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
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}
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}