?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ipv6: frags: rewrite ip6_expire_frag_queue()
[linux-stable.git] / net / ipv6 / reassembly.c
blob2127da130dc2cee677059247d5e079a8afbca13d
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
63 static const char ip6_frag_cache_name[] = "ip6-frags";
64
65 struct ip6frag_skb_cb {
66 struct inet6_skb_parm h;
67 int offset;
68 };
69
70 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb *)((skb)->cb))
71
72 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
73 {
74 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
75 }
76
77 static struct inet_frags ip6_frags;
78
79 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
80 struct net_device *dev);
81
82 void ip6_frag_init(struct inet_frag_queue *q, const void *a)
83 {
84 struct frag_queue *fq = container_of(q, struct frag_queue, q);
85 const struct frag_v6_compare_key *key = a;
86
87 q->key.v6 = *key;
88 fq->ecn = 0;
89 }
90 EXPORT_SYMBOL(ip6_frag_init);
91
92 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq)
93 {
94 struct net_device *dev = NULL;
95 struct sk_buff *head;
96
97 rcu_read_lock();
98 spin_lock(&fq->q.lock);
99
100 if (fq->q.flags & INET_FRAG_COMPLETE)
101 goto out;
102
103 inet_frag_kill(&fq->q);
104
105 dev = dev_get_by_index_rcu(net, fq->iif);
106 if (!dev)
107 goto out;
108
109 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
110 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
111
112 /* Don't send error if the first segment did not arrive. */
113 head = fq->q.fragments;
114 if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !head)
115 goto out;
116
117 /* But use as source device on which LAST ARRIVED
118 * segment was received. And do not use fq->dev
119 * pointer directly, device might already disappeared.
120 */
121 head->dev = dev;
122 skb_get(head);
123 spin_unlock(&fq->q.lock);
124
125 icmpv6_send(head, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
126 kfree_skb(head);
127 goto out_rcu_unlock;
128
129 out:
130 spin_unlock(&fq->q.lock);
131 out_rcu_unlock:
132 rcu_read_unlock();
133 inet_frag_put(&fq->q);
134 }
135 EXPORT_SYMBOL(ip6_expire_frag_queue);
136
137 static void ip6_frag_expire(struct timer_list *t)
138 {
139 struct inet_frag_queue *frag = from_timer(frag, t, timer);
140 struct frag_queue *fq;
141 struct net *net;
142
143 fq = container_of(frag, struct frag_queue, q);
144 net = container_of(fq->q.net, struct net, ipv6.frags);
145
146 ip6_expire_frag_queue(net, fq);
147 }
148
149 static struct frag_queue *
150 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
151 {
152 struct frag_v6_compare_key key = {
153 .id = id,
154 .saddr = hdr->saddr,
155 .daddr = hdr->daddr,
156 .user = IP6_DEFRAG_LOCAL_DELIVER,
157 .iif = iif,
158 };
159 struct inet_frag_queue *q;
160
161 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
162 IPV6_ADDR_LINKLOCAL)))
163 key.iif = 0;
164
165 q = inet_frag_find(&net->ipv6.frags, &key);
166 if (!q)
167 return NULL;
168
169 return container_of(q, struct frag_queue, q);
170 }
171
172 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
173 struct frag_hdr *fhdr, int nhoff)
174 {
175 struct sk_buff *prev, *next;
176 struct net_device *dev;
177 int offset, end, fragsize;
178 struct net *net = dev_net(skb_dst(skb)->dev);
179 u8 ecn;
180
181 if (fq->q.flags & INET_FRAG_COMPLETE)
182 goto err;
183
184 offset = ntohs(fhdr->frag_off) & ~0x7;
185 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
186 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
187
188 if ((unsigned int)end > IPV6_MAXPLEN) {
189 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
190 IPSTATS_MIB_INHDRERRORS);
191 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
192 ((u8 *)&fhdr->frag_off -
193 skb_network_header(skb)));
194 return -1;
195 }
196
197 ecn = ip6_frag_ecn(ipv6_hdr(skb));
198
199 if (skb->ip_summed == CHECKSUM_COMPLETE) {
200 const unsigned char *nh = skb_network_header(skb);
201 skb->csum = csum_sub(skb->csum,
202 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
203 0));
204 }
205
206 /* Is this the final fragment? */
207 if (!(fhdr->frag_off & htons(IP6_MF))) {
208 /* If we already have some bits beyond end
209 * or have different end, the segment is corrupted.
210 */
211 if (end < fq->q.len ||
212 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
213 goto err;
214 fq->q.flags |= INET_FRAG_LAST_IN;
215 fq->q.len = end;
216 } else {
217 /* Check if the fragment is rounded to 8 bytes.
218 * Required by the RFC.
219 */
220 if (end & 0x7) {
221 /* RFC2460 says always send parameter problem in
222 * this case. -DaveM
223 */
224 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
225 IPSTATS_MIB_INHDRERRORS);
226 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
227 offsetof(struct ipv6hdr, payload_len));
228 return -1;
229 }
230 if (end > fq->q.len) {
231 /* Some bits beyond end -> corruption. */
232 if (fq->q.flags & INET_FRAG_LAST_IN)
233 goto err;
234 fq->q.len = end;
235 }
236 }
237
238 if (end == offset)
239 goto err;
240
241 /* Point into the IP datagram 'data' part. */
242 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
243 goto err;
244
245 if (pskb_trim_rcsum(skb, end - offset))
246 goto err;
247
248 /* Find out which fragments are in front and at the back of us
249 * in the chain of fragments so far. We must know where to put
250 * this fragment, right?
251 */
252 prev = fq->q.fragments_tail;
253 if (!prev || FRAG6_CB(prev)->offset < offset) {
254 next = NULL;
255 goto found;
256 }
257 prev = NULL;
258 for (next = fq->q.fragments; next != NULL; next = next->next) {
259 if (FRAG6_CB(next)->offset >= offset)
260 break; /* bingo! */
261 prev = next;
262 }
263
264 found:
265 /* RFC5722, Section 4, amended by Errata ID : 3089
266 * When reassembling an IPv6 datagram, if
267 * one or more its constituent fragments is determined to be an
268 * overlapping fragment, the entire datagram (and any constituent
269 * fragments) MUST be silently discarded.
270 */
271
272 /* Check for overlap with preceding fragment. */
273 if (prev &&
274 (FRAG6_CB(prev)->offset + prev->len) > offset)
275 goto discard_fq;
276
277 /* Look for overlap with succeeding segment. */
278 if (next && FRAG6_CB(next)->offset < end)
279 goto discard_fq;
280
281 FRAG6_CB(skb)->offset = offset;
282
283 /* Insert this fragment in the chain of fragments. */
284 skb->next = next;
285 if (!next)
286 fq->q.fragments_tail = skb;
287 if (prev)
288 prev->next = skb;
289 else
290 fq->q.fragments = skb;
291
292 dev = skb->dev;
293 if (dev) {
294 fq->iif = dev->ifindex;
295 skb->dev = NULL;
296 }
297 fq->q.stamp = skb->tstamp;
298 fq->q.meat += skb->len;
299 fq->ecn |= ecn;
300 add_frag_mem_limit(fq->q.net, skb->truesize);
301
302 fragsize = -skb_network_offset(skb) + skb->len;
303 if (fragsize > fq->q.max_size)
304 fq->q.max_size = fragsize;
305
306 /* The first fragment.
307 * nhoffset is obtained from the first fragment, of course.
308 */
309 if (offset == 0) {
310 fq->nhoffset = nhoff;
311 fq->q.flags |= INET_FRAG_FIRST_IN;
312 }
313
314 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
315 fq->q.meat == fq->q.len) {
316 int res;
317 unsigned long orefdst = skb->_skb_refdst;
318
319 skb->_skb_refdst = 0UL;
320 res = ip6_frag_reasm(fq, prev, dev);
321 skb->_skb_refdst = orefdst;
322 return res;
323 }
324
325 skb_dst_drop(skb);
326 return -1;
327
328 discard_fq:
329 inet_frag_kill(&fq->q);
330 err:
331 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
332 IPSTATS_MIB_REASMFAILS);
333 kfree_skb(skb);
334 return -1;
335 }
336
337 /*
338 * Check if this packet is complete.
339 * Returns NULL on failure by any reason, and pointer
340 * to current nexthdr field in reassembled frame.
341 *
342 * It is called with locked fq, and caller must check that
343 * queue is eligible for reassembly i.e. it is not COMPLETE,
344 * the last and the first frames arrived and all the bits are here.
345 */
346 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
347 struct net_device *dev)
348 {
349 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
350 struct sk_buff *fp, *head = fq->q.fragments;
351 int payload_len;
352 unsigned int nhoff;
353 int sum_truesize;
354 u8 ecn;
355
356 inet_frag_kill(&fq->q);
357
358 ecn = ip_frag_ecn_table[fq->ecn];
359 if (unlikely(ecn == 0xff))
360 goto out_fail;
361
362 /* Make the one we just received the head. */
363 if (prev) {
364 head = prev->next;
365 fp = skb_clone(head, GFP_ATOMIC);
366
367 if (!fp)
368 goto out_oom;
369
370 fp->next = head->next;
371 if (!fp->next)
372 fq->q.fragments_tail = fp;
373 prev->next = fp;
374
375 skb_morph(head, fq->q.fragments);
376 head->next = fq->q.fragments->next;
377
378 consume_skb(fq->q.fragments);
379 fq->q.fragments = head;
380 }
381
382 WARN_ON(head == NULL);
383 WARN_ON(FRAG6_CB(head)->offset != 0);
384
385 /* Unfragmented part is taken from the first segment. */
386 payload_len = ((head->data - skb_network_header(head)) -
387 sizeof(struct ipv6hdr) + fq->q.len -
388 sizeof(struct frag_hdr));
389 if (payload_len > IPV6_MAXPLEN)
390 goto out_oversize;
391
392 /* Head of list must not be cloned. */
393 if (skb_unclone(head, GFP_ATOMIC))
394 goto out_oom;
395
396 /* If the first fragment is fragmented itself, we split
397 * it to two chunks: the first with data and paged part
398 * and the second, holding only fragments. */
399 if (skb_has_frag_list(head)) {
400 struct sk_buff *clone;
401 int i, plen = 0;
402
403 clone = alloc_skb(0, GFP_ATOMIC);
404 if (!clone)
405 goto out_oom;
406 clone->next = head->next;
407 head->next = clone;
408 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
409 skb_frag_list_init(head);
410 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
411 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
412 clone->len = clone->data_len = head->data_len - plen;
413 head->data_len -= clone->len;
414 head->len -= clone->len;
415 clone->csum = 0;
416 clone->ip_summed = head->ip_summed;
417 add_frag_mem_limit(fq->q.net, clone->truesize);
418 }
419
420 /* We have to remove fragment header from datagram and to relocate
421 * header in order to calculate ICV correctly. */
422 nhoff = fq->nhoffset;
423 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
424 memmove(head->head + sizeof(struct frag_hdr), head->head,
425 (head->data - head->head) - sizeof(struct frag_hdr));
426 if (skb_mac_header_was_set(head))
427 head->mac_header += sizeof(struct frag_hdr);
428 head->network_header += sizeof(struct frag_hdr);
429
430 skb_reset_transport_header(head);
431 skb_push(head, head->data - skb_network_header(head));
432
433 sum_truesize = head->truesize;
434 for (fp = head->next; fp;) {
435 bool headstolen;
436 int delta;
437 struct sk_buff *next = fp->next;
438
439 sum_truesize += fp->truesize;
440 if (head->ip_summed != fp->ip_summed)
441 head->ip_summed = CHECKSUM_NONE;
442 else if (head->ip_summed == CHECKSUM_COMPLETE)
443 head->csum = csum_add(head->csum, fp->csum);
444
445 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
446 kfree_skb_partial(fp, headstolen);
447 } else {
448 if (!skb_shinfo(head)->frag_list)
449 skb_shinfo(head)->frag_list = fp;
450 head->data_len += fp->len;
451 head->len += fp->len;
452 head->truesize += fp->truesize;
453 }
454 fp = next;
455 }
456 sub_frag_mem_limit(fq->q.net, sum_truesize);
457
458 head->next = NULL;
459 head->dev = dev;
460 head->tstamp = fq->q.stamp;
461 ipv6_hdr(head)->payload_len = htons(payload_len);
462 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
463 IP6CB(head)->nhoff = nhoff;
464 IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
465 IP6CB(head)->frag_max_size = fq->q.max_size;
466
467 /* Yes, and fold redundant checksum back. 8) */
468 skb_postpush_rcsum(head, skb_network_header(head),
469 skb_network_header_len(head));
470
471 rcu_read_lock();
472 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
473 rcu_read_unlock();
474 fq->q.fragments = NULL;
475 fq->q.fragments_tail = NULL;
476 return 1;
477
478 out_oversize:
479 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
480 goto out_fail;
481 out_oom:
482 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
483 out_fail:
484 rcu_read_lock();
485 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
486 rcu_read_unlock();
487 return -1;
488 }
489
490 static int ipv6_frag_rcv(struct sk_buff *skb)
491 {
492 struct frag_hdr *fhdr;
493 struct frag_queue *fq;
494 const struct ipv6hdr *hdr = ipv6_hdr(skb);
495 struct net *net = dev_net(skb_dst(skb)->dev);
496 int iif;
497
498 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
499 goto fail_hdr;
500
501 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
502
503 /* Jumbo payload inhibits frag. header */
504 if (hdr->payload_len == 0)
505 goto fail_hdr;
506
507 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
508 sizeof(struct frag_hdr))))
509 goto fail_hdr;
510
511 hdr = ipv6_hdr(skb);
512 fhdr = (struct frag_hdr *)skb_transport_header(skb);
513
514 if (!(fhdr->frag_off & htons(0xFFF9))) {
515 /* It is not a fragmented frame */
516 skb->transport_header += sizeof(struct frag_hdr);
517 __IP6_INC_STATS(net,
518 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
519
520 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
521 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
522 return 1;
523 }
524
525 iif = skb->dev ? skb->dev->ifindex : 0;
526 fq = fq_find(net, fhdr->identification, hdr, iif);
527 if (fq) {
528 int ret;
529
530 spin_lock(&fq->q.lock);
531
532 fq->iif = iif;
533 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
534
535 spin_unlock(&fq->q.lock);
536 inet_frag_put(&fq->q);
537 return ret;
538 }
539
540 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
541 kfree_skb(skb);
542 return -1;
543
544 fail_hdr:
545 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
546 IPSTATS_MIB_INHDRERRORS);
547 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
548 return -1;
549 }
550
551 static const struct inet6_protocol frag_protocol = {
552 .handler = ipv6_frag_rcv,
553 .flags = INET6_PROTO_NOPOLICY,
554 };
555
556 #ifdef CONFIG_SYSCTL
557 static int zero;
558
559 static struct ctl_table ip6_frags_ns_ctl_table[] = {
560 {
561 .procname = "ip6frag_high_thresh",
562 .data = &init_net.ipv6.frags.high_thresh,
563 .maxlen = sizeof(unsigned long),
564 .mode = 0644,
565 .proc_handler = proc_doulongvec_minmax,
566 .extra1 = &init_net.ipv6.frags.low_thresh
567 },
568 {
569 .procname = "ip6frag_low_thresh",
570 .data = &init_net.ipv6.frags.low_thresh,
571 .maxlen = sizeof(unsigned long),
572 .mode = 0644,
573 .proc_handler = proc_dointvec_minmax,
574 .extra1 = &zero,
575 .extra2 = &init_net.ipv6.frags.high_thresh
576 },
577 {
578 .procname = "ip6frag_time",
579 .data = &init_net.ipv6.frags.timeout,
580 .maxlen = sizeof(int),
581 .mode = 0644,
582 .proc_handler = proc_dointvec_jiffies,
583 },
584 { }
585 };
586
587 /* secret interval has been deprecated */
588 static int ip6_frags_secret_interval_unused;
589 static struct ctl_table ip6_frags_ctl_table[] = {
590 {
591 .procname = "ip6frag_secret_interval",
592 .data = &ip6_frags_secret_interval_unused,
593 .maxlen = sizeof(int),
594 .mode = 0644,
595 .proc_handler = proc_dointvec_jiffies,
596 },
597 { }
598 };
599
600 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
601 {
602 struct ctl_table *table;
603 struct ctl_table_header *hdr;
604
605 table = ip6_frags_ns_ctl_table;
606 if (!net_eq(net, &init_net)) {
607 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
608 if (!table)
609 goto err_alloc;
610
611 table[0].data = &net->ipv6.frags.high_thresh;
612 table[0].extra1 = &net->ipv6.frags.low_thresh;
613 table[0].extra2 = &init_net.ipv6.frags.high_thresh;
614 table[1].data = &net->ipv6.frags.low_thresh;
615 table[1].extra2 = &net->ipv6.frags.high_thresh;
616 table[2].data = &net->ipv6.frags.timeout;
617 }
618
619 hdr = register_net_sysctl(net, "net/ipv6", table);
620 if (!hdr)
621 goto err_reg;
622
623 net->ipv6.sysctl.frags_hdr = hdr;
624 return 0;
625
626 err_reg:
627 if (!net_eq(net, &init_net))
628 kfree(table);
629 err_alloc:
630 return -ENOMEM;
631 }
632
633 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
634 {
635 struct ctl_table *table;
636
637 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
638 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
639 if (!net_eq(net, &init_net))
640 kfree(table);
641 }
642
643 static struct ctl_table_header *ip6_ctl_header;
644
645 static int ip6_frags_sysctl_register(void)
646 {
647 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
648 ip6_frags_ctl_table);
649 return ip6_ctl_header == NULL ? -ENOMEM : 0;
650 }
651
652 static void ip6_frags_sysctl_unregister(void)
653 {
654 unregister_net_sysctl_table(ip6_ctl_header);
655 }
656 #else
657 static int ip6_frags_ns_sysctl_register(struct net *net)
658 {
659 return 0;
660 }
661
662 static void ip6_frags_ns_sysctl_unregister(struct net *net)
663 {
664 }
665
666 static int ip6_frags_sysctl_register(void)
667 {
668 return 0;
669 }
670
671 static void ip6_frags_sysctl_unregister(void)
672 {
673 }
674 #endif
675
676 static int __net_init ipv6_frags_init_net(struct net *net)
677 {
678 int res;
679
680 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
681 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
682 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
683 net->ipv6.frags.f = &ip6_frags;
684
685 res = inet_frags_init_net(&net->ipv6.frags);
686 if (res < 0)
687 return res;
688
689 res = ip6_frags_ns_sysctl_register(net);
690 if (res < 0)
691 inet_frags_exit_net(&net->ipv6.frags);
692 return res;
693 }
694
695 static void __net_exit ipv6_frags_exit_net(struct net *net)
696 {
697 ip6_frags_ns_sysctl_unregister(net);
698 inet_frags_exit_net(&net->ipv6.frags);
699 }
700
701 static struct pernet_operations ip6_frags_ops = {
702 .init = ipv6_frags_init_net,
703 .exit = ipv6_frags_exit_net,
704 };
705
706 static u32 ip6_key_hashfn(const void *data, u32 len, u32 seed)
707 {
708 return jhash2(data,
709 sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
710 }
711
712 static u32 ip6_obj_hashfn(const void *data, u32 len, u32 seed)
713 {
714 const struct inet_frag_queue *fq = data;
715
716 return jhash2((const u32 *)&fq->key.v6,
717 sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
718 }
719
720 static int ip6_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
721 {
722 const struct frag_v6_compare_key *key = arg->key;
723 const struct inet_frag_queue *fq = ptr;
724
725 return !!memcmp(&fq->key, key, sizeof(*key));
726 }
727
728 const struct rhashtable_params ip6_rhash_params = {
729 .head_offset = offsetof(struct inet_frag_queue, node),
730 .hashfn = ip6_key_hashfn,
731 .obj_hashfn = ip6_obj_hashfn,
732 .obj_cmpfn = ip6_obj_cmpfn,
733 .automatic_shrinking = true,
734 };
735 EXPORT_SYMBOL(ip6_rhash_params);
736
737 int __init ipv6_frag_init(void)
738 {
739 int ret;
740
741 ip6_frags.constructor = ip6_frag_init;
742 ip6_frags.destructor = NULL;
743 ip6_frags.qsize = sizeof(struct frag_queue);
744 ip6_frags.frag_expire = ip6_frag_expire;
745 ip6_frags.frags_cache_name = ip6_frag_cache_name;
746 ip6_frags.rhash_params = ip6_rhash_params;
747 ret = inet_frags_init(&ip6_frags);
748 if (ret)
749 goto out;
750
751 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
752 if (ret)
753 goto err_protocol;
754
755 ret = ip6_frags_sysctl_register();
756 if (ret)
757 goto err_sysctl;
758
759 ret = register_pernet_subsys(&ip6_frags_ops);
760 if (ret)
761 goto err_pernet;
762
763 out:
764 return ret;
765
766 err_pernet:
767 ip6_frags_sysctl_unregister();
768 err_sysctl:
769 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
770 err_protocol:
771 inet_frags_fini(&ip6_frags);
772 goto out;
773 }
774
775 void ipv6_frag_exit(void)
776 {
777 inet_frags_fini(&ip6_frags);
778 ip6_frags_sysctl_unregister();
779 unregister_pernet_subsys(&ip6_frags_ops);
780 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
781 }
Linux kernel stable tree