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 dev_replace_track(dev, blackhole_netdev,
1554 &rt->dst.dev_tracker,
1555 GFP_ATOMIC);
1556 list_move(&rt->rt_uncached, &ul->quarantine);
1557 }
1558 spin_unlock_bh(&ul->lock);
1559 }
1560 }
1561
rt_cache_valid(const struct rtable * rt)1562 static bool rt_cache_valid(const struct rtable *rt)
1563 {
1564 return rt &&
1565 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1566 !rt_is_expired(rt);
1567 }
1568
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)1569 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1570 const struct fib_result *res,
1571 struct fib_nh_exception *fnhe,
1572 struct fib_info *fi, u16 type, u32 itag,
1573 const bool do_cache)
1574 {
1575 bool cached = false;
1576
1577 if (fi) {
1578 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1579
1580 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1581 rt->rt_uses_gateway = 1;
1582 rt->rt_gw_family = nhc->nhc_gw_family;
1583 /* only INET and INET6 are supported */
1584 if (likely(nhc->nhc_gw_family == AF_INET))
1585 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1586 else
1587 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1588 }
1589
1590 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1591
1592 #ifdef CONFIG_IP_ROUTE_CLASSID
1593 if (nhc->nhc_family == AF_INET) {
1594 struct fib_nh *nh;
1595
1596 nh = container_of(nhc, struct fib_nh, nh_common);
1597 rt->dst.tclassid = nh->nh_tclassid;
1598 }
1599 #endif
1600 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1601 if (unlikely(fnhe))
1602 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1603 else if (do_cache)
1604 cached = rt_cache_route(nhc, rt);
1605 if (unlikely(!cached)) {
1606 /* Routes we intend to cache in nexthop exception or
1607 * FIB nexthop have the DST_NOCACHE bit clear.
1608 * However, if we are unsuccessful at storing this
1609 * route into the cache we really need to set it.
1610 */
1611 if (!rt->rt_gw4) {
1612 rt->rt_gw_family = AF_INET;
1613 rt->rt_gw4 = daddr;
1614 }
1615 rt_add_uncached_list(rt);
1616 }
1617 } else
1618 rt_add_uncached_list(rt);
1619
1620 #ifdef CONFIG_IP_ROUTE_CLASSID
1621 #ifdef CONFIG_IP_MULTIPLE_TABLES
1622 set_class_tag(rt, res->tclassid);
1623 #endif
1624 set_class_tag(rt, itag);
1625 #endif
1626 }
1627
rt_dst_alloc(struct net_device * dev,unsigned int flags,u16 type,bool nopolicy,bool noxfrm)1628 struct rtable *rt_dst_alloc(struct net_device *dev,
1629 unsigned int flags, u16 type,
1630 bool nopolicy, bool noxfrm)
1631 {
1632 struct rtable *rt;
1633
1634 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1635 (nopolicy ? DST_NOPOLICY : 0) |
1636 (noxfrm ? DST_NOXFRM : 0));
1637
1638 if (rt) {
1639 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1640 rt->rt_flags = flags;
1641 rt->rt_type = type;
1642 rt->rt_is_input = 0;
1643 rt->rt_iif = 0;
1644 rt->rt_pmtu = 0;
1645 rt->rt_mtu_locked = 0;
1646 rt->rt_uses_gateway = 0;
1647 rt->rt_gw_family = 0;
1648 rt->rt_gw4 = 0;
1649 INIT_LIST_HEAD(&rt->rt_uncached);
1650
1651 rt->dst.output = ip_output;
1652 if (flags & RTCF_LOCAL)
1653 rt->dst.input = ip_local_deliver;
1654 }
1655
1656 return rt;
1657 }
1658 EXPORT_SYMBOL(rt_dst_alloc);
1659
rt_dst_clone(struct net_device * dev,struct rtable * rt)1660 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1661 {
1662 struct rtable *new_rt;
1663
1664 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1665 rt->dst.flags);
1666
1667 if (new_rt) {
1668 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1669 new_rt->rt_flags = rt->rt_flags;
1670 new_rt->rt_type = rt->rt_type;
1671 new_rt->rt_is_input = rt->rt_is_input;
1672 new_rt->rt_iif = rt->rt_iif;
1673 new_rt->rt_pmtu = rt->rt_pmtu;
1674 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1675 new_rt->rt_gw_family = rt->rt_gw_family;
1676 if (rt->rt_gw_family == AF_INET)
1677 new_rt->rt_gw4 = rt->rt_gw4;
1678 else if (rt->rt_gw_family == AF_INET6)
1679 new_rt->rt_gw6 = rt->rt_gw6;
1680 INIT_LIST_HEAD(&new_rt->rt_uncached);
1681
1682 new_rt->dst.input = rt->dst.input;
1683 new_rt->dst.output = rt->dst.output;
1684 new_rt->dst.error = rt->dst.error;
1685 new_rt->dst.lastuse = jiffies;
1686 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1687 }
1688 return new_rt;
1689 }
1690 EXPORT_SYMBOL(rt_dst_clone);
1691
1692 /* 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)1693 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1694 u8 tos, struct net_device *dev,
1695 struct in_device *in_dev, u32 *itag)
1696 {
1697 int err;
1698
1699 /* Primary sanity checks. */
1700 if (!in_dev)
1701 return -EINVAL;
1702
1703 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1704 skb->protocol != htons(ETH_P_IP))
1705 return -EINVAL;
1706
1707 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1708 return -EINVAL;
1709
1710 if (ipv4_is_zeronet(saddr)) {
1711 if (!ipv4_is_local_multicast(daddr) &&
1712 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1713 return -EINVAL;
1714 } else {
1715 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1716 in_dev, itag);
1717 if (err < 0)
1718 return err;
1719 }
1720 return 0;
1721 }
1722
1723 /* 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)1724 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1725 u8 tos, struct net_device *dev, int our)
1726 {
1727 struct in_device *in_dev = __in_dev_get_rcu(dev);
1728 unsigned int flags = RTCF_MULTICAST;
1729 struct rtable *rth;
1730 bool no_policy;
1731 u32 itag = 0;
1732 int err;
1733
1734 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1735 if (err)
1736 return err;
1737
1738 if (our)
1739 flags |= RTCF_LOCAL;
1740
1741 no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
1742 if (no_policy)
1743 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1744
1745 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1746 no_policy, false);
1747 if (!rth)
1748 return -ENOBUFS;
1749
1750 #ifdef CONFIG_IP_ROUTE_CLASSID
1751 rth->dst.tclassid = itag;
1752 #endif
1753 rth->dst.output = ip_rt_bug;
1754 rth->rt_is_input= 1;
1755
1756 #ifdef CONFIG_IP_MROUTE
1757 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1758 rth->dst.input = ip_mr_input;
1759 #endif
1760 RT_CACHE_STAT_INC(in_slow_mc);
1761
1762 skb_dst_drop(skb);
1763 skb_dst_set(skb, &rth->dst);
1764 return 0;
1765 }
1766
1767
ip_handle_martian_source(struct net_device * dev,struct in_device * in_dev,struct sk_buff * skb,__be32 daddr,__be32 saddr)1768 static void ip_handle_martian_source(struct net_device *dev,
1769 struct in_device *in_dev,
1770 struct sk_buff *skb,
1771 __be32 daddr,
1772 __be32 saddr)
1773 {
1774 RT_CACHE_STAT_INC(in_martian_src);
1775 #ifdef CONFIG_IP_ROUTE_VERBOSE
1776 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1777 /*
1778 * RFC1812 recommendation, if source is martian,
1779 * the only hint is MAC header.
1780 */
1781 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1782 &daddr, &saddr, dev->name);
1783 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1784 print_hex_dump(KERN_WARNING, "ll header: ",
1785 DUMP_PREFIX_OFFSET, 16, 1,
1786 skb_mac_header(skb),
1787 dev->hard_header_len, false);
1788 }
1789 }
1790 #endif
1791 }
1792
1793 /* 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)1794 static int __mkroute_input(struct sk_buff *skb,
1795 const struct fib_result *res,
1796 struct in_device *in_dev,
1797 __be32 daddr, __be32 saddr, u32 tos)
1798 {
1799 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1800 struct net_device *dev = nhc->nhc_dev;
1801 struct fib_nh_exception *fnhe;
1802 struct rtable *rth;
1803 int err;
1804 struct in_device *out_dev;
1805 bool do_cache, no_policy;
1806 u32 itag = 0;
1807
1808 /* get a working reference to the output device */
1809 out_dev = __in_dev_get_rcu(dev);
1810 if (!out_dev) {
1811 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1812 return -EINVAL;
1813 }
1814
1815 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1816 in_dev->dev, in_dev, &itag);
1817 if (err < 0) {
1818 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1819 saddr);
1820
1821 goto cleanup;
1822 }
1823
1824 do_cache = res->fi && !itag;
1825 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1826 skb->protocol == htons(ETH_P_IP)) {
1827 __be32 gw;
1828
1829 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1830 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1831 inet_addr_onlink(out_dev, saddr, gw))
1832 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1833 }
1834
1835 if (skb->protocol != htons(ETH_P_IP)) {
1836 /* Not IP (i.e. ARP). Do not create route, if it is
1837 * invalid for proxy arp. DNAT routes are always valid.
1838 *
1839 * Proxy arp feature have been extended to allow, ARP
1840 * replies back to the same interface, to support
1841 * Private VLAN switch technologies. See arp.c.
1842 */
1843 if (out_dev == in_dev &&
1844 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1845 err = -EINVAL;
1846 goto cleanup;
1847 }
1848 }
1849
1850 no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
1851 if (no_policy)
1852 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1853
1854 fnhe = find_exception(nhc, daddr);
1855 if (do_cache) {
1856 if (fnhe)
1857 rth = rcu_dereference(fnhe->fnhe_rth_input);
1858 else
1859 rth = rcu_dereference(nhc->nhc_rth_input);
1860 if (rt_cache_valid(rth)) {
1861 skb_dst_set_noref(skb, &rth->dst);
1862 goto out;
1863 }
1864 }
1865
1866 rth = rt_dst_alloc(out_dev->dev, 0, res->type, no_policy,
1867 IN_DEV_ORCONF(out_dev, NOXFRM));
1868 if (!rth) {
1869 err = -ENOBUFS;
1870 goto cleanup;
1871 }
1872
1873 rth->rt_is_input = 1;
1874 RT_CACHE_STAT_INC(in_slow_tot);
1875
1876 rth->dst.input = ip_forward;
1877
1878 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1879 do_cache);
1880 lwtunnel_set_redirect(&rth->dst);
1881 skb_dst_set(skb, &rth->dst);
1882 out:
1883 err = 0;
1884 cleanup:
1885 return err;
1886 }
1887
1888 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1889 /* To make ICMP packets follow the right flow, the multipath hash is
1890 * calculated from the inner IP addresses.
1891 */
ip_multipath_l3_keys(const struct sk_buff * skb,struct flow_keys * hash_keys)1892 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1893 struct flow_keys *hash_keys)
1894 {
1895 const struct iphdr *outer_iph = ip_hdr(skb);
1896 const struct iphdr *key_iph = outer_iph;
1897 const struct iphdr *inner_iph;
1898 const struct icmphdr *icmph;
1899 struct iphdr _inner_iph;
1900 struct icmphdr _icmph;
1901
1902 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1903 goto out;
1904
1905 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1906 goto out;
1907
1908 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1909 &_icmph);
1910 if (!icmph)
1911 goto out;
1912
1913 if (!icmp_is_err(icmph->type))
1914 goto out;
1915
1916 inner_iph = skb_header_pointer(skb,
1917 outer_iph->ihl * 4 + sizeof(_icmph),
1918 sizeof(_inner_iph), &_inner_iph);
1919 if (!inner_iph)
1920 goto out;
1921
1922 key_iph = inner_iph;
1923 out:
1924 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1925 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1926 }
1927
fib_multipath_custom_hash_outer(const struct net * net,const struct sk_buff * skb,bool * p_has_inner)1928 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1929 const struct sk_buff *skb,
1930 bool *p_has_inner)
1931 {
1932 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1933 struct flow_keys keys, hash_keys;
1934
1935 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1936 return 0;
1937
1938 memset(&hash_keys, 0, sizeof(hash_keys));
1939 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1940
1941 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1942 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1943 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1944 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1945 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1946 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1947 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1948 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1949 hash_keys.ports.src = keys.ports.src;
1950 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1951 hash_keys.ports.dst = keys.ports.dst;
1952
1953 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1954 return flow_hash_from_keys(&hash_keys);
1955 }
1956
fib_multipath_custom_hash_inner(const struct net * net,const struct sk_buff * skb,bool has_inner)1957 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1958 const struct sk_buff *skb,
1959 bool has_inner)
1960 {
1961 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1962 struct flow_keys keys, hash_keys;
1963
1964 /* We assume the packet carries an encapsulation, but if none was
1965 * encountered during dissection of the outer flow, then there is no
1966 * point in calling the flow dissector again.
1967 */
1968 if (!has_inner)
1969 return 0;
1970
1971 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1972 return 0;
1973
1974 memset(&hash_keys, 0, sizeof(hash_keys));
1975 skb_flow_dissect_flow_keys(skb, &keys, 0);
1976
1977 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1978 return 0;
1979
1980 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1981 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1982 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1983 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1984 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1985 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1986 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1987 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1988 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1989 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1990 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1991 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1992 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1993 hash_keys.tags.flow_label = keys.tags.flow_label;
1994 }
1995
1996 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1997 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1998 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1999 hash_keys.ports.src = keys.ports.src;
2000 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2001 hash_keys.ports.dst = keys.ports.dst;
2002
2003 return flow_hash_from_keys(&hash_keys);
2004 }
2005
fib_multipath_custom_hash_skb(const struct net * net,const struct sk_buff * skb)2006 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2007 const struct sk_buff *skb)
2008 {
2009 u32 mhash, mhash_inner;
2010 bool has_inner = true;
2011
2012 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2013 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2014
2015 return jhash_2words(mhash, mhash_inner, 0);
2016 }
2017
fib_multipath_custom_hash_fl4(const struct net * net,const struct flowi4 * fl4)2018 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2019 const struct flowi4 *fl4)
2020 {
2021 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2022 struct flow_keys hash_keys;
2023
2024 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2025 return 0;
2026
2027 memset(&hash_keys, 0, sizeof(hash_keys));
2028 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2029 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2030 hash_keys.addrs.v4addrs.src = fl4->saddr;
2031 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2032 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2033 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2034 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2035 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2036 hash_keys.ports.src = fl4->fl4_sport;
2037 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2038 hash_keys.ports.dst = fl4->fl4_dport;
2039
2040 return flow_hash_from_keys(&hash_keys);
2041 }
2042
2043 /* 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)2044 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2045 const struct sk_buff *skb, struct flow_keys *flkeys)
2046 {
2047 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2048 struct flow_keys hash_keys;
2049 u32 mhash = 0;
2050
2051 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2052 case 0:
2053 memset(&hash_keys, 0, sizeof(hash_keys));
2054 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2055 if (skb) {
2056 ip_multipath_l3_keys(skb, &hash_keys);
2057 } else {
2058 hash_keys.addrs.v4addrs.src = fl4->saddr;
2059 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2060 }
2061 mhash = flow_hash_from_keys(&hash_keys);
2062 break;
2063 case 1:
2064 /* skb is currently provided only when forwarding */
2065 if (skb) {
2066 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2067 struct flow_keys keys;
2068
2069 /* short-circuit if we already have L4 hash present */
2070 if (skb->l4_hash)
2071 return skb_get_hash_raw(skb) >> 1;
2072
2073 memset(&hash_keys, 0, sizeof(hash_keys));
2074
2075 if (!flkeys) {
2076 skb_flow_dissect_flow_keys(skb, &keys, flag);
2077 flkeys = &keys;
2078 }
2079
2080 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2081 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2082 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2083 hash_keys.ports.src = flkeys->ports.src;
2084 hash_keys.ports.dst = flkeys->ports.dst;
2085 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2086 } else {
2087 memset(&hash_keys, 0, sizeof(hash_keys));
2088 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2089 hash_keys.addrs.v4addrs.src = fl4->saddr;
2090 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2091 hash_keys.ports.src = fl4->fl4_sport;
2092 hash_keys.ports.dst = fl4->fl4_dport;
2093 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2094 }
2095 mhash = flow_hash_from_keys(&hash_keys);
2096 break;
2097 case 2:
2098 memset(&hash_keys, 0, sizeof(hash_keys));
2099 /* skb is currently provided only when forwarding */
2100 if (skb) {
2101 struct flow_keys keys;
2102
2103 skb_flow_dissect_flow_keys(skb, &keys, 0);
2104 /* Inner can be v4 or v6 */
2105 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2106 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2107 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2108 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2109 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2110 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2111 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2112 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2113 hash_keys.tags.flow_label = keys.tags.flow_label;
2114 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2115 } else {
2116 /* Same as case 0 */
2117 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2118 ip_multipath_l3_keys(skb, &hash_keys);
2119 }
2120 } else {
2121 /* Same as case 0 */
2122 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2123 hash_keys.addrs.v4addrs.src = fl4->saddr;
2124 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2125 }
2126 mhash = flow_hash_from_keys(&hash_keys);
2127 break;
2128 case 3:
2129 if (skb)
2130 mhash = fib_multipath_custom_hash_skb(net, skb);
2131 else
2132 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2133 break;
2134 }
2135
2136 if (multipath_hash)
2137 mhash = jhash_2words(mhash, multipath_hash, 0);
2138
2139 return mhash >> 1;
2140 }
2141 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2142
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)2143 static int ip_mkroute_input(struct sk_buff *skb,
2144 struct fib_result *res,
2145 struct in_device *in_dev,
2146 __be32 daddr, __be32 saddr, u32 tos,
2147 struct flow_keys *hkeys)
2148 {
2149 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2150 if (res->fi && fib_info_num_path(res->fi) > 1) {
2151 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2152
2153 fib_select_multipath(res, h);
2154 }
2155 #endif
2156
2157 /* create a routing cache entry */
2158 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2159 }
2160
2161 /* Implements all the saddr-related checks as ip_route_input_slow(),
2162 * assuming daddr is valid and the destination is not a local broadcast one.
2163 * Uses the provided hint instead of performing a route lookup.
2164 */
ip_route_use_hint(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev,const struct sk_buff * hint)2165 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2166 u8 tos, struct net_device *dev,
2167 const struct sk_buff *hint)
2168 {
2169 struct in_device *in_dev = __in_dev_get_rcu(dev);
2170 struct rtable *rt = skb_rtable(hint);
2171 struct net *net = dev_net(dev);
2172 int err = -EINVAL;
2173 u32 tag = 0;
2174
2175 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2176 goto martian_source;
2177
2178 if (ipv4_is_zeronet(saddr))
2179 goto martian_source;
2180
2181 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2182 goto martian_source;
2183
2184 if (rt->rt_type != RTN_LOCAL)
2185 goto skip_validate_source;
2186
2187 tos &= IPTOS_RT_MASK;
2188 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2189 if (err < 0)
2190 goto martian_source;
2191
2192 skip_validate_source:
2193 skb_dst_copy(skb, hint);
2194 return 0;
2195
2196 martian_source:
2197 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2198 return err;
2199 }
2200
2201 /* get device for dst_alloc with local routes */
ip_rt_get_dev(struct net * net,const struct fib_result * res)2202 static struct net_device *ip_rt_get_dev(struct net *net,
2203 const struct fib_result *res)
2204 {
2205 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2206 struct net_device *dev = NULL;
2207
2208 if (nhc)
2209 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2210
2211 return dev ? : net->loopback_dev;
2212 }
2213
2214 /*
2215 * NOTE. We drop all the packets that has local source
2216 * addresses, because every properly looped back packet
2217 * must have correct destination already attached by output routine.
2218 * Changes in the enforced policies must be applied also to
2219 * ip_route_use_hint().
2220 *
2221 * Such approach solves two big problems:
2222 * 1. Not simplex devices are handled properly.
2223 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2224 * called with rcu_read_lock()
2225 */
2226
ip_route_input_slow(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev,struct fib_result * res)2227 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2228 u8 tos, struct net_device *dev,
2229 struct fib_result *res)
2230 {
2231 struct in_device *in_dev = __in_dev_get_rcu(dev);
2232 struct flow_keys *flkeys = NULL, _flkeys;
2233 struct net *net = dev_net(dev);
2234 struct ip_tunnel_info *tun_info;
2235 int err = -EINVAL;
2236 unsigned int flags = 0;
2237 u32 itag = 0;
2238 struct rtable *rth;
2239 struct flowi4 fl4;
2240 bool do_cache = true;
2241 bool no_policy;
2242
2243 /* IP on this device is disabled. */
2244
2245 if (!in_dev)
2246 goto out;
2247
2248 /* Check for the most weird martians, which can be not detected
2249 * by fib_lookup.
2250 */
2251
2252 tun_info = skb_tunnel_info(skb);
2253 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2254 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2255 else
2256 fl4.flowi4_tun_key.tun_id = 0;
2257 skb_dst_drop(skb);
2258
2259 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2260 goto martian_source;
2261
2262 res->fi = NULL;
2263 res->table = NULL;
2264 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2265 goto brd_input;
2266
2267 /* Accept zero addresses only to limited broadcast;
2268 * I even do not know to fix it or not. Waiting for complains :-)
2269 */
2270 if (ipv4_is_zeronet(saddr))
2271 goto martian_source;
2272
2273 if (ipv4_is_zeronet(daddr))
2274 goto martian_destination;
2275
2276 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2277 * and call it once if daddr or/and saddr are loopback addresses
2278 */
2279 if (ipv4_is_loopback(daddr)) {
2280 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2281 goto martian_destination;
2282 } else if (ipv4_is_loopback(saddr)) {
2283 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2284 goto martian_source;
2285 }
2286
2287 /*
2288 * Now we are ready to route packet.
2289 */
2290 fl4.flowi4_l3mdev = 0;
2291 fl4.flowi4_oif = 0;
2292 fl4.flowi4_iif = dev->ifindex;
2293 fl4.flowi4_mark = skb->mark;
2294 fl4.flowi4_tos = tos;
2295 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2296 fl4.flowi4_flags = 0;
2297 fl4.daddr = daddr;
2298 fl4.saddr = saddr;
2299 fl4.flowi4_uid = sock_net_uid(net, NULL);
2300 fl4.flowi4_multipath_hash = 0;
2301
2302 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2303 flkeys = &_flkeys;
2304 } else {
2305 fl4.flowi4_proto = 0;
2306 fl4.fl4_sport = 0;
2307 fl4.fl4_dport = 0;
2308 }
2309
2310 err = fib_lookup(net, &fl4, res, 0);
2311 if (err != 0) {
2312 if (!IN_DEV_FORWARD(in_dev))
2313 err = -EHOSTUNREACH;
2314 goto no_route;
2315 }
2316
2317 if (res->type == RTN_BROADCAST) {
2318 if (IN_DEV_BFORWARD(in_dev))
2319 goto make_route;
2320 /* not do cache if bc_forwarding is enabled */
2321 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2322 do_cache = false;
2323 goto brd_input;
2324 }
2325
2326 if (res->type == RTN_LOCAL) {
2327 err = fib_validate_source(skb, saddr, daddr, tos,
2328 0, dev, in_dev, &itag);
2329 if (err < 0)
2330 goto martian_source;
2331 goto local_input;
2332 }
2333
2334 if (!IN_DEV_FORWARD(in_dev)) {
2335 err = -EHOSTUNREACH;
2336 goto no_route;
2337 }
2338 if (res->type != RTN_UNICAST)
2339 goto martian_destination;
2340
2341 make_route:
2342 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2343 out: return err;
2344
2345 brd_input:
2346 if (skb->protocol != htons(ETH_P_IP))
2347 goto e_inval;
2348
2349 if (!ipv4_is_zeronet(saddr)) {
2350 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2351 in_dev, &itag);
2352 if (err < 0)
2353 goto martian_source;
2354 }
2355 flags |= RTCF_BROADCAST;
2356 res->type = RTN_BROADCAST;
2357 RT_CACHE_STAT_INC(in_brd);
2358
2359 local_input:
2360 no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
2361 if (no_policy)
2362 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2363
2364 do_cache &= res->fi && !itag;
2365 if (do_cache) {
2366 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2367
2368 rth = rcu_dereference(nhc->nhc_rth_input);
2369 if (rt_cache_valid(rth)) {
2370 skb_dst_set_noref(skb, &rth->dst);
2371 err = 0;
2372 goto out;
2373 }
2374 }
2375
2376 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2377 flags | RTCF_LOCAL, res->type,
2378 no_policy, false);
2379 if (!rth)
2380 goto e_nobufs;
2381
2382 rth->dst.output= ip_rt_bug;
2383 #ifdef CONFIG_IP_ROUTE_CLASSID
2384 rth->dst.tclassid = itag;
2385 #endif
2386 rth->rt_is_input = 1;
2387
2388 RT_CACHE_STAT_INC(in_slow_tot);
2389 if (res->type == RTN_UNREACHABLE) {
2390 rth->dst.input= ip_error;
2391 rth->dst.error= -err;
2392 rth->rt_flags &= ~RTCF_LOCAL;
2393 }
2394
2395 if (do_cache) {
2396 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2397
2398 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2399 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2400 WARN_ON(rth->dst.input == lwtunnel_input);
2401 rth->dst.lwtstate->orig_input = rth->dst.input;
2402 rth->dst.input = lwtunnel_input;
2403 }
2404
2405 if (unlikely(!rt_cache_route(nhc, rth)))
2406 rt_add_uncached_list(rth);
2407 }
2408 skb_dst_set(skb, &rth->dst);
2409 err = 0;
2410 goto out;
2411
2412 no_route:
2413 RT_CACHE_STAT_INC(in_no_route);
2414 res->type = RTN_UNREACHABLE;
2415 res->fi = NULL;
2416 res->table = NULL;
2417 goto local_input;
2418
2419 /*
2420 * Do not cache martian addresses: they should be logged (RFC1812)
2421 */
2422 martian_destination:
2423 RT_CACHE_STAT_INC(in_martian_dst);
2424 #ifdef CONFIG_IP_ROUTE_VERBOSE
2425 if (IN_DEV_LOG_MARTIANS(in_dev))
2426 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2427 &daddr, &saddr, dev->name);
2428 #endif
2429
2430 e_inval:
2431 err = -EINVAL;
2432 goto out;
2433
2434 e_nobufs:
2435 err = -ENOBUFS;
2436 goto out;
2437
2438 martian_source:
2439 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2440 goto out;
2441 }
2442
ip_route_input_noref(struct sk_buff * skb,__be32 daddr,__be32 saddr,u8 tos,struct net_device * dev)2443 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2444 u8 tos, struct net_device *dev)
2445 {
2446 struct fib_result res;
2447 int err;
2448
2449 tos &= IPTOS_RT_MASK;
2450 rcu_read_lock();
2451 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2452 rcu_read_unlock();
2453
2454 return err;
2455 }
2456 EXPORT_SYMBOL(ip_route_input_noref);
2457
2458 /* 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)2459 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2460 u8 tos, struct net_device *dev, struct fib_result *res)
2461 {
2462 /* Multicast recognition logic is moved from route cache to here.
2463 * The problem was that too many Ethernet cards have broken/missing
2464 * hardware multicast filters :-( As result the host on multicasting
2465 * network acquires a lot of useless route cache entries, sort of
2466 * SDR messages from all the world. Now we try to get rid of them.
2467 * Really, provided software IP multicast filter is organized
2468 * reasonably (at least, hashed), it does not result in a slowdown
2469 * comparing with route cache reject entries.
2470 * Note, that multicast routers are not affected, because
2471 * route cache entry is created eventually.
2472 */
2473 if (ipv4_is_multicast(daddr)) {
2474 struct in_device *in_dev = __in_dev_get_rcu(dev);
2475 int our = 0;
2476 int err = -EINVAL;
2477
2478 if (!in_dev)
2479 return err;
2480 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2481 ip_hdr(skb)->protocol);
2482
2483 /* check l3 master if no match yet */
2484 if (!our && netif_is_l3_slave(dev)) {
2485 struct in_device *l3_in_dev;
2486
2487 l3_in_dev = __in_dev_get_rcu(skb->dev);
2488 if (l3_in_dev)
2489 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2490 ip_hdr(skb)->protocol);
2491 }
2492
2493 if (our
2494 #ifdef CONFIG_IP_MROUTE
2495 ||
2496 (!ipv4_is_local_multicast(daddr) &&
2497 IN_DEV_MFORWARD(in_dev))
2498 #endif
2499 ) {
2500 err = ip_route_input_mc(skb, daddr, saddr,
2501 tos, dev, our);
2502 }
2503 return err;
2504 }
2505
2506 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2507 }
2508
2509 /* 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)2510 static struct rtable *__mkroute_output(const struct fib_result *res,
2511 const struct flowi4 *fl4, int orig_oif,
2512 struct net_device *dev_out,
2513 unsigned int flags)
2514 {
2515 struct fib_info *fi = res->fi;
2516 struct fib_nh_exception *fnhe;
2517 struct in_device *in_dev;
2518 u16 type = res->type;
2519 struct rtable *rth;
2520 bool do_cache;
2521
2522 in_dev = __in_dev_get_rcu(dev_out);
2523 if (!in_dev)
2524 return ERR_PTR(-EINVAL);
2525
2526 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2527 if (ipv4_is_loopback(fl4->saddr) &&
2528 !(dev_out->flags & IFF_LOOPBACK) &&
2529 !netif_is_l3_master(dev_out))
2530 return ERR_PTR(-EINVAL);
2531
2532 if (ipv4_is_lbcast(fl4->daddr))
2533 type = RTN_BROADCAST;
2534 else if (ipv4_is_multicast(fl4->daddr))
2535 type = RTN_MULTICAST;
2536 else if (ipv4_is_zeronet(fl4->daddr))
2537 return ERR_PTR(-EINVAL);
2538
2539 if (dev_out->flags & IFF_LOOPBACK)
2540 flags |= RTCF_LOCAL;
2541
2542 do_cache = true;
2543 if (type == RTN_BROADCAST) {
2544 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2545 fi = NULL;
2546 } else if (type == RTN_MULTICAST) {
2547 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2548 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2549 fl4->flowi4_proto))
2550 flags &= ~RTCF_LOCAL;
2551 else
2552 do_cache = false;
2553 /* If multicast route do not exist use
2554 * default one, but do not gateway in this case.
2555 * Yes, it is hack.
2556 */
2557 if (fi && res->prefixlen < 4)
2558 fi = NULL;
2559 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2560 (orig_oif != dev_out->ifindex)) {
2561 /* For local routes that require a particular output interface
2562 * we do not want to cache the result. Caching the result
2563 * causes incorrect behaviour when there are multiple source
2564 * addresses on the interface, the end result being that if the
2565 * intended recipient is waiting on that interface for the
2566 * packet he won't receive it because it will be delivered on
2567 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2568 * be set to the loopback interface as well.
2569 */
2570 do_cache = false;
2571 }
2572
2573 fnhe = NULL;
2574 do_cache &= fi != NULL;
2575 if (fi) {
2576 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2577 struct rtable __rcu **prth;
2578
2579 fnhe = find_exception(nhc, fl4->daddr);
2580 if (!do_cache)
2581 goto add;
2582 if (fnhe) {
2583 prth = &fnhe->fnhe_rth_output;
2584 } else {
2585 if (unlikely(fl4->flowi4_flags &
2586 FLOWI_FLAG_KNOWN_NH &&
2587 !(nhc->nhc_gw_family &&
2588 nhc->nhc_scope == RT_SCOPE_LINK))) {
2589 do_cache = false;
2590 goto add;
2591 }
2592 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2593 }
2594 rth = rcu_dereference(*prth);
2595 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2596 return rth;
2597 }
2598
2599 add:
2600 rth = rt_dst_alloc(dev_out, flags, type,
2601 IN_DEV_ORCONF(in_dev, NOPOLICY),
2602 IN_DEV_ORCONF(in_dev, NOXFRM));
2603 if (!rth)
2604 return ERR_PTR(-ENOBUFS);
2605
2606 rth->rt_iif = orig_oif;
2607
2608 RT_CACHE_STAT_INC(out_slow_tot);
2609
2610 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2611 if (flags & RTCF_LOCAL &&
2612 !(dev_out->flags & IFF_LOOPBACK)) {
2613 rth->dst.output = ip_mc_output;
2614 RT_CACHE_STAT_INC(out_slow_mc);
2615 }
2616 #ifdef CONFIG_IP_MROUTE
2617 if (type == RTN_MULTICAST) {
2618 if (IN_DEV_MFORWARD(in_dev) &&
2619 !ipv4_is_local_multicast(fl4->daddr)) {
2620 rth->dst.input = ip_mr_input;
2621 rth->dst.output = ip_mc_output;
2622 }
2623 }
2624 #endif
2625 }
2626
2627 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2628 lwtunnel_set_redirect(&rth->dst);
2629
2630 return rth;
2631 }
2632
2633 /*
2634 * Major route resolver routine.
2635 */
2636
ip_route_output_key_hash(struct net * net,struct flowi4 * fl4,const struct sk_buff * skb)2637 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2638 const struct sk_buff *skb)
2639 {
2640 struct fib_result res = {
2641 .type = RTN_UNSPEC,
2642 .fi = NULL,
2643 .table = NULL,
2644 .tclassid = 0,
2645 };
2646 struct rtable *rth;
2647
2648 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2649 ip_rt_fix_tos(fl4);
2650
2651 rcu_read_lock();
2652 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2653 rcu_read_unlock();
2654
2655 return rth;
2656 }
2657 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2658
ip_route_output_key_hash_rcu(struct net * net,struct flowi4 * fl4,struct fib_result * res,const struct sk_buff * skb)2659 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2660 struct fib_result *res,
2661 const struct sk_buff *skb)
2662 {
2663 struct net_device *dev_out = NULL;
2664 int orig_oif = fl4->flowi4_oif;
2665 unsigned int flags = 0;
2666 struct rtable *rth;
2667 int err;
2668
2669 if (fl4->saddr) {
2670 if (ipv4_is_multicast(fl4->saddr) ||
2671 ipv4_is_lbcast(fl4->saddr) ||
2672 ipv4_is_zeronet(fl4->saddr)) {
2673 rth = ERR_PTR(-EINVAL);
2674 goto out;
2675 }
2676
2677 rth = ERR_PTR(-ENETUNREACH);
2678
2679 /* I removed check for oif == dev_out->oif here.
2680 * It was wrong for two reasons:
2681 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2682 * is assigned to multiple interfaces.
2683 * 2. Moreover, we are allowed to send packets with saddr
2684 * of another iface. --ANK
2685 */
2686
2687 if (fl4->flowi4_oif == 0 &&
2688 (ipv4_is_multicast(fl4->daddr) ||
2689 ipv4_is_lbcast(fl4->daddr))) {
2690 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2691 dev_out = __ip_dev_find(net, fl4->saddr, false);
2692 if (!dev_out)
2693 goto out;
2694
2695 /* Special hack: user can direct multicasts
2696 * and limited broadcast via necessary interface
2697 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2698 * This hack is not just for fun, it allows
2699 * vic,vat and friends to work.
2700 * They bind socket to loopback, set ttl to zero
2701 * and expect that it will work.
2702 * From the viewpoint of routing cache they are broken,
2703 * because we are not allowed to build multicast path
2704 * with loopback source addr (look, routing cache
2705 * cannot know, that ttl is zero, so that packet
2706 * will not leave this host and route is valid).
2707 * Luckily, this hack is good workaround.
2708 */
2709
2710 fl4->flowi4_oif = dev_out->ifindex;
2711 goto make_route;
2712 }
2713
2714 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2715 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2716 if (!__ip_dev_find(net, fl4->saddr, false))
2717 goto out;
2718 }
2719 }
2720
2721
2722 if (fl4->flowi4_oif) {
2723 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2724 rth = ERR_PTR(-ENODEV);
2725 if (!dev_out)
2726 goto out;
2727
2728 /* RACE: Check return value of inet_select_addr instead. */
2729 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2730 rth = ERR_PTR(-ENETUNREACH);
2731 goto out;
2732 }
2733 if (ipv4_is_local_multicast(fl4->daddr) ||
2734 ipv4_is_lbcast(fl4->daddr) ||
2735 fl4->flowi4_proto == IPPROTO_IGMP) {
2736 if (!fl4->saddr)
2737 fl4->saddr = inet_select_addr(dev_out, 0,
2738 RT_SCOPE_LINK);
2739 goto make_route;
2740 }
2741 if (!fl4->saddr) {
2742 if (ipv4_is_multicast(fl4->daddr))
2743 fl4->saddr = inet_select_addr(dev_out, 0,
2744 fl4->flowi4_scope);
2745 else if (!fl4->daddr)
2746 fl4->saddr = inet_select_addr(dev_out, 0,
2747 RT_SCOPE_HOST);
2748 }
2749 }
2750
2751 if (!fl4->daddr) {
2752 fl4->daddr = fl4->saddr;
2753 if (!fl4->daddr)
2754 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2755 dev_out = net->loopback_dev;
2756 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2757 res->type = RTN_LOCAL;
2758 flags |= RTCF_LOCAL;
2759 goto make_route;
2760 }
2761
2762 err = fib_lookup(net, fl4, res, 0);
2763 if (err) {
2764 res->fi = NULL;
2765 res->table = NULL;
2766 if (fl4->flowi4_oif &&
2767 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2768 /* Apparently, routing tables are wrong. Assume,
2769 * that the destination is on link.
2770 *
2771 * WHY? DW.
2772 * Because we are allowed to send to iface
2773 * even if it has NO routes and NO assigned
2774 * addresses. When oif is specified, routing
2775 * tables are looked up with only one purpose:
2776 * to catch if destination is gatewayed, rather than
2777 * direct. Moreover, if MSG_DONTROUTE is set,
2778 * we send packet, ignoring both routing tables
2779 * and ifaddr state. --ANK
2780 *
2781 *
2782 * We could make it even if oif is unknown,
2783 * likely IPv6, but we do not.
2784 */
2785
2786 if (fl4->saddr == 0)
2787 fl4->saddr = inet_select_addr(dev_out, 0,
2788 RT_SCOPE_LINK);
2789 res->type = RTN_UNICAST;
2790 goto make_route;
2791 }
2792 rth = ERR_PTR(err);
2793 goto out;
2794 }
2795
2796 if (res->type == RTN_LOCAL) {
2797 if (!fl4->saddr) {
2798 if (res->fi->fib_prefsrc)
2799 fl4->saddr = res->fi->fib_prefsrc;
2800 else
2801 fl4->saddr = fl4->daddr;
2802 }
2803
2804 /* L3 master device is the loopback for that domain */
2805 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2806 net->loopback_dev;
2807
2808 /* make sure orig_oif points to fib result device even
2809 * though packet rx/tx happens over loopback or l3mdev
2810 */
2811 orig_oif = FIB_RES_OIF(*res);
2812
2813 fl4->flowi4_oif = dev_out->ifindex;
2814 flags |= RTCF_LOCAL;
2815 goto make_route;
2816 }
2817
2818 fib_select_path(net, res, fl4, skb);
2819
2820 dev_out = FIB_RES_DEV(*res);
2821
2822 make_route:
2823 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2824
2825 out:
2826 return rth;
2827 }
2828
2829 static struct dst_ops ipv4_dst_blackhole_ops = {
2830 .family = AF_INET,
2831 .default_advmss = ipv4_default_advmss,
2832 .neigh_lookup = ipv4_neigh_lookup,
2833 .check = dst_blackhole_check,
2834 .cow_metrics = dst_blackhole_cow_metrics,
2835 .update_pmtu = dst_blackhole_update_pmtu,
2836 .redirect = dst_blackhole_redirect,
2837 .mtu = dst_blackhole_mtu,
2838 };
2839
ipv4_blackhole_route(struct net * net,struct dst_entry * dst_orig)2840 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2841 {
2842 struct rtable *ort = (struct rtable *) dst_orig;
2843 struct rtable *rt;
2844
2845 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2846 if (rt) {
2847 struct dst_entry *new = &rt->dst;
2848
2849 new->__use = 1;
2850 new->input = dst_discard;
2851 new->output = dst_discard_out;
2852
2853 new->dev = net->loopback_dev;
2854 dev_hold_track(new->dev, &new->dev_tracker, GFP_ATOMIC);
2855
2856 rt->rt_is_input = ort->rt_is_input;
2857 rt->rt_iif = ort->rt_iif;
2858 rt->rt_pmtu = ort->rt_pmtu;
2859 rt->rt_mtu_locked = ort->rt_mtu_locked;
2860
2861 rt->rt_genid = rt_genid_ipv4(net);
2862 rt->rt_flags = ort->rt_flags;
2863 rt->rt_type = ort->rt_type;
2864 rt->rt_uses_gateway = ort->rt_uses_gateway;
2865 rt->rt_gw_family = ort->rt_gw_family;
2866 if (rt->rt_gw_family == AF_INET)
2867 rt->rt_gw4 = ort->rt_gw4;
2868 else if (rt->rt_gw_family == AF_INET6)
2869 rt->rt_gw6 = ort->rt_gw6;
2870
2871 INIT_LIST_HEAD(&rt->rt_uncached);
2872 }
2873
2874 dst_release(dst_orig);
2875
2876 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2877 }
2878
ip_route_output_flow(struct net * net,struct flowi4 * flp4,const struct sock * sk)2879 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2880 const struct sock *sk)
2881 {
2882 struct rtable *rt = __ip_route_output_key(net, flp4);
2883
2884 if (IS_ERR(rt))
2885 return rt;
2886
2887 if (flp4->flowi4_proto) {
2888 flp4->flowi4_oif = rt->dst.dev->ifindex;
2889 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2890 flowi4_to_flowi(flp4),
2891 sk, 0);
2892 }
2893
2894 return rt;
2895 }
2896 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2897
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)2898 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2899 struct net_device *dev,
2900 struct net *net, __be32 *saddr,
2901 const struct ip_tunnel_info *info,
2902 u8 protocol, bool use_cache)
2903 {
2904 #ifdef CONFIG_DST_CACHE
2905 struct dst_cache *dst_cache;
2906 #endif
2907 struct rtable *rt = NULL;
2908 struct flowi4 fl4;
2909 __u8 tos;
2910
2911 #ifdef CONFIG_DST_CACHE
2912 dst_cache = (struct dst_cache *)&info->dst_cache;
2913 if (use_cache) {
2914 rt = dst_cache_get_ip4(dst_cache, saddr);
2915 if (rt)
2916 return rt;
2917 }
2918 #endif
2919 memset(&fl4, 0, sizeof(fl4));
2920 fl4.flowi4_mark = skb->mark;
2921 fl4.flowi4_proto = protocol;
2922 fl4.daddr = info->key.u.ipv4.dst;
2923 fl4.saddr = info->key.u.ipv4.src;
2924 tos = info->key.tos;
2925 fl4.flowi4_tos = RT_TOS(tos);
2926
2927 rt = ip_route_output_key(net, &fl4);
2928 if (IS_ERR(rt)) {
2929 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2930 return ERR_PTR(-ENETUNREACH);
2931 }
2932 if (rt->dst.dev == dev) { /* is this necessary? */
2933 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2934 ip_rt_put(rt);
2935 return ERR_PTR(-ELOOP);
2936 }
2937 #ifdef CONFIG_DST_CACHE
2938 if (use_cache)
2939 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2940 #endif
2941 *saddr = fl4.saddr;
2942 return rt;
2943 }
2944 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2945
2946 /* 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)2947 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2948 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2949 struct sk_buff *skb, u32 portid, u32 seq,
2950 unsigned int flags)
2951 {
2952 struct rtmsg *r;
2953 struct nlmsghdr *nlh;
2954 unsigned long expires = 0;
2955 u32 error;
2956 u32 metrics[RTAX_MAX];
2957
2958 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2959 if (!nlh)
2960 return -EMSGSIZE;
2961
2962 r = nlmsg_data(nlh);
2963 r->rtm_family = AF_INET;
2964 r->rtm_dst_len = 32;
2965 r->rtm_src_len = 0;
2966 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2967 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2968 if (nla_put_u32(skb, RTA_TABLE, table_id))
2969 goto nla_put_failure;
2970 r->rtm_type = rt->rt_type;
2971 r->rtm_scope = RT_SCOPE_UNIVERSE;
2972 r->rtm_protocol = RTPROT_UNSPEC;
2973 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2974 if (rt->rt_flags & RTCF_NOTIFY)
2975 r->rtm_flags |= RTM_F_NOTIFY;
2976 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2977 r->rtm_flags |= RTCF_DOREDIRECT;
2978
2979 if (nla_put_in_addr(skb, RTA_DST, dst))
2980 goto nla_put_failure;
2981 if (src) {
2982 r->rtm_src_len = 32;
2983 if (nla_put_in_addr(skb, RTA_SRC, src))
2984 goto nla_put_failure;
2985 }
2986 if (rt->dst.dev &&
2987 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2988 goto nla_put_failure;
2989 if (rt->dst.lwtstate &&
2990 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2991 goto nla_put_failure;
2992 #ifdef CONFIG_IP_ROUTE_CLASSID
2993 if (rt->dst.tclassid &&
2994 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2995 goto nla_put_failure;
2996 #endif
2997 if (fl4 && !rt_is_input_route(rt) &&
2998 fl4->saddr != src) {
2999 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
3000 goto nla_put_failure;
3001 }
3002 if (rt->rt_uses_gateway) {
3003 if (rt->rt_gw_family == AF_INET &&
3004 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3005 goto nla_put_failure;
3006 } else if (rt->rt_gw_family == AF_INET6) {
3007 int alen = sizeof(struct in6_addr);
3008 struct nlattr *nla;
3009 struct rtvia *via;
3010
3011 nla = nla_reserve(skb, RTA_VIA, alen + 2);
3012 if (!nla)
3013 goto nla_put_failure;
3014
3015 via = nla_data(nla);
3016 via->rtvia_family = AF_INET6;
3017 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3018 }
3019 }
3020
3021 expires = rt->dst.expires;
3022 if (expires) {
3023 unsigned long now = jiffies;
3024
3025 if (time_before(now, expires))
3026 expires -= now;
3027 else
3028 expires = 0;
3029 }
3030
3031 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3032 if (rt->rt_pmtu && expires)
3033 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3034 if (rt->rt_mtu_locked && expires)
3035 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3036 if (rtnetlink_put_metrics(skb, metrics) < 0)
3037 goto nla_put_failure;
3038
3039 if (fl4) {
3040 if (fl4->flowi4_mark &&
3041 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3042 goto nla_put_failure;
3043
3044 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3045 nla_put_u32(skb, RTA_UID,
3046 from_kuid_munged(current_user_ns(),
3047 fl4->flowi4_uid)))
3048 goto nla_put_failure;
3049
3050 if (rt_is_input_route(rt)) {
3051 #ifdef CONFIG_IP_MROUTE
3052 if (ipv4_is_multicast(dst) &&
3053 !ipv4_is_local_multicast(dst) &&
3054 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3055 int err = ipmr_get_route(net, skb,
3056 fl4->saddr, fl4->daddr,
3057 r, portid);
3058
3059 if (err <= 0) {
3060 if (err == 0)
3061 return 0;
3062 goto nla_put_failure;
3063 }
3064 } else
3065 #endif
3066 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3067 goto nla_put_failure;
3068 }
3069 }
3070
3071 error = rt->dst.error;
3072
3073 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3074 goto nla_put_failure;
3075
3076 nlmsg_end(skb, nlh);
3077 return 0;
3078
3079 nla_put_failure:
3080 nlmsg_cancel(skb, nlh);
3081 return -EMSGSIZE;
3082 }
3083
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)3084 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3085 struct netlink_callback *cb, u32 table_id,
3086 struct fnhe_hash_bucket *bucket, int genid,
3087 int *fa_index, int fa_start, unsigned int flags)
3088 {
3089 int i;
3090
3091 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3092 struct fib_nh_exception *fnhe;
3093
3094 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3095 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3096 struct rtable *rt;
3097 int err;
3098
3099 if (*fa_index < fa_start)
3100 goto next;
3101
3102 if (fnhe->fnhe_genid != genid)
3103 goto next;
3104
3105 if (fnhe->fnhe_expires &&
3106 time_after(jiffies, fnhe->fnhe_expires))
3107 goto next;
3108
3109 rt = rcu_dereference(fnhe->fnhe_rth_input);
3110 if (!rt)
3111 rt = rcu_dereference(fnhe->fnhe_rth_output);
3112 if (!rt)
3113 goto next;
3114
3115 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3116 table_id, NULL, skb,
3117 NETLINK_CB(cb->skb).portid,
3118 cb->nlh->nlmsg_seq, flags);
3119 if (err)
3120 return err;
3121 next:
3122 (*fa_index)++;
3123 }
3124 }
3125
3126 return 0;
3127 }
3128
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)3129 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3130 u32 table_id, struct fib_info *fi,
3131 int *fa_index, int fa_start, unsigned int flags)
3132 {
3133 struct net *net = sock_net(cb->skb->sk);
3134 int nhsel, genid = fnhe_genid(net);
3135
3136 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3137 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3138 struct fnhe_hash_bucket *bucket;
3139 int err;
3140
3141 if (nhc->nhc_flags & RTNH_F_DEAD)
3142 continue;
3143
3144 rcu_read_lock();
3145 bucket = rcu_dereference(nhc->nhc_exceptions);
3146 err = 0;
3147 if (bucket)
3148 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3149 genid, fa_index, fa_start,
3150 flags);
3151 rcu_read_unlock();
3152 if (err)
3153 return err;
3154 }
3155
3156 return 0;
3157 }
3158
inet_rtm_getroute_build_skb(__be32 src,__be32 dst,u8 ip_proto,__be16 sport,__be16 dport)3159 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3160 u8 ip_proto, __be16 sport,
3161 __be16 dport)
3162 {
3163 struct sk_buff *skb;
3164 struct iphdr *iph;
3165
3166 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3167 if (!skb)
3168 return NULL;
3169
3170 /* Reserve room for dummy headers, this skb can pass
3171 * through good chunk of routing engine.
3172 */
3173 skb_reset_mac_header(skb);
3174 skb_reset_network_header(skb);
3175 skb->protocol = htons(ETH_P_IP);
3176 iph = skb_put(skb, sizeof(struct iphdr));
3177 iph->protocol = ip_proto;
3178 iph->saddr = src;
3179 iph->daddr = dst;
3180 iph->version = 0x4;
3181 iph->frag_off = 0;
3182 iph->ihl = 0x5;
3183 skb_set_transport_header(skb, skb->len);
3184
3185 switch (iph->protocol) {
3186 case IPPROTO_UDP: {
3187 struct udphdr *udph;
3188
3189 udph = skb_put_zero(skb, sizeof(struct udphdr));
3190 udph->source = sport;
3191 udph->dest = dport;
3192 udph->len = htons(sizeof(struct udphdr));
3193 udph->check = 0;
3194 break;
3195 }
3196 case IPPROTO_TCP: {
3197 struct tcphdr *tcph;
3198
3199 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3200 tcph->source = sport;
3201 tcph->dest = dport;
3202 tcph->doff = sizeof(struct tcphdr) / 4;
3203 tcph->rst = 1;
3204 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3205 src, dst, 0);
3206 break;
3207 }
3208 case IPPROTO_ICMP: {
3209 struct icmphdr *icmph;
3210
3211 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3212 icmph->type = ICMP_ECHO;
3213 icmph->code = 0;
3214 }
3215 }
3216
3217 return skb;
3218 }
3219
inet_rtm_valid_getroute_req(struct sk_buff * skb,const struct nlmsghdr * nlh,struct nlattr ** tb,struct netlink_ext_ack * extack)3220 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3221 const struct nlmsghdr *nlh,
3222 struct nlattr **tb,
3223 struct netlink_ext_ack *extack)
3224 {
3225 struct rtmsg *rtm;
3226 int i, err;
3227
3228 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3229 NL_SET_ERR_MSG(extack,
3230 "ipv4: Invalid header for route get request");
3231 return -EINVAL;
3232 }
3233
3234 if (!netlink_strict_get_check(skb))
3235 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3236 rtm_ipv4_policy, extack);
3237
3238 rtm = nlmsg_data(nlh);
3239 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3240 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3241 rtm->rtm_table || rtm->rtm_protocol ||
3242 rtm->rtm_scope || rtm->rtm_type) {
3243 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3244 return -EINVAL;
3245 }
3246
3247 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3248 RTM_F_LOOKUP_TABLE |
3249 RTM_F_FIB_MATCH)) {
3250 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3251 return -EINVAL;
3252 }
3253
3254 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3255 rtm_ipv4_policy, extack);
3256 if (err)
3257 return err;
3258
3259 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3260 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3261 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3262 return -EINVAL;
3263 }
3264
3265 for (i = 0; i <= RTA_MAX; i++) {
3266 if (!tb[i])
3267 continue;
3268
3269 switch (i) {
3270 case RTA_IIF:
3271 case RTA_OIF:
3272 case RTA_SRC:
3273 case RTA_DST:
3274 case RTA_IP_PROTO:
3275 case RTA_SPORT:
3276 case RTA_DPORT:
3277 case RTA_MARK:
3278 case RTA_UID:
3279 break;
3280 default:
3281 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3282 return -EINVAL;
3283 }
3284 }
3285
3286 return 0;
3287 }
3288
inet_rtm_getroute(struct sk_buff * in_skb,struct nlmsghdr * nlh,struct netlink_ext_ack * extack)3289 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3290 struct netlink_ext_ack *extack)
3291 {
3292 struct net *net = sock_net(in_skb->sk);
3293 struct nlattr *tb[RTA_MAX+1];
3294 u32 table_id = RT_TABLE_MAIN;
3295 __be16 sport = 0, dport = 0;
3296 struct fib_result res = {};
3297 u8 ip_proto = IPPROTO_UDP;
3298 struct rtable *rt = NULL;
3299 struct sk_buff *skb;
3300 struct rtmsg *rtm;
3301 struct flowi4 fl4 = {};
3302 __be32 dst = 0;
3303 __be32 src = 0;
3304 kuid_t uid;
3305 u32 iif;
3306 int err;
3307 int mark;
3308
3309 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3310 if (err < 0)
3311 return err;
3312
3313 rtm = nlmsg_data(nlh);
3314 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3315 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3316 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3317 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3318 if (tb[RTA_UID])
3319 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3320 else
3321 uid = (iif ? INVALID_UID : current_uid());
3322
3323 if (tb[RTA_IP_PROTO]) {
3324 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3325 &ip_proto, AF_INET, extack);
3326 if (err)
3327 return err;
3328 }
3329
3330 if (tb[RTA_SPORT])
3331 sport = nla_get_be16(tb[RTA_SPORT]);
3332
3333 if (tb[RTA_DPORT])
3334 dport = nla_get_be16(tb[RTA_DPORT]);
3335
3336 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3337 if (!skb)
3338 return -ENOBUFS;
3339
3340 fl4.daddr = dst;
3341 fl4.saddr = src;
3342 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3343 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3344 fl4.flowi4_mark = mark;
3345 fl4.flowi4_uid = uid;
3346 if (sport)
3347 fl4.fl4_sport = sport;
3348 if (dport)
3349 fl4.fl4_dport = dport;
3350 fl4.flowi4_proto = ip_proto;
3351
3352 rcu_read_lock();
3353
3354 if (iif) {
3355 struct net_device *dev;
3356
3357 dev = dev_get_by_index_rcu(net, iif);
3358 if (!dev) {
3359 err = -ENODEV;
3360 goto errout_rcu;
3361 }
3362
3363 fl4.flowi4_iif = iif; /* for rt_fill_info */
3364 skb->dev = dev;
3365 skb->mark = mark;
3366 err = ip_route_input_rcu(skb, dst, src,
3367 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3368 &res);
3369
3370 rt = skb_rtable(skb);
3371 if (err == 0 && rt->dst.error)
3372 err = -rt->dst.error;
3373 } else {
3374 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3375 skb->dev = net->loopback_dev;
3376 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3377 err = 0;
3378 if (IS_ERR(rt))
3379 err = PTR_ERR(rt);
3380 else
3381 skb_dst_set(skb, &rt->dst);
3382 }
3383
3384 if (err)
3385 goto errout_rcu;
3386
3387 if (rtm->rtm_flags & RTM_F_NOTIFY)
3388 rt->rt_flags |= RTCF_NOTIFY;
3389
3390 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3391 table_id = res.table ? res.table->tb_id : 0;
3392
3393 /* reset skb for netlink reply msg */
3394 skb_trim(skb, 0);
3395 skb_reset_network_header(skb);
3396 skb_reset_transport_header(skb);
3397 skb_reset_mac_header(skb);
3398
3399 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3400 struct fib_rt_info fri;
3401
3402 if (!res.fi) {
3403 err = fib_props[res.type].error;
3404 if (!err)
3405 err = -EHOSTUNREACH;
3406 goto errout_rcu;
3407 }
3408 fri.fi = res.fi;
3409 fri.tb_id = table_id;
3410 fri.dst = res.prefix;
3411 fri.dst_len = res.prefixlen;
3412 fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3413 fri.type = rt->rt_type;
3414 fri.offload = 0;
3415 fri.trap = 0;
3416 fri.offload_failed = 0;
3417 if (res.fa_head) {
3418 struct fib_alias *fa;
3419
3420 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3421 u8 slen = 32 - fri.dst_len;
3422
3423 if (fa->fa_slen == slen &&
3424 fa->tb_id == fri.tb_id &&
3425 fa->fa_dscp == fri.dscp &&
3426 fa->fa_info == res.fi &&
3427 fa->fa_type == fri.type) {
3428 fri.offload = READ_ONCE(fa->offload);
3429 fri.trap = READ_ONCE(fa->trap);
3430 break;
3431 }
3432 }
3433 }
3434 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3435 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3436 } else {
3437 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3438 NETLINK_CB(in_skb).portid,
3439 nlh->nlmsg_seq, 0);
3440 }
3441 if (err < 0)
3442 goto errout_rcu;
3443
3444 rcu_read_unlock();
3445
3446 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3447
3448 errout_free:
3449 return err;
3450 errout_rcu:
3451 rcu_read_unlock();
3452 kfree_skb(skb);
3453 goto errout_free;
3454 }
3455
ip_rt_multicast_event(struct in_device * in_dev)3456 void ip_rt_multicast_event(struct in_device *in_dev)
3457 {
3458 rt_cache_flush(dev_net(in_dev->dev));
3459 }
3460
3461 #ifdef CONFIG_SYSCTL
3462 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3463 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3464 static int ip_rt_gc_elasticity __read_mostly = 8;
3465 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3466
ipv4_sysctl_rtcache_flush(struct ctl_table * __ctl,int write,void * buffer,size_t * lenp,loff_t * ppos)3467 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3468 void *buffer, size_t *lenp, loff_t *ppos)
3469 {
3470 struct net *net = (struct net *)__ctl->extra1;
3471
3472 if (write) {
3473 rt_cache_flush(net);
3474 fnhe_genid_bump(net);
3475 return 0;
3476 }
3477
3478 return -EINVAL;
3479 }
3480
3481 static struct ctl_table ipv4_route_table[] = {
3482 {
3483 .procname = "gc_thresh",
3484 .data = &ipv4_dst_ops.gc_thresh,
3485 .maxlen = sizeof(int),
3486 .mode = 0644,
3487 .proc_handler = proc_dointvec,
3488 },
3489 {
3490 .procname = "max_size",
3491 .data = &ip_rt_max_size,
3492 .maxlen = sizeof(int),
3493 .mode = 0644,
3494 .proc_handler = proc_dointvec,
3495 },
3496 {
3497 /* Deprecated. Use gc_min_interval_ms */
3498
3499 .procname = "gc_min_interval",
3500 .data = &ip_rt_gc_min_interval,
3501 .maxlen = sizeof(int),
3502 .mode = 0644,
3503 .proc_handler = proc_dointvec_jiffies,
3504 },
3505 {
3506 .procname = "gc_min_interval_ms",
3507 .data = &ip_rt_gc_min_interval,
3508 .maxlen = sizeof(int),
3509 .mode = 0644,
3510 .proc_handler = proc_dointvec_ms_jiffies,
3511 },
3512 {
3513 .procname = "gc_timeout",
3514 .data = &ip_rt_gc_timeout,
3515 .maxlen = sizeof(int),
3516 .mode = 0644,
3517 .proc_handler = proc_dointvec_jiffies,
3518 },
3519 {
3520 .procname = "gc_interval",
3521 .data = &ip_rt_gc_interval,
3522 .maxlen = sizeof(int),
3523 .mode = 0644,
3524 .proc_handler = proc_dointvec_jiffies,
3525 },
3526 {
3527 .procname = "redirect_load",
3528 .data = &ip_rt_redirect_load,
3529 .maxlen = sizeof(int),
3530 .mode = 0644,
3531 .proc_handler = proc_dointvec,
3532 },
3533 {
3534 .procname = "redirect_number",
3535 .data = &ip_rt_redirect_number,
3536 .maxlen = sizeof(int),
3537 .mode = 0644,
3538 .proc_handler = proc_dointvec,
3539 },
3540 {
3541 .procname = "redirect_silence",
3542 .data = &ip_rt_redirect_silence,
3543 .maxlen = sizeof(int),
3544 .mode = 0644,
3545 .proc_handler = proc_dointvec,
3546 },
3547 {
3548 .procname = "error_cost",
3549 .data = &ip_rt_error_cost,
3550 .maxlen = sizeof(int),
3551 .mode = 0644,
3552 .proc_handler = proc_dointvec,
3553 },
3554 {
3555 .procname = "error_burst",
3556 .data = &ip_rt_error_burst,
3557 .maxlen = sizeof(int),
3558 .mode = 0644,
3559 .proc_handler = proc_dointvec,
3560 },
3561 {
3562 .procname = "gc_elasticity",
3563 .data = &ip_rt_gc_elasticity,
3564 .maxlen = sizeof(int),
3565 .mode = 0644,
3566 .proc_handler = proc_dointvec,
3567 },
3568 { }
3569 };
3570
3571 static const char ipv4_route_flush_procname[] = "flush";
3572
3573 static struct ctl_table ipv4_route_netns_table[] = {
3574 {
3575 .procname = ipv4_route_flush_procname,
3576 .maxlen = sizeof(int),
3577 .mode = 0200,
3578 .proc_handler = ipv4_sysctl_rtcache_flush,
3579 },
3580 {
3581 .procname = "min_pmtu",
3582 .data = &init_net.ipv4.ip_rt_min_pmtu,
3583 .maxlen = sizeof(int),
3584 .mode = 0644,
3585 .proc_handler = proc_dointvec_minmax,
3586 .extra1 = &ip_min_valid_pmtu,
3587 },
3588 {
3589 .procname = "mtu_expires",
3590 .data = &init_net.ipv4.ip_rt_mtu_expires,
3591 .maxlen = sizeof(int),
3592 .mode = 0644,
3593 .proc_handler = proc_dointvec_jiffies,
3594 },
3595 {
3596 .procname = "min_adv_mss",
3597 .data = &init_net.ipv4.ip_rt_min_advmss,
3598 .maxlen = sizeof(int),
3599 .mode = 0644,
3600 .proc_handler = proc_dointvec,
3601 },
3602 { },
3603 };
3604
sysctl_route_net_init(struct net * net)3605 static __net_init int sysctl_route_net_init(struct net *net)
3606 {
3607 struct ctl_table *tbl;
3608
3609 tbl = ipv4_route_netns_table;
3610 if (!net_eq(net, &init_net)) {
3611 int i;
3612
3613 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3614 if (!tbl)
3615 goto err_dup;
3616
3617 /* Don't export non-whitelisted sysctls to unprivileged users */
3618 if (net->user_ns != &init_user_ns) {
3619 if (tbl[0].procname != ipv4_route_flush_procname)
3620 tbl[0].procname = NULL;
3621 }
3622
3623 /* Update the variables to point into the current struct net
3624 * except for the first element flush
3625 */
3626 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3627 tbl[i].data += (void *)net - (void *)&init_net;
3628 }
3629 tbl[0].extra1 = net;
3630
3631 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3632 if (!net->ipv4.route_hdr)
3633 goto err_reg;
3634 return 0;
3635
3636 err_reg:
3637 if (tbl != ipv4_route_netns_table)
3638 kfree(tbl);
3639 err_dup:
3640 return -ENOMEM;
3641 }
3642
sysctl_route_net_exit(struct net * net)3643 static __net_exit void sysctl_route_net_exit(struct net *net)
3644 {
3645 struct ctl_table *tbl;
3646
3647 tbl = net->ipv4.route_hdr->ctl_table_arg;
3648 unregister_net_sysctl_table(net->ipv4.route_hdr);
3649 BUG_ON(tbl == ipv4_route_netns_table);
3650 kfree(tbl);
3651 }
3652
3653 static __net_initdata struct pernet_operations sysctl_route_ops = {
3654 .init = sysctl_route_net_init,
3655 .exit = sysctl_route_net_exit,
3656 };
3657 #endif
3658
netns_ip_rt_init(struct net * net)3659 static __net_init int netns_ip_rt_init(struct net *net)
3660 {
3661 /* Set default value for namespaceified sysctls */
3662 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3663 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3664 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3665 return 0;
3666 }
3667
3668 static struct pernet_operations __net_initdata ip_rt_ops = {
3669 .init = netns_ip_rt_init,
3670 };
3671
rt_genid_init(struct net * net)3672 static __net_init int rt_genid_init(struct net *net)
3673 {
3674 atomic_set(&net->ipv4.rt_genid, 0);
3675 atomic_set(&net->fnhe_genid, 0);
3676 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3677 return 0;
3678 }
3679
3680 static __net_initdata struct pernet_operations rt_genid_ops = {
3681 .init = rt_genid_init,
3682 };
3683
ipv4_inetpeer_init(struct net * net)3684 static int __net_init ipv4_inetpeer_init(struct net *net)
3685 {
3686 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3687
3688 if (!bp)
3689 return -ENOMEM;
3690 inet_peer_base_init(bp);
3691 net->ipv4.peers = bp;
3692 return 0;
3693 }
3694
ipv4_inetpeer_exit(struct net * net)3695 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3696 {
3697 struct inet_peer_base *bp = net->ipv4.peers;
3698
3699 net->ipv4.peers = NULL;
3700 inetpeer_invalidate_tree(bp);
3701 kfree(bp);
3702 }
3703
3704 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3705 .init = ipv4_inetpeer_init,
3706 .exit = ipv4_inetpeer_exit,
3707 };
3708
3709 #ifdef CONFIG_IP_ROUTE_CLASSID
3710 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3711 #endif /* CONFIG_IP_ROUTE_CLASSID */
3712
ip_rt_init(void)3713 int __init ip_rt_init(void)
3714 {
3715 void *idents_hash;
3716 int cpu;
3717
3718 /* For modern hosts, this will use 2 MB of memory */
3719 idents_hash = alloc_large_system_hash("IP idents",
3720 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3721 0,
3722 16, /* one bucket per 64 KB */
3723 HASH_ZERO,
3724 NULL,
3725 &ip_idents_mask,
3726 2048,
3727 256*1024);
3728
3729 ip_idents = idents_hash;
3730
3731 prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3732
3733 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3734
3735 for_each_possible_cpu(cpu) {
3736 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3737
3738 INIT_LIST_HEAD(&ul->head);
3739 INIT_LIST_HEAD(&ul->quarantine);
3740 spin_lock_init(&ul->lock);
3741 }
3742 #ifdef CONFIG_IP_ROUTE_CLASSID
3743 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3744 if (!ip_rt_acct)
3745 panic("IP: failed to allocate ip_rt_acct\n");
3746 #endif
3747
3748 ipv4_dst_ops.kmem_cachep =
3749 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3750 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3751
3752 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3753
3754 if (dst_entries_init(&ipv4_dst_ops) < 0)
3755 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3756
3757 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3758 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3759
3760 ipv4_dst_ops.gc_thresh = ~0;
3761 ip_rt_max_size = INT_MAX;
3762
3763 devinet_init();
3764 ip_fib_init();
3765
3766 if (ip_rt_proc_init())
3767 pr_err("Unable to create route proc files\n");
3768 #ifdef CONFIG_XFRM
3769 xfrm_init();
3770 xfrm4_init();
3771 #endif
3772 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3773 RTNL_FLAG_DOIT_UNLOCKED);
3774
3775 #ifdef CONFIG_SYSCTL
3776 register_pernet_subsys(&sysctl_route_ops);
3777 #endif
3778 register_pernet_subsys(&ip_rt_ops);
3779 register_pernet_subsys(&rt_genid_ops);
3780 register_pernet_subsys(&ipv4_inetpeer_ops);
3781 return 0;
3782 }
3783
3784 #ifdef CONFIG_SYSCTL
3785 /*
3786 * We really need to sanitize the damn ipv4 init order, then all
3787 * this nonsense will go away.
3788 */
ip_static_sysctl_init(void)3789 void __init ip_static_sysctl_init(void)
3790 {
3791 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3792 }
3793 #endif
3794