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>
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 */
67 struct in6_addr saddr
;
68 struct in6_addr daddr
;
74 static struct inet_frags nf_frags
;
75 static struct netns_frags nf_init_frags
;
78 struct ctl_table nf_ct_ipv6_sysctl_table
[] = {
80 .procname
= "nf_conntrack_frag6_timeout",
81 .data
= &nf_init_frags
.timeout
,
82 .maxlen
= sizeof(unsigned int),
84 .proc_handler
= proc_dointvec_jiffies
,
87 .procname
= "nf_conntrack_frag6_low_thresh",
88 .data
= &nf_init_frags
.low_thresh
,
89 .maxlen
= sizeof(unsigned int),
91 .proc_handler
= proc_dointvec
,
94 .procname
= "nf_conntrack_frag6_high_thresh",
95 .data
= &nf_init_frags
.high_thresh
,
96 .maxlen
= sizeof(unsigned int),
98 .proc_handler
= proc_dointvec
,
104 static unsigned int nf_hashfn(struct inet_frag_queue
*q
)
106 const struct nf_ct_frag6_queue
*nq
;
108 nq
= container_of(q
, struct nf_ct_frag6_queue
, q
);
109 return inet6_hash_frag(nq
->id
, &nq
->saddr
, &nq
->daddr
, nf_frags
.rnd
);
112 static void nf_skb_free(struct sk_buff
*skb
)
114 if (NFCT_FRAG6_CB(skb
)->orig
)
115 kfree_skb(NFCT_FRAG6_CB(skb
)->orig
);
118 /* Memory Tracking Functions. */
119 static inline void frag_kfree_skb(struct sk_buff
*skb
, unsigned int *work
)
122 *work
-= skb
->truesize
;
123 atomic_sub(skb
->truesize
, &nf_init_frags
.mem
);
128 /* Destruction primitives. */
130 static __inline__
void fq_put(struct nf_ct_frag6_queue
*fq
)
132 inet_frag_put(&fq
->q
, &nf_frags
);
135 /* Kill fq entry. It is not destroyed immediately,
136 * because caller (and someone more) holds reference count.
138 static __inline__
void fq_kill(struct nf_ct_frag6_queue
*fq
)
140 inet_frag_kill(&fq
->q
, &nf_frags
);
143 static void nf_ct_frag6_evictor(void)
146 inet_frag_evictor(&nf_init_frags
, &nf_frags
);
150 static void nf_ct_frag6_expire(unsigned long data
)
152 struct nf_ct_frag6_queue
*fq
;
154 fq
= container_of((struct inet_frag_queue
*)data
,
155 struct nf_ct_frag6_queue
, q
);
157 spin_lock(&fq
->q
.lock
);
159 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
165 spin_unlock(&fq
->q
.lock
);
169 /* Creation primitives. */
171 static __inline__
struct nf_ct_frag6_queue
*
172 fq_find(__be32 id
, u32 user
, struct in6_addr
*src
, struct in6_addr
*dst
)
174 struct inet_frag_queue
*q
;
175 struct ip6_create_arg arg
;
183 read_lock_bh(&nf_frags
.lock
);
184 hash
= inet6_hash_frag(id
, src
, dst
, nf_frags
.rnd
);
186 q
= inet_frag_find(&nf_init_frags
, &nf_frags
, &arg
, hash
);
191 return container_of(q
, struct nf_ct_frag6_queue
, q
);
194 pr_debug("Can't alloc new queue\n");
199 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue
*fq
, struct sk_buff
*skb
,
200 const struct frag_hdr
*fhdr
, int nhoff
)
202 struct sk_buff
*prev
, *next
;
205 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
) {
206 pr_debug("Allready completed\n");
210 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
211 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
212 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
214 if ((unsigned int)end
> IPV6_MAXPLEN
) {
215 pr_debug("offset is too large.\n");
219 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
220 const unsigned char *nh
= skb_network_header(skb
);
221 skb
->csum
= csum_sub(skb
->csum
,
222 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
226 /* Is this the final fragment? */
227 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
228 /* If we already have some bits beyond end
229 * or have different end, the segment is corrupted.
231 if (end
< fq
->q
.len
||
232 ((fq
->q
.last_in
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
)) {
233 pr_debug("already received last fragment\n");
236 fq
->q
.last_in
|= INET_FRAG_LAST_IN
;
239 /* Check if the fragment is rounded to 8 bytes.
240 * Required by the RFC.
243 /* RFC2460 says always send parameter problem in
246 pr_debug("end of fragment not rounded to 8 bytes.\n");
249 if (end
> fq
->q
.len
) {
250 /* Some bits beyond end -> corruption. */
251 if (fq
->q
.last_in
& INET_FRAG_LAST_IN
) {
252 pr_debug("last packet already reached.\n");
262 /* Point into the IP datagram 'data' part. */
263 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
)) {
264 pr_debug("queue: message is too short.\n");
267 if (pskb_trim_rcsum(skb
, end
- offset
)) {
268 pr_debug("Can't trim\n");
272 /* Find out which fragments are in front and at the back of us
273 * in the chain of fragments so far. We must know where to put
274 * this fragment, right?
277 for (next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
278 if (NFCT_FRAG6_CB(next
)->offset
>= offset
)
283 /* We found where to put this one. Check for overlap with
284 * preceding fragment, and, if needed, align things so that
285 * any overlaps are eliminated.
288 int i
= (NFCT_FRAG6_CB(prev
)->offset
+ prev
->len
) - offset
;
293 pr_debug("overlap\n");
296 if (!pskb_pull(skb
, i
)) {
297 pr_debug("Can't pull\n");
300 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
)
301 skb
->ip_summed
= CHECKSUM_NONE
;
305 /* Look for overlap with succeeding segments.
306 * If we can merge fragments, do it.
308 while (next
&& NFCT_FRAG6_CB(next
)->offset
< end
) {
309 /* overlap is 'i' bytes */
310 int i
= end
- NFCT_FRAG6_CB(next
)->offset
;
313 /* Eat head of the next overlapped fragment
314 * and leave the loop. The next ones cannot overlap.
316 pr_debug("Eat head of the overlapped parts.: %d", i
);
317 if (!pskb_pull(next
, i
))
321 NFCT_FRAG6_CB(next
)->offset
+= i
;
323 if (next
->ip_summed
!= CHECKSUM_UNNECESSARY
)
324 next
->ip_summed
= CHECKSUM_NONE
;
327 struct sk_buff
*free_it
= next
;
329 /* Old fragmnet is completely overridden with
337 fq
->q
.fragments
= next
;
339 fq
->q
.meat
-= free_it
->len
;
340 frag_kfree_skb(free_it
, NULL
);
344 NFCT_FRAG6_CB(skb
)->offset
= offset
;
346 /* Insert this fragment in the chain of fragments. */
351 fq
->q
.fragments
= skb
;
354 fq
->q
.stamp
= skb
->tstamp
;
355 fq
->q
.meat
+= skb
->len
;
356 atomic_add(skb
->truesize
, &nf_init_frags
.mem
);
358 /* The first fragment.
359 * nhoffset is obtained from the first fragment, of course.
362 fq
->nhoffset
= nhoff
;
363 fq
->q
.last_in
|= INET_FRAG_FIRST_IN
;
365 write_lock(&nf_frags
.lock
);
366 list_move_tail(&fq
->q
.lru_list
, &nf_init_frags
.lru_list
);
367 write_unlock(&nf_frags
.lock
);
375 * Check if this packet is complete.
376 * Returns NULL on failure by any reason, and pointer
377 * to current nexthdr field in reassembled frame.
379 * It is called with locked fq, and caller must check that
380 * queue is eligible for reassembly i.e. it is not COMPLETE,
381 * the last and the first frames arrived and all the bits are here.
383 static struct sk_buff
*
384 nf_ct_frag6_reasm(struct nf_ct_frag6_queue
*fq
, struct net_device
*dev
)
386 struct sk_buff
*fp
, *op
, *head
= fq
->q
.fragments
;
391 WARN_ON(head
== NULL
);
392 WARN_ON(NFCT_FRAG6_CB(head
)->offset
!= 0);
394 /* Unfragmented part is taken from the first segment. */
395 payload_len
= ((head
->data
- skb_network_header(head
)) -
396 sizeof(struct ipv6hdr
) + fq
->q
.len
-
397 sizeof(struct frag_hdr
));
398 if (payload_len
> IPV6_MAXPLEN
) {
399 pr_debug("payload len is too large.\n");
403 /* Head of list must not be cloned. */
404 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
)) {
405 pr_debug("skb is cloned but can't expand head");
409 /* If the first fragment is fragmented itself, we split
410 * it to two chunks: the first with data and paged part
411 * and the second, holding only fragments. */
412 if (skb_has_frags(head
)) {
413 struct sk_buff
*clone
;
416 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
) {
417 pr_debug("Can't alloc skb\n");
420 clone
->next
= head
->next
;
422 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
423 skb_frag_list_init(head
);
424 for (i
=0; i
<skb_shinfo(head
)->nr_frags
; i
++)
425 plen
+= skb_shinfo(head
)->frags
[i
].size
;
426 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
427 head
->data_len
-= clone
->len
;
428 head
->len
-= clone
->len
;
430 clone
->ip_summed
= head
->ip_summed
;
432 NFCT_FRAG6_CB(clone
)->orig
= NULL
;
433 atomic_add(clone
->truesize
, &nf_init_frags
.mem
);
436 /* We have to remove fragment header from datagram and to relocate
437 * header in order to calculate ICV correctly. */
438 skb_network_header(head
)[fq
->nhoffset
] = skb_transport_header(head
)[0];
439 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
440 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
441 head
->mac_header
+= sizeof(struct frag_hdr
);
442 head
->network_header
+= sizeof(struct frag_hdr
);
444 skb_shinfo(head
)->frag_list
= head
->next
;
445 skb_reset_transport_header(head
);
446 skb_push(head
, head
->data
- skb_network_header(head
));
447 atomic_sub(head
->truesize
, &nf_init_frags
.mem
);
449 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
450 head
->data_len
+= fp
->len
;
451 head
->len
+= fp
->len
;
452 if (head
->ip_summed
!= fp
->ip_summed
)
453 head
->ip_summed
= CHECKSUM_NONE
;
454 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
455 head
->csum
= csum_add(head
->csum
, fp
->csum
);
456 head
->truesize
+= fp
->truesize
;
457 atomic_sub(fp
->truesize
, &nf_init_frags
.mem
);
462 head
->tstamp
= fq
->q
.stamp
;
463 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
465 /* Yes, and fold redundant checksum back. 8) */
466 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
467 head
->csum
= csum_partial(skb_network_header(head
),
468 skb_network_header_len(head
),
471 fq
->q
.fragments
= NULL
;
473 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
474 fp
= skb_shinfo(head
)->frag_list
;
475 if (NFCT_FRAG6_CB(fp
)->orig
== NULL
)
476 /* at above code, head skb is divided into two skbs. */
479 op
= NFCT_FRAG6_CB(head
)->orig
;
480 for (; fp
; fp
= fp
->next
) {
481 struct sk_buff
*orig
= NFCT_FRAG6_CB(fp
)->orig
;
485 NFCT_FRAG6_CB(fp
)->orig
= NULL
;
492 printk(KERN_DEBUG
"nf_ct_frag6_reasm: payload len = %d\n", payload_len
);
496 printk(KERN_DEBUG
"nf_ct_frag6_reasm: no memory for reassembly\n");
502 * find the header just before Fragment Header.
504 * if success return 0 and set ...
505 * (*prevhdrp): the value of "Next Header Field" in the header
506 * just before Fragment Header.
507 * (*prevhoff): the offset of "Next Header Field" in the header
508 * just before Fragment Header.
509 * (*fhoff) : the offset of Fragment Header.
511 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
515 find_prev_fhdr(struct sk_buff
*skb
, u8
*prevhdrp
, int *prevhoff
, int *fhoff
)
517 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
518 const int netoff
= skb_network_offset(skb
);
519 u8 prev_nhoff
= netoff
+ offsetof(struct ipv6hdr
, nexthdr
);
520 int start
= netoff
+ sizeof(struct ipv6hdr
);
521 int len
= skb
->len
- start
;
522 u8 prevhdr
= NEXTHDR_IPV6
;
524 while (nexthdr
!= NEXTHDR_FRAGMENT
) {
525 struct ipv6_opt_hdr hdr
;
528 if (!ipv6_ext_hdr(nexthdr
)) {
531 if (nexthdr
== NEXTHDR_NONE
) {
532 pr_debug("next header is none\n");
535 if (len
< (int)sizeof(struct ipv6_opt_hdr
)) {
536 pr_debug("too short\n");
539 if (skb_copy_bits(skb
, start
, &hdr
, sizeof(hdr
)))
541 if (nexthdr
== NEXTHDR_AUTH
)
542 hdrlen
= (hdr
.hdrlen
+2)<<2;
544 hdrlen
= ipv6_optlen(&hdr
);
549 nexthdr
= hdr
.nexthdr
;
558 *prevhoff
= prev_nhoff
;
564 struct sk_buff
*nf_ct_frag6_gather(struct sk_buff
*skb
, u32 user
)
566 struct sk_buff
*clone
;
567 struct net_device
*dev
= skb
->dev
;
568 struct frag_hdr
*fhdr
;
569 struct nf_ct_frag6_queue
*fq
;
573 struct sk_buff
*ret_skb
= NULL
;
575 /* Jumbo payload inhibits frag. header */
576 if (ipv6_hdr(skb
)->payload_len
== 0) {
577 pr_debug("payload len = 0\n");
581 if (find_prev_fhdr(skb
, &prevhdr
, &nhoff
, &fhoff
) < 0)
584 clone
= skb_clone(skb
, GFP_ATOMIC
);
586 pr_debug("Can't clone skb\n");
590 NFCT_FRAG6_CB(clone
)->orig
= skb
;
592 if (!pskb_may_pull(clone
, fhoff
+ sizeof(*fhdr
))) {
593 pr_debug("message is too short.\n");
597 skb_set_transport_header(clone
, fhoff
);
598 hdr
= ipv6_hdr(clone
);
599 fhdr
= (struct frag_hdr
*)skb_transport_header(clone
);
601 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
602 pr_debug("Invalid fragment offset\n");
603 /* It is not a fragmented frame */
607 if (atomic_read(&nf_init_frags
.mem
) > nf_init_frags
.high_thresh
)
608 nf_ct_frag6_evictor();
610 fq
= fq_find(fhdr
->identification
, user
, &hdr
->saddr
, &hdr
->daddr
);
612 pr_debug("Can't find and can't create new queue\n");
616 spin_lock_bh(&fq
->q
.lock
);
618 if (nf_ct_frag6_queue(fq
, clone
, fhdr
, nhoff
) < 0) {
619 spin_unlock_bh(&fq
->q
.lock
);
620 pr_debug("Can't insert skb to queue\n");
625 if (fq
->q
.last_in
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
626 fq
->q
.meat
== fq
->q
.len
) {
627 ret_skb
= nf_ct_frag6_reasm(fq
, dev
);
629 pr_debug("Can't reassemble fragmented packets\n");
631 spin_unlock_bh(&fq
->q
.lock
);
641 void nf_ct_frag6_output(unsigned int hooknum
, struct sk_buff
*skb
,
642 struct net_device
*in
, struct net_device
*out
,
643 int (*okfn
)(struct sk_buff
*))
645 struct sk_buff
*s
, *s2
;
647 for (s
= NFCT_FRAG6_CB(skb
)->orig
; s
;) {
648 nf_conntrack_put_reasm(s
->nfct_reasm
);
649 nf_conntrack_get_reasm(skb
);
655 NF_HOOK_THRESH(PF_INET6
, hooknum
, s
, in
, out
, okfn
,
656 NF_IP6_PRI_CONNTRACK_DEFRAG
+ 1);
659 nf_conntrack_put_reasm(skb
);
662 int nf_ct_frag6_init(void)
664 nf_frags
.hashfn
= nf_hashfn
;
665 nf_frags
.constructor
= ip6_frag_init
;
666 nf_frags
.destructor
= NULL
;
667 nf_frags
.skb_free
= nf_skb_free
;
668 nf_frags
.qsize
= sizeof(struct nf_ct_frag6_queue
);
669 nf_frags
.match
= ip6_frag_match
;
670 nf_frags
.frag_expire
= nf_ct_frag6_expire
;
671 nf_frags
.secret_interval
= 10 * 60 * HZ
;
672 nf_init_frags
.timeout
= IPV6_FRAG_TIMEOUT
;
673 nf_init_frags
.high_thresh
= 256 * 1024;
674 nf_init_frags
.low_thresh
= 192 * 1024;
675 inet_frags_init_net(&nf_init_frags
);
676 inet_frags_init(&nf_frags
);
681 void nf_ct_frag6_cleanup(void)
683 inet_frags_fini(&nf_frags
);
685 nf_init_frags
.low_thresh
= 0;
686 nf_ct_frag6_evictor();