2 * IPv6 fragment reassembly
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
10 * Based on: net/ipv4/ip_fragment.c
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
20 * Andi Kleen Make it work with multiple hosts.
21 * More RFC compliance.
23 * Horst von Brand Add missing #include <linux/string.h>
24 * Alexey Kuznetsov SMP races, threading, cleanup.
25 * Patrick McHardy LRU queue of frag heads for evictor.
26 * Mitsuru KANDA @USAGI Register inet6_protocol{}.
28 * YOSHIFUJI,H. @USAGI Always remove fragment header to
29 * calculate ICV correctly.
31 #include <linux/errno.h>
32 #include <linux/types.h>
33 #include <linux/string.h>
34 #include <linux/socket.h>
35 #include <linux/sockios.h>
36 #include <linux/jiffies.h>
37 #include <linux/net.h>
38 #include <linux/list.h>
39 #include <linux/netdevice.h>
40 #include <linux/in6.h>
41 #include <linux/ipv6.h>
42 #include <linux/icmpv6.h>
43 #include <linux/random.h>
44 #include <linux/jhash.h>
50 #include <net/protocol.h>
51 #include <net/transp_v6.h>
52 #include <net/rawv6.h>
53 #include <net/ndisc.h>
54 #include <net/addrconf.h>
56 int sysctl_ip6frag_high_thresh __read_mostly
= 256*1024;
57 int sysctl_ip6frag_low_thresh __read_mostly
= 192*1024;
59 int sysctl_ip6frag_time __read_mostly
= IPV6_FRAG_TIMEOUT
;
63 struct inet6_skb_parm h
;
67 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
71 * Equivalent of ipv4 struct ipq
76 struct hlist_node list
;
77 struct list_head lru_list
; /* lru list member */
79 __u32 id
; /* fragment id */
80 struct in6_addr saddr
;
81 struct in6_addr daddr
;
85 struct timer_list timer
; /* expire timer */
86 struct sk_buff
*fragments
;
92 __u8 last_in
; /* has first/last segment arrived? */
101 #define IP6Q_HASHSZ 64
103 static struct hlist_head ip6_frag_hash
[IP6Q_HASHSZ
];
104 static DEFINE_RWLOCK(ip6_frag_lock
);
105 static u32 ip6_frag_hash_rnd
;
106 static LIST_HEAD(ip6_frag_lru_list
);
107 int ip6_frag_nqueues
= 0;
109 static __inline__
void __fq_unlink(struct frag_queue
*fq
)
111 hlist_del(&fq
->list
);
112 list_del(&fq
->lru_list
);
116 static __inline__
void fq_unlink(struct frag_queue
*fq
)
118 write_lock(&ip6_frag_lock
);
120 write_unlock(&ip6_frag_lock
);
124 * callers should be careful not to use the hash value outside the ipfrag_lock
125 * as doing so could race with ipfrag_hash_rnd being recalculated.
127 static unsigned int ip6qhashfn(u32 id
, struct in6_addr
*saddr
,
128 struct in6_addr
*daddr
)
132 a
= saddr
->s6_addr32
[0];
133 b
= saddr
->s6_addr32
[1];
134 c
= saddr
->s6_addr32
[2];
136 a
+= JHASH_GOLDEN_RATIO
;
137 b
+= JHASH_GOLDEN_RATIO
;
138 c
+= ip6_frag_hash_rnd
;
139 __jhash_mix(a
, b
, c
);
141 a
+= saddr
->s6_addr32
[3];
142 b
+= daddr
->s6_addr32
[0];
143 c
+= daddr
->s6_addr32
[1];
144 __jhash_mix(a
, b
, c
);
146 a
+= daddr
->s6_addr32
[2];
147 b
+= daddr
->s6_addr32
[3];
149 __jhash_mix(a
, b
, c
);
151 return c
& (IP6Q_HASHSZ
- 1);
154 static struct timer_list ip6_frag_secret_timer
;
155 int sysctl_ip6frag_secret_interval __read_mostly
= 10 * 60 * HZ
;
157 static void ip6_frag_secret_rebuild(unsigned long dummy
)
159 unsigned long now
= jiffies
;
162 write_lock(&ip6_frag_lock
);
163 get_random_bytes(&ip6_frag_hash_rnd
, sizeof(u32
));
164 for (i
= 0; i
< IP6Q_HASHSZ
; i
++) {
165 struct frag_queue
*q
;
166 struct hlist_node
*p
, *n
;
168 hlist_for_each_entry_safe(q
, p
, n
, &ip6_frag_hash
[i
], list
) {
169 unsigned int hval
= ip6qhashfn(q
->id
,
176 /* Relink to new hash chain. */
177 hlist_add_head(&q
->list
,
178 &ip6_frag_hash
[hval
]);
183 write_unlock(&ip6_frag_lock
);
185 mod_timer(&ip6_frag_secret_timer
, now
+ sysctl_ip6frag_secret_interval
);
188 atomic_t ip6_frag_mem
= ATOMIC_INIT(0);
190 /* Memory Tracking Functions. */
191 static inline void frag_kfree_skb(struct sk_buff
*skb
, int *work
)
194 *work
-= skb
->truesize
;
195 atomic_sub(skb
->truesize
, &ip6_frag_mem
);
199 static inline void frag_free_queue(struct frag_queue
*fq
, int *work
)
202 *work
-= sizeof(struct frag_queue
);
203 atomic_sub(sizeof(struct frag_queue
), &ip6_frag_mem
);
207 static inline struct frag_queue
*frag_alloc_queue(void)
209 struct frag_queue
*fq
= kzalloc(sizeof(struct frag_queue
), GFP_ATOMIC
);
213 atomic_add(sizeof(struct frag_queue
), &ip6_frag_mem
);
217 /* Destruction primitives. */
219 /* Complete destruction of fq. */
220 static void ip6_frag_destroy(struct frag_queue
*fq
, int *work
)
224 BUG_TRAP(fq
->last_in
&COMPLETE
);
225 BUG_TRAP(del_timer(&fq
->timer
) == 0);
227 /* Release all fragment data. */
230 struct sk_buff
*xp
= fp
->next
;
232 frag_kfree_skb(fp
, work
);
236 frag_free_queue(fq
, work
);
239 static __inline__
void fq_put(struct frag_queue
*fq
, int *work
)
241 if (atomic_dec_and_test(&fq
->refcnt
))
242 ip6_frag_destroy(fq
, work
);
245 /* Kill fq entry. It is not destroyed immediately,
246 * because caller (and someone more) holds reference count.
248 static __inline__
void fq_kill(struct frag_queue
*fq
)
250 if (del_timer(&fq
->timer
))
251 atomic_dec(&fq
->refcnt
);
253 if (!(fq
->last_in
& COMPLETE
)) {
255 atomic_dec(&fq
->refcnt
);
256 fq
->last_in
|= COMPLETE
;
260 static void ip6_evictor(void)
262 struct frag_queue
*fq
;
263 struct list_head
*tmp
;
266 work
= atomic_read(&ip6_frag_mem
) - sysctl_ip6frag_low_thresh
;
271 read_lock(&ip6_frag_lock
);
272 if (list_empty(&ip6_frag_lru_list
)) {
273 read_unlock(&ip6_frag_lock
);
276 tmp
= ip6_frag_lru_list
.next
;
277 fq
= list_entry(tmp
, struct frag_queue
, lru_list
);
278 atomic_inc(&fq
->refcnt
);
279 read_unlock(&ip6_frag_lock
);
281 spin_lock(&fq
->lock
);
282 if (!(fq
->last_in
&COMPLETE
))
284 spin_unlock(&fq
->lock
);
287 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS
);
291 static void ip6_frag_expire(unsigned long data
)
293 struct frag_queue
*fq
= (struct frag_queue
*) data
;
294 struct net_device
*dev
;
296 spin_lock(&fq
->lock
);
298 if (fq
->last_in
& COMPLETE
)
303 IP6_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT
);
304 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS
);
306 /* Don't send error if the first segment did not arrive. */
307 if (!(fq
->last_in
&FIRST_IN
) || !fq
->fragments
)
310 dev
= dev_get_by_index(fq
->iif
);
315 But use as source device on which LAST ARRIVED
316 segment was received. And do not use fq->dev
317 pointer directly, device might already disappeared.
319 fq
->fragments
->dev
= dev
;
320 icmpv6_send(fq
->fragments
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_FRAGTIME
, 0, dev
);
323 spin_unlock(&fq
->lock
);
327 /* Creation primitives. */
330 static struct frag_queue
*ip6_frag_intern(struct frag_queue
*fq_in
)
332 struct frag_queue
*fq
;
335 struct hlist_node
*n
;
338 write_lock(&ip6_frag_lock
);
339 hash
= ip6qhashfn(fq_in
->id
, &fq_in
->saddr
, &fq_in
->daddr
);
341 hlist_for_each_entry(fq
, n
, &ip6_frag_hash
[hash
], list
) {
342 if (fq
->id
== fq_in
->id
&&
343 ipv6_addr_equal(&fq_in
->saddr
, &fq
->saddr
) &&
344 ipv6_addr_equal(&fq_in
->daddr
, &fq
->daddr
)) {
345 atomic_inc(&fq
->refcnt
);
346 write_unlock(&ip6_frag_lock
);
347 fq_in
->last_in
|= COMPLETE
;
355 if (!mod_timer(&fq
->timer
, jiffies
+ sysctl_ip6frag_time
))
356 atomic_inc(&fq
->refcnt
);
358 atomic_inc(&fq
->refcnt
);
359 hlist_add_head(&fq
->list
, &ip6_frag_hash
[hash
]);
360 INIT_LIST_HEAD(&fq
->lru_list
);
361 list_add_tail(&fq
->lru_list
, &ip6_frag_lru_list
);
363 write_unlock(&ip6_frag_lock
);
368 static struct frag_queue
*
369 ip6_frag_create(u32 id
, struct in6_addr
*src
, struct in6_addr
*dst
)
371 struct frag_queue
*fq
;
373 if ((fq
= frag_alloc_queue()) == NULL
)
377 ipv6_addr_copy(&fq
->saddr
, src
);
378 ipv6_addr_copy(&fq
->daddr
, dst
);
380 init_timer(&fq
->timer
);
381 fq
->timer
.function
= ip6_frag_expire
;
382 fq
->timer
.data
= (long) fq
;
383 spin_lock_init(&fq
->lock
);
384 atomic_set(&fq
->refcnt
, 1);
386 return ip6_frag_intern(fq
);
389 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS
);
393 static __inline__
struct frag_queue
*
394 fq_find(u32 id
, struct in6_addr
*src
, struct in6_addr
*dst
)
396 struct frag_queue
*fq
;
397 struct hlist_node
*n
;
400 read_lock(&ip6_frag_lock
);
401 hash
= ip6qhashfn(id
, src
, dst
);
402 hlist_for_each_entry(fq
, n
, &ip6_frag_hash
[hash
], list
) {
404 ipv6_addr_equal(src
, &fq
->saddr
) &&
405 ipv6_addr_equal(dst
, &fq
->daddr
)) {
406 atomic_inc(&fq
->refcnt
);
407 read_unlock(&ip6_frag_lock
);
411 read_unlock(&ip6_frag_lock
);
413 return ip6_frag_create(id
, src
, dst
);
417 static void ip6_frag_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
418 struct frag_hdr
*fhdr
, int nhoff
)
420 struct sk_buff
*prev
, *next
;
423 if (fq
->last_in
& COMPLETE
)
426 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
427 end
= offset
+ (ntohs(skb
->nh
.ipv6h
->payload_len
) -
428 ((u8
*) (fhdr
+ 1) - (u8
*) (skb
->nh
.ipv6h
+ 1)));
430 if ((unsigned int)end
> IPV6_MAXPLEN
) {
431 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS
);
432 icmpv6_param_prob(skb
,ICMPV6_HDR_FIELD
, (u8
*)&fhdr
->frag_off
- skb
->nh
.raw
);
436 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
437 skb
->csum
= csum_sub(skb
->csum
,
438 csum_partial(skb
->nh
.raw
, (u8
*)(fhdr
+1)-skb
->nh
.raw
, 0));
440 /* Is this the final fragment? */
441 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
442 /* If we already have some bits beyond end
443 * or have different end, the segment is corrupted.
446 ((fq
->last_in
& LAST_IN
) && end
!= fq
->len
))
448 fq
->last_in
|= LAST_IN
;
451 /* Check if the fragment is rounded to 8 bytes.
452 * Required by the RFC.
455 /* RFC2460 says always send parameter problem in
458 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS
);
459 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
460 offsetof(struct ipv6hdr
, payload_len
));
464 /* Some bits beyond end -> corruption. */
465 if (fq
->last_in
& LAST_IN
)
474 /* Point into the IP datagram 'data' part. */
475 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
))
478 if (pskb_trim_rcsum(skb
, end
- offset
))
481 /* Find out which fragments are in front and at the back of us
482 * in the chain of fragments so far. We must know where to put
483 * this fragment, right?
486 for(next
= fq
->fragments
; next
!= NULL
; next
= next
->next
) {
487 if (FRAG6_CB(next
)->offset
>= offset
)
492 /* We found where to put this one. Check for overlap with
493 * preceding fragment, and, if needed, align things so that
494 * any overlaps are eliminated.
497 int i
= (FRAG6_CB(prev
)->offset
+ prev
->len
) - offset
;
503 if (!pskb_pull(skb
, i
))
505 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
)
506 skb
->ip_summed
= CHECKSUM_NONE
;
510 /* Look for overlap with succeeding segments.
511 * If we can merge fragments, do it.
513 while (next
&& FRAG6_CB(next
)->offset
< end
) {
514 int i
= end
- FRAG6_CB(next
)->offset
; /* overlap is 'i' bytes */
517 /* Eat head of the next overlapped fragment
518 * and leave the loop. The next ones cannot overlap.
520 if (!pskb_pull(next
, i
))
522 FRAG6_CB(next
)->offset
+= i
; /* next fragment */
524 if (next
->ip_summed
!= CHECKSUM_UNNECESSARY
)
525 next
->ip_summed
= CHECKSUM_NONE
;
528 struct sk_buff
*free_it
= next
;
530 /* Old fragment is completely overridden with
538 fq
->fragments
= next
;
540 fq
->meat
-= free_it
->len
;
541 frag_kfree_skb(free_it
, NULL
);
545 FRAG6_CB(skb
)->offset
= offset
;
547 /* Insert this fragment in the chain of fragments. */
555 fq
->iif
= skb
->dev
->ifindex
;
557 skb_get_timestamp(skb
, &fq
->stamp
);
558 fq
->meat
+= skb
->len
;
559 atomic_add(skb
->truesize
, &ip6_frag_mem
);
561 /* The first fragment.
562 * nhoffset is obtained from the first fragment, of course.
565 fq
->nhoffset
= nhoff
;
566 fq
->last_in
|= FIRST_IN
;
568 write_lock(&ip6_frag_lock
);
569 list_move_tail(&fq
->lru_list
, &ip6_frag_lru_list
);
570 write_unlock(&ip6_frag_lock
);
574 IP6_INC_STATS(IPSTATS_MIB_REASMFAILS
);
579 * Check if this packet is complete.
580 * Returns NULL on failure by any reason, and pointer
581 * to current nexthdr field in reassembled frame.
583 * It is called with locked fq, and caller must check that
584 * queue is eligible for reassembly i.e. it is not COMPLETE,
585 * the last and the first frames arrived and all the bits are here.
587 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
**skb_in
,
588 struct net_device
*dev
)
590 struct sk_buff
*fp
, *head
= fq
->fragments
;
596 BUG_TRAP(head
!= NULL
);
597 BUG_TRAP(FRAG6_CB(head
)->offset
== 0);
599 /* Unfragmented part is taken from the first segment. */
600 payload_len
= (head
->data
- head
->nh
.raw
) - sizeof(struct ipv6hdr
) + fq
->len
- sizeof(struct frag_hdr
);
601 if (payload_len
> IPV6_MAXPLEN
)
604 /* Head of list must not be cloned. */
605 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
))
608 /* If the first fragment is fragmented itself, we split
609 * it to two chunks: the first with data and paged part
610 * and the second, holding only fragments. */
611 if (skb_shinfo(head
)->frag_list
) {
612 struct sk_buff
*clone
;
615 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
)
617 clone
->next
= head
->next
;
619 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
620 skb_shinfo(head
)->frag_list
= NULL
;
621 for (i
=0; i
<skb_shinfo(head
)->nr_frags
; i
++)
622 plen
+= skb_shinfo(head
)->frags
[i
].size
;
623 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
624 head
->data_len
-= clone
->len
;
625 head
->len
-= clone
->len
;
627 clone
->ip_summed
= head
->ip_summed
;
628 atomic_add(clone
->truesize
, &ip6_frag_mem
);
631 /* We have to remove fragment header from datagram and to relocate
632 * header in order to calculate ICV correctly. */
633 nhoff
= fq
->nhoffset
;
634 head
->nh
.raw
[nhoff
] = head
->h
.raw
[0];
635 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
636 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
637 head
->mac
.raw
+= sizeof(struct frag_hdr
);
638 head
->nh
.raw
+= sizeof(struct frag_hdr
);
640 skb_shinfo(head
)->frag_list
= head
->next
;
641 head
->h
.raw
= head
->data
;
642 skb_push(head
, head
->data
- head
->nh
.raw
);
643 atomic_sub(head
->truesize
, &ip6_frag_mem
);
645 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
646 head
->data_len
+= fp
->len
;
647 head
->len
+= fp
->len
;
648 if (head
->ip_summed
!= fp
->ip_summed
)
649 head
->ip_summed
= CHECKSUM_NONE
;
650 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
651 head
->csum
= csum_add(head
->csum
, fp
->csum
);
652 head
->truesize
+= fp
->truesize
;
653 atomic_sub(fp
->truesize
, &ip6_frag_mem
);
658 skb_set_timestamp(head
, &fq
->stamp
);
659 head
->nh
.ipv6h
->payload_len
= htons(payload_len
);
660 IP6CB(head
)->nhoff
= nhoff
;
664 /* Yes, and fold redundant checksum back. 8) */
665 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
666 head
->csum
= csum_partial(head
->nh
.raw
, head
->h
.raw
-head
->nh
.raw
, head
->csum
);
668 IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS
);
669 fq
->fragments
= NULL
;
674 printk(KERN_DEBUG
"ip6_frag_reasm: payload len = %d\n", payload_len
);
678 printk(KERN_DEBUG
"ip6_frag_reasm: no memory for reassembly\n");
680 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS
);
684 static int ipv6_frag_rcv(struct sk_buff
**skbp
)
686 struct sk_buff
*skb
= *skbp
;
687 struct net_device
*dev
= skb
->dev
;
688 struct frag_hdr
*fhdr
;
689 struct frag_queue
*fq
;
694 IP6_INC_STATS_BH(IPSTATS_MIB_REASMREQDS
);
696 /* Jumbo payload inhibits frag. header */
697 if (hdr
->payload_len
==0) {
698 IP6_INC_STATS(IPSTATS_MIB_INHDRERRORS
);
699 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb
->h
.raw
-skb
->nh
.raw
);
702 if (!pskb_may_pull(skb
, (skb
->h
.raw
-skb
->data
)+sizeof(struct frag_hdr
))) {
703 IP6_INC_STATS(IPSTATS_MIB_INHDRERRORS
);
704 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb
->h
.raw
-skb
->nh
.raw
);
709 fhdr
= (struct frag_hdr
*)skb
->h
.raw
;
711 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
712 /* It is not a fragmented frame */
713 skb
->h
.raw
+= sizeof(struct frag_hdr
);
714 IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS
);
716 IP6CB(skb
)->nhoff
= (u8
*)fhdr
- skb
->nh
.raw
;
720 if (atomic_read(&ip6_frag_mem
) > sysctl_ip6frag_high_thresh
)
723 if ((fq
= fq_find(fhdr
->identification
, &hdr
->saddr
, &hdr
->daddr
)) != NULL
) {
726 spin_lock(&fq
->lock
);
728 ip6_frag_queue(fq
, skb
, fhdr
, IP6CB(skb
)->nhoff
);
730 if (fq
->last_in
== (FIRST_IN
|LAST_IN
) &&
732 ret
= ip6_frag_reasm(fq
, skbp
, dev
);
734 spin_unlock(&fq
->lock
);
739 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS
);
744 static struct inet6_protocol frag_protocol
=
746 .handler
= ipv6_frag_rcv
,
747 .flags
= INET6_PROTO_NOPOLICY
,
750 void __init
ipv6_frag_init(void)
752 if (inet6_add_protocol(&frag_protocol
, IPPROTO_FRAGMENT
) < 0)
753 printk(KERN_ERR
"ipv6_frag_init: Could not register protocol\n");
755 ip6_frag_hash_rnd
= (u32
) ((num_physpages
^ (num_physpages
>>7)) ^
756 (jiffies
^ (jiffies
>> 6)));
758 init_timer(&ip6_frag_secret_timer
);
759 ip6_frag_secret_timer
.function
= ip6_frag_secret_rebuild
;
760 ip6_frag_secret_timer
.expires
= jiffies
+ sysctl_ip6frag_secret_interval
;
761 add_timer(&ip6_frag_secret_timer
);