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_GC_TIMEOUT (300*HZ)
110
111 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
112 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
113 #define DEFAULT_MIN_ADVMSS 256
114 static int ip_rt_max_size;
115 static int ip_rt_redirect_number __read_mostly = 9;
116 static int ip_rt_redirect_load __read_mostly = HZ / 50;
117 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
118 static int ip_rt_error_cost __read_mostly = HZ;
119 static int ip_rt_error_burst __read_mostly = 5 * HZ;
120
121 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
122
123 /*
124 * Interface to generic destination cache.
125 */
126
127 INDIRECT_CALLABLE_SCOPE
128 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
129 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
130 INDIRECT_CALLABLE_SCOPE
131 unsigned int ipv4_mtu(const struct dst_entry *dst);
132 static void ipv4_negative_advice(struct sock *sk,
133 struct dst_entry *dst);
134 static void ipv4_link_failure(struct sk_buff *skb);
135 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
136 struct sk_buff *skb, u32 mtu,
137 bool confirm_neigh);
138 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
139 struct sk_buff *skb);
140 static void ipv4_dst_destroy(struct dst_entry *dst);
141
ipv4_cow_metrics(struct dst_entry * dst,unsigned long old)142 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
143 {
144 WARN_ON(1);
145 return NULL;
146 }
147
148 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
149 struct sk_buff *skb,
150 const void *daddr);
151 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
152
153 static struct dst_ops ipv4_dst_ops = {
154 .family = AF_INET,
155 .check = ipv4_dst_check,
156 .default_advmss = ipv4_default_advmss,
157 .mtu = ipv4_mtu,
158 .cow_metrics = ipv4_cow_metrics,
159 .destroy = ipv4_dst_destroy,
160 .negative_advice = ipv4_negative_advice,
161 .link_failure = ipv4_link_failure,
162 .update_pmtu = ip_rt_update_pmtu,
163 .redirect = ip_do_redirect,
164 .local_out = __ip_local_out,
165 .neigh_lookup = ipv4_neigh_lookup,
166 .confirm_neigh = ipv4_confirm_neigh,
167 };
168
169 #define ECN_OR_COST(class) TC_PRIO_##class
170
171 const __u8 ip_tos2prio[16] = {
172 TC_PRIO_BESTEFFORT,
173 ECN_OR_COST(BESTEFFORT),
174 TC_PRIO_BESTEFFORT,
175 ECN_OR_COST(BESTEFFORT),
176 TC_PRIO_BULK,
177 ECN_OR_COST(BULK),
178 TC_PRIO_BULK,
179 ECN_OR_COST(BULK),
180 TC_PRIO_INTERACTIVE,
181 ECN_OR_COST(INTERACTIVE),
182 TC_PRIO_INTERACTIVE,
183 ECN_OR_COST(INTERACTIVE),
184 TC_PRIO_INTERACTIVE_BULK,
185 ECN_OR_COST(INTERACTIVE_BULK),
186 TC_PRIO_INTERACTIVE_BULK,
187 ECN_OR_COST(INTERACTIVE_BULK)
188 };
189 EXPORT_SYMBOL(ip_tos2prio);
190
191 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
192 #ifndef CONFIG_PREEMPT_RT
193 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
194 #else
195 #define RT_CACHE_STAT_INC(field) this_cpu_inc(rt_cache_stat.field)
196 #endif
197
198 #ifdef CONFIG_PROC_FS
rt_cache_seq_start(struct seq_file * seq,loff_t * pos)199 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
200 {
201 if (*pos)
202 return NULL;
203 return SEQ_START_TOKEN;
204 }
205
rt_cache_seq_next(struct seq_file * seq,void * v,loff_t * pos)206 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
207 {
208 ++*pos;
209 return NULL;
210 }
211
rt_cache_seq_stop(struct seq_file * seq,void * v)212 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
213 {
214 }
215
rt_cache_seq_show(struct seq_file * seq,void * v)216 static int rt_cache_seq_show(struct seq_file *seq, void *v)
217 {
218 if (v == SEQ_START_TOKEN)
219 seq_printf(seq, "%-127s\n",
220 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
221 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
222 "HHUptod\tSpecDst");
223 return 0;
224 }
225
226 static const struct seq_operations rt_cache_seq_ops = {
227 .start = rt_cache_seq_start,
228 .next = rt_cache_seq_next,
229 .stop = rt_cache_seq_stop,
230 .show = rt_cache_seq_show,
231 };
232
rt_cpu_seq_start(struct seq_file * seq,loff_t * pos)233 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
234 {
235 int cpu;
236
237 if (*pos == 0)
238 return SEQ_START_TOKEN;
239
240 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
241 if (!cpu_possible(cpu))
242 continue;
243 *pos = cpu+1;
244 return &per_cpu(rt_cache_stat, cpu);
245 }
246 return NULL;
247 }
248
rt_cpu_seq_next(struct seq_file * seq,void * v,loff_t * pos)249 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
250 {
251 int cpu;
252
253 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
254 if (!cpu_possible(cpu))
255 continue;
256 *pos = cpu+1;
257 return &per_cpu(rt_cache_stat, cpu);
258 }
259 (*pos)++;
260 return NULL;
261
262 }
263
rt_cpu_seq_stop(struct seq_file * seq,void * v)264 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
265 {
266
267 }
268
rt_cpu_seq_show(struct seq_file * seq,void * v)269 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
270 {
271 struct rt_cache_stat *st = v;
272
273 if (v == SEQ_START_TOKEN) {
274 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");
275 return 0;
276 }
277
278 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x "
279 "%08x %08x %08x %08x %08x %08x "
280 "%08x %08x %08x %08x\n",
281 dst_entries_get_slow(&ipv4_dst_ops),
282 0, /* st->in_hit */
283 st->in_slow_tot,
284 st->in_slow_mc,
285 st->in_no_route,
286 st->in_brd,
287 st->in_martian_dst,
288 st->in_martian_src,
289
290 0, /* st->out_hit */
291 st->out_slow_tot,
292 st->out_slow_mc,
293
294 0, /* st->gc_total */
295 0, /* st->gc_ignored */
296 0, /* st->gc_goal_miss */
297 0, /* st->gc_dst_overflow */
298 0, /* st->in_hlist_search */
299 0 /* st->out_hlist_search */
300 );
301 return 0;
302 }
303
304 static const struct seq_operations rt_cpu_seq_ops = {
305 .start = rt_cpu_seq_start,
306 .next = rt_cpu_seq_next,
307 .stop = rt_cpu_seq_stop,
308 .show = rt_cpu_seq_show,
309 };
310
311 #ifdef CONFIG_IP_ROUTE_CLASSID
rt_acct_proc_show(struct seq_file * m,void * v)312 static int rt_acct_proc_show(struct seq_file *m, void *v)
313 {
314 struct ip_rt_acct *dst, *src;
315 unsigned int i, j;
316
317 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
318 if (!dst)
319 return -ENOMEM;
320
321 for_each_possible_cpu(i) {
322 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
323 for (j = 0; j < 256; j++) {
324 dst[j].o_bytes += src[j].o_bytes;
325 dst[j].o_packets += src[j].o_packets;
326 dst[j].i_bytes += src[j].i_bytes;
327 dst[j].i_packets += src[j].i_packets;
328 }
329 }
330
331 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
332 kfree(dst);
333 return 0;
334 }
335 #endif
336
ip_rt_do_proc_init(struct net * net)337 static int __net_init ip_rt_do_proc_init(struct net *net)
338 {
339 struct proc_dir_entry *pde;
340
341 pde = proc_create_seq("rt_cache", 0444, net->proc_net,
342 &rt_cache_seq_ops);
343 if (!pde)
344 goto err1;
345
346 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
347 &rt_cpu_seq_ops);
348 if (!pde)
349 goto err2;
350
351 #ifdef CONFIG_IP_ROUTE_CLASSID
352 pde = proc_create_single("rt_acct", 0, net->proc_net,
353 rt_acct_proc_show);
354 if (!pde)
355 goto err3;
356 #endif
357 return 0;
358
359 #ifdef CONFIG_IP_ROUTE_CLASSID
360 err3:
361 remove_proc_entry("rt_cache", net->proc_net_stat);
362 #endif
363 err2:
364 remove_proc_entry("rt_cache", net->proc_net);
365 err1:
366 return -ENOMEM;
367 }
368
ip_rt_do_proc_exit(struct net * net)369 static void __net_exit ip_rt_do_proc_exit(struct net *net)
370 {
371 remove_proc_entry("rt_cache", net->proc_net_stat);
372 remove_proc_entry("rt_cache", net->proc_net);
373 #ifdef CONFIG_IP_ROUTE_CLASSID
374 remove_proc_entry("rt_acct", net->proc_net);
375 #endif
376 }
377
378 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
379 .init = ip_rt_do_proc_init,
380 .exit = ip_rt_do_proc_exit,
381 };
382
ip_rt_proc_init(void)383 static int __init ip_rt_proc_init(void)
384 {
385 return register_pernet_subsys(&ip_rt_proc_ops);
386 }
387
388 #else
ip_rt_proc_init(void)389 static inline int ip_rt_proc_init(void)
390 {
391 return 0;
392 }
393 #endif /* CONFIG_PROC_FS */
394
rt_is_expired(const struct rtable * rth)395 static inline bool rt_is_expired(const struct rtable *rth)
396 {
397 bool res;
398
399 rcu_read_lock();
400 res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev));
401 rcu_read_unlock();
402
403 return res;
404 }
405
rt_cache_flush(struct net * net)406 void rt_cache_flush(struct net *net)
407 {
408 rt_genid_bump_ipv4(net);
409 }
410
ipv4_neigh_lookup(const struct dst_entry * dst,struct sk_buff * skb,const void * daddr)411 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
412 struct sk_buff *skb,
413 const void *daddr)
414 {
415 const struct rtable *rt = container_of(dst, struct rtable, dst);
416 struct net_device *dev = dst_dev(dst);
417 struct neighbour *n;
418
419 rcu_read_lock();
420
421 if (likely(rt->rt_gw_family == AF_INET)) {
422 n = ip_neigh_gw4(dev, rt->rt_gw4);
423 } else if (rt->rt_gw_family == AF_INET6) {
424 n = ip_neigh_gw6(dev, &rt->rt_gw6);
425 } else {
426 __be32 pkey;
427
428 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
429 n = ip_neigh_gw4(dev, pkey);
430 }
431
432 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
433 n = NULL;
434
435 rcu_read_unlock();
436
437 return n;
438 }
439
ipv4_confirm_neigh(const struct dst_entry * dst,const void * daddr)440 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
441 {
442 const struct rtable *rt = container_of(dst, struct rtable, dst);
443 struct net_device *dev = dst_dev(dst);
444 const __be32 *pkey = daddr;
445
446 if (rt->rt_gw_family == AF_INET) {
447 pkey = (const __be32 *)&rt->rt_gw4;
448 } else if (rt->rt_gw_family == AF_INET6) {
449 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
450 } else if (!daddr ||
451 (rt->rt_flags &
452 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
453 return;
454 }
455 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
456 }
457
458 /* Hash tables of size 2048..262144 depending on RAM size.
459 * Each bucket uses 8 bytes.
460 */
461 static u32 ip_idents_mask __read_mostly;
462 static atomic_t *ip_idents __read_mostly;
463 static u32 *ip_tstamps __read_mostly;
464
465 /* In order to protect privacy, we add a perturbation to identifiers
466 * if one generator is seldom used. This makes hard for an attacker
467 * to infer how many packets were sent between two points in time.
468 */
ip_idents_reserve(u32 hash,int segs)469 static u32 ip_idents_reserve(u32 hash, int segs)
470 {
471 u32 bucket, old, now = (u32)jiffies;
472 atomic_t *p_id;
473 u32 *p_tstamp;
474 u32 delta = 0;
475
476 bucket = hash & ip_idents_mask;
477 p_tstamp = ip_tstamps + bucket;
478 p_id = ip_idents + bucket;
479 old = READ_ONCE(*p_tstamp);
480
481 if (old != now && cmpxchg(p_tstamp, old, now) == old)
482 delta = get_random_u32_below(now - old);
483
484 /* If UBSAN reports an error there, please make sure your compiler
485 * supports -fno-strict-overflow before reporting it that was a bug
486 * in UBSAN, and it has been fixed in GCC-8.
487 */
488 return atomic_add_return(segs + delta, p_id) - segs;
489 }
490
__ip_select_ident(struct net * net,struct iphdr * iph,int segs)491 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
492 {
493 u32 hash, id;
494
495 /* Note the following code is not safe, but this is okay. */
496 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
497 get_random_bytes(&net->ipv4.ip_id_key,
498 sizeof(net->ipv4.ip_id_key));
499
500 hash = siphash_3u32((__force u32)iph->daddr,
501 (__force u32)iph->saddr,
502 iph->protocol,
503 &net->ipv4.ip_id_key);
504 id = ip_idents_reserve(hash, segs);
505 iph->id = htons(id);
506 }
507 EXPORT_SYMBOL(__ip_select_ident);
508
__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)509 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
510 const struct sock *sk, const struct iphdr *iph,
511 int oif, __u8 tos, u8 prot, u32 mark,
512 int flow_flags)
513 {
514 __u8 scope = RT_SCOPE_UNIVERSE;
515
516 if (sk) {
517 oif = sk->sk_bound_dev_if;
518 mark = READ_ONCE(sk->sk_mark);
519 tos = ip_sock_rt_tos(sk);
520 scope = ip_sock_rt_scope(sk);
521 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
522 sk->sk_protocol;
523 }
524
525 flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
526 prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
527 sock_net_uid(net, sk));
528 }
529
build_skb_flow_key(struct flowi4 * fl4,const struct sk_buff * skb,const struct sock * sk)530 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
531 const struct sock *sk)
532 {
533 const struct net *net = dev_net(skb->dev);
534 const struct iphdr *iph = ip_hdr(skb);
535 int oif = skb->dev->ifindex;
536 u8 prot = iph->protocol;
537 u32 mark = skb->mark;
538 __u8 tos = iph->tos;
539
540 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
541 }
542
build_sk_flow_key(struct flowi4 * fl4,const struct sock * sk)543 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
544 {
545 const struct inet_sock *inet = inet_sk(sk);
546 const struct ip_options_rcu *inet_opt;
547 __be32 daddr = inet->inet_daddr;
548
549 rcu_read_lock();
550 inet_opt = rcu_dereference(inet->inet_opt);
551 if (inet_opt && inet_opt->opt.srr)
552 daddr = inet_opt->opt.faddr;
553 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
554 ip_sock_rt_tos(sk),
555 ip_sock_rt_scope(sk),
556 inet_test_bit(HDRINCL, sk) ?
557 IPPROTO_RAW : sk->sk_protocol,
558 inet_sk_flowi_flags(sk),
559 daddr, inet->inet_saddr, 0, 0,
560 sk_uid(sk));
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 get_random_u32_below(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 WRITE_ONCE(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 WRITE_ONCE(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, (__force u32)new_gw);
784 if (!n)
785 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
786 if (!IS_ERR(n)) {
787 if (!(READ_ONCE(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 WRITE_ONCE(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 = dst_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 sock * sk,struct dst_entry * dst)840 static void ipv4_negative_advice(struct sock *sk,
841 struct dst_entry *dst)
842 {
843 struct rtable *rt = dst_rtable(dst);
844
845 if ((READ_ONCE(dst->obsolete) > 0) ||
846 (rt->rt_flags & RTCF_REDIRECTED) ||
847 READ_ONCE(rt->dst.expires))
848 sk_dst_reset(sk);
849 }
850
851 /*
852 * Algorithm:
853 * 1. The first ip_rt_redirect_number redirects are sent
854 * with exponential backoff, then we stop sending them at all,
855 * assuming that the host ignores our redirects.
856 * 2. If we did not see packets requiring redirects
857 * during ip_rt_redirect_silence, we assume that the host
858 * forgot redirected route and start to send redirects again.
859 *
860 * This algorithm is much cheaper and more intelligent than dumb load limiting
861 * in icmp.c.
862 *
863 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
864 * and "frag. need" (breaks PMTU discovery) in icmp.c.
865 */
866
ip_rt_send_redirect(struct sk_buff * skb)867 void ip_rt_send_redirect(struct sk_buff *skb)
868 {
869 struct rtable *rt = skb_rtable(skb);
870 struct in_device *in_dev;
871 struct inet_peer *peer;
872 struct net *net;
873 int log_martians;
874 int vif;
875
876 rcu_read_lock();
877 in_dev = __in_dev_get_rcu(rt->dst.dev);
878 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
879 rcu_read_unlock();
880 return;
881 }
882 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
883 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
884
885 net = dev_net(rt->dst.dev);
886 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
887 if (!peer) {
888 rcu_read_unlock();
889 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
890 rt_nexthop(rt, ip_hdr(skb)->daddr));
891 return;
892 }
893
894 /* No redirected packets during ip_rt_redirect_silence;
895 * reset the algorithm.
896 */
897 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
898 peer->rate_tokens = 0;
899 peer->n_redirects = 0;
900 }
901
902 /* Too many ignored redirects; do not send anything
903 * set dst.rate_last to the last seen redirected packet.
904 */
905 if (peer->n_redirects >= ip_rt_redirect_number) {
906 peer->rate_last = jiffies;
907 goto out_unlock;
908 }
909
910 /* Check for load limit; set rate_last to the latest sent
911 * redirect.
912 */
913 if (peer->n_redirects == 0 ||
914 time_after(jiffies,
915 (peer->rate_last +
916 (ip_rt_redirect_load << peer->n_redirects)))) {
917 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
918
919 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
920 peer->rate_last = jiffies;
921 ++peer->n_redirects;
922 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
923 peer->n_redirects == ip_rt_redirect_number)
924 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
925 &ip_hdr(skb)->saddr, inet_iif(skb),
926 &ip_hdr(skb)->daddr, &gw);
927 }
928 out_unlock:
929 rcu_read_unlock();
930 }
931
ip_error(struct sk_buff * skb)932 static int ip_error(struct sk_buff *skb)
933 {
934 struct rtable *rt = skb_rtable(skb);
935 struct net_device *dev = skb->dev;
936 struct in_device *in_dev;
937 struct inet_peer *peer;
938 unsigned long now;
939 struct net *net;
940 SKB_DR(reason);
941 bool send;
942 int code;
943
944 if (netif_is_l3_master(skb->dev)) {
945 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
946 if (!dev)
947 goto out;
948 }
949
950 in_dev = __in_dev_get_rcu(dev);
951
952 /* IP on this device is disabled. */
953 if (!in_dev)
954 goto out;
955
956 net = dev_net(rt->dst.dev);
957 if (!IN_DEV_FORWARD(in_dev)) {
958 switch (rt->dst.error) {
959 case EHOSTUNREACH:
960 SKB_DR_SET(reason, IP_INADDRERRORS);
961 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
962 break;
963
964 case ENETUNREACH:
965 SKB_DR_SET(reason, IP_INNOROUTES);
966 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
967 break;
968 }
969 goto out;
970 }
971
972 switch (rt->dst.error) {
973 case EINVAL:
974 default:
975 goto out;
976 case EHOSTUNREACH:
977 code = ICMP_HOST_UNREACH;
978 break;
979 case ENETUNREACH:
980 code = ICMP_NET_UNREACH;
981 SKB_DR_SET(reason, IP_INNOROUTES);
982 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
983 break;
984 case EACCES:
985 code = ICMP_PKT_FILTERED;
986 break;
987 }
988
989 rcu_read_lock();
990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
991 l3mdev_master_ifindex_rcu(skb->dev));
992 send = true;
993 if (peer) {
994 now = jiffies;
995 peer->rate_tokens += now - peer->rate_last;
996 if (peer->rate_tokens > ip_rt_error_burst)
997 peer->rate_tokens = ip_rt_error_burst;
998 peer->rate_last = now;
999 if (peer->rate_tokens >= ip_rt_error_cost)
1000 peer->rate_tokens -= ip_rt_error_cost;
1001 else
1002 send = false;
1003 }
1004 rcu_read_unlock();
1005
1006 if (send)
1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1008
1009 out: kfree_skb_reason(skb, reason);
1010 return 0;
1011 }
1012
__ip_rt_update_pmtu(struct rtable * rt,struct flowi4 * fl4,u32 mtu)1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1014 {
1015 struct dst_entry *dst = &rt->dst;
1016 struct fib_result res;
1017 bool lock = false;
1018 struct net *net;
1019 u32 old_mtu;
1020
1021 if (ip_mtu_locked(dst))
1022 return;
1023
1024 old_mtu = ipv4_mtu(dst);
1025 if (old_mtu < mtu)
1026 return;
1027
1028 rcu_read_lock();
1029 net = dev_net_rcu(dst_dev(dst));
1030 if (mtu < net->ipv4.ip_rt_min_pmtu) {
1031 lock = true;
1032 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1033 }
1034
1035 if (rt->rt_pmtu == mtu && !lock &&
1036 time_before(jiffies, READ_ONCE(dst->expires) -
1037 net->ipv4.ip_rt_mtu_expires / 2))
1038 goto out;
1039
1040 if (fib_lookup(net, fl4, &res, 0) == 0) {
1041 struct fib_nh_common *nhc;
1042
1043 fib_select_path(net, &res, fl4, NULL);
1044 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1045 if (fib_info_num_path(res.fi) > 1) {
1046 int nhsel;
1047
1048 for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1049 nhc = fib_info_nhc(res.fi, nhsel);
1050 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1051 jiffies + net->ipv4.ip_rt_mtu_expires);
1052 }
1053 goto out;
1054 }
1055 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1056 nhc = FIB_RES_NHC(res);
1057 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1058 jiffies + net->ipv4.ip_rt_mtu_expires);
1059 }
1060 out:
1061 rcu_read_unlock();
1062 }
1063
ip_rt_update_pmtu(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb,u32 mtu,bool confirm_neigh)1064 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1065 struct sk_buff *skb, u32 mtu,
1066 bool confirm_neigh)
1067 {
1068 struct rtable *rt = dst_rtable(dst);
1069 struct flowi4 fl4;
1070
1071 ip_rt_build_flow_key(&fl4, sk, skb);
1072
1073 /* Don't make lookup fail for bridged encapsulations */
1074 if (skb && netif_is_any_bridge_port(skb->dev))
1075 fl4.flowi4_oif = 0;
1076
1077 __ip_rt_update_pmtu(rt, &fl4, mtu);
1078 }
1079
ipv4_update_pmtu(struct sk_buff * skb,struct net * net,u32 mtu,int oif,u8 protocol)1080 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1081 int oif, u8 protocol)
1082 {
1083 const struct iphdr *iph = (const struct iphdr *)skb->data;
1084 struct flowi4 fl4;
1085 struct rtable *rt;
1086 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1087
1088 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1089 0);
1090 rt = __ip_route_output_key(net, &fl4);
1091 if (!IS_ERR(rt)) {
1092 __ip_rt_update_pmtu(rt, &fl4, mtu);
1093 ip_rt_put(rt);
1094 }
1095 }
1096 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1097
__ipv4_sk_update_pmtu(struct sk_buff * skb,struct sock * sk,u32 mtu)1098 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1099 {
1100 const struct iphdr *iph = (const struct iphdr *)skb->data;
1101 struct flowi4 fl4;
1102 struct rtable *rt;
1103
1104 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1105
1106 if (!fl4.flowi4_mark)
1107 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1108
1109 rt = __ip_route_output_key(sock_net(sk), &fl4);
1110 if (!IS_ERR(rt)) {
1111 __ip_rt_update_pmtu(rt, &fl4, mtu);
1112 ip_rt_put(rt);
1113 }
1114 }
1115
ipv4_sk_update_pmtu(struct sk_buff * skb,struct sock * sk,u32 mtu)1116 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1117 {
1118 const struct iphdr *iph = (const struct iphdr *)skb->data;
1119 struct flowi4 fl4;
1120 struct rtable *rt;
1121 struct dst_entry *odst = NULL;
1122 bool new = false;
1123 struct net *net = sock_net(sk);
1124
1125 bh_lock_sock(sk);
1126
1127 if (!ip_sk_accept_pmtu(sk))
1128 goto out;
1129
1130 odst = sk_dst_get(sk);
1131
1132 if (sock_owned_by_user(sk) || !odst) {
1133 __ipv4_sk_update_pmtu(skb, sk, mtu);
1134 goto out;
1135 }
1136
1137 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1138
1139 rt = dst_rtable(odst);
1140 if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) {
1141 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1142 if (IS_ERR(rt))
1143 goto out;
1144
1145 new = true;
1146 }
1147
1148 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1149
1150 if (!dst_check(&rt->dst, 0)) {
1151 if (new)
1152 dst_release(&rt->dst);
1153
1154 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1155 if (IS_ERR(rt))
1156 goto out;
1157
1158 new = true;
1159 }
1160
1161 if (new)
1162 sk_dst_set(sk, &rt->dst);
1163
1164 out:
1165 bh_unlock_sock(sk);
1166 dst_release(odst);
1167 }
1168 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1169
ipv4_redirect(struct sk_buff * skb,struct net * net,int oif,u8 protocol)1170 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1171 int oif, u8 protocol)
1172 {
1173 const struct iphdr *iph = (const struct iphdr *)skb->data;
1174 struct flowi4 fl4;
1175 struct rtable *rt;
1176
1177 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1178 rt = __ip_route_output_key(net, &fl4);
1179 if (!IS_ERR(rt)) {
1180 __ip_do_redirect(rt, skb, &fl4, false);
1181 ip_rt_put(rt);
1182 }
1183 }
1184 EXPORT_SYMBOL_GPL(ipv4_redirect);
1185
ipv4_sk_redirect(struct sk_buff * skb,struct sock * sk)1186 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1187 {
1188 const struct iphdr *iph = (const struct iphdr *)skb->data;
1189 struct flowi4 fl4;
1190 struct rtable *rt;
1191 struct net *net = sock_net(sk);
1192
1193 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1194 rt = __ip_route_output_key(net, &fl4);
1195 if (!IS_ERR(rt)) {
1196 __ip_do_redirect(rt, skb, &fl4, false);
1197 ip_rt_put(rt);
1198 }
1199 }
1200 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1201
ipv4_dst_check(struct dst_entry * dst,u32 cookie)1202 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1203 u32 cookie)
1204 {
1205 struct rtable *rt = dst_rtable(dst);
1206
1207 /* All IPV4 dsts are created with ->obsolete set to the value
1208 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1209 * into this function always.
1210 *
1211 * When a PMTU/redirect information update invalidates a route,
1212 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1213 * DST_OBSOLETE_DEAD.
1214 */
1215 if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK ||
1216 rt_is_expired(rt))
1217 return NULL;
1218 return dst;
1219 }
1220 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1221
ipv4_send_dest_unreach(struct sk_buff * skb)1222 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1223 {
1224 struct net_device *dev;
1225 struct ip_options opt;
1226 int res;
1227
1228 /* Recompile ip options since IPCB may not be valid anymore.
1229 * Also check we have a reasonable ipv4 header.
1230 */
1231 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1232 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1233 return;
1234
1235 memset(&opt, 0, sizeof(opt));
1236 if (ip_hdr(skb)->ihl > 5) {
1237 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1238 return;
1239 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1240
1241 rcu_read_lock();
1242 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1243 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1244 rcu_read_unlock();
1245
1246 if (res)
1247 return;
1248 }
1249 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1250 }
1251
ipv4_link_failure(struct sk_buff * skb)1252 static void ipv4_link_failure(struct sk_buff *skb)
1253 {
1254 struct rtable *rt;
1255
1256 ipv4_send_dest_unreach(skb);
1257
1258 rt = skb_rtable(skb);
1259 if (rt)
1260 dst_set_expires(&rt->dst, 0);
1261 }
1262
ip_rt_bug(struct net * net,struct sock * sk,struct sk_buff * skb)1263 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1264 {
1265 pr_debug("%s: %pI4 -> %pI4, %s\n",
1266 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1267 skb->dev ? skb->dev->name : "?");
1268 kfree_skb(skb);
1269 WARN_ON(1);
1270 return 0;
1271 }
1272
1273 /*
1274 * We do not cache source address of outgoing interface,
1275 * because it is used only by IP RR, TS and SRR options,
1276 * so that it out of fast path.
1277 *
1278 * BTW remember: "addr" is allowed to be not aligned
1279 * in IP options!
1280 */
1281
ip_rt_get_source(u8 * addr,struct sk_buff * skb,struct rtable * rt)1282 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1283 {
1284 __be32 src;
1285
1286 if (rt_is_output_route(rt))
1287 src = ip_hdr(skb)->saddr;
1288 else {
1289 struct fib_result res;
1290 struct iphdr *iph = ip_hdr(skb);
1291 struct flowi4 fl4 = {
1292 .daddr = iph->daddr,
1293 .saddr = iph->saddr,
1294 .flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(iph)),
1295 .flowi4_oif = rt->dst.dev->ifindex,
1296 .flowi4_iif = skb->dev->ifindex,
1297 .flowi4_mark = skb->mark,
1298 };
1299
1300 rcu_read_lock();
1301 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1302 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1303 else
1304 src = inet_select_addr(rt->dst.dev,
1305 rt_nexthop(rt, iph->daddr),
1306 RT_SCOPE_UNIVERSE);
1307 rcu_read_unlock();
1308 }
1309 memcpy(addr, &src, 4);
1310 }
1311
1312 #ifdef CONFIG_IP_ROUTE_CLASSID
set_class_tag(struct rtable * rt,u32 tag)1313 static void set_class_tag(struct rtable *rt, u32 tag)
1314 {
1315 if (!(rt->dst.tclassid & 0xFFFF))
1316 rt->dst.tclassid |= tag & 0xFFFF;
1317 if (!(rt->dst.tclassid & 0xFFFF0000))
1318 rt->dst.tclassid |= tag & 0xFFFF0000;
1319 }
1320 #endif
1321
ipv4_default_advmss(const struct dst_entry * dst)1322 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1323 {
1324 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1325 unsigned int advmss;
1326 struct net *net;
1327
1328 rcu_read_lock();
1329 net = dev_net_rcu(dst_dev(dst));
1330 advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1331 net->ipv4.ip_rt_min_advmss);
1332 rcu_read_unlock();
1333
1334 return min(advmss, IPV4_MAX_PMTU - header_size);
1335 }
1336
ipv4_mtu(const struct dst_entry * dst)1337 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1338 {
1339 return ip_dst_mtu_maybe_forward(dst, false);
1340 }
1341 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1342
ip_del_fnhe(struct fib_nh_common * nhc,__be32 daddr)1343 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1344 {
1345 struct fnhe_hash_bucket *hash;
1346 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1347 u32 hval = fnhe_hashfun(daddr);
1348
1349 spin_lock_bh(&fnhe_lock);
1350
1351 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1352 lockdep_is_held(&fnhe_lock));
1353 hash += hval;
1354
1355 fnhe_p = &hash->chain;
1356 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1357 while (fnhe) {
1358 if (fnhe->fnhe_daddr == daddr) {
1359 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1360 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1361 /* set fnhe_daddr to 0 to ensure it won't bind with
1362 * new dsts in rt_bind_exception().
1363 */
1364 fnhe->fnhe_daddr = 0;
1365 fnhe_flush_routes(fnhe);
1366 kfree_rcu(fnhe, rcu);
1367 break;
1368 }
1369 fnhe_p = &fnhe->fnhe_next;
1370 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1371 lockdep_is_held(&fnhe_lock));
1372 }
1373
1374 spin_unlock_bh(&fnhe_lock);
1375 }
1376
find_exception(struct fib_nh_common * nhc,__be32 daddr)1377 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1378 __be32 daddr)
1379 {
1380 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1381 struct fib_nh_exception *fnhe;
1382 u32 hval;
1383
1384 if (!hash)
1385 return NULL;
1386
1387 hval = fnhe_hashfun(daddr);
1388
1389 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1390 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1391 if (fnhe->fnhe_daddr == daddr) {
1392 if (fnhe->fnhe_expires &&
1393 time_after(jiffies, fnhe->fnhe_expires)) {
1394 ip_del_fnhe(nhc, daddr);
1395 break;
1396 }
1397 return fnhe;
1398 }
1399 }
1400 return NULL;
1401 }
1402
1403 /* MTU selection:
1404 * 1. mtu on route is locked - use it
1405 * 2. mtu from nexthop exception
1406 * 3. mtu from egress device
1407 */
1408
ip_mtu_from_fib_result(struct fib_result * res,__be32 daddr)1409 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1410 {
1411 struct fib_nh_common *nhc = res->nhc;
1412 struct net_device *dev = nhc->nhc_dev;
1413 struct fib_info *fi = res->fi;
1414 u32 mtu = 0;
1415
1416 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1417 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1418 mtu = fi->fib_mtu;
1419
1420 if (likely(!mtu)) {
1421 struct fib_nh_exception *fnhe;
1422
1423 fnhe = find_exception(nhc, daddr);
1424 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1425 mtu = fnhe->fnhe_pmtu;
1426 }
1427
1428 if (likely(!mtu))
1429 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1430
1431 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1432 }
1433
rt_bind_exception(struct rtable * rt,struct fib_nh_exception * fnhe,__be32 daddr,const bool do_cache)1434 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1435 __be32 daddr, const bool do_cache)
1436 {
1437 bool ret = false;
1438
1439 spin_lock_bh(&fnhe_lock);
1440
1441 if (daddr == fnhe->fnhe_daddr) {
1442 struct rtable __rcu **porig;
1443 struct rtable *orig;
1444 int genid = fnhe_genid(dev_net(rt->dst.dev));
1445
1446 if (rt_is_input_route(rt))
1447 porig = &fnhe->fnhe_rth_input;
1448 else
1449 porig = &fnhe->fnhe_rth_output;
1450 orig = rcu_dereference(*porig);
1451
1452 if (fnhe->fnhe_genid != genid) {
1453 fnhe->fnhe_genid = genid;
1454 fnhe->fnhe_gw = 0;
1455 fnhe->fnhe_pmtu = 0;
1456 fnhe->fnhe_expires = 0;
1457 fnhe->fnhe_mtu_locked = false;
1458 fnhe_flush_routes(fnhe);
1459 orig = NULL;
1460 }
1461 fill_route_from_fnhe(rt, fnhe);
1462 if (!rt->rt_gw4) {
1463 rt->rt_gw4 = daddr;
1464 rt->rt_gw_family = AF_INET;
1465 }
1466
1467 if (do_cache) {
1468 dst_hold(&rt->dst);
1469 rcu_assign_pointer(*porig, rt);
1470 if (orig) {
1471 dst_dev_put(&orig->dst);
1472 dst_release(&orig->dst);
1473 }
1474 ret = true;
1475 }
1476
1477 fnhe->fnhe_stamp = jiffies;
1478 }
1479 spin_unlock_bh(&fnhe_lock);
1480
1481 return ret;
1482 }
1483
rt_cache_route(struct fib_nh_common * nhc,struct rtable * rt)1484 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1485 {
1486 struct rtable *orig, *prev, **p;
1487 bool ret = true;
1488
1489 if (rt_is_input_route(rt)) {
1490 p = (struct rtable **)&nhc->nhc_rth_input;
1491 } else {
1492 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1493 }
1494 orig = *p;
1495
1496 /* hold dst before doing cmpxchg() to avoid race condition
1497 * on this dst
1498 */
1499 dst_hold(&rt->dst);
1500 prev = cmpxchg(p, orig, rt);
1501 if (prev == orig) {
1502 if (orig) {
1503 rt_add_uncached_list(orig);
1504 dst_release(&orig->dst);
1505 }
1506 } else {
1507 dst_release(&rt->dst);
1508 ret = false;
1509 }
1510
1511 return ret;
1512 }
1513
1514 struct uncached_list {
1515 spinlock_t lock;
1516 struct list_head head;
1517 };
1518
1519 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1520
rt_add_uncached_list(struct rtable * rt)1521 void rt_add_uncached_list(struct rtable *rt)
1522 {
1523 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1524
1525 rt->dst.rt_uncached_list = ul;
1526
1527 spin_lock_bh(&ul->lock);
1528 list_add_tail(&rt->dst.rt_uncached, &ul->head);
1529 spin_unlock_bh(&ul->lock);
1530 }
1531
rt_del_uncached_list(struct rtable * rt)1532 void rt_del_uncached_list(struct rtable *rt)
1533 {
1534 if (!list_empty(&rt->dst.rt_uncached)) {
1535 struct uncached_list *ul = rt->dst.rt_uncached_list;
1536
1537 spin_lock_bh(&ul->lock);
1538 list_del_init(&rt->dst.rt_uncached);
1539 spin_unlock_bh(&ul->lock);
1540 }
1541 }
1542
ipv4_dst_destroy(struct dst_entry * dst)1543 static void ipv4_dst_destroy(struct dst_entry *dst)
1544 {
1545 ip_dst_metrics_put(dst);
1546 rt_del_uncached_list(dst_rtable(dst));
1547 }
1548
rt_flush_dev(struct net_device * dev)1549 void rt_flush_dev(struct net_device *dev)
1550 {
1551 struct rtable *rt, *safe;
1552 int cpu;
1553
1554 for_each_possible_cpu(cpu) {
1555 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1556
1557 if (list_empty(&ul->head))
1558 continue;
1559
1560 spin_lock_bh(&ul->lock);
1561 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1562 if (rt->dst.dev != dev)
1563 continue;
1564 rt->dst.dev = blackhole_netdev;
1565 netdev_ref_replace(dev, blackhole_netdev,
1566 &rt->dst.dev_tracker, GFP_ATOMIC);
1567 list_del_init(&rt->dst.rt_uncached);
1568 }
1569 spin_unlock_bh(&ul->lock);
1570 }
1571 }
1572
rt_cache_valid(const struct rtable * rt)1573 static bool rt_cache_valid(const struct rtable *rt)
1574 {
1575 return rt &&
1576 READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK &&
1577 !rt_is_expired(rt);
1578 }
1579
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)1580 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1581 const struct fib_result *res,
1582 struct fib_nh_exception *fnhe,
1583 struct fib_info *fi, u16 type, u32 itag,
1584 const bool do_cache)
1585 {
1586 bool cached = false;
1587
1588 if (fi) {
1589 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1590
1591 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1592 rt->rt_uses_gateway = 1;
1593 rt->rt_gw_family = nhc->nhc_gw_family;
1594 /* only INET and INET6 are supported */
1595 if (likely(nhc->nhc_gw_family == AF_INET))
1596 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1597 else
1598 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1599 }
1600
1601 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1602
1603 #ifdef CONFIG_IP_ROUTE_CLASSID
1604 if (nhc->nhc_family == AF_INET) {
1605 struct fib_nh *nh;
1606
1607 nh = container_of(nhc, struct fib_nh, nh_common);
1608 rt->dst.tclassid = nh->nh_tclassid;
1609 }
1610 #endif
1611 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1612 if (unlikely(fnhe))
1613 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1614 else if (do_cache)
1615 cached = rt_cache_route(nhc, rt);
1616 if (unlikely(!cached)) {
1617 /* Routes we intend to cache in nexthop exception or
1618 * FIB nexthop have the DST_NOCACHE bit clear.
1619 * However, if we are unsuccessful at storing this
1620 * route into the cache we really need to set it.
1621 */
1622 if (!rt->rt_gw4) {
1623 rt->rt_gw_family = AF_INET;
1624 rt->rt_gw4 = daddr;
1625 }
1626 rt_add_uncached_list(rt);
1627 }
1628 } else
1629 rt_add_uncached_list(rt);
1630
1631 #ifdef CONFIG_IP_ROUTE_CLASSID
1632 #ifdef CONFIG_IP_MULTIPLE_TABLES
1633 set_class_tag(rt, res->tclassid);
1634 #endif
1635 set_class_tag(rt, itag);
1636 #endif
1637 }
1638
rt_dst_alloc(struct net_device * dev,unsigned int flags,u16 type,bool noxfrm)1639 struct rtable *rt_dst_alloc(struct net_device *dev,
1640 unsigned int flags, u16 type,
1641 bool noxfrm)
1642 {
1643 struct rtable *rt;
1644
1645 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1646 (noxfrm ? DST_NOXFRM : 0));
1647
1648 if (rt) {
1649 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1650 rt->rt_flags = flags;
1651 rt->rt_type = type;
1652 rt->rt_is_input = 0;
1653 rt->rt_iif = 0;
1654 rt->rt_pmtu = 0;
1655 rt->rt_mtu_locked = 0;
1656 rt->rt_uses_gateway = 0;
1657 rt->rt_gw_family = 0;
1658 rt->rt_gw4 = 0;
1659
1660 rt->dst.output = ip_output;
1661 if (flags & RTCF_LOCAL)
1662 rt->dst.input = ip_local_deliver;
1663 }
1664
1665 return rt;
1666 }
1667 EXPORT_SYMBOL(rt_dst_alloc);
1668
rt_dst_clone(struct net_device * dev,struct rtable * rt)1669 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1670 {
1671 struct rtable *new_rt;
1672
1673 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1674 rt->dst.flags);
1675
1676 if (new_rt) {
1677 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1678 new_rt->rt_flags = rt->rt_flags;
1679 new_rt->rt_type = rt->rt_type;
1680 new_rt->rt_is_input = rt->rt_is_input;
1681 new_rt->rt_iif = rt->rt_iif;
1682 new_rt->rt_pmtu = rt->rt_pmtu;
1683 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1684 new_rt->rt_gw_family = rt->rt_gw_family;
1685 if (rt->rt_gw_family == AF_INET)
1686 new_rt->rt_gw4 = rt->rt_gw4;
1687 else if (rt->rt_gw_family == AF_INET6)
1688 new_rt->rt_gw6 = rt->rt_gw6;
1689
1690 new_rt->dst.input = READ_ONCE(rt->dst.input);
1691 new_rt->dst.output = READ_ONCE(rt->dst.output);
1692 new_rt->dst.error = rt->dst.error;
1693 new_rt->dst.lastuse = jiffies;
1694 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1695 }
1696 return new_rt;
1697 }
1698 EXPORT_SYMBOL(rt_dst_clone);
1699
1700 /* called in rcu_read_lock() section */
1701 enum skb_drop_reason
ip_mc_validate_source(struct sk_buff * skb,__be32 daddr,__be32 saddr,dscp_t dscp,struct net_device * dev,struct in_device * in_dev,u32 * itag)1702 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1703 dscp_t dscp, struct net_device *dev,
1704 struct in_device *in_dev, u32 *itag)
1705 {
1706 enum skb_drop_reason reason;
1707
1708 /* Primary sanity checks. */
1709 if (!in_dev)
1710 return SKB_DROP_REASON_NOT_SPECIFIED;
1711
1712 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1713 return SKB_DROP_REASON_IP_INVALID_SOURCE;
1714
1715 if (skb->protocol != htons(ETH_P_IP))
1716 return SKB_DROP_REASON_INVALID_PROTO;
1717
1718 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1719 return SKB_DROP_REASON_IP_LOCALNET;
1720
1721 if (ipv4_is_zeronet(saddr)) {
1722 if (!ipv4_is_local_multicast(daddr) &&
1723 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1724 return SKB_DROP_REASON_IP_INVALID_SOURCE;
1725 } else {
1726 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1727 dev, in_dev, itag);
1728 if (reason)
1729 return reason;
1730 }
1731 return SKB_NOT_DROPPED_YET;
1732 }
1733
1734 /* called in rcu_read_lock() section */
1735 static enum skb_drop_reason
ip_route_input_mc(struct sk_buff * skb,__be32 daddr,__be32 saddr,dscp_t dscp,struct net_device * dev,int our)1736 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1737 dscp_t dscp, struct net_device *dev, int our)
1738 {
1739 struct in_device *in_dev = __in_dev_get_rcu(dev);
1740 unsigned int flags = RTCF_MULTICAST;
1741 enum skb_drop_reason reason;
1742 struct rtable *rth;
1743 u32 itag = 0;
1744
1745 reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1746 &itag);
1747 if (reason)
1748 return reason;
1749
1750 if (our)
1751 flags |= RTCF_LOCAL;
1752
1753 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1754 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1755
1756 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1757 false);
1758 if (!rth)
1759 return SKB_DROP_REASON_NOMEM;
1760
1761 #ifdef CONFIG_IP_ROUTE_CLASSID
1762 rth->dst.tclassid = itag;
1763 #endif
1764 rth->dst.output = ip_rt_bug;
1765 rth->rt_is_input= 1;
1766
1767 #ifdef CONFIG_IP_MROUTE
1768 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1769 rth->dst.input = ip_mr_input;
1770 #endif
1771 RT_CACHE_STAT_INC(in_slow_mc);
1772
1773 skb_dst_drop(skb);
1774 skb_dst_set(skb, &rth->dst);
1775 return SKB_NOT_DROPPED_YET;
1776 }
1777
1778
ip_handle_martian_source(struct net_device * dev,struct in_device * in_dev,struct sk_buff * skb,__be32 daddr,__be32 saddr)1779 static void ip_handle_martian_source(struct net_device *dev,
1780 struct in_device *in_dev,
1781 struct sk_buff *skb,
1782 __be32 daddr,
1783 __be32 saddr)
1784 {
1785 RT_CACHE_STAT_INC(in_martian_src);
1786 #ifdef CONFIG_IP_ROUTE_VERBOSE
1787 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1788 /*
1789 * RFC1812 recommendation, if source is martian,
1790 * the only hint is MAC header.
1791 */
1792 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1793 &daddr, &saddr, dev->name);
1794 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1795 print_hex_dump(KERN_WARNING, "ll header: ",
1796 DUMP_PREFIX_OFFSET, 16, 1,
1797 skb_mac_header(skb),
1798 dev->hard_header_len, false);
1799 }
1800 }
1801 #endif
1802 }
1803
1804 /* called in rcu_read_lock() section */
1805 static enum skb_drop_reason
__mkroute_input(struct sk_buff * skb,const struct fib_result * res,struct in_device * in_dev,__be32 daddr,__be32 saddr,dscp_t dscp)1806 __mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1807 struct in_device *in_dev, __be32 daddr,
1808 __be32 saddr, dscp_t dscp)
1809 {
1810 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1811 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1812 struct net_device *dev = nhc->nhc_dev;
1813 struct fib_nh_exception *fnhe;
1814 struct rtable *rth;
1815 int err;
1816 struct in_device *out_dev;
1817 bool do_cache;
1818 u32 itag = 0;
1819
1820 /* get a working reference to the output device */
1821 out_dev = __in_dev_get_rcu(dev);
1822 if (!out_dev) {
1823 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1824 return reason;
1825 }
1826
1827 err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1828 in_dev->dev, in_dev, &itag);
1829 if (err < 0) {
1830 reason = -err;
1831 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1832 saddr);
1833
1834 goto cleanup;
1835 }
1836
1837 do_cache = res->fi && !itag;
1838 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1839 skb->protocol == htons(ETH_P_IP)) {
1840 __be32 gw;
1841
1842 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1843 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1844 inet_addr_onlink(out_dev, saddr, gw))
1845 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1846 }
1847
1848 if (skb->protocol != htons(ETH_P_IP)) {
1849 /* Not IP (i.e. ARP). Do not create route, if it is
1850 * invalid for proxy arp. DNAT routes are always valid.
1851 *
1852 * Proxy arp feature have been extended to allow, ARP
1853 * replies back to the same interface, to support
1854 * Private VLAN switch technologies. See arp.c.
1855 */
1856 if (out_dev == in_dev &&
1857 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1858 reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1859 goto cleanup;
1860 }
1861 }
1862
1863 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1864 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1865
1866 fnhe = find_exception(nhc, daddr);
1867 if (do_cache) {
1868 if (fnhe)
1869 rth = rcu_dereference(fnhe->fnhe_rth_input);
1870 else
1871 rth = rcu_dereference(nhc->nhc_rth_input);
1872 if (rt_cache_valid(rth)) {
1873 skb_dst_set_noref(skb, &rth->dst);
1874 goto out;
1875 }
1876 }
1877
1878 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1879 IN_DEV_ORCONF(out_dev, NOXFRM));
1880 if (!rth) {
1881 reason = SKB_DROP_REASON_NOMEM;
1882 goto cleanup;
1883 }
1884
1885 rth->rt_is_input = 1;
1886 RT_CACHE_STAT_INC(in_slow_tot);
1887
1888 rth->dst.input = ip_forward;
1889
1890 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1891 do_cache);
1892 lwtunnel_set_redirect(&rth->dst);
1893 skb_dst_set(skb, &rth->dst);
1894 out:
1895 reason = SKB_NOT_DROPPED_YET;
1896 cleanup:
1897 return reason;
1898 }
1899
1900 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1901 /* To make ICMP packets follow the right flow, the multipath hash is
1902 * calculated from the inner IP addresses.
1903 */
ip_multipath_l3_keys(const struct sk_buff * skb,struct flow_keys * hash_keys)1904 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1905 struct flow_keys *hash_keys)
1906 {
1907 const struct iphdr *outer_iph = ip_hdr(skb);
1908 const struct iphdr *key_iph = outer_iph;
1909 const struct iphdr *inner_iph;
1910 const struct icmphdr *icmph;
1911 struct iphdr _inner_iph;
1912 struct icmphdr _icmph;
1913
1914 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1915 goto out;
1916
1917 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1918 goto out;
1919
1920 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1921 &_icmph);
1922 if (!icmph)
1923 goto out;
1924
1925 if (!icmp_is_err(icmph->type))
1926 goto out;
1927
1928 inner_iph = skb_header_pointer(skb,
1929 outer_iph->ihl * 4 + sizeof(_icmph),
1930 sizeof(_inner_iph), &_inner_iph);
1931 if (!inner_iph)
1932 goto out;
1933
1934 key_iph = inner_iph;
1935 out:
1936 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1937 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1938 }
1939
fib_multipath_custom_hash_outer(const struct net * net,const struct sk_buff * skb,bool * p_has_inner)1940 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1941 const struct sk_buff *skb,
1942 bool *p_has_inner)
1943 {
1944 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1945 struct flow_keys keys, hash_keys;
1946
1947 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1948 return 0;
1949
1950 memset(&hash_keys, 0, sizeof(hash_keys));
1951 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1952
1953 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1954 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1955 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1956 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1957 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1958 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1959 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1960 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1961 hash_keys.ports.src = keys.ports.src;
1962 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1963 hash_keys.ports.dst = keys.ports.dst;
1964
1965 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1966 return fib_multipath_hash_from_keys(net, &hash_keys);
1967 }
1968
fib_multipath_custom_hash_inner(const struct net * net,const struct sk_buff * skb,bool has_inner)1969 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1970 const struct sk_buff *skb,
1971 bool has_inner)
1972 {
1973 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1974 struct flow_keys keys, hash_keys;
1975
1976 /* We assume the packet carries an encapsulation, but if none was
1977 * encountered during dissection of the outer flow, then there is no
1978 * point in calling the flow dissector again.
1979 */
1980 if (!has_inner)
1981 return 0;
1982
1983 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1984 return 0;
1985
1986 memset(&hash_keys, 0, sizeof(hash_keys));
1987 skb_flow_dissect_flow_keys(skb, &keys, 0);
1988
1989 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1990 return 0;
1991
1992 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1993 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1994 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1995 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1996 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1997 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1998 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1999 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2000 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2001 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2002 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2003 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2004 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2005 hash_keys.tags.flow_label = keys.tags.flow_label;
2006 }
2007
2008 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2009 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2010 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2011 hash_keys.ports.src = keys.ports.src;
2012 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2013 hash_keys.ports.dst = keys.ports.dst;
2014
2015 return fib_multipath_hash_from_keys(net, &hash_keys);
2016 }
2017
fib_multipath_custom_hash_skb(const struct net * net,const struct sk_buff * skb)2018 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2019 const struct sk_buff *skb)
2020 {
2021 u32 mhash, mhash_inner;
2022 bool has_inner = true;
2023
2024 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2025 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2026
2027 return jhash_2words(mhash, mhash_inner, 0);
2028 }
2029
fib_multipath_custom_hash_fl4(const struct net * net,const struct flowi4 * fl4)2030 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2031 const struct flowi4 *fl4)
2032 {
2033 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2034 struct flow_keys hash_keys;
2035
2036 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2037 return 0;
2038
2039 memset(&hash_keys, 0, sizeof(hash_keys));
2040 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2041 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2042 hash_keys.addrs.v4addrs.src = fl4->saddr;
2043 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2044 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2045 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2046 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2047 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2048 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2049 hash_keys.ports.src = (__force __be16)get_random_u16();
2050 else
2051 hash_keys.ports.src = fl4->fl4_sport;
2052 }
2053 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2054 hash_keys.ports.dst = fl4->fl4_dport;
2055
2056 return fib_multipath_hash_from_keys(net, &hash_keys);
2057 }
2058
2059 /* 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)2060 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2061 const struct sk_buff *skb, struct flow_keys *flkeys)
2062 {
2063 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2064 struct flow_keys hash_keys;
2065 u32 mhash = 0;
2066
2067 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2068 case 0:
2069 memset(&hash_keys, 0, sizeof(hash_keys));
2070 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2071 if (skb) {
2072 ip_multipath_l3_keys(skb, &hash_keys);
2073 } else {
2074 hash_keys.addrs.v4addrs.src = fl4->saddr;
2075 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2076 }
2077 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2078 break;
2079 case 1:
2080 /* skb is currently provided only when forwarding */
2081 if (skb) {
2082 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2083 struct flow_keys keys;
2084
2085 /* short-circuit if we already have L4 hash present */
2086 if (skb->l4_hash)
2087 return skb_get_hash_raw(skb) >> 1;
2088
2089 memset(&hash_keys, 0, sizeof(hash_keys));
2090
2091 if (!flkeys) {
2092 skb_flow_dissect_flow_keys(skb, &keys, flag);
2093 flkeys = &keys;
2094 }
2095
2096 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2097 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2098 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2099 hash_keys.ports.src = flkeys->ports.src;
2100 hash_keys.ports.dst = flkeys->ports.dst;
2101 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2102 } else {
2103 memset(&hash_keys, 0, sizeof(hash_keys));
2104 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2105 hash_keys.addrs.v4addrs.src = fl4->saddr;
2106 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2107 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2108 hash_keys.ports.src = (__force __be16)get_random_u16();
2109 else
2110 hash_keys.ports.src = fl4->fl4_sport;
2111 hash_keys.ports.dst = fl4->fl4_dport;
2112 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2113 }
2114 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2115 break;
2116 case 2:
2117 memset(&hash_keys, 0, sizeof(hash_keys));
2118 /* skb is currently provided only when forwarding */
2119 if (skb) {
2120 struct flow_keys keys;
2121
2122 skb_flow_dissect_flow_keys(skb, &keys, 0);
2123 /* Inner can be v4 or v6 */
2124 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2125 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2126 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2127 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2128 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2129 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2130 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2131 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2132 hash_keys.tags.flow_label = keys.tags.flow_label;
2133 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2134 } else {
2135 /* Same as case 0 */
2136 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2137 ip_multipath_l3_keys(skb, &hash_keys);
2138 }
2139 } else {
2140 /* Same as case 0 */
2141 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2142 hash_keys.addrs.v4addrs.src = fl4->saddr;
2143 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2144 }
2145 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2146 break;
2147 case 3:
2148 if (skb)
2149 mhash = fib_multipath_custom_hash_skb(net, skb);
2150 else
2151 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2152 break;
2153 }
2154
2155 if (multipath_hash)
2156 mhash = jhash_2words(mhash, multipath_hash, 0);
2157
2158 return mhash >> 1;
2159 }
2160 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2161
2162 static enum skb_drop_reason
ip_mkroute_input(struct sk_buff * skb,struct fib_result * res,struct in_device * in_dev,__be32 daddr,__be32 saddr,dscp_t dscp,struct flow_keys * hkeys)2163 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2164 struct in_device *in_dev, __be32 daddr,
2165 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2166 {
2167 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2168 if (res->fi && fib_info_num_path(res->fi) > 1) {
2169 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2170
2171 fib_select_multipath(res, h, NULL);
2172 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2173 }
2174 #endif
2175
2176 /* create a routing cache entry */
2177 return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2178 }
2179
2180 /* Implements all the saddr-related checks as ip_route_input_slow(),
2181 * assuming daddr is valid and the destination is not a local broadcast one.
2182 * Uses the provided hint instead of performing a route lookup.
2183 */
2184 enum skb_drop_reason
ip_route_use_hint(struct sk_buff * skb,__be32 daddr,__be32 saddr,dscp_t dscp,struct net_device * dev,const struct sk_buff * hint)2185 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2186 dscp_t dscp, struct net_device *dev,
2187 const struct sk_buff *hint)
2188 {
2189 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2190 struct in_device *in_dev = __in_dev_get_rcu(dev);
2191 struct rtable *rt = skb_rtable(hint);
2192 struct net *net = dev_net(dev);
2193 u32 tag = 0;
2194
2195 if (!in_dev)
2196 return reason;
2197
2198 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2199 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2200 goto martian_source;
2201 }
2202
2203 if (ipv4_is_zeronet(saddr)) {
2204 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2205 goto martian_source;
2206 }
2207
2208 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2209 reason = SKB_DROP_REASON_IP_LOCALNET;
2210 goto martian_source;
2211 }
2212
2213 if (rt->rt_type != RTN_LOCAL)
2214 goto skip_validate_source;
2215
2216 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2217 in_dev, &tag);
2218 if (reason)
2219 goto martian_source;
2220
2221 skip_validate_source:
2222 skb_dst_copy(skb, hint);
2223 return SKB_NOT_DROPPED_YET;
2224
2225 martian_source:
2226 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2227 return reason;
2228 }
2229
2230 /* get device for dst_alloc with local routes */
ip_rt_get_dev(struct net * net,const struct fib_result * res)2231 static struct net_device *ip_rt_get_dev(struct net *net,
2232 const struct fib_result *res)
2233 {
2234 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2235 struct net_device *dev = NULL;
2236
2237 if (nhc)
2238 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2239
2240 return dev ? : net->loopback_dev;
2241 }
2242
2243 /*
2244 * NOTE. We drop all the packets that has local source
2245 * addresses, because every properly looped back packet
2246 * must have correct destination already attached by output routine.
2247 * Changes in the enforced policies must be applied also to
2248 * ip_route_use_hint().
2249 *
2250 * Such approach solves two big problems:
2251 * 1. Not simplex devices are handled properly.
2252 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2253 * called with rcu_read_lock()
2254 */
2255
2256 static enum skb_drop_reason
ip_route_input_slow(struct sk_buff * skb,__be32 daddr,__be32 saddr,dscp_t dscp,struct net_device * dev,struct fib_result * res)2257 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2258 dscp_t dscp, struct net_device *dev,
2259 struct fib_result *res)
2260 {
2261 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2262 struct in_device *in_dev = __in_dev_get_rcu(dev);
2263 struct flow_keys *flkeys = NULL, _flkeys;
2264 struct net *net = dev_net(dev);
2265 struct ip_tunnel_info *tun_info;
2266 int err = -EINVAL;
2267 unsigned int flags = 0;
2268 u32 itag = 0;
2269 struct rtable *rth;
2270 struct flowi4 fl4;
2271 bool do_cache = true;
2272
2273 /* IP on this device is disabled. */
2274
2275 if (!in_dev)
2276 goto out;
2277
2278 /* Check for the most weird martians, which can be not detected
2279 * by fib_lookup.
2280 */
2281
2282 tun_info = skb_tunnel_info(skb);
2283 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2284 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2285 else
2286 fl4.flowi4_tun_key.tun_id = 0;
2287 skb_dst_drop(skb);
2288
2289 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2290 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2291 goto martian_source;
2292 }
2293
2294 res->fi = NULL;
2295 res->table = NULL;
2296 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2297 goto brd_input;
2298
2299 /* Accept zero addresses only to limited broadcast;
2300 * I even do not know to fix it or not. Waiting for complains :-)
2301 */
2302 if (ipv4_is_zeronet(saddr)) {
2303 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2304 goto martian_source;
2305 }
2306
2307 if (ipv4_is_zeronet(daddr)) {
2308 reason = SKB_DROP_REASON_IP_INVALID_DEST;
2309 goto martian_destination;
2310 }
2311
2312 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2313 * and call it once if daddr or/and saddr are loopback addresses
2314 */
2315 if (ipv4_is_loopback(daddr)) {
2316 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2317 reason = SKB_DROP_REASON_IP_LOCALNET;
2318 goto martian_destination;
2319 }
2320 } else if (ipv4_is_loopback(saddr)) {
2321 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2322 reason = SKB_DROP_REASON_IP_LOCALNET;
2323 goto martian_source;
2324 }
2325 }
2326
2327 /*
2328 * Now we are ready to route packet.
2329 */
2330 fl4.flowi4_l3mdev = 0;
2331 fl4.flowi4_oif = 0;
2332 fl4.flowi4_iif = dev->ifindex;
2333 fl4.flowi4_mark = skb->mark;
2334 fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
2335 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2336 fl4.flowi4_flags = 0;
2337 fl4.daddr = daddr;
2338 fl4.saddr = saddr;
2339 fl4.flowi4_uid = sock_net_uid(net, NULL);
2340 fl4.flowi4_multipath_hash = 0;
2341
2342 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2343 flkeys = &_flkeys;
2344 } else {
2345 fl4.flowi4_proto = 0;
2346 fl4.fl4_sport = 0;
2347 fl4.fl4_dport = 0;
2348 }
2349
2350 err = fib_lookup(net, &fl4, res, 0);
2351 if (err != 0) {
2352 if (!IN_DEV_FORWARD(in_dev))
2353 err = -EHOSTUNREACH;
2354 goto no_route;
2355 }
2356
2357 if (res->type == RTN_BROADCAST) {
2358 if (IN_DEV_BFORWARD(in_dev))
2359 goto make_route;
2360 /* not do cache if bc_forwarding is enabled */
2361 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2362 do_cache = false;
2363 goto brd_input;
2364 }
2365
2366 err = -EINVAL;
2367 if (res->type == RTN_LOCAL) {
2368 reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2369 0, dev, in_dev, &itag);
2370 if (reason)
2371 goto martian_source;
2372 goto local_input;
2373 }
2374
2375 if (!IN_DEV_FORWARD(in_dev)) {
2376 err = -EHOSTUNREACH;
2377 goto no_route;
2378 }
2379 if (res->type != RTN_UNICAST) {
2380 reason = SKB_DROP_REASON_IP_INVALID_DEST;
2381 goto martian_destination;
2382 }
2383
2384 make_route:
2385 reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2386 flkeys);
2387
2388 out:
2389 return reason;
2390
2391 brd_input:
2392 if (skb->protocol != htons(ETH_P_IP)) {
2393 reason = SKB_DROP_REASON_INVALID_PROTO;
2394 goto out;
2395 }
2396
2397 if (!ipv4_is_zeronet(saddr)) {
2398 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2399 dev, in_dev, &itag);
2400 if (reason)
2401 goto martian_source;
2402 }
2403 flags |= RTCF_BROADCAST;
2404 res->type = RTN_BROADCAST;
2405 RT_CACHE_STAT_INC(in_brd);
2406
2407 local_input:
2408 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2409 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2410
2411 do_cache &= res->fi && !itag;
2412 if (do_cache) {
2413 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2414
2415 rth = rcu_dereference(nhc->nhc_rth_input);
2416 if (rt_cache_valid(rth)) {
2417 skb_dst_set_noref(skb, &rth->dst);
2418 reason = SKB_NOT_DROPPED_YET;
2419 goto out;
2420 }
2421 }
2422
2423 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2424 flags | RTCF_LOCAL, res->type, false);
2425 if (!rth)
2426 goto e_nobufs;
2427
2428 rth->dst.output= ip_rt_bug;
2429 #ifdef CONFIG_IP_ROUTE_CLASSID
2430 rth->dst.tclassid = itag;
2431 #endif
2432 rth->rt_is_input = 1;
2433
2434 RT_CACHE_STAT_INC(in_slow_tot);
2435 if (res->type == RTN_UNREACHABLE) {
2436 rth->dst.input= ip_error;
2437 rth->dst.error= -err;
2438 rth->rt_flags &= ~RTCF_LOCAL;
2439 }
2440
2441 if (do_cache) {
2442 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2443
2444 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2445 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2446 WARN_ON(rth->dst.input == lwtunnel_input);
2447 rth->dst.lwtstate->orig_input = rth->dst.input;
2448 rth->dst.input = lwtunnel_input;
2449 }
2450
2451 if (unlikely(!rt_cache_route(nhc, rth)))
2452 rt_add_uncached_list(rth);
2453 }
2454 skb_dst_set(skb, &rth->dst);
2455 reason = SKB_NOT_DROPPED_YET;
2456 goto out;
2457
2458 no_route:
2459 RT_CACHE_STAT_INC(in_no_route);
2460 res->type = RTN_UNREACHABLE;
2461 res->fi = NULL;
2462 res->table = NULL;
2463 goto local_input;
2464
2465 /*
2466 * Do not cache martian addresses: they should be logged (RFC1812)
2467 */
2468 martian_destination:
2469 RT_CACHE_STAT_INC(in_martian_dst);
2470 #ifdef CONFIG_IP_ROUTE_VERBOSE
2471 if (IN_DEV_LOG_MARTIANS(in_dev))
2472 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2473 &daddr, &saddr, dev->name);
2474 #endif
2475 goto out;
2476
2477 e_nobufs:
2478 reason = SKB_DROP_REASON_NOMEM;
2479 goto out;
2480
2481 martian_source:
2482 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2483 goto out;
2484 }
2485
2486 /* called with rcu_read_lock held */
2487 static enum skb_drop_reason
ip_route_input_rcu(struct sk_buff * skb,__be32 daddr,__be32 saddr,dscp_t dscp,struct net_device * dev,struct fib_result * res)2488 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2489 dscp_t dscp, struct net_device *dev,
2490 struct fib_result *res)
2491 {
2492 /* Multicast recognition logic is moved from route cache to here.
2493 * The problem was that too many Ethernet cards have broken/missing
2494 * hardware multicast filters :-( As result the host on multicasting
2495 * network acquires a lot of useless route cache entries, sort of
2496 * SDR messages from all the world. Now we try to get rid of them.
2497 * Really, provided software IP multicast filter is organized
2498 * reasonably (at least, hashed), it does not result in a slowdown
2499 * comparing with route cache reject entries.
2500 * Note, that multicast routers are not affected, because
2501 * route cache entry is created eventually.
2502 */
2503 if (ipv4_is_multicast(daddr)) {
2504 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2505 struct in_device *in_dev = __in_dev_get_rcu(dev);
2506 int our = 0;
2507
2508 if (!in_dev)
2509 return reason;
2510
2511 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2512 ip_hdr(skb)->protocol);
2513
2514 /* check l3 master if no match yet */
2515 if (!our && netif_is_l3_slave(dev)) {
2516 struct in_device *l3_in_dev;
2517
2518 l3_in_dev = __in_dev_get_rcu(skb->dev);
2519 if (l3_in_dev)
2520 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2521 ip_hdr(skb)->protocol);
2522 }
2523
2524 if (our
2525 #ifdef CONFIG_IP_MROUTE
2526 ||
2527 (!ipv4_is_local_multicast(daddr) &&
2528 IN_DEV_MFORWARD(in_dev))
2529 #endif
2530 ) {
2531 reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2532 dev, our);
2533 }
2534 return reason;
2535 }
2536
2537 return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2538 }
2539
ip_route_input_noref(struct sk_buff * skb,__be32 daddr,__be32 saddr,dscp_t dscp,struct net_device * dev)2540 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2541 __be32 saddr, dscp_t dscp,
2542 struct net_device *dev)
2543 {
2544 enum skb_drop_reason reason;
2545 struct fib_result res;
2546
2547 rcu_read_lock();
2548 reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2549 rcu_read_unlock();
2550
2551 return reason;
2552 }
2553 EXPORT_SYMBOL(ip_route_input_noref);
2554
2555 /* 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)2556 static struct rtable *__mkroute_output(const struct fib_result *res,
2557 const struct flowi4 *fl4, int orig_oif,
2558 struct net_device *dev_out,
2559 unsigned int flags)
2560 {
2561 struct fib_info *fi = res->fi;
2562 struct fib_nh_exception *fnhe;
2563 struct in_device *in_dev;
2564 u16 type = res->type;
2565 struct rtable *rth;
2566 bool do_cache;
2567
2568 in_dev = __in_dev_get_rcu(dev_out);
2569 if (!in_dev)
2570 return ERR_PTR(-EINVAL);
2571
2572 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2573 if (ipv4_is_loopback(fl4->saddr) &&
2574 !(dev_out->flags & IFF_LOOPBACK) &&
2575 !netif_is_l3_master(dev_out))
2576 return ERR_PTR(-EINVAL);
2577
2578 if (ipv4_is_lbcast(fl4->daddr))
2579 type = RTN_BROADCAST;
2580 else if (ipv4_is_multicast(fl4->daddr))
2581 type = RTN_MULTICAST;
2582 else if (ipv4_is_zeronet(fl4->daddr))
2583 return ERR_PTR(-EINVAL);
2584
2585 if (dev_out->flags & IFF_LOOPBACK)
2586 flags |= RTCF_LOCAL;
2587
2588 do_cache = true;
2589 if (type == RTN_BROADCAST) {
2590 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2591 } else if (type == RTN_MULTICAST) {
2592 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2593 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2594 fl4->flowi4_proto))
2595 flags &= ~RTCF_LOCAL;
2596 else
2597 do_cache = false;
2598 /* If multicast route do not exist use
2599 * default one, but do not gateway in this case.
2600 * Yes, it is hack.
2601 */
2602 if (fi && res->prefixlen < 4)
2603 fi = NULL;
2604 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2605 (orig_oif != dev_out->ifindex)) {
2606 /* For local routes that require a particular output interface
2607 * we do not want to cache the result. Caching the result
2608 * causes incorrect behaviour when there are multiple source
2609 * addresses on the interface, the end result being that if the
2610 * intended recipient is waiting on that interface for the
2611 * packet he won't receive it because it will be delivered on
2612 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2613 * be set to the loopback interface as well.
2614 */
2615 do_cache = false;
2616 }
2617
2618 fnhe = NULL;
2619 do_cache &= fi != NULL;
2620 if (fi) {
2621 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2622 struct rtable __rcu **prth;
2623
2624 fnhe = find_exception(nhc, fl4->daddr);
2625 if (!do_cache)
2626 goto add;
2627 if (fnhe) {
2628 prth = &fnhe->fnhe_rth_output;
2629 } else {
2630 if (unlikely(fl4->flowi4_flags &
2631 FLOWI_FLAG_KNOWN_NH &&
2632 !(nhc->nhc_gw_family &&
2633 nhc->nhc_scope == RT_SCOPE_LINK))) {
2634 do_cache = false;
2635 goto add;
2636 }
2637 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2638 }
2639 rth = rcu_dereference(*prth);
2640 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2641 return rth;
2642 }
2643
2644 add:
2645 rth = rt_dst_alloc(dev_out, flags, type,
2646 IN_DEV_ORCONF(in_dev, NOXFRM));
2647 if (!rth)
2648 return ERR_PTR(-ENOBUFS);
2649
2650 rth->rt_iif = orig_oif;
2651
2652 RT_CACHE_STAT_INC(out_slow_tot);
2653
2654 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2655 if (flags & RTCF_LOCAL &&
2656 !(dev_out->flags & IFF_LOOPBACK)) {
2657 rth->dst.output = ip_mc_output;
2658 RT_CACHE_STAT_INC(out_slow_mc);
2659 }
2660 #ifdef CONFIG_IP_MROUTE
2661 if (type == RTN_MULTICAST) {
2662 if (IN_DEV_MFORWARD(in_dev) &&
2663 !ipv4_is_local_multicast(fl4->daddr)) {
2664 rth->dst.input = ip_mr_input;
2665 rth->dst.output = ip_mr_output;
2666 }
2667 }
2668 #endif
2669 }
2670
2671 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2672 lwtunnel_set_redirect(&rth->dst);
2673
2674 return rth;
2675 }
2676
2677 /*
2678 * Major route resolver routine.
2679 */
2680
ip_route_output_key_hash(struct net * net,struct flowi4 * fl4,const struct sk_buff * skb)2681 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2682 const struct sk_buff *skb)
2683 {
2684 struct fib_result res = {
2685 .type = RTN_UNSPEC,
2686 .fi = NULL,
2687 .table = NULL,
2688 .tclassid = 0,
2689 };
2690 struct rtable *rth;
2691
2692 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2693 fl4->flowi4_tos &= INET_DSCP_MASK;
2694
2695 rcu_read_lock();
2696 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2697 rcu_read_unlock();
2698
2699 return rth;
2700 }
2701 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2702
ip_route_output_key_hash_rcu(struct net * net,struct flowi4 * fl4,struct fib_result * res,const struct sk_buff * skb)2703 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2704 struct fib_result *res,
2705 const struct sk_buff *skb)
2706 {
2707 struct net_device *dev_out = NULL;
2708 int orig_oif = fl4->flowi4_oif;
2709 unsigned int flags = 0;
2710 struct rtable *rth;
2711 int err;
2712
2713 if (fl4->saddr) {
2714 if (ipv4_is_multicast(fl4->saddr) ||
2715 ipv4_is_lbcast(fl4->saddr)) {
2716 rth = ERR_PTR(-EINVAL);
2717 goto out;
2718 }
2719
2720 rth = ERR_PTR(-ENETUNREACH);
2721
2722 /* I removed check for oif == dev_out->oif here.
2723 * It was wrong for two reasons:
2724 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2725 * is assigned to multiple interfaces.
2726 * 2. Moreover, we are allowed to send packets with saddr
2727 * of another iface. --ANK
2728 */
2729
2730 if (fl4->flowi4_oif == 0 &&
2731 (ipv4_is_multicast(fl4->daddr) ||
2732 ipv4_is_lbcast(fl4->daddr))) {
2733 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2734 dev_out = __ip_dev_find(net, fl4->saddr, false);
2735 if (!dev_out)
2736 goto out;
2737
2738 /* Special hack: user can direct multicasts
2739 * and limited broadcast via necessary interface
2740 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2741 * This hack is not just for fun, it allows
2742 * vic,vat and friends to work.
2743 * They bind socket to loopback, set ttl to zero
2744 * and expect that it will work.
2745 * From the viewpoint of routing cache they are broken,
2746 * because we are not allowed to build multicast path
2747 * with loopback source addr (look, routing cache
2748 * cannot know, that ttl is zero, so that packet
2749 * will not leave this host and route is valid).
2750 * Luckily, this hack is good workaround.
2751 */
2752
2753 fl4->flowi4_oif = dev_out->ifindex;
2754 goto make_route;
2755 }
2756
2757 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2758 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2759 if (!__ip_dev_find(net, fl4->saddr, false))
2760 goto out;
2761 }
2762 }
2763
2764
2765 if (fl4->flowi4_oif) {
2766 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2767 rth = ERR_PTR(-ENODEV);
2768 if (!dev_out)
2769 goto out;
2770
2771 /* RACE: Check return value of inet_select_addr instead. */
2772 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2773 rth = ERR_PTR(-ENETUNREACH);
2774 goto out;
2775 }
2776 if (ipv4_is_local_multicast(fl4->daddr) ||
2777 ipv4_is_lbcast(fl4->daddr) ||
2778 fl4->flowi4_proto == IPPROTO_IGMP) {
2779 if (!fl4->saddr)
2780 fl4->saddr = inet_select_addr(dev_out, 0,
2781 RT_SCOPE_LINK);
2782 goto make_route;
2783 }
2784 if (!fl4->saddr) {
2785 if (ipv4_is_multicast(fl4->daddr))
2786 fl4->saddr = inet_select_addr(dev_out, 0,
2787 fl4->flowi4_scope);
2788 else if (!fl4->daddr)
2789 fl4->saddr = inet_select_addr(dev_out, 0,
2790 RT_SCOPE_HOST);
2791 }
2792 }
2793
2794 if (!fl4->daddr) {
2795 fl4->daddr = fl4->saddr;
2796 if (!fl4->daddr)
2797 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2798 dev_out = net->loopback_dev;
2799 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2800 res->type = RTN_LOCAL;
2801 flags |= RTCF_LOCAL;
2802 goto make_route;
2803 }
2804
2805 err = fib_lookup(net, fl4, res, 0);
2806 if (err) {
2807 res->fi = NULL;
2808 res->table = NULL;
2809 if (fl4->flowi4_oif &&
2810 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2811 /* Apparently, routing tables are wrong. Assume,
2812 * that the destination is on link.
2813 *
2814 * WHY? DW.
2815 * Because we are allowed to send to iface
2816 * even if it has NO routes and NO assigned
2817 * addresses. When oif is specified, routing
2818 * tables are looked up with only one purpose:
2819 * to catch if destination is gatewayed, rather than
2820 * direct. Moreover, if MSG_DONTROUTE is set,
2821 * we send packet, ignoring both routing tables
2822 * and ifaddr state. --ANK
2823 *
2824 *
2825 * We could make it even if oif is unknown,
2826 * likely IPv6, but we do not.
2827 */
2828
2829 if (fl4->saddr == 0)
2830 fl4->saddr = inet_select_addr(dev_out, 0,
2831 RT_SCOPE_LINK);
2832 res->type = RTN_UNICAST;
2833 goto make_route;
2834 }
2835 rth = ERR_PTR(err);
2836 goto out;
2837 }
2838
2839 if (res->type == RTN_LOCAL) {
2840 if (!fl4->saddr) {
2841 if (res->fi->fib_prefsrc)
2842 fl4->saddr = res->fi->fib_prefsrc;
2843 else
2844 fl4->saddr = fl4->daddr;
2845 }
2846
2847 /* L3 master device is the loopback for that domain */
2848 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2849 net->loopback_dev;
2850
2851 /* make sure orig_oif points to fib result device even
2852 * though packet rx/tx happens over loopback or l3mdev
2853 */
2854 orig_oif = FIB_RES_OIF(*res);
2855
2856 fl4->flowi4_oif = dev_out->ifindex;
2857 flags |= RTCF_LOCAL;
2858 goto make_route;
2859 }
2860
2861 fib_select_path(net, res, fl4, skb);
2862
2863 dev_out = FIB_RES_DEV(*res);
2864
2865 make_route:
2866 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2867
2868 out:
2869 return rth;
2870 }
2871
2872 static struct dst_ops ipv4_dst_blackhole_ops = {
2873 .family = AF_INET,
2874 .default_advmss = ipv4_default_advmss,
2875 .neigh_lookup = ipv4_neigh_lookup,
2876 .check = dst_blackhole_check,
2877 .cow_metrics = dst_blackhole_cow_metrics,
2878 .update_pmtu = dst_blackhole_update_pmtu,
2879 .redirect = dst_blackhole_redirect,
2880 .mtu = dst_blackhole_mtu,
2881 };
2882
ipv4_blackhole_route(struct net * net,struct dst_entry * dst_orig)2883 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2884 {
2885 struct rtable *ort = dst_rtable(dst_orig);
2886 struct rtable *rt;
2887
2888 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2889 if (rt) {
2890 struct dst_entry *new = &rt->dst;
2891
2892 new->__use = 1;
2893 new->input = dst_discard;
2894 new->output = dst_discard_out;
2895
2896 new->dev = net->loopback_dev;
2897 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2898
2899 rt->rt_is_input = ort->rt_is_input;
2900 rt->rt_iif = ort->rt_iif;
2901 rt->rt_pmtu = ort->rt_pmtu;
2902 rt->rt_mtu_locked = ort->rt_mtu_locked;
2903
2904 rt->rt_genid = rt_genid_ipv4(net);
2905 rt->rt_flags = ort->rt_flags;
2906 rt->rt_type = ort->rt_type;
2907 rt->rt_uses_gateway = ort->rt_uses_gateway;
2908 rt->rt_gw_family = ort->rt_gw_family;
2909 if (rt->rt_gw_family == AF_INET)
2910 rt->rt_gw4 = ort->rt_gw4;
2911 else if (rt->rt_gw_family == AF_INET6)
2912 rt->rt_gw6 = ort->rt_gw6;
2913 }
2914
2915 dst_release(dst_orig);
2916
2917 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2918 }
2919
ip_route_output_flow(struct net * net,struct flowi4 * flp4,const struct sock * sk)2920 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2921 const struct sock *sk)
2922 {
2923 struct rtable *rt = __ip_route_output_key(net, flp4);
2924
2925 if (IS_ERR(rt))
2926 return rt;
2927
2928 if (flp4->flowi4_proto) {
2929 flp4->flowi4_oif = rt->dst.dev->ifindex;
2930 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2931 flowi4_to_flowi(flp4),
2932 sk, 0));
2933 }
2934
2935 return rt;
2936 }
2937 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2938
2939 /* called with rcu_read_lock held */
rt_fill_info(struct net * net,__be32 dst,__be32 src,struct rtable * rt,u32 table_id,dscp_t dscp,struct flowi4 * fl4,struct sk_buff * skb,u32 portid,u32 seq,unsigned int flags)2940 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2941 struct rtable *rt, u32 table_id, dscp_t dscp,
2942 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2943 u32 seq, unsigned int flags)
2944 {
2945 struct rtmsg *r;
2946 struct nlmsghdr *nlh;
2947 unsigned long expires = 0;
2948 u32 error;
2949 u32 metrics[RTAX_MAX];
2950
2951 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2952 if (!nlh)
2953 return -EMSGSIZE;
2954
2955 r = nlmsg_data(nlh);
2956 r->rtm_family = AF_INET;
2957 r->rtm_dst_len = 32;
2958 r->rtm_src_len = 0;
2959 r->rtm_tos = inet_dscp_to_dsfield(dscp);
2960 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2961 if (nla_put_u32(skb, RTA_TABLE, table_id))
2962 goto nla_put_failure;
2963 r->rtm_type = rt->rt_type;
2964 r->rtm_scope = RT_SCOPE_UNIVERSE;
2965 r->rtm_protocol = RTPROT_UNSPEC;
2966 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2967 if (rt->rt_flags & RTCF_NOTIFY)
2968 r->rtm_flags |= RTM_F_NOTIFY;
2969 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2970 r->rtm_flags |= RTCF_DOREDIRECT;
2971
2972 if (nla_put_in_addr(skb, RTA_DST, dst))
2973 goto nla_put_failure;
2974 if (src) {
2975 r->rtm_src_len = 32;
2976 if (nla_put_in_addr(skb, RTA_SRC, src))
2977 goto nla_put_failure;
2978 }
2979 if (rt->dst.dev &&
2980 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2981 goto nla_put_failure;
2982 if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2983 goto nla_put_failure;
2984 #ifdef CONFIG_IP_ROUTE_CLASSID
2985 if (rt->dst.tclassid &&
2986 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2987 goto nla_put_failure;
2988 #endif
2989 if (fl4 && !rt_is_input_route(rt) &&
2990 fl4->saddr != src) {
2991 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2992 goto nla_put_failure;
2993 }
2994 if (rt->rt_uses_gateway) {
2995 if (rt->rt_gw_family == AF_INET &&
2996 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2997 goto nla_put_failure;
2998 } else if (rt->rt_gw_family == AF_INET6) {
2999 int alen = sizeof(struct in6_addr);
3000 struct nlattr *nla;
3001 struct rtvia *via;
3002
3003 nla = nla_reserve(skb, RTA_VIA, alen + 2);
3004 if (!nla)
3005 goto nla_put_failure;
3006
3007 via = nla_data(nla);
3008 via->rtvia_family = AF_INET6;
3009 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3010 }
3011 }
3012
3013 expires = READ_ONCE(rt->dst.expires);
3014 if (expires) {
3015 unsigned long now = jiffies;
3016
3017 if (time_before(now, expires))
3018 expires -= now;
3019 else
3020 expires = 0;
3021 }
3022
3023 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3024 if (rt->rt_pmtu && expires)
3025 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3026 if (rt->rt_mtu_locked && expires)
3027 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3028 if (rtnetlink_put_metrics(skb, metrics) < 0)
3029 goto nla_put_failure;
3030
3031 if (fl4) {
3032 if (fl4->flowi4_mark &&
3033 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3034 goto nla_put_failure;
3035
3036 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3037 nla_put_u32(skb, RTA_UID,
3038 from_kuid_munged(current_user_ns(),
3039 fl4->flowi4_uid)))
3040 goto nla_put_failure;
3041
3042 if (rt_is_input_route(rt)) {
3043 #ifdef CONFIG_IP_MROUTE
3044 if (ipv4_is_multicast(dst) &&
3045 !ipv4_is_local_multicast(dst) &&
3046 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3047 int err = ipmr_get_route(net, skb,
3048 fl4->saddr, fl4->daddr,
3049 r, portid);
3050
3051 if (err <= 0) {
3052 if (err == 0)
3053 return 0;
3054 goto nla_put_failure;
3055 }
3056 } else
3057 #endif
3058 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3059 goto nla_put_failure;
3060 }
3061 }
3062
3063 error = rt->dst.error;
3064
3065 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3066 goto nla_put_failure;
3067
3068 nlmsg_end(skb, nlh);
3069 return 0;
3070
3071 nla_put_failure:
3072 nlmsg_cancel(skb, nlh);
3073 return -EMSGSIZE;
3074 }
3075
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)3076 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3077 struct netlink_callback *cb, u32 table_id,
3078 struct fnhe_hash_bucket *bucket, int genid,
3079 int *fa_index, int fa_start, unsigned int flags)
3080 {
3081 int i;
3082
3083 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3084 struct fib_nh_exception *fnhe;
3085
3086 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3087 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3088 struct rtable *rt;
3089 int err;
3090
3091 if (*fa_index < fa_start)
3092 goto next;
3093
3094 if (fnhe->fnhe_genid != genid)
3095 goto next;
3096
3097 if (fnhe->fnhe_expires &&
3098 time_after(jiffies, fnhe->fnhe_expires))
3099 goto next;
3100
3101 rt = rcu_dereference(fnhe->fnhe_rth_input);
3102 if (!rt)
3103 rt = rcu_dereference(fnhe->fnhe_rth_output);
3104 if (!rt)
3105 goto next;
3106
3107 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3108 table_id, 0, NULL, skb,
3109 NETLINK_CB(cb->skb).portid,
3110 cb->nlh->nlmsg_seq, flags);
3111 if (err)
3112 return err;
3113 next:
3114 (*fa_index)++;
3115 }
3116 }
3117
3118 return 0;
3119 }
3120
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)3121 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3122 u32 table_id, struct fib_info *fi,
3123 int *fa_index, int fa_start, unsigned int flags)
3124 {
3125 struct net *net = sock_net(cb->skb->sk);
3126 int nhsel, genid = fnhe_genid(net);
3127
3128 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3129 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3130 struct fnhe_hash_bucket *bucket;
3131 int err;
3132
3133 if (nhc->nhc_flags & RTNH_F_DEAD)
3134 continue;
3135
3136 rcu_read_lock();
3137 bucket = rcu_dereference(nhc->nhc_exceptions);
3138 err = 0;
3139 if (bucket)
3140 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3141 genid, fa_index, fa_start,
3142 flags);
3143 rcu_read_unlock();
3144 if (err)
3145 return err;
3146 }
3147
3148 return 0;
3149 }
3150
inet_rtm_getroute_build_skb(__be32 src,__be32 dst,u8 ip_proto,__be16 sport,__be16 dport)3151 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3152 u8 ip_proto, __be16 sport,
3153 __be16 dport)
3154 {
3155 struct sk_buff *skb;
3156 struct iphdr *iph;
3157
3158 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3159 if (!skb)
3160 return NULL;
3161
3162 /* Reserve room for dummy headers, this skb can pass
3163 * through good chunk of routing engine.
3164 */
3165 skb_reset_mac_header(skb);
3166 skb_reset_network_header(skb);
3167 skb->protocol = htons(ETH_P_IP);
3168 iph = skb_put(skb, sizeof(struct iphdr));
3169 iph->protocol = ip_proto;
3170 iph->saddr = src;
3171 iph->daddr = dst;
3172 iph->version = 0x4;
3173 iph->frag_off = 0;
3174 iph->ihl = 0x5;
3175 skb_set_transport_header(skb, skb->len);
3176
3177 switch (iph->protocol) {
3178 case IPPROTO_UDP: {
3179 struct udphdr *udph;
3180
3181 udph = skb_put_zero(skb, sizeof(struct udphdr));
3182 udph->source = sport;
3183 udph->dest = dport;
3184 udph->len = htons(sizeof(struct udphdr));
3185 udph->check = 0;
3186 break;
3187 }
3188 case IPPROTO_TCP: {
3189 struct tcphdr *tcph;
3190
3191 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3192 tcph->source = sport;
3193 tcph->dest = dport;
3194 tcph->doff = sizeof(struct tcphdr) / 4;
3195 tcph->rst = 1;
3196 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3197 src, dst, 0);
3198 break;
3199 }
3200 case IPPROTO_ICMP: {
3201 struct icmphdr *icmph;
3202
3203 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3204 icmph->type = ICMP_ECHO;
3205 icmph->code = 0;
3206 }
3207 }
3208
3209 return skb;
3210 }
3211
inet_rtm_valid_getroute_req(struct sk_buff * skb,const struct nlmsghdr * nlh,struct nlattr ** tb,struct netlink_ext_ack * extack)3212 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3213 const struct nlmsghdr *nlh,
3214 struct nlattr **tb,
3215 struct netlink_ext_ack *extack)
3216 {
3217 struct rtmsg *rtm;
3218 int i, err;
3219
3220 rtm = nlmsg_payload(nlh, sizeof(*rtm));
3221 if (!rtm) {
3222 NL_SET_ERR_MSG(extack,
3223 "ipv4: Invalid header for route get request");
3224 return -EINVAL;
3225 }
3226
3227 if (!netlink_strict_get_check(skb))
3228 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3229 rtm_ipv4_policy, extack);
3230
3231 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3232 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3233 rtm->rtm_table || rtm->rtm_protocol ||
3234 rtm->rtm_scope || rtm->rtm_type) {
3235 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3236 return -EINVAL;
3237 }
3238
3239 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3240 RTM_F_LOOKUP_TABLE |
3241 RTM_F_FIB_MATCH)) {
3242 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3243 return -EINVAL;
3244 }
3245
3246 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3247 rtm_ipv4_policy, extack);
3248 if (err)
3249 return err;
3250
3251 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3252 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3253 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3254 return -EINVAL;
3255 }
3256
3257 for (i = 0; i <= RTA_MAX; i++) {
3258 if (!tb[i])
3259 continue;
3260
3261 switch (i) {
3262 case RTA_IIF:
3263 case RTA_OIF:
3264 case RTA_SRC:
3265 case RTA_DST:
3266 case RTA_IP_PROTO:
3267 case RTA_SPORT:
3268 case RTA_DPORT:
3269 case RTA_MARK:
3270 case RTA_UID:
3271 break;
3272 default:
3273 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3274 return -EINVAL;
3275 }
3276 }
3277
3278 return 0;
3279 }
3280
inet_rtm_getroute(struct sk_buff * in_skb,struct nlmsghdr * nlh,struct netlink_ext_ack * extack)3281 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3282 struct netlink_ext_ack *extack)
3283 {
3284 struct net *net = sock_net(in_skb->sk);
3285 struct nlattr *tb[RTA_MAX+1];
3286 u32 table_id = RT_TABLE_MAIN;
3287 __be16 sport = 0, dport = 0;
3288 struct fib_result res = {};
3289 u8 ip_proto = IPPROTO_UDP;
3290 struct rtable *rt = NULL;
3291 struct sk_buff *skb;
3292 struct rtmsg *rtm;
3293 struct flowi4 fl4 = {};
3294 __be32 dst = 0;
3295 __be32 src = 0;
3296 dscp_t dscp;
3297 kuid_t uid;
3298 u32 iif;
3299 int err;
3300 int mark;
3301
3302 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3303 if (err < 0)
3304 return err;
3305
3306 rtm = nlmsg_data(nlh);
3307 src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3308 dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3309 iif = nla_get_u32_default(tb[RTA_IIF], 0);
3310 mark = nla_get_u32_default(tb[RTA_MARK], 0);
3311 dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
3312 if (tb[RTA_UID])
3313 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3314 else
3315 uid = (iif ? INVALID_UID : current_uid());
3316
3317 if (tb[RTA_IP_PROTO]) {
3318 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3319 &ip_proto, AF_INET, extack);
3320 if (err)
3321 return err;
3322 }
3323
3324 if (tb[RTA_SPORT])
3325 sport = nla_get_be16(tb[RTA_SPORT]);
3326
3327 if (tb[RTA_DPORT])
3328 dport = nla_get_be16(tb[RTA_DPORT]);
3329
3330 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3331 if (!skb)
3332 return -ENOBUFS;
3333
3334 fl4.daddr = dst;
3335 fl4.saddr = src;
3336 fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
3337 fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3338 fl4.flowi4_mark = mark;
3339 fl4.flowi4_uid = uid;
3340 if (sport)
3341 fl4.fl4_sport = sport;
3342 if (dport)
3343 fl4.fl4_dport = dport;
3344 fl4.flowi4_proto = ip_proto;
3345
3346 rcu_read_lock();
3347
3348 if (iif) {
3349 struct net_device *dev;
3350
3351 dev = dev_get_by_index_rcu(net, iif);
3352 if (!dev) {
3353 err = -ENODEV;
3354 goto errout_rcu;
3355 }
3356
3357 fl4.flowi4_iif = iif; /* for rt_fill_info */
3358 skb->dev = dev;
3359 skb->mark = mark;
3360 err = ip_route_input_rcu(skb, dst, src, dscp, dev,
3361 &res) ? -EINVAL : 0;
3362
3363 rt = skb_rtable(skb);
3364 if (err == 0 && rt->dst.error)
3365 err = -rt->dst.error;
3366 } else {
3367 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3368 skb->dev = net->loopback_dev;
3369 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3370 err = 0;
3371 if (IS_ERR(rt))
3372 err = PTR_ERR(rt);
3373 else
3374 skb_dst_set(skb, &rt->dst);
3375 }
3376
3377 if (err)
3378 goto errout_rcu;
3379
3380 if (rtm->rtm_flags & RTM_F_NOTIFY)
3381 rt->rt_flags |= RTCF_NOTIFY;
3382
3383 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3384 table_id = res.table ? res.table->tb_id : 0;
3385
3386 /* reset skb for netlink reply msg */
3387 skb_trim(skb, 0);
3388 skb_reset_network_header(skb);
3389 skb_reset_transport_header(skb);
3390 skb_reset_mac_header(skb);
3391
3392 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3393 struct fib_rt_info fri;
3394
3395 if (!res.fi) {
3396 err = fib_props[res.type].error;
3397 if (!err)
3398 err = -EHOSTUNREACH;
3399 goto errout_rcu;
3400 }
3401 fri.fi = res.fi;
3402 fri.tb_id = table_id;
3403 fri.dst = res.prefix;
3404 fri.dst_len = res.prefixlen;
3405 fri.dscp = res.dscp;
3406 fri.type = rt->rt_type;
3407 fri.offload = 0;
3408 fri.trap = 0;
3409 fri.offload_failed = 0;
3410 if (res.fa_head) {
3411 struct fib_alias *fa;
3412
3413 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3414 u8 slen = 32 - fri.dst_len;
3415
3416 if (fa->fa_slen == slen &&
3417 fa->tb_id == fri.tb_id &&
3418 fa->fa_dscp == fri.dscp &&
3419 fa->fa_info == res.fi &&
3420 fa->fa_type == fri.type) {
3421 fri.offload = READ_ONCE(fa->offload);
3422 fri.trap = READ_ONCE(fa->trap);
3423 fri.offload_failed =
3424 READ_ONCE(fa->offload_failed);
3425 break;
3426 }
3427 }
3428 }
3429 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3430 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3431 } else {
3432 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3433 skb, NETLINK_CB(in_skb).portid,
3434 nlh->nlmsg_seq, 0);
3435 }
3436 if (err < 0)
3437 goto errout_rcu;
3438
3439 rcu_read_unlock();
3440
3441 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3442
3443 errout_free:
3444 return err;
3445 errout_rcu:
3446 rcu_read_unlock();
3447 kfree_skb(skb);
3448 goto errout_free;
3449 }
3450
ip_rt_multicast_event(struct in_device * in_dev)3451 void ip_rt_multicast_event(struct in_device *in_dev)
3452 {
3453 rt_cache_flush(dev_net(in_dev->dev));
3454 }
3455
3456 #ifdef CONFIG_SYSCTL
3457 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3458 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3459 static int ip_rt_gc_elasticity __read_mostly = 8;
3460 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3461
ipv4_sysctl_rtcache_flush(const struct ctl_table * __ctl,int write,void * buffer,size_t * lenp,loff_t * ppos)3462 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3463 void *buffer, size_t *lenp, loff_t *ppos)
3464 {
3465 struct net *net = (struct net *)__ctl->extra1;
3466
3467 if (write) {
3468 rt_cache_flush(net);
3469 fnhe_genid_bump(net);
3470 return 0;
3471 }
3472
3473 return -EINVAL;
3474 }
3475
3476 static struct ctl_table ipv4_route_table[] = {
3477 {
3478 .procname = "gc_thresh",
3479 .data = &ipv4_dst_ops.gc_thresh,
3480 .maxlen = sizeof(int),
3481 .mode = 0644,
3482 .proc_handler = proc_dointvec,
3483 },
3484 {
3485 .procname = "max_size",
3486 .data = &ip_rt_max_size,
3487 .maxlen = sizeof(int),
3488 .mode = 0644,
3489 .proc_handler = proc_dointvec,
3490 },
3491 {
3492 /* Deprecated. Use gc_min_interval_ms */
3493
3494 .procname = "gc_min_interval",
3495 .data = &ip_rt_gc_min_interval,
3496 .maxlen = sizeof(int),
3497 .mode = 0644,
3498 .proc_handler = proc_dointvec_jiffies,
3499 },
3500 {
3501 .procname = "gc_min_interval_ms",
3502 .data = &ip_rt_gc_min_interval,
3503 .maxlen = sizeof(int),
3504 .mode = 0644,
3505 .proc_handler = proc_dointvec_ms_jiffies,
3506 },
3507 {
3508 .procname = "gc_timeout",
3509 .data = &ip_rt_gc_timeout,
3510 .maxlen = sizeof(int),
3511 .mode = 0644,
3512 .proc_handler = proc_dointvec_jiffies,
3513 },
3514 {
3515 .procname = "gc_interval",
3516 .data = &ip_rt_gc_interval,
3517 .maxlen = sizeof(int),
3518 .mode = 0644,
3519 .proc_handler = proc_dointvec_jiffies,
3520 },
3521 {
3522 .procname = "redirect_load",
3523 .data = &ip_rt_redirect_load,
3524 .maxlen = sizeof(int),
3525 .mode = 0644,
3526 .proc_handler = proc_dointvec,
3527 },
3528 {
3529 .procname = "redirect_number",
3530 .data = &ip_rt_redirect_number,
3531 .maxlen = sizeof(int),
3532 .mode = 0644,
3533 .proc_handler = proc_dointvec,
3534 },
3535 {
3536 .procname = "redirect_silence",
3537 .data = &ip_rt_redirect_silence,
3538 .maxlen = sizeof(int),
3539 .mode = 0644,
3540 .proc_handler = proc_dointvec,
3541 },
3542 {
3543 .procname = "error_cost",
3544 .data = &ip_rt_error_cost,
3545 .maxlen = sizeof(int),
3546 .mode = 0644,
3547 .proc_handler = proc_dointvec,
3548 },
3549 {
3550 .procname = "error_burst",
3551 .data = &ip_rt_error_burst,
3552 .maxlen = sizeof(int),
3553 .mode = 0644,
3554 .proc_handler = proc_dointvec,
3555 },
3556 {
3557 .procname = "gc_elasticity",
3558 .data = &ip_rt_gc_elasticity,
3559 .maxlen = sizeof(int),
3560 .mode = 0644,
3561 .proc_handler = proc_dointvec,
3562 },
3563 };
3564
3565 static const char ipv4_route_flush_procname[] = "flush";
3566
3567 static struct ctl_table ipv4_route_netns_table[] = {
3568 {
3569 .procname = ipv4_route_flush_procname,
3570 .maxlen = sizeof(int),
3571 .mode = 0200,
3572 .proc_handler = ipv4_sysctl_rtcache_flush,
3573 },
3574 {
3575 .procname = "min_pmtu",
3576 .data = &init_net.ipv4.ip_rt_min_pmtu,
3577 .maxlen = sizeof(int),
3578 .mode = 0644,
3579 .proc_handler = proc_dointvec_minmax,
3580 .extra1 = &ip_min_valid_pmtu,
3581 },
3582 {
3583 .procname = "mtu_expires",
3584 .data = &init_net.ipv4.ip_rt_mtu_expires,
3585 .maxlen = sizeof(int),
3586 .mode = 0644,
3587 .proc_handler = proc_dointvec_jiffies,
3588 },
3589 {
3590 .procname = "min_adv_mss",
3591 .data = &init_net.ipv4.ip_rt_min_advmss,
3592 .maxlen = sizeof(int),
3593 .mode = 0644,
3594 .proc_handler = proc_dointvec,
3595 },
3596 };
3597
sysctl_route_net_init(struct net * net)3598 static __net_init int sysctl_route_net_init(struct net *net)
3599 {
3600 struct ctl_table *tbl;
3601 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3602
3603 tbl = ipv4_route_netns_table;
3604 if (!net_eq(net, &init_net)) {
3605 int i;
3606
3607 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3608 if (!tbl)
3609 goto err_dup;
3610
3611 /* Don't export non-whitelisted sysctls to unprivileged users */
3612 if (net->user_ns != &init_user_ns) {
3613 if (tbl[0].procname != ipv4_route_flush_procname)
3614 table_size = 0;
3615 }
3616
3617 /* Update the variables to point into the current struct net
3618 * except for the first element flush
3619 */
3620 for (i = 1; i < table_size; i++)
3621 tbl[i].data += (void *)net - (void *)&init_net;
3622 }
3623 tbl[0].extra1 = net;
3624
3625 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3626 tbl, table_size);
3627 if (!net->ipv4.route_hdr)
3628 goto err_reg;
3629 return 0;
3630
3631 err_reg:
3632 if (tbl != ipv4_route_netns_table)
3633 kfree(tbl);
3634 err_dup:
3635 return -ENOMEM;
3636 }
3637
sysctl_route_net_exit(struct net * net)3638 static __net_exit void sysctl_route_net_exit(struct net *net)
3639 {
3640 const struct ctl_table *tbl;
3641
3642 tbl = net->ipv4.route_hdr->ctl_table_arg;
3643 unregister_net_sysctl_table(net->ipv4.route_hdr);
3644 BUG_ON(tbl == ipv4_route_netns_table);
3645 kfree(tbl);
3646 }
3647
3648 static __net_initdata struct pernet_operations sysctl_route_ops = {
3649 .init = sysctl_route_net_init,
3650 .exit = sysctl_route_net_exit,
3651 };
3652 #endif
3653
netns_ip_rt_init(struct net * net)3654 static __net_init int netns_ip_rt_init(struct net *net)
3655 {
3656 /* Set default value for namespaceified sysctls */
3657 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3658 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3659 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3660 return 0;
3661 }
3662
3663 static struct pernet_operations __net_initdata ip_rt_ops = {
3664 .init = netns_ip_rt_init,
3665 };
3666
rt_genid_init(struct net * net)3667 static __net_init int rt_genid_init(struct net *net)
3668 {
3669 atomic_set(&net->ipv4.rt_genid, 0);
3670 atomic_set(&net->fnhe_genid, 0);
3671 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3672 return 0;
3673 }
3674
3675 static __net_initdata struct pernet_operations rt_genid_ops = {
3676 .init = rt_genid_init,
3677 };
3678
ipv4_inetpeer_init(struct net * net)3679 static int __net_init ipv4_inetpeer_init(struct net *net)
3680 {
3681 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3682
3683 if (!bp)
3684 return -ENOMEM;
3685 inet_peer_base_init(bp);
3686 net->ipv4.peers = bp;
3687 return 0;
3688 }
3689
ipv4_inetpeer_exit(struct net * net)3690 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3691 {
3692 struct inet_peer_base *bp = net->ipv4.peers;
3693
3694 net->ipv4.peers = NULL;
3695 inetpeer_invalidate_tree(bp);
3696 kfree(bp);
3697 }
3698
3699 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3700 .init = ipv4_inetpeer_init,
3701 .exit = ipv4_inetpeer_exit,
3702 };
3703
3704 #ifdef CONFIG_IP_ROUTE_CLASSID
3705 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3706 #endif /* CONFIG_IP_ROUTE_CLASSID */
3707
3708 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3709 {.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3710 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3711 };
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 get_random_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 spin_lock_init(&ul->lock);
3740 }
3741 #ifdef CONFIG_IP_ROUTE_CLASSID
3742 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3743 if (!ip_rt_acct)
3744 panic("IP: failed to allocate ip_rt_acct\n");
3745 #endif
3746
3747 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3748 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3749
3750 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3751
3752 if (dst_entries_init(&ipv4_dst_ops) < 0)
3753 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3754
3755 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3756 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3757
3758 ipv4_dst_ops.gc_thresh = ~0;
3759 ip_rt_max_size = INT_MAX;
3760
3761 devinet_init();
3762 ip_fib_init();
3763
3764 if (ip_rt_proc_init())
3765 pr_err("Unable to create route proc files\n");
3766 #ifdef CONFIG_XFRM
3767 xfrm_init();
3768 xfrm4_init();
3769 #endif
3770 rtnl_register_many(ip_rt_rtnl_msg_handlers);
3771
3772 #ifdef CONFIG_SYSCTL
3773 register_pernet_subsys(&sysctl_route_ops);
3774 #endif
3775 register_pernet_subsys(&ip_rt_ops);
3776 register_pernet_subsys(&rt_genid_ops);
3777 register_pernet_subsys(&ipv4_inetpeer_ops);
3778 return 0;
3779 }
3780
3781 #ifdef CONFIG_SYSCTL
3782 /*
3783 * We really need to sanitize the damn ipv4 init order, then all
3784 * this nonsense will go away.
3785 */
ip_static_sysctl_init(void)3786 void __init ip_static_sysctl_init(void)
3787 {
3788 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3789 }
3790 #endif
3791