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 #include <linux/errno.h>
18 #include <linux/types.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/jiffies.h>
23 #include <linux/net.h>
24 #include <linux/list.h>
25 #include <linux/netdevice.h>
26 #include <linux/in6.h>
27 #include <linux/ipv6.h>
28 #include <linux/icmpv6.h>
29 #include <linux/random.h>
33 #include <net/inet_frag.h>
36 #include <net/protocol.h>
37 #include <net/transp_v6.h>
38 #include <net/rawv6.h>
39 #include <net/ndisc.h>
40 #include <net/addrconf.h>
41 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
42 #include <linux/sysctl.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45 #include <linux/kernel.h>
46 #include <linux/module.h>
49 struct nf_ct_frag6_skb_cb
51 struct inet6_skb_parm h;
56 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
58 struct nf_ct_frag6_queue
60 struct inet_frag_queue q;
62 __be32 id; /* fragment id */
64 struct in6_addr saddr;
65 struct in6_addr daddr;
71 static struct inet_frags nf_frags;
72 static struct netns_frags nf_init_frags;
75 struct ctl_table nf_ct_ipv6_sysctl_table[] = {
77 .procname = "nf_conntrack_frag6_timeout",
78 .data = &nf_init_frags.timeout,
79 .maxlen = sizeof(unsigned int),
81 .proc_handler = proc_dointvec_jiffies,
84 .procname = "nf_conntrack_frag6_low_thresh",
85 .data = &nf_init_frags.low_thresh,
86 .maxlen = sizeof(unsigned int),
88 .proc_handler = proc_dointvec,
91 .procname = "nf_conntrack_frag6_high_thresh",
92 .data = &nf_init_frags.high_thresh,
93 .maxlen = sizeof(unsigned int),
95 .proc_handler = proc_dointvec,
101 static unsigned int nf_hashfn(struct inet_frag_queue *q)
103 const struct nf_ct_frag6_queue *nq;
105 nq = container_of(q, struct nf_ct_frag6_queue, q);
106 return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
109 static void nf_skb_free(struct sk_buff *skb)
111 if (NFCT_FRAG6_CB(skb)->orig)
112 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
115 /* Memory Tracking Functions. */
116 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
119 *work -= skb->truesize;
120 atomic_sub(skb->truesize, &nf_init_frags.mem);
125 /* Destruction primitives. */
127 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
129 inet_frag_put(&fq->q, &nf_frags);
132 /* Kill fq entry. It is not destroyed immediately,
133 * because caller (and someone more) holds reference count.
135 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
137 inet_frag_kill(&fq->q, &nf_frags);
140 static void nf_ct_frag6_evictor(void)
143 inet_frag_evictor(&nf_init_frags, &nf_frags);
147 static void nf_ct_frag6_expire(unsigned long data)
149 struct nf_ct_frag6_queue *fq;
151 fq = container_of((struct inet_frag_queue *)data,
152 struct nf_ct_frag6_queue, q);
154 spin_lock(&fq->q.lock);
156 if (fq->q.last_in & INET_FRAG_COMPLETE)
162 spin_unlock(&fq->q.lock);
166 /* Creation primitives. */
168 static __inline__ struct nf_ct_frag6_queue *
169 fq_find(__be32 id, u32 user, struct in6_addr *src, struct in6_addr *dst)
171 struct inet_frag_queue *q;
172 struct ip6_create_arg arg;
180 read_lock_bh(&nf_frags.lock);
181 hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
183 q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash);
188 return container_of(q, struct nf_ct_frag6_queue, q);
191 pr_debug("Can't alloc new queue\n");
196 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
197 const struct frag_hdr *fhdr, int nhoff)
199 struct sk_buff *prev, *next;
202 if (fq->q.last_in & INET_FRAG_COMPLETE) {
203 pr_debug("Allready completed\n");
207 offset = ntohs(fhdr->frag_off) & ~0x7;
208 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
209 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
211 if ((unsigned int)end > IPV6_MAXPLEN) {
212 pr_debug("offset is too large.\n");
216 if (skb->ip_summed == CHECKSUM_COMPLETE) {
217 const unsigned char *nh = skb_network_header(skb);
218 skb->csum = csum_sub(skb->csum,
219 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
223 /* Is this the final fragment? */
224 if (!(fhdr->frag_off & htons(IP6_MF))) {
225 /* If we already have some bits beyond end
226 * or have different end, the segment is corrupted.
228 if (end < fq->q.len ||
229 ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
230 pr_debug("already received last fragment\n");
233 fq->q.last_in |= INET_FRAG_LAST_IN;
236 /* Check if the fragment is rounded to 8 bytes.
237 * Required by the RFC.
240 /* RFC2460 says always send parameter problem in
243 pr_debug("end of fragment not rounded to 8 bytes.\n");
246 if (end > fq->q.len) {
247 /* Some bits beyond end -> corruption. */
248 if (fq->q.last_in & INET_FRAG_LAST_IN) {
249 pr_debug("last packet already reached.\n");
259 /* Point into the IP datagram 'data' part. */
260 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
261 pr_debug("queue: message is too short.\n");
264 if (pskb_trim_rcsum(skb, end - offset)) {
265 pr_debug("Can't trim\n");
269 /* Find out which fragments are in front and at the back of us
270 * in the chain of fragments so far. We must know where to put
271 * this fragment, right?
274 for (next = fq->q.fragments; next != NULL; next = next->next) {
275 if (NFCT_FRAG6_CB(next)->offset >= offset)
280 /* We found where to put this one. Check for overlap with
281 * preceding fragment, and, if needed, align things so that
282 * any overlaps are eliminated.
285 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
290 pr_debug("overlap\n");
293 if (!pskb_pull(skb, i)) {
294 pr_debug("Can't pull\n");
297 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
298 skb->ip_summed = CHECKSUM_NONE;
302 /* Look for overlap with succeeding segments.
303 * If we can merge fragments, do it.
305 while (next && NFCT_FRAG6_CB(next)->offset < end) {
306 /* overlap is 'i' bytes */
307 int i = end - NFCT_FRAG6_CB(next)->offset;
310 /* Eat head of the next overlapped fragment
311 * and leave the loop. The next ones cannot overlap.
313 pr_debug("Eat head of the overlapped parts.: %d", i);
314 if (!pskb_pull(next, i))
318 NFCT_FRAG6_CB(next)->offset += i;
320 if (next->ip_summed != CHECKSUM_UNNECESSARY)
321 next->ip_summed = CHECKSUM_NONE;
324 struct sk_buff *free_it = next;
326 /* Old fragmnet is completely overridden with
334 fq->q.fragments = next;
336 fq->q.meat -= free_it->len;
337 frag_kfree_skb(free_it, NULL);
341 NFCT_FRAG6_CB(skb)->offset = offset;
343 /* Insert this fragment in the chain of fragments. */
348 fq->q.fragments = skb;
351 fq->q.stamp = skb->tstamp;
352 fq->q.meat += skb->len;
353 atomic_add(skb->truesize, &nf_init_frags.mem);
355 /* The first fragment.
356 * nhoffset is obtained from the first fragment, of course.
359 fq->nhoffset = nhoff;
360 fq->q.last_in |= INET_FRAG_FIRST_IN;
362 write_lock(&nf_frags.lock);
363 list_move_tail(&fq->q.lru_list, &nf_init_frags.lru_list);
364 write_unlock(&nf_frags.lock);
372 * Check if this packet is complete.
373 * Returns NULL on failure by any reason, and pointer
374 * to current nexthdr field in reassembled frame.
376 * It is called with locked fq, and caller must check that
377 * queue is eligible for reassembly i.e. it is not COMPLETE,
378 * the last and the first frames arrived and all the bits are here.
380 static struct sk_buff *
381 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
383 struct sk_buff *fp, *op, *head = fq->q.fragments;
388 WARN_ON(head == NULL);
389 WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
391 /* Unfragmented part is taken from the first segment. */
392 payload_len = ((head->data - skb_network_header(head)) -
393 sizeof(struct ipv6hdr) + fq->q.len -
394 sizeof(struct frag_hdr));
395 if (payload_len > IPV6_MAXPLEN) {
396 pr_debug("payload len is too large.\n");
400 /* Head of list must not be cloned. */
401 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
402 pr_debug("skb is cloned but can't expand head");
406 /* If the first fragment is fragmented itself, we split
407 * it to two chunks: the first with data and paged part
408 * and the second, holding only fragments. */
409 if (skb_has_frags(head)) {
410 struct sk_buff *clone;
413 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
414 pr_debug("Can't alloc skb\n");
417 clone->next = head->next;
419 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
420 skb_frag_list_init(head);
421 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
422 plen += skb_shinfo(head)->frags[i].size;
423 clone->len = clone->data_len = head->data_len - plen;
424 head->data_len -= clone->len;
425 head->len -= clone->len;
427 clone->ip_summed = head->ip_summed;
429 NFCT_FRAG6_CB(clone)->orig = NULL;
430 atomic_add(clone->truesize, &nf_init_frags.mem);
433 /* We have to remove fragment header from datagram and to relocate
434 * header in order to calculate ICV correctly. */
435 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
436 memmove(head->head + sizeof(struct frag_hdr), head->head,
437 (head->data - head->head) - sizeof(struct frag_hdr));
438 head->mac_header += sizeof(struct frag_hdr);
439 head->network_header += sizeof(struct frag_hdr);
441 skb_shinfo(head)->frag_list = head->next;
442 skb_reset_transport_header(head);
443 skb_push(head, head->data - skb_network_header(head));
444 atomic_sub(head->truesize, &nf_init_frags.mem);
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 atomic_sub(fp->truesize, &nf_init_frags.mem);
459 head->tstamp = fq->q.stamp;
460 ipv6_hdr(head)->payload_len = htons(payload_len);
462 /* Yes, and fold redundant checksum back. 8) */
463 if (head->ip_summed == CHECKSUM_COMPLETE)
464 head->csum = csum_partial(skb_network_header(head),
465 skb_network_header_len(head),
468 fq->q.fragments = NULL;
470 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
471 fp = skb_shinfo(head)->frag_list;
472 if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
473 /* at above code, head skb is divided into two skbs. */
476 op = NFCT_FRAG6_CB(head)->orig;
477 for (; fp; fp = fp->next) {
478 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
482 NFCT_FRAG6_CB(fp)->orig = NULL;
489 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
493 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
499 * find the header just before Fragment Header.
501 * if success return 0 and set ...
502 * (*prevhdrp): the value of "Next Header Field" in the header
503 * just before Fragment Header.
504 * (*prevhoff): the offset of "Next Header Field" in the header
505 * just before Fragment Header.
506 * (*fhoff) : the offset of Fragment Header.
508 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
512 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
514 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
515 const int netoff = skb_network_offset(skb);
516 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
517 int start = netoff + sizeof(struct ipv6hdr);
518 int len = skb->len - start;
519 u8 prevhdr = NEXTHDR_IPV6;
521 while (nexthdr != NEXTHDR_FRAGMENT) {
522 struct ipv6_opt_hdr hdr;
525 if (!ipv6_ext_hdr(nexthdr)) {
528 if (nexthdr == NEXTHDR_NONE) {
529 pr_debug("next header is none\n");
532 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
533 pr_debug("too short\n");
536 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
538 if (nexthdr == NEXTHDR_AUTH)
539 hdrlen = (hdr.hdrlen+2)<<2;
541 hdrlen = ipv6_optlen(&hdr);
546 nexthdr = hdr.nexthdr;
555 *prevhoff = prev_nhoff;
561 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
563 struct sk_buff *clone;
564 struct net_device *dev = skb->dev;
565 struct frag_hdr *fhdr;
566 struct nf_ct_frag6_queue *fq;
570 struct sk_buff *ret_skb = NULL;
572 /* Jumbo payload inhibits frag. header */
573 if (ipv6_hdr(skb)->payload_len == 0) {
574 pr_debug("payload len = 0\n");
578 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
581 clone = skb_clone(skb, GFP_ATOMIC);
583 pr_debug("Can't clone skb\n");
587 NFCT_FRAG6_CB(clone)->orig = skb;
589 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
590 pr_debug("message is too short.\n");
594 skb_set_transport_header(clone, fhoff);
595 hdr = ipv6_hdr(clone);
596 fhdr = (struct frag_hdr *)skb_transport_header(clone);
598 if (atomic_read(&nf_init_frags.mem) > nf_init_frags.high_thresh)
599 nf_ct_frag6_evictor();
601 fq = fq_find(fhdr->identification, user, &hdr->saddr, &hdr->daddr);
603 pr_debug("Can't find and can't create new queue\n");
607 spin_lock_bh(&fq->q.lock);
609 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
610 spin_unlock_bh(&fq->q.lock);
611 pr_debug("Can't insert skb to queue\n");
616 if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
617 fq->q.meat == fq->q.len) {
618 ret_skb = nf_ct_frag6_reasm(fq, dev);
620 pr_debug("Can't reassemble fragmented packets\n");
622 spin_unlock_bh(&fq->q.lock);
632 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
633 struct net_device *in, struct net_device *out,
634 int (*okfn)(struct sk_buff *))
636 struct sk_buff *s, *s2;
638 for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
639 nf_conntrack_put_reasm(s->nfct_reasm);
640 nf_conntrack_get_reasm(skb);
646 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
647 NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
650 nf_conntrack_put_reasm(skb);
653 int nf_ct_frag6_init(void)
655 nf_frags.hashfn = nf_hashfn;
656 nf_frags.constructor = ip6_frag_init;
657 nf_frags.destructor = NULL;
658 nf_frags.skb_free = nf_skb_free;
659 nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
660 nf_frags.match = ip6_frag_match;
661 nf_frags.frag_expire = nf_ct_frag6_expire;
662 nf_frags.secret_interval = 10 * 60 * HZ;
663 nf_init_frags.timeout = IPV6_FRAG_TIMEOUT;
664 nf_init_frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
665 nf_init_frags.low_thresh = IPV6_FRAG_LOW_THRESH;
666 inet_frags_init_net(&nf_init_frags);
667 inet_frags_init(&nf_frags);
672 void nf_ct_frag6_cleanup(void)
674 inet_frags_fini(&nf_frags);
676 nf_init_frags.low_thresh = 0;
677 nf_ct_frag6_evictor();