2 * IPv6 fragment reassembly
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on: net/ipv4/ip_fragment.c
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
18 * Andi Kleen Make it work with multiple hosts.
19 * More RFC compliance.
21 * Horst von Brand Add missing #include <linux/string.h>
22 * Alexey Kuznetsov SMP races, threading, cleanup.
23 * Patrick McHardy LRU queue of frag heads for evictor.
24 * Mitsuru KANDA @USAGI Register inet6_protocol{}.
26 * YOSHIFUJI,H. @USAGI Always remove fragment header to
27 * calculate ICV correctly.
30 #define pr_fmt(fmt) "IPv6: " fmt
32 #include <linux/errno.h>
33 #include <linux/types.h>
34 #include <linux/string.h>
35 #include <linux/socket.h>
36 #include <linux/sockios.h>
37 #include <linux/jiffies.h>
38 #include <linux/net.h>
39 #include <linux/list.h>
40 #include <linux/netdevice.h>
41 #include <linux/in6.h>
42 #include <linux/ipv6.h>
43 #include <linux/icmpv6.h>
44 #include <linux/random.h>
45 #include <linux/jhash.h>
46 #include <linux/skbuff.h>
47 #include <linux/slab.h>
48 #include <linux/export.h>
54 #include <net/ip6_route.h>
55 #include <net/protocol.h>
56 #include <net/transp_v6.h>
57 #include <net/rawv6.h>
58 #include <net/ndisc.h>
59 #include <net/addrconf.h>
60 #include <net/inet_frag.h>
64 struct inet6_skb_parm h
;
68 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
71 static struct inet_frags ip6_frags
;
73 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
74 struct net_device
*dev
);
77 * callers should be careful not to use the hash value outside the ipfrag_lock
78 * as doing so could race with ipfrag_hash_rnd being recalculated.
80 unsigned int inet6_hash_frag(__be32 id
, const struct in6_addr
*saddr
,
81 const struct in6_addr
*daddr
, u32 rnd
)
85 c
= jhash_3words(ipv6_addr_hash(saddr
), ipv6_addr_hash(daddr
),
86 (__force u32
)id
, rnd
);
88 return c
& (INETFRAGS_HASHSZ
- 1);
90 EXPORT_SYMBOL_GPL(inet6_hash_frag
);
92 static unsigned int ip6_hashfn(struct inet_frag_queue
*q
)
94 struct frag_queue
*fq
;
96 fq
= container_of(q
, struct frag_queue
, q
);
97 return inet6_hash_frag(fq
->id
, &fq
->saddr
, &fq
->daddr
, ip6_frags
.rnd
);
100 bool ip6_frag_match(struct inet_frag_queue
*q
, void *a
)
102 struct frag_queue
*fq
;
103 struct ip6_create_arg
*arg
= a
;
105 fq
= container_of(q
, struct frag_queue
, q
);
106 return fq
->id
== arg
->id
&&
107 fq
->user
== arg
->user
&&
108 ipv6_addr_equal(&fq
->saddr
, arg
->src
) &&
109 ipv6_addr_equal(&fq
->daddr
, arg
->dst
);
111 EXPORT_SYMBOL(ip6_frag_match
);
113 void ip6_frag_init(struct inet_frag_queue
*q
, void *a
)
115 struct frag_queue
*fq
= container_of(q
, struct frag_queue
, q
);
116 struct ip6_create_arg
*arg
= a
;
119 fq
->user
= arg
->user
;
120 fq
->saddr
= *arg
->src
;
121 fq
->daddr
= *arg
->dst
;
123 EXPORT_SYMBOL(ip6_frag_init
);
125 void ip6_expire_frag_queue(struct net
*net
, struct frag_queue
*fq
,
126 struct inet_frags
*frags
)
128 struct net_device
*dev
= NULL
;
130 spin_lock(&fq
->q
.lock
);
132 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
135 inet_frag_kill(&fq
->q
, frags
);
138 dev
= dev_get_by_index_rcu(net
, fq
->iif
);
142 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMTIMEOUT
);
143 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
145 /* Don't send error if the first segment did not arrive. */
146 if (!(fq
->q
.last_in
& INET_FRAG_FIRST_IN
) || !fq
->q
.fragments
)
150 But use as source device on which LAST ARRIVED
151 segment was received. And do not use fq->dev
152 pointer directly, device might already disappeared.
154 fq
->q
.fragments
->dev
= dev
;
155 icmpv6_send(fq
->q
.fragments
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_FRAGTIME
, 0);
159 spin_unlock(&fq
->q
.lock
);
160 inet_frag_put(&fq
->q
, frags
);
162 EXPORT_SYMBOL(ip6_expire_frag_queue
);
164 static void ip6_frag_expire(unsigned long data
)
166 struct frag_queue
*fq
;
169 fq
= container_of((struct inet_frag_queue
*)data
, struct frag_queue
, q
);
170 net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
172 ip6_expire_frag_queue(net
, fq
, &ip6_frags
);
175 static __inline__
struct frag_queue
*
176 fq_find(struct net
*net
, __be32 id
, const struct in6_addr
*src
, const struct in6_addr
*dst
)
178 struct inet_frag_queue
*q
;
179 struct ip6_create_arg arg
;
183 arg
.user
= IP6_DEFRAG_LOCAL_DELIVER
;
187 read_lock(&ip6_frags
.lock
);
188 hash
= inet6_hash_frag(id
, src
, dst
, ip6_frags
.rnd
);
190 q
= inet_frag_find(&net
->ipv6
.frags
, &ip6_frags
, &arg
, hash
);
191 if (IS_ERR_OR_NULL(q
)) {
192 inet_frag_maybe_warn_overflow(q
, pr_fmt());
195 return container_of(q
, struct frag_queue
, q
);
198 static int ip6_frag_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
199 struct frag_hdr
*fhdr
, int nhoff
)
201 struct sk_buff
*prev
, *next
;
202 struct net_device
*dev
;
204 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
206 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
209 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
210 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
211 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
213 if ((unsigned int)end
> IPV6_MAXPLEN
) {
214 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
215 IPSTATS_MIB_INHDRERRORS
);
216 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
217 ((u8
*)&fhdr
->frag_off
-
218 skb_network_header(skb
)));
222 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
223 const unsigned char *nh
= skb_network_header(skb
);
224 skb
->csum
= csum_sub(skb
->csum
,
225 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
229 /* Is this the final fragment? */
230 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
231 /* If we already have some bits beyond end
232 * or have different end, the segment is corrupted.
234 if (end
< fq
->q
.len
||
235 ((fq
->q
.last_in
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
))
237 fq
->q
.last_in
|= INET_FRAG_LAST_IN
;
240 /* Check if the fragment is rounded to 8 bytes.
241 * Required by the RFC.
244 /* RFC2460 says always send parameter problem in
247 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
248 IPSTATS_MIB_INHDRERRORS
);
249 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
250 offsetof(struct ipv6hdr
, payload_len
));
253 if (end
> fq
->q
.len
) {
254 /* Some bits beyond end -> corruption. */
255 if (fq
->q
.last_in
& INET_FRAG_LAST_IN
)
264 /* Point into the IP datagram 'data' part. */
265 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
))
268 if (pskb_trim_rcsum(skb
, end
- offset
))
271 /* Find out which fragments are in front and at the back of us
272 * in the chain of fragments so far. We must know where to put
273 * this fragment, right?
275 prev
= fq
->q
.fragments_tail
;
276 if (!prev
|| FRAG6_CB(prev
)->offset
< offset
) {
281 for(next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
282 if (FRAG6_CB(next
)->offset
>= offset
)
288 /* RFC5722, Section 4, amended by Errata ID : 3089
289 * When reassembling an IPv6 datagram, if
290 * one or more its constituent fragments is determined to be an
291 * overlapping fragment, the entire datagram (and any constituent
292 * fragments) MUST be silently discarded.
295 /* Check for overlap with preceding fragment. */
297 (FRAG6_CB(prev
)->offset
+ prev
->len
) > offset
)
300 /* Look for overlap with succeeding segment. */
301 if (next
&& FRAG6_CB(next
)->offset
< end
)
304 FRAG6_CB(skb
)->offset
= offset
;
306 /* Insert this fragment in the chain of fragments. */
309 fq
->q
.fragments_tail
= skb
;
313 fq
->q
.fragments
= skb
;
317 fq
->iif
= dev
->ifindex
;
320 fq
->q
.stamp
= skb
->tstamp
;
321 fq
->q
.meat
+= skb
->len
;
322 add_frag_mem_limit(&fq
->q
, skb
->truesize
);
324 /* The first fragment.
325 * nhoffset is obtained from the first fragment, of course.
328 fq
->nhoffset
= nhoff
;
329 fq
->q
.last_in
|= INET_FRAG_FIRST_IN
;
332 if (fq
->q
.last_in
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
333 fq
->q
.meat
== fq
->q
.len
)
334 return ip6_frag_reasm(fq
, prev
, dev
);
336 inet_frag_lru_move(&fq
->q
);
340 inet_frag_kill(&fq
->q
, &ip6_frags
);
342 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
343 IPSTATS_MIB_REASMFAILS
);
349 * Check if this packet is complete.
350 * Returns NULL on failure by any reason, and pointer
351 * to current nexthdr field in reassembled frame.
353 * It is called with locked fq, and caller must check that
354 * queue is eligible for reassembly i.e. it is not COMPLETE,
355 * the last and the first frames arrived and all the bits are here.
357 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
358 struct net_device
*dev
)
360 struct net
*net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
361 struct sk_buff
*fp
, *head
= fq
->q
.fragments
;
366 inet_frag_kill(&fq
->q
, &ip6_frags
);
368 /* Make the one we just received the head. */
371 fp
= skb_clone(head
, GFP_ATOMIC
);
376 fp
->next
= head
->next
;
378 fq
->q
.fragments_tail
= fp
;
381 skb_morph(head
, fq
->q
.fragments
);
382 head
->next
= fq
->q
.fragments
->next
;
384 consume_skb(fq
->q
.fragments
);
385 fq
->q
.fragments
= head
;
388 WARN_ON(head
== NULL
);
389 WARN_ON(FRAG6_CB(head
)->offset
!= 0);
391 /* Unfragmented part is taken from the first segment. */
392 payload_len
= ((head
->data
- skb_network_header(head
)) -
393 sizeof(struct ipv6hdr
) + fq
->q
.len
-
394 sizeof(struct frag_hdr
));
395 if (payload_len
> IPV6_MAXPLEN
)
398 /* Head of list must not be cloned. */
399 if (skb_unclone(head
, GFP_ATOMIC
))
402 /* If the first fragment is fragmented itself, we split
403 * it to two chunks: the first with data and paged part
404 * and the second, holding only fragments. */
405 if (skb_has_frag_list(head
)) {
406 struct sk_buff
*clone
;
409 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
)
411 clone
->next
= head
->next
;
413 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
414 skb_frag_list_init(head
);
415 for (i
= 0; i
< skb_shinfo(head
)->nr_frags
; i
++)
416 plen
+= skb_frag_size(&skb_shinfo(head
)->frags
[i
]);
417 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
418 head
->data_len
-= clone
->len
;
419 head
->len
-= clone
->len
;
421 clone
->ip_summed
= head
->ip_summed
;
422 add_frag_mem_limit(&fq
->q
, clone
->truesize
);
425 /* We have to remove fragment header from datagram and to relocate
426 * header in order to calculate ICV correctly. */
427 nhoff
= fq
->nhoffset
;
428 skb_network_header(head
)[nhoff
] = skb_transport_header(head
)[0];
429 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
430 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
431 head
->mac_header
+= sizeof(struct frag_hdr
);
432 head
->network_header
+= sizeof(struct frag_hdr
);
434 skb_reset_transport_header(head
);
435 skb_push(head
, head
->data
- skb_network_header(head
));
437 sum_truesize
= head
->truesize
;
438 for (fp
= head
->next
; fp
;) {
441 struct sk_buff
*next
= fp
->next
;
443 sum_truesize
+= fp
->truesize
;
444 if (head
->ip_summed
!= fp
->ip_summed
)
445 head
->ip_summed
= CHECKSUM_NONE
;
446 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
447 head
->csum
= csum_add(head
->csum
, fp
->csum
);
449 if (skb_try_coalesce(head
, fp
, &headstolen
, &delta
)) {
450 kfree_skb_partial(fp
, headstolen
);
452 if (!skb_shinfo(head
)->frag_list
)
453 skb_shinfo(head
)->frag_list
= fp
;
454 head
->data_len
+= fp
->len
;
455 head
->len
+= fp
->len
;
456 head
->truesize
+= fp
->truesize
;
460 sub_frag_mem_limit(&fq
->q
, sum_truesize
);
464 head
->tstamp
= fq
->q
.stamp
;
465 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
466 IP6CB(head
)->nhoff
= nhoff
;
468 /* Yes, and fold redundant checksum back. 8) */
469 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
470 head
->csum
= csum_partial(skb_network_header(head
),
471 skb_network_header_len(head
),
475 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMOKS
);
477 fq
->q
.fragments
= NULL
;
478 fq
->q
.fragments_tail
= NULL
;
482 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len
);
485 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
488 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
493 static int ipv6_frag_rcv(struct sk_buff
*skb
)
495 struct frag_hdr
*fhdr
;
496 struct frag_queue
*fq
;
497 const struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
498 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
501 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMREQDS
);
503 /* Jumbo payload inhibits frag. header */
504 if (hdr
->payload_len
==0)
507 if (!pskb_may_pull(skb
, (skb_transport_offset(skb
) +
508 sizeof(struct frag_hdr
))))
512 fhdr
= (struct frag_hdr
*)skb_transport_header(skb
);
514 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
515 /* It is not a fragmented frame */
516 skb
->transport_header
+= sizeof(struct frag_hdr
);
517 IP6_INC_STATS_BH(net
,
518 ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMOKS
);
520 IP6CB(skb
)->nhoff
= (u8
*)fhdr
- skb_network_header(skb
);
524 evicted
= inet_frag_evictor(&net
->ipv6
.frags
, &ip6_frags
, false);
526 IP6_ADD_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
527 IPSTATS_MIB_REASMFAILS
, evicted
);
529 fq
= fq_find(net
, fhdr
->identification
, &hdr
->saddr
, &hdr
->daddr
);
533 spin_lock(&fq
->q
.lock
);
535 ret
= ip6_frag_queue(fq
, skb
, fhdr
, IP6CB(skb
)->nhoff
);
537 spin_unlock(&fq
->q
.lock
);
538 inet_frag_put(&fq
->q
, &ip6_frags
);
542 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMFAILS
);
547 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_INHDRERRORS
);
548 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb_network_header_len(skb
));
552 static const struct inet6_protocol frag_protocol
=
554 .handler
= ipv6_frag_rcv
,
555 .flags
= INET6_PROTO_NOPOLICY
,
559 static struct ctl_table ip6_frags_ns_ctl_table
[] = {
561 .procname
= "ip6frag_high_thresh",
562 .data
= &init_net
.ipv6
.frags
.high_thresh
,
563 .maxlen
= sizeof(int),
565 .proc_handler
= proc_dointvec
568 .procname
= "ip6frag_low_thresh",
569 .data
= &init_net
.ipv6
.frags
.low_thresh
,
570 .maxlen
= sizeof(int),
572 .proc_handler
= proc_dointvec
575 .procname
= "ip6frag_time",
576 .data
= &init_net
.ipv6
.frags
.timeout
,
577 .maxlen
= sizeof(int),
579 .proc_handler
= proc_dointvec_jiffies
,
584 static struct ctl_table ip6_frags_ctl_table
[] = {
586 .procname
= "ip6frag_secret_interval",
587 .data
= &ip6_frags
.secret_interval
,
588 .maxlen
= sizeof(int),
590 .proc_handler
= proc_dointvec_jiffies
,
595 static int __net_init
ip6_frags_ns_sysctl_register(struct net
*net
)
597 struct ctl_table
*table
;
598 struct ctl_table_header
*hdr
;
600 table
= ip6_frags_ns_ctl_table
;
601 if (!net_eq(net
, &init_net
)) {
602 table
= kmemdup(table
, sizeof(ip6_frags_ns_ctl_table
), GFP_KERNEL
);
606 table
[0].data
= &net
->ipv6
.frags
.high_thresh
;
607 table
[1].data
= &net
->ipv6
.frags
.low_thresh
;
608 table
[2].data
= &net
->ipv6
.frags
.timeout
;
610 /* Don't export sysctls to unprivileged users */
611 if (net
->user_ns
!= &init_user_ns
)
612 table
[0].procname
= NULL
;
615 hdr
= register_net_sysctl(net
, "net/ipv6", table
);
619 net
->ipv6
.sysctl
.frags_hdr
= hdr
;
623 if (!net_eq(net
, &init_net
))
629 static void __net_exit
ip6_frags_ns_sysctl_unregister(struct net
*net
)
631 struct ctl_table
*table
;
633 table
= net
->ipv6
.sysctl
.frags_hdr
->ctl_table_arg
;
634 unregister_net_sysctl_table(net
->ipv6
.sysctl
.frags_hdr
);
635 if (!net_eq(net
, &init_net
))
639 static struct ctl_table_header
*ip6_ctl_header
;
641 static int ip6_frags_sysctl_register(void)
643 ip6_ctl_header
= register_net_sysctl(&init_net
, "net/ipv6",
644 ip6_frags_ctl_table
);
645 return ip6_ctl_header
== NULL
? -ENOMEM
: 0;
648 static void ip6_frags_sysctl_unregister(void)
650 unregister_net_sysctl_table(ip6_ctl_header
);
653 static inline int ip6_frags_ns_sysctl_register(struct net
*net
)
658 static inline void ip6_frags_ns_sysctl_unregister(struct net
*net
)
662 static inline int ip6_frags_sysctl_register(void)
667 static inline void ip6_frags_sysctl_unregister(void)
672 static int __net_init
ipv6_frags_init_net(struct net
*net
)
674 net
->ipv6
.frags
.high_thresh
= IPV6_FRAG_HIGH_THRESH
;
675 net
->ipv6
.frags
.low_thresh
= IPV6_FRAG_LOW_THRESH
;
676 net
->ipv6
.frags
.timeout
= IPV6_FRAG_TIMEOUT
;
678 inet_frags_init_net(&net
->ipv6
.frags
);
680 return ip6_frags_ns_sysctl_register(net
);
683 static void __net_exit
ipv6_frags_exit_net(struct net
*net
)
685 ip6_frags_ns_sysctl_unregister(net
);
686 inet_frags_exit_net(&net
->ipv6
.frags
, &ip6_frags
);
689 static struct pernet_operations ip6_frags_ops
= {
690 .init
= ipv6_frags_init_net
,
691 .exit
= ipv6_frags_exit_net
,
694 int __init
ipv6_frag_init(void)
698 ret
= inet6_add_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
702 ret
= ip6_frags_sysctl_register();
706 ret
= register_pernet_subsys(&ip6_frags_ops
);
710 ip6_frags
.hashfn
= ip6_hashfn
;
711 ip6_frags
.constructor
= ip6_frag_init
;
712 ip6_frags
.destructor
= NULL
;
713 ip6_frags
.skb_free
= NULL
;
714 ip6_frags
.qsize
= sizeof(struct frag_queue
);
715 ip6_frags
.match
= ip6_frag_match
;
716 ip6_frags
.frag_expire
= ip6_frag_expire
;
717 ip6_frags
.secret_interval
= 10 * 60 * HZ
;
718 inet_frags_init(&ip6_frags
);
723 ip6_frags_sysctl_unregister();
725 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
729 void ipv6_frag_exit(void)
731 inet_frags_fini(&ip6_frags
);
732 ip6_frags_sysctl_unregister();
733 unregister_pernet_subsys(&ip6_frags_ops
);
734 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);