,
542 iph->daddr, iph->saddr, 0, 0,
543 sock_net_uid(net, sk));
544}
545
546static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
547 const struct sock *sk)
548{
549 const struct net *net = dev_net(skb->dev);
550 const struct iphdr *iph = ip_hdr(skb);
551 int oif = skb->dev->ifindex;
552 u8 tos = RT_TOS(iph->tos);
553 u8 prot = iph->protocol;
554 u32 mark = skb->mark;
555
556 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
557}
558
559static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
560{
561 const struct inet_sock *inet = inet_sk(sk);
562 const struct ip_options_rcu *inet_opt;
563 __be32 daddr = inet->inet_daddr;
564
565 rcu_read_lock();
566 inet_opt = rcu_dereference(inet->inet_opt);
567 if (inet_opt && inet_opt->opt.srr)
568 daddr = inet_opt->opt.faddr;
569 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
570 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
571 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
572 inet_sk_flowi_flags(sk),
573 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
574 rcu_read_unlock();
575}
576
577static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
578 const struct sk_buff *skb)
579{
580 if (skb)
581 build_skb_flow_key(fl4, skb, sk);
582 else
583 build_sk_flow_key(fl4, sk);
584}
585
586static DEFINE_SPINLOCK(fnhe_lock);
587
588static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
589{
590 struct rtable *rt;
591
592 rt = rcu_dereference(fnhe->fnhe_rth_input);
593 if (rt) {
594 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
595 dst_dev_put(&rt->dst);
596 dst_release(&rt->dst);
597 }
598 rt = rcu_dereference(fnhe->fnhe_rth_output);
599 if (rt) {
600 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
601 dst_dev_put(&rt->dst);
602 dst_release(&rt->dst);
603 }
604}
605
606static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
607{
608 struct fib_nh_exception *fnhe, *oldest;
609
610 oldest = rcu_dereference(hash->chain);
611 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
612 fnhe = rcu_dereference(fnhe->fnhe_next)) {
613 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
614 oldest = fnhe;
615 }
616 fnhe_flush_routes(oldest);
617 return oldest;
618}
619
620static inline u32 fnhe_hashfun(__be32 daddr)
621{
622 static u32 fnhe_hashrnd __read_mostly;
623 u32 hval;
624
625 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
626 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
627 return hash_32(hval, FNHE_HASH_SHIFT);
628}
629
630static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
631{
632 rt->rt_pmtu = fnhe->fnhe_pmtu;
633 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
634 rt->dst.expires = fnhe->fnhe_expires;
635
636 if (fnhe->fnhe_gw) {
637 rt->rt_flags |= RTCF_REDIRECTED;
638 rt->rt_uses_gateway = 1;
639 rt->rt_gw_family = AF_INET;
640 rt->rt_gw4 = fnhe->fnhe_gw;
641 }
642}
643
644static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
645 __be32 gw, u32 pmtu, bool lock,
646 unsigned long expires)
647{
648 struct fnhe_hash_bucket *hash;
649 struct fib_nh_exception *fnhe;
650 struct rtable *rt;
651 u32 genid, hval;
652 unsigned int i;
653 int depth;
654
655 genid = fnhe_genid(dev_net(nhc->nhc_dev));
656 hval = fnhe_hashfun(daddr);
657
658 spin_lock_bh(&fnhe_lock);
659
660 hash = rcu_dereference(nhc->nhc_exceptions);
661 if (!hash) {
662 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
663 if (!hash)
664 goto out_unlock;
665 rcu_assign_pointer(nhc->nhc_exceptions, hash);
666 }
667
668 hash += hval;
669
670 depth = 0;
671 for (fnhe = rcu_dereference(hash->chain); fnhe;
672 fnhe = rcu_dereference(fnhe->fnhe_next)) {
673 if (fnhe->fnhe_daddr == daddr)
674 break;
675 depth++;
676 }
677
678 if (fnhe) {
679 if (fnhe->fnhe_genid != genid)
680 fnhe->fnhe_genid = genid;
681 if (gw)
682 fnhe->fnhe_gw = gw;
683 if (pmtu) {
684 fnhe->fnhe_pmtu = pmtu;
685 fnhe->fnhe_mtu_locked = lock;
686 }
687 fnhe->fnhe_expires = max(1UL, expires);
688 /* Update all cached dsts too */
689 rt = rcu_dereference(fnhe->fnhe_rth_input);
690 if (rt)
691 fill_route_from_fnhe(rt, fnhe);
692 rt = rcu_dereference(fnhe->fnhe_rth_output);
693 if (rt)
694 fill_route_from_fnhe(rt, fnhe);
695 } else {
696 if (depth > FNHE_RECLAIM_DEPTH)
697 fnhe = fnhe_oldest(hash);
698 else {
699 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
700 if (!fnhe)
701 goto out_unlock;
702
703 fnhe->fnhe_next = hash->chain;
704 rcu_assign_pointer(hash->chain, fnhe);
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 /* Exception created; mark the cached routes for the nexthop
714 * stale, so anyone caching it rechecks if this exception
715 * applies to them.
716 */
717 rt = rcu_dereference(nhc->nhc_rth_input);
718 if (rt)
719 rt->dst.obsolete = DST_OBSOLETE_KILL;
720
721 for_each_possible_cpu(i) {
722 struct rtable __rcu **prt;
723 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
724 rt = rcu_dereference(*prt);
725 if (rt)
726 rt->dst.obsolete = DST_OBSOLETE_KILL;
727 }
728 }
729
730 fnhe->fnhe_stamp = jiffies;
731
732out_unlock:
733 spin_unlock_bh(&fnhe_lock);
734}
735
736static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
737 bool kill_route)
738{
739 __be32 new_gw = icmp_hdr(skb)->un.gateway;
740 __be32 old_gw = ip_hdr(skb)->saddr;
741 struct net_device *dev = skb->dev;
742 struct in_device *in_dev;
743 struct fib_result res;
744 struct neighbour *n;
745 struct net *net;
746
747 switch (icmp_hdr(skb)->code & 7) {
748 case ICMP_REDIR_NET:
749 case ICMP_REDIR_NETTOS:
750 case ICMP_REDIR_HOST:
751 case ICMP_REDIR_HOSTTOS:
752 break;
753
754 default:
755 return;
756 }
757
758 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
759 return;
760
761 in_dev = __in_dev_get_rcu(dev);
762 if (!in_dev)
763 return;
764
765 net = dev_net(dev);
766 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
767 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
768 ipv4_is_zeronet(new_gw))
769 goto reject_redirect;
770
771 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
772 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
773 goto reject_redirect;
774 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
775 goto reject_redirect;
776 } else {
777 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
778 goto reject_redirect;
779 }
780
781 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
782 if (!n)
783 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
784 if (!IS_ERR(n)) {
785 if (!(n->nud_state & NUD_VALID)) {
786 neigh_event_send(n, NULL);
787 } else {
788 if (fib_lookup(net, fl4, &res, 0) == 0) {
789 struct fib_nh_common *nhc = FIB_RES_NHC(res);
790
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->rate_tokens == 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 u32 old_mtu = ipv4_mtu(dst);
1017 struct fib_result res;
1018 bool lock = false;
1019
1020 if (ip_mtu_locked(dst))
1021 return;
1022
1023 if (old_mtu < mtu)
1024 return;
1025
1026 if (mtu < ip_rt_min_pmtu) {
1027 lock = true;
1028 mtu = min(old_mtu, ip_rt_min_pmtu);
1029 }
1030
1031 if (rt->rt_pmtu == mtu && !lock &&
1032 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1033 return;
1034
1035 rcu_read_lock();
1036 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1037 struct fib_nh_common *nhc = FIB_RES_NHC(res);
1038
1039 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1040 jiffies + ip_rt_mtu_expires);
1041 }
1042 rcu_read_unlock();
1043}
1044
1045static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1046 struct sk_buff *skb, u32 mtu)
1047{
1048 struct rtable *rt = (struct rtable *) dst;
1049 struct flowi4 fl4;
1050
1051 ip_rt_build_flow_key(&fl4, sk, skb);
1052 __ip_rt_update_pmtu(rt, &fl4, mtu);
1053}
1054
1055void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1056 int oif, u8 protocol)
1057{
1058 const struct iphdr *iph = (const struct iphdr *) skb->data;
1059 struct flowi4 fl4;
1060 struct rtable *rt;
1061 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1062
1063 __build_flow_key(net, &fl4, NULL, iph, oif,
1064 RT_TOS(iph->tos), protocol, mark, 0);
1065 rt = __ip_route_output_key(net, &fl4);
1066 if (!IS_ERR(rt)) {
1067 __ip_rt_update_pmtu(rt, &fl4, mtu);
1068 ip_rt_put(rt);
1069 }
1070}
1071EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1072
1073static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1074{
1075 const struct iphdr *iph = (const struct iphdr *) skb->data;
1076 struct flowi4 fl4;
1077 struct rtable *rt;
1078
1079 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1080
1081 if (!fl4.flowi4_mark)
1082 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1083
1084 rt = __ip_route_output_key(sock_net(sk), &fl4);
1085 if (!IS_ERR(rt)) {
1086 __ip_rt_update_pmtu(rt, &fl4, mtu);
1087 ip_rt_put(rt);
1088 }
1089}
1090
1091void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1092{
1093 const struct iphdr *iph = (const struct iphdr *) skb->data;
1094 struct flowi4 fl4;
1095 struct rtable *rt;
1096 struct dst_entry *odst = NULL;
1097 bool new = false;
1098 struct net *net = sock_net(sk);
1099
1100 bh_lock_sock(sk);
1101
1102 if (!ip_sk_accept_pmtu(sk))
1103 goto out;
1104
1105 odst = sk_dst_get(sk);
1106
1107 if (sock_owned_by_user(sk) || !odst) {
1108 __ipv4_sk_update_pmtu(skb, sk, mtu);
1109 goto out;
1110 }
1111
1112 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1113
1114 rt = (struct rtable *)odst;
1115 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1116 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1117 if (IS_ERR(rt))
1118 goto out;
1119
1120 new = true;
1121 }
1122
1123 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1124
1125 if (!dst_check(&rt->dst, 0)) {
1126 if (new)
1127 dst_release(&rt->dst);
1128
1129 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1130 if (IS_ERR(rt))
1131 goto out;
1132
1133 new = true;
1134 }
1135
1136 if (new)
1137 sk_dst_set(sk, &rt->dst);
1138
1139out:
1140 bh_unlock_sock(sk);
1141 dst_release(odst);
1142}
1143EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1144
1145void ipv4_redirect(struct sk_buff *skb, struct net *net,
1146 int oif, u8 protocol)
1147{
1148 const struct iphdr *iph = (const struct iphdr *) skb->data;
1149 struct flowi4 fl4;
1150 struct rtable *rt;
1151
1152 __build_flow_key(net, &fl4, NULL, iph, oif,
1153 RT_TOS(iph->tos), protocol, 0, 0);
1154 rt = __ip_route_output_key(net, &fl4);
1155 if (!IS_ERR(rt)) {
1156 __ip_do_redirect(rt, skb, &fl4, false);
1157 ip_rt_put(rt);
1158 }
1159}
1160EXPORT_SYMBOL_GPL(ipv4_redirect);
1161
1162void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1163{
1164 const struct iphdr *iph = (const struct iphdr *) skb->data;
1165 struct flowi4 fl4;
1166 struct rtable *rt;
1167 struct net *net = sock_net(sk);
1168
1169 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1170 rt = __ip_route_output_key(net, &fl4);
1171 if (!IS_ERR(rt)) {
1172 __ip_do_redirect(rt, skb, &fl4, false);
1173 ip_rt_put(rt);
1174 }
1175}
1176EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1177
1178static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1179{
1180 struct rtable *rt = (struct rtable *) dst;
1181
1182 /* All IPV4 dsts are created with ->obsolete set to the value
1183 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1184 * into this function always.
1185 *
1186 * When a PMTU/redirect information update invalidates a route,
1187 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1188 * DST_OBSOLETE_DEAD.
1189 */
1190 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1191 return NULL;
1192 return dst;
1193}
1194
1195static void ipv4_send_dest_unreach(struct sk_buff *skb)
1196{
1197 struct ip_options opt;
1198 int res;
1199
1200 /* Recompile ip options since IPCB may not be valid anymore.
1201 * Also check we have a reasonable ipv4 header.
1202 */
1203 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1204 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1205 return;
1206
1207 memset(&opt, 0, sizeof(opt));
1208 if (ip_hdr(skb)->ihl > 5) {
1209 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1210 return;
1211 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1212
1213 rcu_read_lock();
1214 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1215 rcu_read_unlock();
1216
1217 if (res)
1218 return;
1219 }
1220 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1221}
1222
1223static void ipv4_link_failure(struct sk_buff *skb)
1224{
1225 struct rtable *rt;
1226
1227 ipv4_send_dest_unreach(skb);
1228
1229 rt = skb_rtable(skb);
1230 if (rt)
1231 dst_set_expires(&rt->dst, 0);
1232}
1233
1234static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1235{
1236 pr_debug("%s: %pI4 -> %pI4, %s\n",
1237 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1238 skb->dev ? skb->dev->name : "?");
1239 kfree_skb(skb);
1240 WARN_ON(1);
1241 return 0;
1242}
1243
1244/*
1245 We do not cache source address of outgoing interface,
1246 because it is used only by IP RR, TS and SRR options,
1247 so that it out of fast path.
1248
1249 BTW remember: "addr" is allowed to be not aligned
1250 in IP options!
1251 */
1252
1253void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1254{
1255 __be32 src;
1256
1257 if (rt_is_output_route(rt))
1258 src = ip_hdr(skb)->saddr;
1259 else {
1260 struct fib_result res;
1261 struct iphdr *iph = ip_hdr(skb);
1262 struct flowi4 fl4 = {
1263 .daddr = iph->daddr,
1264 .saddr = iph->saddr,
1265 .flowi4_tos = RT_TOS(iph->tos),
1266 .flowi4_oif = rt->dst.dev->ifindex,
1267 .flowi4_iif = skb->dev->ifindex,
1268 .flowi4_mark = skb->mark,
1269 };
1270
1271 rcu_read_lock();
1272 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1273 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1274 else
1275 src = inet_select_addr(rt->dst.dev,
1276 rt_nexthop(rt, iph->daddr),
1277 RT_SCOPE_UNIVERSE);
1278 rcu_read_unlock();
1279 }
1280 memcpy(addr, &src, 4);
1281}
1282
1283#ifdef CONFIG_IP_ROUTE_CLASSID
1284static void set_class_tag(struct rtable *rt, u32 tag)
1285{
1286 if (!(rt->dst.tclassid & 0xFFFF))
1287 rt->dst.tclassid |= tag & 0xFFFF;
1288 if (!(rt->dst.tclassid & 0xFFFF0000))
1289 rt->dst.tclassid |= tag & 0xFFFF0000;
1290}
1291#endif
1292
1293static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1294{
1295 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1296 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1297 ip_rt_min_advmss);
1298
1299 return min(advmss, IPV4_MAX_PMTU - header_size);
1300}
1301
1302static unsigned int ipv4_mtu(const struct dst_entry *dst)
1303{
1304 const struct rtable *rt = (const struct rtable *) dst;
1305 unsigned int mtu = rt->rt_pmtu;
1306
1307 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1308 mtu = dst_metric_raw(dst, RTAX_MTU);
1309
1310 if (mtu)
1311 return mtu;
1312
1313 mtu = READ_ONCE(dst->dev->mtu);
1314
1315 if (unlikely(ip_mtu_locked(dst))) {
1316 if (rt->rt_uses_gateway && mtu > 576)
1317 mtu = 576;
1318 }
1319
1320 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1321
1322 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1323}
1324
1325static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1326{
1327 struct fnhe_hash_bucket *hash;
1328 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1329 u32 hval = fnhe_hashfun(daddr);
1330
1331 spin_lock_bh(&fnhe_lock);
1332
1333 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1334 lockdep_is_held(&fnhe_lock));
1335 hash += hval;
1336
1337 fnhe_p = &hash->chain;
1338 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1339 while (fnhe) {
1340 if (fnhe->fnhe_daddr == daddr) {
1341 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1342 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1343 /* set fnhe_daddr to 0 to ensure it won't bind with
1344 * new dsts in rt_bind_exception().
1345 */
1346 fnhe->fnhe_daddr = 0;
1347 fnhe_flush_routes(fnhe);
1348 kfree_rcu(fnhe, rcu);
1349 break;
1350 }
1351 fnhe_p = &fnhe->fnhe_next;
1352 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1353 lockdep_is_held(&fnhe_lock));
1354 }
1355
1356 spin_unlock_bh(&fnhe_lock);
1357}
1358
1359static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1360 __be32 daddr)
1361{
1362 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1363 struct fib_nh_exception *fnhe;
1364 u32 hval;
1365
1366 if (!hash)
1367 return NULL;
1368
1369 hval = fnhe_hashfun(daddr);
1370
1371 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1372 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1373 if (fnhe->fnhe_daddr == daddr) {
1374 if (fnhe->fnhe_expires &&
1375 time_after(jiffies, fnhe->fnhe_expires)) {
1376 ip_del_fnhe(nhc, daddr);
1377 break;
1378 }
1379 return fnhe;
1380 }
1381 }
1382 return NULL;
1383}
1384
1385/* MTU selection:
1386 * 1. mtu on route is locked - use it
1387 * 2. mtu from nexthop exception
1388 * 3. mtu from egress device
1389 */
1390
1391u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1392{
1393 struct fib_nh_common *nhc = res->nhc;
1394 struct net_device *dev = nhc->nhc_dev;
1395 struct fib_info *fi = res->fi;
1396 u32 mtu = 0;
1397
1398 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1399 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1400 mtu = fi->fib_mtu;
1401
1402 if (likely(!mtu)) {
1403 struct fib_nh_exception *fnhe;
1404
1405 fnhe = find_exception(nhc, daddr);
1406 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1407 mtu = fnhe->fnhe_pmtu;
1408 }
1409
1410 if (likely(!mtu))
1411 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1412
1413 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1414}
1415
1416static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1417 __be32 daddr, const bool do_cache)
1418{
1419 bool ret = false;
1420
1421 spin_lock_bh(&fnhe_lock);
1422
1423 if (daddr == fnhe->fnhe_daddr) {
1424 struct rtable __rcu **porig;
1425 struct rtable *orig;
1426 int genid = fnhe_genid(dev_net(rt->dst.dev));
1427
1428 if (rt_is_input_route(rt))
1429 porig = &fnhe->fnhe_rth_input;
1430 else
1431 porig = &fnhe->fnhe_rth_output;
1432 orig = rcu_dereference(*porig);
1433
1434 if (fnhe->fnhe_genid != genid) {
1435 fnhe->fnhe_genid = genid;
1436 fnhe->fnhe_gw = 0;
1437 fnhe->fnhe_pmtu = 0;
1438 fnhe->fnhe_expires = 0;
1439 fnhe->fnhe_mtu_locked = false;
1440 fnhe_flush_routes(fnhe);
1441 orig = NULL;
1442 }
1443 fill_route_from_fnhe(rt, fnhe);
1444 if (!rt->rt_gw4) {
1445 rt->rt_gw4 = daddr;
1446 rt->rt_gw_family = AF_INET;
1447 }
1448
1449 if (do_cache) {
1450 dst_hold(&rt->dst);
1451 rcu_assign_pointer(*porig, rt);
1452 if (orig) {
1453 dst_dev_put(&orig->dst);
1454 dst_release(&orig->dst);
1455 }
1456 ret = true;
1457 }
1458
1459 fnhe->fnhe_stamp = jiffies;
1460 }
1461 spin_unlock_bh(&fnhe_lock);
1462
1463 return ret;
1464}
1465
1466static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1467{
1468 struct rtable *orig, *prev, **p;
1469 bool ret = true;
1470
1471 if (rt_is_input_route(rt)) {
1472 p = (struct rtable **)&nhc->nhc_rth_input;
1473 } else {
1474 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1475 }
1476 orig = *p;
1477
1478 /* hold dst before doing cmpxchg() to avoid race condition
1479 * on this dst
1480 */
1481 dst_hold(&rt->dst);
1482 prev = cmpxchg(p, orig, rt);
1483 if (prev == orig) {
1484 if (orig) {
1485 rt_add_uncached_list(orig);
1486 dst_release(&orig->dst);
1487 }
1488 } else {
1489 dst_release(&rt->dst);
1490 ret = false;
1491 }
1492
1493 return ret;
1494}
1495
1496struct uncached_list {
1497 spinlock_t lock;
1498 struct list_head head;
1499};
1500
1501static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1502
1503void rt_add_uncached_list(struct rtable *rt)
1504{
1505 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1506
1507 rt->rt_uncached_list = ul;
1508
1509 spin_lock_bh(&ul->lock);
1510 list_add_tail(&rt->rt_uncached, &ul->head);
1511 spin_unlock_bh(&ul->lock);
1512}
1513
1514void rt_del_uncached_list(struct rtable *rt)
1515{
1516 if (!list_empty(&rt->rt_uncached)) {
1517 struct uncached_list *ul = rt->rt_uncached_list;
1518
1519 spin_lock_bh(&ul->lock);
1520 list_del(&rt->rt_uncached);
1521 spin_unlock_bh(&ul->lock);
1522 }
1523}
1524
1525static void ipv4_dst_destroy(struct dst_entry *dst)
1526{
1527 struct rtable *rt = (struct rtable *)dst;
1528
1529 ip_dst_metrics_put(dst);
1530 rt_del_uncached_list(rt);
1531}
1532
1533void rt_flush_dev(struct net_device *dev)
1534{
1535 struct rtable *rt;
1536 int cpu;
1537
1538 for_each_possible_cpu(cpu) {
1539 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1540
1541 spin_lock_bh(&ul->lock);
1542 list_for_each_entry(rt, &ul->head, rt_uncached) {
1543 if (rt->dst.dev != dev)
1544 continue;
1545 rt->dst.dev = blackhole_netdev;
1546 dev_hold(rt->dst.dev);
1547 dev_put(dev);
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 nopolicy, bool noxfrm, bool will_cache)
1622{
1623 struct rtable *rt;
1624
1625 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1626 (will_cache ? 0 : DST_HOST) |
1627 (nopolicy ? DST_NOPOLICY : 0) |
1628 (noxfrm ? DST_NOXFRM : 0));
1629
1630 if (rt) {
1631 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1632 rt->rt_flags = flags;
1633 rt->rt_type = type;
1634 rt->rt_is_input = 0;
1635 rt->rt_iif = 0;
1636 rt->rt_pmtu = 0;
1637 rt->rt_mtu_locked = 0;
1638 rt->rt_uses_gateway = 0;
1639 rt->rt_gw_family = 0;
1640 rt->rt_gw4 = 0;
1641 INIT_LIST_HEAD(&rt->rt_uncached);
1642
1643 rt->dst.output = ip_output;
1644 if (flags & RTCF_LOCAL)
1645 rt->dst.input = ip_local_deliver;
1646 }
1647
1648 return rt;
1649}
1650EXPORT_SYMBOL(rt_dst_alloc);
1651
1652struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1653{
1654 struct rtable *new_rt;
1655
1656 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1657 rt->dst.flags);
1658
1659 if (new_rt) {
1660 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1661 new_rt->rt_flags = rt->rt_flags;
1662 new_rt->rt_type = rt->rt_type;
1663 new_rt->rt_is_input = rt->rt_is_input;
1664 new_rt->rt_iif = rt->rt_iif;
1665 new_rt->rt_pmtu = rt->rt_pmtu;
1666 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1667 new_rt->rt_gw_family = rt->rt_gw_family;
1668 if (rt->rt_gw_family == AF_INET)
1669 new_rt->rt_gw4 = rt->rt_gw4;
1670 else if (rt->rt_gw_family == AF_INET6)
1671 new_rt->rt_gw6 = rt->rt_gw6;
1672 INIT_LIST_HEAD(&new_rt->rt_uncached);
1673
1674 new_rt->dst.flags |= DST_HOST;
1675 new_rt->dst.input = rt->dst.input;
1676 new_rt->dst.output = rt->dst.output;
1677 new_rt->dst.error = rt->dst.error;
1678 new_rt->dst.lastuse = jiffies;
1679 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1680 }
1681 return new_rt;
1682}
1683EXPORT_SYMBOL(rt_dst_clone);
1684
1685/* called in rcu_read_lock() section */
1686int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1687 u8 tos, struct net_device *dev,
1688 struct in_device *in_dev, u32 *itag)
1689{
1690 int err;
1691
1692 /* Primary sanity checks. */
1693 if (!in_dev)
1694 return -EINVAL;
1695
1696 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1697 skb->protocol != htons(ETH_P_IP))
1698 return -EINVAL;
1699
1700 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1701 return -EINVAL;
1702
1703 if (ipv4_is_zeronet(saddr)) {
1704 if (!ipv4_is_local_multicast(daddr) &&
1705 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1706 return -EINVAL;
1707 } else {
1708 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1709 in_dev, itag);
1710 if (err < 0)
1711 return err;
1712 }
1713 return 0;
1714}
1715
1716/* called in rcu_read_lock() section */
1717static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1718 u8 tos, struct net_device *dev, int our)
1719{
1720 struct in_device *in_dev = __in_dev_get_rcu(dev);
1721 unsigned int flags = RTCF_MULTICAST;
1722 struct rtable *rth;
1723 u32 itag = 0;
1724 int err;
1725
1726 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1727 if (err)
1728 return err;
1729
1730 if (our)
1731 flags |= RTCF_LOCAL;
1732
1733 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1734 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1735 if (!rth)
1736 return -ENOBUFS;
1737
1738#ifdef CONFIG_IP_ROUTE_CLASSID
1739 rth->dst.tclassid = itag;
1740#endif
1741 rth->dst.output = ip_rt_bug;
1742 rth->rt_is_input= 1;
1743
1744#ifdef CONFIG_IP_MROUTE
1745 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1746 rth->dst.input = ip_mr_input;
1747#endif
1748 RT_CACHE_STAT_INC(in_slow_mc);
1749
1750 skb_dst_set(skb, &rth->dst);
1751 return 0;
1752}
1753
1754
1755static void ip_handle_martian_source(struct net_device *dev,
1756 struct in_device *in_dev,
1757 struct sk_buff *skb,
1758 __be32 daddr,
1759 __be32 saddr)
1760{
1761 RT_CACHE_STAT_INC(in_martian_src);
1762#ifdef CONFIG_IP_ROUTE_VERBOSE
1763 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1764 /*
1765 * RFC1812 recommendation, if source is martian,
1766 * the only hint is MAC header.
1767 */
1768 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1769 &daddr, &saddr, dev->name);
1770 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1771 print_hex_dump(KERN_WARNING, "ll header: ",
1772 DUMP_PREFIX_OFFSET, 16, 1,
1773 skb_mac_header(skb),
1774 dev->hard_header_len, false);
1775 }
1776 }
1777#endif
1778}
1779
1780/* called in rcu_read_lock() section */
1781static int __mkroute_input(struct sk_buff *skb,
1782 const struct fib_result *res,
1783 struct in_device *in_dev,
1784 __be32 daddr, __be32 saddr, u32 tos)
1785{
1786 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1787 struct net_device *dev = nhc->nhc_dev;
1788 struct fib_nh_exception *fnhe;
1789 struct rtable *rth;
1790 int err;
1791 struct in_device *out_dev;
1792 bool do_cache;
1793 u32 itag = 0;
1794
1795 /* get a working reference to the output device */
1796 out_dev = __in_dev_get_rcu(dev);
1797 if (!out_dev) {
1798 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1799 return -EINVAL;
1800 }
1801
1802 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1803 in_dev->dev, in_dev, &itag);
1804 if (err < 0) {
1805 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1806 saddr);
1807
1808 goto cleanup;
1809 }
1810
1811 do_cache = res->fi && !itag;
1812 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1813 skb->protocol == htons(ETH_P_IP)) {
1814 __be32 gw;
1815
1816 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1817 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1818 inet_addr_onlink(out_dev, saddr, gw))
1819 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1820 }
1821
1822 if (skb->protocol != htons(ETH_P_IP)) {
1823 /* Not IP (i.e. ARP). Do not create route, if it is
1824 * invalid for proxy arp. DNAT routes are always valid.
1825 *
1826 * Proxy arp feature have been extended to allow, ARP
1827 * replies back to the same interface, to support
1828 * Private VLAN switch technologies. See arp.c.
1829 */
1830 if (out_dev == in_dev &&
1831 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1832 err = -EINVAL;
1833 goto cleanup;
1834 }
1835 }
1836
1837 fnhe = find_exception(nhc, daddr);
1838 if (do_cache) {
1839 if (fnhe)
1840 rth = rcu_dereference(fnhe->fnhe_rth_input);
1841 else
1842 rth = rcu_dereference(nhc->nhc_rth_input);
1843 if (rt_cache_valid(rth)) {
1844 skb_dst_set_noref(skb, &rth->dst);
1845 goto out;
1846 }
1847 }
1848
1849 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1850 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1851 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1852 if (!rth) {
1853 err = -ENOBUFS;
1854 goto cleanup;
1855 }
1856
1857 rth->rt_is_input = 1;
1858 RT_CACHE_STAT_INC(in_slow_tot);
1859
1860 rth->dst.input = ip_forward;
1861
1862 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1863 do_cache);
1864 lwtunnel_set_redirect(&rth->dst);
1865 skb_dst_set(skb, &rth->dst);
1866out:
1867 err = 0;
1868 cleanup:
1869 return err;
1870}
1871
1872#ifdef CONFIG_IP_ROUTE_MULTIPATH
1873/* To make ICMP packets follow the right flow, the multipath hash is
1874 * calculated from the inner IP addresses.
1875 */
1876static void ip_multipath_l3_keys(const struct sk_buff *skb,
1877 struct flow_keys *hash_keys)
1878{
1879 const struct iphdr *outer_iph = ip_hdr(skb);
1880 const struct iphdr *key_iph = outer_iph;
1881 const struct iphdr *inner_iph;
1882 const struct icmphdr *icmph;
1883 struct iphdr _inner_iph;
1884 struct icmphdr _icmph;
1885
1886 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1887 goto out;
1888
1889 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1890 goto out;
1891
1892 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1893 &_icmph);
1894 if (!icmph)
1895 goto out;
1896
1897 if (icmph->type != ICMP_DEST_UNREACH &&
1898 icmph->type != ICMP_REDIRECT &&
1899 icmph->type != ICMP_TIME_EXCEEDED &&
1900 icmph->type != ICMP_PARAMETERPROB)
1901 goto out;
1902
1903 inner_iph = skb_header_pointer(skb,
1904 outer_iph->ihl * 4 + sizeof(_icmph),
1905 sizeof(_inner_iph), &_inner_iph);
1906 if (!inner_iph)
1907 goto out;
1908
1909 key_iph = inner_iph;
1910out:
1911 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1912 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1913}
1914
1915/* if skb is set it will be used and fl4 can be NULL */
1916int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1917 const struct sk_buff *skb, struct flow_keys *flkeys)
1918{
1919 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1920 struct flow_keys hash_keys;
1921 u32 mhash;
1922
1923 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1924 case 0:
1925 memset(&hash_keys, 0, sizeof(hash_keys));
1926 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1927 if (skb) {
1928 ip_multipath_l3_keys(skb, &hash_keys);
1929 } else {
1930 hash_keys.addrs.v4addrs.src = fl4->saddr;
1931 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1932 }
1933 break;
1934 case 1:
1935 /* skb is currently provided only when forwarding */
1936 if (skb) {
1937 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1938 struct flow_keys keys;
1939
1940 /* short-circuit if we already have L4 hash present */
1941 if (skb->l4_hash)
1942 return skb_get_hash_raw(skb) >> 1;
1943
1944 memset(&hash_keys, 0, sizeof(hash_keys));
1945
1946 if (!flkeys) {
1947 skb_flow_dissect_flow_keys(skb, &keys, flag);
1948 flkeys = &keys;
1949 }
1950
1951 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1952 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1953 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1954 hash_keys.ports.src = flkeys->ports.src;
1955 hash_keys.ports.dst = flkeys->ports.dst;
1956 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1957 } else {
1958 memset(&hash_keys, 0, sizeof(hash_keys));
1959 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1960 hash_keys.addrs.v4addrs.src = fl4->saddr;
1961 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1962 hash_keys.ports.src = fl4->fl4_sport;
1963 hash_keys.ports.dst = fl4->fl4_dport;
1964 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1965 }
1966 break;
1967 case 2:
1968 memset(&hash_keys, 0, sizeof(hash_keys));
1969 /* skb is currently provided only when forwarding */
1970 if (skb) {
1971 struct flow_keys keys;
1972
1973 skb_flow_dissect_flow_keys(skb, &keys, 0);
1974 /* Inner can be v4 or v6 */
1975 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1976 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1977 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1978 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1979 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1980 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1981 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1982 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1983 hash_keys.tags.flow_label = keys.tags.flow_label;
1984 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1985 } else {
1986 /* Same as case 0 */
1987 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1988 ip_multipath_l3_keys(skb, &hash_keys);
1989 }
1990 } else {
1991 /* Same as case 0 */
1992 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1993 hash_keys.addrs.v4addrs.src = fl4->saddr;
1994 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1995 }
1996 break;
1997 }
1998 mhash = flow_hash_from_keys(&hash_keys);
1999
2000 if (multipath_hash)
2001 mhash = jhash_2words(mhash, multipath_hash, 0);
2002
2003 return mhash >> 1;
2004}
2005#endif /* CONFIG_IP_ROUTE_MULTIPATH */
2006
2007static int ip_mkroute_input(struct sk_buff *skb,
2008 struct fib_result *res,
2009 struct in_device *in_dev,
2010 __be32 daddr, __be32 saddr, u32 tos,
2011 struct flow_keys *hkeys)
2012{
2013#ifdef CONFIG_IP_ROUTE_MULTIPATH
2014 if (res->fi && fib_info_num_path(res->fi) > 1) {
2015 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2016
2017 fib_select_multipath(res, h);
2018 }
2019#endif
2020
2021 /* create a routing cache entry */
2022 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2023}
2024
2025/*
2026 * NOTE. We drop all the packets that has local source
2027 * addresses, because every properly looped back packet
2028 * must have correct destination already attached by output routine.
2029 *
2030 * Such approach solves two big problems:
2031 * 1. Not simplex devices are handled properly.
2032 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2033 * called with rcu_read_lock()
2034 */
2035
2036static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2037 u8 tos, struct net_device *dev,
2038 struct fib_result *res)
2039{
2040 struct in_device *in_dev = __in_dev_get_rcu(dev);
2041 struct flow_keys *flkeys = NULL, _flkeys;
2042 struct net *net = dev_net(dev);
2043 struct ip_tunnel_info *tun_info;
2044 int err = -EINVAL;
2045 unsigned int flags = 0;
2046 u32 itag = 0;
2047 struct rtable *rth;
2048 struct flowi4 fl4;
2049 bool do_cache = true;
2050
2051 /* IP on this device is disabled. */
2052
2053 if (!in_dev)
2054 goto out;
2055
2056 /* Check for the most weird martians, which can be not detected
2057 by fib_lookup.
2058 */
2059
2060 tun_info = skb_tunnel_info(skb);
2061 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2062 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2063 else
2064 fl4.flowi4_tun_key.tun_id = 0;
2065 skb_dst_drop(skb);
2066
2067 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2068 goto martian_source;
2069
2070 res->fi = NULL;
2071 res->table = NULL;
2072 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2073 goto brd_input;
2074
2075 /* Accept zero addresses only to limited broadcast;
2076 * I even do not know to fix it or not. Waiting for complains :-)
2077 */
2078 if (ipv4_is_zeronet(saddr))
2079 goto martian_source;
2080
2081 if (ipv4_is_zeronet(daddr))
2082 goto martian_destination;
2083
2084 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2085 * and call it once if daddr or/and saddr are loopback addresses
2086 */
2087 if (ipv4_is_loopback(daddr)) {
2088 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2089 goto martian_destination;
2090 } else if (ipv4_is_loopback(saddr)) {
2091 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2092 goto martian_source;
2093 }
2094
2095 /*
2096 * Now we are ready to route packet.
2097 */
2098 fl4.flowi4_oif = 0;
2099 fl4.flowi4_iif = dev->ifindex;
2100 fl4.flowi4_mark = skb->mark;
2101 fl4.flowi4_tos = tos;
2102 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2103 fl4.flowi4_flags = 0;
2104 fl4.daddr = daddr;
2105 fl4.saddr = saddr;
2106 fl4.flowi4_uid = sock_net_uid(net, NULL);
2107
2108 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2109 flkeys = &_flkeys;
2110 } else {
2111 fl4.flowi4_proto = 0;
2112 fl4.fl4_sport = 0;
2113 fl4.fl4_dport = 0;
2114 }
2115
2116 err = fib_lookup(net, &fl4, res, 0);
2117 if (err != 0) {
2118 if (!IN_DEV_FORWARD(in_dev))
2119 err = -EHOSTUNREACH;
2120 goto no_route;
2121 }
2122
2123 if (res->type == RTN_BROADCAST) {
2124 if (IN_DEV_BFORWARD(in_dev))
2125 goto make_route;
2126 /* not do cache if bc_forwarding is enabled */
2127 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2128 do_cache = false;
2129 goto brd_input;
2130 }
2131
2132 if (res->type == RTN_LOCAL) {
2133 err = fib_validate_source(skb, saddr, daddr, tos,
2134 0, dev, in_dev, &itag);
2135 if (err < 0)
2136 goto martian_source;
2137 goto local_input;
2138 }
2139
2140 if (!IN_DEV_FORWARD(in_dev)) {
2141 err = -EHOSTUNREACH;
2142 goto no_route;
2143 }
2144 if (res->type != RTN_UNICAST)
2145 goto martian_destination;
2146
2147make_route:
2148 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2149out: return err;
2150
2151brd_input:
2152 if (skb->protocol != htons(ETH_P_IP))
2153 goto e_inval;
2154
2155 if (!ipv4_is_zeronet(saddr)) {
2156 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2157 in_dev, &itag);
2158 if (err < 0)
2159 goto martian_source;
2160 }
2161 flags |= RTCF_BROADCAST;
2162 res->type = RTN_BROADCAST;
2163 RT_CACHE_STAT_INC(in_brd);
2164
2165local_input:
2166 do_cache &= res->fi && !itag;
2167 if (do_cache) {
2168 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2169
2170 rth = rcu_dereference(nhc->nhc_rth_input);
2171 if (rt_cache_valid(rth)) {
2172 skb_dst_set_noref(skb, &rth->dst);
2173 err = 0;
2174 goto out;
2175 }
2176 }
2177
2178 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2179 flags | RTCF_LOCAL, res->type,
2180 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2181 if (!rth)
2182 goto e_nobufs;
2183
2184 rth->dst.output= ip_rt_bug;
2185#ifdef CONFIG_IP_ROUTE_CLASSID
2186 rth->dst.tclassid = itag;
2187#endif
2188 rth->rt_is_input = 1;
2189
2190 RT_CACHE_STAT_INC(in_slow_tot);
2191 if (res->type == RTN_UNREACHABLE) {
2192 rth->dst.input= ip_error;
2193 rth->dst.error= -err;
2194 rth->rt_flags &= ~RTCF_LOCAL;
2195 }
2196
2197 if (do_cache) {
2198 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2199
2200 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2201 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2202 WARN_ON(rth->dst.input == lwtunnel_input);
2203 rth->dst.lwtstate->orig_input = rth->dst.input;
2204 rth->dst.input = lwtunnel_input;
2205 }
2206
2207 if (unlikely(!rt_cache_route(nhc, rth)))
2208 rt_add_uncached_list(rth);
2209 }
2210 skb_dst_set(skb, &rth->dst);
2211 err = 0;
2212 goto out;
2213
2214no_route:
2215 RT_CACHE_STAT_INC(in_no_route);
2216 res->type = RTN_UNREACHABLE;
2217 res->fi = NULL;
2218 res->table = NULL;
2219 goto local_input;
2220
2221 /*
2222 * Do not cache martian addresses: they should be logged (RFC1812)
2223 */
2224martian_destination:
2225 RT_CACHE_STAT_INC(in_martian_dst);
2226#ifdef CONFIG_IP_ROUTE_VERBOSE
2227 if (IN_DEV_LOG_MARTIANS(in_dev))
2228 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2229 &daddr, &saddr, dev->name);
2230#endif
2231
2232e_inval:
2233 err = -EINVAL;
2234 goto out;
2235
2236e_nobufs:
2237 err = -ENOBUFS;
2238 goto out;
2239
2240martian_source:
2241 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2242 goto out;
2243}
2244
2245int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2246 u8 tos, struct net_device *dev)
2247{
2248 struct fib_result res;
2249 int err;
2250
2251 tos &= IPTOS_RT_MASK;
2252 rcu_read_lock();
2253 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2254 rcu_read_unlock();
2255
2256 return err;
2257}
2258EXPORT_SYMBOL(ip_route_input_noref);
2259
2260/* called with rcu_read_lock held */
2261int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2262 u8 tos, struct net_device *dev, struct fib_result *res)
2263{
2264 /* Multicast recognition logic is moved from route cache to here.
2265 The problem was that too many Ethernet cards have broken/missing
2266 hardware multicast filters :-( As result the host on multicasting
2267 network acquires a lot of useless route cache entries, sort of
2268 SDR messages from all the world. Now we try to get rid of them.
2269 Really, provided software IP multicast filter is organized
2270 reasonably (at least, hashed), it does not result in a slowdown
2271 comparing with route cache reject entries.
2272 Note, that multicast routers are not affected, because
2273 route cache entry is created eventually.
2274 */
2275 if (ipv4_is_multicast(daddr)) {
2276 struct in_device *in_dev = __in_dev_get_rcu(dev);
2277 int our = 0;
2278 int err = -EINVAL;
2279
2280 if (!in_dev)
2281 return err;
2282 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2283 ip_hdr(skb)->protocol);
2284
2285 /* check l3 master if no match yet */
2286 if (!our && netif_is_l3_slave(dev)) {
2287 struct in_device *l3_in_dev;
2288
2289 l3_in_dev = __in_dev_get_rcu(skb->dev);
2290 if (l3_in_dev)
2291 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2292 ip_hdr(skb)->protocol);
2293 }
2294
2295 if (our
2296#ifdef CONFIG_IP_MROUTE
2297 ||
2298 (!ipv4_is_local_multicast(daddr) &&
2299 IN_DEV_MFORWARD(in_dev))
2300#endif
2301 ) {
2302 err = ip_route_input_mc(skb, daddr, saddr,
2303 tos, dev, our);
2304 }
2305 return err;
2306 }
2307
2308 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2309}
2310
2311/* called with rcu_read_lock() */
2312static struct rtable *__mkroute_output(const struct fib_result *res,
2313 const struct flowi4 *fl4, int orig_oif,
2314 struct net_device *dev_out,
2315 unsigned int flags)
2316{
2317 struct fib_info *fi = res->fi;
2318 struct fib_nh_exception *fnhe;
2319 struct in_device *in_dev;
2320 u16 type = res->type;
2321 struct rtable *rth;
2322 bool do_cache;
2323
2324 in_dev = __in_dev_get_rcu(dev_out);
2325 if (!in_dev)
2326 return ERR_PTR(-EINVAL);
2327
2328 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2329 if (ipv4_is_loopback(fl4->saddr) &&
2330 !(dev_out->flags & IFF_LOOPBACK) &&
2331 !netif_is_l3_master(dev_out))
2332 return ERR_PTR(-EINVAL);
2333
2334 if (ipv4_is_lbcast(fl4->daddr))
2335 type = RTN_BROADCAST;
2336 else if (ipv4_is_multicast(fl4->daddr))
2337 type = RTN_MULTICAST;
2338 else if (ipv4_is_zeronet(fl4->daddr))
2339 return ERR_PTR(-EINVAL);
2340
2341 if (dev_out->flags & IFF_LOOPBACK)
2342 flags |= RTCF_LOCAL;
2343
2344 do_cache = true;
2345 if (type == RTN_BROADCAST) {
2346 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2347 fi = NULL;
2348 } else if (type == RTN_MULTICAST) {
2349 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2350 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2351 fl4->flowi4_proto))
2352 flags &= ~RTCF_LOCAL;
2353 else
2354 do_cache = false;
2355 /* If multicast route do not exist use
2356 * default one, but do not gateway in this case.
2357 * Yes, it is hack.
2358 */
2359 if (fi && res->prefixlen < 4)
2360 fi = NULL;
2361 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2362 (orig_oif != dev_out->ifindex)) {
2363 /* For local routes that require a particular output interface
2364 * we do not want to cache the result. Caching the result
2365 * causes incorrect behaviour when there are multiple source
2366 * addresses on the interface, the end result being that if the
2367 * intended recipient is waiting on that interface for the
2368 * packet he won't receive it because it will be delivered on
2369 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2370 * be set to the loopback interface as well.
2371 */
2372 do_cache = false;
2373 }
2374
2375 fnhe = NULL;
2376 do_cache &= fi != NULL;
2377 if (fi) {
2378 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2379 struct rtable __rcu **prth;
2380
2381 fnhe = find_exception(nhc, fl4->daddr);
2382 if (!do_cache)
2383 goto add;
2384 if (fnhe) {
2385 prth = &fnhe->fnhe_rth_output;
2386 } else {
2387 if (unlikely(fl4->flowi4_flags &
2388 FLOWI_FLAG_KNOWN_NH &&
2389 !(nhc->nhc_gw_family &&
2390 nhc->nhc_scope == RT_SCOPE_LINK))) {
2391 do_cache = false;
2392 goto add;
2393 }
2394 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2395 }
2396 rth = rcu_dereference(*prth);
2397 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2398 return rth;
2399 }
2400
2401add:
2402 rth = rt_dst_alloc(dev_out, flags, type,
2403 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2404 IN_DEV_CONF_GET(in_dev, NOXFRM),
2405 do_cache);
2406 if (!rth)
2407 return ERR_PTR(-ENOBUFS);
2408
2409 rth->rt_iif = orig_oif;
2410
2411 RT_CACHE_STAT_INC(out_slow_tot);
2412
2413 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2414 if (flags & RTCF_LOCAL &&
2415 !(dev_out->flags & IFF_LOOPBACK)) {
2416 rth->dst.output = ip_mc_output;
2417 RT_CACHE_STAT_INC(out_slow_mc);
2418 }
2419#ifdef CONFIG_IP_MROUTE
2420 if (type == RTN_MULTICAST) {
2421 if (IN_DEV_MFORWARD(in_dev) &&
2422 !ipv4_is_local_multicast(fl4->daddr)) {
2423 rth->dst.input = ip_mr_input;
2424 rth->dst.output = ip_mc_output;
2425 }
2426 }
2427#endif
2428 }
2429
2430 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2431 lwtunnel_set_redirect(&rth->dst);
2432
2433 return rth;
2434}
2435
2436/*
2437 * Major route resolver routine.
2438 */
2439
2440struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2441 const struct sk_buff *skb)
2442{
2443 __u8 tos = RT_FL_TOS(fl4);
2444 struct fib_result res = {
2445 .type = RTN_UNSPEC,
2446 .fi = NULL,
2447 .table = NULL,
2448 .tclassid = 0,
2449 };
2450 struct rtable *rth;
2451
2452 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2453 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2454 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2455 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2456
2457 rcu_read_lock();
2458 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2459 rcu_read_unlock();
2460
2461 return rth;
2462}
2463EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2464
2465struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2466 struct fib_result *res,
2467 const struct sk_buff *skb)
2468{
2469 struct net_device *dev_out = NULL;
2470 int orig_oif = fl4->flowi4_oif;
2471 unsigned int flags = 0;
2472 struct rtable *rth;
2473 int err;
2474
2475 if (fl4->saddr) {
2476 if (ipv4_is_multicast(fl4->saddr) ||
2477 ipv4_is_lbcast(fl4->saddr) ||
2478 ipv4_is_zeronet(fl4->saddr)) {
2479 rth = ERR_PTR(-EINVAL);
2480 goto out;
2481 }
2482
2483 rth = ERR_PTR(-ENETUNREACH);
2484
2485 /* I removed check for oif == dev_out->oif here.
2486 It was wrong for two reasons:
2487 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2488 is assigned to multiple interfaces.
2489 2. Moreover, we are allowed to send packets with saddr
2490 of another iface. --ANK
2491 */
2492
2493 if (fl4->flowi4_oif == 0 &&
2494 (ipv4_is_multicast(fl4->daddr) ||
2495 ipv4_is_lbcast(fl4->daddr))) {
2496 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2497 dev_out = __ip_dev_find(net, fl4->saddr, false);
2498 if (!dev_out)
2499 goto out;
2500
2501 /* Special hack: user can direct multicasts
2502 and limited broadcast via necessary interface
2503 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2504 This hack is not just for fun, it allows
2505 vic,vat and friends to work.
2506 They bind socket to loopback, set ttl to zero
2507 and expect that it will work.
2508 From the viewpoint of routing cache they are broken,
2509 because we are not allowed to build multicast path
2510 with loopback source addr (look, routing cache
2511 cannot know, that ttl is zero, so that packet
2512 will not leave this host and route is valid).
2513 Luckily, this hack is good workaround.
2514 */
2515
2516 fl4->flowi4_oif = dev_out->ifindex;
2517 goto make_route;
2518 }
2519
2520 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2521 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2522 if (!__ip_dev_find(net, fl4->saddr, false))
2523 goto out;
2524 }
2525 }
2526
2527
2528 if (fl4->flowi4_oif) {
2529 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2530 rth = ERR_PTR(-ENODEV);
2531 if (!dev_out)
2532 goto out;
2533
2534 /* RACE: Check return value of inet_select_addr instead. */
2535 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2536 rth = ERR_PTR(-ENETUNREACH);
2537 goto out;
2538 }
2539 if (ipv4_is_local_multicast(fl4->daddr) ||
2540 ipv4_is_lbcast(fl4->daddr) ||
2541 fl4->flowi4_proto == IPPROTO_IGMP) {
2542 if (!fl4->saddr)
2543 fl4->saddr = inet_select_addr(dev_out, 0,
2544 RT_SCOPE_LINK);
2545 goto make_route;
2546 }
2547 if (!fl4->saddr) {
2548 if (ipv4_is_multicast(fl4->daddr))
2549 fl4->saddr = inet_select_addr(dev_out, 0,
2550 fl4->flowi4_scope);
2551 else if (!fl4->daddr)
2552 fl4->saddr = inet_select_addr(dev_out, 0,
2553 RT_SCOPE_HOST);
2554 }
2555 }
2556
2557 if (!fl4->daddr) {
2558 fl4->daddr = fl4->saddr;
2559 if (!fl4->daddr)
2560 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2561 dev_out = net->loopback_dev;
2562 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2563 res->type = RTN_LOCAL;
2564 flags |= RTCF_LOCAL;
2565 goto make_route;
2566 }
2567
2568 err = fib_lookup(net, fl4, res, 0);
2569 if (err) {
2570 res->fi = NULL;
2571 res->table = NULL;
2572 if (fl4->flowi4_oif &&
2573 (ipv4_is_multicast(fl4->daddr) ||
2574 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2575 /* Apparently, routing tables are wrong. Assume,
2576 that the destination is on link.
2577
2578 WHY? DW.
2579 Because we are allowed to send to iface
2580 even if it has NO routes and NO assigned
2581 addresses. When oif is specified, routing
2582 tables are looked up with only one purpose:
2583 to catch if destination is gatewayed, rather than
2584 direct. Moreover, if MSG_DONTROUTE is set,
2585 we send packet, ignoring both routing tables
2586 and ifaddr state. --ANK
2587
2588
2589 We could make it even if oif is unknown,
2590 likely IPv6, but we do not.
2591 */
2592
2593 if (fl4->saddr == 0)
2594 fl4->saddr = inet_select_addr(dev_out, 0,
2595 RT_SCOPE_LINK);
2596 res->type = RTN_UNICAST;
2597 goto make_route;
2598 }
2599 rth = ERR_PTR(err);
2600 goto out;
2601 }
2602
2603 if (res->type == RTN_LOCAL) {
2604 if (!fl4->saddr) {
2605 if (res->fi->fib_prefsrc)
2606 fl4->saddr = res->fi->fib_prefsrc;
2607 else
2608 fl4->saddr = fl4->daddr;
2609 }
2610
2611 /* L3 master device is the loopback for that domain */
2612 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2613 net->loopback_dev;
2614
2615 /* make sure orig_oif points to fib result device even
2616 * though packet rx/tx happens over loopback or l3mdev
2617 */
2618 orig_oif = FIB_RES_OIF(*res);
2619
2620 fl4->flowi4_oif = dev_out->ifindex;
2621 flags |= RTCF_LOCAL;
2622 goto make_route;
2623 }
2624
2625 fib_select_path(net, res, fl4, skb);
2626
2627 dev_out = FIB_RES_DEV(*res);
2628 fl4->flowi4_oif = dev_out->ifindex;
2629
2630
2631make_route:
2632 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2633
2634out:
2635 return rth;
2636}
2637
2638static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2639{
2640 return NULL;
2641}
2642
2643static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2644{
2645 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2646
2647 return mtu ? : dst->dev->mtu;
2648}
2649
2650static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2651 struct sk_buff *skb, u32 mtu)
2652{
2653}
2654
2655static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2656 struct sk_buff *skb)
2657{
2658}
2659
2660static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2661 unsigned long old)
2662{
2663 return NULL;
2664}
2665
2666static struct dst_ops ipv4_dst_blackhole_ops = {
2667 .family = AF_INET,
2668 .check = ipv4_blackhole_dst_check,
2669 .mtu = ipv4_blackhole_mtu,
2670 .default_advmss = ipv4_default_advmss,
2671 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2672 .redirect = ipv4_rt_blackhole_redirect,
2673 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2674 .neigh_lookup = ipv4_neigh_lookup,
2675};
2676
2677struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2678{
2679 struct rtable *ort = (struct rtable *) dst_orig;
2680 struct rtable *rt;
2681
2682 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2683 if (rt) {
2684 struct dst_entry *new = &rt->dst;
2685
2686 new->__use = 1;
2687 new->input = dst_discard;
2688 new->output = dst_discard_out;
2689
2690 new->dev = net->loopback_dev;
2691 if (new->dev)
2692 dev_hold(new->dev);
2693
2694 rt->rt_is_input = ort->rt_is_input;
2695 rt->rt_iif = ort->rt_iif;
2696 rt->rt_pmtu = ort->rt_pmtu;
2697 rt->rt_mtu_locked = ort->rt_mtu_locked;
2698
2699 rt->rt_genid = rt_genid_ipv4(net);
2700 rt->rt_flags = ort->rt_flags;
2701 rt->rt_type = ort->rt_type;
2702 rt->rt_uses_gateway = ort->rt_uses_gateway;
2703 rt->rt_gw_family = ort->rt_gw_family;
2704 if (rt->rt_gw_family == AF_INET)
2705 rt->rt_gw4 = ort->rt_gw4;
2706 else if (rt->rt_gw_family == AF_INET6)
2707 rt->rt_gw6 = ort->rt_gw6;
2708
2709 INIT_LIST_HEAD(&rt->rt_uncached);
2710 }
2711
2712 dst_release(dst_orig);
2713
2714 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2715}
2716
2717struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2718 const struct sock *sk)
2719{
2720 struct rtable *rt = __ip_route_output_key(net, flp4);
2721
2722 if (IS_ERR(rt))
2723 return rt;
2724
2725 if (flp4->flowi4_proto)
2726 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2727 flowi4_to_flowi(flp4),
2728 sk, 0);
2729
2730 return rt;
2731}
2732EXPORT_SYMBOL_GPL(ip_route_output_flow);
2733
2734/* called with rcu_read_lock held */
2735static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2736 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2737 struct sk_buff *skb, u32 portid, u32 seq,
2738 unsigned int flags)
2739{
2740 struct rtmsg *r;
2741 struct nlmsghdr *nlh;
2742 unsigned long expires = 0;
2743 u32 error;
2744 u32 metrics[RTAX_MAX];
2745
2746 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2747 if (!nlh)
2748 return -EMSGSIZE;
2749
2750 r = nlmsg_data(nlh);
2751 r->rtm_family = AF_INET;
2752 r->rtm_dst_len = 32;
2753 r->rtm_src_len = 0;
2754 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2755 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2756 if (nla_put_u32(skb, RTA_TABLE, table_id))
2757 goto nla_put_failure;
2758 r->rtm_type = rt->rt_type;
2759 r->rtm_scope = RT_SCOPE_UNIVERSE;
2760 r->rtm_protocol = RTPROT_UNSPEC;
2761 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2762 if (rt->rt_flags & RTCF_NOTIFY)
2763 r->rtm_flags |= RTM_F_NOTIFY;
2764 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2765 r->rtm_flags |= RTCF_DOREDIRECT;
2766
2767 if (nla_put_in_addr(skb, RTA_DST, dst))
2768 goto nla_put_failure;
2769 if (src) {
2770 r->rtm_src_len = 32;
2771 if (nla_put_in_addr(skb, RTA_SRC, src))
2772 goto nla_put_failure;
2773 }
2774 if (rt->dst.dev &&
2775 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2776 goto nla_put_failure;
2777#ifdef CONFIG_IP_ROUTE_CLASSID
2778 if (rt->dst.tclassid &&
2779 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2780 goto nla_put_failure;
2781#endif
2782 if (fl4 && !rt_is_input_route(rt) &&
2783 fl4->saddr != src) {
2784 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2785 goto nla_put_failure;
2786 }
2787 if (rt->rt_uses_gateway) {
2788 if (rt->rt_gw_family == AF_INET &&
2789 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2790 goto nla_put_failure;
2791 } else if (rt->rt_gw_family == AF_INET6) {
2792 int alen = sizeof(struct in6_addr);
2793 struct nlattr *nla;
2794 struct rtvia *via;
2795
2796 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2797 if (!nla)
2798 goto nla_put_failure;
2799
2800 via = nla_data(nla);
2801 via->rtvia_family = AF_INET6;
2802 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2803 }
2804 }
2805
2806 expires = rt->dst.expires;
2807 if (expires) {
2808 unsigned long now = jiffies;
2809
2810 if (time_before(now, expires))
2811 expires -= now;
2812 else
2813 expires = 0;
2814 }
2815
2816 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2817 if (rt->rt_pmtu && expires)
2818 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2819 if (rt->rt_mtu_locked && expires)
2820 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2821 if (rtnetlink_put_metrics(skb, metrics) < 0)
2822 goto nla_put_failure;
2823
2824 if (fl4) {
2825 if (fl4->flowi4_mark &&
2826 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2827 goto nla_put_failure;
2828
2829 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2830 nla_put_u32(skb, RTA_UID,
2831 from_kuid_munged(current_user_ns(),
2832 fl4->flowi4_uid)))
2833 goto nla_put_failure;
2834
2835 if (rt_is_input_route(rt)) {
2836#ifdef CONFIG_IP_MROUTE
2837 if (ipv4_is_multicast(dst) &&
2838 !ipv4_is_local_multicast(dst) &&
2839 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2840 int err = ipmr_get_route(net, skb,
2841 fl4->saddr, fl4->daddr,
2842 r, portid);
2843
2844 if (err <= 0) {
2845 if (err == 0)
2846 return 0;
2847 goto nla_put_failure;
2848 }
2849 } else
2850#endif
2851 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2852 goto nla_put_failure;
2853 }
2854 }
2855
2856 error = rt->dst.error;
2857
2858 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2859 goto nla_put_failure;
2860
2861 nlmsg_end(skb, nlh);
2862 return 0;
2863
2864nla_put_failure:
2865 nlmsg_cancel(skb, nlh);
2866 return -EMSGSIZE;
2867}
2868
2869static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
2870 struct netlink_callback *cb, u32 table_id,
2871 struct fnhe_hash_bucket *bucket, int genid,
2872 int *fa_index, int fa_start, unsigned int flags)
2873{
2874 int i;
2875
2876 for (i = 0; i < FNHE_HASH_SIZE; i++) {
2877 struct fib_nh_exception *fnhe;
2878
2879 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
2880 fnhe = rcu_dereference(fnhe->fnhe_next)) {
2881 struct rtable *rt;
2882 int err;
2883
2884 if (*fa_index < fa_start)
2885 goto next;
2886
2887 if (fnhe->fnhe_genid != genid)
2888 goto next;
2889
2890 if (fnhe->fnhe_expires &&
2891 time_after(jiffies, fnhe->fnhe_expires))
2892 goto next;
2893
2894 rt = rcu_dereference(fnhe->fnhe_rth_input);
2895 if (!rt)
2896 rt = rcu_dereference(fnhe->fnhe_rth_output);
2897 if (!rt)
2898 goto next;
2899
2900 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
2901 table_id, NULL, skb,
2902 NETLINK_CB(cb->skb).portid,
2903 cb->nlh->nlmsg_seq, flags);
2904 if (err)
2905 return err;
2906next:
2907 (*fa_index)++;
2908 }
2909 }
2910
2911 return 0;
2912}
2913
2914int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
2915 u32 table_id, struct fib_info *fi,
2916 int *fa_index, int fa_start, unsigned int flags)
2917{
2918 struct net *net = sock_net(cb->skb->sk);
2919 int nhsel, genid = fnhe_genid(net);
2920
2921 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
2922 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
2923 struct fnhe_hash_bucket *bucket;
2924 int err;
2925
2926 if (nhc->nhc_flags & RTNH_F_DEAD)
2927 continue;
2928
2929 rcu_read_lock();
2930 bucket = rcu_dereference(nhc->nhc_exceptions);
2931 err = 0;
2932 if (bucket)
2933 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
2934 genid, fa_index, fa_start,
2935 flags);
2936 rcu_read_unlock();
2937 if (err)
2938 return err;
2939 }
2940
2941 return 0;
2942}
2943
2944static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2945 u8 ip_proto, __be16 sport,
2946 __be16 dport)
2947{
2948 struct sk_buff *skb;
2949 struct iphdr *iph;
2950
2951 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2952 if (!skb)
2953 return NULL;
2954
2955 /* Reserve room for dummy headers, this skb can pass
2956 * through good chunk of routing engine.
2957 */
2958 skb_reset_mac_header(skb);
2959 skb_reset_network_header(skb);
2960 skb->protocol = htons(ETH_P_IP);
2961 iph = skb_put(skb, sizeof(struct iphdr));
2962 iph->protocol = ip_proto;
2963 iph->saddr = src;
2964 iph->daddr = dst;
2965 iph->version = 0x4;
2966 iph->frag_off = 0;
2967 iph->ihl = 0x5;
2968 skb_set_transport_header(skb, skb->len);
2969
2970 switch (iph->protocol) {
2971 case IPPROTO_UDP: {
2972 struct udphdr *udph;
2973
2974 udph = skb_put_zero(skb, sizeof(struct udphdr));
2975 udph->source = sport;
2976 udph->dest = dport;
2977 udph->len = sizeof(struct udphdr);
2978 udph->check = 0;
2979 break;
2980 }
2981 case IPPROTO_TCP: {
2982 struct tcphdr *tcph;
2983
2984 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2985 tcph->source = sport;
2986 tcph->dest = dport;
2987 tcph->doff = sizeof(struct tcphdr) / 4;
2988 tcph->rst = 1;
2989 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2990 src, dst, 0);
2991 break;
2992 }
2993 case IPPROTO_ICMP: {
2994 struct icmphdr *icmph;
2995
2996 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2997 icmph->type = ICMP_ECHO;
2998 icmph->code = 0;
2999 }
3000 }
3001
3002 return skb;
3003}
3004
3005static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3006 const struct nlmsghdr *nlh,
3007 struct nlattr **tb,
3008 struct netlink_ext_ack *extack)
3009{
3010 struct rtmsg *rtm;
3011 int i, err;
3012
3013 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3014 NL_SET_ERR_MSG(extack,
3015 "ipv4: Invalid header for route get request");
3016 return -EINVAL;
3017 }
3018
3019 if (!netlink_strict_get_check(skb))
3020 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3021 rtm_ipv4_policy, extack);
3022
3023 rtm = nlmsg_data(nlh);
3024 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3025 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3026 rtm->rtm_table || rtm->rtm_protocol ||
3027 rtm->rtm_scope || rtm->rtm_type) {
3028 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3029 return -EINVAL;
3030 }
3031
3032 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3033 RTM_F_LOOKUP_TABLE |
3034 RTM_F_FIB_MATCH)) {
3035 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3036 return -EINVAL;
3037 }
3038
3039 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3040 rtm_ipv4_policy, extack);
3041 if (err)
3042 return err;
3043
3044 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3045 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3046 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3047 return -EINVAL;
3048 }
3049
3050 for (i = 0; i <= RTA_MAX; i++) {
3051 if (!tb[i])
3052 continue;
3053
3054 switch (i) {
3055 case RTA_IIF:
3056 case RTA_OIF:
3057 case RTA_SRC:
3058 case RTA_DST:
3059 case RTA_IP_PROTO:
3060 case RTA_SPORT:
3061 case RTA_DPORT:
3062 case RTA_MARK:
3063 case RTA_UID:
3064 break;
3065 default:
3066 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3067 return -EINVAL;
3068 }
3069 }
3070
3071 return 0;
3072}
3073
3074static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3075 struct netlink_ext_ack *extack)
3076{
3077 struct net *net = sock_net(in_skb->sk);
3078 struct nlattr *tb[RTA_MAX+1];
3079 u32 table_id = RT_TABLE_MAIN;
3080 __be16 sport = 0, dport = 0;
3081 struct fib_result res = {};
3082 u8 ip_proto = IPPROTO_UDP;
3083 struct rtable *rt = NULL;
3084 struct sk_buff *skb;
3085 struct rtmsg *rtm;
3086 struct flowi4 fl4 = {};
3087 __be32 dst = 0;
3088 __be32 src = 0;
3089 kuid_t uid;
3090 u32 iif;
3091 int err;
3092 int mark;
3093
3094 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3095 if (err < 0)
3096 return err;
3097
3098 rtm = nlmsg_data(nlh);
3099 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3100 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3101 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3102 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3103 if (tb[RTA_UID])
3104 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3105 else
3106 uid = (iif ? INVALID_UID : current_uid());
3107
3108 if (tb[RTA_IP_PROTO]) {
3109 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3110 &ip_proto, AF_INET, extack);
3111 if (err)
3112 return err;
3113 }
3114
3115 if (tb[RTA_SPORT])
3116 sport = nla_get_be16(tb[RTA_SPORT]);
3117
3118 if (tb[RTA_DPORT])
3119 dport = nla_get_be16(tb[RTA_DPORT]);
3120
3121 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3122 if (!skb)
3123 return -ENOBUFS;
3124
3125 fl4.daddr = dst;
3126 fl4.saddr = src;
3127 fl4.flowi4_tos = rtm->rtm_tos;
3128 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3129 fl4.flowi4_mark = mark;
3130 fl4.flowi4_uid = uid;
3131 if (sport)
3132 fl4.fl4_sport = sport;
3133 if (dport)
3134 fl4.fl4_dport = dport;
3135 fl4.flowi4_proto = ip_proto;
3136
3137 rcu_read_lock();
3138
3139 if (iif) {
3140 struct net_device *dev;
3141
3142 dev = dev_get_by_index_rcu(net, iif);
3143 if (!dev) {
3144 err = -ENODEV;
3145 goto errout_rcu;
3146 }
3147
3148 fl4.flowi4_iif = iif; /* for rt_fill_info */
3149 skb->dev = dev;
3150 skb->mark = mark;
3151 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
3152 dev, &res);
3153
3154 rt = skb_rtable(skb);
3155 if (err == 0 && rt->dst.error)
3156 err = -rt->dst.error;
3157 } else {
3158 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3159 skb->dev = net->loopback_dev;
3160 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3161 err = 0;
3162 if (IS_ERR(rt))
3163 err = PTR_ERR(rt);
3164 else
3165 skb_dst_set(skb, &rt->dst);
3166 }
3167
3168 if (err)
3169 goto errout_rcu;
3170
3171 if (rtm->rtm_flags & RTM_F_NOTIFY)
3172 rt->rt_flags |= RTCF_NOTIFY;
3173
3174 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3175 table_id = res.table ? res.table->tb_id : 0;
3176
3177 /* reset skb for netlink reply msg */
3178 skb_trim(skb, 0);
3179 skb_reset_network_header(skb);
3180 skb_reset_transport_header(skb);
3181 skb_reset_mac_header(skb);
3182
3183 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3184 if (!res.fi) {
3185 err = fib_props[res.type].error;
3186 if (!err)
3187 err = -EHOSTUNREACH;
3188 goto errout_rcu;
3189 }
3190 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3191 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
3192 rt->rt_type, res.prefix, res.prefixlen,
3193 fl4.flowi4_tos, res.fi, 0);
3194 } else {
3195 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3196 NETLINK_CB(in_skb).portid,
3197 nlh->nlmsg_seq, 0);
3198 }
3199 if (err < 0)
3200 goto errout_rcu;
3201
3202 rcu_read_unlock();
3203
3204 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3205
3206errout_free:
3207 return err;
3208errout_rcu:
3209 rcu_read_unlock();
3210 kfree_skb(skb);
3211 goto errout_free;
3212}
3213
3214void ip_rt_multicast_event(struct in_device *in_dev)
3215{
3216 rt_cache_flush(dev_net(in_dev->dev));
3217}
3218
3219#ifdef CONFIG_SYSCTL
3220static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3221static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3222static int ip_rt_gc_elasticity __read_mostly = 8;
3223static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3224
3225static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3226 void __user *buffer,
3227 size_t *lenp, loff_t *ppos)
3228{
3229 struct net *net = (struct net *)__ctl->extra1;
3230
3231 if (write) {
3232 rt_cache_flush(net);
3233 fnhe_genid_bump(net);
3234 return 0;
3235 }
3236
3237 return -EINVAL;
3238}
3239
3240static struct ctl_table ipv4_route_table[] = {
3241 {
3242 .procname = "gc_thresh",
3243 .data = &ipv4_dst_ops.gc_thresh,
3244 .maxlen = sizeof(int),
3245 .mode = 0644,
3246 .proc_handler = proc_dointvec,
3247 },
3248 {
3249 .procname = "max_size",
3250 .data = &ip_rt_max_size,
3251 .maxlen = sizeof(int),
3252 .mode = 0644,
3253 .proc_handler = proc_dointvec,
3254 },
3255 {
3256 /* Deprecated. Use gc_min_interval_ms */
3257
3258 .procname = "gc_min_interval",
3259 .data = &ip_rt_gc_min_interval,
3260 .maxlen = sizeof(int),
3261 .mode = 0644,
3262 .proc_handler = proc_dointvec_jiffies,
3263 },
3264 {
3265 .procname = "gc_min_interval_ms",
3266 .data = &ip_rt_gc_min_interval,
3267 .maxlen = sizeof(int),
3268 .mode = 0644,
3269 .proc_handler = proc_dointvec_ms_jiffies,
3270 },
3271 {
3272 .procname = "gc_timeout",
3273 .data = &ip_rt_gc_timeout,
3274 .maxlen = sizeof(int),
3275 .mode = 0644,
3276 .proc_handler = proc_dointvec_jiffies,
3277 },
3278 {
3279 .procname = "gc_interval",
3280 .data = &ip_rt_gc_interval,
3281 .maxlen = sizeof(int),
3282 .mode = 0644,
3283 .proc_handler = proc_dointvec_jiffies,
3284 },
3285 {
3286 .procname = "redirect_load",
3287 .data = &ip_rt_redirect_load,
3288 .maxlen = sizeof(int),
3289 .mode = 0644,
3290 .proc_handler = proc_dointvec,
3291 },
3292 {
3293 .procname = "redirect_number",
3294 .data = &ip_rt_redirect_number,
3295 .maxlen = sizeof(int),
3296 .mode = 0644,
3297 .proc_handler = proc_dointvec,
3298 },
3299 {
3300 .procname = "redirect_silence",
3301 .data = &ip_rt_redirect_silence,
3302 .maxlen = sizeof(int),
3303 .mode = 0644,
3304 .proc_handler = proc_dointvec,
3305 },
3306 {
3307 .procname = "error_cost",
3308 .data = &ip_rt_error_cost,
3309 .maxlen = sizeof(int),
3310 .mode = 0644,
3311 .proc_handler = proc_dointvec,
3312 },
3313 {
3314 .procname = "error_burst",
3315 .data = &ip_rt_error_burst,
3316 .maxlen = sizeof(int),
3317 .mode = 0644,
3318 .proc_handler = proc_dointvec,
3319 },
3320 {
3321 .procname = "gc_elasticity",
3322 .data = &ip_rt_gc_elasticity,
3323 .maxlen = sizeof(int),
3324 .mode = 0644,
3325 .proc_handler = proc_dointvec,
3326 },
3327 {
3328 .procname = "mtu_expires",
3329 .data = &ip_rt_mtu_expires,
3330 .maxlen = sizeof(int),
3331 .mode = 0644,
3332 .proc_handler = proc_dointvec_jiffies,
3333 },
3334 {
3335 .procname = "min_pmtu",
3336 .data = &ip_rt_min_pmtu,
3337 .maxlen = sizeof(int),
3338 .mode = 0644,
3339 .proc_handler = proc_dointvec_minmax,
3340 .extra1 = &ip_min_valid_pmtu,
3341 },
3342 {
3343 .procname = "min_adv_mss",
3344 .data = &ip_rt_min_advmss,
3345 .maxlen = sizeof(int),
3346 .mode = 0644,
3347 .proc_handler = proc_dointvec,
3348 },
3349 { }
3350};
3351
3352static const char ipv4_route_flush_procname[] = "flush";
3353
3354static struct ctl_table ipv4_route_flush_table[] = {
3355 {
3356 .procname = ipv4_route_flush_procname,
3357 .maxlen = sizeof(int),
3358 .mode = 0200,
3359 .proc_handler = ipv4_sysctl_rtcache_flush,
3360 },
3361 { },
3362};
3363
3364static __net_init int sysctl_route_net_init(struct net *net)
3365{
3366 struct ctl_table *tbl;
3367
3368 tbl = ipv4_route_flush_table;
3369 if (!net_eq(net, &init_net)) {
3370 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3371 if (!tbl)
3372 goto err_dup;
3373
3374 /* Don't export non-whitelisted sysctls to unprivileged users */
3375 if (net->user_ns != &init_user_ns) {
3376 if (tbl[0].procname != ipv4_route_flush_procname)
3377 tbl[0].procname = NULL;
3378 }
3379 }
3380 tbl[0].extra1 = net;
3381
3382 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3383 if (!net->ipv4.route_hdr)
3384 goto err_reg;
3385 return 0;
3386
3387err_reg:
3388 if (tbl != ipv4_route_flush_table)
3389 kfree(tbl);
3390err_dup:
3391 return -ENOMEM;
3392}
3393
3394static __net_exit void sysctl_route_net_exit(struct net *net)
3395{
3396 struct ctl_table *tbl;
3397
3398 tbl = net->ipv4.route_hdr->ctl_table_arg;
3399 unregister_net_sysctl_table(net->ipv4.route_hdr);
3400 BUG_ON(tbl == ipv4_route_flush_table);
3401 kfree(tbl);
3402}
3403
3404static __net_initdata struct pernet_operations sysctl_route_ops = {
3405 .init = sysctl_route_net_init,
3406 .exit = sysctl_route_net_exit,
3407};
3408#endif
3409
3410static __net_init int rt_genid_init(struct net *net)
3411{
3412 atomic_set(&net->ipv4.rt_genid, 0);
3413 atomic_set(&net->fnhe_genid, 0);
3414 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3415 return 0;
3416}
3417
3418static __net_initdata struct pernet_operations rt_genid_ops = {
3419 .init = rt_genid_init,
3420};
3421
3422static int __net_init ipv4_inetpeer_init(struct net *net)
3423{
3424 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3425
3426 if (!bp)
3427 return -ENOMEM;
3428 inet_peer_base_init(bp);
3429 net->ipv4.peers = bp;
3430 return 0;
3431}
3432
3433static void __net_exit ipv4_inetpeer_exit(struct net *net)
3434{
3435 struct inet_peer_base *bp = net->ipv4.peers;
3436
3437 net->ipv4.peers = NULL;
3438 inetpeer_invalidate_tree(bp);
3439 kfree(bp);
3440}
3441
3442static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3443 .init = ipv4_inetpeer_init,
3444 .exit = ipv4_inetpeer_exit,
3445};
3446
3447#ifdef CONFIG_IP_ROUTE_CLASSID
3448struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3449#endif /* CONFIG_IP_ROUTE_CLASSID */
3450
3451int __init ip_rt_init(void)
3452{
3453 int cpu;
3454
3455 ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3456 GFP_KERNEL);
3457 if (!ip_idents)
3458 panic("IP: failed to allocate ip_idents\n");
3459
3460 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3461
3462 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3463 if (!ip_tstamps)
3464 panic("IP: failed to allocate ip_tstamps\n");
3465
3466 for_each_possible_cpu(cpu) {
3467 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3468
3469 INIT_LIST_HEAD(&ul->head);
3470 spin_lock_init(&ul->lock);
3471 }
3472#ifdef CONFIG_IP_ROUTE_CLASSID
3473 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3474 if (!ip_rt_acct)
3475 panic("IP: failed to allocate ip_rt_acct\n");
3476#endif
3477
3478 ipv4_dst_ops.kmem_cachep =
3479 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3480 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3481
3482 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3483
3484 if (dst_entries_init(&ipv4_dst_ops) < 0)
3485 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3486
3487 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3488 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3489
3490 ipv4_dst_ops.gc_thresh = ~0;
3491 ip_rt_max_size = INT_MAX;
3492
3493 devinet_init();
3494 ip_fib_init();
3495
3496 if (ip_rt_proc_init())
3497 pr_err("Unable to create route proc files\n");
3498#ifdef CONFIG_XFRM
3499 xfrm_init();
3500 xfrm4_init();
3501#endif
3502 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3503 RTNL_FLAG_DOIT_UNLOCKED);
3504
3505#ifdef CONFIG_SYSCTL
3506 register_pernet_subsys(&sysctl_route_ops);
3507#endif
3508 register_pernet_subsys(&rt_genid_ops);
3509 register_pernet_subsys(&ipv4_inetpeer_ops);
3510 return 0;
3511}
3512
3513#ifdef CONFIG_SYSCTL
3514/*
3515 * We really need to sanitize the damn ipv4 init order, then all
3516 * this nonsense will go away.
3517 */
3518void __init ip_static_sysctl_init(void)
3519{
3520 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3521}
3522#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/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