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/flow.h>
88 #include <net/inet_dscp.h>
89 #include <net/net_namespace.h>
90 #include <net/ip.h>
91 #include <net/route.h>
92 #include <net/inetpeer.h>
93 #include <net/sock.h>
94 #include <net/ip_fib.h>
95 #include <net/nexthop.h>
96 #include <net/tcp.h>
97 #include <net/icmp.h>
98 #include <net/xfrm.h>
99 #include <net/lwtunnel.h>
100 #include <net/netevent.h>
101 #include <net/rtnetlink.h>
102 #ifdef CONFIG_SYSCTL
103 #include <linux/sysctl.h>
104 #endif
105 #include <net/secure_seq.h>
106 #include <net/ip_tunnels.h>
107
108 #include "fib_lookup.h"
109
110 #define RT_GC_TIMEOUT (300*HZ)
111
112 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
113 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
114 #define DEFAULT_MIN_ADVMSS 256
115 static int ip_rt_max_size;
116 static int ip_rt_redirect_number __read_mostly = 9;
117 static int ip_rt_redirect_load __read_mostly = HZ / 50;
118 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
119 static int ip_rt_error_cost __read_mostly = HZ;
120 static int ip_rt_error_burst __read_mostly = 5 * HZ;
121
122 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
123
124 /*
125 * Interface to generic destination cache.
126 */
127
128 INDIRECT_CALLABLE_SCOPE
129 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
130 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
131 INDIRECT_CALLABLE_SCOPE
132 unsigned int ipv4_mtu(const struct dst_entry *dst);
133 static void ipv4_negative_advice(struct sock *sk,
134 struct dst_entry *dst);
135 static void ipv4_link_failure(struct sk_buff *skb);
136 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
137 struct sk_buff *skb, u32 mtu,
138 bool confirm_neigh);
139 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
140 struct sk_buff *skb);
141 static void ipv4_dst_destroy(struct dst_entry *dst);
142
ipv4_cow_metrics(struct dst_entry * dst,unsigned long old)143 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
144 {
145 WARN_ON(1);
146 return NULL;
147 }
148
149 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
150 struct sk_buff *skb,
151 const void *daddr);
152 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
153
154 static struct dst_ops ipv4_dst_ops = {
155 .family = AF_INET,
156 .check = ipv4_dst_check,
157 .default_advmss = ipv4_default_advmss,
158 .mtu = ipv4_mtu,
159 .cow_metrics = ipv4_cow_metrics,
160 .destroy = ipv4_dst_destroy,
161 .negative_advice = ipv4_negative_advice,
162 .link_failure = ipv4_link_failure,
163 .update_pmtu = ip_rt_update_pmtu,
164 .redirect = ip_do_redirect,
165 .local_out = __ip_local_out,
166 .neigh_lookup = ipv4_neigh_lookup,
167 .confirm_neigh = ipv4_confirm_neigh,
168 };
169
170 #define ECN_OR_COST(class) TC_PRIO_##class
171
172 const __u8 ip_tos2prio[16] = {
173 TC_PRIO_BESTEFFORT,
174 ECN_OR_COST(BESTEFFORT),
175 TC_PRIO_BESTEFFORT,
176 ECN_OR_COST(BESTEFFORT),
177 TC_PRIO_BULK,
178 ECN_OR_COST(BULK),
179 TC_PRIO_BULK,
180 ECN_OR_COST(BULK),
181 TC_PRIO_INTERACTIVE,
182 ECN_OR_COST(INTERACTIVE),
183 TC_PRIO_INTERACTIVE,
184 ECN_OR_COST(INTERACTIVE),
185 TC_PRIO_INTERACTIVE_BULK,
186 ECN_OR_COST(INTERACTIVE_BULK),
187 TC_PRIO_INTERACTIVE_BULK,
188 ECN_OR_COST(INTERACTIVE_BULK)
189 };
190 EXPORT_SYMBOL(ip_tos2prio);
191
192 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
193 #ifndef CONFIG_PREEMPT_RT
194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
195 #else
196 #define RT_CACHE_STAT_INC(field) this_cpu_inc(rt_cache_stat.field)
197 #endif
198
199 #ifdef CONFIG_PROC_FS
rt_cache_seq_start(struct seq_file * seq,loff_t * pos)200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
201 {
202 if (*pos)
203 return NULL;
204 return SEQ_START_TOKEN;
205 }
206
rt_cache_seq_next(struct seq_file * seq,void * v,loff_t * pos)207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
208 {
209 ++*pos;
210 return NULL;
211 }
212
rt_cache_seq_stop(struct seq_file * seq,void * v)213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
214 {
215 }
216
rt_cache_seq_show(struct seq_file * seq,void * v)217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
218 {
219 if (v == SEQ_START_TOKEN)
220 seq_printf(seq, "%-127s\n",
221 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
222 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
223 "HHUptod\tSpecDst");
224 return 0;
225 }
226
227 static const struct seq_operations rt_cache_seq_ops = {
228 .start = rt_cache_seq_start,
229 .next = rt_cache_seq_next,
230 .stop = rt_cache_seq_stop,
231 .show = rt_cache_seq_show,
232 };
233
rt_cpu_seq_start(struct seq_file * seq,loff_t * pos)234 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
235 {
236 int cpu;
237
238 if (*pos == 0)
239 return SEQ_START_TOKEN;
240
241 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
242 if (!cpu_possible(cpu))
243 continue;
244 *pos = cpu+1;
245 return &per_cpu(rt_cache_stat, cpu);
246 }
247 return NULL;
248 }
249
rt_cpu_seq_next(struct seq_file * seq,void * v,loff_t * pos)250 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
251 {
252 int cpu;
253
254 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
255 if (!cpu_possible(cpu))
256 continue;
257 *pos = cpu+1;
258 return &per_cpu(rt_cache_stat, cpu);
259 }
260 (*pos)++;
261 return NULL;
262
263 }
264
rt_cpu_seq_stop(struct seq_file * seq,void * v)265 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
266 {
267
268 }
269
rt_cpu_seq_show(struct seq_file * seq,void * v)270 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
271 {
272 struct rt_cache_stat *st = v;
273
274 if (v == SEQ_START_TOKEN) {
275 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");
276 return 0;
277 }
278
279 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x "
280 "%08x %08x %08x %08x %08x %08x "
281 "%08x %08x %08x %08x\n",
282 dst_entries_get_slow(&ipv4_dst_ops),
283 0, /* st->in_hit */
284 st->in_slow_tot,
285 st->in_slow_mc,
286 st->in_no_route,
287 st->in_brd,
288 st->in_martian_dst,
289 st->in_martian_src,
290
291 0, /* st->out_hit */
292 st->out_slow_tot,
293 st->out_slow_mc,
294
295 0, /* st->gc_total */
296 0, /* st->gc_ignored */
297 0, /* st->gc_goal_miss */
298 0, /* st->gc_dst_overflow */
299 0, /* st->in_hlist_search */
300 0 /* st->out_hlist_search */
301 );
302 return 0;
303 }
304
305 static const struct seq_operations rt_cpu_seq_ops = {
306 .start = rt_cpu_seq_start,
307 .next = rt_cpu_seq_next,
308 .stop = rt_cpu_seq_stop,
309 .show = rt_cpu_seq_show,
310 };
311
312 #ifdef CONFIG_IP_ROUTE_CLASSID
rt_acct_proc_show(struct seq_file * m,void * v)313 static int rt_acct_proc_show(struct seq_file *m, void *v)
314 {
315 struct ip_rt_acct *dst, *src;
316 unsigned int i, j;
317
318 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
319 if (!dst)
320 return -ENOMEM;
321
322 for_each_possible_cpu(i) {
323 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
324 for (j = 0; j < 256; j++) {
325 dst[j].o_bytes += src[j].o_bytes;
326 dst[j].o_packets += src[j].o_packets;
327 dst[j].i_bytes += src[j].i_bytes;
328 dst[j].i_packets += src[j].i_packets;
329 }
330 }
331
332 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
333 kfree(dst);
334 return 0;
335 }
336 #endif
337
ip_rt_do_proc_init(struct net * net)338 static int __net_init ip_rt_do_proc_init(struct net *net)
339 {
340 struct proc_dir_entry *pde;
341
342 pde = proc_create_seq("rt_cache", 0444, net->proc_net,
343 &rt_cache_seq_ops);
344 if (!pde)
345 goto err1;
346
347 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
348 &rt_cpu_seq_ops);
349 if (!pde)
350 goto err2;
351
352 #ifdef CONFIG_IP_ROUTE_CLASSID
353 pde = proc_create_single("rt_acct", 0, net->proc_net,
354 rt_acct_proc_show);
355 if (!pde)
356 goto err3;
357 #endif
358 return 0;
359
360 #ifdef CONFIG_IP_ROUTE_CLASSID
361 err3:
362 remove_proc_entry("rt_cache", net->proc_net_stat);
363 #endif
364 err2:
365 remove_proc_entry("rt_cache", net->proc_net);
366 err1:
367 return -ENOMEM;
368 }
369
ip_rt_do_proc_exit(struct net * net)370 static void __net_exit ip_rt_do_proc_exit(struct net *net)
371 {
372 remove_proc_entry("rt_cache", net->proc_net_stat);
373 remove_proc_entry("rt_cache", net->proc_net);
374 #ifdef CONFIG_IP_ROUTE_CLASSID
375 remove_proc_entry("rt_acct", net->proc_net);
376 #endif
377 }
378
379 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
380 .init = ip_rt_do_proc_init,
381 .exit = ip_rt_do_proc_exit,
382 };
383
ip_rt_proc_init(void)384 static int __init ip_rt_proc_init(void)
385 {
386 return register_pernet_subsys(&ip_rt_proc_ops);
387 }
388
389 #else
ip_rt_proc_init(void)390 static inline int ip_rt_proc_init(void)
391 {
392 return 0;
393 }
394 #endif /* CONFIG_PROC_FS */
395
rt_is_expired(const struct rtable * rth)396 static inline bool rt_is_expired(const struct rtable *rth)
397 {
398 bool res;
399
400 rcu_read_lock();
401 res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev));
402 rcu_read_unlock();
403
404 return res;
405 }
406
rt_cache_flush(struct net * net)407 void rt_cache_flush(struct net *net)
408 {
409 rt_genid_bump_ipv4(net);
410 }
411
ipv4_neigh_lookup(const struct dst_entry * dst,struct sk_buff * skb,const void * daddr)412 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
413 struct sk_buff *skb,
414 const void *daddr)
415 {
416 const struct rtable *rt = container_of(dst, struct rtable, dst);
417 struct net_device *dev;
418 struct neighbour *n;
419
420 rcu_read_lock();
421 dev = dst_dev_rcu(dst);
422 if (likely(rt->rt_gw_family == AF_INET)) {
423 n = ip_neigh_gw4(dev, rt->rt_gw4);
424 } else if (rt->rt_gw_family == AF_INET6) {
425 n = ip_neigh_gw6(dev, &rt->rt_gw6);
426 } else {
427 __be32 pkey;
428
429 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
430 n = ip_neigh_gw4(dev, pkey);
431 }
432
433 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
434 n = NULL;
435
436 rcu_read_unlock();
437
438 return n;
439 }
440
ipv4_confirm_neigh(const struct dst_entry * dst,const void * daddr)441 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
442 {
443 const struct rtable *rt = container_of(dst, struct rtable, dst);
444 struct net_device *dev = dst_dev(dst);
445 const __be32 *pkey = daddr;
446
447 if (rt->rt_gw_family == AF_INET) {
448 pkey = (const __be32 *)&rt->rt_gw4;
449 } else if (rt->rt_gw_family == AF_INET6) {
450 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
451 } else if (!daddr ||
452 (rt->rt_flags &
453 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
454 return;
455 }
456 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
457 }
458
459 /* Hash tables of size 2048..262144 depending on RAM size.
460 * Each bucket uses 8 bytes.
461 */
462 static u32 ip_idents_mask __read_mostly;
463 static atomic_t *ip_idents __read_mostly;
464 static u32 *ip_tstamps __read_mostly;
465
466 /* In order to protect privacy, we add a perturbation to identifiers
467 * if one generator is seldom used. This makes hard for an attacker
468 * to infer how many packets were sent between two points in time.
469 */
ip_idents_reserve(u32 hash,int segs)470 static u32 ip_idents_reserve(u32 hash, int segs)
471 {
472 u32 bucket, old, now = (u32)jiffies;
473 atomic_t *p_id;
474 u32 *p_tstamp;
475 u32 delta = 0;
476
477 bucket = hash & ip_idents_mask;
478 p_tstamp = ip_tstamps + bucket;
479 p_id = ip_idents + bucket;
480 old = READ_ONCE(*p_tstamp);
481
482 if (old != now && cmpxchg(p_tstamp, old, now) == old)
483 delta = get_random_u32_below(now - old);
484
485 /* If UBSAN reports an error there, please make sure your compiler
486 * supports -fno-strict-overflow before reporting it that was a bug
487 * in UBSAN, and it has been fixed in GCC-8.
488 */
489 return atomic_add_return(segs + delta, p_id) - segs;
490 }
491
__ip_select_ident(struct net * net,struct iphdr * iph,int segs)492 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
493 {
494 u32 hash, id;
495
496 /* Note the following code is not safe, but this is okay. */
497 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
498 get_random_bytes(&net->ipv4.ip_id_key,
499 sizeof(net->ipv4.ip_id_key));
500
501 hash = siphash_3u32((__force u32)iph->daddr,
502 (__force u32)iph->saddr,
503 iph->protocol,
504 &net->ipv4.ip_id_key);
505 id = ip_idents_reserve(hash, segs);
506 iph->id = htons(id);
507 }
508 EXPORT_SYMBOL(__ip_select_ident);
509
__build_flow_key(const struct net * net,struct flowi4 * fl4,const struct sock * sk,const struct iphdr * iph,int oif,__u8 tos,u8 prot,u32 mark,int flow_flags)510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
511 const struct sock *sk, const struct iphdr *iph,
512 int oif, __u8 tos, u8 prot, u32 mark,
513 int flow_flags)
514 {
515 __u8 scope = RT_SCOPE_UNIVERSE;
516
517 if (sk) {
518 oif = sk->sk_bound_dev_if;
519 mark = READ_ONCE(sk->sk_mark);
520 tos = ip_sock_rt_tos(sk);
521 scope = ip_sock_rt_scope(sk);
522 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
523 sk->sk_protocol;
524 }
525
526 flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope,
527 prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
528 sock_net_uid(net, sk));
529 }
530
build_skb_flow_key(struct flowi4 * fl4,const struct sk_buff * skb,const struct sock * sk)531 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
532 const struct sock *sk)
533 {
534 const struct net *net = dev_net(skb->dev);
535 const struct iphdr *iph = ip_hdr(skb);
536 int oif = skb->dev->ifindex;
537 u8 prot = iph->protocol;
538 u32 mark = skb->mark;
539 __u8 tos = iph->tos;
540
541 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
542 }
543
build_sk_flow_key(struct flowi4 * fl4,const struct sock * sk)544 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
545 {
546 const struct inet_sock *inet = inet_sk(sk);
547 const struct ip_options_rcu *inet_opt;
548 __be32 daddr = inet->inet_daddr;
549
550 rcu_read_lock();
551 inet_opt = rcu_dereference(inet->inet_opt);
552 if (inet_opt && inet_opt->opt.srr)
553 daddr = inet_opt->opt.faddr;
554 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
555 ip_sock_rt_tos(sk),
556 ip_sock_rt_scope(sk),
557 inet_test_bit(HDRINCL, sk) ?
558 IPPROTO_RAW : sk->sk_protocol,
559 inet_sk_flowi_flags(sk),
560 daddr, inet->inet_saddr, 0, 0,
561 sk_uid(sk));
562 rcu_read_unlock();
563 }
564
ip_rt_build_flow_key(struct flowi4 * fl4,const struct sock * sk,const struct sk_buff * skb)565 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
566 const struct sk_buff *skb)
567 {
568 if (skb)
569 build_skb_flow_key(fl4, skb, sk);
570 else
571 build_sk_flow_key(fl4, sk);
572 }
573
574 static DEFINE_SPINLOCK(fnhe_lock);
575
fnhe_flush_routes(struct fib_nh_exception * fnhe)576 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
577 {
578 struct rtable *rt;
579
580 rt = rcu_dereference(fnhe->fnhe_rth_input);
581 if (rt) {
582 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
583 dst_dev_put(&rt->dst);
584 dst_release(&rt->dst);
585 }
586 rt = rcu_dereference(fnhe->fnhe_rth_output);
587 if (rt) {
588 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
589 dst_dev_put(&rt->dst);
590 dst_release(&rt->dst);
591 }
592 }
593
fnhe_remove_oldest(struct fnhe_hash_bucket * hash)594 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
595 {
596 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
597 struct fib_nh_exception *fnhe, *oldest = NULL;
598
599 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
600 fnhe = rcu_dereference_protected(*fnhe_p,
601 lockdep_is_held(&fnhe_lock));
602 if (!fnhe)
603 break;
604 if (!oldest ||
605 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
606 oldest = fnhe;
607 oldest_p = fnhe_p;
608 }
609 }
610
611 /* Clear oldest->fnhe_daddr to prevent this fnhe from being
612 * rebound with new dsts in rt_bind_exception().
613 */
614 oldest->fnhe_daddr = 0;
615 fnhe_flush_routes(oldest);
616 *oldest_p = oldest->fnhe_next;
617 kfree_rcu(oldest, rcu);
618 }
619
fnhe_hashfun(__be32 daddr)620 static u32 fnhe_hashfun(__be32 daddr)
621 {
622 static siphash_aligned_key_t fnhe_hash_key;
623 u64 hval;
624
625 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
626 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
627 return hash_64(hval, FNHE_HASH_SHIFT);
628 }
629
fill_route_from_fnhe(struct rtable * rt,struct fib_nh_exception * fnhe)630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
631 {
632 rt->rt_pmtu = fnhe->fnhe_pmtu;
633 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
634 rt->dst.expires = fnhe->fnhe_expires;
635
636 if (fnhe->fnhe_gw) {
637 rt->rt_flags |= RTCF_REDIRECTED;
638 rt->rt_uses_gateway = 1;
639 rt->rt_gw_family = AF_INET;
640 rt->rt_gw4 = fnhe->fnhe_gw;
641 }
642 }
643
update_or_create_fnhe(struct fib_nh_common * nhc,__be32 daddr,__be32 gw,u32 pmtu,bool lock,unsigned long expires)644 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
645 __be32 gw, u32 pmtu, bool lock,
646 unsigned long expires)
647 {
648 struct fnhe_hash_bucket *hash;
649 struct fib_nh_exception *fnhe;
650 struct rtable *rt;
651 u32 genid, hval;
652 unsigned int i;
653 int depth;
654
655 genid = fnhe_genid(dev_net(nhc->nhc_dev));
656 hval = fnhe_hashfun(daddr);
657
658 spin_lock_bh(&fnhe_lock);
659
660 hash = rcu_dereference(nhc->nhc_exceptions);
661 if (!hash) {
662 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
663 if (!hash)
664 goto out_unlock;
665 rcu_assign_pointer(nhc->nhc_exceptions, hash);
666 }
667
668 hash += hval;
669
670 depth = 0;
671 for (fnhe = rcu_dereference(hash->chain); fnhe;
672 fnhe = rcu_dereference(fnhe->fnhe_next)) {
673 if (fnhe->fnhe_daddr == daddr)
674 break;
675 depth++;
676 }
677
678 if (fnhe) {
679 if (fnhe->fnhe_genid != genid)
680 fnhe->fnhe_genid = genid;
681 if (gw)
682 fnhe->fnhe_gw = gw;
683 if (pmtu) {
684 fnhe->fnhe_pmtu = pmtu;
685 fnhe->fnhe_mtu_locked = lock;
686 }
687 fnhe->fnhe_expires = max(1UL, expires);
688 /* Update all cached dsts too */
689 rt = rcu_dereference(fnhe->fnhe_rth_input);
690 if (rt)
691 fill_route_from_fnhe(rt, fnhe);
692 rt = rcu_dereference(fnhe->fnhe_rth_output);
693 if (rt)
694 fill_route_from_fnhe(rt, fnhe);
695 } else {
696 /* Randomize max depth to avoid some side channels attacks. */
697 int max_depth = FNHE_RECLAIM_DEPTH +
698 get_random_u32_below(FNHE_RECLAIM_DEPTH);
699
700 while (depth > max_depth) {
701 fnhe_remove_oldest(hash);
702 depth--;
703 }
704
705 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
706 if (!fnhe)
707 goto out_unlock;
708
709 fnhe->fnhe_next = hash->chain;
710
711 fnhe->fnhe_genid = genid;
712 fnhe->fnhe_daddr = daddr;
713 fnhe->fnhe_gw = gw;
714 fnhe->fnhe_pmtu = pmtu;
715 fnhe->fnhe_mtu_locked = lock;
716 fnhe->fnhe_expires = max(1UL, expires);
717
718 rcu_assign_pointer(hash->chain, fnhe);
719
720 /* Exception created; mark the cached routes for the nexthop
721 * stale, so anyone caching it rechecks if this exception
722 * applies to them.
723 */
724 rt = rcu_dereference(nhc->nhc_rth_input);
725 if (rt)
726 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
727
728 for_each_possible_cpu(i) {
729 struct rtable __rcu **prt;
730
731 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
732 rt = rcu_dereference(*prt);
733 if (rt)
734 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
735 }
736 }
737
738 fnhe->fnhe_stamp = jiffies;
739
740 out_unlock:
741 spin_unlock_bh(&fnhe_lock);
742 }
743
__ip_do_redirect(struct rtable * rt,struct sk_buff * skb,struct flowi4 * fl4,bool kill_route)744 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
745 bool kill_route)
746 {
747 __be32 new_gw = icmp_hdr(skb)->un.gateway;
748 __be32 old_gw = ip_hdr(skb)->saddr;
749 struct net_device *dev = skb->dev;
750 struct in_device *in_dev;
751 struct fib_result res;
752 struct neighbour *n;
753 struct net *net;
754
755 switch (icmp_hdr(skb)->code & 7) {
756 case ICMP_REDIR_NET:
757 case ICMP_REDIR_NETTOS:
758 case ICMP_REDIR_HOST:
759 case ICMP_REDIR_HOSTTOS:
760 break;
761
762 default:
763 return;
764 }
765
766 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
767 return;
768
769 in_dev = __in_dev_get_rcu(dev);
770 if (!in_dev)
771 return;
772
773 net = dev_net(dev);
774 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
775 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
776 ipv4_is_zeronet(new_gw))
777 goto reject_redirect;
778
779 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
780 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
781 goto reject_redirect;
782 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
783 goto reject_redirect;
784 } else {
785 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
786 goto reject_redirect;
787 }
788
789 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
790 if (!n)
791 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
792 if (!IS_ERR(n)) {
793 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
794 neigh_event_send(n, NULL);
795 } else {
796 if (fib_lookup(net, fl4, &res, 0) == 0) {
797 struct fib_nh_common *nhc;
798
799 fib_select_path(net, &res, fl4, skb);
800 nhc = FIB_RES_NHC(res);
801 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
802 0, false,
803 jiffies + ip_rt_gc_timeout);
804 }
805 if (kill_route)
806 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
807 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
808 }
809 neigh_release(n);
810 }
811 return;
812
813 reject_redirect:
814 #ifdef CONFIG_IP_ROUTE_VERBOSE
815 if (IN_DEV_LOG_MARTIANS(in_dev)) {
816 const struct iphdr *iph = (const struct iphdr *) skb->data;
817 __be32 daddr = iph->daddr;
818 __be32 saddr = iph->saddr;
819
820 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
821 " Advised path = %pI4 -> %pI4\n",
822 &old_gw, dev->name, &new_gw,
823 &saddr, &daddr);
824 }
825 #endif
826 ;
827 }
828
ip_do_redirect(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb)829 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
830 {
831 struct rtable *rt;
832 struct flowi4 fl4;
833 const struct iphdr *iph = (const struct iphdr *) skb->data;
834 struct net *net = dev_net(skb->dev);
835 int oif = skb->dev->ifindex;
836 u8 prot = iph->protocol;
837 u32 mark = skb->mark;
838 __u8 tos = iph->tos;
839
840 rt = dst_rtable(dst);
841
842 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
843 __ip_do_redirect(rt, skb, &fl4, true);
844 }
845
ipv4_negative_advice(struct sock * sk,struct dst_entry * dst)846 static void ipv4_negative_advice(struct sock *sk,
847 struct dst_entry *dst)
848 {
849 struct rtable *rt = dst_rtable(dst);
850
851 if ((READ_ONCE(dst->obsolete) > 0) ||
852 (rt->rt_flags & RTCF_REDIRECTED) ||
853 READ_ONCE(rt->dst.expires))
854 sk_dst_reset(sk);
855 }
856
857 /*
858 * Algorithm:
859 * 1. The first ip_rt_redirect_number redirects are sent
860 * with exponential backoff, then we stop sending them at all,
861 * assuming that the host ignores our redirects.
862 * 2. If we did not see packets requiring redirects
863 * during ip_rt_redirect_silence, we assume that the host
864 * forgot redirected route and start to send redirects again.
865 *
866 * This algorithm is much cheaper and more intelligent than dumb load limiting
867 * in icmp.c.
868 *
869 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
870 * and "frag. need" (breaks PMTU discovery) in icmp.c.
871 */
872
ip_rt_send_redirect(struct sk_buff * skb)873 void ip_rt_send_redirect(struct sk_buff *skb)
874 {
875 struct rtable *rt = skb_rtable(skb);
876 struct in_device *in_dev;
877 struct inet_peer *peer;
878 struct net *net;
879 int log_martians;
880 int vif;
881
882 rcu_read_lock();
883 in_dev = __in_dev_get_rcu(rt->dst.dev);
884 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
885 rcu_read_unlock();
886 return;
887 }
888 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
889 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
890
891 net = dev_net(rt->dst.dev);
892 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
893 if (!peer) {
894 rcu_read_unlock();
895 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
896 rt_nexthop(rt, ip_hdr(skb)->daddr));
897 return;
898 }
899
900 /* No redirected packets during ip_rt_redirect_silence;
901 * reset the algorithm.
902 */
903 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
904 peer->rate_tokens = 0;
905 peer->n_redirects = 0;
906 }
907
908 /* Too many ignored redirects; do not send anything
909 * set dst.rate_last to the last seen redirected packet.
910 */
911 if (peer->n_redirects >= ip_rt_redirect_number) {
912 peer->rate_last = jiffies;
913 goto out_unlock;
914 }
915
916 /* Check for load limit; set rate_last to the latest sent
917 * redirect.
918 */
919 if (peer->n_redirects == 0 ||
920 time_after(jiffies,
921 (peer->rate_last +
922 (ip_rt_redirect_load << peer->n_redirects)))) {
923 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
924
925 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
926 peer->rate_last = jiffies;
927 ++peer->n_redirects;
928 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
929 peer->n_redirects == ip_rt_redirect_number)
930 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
931 &ip_hdr(skb)->saddr, inet_iif(skb),
932 &ip_hdr(skb)->daddr, &gw);
933 }
934 out_unlock:
935 rcu_read_unlock();
936 }
937
ip_error(struct sk_buff * skb)938 static int ip_error(struct sk_buff *skb)
939 {
940 struct rtable *rt = skb_rtable(skb);
941 struct net_device *dev = skb->dev;
942 struct in_device *in_dev;
943 struct inet_peer *peer;
944 unsigned long now;
945 struct net *net;
946 SKB_DR(reason);
947 bool send;
948 int code;
949
950 if (netif_is_l3_master(skb->dev)) {
951 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
952 if (!dev)
953 goto out;
954 }
955
956 in_dev = __in_dev_get_rcu(dev);
957
958 /* IP on this device is disabled. */
959 if (!in_dev)
960 goto out;
961
962 net = dev_net(rt->dst.dev);
963 if (!IN_DEV_FORWARD(in_dev)) {
964 switch (rt->dst.error) {
965 case EHOSTUNREACH:
966 SKB_DR_SET(reason, IP_INADDRERRORS);
967 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
968 break;
969
970 case ENETUNREACH:
971 SKB_DR_SET(reason, IP_INNOROUTES);
972 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
973 break;
974 }
975 goto out;
976 }
977
978 switch (rt->dst.error) {
979 case EINVAL:
980 default:
981 goto out;
982 case EHOSTUNREACH:
983 code = ICMP_HOST_UNREACH;
984 break;
985 case ENETUNREACH:
986 code = ICMP_NET_UNREACH;
987 SKB_DR_SET(reason, IP_INNOROUTES);
988 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
989 break;
990 case EACCES:
991 code = ICMP_PKT_FILTERED;
992 break;
993 }
994
995 rcu_read_lock();
996 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
997 l3mdev_master_ifindex_rcu(skb->dev));
998 send = true;
999 if (peer) {
1000 now = jiffies;
1001 peer->rate_tokens += now - peer->rate_last;
1002 if (peer->rate_tokens > ip_rt_error_burst)
1003 peer->rate_tokens = ip_rt_error_burst;
1004 peer->rate_last = now;
1005 if (peer->rate_tokens >= ip_rt_error_cost)
1006 peer->rate_tokens -= ip_rt_error_cost;
1007 else
1008 send = false;
1009 }
1010 rcu_read_unlock();
1011
1012 if (send)
1013 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1014
1015 out: kfree_skb_reason(skb, reason);
1016 return 0;
1017 }
1018
__ip_rt_update_pmtu(struct rtable * rt,struct flowi4 * fl4,u32 mtu)1019 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1020 {
1021 struct dst_entry *dst = &rt->dst;
1022 struct fib_result res;
1023 bool lock = false;
1024 struct net *net;
1025 u32 old_mtu;
1026
1027 if (ip_mtu_locked(dst))
1028 return;
1029
1030 old_mtu = ipv4_mtu(dst);
1031 if (old_mtu < mtu)
1032 return;
1033
1034 rcu_read_lock();
1035 net = dst_dev_net_rcu(dst);
1036 if (mtu < net->ipv4.ip_rt_min_pmtu) {
1037 lock = true;
1038 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1039 }
1040
1041 if (rt->rt_pmtu == mtu && !lock &&
1042 time_before(jiffies, READ_ONCE(dst->expires) -
1043 net->ipv4.ip_rt_mtu_expires / 2))
1044 goto out;
1045
1046 if (fib_lookup(net, fl4, &res, 0) == 0) {
1047 struct fib_nh_common *nhc;
1048
1049 fib_select_path(net, &res, fl4, NULL);
1050 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1051 if (fib_info_num_path(res.fi) > 1) {
1052 int nhsel;
1053
1054 for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1055 nhc = fib_info_nhc(res.fi, nhsel);
1056 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1057 jiffies + net->ipv4.ip_rt_mtu_expires);
1058 }
1059 goto out;
1060 }
1061 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1062 nhc = FIB_RES_NHC(res);
1063 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1064 jiffies + net->ipv4.ip_rt_mtu_expires);
1065 }
1066 out:
1067 rcu_read_unlock();
1068 }
1069
ip_rt_update_pmtu(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb,u32 mtu,bool confirm_neigh)1070 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1071 struct sk_buff *skb, u32 mtu,
1072 bool confirm_neigh)
1073 {
1074 struct rtable *rt = dst_rtable(dst);
1075 struct flowi4 fl4;
1076
1077 ip_rt_build_flow_key(&fl4, sk, skb);
1078
1079 /* Don't make lookup fail for bridged encapsulations */
1080 if (skb && netif_is_any_bridge_port(skb->dev))
1081 fl4.flowi4_oif = 0;
1082
1083 __ip_rt_update_pmtu(rt, &fl4, mtu);
1084 }
1085
ipv4_update_pmtu(struct sk_buff * skb,struct net * net,u32 mtu,int oif,u8 protocol)1086 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1087 int oif, u8 protocol)
1088 {
1089 const struct iphdr *iph = (const struct iphdr *)skb->data;
1090 struct flowi4 fl4;
1091 struct rtable *rt;
1092 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1093
1094 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1095 0);
1096 rt = __ip_route_output_key(net, &fl4);
1097 if (!IS_ERR(rt)) {
1098 __ip_rt_update_pmtu(rt, &fl4, mtu);
1099 ip_rt_put(rt);
1100 }
1101 }
1102 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1103
__ipv4_sk_update_pmtu(struct sk_buff * skb,struct sock * sk,u32 mtu)1104 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1105 {
1106 const struct iphdr *iph = (const struct iphdr *)skb->data;
1107 struct flowi4 fl4;
1108 struct rtable *rt;
1109
1110 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1111
1112 if (!fl4.flowi4_mark)
1113 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1114
1115 rt = __ip_route_output_key(sock_net(sk), &fl4);
1116 if (!IS_ERR(rt)) {
1117 __ip_rt_update_pmtu(rt, &fl4, mtu);
1118 ip_rt_put(rt);
1119 }
1120 }
1121
ipv4_sk_update_pmtu(struct sk_buff * skb,struct sock * sk,u32 mtu)1122 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1123 {
1124 const struct iphdr *iph = (const struct iphdr *)skb->data;
1125 struct flowi4 fl4;
1126 struct rtable *rt;
1127 struct dst_entry *odst = NULL;
1128 bool new = false;
1129 struct net *net = sock_net(sk);
1130
1131 bh_lock_sock(sk);
1132
1133 if (!ip_sk_accept_pmtu(sk))
1134 goto out;
1135
1136 odst = sk_dst_get(sk);
1137
1138 if (sock_owned_by_user(sk) || !odst) {
1139 __ipv4_sk_update_pmtu(skb, sk, mtu);
1140 goto out;
1141 }
1142
1143 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1144
1145 rt = dst_rtable(odst);
1146 if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) {
1147 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1148 if (IS_ERR(rt))
1149 goto out;
1150
1151 new = true;
1152 }
1153
1154 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1155
1156 if (!dst_check(&rt->dst, 0)) {
1157 if (new)
1158 dst_release(&rt->dst);
1159
1160 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1161 if (IS_ERR(rt))
1162 goto out;
1163
1164 new = true;
1165 }
1166
1167 if (new)
1168 sk_dst_set(sk, &rt->dst);
1169
1170 out:
1171 bh_unlock_sock(sk);
1172 dst_release(odst);
1173 }
1174 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1175
ipv4_redirect(struct sk_buff * skb,struct net * net,int oif,u8 protocol)1176 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1177 int oif, u8 protocol)
1178 {
1179 const struct iphdr *iph = (const struct iphdr *)skb->data;
1180 struct flowi4 fl4;
1181 struct rtable *rt;
1182
1183 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1184 rt = __ip_route_output_key(net, &fl4);
1185 if (!IS_ERR(rt)) {
1186 __ip_do_redirect(rt, skb, &fl4, false);
1187 ip_rt_put(rt);
1188 }
1189 }
1190 EXPORT_SYMBOL_GPL(ipv4_redirect);
1191
ipv4_sk_redirect(struct sk_buff * skb,struct sock * sk)1192 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1193 {
1194 const struct iphdr *iph = (const struct iphdr *)skb->data;
1195 struct flowi4 fl4;
1196 struct rtable *rt;
1197 struct net *net = sock_net(sk);
1198
1199 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1200 rt = __ip_route_output_key(net, &fl4);
1201 if (!IS_ERR(rt)) {
1202 __ip_do_redirect(rt, skb, &fl4, false);
1203 ip_rt_put(rt);
1204 }
1205 }
1206 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1207
ipv4_dst_check(struct dst_entry * dst,u32 cookie)1208 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1209 u32 cookie)
1210 {
1211 struct rtable *rt = dst_rtable(dst);
1212
1213 /* All IPV4 dsts are created with ->obsolete set to the value
1214 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1215 * into this function always.
1216 *
1217 * When a PMTU/redirect information update invalidates a route,
1218 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1219 * DST_OBSOLETE_DEAD.
1220 */
1221 if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK ||
1222 rt_is_expired(rt))
1223 return NULL;
1224 return dst;
1225 }
1226 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1227
ipv4_send_dest_unreach(struct sk_buff * skb)1228 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1229 {
1230 struct inet_skb_parm parm;
1231 struct net_device *dev;
1232 int res;
1233
1234 /* Recompile ip options since IPCB may not be valid anymore.
1235 * Also check we have a reasonable ipv4 header.
1236 */
1237 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1238 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1239 return;
1240
1241 memset(&parm, 0, sizeof(parm));
1242 if (ip_hdr(skb)->ihl > 5) {
1243 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1244 return;
1245 parm.opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1246
1247 rcu_read_lock();
1248 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1249 res = __ip_options_compile(dev_net(dev), &parm.opt, skb, NULL);
1250 rcu_read_unlock();
1251
1252 if (res)
1253 return;
1254 }
1255 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &parm);
1256 }
1257
ipv4_link_failure(struct sk_buff * skb)1258 static void ipv4_link_failure(struct sk_buff *skb)
1259 {
1260 struct rtable *rt;
1261
1262 ipv4_send_dest_unreach(skb);
1263
1264 rt = skb_rtable(skb);
1265 if (rt)
1266 dst_set_expires(&rt->dst, 0);
1267 }
1268
ip_rt_bug(struct net * net,struct sock * sk,struct sk_buff * skb)1269 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1270 {
1271 pr_debug("%s: %pI4 -> %pI4, %s\n",
1272 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1273 skb->dev ? skb->dev->name : "?");
1274 kfree_skb(skb);
1275 WARN_ON(1);
1276 return 0;
1277 }
1278
1279 /*
1280 * We do not cache source address of outgoing interface,
1281 * because it is used only by IP RR, TS and SRR options,
1282 * so that it out of fast path.
1283 *
1284 * BTW remember: "addr" is allowed to be not aligned
1285 * in IP options!
1286 */
1287
ip_rt_get_source(u8 * addr,struct sk_buff * skb,struct rtable * rt)1288 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1289 {
1290 __be32 src;
1291
1292 if (rt_is_output_route(rt))
1293 src = ip_hdr(skb)->saddr;
1294 else {
1295 struct fib_result res;
1296 struct iphdr *iph = ip_hdr(skb);
1297 struct flowi4 fl4 = {
1298 .daddr = iph->daddr,
1299 .saddr = iph->saddr,
1300 .flowi4_dscp = ip4h_dscp(iph),
1301 .flowi4_oif = rt->dst.dev->ifindex,
1302 .flowi4_iif = skb->dev->ifindex,
1303 .flowi4_mark = skb->mark,
1304 };
1305
1306 rcu_read_lock();
1307 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1308 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1309 else
1310 src = inet_select_addr(rt->dst.dev,
1311 rt_nexthop(rt, iph->daddr),
1312 RT_SCOPE_UNIVERSE);
1313 rcu_read_unlock();
1314 }
1315 memcpy(addr, &src, 4);
1316 }
1317
1318 #ifdef CONFIG_IP_ROUTE_CLASSID
set_class_tag(struct rtable * rt,u32 tag)1319 static void set_class_tag(struct rtable *rt, u32 tag)
1320 {
1321 if (!(rt->dst.tclassid & 0xFFFF))
1322 rt->dst.tclassid |= tag & 0xFFFF;
1323 if (!(rt->dst.tclassid & 0xFFFF0000))
1324 rt->dst.tclassid |= tag & 0xFFFF0000;
1325 }
1326 #endif
1327
ipv4_default_advmss(const struct dst_entry * dst)1328 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1329 {
1330 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1331 unsigned int advmss;
1332 struct net *net;
1333
1334 rcu_read_lock();
1335 net = dst_dev_net_rcu(dst);
1336 advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1337 net->ipv4.ip_rt_min_advmss);
1338 rcu_read_unlock();
1339
1340 return min(advmss, IPV4_MAX_PMTU - header_size);
1341 }
1342
ipv4_mtu(const struct dst_entry * dst)1343 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1344 {
1345 return ip_dst_mtu_maybe_forward(dst, false);
1346 }
1347 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1348
ip_del_fnhe(struct fib_nh_common * nhc,__be32 daddr)1349 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1350 {
1351 struct fnhe_hash_bucket *hash;
1352 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1353 u32 hval = fnhe_hashfun(daddr);
1354
1355 spin_lock_bh(&fnhe_lock);
1356
1357 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1358 lockdep_is_held(&fnhe_lock));
1359 hash += hval;
1360
1361 fnhe_p = &hash->chain;
1362 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1363 while (fnhe) {
1364 if (fnhe->fnhe_daddr == daddr) {
1365 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1366 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1367 /* set fnhe_daddr to 0 to ensure it won't bind with
1368 * new dsts in rt_bind_exception().
1369 */
1370 fnhe->fnhe_daddr = 0;
1371 fnhe_flush_routes(fnhe);
1372 kfree_rcu(fnhe, rcu);
1373 break;
1374 }
1375 fnhe_p = &fnhe->fnhe_next;
1376 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1377 lockdep_is_held(&fnhe_lock));
1378 }
1379
1380 spin_unlock_bh(&fnhe_lock);
1381 }
1382
find_exception(struct fib_nh_common * nhc,__be32 daddr)1383 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1384 __be32 daddr)
1385 {
1386 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1387 struct fib_nh_exception *fnhe;
1388 u32 hval;
1389
1390 if (!hash)
1391 return NULL;
1392
1393 hval = fnhe_hashfun(daddr);
1394
1395 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1396 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1397 if (fnhe->fnhe_daddr == daddr) {
1398 if (fnhe->fnhe_expires &&
1399 time_after(jiffies, fnhe->fnhe_expires)) {
1400 ip_del_fnhe(nhc, daddr);
1401 break;
1402 }
1403 return fnhe;
1404 }
1405 }
1406 return NULL;
1407 }
1408
1409 /* MTU selection:
1410 * 1. mtu on route is locked - use it
1411 * 2. mtu from nexthop exception
1412 * 3. mtu from egress device
1413 */
1414
ip_mtu_from_fib_result(struct fib_result * res,__be32 daddr)1415 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1416 {
1417 struct fib_nh_common *nhc = res->nhc;
1418 struct net_device *dev = nhc->nhc_dev;
1419 struct fib_info *fi = res->fi;
1420 u32 mtu = 0;
1421
1422 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1423 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1424 mtu = fi->fib_mtu;
1425
1426 if (likely(!mtu)) {
1427 struct fib_nh_exception *fnhe;
1428
1429 fnhe = find_exception(nhc, daddr);
1430 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1431 mtu = fnhe->fnhe_pmtu;
1432 }
1433
1434 if (likely(!mtu))
1435 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1436
1437 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1438 }
1439
rt_bind_exception(struct rtable * rt,struct fib_nh_exception * fnhe,__be32 daddr,const bool do_cache)1440 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1441 __be32 daddr, const bool do_cache)
1442 {
1443 bool ret = false;
1444
1445 spin_lock_bh(&fnhe_lock);
1446
1447 if (daddr == fnhe->fnhe_daddr) {
1448 struct rtable __rcu **porig;
1449 struct rtable *orig;
1450 int genid = fnhe_genid(dev_net(rt->dst.dev));
1451
1452 if (rt_is_input_route(rt))
1453 porig = &fnhe->fnhe_rth_input;
1454 else
1455 porig = &fnhe->fnhe_rth_output;
1456 orig = rcu_dereference(*porig);
1457
1458 if (fnhe->fnhe_genid != genid) {
1459 fnhe->fnhe_genid = genid;
1460 fnhe->fnhe_gw = 0;
1461 fnhe->fnhe_pmtu = 0;
1462 fnhe->fnhe_expires = 0;
1463 fnhe->fnhe_mtu_locked = false;
1464 fnhe_flush_routes(fnhe);
1465 orig = NULL;
1466 }
1467 fill_route_from_fnhe(rt, fnhe);
1468 if (!rt->rt_gw4) {
1469 rt->rt_gw4 = daddr;
1470 rt->rt_gw_family = AF_INET;
1471 }
1472
1473 if (do_cache) {
1474 dst_hold(&rt->dst);
1475 rcu_assign_pointer(*porig, rt);
1476 if (orig) {
1477 dst_dev_put(&orig->dst);
1478 dst_release(&orig->dst);
1479 }
1480 ret = true;
1481 }
1482
1483 fnhe->fnhe_stamp = jiffies;
1484 }
1485 spin_unlock_bh(&fnhe_lock);
1486
1487 return ret;
1488 }
1489
rt_cache_route(struct fib_nh_common * nhc,struct rtable * rt)1490 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1491 {
1492 struct rtable *orig, *prev, **p;
1493 bool ret = true;
1494
1495 if (rt_is_input_route(rt)) {
1496 p = (struct rtable **)&nhc->nhc_rth_input;
1497 } else {
1498 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1499 }
1500 orig = *p;
1501
1502 /* hold dst before doing cmpxchg() to avoid race condition
1503 * on this dst
1504 */
1505 dst_hold(&rt->dst);
1506 prev = cmpxchg(p, orig, rt);
1507 if (prev == orig) {
1508 if (orig) {
1509 rt_add_uncached_list(orig);
1510 dst_release(&orig->dst);
1511 }
1512 } else {
1513 dst_release(&rt->dst);
1514 ret = false;
1515 }
1516
1517 return ret;
1518 }
1519
1520 struct uncached_list {
1521 spinlock_t lock;
1522 struct list_head head;
1523 };
1524
1525 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1526
rt_add_uncached_list(struct rtable * rt)1527 void rt_add_uncached_list(struct rtable *rt)
1528 {
1529 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1530
1531 rt->dst.rt_uncached_list = ul;
1532
1533 spin_lock_bh(&ul->lock);
1534 list_add_tail(&rt->dst.rt_uncached, &ul->head);
1535 spin_unlock_bh(&ul->lock);
1536 }
1537
rt_del_uncached_list(struct rtable * rt)1538 void rt_del_uncached_list(struct rtable *rt)
1539 {
1540 if (!list_empty(&rt->dst.rt_uncached)) {
1541 struct uncached_list *ul = rt->dst.rt_uncached_list;
1542
1543 spin_lock_bh(&ul->lock);
1544 list_del_init(&rt->dst.rt_uncached);
1545 spin_unlock_bh(&ul->lock);
1546 }
1547 }
1548
ipv4_dst_destroy(struct dst_entry * dst)1549 static void ipv4_dst_destroy(struct dst_entry *dst)
1550 {
1551 ip_dst_metrics_put(dst);
1552 rt_del_uncached_list(dst_rtable(dst));
1553 }
1554
rt_flush_dev(struct net_device * dev)1555 void rt_flush_dev(struct net_device *dev)
1556 {
1557 struct rtable *rt, *safe;
1558 int cpu;
1559
1560 for_each_possible_cpu(cpu) {
1561 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1562
1563 if (list_empty(&ul->head))
1564 continue;
1565
1566 spin_lock_bh(&ul->lock);
1567 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1568 if (rt->dst.dev != dev)
1569 continue;
1570 rt->dst.dev = blackhole_netdev;
1571 netdev_ref_replace(dev, blackhole_netdev,
1572 &rt->dst.dev_tracker, GFP_ATOMIC);
1573 list_del_init(&rt->dst.rt_uncached);
1574 }
1575 spin_unlock_bh(&ul->lock);
1576 }
1577 }
1578
rt_cache_valid(const struct rtable * rt)1579 static bool rt_cache_valid(const struct rtable *rt)
1580 {
1581 return rt &&
1582 READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK &&
1583 !rt_is_expired(rt);
1584 }
1585
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)1586 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1587 const struct fib_result *res,
1588 struct fib_nh_exception *fnhe,
1589 struct fib_info *fi, u16 type, u32 itag,
1590 const bool do_cache)
1591 {
1592 bool cached = false;
1593
1594 if (fi) {
1595 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1596
1597 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1598 rt->rt_uses_gateway = 1;
1599 rt->rt_gw_family = nhc->nhc_gw_family;
1600 /* only INET and INET6 are supported */
1601 if (likely(nhc->nhc_gw_family == AF_INET))
1602 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1603 else
1604 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1605 }
1606
1607 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1608
1609 #ifdef CONFIG_IP_ROUTE_CLASSID
1610 if (nhc->nhc_family == AF_INET) {
1611 struct fib_nh *nh;
1612
1613 nh = container_of(nhc, struct fib_nh, nh_common);
1614 rt->dst.tclassid = nh->nh_tclassid;
1615 }
1616 #endif
1617 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1618 if (unlikely(fnhe))
1619 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1620 else if (do_cache)
1621 cached = rt_cache_route(nhc, rt);
1622 if (unlikely(!cached)) {
1623 /* Routes we intend to cache in nexthop exception or
1624 * FIB nexthop have the DST_NOCACHE bit clear.
1625 * However, if we are unsuccessful at storing this
1626 * route into the cache we really need to set it.
1627 */
1628 if (!rt->rt_gw4) {
1629 rt->rt_gw_family = AF_INET;
1630 rt->rt_gw4 = daddr;
1631 }
1632 rt_add_uncached_list(rt);
1633 }
1634 } else
1635 rt_add_uncached_list(rt);
1636
1637 #ifdef CONFIG_IP_ROUTE_CLASSID
1638 #ifdef CONFIG_IP_MULTIPLE_TABLES
1639 set_class_tag(rt, res->tclassid);
1640 #endif
1641 set_class_tag(rt, itag);
1642 #endif
1643 }
1644
rt_dst_alloc(struct net_device * dev,unsigned int flags,u16 type,bool noxfrm)1645 struct rtable *rt_dst_alloc(struct net_device *dev,
1646 unsigned int flags, u16 type,
1647 bool noxfrm)
1648 {
1649 struct rtable *rt;
1650
1651 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1652 (noxfrm ? DST_NOXFRM : 0));
1653
1654 if (rt) {
1655 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1656 rt->rt_flags = flags;
1657 rt->rt_type = type;
1658 rt->rt_is_input = 0;
1659 rt->rt_iif = 0;
1660 rt->rt_pmtu = 0;
1661 rt->rt_mtu_locked = 0;
1662 rt->rt_uses_gateway = 0;
1663 rt->rt_gw_family = 0;
1664 rt->rt_gw4 = 0;
1665
1666 rt->dst.output = ip_output;
1667 if (flags & RTCF_LOCAL)
1668 rt->dst.input = ip_local_deliver;
1669 }
1670
1671 return rt;
1672 }
1673 EXPORT_SYMBOL(rt_dst_alloc);
1674
rt_dst_clone(struct net_device * dev,struct rtable * rt)1675 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1676 {
1677 struct rtable *new_rt;
1678
1679 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1680 rt->dst.flags);
1681
1682 if (new_rt) {
1683 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1684 new_rt->rt_flags = rt->rt_flags;
1685 new_rt->rt_type = rt->rt_type;
1686 new_rt->rt_is_input = rt->rt_is_input;
1687 new_rt->rt_iif = rt->rt_iif;
1688 new_rt->rt_pmtu = rt->rt_pmtu;
1689 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1690 new_rt->rt_gw_family = rt->rt_gw_family;
1691 if (rt->rt_gw_family == AF_INET)
1692 new_rt->rt_gw4 = rt->rt_gw4;
1693 else if (rt->rt_gw_family == AF_INET6)
1694 new_rt->rt_gw6 = rt->rt_gw6;
1695
1696 new_rt->dst.input = READ_ONCE(rt->dst.input);
1697 new_rt->dst.output = READ_ONCE(rt->dst.output);
1698 new_rt->dst.error = rt->dst.error;
1699 new_rt->dst.lastuse = jiffies;
1700 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1701 }
1702 return new_rt;
1703 }
1704 EXPORT_SYMBOL(rt_dst_clone);
1705
1706 /* called in rcu_read_lock() section */
1707 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)1708 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1709 dscp_t dscp, struct net_device *dev,
1710 struct in_device *in_dev, u32 *itag)
1711 {
1712 enum skb_drop_reason reason;
1713
1714 /* Primary sanity checks. */
1715 if (!in_dev)
1716 return SKB_DROP_REASON_NOT_SPECIFIED;
1717
1718 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1719 return SKB_DROP_REASON_IP_INVALID_SOURCE;
1720
1721 if (skb->protocol != htons(ETH_P_IP))
1722 return SKB_DROP_REASON_INVALID_PROTO;
1723
1724 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1725 return SKB_DROP_REASON_IP_LOCALNET;
1726
1727 if (ipv4_is_zeronet(saddr)) {
1728 if (!ipv4_is_local_multicast(daddr) &&
1729 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1730 return SKB_DROP_REASON_IP_INVALID_SOURCE;
1731 } else {
1732 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1733 dev, in_dev, itag);
1734 if (reason)
1735 return reason;
1736 }
1737 return SKB_NOT_DROPPED_YET;
1738 }
1739
1740 /* called in rcu_read_lock() section */
1741 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)1742 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1743 dscp_t dscp, struct net_device *dev, int our)
1744 {
1745 struct in_device *in_dev = __in_dev_get_rcu(dev);
1746 unsigned int flags = RTCF_MULTICAST;
1747 enum skb_drop_reason reason;
1748 struct rtable *rth;
1749 u32 itag = 0;
1750
1751 reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1752 &itag);
1753 if (reason)
1754 return reason;
1755
1756 if (our)
1757 flags |= RTCF_LOCAL;
1758
1759 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1760 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1761
1762 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1763 false);
1764 if (!rth)
1765 return SKB_DROP_REASON_NOMEM;
1766
1767 #ifdef CONFIG_IP_ROUTE_CLASSID
1768 rth->dst.tclassid = itag;
1769 #endif
1770 rth->dst.output = ip_rt_bug;
1771 rth->rt_is_input= 1;
1772
1773 #ifdef CONFIG_IP_MROUTE
1774 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1775 rth->dst.input = ip_mr_input;
1776 #endif
1777 RT_CACHE_STAT_INC(in_slow_mc);
1778
1779 skb_dst_drop(skb);
1780 skb_dst_set(skb, &rth->dst);
1781 return SKB_NOT_DROPPED_YET;
1782 }
1783
1784
ip_handle_martian_source(struct net_device * dev,struct in_device * in_dev,struct sk_buff * skb,__be32 daddr,__be32 saddr)1785 static void ip_handle_martian_source(struct net_device *dev,
1786 struct in_device *in_dev,
1787 struct sk_buff *skb,
1788 __be32 daddr,
1789 __be32 saddr)
1790 {
1791 RT_CACHE_STAT_INC(in_martian_src);
1792 #ifdef CONFIG_IP_ROUTE_VERBOSE
1793 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1794 /*
1795 * RFC1812 recommendation, if source is martian,
1796 * the only hint is MAC header.
1797 */
1798 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1799 &daddr, &saddr, dev->name);
1800 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1801 print_hex_dump(KERN_WARNING, "ll header: ",
1802 DUMP_PREFIX_OFFSET, 16, 1,
1803 skb_mac_header(skb),
1804 dev->hard_header_len, false);
1805 }
1806 }
1807 #endif
1808 }
1809
1810 /* called in rcu_read_lock() section */
1811 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)1812 __mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1813 struct in_device *in_dev, __be32 daddr,
1814 __be32 saddr, dscp_t dscp)
1815 {
1816 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1817 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1818 struct net_device *dev = nhc->nhc_dev;
1819 struct fib_nh_exception *fnhe;
1820 struct rtable *rth;
1821 int err;
1822 struct in_device *out_dev;
1823 bool do_cache;
1824 u32 itag = 0;
1825
1826 /* get a working reference to the output device */
1827 out_dev = __in_dev_get_rcu(dev);
1828 if (!out_dev) {
1829 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1830 return reason;
1831 }
1832
1833 err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1834 in_dev->dev, in_dev, &itag);
1835 if (err < 0) {
1836 reason = -err;
1837 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1838 saddr);
1839
1840 goto cleanup;
1841 }
1842
1843 do_cache = res->fi && !itag;
1844 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1845 skb->protocol == htons(ETH_P_IP)) {
1846 __be32 gw;
1847
1848 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1849 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1850 inet_addr_onlink(out_dev, saddr, gw))
1851 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1852 }
1853
1854 if (skb->protocol != htons(ETH_P_IP)) {
1855 /* Not IP (i.e. ARP). Do not create route, if it is
1856 * invalid for proxy arp. DNAT routes are always valid.
1857 *
1858 * Proxy arp feature have been extended to allow, ARP
1859 * replies back to the same interface, to support
1860 * Private VLAN switch technologies. See arp.c.
1861 */
1862 if (out_dev == in_dev &&
1863 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1864 reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1865 goto cleanup;
1866 }
1867 }
1868
1869 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1870 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1871
1872 fnhe = find_exception(nhc, daddr);
1873 if (do_cache) {
1874 if (fnhe)
1875 rth = rcu_dereference(fnhe->fnhe_rth_input);
1876 else
1877 rth = rcu_dereference(nhc->nhc_rth_input);
1878 if (rt_cache_valid(rth)) {
1879 skb_dst_set_noref(skb, &rth->dst);
1880 goto out;
1881 }
1882 }
1883
1884 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1885 IN_DEV_ORCONF(out_dev, NOXFRM));
1886 if (!rth) {
1887 reason = SKB_DROP_REASON_NOMEM;
1888 goto cleanup;
1889 }
1890
1891 rth->rt_is_input = 1;
1892 RT_CACHE_STAT_INC(in_slow_tot);
1893
1894 rth->dst.input = ip_forward;
1895
1896 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1897 do_cache);
1898 lwtunnel_set_redirect(&rth->dst);
1899 skb_dst_set(skb, &rth->dst);
1900 out:
1901 reason = SKB_NOT_DROPPED_YET;
1902 cleanup:
1903 return reason;
1904 }
1905
1906 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1907 /* To make ICMP packets follow the right flow, the multipath hash is
1908 * calculated from the inner IP addresses.
1909 */
ip_multipath_l3_keys(const struct sk_buff * skb,struct flow_keys * hash_keys)1910 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1911 struct flow_keys *hash_keys)
1912 {
1913 const struct iphdr *outer_iph = ip_hdr(skb);
1914 const struct iphdr *key_iph = outer_iph;
1915 const struct iphdr *inner_iph;
1916 const struct icmphdr *icmph;
1917 struct iphdr _inner_iph;
1918 struct icmphdr _icmph;
1919
1920 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1921 goto out;
1922
1923 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1924 goto out;
1925
1926 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1927 &_icmph);
1928 if (!icmph)
1929 goto out;
1930
1931 if (!icmp_is_err(icmph->type))
1932 goto out;
1933
1934 inner_iph = skb_header_pointer(skb,
1935 outer_iph->ihl * 4 + sizeof(_icmph),
1936 sizeof(_inner_iph), &_inner_iph);
1937 if (!inner_iph)
1938 goto out;
1939
1940 key_iph = inner_iph;
1941 out:
1942 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1943 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1944 }
1945
fib_multipath_custom_hash_outer(const struct net * net,const struct sk_buff * skb,bool * p_has_inner)1946 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1947 const struct sk_buff *skb,
1948 bool *p_has_inner)
1949 {
1950 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1951 struct flow_keys keys, hash_keys;
1952
1953 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1954 return 0;
1955
1956 memset(&hash_keys, 0, sizeof(hash_keys));
1957 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1958
1959 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1960 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1961 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1962 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1963 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1964 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1965 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1966 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1967 hash_keys.ports.src = keys.ports.src;
1968 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1969 hash_keys.ports.dst = keys.ports.dst;
1970
1971 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1972 return fib_multipath_hash_from_keys(net, &hash_keys);
1973 }
1974
fib_multipath_custom_hash_inner(const struct net * net,const struct sk_buff * skb,bool has_inner)1975 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1976 const struct sk_buff *skb,
1977 bool has_inner)
1978 {
1979 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1980 struct flow_keys keys, hash_keys;
1981
1982 /* We assume the packet carries an encapsulation, but if none was
1983 * encountered during dissection of the outer flow, then there is no
1984 * point in calling the flow dissector again.
1985 */
1986 if (!has_inner)
1987 return 0;
1988
1989 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1990 return 0;
1991
1992 memset(&hash_keys, 0, sizeof(hash_keys));
1993 skb_flow_dissect_flow_keys(skb, &keys, 0);
1994
1995 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1996 return 0;
1997
1998 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1999 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2000 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2001 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2002 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2003 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2004 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2005 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2006 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2007 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2008 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2009 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2010 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2011 hash_keys.tags.flow_label = keys.tags.flow_label;
2012 }
2013
2014 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2015 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2016 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2017 hash_keys.ports.src = keys.ports.src;
2018 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2019 hash_keys.ports.dst = keys.ports.dst;
2020
2021 return fib_multipath_hash_from_keys(net, &hash_keys);
2022 }
2023
fib_multipath_custom_hash_skb(const struct net * net,const struct sk_buff * skb)2024 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2025 const struct sk_buff *skb)
2026 {
2027 u32 mhash, mhash_inner;
2028 bool has_inner = true;
2029
2030 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2031 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2032
2033 return jhash_2words(mhash, mhash_inner, 0);
2034 }
2035
fib_multipath_custom_hash_fl4(const struct net * net,const struct flowi4 * fl4)2036 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2037 const struct flowi4 *fl4)
2038 {
2039 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2040 struct flow_keys hash_keys;
2041
2042 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2043 return 0;
2044
2045 memset(&hash_keys, 0, sizeof(hash_keys));
2046 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2047 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2048 hash_keys.addrs.v4addrs.src = fl4->saddr;
2049 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2050 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2051 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2052 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2053 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2054 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2055 hash_keys.ports.src = (__force __be16)get_random_u16();
2056 else
2057 hash_keys.ports.src = fl4->fl4_sport;
2058 }
2059 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2060 hash_keys.ports.dst = fl4->fl4_dport;
2061
2062 return fib_multipath_hash_from_keys(net, &hash_keys);
2063 }
2064
2065 /* 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)2066 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2067 const struct sk_buff *skb, struct flow_keys *flkeys)
2068 {
2069 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2070 struct flow_keys hash_keys;
2071 u32 mhash = 0;
2072
2073 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2074 case 0:
2075 memset(&hash_keys, 0, sizeof(hash_keys));
2076 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2077 if (skb) {
2078 ip_multipath_l3_keys(skb, &hash_keys);
2079 } else {
2080 hash_keys.addrs.v4addrs.src = fl4->saddr;
2081 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2082 }
2083 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2084 break;
2085 case 1:
2086 /* skb is currently provided only when forwarding */
2087 if (skb) {
2088 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2089 struct flow_keys keys;
2090
2091 /* short-circuit if we already have L4 hash present */
2092 if (skb->l4_hash)
2093 return skb_get_hash_raw(skb) >> 1;
2094
2095 memset(&hash_keys, 0, sizeof(hash_keys));
2096
2097 if (!flkeys) {
2098 skb_flow_dissect_flow_keys(skb, &keys, flag);
2099 flkeys = &keys;
2100 }
2101
2102 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2103 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2104 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2105 hash_keys.ports.src = flkeys->ports.src;
2106 hash_keys.ports.dst = flkeys->ports.dst;
2107 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2108 } else {
2109 memset(&hash_keys, 0, sizeof(hash_keys));
2110 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2111 hash_keys.addrs.v4addrs.src = fl4->saddr;
2112 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2113 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2114 hash_keys.ports.src = (__force __be16)get_random_u16();
2115 else
2116 hash_keys.ports.src = fl4->fl4_sport;
2117 hash_keys.ports.dst = fl4->fl4_dport;
2118 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2119 }
2120 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2121 break;
2122 case 2:
2123 memset(&hash_keys, 0, sizeof(hash_keys));
2124 /* skb is currently provided only when forwarding */
2125 if (skb) {
2126 struct flow_keys keys;
2127
2128 skb_flow_dissect_flow_keys(skb, &keys, 0);
2129 /* Inner can be v4 or v6 */
2130 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2131 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2132 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2133 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2134 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2135 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2136 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2137 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2138 hash_keys.tags.flow_label = keys.tags.flow_label;
2139 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2140 } else {
2141 /* Same as case 0 */
2142 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2143 ip_multipath_l3_keys(skb, &hash_keys);
2144 }
2145 } else {
2146 /* Same as case 0 */
2147 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2148 hash_keys.addrs.v4addrs.src = fl4->saddr;
2149 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2150 }
2151 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2152 break;
2153 case 3:
2154 if (skb)
2155 mhash = fib_multipath_custom_hash_skb(net, skb);
2156 else
2157 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2158 break;
2159 }
2160
2161 if (multipath_hash)
2162 mhash = jhash_2words(mhash, multipath_hash, 0);
2163
2164 return mhash >> 1;
2165 }
2166 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2167
2168 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)2169 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2170 struct in_device *in_dev, __be32 daddr,
2171 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2172 {
2173 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2174 if (res->fi && fib_info_num_path(res->fi) > 1) {
2175 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2176
2177 fib_select_multipath(res, h, NULL);
2178 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2179 }
2180 #endif
2181
2182 /* create a routing cache entry */
2183 return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2184 }
2185
2186 /* Implements all the saddr-related checks as ip_route_input_slow(),
2187 * assuming daddr is valid and the destination is not a local broadcast one.
2188 * Uses the provided hint instead of performing a route lookup.
2189 */
2190 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)2191 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2192 dscp_t dscp, struct net_device *dev,
2193 const struct sk_buff *hint)
2194 {
2195 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2196 struct in_device *in_dev = __in_dev_get_rcu(dev);
2197 struct rtable *rt = skb_rtable(hint);
2198 struct net *net = dev_net(dev);
2199 u32 tag = 0;
2200
2201 if (!in_dev)
2202 return reason;
2203
2204 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2205 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2206 goto martian_source;
2207 }
2208
2209 if (ipv4_is_zeronet(saddr)) {
2210 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2211 goto martian_source;
2212 }
2213
2214 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2215 reason = SKB_DROP_REASON_IP_LOCALNET;
2216 goto martian_source;
2217 }
2218
2219 if (!(rt->rt_flags & RTCF_LOCAL))
2220 goto skip_validate_source;
2221
2222 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2223 in_dev, &tag);
2224 if (reason)
2225 goto martian_source;
2226
2227 skip_validate_source:
2228 skb_dst_copy(skb, hint);
2229 return SKB_NOT_DROPPED_YET;
2230
2231 martian_source:
2232 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2233 return reason;
2234 }
2235
2236 /* get device for dst_alloc with local routes */
ip_rt_get_dev(struct net * net,const struct fib_result * res)2237 static struct net_device *ip_rt_get_dev(struct net *net,
2238 const struct fib_result *res)
2239 {
2240 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2241 struct net_device *dev = NULL;
2242
2243 if (nhc)
2244 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2245
2246 return dev ? : net->loopback_dev;
2247 }
2248
2249 /*
2250 * NOTE. We drop all the packets that has local source
2251 * addresses, because every properly looped back packet
2252 * must have correct destination already attached by output routine.
2253 * Changes in the enforced policies must be applied also to
2254 * ip_route_use_hint().
2255 *
2256 * Such approach solves two big problems:
2257 * 1. Not simplex devices are handled properly.
2258 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2259 * called with rcu_read_lock()
2260 */
2261
2262 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)2263 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2264 dscp_t dscp, struct net_device *dev,
2265 struct fib_result *res)
2266 {
2267 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2268 struct in_device *in_dev = __in_dev_get_rcu(dev);
2269 struct flow_keys *flkeys = NULL, _flkeys;
2270 struct net *net = dev_net(dev);
2271 struct ip_tunnel_info *tun_info;
2272 int err = -EINVAL;
2273 unsigned int flags = 0;
2274 u32 itag = 0;
2275 struct rtable *rth;
2276 struct flowi4 fl4;
2277 bool do_cache = true;
2278
2279 /* IP on this device is disabled. */
2280
2281 if (!in_dev)
2282 goto out;
2283
2284 /* Check for the most weird martians, which can be not detected
2285 * by fib_lookup.
2286 */
2287
2288 tun_info = skb_tunnel_info(skb);
2289 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2290 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2291 else
2292 fl4.flowi4_tun_key.tun_id = 0;
2293 skb_dst_drop(skb);
2294
2295 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2296 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2297 goto martian_source;
2298 }
2299
2300 res->fi = NULL;
2301 res->table = NULL;
2302 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2303 goto brd_input;
2304
2305 /* Accept zero addresses only to limited broadcast;
2306 * I even do not know to fix it or not. Waiting for complains :-)
2307 */
2308 if (ipv4_is_zeronet(saddr)) {
2309 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2310 goto martian_source;
2311 }
2312
2313 if (ipv4_is_zeronet(daddr)) {
2314 reason = SKB_DROP_REASON_IP_INVALID_DEST;
2315 goto martian_destination;
2316 }
2317
2318 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2319 * and call it once if daddr or/and saddr are loopback addresses
2320 */
2321 if (ipv4_is_loopback(daddr)) {
2322 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2323 reason = SKB_DROP_REASON_IP_LOCALNET;
2324 goto martian_destination;
2325 }
2326 } else if (ipv4_is_loopback(saddr)) {
2327 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2328 reason = SKB_DROP_REASON_IP_LOCALNET;
2329 goto martian_source;
2330 }
2331 }
2332
2333 /*
2334 * Now we are ready to route packet.
2335 */
2336 fl4.flowi4_l3mdev = 0;
2337 fl4.flowi4_oif = 0;
2338 fl4.flowi4_iif = dev->ifindex;
2339 fl4.flowi4_mark = skb->mark;
2340 fl4.flowi4_dscp = dscp;
2341 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2342 fl4.flowi4_flags = 0;
2343 fl4.daddr = daddr;
2344 fl4.saddr = saddr;
2345 fl4.flowi4_uid = sock_net_uid(net, NULL);
2346 fl4.flowi4_multipath_hash = 0;
2347
2348 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2349 flkeys = &_flkeys;
2350 } else {
2351 fl4.flowi4_proto = 0;
2352 fl4.fl4_sport = 0;
2353 fl4.fl4_dport = 0;
2354 }
2355
2356 err = fib_lookup(net, &fl4, res, 0);
2357 if (err != 0) {
2358 if (!IN_DEV_FORWARD(in_dev))
2359 err = -EHOSTUNREACH;
2360 goto no_route;
2361 }
2362
2363 if (res->type == RTN_BROADCAST) {
2364 if (IN_DEV_BFORWARD(in_dev))
2365 goto make_route;
2366 /* not do cache if bc_forwarding is enabled */
2367 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2368 do_cache = false;
2369 goto brd_input;
2370 }
2371
2372 err = -EINVAL;
2373 if (res->type == RTN_LOCAL) {
2374 reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2375 0, dev, in_dev, &itag);
2376 if (reason)
2377 goto martian_source;
2378 goto local_input;
2379 }
2380
2381 if (!IN_DEV_FORWARD(in_dev)) {
2382 err = -EHOSTUNREACH;
2383 goto no_route;
2384 }
2385 if (res->type != RTN_UNICAST) {
2386 reason = SKB_DROP_REASON_IP_INVALID_DEST;
2387 goto martian_destination;
2388 }
2389
2390 make_route:
2391 reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2392 flkeys);
2393
2394 out:
2395 return reason;
2396
2397 brd_input:
2398 if (skb->protocol != htons(ETH_P_IP)) {
2399 reason = SKB_DROP_REASON_INVALID_PROTO;
2400 goto out;
2401 }
2402
2403 if (!ipv4_is_zeronet(saddr)) {
2404 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2405 dev, in_dev, &itag);
2406 if (reason)
2407 goto martian_source;
2408 }
2409 flags |= RTCF_BROADCAST;
2410 res->type = RTN_BROADCAST;
2411 RT_CACHE_STAT_INC(in_brd);
2412
2413 local_input:
2414 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2415 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2416
2417 do_cache &= res->fi && !itag;
2418 if (do_cache) {
2419 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2420
2421 rth = rcu_dereference(nhc->nhc_rth_input);
2422 if (rt_cache_valid(rth)) {
2423 skb_dst_set_noref(skb, &rth->dst);
2424 reason = SKB_NOT_DROPPED_YET;
2425 goto out;
2426 }
2427 }
2428
2429 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2430 flags | RTCF_LOCAL, res->type, false);
2431 if (!rth)
2432 goto e_nobufs;
2433
2434 rth->dst.output= ip_rt_bug;
2435 #ifdef CONFIG_IP_ROUTE_CLASSID
2436 rth->dst.tclassid = itag;
2437 #endif
2438 rth->rt_is_input = 1;
2439
2440 RT_CACHE_STAT_INC(in_slow_tot);
2441 if (res->type == RTN_UNREACHABLE) {
2442 rth->dst.input= ip_error;
2443 rth->dst.error= -err;
2444 rth->rt_flags &= ~RTCF_LOCAL;
2445 }
2446
2447 if (do_cache) {
2448 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2449
2450 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2451 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2452 WARN_ON(rth->dst.input == lwtunnel_input);
2453 rth->dst.lwtstate->orig_input = rth->dst.input;
2454 rth->dst.input = lwtunnel_input;
2455 }
2456
2457 if (unlikely(!rt_cache_route(nhc, rth)))
2458 rt_add_uncached_list(rth);
2459 }
2460 skb_dst_set(skb, &rth->dst);
2461 reason = SKB_NOT_DROPPED_YET;
2462 goto out;
2463
2464 no_route:
2465 RT_CACHE_STAT_INC(in_no_route);
2466 res->type = RTN_UNREACHABLE;
2467 res->fi = NULL;
2468 res->table = NULL;
2469 goto local_input;
2470
2471 /*
2472 * Do not cache martian addresses: they should be logged (RFC1812)
2473 */
2474 martian_destination:
2475 RT_CACHE_STAT_INC(in_martian_dst);
2476 #ifdef CONFIG_IP_ROUTE_VERBOSE
2477 if (IN_DEV_LOG_MARTIANS(in_dev))
2478 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2479 &daddr, &saddr, dev->name);
2480 #endif
2481 goto out;
2482
2483 e_nobufs:
2484 reason = SKB_DROP_REASON_NOMEM;
2485 goto out;
2486
2487 martian_source:
2488 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2489 goto out;
2490 }
2491
2492 /* called with rcu_read_lock held */
2493 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)2494 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2495 dscp_t dscp, struct net_device *dev,
2496 struct fib_result *res)
2497 {
2498 /* Multicast recognition logic is moved from route cache to here.
2499 * The problem was that too many Ethernet cards have broken/missing
2500 * hardware multicast filters :-( As result the host on multicasting
2501 * network acquires a lot of useless route cache entries, sort of
2502 * SDR messages from all the world. Now we try to get rid of them.
2503 * Really, provided software IP multicast filter is organized
2504 * reasonably (at least, hashed), it does not result in a slowdown
2505 * comparing with route cache reject entries.
2506 * Note, that multicast routers are not affected, because
2507 * route cache entry is created eventually.
2508 */
2509 if (ipv4_is_multicast(daddr)) {
2510 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2511 struct in_device *in_dev = __in_dev_get_rcu(dev);
2512 int our = 0;
2513
2514 if (!in_dev)
2515 return reason;
2516
2517 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2518 ip_hdr(skb)->protocol);
2519
2520 /* check l3 master if no match yet */
2521 if (!our && netif_is_l3_slave(dev)) {
2522 struct in_device *l3_in_dev;
2523
2524 l3_in_dev = __in_dev_get_rcu(skb->dev);
2525 if (l3_in_dev)
2526 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2527 ip_hdr(skb)->protocol);
2528 }
2529
2530 if (our
2531 #ifdef CONFIG_IP_MROUTE
2532 ||
2533 (!ipv4_is_local_multicast(daddr) &&
2534 IN_DEV_MFORWARD(in_dev))
2535 #endif
2536 ) {
2537 reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2538 dev, our);
2539 }
2540 return reason;
2541 }
2542
2543 return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2544 }
2545
ip_route_input_noref(struct sk_buff * skb,__be32 daddr,__be32 saddr,dscp_t dscp,struct net_device * dev)2546 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2547 __be32 saddr, dscp_t dscp,
2548 struct net_device *dev)
2549 {
2550 enum skb_drop_reason reason;
2551 struct fib_result res;
2552
2553 rcu_read_lock();
2554 reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2555 rcu_read_unlock();
2556
2557 return reason;
2558 }
2559 EXPORT_SYMBOL(ip_route_input_noref);
2560
2561 /* 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)2562 static struct rtable *__mkroute_output(const struct fib_result *res,
2563 const struct flowi4 *fl4, int orig_oif,
2564 struct net_device *dev_out,
2565 unsigned int flags)
2566 {
2567 struct fib_info *fi = res->fi;
2568 struct fib_nh_exception *fnhe;
2569 struct in_device *in_dev;
2570 u16 type = res->type;
2571 struct rtable *rth;
2572 bool do_cache;
2573
2574 in_dev = __in_dev_get_rcu(dev_out);
2575 if (!in_dev)
2576 return ERR_PTR(-EINVAL);
2577
2578 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2579 if (ipv4_is_loopback(fl4->saddr) &&
2580 !(dev_out->flags & IFF_LOOPBACK) &&
2581 !netif_is_l3_master(dev_out))
2582 return ERR_PTR(-EINVAL);
2583
2584 if (ipv4_is_lbcast(fl4->daddr)) {
2585 type = RTN_BROADCAST;
2586
2587 /* reset fi to prevent gateway resolution */
2588 fi = NULL;
2589 } else if (ipv4_is_multicast(fl4->daddr)) {
2590 type = RTN_MULTICAST;
2591 } else if (ipv4_is_zeronet(fl4->daddr)) {
2592 return ERR_PTR(-EINVAL);
2593 }
2594
2595 if (dev_out->flags & IFF_LOOPBACK)
2596 flags |= RTCF_LOCAL;
2597
2598 do_cache = true;
2599 if (type == RTN_BROADCAST) {
2600 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2601 } else if (type == RTN_MULTICAST) {
2602 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2603 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2604 fl4->flowi4_proto))
2605 flags &= ~RTCF_LOCAL;
2606 else
2607 do_cache = false;
2608 /* If multicast route do not exist use
2609 * default one, but do not gateway in this case.
2610 * Yes, it is hack.
2611 */
2612 if (fi && res->prefixlen < 4)
2613 fi = NULL;
2614 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2615 (orig_oif != dev_out->ifindex)) {
2616 /* For local routes that require a particular output interface
2617 * we do not want to cache the result. Caching the result
2618 * causes incorrect behaviour when there are multiple source
2619 * addresses on the interface, the end result being that if the
2620 * intended recipient is waiting on that interface for the
2621 * packet he won't receive it because it will be delivered on
2622 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2623 * be set to the loopback interface as well.
2624 */
2625 do_cache = false;
2626 }
2627
2628 fnhe = NULL;
2629 do_cache &= fi != NULL;
2630 if (fi) {
2631 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2632 struct rtable __rcu **prth;
2633
2634 fnhe = find_exception(nhc, fl4->daddr);
2635 if (!do_cache)
2636 goto add;
2637 if (fnhe) {
2638 prth = &fnhe->fnhe_rth_output;
2639 } else {
2640 if (unlikely(fl4->flowi4_flags &
2641 FLOWI_FLAG_KNOWN_NH &&
2642 !(nhc->nhc_gw_family &&
2643 nhc->nhc_scope == RT_SCOPE_LINK))) {
2644 do_cache = false;
2645 goto add;
2646 }
2647 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2648 }
2649 rth = rcu_dereference(*prth);
2650 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2651 return rth;
2652 }
2653
2654 add:
2655 rth = rt_dst_alloc(dev_out, flags, type,
2656 IN_DEV_ORCONF(in_dev, NOXFRM));
2657 if (!rth)
2658 return ERR_PTR(-ENOBUFS);
2659
2660 rth->rt_iif = orig_oif;
2661
2662 RT_CACHE_STAT_INC(out_slow_tot);
2663
2664 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2665 if (flags & RTCF_LOCAL &&
2666 !(dev_out->flags & IFF_LOOPBACK)) {
2667 rth->dst.output = ip_mc_output;
2668 RT_CACHE_STAT_INC(out_slow_mc);
2669 }
2670 #ifdef CONFIG_IP_MROUTE
2671 if (type == RTN_MULTICAST) {
2672 if (IN_DEV_MFORWARD(in_dev) &&
2673 !ipv4_is_local_multicast(fl4->daddr)) {
2674 rth->dst.input = ip_mr_input;
2675 rth->dst.output = ip_mr_output;
2676 }
2677 }
2678 #endif
2679 }
2680
2681 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2682 lwtunnel_set_redirect(&rth->dst);
2683
2684 return rth;
2685 }
2686
2687 /*
2688 * Major route resolver routine.
2689 */
2690
ip_route_output_key_hash(struct net * net,struct flowi4 * fl4,const struct sk_buff * skb)2691 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2692 const struct sk_buff *skb)
2693 {
2694 struct fib_result res = {
2695 .type = RTN_UNSPEC,
2696 .fi = NULL,
2697 .table = NULL,
2698 .tclassid = 0,
2699 };
2700 struct rtable *rth;
2701
2702 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2703
2704 rcu_read_lock();
2705 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2706 rcu_read_unlock();
2707
2708 return rth;
2709 }
2710 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2711
ip_route_output_key_hash_rcu(struct net * net,struct flowi4 * fl4,struct fib_result * res,const struct sk_buff * skb)2712 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2713 struct fib_result *res,
2714 const struct sk_buff *skb)
2715 {
2716 struct net_device *dev_out = NULL;
2717 int orig_oif = fl4->flowi4_oif;
2718 unsigned int flags = 0;
2719 struct rtable *rth;
2720 int err;
2721
2722 if (fl4->saddr) {
2723 if (ipv4_is_multicast(fl4->saddr) ||
2724 ipv4_is_lbcast(fl4->saddr)) {
2725 rth = ERR_PTR(-EINVAL);
2726 goto out;
2727 }
2728
2729 rth = ERR_PTR(-ENETUNREACH);
2730
2731 /* I removed check for oif == dev_out->oif here.
2732 * It was wrong for two reasons:
2733 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2734 * is assigned to multiple interfaces.
2735 * 2. Moreover, we are allowed to send packets with saddr
2736 * of another iface. --ANK
2737 */
2738
2739 if (fl4->flowi4_oif == 0 &&
2740 (ipv4_is_multicast(fl4->daddr) ||
2741 ipv4_is_lbcast(fl4->daddr))) {
2742 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2743 dev_out = __ip_dev_find(net, fl4->saddr, false);
2744 if (!dev_out)
2745 goto out;
2746
2747 /* Special hack: user can direct multicasts
2748 * and limited broadcast via necessary interface
2749 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2750 * This hack is not just for fun, it allows
2751 * vic,vat and friends to work.
2752 * They bind socket to loopback, set ttl to zero
2753 * and expect that it will work.
2754 * From the viewpoint of routing cache they are broken,
2755 * because we are not allowed to build multicast path
2756 * with loopback source addr (look, routing cache
2757 * cannot know, that ttl is zero, so that packet
2758 * will not leave this host and route is valid).
2759 * Luckily, this hack is good workaround.
2760 */
2761
2762 fl4->flowi4_oif = dev_out->ifindex;
2763 goto make_route;
2764 }
2765
2766 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2767 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2768 if (!__ip_dev_find(net, fl4->saddr, false))
2769 goto out;
2770 }
2771 }
2772
2773
2774 if (fl4->flowi4_oif) {
2775 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2776 rth = ERR_PTR(-ENODEV);
2777 if (!dev_out)
2778 goto out;
2779
2780 /* RACE: Check return value of inet_select_addr instead. */
2781 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2782 rth = ERR_PTR(-ENETUNREACH);
2783 goto out;
2784 }
2785 if (ipv4_is_local_multicast(fl4->daddr) ||
2786 ipv4_is_lbcast(fl4->daddr) ||
2787 fl4->flowi4_proto == IPPROTO_IGMP) {
2788 if (!fl4->saddr)
2789 fl4->saddr = inet_select_addr(dev_out, 0,
2790 RT_SCOPE_LINK);
2791 goto make_route;
2792 }
2793 if (!fl4->saddr) {
2794 if (ipv4_is_multicast(fl4->daddr))
2795 fl4->saddr = inet_select_addr(dev_out, 0,
2796 fl4->flowi4_scope);
2797 else if (!fl4->daddr)
2798 fl4->saddr = inet_select_addr(dev_out, 0,
2799 RT_SCOPE_HOST);
2800 }
2801 }
2802
2803 if (!fl4->daddr) {
2804 fl4->daddr = fl4->saddr;
2805 if (!fl4->daddr)
2806 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2807 dev_out = net->loopback_dev;
2808 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2809 res->type = RTN_LOCAL;
2810 flags |= RTCF_LOCAL;
2811 goto make_route;
2812 }
2813
2814 err = fib_lookup(net, fl4, res, 0);
2815 if (err) {
2816 res->fi = NULL;
2817 res->table = NULL;
2818 if (fl4->flowi4_oif &&
2819 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2820 /* Apparently, routing tables are wrong. Assume,
2821 * that the destination is on link.
2822 *
2823 * WHY? DW.
2824 * Because we are allowed to send to iface
2825 * even if it has NO routes and NO assigned
2826 * addresses. When oif is specified, routing
2827 * tables are looked up with only one purpose:
2828 * to catch if destination is gatewayed, rather than
2829 * direct. Moreover, if MSG_DONTROUTE is set,
2830 * we send packet, ignoring both routing tables
2831 * and ifaddr state. --ANK
2832 *
2833 *
2834 * We could make it even if oif is unknown,
2835 * likely IPv6, but we do not.
2836 */
2837
2838 if (fl4->saddr == 0)
2839 fl4->saddr = inet_select_addr(dev_out, 0,
2840 RT_SCOPE_LINK);
2841 res->type = RTN_UNICAST;
2842 goto make_route;
2843 }
2844 rth = ERR_PTR(err);
2845 goto out;
2846 }
2847
2848 if (res->type == RTN_LOCAL) {
2849 if (!fl4->saddr) {
2850 if (res->fi->fib_prefsrc)
2851 fl4->saddr = res->fi->fib_prefsrc;
2852 else
2853 fl4->saddr = fl4->daddr;
2854 }
2855
2856 /* L3 master device is the loopback for that domain */
2857 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2858 net->loopback_dev;
2859
2860 /* make sure orig_oif points to fib result device even
2861 * though packet rx/tx happens over loopback or l3mdev
2862 */
2863 orig_oif = FIB_RES_OIF(*res);
2864
2865 fl4->flowi4_oif = dev_out->ifindex;
2866 flags |= RTCF_LOCAL;
2867 goto make_route;
2868 }
2869
2870 fib_select_path(net, res, fl4, skb);
2871
2872 dev_out = FIB_RES_DEV(*res);
2873
2874 make_route:
2875 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2876
2877 out:
2878 return rth;
2879 }
2880
2881 static struct dst_ops ipv4_dst_blackhole_ops = {
2882 .family = AF_INET,
2883 .default_advmss = ipv4_default_advmss,
2884 .neigh_lookup = ipv4_neigh_lookup,
2885 .check = dst_blackhole_check,
2886 .cow_metrics = dst_blackhole_cow_metrics,
2887 .update_pmtu = dst_blackhole_update_pmtu,
2888 .redirect = dst_blackhole_redirect,
2889 .mtu = dst_blackhole_mtu,
2890 };
2891
ipv4_blackhole_route(struct net * net,struct dst_entry * dst_orig)2892 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2893 {
2894 struct rtable *ort = dst_rtable(dst_orig);
2895 struct rtable *rt;
2896
2897 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2898 if (rt) {
2899 struct dst_entry *new = &rt->dst;
2900
2901 new->__use = 1;
2902 new->input = dst_discard;
2903 new->output = dst_discard_out;
2904
2905 new->dev = net->loopback_dev;
2906 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2907
2908 rt->rt_is_input = ort->rt_is_input;
2909 rt->rt_iif = ort->rt_iif;
2910 rt->rt_pmtu = ort->rt_pmtu;
2911 rt->rt_mtu_locked = ort->rt_mtu_locked;
2912
2913 rt->rt_genid = rt_genid_ipv4(net);
2914 rt->rt_flags = ort->rt_flags;
2915 rt->rt_type = ort->rt_type;
2916 rt->rt_uses_gateway = ort->rt_uses_gateway;
2917 rt->rt_gw_family = ort->rt_gw_family;
2918 if (rt->rt_gw_family == AF_INET)
2919 rt->rt_gw4 = ort->rt_gw4;
2920 else if (rt->rt_gw_family == AF_INET6)
2921 rt->rt_gw6 = ort->rt_gw6;
2922 }
2923
2924 dst_release(dst_orig);
2925
2926 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2927 }
2928
ip_route_output_flow(struct net * net,struct flowi4 * flp4,const struct sock * sk)2929 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2930 const struct sock *sk)
2931 {
2932 struct rtable *rt = __ip_route_output_key(net, flp4);
2933
2934 if (IS_ERR(rt))
2935 return rt;
2936
2937 if (flp4->flowi4_proto) {
2938 flp4->flowi4_oif = rt->dst.dev->ifindex;
2939 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2940 flowi4_to_flowi(flp4),
2941 sk, 0));
2942 }
2943
2944 return rt;
2945 }
2946 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2947
2948 /* 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)2949 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2950 struct rtable *rt, u32 table_id, dscp_t dscp,
2951 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2952 u32 seq, unsigned int flags)
2953 {
2954 struct rtmsg *r;
2955 struct nlmsghdr *nlh;
2956 unsigned long expires = 0;
2957 u32 error;
2958 u32 metrics[RTAX_MAX];
2959
2960 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2961 if (!nlh)
2962 return -EMSGSIZE;
2963
2964 r = nlmsg_data(nlh);
2965 r->rtm_family = AF_INET;
2966 r->rtm_dst_len = 32;
2967 r->rtm_src_len = 0;
2968 r->rtm_tos = inet_dscp_to_dsfield(dscp);
2969 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2970 if (nla_put_u32(skb, RTA_TABLE, table_id))
2971 goto nla_put_failure;
2972 r->rtm_type = rt->rt_type;
2973 r->rtm_scope = RT_SCOPE_UNIVERSE;
2974 r->rtm_protocol = RTPROT_UNSPEC;
2975 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2976 if (rt->rt_flags & RTCF_NOTIFY)
2977 r->rtm_flags |= RTM_F_NOTIFY;
2978 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2979 r->rtm_flags |= RTCF_DOREDIRECT;
2980
2981 if (nla_put_in_addr(skb, RTA_DST, dst))
2982 goto nla_put_failure;
2983 if (src) {
2984 r->rtm_src_len = 32;
2985 if (nla_put_in_addr(skb, RTA_SRC, src))
2986 goto nla_put_failure;
2987 }
2988 if (rt->dst.dev &&
2989 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2990 goto nla_put_failure;
2991 if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2992 goto nla_put_failure;
2993 #ifdef CONFIG_IP_ROUTE_CLASSID
2994 if (rt->dst.tclassid &&
2995 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2996 goto nla_put_failure;
2997 #endif
2998 if (fl4 && !rt_is_input_route(rt) &&
2999 fl4->saddr != src) {
3000 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
3001 goto nla_put_failure;
3002 }
3003 if (rt->rt_uses_gateway) {
3004 if (rt->rt_gw_family == AF_INET &&
3005 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3006 goto nla_put_failure;
3007 } else if (rt->rt_gw_family == AF_INET6) {
3008 int alen = sizeof(struct in6_addr);
3009 struct nlattr *nla;
3010 struct rtvia *via;
3011
3012 nla = nla_reserve(skb, RTA_VIA, alen + 2);
3013 if (!nla)
3014 goto nla_put_failure;
3015
3016 via = nla_data(nla);
3017 via->rtvia_family = AF_INET6;
3018 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3019 }
3020 }
3021
3022 expires = READ_ONCE(rt->dst.expires);
3023 if (expires) {
3024 unsigned long now = jiffies;
3025
3026 if (time_before(now, expires))
3027 expires -= now;
3028 else
3029 expires = 0;
3030 }
3031
3032 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3033 if (rt->rt_pmtu && expires)
3034 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3035 if (rt->rt_mtu_locked && expires)
3036 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3037 if (rtnetlink_put_metrics(skb, metrics) < 0)
3038 goto nla_put_failure;
3039
3040 if (fl4) {
3041 if (fl4->flowi4_mark &&
3042 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3043 goto nla_put_failure;
3044
3045 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3046 nla_put_u32(skb, RTA_UID,
3047 from_kuid_munged(current_user_ns(),
3048 fl4->flowi4_uid)))
3049 goto nla_put_failure;
3050
3051 if (rt_is_input_route(rt)) {
3052 #ifdef CONFIG_IP_MROUTE
3053 if (ipv4_is_multicast(dst) &&
3054 !ipv4_is_local_multicast(dst) &&
3055 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3056 int err = ipmr_get_route(net, skb,
3057 fl4->saddr, fl4->daddr,
3058 r, portid);
3059
3060 if (err <= 0) {
3061 if (err == 0)
3062 return 0;
3063 goto nla_put_failure;
3064 }
3065 } else
3066 #endif
3067 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3068 goto nla_put_failure;
3069 }
3070 }
3071
3072 error = rt->dst.error;
3073
3074 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3075 goto nla_put_failure;
3076
3077 nlmsg_end(skb, nlh);
3078 return 0;
3079
3080 nla_put_failure:
3081 nlmsg_cancel(skb, nlh);
3082 return -EMSGSIZE;
3083 }
3084
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)3085 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3086 struct netlink_callback *cb, u32 table_id,
3087 struct fnhe_hash_bucket *bucket, int genid,
3088 int *fa_index, int fa_start, unsigned int flags)
3089 {
3090 int i;
3091
3092 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3093 struct fib_nh_exception *fnhe;
3094
3095 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3096 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3097 struct rtable *rt;
3098 int err;
3099
3100 if (*fa_index < fa_start)
3101 goto next;
3102
3103 if (fnhe->fnhe_genid != genid)
3104 goto next;
3105
3106 if (fnhe->fnhe_expires &&
3107 time_after(jiffies, fnhe->fnhe_expires))
3108 goto next;
3109
3110 rt = rcu_dereference(fnhe->fnhe_rth_input);
3111 if (!rt)
3112 rt = rcu_dereference(fnhe->fnhe_rth_output);
3113 if (!rt)
3114 goto next;
3115
3116 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3117 table_id, 0, NULL, skb,
3118 NETLINK_CB(cb->skb).portid,
3119 cb->nlh->nlmsg_seq, flags);
3120 if (err)
3121 return err;
3122 next:
3123 (*fa_index)++;
3124 }
3125 }
3126
3127 return 0;
3128 }
3129
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)3130 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3131 u32 table_id, struct fib_info *fi,
3132 int *fa_index, int fa_start, unsigned int flags)
3133 {
3134 struct net *net = sock_net(cb->skb->sk);
3135 int nhsel, genid = fnhe_genid(net);
3136
3137 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3138 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3139 struct fnhe_hash_bucket *bucket;
3140 int err;
3141
3142 if (nhc->nhc_flags & RTNH_F_DEAD)
3143 continue;
3144
3145 rcu_read_lock();
3146 bucket = rcu_dereference(nhc->nhc_exceptions);
3147 err = 0;
3148 if (bucket)
3149 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3150 genid, fa_index, fa_start,
3151 flags);
3152 rcu_read_unlock();
3153 if (err)
3154 return err;
3155 }
3156
3157 return 0;
3158 }
3159
inet_rtm_getroute_build_skb(__be32 src,__be32 dst,u8 ip_proto,__be16 sport,__be16 dport)3160 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3161 u8 ip_proto, __be16 sport,
3162 __be16 dport)
3163 {
3164 struct sk_buff *skb;
3165 struct iphdr *iph;
3166
3167 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3168 if (!skb)
3169 return NULL;
3170
3171 /* Reserve room for dummy headers, this skb can pass
3172 * through good chunk of routing engine.
3173 */
3174 skb_reset_mac_header(skb);
3175 skb_reset_network_header(skb);
3176 skb->protocol = htons(ETH_P_IP);
3177 iph = skb_put(skb, sizeof(struct iphdr));
3178 iph->protocol = ip_proto;
3179 iph->saddr = src;
3180 iph->daddr = dst;
3181 iph->version = 0x4;
3182 iph->frag_off = 0;
3183 iph->ihl = 0x5;
3184 skb_set_transport_header(skb, skb->len);
3185
3186 switch (iph->protocol) {
3187 case IPPROTO_UDP: {
3188 struct udphdr *udph;
3189
3190 udph = skb_put_zero(skb, sizeof(struct udphdr));
3191 udph->source = sport;
3192 udph->dest = dport;
3193 udph->len = htons(sizeof(struct udphdr));
3194 udph->check = 0;
3195 break;
3196 }
3197 case IPPROTO_TCP: {
3198 struct tcphdr *tcph;
3199
3200 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3201 tcph->source = sport;
3202 tcph->dest = dport;
3203 tcph->doff = sizeof(struct tcphdr) / 4;
3204 tcph->rst = 1;
3205 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3206 src, dst, 0);
3207 break;
3208 }
3209 case IPPROTO_ICMP: {
3210 struct icmphdr *icmph;
3211
3212 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3213 icmph->type = ICMP_ECHO;
3214 icmph->code = 0;
3215 }
3216 }
3217
3218 return skb;
3219 }
3220
inet_rtm_valid_getroute_req(struct sk_buff * skb,const struct nlmsghdr * nlh,struct nlattr ** tb,struct netlink_ext_ack * extack)3221 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3222 const struct nlmsghdr *nlh,
3223 struct nlattr **tb,
3224 struct netlink_ext_ack *extack)
3225 {
3226 struct rtmsg *rtm;
3227 int i, err;
3228
3229 rtm = nlmsg_payload(nlh, sizeof(*rtm));
3230 if (!rtm) {
3231 NL_SET_ERR_MSG(extack,
3232 "ipv4: Invalid header for route get request");
3233 return -EINVAL;
3234 }
3235
3236 if (!netlink_strict_get_check(skb))
3237 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3238 rtm_ipv4_policy, extack);
3239
3240 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3241 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3242 rtm->rtm_table || rtm->rtm_protocol ||
3243 rtm->rtm_scope || rtm->rtm_type) {
3244 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3245 return -EINVAL;
3246 }
3247
3248 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3249 RTM_F_LOOKUP_TABLE |
3250 RTM_F_FIB_MATCH)) {
3251 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3252 return -EINVAL;
3253 }
3254
3255 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3256 rtm_ipv4_policy, extack);
3257 if (err)
3258 return err;
3259
3260 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3261 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3262 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3263 return -EINVAL;
3264 }
3265
3266 for (i = 0; i <= RTA_MAX; i++) {
3267 if (!tb[i])
3268 continue;
3269
3270 switch (i) {
3271 case RTA_IIF:
3272 case RTA_OIF:
3273 case RTA_SRC:
3274 case RTA_DST:
3275 case RTA_IP_PROTO:
3276 case RTA_SPORT:
3277 case RTA_DPORT:
3278 case RTA_MARK:
3279 case RTA_UID:
3280 break;
3281 default:
3282 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3283 return -EINVAL;
3284 }
3285 }
3286
3287 return 0;
3288 }
3289
inet_rtm_getroute(struct sk_buff * in_skb,struct nlmsghdr * nlh,struct netlink_ext_ack * extack)3290 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3291 struct netlink_ext_ack *extack)
3292 {
3293 struct net *net = sock_net(in_skb->sk);
3294 struct nlattr *tb[RTA_MAX+1];
3295 u32 table_id = RT_TABLE_MAIN;
3296 __be16 sport = 0, dport = 0;
3297 struct fib_result res = {};
3298 u8 ip_proto = IPPROTO_UDP;
3299 struct rtable *rt = NULL;
3300 struct sk_buff *skb;
3301 struct rtmsg *rtm;
3302 struct flowi4 fl4 = {};
3303 __be32 dst = 0;
3304 __be32 src = 0;
3305 dscp_t dscp;
3306 kuid_t uid;
3307 u32 iif;
3308 int err;
3309 int mark;
3310
3311 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3312 if (err < 0)
3313 return err;
3314
3315 rtm = nlmsg_data(nlh);
3316 src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3317 dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3318 iif = nla_get_u32_default(tb[RTA_IIF], 0);
3319 mark = nla_get_u32_default(tb[RTA_MARK], 0);
3320 dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
3321 if (tb[RTA_UID])
3322 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3323 else
3324 uid = (iif ? INVALID_UID : current_uid());
3325
3326 if (tb[RTA_IP_PROTO]) {
3327 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3328 &ip_proto, AF_INET, extack);
3329 if (err)
3330 return err;
3331 }
3332
3333 if (tb[RTA_SPORT])
3334 sport = nla_get_be16(tb[RTA_SPORT]);
3335
3336 if (tb[RTA_DPORT])
3337 dport = nla_get_be16(tb[RTA_DPORT]);
3338
3339 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3340 if (!skb)
3341 return -ENOBUFS;
3342
3343 fl4.daddr = dst;
3344 fl4.saddr = src;
3345 fl4.flowi4_dscp = dscp;
3346 fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3347 fl4.flowi4_mark = mark;
3348 fl4.flowi4_uid = uid;
3349 if (sport)
3350 fl4.fl4_sport = sport;
3351 if (dport)
3352 fl4.fl4_dport = dport;
3353 fl4.flowi4_proto = ip_proto;
3354
3355 rcu_read_lock();
3356
3357 if (iif) {
3358 struct net_device *dev;
3359
3360 dev = dev_get_by_index_rcu(net, iif);
3361 if (!dev) {
3362 err = -ENODEV;
3363 goto errout_rcu;
3364 }
3365
3366 fl4.flowi4_iif = iif; /* for rt_fill_info */
3367 skb->dev = dev;
3368 skb->mark = mark;
3369 err = ip_route_input_rcu(skb, dst, src, dscp, dev,
3370 &res) ? -EINVAL : 0;
3371
3372 rt = skb_rtable(skb);
3373 if (err == 0 && rt->dst.error)
3374 err = -rt->dst.error;
3375 } else {
3376 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3377 skb->dev = net->loopback_dev;
3378 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3379 err = 0;
3380 if (IS_ERR(rt))
3381 err = PTR_ERR(rt);
3382 else
3383 skb_dst_set(skb, &rt->dst);
3384 }
3385
3386 if (err)
3387 goto errout_rcu;
3388
3389 if (rtm->rtm_flags & RTM_F_NOTIFY)
3390 rt->rt_flags |= RTCF_NOTIFY;
3391
3392 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3393 table_id = res.table ? res.table->tb_id : 0;
3394
3395 /* reset skb for netlink reply msg */
3396 skb_trim(skb, 0);
3397 skb_reset_network_header(skb);
3398 skb_reset_transport_header(skb);
3399 skb_reset_mac_header(skb);
3400
3401 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3402 struct fib_rt_info fri;
3403
3404 if (!res.fi) {
3405 err = fib_props[res.type].error;
3406 if (!err)
3407 err = -EHOSTUNREACH;
3408 goto errout_rcu;
3409 }
3410 fri.fi = res.fi;
3411 fri.tb_id = table_id;
3412 fri.dst = res.prefix;
3413 fri.dst_len = res.prefixlen;
3414 fri.dscp = res.dscp;
3415 fri.type = rt->rt_type;
3416 fri.offload = 0;
3417 fri.trap = 0;
3418 fri.offload_failed = 0;
3419 if (res.fa_head) {
3420 struct fib_alias *fa;
3421
3422 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3423 u8 slen = 32 - fri.dst_len;
3424
3425 if (fa->fa_slen == slen &&
3426 fa->tb_id == fri.tb_id &&
3427 fa->fa_dscp == fri.dscp &&
3428 fa->fa_info == res.fi &&
3429 fa->fa_type == fri.type) {
3430 fri.offload = READ_ONCE(fa->offload);
3431 fri.trap = READ_ONCE(fa->trap);
3432 fri.offload_failed =
3433 READ_ONCE(fa->offload_failed);
3434 break;
3435 }
3436 }
3437 }
3438 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3439 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3440 } else {
3441 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3442 skb, NETLINK_CB(in_skb).portid,
3443 nlh->nlmsg_seq, 0);
3444 }
3445 if (err < 0)
3446 goto errout_rcu;
3447
3448 rcu_read_unlock();
3449
3450 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3451
3452 errout_free:
3453 return err;
3454 errout_rcu:
3455 rcu_read_unlock();
3456 kfree_skb(skb);
3457 goto errout_free;
3458 }
3459
ip_rt_multicast_event(struct in_device * in_dev)3460 void ip_rt_multicast_event(struct in_device *in_dev)
3461 {
3462 rt_cache_flush(dev_net(in_dev->dev));
3463 }
3464
3465 #ifdef CONFIG_SYSCTL
3466 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3467 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3468 static int ip_rt_gc_elasticity __read_mostly = 8;
3469 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3470
ipv4_sysctl_rtcache_flush(const struct ctl_table * __ctl,int write,void * buffer,size_t * lenp,loff_t * ppos)3471 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3472 void *buffer, size_t *lenp, loff_t *ppos)
3473 {
3474 struct net *net = (struct net *)__ctl->extra1;
3475
3476 if (write) {
3477 rt_cache_flush(net);
3478 fnhe_genid_bump(net);
3479 return 0;
3480 }
3481
3482 return -EINVAL;
3483 }
3484
3485 static struct ctl_table ipv4_route_table[] = {
3486 {
3487 .procname = "gc_thresh",
3488 .data = &ipv4_dst_ops.gc_thresh,
3489 .maxlen = sizeof(int),
3490 .mode = 0644,
3491 .proc_handler = proc_dointvec,
3492 },
3493 {
3494 .procname = "max_size",
3495 .data = &ip_rt_max_size,
3496 .maxlen = sizeof(int),
3497 .mode = 0644,
3498 .proc_handler = proc_dointvec,
3499 },
3500 {
3501 /* Deprecated. Use gc_min_interval_ms */
3502
3503 .procname = "gc_min_interval",
3504 .data = &ip_rt_gc_min_interval,
3505 .maxlen = sizeof(int),
3506 .mode = 0644,
3507 .proc_handler = proc_dointvec_jiffies,
3508 },
3509 {
3510 .procname = "gc_min_interval_ms",
3511 .data = &ip_rt_gc_min_interval,
3512 .maxlen = sizeof(int),
3513 .mode = 0644,
3514 .proc_handler = proc_dointvec_ms_jiffies,
3515 },
3516 {
3517 .procname = "gc_timeout",
3518 .data = &ip_rt_gc_timeout,
3519 .maxlen = sizeof(int),
3520 .mode = 0644,
3521 .proc_handler = proc_dointvec_jiffies,
3522 },
3523 {
3524 .procname = "gc_interval",
3525 .data = &ip_rt_gc_interval,
3526 .maxlen = sizeof(int),
3527 .mode = 0644,
3528 .proc_handler = proc_dointvec_jiffies,
3529 },
3530 {
3531 .procname = "redirect_load",
3532 .data = &ip_rt_redirect_load,
3533 .maxlen = sizeof(int),
3534 .mode = 0644,
3535 .proc_handler = proc_dointvec,
3536 },
3537 {
3538 .procname = "redirect_number",
3539 .data = &ip_rt_redirect_number,
3540 .maxlen = sizeof(int),
3541 .mode = 0644,
3542 .proc_handler = proc_dointvec,
3543 },
3544 {
3545 .procname = "redirect_silence",
3546 .data = &ip_rt_redirect_silence,
3547 .maxlen = sizeof(int),
3548 .mode = 0644,
3549 .proc_handler = proc_dointvec,
3550 },
3551 {
3552 .procname = "error_cost",
3553 .data = &ip_rt_error_cost,
3554 .maxlen = sizeof(int),
3555 .mode = 0644,
3556 .proc_handler = proc_dointvec,
3557 },
3558 {
3559 .procname = "error_burst",
3560 .data = &ip_rt_error_burst,
3561 .maxlen = sizeof(int),
3562 .mode = 0644,
3563 .proc_handler = proc_dointvec,
3564 },
3565 {
3566 .procname = "gc_elasticity",
3567 .data = &ip_rt_gc_elasticity,
3568 .maxlen = sizeof(int),
3569 .mode = 0644,
3570 .proc_handler = proc_dointvec,
3571 },
3572 };
3573
3574 static const char ipv4_route_flush_procname[] = "flush";
3575
3576 static struct ctl_table ipv4_route_netns_table[] = {
3577 {
3578 .procname = ipv4_route_flush_procname,
3579 .maxlen = sizeof(int),
3580 .mode = 0200,
3581 .proc_handler = ipv4_sysctl_rtcache_flush,
3582 },
3583 {
3584 .procname = "min_pmtu",
3585 .data = &init_net.ipv4.ip_rt_min_pmtu,
3586 .maxlen = sizeof(int),
3587 .mode = 0644,
3588 .proc_handler = proc_dointvec_minmax,
3589 .extra1 = &ip_min_valid_pmtu,
3590 },
3591 {
3592 .procname = "mtu_expires",
3593 .data = &init_net.ipv4.ip_rt_mtu_expires,
3594 .maxlen = sizeof(int),
3595 .mode = 0644,
3596 .proc_handler = proc_dointvec_jiffies,
3597 },
3598 {
3599 .procname = "min_adv_mss",
3600 .data = &init_net.ipv4.ip_rt_min_advmss,
3601 .maxlen = sizeof(int),
3602 .mode = 0644,
3603 .proc_handler = proc_dointvec,
3604 },
3605 };
3606
sysctl_route_net_init(struct net * net)3607 static __net_init int sysctl_route_net_init(struct net *net)
3608 {
3609 struct ctl_table *tbl;
3610 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3611
3612 tbl = ipv4_route_netns_table;
3613 if (!net_eq(net, &init_net)) {
3614 int i;
3615
3616 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3617 if (!tbl)
3618 goto err_dup;
3619
3620 /* Don't export non-whitelisted sysctls to unprivileged users */
3621 if (net->user_ns != &init_user_ns) {
3622 if (tbl[0].procname != ipv4_route_flush_procname)
3623 table_size = 0;
3624 }
3625
3626 /* Update the variables to point into the current struct net
3627 * except for the first element flush
3628 */
3629 for (i = 1; i < table_size; i++)
3630 tbl[i].data += (void *)net - (void *)&init_net;
3631 }
3632 tbl[0].extra1 = net;
3633
3634 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3635 tbl, table_size);
3636 if (!net->ipv4.route_hdr)
3637 goto err_reg;
3638 return 0;
3639
3640 err_reg:
3641 if (tbl != ipv4_route_netns_table)
3642 kfree(tbl);
3643 err_dup:
3644 return -ENOMEM;
3645 }
3646
sysctl_route_net_exit(struct net * net)3647 static __net_exit void sysctl_route_net_exit(struct net *net)
3648 {
3649 const struct ctl_table *tbl;
3650
3651 tbl = net->ipv4.route_hdr->ctl_table_arg;
3652 unregister_net_sysctl_table(net->ipv4.route_hdr);
3653 BUG_ON(tbl == ipv4_route_netns_table);
3654 kfree(tbl);
3655 }
3656
3657 static __net_initdata struct pernet_operations sysctl_route_ops = {
3658 .init = sysctl_route_net_init,
3659 .exit = sysctl_route_net_exit,
3660 };
3661 #endif
3662
netns_ip_rt_init(struct net * net)3663 static __net_init int netns_ip_rt_init(struct net *net)
3664 {
3665 /* Set default value for namespaceified sysctls */
3666 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3667 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3668 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3669 return 0;
3670 }
3671
3672 static struct pernet_operations __net_initdata ip_rt_ops = {
3673 .init = netns_ip_rt_init,
3674 };
3675
rt_genid_init(struct net * net)3676 static __net_init int rt_genid_init(struct net *net)
3677 {
3678 atomic_set(&net->ipv4.rt_genid, 0);
3679 atomic_set(&net->fnhe_genid, 0);
3680 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3681 return 0;
3682 }
3683
3684 static __net_initdata struct pernet_operations rt_genid_ops = {
3685 .init = rt_genid_init,
3686 };
3687
ipv4_inetpeer_init(struct net * net)3688 static int __net_init ipv4_inetpeer_init(struct net *net)
3689 {
3690 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3691
3692 if (!bp)
3693 return -ENOMEM;
3694 inet_peer_base_init(bp);
3695 net->ipv4.peers = bp;
3696 return 0;
3697 }
3698
ipv4_inetpeer_exit(struct net * net)3699 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3700 {
3701 struct inet_peer_base *bp = net->ipv4.peers;
3702
3703 net->ipv4.peers = NULL;
3704 inetpeer_invalidate_tree(bp);
3705 kfree(bp);
3706 }
3707
3708 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3709 .init = ipv4_inetpeer_init,
3710 .exit = ipv4_inetpeer_exit,
3711 };
3712
3713 #ifdef CONFIG_IP_ROUTE_CLASSID
3714 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3715 #endif /* CONFIG_IP_ROUTE_CLASSID */
3716
3717 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3718 {.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3719 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3720 };
3721
ip_rt_init(void)3722 int __init ip_rt_init(void)
3723 {
3724 void *idents_hash;
3725 int cpu;
3726
3727 /* For modern hosts, this will use 2 MB of memory */
3728 idents_hash = alloc_large_system_hash("IP idents",
3729 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3730 0,
3731 16, /* one bucket per 64 KB */
3732 HASH_ZERO,
3733 NULL,
3734 &ip_idents_mask,
3735 2048,
3736 256*1024);
3737
3738 ip_idents = idents_hash;
3739
3740 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3741
3742 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3743
3744 for_each_possible_cpu(cpu) {
3745 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3746
3747 INIT_LIST_HEAD(&ul->head);
3748 spin_lock_init(&ul->lock);
3749 }
3750 #ifdef CONFIG_IP_ROUTE_CLASSID
3751 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3752 if (!ip_rt_acct)
3753 panic("IP: failed to allocate ip_rt_acct\n");
3754 #endif
3755
3756 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3757 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3758
3759 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3760
3761 if (dst_entries_init(&ipv4_dst_ops) < 0)
3762 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3763
3764 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3765 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3766
3767 ipv4_dst_ops.gc_thresh = ~0;
3768 ip_rt_max_size = INT_MAX;
3769
3770 devinet_init();
3771 ip_fib_init();
3772
3773 if (ip_rt_proc_init())
3774 pr_err("Unable to create route proc files\n");
3775 #ifdef CONFIG_XFRM
3776 xfrm_init();
3777 xfrm4_init();
3778 #endif
3779 rtnl_register_many(ip_rt_rtnl_msg_handlers);
3780
3781 #ifdef CONFIG_SYSCTL
3782 register_pernet_subsys(&sysctl_route_ops);
3783 #endif
3784 register_pernet_subsys(&ip_rt_ops);
3785 register_pernet_subsys(&rt_genid_ops);
3786 register_pernet_subsys(&ipv4_inetpeer_ops);
3787 return 0;
3788 }
3789
3790 #ifdef CONFIG_SYSCTL
3791 /*
3792 * We really need to sanitize the damn ipv4 init order, then all
3793 * this nonsense will go away.
3794 */
ip_static_sysctl_init(void)3795 void __init ip_static_sysctl_init(void)
3796 {
3797 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3798 }
3799 #endif
3800