projects/linux·blob master · refs

3833 lines · 96.7 KiB

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, <[email protected]>
11 * Alan Cox, <[email protected]>
12 * Linus Torvalds, <[email protected]>
13 * Alexey Kuznetsov, <[email protected]>
14 *
15 * Fixes:
16 * Alan Cox : Verify area fixes.
17 * Alan Cox : cli() protects routing changes
18 * Rui Oliveira : ICMP routing table updates
19 * ([email protected]) 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
115static int ip_rt_max_size;
116static int ip_rt_redirect_number __read_mostly = 9;
117static int ip_rt_redirect_load __read_mostly = HZ / 50;
118static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
119static int ip_rt_error_cost __read_mostly = HZ;
120static int ip_rt_error_burst __read_mostly = 5 * HZ;
121
122static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
123
124/*
125 * Interface to generic destination cache.
126 */
127
128INDIRECT_CALLABLE_SCOPE
129struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
130static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
131INDIRECT_CALLABLE_SCOPE
132unsigned int ipv4_mtu(const struct dst_entry *dst);
133static void ipv4_negative_advice(struct sock *sk,
134 struct dst_entry *dst);
135static void ipv4_link_failure(struct sk_buff *skb);
136static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
137 struct sk_buff *skb, u32 mtu,
138 bool confirm_neigh);
139static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
140 struct sk_buff *skb);
141static void ipv4_dst_destroy(struct dst_entry *dst);
142
143static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
144{
145 WARN_ON(1);
146 return NULL;
147}
148
149static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
150 struct sk_buff *skb,
151 const void *daddr);
152static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
153
154static 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
172const __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};
190EXPORT_SYMBOL(ip_tos2prio);
191
192static 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
200static 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
207static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
208{
209 ++*pos;
210 return NULL;
211}
212
213static void rt_cache_seq_stop(struct seq_file *seq, void *v)
214{
215}
216
217static 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
227static 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
234static 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
250static 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
265static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
266{
267
268}
269
270static 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
305static 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
313static 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 = kzalloc_objs(struct ip_rt_acct, 256);
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
338static 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
361err3:
362 remove_proc_entry("rt_cache", net->proc_net_stat);
363#endif
364err2:
365 remove_proc_entry("rt_cache", net->proc_net);
366err1:
367 return -ENOMEM;
368}
369
370static 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
379static 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
384static int __init ip_rt_proc_init(void)
385{
386 return register_pernet_subsys(&ip_rt_proc_ops);
387}
388
389#else
390static inline int ip_rt_proc_init(void)
391{
392 return 0;
393}
394#endif /* CONFIG_PROC_FS */
395
396static 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
407void rt_cache_flush(struct net *net)
408{
409 rt_genid_bump_ipv4(net);
410}
411
412static 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
441static 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 (IS_ENABLED(CONFIG_IPV6) && rt->rt_gw_family == AF_INET6) {
450 return __ipv6_confirm_neigh(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 */
462static u32 ip_idents_mask __read_mostly;
463static atomic_t *ip_idents __read_mostly;
464static 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 */
470static 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
492void __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}
508EXPORT_SYMBOL(__ip_select_ident);
509
510static 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
531static 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
544static 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
565static 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
574static DEFINE_SPINLOCK(fnhe_lock);
575
576static 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
594static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
595{
596 struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
597 struct fib_nh_exception *fnhe, *oldest = NULL;
598
599 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
600 fnhe = rcu_dereference_protected(*fnhe_p,
601 lockdep_is_held(&fnhe_lock));
602 if (!fnhe)
603 break;
604 if (!oldest ||
605 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
606 oldest = fnhe;
607 oldest_p = fnhe_p;
608 }
609 }
610
611 /* Clear oldest->fnhe_daddr to prevent this fnhe from being
612 * rebound with new dsts in rt_bind_exception().
613 */
614 oldest->fnhe_daddr = 0;
615 fnhe_flush_routes(oldest);
616 *oldest_p = oldest->fnhe_next;
617 kfree_rcu(oldest, rcu);
618}
619
620static u32 fnhe_hashfun(__be32 daddr)
621{
622 static siphash_aligned_key_t fnhe_hash_key;
623 u64 hval;
624
625 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
626 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
627 return hash_64(hval, FNHE_HASH_SHIFT);
628}
629
630static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
631{
632 rt->rt_pmtu = fnhe->fnhe_pmtu;
633 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
634 rt->dst.expires = fnhe->fnhe_expires;
635
636 if (fnhe->fnhe_gw) {
637 rt->rt_flags |= RTCF_REDIRECTED;
638 rt->rt_uses_gateway = 1;
639 rt->rt_gw_family = AF_INET;
640 rt->rt_gw4 = fnhe->fnhe_gw;
641 }
642}
643
644static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
645 __be32 gw, u32 pmtu, bool lock,
646 unsigned long expires)
647{
648 struct fnhe_hash_bucket *hash;
649 struct fib_nh_exception *fnhe;
650 struct rtable *rt;
651 u32 genid, hval;
652 unsigned int i;
653 int depth;
654
655 genid = fnhe_genid(dev_net(nhc->nhc_dev));
656 hval = fnhe_hashfun(daddr);
657
658 spin_lock_bh(&fnhe_lock);
659
660 hash = rcu_dereference(nhc->nhc_exceptions);
661 if (!hash) {
662 hash = kzalloc_objs(*hash, FNHE_HASH_SIZE, GFP_ATOMIC);
663 if (!hash)
664 goto out_unlock;
665 rcu_assign_pointer(nhc->nhc_exceptions, hash);
666 }
667
668 hash += hval;
669
670 depth = 0;
671 for (fnhe = rcu_dereference(hash->chain); fnhe;
672 fnhe = rcu_dereference(fnhe->fnhe_next)) {
673 if (fnhe->fnhe_daddr == daddr)
674 break;
675 depth++;
676 }
677
678 if (fnhe) {
679 if (fnhe->fnhe_genid != genid)
680 fnhe->fnhe_genid = genid;
681 if (gw)
682 fnhe->fnhe_gw = gw;
683 if (pmtu) {
684 fnhe->fnhe_pmtu = pmtu;
685 fnhe->fnhe_mtu_locked = lock;
686 }
687 fnhe->fnhe_expires = max(1UL, expires);
688 /* Update all cached dsts too */
689 rt = rcu_dereference(fnhe->fnhe_rth_input);
690 if (rt)
691 fill_route_from_fnhe(rt, fnhe);
692 rt = rcu_dereference(fnhe->fnhe_rth_output);
693 if (rt)
694 fill_route_from_fnhe(rt, fnhe);
695 } else {
696 /* Randomize max depth to avoid some side channels attacks. */
697 int max_depth = FNHE_RECLAIM_DEPTH +
698 get_random_u32_below(FNHE_RECLAIM_DEPTH);
699
700 while (depth > max_depth) {
701 fnhe_remove_oldest(hash);
702 depth--;
703 }
704
705 fnhe = kzalloc_obj(*fnhe, GFP_ATOMIC);
706 if (!fnhe)
707 goto out_unlock;
708
709 fnhe->fnhe_next = hash->chain;
710
711 fnhe->fnhe_genid = genid;
712 fnhe->fnhe_daddr = daddr;
713 fnhe->fnhe_gw = gw;
714 fnhe->fnhe_pmtu = pmtu;
715 fnhe->fnhe_mtu_locked = lock;
716 fnhe->fnhe_expires = max(1UL, expires);
717
718 rcu_assign_pointer(hash->chain, fnhe);
719
720 /* Exception created; mark the cached routes for the nexthop
721 * stale, so anyone caching it rechecks if this exception
722 * applies to them.
723 */
724 rt = rcu_dereference(nhc->nhc_rth_input);
725 if (rt)
726 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
727
728 for_each_possible_cpu(i) {
729 struct rtable __rcu **prt;
730
731 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
732 rt = rcu_dereference(*prt);
733 if (rt)
734 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
735 }
736 }
737
738 fnhe->fnhe_stamp = jiffies;
739
740out_unlock:
741 spin_unlock_bh(&fnhe_lock);
742}
743
744/* Update the PMTU of an exception when:
745 * - the new MTU of the first hop becomes smaller than the PMTU
746 * - the old MTU was the same as the PMTU, and it limited discovery of
747 * larger MTUs on the path. With that limit raised, we can now
748 * discover larger MTUs
749 * A special case is locked exceptions, for which the PMTU is smaller
750 * than the minimal accepted PMTU:
751 * - if the new MTU is greater than the PMTU, don't make any change
752 * - otherwise, unlock and set PMTU
753 *
754 * fnhe_lock keeps fnhe_pmtu and fnhe_mtu_locked consistent against
755 * update_or_create_fnhe(), which sets both under the same lock.
756 */
757void fnhe_update_pmtu(struct fib_nh_exception *fnhe, u32 new, u32 orig)
758{
759 spin_lock_bh(&fnhe_lock);
760
761 if (fnhe->fnhe_mtu_locked) {
762 if (new <= fnhe->fnhe_pmtu) {
763 fnhe->fnhe_pmtu = new;
764 fnhe->fnhe_mtu_locked = false;
765 }
766 } else if (new < fnhe->fnhe_pmtu || orig == fnhe->fnhe_pmtu) {
767 fnhe->fnhe_pmtu = new;
768 }
769
770 spin_unlock_bh(&fnhe_lock);
771}
772
773static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
774 bool kill_route)
775{
776 __be32 new_gw = icmp_hdr(skb)->un.gateway;
777 __be32 old_gw = ip_hdr(skb)->saddr;
778 struct net_device *dev = skb->dev;
779 struct in_device *in_dev;
780 struct fib_result res;
781 struct neighbour *n;
782 struct net *net;
783
784 switch (icmp_hdr(skb)->code & 7) {
785 case ICMP_REDIR_NET:
786 case ICMP_REDIR_NETTOS:
787 case ICMP_REDIR_HOST:
788 case ICMP_REDIR_HOSTTOS:
789 break;
790
791 default:
792 return;
793 }
794
795 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
796 return;
797
798 in_dev = __in_dev_get_rcu(dev);
799 if (!in_dev)
800 return;
801
802 net = dev_net(dev);
803 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
804 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
805 ipv4_is_zeronet(new_gw))
806 goto reject_redirect;
807
808 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
809 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
810 goto reject_redirect;
811 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
812 goto reject_redirect;
813 } else {
814 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
815 goto reject_redirect;
816 }
817
818 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
819 if (!n)
820 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
821 if (!IS_ERR(n)) {
822 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
823 neigh_event_send(n, NULL);
824 } else {
825 if (fib_lookup(net, fl4, &res, 0) == 0) {
826 struct fib_nh_common *nhc;
827
828 fib_select_path(net, &res, fl4, skb);
829 nhc = FIB_RES_NHC(res);
830 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
831 0, false,
832 jiffies + ip_rt_gc_timeout);
833 }
834 if (kill_route)
835 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL);
836 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
837 }
838 neigh_release(n);
839 }
840 return;
841
842reject_redirect:
843#ifdef CONFIG_IP_ROUTE_VERBOSE
844 if (IN_DEV_LOG_MARTIANS(in_dev)) {
845 const struct iphdr *iph = (const struct iphdr *) skb->data;
846 __be32 daddr = iph->daddr;
847 __be32 saddr = iph->saddr;
848
849 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
850 " Advised path = %pI4 -> %pI4\n",
851 &old_gw, dev->name, &new_gw,
852 &saddr, &daddr);
853 }
854#endif
855 ;
856}
857
858static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
859{
860 struct rtable *rt;
861 struct flowi4 fl4;
862 const struct iphdr *iph = (const struct iphdr *) skb->data;
863 struct net *net = dev_net(skb->dev);
864 int oif = skb->dev->ifindex;
865 u8 prot = iph->protocol;
866 u32 mark = skb->mark;
867 __u8 tos = iph->tos;
868
869 rt = dst_rtable(dst);
870
871 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
872 __ip_do_redirect(rt, skb, &fl4, true);
873}
874
875static void ipv4_negative_advice(struct sock *sk,
876 struct dst_entry *dst)
877{
878 struct rtable *rt = dst_rtable(dst);
879
880 if ((READ_ONCE(dst->obsolete) > 0) ||
881 (rt->rt_flags & RTCF_REDIRECTED) ||
882 READ_ONCE(rt->dst.expires))
883 sk_dst_reset(sk);
884}
885
886/*
887 * Algorithm:
888 * 1. The first ip_rt_redirect_number redirects are sent
889 * with exponential backoff, then we stop sending them at all,
890 * assuming that the host ignores our redirects.
891 * 2. If we did not see packets requiring redirects
892 * during ip_rt_redirect_silence, we assume that the host
893 * forgot redirected route and start to send redirects again.
894 *
895 * This algorithm is much cheaper and more intelligent than dumb load limiting
896 * in icmp.c.
897 *
898 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
899 * and "frag. need" (breaks PMTU discovery) in icmp.c.
900 */
901
902void ip_rt_send_redirect(struct sk_buff *skb)
903{
904 struct rtable *rt = skb_rtable(skb);
905 struct in_device *in_dev;
906 struct net_device *dev;
907 struct inet_peer *peer;
908 int log_martians;
909 struct net *net;
910 int vif;
911
912 rcu_read_lock();
913 dev = dst_dev_rcu(&rt->dst);
914 in_dev = __in_dev_get_rcu(dev);
915 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
916 rcu_read_unlock();
917 return;
918 }
919 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
920 vif = l3mdev_master_ifindex_rcu(dev);
921
922 net = dev_net_rcu(dev);
923 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif);
924 if (!peer) {
925 rcu_read_unlock();
926 return;
927 }
928
929 /* No redirected packets during ip_rt_redirect_silence;
930 * reset the algorithm.
931 */
932 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
933 peer->rate_tokens = 0;
934 peer->n_redirects = 0;
935 }
936
937 /* Too many ignored redirects; do not send anything
938 * set dst.rate_last to the last seen redirected packet.
939 */
940 if (peer->n_redirects >= ip_rt_redirect_number) {
941 peer->rate_last = jiffies;
942 goto out_unlock;
943 }
944
945 /* Check for load limit; set rate_last to the latest sent
946 * redirect.
947 */
948 if (peer->n_redirects == 0 ||
949 time_after(jiffies,
950 (peer->rate_last +
951 (ip_rt_redirect_load << peer->n_redirects)))) {
952 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
953
954 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
955 peer->rate_last = jiffies;
956 ++peer->n_redirects;
957 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
958 peer->n_redirects == ip_rt_redirect_number)
959 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
960 &ip_hdr(skb)->saddr, inet_iif(skb),
961 &ip_hdr(skb)->daddr, &gw);
962 }
963out_unlock:
964 rcu_read_unlock();
965}
966
967static int ip_error(struct sk_buff *skb)
968{
969 struct rtable *rt = skb_rtable(skb);
970 struct net_device *dev = skb->dev;
971 struct in_device *in_dev;
972 struct inet_peer *peer;
973 unsigned long now;
974 struct net *net;
975 SKB_DR(reason);
976 bool send;
977 int code;
978
979 if (netif_is_l3_master(skb->dev)) {
980 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
981 if (!dev)
982 goto out;
983 }
984
985 in_dev = __in_dev_get_rcu(dev);
986
987 /* IP on this device is disabled. */
988 if (!in_dev)
989 goto out;
990
991 net = dev_net(rt->dst.dev);
992 if (!IN_DEV_FORWARD(in_dev)) {
993 switch (rt->dst.error) {
994 case EHOSTUNREACH:
995 SKB_DR_SET(reason, IP_INADDRERRORS);
996 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
997 break;
998
999 case ENETUNREACH:
1000 SKB_DR_SET(reason, IP_INNOROUTES);
1001 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
1002 break;
1003 }
1004 goto out;
1005 }
1006
1007 switch (rt->dst.error) {
1008 case EINVAL:
1009 default:
1010 goto out;
1011 case EHOSTUNREACH:
1012 code = ICMP_HOST_UNREACH;
1013 break;
1014 case ENETUNREACH:
1015 code = ICMP_NET_UNREACH;
1016 SKB_DR_SET(reason, IP_INNOROUTES);
1017 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
1018 break;
1019 case EACCES:
1020 code = ICMP_PKT_FILTERED;
1021 break;
1022 }
1023
1024 rcu_read_lock();
1025 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
1026 l3mdev_master_ifindex_rcu(skb->dev));
1027 send = true;
1028 if (peer) {
1029 now = jiffies;
1030 peer->rate_tokens += now - peer->rate_last;
1031 if (peer->rate_tokens > ip_rt_error_burst)
1032 peer->rate_tokens = ip_rt_error_burst;
1033 peer->rate_last = now;
1034 if (peer->rate_tokens >= ip_rt_error_cost)
1035 peer->rate_tokens -= ip_rt_error_cost;
1036 else
1037 send = false;
1038 }
1039 rcu_read_unlock();
1040
1041 if (send)
1042 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1043
1044out: kfree_skb_reason(skb, reason);
1045 return 0;
1046}
1047
1048static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1049{
1050 struct dst_entry *dst = &rt->dst;
1051 struct fib_result res;
1052 bool lock = false;
1053 struct net *net;
1054 u32 old_mtu;
1055
1056 if (ip_mtu_locked(dst))
1057 return;
1058
1059 old_mtu = ipv4_mtu(dst);
1060 if (old_mtu < mtu)
1061 return;
1062
1063 rcu_read_lock();
1064 net = dst_dev_net_rcu(dst);
1065 if (mtu < net->ipv4.ip_rt_min_pmtu) {
1066 lock = true;
1067 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1068 }
1069
1070 if (rt->rt_pmtu == mtu && !lock &&
1071 time_before(jiffies, READ_ONCE(dst->expires) -
1072 net->ipv4.ip_rt_mtu_expires / 2))
1073 goto out;
1074
1075 if (fib_lookup(net, fl4, &res, 0) == 0) {
1076 struct fib_nh_common *nhc;
1077
1078 fib_select_path(net, &res, fl4, NULL);
1079#ifdef CONFIG_IP_ROUTE_MULTIPATH
1080 if (fib_info_num_path(res.fi) > 1) {
1081 int nhsel;
1082
1083 for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) {
1084 nhc = fib_info_nhc(res.fi, nhsel);
1085 if (!nhc)
1086 break;
1087 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1088 jiffies + net->ipv4.ip_rt_mtu_expires);
1089 }
1090 goto out;
1091 }
1092#endif /* CONFIG_IP_ROUTE_MULTIPATH */
1093 nhc = FIB_RES_NHC(res);
1094 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1095 jiffies + net->ipv4.ip_rt_mtu_expires);
1096 }
1097out:
1098 rcu_read_unlock();
1099}
1100
1101static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1102 struct sk_buff *skb, u32 mtu,
1103 bool confirm_neigh)
1104{
1105 struct rtable *rt = dst_rtable(dst);
1106 struct flowi4 fl4;
1107
1108 ip_rt_build_flow_key(&fl4, sk, skb);
1109
1110 /* Don't make lookup fail for bridged encapsulations */
1111 if (skb && netif_is_any_bridge_port(skb->dev))
1112 fl4.flowi4_oif = 0;
1113
1114 __ip_rt_update_pmtu(rt, &fl4, mtu);
1115}
1116
1117void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1118 int oif, u8 protocol)
1119{
1120 const struct iphdr *iph = (const struct iphdr *)skb->data;
1121 struct flowi4 fl4;
1122 struct rtable *rt;
1123 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1124
1125 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1126 0);
1127 rt = __ip_route_output_key(net, &fl4);
1128 if (!IS_ERR(rt)) {
1129 __ip_rt_update_pmtu(rt, &fl4, mtu);
1130 ip_rt_put(rt);
1131 }
1132}
1133EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1134
1135static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1136{
1137 const struct iphdr *iph = (const struct iphdr *)skb->data;
1138 struct flowi4 fl4;
1139 struct rtable *rt;
1140
1141 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1142
1143 if (!fl4.flowi4_mark)
1144 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1145
1146 rt = __ip_route_output_key(sock_net(sk), &fl4);
1147 if (!IS_ERR(rt)) {
1148 __ip_rt_update_pmtu(rt, &fl4, mtu);
1149 ip_rt_put(rt);
1150 }
1151}
1152
1153void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1154{
1155 const struct iphdr *iph = (const struct iphdr *)skb->data;
1156 struct flowi4 fl4;
1157 struct rtable *rt;
1158 struct dst_entry *odst = NULL;
1159 bool new = false;
1160 struct net *net = sock_net(sk);
1161
1162 bh_lock_sock(sk);
1163
1164 if (!ip_sk_accept_pmtu(sk))
1165 goto out;
1166
1167 odst = sk_dst_get(sk);
1168
1169 if (sock_owned_by_user(sk) || !odst) {
1170 __ipv4_sk_update_pmtu(skb, sk, mtu);
1171 goto out;
1172 }
1173
1174 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1175
1176 rt = dst_rtable(odst);
1177 if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) {
1178 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1179 if (IS_ERR(rt))
1180 goto out;
1181
1182 new = true;
1183 }
1184
1185 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1186
1187 if (!dst_check(&rt->dst, 0)) {
1188 if (new)
1189 dst_release(&rt->dst);
1190
1191 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1192 if (IS_ERR(rt))
1193 goto out;
1194
1195 new = true;
1196 }
1197
1198 if (new)
1199 sk_dst_set(sk, &rt->dst);
1200
1201out:
1202 bh_unlock_sock(sk);
1203 dst_release(odst);
1204}
1205
1206void ipv4_redirect(struct sk_buff *skb, struct net *net,
1207 int oif, u8 protocol)
1208{
1209 const struct iphdr *iph = (const struct iphdr *)skb->data;
1210 struct flowi4 fl4;
1211 struct rtable *rt;
1212
1213 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1214 rt = __ip_route_output_key(net, &fl4);
1215 if (!IS_ERR(rt)) {
1216 __ip_do_redirect(rt, skb, &fl4, false);
1217 ip_rt_put(rt);
1218 }
1219}
1220EXPORT_SYMBOL_GPL(ipv4_redirect);
1221
1222void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1223{
1224 const struct iphdr *iph = (const struct iphdr *)skb->data;
1225 struct flowi4 fl4;
1226 struct rtable *rt;
1227 struct net *net = sock_net(sk);
1228
1229 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1230 rt = __ip_route_output_key(net, &fl4);
1231 if (!IS_ERR(rt)) {
1232 __ip_do_redirect(rt, skb, &fl4, false);
1233 ip_rt_put(rt);
1234 }
1235}
1236
1237INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1238 u32 cookie)
1239{
1240 struct rtable *rt = dst_rtable(dst);
1241
1242 /* All IPV4 dsts are created with ->obsolete set to the value
1243 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1244 * into this function always.
1245 *
1246 * When a PMTU/redirect information update invalidates a route,
1247 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1248 * DST_OBSOLETE_DEAD.
1249 */
1250 if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK ||
1251 rt_is_expired(rt))
1252 return NULL;
1253 return dst;
1254}
1255EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1256
1257static void ipv4_send_dest_unreach(struct sk_buff *skb)
1258{
1259 struct inet_skb_parm parm;
1260 struct net_device *dev;
1261 int res;
1262
1263 /* Recompile ip options since IPCB may not be valid anymore.
1264 * Also check we have a reasonable ipv4 header.
1265 */
1266 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1267 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1268 return;
1269
1270 memset(&parm, 0, sizeof(parm));
1271 if (ip_hdr(skb)->ihl > 5) {
1272 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1273 return;
1274 parm.opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1275
1276 rcu_read_lock();
1277 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1278 res = __ip_options_compile(dev_net(dev), &parm.opt, skb, NULL);
1279 rcu_read_unlock();
1280
1281 if (res)
1282 return;
1283 }
1284 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &parm);
1285}
1286
1287static void ipv4_link_failure(struct sk_buff *skb)
1288{
1289 struct rtable *rt;
1290
1291 ipv4_send_dest_unreach(skb);
1292
1293 rt = skb_rtable(skb);
1294 if (rt)
1295 dst_set_expires(&rt->dst, 0);
1296}
1297
1298static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1299{
1300 pr_debug("%s: %pI4 -> %pI4, %s\n",
1301 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1302 skb->dev ? skb->dev->name : "?");
1303 kfree_skb(skb);
1304 WARN_ON_ONCE(1);
1305 return 0;
1306}
1307
1308/*
1309 * We do not cache source address of outgoing interface,
1310 * because it is used only by IP RR, TS and SRR options,
1311 * so that it out of fast path.
1312 *
1313 * BTW remember: "addr" is allowed to be not aligned
1314 * in IP options!
1315 */
1316
1317void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1318{
1319 __be32 src;
1320
1321 rcu_read_lock();
1322 if (rt_is_output_route(rt)) {
1323 src = ip_hdr(skb)->saddr;
1324 } else {
1325 struct net_device *dev = dst_dev_rcu(&rt->dst);
1326 struct net *net = dev_net_rcu(dev);
1327 struct iphdr *iph = ip_hdr(skb);
1328 struct fib_result res;
1329 struct flowi4 fl4 = {
1330 .daddr = iph->daddr,
1331 .saddr = iph->saddr,
1332 .flowi4_dscp = ip4h_dscp(iph),
1333 .flowi4_oif = dev->ifindex,
1334 .flowi4_iif = skb->dev->ifindex,
1335 .flowi4_mark = skb->mark,
1336 };
1337
1338 if (fib_lookup(net, &fl4, &res, 0) == 0)
1339 src = fib_result_prefsrc(net, &res);
1340 else
1341 src = inet_select_addr(dev,
1342 rt_nexthop(rt, iph->daddr),
1343 RT_SCOPE_UNIVERSE);
1344 }
1345 rcu_read_unlock();
1346
1347 memcpy(addr, &src, 4);
1348}
1349
1350#ifdef CONFIG_IP_ROUTE_CLASSID
1351static void set_class_tag(struct rtable *rt, u32 tag)
1352{
1353 if (!(rt->dst.tclassid & 0xFFFF))
1354 rt->dst.tclassid |= tag & 0xFFFF;
1355 if (!(rt->dst.tclassid & 0xFFFF0000))
1356 rt->dst.tclassid |= tag & 0xFFFF0000;
1357}
1358#endif
1359
1360static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1361{
1362 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1363 unsigned int advmss;
1364 struct net *net;
1365
1366 rcu_read_lock();
1367 net = dst_dev_net_rcu(dst);
1368 advmss = max_t(unsigned int, ip_dst_mtu_configured(dst) - header_size,
1369 net->ipv4.ip_rt_min_advmss);
1370 rcu_read_unlock();
1371
1372 return min(advmss, IPV4_MAX_PMTU - header_size);
1373}
1374
1375INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1376{
1377 return ip_dst_mtu_maybe_forward(dst, false);
1378}
1379EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1380
1381static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1382{
1383 struct fnhe_hash_bucket *hash;
1384 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1385 u32 hval = fnhe_hashfun(daddr);
1386
1387 spin_lock_bh(&fnhe_lock);
1388
1389 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1390 lockdep_is_held(&fnhe_lock));
1391 hash += hval;
1392
1393 fnhe_p = &hash->chain;
1394 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1395 while (fnhe) {
1396 if (fnhe->fnhe_daddr == daddr) {
1397 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1398 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1399 /* set fnhe_daddr to 0 to ensure it won't bind with
1400 * new dsts in rt_bind_exception().
1401 */
1402 fnhe->fnhe_daddr = 0;
1403 fnhe_flush_routes(fnhe);
1404 kfree_rcu(fnhe, rcu);
1405 break;
1406 }
1407 fnhe_p = &fnhe->fnhe_next;
1408 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1409 lockdep_is_held(&fnhe_lock));
1410 }
1411
1412 spin_unlock_bh(&fnhe_lock);
1413}
1414
1415static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1416 __be32 daddr)
1417{
1418 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1419 struct fib_nh_exception *fnhe;
1420 u32 hval;
1421
1422 if (!hash)
1423 return NULL;
1424
1425 hval = fnhe_hashfun(daddr);
1426
1427 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1428 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1429 if (fnhe->fnhe_daddr == daddr) {
1430 if (fnhe->fnhe_expires &&
1431 time_after(jiffies, fnhe->fnhe_expires)) {
1432 ip_del_fnhe(nhc, daddr);
1433 break;
1434 }
1435 return fnhe;
1436 }
1437 }
1438 return NULL;
1439}
1440
1441/* MTU selection:
1442 * 1. mtu on route is locked - use it
1443 * 2. mtu from nexthop exception
1444 * 3. mtu from egress device
1445 */
1446
1447u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1448{
1449 struct fib_nh_common *nhc = res->nhc;
1450 struct net_device *dev = nhc->nhc_dev;
1451 struct fib_info *fi = res->fi;
1452 u32 mtu = 0;
1453
1454 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1455 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1456 mtu = fi->fib_mtu;
1457
1458 if (likely(!mtu)) {
1459 struct fib_nh_exception *fnhe;
1460
1461 fnhe = find_exception(nhc, daddr);
1462 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1463 mtu = fnhe->fnhe_pmtu;
1464 }
1465
1466 if (likely(!mtu))
1467 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1468
1469 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1470}
1471
1472static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1473 __be32 daddr, const bool do_cache)
1474{
1475 bool ret = false;
1476
1477 spin_lock_bh(&fnhe_lock);
1478
1479 if (daddr == fnhe->fnhe_daddr) {
1480 struct rtable __rcu **porig;
1481 struct rtable *orig;
1482 int genid = fnhe_genid(dev_net(rt->dst.dev));
1483
1484 if (rt_is_input_route(rt))
1485 porig = &fnhe->fnhe_rth_input;
1486 else
1487 porig = &fnhe->fnhe_rth_output;
1488 orig = rcu_dereference(*porig);
1489
1490 if (fnhe->fnhe_genid != genid) {
1491 fnhe->fnhe_genid = genid;
1492 fnhe->fnhe_gw = 0;
1493 fnhe->fnhe_pmtu = 0;
1494 fnhe->fnhe_expires = 0;
1495 fnhe->fnhe_mtu_locked = false;
1496 fnhe_flush_routes(fnhe);
1497 orig = NULL;
1498 }
1499 fill_route_from_fnhe(rt, fnhe);
1500 if (!rt->rt_gw4) {
1501 rt->rt_gw4 = daddr;
1502 rt->rt_gw_family = AF_INET;
1503 }
1504
1505 if (do_cache) {
1506 dst_hold(&rt->dst);
1507 rcu_assign_pointer(*porig, rt);
1508 if (orig) {
1509 dst_dev_put(&orig->dst);
1510 dst_release(&orig->dst);
1511 }
1512 ret = true;
1513 }
1514
1515 fnhe->fnhe_stamp = jiffies;
1516 }
1517 spin_unlock_bh(&fnhe_lock);
1518
1519 return ret;
1520}
1521
1522static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1523{
1524 struct rtable *orig, *prev, **p;
1525 bool ret = true;
1526
1527 if (rt_is_input_route(rt)) {
1528 p = (struct rtable **)&nhc->nhc_rth_input;
1529 } else {
1530 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1531 }
1532 orig = *p;
1533
1534 /* hold dst before doing cmpxchg() to avoid race condition
1535 * on this dst
1536 */
1537 dst_hold(&rt->dst);
1538 prev = cmpxchg(p, orig, rt);
1539 if (prev == orig) {
1540 if (orig) {
1541 rt_add_uncached_list(orig);
1542 dst_release(&orig->dst);
1543 }
1544 } else {
1545 dst_release(&rt->dst);
1546 ret = false;
1547 }
1548
1549 return ret;
1550}
1551
1552struct uncached_list {
1553 spinlock_t lock;
1554 struct list_head head;
1555};
1556
1557static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1558
1559void rt_add_uncached_list(struct rtable *rt)
1560{
1561 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1562
1563 rt->dst.rt_uncached_list = ul;
1564
1565 spin_lock_bh(&ul->lock);
1566 list_add_tail(&rt->dst.rt_uncached, &ul->head);
1567 spin_unlock_bh(&ul->lock);
1568}
1569
1570void rt_del_uncached_list(struct rtable *rt)
1571{
1572 struct uncached_list *ul = rt->dst.rt_uncached_list;
1573
1574 if (ul) {
1575 spin_lock_bh(&ul->lock);
1576 list_del_init(&rt->dst.rt_uncached);
1577 spin_unlock_bh(&ul->lock);
1578 }
1579}
1580
1581static void ipv4_dst_destroy(struct dst_entry *dst)
1582{
1583 ip_dst_metrics_put(dst);
1584 rt_del_uncached_list(dst_rtable(dst));
1585}
1586
1587void rt_flush_dev(struct net_device *dev)
1588{
1589 struct rtable *rt, *safe;
1590 int cpu;
1591
1592 for_each_possible_cpu(cpu) {
1593 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1594
1595 if (list_empty(&ul->head))
1596 continue;
1597
1598 spin_lock_bh(&ul->lock);
1599 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1600 if (rt->dst.dev != dev)
1601 continue;
1602 rcu_assign_pointer(rt->dst.dev_rcu, blackhole_netdev);
1603 netdev_ref_replace(dev, blackhole_netdev,
1604 &rt->dst.dev_tracker, GFP_ATOMIC);
1605 list_del_init(&rt->dst.rt_uncached);
1606 }
1607 spin_unlock_bh(&ul->lock);
1608 }
1609}
1610
1611static bool rt_cache_valid(const struct rtable *rt)
1612{
1613 return rt &&
1614 READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK &&
1615 !rt_is_expired(rt);
1616}
1617
1618static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1619 const struct fib_result *res,
1620 struct fib_nh_exception *fnhe,
1621 struct fib_info *fi, u16 type, u32 itag,
1622 const bool do_cache)
1623{
1624 bool cached = false;
1625
1626 if (fi) {
1627 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1628
1629 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1630 rt->rt_uses_gateway = 1;
1631 rt->rt_gw_family = nhc->nhc_gw_family;
1632 /* only INET and INET6 are supported */
1633 if (likely(nhc->nhc_gw_family == AF_INET))
1634 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1635 else
1636 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1637 }
1638
1639 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1640
1641#ifdef CONFIG_IP_ROUTE_CLASSID
1642 if (nhc->nhc_family == AF_INET) {
1643 struct fib_nh *nh;
1644
1645 nh = container_of(nhc, struct fib_nh, nh_common);
1646 rt->dst.tclassid = nh->nh_tclassid;
1647 }
1648#endif
1649 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1650 if (unlikely(fnhe))
1651 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1652 else if (do_cache)
1653 cached = rt_cache_route(nhc, rt);
1654 if (unlikely(!cached)) {
1655 /* Routes we intend to cache in nexthop exception or
1656 * FIB nexthop have the DST_NOCACHE bit clear.
1657 * However, if we are unsuccessful at storing this
1658 * route into the cache we really need to set it.
1659 */
1660 if (!rt->rt_gw4) {
1661 rt->rt_gw_family = AF_INET;
1662 rt->rt_gw4 = daddr;
1663 }
1664 rt_add_uncached_list(rt);
1665 }
1666 } else
1667 rt_add_uncached_list(rt);
1668
1669#ifdef CONFIG_IP_ROUTE_CLASSID
1670#ifdef CONFIG_IP_MULTIPLE_TABLES
1671 set_class_tag(rt, res->tclassid);
1672#endif
1673 set_class_tag(rt, itag);
1674#endif
1675}
1676
1677struct rtable *rt_dst_alloc(struct net_device *dev,
1678 unsigned int flags, u16 type,
1679 bool noxfrm)
1680{
1681 struct rtable *rt;
1682
1683 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1684 (noxfrm ? DST_NOXFRM : 0));
1685
1686 if (rt) {
1687 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1688 rt->rt_flags = flags;
1689 rt->rt_type = type;
1690 rt->rt_is_input = 0;
1691 rt->rt_iif = 0;
1692 rt->rt_pmtu = 0;
1693 rt->rt_mtu_locked = 0;
1694 rt->rt_uses_gateway = 0;
1695 rt->rt_gw_family = 0;
1696 rt->rt_gw4 = 0;
1697
1698 rt->dst.output = ip_output;
1699 if (flags & RTCF_LOCAL)
1700 rt->dst.input = ip_local_deliver;
1701 }
1702
1703 return rt;
1704}
1705EXPORT_SYMBOL(rt_dst_alloc);
1706
1707struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1708{
1709 struct rtable *new_rt;
1710
1711 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1712 rt->dst.flags);
1713
1714 if (new_rt) {
1715 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1716 new_rt->rt_flags = rt->rt_flags;
1717 new_rt->rt_type = rt->rt_type;
1718 new_rt->rt_is_input = rt->rt_is_input;
1719 new_rt->rt_iif = rt->rt_iif;
1720 new_rt->rt_pmtu = rt->rt_pmtu;
1721 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1722 new_rt->rt_gw_family = rt->rt_gw_family;
1723 if (rt->rt_gw_family == AF_INET)
1724 new_rt->rt_gw4 = rt->rt_gw4;
1725 else if (rt->rt_gw_family == AF_INET6)
1726 new_rt->rt_gw6 = rt->rt_gw6;
1727
1728 new_rt->dst.input = READ_ONCE(rt->dst.input);
1729 new_rt->dst.output = READ_ONCE(rt->dst.output);
1730 new_rt->dst.error = rt->dst.error;
1731 new_rt->dst.lastuse = jiffies;
1732 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1733 }
1734 return new_rt;
1735}
1736
1737/* called in rcu_read_lock() section */
1738enum skb_drop_reason
1739ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1740 dscp_t dscp, struct net_device *dev,
1741 struct in_device *in_dev, u32 *itag)
1742{
1743 enum skb_drop_reason reason;
1744
1745 /* Primary sanity checks. */
1746 if (!in_dev)
1747 return SKB_DROP_REASON_NOT_SPECIFIED;
1748
1749 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1750 return SKB_DROP_REASON_IP_INVALID_SOURCE;
1751
1752 if (skb->protocol != htons(ETH_P_IP))
1753 return SKB_DROP_REASON_INVALID_PROTO;
1754
1755 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1756 return SKB_DROP_REASON_IP_LOCALNET;
1757
1758 if (ipv4_is_zeronet(saddr)) {
1759 if (!ipv4_is_local_multicast(daddr) &&
1760 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1761 return SKB_DROP_REASON_IP_INVALID_SOURCE;
1762 } else {
1763 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
1764 dev, in_dev, itag);
1765 if (reason)
1766 return reason;
1767 }
1768 return SKB_NOT_DROPPED_YET;
1769}
1770
1771/* called in rcu_read_lock() section */
1772static enum skb_drop_reason
1773ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1774 dscp_t dscp, struct net_device *dev, int our)
1775{
1776 struct in_device *in_dev = __in_dev_get_rcu(dev);
1777 unsigned int flags = RTCF_MULTICAST;
1778 enum skb_drop_reason reason;
1779 struct rtable *rth;
1780 u32 itag = 0;
1781
1782 reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev,
1783 &itag);
1784 if (reason)
1785 return reason;
1786
1787 if (our)
1788 flags |= RTCF_LOCAL;
1789
1790 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1791 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1792
1793 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1794 false);
1795 if (!rth)
1796 return SKB_DROP_REASON_NOMEM;
1797
1798#ifdef CONFIG_IP_ROUTE_CLASSID
1799 rth->dst.tclassid = itag;
1800#endif
1801 rth->dst.output = ip_rt_bug;
1802 rth->rt_is_input= 1;
1803
1804#ifdef CONFIG_IP_MROUTE
1805 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1806 rth->dst.input = ip_mr_input;
1807#endif
1808 RT_CACHE_STAT_INC(in_slow_mc);
1809
1810 skb_dst_drop(skb);
1811 skb_dst_set(skb, &rth->dst);
1812 return SKB_NOT_DROPPED_YET;
1813}
1814
1815
1816static void ip_handle_martian_source(struct net_device *dev,
1817 struct in_device *in_dev,
1818 struct sk_buff *skb,
1819 __be32 daddr,
1820 __be32 saddr)
1821{
1822 RT_CACHE_STAT_INC(in_martian_src);
1823#ifdef CONFIG_IP_ROUTE_VERBOSE
1824 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1825 /*
1826 * RFC1812 recommendation, if source is martian,
1827 * the only hint is MAC header.
1828 */
1829 pr_warn("martian source (src=%pI4, dst=%pI4, dev=%s)\n",
1830 &saddr, &daddr, dev->name);
1831 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1832 print_hex_dump(KERN_WARNING, "ll header: ",
1833 DUMP_PREFIX_OFFSET, 16, 1,
1834 skb_mac_header(skb),
1835 dev->hard_header_len, false);
1836 }
1837 }
1838#endif
1839}
1840
1841/* called in rcu_read_lock() section */
1842static enum skb_drop_reason
1843__mkroute_input(struct sk_buff *skb, const struct fib_result *res,
1844 struct in_device *in_dev, __be32 daddr,
1845 __be32 saddr, dscp_t dscp)
1846{
1847 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1848 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1849 struct net_device *dev = nhc->nhc_dev;
1850 struct fib_nh_exception *fnhe;
1851 struct rtable *rth;
1852 int err;
1853 struct in_device *out_dev;
1854 bool do_cache;
1855 u32 itag = 0;
1856
1857 /* get a working reference to the output device */
1858 out_dev = __in_dev_get_rcu(dev);
1859 if (!out_dev) {
1860 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1861 return reason;
1862 }
1863
1864 err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res),
1865 in_dev->dev, in_dev, &itag);
1866 if (err < 0) {
1867 reason = -err;
1868 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1869 saddr);
1870
1871 goto cleanup;
1872 }
1873
1874 do_cache = res->fi && !itag;
1875 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1876 skb->protocol == htons(ETH_P_IP)) {
1877 __be32 gw;
1878
1879 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1880 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1881 inet_addr_onlink(out_dev, saddr, gw))
1882 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1883 }
1884
1885 if (skb->protocol != htons(ETH_P_IP)) {
1886 /* Not IP (i.e. ARP). Do not create route, if it is
1887 * invalid for proxy arp. DNAT routes are always valid.
1888 *
1889 * Proxy arp feature have been extended to allow, ARP
1890 * replies back to the same interface, to support
1891 * Private VLAN switch technologies. See arp.c.
1892 */
1893 if (out_dev == in_dev &&
1894 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1895 reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE;
1896 goto cleanup;
1897 }
1898 }
1899
1900 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1901 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1902
1903 fnhe = find_exception(nhc, daddr);
1904 if (do_cache) {
1905 if (fnhe)
1906 rth = rcu_dereference(fnhe->fnhe_rth_input);
1907 else
1908 rth = rcu_dereference(nhc->nhc_rth_input);
1909 if (rt_cache_valid(rth)) {
1910 skb_dst_set_noref(skb, &rth->dst);
1911 goto out;
1912 }
1913 }
1914
1915 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1916 IN_DEV_ORCONF(out_dev, NOXFRM));
1917 if (!rth) {
1918 reason = SKB_DROP_REASON_NOMEM;
1919 goto cleanup;
1920 }
1921
1922 rth->rt_is_input = 1;
1923 RT_CACHE_STAT_INC(in_slow_tot);
1924
1925 rth->dst.input = ip_forward;
1926
1927 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1928 do_cache);
1929 lwtunnel_set_redirect(&rth->dst);
1930 skb_dst_set(skb, &rth->dst);
1931out:
1932 reason = SKB_NOT_DROPPED_YET;
1933cleanup:
1934 return reason;
1935}
1936
1937#ifdef CONFIG_IP_ROUTE_MULTIPATH
1938/* To make ICMP packets follow the right flow, the multipath hash is
1939 * calculated from the inner IP addresses.
1940 */
1941static void ip_multipath_l3_keys(const struct sk_buff *skb,
1942 struct flow_keys *hash_keys)
1943{
1944 const struct iphdr *outer_iph = ip_hdr(skb);
1945 const struct iphdr *key_iph = outer_iph;
1946 const struct iphdr *inner_iph;
1947 const struct icmphdr *icmph;
1948 struct iphdr _inner_iph;
1949 struct icmphdr _icmph;
1950
1951 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1952 goto out;
1953
1954 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1955 goto out;
1956
1957 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1958 &_icmph);
1959 if (!icmph)
1960 goto out;
1961
1962 if (!icmp_is_err(icmph->type))
1963 goto out;
1964
1965 inner_iph = skb_header_pointer(skb,
1966 outer_iph->ihl * 4 + sizeof(_icmph),
1967 sizeof(_inner_iph), &_inner_iph);
1968 if (!inner_iph)
1969 goto out;
1970
1971 key_iph = inner_iph;
1972out:
1973 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1974 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1975}
1976
1977static u32 fib_multipath_custom_hash_outer(const struct net *net,
1978 const struct sk_buff *skb,
1979 bool *p_has_inner)
1980{
1981 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1982 struct flow_keys keys, hash_keys;
1983
1984 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1985 return 0;
1986
1987 memset(&hash_keys, 0, sizeof(hash_keys));
1988 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1989
1990 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1991 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1992 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1993 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1994 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1995 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1996 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1997 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1998 hash_keys.ports.src = keys.ports.src;
1999 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2000 hash_keys.ports.dst = keys.ports.dst;
2001
2002 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2003 return fib_multipath_hash_from_keys(net, &hash_keys);
2004}
2005
2006static u32 fib_multipath_custom_hash_inner(const struct net *net,
2007 const struct sk_buff *skb,
2008 bool has_inner)
2009{
2010 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2011 struct flow_keys keys, hash_keys;
2012
2013 /* We assume the packet carries an encapsulation, but if none was
2014 * encountered during dissection of the outer flow, then there is no
2015 * point in calling the flow dissector again.
2016 */
2017 if (!has_inner)
2018 return 0;
2019
2020 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2021 return 0;
2022
2023 memset(&hash_keys, 0, sizeof(hash_keys));
2024 skb_flow_dissect_flow_keys(skb, &keys, 0);
2025
2026 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2027 return 0;
2028
2029 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2030 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2031 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2032 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2033 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2034 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2035 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2036 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2037 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2038 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2039 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2040 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2041 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2042 hash_keys.tags.flow_label = keys.tags.flow_label;
2043 }
2044
2045 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2046 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2047 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2048 hash_keys.ports.src = keys.ports.src;
2049 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2050 hash_keys.ports.dst = keys.ports.dst;
2051
2052 return fib_multipath_hash_from_keys(net, &hash_keys);
2053}
2054
2055static u32 fib_multipath_custom_hash_skb(const struct net *net,
2056 const struct sk_buff *skb)
2057{
2058 u32 mhash, mhash_inner;
2059 bool has_inner = true;
2060
2061 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2062 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2063
2064 return jhash_2words(mhash, mhash_inner, 0);
2065}
2066
2067static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2068 const struct flowi4 *fl4)
2069{
2070 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2071 struct flow_keys hash_keys;
2072
2073 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2074 return 0;
2075
2076 memset(&hash_keys, 0, sizeof(hash_keys));
2077 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2078 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2079 hash_keys.addrs.v4addrs.src = fl4->saddr;
2080 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2081 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2082 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2083 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2084 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2085 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2086 hash_keys.ports.src = (__force __be16)get_random_u16();
2087 else
2088 hash_keys.ports.src = fl4->fl4_sport;
2089 }
2090 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2091 hash_keys.ports.dst = fl4->fl4_dport;
2092
2093 return fib_multipath_hash_from_keys(net, &hash_keys);
2094}
2095
2096/* if skb is set it will be used and fl4 can be NULL */
2097int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2098 const struct sk_buff *skb, struct flow_keys *flkeys)
2099{
2100 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2101 struct flow_keys hash_keys;
2102 u32 mhash = 0;
2103
2104 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2105 case 0:
2106 memset(&hash_keys, 0, sizeof(hash_keys));
2107 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2108 if (skb) {
2109 ip_multipath_l3_keys(skb, &hash_keys);
2110 } else {
2111 hash_keys.addrs.v4addrs.src = fl4->saddr;
2112 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2113 }
2114 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2115 break;
2116 case 1:
2117 /* skb is currently provided only when forwarding */
2118 if (skb) {
2119 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2120 struct flow_keys keys;
2121
2122 /* short-circuit if we already have L4 hash present */
2123 if (skb->l4_hash)
2124 return skb_get_hash_raw(skb) >> 1;
2125
2126 memset(&hash_keys, 0, sizeof(hash_keys));
2127
2128 if (!flkeys) {
2129 skb_flow_dissect_flow_keys(skb, &keys, flag);
2130 flkeys = &keys;
2131 }
2132
2133 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2134 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2135 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2136 hash_keys.ports.src = flkeys->ports.src;
2137 hash_keys.ports.dst = flkeys->ports.dst;
2138 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2139 } else {
2140 memset(&hash_keys, 0, sizeof(hash_keys));
2141 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2142 hash_keys.addrs.v4addrs.src = fl4->saddr;
2143 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2144 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT)
2145 hash_keys.ports.src = (__force __be16)get_random_u16();
2146 else
2147 hash_keys.ports.src = fl4->fl4_sport;
2148 hash_keys.ports.dst = fl4->fl4_dport;
2149 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2150 }
2151 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2152 break;
2153 case 2:
2154 memset(&hash_keys, 0, sizeof(hash_keys));
2155 /* skb is currently provided only when forwarding */
2156 if (skb) {
2157 struct flow_keys keys;
2158
2159 skb_flow_dissect_flow_keys(skb, &keys, 0);
2160 /* Inner can be v4 or v6 */
2161 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2162 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2163 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2164 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2165 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2166 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2167 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2168 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2169 hash_keys.tags.flow_label = keys.tags.flow_label;
2170 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2171 } else {
2172 /* Same as case 0 */
2173 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2174 ip_multipath_l3_keys(skb, &hash_keys);
2175 }
2176 } else {
2177 /* Same as case 0 */
2178 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2179 hash_keys.addrs.v4addrs.src = fl4->saddr;
2180 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2181 }
2182 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2183 break;
2184 case 3:
2185 if (skb)
2186 mhash = fib_multipath_custom_hash_skb(net, skb);
2187 else
2188 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2189 break;
2190 }
2191
2192 if (multipath_hash)
2193 mhash = jhash_2words(mhash, multipath_hash, 0);
2194
2195 return mhash >> 1;
2196}
2197#endif /* CONFIG_IP_ROUTE_MULTIPATH */
2198
2199static enum skb_drop_reason
2200ip_mkroute_input(struct sk_buff *skb, struct fib_result *res,
2201 struct in_device *in_dev, __be32 daddr,
2202 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys)
2203{
2204#ifdef CONFIG_IP_ROUTE_MULTIPATH
2205 if (res->fi && fib_info_num_path(res->fi) > 1) {
2206 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2207
2208 fib_select_multipath(res, h, NULL);
2209 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2210 }
2211#endif
2212
2213 /* create a routing cache entry */
2214 return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp);
2215}
2216
2217/* Implements all the saddr-related checks as ip_route_input_slow(),
2218 * assuming daddr is valid and the destination is not a local broadcast one.
2219 * Uses the provided hint instead of performing a route lookup.
2220 */
2221enum skb_drop_reason
2222ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2223 dscp_t dscp, struct net_device *dev,
2224 const struct sk_buff *hint)
2225{
2226 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2227 struct in_device *in_dev = __in_dev_get_rcu(dev);
2228 struct rtable *rt = skb_rtable(hint);
2229 struct net *net = dev_net(dev);
2230 u32 tag = 0;
2231
2232 if (!in_dev)
2233 return reason;
2234
2235 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2236 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2237 goto martian_source;
2238 }
2239
2240 if (ipv4_is_zeronet(saddr)) {
2241 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2242 goto martian_source;
2243 }
2244
2245 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2246 reason = SKB_DROP_REASON_IP_LOCALNET;
2247 goto martian_source;
2248 }
2249
2250 if (!(rt->rt_flags & RTCF_LOCAL))
2251 goto skip_validate_source;
2252
2253 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev,
2254 in_dev, &tag);
2255 if (reason)
2256 goto martian_source;
2257
2258skip_validate_source:
2259 skb_dst_copy(skb, hint);
2260 return SKB_NOT_DROPPED_YET;
2261
2262martian_source:
2263 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2264 return reason;
2265}
2266
2267/* get device for dst_alloc with local routes */
2268static struct net_device *ip_rt_get_dev(struct net *net,
2269 const struct fib_result *res)
2270{
2271 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2272 struct net_device *dev = NULL;
2273
2274 if (nhc)
2275 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2276
2277 return dev ? : net->loopback_dev;
2278}
2279
2280/*
2281 * NOTE. We drop all the packets that has local source
2282 * addresses, because every properly looped back packet
2283 * must have correct destination already attached by output routine.
2284 * Changes in the enforced policies must be applied also to
2285 * ip_route_use_hint().
2286 *
2287 * Such approach solves two big problems:
2288 * 1. Not simplex devices are handled properly.
2289 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2290 * called with rcu_read_lock()
2291 */
2292
2293static enum skb_drop_reason
2294ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2295 dscp_t dscp, struct net_device *dev,
2296 struct fib_result *res)
2297{
2298 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2299 struct in_device *in_dev = __in_dev_get_rcu(dev);
2300 struct flow_keys *flkeys = NULL, _flkeys;
2301 struct net *net = dev_net(dev);
2302 struct ip_tunnel_info *tun_info;
2303 int err = -EINVAL;
2304 unsigned int flags = 0;
2305 u32 itag = 0;
2306 struct rtable *rth;
2307 struct flowi4 fl4;
2308 bool do_cache = true;
2309
2310 /* IP on this device is disabled. */
2311
2312 if (!in_dev)
2313 goto out;
2314
2315 /* Check for the most weird martians, which can be not detected
2316 * by fib_lookup.
2317 */
2318
2319 tun_info = skb_tunnel_info(skb);
2320 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2321 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2322 else
2323 fl4.flowi4_tun_key.tun_id = 0;
2324 skb_dst_drop(skb);
2325
2326 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) {
2327 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2328 goto martian_source;
2329 }
2330
2331 res->fi = NULL;
2332 res->table = NULL;
2333 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2334 goto brd_input;
2335
2336 /* Accept zero addresses only to limited broadcast;
2337 * I even do not know to fix it or not. Waiting for complains :-)
2338 */
2339 if (ipv4_is_zeronet(saddr)) {
2340 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
2341 goto martian_source;
2342 }
2343
2344 if (ipv4_is_zeronet(daddr)) {
2345 reason = SKB_DROP_REASON_IP_INVALID_DEST;
2346 goto martian_destination;
2347 }
2348
2349 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2350 * and call it once if daddr or/and saddr are loopback addresses
2351 */
2352 if (ipv4_is_loopback(daddr)) {
2353 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2354 reason = SKB_DROP_REASON_IP_LOCALNET;
2355 goto martian_destination;
2356 }
2357 } else if (ipv4_is_loopback(saddr)) {
2358 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) {
2359 reason = SKB_DROP_REASON_IP_LOCALNET;
2360 goto martian_source;
2361 }
2362 }
2363
2364 /*
2365 * Now we are ready to route packet.
2366 */
2367 fl4.flowi4_l3mdev = 0;
2368 fl4.flowi4_oif = 0;
2369 fl4.flowi4_iif = dev->ifindex;
2370 fl4.flowi4_mark = skb->mark;
2371 fl4.flowi4_dscp = dscp;
2372 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2373 fl4.flowi4_flags = 0;
2374 fl4.daddr = daddr;
2375 fl4.saddr = saddr;
2376 fl4.flowi4_uid = sock_net_uid(net, NULL);
2377 fl4.flowi4_multipath_hash = 0;
2378
2379 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2380 flkeys = &_flkeys;
2381 } else {
2382 fl4.flowi4_proto = 0;
2383 fl4.fl4_sport = 0;
2384 fl4.fl4_dport = 0;
2385 }
2386
2387 err = fib_lookup(net, &fl4, res, 0);
2388 if (err != 0) {
2389 if (!IN_DEV_FORWARD(in_dev))
2390 err = -EHOSTUNREACH;
2391 goto no_route;
2392 }
2393
2394 if (res->type == RTN_BROADCAST) {
2395 if (IN_DEV_BFORWARD(in_dev))
2396 goto make_route;
2397 /* not do cache if bc_forwarding is enabled */
2398 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2399 do_cache = false;
2400 goto brd_input;
2401 }
2402
2403 err = -EINVAL;
2404 if (res->type == RTN_LOCAL) {
2405 reason = fib_validate_source_reason(skb, saddr, daddr, dscp,
2406 0, dev, in_dev, &itag);
2407 if (reason)
2408 goto martian_source;
2409 goto local_input;
2410 }
2411
2412 if (!IN_DEV_FORWARD(in_dev)) {
2413 err = -EHOSTUNREACH;
2414 goto no_route;
2415 }
2416 if (res->type != RTN_UNICAST) {
2417 reason = SKB_DROP_REASON_IP_INVALID_DEST;
2418 goto martian_destination;
2419 }
2420
2421make_route:
2422 reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp,
2423 flkeys);
2424
2425out:
2426 return reason;
2427
2428brd_input:
2429 if (skb->protocol != htons(ETH_P_IP)) {
2430 reason = SKB_DROP_REASON_INVALID_PROTO;
2431 goto out;
2432 }
2433
2434 if (!ipv4_is_zeronet(saddr)) {
2435 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0,
2436 dev, in_dev, &itag);
2437 if (reason)
2438 goto martian_source;
2439 }
2440 flags |= RTCF_BROADCAST;
2441 res->type = RTN_BROADCAST;
2442 RT_CACHE_STAT_INC(in_brd);
2443
2444local_input:
2445 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2446 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2447
2448 do_cache &= res->fi && !itag;
2449 if (do_cache) {
2450 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2451
2452 rth = rcu_dereference(nhc->nhc_rth_input);
2453 if (rt_cache_valid(rth)) {
2454 skb_dst_set_noref(skb, &rth->dst);
2455 reason = SKB_NOT_DROPPED_YET;
2456 goto out;
2457 }
2458 }
2459
2460 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2461 flags | RTCF_LOCAL, res->type, false);
2462 if (!rth)
2463 goto e_nobufs;
2464
2465 rth->dst.output= ip_rt_bug;
2466#ifdef CONFIG_IP_ROUTE_CLASSID
2467 rth->dst.tclassid = itag;
2468#endif
2469 rth->rt_is_input = 1;
2470
2471 RT_CACHE_STAT_INC(in_slow_tot);
2472 if (res->type == RTN_UNREACHABLE) {
2473 rth->dst.input= ip_error;
2474 rth->dst.error= -err;
2475 rth->rt_flags &= ~RTCF_LOCAL;
2476 }
2477
2478 if (do_cache) {
2479 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2480
2481 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2482 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2483 WARN_ON(rth->dst.input == lwtunnel_input);
2484 rth->dst.lwtstate->orig_input = rth->dst.input;
2485 rth->dst.input = lwtunnel_input;
2486 }
2487
2488 if (unlikely(!rt_cache_route(nhc, rth)))
2489 rt_add_uncached_list(rth);
2490 }
2491 skb_dst_set(skb, &rth->dst);
2492 reason = SKB_NOT_DROPPED_YET;
2493 goto out;
2494
2495no_route:
2496 RT_CACHE_STAT_INC(in_no_route);
2497 res->type = RTN_UNREACHABLE;
2498 res->fi = NULL;
2499 res->table = NULL;
2500 goto local_input;
2501
2502 /*
2503 * Do not cache martian addresses: they should be logged (RFC1812)
2504 */
2505martian_destination:
2506 RT_CACHE_STAT_INC(in_martian_dst);
2507#ifdef CONFIG_IP_ROUTE_VERBOSE
2508 if (IN_DEV_LOG_MARTIANS(in_dev))
2509 net_warn_ratelimited("martian destination (src=%pI4, dst=%pI4, dev=%s)\n",
2510 &saddr, &daddr, dev->name);
2511#endif
2512 goto out;
2513
2514e_nobufs:
2515 reason = SKB_DROP_REASON_NOMEM;
2516 goto out;
2517
2518martian_source:
2519 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2520 goto out;
2521}
2522
2523/* called with rcu_read_lock held */
2524static enum skb_drop_reason
2525ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2526 dscp_t dscp, struct net_device *dev,
2527 struct fib_result *res)
2528{
2529 /* Multicast recognition logic is moved from route cache to here.
2530 * The problem was that too many Ethernet cards have broken/missing
2531 * hardware multicast filters :-( As result the host on multicasting
2532 * network acquires a lot of useless route cache entries, sort of
2533 * SDR messages from all the world. Now we try to get rid of them.
2534 * Really, provided software IP multicast filter is organized
2535 * reasonably (at least, hashed), it does not result in a slowdown
2536 * comparing with route cache reject entries.
2537 * Note, that multicast routers are not affected, because
2538 * route cache entry is created eventually.
2539 */
2540 if (ipv4_is_multicast(daddr)) {
2541 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
2542 struct in_device *in_dev = __in_dev_get_rcu(dev);
2543 int our = 0;
2544
2545 if (!in_dev)
2546 return reason;
2547
2548 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2549 ip_hdr(skb)->protocol);
2550
2551 /* check l3 master if no match yet */
2552 if (!our && netif_is_l3_slave(dev)) {
2553 struct in_device *l3_in_dev;
2554
2555 l3_in_dev = __in_dev_get_rcu(skb->dev);
2556 if (l3_in_dev)
2557 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2558 ip_hdr(skb)->protocol);
2559 }
2560
2561 if (our
2562#ifdef CONFIG_IP_MROUTE
2563 ||
2564 (!ipv4_is_local_multicast(daddr) &&
2565 IN_DEV_MFORWARD(in_dev))
2566#endif
2567 ) {
2568 reason = ip_route_input_mc(skb, daddr, saddr, dscp,
2569 dev, our);
2570 }
2571 return reason;
2572 }
2573
2574 return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res);
2575}
2576
2577enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr,
2578 __be32 saddr, dscp_t dscp,
2579 struct net_device *dev)
2580{
2581 enum skb_drop_reason reason;
2582 struct fib_result res;
2583
2584 rcu_read_lock();
2585 reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res);
2586 rcu_read_unlock();
2587
2588 return reason;
2589}
2590EXPORT_SYMBOL(ip_route_input_noref);
2591
2592/* called with rcu_read_lock() */
2593static struct rtable *__mkroute_output(const struct fib_result *res,
2594 const struct flowi4 *fl4, int orig_oif,
2595 struct net_device *dev_out,
2596 unsigned int flags)
2597{
2598 struct fib_info *fi = res->fi;
2599 struct fib_nh_exception *fnhe;
2600 struct in_device *in_dev;
2601 u16 type = res->type;
2602 struct rtable *rth;
2603 bool do_cache;
2604
2605 in_dev = __in_dev_get_rcu(dev_out);
2606 if (!in_dev)
2607 return ERR_PTR(-EINVAL);
2608
2609 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2610 if (ipv4_is_loopback(fl4->saddr) &&
2611 !(dev_out->flags & IFF_LOOPBACK) &&
2612 !netif_is_l3_master(dev_out))
2613 return ERR_PTR(-EINVAL);
2614
2615 if (ipv4_is_lbcast(fl4->daddr)) {
2616 type = RTN_BROADCAST;
2617
2618 /* reset fi to prevent gateway resolution */
2619 fi = NULL;
2620 } else if (ipv4_is_multicast(fl4->daddr)) {
2621 type = RTN_MULTICAST;
2622 } else if (ipv4_is_zeronet(fl4->daddr)) {
2623 return ERR_PTR(-EINVAL);
2624 }
2625
2626 if (dev_out->flags & IFF_LOOPBACK)
2627 flags |= RTCF_LOCAL;
2628
2629 do_cache = true;
2630 if (type == RTN_BROADCAST) {
2631 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2632 } else if (type == RTN_MULTICAST) {
2633 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2634 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2635 fl4->flowi4_proto))
2636 flags &= ~RTCF_LOCAL;
2637 else
2638 do_cache = false;
2639 /* If multicast route do not exist use
2640 * default one, but do not gateway in this case.
2641 * Yes, it is hack.
2642 */
2643 if (fi && res->prefixlen < 4)
2644 fi = NULL;
2645 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2646 (orig_oif != dev_out->ifindex)) {
2647 /* For local routes that require a particular output interface
2648 * we do not want to cache the result. Caching the result
2649 * causes incorrect behaviour when there are multiple source
2650 * addresses on the interface, the end result being that if the
2651 * intended recipient is waiting on that interface for the
2652 * packet he won't receive it because it will be delivered on
2653 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2654 * be set to the loopback interface as well.
2655 */
2656 do_cache = false;
2657 }
2658
2659 fnhe = NULL;
2660 do_cache &= fi != NULL;
2661 if (fi) {
2662 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2663 struct rtable __rcu **prth;
2664
2665 fnhe = find_exception(nhc, fl4->daddr);
2666 if (!do_cache)
2667 goto add;
2668 if (fnhe) {
2669 prth = &fnhe->fnhe_rth_output;
2670 } else {
2671 if (unlikely(fl4->flowi4_flags &
2672 FLOWI_FLAG_KNOWN_NH &&
2673 !(nhc->nhc_gw_family &&
2674 nhc->nhc_scope == RT_SCOPE_LINK))) {
2675 do_cache = false;
2676 goto add;
2677 }
2678 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2679 }
2680 rth = rcu_dereference(*prth);
2681 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2682 return rth;
2683 }
2684
2685add:
2686 rth = rt_dst_alloc(dev_out, flags, type,
2687 IN_DEV_ORCONF(in_dev, NOXFRM));
2688 if (!rth)
2689 return ERR_PTR(-ENOBUFS);
2690
2691 rth->rt_iif = orig_oif;
2692
2693 RT_CACHE_STAT_INC(out_slow_tot);
2694
2695 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2696 if (flags & RTCF_LOCAL &&
2697 !(dev_out->flags & IFF_LOOPBACK)) {
2698 rth->dst.output = ip_mc_output;
2699 RT_CACHE_STAT_INC(out_slow_mc);
2700 }
2701#ifdef CONFIG_IP_MROUTE
2702 if (type == RTN_MULTICAST) {
2703 if (IN_DEV_MFORWARD(in_dev) &&
2704 !ipv4_is_local_multicast(fl4->daddr)) {
2705 rth->dst.input = ip_mr_input;
2706 rth->dst.output = ip_mr_output;
2707 }
2708 }
2709#endif
2710 }
2711
2712 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2713 lwtunnel_set_redirect(&rth->dst);
2714
2715 return rth;
2716}
2717
2718/*
2719 * Major route resolver routine.
2720 */
2721
2722struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2723 const struct sk_buff *skb)
2724{
2725 struct fib_result res = {
2726 .type = RTN_UNSPEC,
2727 .fi = NULL,
2728 .table = NULL,
2729 .tclassid = 0,
2730 };
2731 struct rtable *rth;
2732
2733 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2734
2735 rcu_read_lock();
2736 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2737 rcu_read_unlock();
2738
2739 return rth;
2740}
2741EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2742
2743struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2744 struct fib_result *res,
2745 const struct sk_buff *skb)
2746{
2747 struct net_device *dev_out = NULL;
2748 int orig_oif = fl4->flowi4_oif;
2749 unsigned int flags = 0;
2750 struct rtable *rth;
2751 int err;
2752
2753 if (fl4->saddr) {
2754 if (ipv4_is_multicast(fl4->saddr) ||
2755 ipv4_is_lbcast(fl4->saddr)) {
2756 rth = ERR_PTR(-EINVAL);
2757 goto out;
2758 }
2759
2760 rth = ERR_PTR(-ENETUNREACH);
2761
2762 /* I removed check for oif == dev_out->oif here.
2763 * It was wrong for two reasons:
2764 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2765 * is assigned to multiple interfaces.
2766 * 2. Moreover, we are allowed to send packets with saddr
2767 * of another iface. --ANK
2768 */
2769
2770 if (fl4->flowi4_oif == 0 &&
2771 (ipv4_is_multicast(fl4->daddr) ||
2772 ipv4_is_lbcast(fl4->daddr))) {
2773 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2774 dev_out = __ip_dev_find(net, fl4->saddr, false);
2775 if (!dev_out)
2776 goto out;
2777
2778 /* Special hack: user can direct multicasts
2779 * and limited broadcast via necessary interface
2780 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2781 * This hack is not just for fun, it allows
2782 * vic,vat and friends to work.
2783 * They bind socket to loopback, set ttl to zero
2784 * and expect that it will work.
2785 * From the viewpoint of routing cache they are broken,
2786 * because we are not allowed to build multicast path
2787 * with loopback source addr (look, routing cache
2788 * cannot know, that ttl is zero, so that packet
2789 * will not leave this host and route is valid).
2790 * Luckily, this hack is good workaround.
2791 */
2792
2793 fl4->flowi4_oif = dev_out->ifindex;
2794 goto make_route;
2795 }
2796
2797 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2798 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2799 if (!__ip_dev_find(net, fl4->saddr, false))
2800 goto out;
2801 }
2802 }
2803
2804
2805 if (fl4->flowi4_oif) {
2806 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2807 rth = ERR_PTR(-ENODEV);
2808 if (!dev_out)
2809 goto out;
2810
2811 /* RACE: Check return value of inet_select_addr instead. */
2812 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2813 rth = ERR_PTR(-ENETUNREACH);
2814 goto out;
2815 }
2816 if (ipv4_is_local_multicast(fl4->daddr) ||
2817 ipv4_is_lbcast(fl4->daddr) ||
2818 fl4->flowi4_proto == IPPROTO_IGMP) {
2819 if (!fl4->saddr)
2820 fl4->saddr = inet_select_addr(dev_out, 0,
2821 RT_SCOPE_LINK);
2822 goto make_route;
2823 }
2824 if (!fl4->saddr) {
2825 if (ipv4_is_multicast(fl4->daddr))
2826 fl4->saddr = inet_select_addr(dev_out, 0,
2827 fl4->flowi4_scope);
2828 else if (!fl4->daddr)
2829 fl4->saddr = inet_select_addr(dev_out, 0,
2830 RT_SCOPE_HOST);
2831 }
2832 }
2833
2834 if (!fl4->daddr) {
2835 fl4->daddr = fl4->saddr;
2836 if (!fl4->daddr)
2837 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2838 dev_out = net->loopback_dev;
2839 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2840 res->type = RTN_LOCAL;
2841 flags |= RTCF_LOCAL;
2842 goto make_route;
2843 }
2844
2845 err = fib_lookup(net, fl4, res, 0);
2846 if (err) {
2847 res->fi = NULL;
2848 res->table = NULL;
2849 if (fl4->flowi4_oif &&
2850 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2851 /* Apparently, routing tables are wrong. Assume,
2852 * that the destination is on link.
2853 *
2854 * WHY? DW.
2855 * Because we are allowed to send to iface
2856 * even if it has NO routes and NO assigned
2857 * addresses. When oif is specified, routing
2858 * tables are looked up with only one purpose:
2859 * to catch if destination is gatewayed, rather than
2860 * direct. Moreover, if MSG_DONTROUTE is set,
2861 * we send packet, ignoring both routing tables
2862 * and ifaddr state. --ANK
2863 *
2864 *
2865 * We could make it even if oif is unknown,
2866 * likely IPv6, but we do not.
2867 */
2868
2869 if (fl4->saddr == 0)
2870 fl4->saddr = inet_select_addr(dev_out, 0,
2871 RT_SCOPE_LINK);
2872 res->type = RTN_UNICAST;
2873 goto make_route;
2874 }
2875 rth = ERR_PTR(err);
2876 goto out;
2877 }
2878
2879 if (res->type == RTN_LOCAL) {
2880 if (!fl4->saddr) {
2881 if (res->fi->fib_prefsrc)
2882 fl4->saddr = res->fi->fib_prefsrc;
2883 else
2884 fl4->saddr = fl4->daddr;
2885 }
2886
2887 /* L3 master device is the loopback for that domain */
2888 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2889 net->loopback_dev;
2890
2891 /* make sure orig_oif points to fib result device even
2892 * though packet rx/tx happens over loopback or l3mdev
2893 */
2894 orig_oif = FIB_RES_OIF(*res);
2895
2896 fl4->flowi4_oif = dev_out->ifindex;
2897 flags |= RTCF_LOCAL;
2898 goto make_route;
2899 }
2900
2901 fib_select_path(net, res, fl4, skb);
2902
2903 dev_out = FIB_RES_DEV(*res);
2904
2905make_route:
2906 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2907
2908out:
2909 return rth;
2910}
2911
2912static struct dst_ops ipv4_dst_blackhole_ops = {
2913 .family = AF_INET,
2914 .default_advmss = ipv4_default_advmss,
2915 .neigh_lookup = ipv4_neigh_lookup,
2916 .check = dst_blackhole_check,
2917 .cow_metrics = dst_blackhole_cow_metrics,
2918 .update_pmtu = dst_blackhole_update_pmtu,
2919 .redirect = dst_blackhole_redirect,
2920 .mtu = dst_blackhole_mtu,
2921};
2922
2923struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2924{
2925 struct rtable *ort = dst_rtable(dst_orig);
2926 struct rtable *rt;
2927
2928 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2929 if (rt) {
2930 struct dst_entry *new = &rt->dst;
2931
2932 new->__use = 1;
2933 new->input = dst_discard;
2934 new->output = dst_discard_out;
2935
2936 new->dev = net->loopback_dev;
2937 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2938
2939 rt->rt_is_input = ort->rt_is_input;
2940 rt->rt_iif = ort->rt_iif;
2941 rt->rt_pmtu = ort->rt_pmtu;
2942 rt->rt_mtu_locked = ort->rt_mtu_locked;
2943
2944 rt->rt_genid = rt_genid_ipv4(net);
2945 rt->rt_flags = ort->rt_flags;
2946 rt->rt_type = ort->rt_type;
2947 rt->rt_uses_gateway = ort->rt_uses_gateway;
2948 rt->rt_gw_family = ort->rt_gw_family;
2949 if (rt->rt_gw_family == AF_INET)
2950 rt->rt_gw4 = ort->rt_gw4;
2951 else if (rt->rt_gw_family == AF_INET6)
2952 rt->rt_gw6 = ort->rt_gw6;
2953 }
2954
2955 dst_release(dst_orig);
2956
2957 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2958}
2959
2960struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2961 const struct sock *sk)
2962{
2963 struct rtable *rt = __ip_route_output_key(net, flp4);
2964
2965 if (IS_ERR(rt))
2966 return rt;
2967
2968 if (flp4->flowi4_proto) {
2969 flp4->flowi4_oif = rt->dst.dev->ifindex;
2970 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2971 flowi4_to_flowi(flp4),
2972 sk, 0));
2973 }
2974
2975 return rt;
2976}
2977EXPORT_SYMBOL_GPL(ip_route_output_flow);
2978
2979/* called with rcu_read_lock held */
2980static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2981 struct rtable *rt, u32 table_id, dscp_t dscp,
2982 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2983 u32 seq, unsigned int flags)
2984{
2985 struct rtmsg *r;
2986 struct nlmsghdr *nlh;
2987 unsigned long expires = 0;
2988 u32 error;
2989 u32 metrics[RTAX_MAX];
2990
2991 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2992 if (!nlh)
2993 return -EMSGSIZE;
2994
2995 r = nlmsg_data(nlh);
2996 r->rtm_family = AF_INET;
2997 r->rtm_dst_len = 32;
2998 r->rtm_src_len = 0;
2999 r->rtm_tos = inet_dscp_to_dsfield(dscp);
3000 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
3001 if (nla_put_u32(skb, RTA_TABLE, table_id))
3002 goto nla_put_failure;
3003 r->rtm_type = rt->rt_type;
3004 r->rtm_scope = RT_SCOPE_UNIVERSE;
3005 r->rtm_protocol = RTPROT_UNSPEC;
3006 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
3007 if (rt->rt_flags & RTCF_NOTIFY)
3008 r->rtm_flags |= RTM_F_NOTIFY;
3009 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
3010 r->rtm_flags |= RTCF_DOREDIRECT;
3011
3012 if (nla_put_in_addr(skb, RTA_DST, dst))
3013 goto nla_put_failure;
3014 if (src) {
3015 r->rtm_src_len = 32;
3016 if (nla_put_in_addr(skb, RTA_SRC, src))
3017 goto nla_put_failure;
3018 }
3019 if (rt->dst.dev &&
3020 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
3021 goto nla_put_failure;
3022 if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
3023 goto nla_put_failure;
3024#ifdef CONFIG_IP_ROUTE_CLASSID
3025 if (rt->dst.tclassid &&
3026 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
3027 goto nla_put_failure;
3028#endif
3029 if (fl4 && !rt_is_input_route(rt) &&
3030 fl4->saddr != src) {
3031 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
3032 goto nla_put_failure;
3033 }
3034 if (rt->rt_uses_gateway) {
3035 if (rt->rt_gw_family == AF_INET &&
3036 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3037 goto nla_put_failure;
3038 } else if (rt->rt_gw_family == AF_INET6) {
3039 int alen = sizeof(struct in6_addr);
3040 struct nlattr *nla;
3041 struct rtvia *via;
3042
3043 nla = nla_reserve(skb, RTA_VIA, alen + 2);
3044 if (!nla)
3045 goto nla_put_failure;
3046
3047 via = nla_data(nla);
3048 via->rtvia_family = AF_INET6;
3049 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3050 }
3051 }
3052
3053 expires = READ_ONCE(rt->dst.expires);
3054 if (expires) {
3055 unsigned long now = jiffies;
3056
3057 if (time_before(now, expires))
3058 expires -= now;
3059 else
3060 expires = 0;
3061 }
3062
3063 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3064 if (rt->rt_pmtu && expires)
3065 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3066 if (rt->rt_mtu_locked && expires)
3067 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3068 if (rtnetlink_put_metrics(skb, metrics) < 0)
3069 goto nla_put_failure;
3070
3071 if (fl4) {
3072 if (fl4->flowi4_mark &&
3073 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3074 goto nla_put_failure;
3075
3076 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3077 nla_put_u32(skb, RTA_UID,
3078 from_kuid_munged(current_user_ns(),
3079 fl4->flowi4_uid)))
3080 goto nla_put_failure;
3081
3082 if (rt_is_input_route(rt)) {
3083#ifdef CONFIG_IP_MROUTE
3084 if (ipv4_is_multicast(dst) &&
3085 !ipv4_is_local_multicast(dst) &&
3086 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
3087 int err = ipmr_get_route(net, skb,
3088 fl4->saddr, fl4->daddr,
3089 r, portid);
3090
3091 if (err <= 0) {
3092 if (err == 0)
3093 return 0;
3094 goto nla_put_failure;
3095 }
3096 } else
3097#endif
3098 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3099 goto nla_put_failure;
3100 }
3101 }
3102
3103 error = rt->dst.error;
3104
3105 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3106 goto nla_put_failure;
3107
3108 nlmsg_end(skb, nlh);
3109 return 0;
3110
3111nla_put_failure:
3112 nlmsg_cancel(skb, nlh);
3113 return -EMSGSIZE;
3114}
3115
3116static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3117 struct netlink_callback *cb, u32 table_id,
3118 struct fnhe_hash_bucket *bucket, int genid,
3119 int *fa_index, int fa_start, unsigned int flags)
3120{
3121 int i;
3122
3123 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3124 struct fib_nh_exception *fnhe;
3125
3126 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3127 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3128 struct rtable *rt;
3129 int err;
3130
3131 if (*fa_index < fa_start)
3132 goto next;
3133
3134 if (fnhe->fnhe_genid != genid)
3135 goto next;
3136
3137 if (fnhe->fnhe_expires &&
3138 time_after(jiffies, fnhe->fnhe_expires))
3139 goto next;
3140
3141 rt = rcu_dereference(fnhe->fnhe_rth_input);
3142 if (!rt)
3143 rt = rcu_dereference(fnhe->fnhe_rth_output);
3144 if (!rt)
3145 goto next;
3146
3147 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3148 table_id, 0, NULL, skb,
3149 NETLINK_CB(cb->skb).portid,
3150 cb->nlh->nlmsg_seq, flags);
3151 if (err)
3152 return err;
3153next:
3154 (*fa_index)++;
3155 }
3156 }
3157
3158 return 0;
3159}
3160
3161int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3162 u32 table_id, struct fib_info *fi,
3163 int *fa_index, int fa_start, unsigned int flags)
3164{
3165 struct net *net = sock_net(cb->skb->sk);
3166 int nhsel, genid = fnhe_genid(net);
3167
3168 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3169 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3170 struct fnhe_hash_bucket *bucket;
3171 int err;
3172
3173 if (!nhc)
3174 break;
3175
3176 if (nhc->nhc_flags & RTNH_F_DEAD)
3177 continue;
3178
3179 rcu_read_lock();
3180 bucket = rcu_dereference(nhc->nhc_exceptions);
3181 err = 0;
3182 if (bucket)
3183 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3184 genid, fa_index, fa_start,
3185 flags);
3186 rcu_read_unlock();
3187 if (err)
3188 return err;
3189 }
3190
3191 return 0;
3192}
3193
3194static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3195 u8 ip_proto, __be16 sport,
3196 __be16 dport)
3197{
3198 struct sk_buff *skb;
3199 struct iphdr *iph;
3200
3201 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3202 if (!skb)
3203 return NULL;
3204
3205 /* Reserve room for dummy headers, this skb can pass
3206 * through good chunk of routing engine.
3207 */
3208 skb_reset_mac_header(skb);
3209 skb_reset_network_header(skb);
3210 skb->protocol = htons(ETH_P_IP);
3211 iph = skb_put(skb, sizeof(struct iphdr));
3212 iph->protocol = ip_proto;
3213 iph->saddr = src;
3214 iph->daddr = dst;
3215 iph->version = 0x4;
3216 iph->frag_off = 0;
3217 iph->ihl = 0x5;
3218 skb_set_transport_header(skb, skb->len);
3219
3220 switch (iph->protocol) {
3221 case IPPROTO_UDP: {
3222 struct udphdr *udph;
3223
3224 udph = skb_put_zero(skb, sizeof(struct udphdr));
3225 udph->source = sport;
3226 udph->dest = dport;
3227 udp_set_len_short(udph, sizeof(struct udphdr));
3228 udph->check = 0;
3229 break;
3230 }
3231 case IPPROTO_TCP: {
3232 struct tcphdr *tcph;
3233
3234 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3235 tcph->source = sport;
3236 tcph->dest = dport;
3237 tcph->doff = sizeof(struct tcphdr) / 4;
3238 tcph->rst = 1;
3239 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3240 src, dst, 0);
3241 break;
3242 }
3243 case IPPROTO_ICMP: {
3244 struct icmphdr *icmph;
3245
3246 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3247 icmph->type = ICMP_ECHO;
3248 icmph->code = 0;
3249 }
3250 }
3251
3252 return skb;
3253}
3254
3255static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3256 const struct nlmsghdr *nlh,
3257 struct nlattr **tb,
3258 struct netlink_ext_ack *extack)
3259{
3260 struct rtmsg *rtm;
3261 int i, err;
3262
3263 rtm = nlmsg_payload(nlh, sizeof(*rtm));
3264 if (!rtm) {
3265 NL_SET_ERR_MSG(extack,
3266 "ipv4: Invalid header for route get request");
3267 return -EINVAL;
3268 }
3269
3270 if (!netlink_strict_get_check(skb))
3271 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3272 rtm_ipv4_policy, extack);
3273
3274 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3275 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3276 rtm->rtm_table || rtm->rtm_protocol ||
3277 rtm->rtm_scope || rtm->rtm_type) {
3278 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3279 return -EINVAL;
3280 }
3281
3282 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3283 RTM_F_LOOKUP_TABLE |
3284 RTM_F_FIB_MATCH)) {
3285 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3286 return -EINVAL;
3287 }
3288
3289 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3290 rtm_ipv4_policy, extack);
3291 if (err)
3292 return err;
3293
3294 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3295 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3296 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3297 return -EINVAL;
3298 }
3299
3300 for (i = 0; i <= RTA_MAX; i++) {
3301 if (!tb[i])
3302 continue;
3303
3304 switch (i) {
3305 case RTA_IIF:
3306 case RTA_OIF:
3307 case RTA_SRC:
3308 case RTA_DST:
3309 case RTA_IP_PROTO:
3310 case RTA_SPORT:
3311 case RTA_DPORT:
3312 case RTA_MARK:
3313 case RTA_UID:
3314 break;
3315 default:
3316 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3317 return -EINVAL;
3318 }
3319 }
3320
3321 return 0;
3322}
3323
3324static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3325 struct netlink_ext_ack *extack)
3326{
3327 struct net *net = sock_net(in_skb->sk);
3328 struct nlattr *tb[RTA_MAX+1];
3329 u32 table_id = RT_TABLE_MAIN;
3330 __be16 sport = 0, dport = 0;
3331 struct fib_result res = {};
3332 u8 ip_proto = IPPROTO_UDP;
3333 struct rtable *rt = NULL;
3334 struct sk_buff *skb;
3335 struct rtmsg *rtm;
3336 struct flowi4 fl4 = {};
3337 __be32 dst = 0;
3338 __be32 src = 0;
3339 dscp_t dscp;
3340 kuid_t uid;
3341 u32 iif;
3342 int err;
3343 int mark;
3344
3345 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3346 if (err < 0)
3347 return err;
3348
3349 rtm = nlmsg_data(nlh);
3350 src = nla_get_in_addr_default(tb[RTA_SRC], 0);
3351 dst = nla_get_in_addr_default(tb[RTA_DST], 0);
3352 iif = nla_get_u32_default(tb[RTA_IIF], 0);
3353 mark = nla_get_u32_default(tb[RTA_MARK], 0);
3354 dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
3355 if (tb[RTA_UID])
3356 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3357 else
3358 uid = (iif ? INVALID_UID : current_uid());
3359
3360 if (tb[RTA_IP_PROTO]) {
3361 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3362 &ip_proto, AF_INET, extack);
3363 if (err)
3364 return err;
3365 }
3366
3367 if (tb[RTA_SPORT])
3368 sport = nla_get_be16(tb[RTA_SPORT]);
3369
3370 if (tb[RTA_DPORT])
3371 dport = nla_get_be16(tb[RTA_DPORT]);
3372
3373 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3374 if (!skb)
3375 return -ENOBUFS;
3376
3377 fl4.daddr = dst;
3378 fl4.saddr = src;
3379 fl4.flowi4_dscp = dscp;
3380 fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0);
3381 fl4.flowi4_mark = mark;
3382 fl4.flowi4_uid = uid;
3383 if (sport)
3384 fl4.fl4_sport = sport;
3385 if (dport)
3386 fl4.fl4_dport = dport;
3387 fl4.flowi4_proto = ip_proto;
3388
3389 rcu_read_lock();
3390
3391 if (iif) {
3392 struct net_device *dev;
3393
3394 dev = dev_get_by_index_rcu(net, iif);
3395 if (!dev) {
3396 err = -ENODEV;
3397 goto errout_rcu;
3398 }
3399
3400 fl4.flowi4_iif = iif; /* for rt_fill_info */
3401 skb->dev = dev;
3402 skb->mark = mark;
3403 err = ip_route_input_rcu(skb, dst, src, dscp, dev,
3404 &res) ? -EINVAL : 0;
3405
3406 rt = skb_rtable(skb);
3407 if (err == 0 && rt->dst.error)
3408 err = -rt->dst.error;
3409 } else {
3410 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3411 skb->dev = net->loopback_dev;
3412 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3413 err = 0;
3414 if (IS_ERR(rt))
3415 err = PTR_ERR(rt);
3416 else
3417 skb_dst_set(skb, &rt->dst);
3418 }
3419
3420 if (err)
3421 goto errout_rcu;
3422
3423 if (rtm->rtm_flags & RTM_F_NOTIFY)
3424 rt->rt_flags |= RTCF_NOTIFY;
3425
3426 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3427 table_id = res.table ? res.table->tb_id : 0;
3428
3429 /* reset skb for netlink reply msg */
3430 skb_trim(skb, 0);
3431 skb_reset_network_header(skb);
3432 skb_reset_transport_header(skb);
3433 skb_reset_mac_header(skb);
3434
3435 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3436 struct fib_rt_info fri;
3437
3438 if (!res.fi) {
3439 err = fib_props[res.type].error;
3440 if (!err)
3441 err = -EHOSTUNREACH;
3442 goto errout_rcu;
3443 }
3444 fri.fi = res.fi;
3445 fri.tb_id = table_id;
3446 fri.dst = res.prefix;
3447 fri.dst_len = res.prefixlen;
3448 fri.dscp = res.dscp;
3449 fri.type = rt->rt_type;
3450 fri.offload = 0;
3451 fri.trap = 0;
3452 fri.offload_failed = 0;
3453 if (res.fa_head) {
3454 struct fib_alias *fa;
3455
3456 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3457 u8 slen = 32 - fri.dst_len;
3458
3459 if (fa->fa_slen == slen &&
3460 fa->tb_id == fri.tb_id &&
3461 fa->fa_dscp == fri.dscp &&
3462 fa->fa_info == res.fi &&
3463 fa->fa_type == fri.type) {
3464 fri.offload = READ_ONCE(fa->offload);
3465 fri.trap = READ_ONCE(fa->trap);
3466 fri.offload_failed =
3467 READ_ONCE(fa->offload_failed);
3468 break;
3469 }
3470 }
3471 }
3472 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3473 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3474 } else {
3475 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4,
3476 skb, NETLINK_CB(in_skb).portid,
3477 nlh->nlmsg_seq, 0);
3478 }
3479 if (err < 0)
3480 goto errout_rcu;
3481
3482 rcu_read_unlock();
3483
3484 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3485
3486errout_free:
3487 return err;
3488errout_rcu:
3489 rcu_read_unlock();
3490 kfree_skb(skb);
3491 goto errout_free;
3492}
3493
3494void ip_rt_multicast_event(struct in_device *in_dev)
3495{
3496 rt_cache_flush(dev_net(in_dev->dev));
3497}
3498
3499#ifdef CONFIG_SYSCTL
3500static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3501static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3502static int ip_rt_gc_elasticity __read_mostly = 8;
3503static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3504
3505static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write,
3506 void *buffer, size_t *lenp, loff_t *ppos)
3507{
3508 struct net *net = (struct net *)__ctl->extra1;
3509
3510 if (write) {
3511 rt_cache_flush(net);
3512 fnhe_genid_bump(net);
3513 return 0;
3514 }
3515
3516 return -EINVAL;
3517}
3518
3519static struct ctl_table ipv4_route_table[] = {
3520 {
3521 .procname = "gc_thresh",
3522 .data = &ipv4_dst_ops.gc_thresh,
3523 .maxlen = sizeof(int),
3524 .mode = 0644,
3525 .proc_handler = proc_dointvec,
3526 },
3527 {
3528 .procname = "max_size",
3529 .data = &ip_rt_max_size,
3530 .maxlen = sizeof(int),
3531 .mode = 0644,
3532 .proc_handler = proc_dointvec,
3533 },
3534 {
3535 /* Deprecated. Use gc_min_interval_ms */
3536
3537 .procname = "gc_min_interval",
3538 .data = &ip_rt_gc_min_interval,
3539 .maxlen = sizeof(int),
3540 .mode = 0644,
3541 .proc_handler = proc_dointvec_jiffies,
3542 },
3543 {
3544 .procname = "gc_min_interval_ms",
3545 .data = &ip_rt_gc_min_interval,
3546 .maxlen = sizeof(int),
3547 .mode = 0644,
3548 .proc_handler = proc_dointvec_ms_jiffies,
3549 },
3550 {
3551 .procname = "gc_timeout",
3552 .data = &ip_rt_gc_timeout,
3553 .maxlen = sizeof(int),
3554 .mode = 0644,
3555 .proc_handler = proc_dointvec_jiffies,
3556 },
3557 {
3558 .procname = "gc_interval",
3559 .data = &ip_rt_gc_interval,
3560 .maxlen = sizeof(int),
3561 .mode = 0644,
3562 .proc_handler = proc_dointvec_jiffies,
3563 },
3564 {
3565 .procname = "redirect_load",
3566 .data = &ip_rt_redirect_load,
3567 .maxlen = sizeof(int),
3568 .mode = 0644,
3569 .proc_handler = proc_dointvec,
3570 },
3571 {
3572 .procname = "redirect_number",
3573 .data = &ip_rt_redirect_number,
3574 .maxlen = sizeof(int),
3575 .mode = 0644,
3576 .proc_handler = proc_dointvec,
3577 },
3578 {
3579 .procname = "redirect_silence",
3580 .data = &ip_rt_redirect_silence,
3581 .maxlen = sizeof(int),
3582 .mode = 0644,
3583 .proc_handler = proc_dointvec,
3584 },
3585 {
3586 .procname = "error_cost",
3587 .data = &ip_rt_error_cost,
3588 .maxlen = sizeof(int),
3589 .mode = 0644,
3590 .proc_handler = proc_dointvec,
3591 },
3592 {
3593 .procname = "error_burst",
3594 .data = &ip_rt_error_burst,
3595 .maxlen = sizeof(int),
3596 .mode = 0644,
3597 .proc_handler = proc_dointvec,
3598 },
3599 {
3600 .procname = "gc_elasticity",
3601 .data = &ip_rt_gc_elasticity,
3602 .maxlen = sizeof(int),
3603 .mode = 0644,
3604 .proc_handler = proc_dointvec,
3605 },
3606};
3607
3608static const char ipv4_route_flush_procname[] = "flush";
3609
3610static struct ctl_table ipv4_route_netns_table[] = {
3611 {
3612 .procname = ipv4_route_flush_procname,
3613 .maxlen = sizeof(int),
3614 .mode = 0200,
3615 .proc_handler = ipv4_sysctl_rtcache_flush,
3616 },
3617 {
3618 .procname = "min_pmtu",
3619 .data = &init_net.ipv4.ip_rt_min_pmtu,
3620 .maxlen = sizeof(int),
3621 .mode = 0644,
3622 .proc_handler = proc_dointvec_minmax,
3623 .extra1 = &ip_min_valid_pmtu,
3624 },
3625 {
3626 .procname = "mtu_expires",
3627 .data = &init_net.ipv4.ip_rt_mtu_expires,
3628 .maxlen = sizeof(int),
3629 .mode = 0644,
3630 .proc_handler = proc_dointvec_jiffies,
3631 },
3632 {
3633 .procname = "min_adv_mss",
3634 .data = &init_net.ipv4.ip_rt_min_advmss,
3635 .maxlen = sizeof(int),
3636 .mode = 0644,
3637 .proc_handler = proc_dointvec,
3638 },
3639};
3640
3641static __net_init int sysctl_route_net_init(struct net *net)
3642{
3643 struct ctl_table *tbl;
3644 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3645
3646 tbl = ipv4_route_netns_table;
3647 if (!net_eq(net, &init_net)) {
3648 int i;
3649
3650 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3651 if (!tbl)
3652 goto err_dup;
3653
3654 /* Don't export non-whitelisted sysctls to unprivileged users */
3655 if (net->user_ns != &init_user_ns) {
3656 if (tbl[0].procname != ipv4_route_flush_procname)
3657 table_size = 0;
3658 }
3659
3660 /* Update the variables to point into the current struct net
3661 * except for the first element flush
3662 */
3663 for (i = 1; i < table_size; i++)
3664 tbl[i].data += (void *)net - (void *)&init_net;
3665 }
3666 tbl[0].extra1 = net;
3667
3668 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3669 tbl, table_size);
3670 if (!net->ipv4.route_hdr)
3671 goto err_reg;
3672 return 0;
3673
3674err_reg:
3675 if (tbl != ipv4_route_netns_table)
3676 kfree(tbl);
3677err_dup:
3678 return -ENOMEM;
3679}
3680
3681static __net_exit void sysctl_route_net_exit(struct net *net)
3682{
3683 const struct ctl_table *tbl;
3684
3685 tbl = net->ipv4.route_hdr->ctl_table_arg;
3686 unregister_net_sysctl_table(net->ipv4.route_hdr);
3687 BUG_ON(tbl == ipv4_route_netns_table);
3688 kfree(tbl);
3689}
3690
3691static __net_initdata struct pernet_operations sysctl_route_ops = {
3692 .init = sysctl_route_net_init,
3693 .exit = sysctl_route_net_exit,
3694};
3695#endif
3696
3697static __net_init int netns_ip_rt_init(struct net *net)
3698{
3699 /* Set default value for namespaceified sysctls */
3700 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3701 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3702 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3703 return 0;
3704}
3705
3706static struct pernet_operations __net_initdata ip_rt_ops = {
3707 .init = netns_ip_rt_init,
3708};
3709
3710static __net_init int rt_genid_init(struct net *net)
3711{
3712 atomic_set(&net->ipv4.rt_genid, 0);
3713 atomic_set(&net->fnhe_genid, 0);
3714 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3715 return 0;
3716}
3717
3718static __net_initdata struct pernet_operations rt_genid_ops = {
3719 .init = rt_genid_init,
3720};
3721
3722static int __net_init ipv4_inetpeer_init(struct net *net)
3723{
3724 struct inet_peer_base *bp = kmalloc_obj(*bp);
3725
3726 if (!bp)
3727 return -ENOMEM;
3728 inet_peer_base_init(bp);
3729 net->ipv4.peers = bp;
3730 return 0;
3731}
3732
3733static void __net_exit ipv4_inetpeer_exit(struct net *net)
3734{
3735 struct inet_peer_base *bp = net->ipv4.peers;
3736
3737 net->ipv4.peers = NULL;
3738 inetpeer_invalidate_tree(bp);
3739 kfree(bp);
3740}
3741
3742static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3743 .init = ipv4_inetpeer_init,
3744 .exit = ipv4_inetpeer_exit,
3745};
3746
3747#ifdef CONFIG_IP_ROUTE_CLASSID
3748struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3749#endif /* CONFIG_IP_ROUTE_CLASSID */
3750
3751static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = {
3752 {.protocol = PF_INET, .msgtype = RTM_GETROUTE,
3753 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
3754};
3755
3756int __init ip_rt_init(void)
3757{
3758 void *idents_hash;
3759 int cpu;
3760
3761 /* For modern hosts, this will use 2 MB of memory */
3762 idents_hash = alloc_large_system_hash("IP idents",
3763 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3764 0,
3765 16, /* one bucket per 64 KB */
3766 HASH_ZERO,
3767 NULL,
3768 &ip_idents_mask,
3769 2048,
3770 256*1024);
3771
3772 ip_idents = idents_hash;
3773
3774 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3775
3776 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3777
3778 for_each_possible_cpu(cpu) {
3779 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3780
3781 INIT_LIST_HEAD(&ul->head);
3782 spin_lock_init(&ul->lock);
3783 }
3784#ifdef CONFIG_IP_ROUTE_CLASSID
3785 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3786 if (!ip_rt_acct)
3787 panic("IP: failed to allocate ip_rt_acct\n");
3788#endif
3789
3790 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3791 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3792
3793 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3794
3795 if (dst_entries_init(&ipv4_dst_ops) < 0)
3796 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3797
3798 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3799 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3800
3801 ipv4_dst_ops.gc_thresh = ~0;
3802 ip_rt_max_size = INT_MAX;
3803
3804 devinet_init();
3805 ip_fib_init();
3806
3807 if (ip_rt_proc_init())
3808 pr_err("Unable to create route proc files\n");
3809#ifdef CONFIG_XFRM
3810 xfrm_init();
3811 xfrm4_init();
3812#endif
3813 rtnl_register_many(ip_rt_rtnl_msg_handlers);
3814
3815#ifdef CONFIG_SYSCTL
3816 register_pernet_subsys(&sysctl_route_ops);
3817#endif
3818 register_pernet_subsys(&ip_rt_ops);
3819 register_pernet_subsys(&rt_genid_ops);
3820 register_pernet_subsys(&ipv4_inetpeer_ops);
3821 return 0;
3822}
3823
3824#ifdef CONFIG_SYSCTL
3825/*
3826 * We really need to sanitize the damn ipv4 init order, then all
3827 * this nonsense will go away.
3828 */
3829void __init ip_static_sysctl_init(void)
3830{
3831 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3832}
3833#endif