1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * IPv6 fragment reassembly
4 * Linux INET6 implementation
9 * Based on: net/ipv4/ip_fragment.c
14 * Andi Kleen Make it work with multiple hosts.
15 * More RFC compliance.
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{}.
22 * YOSHIFUJI,H. @USAGI Always remove fragment header to
23 * calculate ICV correctly.
26 #define pr_fmt(fmt) "IPv6: " fmt
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>
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>
61 static const char ip6_frag_cache_name[] = "ip6-frags";
63 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
65 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
68 static struct inet_frags ip6_frags;
70 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
71 struct sk_buff *prev_tail, struct net_device *dev);
73 static void ip6_frag_expire(struct timer_list *t)
75 struct inet_frag_queue *frag = from_timer(frag, t, timer);
76 struct frag_queue *fq;
78 fq = container_of(frag, struct frag_queue, q);
80 ip6frag_expire_frag_queue(fq->q.fqdir->net, fq);
83 static struct frag_queue *
84 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
86 struct frag_v6_compare_key key = {
90 .user = IP6_DEFRAG_LOCAL_DELIVER,
93 struct inet_frag_queue *q;
95 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
96 IPV6_ADDR_LINKLOCAL)))
99 q = inet_frag_find(net->ipv6.fqdir, &key);
103 return container_of(q, struct frag_queue, q);
106 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
107 struct frag_hdr *fhdr, int nhoff,
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;
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);
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.mono_delivery_time = skb->mono_delivery_time;
202 fq->q.meat += skb->len;
204 add_frag_mem_limit(fq->q.fqdir, skb->truesize);
206 fragsize = -skb_network_offset(skb) + skb->len;
207 if (fragsize > fq->q.max_size)
208 fq->q.max_size = fragsize;
210 /* The first fragment.
211 * nhoffset is obtained from the first fragment, of course.
214 fq->nhoffset = nhoff;
215 fq->q.flags |= INET_FRAG_FIRST_IN;
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;
222 skb->_skb_refdst = 0UL;
223 err = ip6_frag_reasm(fq, skb, prev_tail, dev);
224 skb->_skb_refdst = orefdst;
232 if (err == IPFRAG_DUP) {
233 SKB_DR_SET(reason, DUP_FRAG);
238 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
239 IPSTATS_MIB_REASM_OVERLAPS);
241 inet_frag_kill(&fq->q);
242 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
243 IPSTATS_MIB_REASMFAILS);
245 kfree_skb_reason(skb, reason);
250 * Check if this packet is complete.
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.
256 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
257 struct sk_buff *prev_tail, struct net_device *dev)
259 struct net *net = fq->q.fqdir->net;
265 inet_frag_kill(&fq->q);
267 ecn = ip_frag_ecn_table[fq->ecn];
268 if (unlikely(ecn == 0xff))
271 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
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)
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);
291 skb_reset_transport_header(skb);
293 inet_frag_reasm_finish(&fq->q, skb, reasm_data, true);
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;
302 /* Yes, and fold redundant checksum back. 8) */
303 skb_postpush_rcsum(skb, skb_network_header(skb),
304 skb_network_header_len(skb));
307 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS);
309 fq->q.rb_fragments = RB_ROOT;
310 fq->q.fragments_tail = NULL;
311 fq->q.last_run_head = NULL;
315 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
318 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
321 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS);
323 inet_frag_kill(&fq->q);
327 static int ipv6_frag_rcv(struct sk_buff *skb)
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);
336 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
339 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
341 /* Jumbo payload inhibits frag. header */
342 if (hdr->payload_len == 0)
345 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
346 sizeof(struct frag_hdr))))
350 fhdr = (struct frag_hdr *)skb_transport_header(skb);
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);
356 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
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);
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.
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);
379 iif = skb->dev ? skb->dev->ifindex : 0;
380 fq = fq_find(net, fhdr->identification, hdr, iif);
385 spin_lock(&fq->q.lock);
388 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
391 spin_unlock(&fq->q.lock);
392 inet_frag_put(&fq->q);
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);
402 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
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));
413 static const struct inet6_protocol frag_protocol = {
414 .handler = ipv6_frag_rcv,
415 .flags = INET6_PROTO_NOPOLICY,
420 static struct ctl_table ip6_frags_ns_ctl_table[] = {
422 .procname = "ip6frag_high_thresh",
423 .maxlen = sizeof(unsigned long),
425 .proc_handler = proc_doulongvec_minmax,
428 .procname = "ip6frag_low_thresh",
429 .maxlen = sizeof(unsigned long),
431 .proc_handler = proc_doulongvec_minmax,
434 .procname = "ip6frag_time",
435 .maxlen = sizeof(int),
437 .proc_handler = proc_dointvec_jiffies,
442 /* secret interval has been deprecated */
443 static int ip6_frags_secret_interval_unused;
444 static struct ctl_table ip6_frags_ctl_table[] = {
446 .procname = "ip6frag_secret_interval",
447 .data = &ip6_frags_secret_interval_unused,
448 .maxlen = sizeof(int),
450 .proc_handler = proc_dointvec_jiffies,
455 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
457 struct ctl_table *table;
458 struct ctl_table_header *hdr;
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);
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;
473 hdr = register_net_sysctl(net, "net/ipv6", table);
477 net->ipv6.sysctl.frags_hdr = hdr;
481 if (!net_eq(net, &init_net))
487 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
489 struct ctl_table *table;
491 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
492 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
493 if (!net_eq(net, &init_net))
497 static struct ctl_table_header *ip6_ctl_header;
499 static int ip6_frags_sysctl_register(void)
501 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
502 ip6_frags_ctl_table);
503 return ip6_ctl_header == NULL ? -ENOMEM : 0;
506 static void ip6_frags_sysctl_unregister(void)
508 unregister_net_sysctl_table(ip6_ctl_header);
511 static int ip6_frags_ns_sysctl_register(struct net *net)
516 static void ip6_frags_ns_sysctl_unregister(struct net *net)
520 static int ip6_frags_sysctl_register(void)
525 static void ip6_frags_sysctl_unregister(void)
530 static int __net_init ipv6_frags_init_net(struct net *net)
534 res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
538 net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
539 net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
540 net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
542 res = ip6_frags_ns_sysctl_register(net);
544 fqdir_exit(net->ipv6.fqdir);
548 static void __net_exit ipv6_frags_pre_exit_net(struct net *net)
550 fqdir_pre_exit(net->ipv6.fqdir);
553 static void __net_exit ipv6_frags_exit_net(struct net *net)
555 ip6_frags_ns_sysctl_unregister(net);
556 fqdir_exit(net->ipv6.fqdir);
559 static struct pernet_operations ip6_frags_ops = {
560 .init = ipv6_frags_init_net,
561 .pre_exit = ipv6_frags_pre_exit_net,
562 .exit = ipv6_frags_exit_net,
565 static const struct rhashtable_params ip6_rhash_params = {
566 .head_offset = offsetof(struct inet_frag_queue, node),
567 .hashfn = ip6frag_key_hashfn,
568 .obj_hashfn = ip6frag_obj_hashfn,
569 .obj_cmpfn = ip6frag_obj_cmpfn,
570 .automatic_shrinking = true,
573 int __init ipv6_frag_init(void)
577 ip6_frags.constructor = ip6frag_init;
578 ip6_frags.destructor = NULL;
579 ip6_frags.qsize = sizeof(struct frag_queue);
580 ip6_frags.frag_expire = ip6_frag_expire;
581 ip6_frags.frags_cache_name = ip6_frag_cache_name;
582 ip6_frags.rhash_params = ip6_rhash_params;
583 ret = inet_frags_init(&ip6_frags);
587 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
591 ret = ip6_frags_sysctl_register();
595 ret = register_pernet_subsys(&ip6_frags_ops);
603 ip6_frags_sysctl_unregister();
605 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
607 inet_frags_fini(&ip6_frags);
611 void ipv6_frag_exit(void)
613 ip6_frags_sysctl_unregister();
614 unregister_pernet_subsys(&ip6_frags_ops);
615 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
616 inet_frags_fini(&ip6_frags);