2 * IPv6 fragment reassembly
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on: net/ipv4/ip_fragment.c
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.
18 * Andi Kleen Make it work with multiple hosts.
19 * More RFC compliance.
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{}.
26 * YOSHIFUJI,H. @USAGI Always remove fragment header to
27 * calculate ICV correctly.
30 #define pr_fmt(fmt) "IPv6: " fmt
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>
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/ipv6_frag.h>
61 #include <net/inet_ecn.h>
63 static const char ip6_frag_cache_name[] = "ip6-frags";
65 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
67 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
70 static struct inet_frags ip6_frags;
72 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
73 struct sk_buff *prev_tail, struct net_device *dev);
75 static void ip6_frag_expire(struct timer_list *t)
77 struct inet_frag_queue *frag = from_timer(frag, t, timer);
78 struct frag_queue *fq;
81 fq = container_of(frag, struct frag_queue, q);
82 net = container_of(fq->q.net, struct net, ipv6.frags);
84 ip6frag_expire_frag_queue(net, fq);
87 static struct frag_queue *
88 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
90 struct frag_v6_compare_key key = {
94 .user = IP6_DEFRAG_LOCAL_DELIVER,
97 struct inet_frag_queue *q;
99 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
100 IPV6_ADDR_LINKLOCAL)))
103 q = inet_frag_find(&net->ipv6.frags, &key);
107 return container_of(q, struct frag_queue, q);
110 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
111 struct frag_hdr *fhdr, int nhoff,
114 struct net *net = dev_net(skb_dst(skb)->dev);
115 int offset, end, fragsize;
116 struct sk_buff *prev_tail;
117 struct net_device *dev;
121 if (fq->q.flags & INET_FRAG_COMPLETE)
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)));
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.
137 ecn = ip6_frag_ecn(ipv6_hdr(skb));
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,
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.
151 if (end < fq->q.len ||
152 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
154 fq->q.flags |= INET_FRAG_LAST_IN;
157 /* Check if the fragment is rounded to 8 bytes.
158 * Required by the RFC.
161 /* RFC2460 says always send parameter problem in
164 *prob_offset = offsetof(struct ipv6hdr, payload_len);
167 if (end > fq->q.len) {
168 /* Some bits beyond end -> corruption. */
169 if (fq->q.flags & INET_FRAG_LAST_IN)
179 /* Point into the IP datagram 'data' part. */
180 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
183 err = pskb_trim_rcsum(skb, end - offset);
187 /* Note : skb->rbnode and skb->dev share the same location. */
189 /* Makes sure compiler wont do silly aliasing games */
192 prev_tail = fq->q.fragments_tail;
193 err = inet_frag_queue_insert(&fq->q, skb, offset, end);
198 fq->iif = dev->ifindex;
200 fq->q.stamp = skb->tstamp;
201 fq->q.meat += skb->len;
203 add_frag_mem_limit(fq->q.net, skb->truesize);
205 fragsize = -skb_network_offset(skb) + skb->len;
206 if (fragsize > fq->q.max_size)
207 fq->q.max_size = fragsize;
209 /* The first fragment.
210 * nhoffset is obtained from the first fragment, of course.
213 fq->nhoffset = nhoff;
214 fq->q.flags |= INET_FRAG_FIRST_IN;
217 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
218 fq->q.meat == fq->q.len) {
219 unsigned long orefdst = skb->_skb_refdst;
221 skb->_skb_refdst = 0UL;
222 err = ip6_frag_reasm(fq, skb, prev_tail, dev);
223 skb->_skb_refdst = orefdst;
231 if (err == IPFRAG_DUP) {
236 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
237 IPSTATS_MIB_REASM_OVERLAPS);
239 inet_frag_kill(&fq->q);
240 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
241 IPSTATS_MIB_REASMFAILS);
248 * Check if this packet is complete.
250 * It is called with locked fq, and caller must check that
251 * queue is eligible for reassembly i.e. it is not COMPLETE,
252 * the last and the first frames arrived and all the bits are here.
254 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
255 struct sk_buff *prev_tail, struct net_device *dev)
257 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
263 inet_frag_kill(&fq->q);
265 ecn = ip_frag_ecn_table[fq->ecn];
266 if (unlikely(ecn == 0xff))
269 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
273 payload_len = ((skb->data - skb_network_header(skb)) -
274 sizeof(struct ipv6hdr) + fq->q.len -
275 sizeof(struct frag_hdr));
276 if (payload_len > IPV6_MAXPLEN)
279 /* We have to remove fragment header from datagram and to relocate
280 * header in order to calculate ICV correctly. */
281 nhoff = fq->nhoffset;
282 skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
283 memmove(skb->head + sizeof(struct frag_hdr), skb->head,
284 (skb->data - skb->head) - sizeof(struct frag_hdr));
285 if (skb_mac_header_was_set(skb))
286 skb->mac_header += sizeof(struct frag_hdr);
287 skb->network_header += sizeof(struct frag_hdr);
289 skb_reset_transport_header(skb);
291 inet_frag_reasm_finish(&fq->q, skb, reasm_data);
294 ipv6_hdr(skb)->payload_len = htons(payload_len);
295 ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
296 IP6CB(skb)->nhoff = nhoff;
297 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
298 IP6CB(skb)->frag_max_size = fq->q.max_size;
300 /* Yes, and fold redundant checksum back. 8) */
301 skb_postpush_rcsum(skb, skb_network_header(skb),
302 skb_network_header_len(skb));
305 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
307 fq->q.rb_fragments = RB_ROOT;
308 fq->q.fragments_tail = NULL;
309 fq->q.last_run_head = NULL;
313 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
316 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
319 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
321 inet_frag_kill(&fq->q);
325 static int ipv6_frag_rcv(struct sk_buff *skb)
327 struct frag_hdr *fhdr;
328 struct frag_queue *fq;
329 const struct ipv6hdr *hdr = ipv6_hdr(skb);
330 struct net *net = dev_net(skb_dst(skb)->dev);
333 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
336 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
338 /* Jumbo payload inhibits frag. header */
339 if (hdr->payload_len == 0)
342 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
343 sizeof(struct frag_hdr))))
347 fhdr = (struct frag_hdr *)skb_transport_header(skb);
349 if (!(fhdr->frag_off & htons(0xFFF9))) {
350 /* It is not a fragmented frame */
351 skb->transport_header += sizeof(struct frag_hdr);
353 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
355 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
356 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
360 iif = skb->dev ? skb->dev->ifindex : 0;
361 fq = fq_find(net, fhdr->identification, hdr, iif);
366 spin_lock(&fq->q.lock);
369 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
372 spin_unlock(&fq->q.lock);
373 inet_frag_put(&fq->q);
375 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
376 IPSTATS_MIB_INHDRERRORS);
377 /* icmpv6_param_prob() calls kfree_skb(skb) */
378 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
383 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
388 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
389 IPSTATS_MIB_INHDRERRORS);
390 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
394 static const struct inet6_protocol frag_protocol = {
395 .handler = ipv6_frag_rcv,
396 .flags = INET6_PROTO_NOPOLICY,
401 static struct ctl_table ip6_frags_ns_ctl_table[] = {
403 .procname = "ip6frag_high_thresh",
404 .data = &init_net.ipv6.frags.high_thresh,
405 .maxlen = sizeof(unsigned long),
407 .proc_handler = proc_doulongvec_minmax,
408 .extra1 = &init_net.ipv6.frags.low_thresh
411 .procname = "ip6frag_low_thresh",
412 .data = &init_net.ipv6.frags.low_thresh,
413 .maxlen = sizeof(unsigned long),
415 .proc_handler = proc_doulongvec_minmax,
416 .extra2 = &init_net.ipv6.frags.high_thresh
419 .procname = "ip6frag_time",
420 .data = &init_net.ipv6.frags.timeout,
421 .maxlen = sizeof(int),
423 .proc_handler = proc_dointvec_jiffies,
428 /* secret interval has been deprecated */
429 static int ip6_frags_secret_interval_unused;
430 static struct ctl_table ip6_frags_ctl_table[] = {
432 .procname = "ip6frag_secret_interval",
433 .data = &ip6_frags_secret_interval_unused,
434 .maxlen = sizeof(int),
436 .proc_handler = proc_dointvec_jiffies,
441 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
443 struct ctl_table *table;
444 struct ctl_table_header *hdr;
446 table = ip6_frags_ns_ctl_table;
447 if (!net_eq(net, &init_net)) {
448 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
452 table[0].data = &net->ipv6.frags.high_thresh;
453 table[0].extra1 = &net->ipv6.frags.low_thresh;
454 table[1].data = &net->ipv6.frags.low_thresh;
455 table[1].extra2 = &net->ipv6.frags.high_thresh;
456 table[2].data = &net->ipv6.frags.timeout;
459 hdr = register_net_sysctl(net, "net/ipv6", table);
463 net->ipv6.sysctl.frags_hdr = hdr;
467 if (!net_eq(net, &init_net))
473 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
475 struct ctl_table *table;
477 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
478 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
479 if (!net_eq(net, &init_net))
483 static struct ctl_table_header *ip6_ctl_header;
485 static int ip6_frags_sysctl_register(void)
487 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
488 ip6_frags_ctl_table);
489 return ip6_ctl_header == NULL ? -ENOMEM : 0;
492 static void ip6_frags_sysctl_unregister(void)
494 unregister_net_sysctl_table(ip6_ctl_header);
497 static int ip6_frags_ns_sysctl_register(struct net *net)
502 static void ip6_frags_ns_sysctl_unregister(struct net *net)
506 static int ip6_frags_sysctl_register(void)
511 static void ip6_frags_sysctl_unregister(void)
516 static int __net_init ipv6_frags_init_net(struct net *net)
520 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
521 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
522 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
523 net->ipv6.frags.f = &ip6_frags;
525 res = inet_frags_init_net(&net->ipv6.frags);
529 res = ip6_frags_ns_sysctl_register(net);
531 inet_frags_exit_net(&net->ipv6.frags);
535 static void __net_exit ipv6_frags_exit_net(struct net *net)
537 ip6_frags_ns_sysctl_unregister(net);
538 inet_frags_exit_net(&net->ipv6.frags);
541 static struct pernet_operations ip6_frags_ops = {
542 .init = ipv6_frags_init_net,
543 .exit = ipv6_frags_exit_net,
546 static const struct rhashtable_params ip6_rhash_params = {
547 .head_offset = offsetof(struct inet_frag_queue, node),
548 .hashfn = ip6frag_key_hashfn,
549 .obj_hashfn = ip6frag_obj_hashfn,
550 .obj_cmpfn = ip6frag_obj_cmpfn,
551 .automatic_shrinking = true,
554 int __init ipv6_frag_init(void)
558 ip6_frags.constructor = ip6frag_init;
559 ip6_frags.destructor = NULL;
560 ip6_frags.qsize = sizeof(struct frag_queue);
561 ip6_frags.frag_expire = ip6_frag_expire;
562 ip6_frags.frags_cache_name = ip6_frag_cache_name;
563 ip6_frags.rhash_params = ip6_rhash_params;
564 ret = inet_frags_init(&ip6_frags);
568 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
572 ret = ip6_frags_sysctl_register();
576 ret = register_pernet_subsys(&ip6_frags_ops);
584 ip6_frags_sysctl_unregister();
586 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
588 inet_frags_fini(&ip6_frags);
592 void ipv6_frag_exit(void)
594 inet_frags_fini(&ip6_frags);
595 ip6_frags_sysctl_unregister();
596 unregister_pernet_subsys(&ip6_frags_ops);
597 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);