1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * NET3: Implementation of the ICMP protocol layer.
4 *
5 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 *
7 * Some of the function names and the icmp unreach table for this
8 * module were derived from [icmp.c 1.0.11 06/02/93] by
9 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
10 * Other than that this module is a complete rewrite.
11 *
12 * Fixes:
13 * Clemens Fruhwirth : introduce global icmp rate limiting
14 * with icmp type masking ability instead
15 * of broken per type icmp timeouts.
16 * Mike Shaver : RFC1122 checks.
17 * Alan Cox : Multicast ping reply as self.
18 * Alan Cox : Fix atomicity lockup in ip_build_xmit
19 * call.
20 * Alan Cox : Added 216,128 byte paths to the MTU
21 * code.
22 * Martin Mares : RFC1812 checks.
23 * Martin Mares : Can be configured to follow redirects
24 * if acting as a router _without_ a
25 * routing protocol (RFC 1812).
26 * Martin Mares : Echo requests may be configured to
27 * be ignored (RFC 1812).
28 * Martin Mares : Limitation of ICMP error message
29 * transmit rate (RFC 1812).
30 * Martin Mares : TOS and Precedence set correctly
31 * (RFC 1812).
32 * Martin Mares : Now copying as much data from the
33 * original packet as we can without
34 * exceeding 576 bytes (RFC 1812).
35 * Willy Konynenberg : Transparent proxying support.
36 * Keith Owens : RFC1191 correction for 4.2BSD based
37 * path MTU bug.
38 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
39 * valid (RFC 1812).
40 * Andi Kleen : Check all packet lengths properly
41 * and moved all kfree_skb() up to
42 * icmp_rcv.
43 * Andi Kleen : Move the rate limit bookkeeping
44 * into the dest entry and use a token
45 * bucket filter (thanks to ANK). Make
46 * the rates sysctl configurable.
47 * Yu Tianli : Fixed two ugly bugs in icmp_send
48 * - IP option length was accounted wrongly
49 * - ICMP header length was not accounted
50 * at all.
51 * Tristan Greaves : Added sysctl option to ignore bogus
52 * broadcast responses from broken routers.
53 *
54 * To Fix:
55 *
56 * - Should use skb_pull() instead of all the manual checking.
57 * This would also greatly simply some upper layer error handlers. --AK
58 */
59
60 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
61
62 #include <linux/module.h>
63 #include <linux/types.h>
64 #include <linux/jiffies.h>
65 #include <linux/kernel.h>
66 #include <linux/fcntl.h>
67 #include <linux/socket.h>
68 #include <linux/in.h>
69 #include <linux/inet.h>
70 #include <linux/inetdevice.h>
71 #include <linux/netdevice.h>
72 #include <linux/string.h>
73 #include <linux/netfilter_ipv4.h>
74 #include <linux/slab.h>
75 #include <net/snmp.h>
76 #include <net/ip.h>
77 #include <net/route.h>
78 #include <net/protocol.h>
79 #include <net/icmp.h>
80 #include <net/tcp.h>
81 #include <net/udp.h>
82 #include <net/raw.h>
83 #include <net/ping.h>
84 #include <linux/skbuff.h>
85 #include <net/sock.h>
86 #include <linux/errno.h>
87 #include <linux/timer.h>
88 #include <linux/init.h>
89 #include <linux/uaccess.h>
90 #include <net/checksum.h>
91 #include <net/xfrm.h>
92 #include <net/inet_common.h>
93 #include <net/ip_fib.h>
94 #include <net/l3mdev.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 __be32 times[3];
108 } data;
109 int head_len;
110 struct ip_options_data replyopts;
111 };
112
113 /* An array of errno for error messages from dest unreach. */
114 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
115
116 const struct icmp_err icmp_err_convert[] = {
117 {
118 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
119 .fatal = 0,
120 },
121 {
122 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
123 .fatal = 0,
124 },
125 {
126 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
127 .fatal = 1,
128 },
129 {
130 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
131 .fatal = 1,
132 },
133 {
134 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
135 .fatal = 0,
136 },
137 {
138 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
139 .fatal = 0,
140 },
141 {
142 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
143 .fatal = 1,
144 },
145 {
146 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
147 .fatal = 1,
148 },
149 {
150 .errno = ENONET, /* ICMP_HOST_ISOLATED */
151 .fatal = 1,
152 },
153 {
154 .errno = ENETUNREACH, /* ICMP_NET_ANO */
155 .fatal = 1,
156 },
157 {
158 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
159 .fatal = 1,
160 },
161 {
162 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
163 .fatal = 0,
164 },
165 {
166 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
167 .fatal = 0,
168 },
169 {
170 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
171 .fatal = 1,
172 },
173 {
174 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
175 .fatal = 1,
176 },
177 {
178 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
179 .fatal = 1,
180 },
181 };
182 EXPORT_SYMBOL(icmp_err_convert);
183
184 /*
185 * ICMP control array. This specifies what to do with each ICMP.
186 */
187
188 struct icmp_control {
189 enum skb_drop_reason (*handler)(struct sk_buff *skb);
190 short error; /* This ICMP is classed as an error message */
191 };
192
193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194
195 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
196
197 /* Called with BH disabled */
icmp_xmit_lock(struct net * net)198 static inline struct sock *icmp_xmit_lock(struct net *net)
199 {
200 struct sock *sk;
201
202 sk = this_cpu_read(ipv4_icmp_sk);
203
204 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
205 /* This can happen if the output path signals a
206 * dst_link_failure() for an outgoing ICMP packet.
207 */
208 return NULL;
209 }
210 sock_net_set(sk, net);
211 return sk;
212 }
213
icmp_xmit_unlock(struct sock * sk)214 static inline void icmp_xmit_unlock(struct sock *sk)
215 {
216 sock_net_set(sk, &init_net);
217 spin_unlock(&sk->sk_lock.slock);
218 }
219
220 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
221 int sysctl_icmp_msgs_burst __read_mostly = 50;
222
223 static struct {
224 spinlock_t lock;
225 u32 credit;
226 u32 stamp;
227 } icmp_global = {
228 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
229 };
230
231 /**
232 * icmp_global_allow - Are we allowed to send one more ICMP message ?
233 *
234 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
235 * Returns false if we reached the limit and can not send another packet.
236 * Note: called with BH disabled
237 */
icmp_global_allow(void)238 bool icmp_global_allow(void)
239 {
240 u32 credit, delta, incr = 0, now = (u32)jiffies;
241 bool rc = false;
242
243 /* Check if token bucket is empty and cannot be refilled
244 * without taking the spinlock. The READ_ONCE() are paired
245 * with the following WRITE_ONCE() in this same function.
246 */
247 if (!READ_ONCE(icmp_global.credit)) {
248 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
249 if (delta < HZ / 50)
250 return false;
251 }
252
253 spin_lock(&icmp_global.lock);
254 delta = min_t(u32, now - icmp_global.stamp, HZ);
255 if (delta >= HZ / 50) {
256 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
257 if (incr)
258 WRITE_ONCE(icmp_global.stamp, now);
259 }
260 credit = min_t(u32, icmp_global.credit + incr,
261 READ_ONCE(sysctl_icmp_msgs_burst));
262 if (credit) {
263 /* We want to use a credit of one in average, but need to randomize
264 * it for security reasons.
265 */
266 credit = max_t(int, credit - prandom_u32_max(3), 0);
267 rc = true;
268 }
269 WRITE_ONCE(icmp_global.credit, credit);
270 spin_unlock(&icmp_global.lock);
271 return rc;
272 }
273 EXPORT_SYMBOL(icmp_global_allow);
274
icmpv4_mask_allow(struct net * net,int type,int code)275 static bool icmpv4_mask_allow(struct net *net, int type, int code)
276 {
277 if (type > NR_ICMP_TYPES)
278 return true;
279
280 /* Don't limit PMTU discovery. */
281 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
282 return true;
283
284 /* Limit if icmp type is enabled in ratemask. */
285 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
286 return true;
287
288 return false;
289 }
290
icmpv4_global_allow(struct net * net,int type,int code)291 static bool icmpv4_global_allow(struct net *net, int type, int code)
292 {
293 if (icmpv4_mask_allow(net, type, code))
294 return true;
295
296 if (icmp_global_allow())
297 return true;
298
299 return false;
300 }
301
302 /*
303 * Send an ICMP frame.
304 */
305
icmpv4_xrlim_allow(struct net * net,struct rtable * rt,struct flowi4 * fl4,int type,int code)306 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
307 struct flowi4 *fl4, int type, int code)
308 {
309 struct dst_entry *dst = &rt->dst;
310 struct inet_peer *peer;
311 bool rc = true;
312 int vif;
313
314 if (icmpv4_mask_allow(net, type, code))
315 goto out;
316
317 /* No rate limit on loopback */
318 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
319 goto out;
320
321 vif = l3mdev_master_ifindex(dst->dev);
322 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
323 rc = inet_peer_xrlim_allow(peer,
324 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
325 if (peer)
326 inet_putpeer(peer);
327 out:
328 return rc;
329 }
330
331 /*
332 * Maintain the counters used in the SNMP statistics for outgoing ICMP
333 */
icmp_out_count(struct net * net,unsigned char type)334 void icmp_out_count(struct net *net, unsigned char type)
335 {
336 ICMPMSGOUT_INC_STATS(net, type);
337 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
338 }
339
340 /*
341 * Checksum each fragment, and on the first include the headers and final
342 * checksum.
343 */
icmp_glue_bits(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb)344 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
345 struct sk_buff *skb)
346 {
347 struct icmp_bxm *icmp_param = from;
348 __wsum csum;
349
350 csum = skb_copy_and_csum_bits(icmp_param->skb,
351 icmp_param->offset + offset,
352 to, len);
353
354 skb->csum = csum_block_add(skb->csum, csum, odd);
355 if (icmp_pointers[icmp_param->data.icmph.type].error)
356 nf_ct_attach(skb, icmp_param->skb);
357 return 0;
358 }
359
icmp_push_reply(struct sock * sk,struct icmp_bxm * icmp_param,struct flowi4 * fl4,struct ipcm_cookie * ipc,struct rtable ** rt)360 static void icmp_push_reply(struct sock *sk,
361 struct icmp_bxm *icmp_param,
362 struct flowi4 *fl4,
363 struct ipcm_cookie *ipc, struct rtable **rt)
364 {
365 struct sk_buff *skb;
366
367 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
368 icmp_param->data_len+icmp_param->head_len,
369 icmp_param->head_len,
370 ipc, rt, MSG_DONTWAIT) < 0) {
371 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
372 ip_flush_pending_frames(sk);
373 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
374 struct icmphdr *icmph = icmp_hdr(skb);
375 __wsum csum;
376 struct sk_buff *skb1;
377
378 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
379 (char *)icmph,
380 icmp_param->head_len);
381 skb_queue_walk(&sk->sk_write_queue, skb1) {
382 csum = csum_add(csum, skb1->csum);
383 }
384 icmph->checksum = csum_fold(csum);
385 skb->ip_summed = CHECKSUM_NONE;
386 ip_push_pending_frames(sk, fl4);
387 }
388 }
389
390 /*
391 * Driving logic for building and sending ICMP messages.
392 */
393
icmp_reply(struct icmp_bxm * icmp_param,struct sk_buff * skb)394 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
395 {
396 struct ipcm_cookie ipc;
397 struct rtable *rt = skb_rtable(skb);
398 struct net *net = dev_net(rt->dst.dev);
399 struct flowi4 fl4;
400 struct sock *sk;
401 struct inet_sock *inet;
402 __be32 daddr, saddr;
403 u32 mark = IP4_REPLY_MARK(net, skb->mark);
404 int type = icmp_param->data.icmph.type;
405 int code = icmp_param->data.icmph.code;
406
407 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
408 return;
409
410 /* Needed by both icmp_global_allow and icmp_xmit_lock */
411 local_bh_disable();
412
413 /* global icmp_msgs_per_sec */
414 if (!icmpv4_global_allow(net, type, code))
415 goto out_bh_enable;
416
417 sk = icmp_xmit_lock(net);
418 if (!sk)
419 goto out_bh_enable;
420 inet = inet_sk(sk);
421
422 icmp_param->data.icmph.checksum = 0;
423
424 ipcm_init(&ipc);
425 inet->tos = ip_hdr(skb)->tos;
426 ipc.sockc.mark = mark;
427 daddr = ipc.addr = ip_hdr(skb)->saddr;
428 saddr = fib_compute_spec_dst(skb);
429
430 if (icmp_param->replyopts.opt.opt.optlen) {
431 ipc.opt = &icmp_param->replyopts.opt;
432 if (ipc.opt->opt.srr)
433 daddr = icmp_param->replyopts.opt.opt.faddr;
434 }
435 memset(&fl4, 0, sizeof(fl4));
436 fl4.daddr = daddr;
437 fl4.saddr = saddr;
438 fl4.flowi4_mark = mark;
439 fl4.flowi4_uid = sock_net_uid(net, NULL);
440 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
441 fl4.flowi4_proto = IPPROTO_ICMP;
442 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
443 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
444 rt = ip_route_output_key(net, &fl4);
445 if (IS_ERR(rt))
446 goto out_unlock;
447 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
448 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
449 ip_rt_put(rt);
450 out_unlock:
451 icmp_xmit_unlock(sk);
452 out_bh_enable:
453 local_bh_enable();
454 }
455
456 /*
457 * The device used for looking up which routing table to use for sending an ICMP
458 * error is preferably the source whenever it is set, which should ensure the
459 * icmp error can be sent to the source host, else lookup using the routing
460 * table of the destination device, else use the main routing table (index 0).
461 */
icmp_get_route_lookup_dev(struct sk_buff * skb)462 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
463 {
464 struct net_device *route_lookup_dev = NULL;
465
466 if (skb->dev)
467 route_lookup_dev = skb->dev;
468 else if (skb_dst(skb))
469 route_lookup_dev = skb_dst(skb)->dev;
470 return route_lookup_dev;
471 }
472
icmp_route_lookup(struct net * net,struct flowi4 * fl4,struct sk_buff * skb_in,const struct iphdr * iph,__be32 saddr,u8 tos,u32 mark,int type,int code,struct icmp_bxm * param)473 static struct rtable *icmp_route_lookup(struct net *net,
474 struct flowi4 *fl4,
475 struct sk_buff *skb_in,
476 const struct iphdr *iph,
477 __be32 saddr, u8 tos, u32 mark,
478 int type, int code,
479 struct icmp_bxm *param)
480 {
481 struct net_device *route_lookup_dev;
482 struct rtable *rt, *rt2;
483 struct flowi4 fl4_dec;
484 int err;
485
486 memset(fl4, 0, sizeof(*fl4));
487 fl4->daddr = (param->replyopts.opt.opt.srr ?
488 param->replyopts.opt.opt.faddr : iph->saddr);
489 fl4->saddr = saddr;
490 fl4->flowi4_mark = mark;
491 fl4->flowi4_uid = sock_net_uid(net, NULL);
492 fl4->flowi4_tos = RT_TOS(tos);
493 fl4->flowi4_proto = IPPROTO_ICMP;
494 fl4->fl4_icmp_type = type;
495 fl4->fl4_icmp_code = code;
496 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
497 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
498
499 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
500 rt = ip_route_output_key_hash(net, fl4, skb_in);
501 if (IS_ERR(rt))
502 return rt;
503
504 /* No need to clone since we're just using its address. */
505 rt2 = rt;
506
507 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
508 flowi4_to_flowi(fl4), NULL, 0);
509 if (!IS_ERR(rt)) {
510 if (rt != rt2)
511 return rt;
512 } else if (PTR_ERR(rt) == -EPERM) {
513 rt = NULL;
514 } else
515 return rt;
516
517 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
518 if (err)
519 goto relookup_failed;
520
521 if (inet_addr_type_dev_table(net, route_lookup_dev,
522 fl4_dec.saddr) == RTN_LOCAL) {
523 rt2 = __ip_route_output_key(net, &fl4_dec);
524 if (IS_ERR(rt2))
525 err = PTR_ERR(rt2);
526 } else {
527 struct flowi4 fl4_2 = {};
528 unsigned long orefdst;
529
530 fl4_2.daddr = fl4_dec.saddr;
531 rt2 = ip_route_output_key(net, &fl4_2);
532 if (IS_ERR(rt2)) {
533 err = PTR_ERR(rt2);
534 goto relookup_failed;
535 }
536 /* Ugh! */
537 orefdst = skb_in->_skb_refdst; /* save old refdst */
538 skb_dst_set(skb_in, NULL);
539 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
540 RT_TOS(tos), rt2->dst.dev);
541
542 dst_release(&rt2->dst);
543 rt2 = skb_rtable(skb_in);
544 skb_in->_skb_refdst = orefdst; /* restore old refdst */
545 }
546
547 if (err)
548 goto relookup_failed;
549
550 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
551 flowi4_to_flowi(&fl4_dec), NULL,
552 XFRM_LOOKUP_ICMP);
553 if (!IS_ERR(rt2)) {
554 dst_release(&rt->dst);
555 memcpy(fl4, &fl4_dec, sizeof(*fl4));
556 rt = rt2;
557 } else if (PTR_ERR(rt2) == -EPERM) {
558 if (rt)
559 dst_release(&rt->dst);
560 return rt2;
561 } else {
562 err = PTR_ERR(rt2);
563 goto relookup_failed;
564 }
565 return rt;
566
567 relookup_failed:
568 if (rt)
569 return rt;
570 return ERR_PTR(err);
571 }
572
573 /*
574 * Send an ICMP message in response to a situation
575 *
576 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
577 * MAY send more (we do).
578 * MUST NOT change this header information.
579 * MUST NOT reply to a multicast/broadcast IP address.
580 * MUST NOT reply to a multicast/broadcast MAC address.
581 * MUST reply to only the first fragment.
582 */
583
__icmp_send(struct sk_buff * skb_in,int type,int code,__be32 info,const struct ip_options * opt)584 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
585 const struct ip_options *opt)
586 {
587 struct iphdr *iph;
588 int room;
589 struct icmp_bxm icmp_param;
590 struct rtable *rt = skb_rtable(skb_in);
591 struct ipcm_cookie ipc;
592 struct flowi4 fl4;
593 __be32 saddr;
594 u8 tos;
595 u32 mark;
596 struct net *net;
597 struct sock *sk;
598
599 if (!rt)
600 goto out;
601
602 if (rt->dst.dev)
603 net = dev_net(rt->dst.dev);
604 else if (skb_in->dev)
605 net = dev_net(skb_in->dev);
606 else
607 goto out;
608
609 /*
610 * Find the original header. It is expected to be valid, of course.
611 * Check this, icmp_send is called from the most obscure devices
612 * sometimes.
613 */
614 iph = ip_hdr(skb_in);
615
616 if ((u8 *)iph < skb_in->head ||
617 (skb_network_header(skb_in) + sizeof(*iph)) >
618 skb_tail_pointer(skb_in))
619 goto out;
620
621 /*
622 * No replies to physical multicast/broadcast
623 */
624 if (skb_in->pkt_type != PACKET_HOST)
625 goto out;
626
627 /*
628 * Now check at the protocol level
629 */
630 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
631 goto out;
632
633 /*
634 * Only reply to fragment 0. We byte re-order the constant
635 * mask for efficiency.
636 */
637 if (iph->frag_off & htons(IP_OFFSET))
638 goto out;
639
640 /*
641 * If we send an ICMP error to an ICMP error a mess would result..
642 */
643 if (icmp_pointers[type].error) {
644 /*
645 * We are an error, check if we are replying to an
646 * ICMP error
647 */
648 if (iph->protocol == IPPROTO_ICMP) {
649 u8 _inner_type, *itp;
650
651 itp = skb_header_pointer(skb_in,
652 skb_network_header(skb_in) +
653 (iph->ihl << 2) +
654 offsetof(struct icmphdr,
655 type) -
656 skb_in->data,
657 sizeof(_inner_type),
658 &_inner_type);
659 if (!itp)
660 goto out;
661
662 /*
663 * Assume any unknown ICMP type is an error. This
664 * isn't specified by the RFC, but think about it..
665 */
666 if (*itp > NR_ICMP_TYPES ||
667 icmp_pointers[*itp].error)
668 goto out;
669 }
670 }
671
672 /* Needed by both icmp_global_allow and icmp_xmit_lock */
673 local_bh_disable();
674
675 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
676 * incoming dev is loopback. If outgoing dev change to not be
677 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
678 */
679 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
680 !icmpv4_global_allow(net, type, code))
681 goto out_bh_enable;
682
683 sk = icmp_xmit_lock(net);
684 if (!sk)
685 goto out_bh_enable;
686
687 /*
688 * Construct source address and options.
689 */
690
691 saddr = iph->daddr;
692 if (!(rt->rt_flags & RTCF_LOCAL)) {
693 struct net_device *dev = NULL;
694
695 rcu_read_lock();
696 if (rt_is_input_route(rt) &&
697 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
698 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
699
700 if (dev)
701 saddr = inet_select_addr(dev, iph->saddr,
702 RT_SCOPE_LINK);
703 else
704 saddr = 0;
705 rcu_read_unlock();
706 }
707
708 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
709 IPTOS_PREC_INTERNETCONTROL) :
710 iph->tos;
711 mark = IP4_REPLY_MARK(net, skb_in->mark);
712
713 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
714 goto out_unlock;
715
716
717 /*
718 * Prepare data for ICMP header.
719 */
720
721 icmp_param.data.icmph.type = type;
722 icmp_param.data.icmph.code = code;
723 icmp_param.data.icmph.un.gateway = info;
724 icmp_param.data.icmph.checksum = 0;
725 icmp_param.skb = skb_in;
726 icmp_param.offset = skb_network_offset(skb_in);
727 inet_sk(sk)->tos = tos;
728 ipcm_init(&ipc);
729 ipc.addr = iph->saddr;
730 ipc.opt = &icmp_param.replyopts.opt;
731 ipc.sockc.mark = mark;
732
733 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
734 type, code, &icmp_param);
735 if (IS_ERR(rt))
736 goto out_unlock;
737
738 /* peer icmp_ratelimit */
739 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
740 goto ende;
741
742 /* RFC says return as much as we can without exceeding 576 bytes. */
743
744 room = dst_mtu(&rt->dst);
745 if (room > 576)
746 room = 576;
747 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
748 room -= sizeof(struct icmphdr);
749 /* Guard against tiny mtu. We need to include at least one
750 * IP network header for this message to make any sense.
751 */
752 if (room <= (int)sizeof(struct iphdr))
753 goto ende;
754
755 icmp_param.data_len = skb_in->len - icmp_param.offset;
756 if (icmp_param.data_len > room)
757 icmp_param.data_len = room;
758 icmp_param.head_len = sizeof(struct icmphdr);
759
760 /* if we don't have a source address at this point, fall back to the
761 * dummy address instead of sending out a packet with a source address
762 * of 0.0.0.0
763 */
764 if (!fl4.saddr)
765 fl4.saddr = htonl(INADDR_DUMMY);
766
767 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
768 ende:
769 ip_rt_put(rt);
770 out_unlock:
771 icmp_xmit_unlock(sk);
772 out_bh_enable:
773 local_bh_enable();
774 out:;
775 }
776 EXPORT_SYMBOL(__icmp_send);
777
778 #if IS_ENABLED(CONFIG_NF_NAT)
779 #include <net/netfilter/nf_conntrack.h>
icmp_ndo_send(struct sk_buff * skb_in,int type,int code,__be32 info)780 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
781 {
782 struct sk_buff *cloned_skb = NULL;
783 struct ip_options opts = { 0 };
784 enum ip_conntrack_info ctinfo;
785 struct nf_conn *ct;
786 __be32 orig_ip;
787
788 ct = nf_ct_get(skb_in, &ctinfo);
789 if (!ct || !(ct->status & IPS_SRC_NAT)) {
790 __icmp_send(skb_in, type, code, info, &opts);
791 return;
792 }
793
794 if (skb_shared(skb_in))
795 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
796
797 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
798 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
799 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
800 skb_network_offset(skb_in) + sizeof(struct iphdr))))
801 goto out;
802
803 orig_ip = ip_hdr(skb_in)->saddr;
804 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
805 __icmp_send(skb_in, type, code, info, &opts);
806 ip_hdr(skb_in)->saddr = orig_ip;
807 out:
808 consume_skb(cloned_skb);
809 }
810 EXPORT_SYMBOL(icmp_ndo_send);
811 #endif
812
icmp_socket_deliver(struct sk_buff * skb,u32 info)813 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
814 {
815 const struct iphdr *iph = (const struct iphdr *)skb->data;
816 const struct net_protocol *ipprot;
817 int protocol = iph->protocol;
818
819 /* Checkin full IP header plus 8 bytes of protocol to
820 * avoid additional coding at protocol handlers.
821 */
822 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
823 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
824 return;
825 }
826
827 raw_icmp_error(skb, protocol, info);
828
829 ipprot = rcu_dereference(inet_protos[protocol]);
830 if (ipprot && ipprot->err_handler)
831 ipprot->err_handler(skb, info);
832 }
833
icmp_tag_validation(int proto)834 static bool icmp_tag_validation(int proto)
835 {
836 bool ok;
837
838 rcu_read_lock();
839 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
840 rcu_read_unlock();
841 return ok;
842 }
843
844 /*
845 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
846 * ICMP_PARAMETERPROB.
847 */
848
icmp_unreach(struct sk_buff * skb)849 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
850 {
851 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
852 const struct iphdr *iph;
853 struct icmphdr *icmph;
854 struct net *net;
855 u32 info = 0;
856
857 net = dev_net(skb_dst(skb)->dev);
858
859 /*
860 * Incomplete header ?
861 * Only checks for the IP header, there should be an
862 * additional check for longer headers in upper levels.
863 */
864
865 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
866 goto out_err;
867
868 icmph = icmp_hdr(skb);
869 iph = (const struct iphdr *)skb->data;
870
871 if (iph->ihl < 5) { /* Mangled header, drop. */
872 reason = SKB_DROP_REASON_IP_INHDR;
873 goto out_err;
874 }
875
876 switch (icmph->type) {
877 case ICMP_DEST_UNREACH:
878 switch (icmph->code & 15) {
879 case ICMP_NET_UNREACH:
880 case ICMP_HOST_UNREACH:
881 case ICMP_PROT_UNREACH:
882 case ICMP_PORT_UNREACH:
883 break;
884 case ICMP_FRAG_NEEDED:
885 /* for documentation of the ip_no_pmtu_disc
886 * values please see
887 * Documentation/networking/ip-sysctl.rst
888 */
889 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
890 default:
891 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
892 &iph->daddr);
893 break;
894 case 2:
895 goto out;
896 case 3:
897 if (!icmp_tag_validation(iph->protocol))
898 goto out;
899 fallthrough;
900 case 0:
901 info = ntohs(icmph->un.frag.mtu);
902 }
903 break;
904 case ICMP_SR_FAILED:
905 net_dbg_ratelimited("%pI4: Source Route Failed\n",
906 &iph->daddr);
907 break;
908 default:
909 break;
910 }
911 if (icmph->code > NR_ICMP_UNREACH)
912 goto out;
913 break;
914 case ICMP_PARAMETERPROB:
915 info = ntohl(icmph->un.gateway) >> 24;
916 break;
917 case ICMP_TIME_EXCEEDED:
918 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
919 if (icmph->code == ICMP_EXC_FRAGTIME)
920 goto out;
921 break;
922 }
923
924 /*
925 * Throw it at our lower layers
926 *
927 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
928 * header.
929 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
930 * transport layer.
931 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
932 * transport layer.
933 */
934
935 /*
936 * Check the other end isn't violating RFC 1122. Some routers send
937 * bogus responses to broadcast frames. If you see this message
938 * first check your netmask matches at both ends, if it does then
939 * get the other vendor to fix their kit.
940 */
941
942 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
943 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
944 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
945 &ip_hdr(skb)->saddr,
946 icmph->type, icmph->code,
947 &iph->daddr, skb->dev->name);
948 goto out;
949 }
950
951 icmp_socket_deliver(skb, info);
952
953 out:
954 return reason;
955 out_err:
956 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
957 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
958 }
959
960
961 /*
962 * Handle ICMP_REDIRECT.
963 */
964
icmp_redirect(struct sk_buff * skb)965 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
966 {
967 if (skb->len < sizeof(struct iphdr)) {
968 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
969 return SKB_DROP_REASON_PKT_TOO_SMALL;
970 }
971
972 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
973 /* there aught to be a stat */
974 return SKB_DROP_REASON_NOMEM;
975 }
976
977 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
978 return SKB_NOT_DROPPED_YET;
979 }
980
981 /*
982 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
983 *
984 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
985 * requests.
986 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
987 * included in the reply.
988 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
989 * echo requests, MUST have default=NOT.
990 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
991 * Extended Echo Functionality, MUST be disabled by default
992 * See also WRT handling of options once they are done and working.
993 */
994
icmp_echo(struct sk_buff * skb)995 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
996 {
997 struct icmp_bxm icmp_param;
998 struct net *net;
999
1000 net = dev_net(skb_dst(skb)->dev);
1001 /* should there be an ICMP stat for ignored echos? */
1002 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1003 return SKB_NOT_DROPPED_YET;
1004
1005 icmp_param.data.icmph = *icmp_hdr(skb);
1006 icmp_param.skb = skb;
1007 icmp_param.offset = 0;
1008 icmp_param.data_len = skb->len;
1009 icmp_param.head_len = sizeof(struct icmphdr);
1010
1011 if (icmp_param.data.icmph.type == ICMP_ECHO)
1012 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1013 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1014 return SKB_NOT_DROPPED_YET;
1015
1016 icmp_reply(&icmp_param, skb);
1017 return SKB_NOT_DROPPED_YET;
1018 }
1019
1020 /* Helper for icmp_echo and icmpv6_echo_reply.
1021 * Searches for net_device that matches PROBE interface identifier
1022 * and builds PROBE reply message in icmphdr.
1023 *
1024 * Returns false if PROBE responses are disabled via sysctl
1025 */
1026
icmp_build_probe(struct sk_buff * skb,struct icmphdr * icmphdr)1027 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1028 {
1029 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1030 struct icmp_ext_echo_iio *iio, _iio;
1031 struct net *net = dev_net(skb->dev);
1032 struct net_device *dev;
1033 char buff[IFNAMSIZ];
1034 u16 ident_len;
1035 u8 status;
1036
1037 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1038 return false;
1039
1040 /* We currently only support probing interfaces on the proxy node
1041 * Check to ensure L-bit is set
1042 */
1043 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1044 return false;
1045 /* Clear status bits in reply message */
1046 icmphdr->un.echo.sequence &= htons(0xFF00);
1047 if (icmphdr->type == ICMP_EXT_ECHO)
1048 icmphdr->type = ICMP_EXT_ECHOREPLY;
1049 else
1050 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1051 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1052 /* Size of iio is class_type dependent.
1053 * Only check header here and assign length based on ctype in the switch statement
1054 */
1055 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1056 if (!ext_hdr || !iio)
1057 goto send_mal_query;
1058 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1059 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1060 goto send_mal_query;
1061 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1062 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1063 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1064 if (!iio)
1065 goto send_mal_query;
1066
1067 status = 0;
1068 dev = NULL;
1069 switch (iio->extobj_hdr.class_type) {
1070 case ICMP_EXT_ECHO_CTYPE_NAME:
1071 if (ident_len >= IFNAMSIZ)
1072 goto send_mal_query;
1073 memset(buff, 0, sizeof(buff));
1074 memcpy(buff, &iio->ident.name, ident_len);
1075 dev = dev_get_by_name(net, buff);
1076 break;
1077 case ICMP_EXT_ECHO_CTYPE_INDEX:
1078 if (ident_len != sizeof(iio->ident.ifindex))
1079 goto send_mal_query;
1080 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1081 break;
1082 case ICMP_EXT_ECHO_CTYPE_ADDR:
1083 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1084 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1085 iio->ident.addr.ctype3_hdr.addrlen)
1086 goto send_mal_query;
1087 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1088 case ICMP_AFI_IP:
1089 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1090 goto send_mal_query;
1091 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1092 break;
1093 #if IS_ENABLED(CONFIG_IPV6)
1094 case ICMP_AFI_IP6:
1095 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1096 goto send_mal_query;
1097 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1098 dev_hold(dev);
1099 break;
1100 #endif
1101 default:
1102 goto send_mal_query;
1103 }
1104 break;
1105 default:
1106 goto send_mal_query;
1107 }
1108 if (!dev) {
1109 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1110 return true;
1111 }
1112 /* Fill bits in reply message */
1113 if (dev->flags & IFF_UP)
1114 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1115 if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1116 status |= ICMP_EXT_ECHOREPLY_IPV4;
1117 if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1118 status |= ICMP_EXT_ECHOREPLY_IPV6;
1119 dev_put(dev);
1120 icmphdr->un.echo.sequence |= htons(status);
1121 return true;
1122 send_mal_query:
1123 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1124 return true;
1125 }
1126 EXPORT_SYMBOL_GPL(icmp_build_probe);
1127
1128 /*
1129 * Handle ICMP Timestamp requests.
1130 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1131 * SHOULD be in the kernel for minimum random latency.
1132 * MUST be accurate to a few minutes.
1133 * MUST be updated at least at 15Hz.
1134 */
icmp_timestamp(struct sk_buff * skb)1135 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1136 {
1137 struct icmp_bxm icmp_param;
1138 /*
1139 * Too short.
1140 */
1141 if (skb->len < 4)
1142 goto out_err;
1143
1144 /*
1145 * Fill in the current time as ms since midnight UT:
1146 */
1147 icmp_param.data.times[1] = inet_current_timestamp();
1148 icmp_param.data.times[2] = icmp_param.data.times[1];
1149
1150 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1151
1152 icmp_param.data.icmph = *icmp_hdr(skb);
1153 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1154 icmp_param.data.icmph.code = 0;
1155 icmp_param.skb = skb;
1156 icmp_param.offset = 0;
1157 icmp_param.data_len = 0;
1158 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1159 icmp_reply(&icmp_param, skb);
1160 return SKB_NOT_DROPPED_YET;
1161
1162 out_err:
1163 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1164 return SKB_DROP_REASON_PKT_TOO_SMALL;
1165 }
1166
icmp_discard(struct sk_buff * skb)1167 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1168 {
1169 /* pretend it was a success */
1170 return SKB_NOT_DROPPED_YET;
1171 }
1172
1173 /*
1174 * Deal with incoming ICMP packets.
1175 */
icmp_rcv(struct sk_buff * skb)1176 int icmp_rcv(struct sk_buff *skb)
1177 {
1178 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1179 struct rtable *rt = skb_rtable(skb);
1180 struct net *net = dev_net(rt->dst.dev);
1181 struct icmphdr *icmph;
1182
1183 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1184 struct sec_path *sp = skb_sec_path(skb);
1185 int nh;
1186
1187 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1188 XFRM_STATE_ICMP)) {
1189 reason = SKB_DROP_REASON_XFRM_POLICY;
1190 goto drop;
1191 }
1192
1193 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1194 goto drop;
1195
1196 nh = skb_network_offset(skb);
1197 skb_set_network_header(skb, sizeof(*icmph));
1198
1199 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1200 skb)) {
1201 reason = SKB_DROP_REASON_XFRM_POLICY;
1202 goto drop;
1203 }
1204
1205 skb_set_network_header(skb, nh);
1206 }
1207
1208 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1209
1210 if (skb_checksum_simple_validate(skb))
1211 goto csum_error;
1212
1213 if (!pskb_pull(skb, sizeof(*icmph)))
1214 goto error;
1215
1216 icmph = icmp_hdr(skb);
1217
1218 ICMPMSGIN_INC_STATS(net, icmph->type);
1219
1220 /* Check for ICMP Extended Echo (PROBE) messages */
1221 if (icmph->type == ICMP_EXT_ECHO) {
1222 /* We can't use icmp_pointers[].handler() because it is an array of
1223 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1224 */
1225 reason = icmp_echo(skb);
1226 goto reason_check;
1227 }
1228
1229 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1230 reason = ping_rcv(skb);
1231 goto reason_check;
1232 }
1233
1234 /*
1235 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1236 *
1237 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1238 * discarded.
1239 */
1240 if (icmph->type > NR_ICMP_TYPES) {
1241 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1242 goto error;
1243 }
1244
1245 /*
1246 * Parse the ICMP message
1247 */
1248
1249 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1250 /*
1251 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1252 * silently ignored (we let user decide with a sysctl).
1253 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1254 * discarded if to broadcast/multicast.
1255 */
1256 if ((icmph->type == ICMP_ECHO ||
1257 icmph->type == ICMP_TIMESTAMP) &&
1258 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1259 reason = SKB_DROP_REASON_INVALID_PROTO;
1260 goto error;
1261 }
1262 if (icmph->type != ICMP_ECHO &&
1263 icmph->type != ICMP_TIMESTAMP &&
1264 icmph->type != ICMP_ADDRESS &&
1265 icmph->type != ICMP_ADDRESSREPLY) {
1266 reason = SKB_DROP_REASON_INVALID_PROTO;
1267 goto error;
1268 }
1269 }
1270
1271 reason = icmp_pointers[icmph->type].handler(skb);
1272 reason_check:
1273 if (!reason) {
1274 consume_skb(skb);
1275 return NET_RX_SUCCESS;
1276 }
1277
1278 drop:
1279 kfree_skb_reason(skb, reason);
1280 return NET_RX_DROP;
1281 csum_error:
1282 reason = SKB_DROP_REASON_ICMP_CSUM;
1283 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1284 error:
1285 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1286 goto drop;
1287 }
1288
ip_icmp_error_rfc4884_validate(const struct sk_buff * skb,int off)1289 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1290 {
1291 struct icmp_extobj_hdr *objh, _objh;
1292 struct icmp_ext_hdr *exth, _exth;
1293 u16 olen;
1294
1295 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1296 if (!exth)
1297 return false;
1298 if (exth->version != 2)
1299 return true;
1300
1301 if (exth->checksum &&
1302 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1303 return false;
1304
1305 off += sizeof(_exth);
1306 while (off < skb->len) {
1307 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1308 if (!objh)
1309 return false;
1310
1311 olen = ntohs(objh->length);
1312 if (olen < sizeof(_objh))
1313 return false;
1314
1315 off += olen;
1316 if (off > skb->len)
1317 return false;
1318 }
1319
1320 return true;
1321 }
1322
ip_icmp_error_rfc4884(const struct sk_buff * skb,struct sock_ee_data_rfc4884 * out,int thlen,int off)1323 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1324 struct sock_ee_data_rfc4884 *out,
1325 int thlen, int off)
1326 {
1327 int hlen;
1328
1329 /* original datagram headers: end of icmph to payload (skb->data) */
1330 hlen = -skb_transport_offset(skb) - thlen;
1331
1332 /* per rfc 4884: minimal datagram length of 128 bytes */
1333 if (off < 128 || off < hlen)
1334 return;
1335
1336 /* kernel has stripped headers: return payload offset in bytes */
1337 off -= hlen;
1338 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1339 return;
1340
1341 out->len = off;
1342
1343 if (!ip_icmp_error_rfc4884_validate(skb, off))
1344 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1345 }
1346 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1347
icmp_err(struct sk_buff * skb,u32 info)1348 int icmp_err(struct sk_buff *skb, u32 info)
1349 {
1350 struct iphdr *iph = (struct iphdr *)skb->data;
1351 int offset = iph->ihl<<2;
1352 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1353 int type = icmp_hdr(skb)->type;
1354 int code = icmp_hdr(skb)->code;
1355 struct net *net = dev_net(skb->dev);
1356
1357 /*
1358 * Use ping_err to handle all icmp errors except those
1359 * triggered by ICMP_ECHOREPLY which sent from kernel.
1360 */
1361 if (icmph->type != ICMP_ECHOREPLY) {
1362 ping_err(skb, offset, info);
1363 return 0;
1364 }
1365
1366 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1367 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1368 else if (type == ICMP_REDIRECT)
1369 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1370
1371 return 0;
1372 }
1373
1374 /*
1375 * This table is the definition of how we handle ICMP.
1376 */
1377 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1378 [ICMP_ECHOREPLY] = {
1379 .handler = ping_rcv,
1380 },
1381 [1] = {
1382 .handler = icmp_discard,
1383 .error = 1,
1384 },
1385 [2] = {
1386 .handler = icmp_discard,
1387 .error = 1,
1388 },
1389 [ICMP_DEST_UNREACH] = {
1390 .handler = icmp_unreach,
1391 .error = 1,
1392 },
1393 [ICMP_SOURCE_QUENCH] = {
1394 .handler = icmp_unreach,
1395 .error = 1,
1396 },
1397 [ICMP_REDIRECT] = {
1398 .handler = icmp_redirect,
1399 .error = 1,
1400 },
1401 [6] = {
1402 .handler = icmp_discard,
1403 .error = 1,
1404 },
1405 [7] = {
1406 .handler = icmp_discard,
1407 .error = 1,
1408 },
1409 [ICMP_ECHO] = {
1410 .handler = icmp_echo,
1411 },
1412 [9] = {
1413 .handler = icmp_discard,
1414 .error = 1,
1415 },
1416 [10] = {
1417 .handler = icmp_discard,
1418 .error = 1,
1419 },
1420 [ICMP_TIME_EXCEEDED] = {
1421 .handler = icmp_unreach,
1422 .error = 1,
1423 },
1424 [ICMP_PARAMETERPROB] = {
1425 .handler = icmp_unreach,
1426 .error = 1,
1427 },
1428 [ICMP_TIMESTAMP] = {
1429 .handler = icmp_timestamp,
1430 },
1431 [ICMP_TIMESTAMPREPLY] = {
1432 .handler = icmp_discard,
1433 },
1434 [ICMP_INFO_REQUEST] = {
1435 .handler = icmp_discard,
1436 },
1437 [ICMP_INFO_REPLY] = {
1438 .handler = icmp_discard,
1439 },
1440 [ICMP_ADDRESS] = {
1441 .handler = icmp_discard,
1442 },
1443 [ICMP_ADDRESSREPLY] = {
1444 .handler = icmp_discard,
1445 },
1446 };
1447
icmp_sk_init(struct net * net)1448 static int __net_init icmp_sk_init(struct net *net)
1449 {
1450 /* Control parameters for ECHO replies. */
1451 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1452 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1453 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1454
1455 /* Control parameter - ignore bogus broadcast responses? */
1456 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1457
1458 /*
1459 * Configurable global rate limit.
1460 *
1461 * ratelimit defines tokens/packet consumed for dst->rate_token
1462 * bucket ratemask defines which icmp types are ratelimited by
1463 * setting it's bit position.
1464 *
1465 * default:
1466 * dest unreachable (3), source quench (4),
1467 * time exceeded (11), parameter problem (12)
1468 */
1469
1470 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1471 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1472 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1473
1474 return 0;
1475 }
1476
1477 static struct pernet_operations __net_initdata icmp_sk_ops = {
1478 .init = icmp_sk_init,
1479 };
1480
icmp_init(void)1481 int __init icmp_init(void)
1482 {
1483 int err, i;
1484
1485 for_each_possible_cpu(i) {
1486 struct sock *sk;
1487
1488 err = inet_ctl_sock_create(&sk, PF_INET,
1489 SOCK_RAW, IPPROTO_ICMP, &init_net);
1490 if (err < 0)
1491 return err;
1492
1493 per_cpu(ipv4_icmp_sk, i) = sk;
1494
1495 /* Enough space for 2 64K ICMP packets, including
1496 * sk_buff/skb_shared_info struct overhead.
1497 */
1498 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1499
1500 /*
1501 * Speedup sock_wfree()
1502 */
1503 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1504 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1505 }
1506 return register_pernet_subsys(&icmp_sk_ops);
1507 }
1508