x86/PCI: remove redundant warnings
[linux-2.6.git] / 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  *      Based on: net/ipv4/ip_fragment.c
9  *
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.
14  */
15
16 /*
17  *      Fixes:
18  *      Andi Kleen      Make it work with multiple hosts.
19  *                      More RFC compliance.
20  *
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{}.
25  *      David Stevens and
26  *      YOSHIFUJI,H. @USAGI     Always remove fragment header to
27  *                              calculate ICV correctly.
28  */
29 #include <linux/errno.h>
30 #include <linux/types.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/sockios.h>
34 #include <linux/jiffies.h>
35 #include <linux/net.h>
36 #include <linux/list.h>
37 #include <linux/netdevice.h>
38 #include <linux/in6.h>
39 #include <linux/ipv6.h>
40 #include <linux/icmpv6.h>
41 #include <linux/random.h>
42 #include <linux/jhash.h>
43 #include <linux/skbuff.h>
44
45 #include <net/sock.h>
46 #include <net/snmp.h>
47
48 #include <net/ipv6.h>
49 #include <net/ip6_route.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>
55 #include <net/inet_frag.h>
56
57 struct ip6frag_skb_cb
58 {
59         struct inet6_skb_parm   h;
60         int                     offset;
61 };
62
63 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
64
65
66 /*
67  *      Equivalent of ipv4 struct ipq
68  */
69
70 struct frag_queue
71 {
72         struct inet_frag_queue  q;
73
74         __be32                  id;             /* fragment id          */
75         u32                     user;
76         struct in6_addr         saddr;
77         struct in6_addr         daddr;
78
79         int                     iif;
80         unsigned int            csum;
81         __u16                   nhoffset;
82 };
83
84 static struct inet_frags ip6_frags;
85
86 int ip6_frag_nqueues(struct net *net)
87 {
88         return net->ipv6.frags.nqueues;
89 }
90
91 int ip6_frag_mem(struct net *net)
92 {
93         return atomic_read(&net->ipv6.frags.mem);
94 }
95
96 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
97                           struct net_device *dev);
98
99 /*
100  * callers should be careful not to use the hash value outside the ipfrag_lock
101  * as doing so could race with ipfrag_hash_rnd being recalculated.
102  */
103 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
104                              const struct in6_addr *daddr, u32 rnd)
105 {
106         u32 a, b, c;
107
108         a = (__force u32)saddr->s6_addr32[0];
109         b = (__force u32)saddr->s6_addr32[1];
110         c = (__force u32)saddr->s6_addr32[2];
111
112         a += JHASH_GOLDEN_RATIO;
113         b += JHASH_GOLDEN_RATIO;
114         c += rnd;
115         __jhash_mix(a, b, c);
116
117         a += (__force u32)saddr->s6_addr32[3];
118         b += (__force u32)daddr->s6_addr32[0];
119         c += (__force u32)daddr->s6_addr32[1];
120         __jhash_mix(a, b, c);
121
122         a += (__force u32)daddr->s6_addr32[2];
123         b += (__force u32)daddr->s6_addr32[3];
124         c += (__force u32)id;
125         __jhash_mix(a, b, c);
126
127         return c & (INETFRAGS_HASHSZ - 1);
128 }
129 EXPORT_SYMBOL_GPL(inet6_hash_frag);
130
131 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
132 {
133         struct frag_queue *fq;
134
135         fq = container_of(q, struct frag_queue, q);
136         return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
137 }
138
139 int ip6_frag_match(struct inet_frag_queue *q, void *a)
140 {
141         struct frag_queue *fq;
142         struct ip6_create_arg *arg = a;
143
144         fq = container_of(q, struct frag_queue, q);
145         return (fq->id == arg->id && fq->user == arg->user &&
146                         ipv6_addr_equal(&fq->saddr, arg->src) &&
147                         ipv6_addr_equal(&fq->daddr, arg->dst));
148 }
149 EXPORT_SYMBOL(ip6_frag_match);
150
151 /* Memory Tracking Functions. */
152 static inline void frag_kfree_skb(struct netns_frags *nf,
153                 struct sk_buff *skb, int *work)
154 {
155         if (work)
156                 *work -= skb->truesize;
157         atomic_sub(skb->truesize, &nf->mem);
158         kfree_skb(skb);
159 }
160
161 void ip6_frag_init(struct inet_frag_queue *q, void *a)
162 {
163         struct frag_queue *fq = container_of(q, struct frag_queue, q);
164         struct ip6_create_arg *arg = a;
165
166         fq->id = arg->id;
167         fq->user = arg->user;
168         ipv6_addr_copy(&fq->saddr, arg->src);
169         ipv6_addr_copy(&fq->daddr, arg->dst);
170 }
171 EXPORT_SYMBOL(ip6_frag_init);
172
173 /* Destruction primitives. */
174
175 static __inline__ void fq_put(struct frag_queue *fq)
176 {
177         inet_frag_put(&fq->q, &ip6_frags);
178 }
179
180 /* Kill fq entry. It is not destroyed immediately,
181  * because caller (and someone more) holds reference count.
182  */
183 static __inline__ void fq_kill(struct frag_queue *fq)
184 {
185         inet_frag_kill(&fq->q, &ip6_frags);
186 }
187
188 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
189 {
190         int evicted;
191
192         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
193         if (evicted)
194                 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
195 }
196
197 static void ip6_frag_expire(unsigned long data)
198 {
199         struct frag_queue *fq;
200         struct net_device *dev = NULL;
201         struct net *net;
202
203         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
204
205         spin_lock(&fq->q.lock);
206
207         if (fq->q.last_in & INET_FRAG_COMPLETE)
208                 goto out;
209
210         fq_kill(fq);
211
212         net = container_of(fq->q.net, struct net, ipv6.frags);
213         rcu_read_lock();
214         dev = dev_get_by_index_rcu(net, fq->iif);
215         if (!dev)
216                 goto out_rcu_unlock;
217
218         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
219         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
220
221         /* Don't send error if the first segment did not arrive. */
222         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
223                 goto out_rcu_unlock;
224
225         /*
226            But use as source device on which LAST ARRIVED
227            segment was received. And do not use fq->dev
228            pointer directly, device might already disappeared.
229          */
230         fq->q.fragments->dev = dev;
231         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
232 out_rcu_unlock:
233         rcu_read_unlock();
234 out:
235         spin_unlock(&fq->q.lock);
236         fq_put(fq);
237 }
238
239 static __inline__ struct frag_queue *
240 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst)
241 {
242         struct inet_frag_queue *q;
243         struct ip6_create_arg arg;
244         unsigned int hash;
245
246         arg.id = id;
247         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
248         arg.src = src;
249         arg.dst = dst;
250
251         read_lock(&ip6_frags.lock);
252         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
253
254         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
255         if (q == NULL)
256                 return NULL;
257
258         return container_of(q, struct frag_queue, q);
259 }
260
261 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
262                            struct frag_hdr *fhdr, int nhoff)
263 {
264         struct sk_buff *prev, *next;
265         struct net_device *dev;
266         int offset, end;
267         struct net *net = dev_net(skb_dst(skb)->dev);
268
269         if (fq->q.last_in & INET_FRAG_COMPLETE)
270                 goto err;
271
272         offset = ntohs(fhdr->frag_off) & ~0x7;
273         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
274                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
275
276         if ((unsigned int)end > IPV6_MAXPLEN) {
277                 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
278                                  IPSTATS_MIB_INHDRERRORS);
279                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
280                                   ((u8 *)&fhdr->frag_off -
281                                    skb_network_header(skb)));
282                 return -1;
283         }
284
285         if (skb->ip_summed == CHECKSUM_COMPLETE) {
286                 const unsigned char *nh = skb_network_header(skb);
287                 skb->csum = csum_sub(skb->csum,
288                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
289                                                   0));
290         }
291
292         /* Is this the final fragment? */
293         if (!(fhdr->frag_off & htons(IP6_MF))) {
294                 /* If we already have some bits beyond end
295                  * or have different end, the segment is corrupted.
296                  */
297                 if (end < fq->q.len ||
298                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
299                         goto err;
300                 fq->q.last_in |= INET_FRAG_LAST_IN;
301                 fq->q.len = end;
302         } else {
303                 /* Check if the fragment is rounded to 8 bytes.
304                  * Required by the RFC.
305                  */
306                 if (end & 0x7) {
307                         /* RFC2460 says always send parameter problem in
308                          * this case. -DaveM
309                          */
310                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
311                                          IPSTATS_MIB_INHDRERRORS);
312                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
313                                           offsetof(struct ipv6hdr, payload_len));
314                         return -1;
315                 }
316                 if (end > fq->q.len) {
317                         /* Some bits beyond end -> corruption. */
318                         if (fq->q.last_in & INET_FRAG_LAST_IN)
319                                 goto err;
320                         fq->q.len = end;
321                 }
322         }
323
324         if (end == offset)
325                 goto err;
326
327         /* Point into the IP datagram 'data' part. */
328         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
329                 goto err;
330
331         if (pskb_trim_rcsum(skb, end - offset))
332                 goto err;
333
334         /* Find out which fragments are in front and at the back of us
335          * in the chain of fragments so far.  We must know where to put
336          * this fragment, right?
337          */
338         prev = NULL;
339         for(next = fq->q.fragments; next != NULL; next = next->next) {
340                 if (FRAG6_CB(next)->offset >= offset)
341                         break;  /* bingo! */
342                 prev = next;
343         }
344
345         /* We found where to put this one.  Check for overlap with
346          * preceding fragment, and, if needed, align things so that
347          * any overlaps are eliminated.
348          */
349         if (prev) {
350                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
351
352                 if (i > 0) {
353                         offset += i;
354                         if (end <= offset)
355                                 goto err;
356                         if (!pskb_pull(skb, i))
357                                 goto err;
358                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
359                                 skb->ip_summed = CHECKSUM_NONE;
360                 }
361         }
362
363         /* Look for overlap with succeeding segments.
364          * If we can merge fragments, do it.
365          */
366         while (next && FRAG6_CB(next)->offset < end) {
367                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
368
369                 if (i < next->len) {
370                         /* Eat head of the next overlapped fragment
371                          * and leave the loop. The next ones cannot overlap.
372                          */
373                         if (!pskb_pull(next, i))
374                                 goto err;
375                         FRAG6_CB(next)->offset += i;    /* next fragment */
376                         fq->q.meat -= i;
377                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
378                                 next->ip_summed = CHECKSUM_NONE;
379                         break;
380                 } else {
381                         struct sk_buff *free_it = next;
382
383                         /* Old fragment is completely overridden with
384                          * new one drop it.
385                          */
386                         next = next->next;
387
388                         if (prev)
389                                 prev->next = next;
390                         else
391                                 fq->q.fragments = next;
392
393                         fq->q.meat -= free_it->len;
394                         frag_kfree_skb(fq->q.net, free_it, NULL);
395                 }
396         }
397
398         FRAG6_CB(skb)->offset = offset;
399
400         /* Insert this fragment in the chain of fragments. */
401         skb->next = next;
402         if (prev)
403                 prev->next = skb;
404         else
405                 fq->q.fragments = skb;
406
407         dev = skb->dev;
408         if (dev) {
409                 fq->iif = dev->ifindex;
410                 skb->dev = NULL;
411         }
412         fq->q.stamp = skb->tstamp;
413         fq->q.meat += skb->len;
414         atomic_add(skb->truesize, &fq->q.net->mem);
415
416         /* The first fragment.
417          * nhoffset is obtained from the first fragment, of course.
418          */
419         if (offset == 0) {
420                 fq->nhoffset = nhoff;
421                 fq->q.last_in |= INET_FRAG_FIRST_IN;
422         }
423
424         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
425             fq->q.meat == fq->q.len)
426                 return ip6_frag_reasm(fq, prev, dev);
427
428         write_lock(&ip6_frags.lock);
429         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
430         write_unlock(&ip6_frags.lock);
431         return -1;
432
433 err:
434         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
435                       IPSTATS_MIB_REASMFAILS);
436         kfree_skb(skb);
437         return -1;
438 }
439
440 /*
441  *      Check if this packet is complete.
442  *      Returns NULL on failure by any reason, and pointer
443  *      to current nexthdr field in reassembled frame.
444  *
445  *      It is called with locked fq, and caller must check that
446  *      queue is eligible for reassembly i.e. it is not COMPLETE,
447  *      the last and the first frames arrived and all the bits are here.
448  */
449 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
450                           struct net_device *dev)
451 {
452         struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
453         struct sk_buff *fp, *head = fq->q.fragments;
454         int    payload_len;
455         unsigned int nhoff;
456
457         fq_kill(fq);
458
459         /* Make the one we just received the head. */
460         if (prev) {
461                 head = prev->next;
462                 fp = skb_clone(head, GFP_ATOMIC);
463
464                 if (!fp)
465                         goto out_oom;
466
467                 fp->next = head->next;
468                 prev->next = fp;
469
470                 skb_morph(head, fq->q.fragments);
471                 head->next = fq->q.fragments->next;
472
473                 kfree_skb(fq->q.fragments);
474                 fq->q.fragments = head;
475         }
476
477         WARN_ON(head == NULL);
478         WARN_ON(FRAG6_CB(head)->offset != 0);
479
480         /* Unfragmented part is taken from the first segment. */
481         payload_len = ((head->data - skb_network_header(head)) -
482                        sizeof(struct ipv6hdr) + fq->q.len -
483                        sizeof(struct frag_hdr));
484         if (payload_len > IPV6_MAXPLEN)
485                 goto out_oversize;
486
487         /* Head of list must not be cloned. */
488         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
489                 goto out_oom;
490
491         /* If the first fragment is fragmented itself, we split
492          * it to two chunks: the first with data and paged part
493          * and the second, holding only fragments. */
494         if (skb_has_frags(head)) {
495                 struct sk_buff *clone;
496                 int i, plen = 0;
497
498                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
499                         goto out_oom;
500                 clone->next = head->next;
501                 head->next = clone;
502                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
503                 skb_frag_list_init(head);
504                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
505                         plen += skb_shinfo(head)->frags[i].size;
506                 clone->len = clone->data_len = head->data_len - plen;
507                 head->data_len -= clone->len;
508                 head->len -= clone->len;
509                 clone->csum = 0;
510                 clone->ip_summed = head->ip_summed;
511                 atomic_add(clone->truesize, &fq->q.net->mem);
512         }
513
514         /* We have to remove fragment header from datagram and to relocate
515          * header in order to calculate ICV correctly. */
516         nhoff = fq->nhoffset;
517         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
518         memmove(head->head + sizeof(struct frag_hdr), head->head,
519                 (head->data - head->head) - sizeof(struct frag_hdr));
520         head->mac_header += sizeof(struct frag_hdr);
521         head->network_header += sizeof(struct frag_hdr);
522
523         skb_shinfo(head)->frag_list = head->next;
524         skb_reset_transport_header(head);
525         skb_push(head, head->data - skb_network_header(head));
526         atomic_sub(head->truesize, &fq->q.net->mem);
527
528         for (fp=head->next; fp; fp = fp->next) {
529                 head->data_len += fp->len;
530                 head->len += fp->len;
531                 if (head->ip_summed != fp->ip_summed)
532                         head->ip_summed = CHECKSUM_NONE;
533                 else if (head->ip_summed == CHECKSUM_COMPLETE)
534                         head->csum = csum_add(head->csum, fp->csum);
535                 head->truesize += fp->truesize;
536                 atomic_sub(fp->truesize, &fq->q.net->mem);
537         }
538
539         head->next = NULL;
540         head->dev = dev;
541         head->tstamp = fq->q.stamp;
542         ipv6_hdr(head)->payload_len = htons(payload_len);
543         IP6CB(head)->nhoff = nhoff;
544
545         /* Yes, and fold redundant checksum back. 8) */
546         if (head->ip_summed == CHECKSUM_COMPLETE)
547                 head->csum = csum_partial(skb_network_header(head),
548                                           skb_network_header_len(head),
549                                           head->csum);
550
551         rcu_read_lock();
552         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
553         rcu_read_unlock();
554         fq->q.fragments = NULL;
555         return 1;
556
557 out_oversize:
558         if (net_ratelimit())
559                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
560         goto out_fail;
561 out_oom:
562         if (net_ratelimit())
563                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
564 out_fail:
565         rcu_read_lock();
566         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
567         rcu_read_unlock();
568         return -1;
569 }
570
571 static int ipv6_frag_rcv(struct sk_buff *skb)
572 {
573         struct frag_hdr *fhdr;
574         struct frag_queue *fq;
575         struct ipv6hdr *hdr = ipv6_hdr(skb);
576         struct net *net = dev_net(skb_dst(skb)->dev);
577
578         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
579
580         /* Jumbo payload inhibits frag. header */
581         if (hdr->payload_len==0)
582                 goto fail_hdr;
583
584         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
585                                  sizeof(struct frag_hdr))))
586                 goto fail_hdr;
587
588         hdr = ipv6_hdr(skb);
589         fhdr = (struct frag_hdr *)skb_transport_header(skb);
590
591         if (!(fhdr->frag_off & htons(0xFFF9))) {
592                 /* It is not a fragmented frame */
593                 skb->transport_header += sizeof(struct frag_hdr);
594                 IP6_INC_STATS_BH(net,
595                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
596
597                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
598                 return 1;
599         }
600
601         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
602                 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
603
604         fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
605         if (fq != NULL) {
606                 int ret;
607
608                 spin_lock(&fq->q.lock);
609
610                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
611
612                 spin_unlock(&fq->q.lock);
613                 fq_put(fq);
614                 return ret;
615         }
616
617         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
618         kfree_skb(skb);
619         return -1;
620
621 fail_hdr:
622         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
623         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
624         return -1;
625 }
626
627 static const struct inet6_protocol frag_protocol =
628 {
629         .handler        =       ipv6_frag_rcv,
630         .flags          =       INET6_PROTO_NOPOLICY,
631 };
632
633 #ifdef CONFIG_SYSCTL
634 static struct ctl_table ip6_frags_ns_ctl_table[] = {
635         {
636                 .procname       = "ip6frag_high_thresh",
637                 .data           = &init_net.ipv6.frags.high_thresh,
638                 .maxlen         = sizeof(int),
639                 .mode           = 0644,
640                 .proc_handler   = proc_dointvec
641         },
642         {
643                 .procname       = "ip6frag_low_thresh",
644                 .data           = &init_net.ipv6.frags.low_thresh,
645                 .maxlen         = sizeof(int),
646                 .mode           = 0644,
647                 .proc_handler   = proc_dointvec
648         },
649         {
650                 .procname       = "ip6frag_time",
651                 .data           = &init_net.ipv6.frags.timeout,
652                 .maxlen         = sizeof(int),
653                 .mode           = 0644,
654                 .proc_handler   = proc_dointvec_jiffies,
655         },
656         { }
657 };
658
659 static struct ctl_table ip6_frags_ctl_table[] = {
660         {
661                 .procname       = "ip6frag_secret_interval",
662                 .data           = &ip6_frags.secret_interval,
663                 .maxlen         = sizeof(int),
664                 .mode           = 0644,
665                 .proc_handler   = proc_dointvec_jiffies,
666         },
667         { }
668 };
669
670 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
671 {
672         struct ctl_table *table;
673         struct ctl_table_header *hdr;
674
675         table = ip6_frags_ns_ctl_table;
676         if (!net_eq(net, &init_net)) {
677                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
678                 if (table == NULL)
679                         goto err_alloc;
680
681                 table[0].data = &net->ipv6.frags.high_thresh;
682                 table[1].data = &net->ipv6.frags.low_thresh;
683                 table[2].data = &net->ipv6.frags.timeout;
684         }
685
686         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
687         if (hdr == NULL)
688                 goto err_reg;
689
690         net->ipv6.sysctl.frags_hdr = hdr;
691         return 0;
692
693 err_reg:
694         if (!net_eq(net, &init_net))
695                 kfree(table);
696 err_alloc:
697         return -ENOMEM;
698 }
699
700 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
701 {
702         struct ctl_table *table;
703
704         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
705         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
706         if (!net_eq(net, &init_net))
707                 kfree(table);
708 }
709
710 static struct ctl_table_header *ip6_ctl_header;
711
712 static int ip6_frags_sysctl_register(void)
713 {
714         ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
715                         ip6_frags_ctl_table);
716         return ip6_ctl_header == NULL ? -ENOMEM : 0;
717 }
718
719 static void ip6_frags_sysctl_unregister(void)
720 {
721         unregister_net_sysctl_table(ip6_ctl_header);
722 }
723 #else
724 static inline int ip6_frags_ns_sysctl_register(struct net *net)
725 {
726         return 0;
727 }
728
729 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
730 {
731 }
732
733 static inline int ip6_frags_sysctl_register(void)
734 {
735         return 0;
736 }
737
738 static inline void ip6_frags_sysctl_unregister(void)
739 {
740 }
741 #endif
742
743 static int __net_init ipv6_frags_init_net(struct net *net)
744 {
745         net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
746         net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
747         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
748
749         inet_frags_init_net(&net->ipv6.frags);
750
751         return ip6_frags_ns_sysctl_register(net);
752 }
753
754 static void __net_exit ipv6_frags_exit_net(struct net *net)
755 {
756         ip6_frags_ns_sysctl_unregister(net);
757         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
758 }
759
760 static struct pernet_operations ip6_frags_ops = {
761         .init = ipv6_frags_init_net,
762         .exit = ipv6_frags_exit_net,
763 };
764
765 int __init ipv6_frag_init(void)
766 {
767         int ret;
768
769         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
770         if (ret)
771                 goto out;
772
773         ret = ip6_frags_sysctl_register();
774         if (ret)
775                 goto err_sysctl;
776
777         ret = register_pernet_subsys(&ip6_frags_ops);
778         if (ret)
779                 goto err_pernet;
780
781         ip6_frags.hashfn = ip6_hashfn;
782         ip6_frags.constructor = ip6_frag_init;
783         ip6_frags.destructor = NULL;
784         ip6_frags.skb_free = NULL;
785         ip6_frags.qsize = sizeof(struct frag_queue);
786         ip6_frags.match = ip6_frag_match;
787         ip6_frags.frag_expire = ip6_frag_expire;
788         ip6_frags.secret_interval = 10 * 60 * HZ;
789         inet_frags_init(&ip6_frags);
790 out:
791         return ret;
792
793 err_pernet:
794         ip6_frags_sysctl_unregister();
795 err_sysctl:
796         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
797         goto out;
798 }
799
800 void ipv6_frag_exit(void)
801 {
802         inet_frags_fini(&ip6_frags);
803         ip6_frags_sysctl_unregister();
804         unregister_pernet_subsys(&ip6_frags_ops);
805         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
806 }