1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * ROUTE - implementation of the IP router.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14 *
15 * Fixes:
16 * Alan Cox : Verify area fixes.
17 * Alan Cox : cli() protects routing changes
18 * Rui Oliveira : ICMP routing table updates
19 * (rco@di.uminho.pt) Routing table insertion and update
20 * Linus Torvalds : Rewrote bits to be sensible
21 * Alan Cox : Added BSD route gw semantics
22 * Alan Cox : Super /proc >4K
23 * Alan Cox : MTU in route table
24 * Alan Cox : MSS actually. Also added the window
25 * clamper.
26 * Sam Lantinga : Fixed route matching in rt_del()
27 * Alan Cox : Routing cache support.
28 * Alan Cox : Removed compatibility cruft.
29 * Alan Cox : RTF_REJECT support.
30 * Alan Cox : TCP irtt support.
31 * Jonathan Naylor : Added Metric support.
32 * Miquel van Smoorenburg : BSD API fixes.
33 * Miquel van Smoorenburg : Metrics.
34 * Alan Cox : Use __u32 properly
35 * Alan Cox : Aligned routing errors more closely with BSD
36 * our system is still very different.
37 * Alan Cox : Faster /proc handling
38 * Alexey Kuznetsov : Massive rework to support tree based routing,
39 * routing caches and better behaviour.
40 *
41 * Olaf Erb : irtt wasn't being copied right.
42 * Bjorn Ekwall : Kerneld route support.
43 * Alan Cox : Multicast fixed (I hope)
44 * Pavel Krauz : Limited broadcast fixed
45 * Mike McLagan : Routing by source
46 * Alexey Kuznetsov : End of old history. Split to fib.c and
47 * route.c and rewritten from scratch.
48 * Andi Kleen : Load-limit warning messages.
49 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
50 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
51 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
52 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
53 * Marc Boucher : routing by fwmark
54 * Robert Olsson : Added rt_cache statistics
55 * Arnaldo C. Melo : Convert proc stuff to seq_file
56 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
57 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
58 * Ilia Sotnikov : Removed TOS from hash calculations
59 */
60
61 #define pr_fmt(fmt) "IPv4: " fmt
62
63 #include <linux/module.h>
64 #include <linux/bitops.h>
65 #include <linux/kernel.h>
66 #include <linux/mm.h>
67 #include <linux/memblock.h>
68 #include <linux/socket.h>
69 #include <linux/errno.h>
70 #include <linux/in.h>
71 #include <linux/inet.h>
72 #include <linux/netdevice.h>
73 #include <linux/proc_fs.h>
74 #include <linux/init.h>
75 #include <linux/skbuff.h>
76 #include <linux/inetdevice.h>
77 #include <linux/igmp.h>
78 #include <linux/pkt_sched.h>
79 #include <linux/mroute.h>
80 #include <linux/netfilter_ipv4.h>
81 #include <linux/random.h>
82 #include <linux/rcupdate.h>
83 #include <linux/slab.h>
84 #include <linux/jhash.h>
85 #include <net/dst.h>
86 #include <net/dst_metadata.h>
87 #include <net/inet_dscp.h>
88 #include <net/net_namespace.h>
89 #include <net/ip.h>
90 #include <net/route.h>
91 #include <net/inetpeer.h>
92 #include <net/sock.h>
93 #include <net/ip_fib.h>
94 #include <net/nexthop.h>
95 #include <net/tcp.h>
96 #include <net/icmp.h>
97 #include <net/xfrm.h>
98 #include <net/lwtunnel.h>
99 #include <net/netevent.h>
100 #include <net/rtnetlink.h>
101 #ifdef CONFIG_SYSCTL
102 #include <linux/sysctl.h>
103 #endif
104 #include <net/secure_seq.h>
105 #include <net/ip_tunnels.h>
106
107 #include "fib_lookup.h"
108
109 #define RT_FL_TOS(oldflp4) \
110 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
111
112 #define RT_GC_TIMEOUT (300*HZ)
113
114 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
116 #define DEFAULT_MIN_ADVMSS 256
117 static int ip_rt_max_size;
118 static int ip_rt_redirect_number __read_mostly = 9;
119 static int ip_rt_redirect_load __read_mostly = HZ / 50;
120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
121 static int ip_rt_error_cost __read_mostly = HZ;
122 static int ip_rt_error_burst __read_mostly = 5 * HZ;
123
124 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125
126 /*
127 * Interface to generic destination cache.
128 */
129
130 INDIRECT_CALLABLE_SCOPE
131 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
132 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
133 INDIRECT_CALLABLE_SCOPE
134 unsigned int ipv4_mtu(const struct dst_entry *dst);
135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
136 static void ipv4_link_failure(struct sk_buff *skb);
137 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
138 struct sk_buff *skb, u32 mtu,
139 bool confirm_neigh);
140 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
141 struct sk_buff *skb);
142 static void ipv4_dst_destroy(struct dst_entry *dst);
143
ipv4_cow_metrics(struct dst_entry * dst,unsigned long old)144 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
145 {
146 WARN_ON(1);
147 return NULL;
148 }
149
150 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
151 struct sk_buff *skb,
152 const void *daddr);
153 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
154
155 static struct dst_ops ipv4_dst_ops = {
156 .family = AF_INET,
157 .check = ipv4_dst_check,
158 .default_advmss = ipv4_default_advmss,
159 .mtu = ipv4_mtu,
160 .cow_metrics = ipv4_cow_metrics,
161 .destroy = ipv4_dst_destroy,
162 .negative_advice = ipv4_negative_advice,
163 .link_failure = ipv4_link_failure,
164 .update_pmtu = ip_rt_update_pmtu,
165 .redirect = ip_do_redirect,
166 .local_out = __ip_local_out,
167 .neigh_lookup = ipv4_neigh_lookup,
168 .confirm_neigh = ipv4_confirm_neigh,
169 };
170
171 #define ECN_OR_COST(class) TC_PRIO_##class
172
173 const __u8 ip_tos2prio[16] = {
174 TC_PRIO_BESTEFFORT,
175 ECN_OR_COST(BESTEFFORT),
176 TC_PRIO_BESTEFFORT,
177 ECN_OR_COST(BESTEFFORT),
178 TC_PRIO_BULK,
179 ECN_OR_COST(BULK),
180 TC_PRIO_BULK,
181 ECN_OR_COST(BULK),
182 TC_PRIO_INTERACTIVE,
183 ECN_OR_COST(INTERACTIVE),
184 TC_PRIO_INTERACTIVE,
185 ECN_OR_COST(INTERACTIVE),
186 TC_PRIO_INTERACTIVE_BULK,
187 ECN_OR_COST(INTERACTIVE_BULK),
188 TC_PRIO_INTERACTIVE_BULK,
189 ECN_OR_COST(INTERACTIVE_BULK)
190 };
191 EXPORT_SYMBOL(ip_tos2prio);
192
193 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
195
196 #ifdef CONFIG_PROC_FS
rt_cache_seq_start(struct seq_file * seq,loff_t * pos)197 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
198 {
199 if (*pos)
200 return NULL;
201 return SEQ_START_TOKEN;
202 }
203
rt_cache_seq_next(struct seq_file * seq,void * v,loff_t * pos)204 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
205 {
206 ++*pos;
207 return NULL;
208 }
209
rt_cache_seq_stop(struct seq_file * seq,void * v)210 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
211 {
212 }
213
rt_cache_seq_show(struct seq_file * seq,void * v)214 static int rt_cache_seq_show(struct seq_file *seq, void *v)
215 {
216 if (v == SEQ_START_TOKEN)
217 seq_printf(seq, "%-127s\n",
218 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
219 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
220 "HHUptod\tSpecDst");
221 return 0;
222 }
223
224 static const struct seq_operations rt_cache_seq_ops = {
225 .start = rt_cache_seq_start,
226 .next = rt_cache_seq_next,
227 .stop = rt_cache_seq_stop,
228 .show = rt_cache_seq_show,
229 };
230
rt_cpu_seq_start(struct seq_file * seq,loff_t * pos)231 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
232 {
233 int cpu;
234
235 if (*pos == 0)
236 return SEQ_START_TOKEN;
237
238 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
239 if (!cpu_possible(cpu))
240 continue;
241 *pos = cpu+1;
242 return &per_cpu(rt_cache_stat, cpu);
243 }
244 return NULL;
245 }
246
rt_cpu_seq_next(struct seq_file * seq,void * v,loff_t * pos)247 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
248 {
249 int cpu;
250
251 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
252 if (!cpu_possible(cpu))
253 continue;
254 *pos = cpu+1;
255 return &per_cpu(rt_cache_stat, cpu);
256 }
257 (*pos)++;
258 return NULL;
259
260 }
261
rt_cpu_seq_stop(struct seq_file * seq,void * v)262 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
263 {
264
265 }
266
rt_cpu_seq_show(struct seq_file * seq,void * v)267 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
268 {
269 struct rt_cache_stat *st = v;
270
271 if (v == SEQ_START_TOKEN) {
272 seq_puts(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
273 return 0;
274 }
275
276 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x "
277 "%08x %08x %08x %08x %08x %08x "
278 "%08x %08x %08x %08x\n",
279 dst_entries_get_slow(&ipv4_dst_ops),
280 0, /* st->in_hit */
281 st->in_slow_tot,
282 st->in_slow_mc,
283 st->in_no_route,
284 st->in_brd,
285 st->in_martian_dst,
286 st->in_martian_src,
287
288 0, /* st->out_hit */
289 st->out_slow_tot,
290 st->out_slow_mc,
291
292 0, /* st->gc_total */
293 0, /* st->gc_ignored */
294 0, /* st->gc_goal_miss */
295 0, /* st->gc_dst_overflow */
296 0, /* st->in_hlist_search */
297 0 /* st->out_hlist_search */
298 );
299 return 0;
300 }
301
302 static const struct seq_operations rt_cpu_seq_ops = {
303 .start = rt_cpu_seq_start,
304 .next = rt_cpu_seq_next,
305 .stop = rt_cpu_seq_stop,
306 .show = rt_cpu_seq_show,
307 };
308
309 #ifdef CONFIG_IP_ROUTE_CLASSID
rt_acct_proc_show(struct seq_file * m,void * v)310 static int rt_acct_proc_show(struct seq_file *m, void *v)
311 {
312 struct ip_rt_acct *dst, *src;
313 unsigned int i, j;
314
315 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
316 if (!dst)
317 return -ENOMEM;
318
319 for_each_possible_cpu(i) {
320 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
321 for (j = 0; j < 256; j++) {
322 dst[j].o_bytes += src[j].o_bytes;
323 dst[j].o_packets += src[j].o_packets;
324 dst[j].i_bytes += src[j].i_bytes;
325 dst[j].i_packets += src[j].i_packets;
326 }
327 }
328
329 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
330 kfree(dst);
331 return 0;
332 }
333 #endif
334
ip_rt_do_proc_init(struct net * net)335 static int __net_init ip_rt_do_proc_init(struct net *net)
336 {
337 struct proc_dir_entry *pde;
338
339 pde = proc_create_seq("rt_cache", 0444, net->proc_net,
340 &rt_cache_seq_ops);
341 if (!pde)
342 goto err1;
343
344 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
345 &rt_cpu_seq_ops);
346 if (!pde)
347 goto err2;
348
349 #ifdef CONFIG_IP_ROUTE_CLASSID
350 pde = proc_create_single("rt_acct", 0, net->proc_net,
351 rt_acct_proc_show);
352 if (!pde)
353 goto err3;
354 #endif
355 return 0;
356
357 #ifdef CONFIG_IP_ROUTE_CLASSID
358 err3:
359 remove_proc_entry("rt_cache", net->proc_net_stat);
360 #endif
361 err2:
362 remove_proc_entry("rt_cache", net->proc_net);
363 err1:
364 return -ENOMEM;
365 }
366
ip_rt_do_proc_exit(struct net * net)367 static void __net_exit ip_rt_do_proc_exit(struct net *net)
368 {
369 remove_proc_entry("rt_cache", net->proc_net_stat);
370 remove_proc_entry("rt_cache", net->proc_net);
371 #ifdef CONFIG_IP_ROUTE_CLASSID
372 remove_proc_entry("rt_acct", net->proc_net);
373 #endif
374 }
375
376 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
377 .init = ip_rt_do_proc_init,
378 .exit = ip_rt_do_proc_exit,
379 };
380
ip_rt_proc_init(void)381 static int __init ip_rt_proc_init(void)
382 {
383 return register_pernet_subsys(&ip_rt_proc_ops);
384 }
385
386 #else
ip_rt_proc_init(void)387 static inline int ip_rt_proc_init(void)
388 {
389 return 0;
390 }
391 #endif /* CONFIG_PROC_FS */
392
rt_is_expired(const struct rtable * rth)393 static inline bool rt_is_expired(const struct rtable *rth)
394 {
395 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
396 }
397
rt_cache_flush(struct net * net)398 void rt_cache_flush(struct net *net)
399 {
400 rt_genid_bump_ipv4(net);
401 }
402
ipv4_neigh_lookup(const struct dst_entry * dst,struct sk_buff * skb,const void * daddr)403 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
404 struct sk_buff *skb,
405 const void *daddr)
406 {
407 const struct rtable *rt = container_of(dst, struct rtable, dst);
408 struct net_device *dev = dst->dev;
409 struct neighbour *n;
410
411 rcu_read_lock_bh();
412
413 if (likely(rt->rt_gw_family == AF_INET)) {
414 n = ip_neigh_gw4(dev, rt->rt_gw4);
415 } else if (rt->rt_gw_family == AF_INET6) {
416 n = ip_neigh_gw6(dev, &rt->rt_gw6);
417 } else {
418 __be32 pkey;
419
420 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
421 n = ip_neigh_gw4(dev, pkey);
422 }
423
424 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
425 n = NULL;
426
427 rcu_read_unlock_bh();
428
429 return n;
430 }
431
ipv4_confirm_neigh(const struct dst_entry * dst,const void * daddr)432 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
433 {
434 const struct rtable *rt = container_of(dst, struct rtable, dst);
435 struct net_device *dev = dst->dev;
436 const __be32 *pkey = daddr;
437
438 if (rt->rt_gw_family == AF_INET) {
439 pkey = (const __be32 *)&rt->rt_gw4;
440 } else if (rt->rt_gw_family == AF_INET6) {
441 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
442 } else if (!daddr ||
443 (rt->rt_flags &
444 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
445 return;
446 }
447 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
448 }
449
450 /* Hash tables of size 2048..262144 depending on RAM size.
451 * Each bucket uses 8 bytes.
452 */
453 static u32 ip_idents_mask __read_mostly;
454 static atomic_t *ip_idents __read_mostly;
455 static u32 *ip_tstamps __read_mostly;
456
457 /* In order to protect privacy, we add a perturbation to identifiers
458 * if one generator is seldom used. This makes hard for an attacker
459 * to infer how many packets were sent between two points in time.
460 */
ip_idents_reserve(u32 hash,int segs)461 static u32 ip_idents_reserve(u32 hash, int segs)
462 {
463 u32 bucket, old, now = (u32)jiffies;
464 atomic_t *p_id;
465 u32 *p_tstamp;
466 u32 delta = 0;
467
468 bucket = hash & ip_idents_mask;
469 p_tstamp = ip_tstamps + bucket;
470 p_id = ip_idents + bucket;
471 old = READ_ONCE(*p_tstamp);
472
473 if (old != now && cmpxchg(p_tstamp, old, now) == old)
474 delta = prandom_u32_max(now - old);
475
476 /* If UBSAN reports an error there, please make sure your compiler
477 * supports -fno-strict-overflow before reporting it that was a bug
478 * in UBSAN, and it has been fixed in GCC-8.
479 */
480 return atomic_add_return(segs + delta, p_id) - segs;
481 }
482
__ip_select_ident(struct net * net,struct iphdr * iph,int segs)483 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
484 {
485 u32 hash, id;
486
487 /* Note the following code is not safe, but this is okay. */
488 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
489 get_random_bytes(&net->ipv4.ip_id_key,
490 sizeof(net->ipv4.ip_id_key));
491
492 hash = siphash_3u32((__force u32)iph->daddr,
493 (__force u32)iph->saddr,
494 iph->protocol,
495 &net->ipv4.ip_id_key);
496 id = ip_idents_reserve(hash, segs);
497 iph->id = htons(id);
498 }
499 EXPORT_SYMBOL(__ip_select_ident);
500
ip_rt_fix_tos(struct flowi4 * fl4)501 static void ip_rt_fix_tos(struct flowi4 *fl4)
502 {
503 __u8 tos = RT_FL_TOS(fl4);
504
505 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
506 if (tos & RTO_ONLINK)
507 fl4->flowi4_scope = RT_SCOPE_LINK;
508 }
509
__build_flow_key(const struct net * net,struct flowi4 * fl4,const struct sock * sk,const struct iphdr * iph,int oif,__u8 tos,u8 prot,u32 mark,int flow_flags)510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
511 const struct sock *sk, const struct iphdr *iph,
512 int oif, __u8 tos, u8 prot, u32 mark,
513 int flow_flags)
514 {
515 __u8 scope = RT_SCOPE_UNIVERSE;
516
517 if (sk) {
518 const struct inet_sock *inet = inet_sk(sk);
519
520 oif = sk->sk_bound_dev_if;
521 mark = sk->sk_mark;
522 tos = ip_sock_rt_tos(sk);
523 scope = ip_sock_rt_scope(sk);
524 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
525 }
526
527 flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
528 prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
529 sock_net_uid(net, sk));
530 }
531
build_skb_flow_key(struct flowi4 * fl4,const struct sk_buff * skb,const struct sock * sk)532 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
533 const struct sock *sk)
534 {
535 const struct net *net = dev_net(skb->dev);
536 const struct iphdr *iph = ip_hdr(skb);
537 int oif = skb->dev->ifindex;
538 u8 prot = iph->protocol;
539 u32 mark = skb->mark;
540 __u8 tos = iph->tos;
541
542 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
543 }
544
build_sk_flow_key(struct flowi4 * fl4,const struct sock * sk)545 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
546 {
547 const struct inet_sock *inet = inet_sk(sk);
548 const struct ip_options_rcu *inet_opt;
549 __be32 daddr = inet->inet_daddr;
550
551 rcu_read_lock();
552 inet_opt = rcu_dereference(inet->inet_opt);
553 if (inet_opt && inet_opt->opt.srr)
554 daddr = inet_opt->opt.faddr;
555 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
556 ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
557 ip_sock_rt_scope(sk),
558 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
559 inet_sk_flowi_flags(sk),
560 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
561 rcu_read_unlock();
562 }
563
ip_rt_build_flow_key(struct flowi4 * fl4,const struct sock * sk,const struct sk_buff * skb)564 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
565 const struct sk_buff *skb)
566 {
567 if (skb)
568 build_skb_flow_key(fl4, skb, sk);
569 else
570 build_sk_flow_key(fl4, sk);
571 }
572
573 static DEFINE_SPINLOCK(fnhe_lock);
574
fnhe_flush_routes(struct fib_nh_exception * fnhe)575 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
576 {
577 struct rtable *rt;
578
579 rt = rcu_dereference(fnhe->fnhe_rth_input);
580 if (rt) {
581 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
582 dst_dev_put(&rt->dst);
583 dst_release(&rt->dst);
584 }
585 rt = rcu_dereference(fnhe->fnhe_rth_output);
586 if (rt) {
587 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
588 dst_dev_put(&rt->dst);
589 dst_release(&rt->dst);
590 }
591 }
592
fnhe_remove_oldest(struct fnhe_hash_bucket * hash)593 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
594 {
595 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
596 struct fib_nh_exception *fnhe, *oldest = NULL;
597
598 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
599 fnhe = rcu_dereference_protected(*fnhe_p,
600 lockdep_is_held(&fnhe_lock));
601 if (!fnhe)
602 break;
603 if (!oldest ||
604 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
605 oldest = fnhe;
606 oldest_p = fnhe_p;
607 }
608 }
609 fnhe_flush_routes(oldest);
610 *oldest_p = oldest->fnhe_next;
611 kfree_rcu(oldest, rcu);
612 }
613
fnhe_hashfun(__be32 daddr)614 static u32 fnhe_hashfun(__be32 daddr)
615 {
616 static siphash_aligned_key_t fnhe_hash_key;
617 u64 hval;
618
619 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
620 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
621 return hash_64(hval, FNHE_HASH_SHIFT);
622 }
623
fill_route_from_fnhe(struct rtable * rt,struct fib_nh_exception * fnhe)624 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
625 {
626 rt->rt_pmtu = fnhe->fnhe_pmtu;
627 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
628 rt->dst.expires = fnhe->fnhe_expires;
629
630 if (fnhe->fnhe_gw) {
631 rt->rt_flags |= RTCF_REDIRECTED;
632 rt->rt_uses_gateway = 1;
633 rt->rt_gw_family = AF_INET;
634 rt->rt_gw4 = fnhe->fnhe_gw;
635 }
636 }
637
update_or_create_fnhe(struct fib_nh_common * nhc,__be32 daddr,__be32 gw,u32 pmtu,bool lock,unsigned long expires)638 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
639 __be32 gw, u32 pmtu, bool lock,
640 unsigned long expires)
641 {
642 struct fnhe_hash_bucket *hash;
643 struct fib_nh_exception *fnhe;
644 struct rtable *rt;
645 u32 genid, hval;
646 unsigned int i;
647 int depth;
648
649 genid = fnhe_genid(dev_net(nhc->nhc_dev));
650 hval = fnhe_hashfun(daddr);
651
652 spin_lock_bh(&fnhe_lock);
653
654 hash = rcu_dereference(nhc->nhc_exceptions);
655 if (!hash) {
656 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
657 if (!hash)
658 goto out_unlock;
659 rcu_assign_pointer(nhc->nhc_exceptions, hash);
660 }
661
662 hash += hval;
663
664 depth = 0;
665 for (fnhe = rcu_dereference(hash->chain); fnhe;
666 fnhe = rcu_dereference(fnhe->fnhe_next)) {
667 if (fnhe->fnhe_daddr == daddr)
668 break;
669 depth++;
670 }
671
672 if (fnhe) {
673 if (fnhe->fnhe_genid != genid)
674 fnhe->fnhe_genid = genid;
675 if (gw)
676 fnhe->fnhe_gw = gw;
677 if (pmtu) {
678 fnhe->fnhe_pmtu = pmtu;
679 fnhe->fnhe_mtu_locked = lock;
680 }
681 fnhe->fnhe_expires = max(1UL, expires);
682 /* Update all cached dsts too */
683 rt = rcu_dereference(fnhe->fnhe_rth_input);
684 if (rt)
685 fill_route_from_fnhe(rt, fnhe);
686 rt = rcu_dereference(fnhe->fnhe_rth_output);
687 if (rt)
688 fill_route_from_fnhe(rt, fnhe);
689 } else {
690 /* Randomize max depth to avoid some side channels attacks. */
691 int max_depth = FNHE_RECLAIM_DEPTH +
692 prandom_u32_max(FNHE_RECLAIM_DEPTH);
693
694 while (depth > max_depth) {
695 fnhe_remove_oldest(hash);
696 depth--;
697 }
698
699 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
700 if (!fnhe)
701 goto out_unlock;
702
703 fnhe->fnhe_next = hash->chain;
704
705 fnhe->fnhe_genid = genid;
706 fnhe->fnhe_daddr = daddr;
707 fnhe->fnhe_gw = gw;
708 fnhe->fnhe_pmtu = pmtu;
709 fnhe->fnhe_mtu_locked = lock;
710 fnhe->fnhe_expires = max(1UL, expires);
711
712 rcu_assign_pointer(hash->chain, fnhe);
713
714 /* Exception created; mark the cached routes for the nexthop
715 * stale, so anyone caching it rechecks if this exception
716 * applies to them.
717 */
718 rt = rcu_dereference(nhc->nhc_rth_input);
719 if (rt)
720 rt->dst.obsolete = DST_OBSOLETE_KILL;
721
722 for_each_possible_cpu(i) {
723 struct rtable __rcu **prt;
724
725 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
726 rt = rcu_dereference(*prt);
727 if (rt)
728 rt->dst.obsolete = DST_OBSOLETE_KILL;
729 }
730 }
731
732 fnhe->fnhe_stamp = jiffies;
733
734 out_unlock:
735 spin_unlock_bh(&fnhe_lock);
736 }
737
__ip_do_redirect(struct rtable * rt,struct sk_buff * skb,struct flowi4 * fl4,bool kill_route)738 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
739 bool kill_route)
740 {
741 __be32 new_gw = icmp_hdr(skb)->un.gateway;
742 __be32 old_gw = ip_hdr(skb)->saddr;
743 struct net_device *dev = skb->dev;
744 struct in_device *in_dev;
745 struct fib_result res;
746 struct neighbour *n;
747 struct net *net;
748
749 switch (icmp_hdr(skb)->code & 7) {
750 case ICMP_REDIR_NET:
751 case ICMP_REDIR_NETTOS:
752 case ICMP_REDIR_HOST:
753 case ICMP_REDIR_HOSTTOS:
754 break;
755
756 default:
757 return;
758 }
759
760 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
761 return;
762
763 in_dev = __in_dev_get_rcu(dev);
764 if (!in_dev)
765 return;
766
767 net = dev_net(dev);
768 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
769 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
770 ipv4_is_zeronet(new_gw))
771 goto reject_redirect;
772
773 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
774 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
775 goto reject_redirect;
776 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
777 goto reject_redirect;
778 } else {
779 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
780 goto reject_redirect;
781 }
782
783 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
784 if (!n)
785 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
786 if (!IS_ERR(n)) {
787 if (!(n->nud_state & NUD_VALID)) {
788 neigh_event_send(n, NULL);
789 } else {
790 if (fib_lookup(net, fl4, &res, 0) == 0) {
791 struct fib_nh_common *nhc;
792
793 fib_select_path(net, &res, fl4, skb);
794 nhc = FIB_RES_NHC(res);
795 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
796 0, false,
797 jiffies + ip_rt_gc_timeout);
798 }
799 if (kill_route)
800 rt->dst.obsolete = DST_OBSOLETE_KILL;
801 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
802 }
803 neigh_release(n);
804 }
805 return;
806
807 reject_redirect:
808 #ifdef CONFIG_IP_ROUTE_VERBOSE
809 if (IN_DEV_LOG_MARTIANS(in_dev)) {
810 const struct iphdr *iph = (const struct iphdr *) skb->data;
811 __be32 daddr = iph->daddr;
812 __be32 saddr = iph->saddr;
813
814 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
815 " Advised path = %pI4 -> %pI4\n",
816 &old_gw, dev->name, &new_gw,
817 &saddr, &daddr);
818 }
819 #endif
820 ;
821 }
822
ip_do_redirect(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb)823 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
824 {
825 struct rtable *rt;
826 struct flowi4 fl4;
827 const struct iphdr *iph = (const struct iphdr *) skb->data;
828 struct net *net = dev_net(skb->dev);
829 int oif = skb->dev->ifindex;
830 u8 prot = iph->protocol;
831 u32 mark = skb->mark;
832 __u8 tos = iph->tos;
833
834 rt = (struct rtable *) dst;
835
836 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
837 __ip_do_redirect(rt, skb, &fl4, true);
838 }
839
ipv4_negative_advice(struct dst_entry * dst)840 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
841 {
842 struct rtable *rt = (struct rtable *)dst;
843 struct dst_entry *ret = dst;
844
845 if (rt) {
846 if (dst->obsolete > 0) {
847 ip_rt_put(rt);
848 ret = NULL;
849 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
850 rt->dst.expires) {
851 ip_rt_put(rt);
852 ret = NULL;
853 }
854 }
855 return ret;
856 }
857
858 /*
859 * Algorithm:
860 * 1. The first ip_rt_redirect_number redirects are sent
861 * with exponential backoff, then we stop sending them at all,
862 * assuming that the host ignores our redirects.
863 * 2. If we did not see packets requiring redirects
864 * during ip_rt_redirect_silence, we assume that the host
865 * forgot redirected route and start to send redirects again.
866 *
867 * This algorithm is much cheaper and more intelligent than dumb load limiting
868 * in icmp.c.
869 *
870 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
871 * and "frag. need" (breaks PMTU discovery) in icmp.c.
872 */
873
ip_rt_send_redirect(struct sk_buff * skb)874 void ip_rt_send_redirect(struct sk_buff *skb)
875 {
876 struct rtable *rt = skb_rtable(skb);
877 struct in_device *in_dev;
878 struct inet_peer *peer;
879 struct net *net;
880 int log_martians;
881 int vif;
882
883 rcu_read_lock();
884 in_dev = __in_dev_get_rcu(rt->dst.dev);
885 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
886 rcu_read_unlock();
887 return;
888 }
889 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
890 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
891 rcu_read_unlock();
892
893 net = dev_net(rt->dst.dev);
894 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
895 if (!peer) {
896 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
897 rt_nexthop(rt, ip_hdr(skb)->daddr));
898 return;
899 }
900
901 /* No redirected packets during ip_rt_redirect_silence;
902 * reset the algorithm.
903 */
904 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
905 peer->rate_tokens = 0;
906 peer->n_redirects = 0;
907 }
908
909 /* Too many ignored redirects; do not send anything
910 * set dst.rate_last to the last seen redirected packet.
911 */
912 if (peer->n_redirects >= ip_rt_redirect_number) {
913 peer->rate_last = jiffies;
914 goto out_put_peer;
915 }
916
917 /* Check for load limit; set rate_last to the latest sent
918 * redirect.
919 */
920 if (peer->n_redirects == 0 ||
921 time_after(jiffies,
922 (peer->rate_last +
923 (ip_rt_redirect_load << peer->n_redirects)))) {
924 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
925
926 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
927 peer->rate_last = jiffies;
928 ++peer->n_redirects;
929 #ifdef CONFIG_IP_ROUTE_VERBOSE
930 if (log_martians &&
931 peer->n_redirects == ip_rt_redirect_number)
932 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
933 &ip_hdr(skb)->saddr, inet_iif(skb),
934 &ip_hdr(skb)->daddr, &gw);
935 #endif
936 }
937 out_put_peer:
938 inet_putpeer(peer);
939 }
940
ip_error(struct sk_buff * skb)941 static int ip_error(struct sk_buff *skb)
942 {
943 struct rtable *rt = skb_rtable(skb);
944 struct net_device *dev = skb->dev;
945 struct in_device *in_dev;
946 struct inet_peer *peer;
947 unsigned long now;
948 struct net *net;
949 SKB_DR(reason);
950 bool send;
951 int code;
952
953 if (netif_is_l3_master(skb->dev)) {
954 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
955 if (!dev)
956 goto out;
957 }
958
959 in_dev = __in_dev_get_rcu(dev);
960
961 /* IP on this device is disabled. */
962 if (!in_dev)
963 goto out;
964
965 net = dev_net(rt->dst.dev);
966 if (!IN_DEV_FORWARD(in_dev)) {
967 switch (rt->dst.error) {
968 case EHOSTUNREACH:
969 SKB_DR_SET(reason, IP_INADDRERRORS);
970 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
971 break;
972
973 case ENETUNREACH:
974 SKB_DR_SET(reason, IP_INNOROUTES);
975 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
976 break;
977 }
978 goto out;
979 }
980
981 switch (rt->dst.error) {
982 case EINVAL:
983 default:
984 goto out;
985 case EHOSTUNREACH:
986 code = ICMP_HOST_UNREACH;
987 break;
988 case ENETUNREACH:
989 code = ICMP_NET_UNREACH;
990 SKB_DR_SET(reason, IP_INNOROUTES);
991 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
992 break;
993 case EACCES:
994 code = ICMP_PKT_FILTERED;
995 break;
996 }
997
998 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
999 l3mdev_master_ifindex(skb->dev), 1);
1000
1001 send = true;
1002 if (peer) {
1003 now = jiffies;
1004 peer->rate_tokens += now - peer->rate_last;
1005 if (peer->rate_tokens > ip_rt_error_burst)
1006 peer->rate_tokens = ip_rt_error_burst;
1007 peer->rate_last = now;
1008 if (peer->rate_tokens >= ip_rt_error_cost)
1009 peer->rate_tokens -= ip_rt_error_cost;
1010 else
1011 send = false;
1012 inet_putpeer(peer);
1013 }
1014 if (send)
1015 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1016
1017 out: kfree_skb_reason(skb, reason);
1018 return 0;
1019 }
1020
__ip_rt_update_pmtu(struct rtable * rt,struct flowi4 * fl4,u32 mtu)1021 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1022 {
1023 struct dst_entry *dst = &rt->dst;
1024 struct net *net = dev_net(dst->dev);
1025 struct fib_result res;
1026 bool lock = false;
1027 u32 old_mtu;
1028
1029 if (ip_mtu_locked(dst))
1030 return;
1031
1032 old_mtu = ipv4_mtu(dst);
1033 if (old_mtu < mtu)
1034 return;
1035
1036 if (mtu < net->ipv4.ip_rt_min_pmtu) {
1037 lock = true;
1038 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1039 }
1040
1041 if (rt->rt_pmtu == mtu && !lock &&
1042 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1043 return;
1044
1045 rcu_read_lock();
1046 if (fib_lookup(net, fl4, &res, 0) == 0) {
1047 struct fib_nh_common *nhc;
1048
1049 fib_select_path(net, &res, fl4, NULL);
1050 nhc = FIB_RES_NHC(res);
1051 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1052 jiffies + net->ipv4.ip_rt_mtu_expires);
1053 }
1054 rcu_read_unlock();
1055 }
1056
ip_rt_update_pmtu(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb,u32 mtu,bool confirm_neigh)1057 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1058 struct sk_buff *skb, u32 mtu,
1059 bool confirm_neigh)
1060 {
1061 struct rtable *rt = (struct rtable *) dst;
1062 struct flowi4 fl4;
1063
1064 ip_rt_build_flow_key(&fl4, sk, skb);
1065
1066 /* Don't make lookup fail for bridged encapsulations */
1067 if (skb && netif_is_any_bridge_port(skb->dev))
1068 fl4.flowi4_oif = 0;
1069
1070 __ip_rt_update_pmtu(rt, &fl4, mtu);
1071 }
1072
ipv4_update_pmtu(struct sk_buff * skb,struct net * net,u32 mtu,int oif,u8 protocol)1073 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1074 int oif, u8 protocol)
1075 {
1076 const struct iphdr *iph = (const struct iphdr *)skb->data;
1077 struct flowi4 fl4;
1078 struct rtable *rt;
1079 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1080
1081 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1082 0);
1083 rt = __ip_route_output_key(net, &fl4);
1084 if (!IS_ERR(rt)) {
1085 __ip_rt_update_pmtu(rt, &fl4, mtu);
1086 ip_rt_put(rt);
1087 }
1088 }
1089 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1090
__ipv4_sk_update_pmtu(struct sk_buff * skb,struct sock * sk,u32 mtu)1091 static void __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
1097 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1098
1099 if (!fl4.flowi4_mark)
1100 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1101
1102 rt = __ip_route_output_key(sock_net(sk), &fl4);
1103 if (!IS_ERR(rt)) {
1104 __ip_rt_update_pmtu(rt, &fl4, mtu);
1105 ip_rt_put(rt);
1106 }
1107 }
1108
ipv4_sk_update_pmtu(struct sk_buff * skb,struct sock * sk,u32 mtu)1109 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1110 {
1111 const struct iphdr *iph = (const struct iphdr *)skb->data;
1112 struct flowi4 fl4;
1113 struct rtable *rt;
1114 struct dst_entry *odst = NULL;
1115 bool new = false;
1116 struct net *net = sock_net(sk);
1117
1118 bh_lock_sock(sk);
1119
1120 if (!ip_sk_accept_pmtu(sk))
1121 goto out;
1122
1123 odst = sk_dst_get(sk);
1124
1125 if (sock_owned_by_user(sk) || !odst) {
1126 __ipv4_sk_update_pmtu(skb, sk, mtu);
1127 goto out;
1128 }
1129
1130 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1131
1132 rt = (struct rtable *)odst;
1133 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1134 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1135 if (IS_ERR(rt))
1136 goto out;
1137
1138 new = true;
1139 }
1140
1141 __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1142
1143 if (!dst_check(&rt->dst, 0)) {
1144 if (new)
1145 dst_release(&rt->dst);
1146
1147 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1148 if (IS_ERR(rt))
1149 goto out;
1150
1151 new = true;
1152 }
1153
1154 if (new)
1155 sk_dst_set(sk, &rt->dst);
1156
1157 out:
1158 bh_unlock_sock(sk);
1159 dst_release(odst);
1160 }
1161 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1162
ipv4_redirect(struct sk_buff * skb,struct net * net,int oif,u8 protocol)1163 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1164 int oif, u8 protocol)
1165 {
1166 const struct iphdr *iph = (const struct iphdr *)skb->data;
1167 struct flowi4 fl4;
1168 struct rtable *rt;
1169
1170 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1171 rt = __ip_route_output_key(net, &fl4);
1172 if (!IS_ERR(rt)) {
1173 __ip_do_redirect(rt, skb, &fl4, false);
1174 ip_rt_put(rt);
1175 }
1176 }
1177 EXPORT_SYMBOL_GPL(ipv4_redirect);
1178
ipv4_sk_redirect(struct sk_buff * skb,struct sock * sk)1179 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1180 {
1181 const struct iphdr *iph = (const struct iphdr *)skb->data;
1182 struct flowi4 fl4;
1183 struct rtable *rt;
1184 struct net *net = sock_net(sk);
1185
1186 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1187 rt = __ip_route_output_key(net, &fl4);
1188 if (!IS_ERR(rt)) {
1189 __ip_do_redirect(rt, skb, &fl4, false);
1190 ip_rt_put(rt);
1191 }
1192 }
1193 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1194
ipv4_dst_check(struct dst_entry * dst,u32 cookie)1195 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1196 u32 cookie)
1197 {
1198 struct rtable *rt = (struct rtable *) dst;
1199
1200 /* All IPV4 dsts are created with ->obsolete set to the value
1201 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1202 * into this function always.
1203 *
1204 * When a PMTU/redirect information update invalidates a route,
1205 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1206 * DST_OBSOLETE_DEAD.
1207 */
1208 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1209 return NULL;
1210 return dst;
1211 }
1212 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1213
ipv4_send_dest_unreach(struct sk_buff * skb)1214 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1215 {
1216 struct ip_options opt;
1217 int res;
1218
1219 /* Recompile ip options since IPCB may not be valid anymore.
1220 * Also check we have a reasonable ipv4 header.
1221 */
1222 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1223 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1224 return;
1225
1226 memset(&opt, 0, sizeof(opt));
1227 if (ip_hdr(skb)->ihl > 5) {
1228 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1229 return;
1230 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1231
1232 rcu_read_lock();
1233 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1234 rcu_read_unlock();
1235
1236 if (res)
1237 return;
1238 }
1239 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1240 }
1241
ipv4_link_failure(struct sk_buff * skb)1242 static void ipv4_link_failure(struct sk_buff *skb)
1243 {
1244 struct rtable *rt;
1245
1246 ipv4_send_dest_unreach(skb);
1247
1248 rt = skb_rtable(skb);
1249 if (rt)
1250 dst_set_expires(&rt->dst, 0);
1251 }
1252
ip_rt_bug(struct net * net,struct sock * sk,struct sk_buff * skb)1253 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1254 {
1255 pr_debug("%s: %pI4 -> %pI4, %s\n",
1256 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1257 skb->dev ? skb->dev->name : "?");
1258 kfree_skb(skb);
1259 WARN_ON(1);
1260 return 0;
1261 }
1262
1263 /*
1264 * We do not cache source address of outgoing interface,
1265 * because it is used only by IP RR, TS and SRR options,
1266 * so that it out of fast path.
1267 *
1268 * BTW remember: "addr" is allowed to be not aligned
1269 * in IP options!
1270 */
1271
ip_rt_get_source(u8 * addr,struct sk_buff * skb,struct rtable * rt)1272 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1273 {
1274 __be32 src;
1275
1276 if (rt_is_output_route(rt))
1277 src = ip_hdr(skb)->saddr;
1278 else {
1279 struct fib_result res;
1280 struct iphdr *iph = ip_hdr(skb);
1281 struct flowi4 fl4 = {
1282 .daddr = iph->daddr,
1283 .saddr = iph->saddr,
1284 .flowi4_tos = RT_TOS(iph->tos),
1285 .flowi4_oif = rt->dst.dev->ifindex,
1286 .flowi4_iif = skb->dev->ifindex,
1287 .flowi4_mark = skb->mark,
1288 };
1289
1290 rcu_read_lock();
1291 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1292 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1293 else
1294 src = inet_select_addr(rt->dst.dev,
1295 rt_nexthop(rt, iph->daddr),
1296 RT_SCOPE_UNIVERSE);
1297 rcu_read_unlock();
1298 }
1299 memcpy(addr, &src, 4);
1300 }
1301
1302 #ifdef CONFIG_IP_ROUTE_CLASSID
set_class_tag(struct rtable * rt,u32 tag)1303 static void set_class_tag(struct rtable *rt, u32 tag)
1304 {
1305 if (!(rt->dst.tclassid & 0xFFFF))
1306 rt->dst.tclassid |= tag & 0xFFFF;
1307 if (!(rt->dst.tclassid & 0xFFFF0000))
1308 rt->dst.tclassid |= tag & 0xFFFF0000;
1309 }
1310 #endif
1311
ipv4_default_advmss(const struct dst_entry * dst)1312 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1313 {
1314 struct net *net = dev_net(dst->dev);
1315 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1316 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1317 net->ipv4.ip_rt_min_advmss);
1318
1319 return min(advmss, IPV4_MAX_PMTU - header_size);
1320 }
1321
ipv4_mtu(const struct dst_entry * dst)1322 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1323 {
1324 return ip_dst_mtu_maybe_forward(dst, false);
1325 }
1326 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1327
ip_del_fnhe(struct fib_nh_common * nhc,__be32 daddr)1328 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1329 {
1330 struct fnhe_hash_bucket *hash;
1331 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1332 u32 hval = fnhe_hashfun(daddr);
1333
1334 spin_lock_bh(&fnhe_lock);
1335
1336 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1337 lockdep_is_held(&fnhe_lock));
1338 hash += hval;
1339
1340 fnhe_p = &hash->chain;
1341 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1342 while (fnhe) {
1343 if (fnhe->fnhe_daddr == daddr) {
1344 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1345 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1346 /* set fnhe_daddr to 0 to ensure it won't bind with
1347 * new dsts in rt_bind_exception().
1348 */
1349 fnhe->fnhe_daddr = 0;
1350 fnhe_flush_routes(fnhe);
1351 kfree_rcu(fnhe, rcu);
1352 break;
1353 }
1354 fnhe_p = &fnhe->fnhe_next;
1355 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1356 lockdep_is_held(&fnhe_lock));
1357 }
1358
1359 spin_unlock_bh(&fnhe_lock);
1360 }
1361
find_exception(struct fib_nh_common * nhc,__be32 daddr)1362 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1363 __be32 daddr)
1364 {
1365 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1366 struct fib_nh_exception *fnhe;
1367 u32 hval;
1368
1369 if (!hash)
1370 return NULL;
1371
1372 hval = fnhe_hashfun(daddr);
1373
1374 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1375 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1376 if (fnhe->fnhe_daddr == daddr) {
1377 if (fnhe->fnhe_expires &&
1378 time_after(jiffies, fnhe->fnhe_expires)) {
1379 ip_del_fnhe(nhc, daddr);
1380 break;
1381 }
1382 return fnhe;
1383 }
1384 }
1385 return NULL;
1386 }
1387
1388 /* MTU selection:
1389 * 1. mtu on route is locked - use it
1390 * 2. mtu from nexthop exception
1391 * 3. mtu from egress device
1392 */
1393
ip_mtu_from_fib_result(struct fib_result * res,__be32 daddr)1394 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1395 {
1396 struct fib_nh_common *nhc = res->nhc;
1397 struct net_device *dev = nhc->nhc_dev;
1398 struct fib_info *fi = res->fi;
1399 u32 mtu = 0;
1400
1401 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1402 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1403 mtu = fi->fib_mtu;
1404
1405 if (likely(!mtu)) {
1406 struct fib_nh_exception *fnhe;
1407
1408 fnhe = find_exception(nhc, daddr);
1409 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1410 mtu = fnhe->fnhe_pmtu;
1411 }
1412
1413 if (likely(!mtu))
1414 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1415
1416 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1417 }
1418
rt_bind_exception(struct rtable * rt,struct fib_nh_exception * fnhe,__be32 daddr,const bool do_cache)1419 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1420 __be32 daddr, const bool do_cache)
1421 {
1422 bool ret = false;
1423
1424 spin_lock_bh(&fnhe_lock);
1425
1426 if (daddr == fnhe->fnhe_daddr) {
1427 struct rtable __rcu **porig;
1428 struct rtable *orig;
1429 int genid = fnhe_genid(dev_net(rt->dst.dev));
1430
1431 if (rt_is_input_route(rt))
1432 porig = &fnhe->fnhe_rth_input;
1433 else
1434 porig = &fnhe->fnhe_rth_output;
1435 orig = rcu_dereference(*porig);
1436
1437 if (fnhe->fnhe_genid != genid) {
1438 fnhe->fnhe_genid = genid;
1439 fnhe->fnhe_gw = 0;
1440 fnhe->fnhe_pmtu = 0;
1441 fnhe->fnhe_expires = 0;
1442 fnhe->fnhe_mtu_locked = false;
1443 fnhe_flush_routes(fnhe);
1444 orig = NULL;
1445 }
1446 fill_route_from_fnhe(rt, fnhe);
1447 if (!rt->rt_gw4) {
1448 rt->rt_gw4 = daddr;
1449 rt->rt_gw_family = AF_INET;
1450 }
1451
1452 if (do_cache) {
1453 dst_hold(&rt->dst);
1454 rcu_assign_pointer(*porig, rt);
1455 if (orig) {
1456 dst_dev_put(&orig->dst);
1457 dst_release(&orig->dst);
1458 }
1459 ret = true;
1460 }
1461
1462 fnhe->fnhe_stamp = jiffies;
1463 }
1464 spin_unlock_bh(&fnhe_lock);
1465
1466 return ret;
1467 }
1468
rt_cache_route(struct fib_nh_common * nhc,struct rtable * rt)1469 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1470 {
1471 struct rtable *orig, *prev, **p;
1472 bool ret = true;
1473
1474 if (rt_is_input_route(rt)) {
1475 p = (struct rtable **)&nhc->nhc_rth_input;
1476 } else {
1477 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1478 }
1479 orig = *p;
1480
1481 /* hold dst before doing cmpxchg() to avoid race condition
1482 * on this dst
1483 */
1484 dst_hold(&rt->dst);
1485 prev = cmpxchg(p, orig, rt);
1486 if (prev == orig) {
1487 if (orig) {
1488 rt_add_uncached_list(orig);
1489 dst_release(&orig->dst);
1490 }
1491 } else {
1492 dst_release(&rt->dst);
1493 ret = false;
1494 }
1495
1496 return ret;
1497 }
1498
1499 struct uncached_list {
1500 spinlock_t lock;
1501 struct list_head head;
1502 struct list_head quarantine;
1503 };
1504
1505 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1506
rt_add_uncached_list(struct rtable * rt)1507 void rt_add_uncached_list(struct rtable *rt)
1508 {
1509 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1510
1511 rt->rt_uncached_list = ul;
1512
1513 spin_lock_bh(&ul->lock);
1514 list_add_tail(&rt->rt_uncached, &ul->head);
1515 spin_unlock_bh(&ul->lock);
1516 }
1517
rt_del_uncached_list(struct rtable * rt)1518 void rt_del_uncached_list(struct rtable *rt)
1519 {
1520 if (!list_empty(&rt->rt_uncached)) {
1521 struct uncached_list *ul = rt->rt_uncached_list;
1522
1523 spin_lock_bh(&ul->lock);
1524 list_del_init(&rt->rt_uncached);
1525 spin_unlock_bh(&ul->lock);
1526 }
1527 }
1528
ipv4_dst_destroy(struct dst_entry * dst)1529 static void ipv4_dst_destroy(struct dst_entry *dst)
1530 {
1531 struct rtable *rt = (struct rtable *)dst;
1532
1533 ip_dst_metrics_put(dst);
1534 rt_del_uncached_list(rt);
1535 }
1536
rt_flush_dev(struct net_device * dev)1537 void rt_flush_dev(struct net_device *dev)
1538 {
1539 struct rtable *rt, *safe;
1540 int cpu;
1541
1542 for_each_possible_cpu(cpu) {
1543 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1544
1545 if (list_empty(&ul->head))
1546 continue;
1547
1548 spin_lock_bh(&ul->lock);
1549 list_for_each_entry_safe(rt, safe, &ul->head, rt_uncached) {
1550 if (rt->dst.dev != dev)
1551 continue;
1552 rt->dst.dev = blackhole_netdev;
1553 netdev_ref_replace(dev, blackhole_netdev,
1554 &rt->dst.dev_tracker, GFP_ATOMIC);
1555 list_move(&rt->rt_uncached, &ul->quarantine);
1556 }
1557 spin_unlock_bh(&ul->lock);
1558 }
1559 }
1560
rt_cache_valid(const struct rtable * rt)1561 static bool rt_cache_valid(const struct rtable *rt)
1562 {
1563 return rt &&
1564 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1565 !rt_is_expired(rt);
1566 }
1567
rt_set_nexthop(struct rtable * rt,__be32 daddr,const struct fib_result * res,struct fib_nh_exception * fnhe,struct fib_info * fi,u16 type,u32 itag,const bool do_cache)1568 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1569 const struct fib_result *res,
1570 struct fib_nh_exception *fnhe,
1571 struct fib_info *fi, u16 type, u32 itag,
1572 const bool do_cache)
1573 {
1574 bool cached = false;
1575
1576 if (fi) {
1577 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1578
1579 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1580 rt->rt_uses_gateway = 1;
1581 rt->rt_gw_family = nhc->nhc_gw_family;
1582 /* only INET and INET6 are supported */
1583 if (likely(nhc->nhc_gw_family == AF_INET))
1584 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1585 else
1586 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1587 }
1588
1589 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1590
1591 #ifdef CONFIG_IP_ROUTE_CLASSID
1592 if (nhc->nhc_family == AF_INET) {
1593 struct fib_nh *nh;
1594
1595 nh = container_of(nhc, struct fib_nh, nh_common);
1596 rt->dst.tclassid = nh->nh_tclassid;
1597 }
1598 #endif
1599 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1600 if (unlikely(fnhe))
1601 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1602 else if (do_cache)
1603 cached = rt_cache_route(nhc, rt);
1604 if (unlikely(!cached)) {
1605 /* Routes we intend to cache in nexthop exception or
1606 * FIB nexthop have the DST_NOCACHE bit clear.
1607 * However, if we are unsuccessful at storing this
1608 * route into the cache we really need to set it.
1609 */
1610 if (!rt->rt_gw4) {
1611 rt->rt_gw_family = AF_INET;
1612 rt->rt_gw4 = daddr;
1613 }
1614 rt_add_uncached_list(rt);
1615 }
1616 } else
1617 rt_add_uncached_list(rt);
1618
1619 #ifdef CONFIG_IP_ROUTE_CLASSID
1620 #ifdef CONFIG_IP_MULTIPLE_TABLES
1621 set_class_tag(rt, res->tclassid);
1622 #endif
1623 set_class_tag(rt, itag);
1624 #endif
1625 }
1626
rt_dst_alloc(struct net_device * dev,unsigned int flags,u16 type,bool noxfrm)1627 struct rtable *rt_dst_alloc(struct net_device *dev,
1628 unsigned int flags, u16 type,
1629 bool noxfrm)
1630 {
1631 struct rtable *rt;
1632
1633 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1634 (noxfrm ? DST_NOXFRM : 0));
1635
1636 if (rt) {
1637 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1638 rt->rt_flags = flags;
1639 rt->rt_type = type;
1640 rt->rt_is_input = 0;
1641 rt->rt_iif = 0;
1642 rt->rt_pmtu = 0;
1643 rt->rt_mtu_locked = 0;
1644 rt->rt_uses_gateway = 0;
1645 rt->rt_gw_family = 0;
1646 rt->rt_gw4 = 0;
1647 INIT_LIST_HEAD(&rt->rt_uncached);
1648
1649 rt->dst.output = ip_output;
1650 if (flags & RTCF_LOCAL)
1651 rt->dst.input = ip_local_deliver;
1652 }
1653
1654 return rt;
1655 }
1656 EXPORT_SYMBOL(rt_dst_alloc);
1657
rt_dst_clone(struct net_device * dev,struct rtable * rt)1658 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1659 {
1660 struct rtable *new_rt;
1661
1662 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1663 rt->dst.flags);
1664
1665 if (new_rt) {
1666 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1667 new_rt->rt_flags = rt->rt_flags;
1668 new_rt->rt_type = rt->rt_type;
1669 new_rt->rt_is_input = rt->rt_is_input;
1670 new_rt->rt_iif = rt->rt_iif;
1671 new_rt->rt_pmtu = rt->rt_pmtu;
1672 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1673 new_rt->rt_gw_family = rt->rt_gw_family;
1674 if (rt->rt_gw_family == AF_INET)
1675 new_rt->rt_gw4 = rt->rt_gw4;
1676 else if (rt->rt_gw_family == AF_INET6)
1677 new_rt->rt_gw6 = rt->rt_gw6;
1678 INIT_LIST_HEAD(&new_rt->rt_uncached);
1679
1680 new_rt->dst.input = rt->dst.input;
1681 new_rt->dst.output = rt->dst.output;
1682 new_rt->dst.error = rt->dst.error;
1683 new_rt->dst.lastuse = jiffies;
1684 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1685 }
1686 return new_rt;
1687 }
1688 EXPORT_SYMBOL(rt_dst_clone);
1689
1690 /* called in rcu_read_lock() section */
ip_mc_validate_source(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev,struct in_device * in_dev,u32 * itag)1691 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1692 u8 tos, struct net_device *dev,
1693 struct in_device *in_dev, u32 *itag)
1694 {
1695 int err;
1696
1697 /* Primary sanity checks. */
1698 if (!in_dev)
1699 return -EINVAL;
1700
1701 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1702 skb->protocol != htons(ETH_P_IP))
1703 return -EINVAL;
1704
1705 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1706 return -EINVAL;
1707
1708 if (ipv4_is_zeronet(saddr)) {
1709 if (!ipv4_is_local_multicast(daddr) &&
1710 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1711 return -EINVAL;
1712 } else {
1713 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1714 in_dev, itag);
1715 if (err < 0)
1716 return err;
1717 }
1718 return 0;
1719 }
1720
1721 /* called in rcu_read_lock() section */
ip_route_input_mc(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev,int our)1722 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1723 u8 tos, struct net_device *dev, int our)
1724 {
1725 struct in_device *in_dev = __in_dev_get_rcu(dev);
1726 unsigned int flags = RTCF_MULTICAST;
1727 struct rtable *rth;
1728 u32 itag = 0;
1729 int err;
1730
1731 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1732 if (err)
1733 return err;
1734
1735 if (our)
1736 flags |= RTCF_LOCAL;
1737
1738 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1739 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1740
1741 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1742 false);
1743 if (!rth)
1744 return -ENOBUFS;
1745
1746 #ifdef CONFIG_IP_ROUTE_CLASSID
1747 rth->dst.tclassid = itag;
1748 #endif
1749 rth->dst.output = ip_rt_bug;
1750 rth->rt_is_input= 1;
1751
1752 #ifdef CONFIG_IP_MROUTE
1753 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1754 rth->dst.input = ip_mr_input;
1755 #endif
1756 RT_CACHE_STAT_INC(in_slow_mc);
1757
1758 skb_dst_drop(skb);
1759 skb_dst_set(skb, &rth->dst);
1760 return 0;
1761 }
1762
1763
ip_handle_martian_source(struct net_device * dev,struct in_device * in_dev,struct sk_buff * skb,__be32 daddr,__be32 saddr)1764 static void ip_handle_martian_source(struct net_device *dev,
1765 struct in_device *in_dev,
1766 struct sk_buff *skb,
1767 __be32 daddr,
1768 __be32 saddr)
1769 {
1770 RT_CACHE_STAT_INC(in_martian_src);
1771 #ifdef CONFIG_IP_ROUTE_VERBOSE
1772 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1773 /*
1774 * RFC1812 recommendation, if source is martian,
1775 * the only hint is MAC header.
1776 */
1777 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1778 &daddr, &saddr, dev->name);
1779 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1780 print_hex_dump(KERN_WARNING, "ll header: ",
1781 DUMP_PREFIX_OFFSET, 16, 1,
1782 skb_mac_header(skb),
1783 dev->hard_header_len, false);
1784 }
1785 }
1786 #endif
1787 }
1788
1789 /* called in rcu_read_lock() section */
__mkroute_input(struct sk_buff * skb,const struct fib_result * res,struct in_device * in_dev,__be32 daddr,__be32 saddr,u32 tos)1790 static int __mkroute_input(struct sk_buff *skb,
1791 const struct fib_result *res,
1792 struct in_device *in_dev,
1793 __be32 daddr, __be32 saddr, u32 tos)
1794 {
1795 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1796 struct net_device *dev = nhc->nhc_dev;
1797 struct fib_nh_exception *fnhe;
1798 struct rtable *rth;
1799 int err;
1800 struct in_device *out_dev;
1801 bool do_cache;
1802 u32 itag = 0;
1803
1804 /* get a working reference to the output device */
1805 out_dev = __in_dev_get_rcu(dev);
1806 if (!out_dev) {
1807 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1808 return -EINVAL;
1809 }
1810
1811 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1812 in_dev->dev, in_dev, &itag);
1813 if (err < 0) {
1814 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1815 saddr);
1816
1817 goto cleanup;
1818 }
1819
1820 do_cache = res->fi && !itag;
1821 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1822 skb->protocol == htons(ETH_P_IP)) {
1823 __be32 gw;
1824
1825 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1826 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1827 inet_addr_onlink(out_dev, saddr, gw))
1828 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1829 }
1830
1831 if (skb->protocol != htons(ETH_P_IP)) {
1832 /* Not IP (i.e. ARP). Do not create route, if it is
1833 * invalid for proxy arp. DNAT routes are always valid.
1834 *
1835 * Proxy arp feature have been extended to allow, ARP
1836 * replies back to the same interface, to support
1837 * Private VLAN switch technologies. See arp.c.
1838 */
1839 if (out_dev == in_dev &&
1840 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1841 err = -EINVAL;
1842 goto cleanup;
1843 }
1844 }
1845
1846 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1847 IPCB(skb)->flags |= IPSKB_NOPOLICY;
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(out_dev, NOXFRM));
1863 if (!rth) {
1864 err = -ENOBUFS;
1865 goto cleanup;
1866 }
1867
1868 rth->rt_is_input = 1;
1869 RT_CACHE_STAT_INC(in_slow_tot);
1870
1871 rth->dst.input = ip_forward;
1872
1873 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1874 do_cache);
1875 lwtunnel_set_redirect(&rth->dst);
1876 skb_dst_set(skb, &rth->dst);
1877 out:
1878 err = 0;
1879 cleanup:
1880 return err;
1881 }
1882
1883 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1884 /* To make ICMP packets follow the right flow, the multipath hash is
1885 * calculated from the inner IP addresses.
1886 */
ip_multipath_l3_keys(const struct sk_buff * skb,struct flow_keys * hash_keys)1887 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1888 struct flow_keys *hash_keys)
1889 {
1890 const struct iphdr *outer_iph = ip_hdr(skb);
1891 const struct iphdr *key_iph = outer_iph;
1892 const struct iphdr *inner_iph;
1893 const struct icmphdr *icmph;
1894 struct iphdr _inner_iph;
1895 struct icmphdr _icmph;
1896
1897 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1898 goto out;
1899
1900 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1901 goto out;
1902
1903 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1904 &_icmph);
1905 if (!icmph)
1906 goto out;
1907
1908 if (!icmp_is_err(icmph->type))
1909 goto out;
1910
1911 inner_iph = skb_header_pointer(skb,
1912 outer_iph->ihl * 4 + sizeof(_icmph),
1913 sizeof(_inner_iph), &_inner_iph);
1914 if (!inner_iph)
1915 goto out;
1916
1917 key_iph = inner_iph;
1918 out:
1919 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1920 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1921 }
1922
fib_multipath_custom_hash_outer(const struct net * net,const struct sk_buff * skb,bool * p_has_inner)1923 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1924 const struct sk_buff *skb,
1925 bool *p_has_inner)
1926 {
1927 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1928 struct flow_keys keys, hash_keys;
1929
1930 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1931 return 0;
1932
1933 memset(&hash_keys, 0, sizeof(hash_keys));
1934 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1935
1936 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1937 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1938 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1939 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1940 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1941 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1942 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1943 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1944 hash_keys.ports.src = keys.ports.src;
1945 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1946 hash_keys.ports.dst = keys.ports.dst;
1947
1948 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1949 return flow_hash_from_keys(&hash_keys);
1950 }
1951
fib_multipath_custom_hash_inner(const struct net * net,const struct sk_buff * skb,bool has_inner)1952 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1953 const struct sk_buff *skb,
1954 bool has_inner)
1955 {
1956 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1957 struct flow_keys keys, hash_keys;
1958
1959 /* We assume the packet carries an encapsulation, but if none was
1960 * encountered during dissection of the outer flow, then there is no
1961 * point in calling the flow dissector again.
1962 */
1963 if (!has_inner)
1964 return 0;
1965
1966 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1967 return 0;
1968
1969 memset(&hash_keys, 0, sizeof(hash_keys));
1970 skb_flow_dissect_flow_keys(skb, &keys, 0);
1971
1972 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1973 return 0;
1974
1975 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1976 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1977 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1978 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1979 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1980 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1981 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1982 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1983 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1984 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1985 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1986 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1987 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1988 hash_keys.tags.flow_label = keys.tags.flow_label;
1989 }
1990
1991 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1992 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1993 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1994 hash_keys.ports.src = keys.ports.src;
1995 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1996 hash_keys.ports.dst = keys.ports.dst;
1997
1998 return flow_hash_from_keys(&hash_keys);
1999 }
2000
fib_multipath_custom_hash_skb(const struct net * net,const struct sk_buff * skb)2001 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2002 const struct sk_buff *skb)
2003 {
2004 u32 mhash, mhash_inner;
2005 bool has_inner = true;
2006
2007 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2008 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2009
2010 return jhash_2words(mhash, mhash_inner, 0);
2011 }
2012
fib_multipath_custom_hash_fl4(const struct net * net,const struct flowi4 * fl4)2013 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2014 const struct flowi4 *fl4)
2015 {
2016 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2017 struct flow_keys hash_keys;
2018
2019 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2020 return 0;
2021
2022 memset(&hash_keys, 0, sizeof(hash_keys));
2023 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2024 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2025 hash_keys.addrs.v4addrs.src = fl4->saddr;
2026 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2027 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2028 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2029 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2030 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2031 hash_keys.ports.src = fl4->fl4_sport;
2032 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2033 hash_keys.ports.dst = fl4->fl4_dport;
2034
2035 return flow_hash_from_keys(&hash_keys);
2036 }
2037
2038 /* if skb is set it will be used and fl4 can be NULL */
fib_multipath_hash(const struct net * net,const struct flowi4 * fl4,const struct sk_buff * skb,struct flow_keys * flkeys)2039 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2040 const struct sk_buff *skb, struct flow_keys *flkeys)
2041 {
2042 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2043 struct flow_keys hash_keys;
2044 u32 mhash = 0;
2045
2046 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2047 case 0:
2048 memset(&hash_keys, 0, sizeof(hash_keys));
2049 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2050 if (skb) {
2051 ip_multipath_l3_keys(skb, &hash_keys);
2052 } else {
2053 hash_keys.addrs.v4addrs.src = fl4->saddr;
2054 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2055 }
2056 mhash = flow_hash_from_keys(&hash_keys);
2057 break;
2058 case 1:
2059 /* skb is currently provided only when forwarding */
2060 if (skb) {
2061 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2062 struct flow_keys keys;
2063
2064 /* short-circuit if we already have L4 hash present */
2065 if (skb->l4_hash)
2066 return skb_get_hash_raw(skb) >> 1;
2067
2068 memset(&hash_keys, 0, sizeof(hash_keys));
2069
2070 if (!flkeys) {
2071 skb_flow_dissect_flow_keys(skb, &keys, flag);
2072 flkeys = &keys;
2073 }
2074
2075 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2076 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2077 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2078 hash_keys.ports.src = flkeys->ports.src;
2079 hash_keys.ports.dst = flkeys->ports.dst;
2080 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2081 } else {
2082 memset(&hash_keys, 0, sizeof(hash_keys));
2083 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2084 hash_keys.addrs.v4addrs.src = fl4->saddr;
2085 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2086 hash_keys.ports.src = fl4->fl4_sport;
2087 hash_keys.ports.dst = fl4->fl4_dport;
2088 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2089 }
2090 mhash = flow_hash_from_keys(&hash_keys);
2091 break;
2092 case 2:
2093 memset(&hash_keys, 0, sizeof(hash_keys));
2094 /* skb is currently provided only when forwarding */
2095 if (skb) {
2096 struct flow_keys keys;
2097
2098 skb_flow_dissect_flow_keys(skb, &keys, 0);
2099 /* Inner can be v4 or v6 */
2100 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2101 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2102 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2103 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2104 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2105 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2106 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2107 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2108 hash_keys.tags.flow_label = keys.tags.flow_label;
2109 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2110 } else {
2111 /* Same as case 0 */
2112 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2113 ip_multipath_l3_keys(skb, &hash_keys);
2114 }
2115 } else {
2116 /* Same as case 0 */
2117 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2118 hash_keys.addrs.v4addrs.src = fl4->saddr;
2119 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2120 }
2121 mhash = flow_hash_from_keys(&hash_keys);
2122 break;
2123 case 3:
2124 if (skb)
2125 mhash = fib_multipath_custom_hash_skb(net, skb);
2126 else
2127 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2128 break;
2129 }
2130
2131 if (multipath_hash)
2132 mhash = jhash_2words(mhash, multipath_hash, 0);
2133
2134 return mhash >> 1;
2135 }
2136 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2137
ip_mkroute_input(struct sk_buff * skb,struct fib_result * res,struct in_device * in_dev,__be32 daddr,__be32 saddr,u32 tos,struct flow_keys * hkeys)2138 static int ip_mkroute_input(struct sk_buff *skb,
2139 struct fib_result *res,
2140 struct in_device *in_dev,
2141 __be32 daddr, __be32 saddr, u32 tos,
2142 struct flow_keys *hkeys)
2143 {
2144 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2145 if (res->fi && fib_info_num_path(res->fi) > 1) {
2146 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2147
2148 fib_select_multipath(res, h);
2149 }
2150 #endif
2151
2152 /* create a routing cache entry */
2153 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2154 }
2155
2156 /* Implements all the saddr-related checks as ip_route_input_slow(),
2157 * assuming daddr is valid and the destination is not a local broadcast one.
2158 * Uses the provided hint instead of performing a route lookup.
2159 */
ip_route_use_hint(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev,const struct sk_buff * hint)2160 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2161 u8 tos, struct net_device *dev,
2162 const struct sk_buff *hint)
2163 {
2164 struct in_device *in_dev = __in_dev_get_rcu(dev);
2165 struct rtable *rt = skb_rtable(hint);
2166 struct net *net = dev_net(dev);
2167 int err = -EINVAL;
2168 u32 tag = 0;
2169
2170 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2171 goto martian_source;
2172
2173 if (ipv4_is_zeronet(saddr))
2174 goto martian_source;
2175
2176 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2177 goto martian_source;
2178
2179 if (rt->rt_type != RTN_LOCAL)
2180 goto skip_validate_source;
2181
2182 tos &= IPTOS_RT_MASK;
2183 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2184 if (err < 0)
2185 goto martian_source;
2186
2187 skip_validate_source:
2188 skb_dst_copy(skb, hint);
2189 return 0;
2190
2191 martian_source:
2192 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2193 return err;
2194 }
2195
2196 /* get device for dst_alloc with local routes */
ip_rt_get_dev(struct net * net,const struct fib_result * res)2197 static struct net_device *ip_rt_get_dev(struct net *net,
2198 const struct fib_result *res)
2199 {
2200 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2201 struct net_device *dev = NULL;
2202
2203 if (nhc)
2204 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2205
2206 return dev ? : net->loopback_dev;
2207 }
2208
2209 /*
2210 * NOTE. We drop all the packets that has local source
2211 * addresses, because every properly looped back packet
2212 * must have correct destination already attached by output routine.
2213 * Changes in the enforced policies must be applied also to
2214 * ip_route_use_hint().
2215 *
2216 * Such approach solves two big problems:
2217 * 1. Not simplex devices are handled properly.
2218 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2219 * called with rcu_read_lock()
2220 */
2221
ip_route_input_slow(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev,struct fib_result * res)2222 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2223 u8 tos, struct net_device *dev,
2224 struct fib_result *res)
2225 {
2226 struct in_device *in_dev = __in_dev_get_rcu(dev);
2227 struct flow_keys *flkeys = NULL, _flkeys;
2228 struct net *net = dev_net(dev);
2229 struct ip_tunnel_info *tun_info;
2230 int err = -EINVAL;
2231 unsigned int flags = 0;
2232 u32 itag = 0;
2233 struct rtable *rth;
2234 struct flowi4 fl4;
2235 bool do_cache = true;
2236
2237 /* IP on this device is disabled. */
2238
2239 if (!in_dev)
2240 goto out;
2241
2242 /* Check for the most weird martians, which can be not detected
2243 * by fib_lookup.
2244 */
2245
2246 tun_info = skb_tunnel_info(skb);
2247 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2248 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2249 else
2250 fl4.flowi4_tun_key.tun_id = 0;
2251 skb_dst_drop(skb);
2252
2253 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2254 goto martian_source;
2255
2256 res->fi = NULL;
2257 res->table = NULL;
2258 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2259 goto brd_input;
2260
2261 /* Accept zero addresses only to limited broadcast;
2262 * I even do not know to fix it or not. Waiting for complains :-)
2263 */
2264 if (ipv4_is_zeronet(saddr))
2265 goto martian_source;
2266
2267 if (ipv4_is_zeronet(daddr))
2268 goto martian_destination;
2269
2270 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2271 * and call it once if daddr or/and saddr are loopback addresses
2272 */
2273 if (ipv4_is_loopback(daddr)) {
2274 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2275 goto martian_destination;
2276 } else if (ipv4_is_loopback(saddr)) {
2277 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2278 goto martian_source;
2279 }
2280
2281 /*
2282 * Now we are ready to route packet.
2283 */
2284 fl4.flowi4_l3mdev = 0;
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
2335 make_route:
2336 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2337 out: return err;
2338
2339 brd_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
2353 local_input:
2354 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2355 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2356
2357 do_cache &= res->fi && !itag;
2358 if (do_cache) {
2359 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2360
2361 rth = rcu_dereference(nhc->nhc_rth_input);
2362 if (rt_cache_valid(rth)) {
2363 skb_dst_set_noref(skb, &rth->dst);
2364 err = 0;
2365 goto out;
2366 }
2367 }
2368
2369 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2370 flags | RTCF_LOCAL, res->type, false);
2371 if (!rth)
2372 goto e_nobufs;
2373
2374 rth->dst.output= ip_rt_bug;
2375 #ifdef CONFIG_IP_ROUTE_CLASSID
2376 rth->dst.tclassid = itag;
2377 #endif
2378 rth->rt_is_input = 1;
2379
2380 RT_CACHE_STAT_INC(in_slow_tot);
2381 if (res->type == RTN_UNREACHABLE) {
2382 rth->dst.input= ip_error;
2383 rth->dst.error= -err;
2384 rth->rt_flags &= ~RTCF_LOCAL;
2385 }
2386
2387 if (do_cache) {
2388 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2389
2390 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2391 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2392 WARN_ON(rth->dst.input == lwtunnel_input);
2393 rth->dst.lwtstate->orig_input = rth->dst.input;
2394 rth->dst.input = lwtunnel_input;
2395 }
2396
2397 if (unlikely(!rt_cache_route(nhc, rth)))
2398 rt_add_uncached_list(rth);
2399 }
2400 skb_dst_set(skb, &rth->dst);
2401 err = 0;
2402 goto out;
2403
2404 no_route:
2405 RT_CACHE_STAT_INC(in_no_route);
2406 res->type = RTN_UNREACHABLE;
2407 res->fi = NULL;
2408 res->table = NULL;
2409 goto local_input;
2410
2411 /*
2412 * Do not cache martian addresses: they should be logged (RFC1812)
2413 */
2414 martian_destination:
2415 RT_CACHE_STAT_INC(in_martian_dst);
2416 #ifdef CONFIG_IP_ROUTE_VERBOSE
2417 if (IN_DEV_LOG_MARTIANS(in_dev))
2418 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2419 &daddr, &saddr, dev->name);
2420 #endif
2421
2422 e_inval:
2423 err = -EINVAL;
2424 goto out;
2425
2426 e_nobufs:
2427 err = -ENOBUFS;
2428 goto out;
2429
2430 martian_source:
2431 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2432 goto out;
2433 }
2434
2435 /* called with rcu_read_lock held */
ip_route_input_rcu(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev,struct fib_result * res)2436 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2437 u8 tos, struct net_device *dev, struct fib_result *res)
2438 {
2439 /* Multicast recognition logic is moved from route cache to here.
2440 * The problem was that too many Ethernet cards have broken/missing
2441 * hardware multicast filters :-( As result the host on multicasting
2442 * network acquires a lot of useless route cache entries, sort of
2443 * SDR messages from all the world. Now we try to get rid of them.
2444 * Really, provided software IP multicast filter is organized
2445 * reasonably (at least, hashed), it does not result in a slowdown
2446 * comparing with route cache reject entries.
2447 * Note, that multicast routers are not affected, because
2448 * route cache entry is created eventually.
2449 */
2450 if (ipv4_is_multicast(daddr)) {
2451 struct in_device *in_dev = __in_dev_get_rcu(dev);
2452 int our = 0;
2453 int err = -EINVAL;
2454
2455 if (!in_dev)
2456 return err;
2457 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2458 ip_hdr(skb)->protocol);
2459
2460 /* check l3 master if no match yet */
2461 if (!our && netif_is_l3_slave(dev)) {
2462 struct in_device *l3_in_dev;
2463
2464 l3_in_dev = __in_dev_get_rcu(skb->dev);
2465 if (l3_in_dev)
2466 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2467 ip_hdr(skb)->protocol);
2468 }
2469
2470 if (our
2471 #ifdef CONFIG_IP_MROUTE
2472 ||
2473 (!ipv4_is_local_multicast(daddr) &&
2474 IN_DEV_MFORWARD(in_dev))
2475 #endif
2476 ) {
2477 err = ip_route_input_mc(skb, daddr, saddr,
2478 tos, dev, our);
2479 }
2480 return err;
2481 }
2482
2483 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2484 }
2485
ip_route_input_noref(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev)2486 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2487 u8 tos, struct net_device *dev)
2488 {
2489 struct fib_result res;
2490 int err;
2491
2492 tos &= IPTOS_RT_MASK;
2493 rcu_read_lock();
2494 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2495 rcu_read_unlock();
2496
2497 return err;
2498 }
2499 EXPORT_SYMBOL(ip_route_input_noref);
2500
2501 /* called with rcu_read_lock() */
__mkroute_output(const struct fib_result * res,const struct flowi4 * fl4,int orig_oif,struct net_device * dev_out,unsigned int flags)2502 static struct rtable *__mkroute_output(const struct fib_result *res,
2503 const struct flowi4 *fl4, int orig_oif,
2504 struct net_device *dev_out,
2505 unsigned int flags)
2506 {
2507 struct fib_info *fi = res->fi;
2508 struct fib_nh_exception *fnhe;
2509 struct in_device *in_dev;
2510 u16 type = res->type;
2511 struct rtable *rth;
2512 bool do_cache;
2513
2514 in_dev = __in_dev_get_rcu(dev_out);
2515 if (!in_dev)
2516 return ERR_PTR(-EINVAL);
2517
2518 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2519 if (ipv4_is_loopback(fl4->saddr) &&
2520 !(dev_out->flags & IFF_LOOPBACK) &&
2521 !netif_is_l3_master(dev_out))
2522 return ERR_PTR(-EINVAL);
2523
2524 if (ipv4_is_lbcast(fl4->daddr))
2525 type = RTN_BROADCAST;
2526 else if (ipv4_is_multicast(fl4->daddr))
2527 type = RTN_MULTICAST;
2528 else if (ipv4_is_zeronet(fl4->daddr))
2529 return ERR_PTR(-EINVAL);
2530
2531 if (dev_out->flags & IFF_LOOPBACK)
2532 flags |= RTCF_LOCAL;
2533
2534 do_cache = true;
2535 if (type == RTN_BROADCAST) {
2536 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2537 fi = NULL;
2538 } else if (type == RTN_MULTICAST) {
2539 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2540 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2541 fl4->flowi4_proto))
2542 flags &= ~RTCF_LOCAL;
2543 else
2544 do_cache = false;
2545 /* If multicast route do not exist use
2546 * default one, but do not gateway in this case.
2547 * Yes, it is hack.
2548 */
2549 if (fi && res->prefixlen < 4)
2550 fi = NULL;
2551 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2552 (orig_oif != dev_out->ifindex)) {
2553 /* For local routes that require a particular output interface
2554 * we do not want to cache the result. Caching the result
2555 * causes incorrect behaviour when there are multiple source
2556 * addresses on the interface, the end result being that if the
2557 * intended recipient is waiting on that interface for the
2558 * packet he won't receive it because it will be delivered on
2559 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2560 * be set to the loopback interface as well.
2561 */
2562 do_cache = false;
2563 }
2564
2565 fnhe = NULL;
2566 do_cache &= fi != NULL;
2567 if (fi) {
2568 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2569 struct rtable __rcu **prth;
2570
2571 fnhe = find_exception(nhc, fl4->daddr);
2572 if (!do_cache)
2573 goto add;
2574 if (fnhe) {
2575 prth = &fnhe->fnhe_rth_output;
2576 } else {
2577 if (unlikely(fl4->flowi4_flags &
2578 FLOWI_FLAG_KNOWN_NH &&
2579 !(nhc->nhc_gw_family &&
2580 nhc->nhc_scope == RT_SCOPE_LINK))) {
2581 do_cache = false;
2582 goto add;
2583 }
2584 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2585 }
2586 rth = rcu_dereference(*prth);
2587 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2588 return rth;
2589 }
2590
2591 add:
2592 rth = rt_dst_alloc(dev_out, flags, type,
2593 IN_DEV_ORCONF(in_dev, NOXFRM));
2594 if (!rth)
2595 return ERR_PTR(-ENOBUFS);
2596
2597 rth->rt_iif = orig_oif;
2598
2599 RT_CACHE_STAT_INC(out_slow_tot);
2600
2601 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2602 if (flags & RTCF_LOCAL &&
2603 !(dev_out->flags & IFF_LOOPBACK)) {
2604 rth->dst.output = ip_mc_output;
2605 RT_CACHE_STAT_INC(out_slow_mc);
2606 }
2607 #ifdef CONFIG_IP_MROUTE
2608 if (type == RTN_MULTICAST) {
2609 if (IN_DEV_MFORWARD(in_dev) &&
2610 !ipv4_is_local_multicast(fl4->daddr)) {
2611 rth->dst.input = ip_mr_input;
2612 rth->dst.output = ip_mc_output;
2613 }
2614 }
2615 #endif
2616 }
2617
2618 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2619 lwtunnel_set_redirect(&rth->dst);
2620
2621 return rth;
2622 }
2623
2624 /*
2625 * Major route resolver routine.
2626 */
2627
ip_route_output_key_hash(struct net * net,struct flowi4 * fl4,const struct sk_buff * skb)2628 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2629 const struct sk_buff *skb)
2630 {
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 ip_rt_fix_tos(fl4);
2641
2642 rcu_read_lock();
2643 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2644 rcu_read_unlock();
2645
2646 return rth;
2647 }
2648 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2649
ip_route_output_key_hash_rcu(struct net * net,struct flowi4 * fl4,struct fib_result * res,const struct sk_buff * skb)2650 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2651 struct fib_result *res,
2652 const struct sk_buff *skb)
2653 {
2654 struct net_device *dev_out = NULL;
2655 int orig_oif = fl4->flowi4_oif;
2656 unsigned int flags = 0;
2657 struct rtable *rth;
2658 int err;
2659
2660 if (fl4->saddr) {
2661 if (ipv4_is_multicast(fl4->saddr) ||
2662 ipv4_is_lbcast(fl4->saddr) ||
2663 ipv4_is_zeronet(fl4->saddr)) {
2664 rth = ERR_PTR(-EINVAL);
2665 goto out;
2666 }
2667
2668 rth = ERR_PTR(-ENETUNREACH);
2669
2670 /* I removed check for oif == dev_out->oif here.
2671 * It was wrong for two reasons:
2672 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2673 * is assigned to multiple interfaces.
2674 * 2. Moreover, we are allowed to send packets with saddr
2675 * of another iface. --ANK
2676 */
2677
2678 if (fl4->flowi4_oif == 0 &&
2679 (ipv4_is_multicast(fl4->daddr) ||
2680 ipv4_is_lbcast(fl4->daddr))) {
2681 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2682 dev_out = __ip_dev_find(net, fl4->saddr, false);
2683 if (!dev_out)
2684 goto out;
2685
2686 /* Special hack: user can direct multicasts
2687 * and limited broadcast via necessary interface
2688 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2689 * This hack is not just for fun, it allows
2690 * vic,vat and friends to work.
2691 * They bind socket to loopback, set ttl to zero
2692 * and expect that it will work.
2693 * From the viewpoint of routing cache they are broken,
2694 * because we are not allowed to build multicast path
2695 * with loopback source addr (look, routing cache
2696 * cannot know, that ttl is zero, so that packet
2697 * will not leave this host and route is valid).
2698 * Luckily, this hack is good workaround.
2699 */
2700
2701 fl4->flowi4_oif = dev_out->ifindex;
2702 goto make_route;
2703 }
2704
2705 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2706 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2707 if (!__ip_dev_find(net, fl4->saddr, false))
2708 goto out;
2709 }
2710 }
2711
2712
2713 if (fl4->flowi4_oif) {
2714 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2715 rth = ERR_PTR(-ENODEV);
2716 if (!dev_out)
2717 goto out;
2718
2719 /* RACE: Check return value of inet_select_addr instead. */
2720 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2721 rth = ERR_PTR(-ENETUNREACH);
2722 goto out;
2723 }
2724 if (ipv4_is_local_multicast(fl4->daddr) ||
2725 ipv4_is_lbcast(fl4->daddr) ||
2726 fl4->flowi4_proto == IPPROTO_IGMP) {
2727 if (!fl4->saddr)
2728 fl4->saddr = inet_select_addr(dev_out, 0,
2729 RT_SCOPE_LINK);
2730 goto make_route;
2731 }
2732 if (!fl4->saddr) {
2733 if (ipv4_is_multicast(fl4->daddr))
2734 fl4->saddr = inet_select_addr(dev_out, 0,
2735 fl4->flowi4_scope);
2736 else if (!fl4->daddr)
2737 fl4->saddr = inet_select_addr(dev_out, 0,
2738 RT_SCOPE_HOST);
2739 }
2740 }
2741
2742 if (!fl4->daddr) {
2743 fl4->daddr = fl4->saddr;
2744 if (!fl4->daddr)
2745 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2746 dev_out = net->loopback_dev;
2747 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2748 res->type = RTN_LOCAL;
2749 flags |= RTCF_LOCAL;
2750 goto make_route;
2751 }
2752
2753 err = fib_lookup(net, fl4, res, 0);
2754 if (err) {
2755 res->fi = NULL;
2756 res->table = NULL;
2757 if (fl4->flowi4_oif &&
2758 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2759 /* Apparently, routing tables are wrong. Assume,
2760 * that the destination is on link.
2761 *
2762 * WHY? DW.
2763 * Because we are allowed to send to iface
2764 * even if it has NO routes and NO assigned
2765 * addresses. When oif is specified, routing
2766 * tables are looked up with only one purpose:
2767 * to catch if destination is gatewayed, rather than
2768 * direct. Moreover, if MSG_DONTROUTE is set,
2769 * we send packet, ignoring both routing tables
2770 * and ifaddr state. --ANK
2771 *
2772 *
2773 * We could make it even if oif is unknown,
2774 * likely IPv6, but we do not.
2775 */
2776
2777 if (fl4->saddr == 0)
2778 fl4->saddr = inet_select_addr(dev_out, 0,
2779 RT_SCOPE_LINK);
2780 res->type = RTN_UNICAST;
2781 goto make_route;
2782 }
2783 rth = ERR_PTR(err);
2784 goto out;
2785 }
2786
2787 if (res->type == RTN_LOCAL) {
2788 if (!fl4->saddr) {
2789 if (res->fi->fib_prefsrc)
2790 fl4->saddr = res->fi->fib_prefsrc;
2791 else
2792 fl4->saddr = fl4->daddr;
2793 }
2794
2795 /* L3 master device is the loopback for that domain */
2796 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2797 net->loopback_dev;
2798
2799 /* make sure orig_oif points to fib result device even
2800 * though packet rx/tx happens over loopback or l3mdev
2801 */
2802 orig_oif = FIB_RES_OIF(*res);
2803
2804 fl4->flowi4_oif = dev_out->ifindex;
2805 flags |= RTCF_LOCAL;
2806 goto make_route;
2807 }
2808
2809 fib_select_path(net, res, fl4, skb);
2810
2811 dev_out = FIB_RES_DEV(*res);
2812
2813 make_route:
2814 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2815
2816 out:
2817 return rth;
2818 }
2819
2820 static struct dst_ops ipv4_dst_blackhole_ops = {
2821 .family = AF_INET,
2822 .default_advmss = ipv4_default_advmss,
2823 .neigh_lookup = ipv4_neigh_lookup,
2824 .check = dst_blackhole_check,
2825 .cow_metrics = dst_blackhole_cow_metrics,
2826 .update_pmtu = dst_blackhole_update_pmtu,
2827 .redirect = dst_blackhole_redirect,
2828 .mtu = dst_blackhole_mtu,
2829 };
2830
ipv4_blackhole_route(struct net * net,struct dst_entry * dst_orig)2831 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2832 {
2833 struct rtable *ort = (struct rtable *) dst_orig;
2834 struct rtable *rt;
2835
2836 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2837 if (rt) {
2838 struct dst_entry *new = &rt->dst;
2839
2840 new->__use = 1;
2841 new->input = dst_discard;
2842 new->output = dst_discard_out;
2843
2844 new->dev = net->loopback_dev;
2845 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2846
2847 rt->rt_is_input = ort->rt_is_input;
2848 rt->rt_iif = ort->rt_iif;
2849 rt->rt_pmtu = ort->rt_pmtu;
2850 rt->rt_mtu_locked = ort->rt_mtu_locked;
2851
2852 rt->rt_genid = rt_genid_ipv4(net);
2853 rt->rt_flags = ort->rt_flags;
2854 rt->rt_type = ort->rt_type;
2855 rt->rt_uses_gateway = ort->rt_uses_gateway;
2856 rt->rt_gw_family = ort->rt_gw_family;
2857 if (rt->rt_gw_family == AF_INET)
2858 rt->rt_gw4 = ort->rt_gw4;
2859 else if (rt->rt_gw_family == AF_INET6)
2860 rt->rt_gw6 = ort->rt_gw6;
2861
2862 INIT_LIST_HEAD(&rt->rt_uncached);
2863 }
2864
2865 dst_release(dst_orig);
2866
2867 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2868 }
2869
ip_route_output_flow(struct net * net,struct flowi4 * flp4,const struct sock * sk)2870 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2871 const struct sock *sk)
2872 {
2873 struct rtable *rt = __ip_route_output_key(net, flp4);
2874
2875 if (IS_ERR(rt))
2876 return rt;
2877
2878 if (flp4->flowi4_proto) {
2879 flp4->flowi4_oif = rt->dst.dev->ifindex;
2880 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2881 flowi4_to_flowi(flp4),
2882 sk, 0);
2883 }
2884
2885 return rt;
2886 }
2887 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2888
ip_route_output_tunnel(struct sk_buff * skb,struct net_device * dev,struct net * net,__be32 * saddr,const struct ip_tunnel_info * info,u8 protocol,bool use_cache)2889 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2890 struct net_device *dev,
2891 struct net *net, __be32 *saddr,
2892 const struct ip_tunnel_info *info,
2893 u8 protocol, bool use_cache)
2894 {
2895 #ifdef CONFIG_DST_CACHE
2896 struct dst_cache *dst_cache;
2897 #endif
2898 struct rtable *rt = NULL;
2899 struct flowi4 fl4;
2900 __u8 tos;
2901
2902 #ifdef CONFIG_DST_CACHE
2903 dst_cache = (struct dst_cache *)&info->dst_cache;
2904 if (use_cache) {
2905 rt = dst_cache_get_ip4(dst_cache, saddr);
2906 if (rt)
2907 return rt;
2908 }
2909 #endif
2910 memset(&fl4, 0, sizeof(fl4));
2911 fl4.flowi4_mark = skb->mark;
2912 fl4.flowi4_proto = protocol;
2913 fl4.daddr = info->key.u.ipv4.dst;
2914 fl4.saddr = info->key.u.ipv4.src;
2915 tos = info->key.tos;
2916 fl4.flowi4_tos = RT_TOS(tos);
2917
2918 rt = ip_route_output_key(net, &fl4);
2919 if (IS_ERR(rt)) {
2920 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2921 return ERR_PTR(-ENETUNREACH);
2922 }
2923 if (rt->dst.dev == dev) { /* is this necessary? */
2924 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2925 ip_rt_put(rt);
2926 return ERR_PTR(-ELOOP);
2927 }
2928 #ifdef CONFIG_DST_CACHE
2929 if (use_cache)
2930 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2931 #endif
2932 *saddr = fl4.saddr;
2933 return rt;
2934 }
2935 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2936
2937 /* called with rcu_read_lock held */
rt_fill_info(struct net * net,__be32 dst,__be32 src,struct rtable * rt,u32 table_id,struct flowi4 * fl4,struct sk_buff * skb,u32 portid,u32 seq,unsigned int flags)2938 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2939 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2940 struct sk_buff *skb, u32 portid, u32 seq,
2941 unsigned int flags)
2942 {
2943 struct rtmsg *r;
2944 struct nlmsghdr *nlh;
2945 unsigned long expires = 0;
2946 u32 error;
2947 u32 metrics[RTAX_MAX];
2948
2949 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2950 if (!nlh)
2951 return -EMSGSIZE;
2952
2953 r = nlmsg_data(nlh);
2954 r->rtm_family = AF_INET;
2955 r->rtm_dst_len = 32;
2956 r->rtm_src_len = 0;
2957 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2958 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2959 if (nla_put_u32(skb, RTA_TABLE, table_id))
2960 goto nla_put_failure;
2961 r->rtm_type = rt->rt_type;
2962 r->rtm_scope = RT_SCOPE_UNIVERSE;
2963 r->rtm_protocol = RTPROT_UNSPEC;
2964 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2965 if (rt->rt_flags & RTCF_NOTIFY)
2966 r->rtm_flags |= RTM_F_NOTIFY;
2967 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2968 r->rtm_flags |= RTCF_DOREDIRECT;
2969
2970 if (nla_put_in_addr(skb, RTA_DST, dst))
2971 goto nla_put_failure;
2972 if (src) {
2973 r->rtm_src_len = 32;
2974 if (nla_put_in_addr(skb, RTA_SRC, src))
2975 goto nla_put_failure;
2976 }
2977 if (rt->dst.dev &&
2978 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2979 goto nla_put_failure;
2980 if (rt->dst.lwtstate &&
2981 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2982 goto nla_put_failure;
2983 #ifdef CONFIG_IP_ROUTE_CLASSID
2984 if (rt->dst.tclassid &&
2985 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2986 goto nla_put_failure;
2987 #endif
2988 if (fl4 && !rt_is_input_route(rt) &&
2989 fl4->saddr != src) {
2990 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2991 goto nla_put_failure;
2992 }
2993 if (rt->rt_uses_gateway) {
2994 if (rt->rt_gw_family == AF_INET &&
2995 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2996 goto nla_put_failure;
2997 } else if (rt->rt_gw_family == AF_INET6) {
2998 int alen = sizeof(struct in6_addr);
2999 struct nlattr *nla;
3000 struct rtvia *via;
3001
3002 nla = nla_reserve(skb, RTA_VIA, alen + 2);
3003 if (!nla)
3004 goto nla_put_failure;
3005
3006 via = nla_data(nla);
3007 via->rtvia_family = AF_INET6;
3008 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3009 }
3010 }
3011
3012 expires = rt->dst.expires;
3013 if (expires) {
3014 unsigned long now = jiffies;
3015
3016 if (time_before(now, expires))
3017 expires -= now;
3018 else
3019 expires = 0;
3020 }
3021
3022 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3023 if (rt->rt_pmtu && expires)
3024 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3025 if (rt->rt_mtu_locked && expires)
3026 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3027 if (rtnetlink_put_metrics(skb, metrics) < 0)
3028 goto nla_put_failure;
3029
3030 if (fl4) {
3031 if (fl4->flowi4_mark &&
3032 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3033 goto nla_put_failure;
3034
3035 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3036 nla_put_u32(skb, RTA_UID,
3037 from_kuid_munged(current_user_ns(),
3038 fl4->flowi4_uid)))
3039 goto nla_put_failure;
3040
3041 if (rt_is_input_route(rt)) {
3042 #ifdef CONFIG_IP_MROUTE
3043 if (ipv4_is_multicast(dst) &&
3044 !ipv4_is_local_multicast(dst) &&
3045 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3046 int err = ipmr_get_route(net, skb,
3047 fl4->saddr, fl4->daddr,
3048 r, portid);
3049
3050 if (err <= 0) {
3051 if (err == 0)
3052 return 0;
3053 goto nla_put_failure;
3054 }
3055 } else
3056 #endif
3057 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3058 goto nla_put_failure;
3059 }
3060 }
3061
3062 error = rt->dst.error;
3063
3064 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3065 goto nla_put_failure;
3066
3067 nlmsg_end(skb, nlh);
3068 return 0;
3069
3070 nla_put_failure:
3071 nlmsg_cancel(skb, nlh);
3072 return -EMSGSIZE;
3073 }
3074
fnhe_dump_bucket(struct net * net,struct sk_buff * skb,struct netlink_callback * cb,u32 table_id,struct fnhe_hash_bucket * bucket,int genid,int * fa_index,int fa_start,unsigned int flags)3075 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3076 struct netlink_callback *cb, u32 table_id,
3077 struct fnhe_hash_bucket *bucket, int genid,
3078 int *fa_index, int fa_start, unsigned int flags)
3079 {
3080 int i;
3081
3082 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3083 struct fib_nh_exception *fnhe;
3084
3085 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3086 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3087 struct rtable *rt;
3088 int err;
3089
3090 if (*fa_index < fa_start)
3091 goto next;
3092
3093 if (fnhe->fnhe_genid != genid)
3094 goto next;
3095
3096 if (fnhe->fnhe_expires &&
3097 time_after(jiffies, fnhe->fnhe_expires))
3098 goto next;
3099
3100 rt = rcu_dereference(fnhe->fnhe_rth_input);
3101 if (!rt)
3102 rt = rcu_dereference(fnhe->fnhe_rth_output);
3103 if (!rt)
3104 goto next;
3105
3106 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3107 table_id, NULL, skb,
3108 NETLINK_CB(cb->skb).portid,
3109 cb->nlh->nlmsg_seq, flags);
3110 if (err)
3111 return err;
3112 next:
3113 (*fa_index)++;
3114 }
3115 }
3116
3117 return 0;
3118 }
3119
fib_dump_info_fnhe(struct sk_buff * skb,struct netlink_callback * cb,u32 table_id,struct fib_info * fi,int * fa_index,int fa_start,unsigned int flags)3120 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3121 u32 table_id, struct fib_info *fi,
3122 int *fa_index, int fa_start, unsigned int flags)
3123 {
3124 struct net *net = sock_net(cb->skb->sk);
3125 int nhsel, genid = fnhe_genid(net);
3126
3127 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3128 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3129 struct fnhe_hash_bucket *bucket;
3130 int err;
3131
3132 if (nhc->nhc_flags & RTNH_F_DEAD)
3133 continue;
3134
3135 rcu_read_lock();
3136 bucket = rcu_dereference(nhc->nhc_exceptions);
3137 err = 0;
3138 if (bucket)
3139 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3140 genid, fa_index, fa_start,
3141 flags);
3142 rcu_read_unlock();
3143 if (err)
3144 return err;
3145 }
3146
3147 return 0;
3148 }
3149
inet_rtm_getroute_build_skb(__be32 src,__be32 dst,u8 ip_proto,__be16 sport,__be16 dport)3150 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3151 u8 ip_proto, __be16 sport,
3152 __be16 dport)
3153 {
3154 struct sk_buff *skb;
3155 struct iphdr *iph;
3156
3157 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3158 if (!skb)
3159 return NULL;
3160
3161 /* Reserve room for dummy headers, this skb can pass
3162 * through good chunk of routing engine.
3163 */
3164 skb_reset_mac_header(skb);
3165 skb_reset_network_header(skb);
3166 skb->protocol = htons(ETH_P_IP);
3167 iph = skb_put(skb, sizeof(struct iphdr));
3168 iph->protocol = ip_proto;
3169 iph->saddr = src;
3170 iph->daddr = dst;
3171 iph->version = 0x4;
3172 iph->frag_off = 0;
3173 iph->ihl = 0x5;
3174 skb_set_transport_header(skb, skb->len);
3175
3176 switch (iph->protocol) {
3177 case IPPROTO_UDP: {
3178 struct udphdr *udph;
3179
3180 udph = skb_put_zero(skb, sizeof(struct udphdr));
3181 udph->source = sport;
3182 udph->dest = dport;
3183 udph->len = htons(sizeof(struct udphdr));
3184 udph->check = 0;
3185 break;
3186 }
3187 case IPPROTO_TCP: {
3188 struct tcphdr *tcph;
3189
3190 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3191 tcph->source = sport;
3192 tcph->dest = dport;
3193 tcph->doff = sizeof(struct tcphdr) / 4;
3194 tcph->rst = 1;
3195 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3196 src, dst, 0);
3197 break;
3198 }
3199 case IPPROTO_ICMP: {
3200 struct icmphdr *icmph;
3201
3202 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3203 icmph->type = ICMP_ECHO;
3204 icmph->code = 0;
3205 }
3206 }
3207
3208 return skb;
3209 }
3210
inet_rtm_valid_getroute_req(struct sk_buff * skb,const struct nlmsghdr * nlh,struct nlattr ** tb,struct netlink_ext_ack * extack)3211 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3212 const struct nlmsghdr *nlh,
3213 struct nlattr **tb,
3214 struct netlink_ext_ack *extack)
3215 {
3216 struct rtmsg *rtm;
3217 int i, err;
3218
3219 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3220 NL_SET_ERR_MSG(extack,
3221 "ipv4: Invalid header for route get request");
3222 return -EINVAL;
3223 }
3224
3225 if (!netlink_strict_get_check(skb))
3226 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3227 rtm_ipv4_policy, extack);
3228
3229 rtm = nlmsg_data(nlh);
3230 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3231 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3232 rtm->rtm_table || rtm->rtm_protocol ||
3233 rtm->rtm_scope || rtm->rtm_type) {
3234 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3235 return -EINVAL;
3236 }
3237
3238 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3239 RTM_F_LOOKUP_TABLE |
3240 RTM_F_FIB_MATCH)) {
3241 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3242 return -EINVAL;
3243 }
3244
3245 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3246 rtm_ipv4_policy, extack);
3247 if (err)
3248 return err;
3249
3250 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3251 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3252 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3253 return -EINVAL;
3254 }
3255
3256 for (i = 0; i <= RTA_MAX; i++) {
3257 if (!tb[i])
3258 continue;
3259
3260 switch (i) {
3261 case RTA_IIF:
3262 case RTA_OIF:
3263 case RTA_SRC:
3264 case RTA_DST:
3265 case RTA_IP_PROTO:
3266 case RTA_SPORT:
3267 case RTA_DPORT:
3268 case RTA_MARK:
3269 case RTA_UID:
3270 break;
3271 default:
3272 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3273 return -EINVAL;
3274 }
3275 }
3276
3277 return 0;
3278 }
3279
inet_rtm_getroute(struct sk_buff * in_skb,struct nlmsghdr * nlh,struct netlink_ext_ack * extack)3280 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3281 struct netlink_ext_ack *extack)
3282 {
3283 struct net *net = sock_net(in_skb->sk);
3284 struct nlattr *tb[RTA_MAX+1];
3285 u32 table_id = RT_TABLE_MAIN;
3286 __be16 sport = 0, dport = 0;
3287 struct fib_result res = {};
3288 u8 ip_proto = IPPROTO_UDP;
3289 struct rtable *rt = NULL;
3290 struct sk_buff *skb;
3291 struct rtmsg *rtm;
3292 struct flowi4 fl4 = {};
3293 __be32 dst = 0;
3294 __be32 src = 0;
3295 kuid_t uid;
3296 u32 iif;
3297 int err;
3298 int mark;
3299
3300 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3301 if (err < 0)
3302 return err;
3303
3304 rtm = nlmsg_data(nlh);
3305 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3306 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3307 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3308 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3309 if (tb[RTA_UID])
3310 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3311 else
3312 uid = (iif ? INVALID_UID : current_uid());
3313
3314 if (tb[RTA_IP_PROTO]) {
3315 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3316 &ip_proto, AF_INET, extack);
3317 if (err)
3318 return err;
3319 }
3320
3321 if (tb[RTA_SPORT])
3322 sport = nla_get_be16(tb[RTA_SPORT]);
3323
3324 if (tb[RTA_DPORT])
3325 dport = nla_get_be16(tb[RTA_DPORT]);
3326
3327 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3328 if (!skb)
3329 return -ENOBUFS;
3330
3331 fl4.daddr = dst;
3332 fl4.saddr = src;
3333 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3334 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3335 fl4.flowi4_mark = mark;
3336 fl4.flowi4_uid = uid;
3337 if (sport)
3338 fl4.fl4_sport = sport;
3339 if (dport)
3340 fl4.fl4_dport = dport;
3341 fl4.flowi4_proto = ip_proto;
3342
3343 rcu_read_lock();
3344
3345 if (iif) {
3346 struct net_device *dev;
3347
3348 dev = dev_get_by_index_rcu(net, iif);
3349 if (!dev) {
3350 err = -ENODEV;
3351 goto errout_rcu;
3352 }
3353
3354 fl4.flowi4_iif = iif; /* for rt_fill_info */
3355 skb->dev = dev;
3356 skb->mark = mark;
3357 err = ip_route_input_rcu(skb, dst, src,
3358 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3359 &res);
3360
3361 rt = skb_rtable(skb);
3362 if (err == 0 && rt->dst.error)
3363 err = -rt->dst.error;
3364 } else {
3365 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3366 skb->dev = net->loopback_dev;
3367 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3368 err = 0;
3369 if (IS_ERR(rt))
3370 err = PTR_ERR(rt);
3371 else
3372 skb_dst_set(skb, &rt->dst);
3373 }
3374
3375 if (err)
3376 goto errout_rcu;
3377
3378 if (rtm->rtm_flags & RTM_F_NOTIFY)
3379 rt->rt_flags |= RTCF_NOTIFY;
3380
3381 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3382 table_id = res.table ? res.table->tb_id : 0;
3383
3384 /* reset skb for netlink reply msg */
3385 skb_trim(skb, 0);
3386 skb_reset_network_header(skb);
3387 skb_reset_transport_header(skb);
3388 skb_reset_mac_header(skb);
3389
3390 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3391 struct fib_rt_info fri;
3392
3393 if (!res.fi) {
3394 err = fib_props[res.type].error;
3395 if (!err)
3396 err = -EHOSTUNREACH;
3397 goto errout_rcu;
3398 }
3399 fri.fi = res.fi;
3400 fri.tb_id = table_id;
3401 fri.dst = res.prefix;
3402 fri.dst_len = res.prefixlen;
3403 fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3404 fri.type = rt->rt_type;
3405 fri.offload = 0;
3406 fri.trap = 0;
3407 fri.offload_failed = 0;
3408 if (res.fa_head) {
3409 struct fib_alias *fa;
3410
3411 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3412 u8 slen = 32 - fri.dst_len;
3413
3414 if (fa->fa_slen == slen &&
3415 fa->tb_id == fri.tb_id &&
3416 fa->fa_dscp == fri.dscp &&
3417 fa->fa_info == res.fi &&
3418 fa->fa_type == fri.type) {
3419 fri.offload = READ_ONCE(fa->offload);
3420 fri.trap = READ_ONCE(fa->trap);
3421 break;
3422 }
3423 }
3424 }
3425 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3426 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3427 } else {
3428 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3429 NETLINK_CB(in_skb).portid,
3430 nlh->nlmsg_seq, 0);
3431 }
3432 if (err < 0)
3433 goto errout_rcu;
3434
3435 rcu_read_unlock();
3436
3437 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3438
3439 errout_free:
3440 return err;
3441 errout_rcu:
3442 rcu_read_unlock();
3443 kfree_skb(skb);
3444 goto errout_free;
3445 }
3446
ip_rt_multicast_event(struct in_device * in_dev)3447 void ip_rt_multicast_event(struct in_device *in_dev)
3448 {
3449 rt_cache_flush(dev_net(in_dev->dev));
3450 }
3451
3452 #ifdef CONFIG_SYSCTL
3453 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3454 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3455 static int ip_rt_gc_elasticity __read_mostly = 8;
3456 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3457
ipv4_sysctl_rtcache_flush(struct ctl_table * __ctl,int write,void * buffer,size_t * lenp,loff_t * ppos)3458 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3459 void *buffer, size_t *lenp, loff_t *ppos)
3460 {
3461 struct net *net = (struct net *)__ctl->extra1;
3462
3463 if (write) {
3464 rt_cache_flush(net);
3465 fnhe_genid_bump(net);
3466 return 0;
3467 }
3468
3469 return -EINVAL;
3470 }
3471
3472 static struct ctl_table ipv4_route_table[] = {
3473 {
3474 .procname = "gc_thresh",
3475 .data = &ipv4_dst_ops.gc_thresh,
3476 .maxlen = sizeof(int),
3477 .mode = 0644,
3478 .proc_handler = proc_dointvec,
3479 },
3480 {
3481 .procname = "max_size",
3482 .data = &ip_rt_max_size,
3483 .maxlen = sizeof(int),
3484 .mode = 0644,
3485 .proc_handler = proc_dointvec,
3486 },
3487 {
3488 /* Deprecated. Use gc_min_interval_ms */
3489
3490 .procname = "gc_min_interval",
3491 .data = &ip_rt_gc_min_interval,
3492 .maxlen = sizeof(int),
3493 .mode = 0644,
3494 .proc_handler = proc_dointvec_jiffies,
3495 },
3496 {
3497 .procname = "gc_min_interval_ms",
3498 .data = &ip_rt_gc_min_interval,
3499 .maxlen = sizeof(int),
3500 .mode = 0644,
3501 .proc_handler = proc_dointvec_ms_jiffies,
3502 },
3503 {
3504 .procname = "gc_timeout",
3505 .data = &ip_rt_gc_timeout,
3506 .maxlen = sizeof(int),
3507 .mode = 0644,
3508 .proc_handler = proc_dointvec_jiffies,
3509 },
3510 {
3511 .procname = "gc_interval",
3512 .data = &ip_rt_gc_interval,
3513 .maxlen = sizeof(int),
3514 .mode = 0644,
3515 .proc_handler = proc_dointvec_jiffies,
3516 },
3517 {
3518 .procname = "redirect_load",
3519 .data = &ip_rt_redirect_load,
3520 .maxlen = sizeof(int),
3521 .mode = 0644,
3522 .proc_handler = proc_dointvec,
3523 },
3524 {
3525 .procname = "redirect_number",
3526 .data = &ip_rt_redirect_number,
3527 .maxlen = sizeof(int),
3528 .mode = 0644,
3529 .proc_handler = proc_dointvec,
3530 },
3531 {
3532 .procname = "redirect_silence",
3533 .data = &ip_rt_redirect_silence,
3534 .maxlen = sizeof(int),
3535 .mode = 0644,
3536 .proc_handler = proc_dointvec,
3537 },
3538 {
3539 .procname = "error_cost",
3540 .data = &ip_rt_error_cost,
3541 .maxlen = sizeof(int),
3542 .mode = 0644,
3543 .proc_handler = proc_dointvec,
3544 },
3545 {
3546 .procname = "error_burst",
3547 .data = &ip_rt_error_burst,
3548 .maxlen = sizeof(int),
3549 .mode = 0644,
3550 .proc_handler = proc_dointvec,
3551 },
3552 {
3553 .procname = "gc_elasticity",
3554 .data = &ip_rt_gc_elasticity,
3555 .maxlen = sizeof(int),
3556 .mode = 0644,
3557 .proc_handler = proc_dointvec,
3558 },
3559 { }
3560 };
3561
3562 static const char ipv4_route_flush_procname[] = "flush";
3563
3564 static struct ctl_table ipv4_route_netns_table[] = {
3565 {
3566 .procname = ipv4_route_flush_procname,
3567 .maxlen = sizeof(int),
3568 .mode = 0200,
3569 .proc_handler = ipv4_sysctl_rtcache_flush,
3570 },
3571 {
3572 .procname = "min_pmtu",
3573 .data = &init_net.ipv4.ip_rt_min_pmtu,
3574 .maxlen = sizeof(int),
3575 .mode = 0644,
3576 .proc_handler = proc_dointvec_minmax,
3577 .extra1 = &ip_min_valid_pmtu,
3578 },
3579 {
3580 .procname = "mtu_expires",
3581 .data = &init_net.ipv4.ip_rt_mtu_expires,
3582 .maxlen = sizeof(int),
3583 .mode = 0644,
3584 .proc_handler = proc_dointvec_jiffies,
3585 },
3586 {
3587 .procname = "min_adv_mss",
3588 .data = &init_net.ipv4.ip_rt_min_advmss,
3589 .maxlen = sizeof(int),
3590 .mode = 0644,
3591 .proc_handler = proc_dointvec,
3592 },
3593 { },
3594 };
3595
sysctl_route_net_init(struct net * net)3596 static __net_init int sysctl_route_net_init(struct net *net)
3597 {
3598 struct ctl_table *tbl;
3599
3600 tbl = ipv4_route_netns_table;
3601 if (!net_eq(net, &init_net)) {
3602 int i;
3603
3604 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3605 if (!tbl)
3606 goto err_dup;
3607
3608 /* Don't export non-whitelisted sysctls to unprivileged users */
3609 if (net->user_ns != &init_user_ns) {
3610 if (tbl[0].procname != ipv4_route_flush_procname)
3611 tbl[0].procname = NULL;
3612 }
3613
3614 /* Update the variables to point into the current struct net
3615 * except for the first element flush
3616 */
3617 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3618 tbl[i].data += (void *)net - (void *)&init_net;
3619 }
3620 tbl[0].extra1 = net;
3621
3622 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3623 if (!net->ipv4.route_hdr)
3624 goto err_reg;
3625 return 0;
3626
3627 err_reg:
3628 if (tbl != ipv4_route_netns_table)
3629 kfree(tbl);
3630 err_dup:
3631 return -ENOMEM;
3632 }
3633
sysctl_route_net_exit(struct net * net)3634 static __net_exit void sysctl_route_net_exit(struct net *net)
3635 {
3636 struct ctl_table *tbl;
3637
3638 tbl = net->ipv4.route_hdr->ctl_table_arg;
3639 unregister_net_sysctl_table(net->ipv4.route_hdr);
3640 BUG_ON(tbl == ipv4_route_netns_table);
3641 kfree(tbl);
3642 }
3643
3644 static __net_initdata struct pernet_operations sysctl_route_ops = {
3645 .init = sysctl_route_net_init,
3646 .exit = sysctl_route_net_exit,
3647 };
3648 #endif
3649
netns_ip_rt_init(struct net * net)3650 static __net_init int netns_ip_rt_init(struct net *net)
3651 {
3652 /* Set default value for namespaceified sysctls */
3653 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3654 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3655 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3656 return 0;
3657 }
3658
3659 static struct pernet_operations __net_initdata ip_rt_ops = {
3660 .init = netns_ip_rt_init,
3661 };
3662
rt_genid_init(struct net * net)3663 static __net_init int rt_genid_init(struct net *net)
3664 {
3665 atomic_set(&net->ipv4.rt_genid, 0);
3666 atomic_set(&net->fnhe_genid, 0);
3667 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3668 return 0;
3669 }
3670
3671 static __net_initdata struct pernet_operations rt_genid_ops = {
3672 .init = rt_genid_init,
3673 };
3674
ipv4_inetpeer_init(struct net * net)3675 static int __net_init ipv4_inetpeer_init(struct net *net)
3676 {
3677 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3678
3679 if (!bp)
3680 return -ENOMEM;
3681 inet_peer_base_init(bp);
3682 net->ipv4.peers = bp;
3683 return 0;
3684 }
3685
ipv4_inetpeer_exit(struct net * net)3686 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3687 {
3688 struct inet_peer_base *bp = net->ipv4.peers;
3689
3690 net->ipv4.peers = NULL;
3691 inetpeer_invalidate_tree(bp);
3692 kfree(bp);
3693 }
3694
3695 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3696 .init = ipv4_inetpeer_init,
3697 .exit = ipv4_inetpeer_exit,
3698 };
3699
3700 #ifdef CONFIG_IP_ROUTE_CLASSID
3701 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3702 #endif /* CONFIG_IP_ROUTE_CLASSID */
3703
ip_rt_init(void)3704 int __init ip_rt_init(void)
3705 {
3706 void *idents_hash;
3707 int cpu;
3708
3709 /* For modern hosts, this will use 2 MB of memory */
3710 idents_hash = alloc_large_system_hash("IP idents",
3711 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3712 0,
3713 16, /* one bucket per 64 KB */
3714 HASH_ZERO,
3715 NULL,
3716 &ip_idents_mask,
3717 2048,
3718 256*1024);
3719
3720 ip_idents = idents_hash;
3721
3722 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3723
3724 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3725
3726 for_each_possible_cpu(cpu) {
3727 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3728
3729 INIT_LIST_HEAD(&ul->head);
3730 INIT_LIST_HEAD(&ul->quarantine);
3731 spin_lock_init(&ul->lock);
3732 }
3733 #ifdef CONFIG_IP_ROUTE_CLASSID
3734 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3735 if (!ip_rt_acct)
3736 panic("IP: failed to allocate ip_rt_acct\n");
3737 #endif
3738
3739 ipv4_dst_ops.kmem_cachep =
3740 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3741 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3742
3743 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3744
3745 if (dst_entries_init(&ipv4_dst_ops) < 0)
3746 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3747
3748 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3749 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3750
3751 ipv4_dst_ops.gc_thresh = ~0;
3752 ip_rt_max_size = INT_MAX;
3753
3754 devinet_init();
3755 ip_fib_init();
3756
3757 if (ip_rt_proc_init())
3758 pr_err("Unable to create route proc files\n");
3759 #ifdef CONFIG_XFRM
3760 xfrm_init();
3761 xfrm4_init();
3762 #endif
3763 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3764 RTNL_FLAG_DOIT_UNLOCKED);
3765
3766 #ifdef CONFIG_SYSCTL
3767 register_pernet_subsys(&sysctl_route_ops);
3768 #endif
3769 register_pernet_subsys(&ip_rt_ops);
3770 register_pernet_subsys(&rt_genid_ops);
3771 register_pernet_subsys(&ipv4_inetpeer_ops);
3772 return 0;
3773 }
3774
3775 #ifdef CONFIG_SYSCTL
3776 /*
3777 * We really need to sanitize the damn ipv4 init order, then all
3778 * this nonsense will go away.
3779 */
ip_static_sysctl_init(void)3780 void __init ip_static_sysctl_init(void)
3781 {
3782 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3783 }
3784 #endif
3785