1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * IPv6 fragment reassembly for connection tracking
5 * Copyright (C)2004 USAGI/WIDE Project
8 * Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
10 * Based on: net/ipv6/reassembly.c
13 #define pr_fmt(fmt) "IPv6-nf: " fmt
15 #include <linux/errno.h>
16 #include <linux/types.h>
17 #include <linux/string.h>
18 #include <linux/net.h>
19 #include <linux/netdevice.h>
20 #include <linux/ipv6.h>
21 #include <linux/slab.h>
23 #include <net/ipv6_frag.h>
25 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
26 #include <linux/sysctl.h>
27 #include <linux/netfilter.h>
28 #include <linux/netfilter_ipv6.h>
29 #include <linux/kernel.h>
30 #include <linux/module.h>
31 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
32 #include <net/netns/generic.h>
34 static const char nf_frags_cache_name[] = "nf-frags";
36 static unsigned int nf_frag_pernet_id __read_mostly;
37 static struct inet_frags nf_frags;
39 static struct nft_ct_frag6_pernet *nf_frag_pernet(struct net *net)
41 return net_generic(net, nf_frag_pernet_id);
46 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
48 .procname = "nf_conntrack_frag6_timeout",
49 .maxlen = sizeof(unsigned int),
51 .proc_handler = proc_dointvec_jiffies,
54 .procname = "nf_conntrack_frag6_low_thresh",
55 .maxlen = sizeof(unsigned long),
57 .proc_handler = proc_doulongvec_minmax,
60 .procname = "nf_conntrack_frag6_high_thresh",
61 .maxlen = sizeof(unsigned long),
63 .proc_handler = proc_doulongvec_minmax,
68 static int nf_ct_frag6_sysctl_register(struct net *net)
70 struct nft_ct_frag6_pernet *nf_frag;
71 struct ctl_table *table;
72 struct ctl_table_header *hdr;
74 table = nf_ct_frag6_sysctl_table;
75 if (!net_eq(net, &init_net)) {
76 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
82 nf_frag = nf_frag_pernet(net);
84 table[0].data = &nf_frag->fqdir->timeout;
85 table[1].data = &nf_frag->fqdir->low_thresh;
86 table[1].extra2 = &nf_frag->fqdir->high_thresh;
87 table[2].data = &nf_frag->fqdir->high_thresh;
88 table[2].extra1 = &nf_frag->fqdir->low_thresh;
90 hdr = register_net_sysctl(net, "net/netfilter", table);
94 nf_frag->nf_frag_frags_hdr = hdr;
98 if (!net_eq(net, &init_net))
104 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
106 struct nft_ct_frag6_pernet *nf_frag = nf_frag_pernet(net);
107 struct ctl_table *table;
109 table = nf_frag->nf_frag_frags_hdr->ctl_table_arg;
110 unregister_net_sysctl_table(nf_frag->nf_frag_frags_hdr);
111 if (!net_eq(net, &init_net))
116 static int nf_ct_frag6_sysctl_register(struct net *net)
120 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
125 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
126 struct sk_buff *prev_tail, struct net_device *dev);
128 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
130 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
133 static void nf_ct_frag6_expire(struct timer_list *t)
135 struct inet_frag_queue *frag = from_timer(frag, t, timer);
136 struct frag_queue *fq;
138 fq = container_of(frag, struct frag_queue, q);
140 ip6frag_expire_frag_queue(fq->q.fqdir->net, fq);
143 /* Creation primitives. */
144 static struct frag_queue *fq_find(struct net *net, __be32 id, u32 user,
145 const struct ipv6hdr *hdr, int iif)
147 struct nft_ct_frag6_pernet *nf_frag = nf_frag_pernet(net);
148 struct frag_v6_compare_key key = {
155 struct inet_frag_queue *q;
157 q = inet_frag_find(nf_frag->fqdir, &key);
161 return container_of(q, struct frag_queue, q);
165 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
166 const struct frag_hdr *fhdr, int nhoff)
168 unsigned int payload_len;
169 struct net_device *dev;
170 struct sk_buff *prev;
171 int offset, end, err;
174 if (fq->q.flags & INET_FRAG_COMPLETE) {
175 pr_debug("Already completed\n");
179 payload_len = ntohs(ipv6_hdr(skb)->payload_len);
181 offset = ntohs(fhdr->frag_off) & ~0x7;
182 end = offset + (payload_len -
183 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
185 if ((unsigned int)end > IPV6_MAXPLEN) {
186 pr_debug("offset is too large.\n");
190 ecn = ip6_frag_ecn(ipv6_hdr(skb));
192 if (skb->ip_summed == CHECKSUM_COMPLETE) {
193 const unsigned char *nh = skb_network_header(skb);
194 skb->csum = csum_sub(skb->csum,
195 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
199 /* Is this the final fragment? */
200 if (!(fhdr->frag_off & htons(IP6_MF))) {
201 /* If we already have some bits beyond end
202 * or have different end, the segment is corrupted.
204 if (end < fq->q.len ||
205 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
206 pr_debug("already received last fragment\n");
209 fq->q.flags |= INET_FRAG_LAST_IN;
212 /* Check if the fragment is rounded to 8 bytes.
213 * Required by the RFC.
216 /* RFC2460 says always send parameter problem in
219 pr_debug("end of fragment not rounded to 8 bytes.\n");
220 inet_frag_kill(&fq->q);
223 if (end > fq->q.len) {
224 /* Some bits beyond end -> corruption. */
225 if (fq->q.flags & INET_FRAG_LAST_IN) {
226 pr_debug("last packet already reached.\n");
236 /* Point into the IP datagram 'data' part. */
237 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
238 pr_debug("queue: message is too short.\n");
241 if (pskb_trim_rcsum(skb, end - offset)) {
242 pr_debug("Can't trim\n");
246 /* Note : skb->rbnode and skb->dev share the same location. */
248 /* Makes sure compiler wont do silly aliasing games */
251 prev = fq->q.fragments_tail;
252 err = inet_frag_queue_insert(&fq->q, skb, offset, end);
254 if (err == IPFRAG_DUP) {
255 /* No error for duplicates, pretend they got queued. */
263 fq->iif = dev->ifindex;
265 fq->q.stamp = skb->tstamp;
266 fq->q.mono_delivery_time = skb->mono_delivery_time;
267 fq->q.meat += skb->len;
269 if (payload_len > fq->q.max_size)
270 fq->q.max_size = payload_len;
271 add_frag_mem_limit(fq->q.fqdir, skb->truesize);
273 /* The first fragment.
274 * nhoffset is obtained from the first fragment, of course.
277 fq->nhoffset = nhoff;
278 fq->q.flags |= INET_FRAG_FIRST_IN;
281 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
282 fq->q.meat == fq->q.len) {
283 unsigned long orefdst = skb->_skb_refdst;
285 skb->_skb_refdst = 0UL;
286 err = nf_ct_frag6_reasm(fq, skb, prev, dev);
287 skb->_skb_refdst = orefdst;
289 /* After queue has assumed skb ownership, only 0 or
290 * -EINPROGRESS must be returned.
292 return err ? -EINPROGRESS : 0;
300 inet_frag_kill(&fq->q);
307 * Check if this packet is complete.
309 * It is called with locked fq, and caller must check that
310 * queue is eligible for reassembly i.e. it is not COMPLETE,
311 * the last and the first frames arrived and all the bits are here.
313 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
314 struct sk_buff *prev_tail, struct net_device *dev)
320 inet_frag_kill(&fq->q);
322 ecn = ip_frag_ecn_table[fq->ecn];
323 if (unlikely(ecn == 0xff))
326 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
330 payload_len = ((skb->data - skb_network_header(skb)) -
331 sizeof(struct ipv6hdr) + fq->q.len -
332 sizeof(struct frag_hdr));
333 if (payload_len > IPV6_MAXPLEN) {
334 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
339 /* We have to remove fragment header from datagram and to relocate
340 * header in order to calculate ICV correctly. */
341 skb_network_header(skb)[fq->nhoffset] = skb_transport_header(skb)[0];
342 memmove(skb->head + sizeof(struct frag_hdr), skb->head,
343 (skb->data - skb->head) - sizeof(struct frag_hdr));
344 skb->mac_header += sizeof(struct frag_hdr);
345 skb->network_header += sizeof(struct frag_hdr);
347 skb_reset_transport_header(skb);
349 inet_frag_reasm_finish(&fq->q, skb, reasm_data, false);
353 ipv6_hdr(skb)->payload_len = htons(payload_len);
354 ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
355 IP6CB(skb)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
356 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
358 /* Yes, and fold redundant checksum back. 8) */
359 if (skb->ip_summed == CHECKSUM_COMPLETE)
360 skb->csum = csum_partial(skb_network_header(skb),
361 skb_network_header_len(skb),
364 fq->q.rb_fragments = RB_ROOT;
365 fq->q.fragments_tail = NULL;
366 fq->q.last_run_head = NULL;
371 inet_frag_kill(&fq->q);
376 * find the header just before Fragment Header.
378 * if success return 0 and set ...
379 * (*prevhdrp): the value of "Next Header Field" in the header
380 * just before Fragment Header.
381 * (*prevhoff): the offset of "Next Header Field" in the header
382 * just before Fragment Header.
383 * (*fhoff) : the offset of Fragment Header.
385 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
389 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
391 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
392 const int netoff = skb_network_offset(skb);
393 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
394 int start = netoff + sizeof(struct ipv6hdr);
395 int len = skb->len - start;
396 u8 prevhdr = NEXTHDR_IPV6;
398 while (nexthdr != NEXTHDR_FRAGMENT) {
399 struct ipv6_opt_hdr hdr;
402 if (!ipv6_ext_hdr(nexthdr)) {
405 if (nexthdr == NEXTHDR_NONE) {
406 pr_debug("next header is none\n");
409 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
410 pr_debug("too short\n");
413 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
415 if (nexthdr == NEXTHDR_AUTH)
416 hdrlen = ipv6_authlen(&hdr);
418 hdrlen = ipv6_optlen(&hdr);
423 nexthdr = hdr.nexthdr;
432 *prevhoff = prev_nhoff;
438 int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
440 u16 savethdr = skb->transport_header;
441 u8 nexthdr = NEXTHDR_FRAGMENT;
442 int fhoff, nhoff, ret;
443 struct frag_hdr *fhdr;
444 struct frag_queue *fq;
448 /* Jumbo payload inhibits frag. header */
449 if (ipv6_hdr(skb)->payload_len == 0) {
450 pr_debug("payload len = 0\n");
454 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
457 /* Discard the first fragment if it does not include all headers
458 * RFC 8200, Section 4.5
460 if (ipv6frag_thdr_truncated(skb, fhoff, &nexthdr)) {
461 pr_debug("Drop incomplete fragment\n");
465 if (!pskb_may_pull(skb, fhoff + sizeof(*fhdr)))
468 skb_set_transport_header(skb, fhoff);
470 fhdr = (struct frag_hdr *)skb_transport_header(skb);
472 fq = fq_find(net, fhdr->identification, user, hdr,
473 skb->dev ? skb->dev->ifindex : 0);
475 pr_debug("Can't find and can't create new queue\n");
479 spin_lock_bh(&fq->q.lock);
481 ret = nf_ct_frag6_queue(fq, skb, fhdr, nhoff);
482 if (ret == -EPROTO) {
483 skb->transport_header = savethdr;
487 spin_unlock_bh(&fq->q.lock);
488 inet_frag_put(&fq->q);
491 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
493 static int nf_ct_net_init(struct net *net)
495 struct nft_ct_frag6_pernet *nf_frag = nf_frag_pernet(net);
498 res = fqdir_init(&nf_frag->fqdir, &nf_frags, net);
502 nf_frag->fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
503 nf_frag->fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
504 nf_frag->fqdir->timeout = IPV6_FRAG_TIMEOUT;
506 res = nf_ct_frag6_sysctl_register(net);
508 fqdir_exit(nf_frag->fqdir);
512 static void nf_ct_net_pre_exit(struct net *net)
514 struct nft_ct_frag6_pernet *nf_frag = nf_frag_pernet(net);
516 fqdir_pre_exit(nf_frag->fqdir);
519 static void nf_ct_net_exit(struct net *net)
521 struct nft_ct_frag6_pernet *nf_frag = nf_frag_pernet(net);
523 nf_ct_frags6_sysctl_unregister(net);
524 fqdir_exit(nf_frag->fqdir);
527 static struct pernet_operations nf_ct_net_ops = {
528 .init = nf_ct_net_init,
529 .pre_exit = nf_ct_net_pre_exit,
530 .exit = nf_ct_net_exit,
531 .id = &nf_frag_pernet_id,
532 .size = sizeof(struct nft_ct_frag6_pernet),
535 static const struct rhashtable_params nfct_rhash_params = {
536 .head_offset = offsetof(struct inet_frag_queue, node),
537 .hashfn = ip6frag_key_hashfn,
538 .obj_hashfn = ip6frag_obj_hashfn,
539 .obj_cmpfn = ip6frag_obj_cmpfn,
540 .automatic_shrinking = true,
543 int nf_ct_frag6_init(void)
547 nf_frags.constructor = ip6frag_init;
548 nf_frags.destructor = NULL;
549 nf_frags.qsize = sizeof(struct frag_queue);
550 nf_frags.frag_expire = nf_ct_frag6_expire;
551 nf_frags.frags_cache_name = nf_frags_cache_name;
552 nf_frags.rhash_params = nfct_rhash_params;
553 ret = inet_frags_init(&nf_frags);
556 ret = register_pernet_subsys(&nf_ct_net_ops);
558 inet_frags_fini(&nf_frags);
564 void nf_ct_frag6_cleanup(void)
566 unregister_pernet_subsys(&nf_ct_net_ops);
567 inet_frags_fini(&nf_frags);