[INET]: Consolidate xxx_the secret_rebuild
[safe/jmp/linux-2.6] / net / ipv6 / reassembly.c
1 /*
2  *      IPv6 fragment reassembly
3  *      Linux INET6 implementation
4  *
5  *      Authors:
6  *      Pedro Roque             <roque@di.fc.ul.pt>
7  *
8  *      $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
9  *
10  *      Based on: net/ipv4/ip_fragment.c
11  *
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.
16  */
17
18 /*
19  *      Fixes:
20  *      Andi Kleen      Make it work with multiple hosts.
21  *                      More RFC compliance.
22  *
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{}.
27  *      David Stevens and
28  *      YOSHIFUJI,H. @USAGI     Always remove fragment header to
29  *                              calculate ICV correctly.
30  */
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>
46
47 #include <net/sock.h>
48 #include <net/snmp.h>
49
50 #include <net/ipv6.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>
58
59 struct ip6frag_skb_cb
60 {
61         struct inet6_skb_parm   h;
62         int                     offset;
63 };
64
65 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
66
67
68 /*
69  *      Equivalent of ipv4 struct ipq
70  */
71
72 struct frag_queue
73 {
74         struct inet_frag_queue  q;
75
76         __be32                  id;             /* fragment id          */
77         struct in6_addr         saddr;
78         struct in6_addr         daddr;
79
80         int                     iif;
81         unsigned int            csum;
82         __u16                   nhoffset;
83 };
84
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,
90 };
91
92 static struct inet_frags ip6_frags;
93
94 int ip6_frag_nqueues(void)
95 {
96         return ip6_frags.nqueues;
97 }
98
99 int ip6_frag_mem(void)
100 {
101         return atomic_read(&ip6_frags.mem);
102 }
103
104 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
105                           struct net_device *dev);
106
107 /*
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.
110  */
111 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
112                                struct in6_addr *daddr)
113 {
114         u32 a, b, c;
115
116         a = (__force u32)saddr->s6_addr32[0];
117         b = (__force u32)saddr->s6_addr32[1];
118         c = (__force u32)saddr->s6_addr32[2];
119
120         a += JHASH_GOLDEN_RATIO;
121         b += JHASH_GOLDEN_RATIO;
122         c += ip6_frags.rnd;
123         __jhash_mix(a, b, c);
124
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);
129
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);
134
135         return c & (INETFRAGS_HASHSZ - 1);
136 }
137
138 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
139 {
140         struct frag_queue *fq;
141
142         fq = container_of(q, struct frag_queue, q);
143         return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
144 }
145
146 /* Memory Tracking Functions. */
147 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
148 {
149         if (work)
150                 *work -= skb->truesize;
151         atomic_sub(skb->truesize, &ip6_frags.mem);
152         kfree_skb(skb);
153 }
154
155 static inline void frag_free_queue(struct frag_queue *fq, int *work)
156 {
157         if (work)
158                 *work -= sizeof(struct frag_queue);
159         atomic_sub(sizeof(struct frag_queue), &ip6_frags.mem);
160         kfree(fq);
161 }
162
163 static inline struct frag_queue *frag_alloc_queue(void)
164 {
165         struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
166
167         if(!fq)
168                 return NULL;
169         atomic_add(sizeof(struct frag_queue), &ip6_frags.mem);
170         return fq;
171 }
172
173 /* Destruction primitives. */
174
175 /* Complete destruction of fq. */
176 static void ip6_frag_destroy(struct frag_queue *fq, int *work)
177 {
178         struct sk_buff *fp;
179
180         BUG_TRAP(fq->q.last_in&COMPLETE);
181         BUG_TRAP(del_timer(&fq->q.timer) == 0);
182
183         /* Release all fragment data. */
184         fp = fq->q.fragments;
185         while (fp) {
186                 struct sk_buff *xp = fp->next;
187
188                 frag_kfree_skb(fp, work);
189                 fp = xp;
190         }
191
192         frag_free_queue(fq, work);
193 }
194
195 static __inline__ void fq_put(struct frag_queue *fq, int *work)
196 {
197         if (atomic_dec_and_test(&fq->q.refcnt))
198                 ip6_frag_destroy(fq, work);
199 }
200
201 /* Kill fq entry. It is not destroyed immediately,
202  * because caller (and someone more) holds reference count.
203  */
204 static __inline__ void fq_kill(struct frag_queue *fq)
205 {
206         inet_frag_kill(&fq->q, &ip6_frags);
207 }
208
209 static void ip6_evictor(struct inet6_dev *idev)
210 {
211         struct frag_queue *fq;
212         struct list_head *tmp;
213         int work;
214
215         work = atomic_read(&ip6_frags.mem) - ip6_frags_ctl.low_thresh;
216         if (work <= 0)
217                 return;
218
219         while(work > 0) {
220                 read_lock(&ip6_frags.lock);
221                 if (list_empty(&ip6_frags.lru_list)) {
222                         read_unlock(&ip6_frags.lock);
223                         return;
224                 }
225                 tmp = ip6_frags.lru_list.next;
226                 fq = list_entry(tmp, struct frag_queue, q.lru_list);
227                 atomic_inc(&fq->q.refcnt);
228                 read_unlock(&ip6_frags.lock);
229
230                 spin_lock(&fq->q.lock);
231                 if (!(fq->q.last_in&COMPLETE))
232                         fq_kill(fq);
233                 spin_unlock(&fq->q.lock);
234
235                 fq_put(fq, &work);
236                 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
237         }
238 }
239
240 static void ip6_frag_expire(unsigned long data)
241 {
242         struct frag_queue *fq = (struct frag_queue *) data;
243         struct net_device *dev = NULL;
244
245         spin_lock(&fq->q.lock);
246
247         if (fq->q.last_in & COMPLETE)
248                 goto out;
249
250         fq_kill(fq);
251
252         dev = dev_get_by_index(&init_net, fq->iif);
253         if (!dev)
254                 goto out;
255
256         rcu_read_lock();
257         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
258         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
259         rcu_read_unlock();
260
261         /* Don't send error if the first segment did not arrive. */
262         if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
263                 goto out;
264
265         /*
266            But use as source device on which LAST ARRIVED
267            segment was received. And do not use fq->dev
268            pointer directly, device might already disappeared.
269          */
270         fq->q.fragments->dev = dev;
271         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
272 out:
273         if (dev)
274                 dev_put(dev);
275         spin_unlock(&fq->q.lock);
276         fq_put(fq, NULL);
277 }
278
279 /* Creation primitives. */
280
281
282 static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in)
283 {
284         struct frag_queue *fq;
285         unsigned int hash;
286 #ifdef CONFIG_SMP
287         struct hlist_node *n;
288 #endif
289
290         write_lock(&ip6_frags.lock);
291         hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr);
292 #ifdef CONFIG_SMP
293         hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
294                 if (fq->id == fq_in->id &&
295                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
296                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
297                         atomic_inc(&fq->q.refcnt);
298                         write_unlock(&ip6_frags.lock);
299                         fq_in->q.last_in |= COMPLETE;
300                         fq_put(fq_in, NULL);
301                         return fq;
302                 }
303         }
304 #endif
305         fq = fq_in;
306
307         if (!mod_timer(&fq->q.timer, jiffies + ip6_frags_ctl.timeout))
308                 atomic_inc(&fq->q.refcnt);
309
310         atomic_inc(&fq->q.refcnt);
311         hlist_add_head(&fq->q.list, &ip6_frags.hash[hash]);
312         INIT_LIST_HEAD(&fq->q.lru_list);
313         list_add_tail(&fq->q.lru_list, &ip6_frags.lru_list);
314         ip6_frags.nqueues++;
315         write_unlock(&ip6_frags.lock);
316         return fq;
317 }
318
319
320 static struct frag_queue *
321 ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst,
322                 struct inet6_dev *idev)
323 {
324         struct frag_queue *fq;
325
326         if ((fq = frag_alloc_queue()) == NULL)
327                 goto oom;
328
329         fq->id = id;
330         ipv6_addr_copy(&fq->saddr, src);
331         ipv6_addr_copy(&fq->daddr, dst);
332
333         init_timer(&fq->q.timer);
334         fq->q.timer.function = ip6_frag_expire;
335         fq->q.timer.data = (long) fq;
336         spin_lock_init(&fq->q.lock);
337         atomic_set(&fq->q.refcnt, 1);
338
339         return ip6_frag_intern(fq);
340
341 oom:
342         IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
343         return NULL;
344 }
345
346 static __inline__ struct frag_queue *
347 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
348         struct inet6_dev *idev)
349 {
350         struct frag_queue *fq;
351         struct hlist_node *n;
352         unsigned int hash;
353
354         read_lock(&ip6_frags.lock);
355         hash = ip6qhashfn(id, src, dst);
356         hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
357                 if (fq->id == id &&
358                     ipv6_addr_equal(src, &fq->saddr) &&
359                     ipv6_addr_equal(dst, &fq->daddr)) {
360                         atomic_inc(&fq->q.refcnt);
361                         read_unlock(&ip6_frags.lock);
362                         return fq;
363                 }
364         }
365         read_unlock(&ip6_frags.lock);
366
367         return ip6_frag_create(id, src, dst, idev);
368 }
369
370
371 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
372                            struct frag_hdr *fhdr, int nhoff)
373 {
374         struct sk_buff *prev, *next;
375         struct net_device *dev;
376         int offset, end;
377
378         if (fq->q.last_in & COMPLETE)
379                 goto err;
380
381         offset = ntohs(fhdr->frag_off) & ~0x7;
382         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
383                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
384
385         if ((unsigned int)end > IPV6_MAXPLEN) {
386                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
387                                  IPSTATS_MIB_INHDRERRORS);
388                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
389                                   ((u8 *)&fhdr->frag_off -
390                                    skb_network_header(skb)));
391                 return -1;
392         }
393
394         if (skb->ip_summed == CHECKSUM_COMPLETE) {
395                 const unsigned char *nh = skb_network_header(skb);
396                 skb->csum = csum_sub(skb->csum,
397                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
398                                                   0));
399         }
400
401         /* Is this the final fragment? */
402         if (!(fhdr->frag_off & htons(IP6_MF))) {
403                 /* If we already have some bits beyond end
404                  * or have different end, the segment is corrupted.
405                  */
406                 if (end < fq->q.len ||
407                     ((fq->q.last_in & LAST_IN) && end != fq->q.len))
408                         goto err;
409                 fq->q.last_in |= LAST_IN;
410                 fq->q.len = end;
411         } else {
412                 /* Check if the fragment is rounded to 8 bytes.
413                  * Required by the RFC.
414                  */
415                 if (end & 0x7) {
416                         /* RFC2460 says always send parameter problem in
417                          * this case. -DaveM
418                          */
419                         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
420                                          IPSTATS_MIB_INHDRERRORS);
421                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
422                                           offsetof(struct ipv6hdr, payload_len));
423                         return -1;
424                 }
425                 if (end > fq->q.len) {
426                         /* Some bits beyond end -> corruption. */
427                         if (fq->q.last_in & LAST_IN)
428                                 goto err;
429                         fq->q.len = end;
430                 }
431         }
432
433         if (end == offset)
434                 goto err;
435
436         /* Point into the IP datagram 'data' part. */
437         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
438                 goto err;
439
440         if (pskb_trim_rcsum(skb, end - offset))
441                 goto err;
442
443         /* Find out which fragments are in front and at the back of us
444          * in the chain of fragments so far.  We must know where to put
445          * this fragment, right?
446          */
447         prev = NULL;
448         for(next = fq->q.fragments; next != NULL; next = next->next) {
449                 if (FRAG6_CB(next)->offset >= offset)
450                         break;  /* bingo! */
451                 prev = next;
452         }
453
454         /* We found where to put this one.  Check for overlap with
455          * preceding fragment, and, if needed, align things so that
456          * any overlaps are eliminated.
457          */
458         if (prev) {
459                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
460
461                 if (i > 0) {
462                         offset += i;
463                         if (end <= offset)
464                                 goto err;
465                         if (!pskb_pull(skb, i))
466                                 goto err;
467                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
468                                 skb->ip_summed = CHECKSUM_NONE;
469                 }
470         }
471
472         /* Look for overlap with succeeding segments.
473          * If we can merge fragments, do it.
474          */
475         while (next && FRAG6_CB(next)->offset < end) {
476                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
477
478                 if (i < next->len) {
479                         /* Eat head of the next overlapped fragment
480                          * and leave the loop. The next ones cannot overlap.
481                          */
482                         if (!pskb_pull(next, i))
483                                 goto err;
484                         FRAG6_CB(next)->offset += i;    /* next fragment */
485                         fq->q.meat -= i;
486                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
487                                 next->ip_summed = CHECKSUM_NONE;
488                         break;
489                 } else {
490                         struct sk_buff *free_it = next;
491
492                         /* Old fragment is completely overridden with
493                          * new one drop it.
494                          */
495                         next = next->next;
496
497                         if (prev)
498                                 prev->next = next;
499                         else
500                                 fq->q.fragments = next;
501
502                         fq->q.meat -= free_it->len;
503                         frag_kfree_skb(free_it, NULL);
504                 }
505         }
506
507         FRAG6_CB(skb)->offset = offset;
508
509         /* Insert this fragment in the chain of fragments. */
510         skb->next = next;
511         if (prev)
512                 prev->next = skb;
513         else
514                 fq->q.fragments = skb;
515
516         dev = skb->dev;
517         if (dev) {
518                 fq->iif = dev->ifindex;
519                 skb->dev = NULL;
520         }
521         fq->q.stamp = skb->tstamp;
522         fq->q.meat += skb->len;
523         atomic_add(skb->truesize, &ip6_frags.mem);
524
525         /* The first fragment.
526          * nhoffset is obtained from the first fragment, of course.
527          */
528         if (offset == 0) {
529                 fq->nhoffset = nhoff;
530                 fq->q.last_in |= FIRST_IN;
531         }
532
533         if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
534                 return ip6_frag_reasm(fq, prev, dev);
535
536         write_lock(&ip6_frags.lock);
537         list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
538         write_unlock(&ip6_frags.lock);
539         return -1;
540
541 err:
542         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
543         kfree_skb(skb);
544         return -1;
545 }
546
547 /*
548  *      Check if this packet is complete.
549  *      Returns NULL on failure by any reason, and pointer
550  *      to current nexthdr field in reassembled frame.
551  *
552  *      It is called with locked fq, and caller must check that
553  *      queue is eligible for reassembly i.e. it is not COMPLETE,
554  *      the last and the first frames arrived and all the bits are here.
555  */
556 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
557                           struct net_device *dev)
558 {
559         struct sk_buff *fp, *head = fq->q.fragments;
560         int    payload_len;
561         unsigned int nhoff;
562
563         fq_kill(fq);
564
565         /* Make the one we just received the head. */
566         if (prev) {
567                 head = prev->next;
568                 fp = skb_clone(head, GFP_ATOMIC);
569
570                 if (!fp)
571                         goto out_oom;
572
573                 fp->next = head->next;
574                 prev->next = fp;
575
576                 skb_morph(head, fq->q.fragments);
577                 head->next = fq->q.fragments->next;
578
579                 kfree_skb(fq->q.fragments);
580                 fq->q.fragments = head;
581         }
582
583         BUG_TRAP(head != NULL);
584         BUG_TRAP(FRAG6_CB(head)->offset == 0);
585
586         /* Unfragmented part is taken from the first segment. */
587         payload_len = ((head->data - skb_network_header(head)) -
588                        sizeof(struct ipv6hdr) + fq->q.len -
589                        sizeof(struct frag_hdr));
590         if (payload_len > IPV6_MAXPLEN)
591                 goto out_oversize;
592
593         /* Head of list must not be cloned. */
594         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
595                 goto out_oom;
596
597         /* If the first fragment is fragmented itself, we split
598          * it to two chunks: the first with data and paged part
599          * and the second, holding only fragments. */
600         if (skb_shinfo(head)->frag_list) {
601                 struct sk_buff *clone;
602                 int i, plen = 0;
603
604                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
605                         goto out_oom;
606                 clone->next = head->next;
607                 head->next = clone;
608                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
609                 skb_shinfo(head)->frag_list = NULL;
610                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
611                         plen += skb_shinfo(head)->frags[i].size;
612                 clone->len = clone->data_len = head->data_len - plen;
613                 head->data_len -= clone->len;
614                 head->len -= clone->len;
615                 clone->csum = 0;
616                 clone->ip_summed = head->ip_summed;
617                 atomic_add(clone->truesize, &ip6_frags.mem);
618         }
619
620         /* We have to remove fragment header from datagram and to relocate
621          * header in order to calculate ICV correctly. */
622         nhoff = fq->nhoffset;
623         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
624         memmove(head->head + sizeof(struct frag_hdr), head->head,
625                 (head->data - head->head) - sizeof(struct frag_hdr));
626         head->mac_header += sizeof(struct frag_hdr);
627         head->network_header += sizeof(struct frag_hdr);
628
629         skb_shinfo(head)->frag_list = head->next;
630         skb_reset_transport_header(head);
631         skb_push(head, head->data - skb_network_header(head));
632         atomic_sub(head->truesize, &ip6_frags.mem);
633
634         for (fp=head->next; fp; fp = fp->next) {
635                 head->data_len += fp->len;
636                 head->len += fp->len;
637                 if (head->ip_summed != fp->ip_summed)
638                         head->ip_summed = CHECKSUM_NONE;
639                 else if (head->ip_summed == CHECKSUM_COMPLETE)
640                         head->csum = csum_add(head->csum, fp->csum);
641                 head->truesize += fp->truesize;
642                 atomic_sub(fp->truesize, &ip6_frags.mem);
643         }
644
645         head->next = NULL;
646         head->dev = dev;
647         head->tstamp = fq->q.stamp;
648         ipv6_hdr(head)->payload_len = htons(payload_len);
649         IP6CB(head)->nhoff = nhoff;
650
651         /* Yes, and fold redundant checksum back. 8) */
652         if (head->ip_summed == CHECKSUM_COMPLETE)
653                 head->csum = csum_partial(skb_network_header(head),
654                                           skb_network_header_len(head),
655                                           head->csum);
656
657         rcu_read_lock();
658         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
659         rcu_read_unlock();
660         fq->q.fragments = NULL;
661         return 1;
662
663 out_oversize:
664         if (net_ratelimit())
665                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
666         goto out_fail;
667 out_oom:
668         if (net_ratelimit())
669                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
670 out_fail:
671         rcu_read_lock();
672         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
673         rcu_read_unlock();
674         return -1;
675 }
676
677 static int ipv6_frag_rcv(struct sk_buff **skbp)
678 {
679         struct sk_buff *skb = *skbp;
680         struct frag_hdr *fhdr;
681         struct frag_queue *fq;
682         struct ipv6hdr *hdr = ipv6_hdr(skb);
683
684         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
685
686         /* Jumbo payload inhibits frag. header */
687         if (hdr->payload_len==0) {
688                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
689                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
690                                   skb_network_header_len(skb));
691                 return -1;
692         }
693         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
694                                  sizeof(struct frag_hdr)))) {
695                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
696                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
697                                   skb_network_header_len(skb));
698                 return -1;
699         }
700
701         hdr = ipv6_hdr(skb);
702         fhdr = (struct frag_hdr *)skb_transport_header(skb);
703
704         if (!(fhdr->frag_off & htons(0xFFF9))) {
705                 /* It is not a fragmented frame */
706                 skb->transport_header += sizeof(struct frag_hdr);
707                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
708
709                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
710                 return 1;
711         }
712
713         if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh)
714                 ip6_evictor(ip6_dst_idev(skb->dst));
715
716         if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
717                           ip6_dst_idev(skb->dst))) != NULL) {
718                 int ret;
719
720                 spin_lock(&fq->q.lock);
721
722                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
723
724                 spin_unlock(&fq->q.lock);
725                 fq_put(fq, NULL);
726                 return ret;
727         }
728
729         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
730         kfree_skb(skb);
731         return -1;
732 }
733
734 static struct inet6_protocol frag_protocol =
735 {
736         .handler        =       ipv6_frag_rcv,
737         .flags          =       INET6_PROTO_NOPOLICY,
738 };
739
740 void __init ipv6_frag_init(void)
741 {
742         if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
743                 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
744
745         ip6_frags.ctl = &ip6_frags_ctl;
746         ip6_frags.hashfn = ip6_hashfn;
747         inet_frags_init(&ip6_frags);
748 }