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// 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#include <linux/tcp.h>
46#include <linux/udp.h>
47
48#include <net/sock.h>
49#include <net/snmp.h>
50
51#include <net/ipv6.h>
52#include <net/ip6_route.h>
53#include <net/protocol.h>
54#include <net/transp_v6.h>
55#include <net/rawv6.h>
56#include <net/ndisc.h>
57#include <net/addrconf.h>
58#include <net/ipv6_frag.h>
59#include <net/inet_ecn.h>
60
61static const char ip6_frag_cache_name[] = "ip6-frags";
62
63static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
64{
65 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
66}
67
68static struct inet_frags ip6_frags;
69
70static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
71 struct sk_buff *prev_tail, struct net_device *dev);
72
73static void ip6_frag_expire(struct timer_list *t)
74{
75 struct inet_frag_queue *frag = from_timer(frag, t, timer);
76 struct frag_queue *fq;
77
78 fq = container_of(frag, struct frag_queue, q);
79
80 ip6frag_expire_frag_queue(fq->q.fqdir->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.fqdir, &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 SKB_DR(reason);
116 u8 ecn;
117
118 /* If reassembly is already done, @skb must be a duplicate frag. */
119 if (fq->q.flags & INET_FRAG_COMPLETE) {
120 SKB_DR_SET(reason, DUP_FRAG);
121 goto err;
122 }
123
124 err = -EINVAL;
125 offset = ntohs(fhdr->frag_off) & ~0x7;
126 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
127 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
128
129 if ((unsigned int)end > IPV6_MAXPLEN) {
130 *prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
131 /* note that if prob_offset is set, the skb is freed elsewhere,
132 * we do not free it here.
133 */
134 return -1;
135 }
136
137 ecn = ip6_frag_ecn(ipv6_hdr(skb));
138
139 if (skb->ip_summed == CHECKSUM_COMPLETE) {
140 const unsigned char *nh = skb_network_header(skb);
141 skb->csum = csum_sub(skb->csum,
142 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
143 0));
144 }
145
146 /* Is this the final fragment? */
147 if (!(fhdr->frag_off & htons(IP6_MF))) {
148 /* If we already have some bits beyond end
149 * or have different end, the segment is corrupted.
150 */
151 if (end < fq->q.len ||
152 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
153 goto discard_fq;
154 fq->q.flags |= INET_FRAG_LAST_IN;
155 fq->q.len = end;
156 } else {
157 /* Check if the fragment is rounded to 8 bytes.
158 * Required by the RFC.
159 */
160 if (end & 0x7) {
161 /* RFC2460 says always send parameter problem in
162 * this case. -DaveM
163 */
164 *prob_offset = offsetof(struct ipv6hdr, payload_len);
165 return -1;
166 }
167 if (end > fq->q.len) {
168 /* Some bits beyond end -> corruption. */
169 if (fq->q.flags & INET_FRAG_LAST_IN)
170 goto discard_fq;
171 fq->q.len = end;
172 }
173 }
174
175 if (end == offset)
176 goto discard_fq;
177
178 err = -ENOMEM;
179 /* Point into the IP datagram 'data' part. */
180 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
181 goto discard_fq;
182
183 err = pskb_trim_rcsum(skb, end - offset);
184 if (err)
185 goto discard_fq;
186
187 /* Note : skb->rbnode and skb->dev share the same location. */
188 dev = skb->dev;
189 /* Makes sure compiler wont do silly aliasing games */
190 barrier();
191
192 prev_tail = fq->q.fragments_tail;
193 err = inet_frag_queue_insert(&fq->q, skb, offset, end);
194 if (err)
195 goto insert_error;
196
197 if (dev)
198 fq->iif = dev->ifindex;
199
200 fq->q.stamp = skb->tstamp;
201 fq->q.mono_delivery_time = skb->mono_delivery_time;
202 fq->q.meat += skb->len;
203 fq->ecn |= ecn;
204 add_frag_mem_limit(fq->q.fqdir, skb->truesize);
205
206 fragsize = -skb_network_offset(skb) + skb->len;
207 if (fragsize > fq->q.max_size)
208 fq->q.max_size = fragsize;
209
210 /* The first fragment.
211 * nhoffset is obtained from the first fragment, of course.
212 */
213 if (offset == 0) {
214 fq->nhoffset = nhoff;
215 fq->q.flags |= INET_FRAG_FIRST_IN;
216 }
217
218 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
219 fq->q.meat == fq->q.len) {
220 unsigned long orefdst = skb->_skb_refdst;
221
222 skb->_skb_refdst = 0UL;
223 err = ip6_frag_reasm(fq, skb, prev_tail, dev);
224 skb->_skb_refdst = orefdst;
225 return err;
226 }
227
228 skb_dst_drop(skb);
229 return -EINPROGRESS;
230
231insert_error:
232 if (err == IPFRAG_DUP) {
233 SKB_DR_SET(reason, DUP_FRAG);
234 err = -EINVAL;
235 goto err;
236 }
237 err = -EINVAL;
238 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
239 IPSTATS_MIB_REASM_OVERLAPS);
240discard_fq:
241 inet_frag_kill(&fq->q);
242 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
243 IPSTATS_MIB_REASMFAILS);
244err:
245 kfree_skb_reason(skb, reason);
246 return err;
247}
248
249/*
250 * Check if this packet is complete.
251 *
252 * It is called with locked fq, and caller must check that
253 * queue is eligible for reassembly i.e. it is not COMPLETE,
254 * the last and the first frames arrived and all the bits are here.
255 */
256static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
257 struct sk_buff *prev_tail, struct net_device *dev)
258{
259 struct net *net = fq->q.fqdir->net;
260 unsigned int nhoff;
261 void *reasm_data;
262 int payload_len;
263 u8 ecn;
264
265 inet_frag_kill(&fq->q);
266
267 ecn = ip_frag_ecn_table[fq->ecn];
268 if (unlikely(ecn == 0xff))
269 goto out_fail;
270
271 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
272 if (!reasm_data)
273 goto out_oom;
274
275 payload_len = ((skb->data - skb_network_header(skb)) -
276 sizeof(struct ipv6hdr) + fq->q.len -
277 sizeof(struct frag_hdr));
278 if (payload_len > IPV6_MAXPLEN)
279 goto out_oversize;
280
281 /* We have to remove fragment header from datagram and to relocate
282 * header in order to calculate ICV correctly. */
283 nhoff = fq->nhoffset;
284 skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
285 memmove(skb->head + sizeof(struct frag_hdr), skb->head,
286 (skb->data - skb->head) - sizeof(struct frag_hdr));
287 if (skb_mac_header_was_set(skb))
288 skb->mac_header += sizeof(struct frag_hdr);
289 skb->network_header += sizeof(struct frag_hdr);
290
291 skb_reset_transport_header(skb);
292
293 inet_frag_reasm_finish(&fq->q, skb, reasm_data, true);
294
295 skb->dev = dev;
296 ipv6_hdr(skb)->payload_len = htons(payload_len);
297 ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
298 IP6CB(skb)->nhoff = nhoff;
299 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
300 IP6CB(skb)->frag_max_size = fq->q.max_size;
301
302 /* Yes, and fold redundant checksum back. 8) */
303 skb_postpush_rcsum(skb, skb_network_header(skb),
304 skb_network_header_len(skb));
305
306 rcu_read_lock();
307 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS);
308 rcu_read_unlock();
309 fq->q.rb_fragments = RB_ROOT;
310 fq->q.fragments_tail = NULL;
311 fq->q.last_run_head = NULL;
312 return 1;
313
314out_oversize:
315 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
316 goto out_fail;
317out_oom:
318 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
319out_fail:
320 rcu_read_lock();
321 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS);
322 rcu_read_unlock();
323 inet_frag_kill(&fq->q);
324 return -1;
325}
326
327static int ipv6_frag_rcv(struct sk_buff *skb)
328{
329 struct frag_hdr *fhdr;
330 struct frag_queue *fq;
331 const struct ipv6hdr *hdr = ipv6_hdr(skb);
332 struct net *net = dev_net(skb_dst(skb)->dev);
333 u8 nexthdr;
334 int iif;
335
336 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
337 goto fail_hdr;
338
339 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
340
341 /* Jumbo payload inhibits frag. header */
342 if (hdr->payload_len == 0)
343 goto fail_hdr;
344
345 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
346 sizeof(struct frag_hdr))))
347 goto fail_hdr;
348
349 hdr = ipv6_hdr(skb);
350 fhdr = (struct frag_hdr *)skb_transport_header(skb);
351
352 if (!(fhdr->frag_off & htons(IP6_OFFSET | IP6_MF))) {
353 /* It is not a fragmented frame */
354 skb->transport_header += sizeof(struct frag_hdr);
355 __IP6_INC_STATS(net,
356 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
357
358 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
359 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
360 IP6CB(skb)->frag_max_size = ntohs(hdr->payload_len) +
361 sizeof(struct ipv6hdr);
362 return 1;
363 }
364
365 /* RFC 8200, Section 4.5 Fragment Header:
366 * If the first fragment does not include all headers through an
367 * Upper-Layer header, then that fragment should be discarded and
368 * an ICMP Parameter Problem, Code 3, message should be sent to
369 * the source of the fragment, with the Pointer field set to zero.
370 */
371 nexthdr = hdr->nexthdr;
372 if (ipv6frag_thdr_truncated(skb, skb_transport_offset(skb), &nexthdr)) {
373 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
374 IPSTATS_MIB_INHDRERRORS);
375 icmpv6_param_prob(skb, ICMPV6_HDR_INCOMP, 0);
376 return -1;
377 }
378
379 iif = skb->dev ? skb->dev->ifindex : 0;
380 fq = fq_find(net, fhdr->identification, hdr, iif);
381 if (fq) {
382 u32 prob_offset = 0;
383 int ret;
384
385 spin_lock(&fq->q.lock);
386
387 fq->iif = iif;
388 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
389 &prob_offset);
390
391 spin_unlock(&fq->q.lock);
392 inet_frag_put(&fq->q);
393 if (prob_offset) {
394 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
395 IPSTATS_MIB_INHDRERRORS);
396 /* icmpv6_param_prob() calls kfree_skb(skb) */
397 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
398 }
399 return ret;
400 }
401
402 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
403 kfree_skb(skb);
404 return -1;
405
406fail_hdr:
407 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
408 IPSTATS_MIB_INHDRERRORS);
409 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
410 return -1;
411}
412
413static const struct inet6_protocol frag_protocol = {
414 .handler = ipv6_frag_rcv,
415 .flags = INET6_PROTO_NOPOLICY,
416};
417
418#ifdef CONFIG_SYSCTL
419
420static struct ctl_table ip6_frags_ns_ctl_table[] = {
421 {
422 .procname = "ip6frag_high_thresh",
423 .maxlen = sizeof(unsigned long),
424 .mode = 0644,
425 .proc_handler = proc_doulongvec_minmax,
426 },
427 {
428 .procname = "ip6frag_low_thresh",
429 .maxlen = sizeof(unsigned long),
430 .mode = 0644,
431 .proc_handler = proc_doulongvec_minmax,
432 },
433 {
434 .procname = "ip6frag_time",
435 .maxlen = sizeof(int),
436 .mode = 0644,
437 .proc_handler = proc_dointvec_jiffies,
438 },
439 { }
440};
441
442/* secret interval has been deprecated */
443static int ip6_frags_secret_interval_unused;
444static struct ctl_table ip6_frags_ctl_table[] = {
445 {
446 .procname = "ip6frag_secret_interval",
447 .data = &ip6_frags_secret_interval_unused,
448 .maxlen = sizeof(int),
449 .mode = 0644,
450 .proc_handler = proc_dointvec_jiffies,
451 },
452 { }
453};
454
455static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
456{
457 struct ctl_table *table;
458 struct ctl_table_header *hdr;
459
460 table = ip6_frags_ns_ctl_table;
461 if (!net_eq(net, &init_net)) {
462 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
463 if (!table)
464 goto err_alloc;
465
466 }
467 table[0].data = &net->ipv6.fqdir->high_thresh;
468 table[0].extra1 = &net->ipv6.fqdir->low_thresh;
469 table[1].data = &net->ipv6.fqdir->low_thresh;
470 table[1].extra2 = &net->ipv6.fqdir->high_thresh;
471 table[2].data = &net->ipv6.fqdir->timeout;
472
473 hdr = register_net_sysctl_sz(net, "net/ipv6", table,
474 ARRAY_SIZE(ip6_frags_ns_ctl_table));
475 if (!hdr)
476 goto err_reg;
477
478 net->ipv6.sysctl.frags_hdr = hdr;
479 return 0;
480
481err_reg:
482 if (!net_eq(net, &init_net))
483 kfree(table);
484err_alloc:
485 return -ENOMEM;
486}
487
488static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
489{
490 struct ctl_table *table;
491
492 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
493 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
494 if (!net_eq(net, &init_net))
495 kfree(table);
496}
497
498static struct ctl_table_header *ip6_ctl_header;
499
500static int ip6_frags_sysctl_register(void)
501{
502 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
503 ip6_frags_ctl_table);
504 return ip6_ctl_header == NULL ? -ENOMEM : 0;
505}
506
507static void ip6_frags_sysctl_unregister(void)
508{
509 unregister_net_sysctl_table(ip6_ctl_header);
510}
511#else
512static int ip6_frags_ns_sysctl_register(struct net *net)
513{
514 return 0;
515}
516
517static void ip6_frags_ns_sysctl_unregister(struct net *net)
518{
519}
520
521static int ip6_frags_sysctl_register(void)
522{
523 return 0;
524}
525
526static void ip6_frags_sysctl_unregister(void)
527{
528}
529#endif
530
531static int __net_init ipv6_frags_init_net(struct net *net)
532{
533 int res;
534
535 res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
536 if (res < 0)
537 return res;
538
539 net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
540 net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
541 net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
542
543 res = ip6_frags_ns_sysctl_register(net);
544 if (res < 0)
545 fqdir_exit(net->ipv6.fqdir);
546 return res;
547}
548
549static void __net_exit ipv6_frags_pre_exit_net(struct net *net)
550{
551 fqdir_pre_exit(net->ipv6.fqdir);
552}
553
554static void __net_exit ipv6_frags_exit_net(struct net *net)
555{
556 ip6_frags_ns_sysctl_unregister(net);
557 fqdir_exit(net->ipv6.fqdir);
558}
559
560static struct pernet_operations ip6_frags_ops = {
561 .init = ipv6_frags_init_net,
562 .pre_exit = ipv6_frags_pre_exit_net,
563 .exit = ipv6_frags_exit_net,
564};
565
566static const struct rhashtable_params ip6_rhash_params = {
567 .head_offset = offsetof(struct inet_frag_queue, node),
568 .hashfn = ip6frag_key_hashfn,
569 .obj_hashfn = ip6frag_obj_hashfn,
570 .obj_cmpfn = ip6frag_obj_cmpfn,
571 .automatic_shrinking = true,
572};
573
574int __init ipv6_frag_init(void)
575{
576 int ret;
577
578 ip6_frags.constructor = ip6frag_init;
579 ip6_frags.destructor = NULL;
580 ip6_frags.qsize = sizeof(struct frag_queue);
581 ip6_frags.frag_expire = ip6_frag_expire;
582 ip6_frags.frags_cache_name = ip6_frag_cache_name;
583 ip6_frags.rhash_params = ip6_rhash_params;
584 ret = inet_frags_init(&ip6_frags);
585 if (ret)
586 goto out;
587
588 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
589 if (ret)
590 goto err_protocol;
591
592 ret = ip6_frags_sysctl_register();
593 if (ret)
594 goto err_sysctl;
595
596 ret = register_pernet_subsys(&ip6_frags_ops);
597 if (ret)
598 goto err_pernet;
599
600out:
601 return ret;
602
603err_pernet:
604 ip6_frags_sysctl_unregister();
605err_sysctl:
606 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
607err_protocol:
608 inet_frags_fini(&ip6_frags);
609 goto out;
610}
611
612void ipv6_frag_exit(void)
613{
614 ip6_frags_sysctl_unregister();
615 unregister_pernet_subsys(&ip6_frags_ops);
616 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
617 inet_frags_fini(&ip6_frags);
618}