]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - net/ipv4/icmp.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[NET] NETNS: Omit net_device->nd_net without CONFIG_NET_NS.
[mirror_ubuntu-artful-kernel.git] / net / ipv4 / icmp.c
1 /*
2 * NET3: Implementation of the ICMP protocol layer.
3 *
4 * Alan Cox, <alan@redhat.com>
5 *
6 * Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * Some of the function names and the icmp unreach table for this
14 * module were derived from [icmp.c 1.0.11 06/02/93] by
15 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
16 * Other than that this module is a complete rewrite.
17 *
18 * Fixes:
19 * Clemens Fruhwirth : introduce global icmp rate limiting
20 * with icmp type masking ability instead
21 * of broken per type icmp timeouts.
22 * Mike Shaver : RFC1122 checks.
23 * Alan Cox : Multicast ping reply as self.
24 * Alan Cox : Fix atomicity lockup in ip_build_xmit
25 * call.
26 * Alan Cox : Added 216,128 byte paths to the MTU
27 * code.
28 * Martin Mares : RFC1812 checks.
29 * Martin Mares : Can be configured to follow redirects
30 * if acting as a router _without_ a
31 * routing protocol (RFC 1812).
32 * Martin Mares : Echo requests may be configured to
33 * be ignored (RFC 1812).
34 * Martin Mares : Limitation of ICMP error message
35 * transmit rate (RFC 1812).
36 * Martin Mares : TOS and Precedence set correctly
37 * (RFC 1812).
38 * Martin Mares : Now copying as much data from the
39 * original packet as we can without
40 * exceeding 576 bytes (RFC 1812).
41 * Willy Konynenberg : Transparent proxying support.
42 * Keith Owens : RFC1191 correction for 4.2BSD based
43 * path MTU bug.
44 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
45 * valid (RFC 1812).
46 * Andi Kleen : Check all packet lengths properly
47 * and moved all kfree_skb() up to
48 * icmp_rcv.
49 * Andi Kleen : Move the rate limit bookkeeping
50 * into the dest entry and use a token
51 * bucket filter (thanks to ANK). Make
52 * the rates sysctl configurable.
53 * Yu Tianli : Fixed two ugly bugs in icmp_send
54 * - IP option length was accounted wrongly
55 * - ICMP header length was not accounted
56 * at all.
57 * Tristan Greaves : Added sysctl option to ignore bogus
58 * broadcast responses from broken routers.
59 *
60 * To Fix:
61 *
62 * - Should use skb_pull() instead of all the manual checking.
63 * This would also greatly simply some upper layer error handlers. --AK
64 *
65 */
66
67 #include <linux/module.h>
68 #include <linux/types.h>
69 #include <linux/jiffies.h>
70 #include <linux/kernel.h>
71 #include <linux/fcntl.h>
72 #include <linux/socket.h>
73 #include <linux/in.h>
74 #include <linux/inet.h>
75 #include <linux/inetdevice.h>
76 #include <linux/netdevice.h>
77 #include <linux/string.h>
78 #include <linux/netfilter_ipv4.h>
79 #include <net/snmp.h>
80 #include <net/ip.h>
81 #include <net/route.h>
82 #include <net/protocol.h>
83 #include <net/icmp.h>
84 #include <net/tcp.h>
85 #include <net/udp.h>
86 #include <net/raw.h>
87 #include <linux/skbuff.h>
88 #include <net/sock.h>
89 #include <linux/errno.h>
90 #include <linux/timer.h>
91 #include <linux/init.h>
92 #include <asm/system.h>
93 #include <asm/uaccess.h>
94 #include <net/checksum.h>
95 #include <net/xfrm.h>
96
97 /*
98 * Build xmit assembly blocks
99 */
100
101 struct icmp_bxm {
102 struct sk_buff *skb;
103 int offset;
104 int data_len;
105
106 struct {
107 struct icmphdr icmph;
108 __be32 times[3];
109 } data;
110 int head_len;
111 struct ip_options replyopts;
112 unsigned char optbuf[40];
113 };
114
115 /*
116 * Statistics
117 */
118 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
119 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
120
121 /* An array of errno for error messages from dest unreach. */
122 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
123
124 struct icmp_err icmp_err_convert[] = {
125 {
126 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
127 .fatal = 0,
128 },
129 {
130 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
131 .fatal = 0,
132 },
133 {
134 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
135 .fatal = 1,
136 },
137 {
138 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
139 .fatal = 1,
140 },
141 {
142 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
143 .fatal = 0,
144 },
145 {
146 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
147 .fatal = 0,
148 },
149 {
150 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
151 .fatal = 1,
152 },
153 {
154 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
155 .fatal = 1,
156 },
157 {
158 .errno = ENONET, /* ICMP_HOST_ISOLATED */
159 .fatal = 1,
160 },
161 {
162 .errno = ENETUNREACH, /* ICMP_NET_ANO */
163 .fatal = 1,
164 },
165 {
166 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
167 .fatal = 1,
168 },
169 {
170 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
171 .fatal = 0,
172 },
173 {
174 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
175 .fatal = 0,
176 },
177 {
178 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
179 .fatal = 1,
180 },
181 {
182 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
183 .fatal = 1,
184 },
185 {
186 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
187 .fatal = 1,
188 },
189 };
190
191 /* Control parameters for ECHO replies. */
192 int sysctl_icmp_echo_ignore_all __read_mostly;
193 int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1;
194
195 /* Control parameter - ignore bogus broadcast responses? */
196 int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1;
197
198 /*
199 * Configurable global rate limit.
200 *
201 * ratelimit defines tokens/packet consumed for dst->rate_token bucket
202 * ratemask defines which icmp types are ratelimited by setting
203 * it's bit position.
204 *
205 * default:
206 * dest unreachable (3), source quench (4),
207 * time exceeded (11), parameter problem (12)
208 */
209
210 int sysctl_icmp_ratelimit __read_mostly = 1 * HZ;
211 int sysctl_icmp_ratemask __read_mostly = 0x1818;
212 int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly;
213
214 /*
215 * ICMP control array. This specifies what to do with each ICMP.
216 */
217
218 struct icmp_control {
219 void (*handler)(struct sk_buff *skb);
220 short error; /* This ICMP is classed as an error message */
221 };
222
223 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
224
225 /*
226 * The ICMP socket(s). This is the most convenient way to flow control
227 * our ICMP output as well as maintain a clean interface throughout
228 * all layers. All Socketless IP sends will soon be gone.
229 *
230 * On SMP we have one ICMP socket per-cpu.
231 */
232 static struct sock *icmp_sk(struct net *net)
233 {
234 return net->ipv4.icmp_sk[smp_processor_id()];
235 }
236
237 static inline int icmp_xmit_lock(struct sock *sk)
238 {
239 local_bh_disable();
240
241 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
242 /* This can happen if the output path signals a
243 * dst_link_failure() for an outgoing ICMP packet.
244 */
245 local_bh_enable();
246 return 1;
247 }
248 return 0;
249 }
250
251 static inline void icmp_xmit_unlock(struct sock *sk)
252 {
253 spin_unlock_bh(&sk->sk_lock.slock);
254 }
255
256 /*
257 * Send an ICMP frame.
258 */
259
260 /*
261 * Check transmit rate limitation for given message.
262 * The rate information is held in the destination cache now.
263 * This function is generic and could be used for other purposes
264 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
265 *
266 * Note that the same dst_entry fields are modified by functions in
267 * route.c too, but these work for packet destinations while xrlim_allow
268 * works for icmp destinations. This means the rate limiting information
269 * for one "ip object" is shared - and these ICMPs are twice limited:
270 * by source and by destination.
271 *
272 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
273 * SHOULD allow setting of rate limits
274 *
275 * Shared between ICMPv4 and ICMPv6.
276 */
277 #define XRLIM_BURST_FACTOR 6
278 int xrlim_allow(struct dst_entry *dst, int timeout)
279 {
280 unsigned long now, token = dst->rate_tokens;
281 int rc = 0;
282
283 now = jiffies;
284 token += now - dst->rate_last;
285 dst->rate_last = now;
286 if (token > XRLIM_BURST_FACTOR * timeout)
287 token = XRLIM_BURST_FACTOR * timeout;
288 if (token >= timeout) {
289 token -= timeout;
290 rc = 1;
291 }
292 dst->rate_tokens = token;
293 return rc;
294 }
295
296 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
297 {
298 struct dst_entry *dst = &rt->u.dst;
299 int rc = 1;
300
301 if (type > NR_ICMP_TYPES)
302 goto out;
303
304 /* Don't limit PMTU discovery. */
305 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
306 goto out;
307
308 /* No rate limit on loopback */
309 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
310 goto out;
311
312 /* Limit if icmp type is enabled in ratemask. */
313 if ((1 << type) & sysctl_icmp_ratemask)
314 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
315 out:
316 return rc;
317 }
318
319 /*
320 * Maintain the counters used in the SNMP statistics for outgoing ICMP
321 */
322 void icmp_out_count(unsigned char type)
323 {
324 ICMPMSGOUT_INC_STATS(type);
325 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
326 }
327
328 /*
329 * Checksum each fragment, and on the first include the headers and final
330 * checksum.
331 */
332 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
333 struct sk_buff *skb)
334 {
335 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
336 __wsum csum;
337
338 csum = skb_copy_and_csum_bits(icmp_param->skb,
339 icmp_param->offset + offset,
340 to, len, 0);
341
342 skb->csum = csum_block_add(skb->csum, csum, odd);
343 if (icmp_pointers[icmp_param->data.icmph.type].error)
344 nf_ct_attach(skb, icmp_param->skb);
345 return 0;
346 }
347
348 static void icmp_push_reply(struct icmp_bxm *icmp_param,
349 struct ipcm_cookie *ipc, struct rtable *rt)
350 {
351 struct sock *sk;
352 struct sk_buff *skb;
353
354 sk = icmp_sk(dev_net(rt->u.dst.dev));
355 if (ip_append_data(sk, icmp_glue_bits, icmp_param,
356 icmp_param->data_len+icmp_param->head_len,
357 icmp_param->head_len,
358 ipc, rt, MSG_DONTWAIT) < 0)
359 ip_flush_pending_frames(sk);
360 else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
361 struct icmphdr *icmph = icmp_hdr(skb);
362 __wsum csum = 0;
363 struct sk_buff *skb1;
364
365 skb_queue_walk(&sk->sk_write_queue, skb1) {
366 csum = csum_add(csum, skb1->csum);
367 }
368 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
369 (char *)icmph,
370 icmp_param->head_len, csum);
371 icmph->checksum = csum_fold(csum);
372 skb->ip_summed = CHECKSUM_NONE;
373 ip_push_pending_frames(sk);
374 }
375 }
376
377 /*
378 * Driving logic for building and sending ICMP messages.
379 */
380
381 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
382 {
383 struct ipcm_cookie ipc;
384 struct rtable *rt = skb->rtable;
385 struct net *net = dev_net(rt->u.dst.dev);
386 struct sock *sk = icmp_sk(net);
387 struct inet_sock *inet = inet_sk(sk);
388 __be32 daddr;
389
390 if (ip_options_echo(&icmp_param->replyopts, skb))
391 return;
392
393 if (icmp_xmit_lock(sk))
394 return;
395
396 icmp_param->data.icmph.checksum = 0;
397
398 inet->tos = ip_hdr(skb)->tos;
399 daddr = ipc.addr = rt->rt_src;
400 ipc.opt = NULL;
401 if (icmp_param->replyopts.optlen) {
402 ipc.opt = &icmp_param->replyopts;
403 if (ipc.opt->srr)
404 daddr = icmp_param->replyopts.faddr;
405 }
406 {
407 struct flowi fl = { .nl_u = { .ip4_u =
408 { .daddr = daddr,
409 .saddr = rt->rt_spec_dst,
410 .tos = RT_TOS(ip_hdr(skb)->tos) } },
411 .proto = IPPROTO_ICMP };
412 security_skb_classify_flow(skb, &fl);
413 if (ip_route_output_key(net, &rt, &fl))
414 goto out_unlock;
415 }
416 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
417 icmp_param->data.icmph.code))
418 icmp_push_reply(icmp_param, &ipc, rt);
419 ip_rt_put(rt);
420 out_unlock:
421 icmp_xmit_unlock(sk);
422 }
423
424
425 /*
426 * Send an ICMP message in response to a situation
427 *
428 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
429 * MAY send more (we do).
430 * MUST NOT change this header information.
431 * MUST NOT reply to a multicast/broadcast IP address.
432 * MUST NOT reply to a multicast/broadcast MAC address.
433 * MUST reply to only the first fragment.
434 */
435
436 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
437 {
438 struct iphdr *iph;
439 int room;
440 struct icmp_bxm icmp_param;
441 struct rtable *rt = skb_in->rtable;
442 struct ipcm_cookie ipc;
443 __be32 saddr;
444 u8 tos;
445 struct net *net;
446 struct sock *sk;
447
448 if (!rt)
449 goto out;
450 net = dev_net(rt->u.dst.dev);
451 sk = icmp_sk(net);
452
453 /*
454 * Find the original header. It is expected to be valid, of course.
455 * Check this, icmp_send is called from the most obscure devices
456 * sometimes.
457 */
458 iph = ip_hdr(skb_in);
459
460 if ((u8 *)iph < skb_in->head ||
461 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
462 goto out;
463
464 /*
465 * No replies to physical multicast/broadcast
466 */
467 if (skb_in->pkt_type != PACKET_HOST)
468 goto out;
469
470 /*
471 * Now check at the protocol level
472 */
473 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
474 goto out;
475
476 /*
477 * Only reply to fragment 0. We byte re-order the constant
478 * mask for efficiency.
479 */
480 if (iph->frag_off & htons(IP_OFFSET))
481 goto out;
482
483 /*
484 * If we send an ICMP error to an ICMP error a mess would result..
485 */
486 if (icmp_pointers[type].error) {
487 /*
488 * We are an error, check if we are replying to an
489 * ICMP error
490 */
491 if (iph->protocol == IPPROTO_ICMP) {
492 u8 _inner_type, *itp;
493
494 itp = skb_header_pointer(skb_in,
495 skb_network_header(skb_in) +
496 (iph->ihl << 2) +
497 offsetof(struct icmphdr,
498 type) -
499 skb_in->data,
500 sizeof(_inner_type),
501 &_inner_type);
502 if (itp == NULL)
503 goto out;
504
505 /*
506 * Assume any unknown ICMP type is an error. This
507 * isn't specified by the RFC, but think about it..
508 */
509 if (*itp > NR_ICMP_TYPES ||
510 icmp_pointers[*itp].error)
511 goto out;
512 }
513 }
514
515 if (icmp_xmit_lock(sk))
516 return;
517
518 /*
519 * Construct source address and options.
520 */
521
522 saddr = iph->daddr;
523 if (!(rt->rt_flags & RTCF_LOCAL)) {
524 struct net_device *dev = NULL;
525
526 if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
527 dev = dev_get_by_index(net, rt->fl.iif);
528
529 if (dev) {
530 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
531 dev_put(dev);
532 } else
533 saddr = 0;
534 }
535
536 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
537 IPTOS_PREC_INTERNETCONTROL) :
538 iph->tos;
539
540 if (ip_options_echo(&icmp_param.replyopts, skb_in))
541 goto out_unlock;
542
543
544 /*
545 * Prepare data for ICMP header.
546 */
547
548 icmp_param.data.icmph.type = type;
549 icmp_param.data.icmph.code = code;
550 icmp_param.data.icmph.un.gateway = info;
551 icmp_param.data.icmph.checksum = 0;
552 icmp_param.skb = skb_in;
553 icmp_param.offset = skb_network_offset(skb_in);
554 inet_sk(sk)->tos = tos;
555 ipc.addr = iph->saddr;
556 ipc.opt = &icmp_param.replyopts;
557
558 {
559 struct flowi fl = {
560 .nl_u = {
561 .ip4_u = {
562 .daddr = icmp_param.replyopts.srr ?
563 icmp_param.replyopts.faddr :
564 iph->saddr,
565 .saddr = saddr,
566 .tos = RT_TOS(tos)
567 }
568 },
569 .proto = IPPROTO_ICMP,
570 .uli_u = {
571 .icmpt = {
572 .type = type,
573 .code = code
574 }
575 }
576 };
577 int err;
578 struct rtable *rt2;
579
580 security_skb_classify_flow(skb_in, &fl);
581 if (__ip_route_output_key(net, &rt, &fl))
582 goto out_unlock;
583
584 /* No need to clone since we're just using its address. */
585 rt2 = rt;
586
587 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
588 switch (err) {
589 case 0:
590 if (rt != rt2)
591 goto route_done;
592 break;
593 case -EPERM:
594 rt = NULL;
595 break;
596 default:
597 goto out_unlock;
598 }
599
600 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
601 goto out_unlock;
602
603 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
604 err = __ip_route_output_key(net, &rt2, &fl);
605 else {
606 struct flowi fl2 = {};
607 struct dst_entry *odst;
608
609 fl2.fl4_dst = fl.fl4_src;
610 if (ip_route_output_key(net, &rt2, &fl2))
611 goto out_unlock;
612
613 /* Ugh! */
614 odst = skb_in->dst;
615 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
616 RT_TOS(tos), rt2->u.dst.dev);
617
618 dst_release(&rt2->u.dst);
619 rt2 = skb_in->rtable;
620 skb_in->dst = odst;
621 }
622
623 if (err)
624 goto out_unlock;
625
626 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
627 XFRM_LOOKUP_ICMP);
628 if (err == -ENOENT) {
629 if (!rt)
630 goto out_unlock;
631 goto route_done;
632 }
633
634 dst_release(&rt->u.dst);
635 rt = rt2;
636
637 if (err)
638 goto out_unlock;
639 }
640
641 route_done:
642 if (!icmpv4_xrlim_allow(rt, type, code))
643 goto ende;
644
645 /* RFC says return as much as we can without exceeding 576 bytes. */
646
647 room = dst_mtu(&rt->u.dst);
648 if (room > 576)
649 room = 576;
650 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
651 room -= sizeof(struct icmphdr);
652
653 icmp_param.data_len = skb_in->len - icmp_param.offset;
654 if (icmp_param.data_len > room)
655 icmp_param.data_len = room;
656 icmp_param.head_len = sizeof(struct icmphdr);
657
658 icmp_push_reply(&icmp_param, &ipc, rt);
659 ende:
660 ip_rt_put(rt);
661 out_unlock:
662 icmp_xmit_unlock(sk);
663 out:;
664 }
665
666
667 /*
668 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
669 */
670
671 static void icmp_unreach(struct sk_buff *skb)
672 {
673 struct iphdr *iph;
674 struct icmphdr *icmph;
675 int hash, protocol;
676 struct net_protocol *ipprot;
677 u32 info = 0;
678 struct net *net;
679
680 net = dev_net(skb->dst->dev);
681
682 /*
683 * Incomplete header ?
684 * Only checks for the IP header, there should be an
685 * additional check for longer headers in upper levels.
686 */
687
688 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
689 goto out_err;
690
691 icmph = icmp_hdr(skb);
692 iph = (struct iphdr *)skb->data;
693
694 if (iph->ihl < 5) /* Mangled header, drop. */
695 goto out_err;
696
697 if (icmph->type == ICMP_DEST_UNREACH) {
698 switch (icmph->code & 15) {
699 case ICMP_NET_UNREACH:
700 case ICMP_HOST_UNREACH:
701 case ICMP_PROT_UNREACH:
702 case ICMP_PORT_UNREACH:
703 break;
704 case ICMP_FRAG_NEEDED:
705 if (ipv4_config.no_pmtu_disc) {
706 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
707 "fragmentation needed "
708 "and DF set.\n",
709 NIPQUAD(iph->daddr));
710 } else {
711 info = ip_rt_frag_needed(net, iph,
712 ntohs(icmph->un.frag.mtu));
713 if (!info)
714 goto out;
715 }
716 break;
717 case ICMP_SR_FAILED:
718 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
719 "Route Failed.\n",
720 NIPQUAD(iph->daddr));
721 break;
722 default:
723 break;
724 }
725 if (icmph->code > NR_ICMP_UNREACH)
726 goto out;
727 } else if (icmph->type == ICMP_PARAMETERPROB)
728 info = ntohl(icmph->un.gateway) >> 24;
729
730 /*
731 * Throw it at our lower layers
732 *
733 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
734 * header.
735 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
736 * transport layer.
737 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
738 * transport layer.
739 */
740
741 /*
742 * Check the other end isnt violating RFC 1122. Some routers send
743 * bogus responses to broadcast frames. If you see this message
744 * first check your netmask matches at both ends, if it does then
745 * get the other vendor to fix their kit.
746 */
747
748 if (!sysctl_icmp_ignore_bogus_error_responses &&
749 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
750 if (net_ratelimit())
751 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
752 "type %u, code %u "
753 "error to a broadcast: %u.%u.%u.%u on %s\n",
754 NIPQUAD(ip_hdr(skb)->saddr),
755 icmph->type, icmph->code,
756 NIPQUAD(iph->daddr),
757 skb->dev->name);
758 goto out;
759 }
760
761 /* Checkin full IP header plus 8 bytes of protocol to
762 * avoid additional coding at protocol handlers.
763 */
764 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
765 goto out;
766
767 iph = (struct iphdr *)skb->data;
768 protocol = iph->protocol;
769
770 /*
771 * Deliver ICMP message to raw sockets. Pretty useless feature?
772 */
773 raw_icmp_error(skb, protocol, info);
774
775 hash = protocol & (MAX_INET_PROTOS - 1);
776 rcu_read_lock();
777 ipprot = rcu_dereference(inet_protos[hash]);
778 if (ipprot && ipprot->err_handler)
779 ipprot->err_handler(skb, info);
780 rcu_read_unlock();
781
782 out:
783 return;
784 out_err:
785 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
786 goto out;
787 }
788
789
790 /*
791 * Handle ICMP_REDIRECT.
792 */
793
794 static void icmp_redirect(struct sk_buff *skb)
795 {
796 struct iphdr *iph;
797
798 if (skb->len < sizeof(struct iphdr))
799 goto out_err;
800
801 /*
802 * Get the copied header of the packet that caused the redirect
803 */
804 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
805 goto out;
806
807 iph = (struct iphdr *)skb->data;
808
809 switch (icmp_hdr(skb)->code & 7) {
810 case ICMP_REDIR_NET:
811 case ICMP_REDIR_NETTOS:
812 /*
813 * As per RFC recommendations now handle it as a host redirect.
814 */
815 case ICMP_REDIR_HOST:
816 case ICMP_REDIR_HOSTTOS:
817 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
818 icmp_hdr(skb)->un.gateway,
819 iph->saddr, skb->dev);
820 break;
821 }
822 out:
823 return;
824 out_err:
825 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
826 goto out;
827 }
828
829 /*
830 * Handle ICMP_ECHO ("ping") requests.
831 *
832 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
833 * requests.
834 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
835 * included in the reply.
836 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
837 * echo requests, MUST have default=NOT.
838 * See also WRT handling of options once they are done and working.
839 */
840
841 static void icmp_echo(struct sk_buff *skb)
842 {
843 if (!sysctl_icmp_echo_ignore_all) {
844 struct icmp_bxm icmp_param;
845
846 icmp_param.data.icmph = *icmp_hdr(skb);
847 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
848 icmp_param.skb = skb;
849 icmp_param.offset = 0;
850 icmp_param.data_len = skb->len;
851 icmp_param.head_len = sizeof(struct icmphdr);
852 icmp_reply(&icmp_param, skb);
853 }
854 }
855
856 /*
857 * Handle ICMP Timestamp requests.
858 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
859 * SHOULD be in the kernel for minimum random latency.
860 * MUST be accurate to a few minutes.
861 * MUST be updated at least at 15Hz.
862 */
863 static void icmp_timestamp(struct sk_buff *skb)
864 {
865 struct timeval tv;
866 struct icmp_bxm icmp_param;
867 /*
868 * Too short.
869 */
870 if (skb->len < 4)
871 goto out_err;
872
873 /*
874 * Fill in the current time as ms since midnight UT:
875 */
876 do_gettimeofday(&tv);
877 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
878 tv.tv_usec / 1000);
879 icmp_param.data.times[2] = icmp_param.data.times[1];
880 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
881 BUG();
882 icmp_param.data.icmph = *icmp_hdr(skb);
883 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
884 icmp_param.data.icmph.code = 0;
885 icmp_param.skb = skb;
886 icmp_param.offset = 0;
887 icmp_param.data_len = 0;
888 icmp_param.head_len = sizeof(struct icmphdr) + 12;
889 icmp_reply(&icmp_param, skb);
890 out:
891 return;
892 out_err:
893 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
894 goto out;
895 }
896
897
898 /*
899 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
900 *
901 * RFC1122 (3.2.2.9). A host MUST only send replies to
902 * ADDRESS_MASK requests if it's been configured as an address mask
903 * agent. Receiving a request doesn't constitute implicit permission to
904 * act as one. Of course, implementing this correctly requires (SHOULD)
905 * a way to turn the functionality on and off. Another one for sysctl(),
906 * I guess. -- MS
907 *
908 * RFC1812 (4.3.3.9). A router MUST implement it.
909 * A router SHOULD have switch turning it on/off.
910 * This switch MUST be ON by default.
911 *
912 * Gratuitous replies, zero-source replies are not implemented,
913 * that complies with RFC. DO NOT implement them!!! All the idea
914 * of broadcast addrmask replies as specified in RFC950 is broken.
915 * The problem is that it is not uncommon to have several prefixes
916 * on one physical interface. Moreover, addrmask agent can even be
917 * not aware of existing another prefixes.
918 * If source is zero, addrmask agent cannot choose correct prefix.
919 * Gratuitous mask announcements suffer from the same problem.
920 * RFC1812 explains it, but still allows to use ADDRMASK,
921 * that is pretty silly. --ANK
922 *
923 * All these rules are so bizarre, that I removed kernel addrmask
924 * support at all. It is wrong, it is obsolete, nobody uses it in
925 * any case. --ANK
926 *
927 * Furthermore you can do it with a usermode address agent program
928 * anyway...
929 */
930
931 static void icmp_address(struct sk_buff *skb)
932 {
933 #if 0
934 if (net_ratelimit())
935 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
936 #endif
937 }
938
939 /*
940 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
941 * loudly if an inconsistency is found.
942 */
943
944 static void icmp_address_reply(struct sk_buff *skb)
945 {
946 struct rtable *rt = skb->rtable;
947 struct net_device *dev = skb->dev;
948 struct in_device *in_dev;
949 struct in_ifaddr *ifa;
950
951 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
952 goto out;
953
954 in_dev = in_dev_get(dev);
955 if (!in_dev)
956 goto out;
957 rcu_read_lock();
958 if (in_dev->ifa_list &&
959 IN_DEV_LOG_MARTIANS(in_dev) &&
960 IN_DEV_FORWARD(in_dev)) {
961 __be32 _mask, *mp;
962
963 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
964 BUG_ON(mp == NULL);
965 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
966 if (*mp == ifa->ifa_mask &&
967 inet_ifa_match(rt->rt_src, ifa))
968 break;
969 }
970 if (!ifa && net_ratelimit()) {
971 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
972 "%s/%u.%u.%u.%u\n",
973 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
974 }
975 }
976 rcu_read_unlock();
977 in_dev_put(in_dev);
978 out:;
979 }
980
981 static void icmp_discard(struct sk_buff *skb)
982 {
983 }
984
985 /*
986 * Deal with incoming ICMP packets.
987 */
988 int icmp_rcv(struct sk_buff *skb)
989 {
990 struct icmphdr *icmph;
991 struct rtable *rt = skb->rtable;
992
993 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
994 int nh;
995
996 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
997 XFRM_STATE_ICMP))
998 goto drop;
999
1000 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1001 goto drop;
1002
1003 nh = skb_network_offset(skb);
1004 skb_set_network_header(skb, sizeof(*icmph));
1005
1006 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1007 goto drop;
1008
1009 skb_set_network_header(skb, nh);
1010 }
1011
1012 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
1013
1014 switch (skb->ip_summed) {
1015 case CHECKSUM_COMPLETE:
1016 if (!csum_fold(skb->csum))
1017 break;
1018 /* fall through */
1019 case CHECKSUM_NONE:
1020 skb->csum = 0;
1021 if (__skb_checksum_complete(skb))
1022 goto error;
1023 }
1024
1025 if (!pskb_pull(skb, sizeof(*icmph)))
1026 goto error;
1027
1028 icmph = icmp_hdr(skb);
1029
1030 ICMPMSGIN_INC_STATS_BH(icmph->type);
1031 /*
1032 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1033 *
1034 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1035 * discarded.
1036 */
1037 if (icmph->type > NR_ICMP_TYPES)
1038 goto error;
1039
1040
1041 /*
1042 * Parse the ICMP message
1043 */
1044
1045 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1046 /*
1047 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1048 * silently ignored (we let user decide with a sysctl).
1049 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1050 * discarded if to broadcast/multicast.
1051 */
1052 if ((icmph->type == ICMP_ECHO ||
1053 icmph->type == ICMP_TIMESTAMP) &&
1054 sysctl_icmp_echo_ignore_broadcasts) {
1055 goto error;
1056 }
1057 if (icmph->type != ICMP_ECHO &&
1058 icmph->type != ICMP_TIMESTAMP &&
1059 icmph->type != ICMP_ADDRESS &&
1060 icmph->type != ICMP_ADDRESSREPLY) {
1061 goto error;
1062 }
1063 }
1064
1065 icmp_pointers[icmph->type].handler(skb);
1066
1067 drop:
1068 kfree_skb(skb);
1069 return 0;
1070 error:
1071 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1072 goto drop;
1073 }
1074
1075 /*
1076 * This table is the definition of how we handle ICMP.
1077 */
1078 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1079 [ICMP_ECHOREPLY] = {
1080 .handler = icmp_discard,
1081 },
1082 [1] = {
1083 .handler = icmp_discard,
1084 .error = 1,
1085 },
1086 [2] = {
1087 .handler = icmp_discard,
1088 .error = 1,
1089 },
1090 [ICMP_DEST_UNREACH] = {
1091 .handler = icmp_unreach,
1092 .error = 1,
1093 },
1094 [ICMP_SOURCE_QUENCH] = {
1095 .handler = icmp_unreach,
1096 .error = 1,
1097 },
1098 [ICMP_REDIRECT] = {
1099 .handler = icmp_redirect,
1100 .error = 1,
1101 },
1102 [6] = {
1103 .handler = icmp_discard,
1104 .error = 1,
1105 },
1106 [7] = {
1107 .handler = icmp_discard,
1108 .error = 1,
1109 },
1110 [ICMP_ECHO] = {
1111 .handler = icmp_echo,
1112 },
1113 [9] = {
1114 .handler = icmp_discard,
1115 .error = 1,
1116 },
1117 [10] = {
1118 .handler = icmp_discard,
1119 .error = 1,
1120 },
1121 [ICMP_TIME_EXCEEDED] = {
1122 .handler = icmp_unreach,
1123 .error = 1,
1124 },
1125 [ICMP_PARAMETERPROB] = {
1126 .handler = icmp_unreach,
1127 .error = 1,
1128 },
1129 [ICMP_TIMESTAMP] = {
1130 .handler = icmp_timestamp,
1131 },
1132 [ICMP_TIMESTAMPREPLY] = {
1133 .handler = icmp_discard,
1134 },
1135 [ICMP_INFO_REQUEST] = {
1136 .handler = icmp_discard,
1137 },
1138 [ICMP_INFO_REPLY] = {
1139 .handler = icmp_discard,
1140 },
1141 [ICMP_ADDRESS] = {
1142 .handler = icmp_address,
1143 },
1144 [ICMP_ADDRESSREPLY] = {
1145 .handler = icmp_address_reply,
1146 },
1147 };
1148
1149 static void __net_exit icmp_sk_exit(struct net *net)
1150 {
1151 int i;
1152
1153 for_each_possible_cpu(i)
1154 sk_release_kernel(net->ipv4.icmp_sk[i]);
1155 kfree(net->ipv4.icmp_sk);
1156 net->ipv4.icmp_sk = NULL;
1157 }
1158
1159 int __net_init icmp_sk_init(struct net *net)
1160 {
1161 int i, err;
1162
1163 net->ipv4.icmp_sk =
1164 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1165 if (net->ipv4.icmp_sk == NULL)
1166 return -ENOMEM;
1167
1168 for_each_possible_cpu(i) {
1169 struct sock *sk;
1170 struct socket *sock;
1171 struct inet_sock *inet;
1172
1173 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP, &sock);
1174 if (err < 0)
1175 goto fail;
1176
1177 net->ipv4.icmp_sk[i] = sk = sock->sk;
1178 sk_change_net(sk, net);
1179
1180 sk->sk_allocation = GFP_ATOMIC;
1181
1182 /* Enough space for 2 64K ICMP packets, including
1183 * sk_buff struct overhead.
1184 */
1185 sk->sk_sndbuf =
1186 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1187
1188 inet = inet_sk(sk);
1189 inet->uc_ttl = -1;
1190 inet->pmtudisc = IP_PMTUDISC_DONT;
1191
1192 /* Unhash it so that IP input processing does not even
1193 * see it, we do not wish this socket to see incoming
1194 * packets.
1195 */
1196 sk->sk_prot->unhash(sk);
1197 }
1198 return 0;
1199
1200 fail:
1201 for_each_possible_cpu(i)
1202 sk_release_kernel(net->ipv4.icmp_sk[i]);
1203 kfree(net->ipv4.icmp_sk);
1204 return err;
1205 }
1206
1207 static struct pernet_operations __net_initdata icmp_sk_ops = {
1208 .init = icmp_sk_init,
1209 .exit = icmp_sk_exit,
1210 };
1211
1212 int __init icmp_init(void)
1213 {
1214 return register_pernet_device(&icmp_sk_ops);
1215 }
1216
1217 EXPORT_SYMBOL(icmp_err_convert);
1218 EXPORT_SYMBOL(icmp_send);
1219 EXPORT_SYMBOL(icmp_statistics);
1220 EXPORT_SYMBOL(xrlim_allow);
Ubuntu Artful kernel mirror