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>
50 #include <net/ip6_route.h>
51 #include <net/protocol.h>
52 #include <net/transp_v6.h>
53 #include <net/rawv6.h>
54 #include <net/ndisc.h>
55 #include <net/addrconf.h>
56 #include <net/inet_frag.h>
60 struct inet6_skb_parm h
;
64 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
68 * Equivalent of ipv4 struct ipq
73 struct inet_frag_queue q
;
75 __be32 id
; /* fragment id */
77 struct in6_addr saddr
;
78 struct in6_addr daddr
;
85 static struct inet_frags ip6_frags
;
87 int ip6_frag_nqueues(struct net
*net
)
89 return net
->ipv6
.frags
.nqueues
;
92 int ip6_frag_mem(struct net
*net
)
94 return atomic_read(&net
->ipv6
.frags
.mem
);
97 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
98 struct net_device
*dev
);
101 * callers should be careful not to use the hash value outside the ipfrag_lock
102 * as doing so could race with ipfrag_hash_rnd being recalculated.
104 unsigned int inet6_hash_frag(__be32 id
, const struct in6_addr
*saddr
,
105 const struct in6_addr
*daddr
, u32 rnd
)
109 c
= jhash_3words((__force u32
)saddr
->s6_addr32
[0],
110 (__force u32
)saddr
->s6_addr32
[1],
111 (__force u32
)saddr
->s6_addr32
[2],
114 c
= jhash_3words((__force u32
)saddr
->s6_addr32
[3],
115 (__force u32
)daddr
->s6_addr32
[0],
116 (__force u32
)daddr
->s6_addr32
[1],
119 c
= jhash_3words((__force u32
)daddr
->s6_addr32
[2],
120 (__force u32
)daddr
->s6_addr32
[3],
124 return c
& (INETFRAGS_HASHSZ
- 1);
126 EXPORT_SYMBOL_GPL(inet6_hash_frag
);
128 static unsigned int ip6_hashfn(struct inet_frag_queue
*q
)
130 struct frag_queue
*fq
;
132 fq
= container_of(q
, struct frag_queue
, q
);
133 return inet6_hash_frag(fq
->id
, &fq
->saddr
, &fq
->daddr
, ip6_frags
.rnd
);
136 int ip6_frag_match(struct inet_frag_queue
*q
, void *a
)
138 struct frag_queue
*fq
;
139 struct ip6_create_arg
*arg
= a
;
141 fq
= container_of(q
, struct frag_queue
, q
);
142 return (fq
->id
== arg
->id
&& fq
->user
== arg
->user
&&
143 ipv6_addr_equal(&fq
->saddr
, arg
->src
) &&
144 ipv6_addr_equal(&fq
->daddr
, arg
->dst
));
146 EXPORT_SYMBOL(ip6_frag_match
);
148 void ip6_frag_init(struct inet_frag_queue
*q
, void *a
)
150 struct frag_queue
*fq
= container_of(q
, struct frag_queue
, q
);
151 struct ip6_create_arg
*arg
= a
;
154 fq
->user
= arg
->user
;
155 ipv6_addr_copy(&fq
->saddr
, arg
->src
);
156 ipv6_addr_copy(&fq
->daddr
, arg
->dst
);
158 EXPORT_SYMBOL(ip6_frag_init
);
160 /* Destruction primitives. */
162 static __inline__
void fq_put(struct frag_queue
*fq
)
164 inet_frag_put(&fq
->q
, &ip6_frags
);
167 /* Kill fq entry. It is not destroyed immediately,
168 * because caller (and someone more) holds reference count.
170 static __inline__
void fq_kill(struct frag_queue
*fq
)
172 inet_frag_kill(&fq
->q
, &ip6_frags
);
175 static void ip6_evictor(struct net
*net
, struct inet6_dev
*idev
)
179 evicted
= inet_frag_evictor(&net
->ipv6
.frags
, &ip6_frags
);
181 IP6_ADD_STATS_BH(net
, idev
, IPSTATS_MIB_REASMFAILS
, evicted
);
184 static void ip6_frag_expire(unsigned long data
)
186 struct frag_queue
*fq
;
187 struct net_device
*dev
= NULL
;
190 fq
= container_of((struct inet_frag_queue
*)data
, struct frag_queue
, q
);
192 spin_lock(&fq
->q
.lock
);
194 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
199 net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
201 dev
= dev_get_by_index_rcu(net
, fq
->iif
);
205 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMTIMEOUT
);
206 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
208 /* Don't send error if the first segment did not arrive. */
209 if (!(fq
->q
.last_in
& INET_FRAG_FIRST_IN
) || !fq
->q
.fragments
)
213 But use as source device on which LAST ARRIVED
214 segment was received. And do not use fq->dev
215 pointer directly, device might already disappeared.
217 fq
->q
.fragments
->dev
= dev
;
218 icmpv6_send(fq
->q
.fragments
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_FRAGTIME
, 0);
222 spin_unlock(&fq
->q
.lock
);
226 static __inline__
struct frag_queue
*
227 fq_find(struct net
*net
, __be32 id
, const struct in6_addr
*src
, const struct in6_addr
*dst
)
229 struct inet_frag_queue
*q
;
230 struct ip6_create_arg arg
;
234 arg
.user
= IP6_DEFRAG_LOCAL_DELIVER
;
238 read_lock(&ip6_frags
.lock
);
239 hash
= inet6_hash_frag(id
, src
, dst
, ip6_frags
.rnd
);
241 q
= inet_frag_find(&net
->ipv6
.frags
, &ip6_frags
, &arg
, hash
);
245 return container_of(q
, struct frag_queue
, q
);
248 static int ip6_frag_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
249 struct frag_hdr
*fhdr
, int nhoff
)
251 struct sk_buff
*prev
, *next
;
252 struct net_device
*dev
;
254 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
256 if (fq
->q
.last_in
& INET_FRAG_COMPLETE
)
259 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
260 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
261 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
263 if ((unsigned int)end
> IPV6_MAXPLEN
) {
264 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
265 IPSTATS_MIB_INHDRERRORS
);
266 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
267 ((u8
*)&fhdr
->frag_off
-
268 skb_network_header(skb
)));
272 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
273 const unsigned char *nh
= skb_network_header(skb
);
274 skb
->csum
= csum_sub(skb
->csum
,
275 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
279 /* Is this the final fragment? */
280 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
281 /* If we already have some bits beyond end
282 * or have different end, the segment is corrupted.
284 if (end
< fq
->q
.len
||
285 ((fq
->q
.last_in
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
))
287 fq
->q
.last_in
|= INET_FRAG_LAST_IN
;
290 /* Check if the fragment is rounded to 8 bytes.
291 * Required by the RFC.
294 /* RFC2460 says always send parameter problem in
297 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)),
298 IPSTATS_MIB_INHDRERRORS
);
299 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
300 offsetof(struct ipv6hdr
, payload_len
));
303 if (end
> fq
->q
.len
) {
304 /* Some bits beyond end -> corruption. */
305 if (fq
->q
.last_in
& INET_FRAG_LAST_IN
)
314 /* Point into the IP datagram 'data' part. */
315 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
))
318 if (pskb_trim_rcsum(skb
, end
- offset
))
321 /* Find out which fragments are in front and at the back of us
322 * in the chain of fragments so far. We must know where to put
323 * this fragment, right?
325 prev
= fq
->q
.fragments_tail
;
326 if (!prev
|| FRAG6_CB(prev
)->offset
< offset
) {
331 for(next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
332 if (FRAG6_CB(next
)->offset
>= offset
)
338 /* RFC5722, Section 4:
339 * When reassembling an IPv6 datagram, if
340 * one or more its constituent fragments is determined to be an
341 * overlapping fragment, the entire datagram (and any constituent
342 * fragments, including those not yet received) MUST be silently
346 /* Check for overlap with preceding fragment. */
348 (FRAG6_CB(prev
)->offset
+ prev
->len
) > offset
)
351 /* Look for overlap with succeeding segment. */
352 if (next
&& FRAG6_CB(next
)->offset
< end
)
355 FRAG6_CB(skb
)->offset
= offset
;
357 /* Insert this fragment in the chain of fragments. */
360 fq
->q
.fragments_tail
= skb
;
364 fq
->q
.fragments
= skb
;
368 fq
->iif
= dev
->ifindex
;
371 fq
->q
.stamp
= skb
->tstamp
;
372 fq
->q
.meat
+= skb
->len
;
373 atomic_add(skb
->truesize
, &fq
->q
.net
->mem
);
375 /* The first fragment.
376 * nhoffset is obtained from the first fragment, of course.
379 fq
->nhoffset
= nhoff
;
380 fq
->q
.last_in
|= INET_FRAG_FIRST_IN
;
383 if (fq
->q
.last_in
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
384 fq
->q
.meat
== fq
->q
.len
)
385 return ip6_frag_reasm(fq
, prev
, dev
);
387 write_lock(&ip6_frags
.lock
);
388 list_move_tail(&fq
->q
.lru_list
, &fq
->q
.net
->lru_list
);
389 write_unlock(&ip6_frags
.lock
);
395 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
396 IPSTATS_MIB_REASMFAILS
);
402 * Check if this packet is complete.
403 * Returns NULL on failure by any reason, and pointer
404 * to current nexthdr field in reassembled frame.
406 * It is called with locked fq, and caller must check that
407 * queue is eligible for reassembly i.e. it is not COMPLETE,
408 * the last and the first frames arrived and all the bits are here.
410 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
411 struct net_device
*dev
)
413 struct net
*net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
414 struct sk_buff
*fp
, *head
= fq
->q
.fragments
;
420 /* Make the one we just received the head. */
423 fp
= skb_clone(head
, GFP_ATOMIC
);
428 fp
->next
= head
->next
;
430 fq
->q
.fragments_tail
= fp
;
433 skb_morph(head
, fq
->q
.fragments
);
434 head
->next
= fq
->q
.fragments
->next
;
436 kfree_skb(fq
->q
.fragments
);
437 fq
->q
.fragments
= head
;
440 WARN_ON(head
== NULL
);
441 WARN_ON(FRAG6_CB(head
)->offset
!= 0);
443 /* Unfragmented part is taken from the first segment. */
444 payload_len
= ((head
->data
- skb_network_header(head
)) -
445 sizeof(struct ipv6hdr
) + fq
->q
.len
-
446 sizeof(struct frag_hdr
));
447 if (payload_len
> IPV6_MAXPLEN
)
450 /* Head of list must not be cloned. */
451 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
))
454 /* If the first fragment is fragmented itself, we split
455 * it to two chunks: the first with data and paged part
456 * and the second, holding only fragments. */
457 if (skb_has_frag_list(head
)) {
458 struct sk_buff
*clone
;
461 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
)
463 clone
->next
= head
->next
;
465 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
466 skb_frag_list_init(head
);
467 for (i
=0; i
<skb_shinfo(head
)->nr_frags
; i
++)
468 plen
+= skb_shinfo(head
)->frags
[i
].size
;
469 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
470 head
->data_len
-= clone
->len
;
471 head
->len
-= clone
->len
;
473 clone
->ip_summed
= head
->ip_summed
;
474 atomic_add(clone
->truesize
, &fq
->q
.net
->mem
);
477 /* We have to remove fragment header from datagram and to relocate
478 * header in order to calculate ICV correctly. */
479 nhoff
= fq
->nhoffset
;
480 skb_network_header(head
)[nhoff
] = skb_transport_header(head
)[0];
481 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
482 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
483 head
->mac_header
+= sizeof(struct frag_hdr
);
484 head
->network_header
+= sizeof(struct frag_hdr
);
486 skb_shinfo(head
)->frag_list
= head
->next
;
487 skb_reset_transport_header(head
);
488 skb_push(head
, head
->data
- skb_network_header(head
));
490 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
491 head
->data_len
+= fp
->len
;
492 head
->len
+= fp
->len
;
493 if (head
->ip_summed
!= fp
->ip_summed
)
494 head
->ip_summed
= CHECKSUM_NONE
;
495 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
496 head
->csum
= csum_add(head
->csum
, fp
->csum
);
497 head
->truesize
+= fp
->truesize
;
499 atomic_sub(head
->truesize
, &fq
->q
.net
->mem
);
503 head
->tstamp
= fq
->q
.stamp
;
504 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
505 IP6CB(head
)->nhoff
= nhoff
;
507 /* Yes, and fold redundant checksum back. 8) */
508 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
509 head
->csum
= csum_partial(skb_network_header(head
),
510 skb_network_header_len(head
),
514 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMOKS
);
516 fq
->q
.fragments
= NULL
;
517 fq
->q
.fragments_tail
= NULL
;
522 printk(KERN_DEBUG
"ip6_frag_reasm: payload len = %d\n", payload_len
);
526 printk(KERN_DEBUG
"ip6_frag_reasm: no memory for reassembly\n");
529 IP6_INC_STATS_BH(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
534 static int ipv6_frag_rcv(struct sk_buff
*skb
)
536 struct frag_hdr
*fhdr
;
537 struct frag_queue
*fq
;
538 const struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
539 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
541 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMREQDS
);
543 /* Jumbo payload inhibits frag. header */
544 if (hdr
->payload_len
==0)
547 if (!pskb_may_pull(skb
, (skb_transport_offset(skb
) +
548 sizeof(struct frag_hdr
))))
552 fhdr
= (struct frag_hdr
*)skb_transport_header(skb
);
554 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
555 /* It is not a fragmented frame */
556 skb
->transport_header
+= sizeof(struct frag_hdr
);
557 IP6_INC_STATS_BH(net
,
558 ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMOKS
);
560 IP6CB(skb
)->nhoff
= (u8
*)fhdr
- skb_network_header(skb
);
564 if (atomic_read(&net
->ipv6
.frags
.mem
) > net
->ipv6
.frags
.high_thresh
)
565 ip6_evictor(net
, ip6_dst_idev(skb_dst(skb
)));
567 fq
= fq_find(net
, fhdr
->identification
, &hdr
->saddr
, &hdr
->daddr
);
571 spin_lock(&fq
->q
.lock
);
573 ret
= ip6_frag_queue(fq
, skb
, fhdr
, IP6CB(skb
)->nhoff
);
575 spin_unlock(&fq
->q
.lock
);
580 IP6_INC_STATS_BH(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMFAILS
);
585 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_INHDRERRORS
);
586 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb_network_header_len(skb
));
590 static const struct inet6_protocol frag_protocol
=
592 .handler
= ipv6_frag_rcv
,
593 .flags
= INET6_PROTO_NOPOLICY
,
597 static struct ctl_table ip6_frags_ns_ctl_table
[] = {
599 .procname
= "ip6frag_high_thresh",
600 .data
= &init_net
.ipv6
.frags
.high_thresh
,
601 .maxlen
= sizeof(int),
603 .proc_handler
= proc_dointvec
606 .procname
= "ip6frag_low_thresh",
607 .data
= &init_net
.ipv6
.frags
.low_thresh
,
608 .maxlen
= sizeof(int),
610 .proc_handler
= proc_dointvec
613 .procname
= "ip6frag_time",
614 .data
= &init_net
.ipv6
.frags
.timeout
,
615 .maxlen
= sizeof(int),
617 .proc_handler
= proc_dointvec_jiffies
,
622 static struct ctl_table ip6_frags_ctl_table
[] = {
624 .procname
= "ip6frag_secret_interval",
625 .data
= &ip6_frags
.secret_interval
,
626 .maxlen
= sizeof(int),
628 .proc_handler
= proc_dointvec_jiffies
,
633 static int __net_init
ip6_frags_ns_sysctl_register(struct net
*net
)
635 struct ctl_table
*table
;
636 struct ctl_table_header
*hdr
;
638 table
= ip6_frags_ns_ctl_table
;
639 if (!net_eq(net
, &init_net
)) {
640 table
= kmemdup(table
, sizeof(ip6_frags_ns_ctl_table
), GFP_KERNEL
);
644 table
[0].data
= &net
->ipv6
.frags
.high_thresh
;
645 table
[1].data
= &net
->ipv6
.frags
.low_thresh
;
646 table
[2].data
= &net
->ipv6
.frags
.timeout
;
649 hdr
= register_net_sysctl_table(net
, net_ipv6_ctl_path
, table
);
653 net
->ipv6
.sysctl
.frags_hdr
= hdr
;
657 if (!net_eq(net
, &init_net
))
663 static void __net_exit
ip6_frags_ns_sysctl_unregister(struct net
*net
)
665 struct ctl_table
*table
;
667 table
= net
->ipv6
.sysctl
.frags_hdr
->ctl_table_arg
;
668 unregister_net_sysctl_table(net
->ipv6
.sysctl
.frags_hdr
);
669 if (!net_eq(net
, &init_net
))
673 static struct ctl_table_header
*ip6_ctl_header
;
675 static int ip6_frags_sysctl_register(void)
677 ip6_ctl_header
= register_net_sysctl_rotable(net_ipv6_ctl_path
,
678 ip6_frags_ctl_table
);
679 return ip6_ctl_header
== NULL
? -ENOMEM
: 0;
682 static void ip6_frags_sysctl_unregister(void)
684 unregister_net_sysctl_table(ip6_ctl_header
);
687 static inline int ip6_frags_ns_sysctl_register(struct net
*net
)
692 static inline void ip6_frags_ns_sysctl_unregister(struct net
*net
)
696 static inline int ip6_frags_sysctl_register(void)
701 static inline void ip6_frags_sysctl_unregister(void)
706 static int __net_init
ipv6_frags_init_net(struct net
*net
)
708 net
->ipv6
.frags
.high_thresh
= IPV6_FRAG_HIGH_THRESH
;
709 net
->ipv6
.frags
.low_thresh
= IPV6_FRAG_LOW_THRESH
;
710 net
->ipv6
.frags
.timeout
= IPV6_FRAG_TIMEOUT
;
712 inet_frags_init_net(&net
->ipv6
.frags
);
714 return ip6_frags_ns_sysctl_register(net
);
717 static void __net_exit
ipv6_frags_exit_net(struct net
*net
)
719 ip6_frags_ns_sysctl_unregister(net
);
720 inet_frags_exit_net(&net
->ipv6
.frags
, &ip6_frags
);
723 static struct pernet_operations ip6_frags_ops
= {
724 .init
= ipv6_frags_init_net
,
725 .exit
= ipv6_frags_exit_net
,
728 int __init
ipv6_frag_init(void)
732 ret
= inet6_add_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
736 ret
= ip6_frags_sysctl_register();
740 ret
= register_pernet_subsys(&ip6_frags_ops
);
744 ip6_frags
.hashfn
= ip6_hashfn
;
745 ip6_frags
.constructor
= ip6_frag_init
;
746 ip6_frags
.destructor
= NULL
;
747 ip6_frags
.skb_free
= NULL
;
748 ip6_frags
.qsize
= sizeof(struct frag_queue
);
749 ip6_frags
.match
= ip6_frag_match
;
750 ip6_frags
.frag_expire
= ip6_frag_expire
;
751 ip6_frags
.secret_interval
= 10 * 60 * HZ
;
752 inet_frags_init(&ip6_frags
);
757 ip6_frags_sysctl_unregister();
759 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
763 void ipv6_frag_exit(void)
765 inet_frags_fini(&ip6_frags
);
766 ip6_frags_sysctl_unregister();
767 unregister_pernet_subsys(&ip6_frags_ops
);
768 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);