[NETNS][FRAGS]: Make thresholds work in namespaces.
[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 static struct inet_frags ip6_frags;
86
87 int ip6_frag_nqueues(struct net *net)
88 {
89         return net->ipv6.frags.nqueues;
90 }
91
92 int ip6_frag_mem(struct net *net)
93 {
94         return atomic_read(&net->ipv6.frags.mem);
95 }
96
97 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
98                           struct net_device *dev);
99
100 /*
101  * callers should be careful not to use the hash value outside the ipfrag_lock
102  * as doing so could race with ipfrag_hash_rnd being recalculated.
103  */
104 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
105                                struct in6_addr *daddr)
106 {
107         u32 a, b, c;
108
109         a = (__force u32)saddr->s6_addr32[0];
110         b = (__force u32)saddr->s6_addr32[1];
111         c = (__force u32)saddr->s6_addr32[2];
112
113         a += JHASH_GOLDEN_RATIO;
114         b += JHASH_GOLDEN_RATIO;
115         c += ip6_frags.rnd;
116         __jhash_mix(a, b, c);
117
118         a += (__force u32)saddr->s6_addr32[3];
119         b += (__force u32)daddr->s6_addr32[0];
120         c += (__force u32)daddr->s6_addr32[1];
121         __jhash_mix(a, b, c);
122
123         a += (__force u32)daddr->s6_addr32[2];
124         b += (__force u32)daddr->s6_addr32[3];
125         c += (__force u32)id;
126         __jhash_mix(a, b, c);
127
128         return c & (INETFRAGS_HASHSZ - 1);
129 }
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 ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
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 &&
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         ipv6_addr_copy(&fq->saddr, arg->src);
168         ipv6_addr_copy(&fq->daddr, arg->dst);
169 }
170 EXPORT_SYMBOL(ip6_frag_init);
171
172 /* Destruction primitives. */
173
174 static __inline__ void fq_put(struct frag_queue *fq)
175 {
176         inet_frag_put(&fq->q, &ip6_frags);
177 }
178
179 /* Kill fq entry. It is not destroyed immediately,
180  * because caller (and someone more) holds reference count.
181  */
182 static __inline__ void fq_kill(struct frag_queue *fq)
183 {
184         inet_frag_kill(&fq->q, &ip6_frags);
185 }
186
187 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
188 {
189         int evicted;
190
191         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
192         if (evicted)
193                 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
194 }
195
196 static void ip6_frag_expire(unsigned long data)
197 {
198         struct frag_queue *fq;
199         struct net_device *dev = NULL;
200
201         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
202
203         spin_lock(&fq->q.lock);
204
205         if (fq->q.last_in & COMPLETE)
206                 goto out;
207
208         fq_kill(fq);
209
210         dev = dev_get_by_index(&init_net, fq->iif);
211         if (!dev)
212                 goto out;
213
214         rcu_read_lock();
215         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
216         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
217         rcu_read_unlock();
218
219         /* Don't send error if the first segment did not arrive. */
220         if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
221                 goto out;
222
223         /*
224            But use as source device on which LAST ARRIVED
225            segment was received. And do not use fq->dev
226            pointer directly, device might already disappeared.
227          */
228         fq->q.fragments->dev = dev;
229         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
230 out:
231         if (dev)
232                 dev_put(dev);
233         spin_unlock(&fq->q.lock);
234         fq_put(fq);
235 }
236
237 static __inline__ struct frag_queue *
238 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
239         struct inet6_dev *idev)
240 {
241         struct inet_frag_queue *q;
242         struct ip6_create_arg arg;
243         unsigned int hash;
244
245         arg.id = id;
246         arg.src = src;
247         arg.dst = dst;
248         hash = ip6qhashfn(id, src, dst);
249
250         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
251         if (q == NULL)
252                 goto oom;
253
254         return container_of(q, struct frag_queue, q);
255
256 oom:
257         IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
258         return NULL;
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
268         if (fq->q.last_in & COMPLETE)
269                 goto err;
270
271         offset = ntohs(fhdr->frag_off) & ~0x7;
272         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
273                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
274
275         if ((unsigned int)end > IPV6_MAXPLEN) {
276                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
277                                  IPSTATS_MIB_INHDRERRORS);
278                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
279                                   ((u8 *)&fhdr->frag_off -
280                                    skb_network_header(skb)));
281                 return -1;
282         }
283
284         if (skb->ip_summed == CHECKSUM_COMPLETE) {
285                 const unsigned char *nh = skb_network_header(skb);
286                 skb->csum = csum_sub(skb->csum,
287                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
288                                                   0));
289         }
290
291         /* Is this the final fragment? */
292         if (!(fhdr->frag_off & htons(IP6_MF))) {
293                 /* If we already have some bits beyond end
294                  * or have different end, the segment is corrupted.
295                  */
296                 if (end < fq->q.len ||
297                     ((fq->q.last_in & LAST_IN) && end != fq->q.len))
298                         goto err;
299                 fq->q.last_in |= LAST_IN;
300                 fq->q.len = end;
301         } else {
302                 /* Check if the fragment is rounded to 8 bytes.
303                  * Required by the RFC.
304                  */
305                 if (end & 0x7) {
306                         /* RFC2460 says always send parameter problem in
307                          * this case. -DaveM
308                          */
309                         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
310                                          IPSTATS_MIB_INHDRERRORS);
311                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
312                                           offsetof(struct ipv6hdr, payload_len));
313                         return -1;
314                 }
315                 if (end > fq->q.len) {
316                         /* Some bits beyond end -> corruption. */
317                         if (fq->q.last_in & LAST_IN)
318                                 goto err;
319                         fq->q.len = end;
320                 }
321         }
322
323         if (end == offset)
324                 goto err;
325
326         /* Point into the IP datagram 'data' part. */
327         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
328                 goto err;
329
330         if (pskb_trim_rcsum(skb, end - offset))
331                 goto err;
332
333         /* Find out which fragments are in front and at the back of us
334          * in the chain of fragments so far.  We must know where to put
335          * this fragment, right?
336          */
337         prev = NULL;
338         for(next = fq->q.fragments; next != NULL; next = next->next) {
339                 if (FRAG6_CB(next)->offset >= offset)
340                         break;  /* bingo! */
341                 prev = next;
342         }
343
344         /* We found where to put this one.  Check for overlap with
345          * preceding fragment, and, if needed, align things so that
346          * any overlaps are eliminated.
347          */
348         if (prev) {
349                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
350
351                 if (i > 0) {
352                         offset += i;
353                         if (end <= offset)
354                                 goto err;
355                         if (!pskb_pull(skb, i))
356                                 goto err;
357                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
358                                 skb->ip_summed = CHECKSUM_NONE;
359                 }
360         }
361
362         /* Look for overlap with succeeding segments.
363          * If we can merge fragments, do it.
364          */
365         while (next && FRAG6_CB(next)->offset < end) {
366                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
367
368                 if (i < next->len) {
369                         /* Eat head of the next overlapped fragment
370                          * and leave the loop. The next ones cannot overlap.
371                          */
372                         if (!pskb_pull(next, i))
373                                 goto err;
374                         FRAG6_CB(next)->offset += i;    /* next fragment */
375                         fq->q.meat -= i;
376                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
377                                 next->ip_summed = CHECKSUM_NONE;
378                         break;
379                 } else {
380                         struct sk_buff *free_it = next;
381
382                         /* Old fragment is completely overridden with
383                          * new one drop it.
384                          */
385                         next = next->next;
386
387                         if (prev)
388                                 prev->next = next;
389                         else
390                                 fq->q.fragments = next;
391
392                         fq->q.meat -= free_it->len;
393                         frag_kfree_skb(fq->q.net, free_it, NULL);
394                 }
395         }
396
397         FRAG6_CB(skb)->offset = offset;
398
399         /* Insert this fragment in the chain of fragments. */
400         skb->next = next;
401         if (prev)
402                 prev->next = skb;
403         else
404                 fq->q.fragments = skb;
405
406         dev = skb->dev;
407         if (dev) {
408                 fq->iif = dev->ifindex;
409                 skb->dev = NULL;
410         }
411         fq->q.stamp = skb->tstamp;
412         fq->q.meat += skb->len;
413         atomic_add(skb->truesize, &fq->q.net->mem);
414
415         /* The first fragment.
416          * nhoffset is obtained from the first fragment, of course.
417          */
418         if (offset == 0) {
419                 fq->nhoffset = nhoff;
420                 fq->q.last_in |= FIRST_IN;
421         }
422
423         if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
424                 return ip6_frag_reasm(fq, prev, dev);
425
426         write_lock(&ip6_frags.lock);
427         list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
428         write_unlock(&ip6_frags.lock);
429         return -1;
430
431 err:
432         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
433         kfree_skb(skb);
434         return -1;
435 }
436
437 /*
438  *      Check if this packet is complete.
439  *      Returns NULL on failure by any reason, and pointer
440  *      to current nexthdr field in reassembled frame.
441  *
442  *      It is called with locked fq, and caller must check that
443  *      queue is eligible for reassembly i.e. it is not COMPLETE,
444  *      the last and the first frames arrived and all the bits are here.
445  */
446 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
447                           struct net_device *dev)
448 {
449         struct sk_buff *fp, *head = fq->q.fragments;
450         int    payload_len;
451         unsigned int nhoff;
452
453         fq_kill(fq);
454
455         /* Make the one we just received the head. */
456         if (prev) {
457                 head = prev->next;
458                 fp = skb_clone(head, GFP_ATOMIC);
459
460                 if (!fp)
461                         goto out_oom;
462
463                 fp->next = head->next;
464                 prev->next = fp;
465
466                 skb_morph(head, fq->q.fragments);
467                 head->next = fq->q.fragments->next;
468
469                 kfree_skb(fq->q.fragments);
470                 fq->q.fragments = head;
471         }
472
473         BUG_TRAP(head != NULL);
474         BUG_TRAP(FRAG6_CB(head)->offset == 0);
475
476         /* Unfragmented part is taken from the first segment. */
477         payload_len = ((head->data - skb_network_header(head)) -
478                        sizeof(struct ipv6hdr) + fq->q.len -
479                        sizeof(struct frag_hdr));
480         if (payload_len > IPV6_MAXPLEN)
481                 goto out_oversize;
482
483         /* Head of list must not be cloned. */
484         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
485                 goto out_oom;
486
487         /* If the first fragment is fragmented itself, we split
488          * it to two chunks: the first with data and paged part
489          * and the second, holding only fragments. */
490         if (skb_shinfo(head)->frag_list) {
491                 struct sk_buff *clone;
492                 int i, plen = 0;
493
494                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
495                         goto out_oom;
496                 clone->next = head->next;
497                 head->next = clone;
498                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
499                 skb_shinfo(head)->frag_list = NULL;
500                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
501                         plen += skb_shinfo(head)->frags[i].size;
502                 clone->len = clone->data_len = head->data_len - plen;
503                 head->data_len -= clone->len;
504                 head->len -= clone->len;
505                 clone->csum = 0;
506                 clone->ip_summed = head->ip_summed;
507                 atomic_add(clone->truesize, &fq->q.net->mem);
508         }
509
510         /* We have to remove fragment header from datagram and to relocate
511          * header in order to calculate ICV correctly. */
512         nhoff = fq->nhoffset;
513         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
514         memmove(head->head + sizeof(struct frag_hdr), head->head,
515                 (head->data - head->head) - sizeof(struct frag_hdr));
516         head->mac_header += sizeof(struct frag_hdr);
517         head->network_header += sizeof(struct frag_hdr);
518
519         skb_shinfo(head)->frag_list = head->next;
520         skb_reset_transport_header(head);
521         skb_push(head, head->data - skb_network_header(head));
522         atomic_sub(head->truesize, &fq->q.net->mem);
523
524         for (fp=head->next; fp; fp = fp->next) {
525                 head->data_len += fp->len;
526                 head->len += fp->len;
527                 if (head->ip_summed != fp->ip_summed)
528                         head->ip_summed = CHECKSUM_NONE;
529                 else if (head->ip_summed == CHECKSUM_COMPLETE)
530                         head->csum = csum_add(head->csum, fp->csum);
531                 head->truesize += fp->truesize;
532                 atomic_sub(fp->truesize, &fq->q.net->mem);
533         }
534
535         head->next = NULL;
536         head->dev = dev;
537         head->tstamp = fq->q.stamp;
538         ipv6_hdr(head)->payload_len = htons(payload_len);
539         IP6CB(head)->nhoff = nhoff;
540
541         /* Yes, and fold redundant checksum back. 8) */
542         if (head->ip_summed == CHECKSUM_COMPLETE)
543                 head->csum = csum_partial(skb_network_header(head),
544                                           skb_network_header_len(head),
545                                           head->csum);
546
547         rcu_read_lock();
548         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
549         rcu_read_unlock();
550         fq->q.fragments = NULL;
551         return 1;
552
553 out_oversize:
554         if (net_ratelimit())
555                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
556         goto out_fail;
557 out_oom:
558         if (net_ratelimit())
559                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
560 out_fail:
561         rcu_read_lock();
562         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
563         rcu_read_unlock();
564         return -1;
565 }
566
567 static int ipv6_frag_rcv(struct sk_buff *skb)
568 {
569         struct frag_hdr *fhdr;
570         struct frag_queue *fq;
571         struct ipv6hdr *hdr = ipv6_hdr(skb);
572         struct net *net;
573
574         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
575
576         /* Jumbo payload inhibits frag. header */
577         if (hdr->payload_len==0) {
578                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
579                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
580                                   skb_network_header_len(skb));
581                 return -1;
582         }
583         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
584                                  sizeof(struct frag_hdr)))) {
585                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
586                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
587                                   skb_network_header_len(skb));
588                 return -1;
589         }
590
591         hdr = ipv6_hdr(skb);
592         fhdr = (struct frag_hdr *)skb_transport_header(skb);
593
594         if (!(fhdr->frag_off & htons(0xFFF9))) {
595                 /* It is not a fragmented frame */
596                 skb->transport_header += sizeof(struct frag_hdr);
597                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
598
599                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
600                 return 1;
601         }
602
603         net = skb->dev->nd_net;
604         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
605                 ip6_evictor(net, ip6_dst_idev(skb->dst));
606
607         if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
608                           ip6_dst_idev(skb->dst))) != NULL) {
609                 int ret;
610
611                 spin_lock(&fq->q.lock);
612
613                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
614
615                 spin_unlock(&fq->q.lock);
616                 fq_put(fq);
617                 return ret;
618         }
619
620         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
621         kfree_skb(skb);
622         return -1;
623 }
624
625 static struct inet6_protocol frag_protocol =
626 {
627         .handler        =       ipv6_frag_rcv,
628         .flags          =       INET6_PROTO_NOPOLICY,
629 };
630
631 #ifdef CONFIG_SYSCTL
632 static struct ctl_table ip6_frags_ctl_table[] = {
633         {
634                 .ctl_name       = NET_IPV6_IP6FRAG_HIGH_THRESH,
635                 .procname       = "ip6frag_high_thresh",
636                 .data           = &init_net.ipv6.frags.high_thresh,
637                 .maxlen         = sizeof(int),
638                 .mode           = 0644,
639                 .proc_handler   = &proc_dointvec
640         },
641         {
642                 .ctl_name       = NET_IPV6_IP6FRAG_LOW_THRESH,
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                 .ctl_name       = NET_IPV6_IP6FRAG_TIME,
651                 .procname       = "ip6frag_time",
652                 .data           = &init_net.ipv6.frags.timeout,
653                 .maxlen         = sizeof(int),
654                 .mode           = 0644,
655                 .proc_handler   = &proc_dointvec_jiffies,
656                 .strategy       = &sysctl_jiffies,
657         },
658         {
659                 .ctl_name       = NET_IPV6_IP6FRAG_SECRET_INTERVAL,
660                 .procname       = "ip6frag_secret_interval",
661                 .data           = &init_net.ipv6.sysctl.frags.secret_interval,
662                 .maxlen         = sizeof(int),
663                 .mode           = 0644,
664                 .proc_handler   = &proc_dointvec_jiffies,
665                 .strategy       = &sysctl_jiffies
666         },
667         { }
668 };
669
670 static int ip6_frags_sysctl_register(struct net *net)
671 {
672         struct ctl_table *table;
673         struct ctl_table_header *hdr;
674
675         table = ip6_frags_ctl_table;
676         if (net != &init_net) {
677                 table = kmemdup(table, sizeof(ip6_frags_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                 table[3].mode &= ~0222;
685         }
686
687         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
688         if (hdr == NULL)
689                 goto err_reg;
690
691         net->ipv6.sysctl.frags_hdr = hdr;
692         return 0;
693
694 err_reg:
695         if (net != &init_net)
696                 kfree(table);
697 err_alloc:
698         return -ENOMEM;
699 }
700
701 static void ip6_frags_sysctl_unregister(struct net *net)
702 {
703         struct ctl_table *table;
704
705         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
706         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
707         kfree(table);
708 }
709 #else
710 static inline int ip6_frags_sysctl_register(struct net *net)
711 {
712         return 0;
713 }
714
715 static inline void ip6_frags_sysctl_unregister(struct net *net)
716 {
717 }
718 #endif
719
720 static int ipv6_frags_init_net(struct net *net)
721 {
722         ip6_frags.ctl = &net->ipv6.sysctl.frags;
723
724         net->ipv6.frags.high_thresh = 256 * 1024;
725         net->ipv6.frags.low_thresh = 192 * 1024;
726         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
727         net->ipv6.sysctl.frags.secret_interval = 10 * 60 * HZ;
728
729         inet_frags_init_net(&net->ipv6.frags);
730
731         return ip6_frags_sysctl_register(net);
732 }
733
734 int __init ipv6_frag_init(void)
735 {
736         int ret;
737
738         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
739         if (ret)
740                 goto out;
741
742         ipv6_frags_init_net(&init_net);
743
744         ip6_frags.hashfn = ip6_hashfn;
745         ip6_frags.constructor = ip6_frag_init;
746         ip6_frags.destructor = NULL;
747         ip6_frags.skb_free = NULL;
748         ip6_frags.qsize = sizeof(struct frag_queue);
749         ip6_frags.match = ip6_frag_match;
750         ip6_frags.frag_expire = ip6_frag_expire;
751         inet_frags_init(&ip6_frags);
752 out:
753         return ret;
754 }
755
756 void ipv6_frag_exit(void)
757 {
758         inet_frags_fini(&ip6_frags);
759         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
760 }