ipfrag : frag_kfree_skb() cleanup
[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 #include <linux/slab.h>
45
46 #include <net/sock.h>
47 #include <net/snmp.h>
48
49 #include <net/ipv6.h>
50 #include <net/ip6_route.h>
51 #include <net/protocol.h>
52 #include <net/transp_v6.h>
53 #include <net/rawv6.h>
54 #include <net/ndisc.h>
55 #include <net/addrconf.h>
56 #include <net/inet_frag.h>
57
58 struct ip6frag_skb_cb
59 {
60         struct inet6_skb_parm   h;
61         int                     offset;
62 };
63
64 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
65
66
67 /*
68  *      Equivalent of ipv4 struct ipq
69  */
70
71 struct frag_queue
72 {
73         struct inet_frag_queue  q;
74
75         __be32                  id;             /* fragment id          */
76         u32                     user;
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 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
105                              const struct in6_addr *daddr, u32 rnd)
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 += 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 EXPORT_SYMBOL_GPL(inet6_hash_frag);
131
132 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
133 {
134         struct frag_queue *fq;
135
136         fq = container_of(q, struct frag_queue, q);
137         return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
138 }
139
140 int ip6_frag_match(struct inet_frag_queue *q, void *a)
141 {
142         struct frag_queue *fq;
143         struct ip6_create_arg *arg = a;
144
145         fq = container_of(q, struct frag_queue, q);
146         return (fq->id == arg->id && fq->user == arg->user &&
147                         ipv6_addr_equal(&fq->saddr, arg->src) &&
148                         ipv6_addr_equal(&fq->daddr, arg->dst));
149 }
150 EXPORT_SYMBOL(ip6_frag_match);
151
152 /* Memory Tracking Functions. */
153 static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
154 {
155         atomic_sub(skb->truesize, &nf->mem);
156         kfree_skb(skb);
157 }
158
159 void ip6_frag_init(struct inet_frag_queue *q, void *a)
160 {
161         struct frag_queue *fq = container_of(q, struct frag_queue, q);
162         struct ip6_create_arg *arg = a;
163
164         fq->id = arg->id;
165         fq->user = arg->user;
166         ipv6_addr_copy(&fq->saddr, arg->src);
167         ipv6_addr_copy(&fq->daddr, arg->dst);
168 }
169 EXPORT_SYMBOL(ip6_frag_init);
170
171 /* Destruction primitives. */
172
173 static __inline__ void fq_put(struct frag_queue *fq)
174 {
175         inet_frag_put(&fq->q, &ip6_frags);
176 }
177
178 /* Kill fq entry. It is not destroyed immediately,
179  * because caller (and someone more) holds reference count.
180  */
181 static __inline__ void fq_kill(struct frag_queue *fq)
182 {
183         inet_frag_kill(&fq->q, &ip6_frags);
184 }
185
186 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
187 {
188         int evicted;
189
190         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
191         if (evicted)
192                 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
193 }
194
195 static void ip6_frag_expire(unsigned long data)
196 {
197         struct frag_queue *fq;
198         struct net_device *dev = NULL;
199         struct net *net;
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 & INET_FRAG_COMPLETE)
206                 goto out;
207
208         fq_kill(fq);
209
210         net = container_of(fq->q.net, struct net, ipv6.frags);
211         rcu_read_lock();
212         dev = dev_get_by_index_rcu(net, fq->iif);
213         if (!dev)
214                 goto out_rcu_unlock;
215
216         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
217         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
218
219         /* Don't send error if the first segment did not arrive. */
220         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
221                 goto out_rcu_unlock;
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);
230 out_rcu_unlock:
231         rcu_read_unlock();
232 out:
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 {
240         struct inet_frag_queue *q;
241         struct ip6_create_arg arg;
242         unsigned int hash;
243
244         arg.id = id;
245         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
246         arg.src = src;
247         arg.dst = dst;
248
249         read_lock(&ip6_frags.lock);
250         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
251
252         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
253         if (q == NULL)
254                 return NULL;
255
256         return container_of(q, struct frag_queue, q);
257 }
258
259 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
260                            struct frag_hdr *fhdr, int nhoff)
261 {
262         struct sk_buff *prev, *next;
263         struct net_device *dev;
264         int offset, end;
265         struct net *net = dev_net(skb_dst(skb)->dev);
266
267         if (fq->q.last_in & INET_FRAG_COMPLETE)
268                 goto err;
269
270         offset = ntohs(fhdr->frag_off) & ~0x7;
271         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
272                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
273
274         if ((unsigned int)end > IPV6_MAXPLEN) {
275                 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
276                                  IPSTATS_MIB_INHDRERRORS);
277                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
278                                   ((u8 *)&fhdr->frag_off -
279                                    skb_network_header(skb)));
280                 return -1;
281         }
282
283         if (skb->ip_summed == CHECKSUM_COMPLETE) {
284                 const unsigned char *nh = skb_network_header(skb);
285                 skb->csum = csum_sub(skb->csum,
286                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
287                                                   0));
288         }
289
290         /* Is this the final fragment? */
291         if (!(fhdr->frag_off & htons(IP6_MF))) {
292                 /* If we already have some bits beyond end
293                  * or have different end, the segment is corrupted.
294                  */
295                 if (end < fq->q.len ||
296                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
297                         goto err;
298                 fq->q.last_in |= INET_FRAG_LAST_IN;
299                 fq->q.len = end;
300         } else {
301                 /* Check if the fragment is rounded to 8 bytes.
302                  * Required by the RFC.
303                  */
304                 if (end & 0x7) {
305                         /* RFC2460 says always send parameter problem in
306                          * this case. -DaveM
307                          */
308                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
309                                          IPSTATS_MIB_INHDRERRORS);
310                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
311                                           offsetof(struct ipv6hdr, payload_len));
312                         return -1;
313                 }
314                 if (end > fq->q.len) {
315                         /* Some bits beyond end -> corruption. */
316                         if (fq->q.last_in & INET_FRAG_LAST_IN)
317                                 goto err;
318                         fq->q.len = end;
319                 }
320         }
321
322         if (end == offset)
323                 goto err;
324
325         /* Point into the IP datagram 'data' part. */
326         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
327                 goto err;
328
329         if (pskb_trim_rcsum(skb, end - offset))
330                 goto err;
331
332         /* Find out which fragments are in front and at the back of us
333          * in the chain of fragments so far.  We must know where to put
334          * this fragment, right?
335          */
336         prev = NULL;
337         for(next = fq->q.fragments; next != NULL; next = next->next) {
338                 if (FRAG6_CB(next)->offset >= offset)
339                         break;  /* bingo! */
340                 prev = next;
341         }
342
343         /* We found where to put this one.  Check for overlap with
344          * preceding fragment, and, if needed, align things so that
345          * any overlaps are eliminated.
346          */
347         if (prev) {
348                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
349
350                 if (i > 0) {
351                         offset += i;
352                         if (end <= offset)
353                                 goto err;
354                         if (!pskb_pull(skb, i))
355                                 goto err;
356                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
357                                 skb->ip_summed = CHECKSUM_NONE;
358                 }
359         }
360
361         /* Look for overlap with succeeding segments.
362          * If we can merge fragments, do it.
363          */
364         while (next && FRAG6_CB(next)->offset < end) {
365                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
366
367                 if (i < next->len) {
368                         /* Eat head of the next overlapped fragment
369                          * and leave the loop. The next ones cannot overlap.
370                          */
371                         if (!pskb_pull(next, i))
372                                 goto err;
373                         FRAG6_CB(next)->offset += i;    /* next fragment */
374                         fq->q.meat -= i;
375                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
376                                 next->ip_summed = CHECKSUM_NONE;
377                         break;
378                 } else {
379                         struct sk_buff *free_it = next;
380
381                         /* Old fragment is completely overridden with
382                          * new one drop it.
383                          */
384                         next = next->next;
385
386                         if (prev)
387                                 prev->next = next;
388                         else
389                                 fq->q.fragments = next;
390
391                         fq->q.meat -= free_it->len;
392                         frag_kfree_skb(fq->q.net, free_it);
393                 }
394         }
395
396         FRAG6_CB(skb)->offset = offset;
397
398         /* Insert this fragment in the chain of fragments. */
399         skb->next = next;
400         if (prev)
401                 prev->next = skb;
402         else
403                 fq->q.fragments = skb;
404
405         dev = skb->dev;
406         if (dev) {
407                 fq->iif = dev->ifindex;
408                 skb->dev = NULL;
409         }
410         fq->q.stamp = skb->tstamp;
411         fq->q.meat += skb->len;
412         atomic_add(skb->truesize, &fq->q.net->mem);
413
414         /* The first fragment.
415          * nhoffset is obtained from the first fragment, of course.
416          */
417         if (offset == 0) {
418                 fq->nhoffset = nhoff;
419                 fq->q.last_in |= INET_FRAG_FIRST_IN;
420         }
421
422         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
423             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, &fq->q.net->lru_list);
428         write_unlock(&ip6_frags.lock);
429         return -1;
430
431 err:
432         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
433                       IPSTATS_MIB_REASMFAILS);
434         kfree_skb(skb);
435         return -1;
436 }
437
438 /*
439  *      Check if this packet is complete.
440  *      Returns NULL on failure by any reason, and pointer
441  *      to current nexthdr field in reassembled frame.
442  *
443  *      It is called with locked fq, and caller must check that
444  *      queue is eligible for reassembly i.e. it is not COMPLETE,
445  *      the last and the first frames arrived and all the bits are here.
446  */
447 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
448                           struct net_device *dev)
449 {
450         struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
451         struct sk_buff *fp, *head = fq->q.fragments;
452         int    payload_len;
453         unsigned int nhoff;
454
455         fq_kill(fq);
456
457         /* Make the one we just received the head. */
458         if (prev) {
459                 head = prev->next;
460                 fp = skb_clone(head, GFP_ATOMIC);
461
462                 if (!fp)
463                         goto out_oom;
464
465                 fp->next = head->next;
466                 prev->next = fp;
467
468                 skb_morph(head, fq->q.fragments);
469                 head->next = fq->q.fragments->next;
470
471                 kfree_skb(fq->q.fragments);
472                 fq->q.fragments = head;
473         }
474
475         WARN_ON(head == NULL);
476         WARN_ON(FRAG6_CB(head)->offset != 0);
477
478         /* Unfragmented part is taken from the first segment. */
479         payload_len = ((head->data - skb_network_header(head)) -
480                        sizeof(struct ipv6hdr) + fq->q.len -
481                        sizeof(struct frag_hdr));
482         if (payload_len > IPV6_MAXPLEN)
483                 goto out_oversize;
484
485         /* Head of list must not be cloned. */
486         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
487                 goto out_oom;
488
489         /* If the first fragment is fragmented itself, we split
490          * it to two chunks: the first with data and paged part
491          * and the second, holding only fragments. */
492         if (skb_has_frags(head)) {
493                 struct sk_buff *clone;
494                 int i, plen = 0;
495
496                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
497                         goto out_oom;
498                 clone->next = head->next;
499                 head->next = clone;
500                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
501                 skb_frag_list_init(head);
502                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
503                         plen += skb_shinfo(head)->frags[i].size;
504                 clone->len = clone->data_len = head->data_len - plen;
505                 head->data_len -= clone->len;
506                 head->len -= clone->len;
507                 clone->csum = 0;
508                 clone->ip_summed = head->ip_summed;
509                 atomic_add(clone->truesize, &fq->q.net->mem);
510         }
511
512         /* We have to remove fragment header from datagram and to relocate
513          * header in order to calculate ICV correctly. */
514         nhoff = fq->nhoffset;
515         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
516         memmove(head->head + sizeof(struct frag_hdr), head->head,
517                 (head->data - head->head) - sizeof(struct frag_hdr));
518         head->mac_header += sizeof(struct frag_hdr);
519         head->network_header += sizeof(struct frag_hdr);
520
521         skb_shinfo(head)->frag_list = head->next;
522         skb_reset_transport_header(head);
523         skb_push(head, head->data - skb_network_header(head));
524
525         for (fp=head->next; fp; fp = fp->next) {
526                 head->data_len += fp->len;
527                 head->len += fp->len;
528                 if (head->ip_summed != fp->ip_summed)
529                         head->ip_summed = CHECKSUM_NONE;
530                 else if (head->ip_summed == CHECKSUM_COMPLETE)
531                         head->csum = csum_add(head->csum, fp->csum);
532                 head->truesize += fp->truesize;
533         }
534         atomic_sub(head->truesize, &fq->q.net->mem);
535
536         head->next = NULL;
537         head->dev = dev;
538         head->tstamp = fq->q.stamp;
539         ipv6_hdr(head)->payload_len = htons(payload_len);
540         IP6CB(head)->nhoff = nhoff;
541
542         /* Yes, and fold redundant checksum back. 8) */
543         if (head->ip_summed == CHECKSUM_COMPLETE)
544                 head->csum = csum_partial(skb_network_header(head),
545                                           skb_network_header_len(head),
546                                           head->csum);
547
548         rcu_read_lock();
549         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
550         rcu_read_unlock();
551         fq->q.fragments = NULL;
552         return 1;
553
554 out_oversize:
555         if (net_ratelimit())
556                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
557         goto out_fail;
558 out_oom:
559         if (net_ratelimit())
560                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
561 out_fail:
562         rcu_read_lock();
563         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
564         rcu_read_unlock();
565         return -1;
566 }
567
568 static int ipv6_frag_rcv(struct sk_buff *skb)
569 {
570         struct frag_hdr *fhdr;
571         struct frag_queue *fq;
572         struct ipv6hdr *hdr = ipv6_hdr(skb);
573         struct net *net = dev_net(skb_dst(skb)->dev);
574
575         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
576
577         /* Jumbo payload inhibits frag. header */
578         if (hdr->payload_len==0)
579                 goto fail_hdr;
580
581         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
582                                  sizeof(struct frag_hdr))))
583                 goto fail_hdr;
584
585         hdr = ipv6_hdr(skb);
586         fhdr = (struct frag_hdr *)skb_transport_header(skb);
587
588         if (!(fhdr->frag_off & htons(0xFFF9))) {
589                 /* It is not a fragmented frame */
590                 skb->transport_header += sizeof(struct frag_hdr);
591                 IP6_INC_STATS_BH(net,
592                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
593
594                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
595                 return 1;
596         }
597
598         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
599                 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
600
601         fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
602         if (fq != NULL) {
603                 int ret;
604
605                 spin_lock(&fq->q.lock);
606
607                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
608
609                 spin_unlock(&fq->q.lock);
610                 fq_put(fq);
611                 return ret;
612         }
613
614         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
615         kfree_skb(skb);
616         return -1;
617
618 fail_hdr:
619         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
620         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
621         return -1;
622 }
623
624 static const struct inet6_protocol frag_protocol =
625 {
626         .handler        =       ipv6_frag_rcv,
627         .flags          =       INET6_PROTO_NOPOLICY,
628 };
629
630 #ifdef CONFIG_SYSCTL
631 static struct ctl_table ip6_frags_ns_ctl_table[] = {
632         {
633                 .procname       = "ip6frag_high_thresh",
634                 .data           = &init_net.ipv6.frags.high_thresh,
635                 .maxlen         = sizeof(int),
636                 .mode           = 0644,
637                 .proc_handler   = proc_dointvec
638         },
639         {
640                 .procname       = "ip6frag_low_thresh",
641                 .data           = &init_net.ipv6.frags.low_thresh,
642                 .maxlen         = sizeof(int),
643                 .mode           = 0644,
644                 .proc_handler   = proc_dointvec
645         },
646         {
647                 .procname       = "ip6frag_time",
648                 .data           = &init_net.ipv6.frags.timeout,
649                 .maxlen         = sizeof(int),
650                 .mode           = 0644,
651                 .proc_handler   = proc_dointvec_jiffies,
652         },
653         { }
654 };
655
656 static struct ctl_table ip6_frags_ctl_table[] = {
657         {
658                 .procname       = "ip6frag_secret_interval",
659                 .data           = &ip6_frags.secret_interval,
660                 .maxlen         = sizeof(int),
661                 .mode           = 0644,
662                 .proc_handler   = proc_dointvec_jiffies,
663         },
664         { }
665 };
666
667 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
668 {
669         struct ctl_table *table;
670         struct ctl_table_header *hdr;
671
672         table = ip6_frags_ns_ctl_table;
673         if (!net_eq(net, &init_net)) {
674                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
675                 if (table == NULL)
676                         goto err_alloc;
677
678                 table[0].data = &net->ipv6.frags.high_thresh;
679                 table[1].data = &net->ipv6.frags.low_thresh;
680                 table[2].data = &net->ipv6.frags.timeout;
681         }
682
683         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
684         if (hdr == NULL)
685                 goto err_reg;
686
687         net->ipv6.sysctl.frags_hdr = hdr;
688         return 0;
689
690 err_reg:
691         if (!net_eq(net, &init_net))
692                 kfree(table);
693 err_alloc:
694         return -ENOMEM;
695 }
696
697 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
698 {
699         struct ctl_table *table;
700
701         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
702         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
703         if (!net_eq(net, &init_net))
704                 kfree(table);
705 }
706
707 static struct ctl_table_header *ip6_ctl_header;
708
709 static int ip6_frags_sysctl_register(void)
710 {
711         ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
712                         ip6_frags_ctl_table);
713         return ip6_ctl_header == NULL ? -ENOMEM : 0;
714 }
715
716 static void ip6_frags_sysctl_unregister(void)
717 {
718         unregister_net_sysctl_table(ip6_ctl_header);
719 }
720 #else
721 static inline int ip6_frags_ns_sysctl_register(struct net *net)
722 {
723         return 0;
724 }
725
726 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
727 {
728 }
729
730 static inline int ip6_frags_sysctl_register(void)
731 {
732         return 0;
733 }
734
735 static inline void ip6_frags_sysctl_unregister(void)
736 {
737 }
738 #endif
739
740 static int __net_init ipv6_frags_init_net(struct net *net)
741 {
742         net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
743         net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
744         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
745
746         inet_frags_init_net(&net->ipv6.frags);
747
748         return ip6_frags_ns_sysctl_register(net);
749 }
750
751 static void __net_exit ipv6_frags_exit_net(struct net *net)
752 {
753         ip6_frags_ns_sysctl_unregister(net);
754         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
755 }
756
757 static struct pernet_operations ip6_frags_ops = {
758         .init = ipv6_frags_init_net,
759         .exit = ipv6_frags_exit_net,
760 };
761
762 int __init ipv6_frag_init(void)
763 {
764         int ret;
765
766         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
767         if (ret)
768                 goto out;
769
770         ret = ip6_frags_sysctl_register();
771         if (ret)
772                 goto err_sysctl;
773
774         ret = register_pernet_subsys(&ip6_frags_ops);
775         if (ret)
776                 goto err_pernet;
777
778         ip6_frags.hashfn = ip6_hashfn;
779         ip6_frags.constructor = ip6_frag_init;
780         ip6_frags.destructor = NULL;
781         ip6_frags.skb_free = NULL;
782         ip6_frags.qsize = sizeof(struct frag_queue);
783         ip6_frags.match = ip6_frag_match;
784         ip6_frags.frag_expire = ip6_frag_expire;
785         ip6_frags.secret_interval = 10 * 60 * HZ;
786         inet_frags_init(&ip6_frags);
787 out:
788         return ret;
789
790 err_pernet:
791         ip6_frags_sysctl_unregister();
792 err_sysctl:
793         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
794         goto out;
795 }
796
797 void ipv6_frag_exit(void)
798 {
799         inet_frags_fini(&ip6_frags);
800         ip6_frags_sysctl_unregister();
801         unregister_pernet_subsys(&ip6_frags_ops);
802         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
803 }