Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * IPv6 fragment reassembly
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Based on: net/ipv4/ip_fragment.c
10 */
11
12/*
13 * Fixes:
14 * Andi Kleen Make it work with multiple hosts.
15 * More RFC compliance.
16 *
17 * Horst von Brand Add missing #include <linux/string.h>
18 * Alexey Kuznetsov SMP races, threading, cleanup.
19 * Patrick McHardy LRU queue of frag heads for evictor.
20 * Mitsuru KANDA @USAGI Register inet6_protocol{}.
21 * David Stevens and
22 * YOSHIFUJI,H. @USAGI Always remove fragment header to
23 * calculate ICV correctly.
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
76 fq = container_of(frag, struct frag_queue, q);
77
78 ip6frag_expire_frag_queue(fq->q.fqdir->net, fq);
79}
80
81static struct frag_queue *
82fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
83{
84 struct frag_v6_compare_key key = {
85 .id = id,
86 .saddr = hdr->saddr,
87 .daddr = hdr->daddr,
88 .user = IP6_DEFRAG_LOCAL_DELIVER,
89 .iif = iif,
90 };
91 struct inet_frag_queue *q;
92
93 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
94 IPV6_ADDR_LINKLOCAL)))
95 key.iif = 0;
96
97 q = inet_frag_find(net->ipv6.fqdir, &key);
98 if (!q)
99 return NULL;
100
101 return container_of(q, struct frag_queue, q);
102}
103
104static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
105 struct frag_hdr *fhdr, int nhoff,
106 u32 *prob_offset)
107{
108 struct net *net = dev_net(skb_dst(skb)->dev);
109 int offset, end, fragsize;
110 struct sk_buff *prev_tail;
111 struct net_device *dev;
112 int err = -ENOENT;
113 u8 ecn;
114
115 if (fq->q.flags & INET_FRAG_COMPLETE)
116 goto err;
117
118 err = -EINVAL;
119 offset = ntohs(fhdr->frag_off) & ~0x7;
120 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
121 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
122
123 if ((unsigned int)end > IPV6_MAXPLEN) {
124 *prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
125 /* note that if prob_offset is set, the skb is freed elsewhere,
126 * we do not free it here.
127 */
128 return -1;
129 }
130
131 ecn = ip6_frag_ecn(ipv6_hdr(skb));
132
133 if (skb->ip_summed == CHECKSUM_COMPLETE) {
134 const unsigned char *nh = skb_network_header(skb);
135 skb->csum = csum_sub(skb->csum,
136 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
137 0));
138 }
139
140 /* Is this the final fragment? */
141 if (!(fhdr->frag_off & htons(IP6_MF))) {
142 /* If we already have some bits beyond end
143 * or have different end, the segment is corrupted.
144 */
145 if (end < fq->q.len ||
146 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
147 goto discard_fq;
148 fq->q.flags |= INET_FRAG_LAST_IN;
149 fq->q.len = end;
150 } else {
151 /* Check if the fragment is rounded to 8 bytes.
152 * Required by the RFC.
153 */
154 if (end & 0x7) {
155 /* RFC2460 says always send parameter problem in
156 * this case. -DaveM
157 */
158 *prob_offset = offsetof(struct ipv6hdr, payload_len);
159 return -1;
160 }
161 if (end > fq->q.len) {
162 /* Some bits beyond end -> corruption. */
163 if (fq->q.flags & INET_FRAG_LAST_IN)
164 goto discard_fq;
165 fq->q.len = end;
166 }
167 }
168
169 if (end == offset)
170 goto discard_fq;
171
172 err = -ENOMEM;
173 /* Point into the IP datagram 'data' part. */
174 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
175 goto discard_fq;
176
177 err = pskb_trim_rcsum(skb, end - offset);
178 if (err)
179 goto discard_fq;
180
181 /* Note : skb->rbnode and skb->dev share the same location. */
182 dev = skb->dev;
183 /* Makes sure compiler wont do silly aliasing games */
184 barrier();
185
186 prev_tail = fq->q.fragments_tail;
187 err = inet_frag_queue_insert(&fq->q, skb, offset, end);
188 if (err)
189 goto insert_error;
190
191 if (dev)
192 fq->iif = dev->ifindex;
193
194 fq->q.stamp = skb->tstamp;
195 fq->q.meat += skb->len;
196 fq->ecn |= ecn;
197 add_frag_mem_limit(fq->q.fqdir, skb->truesize);
198
199 fragsize = -skb_network_offset(skb) + skb->len;
200 if (fragsize > fq->q.max_size)
201 fq->q.max_size = fragsize;
202
203 /* The first fragment.
204 * nhoffset is obtained from the first fragment, of course.
205 */
206 if (offset == 0) {
207 fq->nhoffset = nhoff;
208 fq->q.flags |= INET_FRAG_FIRST_IN;
209 }
210
211 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
212 fq->q.meat == fq->q.len) {
213 unsigned long orefdst = skb->_skb_refdst;
214
215 skb->_skb_refdst = 0UL;
216 err = ip6_frag_reasm(fq, skb, prev_tail, dev);
217 skb->_skb_refdst = orefdst;
218 return err;
219 }
220
221 skb_dst_drop(skb);
222 return -EINPROGRESS;
223
224insert_error:
225 if (err == IPFRAG_DUP) {
226 kfree_skb(skb);
227 return -EINVAL;
228 }
229 err = -EINVAL;
230 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
231 IPSTATS_MIB_REASM_OVERLAPS);
232discard_fq:
233 inet_frag_kill(&fq->q);
234 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
235 IPSTATS_MIB_REASMFAILS);
236err:
237 kfree_skb(skb);
238 return err;
239}
240
241/*
242 * Check if this packet is complete.
243 *
244 * It is called with locked fq, and caller must check that
245 * queue is eligible for reassembly i.e. it is not COMPLETE,
246 * the last and the first frames arrived and all the bits are here.
247 */
248static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
249 struct sk_buff *prev_tail, struct net_device *dev)
250{
251 struct net *net = fq->q.fqdir->net;
252 unsigned int nhoff;
253 void *reasm_data;
254 int payload_len;
255 u8 ecn;
256
257 inet_frag_kill(&fq->q);
258
259 ecn = ip_frag_ecn_table[fq->ecn];
260 if (unlikely(ecn == 0xff))
261 goto out_fail;
262
263 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
264 if (!reasm_data)
265 goto out_oom;
266
267 payload_len = ((skb->data - skb_network_header(skb)) -
268 sizeof(struct ipv6hdr) + fq->q.len -
269 sizeof(struct frag_hdr));
270 if (payload_len > IPV6_MAXPLEN)
271 goto out_oversize;
272
273 /* We have to remove fragment header from datagram and to relocate
274 * header in order to calculate ICV correctly. */
275 nhoff = fq->nhoffset;
276 skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
277 memmove(skb->head + sizeof(struct frag_hdr), skb->head,
278 (skb->data - skb->head) - sizeof(struct frag_hdr));
279 if (skb_mac_header_was_set(skb))
280 skb->mac_header += sizeof(struct frag_hdr);
281 skb->network_header += sizeof(struct frag_hdr);
282
283 skb_reset_transport_header(skb);
284
285 inet_frag_reasm_finish(&fq->q, skb, reasm_data, true);
286
287 skb->dev = dev;
288 ipv6_hdr(skb)->payload_len = htons(payload_len);
289 ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
290 IP6CB(skb)->nhoff = nhoff;
291 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
292 IP6CB(skb)->frag_max_size = fq->q.max_size;
293
294 /* Yes, and fold redundant checksum back. 8) */
295 skb_postpush_rcsum(skb, skb_network_header(skb),
296 skb_network_header_len(skb));
297
298 rcu_read_lock();
299 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS);
300 rcu_read_unlock();
301 fq->q.rb_fragments = RB_ROOT;
302 fq->q.fragments_tail = NULL;
303 fq->q.last_run_head = NULL;
304 return 1;
305
306out_oversize:
307 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
308 goto out_fail;
309out_oom:
310 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
311out_fail:
312 rcu_read_lock();
313 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS);
314 rcu_read_unlock();
315 inet_frag_kill(&fq->q);
316 return -1;
317}
318
319static int ipv6_frag_rcv(struct sk_buff *skb)
320{
321 struct frag_hdr *fhdr;
322 struct frag_queue *fq;
323 const struct ipv6hdr *hdr = ipv6_hdr(skb);
324 struct net *net = dev_net(skb_dst(skb)->dev);
325 int iif;
326
327 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
328 goto fail_hdr;
329
330 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
331
332 /* Jumbo payload inhibits frag. header */
333 if (hdr->payload_len == 0)
334 goto fail_hdr;
335
336 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
337 sizeof(struct frag_hdr))))
338 goto fail_hdr;
339
340 hdr = ipv6_hdr(skb);
341 fhdr = (struct frag_hdr *)skb_transport_header(skb);
342
343 if (!(fhdr->frag_off & htons(0xFFF9))) {
344 /* It is not a fragmented frame */
345 skb->transport_header += sizeof(struct frag_hdr);
346 __IP6_INC_STATS(net,
347 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
348
349 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
350 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
351 return 1;
352 }
353
354 iif = skb->dev ? skb->dev->ifindex : 0;
355 fq = fq_find(net, fhdr->identification, hdr, iif);
356 if (fq) {
357 u32 prob_offset = 0;
358 int ret;
359
360 spin_lock(&fq->q.lock);
361
362 fq->iif = iif;
363 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
364 &prob_offset);
365
366 spin_unlock(&fq->q.lock);
367 inet_frag_put(&fq->q);
368 if (prob_offset) {
369 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
370 IPSTATS_MIB_INHDRERRORS);
371 /* icmpv6_param_prob() calls kfree_skb(skb) */
372 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
373 }
374 return ret;
375 }
376
377 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
378 kfree_skb(skb);
379 return -1;
380
381fail_hdr:
382 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
383 IPSTATS_MIB_INHDRERRORS);
384 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
385 return -1;
386}
387
388static const struct inet6_protocol frag_protocol = {
389 .handler = ipv6_frag_rcv,
390 .flags = INET6_PROTO_NOPOLICY,
391};
392
393#ifdef CONFIG_SYSCTL
394
395static struct ctl_table ip6_frags_ns_ctl_table[] = {
396 {
397 .procname = "ip6frag_high_thresh",
398 .maxlen = sizeof(unsigned long),
399 .mode = 0644,
400 .proc_handler = proc_doulongvec_minmax,
401 },
402 {
403 .procname = "ip6frag_low_thresh",
404 .maxlen = sizeof(unsigned long),
405 .mode = 0644,
406 .proc_handler = proc_doulongvec_minmax,
407 },
408 {
409 .procname = "ip6frag_time",
410 .maxlen = sizeof(int),
411 .mode = 0644,
412 .proc_handler = proc_dointvec_jiffies,
413 },
414 { }
415};
416
417/* secret interval has been deprecated */
418static int ip6_frags_secret_interval_unused;
419static struct ctl_table ip6_frags_ctl_table[] = {
420 {
421 .procname = "ip6frag_secret_interval",
422 .data = &ip6_frags_secret_interval_unused,
423 .maxlen = sizeof(int),
424 .mode = 0644,
425 .proc_handler = proc_dointvec_jiffies,
426 },
427 { }
428};
429
430static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
431{
432 struct ctl_table *table;
433 struct ctl_table_header *hdr;
434
435 table = ip6_frags_ns_ctl_table;
436 if (!net_eq(net, &init_net)) {
437 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
438 if (!table)
439 goto err_alloc;
440
441 }
442 table[0].data = &net->ipv6.fqdir->high_thresh;
443 table[0].extra1 = &net->ipv6.fqdir->low_thresh;
444 table[1].data = &net->ipv6.fqdir->low_thresh;
445 table[1].extra2 = &net->ipv6.fqdir->high_thresh;
446 table[2].data = &net->ipv6.fqdir->timeout;
447
448 hdr = register_net_sysctl(net, "net/ipv6", table);
449 if (!hdr)
450 goto err_reg;
451
452 net->ipv6.sysctl.frags_hdr = hdr;
453 return 0;
454
455err_reg:
456 if (!net_eq(net, &init_net))
457 kfree(table);
458err_alloc:
459 return -ENOMEM;
460}
461
462static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
463{
464 struct ctl_table *table;
465
466 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
467 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
468 if (!net_eq(net, &init_net))
469 kfree(table);
470}
471
472static struct ctl_table_header *ip6_ctl_header;
473
474static int ip6_frags_sysctl_register(void)
475{
476 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
477 ip6_frags_ctl_table);
478 return ip6_ctl_header == NULL ? -ENOMEM : 0;
479}
480
481static void ip6_frags_sysctl_unregister(void)
482{
483 unregister_net_sysctl_table(ip6_ctl_header);
484}
485#else
486static int ip6_frags_ns_sysctl_register(struct net *net)
487{
488 return 0;
489}
490
491static void ip6_frags_ns_sysctl_unregister(struct net *net)
492{
493}
494
495static int ip6_frags_sysctl_register(void)
496{
497 return 0;
498}
499
500static void ip6_frags_sysctl_unregister(void)
501{
502}
503#endif
504
505static int __net_init ipv6_frags_init_net(struct net *net)
506{
507 int res;
508
509 res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
510 if (res < 0)
511 return res;
512
513 net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
514 net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
515 net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
516
517 res = ip6_frags_ns_sysctl_register(net);
518 if (res < 0)
519 fqdir_exit(net->ipv6.fqdir);
520 return res;
521}
522
523static void __net_exit ipv6_frags_pre_exit_net(struct net *net)
524{
525 fqdir_pre_exit(net->ipv6.fqdir);
526}
527
528static void __net_exit ipv6_frags_exit_net(struct net *net)
529{
530 ip6_frags_ns_sysctl_unregister(net);
531 fqdir_exit(net->ipv6.fqdir);
532}
533
534static struct pernet_operations ip6_frags_ops = {
535 .init = ipv6_frags_init_net,
536 .pre_exit = ipv6_frags_pre_exit_net,
537 .exit = ipv6_frags_exit_net,
538};
539
540static const struct rhashtable_params ip6_rhash_params = {
541 .head_offset = offsetof(struct inet_frag_queue, node),
542 .hashfn = ip6frag_key_hashfn,
543 .obj_hashfn = ip6frag_obj_hashfn,
544 .obj_cmpfn = ip6frag_obj_cmpfn,
545 .automatic_shrinking = true,
546};
547
548int __init ipv6_frag_init(void)
549{
550 int ret;
551
552 ip6_frags.constructor = ip6frag_init;
553 ip6_frags.destructor = NULL;
554 ip6_frags.qsize = sizeof(struct frag_queue);
555 ip6_frags.frag_expire = ip6_frag_expire;
556 ip6_frags.frags_cache_name = ip6_frag_cache_name;
557 ip6_frags.rhash_params = ip6_rhash_params;
558 ret = inet_frags_init(&ip6_frags);
559 if (ret)
560 goto out;
561
562 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
563 if (ret)
564 goto err_protocol;
565
566 ret = ip6_frags_sysctl_register();
567 if (ret)
568 goto err_sysctl;
569
570 ret = register_pernet_subsys(&ip6_frags_ops);
571 if (ret)
572 goto err_pernet;
573
574out:
575 return ret;
576
577err_pernet:
578 ip6_frags_sysctl_unregister();
579err_sysctl:
580 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
581err_protocol:
582 inet_frags_fini(&ip6_frags);
583 goto out;
584}
585
586void ipv6_frag_exit(void)
587{
588 ip6_frags_sysctl_unregister();
589 unregister_pernet_subsys(&ip6_frags_ops);
590 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
591 inet_frags_fini(&ip6_frags);
592}
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}