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>
30 #include <linux/jhash.h>
34 #include <net/inet_frag.h>
37 #include <net/protocol.h>
38 #include <net/transp_v6.h>
39 #include <net/rawv6.h>
40 #include <net/ndisc.h>
41 #include <net/addrconf.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>
48 #define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */
49 #define NF_CT_FRAG6_LOW_THRESH 196608 /* == 192*1024 */
50 #define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT
52 struct nf_ct_frag6_skb_cb
54 struct inet6_skb_parm h;
59 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
61 struct nf_ct_frag6_queue
63 struct inet_frag_queue q;
65 __be32 id; /* fragment id */
66 struct in6_addr saddr;
67 struct in6_addr daddr;
73 struct inet_frags_ctl nf_frags_ctl __read_mostly = {
74 .high_thresh = 256 * 1024,
75 .low_thresh = 192 * 1024,
76 .timeout = IPV6_FRAG_TIMEOUT,
77 .secret_interval = 10 * 60 * HZ,
80 static struct inet_frags nf_frags;
82 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
83 struct in6_addr *daddr)
87 a = (__force u32)saddr->s6_addr32[0];
88 b = (__force u32)saddr->s6_addr32[1];
89 c = (__force u32)saddr->s6_addr32[2];
91 a += JHASH_GOLDEN_RATIO;
92 b += JHASH_GOLDEN_RATIO;
96 a += (__force u32)saddr->s6_addr32[3];
97 b += (__force u32)daddr->s6_addr32[0];
98 c += (__force u32)daddr->s6_addr32[1];
101 a += (__force u32)daddr->s6_addr32[2];
102 b += (__force u32)daddr->s6_addr32[3];
103 c += (__force u32)id;
104 __jhash_mix(a, b, c);
106 return c & (INETFRAGS_HASHSZ - 1);
109 static unsigned int nf_hashfn(struct inet_frag_queue *q)
111 struct nf_ct_frag6_queue *nq;
113 nq = container_of(q, struct nf_ct_frag6_queue, q);
114 return ip6qhashfn(nq->id, &nq->saddr, &nq->daddr);
117 static void nf_skb_free(struct sk_buff *skb)
119 if (NFCT_FRAG6_CB(skb)->orig)
120 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
123 /* Memory Tracking Functions. */
124 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
127 *work -= skb->truesize;
128 atomic_sub(skb->truesize, &nf_frags.mem);
133 static void nf_frag_free(struct inet_frag_queue *q)
135 kfree(container_of(q, struct nf_ct_frag6_queue, q));
138 /* Destruction primitives. */
140 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
142 inet_frag_put(&fq->q, &nf_frags);
145 /* Kill fq entry. It is not destroyed immediately,
146 * because caller (and someone more) holds reference count.
148 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
150 inet_frag_kill(&fq->q, &nf_frags);
153 static void nf_ct_frag6_evictor(void)
155 inet_frag_evictor(&nf_frags);
158 static void nf_ct_frag6_expire(unsigned long data)
160 struct nf_ct_frag6_queue *fq;
162 fq = container_of((struct inet_frag_queue *)data,
163 struct nf_ct_frag6_queue, q);
165 spin_lock(&fq->q.lock);
167 if (fq->q.last_in & COMPLETE)
173 spin_unlock(&fq->q.lock);
177 /* Creation primitives. */
179 static __inline__ struct nf_ct_frag6_queue *
180 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
182 struct inet_frag_queue *q;
183 struct ip6_create_arg arg;
189 hash = ip6qhashfn(id, src, dst);
191 q = inet_frag_find(&nf_frags, &arg, hash);
195 return container_of(q, struct nf_ct_frag6_queue, q);
198 pr_debug("Can't alloc new queue\n");
203 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
204 struct frag_hdr *fhdr, int nhoff)
206 struct sk_buff *prev, *next;
209 if (fq->q.last_in & COMPLETE) {
210 pr_debug("Allready completed\n");
214 offset = ntohs(fhdr->frag_off) & ~0x7;
215 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
216 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
218 if ((unsigned int)end > IPV6_MAXPLEN) {
219 pr_debug("offset is too large.\n");
223 if (skb->ip_summed == CHECKSUM_COMPLETE) {
224 const unsigned char *nh = skb_network_header(skb);
225 skb->csum = csum_sub(skb->csum,
226 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
230 /* Is this the final fragment? */
231 if (!(fhdr->frag_off & htons(IP6_MF))) {
232 /* If we already have some bits beyond end
233 * or have different end, the segment is corrupted.
235 if (end < fq->q.len ||
236 ((fq->q.last_in & LAST_IN) && end != fq->q.len)) {
237 pr_debug("already received last fragment\n");
240 fq->q.last_in |= LAST_IN;
243 /* Check if the fragment is rounded to 8 bytes.
244 * Required by the RFC.
247 /* RFC2460 says always send parameter problem in
250 pr_debug("end of fragment not rounded to 8 bytes.\n");
253 if (end > fq->q.len) {
254 /* Some bits beyond end -> corruption. */
255 if (fq->q.last_in & LAST_IN) {
256 pr_debug("last packet already reached.\n");
266 /* Point into the IP datagram 'data' part. */
267 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
268 pr_debug("queue: message is too short.\n");
271 if (pskb_trim_rcsum(skb, end - offset)) {
272 pr_debug("Can't trim\n");
276 /* Find out which fragments are in front and at the back of us
277 * in the chain of fragments so far. We must know where to put
278 * this fragment, right?
281 for (next = fq->q.fragments; next != NULL; next = next->next) {
282 if (NFCT_FRAG6_CB(next)->offset >= offset)
287 /* We found where to put this one. Check for overlap with
288 * preceding fragment, and, if needed, align things so that
289 * any overlaps are eliminated.
292 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
297 pr_debug("overlap\n");
300 if (!pskb_pull(skb, i)) {
301 pr_debug("Can't pull\n");
304 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
305 skb->ip_summed = CHECKSUM_NONE;
309 /* Look for overlap with succeeding segments.
310 * If we can merge fragments, do it.
312 while (next && NFCT_FRAG6_CB(next)->offset < end) {
313 /* overlap is 'i' bytes */
314 int i = end - NFCT_FRAG6_CB(next)->offset;
317 /* Eat head of the next overlapped fragment
318 * and leave the loop. The next ones cannot overlap.
320 pr_debug("Eat head of the overlapped parts.: %d", i);
321 if (!pskb_pull(next, i))
325 NFCT_FRAG6_CB(next)->offset += i;
327 if (next->ip_summed != CHECKSUM_UNNECESSARY)
328 next->ip_summed = CHECKSUM_NONE;
331 struct sk_buff *free_it = next;
333 /* Old fragmnet is completely overridden with
341 fq->q.fragments = next;
343 fq->q.meat -= free_it->len;
344 frag_kfree_skb(free_it, NULL);
348 NFCT_FRAG6_CB(skb)->offset = offset;
350 /* Insert this fragment in the chain of fragments. */
355 fq->q.fragments = skb;
358 fq->q.stamp = skb->tstamp;
359 fq->q.meat += skb->len;
360 atomic_add(skb->truesize, &nf_frags.mem);
362 /* The first fragment.
363 * nhoffset is obtained from the first fragment, of course.
366 fq->nhoffset = nhoff;
367 fq->q.last_in |= FIRST_IN;
369 write_lock(&nf_frags.lock);
370 list_move_tail(&fq->q.lru_list, &nf_frags.lru_list);
371 write_unlock(&nf_frags.lock);
379 * Check if this packet is complete.
380 * Returns NULL on failure by any reason, and pointer
381 * to current nexthdr field in reassembled frame.
383 * It is called with locked fq, and caller must check that
384 * queue is eligible for reassembly i.e. it is not COMPLETE,
385 * the last and the first frames arrived and all the bits are here.
387 static struct sk_buff *
388 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
390 struct sk_buff *fp, *op, *head = fq->q.fragments;
395 BUG_TRAP(head != NULL);
396 BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
398 /* Unfragmented part is taken from the first segment. */
399 payload_len = ((head->data - skb_network_header(head)) -
400 sizeof(struct ipv6hdr) + fq->q.len -
401 sizeof(struct frag_hdr));
402 if (payload_len > IPV6_MAXPLEN) {
403 pr_debug("payload len is too large.\n");
407 /* Head of list must not be cloned. */
408 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
409 pr_debug("skb is cloned but can't expand head");
413 /* If the first fragment is fragmented itself, we split
414 * it to two chunks: the first with data and paged part
415 * and the second, holding only fragments. */
416 if (skb_shinfo(head)->frag_list) {
417 struct sk_buff *clone;
420 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
421 pr_debug("Can't alloc skb\n");
424 clone->next = head->next;
426 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
427 skb_shinfo(head)->frag_list = NULL;
428 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
429 plen += skb_shinfo(head)->frags[i].size;
430 clone->len = clone->data_len = head->data_len - plen;
431 head->data_len -= clone->len;
432 head->len -= clone->len;
434 clone->ip_summed = head->ip_summed;
436 NFCT_FRAG6_CB(clone)->orig = NULL;
437 atomic_add(clone->truesize, &nf_frags.mem);
440 /* We have to remove fragment header from datagram and to relocate
441 * header in order to calculate ICV correctly. */
442 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
443 memmove(head->head + sizeof(struct frag_hdr), head->head,
444 (head->data - head->head) - sizeof(struct frag_hdr));
445 head->mac_header += sizeof(struct frag_hdr);
446 head->network_header += sizeof(struct frag_hdr);
448 skb_shinfo(head)->frag_list = head->next;
449 skb_reset_transport_header(head);
450 skb_push(head, head->data - skb_network_header(head));
451 atomic_sub(head->truesize, &nf_frags.mem);
453 for (fp=head->next; fp; fp = fp->next) {
454 head->data_len += fp->len;
455 head->len += fp->len;
456 if (head->ip_summed != fp->ip_summed)
457 head->ip_summed = CHECKSUM_NONE;
458 else if (head->ip_summed == CHECKSUM_COMPLETE)
459 head->csum = csum_add(head->csum, fp->csum);
460 head->truesize += fp->truesize;
461 atomic_sub(fp->truesize, &nf_frags.mem);
466 head->tstamp = fq->q.stamp;
467 ipv6_hdr(head)->payload_len = htons(payload_len);
469 /* Yes, and fold redundant checksum back. 8) */
470 if (head->ip_summed == CHECKSUM_COMPLETE)
471 head->csum = csum_partial(skb_network_header(head),
472 skb_network_header_len(head),
475 fq->q.fragments = NULL;
477 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
478 fp = skb_shinfo(head)->frag_list;
479 if (NFCT_FRAG6_CB(fp)->orig == NULL)
480 /* at above code, head skb is divided into two skbs. */
483 op = NFCT_FRAG6_CB(head)->orig;
484 for (; fp; fp = fp->next) {
485 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
489 NFCT_FRAG6_CB(fp)->orig = NULL;
496 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
500 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
506 * find the header just before Fragment Header.
508 * if success return 0 and set ...
509 * (*prevhdrp): the value of "Next Header Field" in the header
510 * just before Fragment Header.
511 * (*prevhoff): the offset of "Next Header Field" in the header
512 * just before Fragment Header.
513 * (*fhoff) : the offset of Fragment Header.
515 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
519 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
521 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
522 const int netoff = skb_network_offset(skb);
523 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
524 int start = netoff + sizeof(struct ipv6hdr);
525 int len = skb->len - start;
526 u8 prevhdr = NEXTHDR_IPV6;
528 while (nexthdr != NEXTHDR_FRAGMENT) {
529 struct ipv6_opt_hdr hdr;
532 if (!ipv6_ext_hdr(nexthdr)) {
535 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
536 pr_debug("too short\n");
539 if (nexthdr == NEXTHDR_NONE) {
540 pr_debug("next header is none\n");
543 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
545 if (nexthdr == NEXTHDR_AUTH)
546 hdrlen = (hdr.hdrlen+2)<<2;
548 hdrlen = ipv6_optlen(&hdr);
553 nexthdr = hdr.nexthdr;
562 *prevhoff = prev_nhoff;
568 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
570 struct sk_buff *clone;
571 struct net_device *dev = skb->dev;
572 struct frag_hdr *fhdr;
573 struct nf_ct_frag6_queue *fq;
577 struct sk_buff *ret_skb = NULL;
579 /* Jumbo payload inhibits frag. header */
580 if (ipv6_hdr(skb)->payload_len == 0) {
581 pr_debug("payload len = 0\n");
585 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
588 clone = skb_clone(skb, GFP_ATOMIC);
590 pr_debug("Can't clone skb\n");
594 NFCT_FRAG6_CB(clone)->orig = skb;
596 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
597 pr_debug("message is too short.\n");
601 skb_set_transport_header(clone, fhoff);
602 hdr = ipv6_hdr(clone);
603 fhdr = (struct frag_hdr *)skb_transport_header(clone);
605 if (!(fhdr->frag_off & htons(0xFFF9))) {
606 pr_debug("Invalid fragment offset\n");
607 /* It is not a fragmented frame */
611 if (atomic_read(&nf_frags.mem) > nf_frags_ctl.high_thresh)
612 nf_ct_frag6_evictor();
614 fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
616 pr_debug("Can't find and can't create new queue\n");
620 spin_lock(&fq->q.lock);
622 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
623 spin_unlock(&fq->q.lock);
624 pr_debug("Can't insert skb to queue\n");
629 if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) {
630 ret_skb = nf_ct_frag6_reasm(fq, dev);
632 pr_debug("Can't reassemble fragmented packets\n");
634 spin_unlock(&fq->q.lock);
644 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
645 struct net_device *in, struct net_device *out,
646 int (*okfn)(struct sk_buff *))
648 struct sk_buff *s, *s2;
650 for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
651 nf_conntrack_put_reasm(s->nfct_reasm);
652 nf_conntrack_get_reasm(skb);
658 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
659 NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
662 nf_conntrack_put_reasm(skb);
665 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
667 struct sk_buff *s, *s2;
669 for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
680 int nf_ct_frag6_init(void)
682 nf_frags.ctl = &nf_frags_ctl;
683 nf_frags.hashfn = nf_hashfn;
684 nf_frags.constructor = ip6_frag_init;
685 nf_frags.destructor = nf_frag_free;
686 nf_frags.skb_free = nf_skb_free;
687 nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
688 nf_frags.match = ip6_frag_match;
689 nf_frags.equal = ip6_frag_equal;
690 nf_frags.frag_expire = nf_ct_frag6_expire;
691 inet_frags_init(&nf_frags);
696 void nf_ct_frag6_cleanup(void)
698 inet_frags_fini(&nf_frags);
700 nf_frags_ctl.low_thresh = 0;
701 nf_ct_frag6_evictor();