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>
63 static const char ip6_frag_cache_name
[] = "ip6-frags";
65 static u8
ip6_frag_ecn(const struct ipv6hdr
*ipv6h
)
67 return 1 << (ipv6_get_dsfield(ipv6h
) & INET_ECN_MASK
);
70 static struct inet_frags ip6_frags
;
72 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
73 struct net_device
*dev
);
75 void ip6_frag_init(struct inet_frag_queue
*q
, const void *a
)
77 struct frag_queue
*fq
= container_of(q
, struct frag_queue
, q
);
78 const struct frag_v6_compare_key
*key
= a
;
83 EXPORT_SYMBOL(ip6_frag_init
);
85 void ip6_expire_frag_queue(struct net
*net
, struct frag_queue
*fq
)
87 struct net_device
*dev
= NULL
;
91 spin_lock(&fq
->q
.lock
);
93 if (fq
->q
.flags
& INET_FRAG_COMPLETE
)
96 inet_frag_kill(&fq
->q
);
98 dev
= dev_get_by_index_rcu(net
, fq
->iif
);
102 __IP6_INC_STATS(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
103 __IP6_INC_STATS(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMTIMEOUT
);
105 /* Don't send error if the first segment did not arrive. */
106 head
= fq
->q
.fragments
;
107 if (!(fq
->q
.flags
& INET_FRAG_FIRST_IN
) || !head
)
110 /* But use as source device on which LAST ARRIVED
111 * segment was received. And do not use fq->dev
112 * pointer directly, device might already disappeared.
116 spin_unlock(&fq
->q
.lock
);
118 icmpv6_send(head
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_FRAGTIME
, 0);
123 spin_unlock(&fq
->q
.lock
);
126 inet_frag_put(&fq
->q
);
128 EXPORT_SYMBOL(ip6_expire_frag_queue
);
130 static void ip6_frag_expire(struct timer_list
*t
)
132 struct inet_frag_queue
*frag
= from_timer(frag
, t
, timer
);
133 struct frag_queue
*fq
;
136 fq
= container_of(frag
, struct frag_queue
, q
);
137 net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
139 ip6_expire_frag_queue(net
, fq
);
142 static struct frag_queue
*
143 fq_find(struct net
*net
, __be32 id
, const struct ipv6hdr
*hdr
, int iif
)
145 struct frag_v6_compare_key key
= {
149 .user
= IP6_DEFRAG_LOCAL_DELIVER
,
152 struct inet_frag_queue
*q
;
154 if (!(ipv6_addr_type(&hdr
->daddr
) & (IPV6_ADDR_MULTICAST
|
155 IPV6_ADDR_LINKLOCAL
)))
158 q
= inet_frag_find(&net
->ipv6
.frags
, &key
);
162 return container_of(q
, struct frag_queue
, q
);
165 static int ip6_frag_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
166 struct frag_hdr
*fhdr
, int nhoff
)
168 struct sk_buff
*prev
, *next
;
169 struct net_device
*dev
;
170 int offset
, end
, fragsize
;
171 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
174 if (fq
->q
.flags
& INET_FRAG_COMPLETE
)
177 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
178 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
179 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
181 if ((unsigned int)end
> IPV6_MAXPLEN
) {
182 __IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
183 IPSTATS_MIB_INHDRERRORS
);
184 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
185 ((u8
*)&fhdr
->frag_off
-
186 skb_network_header(skb
)));
190 ecn
= ip6_frag_ecn(ipv6_hdr(skb
));
192 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
193 const unsigned char *nh
= skb_network_header(skb
);
194 skb
->csum
= csum_sub(skb
->csum
,
195 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
199 /* Is this the final fragment? */
200 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
201 /* If we already have some bits beyond end
202 * or have different end, the segment is corrupted.
204 if (end
< fq
->q
.len
||
205 ((fq
->q
.flags
& INET_FRAG_LAST_IN
) && end
!= fq
->q
.len
))
207 fq
->q
.flags
|= INET_FRAG_LAST_IN
;
210 /* Check if the fragment is rounded to 8 bytes.
211 * Required by the RFC.
214 /* RFC2460 says always send parameter problem in
217 __IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
218 IPSTATS_MIB_INHDRERRORS
);
219 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
220 offsetof(struct ipv6hdr
, payload_len
));
223 if (end
> fq
->q
.len
) {
224 /* Some bits beyond end -> corruption. */
225 if (fq
->q
.flags
& INET_FRAG_LAST_IN
)
234 /* Point into the IP datagram 'data' part. */
235 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
))
238 if (pskb_trim_rcsum(skb
, end
- offset
))
241 /* Find out which fragments are in front and at the back of us
242 * in the chain of fragments so far. We must know where to put
243 * this fragment, right?
245 prev
= fq
->q
.fragments_tail
;
246 if (!prev
|| prev
->ip_defrag_offset
< offset
) {
251 for (next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
252 if (next
->ip_defrag_offset
>= offset
)
258 /* RFC5722, Section 4, amended by Errata ID : 3089
259 * When reassembling an IPv6 datagram, if
260 * one or more its constituent fragments is determined to be an
261 * overlapping fragment, the entire datagram (and any constituent
262 * fragments) MUST be silently discarded.
265 /* Check for overlap with preceding fragment. */
267 (prev
->ip_defrag_offset
+ prev
->len
) > offset
)
270 /* Look for overlap with succeeding segment. */
271 if (next
&& next
->ip_defrag_offset
< end
)
274 /* Note : skb->ip_defrag_offset and skb->dev share the same location */
277 fq
->iif
= dev
->ifindex
;
278 /* Makes sure compiler wont do silly aliasing games */
280 skb
->ip_defrag_offset
= offset
;
282 /* Insert this fragment in the chain of fragments. */
285 fq
->q
.fragments_tail
= skb
;
289 fq
->q
.fragments
= skb
;
291 fq
->q
.stamp
= skb
->tstamp
;
292 fq
->q
.meat
+= skb
->len
;
294 add_frag_mem_limit(fq
->q
.net
, skb
->truesize
);
296 fragsize
= -skb_network_offset(skb
) + skb
->len
;
297 if (fragsize
> fq
->q
.max_size
)
298 fq
->q
.max_size
= fragsize
;
300 /* The first fragment.
301 * nhoffset is obtained from the first fragment, of course.
304 fq
->nhoffset
= nhoff
;
305 fq
->q
.flags
|= INET_FRAG_FIRST_IN
;
308 if (fq
->q
.flags
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
309 fq
->q
.meat
== fq
->q
.len
) {
311 unsigned long orefdst
= skb
->_skb_refdst
;
313 skb
->_skb_refdst
= 0UL;
314 res
= ip6_frag_reasm(fq
, prev
, dev
);
315 skb
->_skb_refdst
= orefdst
;
323 inet_frag_kill(&fq
->q
);
325 __IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
326 IPSTATS_MIB_REASMFAILS
);
332 * Check if this packet is complete.
333 * Returns NULL on failure by any reason, and pointer
334 * to current nexthdr field in reassembled frame.
336 * It is called with locked fq, and caller must check that
337 * queue is eligible for reassembly i.e. it is not COMPLETE,
338 * the last and the first frames arrived and all the bits are here.
340 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
341 struct net_device
*dev
)
343 struct net
*net
= container_of(fq
->q
.net
, struct net
, ipv6
.frags
);
344 struct sk_buff
*fp
, *head
= fq
->q
.fragments
;
350 inet_frag_kill(&fq
->q
);
352 ecn
= ip_frag_ecn_table
[fq
->ecn
];
353 if (unlikely(ecn
== 0xff))
356 /* Make the one we just received the head. */
359 fp
= skb_clone(head
, GFP_ATOMIC
);
364 fp
->next
= head
->next
;
366 fq
->q
.fragments_tail
= fp
;
369 skb_morph(head
, fq
->q
.fragments
);
370 head
->next
= fq
->q
.fragments
->next
;
372 consume_skb(fq
->q
.fragments
);
373 fq
->q
.fragments
= head
;
376 WARN_ON(head
== NULL
);
377 WARN_ON(head
->ip_defrag_offset
!= 0);
379 /* Unfragmented part is taken from the first segment. */
380 payload_len
= ((head
->data
- skb_network_header(head
)) -
381 sizeof(struct ipv6hdr
) + fq
->q
.len
-
382 sizeof(struct frag_hdr
));
383 if (payload_len
> IPV6_MAXPLEN
)
386 /* Head of list must not be cloned. */
387 if (skb_unclone(head
, GFP_ATOMIC
))
390 /* If the first fragment is fragmented itself, we split
391 * it to two chunks: the first with data and paged part
392 * and the second, holding only fragments. */
393 if (skb_has_frag_list(head
)) {
394 struct sk_buff
*clone
;
397 clone
= alloc_skb(0, GFP_ATOMIC
);
400 clone
->next
= head
->next
;
402 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
403 skb_frag_list_init(head
);
404 for (i
= 0; i
< skb_shinfo(head
)->nr_frags
; i
++)
405 plen
+= skb_frag_size(&skb_shinfo(head
)->frags
[i
]);
406 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
407 head
->data_len
-= clone
->len
;
408 head
->len
-= clone
->len
;
410 clone
->ip_summed
= head
->ip_summed
;
411 add_frag_mem_limit(fq
->q
.net
, clone
->truesize
);
414 /* We have to remove fragment header from datagram and to relocate
415 * header in order to calculate ICV correctly. */
416 nhoff
= fq
->nhoffset
;
417 skb_network_header(head
)[nhoff
] = skb_transport_header(head
)[0];
418 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
419 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
420 if (skb_mac_header_was_set(head
))
421 head
->mac_header
+= sizeof(struct frag_hdr
);
422 head
->network_header
+= sizeof(struct frag_hdr
);
424 skb_reset_transport_header(head
);
425 skb_push(head
, head
->data
- skb_network_header(head
));
427 sum_truesize
= head
->truesize
;
428 for (fp
= head
->next
; fp
;) {
431 struct sk_buff
*next
= fp
->next
;
433 sum_truesize
+= fp
->truesize
;
434 if (head
->ip_summed
!= fp
->ip_summed
)
435 head
->ip_summed
= CHECKSUM_NONE
;
436 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
437 head
->csum
= csum_add(head
->csum
, fp
->csum
);
439 if (skb_try_coalesce(head
, fp
, &headstolen
, &delta
)) {
440 kfree_skb_partial(fp
, headstolen
);
442 if (!skb_shinfo(head
)->frag_list
)
443 skb_shinfo(head
)->frag_list
= fp
;
444 head
->data_len
+= fp
->len
;
445 head
->len
+= fp
->len
;
446 head
->truesize
+= fp
->truesize
;
450 sub_frag_mem_limit(fq
->q
.net
, sum_truesize
);
454 head
->tstamp
= fq
->q
.stamp
;
455 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
456 ipv6_change_dsfield(ipv6_hdr(head
), 0xff, ecn
);
457 IP6CB(head
)->nhoff
= nhoff
;
458 IP6CB(head
)->flags
|= IP6SKB_FRAGMENTED
;
459 IP6CB(head
)->frag_max_size
= fq
->q
.max_size
;
461 /* Yes, and fold redundant checksum back. 8) */
462 skb_postpush_rcsum(head
, skb_network_header(head
),
463 skb_network_header_len(head
));
466 __IP6_INC_STATS(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMOKS
);
468 fq
->q
.fragments
= NULL
;
469 fq
->q
.fragments_tail
= NULL
;
473 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len
);
476 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
479 __IP6_INC_STATS(net
, __in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
484 static int ipv6_frag_rcv(struct sk_buff
*skb
)
486 struct frag_hdr
*fhdr
;
487 struct frag_queue
*fq
;
488 const struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
489 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
492 if (IP6CB(skb
)->flags
& IP6SKB_FRAGMENTED
)
495 __IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMREQDS
);
497 /* Jumbo payload inhibits frag. header */
498 if (hdr
->payload_len
== 0)
501 if (!pskb_may_pull(skb
, (skb_transport_offset(skb
) +
502 sizeof(struct frag_hdr
))))
506 fhdr
= (struct frag_hdr
*)skb_transport_header(skb
);
508 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
509 /* It is not a fragmented frame */
510 skb
->transport_header
+= sizeof(struct frag_hdr
);
512 ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMOKS
);
514 IP6CB(skb
)->nhoff
= (u8
*)fhdr
- skb_network_header(skb
);
515 IP6CB(skb
)->flags
|= IP6SKB_FRAGMENTED
;
519 iif
= skb
->dev
? skb
->dev
->ifindex
: 0;
520 fq
= fq_find(net
, fhdr
->identification
, hdr
, iif
);
524 spin_lock(&fq
->q
.lock
);
527 ret
= ip6_frag_queue(fq
, skb
, fhdr
, IP6CB(skb
)->nhoff
);
529 spin_unlock(&fq
->q
.lock
);
530 inet_frag_put(&fq
->q
);
534 __IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_REASMFAILS
);
539 __IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
540 IPSTATS_MIB_INHDRERRORS
);
541 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb_network_header_len(skb
));
545 static const struct inet6_protocol frag_protocol
= {
546 .handler
= ipv6_frag_rcv
,
547 .flags
= INET6_PROTO_NOPOLICY
,
552 static struct ctl_table ip6_frags_ns_ctl_table
[] = {
554 .procname
= "ip6frag_high_thresh",
555 .data
= &init_net
.ipv6
.frags
.high_thresh
,
556 .maxlen
= sizeof(unsigned long),
558 .proc_handler
= proc_doulongvec_minmax
,
559 .extra1
= &init_net
.ipv6
.frags
.low_thresh
562 .procname
= "ip6frag_low_thresh",
563 .data
= &init_net
.ipv6
.frags
.low_thresh
,
564 .maxlen
= sizeof(unsigned long),
566 .proc_handler
= proc_doulongvec_minmax
,
567 .extra2
= &init_net
.ipv6
.frags
.high_thresh
570 .procname
= "ip6frag_time",
571 .data
= &init_net
.ipv6
.frags
.timeout
,
572 .maxlen
= sizeof(int),
574 .proc_handler
= proc_dointvec_jiffies
,
579 /* secret interval has been deprecated */
580 static int ip6_frags_secret_interval_unused
;
581 static struct ctl_table ip6_frags_ctl_table
[] = {
583 .procname
= "ip6frag_secret_interval",
584 .data
= &ip6_frags_secret_interval_unused
,
585 .maxlen
= sizeof(int),
587 .proc_handler
= proc_dointvec_jiffies
,
592 static int __net_init
ip6_frags_ns_sysctl_register(struct net
*net
)
594 struct ctl_table
*table
;
595 struct ctl_table_header
*hdr
;
597 table
= ip6_frags_ns_ctl_table
;
598 if (!net_eq(net
, &init_net
)) {
599 table
= kmemdup(table
, sizeof(ip6_frags_ns_ctl_table
), GFP_KERNEL
);
603 table
[0].data
= &net
->ipv6
.frags
.high_thresh
;
604 table
[0].extra1
= &net
->ipv6
.frags
.low_thresh
;
605 table
[0].extra2
= &init_net
.ipv6
.frags
.high_thresh
;
606 table
[1].data
= &net
->ipv6
.frags
.low_thresh
;
607 table
[1].extra2
= &net
->ipv6
.frags
.high_thresh
;
608 table
[2].data
= &net
->ipv6
.frags
.timeout
;
611 hdr
= register_net_sysctl(net
, "net/ipv6", table
);
615 net
->ipv6
.sysctl
.frags_hdr
= hdr
;
619 if (!net_eq(net
, &init_net
))
625 static void __net_exit
ip6_frags_ns_sysctl_unregister(struct net
*net
)
627 struct ctl_table
*table
;
629 table
= net
->ipv6
.sysctl
.frags_hdr
->ctl_table_arg
;
630 unregister_net_sysctl_table(net
->ipv6
.sysctl
.frags_hdr
);
631 if (!net_eq(net
, &init_net
))
635 static struct ctl_table_header
*ip6_ctl_header
;
637 static int ip6_frags_sysctl_register(void)
639 ip6_ctl_header
= register_net_sysctl(&init_net
, "net/ipv6",
640 ip6_frags_ctl_table
);
641 return ip6_ctl_header
== NULL
? -ENOMEM
: 0;
644 static void ip6_frags_sysctl_unregister(void)
646 unregister_net_sysctl_table(ip6_ctl_header
);
649 static int ip6_frags_ns_sysctl_register(struct net
*net
)
654 static void ip6_frags_ns_sysctl_unregister(struct net
*net
)
658 static int ip6_frags_sysctl_register(void)
663 static void ip6_frags_sysctl_unregister(void)
668 static int __net_init
ipv6_frags_init_net(struct net
*net
)
672 net
->ipv6
.frags
.high_thresh
= IPV6_FRAG_HIGH_THRESH
;
673 net
->ipv6
.frags
.low_thresh
= IPV6_FRAG_LOW_THRESH
;
674 net
->ipv6
.frags
.timeout
= IPV6_FRAG_TIMEOUT
;
675 net
->ipv6
.frags
.f
= &ip6_frags
;
677 res
= inet_frags_init_net(&net
->ipv6
.frags
);
681 res
= ip6_frags_ns_sysctl_register(net
);
683 inet_frags_exit_net(&net
->ipv6
.frags
);
687 static void __net_exit
ipv6_frags_exit_net(struct net
*net
)
689 ip6_frags_ns_sysctl_unregister(net
);
690 inet_frags_exit_net(&net
->ipv6
.frags
);
693 static struct pernet_operations ip6_frags_ops
= {
694 .init
= ipv6_frags_init_net
,
695 .exit
= ipv6_frags_exit_net
,
698 static u32
ip6_key_hashfn(const void *data
, u32 len
, u32 seed
)
701 sizeof(struct frag_v6_compare_key
) / sizeof(u32
), seed
);
704 static u32
ip6_obj_hashfn(const void *data
, u32 len
, u32 seed
)
706 const struct inet_frag_queue
*fq
= data
;
708 return jhash2((const u32
*)&fq
->key
.v6
,
709 sizeof(struct frag_v6_compare_key
) / sizeof(u32
), seed
);
712 static int ip6_obj_cmpfn(struct rhashtable_compare_arg
*arg
, const void *ptr
)
714 const struct frag_v6_compare_key
*key
= arg
->key
;
715 const struct inet_frag_queue
*fq
= ptr
;
717 return !!memcmp(&fq
->key
, key
, sizeof(*key
));
720 const struct rhashtable_params ip6_rhash_params
= {
721 .head_offset
= offsetof(struct inet_frag_queue
, node
),
722 .hashfn
= ip6_key_hashfn
,
723 .obj_hashfn
= ip6_obj_hashfn
,
724 .obj_cmpfn
= ip6_obj_cmpfn
,
725 .automatic_shrinking
= true,
727 EXPORT_SYMBOL(ip6_rhash_params
);
729 int __init
ipv6_frag_init(void)
733 ip6_frags
.constructor
= ip6_frag_init
;
734 ip6_frags
.destructor
= NULL
;
735 ip6_frags
.qsize
= sizeof(struct frag_queue
);
736 ip6_frags
.frag_expire
= ip6_frag_expire
;
737 ip6_frags
.frags_cache_name
= ip6_frag_cache_name
;
738 ip6_frags
.rhash_params
= ip6_rhash_params
;
739 ret
= inet_frags_init(&ip6_frags
);
743 ret
= inet6_add_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
747 ret
= ip6_frags_sysctl_register();
751 ret
= register_pernet_subsys(&ip6_frags_ops
);
759 ip6_frags_sysctl_unregister();
761 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);
763 inet_frags_fini(&ip6_frags
);
767 void ipv6_frag_exit(void)
769 inet_frags_fini(&ip6_frags
);
770 ip6_frags_sysctl_unregister();
771 unregister_pernet_subsys(&ip6_frags_ops
);
772 inet6_del_protocol(&frag_protocol
, IPPROTO_FRAGMENT
);