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 #define pr_fmt(fmt) "IPv6-nf: " fmt
19 #include <linux/errno.h>
20 #include <linux/types.h>
21 #include <linux/string.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/jiffies.h>
25 #include <linux/net.h>
26 #include <linux/list.h>
27 #include <linux/netdevice.h>
28 #include <linux/in6.h>
29 #include <linux/ipv6.h>
30 #include <linux/icmpv6.h>
31 #include <linux/random.h>
32 #include <linux/slab.h>
36 #include <net/inet_frag.h>
39 #include <net/protocol.h>
40 #include <net/transp_v6.h>
41 #include <net/rawv6.h>
42 #include <net/ndisc.h>
43 #include <net/addrconf.h>
44 #include <net/inet_ecn.h>
45 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
46 #include <linux/sysctl.h>
47 #include <linux/netfilter.h>
48 #include <linux/netfilter_ipv6.h>
49 #include <linux/kernel.h>
50 #include <linux/module.h>
51 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
53 static const char nf_frags_cache_name
[] = "nf-frags";
55 struct nf_ct_frag6_skb_cb
57 struct inet6_skb_parm h
;
62 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
64 static struct inet_frags nf_frags
;
69 static struct ctl_table nf_ct_frag6_sysctl_table
[] = {
71 .procname
= "nf_conntrack_frag6_timeout",
72 .data
= &init_net
.nf_frag
.frags
.timeout
,
73 .maxlen
= sizeof(unsigned int),
75 .proc_handler
= proc_dointvec_jiffies
,
78 .procname
= "nf_conntrack_frag6_low_thresh",
79 .data
= &init_net
.nf_frag
.frags
.low_thresh
,
80 .maxlen
= sizeof(unsigned int),
82 .proc_handler
= proc_dointvec_minmax
,
84 .extra2
= &init_net
.nf_frag
.frags
.high_thresh
87 .procname
= "nf_conntrack_frag6_high_thresh",
88 .data
= &init_net
.nf_frag
.frags
.high_thresh
,
89 .maxlen
= sizeof(unsigned int),
91 .proc_handler
= proc_dointvec_minmax
,
92 .extra1
= &init_net
.nf_frag
.frags
.low_thresh
97 static int nf_ct_frag6_sysctl_register(struct net
*net
)
99 struct ctl_table
*table
;
100 struct ctl_table_header
*hdr
;
102 table
= nf_ct_frag6_sysctl_table
;
103 if (!net_eq(net
, &init_net
)) {
104 table
= kmemdup(table
, sizeof(nf_ct_frag6_sysctl_table
),
109 table
[0].data
= &net
->nf_frag
.frags
.timeout
;
110 table
[1].data
= &net
->nf_frag
.frags
.low_thresh
;
111 table
[1].extra2
= &net
->nf_frag
.frags
.high_thresh
;
112 table
[2].data
= &net
->nf_frag
.frags
.high_thresh
;
113 table
[2].extra1
= &net
->nf_frag
.frags
.low_thresh
;
114 table
[2].extra2
= &init_net
.nf_frag
.frags
.high_thresh
;
117 hdr
= register_net_sysctl(net
, "net/netfilter", table
);
121 net
->nf_frag
.sysctl
.frags_hdr
= hdr
;
125 if (!net_eq(net
, &init_net
))
131 static void __net_exit
nf_ct_frags6_sysctl_unregister(struct net
*net
)
133 struct ctl_table
*table
;
135 table
= net
->nf_frag
.sysctl
.frags_hdr
->ctl_table_arg
;
136 unregister_net_sysctl_table(net
->nf_frag
.sysctl
.frags_hdr
);
137 if (!net_eq(net
, &init_net
))
142 static int nf_ct_frag6_sysctl_register(struct net
*net
)
146 static void __net_exit
nf_ct_frags6_sysctl_unregister(struct net
*net
)
151 static inline u8
ip6_frag_ecn(const struct ipv6hdr
*ipv6h
)
153 return 1 << (ipv6_get_dsfield(ipv6h
) & INET_ECN_MASK
);
156 static unsigned int nf_hash_frag(__be32 id
, const struct in6_addr
*saddr
,
157 const struct in6_addr
*daddr
)
159 net_get_random_once(&nf_frags
.rnd
, sizeof(nf_frags
.rnd
));
160 return jhash_3words(ipv6_addr_hash(saddr
), ipv6_addr_hash(daddr
),
161 (__force u32
)id
, nf_frags
.rnd
);
165 static unsigned int nf_hashfn(const struct inet_frag_queue
*q
)
167 const struct frag_queue
*nq
;
169 nq
= container_of(q
, struct frag_queue
, q
);
170 return nf_hash_frag(nq
->id
, &nq
->saddr
, &nq
->daddr
);
173 static void nf_skb_free(struct sk_buff
*skb
)
175 if (NFCT_FRAG6_CB(skb
)->orig
)
176 kfree_skb(NFCT_FRAG6_CB(skb
)->orig
);
179 static void nf_ct_frag6_expire(unsigned long data
)
181 struct frag_queue
*fq
;
184 fq
= container_of((struct inet_frag_queue
*)data
, struct frag_queue
, q
);
185 net
= container_of(fq
->q
.net
, struct net
, nf_frag
.frags
);
187 ip6_expire_frag_queue(net
, fq
, &nf_frags
);
190 /* Creation primitives. */
191 static inline struct frag_queue
*fq_find(struct net
*net
, __be32 id
,
192 u32 user
, struct in6_addr
*src
,
193 struct in6_addr
*dst
, u8 ecn
)
195 struct inet_frag_queue
*q
;
196 struct ip6_create_arg arg
;
206 hash
= nf_hash_frag(id
, src
, dst
);
208 q
= inet_frag_find(&net
->nf_frag
.frags
, &nf_frags
, &arg
, hash
);
210 if (IS_ERR_OR_NULL(q
)) {
211 inet_frag_maybe_warn_overflow(q
, pr_fmt());
214 return container_of(q
, struct frag_queue
, q
);
218 static int nf_ct_frag6_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
219 const struct frag_hdr
*fhdr
, int nhoff
)
221 struct sk_buff
*prev
, *next
;
222 unsigned int payload_len
;
226 if (fq
->q
.flags
& INET_FRAG_COMPLETE
) {
227 pr_debug("Already completed\n");
231 payload_len
= ntohs(ipv6_hdr(skb
)->payload_len
);
233 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
234 end
= offset
+ (payload_len
-
235 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
237 if ((unsigned int)end
> IPV6_MAXPLEN
) {
238 pr_debug("offset is too large.\n");
242 ecn
= ip6_frag_ecn(ipv6_hdr(skb
));
244 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
245 const unsigned char *nh
= skb_network_header(skb
);
246 skb
->csum
= csum_sub(skb
->csum
,
247 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
251 /* Is this the final fragment? */
252 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
253 /* If we already have some bits beyond end
254 * or have different end, the segment is corrupted.
256 if (end
< fq
->q
.len
||
257 ((fq
->q
.flags
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
)) {
258 pr_debug("already received last fragment\n");
261 fq
->q
.flags
|= INET_FRAG_LAST_IN
;
264 /* Check if the fragment is rounded to 8 bytes.
265 * Required by the RFC.
268 /* RFC2460 says always send parameter problem in
271 pr_debug("end of fragment not rounded to 8 bytes.\n");
274 if (end
> fq
->q
.len
) {
275 /* Some bits beyond end -> corruption. */
276 if (fq
->q
.flags
& INET_FRAG_LAST_IN
) {
277 pr_debug("last packet already reached.\n");
287 /* Point into the IP datagram 'data' part. */
288 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
)) {
289 pr_debug("queue: message is too short.\n");
292 if (pskb_trim_rcsum(skb
, end
- offset
)) {
293 pr_debug("Can't trim\n");
297 /* Find out which fragments are in front and at the back of us
298 * in the chain of fragments so far. We must know where to put
299 * this fragment, right?
301 prev
= fq
->q
.fragments_tail
;
302 if (!prev
|| NFCT_FRAG6_CB(prev
)->offset
< offset
) {
307 for (next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
308 if (NFCT_FRAG6_CB(next
)->offset
>= offset
)
314 /* RFC5722, Section 4:
315 * When reassembling an IPv6 datagram, if
316 * one or more its constituent fragments is determined to be an
317 * overlapping fragment, the entire datagram (and any constituent
318 * fragments, including those not yet received) MUST be silently
322 /* Check for overlap with preceding fragment. */
324 (NFCT_FRAG6_CB(prev
)->offset
+ prev
->len
) > offset
)
327 /* Look for overlap with succeeding segment. */
328 if (next
&& NFCT_FRAG6_CB(next
)->offset
< end
)
331 NFCT_FRAG6_CB(skb
)->offset
= offset
;
333 /* Insert this fragment in the chain of fragments. */
336 fq
->q
.fragments_tail
= skb
;
340 fq
->q
.fragments
= skb
;
343 fq
->iif
= skb
->dev
->ifindex
;
346 fq
->q
.stamp
= skb
->tstamp
;
347 fq
->q
.meat
+= skb
->len
;
349 if (payload_len
> fq
->q
.max_size
)
350 fq
->q
.max_size
= payload_len
;
351 add_frag_mem_limit(&fq
->q
, skb
->truesize
);
353 /* The first fragment.
354 * nhoffset is obtained from the first fragment, of course.
357 fq
->nhoffset
= nhoff
;
358 fq
->q
.flags
|= INET_FRAG_FIRST_IN
;
364 inet_frag_kill(&fq
->q
, &nf_frags
);
370 * Check if this packet is complete.
371 * Returns NULL on failure by any reason, and pointer
372 * to current nexthdr field in reassembled frame.
374 * It is called with locked fq, and caller must check that
375 * queue is eligible for reassembly i.e. it is not COMPLETE,
376 * the last and the first frames arrived and all the bits are here.
378 static struct sk_buff
*
379 nf_ct_frag6_reasm(struct frag_queue
*fq
, struct net_device
*dev
)
381 struct sk_buff
*fp
, *op
, *head
= fq
->q
.fragments
;
385 inet_frag_kill(&fq
->q
, &nf_frags
);
387 WARN_ON(head
== NULL
);
388 WARN_ON(NFCT_FRAG6_CB(head
)->offset
!= 0);
390 ecn
= ip_frag_ecn_table
[fq
->ecn
];
391 if (unlikely(ecn
== 0xff))
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_unclone(head
, 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_frag_list(head
)) {
413 struct sk_buff
*clone
;
416 clone
= alloc_skb(0, GFP_ATOMIC
);
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_frag_size(&skb_shinfo(head
)->frags
[i
]);
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 add_frag_mem_limit(&fq
->q
, clone
->truesize
);
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
));
448 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
449 head
->data_len
+= fp
->len
;
450 head
->len
+= fp
->len
;
451 if (head
->ip_summed
!= fp
->ip_summed
)
452 head
->ip_summed
= CHECKSUM_NONE
;
453 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
454 head
->csum
= csum_add(head
->csum
, fp
->csum
);
455 head
->truesize
+= fp
->truesize
;
457 sub_frag_mem_limit(&fq
->q
, head
->truesize
);
462 head
->tstamp
= fq
->q
.stamp
;
463 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
464 ipv6_change_dsfield(ipv6_hdr(head
), 0xff, ecn
);
465 IP6CB(head
)->frag_max_size
= sizeof(struct ipv6hdr
) + fq
->q
.max_size
;
467 /* Yes, and fold redundant checksum back. 8) */
468 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
469 head
->csum
= csum_partial(skb_network_header(head
),
470 skb_network_header_len(head
),
473 fq
->q
.fragments
= NULL
;
474 fq
->q
.fragments_tail
= NULL
;
476 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
477 fp
= skb_shinfo(head
)->frag_list
;
478 if (fp
&& NFCT_FRAG6_CB(fp
)->orig
== NULL
)
479 /* at above code, head skb is divided into two skbs. */
482 op
= NFCT_FRAG6_CB(head
)->orig
;
483 for (; fp
; fp
= fp
->next
) {
484 struct sk_buff
*orig
= NFCT_FRAG6_CB(fp
)->orig
;
488 NFCT_FRAG6_CB(fp
)->orig
= NULL
;
494 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
498 net_dbg_ratelimited("nf_ct_frag6_reasm: no memory for reassembly\n");
504 * find the header just before Fragment Header.
506 * if success return 0 and set ...
507 * (*prevhdrp): the value of "Next Header Field" in the header
508 * just before Fragment Header.
509 * (*prevhoff): the offset of "Next Header Field" in the header
510 * just before Fragment Header.
511 * (*fhoff) : the offset of Fragment Header.
513 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
517 find_prev_fhdr(struct sk_buff
*skb
, u8
*prevhdrp
, int *prevhoff
, int *fhoff
)
519 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
520 const int netoff
= skb_network_offset(skb
);
521 u8 prev_nhoff
= netoff
+ offsetof(struct ipv6hdr
, nexthdr
);
522 int start
= netoff
+ sizeof(struct ipv6hdr
);
523 int len
= skb
->len
- start
;
524 u8 prevhdr
= NEXTHDR_IPV6
;
526 while (nexthdr
!= NEXTHDR_FRAGMENT
) {
527 struct ipv6_opt_hdr hdr
;
530 if (!ipv6_ext_hdr(nexthdr
)) {
533 if (nexthdr
== NEXTHDR_NONE
) {
534 pr_debug("next header is none\n");
537 if (len
< (int)sizeof(struct ipv6_opt_hdr
)) {
538 pr_debug("too short\n");
541 if (skb_copy_bits(skb
, start
, &hdr
, sizeof(hdr
)))
543 if (nexthdr
== NEXTHDR_AUTH
)
544 hdrlen
= (hdr
.hdrlen
+2)<<2;
546 hdrlen
= ipv6_optlen(&hdr
);
551 nexthdr
= hdr
.nexthdr
;
560 *prevhoff
= prev_nhoff
;
566 struct sk_buff
*nf_ct_frag6_gather(struct sk_buff
*skb
, u32 user
)
568 struct sk_buff
*clone
;
569 struct net_device
*dev
= skb
->dev
;
570 struct net
*net
= skb_dst(skb
) ? dev_net(skb_dst(skb
)->dev
)
572 struct frag_hdr
*fhdr
;
573 struct frag_queue
*fq
;
577 struct sk_buff
*ret_skb
= NULL
;
579 /* Jumbo payload inhibits frag. header */
580 if (ipv6_hdr(skb
)->payload_len
== 0) {
581 pr_debug("payload len = 0\n");
585 if (find_prev_fhdr(skb
, &prevhdr
, &nhoff
, &fhoff
) < 0)
588 clone
= skb_clone(skb
, GFP_ATOMIC
);
590 pr_debug("Can't clone skb\n");
594 NFCT_FRAG6_CB(clone
)->orig
= skb
;
596 if (!pskb_may_pull(clone
, fhoff
+ sizeof(*fhdr
))) {
597 pr_debug("message is too short.\n");
601 skb_set_transport_header(clone
, fhoff
);
602 hdr
= ipv6_hdr(clone
);
603 fhdr
= (struct frag_hdr
*)skb_transport_header(clone
);
605 fq
= fq_find(net
, fhdr
->identification
, user
, &hdr
->saddr
, &hdr
->daddr
,
608 pr_debug("Can't find and can't create new queue\n");
612 spin_lock_bh(&fq
->q
.lock
);
614 if (nf_ct_frag6_queue(fq
, clone
, fhdr
, nhoff
) < 0) {
615 spin_unlock_bh(&fq
->q
.lock
);
616 pr_debug("Can't insert skb to queue\n");
617 inet_frag_put(&fq
->q
, &nf_frags
);
621 if (fq
->q
.flags
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
622 fq
->q
.meat
== fq
->q
.len
) {
623 ret_skb
= nf_ct_frag6_reasm(fq
, dev
);
625 pr_debug("Can't reassemble fragmented packets\n");
627 spin_unlock_bh(&fq
->q
.lock
);
629 inet_frag_put(&fq
->q
, &nf_frags
);
637 void nf_ct_frag6_consume_orig(struct sk_buff
*skb
)
639 struct sk_buff
*s
, *s2
;
641 for (s
= NFCT_FRAG6_CB(skb
)->orig
; s
;) {
649 static int nf_ct_net_init(struct net
*net
)
651 net
->nf_frag
.frags
.high_thresh
= IPV6_FRAG_HIGH_THRESH
;
652 net
->nf_frag
.frags
.low_thresh
= IPV6_FRAG_LOW_THRESH
;
653 net
->nf_frag
.frags
.timeout
= IPV6_FRAG_TIMEOUT
;
654 inet_frags_init_net(&net
->nf_frag
.frags
);
656 return nf_ct_frag6_sysctl_register(net
);
659 static void nf_ct_net_exit(struct net
*net
)
661 nf_ct_frags6_sysctl_unregister(net
);
662 inet_frags_exit_net(&net
->nf_frag
.frags
, &nf_frags
);
665 static struct pernet_operations nf_ct_net_ops
= {
666 .init
= nf_ct_net_init
,
667 .exit
= nf_ct_net_exit
,
670 int nf_ct_frag6_init(void)
674 nf_frags
.hashfn
= nf_hashfn
;
675 nf_frags
.constructor
= ip6_frag_init
;
676 nf_frags
.destructor
= NULL
;
677 nf_frags
.skb_free
= nf_skb_free
;
678 nf_frags
.qsize
= sizeof(struct frag_queue
);
679 nf_frags
.match
= ip6_frag_match
;
680 nf_frags
.frag_expire
= nf_ct_frag6_expire
;
681 nf_frags
.frags_cache_name
= nf_frags_cache_name
;
682 ret
= inet_frags_init(&nf_frags
);
685 ret
= register_pernet_subsys(&nf_ct_net_ops
);
687 inet_frags_fini(&nf_frags
);
693 void nf_ct_frag6_cleanup(void)
695 unregister_pernet_subsys(&nf_ct_net_ops
);
696 inet_frags_fini(&nf_frags
);