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/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>
57 int sysctl_ip6frag_high_thresh __read_mostly
= 256*1024;
58 int sysctl_ip6frag_low_thresh __read_mostly
= 192*1024;
60 int sysctl_ip6frag_time __read_mostly
= IPV6_FRAG_TIMEOUT
;
64 struct inet6_skb_parm h
;
68 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
72 * Equivalent of ipv4 struct ipq
77 struct hlist_node list
;
78 struct list_head lru_list
; /* lru list member */
80 __be32 id
; /* fragment id */
81 struct in6_addr saddr
;
82 struct in6_addr daddr
;
86 struct timer_list timer
; /* expire timer */
87 struct sk_buff
*fragments
;
93 __u8 last_in
; /* has first/last segment arrived? */
102 #define IP6Q_HASHSZ 64
104 static struct hlist_head ip6_frag_hash
[IP6Q_HASHSZ
];
105 static DEFINE_RWLOCK(ip6_frag_lock
);
106 static u32 ip6_frag_hash_rnd
;
107 static LIST_HEAD(ip6_frag_lru_list
);
108 int ip6_frag_nqueues
= 0;
110 static __inline__
void __fq_unlink(struct frag_queue
*fq
)
112 hlist_del(&fq
->list
);
113 list_del(&fq
->lru_list
);
117 static __inline__
void fq_unlink(struct frag_queue
*fq
)
119 write_lock(&ip6_frag_lock
);
121 write_unlock(&ip6_frag_lock
);
125 * callers should be careful not to use the hash value outside the ipfrag_lock
126 * as doing so could race with ipfrag_hash_rnd being recalculated.
128 static unsigned int ip6qhashfn(__be32 id
, struct in6_addr
*saddr
,
129 struct in6_addr
*daddr
)
133 a
= (__force u32
)saddr
->s6_addr32
[0];
134 b
= (__force u32
)saddr
->s6_addr32
[1];
135 c
= (__force u32
)saddr
->s6_addr32
[2];
137 a
+= JHASH_GOLDEN_RATIO
;
138 b
+= JHASH_GOLDEN_RATIO
;
139 c
+= ip6_frag_hash_rnd
;
140 __jhash_mix(a
, b
, c
);
142 a
+= (__force u32
)saddr
->s6_addr32
[3];
143 b
+= (__force u32
)daddr
->s6_addr32
[0];
144 c
+= (__force u32
)daddr
->s6_addr32
[1];
145 __jhash_mix(a
, b
, c
);
147 a
+= (__force u32
)daddr
->s6_addr32
[2];
148 b
+= (__force u32
)daddr
->s6_addr32
[3];
149 c
+= (__force u32
)id
;
150 __jhash_mix(a
, b
, c
);
152 return c
& (IP6Q_HASHSZ
- 1);
155 static struct timer_list ip6_frag_secret_timer
;
156 int sysctl_ip6frag_secret_interval __read_mostly
= 10 * 60 * HZ
;
158 static void ip6_frag_secret_rebuild(unsigned long dummy
)
160 unsigned long now
= jiffies
;
163 write_lock(&ip6_frag_lock
);
164 get_random_bytes(&ip6_frag_hash_rnd
, sizeof(u32
));
165 for (i
= 0; i
< IP6Q_HASHSZ
; i
++) {
166 struct frag_queue
*q
;
167 struct hlist_node
*p
, *n
;
169 hlist_for_each_entry_safe(q
, p
, n
, &ip6_frag_hash
[i
], list
) {
170 unsigned int hval
= ip6qhashfn(q
->id
,
177 /* Relink to new hash chain. */
178 hlist_add_head(&q
->list
,
179 &ip6_frag_hash
[hval
]);
184 write_unlock(&ip6_frag_lock
);
186 mod_timer(&ip6_frag_secret_timer
, now
+ sysctl_ip6frag_secret_interval
);
189 atomic_t ip6_frag_mem
= ATOMIC_INIT(0);
191 /* Memory Tracking Functions. */
192 static inline void frag_kfree_skb(struct sk_buff
*skb
, int *work
)
195 *work
-= skb
->truesize
;
196 atomic_sub(skb
->truesize
, &ip6_frag_mem
);
200 static inline void frag_free_queue(struct frag_queue
*fq
, int *work
)
203 *work
-= sizeof(struct frag_queue
);
204 atomic_sub(sizeof(struct frag_queue
), &ip6_frag_mem
);
208 static inline struct frag_queue
*frag_alloc_queue(void)
210 struct frag_queue
*fq
= kzalloc(sizeof(struct frag_queue
), GFP_ATOMIC
);
214 atomic_add(sizeof(struct frag_queue
), &ip6_frag_mem
);
218 /* Destruction primitives. */
220 /* Complete destruction of fq. */
221 static void ip6_frag_destroy(struct frag_queue
*fq
, int *work
)
225 BUG_TRAP(fq
->last_in
&COMPLETE
);
226 BUG_TRAP(del_timer(&fq
->timer
) == 0);
228 /* Release all fragment data. */
231 struct sk_buff
*xp
= fp
->next
;
233 frag_kfree_skb(fp
, work
);
237 frag_free_queue(fq
, work
);
240 static __inline__
void fq_put(struct frag_queue
*fq
, int *work
)
242 if (atomic_dec_and_test(&fq
->refcnt
))
243 ip6_frag_destroy(fq
, work
);
246 /* Kill fq entry. It is not destroyed immediately,
247 * because caller (and someone more) holds reference count.
249 static __inline__
void fq_kill(struct frag_queue
*fq
)
251 if (del_timer(&fq
->timer
))
252 atomic_dec(&fq
->refcnt
);
254 if (!(fq
->last_in
& COMPLETE
)) {
256 atomic_dec(&fq
->refcnt
);
257 fq
->last_in
|= COMPLETE
;
261 static void ip6_evictor(struct inet6_dev
*idev
)
263 struct frag_queue
*fq
;
264 struct list_head
*tmp
;
267 work
= atomic_read(&ip6_frag_mem
) - sysctl_ip6frag_low_thresh
;
272 read_lock(&ip6_frag_lock
);
273 if (list_empty(&ip6_frag_lru_list
)) {
274 read_unlock(&ip6_frag_lock
);
277 tmp
= ip6_frag_lru_list
.next
;
278 fq
= list_entry(tmp
, struct frag_queue
, lru_list
);
279 atomic_inc(&fq
->refcnt
);
280 read_unlock(&ip6_frag_lock
);
282 spin_lock(&fq
->lock
);
283 if (!(fq
->last_in
&COMPLETE
))
285 spin_unlock(&fq
->lock
);
288 IP6_INC_STATS_BH(idev
, IPSTATS_MIB_REASMFAILS
);
292 static void ip6_frag_expire(unsigned long data
)
294 struct frag_queue
*fq
= (struct frag_queue
*) data
;
295 struct net_device
*dev
= NULL
;
297 spin_lock(&fq
->lock
);
299 if (fq
->last_in
& COMPLETE
)
304 dev
= dev_get_by_index(fq
->iif
);
309 IP6_INC_STATS_BH(__in6_dev_get(dev
), IPSTATS_MIB_REASMTIMEOUT
);
310 IP6_INC_STATS_BH(__in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
313 /* Don't send error if the first segment did not arrive. */
314 if (!(fq
->last_in
&FIRST_IN
) || !fq
->fragments
)
318 But use as source device on which LAST ARRIVED
319 segment was received. And do not use fq->dev
320 pointer directly, device might already disappeared.
322 fq
->fragments
->dev
= dev
;
323 icmpv6_send(fq
->fragments
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_FRAGTIME
, 0, dev
);
327 spin_unlock(&fq
->lock
);
331 /* Creation primitives. */
334 static struct frag_queue
*ip6_frag_intern(struct frag_queue
*fq_in
)
336 struct frag_queue
*fq
;
339 struct hlist_node
*n
;
342 write_lock(&ip6_frag_lock
);
343 hash
= ip6qhashfn(fq_in
->id
, &fq_in
->saddr
, &fq_in
->daddr
);
345 hlist_for_each_entry(fq
, n
, &ip6_frag_hash
[hash
], list
) {
346 if (fq
->id
== fq_in
->id
&&
347 ipv6_addr_equal(&fq_in
->saddr
, &fq
->saddr
) &&
348 ipv6_addr_equal(&fq_in
->daddr
, &fq
->daddr
)) {
349 atomic_inc(&fq
->refcnt
);
350 write_unlock(&ip6_frag_lock
);
351 fq_in
->last_in
|= COMPLETE
;
359 if (!mod_timer(&fq
->timer
, jiffies
+ sysctl_ip6frag_time
))
360 atomic_inc(&fq
->refcnt
);
362 atomic_inc(&fq
->refcnt
);
363 hlist_add_head(&fq
->list
, &ip6_frag_hash
[hash
]);
364 INIT_LIST_HEAD(&fq
->lru_list
);
365 list_add_tail(&fq
->lru_list
, &ip6_frag_lru_list
);
367 write_unlock(&ip6_frag_lock
);
372 static struct frag_queue
*
373 ip6_frag_create(__be32 id
, struct in6_addr
*src
, struct in6_addr
*dst
,
374 struct inet6_dev
*idev
)
376 struct frag_queue
*fq
;
378 if ((fq
= frag_alloc_queue()) == NULL
)
382 ipv6_addr_copy(&fq
->saddr
, src
);
383 ipv6_addr_copy(&fq
->daddr
, dst
);
385 init_timer(&fq
->timer
);
386 fq
->timer
.function
= ip6_frag_expire
;
387 fq
->timer
.data
= (long) fq
;
388 spin_lock_init(&fq
->lock
);
389 atomic_set(&fq
->refcnt
, 1);
391 return ip6_frag_intern(fq
);
394 IP6_INC_STATS_BH(idev
, IPSTATS_MIB_REASMFAILS
);
398 static __inline__
struct frag_queue
*
399 fq_find(__be32 id
, struct in6_addr
*src
, struct in6_addr
*dst
,
400 struct inet6_dev
*idev
)
402 struct frag_queue
*fq
;
403 struct hlist_node
*n
;
406 read_lock(&ip6_frag_lock
);
407 hash
= ip6qhashfn(id
, src
, dst
);
408 hlist_for_each_entry(fq
, n
, &ip6_frag_hash
[hash
], list
) {
410 ipv6_addr_equal(src
, &fq
->saddr
) &&
411 ipv6_addr_equal(dst
, &fq
->daddr
)) {
412 atomic_inc(&fq
->refcnt
);
413 read_unlock(&ip6_frag_lock
);
417 read_unlock(&ip6_frag_lock
);
419 return ip6_frag_create(id
, src
, dst
, idev
);
423 static void ip6_frag_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
424 struct frag_hdr
*fhdr
, int nhoff
)
426 struct sk_buff
*prev
, *next
;
429 if (fq
->last_in
& COMPLETE
)
432 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
433 end
= offset
+ (ntohs(skb
->nh
.ipv6h
->payload_len
) -
434 ((u8
*) (fhdr
+ 1) - (u8
*) (skb
->nh
.ipv6h
+ 1)));
436 if ((unsigned int)end
> IPV6_MAXPLEN
) {
437 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
),
438 IPSTATS_MIB_INHDRERRORS
);
439 icmpv6_param_prob(skb
,ICMPV6_HDR_FIELD
, (u8
*)&fhdr
->frag_off
- skb
->nh
.raw
);
443 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
444 skb
->csum
= csum_sub(skb
->csum
,
445 csum_partial(skb
->nh
.raw
, (u8
*)(fhdr
+1)-skb
->nh
.raw
, 0));
447 /* Is this the final fragment? */
448 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
449 /* If we already have some bits beyond end
450 * or have different end, the segment is corrupted.
453 ((fq
->last_in
& LAST_IN
) && end
!= fq
->len
))
455 fq
->last_in
|= LAST_IN
;
458 /* Check if the fragment is rounded to 8 bytes.
459 * Required by the RFC.
462 /* RFC2460 says always send parameter problem in
465 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
),
466 IPSTATS_MIB_INHDRERRORS
);
467 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
468 offsetof(struct ipv6hdr
, payload_len
));
472 /* Some bits beyond end -> corruption. */
473 if (fq
->last_in
& LAST_IN
)
482 /* Point into the IP datagram 'data' part. */
483 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
))
486 if (pskb_trim_rcsum(skb
, end
- offset
))
489 /* Find out which fragments are in front and at the back of us
490 * in the chain of fragments so far. We must know where to put
491 * this fragment, right?
494 for(next
= fq
->fragments
; next
!= NULL
; next
= next
->next
) {
495 if (FRAG6_CB(next
)->offset
>= offset
)
500 /* We found where to put this one. Check for overlap with
501 * preceding fragment, and, if needed, align things so that
502 * any overlaps are eliminated.
505 int i
= (FRAG6_CB(prev
)->offset
+ prev
->len
) - offset
;
511 if (!pskb_pull(skb
, i
))
513 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
)
514 skb
->ip_summed
= CHECKSUM_NONE
;
518 /* Look for overlap with succeeding segments.
519 * If we can merge fragments, do it.
521 while (next
&& FRAG6_CB(next
)->offset
< end
) {
522 int i
= end
- FRAG6_CB(next
)->offset
; /* overlap is 'i' bytes */
525 /* Eat head of the next overlapped fragment
526 * and leave the loop. The next ones cannot overlap.
528 if (!pskb_pull(next
, i
))
530 FRAG6_CB(next
)->offset
+= i
; /* next fragment */
532 if (next
->ip_summed
!= CHECKSUM_UNNECESSARY
)
533 next
->ip_summed
= CHECKSUM_NONE
;
536 struct sk_buff
*free_it
= next
;
538 /* Old fragment is completely overridden with
546 fq
->fragments
= next
;
548 fq
->meat
-= free_it
->len
;
549 frag_kfree_skb(free_it
, NULL
);
553 FRAG6_CB(skb
)->offset
= offset
;
555 /* Insert this fragment in the chain of fragments. */
563 fq
->iif
= skb
->dev
->ifindex
;
565 skb_get_timestamp(skb
, &fq
->stamp
);
566 fq
->meat
+= skb
->len
;
567 atomic_add(skb
->truesize
, &ip6_frag_mem
);
569 /* The first fragment.
570 * nhoffset is obtained from the first fragment, of course.
573 fq
->nhoffset
= nhoff
;
574 fq
->last_in
|= FIRST_IN
;
576 write_lock(&ip6_frag_lock
);
577 list_move_tail(&fq
->lru_list
, &ip6_frag_lru_list
);
578 write_unlock(&ip6_frag_lock
);
582 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_REASMFAILS
);
587 * Check if this packet is complete.
588 * Returns NULL on failure by any reason, and pointer
589 * to current nexthdr field in reassembled frame.
591 * It is called with locked fq, and caller must check that
592 * queue is eligible for reassembly i.e. it is not COMPLETE,
593 * the last and the first frames arrived and all the bits are here.
595 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
**skb_in
,
596 struct net_device
*dev
)
598 struct sk_buff
*fp
, *head
= fq
->fragments
;
604 BUG_TRAP(head
!= NULL
);
605 BUG_TRAP(FRAG6_CB(head
)->offset
== 0);
607 /* Unfragmented part is taken from the first segment. */
608 payload_len
= (head
->data
- head
->nh
.raw
) - sizeof(struct ipv6hdr
) + fq
->len
- sizeof(struct frag_hdr
);
609 if (payload_len
> IPV6_MAXPLEN
)
612 /* Head of list must not be cloned. */
613 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
))
616 /* If the first fragment is fragmented itself, we split
617 * it to two chunks: the first with data and paged part
618 * and the second, holding only fragments. */
619 if (skb_shinfo(head
)->frag_list
) {
620 struct sk_buff
*clone
;
623 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
)
625 clone
->next
= head
->next
;
627 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
628 skb_shinfo(head
)->frag_list
= NULL
;
629 for (i
=0; i
<skb_shinfo(head
)->nr_frags
; i
++)
630 plen
+= skb_shinfo(head
)->frags
[i
].size
;
631 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
632 head
->data_len
-= clone
->len
;
633 head
->len
-= clone
->len
;
635 clone
->ip_summed
= head
->ip_summed
;
636 atomic_add(clone
->truesize
, &ip6_frag_mem
);
639 /* We have to remove fragment header from datagram and to relocate
640 * header in order to calculate ICV correctly. */
641 nhoff
= fq
->nhoffset
;
642 head
->nh
.raw
[nhoff
] = head
->h
.raw
[0];
643 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
644 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
645 head
->mac
.raw
+= sizeof(struct frag_hdr
);
646 head
->nh
.raw
+= sizeof(struct frag_hdr
);
648 skb_shinfo(head
)->frag_list
= head
->next
;
649 head
->h
.raw
= head
->data
;
650 skb_push(head
, head
->data
- head
->nh
.raw
);
651 atomic_sub(head
->truesize
, &ip6_frag_mem
);
653 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
654 head
->data_len
+= fp
->len
;
655 head
->len
+= fp
->len
;
656 if (head
->ip_summed
!= fp
->ip_summed
)
657 head
->ip_summed
= CHECKSUM_NONE
;
658 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
659 head
->csum
= csum_add(head
->csum
, fp
->csum
);
660 head
->truesize
+= fp
->truesize
;
661 atomic_sub(fp
->truesize
, &ip6_frag_mem
);
666 skb_set_timestamp(head
, &fq
->stamp
);
667 head
->nh
.ipv6h
->payload_len
= htons(payload_len
);
668 IP6CB(head
)->nhoff
= nhoff
;
672 /* Yes, and fold redundant checksum back. 8) */
673 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
674 head
->csum
= csum_partial(head
->nh
.raw
, head
->h
.raw
-head
->nh
.raw
, head
->csum
);
677 IP6_INC_STATS_BH(__in6_dev_get(dev
), IPSTATS_MIB_REASMOKS
);
679 fq
->fragments
= NULL
;
684 printk(KERN_DEBUG
"ip6_frag_reasm: payload len = %d\n", payload_len
);
688 printk(KERN_DEBUG
"ip6_frag_reasm: no memory for reassembly\n");
691 IP6_INC_STATS_BH(__in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
696 static int ipv6_frag_rcv(struct sk_buff
**skbp
)
698 struct sk_buff
*skb
= *skbp
;
699 struct net_device
*dev
= skb
->dev
;
700 struct frag_hdr
*fhdr
;
701 struct frag_queue
*fq
;
706 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_REASMREQDS
);
708 /* Jumbo payload inhibits frag. header */
709 if (hdr
->payload_len
==0) {
710 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_INHDRERRORS
);
711 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb
->h
.raw
-skb
->nh
.raw
);
714 if (!pskb_may_pull(skb
, (skb
->h
.raw
-skb
->data
)+sizeof(struct frag_hdr
))) {
715 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_INHDRERRORS
);
716 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
, skb
->h
.raw
-skb
->nh
.raw
);
721 fhdr
= (struct frag_hdr
*)skb
->h
.raw
;
723 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
724 /* It is not a fragmented frame */
725 skb
->h
.raw
+= sizeof(struct frag_hdr
);
726 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_REASMOKS
);
728 IP6CB(skb
)->nhoff
= (u8
*)fhdr
- skb
->nh
.raw
;
732 if (atomic_read(&ip6_frag_mem
) > sysctl_ip6frag_high_thresh
)
733 ip6_evictor(ip6_dst_idev(skb
->dst
));
735 if ((fq
= fq_find(fhdr
->identification
, &hdr
->saddr
, &hdr
->daddr
,
736 ip6_dst_idev(skb
->dst
))) != NULL
) {
739 spin_lock(&fq
->lock
);
741 ip6_frag_queue(fq
, skb
, fhdr
, IP6CB(skb
)->nhoff
);
743 if (fq
->last_in
== (FIRST_IN
|LAST_IN
) &&
745 ret
= ip6_frag_reasm(fq
, skbp
, dev
);
747 spin_unlock(&fq
->lock
);
752 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_REASMFAILS
);
757 static struct inet6_protocol frag_protocol
=
759 .handler
= ipv6_frag_rcv
,
760 .flags
= INET6_PROTO_NOPOLICY
,
763 void __init
ipv6_frag_init(void)
765 if (inet6_add_protocol(&frag_protocol
, IPPROTO_FRAGMENT
) < 0)
766 printk(KERN_ERR
"ipv6_frag_init: Could not register protocol\n");
768 ip6_frag_hash_rnd
= (u32
) ((num_physpages
^ (num_physpages
>>7)) ^
769 (jiffies
^ (jiffies
>> 6)));
771 init_timer(&ip6_frag_secret_timer
);
772 ip6_frag_secret_timer
.function
= ip6_frag_secret_rebuild
;
773 ip6_frag_secret_timer
.expires
= jiffies
+ sysctl_ip6frag_secret_interval
;
774 add_timer(&ip6_frag_secret_timer
);