2 * IPv6 fragment reassembly for connection tracking
4 * Copyright (C)2004 USAGI/WIDE Project
7 * Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
9 * Based on: net/ipv6/reassembly.c
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.
17 #define pr_fmt(fmt) "IPv6-nf: " fmt
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>
36 #include <net/ipv6_frag.h>
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>
52 static const char nf_frags_cache_name[] = "nf-frags";
54 static struct inet_frags nf_frags;
58 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
60 .procname = "nf_conntrack_frag6_timeout",
61 .data = &init_net.nf_frag.frags.timeout,
62 .maxlen = sizeof(unsigned int),
64 .proc_handler = proc_dointvec_jiffies,
67 .procname = "nf_conntrack_frag6_low_thresh",
68 .data = &init_net.nf_frag.frags.low_thresh,
69 .maxlen = sizeof(unsigned long),
71 .proc_handler = proc_doulongvec_minmax,
72 .extra2 = &init_net.nf_frag.frags.high_thresh
75 .procname = "nf_conntrack_frag6_high_thresh",
76 .data = &init_net.nf_frag.frags.high_thresh,
77 .maxlen = sizeof(unsigned long),
79 .proc_handler = proc_doulongvec_minmax,
80 .extra1 = &init_net.nf_frag.frags.low_thresh
85 static int nf_ct_frag6_sysctl_register(struct net *net)
87 struct ctl_table *table;
88 struct ctl_table_header *hdr;
90 table = nf_ct_frag6_sysctl_table;
91 if (!net_eq(net, &init_net)) {
92 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
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;
105 hdr = register_net_sysctl(net, "net/netfilter", table);
109 net->nf_frag_frags_hdr = hdr;
113 if (!net_eq(net, &init_net))
119 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
121 struct ctl_table *table;
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))
130 static int nf_ct_frag6_sysctl_register(struct net *net)
134 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
139 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
140 struct sk_buff *prev_tail, struct net_device *dev);
142 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
144 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
147 static void nf_ct_frag6_expire(unsigned long data)
149 struct frag_queue *fq;
152 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
153 net = container_of(fq->q.net, struct net, nf_frag.frags);
155 ip6frag_expire_frag_queue(net, fq);
158 /* Creation primitives. */
159 static struct frag_queue *fq_find(struct net *net, __be32 id, u32 user,
160 const struct ipv6hdr *hdr, int iif)
162 struct frag_v6_compare_key key = {
169 struct inet_frag_queue *q;
171 q = inet_frag_find(&net->nf_frag.frags, &key);
175 return container_of(q, struct frag_queue, q);
179 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
180 const struct frag_hdr *fhdr, int nhoff)
182 unsigned int payload_len;
183 struct net_device *dev;
184 struct sk_buff *prev;
185 int offset, end, err;
188 if (fq->q.flags & INET_FRAG_COMPLETE) {
189 pr_debug("Already completed\n");
193 payload_len = ntohs(ipv6_hdr(skb)->payload_len);
195 offset = ntohs(fhdr->frag_off) & ~0x7;
196 end = offset + (payload_len -
197 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
199 if ((unsigned int)end > IPV6_MAXPLEN) {
200 pr_debug("offset is too large.\n");
204 ecn = ip6_frag_ecn(ipv6_hdr(skb));
206 if (skb->ip_summed == CHECKSUM_COMPLETE) {
207 const unsigned char *nh = skb_network_header(skb);
208 skb->csum = csum_sub(skb->csum,
209 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
213 /* Is this the final fragment? */
214 if (!(fhdr->frag_off & htons(IP6_MF))) {
215 /* If we already have some bits beyond end
216 * or have different end, the segment is corrupted.
218 if (end < fq->q.len ||
219 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
220 pr_debug("already received last fragment\n");
223 fq->q.flags |= INET_FRAG_LAST_IN;
226 /* Check if the fragment is rounded to 8 bytes.
227 * Required by the RFC.
230 /* RFC2460 says always send parameter problem in
233 pr_debug("end of fragment not rounded to 8 bytes.\n");
234 inet_frag_kill(&fq->q);
237 if (end > fq->q.len) {
238 /* Some bits beyond end -> corruption. */
239 if (fq->q.flags & INET_FRAG_LAST_IN) {
240 pr_debug("last packet already reached.\n");
250 /* Point into the IP datagram 'data' part. */
251 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
252 pr_debug("queue: message is too short.\n");
255 if (pskb_trim_rcsum(skb, end - offset)) {
256 pr_debug("Can't trim\n");
260 /* Note : skb->rbnode and skb->dev share the same location. */
262 /* Makes sure compiler wont do silly aliasing games */
265 prev = fq->q.fragments_tail;
266 err = inet_frag_queue_insert(&fq->q, skb, offset, end);
268 if (err == IPFRAG_DUP) {
269 /* No error for duplicates, pretend they got queued. */
277 fq->iif = dev->ifindex;
279 fq->q.stamp = skb->tstamp;
280 fq->q.meat += skb->len;
282 if (payload_len > fq->q.max_size)
283 fq->q.max_size = payload_len;
284 add_frag_mem_limit(fq->q.net, skb->truesize);
286 /* The first fragment.
287 * nhoffset is obtained from the first fragment, of course.
290 fq->nhoffset = nhoff;
291 fq->q.flags |= INET_FRAG_FIRST_IN;
294 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
295 fq->q.meat == fq->q.len) {
296 unsigned long orefdst = skb->_skb_refdst;
298 skb->_skb_refdst = 0UL;
299 err = nf_ct_frag6_reasm(fq, skb, prev, dev);
300 skb->_skb_refdst = orefdst;
302 /* After queue has assumed skb ownership, only 0 or
303 * -EINPROGRESS must be returned.
305 return err ? -EINPROGRESS : 0;
312 inet_frag_kill(&fq->q);
319 * Check if this packet is complete.
321 * It is called with locked fq, and caller must check that
322 * queue is eligible for reassembly i.e. it is not COMPLETE,
323 * the last and the first frames arrived and all the bits are here.
325 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
326 struct sk_buff *prev_tail, struct net_device *dev)
332 inet_frag_kill(&fq->q);
334 ecn = ip_frag_ecn_table[fq->ecn];
335 if (unlikely(ecn == 0xff))
338 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
342 payload_len = ((skb->data - skb_network_header(skb)) -
343 sizeof(struct ipv6hdr) + fq->q.len -
344 sizeof(struct frag_hdr));
345 if (payload_len > IPV6_MAXPLEN) {
346 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
351 /* We have to remove fragment header from datagram and to relocate
352 * header in order to calculate ICV correctly. */
353 skb_network_header(skb)[fq->nhoffset] = skb_transport_header(skb)[0];
354 memmove(skb->head + sizeof(struct frag_hdr), skb->head,
355 (skb->data - skb->head) - sizeof(struct frag_hdr));
356 skb->mac_header += sizeof(struct frag_hdr);
357 skb->network_header += sizeof(struct frag_hdr);
359 skb_reset_transport_header(skb);
361 inet_frag_reasm_finish(&fq->q, skb, reasm_data);
365 ipv6_hdr(skb)->payload_len = htons(payload_len);
366 ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
367 IP6CB(skb)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
369 /* Yes, and fold redundant checksum back. 8) */
370 if (skb->ip_summed == CHECKSUM_COMPLETE)
371 skb->csum = csum_partial(skb_network_header(skb),
372 skb_network_header_len(skb),
375 fq->q.fragments = NULL;
376 fq->q.rb_fragments = RB_ROOT;
377 fq->q.fragments_tail = NULL;
378 fq->q.last_run_head = NULL;
383 inet_frag_kill(&fq->q);
388 * find the header just before Fragment Header.
390 * if success return 0 and set ...
391 * (*prevhdrp): the value of "Next Header Field" in the header
392 * just before Fragment Header.
393 * (*prevhoff): the offset of "Next Header Field" in the header
394 * just before Fragment Header.
395 * (*fhoff) : the offset of Fragment Header.
397 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
401 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
403 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
404 const int netoff = skb_network_offset(skb);
405 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
406 int start = netoff + sizeof(struct ipv6hdr);
407 int len = skb->len - start;
408 u8 prevhdr = NEXTHDR_IPV6;
410 while (nexthdr != NEXTHDR_FRAGMENT) {
411 struct ipv6_opt_hdr hdr;
414 if (!ipv6_ext_hdr(nexthdr)) {
417 if (nexthdr == NEXTHDR_NONE) {
418 pr_debug("next header is none\n");
421 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
422 pr_debug("too short\n");
425 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
427 if (nexthdr == NEXTHDR_AUTH)
428 hdrlen = (hdr.hdrlen+2)<<2;
430 hdrlen = ipv6_optlen(&hdr);
435 nexthdr = hdr.nexthdr;
444 *prevhoff = prev_nhoff;
450 int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
452 u16 savethdr = skb->transport_header;
453 int fhoff, nhoff, ret;
454 struct frag_hdr *fhdr;
455 struct frag_queue *fq;
459 /* Jumbo payload inhibits frag. header */
460 if (ipv6_hdr(skb)->payload_len == 0) {
461 pr_debug("payload len = 0\n");
465 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
468 if (!pskb_may_pull(skb, fhoff + sizeof(*fhdr)))
471 skb_set_transport_header(skb, fhoff);
473 fhdr = (struct frag_hdr *)skb_transport_header(skb);
476 fq = fq_find(net, fhdr->identification, user, hdr,
477 skb->dev ? skb->dev->ifindex : 0);
479 pr_debug("Can't find and can't create new queue\n");
483 spin_lock_bh(&fq->q.lock);
485 ret = nf_ct_frag6_queue(fq, skb, fhdr, nhoff);
486 if (ret == -EPROTO) {
487 skb->transport_header = savethdr;
491 spin_unlock_bh(&fq->q.lock);
492 inet_frag_put(&fq->q);
495 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
497 static int nf_ct_net_init(struct net *net)
501 net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
502 net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
503 net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
504 net->nf_frag.frags.f = &nf_frags;
506 res = inet_frags_init_net(&net->nf_frag.frags);
509 res = nf_ct_frag6_sysctl_register(net);
511 inet_frags_exit_net(&net->nf_frag.frags);
515 static void nf_ct_net_exit(struct net *net)
517 nf_ct_frags6_sysctl_unregister(net);
518 inet_frags_exit_net(&net->nf_frag.frags);
521 static struct pernet_operations nf_ct_net_ops = {
522 .init = nf_ct_net_init,
523 .exit = nf_ct_net_exit,
526 static const struct rhashtable_params nfct_rhash_params = {
527 .head_offset = offsetof(struct inet_frag_queue, node),
528 .hashfn = ip6frag_key_hashfn,
529 .obj_hashfn = ip6frag_obj_hashfn,
530 .obj_cmpfn = ip6frag_obj_cmpfn,
531 .automatic_shrinking = true,
534 int nf_ct_frag6_init(void)
538 nf_frags.constructor = ip6frag_init;
539 nf_frags.destructor = NULL;
540 nf_frags.qsize = sizeof(struct frag_queue);
541 nf_frags.frag_expire = nf_ct_frag6_expire;
542 nf_frags.frags_cache_name = nf_frags_cache_name;
543 nf_frags.rhash_params = nfct_rhash_params;
544 ret = inet_frags_init(&nf_frags);
547 ret = register_pernet_subsys(&nf_ct_net_ops);
549 inet_frags_fini(&nf_frags);
555 void nf_ct_frag6_cleanup(void)
557 unregister_pernet_subsys(&nf_ct_net_ops);
558 inet_frags_fini(&nf_frags);