IB/mthca: Optimize large messages on Sinai HCAs
[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/config.h>
68 #include <linux/module.h>
69 #include <linux/types.h>
70 #include <linux/jiffies.h>
71 #include <linux/kernel.h>
72 #include <linux/fcntl.h>
73 #include <linux/socket.h>
74 #include <linux/in.h>
75 #include <linux/inet.h>
76 #include <linux/inetdevice.h>
77 #include <linux/netdevice.h>
78 #include <linux/string.h>
79 #include <linux/netfilter_ipv4.h>
80 #include <net/snmp.h>
81 #include <net/ip.h>
82 #include <net/route.h>
83 #include <net/protocol.h>
84 #include <net/icmp.h>
85 #include <net/tcp.h>
86 #include <net/udp.h>
87 #include <net/raw.h>
88 #include <linux/skbuff.h>
89 #include <net/sock.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
92 #include <linux/init.h>
93 #include <asm/system.h>
94 #include <asm/uaccess.h>
95 #include <net/checksum.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                 __u32          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
120 /* An array of errno for error messages from dest unreach. */
121 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
122
123 struct icmp_err icmp_err_convert[] = {
124         {
125                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
126                 .fatal = 0,
127         },
128         {
129                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
130                 .fatal = 0,
131         },
132         {
133                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
134                 .fatal = 1,
135         },
136         {
137                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
138                 .fatal = 1,
139         },
140         {
141                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
142                 .fatal = 0,
143         },
144         {
145                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
146                 .fatal = 0,
147         },
148         {
149                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
150                 .fatal = 1,
151         },
152         {
153                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
154                 .fatal = 1,
155         },
156         {
157                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
158                 .fatal = 1,
159         },
160         {
161                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
162                 .fatal = 1,
163         },
164         {
165                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
166                 .fatal = 1,
167         },
168         {
169                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
170                 .fatal = 0,
171         },
172         {
173                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
174                 .fatal = 0,
175         },
176         {
177                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
178                 .fatal = 1,
179         },
180         {
181                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
182                 .fatal = 1,
183         },
184         {
185                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
186                 .fatal = 1,
187         },
188 };
189
190 /* Control parameters for ECHO replies. */
191 int sysctl_icmp_echo_ignore_all;
192 int sysctl_icmp_echo_ignore_broadcasts = 1;
193
194 /* Control parameter - ignore bogus broadcast responses? */
195 int sysctl_icmp_ignore_bogus_error_responses = 1;
196
197 /*
198  *      Configurable global rate limit.
199  *
200  *      ratelimit defines tokens/packet consumed for dst->rate_token bucket
201  *      ratemask defines which icmp types are ratelimited by setting
202  *      it's bit position.
203  *
204  *      default:
205  *      dest unreachable (3), source quench (4),
206  *      time exceeded (11), parameter problem (12)
207  */
208
209 int sysctl_icmp_ratelimit = 1 * HZ;
210 int sysctl_icmp_ratemask = 0x1818;
211 int sysctl_icmp_errors_use_inbound_ifaddr;
212
213 /*
214  *      ICMP control array. This specifies what to do with each ICMP.
215  */
216
217 struct icmp_control {
218         int output_entry;       /* Field for increment on output */
219         int input_entry;        /* Field for increment on input */
220         void (*handler)(struct sk_buff *skb);
221         short   error;          /* This ICMP is classed as an error message */
222 };
223
224 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
225
226 /*
227  *      The ICMP socket(s). This is the most convenient way to flow control
228  *      our ICMP output as well as maintain a clean interface throughout
229  *      all layers. All Socketless IP sends will soon be gone.
230  *
231  *      On SMP we have one ICMP socket per-cpu.
232  */
233 static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL;
234 #define icmp_socket     __get_cpu_var(__icmp_socket)
235
236 static __inline__ int icmp_xmit_lock(void)
237 {
238         local_bh_disable();
239
240         if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) {
241                 /* This can happen if the output path signals a
242                  * dst_link_failure() for an outgoing ICMP packet.
243                  */
244                 local_bh_enable();
245                 return 1;
246         }
247         return 0;
248 }
249
250 static void icmp_xmit_unlock(void)
251 {
252         spin_unlock_bh(&icmp_socket->sk->sk_lock.slock);
253 }
254
255 /*
256  *      Send an ICMP frame.
257  */
258
259 /*
260  *      Check transmit rate limitation for given message.
261  *      The rate information is held in the destination cache now.
262  *      This function is generic and could be used for other purposes
263  *      too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
264  *
265  *      Note that the same dst_entry fields are modified by functions in
266  *      route.c too, but these work for packet destinations while xrlim_allow
267  *      works for icmp destinations. This means the rate limiting information
268  *      for one "ip object" is shared - and these ICMPs are twice limited:
269  *      by source and by destination.
270  *
271  *      RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
272  *                        SHOULD allow setting of rate limits
273  *
274  *      Shared between ICMPv4 and ICMPv6.
275  */
276 #define XRLIM_BURST_FACTOR 6
277 int xrlim_allow(struct dst_entry *dst, int timeout)
278 {
279         unsigned long now;
280         int rc = 0;
281
282         now = jiffies;
283         dst->rate_tokens += now - dst->rate_last;
284         dst->rate_last = now;
285         if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout)
286                 dst->rate_tokens = XRLIM_BURST_FACTOR * timeout;
287         if (dst->rate_tokens >= timeout) {
288                 dst->rate_tokens -= timeout;
289                 rc = 1;
290         }
291         return rc;
292 }
293
294 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
295 {
296         struct dst_entry *dst = &rt->u.dst;
297         int rc = 1;
298
299         if (type > NR_ICMP_TYPES)
300                 goto out;
301
302         /* Don't limit PMTU discovery. */
303         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
304                 goto out;
305
306         /* No rate limit on loopback */
307         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
308                 goto out;
309
310         /* Limit if icmp type is enabled in ratemask. */
311         if ((1 << type) & sysctl_icmp_ratemask)
312                 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
313 out:
314         return rc;
315 }
316
317 /*
318  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
319  */
320 static void icmp_out_count(int type)
321 {
322         if (type <= NR_ICMP_TYPES) {
323                 ICMP_INC_STATS(icmp_pointers[type].output_entry);
324                 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
325         }
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         unsigned int 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 sk_buff *skb;
352
353         if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
354                            icmp_param->data_len+icmp_param->head_len,
355                            icmp_param->head_len,
356                            ipc, rt, MSG_DONTWAIT) < 0)
357                 ip_flush_pending_frames(icmp_socket->sk);
358         else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
359                 struct icmphdr *icmph = skb->h.icmph;
360                 unsigned int csum = 0;
361                 struct sk_buff *skb1;
362
363                 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
364                         csum = csum_add(csum, skb1->csum);
365                 }
366                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
367                                                  (char *)icmph,
368                                                  icmp_param->head_len, csum);
369                 icmph->checksum = csum_fold(csum);
370                 skb->ip_summed = CHECKSUM_NONE;
371                 ip_push_pending_frames(icmp_socket->sk);
372         }
373 }
374
375 /*
376  *      Driving logic for building and sending ICMP messages.
377  */
378
379 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
380 {
381         struct sock *sk = icmp_socket->sk;
382         struct inet_sock *inet = inet_sk(sk);
383         struct ipcm_cookie ipc;
384         struct rtable *rt = (struct rtable *)skb->dst;
385         u32 daddr;
386
387         if (ip_options_echo(&icmp_param->replyopts, skb))
388                 return;
389
390         if (icmp_xmit_lock())
391                 return;
392
393         icmp_param->data.icmph.checksum = 0;
394         icmp_out_count(icmp_param->data.icmph.type);
395
396         inet->tos = skb->nh.iph->tos;
397         daddr = ipc.addr = rt->rt_src;
398         ipc.opt = NULL;
399         if (icmp_param->replyopts.optlen) {
400                 ipc.opt = &icmp_param->replyopts;
401                 if (ipc.opt->srr)
402                         daddr = icmp_param->replyopts.faddr;
403         }
404         {
405                 struct flowi fl = { .nl_u = { .ip4_u =
406                                               { .daddr = daddr,
407                                                 .saddr = rt->rt_spec_dst,
408                                                 .tos = RT_TOS(skb->nh.iph->tos) } },
409                                     .proto = IPPROTO_ICMP };
410                 if (ip_route_output_key(&rt, &fl))
411                         goto out_unlock;
412         }
413         if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
414                                icmp_param->data.icmph.code))
415                 icmp_push_reply(icmp_param, &ipc, rt);
416         ip_rt_put(rt);
417 out_unlock:
418         icmp_xmit_unlock();
419 }
420
421
422 /*
423  *      Send an ICMP message in response to a situation
424  *
425  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
426  *                MAY send more (we do).
427  *                      MUST NOT change this header information.
428  *                      MUST NOT reply to a multicast/broadcast IP address.
429  *                      MUST NOT reply to a multicast/broadcast MAC address.
430  *                      MUST reply to only the first fragment.
431  */
432
433 void icmp_send(struct sk_buff *skb_in, int type, int code, u32 info)
434 {
435         struct iphdr *iph;
436         int room;
437         struct icmp_bxm icmp_param;
438         struct rtable *rt = (struct rtable *)skb_in->dst;
439         struct ipcm_cookie ipc;
440         u32 saddr;
441         u8  tos;
442
443         if (!rt)
444                 goto out;
445
446         /*
447          *      Find the original header. It is expected to be valid, of course.
448          *      Check this, icmp_send is called from the most obscure devices
449          *      sometimes.
450          */
451         iph = skb_in->nh.iph;
452
453         if ((u8 *)iph < skb_in->head || (u8 *)(iph + 1) > skb_in->tail)
454                 goto out;
455
456         /*
457          *      No replies to physical multicast/broadcast
458          */
459         if (skb_in->pkt_type != PACKET_HOST)
460                 goto out;
461
462         /*
463          *      Now check at the protocol level
464          */
465         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
466                 goto out;
467
468         /*
469          *      Only reply to fragment 0. We byte re-order the constant
470          *      mask for efficiency.
471          */
472         if (iph->frag_off & htons(IP_OFFSET))
473                 goto out;
474
475         /*
476          *      If we send an ICMP error to an ICMP error a mess would result..
477          */
478         if (icmp_pointers[type].error) {
479                 /*
480                  *      We are an error, check if we are replying to an
481                  *      ICMP error
482                  */
483                 if (iph->protocol == IPPROTO_ICMP) {
484                         u8 _inner_type, *itp;
485
486                         itp = skb_header_pointer(skb_in,
487                                                  skb_in->nh.raw +
488                                                  (iph->ihl << 2) +
489                                                  offsetof(struct icmphdr,
490                                                           type) -
491                                                  skb_in->data,
492                                                  sizeof(_inner_type),
493                                                  &_inner_type);
494                         if (itp == NULL)
495                                 goto out;
496
497                         /*
498                          *      Assume any unknown ICMP type is an error. This
499                          *      isn't specified by the RFC, but think about it..
500                          */
501                         if (*itp > NR_ICMP_TYPES ||
502                             icmp_pointers[*itp].error)
503                                 goto out;
504                 }
505         }
506
507         if (icmp_xmit_lock())
508                 return;
509
510         /*
511          *      Construct source address and options.
512          */
513
514         saddr = iph->daddr;
515         if (!(rt->rt_flags & RTCF_LOCAL)) {
516                 if (sysctl_icmp_errors_use_inbound_ifaddr)
517                         saddr = inet_select_addr(skb_in->dev, 0, RT_SCOPE_LINK);
518                 else
519                         saddr = 0;
520         }
521
522         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
523                                            IPTOS_PREC_INTERNETCONTROL) :
524                                           iph->tos;
525
526         if (ip_options_echo(&icmp_param.replyopts, skb_in))
527                 goto out_unlock;
528
529
530         /*
531          *      Prepare data for ICMP header.
532          */
533
534         icmp_param.data.icmph.type       = type;
535         icmp_param.data.icmph.code       = code;
536         icmp_param.data.icmph.un.gateway = info;
537         icmp_param.data.icmph.checksum   = 0;
538         icmp_param.skb    = skb_in;
539         icmp_param.offset = skb_in->nh.raw - skb_in->data;
540         icmp_out_count(icmp_param.data.icmph.type);
541         inet_sk(icmp_socket->sk)->tos = tos;
542         ipc.addr = iph->saddr;
543         ipc.opt = &icmp_param.replyopts;
544
545         {
546                 struct flowi fl = {
547                         .nl_u = {
548                                 .ip4_u = {
549                                         .daddr = icmp_param.replyopts.srr ?
550                                                 icmp_param.replyopts.faddr :
551                                                 iph->saddr,
552                                         .saddr = saddr,
553                                         .tos = RT_TOS(tos)
554                                 }
555                         },
556                         .proto = IPPROTO_ICMP,
557                         .uli_u = {
558                                 .icmpt = {
559                                         .type = type,
560                                         .code = code
561                                 }
562                         }
563                 };
564                 if (ip_route_output_key(&rt, &fl))
565                         goto out_unlock;
566         }
567
568         if (!icmpv4_xrlim_allow(rt, type, code))
569                 goto ende;
570
571         /* RFC says return as much as we can without exceeding 576 bytes. */
572
573         room = dst_mtu(&rt->u.dst);
574         if (room > 576)
575                 room = 576;
576         room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
577         room -= sizeof(struct icmphdr);
578
579         icmp_param.data_len = skb_in->len - icmp_param.offset;
580         if (icmp_param.data_len > room)
581                 icmp_param.data_len = room;
582         icmp_param.head_len = sizeof(struct icmphdr);
583
584         icmp_push_reply(&icmp_param, &ipc, rt);
585 ende:
586         ip_rt_put(rt);
587 out_unlock:
588         icmp_xmit_unlock();
589 out:;
590 }
591
592
593 /*
594  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
595  */
596
597 static void icmp_unreach(struct sk_buff *skb)
598 {
599         struct iphdr *iph;
600         struct icmphdr *icmph;
601         int hash, protocol;
602         struct net_protocol *ipprot;
603         struct sock *raw_sk;
604         u32 info = 0;
605
606         /*
607          *      Incomplete header ?
608          *      Only checks for the IP header, there should be an
609          *      additional check for longer headers in upper levels.
610          */
611
612         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
613                 goto out_err;
614
615         icmph = skb->h.icmph;
616         iph   = (struct iphdr *)skb->data;
617
618         if (iph->ihl < 5) /* Mangled header, drop. */
619                 goto out_err;
620
621         if (icmph->type == ICMP_DEST_UNREACH) {
622                 switch (icmph->code & 15) {
623                 case ICMP_NET_UNREACH:
624                 case ICMP_HOST_UNREACH:
625                 case ICMP_PROT_UNREACH:
626                 case ICMP_PORT_UNREACH:
627                         break;
628                 case ICMP_FRAG_NEEDED:
629                         if (ipv4_config.no_pmtu_disc) {
630                                 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
631                                                          "fragmentation needed "
632                                                          "and DF set.\n",
633                                                NIPQUAD(iph->daddr));
634                         } else {
635                                 info = ip_rt_frag_needed(iph,
636                                                      ntohs(icmph->un.frag.mtu));
637                                 if (!info)
638                                         goto out;
639                         }
640                         break;
641                 case ICMP_SR_FAILED:
642                         LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
643                                                  "Route Failed.\n",
644                                        NIPQUAD(iph->daddr));
645                         break;
646                 default:
647                         break;
648                 }
649                 if (icmph->code > NR_ICMP_UNREACH)
650                         goto out;
651         } else if (icmph->type == ICMP_PARAMETERPROB)
652                 info = ntohl(icmph->un.gateway) >> 24;
653
654         /*
655          *      Throw it at our lower layers
656          *
657          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
658          *                header.
659          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
660          *                transport layer.
661          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
662          *                transport layer.
663          */
664
665         /*
666          *      Check the other end isnt violating RFC 1122. Some routers send
667          *      bogus responses to broadcast frames. If you see this message
668          *      first check your netmask matches at both ends, if it does then
669          *      get the other vendor to fix their kit.
670          */
671
672         if (!sysctl_icmp_ignore_bogus_error_responses &&
673             inet_addr_type(iph->daddr) == RTN_BROADCAST) {
674                 if (net_ratelimit())
675                         printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
676                                             "type %u, code %u "
677                                             "error to a broadcast: %u.%u.%u.%u on %s\n",
678                                NIPQUAD(skb->nh.iph->saddr),
679                                icmph->type, icmph->code,
680                                NIPQUAD(iph->daddr),
681                                skb->dev->name);
682                 goto out;
683         }
684
685         /* Checkin full IP header plus 8 bytes of protocol to
686          * avoid additional coding at protocol handlers.
687          */
688         if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
689                 goto out;
690
691         iph = (struct iphdr *)skb->data;
692         protocol = iph->protocol;
693
694         /*
695          *      Deliver ICMP message to raw sockets. Pretty useless feature?
696          */
697
698         /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
699         hash = protocol & (MAX_INET_PROTOS - 1);
700         read_lock(&raw_v4_lock);
701         if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) {
702                 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
703                                                  iph->saddr,
704                                                  skb->dev->ifindex)) != NULL) {
705                         raw_err(raw_sk, skb, info);
706                         raw_sk = sk_next(raw_sk);
707                         iph = (struct iphdr *)skb->data;
708                 }
709         }
710         read_unlock(&raw_v4_lock);
711
712         rcu_read_lock();
713         ipprot = rcu_dereference(inet_protos[hash]);
714         if (ipprot && ipprot->err_handler)
715                 ipprot->err_handler(skb, info);
716         rcu_read_unlock();
717
718 out:
719         return;
720 out_err:
721         ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
722         goto out;
723 }
724
725
726 /*
727  *      Handle ICMP_REDIRECT.
728  */
729
730 static void icmp_redirect(struct sk_buff *skb)
731 {
732         struct iphdr *iph;
733         unsigned long ip;
734
735         if (skb->len < sizeof(struct iphdr))
736                 goto out_err;
737
738         /*
739          *      Get the copied header of the packet that caused the redirect
740          */
741         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
742                 goto out;
743
744         iph = (struct iphdr *)skb->data;
745         ip = iph->daddr;
746
747         switch (skb->h.icmph->code & 7) {
748         case ICMP_REDIR_NET:
749         case ICMP_REDIR_NETTOS:
750                 /*
751                  * As per RFC recommendations now handle it as a host redirect.
752                  */
753         case ICMP_REDIR_HOST:
754         case ICMP_REDIR_HOSTTOS:
755                 ip_rt_redirect(skb->nh.iph->saddr, ip, skb->h.icmph->un.gateway,
756                                iph->saddr, iph->tos, skb->dev);
757                 break;
758         }
759 out:
760         return;
761 out_err:
762         ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
763         goto out;
764 }
765
766 /*
767  *      Handle ICMP_ECHO ("ping") requests.
768  *
769  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
770  *                requests.
771  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
772  *                included in the reply.
773  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
774  *                echo requests, MUST have default=NOT.
775  *      See also WRT handling of options once they are done and working.
776  */
777
778 static void icmp_echo(struct sk_buff *skb)
779 {
780         if (!sysctl_icmp_echo_ignore_all) {
781                 struct icmp_bxm icmp_param;
782
783                 icmp_param.data.icmph      = *skb->h.icmph;
784                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
785                 icmp_param.skb             = skb;
786                 icmp_param.offset          = 0;
787                 icmp_param.data_len        = skb->len;
788                 icmp_param.head_len        = sizeof(struct icmphdr);
789                 icmp_reply(&icmp_param, skb);
790         }
791 }
792
793 /*
794  *      Handle ICMP Timestamp requests.
795  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
796  *                SHOULD be in the kernel for minimum random latency.
797  *                MUST be accurate to a few minutes.
798  *                MUST be updated at least at 15Hz.
799  */
800 static void icmp_timestamp(struct sk_buff *skb)
801 {
802         struct timeval tv;
803         struct icmp_bxm icmp_param;
804         /*
805          *      Too short.
806          */
807         if (skb->len < 4)
808                 goto out_err;
809
810         /*
811          *      Fill in the current time as ms since midnight UT:
812          */
813         do_gettimeofday(&tv);
814         icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
815                                          tv.tv_usec / 1000);
816         icmp_param.data.times[2] = icmp_param.data.times[1];
817         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
818                 BUG();
819         icmp_param.data.icmph      = *skb->h.icmph;
820         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
821         icmp_param.data.icmph.code = 0;
822         icmp_param.skb             = skb;
823         icmp_param.offset          = 0;
824         icmp_param.data_len        = 0;
825         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
826         icmp_reply(&icmp_param, skb);
827 out:
828         return;
829 out_err:
830         ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
831         goto out;
832 }
833
834
835 /*
836  *      Handle ICMP_ADDRESS_MASK requests.  (RFC950)
837  *
838  * RFC1122 (3.2.2.9).  A host MUST only send replies to
839  * ADDRESS_MASK requests if it's been configured as an address mask
840  * agent.  Receiving a request doesn't constitute implicit permission to
841  * act as one. Of course, implementing this correctly requires (SHOULD)
842  * a way to turn the functionality on and off.  Another one for sysctl(),
843  * I guess. -- MS
844  *
845  * RFC1812 (4.3.3.9).   A router MUST implement it.
846  *                      A router SHOULD have switch turning it on/off.
847  *                      This switch MUST be ON by default.
848  *
849  * Gratuitous replies, zero-source replies are not implemented,
850  * that complies with RFC. DO NOT implement them!!! All the idea
851  * of broadcast addrmask replies as specified in RFC950 is broken.
852  * The problem is that it is not uncommon to have several prefixes
853  * on one physical interface. Moreover, addrmask agent can even be
854  * not aware of existing another prefixes.
855  * If source is zero, addrmask agent cannot choose correct prefix.
856  * Gratuitous mask announcements suffer from the same problem.
857  * RFC1812 explains it, but still allows to use ADDRMASK,
858  * that is pretty silly. --ANK
859  *
860  * All these rules are so bizarre, that I removed kernel addrmask
861  * support at all. It is wrong, it is obsolete, nobody uses it in
862  * any case. --ANK
863  *
864  * Furthermore you can do it with a usermode address agent program
865  * anyway...
866  */
867
868 static void icmp_address(struct sk_buff *skb)
869 {
870 #if 0
871         if (net_ratelimit())
872                 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
873 #endif
874 }
875
876 /*
877  * RFC1812 (4.3.3.9).   A router SHOULD listen all replies, and complain
878  *                      loudly if an inconsistency is found.
879  */
880
881 static void icmp_address_reply(struct sk_buff *skb)
882 {
883         struct rtable *rt = (struct rtable *)skb->dst;
884         struct net_device *dev = skb->dev;
885         struct in_device *in_dev;
886         struct in_ifaddr *ifa;
887
888         if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
889                 goto out;
890
891         in_dev = in_dev_get(dev);
892         if (!in_dev)
893                 goto out;
894         rcu_read_lock();
895         if (in_dev->ifa_list &&
896             IN_DEV_LOG_MARTIANS(in_dev) &&
897             IN_DEV_FORWARD(in_dev)) {
898                 u32 _mask, *mp;
899
900                 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
901                 BUG_ON(mp == NULL);
902                 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
903                         if (*mp == ifa->ifa_mask &&
904                             inet_ifa_match(rt->rt_src, ifa))
905                                 break;
906                 }
907                 if (!ifa && net_ratelimit()) {
908                         printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
909                                          "%s/%u.%u.%u.%u\n",
910                                NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
911                 }
912         }
913         rcu_read_unlock();
914         in_dev_put(in_dev);
915 out:;
916 }
917
918 static void icmp_discard(struct sk_buff *skb)
919 {
920 }
921
922 /*
923  *      Deal with incoming ICMP packets.
924  */
925 int icmp_rcv(struct sk_buff *skb)
926 {
927         struct icmphdr *icmph;
928         struct rtable *rt = (struct rtable *)skb->dst;
929
930         ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
931
932         switch (skb->ip_summed) {
933         case CHECKSUM_HW:
934                 if (!(u16)csum_fold(skb->csum))
935                         break;
936                 /* fall through */
937         case CHECKSUM_NONE:
938                 skb->csum = 0;
939                 if (__skb_checksum_complete(skb))
940                         goto error;
941         }
942
943         if (!pskb_pull(skb, sizeof(struct icmphdr)))
944                 goto error;
945
946         icmph = skb->h.icmph;
947
948         /*
949          *      18 is the highest 'known' ICMP type. Anything else is a mystery
950          *
951          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
952          *                discarded.
953          */
954         if (icmph->type > NR_ICMP_TYPES)
955                 goto error;
956
957
958         /*
959          *      Parse the ICMP message
960          */
961
962         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
963                 /*
964                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
965                  *        silently ignored (we let user decide with a sysctl).
966                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
967                  *        discarded if to broadcast/multicast.
968                  */
969                 if ((icmph->type == ICMP_ECHO ||
970                      icmph->type == ICMP_TIMESTAMP) &&
971                     sysctl_icmp_echo_ignore_broadcasts) {
972                         goto error;
973                 }
974                 if (icmph->type != ICMP_ECHO &&
975                     icmph->type != ICMP_TIMESTAMP &&
976                     icmph->type != ICMP_ADDRESS &&
977                     icmph->type != ICMP_ADDRESSREPLY) {
978                         goto error;
979                 }
980         }
981
982         ICMP_INC_STATS_BH(icmp_pointers[icmph->type].input_entry);
983         icmp_pointers[icmph->type].handler(skb);
984
985 drop:
986         kfree_skb(skb);
987         return 0;
988 error:
989         ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
990         goto drop;
991 }
992
993 /*
994  *      This table is the definition of how we handle ICMP.
995  */
996 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
997         [ICMP_ECHOREPLY] = {
998                 .output_entry = ICMP_MIB_OUTECHOREPS,
999                 .input_entry = ICMP_MIB_INECHOREPS,
1000                 .handler = icmp_discard,
1001         },
1002         [1] = {
1003                 .output_entry = ICMP_MIB_DUMMY,
1004                 .input_entry = ICMP_MIB_INERRORS,
1005                 .handler = icmp_discard,
1006                 .error = 1,
1007         },
1008         [2] = {
1009                 .output_entry = ICMP_MIB_DUMMY,
1010                 .input_entry = ICMP_MIB_INERRORS,
1011                 .handler = icmp_discard,
1012                 .error = 1,
1013         },
1014         [ICMP_DEST_UNREACH] = {
1015                 .output_entry = ICMP_MIB_OUTDESTUNREACHS,
1016                 .input_entry = ICMP_MIB_INDESTUNREACHS,
1017                 .handler = icmp_unreach,
1018                 .error = 1,
1019         },
1020         [ICMP_SOURCE_QUENCH] = {
1021                 .output_entry = ICMP_MIB_OUTSRCQUENCHS,
1022                 .input_entry = ICMP_MIB_INSRCQUENCHS,
1023                 .handler = icmp_unreach,
1024                 .error = 1,
1025         },
1026         [ICMP_REDIRECT] = {
1027                 .output_entry = ICMP_MIB_OUTREDIRECTS,
1028                 .input_entry = ICMP_MIB_INREDIRECTS,
1029                 .handler = icmp_redirect,
1030                 .error = 1,
1031         },
1032         [6] = {
1033                 .output_entry = ICMP_MIB_DUMMY,
1034                 .input_entry = ICMP_MIB_INERRORS,
1035                 .handler = icmp_discard,
1036                 .error = 1,
1037         },
1038         [7] = {
1039                 .output_entry = ICMP_MIB_DUMMY,
1040                 .input_entry = ICMP_MIB_INERRORS,
1041                 .handler = icmp_discard,
1042                 .error = 1,
1043         },
1044         [ICMP_ECHO] = {
1045                 .output_entry = ICMP_MIB_OUTECHOS,
1046                 .input_entry = ICMP_MIB_INECHOS,
1047                 .handler = icmp_echo,
1048         },
1049         [9] = {
1050                 .output_entry = ICMP_MIB_DUMMY,
1051                 .input_entry = ICMP_MIB_INERRORS,
1052                 .handler = icmp_discard,
1053                 .error = 1,
1054         },
1055         [10] = {
1056                 .output_entry = ICMP_MIB_DUMMY,
1057                 .input_entry = ICMP_MIB_INERRORS,
1058                 .handler = icmp_discard,
1059                 .error = 1,
1060         },
1061         [ICMP_TIME_EXCEEDED] = {
1062                 .output_entry = ICMP_MIB_OUTTIMEEXCDS,
1063                 .input_entry = ICMP_MIB_INTIMEEXCDS,
1064                 .handler = icmp_unreach,
1065                 .error = 1,
1066         },
1067         [ICMP_PARAMETERPROB] = {
1068                 .output_entry = ICMP_MIB_OUTPARMPROBS,
1069                 .input_entry = ICMP_MIB_INPARMPROBS,
1070                 .handler = icmp_unreach,
1071                 .error = 1,
1072         },
1073         [ICMP_TIMESTAMP] = {
1074                 .output_entry = ICMP_MIB_OUTTIMESTAMPS,
1075                 .input_entry = ICMP_MIB_INTIMESTAMPS,
1076                 .handler = icmp_timestamp,
1077         },
1078         [ICMP_TIMESTAMPREPLY] = {
1079                 .output_entry = ICMP_MIB_OUTTIMESTAMPREPS,
1080                 .input_entry = ICMP_MIB_INTIMESTAMPREPS,
1081                 .handler = icmp_discard,
1082         },
1083         [ICMP_INFO_REQUEST] = {
1084                 .output_entry = ICMP_MIB_DUMMY,
1085                 .input_entry = ICMP_MIB_DUMMY,
1086                 .handler = icmp_discard,
1087         },
1088         [ICMP_INFO_REPLY] = {
1089                 .output_entry = ICMP_MIB_DUMMY,
1090                 .input_entry = ICMP_MIB_DUMMY,
1091                 .handler = icmp_discard,
1092         },
1093         [ICMP_ADDRESS] = {
1094                 .output_entry = ICMP_MIB_OUTADDRMASKS,
1095                 .input_entry = ICMP_MIB_INADDRMASKS,
1096                 .handler = icmp_address,
1097         },
1098         [ICMP_ADDRESSREPLY] = {
1099                 .output_entry = ICMP_MIB_OUTADDRMASKREPS,
1100                 .input_entry = ICMP_MIB_INADDRMASKREPS,
1101                 .handler = icmp_address_reply,
1102         },
1103 };
1104
1105 void __init icmp_init(struct net_proto_family *ops)
1106 {
1107         struct inet_sock *inet;
1108         int i;
1109
1110         for_each_cpu(i) {
1111                 int err;
1112
1113                 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1114                                        &per_cpu(__icmp_socket, i));
1115
1116                 if (err < 0)
1117                         panic("Failed to create the ICMP control socket.\n");
1118
1119                 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1120
1121                 /* Enough space for 2 64K ICMP packets, including
1122                  * sk_buff struct overhead.
1123                  */
1124                 per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1125                         (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1126
1127                 inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1128                 inet->uc_ttl = -1;
1129                 inet->pmtudisc = IP_PMTUDISC_DONT;
1130
1131                 /* Unhash it so that IP input processing does not even
1132                  * see it, we do not wish this socket to see incoming
1133                  * packets.
1134                  */
1135                 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1136         }
1137 }
1138
1139 EXPORT_SYMBOL(icmp_err_convert);
1140 EXPORT_SYMBOL(icmp_send);
1141 EXPORT_SYMBOL(icmp_statistics);
1142 EXPORT_SYMBOL(xrlim_allow);