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