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.
29 #include <linux/errno.h>
30 #include <linux/types.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/sockios.h>
34 #include <linux/jiffies.h>
35 #include <linux/net.h>
36 #include <linux/list.h>
37 #include <linux/netdevice.h>
38 #include <linux/in6.h>
39 #include <linux/ipv6.h>
40 #include <linux/icmpv6.h>
41 #include <linux/random.h>
42 #include <linux/jhash.h>
43 #include <linux/skbuff.h>
44 #include <linux/slab.h>
45 #include <linux/export.h>
51 #include <net/ip6_route.h>
52 #include <net/protocol.h>
53 #include <net/transp_v6.h>
54 #include <net/rawv6.h>
55 #include <net/ndisc.h>
56 #include <net/addrconf.h>
57 #include <net/inet_frag.h>
61 struct inet6_skb_parm h
;
65 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
69 * Equivalent of ipv4 struct ipq
74 struct inet_frag_queue q
;
76 __be32 id
; /* fragment id */
78 struct in6_addr saddr
;
79 struct in6_addr daddr
;
86 static struct inet_frags ip6_frags
;
88 int ip6_frag_nqueues(struct net
*net
)
90 return net
->ipv6
.frags
.nqueues
;
93 int ip6_frag_mem(struct net
*net
)
95 return atomic_read(&net
->ipv6
.frags
.mem
);
98 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
99 struct net_device
*dev
);
102 * callers should be careful not to use the hash value outside the ipfrag_lock
103 * as doing so could race with ipfrag_hash_rnd being recalculated.
105 unsigned int inet6_hash_frag(__be32 id
, const struct in6_addr
*saddr
,
106 const struct in6_addr
*daddr
, u32 rnd
)
110 c
= jhash_3words((__force u32
)saddr
->s6_addr32
[0],
111 (__force u32
)saddr
->s6_addr32
[1],
112 (__force u32
)saddr
->s6_addr32
[2],
115 c
= jhash_3words((__force u32
)saddr
->s6_addr32
[3],
116 (__force u32
)daddr
->s6_addr32
[0],
117 (__force u32
)daddr
->s6_addr32
[1],
120 c
= jhash_3words((__force u32
)daddr
->s6_addr32
[2],
121 (__force u32
)daddr
->s6_addr32
[3],
125 return c
& (INETFRAGS_HASHSZ
- 1);
127 EXPORT_SYMBOL_GPL(inet6_hash_frag
);
129 static unsigned int ip6_hashfn(struct inet_frag_queue
*q
)
131 struct frag_queue
*fq
;
133 fq
= container_of(q
, struct frag_queue
, q
);
134 return inet6_hash_frag(fq
->id
, &fq
->saddr
, &fq
->daddr
, ip6_frags
.rnd
);
137 int ip6_frag_match(struct inet_frag_queue
*q
, void *a
)
139 struct frag_queue
*fq
;
140 struct ip6_create_arg
*arg
= a
;
142 fq
= container_of(q
, struct frag_queue
, q
);
143 return (fq
->id
== arg
->id
&& fq
->user
== arg
->user
&&
144 ipv6_addr_equal(&fq
->saddr
, arg
->src
) &&
145 ipv6_addr_equal(&fq
->daddr
, arg
->dst
));
147 EXPORT_SYMBOL(ip6_frag_match
);
149 void ip6_frag_init(struct inet_frag_queue
*q
, void *a
)
151 struct frag_queue
*fq
= container_of(q
, struct frag_queue
, q
);
152 struct ip6_create_arg
*arg
= a
;
155 fq
->user
= arg
->user
;
156 fq
->saddr
= *arg
->src
;
157 fq
->daddr
= *arg
->dst
;
159 EXPORT_SYMBOL(ip6_frag_init
);
161 /* Destruction primitives. */
163 static __inline__
void fq_put(struct frag_queue
*fq
)
165 inet_frag_put(&fq
->q
, &ip6_frags
);
168 /* Kill fq entry. It is not destroyed immediately,
169 * because caller (and someone more) holds reference count.
171 static __inline__
void fq_kill(struct frag_queue
*fq
)
173 inet_frag_kill(&fq
->q
, &ip6_frags
);
176 static void ip6_evictor(struct net
*net
, struct inet6_dev
*idev
)
180 evicted
= inet_frag_evictor(&net
->ipv6
.frags
, &ip6_frags
);
182 IP6_ADD_STATS_BH(net
, idev
, IPSTATS_MIB_REASMFAILS
, evicted
);
185 static void ip6_frag_expire(unsigned long data
)
187 struct frag_queue
*fq
;
188 struct net_device
*dev
= NULL
;
191 fq
= container_of((struct inet_frag_queue
*)data
, struct frag_queue
, q
);
193 spin_lock(&fq
->q
.lock
);
195 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
200 net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
202 dev
= dev_get_by_index_rcu(net
, fq
->iif
);
206 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMTIMEOUT
);
207 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
209 /* Don't send error if the first segment did not arrive. */
210 if (!(fq
->q
.last_in
& INET_FRAG_FIRST_IN
) || !fq
->q
.fragments
)
214 But use as source device on which LAST ARRIVED
215 segment was received. And do not use fq->dev
216 pointer directly, device might already disappeared.
218 fq
->q
.fragments
->dev
= dev
;
219 icmpv6_send(fq
->q
.fragments
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_FRAGTIME
, 0);
223 spin_unlock(&fq
->q
.lock
);
227 static __inline__
struct frag_queue
*
228 fq_find(struct net
*net
, __be32 id
, const struct in6_addr
*src
, const struct in6_addr
*dst
)
230 struct inet_frag_queue
*q
;
231 struct ip6_create_arg arg
;
235 arg
.user
= IP6_DEFRAG_LOCAL_DELIVER
;
239 read_lock(&ip6_frags
.lock
);
240 hash
= inet6_hash_frag(id
, src
, dst
, ip6_frags
.rnd
);
242 q
= inet_frag_find(&net
->ipv6
.frags
, &ip6_frags
, &arg
, hash
);
246 return container_of(q
, struct frag_queue
, q
);
249 static int ip6_frag_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
250 struct frag_hdr
*fhdr
, int nhoff
)
252 struct sk_buff
*prev
, *next
;
253 struct net_device
*dev
;
255 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
257 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
260 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
261 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
262 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
264 if ((unsigned int)end
> IPV6_MAXPLEN
) {
265 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
266 IPSTATS_MIB_INHDRERRORS
);
267 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
268 ((u8
*)&fhdr
->frag_off
-
269 skb_network_header(skb
)));
273 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
274 const unsigned char *nh
= skb_network_header(skb
);
275 skb
->csum
= csum_sub(skb
->csum
,
276 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
280 /* Is this the final fragment? */
281 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
282 /* If we already have some bits beyond end
283 * or have different end, the segment is corrupted.
285 if (end
< fq
->q
.len
||
286 ((fq
->q
.last_in
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
))
288 fq
->q
.last_in
|= INET_FRAG_LAST_IN
;
291 /* Check if the fragment is rounded to 8 bytes.
292 * Required by the RFC.
295 /* RFC2460 says always send parameter problem in
298 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
299 IPSTATS_MIB_INHDRERRORS
);
300 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
301 offsetof(struct ipv6hdr
, payload_len
));
304 if (end
> fq
->q
.len
) {
305 /* Some bits beyond end -> corruption. */
306 if (fq
->q
.last_in
& INET_FRAG_LAST_IN
)
315 /* Point into the IP datagram 'data' part. */
316 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
))
319 if (pskb_trim_rcsum(skb
, end
- offset
))
322 /* Find out which fragments are in front and at the back of us
323 * in the chain of fragments so far. We must know where to put
324 * this fragment, right?
326 prev
= fq
->q
.fragments_tail
;
327 if (!prev
|| FRAG6_CB(prev
)->offset
< offset
) {
332 for(next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
333 if (FRAG6_CB(next
)->offset
>= offset
)
339 /* RFC5722, Section 4:
340 * When reassembling an IPv6 datagram, if
341 * one or more its constituent fragments is determined to be an
342 * overlapping fragment, the entire datagram (and any constituent
343 * fragments, including those not yet received) MUST be silently
347 /* Check for overlap with preceding fragment. */
349 (FRAG6_CB(prev
)->offset
+ prev
->len
) > offset
)
352 /* Look for overlap with succeeding segment. */
353 if (next
&& FRAG6_CB(next
)->offset
< end
)
356 FRAG6_CB(skb
)->offset
= offset
;
358 /* Insert this fragment in the chain of fragments. */
361 fq
->q
.fragments_tail
= skb
;
365 fq
->q
.fragments
= skb
;
369 fq
->iif
= dev
->ifindex
;
372 fq
->q
.stamp
= skb
->tstamp
;
373 fq
->q
.meat
+= skb
->len
;
374 atomic_add(skb
->truesize
, &fq
->q
.net
->mem
);
376 /* The first fragment.
377 * nhoffset is obtained from the first fragment, of course.
380 fq
->nhoffset
= nhoff
;
381 fq
->q
.last_in
|= INET_FRAG_FIRST_IN
;
384 if (fq
->q
.last_in
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
385 fq
->q
.meat
== fq
->q
.len
)
386 return ip6_frag_reasm(fq
, prev
, dev
);
388 write_lock(&ip6_frags
.lock
);
389 list_move_tail(&fq
->q
.lru_list
, &fq
->q
.net
->lru_list
);
390 write_unlock(&ip6_frags
.lock
);
396 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
397 IPSTATS_MIB_REASMFAILS
);
403 * Check if this packet is complete.
404 * Returns NULL on failure by any reason, and pointer
405 * to current nexthdr field in reassembled frame.
407 * It is called with locked fq, and caller must check that
408 * queue is eligible for reassembly i.e. it is not COMPLETE,
409 * the last and the first frames arrived and all the bits are here.
411 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
412 struct net_device
*dev
)
414 struct net
*net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
415 struct sk_buff
*fp
, *head
= fq
->q
.fragments
;
421 /* Make the one we just received the head. */
424 fp
= skb_clone(head
, GFP_ATOMIC
);
429 fp
->next
= head
->next
;
431 fq
->q
.fragments_tail
= fp
;
434 skb_morph(head
, fq
->q
.fragments
);
435 head
->next
= fq
->q
.fragments
->next
;
437 kfree_skb(fq
->q
.fragments
);
438 fq
->q
.fragments
= head
;
441 WARN_ON(head
== NULL
);
442 WARN_ON(FRAG6_CB(head
)->offset
!= 0);
444 /* Unfragmented part is taken from the first segment. */
445 payload_len
= ((head
->data
- skb_network_header(head
)) -
446 sizeof(struct ipv6hdr
) + fq
->q
.len
-
447 sizeof(struct frag_hdr
));
448 if (payload_len
> IPV6_MAXPLEN
)
451 /* Head of list must not be cloned. */
452 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
))
455 /* If the first fragment is fragmented itself, we split
456 * it to two chunks: the first with data and paged part
457 * and the second, holding only fragments. */
458 if (skb_has_frag_list(head
)) {
459 struct sk_buff
*clone
;
462 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
)
464 clone
->next
= head
->next
;
466 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
467 skb_frag_list_init(head
);
468 for (i
= 0; i
< skb_shinfo(head
)->nr_frags
; i
++)
469 plen
+= skb_frag_size(&skb_shinfo(head
)->frags
[i
]);
470 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
471 head
->data_len
-= clone
->len
;
472 head
->len
-= clone
->len
;
474 clone
->ip_summed
= head
->ip_summed
;
475 atomic_add(clone
->truesize
, &fq
->q
.net
->mem
);
478 /* We have to remove fragment header from datagram and to relocate
479 * header in order to calculate ICV correctly. */
480 nhoff
= fq
->nhoffset
;
481 skb_network_header(head
)[nhoff
] = skb_transport_header(head
)[0];
482 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
483 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
484 head
->mac_header
+= sizeof(struct frag_hdr
);
485 head
->network_header
+= sizeof(struct frag_hdr
);
487 skb_shinfo(head
)->frag_list
= head
->next
;
488 skb_reset_transport_header(head
);
489 skb_push(head
, head
->data
- skb_network_header(head
));
491 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
492 head
->data_len
+= fp
->len
;
493 head
->len
+= fp
->len
;
494 if (head
->ip_summed
!= fp
->ip_summed
)
495 head
->ip_summed
= CHECKSUM_NONE
;
496 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
497 head
->csum
= csum_add(head
->csum
, fp
->csum
);
498 head
->truesize
+= fp
->truesize
;
500 atomic_sub(head
->truesize
, &fq
->q
.net
->mem
);
504 head
->tstamp
= fq
->q
.stamp
;
505 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
506 IP6CB(head
)->nhoff
= nhoff
;
508 /* Yes, and fold redundant checksum back. 8) */
509 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
510 head
->csum
= csum_partial(skb_network_header(head
),
511 skb_network_header_len(head
),
515 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMOKS
);
517 fq
->q
.fragments
= NULL
;
518 fq
->q
.fragments_tail
= NULL
;
523 printk(KERN_DEBUG
"ip6_frag_reasm: payload len = %d\n", payload_len
);
527 printk(KERN_DEBUG
"ip6_frag_reasm: no memory for reassembly\n");
530 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
535 static int ipv6_frag_rcv(struct sk_buff
*skb
)
537 struct frag_hdr
*fhdr
;
538 struct frag_queue
*fq
;
539 const struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
540 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
542 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMREQDS
);
544 /* Jumbo payload inhibits frag. header */
545 if (hdr
->payload_len
==0)
548 if (!pskb_may_pull(skb
, (skb_transport_offset(skb
) +
549 sizeof(struct frag_hdr
))))
553 fhdr
= (struct frag_hdr
*)skb_transport_header(skb
);
555 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
556 /* It is not a fragmented frame */
557 skb
->transport_header
+= sizeof(struct frag_hdr
);
558 IP6_INC_STATS_BH(net
,
559 ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMOKS
);
561 IP6CB(skb
)->nhoff
= (u8
*)fhdr
- skb_network_header(skb
);
565 if (atomic_read(&net
->ipv6
.frags
.mem
) > net
->ipv6
.frags
.high_thresh
)
566 ip6_evictor(net
, ip6_dst_idev(skb_dst(skb
)));
568 fq
= fq_find(net
, fhdr
->identification
, &hdr
->saddr
, &hdr
->daddr
);
572 spin_lock(&fq
->q
.lock
);
574 ret
= ip6_frag_queue(fq
, skb
, fhdr
, IP6CB(skb
)->nhoff
);
576 spin_unlock(&fq
->q
.lock
);
581 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMFAILS
);
586 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_INHDRERRORS
);
587 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb_network_header_len(skb
));
591 static const struct inet6_protocol frag_protocol
=
593 .handler
= ipv6_frag_rcv
,
594 .flags
= INET6_PROTO_NOPOLICY
,
598 static struct ctl_table ip6_frags_ns_ctl_table
[] = {
600 .procname
= "ip6frag_high_thresh",
601 .data
= &init_net
.ipv6
.frags
.high_thresh
,
602 .maxlen
= sizeof(int),
604 .proc_handler
= proc_dointvec
607 .procname
= "ip6frag_low_thresh",
608 .data
= &init_net
.ipv6
.frags
.low_thresh
,
609 .maxlen
= sizeof(int),
611 .proc_handler
= proc_dointvec
614 .procname
= "ip6frag_time",
615 .data
= &init_net
.ipv6
.frags
.timeout
,
616 .maxlen
= sizeof(int),
618 .proc_handler
= proc_dointvec_jiffies
,
623 static struct ctl_table ip6_frags_ctl_table
[] = {
625 .procname
= "ip6frag_secret_interval",
626 .data
= &ip6_frags
.secret_interval
,
627 .maxlen
= sizeof(int),
629 .proc_handler
= proc_dointvec_jiffies
,
634 static int __net_init
ip6_frags_ns_sysctl_register(struct net
*net
)
636 struct ctl_table
*table
;
637 struct ctl_table_header
*hdr
;
639 table
= ip6_frags_ns_ctl_table
;
640 if (!net_eq(net
, &init_net
)) {
641 table
= kmemdup(table
, sizeof(ip6_frags_ns_ctl_table
), GFP_KERNEL
);
645 table
[0].data
= &net
->ipv6
.frags
.high_thresh
;
646 table
[1].data
= &net
->ipv6
.frags
.low_thresh
;
647 table
[2].data
= &net
->ipv6
.frags
.timeout
;
650 hdr
= register_net_sysctl_table(net
, net_ipv6_ctl_path
, table
);
654 net
->ipv6
.sysctl
.frags_hdr
= hdr
;
658 if (!net_eq(net
, &init_net
))
664 static void __net_exit
ip6_frags_ns_sysctl_unregister(struct net
*net
)
666 struct ctl_table
*table
;
668 table
= net
->ipv6
.sysctl
.frags_hdr
->ctl_table_arg
;
669 unregister_net_sysctl_table(net
->ipv6
.sysctl
.frags_hdr
);
670 if (!net_eq(net
, &init_net
))
674 static struct ctl_table_header
*ip6_ctl_header
;
676 static int ip6_frags_sysctl_register(void)
678 ip6_ctl_header
= register_net_sysctl_rotable(net_ipv6_ctl_path
,
679 ip6_frags_ctl_table
);
680 return ip6_ctl_header
== NULL
? -ENOMEM
: 0;
683 static void ip6_frags_sysctl_unregister(void)
685 unregister_net_sysctl_table(ip6_ctl_header
);
688 static inline int ip6_frags_ns_sysctl_register(struct net
*net
)
693 static inline void ip6_frags_ns_sysctl_unregister(struct net
*net
)
697 static inline int ip6_frags_sysctl_register(void)
702 static inline void ip6_frags_sysctl_unregister(void)
707 static int __net_init
ipv6_frags_init_net(struct net
*net
)
709 net
->ipv6
.frags
.high_thresh
= IPV6_FRAG_HIGH_THRESH
;
710 net
->ipv6
.frags
.low_thresh
= IPV6_FRAG_LOW_THRESH
;
711 net
->ipv6
.frags
.timeout
= IPV6_FRAG_TIMEOUT
;
713 inet_frags_init_net(&net
->ipv6
.frags
);
715 return ip6_frags_ns_sysctl_register(net
);
718 static void __net_exit
ipv6_frags_exit_net(struct net
*net
)
720 ip6_frags_ns_sysctl_unregister(net
);
721 inet_frags_exit_net(&net
->ipv6
.frags
, &ip6_frags
);
724 static struct pernet_operations ip6_frags_ops
= {
725 .init
= ipv6_frags_init_net
,
726 .exit
= ipv6_frags_exit_net
,
729 int __init
ipv6_frag_init(void)
733 ret
= inet6_add_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
737 ret
= ip6_frags_sysctl_register();
741 ret
= register_pernet_subsys(&ip6_frags_ops
);
745 ip6_frags
.hashfn
= ip6_hashfn
;
746 ip6_frags
.constructor
= ip6_frag_init
;
747 ip6_frags
.destructor
= NULL
;
748 ip6_frags
.skb_free
= NULL
;
749 ip6_frags
.qsize
= sizeof(struct frag_queue
);
750 ip6_frags
.match
= ip6_frag_match
;
751 ip6_frags
.frag_expire
= ip6_frag_expire
;
752 ip6_frags
.secret_interval
= 10 * 60 * HZ
;
753 inet_frags_init(&ip6_frags
);
758 ip6_frags_sysctl_unregister();
760 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
764 void ipv6_frag_exit(void)
766 inet_frags_fini(&ip6_frags
);
767 ip6_frags_sysctl_unregister();
768 unregister_pernet_subsys(&ip6_frags_ops
);
769 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);