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>
61 #include <net/inet_ecn.h>
65 struct inet6_skb_parm h
;
69 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
71 static inline u8
ip6_frag_ecn(const struct ipv6hdr
*ipv6h
)
73 return 1 << (ipv6_get_dsfield(ipv6h
) & INET_ECN_MASK
);
76 static struct inet_frags ip6_frags
;
78 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
79 struct net_device
*dev
);
82 * callers should be careful not to use the hash value outside the ipfrag_lock
83 * as doing so could race with ipfrag_hash_rnd being recalculated.
85 static unsigned int inet6_hash_frag(__be32 id
, const struct in6_addr
*saddr
,
86 const struct in6_addr
*daddr
)
88 net_get_random_once(&ip6_frags
.rnd
, sizeof(ip6_frags
.rnd
));
89 return jhash_3words(ipv6_addr_hash(saddr
), ipv6_addr_hash(daddr
),
90 (__force u32
)id
, ip6_frags
.rnd
);
93 static unsigned int ip6_hashfn(const struct inet_frag_queue
*q
)
95 const struct frag_queue
*fq
;
97 fq
= container_of(q
, struct frag_queue
, q
);
98 return inet6_hash_frag(fq
->id
, &fq
->saddr
, &fq
->daddr
);
101 bool ip6_frag_match(const struct inet_frag_queue
*q
, const void *a
)
103 const struct frag_queue
*fq
;
104 const struct ip6_create_arg
*arg
= a
;
106 fq
= container_of(q
, struct frag_queue
, q
);
107 return fq
->id
== arg
->id
&&
108 fq
->user
== arg
->user
&&
109 ipv6_addr_equal(&fq
->saddr
, arg
->src
) &&
110 ipv6_addr_equal(&fq
->daddr
, arg
->dst
);
112 EXPORT_SYMBOL(ip6_frag_match
);
114 void ip6_frag_init(struct inet_frag_queue
*q
, const void *a
)
116 struct frag_queue
*fq
= container_of(q
, struct frag_queue
, q
);
117 const struct ip6_create_arg
*arg
= a
;
120 fq
->user
= arg
->user
;
121 fq
->saddr
= *arg
->src
;
122 fq
->daddr
= *arg
->dst
;
125 EXPORT_SYMBOL(ip6_frag_init
);
127 void ip6_expire_frag_queue(struct net
*net
, struct frag_queue
*fq
,
128 struct inet_frags
*frags
)
130 struct net_device
*dev
= NULL
;
132 spin_lock(&fq
->q
.lock
);
134 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
137 inet_frag_kill(&fq
->q
, frags
);
140 dev
= dev_get_by_index_rcu(net
, fq
->iif
);
144 if (!(fq
->q
.last_in
& INET_FRAG_EVICTED
))
145 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
),
146 IPSTATS_MIB_REASMTIMEOUT
);
147 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
149 /* Don't send error if the first segment did not arrive. */
150 if (!(fq
->q
.last_in
& INET_FRAG_FIRST_IN
) || !fq
->q
.fragments
)
154 But use as source device on which LAST ARRIVED
155 segment was received. And do not use fq->dev
156 pointer directly, device might already disappeared.
158 fq
->q
.fragments
->dev
= dev
;
159 icmpv6_send(fq
->q
.fragments
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_FRAGTIME
, 0);
163 spin_unlock(&fq
->q
.lock
);
164 inet_frag_put(&fq
->q
, frags
);
166 EXPORT_SYMBOL(ip6_expire_frag_queue
);
168 static void ip6_frag_expire(unsigned long data
)
170 struct frag_queue
*fq
;
173 fq
= container_of((struct inet_frag_queue
*)data
, struct frag_queue
, q
);
174 net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
176 ip6_expire_frag_queue(net
, fq
, &ip6_frags
);
179 static __inline__
struct frag_queue
*
180 fq_find(struct net
*net
, __be32 id
, const struct in6_addr
*src
,
181 const struct in6_addr
*dst
, u8 ecn
)
183 struct inet_frag_queue
*q
;
184 struct ip6_create_arg arg
;
188 arg
.user
= IP6_DEFRAG_LOCAL_DELIVER
;
193 read_lock(&ip6_frags
.lock
);
194 hash
= inet6_hash_frag(id
, src
, dst
);
196 q
= inet_frag_find(&net
->ipv6
.frags
, &ip6_frags
, &arg
, hash
);
197 if (IS_ERR_OR_NULL(q
)) {
198 inet_frag_maybe_warn_overflow(q
, pr_fmt());
201 return container_of(q
, struct frag_queue
, q
);
204 static int ip6_frag_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
205 struct frag_hdr
*fhdr
, int nhoff
)
207 struct sk_buff
*prev
, *next
;
208 struct net_device
*dev
;
210 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
213 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
216 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
217 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
218 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
220 if ((unsigned int)end
> IPV6_MAXPLEN
) {
221 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
222 IPSTATS_MIB_INHDRERRORS
);
223 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
224 ((u8
*)&fhdr
->frag_off
-
225 skb_network_header(skb
)));
229 ecn
= ip6_frag_ecn(ipv6_hdr(skb
));
231 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
232 const unsigned char *nh
= skb_network_header(skb
);
233 skb
->csum
= csum_sub(skb
->csum
,
234 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
238 /* Is this the final fragment? */
239 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
240 /* If we already have some bits beyond end
241 * or have different end, the segment is corrupted.
243 if (end
< fq
->q
.len
||
244 ((fq
->q
.last_in
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
))
246 fq
->q
.last_in
|= INET_FRAG_LAST_IN
;
249 /* Check if the fragment is rounded to 8 bytes.
250 * Required by the RFC.
253 /* RFC2460 says always send parameter problem in
256 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
257 IPSTATS_MIB_INHDRERRORS
);
258 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
259 offsetof(struct ipv6hdr
, payload_len
));
262 if (end
> fq
->q
.len
) {
263 /* Some bits beyond end -> corruption. */
264 if (fq
->q
.last_in
& INET_FRAG_LAST_IN
)
273 /* Point into the IP datagram 'data' part. */
274 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
))
277 if (pskb_trim_rcsum(skb
, end
- offset
))
280 /* Find out which fragments are in front and at the back of us
281 * in the chain of fragments so far. We must know where to put
282 * this fragment, right?
284 prev
= fq
->q
.fragments_tail
;
285 if (!prev
|| FRAG6_CB(prev
)->offset
< offset
) {
290 for(next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
291 if (FRAG6_CB(next
)->offset
>= offset
)
297 /* RFC5722, Section 4, amended by Errata ID : 3089
298 * When reassembling an IPv6 datagram, if
299 * one or more its constituent fragments is determined to be an
300 * overlapping fragment, the entire datagram (and any constituent
301 * fragments) MUST be silently discarded.
304 /* Check for overlap with preceding fragment. */
306 (FRAG6_CB(prev
)->offset
+ prev
->len
) > offset
)
309 /* Look for overlap with succeeding segment. */
310 if (next
&& FRAG6_CB(next
)->offset
< end
)
313 FRAG6_CB(skb
)->offset
= offset
;
315 /* Insert this fragment in the chain of fragments. */
318 fq
->q
.fragments_tail
= skb
;
322 fq
->q
.fragments
= skb
;
326 fq
->iif
= dev
->ifindex
;
329 fq
->q
.stamp
= skb
->tstamp
;
330 fq
->q
.meat
+= skb
->len
;
332 add_frag_mem_limit(&fq
->q
, skb
->truesize
);
334 /* The first fragment.
335 * nhoffset is obtained from the first fragment, of course.
338 fq
->nhoffset
= nhoff
;
339 fq
->q
.last_in
|= INET_FRAG_FIRST_IN
;
342 if (fq
->q
.last_in
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
343 fq
->q
.meat
== fq
->q
.len
) {
345 unsigned long orefdst
= skb
->_skb_refdst
;
347 skb
->_skb_refdst
= 0UL;
348 res
= ip6_frag_reasm(fq
, prev
, dev
);
349 skb
->_skb_refdst
= orefdst
;
357 inet_frag_kill(&fq
->q
, &ip6_frags
);
359 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
360 IPSTATS_MIB_REASMFAILS
);
366 * Check if this packet is complete.
367 * Returns NULL on failure by any reason, and pointer
368 * to current nexthdr field in reassembled frame.
370 * It is called with locked fq, and caller must check that
371 * queue is eligible for reassembly i.e. it is not COMPLETE,
372 * the last and the first frames arrived and all the bits are here.
374 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
375 struct net_device
*dev
)
377 struct net
*net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
378 struct sk_buff
*fp
, *head
= fq
->q
.fragments
;
384 inet_frag_kill(&fq
->q
, &ip6_frags
);
386 ecn
= ip_frag_ecn_table
[fq
->ecn
];
387 if (unlikely(ecn
== 0xff))
390 /* Make the one we just received the head. */
393 fp
= skb_clone(head
, GFP_ATOMIC
);
398 fp
->next
= head
->next
;
400 fq
->q
.fragments_tail
= fp
;
403 skb_morph(head
, fq
->q
.fragments
);
404 head
->next
= fq
->q
.fragments
->next
;
406 consume_skb(fq
->q
.fragments
);
407 fq
->q
.fragments
= head
;
410 WARN_ON(head
== NULL
);
411 WARN_ON(FRAG6_CB(head
)->offset
!= 0);
413 /* Unfragmented part is taken from the first segment. */
414 payload_len
= ((head
->data
- skb_network_header(head
)) -
415 sizeof(struct ipv6hdr
) + fq
->q
.len
-
416 sizeof(struct frag_hdr
));
417 if (payload_len
> IPV6_MAXPLEN
)
420 /* Head of list must not be cloned. */
421 if (skb_unclone(head
, GFP_ATOMIC
))
424 /* If the first fragment is fragmented itself, we split
425 * it to two chunks: the first with data and paged part
426 * and the second, holding only fragments. */
427 if (skb_has_frag_list(head
)) {
428 struct sk_buff
*clone
;
431 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
)
433 clone
->next
= head
->next
;
435 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
436 skb_frag_list_init(head
);
437 for (i
= 0; i
< skb_shinfo(head
)->nr_frags
; i
++)
438 plen
+= skb_frag_size(&skb_shinfo(head
)->frags
[i
]);
439 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
440 head
->data_len
-= clone
->len
;
441 head
->len
-= clone
->len
;
443 clone
->ip_summed
= head
->ip_summed
;
444 add_frag_mem_limit(&fq
->q
, clone
->truesize
);
447 /* We have to remove fragment header from datagram and to relocate
448 * header in order to calculate ICV correctly. */
449 nhoff
= fq
->nhoffset
;
450 skb_network_header(head
)[nhoff
] = skb_transport_header(head
)[0];
451 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
452 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
453 head
->mac_header
+= sizeof(struct frag_hdr
);
454 head
->network_header
+= sizeof(struct frag_hdr
);
456 skb_reset_transport_header(head
);
457 skb_push(head
, head
->data
- skb_network_header(head
));
459 sum_truesize
= head
->truesize
;
460 for (fp
= head
->next
; fp
;) {
463 struct sk_buff
*next
= fp
->next
;
465 sum_truesize
+= fp
->truesize
;
466 if (head
->ip_summed
!= fp
->ip_summed
)
467 head
->ip_summed
= CHECKSUM_NONE
;
468 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
469 head
->csum
= csum_add(head
->csum
, fp
->csum
);
471 if (skb_try_coalesce(head
, fp
, &headstolen
, &delta
)) {
472 kfree_skb_partial(fp
, headstolen
);
474 if (!skb_shinfo(head
)->frag_list
)
475 skb_shinfo(head
)->frag_list
= fp
;
476 head
->data_len
+= fp
->len
;
477 head
->len
+= fp
->len
;
478 head
->truesize
+= fp
->truesize
;
482 sub_frag_mem_limit(&fq
->q
, sum_truesize
);
486 head
->tstamp
= fq
->q
.stamp
;
487 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
488 ipv6_change_dsfield(ipv6_hdr(head
), 0xff, ecn
);
489 IP6CB(head
)->nhoff
= nhoff
;
490 IP6CB(head
)->flags
|= IP6SKB_FRAGMENTED
;
492 /* Yes, and fold redundant checksum back. 8) */
493 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
494 head
->csum
= csum_partial(skb_network_header(head
),
495 skb_network_header_len(head
),
499 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMOKS
);
501 fq
->q
.fragments
= NULL
;
502 fq
->q
.fragments_tail
= NULL
;
506 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len
);
509 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
512 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
517 static int ipv6_frag_rcv(struct sk_buff
*skb
)
519 struct frag_hdr
*fhdr
;
520 struct frag_queue
*fq
;
521 const struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
522 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
524 if (IP6CB(skb
)->flags
& IP6SKB_FRAGMENTED
)
527 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMREQDS
);
529 /* Jumbo payload inhibits frag. header */
530 if (hdr
->payload_len
==0)
533 if (!pskb_may_pull(skb
, (skb_transport_offset(skb
) +
534 sizeof(struct frag_hdr
))))
538 fhdr
= (struct frag_hdr
*)skb_transport_header(skb
);
540 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
541 /* It is not a fragmented frame */
542 skb
->transport_header
+= sizeof(struct frag_hdr
);
543 IP6_INC_STATS_BH(net
,
544 ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMOKS
);
546 IP6CB(skb
)->nhoff
= (u8
*)fhdr
- skb_network_header(skb
);
547 IP6CB(skb
)->flags
|= IP6SKB_FRAGMENTED
;
551 fq
= fq_find(net
, fhdr
->identification
, &hdr
->saddr
, &hdr
->daddr
,
556 spin_lock(&fq
->q
.lock
);
558 ret
= ip6_frag_queue(fq
, skb
, fhdr
, IP6CB(skb
)->nhoff
);
560 spin_unlock(&fq
->q
.lock
);
561 inet_frag_put(&fq
->q
, &ip6_frags
);
565 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMFAILS
);
570 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_INHDRERRORS
);
571 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb_network_header_len(skb
));
575 static const struct inet6_protocol frag_protocol
=
577 .handler
= ipv6_frag_rcv
,
578 .flags
= INET6_PROTO_NOPOLICY
,
582 static struct ctl_table ip6_frags_ns_ctl_table
[] = {
584 .procname
= "ip6frag_high_thresh",
585 .data
= &init_net
.ipv6
.frags
.high_thresh
,
586 .maxlen
= sizeof(int),
588 .proc_handler
= proc_dointvec
591 .procname
= "ip6frag_low_thresh",
592 .data
= &init_net
.ipv6
.frags
.low_thresh
,
593 .maxlen
= sizeof(int),
595 .proc_handler
= proc_dointvec
598 .procname
= "ip6frag_time",
599 .data
= &init_net
.ipv6
.frags
.timeout
,
600 .maxlen
= sizeof(int),
602 .proc_handler
= proc_dointvec_jiffies
,
607 static struct ctl_table ip6_frags_ctl_table
[] = {
609 .procname
= "ip6frag_secret_interval",
610 .data
= &ip6_frags
.secret_interval
,
611 .maxlen
= sizeof(int),
613 .proc_handler
= proc_dointvec_jiffies
,
618 static int __net_init
ip6_frags_ns_sysctl_register(struct net
*net
)
620 struct ctl_table
*table
;
621 struct ctl_table_header
*hdr
;
623 table
= ip6_frags_ns_ctl_table
;
624 if (!net_eq(net
, &init_net
)) {
625 table
= kmemdup(table
, sizeof(ip6_frags_ns_ctl_table
), GFP_KERNEL
);
629 table
[0].data
= &net
->ipv6
.frags
.high_thresh
;
630 table
[1].data
= &net
->ipv6
.frags
.low_thresh
;
631 table
[2].data
= &net
->ipv6
.frags
.timeout
;
633 /* Don't export sysctls to unprivileged users */
634 if (net
->user_ns
!= &init_user_ns
)
635 table
[0].procname
= NULL
;
638 hdr
= register_net_sysctl(net
, "net/ipv6", table
);
642 net
->ipv6
.sysctl
.frags_hdr
= hdr
;
646 if (!net_eq(net
, &init_net
))
652 static void __net_exit
ip6_frags_ns_sysctl_unregister(struct net
*net
)
654 struct ctl_table
*table
;
656 table
= net
->ipv6
.sysctl
.frags_hdr
->ctl_table_arg
;
657 unregister_net_sysctl_table(net
->ipv6
.sysctl
.frags_hdr
);
658 if (!net_eq(net
, &init_net
))
662 static struct ctl_table_header
*ip6_ctl_header
;
664 static int ip6_frags_sysctl_register(void)
666 ip6_ctl_header
= register_net_sysctl(&init_net
, "net/ipv6",
667 ip6_frags_ctl_table
);
668 return ip6_ctl_header
== NULL
? -ENOMEM
: 0;
671 static void ip6_frags_sysctl_unregister(void)
673 unregister_net_sysctl_table(ip6_ctl_header
);
676 static inline int ip6_frags_ns_sysctl_register(struct net
*net
)
681 static inline void ip6_frags_ns_sysctl_unregister(struct net
*net
)
685 static inline int ip6_frags_sysctl_register(void)
690 static inline void ip6_frags_sysctl_unregister(void)
695 static int __net_init
ipv6_frags_init_net(struct net
*net
)
697 net
->ipv6
.frags
.high_thresh
= IPV6_FRAG_HIGH_THRESH
;
698 net
->ipv6
.frags
.low_thresh
= IPV6_FRAG_LOW_THRESH
;
699 net
->ipv6
.frags
.timeout
= IPV6_FRAG_TIMEOUT
;
701 inet_frags_init_net(&net
->ipv6
.frags
);
703 return ip6_frags_ns_sysctl_register(net
);
706 static void __net_exit
ipv6_frags_exit_net(struct net
*net
)
708 ip6_frags_ns_sysctl_unregister(net
);
709 inet_frags_exit_net(&net
->ipv6
.frags
, &ip6_frags
);
712 static struct pernet_operations ip6_frags_ops
= {
713 .init
= ipv6_frags_init_net
,
714 .exit
= ipv6_frags_exit_net
,
717 int __init
ipv6_frag_init(void)
721 ret
= inet6_add_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
725 ret
= ip6_frags_sysctl_register();
729 ret
= register_pernet_subsys(&ip6_frags_ops
);
733 ip6_frags
.hashfn
= ip6_hashfn
;
734 ip6_frags
.constructor
= ip6_frag_init
;
735 ip6_frags
.destructor
= NULL
;
736 ip6_frags
.skb_free
= NULL
;
737 ip6_frags
.qsize
= sizeof(struct frag_queue
);
738 ip6_frags
.match
= ip6_frag_match
;
739 ip6_frags
.frag_expire
= ip6_frag_expire
;
740 ip6_frags
.secret_interval
= 10 * 60 * HZ
;
741 inet_frags_init(&ip6_frags
);
746 ip6_frags_sysctl_unregister();
748 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
752 void ipv6_frag_exit(void)
754 inet_frags_fini(&ip6_frags
);
755 ip6_frags_sysctl_unregister();
756 unregister_pernet_subsys(&ip6_frags_ops
);
757 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);