![[first]](../icons/first.png){kind=icon}
![[previous]](../icons/n_left.png){kind=icon}
![[next]](../icons/n_right.png){kind=icon}
![[last]](../icons/last.png){kind=icon}
![[top]](../icons/n_top.png){kind=icon}
![[bottom]](../icons/bottom.png){kind=icon}
![[index]](../icons/index.png){kind=icon}
![[help]](../icons/help.png){kind=icon}
*/
1 /*
2 * NET3: Implementation of the ICMP protocol layer.
3 *
4 * Alan Cox, <alan@cymru.net>
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 * Some of the function names and the icmp unreach table for this
12 * module were derived from [icmp.c 1.0.11 06/02/93] by
13 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
14 * Other than that this module is a complete rewrite.
15 *
16 * Fixes:
17 * Mike Shaver : RFC1122 checks.
18 * Alan Cox : Multicast ping reply as self.
19 * Alan Cox : Fix atomicity lockup in ip_build_xmit call
20 * Alan Cox : Added 216,128 byte paths to the MTU code.
21 * Martin Mares : RFC1812 checks.
22 * Martin Mares : Can be configured to follow redirects if acting
23 * as a router _without_ a routing protocol (RFC 1812).
24 * Martin Mares : Echo requests may be configured to be ignored (RFC 1812).
25 * Martin Mares : Limitation of ICMP error message transmit rate (RFC 1812).
26 * Martin Mares : TOS and Precedence set correctly (RFC 1812).
27 *
28 *
29 *
30 * RFC1122 (Host Requirements -- Comm. Layer) Status:
31 * (boy, are there a lot of rules for ICMP)
32 * 3.2.2 (Generic ICMP stuff)
33 * MUST discard messages of unknown type. (OK)
34 * MUST copy at least the first 8 bytes from the offending packet
35 * when sending ICMP errors. (OBSOLETE -- see RFC1812)
36 * MUST pass received ICMP errors up to protocol level. (OK)
37 * SHOULD send ICMP errors with TOS == 0. (OBSOLETE -- see RFC1812)
38 * MUST NOT send ICMP errors in reply to:
39 * ICMP errors (OK)
40 * Broadcast/multicast datagrams (OK)
41 * MAC broadcasts (OK)
42 * Non-initial fragments (OK)
43 * Datagram with a source address that isn't a single host. (OK)
44 * 3.2.2.1 (Destination Unreachable)
45 * All the rules govern the IP layer, and are dealt with in ip.c, not here.
46 * 3.2.2.2 (Redirect)
47 * Host SHOULD NOT send ICMP_REDIRECTs. (OK)
48 * MUST update routing table in response to host or network redirects.
49 * (host OK, network OBSOLETE)
50 * SHOULD drop redirects if they're not from directly connected gateway
51 * (OK -- we drop it if it's not from our old gateway, which is close
52 * enough)
53 * 3.2.2.3 (Source Quench)
54 * MUST pass incoming SOURCE_QUENCHs to transport layer (OK)
55 * Other requirements are dealt with at the transport layer.
56 * 3.2.2.4 (Time Exceeded)
57 * MUST pass TIME_EXCEEDED to transport layer (OK)
58 * Other requirements dealt with at IP (generating TIME_EXCEEDED).
59 * 3.2.2.5 (Parameter Problem)
60 * SHOULD generate these (OK)
61 * MUST pass received PARAMPROBLEM to transport layer (NOT YET)
62 * [Solaris 2.X seems to assert EPROTO when this occurs] -- AC
63 * 3.2.2.6 (Echo Request/Reply)
64 * MUST reply to ECHO_REQUEST, and give app to do ECHO stuff (OK, OK)
65 * MAY discard broadcast ECHO_REQUESTs. (We don't, but that's OK.)
66 * MUST reply using same source address as the request was sent to.
67 * We're OK for unicast ECHOs, and it doesn't say anything about
68 * how to handle broadcast ones, since it's optional.
69 * MUST copy data from REQUEST to REPLY (OK)
70 * unless it would require illegal fragmentation (OK)
71 * MUST pass REPLYs to transport/user layer (OK)
72 * MUST use any provided source route (reversed) for REPLY. (NOT YET)
73 * 3.2.2.7 (Information Request/Reply)
74 * MUST NOT implement this. (I guess that means silently discard...?) (OK)
75 * 3.2.2.8 (Timestamp Request/Reply)
76 * MAY implement (OK)
77 * SHOULD be in-kernel for "minimum variability" (OK)
78 * MAY discard broadcast REQUESTs. (OK, but see source for inconsistency)
79 * MUST reply using same source address as the request was sent to. (OK)
80 * MUST reverse source route, as per ECHO (NOT YET)
81 * MUST pass REPLYs to transport/user layer (requires RAW, just like ECHO) (OK)
82 * MUST update clock for timestamp at least 15 times/sec (OK)
83 * MUST be "correct within a few minutes" (OK)
84 * 3.2.2.9 (Address Mask Request/Reply)
85 * MAY implement (OK)
86 * MUST send a broadcast REQUEST if using this system to set netmask
87 * (OK... we don't use it)
88 * MUST discard received REPLYs if not using this system (OK)
89 * MUST NOT send replies unless specifically made agent for this sort
90 * of thing. (OK)
91 *
92 *
93 * RFC 1812 (IPv4 Router Requirements) Status (even longer):
94 * 4.3.2.1 (Unknown Message Types)
95 * MUST pass messages of unknown type to ICMP user iface or silently discard
96 * them (OK)
97 * 4.3.2.2 (ICMP Message TTL)
98 * MUST initialize TTL when originating an ICMP message (OK)
99 * 4.3.2.3 (Original Message Header)
100 * SHOULD copy as much data from the offending packet as possible without
101 * the length of the ICMP datagram exceeding 576 bytes (NOT YET)
102 * MUST leave original IP header of the offending packet, but we're not
103 * required to undo modifications made (OK)
104 * 4.3.2.4 (Original Message Source Address)
105 * MUST use one of addresses for the interface the orig. packet arrived as
106 * source address (OK)
107 * 4.3.2.5 (TOS and Precedence)
108 * SHOULD leave TOS set to the same value unless the packet would be discarded
109 * for that reason (OK)
110 * MUST use TOS=0 if not possible to leave original value (OK)
111 * MUST leave IP Precedence for Source Quench messages (OK -- not sent at all)
112 * SHOULD use IP Precedence = 6 (Internetwork Control) or 7 (Network Control)
113 * for all other error messages (OK, we use 6)
114 * MAY allow configuration of IP Precedence (OK -- not done)
115 * MUST leave IP Precedence and TOS for reply messages (OK)
116 * 4.3.2.6 (Source Route)
117 * SHOULD use reverse source route UNLESS sending Parameter Problem on source
118 * routing and UNLESS the packet would be immediately discarded (NOT YET)
119 * 4.3.2.7 (When Not to Send ICMP Errors)
120 * MUST NOT send ICMP errors in reply to:
121 * ICMP errors (OK)
122 * Packets failing IP header validation tests unless otherwise noted (OK)
123 * Broadcast/multicast datagrams (OK)
124 * MAC broadcasts (OK)
125 * Non-initial fragments (OK)
126 * Datagram with a source address that isn't a single host. (OK)
127 * 4.3.2.8 (Rate Limiting)
128 * SHOULD be able to limit error message rate (OK)
129 * SHOULD allow setting of rate limits (OK, in the source)
130 * 4.3.3.1 (Destination Unreachable)
131 * All the rules govern the IP layer, and are dealt with in ip.c, not here.
132 * 4.3.3.2 (Redirect)
133 * MAY ignore ICMP Redirects if running a routing protocol or if forwarding
134 * is enabled on the interface (OK -- ignores)
135 * 4.3.3.3 (Source Quench)
136 * SHOULD NOT originate SQ messages (OK)
137 * MUST be able to limit SQ rate if originates them (OK as we don't send them)
138 * MAY ignore SQ messages it receives (OK -- we don't)
139 * 4.3.3.4 (Time Exceeded)
140 * Requirements dealt with at IP (generating TIME_EXCEEDED).
141 * 4.3.3.5 (Parameter Problem)
142 * MUST generate these for all errors not covered by other messages (OK)
143 * MUST include original value of the value pointed by (OK)
144 * 4.3.3.6 (Echo Request)
145 * MUST implement echo server function (OK)
146 * MUST process at ER of at least max(576, MTU) (OK)
147 * MAY reject broadcast/multicast ER's (We don't, but that's OK)
148 * SHOULD have a config option for silently ignoring ER's (OK)
149 * MUST have a default value for the above switch = NO (OK)
150 * MUST have application layer interface for Echo Request/Reply (OK)
151 * MUST reply using same source address as the request was sent to.
152 * We're OK for unicast ECHOs, and it doesn't say anything about
153 * how to handle broadcast ones, since it's optional.
154 * MUST copy data from Request to Reply (OK)
155 * SHOULD update Record Route / Timestamp options (??)
156 * MUST use reversed Source Route for Reply if possible (NOT YET)
157 * 4.3.3.7 (Information Request/Reply)
158 * SHOULD NOT originate or respond to these (OK)
159 * 4.3.3.8 (Timestamp / Timestamp Reply)
160 * MAY implement (OK)
161 * MUST reply to every Timestamp message received (OK)
162 * MAY discard broadcast REQUESTs. (OK, but see source for inconsistency)
163 * MUST reply using same source address as the request was sent to. (OK)
164 * MUST use reversed Source Route if possible (NOT YET)
165 * SHOULD update Record Route / Timestamp options (??)
166 * MUST pass REPLYs to transport/user layer (requires RAW, just like ECHO) (OK)
167 * MUST update clock for timestamp at least 16 times/sec (OK)
168 * MUST be "correct within a few minutes" (OK)
169 * 4.3.3.9 (Address Mask Request/Reply)
170 * MUST have support for receiving AMRq and responding with AMRe (OK, but only as a
171 * compile-time option)
172 * SHOULD have option for each interface for AMRe's, MUST default to NO (NOT YET)
173 * MUST NOT reply to AMRq before knows the correct AM (OK)
174 * MUST NOT respond to AMRq with source address 0.0.0.0 and the AM's for
175 * logical i-faces for the physical i-face are not the same (NOT YET)
176 * SHOULD examine all AMRe's it receives and check them (NOT YET)
177 * SHOULD log invalid AMRe's (AM+sender) (NOT YET)
178 * MUST NOT use contents of AMRe to determine correct AM (OK)
179 * MAY broadcast AMRe's after having configured address masks (OK -- doesn't)
180 * MUST NOT do broadcast AMRe's if not set by extra option (OK, no option)
181 * MUST use the { <NetPrefix>, -1 } form of broadcast addresses (OK)
182 * 4.3.3.10 (Router Advertisement and Solicitations)
183 * MUST support router part of Router Discovery Protocol on all networks we
184 * support broadcast or multicast addressing. (OK -- done by gated)
185 * MUST have all config parameters with the respective defaults (OK)
186 * 5.2.7.1 (Destination Unreachable)
187 * MUST generate DU's (OK)
188 * SHOULD choose a best-match response code (OK)
189 * SHOULD NOT generate Host Isolated codes (OK)
190 * SHOULD use Communication Administratively Prohibited when administratively
191 * filtering packets (NOT YET)
192 * MAY include config option for not generating the above and silently discard
193 * the packets instead (OK)
194 * MAY include config option for not generating Precedence Violation and
195 * Precedence Cutoff messages (OK as we don't generate them at all)
196 * MUST use Host Unreachable or Dest. Host Unknown codes whenever other hosts
197 * on the same network might be reachable (OK -- no net unreach's at all)
198 * MUST use new form of Fragmentation Needed and DF Set messages (OK)
199 * 5.2.7.2 (Redirect)
200 * MUST NOT generate network redirects (OK)
201 * MUST be able to generate host redirects (OK)
202 * SHOULD be able to generate Host+TOS redirects (NO as we don't use TOS)
203 * MUST have an option to use Host redirects instead of Host+TOS ones (OK as
204 * no Host+TOS Redirects are used)
205 * MUST NOT generate redirects unless forwarding to the same i-face and the
206 * dest. address is on the same subnet as the src. address and no source
207 * routing is in use. (OK)
208 * MUST NOT follow redirects when using a routing protocol (OK)
209 * MAY use redirects if not using a routing protocol (OK, compile-time option)
210 * MUST comply to Host Requirements when not acting as a router (OK)
211 * 5.2.7.3 (Time Exceeded)
212 * MUST generate Time Exceeded Code 0 when discarding packet due to TTL=0 (OK)
213 * MAY have a per-interface option to disable origination of TE messages, but
214 * it MUST default to "originate" (OK -- we don't support it)
215 */
216
217 #include <linux/config.h>
218 #include <linux/types.h>
219 #include <linux/sched.h>
220 #include <linux/kernel.h>
221 #include <linux/fcntl.h>
222 #include <linux/socket.h>
223 #include <linux/in.h>
224 #include <linux/inet.h>
225 #include <linux/netdevice.h>
226 #include <linux/string.h>
227 #include <net/snmp.h>
228 #include <net/ip.h>
229 #include <net/route.h>
230 #include <net/protocol.h>
231 #include <net/icmp.h>
232 #include <net/tcp.h>
233 #include <net/snmp.h>
234 #include <linux/skbuff.h>
235 #include <net/sock.h>
236 #include <linux/errno.h>
237 #include <linux/timer.h>
238 #include <asm/system.h>
239 #include <asm/segment.h>
240 #include <net/checksum.h>
241
242 #define min(a,b) ((a)<(b)?(a):(b))
243
244 /*
245 * Statistics
246 */
247
248 struct icmp_mib icmp_statistics;
249
250 /* An array of errno for error messages from dest unreach. */
251 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOS_UNREACH and SR_FAIELD MUST be considered 'transient errs'. */
252
253 struct icmp_err icmp_err_convert[] = {
254 { ENETUNREACH, 0 }, /* ICMP_NET_UNREACH */
255 { EHOSTUNREACH, 0 }, /* ICMP_HOST_UNREACH */
256 { ENOPROTOOPT, 1 }, /* ICMP_PROT_UNREACH */
257 { ECONNREFUSED, 1 }, /* ICMP_PORT_UNREACH */
258 { EOPNOTSUPP, 0 }, /* ICMP_FRAG_NEEDED */
259 { EOPNOTSUPP, 0 }, /* ICMP_SR_FAILED */
260 { ENETUNREACH, 1 }, /* ICMP_NET_UNKNOWN */
261 { EHOSTDOWN, 1 }, /* ICMP_HOST_UNKNOWN */
262 { ENONET, 1 }, /* ICMP_HOST_ISOLATED */
263 { ENETUNREACH, 1 }, /* ICMP_NET_ANO */
264 { EHOSTUNREACH, 1 }, /* ICMP_HOST_ANO */
265 { EOPNOTSUPP, 0 }, /* ICMP_NET_UNR_TOS */
266 { EOPNOTSUPP, 0 } /* ICMP_HOST_UNR_TOS */
267 };
268
269 /*
270 * A spare long used to speed up statistics updating
271 */
272
273 unsigned long dummy;
274
275 /*
276 * ICMP transmit rate limit control structures. We use a relatively simple
277 * approach to the problem: For each type of ICMP message with rate limit
278 * we count the number of messages sent during some time quantum. If this
279 * count exceeds given maximal value, we ignore all messages not separated
280 * from the last message sent at least by specified time.
281 */
282
283 #define XRLIM_CACHE_SIZE 16 /* How many destination hosts do we cache */
284
285 struct icmp_xrl_cache /* One entry of the ICMP rate cache */
286 {
287 __u32 daddr; /* Destination address */
288 unsigned long counter; /* Message counter */
289 unsigned long next_reset; /* Time of next reset of the counter */
290 unsigned long last_access; /* Time of last access to this entry (LRU) */
291 unsigned int restricted; /* Set if we're in restricted mode */
292 unsigned long next_packet; /* When we'll allow a next packet if restricted */
293 };
294
295 struct icmp_xrlim
296 {
297 unsigned long timeout; /* Time quantum for rate measuring */
298 unsigned long limit; /* Maximal number of messages per time quantum allowed */
299 unsigned long delay; /* How long we wait between packets when restricting */
300 struct icmp_xrl_cache cache[XRLIM_CACHE_SIZE]; /* Rate cache */
301 };
302
303 /*
304 * ICMP control array. This specifies what to do with each ICMP.
305 */
306
307 struct icmp_control
308 {
309 unsigned long *output; /* Address to increment on output */
310 unsigned long *input; /* Address to increment on input */
311 void (*handler)(struct icmphdr *icmph, struct sk_buff *skb, struct device *dev, __u32 saddr, __u32 daddr, int len);
312 unsigned long error; /* This ICMP is classed as an error message */
313 struct icmp_xrlim *xrlim; /* Transmit rate limit control structure or NULL for no limits */
314 };
315
316 static struct icmp_control icmp_pointers[19];
317
318 /*
319 * Build xmit assembly blocks
320 */
321
322 struct icmp_bxm
323 {
324 void *data_ptr;
325 int data_len;
326 struct icmphdr icmph;
327 unsigned long csum;
328 struct options replyopts;
329 unsigned char optbuf[40];
330 };
331
332 /*
333 * The ICMP socket. This is the most convenient way to flow control
334 * our ICMP output as well as maintain a clean interface throughout
335 * all layers. All Socketless IP sends will soon be gone.
336 */
337
338 struct socket icmp_socket;
339
340 /*
341 * Send an ICMP frame.
342 */
343
344
345 /*
346 * Initialize the transmit rate limitation mechanism.
347 */
348
349 #ifndef CONFIG_NO_ICMP_LIMIT
350
351 static void xrlim_init(void)
/* ![[next]](../icons/right.png)
![[first]](../icons/n_first.png)
![[top]](../icons/top.png)
*/
352 {
353 int type, entry;
354 struct icmp_xrlim *xr;
355
356 for (type=0; type<=18; type++) {
357 xr = icmp_pointers[type].xrlim;
358 if (xr) {
359 for (entry=0; entry<XRLIM_CACHE_SIZE; entry++)
360 xr->cache[entry].daddr = INADDR_NONE;
361 }
362 }
363 }
364
365 /*
366 * Check transmit rate limitation for given message.
367 *
368 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
369 * SHOULD allow setting of rate limits (we allow in the source)
370 */
371
372 static int xrlim_allow(int type, __u32 addr)
/* ![[previous]](../icons/left.png)
*/
373 {
374 struct icmp_xrlim *r;
375 struct icmp_xrl_cache *c;
376 unsigned long now;
377
378 if (type > 18) /* No time limit present */
379 return 1;
380 r = icmp_pointers[type].xrlim;
381 if (!r)
382 return 1;
383
384 for (c = r->cache; c < &r->cache[XRLIM_CACHE_SIZE]; c++) /* Cache lookup */
385 if (c->daddr == addr)
386 break;
387
388 now = jiffies; /* Cache current time (saves accesses to volatile variable) */
389
390 if (c == &r->cache[XRLIM_CACHE_SIZE]) { /* Cache miss */
391 unsigned long oldest = now; /* Find the oldest entry to replace */
392 struct icmp_xrl_cache *d;
393 c = r->cache;
394 for (d = r->cache; d < &r->cache[XRLIM_CACHE_SIZE]; d++)
395 if (!d->daddr) { /* Unused entry */
396 c = d;
397 break;
398 } else if (d->last_access < oldest) {
399 oldest = d->last_access;
400 c = d;
401 }
402 c->last_access = now; /* Fill the entry with new data */
403 c->daddr = addr;
404 c->counter = 1;
405 c->next_reset = now + r->timeout;
406 c->restricted = 0;
407 return 1;
408 }
409
410 c->last_access = now;
411 if (c->next_reset > now) { /* Let's increment the counter */
412 c->counter++;
413 if (c->counter == r->limit) { /* Limit exceeded, start restrictions */
414 c->restricted = 1;
415 c->next_packet = now + r->delay;
416 return 0;
417 }
418 if (c->restricted) { /* Any restrictions pending? */
419 if (c->next_packet > now)
420 return 0;
421 c->next_packet = now + r->delay;
422 return 1;
423 }
424 } else { /* Reset the counter */
425 if (c->counter < r->limit) /* Switch off all restrictions */
426 c->restricted = 0;
427 c->next_reset = now + r->timeout;
428 c->counter = 0;
429 }
430
431 return 1; /* Send the packet */
432 }
433
434 #endif /* CONFIG_NO_ICMP_LIMIT */
435
436 /*
437 * Maintain the counters used in the SNMP statistics for outgoing ICMP
438 */
439
440 static void icmp_out_count(int type)
/* */
441 {
442 if(type>18)
443 return;
444 (*icmp_pointers[type].output)++;
445 icmp_statistics.IcmpOutMsgs++;
446 }
447
448 /*
449 * Checksum each fragment, and on the first include the headers and final checksum.
450 */
451
452 static void icmp_glue_bits(const void *p, __u32 saddr, char *to, unsigned int offset, unsigned int fraglen)
/* */
453 {
454 struct icmp_bxm *icmp_param = (struct icmp_bxm *)p;
455 struct icmphdr *icmph;
456 unsigned long csum;
457
458 if (offset) {
459 icmp_param->csum=csum_partial_copy(icmp_param->data_ptr+offset-sizeof(struct icmphdr),
460 to, fraglen,icmp_param->csum);
461 return;
462 }
463
464 /*
465 * First fragment includes header. Note that we've done
466 * the other fragments first, so that we get the checksum
467 * for the whole packet here.
468 */
469 csum = csum_partial_copy((void *)&icmp_param->icmph,
470 to, sizeof(struct icmphdr),
471 icmp_param->csum);
472 csum = csum_partial_copy(icmp_param->data_ptr,
473 to+sizeof(struct icmphdr),
474 fraglen-sizeof(struct icmphdr), csum);
475 icmph=(struct icmphdr *)to;
476 icmph->checksum = csum_fold(csum);
477 }
478
479 /*
480 * Driving logic for building and sending ICMP messages.
481 */
482
483 static void icmp_build_xmit(struct icmp_bxm *icmp_param, __u32 saddr, __u32 daddr, __u8 tos)
/* */
484 {
485 struct sock *sk=icmp_socket.data;
486 icmp_param->icmph.checksum=0;
487 icmp_param->csum=0;
488 icmp_out_count(icmp_param->icmph.type);
489 sk->ip_tos = tos;
490 ip_build_xmit(sk, icmp_glue_bits, icmp_param,
491 icmp_param->data_len+sizeof(struct icmphdr),
492 daddr, saddr, &icmp_param->replyopts, 0, IPPROTO_ICMP, 1);
493 }
494
495
496 /*
497 * Send an ICMP message in response to a situation
498 *
499 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. MAY send more (we don't).
500 * MUST NOT change this header information.
501 * MUST NOT reply to a multicast/broadcast IP address.
502 * MUST NOT reply to a multicast/broadcast MAC address.
503 * MUST reply to only the first fragment.
504 */
505
506 void icmp_send(struct sk_buff *skb_in, int type, int code, unsigned long info, struct device *dev)
/* */
507 {
508 struct iphdr *iph;
509 struct icmphdr *icmph;
510 int atype;
511 struct icmp_bxm icmp_param;
512 __u32 saddr;
513
514 /*
515 * Find the original header
516 */
517
518 iph = skb_in->ip_hdr;
519
520 /*
521 * No replies to physical multicast/broadcast
522 */
523
524 if(skb_in->pkt_type!=PACKET_HOST)
525 return;
526
527 /*
528 * Now check at the protocol level
529 */
530
531 atype=ip_chk_addr(iph->daddr);
532 if(atype==IS_BROADCAST||atype==IS_MULTICAST)
533 return;
534
535 /*
536 * Only reply to fragment 0. We byte re-order the constant
537 * mask for efficiency.
538 */
539
540 if(iph->frag_off&htons(IP_OFFSET))
541 return;
542
543 /*
544 * If we send an ICMP error to an ICMP error a mess would result..
545 */
546
547 if(icmp_pointers[type].error)
548 {
549 /*
550 * We are an error, check if we are replying to an ICMP error
551 */
552
553 if(iph->protocol==IPPROTO_ICMP)
554 {
555 icmph = (struct icmphdr *)((char *)iph + (iph->ihl<<2));
556 /*
557 * Assume any unknown ICMP type is an error. This isn't
558 * specified by the RFC, but think about it..
559 */
560 if(icmph->type>18 || icmp_pointers[icmph->type].error)
561 return;
562 }
563 }
564
565 /*
566 * Check the rate limit
567 */
568
569 #ifndef CONFIG_NO_ICMP_LIMIT
570 if (!xrlim_allow(type, iph->saddr))
571 return;
572 #endif
573
574 /*
575 * Tell our driver what to send
576 */
577
578 saddr=iph->daddr;
579 if(saddr!=dev->pa_addr && ip_chk_addr(saddr)!=IS_MYADDR)
580 saddr=dev->pa_addr;
581
582 icmp_param.icmph.type=type;
583 icmp_param.icmph.code=code;
584 icmp_param.icmph.un.gateway = info;
585 icmp_param.data_ptr=iph;
586 icmp_param.data_len=(iph->ihl<<2)+8; /* RFC says return header + 8 bytes */
587
588 /*
589 * Set it to build.
590 */
591
592 if (ip_options_echo(&icmp_param.replyopts, NULL, saddr, iph->saddr, skb_in) == 0)
593 icmp_build_xmit(&icmp_param, saddr, iph->saddr, ((iph->tos & 0x38) | 6));
594 }
595
596
597 /*
598 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
599 */
600
601 static void icmp_unreach(struct icmphdr *icmph, struct sk_buff *skb, struct device *dev, __u32 saddr, __u32 daddr, int len)
/* */
602 {
603 struct iphdr *iph;
604 int hash;
605 struct inet_protocol *ipprot;
606 unsigned char *dp;
607
608 iph = (struct iphdr *) (icmph + 1);
609
610 dp= ((unsigned char *)iph)+(iph->ihl<<2);
611
612 if(icmph->type==ICMP_DEST_UNREACH)
613 {
614 switch(icmph->code & 15)
615 {
616 case ICMP_NET_UNREACH:
617 break;
618 case ICMP_HOST_UNREACH:
619 break;
620 case ICMP_PROT_UNREACH:
621 /* printk(KERN_INFO "ICMP: %s:%d: protocol unreachable.\n",
622 in_ntoa(iph->daddr), (int)iph->protocol);*/
623 break;
624 case ICMP_PORT_UNREACH:
625 break;
626 case ICMP_FRAG_NEEDED:
627 #ifdef CONFIG_NO_PATH_MTU_DISCOVERY
628 printk(KERN_INFO "ICMP: %s: fragmentation needed and DF set.\n",
629 in_ntoa(iph->daddr));
630 break;
631 #else
632 {
633 unsigned short old_mtu = ntohs(iph->tot_len);
634 unsigned short new_mtu = ntohs(icmph->un.echo.sequence);
635
636 if (new_mtu < 68 || new_mtu >= old_mtu)
637 {
638 /*
639 * It is either dumb router, which does not
640 * understand Path MTU Disc. protocol
641 * or broken (f.e. Linux<=1.3.37 8) router.
642 * Try to guess...
643 * The table is taken from RFC-1191.
644 */
645 if (old_mtu > 32000)
646 new_mtu = 32000;
647 else if (old_mtu > 17914)
648 new_mtu = 17914;
649 else if (old_mtu > 8166)
650 new_mtu = 8166;
651 else if (old_mtu > 4352)
652 new_mtu = 4352;
653 else if (old_mtu > 2002)
654 new_mtu = 2002;
655 else if (old_mtu > 1492)
656 new_mtu = 1492;
657 else if (old_mtu > 576)
658 new_mtu = 576;
659 else if (old_mtu > 296)
660 new_mtu = 296;
661 /*
662 * These two are not from the RFC but
663 * are needed for AMPRnet AX.25 paths.
664 */
665 else if (old_mtu > 216)
666 new_mtu = 216;
667 else if (old_mtu > 128)
668 new_mtu = 128;
669 else
670 /*
671 * Despair..
672 */
673 new_mtu = 68;
674 }
675 /*
676 * Ugly trick to pass MTU to protocol layer.
677 * Really we should add argument "info" to error handler.
678 */
679 iph->id = htons(new_mtu);
680 break;
681 }
682 #endif
683 case ICMP_SR_FAILED:
684 printk(KERN_INFO "ICMP: %s: Source Route Failed.\n", in_ntoa(iph->daddr));
685 break;
686 default:
687 break;
688 }
689 if(icmph->code>12) /* Invalid type */
690 return;
691 }
692
693 /*
694 * Throw it at our lower layers
695 *
696 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed header.
697 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the transport layer.
698 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to transport layer.
699 */
700
701 /*
702 * Get the protocol(s).
703 */
704
705 hash = iph->protocol & (MAX_INET_PROTOS -1);
706
707 /*
708 * This can't change while we are doing it.
709 *
710 * FIXME: Deliver to appropriate raw sockets too.
711 */
712
713 ipprot = (struct inet_protocol *) inet_protos[hash];
714 while(ipprot != NULL)
715 {
716 struct inet_protocol *nextip;
717
718 nextip = (struct inet_protocol *) ipprot->next;
719
720 /*
721 * Pass it off to everyone who wants it.
722 */
723
724 /* RFC1122: OK. Passes appropriate ICMP errors to the */
725 /* appropriate protocol layer (MUST), as per 3.2.2. */
726
727 if (iph->protocol == ipprot->protocol && ipprot->err_handler)
728 {
729 ipprot->err_handler(icmph->type, icmph->code, dp,
730 iph->daddr, iph->saddr, ipprot);
731 }
732
733 ipprot = nextip;
734 }
735 kfree_skb(skb, FREE_READ);
736 }
737
738
739 /*
740 * Handle ICMP_REDIRECT.
741 */
742
743 static void icmp_redirect(struct icmphdr *icmph, struct sk_buff *skb, struct device *dev, __u32 source, __u32 daddr, int len)
/* */
744 {
745 struct iphdr *iph;
746 unsigned long ip;
747
748 /*
749 * Get the copied header of the packet that caused the redirect
750 */
751
752 iph = (struct iphdr *) (icmph + 1);
753 ip = iph->daddr;
754
755 /*
756 * If we are a router and we run a routing protocol, we MUST NOT follow redirects.
757 * When using no routing protocol, we MAY follow redirects. (RFC 1812, 5.2.7.2)
758 */
759
760 #if defined(CONFIG_IP_FORWARD) && !defined(CONFIG_IP_DUMB_ROUTER)
761 printk(KERN_INFO "icmp: ICMP redirect ignored. dest = %s, "
762 "orig gw = %s, \"new\" gw = %s, device = %s.\n", in_ntoa(ip),
763 in_ntoa(source), in_ntoa(icmph->un.gateway), dev->name);
764 #else
765 switch(icmph->code & 7)
766 {
767 case ICMP_REDIR_NET:
768 /*
769 * This causes a problem with subnetted networks. What we should do
770 * is use ICMP_ADDRESS to get the subnet mask of the problem route
771 * and set both. But we don't.. [RFC1812 says routers MUST NOT
772 * generate Network Redirects]
773 */
774 #ifdef not_a_good_idea
775 ip_rt_add((RTF_DYNAMIC | RTF_MODIFIED | RTF_GATEWAY),
776 ip, 0, icmph->un.gateway, dev,0, 0, 0);
777 #endif
778 /*
779 * As per RFC recommendations now handle it as
780 * a host redirect.
781 */
782
783 case ICMP_REDIR_HOST:
784 /*
785 * Add better route to host.
786 * But first check that the redirect
787 * comes from the old gateway..
788 * And make sure it's an ok host address
789 * (not some confused thing sending our
790 * address)
791 */
792 printk(KERN_INFO "ICMP redirect from %s\n", in_ntoa(source));
793 ip_rt_redirect(source, ip, icmph->un.gateway, dev);
794 break;
795 case ICMP_REDIR_NETTOS:
796 case ICMP_REDIR_HOSTTOS:
797 printk(KERN_INFO "ICMP: cannot handle TOS redirects yet!\n");
798 break;
799 default:
800 break;
801 }
802 #endif
803 /*
804 * Discard the original packet
805 */
806
807 kfree_skb(skb, FREE_READ);
808 }
809
810 /*
811 * Handle ICMP_ECHO ("ping") requests.
812 *
813 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo requests.
814 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be included in the reply.
815 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring echo requests, MUST have default=NOT.
816 * See also WRT handling of options once they are done and working.
817 */
818
819 static void icmp_echo(struct icmphdr *icmph, struct sk_buff *skb, struct device *dev, __u32 saddr, __u32 daddr, int len)
/* */
820 {
821 #ifndef CONFIG_IP_IGNORE_ECHO_REQUESTS
822 struct icmp_bxm icmp_param;
823 icmp_param.icmph=*icmph;
824 icmp_param.icmph.type=ICMP_ECHOREPLY;
825 icmp_param.data_ptr=(icmph+1);
826 icmp_param.data_len=len;
827 if (ip_options_echo(&icmp_param.replyopts, NULL, daddr, saddr, skb)==0)
828 icmp_build_xmit(&icmp_param, daddr, saddr, skb->ip_hdr->tos);
829 #endif
830 kfree_skb(skb, FREE_READ);
831 }
832
833 /*
834 * Handle ICMP Timestamp requests.
835 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
836 * SHOULD be in the kernel for minimum random latency.
837 * MUST be accurate to a few minutes.
838 * MUST be updated at least at 15Hz.
839 */
840
841 static void icmp_timestamp(struct icmphdr *icmph, struct sk_buff *skb, struct device *dev, __u32 saddr, __u32 daddr, int len)
/* */
842 {
843 __u32 times[3]; /* So the new timestamp works on ALPHA's.. */
844 struct icmp_bxm icmp_param;
845
846 /*
847 * Too short.
848 */
849
850 if(len<12)
851 {
852 icmp_statistics.IcmpInErrors++;
853 kfree_skb(skb, FREE_READ);
854 return;
855 }
856
857 /*
858 * Fill in the current time as ms since midnight UT:
859 */
860
861 {
862 struct timeval tv;
863 do_gettimeofday(&tv);
864 times[1] = htonl((tv.tv_sec % 86400) * 1000 + tv.tv_usec / 1000);
865 }
866 times[2] = times[1];
867 memcpy((void *)×[0], icmph+1, 4); /* Incoming stamp */
868 icmp_param.icmph=*icmph;
869 icmp_param.icmph.type=ICMP_TIMESTAMPREPLY;
870 icmp_param.icmph.code=0;
871 icmp_param.data_ptr=×
872 icmp_param.data_len=12;
873 if (ip_options_echo(&icmp_param.replyopts, NULL, daddr, saddr, skb)==0)
874 icmp_build_xmit(&icmp_param, daddr, saddr, skb->ip_hdr->tos);
875 kfree_skb(skb,FREE_READ);
876 }
877
878
879 /*
880 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
881 *
882 * RFC1122 (3.2.2.9). A host MUST only send replies to
883 * ADDRESS_MASK requests if it's been configured as an address mask
884 * agent. Receiving a request doesn't constitute implicit permission to
885 * act as one. Of course, implementing this correctly requires (SHOULD)
886 * a way to turn the functionality on and off. Another one for sysctl(),
887 * I guess. -- MS
888 * Botched with a CONFIG option for now - Linus add scts sysctl please..
889 */
890
891 static void icmp_address(struct icmphdr *icmph, struct sk_buff *skb, struct device *dev, __u32 saddr, __u32 daddr, int len)
/* */
892 {
893 #ifdef CONFIG_IP_ADDR_AGENT /* Don't use, broken */
894 struct icmp_bxm icmp_param;
895 icmp_param.icmph.type=ICMP_ADDRESSREPLY;
896 icmp_param.icmph.code=0;
897 icmp_param.icmph.un.echo.id = icmph->un.echo.id;
898 icmp_param.icmph.un.echo.sequence = icmph->un.echo.sequence;
899 icmp_param.data_ptr=&dev->pa_mask;
900 icmp_param.data_len=4;
901 if (ip_options_echo(&icmp_param.replyopts, NULL, daddr, saddr, skb)==0)
902 icmp_build_xmit(&icmp_param, daddr, saddr, skb->iph->tos);
903 #endif
904 kfree_skb(skb, FREE_READ);
905 }
906
907 static void icmp_discard(struct icmphdr *icmph, struct sk_buff *skb, struct device *dev, __u32 saddr, __u32 daddr, int len)
/* */
908 {
909 kfree_skb(skb, FREE_READ);
910 }
911
912 /*
913 * Deal with incoming ICMP packets.
914 */
915
916 int icmp_rcv(struct sk_buff *skb, struct device *dev, struct options *opt,
/* */
917 __u32 daddr, unsigned short len,
918 __u32 saddr, int redo, struct inet_protocol *protocol)
919 {
920 struct icmphdr *icmph=(void *)skb->h.raw;
921 icmp_statistics.IcmpInMsgs++;
922
923 /*
924 * Validate the packet
925 */
926
927 if (ip_compute_csum((unsigned char *) icmph, len))
928 {
929 /* Failed checksum! */
930 icmp_statistics.IcmpInErrors++;
931 printk(KERN_INFO "ICMP: failed checksum from %s!\n", in_ntoa(saddr));
932 kfree_skb(skb, FREE_READ);
933 return(0);
934 }
935
936 /*
937 * 18 is the highest 'known' ICMP type. Anything else is a mystery
938 *
939 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently discarded.
940 */
941
942 if(icmph->type > 18)
943 {
944 icmp_statistics.IcmpInErrors++; /* Is this right - or do we ignore ? */
945 kfree_skb(skb,FREE_READ);
946 return(0);
947 }
948
949 /*
950 * Parse the ICMP message
951 */
952
953 if (daddr!=dev->pa_addr && ip_chk_addr(daddr) != IS_MYADDR)
954 {
955 /*
956 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be silently ignored (we don't as it is used
957 * by some network mapping tools).
958 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently discarded if to broadcast/multicast.
959 */
960 if (icmph->type != ICMP_ECHO)
961 {
962 icmp_statistics.IcmpInErrors++;
963 kfree_skb(skb, FREE_READ);
964 return(0);
965 }
966 /*
967 * Reply the multicast/broadcast using a legal
968 * interface - in this case the device we got
969 * it from.
970 */
971 daddr=dev->pa_addr;
972 }
973
974 len-=sizeof(struct icmphdr);
975 (*icmp_pointers[icmph->type].input)++;
976 (icmp_pointers[icmph->type].handler)(icmph,skb,skb->dev,saddr,daddr,len);
977 return 0;
978 }
979
980 /*
981 * This table defined limits of ICMP sending rate for various ICMP messages.
982 */
983
984 static struct icmp_xrlim
985 xrl_unreach = { 4*HZ, 80, HZ/4 }, /* Host Unreachable */
986 xrl_redirect = { 2*HZ, 10, HZ/2 }, /* Redirect */
987 xrl_generic = { 3*HZ, 30, HZ/4 }; /* All other errors */
988
989 /*
990 * This table is the definition of how we handle ICMP.
991 */
992
993 static struct icmp_control icmp_pointers[19] = {
994 /* ECHO REPLY (0) */
995 { &icmp_statistics.IcmpOutEchoReps, &icmp_statistics.IcmpInEchoReps, icmp_discard, 0, NULL },
996 { &dummy, &icmp_statistics.IcmpInErrors, icmp_discard, 1, NULL },
997 { &dummy, &icmp_statistics.IcmpInErrors, icmp_discard, 1, NULL },
998 /* DEST UNREACH (3) */
999 { &icmp_statistics.IcmpOutDestUnreachs, &icmp_statistics.IcmpInDestUnreachs, icmp_unreach, 1, &xrl_unreach },
1000 /* SOURCE QUENCH (4) */
1001 { &icmp_statistics.IcmpOutSrcQuenchs, &icmp_statistics.IcmpInSrcQuenchs, icmp_unreach, 1, NULL },
1002 /* REDIRECT (5) */
1003 { &icmp_statistics.IcmpOutRedirects, &icmp_statistics.IcmpInRedirects, icmp_redirect, 1, &xrl_redirect },
1004 { &dummy, &icmp_statistics.IcmpInErrors, icmp_discard, 1, NULL },
1005 { &dummy, &icmp_statistics.IcmpInErrors, icmp_discard, 1, NULL },
1006 /* ECHO (8) */
1007 { &icmp_statistics.IcmpOutEchos, &icmp_statistics.IcmpInEchos, icmp_echo, 0, NULL },
1008 { &dummy, &icmp_statistics.IcmpInErrors, icmp_discard, 1, NULL },
1009 { &dummy, &icmp_statistics.IcmpInErrors, icmp_discard, 1, NULL },
1010 /* TIME EXCEEDED (11) */
1011 { &icmp_statistics.IcmpOutTimeExcds, &icmp_statistics.IcmpInTimeExcds, icmp_unreach, 1, &xrl_generic },
1012 /* PARAMETER PROBLEM (12) */
1013 /* FIXME: RFC1122 3.2.2.5 - MUST pass PARAM_PROB messages to transport layer */
1014 { &icmp_statistics.IcmpOutParmProbs, &icmp_statistics.IcmpInParmProbs, icmp_discard, 1, &xrl_generic },
1015 /* TIMESTAMP (13) */
1016 { &icmp_statistics.IcmpOutTimestamps, &icmp_statistics.IcmpInTimestamps, icmp_timestamp, 0, NULL },
1017 /* TIMESTAMP REPLY (14) */
1018 { &icmp_statistics.IcmpOutTimestampReps, &icmp_statistics.IcmpInTimestampReps, icmp_discard, 0, NULL },
1019 /* INFO (15) */
1020 { &dummy, &dummy, icmp_discard, 0, NULL },
1021 /* INFO REPLY (16) */
1022 { &dummy, &dummy, icmp_discard, 0, NULL },
1023 /* ADDR MASK (17) */
1024 { &icmp_statistics.IcmpOutAddrMasks, &icmp_statistics.IcmpInAddrMasks, icmp_address, 0, NULL },
1025 /* ADDR MASK REPLY (18) */
1026 { &icmp_statistics.IcmpOutAddrMaskReps, &icmp_statistics.IcmpInAddrMaskReps, icmp_discard, 0, NULL }
1027 };
1028
1029 void icmp_init(struct proto_ops *ops)
/* ![[last]](../icons/n_last.png)
*/
1030 {
1031 struct sock *sk;
1032 int err;
1033 icmp_socket.type=SOCK_RAW;
1034 icmp_socket.ops=ops;
1035 if((err=ops->create(&icmp_socket, IPPROTO_ICMP))<0)
1036 panic("Failed to create the ICMP control socket.\n");
1037 sk=icmp_socket.data;
1038 sk->allocation=GFP_ATOMIC;
1039 sk->num = 256; /* Don't receive any data */
1040 #ifndef CONFIG_NO_ICMP_LIMIT
1041 xrlim_init();
1042 #endif
1043 }
1044
![[bottom]](../icons/n_bottom.png){kind=icon}
*/