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>
45 #include <linux/skbuff.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 */
77 struct in6_addr saddr
;
78 struct in6_addr daddr
;
85 struct inet_frags_ctl ip6_frags_ctl __read_mostly
= {
86 .high_thresh
= 256 * 1024,
87 .low_thresh
= 192 * 1024,
88 .timeout
= IPV6_FRAG_TIMEOUT
,
89 .secret_interval
= 10 * 60 * HZ
,
92 static struct inet_frags ip6_frags
;
94 int ip6_frag_nqueues(void)
96 return ip6_frags
.nqueues
;
99 int ip6_frag_mem(void)
101 return atomic_read(&ip6_frags
.mem
);
104 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
105 struct net_device
*dev
);
108 * callers should be careful not to use the hash value outside the ipfrag_lock
109 * as doing so could race with ipfrag_hash_rnd being recalculated.
111 static unsigned int ip6qhashfn(__be32 id
, struct in6_addr
*saddr
,
112 struct in6_addr
*daddr
)
116 a
= (__force u32
)saddr
->s6_addr32
[0];
117 b
= (__force u32
)saddr
->s6_addr32
[1];
118 c
= (__force u32
)saddr
->s6_addr32
[2];
120 a
+= JHASH_GOLDEN_RATIO
;
121 b
+= JHASH_GOLDEN_RATIO
;
123 __jhash_mix(a
, b
, c
);
125 a
+= (__force u32
)saddr
->s6_addr32
[3];
126 b
+= (__force u32
)daddr
->s6_addr32
[0];
127 c
+= (__force u32
)daddr
->s6_addr32
[1];
128 __jhash_mix(a
, b
, c
);
130 a
+= (__force u32
)daddr
->s6_addr32
[2];
131 b
+= (__force u32
)daddr
->s6_addr32
[3];
132 c
+= (__force u32
)id
;
133 __jhash_mix(a
, b
, c
);
135 return c
& (INETFRAGS_HASHSZ
- 1);
138 static unsigned int ip6_hashfn(struct inet_frag_queue
*q
)
140 struct frag_queue
*fq
;
142 fq
= container_of(q
, struct frag_queue
, q
);
143 return ip6qhashfn(fq
->id
, &fq
->saddr
, &fq
->daddr
);
146 int ip6_frag_equal(struct inet_frag_queue
*q1
, struct inet_frag_queue
*q2
)
148 struct frag_queue
*fq1
, *fq2
;
150 fq1
= container_of(q1
, struct frag_queue
, q
);
151 fq2
= container_of(q2
, struct frag_queue
, q
);
152 return (fq1
->id
== fq2
->id
&&
153 ipv6_addr_equal(&fq2
->saddr
, &fq1
->saddr
) &&
154 ipv6_addr_equal(&fq2
->daddr
, &fq1
->daddr
));
156 EXPORT_SYMBOL(ip6_frag_equal
);
158 /* Memory Tracking Functions. */
159 static inline void frag_kfree_skb(struct sk_buff
*skb
, int *work
)
162 *work
-= skb
->truesize
;
163 atomic_sub(skb
->truesize
, &ip6_frags
.mem
);
167 static void ip6_frag_free(struct inet_frag_queue
*fq
)
169 kfree(container_of(fq
, struct frag_queue
, q
));
172 static inline struct frag_queue
*frag_alloc_queue(void)
174 struct inet_frag_queue
*q
;
176 q
= inet_frag_alloc(&ip6_frags
);
177 return q
? container_of(q
, struct frag_queue
, q
) : NULL
;
180 /* Destruction primitives. */
182 static __inline__
void fq_put(struct frag_queue
*fq
)
184 inet_frag_put(&fq
->q
, &ip6_frags
);
187 /* Kill fq entry. It is not destroyed immediately,
188 * because caller (and someone more) holds reference count.
190 static __inline__
void fq_kill(struct frag_queue
*fq
)
192 inet_frag_kill(&fq
->q
, &ip6_frags
);
195 static void ip6_evictor(struct inet6_dev
*idev
)
199 evicted
= inet_frag_evictor(&ip6_frags
);
201 IP6_ADD_STATS_BH(idev
, IPSTATS_MIB_REASMFAILS
, evicted
);
204 static void ip6_frag_expire(unsigned long data
)
206 struct frag_queue
*fq
;
207 struct net_device
*dev
= NULL
;
209 fq
= container_of((struct inet_frag_queue
*)data
, struct frag_queue
, q
);
211 spin_lock(&fq
->q
.lock
);
213 if (fq
->q
.last_in
& COMPLETE
)
218 dev
= dev_get_by_index(&init_net
, fq
->iif
);
223 IP6_INC_STATS_BH(__in6_dev_get(dev
), IPSTATS_MIB_REASMTIMEOUT
);
224 IP6_INC_STATS_BH(__in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
227 /* Don't send error if the first segment did not arrive. */
228 if (!(fq
->q
.last_in
&FIRST_IN
) || !fq
->q
.fragments
)
232 But use as source device on which LAST ARRIVED
233 segment was received. And do not use fq->dev
234 pointer directly, device might already disappeared.
236 fq
->q
.fragments
->dev
= dev
;
237 icmpv6_send(fq
->q
.fragments
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_FRAGTIME
, 0, dev
);
241 spin_unlock(&fq
->q
.lock
);
245 /* Creation primitives. */
248 static struct frag_queue
*ip6_frag_intern(struct frag_queue
*fq_in
,
251 struct inet_frag_queue
*q
;
253 q
= inet_frag_intern(&fq_in
->q
, &ip6_frags
, hash
);
254 return container_of(q
, struct frag_queue
, q
);
258 static struct frag_queue
*
259 ip6_frag_create(__be32 id
, struct in6_addr
*src
, struct in6_addr
*dst
,
260 struct inet6_dev
*idev
, unsigned int hash
)
262 struct frag_queue
*fq
;
264 if ((fq
= frag_alloc_queue()) == NULL
)
268 ipv6_addr_copy(&fq
->saddr
, src
);
269 ipv6_addr_copy(&fq
->daddr
, dst
);
271 return ip6_frag_intern(fq
, hash
);
274 IP6_INC_STATS_BH(idev
, IPSTATS_MIB_REASMFAILS
);
278 static __inline__
struct frag_queue
*
279 fq_find(__be32 id
, struct in6_addr
*src
, struct in6_addr
*dst
,
280 struct inet6_dev
*idev
)
282 struct frag_queue
*fq
;
283 struct hlist_node
*n
;
286 read_lock(&ip6_frags
.lock
);
287 hash
= ip6qhashfn(id
, src
, dst
);
288 hlist_for_each_entry(fq
, n
, &ip6_frags
.hash
[hash
], q
.list
) {
290 ipv6_addr_equal(src
, &fq
->saddr
) &&
291 ipv6_addr_equal(dst
, &fq
->daddr
)) {
292 atomic_inc(&fq
->q
.refcnt
);
293 read_unlock(&ip6_frags
.lock
);
297 read_unlock(&ip6_frags
.lock
);
299 return ip6_frag_create(id
, src
, dst
, idev
, hash
);
303 static int ip6_frag_queue(struct frag_queue
*fq
, struct sk_buff
*skb
,
304 struct frag_hdr
*fhdr
, int nhoff
)
306 struct sk_buff
*prev
, *next
;
307 struct net_device
*dev
;
310 if (fq
->q
.last_in
& COMPLETE
)
313 offset
= ntohs(fhdr
->frag_off
) & ~0x7;
314 end
= offset
+ (ntohs(ipv6_hdr(skb
)->payload_len
) -
315 ((u8
*)(fhdr
+ 1) - (u8
*)(ipv6_hdr(skb
) + 1)));
317 if ((unsigned int)end
> IPV6_MAXPLEN
) {
318 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
),
319 IPSTATS_MIB_INHDRERRORS
);
320 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
321 ((u8
*)&fhdr
->frag_off
-
322 skb_network_header(skb
)));
326 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
327 const unsigned char *nh
= skb_network_header(skb
);
328 skb
->csum
= csum_sub(skb
->csum
,
329 csum_partial(nh
, (u8
*)(fhdr
+ 1) - nh
,
333 /* Is this the final fragment? */
334 if (!(fhdr
->frag_off
& htons(IP6_MF
))) {
335 /* If we already have some bits beyond end
336 * or have different end, the segment is corrupted.
338 if (end
< fq
->q
.len
||
339 ((fq
->q
.last_in
& LAST_IN
) && end
!= fq
->q
.len
))
341 fq
->q
.last_in
|= LAST_IN
;
344 /* Check if the fragment is rounded to 8 bytes.
345 * Required by the RFC.
348 /* RFC2460 says always send parameter problem in
351 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
),
352 IPSTATS_MIB_INHDRERRORS
);
353 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
354 offsetof(struct ipv6hdr
, payload_len
));
357 if (end
> fq
->q
.len
) {
358 /* Some bits beyond end -> corruption. */
359 if (fq
->q
.last_in
& LAST_IN
)
368 /* Point into the IP datagram 'data' part. */
369 if (!pskb_pull(skb
, (u8
*) (fhdr
+ 1) - skb
->data
))
372 if (pskb_trim_rcsum(skb
, end
- offset
))
375 /* Find out which fragments are in front and at the back of us
376 * in the chain of fragments so far. We must know where to put
377 * this fragment, right?
380 for(next
= fq
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
381 if (FRAG6_CB(next
)->offset
>= offset
)
386 /* We found where to put this one. Check for overlap with
387 * preceding fragment, and, if needed, align things so that
388 * any overlaps are eliminated.
391 int i
= (FRAG6_CB(prev
)->offset
+ prev
->len
) - offset
;
397 if (!pskb_pull(skb
, i
))
399 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
)
400 skb
->ip_summed
= CHECKSUM_NONE
;
404 /* Look for overlap with succeeding segments.
405 * If we can merge fragments, do it.
407 while (next
&& FRAG6_CB(next
)->offset
< end
) {
408 int i
= end
- FRAG6_CB(next
)->offset
; /* overlap is 'i' bytes */
411 /* Eat head of the next overlapped fragment
412 * and leave the loop. The next ones cannot overlap.
414 if (!pskb_pull(next
, i
))
416 FRAG6_CB(next
)->offset
+= i
; /* next fragment */
418 if (next
->ip_summed
!= CHECKSUM_UNNECESSARY
)
419 next
->ip_summed
= CHECKSUM_NONE
;
422 struct sk_buff
*free_it
= next
;
424 /* Old fragment is completely overridden with
432 fq
->q
.fragments
= next
;
434 fq
->q
.meat
-= free_it
->len
;
435 frag_kfree_skb(free_it
, NULL
);
439 FRAG6_CB(skb
)->offset
= offset
;
441 /* Insert this fragment in the chain of fragments. */
446 fq
->q
.fragments
= skb
;
450 fq
->iif
= dev
->ifindex
;
453 fq
->q
.stamp
= skb
->tstamp
;
454 fq
->q
.meat
+= skb
->len
;
455 atomic_add(skb
->truesize
, &ip6_frags
.mem
);
457 /* The first fragment.
458 * nhoffset is obtained from the first fragment, of course.
461 fq
->nhoffset
= nhoff
;
462 fq
->q
.last_in
|= FIRST_IN
;
465 if (fq
->q
.last_in
== (FIRST_IN
| LAST_IN
) && fq
->q
.meat
== fq
->q
.len
)
466 return ip6_frag_reasm(fq
, prev
, dev
);
468 write_lock(&ip6_frags
.lock
);
469 list_move_tail(&fq
->q
.lru_list
, &ip6_frags
.lru_list
);
470 write_unlock(&ip6_frags
.lock
);
474 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_REASMFAILS
);
480 * Check if this packet is complete.
481 * Returns NULL on failure by any reason, and pointer
482 * to current nexthdr field in reassembled frame.
484 * It is called with locked fq, and caller must check that
485 * queue is eligible for reassembly i.e. it is not COMPLETE,
486 * the last and the first frames arrived and all the bits are here.
488 static int ip6_frag_reasm(struct frag_queue
*fq
, struct sk_buff
*prev
,
489 struct net_device
*dev
)
491 struct sk_buff
*fp
, *head
= fq
->q
.fragments
;
497 /* Make the one we just received the head. */
500 fp
= skb_clone(head
, GFP_ATOMIC
);
505 fp
->next
= head
->next
;
508 skb_morph(head
, fq
->q
.fragments
);
509 head
->next
= fq
->q
.fragments
->next
;
511 kfree_skb(fq
->q
.fragments
);
512 fq
->q
.fragments
= head
;
515 BUG_TRAP(head
!= NULL
);
516 BUG_TRAP(FRAG6_CB(head
)->offset
== 0);
518 /* Unfragmented part is taken from the first segment. */
519 payload_len
= ((head
->data
- skb_network_header(head
)) -
520 sizeof(struct ipv6hdr
) + fq
->q
.len
-
521 sizeof(struct frag_hdr
));
522 if (payload_len
> IPV6_MAXPLEN
)
525 /* Head of list must not be cloned. */
526 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
))
529 /* If the first fragment is fragmented itself, we split
530 * it to two chunks: the first with data and paged part
531 * and the second, holding only fragments. */
532 if (skb_shinfo(head
)->frag_list
) {
533 struct sk_buff
*clone
;
536 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
)
538 clone
->next
= head
->next
;
540 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
541 skb_shinfo(head
)->frag_list
= NULL
;
542 for (i
=0; i
<skb_shinfo(head
)->nr_frags
; i
++)
543 plen
+= skb_shinfo(head
)->frags
[i
].size
;
544 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
545 head
->data_len
-= clone
->len
;
546 head
->len
-= clone
->len
;
548 clone
->ip_summed
= head
->ip_summed
;
549 atomic_add(clone
->truesize
, &ip6_frags
.mem
);
552 /* We have to remove fragment header from datagram and to relocate
553 * header in order to calculate ICV correctly. */
554 nhoff
= fq
->nhoffset
;
555 skb_network_header(head
)[nhoff
] = skb_transport_header(head
)[0];
556 memmove(head
->head
+ sizeof(struct frag_hdr
), head
->head
,
557 (head
->data
- head
->head
) - sizeof(struct frag_hdr
));
558 head
->mac_header
+= sizeof(struct frag_hdr
);
559 head
->network_header
+= sizeof(struct frag_hdr
);
561 skb_shinfo(head
)->frag_list
= head
->next
;
562 skb_reset_transport_header(head
);
563 skb_push(head
, head
->data
- skb_network_header(head
));
564 atomic_sub(head
->truesize
, &ip6_frags
.mem
);
566 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
567 head
->data_len
+= fp
->len
;
568 head
->len
+= fp
->len
;
569 if (head
->ip_summed
!= fp
->ip_summed
)
570 head
->ip_summed
= CHECKSUM_NONE
;
571 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
572 head
->csum
= csum_add(head
->csum
, fp
->csum
);
573 head
->truesize
+= fp
->truesize
;
574 atomic_sub(fp
->truesize
, &ip6_frags
.mem
);
579 head
->tstamp
= fq
->q
.stamp
;
580 ipv6_hdr(head
)->payload_len
= htons(payload_len
);
581 IP6CB(head
)->nhoff
= nhoff
;
583 /* Yes, and fold redundant checksum back. 8) */
584 if (head
->ip_summed
== CHECKSUM_COMPLETE
)
585 head
->csum
= csum_partial(skb_network_header(head
),
586 skb_network_header_len(head
),
590 IP6_INC_STATS_BH(__in6_dev_get(dev
), IPSTATS_MIB_REASMOKS
);
592 fq
->q
.fragments
= NULL
;
597 printk(KERN_DEBUG
"ip6_frag_reasm: payload len = %d\n", payload_len
);
601 printk(KERN_DEBUG
"ip6_frag_reasm: no memory for reassembly\n");
604 IP6_INC_STATS_BH(__in6_dev_get(dev
), IPSTATS_MIB_REASMFAILS
);
609 static int ipv6_frag_rcv(struct sk_buff
*skb
)
611 struct frag_hdr
*fhdr
;
612 struct frag_queue
*fq
;
613 struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
615 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_REASMREQDS
);
617 /* Jumbo payload inhibits frag. header */
618 if (hdr
->payload_len
==0) {
619 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_INHDRERRORS
);
620 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
621 skb_network_header_len(skb
));
624 if (!pskb_may_pull(skb
, (skb_transport_offset(skb
) +
625 sizeof(struct frag_hdr
)))) {
626 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_INHDRERRORS
);
627 icmpv6_param_prob(skb
, ICMPV6_HDR_FIELD
,
628 skb_network_header_len(skb
));
633 fhdr
= (struct frag_hdr
*)skb_transport_header(skb
);
635 if (!(fhdr
->frag_off
& htons(0xFFF9))) {
636 /* It is not a fragmented frame */
637 skb
->transport_header
+= sizeof(struct frag_hdr
);
638 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_REASMOKS
);
640 IP6CB(skb
)->nhoff
= (u8
*)fhdr
- skb_network_header(skb
);
644 if (atomic_read(&ip6_frags
.mem
) > ip6_frags_ctl
.high_thresh
)
645 ip6_evictor(ip6_dst_idev(skb
->dst
));
647 if ((fq
= fq_find(fhdr
->identification
, &hdr
->saddr
, &hdr
->daddr
,
648 ip6_dst_idev(skb
->dst
))) != NULL
) {
651 spin_lock(&fq
->q
.lock
);
653 ret
= ip6_frag_queue(fq
, skb
, fhdr
, IP6CB(skb
)->nhoff
);
655 spin_unlock(&fq
->q
.lock
);
660 IP6_INC_STATS_BH(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_REASMFAILS
);
665 static struct inet6_protocol frag_protocol
=
667 .handler
= ipv6_frag_rcv
,
668 .flags
= INET6_PROTO_NOPOLICY
,
671 void __init
ipv6_frag_init(void)
673 if (inet6_add_protocol(&frag_protocol
, IPPROTO_FRAGMENT
) < 0)
674 printk(KERN_ERR
"ipv6_frag_init: Could not register protocol\n");
676 ip6_frags
.ctl
= &ip6_frags_ctl
;
677 ip6_frags
.hashfn
= ip6_hashfn
;
678 ip6_frags
.destructor
= ip6_frag_free
;
679 ip6_frags
.skb_free
= NULL
;
680 ip6_frags
.qsize
= sizeof(struct frag_queue
);
681 ip6_frags
.equal
= ip6_frag_equal
;
682 ip6_frags
.frag_expire
= ip6_frag_expire
;
683 inet_frags_init(&ip6_frags
);