?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[IPv4]: Convert route get to new netlink api
[hh.org.git] / net / ipv4 / route.c
bloba4d4cb85a16c27fa81600fdf6b4d67a42425f820
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/sched.h>
74 #include <linux/mm.h>
75 #include <linux/bootmem.h>
76 #include <linux/string.h>
77 #include <linux/socket.h>
78 #include <linux/sockios.h>
79 #include <linux/errno.h>
80 #include <linux/in.h>
81 #include <linux/inet.h>
82 #include <linux/netdevice.h>
83 #include <linux/proc_fs.h>
84 #include <linux/init.h>
85 #include <linux/skbuff.h>
86 #include <linux/rtnetlink.h>
87 #include <linux/inetdevice.h>
88 #include <linux/igmp.h>
89 #include <linux/pkt_sched.h>
90 #include <linux/mroute.h>
91 #include <linux/netfilter_ipv4.h>
92 #include <linux/random.h>
93 #include <linux/jhash.h>
94 #include <linux/rcupdate.h>
95 #include <linux/times.h>
96 #include <net/protocol.h>
97 #include <net/ip.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
102 #include <net/arp.h>
103 #include <net/tcp.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/ip_mp_alg.h>
107 #include <net/netevent.h>
108 #ifdef CONFIG_SYSCTL
109 #include <linux/sysctl.h>
110 #endif
111
112 #define RT_FL_TOS(oldflp) \
113 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
114
115 #define IP_MAX_MTU 0xFFF0
116
117 #define RT_GC_TIMEOUT (300*HZ)
118
119 static int ip_rt_min_delay = 2 * HZ;
120 static int ip_rt_max_delay = 10 * HZ;
121 static int ip_rt_max_size;
122 static int ip_rt_gc_timeout = RT_GC_TIMEOUT;
123 static int ip_rt_gc_interval = 60 * HZ;
124 static int ip_rt_gc_min_interval = HZ / 2;
125 static int ip_rt_redirect_number = 9;
126 static int ip_rt_redirect_load = HZ / 50;
127 static int ip_rt_redirect_silence = ((HZ / 50) << (9 + 1));
128 static int ip_rt_error_cost = HZ;
129 static int ip_rt_error_burst = 5 * HZ;
130 static int ip_rt_gc_elasticity = 8;
131 static int ip_rt_mtu_expires = 10 * 60 * HZ;
132 static int ip_rt_min_pmtu = 512 + 20 + 20;
133 static int ip_rt_min_advmss = 256;
134 static int ip_rt_secret_interval = 10 * 60 * HZ;
135 static unsigned long rt_deadline;
136
137 #define RTprint(a...) printk(KERN_DEBUG a)
138
139 static struct timer_list rt_flush_timer;
140 static struct timer_list rt_periodic_timer;
141 static struct timer_list rt_secret_timer;
142
143 /*
144 * Interface to generic destination cache.
145 */
146
147 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
148 static void ipv4_dst_destroy(struct dst_entry *dst);
149 static void ipv4_dst_ifdown(struct dst_entry *dst,
150 struct net_device *dev, int how);
151 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
152 static void ipv4_link_failure(struct sk_buff *skb);
153 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
154 static int rt_garbage_collect(void);
155
156
157 static struct dst_ops ipv4_dst_ops = {
158 .family = AF_INET,
159 .protocol = __constant_htons(ETH_P_IP),
160 .gc = rt_garbage_collect,
161 .check = ipv4_dst_check,
162 .destroy = ipv4_dst_destroy,
163 .ifdown = ipv4_dst_ifdown,
164 .negative_advice = ipv4_negative_advice,
165 .link_failure = ipv4_link_failure,
166 .update_pmtu = ip_rt_update_pmtu,
167 .entry_size = sizeof(struct rtable),
168 };
169
170 #define ECN_OR_COST(class) TC_PRIO_##class
171
172 __u8 ip_tos2prio[16] = {
173 TC_PRIO_BESTEFFORT,
174 ECN_OR_COST(FILLER),
175 TC_PRIO_BESTEFFORT,
176 ECN_OR_COST(BESTEFFORT),
177 TC_PRIO_BULK,
178 ECN_OR_COST(BULK),
179 TC_PRIO_BULK,
180 ECN_OR_COST(BULK),
181 TC_PRIO_INTERACTIVE,
182 ECN_OR_COST(INTERACTIVE),
183 TC_PRIO_INTERACTIVE,
184 ECN_OR_COST(INTERACTIVE),
185 TC_PRIO_INTERACTIVE_BULK,
186 ECN_OR_COST(INTERACTIVE_BULK),
187 TC_PRIO_INTERACTIVE_BULK,
188 ECN_OR_COST(INTERACTIVE_BULK)
189 };
190
191
192 /*
193 * Route cache.
194 */
195
196 /* The locking scheme is rather straight forward:
197 *
198 * 1) Read-Copy Update protects the buckets of the central route hash.
199 * 2) Only writers remove entries, and they hold the lock
200 * as they look at rtable reference counts.
201 * 3) Only readers acquire references to rtable entries,
202 * they do so with atomic increments and with the
203 * lock held.
204 */
205
206 struct rt_hash_bucket {
207 struct rtable *chain;
208 };
209 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
210 defined(CONFIG_PROVE_LOCKING)
211 /*
212 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
213 * The size of this table is a power of two and depends on the number of CPUS.
214 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
215 */
216 #ifdef CONFIG_LOCKDEP
217 # define RT_HASH_LOCK_SZ 256
218 #else
219 # if NR_CPUS >= 32
220 # define RT_HASH_LOCK_SZ 4096
221 # elif NR_CPUS >= 16
222 # define RT_HASH_LOCK_SZ 2048
223 # elif NR_CPUS >= 8
224 # define RT_HASH_LOCK_SZ 1024
225 # elif NR_CPUS >= 4
226 # define RT_HASH_LOCK_SZ 512
227 # else
228 # define RT_HASH_LOCK_SZ 256
229 # endif
230 #endif
231
232 static spinlock_t *rt_hash_locks;
233 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
234 # define rt_hash_lock_init() { \
235 int i; \
236 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ, GFP_KERNEL); \
237 if (!rt_hash_locks) panic("IP: failed to allocate rt_hash_locks\n"); \
238 for (i = 0; i < RT_HASH_LOCK_SZ; i++) \
239 spin_lock_init(&rt_hash_locks[i]); \
240 }
241 #else
242 # define rt_hash_lock_addr(slot) NULL
243 # define rt_hash_lock_init()
244 #endif
245
246 static struct rt_hash_bucket *rt_hash_table;
247 static unsigned rt_hash_mask;
248 static int rt_hash_log;
249 static unsigned int rt_hash_rnd;
250
251 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
252 #define RT_CACHE_STAT_INC(field) \
253 (__raw_get_cpu_var(rt_cache_stat).field++)
254
255 static int rt_intern_hash(unsigned hash, struct rtable *rth,
256 struct rtable **res);
257
258 static unsigned int rt_hash_code(u32 daddr, u32 saddr)
259 {
260 return (jhash_2words(daddr, saddr, rt_hash_rnd)
261 & rt_hash_mask);
262 }
263
264 #ifdef CONFIG_PROC_FS
265 struct rt_cache_iter_state {
266 int bucket;
267 };
268
269 static struct rtable *rt_cache_get_first(struct seq_file *seq)
270 {
271 struct rtable *r = NULL;
272 struct rt_cache_iter_state *st = seq->private;
273
274 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
275 rcu_read_lock_bh();
276 r = rt_hash_table[st->bucket].chain;
277 if (r)
278 break;
279 rcu_read_unlock_bh();
280 }
281 return r;
282 }
283
284 static struct rtable *rt_cache_get_next(struct seq_file *seq, struct rtable *r)
285 {
286 struct rt_cache_iter_state *st = rcu_dereference(seq->private);
287
288 r = r->u.rt_next;
289 while (!r) {
290 rcu_read_unlock_bh();
291 if (--st->bucket < 0)
292 break;
293 rcu_read_lock_bh();
294 r = rt_hash_table[st->bucket].chain;
295 }
296 return r;
297 }
298
299 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
300 {
301 struct rtable *r = rt_cache_get_first(seq);
302
303 if (r)
304 while (pos && (r = rt_cache_get_next(seq, r)))
305 --pos;
306 return pos ? NULL : r;
307 }
308
309 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
310 {
311 return *pos ? rt_cache_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
312 }
313
314 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
315 {
316 struct rtable *r = NULL;
317
318 if (v == SEQ_START_TOKEN)
319 r = rt_cache_get_first(seq);
320 else
321 r = rt_cache_get_next(seq, v);
322 ++*pos;
323 return r;
324 }
325
326 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
327 {
328 if (v && v != SEQ_START_TOKEN)
329 rcu_read_unlock_bh();
330 }
331
332 static int rt_cache_seq_show(struct seq_file *seq, void *v)
333 {
334 if (v == SEQ_START_TOKEN)
335 seq_printf(seq, "%-127s\n",
336 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
337 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
338 "HHUptod\tSpecDst");
339 else {
340 struct rtable *r = v;
341 char temp[256];
342
343 sprintf(temp, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
344 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X",
345 r->u.dst.dev ? r->u.dst.dev->name : "*",
346 (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
347 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
348 r->u.dst.__use, 0, (unsigned long)r->rt_src,
349 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
350 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
351 dst_metric(&r->u.dst, RTAX_WINDOW),
352 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
353 dst_metric(&r->u.dst, RTAX_RTTVAR)),
354 r->fl.fl4_tos,
355 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
356 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
357 dev_queue_xmit) : 0,
358 r->rt_spec_dst);
359 seq_printf(seq, "%-127s\n", temp);
360 }
361 return 0;
362 }
363
364 static struct seq_operations rt_cache_seq_ops = {
365 .start = rt_cache_seq_start,
366 .next = rt_cache_seq_next,
367 .stop = rt_cache_seq_stop,
368 .show = rt_cache_seq_show,
369 };
370
371 static int rt_cache_seq_open(struct inode *inode, struct file *file)
372 {
373 struct seq_file *seq;
374 int rc = -ENOMEM;
375 struct rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
376
377 if (!s)
378 goto out;
379 rc = seq_open(file, &rt_cache_seq_ops);
380 if (rc)
381 goto out_kfree;
382 seq = file->private_data;
383 seq->private = s;
384 memset(s, 0, sizeof(*s));
385 out:
386 return rc;
387 out_kfree:
388 kfree(s);
389 goto out;
390 }
391
392 static struct file_operations rt_cache_seq_fops = {
393 .owner = THIS_MODULE,
394 .open = rt_cache_seq_open,
395 .read = seq_read,
396 .llseek = seq_lseek,
397 .release = seq_release_private,
398 };
399
400
401 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
402 {
403 int cpu;
404
405 if (*pos == 0)
406 return SEQ_START_TOKEN;
407
408 for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
409 if (!cpu_possible(cpu))
410 continue;
411 *pos = cpu+1;
412 return &per_cpu(rt_cache_stat, cpu);
413 }
414 return NULL;
415 }
416
417 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
418 {
419 int cpu;
420
421 for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
422 if (!cpu_possible(cpu))
423 continue;
424 *pos = cpu+1;
425 return &per_cpu(rt_cache_stat, cpu);
426 }
427 return NULL;
428
429 }
430
431 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
432 {
433
434 }
435
436 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
437 {
438 struct rt_cache_stat *st = v;
439
440 if (v == SEQ_START_TOKEN) {
441 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");
442 return 0;
443 }
444
445 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
446 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
447 atomic_read(&ipv4_dst_ops.entries),
448 st->in_hit,
449 st->in_slow_tot,
450 st->in_slow_mc,
451 st->in_no_route,
452 st->in_brd,
453 st->in_martian_dst,
454 st->in_martian_src,
455
456 st->out_hit,
457 st->out_slow_tot,
458 st->out_slow_mc,
459
460 st->gc_total,
461 st->gc_ignored,
462 st->gc_goal_miss,
463 st->gc_dst_overflow,
464 st->in_hlist_search,
465 st->out_hlist_search
466 );
467 return 0;
468 }
469
470 static struct seq_operations rt_cpu_seq_ops = {
471 .start = rt_cpu_seq_start,
472 .next = rt_cpu_seq_next,
473 .stop = rt_cpu_seq_stop,
474 .show = rt_cpu_seq_show,
475 };
476
477
478 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
479 {
480 return seq_open(file, &rt_cpu_seq_ops);
481 }
482
483 static struct file_operations rt_cpu_seq_fops = {
484 .owner = THIS_MODULE,
485 .open = rt_cpu_seq_open,
486 .read = seq_read,
487 .llseek = seq_lseek,
488 .release = seq_release,
489 };
490
491 #endif /* CONFIG_PROC_FS */
492
493 static __inline__ void rt_free(struct rtable *rt)
494 {
495 multipath_remove(rt);
496 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
497 }
498
499 static __inline__ void rt_drop(struct rtable *rt)
500 {
501 multipath_remove(rt);
502 ip_rt_put(rt);
503 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
504 }
505
506 static __inline__ int rt_fast_clean(struct rtable *rth)
507 {
508 /* Kill broadcast/multicast entries very aggresively, if they
509 collide in hash table with more useful entries */
510 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
511 rth->fl.iif && rth->u.rt_next;
512 }
513
514 static __inline__ int rt_valuable(struct rtable *rth)
515 {
516 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
517 rth->u.dst.expires;
518 }
519
520 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
521 {
522 unsigned long age;
523 int ret = 0;
524
525 if (atomic_read(&rth->u.dst.__refcnt))
526 goto out;
527
528 ret = 1;
529 if (rth->u.dst.expires &&
530 time_after_eq(jiffies, rth->u.dst.expires))
531 goto out;
532
533 age = jiffies - rth->u.dst.lastuse;
534 ret = 0;
535 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
536 (age <= tmo2 && rt_valuable(rth)))
537 goto out;
538 ret = 1;
539 out: return ret;
540 }
541
542 /* Bits of score are:
543 * 31: very valuable
544 * 30: not quite useless
545 * 29..0: usage counter
546 */
547 static inline u32 rt_score(struct rtable *rt)
548 {
549 u32 score = jiffies - rt->u.dst.lastuse;
550
551 score = ~score & ~(3<<30);
552
553 if (rt_valuable(rt))
554 score |= (1<<31);
555
556 if (!rt->fl.iif ||
557 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
558 score |= (1<<30);
559
560 return score;
561 }
562
563 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
564 {
565 return memcmp(&fl1->nl_u.ip4_u, &fl2->nl_u.ip4_u, sizeof(fl1->nl_u.ip4_u)) == 0 &&
566 fl1->oif == fl2->oif &&
567 fl1->iif == fl2->iif;
568 }
569
570 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
571 static struct rtable **rt_remove_balanced_route(struct rtable **chain_head,
572 struct rtable *expentry,
573 int *removed_count)
574 {
575 int passedexpired = 0;
576 struct rtable **nextstep = NULL;
577 struct rtable **rthp = chain_head;
578 struct rtable *rth;
579
580 if (removed_count)
581 *removed_count = 0;
582
583 while ((rth = *rthp) != NULL) {
584 if (rth == expentry)
585 passedexpired = 1;
586
587 if (((*rthp)->u.dst.flags & DST_BALANCED) != 0 &&
588 compare_keys(&(*rthp)->fl, &expentry->fl)) {
589 if (*rthp == expentry) {
590 *rthp = rth->u.rt_next;
591 continue;
592 } else {
593 *rthp = rth->u.rt_next;
594 rt_free(rth);
595 if (removed_count)
596 ++(*removed_count);
597 }
598 } else {
599 if (!((*rthp)->u.dst.flags & DST_BALANCED) &&
600 passedexpired && !nextstep)
601 nextstep = &rth->u.rt_next;
602
603 rthp = &rth->u.rt_next;
604 }
605 }
606
607 rt_free(expentry);
608 if (removed_count)
609 ++(*removed_count);
610
611 return nextstep;
612 }
613 #endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
614
615
616 /* This runs via a timer and thus is always in BH context. */
617 static void rt_check_expire(unsigned long dummy)
618 {
619 static unsigned int rover;
620 unsigned int i = rover, goal;
621 struct rtable *rth, **rthp;
622 unsigned long now = jiffies;
623 u64 mult;
624
625 mult = ((u64)ip_rt_gc_interval) << rt_hash_log;
626 if (ip_rt_gc_timeout > 1)
627 do_div(mult, ip_rt_gc_timeout);
628 goal = (unsigned int)mult;
629 if (goal > rt_hash_mask) goal = rt_hash_mask + 1;
630 for (; goal > 0; goal--) {
631 unsigned long tmo = ip_rt_gc_timeout;
632
633 i = (i + 1) & rt_hash_mask;
634 rthp = &rt_hash_table[i].chain;
635
636 if (*rthp == 0)
637 continue;
638 spin_lock(rt_hash_lock_addr(i));
639 while ((rth = *rthp) != NULL) {
640 if (rth->u.dst.expires) {
641 /* Entry is expired even if it is in use */
642 if (time_before_eq(now, rth->u.dst.expires)) {
643 tmo >>= 1;
644 rthp = &rth->u.rt_next;
645 continue;
646 }
647 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
648 tmo >>= 1;
649 rthp = &rth->u.rt_next;
650 continue;
651 }
652
653 /* Cleanup aged off entries. */
654 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
655 /* remove all related balanced entries if necessary */
656 if (rth->u.dst.flags & DST_BALANCED) {
657 rthp = rt_remove_balanced_route(
658 &rt_hash_table[i].chain,
659 rth, NULL);
660 if (!rthp)
661 break;
662 } else {
663 *rthp = rth->u.rt_next;
664 rt_free(rth);
665 }
666 #else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
667 *rthp = rth->u.rt_next;
668 rt_free(rth);
669 #endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
670 }
671 spin_unlock(rt_hash_lock_addr(i));
672
673 /* Fallback loop breaker. */
674 if (time_after(jiffies, now))
675 break;
676 }
677 rover = i;
678 mod_timer(&rt_periodic_timer, jiffies + ip_rt_gc_interval);
679 }
680
681 /* This can run from both BH and non-BH contexts, the latter
682 * in the case of a forced flush event.
683 */
684 static void rt_run_flush(unsigned long dummy)
685 {
686 int i;
687 struct rtable *rth, *next;
688
689 rt_deadline = 0;
690
691 get_random_bytes(&rt_hash_rnd, 4);
692
693 for (i = rt_hash_mask; i >= 0; i--) {
694 spin_lock_bh(rt_hash_lock_addr(i));
695 rth = rt_hash_table[i].chain;
696 if (rth)
697 rt_hash_table[i].chain = NULL;
698 spin_unlock_bh(rt_hash_lock_addr(i));
699
700 for (; rth; rth = next) {
701 next = rth->u.rt_next;
702 rt_free(rth);
703 }
704 }
705 }
706
707 static DEFINE_SPINLOCK(rt_flush_lock);
708
709 void rt_cache_flush(int delay)
710 {
711 unsigned long now = jiffies;
712 int user_mode = !in_softirq();
713
714 if (delay < 0)
715 delay = ip_rt_min_delay;
716
717 /* flush existing multipath state*/
718 multipath_flush();
719
720 spin_lock_bh(&rt_flush_lock);
721
722 if (del_timer(&rt_flush_timer) && delay > 0 && rt_deadline) {
723 long tmo = (long)(rt_deadline - now);
724
725 /* If flush timer is already running
726 and flush request is not immediate (delay > 0):
727
728 if deadline is not achieved, prolongate timer to "delay",
729 otherwise fire it at deadline time.
730 */
731
732 if (user_mode && tmo < ip_rt_max_delay-ip_rt_min_delay)
733 tmo = 0;
734
735 if (delay > tmo)
736 delay = tmo;
737 }
738
739 if (delay <= 0) {
740 spin_unlock_bh(&rt_flush_lock);
741 rt_run_flush(0);
742 return;
743 }
744
745 if (rt_deadline == 0)
746 rt_deadline = now + ip_rt_max_delay;
747
748 mod_timer(&rt_flush_timer, now+delay);
749 spin_unlock_bh(&rt_flush_lock);
750 }
751
752 static void rt_secret_rebuild(unsigned long dummy)
753 {
754 unsigned long now = jiffies;
755
756 rt_cache_flush(0);
757 mod_timer(&rt_secret_timer, now + ip_rt_secret_interval);
758 }
759
760 /*
761 Short description of GC goals.
762
763 We want to build algorithm, which will keep routing cache
764 at some equilibrium point, when number of aged off entries
765 is kept approximately equal to newly generated ones.
766
767 Current expiration strength is variable "expire".
768 We try to adjust it dynamically, so that if networking
769 is idle expires is large enough to keep enough of warm entries,
770 and when load increases it reduces to limit cache size.
771 */
772
773 static int rt_garbage_collect(void)
774 {
775 static unsigned long expire = RT_GC_TIMEOUT;
776 static unsigned long last_gc;
777 static int rover;
778 static int equilibrium;
779 struct rtable *rth, **rthp;
780 unsigned long now = jiffies;
781 int goal;
782
783 /*
784 * Garbage collection is pretty expensive,
785 * do not make it too frequently.
786 */
787
788 RT_CACHE_STAT_INC(gc_total);
789
790 if (now - last_gc < ip_rt_gc_min_interval &&
791 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
792 RT_CACHE_STAT_INC(gc_ignored);
793 goto out;
794 }
795
796 /* Calculate number of entries, which we want to expire now. */
797 goal = atomic_read(&ipv4_dst_ops.entries) -
798 (ip_rt_gc_elasticity << rt_hash_log);
799 if (goal <= 0) {
800 if (equilibrium < ipv4_dst_ops.gc_thresh)
801 equilibrium = ipv4_dst_ops.gc_thresh;
802 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
803 if (goal > 0) {
804 equilibrium += min_t(unsigned int, goal / 2, rt_hash_mask + 1);
805 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
806 }
807 } else {
808 /* We are in dangerous area. Try to reduce cache really
809 * aggressively.
810 */
811 goal = max_t(unsigned int, goal / 2, rt_hash_mask + 1);
812 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
813 }
814
815 if (now - last_gc >= ip_rt_gc_min_interval)
816 last_gc = now;
817
818 if (goal <= 0) {
819 equilibrium += goal;
820 goto work_done;
821 }
822
823 do {
824 int i, k;
825
826 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
827 unsigned long tmo = expire;
828
829 k = (k + 1) & rt_hash_mask;
830 rthp = &rt_hash_table[k].chain;
831 spin_lock_bh(rt_hash_lock_addr(k));
832 while ((rth = *rthp) != NULL) {
833 if (!rt_may_expire(rth, tmo, expire)) {
834 tmo >>= 1;
835 rthp = &rth->u.rt_next;
836 continue;
837 }
838 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
839 /* remove all related balanced entries
840 * if necessary
841 */
842 if (rth->u.dst.flags & DST_BALANCED) {
843 int r;
844
845 rthp = rt_remove_balanced_route(
846 &rt_hash_table[k].chain,
847 rth,
848 &r);
849 goal -= r;
850 if (!rthp)
851 break;
852 } else {
853 *rthp = rth->u.rt_next;
854 rt_free(rth);
855 goal--;
856 }
857 #else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
858 *rthp = rth->u.rt_next;
859 rt_free(rth);
860 goal--;
861 #endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
862 }
863 spin_unlock_bh(rt_hash_lock_addr(k));
864 if (goal <= 0)
865 break;
866 }
867 rover = k;
868
869 if (goal <= 0)
870 goto work_done;
871
872 /* Goal is not achieved. We stop process if:
873
874 - if expire reduced to zero. Otherwise, expire is halfed.
875 - if table is not full.
876 - if we are called from interrupt.
877 - jiffies check is just fallback/debug loop breaker.
878 We will not spin here for long time in any case.
879 */
880
881 RT_CACHE_STAT_INC(gc_goal_miss);
882
883 if (expire == 0)
884 break;
885
886 expire >>= 1;
887 #if RT_CACHE_DEBUG >= 2
888 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
889 atomic_read(&ipv4_dst_ops.entries), goal, i);
890 #endif
891
892 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
893 goto out;
894 } while (!in_softirq() && time_before_eq(jiffies, now));
895
896 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
897 goto out;
898 if (net_ratelimit())
899 printk(KERN_WARNING "dst cache overflow\n");
900 RT_CACHE_STAT_INC(gc_dst_overflow);
901 return 1;
902
903 work_done:
904 expire += ip_rt_gc_min_interval;
905 if (expire > ip_rt_gc_timeout ||
906 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
907 expire = ip_rt_gc_timeout;
908 #if RT_CACHE_DEBUG >= 2
909 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
910 atomic_read(&ipv4_dst_ops.entries), goal, rover);
911 #endif
912 out: return 0;
913 }
914
915 static int rt_intern_hash(unsigned hash, struct rtable *rt, struct rtable **rp)
916 {
917 struct rtable *rth, **rthp;
918 unsigned long now;
919 struct rtable *cand, **candp;
920 u32 min_score;
921 int chain_length;
922 int attempts = !in_softirq();
923
924 restart:
925 chain_length = 0;
926 min_score = ~(u32)0;
927 cand = NULL;
928 candp = NULL;
929 now = jiffies;
930
931 rthp = &rt_hash_table[hash].chain;
932
933 spin_lock_bh(rt_hash_lock_addr(hash));
934 while ((rth = *rthp) != NULL) {
935 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
936 if (!(rth->u.dst.flags & DST_BALANCED) &&
937 compare_keys(&rth->fl, &rt->fl)) {
938 #else
939 if (compare_keys(&rth->fl, &rt->fl)) {
940 #endif
941 /* Put it first */
942 *rthp = rth->u.rt_next;
943 /*
944 * Since lookup is lockfree, the deletion
945 * must be visible to another weakly ordered CPU before
946 * the insertion at the start of the hash chain.
947 */
948 rcu_assign_pointer(rth->u.rt_next,
949 rt_hash_table[hash].chain);
950 /*
951 * Since lookup is lockfree, the update writes
952 * must be ordered for consistency on SMP.
953 */
954 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
955
956 rth->u.dst.__use++;
957 dst_hold(&rth->u.dst);
958 rth->u.dst.lastuse = now;
959 spin_unlock_bh(rt_hash_lock_addr(hash));
960
961 rt_drop(rt);
962 *rp = rth;
963 return 0;
964 }
965
966 if (!atomic_read(&rth->u.dst.__refcnt)) {
967 u32 score = rt_score(rth);
968
969 if (score <= min_score) {
970 cand = rth;
971 candp = rthp;
972 min_score = score;
973 }
974 }
975
976 chain_length++;
977
978 rthp = &rth->u.rt_next;
979 }
980
981 if (cand) {
982 /* ip_rt_gc_elasticity used to be average length of chain
983 * length, when exceeded gc becomes really aggressive.
984 *
985 * The second limit is less certain. At the moment it allows
986 * only 2 entries per bucket. We will see.
987 */
988 if (chain_length > ip_rt_gc_elasticity) {
989 *candp = cand->u.rt_next;
990 rt_free(cand);
991 }
992 }
993
994 /* Try to bind route to arp only if it is output
995 route or unicast forwarding path.
996 */
997 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
998 int err = arp_bind_neighbour(&rt->u.dst);
999 if (err) {
1000 spin_unlock_bh(rt_hash_lock_addr(hash));
1001
1002 if (err != -ENOBUFS) {
1003 rt_drop(rt);
1004 return err;
1005 }
1006
1007 /* Neighbour tables are full and nothing
1008 can be released. Try to shrink route cache,
1009 it is most likely it holds some neighbour records.
1010 */
1011 if (attempts-- > 0) {
1012 int saved_elasticity = ip_rt_gc_elasticity;
1013 int saved_int = ip_rt_gc_min_interval;
1014 ip_rt_gc_elasticity = 1;
1015 ip_rt_gc_min_interval = 0;
1016 rt_garbage_collect();
1017 ip_rt_gc_min_interval = saved_int;
1018 ip_rt_gc_elasticity = saved_elasticity;
1019 goto restart;
1020 }
1021
1022 if (net_ratelimit())
1023 printk(KERN_WARNING "Neighbour table overflow.\n");
1024 rt_drop(rt);
1025 return -ENOBUFS;
1026 }
1027 }
1028
1029 rt->u.rt_next = rt_hash_table[hash].chain;
1030 #if RT_CACHE_DEBUG >= 2
1031 if (rt->u.rt_next) {
1032 struct rtable *trt;
1033 printk(KERN_DEBUG "rt_cache @%02x: %u.%u.%u.%u", hash,
1034 NIPQUAD(rt->rt_dst));
1035 for (trt = rt->u.rt_next; trt; trt = trt->u.rt_next)
1036 printk(" . %u.%u.%u.%u", NIPQUAD(trt->rt_dst));
1037 printk("\n");
1038 }
1039 #endif
1040 rt_hash_table[hash].chain = rt;
1041 spin_unlock_bh(rt_hash_lock_addr(hash));
1042 *rp = rt;
1043 return 0;
1044 }
1045
1046 void rt_bind_peer(struct rtable *rt, int create)
1047 {
1048 static DEFINE_SPINLOCK(rt_peer_lock);
1049 struct inet_peer *peer;
1050
1051 peer = inet_getpeer(rt->rt_dst, create);
1052
1053 spin_lock_bh(&rt_peer_lock);
1054 if (rt->peer == NULL) {
1055 rt->peer = peer;
1056 peer = NULL;
1057 }
1058 spin_unlock_bh(&rt_peer_lock);
1059 if (peer)
1060 inet_putpeer(peer);
1061 }
1062
1063 /*
1064 * Peer allocation may fail only in serious out-of-memory conditions. However
1065 * we still can generate some output.
1066 * Random ID selection looks a bit dangerous because we have no chances to
1067 * select ID being unique in a reasonable period of time.
1068 * But broken packet identifier may be better than no packet at all.
1069 */
1070 static void ip_select_fb_ident(struct iphdr *iph)
1071 {
1072 static DEFINE_SPINLOCK(ip_fb_id_lock);
1073 static u32 ip_fallback_id;
1074 u32 salt;
1075
1076 spin_lock_bh(&ip_fb_id_lock);
1077 salt = secure_ip_id(ip_fallback_id ^ iph->daddr);
1078 iph->id = htons(salt & 0xFFFF);
1079 ip_fallback_id = salt;
1080 spin_unlock_bh(&ip_fb_id_lock);
1081 }
1082
1083 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1084 {
1085 struct rtable *rt = (struct rtable *) dst;
1086
1087 if (rt) {
1088 if (rt->peer == NULL)
1089 rt_bind_peer(rt, 1);
1090
1091 /* If peer is attached to destination, it is never detached,
1092 so that we need not to grab a lock to dereference it.
1093 */
1094 if (rt->peer) {
1095 iph->id = htons(inet_getid(rt->peer, more));
1096 return;
1097 }
1098 } else
1099 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1100 __builtin_return_address(0));
1101
1102 ip_select_fb_ident(iph);
1103 }
1104
1105 static void rt_del(unsigned hash, struct rtable *rt)
1106 {
1107 struct rtable **rthp;
1108
1109 spin_lock_bh(rt_hash_lock_addr(hash));
1110 ip_rt_put(rt);
1111 for (rthp = &rt_hash_table[hash].chain; *rthp;
1112 rthp = &(*rthp)->u.rt_next)
1113 if (*rthp == rt) {
1114 *rthp = rt->u.rt_next;
1115 rt_free(rt);
1116 break;
1117 }
1118 spin_unlock_bh(rt_hash_lock_addr(hash));
1119 }
1120
1121 void ip_rt_redirect(u32 old_gw, u32 daddr, u32 new_gw,
1122 u32 saddr, struct net_device *dev)
1123 {
1124 int i, k;
1125 struct in_device *in_dev = in_dev_get(dev);
1126 struct rtable *rth, **rthp;
1127 u32 skeys[2] = { saddr, 0 };
1128 int ikeys[2] = { dev->ifindex, 0 };
1129 struct netevent_redirect netevent;
1130
1131 if (!in_dev)
1132 return;
1133
1134 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev)
1135 || MULTICAST(new_gw) || BADCLASS(new_gw) || ZERONET(new_gw))
1136 goto reject_redirect;
1137
1138 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1139 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1140 goto reject_redirect;
1141 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1142 goto reject_redirect;
1143 } else {
1144 if (inet_addr_type(new_gw) != RTN_UNICAST)
1145 goto reject_redirect;
1146 }
1147
1148 for (i = 0; i < 2; i++) {
1149 for (k = 0; k < 2; k++) {
1150 unsigned hash = rt_hash_code(daddr,
1151 skeys[i] ^ (ikeys[k] << 5));
1152
1153 rthp=&rt_hash_table[hash].chain;
1154
1155 rcu_read_lock();
1156 while ((rth = rcu_dereference(*rthp)) != NULL) {
1157 struct rtable *rt;
1158
1159 if (rth->fl.fl4_dst != daddr ||
1160 rth->fl.fl4_src != skeys[i] ||
1161 rth->fl.oif != ikeys[k] ||
1162 rth->fl.iif != 0) {
1163 rthp = &rth->u.rt_next;
1164 continue;
1165 }
1166
1167 if (rth->rt_dst != daddr ||
1168 rth->rt_src != saddr ||
1169 rth->u.dst.error ||
1170 rth->rt_gateway != old_gw ||
1171 rth->u.dst.dev != dev)
1172 break;
1173
1174 dst_hold(&rth->u.dst);
1175 rcu_read_unlock();
1176
1177 rt = dst_alloc(&ipv4_dst_ops);
1178 if (rt == NULL) {
1179 ip_rt_put(rth);
1180 in_dev_put(in_dev);
1181 return;
1182 }
1183
1184 /* Copy all the information. */
1185 *rt = *rth;
1186 INIT_RCU_HEAD(&rt->u.dst.rcu_head);
1187 rt->u.dst.__use = 1;
1188 atomic_set(&rt->u.dst.__refcnt, 1);
1189 rt->u.dst.child = NULL;
1190 if (rt->u.dst.dev)
1191 dev_hold(rt->u.dst.dev);
1192 if (rt->idev)
1193 in_dev_hold(rt->idev);
1194 rt->u.dst.obsolete = 0;
1195 rt->u.dst.lastuse = jiffies;
1196 rt->u.dst.path = &rt->u.dst;
1197 rt->u.dst.neighbour = NULL;
1198 rt->u.dst.hh = NULL;
1199 rt->u.dst.xfrm = NULL;
1200
1201 rt->rt_flags |= RTCF_REDIRECTED;
1202
1203 /* Gateway is different ... */
1204 rt->rt_gateway = new_gw;
1205
1206 /* Redirect received -> path was valid */
1207 dst_confirm(&rth->u.dst);
1208
1209 if (rt->peer)
1210 atomic_inc(&rt->peer->refcnt);
1211
1212 if (arp_bind_neighbour(&rt->u.dst) ||
1213 !(rt->u.dst.neighbour->nud_state &
1214 NUD_VALID)) {
1215 if (rt->u.dst.neighbour)
1216 neigh_event_send(rt->u.dst.neighbour, NULL);
1217 ip_rt_put(rth);
1218 rt_drop(rt);
1219 goto do_next;
1220 }
1221
1222 netevent.old = &rth->u.dst;
1223 netevent.new = &rt->u.dst;
1224 call_netevent_notifiers(NETEVENT_REDIRECT,
1225 &netevent);
1226
1227 rt_del(hash, rth);
1228 if (!rt_intern_hash(hash, rt, &rt))
1229 ip_rt_put(rt);
1230 goto do_next;
1231 }
1232 rcu_read_unlock();
1233 do_next:
1234 ;
1235 }
1236 }
1237 in_dev_put(in_dev);
1238 return;
1239
1240 reject_redirect:
1241 #ifdef CONFIG_IP_ROUTE_VERBOSE
1242 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1243 printk(KERN_INFO "Redirect from %u.%u.%u.%u on %s about "
1244 "%u.%u.%u.%u ignored.\n"
1245 " Advised path = %u.%u.%u.%u -> %u.%u.%u.%u\n",
1246 NIPQUAD(old_gw), dev->name, NIPQUAD(new_gw),
1247 NIPQUAD(saddr), NIPQUAD(daddr));
1248 #endif
1249 in_dev_put(in_dev);
1250 }
1251
1252 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1253 {
1254 struct rtable *rt = (struct rtable*)dst;
1255 struct dst_entry *ret = dst;
1256
1257 if (rt) {
1258 if (dst->obsolete) {
1259 ip_rt_put(rt);
1260 ret = NULL;
1261 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1262 rt->u.dst.expires) {
1263 unsigned hash = rt_hash_code(rt->fl.fl4_dst,
1264 rt->fl.fl4_src ^
1265 (rt->fl.oif << 5));
1266 #if RT_CACHE_DEBUG >= 1
1267 printk(KERN_DEBUG "ip_rt_advice: redirect to "
1268 "%u.%u.%u.%u/%02x dropped\n",
1269 NIPQUAD(rt->rt_dst), rt->fl.fl4_tos);
1270 #endif
1271 rt_del(hash, rt);
1272 ret = NULL;
1273 }
1274 }
1275 return ret;
1276 }
1277
1278 /*
1279 * Algorithm:
1280 * 1. The first ip_rt_redirect_number redirects are sent
1281 * with exponential backoff, then we stop sending them at all,
1282 * assuming that the host ignores our redirects.
1283 * 2. If we did not see packets requiring redirects
1284 * during ip_rt_redirect_silence, we assume that the host
1285 * forgot redirected route and start to send redirects again.
1286 *
1287 * This algorithm is much cheaper and more intelligent than dumb load limiting
1288 * in icmp.c.
1289 *
1290 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1291 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1292 */
1293
1294 void ip_rt_send_redirect(struct sk_buff *skb)
1295 {
1296 struct rtable *rt = (struct rtable*)skb->dst;
1297 struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
1298
1299 if (!in_dev)
1300 return;
1301
1302 if (!IN_DEV_TX_REDIRECTS(in_dev))
1303 goto out;
1304
1305 /* No redirected packets during ip_rt_redirect_silence;
1306 * reset the algorithm.
1307 */
1308 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1309 rt->u.dst.rate_tokens = 0;
1310
1311 /* Too many ignored redirects; do not send anything
1312 * set u.dst.rate_last to the last seen redirected packet.
1313 */
1314 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1315 rt->u.dst.rate_last = jiffies;
1316 goto out;
1317 }
1318
1319 /* Check for load limit; set rate_last to the latest sent
1320 * redirect.
1321 */
1322 if (time_after(jiffies,
1323 (rt->u.dst.rate_last +
1324 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1325 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1326 rt->u.dst.rate_last = jiffies;
1327 ++rt->u.dst.rate_tokens;
1328 #ifdef CONFIG_IP_ROUTE_VERBOSE
1329 if (IN_DEV_LOG_MARTIANS(in_dev) &&
1330 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1331 net_ratelimit())
1332 printk(KERN_WARNING "host %u.%u.%u.%u/if%d ignores "
1333 "redirects for %u.%u.%u.%u to %u.%u.%u.%u.\n",
1334 NIPQUAD(rt->rt_src), rt->rt_iif,
1335 NIPQUAD(rt->rt_dst), NIPQUAD(rt->rt_gateway));
1336 #endif
1337 }
1338 out:
1339 in_dev_put(in_dev);
1340 }
1341
1342 static int ip_error(struct sk_buff *skb)
1343 {
1344 struct rtable *rt = (struct rtable*)skb->dst;
1345 unsigned long now;
1346 int code;
1347
1348 switch (rt->u.dst.error) {
1349 case EINVAL:
1350 default:
1351 goto out;
1352 case EHOSTUNREACH:
1353 code = ICMP_HOST_UNREACH;
1354 break;
1355 case ENETUNREACH:
1356 code = ICMP_NET_UNREACH;
1357 break;
1358 case EACCES:
1359 code = ICMP_PKT_FILTERED;
1360 break;
1361 }
1362
1363 now = jiffies;
1364 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1365 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1366 rt->u.dst.rate_tokens = ip_rt_error_burst;
1367 rt->u.dst.rate_last = now;
1368 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1369 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1370 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1371 }
1372
1373 out: kfree_skb(skb);
1374 return 0;
1375 }
1376
1377 /*
1378 * The last two values are not from the RFC but
1379 * are needed for AMPRnet AX.25 paths.
1380 */
1381
1382 static const unsigned short mtu_plateau[] =
1383 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1384
1385 static __inline__ unsigned short guess_mtu(unsigned short old_mtu)
1386 {
1387 int i;
1388
1389 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1390 if (old_mtu > mtu_plateau[i])
1391 return mtu_plateau[i];
1392 return 68;
1393 }
1394
1395 unsigned short ip_rt_frag_needed(struct iphdr *iph, unsigned short new_mtu)
1396 {
1397 int i;
1398 unsigned short old_mtu = ntohs(iph->tot_len);
1399 struct rtable *rth;
1400 u32 skeys[2] = { iph->saddr, 0, };
1401 u32 daddr = iph->daddr;
1402 unsigned short est_mtu = 0;
1403
1404 if (ipv4_config.no_pmtu_disc)
1405 return 0;
1406
1407 for (i = 0; i < 2; i++) {
1408 unsigned hash = rt_hash_code(daddr, skeys[i]);
1409
1410 rcu_read_lock();
1411 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1412 rth = rcu_dereference(rth->u.rt_next)) {
1413 if (rth->fl.fl4_dst == daddr &&
1414 rth->fl.fl4_src == skeys[i] &&
1415 rth->rt_dst == daddr &&
1416 rth->rt_src == iph->saddr &&
1417 rth->fl.iif == 0 &&
1418 !(dst_metric_locked(&rth->u.dst, RTAX_MTU))) {
1419 unsigned short mtu = new_mtu;
1420
1421 if (new_mtu < 68 || new_mtu >= old_mtu) {
1422
1423 /* BSD 4.2 compatibility hack :-( */
1424 if (mtu == 0 &&
1425 old_mtu >= rth->u.dst.metrics[RTAX_MTU-1] &&
1426 old_mtu >= 68 + (iph->ihl << 2))
1427 old_mtu -= iph->ihl << 2;
1428
1429 mtu = guess_mtu(old_mtu);
1430 }
1431 if (mtu <= rth->u.dst.metrics[RTAX_MTU-1]) {
1432 if (mtu < rth->u.dst.metrics[RTAX_MTU-1]) {
1433 dst_confirm(&rth->u.dst);
1434 if (mtu < ip_rt_min_pmtu) {
1435 mtu = ip_rt_min_pmtu;
1436 rth->u.dst.metrics[RTAX_LOCK-1] |=
1437 (1 << RTAX_MTU);
1438 }
1439 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1440 dst_set_expires(&rth->u.dst,
1441 ip_rt_mtu_expires);
1442 }
1443 est_mtu = mtu;
1444 }
1445 }
1446 }
1447 rcu_read_unlock();
1448 }
1449 return est_mtu ? : new_mtu;
1450 }
1451
1452 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1453 {
1454 if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= 68 &&
1455 !(dst_metric_locked(dst, RTAX_MTU))) {
1456 if (mtu < ip_rt_min_pmtu) {
1457 mtu = ip_rt_min_pmtu;
1458 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1459 }
1460 dst->metrics[RTAX_MTU-1] = mtu;
1461 dst_set_expires(dst, ip_rt_mtu_expires);
1462 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1463 }
1464 }
1465
1466 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1467 {
1468 return NULL;
1469 }
1470
1471 static void ipv4_dst_destroy(struct dst_entry *dst)
1472 {
1473 struct rtable *rt = (struct rtable *) dst;
1474 struct inet_peer *peer = rt->peer;
1475 struct in_device *idev = rt->idev;
1476
1477 if (peer) {
1478 rt->peer = NULL;
1479 inet_putpeer(peer);
1480 }
1481
1482 if (idev) {
1483 rt->idev = NULL;
1484 in_dev_put(idev);
1485 }
1486 }
1487
1488 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1489 int how)
1490 {
1491 struct rtable *rt = (struct rtable *) dst;
1492 struct in_device *idev = rt->idev;
1493 if (dev != &loopback_dev && idev && idev->dev == dev) {
1494 struct in_device *loopback_idev = in_dev_get(&loopback_dev);
1495 if (loopback_idev) {
1496 rt->idev = loopback_idev;
1497 in_dev_put(idev);
1498 }
1499 }
1500 }
1501
1502 static void ipv4_link_failure(struct sk_buff *skb)
1503 {
1504 struct rtable *rt;
1505
1506 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1507
1508 rt = (struct rtable *) skb->dst;
1509 if (rt)
1510 dst_set_expires(&rt->u.dst, 0);
1511 }
1512
1513 static int ip_rt_bug(struct sk_buff *skb)
1514 {
1515 printk(KERN_DEBUG "ip_rt_bug: %u.%u.%u.%u -> %u.%u.%u.%u, %s\n",
1516 NIPQUAD(skb->nh.iph->saddr), NIPQUAD(skb->nh.iph->daddr),
1517 skb->dev ? skb->dev->name : "?");
1518 kfree_skb(skb);
1519 return 0;
1520 }
1521
1522 /*
1523 We do not cache source address of outgoing interface,
1524 because it is used only by IP RR, TS and SRR options,
1525 so that it out of fast path.
1526
1527 BTW remember: "addr" is allowed to be not aligned
1528 in IP options!
1529 */
1530
1531 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1532 {
1533 u32 src;
1534 struct fib_result res;
1535
1536 if (rt->fl.iif == 0)
1537 src = rt->rt_src;
1538 else if (fib_lookup(&rt->fl, &res) == 0) {
1539 src = FIB_RES_PREFSRC(res);
1540 fib_res_put(&res);
1541 } else
1542 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1543 RT_SCOPE_UNIVERSE);
1544 memcpy(addr, &src, 4);
1545 }
1546
1547 #ifdef CONFIG_NET_CLS_ROUTE
1548 static void set_class_tag(struct rtable *rt, u32 tag)
1549 {
1550 if (!(rt->u.dst.tclassid & 0xFFFF))
1551 rt->u.dst.tclassid |= tag & 0xFFFF;
1552 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1553 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1554 }
1555 #endif
1556
1557 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1558 {
1559 struct fib_info *fi = res->fi;
1560
1561 if (fi) {
1562 if (FIB_RES_GW(*res) &&
1563 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1564 rt->rt_gateway = FIB_RES_GW(*res);
1565 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1566 sizeof(rt->u.dst.metrics));
1567 if (fi->fib_mtu == 0) {
1568 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1569 if (rt->u.dst.metrics[RTAX_LOCK-1] & (1 << RTAX_MTU) &&
1570 rt->rt_gateway != rt->rt_dst &&
1571 rt->u.dst.dev->mtu > 576)
1572 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1573 }
1574 #ifdef CONFIG_NET_CLS_ROUTE
1575 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1576 #endif
1577 } else
1578 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1579
1580 if (rt->u.dst.metrics[RTAX_HOPLIMIT-1] == 0)
1581 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1582 if (rt->u.dst.metrics[RTAX_MTU-1] > IP_MAX_MTU)
1583 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1584 if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0)
1585 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1586 ip_rt_min_advmss);
1587 if (rt->u.dst.metrics[RTAX_ADVMSS-1] > 65535 - 40)
1588 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1589
1590 #ifdef CONFIG_NET_CLS_ROUTE
1591 #ifdef CONFIG_IP_MULTIPLE_TABLES
1592 set_class_tag(rt, fib_rules_tclass(res));
1593 #endif
1594 set_class_tag(rt, itag);
1595 #endif
1596 rt->rt_type = res->type;
1597 }
1598
1599 static int ip_route_input_mc(struct sk_buff *skb, u32 daddr, u32 saddr,
1600 u8 tos, struct net_device *dev, int our)
1601 {
1602 unsigned hash;
1603 struct rtable *rth;
1604 u32 spec_dst;
1605 struct in_device *in_dev = in_dev_get(dev);
1606 u32 itag = 0;
1607
1608 /* Primary sanity checks. */
1609
1610 if (in_dev == NULL)
1611 return -EINVAL;
1612
1613 if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr) ||
1614 skb->protocol != htons(ETH_P_IP))
1615 goto e_inval;
1616
1617 if (ZERONET(saddr)) {
1618 if (!LOCAL_MCAST(daddr))
1619 goto e_inval;
1620 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1621 } else if (fib_validate_source(saddr, 0, tos, 0,
1622 dev, &spec_dst, &itag) < 0)
1623 goto e_inval;
1624
1625 rth = dst_alloc(&ipv4_dst_ops);
1626 if (!rth)
1627 goto e_nobufs;
1628
1629 rth->u.dst.output= ip_rt_bug;
1630
1631 atomic_set(&rth->u.dst.__refcnt, 1);
1632 rth->u.dst.flags= DST_HOST;
1633 if (in_dev->cnf.no_policy)
1634 rth->u.dst.flags |= DST_NOPOLICY;
1635 rth->fl.fl4_dst = daddr;
1636 rth->rt_dst = daddr;
1637 rth->fl.fl4_tos = tos;
1638 #ifdef CONFIG_IP_ROUTE_FWMARK
1639 rth->fl.fl4_fwmark= skb->nfmark;
1640 #endif
1641 rth->fl.fl4_src = saddr;
1642 rth->rt_src = saddr;
1643 #ifdef CONFIG_NET_CLS_ROUTE
1644 rth->u.dst.tclassid = itag;
1645 #endif
1646 rth->rt_iif =
1647 rth->fl.iif = dev->ifindex;
1648 rth->u.dst.dev = &loopback_dev;
1649 dev_hold(rth->u.dst.dev);
1650 rth->idev = in_dev_get(rth->u.dst.dev);
1651 rth->fl.oif = 0;
1652 rth->rt_gateway = daddr;
1653 rth->rt_spec_dst= spec_dst;
1654 rth->rt_type = RTN_MULTICAST;
1655 rth->rt_flags = RTCF_MULTICAST;
1656 if (our) {
1657 rth->u.dst.input= ip_local_deliver;
1658 rth->rt_flags |= RTCF_LOCAL;
1659 }
1660
1661 #ifdef CONFIG_IP_MROUTE
1662 if (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
1663 rth->u.dst.input = ip_mr_input;
1664 #endif
1665 RT_CACHE_STAT_INC(in_slow_mc);
1666
1667 in_dev_put(in_dev);
1668 hash = rt_hash_code(daddr, saddr ^ (dev->ifindex << 5));
1669 return rt_intern_hash(hash, rth, (struct rtable**) &skb->dst);
1670
1671 e_nobufs:
1672 in_dev_put(in_dev);
1673 return -ENOBUFS;
1674
1675 e_inval:
1676 in_dev_put(in_dev);
1677 return -EINVAL;
1678 }
1679
1680
1681 static void ip_handle_martian_source(struct net_device *dev,
1682 struct in_device *in_dev,
1683 struct sk_buff *skb,
1684 u32 daddr,
1685 u32 saddr)
1686 {
1687 RT_CACHE_STAT_INC(in_martian_src);
1688 #ifdef CONFIG_IP_ROUTE_VERBOSE
1689 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1690 /*
1691 * RFC1812 recommendation, if source is martian,
1692 * the only hint is MAC header.
1693 */
1694 printk(KERN_WARNING "martian source %u.%u.%u.%u from "
1695 "%u.%u.%u.%u, on dev %s\n",
1696 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
1697 if (dev->hard_header_len && skb->mac.raw) {
1698 int i;
1699 unsigned char *p = skb->mac.raw;
1700 printk(KERN_WARNING "ll header: ");
1701 for (i = 0; i < dev->hard_header_len; i++, p++) {
1702 printk("%02x", *p);
1703 if (i < (dev->hard_header_len - 1))
1704 printk(":");
1705 }
1706 printk("\n");
1707 }
1708 }
1709 #endif
1710 }
1711
1712 static inline int __mkroute_input(struct sk_buff *skb,
1713 struct fib_result* res,
1714 struct in_device *in_dev,
1715 u32 daddr, u32 saddr, u32 tos,
1716 struct rtable **result)
1717 {
1718
1719 struct rtable *rth;
1720 int err;
1721 struct in_device *out_dev;
1722 unsigned flags = 0;
1723 u32 spec_dst, itag;
1724
1725 /* get a working reference to the output device */
1726 out_dev = in_dev_get(FIB_RES_DEV(*res));
1727 if (out_dev == NULL) {
1728 if (net_ratelimit())
1729 printk(KERN_CRIT "Bug in ip_route_input" \
1730 "_slow(). Please, report\n");
1731 return -EINVAL;
1732 }
1733
1734
1735 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1736 in_dev->dev, &spec_dst, &itag);
1737 if (err < 0) {
1738 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1739 saddr);
1740
1741 err = -EINVAL;
1742 goto cleanup;
1743 }
1744
1745 if (err)
1746 flags |= RTCF_DIRECTSRC;
1747
1748 if (out_dev == in_dev && err && !(flags & (RTCF_NAT | RTCF_MASQ)) &&
1749 (IN_DEV_SHARED_MEDIA(out_dev) ||
1750 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1751 flags |= RTCF_DOREDIRECT;
1752
1753 if (skb->protocol != htons(ETH_P_IP)) {
1754 /* Not IP (i.e. ARP). Do not create route, if it is
1755 * invalid for proxy arp. DNAT routes are always valid.
1756 */
1757 if (out_dev == in_dev && !(flags & RTCF_DNAT)) {
1758 err = -EINVAL;
1759 goto cleanup;
1760 }
1761 }
1762
1763
1764 rth = dst_alloc(&ipv4_dst_ops);
1765 if (!rth) {
1766 err = -ENOBUFS;
1767 goto cleanup;
1768 }
1769
1770 atomic_set(&rth->u.dst.__refcnt, 1);
1771 rth->u.dst.flags= DST_HOST;
1772 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
1773 if (res->fi->fib_nhs > 1)
1774 rth->u.dst.flags |= DST_BALANCED;
1775 #endif
1776 if (in_dev->cnf.no_policy)
1777 rth->u.dst.flags |= DST_NOPOLICY;
1778 if (in_dev->cnf.no_xfrm)
1779 rth->u.dst.flags |= DST_NOXFRM;
1780 rth->fl.fl4_dst = daddr;
1781 rth->rt_dst = daddr;
1782 rth->fl.fl4_tos = tos;
1783 #ifdef CONFIG_IP_ROUTE_FWMARK
1784 rth->fl.fl4_fwmark= skb->nfmark;
1785 #endif
1786 rth->fl.fl4_src = saddr;
1787 rth->rt_src = saddr;
1788 rth->rt_gateway = daddr;
1789 rth->rt_iif =
1790 rth->fl.iif = in_dev->dev->ifindex;
1791 rth->u.dst.dev = (out_dev)->dev;
1792 dev_hold(rth->u.dst.dev);
1793 rth->idev = in_dev_get(rth->u.dst.dev);
1794 rth->fl.oif = 0;
1795 rth->rt_spec_dst= spec_dst;
1796
1797 rth->u.dst.input = ip_forward;
1798 rth->u.dst.output = ip_output;
1799
1800 rt_set_nexthop(rth, res, itag);
1801
1802 rth->rt_flags = flags;
1803
1804 *result = rth;
1805 err = 0;
1806 cleanup:
1807 /* release the working reference to the output device */
1808 in_dev_put(out_dev);
1809 return err;
1810 }
1811
1812 static inline int ip_mkroute_input_def(struct sk_buff *skb,
1813 struct fib_result* res,
1814 const struct flowi *fl,
1815 struct in_device *in_dev,
1816 u32 daddr, u32 saddr, u32 tos)
1817 {
1818 struct rtable* rth = NULL;
1819 int err;
1820 unsigned hash;
1821
1822 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1823 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
1824 fib_select_multipath(fl, res);
1825 #endif
1826
1827 /* create a routing cache entry */
1828 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1829 if (err)
1830 return err;
1831
1832 /* put it into the cache */
1833 hash = rt_hash_code(daddr, saddr ^ (fl->iif << 5));
1834 return rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
1835 }
1836
1837 static inline int ip_mkroute_input(struct sk_buff *skb,
1838 struct fib_result* res,
1839 const struct flowi *fl,
1840 struct in_device *in_dev,
1841 u32 daddr, u32 saddr, u32 tos)
1842 {
1843 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
1844 struct rtable* rth = NULL, *rtres;
1845 unsigned char hop, hopcount;
1846 int err = -EINVAL;
1847 unsigned int hash;
1848
1849 if (res->fi)
1850 hopcount = res->fi->fib_nhs;
1851 else
1852 hopcount = 1;
1853
1854 /* distinguish between multipath and singlepath */
1855 if (hopcount < 2)
1856 return ip_mkroute_input_def(skb, res, fl, in_dev, daddr,
1857 saddr, tos);
1858
1859 /* add all alternatives to the routing cache */
1860 for (hop = 0; hop < hopcount; hop++) {
1861 res->nh_sel = hop;
1862
1863 /* put reference to previous result */
1864 if (hop)
1865 ip_rt_put(rtres);
1866
1867 /* create a routing cache entry */
1868 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos,
1869 &rth);
1870 if (err)
1871 return err;
1872
1873 /* put it into the cache */
1874 hash = rt_hash_code(daddr, saddr ^ (fl->iif << 5));
1875 err = rt_intern_hash(hash, rth, &rtres);
1876 if (err)
1877 return err;
1878
1879 /* forward hop information to multipath impl. */
1880 multipath_set_nhinfo(rth,
1881 FIB_RES_NETWORK(*res),
1882 FIB_RES_NETMASK(*res),
1883 res->prefixlen,
1884 &FIB_RES_NH(*res));
1885 }
1886 skb->dst = &rtres->u.dst;
1887 return err;
1888 #else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
1889 return ip_mkroute_input_def(skb, res, fl, in_dev, daddr, saddr, tos);
1890 #endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
1891 }
1892
1893
1894 /*
1895 * NOTE. We drop all the packets that has local source
1896 * addresses, because every properly looped back packet
1897 * must have correct destination already attached by output routine.
1898 *
1899 * Such approach solves two big problems:
1900 * 1. Not simplex devices are handled properly.
1901 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1902 */
1903
1904 static int ip_route_input_slow(struct sk_buff *skb, u32 daddr, u32 saddr,
1905 u8 tos, struct net_device *dev)
1906 {
1907 struct fib_result res;
1908 struct in_device *in_dev = in_dev_get(dev);
1909 struct flowi fl = { .nl_u = { .ip4_u =
1910 { .daddr = daddr,
1911 .saddr = saddr,
1912 .tos = tos,
1913 .scope = RT_SCOPE_UNIVERSE,
1914 #ifdef CONFIG_IP_ROUTE_FWMARK
1915 .fwmark = skb->nfmark
1916 #endif
1917 } },
1918 .iif = dev->ifindex };
1919 unsigned flags = 0;
1920 u32 itag = 0;
1921 struct rtable * rth;
1922 unsigned hash;
1923 u32 spec_dst;
1924 int err = -EINVAL;
1925 int free_res = 0;
1926
1927 /* IP on this device is disabled. */
1928
1929 if (!in_dev)
1930 goto out;
1931
1932 /* Check for the most weird martians, which can be not detected
1933 by fib_lookup.
1934 */
1935
1936 if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr))
1937 goto martian_source;
1938
1939 if (daddr == 0xFFFFFFFF || (saddr == 0 && daddr == 0))
1940 goto brd_input;
1941
1942 /* Accept zero addresses only to limited broadcast;
1943 * I even do not know to fix it or not. Waiting for complains :-)
1944 */
1945 if (ZERONET(saddr))
1946 goto martian_source;
1947
1948 if (BADCLASS(daddr) || ZERONET(daddr) || LOOPBACK(daddr))
1949 goto martian_destination;
1950
1951 /*
1952 * Now we are ready to route packet.
1953 */
1954 if ((err = fib_lookup(&fl, &res)) != 0) {
1955 if (!IN_DEV_FORWARD(in_dev))
1956 goto e_hostunreach;
1957 goto no_route;
1958 }
1959 free_res = 1;
1960
1961 RT_CACHE_STAT_INC(in_slow_tot);
1962
1963 if (res.type == RTN_BROADCAST)
1964 goto brd_input;
1965
1966 if (res.type == RTN_LOCAL) {
1967 int result;
1968 result = fib_validate_source(saddr, daddr, tos,
1969 loopback_dev.ifindex,
1970 dev, &spec_dst, &itag);
1971 if (result < 0)
1972 goto martian_source;
1973 if (result)
1974 flags |= RTCF_DIRECTSRC;
1975 spec_dst = daddr;
1976 goto local_input;
1977 }
1978
1979 if (!IN_DEV_FORWARD(in_dev))
1980 goto e_hostunreach;
1981 if (res.type != RTN_UNICAST)
1982 goto martian_destination;
1983
1984 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
1985 if (err == -ENOBUFS)
1986 goto e_nobufs;
1987 if (err == -EINVAL)
1988 goto e_inval;
1989
1990 done:
1991 in_dev_put(in_dev);
1992 if (free_res)
1993 fib_res_put(&res);
1994 out: return err;
1995
1996 brd_input:
1997 if (skb->protocol != htons(ETH_P_IP))
1998 goto e_inval;
1999
2000 if (ZERONET(saddr))
2001 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2002 else {
2003 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2004 &itag);
2005 if (err < 0)
2006 goto martian_source;
2007 if (err)
2008 flags |= RTCF_DIRECTSRC;
2009 }
2010 flags |= RTCF_BROADCAST;
2011 res.type = RTN_BROADCAST;
2012 RT_CACHE_STAT_INC(in_brd);
2013
2014 local_input:
2015 rth = dst_alloc(&ipv4_dst_ops);
2016 if (!rth)
2017 goto e_nobufs;
2018
2019 rth->u.dst.output= ip_rt_bug;
2020
2021 atomic_set(&rth->u.dst.__refcnt, 1);
2022 rth->u.dst.flags= DST_HOST;
2023 if (in_dev->cnf.no_policy)
2024 rth->u.dst.flags |= DST_NOPOLICY;
2025 rth->fl.fl4_dst = daddr;
2026 rth->rt_dst = daddr;
2027 rth->fl.fl4_tos = tos;
2028 #ifdef CONFIG_IP_ROUTE_FWMARK
2029 rth->fl.fl4_fwmark= skb->nfmark;
2030 #endif
2031 rth->fl.fl4_src = saddr;
2032 rth->rt_src = saddr;
2033 #ifdef CONFIG_NET_CLS_ROUTE
2034 rth->u.dst.tclassid = itag;
2035 #endif
2036 rth->rt_iif =
2037 rth->fl.iif = dev->ifindex;
2038 rth->u.dst.dev = &loopback_dev;
2039 dev_hold(rth->u.dst.dev);
2040 rth->idev = in_dev_get(rth->u.dst.dev);
2041 rth->rt_gateway = daddr;
2042 rth->rt_spec_dst= spec_dst;
2043 rth->u.dst.input= ip_local_deliver;
2044 rth->rt_flags = flags|RTCF_LOCAL;
2045 if (res.type == RTN_UNREACHABLE) {
2046 rth->u.dst.input= ip_error;
2047 rth->u.dst.error= -err;
2048 rth->rt_flags &= ~RTCF_LOCAL;
2049 }
2050 rth->rt_type = res.type;
2051 hash = rt_hash_code(daddr, saddr ^ (fl.iif << 5));
2052 err = rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
2053 goto done;
2054
2055 no_route:
2056 RT_CACHE_STAT_INC(in_no_route);
2057 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2058 res.type = RTN_UNREACHABLE;
2059 goto local_input;
2060
2061 /*
2062 * Do not cache martian addresses: they should be logged (RFC1812)
2063 */
2064 martian_destination:
2065 RT_CACHE_STAT_INC(in_martian_dst);
2066 #ifdef CONFIG_IP_ROUTE_VERBOSE
2067 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2068 printk(KERN_WARNING "martian destination %u.%u.%u.%u from "
2069 "%u.%u.%u.%u, dev %s\n",
2070 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
2071 #endif
2072
2073 e_hostunreach:
2074 err = -EHOSTUNREACH;
2075 goto done;
2076
2077 e_inval:
2078 err = -EINVAL;
2079 goto done;
2080
2081 e_nobufs:
2082 err = -ENOBUFS;
2083 goto done;
2084
2085 martian_source:
2086 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2087 goto e_inval;
2088 }
2089
2090 int ip_route_input(struct sk_buff *skb, u32 daddr, u32 saddr,
2091 u8 tos, struct net_device *dev)
2092 {
2093 struct rtable * rth;
2094 unsigned hash;
2095 int iif = dev->ifindex;
2096
2097 tos &= IPTOS_RT_MASK;
2098 hash = rt_hash_code(daddr, saddr ^ (iif << 5));
2099
2100 rcu_read_lock();
2101 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2102 rth = rcu_dereference(rth->u.rt_next)) {
2103 if (rth->fl.fl4_dst == daddr &&
2104 rth->fl.fl4_src == saddr &&
2105 rth->fl.iif == iif &&
2106 rth->fl.oif == 0 &&
2107 #ifdef CONFIG_IP_ROUTE_FWMARK
2108 rth->fl.fl4_fwmark == skb->nfmark &&
2109 #endif
2110 rth->fl.fl4_tos == tos) {
2111 rth->u.dst.lastuse = jiffies;
2112 dst_hold(&rth->u.dst);
2113 rth->u.dst.__use++;
2114 RT_CACHE_STAT_INC(in_hit);
2115 rcu_read_unlock();
2116 skb->dst = (struct dst_entry*)rth;
2117 return 0;
2118 }
2119 RT_CACHE_STAT_INC(in_hlist_search);
2120 }
2121 rcu_read_unlock();
2122
2123 /* Multicast recognition logic is moved from route cache to here.
2124 The problem was that too many Ethernet cards have broken/missing
2125 hardware multicast filters :-( As result the host on multicasting
2126 network acquires a lot of useless route cache entries, sort of
2127 SDR messages from all the world. Now we try to get rid of them.
2128 Really, provided software IP multicast filter is organized
2129 reasonably (at least, hashed), it does not result in a slowdown
2130 comparing with route cache reject entries.
2131 Note, that multicast routers are not affected, because
2132 route cache entry is created eventually.
2133 */
2134 if (MULTICAST(daddr)) {
2135 struct in_device *in_dev;
2136
2137 rcu_read_lock();
2138 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2139 int our = ip_check_mc(in_dev, daddr, saddr,
2140 skb->nh.iph->protocol);
2141 if (our
2142 #ifdef CONFIG_IP_MROUTE
2143 || (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
2144 #endif
2145 ) {
2146 rcu_read_unlock();
2147 return ip_route_input_mc(skb, daddr, saddr,
2148 tos, dev, our);
2149 }
2150 }
2151 rcu_read_unlock();
2152 return -EINVAL;
2153 }
2154 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2155 }
2156
2157 static inline int __mkroute_output(struct rtable **result,
2158 struct fib_result* res,
2159 const struct flowi *fl,
2160 const struct flowi *oldflp,
2161 struct net_device *dev_out,
2162 unsigned flags)
2163 {
2164 struct rtable *rth;
2165 struct in_device *in_dev;
2166 u32 tos = RT_FL_TOS(oldflp);
2167 int err = 0;
2168
2169 if (LOOPBACK(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2170 return -EINVAL;
2171
2172 if (fl->fl4_dst == 0xFFFFFFFF)
2173 res->type = RTN_BROADCAST;
2174 else if (MULTICAST(fl->fl4_dst))
2175 res->type = RTN_MULTICAST;
2176 else if (BADCLASS(fl->fl4_dst) || ZERONET(fl->fl4_dst))
2177 return -EINVAL;
2178
2179 if (dev_out->flags & IFF_LOOPBACK)
2180 flags |= RTCF_LOCAL;
2181
2182 /* get work reference to inet device */
2183 in_dev = in_dev_get(dev_out);
2184 if (!in_dev)
2185 return -EINVAL;
2186
2187 if (res->type == RTN_BROADCAST) {
2188 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2189 if (res->fi) {
2190 fib_info_put(res->fi);
2191 res->fi = NULL;
2192 }
2193 } else if (res->type == RTN_MULTICAST) {
2194 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2195 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2196 oldflp->proto))
2197 flags &= ~RTCF_LOCAL;
2198 /* If multicast route do not exist use
2199 default one, but do not gateway in this case.
2200 Yes, it is hack.
2201 */
2202 if (res->fi && res->prefixlen < 4) {
2203 fib_info_put(res->fi);
2204 res->fi = NULL;
2205 }
2206 }
2207
2208
2209 rth = dst_alloc(&ipv4_dst_ops);
2210 if (!rth) {
2211 err = -ENOBUFS;
2212 goto cleanup;
2213 }
2214
2215 atomic_set(&rth->u.dst.__refcnt, 1);
2216 rth->u.dst.flags= DST_HOST;
2217 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
2218 if (res->fi) {
2219 rth->rt_multipath_alg = res->fi->fib_mp_alg;
2220 if (res->fi->fib_nhs > 1)
2221 rth->u.dst.flags |= DST_BALANCED;
2222 }
2223 #endif
2224 if (in_dev->cnf.no_xfrm)
2225 rth->u.dst.flags |= DST_NOXFRM;
2226 if (in_dev->cnf.no_policy)
2227 rth->u.dst.flags |= DST_NOPOLICY;
2228
2229 rth->fl.fl4_dst = oldflp->fl4_dst;
2230 rth->fl.fl4_tos = tos;
2231 rth->fl.fl4_src = oldflp->fl4_src;
2232 rth->fl.oif = oldflp->oif;
2233 #ifdef CONFIG_IP_ROUTE_FWMARK
2234 rth->fl.fl4_fwmark= oldflp->fl4_fwmark;
2235 #endif
2236 rth->rt_dst = fl->fl4_dst;
2237 rth->rt_src = fl->fl4_src;
2238 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2239 /* get references to the devices that are to be hold by the routing
2240 cache entry */
2241 rth->u.dst.dev = dev_out;
2242 dev_hold(dev_out);
2243 rth->idev = in_dev_get(dev_out);
2244 rth->rt_gateway = fl->fl4_dst;
2245 rth->rt_spec_dst= fl->fl4_src;
2246
2247 rth->u.dst.output=ip_output;
2248
2249 RT_CACHE_STAT_INC(out_slow_tot);
2250
2251 if (flags & RTCF_LOCAL) {
2252 rth->u.dst.input = ip_local_deliver;
2253 rth->rt_spec_dst = fl->fl4_dst;
2254 }
2255 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2256 rth->rt_spec_dst = fl->fl4_src;
2257 if (flags & RTCF_LOCAL &&
2258 !(dev_out->flags & IFF_LOOPBACK)) {
2259 rth->u.dst.output = ip_mc_output;
2260 RT_CACHE_STAT_INC(out_slow_mc);
2261 }
2262 #ifdef CONFIG_IP_MROUTE
2263 if (res->type == RTN_MULTICAST) {
2264 if (IN_DEV_MFORWARD(in_dev) &&
2265 !LOCAL_MCAST(oldflp->fl4_dst)) {
2266 rth->u.dst.input = ip_mr_input;
2267 rth->u.dst.output = ip_mc_output;
2268 }
2269 }
2270 #endif
2271 }
2272
2273 rt_set_nexthop(rth, res, 0);
2274
2275 rth->rt_flags = flags;
2276
2277 *result = rth;
2278 cleanup:
2279 /* release work reference to inet device */
2280 in_dev_put(in_dev);
2281
2282 return err;
2283 }
2284
2285 static inline int ip_mkroute_output_def(struct rtable **rp,
2286 struct fib_result* res,
2287 const struct flowi *fl,
2288 const struct flowi *oldflp,
2289 struct net_device *dev_out,
2290 unsigned flags)
2291 {
2292 struct rtable *rth = NULL;
2293 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2294 unsigned hash;
2295 if (err == 0) {
2296 hash = rt_hash_code(oldflp->fl4_dst,
2297 oldflp->fl4_src ^ (oldflp->oif << 5));
2298 err = rt_intern_hash(hash, rth, rp);
2299 }
2300
2301 return err;
2302 }
2303
2304 static inline int ip_mkroute_output(struct rtable** rp,
2305 struct fib_result* res,
2306 const struct flowi *fl,
2307 const struct flowi *oldflp,
2308 struct net_device *dev_out,
2309 unsigned flags)
2310 {
2311 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
2312 unsigned char hop;
2313 unsigned hash;
2314 int err = -EINVAL;
2315 struct rtable *rth = NULL;
2316
2317 if (res->fi && res->fi->fib_nhs > 1) {
2318 unsigned char hopcount = res->fi->fib_nhs;
2319
2320 for (hop = 0; hop < hopcount; hop++) {
2321 struct net_device *dev2nexthop;
2322
2323 res->nh_sel = hop;
2324
2325 /* hold a work reference to the output device */
2326 dev2nexthop = FIB_RES_DEV(*res);
2327 dev_hold(dev2nexthop);
2328
2329 /* put reference to previous result */
2330 if (hop)
2331 ip_rt_put(*rp);
2332
2333 err = __mkroute_output(&rth, res, fl, oldflp,
2334 dev2nexthop, flags);
2335
2336 if (err != 0)
2337 goto cleanup;
2338
2339 hash = rt_hash_code(oldflp->fl4_dst,
2340 oldflp->fl4_src ^
2341 (oldflp->oif << 5));
2342 err = rt_intern_hash(hash, rth, rp);
2343
2344 /* forward hop information to multipath impl. */
2345 multipath_set_nhinfo(rth,
2346 FIB_RES_NETWORK(*res),
2347 FIB_RES_NETMASK(*res),
2348 res->prefixlen,
2349 &FIB_RES_NH(*res));
2350 cleanup:
2351 /* release work reference to output device */
2352 dev_put(dev2nexthop);
2353
2354 if (err != 0)
2355 return err;
2356 }
2357 return err;
2358 } else {
2359 return ip_mkroute_output_def(rp, res, fl, oldflp, dev_out,
2360 flags);
2361 }
2362 #else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
2363 return ip_mkroute_output_def(rp, res, fl, oldflp, dev_out, flags);
2364 #endif
2365 }
2366
2367 /*
2368 * Major route resolver routine.
2369 */
2370
2371 static int ip_route_output_slow(struct rtable **rp, const struct flowi *oldflp)
2372 {
2373 u32 tos = RT_FL_TOS(oldflp);
2374 struct flowi fl = { .nl_u = { .ip4_u =
2375 { .daddr = oldflp->fl4_dst,
2376 .saddr = oldflp->fl4_src,
2377 .tos = tos & IPTOS_RT_MASK,
2378 .scope = ((tos & RTO_ONLINK) ?
2379 RT_SCOPE_LINK :
2380 RT_SCOPE_UNIVERSE),
2381 #ifdef CONFIG_IP_ROUTE_FWMARK
2382 .fwmark = oldflp->fl4_fwmark
2383 #endif
2384 } },
2385 .iif = loopback_dev.ifindex,
2386 .oif = oldflp->oif };
2387 struct fib_result res;
2388 unsigned flags = 0;
2389 struct net_device *dev_out = NULL;
2390 int free_res = 0;
2391 int err;
2392
2393
2394 res.fi = NULL;
2395 #ifdef CONFIG_IP_MULTIPLE_TABLES
2396 res.r = NULL;
2397 #endif
2398
2399 if (oldflp->fl4_src) {
2400 err = -EINVAL;
2401 if (MULTICAST(oldflp->fl4_src) ||
2402 BADCLASS(oldflp->fl4_src) ||
2403 ZERONET(oldflp->fl4_src))
2404 goto out;
2405
2406 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2407 dev_out = ip_dev_find(oldflp->fl4_src);
2408 if (dev_out == NULL)
2409 goto out;
2410
2411 /* I removed check for oif == dev_out->oif here.
2412 It was wrong for two reasons:
2413 1. ip_dev_find(saddr) can return wrong iface, if saddr is
2414 assigned to multiple interfaces.
2415 2. Moreover, we are allowed to send packets with saddr
2416 of another iface. --ANK
2417 */
2418
2419 if (oldflp->oif == 0
2420 && (MULTICAST(oldflp->fl4_dst) || oldflp->fl4_dst == 0xFFFFFFFF)) {
2421 /* Special hack: user can direct multicasts
2422 and limited broadcast via necessary interface
2423 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2424 This hack is not just for fun, it allows
2425 vic,vat and friends to work.
2426 They bind socket to loopback, set ttl to zero
2427 and expect that it will work.
2428 From the viewpoint of routing cache they are broken,
2429 because we are not allowed to build multicast path
2430 with loopback source addr (look, routing cache
2431 cannot know, that ttl is zero, so that packet
2432 will not leave this host and route is valid).
2433 Luckily, this hack is good workaround.
2434 */
2435
2436 fl.oif = dev_out->ifindex;
2437 goto make_route;
2438 }
2439 if (dev_out)
2440 dev_put(dev_out);
2441 dev_out = NULL;
2442 }
2443
2444
2445 if (oldflp->oif) {
2446 dev_out = dev_get_by_index(oldflp->oif);
2447 err = -ENODEV;
2448 if (dev_out == NULL)
2449 goto out;
2450
2451 /* RACE: Check return value of inet_select_addr instead. */
2452 if (__in_dev_get_rtnl(dev_out) == NULL) {
2453 dev_put(dev_out);
2454 goto out; /* Wrong error code */
2455 }
2456
2457 if (LOCAL_MCAST(oldflp->fl4_dst) || oldflp->fl4_dst == 0xFFFFFFFF) {
2458 if (!fl.fl4_src)
2459 fl.fl4_src = inet_select_addr(dev_out, 0,
2460 RT_SCOPE_LINK);
2461 goto make_route;
2462 }
2463 if (!fl.fl4_src) {
2464 if (MULTICAST(oldflp->fl4_dst))
2465 fl.fl4_src = inet_select_addr(dev_out, 0,
2466 fl.fl4_scope);
2467 else if (!oldflp->fl4_dst)
2468 fl.fl4_src = inet_select_addr(dev_out, 0,
2469 RT_SCOPE_HOST);
2470 }
2471 }
2472
2473 if (!fl.fl4_dst) {
2474 fl.fl4_dst = fl.fl4_src;
2475 if (!fl.fl4_dst)
2476 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2477 if (dev_out)
2478 dev_put(dev_out);
2479 dev_out = &loopback_dev;
2480 dev_hold(dev_out);
2481 fl.oif = loopback_dev.ifindex;
2482 res.type = RTN_LOCAL;
2483 flags |= RTCF_LOCAL;
2484 goto make_route;
2485 }
2486
2487 if (fib_lookup(&fl, &res)) {
2488 res.fi = NULL;
2489 if (oldflp->oif) {
2490 /* Apparently, routing tables are wrong. Assume,
2491 that the destination is on link.
2492
2493 WHY? DW.
2494 Because we are allowed to send to iface
2495 even if it has NO routes and NO assigned
2496 addresses. When oif is specified, routing
2497 tables are looked up with only one purpose:
2498 to catch if destination is gatewayed, rather than
2499 direct. Moreover, if MSG_DONTROUTE is set,
2500 we send packet, ignoring both routing tables
2501 and ifaddr state. --ANK
2502
2503
2504 We could make it even if oif is unknown,
2505 likely IPv6, but we do not.
2506 */
2507
2508 if (fl.fl4_src == 0)
2509 fl.fl4_src = inet_select_addr(dev_out, 0,
2510 RT_SCOPE_LINK);
2511 res.type = RTN_UNICAST;
2512 goto make_route;
2513 }
2514 if (dev_out)
2515 dev_put(dev_out);
2516 err = -ENETUNREACH;
2517 goto out;
2518 }
2519 free_res = 1;
2520
2521 if (res.type == RTN_LOCAL) {
2522 if (!fl.fl4_src)
2523 fl.fl4_src = fl.fl4_dst;
2524 if (dev_out)
2525 dev_put(dev_out);
2526 dev_out = &loopback_dev;
2527 dev_hold(dev_out);
2528 fl.oif = dev_out->ifindex;
2529 if (res.fi)
2530 fib_info_put(res.fi);
2531 res.fi = NULL;
2532 flags |= RTCF_LOCAL;
2533 goto make_route;
2534 }
2535
2536 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2537 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2538 fib_select_multipath(&fl, &res);
2539 else
2540 #endif
2541 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2542 fib_select_default(&fl, &res);
2543
2544 if (!fl.fl4_src)
2545 fl.fl4_src = FIB_RES_PREFSRC(res);
2546
2547 if (dev_out)
2548 dev_put(dev_out);
2549 dev_out = FIB_RES_DEV(res);
2550 dev_hold(dev_out);
2551 fl.oif = dev_out->ifindex;
2552
2553
2554 make_route:
2555 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2556
2557
2558 if (free_res)
2559 fib_res_put(&res);
2560 if (dev_out)
2561 dev_put(dev_out);
2562 out: return err;
2563 }
2564
2565 int __ip_route_output_key(struct rtable **rp, const struct flowi *flp)
2566 {
2567 unsigned hash;
2568 struct rtable *rth;
2569
2570 hash = rt_hash_code(flp->fl4_dst, flp->fl4_src ^ (flp->oif << 5));
2571
2572 rcu_read_lock_bh();
2573 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2574 rth = rcu_dereference(rth->u.rt_next)) {
2575 if (rth->fl.fl4_dst == flp->fl4_dst &&
2576 rth->fl.fl4_src == flp->fl4_src &&
2577 rth->fl.iif == 0 &&
2578 rth->fl.oif == flp->oif &&
2579 #ifdef CONFIG_IP_ROUTE_FWMARK
2580 rth->fl.fl4_fwmark == flp->fl4_fwmark &&
2581 #endif
2582 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2583 (IPTOS_RT_MASK | RTO_ONLINK))) {
2584
2585 /* check for multipath routes and choose one if
2586 * necessary
2587 */
2588 if (multipath_select_route(flp, rth, rp)) {
2589 dst_hold(&(*rp)->u.dst);
2590 RT_CACHE_STAT_INC(out_hit);
2591 rcu_read_unlock_bh();
2592 return 0;
2593 }
2594
2595 rth->u.dst.lastuse = jiffies;
2596 dst_hold(&rth->u.dst);
2597 rth->u.dst.__use++;
2598 RT_CACHE_STAT_INC(out_hit);
2599 rcu_read_unlock_bh();
2600 *rp = rth;
2601 return 0;
2602 }
2603 RT_CACHE_STAT_INC(out_hlist_search);
2604 }
2605 rcu_read_unlock_bh();
2606
2607 return ip_route_output_slow(rp, flp);
2608 }
2609
2610 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2611
2612 int ip_route_output_flow(struct rtable **rp, struct flowi *flp, struct sock *sk, int flags)
2613 {
2614 int err;
2615
2616 if ((err = __ip_route_output_key(rp, flp)) != 0)
2617 return err;
2618
2619 if (flp->proto) {
2620 if (!flp->fl4_src)
2621 flp->fl4_src = (*rp)->rt_src;
2622 if (!flp->fl4_dst)
2623 flp->fl4_dst = (*rp)->rt_dst;
2624 return xfrm_lookup((struct dst_entry **)rp, flp, sk, flags);
2625 }
2626
2627 return 0;
2628 }
2629
2630 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2631
2632 int ip_route_output_key(struct rtable **rp, struct flowi *flp)
2633 {
2634 return ip_route_output_flow(rp, flp, NULL, 0);
2635 }
2636
2637 static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
2638 int nowait, unsigned int flags)
2639 {
2640 struct rtable *rt = (struct rtable*)skb->dst;
2641 struct rtmsg *r;
2642 struct nlmsghdr *nlh;
2643 struct rta_cacheinfo ci;
2644
2645 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2646 if (nlh == NULL)
2647 return -ENOBUFS;
2648
2649 r = nlmsg_data(nlh);
2650 r->rtm_family = AF_INET;
2651 r->rtm_dst_len = 32;
2652 r->rtm_src_len = 0;
2653 r->rtm_tos = rt->fl.fl4_tos;
2654 r->rtm_table = RT_TABLE_MAIN;
2655 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2656 r->rtm_type = rt->rt_type;
2657 r->rtm_scope = RT_SCOPE_UNIVERSE;
2658 r->rtm_protocol = RTPROT_UNSPEC;
2659 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2660 if (rt->rt_flags & RTCF_NOTIFY)
2661 r->rtm_flags |= RTM_F_NOTIFY;
2662
2663 NLA_PUT_U32(skb, RTA_DST, rt->rt_dst);
2664
2665 if (rt->fl.fl4_src) {
2666 r->rtm_src_len = 32;
2667 NLA_PUT_U32(skb, RTA_SRC, rt->fl.fl4_src);
2668 }
2669 if (rt->u.dst.dev)
2670 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2671 #ifdef CONFIG_NET_CLS_ROUTE
2672 if (rt->u.dst.tclassid)
2673 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2674 #endif
2675 #ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
2676 if (rt->rt_multipath_alg != IP_MP_ALG_NONE)
2677 NLA_PUT_U32(skb, RTA_MP_ALGO, rt->rt_multipath_alg);
2678 #endif
2679 if (rt->fl.iif)
2680 NLA_PUT_U32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2681 else if (rt->rt_src != rt->fl.fl4_src)
2682 NLA_PUT_U32(skb, RTA_PREFSRC, rt->rt_src);
2683
2684 if (rt->rt_dst != rt->rt_gateway)
2685 NLA_PUT_U32(skb, RTA_GATEWAY, rt->rt_gateway);
2686
2687 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2688 goto nla_put_failure;
2689
2690 ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
2691 ci.rta_used = rt->u.dst.__use;
2692 ci.rta_clntref = atomic_read(&rt->u.dst.__refcnt);
2693 if (rt->u.dst.expires)
2694 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
2695 else
2696 ci.rta_expires = 0;
2697 ci.rta_error = rt->u.dst.error;
2698 ci.rta_id = ci.rta_ts = ci.rta_tsage = 0;
2699 if (rt->peer) {
2700 ci.rta_id = rt->peer->ip_id_count;
2701 if (rt->peer->tcp_ts_stamp) {
2702 ci.rta_ts = rt->peer->tcp_ts;
2703 ci.rta_tsage = xtime.tv_sec - rt->peer->tcp_ts_stamp;
2704 }
2705 }
2706
2707 if (rt->fl.iif) {
2708 #ifdef CONFIG_IP_MROUTE
2709 u32 dst = rt->rt_dst;
2710
2711 if (MULTICAST(dst) && !LOCAL_MCAST(dst) &&
2712 ipv4_devconf.mc_forwarding) {
2713 int err = ipmr_get_route(skb, r, nowait);
2714 if (err <= 0) {
2715 if (!nowait) {
2716 if (err == 0)
2717 return 0;
2718 goto nla_put_failure;
2719 } else {
2720 if (err == -EMSGSIZE)
2721 goto nla_put_failure;
2722 ci.rta_error = err;
2723 }
2724 }
2725 } else
2726 #endif
2727 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2728 }
2729
2730 NLA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
2731
2732 return nlmsg_end(skb, nlh);
2733
2734 nla_put_failure:
2735 return nlmsg_cancel(skb, nlh);
2736 }
2737
2738 int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2739 {
2740 struct rtmsg *rtm;
2741 struct nlattr *tb[RTA_MAX+1];
2742 struct rtable *rt = NULL;
2743 u32 dst, src, iif;
2744 int err;
2745 struct sk_buff *skb;
2746
2747 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2748 if (err < 0)
2749 goto errout;
2750
2751 rtm = nlmsg_data(nlh);
2752
2753 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2754 if (skb == NULL) {
2755 err = -ENOBUFS;
2756 goto errout;
2757 }
2758
2759 /* Reserve room for dummy headers, this skb can pass
2760 through good chunk of routing engine.
2761 */
2762 skb->mac.raw = skb->nh.raw = skb->data;
2763
2764 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2765 skb->nh.iph->protocol = IPPROTO_ICMP;
2766 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2767
2768 src = tb[RTA_SRC] ? nla_get_u32(tb[RTA_SRC]) : 0;
2769 dst = tb[RTA_DST] ? nla_get_u32(tb[RTA_DST]) : 0;
2770 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2771
2772 if (iif) {
2773 struct net_device *dev;
2774
2775 dev = __dev_get_by_index(iif);
2776 if (dev == NULL) {
2777 err = -ENODEV;
2778 goto errout_free;
2779 }
2780
2781 skb->protocol = htons(ETH_P_IP);
2782 skb->dev = dev;
2783 local_bh_disable();
2784 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2785 local_bh_enable();
2786
2787 rt = (struct rtable*) skb->dst;
2788 if (err == 0 && rt->u.dst.error)
2789 err = -rt->u.dst.error;
2790 } else {
2791 struct flowi fl = {
2792 .nl_u = {
2793 .ip4_u = {
2794 .daddr = dst,
2795 .saddr = src,
2796 .tos = rtm->rtm_tos,
2797 },
2798 },
2799 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2800 };
2801 err = ip_route_output_key(&rt, &fl);
2802 }
2803
2804 if (err)
2805 goto errout_free;
2806
2807 skb->dst = &rt->u.dst;
2808 if (rtm->rtm_flags & RTM_F_NOTIFY)
2809 rt->rt_flags |= RTCF_NOTIFY;
2810
2811 err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2812 RTM_NEWROUTE, 0, 0);
2813 if (err <= 0)
2814 goto errout_free;
2815
2816 err = rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
2817 errout:
2818 return err;
2819
2820 errout_free:
2821 kfree_skb(skb);
2822 goto errout;
2823 }
2824
2825 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2826 {
2827 struct rtable *rt;
2828 int h, s_h;
2829 int idx, s_idx;
2830
2831 s_h = cb->args[0];
2832 s_idx = idx = cb->args[1];
2833 for (h = 0; h <= rt_hash_mask; h++) {
2834 if (h < s_h) continue;
2835 if (h > s_h)
2836 s_idx = 0;
2837 rcu_read_lock_bh();
2838 for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt;
2839 rt = rcu_dereference(rt->u.rt_next), idx++) {
2840 if (idx < s_idx)
2841 continue;
2842 skb->dst = dst_clone(&rt->u.dst);
2843 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
2844 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2845 1, NLM_F_MULTI) <= 0) {
2846 dst_release(xchg(&skb->dst, NULL));
2847 rcu_read_unlock_bh();
2848 goto done;
2849 }
2850 dst_release(xchg(&skb->dst, NULL));
2851 }
2852 rcu_read_unlock_bh();
2853 }
2854
2855 done:
2856 cb->args[0] = h;
2857 cb->args[1] = idx;
2858 return skb->len;
2859 }
2860
2861 void ip_rt_multicast_event(struct in_device *in_dev)
2862 {
2863 rt_cache_flush(0);
2864 }
2865
2866 #ifdef CONFIG_SYSCTL
2867 static int flush_delay;
2868
2869 static int ipv4_sysctl_rtcache_flush(ctl_table *ctl, int write,
2870 struct file *filp, void __user *buffer,
2871 size_t *lenp, loff_t *ppos)
2872 {
2873 if (write) {
2874 proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
2875 rt_cache_flush(flush_delay);
2876 return 0;
2877 }
2878
2879 return -EINVAL;
2880 }
2881
2882 static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table,
2883 int __user *name,
2884 int nlen,
2885 void __user *oldval,
2886 size_t __user *oldlenp,
2887 void __user *newval,
2888 size_t newlen,
2889 void **context)
2890 {
2891 int delay;
2892 if (newlen != sizeof(int))
2893 return -EINVAL;
2894 if (get_user(delay, (int __user *)newval))
2895 return -EFAULT;
2896 rt_cache_flush(delay);
2897 return 0;
2898 }
2899
2900 ctl_table ipv4_route_table[] = {
2901 {
2902 .ctl_name = NET_IPV4_ROUTE_FLUSH,
2903 .procname = "flush",
2904 .data = &flush_delay,
2905 .maxlen = sizeof(int),
2906 .mode = 0200,
2907 .proc_handler = &ipv4_sysctl_rtcache_flush,
2908 .strategy = &ipv4_sysctl_rtcache_flush_strategy,
2909 },
2910 {
2911 .ctl_name = NET_IPV4_ROUTE_MIN_DELAY,
2912 .procname = "min_delay",
2913 .data = &ip_rt_min_delay,
2914 .maxlen = sizeof(int),
2915 .mode = 0644,
2916 .proc_handler = &proc_dointvec_jiffies,
2917 .strategy = &sysctl_jiffies,
2918 },
2919 {
2920 .ctl_name = NET_IPV4_ROUTE_MAX_DELAY,
2921 .procname = "max_delay",
2922 .data = &ip_rt_max_delay,
2923 .maxlen = sizeof(int),
2924 .mode = 0644,
2925 .proc_handler = &proc_dointvec_jiffies,
2926 .strategy = &sysctl_jiffies,
2927 },
2928 {
2929 .ctl_name = NET_IPV4_ROUTE_GC_THRESH,
2930 .procname = "gc_thresh",
2931 .data = &ipv4_dst_ops.gc_thresh,
2932 .maxlen = sizeof(int),
2933 .mode = 0644,
2934 .proc_handler = &proc_dointvec,
2935 },
2936 {
2937 .ctl_name = NET_IPV4_ROUTE_MAX_SIZE,
2938 .procname = "max_size",
2939 .data = &ip_rt_max_size,
2940 .maxlen = sizeof(int),
2941 .mode = 0644,
2942 .proc_handler = &proc_dointvec,
2943 },
2944 {
2945 /* Deprecated. Use gc_min_interval_ms */
2946
2947 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL,
2948 .procname = "gc_min_interval",
2949 .data = &ip_rt_gc_min_interval,
2950 .maxlen = sizeof(int),
2951 .mode = 0644,
2952 .proc_handler = &proc_dointvec_jiffies,
2953 .strategy = &sysctl_jiffies,
2954 },
2955 {
2956 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS,
2957 .procname = "gc_min_interval_ms",
2958 .data = &ip_rt_gc_min_interval,
2959 .maxlen = sizeof(int),
2960 .mode = 0644,
2961 .proc_handler = &proc_dointvec_ms_jiffies,
2962 .strategy = &sysctl_ms_jiffies,
2963 },
2964 {
2965 .ctl_name = NET_IPV4_ROUTE_GC_TIMEOUT,
2966 .procname = "gc_timeout",
2967 .data = &ip_rt_gc_timeout,
2968 .maxlen = sizeof(int),
2969 .mode = 0644,
2970 .proc_handler = &proc_dointvec_jiffies,
2971 .strategy = &sysctl_jiffies,
2972 },
2973 {
2974 .ctl_name = NET_IPV4_ROUTE_GC_INTERVAL,
2975 .procname = "gc_interval",
2976 .data = &ip_rt_gc_interval,
2977 .maxlen = sizeof(int),
2978 .mode = 0644,
2979 .proc_handler = &proc_dointvec_jiffies,
2980 .strategy = &sysctl_jiffies,
2981 },
2982 {
2983 .ctl_name = NET_IPV4_ROUTE_REDIRECT_LOAD,
2984 .procname = "redirect_load",
2985 .data = &ip_rt_redirect_load,
2986 .maxlen = sizeof(int),
2987 .mode = 0644,
2988 .proc_handler = &proc_dointvec,
2989 },
2990 {
2991 .ctl_name = NET_IPV4_ROUTE_REDIRECT_NUMBER,
2992 .procname = "redirect_number",
2993 .data = &ip_rt_redirect_number,
2994 .maxlen = sizeof(int),
2995 .mode = 0644,
2996 .proc_handler = &proc_dointvec,
2997 },
2998 {
2999 .ctl_name = NET_IPV4_ROUTE_REDIRECT_SILENCE,
3000 .procname = "redirect_silence",
3001 .data = &ip_rt_redirect_silence,
3002 .maxlen = sizeof(int),
3003 .mode = 0644,
3004 .proc_handler = &proc_dointvec,
3005 },
3006 {
3007 .ctl_name = NET_IPV4_ROUTE_ERROR_COST,
3008 .procname = "error_cost",
3009 .data = &ip_rt_error_cost,
3010 .maxlen = sizeof(int),
3011 .mode = 0644,
3012 .proc_handler = &proc_dointvec,
3013 },
3014 {
3015 .ctl_name = NET_IPV4_ROUTE_ERROR_BURST,
3016 .procname = "error_burst",
3017 .data = &ip_rt_error_burst,
3018 .maxlen = sizeof(int),
3019 .mode = 0644,
3020 .proc_handler = &proc_dointvec,
3021 },
3022 {
3023 .ctl_name = NET_IPV4_ROUTE_GC_ELASTICITY,
3024 .procname = "gc_elasticity",
3025 .data = &ip_rt_gc_elasticity,
3026 .maxlen = sizeof(int),
3027 .mode = 0644,
3028 .proc_handler = &proc_dointvec,
3029 },
3030 {
3031 .ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES,
3032 .procname = "mtu_expires",
3033 .data = &ip_rt_mtu_expires,
3034 .maxlen = sizeof(int),
3035 .mode = 0644,
3036 .proc_handler = &proc_dointvec_jiffies,
3037 .strategy = &sysctl_jiffies,
3038 },
3039 {
3040 .ctl_name = NET_IPV4_ROUTE_MIN_PMTU,
3041 .procname = "min_pmtu",
3042 .data = &ip_rt_min_pmtu,
3043 .maxlen = sizeof(int),
3044 .mode = 0644,
3045 .proc_handler = &proc_dointvec,
3046 },
3047 {
3048 .ctl_name = NET_IPV4_ROUTE_MIN_ADVMSS,
3049 .procname = "min_adv_mss",
3050 .data = &ip_rt_min_advmss,
3051 .maxlen = sizeof(int),
3052 .mode = 0644,
3053 .proc_handler = &proc_dointvec,
3054 },
3055 {
3056 .ctl_name = NET_IPV4_ROUTE_SECRET_INTERVAL,
3057 .procname = "secret_interval",
3058 .data = &ip_rt_secret_interval,
3059 .maxlen = sizeof(int),
3060 .mode = 0644,
3061 .proc_handler = &proc_dointvec_jiffies,
3062 .strategy = &sysctl_jiffies,
3063 },
3064 { .ctl_name = 0 }
3065 };
3066 #endif
3067
3068 #ifdef CONFIG_NET_CLS_ROUTE
3069 struct ip_rt_acct *ip_rt_acct;
3070
3071 /* This code sucks. But you should have seen it before! --RR */
3072
3073 /* IP route accounting ptr for this logical cpu number. */
3074 #define IP_RT_ACCT_CPU(i) (ip_rt_acct + i * 256)
3075
3076 #ifdef CONFIG_PROC_FS
3077 static int ip_rt_acct_read(char *buffer, char **start, off_t offset,
3078 int length, int *eof, void *data)
3079 {
3080 unsigned int i;
3081
3082 if ((offset & 3) || (length & 3))
3083 return -EIO;
3084
3085 if (offset >= sizeof(struct ip_rt_acct) * 256) {
3086 *eof = 1;
3087 return 0;
3088 }
3089
3090 if (offset + length >= sizeof(struct ip_rt_acct) * 256) {
3091 length = sizeof(struct ip_rt_acct) * 256 - offset;
3092 *eof = 1;
3093 }
3094
3095 offset /= sizeof(u32);
3096
3097 if (length > 0) {
3098 u32 *src = ((u32 *) IP_RT_ACCT_CPU(0)) + offset;
3099 u32 *dst = (u32 *) buffer;
3100
3101 /* Copy first cpu. */
3102 *start = buffer;
3103 memcpy(dst, src, length);
3104
3105 /* Add the other cpus in, one int at a time */
3106 for_each_possible_cpu(i) {
3107 unsigned int j;
3108
3109 src = ((u32 *) IP_RT_ACCT_CPU(i)) + offset;
3110
3111 for (j = 0; j < length/4; j++)
3112 dst[j] += src[j];
3113 }
3114 }
3115 return length;
3116 }
3117 #endif /* CONFIG_PROC_FS */
3118 #endif /* CONFIG_NET_CLS_ROUTE */
3119
3120 static __initdata unsigned long rhash_entries;
3121 static int __init set_rhash_entries(char *str)
3122 {
3123 if (!str)
3124 return 0;
3125 rhash_entries = simple_strtoul(str, &str, 0);
3126 return 1;
3127 }
3128 __setup("rhash_entries=", set_rhash_entries);
3129
3130 int __init ip_rt_init(void)
3131 {
3132 int rc = 0;
3133
3134 rt_hash_rnd = (int) ((num_physpages ^ (num_physpages>>8)) ^
3135 (jiffies ^ (jiffies >> 7)));
3136
3137 #ifdef CONFIG_NET_CLS_ROUTE
3138 {
3139 int order;
3140 for (order = 0;
3141 (PAGE_SIZE << order) < 256 * sizeof(struct ip_rt_acct) * NR_CPUS; order++)
3142 /* NOTHING */;
3143 ip_rt_acct = (struct ip_rt_acct *)__get_free_pages(GFP_KERNEL, order);
3144 if (!ip_rt_acct)
3145 panic("IP: failed to allocate ip_rt_acct\n");
3146 memset(ip_rt_acct, 0, PAGE_SIZE << order);
3147 }
3148 #endif
3149
3150 ipv4_dst_ops.kmem_cachep = kmem_cache_create("ip_dst_cache",
3151 sizeof(struct rtable),
3152 0, SLAB_HWCACHE_ALIGN,
3153 NULL, NULL);
3154
3155 if (!ipv4_dst_ops.kmem_cachep)
3156 panic("IP: failed to allocate ip_dst_cache\n");
3157
3158 rt_hash_table = (struct rt_hash_bucket *)
3159 alloc_large_system_hash("IP route cache",
3160 sizeof(struct rt_hash_bucket),
3161 rhash_entries,
3162 (num_physpages >= 128 * 1024) ?
3163 15 : 17,
3164 0,
3165 &rt_hash_log,
3166 &rt_hash_mask,
3167 0);
3168 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3169 rt_hash_lock_init();
3170
3171 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3172 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3173
3174 devinet_init();
3175 ip_fib_init();
3176
3177 init_timer(&rt_flush_timer);
3178 rt_flush_timer.function = rt_run_flush;
3179 init_timer(&rt_periodic_timer);
3180 rt_periodic_timer.function = rt_check_expire;
3181 init_timer(&rt_secret_timer);
3182 rt_secret_timer.function = rt_secret_rebuild;
3183
3184 /* All the timers, started at system startup tend
3185 to synchronize. Perturb it a bit.
3186 */
3187 rt_periodic_timer.expires = jiffies + net_random() % ip_rt_gc_interval +
3188 ip_rt_gc_interval;
3189 add_timer(&rt_periodic_timer);
3190
3191 rt_secret_timer.expires = jiffies + net_random() % ip_rt_secret_interval +
3192 ip_rt_secret_interval;
3193 add_timer(&rt_secret_timer);
3194
3195 #ifdef CONFIG_PROC_FS
3196 {
3197 struct proc_dir_entry *rtstat_pde = NULL; /* keep gcc happy */
3198 if (!proc_net_fops_create("rt_cache", S_IRUGO, &rt_cache_seq_fops) ||
3199 !(rtstat_pde = create_proc_entry("rt_cache", S_IRUGO,
3200 proc_net_stat))) {
3201 return -ENOMEM;
3202 }
3203 rtstat_pde->proc_fops = &rt_cpu_seq_fops;
3204 }
3205 #ifdef CONFIG_NET_CLS_ROUTE
3206 create_proc_read_entry("rt_acct", 0, proc_net, ip_rt_acct_read, NULL);
3207 #endif
3208 #endif
3209 #ifdef CONFIG_XFRM
3210 xfrm_init();
3211 xfrm4_init();
3212 #endif
3213 return rc;
3214 }
3215
3216 EXPORT_SYMBOL(__ip_select_ident);
3217 EXPORT_SYMBOL(ip_route_input);
3218 EXPORT_SYMBOL(ip_route_output_key);
hh.org GIT migration