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/slab.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 <net/netfilter/ipv6/nf_conntrack_ipv6.h>
43 #include <linux/sysctl.h>
44 #include <linux/netfilter.h>
45 #include <linux/netfilter_ipv6.h>
46 #include <linux/kernel.h>
47 #include <linux/module.h>
48 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
51 struct nf_ct_frag6_skb_cb
53 struct inet6_skb_parm h
;
58 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
60 struct nf_ct_frag6_queue
62 struct inet_frag_queue q
;
64 __be32 id
; /* fragment id */
66 struct in6_addr saddr
;
67 struct in6_addr daddr
;
73 static struct inet_frags nf_frags
;
74 static struct netns_frags nf_init_frags
;
77 static struct ctl_table nf_ct_frag6_sysctl_table
[] = {
79 .procname
= "nf_conntrack_frag6_timeout",
80 .data
= &nf_init_frags
.timeout
,
81 .maxlen
= sizeof(unsigned int),
83 .proc_handler
= proc_dointvec_jiffies
,
86 .procname
= "nf_conntrack_frag6_low_thresh",
87 .data
= &nf_init_frags
.low_thresh
,
88 .maxlen
= sizeof(unsigned int),
90 .proc_handler
= proc_dointvec
,
93 .procname
= "nf_conntrack_frag6_high_thresh",
94 .data
= &nf_init_frags
.high_thresh
,
95 .maxlen
= sizeof(unsigned int),
97 .proc_handler
= proc_dointvec
,
102 static struct ctl_table_header
*nf_ct_frag6_sysctl_header
;
105 static unsigned int nf_hashfn(struct inet_frag_queue
*q
)
107 const struct nf_ct_frag6_queue
*nq
;
109 nq
= container_of(q
, struct nf_ct_frag6_queue
, q
);
110 return inet6_hash_frag(nq
->id
, &nq
->saddr
, &nq
->daddr
, nf_frags
.rnd
);
113 static void nf_skb_free(struct sk_buff
*skb
)
115 if (NFCT_FRAG6_CB(skb
)->orig
)
116 kfree_skb(NFCT_FRAG6_CB(skb
)->orig
);
119 /* Destruction primitives. */
121 static __inline__
void fq_put(struct nf_ct_frag6_queue
*fq
)
123 inet_frag_put(&fq
->q
, &nf_frags
);
126 /* Kill fq entry. It is not destroyed immediately,
127 * because caller (and someone more) holds reference count.
129 static __inline__
void fq_kill(struct nf_ct_frag6_queue
*fq
)
131 inet_frag_kill(&fq
->q
, &nf_frags
);
134 static void nf_ct_frag6_evictor(void)
137 inet_frag_evictor(&nf_init_frags
, &nf_frags
);
141 static void nf_ct_frag6_expire(unsigned long data
)
143 struct nf_ct_frag6_queue
*fq
;
145 fq
= container_of((struct inet_frag_queue
*)data
,
146 struct nf_ct_frag6_queue
, q
);
148 spin_lock(&fq
->q
.lock
);
150 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
156 spin_unlock(&fq
->q
.lock
);
160 /* Creation primitives. */
162 static __inline__
struct nf_ct_frag6_queue
*
163 fq_find(__be32 id
, u32 user
, struct in6_addr
*src
, struct in6_addr
*dst
)
165 struct inet_frag_queue
*q
;
166 struct ip6_create_arg arg
;
174 read_lock_bh(&nf_frags
.lock
);
175 hash
= inet6_hash_frag(id
, src
, dst
, nf_frags
.rnd
);
177 q
= inet_frag_find(&nf_init_frags
, &nf_frags
, &arg
, hash
);
182 return container_of(q
, struct nf_ct_frag6_queue
, q
);
189 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue
*fq
, struct sk_buff
*skb
,
190 const struct frag_hdr
*fhdr
, int nhoff
)
192 struct sk_buff
*prev
, *next
;
195 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
) {
196 pr_debug("Already completed\n");
200 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
201 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
202 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
204 if ((unsigned int)end
> IPV6_MAXPLEN
) {
205 pr_debug("offset is too large.\n");
209 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
210 const unsigned char *nh
= skb_network_header(skb
);
211 skb
->csum
= csum_sub(skb
->csum
,
212 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
216 /* Is this the final fragment? */
217 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
218 /* If we already have some bits beyond end
219 * or have different end, the segment is corrupted.
221 if (end
< fq
->q
.len
||
222 ((fq
->q
.last_in
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
)) {
223 pr_debug("already received last fragment\n");
226 fq
->q
.last_in
|= INET_FRAG_LAST_IN
;
229 /* Check if the fragment is rounded to 8 bytes.
230 * Required by the RFC.
233 /* RFC2460 says always send parameter problem in
236 pr_debug("end of fragment not rounded to 8 bytes.\n");
239 if (end
> fq
->q
.len
) {
240 /* Some bits beyond end -> corruption. */
241 if (fq
->q
.last_in
& INET_FRAG_LAST_IN
) {
242 pr_debug("last packet already reached.\n");
252 /* Point into the IP datagram 'data' part. */
253 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
)) {
254 pr_debug("queue: message is too short.\n");
257 if (pskb_trim_rcsum(skb
, end
- offset
)) {
258 pr_debug("Can't trim\n");
262 /* Find out which fragments are in front and at the back of us
263 * in the chain of fragments so far. We must know where to put
264 * this fragment, right?
266 prev
= fq
->q
.fragments_tail
;
267 if (!prev
|| NFCT_FRAG6_CB(prev
)->offset
< offset
) {
272 for (next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
273 if (NFCT_FRAG6_CB(next
)->offset
>= offset
)
279 /* RFC5722, Section 4:
280 * When reassembling an IPv6 datagram, if
281 * one or more its constituent fragments is determined to be an
282 * overlapping fragment, the entire datagram (and any constituent
283 * fragments, including those not yet received) MUST be silently
287 /* Check for overlap with preceding fragment. */
289 (NFCT_FRAG6_CB(prev
)->offset
+ prev
->len
) > offset
)
292 /* Look for overlap with succeeding segment. */
293 if (next
&& NFCT_FRAG6_CB(next
)->offset
< end
)
296 NFCT_FRAG6_CB(skb
)->offset
= offset
;
298 /* Insert this fragment in the chain of fragments. */
301 fq
->q
.fragments_tail
= skb
;
305 fq
->q
.fragments
= skb
;
308 fq
->q
.stamp
= skb
->tstamp
;
309 fq
->q
.meat
+= skb
->len
;
310 atomic_add(skb
->truesize
, &nf_init_frags
.mem
);
312 /* The first fragment.
313 * nhoffset is obtained from the first fragment, of course.
316 fq
->nhoffset
= nhoff
;
317 fq
->q
.last_in
|= INET_FRAG_FIRST_IN
;
319 write_lock(&nf_frags
.lock
);
320 list_move_tail(&fq
->q
.lru_list
, &nf_init_frags
.lru_list
);
321 write_unlock(&nf_frags
.lock
);
331 * Check if this packet is complete.
332 * Returns NULL on failure by any reason, and pointer
333 * to current nexthdr field in reassembled frame.
335 * It is called with locked fq, and caller must check that
336 * queue is eligible for reassembly i.e. it is not COMPLETE,
337 * the last and the first frames arrived and all the bits are here.
339 static struct sk_buff
*
340 nf_ct_frag6_reasm(struct nf_ct_frag6_queue
*fq
, struct net_device
*dev
)
342 struct sk_buff
*fp
, *op
, *head
= fq
->q
.fragments
;
347 WARN_ON(head
== NULL
);
348 WARN_ON(NFCT_FRAG6_CB(head
)->offset
!= 0);
350 /* Unfragmented part is taken from the first segment. */
351 payload_len
= ((head
->data
- skb_network_header(head
)) -
352 sizeof(struct ipv6hdr
) + fq
->q
.len
-
353 sizeof(struct frag_hdr
));
354 if (payload_len
> IPV6_MAXPLEN
) {
355 pr_debug("payload len is too large.\n");
359 /* Head of list must not be cloned. */
360 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
)) {
361 pr_debug("skb is cloned but can't expand head");
365 /* If the first fragment is fragmented itself, we split
366 * it to two chunks: the first with data and paged part
367 * and the second, holding only fragments. */
368 if (skb_has_frag_list(head
)) {
369 struct sk_buff
*clone
;
372 clone
= alloc_skb(0, GFP_ATOMIC
);
376 clone
->next
= head
->next
;
378 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
379 skb_frag_list_init(head
);
380 for (i
= 0; i
< skb_shinfo(head
)->nr_frags
; i
++)
381 plen
+= skb_frag_size(&skb_shinfo(head
)->frags
[i
]);
382 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
383 head
->data_len
-= clone
->len
;
384 head
->len
-= clone
->len
;
386 clone
->ip_summed
= head
->ip_summed
;
388 NFCT_FRAG6_CB(clone
)->orig
= NULL
;
389 atomic_add(clone
->truesize
, &nf_init_frags
.mem
);
392 /* We have to remove fragment header from datagram and to relocate
393 * header in order to calculate ICV correctly. */
394 skb_network_header(head
)[fq
->nhoffset
] = skb_transport_header(head
)[0];
395 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
396 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
397 head
->mac_header
+= sizeof(struct frag_hdr
);
398 head
->network_header
+= sizeof(struct frag_hdr
);
400 skb_shinfo(head
)->frag_list
= head
->next
;
401 skb_reset_transport_header(head
);
402 skb_push(head
, head
->data
- skb_network_header(head
));
404 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
405 head
->data_len
+= fp
->len
;
406 head
->len
+= fp
->len
;
407 if (head
->ip_summed
!= fp
->ip_summed
)
408 head
->ip_summed
= CHECKSUM_NONE
;
409 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
410 head
->csum
= csum_add(head
->csum
, fp
->csum
);
411 head
->truesize
+= fp
->truesize
;
413 atomic_sub(head
->truesize
, &nf_init_frags
.mem
);
417 head
->tstamp
= fq
->q
.stamp
;
418 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
420 /* Yes, and fold redundant checksum back. 8) */
421 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
422 head
->csum
= csum_partial(skb_network_header(head
),
423 skb_network_header_len(head
),
426 fq
->q
.fragments
= NULL
;
427 fq
->q
.fragments_tail
= NULL
;
429 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
430 fp
= skb_shinfo(head
)->frag_list
;
431 if (fp
&& NFCT_FRAG6_CB(fp
)->orig
== NULL
)
432 /* at above code, head skb is divided into two skbs. */
435 op
= NFCT_FRAG6_CB(head
)->orig
;
436 for (; fp
; fp
= fp
->next
) {
437 struct sk_buff
*orig
= NFCT_FRAG6_CB(fp
)->orig
;
441 NFCT_FRAG6_CB(fp
)->orig
= NULL
;
448 printk(KERN_DEBUG
"nf_ct_frag6_reasm: payload len = %d\n", payload_len
);
452 printk(KERN_DEBUG
"nf_ct_frag6_reasm: no memory for reassembly\n");
458 * find the header just before Fragment Header.
460 * if success return 0 and set ...
461 * (*prevhdrp): the value of "Next Header Field" in the header
462 * just before Fragment Header.
463 * (*prevhoff): the offset of "Next Header Field" in the header
464 * just before Fragment Header.
465 * (*fhoff) : the offset of Fragment Header.
467 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
471 find_prev_fhdr(struct sk_buff
*skb
, u8
*prevhdrp
, int *prevhoff
, int *fhoff
)
473 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
474 const int netoff
= skb_network_offset(skb
);
475 u8 prev_nhoff
= netoff
+ offsetof(struct ipv6hdr
, nexthdr
);
476 int start
= netoff
+ sizeof(struct ipv6hdr
);
477 int len
= skb
->len
- start
;
478 u8 prevhdr
= NEXTHDR_IPV6
;
480 while (nexthdr
!= NEXTHDR_FRAGMENT
) {
481 struct ipv6_opt_hdr hdr
;
484 if (!ipv6_ext_hdr(nexthdr
)) {
487 if (nexthdr
== NEXTHDR_NONE
) {
488 pr_debug("next header is none\n");
491 if (len
< (int)sizeof(struct ipv6_opt_hdr
)) {
492 pr_debug("too short\n");
495 if (skb_copy_bits(skb
, start
, &hdr
, sizeof(hdr
)))
497 if (nexthdr
== NEXTHDR_AUTH
)
498 hdrlen
= (hdr
.hdrlen
+2)<<2;
500 hdrlen
= ipv6_optlen(&hdr
);
505 nexthdr
= hdr
.nexthdr
;
514 *prevhoff
= prev_nhoff
;
520 struct sk_buff
*nf_ct_frag6_gather(struct sk_buff
*skb
, u32 user
)
522 struct sk_buff
*clone
;
523 struct net_device
*dev
= skb
->dev
;
524 struct frag_hdr
*fhdr
;
525 struct nf_ct_frag6_queue
*fq
;
529 struct sk_buff
*ret_skb
= NULL
;
531 /* Jumbo payload inhibits frag. header */
532 if (ipv6_hdr(skb
)->payload_len
== 0) {
533 pr_debug("payload len = 0\n");
537 if (find_prev_fhdr(skb
, &prevhdr
, &nhoff
, &fhoff
) < 0)
540 clone
= skb_clone(skb
, GFP_ATOMIC
);
542 pr_debug("Can't clone skb\n");
546 NFCT_FRAG6_CB(clone
)->orig
= skb
;
548 if (!pskb_may_pull(clone
, fhoff
+ sizeof(*fhdr
))) {
549 pr_debug("message is too short.\n");
553 skb_set_transport_header(clone
, fhoff
);
554 hdr
= ipv6_hdr(clone
);
555 fhdr
= (struct frag_hdr
*)skb_transport_header(clone
);
557 if (atomic_read(&nf_init_frags
.mem
) > nf_init_frags
.high_thresh
)
558 nf_ct_frag6_evictor();
560 fq
= fq_find(fhdr
->identification
, user
, &hdr
->saddr
, &hdr
->daddr
);
562 pr_debug("Can't find and can't create new queue\n");
566 spin_lock_bh(&fq
->q
.lock
);
568 if (nf_ct_frag6_queue(fq
, clone
, fhdr
, nhoff
) < 0) {
569 spin_unlock_bh(&fq
->q
.lock
);
570 pr_debug("Can't insert skb to queue\n");
575 if (fq
->q
.last_in
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
576 fq
->q
.meat
== fq
->q
.len
) {
577 ret_skb
= nf_ct_frag6_reasm(fq
, dev
);
579 pr_debug("Can't reassemble fragmented packets\n");
581 spin_unlock_bh(&fq
->q
.lock
);
591 void nf_ct_frag6_output(unsigned int hooknum
, struct sk_buff
*skb
,
592 struct net_device
*in
, struct net_device
*out
,
593 int (*okfn
)(struct sk_buff
*))
595 struct sk_buff
*s
, *s2
;
597 for (s
= NFCT_FRAG6_CB(skb
)->orig
; s
;) {
598 nf_conntrack_put_reasm(s
->nfct_reasm
);
599 nf_conntrack_get_reasm(skb
);
605 NF_HOOK_THRESH(NFPROTO_IPV6
, hooknum
, s
, in
, out
, okfn
,
606 NF_IP6_PRI_CONNTRACK_DEFRAG
+ 1);
609 nf_conntrack_put_reasm(skb
);
612 int nf_ct_frag6_init(void)
614 nf_frags
.hashfn
= nf_hashfn
;
615 nf_frags
.constructor
= ip6_frag_init
;
616 nf_frags
.destructor
= NULL
;
617 nf_frags
.skb_free
= nf_skb_free
;
618 nf_frags
.qsize
= sizeof(struct nf_ct_frag6_queue
);
619 nf_frags
.match
= ip6_frag_match
;
620 nf_frags
.frag_expire
= nf_ct_frag6_expire
;
621 nf_frags
.secret_interval
= 10 * 60 * HZ
;
622 nf_init_frags
.timeout
= IPV6_FRAG_TIMEOUT
;
623 nf_init_frags
.high_thresh
= IPV6_FRAG_HIGH_THRESH
;
624 nf_init_frags
.low_thresh
= IPV6_FRAG_LOW_THRESH
;
625 inet_frags_init_net(&nf_init_frags
);
626 inet_frags_init(&nf_frags
);
629 nf_ct_frag6_sysctl_header
= register_sysctl_paths(nf_net_netfilter_sysctl_path
,
630 nf_ct_frag6_sysctl_table
);
631 if (!nf_ct_frag6_sysctl_header
) {
632 inet_frags_fini(&nf_frags
);
640 void nf_ct_frag6_cleanup(void)
643 unregister_sysctl_table(nf_ct_frag6_sysctl_header
);
644 nf_ct_frag6_sysctl_header
= NULL
;
646 inet_frags_fini(&nf_frags
);
648 nf_init_frags
.low_thresh
= 0;
649 nf_ct_frag6_evictor();