]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
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 | |
e905a9ed | 23 | * Alan Cox : Super /proc >4K |
1da177e4 LT |
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. | |
e905a9ed | 41 | * |
1da177e4 LT |
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 | |
bb1d23b0 | 57 | * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes. |
cef2685e IS |
58 | * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect |
59 | * Ilia Sotnikov : Removed TOS from hash calculations | |
1da177e4 LT |
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 | ||
1da177e4 LT |
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> | |
1da177e4 | 73 | #include <linux/mm.h> |
424c4b70 | 74 | #include <linux/bootmem.h> |
1da177e4 LT |
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> | |
39c90ece | 84 | #include <linux/workqueue.h> |
1da177e4 | 85 | #include <linux/skbuff.h> |
1da177e4 LT |
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> | |
352e512c | 95 | #include <net/dst.h> |
457c4cbc | 96 | #include <net/net_namespace.h> |
1da177e4 LT |
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> | |
8d71740c | 107 | #include <net/netevent.h> |
63f3444f | 108 | #include <net/rtnetlink.h> |
1da177e4 LT |
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; | |
beb659bd | 136 | static int ip_rt_flush_expected; |
1da177e4 LT |
137 | static unsigned long rt_deadline; |
138 | ||
139 | #define RTprint(a...) printk(KERN_DEBUG a) | |
140 | ||
141 | static struct timer_list rt_flush_timer; | |
beb659bd ED |
142 | static void rt_worker_func(struct work_struct *work); |
143 | static DECLARE_DELAYED_WORK(expires_work, rt_worker_func); | |
1da177e4 LT |
144 | static struct timer_list rt_secret_timer; |
145 | ||
146 | /* | |
147 | * Interface to generic destination cache. | |
148 | */ | |
149 | ||
150 | static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); | |
151 | static void ipv4_dst_destroy(struct dst_entry *dst); | |
152 | static void ipv4_dst_ifdown(struct dst_entry *dst, | |
153 | struct net_device *dev, int how); | |
154 | static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst); | |
155 | static void ipv4_link_failure(struct sk_buff *skb); | |
156 | static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu); | |
157 | static int rt_garbage_collect(void); | |
158 | ||
159 | ||
160 | static struct dst_ops ipv4_dst_ops = { | |
161 | .family = AF_INET, | |
162 | .protocol = __constant_htons(ETH_P_IP), | |
163 | .gc = rt_garbage_collect, | |
164 | .check = ipv4_dst_check, | |
165 | .destroy = ipv4_dst_destroy, | |
166 | .ifdown = ipv4_dst_ifdown, | |
167 | .negative_advice = ipv4_negative_advice, | |
168 | .link_failure = ipv4_link_failure, | |
169 | .update_pmtu = ip_rt_update_pmtu, | |
862b82c6 | 170 | .local_out = ip_local_out, |
1da177e4 LT |
171 | .entry_size = sizeof(struct rtable), |
172 | }; | |
173 | ||
174 | #define ECN_OR_COST(class) TC_PRIO_##class | |
175 | ||
4839c52b | 176 | const __u8 ip_tos2prio[16] = { |
1da177e4 LT |
177 | TC_PRIO_BESTEFFORT, |
178 | ECN_OR_COST(FILLER), | |
179 | TC_PRIO_BESTEFFORT, | |
180 | ECN_OR_COST(BESTEFFORT), | |
181 | TC_PRIO_BULK, | |
182 | ECN_OR_COST(BULK), | |
183 | TC_PRIO_BULK, | |
184 | ECN_OR_COST(BULK), | |
185 | TC_PRIO_INTERACTIVE, | |
186 | ECN_OR_COST(INTERACTIVE), | |
187 | TC_PRIO_INTERACTIVE, | |
188 | ECN_OR_COST(INTERACTIVE), | |
189 | TC_PRIO_INTERACTIVE_BULK, | |
190 | ECN_OR_COST(INTERACTIVE_BULK), | |
191 | TC_PRIO_INTERACTIVE_BULK, | |
192 | ECN_OR_COST(INTERACTIVE_BULK) | |
193 | }; | |
194 | ||
195 | ||
196 | /* | |
197 | * Route cache. | |
198 | */ | |
199 | ||
200 | /* The locking scheme is rather straight forward: | |
201 | * | |
202 | * 1) Read-Copy Update protects the buckets of the central route hash. | |
203 | * 2) Only writers remove entries, and they hold the lock | |
204 | * as they look at rtable reference counts. | |
205 | * 3) Only readers acquire references to rtable entries, | |
206 | * they do so with atomic increments and with the | |
207 | * lock held. | |
208 | */ | |
209 | ||
210 | struct rt_hash_bucket { | |
211 | struct rtable *chain; | |
22c047cc | 212 | }; |
8a25d5de IM |
213 | #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \ |
214 | defined(CONFIG_PROVE_LOCKING) | |
22c047cc ED |
215 | /* |
216 | * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks | |
217 | * The size of this table is a power of two and depends on the number of CPUS. | |
62051200 | 218 | * (on lockdep we have a quite big spinlock_t, so keep the size down there) |
22c047cc | 219 | */ |
62051200 IM |
220 | #ifdef CONFIG_LOCKDEP |
221 | # define RT_HASH_LOCK_SZ 256 | |
22c047cc | 222 | #else |
62051200 IM |
223 | # if NR_CPUS >= 32 |
224 | # define RT_HASH_LOCK_SZ 4096 | |
225 | # elif NR_CPUS >= 16 | |
226 | # define RT_HASH_LOCK_SZ 2048 | |
227 | # elif NR_CPUS >= 8 | |
228 | # define RT_HASH_LOCK_SZ 1024 | |
229 | # elif NR_CPUS >= 4 | |
230 | # define RT_HASH_LOCK_SZ 512 | |
231 | # else | |
232 | # define RT_HASH_LOCK_SZ 256 | |
233 | # endif | |
22c047cc ED |
234 | #endif |
235 | ||
236 | static spinlock_t *rt_hash_locks; | |
237 | # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)] | |
238 | # define rt_hash_lock_init() { \ | |
239 | int i; \ | |
240 | rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ, GFP_KERNEL); \ | |
241 | if (!rt_hash_locks) panic("IP: failed to allocate rt_hash_locks\n"); \ | |
242 | for (i = 0; i < RT_HASH_LOCK_SZ; i++) \ | |
243 | spin_lock_init(&rt_hash_locks[i]); \ | |
244 | } | |
245 | #else | |
246 | # define rt_hash_lock_addr(slot) NULL | |
247 | # define rt_hash_lock_init() | |
248 | #endif | |
1da177e4 LT |
249 | |
250 | static struct rt_hash_bucket *rt_hash_table; | |
251 | static unsigned rt_hash_mask; | |
cfcabdcc | 252 | static unsigned int rt_hash_log; |
1da177e4 LT |
253 | static unsigned int rt_hash_rnd; |
254 | ||
2f970d83 | 255 | static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); |
dbd2915c | 256 | #define RT_CACHE_STAT_INC(field) \ |
bfe5d834 | 257 | (__raw_get_cpu_var(rt_cache_stat).field++) |
1da177e4 LT |
258 | |
259 | static int rt_intern_hash(unsigned hash, struct rtable *rth, | |
260 | struct rtable **res); | |
261 | ||
cef2685e | 262 | static unsigned int rt_hash_code(u32 daddr, u32 saddr) |
1da177e4 | 263 | { |
cef2685e | 264 | return (jhash_2words(daddr, saddr, rt_hash_rnd) |
1da177e4 LT |
265 | & rt_hash_mask); |
266 | } | |
267 | ||
8c7bc840 AV |
268 | #define rt_hash(daddr, saddr, idx) \ |
269 | rt_hash_code((__force u32)(__be32)(daddr),\ | |
270 | (__force u32)(__be32)(saddr) ^ ((idx) << 5)) | |
271 | ||
1da177e4 LT |
272 | #ifdef CONFIG_PROC_FS |
273 | struct rt_cache_iter_state { | |
274 | int bucket; | |
275 | }; | |
276 | ||
277 | static struct rtable *rt_cache_get_first(struct seq_file *seq) | |
278 | { | |
279 | struct rtable *r = NULL; | |
280 | struct rt_cache_iter_state *st = seq->private; | |
281 | ||
282 | for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) { | |
283 | rcu_read_lock_bh(); | |
284 | r = rt_hash_table[st->bucket].chain; | |
285 | if (r) | |
286 | break; | |
287 | rcu_read_unlock_bh(); | |
288 | } | |
0bcceadc | 289 | return rcu_dereference(r); |
1da177e4 LT |
290 | } |
291 | ||
292 | static struct rtable *rt_cache_get_next(struct seq_file *seq, struct rtable *r) | |
293 | { | |
0bcceadc | 294 | struct rt_cache_iter_state *st = seq->private; |
1da177e4 | 295 | |
093c2ca4 | 296 | r = r->u.dst.rt_next; |
1da177e4 LT |
297 | while (!r) { |
298 | rcu_read_unlock_bh(); | |
299 | if (--st->bucket < 0) | |
300 | break; | |
301 | rcu_read_lock_bh(); | |
302 | r = rt_hash_table[st->bucket].chain; | |
303 | } | |
0bcceadc | 304 | return rcu_dereference(r); |
1da177e4 LT |
305 | } |
306 | ||
307 | static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos) | |
308 | { | |
309 | struct rtable *r = rt_cache_get_first(seq); | |
310 | ||
311 | if (r) | |
312 | while (pos && (r = rt_cache_get_next(seq, r))) | |
313 | --pos; | |
314 | return pos ? NULL : r; | |
315 | } | |
316 | ||
317 | static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) | |
318 | { | |
319 | return *pos ? rt_cache_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
320 | } | |
321 | ||
322 | static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
323 | { | |
324 | struct rtable *r = NULL; | |
325 | ||
326 | if (v == SEQ_START_TOKEN) | |
327 | r = rt_cache_get_first(seq); | |
328 | else | |
329 | r = rt_cache_get_next(seq, v); | |
330 | ++*pos; | |
331 | return r; | |
332 | } | |
333 | ||
334 | static void rt_cache_seq_stop(struct seq_file *seq, void *v) | |
335 | { | |
336 | if (v && v != SEQ_START_TOKEN) | |
337 | rcu_read_unlock_bh(); | |
338 | } | |
339 | ||
340 | static int rt_cache_seq_show(struct seq_file *seq, void *v) | |
341 | { | |
342 | if (v == SEQ_START_TOKEN) | |
343 | seq_printf(seq, "%-127s\n", | |
344 | "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" | |
345 | "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" | |
346 | "HHUptod\tSpecDst"); | |
347 | else { | |
348 | struct rtable *r = v; | |
349 | char temp[256]; | |
350 | ||
351 | sprintf(temp, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t" | |
352 | "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X", | |
353 | r->u.dst.dev ? r->u.dst.dev->name : "*", | |
354 | (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway, | |
355 | r->rt_flags, atomic_read(&r->u.dst.__refcnt), | |
356 | r->u.dst.__use, 0, (unsigned long)r->rt_src, | |
357 | (dst_metric(&r->u.dst, RTAX_ADVMSS) ? | |
358 | (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0), | |
359 | dst_metric(&r->u.dst, RTAX_WINDOW), | |
360 | (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) + | |
361 | dst_metric(&r->u.dst, RTAX_RTTVAR)), | |
362 | r->fl.fl4_tos, | |
363 | r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1, | |
364 | r->u.dst.hh ? (r->u.dst.hh->hh_output == | |
365 | dev_queue_xmit) : 0, | |
366 | r->rt_spec_dst); | |
367 | seq_printf(seq, "%-127s\n", temp); | |
e905a9ed YH |
368 | } |
369 | return 0; | |
1da177e4 LT |
370 | } |
371 | ||
f690808e | 372 | static const struct seq_operations rt_cache_seq_ops = { |
1da177e4 LT |
373 | .start = rt_cache_seq_start, |
374 | .next = rt_cache_seq_next, | |
375 | .stop = rt_cache_seq_stop, | |
376 | .show = rt_cache_seq_show, | |
377 | }; | |
378 | ||
379 | static int rt_cache_seq_open(struct inode *inode, struct file *file) | |
380 | { | |
cf7732e4 PE |
381 | return seq_open_private(file, &rt_cache_seq_ops, |
382 | sizeof(struct rt_cache_iter_state)); | |
1da177e4 LT |
383 | } |
384 | ||
9a32144e | 385 | static const struct file_operations rt_cache_seq_fops = { |
1da177e4 LT |
386 | .owner = THIS_MODULE, |
387 | .open = rt_cache_seq_open, | |
388 | .read = seq_read, | |
389 | .llseek = seq_lseek, | |
390 | .release = seq_release_private, | |
391 | }; | |
392 | ||
393 | ||
394 | static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) | |
395 | { | |
396 | int cpu; | |
397 | ||
398 | if (*pos == 0) | |
399 | return SEQ_START_TOKEN; | |
400 | ||
401 | for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) { | |
402 | if (!cpu_possible(cpu)) | |
403 | continue; | |
404 | *pos = cpu+1; | |
2f970d83 | 405 | return &per_cpu(rt_cache_stat, cpu); |
1da177e4 LT |
406 | } |
407 | return NULL; | |
408 | } | |
409 | ||
410 | static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
411 | { | |
412 | int cpu; | |
413 | ||
414 | for (cpu = *pos; cpu < NR_CPUS; ++cpu) { | |
415 | if (!cpu_possible(cpu)) | |
416 | continue; | |
417 | *pos = cpu+1; | |
2f970d83 | 418 | return &per_cpu(rt_cache_stat, cpu); |
1da177e4 LT |
419 | } |
420 | return NULL; | |
e905a9ed | 421 | |
1da177e4 LT |
422 | } |
423 | ||
424 | static void rt_cpu_seq_stop(struct seq_file *seq, void *v) | |
425 | { | |
426 | ||
427 | } | |
428 | ||
429 | static int rt_cpu_seq_show(struct seq_file *seq, void *v) | |
430 | { | |
431 | struct rt_cache_stat *st = v; | |
432 | ||
433 | if (v == SEQ_START_TOKEN) { | |
5bec0039 | 434 | 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"); |
1da177e4 LT |
435 | return 0; |
436 | } | |
e905a9ed | 437 | |
1da177e4 LT |
438 | seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x " |
439 | " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n", | |
440 | atomic_read(&ipv4_dst_ops.entries), | |
441 | st->in_hit, | |
442 | st->in_slow_tot, | |
443 | st->in_slow_mc, | |
444 | st->in_no_route, | |
445 | st->in_brd, | |
446 | st->in_martian_dst, | |
447 | st->in_martian_src, | |
448 | ||
449 | st->out_hit, | |
450 | st->out_slow_tot, | |
e905a9ed | 451 | st->out_slow_mc, |
1da177e4 LT |
452 | |
453 | st->gc_total, | |
454 | st->gc_ignored, | |
455 | st->gc_goal_miss, | |
456 | st->gc_dst_overflow, | |
457 | st->in_hlist_search, | |
458 | st->out_hlist_search | |
459 | ); | |
460 | return 0; | |
461 | } | |
462 | ||
f690808e | 463 | static const struct seq_operations rt_cpu_seq_ops = { |
1da177e4 LT |
464 | .start = rt_cpu_seq_start, |
465 | .next = rt_cpu_seq_next, | |
466 | .stop = rt_cpu_seq_stop, | |
467 | .show = rt_cpu_seq_show, | |
468 | }; | |
469 | ||
470 | ||
471 | static int rt_cpu_seq_open(struct inode *inode, struct file *file) | |
472 | { | |
473 | return seq_open(file, &rt_cpu_seq_ops); | |
474 | } | |
475 | ||
9a32144e | 476 | static const struct file_operations rt_cpu_seq_fops = { |
1da177e4 LT |
477 | .owner = THIS_MODULE, |
478 | .open = rt_cpu_seq_open, | |
479 | .read = seq_read, | |
480 | .llseek = seq_lseek, | |
481 | .release = seq_release, | |
482 | }; | |
483 | ||
78c686e9 PE |
484 | #ifdef CONFIG_NET_CLS_ROUTE |
485 | static int ip_rt_acct_read(char *buffer, char **start, off_t offset, | |
486 | int length, int *eof, void *data) | |
487 | { | |
488 | unsigned int i; | |
489 | ||
490 | if ((offset & 3) || (length & 3)) | |
491 | return -EIO; | |
492 | ||
493 | if (offset >= sizeof(struct ip_rt_acct) * 256) { | |
494 | *eof = 1; | |
495 | return 0; | |
496 | } | |
497 | ||
498 | if (offset + length >= sizeof(struct ip_rt_acct) * 256) { | |
499 | length = sizeof(struct ip_rt_acct) * 256 - offset; | |
500 | *eof = 1; | |
501 | } | |
502 | ||
503 | offset /= sizeof(u32); | |
504 | ||
505 | if (length > 0) { | |
506 | u32 *dst = (u32 *) buffer; | |
507 | ||
508 | *start = buffer; | |
509 | memset(dst, 0, length); | |
510 | ||
511 | for_each_possible_cpu(i) { | |
512 | unsigned int j; | |
513 | u32 *src; | |
514 | ||
515 | src = ((u32 *) per_cpu_ptr(ip_rt_acct, i)) + offset; | |
516 | for (j = 0; j < length/4; j++) | |
517 | dst[j] += src[j]; | |
518 | } | |
519 | } | |
520 | return length; | |
521 | } | |
522 | #endif | |
107f1634 PE |
523 | |
524 | static __init int ip_rt_proc_init(struct net *net) | |
525 | { | |
526 | struct proc_dir_entry *pde; | |
527 | ||
528 | pde = proc_net_fops_create(net, "rt_cache", S_IRUGO, | |
529 | &rt_cache_seq_fops); | |
530 | if (!pde) | |
531 | goto err1; | |
532 | ||
533 | pde = create_proc_entry("rt_cache", S_IRUGO, net->proc_net_stat); | |
534 | if (!pde) | |
535 | goto err2; | |
536 | ||
537 | pde->proc_fops = &rt_cpu_seq_fops; | |
538 | ||
539 | #ifdef CONFIG_NET_CLS_ROUTE | |
540 | pde = create_proc_read_entry("rt_acct", 0, net->proc_net, | |
541 | ip_rt_acct_read, NULL); | |
542 | if (!pde) | |
543 | goto err3; | |
544 | #endif | |
545 | return 0; | |
546 | ||
547 | #ifdef CONFIG_NET_CLS_ROUTE | |
548 | err3: | |
549 | remove_proc_entry("rt_cache", net->proc_net_stat); | |
550 | #endif | |
551 | err2: | |
552 | remove_proc_entry("rt_cache", net->proc_net); | |
553 | err1: | |
554 | return -ENOMEM; | |
555 | } | |
556 | #else | |
557 | static inline int ip_rt_proc_init(struct net *net) | |
558 | { | |
559 | return 0; | |
560 | } | |
1da177e4 | 561 | #endif /* CONFIG_PROC_FS */ |
e905a9ed | 562 | |
1da177e4 LT |
563 | static __inline__ void rt_free(struct rtable *rt) |
564 | { | |
1da177e4 LT |
565 | call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free); |
566 | } | |
567 | ||
568 | static __inline__ void rt_drop(struct rtable *rt) | |
569 | { | |
1da177e4 LT |
570 | ip_rt_put(rt); |
571 | call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free); | |
572 | } | |
573 | ||
574 | static __inline__ int rt_fast_clean(struct rtable *rth) | |
575 | { | |
576 | /* Kill broadcast/multicast entries very aggresively, if they | |
577 | collide in hash table with more useful entries */ | |
578 | return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) && | |
093c2ca4 | 579 | rth->fl.iif && rth->u.dst.rt_next; |
1da177e4 LT |
580 | } |
581 | ||
582 | static __inline__ int rt_valuable(struct rtable *rth) | |
583 | { | |
584 | return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) || | |
585 | rth->u.dst.expires; | |
586 | } | |
587 | ||
588 | static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2) | |
589 | { | |
590 | unsigned long age; | |
591 | int ret = 0; | |
592 | ||
593 | if (atomic_read(&rth->u.dst.__refcnt)) | |
594 | goto out; | |
595 | ||
596 | ret = 1; | |
597 | if (rth->u.dst.expires && | |
598 | time_after_eq(jiffies, rth->u.dst.expires)) | |
599 | goto out; | |
600 | ||
601 | age = jiffies - rth->u.dst.lastuse; | |
602 | ret = 0; | |
603 | if ((age <= tmo1 && !rt_fast_clean(rth)) || | |
604 | (age <= tmo2 && rt_valuable(rth))) | |
605 | goto out; | |
606 | ret = 1; | |
607 | out: return ret; | |
608 | } | |
609 | ||
610 | /* Bits of score are: | |
611 | * 31: very valuable | |
612 | * 30: not quite useless | |
613 | * 29..0: usage counter | |
614 | */ | |
615 | static inline u32 rt_score(struct rtable *rt) | |
616 | { | |
617 | u32 score = jiffies - rt->u.dst.lastuse; | |
618 | ||
619 | score = ~score & ~(3<<30); | |
620 | ||
621 | if (rt_valuable(rt)) | |
622 | score |= (1<<31); | |
623 | ||
624 | if (!rt->fl.iif || | |
625 | !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL))) | |
626 | score |= (1<<30); | |
627 | ||
628 | return score; | |
629 | } | |
630 | ||
631 | static inline int compare_keys(struct flowi *fl1, struct flowi *fl2) | |
632 | { | |
714e85be AV |
633 | return ((__force u32)((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) | |
634 | (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr)) | | |
47dcf0cb | 635 | (fl1->mark ^ fl2->mark) | |
8238b218 DM |
636 | (*(u16 *)&fl1->nl_u.ip4_u.tos ^ |
637 | *(u16 *)&fl2->nl_u.ip4_u.tos) | | |
638 | (fl1->oif ^ fl2->oif) | | |
639 | (fl1->iif ^ fl2->iif)) == 0; | |
1da177e4 LT |
640 | } |
641 | ||
beb659bd ED |
642 | /* |
643 | * Perform a full scan of hash table and free all entries. | |
644 | * Can be called by a softirq or a process. | |
645 | * In the later case, we want to be reschedule if necessary | |
646 | */ | |
647 | static void rt_do_flush(int process_context) | |
648 | { | |
649 | unsigned int i; | |
650 | struct rtable *rth, *next; | |
651 | ||
652 | for (i = 0; i <= rt_hash_mask; i++) { | |
653 | if (process_context && need_resched()) | |
654 | cond_resched(); | |
655 | rth = rt_hash_table[i].chain; | |
656 | if (!rth) | |
657 | continue; | |
658 | ||
659 | spin_lock_bh(rt_hash_lock_addr(i)); | |
660 | rth = rt_hash_table[i].chain; | |
661 | rt_hash_table[i].chain = NULL; | |
662 | spin_unlock_bh(rt_hash_lock_addr(i)); | |
663 | ||
664 | for (; rth; rth = next) { | |
665 | next = rth->u.dst.rt_next; | |
666 | rt_free(rth); | |
667 | } | |
668 | } | |
669 | } | |
670 | ||
671 | static void rt_check_expire(void) | |
1da177e4 | 672 | { |
bb1d23b0 ED |
673 | static unsigned int rover; |
674 | unsigned int i = rover, goal; | |
1da177e4 | 675 | struct rtable *rth, **rthp; |
bb1d23b0 ED |
676 | u64 mult; |
677 | ||
678 | mult = ((u64)ip_rt_gc_interval) << rt_hash_log; | |
679 | if (ip_rt_gc_timeout > 1) | |
680 | do_div(mult, ip_rt_gc_timeout); | |
681 | goal = (unsigned int)mult; | |
39c90ece ED |
682 | if (goal > rt_hash_mask) |
683 | goal = rt_hash_mask + 1; | |
bb1d23b0 | 684 | for (; goal > 0; goal--) { |
1da177e4 LT |
685 | unsigned long tmo = ip_rt_gc_timeout; |
686 | ||
687 | i = (i + 1) & rt_hash_mask; | |
688 | rthp = &rt_hash_table[i].chain; | |
689 | ||
d90bf5a9 ED |
690 | if (need_resched()) |
691 | cond_resched(); | |
692 | ||
cfcabdcc | 693 | if (*rthp == NULL) |
bb1d23b0 | 694 | continue; |
39c90ece | 695 | spin_lock_bh(rt_hash_lock_addr(i)); |
1da177e4 LT |
696 | while ((rth = *rthp) != NULL) { |
697 | if (rth->u.dst.expires) { | |
698 | /* Entry is expired even if it is in use */ | |
39c90ece | 699 | if (time_before_eq(jiffies, rth->u.dst.expires)) { |
1da177e4 | 700 | tmo >>= 1; |
093c2ca4 | 701 | rthp = &rth->u.dst.rt_next; |
1da177e4 LT |
702 | continue; |
703 | } | |
704 | } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout)) { | |
705 | tmo >>= 1; | |
093c2ca4 | 706 | rthp = &rth->u.dst.rt_next; |
1da177e4 LT |
707 | continue; |
708 | } | |
709 | ||
710 | /* Cleanup aged off entries. */ | |
093c2ca4 | 711 | *rthp = rth->u.dst.rt_next; |
e905a9ed | 712 | rt_free(rth); |
1da177e4 | 713 | } |
39c90ece | 714 | spin_unlock_bh(rt_hash_lock_addr(i)); |
1da177e4 LT |
715 | } |
716 | rover = i; | |
beb659bd ED |
717 | } |
718 | ||
719 | /* | |
720 | * rt_worker_func() is run in process context. | |
721 | * If a whole flush was scheduled, it is done. | |
722 | * Else, we call rt_check_expire() to scan part of the hash table | |
723 | */ | |
724 | static void rt_worker_func(struct work_struct *work) | |
725 | { | |
726 | if (ip_rt_flush_expected) { | |
727 | ip_rt_flush_expected = 0; | |
728 | rt_do_flush(1); | |
729 | } else | |
730 | rt_check_expire(); | |
39c90ece | 731 | schedule_delayed_work(&expires_work, ip_rt_gc_interval); |
1da177e4 LT |
732 | } |
733 | ||
734 | /* This can run from both BH and non-BH contexts, the latter | |
735 | * in the case of a forced flush event. | |
736 | */ | |
beb659bd | 737 | static void rt_run_flush(unsigned long process_context) |
1da177e4 | 738 | { |
1da177e4 LT |
739 | rt_deadline = 0; |
740 | ||
741 | get_random_bytes(&rt_hash_rnd, 4); | |
742 | ||
beb659bd | 743 | rt_do_flush(process_context); |
1da177e4 LT |
744 | } |
745 | ||
746 | static DEFINE_SPINLOCK(rt_flush_lock); | |
747 | ||
748 | void rt_cache_flush(int delay) | |
749 | { | |
750 | unsigned long now = jiffies; | |
751 | int user_mode = !in_softirq(); | |
752 | ||
753 | if (delay < 0) | |
754 | delay = ip_rt_min_delay; | |
755 | ||
1da177e4 LT |
756 | spin_lock_bh(&rt_flush_lock); |
757 | ||
758 | if (del_timer(&rt_flush_timer) && delay > 0 && rt_deadline) { | |
759 | long tmo = (long)(rt_deadline - now); | |
760 | ||
761 | /* If flush timer is already running | |
762 | and flush request is not immediate (delay > 0): | |
763 | ||
764 | if deadline is not achieved, prolongate timer to "delay", | |
765 | otherwise fire it at deadline time. | |
766 | */ | |
767 | ||
768 | if (user_mode && tmo < ip_rt_max_delay-ip_rt_min_delay) | |
769 | tmo = 0; | |
e905a9ed | 770 | |
1da177e4 LT |
771 | if (delay > tmo) |
772 | delay = tmo; | |
773 | } | |
774 | ||
775 | if (delay <= 0) { | |
776 | spin_unlock_bh(&rt_flush_lock); | |
beb659bd | 777 | rt_run_flush(user_mode); |
1da177e4 LT |
778 | return; |
779 | } | |
780 | ||
781 | if (rt_deadline == 0) | |
782 | rt_deadline = now + ip_rt_max_delay; | |
783 | ||
784 | mod_timer(&rt_flush_timer, now+delay); | |
785 | spin_unlock_bh(&rt_flush_lock); | |
786 | } | |
787 | ||
beb659bd ED |
788 | /* |
789 | * We change rt_hash_rnd and ask next rt_worker_func() invocation | |
790 | * to perform a flush in process context | |
791 | */ | |
1da177e4 LT |
792 | static void rt_secret_rebuild(unsigned long dummy) |
793 | { | |
beb659bd ED |
794 | get_random_bytes(&rt_hash_rnd, 4); |
795 | ip_rt_flush_expected = 1; | |
796 | cancel_delayed_work(&expires_work); | |
797 | schedule_delayed_work(&expires_work, HZ/10); | |
798 | mod_timer(&rt_secret_timer, jiffies + ip_rt_secret_interval); | |
1da177e4 LT |
799 | } |
800 | ||
801 | /* | |
802 | Short description of GC goals. | |
803 | ||
804 | We want to build algorithm, which will keep routing cache | |
805 | at some equilibrium point, when number of aged off entries | |
806 | is kept approximately equal to newly generated ones. | |
807 | ||
808 | Current expiration strength is variable "expire". | |
809 | We try to adjust it dynamically, so that if networking | |
810 | is idle expires is large enough to keep enough of warm entries, | |
811 | and when load increases it reduces to limit cache size. | |
812 | */ | |
813 | ||
814 | static int rt_garbage_collect(void) | |
815 | { | |
816 | static unsigned long expire = RT_GC_TIMEOUT; | |
817 | static unsigned long last_gc; | |
818 | static int rover; | |
819 | static int equilibrium; | |
820 | struct rtable *rth, **rthp; | |
821 | unsigned long now = jiffies; | |
822 | int goal; | |
823 | ||
824 | /* | |
825 | * Garbage collection is pretty expensive, | |
826 | * do not make it too frequently. | |
827 | */ | |
828 | ||
829 | RT_CACHE_STAT_INC(gc_total); | |
830 | ||
831 | if (now - last_gc < ip_rt_gc_min_interval && | |
832 | atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) { | |
833 | RT_CACHE_STAT_INC(gc_ignored); | |
834 | goto out; | |
835 | } | |
836 | ||
837 | /* Calculate number of entries, which we want to expire now. */ | |
838 | goal = atomic_read(&ipv4_dst_ops.entries) - | |
839 | (ip_rt_gc_elasticity << rt_hash_log); | |
840 | if (goal <= 0) { | |
841 | if (equilibrium < ipv4_dst_ops.gc_thresh) | |
842 | equilibrium = ipv4_dst_ops.gc_thresh; | |
843 | goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium; | |
844 | if (goal > 0) { | |
845 | equilibrium += min_t(unsigned int, goal / 2, rt_hash_mask + 1); | |
846 | goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium; | |
847 | } | |
848 | } else { | |
849 | /* We are in dangerous area. Try to reduce cache really | |
850 | * aggressively. | |
851 | */ | |
852 | goal = max_t(unsigned int, goal / 2, rt_hash_mask + 1); | |
853 | equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal; | |
854 | } | |
855 | ||
856 | if (now - last_gc >= ip_rt_gc_min_interval) | |
857 | last_gc = now; | |
858 | ||
859 | if (goal <= 0) { | |
860 | equilibrium += goal; | |
861 | goto work_done; | |
862 | } | |
863 | ||
864 | do { | |
865 | int i, k; | |
866 | ||
867 | for (i = rt_hash_mask, k = rover; i >= 0; i--) { | |
868 | unsigned long tmo = expire; | |
869 | ||
870 | k = (k + 1) & rt_hash_mask; | |
871 | rthp = &rt_hash_table[k].chain; | |
22c047cc | 872 | spin_lock_bh(rt_hash_lock_addr(k)); |
1da177e4 LT |
873 | while ((rth = *rthp) != NULL) { |
874 | if (!rt_may_expire(rth, tmo, expire)) { | |
875 | tmo >>= 1; | |
093c2ca4 | 876 | rthp = &rth->u.dst.rt_next; |
1da177e4 LT |
877 | continue; |
878 | } | |
093c2ca4 | 879 | *rthp = rth->u.dst.rt_next; |
1da177e4 LT |
880 | rt_free(rth); |
881 | goal--; | |
1da177e4 | 882 | } |
22c047cc | 883 | spin_unlock_bh(rt_hash_lock_addr(k)); |
1da177e4 LT |
884 | if (goal <= 0) |
885 | break; | |
886 | } | |
887 | rover = k; | |
888 | ||
889 | if (goal <= 0) | |
890 | goto work_done; | |
891 | ||
892 | /* Goal is not achieved. We stop process if: | |
893 | ||
894 | - if expire reduced to zero. Otherwise, expire is halfed. | |
895 | - if table is not full. | |
896 | - if we are called from interrupt. | |
897 | - jiffies check is just fallback/debug loop breaker. | |
898 | We will not spin here for long time in any case. | |
899 | */ | |
900 | ||
901 | RT_CACHE_STAT_INC(gc_goal_miss); | |
902 | ||
903 | if (expire == 0) | |
904 | break; | |
905 | ||
906 | expire >>= 1; | |
907 | #if RT_CACHE_DEBUG >= 2 | |
908 | printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire, | |
909 | atomic_read(&ipv4_dst_ops.entries), goal, i); | |
910 | #endif | |
911 | ||
912 | if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) | |
913 | goto out; | |
914 | } while (!in_softirq() && time_before_eq(jiffies, now)); | |
915 | ||
916 | if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) | |
917 | goto out; | |
918 | if (net_ratelimit()) | |
919 | printk(KERN_WARNING "dst cache overflow\n"); | |
920 | RT_CACHE_STAT_INC(gc_dst_overflow); | |
921 | return 1; | |
922 | ||
923 | work_done: | |
924 | expire += ip_rt_gc_min_interval; | |
925 | if (expire > ip_rt_gc_timeout || | |
926 | atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh) | |
927 | expire = ip_rt_gc_timeout; | |
928 | #if RT_CACHE_DEBUG >= 2 | |
929 | printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire, | |
930 | atomic_read(&ipv4_dst_ops.entries), goal, rover); | |
931 | #endif | |
932 | out: return 0; | |
933 | } | |
934 | ||
935 | static int rt_intern_hash(unsigned hash, struct rtable *rt, struct rtable **rp) | |
936 | { | |
937 | struct rtable *rth, **rthp; | |
938 | unsigned long now; | |
939 | struct rtable *cand, **candp; | |
940 | u32 min_score; | |
941 | int chain_length; | |
942 | int attempts = !in_softirq(); | |
943 | ||
944 | restart: | |
945 | chain_length = 0; | |
946 | min_score = ~(u32)0; | |
947 | cand = NULL; | |
948 | candp = NULL; | |
949 | now = jiffies; | |
950 | ||
951 | rthp = &rt_hash_table[hash].chain; | |
952 | ||
22c047cc | 953 | spin_lock_bh(rt_hash_lock_addr(hash)); |
1da177e4 | 954 | while ((rth = *rthp) != NULL) { |
1da177e4 | 955 | if (compare_keys(&rth->fl, &rt->fl)) { |
1da177e4 | 956 | /* Put it first */ |
093c2ca4 | 957 | *rthp = rth->u.dst.rt_next; |
1da177e4 LT |
958 | /* |
959 | * Since lookup is lockfree, the deletion | |
960 | * must be visible to another weakly ordered CPU before | |
961 | * the insertion at the start of the hash chain. | |
962 | */ | |
093c2ca4 | 963 | rcu_assign_pointer(rth->u.dst.rt_next, |
1da177e4 LT |
964 | rt_hash_table[hash].chain); |
965 | /* | |
966 | * Since lookup is lockfree, the update writes | |
967 | * must be ordered for consistency on SMP. | |
968 | */ | |
969 | rcu_assign_pointer(rt_hash_table[hash].chain, rth); | |
970 | ||
03f49f34 | 971 | dst_use(&rth->u.dst, now); |
22c047cc | 972 | spin_unlock_bh(rt_hash_lock_addr(hash)); |
1da177e4 LT |
973 | |
974 | rt_drop(rt); | |
975 | *rp = rth; | |
976 | return 0; | |
977 | } | |
978 | ||
979 | if (!atomic_read(&rth->u.dst.__refcnt)) { | |
980 | u32 score = rt_score(rth); | |
981 | ||
982 | if (score <= min_score) { | |
983 | cand = rth; | |
984 | candp = rthp; | |
985 | min_score = score; | |
986 | } | |
987 | } | |
988 | ||
989 | chain_length++; | |
990 | ||
093c2ca4 | 991 | rthp = &rth->u.dst.rt_next; |
1da177e4 LT |
992 | } |
993 | ||
994 | if (cand) { | |
995 | /* ip_rt_gc_elasticity used to be average length of chain | |
996 | * length, when exceeded gc becomes really aggressive. | |
997 | * | |
998 | * The second limit is less certain. At the moment it allows | |
999 | * only 2 entries per bucket. We will see. | |
1000 | */ | |
1001 | if (chain_length > ip_rt_gc_elasticity) { | |
093c2ca4 | 1002 | *candp = cand->u.dst.rt_next; |
1da177e4 LT |
1003 | rt_free(cand); |
1004 | } | |
1005 | } | |
1006 | ||
1007 | /* Try to bind route to arp only if it is output | |
1008 | route or unicast forwarding path. | |
1009 | */ | |
1010 | if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) { | |
1011 | int err = arp_bind_neighbour(&rt->u.dst); | |
1012 | if (err) { | |
22c047cc | 1013 | spin_unlock_bh(rt_hash_lock_addr(hash)); |
1da177e4 LT |
1014 | |
1015 | if (err != -ENOBUFS) { | |
1016 | rt_drop(rt); | |
1017 | return err; | |
1018 | } | |
1019 | ||
1020 | /* Neighbour tables are full and nothing | |
1021 | can be released. Try to shrink route cache, | |
1022 | it is most likely it holds some neighbour records. | |
1023 | */ | |
1024 | if (attempts-- > 0) { | |
1025 | int saved_elasticity = ip_rt_gc_elasticity; | |
1026 | int saved_int = ip_rt_gc_min_interval; | |
1027 | ip_rt_gc_elasticity = 1; | |
1028 | ip_rt_gc_min_interval = 0; | |
1029 | rt_garbage_collect(); | |
1030 | ip_rt_gc_min_interval = saved_int; | |
1031 | ip_rt_gc_elasticity = saved_elasticity; | |
1032 | goto restart; | |
1033 | } | |
1034 | ||
1035 | if (net_ratelimit()) | |
1036 | printk(KERN_WARNING "Neighbour table overflow.\n"); | |
1037 | rt_drop(rt); | |
1038 | return -ENOBUFS; | |
1039 | } | |
1040 | } | |
1041 | ||
093c2ca4 | 1042 | rt->u.dst.rt_next = rt_hash_table[hash].chain; |
1da177e4 | 1043 | #if RT_CACHE_DEBUG >= 2 |
093c2ca4 | 1044 | if (rt->u.dst.rt_next) { |
1da177e4 LT |
1045 | struct rtable *trt; |
1046 | printk(KERN_DEBUG "rt_cache @%02x: %u.%u.%u.%u", hash, | |
1047 | NIPQUAD(rt->rt_dst)); | |
093c2ca4 | 1048 | for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next) |
1da177e4 LT |
1049 | printk(" . %u.%u.%u.%u", NIPQUAD(trt->rt_dst)); |
1050 | printk("\n"); | |
1051 | } | |
1052 | #endif | |
1053 | rt_hash_table[hash].chain = rt; | |
22c047cc | 1054 | spin_unlock_bh(rt_hash_lock_addr(hash)); |
1da177e4 LT |
1055 | *rp = rt; |
1056 | return 0; | |
1057 | } | |
1058 | ||
1059 | void rt_bind_peer(struct rtable *rt, int create) | |
1060 | { | |
1061 | static DEFINE_SPINLOCK(rt_peer_lock); | |
1062 | struct inet_peer *peer; | |
1063 | ||
1064 | peer = inet_getpeer(rt->rt_dst, create); | |
1065 | ||
1066 | spin_lock_bh(&rt_peer_lock); | |
1067 | if (rt->peer == NULL) { | |
1068 | rt->peer = peer; | |
1069 | peer = NULL; | |
1070 | } | |
1071 | spin_unlock_bh(&rt_peer_lock); | |
1072 | if (peer) | |
1073 | inet_putpeer(peer); | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * Peer allocation may fail only in serious out-of-memory conditions. However | |
1078 | * we still can generate some output. | |
1079 | * Random ID selection looks a bit dangerous because we have no chances to | |
1080 | * select ID being unique in a reasonable period of time. | |
1081 | * But broken packet identifier may be better than no packet at all. | |
1082 | */ | |
1083 | static void ip_select_fb_ident(struct iphdr *iph) | |
1084 | { | |
1085 | static DEFINE_SPINLOCK(ip_fb_id_lock); | |
1086 | static u32 ip_fallback_id; | |
1087 | u32 salt; | |
1088 | ||
1089 | spin_lock_bh(&ip_fb_id_lock); | |
e448515c | 1090 | salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr); |
1da177e4 LT |
1091 | iph->id = htons(salt & 0xFFFF); |
1092 | ip_fallback_id = salt; | |
1093 | spin_unlock_bh(&ip_fb_id_lock); | |
1094 | } | |
1095 | ||
1096 | void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more) | |
1097 | { | |
1098 | struct rtable *rt = (struct rtable *) dst; | |
1099 | ||
1100 | if (rt) { | |
1101 | if (rt->peer == NULL) | |
1102 | rt_bind_peer(rt, 1); | |
1103 | ||
1104 | /* If peer is attached to destination, it is never detached, | |
1105 | so that we need not to grab a lock to dereference it. | |
1106 | */ | |
1107 | if (rt->peer) { | |
1108 | iph->id = htons(inet_getid(rt->peer, more)); | |
1109 | return; | |
1110 | } | |
1111 | } else | |
e905a9ed | 1112 | printk(KERN_DEBUG "rt_bind_peer(0) @%p\n", |
9c2b3328 | 1113 | __builtin_return_address(0)); |
1da177e4 LT |
1114 | |
1115 | ip_select_fb_ident(iph); | |
1116 | } | |
1117 | ||
1118 | static void rt_del(unsigned hash, struct rtable *rt) | |
1119 | { | |
1120 | struct rtable **rthp; | |
1121 | ||
22c047cc | 1122 | spin_lock_bh(rt_hash_lock_addr(hash)); |
1da177e4 LT |
1123 | ip_rt_put(rt); |
1124 | for (rthp = &rt_hash_table[hash].chain; *rthp; | |
093c2ca4 | 1125 | rthp = &(*rthp)->u.dst.rt_next) |
1da177e4 | 1126 | if (*rthp == rt) { |
093c2ca4 | 1127 | *rthp = rt->u.dst.rt_next; |
1da177e4 LT |
1128 | rt_free(rt); |
1129 | break; | |
1130 | } | |
22c047cc | 1131 | spin_unlock_bh(rt_hash_lock_addr(hash)); |
1da177e4 LT |
1132 | } |
1133 | ||
f7655229 AV |
1134 | void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw, |
1135 | __be32 saddr, struct net_device *dev) | |
1da177e4 LT |
1136 | { |
1137 | int i, k; | |
1138 | struct in_device *in_dev = in_dev_get(dev); | |
1139 | struct rtable *rth, **rthp; | |
f7655229 | 1140 | __be32 skeys[2] = { saddr, 0 }; |
1da177e4 | 1141 | int ikeys[2] = { dev->ifindex, 0 }; |
8d71740c | 1142 | struct netevent_redirect netevent; |
1da177e4 | 1143 | |
1da177e4 LT |
1144 | if (!in_dev) |
1145 | return; | |
1146 | ||
1147 | if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) | |
1148 | || MULTICAST(new_gw) || BADCLASS(new_gw) || ZERONET(new_gw)) | |
1149 | goto reject_redirect; | |
1150 | ||
1151 | if (!IN_DEV_SHARED_MEDIA(in_dev)) { | |
1152 | if (!inet_addr_onlink(in_dev, new_gw, old_gw)) | |
1153 | goto reject_redirect; | |
1154 | if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) | |
1155 | goto reject_redirect; | |
1156 | } else { | |
1157 | if (inet_addr_type(new_gw) != RTN_UNICAST) | |
1158 | goto reject_redirect; | |
1159 | } | |
1160 | ||
1161 | for (i = 0; i < 2; i++) { | |
1162 | for (k = 0; k < 2; k++) { | |
8c7bc840 | 1163 | unsigned hash = rt_hash(daddr, skeys[i], ikeys[k]); |
1da177e4 LT |
1164 | |
1165 | rthp=&rt_hash_table[hash].chain; | |
1166 | ||
1167 | rcu_read_lock(); | |
1168 | while ((rth = rcu_dereference(*rthp)) != NULL) { | |
1169 | struct rtable *rt; | |
1170 | ||
1171 | if (rth->fl.fl4_dst != daddr || | |
1172 | rth->fl.fl4_src != skeys[i] || | |
1da177e4 LT |
1173 | rth->fl.oif != ikeys[k] || |
1174 | rth->fl.iif != 0) { | |
093c2ca4 | 1175 | rthp = &rth->u.dst.rt_next; |
1da177e4 LT |
1176 | continue; |
1177 | } | |
1178 | ||
1179 | if (rth->rt_dst != daddr || | |
1180 | rth->rt_src != saddr || | |
1181 | rth->u.dst.error || | |
1182 | rth->rt_gateway != old_gw || | |
1183 | rth->u.dst.dev != dev) | |
1184 | break; | |
1185 | ||
1186 | dst_hold(&rth->u.dst); | |
1187 | rcu_read_unlock(); | |
1188 | ||
1189 | rt = dst_alloc(&ipv4_dst_ops); | |
1190 | if (rt == NULL) { | |
1191 | ip_rt_put(rth); | |
1192 | in_dev_put(in_dev); | |
1193 | return; | |
1194 | } | |
1195 | ||
1196 | /* Copy all the information. */ | |
1197 | *rt = *rth; | |
e905a9ed | 1198 | INIT_RCU_HEAD(&rt->u.dst.rcu_head); |
1da177e4 LT |
1199 | rt->u.dst.__use = 1; |
1200 | atomic_set(&rt->u.dst.__refcnt, 1); | |
1201 | rt->u.dst.child = NULL; | |
1202 | if (rt->u.dst.dev) | |
1203 | dev_hold(rt->u.dst.dev); | |
1204 | if (rt->idev) | |
1205 | in_dev_hold(rt->idev); | |
1206 | rt->u.dst.obsolete = 0; | |
1207 | rt->u.dst.lastuse = jiffies; | |
1208 | rt->u.dst.path = &rt->u.dst; | |
1209 | rt->u.dst.neighbour = NULL; | |
1210 | rt->u.dst.hh = NULL; | |
1211 | rt->u.dst.xfrm = NULL; | |
1212 | ||
1213 | rt->rt_flags |= RTCF_REDIRECTED; | |
1214 | ||
1215 | /* Gateway is different ... */ | |
1216 | rt->rt_gateway = new_gw; | |
1217 | ||
1218 | /* Redirect received -> path was valid */ | |
1219 | dst_confirm(&rth->u.dst); | |
1220 | ||
1221 | if (rt->peer) | |
1222 | atomic_inc(&rt->peer->refcnt); | |
1223 | ||
1224 | if (arp_bind_neighbour(&rt->u.dst) || | |
1225 | !(rt->u.dst.neighbour->nud_state & | |
1226 | NUD_VALID)) { | |
1227 | if (rt->u.dst.neighbour) | |
1228 | neigh_event_send(rt->u.dst.neighbour, NULL); | |
1229 | ip_rt_put(rth); | |
1230 | rt_drop(rt); | |
1231 | goto do_next; | |
1232 | } | |
e905a9ed | 1233 | |
8d71740c TT |
1234 | netevent.old = &rth->u.dst; |
1235 | netevent.new = &rt->u.dst; | |
e905a9ed YH |
1236 | call_netevent_notifiers(NETEVENT_REDIRECT, |
1237 | &netevent); | |
1da177e4 LT |
1238 | |
1239 | rt_del(hash, rth); | |
1240 | if (!rt_intern_hash(hash, rt, &rt)) | |
1241 | ip_rt_put(rt); | |
1242 | goto do_next; | |
1243 | } | |
1244 | rcu_read_unlock(); | |
1245 | do_next: | |
1246 | ; | |
1247 | } | |
1248 | } | |
1249 | in_dev_put(in_dev); | |
1250 | return; | |
1251 | ||
1252 | reject_redirect: | |
1253 | #ifdef CONFIG_IP_ROUTE_VERBOSE | |
1254 | if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) | |
1255 | printk(KERN_INFO "Redirect from %u.%u.%u.%u on %s about " | |
1256 | "%u.%u.%u.%u ignored.\n" | |
cef2685e | 1257 | " Advised path = %u.%u.%u.%u -> %u.%u.%u.%u\n", |
1da177e4 | 1258 | NIPQUAD(old_gw), dev->name, NIPQUAD(new_gw), |
cef2685e | 1259 | NIPQUAD(saddr), NIPQUAD(daddr)); |
1da177e4 LT |
1260 | #endif |
1261 | in_dev_put(in_dev); | |
1262 | } | |
1263 | ||
1264 | static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) | |
1265 | { | |
1266 | struct rtable *rt = (struct rtable*)dst; | |
1267 | struct dst_entry *ret = dst; | |
1268 | ||
1269 | if (rt) { | |
1270 | if (dst->obsolete) { | |
1271 | ip_rt_put(rt); | |
1272 | ret = NULL; | |
1273 | } else if ((rt->rt_flags & RTCF_REDIRECTED) || | |
1274 | rt->u.dst.expires) { | |
8c7bc840 AV |
1275 | unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src, |
1276 | rt->fl.oif); | |
1da177e4 | 1277 | #if RT_CACHE_DEBUG >= 1 |
56c99d04 | 1278 | printk(KERN_DEBUG "ipv4_negative_advice: redirect to " |
1da177e4 LT |
1279 | "%u.%u.%u.%u/%02x dropped\n", |
1280 | NIPQUAD(rt->rt_dst), rt->fl.fl4_tos); | |
1281 | #endif | |
1282 | rt_del(hash, rt); | |
1283 | ret = NULL; | |
1284 | } | |
1285 | } | |
1286 | return ret; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * Algorithm: | |
1291 | * 1. The first ip_rt_redirect_number redirects are sent | |
1292 | * with exponential backoff, then we stop sending them at all, | |
1293 | * assuming that the host ignores our redirects. | |
1294 | * 2. If we did not see packets requiring redirects | |
1295 | * during ip_rt_redirect_silence, we assume that the host | |
1296 | * forgot redirected route and start to send redirects again. | |
1297 | * | |
1298 | * This algorithm is much cheaper and more intelligent than dumb load limiting | |
1299 | * in icmp.c. | |
1300 | * | |
1301 | * NOTE. Do not forget to inhibit load limiting for redirects (redundant) | |
1302 | * and "frag. need" (breaks PMTU discovery) in icmp.c. | |
1303 | */ | |
1304 | ||
1305 | void ip_rt_send_redirect(struct sk_buff *skb) | |
1306 | { | |
1307 | struct rtable *rt = (struct rtable*)skb->dst; | |
1308 | struct in_device *in_dev = in_dev_get(rt->u.dst.dev); | |
1309 | ||
1310 | if (!in_dev) | |
1311 | return; | |
1312 | ||
1313 | if (!IN_DEV_TX_REDIRECTS(in_dev)) | |
1314 | goto out; | |
1315 | ||
1316 | /* No redirected packets during ip_rt_redirect_silence; | |
1317 | * reset the algorithm. | |
1318 | */ | |
1319 | if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence)) | |
1320 | rt->u.dst.rate_tokens = 0; | |
1321 | ||
1322 | /* Too many ignored redirects; do not send anything | |
1323 | * set u.dst.rate_last to the last seen redirected packet. | |
1324 | */ | |
1325 | if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) { | |
1326 | rt->u.dst.rate_last = jiffies; | |
1327 | goto out; | |
1328 | } | |
1329 | ||
1330 | /* Check for load limit; set rate_last to the latest sent | |
1331 | * redirect. | |
1332 | */ | |
14fb8a76 LY |
1333 | if (rt->u.dst.rate_tokens == 0 || |
1334 | time_after(jiffies, | |
1da177e4 LT |
1335 | (rt->u.dst.rate_last + |
1336 | (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) { | |
1337 | icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway); | |
1338 | rt->u.dst.rate_last = jiffies; | |
1339 | ++rt->u.dst.rate_tokens; | |
1340 | #ifdef CONFIG_IP_ROUTE_VERBOSE | |
1341 | if (IN_DEV_LOG_MARTIANS(in_dev) && | |
1342 | rt->u.dst.rate_tokens == ip_rt_redirect_number && | |
1343 | net_ratelimit()) | |
1344 | printk(KERN_WARNING "host %u.%u.%u.%u/if%d ignores " | |
1345 | "redirects for %u.%u.%u.%u to %u.%u.%u.%u.\n", | |
1346 | NIPQUAD(rt->rt_src), rt->rt_iif, | |
1347 | NIPQUAD(rt->rt_dst), NIPQUAD(rt->rt_gateway)); | |
1348 | #endif | |
1349 | } | |
1350 | out: | |
e905a9ed | 1351 | in_dev_put(in_dev); |
1da177e4 LT |
1352 | } |
1353 | ||
1354 | static int ip_error(struct sk_buff *skb) | |
1355 | { | |
1356 | struct rtable *rt = (struct rtable*)skb->dst; | |
1357 | unsigned long now; | |
1358 | int code; | |
1359 | ||
1360 | switch (rt->u.dst.error) { | |
1361 | case EINVAL: | |
1362 | default: | |
1363 | goto out; | |
1364 | case EHOSTUNREACH: | |
1365 | code = ICMP_HOST_UNREACH; | |
1366 | break; | |
1367 | case ENETUNREACH: | |
1368 | code = ICMP_NET_UNREACH; | |
7f53878d | 1369 | IP_INC_STATS_BH(IPSTATS_MIB_INNOROUTES); |
1da177e4 LT |
1370 | break; |
1371 | case EACCES: | |
1372 | code = ICMP_PKT_FILTERED; | |
1373 | break; | |
1374 | } | |
1375 | ||
1376 | now = jiffies; | |
1377 | rt->u.dst.rate_tokens += now - rt->u.dst.rate_last; | |
1378 | if (rt->u.dst.rate_tokens > ip_rt_error_burst) | |
1379 | rt->u.dst.rate_tokens = ip_rt_error_burst; | |
1380 | rt->u.dst.rate_last = now; | |
1381 | if (rt->u.dst.rate_tokens >= ip_rt_error_cost) { | |
1382 | rt->u.dst.rate_tokens -= ip_rt_error_cost; | |
1383 | icmp_send(skb, ICMP_DEST_UNREACH, code, 0); | |
1384 | } | |
1385 | ||
1386 | out: kfree_skb(skb); | |
1387 | return 0; | |
e905a9ed | 1388 | } |
1da177e4 LT |
1389 | |
1390 | /* | |
1391 | * The last two values are not from the RFC but | |
1392 | * are needed for AMPRnet AX.25 paths. | |
1393 | */ | |
1394 | ||
9b5b5cff | 1395 | static const unsigned short mtu_plateau[] = |
1da177e4 LT |
1396 | {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 }; |
1397 | ||
1398 | static __inline__ unsigned short guess_mtu(unsigned short old_mtu) | |
1399 | { | |
1400 | int i; | |
e905a9ed | 1401 | |
1da177e4 LT |
1402 | for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++) |
1403 | if (old_mtu > mtu_plateau[i]) | |
1404 | return mtu_plateau[i]; | |
1405 | return 68; | |
1406 | } | |
1407 | ||
1408 | unsigned short ip_rt_frag_needed(struct iphdr *iph, unsigned short new_mtu) | |
1409 | { | |
1410 | int i; | |
1411 | unsigned short old_mtu = ntohs(iph->tot_len); | |
1412 | struct rtable *rth; | |
e448515c AV |
1413 | __be32 skeys[2] = { iph->saddr, 0, }; |
1414 | __be32 daddr = iph->daddr; | |
1da177e4 LT |
1415 | unsigned short est_mtu = 0; |
1416 | ||
1417 | if (ipv4_config.no_pmtu_disc) | |
1418 | return 0; | |
1419 | ||
1420 | for (i = 0; i < 2; i++) { | |
8c7bc840 | 1421 | unsigned hash = rt_hash(daddr, skeys[i], 0); |
1da177e4 LT |
1422 | |
1423 | rcu_read_lock(); | |
1424 | for (rth = rcu_dereference(rt_hash_table[hash].chain); rth; | |
093c2ca4 | 1425 | rth = rcu_dereference(rth->u.dst.rt_next)) { |
1da177e4 LT |
1426 | if (rth->fl.fl4_dst == daddr && |
1427 | rth->fl.fl4_src == skeys[i] && | |
1428 | rth->rt_dst == daddr && | |
1429 | rth->rt_src == iph->saddr && | |
1da177e4 LT |
1430 | rth->fl.iif == 0 && |
1431 | !(dst_metric_locked(&rth->u.dst, RTAX_MTU))) { | |
1432 | unsigned short mtu = new_mtu; | |
1433 | ||
1434 | if (new_mtu < 68 || new_mtu >= old_mtu) { | |
1435 | ||
1436 | /* BSD 4.2 compatibility hack :-( */ | |
1437 | if (mtu == 0 && | |
1438 | old_mtu >= rth->u.dst.metrics[RTAX_MTU-1] && | |
1439 | old_mtu >= 68 + (iph->ihl << 2)) | |
1440 | old_mtu -= iph->ihl << 2; | |
1441 | ||
1442 | mtu = guess_mtu(old_mtu); | |
1443 | } | |
1444 | if (mtu <= rth->u.dst.metrics[RTAX_MTU-1]) { | |
e905a9ed | 1445 | if (mtu < rth->u.dst.metrics[RTAX_MTU-1]) { |
1da177e4 LT |
1446 | dst_confirm(&rth->u.dst); |
1447 | if (mtu < ip_rt_min_pmtu) { | |
1448 | mtu = ip_rt_min_pmtu; | |
1449 | rth->u.dst.metrics[RTAX_LOCK-1] |= | |
1450 | (1 << RTAX_MTU); | |
1451 | } | |
1452 | rth->u.dst.metrics[RTAX_MTU-1] = mtu; | |
1453 | dst_set_expires(&rth->u.dst, | |
1454 | ip_rt_mtu_expires); | |
1455 | } | |
1456 | est_mtu = mtu; | |
1457 | } | |
1458 | } | |
1459 | } | |
1460 | rcu_read_unlock(); | |
1461 | } | |
1462 | return est_mtu ? : new_mtu; | |
1463 | } | |
1464 | ||
1465 | static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu) | |
1466 | { | |
1467 | if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= 68 && | |
1468 | !(dst_metric_locked(dst, RTAX_MTU))) { | |
1469 | if (mtu < ip_rt_min_pmtu) { | |
1470 | mtu = ip_rt_min_pmtu; | |
1471 | dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU); | |
1472 | } | |
1473 | dst->metrics[RTAX_MTU-1] = mtu; | |
1474 | dst_set_expires(dst, ip_rt_mtu_expires); | |
8d71740c | 1475 | call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst); |
1da177e4 LT |
1476 | } |
1477 | } | |
1478 | ||
1479 | static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) | |
1480 | { | |
1481 | return NULL; | |
1482 | } | |
1483 | ||
1484 | static void ipv4_dst_destroy(struct dst_entry *dst) | |
1485 | { | |
1486 | struct rtable *rt = (struct rtable *) dst; | |
1487 | struct inet_peer *peer = rt->peer; | |
1488 | struct in_device *idev = rt->idev; | |
1489 | ||
1490 | if (peer) { | |
1491 | rt->peer = NULL; | |
1492 | inet_putpeer(peer); | |
1493 | } | |
1494 | ||
1495 | if (idev) { | |
1496 | rt->idev = NULL; | |
1497 | in_dev_put(idev); | |
1498 | } | |
1499 | } | |
1500 | ||
1501 | static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev, | |
1502 | int how) | |
1503 | { | |
1504 | struct rtable *rt = (struct rtable *) dst; | |
1505 | struct in_device *idev = rt->idev; | |
2774c7ab EB |
1506 | if (dev != init_net.loopback_dev && idev && idev->dev == dev) { |
1507 | struct in_device *loopback_idev = in_dev_get(init_net.loopback_dev); | |
1da177e4 LT |
1508 | if (loopback_idev) { |
1509 | rt->idev = loopback_idev; | |
1510 | in_dev_put(idev); | |
1511 | } | |
1512 | } | |
1513 | } | |
1514 | ||
1515 | static void ipv4_link_failure(struct sk_buff *skb) | |
1516 | { | |
1517 | struct rtable *rt; | |
1518 | ||
1519 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0); | |
1520 | ||
1521 | rt = (struct rtable *) skb->dst; | |
1522 | if (rt) | |
1523 | dst_set_expires(&rt->u.dst, 0); | |
1524 | } | |
1525 | ||
1526 | static int ip_rt_bug(struct sk_buff *skb) | |
1527 | { | |
1528 | printk(KERN_DEBUG "ip_rt_bug: %u.%u.%u.%u -> %u.%u.%u.%u, %s\n", | |
eddc9ec5 | 1529 | NIPQUAD(ip_hdr(skb)->saddr), NIPQUAD(ip_hdr(skb)->daddr), |
1da177e4 LT |
1530 | skb->dev ? skb->dev->name : "?"); |
1531 | kfree_skb(skb); | |
1532 | return 0; | |
1533 | } | |
1534 | ||
1535 | /* | |
1536 | We do not cache source address of outgoing interface, | |
1537 | because it is used only by IP RR, TS and SRR options, | |
1538 | so that it out of fast path. | |
1539 | ||
1540 | BTW remember: "addr" is allowed to be not aligned | |
1541 | in IP options! | |
1542 | */ | |
1543 | ||
1544 | void ip_rt_get_source(u8 *addr, struct rtable *rt) | |
1545 | { | |
a61ced5d | 1546 | __be32 src; |
1da177e4 LT |
1547 | struct fib_result res; |
1548 | ||
1549 | if (rt->fl.iif == 0) | |
1550 | src = rt->rt_src; | |
1551 | else if (fib_lookup(&rt->fl, &res) == 0) { | |
1552 | src = FIB_RES_PREFSRC(res); | |
1553 | fib_res_put(&res); | |
1554 | } else | |
1555 | src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway, | |
1556 | RT_SCOPE_UNIVERSE); | |
1557 | memcpy(addr, &src, 4); | |
1558 | } | |
1559 | ||
1560 | #ifdef CONFIG_NET_CLS_ROUTE | |
1561 | static void set_class_tag(struct rtable *rt, u32 tag) | |
1562 | { | |
1563 | if (!(rt->u.dst.tclassid & 0xFFFF)) | |
1564 | rt->u.dst.tclassid |= tag & 0xFFFF; | |
1565 | if (!(rt->u.dst.tclassid & 0xFFFF0000)) | |
1566 | rt->u.dst.tclassid |= tag & 0xFFFF0000; | |
1567 | } | |
1568 | #endif | |
1569 | ||
1570 | static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag) | |
1571 | { | |
1572 | struct fib_info *fi = res->fi; | |
1573 | ||
1574 | if (fi) { | |
1575 | if (FIB_RES_GW(*res) && | |
1576 | FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) | |
1577 | rt->rt_gateway = FIB_RES_GW(*res); | |
1578 | memcpy(rt->u.dst.metrics, fi->fib_metrics, | |
1579 | sizeof(rt->u.dst.metrics)); | |
1580 | if (fi->fib_mtu == 0) { | |
1581 | rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu; | |
1582 | if (rt->u.dst.metrics[RTAX_LOCK-1] & (1 << RTAX_MTU) && | |
1583 | rt->rt_gateway != rt->rt_dst && | |
1584 | rt->u.dst.dev->mtu > 576) | |
1585 | rt->u.dst.metrics[RTAX_MTU-1] = 576; | |
1586 | } | |
1587 | #ifdef CONFIG_NET_CLS_ROUTE | |
1588 | rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid; | |
1589 | #endif | |
1590 | } else | |
1591 | rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu; | |
1592 | ||
1593 | if (rt->u.dst.metrics[RTAX_HOPLIMIT-1] == 0) | |
1594 | rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl; | |
1595 | if (rt->u.dst.metrics[RTAX_MTU-1] > IP_MAX_MTU) | |
1596 | rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU; | |
1597 | if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0) | |
1598 | rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40, | |
1599 | ip_rt_min_advmss); | |
1600 | if (rt->u.dst.metrics[RTAX_ADVMSS-1] > 65535 - 40) | |
1601 | rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40; | |
1602 | ||
1603 | #ifdef CONFIG_NET_CLS_ROUTE | |
1604 | #ifdef CONFIG_IP_MULTIPLE_TABLES | |
1605 | set_class_tag(rt, fib_rules_tclass(res)); | |
1606 | #endif | |
1607 | set_class_tag(rt, itag); | |
1608 | #endif | |
e905a9ed | 1609 | rt->rt_type = res->type; |
1da177e4 LT |
1610 | } |
1611 | ||
9e12bb22 | 1612 | static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, |
1da177e4 LT |
1613 | u8 tos, struct net_device *dev, int our) |
1614 | { | |
1615 | unsigned hash; | |
1616 | struct rtable *rth; | |
a61ced5d | 1617 | __be32 spec_dst; |
1da177e4 LT |
1618 | struct in_device *in_dev = in_dev_get(dev); |
1619 | u32 itag = 0; | |
1620 | ||
1621 | /* Primary sanity checks. */ | |
1622 | ||
1623 | if (in_dev == NULL) | |
1624 | return -EINVAL; | |
1625 | ||
1626 | if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr) || | |
1627 | skb->protocol != htons(ETH_P_IP)) | |
1628 | goto e_inval; | |
1629 | ||
1630 | if (ZERONET(saddr)) { | |
1631 | if (!LOCAL_MCAST(daddr)) | |
1632 | goto e_inval; | |
1633 | spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK); | |
1634 | } else if (fib_validate_source(saddr, 0, tos, 0, | |
1635 | dev, &spec_dst, &itag) < 0) | |
1636 | goto e_inval; | |
1637 | ||
1638 | rth = dst_alloc(&ipv4_dst_ops); | |
1639 | if (!rth) | |
1640 | goto e_nobufs; | |
1641 | ||
1642 | rth->u.dst.output= ip_rt_bug; | |
1643 | ||
1644 | atomic_set(&rth->u.dst.__refcnt, 1); | |
1645 | rth->u.dst.flags= DST_HOST; | |
42f811b8 | 1646 | if (IN_DEV_CONF_GET(in_dev, NOPOLICY)) |
1da177e4 LT |
1647 | rth->u.dst.flags |= DST_NOPOLICY; |
1648 | rth->fl.fl4_dst = daddr; | |
1649 | rth->rt_dst = daddr; | |
1650 | rth->fl.fl4_tos = tos; | |
47dcf0cb | 1651 | rth->fl.mark = skb->mark; |
1da177e4 LT |
1652 | rth->fl.fl4_src = saddr; |
1653 | rth->rt_src = saddr; | |
1654 | #ifdef CONFIG_NET_CLS_ROUTE | |
1655 | rth->u.dst.tclassid = itag; | |
1656 | #endif | |
1657 | rth->rt_iif = | |
1658 | rth->fl.iif = dev->ifindex; | |
2774c7ab | 1659 | rth->u.dst.dev = init_net.loopback_dev; |
1da177e4 LT |
1660 | dev_hold(rth->u.dst.dev); |
1661 | rth->idev = in_dev_get(rth->u.dst.dev); | |
1662 | rth->fl.oif = 0; | |
1663 | rth->rt_gateway = daddr; | |
1664 | rth->rt_spec_dst= spec_dst; | |
1665 | rth->rt_type = RTN_MULTICAST; | |
1666 | rth->rt_flags = RTCF_MULTICAST; | |
1667 | if (our) { | |
1668 | rth->u.dst.input= ip_local_deliver; | |
1669 | rth->rt_flags |= RTCF_LOCAL; | |
1670 | } | |
1671 | ||
1672 | #ifdef CONFIG_IP_MROUTE | |
1673 | if (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev)) | |
1674 | rth->u.dst.input = ip_mr_input; | |
1675 | #endif | |
1676 | RT_CACHE_STAT_INC(in_slow_mc); | |
1677 | ||
1678 | in_dev_put(in_dev); | |
8c7bc840 | 1679 | hash = rt_hash(daddr, saddr, dev->ifindex); |
1da177e4 LT |
1680 | return rt_intern_hash(hash, rth, (struct rtable**) &skb->dst); |
1681 | ||
1682 | e_nobufs: | |
1683 | in_dev_put(in_dev); | |
1684 | return -ENOBUFS; | |
1685 | ||
1686 | e_inval: | |
1687 | in_dev_put(in_dev); | |
1688 | return -EINVAL; | |
1689 | } | |
1690 | ||
1691 | ||
1692 | static void ip_handle_martian_source(struct net_device *dev, | |
1693 | struct in_device *in_dev, | |
1694 | struct sk_buff *skb, | |
9e12bb22 AV |
1695 | __be32 daddr, |
1696 | __be32 saddr) | |
1da177e4 LT |
1697 | { |
1698 | RT_CACHE_STAT_INC(in_martian_src); | |
1699 | #ifdef CONFIG_IP_ROUTE_VERBOSE | |
1700 | if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { | |
1701 | /* | |
1702 | * RFC1812 recommendation, if source is martian, | |
1703 | * the only hint is MAC header. | |
1704 | */ | |
1705 | printk(KERN_WARNING "martian source %u.%u.%u.%u from " | |
1706 | "%u.%u.%u.%u, on dev %s\n", | |
1707 | NIPQUAD(daddr), NIPQUAD(saddr), dev->name); | |
98e399f8 | 1708 | if (dev->hard_header_len && skb_mac_header_was_set(skb)) { |
1da177e4 | 1709 | int i; |
98e399f8 | 1710 | const unsigned char *p = skb_mac_header(skb); |
1da177e4 LT |
1711 | printk(KERN_WARNING "ll header: "); |
1712 | for (i = 0; i < dev->hard_header_len; i++, p++) { | |
1713 | printk("%02x", *p); | |
1714 | if (i < (dev->hard_header_len - 1)) | |
1715 | printk(":"); | |
1716 | } | |
1717 | printk("\n"); | |
1718 | } | |
1719 | } | |
1720 | #endif | |
1721 | } | |
1722 | ||
e905a9ed YH |
1723 | static inline int __mkroute_input(struct sk_buff *skb, |
1724 | struct fib_result* res, | |
1725 | struct in_device *in_dev, | |
9e12bb22 | 1726 | __be32 daddr, __be32 saddr, u32 tos, |
e905a9ed | 1727 | struct rtable **result) |
1da177e4 LT |
1728 | { |
1729 | ||
1730 | struct rtable *rth; | |
1731 | int err; | |
1732 | struct in_device *out_dev; | |
1733 | unsigned flags = 0; | |
d9c9df8c AV |
1734 | __be32 spec_dst; |
1735 | u32 itag; | |
1da177e4 LT |
1736 | |
1737 | /* get a working reference to the output device */ | |
1738 | out_dev = in_dev_get(FIB_RES_DEV(*res)); | |
1739 | if (out_dev == NULL) { | |
1740 | if (net_ratelimit()) | |
1741 | printk(KERN_CRIT "Bug in ip_route_input" \ | |
1742 | "_slow(). Please, report\n"); | |
1743 | return -EINVAL; | |
1744 | } | |
1745 | ||
1746 | ||
e905a9ed | 1747 | err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res), |
1da177e4 LT |
1748 | in_dev->dev, &spec_dst, &itag); |
1749 | if (err < 0) { | |
e905a9ed | 1750 | ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, |
1da177e4 | 1751 | saddr); |
e905a9ed | 1752 | |
1da177e4 LT |
1753 | err = -EINVAL; |
1754 | goto cleanup; | |
1755 | } | |
1756 | ||
1757 | if (err) | |
1758 | flags |= RTCF_DIRECTSRC; | |
1759 | ||
1760 | if (out_dev == in_dev && err && !(flags & (RTCF_NAT | RTCF_MASQ)) && | |
1761 | (IN_DEV_SHARED_MEDIA(out_dev) || | |
1762 | inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) | |
1763 | flags |= RTCF_DOREDIRECT; | |
1764 | ||
1765 | if (skb->protocol != htons(ETH_P_IP)) { | |
1766 | /* Not IP (i.e. ARP). Do not create route, if it is | |
1767 | * invalid for proxy arp. DNAT routes are always valid. | |
1768 | */ | |
1769 | if (out_dev == in_dev && !(flags & RTCF_DNAT)) { | |
1770 | err = -EINVAL; | |
1771 | goto cleanup; | |
1772 | } | |
1773 | } | |
1774 | ||
1775 | ||
1776 | rth = dst_alloc(&ipv4_dst_ops); | |
1777 | if (!rth) { | |
1778 | err = -ENOBUFS; | |
1779 | goto cleanup; | |
1780 | } | |
1781 | ||
ce723d8e | 1782 | atomic_set(&rth->u.dst.__refcnt, 1); |
1da177e4 | 1783 | rth->u.dst.flags= DST_HOST; |
42f811b8 | 1784 | if (IN_DEV_CONF_GET(in_dev, NOPOLICY)) |
1da177e4 | 1785 | rth->u.dst.flags |= DST_NOPOLICY; |
42f811b8 | 1786 | if (IN_DEV_CONF_GET(out_dev, NOXFRM)) |
1da177e4 LT |
1787 | rth->u.dst.flags |= DST_NOXFRM; |
1788 | rth->fl.fl4_dst = daddr; | |
1789 | rth->rt_dst = daddr; | |
1790 | rth->fl.fl4_tos = tos; | |
47dcf0cb | 1791 | rth->fl.mark = skb->mark; |
1da177e4 LT |
1792 | rth->fl.fl4_src = saddr; |
1793 | rth->rt_src = saddr; | |
1794 | rth->rt_gateway = daddr; | |
1795 | rth->rt_iif = | |
1796 | rth->fl.iif = in_dev->dev->ifindex; | |
1797 | rth->u.dst.dev = (out_dev)->dev; | |
1798 | dev_hold(rth->u.dst.dev); | |
1799 | rth->idev = in_dev_get(rth->u.dst.dev); | |
1800 | rth->fl.oif = 0; | |
1801 | rth->rt_spec_dst= spec_dst; | |
1802 | ||
1803 | rth->u.dst.input = ip_forward; | |
1804 | rth->u.dst.output = ip_output; | |
1805 | ||
1806 | rt_set_nexthop(rth, res, itag); | |
1807 | ||
1808 | rth->rt_flags = flags; | |
1809 | ||
1810 | *result = rth; | |
1811 | err = 0; | |
1812 | cleanup: | |
1813 | /* release the working reference to the output device */ | |
1814 | in_dev_put(out_dev); | |
1815 | return err; | |
e905a9ed | 1816 | } |
1da177e4 | 1817 | |
e06e7c61 DM |
1818 | static inline int ip_mkroute_input(struct sk_buff *skb, |
1819 | struct fib_result* res, | |
1820 | const struct flowi *fl, | |
1821 | struct in_device *in_dev, | |
1822 | __be32 daddr, __be32 saddr, u32 tos) | |
1da177e4 | 1823 | { |
7abaa27c | 1824 | struct rtable* rth = NULL; |
1da177e4 LT |
1825 | int err; |
1826 | unsigned hash; | |
1827 | ||
1828 | #ifdef CONFIG_IP_ROUTE_MULTIPATH | |
1829 | if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0) | |
1830 | fib_select_multipath(fl, res); | |
1831 | #endif | |
1832 | ||
1833 | /* create a routing cache entry */ | |
1834 | err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth); | |
1835 | if (err) | |
1836 | return err; | |
1da177e4 LT |
1837 | |
1838 | /* put it into the cache */ | |
8c7bc840 | 1839 | hash = rt_hash(daddr, saddr, fl->iif); |
e905a9ed | 1840 | return rt_intern_hash(hash, rth, (struct rtable**)&skb->dst); |
1da177e4 LT |
1841 | } |
1842 | ||
1da177e4 LT |
1843 | /* |
1844 | * NOTE. We drop all the packets that has local source | |
1845 | * addresses, because every properly looped back packet | |
1846 | * must have correct destination already attached by output routine. | |
1847 | * | |
1848 | * Such approach solves two big problems: | |
1849 | * 1. Not simplex devices are handled properly. | |
1850 | * 2. IP spoofing attempts are filtered with 100% of guarantee. | |
1851 | */ | |
1852 | ||
9e12bb22 | 1853 | static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, |
1da177e4 LT |
1854 | u8 tos, struct net_device *dev) |
1855 | { | |
1856 | struct fib_result res; | |
1857 | struct in_device *in_dev = in_dev_get(dev); | |
1858 | struct flowi fl = { .nl_u = { .ip4_u = | |
1859 | { .daddr = daddr, | |
1860 | .saddr = saddr, | |
1861 | .tos = tos, | |
1862 | .scope = RT_SCOPE_UNIVERSE, | |
1da177e4 | 1863 | } }, |
47dcf0cb | 1864 | .mark = skb->mark, |
1da177e4 LT |
1865 | .iif = dev->ifindex }; |
1866 | unsigned flags = 0; | |
1867 | u32 itag = 0; | |
1868 | struct rtable * rth; | |
1869 | unsigned hash; | |
9e12bb22 | 1870 | __be32 spec_dst; |
1da177e4 LT |
1871 | int err = -EINVAL; |
1872 | int free_res = 0; | |
1873 | ||
1874 | /* IP on this device is disabled. */ | |
1875 | ||
1876 | if (!in_dev) | |
1877 | goto out; | |
1878 | ||
1879 | /* Check for the most weird martians, which can be not detected | |
1880 | by fib_lookup. | |
1881 | */ | |
1882 | ||
1883 | if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr)) | |
1884 | goto martian_source; | |
1885 | ||
e448515c | 1886 | if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0)) |
1da177e4 LT |
1887 | goto brd_input; |
1888 | ||
1889 | /* Accept zero addresses only to limited broadcast; | |
1890 | * I even do not know to fix it or not. Waiting for complains :-) | |
1891 | */ | |
1892 | if (ZERONET(saddr)) | |
1893 | goto martian_source; | |
1894 | ||
1895 | if (BADCLASS(daddr) || ZERONET(daddr) || LOOPBACK(daddr)) | |
1896 | goto martian_destination; | |
1897 | ||
1898 | /* | |
1899 | * Now we are ready to route packet. | |
1900 | */ | |
1901 | if ((err = fib_lookup(&fl, &res)) != 0) { | |
1902 | if (!IN_DEV_FORWARD(in_dev)) | |
2c2910a4 | 1903 | goto e_hostunreach; |
1da177e4 LT |
1904 | goto no_route; |
1905 | } | |
1906 | free_res = 1; | |
1907 | ||
1908 | RT_CACHE_STAT_INC(in_slow_tot); | |
1909 | ||
1910 | if (res.type == RTN_BROADCAST) | |
1911 | goto brd_input; | |
1912 | ||
1913 | if (res.type == RTN_LOCAL) { | |
1914 | int result; | |
1915 | result = fib_validate_source(saddr, daddr, tos, | |
2774c7ab | 1916 | init_net.loopback_dev->ifindex, |
1da177e4 LT |
1917 | dev, &spec_dst, &itag); |
1918 | if (result < 0) | |
1919 | goto martian_source; | |
1920 | if (result) | |
1921 | flags |= RTCF_DIRECTSRC; | |
1922 | spec_dst = daddr; | |
1923 | goto local_input; | |
1924 | } | |
1925 | ||
1926 | if (!IN_DEV_FORWARD(in_dev)) | |
2c2910a4 | 1927 | goto e_hostunreach; |
1da177e4 LT |
1928 | if (res.type != RTN_UNICAST) |
1929 | goto martian_destination; | |
1930 | ||
1931 | err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos); | |
1da177e4 LT |
1932 | done: |
1933 | in_dev_put(in_dev); | |
1934 | if (free_res) | |
1935 | fib_res_put(&res); | |
1936 | out: return err; | |
1937 | ||
1938 | brd_input: | |
1939 | if (skb->protocol != htons(ETH_P_IP)) | |
1940 | goto e_inval; | |
1941 | ||
1942 | if (ZERONET(saddr)) | |
1943 | spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK); | |
1944 | else { | |
1945 | err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst, | |
1946 | &itag); | |
1947 | if (err < 0) | |
1948 | goto martian_source; | |
1949 | if (err) | |
1950 | flags |= RTCF_DIRECTSRC; | |
1951 | } | |
1952 | flags |= RTCF_BROADCAST; | |
1953 | res.type = RTN_BROADCAST; | |
1954 | RT_CACHE_STAT_INC(in_brd); | |
1955 | ||
1956 | local_input: | |
1957 | rth = dst_alloc(&ipv4_dst_ops); | |
1958 | if (!rth) | |
1959 | goto e_nobufs; | |
1960 | ||
1961 | rth->u.dst.output= ip_rt_bug; | |
1962 | ||
1963 | atomic_set(&rth->u.dst.__refcnt, 1); | |
1964 | rth->u.dst.flags= DST_HOST; | |
42f811b8 | 1965 | if (IN_DEV_CONF_GET(in_dev, NOPOLICY)) |
1da177e4 LT |
1966 | rth->u.dst.flags |= DST_NOPOLICY; |
1967 | rth->fl.fl4_dst = daddr; | |
1968 | rth->rt_dst = daddr; | |
1969 | rth->fl.fl4_tos = tos; | |
47dcf0cb | 1970 | rth->fl.mark = skb->mark; |
1da177e4 LT |
1971 | rth->fl.fl4_src = saddr; |
1972 | rth->rt_src = saddr; | |
1973 | #ifdef CONFIG_NET_CLS_ROUTE | |
1974 | rth->u.dst.tclassid = itag; | |
1975 | #endif | |
1976 | rth->rt_iif = | |
1977 | rth->fl.iif = dev->ifindex; | |
2774c7ab | 1978 | rth->u.dst.dev = init_net.loopback_dev; |
1da177e4 LT |
1979 | dev_hold(rth->u.dst.dev); |
1980 | rth->idev = in_dev_get(rth->u.dst.dev); | |
1981 | rth->rt_gateway = daddr; | |
1982 | rth->rt_spec_dst= spec_dst; | |
1983 | rth->u.dst.input= ip_local_deliver; | |
1984 | rth->rt_flags = flags|RTCF_LOCAL; | |
1985 | if (res.type == RTN_UNREACHABLE) { | |
1986 | rth->u.dst.input= ip_error; | |
1987 | rth->u.dst.error= -err; | |
1988 | rth->rt_flags &= ~RTCF_LOCAL; | |
1989 | } | |
1990 | rth->rt_type = res.type; | |
8c7bc840 | 1991 | hash = rt_hash(daddr, saddr, fl.iif); |
1da177e4 LT |
1992 | err = rt_intern_hash(hash, rth, (struct rtable**)&skb->dst); |
1993 | goto done; | |
1994 | ||
1995 | no_route: | |
1996 | RT_CACHE_STAT_INC(in_no_route); | |
1997 | spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); | |
1998 | res.type = RTN_UNREACHABLE; | |
7f53878d MC |
1999 | if (err == -ESRCH) |
2000 | err = -ENETUNREACH; | |
1da177e4 LT |
2001 | goto local_input; |
2002 | ||
2003 | /* | |
2004 | * Do not cache martian addresses: they should be logged (RFC1812) | |
2005 | */ | |
2006 | martian_destination: | |
2007 | RT_CACHE_STAT_INC(in_martian_dst); | |
2008 | #ifdef CONFIG_IP_ROUTE_VERBOSE | |
2009 | if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) | |
2010 | printk(KERN_WARNING "martian destination %u.%u.%u.%u from " | |
2011 | "%u.%u.%u.%u, dev %s\n", | |
2012 | NIPQUAD(daddr), NIPQUAD(saddr), dev->name); | |
2013 | #endif | |
2c2910a4 DE |
2014 | |
2015 | e_hostunreach: | |
e905a9ed YH |
2016 | err = -EHOSTUNREACH; |
2017 | goto done; | |
2c2910a4 | 2018 | |
1da177e4 LT |
2019 | e_inval: |
2020 | err = -EINVAL; | |
2021 | goto done; | |
2022 | ||
2023 | e_nobufs: | |
2024 | err = -ENOBUFS; | |
2025 | goto done; | |
2026 | ||
2027 | martian_source: | |
2028 | ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); | |
2029 | goto e_inval; | |
2030 | } | |
2031 | ||
9e12bb22 | 2032 | int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr, |
1da177e4 LT |
2033 | u8 tos, struct net_device *dev) |
2034 | { | |
2035 | struct rtable * rth; | |
2036 | unsigned hash; | |
2037 | int iif = dev->ifindex; | |
2038 | ||
2039 | tos &= IPTOS_RT_MASK; | |
8c7bc840 | 2040 | hash = rt_hash(daddr, saddr, iif); |
1da177e4 LT |
2041 | |
2042 | rcu_read_lock(); | |
2043 | for (rth = rcu_dereference(rt_hash_table[hash].chain); rth; | |
093c2ca4 | 2044 | rth = rcu_dereference(rth->u.dst.rt_next)) { |
1da177e4 LT |
2045 | if (rth->fl.fl4_dst == daddr && |
2046 | rth->fl.fl4_src == saddr && | |
2047 | rth->fl.iif == iif && | |
2048 | rth->fl.oif == 0 && | |
47dcf0cb | 2049 | rth->fl.mark == skb->mark && |
1da177e4 | 2050 | rth->fl.fl4_tos == tos) { |
03f49f34 | 2051 | dst_use(&rth->u.dst, jiffies); |
1da177e4 LT |
2052 | RT_CACHE_STAT_INC(in_hit); |
2053 | rcu_read_unlock(); | |
2054 | skb->dst = (struct dst_entry*)rth; | |
2055 | return 0; | |
2056 | } | |
2057 | RT_CACHE_STAT_INC(in_hlist_search); | |
2058 | } | |
2059 | rcu_read_unlock(); | |
2060 | ||
2061 | /* Multicast recognition logic is moved from route cache to here. | |
2062 | The problem was that too many Ethernet cards have broken/missing | |
2063 | hardware multicast filters :-( As result the host on multicasting | |
2064 | network acquires a lot of useless route cache entries, sort of | |
2065 | SDR messages from all the world. Now we try to get rid of them. | |
2066 | Really, provided software IP multicast filter is organized | |
2067 | reasonably (at least, hashed), it does not result in a slowdown | |
2068 | comparing with route cache reject entries. | |
2069 | Note, that multicast routers are not affected, because | |
2070 | route cache entry is created eventually. | |
2071 | */ | |
2072 | if (MULTICAST(daddr)) { | |
2073 | struct in_device *in_dev; | |
2074 | ||
2075 | rcu_read_lock(); | |
e5ed6399 | 2076 | if ((in_dev = __in_dev_get_rcu(dev)) != NULL) { |
1da177e4 | 2077 | int our = ip_check_mc(in_dev, daddr, saddr, |
eddc9ec5 | 2078 | ip_hdr(skb)->protocol); |
1da177e4 LT |
2079 | if (our |
2080 | #ifdef CONFIG_IP_MROUTE | |
2081 | || (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev)) | |
2082 | #endif | |
2083 | ) { | |
2084 | rcu_read_unlock(); | |
2085 | return ip_route_input_mc(skb, daddr, saddr, | |
2086 | tos, dev, our); | |
2087 | } | |
2088 | } | |
2089 | rcu_read_unlock(); | |
2090 | return -EINVAL; | |
2091 | } | |
2092 | return ip_route_input_slow(skb, daddr, saddr, tos, dev); | |
2093 | } | |
2094 | ||
2095 | static inline int __mkroute_output(struct rtable **result, | |
e905a9ed | 2096 | struct fib_result* res, |
1da177e4 | 2097 | const struct flowi *fl, |
e905a9ed YH |
2098 | const struct flowi *oldflp, |
2099 | struct net_device *dev_out, | |
2100 | unsigned flags) | |
1da177e4 LT |
2101 | { |
2102 | struct rtable *rth; | |
2103 | struct in_device *in_dev; | |
2104 | u32 tos = RT_FL_TOS(oldflp); | |
2105 | int err = 0; | |
2106 | ||
2107 | if (LOOPBACK(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK)) | |
2108 | return -EINVAL; | |
2109 | ||
e448515c | 2110 | if (fl->fl4_dst == htonl(0xFFFFFFFF)) |
1da177e4 LT |
2111 | res->type = RTN_BROADCAST; |
2112 | else if (MULTICAST(fl->fl4_dst)) | |
2113 | res->type = RTN_MULTICAST; | |
2114 | else if (BADCLASS(fl->fl4_dst) || ZERONET(fl->fl4_dst)) | |
2115 | return -EINVAL; | |
2116 | ||
2117 | if (dev_out->flags & IFF_LOOPBACK) | |
2118 | flags |= RTCF_LOCAL; | |
2119 | ||
2120 | /* get work reference to inet device */ | |
2121 | in_dev = in_dev_get(dev_out); | |
2122 | if (!in_dev) | |
2123 | return -EINVAL; | |
2124 | ||
2125 | if (res->type == RTN_BROADCAST) { | |
2126 | flags |= RTCF_BROADCAST | RTCF_LOCAL; | |
2127 | if (res->fi) { | |
2128 | fib_info_put(res->fi); | |
2129 | res->fi = NULL; | |
2130 | } | |
2131 | } else if (res->type == RTN_MULTICAST) { | |
2132 | flags |= RTCF_MULTICAST|RTCF_LOCAL; | |
e905a9ed | 2133 | if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src, |
1da177e4 LT |
2134 | oldflp->proto)) |
2135 | flags &= ~RTCF_LOCAL; | |
2136 | /* If multicast route do not exist use | |
2137 | default one, but do not gateway in this case. | |
2138 | Yes, it is hack. | |
2139 | */ | |
2140 | if (res->fi && res->prefixlen < 4) { | |
2141 | fib_info_put(res->fi); | |
2142 | res->fi = NULL; | |
2143 | } | |
2144 | } | |
2145 | ||
2146 | ||
2147 | rth = dst_alloc(&ipv4_dst_ops); | |
2148 | if (!rth) { | |
2149 | err = -ENOBUFS; | |
2150 | goto cleanup; | |
e905a9ed | 2151 | } |
1da177e4 | 2152 | |
ce723d8e | 2153 | atomic_set(&rth->u.dst.__refcnt, 1); |
1da177e4 | 2154 | rth->u.dst.flags= DST_HOST; |
42f811b8 | 2155 | if (IN_DEV_CONF_GET(in_dev, NOXFRM)) |
1da177e4 | 2156 | rth->u.dst.flags |= DST_NOXFRM; |
42f811b8 | 2157 | if (IN_DEV_CONF_GET(in_dev, NOPOLICY)) |
1da177e4 LT |
2158 | rth->u.dst.flags |= DST_NOPOLICY; |
2159 | ||
2160 | rth->fl.fl4_dst = oldflp->fl4_dst; | |
2161 | rth->fl.fl4_tos = tos; | |
2162 | rth->fl.fl4_src = oldflp->fl4_src; | |
2163 | rth->fl.oif = oldflp->oif; | |
47dcf0cb | 2164 | rth->fl.mark = oldflp->mark; |
1da177e4 LT |
2165 | rth->rt_dst = fl->fl4_dst; |
2166 | rth->rt_src = fl->fl4_src; | |
2167 | rth->rt_iif = oldflp->oif ? : dev_out->ifindex; | |
e905a9ed | 2168 | /* get references to the devices that are to be hold by the routing |
1da177e4 LT |
2169 | cache entry */ |
2170 | rth->u.dst.dev = dev_out; | |
2171 | dev_hold(dev_out); | |
2172 | rth->idev = in_dev_get(dev_out); | |
2173 | rth->rt_gateway = fl->fl4_dst; | |
2174 | rth->rt_spec_dst= fl->fl4_src; | |
2175 | ||
2176 | rth->u.dst.output=ip_output; | |
2177 | ||
2178 | RT_CACHE_STAT_INC(out_slow_tot); | |
2179 | ||
2180 | if (flags & RTCF_LOCAL) { | |
2181 | rth->u.dst.input = ip_local_deliver; | |
2182 | rth->rt_spec_dst = fl->fl4_dst; | |
2183 | } | |
2184 | if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { | |
2185 | rth->rt_spec_dst = fl->fl4_src; | |
e905a9ed | 2186 | if (flags & RTCF_LOCAL && |
1da177e4 LT |
2187 | !(dev_out->flags & IFF_LOOPBACK)) { |
2188 | rth->u.dst.output = ip_mc_output; | |
2189 | RT_CACHE_STAT_INC(out_slow_mc); | |
2190 | } | |
2191 | #ifdef CONFIG_IP_MROUTE | |
2192 | if (res->type == RTN_MULTICAST) { | |
2193 | if (IN_DEV_MFORWARD(in_dev) && | |
2194 | !LOCAL_MCAST(oldflp->fl4_dst)) { | |
2195 | rth->u.dst.input = ip_mr_input; | |
2196 | rth->u.dst.output = ip_mc_output; | |
2197 | } | |
2198 | } | |
2199 | #endif | |
2200 | } | |
2201 | ||
2202 | rt_set_nexthop(rth, res, 0); | |
2203 | ||
2204 | rth->rt_flags = flags; | |
2205 | ||
2206 | *result = rth; | |
2207 | cleanup: | |
2208 | /* release work reference to inet device */ | |
2209 | in_dev_put(in_dev); | |
2210 | ||
2211 | return err; | |
2212 | } | |
2213 | ||
e06e7c61 DM |
2214 | static inline int ip_mkroute_output(struct rtable **rp, |
2215 | struct fib_result* res, | |
2216 | const struct flowi *fl, | |
2217 | const struct flowi *oldflp, | |
2218 | struct net_device *dev_out, | |
2219 | unsigned flags) | |
1da177e4 | 2220 | { |
7abaa27c | 2221 | struct rtable *rth = NULL; |
1da177e4 LT |
2222 | int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags); |
2223 | unsigned hash; | |
2224 | if (err == 0) { | |
8c7bc840 | 2225 | hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif); |
1da177e4 LT |
2226 | err = rt_intern_hash(hash, rth, rp); |
2227 | } | |
e905a9ed | 2228 | |
1da177e4 LT |
2229 | return err; |
2230 | } | |
2231 | ||
1da177e4 LT |
2232 | /* |
2233 | * Major route resolver routine. | |
2234 | */ | |
2235 | ||
2236 | static int ip_route_output_slow(struct rtable **rp, const struct flowi *oldflp) | |
2237 | { | |
2238 | u32 tos = RT_FL_TOS(oldflp); | |
2239 | struct flowi fl = { .nl_u = { .ip4_u = | |
2240 | { .daddr = oldflp->fl4_dst, | |
2241 | .saddr = oldflp->fl4_src, | |
2242 | .tos = tos & IPTOS_RT_MASK, | |
2243 | .scope = ((tos & RTO_ONLINK) ? | |
2244 | RT_SCOPE_LINK : | |
2245 | RT_SCOPE_UNIVERSE), | |
1da177e4 | 2246 | } }, |
47dcf0cb | 2247 | .mark = oldflp->mark, |
2774c7ab | 2248 | .iif = init_net.loopback_dev->ifindex, |
1da177e4 LT |
2249 | .oif = oldflp->oif }; |
2250 | struct fib_result res; | |
2251 | unsigned flags = 0; | |
2252 | struct net_device *dev_out = NULL; | |
2253 | int free_res = 0; | |
2254 | int err; | |
2255 | ||
2256 | ||
2257 | res.fi = NULL; | |
2258 | #ifdef CONFIG_IP_MULTIPLE_TABLES | |
2259 | res.r = NULL; | |
2260 | #endif | |
2261 | ||
2262 | if (oldflp->fl4_src) { | |
2263 | err = -EINVAL; | |
2264 | if (MULTICAST(oldflp->fl4_src) || | |
2265 | BADCLASS(oldflp->fl4_src) || | |
2266 | ZERONET(oldflp->fl4_src)) | |
2267 | goto out; | |
2268 | ||
2269 | /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ | |
2270 | dev_out = ip_dev_find(oldflp->fl4_src); | |
f6c5d736 | 2271 | if (dev_out == NULL) |
1da177e4 LT |
2272 | goto out; |
2273 | ||
2274 | /* I removed check for oif == dev_out->oif here. | |
2275 | It was wrong for two reasons: | |
2276 | 1. ip_dev_find(saddr) can return wrong iface, if saddr is | |
2277 | assigned to multiple interfaces. | |
2278 | 2. Moreover, we are allowed to send packets with saddr | |
2279 | of another iface. --ANK | |
2280 | */ | |
2281 | ||
f6c5d736 | 2282 | if (oldflp->oif == 0 |
e448515c | 2283 | && (MULTICAST(oldflp->fl4_dst) || oldflp->fl4_dst == htonl(0xFFFFFFFF))) { |
1da177e4 LT |
2284 | /* Special hack: user can direct multicasts |
2285 | and limited broadcast via necessary interface | |
2286 | without fiddling with IP_MULTICAST_IF or IP_PKTINFO. | |
2287 | This hack is not just for fun, it allows | |
2288 | vic,vat and friends to work. | |
2289 | They bind socket to loopback, set ttl to zero | |
2290 | and expect that it will work. | |
2291 | From the viewpoint of routing cache they are broken, | |
2292 | because we are not allowed to build multicast path | |
2293 | with loopback source addr (look, routing cache | |
2294 | cannot know, that ttl is zero, so that packet | |
2295 | will not leave this host and route is valid). | |
2296 | Luckily, this hack is good workaround. | |
2297 | */ | |
2298 | ||
2299 | fl.oif = dev_out->ifindex; | |
2300 | goto make_route; | |
2301 | } | |
2302 | if (dev_out) | |
2303 | dev_put(dev_out); | |
2304 | dev_out = NULL; | |
2305 | } | |
2306 | ||
2307 | ||
2308 | if (oldflp->oif) { | |
881d966b | 2309 | dev_out = dev_get_by_index(&init_net, oldflp->oif); |
1da177e4 LT |
2310 | err = -ENODEV; |
2311 | if (dev_out == NULL) | |
2312 | goto out; | |
e5ed6399 HX |
2313 | |
2314 | /* RACE: Check return value of inet_select_addr instead. */ | |
2315 | if (__in_dev_get_rtnl(dev_out) == NULL) { | |
1da177e4 LT |
2316 | dev_put(dev_out); |
2317 | goto out; /* Wrong error code */ | |
2318 | } | |
2319 | ||
e448515c | 2320 | if (LOCAL_MCAST(oldflp->fl4_dst) || oldflp->fl4_dst == htonl(0xFFFFFFFF)) { |
1da177e4 LT |
2321 | if (!fl.fl4_src) |
2322 | fl.fl4_src = inet_select_addr(dev_out, 0, | |
2323 | RT_SCOPE_LINK); | |
2324 | goto make_route; | |
2325 | } | |
2326 | if (!fl.fl4_src) { | |
2327 | if (MULTICAST(oldflp->fl4_dst)) | |
2328 | fl.fl4_src = inet_select_addr(dev_out, 0, | |
2329 | fl.fl4_scope); | |
2330 | else if (!oldflp->fl4_dst) | |
2331 | fl.fl4_src = inet_select_addr(dev_out, 0, | |
2332 | RT_SCOPE_HOST); | |
2333 | } | |
2334 | } | |
2335 | ||
2336 | if (!fl.fl4_dst) { | |
2337 | fl.fl4_dst = fl.fl4_src; | |
2338 | if (!fl.fl4_dst) | |
2339 | fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK); | |
2340 | if (dev_out) | |
2341 | dev_put(dev_out); | |
2774c7ab | 2342 | dev_out = init_net.loopback_dev; |
1da177e4 | 2343 | dev_hold(dev_out); |
2774c7ab | 2344 | fl.oif = init_net.loopback_dev->ifindex; |
1da177e4 LT |
2345 | res.type = RTN_LOCAL; |
2346 | flags |= RTCF_LOCAL; | |
2347 | goto make_route; | |
2348 | } | |
2349 | ||
2350 | if (fib_lookup(&fl, &res)) { | |
2351 | res.fi = NULL; | |
2352 | if (oldflp->oif) { | |
2353 | /* Apparently, routing tables are wrong. Assume, | |
2354 | that the destination is on link. | |
2355 | ||
2356 | WHY? DW. | |
2357 | Because we are allowed to send to iface | |
2358 | even if it has NO routes and NO assigned | |
2359 | addresses. When oif is specified, routing | |
2360 | tables are looked up with only one purpose: | |
2361 | to catch if destination is gatewayed, rather than | |
2362 | direct. Moreover, if MSG_DONTROUTE is set, | |
2363 | we send packet, ignoring both routing tables | |
2364 | and ifaddr state. --ANK | |
2365 | ||
2366 | ||
2367 | We could make it even if oif is unknown, | |
2368 | likely IPv6, but we do not. | |
2369 | */ | |
2370 | ||
2371 | if (fl.fl4_src == 0) | |
2372 | fl.fl4_src = inet_select_addr(dev_out, 0, | |
2373 | RT_SCOPE_LINK); | |
2374 | res.type = RTN_UNICAST; | |
2375 | goto make_route; | |
2376 | } | |
2377 | if (dev_out) | |
2378 | dev_put(dev_out); | |
2379 | err = -ENETUNREACH; | |
2380 | goto out; | |
2381 | } | |
2382 | free_res = 1; | |
2383 | ||
2384 | if (res.type == RTN_LOCAL) { | |
2385 | if (!fl.fl4_src) | |
2386 | fl.fl4_src = fl.fl4_dst; | |
2387 | if (dev_out) | |
2388 | dev_put(dev_out); | |
2774c7ab | 2389 | dev_out = init_net.loopback_dev; |
1da177e4 LT |
2390 | dev_hold(dev_out); |
2391 | fl.oif = dev_out->ifindex; | |
2392 | if (res.fi) | |
2393 | fib_info_put(res.fi); | |
2394 | res.fi = NULL; | |
2395 | flags |= RTCF_LOCAL; | |
2396 | goto make_route; | |
2397 | } | |
2398 | ||
2399 | #ifdef CONFIG_IP_ROUTE_MULTIPATH | |
2400 | if (res.fi->fib_nhs > 1 && fl.oif == 0) | |
2401 | fib_select_multipath(&fl, &res); | |
2402 | else | |
2403 | #endif | |
2404 | if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif) | |
2405 | fib_select_default(&fl, &res); | |
2406 | ||
2407 | if (!fl.fl4_src) | |
2408 | fl.fl4_src = FIB_RES_PREFSRC(res); | |
2409 | ||
2410 | if (dev_out) | |
2411 | dev_put(dev_out); | |
2412 | dev_out = FIB_RES_DEV(res); | |
2413 | dev_hold(dev_out); | |
2414 | fl.oif = dev_out->ifindex; | |
2415 | ||
2416 | ||
2417 | make_route: | |
2418 | err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags); | |
2419 | ||
2420 | ||
2421 | if (free_res) | |
2422 | fib_res_put(&res); | |
2423 | if (dev_out) | |
2424 | dev_put(dev_out); | |
2425 | out: return err; | |
2426 | } | |
2427 | ||
2428 | int __ip_route_output_key(struct rtable **rp, const struct flowi *flp) | |
2429 | { | |
2430 | unsigned hash; | |
2431 | struct rtable *rth; | |
2432 | ||
8c7bc840 | 2433 | hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif); |
1da177e4 LT |
2434 | |
2435 | rcu_read_lock_bh(); | |
2436 | for (rth = rcu_dereference(rt_hash_table[hash].chain); rth; | |
093c2ca4 | 2437 | rth = rcu_dereference(rth->u.dst.rt_next)) { |
1da177e4 LT |
2438 | if (rth->fl.fl4_dst == flp->fl4_dst && |
2439 | rth->fl.fl4_src == flp->fl4_src && | |
2440 | rth->fl.iif == 0 && | |
2441 | rth->fl.oif == flp->oif && | |
47dcf0cb | 2442 | rth->fl.mark == flp->mark && |
1da177e4 LT |
2443 | !((rth->fl.fl4_tos ^ flp->fl4_tos) & |
2444 | (IPTOS_RT_MASK | RTO_ONLINK))) { | |
03f49f34 | 2445 | dst_use(&rth->u.dst, jiffies); |
1da177e4 LT |
2446 | RT_CACHE_STAT_INC(out_hit); |
2447 | rcu_read_unlock_bh(); | |
2448 | *rp = rth; | |
2449 | return 0; | |
2450 | } | |
2451 | RT_CACHE_STAT_INC(out_hlist_search); | |
2452 | } | |
2453 | rcu_read_unlock_bh(); | |
2454 | ||
2455 | return ip_route_output_slow(rp, flp); | |
2456 | } | |
2457 | ||
d8c97a94 ACM |
2458 | EXPORT_SYMBOL_GPL(__ip_route_output_key); |
2459 | ||
14e50e57 DM |
2460 | static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu) |
2461 | { | |
2462 | } | |
2463 | ||
2464 | static struct dst_ops ipv4_dst_blackhole_ops = { | |
2465 | .family = AF_INET, | |
2466 | .protocol = __constant_htons(ETH_P_IP), | |
2467 | .destroy = ipv4_dst_destroy, | |
2468 | .check = ipv4_dst_check, | |
2469 | .update_pmtu = ipv4_rt_blackhole_update_pmtu, | |
2470 | .entry_size = sizeof(struct rtable), | |
2471 | }; | |
2472 | ||
2473 | ||
14e50e57 DM |
2474 | static int ipv4_dst_blackhole(struct rtable **rp, struct flowi *flp, struct sock *sk) |
2475 | { | |
2476 | struct rtable *ort = *rp; | |
2477 | struct rtable *rt = (struct rtable *) | |
2478 | dst_alloc(&ipv4_dst_blackhole_ops); | |
2479 | ||
2480 | if (rt) { | |
2481 | struct dst_entry *new = &rt->u.dst; | |
2482 | ||
2483 | atomic_set(&new->__refcnt, 1); | |
2484 | new->__use = 1; | |
352e512c HX |
2485 | new->input = dst_discard; |
2486 | new->output = dst_discard; | |
14e50e57 DM |
2487 | memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32)); |
2488 | ||
2489 | new->dev = ort->u.dst.dev; | |
2490 | if (new->dev) | |
2491 | dev_hold(new->dev); | |
2492 | ||
2493 | rt->fl = ort->fl; | |
2494 | ||
2495 | rt->idev = ort->idev; | |
2496 | if (rt->idev) | |
2497 | in_dev_hold(rt->idev); | |
2498 | rt->rt_flags = ort->rt_flags; | |
2499 | rt->rt_type = ort->rt_type; | |
2500 | rt->rt_dst = ort->rt_dst; | |
2501 | rt->rt_src = ort->rt_src; | |
2502 | rt->rt_iif = ort->rt_iif; | |
2503 | rt->rt_gateway = ort->rt_gateway; | |
2504 | rt->rt_spec_dst = ort->rt_spec_dst; | |
2505 | rt->peer = ort->peer; | |
2506 | if (rt->peer) | |
2507 | atomic_inc(&rt->peer->refcnt); | |
2508 | ||
2509 | dst_free(new); | |
2510 | } | |
2511 | ||
2512 | dst_release(&(*rp)->u.dst); | |
2513 | *rp = rt; | |
2514 | return (rt ? 0 : -ENOMEM); | |
2515 | } | |
2516 | ||
1da177e4 LT |
2517 | int ip_route_output_flow(struct rtable **rp, struct flowi *flp, struct sock *sk, int flags) |
2518 | { | |
2519 | int err; | |
2520 | ||
2521 | if ((err = __ip_route_output_key(rp, flp)) != 0) | |
2522 | return err; | |
2523 | ||
2524 | if (flp->proto) { | |
2525 | if (!flp->fl4_src) | |
2526 | flp->fl4_src = (*rp)->rt_src; | |
2527 | if (!flp->fl4_dst) | |
2528 | flp->fl4_dst = (*rp)->rt_dst; | |
14e50e57 DM |
2529 | err = __xfrm_lookup((struct dst_entry **)rp, flp, sk, flags); |
2530 | if (err == -EREMOTE) | |
2531 | err = ipv4_dst_blackhole(rp, flp, sk); | |
2532 | ||
2533 | return err; | |
1da177e4 LT |
2534 | } |
2535 | ||
2536 | return 0; | |
2537 | } | |
2538 | ||
d8c97a94 ACM |
2539 | EXPORT_SYMBOL_GPL(ip_route_output_flow); |
2540 | ||
1da177e4 LT |
2541 | int ip_route_output_key(struct rtable **rp, struct flowi *flp) |
2542 | { | |
2543 | return ip_route_output_flow(rp, flp, NULL, 0); | |
2544 | } | |
2545 | ||
2546 | static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event, | |
b6544c0b | 2547 | int nowait, unsigned int flags) |
1da177e4 LT |
2548 | { |
2549 | struct rtable *rt = (struct rtable*)skb->dst; | |
2550 | struct rtmsg *r; | |
be403ea1 | 2551 | struct nlmsghdr *nlh; |
e3703b3d TG |
2552 | long expires; |
2553 | u32 id = 0, ts = 0, tsage = 0, error; | |
be403ea1 TG |
2554 | |
2555 | nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags); | |
2556 | if (nlh == NULL) | |
26932566 | 2557 | return -EMSGSIZE; |
be403ea1 TG |
2558 | |
2559 | r = nlmsg_data(nlh); | |
1da177e4 LT |
2560 | r->rtm_family = AF_INET; |
2561 | r->rtm_dst_len = 32; | |
2562 | r->rtm_src_len = 0; | |
2563 | r->rtm_tos = rt->fl.fl4_tos; | |
2564 | r->rtm_table = RT_TABLE_MAIN; | |
be403ea1 | 2565 | NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN); |
1da177e4 LT |
2566 | r->rtm_type = rt->rt_type; |
2567 | r->rtm_scope = RT_SCOPE_UNIVERSE; | |
2568 | r->rtm_protocol = RTPROT_UNSPEC; | |
2569 | r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; | |
2570 | if (rt->rt_flags & RTCF_NOTIFY) | |
2571 | r->rtm_flags |= RTM_F_NOTIFY; | |
be403ea1 | 2572 | |
17fb2c64 | 2573 | NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst); |
be403ea1 | 2574 | |
1da177e4 LT |
2575 | if (rt->fl.fl4_src) { |
2576 | r->rtm_src_len = 32; | |
17fb2c64 | 2577 | NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src); |
1da177e4 LT |
2578 | } |
2579 | if (rt->u.dst.dev) | |
be403ea1 | 2580 | NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex); |
1da177e4 LT |
2581 | #ifdef CONFIG_NET_CLS_ROUTE |
2582 | if (rt->u.dst.tclassid) | |
be403ea1 | 2583 | NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid); |
1da177e4 LT |
2584 | #endif |
2585 | if (rt->fl.iif) | |
17fb2c64 | 2586 | NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst); |
1da177e4 | 2587 | else if (rt->rt_src != rt->fl.fl4_src) |
17fb2c64 | 2588 | NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src); |
be403ea1 | 2589 | |
1da177e4 | 2590 | if (rt->rt_dst != rt->rt_gateway) |
17fb2c64 | 2591 | NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway); |
be403ea1 | 2592 | |
1da177e4 | 2593 | if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0) |
be403ea1 TG |
2594 | goto nla_put_failure; |
2595 | ||
e3703b3d TG |
2596 | error = rt->u.dst.error; |
2597 | expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0; | |
1da177e4 | 2598 | if (rt->peer) { |
e3703b3d | 2599 | id = rt->peer->ip_id_count; |
1da177e4 | 2600 | if (rt->peer->tcp_ts_stamp) { |
e3703b3d | 2601 | ts = rt->peer->tcp_ts; |
9d729f72 | 2602 | tsage = get_seconds() - rt->peer->tcp_ts_stamp; |
1da177e4 LT |
2603 | } |
2604 | } | |
be403ea1 | 2605 | |
1da177e4 LT |
2606 | if (rt->fl.iif) { |
2607 | #ifdef CONFIG_IP_MROUTE | |
e448515c | 2608 | __be32 dst = rt->rt_dst; |
1da177e4 LT |
2609 | |
2610 | if (MULTICAST(dst) && !LOCAL_MCAST(dst) && | |
42f811b8 | 2611 | IPV4_DEVCONF_ALL(MC_FORWARDING)) { |
1da177e4 LT |
2612 | int err = ipmr_get_route(skb, r, nowait); |
2613 | if (err <= 0) { | |
2614 | if (!nowait) { | |
2615 | if (err == 0) | |
2616 | return 0; | |
be403ea1 | 2617 | goto nla_put_failure; |
1da177e4 LT |
2618 | } else { |
2619 | if (err == -EMSGSIZE) | |
be403ea1 | 2620 | goto nla_put_failure; |
e3703b3d | 2621 | error = err; |
1da177e4 LT |
2622 | } |
2623 | } | |
2624 | } else | |
2625 | #endif | |
be403ea1 | 2626 | NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif); |
1da177e4 LT |
2627 | } |
2628 | ||
e3703b3d TG |
2629 | if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage, |
2630 | expires, error) < 0) | |
2631 | goto nla_put_failure; | |
be403ea1 TG |
2632 | |
2633 | return nlmsg_end(skb, nlh); | |
1da177e4 | 2634 | |
be403ea1 | 2635 | nla_put_failure: |
26932566 PM |
2636 | nlmsg_cancel(skb, nlh); |
2637 | return -EMSGSIZE; | |
1da177e4 LT |
2638 | } |
2639 | ||
63f3444f | 2640 | static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg) |
1da177e4 | 2641 | { |
b854272b | 2642 | struct net *net = in_skb->sk->sk_net; |
d889ce3b TG |
2643 | struct rtmsg *rtm; |
2644 | struct nlattr *tb[RTA_MAX+1]; | |
1da177e4 | 2645 | struct rtable *rt = NULL; |
9e12bb22 AV |
2646 | __be32 dst = 0; |
2647 | __be32 src = 0; | |
2648 | u32 iif; | |
d889ce3b | 2649 | int err; |
1da177e4 LT |
2650 | struct sk_buff *skb; |
2651 | ||
b854272b DL |
2652 | if (net != &init_net) |
2653 | return -EINVAL; | |
2654 | ||
d889ce3b TG |
2655 | err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy); |
2656 | if (err < 0) | |
2657 | goto errout; | |
2658 | ||
2659 | rtm = nlmsg_data(nlh); | |
2660 | ||
1da177e4 | 2661 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); |
d889ce3b TG |
2662 | if (skb == NULL) { |
2663 | err = -ENOBUFS; | |
2664 | goto errout; | |
2665 | } | |
1da177e4 LT |
2666 | |
2667 | /* Reserve room for dummy headers, this skb can pass | |
2668 | through good chunk of routing engine. | |
2669 | */ | |
459a98ed | 2670 | skb_reset_mac_header(skb); |
c1d2bbe1 | 2671 | skb_reset_network_header(skb); |
d2c962b8 SH |
2672 | |
2673 | /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */ | |
eddc9ec5 | 2674 | ip_hdr(skb)->protocol = IPPROTO_ICMP; |
1da177e4 LT |
2675 | skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr)); |
2676 | ||
17fb2c64 AV |
2677 | src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0; |
2678 | dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0; | |
d889ce3b | 2679 | iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; |
1da177e4 LT |
2680 | |
2681 | if (iif) { | |
d889ce3b TG |
2682 | struct net_device *dev; |
2683 | ||
881d966b | 2684 | dev = __dev_get_by_index(&init_net, iif); |
d889ce3b TG |
2685 | if (dev == NULL) { |
2686 | err = -ENODEV; | |
2687 | goto errout_free; | |
2688 | } | |
2689 | ||
1da177e4 LT |
2690 | skb->protocol = htons(ETH_P_IP); |
2691 | skb->dev = dev; | |
2692 | local_bh_disable(); | |
2693 | err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev); | |
2694 | local_bh_enable(); | |
d889ce3b TG |
2695 | |
2696 | rt = (struct rtable*) skb->dst; | |
2697 | if (err == 0 && rt->u.dst.error) | |
1da177e4 LT |
2698 | err = -rt->u.dst.error; |
2699 | } else { | |
d889ce3b TG |
2700 | struct flowi fl = { |
2701 | .nl_u = { | |
2702 | .ip4_u = { | |
2703 | .daddr = dst, | |
2704 | .saddr = src, | |
2705 | .tos = rtm->rtm_tos, | |
2706 | }, | |
2707 | }, | |
2708 | .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0, | |
2709 | }; | |
1da177e4 LT |
2710 | err = ip_route_output_key(&rt, &fl); |
2711 | } | |
d889ce3b | 2712 | |
1da177e4 | 2713 | if (err) |
d889ce3b | 2714 | goto errout_free; |
1da177e4 LT |
2715 | |
2716 | skb->dst = &rt->u.dst; | |
2717 | if (rtm->rtm_flags & RTM_F_NOTIFY) | |
2718 | rt->rt_flags |= RTCF_NOTIFY; | |
2719 | ||
1da177e4 | 2720 | err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, |
b6544c0b | 2721 | RTM_NEWROUTE, 0, 0); |
d889ce3b TG |
2722 | if (err <= 0) |
2723 | goto errout_free; | |
1da177e4 | 2724 | |
97c53cac | 2725 | err = rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid); |
d889ce3b | 2726 | errout: |
2942e900 | 2727 | return err; |
1da177e4 | 2728 | |
d889ce3b | 2729 | errout_free: |
1da177e4 | 2730 | kfree_skb(skb); |
d889ce3b | 2731 | goto errout; |
1da177e4 LT |
2732 | } |
2733 | ||
2734 | int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb) | |
2735 | { | |
2736 | struct rtable *rt; | |
2737 | int h, s_h; | |
2738 | int idx, s_idx; | |
2739 | ||
2740 | s_h = cb->args[0]; | |
d8c92830 ED |
2741 | if (s_h < 0) |
2742 | s_h = 0; | |
1da177e4 | 2743 | s_idx = idx = cb->args[1]; |
d8c92830 | 2744 | for (h = s_h; h <= rt_hash_mask; h++) { |
1da177e4 LT |
2745 | rcu_read_lock_bh(); |
2746 | for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt; | |
093c2ca4 | 2747 | rt = rcu_dereference(rt->u.dst.rt_next), idx++) { |
1da177e4 LT |
2748 | if (idx < s_idx) |
2749 | continue; | |
2750 | skb->dst = dst_clone(&rt->u.dst); | |
2751 | if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid, | |
e905a9ed | 2752 | cb->nlh->nlmsg_seq, RTM_NEWROUTE, |
b6544c0b | 2753 | 1, NLM_F_MULTI) <= 0) { |
1da177e4 LT |
2754 | dst_release(xchg(&skb->dst, NULL)); |
2755 | rcu_read_unlock_bh(); | |
2756 | goto done; | |
2757 | } | |
2758 | dst_release(xchg(&skb->dst, NULL)); | |
2759 | } | |
2760 | rcu_read_unlock_bh(); | |
d8c92830 | 2761 | s_idx = 0; |
1da177e4 LT |
2762 | } |
2763 | ||
2764 | done: | |
2765 | cb->args[0] = h; | |
2766 | cb->args[1] = idx; | |
2767 | return skb->len; | |
2768 | } | |
2769 | ||
2770 | void ip_rt_multicast_event(struct in_device *in_dev) | |
2771 | { | |
2772 | rt_cache_flush(0); | |
2773 | } | |
2774 | ||
2775 | #ifdef CONFIG_SYSCTL | |
2776 | static int flush_delay; | |
2777 | ||
2778 | static int ipv4_sysctl_rtcache_flush(ctl_table *ctl, int write, | |
2779 | struct file *filp, void __user *buffer, | |
2780 | size_t *lenp, loff_t *ppos) | |
2781 | { | |
2782 | if (write) { | |
2783 | proc_dointvec(ctl, write, filp, buffer, lenp, ppos); | |
2784 | rt_cache_flush(flush_delay); | |
2785 | return 0; | |
e905a9ed | 2786 | } |
1da177e4 LT |
2787 | |
2788 | return -EINVAL; | |
2789 | } | |
2790 | ||
2791 | static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table, | |
2792 | int __user *name, | |
2793 | int nlen, | |
2794 | void __user *oldval, | |
2795 | size_t __user *oldlenp, | |
2796 | void __user *newval, | |
1f29bcd7 | 2797 | size_t newlen) |
1da177e4 LT |
2798 | { |
2799 | int delay; | |
2800 | if (newlen != sizeof(int)) | |
2801 | return -EINVAL; | |
2802 | if (get_user(delay, (int __user *)newval)) | |
e905a9ed YH |
2803 | return -EFAULT; |
2804 | rt_cache_flush(delay); | |
1da177e4 LT |
2805 | return 0; |
2806 | } | |
2807 | ||
2808 | ctl_table ipv4_route_table[] = { | |
e905a9ed | 2809 | { |
1da177e4 LT |
2810 | .ctl_name = NET_IPV4_ROUTE_FLUSH, |
2811 | .procname = "flush", | |
2812 | .data = &flush_delay, | |
2813 | .maxlen = sizeof(int), | |
7e3e0360 | 2814 | .mode = 0200, |
1da177e4 LT |
2815 | .proc_handler = &ipv4_sysctl_rtcache_flush, |
2816 | .strategy = &ipv4_sysctl_rtcache_flush_strategy, | |
2817 | }, | |
2818 | { | |
2819 | .ctl_name = NET_IPV4_ROUTE_MIN_DELAY, | |
2820 | .procname = "min_delay", | |
2821 | .data = &ip_rt_min_delay, | |
2822 | .maxlen = sizeof(int), | |
2823 | .mode = 0644, | |
2824 | .proc_handler = &proc_dointvec_jiffies, | |
2825 | .strategy = &sysctl_jiffies, | |
2826 | }, | |
2827 | { | |
2828 | .ctl_name = NET_IPV4_ROUTE_MAX_DELAY, | |
2829 | .procname = "max_delay", | |
2830 | .data = &ip_rt_max_delay, | |
2831 | .maxlen = sizeof(int), | |
2832 | .mode = 0644, | |
2833 | .proc_handler = &proc_dointvec_jiffies, | |
2834 | .strategy = &sysctl_jiffies, | |
2835 | }, | |
2836 | { | |
2837 | .ctl_name = NET_IPV4_ROUTE_GC_THRESH, | |
2838 | .procname = "gc_thresh", | |
2839 | .data = &ipv4_dst_ops.gc_thresh, | |
2840 | .maxlen = sizeof(int), | |
2841 | .mode = 0644, | |
2842 | .proc_handler = &proc_dointvec, | |
2843 | }, | |
2844 | { | |
2845 | .ctl_name = NET_IPV4_ROUTE_MAX_SIZE, | |
2846 | .procname = "max_size", | |
2847 | .data = &ip_rt_max_size, | |
2848 | .maxlen = sizeof(int), | |
2849 | .mode = 0644, | |
2850 | .proc_handler = &proc_dointvec, | |
2851 | }, | |
2852 | { | |
2853 | /* Deprecated. Use gc_min_interval_ms */ | |
e905a9ed | 2854 | |
1da177e4 LT |
2855 | .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL, |
2856 | .procname = "gc_min_interval", | |
2857 | .data = &ip_rt_gc_min_interval, | |
2858 | .maxlen = sizeof(int), | |
2859 | .mode = 0644, | |
2860 | .proc_handler = &proc_dointvec_jiffies, | |
2861 | .strategy = &sysctl_jiffies, | |
2862 | }, | |
2863 | { | |
2864 | .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS, | |
2865 | .procname = "gc_min_interval_ms", | |
2866 | .data = &ip_rt_gc_min_interval, | |
2867 | .maxlen = sizeof(int), | |
2868 | .mode = 0644, | |
2869 | .proc_handler = &proc_dointvec_ms_jiffies, | |
2870 | .strategy = &sysctl_ms_jiffies, | |
2871 | }, | |
2872 | { | |
2873 | .ctl_name = NET_IPV4_ROUTE_GC_TIMEOUT, | |
2874 | .procname = "gc_timeout", | |
2875 | .data = &ip_rt_gc_timeout, | |
2876 | .maxlen = sizeof(int), | |
2877 | .mode = 0644, | |
2878 | .proc_handler = &proc_dointvec_jiffies, | |
2879 | .strategy = &sysctl_jiffies, | |
2880 | }, | |
2881 | { | |
2882 | .ctl_name = NET_IPV4_ROUTE_GC_INTERVAL, | |
2883 | .procname = "gc_interval", | |
2884 | .data = &ip_rt_gc_interval, | |
2885 | .maxlen = sizeof(int), | |
2886 | .mode = 0644, | |
2887 | .proc_handler = &proc_dointvec_jiffies, | |
2888 | .strategy = &sysctl_jiffies, | |
2889 | }, | |
2890 | { | |
2891 | .ctl_name = NET_IPV4_ROUTE_REDIRECT_LOAD, | |
2892 | .procname = "redirect_load", | |
2893 | .data = &ip_rt_redirect_load, | |
2894 | .maxlen = sizeof(int), | |
2895 | .mode = 0644, | |
2896 | .proc_handler = &proc_dointvec, | |
2897 | }, | |
2898 | { | |
2899 | .ctl_name = NET_IPV4_ROUTE_REDIRECT_NUMBER, | |
2900 | .procname = "redirect_number", | |
2901 | .data = &ip_rt_redirect_number, | |
2902 | .maxlen = sizeof(int), | |
2903 | .mode = 0644, | |
2904 | .proc_handler = &proc_dointvec, | |
2905 | }, | |
2906 | { | |
2907 | .ctl_name = NET_IPV4_ROUTE_REDIRECT_SILENCE, | |
2908 | .procname = "redirect_silence", | |
2909 | .data = &ip_rt_redirect_silence, | |
2910 | .maxlen = sizeof(int), | |
2911 | .mode = 0644, | |
2912 | .proc_handler = &proc_dointvec, | |
2913 | }, | |
2914 | { | |
2915 | .ctl_name = NET_IPV4_ROUTE_ERROR_COST, | |
2916 | .procname = "error_cost", | |
2917 | .data = &ip_rt_error_cost, | |
2918 | .maxlen = sizeof(int), | |
2919 | .mode = 0644, | |
2920 | .proc_handler = &proc_dointvec, | |
2921 | }, | |
2922 | { | |
2923 | .ctl_name = NET_IPV4_ROUTE_ERROR_BURST, | |
2924 | .procname = "error_burst", | |
2925 | .data = &ip_rt_error_burst, | |
2926 | .maxlen = sizeof(int), | |
2927 | .mode = 0644, | |
2928 | .proc_handler = &proc_dointvec, | |
2929 | }, | |
2930 | { | |
2931 | .ctl_name = NET_IPV4_ROUTE_GC_ELASTICITY, | |
2932 | .procname = "gc_elasticity", | |
2933 | .data = &ip_rt_gc_elasticity, | |
2934 | .maxlen = sizeof(int), | |
2935 | .mode = 0644, | |
2936 | .proc_handler = &proc_dointvec, | |
2937 | }, | |
2938 | { | |
2939 | .ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES, | |
2940 | .procname = "mtu_expires", | |
2941 | .data = &ip_rt_mtu_expires, | |
2942 | .maxlen = sizeof(int), | |
2943 | .mode = 0644, | |
2944 | .proc_handler = &proc_dointvec_jiffies, | |
2945 | .strategy = &sysctl_jiffies, | |
2946 | }, | |
2947 | { | |
2948 | .ctl_name = NET_IPV4_ROUTE_MIN_PMTU, | |
2949 | .procname = "min_pmtu", | |
2950 | .data = &ip_rt_min_pmtu, | |
2951 | .maxlen = sizeof(int), | |
2952 | .mode = 0644, | |
2953 | .proc_handler = &proc_dointvec, | |
2954 | }, | |
2955 | { | |
2956 | .ctl_name = NET_IPV4_ROUTE_MIN_ADVMSS, | |
2957 | .procname = "min_adv_mss", | |
2958 | .data = &ip_rt_min_advmss, | |
2959 | .maxlen = sizeof(int), | |
2960 | .mode = 0644, | |
2961 | .proc_handler = &proc_dointvec, | |
2962 | }, | |
2963 | { | |
2964 | .ctl_name = NET_IPV4_ROUTE_SECRET_INTERVAL, | |
2965 | .procname = "secret_interval", | |
2966 | .data = &ip_rt_secret_interval, | |
2967 | .maxlen = sizeof(int), | |
2968 | .mode = 0644, | |
2969 | .proc_handler = &proc_dointvec_jiffies, | |
2970 | .strategy = &sysctl_jiffies, | |
2971 | }, | |
2972 | { .ctl_name = 0 } | |
2973 | }; | |
2974 | #endif | |
2975 | ||
2976 | #ifdef CONFIG_NET_CLS_ROUTE | |
8dbde28d | 2977 | struct ip_rt_acct *ip_rt_acct __read_mostly; |
1da177e4 LT |
2978 | #endif /* CONFIG_NET_CLS_ROUTE */ |
2979 | ||
2980 | static __initdata unsigned long rhash_entries; | |
2981 | static int __init set_rhash_entries(char *str) | |
2982 | { | |
2983 | if (!str) | |
2984 | return 0; | |
2985 | rhash_entries = simple_strtoul(str, &str, 0); | |
2986 | return 1; | |
2987 | } | |
2988 | __setup("rhash_entries=", set_rhash_entries); | |
2989 | ||
2990 | int __init ip_rt_init(void) | |
2991 | { | |
424c4b70 | 2992 | int rc = 0; |
1da177e4 LT |
2993 | |
2994 | rt_hash_rnd = (int) ((num_physpages ^ (num_physpages>>8)) ^ | |
2995 | (jiffies ^ (jiffies >> 7))); | |
2996 | ||
2997 | #ifdef CONFIG_NET_CLS_ROUTE | |
8dbde28d | 2998 | ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct)); |
1da177e4 LT |
2999 | if (!ip_rt_acct) |
3000 | panic("IP: failed to allocate ip_rt_acct\n"); | |
1da177e4 LT |
3001 | #endif |
3002 | ||
e5d679f3 AD |
3003 | ipv4_dst_ops.kmem_cachep = |
3004 | kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, | |
20c2df83 | 3005 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3006 | |
14e50e57 DM |
3007 | ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; |
3008 | ||
424c4b70 ED |
3009 | rt_hash_table = (struct rt_hash_bucket *) |
3010 | alloc_large_system_hash("IP route cache", | |
3011 | sizeof(struct rt_hash_bucket), | |
3012 | rhash_entries, | |
3013 | (num_physpages >= 128 * 1024) ? | |
18955cfc | 3014 | 15 : 17, |
8d1502de | 3015 | 0, |
424c4b70 ED |
3016 | &rt_hash_log, |
3017 | &rt_hash_mask, | |
3018 | 0); | |
22c047cc ED |
3019 | memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket)); |
3020 | rt_hash_lock_init(); | |
1da177e4 LT |
3021 | |
3022 | ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1); | |
3023 | ip_rt_max_size = (rt_hash_mask + 1) * 16; | |
3024 | ||
1da177e4 LT |
3025 | devinet_init(); |
3026 | ip_fib_init(); | |
3027 | ||
b24b8a24 PE |
3028 | setup_timer(&rt_flush_timer, rt_run_flush, 0); |
3029 | setup_timer(&rt_secret_timer, rt_secret_rebuild, 0); | |
1da177e4 LT |
3030 | |
3031 | /* All the timers, started at system startup tend | |
3032 | to synchronize. Perturb it a bit. | |
3033 | */ | |
39c90ece ED |
3034 | schedule_delayed_work(&expires_work, |
3035 | net_random() % ip_rt_gc_interval + ip_rt_gc_interval); | |
1da177e4 LT |
3036 | |
3037 | rt_secret_timer.expires = jiffies + net_random() % ip_rt_secret_interval + | |
3038 | ip_rt_secret_interval; | |
3039 | add_timer(&rt_secret_timer); | |
3040 | ||
107f1634 PE |
3041 | if (ip_rt_proc_init(&init_net)) |
3042 | printk(KERN_ERR "Unable to create route proc files\n"); | |
1da177e4 LT |
3043 | #ifdef CONFIG_XFRM |
3044 | xfrm_init(); | |
3045 | xfrm4_init(); | |
3046 | #endif | |
63f3444f TG |
3047 | rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL); |
3048 | ||
1da177e4 LT |
3049 | return rc; |
3050 | } | |
3051 | ||
3052 | EXPORT_SYMBOL(__ip_select_ident); | |
3053 | EXPORT_SYMBOL(ip_route_input); | |
3054 | EXPORT_SYMBOL(ip_route_output_key); |