?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / net / ipv4 / route.c
blob6b95f74a91d3f87c5d6b4f2bcef7dc35b671882d
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 #include <linux/module.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <linux/bitops.h>
69 #include <linux/types.h>
70 #include <linux/kernel.h>
71 #include <linux/mm.h>
72 #include <linux/bootmem.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
77 #include <linux/in.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/workqueue.h>
83 #include <linux/skbuff.h>
84 #include <linux/inetdevice.h>
85 #include <linux/igmp.h>
86 #include <linux/pkt_sched.h>
87 #include <linux/mroute.h>
88 #include <linux/netfilter_ipv4.h>
89 #include <linux/random.h>
90 #include <linux/jhash.h>
91 #include <linux/rcupdate.h>
92 #include <linux/times.h>
93 #include <linux/slab.h>
94 #include <linux/prefetch.h>
95 #include <net/dst.h>
96 #include <net/net_namespace.h>
97 #include <net/protocol.h>
98 #include <net/ip.h>
99 #include <net/route.h>
100 #include <net/inetpeer.h>
101 #include <net/sock.h>
102 #include <net/ip_fib.h>
103 #include <net/arp.h>
104 #include <net/tcp.h>
105 #include <net/icmp.h>
106 #include <net/xfrm.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
109 #ifdef CONFIG_SYSCTL
110 #include <linux/sysctl.h>
111 #endif
112 #include <net/secure_seq.h>
113
114 #define RT_FL_TOS(oldflp4) \
115 ((u32)(oldflp4->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
116
117 #define IP_MAX_MTU 0xFFF0
118
119 #define RT_GC_TIMEOUT (300*HZ)
120
121 static int ip_rt_max_size;
122 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
123 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
124 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
125 static int ip_rt_redirect_number __read_mostly = 9;
126 static int ip_rt_redirect_load __read_mostly = HZ / 50;
127 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
128 static int ip_rt_error_cost __read_mostly = HZ;
129 static int ip_rt_error_burst __read_mostly = 5 * HZ;
130 static int ip_rt_gc_elasticity __read_mostly = 8;
131 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
132 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
133 static int ip_rt_min_advmss __read_mostly = 256;
134 static int rt_chain_length_max __read_mostly = 20;
135
136 static struct delayed_work expires_work;
137 static unsigned long expires_ljiffies;
138
139 /*
140 * Interface to generic destination cache.
141 */
142
143 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
144 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
145 static unsigned int ipv4_default_mtu(const struct dst_entry *dst);
146 static void ipv4_dst_destroy(struct dst_entry *dst);
147 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
148 static void ipv4_link_failure(struct sk_buff *skb);
149 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
150 static int rt_garbage_collect(struct dst_ops *ops);
151
152 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
153 int how)
154 {
155 }
156
157 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
158 {
159 struct rtable *rt = (struct rtable *) dst;
160 struct inet_peer *peer;
161 u32 *p = NULL;
162
163 if (!rt->peer)
164 rt_bind_peer(rt, rt->rt_dst, 1);
165
166 peer = rt->peer;
167 if (peer) {
168 u32 *old_p = __DST_METRICS_PTR(old);
169 unsigned long prev, new;
170
171 p = peer->metrics;
172 if (inet_metrics_new(peer))
173 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
174
175 new = (unsigned long) p;
176 prev = cmpxchg(&dst->_metrics, old, new);
177
178 if (prev != old) {
179 p = __DST_METRICS_PTR(prev);
180 if (prev & DST_METRICS_READ_ONLY)
181 p = NULL;
182 } else {
183 if (rt->fi) {
184 fib_info_put(rt->fi);
185 rt->fi = NULL;
186 }
187 }
188 }
189 return p;
190 }
191
192 static struct dst_ops ipv4_dst_ops = {
193 .family = AF_INET,
194 .protocol = cpu_to_be16(ETH_P_IP),
195 .gc = rt_garbage_collect,
196 .check = ipv4_dst_check,
197 .default_advmss = ipv4_default_advmss,
198 .default_mtu = ipv4_default_mtu,
199 .cow_metrics = ipv4_cow_metrics,
200 .destroy = ipv4_dst_destroy,
201 .ifdown = ipv4_dst_ifdown,
202 .negative_advice = ipv4_negative_advice,
203 .link_failure = ipv4_link_failure,
204 .update_pmtu = ip_rt_update_pmtu,
205 .local_out = __ip_local_out,
206 };
207
208 #define ECN_OR_COST(class) TC_PRIO_##class
209
210 const __u8 ip_tos2prio[16] = {
211 TC_PRIO_BESTEFFORT,
212 ECN_OR_COST(BESTEFFORT),
213 TC_PRIO_BESTEFFORT,
214 ECN_OR_COST(BESTEFFORT),
215 TC_PRIO_BULK,
216 ECN_OR_COST(BULK),
217 TC_PRIO_BULK,
218 ECN_OR_COST(BULK),
219 TC_PRIO_INTERACTIVE,
220 ECN_OR_COST(INTERACTIVE),
221 TC_PRIO_INTERACTIVE,
222 ECN_OR_COST(INTERACTIVE),
223 TC_PRIO_INTERACTIVE_BULK,
224 ECN_OR_COST(INTERACTIVE_BULK),
225 TC_PRIO_INTERACTIVE_BULK,
226 ECN_OR_COST(INTERACTIVE_BULK)
227 };
228
229
230 /*
231 * Route cache.
232 */
233
234 /* The locking scheme is rather straight forward:
235 *
236 * 1) Read-Copy Update protects the buckets of the central route hash.
237 * 2) Only writers remove entries, and they hold the lock
238 * as they look at rtable reference counts.
239 * 3) Only readers acquire references to rtable entries,
240 * they do so with atomic increments and with the
241 * lock held.
242 */
243
244 struct rt_hash_bucket {
245 struct rtable __rcu *chain;
246 };
247
248 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
249 defined(CONFIG_PROVE_LOCKING)
250 /*
251 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
252 * The size of this table is a power of two and depends on the number of CPUS.
253 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
254 */
255 #ifdef CONFIG_LOCKDEP
256 # define RT_HASH_LOCK_SZ 256
257 #else
258 # if NR_CPUS >= 32
259 # define RT_HASH_LOCK_SZ 4096
260 # elif NR_CPUS >= 16
261 # define RT_HASH_LOCK_SZ 2048
262 # elif NR_CPUS >= 8
263 # define RT_HASH_LOCK_SZ 1024
264 # elif NR_CPUS >= 4
265 # define RT_HASH_LOCK_SZ 512
266 # else
267 # define RT_HASH_LOCK_SZ 256
268 # endif
269 #endif
270
271 static spinlock_t *rt_hash_locks;
272 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
273
274 static __init void rt_hash_lock_init(void)
275 {
276 int i;
277
278 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
279 GFP_KERNEL);
280 if (!rt_hash_locks)
281 panic("IP: failed to allocate rt_hash_locks\n");
282
283 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
284 spin_lock_init(&rt_hash_locks[i]);
285 }
286 #else
287 # define rt_hash_lock_addr(slot) NULL
288
289 static inline void rt_hash_lock_init(void)
290 {
291 }
292 #endif
293
294 static struct rt_hash_bucket *rt_hash_table __read_mostly;
295 static unsigned rt_hash_mask __read_mostly;
296 static unsigned int rt_hash_log __read_mostly;
297
298 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
299 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
300
301 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
302 int genid)
303 {
304 return jhash_3words((__force u32)daddr, (__force u32)saddr,
305 idx, genid)
306 & rt_hash_mask;
307 }
308
309 static inline int rt_genid(struct net *net)
310 {
311 return atomic_read(&net->ipv4.rt_genid);
312 }
313
314 #ifdef CONFIG_PROC_FS
315 struct rt_cache_iter_state {
316 struct seq_net_private p;
317 int bucket;
318 int genid;
319 };
320
321 static struct rtable *rt_cache_get_first(struct seq_file *seq)
322 {
323 struct rt_cache_iter_state *st = seq->private;
324 struct rtable *r = NULL;
325
326 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
327 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
328 continue;
329 rcu_read_lock_bh();
330 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
331 while (r) {
332 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
333 r->rt_genid == st->genid)
334 return r;
335 r = rcu_dereference_bh(r->dst.rt_next);
336 }
337 rcu_read_unlock_bh();
338 }
339 return r;
340 }
341
342 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
343 struct rtable *r)
344 {
345 struct rt_cache_iter_state *st = seq->private;
346
347 r = rcu_dereference_bh(r->dst.rt_next);
348 while (!r) {
349 rcu_read_unlock_bh();
350 do {
351 if (--st->bucket < 0)
352 return NULL;
353 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
354 rcu_read_lock_bh();
355 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
356 }
357 return r;
358 }
359
360 static struct rtable *rt_cache_get_next(struct seq_file *seq,
361 struct rtable *r)
362 {
363 struct rt_cache_iter_state *st = seq->private;
364 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
365 if (dev_net(r->dst.dev) != seq_file_net(seq))
366 continue;
367 if (r->rt_genid == st->genid)
368 break;
369 }
370 return r;
371 }
372
373 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
374 {
375 struct rtable *r = rt_cache_get_first(seq);
376
377 if (r)
378 while (pos && (r = rt_cache_get_next(seq, r)))
379 --pos;
380 return pos ? NULL : r;
381 }
382
383 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
384 {
385 struct rt_cache_iter_state *st = seq->private;
386 if (*pos)
387 return rt_cache_get_idx(seq, *pos - 1);
388 st->genid = rt_genid(seq_file_net(seq));
389 return SEQ_START_TOKEN;
390 }
391
392 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
393 {
394 struct rtable *r;
395
396 if (v == SEQ_START_TOKEN)
397 r = rt_cache_get_first(seq);
398 else
399 r = rt_cache_get_next(seq, v);
400 ++*pos;
401 return r;
402 }
403
404 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
405 {
406 if (v && v != SEQ_START_TOKEN)
407 rcu_read_unlock_bh();
408 }
409
410 static int rt_cache_seq_show(struct seq_file *seq, void *v)
411 {
412 if (v == SEQ_START_TOKEN)
413 seq_printf(seq, "%-127s\n",
414 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
415 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
416 "HHUptod\tSpecDst");
417 else {
418 struct rtable *r = v;
419 struct neighbour *n;
420 int len, HHUptod;
421
422 rcu_read_lock();
423 n = dst_get_neighbour(&r->dst);
424 HHUptod = (n && (n->nud_state & NUD_CONNECTED)) ? 1 : 0;
425 rcu_read_unlock();
426
427 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
428 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
429 r->dst.dev ? r->dst.dev->name : "*",
430 (__force u32)r->rt_dst,
431 (__force u32)r->rt_gateway,
432 r->rt_flags, atomic_read(&r->dst.__refcnt),
433 r->dst.__use, 0, (__force u32)r->rt_src,
434 dst_metric_advmss(&r->dst) + 40,
435 dst_metric(&r->dst, RTAX_WINDOW),
436 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
437 dst_metric(&r->dst, RTAX_RTTVAR)),
438 r->rt_key_tos,
439 r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
440 HHUptod,
441 r->rt_spec_dst, &len);
442
443 seq_printf(seq, "%*s\n", 127 - len, "");
444 }
445 return 0;
446 }
447
448 static const struct seq_operations rt_cache_seq_ops = {
449 .start = rt_cache_seq_start,
450 .next = rt_cache_seq_next,
451 .stop = rt_cache_seq_stop,
452 .show = rt_cache_seq_show,
453 };
454
455 static int rt_cache_seq_open(struct inode *inode, struct file *file)
456 {
457 return seq_open_net(inode, file, &rt_cache_seq_ops,
458 sizeof(struct rt_cache_iter_state));
459 }
460
461 static const struct file_operations rt_cache_seq_fops = {
462 .owner = THIS_MODULE,
463 .open = rt_cache_seq_open,
464 .read = seq_read,
465 .llseek = seq_lseek,
466 .release = seq_release_net,
467 };
468
469
470 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
471 {
472 int cpu;
473
474 if (*pos == 0)
475 return SEQ_START_TOKEN;
476
477 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
478 if (!cpu_possible(cpu))
479 continue;
480 *pos = cpu+1;
481 return &per_cpu(rt_cache_stat, cpu);
482 }
483 return NULL;
484 }
485
486 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
487 {
488 int cpu;
489
490 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
491 if (!cpu_possible(cpu))
492 continue;
493 *pos = cpu+1;
494 return &per_cpu(rt_cache_stat, cpu);
495 }
496 return NULL;
497
498 }
499
500 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
501 {
502
503 }
504
505 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
506 {
507 struct rt_cache_stat *st = v;
508
509 if (v == SEQ_START_TOKEN) {
510 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");
511 return 0;
512 }
513
514 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
515 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
516 dst_entries_get_slow(&ipv4_dst_ops),
517 st->in_hit,
518 st->in_slow_tot,
519 st->in_slow_mc,
520 st->in_no_route,
521 st->in_brd,
522 st->in_martian_dst,
523 st->in_martian_src,
524
525 st->out_hit,
526 st->out_slow_tot,
527 st->out_slow_mc,
528
529 st->gc_total,
530 st->gc_ignored,
531 st->gc_goal_miss,
532 st->gc_dst_overflow,
533 st->in_hlist_search,
534 st->out_hlist_search
535 );
536 return 0;
537 }
538
539 static const struct seq_operations rt_cpu_seq_ops = {
540 .start = rt_cpu_seq_start,
541 .next = rt_cpu_seq_next,
542 .stop = rt_cpu_seq_stop,
543 .show = rt_cpu_seq_show,
544 };
545
546
547 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
548 {
549 return seq_open(file, &rt_cpu_seq_ops);
550 }
551
552 static const struct file_operations rt_cpu_seq_fops = {
553 .owner = THIS_MODULE,
554 .open = rt_cpu_seq_open,
555 .read = seq_read,
556 .llseek = seq_lseek,
557 .release = seq_release,
558 };
559
560 #ifdef CONFIG_IP_ROUTE_CLASSID
561 static int rt_acct_proc_show(struct seq_file *m, void *v)
562 {
563 struct ip_rt_acct *dst, *src;
564 unsigned int i, j;
565
566 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
567 if (!dst)
568 return -ENOMEM;
569
570 for_each_possible_cpu(i) {
571 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
572 for (j = 0; j < 256; j++) {
573 dst[j].o_bytes += src[j].o_bytes;
574 dst[j].o_packets += src[j].o_packets;
575 dst[j].i_bytes += src[j].i_bytes;
576 dst[j].i_packets += src[j].i_packets;
577 }
578 }
579
580 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
581 kfree(dst);
582 return 0;
583 }
584
585 static int rt_acct_proc_open(struct inode *inode, struct file *file)
586 {
587 return single_open(file, rt_acct_proc_show, NULL);
588 }
589
590 static const struct file_operations rt_acct_proc_fops = {
591 .owner = THIS_MODULE,
592 .open = rt_acct_proc_open,
593 .read = seq_read,
594 .llseek = seq_lseek,
595 .release = single_release,
596 };
597 #endif
598
599 static int __net_init ip_rt_do_proc_init(struct net *net)
600 {
601 struct proc_dir_entry *pde;
602
603 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
604 &rt_cache_seq_fops);
605 if (!pde)
606 goto err1;
607
608 pde = proc_create("rt_cache", S_IRUGO,
609 net->proc_net_stat, &rt_cpu_seq_fops);
610 if (!pde)
611 goto err2;
612
613 #ifdef CONFIG_IP_ROUTE_CLASSID
614 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
615 if (!pde)
616 goto err3;
617 #endif
618 return 0;
619
620 #ifdef CONFIG_IP_ROUTE_CLASSID
621 err3:
622 remove_proc_entry("rt_cache", net->proc_net_stat);
623 #endif
624 err2:
625 remove_proc_entry("rt_cache", net->proc_net);
626 err1:
627 return -ENOMEM;
628 }
629
630 static void __net_exit ip_rt_do_proc_exit(struct net *net)
631 {
632 remove_proc_entry("rt_cache", net->proc_net_stat);
633 remove_proc_entry("rt_cache", net->proc_net);
634 #ifdef CONFIG_IP_ROUTE_CLASSID
635 remove_proc_entry("rt_acct", net->proc_net);
636 #endif
637 }
638
639 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
640 .init = ip_rt_do_proc_init,
641 .exit = ip_rt_do_proc_exit,
642 };
643
644 static int __init ip_rt_proc_init(void)
645 {
646 return register_pernet_subsys(&ip_rt_proc_ops);
647 }
648
649 #else
650 static inline int ip_rt_proc_init(void)
651 {
652 return 0;
653 }
654 #endif /* CONFIG_PROC_FS */
655
656 static inline void rt_free(struct rtable *rt)
657 {
658 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
659 }
660
661 static inline void rt_drop(struct rtable *rt)
662 {
663 ip_rt_put(rt);
664 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
665 }
666
667 static inline int rt_fast_clean(struct rtable *rth)
668 {
669 /* Kill broadcast/multicast entries very aggresively, if they
670 collide in hash table with more useful entries */
671 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
672 rt_is_input_route(rth) && rth->dst.rt_next;
673 }
674
675 static inline int rt_valuable(struct rtable *rth)
676 {
677 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
678 (rth->peer && rth->peer->pmtu_expires);
679 }
680
681 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
682 {
683 unsigned long age;
684 int ret = 0;
685
686 if (atomic_read(&rth->dst.__refcnt))
687 goto out;
688
689 age = jiffies - rth->dst.lastuse;
690 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
691 (age <= tmo2 && rt_valuable(rth)))
692 goto out;
693 ret = 1;
694 out: return ret;
695 }
696
697 /* Bits of score are:
698 * 31: very valuable
699 * 30: not quite useless
700 * 29..0: usage counter
701 */
702 static inline u32 rt_score(struct rtable *rt)
703 {
704 u32 score = jiffies - rt->dst.lastuse;
705
706 score = ~score & ~(3<<30);
707
708 if (rt_valuable(rt))
709 score |= (1<<31);
710
711 if (rt_is_output_route(rt) ||
712 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
713 score |= (1<<30);
714
715 return score;
716 }
717
718 static inline bool rt_caching(const struct net *net)
719 {
720 return net->ipv4.current_rt_cache_rebuild_count <=
721 net->ipv4.sysctl_rt_cache_rebuild_count;
722 }
723
724 static inline bool compare_hash_inputs(const struct rtable *rt1,
725 const struct rtable *rt2)
726 {
727 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
728 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
729 (rt1->rt_route_iif ^ rt2->rt_route_iif)) == 0);
730 }
731
732 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
733 {
734 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
735 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
736 (rt1->rt_mark ^ rt2->rt_mark) |
737 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
738 (rt1->rt_route_iif ^ rt2->rt_route_iif) |
739 (rt1->rt_oif ^ rt2->rt_oif)) == 0;
740 }
741
742 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
743 {
744 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
745 }
746
747 static inline int rt_is_expired(struct rtable *rth)
748 {
749 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
750 }
751
752 /*
753 * Perform a full scan of hash table and free all entries.
754 * Can be called by a softirq or a process.
755 * In the later case, we want to be reschedule if necessary
756 */
757 static void rt_do_flush(struct net *net, int process_context)
758 {
759 unsigned int i;
760 struct rtable *rth, *next;
761
762 for (i = 0; i <= rt_hash_mask; i++) {
763 struct rtable __rcu **pprev;
764 struct rtable *list;
765
766 if (process_context && need_resched())
767 cond_resched();
768 rth = rcu_dereference_raw(rt_hash_table[i].chain);
769 if (!rth)
770 continue;
771
772 spin_lock_bh(rt_hash_lock_addr(i));
773
774 list = NULL;
775 pprev = &rt_hash_table[i].chain;
776 rth = rcu_dereference_protected(*pprev,
777 lockdep_is_held(rt_hash_lock_addr(i)));
778
779 while (rth) {
780 next = rcu_dereference_protected(rth->dst.rt_next,
781 lockdep_is_held(rt_hash_lock_addr(i)));
782
783 if (!net ||
784 net_eq(dev_net(rth->dst.dev), net)) {
785 rcu_assign_pointer(*pprev, next);
786 rcu_assign_pointer(rth->dst.rt_next, list);
787 list = rth;
788 } else {
789 pprev = &rth->dst.rt_next;
790 }
791 rth = next;
792 }
793
794 spin_unlock_bh(rt_hash_lock_addr(i));
795
796 for (; list; list = next) {
797 next = rcu_dereference_protected(list->dst.rt_next, 1);
798 rt_free(list);
799 }
800 }
801 }
802
803 /*
804 * While freeing expired entries, we compute average chain length
805 * and standard deviation, using fixed-point arithmetic.
806 * This to have an estimation of rt_chain_length_max
807 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
808 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
809 */
810
811 #define FRACT_BITS 3
812 #define ONE (1UL << FRACT_BITS)
813
814 /*
815 * Given a hash chain and an item in this hash chain,
816 * find if a previous entry has the same hash_inputs
817 * (but differs on tos, mark or oif)
818 * Returns 0 if an alias is found.
819 * Returns ONE if rth has no alias before itself.
820 */
821 static int has_noalias(const struct rtable *head, const struct rtable *rth)
822 {
823 const struct rtable *aux = head;
824
825 while (aux != rth) {
826 if (compare_hash_inputs(aux, rth))
827 return 0;
828 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
829 }
830 return ONE;
831 }
832
833 static void rt_check_expire(void)
834 {
835 static unsigned int rover;
836 unsigned int i = rover, goal;
837 struct rtable *rth;
838 struct rtable __rcu **rthp;
839 unsigned long samples = 0;
840 unsigned long sum = 0, sum2 = 0;
841 unsigned long delta;
842 u64 mult;
843
844 delta = jiffies - expires_ljiffies;
845 expires_ljiffies = jiffies;
846 mult = ((u64)delta) << rt_hash_log;
847 if (ip_rt_gc_timeout > 1)
848 do_div(mult, ip_rt_gc_timeout);
849 goal = (unsigned int)mult;
850 if (goal > rt_hash_mask)
851 goal = rt_hash_mask + 1;
852 for (; goal > 0; goal--) {
853 unsigned long tmo = ip_rt_gc_timeout;
854 unsigned long length;
855
856 i = (i + 1) & rt_hash_mask;
857 rthp = &rt_hash_table[i].chain;
858
859 if (need_resched())
860 cond_resched();
861
862 samples++;
863
864 if (rcu_dereference_raw(*rthp) == NULL)
865 continue;
866 length = 0;
867 spin_lock_bh(rt_hash_lock_addr(i));
868 while ((rth = rcu_dereference_protected(*rthp,
869 lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
870 prefetch(rth->dst.rt_next);
871 if (rt_is_expired(rth)) {
872 *rthp = rth->dst.rt_next;
873 rt_free(rth);
874 continue;
875 }
876 if (rth->dst.expires) {
877 /* Entry is expired even if it is in use */
878 if (time_before_eq(jiffies, rth->dst.expires)) {
879 nofree:
880 tmo >>= 1;
881 rthp = &rth->dst.rt_next;
882 /*
883 * We only count entries on
884 * a chain with equal hash inputs once
885 * so that entries for different QOS
886 * levels, and other non-hash input
887 * attributes don't unfairly skew
888 * the length computation
889 */
890 length += has_noalias(rt_hash_table[i].chain, rth);
891 continue;
892 }
893 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
894 goto nofree;
895
896 /* Cleanup aged off entries. */
897 *rthp = rth->dst.rt_next;
898 rt_free(rth);
899 }
900 spin_unlock_bh(rt_hash_lock_addr(i));
901 sum += length;
902 sum2 += length*length;
903 }
904 if (samples) {
905 unsigned long avg = sum / samples;
906 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
907 rt_chain_length_max = max_t(unsigned long,
908 ip_rt_gc_elasticity,
909 (avg + 4*sd) >> FRACT_BITS);
910 }
911 rover = i;
912 }
913
914 /*
915 * rt_worker_func() is run in process context.
916 * we call rt_check_expire() to scan part of the hash table
917 */
918 static void rt_worker_func(struct work_struct *work)
919 {
920 rt_check_expire();
921 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
922 }
923
924 /*
925 * Perturbation of rt_genid by a small quantity [1..256]
926 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
927 * many times (2^24) without giving recent rt_genid.
928 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
929 */
930 static void rt_cache_invalidate(struct net *net)
931 {
932 unsigned char shuffle;
933
934 get_random_bytes(&shuffle, sizeof(shuffle));
935 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
936 }
937
938 /*
939 * delay < 0 : invalidate cache (fast : entries will be deleted later)
940 * delay >= 0 : invalidate & flush cache (can be long)
941 */
942 void rt_cache_flush(struct net *net, int delay)
943 {
944 rt_cache_invalidate(net);
945 if (delay >= 0)
946 rt_do_flush(net, !in_softirq());
947 }
948
949 /* Flush previous cache invalidated entries from the cache */
950 void rt_cache_flush_batch(struct net *net)
951 {
952 rt_do_flush(net, !in_softirq());
953 }
954
955 static void rt_emergency_hash_rebuild(struct net *net)
956 {
957 if (net_ratelimit())
958 printk(KERN_WARNING "Route hash chain too long!\n");
959 rt_cache_invalidate(net);
960 }
961
962 /*
963 Short description of GC goals.
964
965 We want to build algorithm, which will keep routing cache
966 at some equilibrium point, when number of aged off entries
967 is kept approximately equal to newly generated ones.
968
969 Current expiration strength is variable "expire".
970 We try to adjust it dynamically, so that if networking
971 is idle expires is large enough to keep enough of warm entries,
972 and when load increases it reduces to limit cache size.
973 */
974
975 static int rt_garbage_collect(struct dst_ops *ops)
976 {
977 static unsigned long expire = RT_GC_TIMEOUT;
978 static unsigned long last_gc;
979 static int rover;
980 static int equilibrium;
981 struct rtable *rth;
982 struct rtable __rcu **rthp;
983 unsigned long now = jiffies;
984 int goal;
985 int entries = dst_entries_get_fast(&ipv4_dst_ops);
986
987 /*
988 * Garbage collection is pretty expensive,
989 * do not make it too frequently.
990 */
991
992 RT_CACHE_STAT_INC(gc_total);
993
994 if (now - last_gc < ip_rt_gc_min_interval &&
995 entries < ip_rt_max_size) {
996 RT_CACHE_STAT_INC(gc_ignored);
997 goto out;
998 }
999
1000 entries = dst_entries_get_slow(&ipv4_dst_ops);
1001 /* Calculate number of entries, which we want to expire now. */
1002 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
1003 if (goal <= 0) {
1004 if (equilibrium < ipv4_dst_ops.gc_thresh)
1005 equilibrium = ipv4_dst_ops.gc_thresh;
1006 goal = entries - equilibrium;
1007 if (goal > 0) {
1008 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1009 goal = entries - equilibrium;
1010 }
1011 } else {
1012 /* We are in dangerous area. Try to reduce cache really
1013 * aggressively.
1014 */
1015 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
1016 equilibrium = entries - goal;
1017 }
1018
1019 if (now - last_gc >= ip_rt_gc_min_interval)
1020 last_gc = now;
1021
1022 if (goal <= 0) {
1023 equilibrium += goal;
1024 goto work_done;
1025 }
1026
1027 do {
1028 int i, k;
1029
1030 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1031 unsigned long tmo = expire;
1032
1033 k = (k + 1) & rt_hash_mask;
1034 rthp = &rt_hash_table[k].chain;
1035 spin_lock_bh(rt_hash_lock_addr(k));
1036 while ((rth = rcu_dereference_protected(*rthp,
1037 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1038 if (!rt_is_expired(rth) &&
1039 !rt_may_expire(rth, tmo, expire)) {
1040 tmo >>= 1;
1041 rthp = &rth->dst.rt_next;
1042 continue;
1043 }
1044 *rthp = rth->dst.rt_next;
1045 rt_free(rth);
1046 goal--;
1047 }
1048 spin_unlock_bh(rt_hash_lock_addr(k));
1049 if (goal <= 0)
1050 break;
1051 }
1052 rover = k;
1053
1054 if (goal <= 0)
1055 goto work_done;
1056
1057 /* Goal is not achieved. We stop process if:
1058
1059 - if expire reduced to zero. Otherwise, expire is halfed.
1060 - if table is not full.
1061 - if we are called from interrupt.
1062 - jiffies check is just fallback/debug loop breaker.
1063 We will not spin here for long time in any case.
1064 */
1065
1066 RT_CACHE_STAT_INC(gc_goal_miss);
1067
1068 if (expire == 0)
1069 break;
1070
1071 expire >>= 1;
1072
1073 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1074 goto out;
1075 } while (!in_softirq() && time_before_eq(jiffies, now));
1076
1077 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1078 goto out;
1079 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1080 goto out;
1081 if (net_ratelimit())
1082 printk(KERN_WARNING "dst cache overflow\n");
1083 RT_CACHE_STAT_INC(gc_dst_overflow);
1084 return 1;
1085
1086 work_done:
1087 expire += ip_rt_gc_min_interval;
1088 if (expire > ip_rt_gc_timeout ||
1089 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1090 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1091 expire = ip_rt_gc_timeout;
1092 out: return 0;
1093 }
1094
1095 /*
1096 * Returns number of entries in a hash chain that have different hash_inputs
1097 */
1098 static int slow_chain_length(const struct rtable *head)
1099 {
1100 int length = 0;
1101 const struct rtable *rth = head;
1102
1103 while (rth) {
1104 length += has_noalias(head, rth);
1105 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1106 }
1107 return length >> FRACT_BITS;
1108 }
1109
1110 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt,
1111 struct sk_buff *skb, int ifindex)
1112 {
1113 struct rtable *rth, *cand;
1114 struct rtable __rcu **rthp, **candp;
1115 unsigned long now;
1116 u32 min_score;
1117 int chain_length;
1118 int attempts = !in_softirq();
1119
1120 restart:
1121 chain_length = 0;
1122 min_score = ~(u32)0;
1123 cand = NULL;
1124 candp = NULL;
1125 now = jiffies;
1126
1127 if (!rt_caching(dev_net(rt->dst.dev))) {
1128 /*
1129 * If we're not caching, just tell the caller we
1130 * were successful and don't touch the route. The
1131 * caller hold the sole reference to the cache entry, and
1132 * it will be released when the caller is done with it.
1133 * If we drop it here, the callers have no way to resolve routes
1134 * when we're not caching. Instead, just point *rp at rt, so
1135 * the caller gets a single use out of the route
1136 * Note that we do rt_free on this new route entry, so that
1137 * once its refcount hits zero, we are still able to reap it
1138 * (Thanks Alexey)
1139 * Note: To avoid expensive rcu stuff for this uncached dst,
1140 * we set DST_NOCACHE so that dst_release() can free dst without
1141 * waiting a grace period.
1142 */
1143
1144 rt->dst.flags |= DST_NOCACHE;
1145 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1146 int err = arp_bind_neighbour(&rt->dst);
1147 if (err) {
1148 if (net_ratelimit())
1149 printk(KERN_WARNING
1150 "Neighbour table failure & not caching routes.\n");
1151 ip_rt_put(rt);
1152 return ERR_PTR(err);
1153 }
1154 }
1155
1156 goto skip_hashing;
1157 }
1158
1159 rthp = &rt_hash_table[hash].chain;
1160
1161 spin_lock_bh(rt_hash_lock_addr(hash));
1162 while ((rth = rcu_dereference_protected(*rthp,
1163 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1164 if (rt_is_expired(rth)) {
1165 *rthp = rth->dst.rt_next;
1166 rt_free(rth);
1167 continue;
1168 }
1169 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1170 /* Put it first */
1171 *rthp = rth->dst.rt_next;
1172 /*
1173 * Since lookup is lockfree, the deletion
1174 * must be visible to another weakly ordered CPU before
1175 * the insertion at the start of the hash chain.
1176 */
1177 rcu_assign_pointer(rth->dst.rt_next,
1178 rt_hash_table[hash].chain);
1179 /*
1180 * Since lookup is lockfree, the update writes
1181 * must be ordered for consistency on SMP.
1182 */
1183 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1184
1185 dst_use(&rth->dst, now);
1186 spin_unlock_bh(rt_hash_lock_addr(hash));
1187
1188 rt_drop(rt);
1189 if (skb)
1190 skb_dst_set(skb, &rth->dst);
1191 return rth;
1192 }
1193
1194 if (!atomic_read(&rth->dst.__refcnt)) {
1195 u32 score = rt_score(rth);
1196
1197 if (score <= min_score) {
1198 cand = rth;
1199 candp = rthp;
1200 min_score = score;
1201 }
1202 }
1203
1204 chain_length++;
1205
1206 rthp = &rth->dst.rt_next;
1207 }
1208
1209 if (cand) {
1210 /* ip_rt_gc_elasticity used to be average length of chain
1211 * length, when exceeded gc becomes really aggressive.
1212 *
1213 * The second limit is less certain. At the moment it allows
1214 * only 2 entries per bucket. We will see.
1215 */
1216 if (chain_length > ip_rt_gc_elasticity) {
1217 *candp = cand->dst.rt_next;
1218 rt_free(cand);
1219 }
1220 } else {
1221 if (chain_length > rt_chain_length_max &&
1222 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1223 struct net *net = dev_net(rt->dst.dev);
1224 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1225 if (!rt_caching(net)) {
1226 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1227 rt->dst.dev->name, num);
1228 }
1229 rt_emergency_hash_rebuild(net);
1230 spin_unlock_bh(rt_hash_lock_addr(hash));
1231
1232 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1233 ifindex, rt_genid(net));
1234 goto restart;
1235 }
1236 }
1237
1238 /* Try to bind route to arp only if it is output
1239 route or unicast forwarding path.
1240 */
1241 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1242 int err = arp_bind_neighbour(&rt->dst);
1243 if (err) {
1244 spin_unlock_bh(rt_hash_lock_addr(hash));
1245
1246 if (err != -ENOBUFS) {
1247 rt_drop(rt);
1248 return ERR_PTR(err);
1249 }
1250
1251 /* Neighbour tables are full and nothing
1252 can be released. Try to shrink route cache,
1253 it is most likely it holds some neighbour records.
1254 */
1255 if (attempts-- > 0) {
1256 int saved_elasticity = ip_rt_gc_elasticity;
1257 int saved_int = ip_rt_gc_min_interval;
1258 ip_rt_gc_elasticity = 1;
1259 ip_rt_gc_min_interval = 0;
1260 rt_garbage_collect(&ipv4_dst_ops);
1261 ip_rt_gc_min_interval = saved_int;
1262 ip_rt_gc_elasticity = saved_elasticity;
1263 goto restart;
1264 }
1265
1266 if (net_ratelimit())
1267 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1268 rt_drop(rt);
1269 return ERR_PTR(-ENOBUFS);
1270 }
1271 }
1272
1273 rt->dst.rt_next = rt_hash_table[hash].chain;
1274
1275 /*
1276 * Since lookup is lockfree, we must make sure
1277 * previous writes to rt are committed to memory
1278 * before making rt visible to other CPUS.
1279 */
1280 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1281
1282 spin_unlock_bh(rt_hash_lock_addr(hash));
1283
1284 skip_hashing:
1285 if (skb)
1286 skb_dst_set(skb, &rt->dst);
1287 return rt;
1288 }
1289
1290 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1291
1292 static u32 rt_peer_genid(void)
1293 {
1294 return atomic_read(&__rt_peer_genid);
1295 }
1296
1297 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1298 {
1299 struct inet_peer *peer;
1300
1301 peer = inet_getpeer_v4(daddr, create);
1302
1303 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1304 inet_putpeer(peer);
1305 else
1306 rt->rt_peer_genid = rt_peer_genid();
1307 }
1308
1309 /*
1310 * Peer allocation may fail only in serious out-of-memory conditions. However
1311 * we still can generate some output.
1312 * Random ID selection looks a bit dangerous because we have no chances to
1313 * select ID being unique in a reasonable period of time.
1314 * But broken packet identifier may be better than no packet at all.
1315 */
1316 static void ip_select_fb_ident(struct iphdr *iph)
1317 {
1318 static DEFINE_SPINLOCK(ip_fb_id_lock);
1319 static u32 ip_fallback_id;
1320 u32 salt;
1321
1322 spin_lock_bh(&ip_fb_id_lock);
1323 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1324 iph->id = htons(salt & 0xFFFF);
1325 ip_fallback_id = salt;
1326 spin_unlock_bh(&ip_fb_id_lock);
1327 }
1328
1329 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1330 {
1331 struct rtable *rt = (struct rtable *) dst;
1332
1333 if (rt) {
1334 if (rt->peer == NULL)
1335 rt_bind_peer(rt, rt->rt_dst, 1);
1336
1337 /* If peer is attached to destination, it is never detached,
1338 so that we need not to grab a lock to dereference it.
1339 */
1340 if (rt->peer) {
1341 iph->id = htons(inet_getid(rt->peer, more));
1342 return;
1343 }
1344 } else
1345 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1346 __builtin_return_address(0));
1347
1348 ip_select_fb_ident(iph);
1349 }
1350 EXPORT_SYMBOL(__ip_select_ident);
1351
1352 static void rt_del(unsigned hash, struct rtable *rt)
1353 {
1354 struct rtable __rcu **rthp;
1355 struct rtable *aux;
1356
1357 rthp = &rt_hash_table[hash].chain;
1358 spin_lock_bh(rt_hash_lock_addr(hash));
1359 ip_rt_put(rt);
1360 while ((aux = rcu_dereference_protected(*rthp,
1361 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1362 if (aux == rt || rt_is_expired(aux)) {
1363 *rthp = aux->dst.rt_next;
1364 rt_free(aux);
1365 continue;
1366 }
1367 rthp = &aux->dst.rt_next;
1368 }
1369 spin_unlock_bh(rt_hash_lock_addr(hash));
1370 }
1371
1372 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1373 {
1374 struct rtable *rt = (struct rtable *) dst;
1375 __be32 orig_gw = rt->rt_gateway;
1376 struct neighbour *n, *old_n;
1377
1378 dst_confirm(&rt->dst);
1379
1380 rt->rt_gateway = peer->redirect_learned.a4;
1381 n = __arp_bind_neighbour(&rt->dst, rt->rt_gateway);
1382 if (IS_ERR(n))
1383 return PTR_ERR(n);
1384 old_n = xchg(&rt->dst._neighbour, n);
1385 if (old_n)
1386 neigh_release(old_n);
1387 if (!n || !(n->nud_state & NUD_VALID)) {
1388 if (n)
1389 neigh_event_send(n, NULL);
1390 rt->rt_gateway = orig_gw;
1391 return -EAGAIN;
1392 } else {
1393 rt->rt_flags |= RTCF_REDIRECTED;
1394 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
1395 }
1396 return 0;
1397 }
1398
1399 /* called in rcu_read_lock() section */
1400 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1401 __be32 saddr, struct net_device *dev)
1402 {
1403 int s, i;
1404 struct in_device *in_dev = __in_dev_get_rcu(dev);
1405 __be32 skeys[2] = { saddr, 0 };
1406 int ikeys[2] = { dev->ifindex, 0 };
1407 struct inet_peer *peer;
1408 struct net *net;
1409
1410 if (!in_dev)
1411 return;
1412
1413 net = dev_net(dev);
1414 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1415 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1416 ipv4_is_zeronet(new_gw))
1417 goto reject_redirect;
1418
1419 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1420 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1421 goto reject_redirect;
1422 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1423 goto reject_redirect;
1424 } else {
1425 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1426 goto reject_redirect;
1427 }
1428
1429 for (s = 0; s < 2; s++) {
1430 for (i = 0; i < 2; i++) {
1431 unsigned int hash;
1432 struct rtable __rcu **rthp;
1433 struct rtable *rt;
1434
1435 hash = rt_hash(daddr, skeys[s], ikeys[i], rt_genid(net));
1436
1437 rthp = &rt_hash_table[hash].chain;
1438
1439 while ((rt = rcu_dereference(*rthp)) != NULL) {
1440 rthp = &rt->dst.rt_next;
1441
1442 if (rt->rt_key_dst != daddr ||
1443 rt->rt_key_src != skeys[s] ||
1444 rt->rt_oif != ikeys[i] ||
1445 rt_is_input_route(rt) ||
1446 rt_is_expired(rt) ||
1447 !net_eq(dev_net(rt->dst.dev), net) ||
1448 rt->dst.error ||
1449 rt->dst.dev != dev ||
1450 rt->rt_gateway != old_gw)
1451 continue;
1452
1453 if (!rt->peer)
1454 rt_bind_peer(rt, rt->rt_dst, 1);
1455
1456 peer = rt->peer;
1457 if (peer) {
1458 if (peer->redirect_learned.a4 != new_gw) {
1459 peer->redirect_learned.a4 = new_gw;
1460 atomic_inc(&__rt_peer_genid);
1461 }
1462 check_peer_redir(&rt->dst, peer);
1463 }
1464 }
1465 }
1466 }
1467 return;
1468
1469 reject_redirect:
1470 #ifdef CONFIG_IP_ROUTE_VERBOSE
1471 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1472 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1473 " Advised path = %pI4 -> %pI4\n",
1474 &old_gw, dev->name, &new_gw,
1475 &saddr, &daddr);
1476 #endif
1477 ;
1478 }
1479
1480 static bool peer_pmtu_expired(struct inet_peer *peer)
1481 {
1482 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1483
1484 return orig &&
1485 time_after_eq(jiffies, orig) &&
1486 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1487 }
1488
1489 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1490 {
1491 unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1492
1493 return orig &&
1494 cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1495 }
1496
1497 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1498 {
1499 struct rtable *rt = (struct rtable *)dst;
1500 struct dst_entry *ret = dst;
1501
1502 if (rt) {
1503 if (dst->obsolete > 0) {
1504 ip_rt_put(rt);
1505 ret = NULL;
1506 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1507 unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1508 rt->rt_oif,
1509 rt_genid(dev_net(dst->dev)));
1510 rt_del(hash, rt);
1511 ret = NULL;
1512 } else if (rt->peer && peer_pmtu_expired(rt->peer)) {
1513 dst_metric_set(dst, RTAX_MTU, rt->peer->pmtu_orig);
1514 }
1515 }
1516 return ret;
1517 }
1518
1519 /*
1520 * Algorithm:
1521 * 1. The first ip_rt_redirect_number redirects are sent
1522 * with exponential backoff, then we stop sending them at all,
1523 * assuming that the host ignores our redirects.
1524 * 2. If we did not see packets requiring redirects
1525 * during ip_rt_redirect_silence, we assume that the host
1526 * forgot redirected route and start to send redirects again.
1527 *
1528 * This algorithm is much cheaper and more intelligent than dumb load limiting
1529 * in icmp.c.
1530 *
1531 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1532 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1533 */
1534
1535 void ip_rt_send_redirect(struct sk_buff *skb)
1536 {
1537 struct rtable *rt = skb_rtable(skb);
1538 struct in_device *in_dev;
1539 struct inet_peer *peer;
1540 int log_martians;
1541
1542 rcu_read_lock();
1543 in_dev = __in_dev_get_rcu(rt->dst.dev);
1544 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1545 rcu_read_unlock();
1546 return;
1547 }
1548 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1549 rcu_read_unlock();
1550
1551 if (!rt->peer)
1552 rt_bind_peer(rt, rt->rt_dst, 1);
1553 peer = rt->peer;
1554 if (!peer) {
1555 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1556 return;
1557 }
1558
1559 /* No redirected packets during ip_rt_redirect_silence;
1560 * reset the algorithm.
1561 */
1562 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1563 peer->rate_tokens = 0;
1564
1565 /* Too many ignored redirects; do not send anything
1566 * set dst.rate_last to the last seen redirected packet.
1567 */
1568 if (peer->rate_tokens >= ip_rt_redirect_number) {
1569 peer->rate_last = jiffies;
1570 return;
1571 }
1572
1573 /* Check for load limit; set rate_last to the latest sent
1574 * redirect.
1575 */
1576 if (peer->rate_tokens == 0 ||
1577 time_after(jiffies,
1578 (peer->rate_last +
1579 (ip_rt_redirect_load << peer->rate_tokens)))) {
1580 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1581 peer->rate_last = jiffies;
1582 ++peer->rate_tokens;
1583 #ifdef CONFIG_IP_ROUTE_VERBOSE
1584 if (log_martians &&
1585 peer->rate_tokens == ip_rt_redirect_number &&
1586 net_ratelimit())
1587 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1588 &ip_hdr(skb)->saddr, rt->rt_iif,
1589 &rt->rt_dst, &rt->rt_gateway);
1590 #endif
1591 }
1592 }
1593
1594 static int ip_error(struct sk_buff *skb)
1595 {
1596 struct rtable *rt = skb_rtable(skb);
1597 struct inet_peer *peer;
1598 unsigned long now;
1599 bool send;
1600 int code;
1601
1602 switch (rt->dst.error) {
1603 case EINVAL:
1604 default:
1605 goto out;
1606 case EHOSTUNREACH:
1607 code = ICMP_HOST_UNREACH;
1608 break;
1609 case ENETUNREACH:
1610 code = ICMP_NET_UNREACH;
1611 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1612 IPSTATS_MIB_INNOROUTES);
1613 break;
1614 case EACCES:
1615 code = ICMP_PKT_FILTERED;
1616 break;
1617 }
1618
1619 if (!rt->peer)
1620 rt_bind_peer(rt, rt->rt_dst, 1);
1621 peer = rt->peer;
1622
1623 send = true;
1624 if (peer) {
1625 now = jiffies;
1626 peer->rate_tokens += now - peer->rate_last;
1627 if (peer->rate_tokens > ip_rt_error_burst)
1628 peer->rate_tokens = ip_rt_error_burst;
1629 peer->rate_last = now;
1630 if (peer->rate_tokens >= ip_rt_error_cost)
1631 peer->rate_tokens -= ip_rt_error_cost;
1632 else
1633 send = false;
1634 }
1635 if (send)
1636 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1637
1638 out: kfree_skb(skb);
1639 return 0;
1640 }
1641
1642 /*
1643 * The last two values are not from the RFC but
1644 * are needed for AMPRnet AX.25 paths.
1645 */
1646
1647 static const unsigned short mtu_plateau[] =
1648 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1649
1650 static inline unsigned short guess_mtu(unsigned short old_mtu)
1651 {
1652 int i;
1653
1654 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1655 if (old_mtu > mtu_plateau[i])
1656 return mtu_plateau[i];
1657 return 68;
1658 }
1659
1660 unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1661 unsigned short new_mtu,
1662 struct net_device *dev)
1663 {
1664 unsigned short old_mtu = ntohs(iph->tot_len);
1665 unsigned short est_mtu = 0;
1666 struct inet_peer *peer;
1667
1668 peer = inet_getpeer_v4(iph->daddr, 1);
1669 if (peer) {
1670 unsigned short mtu = new_mtu;
1671
1672 if (new_mtu < 68 || new_mtu >= old_mtu) {
1673 /* BSD 4.2 derived systems incorrectly adjust
1674 * tot_len by the IP header length, and report
1675 * a zero MTU in the ICMP message.
1676 */
1677 if (mtu == 0 &&
1678 old_mtu >= 68 + (iph->ihl << 2))
1679 old_mtu -= iph->ihl << 2;
1680 mtu = guess_mtu(old_mtu);
1681 }
1682
1683 if (mtu < ip_rt_min_pmtu)
1684 mtu = ip_rt_min_pmtu;
1685 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1686 unsigned long pmtu_expires;
1687
1688 pmtu_expires = jiffies + ip_rt_mtu_expires;
1689 if (!pmtu_expires)
1690 pmtu_expires = 1UL;
1691
1692 est_mtu = mtu;
1693 peer->pmtu_learned = mtu;
1694 peer->pmtu_expires = pmtu_expires;
1695 }
1696
1697 inet_putpeer(peer);
1698
1699 atomic_inc(&__rt_peer_genid);
1700 }
1701 return est_mtu ? : new_mtu;
1702 }
1703
1704 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1705 {
1706 unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1707
1708 if (!expires)
1709 return;
1710 if (time_before(jiffies, expires)) {
1711 u32 orig_dst_mtu = dst_mtu(dst);
1712 if (peer->pmtu_learned < orig_dst_mtu) {
1713 if (!peer->pmtu_orig)
1714 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1715 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1716 }
1717 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1718 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1719 }
1720
1721 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1722 {
1723 struct rtable *rt = (struct rtable *) dst;
1724 struct inet_peer *peer;
1725
1726 dst_confirm(dst);
1727
1728 if (!rt->peer)
1729 rt_bind_peer(rt, rt->rt_dst, 1);
1730 peer = rt->peer;
1731 if (peer) {
1732 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1733
1734 if (mtu < ip_rt_min_pmtu)
1735 mtu = ip_rt_min_pmtu;
1736 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1737
1738 pmtu_expires = jiffies + ip_rt_mtu_expires;
1739 if (!pmtu_expires)
1740 pmtu_expires = 1UL;
1741
1742 peer->pmtu_learned = mtu;
1743 peer->pmtu_expires = pmtu_expires;
1744
1745 atomic_inc(&__rt_peer_genid);
1746 rt->rt_peer_genid = rt_peer_genid();
1747 }
1748 check_peer_pmtu(dst, peer);
1749 }
1750 }
1751
1752 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1753 {
1754 struct rtable *rt = (struct rtable *) dst;
1755
1756 if (rt_is_expired(rt))
1757 return NULL;
1758 if (rt->rt_peer_genid != rt_peer_genid()) {
1759 struct inet_peer *peer;
1760
1761 if (!rt->peer)
1762 rt_bind_peer(rt, rt->rt_dst, 0);
1763
1764 peer = rt->peer;
1765 if (peer) {
1766 check_peer_pmtu(dst, peer);
1767
1768 if (peer->redirect_learned.a4 &&
1769 peer->redirect_learned.a4 != rt->rt_gateway) {
1770 if (check_peer_redir(dst, peer))
1771 return NULL;
1772 }
1773 }
1774
1775 rt->rt_peer_genid = rt_peer_genid();
1776 }
1777 return dst;
1778 }
1779
1780 static void ipv4_dst_destroy(struct dst_entry *dst)
1781 {
1782 struct rtable *rt = (struct rtable *) dst;
1783 struct inet_peer *peer = rt->peer;
1784
1785 if (rt->fi) {
1786 fib_info_put(rt->fi);
1787 rt->fi = NULL;
1788 }
1789 if (peer) {
1790 rt->peer = NULL;
1791 inet_putpeer(peer);
1792 }
1793 }
1794
1795
1796 static void ipv4_link_failure(struct sk_buff *skb)
1797 {
1798 struct rtable *rt;
1799
1800 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1801
1802 rt = skb_rtable(skb);
1803 if (rt && rt->peer && peer_pmtu_cleaned(rt->peer))
1804 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1805 }
1806
1807 static int ip_rt_bug(struct sk_buff *skb)
1808 {
1809 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1810 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1811 skb->dev ? skb->dev->name : "?");
1812 kfree_skb(skb);
1813 WARN_ON(1);
1814 return 0;
1815 }
1816
1817 /*
1818 We do not cache source address of outgoing interface,
1819 because it is used only by IP RR, TS and SRR options,
1820 so that it out of fast path.
1821
1822 BTW remember: "addr" is allowed to be not aligned
1823 in IP options!
1824 */
1825
1826 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1827 {
1828 __be32 src;
1829
1830 if (rt_is_output_route(rt))
1831 src = ip_hdr(skb)->saddr;
1832 else {
1833 struct fib_result res;
1834 struct flowi4 fl4;
1835 struct iphdr *iph;
1836
1837 iph = ip_hdr(skb);
1838
1839 memset(&fl4, 0, sizeof(fl4));
1840 fl4.daddr = iph->daddr;
1841 fl4.saddr = iph->saddr;
1842 fl4.flowi4_tos = RT_TOS(iph->tos);
1843 fl4.flowi4_oif = rt->dst.dev->ifindex;
1844 fl4.flowi4_iif = skb->dev->ifindex;
1845 fl4.flowi4_mark = skb->mark;
1846
1847 rcu_read_lock();
1848 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1849 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1850 else
1851 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1852 RT_SCOPE_UNIVERSE);
1853 rcu_read_unlock();
1854 }
1855 memcpy(addr, &src, 4);
1856 }
1857
1858 #ifdef CONFIG_IP_ROUTE_CLASSID
1859 static void set_class_tag(struct rtable *rt, u32 tag)
1860 {
1861 if (!(rt->dst.tclassid & 0xFFFF))
1862 rt->dst.tclassid |= tag & 0xFFFF;
1863 if (!(rt->dst.tclassid & 0xFFFF0000))
1864 rt->dst.tclassid |= tag & 0xFFFF0000;
1865 }
1866 #endif
1867
1868 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1869 {
1870 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1871
1872 if (advmss == 0) {
1873 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1874 ip_rt_min_advmss);
1875 if (advmss > 65535 - 40)
1876 advmss = 65535 - 40;
1877 }
1878 return advmss;
1879 }
1880
1881 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1882 {
1883 unsigned int mtu = dst->dev->mtu;
1884
1885 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1886 const struct rtable *rt = (const struct rtable *) dst;
1887
1888 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1889 mtu = 576;
1890 }
1891
1892 if (mtu > IP_MAX_MTU)
1893 mtu = IP_MAX_MTU;
1894
1895 return mtu;
1896 }
1897
1898 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1899 struct fib_info *fi)
1900 {
1901 struct inet_peer *peer;
1902 int create = 0;
1903
1904 /* If a peer entry exists for this destination, we must hook
1905 * it up in order to get at cached metrics.
1906 */
1907 if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1908 create = 1;
1909
1910 rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1911 if (peer) {
1912 rt->rt_peer_genid = rt_peer_genid();
1913 if (inet_metrics_new(peer))
1914 memcpy(peer->metrics, fi->fib_metrics,
1915 sizeof(u32) * RTAX_MAX);
1916 dst_init_metrics(&rt->dst, peer->metrics, false);
1917
1918 check_peer_pmtu(&rt->dst, peer);
1919 if (peer->redirect_learned.a4 &&
1920 peer->redirect_learned.a4 != rt->rt_gateway) {
1921 rt->rt_gateway = peer->redirect_learned.a4;
1922 rt->rt_flags |= RTCF_REDIRECTED;
1923 }
1924 } else {
1925 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1926 rt->fi = fi;
1927 atomic_inc(&fi->fib_clntref);
1928 }
1929 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1930 }
1931 }
1932
1933 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1934 const struct fib_result *res,
1935 struct fib_info *fi, u16 type, u32 itag)
1936 {
1937 struct dst_entry *dst = &rt->dst;
1938
1939 if (fi) {
1940 if (FIB_RES_GW(*res) &&
1941 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1942 rt->rt_gateway = FIB_RES_GW(*res);
1943 rt_init_metrics(rt, fl4, fi);
1944 #ifdef CONFIG_IP_ROUTE_CLASSID
1945 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1946 #endif
1947 }
1948
1949 if (dst_mtu(dst) > IP_MAX_MTU)
1950 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1951 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1952 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1953
1954 #ifdef CONFIG_IP_ROUTE_CLASSID
1955 #ifdef CONFIG_IP_MULTIPLE_TABLES
1956 set_class_tag(rt, fib_rules_tclass(res));
1957 #endif
1958 set_class_tag(rt, itag);
1959 #endif
1960 }
1961
1962 static struct rtable *rt_dst_alloc(struct net_device *dev,
1963 bool nopolicy, bool noxfrm)
1964 {
1965 return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1966 DST_HOST |
1967 (nopolicy ? DST_NOPOLICY : 0) |
1968 (noxfrm ? DST_NOXFRM : 0));
1969 }
1970
1971 /* called in rcu_read_lock() section */
1972 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1973 u8 tos, struct net_device *dev, int our)
1974 {
1975 unsigned int hash;
1976 struct rtable *rth;
1977 __be32 spec_dst;
1978 struct in_device *in_dev = __in_dev_get_rcu(dev);
1979 u32 itag = 0;
1980 int err;
1981
1982 /* Primary sanity checks. */
1983
1984 if (in_dev == NULL)
1985 return -EINVAL;
1986
1987 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1988 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1989 goto e_inval;
1990
1991 if (ipv4_is_zeronet(saddr)) {
1992 if (!ipv4_is_local_multicast(daddr))
1993 goto e_inval;
1994 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1995 } else {
1996 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
1997 &itag);
1998 if (err < 0)
1999 goto e_err;
2000 }
2001 rth = rt_dst_alloc(init_net.loopback_dev,
2002 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2003 if (!rth)
2004 goto e_nobufs;
2005
2006 #ifdef CONFIG_IP_ROUTE_CLASSID
2007 rth->dst.tclassid = itag;
2008 #endif
2009 rth->dst.output = ip_rt_bug;
2010
2011 rth->rt_key_dst = daddr;
2012 rth->rt_key_src = saddr;
2013 rth->rt_genid = rt_genid(dev_net(dev));
2014 rth->rt_flags = RTCF_MULTICAST;
2015 rth->rt_type = RTN_MULTICAST;
2016 rth->rt_key_tos = tos;
2017 rth->rt_dst = daddr;
2018 rth->rt_src = saddr;
2019 rth->rt_route_iif = dev->ifindex;
2020 rth->rt_iif = dev->ifindex;
2021 rth->rt_oif = 0;
2022 rth->rt_mark = skb->mark;
2023 rth->rt_gateway = daddr;
2024 rth->rt_spec_dst= spec_dst;
2025 rth->rt_peer_genid = 0;
2026 rth->peer = NULL;
2027 rth->fi = NULL;
2028 if (our) {
2029 rth->dst.input= ip_local_deliver;
2030 rth->rt_flags |= RTCF_LOCAL;
2031 }
2032
2033 #ifdef CONFIG_IP_MROUTE
2034 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
2035 rth->dst.input = ip_mr_input;
2036 #endif
2037 RT_CACHE_STAT_INC(in_slow_mc);
2038
2039 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
2040 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
2041 return IS_ERR(rth) ? PTR_ERR(rth) : 0;
2042
2043 e_nobufs:
2044 return -ENOBUFS;
2045 e_inval:
2046 return -EINVAL;
2047 e_err:
2048 return err;
2049 }
2050
2051
2052 static void ip_handle_martian_source(struct net_device *dev,
2053 struct in_device *in_dev,
2054 struct sk_buff *skb,
2055 __be32 daddr,
2056 __be32 saddr)
2057 {
2058 RT_CACHE_STAT_INC(in_martian_src);
2059 #ifdef CONFIG_IP_ROUTE_VERBOSE
2060 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
2061 /*
2062 * RFC1812 recommendation, if source is martian,
2063 * the only hint is MAC header.
2064 */
2065 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
2066 &daddr, &saddr, dev->name);
2067 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
2068 int i;
2069 const unsigned char *p = skb_mac_header(skb);
2070 printk(KERN_WARNING "ll header: ");
2071 for (i = 0; i < dev->hard_header_len; i++, p++) {
2072 printk("%02x", *p);
2073 if (i < (dev->hard_header_len - 1))
2074 printk(":");
2075 }
2076 printk("\n");
2077 }
2078 }
2079 #endif
2080 }
2081
2082 /* called in rcu_read_lock() section */
2083 static int __mkroute_input(struct sk_buff *skb,
2084 const struct fib_result *res,
2085 struct in_device *in_dev,
2086 __be32 daddr, __be32 saddr, u32 tos,
2087 struct rtable **result)
2088 {
2089 struct rtable *rth;
2090 int err;
2091 struct in_device *out_dev;
2092 unsigned int flags = 0;
2093 __be32 spec_dst;
2094 u32 itag;
2095
2096 /* get a working reference to the output device */
2097 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
2098 if (out_dev == NULL) {
2099 if (net_ratelimit())
2100 printk(KERN_CRIT "Bug in ip_route_input" \
2101 "_slow(). Please, report\n");
2102 return -EINVAL;
2103 }
2104
2105
2106 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
2107 in_dev->dev, &spec_dst, &itag);
2108 if (err < 0) {
2109 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2110 saddr);
2111
2112 goto cleanup;
2113 }
2114
2115 if (err)
2116 flags |= RTCF_DIRECTSRC;
2117
2118 if (out_dev == in_dev && err &&
2119 (IN_DEV_SHARED_MEDIA(out_dev) ||
2120 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2121 flags |= RTCF_DOREDIRECT;
2122
2123 if (skb->protocol != htons(ETH_P_IP)) {
2124 /* Not IP (i.e. ARP). Do not create route, if it is
2125 * invalid for proxy arp. DNAT routes are always valid.
2126 *
2127 * Proxy arp feature have been extended to allow, ARP
2128 * replies back to the same interface, to support
2129 * Private VLAN switch technologies. See arp.c.
2130 */
2131 if (out_dev == in_dev &&
2132 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2133 err = -EINVAL;
2134 goto cleanup;
2135 }
2136 }
2137
2138 rth = rt_dst_alloc(out_dev->dev,
2139 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2140 IN_DEV_CONF_GET(out_dev, NOXFRM));
2141 if (!rth) {
2142 err = -ENOBUFS;
2143 goto cleanup;
2144 }
2145
2146 rth->rt_key_dst = daddr;
2147 rth->rt_key_src = saddr;
2148 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2149 rth->rt_flags = flags;
2150 rth->rt_type = res->type;
2151 rth->rt_key_tos = tos;
2152 rth->rt_dst = daddr;
2153 rth->rt_src = saddr;
2154 rth->rt_route_iif = in_dev->dev->ifindex;
2155 rth->rt_iif = in_dev->dev->ifindex;
2156 rth->rt_oif = 0;
2157 rth->rt_mark = skb->mark;
2158 rth->rt_gateway = daddr;
2159 rth->rt_spec_dst= spec_dst;
2160 rth->rt_peer_genid = 0;
2161 rth->peer = NULL;
2162 rth->fi = NULL;
2163
2164 rth->dst.input = ip_forward;
2165 rth->dst.output = ip_output;
2166
2167 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2168
2169 *result = rth;
2170 err = 0;
2171 cleanup:
2172 return err;
2173 }
2174
2175 static int ip_mkroute_input(struct sk_buff *skb,
2176 struct fib_result *res,
2177 const struct flowi4 *fl4,
2178 struct in_device *in_dev,
2179 __be32 daddr, __be32 saddr, u32 tos)
2180 {
2181 struct rtable* rth = NULL;
2182 int err;
2183 unsigned hash;
2184
2185 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2186 if (res->fi && res->fi->fib_nhs > 1)
2187 fib_select_multipath(res);
2188 #endif
2189
2190 /* create a routing cache entry */
2191 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2192 if (err)
2193 return err;
2194
2195 /* put it into the cache */
2196 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2197 rt_genid(dev_net(rth->dst.dev)));
2198 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2199 if (IS_ERR(rth))
2200 return PTR_ERR(rth);
2201 return 0;
2202 }
2203
2204 /*
2205 * NOTE. We drop all the packets that has local source
2206 * addresses, because every properly looped back packet
2207 * must have correct destination already attached by output routine.
2208 *
2209 * Such approach solves two big problems:
2210 * 1. Not simplex devices are handled properly.
2211 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2212 * called with rcu_read_lock()
2213 */
2214
2215 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2216 u8 tos, struct net_device *dev)
2217 {
2218 struct fib_result res;
2219 struct in_device *in_dev = __in_dev_get_rcu(dev);
2220 struct flowi4 fl4;
2221 unsigned flags = 0;
2222 u32 itag = 0;
2223 struct rtable * rth;
2224 unsigned hash;
2225 __be32 spec_dst;
2226 int err = -EINVAL;
2227 struct net * net = dev_net(dev);
2228
2229 /* IP on this device is disabled. */
2230
2231 if (!in_dev)
2232 goto out;
2233
2234 /* Check for the most weird martians, which can be not detected
2235 by fib_lookup.
2236 */
2237
2238 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2239 ipv4_is_loopback(saddr))
2240 goto martian_source;
2241
2242 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2243 goto brd_input;
2244
2245 /* Accept zero addresses only to limited broadcast;
2246 * I even do not know to fix it or not. Waiting for complains :-)
2247 */
2248 if (ipv4_is_zeronet(saddr))
2249 goto martian_source;
2250
2251 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2252 goto martian_destination;
2253
2254 /*
2255 * Now we are ready to route packet.
2256 */
2257 fl4.flowi4_oif = 0;
2258 fl4.flowi4_iif = dev->ifindex;
2259 fl4.flowi4_mark = skb->mark;
2260 fl4.flowi4_tos = tos;
2261 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2262 fl4.daddr = daddr;
2263 fl4.saddr = saddr;
2264 err = fib_lookup(net, &fl4, &res);
2265 if (err != 0) {
2266 if (!IN_DEV_FORWARD(in_dev))
2267 goto e_hostunreach;
2268 goto no_route;
2269 }
2270
2271 RT_CACHE_STAT_INC(in_slow_tot);
2272
2273 if (res.type == RTN_BROADCAST)
2274 goto brd_input;
2275
2276 if (res.type == RTN_LOCAL) {
2277 err = fib_validate_source(skb, saddr, daddr, tos,
2278 net->loopback_dev->ifindex,
2279 dev, &spec_dst, &itag);
2280 if (err < 0)
2281 goto martian_source_keep_err;
2282 if (err)
2283 flags |= RTCF_DIRECTSRC;
2284 spec_dst = daddr;
2285 goto local_input;
2286 }
2287
2288 if (!IN_DEV_FORWARD(in_dev))
2289 goto e_hostunreach;
2290 if (res.type != RTN_UNICAST)
2291 goto martian_destination;
2292
2293 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2294 out: return err;
2295
2296 brd_input:
2297 if (skb->protocol != htons(ETH_P_IP))
2298 goto e_inval;
2299
2300 if (ipv4_is_zeronet(saddr))
2301 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2302 else {
2303 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2304 &itag);
2305 if (err < 0)
2306 goto martian_source_keep_err;
2307 if (err)
2308 flags |= RTCF_DIRECTSRC;
2309 }
2310 flags |= RTCF_BROADCAST;
2311 res.type = RTN_BROADCAST;
2312 RT_CACHE_STAT_INC(in_brd);
2313
2314 local_input:
2315 rth = rt_dst_alloc(net->loopback_dev,
2316 IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2317 if (!rth)
2318 goto e_nobufs;
2319
2320 rth->dst.input= ip_local_deliver;
2321 rth->dst.output= ip_rt_bug;
2322 #ifdef CONFIG_IP_ROUTE_CLASSID
2323 rth->dst.tclassid = itag;
2324 #endif
2325
2326 rth->rt_key_dst = daddr;
2327 rth->rt_key_src = saddr;
2328 rth->rt_genid = rt_genid(net);
2329 rth->rt_flags = flags|RTCF_LOCAL;
2330 rth->rt_type = res.type;
2331 rth->rt_key_tos = tos;
2332 rth->rt_dst = daddr;
2333 rth->rt_src = saddr;
2334 #ifdef CONFIG_IP_ROUTE_CLASSID
2335 rth->dst.tclassid = itag;
2336 #endif
2337 rth->rt_route_iif = dev->ifindex;
2338 rth->rt_iif = dev->ifindex;
2339 rth->rt_oif = 0;
2340 rth->rt_mark = skb->mark;
2341 rth->rt_gateway = daddr;
2342 rth->rt_spec_dst= spec_dst;
2343 rth->rt_peer_genid = 0;
2344 rth->peer = NULL;
2345 rth->fi = NULL;
2346 if (res.type == RTN_UNREACHABLE) {
2347 rth->dst.input= ip_error;
2348 rth->dst.error= -err;
2349 rth->rt_flags &= ~RTCF_LOCAL;
2350 }
2351 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2352 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2353 err = 0;
2354 if (IS_ERR(rth))
2355 err = PTR_ERR(rth);
2356 goto out;
2357
2358 no_route:
2359 RT_CACHE_STAT_INC(in_no_route);
2360 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2361 res.type = RTN_UNREACHABLE;
2362 if (err == -ESRCH)
2363 err = -ENETUNREACH;
2364 goto local_input;
2365
2366 /*
2367 * Do not cache martian addresses: they should be logged (RFC1812)
2368 */
2369 martian_destination:
2370 RT_CACHE_STAT_INC(in_martian_dst);
2371 #ifdef CONFIG_IP_ROUTE_VERBOSE
2372 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2373 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2374 &daddr, &saddr, dev->name);
2375 #endif
2376
2377 e_hostunreach:
2378 err = -EHOSTUNREACH;
2379 goto out;
2380
2381 e_inval:
2382 err = -EINVAL;
2383 goto out;
2384
2385 e_nobufs:
2386 err = -ENOBUFS;
2387 goto out;
2388
2389 martian_source:
2390 err = -EINVAL;
2391 martian_source_keep_err:
2392 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2393 goto out;
2394 }
2395
2396 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2397 u8 tos, struct net_device *dev, bool noref)
2398 {
2399 struct rtable * rth;
2400 unsigned hash;
2401 int iif = dev->ifindex;
2402 struct net *net;
2403 int res;
2404
2405 net = dev_net(dev);
2406
2407 rcu_read_lock();
2408
2409 if (!rt_caching(net))
2410 goto skip_cache;
2411
2412 tos &= IPTOS_RT_MASK;
2413 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2414
2415 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2416 rth = rcu_dereference(rth->dst.rt_next)) {
2417 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2418 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2419 (rth->rt_route_iif ^ iif) |
2420 (rth->rt_key_tos ^ tos)) == 0 &&
2421 rth->rt_mark == skb->mark &&
2422 net_eq(dev_net(rth->dst.dev), net) &&
2423 !rt_is_expired(rth)) {
2424 if (noref) {
2425 dst_use_noref(&rth->dst, jiffies);
2426 skb_dst_set_noref(skb, &rth->dst);
2427 } else {
2428 dst_use(&rth->dst, jiffies);
2429 skb_dst_set(skb, &rth->dst);
2430 }
2431 RT_CACHE_STAT_INC(in_hit);
2432 rcu_read_unlock();
2433 return 0;
2434 }
2435 RT_CACHE_STAT_INC(in_hlist_search);
2436 }
2437
2438 skip_cache:
2439 /* Multicast recognition logic is moved from route cache to here.
2440 The problem was that too many Ethernet cards have broken/missing
2441 hardware multicast filters :-( As result the host on multicasting
2442 network acquires a lot of useless route cache entries, sort of
2443 SDR messages from all the world. Now we try to get rid of them.
2444 Really, provided software IP multicast filter is organized
2445 reasonably (at least, hashed), it does not result in a slowdown
2446 comparing with route cache reject entries.
2447 Note, that multicast routers are not affected, because
2448 route cache entry is created eventually.
2449 */
2450 if (ipv4_is_multicast(daddr)) {
2451 struct in_device *in_dev = __in_dev_get_rcu(dev);
2452
2453 if (in_dev) {
2454 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2455 ip_hdr(skb)->protocol);
2456 if (our
2457 #ifdef CONFIG_IP_MROUTE
2458 ||
2459 (!ipv4_is_local_multicast(daddr) &&
2460 IN_DEV_MFORWARD(in_dev))
2461 #endif
2462 ) {
2463 int res = ip_route_input_mc(skb, daddr, saddr,
2464 tos, dev, our);
2465 rcu_read_unlock();
2466 return res;
2467 }
2468 }
2469 rcu_read_unlock();
2470 return -EINVAL;
2471 }
2472 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2473 rcu_read_unlock();
2474 return res;
2475 }
2476 EXPORT_SYMBOL(ip_route_input_common);
2477
2478 /* called with rcu_read_lock() */
2479 static struct rtable *__mkroute_output(const struct fib_result *res,
2480 const struct flowi4 *fl4,
2481 __be32 orig_daddr, __be32 orig_saddr,
2482 int orig_oif, struct net_device *dev_out,
2483 unsigned int flags)
2484 {
2485 struct fib_info *fi = res->fi;
2486 u32 tos = RT_FL_TOS(fl4);
2487 struct in_device *in_dev;
2488 u16 type = res->type;
2489 struct rtable *rth;
2490
2491 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2492 return ERR_PTR(-EINVAL);
2493
2494 if (ipv4_is_lbcast(fl4->daddr))
2495 type = RTN_BROADCAST;
2496 else if (ipv4_is_multicast(fl4->daddr))
2497 type = RTN_MULTICAST;
2498 else if (ipv4_is_zeronet(fl4->daddr))
2499 return ERR_PTR(-EINVAL);
2500
2501 if (dev_out->flags & IFF_LOOPBACK)
2502 flags |= RTCF_LOCAL;
2503
2504 in_dev = __in_dev_get_rcu(dev_out);
2505 if (!in_dev)
2506 return ERR_PTR(-EINVAL);
2507
2508 if (type == RTN_BROADCAST) {
2509 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2510 fi = NULL;
2511 } else if (type == RTN_MULTICAST) {
2512 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2513 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2514 fl4->flowi4_proto))
2515 flags &= ~RTCF_LOCAL;
2516 /* If multicast route do not exist use
2517 * default one, but do not gateway in this case.
2518 * Yes, it is hack.
2519 */
2520 if (fi && res->prefixlen < 4)
2521 fi = NULL;
2522 }
2523
2524 rth = rt_dst_alloc(dev_out,
2525 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2526 IN_DEV_CONF_GET(in_dev, NOXFRM));
2527 if (!rth)
2528 return ERR_PTR(-ENOBUFS);
2529
2530 rth->dst.output = ip_output;
2531
2532 rth->rt_key_dst = orig_daddr;
2533 rth->rt_key_src = orig_saddr;
2534 rth->rt_genid = rt_genid(dev_net(dev_out));
2535 rth->rt_flags = flags;
2536 rth->rt_type = type;
2537 rth->rt_key_tos = tos;
2538 rth->rt_dst = fl4->daddr;
2539 rth->rt_src = fl4->saddr;
2540 rth->rt_route_iif = 0;
2541 rth->rt_iif = orig_oif ? : dev_out->ifindex;
2542 rth->rt_oif = orig_oif;
2543 rth->rt_mark = fl4->flowi4_mark;
2544 rth->rt_gateway = fl4->daddr;
2545 rth->rt_spec_dst= fl4->saddr;
2546 rth->rt_peer_genid = 0;
2547 rth->peer = NULL;
2548 rth->fi = NULL;
2549
2550 RT_CACHE_STAT_INC(out_slow_tot);
2551
2552 if (flags & RTCF_LOCAL) {
2553 rth->dst.input = ip_local_deliver;
2554 rth->rt_spec_dst = fl4->daddr;
2555 }
2556 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2557 rth->rt_spec_dst = fl4->saddr;
2558 if (flags & RTCF_LOCAL &&
2559 !(dev_out->flags & IFF_LOOPBACK)) {
2560 rth->dst.output = ip_mc_output;
2561 RT_CACHE_STAT_INC(out_slow_mc);
2562 }
2563 #ifdef CONFIG_IP_MROUTE
2564 if (type == RTN_MULTICAST) {
2565 if (IN_DEV_MFORWARD(in_dev) &&
2566 !ipv4_is_local_multicast(fl4->daddr)) {
2567 rth->dst.input = ip_mr_input;
2568 rth->dst.output = ip_mc_output;
2569 }
2570 }
2571 #endif
2572 }
2573
2574 rt_set_nexthop(rth, fl4, res, fi, type, 0);
2575
2576 return rth;
2577 }
2578
2579 /*
2580 * Major route resolver routine.
2581 * called with rcu_read_lock();
2582 */
2583
2584 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2585 {
2586 struct net_device *dev_out = NULL;
2587 u32 tos = RT_FL_TOS(fl4);
2588 unsigned int flags = 0;
2589 struct fib_result res;
2590 struct rtable *rth;
2591 __be32 orig_daddr;
2592 __be32 orig_saddr;
2593 int orig_oif;
2594
2595 res.fi = NULL;
2596 #ifdef CONFIG_IP_MULTIPLE_TABLES
2597 res.r = NULL;
2598 #endif
2599
2600 orig_daddr = fl4->daddr;
2601 orig_saddr = fl4->saddr;
2602 orig_oif = fl4->flowi4_oif;
2603
2604 fl4->flowi4_iif = net->loopback_dev->ifindex;
2605 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2606 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2607 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2608
2609 rcu_read_lock();
2610 if (fl4->saddr) {
2611 rth = ERR_PTR(-EINVAL);
2612 if (ipv4_is_multicast(fl4->saddr) ||
2613 ipv4_is_lbcast(fl4->saddr) ||
2614 ipv4_is_zeronet(fl4->saddr))
2615 goto out;
2616
2617 /* I removed check for oif == dev_out->oif here.
2618 It was wrong for two reasons:
2619 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2620 is assigned to multiple interfaces.
2621 2. Moreover, we are allowed to send packets with saddr
2622 of another iface. --ANK
2623 */
2624
2625 if (fl4->flowi4_oif == 0 &&
2626 (ipv4_is_multicast(fl4->daddr) ||
2627 ipv4_is_lbcast(fl4->daddr))) {
2628 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2629 dev_out = __ip_dev_find(net, fl4->saddr, false);
2630 if (dev_out == NULL)
2631 goto out;
2632
2633 /* Special hack: user can direct multicasts
2634 and limited broadcast via necessary interface
2635 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2636 This hack is not just for fun, it allows
2637 vic,vat and friends to work.
2638 They bind socket to loopback, set ttl to zero
2639 and expect that it will work.
2640 From the viewpoint of routing cache they are broken,
2641 because we are not allowed to build multicast path
2642 with loopback source addr (look, routing cache
2643 cannot know, that ttl is zero, so that packet
2644 will not leave this host and route is valid).
2645 Luckily, this hack is good workaround.
2646 */
2647
2648 fl4->flowi4_oif = dev_out->ifindex;
2649 goto make_route;
2650 }
2651
2652 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2653 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2654 if (!__ip_dev_find(net, fl4->saddr, false))
2655 goto out;
2656 }
2657 }
2658
2659
2660 if (fl4->flowi4_oif) {
2661 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2662 rth = ERR_PTR(-ENODEV);
2663 if (dev_out == NULL)
2664 goto out;
2665
2666 /* RACE: Check return value of inet_select_addr instead. */
2667 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2668 rth = ERR_PTR(-ENETUNREACH);
2669 goto out;
2670 }
2671 if (ipv4_is_local_multicast(fl4->daddr) ||
2672 ipv4_is_lbcast(fl4->daddr)) {
2673 if (!fl4->saddr)
2674 fl4->saddr = inet_select_addr(dev_out, 0,
2675 RT_SCOPE_LINK);
2676 goto make_route;
2677 }
2678 if (fl4->saddr) {
2679 if (ipv4_is_multicast(fl4->daddr))
2680 fl4->saddr = inet_select_addr(dev_out, 0,
2681 fl4->flowi4_scope);
2682 else if (!fl4->daddr)
2683 fl4->saddr = inet_select_addr(dev_out, 0,
2684 RT_SCOPE_HOST);
2685 }
2686 }
2687
2688 if (!fl4->daddr) {
2689 fl4->daddr = fl4->saddr;
2690 if (!fl4->daddr)
2691 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2692 dev_out = net->loopback_dev;
2693 fl4->flowi4_oif = net->loopback_dev->ifindex;
2694 res.type = RTN_LOCAL;
2695 flags |= RTCF_LOCAL;
2696 goto make_route;
2697 }
2698
2699 if (fib_lookup(net, fl4, &res)) {
2700 res.fi = NULL;
2701 if (fl4->flowi4_oif) {
2702 /* Apparently, routing tables are wrong. Assume,
2703 that the destination is on link.
2704
2705 WHY? DW.
2706 Because we are allowed to send to iface
2707 even if it has NO routes and NO assigned
2708 addresses. When oif is specified, routing
2709 tables are looked up with only one purpose:
2710 to catch if destination is gatewayed, rather than
2711 direct. Moreover, if MSG_DONTROUTE is set,
2712 we send packet, ignoring both routing tables
2713 and ifaddr state. --ANK
2714
2715
2716 We could make it even if oif is unknown,
2717 likely IPv6, but we do not.
2718 */
2719
2720 if (fl4->saddr == 0)
2721 fl4->saddr = inet_select_addr(dev_out, 0,
2722 RT_SCOPE_LINK);
2723 res.type = RTN_UNICAST;
2724 goto make_route;
2725 }
2726 rth = ERR_PTR(-ENETUNREACH);
2727 goto out;
2728 }
2729
2730 if (res.type == RTN_LOCAL) {
2731 if (!fl4->saddr) {
2732 if (res.fi->fib_prefsrc)
2733 fl4->saddr = res.fi->fib_prefsrc;
2734 else
2735 fl4->saddr = fl4->daddr;
2736 }
2737 dev_out = net->loopback_dev;
2738 fl4->flowi4_oif = dev_out->ifindex;
2739 res.fi = NULL;
2740 flags |= RTCF_LOCAL;
2741 goto make_route;
2742 }
2743
2744 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2745 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2746 fib_select_multipath(&res);
2747 else
2748 #endif
2749 if (!res.prefixlen &&
2750 res.table->tb_num_default > 1 &&
2751 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2752 fib_select_default(&res);
2753
2754 if (!fl4->saddr)
2755 fl4->saddr = FIB_RES_PREFSRC(net, res);
2756
2757 dev_out = FIB_RES_DEV(res);
2758 fl4->flowi4_oif = dev_out->ifindex;
2759
2760
2761 make_route:
2762 rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2763 dev_out, flags);
2764 if (!IS_ERR(rth)) {
2765 unsigned int hash;
2766
2767 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2768 rt_genid(dev_net(dev_out)));
2769 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2770 }
2771
2772 out:
2773 rcu_read_unlock();
2774 return rth;
2775 }
2776
2777 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2778 {
2779 struct rtable *rth;
2780 unsigned int hash;
2781
2782 if (!rt_caching(net))
2783 goto slow_output;
2784
2785 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2786
2787 rcu_read_lock_bh();
2788 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2789 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2790 if (rth->rt_key_dst == flp4->daddr &&
2791 rth->rt_key_src == flp4->saddr &&
2792 rt_is_output_route(rth) &&
2793 rth->rt_oif == flp4->flowi4_oif &&
2794 rth->rt_mark == flp4->flowi4_mark &&
2795 !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2796 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2797 net_eq(dev_net(rth->dst.dev), net) &&
2798 !rt_is_expired(rth)) {
2799 dst_use(&rth->dst, jiffies);
2800 RT_CACHE_STAT_INC(out_hit);
2801 rcu_read_unlock_bh();
2802 if (!flp4->saddr)
2803 flp4->saddr = rth->rt_src;
2804 if (!flp4->daddr)
2805 flp4->daddr = rth->rt_dst;
2806 return rth;
2807 }
2808 RT_CACHE_STAT_INC(out_hlist_search);
2809 }
2810 rcu_read_unlock_bh();
2811
2812 slow_output:
2813 return ip_route_output_slow(net, flp4);
2814 }
2815 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2816
2817 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2818 {
2819 return NULL;
2820 }
2821
2822 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2823 {
2824 return 0;
2825 }
2826
2827 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2828 {
2829 }
2830
2831 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2832 unsigned long old)
2833 {
2834 return NULL;
2835 }
2836
2837 static struct dst_ops ipv4_dst_blackhole_ops = {
2838 .family = AF_INET,
2839 .protocol = cpu_to_be16(ETH_P_IP),
2840 .destroy = ipv4_dst_destroy,
2841 .check = ipv4_blackhole_dst_check,
2842 .default_mtu = ipv4_blackhole_default_mtu,
2843 .default_advmss = ipv4_default_advmss,
2844 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2845 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2846 };
2847
2848 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2849 {
2850 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2851 struct rtable *ort = (struct rtable *) dst_orig;
2852
2853 if (rt) {
2854 struct dst_entry *new = &rt->dst;
2855
2856 new->__use = 1;
2857 new->input = dst_discard;
2858 new->output = dst_discard;
2859 dst_copy_metrics(new, &ort->dst);
2860
2861 new->dev = ort->dst.dev;
2862 if (new->dev)
2863 dev_hold(new->dev);
2864
2865 rt->rt_key_dst = ort->rt_key_dst;
2866 rt->rt_key_src = ort->rt_key_src;
2867 rt->rt_key_tos = ort->rt_key_tos;
2868 rt->rt_route_iif = ort->rt_route_iif;
2869 rt->rt_iif = ort->rt_iif;
2870 rt->rt_oif = ort->rt_oif;
2871 rt->rt_mark = ort->rt_mark;
2872
2873 rt->rt_genid = rt_genid(net);
2874 rt->rt_flags = ort->rt_flags;
2875 rt->rt_type = ort->rt_type;
2876 rt->rt_dst = ort->rt_dst;
2877 rt->rt_src = ort->rt_src;
2878 rt->rt_gateway = ort->rt_gateway;
2879 rt->rt_spec_dst = ort->rt_spec_dst;
2880 rt->peer = ort->peer;
2881 if (rt->peer)
2882 atomic_inc(&rt->peer->refcnt);
2883 rt->fi = ort->fi;
2884 if (rt->fi)
2885 atomic_inc(&rt->fi->fib_clntref);
2886
2887 dst_free(new);
2888 }
2889
2890 dst_release(dst_orig);
2891
2892 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2893 }
2894
2895 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2896 struct sock *sk)
2897 {
2898 struct rtable *rt = __ip_route_output_key(net, flp4);
2899
2900 if (IS_ERR(rt))
2901 return rt;
2902
2903 if (flp4->flowi4_proto)
2904 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2905 flowi4_to_flowi(flp4),
2906 sk, 0);
2907
2908 return rt;
2909 }
2910 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2911
2912 static int rt_fill_info(struct net *net,
2913 struct sk_buff *skb, u32 pid, u32 seq, int event,
2914 int nowait, unsigned int flags)
2915 {
2916 struct rtable *rt = skb_rtable(skb);
2917 struct rtmsg *r;
2918 struct nlmsghdr *nlh;
2919 long expires = 0;
2920 const struct inet_peer *peer = rt->peer;
2921 u32 id = 0, ts = 0, tsage = 0, error;
2922
2923 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2924 if (nlh == NULL)
2925 return -EMSGSIZE;
2926
2927 r = nlmsg_data(nlh);
2928 r->rtm_family = AF_INET;
2929 r->rtm_dst_len = 32;
2930 r->rtm_src_len = 0;
2931 r->rtm_tos = rt->rt_key_tos;
2932 r->rtm_table = RT_TABLE_MAIN;
2933 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2934 r->rtm_type = rt->rt_type;
2935 r->rtm_scope = RT_SCOPE_UNIVERSE;
2936 r->rtm_protocol = RTPROT_UNSPEC;
2937 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2938 if (rt->rt_flags & RTCF_NOTIFY)
2939 r->rtm_flags |= RTM_F_NOTIFY;
2940
2941 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2942
2943 if (rt->rt_key_src) {
2944 r->rtm_src_len = 32;
2945 NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
2946 }
2947 if (rt->dst.dev)
2948 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2949 #ifdef CONFIG_IP_ROUTE_CLASSID
2950 if (rt->dst.tclassid)
2951 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2952 #endif
2953 if (rt_is_input_route(rt))
2954 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2955 else if (rt->rt_src != rt->rt_key_src)
2956 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2957
2958 if (rt->rt_dst != rt->rt_gateway)
2959 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2960
2961 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2962 goto nla_put_failure;
2963
2964 if (rt->rt_mark)
2965 NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
2966
2967 error = rt->dst.error;
2968 if (peer) {
2969 inet_peer_refcheck(rt->peer);
2970 id = atomic_read(&peer->ip_id_count) & 0xffff;
2971 if (peer->tcp_ts_stamp) {
2972 ts = peer->tcp_ts;
2973 tsage = get_seconds() - peer->tcp_ts_stamp;
2974 }
2975 expires = ACCESS_ONCE(peer->pmtu_expires);
2976 if (expires)
2977 expires -= jiffies;
2978 }
2979
2980 if (rt_is_input_route(rt)) {
2981 #ifdef CONFIG_IP_MROUTE
2982 __be32 dst = rt->rt_dst;
2983
2984 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2985 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2986 int err = ipmr_get_route(net, skb,
2987 rt->rt_src, rt->rt_dst,
2988 r, nowait);
2989 if (err <= 0) {
2990 if (!nowait) {
2991 if (err == 0)
2992 return 0;
2993 goto nla_put_failure;
2994 } else {
2995 if (err == -EMSGSIZE)
2996 goto nla_put_failure;
2997 error = err;
2998 }
2999 }
3000 } else
3001 #endif
3002 NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
3003 }
3004
3005 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
3006 expires, error) < 0)
3007 goto nla_put_failure;
3008
3009 return nlmsg_end(skb, nlh);
3010
3011 nla_put_failure:
3012 nlmsg_cancel(skb, nlh);
3013 return -EMSGSIZE;
3014 }
3015
3016 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
3017 {
3018 struct net *net = sock_net(in_skb->sk);
3019 struct rtmsg *rtm;
3020 struct nlattr *tb[RTA_MAX+1];
3021 struct rtable *rt = NULL;
3022 __be32 dst = 0;
3023 __be32 src = 0;
3024 u32 iif;
3025 int err;
3026 int mark;
3027 struct sk_buff *skb;
3028
3029 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
3030 if (err < 0)
3031 goto errout;
3032
3033 rtm = nlmsg_data(nlh);
3034
3035 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3036 if (skb == NULL) {
3037 err = -ENOBUFS;
3038 goto errout;
3039 }
3040
3041 /* Reserve room for dummy headers, this skb can pass
3042 through good chunk of routing engine.
3043 */
3044 skb_reset_mac_header(skb);
3045 skb_reset_network_header(skb);
3046
3047 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3048 ip_hdr(skb)->protocol = IPPROTO_ICMP;
3049 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
3050
3051 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
3052 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
3053 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3054 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3055
3056 if (iif) {
3057 struct net_device *dev;
3058
3059 dev = __dev_get_by_index(net, iif);
3060 if (dev == NULL) {
3061 err = -ENODEV;
3062 goto errout_free;
3063 }
3064
3065 skb->protocol = htons(ETH_P_IP);
3066 skb->dev = dev;
3067 skb->mark = mark;
3068 local_bh_disable();
3069 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
3070 local_bh_enable();
3071
3072 rt = skb_rtable(skb);
3073 if (err == 0 && rt->dst.error)
3074 err = -rt->dst.error;
3075 } else {
3076 struct flowi4 fl4 = {
3077 .daddr = dst,
3078 .saddr = src,
3079 .flowi4_tos = rtm->rtm_tos,
3080 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3081 .flowi4_mark = mark,
3082 };
3083 rt = ip_route_output_key(net, &fl4);
3084
3085 err = 0;
3086 if (IS_ERR(rt))
3087 err = PTR_ERR(rt);
3088 }
3089
3090 if (err)
3091 goto errout_free;
3092
3093 skb_dst_set(skb, &rt->dst);
3094 if (rtm->rtm_flags & RTM_F_NOTIFY)
3095 rt->rt_flags |= RTCF_NOTIFY;
3096
3097 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3098 RTM_NEWROUTE, 0, 0);
3099 if (err <= 0)
3100 goto errout_free;
3101
3102 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3103 errout:
3104 return err;
3105
3106 errout_free:
3107 kfree_skb(skb);
3108 goto errout;
3109 }
3110
3111 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
3112 {
3113 struct rtable *rt;
3114 int h, s_h;
3115 int idx, s_idx;
3116 struct net *net;
3117
3118 net = sock_net(skb->sk);
3119
3120 s_h = cb->args[0];
3121 if (s_h < 0)
3122 s_h = 0;
3123 s_idx = idx = cb->args[1];
3124 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3125 if (!rt_hash_table[h].chain)
3126 continue;
3127 rcu_read_lock_bh();
3128 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3129 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3130 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3131 continue;
3132 if (rt_is_expired(rt))
3133 continue;
3134 skb_dst_set_noref(skb, &rt->dst);
3135 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3136 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3137 1, NLM_F_MULTI) <= 0) {
3138 skb_dst_drop(skb);
3139 rcu_read_unlock_bh();
3140 goto done;
3141 }
3142 skb_dst_drop(skb);
3143 }
3144 rcu_read_unlock_bh();
3145 }
3146
3147 done:
3148 cb->args[0] = h;
3149 cb->args[1] = idx;
3150 return skb->len;
3151 }
3152
3153 void ip_rt_multicast_event(struct in_device *in_dev)
3154 {
3155 rt_cache_flush(dev_net(in_dev->dev), 0);
3156 }
3157
3158 #ifdef CONFIG_SYSCTL
3159 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3160 void __user *buffer,
3161 size_t *lenp, loff_t *ppos)
3162 {
3163 if (write) {
3164 int flush_delay;
3165 ctl_table ctl;
3166 struct net *net;
3167
3168 memcpy(&ctl, __ctl, sizeof(ctl));
3169 ctl.data = &flush_delay;
3170 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3171
3172 net = (struct net *)__ctl->extra1;
3173 rt_cache_flush(net, flush_delay);
3174 return 0;
3175 }
3176
3177 return -EINVAL;
3178 }
3179
3180 static ctl_table ipv4_route_table[] = {
3181 {
3182 .procname = "gc_thresh",
3183 .data = &ipv4_dst_ops.gc_thresh,
3184 .maxlen = sizeof(int),
3185 .mode = 0644,
3186 .proc_handler = proc_dointvec,
3187 },
3188 {
3189 .procname = "max_size",
3190 .data = &ip_rt_max_size,
3191 .maxlen = sizeof(int),
3192 .mode = 0644,
3193 .proc_handler = proc_dointvec,
3194 },
3195 {
3196 /* Deprecated. Use gc_min_interval_ms */
3197
3198 .procname = "gc_min_interval",
3199 .data = &ip_rt_gc_min_interval,
3200 .maxlen = sizeof(int),
3201 .mode = 0644,
3202 .proc_handler = proc_dointvec_jiffies,
3203 },
3204 {
3205 .procname = "gc_min_interval_ms",
3206 .data = &ip_rt_gc_min_interval,
3207 .maxlen = sizeof(int),
3208 .mode = 0644,
3209 .proc_handler = proc_dointvec_ms_jiffies,
3210 },
3211 {
3212 .procname = "gc_timeout",
3213 .data = &ip_rt_gc_timeout,
3214 .maxlen = sizeof(int),
3215 .mode = 0644,
3216 .proc_handler = proc_dointvec_jiffies,
3217 },
3218 {
3219 .procname = "gc_interval",
3220 .data = &ip_rt_gc_interval,
3221 .maxlen = sizeof(int),
3222 .mode = 0644,
3223 .proc_handler = proc_dointvec_jiffies,
3224 },
3225 {
3226 .procname = "gc_interval",
3227 .data = &ip_rt_gc_interval,
3228 .maxlen = sizeof(int),
3229 .mode = 0644,
3230 .proc_handler = proc_dointvec_jiffies,
3231 },
3232 {
3233 .procname = "redirect_load",
3234 .data = &ip_rt_redirect_load,
3235 .maxlen = sizeof(int),
3236 .mode = 0644,
3237 .proc_handler = proc_dointvec,
3238 },
3239 {
3240 .procname = "redirect_number",
3241 .data = &ip_rt_redirect_number,
3242 .maxlen = sizeof(int),
3243 .mode = 0644,
3244 .proc_handler = proc_dointvec,
3245 },
3246 {
3247 .procname = "redirect_silence",
3248 .data = &ip_rt_redirect_silence,
3249 .maxlen = sizeof(int),
3250 .mode = 0644,
3251 .proc_handler = proc_dointvec,
3252 },
3253 {
3254 .procname = "error_cost",
3255 .data = &ip_rt_error_cost,
3256 .maxlen = sizeof(int),
3257 .mode = 0644,
3258 .proc_handler = proc_dointvec,
3259 },
3260 {
3261 .procname = "error_burst",
3262 .data = &ip_rt_error_burst,
3263 .maxlen = sizeof(int),
3264 .mode = 0644,
3265 .proc_handler = proc_dointvec,
3266 },
3267 {
3268 .procname = "gc_elasticity",
3269 .data = &ip_rt_gc_elasticity,
3270 .maxlen = sizeof(int),
3271 .mode = 0644,
3272 .proc_handler = proc_dointvec,
3273 },
3274 {
3275 .procname = "mtu_expires",
3276 .data = &ip_rt_mtu_expires,
3277 .maxlen = sizeof(int),
3278 .mode = 0644,
3279 .proc_handler = proc_dointvec_jiffies,
3280 },
3281 {
3282 .procname = "min_pmtu",
3283 .data = &ip_rt_min_pmtu,
3284 .maxlen = sizeof(int),
3285 .mode = 0644,
3286 .proc_handler = proc_dointvec,
3287 },
3288 {
3289 .procname = "min_adv_mss",
3290 .data = &ip_rt_min_advmss,
3291 .maxlen = sizeof(int),
3292 .mode = 0644,
3293 .proc_handler = proc_dointvec,
3294 },
3295 { }
3296 };
3297
3298 static struct ctl_table empty[1];
3299
3300 static struct ctl_table ipv4_skeleton[] =
3301 {
3302 { .procname = "route",
3303 .mode = 0555, .child = ipv4_route_table},
3304 { .procname = "neigh",
3305 .mode = 0555, .child = empty},
3306 { }
3307 };
3308
3309 static __net_initdata struct ctl_path ipv4_path[] = {
3310 { .procname = "net", },
3311 { .procname = "ipv4", },
3312 { },
3313 };
3314
3315 static struct ctl_table ipv4_route_flush_table[] = {
3316 {
3317 .procname = "flush",
3318 .maxlen = sizeof(int),
3319 .mode = 0200,
3320 .proc_handler = ipv4_sysctl_rtcache_flush,
3321 },
3322 { },
3323 };
3324
3325 static __net_initdata struct ctl_path ipv4_route_path[] = {
3326 { .procname = "net", },
3327 { .procname = "ipv4", },
3328 { .procname = "route", },
3329 { },
3330 };
3331
3332 static __net_init int sysctl_route_net_init(struct net *net)
3333 {
3334 struct ctl_table *tbl;
3335
3336 tbl = ipv4_route_flush_table;
3337 if (!net_eq(net, &init_net)) {
3338 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3339 if (tbl == NULL)
3340 goto err_dup;
3341 }
3342 tbl[0].extra1 = net;
3343
3344 net->ipv4.route_hdr =
3345 register_net_sysctl_table(net, ipv4_route_path, tbl);
3346 if (net->ipv4.route_hdr == NULL)
3347 goto err_reg;
3348 return 0;
3349
3350 err_reg:
3351 if (tbl != ipv4_route_flush_table)
3352 kfree(tbl);
3353 err_dup:
3354 return -ENOMEM;
3355 }
3356
3357 static __net_exit void sysctl_route_net_exit(struct net *net)
3358 {
3359 struct ctl_table *tbl;
3360
3361 tbl = net->ipv4.route_hdr->ctl_table_arg;
3362 unregister_net_sysctl_table(net->ipv4.route_hdr);
3363 BUG_ON(tbl == ipv4_route_flush_table);
3364 kfree(tbl);
3365 }
3366
3367 static __net_initdata struct pernet_operations sysctl_route_ops = {
3368 .init = sysctl_route_net_init,
3369 .exit = sysctl_route_net_exit,
3370 };
3371 #endif
3372
3373 static __net_init int rt_genid_init(struct net *net)
3374 {
3375 get_random_bytes(&net->ipv4.rt_genid,
3376 sizeof(net->ipv4.rt_genid));
3377 get_random_bytes(&net->ipv4.dev_addr_genid,
3378 sizeof(net->ipv4.dev_addr_genid));
3379 return 0;
3380 }
3381
3382 static __net_initdata struct pernet_operations rt_genid_ops = {
3383 .init = rt_genid_init,
3384 };
3385
3386
3387 #ifdef CONFIG_IP_ROUTE_CLASSID
3388 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3389 #endif /* CONFIG_IP_ROUTE_CLASSID */
3390
3391 static __initdata unsigned long rhash_entries;
3392 static int __init set_rhash_entries(char *str)
3393 {
3394 if (!str)
3395 return 0;
3396 rhash_entries = simple_strtoul(str, &str, 0);
3397 return 1;
3398 }
3399 __setup("rhash_entries=", set_rhash_entries);
3400
3401 int __init ip_rt_init(void)
3402 {
3403 int rc = 0;
3404
3405 #ifdef CONFIG_IP_ROUTE_CLASSID
3406 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3407 if (!ip_rt_acct)
3408 panic("IP: failed to allocate ip_rt_acct\n");
3409 #endif
3410
3411 ipv4_dst_ops.kmem_cachep =
3412 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3413 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3414
3415 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3416
3417 if (dst_entries_init(&ipv4_dst_ops) < 0)
3418 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3419
3420 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3421 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3422
3423 rt_hash_table = (struct rt_hash_bucket *)
3424 alloc_large_system_hash("IP route cache",
3425 sizeof(struct rt_hash_bucket),
3426 rhash_entries,
3427 (totalram_pages >= 128 * 1024) ?
3428 15 : 17,
3429 0,
3430 &rt_hash_log,
3431 &rt_hash_mask,
3432 rhash_entries ? 0 : 512 * 1024);
3433 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3434 rt_hash_lock_init();
3435
3436 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3437 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3438
3439 devinet_init();
3440 ip_fib_init();
3441
3442 INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3443 expires_ljiffies = jiffies;
3444 schedule_delayed_work(&expires_work,
3445 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3446
3447 if (ip_rt_proc_init())
3448 printk(KERN_ERR "Unable to create route proc files\n");
3449 #ifdef CONFIG_XFRM
3450 xfrm_init();
3451 xfrm4_init(ip_rt_max_size);
3452 #endif
3453 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3454
3455 #ifdef CONFIG_SYSCTL
3456 register_pernet_subsys(&sysctl_route_ops);
3457 #endif
3458 register_pernet_subsys(&rt_genid_ops);
3459 return rc;
3460 }
3461
3462 #ifdef CONFIG_SYSCTL
3463 /*
3464 * We really need to sanitize the damn ipv4 init order, then all
3465 * this nonsense will go away.
3466 */
3467 void __init ip_static_sysctl_init(void)
3468 {
3469 register_sysctl_paths(ipv4_path, ipv4_skeleton);
3470 }
3471 #endif
Linux with added FPC-III support