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 static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
140 struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
144 atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_frags.mem);
148 /* Destruction primitives. */
150 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
152 inet_frag_put(&fq->q, &nf_frags);
155 /* Kill fq entry. It is not destroyed immediately,
156 * because caller (and someone more) holds reference count.
158 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
160 inet_frag_kill(&fq->q, &nf_frags);
163 static void nf_ct_frag6_evictor(void)
165 inet_frag_evictor(&nf_frags);
168 static void nf_ct_frag6_expire(unsigned long data)
170 struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
172 spin_lock(&fq->q.lock);
174 if (fq->q.last_in & COMPLETE)
180 spin_unlock(&fq->q.lock);
184 /* Creation primitives. */
186 static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
187 struct nf_ct_frag6_queue *fq_in)
189 struct nf_ct_frag6_queue *fq;
191 struct hlist_node *n;
194 write_lock(&nf_frags.lock);
196 hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
197 if (fq->id == fq_in->id &&
198 ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
199 ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
200 atomic_inc(&fq->q.refcnt);
201 write_unlock(&nf_frags.lock);
202 fq_in->q.last_in |= COMPLETE;
210 if (!mod_timer(&fq->q.timer, jiffies + nf_frags_ctl.timeout))
211 atomic_inc(&fq->q.refcnt);
213 atomic_inc(&fq->q.refcnt);
214 hlist_add_head(&fq->q.list, &nf_frags.hash[hash]);
215 INIT_LIST_HEAD(&fq->q.lru_list);
216 list_add_tail(&fq->q.lru_list, &nf_frags.lru_list);
218 write_unlock(&nf_frags.lock);
223 static struct nf_ct_frag6_queue *
224 nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src, struct in6_addr *dst)
226 struct nf_ct_frag6_queue *fq;
228 if ((fq = frag_alloc_queue()) == NULL) {
229 pr_debug("Can't alloc new queue\n");
233 memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
236 ipv6_addr_copy(&fq->saddr, src);
237 ipv6_addr_copy(&fq->daddr, dst);
239 setup_timer(&fq->q.timer, nf_ct_frag6_expire, (unsigned long)fq);
240 spin_lock_init(&fq->q.lock);
241 atomic_set(&fq->q.refcnt, 1);
243 return nf_ct_frag6_intern(hash, fq);
249 static __inline__ struct nf_ct_frag6_queue *
250 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
252 struct nf_ct_frag6_queue *fq;
253 struct hlist_node *n;
254 unsigned int hash = ip6qhashfn(id, src, dst);
256 read_lock(&nf_frags.lock);
257 hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
259 ipv6_addr_equal(src, &fq->saddr) &&
260 ipv6_addr_equal(dst, &fq->daddr)) {
261 atomic_inc(&fq->q.refcnt);
262 read_unlock(&nf_frags.lock);
266 read_unlock(&nf_frags.lock);
268 return nf_ct_frag6_create(hash, id, src, dst);
272 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
273 struct frag_hdr *fhdr, int nhoff)
275 struct sk_buff *prev, *next;
278 if (fq->q.last_in & COMPLETE) {
279 pr_debug("Allready completed\n");
283 offset = ntohs(fhdr->frag_off) & ~0x7;
284 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
285 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
287 if ((unsigned int)end > IPV6_MAXPLEN) {
288 pr_debug("offset is too large.\n");
292 if (skb->ip_summed == CHECKSUM_COMPLETE) {
293 const unsigned char *nh = skb_network_header(skb);
294 skb->csum = csum_sub(skb->csum,
295 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
299 /* Is this the final fragment? */
300 if (!(fhdr->frag_off & htons(IP6_MF))) {
301 /* If we already have some bits beyond end
302 * or have different end, the segment is corrupted.
304 if (end < fq->q.len ||
305 ((fq->q.last_in & LAST_IN) && end != fq->q.len)) {
306 pr_debug("already received last fragment\n");
309 fq->q.last_in |= LAST_IN;
312 /* Check if the fragment is rounded to 8 bytes.
313 * Required by the RFC.
316 /* RFC2460 says always send parameter problem in
319 pr_debug("end of fragment not rounded to 8 bytes.\n");
322 if (end > fq->q.len) {
323 /* Some bits beyond end -> corruption. */
324 if (fq->q.last_in & LAST_IN) {
325 pr_debug("last packet already reached.\n");
335 /* Point into the IP datagram 'data' part. */
336 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
337 pr_debug("queue: message is too short.\n");
340 if (pskb_trim_rcsum(skb, end - offset)) {
341 pr_debug("Can't trim\n");
345 /* Find out which fragments are in front and at the back of us
346 * in the chain of fragments so far. We must know where to put
347 * this fragment, right?
350 for (next = fq->q.fragments; next != NULL; next = next->next) {
351 if (NFCT_FRAG6_CB(next)->offset >= offset)
356 /* We found where to put this one. Check for overlap with
357 * preceding fragment, and, if needed, align things so that
358 * any overlaps are eliminated.
361 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
366 pr_debug("overlap\n");
369 if (!pskb_pull(skb, i)) {
370 pr_debug("Can't pull\n");
373 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
374 skb->ip_summed = CHECKSUM_NONE;
378 /* Look for overlap with succeeding segments.
379 * If we can merge fragments, do it.
381 while (next && NFCT_FRAG6_CB(next)->offset < end) {
382 /* overlap is 'i' bytes */
383 int i = end - NFCT_FRAG6_CB(next)->offset;
386 /* Eat head of the next overlapped fragment
387 * and leave the loop. The next ones cannot overlap.
389 pr_debug("Eat head of the overlapped parts.: %d", i);
390 if (!pskb_pull(next, i))
394 NFCT_FRAG6_CB(next)->offset += i;
396 if (next->ip_summed != CHECKSUM_UNNECESSARY)
397 next->ip_summed = CHECKSUM_NONE;
400 struct sk_buff *free_it = next;
402 /* Old fragmnet is completely overridden with
410 fq->q.fragments = next;
412 fq->q.meat -= free_it->len;
413 frag_kfree_skb(free_it, NULL);
417 NFCT_FRAG6_CB(skb)->offset = offset;
419 /* Insert this fragment in the chain of fragments. */
424 fq->q.fragments = skb;
427 fq->q.stamp = skb->tstamp;
428 fq->q.meat += skb->len;
429 atomic_add(skb->truesize, &nf_frags.mem);
431 /* The first fragment.
432 * nhoffset is obtained from the first fragment, of course.
435 fq->nhoffset = nhoff;
436 fq->q.last_in |= FIRST_IN;
438 write_lock(&nf_frags.lock);
439 list_move_tail(&fq->q.lru_list, &nf_frags.lru_list);
440 write_unlock(&nf_frags.lock);
448 * Check if this packet is complete.
449 * Returns NULL on failure by any reason, and pointer
450 * to current nexthdr field in reassembled frame.
452 * It is called with locked fq, and caller must check that
453 * queue is eligible for reassembly i.e. it is not COMPLETE,
454 * the last and the first frames arrived and all the bits are here.
456 static struct sk_buff *
457 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
459 struct sk_buff *fp, *op, *head = fq->q.fragments;
464 BUG_TRAP(head != NULL);
465 BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
467 /* Unfragmented part is taken from the first segment. */
468 payload_len = ((head->data - skb_network_header(head)) -
469 sizeof(struct ipv6hdr) + fq->q.len -
470 sizeof(struct frag_hdr));
471 if (payload_len > IPV6_MAXPLEN) {
472 pr_debug("payload len is too large.\n");
476 /* Head of list must not be cloned. */
477 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
478 pr_debug("skb is cloned but can't expand head");
482 /* If the first fragment is fragmented itself, we split
483 * it to two chunks: the first with data and paged part
484 * and the second, holding only fragments. */
485 if (skb_shinfo(head)->frag_list) {
486 struct sk_buff *clone;
489 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
490 pr_debug("Can't alloc skb\n");
493 clone->next = head->next;
495 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
496 skb_shinfo(head)->frag_list = NULL;
497 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
498 plen += skb_shinfo(head)->frags[i].size;
499 clone->len = clone->data_len = head->data_len - plen;
500 head->data_len -= clone->len;
501 head->len -= clone->len;
503 clone->ip_summed = head->ip_summed;
505 NFCT_FRAG6_CB(clone)->orig = NULL;
506 atomic_add(clone->truesize, &nf_frags.mem);
509 /* We have to remove fragment header from datagram and to relocate
510 * header in order to calculate ICV correctly. */
511 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
512 memmove(head->head + sizeof(struct frag_hdr), head->head,
513 (head->data - head->head) - sizeof(struct frag_hdr));
514 head->mac_header += sizeof(struct frag_hdr);
515 head->network_header += sizeof(struct frag_hdr);
517 skb_shinfo(head)->frag_list = head->next;
518 skb_reset_transport_header(head);
519 skb_push(head, head->data - skb_network_header(head));
520 atomic_sub(head->truesize, &nf_frags.mem);
522 for (fp=head->next; fp; fp = fp->next) {
523 head->data_len += fp->len;
524 head->len += fp->len;
525 if (head->ip_summed != fp->ip_summed)
526 head->ip_summed = CHECKSUM_NONE;
527 else if (head->ip_summed == CHECKSUM_COMPLETE)
528 head->csum = csum_add(head->csum, fp->csum);
529 head->truesize += fp->truesize;
530 atomic_sub(fp->truesize, &nf_frags.mem);
535 head->tstamp = fq->q.stamp;
536 ipv6_hdr(head)->payload_len = htons(payload_len);
538 /* Yes, and fold redundant checksum back. 8) */
539 if (head->ip_summed == CHECKSUM_COMPLETE)
540 head->csum = csum_partial(skb_network_header(head),
541 skb_network_header_len(head),
544 fq->q.fragments = NULL;
546 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
547 fp = skb_shinfo(head)->frag_list;
548 if (NFCT_FRAG6_CB(fp)->orig == NULL)
549 /* at above code, head skb is divided into two skbs. */
552 op = NFCT_FRAG6_CB(head)->orig;
553 for (; fp; fp = fp->next) {
554 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
558 NFCT_FRAG6_CB(fp)->orig = NULL;
565 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
569 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
575 * find the header just before Fragment Header.
577 * if success return 0 and set ...
578 * (*prevhdrp): the value of "Next Header Field" in the header
579 * just before Fragment Header.
580 * (*prevhoff): the offset of "Next Header Field" in the header
581 * just before Fragment Header.
582 * (*fhoff) : the offset of Fragment Header.
584 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
588 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
590 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
591 const int netoff = skb_network_offset(skb);
592 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
593 int start = netoff + sizeof(struct ipv6hdr);
594 int len = skb->len - start;
595 u8 prevhdr = NEXTHDR_IPV6;
597 while (nexthdr != NEXTHDR_FRAGMENT) {
598 struct ipv6_opt_hdr hdr;
601 if (!ipv6_ext_hdr(nexthdr)) {
604 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
605 pr_debug("too short\n");
608 if (nexthdr == NEXTHDR_NONE) {
609 pr_debug("next header is none\n");
612 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
614 if (nexthdr == NEXTHDR_AUTH)
615 hdrlen = (hdr.hdrlen+2)<<2;
617 hdrlen = ipv6_optlen(&hdr);
622 nexthdr = hdr.nexthdr;
631 *prevhoff = prev_nhoff;
637 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
639 struct sk_buff *clone;
640 struct net_device *dev = skb->dev;
641 struct frag_hdr *fhdr;
642 struct nf_ct_frag6_queue *fq;
646 struct sk_buff *ret_skb = NULL;
648 /* Jumbo payload inhibits frag. header */
649 if (ipv6_hdr(skb)->payload_len == 0) {
650 pr_debug("payload len = 0\n");
654 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
657 clone = skb_clone(skb, GFP_ATOMIC);
659 pr_debug("Can't clone skb\n");
663 NFCT_FRAG6_CB(clone)->orig = skb;
665 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
666 pr_debug("message is too short.\n");
670 skb_set_transport_header(clone, fhoff);
671 hdr = ipv6_hdr(clone);
672 fhdr = (struct frag_hdr *)skb_transport_header(clone);
674 if (!(fhdr->frag_off & htons(0xFFF9))) {
675 pr_debug("Invalid fragment offset\n");
676 /* It is not a fragmented frame */
680 if (atomic_read(&nf_frags.mem) > nf_frags_ctl.high_thresh)
681 nf_ct_frag6_evictor();
683 fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
685 pr_debug("Can't find and can't create new queue\n");
689 spin_lock(&fq->q.lock);
691 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
692 spin_unlock(&fq->q.lock);
693 pr_debug("Can't insert skb to queue\n");
698 if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) {
699 ret_skb = nf_ct_frag6_reasm(fq, dev);
701 pr_debug("Can't reassemble fragmented packets\n");
703 spin_unlock(&fq->q.lock);
713 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
714 struct net_device *in, struct net_device *out,
715 int (*okfn)(struct sk_buff *))
717 struct sk_buff *s, *s2;
719 for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
720 nf_conntrack_put_reasm(s->nfct_reasm);
721 nf_conntrack_get_reasm(skb);
727 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
728 NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
731 nf_conntrack_put_reasm(skb);
734 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
736 struct sk_buff *s, *s2;
738 for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
749 int nf_ct_frag6_init(void)
751 nf_frags.ctl = &nf_frags_ctl;
752 nf_frags.hashfn = nf_hashfn;
753 nf_frags.destructor = nf_frag_free;
754 nf_frags.skb_free = nf_skb_free;
755 nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
756 inet_frags_init(&nf_frags);
761 void nf_ct_frag6_cleanup(void)
763 inet_frags_fini(&nf_frags);
765 nf_frags_ctl.low_thresh = 0;
766 nf_ct_frag6_evictor();