1 /*
2 * IPv6 fragment reassembly for connection tracking
3 *
4 * Copyright (C)2004 USAGI/WIDE Project
5 *
6 * Author:
7 * Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
8 *
9 * Based on: net/ipv6/reassembly.c
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
15 */
16
17 #define pr_fmt(fmt) "IPv6-nf: " fmt
18
19 #include <linux/errno.h>
20 #include <linux/types.h>
21 #include <linux/string.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/jiffies.h>
25 #include <linux/net.h>
26 #include <linux/list.h>
27 #include <linux/netdevice.h>
28 #include <linux/in6.h>
29 #include <linux/ipv6.h>
30 #include <linux/icmpv6.h>
31 #include <linux/random.h>
32 #include <linux/slab.h>
33
34 #include <net/sock.h>
35 #include <net/snmp.h>
36 #include <net/ipv6_frag.h>
37
38 #include <net/protocol.h>
39 #include <net/transp_v6.h>
40 #include <net/rawv6.h>
41 #include <net/ndisc.h>
42 #include <net/addrconf.h>
43 #include <net/inet_ecn.h>
44 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
45 #include <linux/sysctl.h>
46 #include <linux/netfilter.h>
47 #include <linux/netfilter_ipv6.h>
48 #include <linux/kernel.h>
49 #include <linux/module.h>
50 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
51
52 static const char nf_frags_cache_name[] = "nf-frags";
53
54 static struct inet_frags nf_frags;
55
56 #ifdef CONFIG_SYSCTL
57
58 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
59 {
60 .procname = "nf_conntrack_frag6_timeout",
61 .data = &init_net.nf_frag.frags.timeout,
62 .maxlen = sizeof(unsigned int),
63 .mode = 0644,
64 .proc_handler = proc_dointvec_jiffies,
65 },
66 {
67 .procname = "nf_conntrack_frag6_low_thresh",
68 .data = &init_net.nf_frag.frags.low_thresh,
69 .maxlen = sizeof(unsigned long),
70 .mode = 0644,
71 .proc_handler = proc_doulongvec_minmax,
72 .extra2 = &init_net.nf_frag.frags.high_thresh
73 },
74 {
75 .procname = "nf_conntrack_frag6_high_thresh",
76 .data = &init_net.nf_frag.frags.high_thresh,
77 .maxlen = sizeof(unsigned long),
78 .mode = 0644,
79 .proc_handler = proc_doulongvec_minmax,
80 .extra1 = &init_net.nf_frag.frags.low_thresh
81 },
82 { }
83 };
84
85 static int nf_ct_frag6_sysctl_register(struct net *net)
86 {
87 struct ctl_table *table;
88 struct ctl_table_header *hdr;
89
90 table = nf_ct_frag6_sysctl_table;
91 if (!net_eq(net, &init_net)) {
92 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
93 GFP_KERNEL);
94 if (table == NULL)
95 goto err_alloc;
96
97 table[0].data = &net->nf_frag.frags.timeout;
98 table[1].data = &net->nf_frag.frags.low_thresh;
99 table[1].extra2 = &net->nf_frag.frags.high_thresh;
100 table[2].data = &net->nf_frag.frags.high_thresh;
101 table[2].extra1 = &net->nf_frag.frags.low_thresh;
102 table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
103 }
104
105 hdr = register_net_sysctl(net, "net/netfilter", table);
106 if (hdr == NULL)
107 goto err_reg;
108
109 net->nf_frag_frags_hdr = hdr;
110 return 0;
111
112 err_reg:
113 if (!net_eq(net, &init_net))
114 kfree(table);
115 err_alloc:
116 return -ENOMEM;
117 }
118
119 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
120 {
121 struct ctl_table *table;
122
123 table = net->nf_frag_frags_hdr->ctl_table_arg;
124 unregister_net_sysctl_table(net->nf_frag_frags_hdr);
125 if (!net_eq(net, &init_net))
126 kfree(table);
127 }
128
129 #else
130 static int nf_ct_frag6_sysctl_register(struct net *net)
131 {
132 return 0;
133 }
134 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
135 {
136 }
137 #endif
138
139 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
140 {
141 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
142 }
143
144 static void nf_ct_frag6_expire(struct timer_list *t)
145 {
146 struct inet_frag_queue *frag = from_timer(frag, t, timer);
147 struct frag_queue *fq;
148 struct net *net;
149
150 fq = container_of(frag, struct frag_queue, q);
151 net = container_of(fq->q.net, struct net, nf_frag.frags);
152
153 ip6frag_expire_frag_queue(net, fq);
154 }
155
156 /* Creation primitives. */
157 static struct frag_queue *fq_find(struct net *net, __be32 id, u32 user,
158 const struct ipv6hdr *hdr, int iif)
159 {
160 struct frag_v6_compare_key key = {
161 .id = id,
162 .saddr = hdr->saddr,
163 .daddr = hdr->daddr,
164 .user = user,
165 .iif = iif,
166 };
167 struct inet_frag_queue *q;
168
169 q = inet_frag_find(&net->nf_frag.frags, &key);
170 if (!q)
171 return NULL;
172
173 return container_of(q, struct frag_queue, q);
174 }
175
176
177 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
178 const struct frag_hdr *fhdr, int nhoff)
179 {
180 struct sk_buff *prev, *next;
181 unsigned int payload_len;
182 int offset, end;
183 u8 ecn;
184
185 if (fq->q.flags & INET_FRAG_COMPLETE) {
186 pr_debug("Already completed\n");
187 goto err;
188 }
189
190 payload_len = ntohs(ipv6_hdr(skb)->payload_len);
191
192 offset = ntohs(fhdr->frag_off) & ~0x7;
193 end = offset + (payload_len -
194 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
195
196 if ((unsigned int)end > IPV6_MAXPLEN) {
197 pr_debug("offset is too large.\n");
198 return -EINVAL;
199 }
200
201 ecn = ip6_frag_ecn(ipv6_hdr(skb));
202
203 if (skb->ip_summed == CHECKSUM_COMPLETE) {
204 const unsigned char *nh = skb_network_header(skb);
205 skb->csum = csum_sub(skb->csum,
206 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
207 0));
208 }
209
210 /* Is this the final fragment? */
211 if (!(fhdr->frag_off & htons(IP6_MF))) {
212 /* If we already have some bits beyond end
213 * or have different end, the segment is corrupted.
214 */
215 if (end < fq->q.len ||
216 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
217 pr_debug("already received last fragment\n");
218 goto err;
219 }
220 fq->q.flags |= INET_FRAG_LAST_IN;
221 fq->q.len = end;
222 } else {
223 /* Check if the fragment is rounded to 8 bytes.
224 * Required by the RFC.
225 */
226 if (end & 0x7) {
227 /* RFC2460 says always send parameter problem in
228 * this case. -DaveM
229 */
230 pr_debug("end of fragment not rounded to 8 bytes.\n");
231 inet_frag_kill(&fq->q);
232 return -EPROTO;
233 }
234 if (end > fq->q.len) {
235 /* Some bits beyond end -> corruption. */
236 if (fq->q.flags & INET_FRAG_LAST_IN) {
237 pr_debug("last packet already reached.\n");
238 goto err;
239 }
240 fq->q.len = end;
241 }
242 }
243
244 if (end == offset)
245 goto err;
246
247 /* Point into the IP datagram 'data' part. */
248 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
249 pr_debug("queue: message is too short.\n");
250 goto err;
251 }
252 if (pskb_trim_rcsum(skb, end - offset)) {
253 pr_debug("Can't trim\n");
254 goto err;
255 }
256
257 /* Find out which fragments are in front and at the back of us
258 * in the chain of fragments so far. We must know where to put
259 * this fragment, right?
260 */
261 prev = fq->q.fragments_tail;
262 if (!prev || prev->ip_defrag_offset < offset) {
263 next = NULL;
264 goto found;
265 }
266 prev = NULL;
267 for (next = fq->q.fragments; next != NULL; next = next->next) {
268 if (next->ip_defrag_offset >= offset)
269 break; /* bingo! */
270 prev = next;
271 }
272
273 found:
274 /* RFC5722, Section 4:
275 * When reassembling an IPv6 datagram, if
276 * one or more its constituent fragments is determined to be an
277 * overlapping fragment, the entire datagram (and any constituent
278 * fragments, including those not yet received) MUST be silently
279 * discarded.
280 */
281
282 /* Check for overlap with preceding fragment. */
283 if (prev &&
284 (prev->ip_defrag_offset + prev->len) > offset)
285 goto discard_fq;
286
287 /* Look for overlap with succeeding segment. */
288 if (next && next->ip_defrag_offset < end)
289 goto discard_fq;
290
291 /* Note : skb->ip_defrag_offset and skb->dev share the same location */
292 if (skb->dev)
293 fq->iif = skb->dev->ifindex;
294 /* Makes sure compiler wont do silly aliasing games */
295 barrier();
296 skb->ip_defrag_offset = offset;
297
298 /* Insert this fragment in the chain of fragments. */
299 skb->next = next;
300 if (!next)
301 fq->q.fragments_tail = skb;
302 if (prev)
303 prev->next = skb;
304 else
305 fq->q.fragments = skb;
306
307 fq->q.stamp = skb->tstamp;
308 fq->q.meat += skb->len;
309 fq->ecn |= ecn;
310 if (payload_len > fq->q.max_size)
311 fq->q.max_size = payload_len;
312 add_frag_mem_limit(fq->q.net, skb->truesize);
313
314 /* The first fragment.
315 * nhoffset is obtained from the first fragment, of course.
316 */
317 if (offset == 0) {
318 fq->nhoffset = nhoff;
319 fq->q.flags |= INET_FRAG_FIRST_IN;
320 }
321
322 return 0;
323
324 discard_fq:
325 inet_frag_kill(&fq->q);
326 err:
327 return -EINVAL;
328 }
329
330 /*
331 * Check if this packet is complete.
332 *
333 * It is called with locked fq, and caller must check that
334 * queue is eligible for reassembly i.e. it is not COMPLETE,
335 * the last and the first frames arrived and all the bits are here.
336 *
337 * returns true if *prev skb has been transformed into the reassembled
338 * skb, false otherwise.
339 */
340 static bool
341 nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *prev, struct net_device *dev)
342 {
343 struct sk_buff *fp, *head = fq->q.fragments;
344 int payload_len, delta;
345 u8 ecn;
346
347 inet_frag_kill(&fq->q);
348
349 WARN_ON(head == NULL);
350 WARN_ON(head->ip_defrag_offset != 0);
351
352 ecn = ip_frag_ecn_table[fq->ecn];
353 if (unlikely(ecn == 0xff))
354 return false;
355
356 /* Unfragmented part is taken from the first segment. */
357 payload_len = ((head->data - skb_network_header(head)) -
358 sizeof(struct ipv6hdr) + fq->q.len -
359 sizeof(struct frag_hdr));
360 if (payload_len > IPV6_MAXPLEN) {
361 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
362 payload_len);
363 return false;
364 }
365
366 delta = - head->truesize;
367
368 /* Head of list must not be cloned. */
369 if (skb_unclone(head, GFP_ATOMIC))
370 return false;
371
372 delta += head->truesize;
373 if (delta)
374 add_frag_mem_limit(fq->q.net, delta);
375
376 /* If the first fragment is fragmented itself, we split
377 * it to two chunks: the first with data and paged part
378 * and the second, holding only fragments. */
379 if (skb_has_frag_list(head)) {
380 struct sk_buff *clone;
381 int i, plen = 0;
382
383 clone = alloc_skb(0, GFP_ATOMIC);
384 if (clone == NULL)
385 return false;
386
387 clone->next = head->next;
388 head->next = clone;
389 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
390 skb_frag_list_init(head);
391 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
392 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
393 clone->len = clone->data_len = head->data_len - plen;
394 head->data_len -= clone->len;
395 head->len -= clone->len;
396 clone->csum = 0;
397 clone->ip_summed = head->ip_summed;
398
399 add_frag_mem_limit(fq->q.net, clone->truesize);
400 }
401
402 /* morph head into last received skb: prev.
403 *
404 * This allows callers of ipv6 conntrack defrag to continue
405 * to use the last skb(frag) passed into the reasm engine.
406 * The last skb frag 'silently' turns into the full reassembled skb.
407 *
408 * Since prev is also part of q->fragments we have to clone it first.
409 */
410 if (head != prev) {
411 struct sk_buff *iter;
412
413 fp = skb_clone(prev, GFP_ATOMIC);
414 if (!fp)
415 return false;
416
417 fp->next = prev->next;
418
419 iter = head;
420 while (iter) {
421 if (iter->next == prev) {
422 iter->next = fp;
423 break;
424 }
425 iter = iter->next;
426 }
427
428 skb_morph(prev, head);
429 prev->next = head->next;
430 consume_skb(head);
431 head = prev;
432 }
433
434 /* We have to remove fragment header from datagram and to relocate
435 * header in order to calculate ICV correctly. */
436 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
437 memmove(head->head + sizeof(struct frag_hdr), head->head,
438 (head->data - head->head) - sizeof(struct frag_hdr));
439 head->mac_header += sizeof(struct frag_hdr);
440 head->network_header += sizeof(struct frag_hdr);
441
442 skb_shinfo(head)->frag_list = head->next;
443 skb_reset_transport_header(head);
444 skb_push(head, head->data - skb_network_header(head));
445
446 for (fp = head->next; fp; fp = fp->next) {
447 head->data_len += fp->len;
448 head->len += fp->len;
449 if (head->ip_summed != fp->ip_summed)
450 head->ip_summed = CHECKSUM_NONE;
451 else if (head->ip_summed == CHECKSUM_COMPLETE)
452 head->csum = csum_add(head->csum, fp->csum);
453 head->truesize += fp->truesize;
454 fp->sk = NULL;
455 }
456 sub_frag_mem_limit(fq->q.net, head->truesize);
457
458 head->ignore_df = 1;
459 skb_mark_not_on_list(head);
460 head->dev = dev;
461 head->tstamp = fq->q.stamp;
462 ipv6_hdr(head)->payload_len = htons(payload_len);
463 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
464 IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
465
466 /* Yes, and fold redundant checksum back. 8) */
467 if (head->ip_summed == CHECKSUM_COMPLETE)
468 head->csum = csum_partial(skb_network_header(head),
469 skb_network_header_len(head),
470 head->csum);
471
472 fq->q.fragments = NULL;
473 fq->q.rb_fragments = RB_ROOT;
474 fq->q.fragments_tail = NULL;
475
476 return true;
477 }
478
479 /*
480 * find the header just before Fragment Header.
481 *
482 * if success return 0 and set ...
483 * (*prevhdrp): the value of "Next Header Field" in the header
484 * just before Fragment Header.
485 * (*prevhoff): the offset of "Next Header Field" in the header
486 * just before Fragment Header.
487 * (*fhoff) : the offset of Fragment Header.
488 *
489 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
490 *
491 */
492 static int
493 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
494 {
495 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
496 const int netoff = skb_network_offset(skb);
497 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
498 int start = netoff + sizeof(struct ipv6hdr);
499 int len = skb->len - start;
500 u8 prevhdr = NEXTHDR_IPV6;
501
502 while (nexthdr != NEXTHDR_FRAGMENT) {
503 struct ipv6_opt_hdr hdr;
504 int hdrlen;
505
506 if (!ipv6_ext_hdr(nexthdr)) {
507 return -1;
508 }
509 if (nexthdr == NEXTHDR_NONE) {
510 pr_debug("next header is none\n");
511 return -1;
512 }
513 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
514 pr_debug("too short\n");
515 return -1;
516 }
517 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
518 BUG();
519 if (nexthdr == NEXTHDR_AUTH)
520 hdrlen = (hdr.hdrlen+2)<<2;
521 else
522 hdrlen = ipv6_optlen(&hdr);
523
524 prevhdr = nexthdr;
525 prev_nhoff = start;
526
527 nexthdr = hdr.nexthdr;
528 len -= hdrlen;
529 start += hdrlen;
530 }
531
532 if (len < 0)
533 return -1;
534
535 *prevhdrp = prevhdr;
536 *prevhoff = prev_nhoff;
537 *fhoff = start;
538
539 return 0;
540 }
541
542 int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
543 {
544 u16 savethdr = skb->transport_header;
545 struct net_device *dev = skb->dev;
546 int fhoff, nhoff, ret;
547 struct frag_hdr *fhdr;
548 struct frag_queue *fq;
549 struct ipv6hdr *hdr;
550 u8 prevhdr;
551
552 /* Jumbo payload inhibits frag. header */
553 if (ipv6_hdr(skb)->payload_len == 0) {
554 pr_debug("payload len = 0\n");
555 return 0;
556 }
557
558 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
559 return 0;
560
561 if (!pskb_may_pull(skb, fhoff + sizeof(*fhdr)))
562 return -ENOMEM;
563
564 skb_set_transport_header(skb, fhoff);
565 hdr = ipv6_hdr(skb);
566 fhdr = (struct frag_hdr *)skb_transport_header(skb);
567
568 if (skb->len - skb_network_offset(skb) < IPV6_MIN_MTU &&
569 fhdr->frag_off & htons(IP6_MF))
570 return -EINVAL;
571
572 skb_orphan(skb);
573 fq = fq_find(net, fhdr->identification, user, hdr,
574 skb->dev ? skb->dev->ifindex : 0);
575 if (fq == NULL) {
576 pr_debug("Can't find and can't create new queue\n");
577 return -ENOMEM;
578 }
579
580 spin_lock_bh(&fq->q.lock);
581
582 ret = nf_ct_frag6_queue(fq, skb, fhdr, nhoff);
583 if (ret < 0) {
584 if (ret == -EPROTO) {
585 skb->transport_header = savethdr;
586 ret = 0;
587 }
588 goto out_unlock;
589 }
590
591 /* after queue has assumed skb ownership, only 0 or -EINPROGRESS
592 * must be returned.
593 */
594 ret = -EINPROGRESS;
595 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
596 fq->q.meat == fq->q.len) {
597 unsigned long orefdst = skb->_skb_refdst;
598
599 skb->_skb_refdst = 0UL;
600 if (nf_ct_frag6_reasm(fq, skb, dev))
601 ret = 0;
602 skb->_skb_refdst = orefdst;
603 } else {
604 skb_dst_drop(skb);
605 }
606
607 out_unlock:
608 spin_unlock_bh(&fq->q.lock);
609 inet_frag_put(&fq->q);
610 return ret;
611 }
612 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
613
614 static int nf_ct_net_init(struct net *net)
615 {
616 int res;
617
618 net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
619 net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
620 net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
621 net->nf_frag.frags.f = &nf_frags;
622
623 res = inet_frags_init_net(&net->nf_frag.frags);
624 if (res < 0)
625 return res;
626 res = nf_ct_frag6_sysctl_register(net);
627 if (res < 0)
628 inet_frags_exit_net(&net->nf_frag.frags);
629 return res;
630 }
631
632 static void nf_ct_net_exit(struct net *net)
633 {
634 nf_ct_frags6_sysctl_unregister(net);
635 inet_frags_exit_net(&net->nf_frag.frags);
636 }
637
638 static struct pernet_operations nf_ct_net_ops = {
639 .init = nf_ct_net_init,
640 .exit = nf_ct_net_exit,
641 };
642
643 static const struct rhashtable_params nfct_rhash_params = {
644 .head_offset = offsetof(struct inet_frag_queue, node),
645 .hashfn = ip6frag_key_hashfn,
646 .obj_hashfn = ip6frag_obj_hashfn,
647 .obj_cmpfn = ip6frag_obj_cmpfn,
648 .automatic_shrinking = true,
649 };
650
651 int nf_ct_frag6_init(void)
652 {
653 int ret = 0;
654
655 nf_frags.constructor = ip6frag_init;
656 nf_frags.destructor = NULL;
657 nf_frags.qsize = sizeof(struct frag_queue);
658 nf_frags.frag_expire = nf_ct_frag6_expire;
659 nf_frags.frags_cache_name = nf_frags_cache_name;
660 nf_frags.rhash_params = nfct_rhash_params;
661 ret = inet_frags_init(&nf_frags);
662 if (ret)
663 goto out;
664 ret = register_pernet_subsys(&nf_ct_net_ops);
665 if (ret)
666 inet_frags_fini(&nf_frags);
667
668 out:
669 return ret;
670 }
671
672 void nf_ct_frag6_cleanup(void)
673 {
674 unregister_pernet_subsys(&nf_ct_net_ops);
675 inet_frags_fini(&nf_frags);
676 }
677
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.