Merge commit master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6 of HEAD
[linux-2.6] / 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
96 /*
97  *      Build xmit assembly blocks
98  */
99
100 struct icmp_bxm {
101         struct sk_buff *skb;
102         int offset;
103         int data_len;
104
105         struct {
106                 struct icmphdr icmph;
107                 __u32          times[3];
108         } data;
109         int head_len;
110         struct ip_options replyopts;
111         unsigned char  optbuf[40];
112 };
113
114 /*
115  *      Statistics
116  */
117 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
118
119 /* An array of errno for error messages from dest unreach. */
120 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
121
122 struct icmp_err icmp_err_convert[] = {
123         {
124                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
125                 .fatal = 0,
126         },
127         {
128                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
129                 .fatal = 0,
130         },
131         {
132                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
133                 .fatal = 1,
134         },
135         {
136                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
137                 .fatal = 1,
138         },
139         {
140                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
141                 .fatal = 0,
142         },
143         {
144                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
145                 .fatal = 0,
146         },
147         {
148                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
149                 .fatal = 1,
150         },
151         {
152                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
153                 .fatal = 1,
154         },
155         {
156                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
157                 .fatal = 1,
158         },
159         {
160                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
161                 .fatal = 1,
162         },
163         {
164                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
165                 .fatal = 1,
166         },
167         {
168                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
169                 .fatal = 0,
170         },
171         {
172                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
173                 .fatal = 0,
174         },
175         {
176                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
177                 .fatal = 1,
178         },
179         {
180                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
181                 .fatal = 1,
182         },
183         {
184                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
185                 .fatal = 1,
186         },
187 };
188
189 /* Control parameters for ECHO replies. */
190 int sysctl_icmp_echo_ignore_all;
191 int sysctl_icmp_echo_ignore_broadcasts = 1;
192
193 /* Control parameter - ignore bogus broadcast responses? */
194 int sysctl_icmp_ignore_bogus_error_responses = 1;
195
196 /*
197  *      Configurable global rate limit.
198  *
199  *      ratelimit defines tokens/packet consumed for dst->rate_token bucket
200  *      ratemask defines which icmp types are ratelimited by setting
201  *      it's bit position.
202  *
203  *      default:
204  *      dest unreachable (3), source quench (4),
205  *      time exceeded (11), parameter problem (12)
206  */
207
208 int sysctl_icmp_ratelimit = 1 * HZ;
209 int sysctl_icmp_ratemask = 0x1818;
210 int sysctl_icmp_errors_use_inbound_ifaddr;
211
212 /*
213  *      ICMP control array. This specifies what to do with each ICMP.
214  */
215
216 struct icmp_control {
217         int output_entry;       /* Field for increment on output */
218         int input_entry;        /* Field for increment on input */
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 DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL;
233 #define icmp_socket     __get_cpu_var(__icmp_socket)
234
235 static __inline__ int icmp_xmit_lock(void)
236 {
237         local_bh_disable();
238
239         if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) {
240                 /* This can happen if the output path signals a
241                  * dst_link_failure() for an outgoing ICMP packet.
242                  */
243                 local_bh_enable();
244                 return 1;
245         }
246         return 0;
247 }
248
249 static void icmp_xmit_unlock(void)
250 {
251         spin_unlock_bh(&icmp_socket->sk->sk_lock.slock);
252 }
253
254 /*
255  *      Send an ICMP frame.
256  */
257
258 /*
259  *      Check transmit rate limitation for given message.
260  *      The rate information is held in the destination cache now.
261  *      This function is generic and could be used for other purposes
262  *      too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
263  *
264  *      Note that the same dst_entry fields are modified by functions in
265  *      route.c too, but these work for packet destinations while xrlim_allow
266  *      works for icmp destinations. This means the rate limiting information
267  *      for one "ip object" is shared - and these ICMPs are twice limited:
268  *      by source and by destination.
269  *
270  *      RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
271  *                        SHOULD allow setting of rate limits
272  *
273  *      Shared between ICMPv4 and ICMPv6.
274  */
275 #define XRLIM_BURST_FACTOR 6
276 int xrlim_allow(struct dst_entry *dst, int timeout)
277 {
278         unsigned long now;
279         int rc = 0;
280
281         now = jiffies;
282         dst->rate_tokens += now - dst->rate_last;
283         dst->rate_last = now;
284         if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout)
285                 dst->rate_tokens = XRLIM_BURST_FACTOR * timeout;
286         if (dst->rate_tokens >= timeout) {
287                 dst->rate_tokens -= timeout;
288                 rc = 1;
289         }
290         return rc;
291 }
292
293 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
294 {
295         struct dst_entry *dst = &rt->u.dst;
296         int rc = 1;
297
298         if (type > NR_ICMP_TYPES)
299                 goto out;
300
301         /* Don't limit PMTU discovery. */
302         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
303                 goto out;
304
305         /* No rate limit on loopback */
306         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
307                 goto out;
308
309         /* Limit if icmp type is enabled in ratemask. */
310         if ((1 << type) & sysctl_icmp_ratemask)
311                 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
312 out:
313         return rc;
314 }
315
316 /*
317  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
318  */
319 static void icmp_out_count(int type)
320 {
321         if (type <= NR_ICMP_TYPES) {
322                 ICMP_INC_STATS(icmp_pointers[type].output_entry);
323                 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
324         }
325 }
326
327 /*
328  *      Checksum each fragment, and on the first include the headers and final
329  *      checksum.
330  */
331 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
332                           struct sk_buff *skb)
333 {
334         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
335         unsigned int csum;
336
337         csum = skb_copy_and_csum_bits(icmp_param->skb,
338                                       icmp_param->offset + offset,
339                                       to, len, 0);
340
341         skb->csum = csum_block_add(skb->csum, csum, odd);
342         if (icmp_pointers[icmp_param->data.icmph.type].error)
343                 nf_ct_attach(skb, icmp_param->skb);
344         return 0;
345 }
346
347 static void icmp_push_reply(struct icmp_bxm *icmp_param,
348                             struct ipcm_cookie *ipc, struct rtable *rt)
349 {
350         struct sk_buff *skb;
351
352         if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
353                            icmp_param->data_len+icmp_param->head_len,
354                            icmp_param->head_len,
355                            ipc, rt, MSG_DONTWAIT) < 0)
356                 ip_flush_pending_frames(icmp_socket->sk);
357         else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
358                 struct icmphdr *icmph = skb->h.icmph;
359                 unsigned int csum = 0;
360                 struct sk_buff *skb1;
361
362                 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
363                         csum = csum_add(csum, skb1->csum);
364                 }
365                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
366                                                  (char *)icmph,
367                                                  icmp_param->head_len, csum);
368                 icmph->checksum = csum_fold(csum);
369                 skb->ip_summed = CHECKSUM_NONE;
370                 ip_push_pending_frames(icmp_socket->sk);
371         }
372 }
373
374 /*
375  *      Driving logic for building and sending ICMP messages.
376  */
377
378 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
379 {
380         struct sock *sk = icmp_socket->sk;
381         struct inet_sock *inet = inet_sk(sk);
382         struct ipcm_cookie ipc;
383         struct rtable *rt = (struct rtable *)skb->dst;
384         u32 daddr;
385
386         if (ip_options_echo(&icmp_param->replyopts, skb))
387                 return;
388
389         if (icmp_xmit_lock())
390                 return;
391
392         icmp_param->data.icmph.checksum = 0;
393         icmp_out_count(icmp_param->data.icmph.type);
394
395         inet->tos = skb->nh.iph->tos;
396         daddr = ipc.addr = rt->rt_src;
397         ipc.opt = NULL;
398         if (icmp_param->replyopts.optlen) {
399                 ipc.opt = &icmp_param->replyopts;
400                 if (ipc.opt->srr)
401                         daddr = icmp_param->replyopts.faddr;
402         }
403         {
404                 struct flowi fl = { .nl_u = { .ip4_u =
405                                               { .daddr = daddr,
406                                                 .saddr = rt->rt_spec_dst,
407                                                 .tos = RT_TOS(skb->nh.iph->tos) } },
408                                     .proto = IPPROTO_ICMP };
409                 if (ip_route_output_key(&rt, &fl))
410                         goto out_unlock;
411         }
412         if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
413                                icmp_param->data.icmph.code))
414                 icmp_push_reply(icmp_param, &ipc, rt);
415         ip_rt_put(rt);
416 out_unlock:
417         icmp_xmit_unlock();
418 }
419
420
421 /*
422  *      Send an ICMP message in response to a situation
423  *
424  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
425  *                MAY send more (we do).
426  *                      MUST NOT change this header information.
427  *                      MUST NOT reply to a multicast/broadcast IP address.
428  *                      MUST NOT reply to a multicast/broadcast MAC address.
429  *                      MUST reply to only the first fragment.
430  */
431
432 void icmp_send(struct sk_buff *skb_in, int type, int code, u32 info)
433 {
434         struct iphdr *iph;
435         int room;
436         struct icmp_bxm icmp_param;
437         struct rtable *rt = (struct rtable *)skb_in->dst;
438         struct ipcm_cookie ipc;
439         u32 saddr;
440         u8  tos;
441
442         if (!rt)
443                 goto out;
444
445         /*
446          *      Find the original header. It is expected to be valid, of course.
447          *      Check this, icmp_send is called from the most obscure devices
448          *      sometimes.
449          */
450         iph = skb_in->nh.iph;
451
452         if ((u8 *)iph < skb_in->head || (u8 *)(iph + 1) > skb_in->tail)
453                 goto out;
454
455         /*
456          *      No replies to physical multicast/broadcast
457          */
458         if (skb_in->pkt_type != PACKET_HOST)
459                 goto out;
460
461         /*
462          *      Now check at the protocol level
463          */
464         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
465                 goto out;
466
467         /*
468          *      Only reply to fragment 0. We byte re-order the constant
469          *      mask for efficiency.
470          */
471         if (iph->frag_off & htons(IP_OFFSET))
472                 goto out;
473
474         /*
475          *      If we send an ICMP error to an ICMP error a mess would result..
476          */
477         if (icmp_pointers[type].error) {
478                 /*
479                  *      We are an error, check if we are replying to an
480                  *      ICMP error
481                  */
482                 if (iph->protocol == IPPROTO_ICMP) {
483                         u8 _inner_type, *itp;
484
485                         itp = skb_header_pointer(skb_in,
486                                                  skb_in->nh.raw +
487                                                  (iph->ihl << 2) +
488                                                  offsetof(struct icmphdr,
489                                                           type) -
490                                                  skb_in->data,
491                                                  sizeof(_inner_type),
492                                                  &_inner_type);
493                         if (itp == NULL)
494                                 goto out;
495
496                         /*
497                          *      Assume any unknown ICMP type is an error. This
498                          *      isn't specified by the RFC, but think about it..
499                          */
500                         if (*itp > NR_ICMP_TYPES ||
501                             icmp_pointers[*itp].error)
502                                 goto out;
503                 }
504         }
505
506         if (icmp_xmit_lock())
507                 return;
508
509         /*
510          *      Construct source address and options.
511          */
512
513         saddr = iph->daddr;
514         if (!(rt->rt_flags & RTCF_LOCAL)) {
515                 if (sysctl_icmp_errors_use_inbound_ifaddr)
516                         saddr = inet_select_addr(skb_in->dev, 0, RT_SCOPE_LINK);
517                 else
518                         saddr = 0;
519         }
520
521         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
522                                            IPTOS_PREC_INTERNETCONTROL) :
523                                           iph->tos;
524
525         if (ip_options_echo(&icmp_param.replyopts, skb_in))
526                 goto out_unlock;
527
528
529         /*
530          *      Prepare data for ICMP header.
531          */
532
533         icmp_param.data.icmph.type       = type;
534         icmp_param.data.icmph.code       = code;
535         icmp_param.data.icmph.un.gateway = info;
536         icmp_param.data.icmph.checksum   = 0;
537         icmp_param.skb    = skb_in;
538         icmp_param.offset = skb_in->nh.raw - skb_in->data;
539         icmp_out_count(icmp_param.data.icmph.type);
540         inet_sk(icmp_socket->sk)->tos = tos;
541         ipc.addr = iph->saddr;
542         ipc.opt = &icmp_param.replyopts;
543
544         {
545                 struct flowi fl = {
546                         .nl_u = {
547                                 .ip4_u = {
548                                         .daddr = icmp_param.replyopts.srr ?
549                                                 icmp_param.replyopts.faddr :
550                                                 iph->saddr,
551                                         .saddr = saddr,
552                                         .tos = RT_TOS(tos)
553                                 }
554                         },
555                         .proto = IPPROTO_ICMP,
556                         .uli_u = {
557                                 .icmpt = {
558                                         .type = type,
559                                         .code = code
560                                 }
561                         }
562                 };
563                 if (ip_route_output_key(&rt, &fl))
564                         goto out_unlock;
565         }
566
567         if (!icmpv4_xrlim_allow(rt, type, code))
568                 goto ende;
569
570         /* RFC says return as much as we can without exceeding 576 bytes. */
571
572         room = dst_mtu(&rt->u.dst);
573         if (room > 576)
574                 room = 576;
575         room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
576         room -= sizeof(struct icmphdr);
577
578         icmp_param.data_len = skb_in->len - icmp_param.offset;
579         if (icmp_param.data_len > room)
580                 icmp_param.data_len = room;
581         icmp_param.head_len = sizeof(struct icmphdr);
582
583         icmp_push_reply(&icmp_param, &ipc, rt);
584 ende:
585         ip_rt_put(rt);
586 out_unlock:
587         icmp_xmit_unlock();
588 out:;
589 }
590
591
592 /*
593  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
594  */
595
596 static void icmp_unreach(struct sk_buff *skb)
597 {
598         struct iphdr *iph;
599         struct icmphdr *icmph;
600         int hash, protocol;
601         struct net_protocol *ipprot;
602         struct sock *raw_sk;
603         u32 info = 0;
604
605         /*
606          *      Incomplete header ?
607          *      Only checks for the IP header, there should be an
608          *      additional check for longer headers in upper levels.
609          */
610
611         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
612                 goto out_err;
613
614         icmph = skb->h.icmph;
615         iph   = (struct iphdr *)skb->data;
616
617         if (iph->ihl < 5) /* Mangled header, drop. */
618                 goto out_err;
619
620         if (icmph->type == ICMP_DEST_UNREACH) {
621                 switch (icmph->code & 15) {
622                 case ICMP_NET_UNREACH:
623                 case ICMP_HOST_UNREACH:
624                 case ICMP_PROT_UNREACH:
625                 case ICMP_PORT_UNREACH:
626                         break;
627                 case ICMP_FRAG_NEEDED:
628                         if (ipv4_config.no_pmtu_disc) {
629                                 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
630                                                          "fragmentation needed "
631                                                          "and DF set.\n",
632                                                NIPQUAD(iph->daddr));
633                         } else {
634                                 info = ip_rt_frag_needed(iph,
635                                                      ntohs(icmph->un.frag.mtu));
636                                 if (!info)
637                                         goto out;
638                         }
639                         break;
640                 case ICMP_SR_FAILED:
641                         LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
642                                                  "Route Failed.\n",
643                                        NIPQUAD(iph->daddr));
644                         break;
645                 default:
646                         break;
647                 }
648                 if (icmph->code > NR_ICMP_UNREACH)
649                         goto out;
650         } else if (icmph->type == ICMP_PARAMETERPROB)
651                 info = ntohl(icmph->un.gateway) >> 24;
652
653         /*
654          *      Throw it at our lower layers
655          *
656          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
657          *                header.
658          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
659          *                transport layer.
660          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
661          *                transport layer.
662          */
663
664         /*
665          *      Check the other end isnt violating RFC 1122. Some routers send
666          *      bogus responses to broadcast frames. If you see this message
667          *      first check your netmask matches at both ends, if it does then
668          *      get the other vendor to fix their kit.
669          */
670
671         if (!sysctl_icmp_ignore_bogus_error_responses &&
672             inet_addr_type(iph->daddr) == RTN_BROADCAST) {
673                 if (net_ratelimit())
674                         printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
675                                             "type %u, code %u "
676                                             "error to a broadcast: %u.%u.%u.%u on %s\n",
677                                NIPQUAD(skb->nh.iph->saddr),
678                                icmph->type, icmph->code,
679                                NIPQUAD(iph->daddr),
680                                skb->dev->name);
681                 goto out;
682         }
683
684         /* Checkin full IP header plus 8 bytes of protocol to
685          * avoid additional coding at protocol handlers.
686          */
687         if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
688                 goto out;
689
690         iph = (struct iphdr *)skb->data;
691         protocol = iph->protocol;
692
693         /*
694          *      Deliver ICMP message to raw sockets. Pretty useless feature?
695          */
696
697         /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
698         hash = protocol & (MAX_INET_PROTOS - 1);
699         read_lock(&raw_v4_lock);
700         if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) {
701                 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
702                                                  iph->saddr,
703                                                  skb->dev->ifindex)) != NULL) {
704                         raw_err(raw_sk, skb, info);
705                         raw_sk = sk_next(raw_sk);
706                         iph = (struct iphdr *)skb->data;
707                 }
708         }
709         read_unlock(&raw_v4_lock);
710
711         rcu_read_lock();
712         ipprot = rcu_dereference(inet_protos[hash]);
713         if (ipprot && ipprot->err_handler)
714                 ipprot->err_handler(skb, info);
715         rcu_read_unlock();
716
717 out:
718         return;
719 out_err:
720         ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
721         goto out;
722 }
723
724
725 /*
726  *      Handle ICMP_REDIRECT.
727  */
728
729 static void icmp_redirect(struct sk_buff *skb)
730 {
731         struct iphdr *iph;
732
733         if (skb->len < sizeof(struct iphdr))
734                 goto out_err;
735
736         /*
737          *      Get the copied header of the packet that caused the redirect
738          */
739         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
740                 goto out;
741
742         iph = (struct iphdr *)skb->data;
743
744         switch (skb->h.icmph->code & 7) {
745         case ICMP_REDIR_NET:
746         case ICMP_REDIR_NETTOS:
747                 /*
748                  * As per RFC recommendations now handle it as a host redirect.
749                  */
750         case ICMP_REDIR_HOST:
751         case ICMP_REDIR_HOSTTOS:
752                 ip_rt_redirect(skb->nh.iph->saddr, iph->daddr,
753                                skb->h.icmph->un.gateway,
754                                iph->saddr, skb->dev);
755                 break;
756         }
757 out:
758         return;
759 out_err:
760         ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
761         goto out;
762 }
763
764 /*
765  *      Handle ICMP_ECHO ("ping") requests.
766  *
767  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
768  *                requests.
769  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
770  *                included in the reply.
771  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
772  *                echo requests, MUST have default=NOT.
773  *      See also WRT handling of options once they are done and working.
774  */
775
776 static void icmp_echo(struct sk_buff *skb)
777 {
778         if (!sysctl_icmp_echo_ignore_all) {
779                 struct icmp_bxm icmp_param;
780
781                 icmp_param.data.icmph      = *skb->h.icmph;
782                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
783                 icmp_param.skb             = skb;
784                 icmp_param.offset          = 0;
785                 icmp_param.data_len        = skb->len;
786                 icmp_param.head_len        = sizeof(struct icmphdr);
787                 icmp_reply(&icmp_param, skb);
788         }
789 }
790
791 /*
792  *      Handle ICMP Timestamp requests.
793  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
794  *                SHOULD be in the kernel for minimum random latency.
795  *                MUST be accurate to a few minutes.
796  *                MUST be updated at least at 15Hz.
797  */
798 static void icmp_timestamp(struct sk_buff *skb)
799 {
800         struct timeval tv;
801         struct icmp_bxm icmp_param;
802         /*
803          *      Too short.
804          */
805         if (skb->len < 4)
806                 goto out_err;
807
808         /*
809          *      Fill in the current time as ms since midnight UT:
810          */
811         do_gettimeofday(&tv);
812         icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
813                                          tv.tv_usec / 1000);
814         icmp_param.data.times[2] = icmp_param.data.times[1];
815         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
816                 BUG();
817         icmp_param.data.icmph      = *skb->h.icmph;
818         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
819         icmp_param.data.icmph.code = 0;
820         icmp_param.skb             = skb;
821         icmp_param.offset          = 0;
822         icmp_param.data_len        = 0;
823         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
824         icmp_reply(&icmp_param, skb);
825 out:
826         return;
827 out_err:
828         ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
829         goto out;
830 }
831
832
833 /*
834  *      Handle ICMP_ADDRESS_MASK requests.  (RFC950)
835  *
836  * RFC1122 (3.2.2.9).  A host MUST only send replies to
837  * ADDRESS_MASK requests if it's been configured as an address mask
838  * agent.  Receiving a request doesn't constitute implicit permission to
839  * act as one. Of course, implementing this correctly requires (SHOULD)
840  * a way to turn the functionality on and off.  Another one for sysctl(),
841  * I guess. -- MS
842  *
843  * RFC1812 (4.3.3.9).   A router MUST implement it.
844  *                      A router SHOULD have switch turning it on/off.
845  *                      This switch MUST be ON by default.
846  *
847  * Gratuitous replies, zero-source replies are not implemented,
848  * that complies with RFC. DO NOT implement them!!! All the idea
849  * of broadcast addrmask replies as specified in RFC950 is broken.
850  * The problem is that it is not uncommon to have several prefixes
851  * on one physical interface. Moreover, addrmask agent can even be
852  * not aware of existing another prefixes.
853  * If source is zero, addrmask agent cannot choose correct prefix.
854  * Gratuitous mask announcements suffer from the same problem.
855  * RFC1812 explains it, but still allows to use ADDRMASK,
856  * that is pretty silly. --ANK
857  *
858  * All these rules are so bizarre, that I removed kernel addrmask
859  * support at all. It is wrong, it is obsolete, nobody uses it in
860  * any case. --ANK
861  *
862  * Furthermore you can do it with a usermode address agent program
863  * anyway...
864  */
865
866 static void icmp_address(struct sk_buff *skb)
867 {
868 #if 0
869         if (net_ratelimit())
870                 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
871 #endif
872 }
873
874 /*
875  * RFC1812 (4.3.3.9).   A router SHOULD listen all replies, and complain
876  *                      loudly if an inconsistency is found.
877  */
878
879 static void icmp_address_reply(struct sk_buff *skb)
880 {
881         struct rtable *rt = (struct rtable *)skb->dst;
882         struct net_device *dev = skb->dev;
883         struct in_device *in_dev;
884         struct in_ifaddr *ifa;
885
886         if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
887                 goto out;
888
889         in_dev = in_dev_get(dev);
890         if (!in_dev)
891                 goto out;
892         rcu_read_lock();
893         if (in_dev->ifa_list &&
894             IN_DEV_LOG_MARTIANS(in_dev) &&
895             IN_DEV_FORWARD(in_dev)) {
896                 u32 _mask, *mp;
897
898                 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
899                 BUG_ON(mp == NULL);
900                 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
901                         if (*mp == ifa->ifa_mask &&
902                             inet_ifa_match(rt->rt_src, ifa))
903                                 break;
904                 }
905                 if (!ifa && net_ratelimit()) {
906                         printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
907                                          "%s/%u.%u.%u.%u\n",
908                                NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
909                 }
910         }
911         rcu_read_unlock();
912         in_dev_put(in_dev);
913 out:;
914 }
915
916 static void icmp_discard(struct sk_buff *skb)
917 {
918 }
919
920 /*
921  *      Deal with incoming ICMP packets.
922  */
923 int icmp_rcv(struct sk_buff *skb)
924 {
925         struct icmphdr *icmph;
926         struct rtable *rt = (struct rtable *)skb->dst;
927
928         ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
929
930         switch (skb->ip_summed) {
931         case CHECKSUM_HW:
932                 if (!(u16)csum_fold(skb->csum))
933                         break;
934                 /* fall through */
935         case CHECKSUM_NONE:
936                 skb->csum = 0;
937                 if (__skb_checksum_complete(skb))
938                         goto error;
939         }
940
941         if (!pskb_pull(skb, sizeof(struct icmphdr)))
942                 goto error;
943
944         icmph = skb->h.icmph;
945
946         /*
947          *      18 is the highest 'known' ICMP type. Anything else is a mystery
948          *
949          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
950          *                discarded.
951          */
952         if (icmph->type > NR_ICMP_TYPES)
953                 goto error;
954
955
956         /*
957          *      Parse the ICMP message
958          */
959
960         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
961                 /*
962                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
963                  *        silently ignored (we let user decide with a sysctl).
964                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
965                  *        discarded if to broadcast/multicast.
966                  */
967                 if ((icmph->type == ICMP_ECHO ||
968                      icmph->type == ICMP_TIMESTAMP) &&
969                     sysctl_icmp_echo_ignore_broadcasts) {
970                         goto error;
971                 }
972                 if (icmph->type != ICMP_ECHO &&
973                     icmph->type != ICMP_TIMESTAMP &&
974                     icmph->type != ICMP_ADDRESS &&
975                     icmph->type != ICMP_ADDRESSREPLY) {
976                         goto error;
977                 }
978         }
979
980         ICMP_INC_STATS_BH(icmp_pointers[icmph->type].input_entry);
981         icmp_pointers[icmph->type].handler(skb);
982
983 drop:
984         kfree_skb(skb);
985         return 0;
986 error:
987         ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
988         goto drop;
989 }
990
991 /*
992  *      This table is the definition of how we handle ICMP.
993  */
994 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
995         [ICMP_ECHOREPLY] = {
996                 .output_entry = ICMP_MIB_OUTECHOREPS,
997                 .input_entry = ICMP_MIB_INECHOREPS,
998                 .handler = icmp_discard,
999         },
1000         [1] = {
1001                 .output_entry = ICMP_MIB_DUMMY,
1002                 .input_entry = ICMP_MIB_INERRORS,
1003                 .handler = icmp_discard,
1004                 .error = 1,
1005         },
1006         [2] = {
1007                 .output_entry = ICMP_MIB_DUMMY,
1008                 .input_entry = ICMP_MIB_INERRORS,
1009                 .handler = icmp_discard,
1010                 .error = 1,
1011         },
1012         [ICMP_DEST_UNREACH] = {
1013                 .output_entry = ICMP_MIB_OUTDESTUNREACHS,
1014                 .input_entry = ICMP_MIB_INDESTUNREACHS,
1015                 .handler = icmp_unreach,
1016                 .error = 1,
1017         },
1018         [ICMP_SOURCE_QUENCH] = {
1019                 .output_entry = ICMP_MIB_OUTSRCQUENCHS,
1020                 .input_entry = ICMP_MIB_INSRCQUENCHS,
1021                 .handler = icmp_unreach,
1022                 .error = 1,
1023         },
1024         [ICMP_REDIRECT] = {
1025                 .output_entry = ICMP_MIB_OUTREDIRECTS,
1026                 .input_entry = ICMP_MIB_INREDIRECTS,
1027                 .handler = icmp_redirect,
1028                 .error = 1,
1029         },
1030         [6] = {
1031                 .output_entry = ICMP_MIB_DUMMY,
1032                 .input_entry = ICMP_MIB_INERRORS,
1033                 .handler = icmp_discard,
1034                 .error = 1,
1035         },
1036         [7] = {
1037                 .output_entry = ICMP_MIB_DUMMY,
1038                 .input_entry = ICMP_MIB_INERRORS,
1039                 .handler = icmp_discard,
1040                 .error = 1,
1041         },
1042         [ICMP_ECHO] = {
1043                 .output_entry = ICMP_MIB_OUTECHOS,
1044                 .input_entry = ICMP_MIB_INECHOS,
1045                 .handler = icmp_echo,
1046         },
1047         [9] = {
1048                 .output_entry = ICMP_MIB_DUMMY,
1049                 .input_entry = ICMP_MIB_INERRORS,
1050                 .handler = icmp_discard,
1051                 .error = 1,
1052         },
1053         [10] = {
1054                 .output_entry = ICMP_MIB_DUMMY,
1055                 .input_entry = ICMP_MIB_INERRORS,
1056                 .handler = icmp_discard,
1057                 .error = 1,
1058         },
1059         [ICMP_TIME_EXCEEDED] = {
1060                 .output_entry = ICMP_MIB_OUTTIMEEXCDS,
1061                 .input_entry = ICMP_MIB_INTIMEEXCDS,
1062                 .handler = icmp_unreach,
1063                 .error = 1,
1064         },
1065         [ICMP_PARAMETERPROB] = {
1066                 .output_entry = ICMP_MIB_OUTPARMPROBS,
1067                 .input_entry = ICMP_MIB_INPARMPROBS,
1068                 .handler = icmp_unreach,
1069                 .error = 1,
1070         },
1071         [ICMP_TIMESTAMP] = {
1072                 .output_entry = ICMP_MIB_OUTTIMESTAMPS,
1073                 .input_entry = ICMP_MIB_INTIMESTAMPS,
1074                 .handler = icmp_timestamp,
1075         },
1076         [ICMP_TIMESTAMPREPLY] = {
1077                 .output_entry = ICMP_MIB_OUTTIMESTAMPREPS,
1078                 .input_entry = ICMP_MIB_INTIMESTAMPREPS,
1079                 .handler = icmp_discard,
1080         },
1081         [ICMP_INFO_REQUEST] = {
1082                 .output_entry = ICMP_MIB_DUMMY,
1083                 .input_entry = ICMP_MIB_DUMMY,
1084                 .handler = icmp_discard,
1085         },
1086         [ICMP_INFO_REPLY] = {
1087                 .output_entry = ICMP_MIB_DUMMY,
1088                 .input_entry = ICMP_MIB_DUMMY,
1089                 .handler = icmp_discard,
1090         },
1091         [ICMP_ADDRESS] = {
1092                 .output_entry = ICMP_MIB_OUTADDRMASKS,
1093                 .input_entry = ICMP_MIB_INADDRMASKS,
1094                 .handler = icmp_address,
1095         },
1096         [ICMP_ADDRESSREPLY] = {
1097                 .output_entry = ICMP_MIB_OUTADDRMASKREPS,
1098                 .input_entry = ICMP_MIB_INADDRMASKREPS,
1099                 .handler = icmp_address_reply,
1100         },
1101 };
1102
1103 void __init icmp_init(struct net_proto_family *ops)
1104 {
1105         struct inet_sock *inet;
1106         int i;
1107
1108         for_each_possible_cpu(i) {
1109                 int err;
1110
1111                 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1112                                        &per_cpu(__icmp_socket, i));
1113
1114                 if (err < 0)
1115                         panic("Failed to create the ICMP control socket.\n");
1116
1117                 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1118
1119                 /* Enough space for 2 64K ICMP packets, including
1120                  * sk_buff struct overhead.
1121                  */
1122                 per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1123                         (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1124
1125                 inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1126                 inet->uc_ttl = -1;
1127                 inet->pmtudisc = IP_PMTUDISC_DONT;
1128
1129                 /* Unhash it so that IP input processing does not even
1130                  * see it, we do not wish this socket to see incoming
1131                  * packets.
1132                  */
1133                 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1134         }
1135 }
1136
1137 EXPORT_SYMBOL(icmp_err_convert);
1138 EXPORT_SYMBOL(icmp_send);
1139 EXPORT_SYMBOL(icmp_statistics);
1140 EXPORT_SYMBOL(xrlim_allow);