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
30#define pr_fmt(fmt) "IPv6: " fmt
31
32#include <linux/errno.h>
33#include <linux/types.h>
34#include <linux/string.h>
35#include <linux/socket.h>
36#include <linux/sockios.h>
37#include <linux/jiffies.h>
38#include <linux/net.h>
39#include <linux/list.h>
40#include <linux/netdevice.h>
41#include <linux/in6.h>
42#include <linux/ipv6.h>
43#include <linux/icmpv6.h>
44#include <linux/random.h>
45#include <linux/jhash.h>
46#include <linux/skbuff.h>
47#include <linux/slab.h>
48#include <linux/export.h>
49
50#include <net/sock.h>
51#include <net/snmp.h>
52
53#include <net/ipv6.h>
54#include <net/ip6_route.h>
55#include <net/protocol.h>
56#include <net/transp_v6.h>
57#include <net/rawv6.h>
58#include <net/ndisc.h>
59#include <net/addrconf.h>
60#include <net/inet_frag.h>
61#include <net/inet_ecn.h>
62
63struct ip6frag_skb_cb
64{
65 struct inet6_skb_parm h;
66 int offset;
67};
68
69#define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
70
71static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
72{
73 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
74}
75
76static struct inet_frags ip6_frags;
77
78static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
79 struct net_device *dev);
80
81/*
82 * callers should be careful not to use the hash value outside the ipfrag_lock
83 * as doing so could race with ipfrag_hash_rnd being recalculated.
84 */
85static unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
86 const struct in6_addr *daddr)
87{
88 u32 c;
89
90 net_get_random_once(&ip6_frags.rnd, sizeof(ip6_frags.rnd));
91 c = jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
92 (__force u32)id, ip6_frags.rnd);
93
94 return c & (INETFRAGS_HASHSZ - 1);
95}
96
97static unsigned int ip6_hashfn(struct inet_frag_queue *q)
98{
99 struct frag_queue *fq;
100
101 fq = container_of(q, struct frag_queue, q);
102 return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr);
103}
104
105bool ip6_frag_match(struct inet_frag_queue *q, void *a)
106{
107 struct frag_queue *fq;
108 struct ip6_create_arg *arg = a;
109
110 fq = container_of(q, struct frag_queue, q);
111 return fq->id == arg->id &&
112 fq->user == arg->user &&
113 ipv6_addr_equal(&fq->saddr, arg->src) &&
114 ipv6_addr_equal(&fq->daddr, arg->dst);
115}
116EXPORT_SYMBOL(ip6_frag_match);
117
118void ip6_frag_init(struct inet_frag_queue *q, void *a)
119{
120 struct frag_queue *fq = container_of(q, struct frag_queue, q);
121 struct ip6_create_arg *arg = a;
122
123 fq->id = arg->id;
124 fq->user = arg->user;
125 fq->saddr = *arg->src;
126 fq->daddr = *arg->dst;
127 fq->ecn = arg->ecn;
128}
129EXPORT_SYMBOL(ip6_frag_init);
130
131void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
132 struct inet_frags *frags)
133{
134 struct net_device *dev = NULL;
135
136 spin_lock(&fq->q.lock);
137
138 if (fq->q.last_in & INET_FRAG_COMPLETE)
139 goto out;
140
141 inet_frag_kill(&fq->q, frags);
142
143 rcu_read_lock();
144 dev = dev_get_by_index_rcu(net, fq->iif);
145 if (!dev)
146 goto out_rcu_unlock;
147
148 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
149 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
150
151 /* Don't send error if the first segment did not arrive. */
152 if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
153 goto out_rcu_unlock;
154
155 /*
156 But use as source device on which LAST ARRIVED
157 segment was received. And do not use fq->dev
158 pointer directly, device might already disappeared.
159 */
160 fq->q.fragments->dev = dev;
161 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
162out_rcu_unlock:
163 rcu_read_unlock();
164out:
165 spin_unlock(&fq->q.lock);
166 inet_frag_put(&fq->q, frags);
167}
168EXPORT_SYMBOL(ip6_expire_frag_queue);
169
170static void ip6_frag_expire(unsigned long data)
171{
172 struct frag_queue *fq;
173 struct net *net;
174
175 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
176 net = container_of(fq->q.net, struct net, ipv6.frags);
177
178 ip6_expire_frag_queue(net, fq, &ip6_frags);
179}
180
181static __inline__ struct frag_queue *
182fq_find(struct net *net, __be32 id, const struct in6_addr *src,
183 const struct in6_addr *dst, u8 ecn)
184{
185 struct inet_frag_queue *q;
186 struct ip6_create_arg arg;
187 unsigned int hash;
188
189 arg.id = id;
190 arg.user = IP6_DEFRAG_LOCAL_DELIVER;
191 arg.src = src;
192 arg.dst = dst;
193 arg.ecn = ecn;
194
195 read_lock(&ip6_frags.lock);
196 hash = inet6_hash_frag(id, src, dst);
197
198 q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
199 if (IS_ERR_OR_NULL(q)) {
200 inet_frag_maybe_warn_overflow(q, pr_fmt());
201 return NULL;
202 }
203 return container_of(q, struct frag_queue, q);
204}
205
206static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
207 struct frag_hdr *fhdr, int nhoff)
208{
209 struct sk_buff *prev, *next;
210 struct net_device *dev;
211 int offset, end;
212 struct net *net = dev_net(skb_dst(skb)->dev);
213 u8 ecn;
214
215 if (fq->q.last_in & INET_FRAG_COMPLETE)
216 goto err;
217
218 offset = ntohs(fhdr->frag_off) & ~0x7;
219 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
220 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
221
222 if ((unsigned int)end > IPV6_MAXPLEN) {
223 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
224 IPSTATS_MIB_INHDRERRORS);
225 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
226 ((u8 *)&fhdr->frag_off -
227 skb_network_header(skb)));
228 return -1;
229 }
230
231 ecn = ip6_frag_ecn(ipv6_hdr(skb));
232
233 if (skb->ip_summed == CHECKSUM_COMPLETE) {
234 const unsigned char *nh = skb_network_header(skb);
235 skb->csum = csum_sub(skb->csum,
236 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
237 0));
238 }
239
240 /* Is this the final fragment? */
241 if (!(fhdr->frag_off & htons(IP6_MF))) {
242 /* If we already have some bits beyond end
243 * or have different end, the segment is corrupted.
244 */
245 if (end < fq->q.len ||
246 ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
247 goto err;
248 fq->q.last_in |= INET_FRAG_LAST_IN;
249 fq->q.len = end;
250 } else {
251 /* Check if the fragment is rounded to 8 bytes.
252 * Required by the RFC.
253 */
254 if (end & 0x7) {
255 /* RFC2460 says always send parameter problem in
256 * this case. -DaveM
257 */
258 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
259 IPSTATS_MIB_INHDRERRORS);
260 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
261 offsetof(struct ipv6hdr, payload_len));
262 return -1;
263 }
264 if (end > fq->q.len) {
265 /* Some bits beyond end -> corruption. */
266 if (fq->q.last_in & INET_FRAG_LAST_IN)
267 goto err;
268 fq->q.len = end;
269 }
270 }
271
272 if (end == offset)
273 goto err;
274
275 /* Point into the IP datagram 'data' part. */
276 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
277 goto err;
278
279 if (pskb_trim_rcsum(skb, end - offset))
280 goto err;
281
282 /* Find out which fragments are in front and at the back of us
283 * in the chain of fragments so far. We must know where to put
284 * this fragment, right?
285 */
286 prev = fq->q.fragments_tail;
287 if (!prev || FRAG6_CB(prev)->offset < offset) {
288 next = NULL;
289 goto found;
290 }
291 prev = NULL;
292 for(next = fq->q.fragments; next != NULL; next = next->next) {
293 if (FRAG6_CB(next)->offset >= offset)
294 break; /* bingo! */
295 prev = next;
296 }
297
298found:
299 /* RFC5722, Section 4, amended by Errata ID : 3089
300 * When reassembling an IPv6 datagram, if
301 * one or more its constituent fragments is determined to be an
302 * overlapping fragment, the entire datagram (and any constituent
303 * fragments) MUST be silently discarded.
304 */
305
306 /* Check for overlap with preceding fragment. */
307 if (prev &&
308 (FRAG6_CB(prev)->offset + prev->len) > offset)
309 goto discard_fq;
310
311 /* Look for overlap with succeeding segment. */
312 if (next && FRAG6_CB(next)->offset < end)
313 goto discard_fq;
314
315 FRAG6_CB(skb)->offset = offset;
316
317 /* Insert this fragment in the chain of fragments. */
318 skb->next = next;
319 if (!next)
320 fq->q.fragments_tail = skb;
321 if (prev)
322 prev->next = skb;
323 else
324 fq->q.fragments = skb;
325
326 dev = skb->dev;
327 if (dev) {
328 fq->iif = dev->ifindex;
329 skb->dev = NULL;
330 }
331 fq->q.stamp = skb->tstamp;
332 fq->q.meat += skb->len;
333 fq->ecn |= ecn;
334 add_frag_mem_limit(&fq->q, skb->truesize);
335
336 /* The first fragment.
337 * nhoffset is obtained from the first fragment, of course.
338 */
339 if (offset == 0) {
340 fq->nhoffset = nhoff;
341 fq->q.last_in |= INET_FRAG_FIRST_IN;
342 }
343
344 if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
345 fq->q.meat == fq->q.len) {
346 int res;
347 unsigned long orefdst = skb->_skb_refdst;
348
349 skb->_skb_refdst = 0UL;
350 res = ip6_frag_reasm(fq, prev, dev);
351 skb->_skb_refdst = orefdst;
352 return res;
353 }
354
355 skb_dst_drop(skb);
356 inet_frag_lru_move(&fq->q);
357 return -1;
358
359discard_fq:
360 inet_frag_kill(&fq->q, &ip6_frags);
361err:
362 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
363 IPSTATS_MIB_REASMFAILS);
364 kfree_skb(skb);
365 return -1;
366}
367
368/*
369 * Check if this packet is complete.
370 * Returns NULL on failure by any reason, and pointer
371 * to current nexthdr field in reassembled frame.
372 *
373 * It is called with locked fq, and caller must check that
374 * queue is eligible for reassembly i.e. it is not COMPLETE,
375 * the last and the first frames arrived and all the bits are here.
376 */
377static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
378 struct net_device *dev)
379{
380 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
381 struct sk_buff *fp, *head = fq->q.fragments;
382 int payload_len;
383 unsigned int nhoff;
384 int sum_truesize;
385 u8 ecn;
386
387 inet_frag_kill(&fq->q, &ip6_frags);
388
389 ecn = ip_frag_ecn_table[fq->ecn];
390 if (unlikely(ecn == 0xff))
391 goto out_fail;
392
393 /* Make the one we just received the head. */
394 if (prev) {
395 head = prev->next;
396 fp = skb_clone(head, GFP_ATOMIC);
397
398 if (!fp)
399 goto out_oom;
400
401 fp->next = head->next;
402 if (!fp->next)
403 fq->q.fragments_tail = fp;
404 prev->next = fp;
405
406 skb_morph(head, fq->q.fragments);
407 head->next = fq->q.fragments->next;
408
409 consume_skb(fq->q.fragments);
410 fq->q.fragments = head;
411 }
412
413 WARN_ON(head == NULL);
414 WARN_ON(FRAG6_CB(head)->offset != 0);
415
416 /* Unfragmented part is taken from the first segment. */
417 payload_len = ((head->data - skb_network_header(head)) -
418 sizeof(struct ipv6hdr) + fq->q.len -
419 sizeof(struct frag_hdr));
420 if (payload_len > IPV6_MAXPLEN)
421 goto out_oversize;
422
423 /* Head of list must not be cloned. */
424 if (skb_unclone(head, GFP_ATOMIC))
425 goto out_oom;
426
427 /* If the first fragment is fragmented itself, we split
428 * it to two chunks: the first with data and paged part
429 * and the second, holding only fragments. */
430 if (skb_has_frag_list(head)) {
431 struct sk_buff *clone;
432 int i, plen = 0;
433
434 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
435 goto out_oom;
436 clone->next = head->next;
437 head->next = clone;
438 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
439 skb_frag_list_init(head);
440 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
441 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
442 clone->len = clone->data_len = head->data_len - plen;
443 head->data_len -= clone->len;
444 head->len -= clone->len;
445 clone->csum = 0;
446 clone->ip_summed = head->ip_summed;
447 add_frag_mem_limit(&fq->q, clone->truesize);
448 }
449
450 /* We have to remove fragment header from datagram and to relocate
451 * header in order to calculate ICV correctly. */
452 nhoff = fq->nhoffset;
453 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
454 memmove(head->head + sizeof(struct frag_hdr), head->head,
455 (head->data - head->head) - sizeof(struct frag_hdr));
456 head->mac_header += sizeof(struct frag_hdr);
457 head->network_header += sizeof(struct frag_hdr);
458
459 skb_reset_transport_header(head);
460 skb_push(head, head->data - skb_network_header(head));
461
462 sum_truesize = head->truesize;
463 for (fp = head->next; fp;) {
464 bool headstolen;
465 int delta;
466 struct sk_buff *next = fp->next;
467
468 sum_truesize += fp->truesize;
469 if (head->ip_summed != fp->ip_summed)
470 head->ip_summed = CHECKSUM_NONE;
471 else if (head->ip_summed == CHECKSUM_COMPLETE)
472 head->csum = csum_add(head->csum, fp->csum);
473
474 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
475 kfree_skb_partial(fp, headstolen);
476 } else {
477 if (!skb_shinfo(head)->frag_list)
478 skb_shinfo(head)->frag_list = fp;
479 head->data_len += fp->len;
480 head->len += fp->len;
481 head->truesize += fp->truesize;
482 }
483 fp = next;
484 }
485 sub_frag_mem_limit(&fq->q, sum_truesize);
486
487 head->next = NULL;
488 head->dev = dev;
489 head->tstamp = fq->q.stamp;
490 ipv6_hdr(head)->payload_len = htons(payload_len);
491 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
492 IP6CB(head)->nhoff = nhoff;
493 IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
494
495 /* Yes, and fold redundant checksum back. 8) */
496 if (head->ip_summed == CHECKSUM_COMPLETE)
497 head->csum = csum_partial(skb_network_header(head),
498 skb_network_header_len(head),
499 head->csum);
500
501 rcu_read_lock();
502 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
503 rcu_read_unlock();
504 fq->q.fragments = NULL;
505 fq->q.fragments_tail = NULL;
506 return 1;
507
508out_oversize:
509 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
510 goto out_fail;
511out_oom:
512 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
513out_fail:
514 rcu_read_lock();
515 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
516 rcu_read_unlock();
517 return -1;
518}
519
520static int ipv6_frag_rcv(struct sk_buff *skb)
521{
522 struct frag_hdr *fhdr;
523 struct frag_queue *fq;
524 const struct ipv6hdr *hdr = ipv6_hdr(skb);
525 struct net *net = dev_net(skb_dst(skb)->dev);
526 int evicted;
527
528 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
529 goto fail_hdr;
530
531 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
532
533 /* Jumbo payload inhibits frag. header */
534 if (hdr->payload_len==0)
535 goto fail_hdr;
536
537 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
538 sizeof(struct frag_hdr))))
539 goto fail_hdr;
540
541 hdr = ipv6_hdr(skb);
542 fhdr = (struct frag_hdr *)skb_transport_header(skb);
543
544 if (!(fhdr->frag_off & htons(0xFFF9))) {
545 /* It is not a fragmented frame */
546 skb->transport_header += sizeof(struct frag_hdr);
547 IP6_INC_STATS_BH(net,
548 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
549
550 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
551 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
552 return 1;
553 }
554
555 evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags, false);
556 if (evicted)
557 IP6_ADD_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
558 IPSTATS_MIB_REASMFAILS, evicted);
559
560 fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
561 ip6_frag_ecn(hdr));
562 if (fq != NULL) {
563 int ret;
564
565 spin_lock(&fq->q.lock);
566
567 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
568
569 spin_unlock(&fq->q.lock);
570 inet_frag_put(&fq->q, &ip6_frags);
571 return ret;
572 }
573
574 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
575 kfree_skb(skb);
576 return -1;
577
578fail_hdr:
579 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
580 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
581 return -1;
582}
583
584static const struct inet6_protocol frag_protocol =
585{
586 .handler = ipv6_frag_rcv,
587 .flags = INET6_PROTO_NOPOLICY,
588};
589
590#ifdef CONFIG_SYSCTL
591static struct ctl_table ip6_frags_ns_ctl_table[] = {
592 {
593 .procname = "ip6frag_high_thresh",
594 .data = &init_net.ipv6.frags.high_thresh,
595 .maxlen = sizeof(int),
596 .mode = 0644,
597 .proc_handler = proc_dointvec
598 },
599 {
600 .procname = "ip6frag_low_thresh",
601 .data = &init_net.ipv6.frags.low_thresh,
602 .maxlen = sizeof(int),
603 .mode = 0644,
604 .proc_handler = proc_dointvec
605 },
606 {
607 .procname = "ip6frag_time",
608 .data = &init_net.ipv6.frags.timeout,
609 .maxlen = sizeof(int),
610 .mode = 0644,
611 .proc_handler = proc_dointvec_jiffies,
612 },
613 { }
614};
615
616static struct ctl_table ip6_frags_ctl_table[] = {
617 {
618 .procname = "ip6frag_secret_interval",
619 .data = &ip6_frags.secret_interval,
620 .maxlen = sizeof(int),
621 .mode = 0644,
622 .proc_handler = proc_dointvec_jiffies,
623 },
624 { }
625};
626
627static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
628{
629 struct ctl_table *table;
630 struct ctl_table_header *hdr;
631
632 table = ip6_frags_ns_ctl_table;
633 if (!net_eq(net, &init_net)) {
634 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
635 if (table == NULL)
636 goto err_alloc;
637
638 table[0].data = &net->ipv6.frags.high_thresh;
639 table[1].data = &net->ipv6.frags.low_thresh;
640 table[2].data = &net->ipv6.frags.timeout;
641
642 /* Don't export sysctls to unprivileged users */
643 if (net->user_ns != &init_user_ns)
644 table[0].procname = NULL;
645 }
646
647 hdr = register_net_sysctl(net, "net/ipv6", table);
648 if (hdr == NULL)
649 goto err_reg;
650
651 net->ipv6.sysctl.frags_hdr = hdr;
652 return 0;
653
654err_reg:
655 if (!net_eq(net, &init_net))
656 kfree(table);
657err_alloc:
658 return -ENOMEM;
659}
660
661static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
662{
663 struct ctl_table *table;
664
665 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
666 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
667 if (!net_eq(net, &init_net))
668 kfree(table);
669}
670
671static struct ctl_table_header *ip6_ctl_header;
672
673static int ip6_frags_sysctl_register(void)
674{
675 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
676 ip6_frags_ctl_table);
677 return ip6_ctl_header == NULL ? -ENOMEM : 0;
678}
679
680static void ip6_frags_sysctl_unregister(void)
681{
682 unregister_net_sysctl_table(ip6_ctl_header);
683}
684#else
685static inline int ip6_frags_ns_sysctl_register(struct net *net)
686{
687 return 0;
688}
689
690static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
691{
692}
693
694static inline int ip6_frags_sysctl_register(void)
695{
696 return 0;
697}
698
699static inline void ip6_frags_sysctl_unregister(void)
700{
701}
702#endif
703
704static int __net_init ipv6_frags_init_net(struct net *net)
705{
706 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
707 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
708 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
709
710 inet_frags_init_net(&net->ipv6.frags);
711
712 return ip6_frags_ns_sysctl_register(net);
713}
714
715static void __net_exit ipv6_frags_exit_net(struct net *net)
716{
717 ip6_frags_ns_sysctl_unregister(net);
718 inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
719}
720
721static struct pernet_operations ip6_frags_ops = {
722 .init = ipv6_frags_init_net,
723 .exit = ipv6_frags_exit_net,
724};
725
726int __init ipv6_frag_init(void)
727{
728 int ret;
729
730 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
731 if (ret)
732 goto out;
733
734 ret = ip6_frags_sysctl_register();
735 if (ret)
736 goto err_sysctl;
737
738 ret = register_pernet_subsys(&ip6_frags_ops);
739 if (ret)
740 goto err_pernet;
741
742 ip6_frags.hashfn = ip6_hashfn;
743 ip6_frags.constructor = ip6_frag_init;
744 ip6_frags.destructor = NULL;
745 ip6_frags.skb_free = NULL;
746 ip6_frags.qsize = sizeof(struct frag_queue);
747 ip6_frags.match = ip6_frag_match;
748 ip6_frags.frag_expire = ip6_frag_expire;
749 ip6_frags.secret_interval = 10 * 60 * HZ;
750 inet_frags_init(&ip6_frags);
751out:
752 return ret;
753
754err_pernet:
755 ip6_frags_sysctl_unregister();
756err_sysctl:
757 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
758 goto out;
759}
760
761void ipv6_frag_exit(void)
762{
763 inet_frags_fini(&ip6_frags);
764 ip6_frags_sysctl_unregister();
765 unregister_pernet_subsys(&ip6_frags_ops);
766 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
767}