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/socket.h>
19 #include <linux/sockios.h>
20 #include <linux/jiffies.h>
21 #include <linux/net.h>
22 #include <linux/list.h>
23 #include <linux/netdevice.h>
24 #include <linux/in6.h>
25 #include <linux/ipv6.h>
26 #include <linux/icmpv6.h>
27 #include <linux/random.h>
28 #include <linux/slab.h>
32 #include <net/ipv6_frag.h>
34 #include <net/protocol.h>
35 #include <net/transp_v6.h>
36 #include <net/rawv6.h>
37 #include <net/ndisc.h>
38 #include <net/addrconf.h>
39 #include <net/inet_ecn.h>
40 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
41 #include <linux/sysctl.h>
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
44 #include <linux/kernel.h>
45 #include <linux/module.h>
46 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
48 static const char nf_frags_cache_name[] = "nf-frags";
50 static struct inet_frags nf_frags;
54 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
56 .procname = "nf_conntrack_frag6_timeout",
57 .data = &init_net.nf_frag.frags.timeout,
58 .maxlen = sizeof(unsigned int),
60 .proc_handler = proc_dointvec_jiffies,
63 .procname = "nf_conntrack_frag6_low_thresh",
64 .data = &init_net.nf_frag.frags.low_thresh,
65 .maxlen = sizeof(unsigned long),
67 .proc_handler = proc_doulongvec_minmax,
68 .extra2 = &init_net.nf_frag.frags.high_thresh
71 .procname = "nf_conntrack_frag6_high_thresh",
72 .data = &init_net.nf_frag.frags.high_thresh,
73 .maxlen = sizeof(unsigned long),
75 .proc_handler = proc_doulongvec_minmax,
76 .extra1 = &init_net.nf_frag.frags.low_thresh
81 static int nf_ct_frag6_sysctl_register(struct net *net)
83 struct ctl_table *table;
84 struct ctl_table_header *hdr;
86 table = nf_ct_frag6_sysctl_table;
87 if (!net_eq(net, &init_net)) {
88 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
93 table[0].data = &net->nf_frag.frags.timeout;
94 table[1].data = &net->nf_frag.frags.low_thresh;
95 table[1].extra2 = &net->nf_frag.frags.high_thresh;
96 table[2].data = &net->nf_frag.frags.high_thresh;
97 table[2].extra1 = &net->nf_frag.frags.low_thresh;
98 table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
101 hdr = register_net_sysctl(net, "net/netfilter", table);
105 net->nf_frag_frags_hdr = hdr;
109 if (!net_eq(net, &init_net))
115 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
117 struct ctl_table *table;
119 table = net->nf_frag_frags_hdr->ctl_table_arg;
120 unregister_net_sysctl_table(net->nf_frag_frags_hdr);
121 if (!net_eq(net, &init_net))
126 static int nf_ct_frag6_sysctl_register(struct net *net)
130 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
135 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
136 struct sk_buff *prev_tail, struct net_device *dev);
138 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
140 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
143 static void nf_ct_frag6_expire(struct timer_list *t)
145 struct inet_frag_queue *frag = from_timer(frag, t, timer);
146 struct frag_queue *fq;
149 fq = container_of(frag, struct frag_queue, q);
150 net = container_of(fq->q.net, struct net, nf_frag.frags);
152 ip6frag_expire_frag_queue(net, fq);
155 /* Creation primitives. */
156 static struct frag_queue *fq_find(struct net *net, __be32 id, u32 user,
157 const struct ipv6hdr *hdr, int iif)
159 struct frag_v6_compare_key key = {
166 struct inet_frag_queue *q;
168 q = inet_frag_find(&net->nf_frag.frags, &key);
172 return container_of(q, struct frag_queue, q);
176 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
177 const struct frag_hdr *fhdr, int nhoff)
179 unsigned int payload_len;
180 struct net_device *dev;
181 struct sk_buff *prev;
182 int offset, end, err;
185 if (fq->q.flags & INET_FRAG_COMPLETE) {
186 pr_debug("Already completed\n");
190 payload_len = ntohs(ipv6_hdr(skb)->payload_len);
192 offset = ntohs(fhdr->frag_off) & ~0x7;
193 end = offset + (payload_len -
194 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
196 if ((unsigned int)end > IPV6_MAXPLEN) {
197 pr_debug("offset is too large.\n");
201 ecn = ip6_frag_ecn(ipv6_hdr(skb));
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,
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.
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");
220 fq->q.flags |= INET_FRAG_LAST_IN;
223 /* Check if the fragment is rounded to 8 bytes.
224 * Required by the RFC.
227 /* RFC2460 says always send parameter problem in
230 pr_debug("end of fragment not rounded to 8 bytes.\n");
231 inet_frag_kill(&fq->q);
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");
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");
252 if (pskb_trim_rcsum(skb, end - offset)) {
253 pr_debug("Can't trim\n");
257 /* Note : skb->rbnode and skb->dev share the same location. */
259 /* Makes sure compiler wont do silly aliasing games */
262 prev = fq->q.fragments_tail;
263 err = inet_frag_queue_insert(&fq->q, skb, offset, end);
265 if (err == IPFRAG_DUP) {
266 /* No error for duplicates, pretend they got queued. */
274 fq->iif = dev->ifindex;
276 fq->q.stamp = skb->tstamp;
277 fq->q.meat += skb->len;
279 if (payload_len > fq->q.max_size)
280 fq->q.max_size = payload_len;
281 add_frag_mem_limit(fq->q.net, skb->truesize);
283 /* The first fragment.
284 * nhoffset is obtained from the first fragment, of course.
287 fq->nhoffset = nhoff;
288 fq->q.flags |= INET_FRAG_FIRST_IN;
291 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
292 fq->q.meat == fq->q.len) {
293 unsigned long orefdst = skb->_skb_refdst;
295 skb->_skb_refdst = 0UL;
296 err = nf_ct_frag6_reasm(fq, skb, prev, dev);
297 skb->_skb_refdst = orefdst;
299 /* After queue has assumed skb ownership, only 0 or
300 * -EINPROGRESS must be returned.
302 return err ? -EINPROGRESS : 0;
309 inet_frag_kill(&fq->q);
316 * Check if this packet is complete.
318 * It is called with locked fq, and caller must check that
319 * queue is eligible for reassembly i.e. it is not COMPLETE,
320 * the last and the first frames arrived and all the bits are here.
322 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
323 struct sk_buff *prev_tail, struct net_device *dev)
329 inet_frag_kill(&fq->q);
331 ecn = ip_frag_ecn_table[fq->ecn];
332 if (unlikely(ecn == 0xff))
335 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
339 payload_len = ((skb->data - skb_network_header(skb)) -
340 sizeof(struct ipv6hdr) + fq->q.len -
341 sizeof(struct frag_hdr));
342 if (payload_len > IPV6_MAXPLEN) {
343 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
348 /* We have to remove fragment header from datagram and to relocate
349 * header in order to calculate ICV correctly. */
350 skb_network_header(skb)[fq->nhoffset] = skb_transport_header(skb)[0];
351 memmove(skb->head + sizeof(struct frag_hdr), skb->head,
352 (skb->data - skb->head) - sizeof(struct frag_hdr));
353 skb->mac_header += sizeof(struct frag_hdr);
354 skb->network_header += sizeof(struct frag_hdr);
356 skb_reset_transport_header(skb);
358 inet_frag_reasm_finish(&fq->q, skb, reasm_data);
362 ipv6_hdr(skb)->payload_len = htons(payload_len);
363 ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
364 IP6CB(skb)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
366 /* Yes, and fold redundant checksum back. 8) */
367 if (skb->ip_summed == CHECKSUM_COMPLETE)
368 skb->csum = csum_partial(skb_network_header(skb),
369 skb_network_header_len(skb),
372 fq->q.rb_fragments = RB_ROOT;
373 fq->q.fragments_tail = NULL;
374 fq->q.last_run_head = NULL;
379 inet_frag_kill(&fq->q);
384 * find the header just before Fragment Header.
386 * if success return 0 and set ...
387 * (*prevhdrp): the value of "Next Header Field" in the header
388 * just before Fragment Header.
389 * (*prevhoff): the offset of "Next Header Field" in the header
390 * just before Fragment Header.
391 * (*fhoff) : the offset of Fragment Header.
393 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
397 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
399 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
400 const int netoff = skb_network_offset(skb);
401 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
402 int start = netoff + sizeof(struct ipv6hdr);
403 int len = skb->len - start;
404 u8 prevhdr = NEXTHDR_IPV6;
406 while (nexthdr != NEXTHDR_FRAGMENT) {
407 struct ipv6_opt_hdr hdr;
410 if (!ipv6_ext_hdr(nexthdr)) {
413 if (nexthdr == NEXTHDR_NONE) {
414 pr_debug("next header is none\n");
417 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
418 pr_debug("too short\n");
421 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
423 if (nexthdr == NEXTHDR_AUTH)
424 hdrlen = (hdr.hdrlen+2)<<2;
426 hdrlen = ipv6_optlen(&hdr);
431 nexthdr = hdr.nexthdr;
440 *prevhoff = prev_nhoff;
446 int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
448 u16 savethdr = skb->transport_header;
449 int fhoff, nhoff, ret;
450 struct frag_hdr *fhdr;
451 struct frag_queue *fq;
455 /* Jumbo payload inhibits frag. header */
456 if (ipv6_hdr(skb)->payload_len == 0) {
457 pr_debug("payload len = 0\n");
461 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
464 if (!pskb_may_pull(skb, fhoff + sizeof(*fhdr)))
467 skb_set_transport_header(skb, fhoff);
469 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)
497 net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
498 net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
499 net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
500 net->nf_frag.frags.f = &nf_frags;
502 res = inet_frags_init_net(&net->nf_frag.frags);
505 res = nf_ct_frag6_sysctl_register(net);
507 inet_frags_exit_net(&net->nf_frag.frags);
511 static void nf_ct_net_exit(struct net *net)
513 nf_ct_frags6_sysctl_unregister(net);
514 inet_frags_exit_net(&net->nf_frag.frags);
517 static struct pernet_operations nf_ct_net_ops = {
518 .init = nf_ct_net_init,
519 .exit = nf_ct_net_exit,
522 static const struct rhashtable_params nfct_rhash_params = {
523 .head_offset = offsetof(struct inet_frag_queue, node),
524 .hashfn = ip6frag_key_hashfn,
525 .obj_hashfn = ip6frag_obj_hashfn,
526 .obj_cmpfn = ip6frag_obj_cmpfn,
527 .automatic_shrinking = true,
530 int nf_ct_frag6_init(void)
534 nf_frags.constructor = ip6frag_init;
535 nf_frags.destructor = NULL;
536 nf_frags.qsize = sizeof(struct frag_queue);
537 nf_frags.frag_expire = nf_ct_frag6_expire;
538 nf_frags.frags_cache_name = nf_frags_cache_name;
539 nf_frags.rhash_params = nfct_rhash_params;
540 ret = inet_frags_init(&nf_frags);
543 ret = register_pernet_subsys(&nf_ct_net_ops);
545 inet_frags_fini(&nf_frags);
551 void nf_ct_frag6_cleanup(void)
553 unregister_pernet_subsys(&nf_ct_net_ops);
554 inet_frags_fini(&nf_frags);