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 static struct inet_frags nf_frags
;
74 static struct netns_frags nf_init_frags
;
77 struct ctl_table nf_ct_ipv6_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 .ctl_name
= NET_NF_CONNTRACK_FRAG6_LOW_THRESH
,
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 .ctl_name
= NET_NF_CONNTRACK_FRAG6_HIGH_THRESH
,
95 .procname
= "nf_conntrack_frag6_high_thresh",
96 .data
= &nf_init_frags
.high_thresh
,
97 .maxlen
= sizeof(unsigned int),
99 .proc_handler
= &proc_dointvec
,
105 static unsigned int ip6qhashfn(__be32 id
, struct in6_addr
*saddr
,
106 struct in6_addr
*daddr
)
110 a
= (__force u32
)saddr
->s6_addr32
[0];
111 b
= (__force u32
)saddr
->s6_addr32
[1];
112 c
= (__force u32
)saddr
->s6_addr32
[2];
114 a
+= JHASH_GOLDEN_RATIO
;
115 b
+= JHASH_GOLDEN_RATIO
;
117 __jhash_mix(a
, b
, c
);
119 a
+= (__force u32
)saddr
->s6_addr32
[3];
120 b
+= (__force u32
)daddr
->s6_addr32
[0];
121 c
+= (__force u32
)daddr
->s6_addr32
[1];
122 __jhash_mix(a
, b
, c
);
124 a
+= (__force u32
)daddr
->s6_addr32
[2];
125 b
+= (__force u32
)daddr
->s6_addr32
[3];
126 c
+= (__force u32
)id
;
127 __jhash_mix(a
, b
, c
);
129 return c
& (INETFRAGS_HASHSZ
- 1);
132 static unsigned int nf_hashfn(struct inet_frag_queue
*q
)
134 struct nf_ct_frag6_queue
*nq
;
136 nq
= container_of(q
, struct nf_ct_frag6_queue
, q
);
137 return ip6qhashfn(nq
->id
, &nq
->saddr
, &nq
->daddr
);
140 static void nf_skb_free(struct sk_buff
*skb
)
142 if (NFCT_FRAG6_CB(skb
)->orig
)
143 kfree_skb(NFCT_FRAG6_CB(skb
)->orig
);
146 /* Memory Tracking Functions. */
147 static inline void frag_kfree_skb(struct sk_buff
*skb
, unsigned int *work
)
150 *work
-= skb
->truesize
;
151 atomic_sub(skb
->truesize
, &nf_init_frags
.mem
);
156 /* Destruction primitives. */
158 static __inline__
void fq_put(struct nf_ct_frag6_queue
*fq
)
160 inet_frag_put(&fq
->q
, &nf_frags
);
163 /* Kill fq entry. It is not destroyed immediately,
164 * because caller (and someone more) holds reference count.
166 static __inline__
void fq_kill(struct nf_ct_frag6_queue
*fq
)
168 inet_frag_kill(&fq
->q
, &nf_frags
);
171 static void nf_ct_frag6_evictor(void)
173 inet_frag_evictor(&nf_init_frags
, &nf_frags
);
176 static void nf_ct_frag6_expire(unsigned long data
)
178 struct nf_ct_frag6_queue
*fq
;
180 fq
= container_of((struct inet_frag_queue
*)data
,
181 struct nf_ct_frag6_queue
, q
);
183 spin_lock(&fq
->q
.lock
);
185 if (fq
->q
.last_in
& COMPLETE
)
191 spin_unlock(&fq
->q
.lock
);
195 /* Creation primitives. */
197 static __inline__
struct nf_ct_frag6_queue
*
198 fq_find(__be32 id
, struct in6_addr
*src
, struct in6_addr
*dst
)
200 struct inet_frag_queue
*q
;
201 struct ip6_create_arg arg
;
207 hash
= ip6qhashfn(id
, src
, dst
);
209 q
= inet_frag_find(&nf_init_frags
, &nf_frags
, &arg
, hash
);
213 return container_of(q
, struct nf_ct_frag6_queue
, q
);
216 pr_debug("Can't alloc new queue\n");
221 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue
*fq
, struct sk_buff
*skb
,
222 struct frag_hdr
*fhdr
, int nhoff
)
224 struct sk_buff
*prev
, *next
;
227 if (fq
->q
.last_in
& COMPLETE
) {
228 pr_debug("Allready completed\n");
232 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
233 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
234 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
236 if ((unsigned int)end
> IPV6_MAXPLEN
) {
237 pr_debug("offset is too large.\n");
241 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
242 const unsigned char *nh
= skb_network_header(skb
);
243 skb
->csum
= csum_sub(skb
->csum
,
244 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
248 /* Is this the final fragment? */
249 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
250 /* If we already have some bits beyond end
251 * or have different end, the segment is corrupted.
253 if (end
< fq
->q
.len
||
254 ((fq
->q
.last_in
& LAST_IN
) && end
!= fq
->q
.len
)) {
255 pr_debug("already received last fragment\n");
258 fq
->q
.last_in
|= LAST_IN
;
261 /* Check if the fragment is rounded to 8 bytes.
262 * Required by the RFC.
265 /* RFC2460 says always send parameter problem in
268 pr_debug("end of fragment not rounded to 8 bytes.\n");
271 if (end
> fq
->q
.len
) {
272 /* Some bits beyond end -> corruption. */
273 if (fq
->q
.last_in
& LAST_IN
) {
274 pr_debug("last packet already reached.\n");
284 /* Point into the IP datagram 'data' part. */
285 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
)) {
286 pr_debug("queue: message is too short.\n");
289 if (pskb_trim_rcsum(skb
, end
- offset
)) {
290 pr_debug("Can't trim\n");
294 /* Find out which fragments are in front and at the back of us
295 * in the chain of fragments so far. We must know where to put
296 * this fragment, right?
299 for (next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
300 if (NFCT_FRAG6_CB(next
)->offset
>= offset
)
305 /* We found where to put this one. Check for overlap with
306 * preceding fragment, and, if needed, align things so that
307 * any overlaps are eliminated.
310 int i
= (NFCT_FRAG6_CB(prev
)->offset
+ prev
->len
) - offset
;
315 pr_debug("overlap\n");
318 if (!pskb_pull(skb
, i
)) {
319 pr_debug("Can't pull\n");
322 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
)
323 skb
->ip_summed
= CHECKSUM_NONE
;
327 /* Look for overlap with succeeding segments.
328 * If we can merge fragments, do it.
330 while (next
&& NFCT_FRAG6_CB(next
)->offset
< end
) {
331 /* overlap is 'i' bytes */
332 int i
= end
- NFCT_FRAG6_CB(next
)->offset
;
335 /* Eat head of the next overlapped fragment
336 * and leave the loop. The next ones cannot overlap.
338 pr_debug("Eat head of the overlapped parts.: %d", i
);
339 if (!pskb_pull(next
, i
))
343 NFCT_FRAG6_CB(next
)->offset
+= i
;
345 if (next
->ip_summed
!= CHECKSUM_UNNECESSARY
)
346 next
->ip_summed
= CHECKSUM_NONE
;
349 struct sk_buff
*free_it
= next
;
351 /* Old fragmnet is completely overridden with
359 fq
->q
.fragments
= next
;
361 fq
->q
.meat
-= free_it
->len
;
362 frag_kfree_skb(free_it
, NULL
);
366 NFCT_FRAG6_CB(skb
)->offset
= offset
;
368 /* Insert this fragment in the chain of fragments. */
373 fq
->q
.fragments
= skb
;
376 fq
->q
.stamp
= skb
->tstamp
;
377 fq
->q
.meat
+= skb
->len
;
378 atomic_add(skb
->truesize
, &nf_init_frags
.mem
);
380 /* The first fragment.
381 * nhoffset is obtained from the first fragment, of course.
384 fq
->nhoffset
= nhoff
;
385 fq
->q
.last_in
|= FIRST_IN
;
387 write_lock(&nf_frags
.lock
);
388 list_move_tail(&fq
->q
.lru_list
, &nf_frags
.lru_list
);
389 write_unlock(&nf_frags
.lock
);
397 * Check if this packet is complete.
398 * Returns NULL on failure by any reason, and pointer
399 * to current nexthdr field in reassembled frame.
401 * It is called with locked fq, and caller must check that
402 * queue is eligible for reassembly i.e. it is not COMPLETE,
403 * the last and the first frames arrived and all the bits are here.
405 static struct sk_buff
*
406 nf_ct_frag6_reasm(struct nf_ct_frag6_queue
*fq
, struct net_device
*dev
)
408 struct sk_buff
*fp
, *op
, *head
= fq
->q
.fragments
;
413 BUG_TRAP(head
!= NULL
);
414 BUG_TRAP(NFCT_FRAG6_CB(head
)->offset
== 0);
416 /* Unfragmented part is taken from the first segment. */
417 payload_len
= ((head
->data
- skb_network_header(head
)) -
418 sizeof(struct ipv6hdr
) + fq
->q
.len
-
419 sizeof(struct frag_hdr
));
420 if (payload_len
> IPV6_MAXPLEN
) {
421 pr_debug("payload len is too large.\n");
425 /* Head of list must not be cloned. */
426 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
)) {
427 pr_debug("skb is cloned but can't expand head");
431 /* If the first fragment is fragmented itself, we split
432 * it to two chunks: the first with data and paged part
433 * and the second, holding only fragments. */
434 if (skb_shinfo(head
)->frag_list
) {
435 struct sk_buff
*clone
;
438 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
) {
439 pr_debug("Can't alloc skb\n");
442 clone
->next
= head
->next
;
444 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
445 skb_shinfo(head
)->frag_list
= NULL
;
446 for (i
=0; i
<skb_shinfo(head
)->nr_frags
; i
++)
447 plen
+= skb_shinfo(head
)->frags
[i
].size
;
448 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
449 head
->data_len
-= clone
->len
;
450 head
->len
-= clone
->len
;
452 clone
->ip_summed
= head
->ip_summed
;
454 NFCT_FRAG6_CB(clone
)->orig
= NULL
;
455 atomic_add(clone
->truesize
, &nf_init_frags
.mem
);
458 /* We have to remove fragment header from datagram and to relocate
459 * header in order to calculate ICV correctly. */
460 skb_network_header(head
)[fq
->nhoffset
] = skb_transport_header(head
)[0];
461 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
462 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
463 head
->mac_header
+= sizeof(struct frag_hdr
);
464 head
->network_header
+= sizeof(struct frag_hdr
);
466 skb_shinfo(head
)->frag_list
= head
->next
;
467 skb_reset_transport_header(head
);
468 skb_push(head
, head
->data
- skb_network_header(head
));
469 atomic_sub(head
->truesize
, &nf_init_frags
.mem
);
471 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
472 head
->data_len
+= fp
->len
;
473 head
->len
+= fp
->len
;
474 if (head
->ip_summed
!= fp
->ip_summed
)
475 head
->ip_summed
= CHECKSUM_NONE
;
476 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
477 head
->csum
= csum_add(head
->csum
, fp
->csum
);
478 head
->truesize
+= fp
->truesize
;
479 atomic_sub(fp
->truesize
, &nf_init_frags
.mem
);
484 head
->tstamp
= fq
->q
.stamp
;
485 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
487 /* Yes, and fold redundant checksum back. 8) */
488 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
489 head
->csum
= csum_partial(skb_network_header(head
),
490 skb_network_header_len(head
),
493 fq
->q
.fragments
= NULL
;
495 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
496 fp
= skb_shinfo(head
)->frag_list
;
497 if (NFCT_FRAG6_CB(fp
)->orig
== NULL
)
498 /* at above code, head skb is divided into two skbs. */
501 op
= NFCT_FRAG6_CB(head
)->orig
;
502 for (; fp
; fp
= fp
->next
) {
503 struct sk_buff
*orig
= NFCT_FRAG6_CB(fp
)->orig
;
507 NFCT_FRAG6_CB(fp
)->orig
= NULL
;
514 printk(KERN_DEBUG
"nf_ct_frag6_reasm: payload len = %d\n", payload_len
);
518 printk(KERN_DEBUG
"nf_ct_frag6_reasm: no memory for reassembly\n");
524 * find the header just before Fragment Header.
526 * if success return 0 and set ...
527 * (*prevhdrp): the value of "Next Header Field" in the header
528 * just before Fragment Header.
529 * (*prevhoff): the offset of "Next Header Field" in the header
530 * just before Fragment Header.
531 * (*fhoff) : the offset of Fragment Header.
533 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
537 find_prev_fhdr(struct sk_buff
*skb
, u8
*prevhdrp
, int *prevhoff
, int *fhoff
)
539 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
540 const int netoff
= skb_network_offset(skb
);
541 u8 prev_nhoff
= netoff
+ offsetof(struct ipv6hdr
, nexthdr
);
542 int start
= netoff
+ sizeof(struct ipv6hdr
);
543 int len
= skb
->len
- start
;
544 u8 prevhdr
= NEXTHDR_IPV6
;
546 while (nexthdr
!= NEXTHDR_FRAGMENT
) {
547 struct ipv6_opt_hdr hdr
;
550 if (!ipv6_ext_hdr(nexthdr
)) {
553 if (len
< (int)sizeof(struct ipv6_opt_hdr
)) {
554 pr_debug("too short\n");
557 if (nexthdr
== NEXTHDR_NONE
) {
558 pr_debug("next header is none\n");
561 if (skb_copy_bits(skb
, start
, &hdr
, sizeof(hdr
)))
563 if (nexthdr
== NEXTHDR_AUTH
)
564 hdrlen
= (hdr
.hdrlen
+2)<<2;
566 hdrlen
= ipv6_optlen(&hdr
);
571 nexthdr
= hdr
.nexthdr
;
580 *prevhoff
= prev_nhoff
;
586 struct sk_buff
*nf_ct_frag6_gather(struct sk_buff
*skb
)
588 struct sk_buff
*clone
;
589 struct net_device
*dev
= skb
->dev
;
590 struct frag_hdr
*fhdr
;
591 struct nf_ct_frag6_queue
*fq
;
595 struct sk_buff
*ret_skb
= NULL
;
597 /* Jumbo payload inhibits frag. header */
598 if (ipv6_hdr(skb
)->payload_len
== 0) {
599 pr_debug("payload len = 0\n");
603 if (find_prev_fhdr(skb
, &prevhdr
, &nhoff
, &fhoff
) < 0)
606 clone
= skb_clone(skb
, GFP_ATOMIC
);
608 pr_debug("Can't clone skb\n");
612 NFCT_FRAG6_CB(clone
)->orig
= skb
;
614 if (!pskb_may_pull(clone
, fhoff
+ sizeof(*fhdr
))) {
615 pr_debug("message is too short.\n");
619 skb_set_transport_header(clone
, fhoff
);
620 hdr
= ipv6_hdr(clone
);
621 fhdr
= (struct frag_hdr
*)skb_transport_header(clone
);
623 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
624 pr_debug("Invalid fragment offset\n");
625 /* It is not a fragmented frame */
629 if (atomic_read(&nf_init_frags
.mem
) > nf_init_frags
.high_thresh
)
630 nf_ct_frag6_evictor();
632 fq
= fq_find(fhdr
->identification
, &hdr
->saddr
, &hdr
->daddr
);
634 pr_debug("Can't find and can't create new queue\n");
638 spin_lock(&fq
->q
.lock
);
640 if (nf_ct_frag6_queue(fq
, clone
, fhdr
, nhoff
) < 0) {
641 spin_unlock(&fq
->q
.lock
);
642 pr_debug("Can't insert skb to queue\n");
647 if (fq
->q
.last_in
== (FIRST_IN
|LAST_IN
) && fq
->q
.meat
== fq
->q
.len
) {
648 ret_skb
= nf_ct_frag6_reasm(fq
, dev
);
650 pr_debug("Can't reassemble fragmented packets\n");
652 spin_unlock(&fq
->q
.lock
);
662 void nf_ct_frag6_output(unsigned int hooknum
, struct sk_buff
*skb
,
663 struct net_device
*in
, struct net_device
*out
,
664 int (*okfn
)(struct sk_buff
*))
666 struct sk_buff
*s
, *s2
;
668 for (s
= NFCT_FRAG6_CB(skb
)->orig
; s
;) {
669 nf_conntrack_put_reasm(s
->nfct_reasm
);
670 nf_conntrack_get_reasm(skb
);
676 NF_HOOK_THRESH(PF_INET6
, hooknum
, s
, in
, out
, okfn
,
677 NF_IP6_PRI_CONNTRACK_DEFRAG
+ 1);
680 nf_conntrack_put_reasm(skb
);
683 int nf_ct_frag6_kfree_frags(struct sk_buff
*skb
)
685 struct sk_buff
*s
, *s2
;
687 for (s
= NFCT_FRAG6_CB(skb
)->orig
; s
; s
= s2
) {
698 int nf_ct_frag6_init(void)
700 nf_frags
.hashfn
= nf_hashfn
;
701 nf_frags
.constructor
= ip6_frag_init
;
702 nf_frags
.destructor
= NULL
;
703 nf_frags
.skb_free
= nf_skb_free
;
704 nf_frags
.qsize
= sizeof(struct nf_ct_frag6_queue
);
705 nf_frags
.match
= ip6_frag_match
;
706 nf_frags
.frag_expire
= nf_ct_frag6_expire
;
707 nf_frags
.secret_interval
= 10 * 60 * HZ
;
708 nf_init_frags
.timeout
= IPV6_FRAG_TIMEOUT
;
709 nf_init_frags
.high_thresh
= 256 * 1024;
710 nf_init_frags
.low_thresh
= 192 * 1024;
711 inet_frags_init_net(&nf_init_frags
);
712 inet_frags_init(&nf_frags
);
717 void nf_ct_frag6_cleanup(void)
719 inet_frags_fini(&nf_frags
);
721 nf_init_frags
.low_thresh
= 0;
722 nf_ct_frag6_evictor();