Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * ROUTE - implementation of the IP router.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13 *
14 * Fixes:
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
24 * clamper.
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
39 *
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
58 *
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
63 */
64
65#define pr_fmt(fmt) "IPv4: " fmt
66
67#include <linux/module.h>
68#include <asm/uaccess.h>
69#include <linux/bitops.h>
70#include <linux/types.h>
71#include <linux/kernel.h>
72#include <linux/mm.h>
73#include <linux/bootmem.h>
74#include <linux/string.h>
75#include <linux/socket.h>
76#include <linux/sockios.h>
77#include <linux/errno.h>
78#include <linux/in.h>
79#include <linux/inet.h>
80#include <linux/netdevice.h>
81#include <linux/proc_fs.h>
82#include <linux/init.h>
83#include <linux/workqueue.h>
84#include <linux/skbuff.h>
85#include <linux/inetdevice.h>
86#include <linux/igmp.h>
87#include <linux/pkt_sched.h>
88#include <linux/mroute.h>
89#include <linux/netfilter_ipv4.h>
90#include <linux/random.h>
91#include <linux/jhash.h>
92#include <linux/rcupdate.h>
93#include <linux/times.h>
94#include <linux/slab.h>
95#include <linux/prefetch.h>
96#include <net/dst.h>
97#include <net/net_namespace.h>
98#include <net/protocol.h>
99#include <net/ip.h>
100#include <net/route.h>
101#include <net/inetpeer.h>
102#include <net/sock.h>
103#include <net/ip_fib.h>
104#include <net/arp.h>
105#include <net/tcp.h>
106#include <net/icmp.h>
107#include <net/xfrm.h>
108#include <net/netevent.h>
109#include <net/rtnetlink.h>
110#ifdef CONFIG_SYSCTL
111#include <linux/sysctl.h>
112#include <linux/kmemleak.h>
113#endif
114#include <net/secure_seq.h>
115
116#define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
118
119#define IP_MAX_MTU 0xFFF0
120
121#define RT_GC_TIMEOUT (300*HZ)
122
123static int ip_rt_max_size;
124static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125static int ip_rt_gc_interval __read_mostly = 60 * HZ;
126static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
127static int ip_rt_redirect_number __read_mostly = 9;
128static int ip_rt_redirect_load __read_mostly = HZ / 50;
129static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
130static int ip_rt_error_cost __read_mostly = HZ;
131static int ip_rt_error_burst __read_mostly = 5 * HZ;
132static int ip_rt_gc_elasticity __read_mostly = 8;
133static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
134static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
135static int ip_rt_min_advmss __read_mostly = 256;
136static int rt_chain_length_max __read_mostly = 20;
137
138static struct delayed_work expires_work;
139static unsigned long expires_ljiffies;
140
141/*
142 * Interface to generic destination cache.
143 */
144
145static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
146static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
147static unsigned int ipv4_mtu(const struct dst_entry *dst);
148static void ipv4_dst_destroy(struct dst_entry *dst);
149static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
150static void ipv4_link_failure(struct sk_buff *skb);
151static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
152static int rt_garbage_collect(struct dst_ops *ops);
153
154static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
155 int how)
156{
157}
158
159static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
160{
161 struct rtable *rt = (struct rtable *) dst;
162 struct inet_peer *peer;
163 u32 *p = NULL;
164
165 if (!rt->peer)
166 rt_bind_peer(rt, rt->rt_dst, 1);
167
168 peer = rt->peer;
169 if (peer) {
170 u32 *old_p = __DST_METRICS_PTR(old);
171 unsigned long prev, new;
172
173 p = peer->metrics;
174 if (inet_metrics_new(peer))
175 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
176
177 new = (unsigned long) p;
178 prev = cmpxchg(&dst->_metrics, old, new);
179
180 if (prev != old) {
181 p = __DST_METRICS_PTR(prev);
182 if (prev & DST_METRICS_READ_ONLY)
183 p = NULL;
184 } else {
185 if (rt->fi) {
186 fib_info_put(rt->fi);
187 rt->fi = NULL;
188 }
189 }
190 }
191 return p;
192}
193
194static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr);
195
196static struct dst_ops ipv4_dst_ops = {
197 .family = AF_INET,
198 .protocol = cpu_to_be16(ETH_P_IP),
199 .gc = rt_garbage_collect,
200 .check = ipv4_dst_check,
201 .default_advmss = ipv4_default_advmss,
202 .mtu = ipv4_mtu,
203 .cow_metrics = ipv4_cow_metrics,
204 .destroy = ipv4_dst_destroy,
205 .ifdown = ipv4_dst_ifdown,
206 .negative_advice = ipv4_negative_advice,
207 .link_failure = ipv4_link_failure,
208 .update_pmtu = ip_rt_update_pmtu,
209 .local_out = __ip_local_out,
210 .neigh_lookup = ipv4_neigh_lookup,
211};
212
213#define ECN_OR_COST(class) TC_PRIO_##class
214
215const __u8 ip_tos2prio[16] = {
216 TC_PRIO_BESTEFFORT,
217 ECN_OR_COST(BESTEFFORT),
218 TC_PRIO_BESTEFFORT,
219 ECN_OR_COST(BESTEFFORT),
220 TC_PRIO_BULK,
221 ECN_OR_COST(BULK),
222 TC_PRIO_BULK,
223 ECN_OR_COST(BULK),
224 TC_PRIO_INTERACTIVE,
225 ECN_OR_COST(INTERACTIVE),
226 TC_PRIO_INTERACTIVE,
227 ECN_OR_COST(INTERACTIVE),
228 TC_PRIO_INTERACTIVE_BULK,
229 ECN_OR_COST(INTERACTIVE_BULK),
230 TC_PRIO_INTERACTIVE_BULK,
231 ECN_OR_COST(INTERACTIVE_BULK)
232};
233EXPORT_SYMBOL(ip_tos2prio);
234
235/*
236 * Route cache.
237 */
238
239/* The locking scheme is rather straight forward:
240 *
241 * 1) Read-Copy Update protects the buckets of the central route hash.
242 * 2) Only writers remove entries, and they hold the lock
243 * as they look at rtable reference counts.
244 * 3) Only readers acquire references to rtable entries,
245 * they do so with atomic increments and with the
246 * lock held.
247 */
248
249struct rt_hash_bucket {
250 struct rtable __rcu *chain;
251};
252
253#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
254 defined(CONFIG_PROVE_LOCKING)
255/*
256 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
257 * The size of this table is a power of two and depends on the number of CPUS.
258 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
259 */
260#ifdef CONFIG_LOCKDEP
261# define RT_HASH_LOCK_SZ 256
262#else
263# if NR_CPUS >= 32
264# define RT_HASH_LOCK_SZ 4096
265# elif NR_CPUS >= 16
266# define RT_HASH_LOCK_SZ 2048
267# elif NR_CPUS >= 8
268# define RT_HASH_LOCK_SZ 1024
269# elif NR_CPUS >= 4
270# define RT_HASH_LOCK_SZ 512
271# else
272# define RT_HASH_LOCK_SZ 256
273# endif
274#endif
275
276static spinlock_t *rt_hash_locks;
277# define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
278
279static __init void rt_hash_lock_init(void)
280{
281 int i;
282
283 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
284 GFP_KERNEL);
285 if (!rt_hash_locks)
286 panic("IP: failed to allocate rt_hash_locks\n");
287
288 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
289 spin_lock_init(&rt_hash_locks[i]);
290}
291#else
292# define rt_hash_lock_addr(slot) NULL
293
294static inline void rt_hash_lock_init(void)
295{
296}
297#endif
298
299static struct rt_hash_bucket *rt_hash_table __read_mostly;
300static unsigned int rt_hash_mask __read_mostly;
301static unsigned int rt_hash_log __read_mostly;
302
303static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
304#define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
305
306static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
307 int genid)
308{
309 return jhash_3words((__force u32)daddr, (__force u32)saddr,
310 idx, genid)
311 & rt_hash_mask;
312}
313
314static inline int rt_genid(struct net *net)
315{
316 return atomic_read(&net->ipv4.rt_genid);
317}
318
319#ifdef CONFIG_PROC_FS
320struct rt_cache_iter_state {
321 struct seq_net_private p;
322 int bucket;
323 int genid;
324};
325
326static struct rtable *rt_cache_get_first(struct seq_file *seq)
327{
328 struct rt_cache_iter_state *st = seq->private;
329 struct rtable *r = NULL;
330
331 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
332 if (!rcu_access_pointer(rt_hash_table[st->bucket].chain))
333 continue;
334 rcu_read_lock_bh();
335 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
336 while (r) {
337 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
338 r->rt_genid == st->genid)
339 return r;
340 r = rcu_dereference_bh(r->dst.rt_next);
341 }
342 rcu_read_unlock_bh();
343 }
344 return r;
345}
346
347static struct rtable *__rt_cache_get_next(struct seq_file *seq,
348 struct rtable *r)
349{
350 struct rt_cache_iter_state *st = seq->private;
351
352 r = rcu_dereference_bh(r->dst.rt_next);
353 while (!r) {
354 rcu_read_unlock_bh();
355 do {
356 if (--st->bucket < 0)
357 return NULL;
358 } while (!rcu_access_pointer(rt_hash_table[st->bucket].chain));
359 rcu_read_lock_bh();
360 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
361 }
362 return r;
363}
364
365static struct rtable *rt_cache_get_next(struct seq_file *seq,
366 struct rtable *r)
367{
368 struct rt_cache_iter_state *st = seq->private;
369 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
370 if (dev_net(r->dst.dev) != seq_file_net(seq))
371 continue;
372 if (r->rt_genid == st->genid)
373 break;
374 }
375 return r;
376}
377
378static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
379{
380 struct rtable *r = rt_cache_get_first(seq);
381
382 if (r)
383 while (pos && (r = rt_cache_get_next(seq, r)))
384 --pos;
385 return pos ? NULL : r;
386}
387
388static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
389{
390 struct rt_cache_iter_state *st = seq->private;
391 if (*pos)
392 return rt_cache_get_idx(seq, *pos - 1);
393 st->genid = rt_genid(seq_file_net(seq));
394 return SEQ_START_TOKEN;
395}
396
397static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
398{
399 struct rtable *r;
400
401 if (v == SEQ_START_TOKEN)
402 r = rt_cache_get_first(seq);
403 else
404 r = rt_cache_get_next(seq, v);
405 ++*pos;
406 return r;
407}
408
409static void rt_cache_seq_stop(struct seq_file *seq, void *v)
410{
411 if (v && v != SEQ_START_TOKEN)
412 rcu_read_unlock_bh();
413}
414
415static int rt_cache_seq_show(struct seq_file *seq, void *v)
416{
417 if (v == SEQ_START_TOKEN)
418 seq_printf(seq, "%-127s\n",
419 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
420 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
421 "HHUptod\tSpecDst");
422 else {
423 struct rtable *r = v;
424 struct neighbour *n;
425 int len, HHUptod;
426
427 rcu_read_lock();
428 n = dst_get_neighbour_noref(&r->dst);
429 HHUptod = (n && (n->nud_state & NUD_CONNECTED)) ? 1 : 0;
430 rcu_read_unlock();
431
432 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
433 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
434 r->dst.dev ? r->dst.dev->name : "*",
435 (__force u32)r->rt_dst,
436 (__force u32)r->rt_gateway,
437 r->rt_flags, atomic_read(&r->dst.__refcnt),
438 r->dst.__use, 0, (__force u32)r->rt_src,
439 dst_metric_advmss(&r->dst) + 40,
440 dst_metric(&r->dst, RTAX_WINDOW),
441 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
442 dst_metric(&r->dst, RTAX_RTTVAR)),
443 r->rt_key_tos,
444 -1,
445 HHUptod,
446 r->rt_spec_dst, &len);
447
448 seq_printf(seq, "%*s\n", 127 - len, "");
449 }
450 return 0;
451}
452
453static const struct seq_operations rt_cache_seq_ops = {
454 .start = rt_cache_seq_start,
455 .next = rt_cache_seq_next,
456 .stop = rt_cache_seq_stop,
457 .show = rt_cache_seq_show,
458};
459
460static int rt_cache_seq_open(struct inode *inode, struct file *file)
461{
462 return seq_open_net(inode, file, &rt_cache_seq_ops,
463 sizeof(struct rt_cache_iter_state));
464}
465
466static const struct file_operations rt_cache_seq_fops = {
467 .owner = THIS_MODULE,
468 .open = rt_cache_seq_open,
469 .read = seq_read,
470 .llseek = seq_lseek,
471 .release = seq_release_net,
472};
473
474
475static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
476{
477 int cpu;
478
479 if (*pos == 0)
480 return SEQ_START_TOKEN;
481
482 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
483 if (!cpu_possible(cpu))
484 continue;
485 *pos = cpu+1;
486 return &per_cpu(rt_cache_stat, cpu);
487 }
488 return NULL;
489}
490
491static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
492{
493 int cpu;
494
495 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
496 if (!cpu_possible(cpu))
497 continue;
498 *pos = cpu+1;
499 return &per_cpu(rt_cache_stat, cpu);
500 }
501 return NULL;
502
503}
504
505static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
506{
507
508}
509
510static int rt_cpu_seq_show(struct seq_file *seq, void *v)
511{
512 struct rt_cache_stat *st = v;
513
514 if (v == SEQ_START_TOKEN) {
515 seq_printf(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");
516 return 0;
517 }
518
519 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
520 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
521 dst_entries_get_slow(&ipv4_dst_ops),
522 st->in_hit,
523 st->in_slow_tot,
524 st->in_slow_mc,
525 st->in_no_route,
526 st->in_brd,
527 st->in_martian_dst,
528 st->in_martian_src,
529
530 st->out_hit,
531 st->out_slow_tot,
532 st->out_slow_mc,
533
534 st->gc_total,
535 st->gc_ignored,
536 st->gc_goal_miss,
537 st->gc_dst_overflow,
538 st->in_hlist_search,
539 st->out_hlist_search
540 );
541 return 0;
542}
543
544static const struct seq_operations rt_cpu_seq_ops = {
545 .start = rt_cpu_seq_start,
546 .next = rt_cpu_seq_next,
547 .stop = rt_cpu_seq_stop,
548 .show = rt_cpu_seq_show,
549};
550
551
552static int rt_cpu_seq_open(struct inode *inode, struct file *file)
553{
554 return seq_open(file, &rt_cpu_seq_ops);
555}
556
557static const struct file_operations rt_cpu_seq_fops = {
558 .owner = THIS_MODULE,
559 .open = rt_cpu_seq_open,
560 .read = seq_read,
561 .llseek = seq_lseek,
562 .release = seq_release,
563};
564
565#ifdef CONFIG_IP_ROUTE_CLASSID
566static int rt_acct_proc_show(struct seq_file *m, void *v)
567{
568 struct ip_rt_acct *dst, *src;
569 unsigned int i, j;
570
571 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
572 if (!dst)
573 return -ENOMEM;
574
575 for_each_possible_cpu(i) {
576 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
577 for (j = 0; j < 256; j++) {
578 dst[j].o_bytes += src[j].o_bytes;
579 dst[j].o_packets += src[j].o_packets;
580 dst[j].i_bytes += src[j].i_bytes;
581 dst[j].i_packets += src[j].i_packets;
582 }
583 }
584
585 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
586 kfree(dst);
587 return 0;
588}
589
590static int rt_acct_proc_open(struct inode *inode, struct file *file)
591{
592 return single_open(file, rt_acct_proc_show, NULL);
593}
594
595static const struct file_operations rt_acct_proc_fops = {
596 .owner = THIS_MODULE,
597 .open = rt_acct_proc_open,
598 .read = seq_read,
599 .llseek = seq_lseek,
600 .release = single_release,
601};
602#endif
603
604static int __net_init ip_rt_do_proc_init(struct net *net)
605{
606 struct proc_dir_entry *pde;
607
608 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
609 &rt_cache_seq_fops);
610 if (!pde)
611 goto err1;
612
613 pde = proc_create("rt_cache", S_IRUGO,
614 net->proc_net_stat, &rt_cpu_seq_fops);
615 if (!pde)
616 goto err2;
617
618#ifdef CONFIG_IP_ROUTE_CLASSID
619 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
620 if (!pde)
621 goto err3;
622#endif
623 return 0;
624
625#ifdef CONFIG_IP_ROUTE_CLASSID
626err3:
627 remove_proc_entry("rt_cache", net->proc_net_stat);
628#endif
629err2:
630 remove_proc_entry("rt_cache", net->proc_net);
631err1:
632 return -ENOMEM;
633}
634
635static void __net_exit ip_rt_do_proc_exit(struct net *net)
636{
637 remove_proc_entry("rt_cache", net->proc_net_stat);
638 remove_proc_entry("rt_cache", net->proc_net);
639#ifdef CONFIG_IP_ROUTE_CLASSID
640 remove_proc_entry("rt_acct", net->proc_net);
641#endif
642}
643
644static struct pernet_operations ip_rt_proc_ops __net_initdata = {
645 .init = ip_rt_do_proc_init,
646 .exit = ip_rt_do_proc_exit,
647};
648
649static int __init ip_rt_proc_init(void)
650{
651 return register_pernet_subsys(&ip_rt_proc_ops);
652}
653
654#else
655static inline int ip_rt_proc_init(void)
656{
657 return 0;
658}
659#endif /* CONFIG_PROC_FS */
660
661static inline void rt_free(struct rtable *rt)
662{
663 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
664}
665
666static inline void rt_drop(struct rtable *rt)
667{
668 ip_rt_put(rt);
669 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
670}
671
672static inline int rt_fast_clean(struct rtable *rth)
673{
674 /* Kill broadcast/multicast entries very aggresively, if they
675 collide in hash table with more useful entries */
676 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
677 rt_is_input_route(rth) && rth->dst.rt_next;
678}
679
680static inline int rt_valuable(struct rtable *rth)
681{
682 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
683 (rth->peer && rth->peer->pmtu_expires);
684}
685
686static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
687{
688 unsigned long age;
689 int ret = 0;
690
691 if (atomic_read(&rth->dst.__refcnt))
692 goto out;
693
694 age = jiffies - rth->dst.lastuse;
695 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
696 (age <= tmo2 && rt_valuable(rth)))
697 goto out;
698 ret = 1;
699out: return ret;
700}
701
702/* Bits of score are:
703 * 31: very valuable
704 * 30: not quite useless
705 * 29..0: usage counter
706 */
707static inline u32 rt_score(struct rtable *rt)
708{
709 u32 score = jiffies - rt->dst.lastuse;
710
711 score = ~score & ~(3<<30);
712
713 if (rt_valuable(rt))
714 score |= (1<<31);
715
716 if (rt_is_output_route(rt) ||
717 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
718 score |= (1<<30);
719
720 return score;
721}
722
723static inline bool rt_caching(const struct net *net)
724{
725 return net->ipv4.current_rt_cache_rebuild_count <=
726 net->ipv4.sysctl_rt_cache_rebuild_count;
727}
728
729static inline bool compare_hash_inputs(const struct rtable *rt1,
730 const struct rtable *rt2)
731{
732 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
733 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
734 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
735}
736
737static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
738{
739 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
740 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
741 (rt1->rt_mark ^ rt2->rt_mark) |
742 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
743 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
744 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
745}
746
747static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
748{
749 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
750}
751
752static inline int rt_is_expired(struct rtable *rth)
753{
754 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
755}
756
757/*
758 * Perform a full scan of hash table and free all entries.
759 * Can be called by a softirq or a process.
760 * In the later case, we want to be reschedule if necessary
761 */
762static void rt_do_flush(struct net *net, int process_context)
763{
764 unsigned int i;
765 struct rtable *rth, *next;
766
767 for (i = 0; i <= rt_hash_mask; i++) {
768 struct rtable __rcu **pprev;
769 struct rtable *list;
770
771 if (process_context && need_resched())
772 cond_resched();
773 rth = rcu_access_pointer(rt_hash_table[i].chain);
774 if (!rth)
775 continue;
776
777 spin_lock_bh(rt_hash_lock_addr(i));
778
779 list = NULL;
780 pprev = &rt_hash_table[i].chain;
781 rth = rcu_dereference_protected(*pprev,
782 lockdep_is_held(rt_hash_lock_addr(i)));
783
784 while (rth) {
785 next = rcu_dereference_protected(rth->dst.rt_next,
786 lockdep_is_held(rt_hash_lock_addr(i)));
787
788 if (!net ||
789 net_eq(dev_net(rth->dst.dev), net)) {
790 rcu_assign_pointer(*pprev, next);
791 rcu_assign_pointer(rth->dst.rt_next, list);
792 list = rth;
793 } else {
794 pprev = &rth->dst.rt_next;
795 }
796 rth = next;
797 }
798
799 spin_unlock_bh(rt_hash_lock_addr(i));
800
801 for (; list; list = next) {
802 next = rcu_dereference_protected(list->dst.rt_next, 1);
803 rt_free(list);
804 }
805 }
806}
807
808/*
809 * While freeing expired entries, we compute average chain length
810 * and standard deviation, using fixed-point arithmetic.
811 * This to have an estimation of rt_chain_length_max
812 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
813 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
814 */
815
816#define FRACT_BITS 3
817#define ONE (1UL << FRACT_BITS)
818
819/*
820 * Given a hash chain and an item in this hash chain,
821 * find if a previous entry has the same hash_inputs
822 * (but differs on tos, mark or oif)
823 * Returns 0 if an alias is found.
824 * Returns ONE if rth has no alias before itself.
825 */
826static int has_noalias(const struct rtable *head, const struct rtable *rth)
827{
828 const struct rtable *aux = head;
829
830 while (aux != rth) {
831 if (compare_hash_inputs(aux, rth))
832 return 0;
833 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
834 }
835 return ONE;
836}
837
838static void rt_check_expire(void)
839{
840 static unsigned int rover;
841 unsigned int i = rover, goal;
842 struct rtable *rth;
843 struct rtable __rcu **rthp;
844 unsigned long samples = 0;
845 unsigned long sum = 0, sum2 = 0;
846 unsigned long delta;
847 u64 mult;
848
849 delta = jiffies - expires_ljiffies;
850 expires_ljiffies = jiffies;
851 mult = ((u64)delta) << rt_hash_log;
852 if (ip_rt_gc_timeout > 1)
853 do_div(mult, ip_rt_gc_timeout);
854 goal = (unsigned int)mult;
855 if (goal > rt_hash_mask)
856 goal = rt_hash_mask + 1;
857 for (; goal > 0; goal--) {
858 unsigned long tmo = ip_rt_gc_timeout;
859 unsigned long length;
860
861 i = (i + 1) & rt_hash_mask;
862 rthp = &rt_hash_table[i].chain;
863
864 if (need_resched())
865 cond_resched();
866
867 samples++;
868
869 if (rcu_dereference_raw(*rthp) == NULL)
870 continue;
871 length = 0;
872 spin_lock_bh(rt_hash_lock_addr(i));
873 while ((rth = rcu_dereference_protected(*rthp,
874 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
875 prefetch(rth->dst.rt_next);
876 if (rt_is_expired(rth)) {
877 *rthp = rth->dst.rt_next;
878 rt_free(rth);
879 continue;
880 }
881 if (rth->dst.expires) {
882 /* Entry is expired even if it is in use */
883 if (time_before_eq(jiffies, rth->dst.expires)) {
884nofree:
885 tmo >>= 1;
886 rthp = &rth->dst.rt_next;
887 /*
888 * We only count entries on
889 * a chain with equal hash inputs once
890 * so that entries for different QOS
891 * levels, and other non-hash input
892 * attributes don't unfairly skew
893 * the length computation
894 */
895 length += has_noalias(rt_hash_table[i].chain, rth);
896 continue;
897 }
898 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
899 goto nofree;
900
901 /* Cleanup aged off entries. */
902 *rthp = rth->dst.rt_next;
903 rt_free(rth);
904 }
905 spin_unlock_bh(rt_hash_lock_addr(i));
906 sum += length;
907 sum2 += length*length;
908 }
909 if (samples) {
910 unsigned long avg = sum / samples;
911 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
912 rt_chain_length_max = max_t(unsigned long,
913 ip_rt_gc_elasticity,
914 (avg + 4*sd) >> FRACT_BITS);
915 }
916 rover = i;
917}
918
919/*
920 * rt_worker_func() is run in process context.
921 * we call rt_check_expire() to scan part of the hash table
922 */
923static void rt_worker_func(struct work_struct *work)
924{
925 rt_check_expire();
926 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
927}
928
929/*
930 * Perturbation of rt_genid by a small quantity [1..256]
931 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
932 * many times (2^24) without giving recent rt_genid.
933 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
934 */
935static void rt_cache_invalidate(struct net *net)
936{
937 unsigned char shuffle;
938
939 get_random_bytes(&shuffle, sizeof(shuffle));
940 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
941 inetpeer_invalidate_tree(AF_INET);
942}
943
944/*
945 * delay < 0 : invalidate cache (fast : entries will be deleted later)
946 * delay >= 0 : invalidate & flush cache (can be long)
947 */
948void rt_cache_flush(struct net *net, int delay)
949{
950 rt_cache_invalidate(net);
951 if (delay >= 0)
952 rt_do_flush(net, !in_softirq());
953}
954
955/* Flush previous cache invalidated entries from the cache */
956void rt_cache_flush_batch(struct net *net)
957{
958 rt_do_flush(net, !in_softirq());
959}
960
961static void rt_emergency_hash_rebuild(struct net *net)
962{
963 net_warn_ratelimited("Route hash chain too long!\n");
964 rt_cache_invalidate(net);
965}
966
967/*
968 Short description of GC goals.
969
970 We want to build algorithm, which will keep routing cache
971 at some equilibrium point, when number of aged off entries
972 is kept approximately equal to newly generated ones.
973
974 Current expiration strength is variable "expire".
975 We try to adjust it dynamically, so that if networking
976 is idle expires is large enough to keep enough of warm entries,
977 and when load increases it reduces to limit cache size.
978 */
979
980static int rt_garbage_collect(struct dst_ops *ops)
981{
982 static unsigned long expire = RT_GC_TIMEOUT;
983 static unsigned long last_gc;
984 static int rover;
985 static int equilibrium;
986 struct rtable *rth;
987 struct rtable __rcu **rthp;
988 unsigned long now = jiffies;
989 int goal;
990 int entries = dst_entries_get_fast(&ipv4_dst_ops);
991
992 /*
993 * Garbage collection is pretty expensive,
994 * do not make it too frequently.
995 */
996
997 RT_CACHE_STAT_INC(gc_total);
998
999 if (now - last_gc < ip_rt_gc_min_interval &&
1000 entries < ip_rt_max_size) {
1001 RT_CACHE_STAT_INC(gc_ignored);
1002 goto out;
1003 }
1004
1005 entries = dst_entries_get_slow(&ipv4_dst_ops);
1006 /* Calculate number of entries, which we want to expire now. */
1007 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
1008 if (goal <= 0) {
1009 if (equilibrium < ipv4_dst_ops.gc_thresh)
1010 equilibrium = ipv4_dst_ops.gc_thresh;
1011 goal = entries - equilibrium;
1012 if (goal > 0) {
1013 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1014 goal = entries - equilibrium;
1015 }
1016 } else {
1017 /* We are in dangerous area. Try to reduce cache really
1018 * aggressively.
1019 */
1020 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1021 equilibrium = entries - goal;
1022 }
1023
1024 if (now - last_gc >= ip_rt_gc_min_interval)
1025 last_gc = now;
1026
1027 if (goal <= 0) {
1028 equilibrium += goal;
1029 goto work_done;
1030 }
1031
1032 do {
1033 int i, k;
1034
1035 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1036 unsigned long tmo = expire;
1037
1038 k = (k + 1) & rt_hash_mask;
1039 rthp = &rt_hash_table[k].chain;
1040 spin_lock_bh(rt_hash_lock_addr(k));
1041 while ((rth = rcu_dereference_protected(*rthp,
1042 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1043 if (!rt_is_expired(rth) &&
1044 !rt_may_expire(rth, tmo, expire)) {
1045 tmo >>= 1;
1046 rthp = &rth->dst.rt_next;
1047 continue;
1048 }
1049 *rthp = rth->dst.rt_next;
1050 rt_free(rth);
1051 goal--;
1052 }
1053 spin_unlock_bh(rt_hash_lock_addr(k));
1054 if (goal <= 0)
1055 break;
1056 }
1057 rover = k;
1058
1059 if (goal <= 0)
1060 goto work_done;
1061
1062 /* Goal is not achieved. We stop process if:
1063
1064 - if expire reduced to zero. Otherwise, expire is halfed.
1065 - if table is not full.
1066 - if we are called from interrupt.
1067 - jiffies check is just fallback/debug loop breaker.
1068 We will not spin here for long time in any case.
1069 */
1070
1071 RT_CACHE_STAT_INC(gc_goal_miss);
1072
1073 if (expire == 0)
1074 break;
1075
1076 expire >>= 1;
1077
1078 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1079 goto out;
1080 } while (!in_softirq() && time_before_eq(jiffies, now));
1081
1082 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1083 goto out;
1084 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1085 goto out;
1086 net_warn_ratelimited("dst cache overflow\n");
1087 RT_CACHE_STAT_INC(gc_dst_overflow);
1088 return 1;
1089
1090work_done:
1091 expire += ip_rt_gc_min_interval;
1092 if (expire > ip_rt_gc_timeout ||
1093 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1094 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1095 expire = ip_rt_gc_timeout;
1096out: return 0;
1097}
1098
1099/*
1100 * Returns number of entries in a hash chain that have different hash_inputs
1101 */
1102static int slow_chain_length(const struct rtable *head)
1103{
1104 int length = 0;
1105 const struct rtable *rth = head;
1106
1107 while (rth) {
1108 length += has_noalias(head, rth);
1109 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1110 }
1111 return length >> FRACT_BITS;
1112}
1113
1114static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, const void *daddr)
1115{
1116 static const __be32 inaddr_any = 0;
1117 struct net_device *dev = dst->dev;
1118 const __be32 *pkey = daddr;
1119 const struct rtable *rt;
1120 struct neighbour *n;
1121
1122 rt = (const struct rtable *) dst;
1123
1124 if (dev->flags & (IFF_LOOPBACK | IFF_POINTOPOINT))
1125 pkey = &inaddr_any;
1126 else if (rt->rt_gateway)
1127 pkey = (const __be32 *) &rt->rt_gateway;
1128
1129 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
1130 if (n)
1131 return n;
1132 return neigh_create(&arp_tbl, pkey, dev);
1133}
1134
1135static int rt_bind_neighbour(struct rtable *rt)
1136{
1137 struct neighbour *n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1138 if (IS_ERR(n))
1139 return PTR_ERR(n);
1140 dst_set_neighbour(&rt->dst, n);
1141
1142 return 0;
1143}
1144
1145static struct rtable *rt_intern_hash(unsigned int hash, struct rtable *rt,
1146 struct sk_buff *skb, int ifindex)
1147{
1148 struct rtable *rth, *cand;
1149 struct rtable __rcu **rthp, **candp;
1150 unsigned long now;
1151 u32 min_score;
1152 int chain_length;
1153 int attempts = !in_softirq();
1154
1155restart:
1156 chain_length = 0;
1157 min_score = ~(u32)0;
1158 cand = NULL;
1159 candp = NULL;
1160 now = jiffies;
1161
1162 if (!rt_caching(dev_net(rt->dst.dev))) {
1163 /*
1164 * If we're not caching, just tell the caller we
1165 * were successful and don't touch the route. The
1166 * caller hold the sole reference to the cache entry, and
1167 * it will be released when the caller is done with it.
1168 * If we drop it here, the callers have no way to resolve routes
1169 * when we're not caching. Instead, just point *rp at rt, so
1170 * the caller gets a single use out of the route
1171 * Note that we do rt_free on this new route entry, so that
1172 * once its refcount hits zero, we are still able to reap it
1173 * (Thanks Alexey)
1174 * Note: To avoid expensive rcu stuff for this uncached dst,
1175 * we set DST_NOCACHE so that dst_release() can free dst without
1176 * waiting a grace period.
1177 */
1178
1179 rt->dst.flags |= DST_NOCACHE;
1180 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1181 int err = rt_bind_neighbour(rt);
1182 if (err) {
1183 net_warn_ratelimited("Neighbour table failure & not caching routes\n");
1184 ip_rt_put(rt);
1185 return ERR_PTR(err);
1186 }
1187 }
1188
1189 goto skip_hashing;
1190 }
1191
1192 rthp = &rt_hash_table[hash].chain;
1193
1194 spin_lock_bh(rt_hash_lock_addr(hash));
1195 while ((rth = rcu_dereference_protected(*rthp,
1196 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1197 if (rt_is_expired(rth)) {
1198 *rthp = rth->dst.rt_next;
1199 rt_free(rth);
1200 continue;
1201 }
1202 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1203 /* Put it first */
1204 *rthp = rth->dst.rt_next;
1205 /*
1206 * Since lookup is lockfree, the deletion
1207 * must be visible to another weakly ordered CPU before
1208 * the insertion at the start of the hash chain.
1209 */
1210 rcu_assign_pointer(rth->dst.rt_next,
1211 rt_hash_table[hash].chain);
1212 /*
1213 * Since lookup is lockfree, the update writes
1214 * must be ordered for consistency on SMP.
1215 */
1216 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1217
1218 dst_use(&rth->dst, now);
1219 spin_unlock_bh(rt_hash_lock_addr(hash));
1220
1221 rt_drop(rt);
1222 if (skb)
1223 skb_dst_set(skb, &rth->dst);
1224 return rth;
1225 }
1226
1227 if (!atomic_read(&rth->dst.__refcnt)) {
1228 u32 score = rt_score(rth);
1229
1230 if (score <= min_score) {
1231 cand = rth;
1232 candp = rthp;
1233 min_score = score;
1234 }
1235 }
1236
1237 chain_length++;
1238
1239 rthp = &rth->dst.rt_next;
1240 }
1241
1242 if (cand) {
1243 /* ip_rt_gc_elasticity used to be average length of chain
1244 * length, when exceeded gc becomes really aggressive.
1245 *
1246 * The second limit is less certain. At the moment it allows
1247 * only 2 entries per bucket. We will see.
1248 */
1249 if (chain_length > ip_rt_gc_elasticity) {
1250 *candp = cand->dst.rt_next;
1251 rt_free(cand);
1252 }
1253 } else {
1254 if (chain_length > rt_chain_length_max &&
1255 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1256 struct net *net = dev_net(rt->dst.dev);
1257 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1258 if (!rt_caching(net)) {
1259 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1260 rt->dst.dev->name, num);
1261 }
1262 rt_emergency_hash_rebuild(net);
1263 spin_unlock_bh(rt_hash_lock_addr(hash));
1264
1265 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1266 ifindex, rt_genid(net));
1267 goto restart;
1268 }
1269 }
1270
1271 /* Try to bind route to arp only if it is output
1272 route or unicast forwarding path.
1273 */
1274 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1275 int err = rt_bind_neighbour(rt);
1276 if (err) {
1277 spin_unlock_bh(rt_hash_lock_addr(hash));
1278
1279 if (err != -ENOBUFS) {
1280 rt_drop(rt);
1281 return ERR_PTR(err);
1282 }
1283
1284 /* Neighbour tables are full and nothing
1285 can be released. Try to shrink route cache,
1286 it is most likely it holds some neighbour records.
1287 */
1288 if (attempts-- > 0) {
1289 int saved_elasticity = ip_rt_gc_elasticity;
1290 int saved_int = ip_rt_gc_min_interval;
1291 ip_rt_gc_elasticity = 1;
1292 ip_rt_gc_min_interval = 0;
1293 rt_garbage_collect(&ipv4_dst_ops);
1294 ip_rt_gc_min_interval = saved_int;
1295 ip_rt_gc_elasticity = saved_elasticity;
1296 goto restart;
1297 }
1298
1299 net_warn_ratelimited("Neighbour table overflow\n");
1300 rt_drop(rt);
1301 return ERR_PTR(-ENOBUFS);
1302 }
1303 }
1304
1305 rt->dst.rt_next = rt_hash_table[hash].chain;
1306
1307 /*
1308 * Since lookup is lockfree, we must make sure
1309 * previous writes to rt are committed to memory
1310 * before making rt visible to other CPUS.
1311 */
1312 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1313
1314 spin_unlock_bh(rt_hash_lock_addr(hash));
1315
1316skip_hashing:
1317 if (skb)
1318 skb_dst_set(skb, &rt->dst);
1319 return rt;
1320}
1321
1322static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1323
1324static u32 rt_peer_genid(void)
1325{
1326 return atomic_read(&__rt_peer_genid);
1327}
1328
1329void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1330{
1331 struct inet_peer *peer;
1332
1333 peer = inet_getpeer_v4(daddr, create);
1334
1335 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1336 inet_putpeer(peer);
1337 else
1338 rt->rt_peer_genid = rt_peer_genid();
1339}
1340
1341/*
1342 * Peer allocation may fail only in serious out-of-memory conditions. However
1343 * we still can generate some output.
1344 * Random ID selection looks a bit dangerous because we have no chances to
1345 * select ID being unique in a reasonable period of time.
1346 * But broken packet identifier may be better than no packet at all.
1347 */
1348static void ip_select_fb_ident(struct iphdr *iph)
1349{
1350 static DEFINE_SPINLOCK(ip_fb_id_lock);
1351 static u32 ip_fallback_id;
1352 u32 salt;
1353
1354 spin_lock_bh(&ip_fb_id_lock);
1355 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1356 iph->id = htons(salt & 0xFFFF);
1357 ip_fallback_id = salt;
1358 spin_unlock_bh(&ip_fb_id_lock);
1359}
1360
1361void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1362{
1363 struct rtable *rt = (struct rtable *) dst;
1364
1365 if (rt && !(rt->dst.flags & DST_NOPEER)) {
1366 if (rt->peer == NULL)
1367 rt_bind_peer(rt, rt->rt_dst, 1);
1368
1369 /* If peer is attached to destination, it is never detached,
1370 so that we need not to grab a lock to dereference it.
1371 */
1372 if (rt->peer) {
1373 iph->id = htons(inet_getid(rt->peer, more));
1374 return;
1375 }
1376 } else if (!rt)
1377 pr_debug("rt_bind_peer(0) @%p\n", __builtin_return_address(0));
1378
1379 ip_select_fb_ident(iph);
1380}
1381EXPORT_SYMBOL(__ip_select_ident);
1382
1383static void rt_del(unsigned int hash, struct rtable *rt)
1384{
1385 struct rtable __rcu **rthp;
1386 struct rtable *aux;
1387
1388 rthp = &rt_hash_table[hash].chain;
1389 spin_lock_bh(rt_hash_lock_addr(hash));
1390 ip_rt_put(rt);
1391 while ((aux = rcu_dereference_protected(*rthp,
1392 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1393 if (aux == rt || rt_is_expired(aux)) {
1394 *rthp = aux->dst.rt_next;
1395 rt_free(aux);
1396 continue;
1397 }
1398 rthp = &aux->dst.rt_next;
1399 }
1400 spin_unlock_bh(rt_hash_lock_addr(hash));
1401}
1402
1403static void check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1404{
1405 struct rtable *rt = (struct rtable *) dst;
1406 __be32 orig_gw = rt->rt_gateway;
1407 struct neighbour *n, *old_n;
1408
1409 dst_confirm(&rt->dst);
1410
1411 rt->rt_gateway = peer->redirect_learned.a4;
1412
1413 n = ipv4_neigh_lookup(&rt->dst, &rt->rt_gateway);
1414 if (IS_ERR(n)) {
1415 rt->rt_gateway = orig_gw;
1416 return;
1417 }
1418 old_n = xchg(&rt->dst._neighbour, n);
1419 if (old_n)
1420 neigh_release(old_n);
1421 if (!(n->nud_state & NUD_VALID)) {
1422 neigh_event_send(n, NULL);
1423 } else {
1424 rt->rt_flags |= RTCF_REDIRECTED;
1425 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1426 }
1427}
1428
1429/* called in rcu_read_lock() section */
1430void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1431 __be32 saddr, struct net_device *dev)
1432{
1433 int s, i;
1434 struct in_device *in_dev = __in_dev_get_rcu(dev);
1435 __be32 skeys[2] = { saddr, 0 };
1436 int ikeys[2] = { dev->ifindex, 0 };
1437 struct inet_peer *peer;
1438 struct net *net;
1439
1440 if (!in_dev)
1441 return;
1442
1443 net = dev_net(dev);
1444 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1445 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1446 ipv4_is_zeronet(new_gw))
1447 goto reject_redirect;
1448
1449 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1450 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1451 goto reject_redirect;
1452 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1453 goto reject_redirect;
1454 } else {
1455 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1456 goto reject_redirect;
1457 }
1458
1459 for (s = 0; s < 2; s++) {
1460 for (i = 0; i < 2; i++) {
1461 unsigned int hash;
1462 struct rtable __rcu **rthp;
1463 struct rtable *rt;
1464
1465 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1466
1467 rthp = &rt_hash_table[hash].chain;
1468
1469 while ((rt = rcu_dereference(*rthp)) != NULL) {
1470 rthp = &rt->dst.rt_next;
1471
1472 if (rt->rt_key_dst != daddr ||
1473 rt->rt_key_src != skeys[s] ||
1474 rt->rt_oif != ikeys[i] ||
1475 rt_is_input_route(rt) ||
1476 rt_is_expired(rt) ||
1477 !net_eq(dev_net(rt->dst.dev), net) ||
1478 rt->dst.error ||
1479 rt->dst.dev != dev ||
1480 rt->rt_gateway != old_gw)
1481 continue;
1482
1483 if (!rt->peer)
1484 rt_bind_peer(rt, rt->rt_dst, 1);
1485
1486 peer = rt->peer;
1487 if (peer) {
1488 if (peer->redirect_learned.a4 != new_gw) {
1489 peer->redirect_learned.a4 = new_gw;
1490 atomic_inc(&__rt_peer_genid);
1491 }
1492 check_peer_redir(&rt->dst, peer);
1493 }
1494 }
1495 }
1496 }
1497 return;
1498
1499reject_redirect:
1500#ifdef CONFIG_IP_ROUTE_VERBOSE
1501 if (IN_DEV_LOG_MARTIANS(in_dev))
1502 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1503 " Advised path = %pI4 -> %pI4\n",
1504 &old_gw, dev->name, &new_gw,
1505 &saddr, &daddr);
1506#endif
1507 ;
1508}
1509
1510static bool peer_pmtu_expired(struct inet_peer *peer)
1511{
1512 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1513
1514 return orig &&
1515 time_after_eq(jiffies, orig) &&
1516 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1517}
1518
1519static bool peer_pmtu_cleaned(struct inet_peer *peer)
1520{
1521 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1522
1523 return orig &&
1524 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1525}
1526
1527static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1528{
1529 struct rtable *rt = (struct rtable *)dst;
1530 struct dst_entry *ret = dst;
1531
1532 if (rt) {
1533 if (dst->obsolete > 0) {
1534 ip_rt_put(rt);
1535 ret = NULL;
1536 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1537 unsigned int hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1538 rt->rt_oif,
1539 rt_genid(dev_net(dst->dev)));
1540 rt_del(hash, rt);
1541 ret = NULL;
1542 } else if (rt->peer && peer_pmtu_expired(rt->peer)) {
1543 dst_metric_set(dst, RTAX_MTU, rt->peer->pmtu_orig);
1544 }
1545 }
1546 return ret;
1547}
1548
1549/*
1550 * Algorithm:
1551 * 1. The first ip_rt_redirect_number redirects are sent
1552 * with exponential backoff, then we stop sending them at all,
1553 * assuming that the host ignores our redirects.
1554 * 2. If we did not see packets requiring redirects
1555 * during ip_rt_redirect_silence, we assume that the host
1556 * forgot redirected route and start to send redirects again.
1557 *
1558 * This algorithm is much cheaper and more intelligent than dumb load limiting
1559 * in icmp.c.
1560 *
1561 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1562 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1563 */
1564
1565void ip_rt_send_redirect(struct sk_buff *skb)
1566{
1567 struct rtable *rt = skb_rtable(skb);
1568 struct in_device *in_dev;
1569 struct inet_peer *peer;
1570 int log_martians;
1571
1572 rcu_read_lock();
1573 in_dev = __in_dev_get_rcu(rt->dst.dev);
1574 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1575 rcu_read_unlock();
1576 return;
1577 }
1578 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1579 rcu_read_unlock();
1580
1581 if (!rt->peer)
1582 rt_bind_peer(rt, rt->rt_dst, 1);
1583 peer = rt->peer;
1584 if (!peer) {
1585 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1586 return;
1587 }
1588
1589 /* No redirected packets during ip_rt_redirect_silence;
1590 * reset the algorithm.
1591 */
1592 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1593 peer->rate_tokens = 0;
1594
1595 /* Too many ignored redirects; do not send anything
1596 * set dst.rate_last to the last seen redirected packet.
1597 */
1598 if (peer->rate_tokens >= ip_rt_redirect_number) {
1599 peer->rate_last = jiffies;
1600 return;
1601 }
1602
1603 /* Check for load limit; set rate_last to the latest sent
1604 * redirect.
1605 */
1606 if (peer->rate_tokens == 0 ||
1607 time_after(jiffies,
1608 (peer->rate_last +
1609 (ip_rt_redirect_load << peer->rate_tokens)))) {
1610 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1611 peer->rate_last = jiffies;
1612 ++peer->rate_tokens;
1613#ifdef CONFIG_IP_ROUTE_VERBOSE
1614 if (log_martians &&
1615 peer->rate_tokens == ip_rt_redirect_number)
1616 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1617 &ip_hdr(skb)->saddr, rt->rt_iif,
1618 &rt->rt_dst, &rt->rt_gateway);
1619#endif
1620 }
1621}
1622
1623static int ip_error(struct sk_buff *skb)
1624{
1625 struct rtable *rt = skb_rtable(skb);
1626 struct inet_peer *peer;
1627 unsigned long now;
1628 bool send;
1629 int code;
1630
1631 switch (rt->dst.error) {
1632 case EINVAL:
1633 default:
1634 goto out;
1635 case EHOSTUNREACH:
1636 code = ICMP_HOST_UNREACH;
1637 break;
1638 case ENETUNREACH:
1639 code = ICMP_NET_UNREACH;
1640 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1641 IPSTATS_MIB_INNOROUTES);
1642 break;
1643 case EACCES:
1644 code = ICMP_PKT_FILTERED;
1645 break;
1646 }
1647
1648 if (!rt->peer)
1649 rt_bind_peer(rt, rt->rt_dst, 1);
1650 peer = rt->peer;
1651
1652 send = true;
1653 if (peer) {
1654 now = jiffies;
1655 peer->rate_tokens += now - peer->rate_last;
1656 if (peer->rate_tokens > ip_rt_error_burst)
1657 peer->rate_tokens = ip_rt_error_burst;
1658 peer->rate_last = now;
1659 if (peer->rate_tokens >= ip_rt_error_cost)
1660 peer->rate_tokens -= ip_rt_error_cost;
1661 else
1662 send = false;
1663 }
1664 if (send)
1665 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1666
1667out: kfree_skb(skb);
1668 return 0;
1669}
1670
1671/*
1672 * The last two values are not from the RFC but
1673 * are needed for AMPRnet AX.25 paths.
1674 */
1675
1676static const unsigned short mtu_plateau[] =
1677{32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1678
1679static inline unsigned short guess_mtu(unsigned short old_mtu)
1680{
1681 int i;
1682
1683 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1684 if (old_mtu > mtu_plateau[i])
1685 return mtu_plateau[i];
1686 return 68;
1687}
1688
1689unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1690 unsigned short new_mtu,
1691 struct net_device *dev)
1692{
1693 unsigned short old_mtu = ntohs(iph->tot_len);
1694 unsigned short est_mtu = 0;
1695 struct inet_peer *peer;
1696
1697 peer = inet_getpeer_v4(iph->daddr, 1);
1698 if (peer) {
1699 unsigned short mtu = new_mtu;
1700
1701 if (new_mtu < 68 || new_mtu >= old_mtu) {
1702 /* BSD 4.2 derived systems incorrectly adjust
1703 * tot_len by the IP header length, and report
1704 * a zero MTU in the ICMP message.
1705 */
1706 if (mtu == 0 &&
1707 old_mtu >= 68 + (iph->ihl << 2))
1708 old_mtu -= iph->ihl << 2;
1709 mtu = guess_mtu(old_mtu);
1710 }
1711
1712 if (mtu < ip_rt_min_pmtu)
1713 mtu = ip_rt_min_pmtu;
1714 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1715 unsigned long pmtu_expires;
1716
1717 pmtu_expires = jiffies + ip_rt_mtu_expires;
1718 if (!pmtu_expires)
1719 pmtu_expires = 1UL;
1720
1721 est_mtu = mtu;
1722 peer->pmtu_learned = mtu;
1723 peer->pmtu_expires = pmtu_expires;
1724 atomic_inc(&__rt_peer_genid);
1725 }
1726
1727 inet_putpeer(peer);
1728 }
1729 return est_mtu ? : new_mtu;
1730}
1731
1732static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1733{
1734 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1735
1736 if (!expires)
1737 return;
1738 if (time_before(jiffies, expires)) {
1739 u32 orig_dst_mtu = dst_mtu(dst);
1740 if (peer->pmtu_learned < orig_dst_mtu) {
1741 if (!peer->pmtu_orig)
1742 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1743 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1744 }
1745 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1746 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1747}
1748
1749static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1750{
1751 struct rtable *rt = (struct rtable *) dst;
1752 struct inet_peer *peer;
1753
1754 dst_confirm(dst);
1755
1756 if (!rt->peer)
1757 rt_bind_peer(rt, rt->rt_dst, 1);
1758 peer = rt->peer;
1759 if (peer) {
1760 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1761
1762 if (mtu < ip_rt_min_pmtu)
1763 mtu = ip_rt_min_pmtu;
1764 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1765
1766 pmtu_expires = jiffies + ip_rt_mtu_expires;
1767 if (!pmtu_expires)
1768 pmtu_expires = 1UL;
1769
1770 peer->pmtu_learned = mtu;
1771 peer->pmtu_expires = pmtu_expires;
1772
1773 atomic_inc(&__rt_peer_genid);
1774 rt->rt_peer_genid = rt_peer_genid();
1775 }
1776 check_peer_pmtu(dst, peer);
1777 }
1778}
1779
1780
1781static void ipv4_validate_peer(struct rtable *rt)
1782{
1783 if (rt->rt_peer_genid != rt_peer_genid()) {
1784 struct inet_peer *peer;
1785
1786 if (!rt->peer)
1787 rt_bind_peer(rt, rt->rt_dst, 0);
1788
1789 peer = rt->peer;
1790 if (peer) {
1791 check_peer_pmtu(&rt->dst, peer);
1792
1793 if (peer->redirect_learned.a4 &&
1794 peer->redirect_learned.a4 != rt->rt_gateway)
1795 check_peer_redir(&rt->dst, peer);
1796 }
1797
1798 rt->rt_peer_genid = rt_peer_genid();
1799 }
1800}
1801
1802static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1803{
1804 struct rtable *rt = (struct rtable *) dst;
1805
1806 if (rt_is_expired(rt))
1807 return NULL;
1808 ipv4_validate_peer(rt);
1809 return dst;
1810}
1811
1812static void ipv4_dst_destroy(struct dst_entry *dst)
1813{
1814 struct rtable *rt = (struct rtable *) dst;
1815 struct inet_peer *peer = rt->peer;
1816
1817 if (rt->fi) {
1818 fib_info_put(rt->fi);
1819 rt->fi = NULL;
1820 }
1821 if (peer) {
1822 rt->peer = NULL;
1823 inet_putpeer(peer);
1824 }
1825}
1826
1827
1828static void ipv4_link_failure(struct sk_buff *skb)
1829{
1830 struct rtable *rt;
1831
1832 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1833
1834 rt = skb_rtable(skb);
1835 if (rt && rt->peer && peer_pmtu_cleaned(rt->peer))
1836 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1837}
1838
1839static int ip_rt_bug(struct sk_buff *skb)
1840{
1841 pr_debug("%s: %pI4 -> %pI4, %s\n",
1842 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1843 skb->dev ? skb->dev->name : "?");
1844 kfree_skb(skb);
1845 WARN_ON(1);
1846 return 0;
1847}
1848
1849/*
1850 We do not cache source address of outgoing interface,
1851 because it is used only by IP RR, TS and SRR options,
1852 so that it out of fast path.
1853
1854 BTW remember: "addr" is allowed to be not aligned
1855 in IP options!
1856 */
1857
1858void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1859{
1860 __be32 src;
1861
1862 if (rt_is_output_route(rt))
1863 src = ip_hdr(skb)->saddr;
1864 else {
1865 struct fib_result res;
1866 struct flowi4 fl4;
1867 struct iphdr *iph;
1868
1869 iph = ip_hdr(skb);
1870
1871 memset(&fl4, 0, sizeof(fl4));
1872 fl4.daddr = iph->daddr;
1873 fl4.saddr = iph->saddr;
1874 fl4.flowi4_tos = RT_TOS(iph->tos);
1875 fl4.flowi4_oif = rt->dst.dev->ifindex;
1876 fl4.flowi4_iif = skb->dev->ifindex;
1877 fl4.flowi4_mark = skb->mark;
1878
1879 rcu_read_lock();
1880 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1881 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1882 else
1883 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1884 RT_SCOPE_UNIVERSE);
1885 rcu_read_unlock();
1886 }
1887 memcpy(addr, &src, 4);
1888}
1889
1890#ifdef CONFIG_IP_ROUTE_CLASSID
1891static void set_class_tag(struct rtable *rt, u32 tag)
1892{
1893 if (!(rt->dst.tclassid & 0xFFFF))
1894 rt->dst.tclassid |= tag & 0xFFFF;
1895 if (!(rt->dst.tclassid & 0xFFFF0000))
1896 rt->dst.tclassid |= tag & 0xFFFF0000;
1897}
1898#endif
1899
1900static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1901{
1902 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1903
1904 if (advmss == 0) {
1905 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1906 ip_rt_min_advmss);
1907 if (advmss > 65535 - 40)
1908 advmss = 65535 - 40;
1909 }
1910 return advmss;
1911}
1912
1913static unsigned int ipv4_mtu(const struct dst_entry *dst)
1914{
1915 const struct rtable *rt = (const struct rtable *) dst;
1916 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1917
1918 if (mtu && rt_is_output_route(rt))
1919 return mtu;
1920
1921 mtu = dst->dev->mtu;
1922
1923 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1924
1925 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1926 mtu = 576;
1927 }
1928
1929 if (mtu > IP_MAX_MTU)
1930 mtu = IP_MAX_MTU;
1931
1932 return mtu;
1933}
1934
1935static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1936 struct fib_info *fi)
1937{
1938 struct inet_peer *peer;
1939 int create = 0;
1940
1941 /* If a peer entry exists for this destination, we must hook
1942 * it up in order to get at cached metrics.
1943 */
1944 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1945 create = 1;
1946
1947 rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1948 if (peer) {
1949 rt->rt_peer_genid = rt_peer_genid();
1950 if (inet_metrics_new(peer))
1951 memcpy(peer->metrics, fi->fib_metrics,
1952 sizeof(u32) * RTAX_MAX);
1953 dst_init_metrics(&rt->dst, peer->metrics, false);
1954
1955 check_peer_pmtu(&rt->dst, peer);
1956
1957 if (peer->redirect_learned.a4 &&
1958 peer->redirect_learned.a4 != rt->rt_gateway) {
1959 rt->rt_gateway = peer->redirect_learned.a4;
1960 rt->rt_flags |= RTCF_REDIRECTED;
1961 }
1962 } else {
1963 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1964 rt->fi = fi;
1965 atomic_inc(&fi->fib_clntref);
1966 }
1967 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1968 }
1969}
1970
1971static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1972 const struct fib_result *res,
1973 struct fib_info *fi, u16 type, u32 itag)
1974{
1975 struct dst_entry *dst = &rt->dst;
1976
1977 if (fi) {
1978 if (FIB_RES_GW(*res) &&
1979 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1980 rt->rt_gateway = FIB_RES_GW(*res);
1981 rt_init_metrics(rt, fl4, fi);
1982#ifdef CONFIG_IP_ROUTE_CLASSID
1983 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1984#endif
1985 }
1986
1987 if (dst_mtu(dst) > IP_MAX_MTU)
1988 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1989 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1990 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1991
1992#ifdef CONFIG_IP_ROUTE_CLASSID
1993#ifdef CONFIG_IP_MULTIPLE_TABLES
1994 set_class_tag(rt, fib_rules_tclass(res));
1995#endif
1996 set_class_tag(rt, itag);
1997#endif
1998}
1999
2000static struct rtable *rt_dst_alloc(struct net_device *dev,
2001 bool nopolicy, bool noxfrm)
2002{
2003 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
2004 DST_HOST |
2005 (nopolicy ? DST_NOPOLICY : 0) |
2006 (noxfrm ? DST_NOXFRM : 0));
2007}
2008
2009/* called in rcu_read_lock() section */
2010static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2011 u8 tos, struct net_device *dev, int our)
2012{
2013 unsigned int hash;
2014 struct rtable *rth;
2015 __be32 spec_dst;
2016 struct in_device *in_dev = __in_dev_get_rcu(dev);
2017 u32 itag = 0;
2018 int err;
2019
2020 /* Primary sanity checks. */
2021
2022 if (in_dev == NULL)
2023 return -EINVAL;
2024
2025 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2026 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
2027 goto e_inval;
2028
2029 if (ipv4_is_zeronet(saddr)) {
2030 if (!ipv4_is_local_multicast(daddr))
2031 goto e_inval;
2032 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2033 } else {
2034 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2035 &itag);
2036 if (err < 0)
2037 goto e_err;
2038 }
2039 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
2040 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2041 if (!rth)
2042 goto e_nobufs;
2043
2044#ifdef CONFIG_IP_ROUTE_CLASSID
2045 rth->dst.tclassid = itag;
2046#endif
2047 rth->dst.output = ip_rt_bug;
2048
2049 rth->rt_key_dst = daddr;
2050 rth->rt_key_src = saddr;
2051 rth->rt_genid = rt_genid(dev_net(dev));
2052 rth->rt_flags = RTCF_MULTICAST;
2053 rth->rt_type = RTN_MULTICAST;
2054 rth->rt_key_tos = tos;
2055 rth->rt_dst = daddr;
2056 rth->rt_src = saddr;
2057 rth->rt_route_iif = dev->ifindex;
2058 rth->rt_iif = dev->ifindex;
2059 rth->rt_oif = 0;
2060 rth->rt_mark = skb->mark;
2061 rth->rt_gateway = daddr;
2062 rth->rt_spec_dst= spec_dst;
2063 rth->rt_peer_genid = 0;
2064 rth->peer = NULL;
2065 rth->fi = NULL;
2066 if (our) {
2067 rth->dst.input= ip_local_deliver;
2068 rth->rt_flags |= RTCF_LOCAL;
2069 }
2070
2071#ifdef CONFIG_IP_MROUTE
2072 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
2073 rth->dst.input = ip_mr_input;
2074#endif
2075 RT_CACHE_STAT_INC(in_slow_mc);
2076
2077 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
2078 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
2079 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
2080
2081e_nobufs:
2082 return -ENOBUFS;
2083e_inval:
2084 return -EINVAL;
2085e_err:
2086 return err;
2087}
2088
2089
2090static void ip_handle_martian_source(struct net_device *dev,
2091 struct in_device *in_dev,
2092 struct sk_buff *skb,
2093 __be32 daddr,
2094 __be32 saddr)
2095{
2096 RT_CACHE_STAT_INC(in_martian_src);
2097#ifdef CONFIG_IP_ROUTE_VERBOSE
2098 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2099 /*
2100 * RFC1812 recommendation, if source is martian,
2101 * the only hint is MAC header.
2102 */
2103 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
2104 &daddr, &saddr, dev->name);
2105 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2106 print_hex_dump(KERN_WARNING, "ll header: ",
2107 DUMP_PREFIX_OFFSET, 16, 1,
2108 skb_mac_header(skb),
2109 dev->hard_header_len, true);
2110 }
2111 }
2112#endif
2113}
2114
2115/* called in rcu_read_lock() section */
2116static int __mkroute_input(struct sk_buff *skb,
2117 const struct fib_result *res,
2118 struct in_device *in_dev,
2119 __be32 daddr, __be32 saddr, u32 tos,
2120 struct rtable **result)
2121{
2122 struct rtable *rth;
2123 int err;
2124 struct in_device *out_dev;
2125 unsigned int flags = 0;
2126 __be32 spec_dst;
2127 u32 itag;
2128
2129 /* get a working reference to the output device */
2130 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2131 if (out_dev == NULL) {
2132 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
2133 return -EINVAL;
2134 }
2135
2136
2137 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2138 in_dev->dev, &spec_dst, &itag);
2139 if (err < 0) {
2140 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2141 saddr);
2142
2143 goto cleanup;
2144 }
2145
2146 if (err)
2147 flags |= RTCF_DIRECTSRC;
2148
2149 if (out_dev == in_dev && err &&
2150 (IN_DEV_SHARED_MEDIA(out_dev) ||
2151 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2152 flags |= RTCF_DOREDIRECT;
2153
2154 if (skb->protocol != htons(ETH_P_IP)) {
2155 /* Not IP (i.e. ARP). Do not create route, if it is
2156 * invalid for proxy arp. DNAT routes are always valid.
2157 *
2158 * Proxy arp feature have been extended to allow, ARP
2159 * replies back to the same interface, to support
2160 * Private VLAN switch technologies. See arp.c.
2161 */
2162 if (out_dev == in_dev &&
2163 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2164 err = -EINVAL;
2165 goto cleanup;
2166 }
2167 }
2168
2169 rth = rt_dst_alloc(out_dev->dev,
2170 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2171 IN_DEV_CONF_GET(out_dev, NOXFRM));
2172 if (!rth) {
2173 err = -ENOBUFS;
2174 goto cleanup;
2175 }
2176
2177 rth->rt_key_dst = daddr;
2178 rth->rt_key_src = saddr;
2179 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2180 rth->rt_flags = flags;
2181 rth->rt_type = res->type;
2182 rth->rt_key_tos = tos;
2183 rth->rt_dst = daddr;
2184 rth->rt_src = saddr;
2185 rth->rt_route_iif = in_dev->dev->ifindex;
2186 rth->rt_iif = in_dev->dev->ifindex;
2187 rth->rt_oif = 0;
2188 rth->rt_mark = skb->mark;
2189 rth->rt_gateway = daddr;
2190 rth->rt_spec_dst= spec_dst;
2191 rth->rt_peer_genid = 0;
2192 rth->peer = NULL;
2193 rth->fi = NULL;
2194
2195 rth->dst.input = ip_forward;
2196 rth->dst.output = ip_output;
2197
2198 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2199
2200 *result = rth;
2201 err = 0;
2202 cleanup:
2203 return err;
2204}
2205
2206static int ip_mkroute_input(struct sk_buff *skb,
2207 struct fib_result *res,
2208 const struct flowi4 *fl4,
2209 struct in_device *in_dev,
2210 __be32 daddr, __be32 saddr, u32 tos)
2211{
2212 struct rtable *rth = NULL;
2213 int err;
2214 unsigned int hash;
2215
2216#ifdef CONFIG_IP_ROUTE_MULTIPATH
2217 if (res->fi && res->fi->fib_nhs > 1)
2218 fib_select_multipath(res);
2219#endif
2220
2221 /* create a routing cache entry */
2222 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2223 if (err)
2224 return err;
2225
2226 /* put it into the cache */
2227 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2228 rt_genid(dev_net(rth->dst.dev)));
2229 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2230 if (IS_ERR(rth))
2231 return PTR_ERR(rth);
2232 return 0;
2233}
2234
2235/*
2236 * NOTE. We drop all the packets that has local source
2237 * addresses, because every properly looped back packet
2238 * must have correct destination already attached by output routine.
2239 *
2240 * Such approach solves two big problems:
2241 * 1. Not simplex devices are handled properly.
2242 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2243 * called with rcu_read_lock()
2244 */
2245
2246static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2247 u8 tos, struct net_device *dev)
2248{
2249 struct fib_result res;
2250 struct in_device *in_dev = __in_dev_get_rcu(dev);
2251 struct flowi4 fl4;
2252 unsigned int flags = 0;
2253 u32 itag = 0;
2254 struct rtable *rth;
2255 unsigned int hash;
2256 __be32 spec_dst;
2257 int err = -EINVAL;
2258 struct net *net = dev_net(dev);
2259
2260 /* IP on this device is disabled. */
2261
2262 if (!in_dev)
2263 goto out;
2264
2265 /* Check for the most weird martians, which can be not detected
2266 by fib_lookup.
2267 */
2268
2269 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2270 ipv4_is_loopback(saddr))
2271 goto martian_source;
2272
2273 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2274 goto brd_input;
2275
2276 /* Accept zero addresses only to limited broadcast;
2277 * I even do not know to fix it or not. Waiting for complains :-)
2278 */
2279 if (ipv4_is_zeronet(saddr))
2280 goto martian_source;
2281
2282 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2283 goto martian_destination;
2284
2285 /*
2286 * Now we are ready to route packet.
2287 */
2288 fl4.flowi4_oif = 0;
2289 fl4.flowi4_iif = dev->ifindex;
2290 fl4.flowi4_mark = skb->mark;
2291 fl4.flowi4_tos = tos;
2292 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2293 fl4.daddr = daddr;
2294 fl4.saddr = saddr;
2295 err = fib_lookup(net, &fl4, &res);
2296 if (err != 0) {
2297 if (!IN_DEV_FORWARD(in_dev))
2298 goto e_hostunreach;
2299 goto no_route;
2300 }
2301
2302 RT_CACHE_STAT_INC(in_slow_tot);
2303
2304 if (res.type == RTN_BROADCAST)
2305 goto brd_input;
2306
2307 if (res.type == RTN_LOCAL) {
2308 err = fib_validate_source(skb, saddr, daddr, tos,
2309 net->loopback_dev->ifindex,
2310 dev, &spec_dst, &itag);
2311 if (err < 0)
2312 goto martian_source_keep_err;
2313 if (err)
2314 flags |= RTCF_DIRECTSRC;
2315 spec_dst = daddr;
2316 goto local_input;
2317 }
2318
2319 if (!IN_DEV_FORWARD(in_dev))
2320 goto e_hostunreach;
2321 if (res.type != RTN_UNICAST)
2322 goto martian_destination;
2323
2324 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2325out: return err;
2326
2327brd_input:
2328 if (skb->protocol != htons(ETH_P_IP))
2329 goto e_inval;
2330
2331 if (ipv4_is_zeronet(saddr))
2332 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2333 else {
2334 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2335 &itag);
2336 if (err < 0)
2337 goto martian_source_keep_err;
2338 if (err)
2339 flags |= RTCF_DIRECTSRC;
2340 }
2341 flags |= RTCF_BROADCAST;
2342 res.type = RTN_BROADCAST;
2343 RT_CACHE_STAT_INC(in_brd);
2344
2345local_input:
2346 rth = rt_dst_alloc(net->loopback_dev,
2347 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2348 if (!rth)
2349 goto e_nobufs;
2350
2351 rth->dst.input= ip_local_deliver;
2352 rth->dst.output= ip_rt_bug;
2353#ifdef CONFIG_IP_ROUTE_CLASSID
2354 rth->dst.tclassid = itag;
2355#endif
2356
2357 rth->rt_key_dst = daddr;
2358 rth->rt_key_src = saddr;
2359 rth->rt_genid = rt_genid(net);
2360 rth->rt_flags = flags|RTCF_LOCAL;
2361 rth->rt_type = res.type;
2362 rth->rt_key_tos = tos;
2363 rth->rt_dst = daddr;
2364 rth->rt_src = saddr;
2365#ifdef CONFIG_IP_ROUTE_CLASSID
2366 rth->dst.tclassid = itag;
2367#endif
2368 rth->rt_route_iif = dev->ifindex;
2369 rth->rt_iif = dev->ifindex;
2370 rth->rt_oif = 0;
2371 rth->rt_mark = skb->mark;
2372 rth->rt_gateway = daddr;
2373 rth->rt_spec_dst= spec_dst;
2374 rth->rt_peer_genid = 0;
2375 rth->peer = NULL;
2376 rth->fi = NULL;
2377 if (res.type == RTN_UNREACHABLE) {
2378 rth->dst.input= ip_error;
2379 rth->dst.error= -err;
2380 rth->rt_flags &= ~RTCF_LOCAL;
2381 }
2382 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2383 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2384 err = 0;
2385 if (IS_ERR(rth))
2386 err = PTR_ERR(rth);
2387 goto out;
2388
2389no_route:
2390 RT_CACHE_STAT_INC(in_no_route);
2391 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2392 res.type = RTN_UNREACHABLE;
2393 if (err == -ESRCH)
2394 err = -ENETUNREACH;
2395 goto local_input;
2396
2397 /*
2398 * Do not cache martian addresses: they should be logged (RFC1812)
2399 */
2400martian_destination:
2401 RT_CACHE_STAT_INC(in_martian_dst);
2402#ifdef CONFIG_IP_ROUTE_VERBOSE
2403 if (IN_DEV_LOG_MARTIANS(in_dev))
2404 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2405 &daddr, &saddr, dev->name);
2406#endif
2407
2408e_hostunreach:
2409 err = -EHOSTUNREACH;
2410 goto out;
2411
2412e_inval:
2413 err = -EINVAL;
2414 goto out;
2415
2416e_nobufs:
2417 err = -ENOBUFS;
2418 goto out;
2419
2420martian_source:
2421 err = -EINVAL;
2422martian_source_keep_err:
2423 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2424 goto out;
2425}
2426
2427int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2428 u8 tos, struct net_device *dev, bool noref)
2429{
2430 struct rtable *rth;
2431 unsigned int hash;
2432 int iif = dev->ifindex;
2433 struct net *net;
2434 int res;
2435
2436 net = dev_net(dev);
2437
2438 rcu_read_lock();
2439
2440 if (!rt_caching(net))
2441 goto skip_cache;
2442
2443 tos &= IPTOS_RT_MASK;
2444 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2445
2446 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2447 rth = rcu_dereference(rth->dst.rt_next)) {
2448 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2449 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2450 (rth->rt_route_iif ^ iif) |
2451 (rth->rt_key_tos ^ tos)) == 0 &&
2452 rth->rt_mark == skb->mark &&
2453 net_eq(dev_net(rth->dst.dev), net) &&
2454 !rt_is_expired(rth)) {
2455 ipv4_validate_peer(rth);
2456 if (noref) {
2457 dst_use_noref(&rth->dst, jiffies);
2458 skb_dst_set_noref(skb, &rth->dst);
2459 } else {
2460 dst_use(&rth->dst, jiffies);
2461 skb_dst_set(skb, &rth->dst);
2462 }
2463 RT_CACHE_STAT_INC(in_hit);
2464 rcu_read_unlock();
2465 return 0;
2466 }
2467 RT_CACHE_STAT_INC(in_hlist_search);
2468 }
2469
2470skip_cache:
2471 /* Multicast recognition logic is moved from route cache to here.
2472 The problem was that too many Ethernet cards have broken/missing
2473 hardware multicast filters :-( As result the host on multicasting
2474 network acquires a lot of useless route cache entries, sort of
2475 SDR messages from all the world. Now we try to get rid of them.
2476 Really, provided software IP multicast filter is organized
2477 reasonably (at least, hashed), it does not result in a slowdown
2478 comparing with route cache reject entries.
2479 Note, that multicast routers are not affected, because
2480 route cache entry is created eventually.
2481 */
2482 if (ipv4_is_multicast(daddr)) {
2483 struct in_device *in_dev = __in_dev_get_rcu(dev);
2484
2485 if (in_dev) {
2486 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2487 ip_hdr(skb)->protocol);
2488 if (our
2489#ifdef CONFIG_IP_MROUTE
2490 ||
2491 (!ipv4_is_local_multicast(daddr) &&
2492 IN_DEV_MFORWARD(in_dev))
2493#endif
2494 ) {
2495 int res = ip_route_input_mc(skb, daddr, saddr,
2496 tos, dev, our);
2497 rcu_read_unlock();
2498 return res;
2499 }
2500 }
2501 rcu_read_unlock();
2502 return -EINVAL;
2503 }
2504 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2505 rcu_read_unlock();
2506 return res;
2507}
2508EXPORT_SYMBOL(ip_route_input_common);
2509
2510/* called with rcu_read_lock() */
2511static struct rtable *__mkroute_output(const struct fib_result *res,
2512 const struct flowi4 *fl4,
2513 __be32 orig_daddr, __be32 orig_saddr,
2514 int orig_oif, __u8 orig_rtos,
2515 struct net_device *dev_out,
2516 unsigned int flags)
2517{
2518 struct fib_info *fi = res->fi;
2519 struct in_device *in_dev;
2520 u16 type = res->type;
2521 struct rtable *rth;
2522
2523 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2524 return ERR_PTR(-EINVAL);
2525
2526 if (ipv4_is_lbcast(fl4->daddr))
2527 type = RTN_BROADCAST;
2528 else if (ipv4_is_multicast(fl4->daddr))
2529 type = RTN_MULTICAST;
2530 else if (ipv4_is_zeronet(fl4->daddr))
2531 return ERR_PTR(-EINVAL);
2532
2533 if (dev_out->flags & IFF_LOOPBACK)
2534 flags |= RTCF_LOCAL;
2535
2536 in_dev = __in_dev_get_rcu(dev_out);
2537 if (!in_dev)
2538 return ERR_PTR(-EINVAL);
2539
2540 if (type == RTN_BROADCAST) {
2541 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2542 fi = NULL;
2543 } else if (type == RTN_MULTICAST) {
2544 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2545 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2546 fl4->flowi4_proto))
2547 flags &= ~RTCF_LOCAL;
2548 /* If multicast route do not exist use
2549 * default one, but do not gateway in this case.
2550 * Yes, it is hack.
2551 */
2552 if (fi && res->prefixlen < 4)
2553 fi = NULL;
2554 }
2555
2556 rth = rt_dst_alloc(dev_out,
2557 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2558 IN_DEV_CONF_GET(in_dev, NOXFRM));
2559 if (!rth)
2560 return ERR_PTR(-ENOBUFS);
2561
2562 rth->dst.output = ip_output;
2563
2564 rth->rt_key_dst = orig_daddr;
2565 rth->rt_key_src = orig_saddr;
2566 rth->rt_genid = rt_genid(dev_net(dev_out));
2567 rth->rt_flags = flags;
2568 rth->rt_type = type;
2569 rth->rt_key_tos = orig_rtos;
2570 rth->rt_dst = fl4->daddr;
2571 rth->rt_src = fl4->saddr;
2572 rth->rt_route_iif = 0;
2573 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2574 rth->rt_oif = orig_oif;
2575 rth->rt_mark = fl4->flowi4_mark;
2576 rth->rt_gateway = fl4->daddr;
2577 rth->rt_spec_dst= fl4->saddr;
2578 rth->rt_peer_genid = 0;
2579 rth->peer = NULL;
2580 rth->fi = NULL;
2581
2582 RT_CACHE_STAT_INC(out_slow_tot);
2583
2584 if (flags & RTCF_LOCAL) {
2585 rth->dst.input = ip_local_deliver;
2586 rth->rt_spec_dst = fl4->daddr;
2587 }
2588 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2589 rth->rt_spec_dst = fl4->saddr;
2590 if (flags & RTCF_LOCAL &&
2591 !(dev_out->flags & IFF_LOOPBACK)) {
2592 rth->dst.output = ip_mc_output;
2593 RT_CACHE_STAT_INC(out_slow_mc);
2594 }
2595#ifdef CONFIG_IP_MROUTE
2596 if (type == RTN_MULTICAST) {
2597 if (IN_DEV_MFORWARD(in_dev) &&
2598 !ipv4_is_local_multicast(fl4->daddr)) {
2599 rth->dst.input = ip_mr_input;
2600 rth->dst.output = ip_mc_output;
2601 }
2602 }
2603#endif
2604 }
2605
2606 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2607
2608 return rth;
2609}
2610
2611/*
2612 * Major route resolver routine.
2613 * called with rcu_read_lock();
2614 */
2615
2616static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2617{
2618 struct net_device *dev_out = NULL;
2619 __u8 tos = RT_FL_TOS(fl4);
2620 unsigned int flags = 0;
2621 struct fib_result res;
2622 struct rtable *rth;
2623 __be32 orig_daddr;
2624 __be32 orig_saddr;
2625 int orig_oif;
2626
2627 res.fi = NULL;
2628#ifdef CONFIG_IP_MULTIPLE_TABLES
2629 res.r = NULL;
2630#endif
2631
2632 orig_daddr = fl4->daddr;
2633 orig_saddr = fl4->saddr;
2634 orig_oif = fl4->flowi4_oif;
2635
2636 fl4->flowi4_iif = net->loopback_dev->ifindex;
2637 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2638 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2639 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2640
2641 rcu_read_lock();
2642 if (fl4->saddr) {
2643 rth = ERR_PTR(-EINVAL);
2644 if (ipv4_is_multicast(fl4->saddr) ||
2645 ipv4_is_lbcast(fl4->saddr) ||
2646 ipv4_is_zeronet(fl4->saddr))
2647 goto out;
2648
2649 /* I removed check for oif == dev_out->oif here.
2650 It was wrong for two reasons:
2651 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2652 is assigned to multiple interfaces.
2653 2. Moreover, we are allowed to send packets with saddr
2654 of another iface. --ANK
2655 */
2656
2657 if (fl4->flowi4_oif == 0 &&
2658 (ipv4_is_multicast(fl4->daddr) ||
2659 ipv4_is_lbcast(fl4->daddr))) {
2660 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2661 dev_out = __ip_dev_find(net, fl4->saddr, false);
2662 if (dev_out == NULL)
2663 goto out;
2664
2665 /* Special hack: user can direct multicasts
2666 and limited broadcast via necessary interface
2667 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2668 This hack is not just for fun, it allows
2669 vic,vat and friends to work.
2670 They bind socket to loopback, set ttl to zero
2671 and expect that it will work.
2672 From the viewpoint of routing cache they are broken,
2673 because we are not allowed to build multicast path
2674 with loopback source addr (look, routing cache
2675 cannot know, that ttl is zero, so that packet
2676 will not leave this host and route is valid).
2677 Luckily, this hack is good workaround.
2678 */
2679
2680 fl4->flowi4_oif = dev_out->ifindex;
2681 goto make_route;
2682 }
2683
2684 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2685 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2686 if (!__ip_dev_find(net, fl4->saddr, false))
2687 goto out;
2688 }
2689 }
2690
2691
2692 if (fl4->flowi4_oif) {
2693 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2694 rth = ERR_PTR(-ENODEV);
2695 if (dev_out == NULL)
2696 goto out;
2697
2698 /* RACE: Check return value of inet_select_addr instead. */
2699 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2700 rth = ERR_PTR(-ENETUNREACH);
2701 goto out;
2702 }
2703 if (ipv4_is_local_multicast(fl4->daddr) ||
2704 ipv4_is_lbcast(fl4->daddr)) {
2705 if (!fl4->saddr)
2706 fl4->saddr = inet_select_addr(dev_out, 0,
2707 RT_SCOPE_LINK);
2708 goto make_route;
2709 }
2710 if (fl4->saddr) {
2711 if (ipv4_is_multicast(fl4->daddr))
2712 fl4->saddr = inet_select_addr(dev_out, 0,
2713 fl4->flowi4_scope);
2714 else if (!fl4->daddr)
2715 fl4->saddr = inet_select_addr(dev_out, 0,
2716 RT_SCOPE_HOST);
2717 }
2718 }
2719
2720 if (!fl4->daddr) {
2721 fl4->daddr = fl4->saddr;
2722 if (!fl4->daddr)
2723 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2724 dev_out = net->loopback_dev;
2725 fl4->flowi4_oif = net->loopback_dev->ifindex;
2726 res.type = RTN_LOCAL;
2727 flags |= RTCF_LOCAL;
2728 goto make_route;
2729 }
2730
2731 if (fib_lookup(net, fl4, &res)) {
2732 res.fi = NULL;
2733 if (fl4->flowi4_oif) {
2734 /* Apparently, routing tables are wrong. Assume,
2735 that the destination is on link.
2736
2737 WHY? DW.
2738 Because we are allowed to send to iface
2739 even if it has NO routes and NO assigned
2740 addresses. When oif is specified, routing
2741 tables are looked up with only one purpose:
2742 to catch if destination is gatewayed, rather than
2743 direct. Moreover, if MSG_DONTROUTE is set,
2744 we send packet, ignoring both routing tables
2745 and ifaddr state. --ANK
2746
2747
2748 We could make it even if oif is unknown,
2749 likely IPv6, but we do not.
2750 */
2751
2752 if (fl4->saddr == 0)
2753 fl4->saddr = inet_select_addr(dev_out, 0,
2754 RT_SCOPE_LINK);
2755 res.type = RTN_UNICAST;
2756 goto make_route;
2757 }
2758 rth = ERR_PTR(-ENETUNREACH);
2759 goto out;
2760 }
2761
2762 if (res.type == RTN_LOCAL) {
2763 if (!fl4->saddr) {
2764 if (res.fi->fib_prefsrc)
2765 fl4->saddr = res.fi->fib_prefsrc;
2766 else
2767 fl4->saddr = fl4->daddr;
2768 }
2769 dev_out = net->loopback_dev;
2770 fl4->flowi4_oif = dev_out->ifindex;
2771 res.fi = NULL;
2772 flags |= RTCF_LOCAL;
2773 goto make_route;
2774 }
2775
2776#ifdef CONFIG_IP_ROUTE_MULTIPATH
2777 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2778 fib_select_multipath(&res);
2779 else
2780#endif
2781 if (!res.prefixlen &&
2782 res.table->tb_num_default > 1 &&
2783 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2784 fib_select_default(&res);
2785
2786 if (!fl4->saddr)
2787 fl4->saddr = FIB_RES_PREFSRC(net, res);
2788
2789 dev_out = FIB_RES_DEV(res);
2790 fl4->flowi4_oif = dev_out->ifindex;
2791
2792
2793make_route:
2794 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2795 tos, dev_out, flags);
2796 if (!IS_ERR(rth)) {
2797 unsigned int hash;
2798
2799 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2800 rt_genid(dev_net(dev_out)));
2801 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2802 }
2803
2804out:
2805 rcu_read_unlock();
2806 return rth;
2807}
2808
2809struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2810{
2811 struct rtable *rth;
2812 unsigned int hash;
2813
2814 if (!rt_caching(net))
2815 goto slow_output;
2816
2817 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2818
2819 rcu_read_lock_bh();
2820 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2821 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2822 if (rth->rt_key_dst == flp4->daddr &&
2823 rth->rt_key_src == flp4->saddr &&
2824 rt_is_output_route(rth) &&
2825 rth->rt_oif == flp4->flowi4_oif &&
2826 rth->rt_mark == flp4->flowi4_mark &&
2827 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2828 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2829 net_eq(dev_net(rth->dst.dev), net) &&
2830 !rt_is_expired(rth)) {
2831 ipv4_validate_peer(rth);
2832 dst_use(&rth->dst, jiffies);
2833 RT_CACHE_STAT_INC(out_hit);
2834 rcu_read_unlock_bh();
2835 if (!flp4->saddr)
2836 flp4->saddr = rth->rt_src;
2837 if (!flp4->daddr)
2838 flp4->daddr = rth->rt_dst;
2839 return rth;
2840 }
2841 RT_CACHE_STAT_INC(out_hlist_search);
2842 }
2843 rcu_read_unlock_bh();
2844
2845slow_output:
2846 return ip_route_output_slow(net, flp4);
2847}
2848EXPORT_SYMBOL_GPL(__ip_route_output_key);
2849
2850static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2851{
2852 return NULL;
2853}
2854
2855static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2856{
2857 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2858
2859 return mtu ? : dst->dev->mtu;
2860}
2861
2862static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2863{
2864}
2865
2866static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2867 unsigned long old)
2868{
2869 return NULL;
2870}
2871
2872static struct dst_ops ipv4_dst_blackhole_ops = {
2873 .family = AF_INET,
2874 .protocol = cpu_to_be16(ETH_P_IP),
2875 .destroy = ipv4_dst_destroy,
2876 .check = ipv4_blackhole_dst_check,
2877 .mtu = ipv4_blackhole_mtu,
2878 .default_advmss = ipv4_default_advmss,
2879 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2880 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2881 .neigh_lookup = ipv4_neigh_lookup,
2882};
2883
2884struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2885{
2886 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2887 struct rtable *ort = (struct rtable *) dst_orig;
2888
2889 if (rt) {
2890 struct dst_entry *new = &rt->dst;
2891
2892 new->__use = 1;
2893 new->input = dst_discard;
2894 new->output = dst_discard;
2895 dst_copy_metrics(new, &ort->dst);
2896
2897 new->dev = ort->dst.dev;
2898 if (new->dev)
2899 dev_hold(new->dev);
2900
2901 rt->rt_key_dst = ort->rt_key_dst;
2902 rt->rt_key_src = ort->rt_key_src;
2903 rt->rt_key_tos = ort->rt_key_tos;
2904 rt->rt_route_iif = ort->rt_route_iif;
2905 rt->rt_iif = ort->rt_iif;
2906 rt->rt_oif = ort->rt_oif;
2907 rt->rt_mark = ort->rt_mark;
2908
2909 rt->rt_genid = rt_genid(net);
2910 rt->rt_flags = ort->rt_flags;
2911 rt->rt_type = ort->rt_type;
2912 rt->rt_dst = ort->rt_dst;
2913 rt->rt_src = ort->rt_src;
2914 rt->rt_gateway = ort->rt_gateway;
2915 rt->rt_spec_dst = ort->rt_spec_dst;
2916 rt->peer = ort->peer;
2917 if (rt->peer)
2918 atomic_inc(&rt->peer->refcnt);
2919 rt->fi = ort->fi;
2920 if (rt->fi)
2921 atomic_inc(&rt->fi->fib_clntref);
2922
2923 dst_free(new);
2924 }
2925
2926 dst_release(dst_orig);
2927
2928 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2929}
2930
2931struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2932 struct sock *sk)
2933{
2934 struct rtable *rt = __ip_route_output_key(net, flp4);
2935
2936 if (IS_ERR(rt))
2937 return rt;
2938
2939 if (flp4->flowi4_proto)
2940 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2941 flowi4_to_flowi(flp4),
2942 sk, 0);
2943
2944 return rt;
2945}
2946EXPORT_SYMBOL_GPL(ip_route_output_flow);
2947
2948static int rt_fill_info(struct net *net,
2949 struct sk_buff *skb, u32 pid, u32 seq, int event,
2950 int nowait, unsigned int flags)
2951{
2952 struct rtable *rt = skb_rtable(skb);
2953 struct rtmsg *r;
2954 struct nlmsghdr *nlh;
2955 unsigned long expires = 0;
2956 const struct inet_peer *peer = rt->peer;
2957 u32 id = 0, ts = 0, tsage = 0, error;
2958
2959 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2960 if (nlh == NULL)
2961 return -EMSGSIZE;
2962
2963 r = nlmsg_data(nlh);
2964 r->rtm_family = AF_INET;
2965 r->rtm_dst_len = 32;
2966 r->rtm_src_len = 0;
2967 r->rtm_tos = rt->rt_key_tos;
2968 r->rtm_table = RT_TABLE_MAIN;
2969 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2970 goto nla_put_failure;
2971 r->rtm_type = rt->rt_type;
2972 r->rtm_scope = RT_SCOPE_UNIVERSE;
2973 r->rtm_protocol = RTPROT_UNSPEC;
2974 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2975 if (rt->rt_flags & RTCF_NOTIFY)
2976 r->rtm_flags |= RTM_F_NOTIFY;
2977
2978 if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2979 goto nla_put_failure;
2980 if (rt->rt_key_src) {
2981 r->rtm_src_len = 32;
2982 if (nla_put_be32(skb, RTA_SRC, rt->rt_key_src))
2983 goto nla_put_failure;
2984 }
2985 if (rt->dst.dev &&
2986 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2987 goto nla_put_failure;
2988#ifdef CONFIG_IP_ROUTE_CLASSID
2989 if (rt->dst.tclassid &&
2990 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2991 goto nla_put_failure;
2992#endif
2993 if (rt_is_input_route(rt)) {
2994 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_spec_dst))
2995 goto nla_put_failure;
2996 } else if (rt->rt_src != rt->rt_key_src) {
2997 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2998 goto nla_put_failure;
2999 }
3000 if (rt->rt_dst != rt->rt_gateway &&
3001 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
3002 goto nla_put_failure;
3003
3004 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
3005 goto nla_put_failure;
3006
3007 if (rt->rt_mark &&
3008 nla_put_be32(skb, RTA_MARK, rt->rt_mark))
3009 goto nla_put_failure;
3010
3011 error = rt->dst.error;
3012 if (peer) {
3013 inet_peer_refcheck(rt->peer);
3014 id = atomic_read(&peer->ip_id_count) & 0xffff;
3015 if (peer->tcp_ts_stamp) {
3016 ts = peer->tcp_ts;
3017 tsage = get_seconds() - peer->tcp_ts_stamp;
3018 }
3019 expires = ACCESS_ONCE(peer->pmtu_expires);
3020 if (expires) {
3021 if (time_before(jiffies, expires))
3022 expires -= jiffies;
3023 else
3024 expires = 0;
3025 }
3026 }
3027
3028 if (rt_is_input_route(rt)) {
3029#ifdef CONFIG_IP_MROUTE
3030 __be32 dst = rt->rt_dst;
3031
3032 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
3033 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3034 int err = ipmr_get_route(net, skb,
3035 rt->rt_src, rt->rt_dst,
3036 r, nowait);
3037 if (err <= 0) {
3038 if (!nowait) {
3039 if (err == 0)
3040 return 0;
3041 goto nla_put_failure;
3042 } else {
3043 if (err == -EMSGSIZE)
3044 goto nla_put_failure;
3045 error = err;
3046 }
3047 }
3048 } else
3049#endif
3050 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
3051 goto nla_put_failure;
3052 }
3053
3054 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
3055 expires, error) < 0)
3056 goto nla_put_failure;
3057
3058 return nlmsg_end(skb, nlh);
3059
3060nla_put_failure:
3061 nlmsg_cancel(skb, nlh);
3062 return -EMSGSIZE;
3063}
3064
3065static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
3066{
3067 struct net *net = sock_net(in_skb->sk);
3068 struct rtmsg *rtm;
3069 struct nlattr *tb[RTA_MAX+1];
3070 struct rtable *rt = NULL;
3071 __be32 dst = 0;
3072 __be32 src = 0;
3073 u32 iif;
3074 int err;
3075 int mark;
3076 struct sk_buff *skb;
3077
3078 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
3079 if (err < 0)
3080 goto errout;
3081
3082 rtm = nlmsg_data(nlh);
3083
3084 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3085 if (skb == NULL) {
3086 err = -ENOBUFS;
3087 goto errout;
3088 }
3089
3090 /* Reserve room for dummy headers, this skb can pass
3091 through good chunk of routing engine.
3092 */
3093 skb_reset_mac_header(skb);
3094 skb_reset_network_header(skb);
3095
3096 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3097 ip_hdr(skb)->protocol = IPPROTO_ICMP;
3098 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
3099
3100 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
3101 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
3102 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3103 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3104
3105 if (iif) {
3106 struct net_device *dev;
3107
3108 dev = __dev_get_by_index(net, iif);
3109 if (dev == NULL) {
3110 err = -ENODEV;
3111 goto errout_free;
3112 }
3113
3114 skb->protocol = htons(ETH_P_IP);
3115 skb->dev = dev;
3116 skb->mark = mark;
3117 local_bh_disable();
3118 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3119 local_bh_enable();
3120
3121 rt = skb_rtable(skb);
3122 if (err == 0 && rt->dst.error)
3123 err = -rt->dst.error;
3124 } else {
3125 struct flowi4 fl4 = {
3126 .daddr = dst,
3127 .saddr = src,
3128 .flowi4_tos = rtm->rtm_tos,
3129 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3130 .flowi4_mark = mark,
3131 };
3132 rt = ip_route_output_key(net, &fl4);
3133
3134 err = 0;
3135 if (IS_ERR(rt))
3136 err = PTR_ERR(rt);
3137 }
3138
3139 if (err)
3140 goto errout_free;
3141
3142 skb_dst_set(skb, &rt->dst);
3143 if (rtm->rtm_flags & RTM_F_NOTIFY)
3144 rt->rt_flags |= RTCF_NOTIFY;
3145
3146 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3147 RTM_NEWROUTE, 0, 0);
3148 if (err <= 0)
3149 goto errout_free;
3150
3151 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3152errout:
3153 return err;
3154
3155errout_free:
3156 kfree_skb(skb);
3157 goto errout;
3158}
3159
3160int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3161{
3162 struct rtable *rt;
3163 int h, s_h;
3164 int idx, s_idx;
3165 struct net *net;
3166
3167 net = sock_net(skb->sk);
3168
3169 s_h = cb->args[0];
3170 if (s_h < 0)
3171 s_h = 0;
3172 s_idx = idx = cb->args[1];
3173 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3174 if (!rt_hash_table[h].chain)
3175 continue;
3176 rcu_read_lock_bh();
3177 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3178 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3179 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3180 continue;
3181 if (rt_is_expired(rt))
3182 continue;
3183 skb_dst_set_noref(skb, &rt->dst);
3184 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3185 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3186 1, NLM_F_MULTI) <= 0) {
3187 skb_dst_drop(skb);
3188 rcu_read_unlock_bh();
3189 goto done;
3190 }
3191 skb_dst_drop(skb);
3192 }
3193 rcu_read_unlock_bh();
3194 }
3195
3196done:
3197 cb->args[0] = h;
3198 cb->args[1] = idx;
3199 return skb->len;
3200}
3201
3202void ip_rt_multicast_event(struct in_device *in_dev)
3203{
3204 rt_cache_flush(dev_net(in_dev->dev), 0);
3205}
3206
3207#ifdef CONFIG_SYSCTL
3208static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3209 void __user *buffer,
3210 size_t *lenp, loff_t *ppos)
3211{
3212 if (write) {
3213 int flush_delay;
3214 ctl_table ctl;
3215 struct net *net;
3216
3217 memcpy(&ctl, __ctl, sizeof(ctl));
3218 ctl.data = &flush_delay;
3219 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3220
3221 net = (struct net *)__ctl->extra1;
3222 rt_cache_flush(net, flush_delay);
3223 return 0;
3224 }
3225
3226 return -EINVAL;
3227}
3228
3229static ctl_table ipv4_route_table[] = {
3230 {
3231 .procname = "gc_thresh",
3232 .data = &ipv4_dst_ops.gc_thresh,
3233 .maxlen = sizeof(int),
3234 .mode = 0644,
3235 .proc_handler = proc_dointvec,
3236 },
3237 {
3238 .procname = "max_size",
3239 .data = &ip_rt_max_size,
3240 .maxlen = sizeof(int),
3241 .mode = 0644,
3242 .proc_handler = proc_dointvec,
3243 },
3244 {
3245 /* Deprecated. Use gc_min_interval_ms */
3246
3247 .procname = "gc_min_interval",
3248 .data = &ip_rt_gc_min_interval,
3249 .maxlen = sizeof(int),
3250 .mode = 0644,
3251 .proc_handler = proc_dointvec_jiffies,
3252 },
3253 {
3254 .procname = "gc_min_interval_ms",
3255 .data = &ip_rt_gc_min_interval,
3256 .maxlen = sizeof(int),
3257 .mode = 0644,
3258 .proc_handler = proc_dointvec_ms_jiffies,
3259 },
3260 {
3261 .procname = "gc_timeout",
3262 .data = &ip_rt_gc_timeout,
3263 .maxlen = sizeof(int),
3264 .mode = 0644,
3265 .proc_handler = proc_dointvec_jiffies,
3266 },
3267 {
3268 .procname = "gc_interval",
3269 .data = &ip_rt_gc_interval,
3270 .maxlen = sizeof(int),
3271 .mode = 0644,
3272 .proc_handler = proc_dointvec_jiffies,
3273 },
3274 {
3275 .procname = "redirect_load",
3276 .data = &ip_rt_redirect_load,
3277 .maxlen = sizeof(int),
3278 .mode = 0644,
3279 .proc_handler = proc_dointvec,
3280 },
3281 {
3282 .procname = "redirect_number",
3283 .data = &ip_rt_redirect_number,
3284 .maxlen = sizeof(int),
3285 .mode = 0644,
3286 .proc_handler = proc_dointvec,
3287 },
3288 {
3289 .procname = "redirect_silence",
3290 .data = &ip_rt_redirect_silence,
3291 .maxlen = sizeof(int),
3292 .mode = 0644,
3293 .proc_handler = proc_dointvec,
3294 },
3295 {
3296 .procname = "error_cost",
3297 .data = &ip_rt_error_cost,
3298 .maxlen = sizeof(int),
3299 .mode = 0644,
3300 .proc_handler = proc_dointvec,
3301 },
3302 {
3303 .procname = "error_burst",
3304 .data = &ip_rt_error_burst,
3305 .maxlen = sizeof(int),
3306 .mode = 0644,
3307 .proc_handler = proc_dointvec,
3308 },
3309 {
3310 .procname = "gc_elasticity",
3311 .data = &ip_rt_gc_elasticity,
3312 .maxlen = sizeof(int),
3313 .mode = 0644,
3314 .proc_handler = proc_dointvec,
3315 },
3316 {
3317 .procname = "mtu_expires",
3318 .data = &ip_rt_mtu_expires,
3319 .maxlen = sizeof(int),
3320 .mode = 0644,
3321 .proc_handler = proc_dointvec_jiffies,
3322 },
3323 {
3324 .procname = "min_pmtu",
3325 .data = &ip_rt_min_pmtu,
3326 .maxlen = sizeof(int),
3327 .mode = 0644,
3328 .proc_handler = proc_dointvec,
3329 },
3330 {
3331 .procname = "min_adv_mss",
3332 .data = &ip_rt_min_advmss,
3333 .maxlen = sizeof(int),
3334 .mode = 0644,
3335 .proc_handler = proc_dointvec,
3336 },
3337 { }
3338};
3339
3340static struct ctl_table ipv4_route_flush_table[] = {
3341 {
3342 .procname = "flush",
3343 .maxlen = sizeof(int),
3344 .mode = 0200,
3345 .proc_handler = ipv4_sysctl_rtcache_flush,
3346 },
3347 { },
3348};
3349
3350static __net_init int sysctl_route_net_init(struct net *net)
3351{
3352 struct ctl_table *tbl;
3353
3354 tbl = ipv4_route_flush_table;
3355 if (!net_eq(net, &init_net)) {
3356 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3357 if (tbl == NULL)
3358 goto err_dup;
3359 }
3360 tbl[0].extra1 = net;
3361
3362 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3363 if (net->ipv4.route_hdr == NULL)
3364 goto err_reg;
3365 return 0;
3366
3367err_reg:
3368 if (tbl != ipv4_route_flush_table)
3369 kfree(tbl);
3370err_dup:
3371 return -ENOMEM;
3372}
3373
3374static __net_exit void sysctl_route_net_exit(struct net *net)
3375{
3376 struct ctl_table *tbl;
3377
3378 tbl = net->ipv4.route_hdr->ctl_table_arg;
3379 unregister_net_sysctl_table(net->ipv4.route_hdr);
3380 BUG_ON(tbl == ipv4_route_flush_table);
3381 kfree(tbl);
3382}
3383
3384static __net_initdata struct pernet_operations sysctl_route_ops = {
3385 .init = sysctl_route_net_init,
3386 .exit = sysctl_route_net_exit,
3387};
3388#endif
3389
3390static __net_init int rt_genid_init(struct net *net)
3391{
3392 get_random_bytes(&net->ipv4.rt_genid,
3393 sizeof(net->ipv4.rt_genid));
3394 get_random_bytes(&net->ipv4.dev_addr_genid,
3395 sizeof(net->ipv4.dev_addr_genid));
3396 return 0;
3397}
3398
3399static __net_initdata struct pernet_operations rt_genid_ops = {
3400 .init = rt_genid_init,
3401};
3402
3403
3404#ifdef CONFIG_IP_ROUTE_CLASSID
3405struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3406#endif /* CONFIG_IP_ROUTE_CLASSID */
3407
3408static __initdata unsigned long rhash_entries;
3409static int __init set_rhash_entries(char *str)
3410{
3411 ssize_t ret;
3412
3413 if (!str)
3414 return 0;
3415
3416 ret = kstrtoul(str, 0, &rhash_entries);
3417 if (ret)
3418 return 0;
3419
3420 return 1;
3421}
3422__setup("rhash_entries=", set_rhash_entries);
3423
3424int __init ip_rt_init(void)
3425{
3426 int rc = 0;
3427
3428#ifdef CONFIG_IP_ROUTE_CLASSID
3429 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3430 if (!ip_rt_acct)
3431 panic("IP: failed to allocate ip_rt_acct\n");
3432#endif
3433
3434 ipv4_dst_ops.kmem_cachep =
3435 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3436 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3437
3438 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3439
3440 if (dst_entries_init(&ipv4_dst_ops) < 0)
3441 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3442
3443 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3444 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3445
3446 rt_hash_table = (struct rt_hash_bucket *)
3447 alloc_large_system_hash("IP route cache",
3448 sizeof(struct rt_hash_bucket),
3449 rhash_entries,
3450 (totalram_pages >= 128 * 1024) ?
3451 15 : 17,
3452 0,
3453 &rt_hash_log,
3454 &rt_hash_mask,
3455 0,
3456 rhash_entries ? 0 : 512 * 1024);
3457 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3458 rt_hash_lock_init();
3459
3460 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3461 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3462
3463 devinet_init();
3464 ip_fib_init();
3465
3466 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3467 expires_ljiffies = jiffies;
3468 schedule_delayed_work(&expires_work,
3469 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3470
3471 if (ip_rt_proc_init())
3472 pr_err("Unable to create route proc files\n");
3473#ifdef CONFIG_XFRM
3474 xfrm_init();
3475 xfrm4_init(ip_rt_max_size);
3476#endif
3477 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3478
3479#ifdef CONFIG_SYSCTL
3480 register_pernet_subsys(&sysctl_route_ops);
3481#endif
3482 register_pernet_subsys(&rt_genid_ops);
3483 return rc;
3484}
3485
3486#ifdef CONFIG_SYSCTL
3487/*
3488 * We really need to sanitize the damn ipv4 init order, then all
3489 * this nonsense will go away.
3490 */
3491void __init ip_static_sysctl_init(void)
3492{
3493 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3494}
3495#endif
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * ROUTE - implementation of the IP router.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
13 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
14 *
15 * Fixes:
16 * Alan Cox : Verify area fixes.
17 * Alan Cox : cli() protects routing changes
18 * Rui Oliveira : ICMP routing table updates
19 * (rco@di.uminho.pt) Routing table insertion and update
20 * Linus Torvalds : Rewrote bits to be sensible
21 * Alan Cox : Added BSD route gw semantics
22 * Alan Cox : Super /proc >4K
23 * Alan Cox : MTU in route table
24 * Alan Cox : MSS actually. Also added the window
25 * clamper.
26 * Sam Lantinga : Fixed route matching in rt_del()
27 * Alan Cox : Routing cache support.
28 * Alan Cox : Removed compatibility cruft.
29 * Alan Cox : RTF_REJECT support.
30 * Alan Cox : TCP irtt support.
31 * Jonathan Naylor : Added Metric support.
32 * Miquel van Smoorenburg : BSD API fixes.
33 * Miquel van Smoorenburg : Metrics.
34 * Alan Cox : Use __u32 properly
35 * Alan Cox : Aligned routing errors more closely with BSD
36 * our system is still very different.
37 * Alan Cox : Faster /proc handling
38 * Alexey Kuznetsov : Massive rework to support tree based routing,
39 * routing caches and better behaviour.
40 *
41 * Olaf Erb : irtt wasn't being copied right.
42 * Bjorn Ekwall : Kerneld route support.
43 * Alan Cox : Multicast fixed (I hope)
44 * Pavel Krauz : Limited broadcast fixed
45 * Mike McLagan : Routing by source
46 * Alexey Kuznetsov : End of old history. Split to fib.c and
47 * route.c and rewritten from scratch.
48 * Andi Kleen : Load-limit warning messages.
49 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
50 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
51 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
52 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
53 * Marc Boucher : routing by fwmark
54 * Robert Olsson : Added rt_cache statistics
55 * Arnaldo C. Melo : Convert proc stuff to seq_file
56 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
57 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
58 * Ilia Sotnikov : Removed TOS from hash calculations
59 */
60
61#define pr_fmt(fmt) "IPv4: " fmt
62
63#include <linux/module.h>
64#include <linux/bitops.h>
65#include <linux/kernel.h>
66#include <linux/mm.h>
67#include <linux/memblock.h>
68#include <linux/socket.h>
69#include <linux/errno.h>
70#include <linux/in.h>
71#include <linux/inet.h>
72#include <linux/netdevice.h>
73#include <linux/proc_fs.h>
74#include <linux/init.h>
75#include <linux/skbuff.h>
76#include <linux/inetdevice.h>
77#include <linux/igmp.h>
78#include <linux/pkt_sched.h>
79#include <linux/mroute.h>
80#include <linux/netfilter_ipv4.h>
81#include <linux/random.h>
82#include <linux/rcupdate.h>
83#include <linux/slab.h>
84#include <linux/jhash.h>
85#include <net/dst.h>
86#include <net/dst_metadata.h>
87#include <net/inet_dscp.h>
88#include <net/net_namespace.h>
89#include <net/ip.h>
90#include <net/route.h>
91#include <net/inetpeer.h>
92#include <net/sock.h>
93#include <net/ip_fib.h>
94#include <net/nexthop.h>
95#include <net/tcp.h>
96#include <net/icmp.h>
97#include <net/xfrm.h>
98#include <net/lwtunnel.h>
99#include <net/netevent.h>
100#include <net/rtnetlink.h>
101#ifdef CONFIG_SYSCTL
102#include <linux/sysctl.h>
103#endif
104#include <net/secure_seq.h>
105#include <net/ip_tunnels.h>
106
107#include "fib_lookup.h"
108
109#define RT_FL_TOS(oldflp4) \
110 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
111
112#define RT_GC_TIMEOUT (300*HZ)
113
114#define DEFAULT_MIN_PMTU (512 + 20 + 20)
115#define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
116#define DEFAULT_MIN_ADVMSS 256
117static int ip_rt_max_size;
118static int ip_rt_redirect_number __read_mostly = 9;
119static int ip_rt_redirect_load __read_mostly = HZ / 50;
120static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
121static int ip_rt_error_cost __read_mostly = HZ;
122static int ip_rt_error_burst __read_mostly = 5 * HZ;
123
124static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
125
126/*
127 * Interface to generic destination cache.
128 */
129
130INDIRECT_CALLABLE_SCOPE
131struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
132static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
133INDIRECT_CALLABLE_SCOPE
134unsigned int ipv4_mtu(const struct dst_entry *dst);
135static void ipv4_negative_advice(struct sock *sk,
136 struct dst_entry *dst);
137static void ipv4_link_failure(struct sk_buff *skb);
138static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
139 struct sk_buff *skb, u32 mtu,
140 bool confirm_neigh);
141static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
142 struct sk_buff *skb);
143static void ipv4_dst_destroy(struct dst_entry *dst);
144
145static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
146{
147 WARN_ON(1);
148 return NULL;
149}
150
151static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
152 struct sk_buff *skb,
153 const void *daddr);
154static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
155
156static struct dst_ops ipv4_dst_ops = {
157 .family = AF_INET,
158 .check = ipv4_dst_check,
159 .default_advmss = ipv4_default_advmss,
160 .mtu = ipv4_mtu,
161 .cow_metrics = ipv4_cow_metrics,
162 .destroy = ipv4_dst_destroy,
163 .negative_advice = ipv4_negative_advice,
164 .link_failure = ipv4_link_failure,
165 .update_pmtu = ip_rt_update_pmtu,
166 .redirect = ip_do_redirect,
167 .local_out = __ip_local_out,
168 .neigh_lookup = ipv4_neigh_lookup,
169 .confirm_neigh = ipv4_confirm_neigh,
170};
171
172#define ECN_OR_COST(class) TC_PRIO_##class
173
174const __u8 ip_tos2prio[16] = {
175 TC_PRIO_BESTEFFORT,
176 ECN_OR_COST(BESTEFFORT),
177 TC_PRIO_BESTEFFORT,
178 ECN_OR_COST(BESTEFFORT),
179 TC_PRIO_BULK,
180 ECN_OR_COST(BULK),
181 TC_PRIO_BULK,
182 ECN_OR_COST(BULK),
183 TC_PRIO_INTERACTIVE,
184 ECN_OR_COST(INTERACTIVE),
185 TC_PRIO_INTERACTIVE,
186 ECN_OR_COST(INTERACTIVE),
187 TC_PRIO_INTERACTIVE_BULK,
188 ECN_OR_COST(INTERACTIVE_BULK),
189 TC_PRIO_INTERACTIVE_BULK,
190 ECN_OR_COST(INTERACTIVE_BULK)
191};
192EXPORT_SYMBOL(ip_tos2prio);
193
194static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
195#define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
196
197#ifdef CONFIG_PROC_FS
198static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
199{
200 if (*pos)
201 return NULL;
202 return SEQ_START_TOKEN;
203}
204
205static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
206{
207 ++*pos;
208 return NULL;
209}
210
211static void rt_cache_seq_stop(struct seq_file *seq, void *v)
212{
213}
214
215static int rt_cache_seq_show(struct seq_file *seq, void *v)
216{
217 if (v == SEQ_START_TOKEN)
218 seq_printf(seq, "%-127s\n",
219 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
220 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
221 "HHUptod\tSpecDst");
222 return 0;
223}
224
225static const struct seq_operations rt_cache_seq_ops = {
226 .start = rt_cache_seq_start,
227 .next = rt_cache_seq_next,
228 .stop = rt_cache_seq_stop,
229 .show = rt_cache_seq_show,
230};
231
232static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
233{
234 int cpu;
235
236 if (*pos == 0)
237 return SEQ_START_TOKEN;
238
239 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
240 if (!cpu_possible(cpu))
241 continue;
242 *pos = cpu+1;
243 return &per_cpu(rt_cache_stat, cpu);
244 }
245 return NULL;
246}
247
248static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
249{
250 int cpu;
251
252 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
253 if (!cpu_possible(cpu))
254 continue;
255 *pos = cpu+1;
256 return &per_cpu(rt_cache_stat, cpu);
257 }
258 (*pos)++;
259 return NULL;
260
261}
262
263static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
264{
265
266}
267
268static int rt_cpu_seq_show(struct seq_file *seq, void *v)
269{
270 struct rt_cache_stat *st = v;
271
272 if (v == SEQ_START_TOKEN) {
273 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");
274 return 0;
275 }
276
277 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x "
278 "%08x %08x %08x %08x %08x %08x "
279 "%08x %08x %08x %08x\n",
280 dst_entries_get_slow(&ipv4_dst_ops),
281 0, /* st->in_hit */
282 st->in_slow_tot,
283 st->in_slow_mc,
284 st->in_no_route,
285 st->in_brd,
286 st->in_martian_dst,
287 st->in_martian_src,
288
289 0, /* st->out_hit */
290 st->out_slow_tot,
291 st->out_slow_mc,
292
293 0, /* st->gc_total */
294 0, /* st->gc_ignored */
295 0, /* st->gc_goal_miss */
296 0, /* st->gc_dst_overflow */
297 0, /* st->in_hlist_search */
298 0 /* st->out_hlist_search */
299 );
300 return 0;
301}
302
303static const struct seq_operations rt_cpu_seq_ops = {
304 .start = rt_cpu_seq_start,
305 .next = rt_cpu_seq_next,
306 .stop = rt_cpu_seq_stop,
307 .show = rt_cpu_seq_show,
308};
309
310#ifdef CONFIG_IP_ROUTE_CLASSID
311static int rt_acct_proc_show(struct seq_file *m, void *v)
312{
313 struct ip_rt_acct *dst, *src;
314 unsigned int i, j;
315
316 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
317 if (!dst)
318 return -ENOMEM;
319
320 for_each_possible_cpu(i) {
321 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
322 for (j = 0; j < 256; j++) {
323 dst[j].o_bytes += src[j].o_bytes;
324 dst[j].o_packets += src[j].o_packets;
325 dst[j].i_bytes += src[j].i_bytes;
326 dst[j].i_packets += src[j].i_packets;
327 }
328 }
329
330 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
331 kfree(dst);
332 return 0;
333}
334#endif
335
336static int __net_init ip_rt_do_proc_init(struct net *net)
337{
338 struct proc_dir_entry *pde;
339
340 pde = proc_create_seq("rt_cache", 0444, net->proc_net,
341 &rt_cache_seq_ops);
342 if (!pde)
343 goto err1;
344
345 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
346 &rt_cpu_seq_ops);
347 if (!pde)
348 goto err2;
349
350#ifdef CONFIG_IP_ROUTE_CLASSID
351 pde = proc_create_single("rt_acct", 0, net->proc_net,
352 rt_acct_proc_show);
353 if (!pde)
354 goto err3;
355#endif
356 return 0;
357
358#ifdef CONFIG_IP_ROUTE_CLASSID
359err3:
360 remove_proc_entry("rt_cache", net->proc_net_stat);
361#endif
362err2:
363 remove_proc_entry("rt_cache", net->proc_net);
364err1:
365 return -ENOMEM;
366}
367
368static void __net_exit ip_rt_do_proc_exit(struct net *net)
369{
370 remove_proc_entry("rt_cache", net->proc_net_stat);
371 remove_proc_entry("rt_cache", net->proc_net);
372#ifdef CONFIG_IP_ROUTE_CLASSID
373 remove_proc_entry("rt_acct", net->proc_net);
374#endif
375}
376
377static struct pernet_operations ip_rt_proc_ops __net_initdata = {
378 .init = ip_rt_do_proc_init,
379 .exit = ip_rt_do_proc_exit,
380};
381
382static int __init ip_rt_proc_init(void)
383{
384 return register_pernet_subsys(&ip_rt_proc_ops);
385}
386
387#else
388static inline int ip_rt_proc_init(void)
389{
390 return 0;
391}
392#endif /* CONFIG_PROC_FS */
393
394static inline bool rt_is_expired(const struct rtable *rth)
395{
396 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
397}
398
399void rt_cache_flush(struct net *net)
400{
401 rt_genid_bump_ipv4(net);
402}
403
404static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
405 struct sk_buff *skb,
406 const void *daddr)
407{
408 const struct rtable *rt = container_of(dst, struct rtable, dst);
409 struct net_device *dev = dst->dev;
410 struct neighbour *n;
411
412 rcu_read_lock();
413
414 if (likely(rt->rt_gw_family == AF_INET)) {
415 n = ip_neigh_gw4(dev, rt->rt_gw4);
416 } else if (rt->rt_gw_family == AF_INET6) {
417 n = ip_neigh_gw6(dev, &rt->rt_gw6);
418 } else {
419 __be32 pkey;
420
421 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
422 n = ip_neigh_gw4(dev, pkey);
423 }
424
425 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
426 n = NULL;
427
428 rcu_read_unlock();
429
430 return n;
431}
432
433static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
434{
435 const struct rtable *rt = container_of(dst, struct rtable, dst);
436 struct net_device *dev = dst->dev;
437 const __be32 *pkey = daddr;
438
439 if (rt->rt_gw_family == AF_INET) {
440 pkey = (const __be32 *)&rt->rt_gw4;
441 } else if (rt->rt_gw_family == AF_INET6) {
442 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
443 } else if (!daddr ||
444 (rt->rt_flags &
445 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
446 return;
447 }
448 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
449}
450
451/* Hash tables of size 2048..262144 depending on RAM size.
452 * Each bucket uses 8 bytes.
453 */
454static u32 ip_idents_mask __read_mostly;
455static atomic_t *ip_idents __read_mostly;
456static u32 *ip_tstamps __read_mostly;
457
458/* In order to protect privacy, we add a perturbation to identifiers
459 * if one generator is seldom used. This makes hard for an attacker
460 * to infer how many packets were sent between two points in time.
461 */
462static u32 ip_idents_reserve(u32 hash, int segs)
463{
464 u32 bucket, old, now = (u32)jiffies;
465 atomic_t *p_id;
466 u32 *p_tstamp;
467 u32 delta = 0;
468
469 bucket = hash & ip_idents_mask;
470 p_tstamp = ip_tstamps + bucket;
471 p_id = ip_idents + bucket;
472 old = READ_ONCE(*p_tstamp);
473
474 if (old != now && cmpxchg(p_tstamp, old, now) == old)
475 delta = get_random_u32_below(now - old);
476
477 /* If UBSAN reports an error there, please make sure your compiler
478 * supports -fno-strict-overflow before reporting it that was a bug
479 * in UBSAN, and it has been fixed in GCC-8.
480 */
481 return atomic_add_return(segs + delta, p_id) - segs;
482}
483
484void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
485{
486 u32 hash, id;
487
488 /* Note the following code is not safe, but this is okay. */
489 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
490 get_random_bytes(&net->ipv4.ip_id_key,
491 sizeof(net->ipv4.ip_id_key));
492
493 hash = siphash_3u32((__force u32)iph->daddr,
494 (__force u32)iph->saddr,
495 iph->protocol,
496 &net->ipv4.ip_id_key);
497 id = ip_idents_reserve(hash, segs);
498 iph->id = htons(id);
499}
500EXPORT_SYMBOL(__ip_select_ident);
501
502static void ip_rt_fix_tos(struct flowi4 *fl4)
503{
504 __u8 tos = RT_FL_TOS(fl4);
505
506 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
507 if (tos & RTO_ONLINK)
508 fl4->flowi4_scope = RT_SCOPE_LINK;
509}
510
511static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
512 const struct sock *sk, const struct iphdr *iph,
513 int oif, __u8 tos, u8 prot, u32 mark,
514 int flow_flags)
515{
516 __u8 scope = RT_SCOPE_UNIVERSE;
517
518 if (sk) {
519 oif = sk->sk_bound_dev_if;
520 mark = READ_ONCE(sk->sk_mark);
521 tos = ip_sock_rt_tos(sk);
522 scope = ip_sock_rt_scope(sk);
523 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
524 sk->sk_protocol;
525 }
526
527 flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
528 prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
529 sock_net_uid(net, sk));
530}
531
532static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
533 const struct sock *sk)
534{
535 const struct net *net = dev_net(skb->dev);
536 const struct iphdr *iph = ip_hdr(skb);
537 int oif = skb->dev->ifindex;
538 u8 prot = iph->protocol;
539 u32 mark = skb->mark;
540 __u8 tos = iph->tos;
541
542 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
543}
544
545static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
546{
547 const struct inet_sock *inet = inet_sk(sk);
548 const struct ip_options_rcu *inet_opt;
549 __be32 daddr = inet->inet_daddr;
550
551 rcu_read_lock();
552 inet_opt = rcu_dereference(inet->inet_opt);
553 if (inet_opt && inet_opt->opt.srr)
554 daddr = inet_opt->opt.faddr;
555 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
556 ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
557 ip_sock_rt_scope(sk),
558 inet_test_bit(HDRINCL, sk) ?
559 IPPROTO_RAW : sk->sk_protocol,
560 inet_sk_flowi_flags(sk),
561 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
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 fnhe_flush_routes(oldest);
611 *oldest_p = oldest->fnhe_next;
612 kfree_rcu(oldest, rcu);
613}
614
615static u32 fnhe_hashfun(__be32 daddr)
616{
617 static siphash_aligned_key_t fnhe_hash_key;
618 u64 hval;
619
620 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
621 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
622 return hash_64(hval, FNHE_HASH_SHIFT);
623}
624
625static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
626{
627 rt->rt_pmtu = fnhe->fnhe_pmtu;
628 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
629 rt->dst.expires = fnhe->fnhe_expires;
630
631 if (fnhe->fnhe_gw) {
632 rt->rt_flags |= RTCF_REDIRECTED;
633 rt->rt_uses_gateway = 1;
634 rt->rt_gw_family = AF_INET;
635 rt->rt_gw4 = fnhe->fnhe_gw;
636 }
637}
638
639static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
640 __be32 gw, u32 pmtu, bool lock,
641 unsigned long expires)
642{
643 struct fnhe_hash_bucket *hash;
644 struct fib_nh_exception *fnhe;
645 struct rtable *rt;
646 u32 genid, hval;
647 unsigned int i;
648 int depth;
649
650 genid = fnhe_genid(dev_net(nhc->nhc_dev));
651 hval = fnhe_hashfun(daddr);
652
653 spin_lock_bh(&fnhe_lock);
654
655 hash = rcu_dereference(nhc->nhc_exceptions);
656 if (!hash) {
657 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
658 if (!hash)
659 goto out_unlock;
660 rcu_assign_pointer(nhc->nhc_exceptions, hash);
661 }
662
663 hash += hval;
664
665 depth = 0;
666 for (fnhe = rcu_dereference(hash->chain); fnhe;
667 fnhe = rcu_dereference(fnhe->fnhe_next)) {
668 if (fnhe->fnhe_daddr == daddr)
669 break;
670 depth++;
671 }
672
673 if (fnhe) {
674 if (fnhe->fnhe_genid != genid)
675 fnhe->fnhe_genid = genid;
676 if (gw)
677 fnhe->fnhe_gw = gw;
678 if (pmtu) {
679 fnhe->fnhe_pmtu = pmtu;
680 fnhe->fnhe_mtu_locked = lock;
681 }
682 fnhe->fnhe_expires = max(1UL, expires);
683 /* Update all cached dsts too */
684 rt = rcu_dereference(fnhe->fnhe_rth_input);
685 if (rt)
686 fill_route_from_fnhe(rt, fnhe);
687 rt = rcu_dereference(fnhe->fnhe_rth_output);
688 if (rt)
689 fill_route_from_fnhe(rt, fnhe);
690 } else {
691 /* Randomize max depth to avoid some side channels attacks. */
692 int max_depth = FNHE_RECLAIM_DEPTH +
693 get_random_u32_below(FNHE_RECLAIM_DEPTH);
694
695 while (depth > max_depth) {
696 fnhe_remove_oldest(hash);
697 depth--;
698 }
699
700 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
701 if (!fnhe)
702 goto out_unlock;
703
704 fnhe->fnhe_next = hash->chain;
705
706 fnhe->fnhe_genid = genid;
707 fnhe->fnhe_daddr = daddr;
708 fnhe->fnhe_gw = gw;
709 fnhe->fnhe_pmtu = pmtu;
710 fnhe->fnhe_mtu_locked = lock;
711 fnhe->fnhe_expires = max(1UL, expires);
712
713 rcu_assign_pointer(hash->chain, fnhe);
714
715 /* Exception created; mark the cached routes for the nexthop
716 * stale, so anyone caching it rechecks if this exception
717 * applies to them.
718 */
719 rt = rcu_dereference(nhc->nhc_rth_input);
720 if (rt)
721 rt->dst.obsolete = DST_OBSOLETE_KILL;
722
723 for_each_possible_cpu(i) {
724 struct rtable __rcu **prt;
725
726 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
727 rt = rcu_dereference(*prt);
728 if (rt)
729 rt->dst.obsolete = DST_OBSOLETE_KILL;
730 }
731 }
732
733 fnhe->fnhe_stamp = jiffies;
734
735out_unlock:
736 spin_unlock_bh(&fnhe_lock);
737}
738
739static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
740 bool kill_route)
741{
742 __be32 new_gw = icmp_hdr(skb)->un.gateway;
743 __be32 old_gw = ip_hdr(skb)->saddr;
744 struct net_device *dev = skb->dev;
745 struct in_device *in_dev;
746 struct fib_result res;
747 struct neighbour *n;
748 struct net *net;
749
750 switch (icmp_hdr(skb)->code & 7) {
751 case ICMP_REDIR_NET:
752 case ICMP_REDIR_NETTOS:
753 case ICMP_REDIR_HOST:
754 case ICMP_REDIR_HOSTTOS:
755 break;
756
757 default:
758 return;
759 }
760
761 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
762 return;
763
764 in_dev = __in_dev_get_rcu(dev);
765 if (!in_dev)
766 return;
767
768 net = dev_net(dev);
769 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
770 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
771 ipv4_is_zeronet(new_gw))
772 goto reject_redirect;
773
774 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
775 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
776 goto reject_redirect;
777 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
778 goto reject_redirect;
779 } else {
780 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
781 goto reject_redirect;
782 }
783
784 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
785 if (!n)
786 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
787 if (!IS_ERR(n)) {
788 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
789 neigh_event_send(n, NULL);
790 } else {
791 if (fib_lookup(net, fl4, &res, 0) == 0) {
792 struct fib_nh_common *nhc;
793
794 fib_select_path(net, &res, fl4, skb);
795 nhc = FIB_RES_NHC(res);
796 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
797 0, false,
798 jiffies + ip_rt_gc_timeout);
799 }
800 if (kill_route)
801 rt->dst.obsolete = DST_OBSOLETE_KILL;
802 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
803 }
804 neigh_release(n);
805 }
806 return;
807
808reject_redirect:
809#ifdef CONFIG_IP_ROUTE_VERBOSE
810 if (IN_DEV_LOG_MARTIANS(in_dev)) {
811 const struct iphdr *iph = (const struct iphdr *) skb->data;
812 __be32 daddr = iph->daddr;
813 __be32 saddr = iph->saddr;
814
815 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
816 " Advised path = %pI4 -> %pI4\n",
817 &old_gw, dev->name, &new_gw,
818 &saddr, &daddr);
819 }
820#endif
821 ;
822}
823
824static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
825{
826 struct rtable *rt;
827 struct flowi4 fl4;
828 const struct iphdr *iph = (const struct iphdr *) skb->data;
829 struct net *net = dev_net(skb->dev);
830 int oif = skb->dev->ifindex;
831 u8 prot = iph->protocol;
832 u32 mark = skb->mark;
833 __u8 tos = iph->tos;
834
835 rt = dst_rtable(dst);
836
837 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
838 __ip_do_redirect(rt, skb, &fl4, true);
839}
840
841static void ipv4_negative_advice(struct sock *sk,
842 struct dst_entry *dst)
843{
844 struct rtable *rt = dst_rtable(dst);
845
846 if ((dst->obsolete > 0) ||
847 (rt->rt_flags & RTCF_REDIRECTED) ||
848 rt->dst.expires)
849 sk_dst_reset(sk);
850}
851
852/*
853 * Algorithm:
854 * 1. The first ip_rt_redirect_number redirects are sent
855 * with exponential backoff, then we stop sending them at all,
856 * assuming that the host ignores our redirects.
857 * 2. If we did not see packets requiring redirects
858 * during ip_rt_redirect_silence, we assume that the host
859 * forgot redirected route and start to send redirects again.
860 *
861 * This algorithm is much cheaper and more intelligent than dumb load limiting
862 * in icmp.c.
863 *
864 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
865 * and "frag. need" (breaks PMTU discovery) in icmp.c.
866 */
867
868void ip_rt_send_redirect(struct sk_buff *skb)
869{
870 struct rtable *rt = skb_rtable(skb);
871 struct in_device *in_dev;
872 struct inet_peer *peer;
873 struct net *net;
874 int log_martians;
875 int vif;
876
877 rcu_read_lock();
878 in_dev = __in_dev_get_rcu(rt->dst.dev);
879 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
880 rcu_read_unlock();
881 return;
882 }
883 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
884 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
885 rcu_read_unlock();
886
887 net = dev_net(rt->dst.dev);
888 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
889 if (!peer) {
890 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
891 rt_nexthop(rt, ip_hdr(skb)->daddr));
892 return;
893 }
894
895 /* No redirected packets during ip_rt_redirect_silence;
896 * reset the algorithm.
897 */
898 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
899 peer->rate_tokens = 0;
900 peer->n_redirects = 0;
901 }
902
903 /* Too many ignored redirects; do not send anything
904 * set dst.rate_last to the last seen redirected packet.
905 */
906 if (peer->n_redirects >= ip_rt_redirect_number) {
907 peer->rate_last = jiffies;
908 goto out_put_peer;
909 }
910
911 /* Check for load limit; set rate_last to the latest sent
912 * redirect.
913 */
914 if (peer->n_redirects == 0 ||
915 time_after(jiffies,
916 (peer->rate_last +
917 (ip_rt_redirect_load << peer->n_redirects)))) {
918 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
919
920 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
921 peer->rate_last = jiffies;
922 ++peer->n_redirects;
923 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
924 peer->n_redirects == ip_rt_redirect_number)
925 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
926 &ip_hdr(skb)->saddr, inet_iif(skb),
927 &ip_hdr(skb)->daddr, &gw);
928 }
929out_put_peer:
930 inet_putpeer(peer);
931}
932
933static int ip_error(struct sk_buff *skb)
934{
935 struct rtable *rt = skb_rtable(skb);
936 struct net_device *dev = skb->dev;
937 struct in_device *in_dev;
938 struct inet_peer *peer;
939 unsigned long now;
940 struct net *net;
941 SKB_DR(reason);
942 bool send;
943 int code;
944
945 if (netif_is_l3_master(skb->dev)) {
946 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
947 if (!dev)
948 goto out;
949 }
950
951 in_dev = __in_dev_get_rcu(dev);
952
953 /* IP on this device is disabled. */
954 if (!in_dev)
955 goto out;
956
957 net = dev_net(rt->dst.dev);
958 if (!IN_DEV_FORWARD(in_dev)) {
959 switch (rt->dst.error) {
960 case EHOSTUNREACH:
961 SKB_DR_SET(reason, IP_INADDRERRORS);
962 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
963 break;
964
965 case ENETUNREACH:
966 SKB_DR_SET(reason, IP_INNOROUTES);
967 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
968 break;
969 }
970 goto out;
971 }
972
973 switch (rt->dst.error) {
974 case EINVAL:
975 default:
976 goto out;
977 case EHOSTUNREACH:
978 code = ICMP_HOST_UNREACH;
979 break;
980 case ENETUNREACH:
981 code = ICMP_NET_UNREACH;
982 SKB_DR_SET(reason, IP_INNOROUTES);
983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
984 break;
985 case EACCES:
986 code = ICMP_PKT_FILTERED;
987 break;
988 }
989
990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
991 l3mdev_master_ifindex(skb->dev), 1);
992
993 send = true;
994 if (peer) {
995 now = jiffies;
996 peer->rate_tokens += now - peer->rate_last;
997 if (peer->rate_tokens > ip_rt_error_burst)
998 peer->rate_tokens = ip_rt_error_burst;
999 peer->rate_last = now;
1000 if (peer->rate_tokens >= ip_rt_error_cost)
1001 peer->rate_tokens -= ip_rt_error_cost;
1002 else
1003 send = false;
1004 inet_putpeer(peer);
1005 }
1006 if (send)
1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1008
1009out: kfree_skb_reason(skb, reason);
1010 return 0;
1011}
1012
1013static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1014{
1015 struct dst_entry *dst = &rt->dst;
1016 struct net *net = dev_net(dst->dev);
1017 struct fib_result res;
1018 bool lock = false;
1019 u32 old_mtu;
1020
1021 if (ip_mtu_locked(dst))
1022 return;
1023
1024 old_mtu = ipv4_mtu(dst);
1025 if (old_mtu < mtu)
1026 return;
1027
1028 if (mtu < net->ipv4.ip_rt_min_pmtu) {
1029 lock = true;
1030 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1031 }
1032
1033 if (rt->rt_pmtu == mtu && !lock &&
1034 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1035 return;
1036
1037 rcu_read_lock();
1038 if (fib_lookup(net, fl4, &res, 0) == 0) {
1039 struct fib_nh_common *nhc;
1040
1041 fib_select_path(net, &res, fl4, NULL);
1042 nhc = FIB_RES_NHC(res);
1043 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1044 jiffies + net->ipv4.ip_rt_mtu_expires);
1045 }
1046 rcu_read_unlock();
1047}
1048
1049static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1050 struct sk_buff *skb, u32 mtu,
1051 bool confirm_neigh)
1052{
1053 struct rtable *rt = dst_rtable(dst);
1054 struct flowi4 fl4;
1055
1056 ip_rt_build_flow_key(&fl4, sk, skb);
1057
1058 /* Don't make lookup fail for bridged encapsulations */
1059 if (skb && netif_is_any_bridge_port(skb->dev))
1060 fl4.flowi4_oif = 0;
1061
1062 __ip_rt_update_pmtu(rt, &fl4, mtu);
1063}
1064
1065void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1066 int oif, u8 protocol)
1067{
1068 const struct iphdr *iph = (const struct iphdr *)skb->data;
1069 struct flowi4 fl4;
1070 struct rtable *rt;
1071 u32 mark = IP4_REPLY_MARK(net, skb->mark);
1072
1073 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1074 0);
1075 rt = __ip_route_output_key(net, &fl4);
1076 if (!IS_ERR(rt)) {
1077 __ip_rt_update_pmtu(rt, &fl4, mtu);
1078 ip_rt_put(rt);
1079 }
1080}
1081EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1082
1083static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1084{
1085 const struct iphdr *iph = (const struct iphdr *)skb->data;
1086 struct flowi4 fl4;
1087 struct rtable *rt;
1088
1089 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1090
1091 if (!fl4.flowi4_mark)
1092 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1093
1094 rt = __ip_route_output_key(sock_net(sk), &fl4);
1095 if (!IS_ERR(rt)) {
1096 __ip_rt_update_pmtu(rt, &fl4, mtu);
1097 ip_rt_put(rt);
1098 }
1099}
1100
1101void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1102{
1103 const struct iphdr *iph = (const struct iphdr *)skb->data;
1104 struct flowi4 fl4;
1105 struct rtable *rt;
1106 struct dst_entry *odst = NULL;
1107 bool new = false;
1108 struct net *net = sock_net(sk);
1109
1110 bh_lock_sock(sk);
1111
1112 if (!ip_sk_accept_pmtu(sk))
1113 goto out;
1114
1115 odst = sk_dst_get(sk);
1116
1117 if (sock_owned_by_user(sk) || !odst) {
1118 __ipv4_sk_update_pmtu(skb, sk, mtu);
1119 goto out;
1120 }
1121
1122 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1123
1124 rt = dst_rtable(odst);
1125 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1126 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1127 if (IS_ERR(rt))
1128 goto out;
1129
1130 new = true;
1131 }
1132
1133 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1134
1135 if (!dst_check(&rt->dst, 0)) {
1136 if (new)
1137 dst_release(&rt->dst);
1138
1139 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1140 if (IS_ERR(rt))
1141 goto out;
1142
1143 new = true;
1144 }
1145
1146 if (new)
1147 sk_dst_set(sk, &rt->dst);
1148
1149out:
1150 bh_unlock_sock(sk);
1151 dst_release(odst);
1152}
1153EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1154
1155void ipv4_redirect(struct sk_buff *skb, struct net *net,
1156 int oif, u8 protocol)
1157{
1158 const struct iphdr *iph = (const struct iphdr *)skb->data;
1159 struct flowi4 fl4;
1160 struct rtable *rt;
1161
1162 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1163 rt = __ip_route_output_key(net, &fl4);
1164 if (!IS_ERR(rt)) {
1165 __ip_do_redirect(rt, skb, &fl4, false);
1166 ip_rt_put(rt);
1167 }
1168}
1169EXPORT_SYMBOL_GPL(ipv4_redirect);
1170
1171void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1172{
1173 const struct iphdr *iph = (const struct iphdr *)skb->data;
1174 struct flowi4 fl4;
1175 struct rtable *rt;
1176 struct net *net = sock_net(sk);
1177
1178 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1179 rt = __ip_route_output_key(net, &fl4);
1180 if (!IS_ERR(rt)) {
1181 __ip_do_redirect(rt, skb, &fl4, false);
1182 ip_rt_put(rt);
1183 }
1184}
1185EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1186
1187INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1188 u32 cookie)
1189{
1190 struct rtable *rt = dst_rtable(dst);
1191
1192 /* All IPV4 dsts are created with ->obsolete set to the value
1193 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1194 * into this function always.
1195 *
1196 * When a PMTU/redirect information update invalidates a route,
1197 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1198 * DST_OBSOLETE_DEAD.
1199 */
1200 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1201 return NULL;
1202 return dst;
1203}
1204EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1205
1206static void ipv4_send_dest_unreach(struct sk_buff *skb)
1207{
1208 struct net_device *dev;
1209 struct ip_options opt;
1210 int res;
1211
1212 /* Recompile ip options since IPCB may not be valid anymore.
1213 * Also check we have a reasonable ipv4 header.
1214 */
1215 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1216 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1217 return;
1218
1219 memset(&opt, 0, sizeof(opt));
1220 if (ip_hdr(skb)->ihl > 5) {
1221 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1222 return;
1223 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1224
1225 rcu_read_lock();
1226 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1227 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1228 rcu_read_unlock();
1229
1230 if (res)
1231 return;
1232 }
1233 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1234}
1235
1236static void ipv4_link_failure(struct sk_buff *skb)
1237{
1238 struct rtable *rt;
1239
1240 ipv4_send_dest_unreach(skb);
1241
1242 rt = skb_rtable(skb);
1243 if (rt)
1244 dst_set_expires(&rt->dst, 0);
1245}
1246
1247static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1248{
1249 pr_debug("%s: %pI4 -> %pI4, %s\n",
1250 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1251 skb->dev ? skb->dev->name : "?");
1252 kfree_skb(skb);
1253 WARN_ON(1);
1254 return 0;
1255}
1256
1257/*
1258 * We do not cache source address of outgoing interface,
1259 * because it is used only by IP RR, TS and SRR options,
1260 * so that it out of fast path.
1261 *
1262 * BTW remember: "addr" is allowed to be not aligned
1263 * in IP options!
1264 */
1265
1266void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1267{
1268 __be32 src;
1269
1270 if (rt_is_output_route(rt))
1271 src = ip_hdr(skb)->saddr;
1272 else {
1273 struct fib_result res;
1274 struct iphdr *iph = ip_hdr(skb);
1275 struct flowi4 fl4 = {
1276 .daddr = iph->daddr,
1277 .saddr = iph->saddr,
1278 .flowi4_tos = RT_TOS(iph->tos),
1279 .flowi4_oif = rt->dst.dev->ifindex,
1280 .flowi4_iif = skb->dev->ifindex,
1281 .flowi4_mark = skb->mark,
1282 };
1283
1284 rcu_read_lock();
1285 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1286 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1287 else
1288 src = inet_select_addr(rt->dst.dev,
1289 rt_nexthop(rt, iph->daddr),
1290 RT_SCOPE_UNIVERSE);
1291 rcu_read_unlock();
1292 }
1293 memcpy(addr, &src, 4);
1294}
1295
1296#ifdef CONFIG_IP_ROUTE_CLASSID
1297static void set_class_tag(struct rtable *rt, u32 tag)
1298{
1299 if (!(rt->dst.tclassid & 0xFFFF))
1300 rt->dst.tclassid |= tag & 0xFFFF;
1301 if (!(rt->dst.tclassid & 0xFFFF0000))
1302 rt->dst.tclassid |= tag & 0xFFFF0000;
1303}
1304#endif
1305
1306static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1307{
1308 struct net *net = dev_net(dst->dev);
1309 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1310 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1311 net->ipv4.ip_rt_min_advmss);
1312
1313 return min(advmss, IPV4_MAX_PMTU - header_size);
1314}
1315
1316INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1317{
1318 return ip_dst_mtu_maybe_forward(dst, false);
1319}
1320EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1321
1322static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1323{
1324 struct fnhe_hash_bucket *hash;
1325 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1326 u32 hval = fnhe_hashfun(daddr);
1327
1328 spin_lock_bh(&fnhe_lock);
1329
1330 hash = rcu_dereference_protected(nhc->nhc_exceptions,
1331 lockdep_is_held(&fnhe_lock));
1332 hash += hval;
1333
1334 fnhe_p = &hash->chain;
1335 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1336 while (fnhe) {
1337 if (fnhe->fnhe_daddr == daddr) {
1338 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1339 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1340 /* set fnhe_daddr to 0 to ensure it won't bind with
1341 * new dsts in rt_bind_exception().
1342 */
1343 fnhe->fnhe_daddr = 0;
1344 fnhe_flush_routes(fnhe);
1345 kfree_rcu(fnhe, rcu);
1346 break;
1347 }
1348 fnhe_p = &fnhe->fnhe_next;
1349 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1350 lockdep_is_held(&fnhe_lock));
1351 }
1352
1353 spin_unlock_bh(&fnhe_lock);
1354}
1355
1356static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1357 __be32 daddr)
1358{
1359 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1360 struct fib_nh_exception *fnhe;
1361 u32 hval;
1362
1363 if (!hash)
1364 return NULL;
1365
1366 hval = fnhe_hashfun(daddr);
1367
1368 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1369 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1370 if (fnhe->fnhe_daddr == daddr) {
1371 if (fnhe->fnhe_expires &&
1372 time_after(jiffies, fnhe->fnhe_expires)) {
1373 ip_del_fnhe(nhc, daddr);
1374 break;
1375 }
1376 return fnhe;
1377 }
1378 }
1379 return NULL;
1380}
1381
1382/* MTU selection:
1383 * 1. mtu on route is locked - use it
1384 * 2. mtu from nexthop exception
1385 * 3. mtu from egress device
1386 */
1387
1388u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1389{
1390 struct fib_nh_common *nhc = res->nhc;
1391 struct net_device *dev = nhc->nhc_dev;
1392 struct fib_info *fi = res->fi;
1393 u32 mtu = 0;
1394
1395 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1396 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1397 mtu = fi->fib_mtu;
1398
1399 if (likely(!mtu)) {
1400 struct fib_nh_exception *fnhe;
1401
1402 fnhe = find_exception(nhc, daddr);
1403 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1404 mtu = fnhe->fnhe_pmtu;
1405 }
1406
1407 if (likely(!mtu))
1408 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1409
1410 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1411}
1412
1413static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1414 __be32 daddr, const bool do_cache)
1415{
1416 bool ret = false;
1417
1418 spin_lock_bh(&fnhe_lock);
1419
1420 if (daddr == fnhe->fnhe_daddr) {
1421 struct rtable __rcu **porig;
1422 struct rtable *orig;
1423 int genid = fnhe_genid(dev_net(rt->dst.dev));
1424
1425 if (rt_is_input_route(rt))
1426 porig = &fnhe->fnhe_rth_input;
1427 else
1428 porig = &fnhe->fnhe_rth_output;
1429 orig = rcu_dereference(*porig);
1430
1431 if (fnhe->fnhe_genid != genid) {
1432 fnhe->fnhe_genid = genid;
1433 fnhe->fnhe_gw = 0;
1434 fnhe->fnhe_pmtu = 0;
1435 fnhe->fnhe_expires = 0;
1436 fnhe->fnhe_mtu_locked = false;
1437 fnhe_flush_routes(fnhe);
1438 orig = NULL;
1439 }
1440 fill_route_from_fnhe(rt, fnhe);
1441 if (!rt->rt_gw4) {
1442 rt->rt_gw4 = daddr;
1443 rt->rt_gw_family = AF_INET;
1444 }
1445
1446 if (do_cache) {
1447 dst_hold(&rt->dst);
1448 rcu_assign_pointer(*porig, rt);
1449 if (orig) {
1450 dst_dev_put(&orig->dst);
1451 dst_release(&orig->dst);
1452 }
1453 ret = true;
1454 }
1455
1456 fnhe->fnhe_stamp = jiffies;
1457 }
1458 spin_unlock_bh(&fnhe_lock);
1459
1460 return ret;
1461}
1462
1463static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1464{
1465 struct rtable *orig, *prev, **p;
1466 bool ret = true;
1467
1468 if (rt_is_input_route(rt)) {
1469 p = (struct rtable **)&nhc->nhc_rth_input;
1470 } else {
1471 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1472 }
1473 orig = *p;
1474
1475 /* hold dst before doing cmpxchg() to avoid race condition
1476 * on this dst
1477 */
1478 dst_hold(&rt->dst);
1479 prev = cmpxchg(p, orig, rt);
1480 if (prev == orig) {
1481 if (orig) {
1482 rt_add_uncached_list(orig);
1483 dst_release(&orig->dst);
1484 }
1485 } else {
1486 dst_release(&rt->dst);
1487 ret = false;
1488 }
1489
1490 return ret;
1491}
1492
1493struct uncached_list {
1494 spinlock_t lock;
1495 struct list_head head;
1496 struct list_head quarantine;
1497};
1498
1499static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1500
1501void rt_add_uncached_list(struct rtable *rt)
1502{
1503 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1504
1505 rt->dst.rt_uncached_list = ul;
1506
1507 spin_lock_bh(&ul->lock);
1508 list_add_tail(&rt->dst.rt_uncached, &ul->head);
1509 spin_unlock_bh(&ul->lock);
1510}
1511
1512void rt_del_uncached_list(struct rtable *rt)
1513{
1514 if (!list_empty(&rt->dst.rt_uncached)) {
1515 struct uncached_list *ul = rt->dst.rt_uncached_list;
1516
1517 spin_lock_bh(&ul->lock);
1518 list_del_init(&rt->dst.rt_uncached);
1519 spin_unlock_bh(&ul->lock);
1520 }
1521}
1522
1523static void ipv4_dst_destroy(struct dst_entry *dst)
1524{
1525 ip_dst_metrics_put(dst);
1526 rt_del_uncached_list(dst_rtable(dst));
1527}
1528
1529void rt_flush_dev(struct net_device *dev)
1530{
1531 struct rtable *rt, *safe;
1532 int cpu;
1533
1534 for_each_possible_cpu(cpu) {
1535 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1536
1537 if (list_empty(&ul->head))
1538 continue;
1539
1540 spin_lock_bh(&ul->lock);
1541 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1542 if (rt->dst.dev != dev)
1543 continue;
1544 rt->dst.dev = blackhole_netdev;
1545 netdev_ref_replace(dev, blackhole_netdev,
1546 &rt->dst.dev_tracker, GFP_ATOMIC);
1547 list_move(&rt->dst.rt_uncached, &ul->quarantine);
1548 }
1549 spin_unlock_bh(&ul->lock);
1550 }
1551}
1552
1553static bool rt_cache_valid(const struct rtable *rt)
1554{
1555 return rt &&
1556 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1557 !rt_is_expired(rt);
1558}
1559
1560static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1561 const struct fib_result *res,
1562 struct fib_nh_exception *fnhe,
1563 struct fib_info *fi, u16 type, u32 itag,
1564 const bool do_cache)
1565{
1566 bool cached = false;
1567
1568 if (fi) {
1569 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1570
1571 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1572 rt->rt_uses_gateway = 1;
1573 rt->rt_gw_family = nhc->nhc_gw_family;
1574 /* only INET and INET6 are supported */
1575 if (likely(nhc->nhc_gw_family == AF_INET))
1576 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1577 else
1578 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1579 }
1580
1581 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1582
1583#ifdef CONFIG_IP_ROUTE_CLASSID
1584 if (nhc->nhc_family == AF_INET) {
1585 struct fib_nh *nh;
1586
1587 nh = container_of(nhc, struct fib_nh, nh_common);
1588 rt->dst.tclassid = nh->nh_tclassid;
1589 }
1590#endif
1591 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1592 if (unlikely(fnhe))
1593 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1594 else if (do_cache)
1595 cached = rt_cache_route(nhc, rt);
1596 if (unlikely(!cached)) {
1597 /* Routes we intend to cache in nexthop exception or
1598 * FIB nexthop have the DST_NOCACHE bit clear.
1599 * However, if we are unsuccessful at storing this
1600 * route into the cache we really need to set it.
1601 */
1602 if (!rt->rt_gw4) {
1603 rt->rt_gw_family = AF_INET;
1604 rt->rt_gw4 = daddr;
1605 }
1606 rt_add_uncached_list(rt);
1607 }
1608 } else
1609 rt_add_uncached_list(rt);
1610
1611#ifdef CONFIG_IP_ROUTE_CLASSID
1612#ifdef CONFIG_IP_MULTIPLE_TABLES
1613 set_class_tag(rt, res->tclassid);
1614#endif
1615 set_class_tag(rt, itag);
1616#endif
1617}
1618
1619struct rtable *rt_dst_alloc(struct net_device *dev,
1620 unsigned int flags, u16 type,
1621 bool noxfrm)
1622{
1623 struct rtable *rt;
1624
1625 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1626 (noxfrm ? DST_NOXFRM : 0));
1627
1628 if (rt) {
1629 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1630 rt->rt_flags = flags;
1631 rt->rt_type = type;
1632 rt->rt_is_input = 0;
1633 rt->rt_iif = 0;
1634 rt->rt_pmtu = 0;
1635 rt->rt_mtu_locked = 0;
1636 rt->rt_uses_gateway = 0;
1637 rt->rt_gw_family = 0;
1638 rt->rt_gw4 = 0;
1639
1640 rt->dst.output = ip_output;
1641 if (flags & RTCF_LOCAL)
1642 rt->dst.input = ip_local_deliver;
1643 }
1644
1645 return rt;
1646}
1647EXPORT_SYMBOL(rt_dst_alloc);
1648
1649struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1650{
1651 struct rtable *new_rt;
1652
1653 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1654 rt->dst.flags);
1655
1656 if (new_rt) {
1657 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1658 new_rt->rt_flags = rt->rt_flags;
1659 new_rt->rt_type = rt->rt_type;
1660 new_rt->rt_is_input = rt->rt_is_input;
1661 new_rt->rt_iif = rt->rt_iif;
1662 new_rt->rt_pmtu = rt->rt_pmtu;
1663 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1664 new_rt->rt_gw_family = rt->rt_gw_family;
1665 if (rt->rt_gw_family == AF_INET)
1666 new_rt->rt_gw4 = rt->rt_gw4;
1667 else if (rt->rt_gw_family == AF_INET6)
1668 new_rt->rt_gw6 = rt->rt_gw6;
1669
1670 new_rt->dst.input = rt->dst.input;
1671 new_rt->dst.output = rt->dst.output;
1672 new_rt->dst.error = rt->dst.error;
1673 new_rt->dst.lastuse = jiffies;
1674 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1675 }
1676 return new_rt;
1677}
1678EXPORT_SYMBOL(rt_dst_clone);
1679
1680/* called in rcu_read_lock() section */
1681int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1682 u8 tos, struct net_device *dev,
1683 struct in_device *in_dev, u32 *itag)
1684{
1685 int err;
1686
1687 /* Primary sanity checks. */
1688 if (!in_dev)
1689 return -EINVAL;
1690
1691 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1692 skb->protocol != htons(ETH_P_IP))
1693 return -EINVAL;
1694
1695 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1696 return -EINVAL;
1697
1698 if (ipv4_is_zeronet(saddr)) {
1699 if (!ipv4_is_local_multicast(daddr) &&
1700 ip_hdr(skb)->protocol != IPPROTO_IGMP)
1701 return -EINVAL;
1702 } else {
1703 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1704 in_dev, itag);
1705 if (err < 0)
1706 return err;
1707 }
1708 return 0;
1709}
1710
1711/* called in rcu_read_lock() section */
1712static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1713 u8 tos, struct net_device *dev, int our)
1714{
1715 struct in_device *in_dev = __in_dev_get_rcu(dev);
1716 unsigned int flags = RTCF_MULTICAST;
1717 struct rtable *rth;
1718 u32 itag = 0;
1719 int err;
1720
1721 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1722 if (err)
1723 return err;
1724
1725 if (our)
1726 flags |= RTCF_LOCAL;
1727
1728 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1729 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1730
1731 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1732 false);
1733 if (!rth)
1734 return -ENOBUFS;
1735
1736#ifdef CONFIG_IP_ROUTE_CLASSID
1737 rth->dst.tclassid = itag;
1738#endif
1739 rth->dst.output = ip_rt_bug;
1740 rth->rt_is_input= 1;
1741
1742#ifdef CONFIG_IP_MROUTE
1743 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1744 rth->dst.input = ip_mr_input;
1745#endif
1746 RT_CACHE_STAT_INC(in_slow_mc);
1747
1748 skb_dst_drop(skb);
1749 skb_dst_set(skb, &rth->dst);
1750 return 0;
1751}
1752
1753
1754static void ip_handle_martian_source(struct net_device *dev,
1755 struct in_device *in_dev,
1756 struct sk_buff *skb,
1757 __be32 daddr,
1758 __be32 saddr)
1759{
1760 RT_CACHE_STAT_INC(in_martian_src);
1761#ifdef CONFIG_IP_ROUTE_VERBOSE
1762 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1763 /*
1764 * RFC1812 recommendation, if source is martian,
1765 * the only hint is MAC header.
1766 */
1767 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1768 &daddr, &saddr, dev->name);
1769 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1770 print_hex_dump(KERN_WARNING, "ll header: ",
1771 DUMP_PREFIX_OFFSET, 16, 1,
1772 skb_mac_header(skb),
1773 dev->hard_header_len, false);
1774 }
1775 }
1776#endif
1777}
1778
1779/* called in rcu_read_lock() section */
1780static int __mkroute_input(struct sk_buff *skb,
1781 const struct fib_result *res,
1782 struct in_device *in_dev,
1783 __be32 daddr, __be32 saddr, u32 tos)
1784{
1785 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1786 struct net_device *dev = nhc->nhc_dev;
1787 struct fib_nh_exception *fnhe;
1788 struct rtable *rth;
1789 int err;
1790 struct in_device *out_dev;
1791 bool do_cache;
1792 u32 itag = 0;
1793
1794 /* get a working reference to the output device */
1795 out_dev = __in_dev_get_rcu(dev);
1796 if (!out_dev) {
1797 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1798 return -EINVAL;
1799 }
1800
1801 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1802 in_dev->dev, in_dev, &itag);
1803 if (err < 0) {
1804 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1805 saddr);
1806
1807 goto cleanup;
1808 }
1809
1810 do_cache = res->fi && !itag;
1811 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1812 skb->protocol == htons(ETH_P_IP)) {
1813 __be32 gw;
1814
1815 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1816 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1817 inet_addr_onlink(out_dev, saddr, gw))
1818 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1819 }
1820
1821 if (skb->protocol != htons(ETH_P_IP)) {
1822 /* Not IP (i.e. ARP). Do not create route, if it is
1823 * invalid for proxy arp. DNAT routes are always valid.
1824 *
1825 * Proxy arp feature have been extended to allow, ARP
1826 * replies back to the same interface, to support
1827 * Private VLAN switch technologies. See arp.c.
1828 */
1829 if (out_dev == in_dev &&
1830 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1831 err = -EINVAL;
1832 goto cleanup;
1833 }
1834 }
1835
1836 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1837 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1838
1839 fnhe = find_exception(nhc, daddr);
1840 if (do_cache) {
1841 if (fnhe)
1842 rth = rcu_dereference(fnhe->fnhe_rth_input);
1843 else
1844 rth = rcu_dereference(nhc->nhc_rth_input);
1845 if (rt_cache_valid(rth)) {
1846 skb_dst_set_noref(skb, &rth->dst);
1847 goto out;
1848 }
1849 }
1850
1851 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1852 IN_DEV_ORCONF(out_dev, NOXFRM));
1853 if (!rth) {
1854 err = -ENOBUFS;
1855 goto cleanup;
1856 }
1857
1858 rth->rt_is_input = 1;
1859 RT_CACHE_STAT_INC(in_slow_tot);
1860
1861 rth->dst.input = ip_forward;
1862
1863 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1864 do_cache);
1865 lwtunnel_set_redirect(&rth->dst);
1866 skb_dst_set(skb, &rth->dst);
1867out:
1868 err = 0;
1869 cleanup:
1870 return err;
1871}
1872
1873#ifdef CONFIG_IP_ROUTE_MULTIPATH
1874/* To make ICMP packets follow the right flow, the multipath hash is
1875 * calculated from the inner IP addresses.
1876 */
1877static void ip_multipath_l3_keys(const struct sk_buff *skb,
1878 struct flow_keys *hash_keys)
1879{
1880 const struct iphdr *outer_iph = ip_hdr(skb);
1881 const struct iphdr *key_iph = outer_iph;
1882 const struct iphdr *inner_iph;
1883 const struct icmphdr *icmph;
1884 struct iphdr _inner_iph;
1885 struct icmphdr _icmph;
1886
1887 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1888 goto out;
1889
1890 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1891 goto out;
1892
1893 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1894 &_icmph);
1895 if (!icmph)
1896 goto out;
1897
1898 if (!icmp_is_err(icmph->type))
1899 goto out;
1900
1901 inner_iph = skb_header_pointer(skb,
1902 outer_iph->ihl * 4 + sizeof(_icmph),
1903 sizeof(_inner_iph), &_inner_iph);
1904 if (!inner_iph)
1905 goto out;
1906
1907 key_iph = inner_iph;
1908out:
1909 hash_keys->addrs.v4addrs.src = key_iph->saddr;
1910 hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1911}
1912
1913static u32 fib_multipath_custom_hash_outer(const struct net *net,
1914 const struct sk_buff *skb,
1915 bool *p_has_inner)
1916{
1917 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1918 struct flow_keys keys, hash_keys;
1919
1920 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1921 return 0;
1922
1923 memset(&hash_keys, 0, sizeof(hash_keys));
1924 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1925
1926 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1927 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1928 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1929 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1930 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1931 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1932 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1933 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1934 hash_keys.ports.src = keys.ports.src;
1935 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1936 hash_keys.ports.dst = keys.ports.dst;
1937
1938 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1939 return flow_hash_from_keys(&hash_keys);
1940}
1941
1942static u32 fib_multipath_custom_hash_inner(const struct net *net,
1943 const struct sk_buff *skb,
1944 bool has_inner)
1945{
1946 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1947 struct flow_keys keys, hash_keys;
1948
1949 /* We assume the packet carries an encapsulation, but if none was
1950 * encountered during dissection of the outer flow, then there is no
1951 * point in calling the flow dissector again.
1952 */
1953 if (!has_inner)
1954 return 0;
1955
1956 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1957 return 0;
1958
1959 memset(&hash_keys, 0, sizeof(hash_keys));
1960 skb_flow_dissect_flow_keys(skb, &keys, 0);
1961
1962 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1963 return 0;
1964
1965 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1966 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1967 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1968 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1969 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1970 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1971 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1972 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1973 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1974 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1975 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1976 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1977 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1978 hash_keys.tags.flow_label = keys.tags.flow_label;
1979 }
1980
1981 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1982 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1983 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1984 hash_keys.ports.src = keys.ports.src;
1985 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1986 hash_keys.ports.dst = keys.ports.dst;
1987
1988 return flow_hash_from_keys(&hash_keys);
1989}
1990
1991static u32 fib_multipath_custom_hash_skb(const struct net *net,
1992 const struct sk_buff *skb)
1993{
1994 u32 mhash, mhash_inner;
1995 bool has_inner = true;
1996
1997 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1998 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1999
2000 return jhash_2words(mhash, mhash_inner, 0);
2001}
2002
2003static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2004 const struct flowi4 *fl4)
2005{
2006 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2007 struct flow_keys hash_keys;
2008
2009 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2010 return 0;
2011
2012 memset(&hash_keys, 0, sizeof(hash_keys));
2013 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2014 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2015 hash_keys.addrs.v4addrs.src = fl4->saddr;
2016 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2017 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2018 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2019 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2020 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2021 hash_keys.ports.src = fl4->fl4_sport;
2022 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2023 hash_keys.ports.dst = fl4->fl4_dport;
2024
2025 return flow_hash_from_keys(&hash_keys);
2026}
2027
2028/* if skb is set it will be used and fl4 can be NULL */
2029int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2030 const struct sk_buff *skb, struct flow_keys *flkeys)
2031{
2032 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2033 struct flow_keys hash_keys;
2034 u32 mhash = 0;
2035
2036 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2037 case 0:
2038 memset(&hash_keys, 0, sizeof(hash_keys));
2039 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2040 if (skb) {
2041 ip_multipath_l3_keys(skb, &hash_keys);
2042 } else {
2043 hash_keys.addrs.v4addrs.src = fl4->saddr;
2044 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2045 }
2046 mhash = flow_hash_from_keys(&hash_keys);
2047 break;
2048 case 1:
2049 /* skb is currently provided only when forwarding */
2050 if (skb) {
2051 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2052 struct flow_keys keys;
2053
2054 /* short-circuit if we already have L4 hash present */
2055 if (skb->l4_hash)
2056 return skb_get_hash_raw(skb) >> 1;
2057
2058 memset(&hash_keys, 0, sizeof(hash_keys));
2059
2060 if (!flkeys) {
2061 skb_flow_dissect_flow_keys(skb, &keys, flag);
2062 flkeys = &keys;
2063 }
2064
2065 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2066 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2067 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2068 hash_keys.ports.src = flkeys->ports.src;
2069 hash_keys.ports.dst = flkeys->ports.dst;
2070 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2071 } else {
2072 memset(&hash_keys, 0, sizeof(hash_keys));
2073 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2074 hash_keys.addrs.v4addrs.src = fl4->saddr;
2075 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2076 hash_keys.ports.src = fl4->fl4_sport;
2077 hash_keys.ports.dst = fl4->fl4_dport;
2078 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2079 }
2080 mhash = flow_hash_from_keys(&hash_keys);
2081 break;
2082 case 2:
2083 memset(&hash_keys, 0, sizeof(hash_keys));
2084 /* skb is currently provided only when forwarding */
2085 if (skb) {
2086 struct flow_keys keys;
2087
2088 skb_flow_dissect_flow_keys(skb, &keys, 0);
2089 /* Inner can be v4 or v6 */
2090 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2091 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2092 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2093 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2094 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2095 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2096 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2097 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2098 hash_keys.tags.flow_label = keys.tags.flow_label;
2099 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2100 } else {
2101 /* Same as case 0 */
2102 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2103 ip_multipath_l3_keys(skb, &hash_keys);
2104 }
2105 } else {
2106 /* Same as case 0 */
2107 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2108 hash_keys.addrs.v4addrs.src = fl4->saddr;
2109 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2110 }
2111 mhash = flow_hash_from_keys(&hash_keys);
2112 break;
2113 case 3:
2114 if (skb)
2115 mhash = fib_multipath_custom_hash_skb(net, skb);
2116 else
2117 mhash = fib_multipath_custom_hash_fl4(net, fl4);
2118 break;
2119 }
2120
2121 if (multipath_hash)
2122 mhash = jhash_2words(mhash, multipath_hash, 0);
2123
2124 return mhash >> 1;
2125}
2126#endif /* CONFIG_IP_ROUTE_MULTIPATH */
2127
2128static int ip_mkroute_input(struct sk_buff *skb,
2129 struct fib_result *res,
2130 struct in_device *in_dev,
2131 __be32 daddr, __be32 saddr, u32 tos,
2132 struct flow_keys *hkeys)
2133{
2134#ifdef CONFIG_IP_ROUTE_MULTIPATH
2135 if (res->fi && fib_info_num_path(res->fi) > 1) {
2136 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2137
2138 fib_select_multipath(res, h);
2139 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2140 }
2141#endif
2142
2143 /* create a routing cache entry */
2144 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2145}
2146
2147/* Implements all the saddr-related checks as ip_route_input_slow(),
2148 * assuming daddr is valid and the destination is not a local broadcast one.
2149 * Uses the provided hint instead of performing a route lookup.
2150 */
2151int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2152 u8 tos, struct net_device *dev,
2153 const struct sk_buff *hint)
2154{
2155 struct in_device *in_dev = __in_dev_get_rcu(dev);
2156 struct rtable *rt = skb_rtable(hint);
2157 struct net *net = dev_net(dev);
2158 int err = -EINVAL;
2159 u32 tag = 0;
2160
2161 if (!in_dev)
2162 return -EINVAL;
2163
2164 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2165 goto martian_source;
2166
2167 if (ipv4_is_zeronet(saddr))
2168 goto martian_source;
2169
2170 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2171 goto martian_source;
2172
2173 if (rt->rt_type != RTN_LOCAL)
2174 goto skip_validate_source;
2175
2176 tos &= IPTOS_RT_MASK;
2177 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2178 if (err < 0)
2179 goto martian_source;
2180
2181skip_validate_source:
2182 skb_dst_copy(skb, hint);
2183 return 0;
2184
2185martian_source:
2186 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2187 return err;
2188}
2189
2190/* get device for dst_alloc with local routes */
2191static struct net_device *ip_rt_get_dev(struct net *net,
2192 const struct fib_result *res)
2193{
2194 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2195 struct net_device *dev = NULL;
2196
2197 if (nhc)
2198 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2199
2200 return dev ? : net->loopback_dev;
2201}
2202
2203/*
2204 * NOTE. We drop all the packets that has local source
2205 * addresses, because every properly looped back packet
2206 * must have correct destination already attached by output routine.
2207 * Changes in the enforced policies must be applied also to
2208 * ip_route_use_hint().
2209 *
2210 * Such approach solves two big problems:
2211 * 1. Not simplex devices are handled properly.
2212 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2213 * called with rcu_read_lock()
2214 */
2215
2216static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2217 u8 tos, struct net_device *dev,
2218 struct fib_result *res)
2219{
2220 struct in_device *in_dev = __in_dev_get_rcu(dev);
2221 struct flow_keys *flkeys = NULL, _flkeys;
2222 struct net *net = dev_net(dev);
2223 struct ip_tunnel_info *tun_info;
2224 int err = -EINVAL;
2225 unsigned int flags = 0;
2226 u32 itag = 0;
2227 struct rtable *rth;
2228 struct flowi4 fl4;
2229 bool do_cache = true;
2230
2231 /* IP on this device is disabled. */
2232
2233 if (!in_dev)
2234 goto out;
2235
2236 /* Check for the most weird martians, which can be not detected
2237 * by fib_lookup.
2238 */
2239
2240 tun_info = skb_tunnel_info(skb);
2241 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2242 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2243 else
2244 fl4.flowi4_tun_key.tun_id = 0;
2245 skb_dst_drop(skb);
2246
2247 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2248 goto martian_source;
2249
2250 res->fi = NULL;
2251 res->table = NULL;
2252 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2253 goto brd_input;
2254
2255 /* Accept zero addresses only to limited broadcast;
2256 * I even do not know to fix it or not. Waiting for complains :-)
2257 */
2258 if (ipv4_is_zeronet(saddr))
2259 goto martian_source;
2260
2261 if (ipv4_is_zeronet(daddr))
2262 goto martian_destination;
2263
2264 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2265 * and call it once if daddr or/and saddr are loopback addresses
2266 */
2267 if (ipv4_is_loopback(daddr)) {
2268 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2269 goto martian_destination;
2270 } else if (ipv4_is_loopback(saddr)) {
2271 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2272 goto martian_source;
2273 }
2274
2275 /*
2276 * Now we are ready to route packet.
2277 */
2278 fl4.flowi4_l3mdev = 0;
2279 fl4.flowi4_oif = 0;
2280 fl4.flowi4_iif = dev->ifindex;
2281 fl4.flowi4_mark = skb->mark;
2282 fl4.flowi4_tos = tos;
2283 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2284 fl4.flowi4_flags = 0;
2285 fl4.daddr = daddr;
2286 fl4.saddr = saddr;
2287 fl4.flowi4_uid = sock_net_uid(net, NULL);
2288 fl4.flowi4_multipath_hash = 0;
2289
2290 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2291 flkeys = &_flkeys;
2292 } else {
2293 fl4.flowi4_proto = 0;
2294 fl4.fl4_sport = 0;
2295 fl4.fl4_dport = 0;
2296 }
2297
2298 err = fib_lookup(net, &fl4, res, 0);
2299 if (err != 0) {
2300 if (!IN_DEV_FORWARD(in_dev))
2301 err = -EHOSTUNREACH;
2302 goto no_route;
2303 }
2304
2305 if (res->type == RTN_BROADCAST) {
2306 if (IN_DEV_BFORWARD(in_dev))
2307 goto make_route;
2308 /* not do cache if bc_forwarding is enabled */
2309 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2310 do_cache = false;
2311 goto brd_input;
2312 }
2313
2314 if (res->type == RTN_LOCAL) {
2315 err = fib_validate_source(skb, saddr, daddr, tos,
2316 0, dev, in_dev, &itag);
2317 if (err < 0)
2318 goto martian_source;
2319 goto local_input;
2320 }
2321
2322 if (!IN_DEV_FORWARD(in_dev)) {
2323 err = -EHOSTUNREACH;
2324 goto no_route;
2325 }
2326 if (res->type != RTN_UNICAST)
2327 goto martian_destination;
2328
2329make_route:
2330 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2331out: return err;
2332
2333brd_input:
2334 if (skb->protocol != htons(ETH_P_IP))
2335 goto e_inval;
2336
2337 if (!ipv4_is_zeronet(saddr)) {
2338 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2339 in_dev, &itag);
2340 if (err < 0)
2341 goto martian_source;
2342 }
2343 flags |= RTCF_BROADCAST;
2344 res->type = RTN_BROADCAST;
2345 RT_CACHE_STAT_INC(in_brd);
2346
2347local_input:
2348 if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2349 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2350
2351 do_cache &= res->fi && !itag;
2352 if (do_cache) {
2353 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2354
2355 rth = rcu_dereference(nhc->nhc_rth_input);
2356 if (rt_cache_valid(rth)) {
2357 skb_dst_set_noref(skb, &rth->dst);
2358 err = 0;
2359 goto out;
2360 }
2361 }
2362
2363 rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2364 flags | RTCF_LOCAL, res->type, false);
2365 if (!rth)
2366 goto e_nobufs;
2367
2368 rth->dst.output= ip_rt_bug;
2369#ifdef CONFIG_IP_ROUTE_CLASSID
2370 rth->dst.tclassid = itag;
2371#endif
2372 rth->rt_is_input = 1;
2373
2374 RT_CACHE_STAT_INC(in_slow_tot);
2375 if (res->type == RTN_UNREACHABLE) {
2376 rth->dst.input= ip_error;
2377 rth->dst.error= -err;
2378 rth->rt_flags &= ~RTCF_LOCAL;
2379 }
2380
2381 if (do_cache) {
2382 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2383
2384 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2385 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2386 WARN_ON(rth->dst.input == lwtunnel_input);
2387 rth->dst.lwtstate->orig_input = rth->dst.input;
2388 rth->dst.input = lwtunnel_input;
2389 }
2390
2391 if (unlikely(!rt_cache_route(nhc, rth)))
2392 rt_add_uncached_list(rth);
2393 }
2394 skb_dst_set(skb, &rth->dst);
2395 err = 0;
2396 goto out;
2397
2398no_route:
2399 RT_CACHE_STAT_INC(in_no_route);
2400 res->type = RTN_UNREACHABLE;
2401 res->fi = NULL;
2402 res->table = NULL;
2403 goto local_input;
2404
2405 /*
2406 * Do not cache martian addresses: they should be logged (RFC1812)
2407 */
2408martian_destination:
2409 RT_CACHE_STAT_INC(in_martian_dst);
2410#ifdef CONFIG_IP_ROUTE_VERBOSE
2411 if (IN_DEV_LOG_MARTIANS(in_dev))
2412 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2413 &daddr, &saddr, dev->name);
2414#endif
2415
2416e_inval:
2417 err = -EINVAL;
2418 goto out;
2419
2420e_nobufs:
2421 err = -ENOBUFS;
2422 goto out;
2423
2424martian_source:
2425 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2426 goto out;
2427}
2428
2429/* called with rcu_read_lock held */
2430static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2431 u8 tos, struct net_device *dev, struct fib_result *res)
2432{
2433 /* Multicast recognition logic is moved from route cache to here.
2434 * The problem was that too many Ethernet cards have broken/missing
2435 * hardware multicast filters :-( As result the host on multicasting
2436 * network acquires a lot of useless route cache entries, sort of
2437 * SDR messages from all the world. Now we try to get rid of them.
2438 * Really, provided software IP multicast filter is organized
2439 * reasonably (at least, hashed), it does not result in a slowdown
2440 * comparing with route cache reject entries.
2441 * Note, that multicast routers are not affected, because
2442 * route cache entry is created eventually.
2443 */
2444 if (ipv4_is_multicast(daddr)) {
2445 struct in_device *in_dev = __in_dev_get_rcu(dev);
2446 int our = 0;
2447 int err = -EINVAL;
2448
2449 if (!in_dev)
2450 return err;
2451 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2452 ip_hdr(skb)->protocol);
2453
2454 /* check l3 master if no match yet */
2455 if (!our && netif_is_l3_slave(dev)) {
2456 struct in_device *l3_in_dev;
2457
2458 l3_in_dev = __in_dev_get_rcu(skb->dev);
2459 if (l3_in_dev)
2460 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2461 ip_hdr(skb)->protocol);
2462 }
2463
2464 if (our
2465#ifdef CONFIG_IP_MROUTE
2466 ||
2467 (!ipv4_is_local_multicast(daddr) &&
2468 IN_DEV_MFORWARD(in_dev))
2469#endif
2470 ) {
2471 err = ip_route_input_mc(skb, daddr, saddr,
2472 tos, dev, our);
2473 }
2474 return err;
2475 }
2476
2477 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2478}
2479
2480int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2481 u8 tos, struct net_device *dev)
2482{
2483 struct fib_result res;
2484 int err;
2485
2486 tos &= IPTOS_RT_MASK;
2487 rcu_read_lock();
2488 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2489 rcu_read_unlock();
2490
2491 return err;
2492}
2493EXPORT_SYMBOL(ip_route_input_noref);
2494
2495/* called with rcu_read_lock() */
2496static struct rtable *__mkroute_output(const struct fib_result *res,
2497 const struct flowi4 *fl4, int orig_oif,
2498 struct net_device *dev_out,
2499 unsigned int flags)
2500{
2501 struct fib_info *fi = res->fi;
2502 struct fib_nh_exception *fnhe;
2503 struct in_device *in_dev;
2504 u16 type = res->type;
2505 struct rtable *rth;
2506 bool do_cache;
2507
2508 in_dev = __in_dev_get_rcu(dev_out);
2509 if (!in_dev)
2510 return ERR_PTR(-EINVAL);
2511
2512 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2513 if (ipv4_is_loopback(fl4->saddr) &&
2514 !(dev_out->flags & IFF_LOOPBACK) &&
2515 !netif_is_l3_master(dev_out))
2516 return ERR_PTR(-EINVAL);
2517
2518 if (ipv4_is_lbcast(fl4->daddr))
2519 type = RTN_BROADCAST;
2520 else if (ipv4_is_multicast(fl4->daddr))
2521 type = RTN_MULTICAST;
2522 else if (ipv4_is_zeronet(fl4->daddr))
2523 return ERR_PTR(-EINVAL);
2524
2525 if (dev_out->flags & IFF_LOOPBACK)
2526 flags |= RTCF_LOCAL;
2527
2528 do_cache = true;
2529 if (type == RTN_BROADCAST) {
2530 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2531 fi = NULL;
2532 } else if (type == RTN_MULTICAST) {
2533 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2534 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2535 fl4->flowi4_proto))
2536 flags &= ~RTCF_LOCAL;
2537 else
2538 do_cache = false;
2539 /* If multicast route do not exist use
2540 * default one, but do not gateway in this case.
2541 * Yes, it is hack.
2542 */
2543 if (fi && res->prefixlen < 4)
2544 fi = NULL;
2545 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2546 (orig_oif != dev_out->ifindex)) {
2547 /* For local routes that require a particular output interface
2548 * we do not want to cache the result. Caching the result
2549 * causes incorrect behaviour when there are multiple source
2550 * addresses on the interface, the end result being that if the
2551 * intended recipient is waiting on that interface for the
2552 * packet he won't receive it because it will be delivered on
2553 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2554 * be set to the loopback interface as well.
2555 */
2556 do_cache = false;
2557 }
2558
2559 fnhe = NULL;
2560 do_cache &= fi != NULL;
2561 if (fi) {
2562 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2563 struct rtable __rcu **prth;
2564
2565 fnhe = find_exception(nhc, fl4->daddr);
2566 if (!do_cache)
2567 goto add;
2568 if (fnhe) {
2569 prth = &fnhe->fnhe_rth_output;
2570 } else {
2571 if (unlikely(fl4->flowi4_flags &
2572 FLOWI_FLAG_KNOWN_NH &&
2573 !(nhc->nhc_gw_family &&
2574 nhc->nhc_scope == RT_SCOPE_LINK))) {
2575 do_cache = false;
2576 goto add;
2577 }
2578 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2579 }
2580 rth = rcu_dereference(*prth);
2581 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2582 return rth;
2583 }
2584
2585add:
2586 rth = rt_dst_alloc(dev_out, flags, type,
2587 IN_DEV_ORCONF(in_dev, NOXFRM));
2588 if (!rth)
2589 return ERR_PTR(-ENOBUFS);
2590
2591 rth->rt_iif = orig_oif;
2592
2593 RT_CACHE_STAT_INC(out_slow_tot);
2594
2595 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2596 if (flags & RTCF_LOCAL &&
2597 !(dev_out->flags & IFF_LOOPBACK)) {
2598 rth->dst.output = ip_mc_output;
2599 RT_CACHE_STAT_INC(out_slow_mc);
2600 }
2601#ifdef CONFIG_IP_MROUTE
2602 if (type == RTN_MULTICAST) {
2603 if (IN_DEV_MFORWARD(in_dev) &&
2604 !ipv4_is_local_multicast(fl4->daddr)) {
2605 rth->dst.input = ip_mr_input;
2606 rth->dst.output = ip_mc_output;
2607 }
2608 }
2609#endif
2610 }
2611
2612 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2613 lwtunnel_set_redirect(&rth->dst);
2614
2615 return rth;
2616}
2617
2618/*
2619 * Major route resolver routine.
2620 */
2621
2622struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2623 const struct sk_buff *skb)
2624{
2625 struct fib_result res = {
2626 .type = RTN_UNSPEC,
2627 .fi = NULL,
2628 .table = NULL,
2629 .tclassid = 0,
2630 };
2631 struct rtable *rth;
2632
2633 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2634 ip_rt_fix_tos(fl4);
2635
2636 rcu_read_lock();
2637 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2638 rcu_read_unlock();
2639
2640 return rth;
2641}
2642EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2643
2644struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2645 struct fib_result *res,
2646 const struct sk_buff *skb)
2647{
2648 struct net_device *dev_out = NULL;
2649 int orig_oif = fl4->flowi4_oif;
2650 unsigned int flags = 0;
2651 struct rtable *rth;
2652 int err;
2653
2654 if (fl4->saddr) {
2655 if (ipv4_is_multicast(fl4->saddr) ||
2656 ipv4_is_lbcast(fl4->saddr) ||
2657 ipv4_is_zeronet(fl4->saddr)) {
2658 rth = ERR_PTR(-EINVAL);
2659 goto out;
2660 }
2661
2662 rth = ERR_PTR(-ENETUNREACH);
2663
2664 /* I removed check for oif == dev_out->oif here.
2665 * It was wrong for two reasons:
2666 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2667 * is assigned to multiple interfaces.
2668 * 2. Moreover, we are allowed to send packets with saddr
2669 * of another iface. --ANK
2670 */
2671
2672 if (fl4->flowi4_oif == 0 &&
2673 (ipv4_is_multicast(fl4->daddr) ||
2674 ipv4_is_lbcast(fl4->daddr))) {
2675 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2676 dev_out = __ip_dev_find(net, fl4->saddr, false);
2677 if (!dev_out)
2678 goto out;
2679
2680 /* Special hack: user can direct multicasts
2681 * and limited broadcast via necessary interface
2682 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2683 * This hack is not just for fun, it allows
2684 * vic,vat and friends to work.
2685 * They bind socket to loopback, set ttl to zero
2686 * and expect that it will work.
2687 * From the viewpoint of routing cache they are broken,
2688 * because we are not allowed to build multicast path
2689 * with loopback source addr (look, routing cache
2690 * cannot know, that ttl is zero, so that packet
2691 * will not leave this host and route is valid).
2692 * Luckily, this hack is good workaround.
2693 */
2694
2695 fl4->flowi4_oif = dev_out->ifindex;
2696 goto make_route;
2697 }
2698
2699 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2700 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2701 if (!__ip_dev_find(net, fl4->saddr, false))
2702 goto out;
2703 }
2704 }
2705
2706
2707 if (fl4->flowi4_oif) {
2708 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2709 rth = ERR_PTR(-ENODEV);
2710 if (!dev_out)
2711 goto out;
2712
2713 /* RACE: Check return value of inet_select_addr instead. */
2714 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2715 rth = ERR_PTR(-ENETUNREACH);
2716 goto out;
2717 }
2718 if (ipv4_is_local_multicast(fl4->daddr) ||
2719 ipv4_is_lbcast(fl4->daddr) ||
2720 fl4->flowi4_proto == IPPROTO_IGMP) {
2721 if (!fl4->saddr)
2722 fl4->saddr = inet_select_addr(dev_out, 0,
2723 RT_SCOPE_LINK);
2724 goto make_route;
2725 }
2726 if (!fl4->saddr) {
2727 if (ipv4_is_multicast(fl4->daddr))
2728 fl4->saddr = inet_select_addr(dev_out, 0,
2729 fl4->flowi4_scope);
2730 else if (!fl4->daddr)
2731 fl4->saddr = inet_select_addr(dev_out, 0,
2732 RT_SCOPE_HOST);
2733 }
2734 }
2735
2736 if (!fl4->daddr) {
2737 fl4->daddr = fl4->saddr;
2738 if (!fl4->daddr)
2739 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2740 dev_out = net->loopback_dev;
2741 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2742 res->type = RTN_LOCAL;
2743 flags |= RTCF_LOCAL;
2744 goto make_route;
2745 }
2746
2747 err = fib_lookup(net, fl4, res, 0);
2748 if (err) {
2749 res->fi = NULL;
2750 res->table = NULL;
2751 if (fl4->flowi4_oif &&
2752 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2753 /* Apparently, routing tables are wrong. Assume,
2754 * that the destination is on link.
2755 *
2756 * WHY? DW.
2757 * Because we are allowed to send to iface
2758 * even if it has NO routes and NO assigned
2759 * addresses. When oif is specified, routing
2760 * tables are looked up with only one purpose:
2761 * to catch if destination is gatewayed, rather than
2762 * direct. Moreover, if MSG_DONTROUTE is set,
2763 * we send packet, ignoring both routing tables
2764 * and ifaddr state. --ANK
2765 *
2766 *
2767 * We could make it even if oif is unknown,
2768 * likely IPv6, but we do not.
2769 */
2770
2771 if (fl4->saddr == 0)
2772 fl4->saddr = inet_select_addr(dev_out, 0,
2773 RT_SCOPE_LINK);
2774 res->type = RTN_UNICAST;
2775 goto make_route;
2776 }
2777 rth = ERR_PTR(err);
2778 goto out;
2779 }
2780
2781 if (res->type == RTN_LOCAL) {
2782 if (!fl4->saddr) {
2783 if (res->fi->fib_prefsrc)
2784 fl4->saddr = res->fi->fib_prefsrc;
2785 else
2786 fl4->saddr = fl4->daddr;
2787 }
2788
2789 /* L3 master device is the loopback for that domain */
2790 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2791 net->loopback_dev;
2792
2793 /* make sure orig_oif points to fib result device even
2794 * though packet rx/tx happens over loopback or l3mdev
2795 */
2796 orig_oif = FIB_RES_OIF(*res);
2797
2798 fl4->flowi4_oif = dev_out->ifindex;
2799 flags |= RTCF_LOCAL;
2800 goto make_route;
2801 }
2802
2803 fib_select_path(net, res, fl4, skb);
2804
2805 dev_out = FIB_RES_DEV(*res);
2806
2807make_route:
2808 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2809
2810out:
2811 return rth;
2812}
2813
2814static struct dst_ops ipv4_dst_blackhole_ops = {
2815 .family = AF_INET,
2816 .default_advmss = ipv4_default_advmss,
2817 .neigh_lookup = ipv4_neigh_lookup,
2818 .check = dst_blackhole_check,
2819 .cow_metrics = dst_blackhole_cow_metrics,
2820 .update_pmtu = dst_blackhole_update_pmtu,
2821 .redirect = dst_blackhole_redirect,
2822 .mtu = dst_blackhole_mtu,
2823};
2824
2825struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2826{
2827 struct rtable *ort = dst_rtable(dst_orig);
2828 struct rtable *rt;
2829
2830 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2831 if (rt) {
2832 struct dst_entry *new = &rt->dst;
2833
2834 new->__use = 1;
2835 new->input = dst_discard;
2836 new->output = dst_discard_out;
2837
2838 new->dev = net->loopback_dev;
2839 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2840
2841 rt->rt_is_input = ort->rt_is_input;
2842 rt->rt_iif = ort->rt_iif;
2843 rt->rt_pmtu = ort->rt_pmtu;
2844 rt->rt_mtu_locked = ort->rt_mtu_locked;
2845
2846 rt->rt_genid = rt_genid_ipv4(net);
2847 rt->rt_flags = ort->rt_flags;
2848 rt->rt_type = ort->rt_type;
2849 rt->rt_uses_gateway = ort->rt_uses_gateway;
2850 rt->rt_gw_family = ort->rt_gw_family;
2851 if (rt->rt_gw_family == AF_INET)
2852 rt->rt_gw4 = ort->rt_gw4;
2853 else if (rt->rt_gw_family == AF_INET6)
2854 rt->rt_gw6 = ort->rt_gw6;
2855 }
2856
2857 dst_release(dst_orig);
2858
2859 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2860}
2861
2862struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2863 const struct sock *sk)
2864{
2865 struct rtable *rt = __ip_route_output_key(net, flp4);
2866
2867 if (IS_ERR(rt))
2868 return rt;
2869
2870 if (flp4->flowi4_proto) {
2871 flp4->flowi4_oif = rt->dst.dev->ifindex;
2872 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2873 flowi4_to_flowi(flp4),
2874 sk, 0));
2875 }
2876
2877 return rt;
2878}
2879EXPORT_SYMBOL_GPL(ip_route_output_flow);
2880
2881/* called with rcu_read_lock held */
2882static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2883 struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2884 struct sk_buff *skb, u32 portid, u32 seq,
2885 unsigned int flags)
2886{
2887 struct rtmsg *r;
2888 struct nlmsghdr *nlh;
2889 unsigned long expires = 0;
2890 u32 error;
2891 u32 metrics[RTAX_MAX];
2892
2893 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2894 if (!nlh)
2895 return -EMSGSIZE;
2896
2897 r = nlmsg_data(nlh);
2898 r->rtm_family = AF_INET;
2899 r->rtm_dst_len = 32;
2900 r->rtm_src_len = 0;
2901 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0;
2902 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2903 if (nla_put_u32(skb, RTA_TABLE, table_id))
2904 goto nla_put_failure;
2905 r->rtm_type = rt->rt_type;
2906 r->rtm_scope = RT_SCOPE_UNIVERSE;
2907 r->rtm_protocol = RTPROT_UNSPEC;
2908 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2909 if (rt->rt_flags & RTCF_NOTIFY)
2910 r->rtm_flags |= RTM_F_NOTIFY;
2911 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2912 r->rtm_flags |= RTCF_DOREDIRECT;
2913
2914 if (nla_put_in_addr(skb, RTA_DST, dst))
2915 goto nla_put_failure;
2916 if (src) {
2917 r->rtm_src_len = 32;
2918 if (nla_put_in_addr(skb, RTA_SRC, src))
2919 goto nla_put_failure;
2920 }
2921 if (rt->dst.dev &&
2922 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2923 goto nla_put_failure;
2924 if (rt->dst.lwtstate &&
2925 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2926 goto nla_put_failure;
2927#ifdef CONFIG_IP_ROUTE_CLASSID
2928 if (rt->dst.tclassid &&
2929 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2930 goto nla_put_failure;
2931#endif
2932 if (fl4 && !rt_is_input_route(rt) &&
2933 fl4->saddr != src) {
2934 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2935 goto nla_put_failure;
2936 }
2937 if (rt->rt_uses_gateway) {
2938 if (rt->rt_gw_family == AF_INET &&
2939 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2940 goto nla_put_failure;
2941 } else if (rt->rt_gw_family == AF_INET6) {
2942 int alen = sizeof(struct in6_addr);
2943 struct nlattr *nla;
2944 struct rtvia *via;
2945
2946 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2947 if (!nla)
2948 goto nla_put_failure;
2949
2950 via = nla_data(nla);
2951 via->rtvia_family = AF_INET6;
2952 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2953 }
2954 }
2955
2956 expires = rt->dst.expires;
2957 if (expires) {
2958 unsigned long now = jiffies;
2959
2960 if (time_before(now, expires))
2961 expires -= now;
2962 else
2963 expires = 0;
2964 }
2965
2966 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2967 if (rt->rt_pmtu && expires)
2968 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2969 if (rt->rt_mtu_locked && expires)
2970 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2971 if (rtnetlink_put_metrics(skb, metrics) < 0)
2972 goto nla_put_failure;
2973
2974 if (fl4) {
2975 if (fl4->flowi4_mark &&
2976 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2977 goto nla_put_failure;
2978
2979 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2980 nla_put_u32(skb, RTA_UID,
2981 from_kuid_munged(current_user_ns(),
2982 fl4->flowi4_uid)))
2983 goto nla_put_failure;
2984
2985 if (rt_is_input_route(rt)) {
2986#ifdef CONFIG_IP_MROUTE
2987 if (ipv4_is_multicast(dst) &&
2988 !ipv4_is_local_multicast(dst) &&
2989 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
2990 int err = ipmr_get_route(net, skb,
2991 fl4->saddr, fl4->daddr,
2992 r, portid);
2993
2994 if (err <= 0) {
2995 if (err == 0)
2996 return 0;
2997 goto nla_put_failure;
2998 }
2999 } else
3000#endif
3001 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3002 goto nla_put_failure;
3003 }
3004 }
3005
3006 error = rt->dst.error;
3007
3008 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3009 goto nla_put_failure;
3010
3011 nlmsg_end(skb, nlh);
3012 return 0;
3013
3014nla_put_failure:
3015 nlmsg_cancel(skb, nlh);
3016 return -EMSGSIZE;
3017}
3018
3019static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3020 struct netlink_callback *cb, u32 table_id,
3021 struct fnhe_hash_bucket *bucket, int genid,
3022 int *fa_index, int fa_start, unsigned int flags)
3023{
3024 int i;
3025
3026 for (i = 0; i < FNHE_HASH_SIZE; i++) {
3027 struct fib_nh_exception *fnhe;
3028
3029 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3030 fnhe = rcu_dereference(fnhe->fnhe_next)) {
3031 struct rtable *rt;
3032 int err;
3033
3034 if (*fa_index < fa_start)
3035 goto next;
3036
3037 if (fnhe->fnhe_genid != genid)
3038 goto next;
3039
3040 if (fnhe->fnhe_expires &&
3041 time_after(jiffies, fnhe->fnhe_expires))
3042 goto next;
3043
3044 rt = rcu_dereference(fnhe->fnhe_rth_input);
3045 if (!rt)
3046 rt = rcu_dereference(fnhe->fnhe_rth_output);
3047 if (!rt)
3048 goto next;
3049
3050 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3051 table_id, NULL, skb,
3052 NETLINK_CB(cb->skb).portid,
3053 cb->nlh->nlmsg_seq, flags);
3054 if (err)
3055 return err;
3056next:
3057 (*fa_index)++;
3058 }
3059 }
3060
3061 return 0;
3062}
3063
3064int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3065 u32 table_id, struct fib_info *fi,
3066 int *fa_index, int fa_start, unsigned int flags)
3067{
3068 struct net *net = sock_net(cb->skb->sk);
3069 int nhsel, genid = fnhe_genid(net);
3070
3071 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3072 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3073 struct fnhe_hash_bucket *bucket;
3074 int err;
3075
3076 if (nhc->nhc_flags & RTNH_F_DEAD)
3077 continue;
3078
3079 rcu_read_lock();
3080 bucket = rcu_dereference(nhc->nhc_exceptions);
3081 err = 0;
3082 if (bucket)
3083 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3084 genid, fa_index, fa_start,
3085 flags);
3086 rcu_read_unlock();
3087 if (err)
3088 return err;
3089 }
3090
3091 return 0;
3092}
3093
3094static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3095 u8 ip_proto, __be16 sport,
3096 __be16 dport)
3097{
3098 struct sk_buff *skb;
3099 struct iphdr *iph;
3100
3101 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3102 if (!skb)
3103 return NULL;
3104
3105 /* Reserve room for dummy headers, this skb can pass
3106 * through good chunk of routing engine.
3107 */
3108 skb_reset_mac_header(skb);
3109 skb_reset_network_header(skb);
3110 skb->protocol = htons(ETH_P_IP);
3111 iph = skb_put(skb, sizeof(struct iphdr));
3112 iph->protocol = ip_proto;
3113 iph->saddr = src;
3114 iph->daddr = dst;
3115 iph->version = 0x4;
3116 iph->frag_off = 0;
3117 iph->ihl = 0x5;
3118 skb_set_transport_header(skb, skb->len);
3119
3120 switch (iph->protocol) {
3121 case IPPROTO_UDP: {
3122 struct udphdr *udph;
3123
3124 udph = skb_put_zero(skb, sizeof(struct udphdr));
3125 udph->source = sport;
3126 udph->dest = dport;
3127 udph->len = htons(sizeof(struct udphdr));
3128 udph->check = 0;
3129 break;
3130 }
3131 case IPPROTO_TCP: {
3132 struct tcphdr *tcph;
3133
3134 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3135 tcph->source = sport;
3136 tcph->dest = dport;
3137 tcph->doff = sizeof(struct tcphdr) / 4;
3138 tcph->rst = 1;
3139 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3140 src, dst, 0);
3141 break;
3142 }
3143 case IPPROTO_ICMP: {
3144 struct icmphdr *icmph;
3145
3146 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3147 icmph->type = ICMP_ECHO;
3148 icmph->code = 0;
3149 }
3150 }
3151
3152 return skb;
3153}
3154
3155static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3156 const struct nlmsghdr *nlh,
3157 struct nlattr **tb,
3158 struct netlink_ext_ack *extack)
3159{
3160 struct rtmsg *rtm;
3161 int i, err;
3162
3163 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3164 NL_SET_ERR_MSG(extack,
3165 "ipv4: Invalid header for route get request");
3166 return -EINVAL;
3167 }
3168
3169 if (!netlink_strict_get_check(skb))
3170 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3171 rtm_ipv4_policy, extack);
3172
3173 rtm = nlmsg_data(nlh);
3174 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3175 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3176 rtm->rtm_table || rtm->rtm_protocol ||
3177 rtm->rtm_scope || rtm->rtm_type) {
3178 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3179 return -EINVAL;
3180 }
3181
3182 if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3183 RTM_F_LOOKUP_TABLE |
3184 RTM_F_FIB_MATCH)) {
3185 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3186 return -EINVAL;
3187 }
3188
3189 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3190 rtm_ipv4_policy, extack);
3191 if (err)
3192 return err;
3193
3194 if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3195 (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3196 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3197 return -EINVAL;
3198 }
3199
3200 for (i = 0; i <= RTA_MAX; i++) {
3201 if (!tb[i])
3202 continue;
3203
3204 switch (i) {
3205 case RTA_IIF:
3206 case RTA_OIF:
3207 case RTA_SRC:
3208 case RTA_DST:
3209 case RTA_IP_PROTO:
3210 case RTA_SPORT:
3211 case RTA_DPORT:
3212 case RTA_MARK:
3213 case RTA_UID:
3214 break;
3215 default:
3216 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3217 return -EINVAL;
3218 }
3219 }
3220
3221 return 0;
3222}
3223
3224static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3225 struct netlink_ext_ack *extack)
3226{
3227 struct net *net = sock_net(in_skb->sk);
3228 struct nlattr *tb[RTA_MAX+1];
3229 u32 table_id = RT_TABLE_MAIN;
3230 __be16 sport = 0, dport = 0;
3231 struct fib_result res = {};
3232 u8 ip_proto = IPPROTO_UDP;
3233 struct rtable *rt = NULL;
3234 struct sk_buff *skb;
3235 struct rtmsg *rtm;
3236 struct flowi4 fl4 = {};
3237 __be32 dst = 0;
3238 __be32 src = 0;
3239 kuid_t uid;
3240 u32 iif;
3241 int err;
3242 int mark;
3243
3244 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3245 if (err < 0)
3246 return err;
3247
3248 rtm = nlmsg_data(nlh);
3249 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3250 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3251 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3252 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3253 if (tb[RTA_UID])
3254 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3255 else
3256 uid = (iif ? INVALID_UID : current_uid());
3257
3258 if (tb[RTA_IP_PROTO]) {
3259 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3260 &ip_proto, AF_INET, extack);
3261 if (err)
3262 return err;
3263 }
3264
3265 if (tb[RTA_SPORT])
3266 sport = nla_get_be16(tb[RTA_SPORT]);
3267
3268 if (tb[RTA_DPORT])
3269 dport = nla_get_be16(tb[RTA_DPORT]);
3270
3271 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3272 if (!skb)
3273 return -ENOBUFS;
3274
3275 fl4.daddr = dst;
3276 fl4.saddr = src;
3277 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3278 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3279 fl4.flowi4_mark = mark;
3280 fl4.flowi4_uid = uid;
3281 if (sport)
3282 fl4.fl4_sport = sport;
3283 if (dport)
3284 fl4.fl4_dport = dport;
3285 fl4.flowi4_proto = ip_proto;
3286
3287 rcu_read_lock();
3288
3289 if (iif) {
3290 struct net_device *dev;
3291
3292 dev = dev_get_by_index_rcu(net, iif);
3293 if (!dev) {
3294 err = -ENODEV;
3295 goto errout_rcu;
3296 }
3297
3298 fl4.flowi4_iif = iif; /* for rt_fill_info */
3299 skb->dev = dev;
3300 skb->mark = mark;
3301 err = ip_route_input_rcu(skb, dst, src,
3302 rtm->rtm_tos & IPTOS_RT_MASK, dev,
3303 &res);
3304
3305 rt = skb_rtable(skb);
3306 if (err == 0 && rt->dst.error)
3307 err = -rt->dst.error;
3308 } else {
3309 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3310 skb->dev = net->loopback_dev;
3311 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3312 err = 0;
3313 if (IS_ERR(rt))
3314 err = PTR_ERR(rt);
3315 else
3316 skb_dst_set(skb, &rt->dst);
3317 }
3318
3319 if (err)
3320 goto errout_rcu;
3321
3322 if (rtm->rtm_flags & RTM_F_NOTIFY)
3323 rt->rt_flags |= RTCF_NOTIFY;
3324
3325 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3326 table_id = res.table ? res.table->tb_id : 0;
3327
3328 /* reset skb for netlink reply msg */
3329 skb_trim(skb, 0);
3330 skb_reset_network_header(skb);
3331 skb_reset_transport_header(skb);
3332 skb_reset_mac_header(skb);
3333
3334 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3335 struct fib_rt_info fri;
3336
3337 if (!res.fi) {
3338 err = fib_props[res.type].error;
3339 if (!err)
3340 err = -EHOSTUNREACH;
3341 goto errout_rcu;
3342 }
3343 fri.fi = res.fi;
3344 fri.tb_id = table_id;
3345 fri.dst = res.prefix;
3346 fri.dst_len = res.prefixlen;
3347 fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3348 fri.type = rt->rt_type;
3349 fri.offload = 0;
3350 fri.trap = 0;
3351 fri.offload_failed = 0;
3352 if (res.fa_head) {
3353 struct fib_alias *fa;
3354
3355 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3356 u8 slen = 32 - fri.dst_len;
3357
3358 if (fa->fa_slen == slen &&
3359 fa->tb_id == fri.tb_id &&
3360 fa->fa_dscp == fri.dscp &&
3361 fa->fa_info == res.fi &&
3362 fa->fa_type == fri.type) {
3363 fri.offload = READ_ONCE(fa->offload);
3364 fri.trap = READ_ONCE(fa->trap);
3365 fri.offload_failed =
3366 READ_ONCE(fa->offload_failed);
3367 break;
3368 }
3369 }
3370 }
3371 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3372 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3373 } else {
3374 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3375 NETLINK_CB(in_skb).portid,
3376 nlh->nlmsg_seq, 0);
3377 }
3378 if (err < 0)
3379 goto errout_rcu;
3380
3381 rcu_read_unlock();
3382
3383 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3384
3385errout_free:
3386 return err;
3387errout_rcu:
3388 rcu_read_unlock();
3389 kfree_skb(skb);
3390 goto errout_free;
3391}
3392
3393void ip_rt_multicast_event(struct in_device *in_dev)
3394{
3395 rt_cache_flush(dev_net(in_dev->dev));
3396}
3397
3398#ifdef CONFIG_SYSCTL
3399static int ip_rt_gc_interval __read_mostly = 60 * HZ;
3400static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
3401static int ip_rt_gc_elasticity __read_mostly = 8;
3402static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
3403
3404static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3405 void *buffer, size_t *lenp, loff_t *ppos)
3406{
3407 struct net *net = (struct net *)__ctl->extra1;
3408
3409 if (write) {
3410 rt_cache_flush(net);
3411 fnhe_genid_bump(net);
3412 return 0;
3413 }
3414
3415 return -EINVAL;
3416}
3417
3418static struct ctl_table ipv4_route_table[] = {
3419 {
3420 .procname = "gc_thresh",
3421 .data = &ipv4_dst_ops.gc_thresh,
3422 .maxlen = sizeof(int),
3423 .mode = 0644,
3424 .proc_handler = proc_dointvec,
3425 },
3426 {
3427 .procname = "max_size",
3428 .data = &ip_rt_max_size,
3429 .maxlen = sizeof(int),
3430 .mode = 0644,
3431 .proc_handler = proc_dointvec,
3432 },
3433 {
3434 /* Deprecated. Use gc_min_interval_ms */
3435
3436 .procname = "gc_min_interval",
3437 .data = &ip_rt_gc_min_interval,
3438 .maxlen = sizeof(int),
3439 .mode = 0644,
3440 .proc_handler = proc_dointvec_jiffies,
3441 },
3442 {
3443 .procname = "gc_min_interval_ms",
3444 .data = &ip_rt_gc_min_interval,
3445 .maxlen = sizeof(int),
3446 .mode = 0644,
3447 .proc_handler = proc_dointvec_ms_jiffies,
3448 },
3449 {
3450 .procname = "gc_timeout",
3451 .data = &ip_rt_gc_timeout,
3452 .maxlen = sizeof(int),
3453 .mode = 0644,
3454 .proc_handler = proc_dointvec_jiffies,
3455 },
3456 {
3457 .procname = "gc_interval",
3458 .data = &ip_rt_gc_interval,
3459 .maxlen = sizeof(int),
3460 .mode = 0644,
3461 .proc_handler = proc_dointvec_jiffies,
3462 },
3463 {
3464 .procname = "redirect_load",
3465 .data = &ip_rt_redirect_load,
3466 .maxlen = sizeof(int),
3467 .mode = 0644,
3468 .proc_handler = proc_dointvec,
3469 },
3470 {
3471 .procname = "redirect_number",
3472 .data = &ip_rt_redirect_number,
3473 .maxlen = sizeof(int),
3474 .mode = 0644,
3475 .proc_handler = proc_dointvec,
3476 },
3477 {
3478 .procname = "redirect_silence",
3479 .data = &ip_rt_redirect_silence,
3480 .maxlen = sizeof(int),
3481 .mode = 0644,
3482 .proc_handler = proc_dointvec,
3483 },
3484 {
3485 .procname = "error_cost",
3486 .data = &ip_rt_error_cost,
3487 .maxlen = sizeof(int),
3488 .mode = 0644,
3489 .proc_handler = proc_dointvec,
3490 },
3491 {
3492 .procname = "error_burst",
3493 .data = &ip_rt_error_burst,
3494 .maxlen = sizeof(int),
3495 .mode = 0644,
3496 .proc_handler = proc_dointvec,
3497 },
3498 {
3499 .procname = "gc_elasticity",
3500 .data = &ip_rt_gc_elasticity,
3501 .maxlen = sizeof(int),
3502 .mode = 0644,
3503 .proc_handler = proc_dointvec,
3504 },
3505 { }
3506};
3507
3508static const char ipv4_route_flush_procname[] = "flush";
3509
3510static struct ctl_table ipv4_route_netns_table[] = {
3511 {
3512 .procname = ipv4_route_flush_procname,
3513 .maxlen = sizeof(int),
3514 .mode = 0200,
3515 .proc_handler = ipv4_sysctl_rtcache_flush,
3516 },
3517 {
3518 .procname = "min_pmtu",
3519 .data = &init_net.ipv4.ip_rt_min_pmtu,
3520 .maxlen = sizeof(int),
3521 .mode = 0644,
3522 .proc_handler = proc_dointvec_minmax,
3523 .extra1 = &ip_min_valid_pmtu,
3524 },
3525 {
3526 .procname = "mtu_expires",
3527 .data = &init_net.ipv4.ip_rt_mtu_expires,
3528 .maxlen = sizeof(int),
3529 .mode = 0644,
3530 .proc_handler = proc_dointvec_jiffies,
3531 },
3532 {
3533 .procname = "min_adv_mss",
3534 .data = &init_net.ipv4.ip_rt_min_advmss,
3535 .maxlen = sizeof(int),
3536 .mode = 0644,
3537 .proc_handler = proc_dointvec,
3538 },
3539 { },
3540};
3541
3542static __net_init int sysctl_route_net_init(struct net *net)
3543{
3544 struct ctl_table *tbl;
3545 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3546
3547 tbl = ipv4_route_netns_table;
3548 if (!net_eq(net, &init_net)) {
3549 int i;
3550
3551 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3552 if (!tbl)
3553 goto err_dup;
3554
3555 /* Don't export non-whitelisted sysctls to unprivileged users */
3556 if (net->user_ns != &init_user_ns) {
3557 if (tbl[0].procname != ipv4_route_flush_procname) {
3558 tbl[0].procname = NULL;
3559 table_size = 0;
3560 }
3561 }
3562
3563 /* Update the variables to point into the current struct net
3564 * except for the first element flush
3565 */
3566 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3567 tbl[i].data += (void *)net - (void *)&init_net;
3568 }
3569 tbl[0].extra1 = net;
3570
3571 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3572 tbl, table_size);
3573 if (!net->ipv4.route_hdr)
3574 goto err_reg;
3575 return 0;
3576
3577err_reg:
3578 if (tbl != ipv4_route_netns_table)
3579 kfree(tbl);
3580err_dup:
3581 return -ENOMEM;
3582}
3583
3584static __net_exit void sysctl_route_net_exit(struct net *net)
3585{
3586 struct ctl_table *tbl;
3587
3588 tbl = net->ipv4.route_hdr->ctl_table_arg;
3589 unregister_net_sysctl_table(net->ipv4.route_hdr);
3590 BUG_ON(tbl == ipv4_route_netns_table);
3591 kfree(tbl);
3592}
3593
3594static __net_initdata struct pernet_operations sysctl_route_ops = {
3595 .init = sysctl_route_net_init,
3596 .exit = sysctl_route_net_exit,
3597};
3598#endif
3599
3600static __net_init int netns_ip_rt_init(struct net *net)
3601{
3602 /* Set default value for namespaceified sysctls */
3603 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3604 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3605 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3606 return 0;
3607}
3608
3609static struct pernet_operations __net_initdata ip_rt_ops = {
3610 .init = netns_ip_rt_init,
3611};
3612
3613static __net_init int rt_genid_init(struct net *net)
3614{
3615 atomic_set(&net->ipv4.rt_genid, 0);
3616 atomic_set(&net->fnhe_genid, 0);
3617 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3618 return 0;
3619}
3620
3621static __net_initdata struct pernet_operations rt_genid_ops = {
3622 .init = rt_genid_init,
3623};
3624
3625static int __net_init ipv4_inetpeer_init(struct net *net)
3626{
3627 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3628
3629 if (!bp)
3630 return -ENOMEM;
3631 inet_peer_base_init(bp);
3632 net->ipv4.peers = bp;
3633 return 0;
3634}
3635
3636static void __net_exit ipv4_inetpeer_exit(struct net *net)
3637{
3638 struct inet_peer_base *bp = net->ipv4.peers;
3639
3640 net->ipv4.peers = NULL;
3641 inetpeer_invalidate_tree(bp);
3642 kfree(bp);
3643}
3644
3645static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3646 .init = ipv4_inetpeer_init,
3647 .exit = ipv4_inetpeer_exit,
3648};
3649
3650#ifdef CONFIG_IP_ROUTE_CLASSID
3651struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3652#endif /* CONFIG_IP_ROUTE_CLASSID */
3653
3654int __init ip_rt_init(void)
3655{
3656 void *idents_hash;
3657 int cpu;
3658
3659 /* For modern hosts, this will use 2 MB of memory */
3660 idents_hash = alloc_large_system_hash("IP idents",
3661 sizeof(*ip_idents) + sizeof(*ip_tstamps),
3662 0,
3663 16, /* one bucket per 64 KB */
3664 HASH_ZERO,
3665 NULL,
3666 &ip_idents_mask,
3667 2048,
3668 256*1024);
3669
3670 ip_idents = idents_hash;
3671
3672 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3673
3674 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3675
3676 for_each_possible_cpu(cpu) {
3677 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3678
3679 INIT_LIST_HEAD(&ul->head);
3680 INIT_LIST_HEAD(&ul->quarantine);
3681 spin_lock_init(&ul->lock);
3682 }
3683#ifdef CONFIG_IP_ROUTE_CLASSID
3684 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3685 if (!ip_rt_acct)
3686 panic("IP: failed to allocate ip_rt_acct\n");
3687#endif
3688
3689 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3690 SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3691
3692 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3693
3694 if (dst_entries_init(&ipv4_dst_ops) < 0)
3695 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3696
3697 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3698 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3699
3700 ipv4_dst_ops.gc_thresh = ~0;
3701 ip_rt_max_size = INT_MAX;
3702
3703 devinet_init();
3704 ip_fib_init();
3705
3706 if (ip_rt_proc_init())
3707 pr_err("Unable to create route proc files\n");
3708#ifdef CONFIG_XFRM
3709 xfrm_init();
3710 xfrm4_init();
3711#endif
3712 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3713 RTNL_FLAG_DOIT_UNLOCKED);
3714
3715#ifdef CONFIG_SYSCTL
3716 register_pernet_subsys(&sysctl_route_ops);
3717#endif
3718 register_pernet_subsys(&ip_rt_ops);
3719 register_pernet_subsys(&rt_genid_ops);
3720 register_pernet_subsys(&ipv4_inetpeer_ops);
3721 return 0;
3722}
3723
3724#ifdef CONFIG_SYSCTL
3725/*
3726 * We really need to sanitize the damn ipv4 init order, then all
3727 * this nonsense will go away.
3728 */
3729void __init ip_static_sysctl_init(void)
3730{
3731 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3732}
3733#endif