?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[CAN]: Add raw protocol
[linux-2.6/kvm.git] / net / ipv4 / route.c
bloba21021bf1409179e59d0be454df4f51e3e347c4e
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 * Version: $Id: route.c,v 1.103 2002/01/12 07:44:09 davem Exp $
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
14 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 *
16 * Fixes:
17 * Alan Cox : Verify area fixes.
18 * Alan Cox : cli() protects routing changes
19 * Rui Oliveira : ICMP routing table updates
20 * (rco@di.uminho.pt) Routing table insertion and update
21 * Linus Torvalds : Rewrote bits to be sensible
22 * Alan Cox : Added BSD route gw semantics
23 * Alan Cox : Super /proc >4K
24 * Alan Cox : MTU in route table
25 * Alan Cox : MSS actually. Also added the window
26 * clamper.
27 * Sam Lantinga : Fixed route matching in rt_del()
28 * Alan Cox : Routing cache support.
29 * Alan Cox : Removed compatibility cruft.
30 * Alan Cox : RTF_REJECT support.
31 * Alan Cox : TCP irtt support.
32 * Jonathan Naylor : Added Metric support.
33 * Miquel van Smoorenburg : BSD API fixes.
34 * Miquel van Smoorenburg : Metrics.
35 * Alan Cox : Use __u32 properly
36 * Alan Cox : Aligned routing errors more closely with BSD
37 * our system is still very different.
38 * Alan Cox : Faster /proc handling
39 * Alexey Kuznetsov : Massive rework to support tree based routing,
40 * routing caches and better behaviour.
41 *
42 * Olaf Erb : irtt wasn't being copied right.
43 * Bjorn Ekwall : Kerneld route support.
44 * Alan Cox : Multicast fixed (I hope)
45 * Pavel Krauz : Limited broadcast fixed
46 * Mike McLagan : Routing by source
47 * Alexey Kuznetsov : End of old history. Split to fib.c and
48 * route.c and rewritten from scratch.
49 * Andi Kleen : Load-limit warning messages.
50 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
51 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
52 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
53 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
54 * Marc Boucher : routing by fwmark
55 * Robert Olsson : Added rt_cache statistics
56 * Arnaldo C. Melo : Convert proc stuff to seq_file
57 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
58 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
59 * Ilia Sotnikov : Removed TOS from hash calculations
60 *
61 * This program is free software; you can redistribute it and/or
62 * modify it under the terms of the GNU General Public License
63 * as published by the Free Software Foundation; either version
64 * 2 of the License, or (at your option) any later version.
65 */
66
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <linux/bitops.h>
71 #include <linux/types.h>
72 #include <linux/kernel.h>
73 #include <linux/mm.h>
74 #include <linux/bootmem.h>
75 #include <linux/string.h>
76 #include <linux/socket.h>
77 #include <linux/sockios.h>
78 #include <linux/errno.h>
79 #include <linux/in.h>
80 #include <linux/inet.h>
81 #include <linux/netdevice.h>
82 #include <linux/proc_fs.h>
83 #include <linux/init.h>
84 #include <linux/workqueue.h>
85 #include <linux/skbuff.h>
86 #include <linux/inetdevice.h>
87 #include <linux/igmp.h>
88 #include <linux/pkt_sched.h>
89 #include <linux/mroute.h>
90 #include <linux/netfilter_ipv4.h>
91 #include <linux/random.h>
92 #include <linux/jhash.h>
93 #include <linux/rcupdate.h>
94 #include <linux/times.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
113 #define RT_FL_TOS(oldflp) \
114 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
115
116 #define IP_MAX_MTU 0xFFF0
117
118 #define RT_GC_TIMEOUT (300*HZ)
119
120 static int ip_rt_min_delay = 2 * HZ;
121 static int ip_rt_max_delay = 10 * HZ;
122 static int ip_rt_max_size;
123 static int ip_rt_gc_timeout = RT_GC_TIMEOUT;
124 static int ip_rt_gc_interval = 60 * HZ;
125 static int ip_rt_gc_min_interval = HZ / 2;
126 static int ip_rt_redirect_number = 9;
127 static int ip_rt_redirect_load = HZ / 50;
128 static int ip_rt_redirect_silence = ((HZ / 50) << (9 + 1));
129 static int ip_rt_error_cost = HZ;
130 static int ip_rt_error_burst = 5 * HZ;
131 static int ip_rt_gc_elasticity = 8;
132 static int ip_rt_mtu_expires = 10 * 60 * HZ;
133 static int ip_rt_min_pmtu = 512 + 20 + 20;
134 static int ip_rt_min_advmss = 256;
135 static int ip_rt_secret_interval = 10 * 60 * HZ;
136 static unsigned long rt_deadline;
137
138 #define RTprint(a...) printk(KERN_DEBUG a)
139
140 static struct timer_list rt_flush_timer;
141 static void rt_check_expire(struct work_struct *work);
142 static DECLARE_DELAYED_WORK(expires_work, rt_check_expire);
143 static struct timer_list rt_secret_timer;
144
145 /*
146 * Interface to generic destination cache.
147 */
148
149 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
150 static void ipv4_dst_destroy(struct dst_entry *dst);
151 static void ipv4_dst_ifdown(struct dst_entry *dst,
152 struct net_device *dev, int how);
153 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
154 static void ipv4_link_failure(struct sk_buff *skb);
155 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
156 static int rt_garbage_collect(void);
157
158
159 static struct dst_ops ipv4_dst_ops = {
160 .family = AF_INET,
161 .protocol = __constant_htons(ETH_P_IP),
162 .gc = rt_garbage_collect,
163 .check = ipv4_dst_check,
164 .destroy = ipv4_dst_destroy,
165 .ifdown = ipv4_dst_ifdown,
166 .negative_advice = ipv4_negative_advice,
167 .link_failure = ipv4_link_failure,
168 .update_pmtu = ip_rt_update_pmtu,
169 .local_out = ip_local_out,
170 .entry_size = sizeof(struct rtable),
171 };
172
173 #define ECN_OR_COST(class) TC_PRIO_##class
174
175 const __u8 ip_tos2prio[16] = {
176 TC_PRIO_BESTEFFORT,
177 ECN_OR_COST(FILLER),
178 TC_PRIO_BESTEFFORT,
179 ECN_OR_COST(BESTEFFORT),
180 TC_PRIO_BULK,
181 ECN_OR_COST(BULK),
182 TC_PRIO_BULK,
183 ECN_OR_COST(BULK),
184 TC_PRIO_INTERACTIVE,
185 ECN_OR_COST(INTERACTIVE),
186 TC_PRIO_INTERACTIVE,
187 ECN_OR_COST(INTERACTIVE),
188 TC_PRIO_INTERACTIVE_BULK,
189 ECN_OR_COST(INTERACTIVE_BULK),
190 TC_PRIO_INTERACTIVE_BULK,
191 ECN_OR_COST(INTERACTIVE_BULK)
192 };
193
194
195 /*
196 * Route cache.
197 */
198
199 /* The locking scheme is rather straight forward:
200 *
201 * 1) Read-Copy Update protects the buckets of the central route hash.
202 * 2) Only writers remove entries, and they hold the lock
203 * as they look at rtable reference counts.
204 * 3) Only readers acquire references to rtable entries,
205 * they do so with atomic increments and with the
206 * lock held.
207 */
208
209 struct rt_hash_bucket {
210 struct rtable *chain;
211 };
212 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
213 defined(CONFIG_PROVE_LOCKING)
214 /*
215 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
216 * The size of this table is a power of two and depends on the number of CPUS.
217 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
218 */
219 #ifdef CONFIG_LOCKDEP
220 # define RT_HASH_LOCK_SZ 256
221 #else
222 # if NR_CPUS >= 32
223 # define RT_HASH_LOCK_SZ 4096
224 # elif NR_CPUS >= 16
225 # define RT_HASH_LOCK_SZ 2048
226 # elif NR_CPUS >= 8
227 # define RT_HASH_LOCK_SZ 1024
228 # elif NR_CPUS >= 4
229 # define RT_HASH_LOCK_SZ 512
230 # else
231 # define RT_HASH_LOCK_SZ 256
232 # endif
233 #endif
234
235 static spinlock_t *rt_hash_locks;
236 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
237 # define rt_hash_lock_init() { \
238 int i; \
239 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ, GFP_KERNEL); \
240 if (!rt_hash_locks) panic("IP: failed to allocate rt_hash_locks\n"); \
241 for (i = 0; i < RT_HASH_LOCK_SZ; i++) \
242 spin_lock_init(&rt_hash_locks[i]); \
243 }
244 #else
245 # define rt_hash_lock_addr(slot) NULL
246 # define rt_hash_lock_init()
247 #endif
248
249 static struct rt_hash_bucket *rt_hash_table;
250 static unsigned rt_hash_mask;
251 static unsigned int rt_hash_log;
252 static unsigned int rt_hash_rnd;
253
254 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
255 #define RT_CACHE_STAT_INC(field) \
256 (__raw_get_cpu_var(rt_cache_stat).field++)
257
258 static int rt_intern_hash(unsigned hash, struct rtable *rth,
259 struct rtable **res);
260
261 static unsigned int rt_hash_code(u32 daddr, u32 saddr)
262 {
263 return (jhash_2words(daddr, saddr, rt_hash_rnd)
264 & rt_hash_mask);
265 }
266
267 #define rt_hash(daddr, saddr, idx) \
268 rt_hash_code((__force u32)(__be32)(daddr),\
269 (__force u32)(__be32)(saddr) ^ ((idx) << 5))
270
271 #ifdef CONFIG_PROC_FS
272 struct rt_cache_iter_state {
273 int bucket;
274 };
275
276 static struct rtable *rt_cache_get_first(struct seq_file *seq)
277 {
278 struct rtable *r = NULL;
279 struct rt_cache_iter_state *st = seq->private;
280
281 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
282 rcu_read_lock_bh();
283 r = rt_hash_table[st->bucket].chain;
284 if (r)
285 break;
286 rcu_read_unlock_bh();
287 }
288 return rcu_dereference(r);
289 }
290
291 static struct rtable *rt_cache_get_next(struct seq_file *seq, struct rtable *r)
292 {
293 struct rt_cache_iter_state *st = seq->private;
294
295 r = r->u.dst.rt_next;
296 while (!r) {
297 rcu_read_unlock_bh();
298 if (--st->bucket < 0)
299 break;
300 rcu_read_lock_bh();
301 r = rt_hash_table[st->bucket].chain;
302 }
303 return rcu_dereference(r);
304 }
305
306 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
307 {
308 struct rtable *r = rt_cache_get_first(seq);
309
310 if (r)
311 while (pos && (r = rt_cache_get_next(seq, r)))
312 --pos;
313 return pos ? NULL : r;
314 }
315
316 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
317 {
318 return *pos ? rt_cache_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
319 }
320
321 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
322 {
323 struct rtable *r = NULL;
324
325 if (v == SEQ_START_TOKEN)
326 r = rt_cache_get_first(seq);
327 else
328 r = rt_cache_get_next(seq, v);
329 ++*pos;
330 return r;
331 }
332
333 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
334 {
335 if (v && v != SEQ_START_TOKEN)
336 rcu_read_unlock_bh();
337 }
338
339 static int rt_cache_seq_show(struct seq_file *seq, void *v)
340 {
341 if (v == SEQ_START_TOKEN)
342 seq_printf(seq, "%-127s\n",
343 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
344 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
345 "HHUptod\tSpecDst");
346 else {
347 struct rtable *r = v;
348 char temp[256];
349
350 sprintf(temp, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
351 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X",
352 r->u.dst.dev ? r->u.dst.dev->name : "*",
353 (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
354 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
355 r->u.dst.__use, 0, (unsigned long)r->rt_src,
356 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
357 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
358 dst_metric(&r->u.dst, RTAX_WINDOW),
359 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
360 dst_metric(&r->u.dst, RTAX_RTTVAR)),
361 r->fl.fl4_tos,
362 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
363 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
364 dev_queue_xmit) : 0,
365 r->rt_spec_dst);
366 seq_printf(seq, "%-127s\n", temp);
367 }
368 return 0;
369 }
370
371 static const struct seq_operations rt_cache_seq_ops = {
372 .start = rt_cache_seq_start,
373 .next = rt_cache_seq_next,
374 .stop = rt_cache_seq_stop,
375 .show = rt_cache_seq_show,
376 };
377
378 static int rt_cache_seq_open(struct inode *inode, struct file *file)
379 {
380 return seq_open_private(file, &rt_cache_seq_ops,
381 sizeof(struct rt_cache_iter_state));
382 }
383
384 static const struct file_operations rt_cache_seq_fops = {
385 .owner = THIS_MODULE,
386 .open = rt_cache_seq_open,
387 .read = seq_read,
388 .llseek = seq_lseek,
389 .release = seq_release_private,
390 };
391
392
393 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
394 {
395 int cpu;
396
397 if (*pos == 0)
398 return SEQ_START_TOKEN;
399
400 for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
401 if (!cpu_possible(cpu))
402 continue;
403 *pos = cpu+1;
404 return &per_cpu(rt_cache_stat, cpu);
405 }
406 return NULL;
407 }
408
409 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
410 {
411 int cpu;
412
413 for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
414 if (!cpu_possible(cpu))
415 continue;
416 *pos = cpu+1;
417 return &per_cpu(rt_cache_stat, cpu);
418 }
419 return NULL;
420
421 }
422
423 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
424 {
425
426 }
427
428 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
429 {
430 struct rt_cache_stat *st = v;
431
432 if (v == SEQ_START_TOKEN) {
433 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");
434 return 0;
435 }
436
437 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
438 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
439 atomic_read(&ipv4_dst_ops.entries),
440 st->in_hit,
441 st->in_slow_tot,
442 st->in_slow_mc,
443 st->in_no_route,
444 st->in_brd,
445 st->in_martian_dst,
446 st->in_martian_src,
447
448 st->out_hit,
449 st->out_slow_tot,
450 st->out_slow_mc,
451
452 st->gc_total,
453 st->gc_ignored,
454 st->gc_goal_miss,
455 st->gc_dst_overflow,
456 st->in_hlist_search,
457 st->out_hlist_search
458 );
459 return 0;
460 }
461
462 static const struct seq_operations rt_cpu_seq_ops = {
463 .start = rt_cpu_seq_start,
464 .next = rt_cpu_seq_next,
465 .stop = rt_cpu_seq_stop,
466 .show = rt_cpu_seq_show,
467 };
468
469
470 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
471 {
472 return seq_open(file, &rt_cpu_seq_ops);
473 }
474
475 static const struct file_operations rt_cpu_seq_fops = {
476 .owner = THIS_MODULE,
477 .open = rt_cpu_seq_open,
478 .read = seq_read,
479 .llseek = seq_lseek,
480 .release = seq_release,
481 };
482
483 #endif /* CONFIG_PROC_FS */
484
485 static __inline__ void rt_free(struct rtable *rt)
486 {
487 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
488 }
489
490 static __inline__ void rt_drop(struct rtable *rt)
491 {
492 ip_rt_put(rt);
493 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
494 }
495
496 static __inline__ int rt_fast_clean(struct rtable *rth)
497 {
498 /* Kill broadcast/multicast entries very aggresively, if they
499 collide in hash table with more useful entries */
500 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
501 rth->fl.iif && rth->u.dst.rt_next;
502 }
503
504 static __inline__ int rt_valuable(struct rtable *rth)
505 {
506 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
507 rth->u.dst.expires;
508 }
509
510 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
511 {
512 unsigned long age;
513 int ret = 0;
514
515 if (atomic_read(&rth->u.dst.__refcnt))
516 goto out;
517
518 ret = 1;
519 if (rth->u.dst.expires &&
520 time_after_eq(jiffies, rth->u.dst.expires))
521 goto out;
522
523 age = jiffies - rth->u.dst.lastuse;
524 ret = 0;
525 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
526 (age <= tmo2 && rt_valuable(rth)))
527 goto out;
528 ret = 1;
529 out: return ret;
530 }
531
532 /* Bits of score are:
533 * 31: very valuable
534 * 30: not quite useless
535 * 29..0: usage counter
536 */
537 static inline u32 rt_score(struct rtable *rt)
538 {
539 u32 score = jiffies - rt->u.dst.lastuse;
540
541 score = ~score & ~(3<<30);
542
543 if (rt_valuable(rt))
544 score |= (1<<31);
545
546 if (!rt->fl.iif ||
547 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
548 score |= (1<<30);
549
550 return score;
551 }
552
553 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
554 {
555 return ((__force u32)((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
556 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr)) |
557 (fl1->mark ^ fl2->mark) |
558 (*(u16 *)&fl1->nl_u.ip4_u.tos ^
559 *(u16 *)&fl2->nl_u.ip4_u.tos) |
560 (fl1->oif ^ fl2->oif) |
561 (fl1->iif ^ fl2->iif)) == 0;
562 }
563
564 static void rt_check_expire(struct work_struct *work)
565 {
566 static unsigned int rover;
567 unsigned int i = rover, goal;
568 struct rtable *rth, **rthp;
569 u64 mult;
570
571 mult = ((u64)ip_rt_gc_interval) << rt_hash_log;
572 if (ip_rt_gc_timeout > 1)
573 do_div(mult, ip_rt_gc_timeout);
574 goal = (unsigned int)mult;
575 if (goal > rt_hash_mask)
576 goal = rt_hash_mask + 1;
577 for (; goal > 0; goal--) {
578 unsigned long tmo = ip_rt_gc_timeout;
579
580 i = (i + 1) & rt_hash_mask;
581 rthp = &rt_hash_table[i].chain;
582
583 if (need_resched())
584 cond_resched();
585
586 if (*rthp == NULL)
587 continue;
588 spin_lock_bh(rt_hash_lock_addr(i));
589 while ((rth = *rthp) != NULL) {
590 if (rth->u.dst.expires) {
591 /* Entry is expired even if it is in use */
592 if (time_before_eq(jiffies, rth->u.dst.expires)) {
593 tmo >>= 1;
594 rthp = &rth->u.dst.rt_next;
595 continue;
596 }
597 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
598 tmo >>= 1;
599 rthp = &rth->u.dst.rt_next;
600 continue;
601 }
602
603 /* Cleanup aged off entries. */
604 *rthp = rth->u.dst.rt_next;
605 rt_free(rth);
606 }
607 spin_unlock_bh(rt_hash_lock_addr(i));
608 }
609 rover = i;
610 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
611 }
612
613 /* This can run from both BH and non-BH contexts, the latter
614 * in the case of a forced flush event.
615 */
616 static void rt_run_flush(unsigned long dummy)
617 {
618 int i;
619 struct rtable *rth, *next;
620
621 rt_deadline = 0;
622
623 get_random_bytes(&rt_hash_rnd, 4);
624
625 for (i = rt_hash_mask; i >= 0; i--) {
626 spin_lock_bh(rt_hash_lock_addr(i));
627 rth = rt_hash_table[i].chain;
628 if (rth)
629 rt_hash_table[i].chain = NULL;
630 spin_unlock_bh(rt_hash_lock_addr(i));
631
632 for (; rth; rth = next) {
633 next = rth->u.dst.rt_next;
634 rt_free(rth);
635 }
636 }
637 }
638
639 static DEFINE_SPINLOCK(rt_flush_lock);
640
641 void rt_cache_flush(int delay)
642 {
643 unsigned long now = jiffies;
644 int user_mode = !in_softirq();
645
646 if (delay < 0)
647 delay = ip_rt_min_delay;
648
649 spin_lock_bh(&rt_flush_lock);
650
651 if (del_timer(&rt_flush_timer) && delay > 0 && rt_deadline) {
652 long tmo = (long)(rt_deadline - now);
653
654 /* If flush timer is already running
655 and flush request is not immediate (delay > 0):
656
657 if deadline is not achieved, prolongate timer to "delay",
658 otherwise fire it at deadline time.
659 */
660
661 if (user_mode && tmo < ip_rt_max_delay-ip_rt_min_delay)
662 tmo = 0;
663
664 if (delay > tmo)
665 delay = tmo;
666 }
667
668 if (delay <= 0) {
669 spin_unlock_bh(&rt_flush_lock);
670 rt_run_flush(0);
671 return;
672 }
673
674 if (rt_deadline == 0)
675 rt_deadline = now + ip_rt_max_delay;
676
677 mod_timer(&rt_flush_timer, now+delay);
678 spin_unlock_bh(&rt_flush_lock);
679 }
680
681 static void rt_secret_rebuild(unsigned long dummy)
682 {
683 unsigned long now = jiffies;
684
685 rt_cache_flush(0);
686 mod_timer(&rt_secret_timer, now + ip_rt_secret_interval);
687 }
688
689 /*
690 Short description of GC goals.
691
692 We want to build algorithm, which will keep routing cache
693 at some equilibrium point, when number of aged off entries
694 is kept approximately equal to newly generated ones.
695
696 Current expiration strength is variable "expire".
697 We try to adjust it dynamically, so that if networking
698 is idle expires is large enough to keep enough of warm entries,
699 and when load increases it reduces to limit cache size.
700 */
701
702 static int rt_garbage_collect(void)
703 {
704 static unsigned long expire = RT_GC_TIMEOUT;
705 static unsigned long last_gc;
706 static int rover;
707 static int equilibrium;
708 struct rtable *rth, **rthp;
709 unsigned long now = jiffies;
710 int goal;
711
712 /*
713 * Garbage collection is pretty expensive,
714 * do not make it too frequently.
715 */
716
717 RT_CACHE_STAT_INC(gc_total);
718
719 if (now - last_gc < ip_rt_gc_min_interval &&
720 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
721 RT_CACHE_STAT_INC(gc_ignored);
722 goto out;
723 }
724
725 /* Calculate number of entries, which we want to expire now. */
726 goal = atomic_read(&ipv4_dst_ops.entries) -
727 (ip_rt_gc_elasticity << rt_hash_log);
728 if (goal <= 0) {
729 if (equilibrium < ipv4_dst_ops.gc_thresh)
730 equilibrium = ipv4_dst_ops.gc_thresh;
731 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
732 if (goal > 0) {
733 equilibrium += min_t(unsigned int, goal / 2, rt_hash_mask + 1);
734 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
735 }
736 } else {
737 /* We are in dangerous area. Try to reduce cache really
738 * aggressively.
739 */
740 goal = max_t(unsigned int, goal / 2, rt_hash_mask + 1);
741 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
742 }
743
744 if (now - last_gc >= ip_rt_gc_min_interval)
745 last_gc = now;
746
747 if (goal <= 0) {
748 equilibrium += goal;
749 goto work_done;
750 }
751
752 do {
753 int i, k;
754
755 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
756 unsigned long tmo = expire;
757
758 k = (k + 1) & rt_hash_mask;
759 rthp = &rt_hash_table[k].chain;
760 spin_lock_bh(rt_hash_lock_addr(k));
761 while ((rth = *rthp) != NULL) {
762 if (!rt_may_expire(rth, tmo, expire)) {
763 tmo >>= 1;
764 rthp = &rth->u.dst.rt_next;
765 continue;
766 }
767 *rthp = rth->u.dst.rt_next;
768 rt_free(rth);
769 goal--;
770 }
771 spin_unlock_bh(rt_hash_lock_addr(k));
772 if (goal <= 0)
773 break;
774 }
775 rover = k;
776
777 if (goal <= 0)
778 goto work_done;
779
780 /* Goal is not achieved. We stop process if:
781
782 - if expire reduced to zero. Otherwise, expire is halfed.
783 - if table is not full.
784 - if we are called from interrupt.
785 - jiffies check is just fallback/debug loop breaker.
786 We will not spin here for long time in any case.
787 */
788
789 RT_CACHE_STAT_INC(gc_goal_miss);
790
791 if (expire == 0)
792 break;
793
794 expire >>= 1;
795 #if RT_CACHE_DEBUG >= 2
796 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
797 atomic_read(&ipv4_dst_ops.entries), goal, i);
798 #endif
799
800 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
801 goto out;
802 } while (!in_softirq() && time_before_eq(jiffies, now));
803
804 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
805 goto out;
806 if (net_ratelimit())
807 printk(KERN_WARNING "dst cache overflow\n");
808 RT_CACHE_STAT_INC(gc_dst_overflow);
809 return 1;
810
811 work_done:
812 expire += ip_rt_gc_min_interval;
813 if (expire > ip_rt_gc_timeout ||
814 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
815 expire = ip_rt_gc_timeout;
816 #if RT_CACHE_DEBUG >= 2
817 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
818 atomic_read(&ipv4_dst_ops.entries), goal, rover);
819 #endif
820 out: return 0;
821 }
822
823 static int rt_intern_hash(unsigned hash, struct rtable *rt, struct rtable **rp)
824 {
825 struct rtable *rth, **rthp;
826 unsigned long now;
827 struct rtable *cand, **candp;
828 u32 min_score;
829 int chain_length;
830 int attempts = !in_softirq();
831
832 restart:
833 chain_length = 0;
834 min_score = ~(u32)0;
835 cand = NULL;
836 candp = NULL;
837 now = jiffies;
838
839 rthp = &rt_hash_table[hash].chain;
840
841 spin_lock_bh(rt_hash_lock_addr(hash));
842 while ((rth = *rthp) != NULL) {
843 if (compare_keys(&rth->fl, &rt->fl)) {
844 /* Put it first */
845 *rthp = rth->u.dst.rt_next;
846 /*
847 * Since lookup is lockfree, the deletion
848 * must be visible to another weakly ordered CPU before
849 * the insertion at the start of the hash chain.
850 */
851 rcu_assign_pointer(rth->u.dst.rt_next,
852 rt_hash_table[hash].chain);
853 /*
854 * Since lookup is lockfree, the update writes
855 * must be ordered for consistency on SMP.
856 */
857 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
858
859 dst_use(&rth->u.dst, now);
860 spin_unlock_bh(rt_hash_lock_addr(hash));
861
862 rt_drop(rt);
863 *rp = rth;
864 return 0;
865 }
866
867 if (!atomic_read(&rth->u.dst.__refcnt)) {
868 u32 score = rt_score(rth);
869
870 if (score <= min_score) {
871 cand = rth;
872 candp = rthp;
873 min_score = score;
874 }
875 }
876
877 chain_length++;
878
879 rthp = &rth->u.dst.rt_next;
880 }
881
882 if (cand) {
883 /* ip_rt_gc_elasticity used to be average length of chain
884 * length, when exceeded gc becomes really aggressive.
885 *
886 * The second limit is less certain. At the moment it allows
887 * only 2 entries per bucket. We will see.
888 */
889 if (chain_length > ip_rt_gc_elasticity) {
890 *candp = cand->u.dst.rt_next;
891 rt_free(cand);
892 }
893 }
894
895 /* Try to bind route to arp only if it is output
896 route or unicast forwarding path.
897 */
898 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
899 int err = arp_bind_neighbour(&rt->u.dst);
900 if (err) {
901 spin_unlock_bh(rt_hash_lock_addr(hash));
902
903 if (err != -ENOBUFS) {
904 rt_drop(rt);
905 return err;
906 }
907
908 /* Neighbour tables are full and nothing
909 can be released. Try to shrink route cache,
910 it is most likely it holds some neighbour records.
911 */
912 if (attempts-- > 0) {
913 int saved_elasticity = ip_rt_gc_elasticity;
914 int saved_int = ip_rt_gc_min_interval;
915 ip_rt_gc_elasticity = 1;
916 ip_rt_gc_min_interval = 0;
917 rt_garbage_collect();
918 ip_rt_gc_min_interval = saved_int;
919 ip_rt_gc_elasticity = saved_elasticity;
920 goto restart;
921 }
922
923 if (net_ratelimit())
924 printk(KERN_WARNING "Neighbour table overflow.\n");
925 rt_drop(rt);
926 return -ENOBUFS;
927 }
928 }
929
930 rt->u.dst.rt_next = rt_hash_table[hash].chain;
931 #if RT_CACHE_DEBUG >= 2
932 if (rt->u.dst.rt_next) {
933 struct rtable *trt;
934 printk(KERN_DEBUG "rt_cache @%02x: %u.%u.%u.%u", hash,
935 NIPQUAD(rt->rt_dst));
936 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
937 printk(" . %u.%u.%u.%u", NIPQUAD(trt->rt_dst));
938 printk("\n");
939 }
940 #endif
941 rt_hash_table[hash].chain = rt;
942 spin_unlock_bh(rt_hash_lock_addr(hash));
943 *rp = rt;
944 return 0;
945 }
946
947 void rt_bind_peer(struct rtable *rt, int create)
948 {
949 static DEFINE_SPINLOCK(rt_peer_lock);
950 struct inet_peer *peer;
951
952 peer = inet_getpeer(rt->rt_dst, create);
953
954 spin_lock_bh(&rt_peer_lock);
955 if (rt->peer == NULL) {
956 rt->peer = peer;
957 peer = NULL;
958 }
959 spin_unlock_bh(&rt_peer_lock);
960 if (peer)
961 inet_putpeer(peer);
962 }
963
964 /*
965 * Peer allocation may fail only in serious out-of-memory conditions. However
966 * we still can generate some output.
967 * Random ID selection looks a bit dangerous because we have no chances to
968 * select ID being unique in a reasonable period of time.
969 * But broken packet identifier may be better than no packet at all.
970 */
971 static void ip_select_fb_ident(struct iphdr *iph)
972 {
973 static DEFINE_SPINLOCK(ip_fb_id_lock);
974 static u32 ip_fallback_id;
975 u32 salt;
976
977 spin_lock_bh(&ip_fb_id_lock);
978 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
979 iph->id = htons(salt & 0xFFFF);
980 ip_fallback_id = salt;
981 spin_unlock_bh(&ip_fb_id_lock);
982 }
983
984 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
985 {
986 struct rtable *rt = (struct rtable *) dst;
987
988 if (rt) {
989 if (rt->peer == NULL)
990 rt_bind_peer(rt, 1);
991
992 /* If peer is attached to destination, it is never detached,
993 so that we need not to grab a lock to dereference it.
994 */
995 if (rt->peer) {
996 iph->id = htons(inet_getid(rt->peer, more));
997 return;
998 }
999 } else
1000 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1001 __builtin_return_address(0));
1002
1003 ip_select_fb_ident(iph);
1004 }
1005
1006 static void rt_del(unsigned hash, struct rtable *rt)
1007 {
1008 struct rtable **rthp;
1009
1010 spin_lock_bh(rt_hash_lock_addr(hash));
1011 ip_rt_put(rt);
1012 for (rthp = &rt_hash_table[hash].chain; *rthp;
1013 rthp = &(*rthp)->u.dst.rt_next)
1014 if (*rthp == rt) {
1015 *rthp = rt->u.dst.rt_next;
1016 rt_free(rt);
1017 break;
1018 }
1019 spin_unlock_bh(rt_hash_lock_addr(hash));
1020 }
1021
1022 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1023 __be32 saddr, struct net_device *dev)
1024 {
1025 int i, k;
1026 struct in_device *in_dev = in_dev_get(dev);
1027 struct rtable *rth, **rthp;
1028 __be32 skeys[2] = { saddr, 0 };
1029 int ikeys[2] = { dev->ifindex, 0 };
1030 struct netevent_redirect netevent;
1031
1032 if (!in_dev)
1033 return;
1034
1035 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev)
1036 || MULTICAST(new_gw) || BADCLASS(new_gw) || ZERONET(new_gw))
1037 goto reject_redirect;
1038
1039 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1040 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1041 goto reject_redirect;
1042 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1043 goto reject_redirect;
1044 } else {
1045 if (inet_addr_type(new_gw) != RTN_UNICAST)
1046 goto reject_redirect;
1047 }
1048
1049 for (i = 0; i < 2; i++) {
1050 for (k = 0; k < 2; k++) {
1051 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k]);
1052
1053 rthp=&rt_hash_table[hash].chain;
1054
1055 rcu_read_lock();
1056 while ((rth = rcu_dereference(*rthp)) != NULL) {
1057 struct rtable *rt;
1058
1059 if (rth->fl.fl4_dst != daddr ||
1060 rth->fl.fl4_src != skeys[i] ||
1061 rth->fl.oif != ikeys[k] ||
1062 rth->fl.iif != 0) {
1063 rthp = &rth->u.dst.rt_next;
1064 continue;
1065 }
1066
1067 if (rth->rt_dst != daddr ||
1068 rth->rt_src != saddr ||
1069 rth->u.dst.error ||
1070 rth->rt_gateway != old_gw ||
1071 rth->u.dst.dev != dev)
1072 break;
1073
1074 dst_hold(&rth->u.dst);
1075 rcu_read_unlock();
1076
1077 rt = dst_alloc(&ipv4_dst_ops);
1078 if (rt == NULL) {
1079 ip_rt_put(rth);
1080 in_dev_put(in_dev);
1081 return;
1082 }
1083
1084 /* Copy all the information. */
1085 *rt = *rth;
1086 INIT_RCU_HEAD(&rt->u.dst.rcu_head);
1087 rt->u.dst.__use = 1;
1088 atomic_set(&rt->u.dst.__refcnt, 1);
1089 rt->u.dst.child = NULL;
1090 if (rt->u.dst.dev)
1091 dev_hold(rt->u.dst.dev);
1092 if (rt->idev)
1093 in_dev_hold(rt->idev);
1094 rt->u.dst.obsolete = 0;
1095 rt->u.dst.lastuse = jiffies;
1096 rt->u.dst.path = &rt->u.dst;
1097 rt->u.dst.neighbour = NULL;
1098 rt->u.dst.hh = NULL;
1099 rt->u.dst.xfrm = NULL;
1100
1101 rt->rt_flags |= RTCF_REDIRECTED;
1102
1103 /* Gateway is different ... */
1104 rt->rt_gateway = new_gw;
1105
1106 /* Redirect received -> path was valid */
1107 dst_confirm(&rth->u.dst);
1108
1109 if (rt->peer)
1110 atomic_inc(&rt->peer->refcnt);
1111
1112 if (arp_bind_neighbour(&rt->u.dst) ||
1113 !(rt->u.dst.neighbour->nud_state &
1114 NUD_VALID)) {
1115 if (rt->u.dst.neighbour)
1116 neigh_event_send(rt->u.dst.neighbour, NULL);
1117 ip_rt_put(rth);
1118 rt_drop(rt);
1119 goto do_next;
1120 }
1121
1122 netevent.old = &rth->u.dst;
1123 netevent.new = &rt->u.dst;
1124 call_netevent_notifiers(NETEVENT_REDIRECT,
1125 &netevent);
1126
1127 rt_del(hash, rth);
1128 if (!rt_intern_hash(hash, rt, &rt))
1129 ip_rt_put(rt);
1130 goto do_next;
1131 }
1132 rcu_read_unlock();
1133 do_next:
1134 ;
1135 }
1136 }
1137 in_dev_put(in_dev);
1138 return;
1139
1140 reject_redirect:
1141 #ifdef CONFIG_IP_ROUTE_VERBOSE
1142 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1143 printk(KERN_INFO "Redirect from %u.%u.%u.%u on %s about "
1144 "%u.%u.%u.%u ignored.\n"
1145 " Advised path = %u.%u.%u.%u -> %u.%u.%u.%u\n",
1146 NIPQUAD(old_gw), dev->name, NIPQUAD(new_gw),
1147 NIPQUAD(saddr), NIPQUAD(daddr));
1148 #endif
1149 in_dev_put(in_dev);
1150 }
1151
1152 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1153 {
1154 struct rtable *rt = (struct rtable*)dst;
1155 struct dst_entry *ret = dst;
1156
1157 if (rt) {
1158 if (dst->obsolete) {
1159 ip_rt_put(rt);
1160 ret = NULL;
1161 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1162 rt->u.dst.expires) {
1163 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1164 rt->fl.oif);
1165 #if RT_CACHE_DEBUG >= 1
1166 printk(KERN_DEBUG "ipv4_negative_advice: redirect to "
1167 "%u.%u.%u.%u/%02x dropped\n",
1168 NIPQUAD(rt->rt_dst), rt->fl.fl4_tos);
1169 #endif
1170 rt_del(hash, rt);
1171 ret = NULL;
1172 }
1173 }
1174 return ret;
1175 }
1176
1177 /*
1178 * Algorithm:
1179 * 1. The first ip_rt_redirect_number redirects are sent
1180 * with exponential backoff, then we stop sending them at all,
1181 * assuming that the host ignores our redirects.
1182 * 2. If we did not see packets requiring redirects
1183 * during ip_rt_redirect_silence, we assume that the host
1184 * forgot redirected route and start to send redirects again.
1185 *
1186 * This algorithm is much cheaper and more intelligent than dumb load limiting
1187 * in icmp.c.
1188 *
1189 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1190 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1191 */
1192
1193 void ip_rt_send_redirect(struct sk_buff *skb)
1194 {
1195 struct rtable *rt = (struct rtable*)skb->dst;
1196 struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
1197
1198 if (!in_dev)
1199 return;
1200
1201 if (!IN_DEV_TX_REDIRECTS(in_dev))
1202 goto out;
1203
1204 /* No redirected packets during ip_rt_redirect_silence;
1205 * reset the algorithm.
1206 */
1207 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1208 rt->u.dst.rate_tokens = 0;
1209
1210 /* Too many ignored redirects; do not send anything
1211 * set u.dst.rate_last to the last seen redirected packet.
1212 */
1213 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1214 rt->u.dst.rate_last = jiffies;
1215 goto out;
1216 }
1217
1218 /* Check for load limit; set rate_last to the latest sent
1219 * redirect.
1220 */
1221 if (rt->u.dst.rate_tokens == 0 ||
1222 time_after(jiffies,
1223 (rt->u.dst.rate_last +
1224 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1225 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1226 rt->u.dst.rate_last = jiffies;
1227 ++rt->u.dst.rate_tokens;
1228 #ifdef CONFIG_IP_ROUTE_VERBOSE
1229 if (IN_DEV_LOG_MARTIANS(in_dev) &&
1230 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1231 net_ratelimit())
1232 printk(KERN_WARNING "host %u.%u.%u.%u/if%d ignores "
1233 "redirects for %u.%u.%u.%u to %u.%u.%u.%u.\n",
1234 NIPQUAD(rt->rt_src), rt->rt_iif,
1235 NIPQUAD(rt->rt_dst), NIPQUAD(rt->rt_gateway));
1236 #endif
1237 }
1238 out:
1239 in_dev_put(in_dev);
1240 }
1241
1242 static int ip_error(struct sk_buff *skb)
1243 {
1244 struct rtable *rt = (struct rtable*)skb->dst;
1245 unsigned long now;
1246 int code;
1247
1248 switch (rt->u.dst.error) {
1249 case EINVAL:
1250 default:
1251 goto out;
1252 case EHOSTUNREACH:
1253 code = ICMP_HOST_UNREACH;
1254 break;
1255 case ENETUNREACH:
1256 code = ICMP_NET_UNREACH;
1257 IP_INC_STATS_BH(IPSTATS_MIB_INNOROUTES);
1258 break;
1259 case EACCES:
1260 code = ICMP_PKT_FILTERED;
1261 break;
1262 }
1263
1264 now = jiffies;
1265 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1266 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1267 rt->u.dst.rate_tokens = ip_rt_error_burst;
1268 rt->u.dst.rate_last = now;
1269 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1270 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1271 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1272 }
1273
1274 out: kfree_skb(skb);
1275 return 0;
1276 }
1277
1278 /*
1279 * The last two values are not from the RFC but
1280 * are needed for AMPRnet AX.25 paths.
1281 */
1282
1283 static const unsigned short mtu_plateau[] =
1284 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1285
1286 static __inline__ unsigned short guess_mtu(unsigned short old_mtu)
1287 {
1288 int i;
1289
1290 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1291 if (old_mtu > mtu_plateau[i])
1292 return mtu_plateau[i];
1293 return 68;
1294 }
1295
1296 unsigned short ip_rt_frag_needed(struct iphdr *iph, unsigned short new_mtu)
1297 {
1298 int i;
1299 unsigned short old_mtu = ntohs(iph->tot_len);
1300 struct rtable *rth;
1301 __be32 skeys[2] = { iph->saddr, 0, };
1302 __be32 daddr = iph->daddr;
1303 unsigned short est_mtu = 0;
1304
1305 if (ipv4_config.no_pmtu_disc)
1306 return 0;
1307
1308 for (i = 0; i < 2; i++) {
1309 unsigned hash = rt_hash(daddr, skeys[i], 0);
1310
1311 rcu_read_lock();
1312 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1313 rth = rcu_dereference(rth->u.dst.rt_next)) {
1314 if (rth->fl.fl4_dst == daddr &&
1315 rth->fl.fl4_src == skeys[i] &&
1316 rth->rt_dst == daddr &&
1317 rth->rt_src == iph->saddr &&
1318 rth->fl.iif == 0 &&
1319 !(dst_metric_locked(&rth->u.dst, RTAX_MTU))) {
1320 unsigned short mtu = new_mtu;
1321
1322 if (new_mtu < 68 || new_mtu >= old_mtu) {
1323
1324 /* BSD 4.2 compatibility hack :-( */
1325 if (mtu == 0 &&
1326 old_mtu >= rth->u.dst.metrics[RTAX_MTU-1] &&
1327 old_mtu >= 68 + (iph->ihl << 2))
1328 old_mtu -= iph->ihl << 2;
1329
1330 mtu = guess_mtu(old_mtu);
1331 }
1332 if (mtu <= rth->u.dst.metrics[RTAX_MTU-1]) {
1333 if (mtu < rth->u.dst.metrics[RTAX_MTU-1]) {
1334 dst_confirm(&rth->u.dst);
1335 if (mtu < ip_rt_min_pmtu) {
1336 mtu = ip_rt_min_pmtu;
1337 rth->u.dst.metrics[RTAX_LOCK-1] |=
1338 (1 << RTAX_MTU);
1339 }
1340 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1341 dst_set_expires(&rth->u.dst,
1342 ip_rt_mtu_expires);
1343 }
1344 est_mtu = mtu;
1345 }
1346 }
1347 }
1348 rcu_read_unlock();
1349 }
1350 return est_mtu ? : new_mtu;
1351 }
1352
1353 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1354 {
1355 if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= 68 &&
1356 !(dst_metric_locked(dst, RTAX_MTU))) {
1357 if (mtu < ip_rt_min_pmtu) {
1358 mtu = ip_rt_min_pmtu;
1359 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1360 }
1361 dst->metrics[RTAX_MTU-1] = mtu;
1362 dst_set_expires(dst, ip_rt_mtu_expires);
1363 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1364 }
1365 }
1366
1367 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1368 {
1369 return NULL;
1370 }
1371
1372 static void ipv4_dst_destroy(struct dst_entry *dst)
1373 {
1374 struct rtable *rt = (struct rtable *) dst;
1375 struct inet_peer *peer = rt->peer;
1376 struct in_device *idev = rt->idev;
1377
1378 if (peer) {
1379 rt->peer = NULL;
1380 inet_putpeer(peer);
1381 }
1382
1383 if (idev) {
1384 rt->idev = NULL;
1385 in_dev_put(idev);
1386 }
1387 }
1388
1389 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1390 int how)
1391 {
1392 struct rtable *rt = (struct rtable *) dst;
1393 struct in_device *idev = rt->idev;
1394 if (dev != init_net.loopback_dev && idev && idev->dev == dev) {
1395 struct in_device *loopback_idev = in_dev_get(init_net.loopback_dev);
1396 if (loopback_idev) {
1397 rt->idev = loopback_idev;
1398 in_dev_put(idev);
1399 }
1400 }
1401 }
1402
1403 static void ipv4_link_failure(struct sk_buff *skb)
1404 {
1405 struct rtable *rt;
1406
1407 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1408
1409 rt = (struct rtable *) skb->dst;
1410 if (rt)
1411 dst_set_expires(&rt->u.dst, 0);
1412 }
1413
1414 static int ip_rt_bug(struct sk_buff *skb)
1415 {
1416 printk(KERN_DEBUG "ip_rt_bug: %u.%u.%u.%u -> %u.%u.%u.%u, %s\n",
1417 NIPQUAD(ip_hdr(skb)->saddr), NIPQUAD(ip_hdr(skb)->daddr),
1418 skb->dev ? skb->dev->name : "?");
1419 kfree_skb(skb);
1420 return 0;
1421 }
1422
1423 /*
1424 We do not cache source address of outgoing interface,
1425 because it is used only by IP RR, TS and SRR options,
1426 so that it out of fast path.
1427
1428 BTW remember: "addr" is allowed to be not aligned
1429 in IP options!
1430 */
1431
1432 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1433 {
1434 __be32 src;
1435 struct fib_result res;
1436
1437 if (rt->fl.iif == 0)
1438 src = rt->rt_src;
1439 else if (fib_lookup(&rt->fl, &res) == 0) {
1440 src = FIB_RES_PREFSRC(res);
1441 fib_res_put(&res);
1442 } else
1443 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1444 RT_SCOPE_UNIVERSE);
1445 memcpy(addr, &src, 4);
1446 }
1447
1448 #ifdef CONFIG_NET_CLS_ROUTE
1449 static void set_class_tag(struct rtable *rt, u32 tag)
1450 {
1451 if (!(rt->u.dst.tclassid & 0xFFFF))
1452 rt->u.dst.tclassid |= tag & 0xFFFF;
1453 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1454 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1455 }
1456 #endif
1457
1458 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1459 {
1460 struct fib_info *fi = res->fi;
1461
1462 if (fi) {
1463 if (FIB_RES_GW(*res) &&
1464 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1465 rt->rt_gateway = FIB_RES_GW(*res);
1466 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1467 sizeof(rt->u.dst.metrics));
1468 if (fi->fib_mtu == 0) {
1469 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1470 if (rt->u.dst.metrics[RTAX_LOCK-1] & (1 << RTAX_MTU) &&
1471 rt->rt_gateway != rt->rt_dst &&
1472 rt->u.dst.dev->mtu > 576)
1473 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1474 }
1475 #ifdef CONFIG_NET_CLS_ROUTE
1476 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1477 #endif
1478 } else
1479 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1480
1481 if (rt->u.dst.metrics[RTAX_HOPLIMIT-1] == 0)
1482 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1483 if (rt->u.dst.metrics[RTAX_MTU-1] > IP_MAX_MTU)
1484 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1485 if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0)
1486 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1487 ip_rt_min_advmss);
1488 if (rt->u.dst.metrics[RTAX_ADVMSS-1] > 65535 - 40)
1489 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1490
1491 #ifdef CONFIG_NET_CLS_ROUTE
1492 #ifdef CONFIG_IP_MULTIPLE_TABLES
1493 set_class_tag(rt, fib_rules_tclass(res));
1494 #endif
1495 set_class_tag(rt, itag);
1496 #endif
1497 rt->rt_type = res->type;
1498 }
1499
1500 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1501 u8 tos, struct net_device *dev, int our)
1502 {
1503 unsigned hash;
1504 struct rtable *rth;
1505 __be32 spec_dst;
1506 struct in_device *in_dev = in_dev_get(dev);
1507 u32 itag = 0;
1508
1509 /* Primary sanity checks. */
1510
1511 if (in_dev == NULL)
1512 return -EINVAL;
1513
1514 if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr) ||
1515 skb->protocol != htons(ETH_P_IP))
1516 goto e_inval;
1517
1518 if (ZERONET(saddr)) {
1519 if (!LOCAL_MCAST(daddr))
1520 goto e_inval;
1521 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1522 } else if (fib_validate_source(saddr, 0, tos, 0,
1523 dev, &spec_dst, &itag) < 0)
1524 goto e_inval;
1525
1526 rth = dst_alloc(&ipv4_dst_ops);
1527 if (!rth)
1528 goto e_nobufs;
1529
1530 rth->u.dst.output= ip_rt_bug;
1531
1532 atomic_set(&rth->u.dst.__refcnt, 1);
1533 rth->u.dst.flags= DST_HOST;
1534 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1535 rth->u.dst.flags |= DST_NOPOLICY;
1536 rth->fl.fl4_dst = daddr;
1537 rth->rt_dst = daddr;
1538 rth->fl.fl4_tos = tos;
1539 rth->fl.mark = skb->mark;
1540 rth->fl.fl4_src = saddr;
1541 rth->rt_src = saddr;
1542 #ifdef CONFIG_NET_CLS_ROUTE
1543 rth->u.dst.tclassid = itag;
1544 #endif
1545 rth->rt_iif =
1546 rth->fl.iif = dev->ifindex;
1547 rth->u.dst.dev = init_net.loopback_dev;
1548 dev_hold(rth->u.dst.dev);
1549 rth->idev = in_dev_get(rth->u.dst.dev);
1550 rth->fl.oif = 0;
1551 rth->rt_gateway = daddr;
1552 rth->rt_spec_dst= spec_dst;
1553 rth->rt_type = RTN_MULTICAST;
1554 rth->rt_flags = RTCF_MULTICAST;
1555 if (our) {
1556 rth->u.dst.input= ip_local_deliver;
1557 rth->rt_flags |= RTCF_LOCAL;
1558 }
1559
1560 #ifdef CONFIG_IP_MROUTE
1561 if (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
1562 rth->u.dst.input = ip_mr_input;
1563 #endif
1564 RT_CACHE_STAT_INC(in_slow_mc);
1565
1566 in_dev_put(in_dev);
1567 hash = rt_hash(daddr, saddr, dev->ifindex);
1568 return rt_intern_hash(hash, rth, (struct rtable**) &skb->dst);
1569
1570 e_nobufs:
1571 in_dev_put(in_dev);
1572 return -ENOBUFS;
1573
1574 e_inval:
1575 in_dev_put(in_dev);
1576 return -EINVAL;
1577 }
1578
1579
1580 static void ip_handle_martian_source(struct net_device *dev,
1581 struct in_device *in_dev,
1582 struct sk_buff *skb,
1583 __be32 daddr,
1584 __be32 saddr)
1585 {
1586 RT_CACHE_STAT_INC(in_martian_src);
1587 #ifdef CONFIG_IP_ROUTE_VERBOSE
1588 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1589 /*
1590 * RFC1812 recommendation, if source is martian,
1591 * the only hint is MAC header.
1592 */
1593 printk(KERN_WARNING "martian source %u.%u.%u.%u from "
1594 "%u.%u.%u.%u, on dev %s\n",
1595 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
1596 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1597 int i;
1598 const unsigned char *p = skb_mac_header(skb);
1599 printk(KERN_WARNING "ll header: ");
1600 for (i = 0; i < dev->hard_header_len; i++, p++) {
1601 printk("%02x", *p);
1602 if (i < (dev->hard_header_len - 1))
1603 printk(":");
1604 }
1605 printk("\n");
1606 }
1607 }
1608 #endif
1609 }
1610
1611 static inline int __mkroute_input(struct sk_buff *skb,
1612 struct fib_result* res,
1613 struct in_device *in_dev,
1614 __be32 daddr, __be32 saddr, u32 tos,
1615 struct rtable **result)
1616 {
1617
1618 struct rtable *rth;
1619 int err;
1620 struct in_device *out_dev;
1621 unsigned flags = 0;
1622 __be32 spec_dst;
1623 u32 itag;
1624
1625 /* get a working reference to the output device */
1626 out_dev = in_dev_get(FIB_RES_DEV(*res));
1627 if (out_dev == NULL) {
1628 if (net_ratelimit())
1629 printk(KERN_CRIT "Bug in ip_route_input" \
1630 "_slow(). Please, report\n");
1631 return -EINVAL;
1632 }
1633
1634
1635 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1636 in_dev->dev, &spec_dst, &itag);
1637 if (err < 0) {
1638 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1639 saddr);
1640
1641 err = -EINVAL;
1642 goto cleanup;
1643 }
1644
1645 if (err)
1646 flags |= RTCF_DIRECTSRC;
1647
1648 if (out_dev == in_dev && err && !(flags & (RTCF_NAT | RTCF_MASQ)) &&
1649 (IN_DEV_SHARED_MEDIA(out_dev) ||
1650 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1651 flags |= RTCF_DOREDIRECT;
1652
1653 if (skb->protocol != htons(ETH_P_IP)) {
1654 /* Not IP (i.e. ARP). Do not create route, if it is
1655 * invalid for proxy arp. DNAT routes are always valid.
1656 */
1657 if (out_dev == in_dev && !(flags & RTCF_DNAT)) {
1658 err = -EINVAL;
1659 goto cleanup;
1660 }
1661 }
1662
1663
1664 rth = dst_alloc(&ipv4_dst_ops);
1665 if (!rth) {
1666 err = -ENOBUFS;
1667 goto cleanup;
1668 }
1669
1670 atomic_set(&rth->u.dst.__refcnt, 1);
1671 rth->u.dst.flags= DST_HOST;
1672 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1673 rth->u.dst.flags |= DST_NOPOLICY;
1674 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
1675 rth->u.dst.flags |= DST_NOXFRM;
1676 rth->fl.fl4_dst = daddr;
1677 rth->rt_dst = daddr;
1678 rth->fl.fl4_tos = tos;
1679 rth->fl.mark = skb->mark;
1680 rth->fl.fl4_src = saddr;
1681 rth->rt_src = saddr;
1682 rth->rt_gateway = daddr;
1683 rth->rt_iif =
1684 rth->fl.iif = in_dev->dev->ifindex;
1685 rth->u.dst.dev = (out_dev)->dev;
1686 dev_hold(rth->u.dst.dev);
1687 rth->idev = in_dev_get(rth->u.dst.dev);
1688 rth->fl.oif = 0;
1689 rth->rt_spec_dst= spec_dst;
1690
1691 rth->u.dst.input = ip_forward;
1692 rth->u.dst.output = ip_output;
1693
1694 rt_set_nexthop(rth, res, itag);
1695
1696 rth->rt_flags = flags;
1697
1698 *result = rth;
1699 err = 0;
1700 cleanup:
1701 /* release the working reference to the output device */
1702 in_dev_put(out_dev);
1703 return err;
1704 }
1705
1706 static inline int ip_mkroute_input(struct sk_buff *skb,
1707 struct fib_result* res,
1708 const struct flowi *fl,
1709 struct in_device *in_dev,
1710 __be32 daddr, __be32 saddr, u32 tos)
1711 {
1712 struct rtable* rth = NULL;
1713 int err;
1714 unsigned hash;
1715
1716 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1717 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
1718 fib_select_multipath(fl, res);
1719 #endif
1720
1721 /* create a routing cache entry */
1722 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1723 if (err)
1724 return err;
1725
1726 /* put it into the cache */
1727 hash = rt_hash(daddr, saddr, fl->iif);
1728 return rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
1729 }
1730
1731 /*
1732 * NOTE. We drop all the packets that has local source
1733 * addresses, because every properly looped back packet
1734 * must have correct destination already attached by output routine.
1735 *
1736 * Such approach solves two big problems:
1737 * 1. Not simplex devices are handled properly.
1738 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1739 */
1740
1741 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1742 u8 tos, struct net_device *dev)
1743 {
1744 struct fib_result res;
1745 struct in_device *in_dev = in_dev_get(dev);
1746 struct flowi fl = { .nl_u = { .ip4_u =
1747 { .daddr = daddr,
1748 .saddr = saddr,
1749 .tos = tos,
1750 .scope = RT_SCOPE_UNIVERSE,
1751 } },
1752 .mark = skb->mark,
1753 .iif = dev->ifindex };
1754 unsigned flags = 0;
1755 u32 itag = 0;
1756 struct rtable * rth;
1757 unsigned hash;
1758 __be32 spec_dst;
1759 int err = -EINVAL;
1760 int free_res = 0;
1761
1762 /* IP on this device is disabled. */
1763
1764 if (!in_dev)
1765 goto out;
1766
1767 /* Check for the most weird martians, which can be not detected
1768 by fib_lookup.
1769 */
1770
1771 if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr))
1772 goto martian_source;
1773
1774 if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
1775 goto brd_input;
1776
1777 /* Accept zero addresses only to limited broadcast;
1778 * I even do not know to fix it or not. Waiting for complains :-)
1779 */
1780 if (ZERONET(saddr))
1781 goto martian_source;
1782
1783 if (BADCLASS(daddr) || ZERONET(daddr) || LOOPBACK(daddr))
1784 goto martian_destination;
1785
1786 /*
1787 * Now we are ready to route packet.
1788 */
1789 if ((err = fib_lookup(&fl, &res)) != 0) {
1790 if (!IN_DEV_FORWARD(in_dev))
1791 goto e_hostunreach;
1792 goto no_route;
1793 }
1794 free_res = 1;
1795
1796 RT_CACHE_STAT_INC(in_slow_tot);
1797
1798 if (res.type == RTN_BROADCAST)
1799 goto brd_input;
1800
1801 if (res.type == RTN_LOCAL) {
1802 int result;
1803 result = fib_validate_source(saddr, daddr, tos,
1804 init_net.loopback_dev->ifindex,
1805 dev, &spec_dst, &itag);
1806 if (result < 0)
1807 goto martian_source;
1808 if (result)
1809 flags |= RTCF_DIRECTSRC;
1810 spec_dst = daddr;
1811 goto local_input;
1812 }
1813
1814 if (!IN_DEV_FORWARD(in_dev))
1815 goto e_hostunreach;
1816 if (res.type != RTN_UNICAST)
1817 goto martian_destination;
1818
1819 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
1820 done:
1821 in_dev_put(in_dev);
1822 if (free_res)
1823 fib_res_put(&res);
1824 out: return err;
1825
1826 brd_input:
1827 if (skb->protocol != htons(ETH_P_IP))
1828 goto e_inval;
1829
1830 if (ZERONET(saddr))
1831 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1832 else {
1833 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1834 &itag);
1835 if (err < 0)
1836 goto martian_source;
1837 if (err)
1838 flags |= RTCF_DIRECTSRC;
1839 }
1840 flags |= RTCF_BROADCAST;
1841 res.type = RTN_BROADCAST;
1842 RT_CACHE_STAT_INC(in_brd);
1843
1844 local_input:
1845 rth = dst_alloc(&ipv4_dst_ops);
1846 if (!rth)
1847 goto e_nobufs;
1848
1849 rth->u.dst.output= ip_rt_bug;
1850
1851 atomic_set(&rth->u.dst.__refcnt, 1);
1852 rth->u.dst.flags= DST_HOST;
1853 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1854 rth->u.dst.flags |= DST_NOPOLICY;
1855 rth->fl.fl4_dst = daddr;
1856 rth->rt_dst = daddr;
1857 rth->fl.fl4_tos = tos;
1858 rth->fl.mark = skb->mark;
1859 rth->fl.fl4_src = saddr;
1860 rth->rt_src = saddr;
1861 #ifdef CONFIG_NET_CLS_ROUTE
1862 rth->u.dst.tclassid = itag;
1863 #endif
1864 rth->rt_iif =
1865 rth->fl.iif = dev->ifindex;
1866 rth->u.dst.dev = init_net.loopback_dev;
1867 dev_hold(rth->u.dst.dev);
1868 rth->idev = in_dev_get(rth->u.dst.dev);
1869 rth->rt_gateway = daddr;
1870 rth->rt_spec_dst= spec_dst;
1871 rth->u.dst.input= ip_local_deliver;
1872 rth->rt_flags = flags|RTCF_LOCAL;
1873 if (res.type == RTN_UNREACHABLE) {
1874 rth->u.dst.input= ip_error;
1875 rth->u.dst.error= -err;
1876 rth->rt_flags &= ~RTCF_LOCAL;
1877 }
1878 rth->rt_type = res.type;
1879 hash = rt_hash(daddr, saddr, fl.iif);
1880 err = rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
1881 goto done;
1882
1883 no_route:
1884 RT_CACHE_STAT_INC(in_no_route);
1885 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1886 res.type = RTN_UNREACHABLE;
1887 if (err == -ESRCH)
1888 err = -ENETUNREACH;
1889 goto local_input;
1890
1891 /*
1892 * Do not cache martian addresses: they should be logged (RFC1812)
1893 */
1894 martian_destination:
1895 RT_CACHE_STAT_INC(in_martian_dst);
1896 #ifdef CONFIG_IP_ROUTE_VERBOSE
1897 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1898 printk(KERN_WARNING "martian destination %u.%u.%u.%u from "
1899 "%u.%u.%u.%u, dev %s\n",
1900 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
1901 #endif
1902
1903 e_hostunreach:
1904 err = -EHOSTUNREACH;
1905 goto done;
1906
1907 e_inval:
1908 err = -EINVAL;
1909 goto done;
1910
1911 e_nobufs:
1912 err = -ENOBUFS;
1913 goto done;
1914
1915 martian_source:
1916 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1917 goto e_inval;
1918 }
1919
1920 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1921 u8 tos, struct net_device *dev)
1922 {
1923 struct rtable * rth;
1924 unsigned hash;
1925 int iif = dev->ifindex;
1926
1927 tos &= IPTOS_RT_MASK;
1928 hash = rt_hash(daddr, saddr, iif);
1929
1930 rcu_read_lock();
1931 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1932 rth = rcu_dereference(rth->u.dst.rt_next)) {
1933 if (rth->fl.fl4_dst == daddr &&
1934 rth->fl.fl4_src == saddr &&
1935 rth->fl.iif == iif &&
1936 rth->fl.oif == 0 &&
1937 rth->fl.mark == skb->mark &&
1938 rth->fl.fl4_tos == tos) {
1939 dst_use(&rth->u.dst, jiffies);
1940 RT_CACHE_STAT_INC(in_hit);
1941 rcu_read_unlock();
1942 skb->dst = (struct dst_entry*)rth;
1943 return 0;
1944 }
1945 RT_CACHE_STAT_INC(in_hlist_search);
1946 }
1947 rcu_read_unlock();
1948
1949 /* Multicast recognition logic is moved from route cache to here.
1950 The problem was that too many Ethernet cards have broken/missing
1951 hardware multicast filters :-( As result the host on multicasting
1952 network acquires a lot of useless route cache entries, sort of
1953 SDR messages from all the world. Now we try to get rid of them.
1954 Really, provided software IP multicast filter is organized
1955 reasonably (at least, hashed), it does not result in a slowdown
1956 comparing with route cache reject entries.
1957 Note, that multicast routers are not affected, because
1958 route cache entry is created eventually.
1959 */
1960 if (MULTICAST(daddr)) {
1961 struct in_device *in_dev;
1962
1963 rcu_read_lock();
1964 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
1965 int our = ip_check_mc(in_dev, daddr, saddr,
1966 ip_hdr(skb)->protocol);
1967 if (our
1968 #ifdef CONFIG_IP_MROUTE
1969 || (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
1970 #endif
1971 ) {
1972 rcu_read_unlock();
1973 return ip_route_input_mc(skb, daddr, saddr,
1974 tos, dev, our);
1975 }
1976 }
1977 rcu_read_unlock();
1978 return -EINVAL;
1979 }
1980 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
1981 }
1982
1983 static inline int __mkroute_output(struct rtable **result,
1984 struct fib_result* res,
1985 const struct flowi *fl,
1986 const struct flowi *oldflp,
1987 struct net_device *dev_out,
1988 unsigned flags)
1989 {
1990 struct rtable *rth;
1991 struct in_device *in_dev;
1992 u32 tos = RT_FL_TOS(oldflp);
1993 int err = 0;
1994
1995 if (LOOPBACK(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
1996 return -EINVAL;
1997
1998 if (fl->fl4_dst == htonl(0xFFFFFFFF))
1999 res->type = RTN_BROADCAST;
2000 else if (MULTICAST(fl->fl4_dst))
2001 res->type = RTN_MULTICAST;
2002 else if (BADCLASS(fl->fl4_dst) || ZERONET(fl->fl4_dst))
2003 return -EINVAL;
2004
2005 if (dev_out->flags & IFF_LOOPBACK)
2006 flags |= RTCF_LOCAL;
2007
2008 /* get work reference to inet device */
2009 in_dev = in_dev_get(dev_out);
2010 if (!in_dev)
2011 return -EINVAL;
2012
2013 if (res->type == RTN_BROADCAST) {
2014 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2015 if (res->fi) {
2016 fib_info_put(res->fi);
2017 res->fi = NULL;
2018 }
2019 } else if (res->type == RTN_MULTICAST) {
2020 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2021 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2022 oldflp->proto))
2023 flags &= ~RTCF_LOCAL;
2024 /* If multicast route do not exist use
2025 default one, but do not gateway in this case.
2026 Yes, it is hack.
2027 */
2028 if (res->fi && res->prefixlen < 4) {
2029 fib_info_put(res->fi);
2030 res->fi = NULL;
2031 }
2032 }
2033
2034
2035 rth = dst_alloc(&ipv4_dst_ops);
2036 if (!rth) {
2037 err = -ENOBUFS;
2038 goto cleanup;
2039 }
2040
2041 atomic_set(&rth->u.dst.__refcnt, 1);
2042 rth->u.dst.flags= DST_HOST;
2043 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2044 rth->u.dst.flags |= DST_NOXFRM;
2045 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2046 rth->u.dst.flags |= DST_NOPOLICY;
2047
2048 rth->fl.fl4_dst = oldflp->fl4_dst;
2049 rth->fl.fl4_tos = tos;
2050 rth->fl.fl4_src = oldflp->fl4_src;
2051 rth->fl.oif = oldflp->oif;
2052 rth->fl.mark = oldflp->mark;
2053 rth->rt_dst = fl->fl4_dst;
2054 rth->rt_src = fl->fl4_src;
2055 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2056 /* get references to the devices that are to be hold by the routing
2057 cache entry */
2058 rth->u.dst.dev = dev_out;
2059 dev_hold(dev_out);
2060 rth->idev = in_dev_get(dev_out);
2061 rth->rt_gateway = fl->fl4_dst;
2062 rth->rt_spec_dst= fl->fl4_src;
2063
2064 rth->u.dst.output=ip_output;
2065
2066 RT_CACHE_STAT_INC(out_slow_tot);
2067
2068 if (flags & RTCF_LOCAL) {
2069 rth->u.dst.input = ip_local_deliver;
2070 rth->rt_spec_dst = fl->fl4_dst;
2071 }
2072 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2073 rth->rt_spec_dst = fl->fl4_src;
2074 if (flags & RTCF_LOCAL &&
2075 !(dev_out->flags & IFF_LOOPBACK)) {
2076 rth->u.dst.output = ip_mc_output;
2077 RT_CACHE_STAT_INC(out_slow_mc);
2078 }
2079 #ifdef CONFIG_IP_MROUTE
2080 if (res->type == RTN_MULTICAST) {
2081 if (IN_DEV_MFORWARD(in_dev) &&
2082 !LOCAL_MCAST(oldflp->fl4_dst)) {
2083 rth->u.dst.input = ip_mr_input;
2084 rth->u.dst.output = ip_mc_output;
2085 }
2086 }
2087 #endif
2088 }
2089
2090 rt_set_nexthop(rth, res, 0);
2091
2092 rth->rt_flags = flags;
2093
2094 *result = rth;
2095 cleanup:
2096 /* release work reference to inet device */
2097 in_dev_put(in_dev);
2098
2099 return err;
2100 }
2101
2102 static inline int ip_mkroute_output(struct rtable **rp,
2103 struct fib_result* res,
2104 const struct flowi *fl,
2105 const struct flowi *oldflp,
2106 struct net_device *dev_out,
2107 unsigned flags)
2108 {
2109 struct rtable *rth = NULL;
2110 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2111 unsigned hash;
2112 if (err == 0) {
2113 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif);
2114 err = rt_intern_hash(hash, rth, rp);
2115 }
2116
2117 return err;
2118 }
2119
2120 /*
2121 * Major route resolver routine.
2122 */
2123
2124 static int ip_route_output_slow(struct rtable **rp, const struct flowi *oldflp)
2125 {
2126 u32 tos = RT_FL_TOS(oldflp);
2127 struct flowi fl = { .nl_u = { .ip4_u =
2128 { .daddr = oldflp->fl4_dst,
2129 .saddr = oldflp->fl4_src,
2130 .tos = tos & IPTOS_RT_MASK,
2131 .scope = ((tos & RTO_ONLINK) ?
2132 RT_SCOPE_LINK :
2133 RT_SCOPE_UNIVERSE),
2134 } },
2135 .mark = oldflp->mark,
2136 .iif = init_net.loopback_dev->ifindex,
2137 .oif = oldflp->oif };
2138 struct fib_result res;
2139 unsigned flags = 0;
2140 struct net_device *dev_out = NULL;
2141 int free_res = 0;
2142 int err;
2143
2144
2145 res.fi = NULL;
2146 #ifdef CONFIG_IP_MULTIPLE_TABLES
2147 res.r = NULL;
2148 #endif
2149
2150 if (oldflp->fl4_src) {
2151 err = -EINVAL;
2152 if (MULTICAST(oldflp->fl4_src) ||
2153 BADCLASS(oldflp->fl4_src) ||
2154 ZERONET(oldflp->fl4_src))
2155 goto out;
2156
2157 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2158 dev_out = ip_dev_find(oldflp->fl4_src);
2159 if (dev_out == NULL)
2160 goto out;
2161
2162 /* I removed check for oif == dev_out->oif here.
2163 It was wrong for two reasons:
2164 1. ip_dev_find(saddr) can return wrong iface, if saddr is
2165 assigned to multiple interfaces.
2166 2. Moreover, we are allowed to send packets with saddr
2167 of another iface. --ANK
2168 */
2169
2170 if (oldflp->oif == 0
2171 && (MULTICAST(oldflp->fl4_dst) || oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2172 /* Special hack: user can direct multicasts
2173 and limited broadcast via necessary interface
2174 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2175 This hack is not just for fun, it allows
2176 vic,vat and friends to work.
2177 They bind socket to loopback, set ttl to zero
2178 and expect that it will work.
2179 From the viewpoint of routing cache they are broken,
2180 because we are not allowed to build multicast path
2181 with loopback source addr (look, routing cache
2182 cannot know, that ttl is zero, so that packet
2183 will not leave this host and route is valid).
2184 Luckily, this hack is good workaround.
2185 */
2186
2187 fl.oif = dev_out->ifindex;
2188 goto make_route;
2189 }
2190 if (dev_out)
2191 dev_put(dev_out);
2192 dev_out = NULL;
2193 }
2194
2195
2196 if (oldflp->oif) {
2197 dev_out = dev_get_by_index(&init_net, oldflp->oif);
2198 err = -ENODEV;
2199 if (dev_out == NULL)
2200 goto out;
2201
2202 /* RACE: Check return value of inet_select_addr instead. */
2203 if (__in_dev_get_rtnl(dev_out) == NULL) {
2204 dev_put(dev_out);
2205 goto out; /* Wrong error code */
2206 }
2207
2208 if (LOCAL_MCAST(oldflp->fl4_dst) || oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2209 if (!fl.fl4_src)
2210 fl.fl4_src = inet_select_addr(dev_out, 0,
2211 RT_SCOPE_LINK);
2212 goto make_route;
2213 }
2214 if (!fl.fl4_src) {
2215 if (MULTICAST(oldflp->fl4_dst))
2216 fl.fl4_src = inet_select_addr(dev_out, 0,
2217 fl.fl4_scope);
2218 else if (!oldflp->fl4_dst)
2219 fl.fl4_src = inet_select_addr(dev_out, 0,
2220 RT_SCOPE_HOST);
2221 }
2222 }
2223
2224 if (!fl.fl4_dst) {
2225 fl.fl4_dst = fl.fl4_src;
2226 if (!fl.fl4_dst)
2227 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2228 if (dev_out)
2229 dev_put(dev_out);
2230 dev_out = init_net.loopback_dev;
2231 dev_hold(dev_out);
2232 fl.oif = init_net.loopback_dev->ifindex;
2233 res.type = RTN_LOCAL;
2234 flags |= RTCF_LOCAL;
2235 goto make_route;
2236 }
2237
2238 if (fib_lookup(&fl, &res)) {
2239 res.fi = NULL;
2240 if (oldflp->oif) {
2241 /* Apparently, routing tables are wrong. Assume,
2242 that the destination is on link.
2243
2244 WHY? DW.
2245 Because we are allowed to send to iface
2246 even if it has NO routes and NO assigned
2247 addresses. When oif is specified, routing
2248 tables are looked up with only one purpose:
2249 to catch if destination is gatewayed, rather than
2250 direct. Moreover, if MSG_DONTROUTE is set,
2251 we send packet, ignoring both routing tables
2252 and ifaddr state. --ANK
2253
2254
2255 We could make it even if oif is unknown,
2256 likely IPv6, but we do not.
2257 */
2258
2259 if (fl.fl4_src == 0)
2260 fl.fl4_src = inet_select_addr(dev_out, 0,
2261 RT_SCOPE_LINK);
2262 res.type = RTN_UNICAST;
2263 goto make_route;
2264 }
2265 if (dev_out)
2266 dev_put(dev_out);
2267 err = -ENETUNREACH;
2268 goto out;
2269 }
2270 free_res = 1;
2271
2272 if (res.type == RTN_LOCAL) {
2273 if (!fl.fl4_src)
2274 fl.fl4_src = fl.fl4_dst;
2275 if (dev_out)
2276 dev_put(dev_out);
2277 dev_out = init_net.loopback_dev;
2278 dev_hold(dev_out);
2279 fl.oif = dev_out->ifindex;
2280 if (res.fi)
2281 fib_info_put(res.fi);
2282 res.fi = NULL;
2283 flags |= RTCF_LOCAL;
2284 goto make_route;
2285 }
2286
2287 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2288 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2289 fib_select_multipath(&fl, &res);
2290 else
2291 #endif
2292 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2293 fib_select_default(&fl, &res);
2294
2295 if (!fl.fl4_src)
2296 fl.fl4_src = FIB_RES_PREFSRC(res);
2297
2298 if (dev_out)
2299 dev_put(dev_out);
2300 dev_out = FIB_RES_DEV(res);
2301 dev_hold(dev_out);
2302 fl.oif = dev_out->ifindex;
2303
2304
2305 make_route:
2306 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2307
2308
2309 if (free_res)
2310 fib_res_put(&res);
2311 if (dev_out)
2312 dev_put(dev_out);
2313 out: return err;
2314 }
2315
2316 int __ip_route_output_key(struct rtable **rp, const struct flowi *flp)
2317 {
2318 unsigned hash;
2319 struct rtable *rth;
2320
2321 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif);
2322
2323 rcu_read_lock_bh();
2324 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2325 rth = rcu_dereference(rth->u.dst.rt_next)) {
2326 if (rth->fl.fl4_dst == flp->fl4_dst &&
2327 rth->fl.fl4_src == flp->fl4_src &&
2328 rth->fl.iif == 0 &&
2329 rth->fl.oif == flp->oif &&
2330 rth->fl.mark == flp->mark &&
2331 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2332 (IPTOS_RT_MASK | RTO_ONLINK))) {
2333 dst_use(&rth->u.dst, jiffies);
2334 RT_CACHE_STAT_INC(out_hit);
2335 rcu_read_unlock_bh();
2336 *rp = rth;
2337 return 0;
2338 }
2339 RT_CACHE_STAT_INC(out_hlist_search);
2340 }
2341 rcu_read_unlock_bh();
2342
2343 return ip_route_output_slow(rp, flp);
2344 }
2345
2346 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2347
2348 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2349 {
2350 }
2351
2352 static struct dst_ops ipv4_dst_blackhole_ops = {
2353 .family = AF_INET,
2354 .protocol = __constant_htons(ETH_P_IP),
2355 .destroy = ipv4_dst_destroy,
2356 .check = ipv4_dst_check,
2357 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2358 .entry_size = sizeof(struct rtable),
2359 };
2360
2361
2362 static int ipv4_dst_blackhole(struct rtable **rp, struct flowi *flp, struct sock *sk)
2363 {
2364 struct rtable *ort = *rp;
2365 struct rtable *rt = (struct rtable *)
2366 dst_alloc(&ipv4_dst_blackhole_ops);
2367
2368 if (rt) {
2369 struct dst_entry *new = &rt->u.dst;
2370
2371 atomic_set(&new->__refcnt, 1);
2372 new->__use = 1;
2373 new->input = dst_discard;
2374 new->output = dst_discard;
2375 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2376
2377 new->dev = ort->u.dst.dev;
2378 if (new->dev)
2379 dev_hold(new->dev);
2380
2381 rt->fl = ort->fl;
2382
2383 rt->idev = ort->idev;
2384 if (rt->idev)
2385 in_dev_hold(rt->idev);
2386 rt->rt_flags = ort->rt_flags;
2387 rt->rt_type = ort->rt_type;
2388 rt->rt_dst = ort->rt_dst;
2389 rt->rt_src = ort->rt_src;
2390 rt->rt_iif = ort->rt_iif;
2391 rt->rt_gateway = ort->rt_gateway;
2392 rt->rt_spec_dst = ort->rt_spec_dst;
2393 rt->peer = ort->peer;
2394 if (rt->peer)
2395 atomic_inc(&rt->peer->refcnt);
2396
2397 dst_free(new);
2398 }
2399
2400 dst_release(&(*rp)->u.dst);
2401 *rp = rt;
2402 return (rt ? 0 : -ENOMEM);
2403 }
2404
2405 int ip_route_output_flow(struct rtable **rp, struct flowi *flp, struct sock *sk, int flags)
2406 {
2407 int err;
2408
2409 if ((err = __ip_route_output_key(rp, flp)) != 0)
2410 return err;
2411
2412 if (flp->proto) {
2413 if (!flp->fl4_src)
2414 flp->fl4_src = (*rp)->rt_src;
2415 if (!flp->fl4_dst)
2416 flp->fl4_dst = (*rp)->rt_dst;
2417 err = __xfrm_lookup((struct dst_entry **)rp, flp, sk, flags);
2418 if (err == -EREMOTE)
2419 err = ipv4_dst_blackhole(rp, flp, sk);
2420
2421 return err;
2422 }
2423
2424 return 0;
2425 }
2426
2427 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2428
2429 int ip_route_output_key(struct rtable **rp, struct flowi *flp)
2430 {
2431 return ip_route_output_flow(rp, flp, NULL, 0);
2432 }
2433
2434 static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
2435 int nowait, unsigned int flags)
2436 {
2437 struct rtable *rt = (struct rtable*)skb->dst;
2438 struct rtmsg *r;
2439 struct nlmsghdr *nlh;
2440 long expires;
2441 u32 id = 0, ts = 0, tsage = 0, error;
2442
2443 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2444 if (nlh == NULL)
2445 return -EMSGSIZE;
2446
2447 r = nlmsg_data(nlh);
2448 r->rtm_family = AF_INET;
2449 r->rtm_dst_len = 32;
2450 r->rtm_src_len = 0;
2451 r->rtm_tos = rt->fl.fl4_tos;
2452 r->rtm_table = RT_TABLE_MAIN;
2453 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2454 r->rtm_type = rt->rt_type;
2455 r->rtm_scope = RT_SCOPE_UNIVERSE;
2456 r->rtm_protocol = RTPROT_UNSPEC;
2457 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2458 if (rt->rt_flags & RTCF_NOTIFY)
2459 r->rtm_flags |= RTM_F_NOTIFY;
2460
2461 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2462
2463 if (rt->fl.fl4_src) {
2464 r->rtm_src_len = 32;
2465 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2466 }
2467 if (rt->u.dst.dev)
2468 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2469 #ifdef CONFIG_NET_CLS_ROUTE
2470 if (rt->u.dst.tclassid)
2471 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2472 #endif
2473 if (rt->fl.iif)
2474 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2475 else if (rt->rt_src != rt->fl.fl4_src)
2476 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2477
2478 if (rt->rt_dst != rt->rt_gateway)
2479 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2480
2481 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2482 goto nla_put_failure;
2483
2484 error = rt->u.dst.error;
2485 expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2486 if (rt->peer) {
2487 id = rt->peer->ip_id_count;
2488 if (rt->peer->tcp_ts_stamp) {
2489 ts = rt->peer->tcp_ts;
2490 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2491 }
2492 }
2493
2494 if (rt->fl.iif) {
2495 #ifdef CONFIG_IP_MROUTE
2496 __be32 dst = rt->rt_dst;
2497
2498 if (MULTICAST(dst) && !LOCAL_MCAST(dst) &&
2499 IPV4_DEVCONF_ALL(MC_FORWARDING)) {
2500 int err = ipmr_get_route(skb, r, nowait);
2501 if (err <= 0) {
2502 if (!nowait) {
2503 if (err == 0)
2504 return 0;
2505 goto nla_put_failure;
2506 } else {
2507 if (err == -EMSGSIZE)
2508 goto nla_put_failure;
2509 error = err;
2510 }
2511 }
2512 } else
2513 #endif
2514 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2515 }
2516
2517 if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2518 expires, error) < 0)
2519 goto nla_put_failure;
2520
2521 return nlmsg_end(skb, nlh);
2522
2523 nla_put_failure:
2524 nlmsg_cancel(skb, nlh);
2525 return -EMSGSIZE;
2526 }
2527
2528 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2529 {
2530 struct rtmsg *rtm;
2531 struct nlattr *tb[RTA_MAX+1];
2532 struct rtable *rt = NULL;
2533 __be32 dst = 0;
2534 __be32 src = 0;
2535 u32 iif;
2536 int err;
2537 struct sk_buff *skb;
2538
2539 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2540 if (err < 0)
2541 goto errout;
2542
2543 rtm = nlmsg_data(nlh);
2544
2545 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2546 if (skb == NULL) {
2547 err = -ENOBUFS;
2548 goto errout;
2549 }
2550
2551 /* Reserve room for dummy headers, this skb can pass
2552 through good chunk of routing engine.
2553 */
2554 skb_reset_mac_header(skb);
2555 skb_reset_network_header(skb);
2556
2557 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2558 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2559 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2560
2561 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2562 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2563 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2564
2565 if (iif) {
2566 struct net_device *dev;
2567
2568 dev = __dev_get_by_index(&init_net, iif);
2569 if (dev == NULL) {
2570 err = -ENODEV;
2571 goto errout_free;
2572 }
2573
2574 skb->protocol = htons(ETH_P_IP);
2575 skb->dev = dev;
2576 local_bh_disable();
2577 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2578 local_bh_enable();
2579
2580 rt = (struct rtable*) skb->dst;
2581 if (err == 0 && rt->u.dst.error)
2582 err = -rt->u.dst.error;
2583 } else {
2584 struct flowi fl = {
2585 .nl_u = {
2586 .ip4_u = {
2587 .daddr = dst,
2588 .saddr = src,
2589 .tos = rtm->rtm_tos,
2590 },
2591 },
2592 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2593 };
2594 err = ip_route_output_key(&rt, &fl);
2595 }
2596
2597 if (err)
2598 goto errout_free;
2599
2600 skb->dst = &rt->u.dst;
2601 if (rtm->rtm_flags & RTM_F_NOTIFY)
2602 rt->rt_flags |= RTCF_NOTIFY;
2603
2604 err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2605 RTM_NEWROUTE, 0, 0);
2606 if (err <= 0)
2607 goto errout_free;
2608
2609 err = rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
2610 errout:
2611 return err;
2612
2613 errout_free:
2614 kfree_skb(skb);
2615 goto errout;
2616 }
2617
2618 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2619 {
2620 struct rtable *rt;
2621 int h, s_h;
2622 int idx, s_idx;
2623
2624 s_h = cb->args[0];
2625 if (s_h < 0)
2626 s_h = 0;
2627 s_idx = idx = cb->args[1];
2628 for (h = s_h; h <= rt_hash_mask; h++) {
2629 rcu_read_lock_bh();
2630 for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt;
2631 rt = rcu_dereference(rt->u.dst.rt_next), idx++) {
2632 if (idx < s_idx)
2633 continue;
2634 skb->dst = dst_clone(&rt->u.dst);
2635 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
2636 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2637 1, NLM_F_MULTI) <= 0) {
2638 dst_release(xchg(&skb->dst, NULL));
2639 rcu_read_unlock_bh();
2640 goto done;
2641 }
2642 dst_release(xchg(&skb->dst, NULL));
2643 }
2644 rcu_read_unlock_bh();
2645 s_idx = 0;
2646 }
2647
2648 done:
2649 cb->args[0] = h;
2650 cb->args[1] = idx;
2651 return skb->len;
2652 }
2653
2654 void ip_rt_multicast_event(struct in_device *in_dev)
2655 {
2656 rt_cache_flush(0);
2657 }
2658
2659 #ifdef CONFIG_SYSCTL
2660 static int flush_delay;
2661
2662 static int ipv4_sysctl_rtcache_flush(ctl_table *ctl, int write,
2663 struct file *filp, void __user *buffer,
2664 size_t *lenp, loff_t *ppos)
2665 {
2666 if (write) {
2667 proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
2668 rt_cache_flush(flush_delay);
2669 return 0;
2670 }
2671
2672 return -EINVAL;
2673 }
2674
2675 static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table,
2676 int __user *name,
2677 int nlen,
2678 void __user *oldval,
2679 size_t __user *oldlenp,
2680 void __user *newval,
2681 size_t newlen)
2682 {
2683 int delay;
2684 if (newlen != sizeof(int))
2685 return -EINVAL;
2686 if (get_user(delay, (int __user *)newval))
2687 return -EFAULT;
2688 rt_cache_flush(delay);
2689 return 0;
2690 }
2691
2692 ctl_table ipv4_route_table[] = {
2693 {
2694 .ctl_name = NET_IPV4_ROUTE_FLUSH,
2695 .procname = "flush",
2696 .data = &flush_delay,
2697 .maxlen = sizeof(int),
2698 .mode = 0200,
2699 .proc_handler = &ipv4_sysctl_rtcache_flush,
2700 .strategy = &ipv4_sysctl_rtcache_flush_strategy,
2701 },
2702 {
2703 .ctl_name = NET_IPV4_ROUTE_MIN_DELAY,
2704 .procname = "min_delay",
2705 .data = &ip_rt_min_delay,
2706 .maxlen = sizeof(int),
2707 .mode = 0644,
2708 .proc_handler = &proc_dointvec_jiffies,
2709 .strategy = &sysctl_jiffies,
2710 },
2711 {
2712 .ctl_name = NET_IPV4_ROUTE_MAX_DELAY,
2713 .procname = "max_delay",
2714 .data = &ip_rt_max_delay,
2715 .maxlen = sizeof(int),
2716 .mode = 0644,
2717 .proc_handler = &proc_dointvec_jiffies,
2718 .strategy = &sysctl_jiffies,
2719 },
2720 {
2721 .ctl_name = NET_IPV4_ROUTE_GC_THRESH,
2722 .procname = "gc_thresh",
2723 .data = &ipv4_dst_ops.gc_thresh,
2724 .maxlen = sizeof(int),
2725 .mode = 0644,
2726 .proc_handler = &proc_dointvec,
2727 },
2728 {
2729 .ctl_name = NET_IPV4_ROUTE_MAX_SIZE,
2730 .procname = "max_size",
2731 .data = &ip_rt_max_size,
2732 .maxlen = sizeof(int),
2733 .mode = 0644,
2734 .proc_handler = &proc_dointvec,
2735 },
2736 {
2737 /* Deprecated. Use gc_min_interval_ms */
2738
2739 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL,
2740 .procname = "gc_min_interval",
2741 .data = &ip_rt_gc_min_interval,
2742 .maxlen = sizeof(int),
2743 .mode = 0644,
2744 .proc_handler = &proc_dointvec_jiffies,
2745 .strategy = &sysctl_jiffies,
2746 },
2747 {
2748 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS,
2749 .procname = "gc_min_interval_ms",
2750 .data = &ip_rt_gc_min_interval,
2751 .maxlen = sizeof(int),
2752 .mode = 0644,
2753 .proc_handler = &proc_dointvec_ms_jiffies,
2754 .strategy = &sysctl_ms_jiffies,
2755 },
2756 {
2757 .ctl_name = NET_IPV4_ROUTE_GC_TIMEOUT,
2758 .procname = "gc_timeout",
2759 .data = &ip_rt_gc_timeout,
2760 .maxlen = sizeof(int),
2761 .mode = 0644,
2762 .proc_handler = &proc_dointvec_jiffies,
2763 .strategy = &sysctl_jiffies,
2764 },
2765 {
2766 .ctl_name = NET_IPV4_ROUTE_GC_INTERVAL,
2767 .procname = "gc_interval",
2768 .data = &ip_rt_gc_interval,
2769 .maxlen = sizeof(int),
2770 .mode = 0644,
2771 .proc_handler = &proc_dointvec_jiffies,
2772 .strategy = &sysctl_jiffies,
2773 },
2774 {
2775 .ctl_name = NET_IPV4_ROUTE_REDIRECT_LOAD,
2776 .procname = "redirect_load",
2777 .data = &ip_rt_redirect_load,
2778 .maxlen = sizeof(int),
2779 .mode = 0644,
2780 .proc_handler = &proc_dointvec,
2781 },
2782 {
2783 .ctl_name = NET_IPV4_ROUTE_REDIRECT_NUMBER,
2784 .procname = "redirect_number",
2785 .data = &ip_rt_redirect_number,
2786 .maxlen = sizeof(int),
2787 .mode = 0644,
2788 .proc_handler = &proc_dointvec,
2789 },
2790 {
2791 .ctl_name = NET_IPV4_ROUTE_REDIRECT_SILENCE,
2792 .procname = "redirect_silence",
2793 .data = &ip_rt_redirect_silence,
2794 .maxlen = sizeof(int),
2795 .mode = 0644,
2796 .proc_handler = &proc_dointvec,
2797 },
2798 {
2799 .ctl_name = NET_IPV4_ROUTE_ERROR_COST,
2800 .procname = "error_cost",
2801 .data = &ip_rt_error_cost,
2802 .maxlen = sizeof(int),
2803 .mode = 0644,
2804 .proc_handler = &proc_dointvec,
2805 },
2806 {
2807 .ctl_name = NET_IPV4_ROUTE_ERROR_BURST,
2808 .procname = "error_burst",
2809 .data = &ip_rt_error_burst,
2810 .maxlen = sizeof(int),
2811 .mode = 0644,
2812 .proc_handler = &proc_dointvec,
2813 },
2814 {
2815 .ctl_name = NET_IPV4_ROUTE_GC_ELASTICITY,
2816 .procname = "gc_elasticity",
2817 .data = &ip_rt_gc_elasticity,
2818 .maxlen = sizeof(int),
2819 .mode = 0644,
2820 .proc_handler = &proc_dointvec,
2821 },
2822 {
2823 .ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES,
2824 .procname = "mtu_expires",
2825 .data = &ip_rt_mtu_expires,
2826 .maxlen = sizeof(int),
2827 .mode = 0644,
2828 .proc_handler = &proc_dointvec_jiffies,
2829 .strategy = &sysctl_jiffies,
2830 },
2831 {
2832 .ctl_name = NET_IPV4_ROUTE_MIN_PMTU,
2833 .procname = "min_pmtu",
2834 .data = &ip_rt_min_pmtu,
2835 .maxlen = sizeof(int),
2836 .mode = 0644,
2837 .proc_handler = &proc_dointvec,
2838 },
2839 {
2840 .ctl_name = NET_IPV4_ROUTE_MIN_ADVMSS,
2841 .procname = "min_adv_mss",
2842 .data = &ip_rt_min_advmss,
2843 .maxlen = sizeof(int),
2844 .mode = 0644,
2845 .proc_handler = &proc_dointvec,
2846 },
2847 {
2848 .ctl_name = NET_IPV4_ROUTE_SECRET_INTERVAL,
2849 .procname = "secret_interval",
2850 .data = &ip_rt_secret_interval,
2851 .maxlen = sizeof(int),
2852 .mode = 0644,
2853 .proc_handler = &proc_dointvec_jiffies,
2854 .strategy = &sysctl_jiffies,
2855 },
2856 { .ctl_name = 0 }
2857 };
2858 #endif
2859
2860 #ifdef CONFIG_NET_CLS_ROUTE
2861 struct ip_rt_acct *ip_rt_acct __read_mostly;
2862
2863 /* IP route accounting ptr for this logical cpu number. */
2864 #define IP_RT_ACCT_CPU(cpu) (per_cpu_ptr(ip_rt_acct, cpu))
2865
2866 #ifdef CONFIG_PROC_FS
2867 static int ip_rt_acct_read(char *buffer, char **start, off_t offset,
2868 int length, int *eof, void *data)
2869 {
2870 unsigned int i;
2871
2872 if ((offset & 3) || (length & 3))
2873 return -EIO;
2874
2875 if (offset >= sizeof(struct ip_rt_acct) * 256) {
2876 *eof = 1;
2877 return 0;
2878 }
2879
2880 if (offset + length >= sizeof(struct ip_rt_acct) * 256) {
2881 length = sizeof(struct ip_rt_acct) * 256 - offset;
2882 *eof = 1;
2883 }
2884
2885 offset /= sizeof(u32);
2886
2887 if (length > 0) {
2888 u32 *dst = (u32 *) buffer;
2889
2890 *start = buffer;
2891 memset(dst, 0, length);
2892
2893 for_each_possible_cpu(i) {
2894 unsigned int j;
2895 u32 *src = ((u32 *) IP_RT_ACCT_CPU(i)) + offset;
2896
2897 for (j = 0; j < length/4; j++)
2898 dst[j] += src[j];
2899 }
2900 }
2901 return length;
2902 }
2903 #endif /* CONFIG_PROC_FS */
2904 #endif /* CONFIG_NET_CLS_ROUTE */
2905
2906 static __initdata unsigned long rhash_entries;
2907 static int __init set_rhash_entries(char *str)
2908 {
2909 if (!str)
2910 return 0;
2911 rhash_entries = simple_strtoul(str, &str, 0);
2912 return 1;
2913 }
2914 __setup("rhash_entries=", set_rhash_entries);
2915
2916 int __init ip_rt_init(void)
2917 {
2918 int rc = 0;
2919
2920 rt_hash_rnd = (int) ((num_physpages ^ (num_physpages>>8)) ^
2921 (jiffies ^ (jiffies >> 7)));
2922
2923 #ifdef CONFIG_NET_CLS_ROUTE
2924 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct));
2925 if (!ip_rt_acct)
2926 panic("IP: failed to allocate ip_rt_acct\n");
2927 #endif
2928
2929 ipv4_dst_ops.kmem_cachep =
2930 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2931 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2932
2933 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2934
2935 rt_hash_table = (struct rt_hash_bucket *)
2936 alloc_large_system_hash("IP route cache",
2937 sizeof(struct rt_hash_bucket),
2938 rhash_entries,
2939 (num_physpages >= 128 * 1024) ?
2940 15 : 17,
2941 0,
2942 &rt_hash_log,
2943 &rt_hash_mask,
2944 0);
2945 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
2946 rt_hash_lock_init();
2947
2948 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
2949 ip_rt_max_size = (rt_hash_mask + 1) * 16;
2950
2951 devinet_init();
2952 ip_fib_init();
2953
2954 setup_timer(&rt_flush_timer, rt_run_flush, 0);
2955 setup_timer(&rt_secret_timer, rt_secret_rebuild, 0);
2956
2957 /* All the timers, started at system startup tend
2958 to synchronize. Perturb it a bit.
2959 */
2960 schedule_delayed_work(&expires_work,
2961 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
2962
2963 rt_secret_timer.expires = jiffies + net_random() % ip_rt_secret_interval +
2964 ip_rt_secret_interval;
2965 add_timer(&rt_secret_timer);
2966
2967 #ifdef CONFIG_PROC_FS
2968 {
2969 struct proc_dir_entry *rtstat_pde = NULL; /* keep gcc happy */
2970 if (!proc_net_fops_create(&init_net, "rt_cache", S_IRUGO, &rt_cache_seq_fops) ||
2971 !(rtstat_pde = create_proc_entry("rt_cache", S_IRUGO,
2972 init_net.proc_net_stat))) {
2973 return -ENOMEM;
2974 }
2975 rtstat_pde->proc_fops = &rt_cpu_seq_fops;
2976 }
2977 #ifdef CONFIG_NET_CLS_ROUTE
2978 create_proc_read_entry("rt_acct", 0, init_net.proc_net, ip_rt_acct_read, NULL);
2979 #endif
2980 #endif
2981 #ifdef CONFIG_XFRM
2982 xfrm_init();
2983 xfrm4_init();
2984 #endif
2985 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
2986
2987 return rc;
2988 }
2989
2990 EXPORT_SYMBOL(__ip_select_ident);
2991 EXPORT_SYMBOL(ip_route_input);
2992 EXPORT_SYMBOL(ip_route_output_key);
kernel-based virtual machine