projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[deliverable/linux.git] / net / ipv4 / ip_options.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The options processing module for ip.c
7 *
8 * Authors: A.N.Kuznetsov
9 *
10 */
11
12 #define pr_fmt(fmt) "IPv4: " fmt
13
14 #include <linux/capability.h>
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
18 #include <asm/uaccess.h>
19 #include <asm/unaligned.h>
20 #include <linux/skbuff.h>
21 #include <linux/ip.h>
22 #include <linux/icmp.h>
23 #include <linux/netdevice.h>
24 #include <linux/rtnetlink.h>
25 #include <net/sock.h>
26 #include <net/ip.h>
27 #include <net/icmp.h>
28 #include <net/route.h>
29 #include <net/cipso_ipv4.h>
30 #include <net/ip_fib.h>
31
32 /*
33 * Write options to IP header, record destination address to
34 * source route option, address of outgoing interface
35 * (we should already know it, so that this function is allowed be
36 * called only after routing decision) and timestamp,
37 * if we originate this datagram.
38 *
39 * daddr is real destination address, next hop is recorded in IP header.
40 * saddr is address of outgoing interface.
41 */
42
43 void ip_options_build(struct sk_buff *skb, struct ip_options *opt,
44 __be32 daddr, struct rtable *rt, int is_frag)
45 {
46 unsigned char *iph = skb_network_header(skb);
47
48 memcpy(&(IPCB(skb)->opt), opt, sizeof(struct ip_options));
49 memcpy(iph+sizeof(struct iphdr), opt->__data, opt->optlen);
50 opt = &(IPCB(skb)->opt);
51
52 if (opt->srr)
53 memcpy(iph+opt->srr+iph[opt->srr+1]-4, &daddr, 4);
54
55 if (!is_frag) {
56 if (opt->rr_needaddr)
57 ip_rt_get_source(iph+opt->rr+iph[opt->rr+2]-5, skb, rt);
58 if (opt->ts_needaddr)
59 ip_rt_get_source(iph+opt->ts+iph[opt->ts+2]-9, skb, rt);
60 if (opt->ts_needtime) {
61 struct timespec tv;
62 __be32 midtime;
63 getnstimeofday(&tv);
64 midtime = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC + tv.tv_nsec / NSEC_PER_MSEC);
65 memcpy(iph+opt->ts+iph[opt->ts+2]-5, &midtime, 4);
66 }
67 return;
68 }
69 if (opt->rr) {
70 memset(iph+opt->rr, IPOPT_NOP, iph[opt->rr+1]);
71 opt->rr = 0;
72 opt->rr_needaddr = 0;
73 }
74 if (opt->ts) {
75 memset(iph+opt->ts, IPOPT_NOP, iph[opt->ts+1]);
76 opt->ts = 0;
77 opt->ts_needaddr = opt->ts_needtime = 0;
78 }
79 }
80
81 /*
82 * Provided (sopt, skb) points to received options,
83 * build in dopt compiled option set appropriate for answering.
84 * i.e. invert SRR option, copy anothers,
85 * and grab room in RR/TS options.
86 *
87 * NOTE: dopt cannot point to skb.
88 */
89
90 int ip_options_echo(struct ip_options *dopt, struct sk_buff *skb)
91 {
92 const struct ip_options *sopt;
93 unsigned char *sptr, *dptr;
94 int soffset, doffset;
95 int optlen;
96
97 memset(dopt, 0, sizeof(struct ip_options));
98
99 sopt = &(IPCB(skb)->opt);
100
101 if (sopt->optlen == 0)
102 return 0;
103
104 sptr = skb_network_header(skb);
105 dptr = dopt->__data;
106
107 if (sopt->rr) {
108 optlen = sptr[sopt->rr+1];
109 soffset = sptr[sopt->rr+2];
110 dopt->rr = dopt->optlen + sizeof(struct iphdr);
111 memcpy(dptr, sptr+sopt->rr, optlen);
112 if (sopt->rr_needaddr && soffset <= optlen) {
113 if (soffset + 3 > optlen)
114 return -EINVAL;
115 dptr[2] = soffset + 4;
116 dopt->rr_needaddr = 1;
117 }
118 dptr += optlen;
119 dopt->optlen += optlen;
120 }
121 if (sopt->ts) {
122 optlen = sptr[sopt->ts+1];
123 soffset = sptr[sopt->ts+2];
124 dopt->ts = dopt->optlen + sizeof(struct iphdr);
125 memcpy(dptr, sptr+sopt->ts, optlen);
126 if (soffset <= optlen) {
127 if (sopt->ts_needaddr) {
128 if (soffset + 3 > optlen)
129 return -EINVAL;
130 dopt->ts_needaddr = 1;
131 soffset += 4;
132 }
133 if (sopt->ts_needtime) {
134 if (soffset + 3 > optlen)
135 return -EINVAL;
136 if ((dptr[3]&0xF) != IPOPT_TS_PRESPEC) {
137 dopt->ts_needtime = 1;
138 soffset += 4;
139 } else {
140 dopt->ts_needtime = 0;
141
142 if (soffset + 7 <= optlen) {
143 __be32 addr;
144
145 memcpy(&addr, dptr+soffset-1, 4);
146 if (inet_addr_type(dev_net(skb_dst(skb)->dev), addr) != RTN_UNICAST) {
147 dopt->ts_needtime = 1;
148 soffset += 8;
149 }
150 }
151 }
152 }
153 dptr[2] = soffset;
154 }
155 dptr += optlen;
156 dopt->optlen += optlen;
157 }
158 if (sopt->srr) {
159 unsigned char *start = sptr+sopt->srr;
160 __be32 faddr;
161
162 optlen = start[1];
163 soffset = start[2];
164 doffset = 0;
165 if (soffset > optlen)
166 soffset = optlen + 1;
167 soffset -= 4;
168 if (soffset > 3) {
169 memcpy(&faddr, &start[soffset-1], 4);
170 for (soffset -= 4, doffset = 4; soffset > 3; soffset -= 4, doffset += 4)
171 memcpy(&dptr[doffset-1], &start[soffset-1], 4);
172 /*
173 * RFC1812 requires to fix illegal source routes.
174 */
175 if (memcmp(&ip_hdr(skb)->saddr,
176 &start[soffset + 3], 4) == 0)
177 doffset -= 4;
178 }
179 if (doffset > 3) {
180 __be32 daddr = fib_compute_spec_dst(skb);
181
182 memcpy(&start[doffset-1], &daddr, 4);
183 dopt->faddr = faddr;
184 dptr[0] = start[0];
185 dptr[1] = doffset+3;
186 dptr[2] = 4;
187 dptr += doffset+3;
188 dopt->srr = dopt->optlen + sizeof(struct iphdr);
189 dopt->optlen += doffset+3;
190 dopt->is_strictroute = sopt->is_strictroute;
191 }
192 }
193 if (sopt->cipso) {
194 optlen = sptr[sopt->cipso+1];
195 dopt->cipso = dopt->optlen+sizeof(struct iphdr);
196 memcpy(dptr, sptr+sopt->cipso, optlen);
197 dptr += optlen;
198 dopt->optlen += optlen;
199 }
200 while (dopt->optlen & 3) {
201 *dptr++ = IPOPT_END;
202 dopt->optlen++;
203 }
204 return 0;
205 }
206
207 /*
208 * Options "fragmenting", just fill options not
209 * allowed in fragments with NOOPs.
210 * Simple and stupid 8), but the most efficient way.
211 */
212
213 void ip_options_fragment(struct sk_buff *skb)
214 {
215 unsigned char *optptr = skb_network_header(skb) + sizeof(struct iphdr);
216 struct ip_options *opt = &(IPCB(skb)->opt);
217 int l = opt->optlen;
218 int optlen;
219
220 while (l > 0) {
221 switch (*optptr) {
222 case IPOPT_END:
223 return;
224 case IPOPT_NOOP:
225 l--;
226 optptr++;
227 continue;
228 }
229 optlen = optptr[1];
230 if (optlen < 2 || optlen > l)
231 return;
232 if (!IPOPT_COPIED(*optptr))
233 memset(optptr, IPOPT_NOOP, optlen);
234 l -= optlen;
235 optptr += optlen;
236 }
237 opt->ts = 0;
238 opt->rr = 0;
239 opt->rr_needaddr = 0;
240 opt->ts_needaddr = 0;
241 opt->ts_needtime = 0;
242 }
243
244 /* helper used by ip_options_compile() to call fib_compute_spec_dst()
245 * at most one time.
246 */
247 static void spec_dst_fill(__be32 *spec_dst, struct sk_buff *skb)
248 {
249 if (*spec_dst == htonl(INADDR_ANY))
250 *spec_dst = fib_compute_spec_dst(skb);
251 }
252
253 /*
254 * Verify options and fill pointers in struct options.
255 * Caller should clear *opt, and set opt->data.
256 * If opt == NULL, then skb->data should point to IP header.
257 */
258
259 int ip_options_compile(struct net *net,
260 struct ip_options *opt, struct sk_buff *skb)
261 {
262 __be32 spec_dst = htonl(INADDR_ANY);
263 unsigned char *pp_ptr = NULL;
264 struct rtable *rt = NULL;
265 unsigned char *optptr;
266 unsigned char *iph;
267 int optlen, l;
268
269 if (skb != NULL) {
270 rt = skb_rtable(skb);
271 optptr = (unsigned char *)&(ip_hdr(skb)[1]);
272 } else
273 optptr = opt->__data;
274 iph = optptr - sizeof(struct iphdr);
275
276 for (l = opt->optlen; l > 0; ) {
277 switch (*optptr) {
278 case IPOPT_END:
279 for (optptr++, l--; l > 0; optptr++, l--) {
280 if (*optptr != IPOPT_END) {
281 *optptr = IPOPT_END;
282 opt->is_changed = 1;
283 }
284 }
285 goto eol;
286 case IPOPT_NOOP:
287 l--;
288 optptr++;
289 continue;
290 }
291 if (unlikely(l < 2)) {
292 pp_ptr = optptr;
293 goto error;
294 }
295 optlen = optptr[1];
296 if (optlen < 2 || optlen > l) {
297 pp_ptr = optptr;
298 goto error;
299 }
300 switch (*optptr) {
301 case IPOPT_SSRR:
302 case IPOPT_LSRR:
303 if (optlen < 3) {
304 pp_ptr = optptr + 1;
305 goto error;
306 }
307 if (optptr[2] < 4) {
308 pp_ptr = optptr + 2;
309 goto error;
310 }
311 /* NB: cf RFC-1812 5.2.4.1 */
312 if (opt->srr) {
313 pp_ptr = optptr;
314 goto error;
315 }
316 if (!skb) {
317 if (optptr[2] != 4 || optlen < 7 || ((optlen-3) & 3)) {
318 pp_ptr = optptr + 1;
319 goto error;
320 }
321 memcpy(&opt->faddr, &optptr[3], 4);
322 if (optlen > 7)
323 memmove(&optptr[3], &optptr[7], optlen-7);
324 }
325 opt->is_strictroute = (optptr[0] == IPOPT_SSRR);
326 opt->srr = optptr - iph;
327 break;
328 case IPOPT_RR:
329 if (opt->rr) {
330 pp_ptr = optptr;
331 goto error;
332 }
333 if (optlen < 3) {
334 pp_ptr = optptr + 1;
335 goto error;
336 }
337 if (optptr[2] < 4) {
338 pp_ptr = optptr + 2;
339 goto error;
340 }
341 if (optptr[2] <= optlen) {
342 if (optptr[2]+3 > optlen) {
343 pp_ptr = optptr + 2;
344 goto error;
345 }
346 if (rt) {
347 spec_dst_fill(&spec_dst, skb);
348 memcpy(&optptr[optptr[2]-1], &spec_dst, 4);
349 opt->is_changed = 1;
350 }
351 optptr[2] += 4;
352 opt->rr_needaddr = 1;
353 }
354 opt->rr = optptr - iph;
355 break;
356 case IPOPT_TIMESTAMP:
357 if (opt->ts) {
358 pp_ptr = optptr;
359 goto error;
360 }
361 if (optlen < 4) {
362 pp_ptr = optptr + 1;
363 goto error;
364 }
365 if (optptr[2] < 5) {
366 pp_ptr = optptr + 2;
367 goto error;
368 }
369 if (optptr[2] <= optlen) {
370 unsigned char *timeptr = NULL;
371 if (optptr[2]+3 > optlen) {
372 pp_ptr = optptr + 2;
373 goto error;
374 }
375 switch (optptr[3]&0xF) {
376 case IPOPT_TS_TSONLY:
377 if (skb)
378 timeptr = &optptr[optptr[2]-1];
379 opt->ts_needtime = 1;
380 optptr[2] += 4;
381 break;
382 case IPOPT_TS_TSANDADDR:
383 if (optptr[2]+7 > optlen) {
384 pp_ptr = optptr + 2;
385 goto error;
386 }
387 if (rt) {
388 spec_dst_fill(&spec_dst, skb);
389 memcpy(&optptr[optptr[2]-1], &spec_dst, 4);
390 timeptr = &optptr[optptr[2]+3];
391 }
392 opt->ts_needaddr = 1;
393 opt->ts_needtime = 1;
394 optptr[2] += 8;
395 break;
396 case IPOPT_TS_PRESPEC:
397 if (optptr[2]+7 > optlen) {
398 pp_ptr = optptr + 2;
399 goto error;
400 }
401 {
402 __be32 addr;
403 memcpy(&addr, &optptr[optptr[2]-1], 4);
404 if (inet_addr_type(net, addr) == RTN_UNICAST)
405 break;
406 if (skb)
407 timeptr = &optptr[optptr[2]+3];
408 }
409 opt->ts_needtime = 1;
410 optptr[2] += 8;
411 break;
412 default:
413 if (!skb && !ns_capable(net->user_ns, CAP_NET_RAW)) {
414 pp_ptr = optptr + 3;
415 goto error;
416 }
417 break;
418 }
419 if (timeptr) {
420 struct timespec tv;
421 u32 midtime;
422 getnstimeofday(&tv);
423 midtime = (tv.tv_sec % 86400) * MSEC_PER_SEC + tv.tv_nsec / NSEC_PER_MSEC;
424 put_unaligned_be32(midtime, timeptr);
425 opt->is_changed = 1;
426 }
427 } else if ((optptr[3]&0xF) != IPOPT_TS_PRESPEC) {
428 unsigned int overflow = optptr[3]>>4;
429 if (overflow == 15) {
430 pp_ptr = optptr + 3;
431 goto error;
432 }
433 if (skb) {
434 optptr[3] = (optptr[3]&0xF)|((overflow+1)<<4);
435 opt->is_changed = 1;
436 }
437 }
438 opt->ts = optptr - iph;
439 break;
440 case IPOPT_RA:
441 if (optlen < 4) {
442 pp_ptr = optptr + 1;
443 goto error;
444 }
445 if (optptr[2] == 0 && optptr[3] == 0)
446 opt->router_alert = optptr - iph;
447 break;
448 case IPOPT_CIPSO:
449 if ((!skb && !ns_capable(net->user_ns, CAP_NET_RAW)) || opt->cipso) {
450 pp_ptr = optptr;
451 goto error;
452 }
453 opt->cipso = optptr - iph;
454 if (cipso_v4_validate(skb, &optptr)) {
455 pp_ptr = optptr;
456 goto error;
457 }
458 break;
459 case IPOPT_SEC:
460 case IPOPT_SID:
461 default:
462 if (!skb && !ns_capable(net->user_ns, CAP_NET_RAW)) {
463 pp_ptr = optptr;
464 goto error;
465 }
466 break;
467 }
468 l -= optlen;
469 optptr += optlen;
470 }
471
472 eol:
473 if (!pp_ptr)
474 return 0;
475
476 error:
477 if (skb) {
478 icmp_send(skb, ICMP_PARAMETERPROB, 0, htonl((pp_ptr-iph)<<24));
479 }
480 return -EINVAL;
481 }
482 EXPORT_SYMBOL(ip_options_compile);
483
484 /*
485 * Undo all the changes done by ip_options_compile().
486 */
487
488 void ip_options_undo(struct ip_options *opt)
489 {
490 if (opt->srr) {
491 unsigned char *optptr = opt->__data+opt->srr-sizeof(struct iphdr);
492 memmove(optptr+7, optptr+3, optptr[1]-7);
493 memcpy(optptr+3, &opt->faddr, 4);
494 }
495 if (opt->rr_needaddr) {
496 unsigned char *optptr = opt->__data+opt->rr-sizeof(struct iphdr);
497 optptr[2] -= 4;
498 memset(&optptr[optptr[2]-1], 0, 4);
499 }
500 if (opt->ts) {
501 unsigned char *optptr = opt->__data+opt->ts-sizeof(struct iphdr);
502 if (opt->ts_needtime) {
503 optptr[2] -= 4;
504 memset(&optptr[optptr[2]-1], 0, 4);
505 if ((optptr[3]&0xF) == IPOPT_TS_PRESPEC)
506 optptr[2] -= 4;
507 }
508 if (opt->ts_needaddr) {
509 optptr[2] -= 4;
510 memset(&optptr[optptr[2]-1], 0, 4);
511 }
512 }
513 }
514
515 static struct ip_options_rcu *ip_options_get_alloc(const int optlen)
516 {
517 return kzalloc(sizeof(struct ip_options_rcu) + ((optlen + 3) & ~3),
518 GFP_KERNEL);
519 }
520
521 static int ip_options_get_finish(struct net *net, struct ip_options_rcu **optp,
522 struct ip_options_rcu *opt, int optlen)
523 {
524 while (optlen & 3)
525 opt->opt.__data[optlen++] = IPOPT_END;
526 opt->opt.optlen = optlen;
527 if (optlen && ip_options_compile(net, &opt->opt, NULL)) {
528 kfree(opt);
529 return -EINVAL;
530 }
531 kfree(*optp);
532 *optp = opt;
533 return 0;
534 }
535
536 int ip_options_get_from_user(struct net *net, struct ip_options_rcu **optp,
537 unsigned char __user *data, int optlen)
538 {
539 struct ip_options_rcu *opt = ip_options_get_alloc(optlen);
540
541 if (!opt)
542 return -ENOMEM;
543 if (optlen && copy_from_user(opt->opt.__data, data, optlen)) {
544 kfree(opt);
545 return -EFAULT;
546 }
547 return ip_options_get_finish(net, optp, opt, optlen);
548 }
549
550 int ip_options_get(struct net *net, struct ip_options_rcu **optp,
551 unsigned char *data, int optlen)
552 {
553 struct ip_options_rcu *opt = ip_options_get_alloc(optlen);
554
555 if (!opt)
556 return -ENOMEM;
557 if (optlen)
558 memcpy(opt->opt.__data, data, optlen);
559 return ip_options_get_finish(net, optp, opt, optlen);
560 }
561
562 void ip_forward_options(struct sk_buff *skb)
563 {
564 struct ip_options *opt = &(IPCB(skb)->opt);
565 unsigned char *optptr;
566 struct rtable *rt = skb_rtable(skb);
567 unsigned char *raw = skb_network_header(skb);
568
569 if (opt->rr_needaddr) {
570 optptr = (unsigned char *)raw + opt->rr;
571 ip_rt_get_source(&optptr[optptr[2]-5], skb, rt);
572 opt->is_changed = 1;
573 }
574 if (opt->srr_is_hit) {
575 int srrptr, srrspace;
576
577 optptr = raw + opt->srr;
578
579 for ( srrptr = optptr[2], srrspace = optptr[1];
580 srrptr <= srrspace;
581 srrptr += 4
582 ) {
583 if (srrptr + 3 > srrspace)
584 break;
585 if (memcmp(&opt->nexthop, &optptr[srrptr-1], 4) == 0)
586 break;
587 }
588 if (srrptr + 3 <= srrspace) {
589 opt->is_changed = 1;
590 ip_hdr(skb)->daddr = opt->nexthop;
591 ip_rt_get_source(&optptr[srrptr-1], skb, rt);
592 optptr[2] = srrptr+4;
593 } else {
594 net_crit_ratelimited("%s(): Argh! Destination lost!\n",
595 __func__);
596 }
597 if (opt->ts_needaddr) {
598 optptr = raw + opt->ts;
599 ip_rt_get_source(&optptr[optptr[2]-9], skb, rt);
600 opt->is_changed = 1;
601 }
602 }
603 if (opt->is_changed) {
604 opt->is_changed = 0;
605 ip_send_check(ip_hdr(skb));
606 }
607 }
608
609 int ip_options_rcv_srr(struct sk_buff *skb)
610 {
611 struct ip_options *opt = &(IPCB(skb)->opt);
612 int srrspace, srrptr;
613 __be32 nexthop;
614 struct iphdr *iph = ip_hdr(skb);
615 unsigned char *optptr = skb_network_header(skb) + opt->srr;
616 struct rtable *rt = skb_rtable(skb);
617 struct rtable *rt2;
618 unsigned long orefdst;
619 int err;
620
621 if (!rt)
622 return 0;
623
624 if (skb->pkt_type != PACKET_HOST)
625 return -EINVAL;
626 if (rt->rt_type == RTN_UNICAST) {
627 if (!opt->is_strictroute)
628 return 0;
629 icmp_send(skb, ICMP_PARAMETERPROB, 0, htonl(16<<24));
630 return -EINVAL;
631 }
632 if (rt->rt_type != RTN_LOCAL)
633 return -EINVAL;
634
635 for (srrptr = optptr[2], srrspace = optptr[1]; srrptr <= srrspace; srrptr += 4) {
636 if (srrptr + 3 > srrspace) {
637 icmp_send(skb, ICMP_PARAMETERPROB, 0, htonl((opt->srr+2)<<24));
638 return -EINVAL;
639 }
640 memcpy(&nexthop, &optptr[srrptr-1], 4);
641
642 orefdst = skb->_skb_refdst;
643 skb_dst_set(skb, NULL);
644 err = ip_route_input(skb, nexthop, iph->saddr, iph->tos, skb->dev);
645 rt2 = skb_rtable(skb);
646 if (err || (rt2->rt_type != RTN_UNICAST && rt2->rt_type != RTN_LOCAL)) {
647 skb_dst_drop(skb);
648 skb->_skb_refdst = orefdst;
649 return -EINVAL;
650 }
651 refdst_drop(orefdst);
652 if (rt2->rt_type != RTN_LOCAL)
653 break;
654 /* Superfast 8) loopback forward */
655 iph->daddr = nexthop;
656 opt->is_changed = 1;
657 }
658 if (srrptr <= srrspace) {
659 opt->srr_is_hit = 1;
660 opt->nexthop = nexthop;
661 opt->is_changed = 1;
662 }
663 return 0;
664 }
665 EXPORT_SYMBOL(ip_options_rcv_srr);
Linux kernel - Deliverables
This page took 0.062942 seconds and 5 git commands to generate.