1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
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
63static const char ip6_frag_cache_name[] = "ip6-frags";
64
65static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
66{
67 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
68}
69
70static struct inet_frags ip6_frags;
71
72static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
73 struct net_device *dev);
74
75void ip6_frag_init(struct inet_frag_queue *q, const void *a)
76{
77 struct frag_queue *fq = container_of(q, struct frag_queue, q);
78 const struct frag_v6_compare_key *key = a;
79
80 q->key.v6 = *key;
81 fq->ecn = 0;
82}
83EXPORT_SYMBOL(ip6_frag_init);
84
85void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq)
86{
87 struct net_device *dev = NULL;
88 struct sk_buff *head;
89
90 rcu_read_lock();
91 spin_lock(&fq->q.lock);
92
93 if (fq->q.flags & INET_FRAG_COMPLETE)
94 goto out;
95
96 inet_frag_kill(&fq->q);
97
98 dev = dev_get_by_index_rcu(net, fq->iif);
99 if (!dev)
100 goto out;
101
102 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
103 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
104
105
106 head = fq->q.fragments;
107 if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !head)
108 goto out;
109
110
111
112
113
114 head->dev = dev;
115 skb_get(head);
116 spin_unlock(&fq->q.lock);
117
118 icmpv6_send(head, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
119 kfree_skb(head);
120 goto out_rcu_unlock;
121
122out:
123 spin_unlock(&fq->q.lock);
124out_rcu_unlock:
125 rcu_read_unlock();
126 inet_frag_put(&fq->q);
127}
128EXPORT_SYMBOL(ip6_expire_frag_queue);
129
130static void ip6_frag_expire(struct timer_list *t)
131{
132 struct inet_frag_queue *frag = from_timer(frag, t, timer);
133 struct frag_queue *fq;
134 struct net *net;
135
136 fq = container_of(frag, struct frag_queue, q);
137 net = container_of(fq->q.net, struct net, ipv6.frags);
138
139 ip6_expire_frag_queue(net, fq);
140}
141
142static struct frag_queue *
143fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
144{
145 struct frag_v6_compare_key key = {
146 .id = id,
147 .saddr = hdr->saddr,
148 .daddr = hdr->daddr,
149 .user = IP6_DEFRAG_LOCAL_DELIVER,
150 .iif = iif,
151 };
152 struct inet_frag_queue *q;
153
154 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
155 IPV6_ADDR_LINKLOCAL)))
156 key.iif = 0;
157
158 q = inet_frag_find(&net->ipv6.frags, &key);
159 if (!q)
160 return NULL;
161
162 return container_of(q, struct frag_queue, q);
163}
164
165static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
166 struct frag_hdr *fhdr, int nhoff)
167{
168 struct sk_buff *prev, *next;
169 struct net_device *dev;
170 int offset, end, fragsize;
171 struct net *net = dev_net(skb_dst(skb)->dev);
172 u8 ecn;
173
174 if (fq->q.flags & INET_FRAG_COMPLETE)
175 goto err;
176
177 offset = ntohs(fhdr->frag_off) & ~0x7;
178 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
179 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
180
181 if ((unsigned int)end > IPV6_MAXPLEN) {
182 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
183 IPSTATS_MIB_INHDRERRORS);
184 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
185 ((u8 *)&fhdr->frag_off -
186 skb_network_header(skb)));
187 return -1;
188 }
189
190 ecn = ip6_frag_ecn(ipv6_hdr(skb));
191
192 if (skb->ip_summed == CHECKSUM_COMPLETE) {
193 const unsigned char *nh = skb_network_header(skb);
194 skb->csum = csum_sub(skb->csum,
195 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
196 0));
197 }
198
199
200 if (!(fhdr->frag_off & htons(IP6_MF))) {
201
202
203
204 if (end < fq->q.len ||
205 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
206 goto err;
207 fq->q.flags |= INET_FRAG_LAST_IN;
208 fq->q.len = end;
209 } else {
210
211
212
213 if (end & 0x7) {
214
215
216
217 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
218 IPSTATS_MIB_INHDRERRORS);
219 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
220 offsetof(struct ipv6hdr, payload_len));
221 return -1;
222 }
223 if (end > fq->q.len) {
224
225 if (fq->q.flags & INET_FRAG_LAST_IN)
226 goto err;
227 fq->q.len = end;
228 }
229 }
230
231 if (end == offset)
232 goto err;
233
234
235 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
236 goto err;
237
238 if (pskb_trim_rcsum(skb, end - offset))
239 goto err;
240
241
242
243
244
245 prev = fq->q.fragments_tail;
246 if (!prev || prev->ip_defrag_offset < offset) {
247 next = NULL;
248 goto found;
249 }
250 prev = NULL;
251 for (next = fq->q.fragments; next != NULL; next = next->next) {
252 if (next->ip_defrag_offset >= offset)
253 break;
254 prev = next;
255 }
256
257found:
258
259
260
261
262
263
264
265
266 if (prev &&
267 (prev->ip_defrag_offset + prev->len) > offset)
268 goto discard_fq;
269
270
271 if (next && next->ip_defrag_offset < end)
272 goto discard_fq;
273
274
275 dev = skb->dev;
276 if (dev)
277 fq->iif = dev->ifindex;
278
279 barrier();
280 skb->ip_defrag_offset = offset;
281
282
283 skb->next = next;
284 if (!next)
285 fq->q.fragments_tail = skb;
286 if (prev)
287 prev->next = skb;
288 else
289 fq->q.fragments = skb;
290
291 fq->q.stamp = skb->tstamp;
292 fq->q.meat += skb->len;
293 fq->ecn |= ecn;
294 add_frag_mem_limit(fq->q.net, skb->truesize);
295
296 fragsize = -skb_network_offset(skb) + skb->len;
297 if (fragsize > fq->q.max_size)
298 fq->q.max_size = fragsize;
299
300
301
302
303 if (offset == 0) {
304 fq->nhoffset = nhoff;
305 fq->q.flags |= INET_FRAG_FIRST_IN;
306 }
307
308 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
309 fq->q.meat == fq->q.len) {
310 int res;
311 unsigned long orefdst = skb->_skb_refdst;
312
313 skb->_skb_refdst = 0UL;
314 res = ip6_frag_reasm(fq, prev, dev);
315 skb->_skb_refdst = orefdst;
316 return res;
317 }
318
319 skb_dst_drop(skb);
320 return -1;
321
322discard_fq:
323 inet_frag_kill(&fq->q);
324err:
325 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
326 IPSTATS_MIB_REASMFAILS);
327 kfree_skb(skb);
328 return -1;
329}
330
331
332
333
334
335
336
337
338
339
340static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
341 struct net_device *dev)
342{
343 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
344 struct sk_buff *fp, *head = fq->q.fragments;
345 int payload_len;
346 unsigned int nhoff;
347 int sum_truesize;
348 u8 ecn;
349
350 inet_frag_kill(&fq->q);
351
352 ecn = ip_frag_ecn_table[fq->ecn];
353 if (unlikely(ecn == 0xff))
354 goto out_fail;
355
356
357 if (prev) {
358 head = prev->next;
359 fp = skb_clone(head, GFP_ATOMIC);
360
361 if (!fp)
362 goto out_oom;
363
364 fp->next = head->next;
365 if (!fp->next)
366 fq->q.fragments_tail = fp;
367 prev->next = fp;
368
369 skb_morph(head, fq->q.fragments);
370 head->next = fq->q.fragments->next;
371
372 consume_skb(fq->q.fragments);
373 fq->q.fragments = head;
374 }
375
376 WARN_ON(head == NULL);
377 WARN_ON(head->ip_defrag_offset != 0);
378
379
380 payload_len = ((head->data - skb_network_header(head)) -
381 sizeof(struct ipv6hdr) + fq->q.len -
382 sizeof(struct frag_hdr));
383 if (payload_len > IPV6_MAXPLEN)
384 goto out_oversize;
385
386
387 if (skb_unclone(head, GFP_ATOMIC))
388 goto out_oom;
389
390
391
392
393 if (skb_has_frag_list(head)) {
394 struct sk_buff *clone;
395 int i, plen = 0;
396
397 clone = alloc_skb(0, GFP_ATOMIC);
398 if (!clone)
399 goto out_oom;
400 clone->next = head->next;
401 head->next = clone;
402 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
403 skb_frag_list_init(head);
404 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
405 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
406 clone->len = clone->data_len = head->data_len - plen;
407 head->data_len -= clone->len;
408 head->len -= clone->len;
409 clone->csum = 0;
410 clone->ip_summed = head->ip_summed;
411 add_frag_mem_limit(fq->q.net, clone->truesize);
412 }
413
414
415
416 nhoff = fq->nhoffset;
417 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
418 memmove(head->head + sizeof(struct frag_hdr), head->head,
419 (head->data - head->head) - sizeof(struct frag_hdr));
420 if (skb_mac_header_was_set(head))
421 head->mac_header += sizeof(struct frag_hdr);
422 head->network_header += sizeof(struct frag_hdr);
423
424 skb_reset_transport_header(head);
425 skb_push(head, head->data - skb_network_header(head));
426
427 sum_truesize = head->truesize;
428 for (fp = head->next; fp;) {
429 bool headstolen;
430 int delta;
431 struct sk_buff *next = fp->next;
432
433 sum_truesize += fp->truesize;
434 if (head->ip_summed != fp->ip_summed)
435 head->ip_summed = CHECKSUM_NONE;
436 else if (head->ip_summed == CHECKSUM_COMPLETE)
437 head->csum = csum_add(head->csum, fp->csum);
438
439 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
440 kfree_skb_partial(fp, headstolen);
441 } else {
442 if (!skb_shinfo(head)->frag_list)
443 skb_shinfo(head)->frag_list = fp;
444 head->data_len += fp->len;
445 head->len += fp->len;
446 head->truesize += fp->truesize;
447 }
448 fp = next;
449 }
450 sub_frag_mem_limit(fq->q.net, sum_truesize);
451
452 head->next = NULL;
453 head->dev = dev;
454 head->tstamp = fq->q.stamp;
455 ipv6_hdr(head)->payload_len = htons(payload_len);
456 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
457 IP6CB(head)->nhoff = nhoff;
458 IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
459 IP6CB(head)->frag_max_size = fq->q.max_size;
460
461
462 skb_postpush_rcsum(head, skb_network_header(head),
463 skb_network_header_len(head));
464
465 rcu_read_lock();
466 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
467 rcu_read_unlock();
468 fq->q.fragments = NULL;
469 fq->q.fragments_tail = NULL;
470 return 1;
471
472out_oversize:
473 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
474 goto out_fail;
475out_oom:
476 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
477out_fail:
478 rcu_read_lock();
479 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
480 rcu_read_unlock();
481 return -1;
482}
483
484static int ipv6_frag_rcv(struct sk_buff *skb)
485{
486 struct frag_hdr *fhdr;
487 struct frag_queue *fq;
488 const struct ipv6hdr *hdr = ipv6_hdr(skb);
489 struct net *net = dev_net(skb_dst(skb)->dev);
490 int iif;
491
492 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
493 goto fail_hdr;
494
495 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
496
497
498 if (hdr->payload_len == 0)
499 goto fail_hdr;
500
501 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
502 sizeof(struct frag_hdr))))
503 goto fail_hdr;
504
505 hdr = ipv6_hdr(skb);
506 fhdr = (struct frag_hdr *)skb_transport_header(skb);
507
508 if (!(fhdr->frag_off & htons(0xFFF9))) {
509
510 skb->transport_header += sizeof(struct frag_hdr);
511 __IP6_INC_STATS(net,
512 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
513
514 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
515 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
516 return 1;
517 }
518
519 iif = skb->dev ? skb->dev->ifindex : 0;
520 fq = fq_find(net, fhdr->identification, hdr, iif);
521 if (fq) {
522 int ret;
523
524 spin_lock(&fq->q.lock);
525
526 fq->iif = iif;
527 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
528
529 spin_unlock(&fq->q.lock);
530 inet_frag_put(&fq->q);
531 return ret;
532 }
533
534 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
535 kfree_skb(skb);
536 return -1;
537
538fail_hdr:
539 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
540 IPSTATS_MIB_INHDRERRORS);
541 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
542 return -1;
543}
544
545static const struct inet6_protocol frag_protocol = {
546 .handler = ipv6_frag_rcv,
547 .flags = INET6_PROTO_NOPOLICY,
548};
549
550#ifdef CONFIG_SYSCTL
551
552static struct ctl_table ip6_frags_ns_ctl_table[] = {
553 {
554 .procname = "ip6frag_high_thresh",
555 .data = &init_net.ipv6.frags.high_thresh,
556 .maxlen = sizeof(unsigned long),
557 .mode = 0644,
558 .proc_handler = proc_doulongvec_minmax,
559 .extra1 = &init_net.ipv6.frags.low_thresh
560 },
561 {
562 .procname = "ip6frag_low_thresh",
563 .data = &init_net.ipv6.frags.low_thresh,
564 .maxlen = sizeof(unsigned long),
565 .mode = 0644,
566 .proc_handler = proc_doulongvec_minmax,
567 .extra2 = &init_net.ipv6.frags.high_thresh
568 },
569 {
570 .procname = "ip6frag_time",
571 .data = &init_net.ipv6.frags.timeout,
572 .maxlen = sizeof(int),
573 .mode = 0644,
574 .proc_handler = proc_dointvec_jiffies,
575 },
576 { }
577};
578
579
580static int ip6_frags_secret_interval_unused;
581static struct ctl_table ip6_frags_ctl_table[] = {
582 {
583 .procname = "ip6frag_secret_interval",
584 .data = &ip6_frags_secret_interval_unused,
585 .maxlen = sizeof(int),
586 .mode = 0644,
587 .proc_handler = proc_dointvec_jiffies,
588 },
589 { }
590};
591
592static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
593{
594 struct ctl_table *table;
595 struct ctl_table_header *hdr;
596
597 table = ip6_frags_ns_ctl_table;
598 if (!net_eq(net, &init_net)) {
599 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
600 if (!table)
601 goto err_alloc;
602
603 table[0].data = &net->ipv6.frags.high_thresh;
604 table[0].extra1 = &net->ipv6.frags.low_thresh;
605 table[0].extra2 = &init_net.ipv6.frags.high_thresh;
606 table[1].data = &net->ipv6.frags.low_thresh;
607 table[1].extra2 = &net->ipv6.frags.high_thresh;
608 table[2].data = &net->ipv6.frags.timeout;
609 }
610
611 hdr = register_net_sysctl(net, "net/ipv6", table);
612 if (!hdr)
613 goto err_reg;
614
615 net->ipv6.sysctl.frags_hdr = hdr;
616 return 0;
617
618err_reg:
619 if (!net_eq(net, &init_net))
620 kfree(table);
621err_alloc:
622 return -ENOMEM;
623}
624
625static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
626{
627 struct ctl_table *table;
628
629 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
630 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
631 if (!net_eq(net, &init_net))
632 kfree(table);
633}
634
635static struct ctl_table_header *ip6_ctl_header;
636
637static int ip6_frags_sysctl_register(void)
638{
639 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
640 ip6_frags_ctl_table);
641 return ip6_ctl_header == NULL ? -ENOMEM : 0;
642}
643
644static void ip6_frags_sysctl_unregister(void)
645{
646 unregister_net_sysctl_table(ip6_ctl_header);
647}
648#else
649static int ip6_frags_ns_sysctl_register(struct net *net)
650{
651 return 0;
652}
653
654static void ip6_frags_ns_sysctl_unregister(struct net *net)
655{
656}
657
658static int ip6_frags_sysctl_register(void)
659{
660 return 0;
661}
662
663static void ip6_frags_sysctl_unregister(void)
664{
665}
666#endif
667
668static int __net_init ipv6_frags_init_net(struct net *net)
669{
670 int res;
671
672 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
673 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
674 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
675 net->ipv6.frags.f = &ip6_frags;
676
677 res = inet_frags_init_net(&net->ipv6.frags);
678 if (res < 0)
679 return res;
680
681 res = ip6_frags_ns_sysctl_register(net);
682 if (res < 0)
683 inet_frags_exit_net(&net->ipv6.frags);
684 return res;
685}
686
687static void __net_exit ipv6_frags_exit_net(struct net *net)
688{
689 ip6_frags_ns_sysctl_unregister(net);
690 inet_frags_exit_net(&net->ipv6.frags);
691}
692
693static struct pernet_operations ip6_frags_ops = {
694 .init = ipv6_frags_init_net,
695 .exit = ipv6_frags_exit_net,
696};
697
698static u32 ip6_key_hashfn(const void *data, u32 len, u32 seed)
699{
700 return jhash2(data,
701 sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
702}
703
704static u32 ip6_obj_hashfn(const void *data, u32 len, u32 seed)
705{
706 const struct inet_frag_queue *fq = data;
707
708 return jhash2((const u32 *)&fq->key.v6,
709 sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
710}
711
712static int ip6_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
713{
714 const struct frag_v6_compare_key *key = arg->key;
715 const struct inet_frag_queue *fq = ptr;
716
717 return !!memcmp(&fq->key, key, sizeof(*key));
718}
719
720const struct rhashtable_params ip6_rhash_params = {
721 .head_offset = offsetof(struct inet_frag_queue, node),
722 .hashfn = ip6_key_hashfn,
723 .obj_hashfn = ip6_obj_hashfn,
724 .obj_cmpfn = ip6_obj_cmpfn,
725 .automatic_shrinking = true,
726};
727EXPORT_SYMBOL(ip6_rhash_params);
728
729int __init ipv6_frag_init(void)
730{
731 int ret;
732
733 ip6_frags.constructor = ip6_frag_init;
734 ip6_frags.destructor = NULL;
735 ip6_frags.qsize = sizeof(struct frag_queue);
736 ip6_frags.frag_expire = ip6_frag_expire;
737 ip6_frags.frags_cache_name = ip6_frag_cache_name;
738 ip6_frags.rhash_params = ip6_rhash_params;
739 ret = inet_frags_init(&ip6_frags);
740 if (ret)
741 goto out;
742
743 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
744 if (ret)
745 goto err_protocol;
746
747 ret = ip6_frags_sysctl_register();
748 if (ret)
749 goto err_sysctl;
750
751 ret = register_pernet_subsys(&ip6_frags_ops);
752 if (ret)
753 goto err_pernet;
754
755out:
756 return ret;
757
758err_pernet:
759 ip6_frags_sysctl_unregister();
760err_sysctl:
761 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
762err_protocol:
763 inet_frags_fini(&ip6_frags);
764 goto out;
765}
766
767void ipv6_frag_exit(void)
768{
769 inet_frags_fini(&ip6_frags);
770 ip6_frags_sysctl_unregister();
771 unregister_pernet_subsys(&ip6_frags_ops);
772 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
773}
774