Loading...
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
58struct 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
71struct 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
85static struct inet_frags ip6_frags;
86
87int ip6_frag_nqueues(struct net *net)
88{
89 return net->ipv6.frags.nqueues;
90}
91
92int ip6_frag_mem(struct net *net)
93{
94 return atomic_read(&net->ipv6.frags.mem);
95}
96
97static 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 */
104unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
105 const struct in6_addr *daddr, u32 rnd)
106{
107 u32 c;
108
109 c = jhash_3words((__force u32)saddr->s6_addr32[0],
110 (__force u32)saddr->s6_addr32[1],
111 (__force u32)saddr->s6_addr32[2],
112 rnd);
113
114 c = jhash_3words((__force u32)saddr->s6_addr32[3],
115 (__force u32)daddr->s6_addr32[0],
116 (__force u32)daddr->s6_addr32[1],
117 c);
118
119 c = jhash_3words((__force u32)daddr->s6_addr32[2],
120 (__force u32)daddr->s6_addr32[3],
121 (__force u32)id,
122 c);
123
124 return c & (INETFRAGS_HASHSZ - 1);
125}
126EXPORT_SYMBOL_GPL(inet6_hash_frag);
127
128static unsigned int ip6_hashfn(struct inet_frag_queue *q)
129{
130 struct frag_queue *fq;
131
132 fq = container_of(q, struct frag_queue, q);
133 return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
134}
135
136int ip6_frag_match(struct inet_frag_queue *q, void *a)
137{
138 struct frag_queue *fq;
139 struct ip6_create_arg *arg = a;
140
141 fq = container_of(q, struct frag_queue, q);
142 return (fq->id == arg->id && fq->user == arg->user &&
143 ipv6_addr_equal(&fq->saddr, arg->src) &&
144 ipv6_addr_equal(&fq->daddr, arg->dst));
145}
146EXPORT_SYMBOL(ip6_frag_match);
147
148void ip6_frag_init(struct inet_frag_queue *q, void *a)
149{
150 struct frag_queue *fq = container_of(q, struct frag_queue, q);
151 struct ip6_create_arg *arg = a;
152
153 fq->id = arg->id;
154 fq->user = arg->user;
155 ipv6_addr_copy(&fq->saddr, arg->src);
156 ipv6_addr_copy(&fq->daddr, arg->dst);
157}
158EXPORT_SYMBOL(ip6_frag_init);
159
160/* Destruction primitives. */
161
162static __inline__ void fq_put(struct frag_queue *fq)
163{
164 inet_frag_put(&fq->q, &ip6_frags);
165}
166
167/* Kill fq entry. It is not destroyed immediately,
168 * because caller (and someone more) holds reference count.
169 */
170static __inline__ void fq_kill(struct frag_queue *fq)
171{
172 inet_frag_kill(&fq->q, &ip6_frags);
173}
174
175static void ip6_evictor(struct net *net, struct inet6_dev *idev)
176{
177 int evicted;
178
179 evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
180 if (evicted)
181 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
182}
183
184static void ip6_frag_expire(unsigned long data)
185{
186 struct frag_queue *fq;
187 struct net_device *dev = NULL;
188 struct net *net;
189
190 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
191
192 spin_lock(&fq->q.lock);
193
194 if (fq->q.last_in & INET_FRAG_COMPLETE)
195 goto out;
196
197 fq_kill(fq);
198
199 net = container_of(fq->q.net, struct net, ipv6.frags);
200 rcu_read_lock();
201 dev = dev_get_by_index_rcu(net, fq->iif);
202 if (!dev)
203 goto out_rcu_unlock;
204
205 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
206 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
207
208 /* Don't send error if the first segment did not arrive. */
209 if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
210 goto out_rcu_unlock;
211
212 /*
213 But use as source device on which LAST ARRIVED
214 segment was received. And do not use fq->dev
215 pointer directly, device might already disappeared.
216 */
217 fq->q.fragments->dev = dev;
218 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
219out_rcu_unlock:
220 rcu_read_unlock();
221out:
222 spin_unlock(&fq->q.lock);
223 fq_put(fq);
224}
225
226static __inline__ struct frag_queue *
227fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
228{
229 struct inet_frag_queue *q;
230 struct ip6_create_arg arg;
231 unsigned int hash;
232
233 arg.id = id;
234 arg.user = IP6_DEFRAG_LOCAL_DELIVER;
235 arg.src = src;
236 arg.dst = dst;
237
238 read_lock(&ip6_frags.lock);
239 hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
240
241 q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
242 if (q == NULL)
243 return NULL;
244
245 return container_of(q, struct frag_queue, q);
246}
247
248static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
249 struct frag_hdr *fhdr, int nhoff)
250{
251 struct sk_buff *prev, *next;
252 struct net_device *dev;
253 int offset, end;
254 struct net *net = dev_net(skb_dst(skb)->dev);
255
256 if (fq->q.last_in & INET_FRAG_COMPLETE)
257 goto err;
258
259 offset = ntohs(fhdr->frag_off) & ~0x7;
260 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
261 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
262
263 if ((unsigned int)end > IPV6_MAXPLEN) {
264 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
265 IPSTATS_MIB_INHDRERRORS);
266 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
267 ((u8 *)&fhdr->frag_off -
268 skb_network_header(skb)));
269 return -1;
270 }
271
272 if (skb->ip_summed == CHECKSUM_COMPLETE) {
273 const unsigned char *nh = skb_network_header(skb);
274 skb->csum = csum_sub(skb->csum,
275 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
276 0));
277 }
278
279 /* Is this the final fragment? */
280 if (!(fhdr->frag_off & htons(IP6_MF))) {
281 /* If we already have some bits beyond end
282 * or have different end, the segment is corrupted.
283 */
284 if (end < fq->q.len ||
285 ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
286 goto err;
287 fq->q.last_in |= INET_FRAG_LAST_IN;
288 fq->q.len = end;
289 } else {
290 /* Check if the fragment is rounded to 8 bytes.
291 * Required by the RFC.
292 */
293 if (end & 0x7) {
294 /* RFC2460 says always send parameter problem in
295 * this case. -DaveM
296 */
297 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
298 IPSTATS_MIB_INHDRERRORS);
299 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
300 offsetof(struct ipv6hdr, payload_len));
301 return -1;
302 }
303 if (end > fq->q.len) {
304 /* Some bits beyond end -> corruption. */
305 if (fq->q.last_in & INET_FRAG_LAST_IN)
306 goto err;
307 fq->q.len = end;
308 }
309 }
310
311 if (end == offset)
312 goto err;
313
314 /* Point into the IP datagram 'data' part. */
315 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
316 goto err;
317
318 if (pskb_trim_rcsum(skb, end - offset))
319 goto err;
320
321 /* Find out which fragments are in front and at the back of us
322 * in the chain of fragments so far. We must know where to put
323 * this fragment, right?
324 */
325 prev = fq->q.fragments_tail;
326 if (!prev || FRAG6_CB(prev)->offset < offset) {
327 next = NULL;
328 goto found;
329 }
330 prev = NULL;
331 for(next = fq->q.fragments; next != NULL; next = next->next) {
332 if (FRAG6_CB(next)->offset >= offset)
333 break; /* bingo! */
334 prev = next;
335 }
336
337found:
338 /* RFC5722, Section 4:
339 * When reassembling an IPv6 datagram, if
340 * one or more its constituent fragments is determined to be an
341 * overlapping fragment, the entire datagram (and any constituent
342 * fragments, including those not yet received) MUST be silently
343 * discarded.
344 */
345
346 /* Check for overlap with preceding fragment. */
347 if (prev &&
348 (FRAG6_CB(prev)->offset + prev->len) > offset)
349 goto discard_fq;
350
351 /* Look for overlap with succeeding segment. */
352 if (next && FRAG6_CB(next)->offset < end)
353 goto discard_fq;
354
355 FRAG6_CB(skb)->offset = offset;
356
357 /* Insert this fragment in the chain of fragments. */
358 skb->next = next;
359 if (!next)
360 fq->q.fragments_tail = skb;
361 if (prev)
362 prev->next = skb;
363 else
364 fq->q.fragments = skb;
365
366 dev = skb->dev;
367 if (dev) {
368 fq->iif = dev->ifindex;
369 skb->dev = NULL;
370 }
371 fq->q.stamp = skb->tstamp;
372 fq->q.meat += skb->len;
373 atomic_add(skb->truesize, &fq->q.net->mem);
374
375 /* The first fragment.
376 * nhoffset is obtained from the first fragment, of course.
377 */
378 if (offset == 0) {
379 fq->nhoffset = nhoff;
380 fq->q.last_in |= INET_FRAG_FIRST_IN;
381 }
382
383 if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
384 fq->q.meat == fq->q.len)
385 return ip6_frag_reasm(fq, prev, dev);
386
387 write_lock(&ip6_frags.lock);
388 list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
389 write_unlock(&ip6_frags.lock);
390 return -1;
391
392discard_fq:
393 fq_kill(fq);
394err:
395 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
396 IPSTATS_MIB_REASMFAILS);
397 kfree_skb(skb);
398 return -1;
399}
400
401/*
402 * Check if this packet is complete.
403 * Returns NULL on failure by any reason, and pointer
404 * to current nexthdr field in reassembled frame.
405 *
406 * It is called with locked fq, and caller must check that
407 * queue is eligible for reassembly i.e. it is not COMPLETE,
408 * the last and the first frames arrived and all the bits are here.
409 */
410static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
411 struct net_device *dev)
412{
413 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
414 struct sk_buff *fp, *head = fq->q.fragments;
415 int payload_len;
416 unsigned int nhoff;
417
418 fq_kill(fq);
419
420 /* Make the one we just received the head. */
421 if (prev) {
422 head = prev->next;
423 fp = skb_clone(head, GFP_ATOMIC);
424
425 if (!fp)
426 goto out_oom;
427
428 fp->next = head->next;
429 if (!fp->next)
430 fq->q.fragments_tail = fp;
431 prev->next = fp;
432
433 skb_morph(head, fq->q.fragments);
434 head->next = fq->q.fragments->next;
435
436 kfree_skb(fq->q.fragments);
437 fq->q.fragments = head;
438 }
439
440 WARN_ON(head == NULL);
441 WARN_ON(FRAG6_CB(head)->offset != 0);
442
443 /* Unfragmented part is taken from the first segment. */
444 payload_len = ((head->data - skb_network_header(head)) -
445 sizeof(struct ipv6hdr) + fq->q.len -
446 sizeof(struct frag_hdr));
447 if (payload_len > IPV6_MAXPLEN)
448 goto out_oversize;
449
450 /* Head of list must not be cloned. */
451 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
452 goto out_oom;
453
454 /* If the first fragment is fragmented itself, we split
455 * it to two chunks: the first with data and paged part
456 * and the second, holding only fragments. */
457 if (skb_has_frag_list(head)) {
458 struct sk_buff *clone;
459 int i, plen = 0;
460
461 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
462 goto out_oom;
463 clone->next = head->next;
464 head->next = clone;
465 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
466 skb_frag_list_init(head);
467 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
468 plen += skb_shinfo(head)->frags[i].size;
469 clone->len = clone->data_len = head->data_len - plen;
470 head->data_len -= clone->len;
471 head->len -= clone->len;
472 clone->csum = 0;
473 clone->ip_summed = head->ip_summed;
474 atomic_add(clone->truesize, &fq->q.net->mem);
475 }
476
477 /* We have to remove fragment header from datagram and to relocate
478 * header in order to calculate ICV correctly. */
479 nhoff = fq->nhoffset;
480 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
481 memmove(head->head + sizeof(struct frag_hdr), head->head,
482 (head->data - head->head) - sizeof(struct frag_hdr));
483 head->mac_header += sizeof(struct frag_hdr);
484 head->network_header += sizeof(struct frag_hdr);
485
486 skb_shinfo(head)->frag_list = head->next;
487 skb_reset_transport_header(head);
488 skb_push(head, head->data - skb_network_header(head));
489
490 for (fp=head->next; fp; fp = fp->next) {
491 head->data_len += fp->len;
492 head->len += fp->len;
493 if (head->ip_summed != fp->ip_summed)
494 head->ip_summed = CHECKSUM_NONE;
495 else if (head->ip_summed == CHECKSUM_COMPLETE)
496 head->csum = csum_add(head->csum, fp->csum);
497 head->truesize += fp->truesize;
498 }
499 atomic_sub(head->truesize, &fq->q.net->mem);
500
501 head->next = NULL;
502 head->dev = dev;
503 head->tstamp = fq->q.stamp;
504 ipv6_hdr(head)->payload_len = htons(payload_len);
505 IP6CB(head)->nhoff = nhoff;
506
507 /* Yes, and fold redundant checksum back. 8) */
508 if (head->ip_summed == CHECKSUM_COMPLETE)
509 head->csum = csum_partial(skb_network_header(head),
510 skb_network_header_len(head),
511 head->csum);
512
513 rcu_read_lock();
514 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
515 rcu_read_unlock();
516 fq->q.fragments = NULL;
517 fq->q.fragments_tail = NULL;
518 return 1;
519
520out_oversize:
521 if (net_ratelimit())
522 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
523 goto out_fail;
524out_oom:
525 if (net_ratelimit())
526 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
527out_fail:
528 rcu_read_lock();
529 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
530 rcu_read_unlock();
531 return -1;
532}
533
534static int ipv6_frag_rcv(struct sk_buff *skb)
535{
536 struct frag_hdr *fhdr;
537 struct frag_queue *fq;
538 const struct ipv6hdr *hdr = ipv6_hdr(skb);
539 struct net *net = dev_net(skb_dst(skb)->dev);
540
541 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
542
543 /* Jumbo payload inhibits frag. header */
544 if (hdr->payload_len==0)
545 goto fail_hdr;
546
547 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
548 sizeof(struct frag_hdr))))
549 goto fail_hdr;
550
551 hdr = ipv6_hdr(skb);
552 fhdr = (struct frag_hdr *)skb_transport_header(skb);
553
554 if (!(fhdr->frag_off & htons(0xFFF9))) {
555 /* It is not a fragmented frame */
556 skb->transport_header += sizeof(struct frag_hdr);
557 IP6_INC_STATS_BH(net,
558 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
559
560 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
561 return 1;
562 }
563
564 if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
565 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
566
567 fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
568 if (fq != NULL) {
569 int ret;
570
571 spin_lock(&fq->q.lock);
572
573 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
574
575 spin_unlock(&fq->q.lock);
576 fq_put(fq);
577 return ret;
578 }
579
580 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
581 kfree_skb(skb);
582 return -1;
583
584fail_hdr:
585 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
586 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
587 return -1;
588}
589
590static const struct inet6_protocol frag_protocol =
591{
592 .handler = ipv6_frag_rcv,
593 .flags = INET6_PROTO_NOPOLICY,
594};
595
596#ifdef CONFIG_SYSCTL
597static struct ctl_table ip6_frags_ns_ctl_table[] = {
598 {
599 .procname = "ip6frag_high_thresh",
600 .data = &init_net.ipv6.frags.high_thresh,
601 .maxlen = sizeof(int),
602 .mode = 0644,
603 .proc_handler = proc_dointvec
604 },
605 {
606 .procname = "ip6frag_low_thresh",
607 .data = &init_net.ipv6.frags.low_thresh,
608 .maxlen = sizeof(int),
609 .mode = 0644,
610 .proc_handler = proc_dointvec
611 },
612 {
613 .procname = "ip6frag_time",
614 .data = &init_net.ipv6.frags.timeout,
615 .maxlen = sizeof(int),
616 .mode = 0644,
617 .proc_handler = proc_dointvec_jiffies,
618 },
619 { }
620};
621
622static struct ctl_table ip6_frags_ctl_table[] = {
623 {
624 .procname = "ip6frag_secret_interval",
625 .data = &ip6_frags.secret_interval,
626 .maxlen = sizeof(int),
627 .mode = 0644,
628 .proc_handler = proc_dointvec_jiffies,
629 },
630 { }
631};
632
633static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
634{
635 struct ctl_table *table;
636 struct ctl_table_header *hdr;
637
638 table = ip6_frags_ns_ctl_table;
639 if (!net_eq(net, &init_net)) {
640 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
641 if (table == NULL)
642 goto err_alloc;
643
644 table[0].data = &net->ipv6.frags.high_thresh;
645 table[1].data = &net->ipv6.frags.low_thresh;
646 table[2].data = &net->ipv6.frags.timeout;
647 }
648
649 hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
650 if (hdr == NULL)
651 goto err_reg;
652
653 net->ipv6.sysctl.frags_hdr = hdr;
654 return 0;
655
656err_reg:
657 if (!net_eq(net, &init_net))
658 kfree(table);
659err_alloc:
660 return -ENOMEM;
661}
662
663static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
664{
665 struct ctl_table *table;
666
667 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
668 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
669 if (!net_eq(net, &init_net))
670 kfree(table);
671}
672
673static struct ctl_table_header *ip6_ctl_header;
674
675static int ip6_frags_sysctl_register(void)
676{
677 ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
678 ip6_frags_ctl_table);
679 return ip6_ctl_header == NULL ? -ENOMEM : 0;
680}
681
682static void ip6_frags_sysctl_unregister(void)
683{
684 unregister_net_sysctl_table(ip6_ctl_header);
685}
686#else
687static inline int ip6_frags_ns_sysctl_register(struct net *net)
688{
689 return 0;
690}
691
692static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
693{
694}
695
696static inline int ip6_frags_sysctl_register(void)
697{
698 return 0;
699}
700
701static inline void ip6_frags_sysctl_unregister(void)
702{
703}
704#endif
705
706static int __net_init ipv6_frags_init_net(struct net *net)
707{
708 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
709 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
710 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
711
712 inet_frags_init_net(&net->ipv6.frags);
713
714 return ip6_frags_ns_sysctl_register(net);
715}
716
717static void __net_exit ipv6_frags_exit_net(struct net *net)
718{
719 ip6_frags_ns_sysctl_unregister(net);
720 inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
721}
722
723static struct pernet_operations ip6_frags_ops = {
724 .init = ipv6_frags_init_net,
725 .exit = ipv6_frags_exit_net,
726};
727
728int __init ipv6_frag_init(void)
729{
730 int ret;
731
732 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
733 if (ret)
734 goto out;
735
736 ret = ip6_frags_sysctl_register();
737 if (ret)
738 goto err_sysctl;
739
740 ret = register_pernet_subsys(&ip6_frags_ops);
741 if (ret)
742 goto err_pernet;
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 ip6_frags.secret_interval = 10 * 60 * HZ;
752 inet_frags_init(&ip6_frags);
753out:
754 return ret;
755
756err_pernet:
757 ip6_frags_sysctl_unregister();
758err_sysctl:
759 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
760 goto out;
761}
762
763void ipv6_frag_exit(void)
764{
765 inet_frags_fini(&ip6_frags);
766 ip6_frags_sysctl_unregister();
767 unregister_pernet_subsys(&ip6_frags_ops);
768 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
769}
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#include <linux/export.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
59struct 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
72struct frag_queue
73{
74 struct inet_frag_queue q;
75
76 __be32 id; /* fragment id */
77 u32 user;
78 struct in6_addr saddr;
79 struct in6_addr daddr;
80
81 int iif;
82 unsigned int csum;
83 __u16 nhoffset;
84};
85
86static struct inet_frags ip6_frags;
87
88int ip6_frag_nqueues(struct net *net)
89{
90 return net->ipv6.frags.nqueues;
91}
92
93int ip6_frag_mem(struct net *net)
94{
95 return atomic_read(&net->ipv6.frags.mem);
96}
97
98static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
99 struct net_device *dev);
100
101/*
102 * callers should be careful not to use the hash value outside the ipfrag_lock
103 * as doing so could race with ipfrag_hash_rnd being recalculated.
104 */
105unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
106 const struct in6_addr *daddr, u32 rnd)
107{
108 u32 c;
109
110 c = jhash_3words((__force u32)saddr->s6_addr32[0],
111 (__force u32)saddr->s6_addr32[1],
112 (__force u32)saddr->s6_addr32[2],
113 rnd);
114
115 c = jhash_3words((__force u32)saddr->s6_addr32[3],
116 (__force u32)daddr->s6_addr32[0],
117 (__force u32)daddr->s6_addr32[1],
118 c);
119
120 c = jhash_3words((__force u32)daddr->s6_addr32[2],
121 (__force u32)daddr->s6_addr32[3],
122 (__force u32)id,
123 c);
124
125 return c & (INETFRAGS_HASHSZ - 1);
126}
127EXPORT_SYMBOL_GPL(inet6_hash_frag);
128
129static unsigned int ip6_hashfn(struct inet_frag_queue *q)
130{
131 struct frag_queue *fq;
132
133 fq = container_of(q, struct frag_queue, q);
134 return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
135}
136
137bool ip6_frag_match(struct inet_frag_queue *q, void *a)
138{
139 struct frag_queue *fq;
140 struct ip6_create_arg *arg = a;
141
142 fq = container_of(q, struct frag_queue, q);
143 return fq->id == arg->id &&
144 fq->user == arg->user &&
145 ipv6_addr_equal(&fq->saddr, arg->src) &&
146 ipv6_addr_equal(&fq->daddr, arg->dst);
147}
148EXPORT_SYMBOL(ip6_frag_match);
149
150void ip6_frag_init(struct inet_frag_queue *q, void *a)
151{
152 struct frag_queue *fq = container_of(q, struct frag_queue, q);
153 struct ip6_create_arg *arg = a;
154
155 fq->id = arg->id;
156 fq->user = arg->user;
157 fq->saddr = *arg->src;
158 fq->daddr = *arg->dst;
159}
160EXPORT_SYMBOL(ip6_frag_init);
161
162/* Destruction primitives. */
163
164static __inline__ void fq_put(struct frag_queue *fq)
165{
166 inet_frag_put(&fq->q, &ip6_frags);
167}
168
169/* Kill fq entry. It is not destroyed immediately,
170 * because caller (and someone more) holds reference count.
171 */
172static __inline__ void fq_kill(struct frag_queue *fq)
173{
174 inet_frag_kill(&fq->q, &ip6_frags);
175}
176
177static void ip6_evictor(struct net *net, struct inet6_dev *idev)
178{
179 int evicted;
180
181 evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
182 if (evicted)
183 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
184}
185
186static void ip6_frag_expire(unsigned long data)
187{
188 struct frag_queue *fq;
189 struct net_device *dev = NULL;
190 struct net *net;
191
192 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
193
194 spin_lock(&fq->q.lock);
195
196 if (fq->q.last_in & INET_FRAG_COMPLETE)
197 goto out;
198
199 fq_kill(fq);
200
201 net = container_of(fq->q.net, struct net, ipv6.frags);
202 rcu_read_lock();
203 dev = dev_get_by_index_rcu(net, fq->iif);
204 if (!dev)
205 goto out_rcu_unlock;
206
207 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
208 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
209
210 /* Don't send error if the first segment did not arrive. */
211 if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
212 goto out_rcu_unlock;
213
214 /*
215 But use as source device on which LAST ARRIVED
216 segment was received. And do not use fq->dev
217 pointer directly, device might already disappeared.
218 */
219 fq->q.fragments->dev = dev;
220 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
221out_rcu_unlock:
222 rcu_read_unlock();
223out:
224 spin_unlock(&fq->q.lock);
225 fq_put(fq);
226}
227
228static __inline__ struct frag_queue *
229fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
230{
231 struct inet_frag_queue *q;
232 struct ip6_create_arg arg;
233 unsigned int hash;
234
235 arg.id = id;
236 arg.user = IP6_DEFRAG_LOCAL_DELIVER;
237 arg.src = src;
238 arg.dst = dst;
239
240 read_lock(&ip6_frags.lock);
241 hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
242
243 q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
244 if (q == NULL)
245 return NULL;
246
247 return container_of(q, struct frag_queue, q);
248}
249
250static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
251 struct frag_hdr *fhdr, int nhoff)
252{
253 struct sk_buff *prev, *next;
254 struct net_device *dev;
255 int offset, end;
256 struct net *net = dev_net(skb_dst(skb)->dev);
257
258 if (fq->q.last_in & INET_FRAG_COMPLETE)
259 goto err;
260
261 offset = ntohs(fhdr->frag_off) & ~0x7;
262 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
263 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
264
265 if ((unsigned int)end > IPV6_MAXPLEN) {
266 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
267 IPSTATS_MIB_INHDRERRORS);
268 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
269 ((u8 *)&fhdr->frag_off -
270 skb_network_header(skb)));
271 return -1;
272 }
273
274 if (skb->ip_summed == CHECKSUM_COMPLETE) {
275 const unsigned char *nh = skb_network_header(skb);
276 skb->csum = csum_sub(skb->csum,
277 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
278 0));
279 }
280
281 /* Is this the final fragment? */
282 if (!(fhdr->frag_off & htons(IP6_MF))) {
283 /* If we already have some bits beyond end
284 * or have different end, the segment is corrupted.
285 */
286 if (end < fq->q.len ||
287 ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
288 goto err;
289 fq->q.last_in |= INET_FRAG_LAST_IN;
290 fq->q.len = end;
291 } else {
292 /* Check if the fragment is rounded to 8 bytes.
293 * Required by the RFC.
294 */
295 if (end & 0x7) {
296 /* RFC2460 says always send parameter problem in
297 * this case. -DaveM
298 */
299 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
300 IPSTATS_MIB_INHDRERRORS);
301 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
302 offsetof(struct ipv6hdr, payload_len));
303 return -1;
304 }
305 if (end > fq->q.len) {
306 /* Some bits beyond end -> corruption. */
307 if (fq->q.last_in & INET_FRAG_LAST_IN)
308 goto err;
309 fq->q.len = end;
310 }
311 }
312
313 if (end == offset)
314 goto err;
315
316 /* Point into the IP datagram 'data' part. */
317 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
318 goto err;
319
320 if (pskb_trim_rcsum(skb, end - offset))
321 goto err;
322
323 /* Find out which fragments are in front and at the back of us
324 * in the chain of fragments so far. We must know where to put
325 * this fragment, right?
326 */
327 prev = fq->q.fragments_tail;
328 if (!prev || FRAG6_CB(prev)->offset < offset) {
329 next = NULL;
330 goto found;
331 }
332 prev = NULL;
333 for(next = fq->q.fragments; next != NULL; next = next->next) {
334 if (FRAG6_CB(next)->offset >= offset)
335 break; /* bingo! */
336 prev = next;
337 }
338
339found:
340 /* RFC5722, Section 4, amended by Errata ID : 3089
341 * When reassembling an IPv6 datagram, if
342 * one or more its constituent fragments is determined to be an
343 * overlapping fragment, the entire datagram (and any constituent
344 * fragments) MUST be silently discarded.
345 */
346
347 /* Check for overlap with preceding fragment. */
348 if (prev &&
349 (FRAG6_CB(prev)->offset + prev->len) > offset)
350 goto discard_fq;
351
352 /* Look for overlap with succeeding segment. */
353 if (next && FRAG6_CB(next)->offset < end)
354 goto discard_fq;
355
356 FRAG6_CB(skb)->offset = offset;
357
358 /* Insert this fragment in the chain of fragments. */
359 skb->next = next;
360 if (!next)
361 fq->q.fragments_tail = skb;
362 if (prev)
363 prev->next = skb;
364 else
365 fq->q.fragments = skb;
366
367 dev = skb->dev;
368 if (dev) {
369 fq->iif = dev->ifindex;
370 skb->dev = NULL;
371 }
372 fq->q.stamp = skb->tstamp;
373 fq->q.meat += skb->len;
374 atomic_add(skb->truesize, &fq->q.net->mem);
375
376 /* The first fragment.
377 * nhoffset is obtained from the first fragment, of course.
378 */
379 if (offset == 0) {
380 fq->nhoffset = nhoff;
381 fq->q.last_in |= INET_FRAG_FIRST_IN;
382 }
383
384 if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
385 fq->q.meat == fq->q.len)
386 return ip6_frag_reasm(fq, prev, dev);
387
388 write_lock(&ip6_frags.lock);
389 list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
390 write_unlock(&ip6_frags.lock);
391 return -1;
392
393discard_fq:
394 fq_kill(fq);
395err:
396 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
397 IPSTATS_MIB_REASMFAILS);
398 kfree_skb(skb);
399 return -1;
400}
401
402/*
403 * Check if this packet is complete.
404 * Returns NULL on failure by any reason, and pointer
405 * to current nexthdr field in reassembled frame.
406 *
407 * It is called with locked fq, and caller must check that
408 * queue is eligible for reassembly i.e. it is not COMPLETE,
409 * the last and the first frames arrived and all the bits are here.
410 */
411static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
412 struct net_device *dev)
413{
414 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
415 struct sk_buff *fp, *head = fq->q.fragments;
416 int payload_len;
417 unsigned int nhoff;
418 int sum_truesize;
419
420 fq_kill(fq);
421
422 /* Make the one we just received the head. */
423 if (prev) {
424 head = prev->next;
425 fp = skb_clone(head, GFP_ATOMIC);
426
427 if (!fp)
428 goto out_oom;
429
430 fp->next = head->next;
431 if (!fp->next)
432 fq->q.fragments_tail = fp;
433 prev->next = fp;
434
435 skb_morph(head, fq->q.fragments);
436 head->next = fq->q.fragments->next;
437
438 consume_skb(fq->q.fragments);
439 fq->q.fragments = head;
440 }
441
442 WARN_ON(head == NULL);
443 WARN_ON(FRAG6_CB(head)->offset != 0);
444
445 /* Unfragmented part is taken from the first segment. */
446 payload_len = ((head->data - skb_network_header(head)) -
447 sizeof(struct ipv6hdr) + fq->q.len -
448 sizeof(struct frag_hdr));
449 if (payload_len > IPV6_MAXPLEN)
450 goto out_oversize;
451
452 /* Head of list must not be cloned. */
453 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
454 goto out_oom;
455
456 /* If the first fragment is fragmented itself, we split
457 * it to two chunks: the first with data and paged part
458 * and the second, holding only fragments. */
459 if (skb_has_frag_list(head)) {
460 struct sk_buff *clone;
461 int i, plen = 0;
462
463 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
464 goto out_oom;
465 clone->next = head->next;
466 head->next = clone;
467 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
468 skb_frag_list_init(head);
469 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
470 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
471 clone->len = clone->data_len = head->data_len - plen;
472 head->data_len -= clone->len;
473 head->len -= clone->len;
474 clone->csum = 0;
475 clone->ip_summed = head->ip_summed;
476 atomic_add(clone->truesize, &fq->q.net->mem);
477 }
478
479 /* We have to remove fragment header from datagram and to relocate
480 * header in order to calculate ICV correctly. */
481 nhoff = fq->nhoffset;
482 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
483 memmove(head->head + sizeof(struct frag_hdr), head->head,
484 (head->data - head->head) - sizeof(struct frag_hdr));
485 head->mac_header += sizeof(struct frag_hdr);
486 head->network_header += sizeof(struct frag_hdr);
487
488 skb_reset_transport_header(head);
489 skb_push(head, head->data - skb_network_header(head));
490
491 sum_truesize = head->truesize;
492 for (fp = head->next; fp;) {
493 bool headstolen;
494 int delta;
495 struct sk_buff *next = fp->next;
496
497 sum_truesize += fp->truesize;
498 if (head->ip_summed != fp->ip_summed)
499 head->ip_summed = CHECKSUM_NONE;
500 else if (head->ip_summed == CHECKSUM_COMPLETE)
501 head->csum = csum_add(head->csum, fp->csum);
502
503 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
504 kfree_skb_partial(fp, headstolen);
505 } else {
506 if (!skb_shinfo(head)->frag_list)
507 skb_shinfo(head)->frag_list = fp;
508 head->data_len += fp->len;
509 head->len += fp->len;
510 head->truesize += fp->truesize;
511 }
512 fp = next;
513 }
514 atomic_sub(sum_truesize, &fq->q.net->mem);
515
516 head->next = NULL;
517 head->dev = dev;
518 head->tstamp = fq->q.stamp;
519 ipv6_hdr(head)->payload_len = htons(payload_len);
520 IP6CB(head)->nhoff = nhoff;
521
522 /* Yes, and fold redundant checksum back. 8) */
523 if (head->ip_summed == CHECKSUM_COMPLETE)
524 head->csum = csum_partial(skb_network_header(head),
525 skb_network_header_len(head),
526 head->csum);
527
528 rcu_read_lock();
529 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
530 rcu_read_unlock();
531 fq->q.fragments = NULL;
532 fq->q.fragments_tail = NULL;
533 return 1;
534
535out_oversize:
536 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
537 goto out_fail;
538out_oom:
539 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
540out_fail:
541 rcu_read_lock();
542 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
543 rcu_read_unlock();
544 return -1;
545}
546
547static int ipv6_frag_rcv(struct sk_buff *skb)
548{
549 struct frag_hdr *fhdr;
550 struct frag_queue *fq;
551 const struct ipv6hdr *hdr = ipv6_hdr(skb);
552 struct net *net = dev_net(skb_dst(skb)->dev);
553
554 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
555
556 /* Jumbo payload inhibits frag. header */
557 if (hdr->payload_len==0)
558 goto fail_hdr;
559
560 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
561 sizeof(struct frag_hdr))))
562 goto fail_hdr;
563
564 hdr = ipv6_hdr(skb);
565 fhdr = (struct frag_hdr *)skb_transport_header(skb);
566
567 if (!(fhdr->frag_off & htons(0xFFF9))) {
568 /* It is not a fragmented frame */
569 skb->transport_header += sizeof(struct frag_hdr);
570 IP6_INC_STATS_BH(net,
571 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
572
573 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
574 return 1;
575 }
576
577 if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
578 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
579
580 fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
581 if (fq != NULL) {
582 int ret;
583
584 spin_lock(&fq->q.lock);
585
586 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
587
588 spin_unlock(&fq->q.lock);
589 fq_put(fq);
590 return ret;
591 }
592
593 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
594 kfree_skb(skb);
595 return -1;
596
597fail_hdr:
598 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
599 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
600 return -1;
601}
602
603static const struct inet6_protocol frag_protocol =
604{
605 .handler = ipv6_frag_rcv,
606 .flags = INET6_PROTO_NOPOLICY,
607};
608
609#ifdef CONFIG_SYSCTL
610static struct ctl_table ip6_frags_ns_ctl_table[] = {
611 {
612 .procname = "ip6frag_high_thresh",
613 .data = &init_net.ipv6.frags.high_thresh,
614 .maxlen = sizeof(int),
615 .mode = 0644,
616 .proc_handler = proc_dointvec
617 },
618 {
619 .procname = "ip6frag_low_thresh",
620 .data = &init_net.ipv6.frags.low_thresh,
621 .maxlen = sizeof(int),
622 .mode = 0644,
623 .proc_handler = proc_dointvec
624 },
625 {
626 .procname = "ip6frag_time",
627 .data = &init_net.ipv6.frags.timeout,
628 .maxlen = sizeof(int),
629 .mode = 0644,
630 .proc_handler = proc_dointvec_jiffies,
631 },
632 { }
633};
634
635static struct ctl_table ip6_frags_ctl_table[] = {
636 {
637 .procname = "ip6frag_secret_interval",
638 .data = &ip6_frags.secret_interval,
639 .maxlen = sizeof(int),
640 .mode = 0644,
641 .proc_handler = proc_dointvec_jiffies,
642 },
643 { }
644};
645
646static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
647{
648 struct ctl_table *table;
649 struct ctl_table_header *hdr;
650
651 table = ip6_frags_ns_ctl_table;
652 if (!net_eq(net, &init_net)) {
653 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
654 if (table == NULL)
655 goto err_alloc;
656
657 table[0].data = &net->ipv6.frags.high_thresh;
658 table[1].data = &net->ipv6.frags.low_thresh;
659 table[2].data = &net->ipv6.frags.timeout;
660 }
661
662 hdr = register_net_sysctl(net, "net/ipv6", table);
663 if (hdr == NULL)
664 goto err_reg;
665
666 net->ipv6.sysctl.frags_hdr = hdr;
667 return 0;
668
669err_reg:
670 if (!net_eq(net, &init_net))
671 kfree(table);
672err_alloc:
673 return -ENOMEM;
674}
675
676static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
677{
678 struct ctl_table *table;
679
680 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
681 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
682 if (!net_eq(net, &init_net))
683 kfree(table);
684}
685
686static struct ctl_table_header *ip6_ctl_header;
687
688static int ip6_frags_sysctl_register(void)
689{
690 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
691 ip6_frags_ctl_table);
692 return ip6_ctl_header == NULL ? -ENOMEM : 0;
693}
694
695static void ip6_frags_sysctl_unregister(void)
696{
697 unregister_net_sysctl_table(ip6_ctl_header);
698}
699#else
700static inline int ip6_frags_ns_sysctl_register(struct net *net)
701{
702 return 0;
703}
704
705static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
706{
707}
708
709static inline int ip6_frags_sysctl_register(void)
710{
711 return 0;
712}
713
714static inline void ip6_frags_sysctl_unregister(void)
715{
716}
717#endif
718
719static int __net_init ipv6_frags_init_net(struct net *net)
720{
721 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
722 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
723 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
724
725 inet_frags_init_net(&net->ipv6.frags);
726
727 return ip6_frags_ns_sysctl_register(net);
728}
729
730static void __net_exit ipv6_frags_exit_net(struct net *net)
731{
732 ip6_frags_ns_sysctl_unregister(net);
733 inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
734}
735
736static struct pernet_operations ip6_frags_ops = {
737 .init = ipv6_frags_init_net,
738 .exit = ipv6_frags_exit_net,
739};
740
741int __init ipv6_frag_init(void)
742{
743 int ret;
744
745 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
746 if (ret)
747 goto out;
748
749 ret = ip6_frags_sysctl_register();
750 if (ret)
751 goto err_sysctl;
752
753 ret = register_pernet_subsys(&ip6_frags_ops);
754 if (ret)
755 goto err_pernet;
756
757 ip6_frags.hashfn = ip6_hashfn;
758 ip6_frags.constructor = ip6_frag_init;
759 ip6_frags.destructor = NULL;
760 ip6_frags.skb_free = NULL;
761 ip6_frags.qsize = sizeof(struct frag_queue);
762 ip6_frags.match = ip6_frag_match;
763 ip6_frags.frag_expire = ip6_frag_expire;
764 ip6_frags.secret_interval = 10 * 60 * HZ;
765 inet_frags_init(&ip6_frags);
766out:
767 return ret;
768
769err_pernet:
770 ip6_frags_sysctl_unregister();
771err_sysctl:
772 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
773 goto out;
774}
775
776void ipv6_frag_exit(void)
777{
778 inet_frags_fini(&ip6_frags);
779 ip6_frags_sysctl_unregister();
780 unregister_pernet_subsys(&ip6_frags_ops);
781 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
782}