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#define pr_fmt(fmt) "IPv6: " fmt
27
28#include <linux/errno.h>
29#include <linux/types.h>
30#include <linux/string.h>
31#include <linux/socket.h>
32#include <linux/sockios.h>
33#include <linux/jiffies.h>
34#include <linux/net.h>
35#include <linux/list.h>
36#include <linux/netdevice.h>
37#include <linux/in6.h>
38#include <linux/ipv6.h>
39#include <linux/icmpv6.h>
40#include <linux/random.h>
41#include <linux/jhash.h>
42#include <linux/skbuff.h>
43#include <linux/slab.h>
44#include <linux/export.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/ipv6_frag.h>
57#include <net/inet_ecn.h>
58
59static const char ip6_frag_cache_name[] = "ip6-frags";
60
61static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
62{
63 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
64}
65
66static struct inet_frags ip6_frags;
67
68static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
69 struct sk_buff *prev_tail, struct net_device *dev);
70
71static void ip6_frag_expire(struct timer_list *t)
72{
73 struct inet_frag_queue *frag = from_timer(frag, t, timer);
74 struct frag_queue *fq;
75 struct net *net;
76
77 fq = container_of(frag, struct frag_queue, q);
78 net = container_of(fq->q.net, struct net, ipv6.frags);
79
80 ip6frag_expire_frag_queue(net, fq);
81}
82
83static struct frag_queue *
84fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
85{
86 struct frag_v6_compare_key key = {
87 .id = id,
88 .saddr = hdr->saddr,
89 .daddr = hdr->daddr,
90 .user = IP6_DEFRAG_LOCAL_DELIVER,
91 .iif = iif,
92 };
93 struct inet_frag_queue *q;
94
95 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
96 IPV6_ADDR_LINKLOCAL)))
97 key.iif = 0;
98
99 q = inet_frag_find(&net->ipv6.frags, &key);
100 if (!q)
101 return NULL;
102
103 return container_of(q, struct frag_queue, q);
104}
105
106static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
107 struct frag_hdr *fhdr, int nhoff,
108 u32 *prob_offset)
109{
110 struct net *net = dev_net(skb_dst(skb)->dev);
111 int offset, end, fragsize;
112 struct sk_buff *prev_tail;
113 struct net_device *dev;
114 int err = -ENOENT;
115 u8 ecn;
116
117 if (fq->q.flags & INET_FRAG_COMPLETE)
118 goto err;
119
120 err = -EINVAL;
121 offset = ntohs(fhdr->frag_off) & ~0x7;
122 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
123 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
124
125 if ((unsigned int)end > IPV6_MAXPLEN) {
126 *prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
127
128
129
130 return -1;
131 }
132
133 ecn = ip6_frag_ecn(ipv6_hdr(skb));
134
135 if (skb->ip_summed == CHECKSUM_COMPLETE) {
136 const unsigned char *nh = skb_network_header(skb);
137 skb->csum = csum_sub(skb->csum,
138 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
139 0));
140 }
141
142
143 if (!(fhdr->frag_off & htons(IP6_MF))) {
144
145
146
147 if (end < fq->q.len ||
148 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
149 goto discard_fq;
150 fq->q.flags |= INET_FRAG_LAST_IN;
151 fq->q.len = end;
152 } else {
153
154
155
156 if (end & 0x7) {
157
158
159
160 *prob_offset = offsetof(struct ipv6hdr, payload_len);
161 return -1;
162 }
163 if (end > fq->q.len) {
164
165 if (fq->q.flags & INET_FRAG_LAST_IN)
166 goto discard_fq;
167 fq->q.len = end;
168 }
169 }
170
171 if (end == offset)
172 goto discard_fq;
173
174 err = -ENOMEM;
175
176 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
177 goto discard_fq;
178
179 err = pskb_trim_rcsum(skb, end - offset);
180 if (err)
181 goto discard_fq;
182
183
184 dev = skb->dev;
185
186 barrier();
187
188 prev_tail = fq->q.fragments_tail;
189 err = inet_frag_queue_insert(&fq->q, skb, offset, end);
190 if (err)
191 goto insert_error;
192
193 if (dev)
194 fq->iif = dev->ifindex;
195
196 fq->q.stamp = skb->tstamp;
197 fq->q.meat += skb->len;
198 fq->ecn |= ecn;
199 add_frag_mem_limit(fq->q.net, skb->truesize);
200
201 fragsize = -skb_network_offset(skb) + skb->len;
202 if (fragsize > fq->q.max_size)
203 fq->q.max_size = fragsize;
204
205
206
207
208 if (offset == 0) {
209 fq->nhoffset = nhoff;
210 fq->q.flags |= INET_FRAG_FIRST_IN;
211 }
212
213 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
214 fq->q.meat == fq->q.len) {
215 unsigned long orefdst = skb->_skb_refdst;
216
217 skb->_skb_refdst = 0UL;
218 err = ip6_frag_reasm(fq, skb, prev_tail, dev);
219 skb->_skb_refdst = orefdst;
220 return err;
221 }
222
223 skb_dst_drop(skb);
224 return -EINPROGRESS;
225
226insert_error:
227 if (err == IPFRAG_DUP) {
228 kfree_skb(skb);
229 return -EINVAL;
230 }
231 err = -EINVAL;
232 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
233 IPSTATS_MIB_REASM_OVERLAPS);
234discard_fq:
235 inet_frag_kill(&fq->q);
236 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
237 IPSTATS_MIB_REASMFAILS);
238err:
239 kfree_skb(skb);
240 return err;
241}
242
243
244
245
246
247
248
249
250static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
251 struct sk_buff *prev_tail, struct net_device *dev)
252{
253 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
254 unsigned int nhoff;
255 void *reasm_data;
256 int payload_len;
257 u8 ecn;
258
259 inet_frag_kill(&fq->q);
260
261 ecn = ip_frag_ecn_table[fq->ecn];
262 if (unlikely(ecn == 0xff))
263 goto out_fail;
264
265 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
266 if (!reasm_data)
267 goto out_oom;
268
269 payload_len = ((skb->data - skb_network_header(skb)) -
270 sizeof(struct ipv6hdr) + fq->q.len -
271 sizeof(struct frag_hdr));
272 if (payload_len > IPV6_MAXPLEN)
273 goto out_oversize;
274
275
276
277 nhoff = fq->nhoffset;
278 skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
279 memmove(skb->head + sizeof(struct frag_hdr), skb->head,
280 (skb->data - skb->head) - sizeof(struct frag_hdr));
281 if (skb_mac_header_was_set(skb))
282 skb->mac_header += sizeof(struct frag_hdr);
283 skb->network_header += sizeof(struct frag_hdr);
284
285 skb_reset_transport_header(skb);
286
287 inet_frag_reasm_finish(&fq->q, skb, reasm_data);
288
289 skb->dev = dev;
290 ipv6_hdr(skb)->payload_len = htons(payload_len);
291 ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
292 IP6CB(skb)->nhoff = nhoff;
293 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
294 IP6CB(skb)->frag_max_size = fq->q.max_size;
295
296
297 skb_postpush_rcsum(skb, skb_network_header(skb),
298 skb_network_header_len(skb));
299
300 rcu_read_lock();
301 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS);
302 rcu_read_unlock();
303 fq->q.rb_fragments = RB_ROOT;
304 fq->q.fragments_tail = NULL;
305 fq->q.last_run_head = NULL;
306 return 1;
307
308out_oversize:
309 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
310 goto out_fail;
311out_oom:
312 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
313out_fail:
314 rcu_read_lock();
315 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS);
316 rcu_read_unlock();
317 inet_frag_kill(&fq->q);
318 return -1;
319}
320
321static int ipv6_frag_rcv(struct sk_buff *skb)
322{
323 struct frag_hdr *fhdr;
324 struct frag_queue *fq;
325 const struct ipv6hdr *hdr = ipv6_hdr(skb);
326 struct net *net = dev_net(skb_dst(skb)->dev);
327 int iif;
328
329 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
330 goto fail_hdr;
331
332 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
333
334
335 if (hdr->payload_len == 0)
336 goto fail_hdr;
337
338 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
339 sizeof(struct frag_hdr))))
340 goto fail_hdr;
341
342 hdr = ipv6_hdr(skb);
343 fhdr = (struct frag_hdr *)skb_transport_header(skb);
344
345 if (!(fhdr->frag_off & htons(0xFFF9))) {
346
347 skb->transport_header += sizeof(struct frag_hdr);
348 __IP6_INC_STATS(net,
349 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
350
351 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
352 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
353 return 1;
354 }
355
356 iif = skb->dev ? skb->dev->ifindex : 0;
357 fq = fq_find(net, fhdr->identification, hdr, iif);
358 if (fq) {
359 u32 prob_offset = 0;
360 int ret;
361
362 spin_lock(&fq->q.lock);
363
364 fq->iif = iif;
365 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
366 &prob_offset);
367
368 spin_unlock(&fq->q.lock);
369 inet_frag_put(&fq->q);
370 if (prob_offset) {
371 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
372 IPSTATS_MIB_INHDRERRORS);
373
374 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
375 }
376 return ret;
377 }
378
379 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
380 kfree_skb(skb);
381 return -1;
382
383fail_hdr:
384 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
385 IPSTATS_MIB_INHDRERRORS);
386 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
387 return -1;
388}
389
390static const struct inet6_protocol frag_protocol = {
391 .handler = ipv6_frag_rcv,
392 .flags = INET6_PROTO_NOPOLICY,
393};
394
395#ifdef CONFIG_SYSCTL
396
397static struct ctl_table ip6_frags_ns_ctl_table[] = {
398 {
399 .procname = "ip6frag_high_thresh",
400 .data = &init_net.ipv6.frags.high_thresh,
401 .maxlen = sizeof(unsigned long),
402 .mode = 0644,
403 .proc_handler = proc_doulongvec_minmax,
404 .extra1 = &init_net.ipv6.frags.low_thresh
405 },
406 {
407 .procname = "ip6frag_low_thresh",
408 .data = &init_net.ipv6.frags.low_thresh,
409 .maxlen = sizeof(unsigned long),
410 .mode = 0644,
411 .proc_handler = proc_doulongvec_minmax,
412 .extra2 = &init_net.ipv6.frags.high_thresh
413 },
414 {
415 .procname = "ip6frag_time",
416 .data = &init_net.ipv6.frags.timeout,
417 .maxlen = sizeof(int),
418 .mode = 0644,
419 .proc_handler = proc_dointvec_jiffies,
420 },
421 { }
422};
423
424
425static int ip6_frags_secret_interval_unused;
426static struct ctl_table ip6_frags_ctl_table[] = {
427 {
428 .procname = "ip6frag_secret_interval",
429 .data = &ip6_frags_secret_interval_unused,
430 .maxlen = sizeof(int),
431 .mode = 0644,
432 .proc_handler = proc_dointvec_jiffies,
433 },
434 { }
435};
436
437static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
438{
439 struct ctl_table *table;
440 struct ctl_table_header *hdr;
441
442 table = ip6_frags_ns_ctl_table;
443 if (!net_eq(net, &init_net)) {
444 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
445 if (!table)
446 goto err_alloc;
447
448 table[0].data = &net->ipv6.frags.high_thresh;
449 table[0].extra1 = &net->ipv6.frags.low_thresh;
450 table[1].data = &net->ipv6.frags.low_thresh;
451 table[1].extra2 = &net->ipv6.frags.high_thresh;
452 table[2].data = &net->ipv6.frags.timeout;
453 }
454
455 hdr = register_net_sysctl(net, "net/ipv6", table);
456 if (!hdr)
457 goto err_reg;
458
459 net->ipv6.sysctl.frags_hdr = hdr;
460 return 0;
461
462err_reg:
463 if (!net_eq(net, &init_net))
464 kfree(table);
465err_alloc:
466 return -ENOMEM;
467}
468
469static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
470{
471 struct ctl_table *table;
472
473 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
474 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
475 if (!net_eq(net, &init_net))
476 kfree(table);
477}
478
479static struct ctl_table_header *ip6_ctl_header;
480
481static int ip6_frags_sysctl_register(void)
482{
483 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
484 ip6_frags_ctl_table);
485 return ip6_ctl_header == NULL ? -ENOMEM : 0;
486}
487
488static void ip6_frags_sysctl_unregister(void)
489{
490 unregister_net_sysctl_table(ip6_ctl_header);
491}
492#else
493static int ip6_frags_ns_sysctl_register(struct net *net)
494{
495 return 0;
496}
497
498static void ip6_frags_ns_sysctl_unregister(struct net *net)
499{
500}
501
502static int ip6_frags_sysctl_register(void)
503{
504 return 0;
505}
506
507static void ip6_frags_sysctl_unregister(void)
508{
509}
510#endif
511
512static int __net_init ipv6_frags_init_net(struct net *net)
513{
514 int res;
515
516 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
517 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
518 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
519 net->ipv6.frags.f = &ip6_frags;
520
521 res = inet_frags_init_net(&net->ipv6.frags);
522 if (res < 0)
523 return res;
524
525 res = ip6_frags_ns_sysctl_register(net);
526 if (res < 0)
527 inet_frags_exit_net(&net->ipv6.frags);
528 return res;
529}
530
531static void __net_exit ipv6_frags_exit_net(struct net *net)
532{
533 ip6_frags_ns_sysctl_unregister(net);
534 inet_frags_exit_net(&net->ipv6.frags);
535}
536
537static struct pernet_operations ip6_frags_ops = {
538 .init = ipv6_frags_init_net,
539 .exit = ipv6_frags_exit_net,
540};
541
542static const struct rhashtable_params ip6_rhash_params = {
543 .head_offset = offsetof(struct inet_frag_queue, node),
544 .hashfn = ip6frag_key_hashfn,
545 .obj_hashfn = ip6frag_obj_hashfn,
546 .obj_cmpfn = ip6frag_obj_cmpfn,
547 .automatic_shrinking = true,
548};
549
550int __init ipv6_frag_init(void)
551{
552 int ret;
553
554 ip6_frags.constructor = ip6frag_init;
555 ip6_frags.destructor = NULL;
556 ip6_frags.qsize = sizeof(struct frag_queue);
557 ip6_frags.frag_expire = ip6_frag_expire;
558 ip6_frags.frags_cache_name = ip6_frag_cache_name;
559 ip6_frags.rhash_params = ip6_rhash_params;
560 ret = inet_frags_init(&ip6_frags);
561 if (ret)
562 goto out;
563
564 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
565 if (ret)
566 goto err_protocol;
567
568 ret = ip6_frags_sysctl_register();
569 if (ret)
570 goto err_sysctl;
571
572 ret = register_pernet_subsys(&ip6_frags_ops);
573 if (ret)
574 goto err_pernet;
575
576out:
577 return ret;
578
579err_pernet:
580 ip6_frags_sysctl_unregister();
581err_sysctl:
582 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
583err_protocol:
584 inet_frags_fini(&ip6_frags);
585 goto out;
586}
587
588void ipv6_frag_exit(void)
589{
590 inet_frags_fini(&ip6_frags);
591 ip6_frags_sysctl_unregister();
592 unregister_pernet_subsys(&ip6_frags_ops);
593 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
594}
595