projects / GitHub / LineageOS / android_kernel_motorola_exynos9610.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
defconfig: exynos9610: Re-add dropped Wi-Fi AP options lost
[GitHub/LineageOS/android_kernel_motorola_exynos9610.git] / net / ipv4 / route.c
... / ...
CommitLineData
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * ROUTE - implementation of the IP router.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13 *
14 * Fixes:
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
24 * clamper.
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
39 *
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
58 *
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
63 */
64
65#define pr_fmt(fmt) "IPv4: " fmt
66
67#include <linux/module.h>
68#include <linux/uaccess.h>
69#include <linux/bitops.h>
70#include <linux/types.h>
71#include <linux/kernel.h>
72#include <linux/mm.h>
73#include <linux/string.h>
74#include <linux/socket.h>
75#include <linux/sockios.h>
76#include <linux/errno.h>
77#include <linux/in.h>
78#include <linux/inet.h>
79#include <linux/netdevice.h>
80#include <linux/proc_fs.h>
81#include <linux/init.h>
82#include <linux/skbuff.h>
83#include <linux/inetdevice.h>
84#include <linux/igmp.h>
85#include <linux/pkt_sched.h>
86#include <linux/mroute.h>
87#include <linux/netfilter_ipv4.h>
88#include <linux/random.h>
89#include <linux/rcupdate.h>
90#include <linux/times.h>
91#include <linux/slab.h>
92#include <linux/jhash.h>
93#include <net/dst.h>
94#include <net/dst_metadata.h>
95#include <net/net_namespace.h>
96#include <net/protocol.h>
97#include <net/ip.h>
98#include <net/route.h>
99#include <net/inetpeer.h>
100#include <net/sock.h>
101#include <net/ip_fib.h>
102#include <net/arp.h>
103#include <net/tcp.h>
104#include <net/icmp.h>
105#include <net/xfrm.h>
106#include <net/lwtunnel.h>
107#include <net/netevent.h>
108#include <net/rtnetlink.h>
109#ifdef CONFIG_SYSCTL
110#include <linux/sysctl.h>
111#include <linux/kmemleak.h>
112#endif
113#include <net/secure_seq.h>
114#include <net/ip_tunnels.h>
115#include <net/l3mdev.h>
116
117#include "fib_lookup.h"
118
119#define RT_FL_TOS(oldflp4) \
120 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
121
122#define RT_GC_TIMEOUT (300*HZ)
123
124static int ip_rt_max_size;
125static int ip_rt_redirect_number __read_mostly = 9;
126static int ip_rt_redirect_load __read_mostly = HZ / 50;
127static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
128static int ip_rt_error_cost __read_mostly = HZ;
129static int ip_rt_error_burst __read_mostly = 5 * HZ;
130static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
131static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
132static int ip_rt_min_advmss __read_mostly = 256;
133
134static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
135
136static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
137
138/*
139 * Interface to generic destination cache.
140 */
141
142static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
143static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
144static unsigned int ipv4_mtu(const struct dst_entry *dst);
145static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
146static void ipv4_link_failure(struct sk_buff *skb);
147static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
148 struct sk_buff *skb, u32 mtu);
149static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
150 struct sk_buff *skb);
151static void ipv4_dst_destroy(struct dst_entry *dst);
152
153static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154{
155 WARN_ON(1);
156 return NULL;
157}
158
159static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
160 struct sk_buff *skb,
161 const void *daddr);
162static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
163
164static struct dst_ops ipv4_dst_ops = {
165 .family = AF_INET,
166 .check = ipv4_dst_check,
167 .default_advmss = ipv4_default_advmss,
168 .mtu = ipv4_mtu,
169 .cow_metrics = ipv4_cow_metrics,
170 .destroy = ipv4_dst_destroy,
171 .negative_advice = ipv4_negative_advice,
172 .link_failure = ipv4_link_failure,
173 .update_pmtu = ip_rt_update_pmtu,
174 .redirect = ip_do_redirect,
175 .local_out = __ip_local_out,
176 .neigh_lookup = ipv4_neigh_lookup,
177 .confirm_neigh = ipv4_confirm_neigh,
178};
179
180#define ECN_OR_COST(class) TC_PRIO_##class
181
182const __u8 ip_tos2prio[16] = {
183 TC_PRIO_BESTEFFORT,
184 ECN_OR_COST(BESTEFFORT),
185 TC_PRIO_BESTEFFORT,
186 ECN_OR_COST(BESTEFFORT),
187 TC_PRIO_BULK,
188 ECN_OR_COST(BULK),
189 TC_PRIO_BULK,
190 ECN_OR_COST(BULK),
191 TC_PRIO_INTERACTIVE,
192 ECN_OR_COST(INTERACTIVE),
193 TC_PRIO_INTERACTIVE,
194 ECN_OR_COST(INTERACTIVE),
195 TC_PRIO_INTERACTIVE_BULK,
196 ECN_OR_COST(INTERACTIVE_BULK),
197 TC_PRIO_INTERACTIVE_BULK,
198 ECN_OR_COST(INTERACTIVE_BULK)
199};
200EXPORT_SYMBOL(ip_tos2prio);
201
202static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
203#define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
204
205#ifdef CONFIG_PROC_FS
206static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
207{
208 if (*pos)
209 return NULL;
210 return SEQ_START_TOKEN;
211}
212
213static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
214{
215 ++*pos;
216 return NULL;
217}
218
219static void rt_cache_seq_stop(struct seq_file *seq, void *v)
220{
221}
222
223static int rt_cache_seq_show(struct seq_file *seq, void *v)
224{
225 if (v == SEQ_START_TOKEN)
226 seq_printf(seq, "%-127s\n",
227 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
228 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
229 "HHUptod\tSpecDst");
230 return 0;
231}
232
233static const struct seq_operations rt_cache_seq_ops = {
234 .start = rt_cache_seq_start,
235 .next = rt_cache_seq_next,
236 .stop = rt_cache_seq_stop,
237 .show = rt_cache_seq_show,
238};
239
240static int rt_cache_seq_open(struct inode *inode, struct file *file)
241{
242 return seq_open(file, &rt_cache_seq_ops);
243}
244
245static const struct file_operations rt_cache_seq_fops = {
246 .owner = THIS_MODULE,
247 .open = rt_cache_seq_open,
248 .read = seq_read,
249 .llseek = seq_lseek,
250 .release = seq_release,
251};
252
253
254static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
255{
256 int cpu;
257
258 if (*pos == 0)
259 return SEQ_START_TOKEN;
260
261 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
262 if (!cpu_possible(cpu))
263 continue;
264 *pos = cpu+1;
265 return &per_cpu(rt_cache_stat, cpu);
266 }
267 return NULL;
268}
269
270static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
271{
272 int cpu;
273
274 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
275 if (!cpu_possible(cpu))
276 continue;
277 *pos = cpu+1;
278 return &per_cpu(rt_cache_stat, cpu);
279 }
280 return NULL;
281
282}
283
284static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
285{
286
287}
288
289static int rt_cpu_seq_show(struct seq_file *seq, void *v)
290{
291 struct rt_cache_stat *st = v;
292
293 if (v == SEQ_START_TOKEN) {
294 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");
295 return 0;
296 }
297
298 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
299 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
300 dst_entries_get_slow(&ipv4_dst_ops),
301 0, /* st->in_hit */
302 st->in_slow_tot,
303 st->in_slow_mc,
304 st->in_no_route,
305 st->in_brd,
306 st->in_martian_dst,
307 st->in_martian_src,
308
309 0, /* st->out_hit */
310 st->out_slow_tot,
311 st->out_slow_mc,
312
313 0, /* st->gc_total */
314 0, /* st->gc_ignored */
315 0, /* st->gc_goal_miss */
316 0, /* st->gc_dst_overflow */
317 0, /* st->in_hlist_search */
318 0 /* st->out_hlist_search */
319 );
320 return 0;
321}
322
323static const struct seq_operations rt_cpu_seq_ops = {
324 .start = rt_cpu_seq_start,
325 .next = rt_cpu_seq_next,
326 .stop = rt_cpu_seq_stop,
327 .show = rt_cpu_seq_show,
328};
329
330
331static int rt_cpu_seq_open(struct inode *inode, struct file *file)
332{
333 return seq_open(file, &rt_cpu_seq_ops);
334}
335
336static const struct file_operations rt_cpu_seq_fops = {
337 .owner = THIS_MODULE,
338 .open = rt_cpu_seq_open,
339 .read = seq_read,
340 .llseek = seq_lseek,
341 .release = seq_release,
342};
343
344#ifdef CONFIG_IP_ROUTE_CLASSID
345static int rt_acct_proc_show(struct seq_file *m, void *v)
346{
347 struct ip_rt_acct *dst, *src;
348 unsigned int i, j;
349
350 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
351 if (!dst)
352 return -ENOMEM;
353
354 for_each_possible_cpu(i) {
355 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
356 for (j = 0; j < 256; j++) {
357 dst[j].o_bytes += src[j].o_bytes;
358 dst[j].o_packets += src[j].o_packets;
359 dst[j].i_bytes += src[j].i_bytes;
360 dst[j].i_packets += src[j].i_packets;
361 }
362 }
363
364 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
365 kfree(dst);
366 return 0;
367}
368
369static int rt_acct_proc_open(struct inode *inode, struct file *file)
370{
371 return single_open(file, rt_acct_proc_show, NULL);
372}
373
374static const struct file_operations rt_acct_proc_fops = {
375 .owner = THIS_MODULE,
376 .open = rt_acct_proc_open,
377 .read = seq_read,
378 .llseek = seq_lseek,
379 .release = single_release,
380};
381#endif
382
383static int __net_init ip_rt_do_proc_init(struct net *net)
384{
385 struct proc_dir_entry *pde;
386
387 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
388 &rt_cache_seq_fops);
389 if (!pde)
390 goto err1;
391
392 pde = proc_create("rt_cache", S_IRUGO,
393 net->proc_net_stat, &rt_cpu_seq_fops);
394 if (!pde)
395 goto err2;
396
397#ifdef CONFIG_IP_ROUTE_CLASSID
398 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
399 if (!pde)
400 goto err3;
401#endif
402 return 0;
403
404#ifdef CONFIG_IP_ROUTE_CLASSID
405err3:
406 remove_proc_entry("rt_cache", net->proc_net_stat);
407#endif
408err2:
409 remove_proc_entry("rt_cache", net->proc_net);
410err1:
411 return -ENOMEM;
412}
413
414static void __net_exit ip_rt_do_proc_exit(struct net *net)
415{
416 remove_proc_entry("rt_cache", net->proc_net_stat);
417 remove_proc_entry("rt_cache", net->proc_net);
418#ifdef CONFIG_IP_ROUTE_CLASSID
419 remove_proc_entry("rt_acct", net->proc_net);
420#endif
421}
422
423static struct pernet_operations ip_rt_proc_ops __net_initdata = {
424 .init = ip_rt_do_proc_init,
425 .exit = ip_rt_do_proc_exit,
426};
427
428static int __init ip_rt_proc_init(void)
429{
430 return register_pernet_subsys(&ip_rt_proc_ops);
431}
432
433#else
434static inline int ip_rt_proc_init(void)
435{
436 return 0;
437}
438#endif /* CONFIG_PROC_FS */
439
440static inline bool rt_is_expired(const struct rtable *rth)
441{
442 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
443}
444
445void rt_cache_flush(struct net *net)
446{
447 rt_genid_bump_ipv4(net);
448}
449
450static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
451 struct sk_buff *skb,
452 const void *daddr)
453{
454 struct net_device *dev = dst->dev;
455 const __be32 *pkey = daddr;
456 const struct rtable *rt;
457 struct neighbour *n;
458
459 rt = (const struct rtable *) dst;
460 if (rt->rt_gateway)
461 pkey = (const __be32 *) &rt->rt_gateway;
462 else if (skb)
463 pkey = &ip_hdr(skb)->daddr;
464
465 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
466 if (n)
467 return n;
468 return neigh_create(&arp_tbl, pkey, dev);
469}
470
471static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
472{
473 struct net_device *dev = dst->dev;
474 const __be32 *pkey = daddr;
475 const struct rtable *rt;
476
477 rt = (const struct rtable *)dst;
478 if (rt->rt_gateway)
479 pkey = (const __be32 *)&rt->rt_gateway;
480 else if (!daddr ||
481 (rt->rt_flags &
482 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
483 return;
484
485 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
486}
487
488#define IP_IDENTS_SZ 2048u
489
490static atomic_t *ip_idents __read_mostly;
491static u32 *ip_tstamps __read_mostly;
492
493/* In order to protect privacy, we add a perturbation to identifiers
494 * if one generator is seldom used. This makes hard for an attacker
495 * to infer how many packets were sent between two points in time.
496 */
497u32 ip_idents_reserve(u32 hash, int segs)
498{
499 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
500 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
501 u32 old = ACCESS_ONCE(*p_tstamp);
502 u32 now = (u32)jiffies;
503 u32 new, delta = 0;
504
505 if (old != now && cmpxchg(p_tstamp, old, now) == old)
506 delta = prandom_u32_max(now - old);
507
508 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
509 do {
510 old = (u32)atomic_read(p_id);
511 new = old + delta + segs;
512 } while (atomic_cmpxchg(p_id, old, new) != old);
513
514 return new - segs;
515}
516EXPORT_SYMBOL(ip_idents_reserve);
517
518void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
519{
520 u32 hash, id;
521
522 /* Note the following code is not safe, but this is okay. */
523 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
524 get_random_bytes(&net->ipv4.ip_id_key,
525 sizeof(net->ipv4.ip_id_key));
526
527 hash = siphash_3u32((__force u32)iph->daddr,
528 (__force u32)iph->saddr,
529 iph->protocol,
530 &net->ipv4.ip_id_key);
531 id = ip_idents_reserve(hash, segs);
532 iph->id = htons(id);
533}
534EXPORT_SYMBOL(__ip_select_ident);
535
536static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
537 const struct sock *sk,
538 const struct iphdr *iph,
539 int oif, u8 tos,
540 u8 prot, u32 mark, int flow_flags)
541{
542 if (sk) {
543 const struct inet_sock *inet = inet_sk(sk);
544
545 oif = sk->sk_bound_dev_if;
546 mark = sk->sk_mark;
547 tos = RT_CONN_FLAGS(sk);
548 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
549 }
550 flowi4_init_output(fl4, oif, mark, tos,
551 RT_SCOPE_UNIVERSE, prot,
552 flow_flags,
553 iph->daddr, iph->saddr, 0, 0,
554 sock_net_uid(net, sk));
555}
556
557static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
558 const struct sock *sk)
559{
560 const struct net *net = dev_net(skb->dev);
561 const struct iphdr *iph = ip_hdr(skb);
562 int oif = skb->dev->ifindex;
563 u8 tos = RT_TOS(iph->tos);
564 u8 prot = iph->protocol;
565 u32 mark = skb->mark;
566
567 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
568}
569
570static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
571{
572 const struct inet_sock *inet = inet_sk(sk);
573 const struct ip_options_rcu *inet_opt;
574 __be32 daddr = inet->inet_daddr;
575
576 rcu_read_lock();
577 inet_opt = rcu_dereference(inet->inet_opt);
578 if (inet_opt && inet_opt->opt.srr)
579 daddr = inet_opt->opt.faddr;
580 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
581 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
582 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
583 inet_sk_flowi_flags(sk),
584 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
585 rcu_read_unlock();
586}
587
588static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
589 const struct sk_buff *skb)
590{
591 if (skb)
592 build_skb_flow_key(fl4, skb, sk);
593 else
594 build_sk_flow_key(fl4, sk);
595}
596
597static DEFINE_SPINLOCK(fnhe_lock);
598
599static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
600{
601 struct rtable *rt;
602
603 rt = rcu_dereference(fnhe->fnhe_rth_input);
604 if (rt) {
605 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
606 dst_dev_put(&rt->dst);
607 dst_release(&rt->dst);
608 }
609 rt = rcu_dereference(fnhe->fnhe_rth_output);
610 if (rt) {
611 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
612 dst_dev_put(&rt->dst);
613 dst_release(&rt->dst);
614 }
615}
616
617static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
618{
619 struct fib_nh_exception *fnhe, *oldest;
620
621 oldest = rcu_dereference(hash->chain);
622 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
623 fnhe = rcu_dereference(fnhe->fnhe_next)) {
624 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
625 oldest = fnhe;
626 }
627 fnhe_flush_routes(oldest);
628 return oldest;
629}
630
631static inline u32 fnhe_hashfun(__be32 daddr)
632{
633 static u32 fnhe_hashrnd __read_mostly;
634 u32 hval;
635
636 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
637 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
638 return hash_32(hval, FNHE_HASH_SHIFT);
639}
640
641static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
642{
643 rt->rt_pmtu = fnhe->fnhe_pmtu;
644 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
645 rt->dst.expires = fnhe->fnhe_expires;
646
647 if (fnhe->fnhe_gw) {
648 rt->rt_flags |= RTCF_REDIRECTED;
649 rt->rt_gateway = fnhe->fnhe_gw;
650 rt->rt_uses_gateway = 1;
651 }
652}
653
654static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
655 u32 pmtu, bool lock, unsigned long expires)
656{
657 struct fnhe_hash_bucket *hash;
658 struct fib_nh_exception *fnhe;
659 struct rtable *rt;
660 u32 genid, hval;
661 unsigned int i;
662 int depth;
663
664 genid = fnhe_genid(dev_net(nh->nh_dev));
665 hval = fnhe_hashfun(daddr);
666
667 spin_lock_bh(&fnhe_lock);
668
669 hash = rcu_dereference(nh->nh_exceptions);
670 if (!hash) {
671 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
672 if (!hash)
673 goto out_unlock;
674 rcu_assign_pointer(nh->nh_exceptions, hash);
675 }
676
677 hash += hval;
678
679 depth = 0;
680 for (fnhe = rcu_dereference(hash->chain); fnhe;
681 fnhe = rcu_dereference(fnhe->fnhe_next)) {
682 if (fnhe->fnhe_daddr == daddr)
683 break;
684 depth++;
685 }
686
687 if (fnhe) {
688 if (fnhe->fnhe_genid != genid)
689 fnhe->fnhe_genid = genid;
690 if (gw)
691 fnhe->fnhe_gw = gw;
692 if (pmtu) {
693 fnhe->fnhe_pmtu = pmtu;
694 fnhe->fnhe_mtu_locked = lock;
695 }
696 fnhe->fnhe_expires = max(1UL, expires);
697 /* Update all cached dsts too */
698 rt = rcu_dereference(fnhe->fnhe_rth_input);
699 if (rt)
700 fill_route_from_fnhe(rt, fnhe);
701 rt = rcu_dereference(fnhe->fnhe_rth_output);
702 if (rt)
703 fill_route_from_fnhe(rt, fnhe);
704 } else {
705 if (depth > FNHE_RECLAIM_DEPTH)
706 fnhe = fnhe_oldest(hash);
707 else {
708 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
709 if (!fnhe)
710 goto out_unlock;
711
712 fnhe->fnhe_next = hash->chain;
713 rcu_assign_pointer(hash->chain, fnhe);
714 }
715 fnhe->fnhe_genid = genid;
716 fnhe->fnhe_daddr = daddr;
717 fnhe->fnhe_gw = gw;
718 fnhe->fnhe_pmtu = pmtu;
719 fnhe->fnhe_mtu_locked = lock;
720 fnhe->fnhe_expires = max(1UL, expires);
721
722 /* Exception created; mark the cached routes for the nexthop
723 * stale, so anyone caching it rechecks if this exception
724 * applies to them.
725 */
726 rt = rcu_dereference(nh->nh_rth_input);
727 if (rt)
728 rt->dst.obsolete = DST_OBSOLETE_KILL;
729
730 for_each_possible_cpu(i) {
731 struct rtable __rcu **prt;
732 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
733 rt = rcu_dereference(*prt);
734 if (rt)
735 rt->dst.obsolete = DST_OBSOLETE_KILL;
736 }
737 }
738
739 fnhe->fnhe_stamp = jiffies;
740
741out_unlock:
742 spin_unlock_bh(&fnhe_lock);
743}
744
745static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
746 bool kill_route)
747{
748 __be32 new_gw = icmp_hdr(skb)->un.gateway;
749 __be32 old_gw = ip_hdr(skb)->saddr;
750 struct net_device *dev = skb->dev;
751 struct in_device *in_dev;
752 struct fib_result res;
753 struct neighbour *n;
754 struct net *net;
755
756 switch (icmp_hdr(skb)->code & 7) {
757 case ICMP_REDIR_NET:
758 case ICMP_REDIR_NETTOS:
759 case ICMP_REDIR_HOST:
760 case ICMP_REDIR_HOSTTOS:
761 break;
762
763 default:
764 return;
765 }
766
767 if (rt->rt_gateway != old_gw)
768 return;
769
770 in_dev = __in_dev_get_rcu(dev);
771 if (!in_dev)
772 return;
773
774 net = dev_net(dev);
775 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
776 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
777 ipv4_is_zeronet(new_gw))
778 goto reject_redirect;
779
780 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
781 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
782 goto reject_redirect;
783 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
784 goto reject_redirect;
785 } else {
786 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
787 goto reject_redirect;
788 }
789
790 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
791 if (!n)
792 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
793 if (!IS_ERR(n)) {
794 if (!(n->nud_state & NUD_VALID)) {
795 neigh_event_send(n, NULL);
796 } else {
797 if (fib_lookup(net, fl4, &res, 0) == 0) {
798 struct fib_nh *nh = &FIB_RES_NH(res);
799
800 update_or_create_fnhe(nh, fl4->daddr, new_gw,
801 0, false,
802 jiffies + ip_rt_gc_timeout);
803 }
804 if (kill_route)
805 rt->dst.obsolete = DST_OBSOLETE_KILL;
806 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
807 }
808 neigh_release(n);
809 }
810 return;
811
812reject_redirect:
813#ifdef CONFIG_IP_ROUTE_VERBOSE
814 if (IN_DEV_LOG_MARTIANS(in_dev)) {
815 const struct iphdr *iph = (const struct iphdr *) skb->data;
816 __be32 daddr = iph->daddr;
817 __be32 saddr = iph->saddr;
818
819 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
820 " Advised path = %pI4 -> %pI4\n",
821 &old_gw, dev->name, &new_gw,
822 &saddr, &daddr);
823 }
824#endif
825 ;
826}
827
828static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
829{
830 struct rtable *rt;
831 struct flowi4 fl4;
832 const struct iphdr *iph = (const struct iphdr *) skb->data;
833 struct net *net = dev_net(skb->dev);
834 int oif = skb->dev->ifindex;
835 u8 tos = RT_TOS(iph->tos);
836 u8 prot = iph->protocol;
837 u32 mark = skb->mark;
838
839 rt = (struct rtable *) dst;
840
841 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
842 __ip_do_redirect(rt, skb, &fl4, true);
843}
844
845static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
846{
847 struct rtable *rt = (struct rtable *)dst;
848 struct dst_entry *ret = dst;
849
850 if (rt) {
851 if (dst->obsolete > 0) {
852 ip_rt_put(rt);
853 ret = NULL;
854 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
855 rt->dst.expires) {
856 ip_rt_put(rt);
857 ret = NULL;
858 }
859 }
860 return ret;
861}
862
863/*
864 * Algorithm:
865 * 1. The first ip_rt_redirect_number redirects are sent
866 * with exponential backoff, then we stop sending them at all,
867 * assuming that the host ignores our redirects.
868 * 2. If we did not see packets requiring redirects
869 * during ip_rt_redirect_silence, we assume that the host
870 * forgot redirected route and start to send redirects again.
871 *
872 * This algorithm is much cheaper and more intelligent than dumb load limiting
873 * in icmp.c.
874 *
875 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
876 * and "frag. need" (breaks PMTU discovery) in icmp.c.
877 */
878
879void ip_rt_send_redirect(struct sk_buff *skb)
880{
881 struct rtable *rt = skb_rtable(skb);
882 struct in_device *in_dev;
883 struct inet_peer *peer;
884 struct net *net;
885 int log_martians;
886 int vif;
887
888 rcu_read_lock();
889 in_dev = __in_dev_get_rcu(rt->dst.dev);
890 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
891 rcu_read_unlock();
892 return;
893 }
894 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
895 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
896 rcu_read_unlock();
897
898 net = dev_net(rt->dst.dev);
899 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
900 if (!peer) {
901 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
902 rt_nexthop(rt, ip_hdr(skb)->daddr));
903 return;
904 }
905
906 /* No redirected packets during ip_rt_redirect_silence;
907 * reset the algorithm.
908 */
909 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
910 peer->rate_tokens = 0;
911 peer->n_redirects = 0;
912 }
913
914 /* Too many ignored redirects; do not send anything
915 * set dst.rate_last to the last seen redirected packet.
916 */
917 if (peer->n_redirects >= ip_rt_redirect_number) {
918 peer->rate_last = jiffies;
919 goto out_put_peer;
920 }
921
922 /* Check for load limit; set rate_last to the latest sent
923 * redirect.
924 */
925 if (peer->rate_tokens == 0 ||
926 time_after(jiffies,
927 (peer->rate_last +
928 (ip_rt_redirect_load << peer->rate_tokens)))) {
929 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
930
931 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
932 peer->rate_last = jiffies;
933 ++peer->rate_tokens;
934 ++peer->n_redirects;
935#ifdef CONFIG_IP_ROUTE_VERBOSE
936 if (log_martians &&
937 peer->rate_tokens == ip_rt_redirect_number)
938 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
939 &ip_hdr(skb)->saddr, inet_iif(skb),
940 &ip_hdr(skb)->daddr, &gw);
941#endif
942 }
943out_put_peer:
944 inet_putpeer(peer);
945}
946
947static int ip_error(struct sk_buff *skb)
948{
949 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
950 struct rtable *rt = skb_rtable(skb);
951 struct inet_peer *peer;
952 unsigned long now;
953 struct net *net;
954 bool send;
955 int code;
956
957 /* IP on this device is disabled. */
958 if (!in_dev)
959 goto out;
960
961 net = dev_net(rt->dst.dev);
962 if (!IN_DEV_FORWARD(in_dev)) {
963 switch (rt->dst.error) {
964 case EHOSTUNREACH:
965 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
966 break;
967
968 case ENETUNREACH:
969 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
970 break;
971 }
972 goto out;
973 }
974
975 switch (rt->dst.error) {
976 case EINVAL:
977 default:
978 goto out;
979 case EHOSTUNREACH:
980 code = ICMP_HOST_UNREACH;
981 break;
982 case ENETUNREACH:
983 code = ICMP_NET_UNREACH;
984 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
985 break;
986 case EACCES:
987 code = ICMP_PKT_FILTERED;
988 break;
989 }
990
991 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
992 l3mdev_master_ifindex(skb->dev), 1);
993
994 send = true;
995 if (peer) {
996 now = jiffies;
997 peer->rate_tokens += now - peer->rate_last;
998 if (peer->rate_tokens > ip_rt_error_burst)
999 peer->rate_tokens = ip_rt_error_burst;
1000 peer->rate_last = now;
1001 if (peer->rate_tokens >= ip_rt_error_cost)
1002 peer->rate_tokens -= ip_rt_error_cost;
1003 else
1004 send = false;
1005 inet_putpeer(peer);
1006 }
1007 if (send)
1008 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1009
1010out: kfree_skb(skb);
1011 return 0;
1012}
1013
1014static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1015{
1016 struct dst_entry *dst = &rt->dst;
1017 struct fib_result res;
1018 bool lock = false;
1019
1020 if (ip_mtu_locked(dst))
1021 return;
1022
1023 if (ipv4_mtu(dst) < mtu)
1024 return;
1025
1026 if (mtu < ip_rt_min_pmtu) {
1027 lock = true;
1028 mtu = ip_rt_min_pmtu;
1029 }
1030
1031 if (rt->rt_pmtu == mtu &&
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 *nh = &FIB_RES_NH(res);
1038
1039 update_or_create_fnhe(nh, 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, u32 mark, u8 protocol, int flow_flags)
1057{
1058 const struct iphdr *iph = (const struct iphdr *) skb->data;
1059 struct flowi4 fl4;
1060 struct rtable *rt;
1061
1062 if (!mark)
1063 mark = IP4_REPLY_MARK(net, skb->mark);
1064
1065 __build_flow_key(net, &fl4, NULL, iph, oif,
1066 RT_TOS(iph->tos), protocol, mark, flow_flags);
1067 rt = __ip_route_output_key(net, &fl4);
1068 if (!IS_ERR(rt)) {
1069 __ip_rt_update_pmtu(rt, &fl4, mtu);
1070 ip_rt_put(rt);
1071 }
1072}
1073EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1074
1075static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1076{
1077 const struct iphdr *iph = (const struct iphdr *) skb->data;
1078 struct flowi4 fl4;
1079 struct rtable *rt;
1080
1081 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1082
1083 if (!fl4.flowi4_mark)
1084 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1085
1086 rt = __ip_route_output_key(sock_net(sk), &fl4);
1087 if (!IS_ERR(rt)) {
1088 __ip_rt_update_pmtu(rt, &fl4, mtu);
1089 ip_rt_put(rt);
1090 }
1091}
1092
1093void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1094{
1095 const struct iphdr *iph = (const struct iphdr *) skb->data;
1096 struct flowi4 fl4;
1097 struct rtable *rt;
1098 struct dst_entry *odst = NULL;
1099 bool new = false;
1100 struct net *net = sock_net(sk);
1101
1102 bh_lock_sock(sk);
1103
1104 if (!ip_sk_accept_pmtu(sk))
1105 goto out;
1106
1107 odst = sk_dst_get(sk);
1108
1109 if (sock_owned_by_user(sk) || !odst) {
1110 __ipv4_sk_update_pmtu(skb, sk, mtu);
1111 goto out;
1112 }
1113
1114 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1115
1116 rt = (struct rtable *)odst;
1117 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1118 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1119 if (IS_ERR(rt))
1120 goto out;
1121
1122 new = true;
1123 }
1124
1125 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1126
1127 if (!dst_check(&rt->dst, 0)) {
1128 if (new)
1129 dst_release(&rt->dst);
1130
1131 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1132 if (IS_ERR(rt))
1133 goto out;
1134
1135 new = true;
1136 }
1137
1138 if (new)
1139 sk_dst_set(sk, &rt->dst);
1140
1141out:
1142 bh_unlock_sock(sk);
1143 dst_release(odst);
1144}
1145EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1146
1147void ipv4_redirect(struct sk_buff *skb, struct net *net,
1148 int oif, u32 mark, u8 protocol, int flow_flags)
1149{
1150 const struct iphdr *iph = (const struct iphdr *) skb->data;
1151 struct flowi4 fl4;
1152 struct rtable *rt;
1153
1154 __build_flow_key(net, &fl4, NULL, iph, oif,
1155 RT_TOS(iph->tos), protocol, mark, flow_flags);
1156 rt = __ip_route_output_key(net, &fl4);
1157 if (!IS_ERR(rt)) {
1158 __ip_do_redirect(rt, skb, &fl4, false);
1159 ip_rt_put(rt);
1160 }
1161}
1162EXPORT_SYMBOL_GPL(ipv4_redirect);
1163
1164void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1165{
1166 const struct iphdr *iph = (const struct iphdr *) skb->data;
1167 struct flowi4 fl4;
1168 struct rtable *rt;
1169 struct net *net = sock_net(sk);
1170
1171 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1172 rt = __ip_route_output_key(net, &fl4);
1173 if (!IS_ERR(rt)) {
1174 __ip_do_redirect(rt, skb, &fl4, false);
1175 ip_rt_put(rt);
1176 }
1177}
1178EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1179
1180static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1181{
1182 struct rtable *rt = (struct rtable *) dst;
1183
1184 /* All IPV4 dsts are created with ->obsolete set to the value
1185 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1186 * into this function always.
1187 *
1188 * When a PMTU/redirect information update invalidates a route,
1189 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1190 * DST_OBSOLETE_DEAD by dst_free().
1191 */
1192 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1193 return NULL;
1194 return dst;
1195}
1196
1197static void ipv4_link_failure(struct sk_buff *skb)
1198{
1199 struct rtable *rt;
1200
1201 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1202
1203 rt = skb_rtable(skb);
1204 if (rt)
1205 dst_set_expires(&rt->dst, 0);
1206}
1207
1208static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1209{
1210 pr_debug("%s: %pI4 -> %pI4, %s\n",
1211 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1212 skb->dev ? skb->dev->name : "?");
1213 kfree_skb(skb);
1214 WARN_ON(1);
1215 return 0;
1216}
1217
1218/*
1219 We do not cache source address of outgoing interface,
1220 because it is used only by IP RR, TS and SRR options,
1221 so that it out of fast path.
1222
1223 BTW remember: "addr" is allowed to be not aligned
1224 in IP options!
1225 */
1226
1227void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1228{
1229 __be32 src;
1230
1231 if (rt_is_output_route(rt))
1232 src = ip_hdr(skb)->saddr;
1233 else {
1234 struct fib_result res;
1235 struct flowi4 fl4;
1236 struct iphdr *iph;
1237
1238 iph = ip_hdr(skb);
1239
1240 memset(&fl4, 0, sizeof(fl4));
1241 fl4.daddr = iph->daddr;
1242 fl4.saddr = iph->saddr;
1243 fl4.flowi4_tos = RT_TOS(iph->tos);
1244 fl4.flowi4_oif = rt->dst.dev->ifindex;
1245 fl4.flowi4_iif = skb->dev->ifindex;
1246 fl4.flowi4_mark = skb->mark;
1247
1248 rcu_read_lock();
1249 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1250 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1251 else
1252 src = inet_select_addr(rt->dst.dev,
1253 rt_nexthop(rt, iph->daddr),
1254 RT_SCOPE_UNIVERSE);
1255 rcu_read_unlock();
1256 }
1257 memcpy(addr, &src, 4);
1258}
1259
1260#ifdef CONFIG_IP_ROUTE_CLASSID
1261static void set_class_tag(struct rtable *rt, u32 tag)
1262{
1263 if (!(rt->dst.tclassid & 0xFFFF))
1264 rt->dst.tclassid |= tag & 0xFFFF;
1265 if (!(rt->dst.tclassid & 0xFFFF0000))
1266 rt->dst.tclassid |= tag & 0xFFFF0000;
1267}
1268#endif
1269
1270static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1271{
1272 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1273 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1274 ip_rt_min_advmss);
1275
1276 return min(advmss, IPV4_MAX_PMTU - header_size);
1277}
1278
1279static unsigned int ipv4_mtu(const struct dst_entry *dst)
1280{
1281 const struct rtable *rt = (const struct rtable *) dst;
1282 unsigned int mtu = rt->rt_pmtu;
1283
1284 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1285 mtu = dst_metric_raw(dst, RTAX_MTU);
1286
1287 if (mtu)
1288 return mtu;
1289
1290 mtu = READ_ONCE(dst->dev->mtu);
1291
1292 if (unlikely(ip_mtu_locked(dst))) {
1293 if (rt->rt_uses_gateway && mtu > 576)
1294 mtu = 576;
1295 }
1296
1297 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1298
1299 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1300}
1301
1302static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1303{
1304 struct fnhe_hash_bucket *hash;
1305 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1306 u32 hval = fnhe_hashfun(daddr);
1307
1308 spin_lock_bh(&fnhe_lock);
1309
1310 hash = rcu_dereference_protected(nh->nh_exceptions,
1311 lockdep_is_held(&fnhe_lock));
1312 hash += hval;
1313
1314 fnhe_p = &hash->chain;
1315 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1316 while (fnhe) {
1317 if (fnhe->fnhe_daddr == daddr) {
1318 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1319 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1320 /* set fnhe_daddr to 0 to ensure it won't bind with
1321 * new dsts in rt_bind_exception().
1322 */
1323 fnhe->fnhe_daddr = 0;
1324 fnhe_flush_routes(fnhe);
1325 kfree_rcu(fnhe, rcu);
1326 break;
1327 }
1328 fnhe_p = &fnhe->fnhe_next;
1329 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1330 lockdep_is_held(&fnhe_lock));
1331 }
1332
1333 spin_unlock_bh(&fnhe_lock);
1334}
1335
1336static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1337{
1338 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1339 struct fib_nh_exception *fnhe;
1340 u32 hval;
1341
1342 if (!hash)
1343 return NULL;
1344
1345 hval = fnhe_hashfun(daddr);
1346
1347 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1348 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1349 if (fnhe->fnhe_daddr == daddr) {
1350 if (fnhe->fnhe_expires &&
1351 time_after(jiffies, fnhe->fnhe_expires)) {
1352 ip_del_fnhe(nh, daddr);
1353 break;
1354 }
1355 return fnhe;
1356 }
1357 }
1358 return NULL;
1359}
1360
1361static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1362 __be32 daddr, const bool do_cache)
1363{
1364 bool ret = false;
1365
1366 spin_lock_bh(&fnhe_lock);
1367
1368 if (daddr == fnhe->fnhe_daddr) {
1369 struct rtable __rcu **porig;
1370 struct rtable *orig;
1371 int genid = fnhe_genid(dev_net(rt->dst.dev));
1372
1373 if (rt_is_input_route(rt))
1374 porig = &fnhe->fnhe_rth_input;
1375 else
1376 porig = &fnhe->fnhe_rth_output;
1377 orig = rcu_dereference(*porig);
1378
1379 if (fnhe->fnhe_genid != genid) {
1380 fnhe->fnhe_genid = genid;
1381 fnhe->fnhe_gw = 0;
1382 fnhe->fnhe_pmtu = 0;
1383 fnhe->fnhe_expires = 0;
1384 fnhe_flush_routes(fnhe);
1385 orig = NULL;
1386 }
1387 fill_route_from_fnhe(rt, fnhe);
1388 if (!rt->rt_gateway)
1389 rt->rt_gateway = daddr;
1390
1391 if (do_cache) {
1392 dst_hold(&rt->dst);
1393 rcu_assign_pointer(*porig, rt);
1394 if (orig) {
1395 dst_dev_put(&orig->dst);
1396 dst_release(&orig->dst);
1397 }
1398 ret = true;
1399 }
1400
1401 fnhe->fnhe_stamp = jiffies;
1402 }
1403 spin_unlock_bh(&fnhe_lock);
1404
1405 return ret;
1406}
1407
1408static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1409{
1410 struct rtable *orig, *prev, **p;
1411 bool ret = true;
1412
1413 if (rt_is_input_route(rt)) {
1414 p = (struct rtable **)&nh->nh_rth_input;
1415 } else {
1416 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1417 }
1418 orig = *p;
1419
1420 /* hold dst before doing cmpxchg() to avoid race condition
1421 * on this dst
1422 */
1423 dst_hold(&rt->dst);
1424 prev = cmpxchg(p, orig, rt);
1425 if (prev == orig) {
1426 if (orig) {
1427 dst_dev_put(&orig->dst);
1428 dst_release(&orig->dst);
1429 }
1430 } else {
1431 dst_release(&rt->dst);
1432 ret = false;
1433 }
1434
1435 return ret;
1436}
1437
1438struct uncached_list {
1439 spinlock_t lock;
1440 struct list_head head;
1441};
1442
1443static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1444
1445static void rt_add_uncached_list(struct rtable *rt)
1446{
1447 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1448
1449 rt->rt_uncached_list = ul;
1450
1451 spin_lock_bh(&ul->lock);
1452 list_add_tail(&rt->rt_uncached, &ul->head);
1453 spin_unlock_bh(&ul->lock);
1454}
1455
1456static void ipv4_dst_destroy(struct dst_entry *dst)
1457{
1458 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1459 struct rtable *rt = (struct rtable *) dst;
1460
1461 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1462 kfree(p);
1463
1464 if (!list_empty(&rt->rt_uncached)) {
1465 struct uncached_list *ul = rt->rt_uncached_list;
1466
1467 spin_lock_bh(&ul->lock);
1468 list_del(&rt->rt_uncached);
1469 spin_unlock_bh(&ul->lock);
1470 }
1471}
1472
1473void rt_flush_dev(struct net_device *dev)
1474{
1475 struct net *net = dev_net(dev);
1476 struct rtable *rt;
1477 int cpu;
1478
1479 for_each_possible_cpu(cpu) {
1480 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1481
1482 spin_lock_bh(&ul->lock);
1483 list_for_each_entry(rt, &ul->head, rt_uncached) {
1484 if (rt->dst.dev != dev)
1485 continue;
1486 rt->dst.dev = net->loopback_dev;
1487 dev_hold(rt->dst.dev);
1488 dev_put(dev);
1489 }
1490 spin_unlock_bh(&ul->lock);
1491 }
1492}
1493
1494static bool rt_cache_valid(const struct rtable *rt)
1495{
1496 return rt &&
1497 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1498 !rt_is_expired(rt);
1499}
1500
1501static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1502 const struct fib_result *res,
1503 struct fib_nh_exception *fnhe,
1504 struct fib_info *fi, u16 type, u32 itag,
1505 const bool do_cache)
1506{
1507 bool cached = false;
1508
1509 if (fi) {
1510 struct fib_nh *nh = &FIB_RES_NH(*res);
1511
1512 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1513 rt->rt_gateway = nh->nh_gw;
1514 rt->rt_uses_gateway = 1;
1515 }
1516 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1517 if (fi->fib_metrics != &dst_default_metrics) {
1518 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1519 refcount_inc(&fi->fib_metrics->refcnt);
1520 }
1521#ifdef CONFIG_IP_ROUTE_CLASSID
1522 rt->dst.tclassid = nh->nh_tclassid;
1523#endif
1524 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1525 if (unlikely(fnhe))
1526 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1527 else if (do_cache)
1528 cached = rt_cache_route(nh, rt);
1529 if (unlikely(!cached)) {
1530 /* Routes we intend to cache in nexthop exception or
1531 * FIB nexthop have the DST_NOCACHE bit clear.
1532 * However, if we are unsuccessful at storing this
1533 * route into the cache we really need to set it.
1534 */
1535 if (!rt->rt_gateway)
1536 rt->rt_gateway = daddr;
1537 rt_add_uncached_list(rt);
1538 }
1539 } else
1540 rt_add_uncached_list(rt);
1541
1542#ifdef CONFIG_IP_ROUTE_CLASSID
1543#ifdef CONFIG_IP_MULTIPLE_TABLES
1544 set_class_tag(rt, res->tclassid);
1545#endif
1546 set_class_tag(rt, itag);
1547#endif
1548}
1549
1550struct rtable *rt_dst_alloc(struct net_device *dev,
1551 unsigned int flags, u16 type,
1552 bool nopolicy, bool noxfrm, bool will_cache)
1553{
1554 struct rtable *rt;
1555
1556 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1557 (will_cache ? 0 : DST_HOST) |
1558 (nopolicy ? DST_NOPOLICY : 0) |
1559 (noxfrm ? DST_NOXFRM : 0));
1560
1561 if (rt) {
1562 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1563 rt->rt_flags = flags;
1564 rt->rt_type = type;
1565 rt->rt_is_input = 0;
1566 rt->rt_iif = 0;
1567 rt->rt_pmtu = 0;
1568 rt->rt_mtu_locked = 0;
1569 rt->rt_gateway = 0;
1570 rt->rt_uses_gateway = 0;
1571 rt->rt_table_id = 0;
1572 INIT_LIST_HEAD(&rt->rt_uncached);
1573
1574 rt->dst.output = ip_output;
1575 if (flags & RTCF_LOCAL)
1576 rt->dst.input = ip_local_deliver;
1577 }
1578
1579 return rt;
1580}
1581EXPORT_SYMBOL(rt_dst_alloc);
1582
1583/* called in rcu_read_lock() section */
1584int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1585 u8 tos, struct net_device *dev,
1586 struct in_device *in_dev, u32 *itag)
1587{
1588 int err;
1589
1590 /* Primary sanity checks. */
1591 if (!in_dev)
1592 return -EINVAL;
1593
1594 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1595 skb->protocol != htons(ETH_P_IP))
1596 return -EINVAL;
1597
1598 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1599 return -EINVAL;
1600
1601 if (ipv4_is_zeronet(saddr)) {
1602 if (!ipv4_is_local_multicast(daddr))
1603 return -EINVAL;
1604 } else {
1605 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1606 in_dev, itag);
1607 if (err < 0)
1608 return err;
1609 }
1610 return 0;
1611}
1612
1613/* called in rcu_read_lock() section */
1614static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1615 u8 tos, struct net_device *dev, int our)
1616{
1617 struct in_device *in_dev = __in_dev_get_rcu(dev);
1618 unsigned int flags = RTCF_MULTICAST;
1619 struct rtable *rth;
1620 u32 itag = 0;
1621 int err;
1622
1623 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1624 if (err)
1625 return err;
1626
1627 if (our)
1628 flags |= RTCF_LOCAL;
1629
1630 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1631 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1632 if (!rth)
1633 return -ENOBUFS;
1634
1635#ifdef CONFIG_IP_ROUTE_CLASSID
1636 rth->dst.tclassid = itag;
1637#endif
1638 rth->dst.output = ip_rt_bug;
1639 rth->rt_is_input= 1;
1640
1641#ifdef CONFIG_IP_MROUTE
1642 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1643 rth->dst.input = ip_mr_input;
1644#endif
1645 RT_CACHE_STAT_INC(in_slow_mc);
1646
1647 skb_dst_set(skb, &rth->dst);
1648 return 0;
1649}
1650
1651
1652static void ip_handle_martian_source(struct net_device *dev,
1653 struct in_device *in_dev,
1654 struct sk_buff *skb,
1655 __be32 daddr,
1656 __be32 saddr)
1657{
1658 RT_CACHE_STAT_INC(in_martian_src);
1659#ifdef CONFIG_IP_ROUTE_VERBOSE
1660 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1661 /*
1662 * RFC1812 recommendation, if source is martian,
1663 * the only hint is MAC header.
1664 */
1665 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1666 &daddr, &saddr, dev->name);
1667 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1668 print_hex_dump(KERN_WARNING, "ll header: ",
1669 DUMP_PREFIX_OFFSET, 16, 1,
1670 skb_mac_header(skb),
1671 dev->hard_header_len, true);
1672 }
1673 }
1674#endif
1675}
1676
1677static void set_lwt_redirect(struct rtable *rth)
1678{
1679 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1680 rth->dst.lwtstate->orig_output = rth->dst.output;
1681 rth->dst.output = lwtunnel_output;
1682 }
1683
1684 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1685 rth->dst.lwtstate->orig_input = rth->dst.input;
1686 rth->dst.input = lwtunnel_input;
1687 }
1688}
1689
1690/* called in rcu_read_lock() section */
1691static int __mkroute_input(struct sk_buff *skb,
1692 const struct fib_result *res,
1693 struct in_device *in_dev,
1694 __be32 daddr, __be32 saddr, u32 tos)
1695{
1696 struct fib_nh_exception *fnhe;
1697 struct rtable *rth;
1698 int err;
1699 struct in_device *out_dev;
1700 bool do_cache;
1701 u32 itag = 0;
1702
1703 /* get a working reference to the output device */
1704 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1705 if (!out_dev) {
1706 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1707 return -EINVAL;
1708 }
1709
1710 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1711 in_dev->dev, in_dev, &itag);
1712 if (err < 0) {
1713 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1714 saddr);
1715
1716 goto cleanup;
1717 }
1718
1719 do_cache = res->fi && !itag;
1720 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1721 skb->protocol == htons(ETH_P_IP) &&
1722 (IN_DEV_SHARED_MEDIA(out_dev) ||
1723 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1724 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1725
1726 if (skb->protocol != htons(ETH_P_IP)) {
1727 /* Not IP (i.e. ARP). Do not create route, if it is
1728 * invalid for proxy arp. DNAT routes are always valid.
1729 *
1730 * Proxy arp feature have been extended to allow, ARP
1731 * replies back to the same interface, to support
1732 * Private VLAN switch technologies. See arp.c.
1733 */
1734 if (out_dev == in_dev &&
1735 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1736 err = -EINVAL;
1737 goto cleanup;
1738 }
1739 }
1740
1741 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1742 if (do_cache) {
1743 if (fnhe)
1744 rth = rcu_dereference(fnhe->fnhe_rth_input);
1745 else
1746 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1747 if (rt_cache_valid(rth)) {
1748 skb_dst_set_noref(skb, &rth->dst);
1749 goto out;
1750 }
1751 }
1752
1753 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1754 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1755 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1756 if (!rth) {
1757 err = -ENOBUFS;
1758 goto cleanup;
1759 }
1760
1761 rth->rt_is_input = 1;
1762 if (res->table)
1763 rth->rt_table_id = res->table->tb_id;
1764 RT_CACHE_STAT_INC(in_slow_tot);
1765
1766 rth->dst.input = ip_forward;
1767
1768 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1769 do_cache);
1770 set_lwt_redirect(rth);
1771 skb_dst_set(skb, &rth->dst);
1772out:
1773 err = 0;
1774 cleanup:
1775 return err;
1776}
1777
1778#ifdef CONFIG_IP_ROUTE_MULTIPATH
1779/* To make ICMP packets follow the right flow, the multipath hash is
1780 * calculated from the inner IP addresses.
1781 */
1782static void ip_multipath_l3_keys(const struct sk_buff *skb,
1783 struct flow_keys *hash_keys)
1784{
1785 const struct iphdr *outer_iph = ip_hdr(skb);
1786 const struct iphdr *inner_iph;
1787 const struct icmphdr *icmph;
1788 struct iphdr _inner_iph;
1789 struct icmphdr _icmph;
1790
1791 hash_keys->addrs.v4addrs.src = outer_iph->saddr;
1792 hash_keys->addrs.v4addrs.dst = outer_iph->daddr;
1793 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1794 return;
1795
1796 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1797 return;
1798
1799 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1800 &_icmph);
1801 if (!icmph)
1802 return;
1803
1804 if (icmph->type != ICMP_DEST_UNREACH &&
1805 icmph->type != ICMP_REDIRECT &&
1806 icmph->type != ICMP_TIME_EXCEEDED &&
1807 icmph->type != ICMP_PARAMETERPROB)
1808 return;
1809
1810 inner_iph = skb_header_pointer(skb,
1811 outer_iph->ihl * 4 + sizeof(_icmph),
1812 sizeof(_inner_iph), &_inner_iph);
1813 if (!inner_iph)
1814 return;
1815 hash_keys->addrs.v4addrs.src = inner_iph->saddr;
1816 hash_keys->addrs.v4addrs.dst = inner_iph->daddr;
1817}
1818
1819/* if skb is set it will be used and fl4 can be NULL */
1820int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
1821 const struct sk_buff *skb)
1822{
1823 struct net *net = fi->fib_net;
1824 struct flow_keys hash_keys;
1825 u32 mhash;
1826
1827 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1828 case 0:
1829 memset(&hash_keys, 0, sizeof(hash_keys));
1830 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1831 if (skb) {
1832 ip_multipath_l3_keys(skb, &hash_keys);
1833 } else {
1834 hash_keys.addrs.v4addrs.src = fl4->saddr;
1835 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1836 }
1837 break;
1838 case 1:
1839 /* skb is currently provided only when forwarding */
1840 if (skb) {
1841 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1842 struct flow_keys keys;
1843
1844 /* short-circuit if we already have L4 hash present */
1845 if (skb->l4_hash)
1846 return skb_get_hash_raw(skb) >> 1;
1847 memset(&hash_keys, 0, sizeof(hash_keys));
1848 skb_flow_dissect_flow_keys(skb, &keys, flag);
1849
1850 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1851 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1852 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1853 hash_keys.ports.src = keys.ports.src;
1854 hash_keys.ports.dst = keys.ports.dst;
1855 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1856 } else {
1857 memset(&hash_keys, 0, sizeof(hash_keys));
1858 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1859 hash_keys.addrs.v4addrs.src = fl4->saddr;
1860 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1861 hash_keys.ports.src = fl4->fl4_sport;
1862 hash_keys.ports.dst = fl4->fl4_dport;
1863 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1864 }
1865 break;
1866 }
1867 mhash = flow_hash_from_keys(&hash_keys);
1868
1869 return mhash >> 1;
1870}
1871EXPORT_SYMBOL_GPL(fib_multipath_hash);
1872#endif /* CONFIG_IP_ROUTE_MULTIPATH */
1873
1874static int ip_mkroute_input(struct sk_buff *skb,
1875 struct fib_result *res,
1876 struct in_device *in_dev,
1877 __be32 daddr, __be32 saddr, u32 tos)
1878{
1879#ifdef CONFIG_IP_ROUTE_MULTIPATH
1880 if (res->fi && res->fi->fib_nhs > 1) {
1881 int h = fib_multipath_hash(res->fi, NULL, skb);
1882
1883 fib_select_multipath(res, h);
1884 }
1885#endif
1886
1887 /* create a routing cache entry */
1888 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1889}
1890
1891/*
1892 * NOTE. We drop all the packets that has local source
1893 * addresses, because every properly looped back packet
1894 * must have correct destination already attached by output routine.
1895 *
1896 * Such approach solves two big problems:
1897 * 1. Not simplex devices are handled properly.
1898 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1899 * called with rcu_read_lock()
1900 */
1901
1902static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1903 u8 tos, struct net_device *dev,
1904 struct fib_result *res)
1905{
1906 struct in_device *in_dev = __in_dev_get_rcu(dev);
1907 struct ip_tunnel_info *tun_info;
1908 struct flowi4 fl4;
1909 unsigned int flags = 0;
1910 u32 itag = 0;
1911 struct rtable *rth;
1912 int err = -EINVAL;
1913 struct net *net = dev_net(dev);
1914 bool do_cache;
1915
1916 /* IP on this device is disabled. */
1917
1918 if (!in_dev)
1919 goto out;
1920
1921 /* Check for the most weird martians, which can be not detected
1922 by fib_lookup.
1923 */
1924
1925 tun_info = skb_tunnel_info(skb);
1926 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1927 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1928 else
1929 fl4.flowi4_tun_key.tun_id = 0;
1930 skb_dst_drop(skb);
1931
1932 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1933 goto martian_source;
1934
1935 res->fi = NULL;
1936 res->table = NULL;
1937 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1938 goto brd_input;
1939
1940 /* Accept zero addresses only to limited broadcast;
1941 * I even do not know to fix it or not. Waiting for complains :-)
1942 */
1943 if (ipv4_is_zeronet(saddr))
1944 goto martian_source;
1945
1946 if (ipv4_is_zeronet(daddr))
1947 goto martian_destination;
1948
1949 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1950 * and call it once if daddr or/and saddr are loopback addresses
1951 */
1952 if (ipv4_is_loopback(daddr)) {
1953 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1954 goto martian_destination;
1955 } else if (ipv4_is_loopback(saddr)) {
1956 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1957 goto martian_source;
1958 }
1959
1960 /*
1961 * Now we are ready to route packet.
1962 */
1963 fl4.flowi4_oif = 0;
1964 fl4.flowi4_iif = dev->ifindex;
1965 fl4.flowi4_mark = skb->mark;
1966 fl4.flowi4_tos = tos;
1967 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1968 fl4.flowi4_flags = 0;
1969 fl4.daddr = daddr;
1970 fl4.saddr = saddr;
1971 fl4.flowi4_uid = sock_net_uid(net, NULL);
1972 err = fib_lookup(net, &fl4, res, 0);
1973 if (err != 0) {
1974 if (!IN_DEV_FORWARD(in_dev))
1975 err = -EHOSTUNREACH;
1976 goto no_route;
1977 }
1978
1979 if (res->type == RTN_BROADCAST)
1980 goto brd_input;
1981
1982 if (res->type == RTN_LOCAL) {
1983 err = fib_validate_source(skb, saddr, daddr, tos,
1984 0, dev, in_dev, &itag);
1985 if (err < 0)
1986 goto martian_source;
1987 goto local_input;
1988 }
1989
1990 if (!IN_DEV_FORWARD(in_dev)) {
1991 err = -EHOSTUNREACH;
1992 goto no_route;
1993 }
1994 if (res->type != RTN_UNICAST)
1995 goto martian_destination;
1996
1997 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1998out: return err;
1999
2000brd_input:
2001 if (skb->protocol != htons(ETH_P_IP))
2002 goto e_inval;
2003
2004 if (!ipv4_is_zeronet(saddr)) {
2005 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2006 in_dev, &itag);
2007 if (err < 0)
2008 goto martian_source;
2009 }
2010 flags |= RTCF_BROADCAST;
2011 res->type = RTN_BROADCAST;
2012 RT_CACHE_STAT_INC(in_brd);
2013
2014local_input:
2015 do_cache = false;
2016 if (res->fi) {
2017 if (!itag) {
2018 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2019 if (rt_cache_valid(rth)) {
2020 skb_dst_set_noref(skb, &rth->dst);
2021 err = 0;
2022 goto out;
2023 }
2024 do_cache = true;
2025 }
2026 }
2027
2028 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2029 flags | RTCF_LOCAL, res->type,
2030 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2031 if (!rth)
2032 goto e_nobufs;
2033
2034 rth->dst.output= ip_rt_bug;
2035#ifdef CONFIG_IP_ROUTE_CLASSID
2036 rth->dst.tclassid = itag;
2037#endif
2038 rth->rt_is_input = 1;
2039 if (res->table)
2040 rth->rt_table_id = res->table->tb_id;
2041
2042 RT_CACHE_STAT_INC(in_slow_tot);
2043 if (res->type == RTN_UNREACHABLE) {
2044 rth->dst.input= ip_error;
2045 rth->dst.error= -err;
2046 rth->rt_flags &= ~RTCF_LOCAL;
2047 }
2048
2049 if (do_cache) {
2050 struct fib_nh *nh = &FIB_RES_NH(*res);
2051
2052 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2053 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2054 WARN_ON(rth->dst.input == lwtunnel_input);
2055 rth->dst.lwtstate->orig_input = rth->dst.input;
2056 rth->dst.input = lwtunnel_input;
2057 }
2058
2059 if (unlikely(!rt_cache_route(nh, rth)))
2060 rt_add_uncached_list(rth);
2061 }
2062 skb_dst_set(skb, &rth->dst);
2063 err = 0;
2064 goto out;
2065
2066no_route:
2067 RT_CACHE_STAT_INC(in_no_route);
2068 res->type = RTN_UNREACHABLE;
2069 res->fi = NULL;
2070 res->table = NULL;
2071 goto local_input;
2072
2073 /*
2074 * Do not cache martian addresses: they should be logged (RFC1812)
2075 */
2076martian_destination:
2077 RT_CACHE_STAT_INC(in_martian_dst);
2078#ifdef CONFIG_IP_ROUTE_VERBOSE
2079 if (IN_DEV_LOG_MARTIANS(in_dev))
2080 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2081 &daddr, &saddr, dev->name);
2082#endif
2083
2084e_inval:
2085 err = -EINVAL;
2086 goto out;
2087
2088e_nobufs:
2089 err = -ENOBUFS;
2090 goto out;
2091
2092martian_source:
2093 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2094 goto out;
2095}
2096
2097int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2098 u8 tos, struct net_device *dev)
2099{
2100 struct fib_result res;
2101 int err;
2102
2103 tos &= IPTOS_RT_MASK;
2104 rcu_read_lock();
2105 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2106 rcu_read_unlock();
2107
2108 return err;
2109}
2110EXPORT_SYMBOL(ip_route_input_noref);
2111
2112/* called with rcu_read_lock held */
2113int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2114 u8 tos, struct net_device *dev, struct fib_result *res)
2115{
2116 /* Multicast recognition logic is moved from route cache to here.
2117 The problem was that too many Ethernet cards have broken/missing
2118 hardware multicast filters :-( As result the host on multicasting
2119 network acquires a lot of useless route cache entries, sort of
2120 SDR messages from all the world. Now we try to get rid of them.
2121 Really, provided software IP multicast filter is organized
2122 reasonably (at least, hashed), it does not result in a slowdown
2123 comparing with route cache reject entries.
2124 Note, that multicast routers are not affected, because
2125 route cache entry is created eventually.
2126 */
2127 if (ipv4_is_multicast(daddr)) {
2128 struct in_device *in_dev = __in_dev_get_rcu(dev);
2129 int our = 0;
2130 int err = -EINVAL;
2131
2132 if (!in_dev)
2133 return err;
2134 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2135 ip_hdr(skb)->protocol);
2136
2137 /* check l3 master if no match yet */
2138 if (!our && netif_is_l3_slave(dev)) {
2139 struct in_device *l3_in_dev;
2140
2141 l3_in_dev = __in_dev_get_rcu(skb->dev);
2142 if (l3_in_dev)
2143 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2144 ip_hdr(skb)->protocol);
2145 }
2146
2147 if (our
2148#ifdef CONFIG_IP_MROUTE
2149 ||
2150 (!ipv4_is_local_multicast(daddr) &&
2151 IN_DEV_MFORWARD(in_dev))
2152#endif
2153 ) {
2154 err = ip_route_input_mc(skb, daddr, saddr,
2155 tos, dev, our);
2156 }
2157 return err;
2158 }
2159
2160 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2161}
2162
2163/* called with rcu_read_lock() */
2164static struct rtable *__mkroute_output(const struct fib_result *res,
2165 const struct flowi4 *fl4, int orig_oif,
2166 struct net_device *dev_out,
2167 unsigned int flags)
2168{
2169 struct fib_info *fi = res->fi;
2170 struct fib_nh_exception *fnhe;
2171 struct in_device *in_dev;
2172 u16 type = res->type;
2173 struct rtable *rth;
2174 bool do_cache;
2175
2176 in_dev = __in_dev_get_rcu(dev_out);
2177 if (!in_dev)
2178 return ERR_PTR(-EINVAL);
2179
2180 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2181 if (ipv4_is_loopback(fl4->saddr) &&
2182 !(dev_out->flags & IFF_LOOPBACK) &&
2183 !netif_is_l3_master(dev_out))
2184 return ERR_PTR(-EINVAL);
2185
2186 if (ipv4_is_lbcast(fl4->daddr))
2187 type = RTN_BROADCAST;
2188 else if (ipv4_is_multicast(fl4->daddr))
2189 type = RTN_MULTICAST;
2190 else if (ipv4_is_zeronet(fl4->daddr))
2191 return ERR_PTR(-EINVAL);
2192
2193 if (dev_out->flags & IFF_LOOPBACK)
2194 flags |= RTCF_LOCAL;
2195
2196 do_cache = true;
2197 if (type == RTN_BROADCAST) {
2198 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2199 fi = NULL;
2200 } else if (type == RTN_MULTICAST) {
2201 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2202 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2203 fl4->flowi4_proto))
2204 flags &= ~RTCF_LOCAL;
2205 else
2206 do_cache = false;
2207 /* If multicast route do not exist use
2208 * default one, but do not gateway in this case.
2209 * Yes, it is hack.
2210 */
2211 if (fi && res->prefixlen < 4)
2212 fi = NULL;
2213 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2214 (orig_oif != dev_out->ifindex)) {
2215 /* For local routes that require a particular output interface
2216 * we do not want to cache the result. Caching the result
2217 * causes incorrect behaviour when there are multiple source
2218 * addresses on the interface, the end result being that if the
2219 * intended recipient is waiting on that interface for the
2220 * packet he won't receive it because it will be delivered on
2221 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2222 * be set to the loopback interface as well.
2223 */
2224 do_cache = false;
2225 }
2226
2227 fnhe = NULL;
2228 do_cache &= fi != NULL;
2229 if (fi) {
2230 struct rtable __rcu **prth;
2231 struct fib_nh *nh = &FIB_RES_NH(*res);
2232
2233 fnhe = find_exception(nh, fl4->daddr);
2234 if (!do_cache)
2235 goto add;
2236 if (fnhe) {
2237 prth = &fnhe->fnhe_rth_output;
2238 } else {
2239 if (unlikely(fl4->flowi4_flags &
2240 FLOWI_FLAG_KNOWN_NH &&
2241 !(nh->nh_gw &&
2242 nh->nh_scope == RT_SCOPE_LINK))) {
2243 do_cache = false;
2244 goto add;
2245 }
2246 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2247 }
2248 rth = rcu_dereference(*prth);
2249 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2250 return rth;
2251 }
2252
2253add:
2254 rth = rt_dst_alloc(dev_out, flags, type,
2255 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2256 IN_DEV_CONF_GET(in_dev, NOXFRM),
2257 do_cache);
2258 if (!rth)
2259 return ERR_PTR(-ENOBUFS);
2260
2261 rth->rt_iif = orig_oif;
2262 if (res->table)
2263 rth->rt_table_id = res->table->tb_id;
2264
2265 RT_CACHE_STAT_INC(out_slow_tot);
2266
2267 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2268 if (flags & RTCF_LOCAL &&
2269 !(dev_out->flags & IFF_LOOPBACK)) {
2270 rth->dst.output = ip_mc_output;
2271 RT_CACHE_STAT_INC(out_slow_mc);
2272 }
2273#ifdef CONFIG_IP_MROUTE
2274 if (type == RTN_MULTICAST) {
2275 if (IN_DEV_MFORWARD(in_dev) &&
2276 !ipv4_is_local_multicast(fl4->daddr)) {
2277 rth->dst.input = ip_mr_input;
2278 rth->dst.output = ip_mc_output;
2279 }
2280 }
2281#endif
2282 }
2283
2284 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2285 set_lwt_redirect(rth);
2286
2287 return rth;
2288}
2289
2290/*
2291 * Major route resolver routine.
2292 */
2293
2294struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2295 const struct sk_buff *skb)
2296{
2297 __u8 tos = RT_FL_TOS(fl4);
2298 struct fib_result res = {
2299 .type = RTN_UNSPEC,
2300 .fi = NULL,
2301 .table = NULL,
2302 .tclassid = 0,
2303 };
2304 struct rtable *rth;
2305
2306 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2307 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2308 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2309 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2310
2311 rcu_read_lock();
2312 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2313 rcu_read_unlock();
2314
2315 return rth;
2316}
2317EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2318
2319struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2320 struct fib_result *res,
2321 const struct sk_buff *skb)
2322{
2323 struct net_device *dev_out = NULL;
2324 int orig_oif = fl4->flowi4_oif;
2325 unsigned int flags = 0;
2326 struct rtable *rth;
2327 int err = -ENETUNREACH;
2328
2329 if (fl4->saddr) {
2330 rth = ERR_PTR(-EINVAL);
2331 if (ipv4_is_multicast(fl4->saddr) ||
2332 ipv4_is_lbcast(fl4->saddr) ||
2333 ipv4_is_zeronet(fl4->saddr))
2334 goto out;
2335
2336 /* I removed check for oif == dev_out->oif here.
2337 It was wrong for two reasons:
2338 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2339 is assigned to multiple interfaces.
2340 2. Moreover, we are allowed to send packets with saddr
2341 of another iface. --ANK
2342 */
2343
2344 if (fl4->flowi4_oif == 0 &&
2345 (ipv4_is_multicast(fl4->daddr) ||
2346 ipv4_is_lbcast(fl4->daddr))) {
2347 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2348 dev_out = __ip_dev_find(net, fl4->saddr, false);
2349 if (!dev_out)
2350 goto out;
2351
2352 /* Special hack: user can direct multicasts
2353 and limited broadcast via necessary interface
2354 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2355 This hack is not just for fun, it allows
2356 vic,vat and friends to work.
2357 They bind socket to loopback, set ttl to zero
2358 and expect that it will work.
2359 From the viewpoint of routing cache they are broken,
2360 because we are not allowed to build multicast path
2361 with loopback source addr (look, routing cache
2362 cannot know, that ttl is zero, so that packet
2363 will not leave this host and route is valid).
2364 Luckily, this hack is good workaround.
2365 */
2366
2367 fl4->flowi4_oif = dev_out->ifindex;
2368 goto make_route;
2369 }
2370
2371 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2372 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2373 if (!__ip_dev_find(net, fl4->saddr, false))
2374 goto out;
2375 }
2376 }
2377
2378
2379 if (fl4->flowi4_oif) {
2380 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2381 rth = ERR_PTR(-ENODEV);
2382 if (!dev_out)
2383 goto out;
2384
2385 /* RACE: Check return value of inet_select_addr instead. */
2386 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2387 rth = ERR_PTR(-ENETUNREACH);
2388 goto out;
2389 }
2390 if (ipv4_is_local_multicast(fl4->daddr) ||
2391 ipv4_is_lbcast(fl4->daddr) ||
2392 fl4->flowi4_proto == IPPROTO_IGMP) {
2393 if (!fl4->saddr)
2394 fl4->saddr = inet_select_addr(dev_out, 0,
2395 RT_SCOPE_LINK);
2396 goto make_route;
2397 }
2398 if (!fl4->saddr) {
2399 if (ipv4_is_multicast(fl4->daddr))
2400 fl4->saddr = inet_select_addr(dev_out, 0,
2401 fl4->flowi4_scope);
2402 else if (!fl4->daddr)
2403 fl4->saddr = inet_select_addr(dev_out, 0,
2404 RT_SCOPE_HOST);
2405 }
2406 }
2407
2408 if (!fl4->daddr) {
2409 fl4->daddr = fl4->saddr;
2410 if (!fl4->daddr)
2411 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2412 dev_out = net->loopback_dev;
2413 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2414 res->type = RTN_LOCAL;
2415 flags |= RTCF_LOCAL;
2416 goto make_route;
2417 }
2418
2419 err = fib_lookup(net, fl4, res, 0);
2420 if (err) {
2421 res->fi = NULL;
2422 res->table = NULL;
2423 if (fl4->flowi4_oif &&
2424 (ipv4_is_multicast(fl4->daddr) ||
2425 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2426 /* Apparently, routing tables are wrong. Assume,
2427 that the destination is on link.
2428
2429 WHY? DW.
2430 Because we are allowed to send to iface
2431 even if it has NO routes and NO assigned
2432 addresses. When oif is specified, routing
2433 tables are looked up with only one purpose:
2434 to catch if destination is gatewayed, rather than
2435 direct. Moreover, if MSG_DONTROUTE is set,
2436 we send packet, ignoring both routing tables
2437 and ifaddr state. --ANK
2438
2439
2440 We could make it even if oif is unknown,
2441 likely IPv6, but we do not.
2442 */
2443
2444 if (fl4->saddr == 0)
2445 fl4->saddr = inet_select_addr(dev_out, 0,
2446 RT_SCOPE_LINK);
2447 res->type = RTN_UNICAST;
2448 goto make_route;
2449 }
2450 rth = ERR_PTR(err);
2451 goto out;
2452 }
2453
2454 if (res->type == RTN_LOCAL) {
2455 if (!fl4->saddr) {
2456 if (res->fi->fib_prefsrc)
2457 fl4->saddr = res->fi->fib_prefsrc;
2458 else
2459 fl4->saddr = fl4->daddr;
2460 }
2461
2462 /* L3 master device is the loopback for that domain */
2463 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2464 net->loopback_dev;
2465
2466 /* make sure orig_oif points to fib result device even
2467 * though packet rx/tx happens over loopback or l3mdev
2468 */
2469 orig_oif = FIB_RES_OIF(*res);
2470
2471 fl4->flowi4_oif = dev_out->ifindex;
2472 flags |= RTCF_LOCAL;
2473 goto make_route;
2474 }
2475
2476 fib_select_path(net, res, fl4, skb);
2477
2478 dev_out = FIB_RES_DEV(*res);
2479 fl4->flowi4_oif = dev_out->ifindex;
2480
2481
2482make_route:
2483 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2484
2485out:
2486 return rth;
2487}
2488
2489static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2490{
2491 return NULL;
2492}
2493
2494static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2495{
2496 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2497
2498 return mtu ? : dst->dev->mtu;
2499}
2500
2501static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2502 struct sk_buff *skb, u32 mtu)
2503{
2504}
2505
2506static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2507 struct sk_buff *skb)
2508{
2509}
2510
2511static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2512 unsigned long old)
2513{
2514 return NULL;
2515}
2516
2517static struct dst_ops ipv4_dst_blackhole_ops = {
2518 .family = AF_INET,
2519 .check = ipv4_blackhole_dst_check,
2520 .mtu = ipv4_blackhole_mtu,
2521 .default_advmss = ipv4_default_advmss,
2522 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2523 .redirect = ipv4_rt_blackhole_redirect,
2524 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2525 .neigh_lookup = ipv4_neigh_lookup,
2526};
2527
2528struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2529{
2530 struct rtable *ort = (struct rtable *) dst_orig;
2531 struct rtable *rt;
2532
2533 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2534 if (rt) {
2535 struct dst_entry *new = &rt->dst;
2536
2537 new->__use = 1;
2538 new->input = dst_discard;
2539 new->output = dst_discard_out;
2540
2541 new->dev = net->loopback_dev;
2542 if (new->dev)
2543 dev_hold(new->dev);
2544
2545 rt->rt_is_input = ort->rt_is_input;
2546 rt->rt_iif = ort->rt_iif;
2547 rt->rt_pmtu = ort->rt_pmtu;
2548 rt->rt_mtu_locked = ort->rt_mtu_locked;
2549
2550 rt->rt_genid = rt_genid_ipv4(net);
2551 rt->rt_flags = ort->rt_flags;
2552 rt->rt_type = ort->rt_type;
2553 rt->rt_gateway = ort->rt_gateway;
2554 rt->rt_uses_gateway = ort->rt_uses_gateway;
2555
2556 INIT_LIST_HEAD(&rt->rt_uncached);
2557 }
2558
2559 dst_release(dst_orig);
2560
2561 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2562}
2563
2564struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2565 const struct sock *sk)
2566{
2567 struct rtable *rt = __ip_route_output_key(net, flp4);
2568
2569 if (IS_ERR(rt))
2570 return rt;
2571
2572 if (flp4->flowi4_proto)
2573 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2574 flowi4_to_flowi(flp4),
2575 sk, 0);
2576
2577 return rt;
2578}
2579EXPORT_SYMBOL_GPL(ip_route_output_flow);
2580
2581/* called with rcu_read_lock held */
2582static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2583 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2584 u32 seq)
2585{
2586 struct rtable *rt = skb_rtable(skb);
2587 struct rtmsg *r;
2588 struct nlmsghdr *nlh;
2589 unsigned long expires = 0;
2590 u32 error;
2591 u32 metrics[RTAX_MAX];
2592
2593 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2594 if (!nlh)
2595 return -EMSGSIZE;
2596
2597 r = nlmsg_data(nlh);
2598 r->rtm_family = AF_INET;
2599 r->rtm_dst_len = 32;
2600 r->rtm_src_len = 0;
2601 r->rtm_tos = fl4->flowi4_tos;
2602 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2603 if (nla_put_u32(skb, RTA_TABLE, table_id))
2604 goto nla_put_failure;
2605 r->rtm_type = rt->rt_type;
2606 r->rtm_scope = RT_SCOPE_UNIVERSE;
2607 r->rtm_protocol = RTPROT_UNSPEC;
2608 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2609 if (rt->rt_flags & RTCF_NOTIFY)
2610 r->rtm_flags |= RTM_F_NOTIFY;
2611 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2612 r->rtm_flags |= RTCF_DOREDIRECT;
2613
2614 if (nla_put_in_addr(skb, RTA_DST, dst))
2615 goto nla_put_failure;
2616 if (src) {
2617 r->rtm_src_len = 32;
2618 if (nla_put_in_addr(skb, RTA_SRC, src))
2619 goto nla_put_failure;
2620 }
2621 if (rt->dst.dev &&
2622 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2623 goto nla_put_failure;
2624#ifdef CONFIG_IP_ROUTE_CLASSID
2625 if (rt->dst.tclassid &&
2626 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2627 goto nla_put_failure;
2628#endif
2629 if (!rt_is_input_route(rt) &&
2630 fl4->saddr != src) {
2631 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2632 goto nla_put_failure;
2633 }
2634 if (rt->rt_uses_gateway &&
2635 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2636 goto nla_put_failure;
2637
2638 expires = rt->dst.expires;
2639 if (expires) {
2640 unsigned long now = jiffies;
2641
2642 if (time_before(now, expires))
2643 expires -= now;
2644 else
2645 expires = 0;
2646 }
2647
2648 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2649 if (rt->rt_pmtu && expires)
2650 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2651 if (rt->rt_mtu_locked && expires)
2652 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2653 if (rtnetlink_put_metrics(skb, metrics) < 0)
2654 goto nla_put_failure;
2655
2656 if (fl4->flowi4_mark &&
2657 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2658 goto nla_put_failure;
2659
2660 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2661 nla_put_u32(skb, RTA_UID,
2662 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2663 goto nla_put_failure;
2664
2665 error = rt->dst.error;
2666
2667 if (rt_is_input_route(rt)) {
2668#ifdef CONFIG_IP_MROUTE
2669 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2670 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2671 int err = ipmr_get_route(net, skb,
2672 fl4->saddr, fl4->daddr,
2673 r, portid);
2674
2675 if (err <= 0) {
2676 if (err == 0)
2677 return 0;
2678 goto nla_put_failure;
2679 }
2680 } else
2681#endif
2682 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2683 goto nla_put_failure;
2684 }
2685
2686 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2687 goto nla_put_failure;
2688
2689 nlmsg_end(skb, nlh);
2690 return 0;
2691
2692nla_put_failure:
2693 nlmsg_cancel(skb, nlh);
2694 return -EMSGSIZE;
2695}
2696
2697static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2698 struct netlink_ext_ack *extack)
2699{
2700 struct net *net = sock_net(in_skb->sk);
2701 struct rtmsg *rtm;
2702 struct nlattr *tb[RTA_MAX+1];
2703 struct fib_result res = {};
2704 struct rtable *rt = NULL;
2705 struct flowi4 fl4;
2706 __be32 dst = 0;
2707 __be32 src = 0;
2708 u32 iif;
2709 int err;
2710 int mark;
2711 struct sk_buff *skb;
2712 u32 table_id = RT_TABLE_MAIN;
2713 kuid_t uid;
2714
2715 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2716 extack);
2717 if (err < 0)
2718 goto errout;
2719
2720 rtm = nlmsg_data(nlh);
2721
2722 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2723 if (!skb) {
2724 err = -ENOBUFS;
2725 goto errout;
2726 }
2727
2728 /* Reserve room for dummy headers, this skb can pass
2729 through good chunk of routing engine.
2730 */
2731 skb_reset_mac_header(skb);
2732 skb_reset_network_header(skb);
2733
2734 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2735 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2736 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2737 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2738 if (tb[RTA_UID])
2739 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2740 else
2741 uid = (iif ? INVALID_UID : current_uid());
2742
2743 /* Bugfix: need to give ip_route_input enough of an IP header to
2744 * not gag.
2745 */
2746 ip_hdr(skb)->protocol = IPPROTO_UDP;
2747 ip_hdr(skb)->saddr = src;
2748 ip_hdr(skb)->daddr = dst;
2749
2750 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2751
2752 memset(&fl4, 0, sizeof(fl4));
2753 fl4.daddr = dst;
2754 fl4.saddr = src;
2755 fl4.flowi4_tos = rtm->rtm_tos;
2756 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2757 fl4.flowi4_mark = mark;
2758 fl4.flowi4_uid = uid;
2759
2760 rcu_read_lock();
2761
2762 if (iif) {
2763 struct net_device *dev;
2764
2765 dev = dev_get_by_index_rcu(net, iif);
2766 if (!dev) {
2767 err = -ENODEV;
2768 goto errout_free;
2769 }
2770
2771 skb->protocol = htons(ETH_P_IP);
2772 skb->dev = dev;
2773 skb->mark = mark;
2774 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2775 dev, &res);
2776
2777 rt = skb_rtable(skb);
2778 if (err == 0 && rt->dst.error)
2779 err = -rt->dst.error;
2780 } else {
2781 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2782 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2783 err = 0;
2784 if (IS_ERR(rt))
2785 err = PTR_ERR(rt);
2786 else
2787 skb_dst_set(skb, &rt->dst);
2788 }
2789
2790 if (err)
2791 goto errout_free;
2792
2793 if (rtm->rtm_flags & RTM_F_NOTIFY)
2794 rt->rt_flags |= RTCF_NOTIFY;
2795
2796 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2797 table_id = rt->rt_table_id;
2798
2799 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2800 if (!res.fi) {
2801 err = fib_props[res.type].error;
2802 if (!err)
2803 err = -EHOSTUNREACH;
2804 goto errout_free;
2805 }
2806 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2807 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2808 rt->rt_type, res.prefix, res.prefixlen,
2809 fl4.flowi4_tos, res.fi, 0);
2810 } else {
2811 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2812 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2813 }
2814 if (err < 0)
2815 goto errout_free;
2816
2817 rcu_read_unlock();
2818
2819 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2820errout:
2821 return err;
2822
2823errout_free:
2824 rcu_read_unlock();
2825 kfree_skb(skb);
2826 goto errout;
2827}
2828
2829void ip_rt_multicast_event(struct in_device *in_dev)
2830{
2831 rt_cache_flush(dev_net(in_dev->dev));
2832}
2833
2834#ifdef CONFIG_SYSCTL
2835static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2836static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2837static int ip_rt_gc_elasticity __read_mostly = 8;
2838
2839static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2840 void __user *buffer,
2841 size_t *lenp, loff_t *ppos)
2842{
2843 struct net *net = (struct net *)__ctl->extra1;
2844
2845 if (write) {
2846 rt_cache_flush(net);
2847 fnhe_genid_bump(net);
2848 return 0;
2849 }
2850
2851 return -EINVAL;
2852}
2853
2854static struct ctl_table ipv4_route_table[] = {
2855 {
2856 .procname = "gc_thresh",
2857 .data = &ipv4_dst_ops.gc_thresh,
2858 .maxlen = sizeof(int),
2859 .mode = 0644,
2860 .proc_handler = proc_dointvec,
2861 },
2862 {
2863 .procname = "max_size",
2864 .data = &ip_rt_max_size,
2865 .maxlen = sizeof(int),
2866 .mode = 0644,
2867 .proc_handler = proc_dointvec,
2868 },
2869 {
2870 /* Deprecated. Use gc_min_interval_ms */
2871
2872 .procname = "gc_min_interval",
2873 .data = &ip_rt_gc_min_interval,
2874 .maxlen = sizeof(int),
2875 .mode = 0644,
2876 .proc_handler = proc_dointvec_jiffies,
2877 },
2878 {
2879 .procname = "gc_min_interval_ms",
2880 .data = &ip_rt_gc_min_interval,
2881 .maxlen = sizeof(int),
2882 .mode = 0644,
2883 .proc_handler = proc_dointvec_ms_jiffies,
2884 },
2885 {
2886 .procname = "gc_timeout",
2887 .data = &ip_rt_gc_timeout,
2888 .maxlen = sizeof(int),
2889 .mode = 0644,
2890 .proc_handler = proc_dointvec_jiffies,
2891 },
2892 {
2893 .procname = "gc_interval",
2894 .data = &ip_rt_gc_interval,
2895 .maxlen = sizeof(int),
2896 .mode = 0644,
2897 .proc_handler = proc_dointvec_jiffies,
2898 },
2899 {
2900 .procname = "redirect_load",
2901 .data = &ip_rt_redirect_load,
2902 .maxlen = sizeof(int),
2903 .mode = 0644,
2904 .proc_handler = proc_dointvec,
2905 },
2906 {
2907 .procname = "redirect_number",
2908 .data = &ip_rt_redirect_number,
2909 .maxlen = sizeof(int),
2910 .mode = 0644,
2911 .proc_handler = proc_dointvec,
2912 },
2913 {
2914 .procname = "redirect_silence",
2915 .data = &ip_rt_redirect_silence,
2916 .maxlen = sizeof(int),
2917 .mode = 0644,
2918 .proc_handler = proc_dointvec,
2919 },
2920 {
2921 .procname = "error_cost",
2922 .data = &ip_rt_error_cost,
2923 .maxlen = sizeof(int),
2924 .mode = 0644,
2925 .proc_handler = proc_dointvec,
2926 },
2927 {
2928 .procname = "error_burst",
2929 .data = &ip_rt_error_burst,
2930 .maxlen = sizeof(int),
2931 .mode = 0644,
2932 .proc_handler = proc_dointvec,
2933 },
2934 {
2935 .procname = "gc_elasticity",
2936 .data = &ip_rt_gc_elasticity,
2937 .maxlen = sizeof(int),
2938 .mode = 0644,
2939 .proc_handler = proc_dointvec,
2940 },
2941 {
2942 .procname = "mtu_expires",
2943 .data = &ip_rt_mtu_expires,
2944 .maxlen = sizeof(int),
2945 .mode = 0644,
2946 .proc_handler = proc_dointvec_jiffies,
2947 },
2948 {
2949 .procname = "min_pmtu",
2950 .data = &ip_rt_min_pmtu,
2951 .maxlen = sizeof(int),
2952 .mode = 0644,
2953 .proc_handler = proc_dointvec_minmax,
2954 .extra1 = &ip_min_valid_pmtu,
2955 },
2956 {
2957 .procname = "min_adv_mss",
2958 .data = &ip_rt_min_advmss,
2959 .maxlen = sizeof(int),
2960 .mode = 0644,
2961 .proc_handler = proc_dointvec,
2962 },
2963 { }
2964};
2965
2966static struct ctl_table ipv4_route_flush_table[] = {
2967 {
2968 .procname = "flush",
2969 .maxlen = sizeof(int),
2970 .mode = 0200,
2971 .proc_handler = ipv4_sysctl_rtcache_flush,
2972 },
2973 { },
2974};
2975
2976static __net_init int sysctl_route_net_init(struct net *net)
2977{
2978 struct ctl_table *tbl;
2979
2980 tbl = ipv4_route_flush_table;
2981 if (!net_eq(net, &init_net)) {
2982 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2983 if (!tbl)
2984 goto err_dup;
2985
2986 /* Don't export sysctls to unprivileged users */
2987 if (net->user_ns != &init_user_ns)
2988 tbl[0].procname = NULL;
2989 }
2990 tbl[0].extra1 = net;
2991
2992 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2993 if (!net->ipv4.route_hdr)
2994 goto err_reg;
2995 return 0;
2996
2997err_reg:
2998 if (tbl != ipv4_route_flush_table)
2999 kfree(tbl);
3000err_dup:
3001 return -ENOMEM;
3002}
3003
3004static __net_exit void sysctl_route_net_exit(struct net *net)
3005{
3006 struct ctl_table *tbl;
3007
3008 tbl = net->ipv4.route_hdr->ctl_table_arg;
3009 unregister_net_sysctl_table(net->ipv4.route_hdr);
3010 BUG_ON(tbl == ipv4_route_flush_table);
3011 kfree(tbl);
3012}
3013
3014static __net_initdata struct pernet_operations sysctl_route_ops = {
3015 .init = sysctl_route_net_init,
3016 .exit = sysctl_route_net_exit,
3017};
3018#endif
3019
3020static __net_init int rt_genid_init(struct net *net)
3021{
3022 atomic_set(&net->ipv4.rt_genid, 0);
3023 atomic_set(&net->fnhe_genid, 0);
3024 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3025 return 0;
3026}
3027
3028static __net_initdata struct pernet_operations rt_genid_ops = {
3029 .init = rt_genid_init,
3030};
3031
3032static int __net_init ipv4_inetpeer_init(struct net *net)
3033{
3034 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3035
3036 if (!bp)
3037 return -ENOMEM;
3038 inet_peer_base_init(bp);
3039 net->ipv4.peers = bp;
3040 return 0;
3041}
3042
3043static void __net_exit ipv4_inetpeer_exit(struct net *net)
3044{
3045 struct inet_peer_base *bp = net->ipv4.peers;
3046
3047 net->ipv4.peers = NULL;
3048 inetpeer_invalidate_tree(bp);
3049 kfree(bp);
3050}
3051
3052static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3053 .init = ipv4_inetpeer_init,
3054 .exit = ipv4_inetpeer_exit,
3055};
3056
3057#ifdef CONFIG_IP_ROUTE_CLASSID
3058struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3059#endif /* CONFIG_IP_ROUTE_CLASSID */
3060
3061int __init ip_rt_init(void)
3062{
3063 int rc = 0;
3064 int cpu;
3065
3066 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
3067 if (!ip_idents)
3068 panic("IP: failed to allocate ip_idents\n");
3069
3070 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3071
3072 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3073 if (!ip_tstamps)
3074 panic("IP: failed to allocate ip_tstamps\n");
3075
3076 for_each_possible_cpu(cpu) {
3077 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3078
3079 INIT_LIST_HEAD(&ul->head);
3080 spin_lock_init(&ul->lock);
3081 }
3082#ifdef CONFIG_IP_ROUTE_CLASSID
3083 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3084 if (!ip_rt_acct)
3085 panic("IP: failed to allocate ip_rt_acct\n");
3086#endif
3087
3088 ipv4_dst_ops.kmem_cachep =
3089 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3090 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3091
3092 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3093
3094 if (dst_entries_init(&ipv4_dst_ops) < 0)
3095 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3096
3097 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3098 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3099
3100 ipv4_dst_ops.gc_thresh = ~0;
3101 ip_rt_max_size = INT_MAX;
3102
3103 devinet_init();
3104 ip_fib_init();
3105
3106 if (ip_rt_proc_init())
3107 pr_err("Unable to create route proc files\n");
3108#ifdef CONFIG_XFRM
3109 xfrm_init();
3110 xfrm4_init();
3111#endif
3112 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3113 RTNL_FLAG_DOIT_UNLOCKED);
3114
3115#ifdef CONFIG_SYSCTL
3116 register_pernet_subsys(&sysctl_route_ops);
3117#endif
3118 register_pernet_subsys(&rt_genid_ops);
3119 register_pernet_subsys(&ipv4_inetpeer_ops);
3120 return rc;
3121}
3122
3123#ifdef CONFIG_SYSCTL
3124/*
3125 * We really need to sanitize the damn ipv4 init order, then all
3126 * this nonsense will go away.
3127 */
3128void __init ip_static_sysctl_init(void)
3129{
3130 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3131}
3132#endif