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>
50 struct nf_ct_frag6_skb_cb
52 struct inet6_skb_parm h
;
57 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
59 struct nf_ct_frag6_queue
61 struct inet_frag_queue q
;
63 __be32 id
; /* fragment id */
65 struct in6_addr saddr
;
66 struct in6_addr daddr
;
72 static struct inet_frags nf_frags
;
73 static struct netns_frags nf_init_frags
;
76 struct ctl_table nf_ct_ipv6_sysctl_table
[] = {
78 .procname
= "nf_conntrack_frag6_timeout",
79 .data
= &nf_init_frags
.timeout
,
80 .maxlen
= sizeof(unsigned int),
82 .proc_handler
= proc_dointvec_jiffies
,
85 .procname
= "nf_conntrack_frag6_low_thresh",
86 .data
= &nf_init_frags
.low_thresh
,
87 .maxlen
= sizeof(unsigned int),
89 .proc_handler
= proc_dointvec
,
92 .procname
= "nf_conntrack_frag6_high_thresh",
93 .data
= &nf_init_frags
.high_thresh
,
94 .maxlen
= sizeof(unsigned int),
96 .proc_handler
= proc_dointvec
,
102 static unsigned int nf_hashfn(struct inet_frag_queue
*q
)
104 const struct nf_ct_frag6_queue
*nq
;
106 nq
= container_of(q
, struct nf_ct_frag6_queue
, q
);
107 return inet6_hash_frag(nq
->id
, &nq
->saddr
, &nq
->daddr
, nf_frags
.rnd
);
110 static void nf_skb_free(struct sk_buff
*skb
)
112 if (NFCT_FRAG6_CB(skb
)->orig
)
113 kfree_skb(NFCT_FRAG6_CB(skb
)->orig
);
116 /* Destruction primitives. */
118 static __inline__
void fq_put(struct nf_ct_frag6_queue
*fq
)
120 inet_frag_put(&fq
->q
, &nf_frags
);
123 /* Kill fq entry. It is not destroyed immediately,
124 * because caller (and someone more) holds reference count.
126 static __inline__
void fq_kill(struct nf_ct_frag6_queue
*fq
)
128 inet_frag_kill(&fq
->q
, &nf_frags
);
131 static void nf_ct_frag6_evictor(void)
134 inet_frag_evictor(&nf_init_frags
, &nf_frags
);
138 static void nf_ct_frag6_expire(unsigned long data
)
140 struct nf_ct_frag6_queue
*fq
;
142 fq
= container_of((struct inet_frag_queue
*)data
,
143 struct nf_ct_frag6_queue
, q
);
145 spin_lock(&fq
->q
.lock
);
147 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
153 spin_unlock(&fq
->q
.lock
);
157 /* Creation primitives. */
159 static __inline__
struct nf_ct_frag6_queue
*
160 fq_find(__be32 id
, u32 user
, struct in6_addr
*src
, struct in6_addr
*dst
)
162 struct inet_frag_queue
*q
;
163 struct ip6_create_arg arg
;
171 read_lock_bh(&nf_frags
.lock
);
172 hash
= inet6_hash_frag(id
, src
, dst
, nf_frags
.rnd
);
174 q
= inet_frag_find(&nf_init_frags
, &nf_frags
, &arg
, hash
);
179 return container_of(q
, struct nf_ct_frag6_queue
, q
);
182 pr_debug("Can't alloc new queue\n");
187 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue
*fq
, struct sk_buff
*skb
,
188 const struct frag_hdr
*fhdr
, int nhoff
)
190 struct sk_buff
*prev
, *next
;
193 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
) {
194 pr_debug("Already completed\n");
198 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
199 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
200 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
202 if ((unsigned int)end
> IPV6_MAXPLEN
) {
203 pr_debug("offset is too large.\n");
207 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
208 const unsigned char *nh
= skb_network_header(skb
);
209 skb
->csum
= csum_sub(skb
->csum
,
210 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
214 /* Is this the final fragment? */
215 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
216 /* If we already have some bits beyond end
217 * or have different end, the segment is corrupted.
219 if (end
< fq
->q
.len
||
220 ((fq
->q
.last_in
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
)) {
221 pr_debug("already received last fragment\n");
224 fq
->q
.last_in
|= INET_FRAG_LAST_IN
;
227 /* Check if the fragment is rounded to 8 bytes.
228 * Required by the RFC.
231 /* RFC2460 says always send parameter problem in
234 pr_debug("end of fragment not rounded to 8 bytes.\n");
237 if (end
> fq
->q
.len
) {
238 /* Some bits beyond end -> corruption. */
239 if (fq
->q
.last_in
& INET_FRAG_LAST_IN
) {
240 pr_debug("last packet already reached.\n");
250 /* Point into the IP datagram 'data' part. */
251 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
)) {
252 pr_debug("queue: message is too short.\n");
255 if (pskb_trim_rcsum(skb
, end
- offset
)) {
256 pr_debug("Can't trim\n");
260 /* Find out which fragments are in front and at the back of us
261 * in the chain of fragments so far. We must know where to put
262 * this fragment, right?
264 prev
= fq
->q
.fragments_tail
;
265 if (!prev
|| NFCT_FRAG6_CB(prev
)->offset
< offset
) {
270 for (next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
271 if (NFCT_FRAG6_CB(next
)->offset
>= offset
)
277 /* RFC5722, Section 4:
278 * When reassembling an IPv6 datagram, if
279 * one or more its constituent fragments is determined to be an
280 * overlapping fragment, the entire datagram (and any constituent
281 * fragments, including those not yet received) MUST be silently
285 /* Check for overlap with preceding fragment. */
287 (NFCT_FRAG6_CB(prev
)->offset
+ prev
->len
) - offset
> 0)
290 /* Look for overlap with succeeding segment. */
291 if (next
&& NFCT_FRAG6_CB(next
)->offset
< end
)
294 NFCT_FRAG6_CB(skb
)->offset
= offset
;
296 /* Insert this fragment in the chain of fragments. */
299 fq
->q
.fragments_tail
= skb
;
303 fq
->q
.fragments
= skb
;
306 fq
->q
.stamp
= skb
->tstamp
;
307 fq
->q
.meat
+= skb
->len
;
308 atomic_add(skb
->truesize
, &nf_init_frags
.mem
);
310 /* The first fragment.
311 * nhoffset is obtained from the first fragment, of course.
314 fq
->nhoffset
= nhoff
;
315 fq
->q
.last_in
|= INET_FRAG_FIRST_IN
;
317 write_lock(&nf_frags
.lock
);
318 list_move_tail(&fq
->q
.lru_list
, &nf_init_frags
.lru_list
);
319 write_unlock(&nf_frags
.lock
);
329 * Check if this packet is complete.
330 * Returns NULL on failure by any reason, and pointer
331 * to current nexthdr field in reassembled frame.
333 * It is called with locked fq, and caller must check that
334 * queue is eligible for reassembly i.e. it is not COMPLETE,
335 * the last and the first frames arrived and all the bits are here.
337 static struct sk_buff
*
338 nf_ct_frag6_reasm(struct nf_ct_frag6_queue
*fq
, struct net_device
*dev
)
340 struct sk_buff
*fp
, *op
, *head
= fq
->q
.fragments
;
345 WARN_ON(head
== NULL
);
346 WARN_ON(NFCT_FRAG6_CB(head
)->offset
!= 0);
348 /* Unfragmented part is taken from the first segment. */
349 payload_len
= ((head
->data
- skb_network_header(head
)) -
350 sizeof(struct ipv6hdr
) + fq
->q
.len
-
351 sizeof(struct frag_hdr
));
352 if (payload_len
> IPV6_MAXPLEN
) {
353 pr_debug("payload len is too large.\n");
357 /* Head of list must not be cloned. */
358 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
)) {
359 pr_debug("skb is cloned but can't expand head");
363 /* If the first fragment is fragmented itself, we split
364 * it to two chunks: the first with data and paged part
365 * and the second, holding only fragments. */
366 if (skb_has_frags(head
)) {
367 struct sk_buff
*clone
;
370 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
) {
371 pr_debug("Can't alloc skb\n");
374 clone
->next
= head
->next
;
376 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
377 skb_frag_list_init(head
);
378 for (i
=0; i
<skb_shinfo(head
)->nr_frags
; i
++)
379 plen
+= skb_shinfo(head
)->frags
[i
].size
;
380 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
381 head
->data_len
-= clone
->len
;
382 head
->len
-= clone
->len
;
384 clone
->ip_summed
= head
->ip_summed
;
386 NFCT_FRAG6_CB(clone
)->orig
= NULL
;
387 atomic_add(clone
->truesize
, &nf_init_frags
.mem
);
390 /* We have to remove fragment header from datagram and to relocate
391 * header in order to calculate ICV correctly. */
392 skb_network_header(head
)[fq
->nhoffset
] = skb_transport_header(head
)[0];
393 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
394 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
395 head
->mac_header
+= sizeof(struct frag_hdr
);
396 head
->network_header
+= sizeof(struct frag_hdr
);
398 skb_shinfo(head
)->frag_list
= head
->next
;
399 skb_reset_transport_header(head
);
400 skb_push(head
, head
->data
- skb_network_header(head
));
402 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
403 head
->data_len
+= fp
->len
;
404 head
->len
+= fp
->len
;
405 if (head
->ip_summed
!= fp
->ip_summed
)
406 head
->ip_summed
= CHECKSUM_NONE
;
407 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
408 head
->csum
= csum_add(head
->csum
, fp
->csum
);
409 head
->truesize
+= fp
->truesize
;
411 atomic_sub(head
->truesize
, &nf_init_frags
.mem
);
415 head
->tstamp
= fq
->q
.stamp
;
416 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
418 /* Yes, and fold redundant checksum back. 8) */
419 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
420 head
->csum
= csum_partial(skb_network_header(head
),
421 skb_network_header_len(head
),
424 fq
->q
.fragments
= NULL
;
425 fq
->q
.fragments_tail
= NULL
;
427 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
428 fp
= skb_shinfo(head
)->frag_list
;
429 if (fp
&& NFCT_FRAG6_CB(fp
)->orig
== NULL
)
430 /* at above code, head skb is divided into two skbs. */
433 op
= NFCT_FRAG6_CB(head
)->orig
;
434 for (; fp
; fp
= fp
->next
) {
435 struct sk_buff
*orig
= NFCT_FRAG6_CB(fp
)->orig
;
439 NFCT_FRAG6_CB(fp
)->orig
= NULL
;
446 printk(KERN_DEBUG
"nf_ct_frag6_reasm: payload len = %d\n", payload_len
);
450 printk(KERN_DEBUG
"nf_ct_frag6_reasm: no memory for reassembly\n");
456 * find the header just before Fragment Header.
458 * if success return 0 and set ...
459 * (*prevhdrp): the value of "Next Header Field" in the header
460 * just before Fragment Header.
461 * (*prevhoff): the offset of "Next Header Field" in the header
462 * just before Fragment Header.
463 * (*fhoff) : the offset of Fragment Header.
465 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
469 find_prev_fhdr(struct sk_buff
*skb
, u8
*prevhdrp
, int *prevhoff
, int *fhoff
)
471 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
472 const int netoff
= skb_network_offset(skb
);
473 u8 prev_nhoff
= netoff
+ offsetof(struct ipv6hdr
, nexthdr
);
474 int start
= netoff
+ sizeof(struct ipv6hdr
);
475 int len
= skb
->len
- start
;
476 u8 prevhdr
= NEXTHDR_IPV6
;
478 while (nexthdr
!= NEXTHDR_FRAGMENT
) {
479 struct ipv6_opt_hdr hdr
;
482 if (!ipv6_ext_hdr(nexthdr
)) {
485 if (nexthdr
== NEXTHDR_NONE
) {
486 pr_debug("next header is none\n");
489 if (len
< (int)sizeof(struct ipv6_opt_hdr
)) {
490 pr_debug("too short\n");
493 if (skb_copy_bits(skb
, start
, &hdr
, sizeof(hdr
)))
495 if (nexthdr
== NEXTHDR_AUTH
)
496 hdrlen
= (hdr
.hdrlen
+2)<<2;
498 hdrlen
= ipv6_optlen(&hdr
);
503 nexthdr
= hdr
.nexthdr
;
512 *prevhoff
= prev_nhoff
;
518 struct sk_buff
*nf_ct_frag6_gather(struct sk_buff
*skb
, u32 user
)
520 struct sk_buff
*clone
;
521 struct net_device
*dev
= skb
->dev
;
522 struct frag_hdr
*fhdr
;
523 struct nf_ct_frag6_queue
*fq
;
527 struct sk_buff
*ret_skb
= NULL
;
529 /* Jumbo payload inhibits frag. header */
530 if (ipv6_hdr(skb
)->payload_len
== 0) {
531 pr_debug("payload len = 0\n");
535 if (find_prev_fhdr(skb
, &prevhdr
, &nhoff
, &fhoff
) < 0)
538 clone
= skb_clone(skb
, GFP_ATOMIC
);
540 pr_debug("Can't clone skb\n");
544 NFCT_FRAG6_CB(clone
)->orig
= skb
;
546 if (!pskb_may_pull(clone
, fhoff
+ sizeof(*fhdr
))) {
547 pr_debug("message is too short.\n");
551 skb_set_transport_header(clone
, fhoff
);
552 hdr
= ipv6_hdr(clone
);
553 fhdr
= (struct frag_hdr
*)skb_transport_header(clone
);
555 if (atomic_read(&nf_init_frags
.mem
) > nf_init_frags
.high_thresh
)
556 nf_ct_frag6_evictor();
558 fq
= fq_find(fhdr
->identification
, user
, &hdr
->saddr
, &hdr
->daddr
);
560 pr_debug("Can't find and can't create new queue\n");
564 spin_lock_bh(&fq
->q
.lock
);
566 if (nf_ct_frag6_queue(fq
, clone
, fhdr
, nhoff
) < 0) {
567 spin_unlock_bh(&fq
->q
.lock
);
568 pr_debug("Can't insert skb to queue\n");
573 if (fq
->q
.last_in
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
574 fq
->q
.meat
== fq
->q
.len
) {
575 ret_skb
= nf_ct_frag6_reasm(fq
, dev
);
577 pr_debug("Can't reassemble fragmented packets\n");
579 spin_unlock_bh(&fq
->q
.lock
);
589 void nf_ct_frag6_output(unsigned int hooknum
, struct sk_buff
*skb
,
590 struct net_device
*in
, struct net_device
*out
,
591 int (*okfn
)(struct sk_buff
*))
593 struct sk_buff
*s
, *s2
;
595 for (s
= NFCT_FRAG6_CB(skb
)->orig
; s
;) {
596 nf_conntrack_put_reasm(s
->nfct_reasm
);
597 nf_conntrack_get_reasm(skb
);
603 NF_HOOK_THRESH(NFPROTO_IPV6
, hooknum
, s
, in
, out
, okfn
,
604 NF_IP6_PRI_CONNTRACK_DEFRAG
+ 1);
607 nf_conntrack_put_reasm(skb
);
610 int nf_ct_frag6_init(void)
612 nf_frags
.hashfn
= nf_hashfn
;
613 nf_frags
.constructor
= ip6_frag_init
;
614 nf_frags
.destructor
= NULL
;
615 nf_frags
.skb_free
= nf_skb_free
;
616 nf_frags
.qsize
= sizeof(struct nf_ct_frag6_queue
);
617 nf_frags
.match
= ip6_frag_match
;
618 nf_frags
.frag_expire
= nf_ct_frag6_expire
;
619 nf_frags
.secret_interval
= 10 * 60 * HZ
;
620 nf_init_frags
.timeout
= IPV6_FRAG_TIMEOUT
;
621 nf_init_frags
.high_thresh
= IPV6_FRAG_HIGH_THRESH
;
622 nf_init_frags
.low_thresh
= IPV6_FRAG_LOW_THRESH
;
623 inet_frags_init_net(&nf_init_frags
);
624 inet_frags_init(&nf_frags
);
629 void nf_ct_frag6_cleanup(void)
631 inet_frags_fini(&nf_frags
);
633 nf_init_frags
.low_thresh
= 0;
634 nf_ct_frag6_evictor();