xref: /openbmc/linux/net/ipv6/reassembly.c (revision 6197e5b7) 
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 
61 static const char ip6_frag_cache_name[] = "ip6-frags";
62 
63 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
64 {
65 	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
66 }
67 
68 static struct inet_frags ip6_frags;
69 
70 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
71 			  struct sk_buff *prev_tail, struct net_device *dev);
72 
73 static 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 
83 static struct frag_queue *
84 fq_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 
106 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
107 			  struct frag_hdr *fhdr, int nhoff,
108 			  u32 *prob_offset)
109 {
110 	struct net *net = dev_net(skb_dst(skb)->dev);
111 	int offset, end, fragsize;
112 	struct sk_buff *prev_tail;
113 	struct net_device *dev;
114 	int err = -ENOENT;
115 	u8 ecn;
116 
117 	if (fq->q.flags & INET_FRAG_COMPLETE)
118 		goto err;
119 
120 	err = -EINVAL;
121 	offset = ntohs(fhdr->frag_off) & ~0x7;
122 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
123 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
124 
125 	if ((unsigned int)end > IPV6_MAXPLEN) {
126 		*prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
127 		/* note that if prob_offset is set, the skb is freed elsewhere,
128 		 * we do not free it here.
129 		 */
130 		return -1;
131 	}
132 
133 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
134 
135 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
136 		const unsigned char *nh = skb_network_header(skb);
137 		skb->csum = csum_sub(skb->csum,
138 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
139 						  0));
140 	}
141 
142 	/* Is this the final fragment? */
143 	if (!(fhdr->frag_off & htons(IP6_MF))) {
144 		/* If we already have some bits beyond end
145 		 * or have different end, the segment is corrupted.
146 		 */
147 		if (end < fq->q.len ||
148 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
149 			goto discard_fq;
150 		fq->q.flags |= INET_FRAG_LAST_IN;
151 		fq->q.len = end;
152 	} else {
153 		/* Check if the fragment is rounded to 8 bytes.
154 		 * Required by the RFC.
155 		 */
156 		if (end & 0x7) {
157 			/* RFC2460 says always send parameter problem in
158 			 * this case. -DaveM
159 			 */
160 			*prob_offset = offsetof(struct ipv6hdr, payload_len);
161 			return -1;
162 		}
163 		if (end > fq->q.len) {
164 			/* Some bits beyond end -> corruption. */
165 			if (fq->q.flags & INET_FRAG_LAST_IN)
166 				goto discard_fq;
167 			fq->q.len = end;
168 		}
169 	}
170 
171 	if (end == offset)
172 		goto discard_fq;
173 
174 	err = -ENOMEM;
175 	/* Point into the IP datagram 'data' part. */
176 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
177 		goto discard_fq;
178 
179 	err = pskb_trim_rcsum(skb, end - offset);
180 	if (err)
181 		goto discard_fq;
182 
183 	/* Note : skb->rbnode and skb->dev share the same location. */
184 	dev = skb->dev;
185 	/* Makes sure compiler wont do silly aliasing games */
186 	barrier();
187 
188 	prev_tail = fq->q.fragments_tail;
189 	err = inet_frag_queue_insert(&fq->q, skb, offset, end);
190 	if (err)
191 		goto insert_error;
192 
193 	if (dev)
194 		fq->iif = dev->ifindex;
195 
196 	fq->q.stamp = skb->tstamp;
197 	fq->q.meat += skb->len;
198 	fq->ecn |= ecn;
199 	add_frag_mem_limit(fq->q.fqdir, skb->truesize);
200 
201 	fragsize = -skb_network_offset(skb) + skb->len;
202 	if (fragsize > fq->q.max_size)
203 		fq->q.max_size = fragsize;
204 
205 	/* The first fragment.
206 	 * nhoffset is obtained from the first fragment, of course.
207 	 */
208 	if (offset == 0) {
209 		fq->nhoffset = nhoff;
210 		fq->q.flags |= INET_FRAG_FIRST_IN;
211 	}
212 
213 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
214 	    fq->q.meat == fq->q.len) {
215 		unsigned long orefdst = skb->_skb_refdst;
216 
217 		skb->_skb_refdst = 0UL;
218 		err = ip6_frag_reasm(fq, skb, prev_tail, dev);
219 		skb->_skb_refdst = orefdst;
220 		return err;
221 	}
222 
223 	skb_dst_drop(skb);
224 	return -EINPROGRESS;
225 
226 insert_error:
227 	if (err == IPFRAG_DUP) {
228 		kfree_skb(skb);
229 		return -EINVAL;
230 	}
231 	err = -EINVAL;
232 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
233 			IPSTATS_MIB_REASM_OVERLAPS);
234 discard_fq:
235 	inet_frag_kill(&fq->q);
236 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
237 			IPSTATS_MIB_REASMFAILS);
238 err:
239 	kfree_skb(skb);
240 	return err;
241 }
242 
243 /*
244  *	Check if this packet is complete.
245  *
246  *	It is called with locked fq, and caller must check that
247  *	queue is eligible for reassembly i.e. it is not COMPLETE,
248  *	the last and the first frames arrived and all the bits are here.
249  */
250 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
251 			  struct sk_buff *prev_tail, struct net_device *dev)
252 {
253 	struct net *net = fq->q.fqdir->net;
254 	unsigned int nhoff;
255 	void *reasm_data;
256 	int payload_len;
257 	u8 ecn;
258 
259 	inet_frag_kill(&fq->q);
260 
261 	ecn = ip_frag_ecn_table[fq->ecn];
262 	if (unlikely(ecn == 0xff))
263 		goto out_fail;
264 
265 	reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
266 	if (!reasm_data)
267 		goto out_oom;
268 
269 	payload_len = ((skb->data - skb_network_header(skb)) -
270 		       sizeof(struct ipv6hdr) + fq->q.len -
271 		       sizeof(struct frag_hdr));
272 	if (payload_len > IPV6_MAXPLEN)
273 		goto out_oversize;
274 
275 	/* We have to remove fragment header from datagram and to relocate
276 	 * header in order to calculate ICV correctly. */
277 	nhoff = fq->nhoffset;
278 	skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
279 	memmove(skb->head + sizeof(struct frag_hdr), skb->head,
280 		(skb->data - skb->head) - sizeof(struct frag_hdr));
281 	if (skb_mac_header_was_set(skb))
282 		skb->mac_header += sizeof(struct frag_hdr);
283 	skb->network_header += sizeof(struct frag_hdr);
284 
285 	skb_reset_transport_header(skb);
286 
287 	inet_frag_reasm_finish(&fq->q, skb, reasm_data, true);
288 
289 	skb->dev = dev;
290 	ipv6_hdr(skb)->payload_len = htons(payload_len);
291 	ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
292 	IP6CB(skb)->nhoff = nhoff;
293 	IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
294 	IP6CB(skb)->frag_max_size = fq->q.max_size;
295 
296 	/* Yes, and fold redundant checksum back. 8) */
297 	skb_postpush_rcsum(skb, skb_network_header(skb),
298 			   skb_network_header_len(skb));
299 
300 	rcu_read_lock();
301 	__IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS);
302 	rcu_read_unlock();
303 	fq->q.rb_fragments = RB_ROOT;
304 	fq->q.fragments_tail = NULL;
305 	fq->q.last_run_head = NULL;
306 	return 1;
307 
308 out_oversize:
309 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
310 	goto out_fail;
311 out_oom:
312 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
313 out_fail:
314 	rcu_read_lock();
315 	__IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS);
316 	rcu_read_unlock();
317 	inet_frag_kill(&fq->q);
318 	return -1;
319 }
320 
321 static int ipv6_frag_rcv(struct sk_buff *skb)
322 {
323 	struct frag_hdr *fhdr;
324 	struct frag_queue *fq;
325 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
326 	struct net *net = dev_net(skb_dst(skb)->dev);
327 	u8 nexthdr;
328 	int iif;
329 
330 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
331 		goto fail_hdr;
332 
333 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
334 
335 	/* Jumbo payload inhibits frag. header */
336 	if (hdr->payload_len == 0)
337 		goto fail_hdr;
338 
339 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
340 				 sizeof(struct frag_hdr))))
341 		goto fail_hdr;
342 
343 	hdr = ipv6_hdr(skb);
344 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
345 
346 	if (!(fhdr->frag_off & htons(0xFFF9))) {
347 		/* It is not a fragmented frame */
348 		skb->transport_header += sizeof(struct frag_hdr);
349 		__IP6_INC_STATS(net,
350 				ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
351 
352 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
353 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
354 		return 1;
355 	}
356 
357 	/* RFC 8200, Section 4.5 Fragment Header:
358 	 * If the first fragment does not include all headers through an
359 	 * Upper-Layer header, then that fragment should be discarded and
360 	 * an ICMP Parameter Problem, Code 3, message should be sent to
361 	 * the source of the fragment, with the Pointer field set to zero.
362 	 */
363 	nexthdr = hdr->nexthdr;
364 	if (ipv6frag_thdr_truncated(skb, skb_transport_offset(skb), &nexthdr)) {
365 		__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
366 				IPSTATS_MIB_INHDRERRORS);
367 		icmpv6_param_prob(skb, ICMPV6_HDR_INCOMP, 0);
368 		return -1;
369 	}
370 
371 	iif = skb->dev ? skb->dev->ifindex : 0;
372 	fq = fq_find(net, fhdr->identification, hdr, iif);
373 	if (fq) {
374 		u32 prob_offset = 0;
375 		int ret;
376 
377 		spin_lock(&fq->q.lock);
378 
379 		fq->iif = iif;
380 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
381 				     &prob_offset);
382 
383 		spin_unlock(&fq->q.lock);
384 		inet_frag_put(&fq->q);
385 		if (prob_offset) {
386 			__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
387 					IPSTATS_MIB_INHDRERRORS);
388 			/* icmpv6_param_prob() calls kfree_skb(skb) */
389 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
390 		}
391 		return ret;
392 	}
393 
394 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
395 	kfree_skb(skb);
396 	return -1;
397 
398 fail_hdr:
399 	__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
400 			IPSTATS_MIB_INHDRERRORS);
401 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
402 	return -1;
403 }
404 
405 static const struct inet6_protocol frag_protocol = {
406 	.handler	=	ipv6_frag_rcv,
407 	.flags		=	INET6_PROTO_NOPOLICY,
408 };
409 
410 #ifdef CONFIG_SYSCTL
411 
412 static struct ctl_table ip6_frags_ns_ctl_table[] = {
413 	{
414 		.procname	= "ip6frag_high_thresh",
415 		.maxlen		= sizeof(unsigned long),
416 		.mode		= 0644,
417 		.proc_handler	= proc_doulongvec_minmax,
418 	},
419 	{
420 		.procname	= "ip6frag_low_thresh",
421 		.maxlen		= sizeof(unsigned long),
422 		.mode		= 0644,
423 		.proc_handler	= proc_doulongvec_minmax,
424 	},
425 	{
426 		.procname	= "ip6frag_time",
427 		.maxlen		= sizeof(int),
428 		.mode		= 0644,
429 		.proc_handler	= proc_dointvec_jiffies,
430 	},
431 	{ }
432 };
433 
434 /* secret interval has been deprecated */
435 static int ip6_frags_secret_interval_unused;
436 static struct ctl_table ip6_frags_ctl_table[] = {
437 	{
438 		.procname	= "ip6frag_secret_interval",
439 		.data		= &ip6_frags_secret_interval_unused,
440 		.maxlen		= sizeof(int),
441 		.mode		= 0644,
442 		.proc_handler	= proc_dointvec_jiffies,
443 	},
444 	{ }
445 };
446 
447 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
448 {
449 	struct ctl_table *table;
450 	struct ctl_table_header *hdr;
451 
452 	table = ip6_frags_ns_ctl_table;
453 	if (!net_eq(net, &init_net)) {
454 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
455 		if (!table)
456 			goto err_alloc;
457 
458 	}
459 	table[0].data	= &net->ipv6.fqdir->high_thresh;
460 	table[0].extra1	= &net->ipv6.fqdir->low_thresh;
461 	table[1].data	= &net->ipv6.fqdir->low_thresh;
462 	table[1].extra2	= &net->ipv6.fqdir->high_thresh;
463 	table[2].data	= &net->ipv6.fqdir->timeout;
464 
465 	hdr = register_net_sysctl(net, "net/ipv6", table);
466 	if (!hdr)
467 		goto err_reg;
468 
469 	net->ipv6.sysctl.frags_hdr = hdr;
470 	return 0;
471 
472 err_reg:
473 	if (!net_eq(net, &init_net))
474 		kfree(table);
475 err_alloc:
476 	return -ENOMEM;
477 }
478 
479 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
480 {
481 	struct ctl_table *table;
482 
483 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
484 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
485 	if (!net_eq(net, &init_net))
486 		kfree(table);
487 }
488 
489 static struct ctl_table_header *ip6_ctl_header;
490 
491 static int ip6_frags_sysctl_register(void)
492 {
493 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
494 			ip6_frags_ctl_table);
495 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
496 }
497 
498 static void ip6_frags_sysctl_unregister(void)
499 {
500 	unregister_net_sysctl_table(ip6_ctl_header);
501 }
502 #else
503 static int ip6_frags_ns_sysctl_register(struct net *net)
504 {
505 	return 0;
506 }
507 
508 static void ip6_frags_ns_sysctl_unregister(struct net *net)
509 {
510 }
511 
512 static int ip6_frags_sysctl_register(void)
513 {
514 	return 0;
515 }
516 
517 static void ip6_frags_sysctl_unregister(void)
518 {
519 }
520 #endif
521 
522 static int __net_init ipv6_frags_init_net(struct net *net)
523 {
524 	int res;
525 
526 	res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
527 	if (res < 0)
528 		return res;
529 
530 	net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
531 	net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
532 	net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
533 
534 	res = ip6_frags_ns_sysctl_register(net);
535 	if (res < 0)
536 		fqdir_exit(net->ipv6.fqdir);
537 	return res;
538 }
539 
540 static void __net_exit ipv6_frags_pre_exit_net(struct net *net)
541 {
542 	fqdir_pre_exit(net->ipv6.fqdir);
543 }
544 
545 static void __net_exit ipv6_frags_exit_net(struct net *net)
546 {
547 	ip6_frags_ns_sysctl_unregister(net);
548 	fqdir_exit(net->ipv6.fqdir);
549 }
550 
551 static struct pernet_operations ip6_frags_ops = {
552 	.init		= ipv6_frags_init_net,
553 	.pre_exit	= ipv6_frags_pre_exit_net,
554 	.exit		= ipv6_frags_exit_net,
555 };
556 
557 static const struct rhashtable_params ip6_rhash_params = {
558 	.head_offset		= offsetof(struct inet_frag_queue, node),
559 	.hashfn			= ip6frag_key_hashfn,
560 	.obj_hashfn		= ip6frag_obj_hashfn,
561 	.obj_cmpfn		= ip6frag_obj_cmpfn,
562 	.automatic_shrinking	= true,
563 };
564 
565 int __init ipv6_frag_init(void)
566 {
567 	int ret;
568 
569 	ip6_frags.constructor = ip6frag_init;
570 	ip6_frags.destructor = NULL;
571 	ip6_frags.qsize = sizeof(struct frag_queue);
572 	ip6_frags.frag_expire = ip6_frag_expire;
573 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
574 	ip6_frags.rhash_params = ip6_rhash_params;
575 	ret = inet_frags_init(&ip6_frags);
576 	if (ret)
577 		goto out;
578 
579 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
580 	if (ret)
581 		goto err_protocol;
582 
583 	ret = ip6_frags_sysctl_register();
584 	if (ret)
585 		goto err_sysctl;
586 
587 	ret = register_pernet_subsys(&ip6_frags_ops);
588 	if (ret)
589 		goto err_pernet;
590 
591 out:
592 	return ret;
593 
594 err_pernet:
595 	ip6_frags_sysctl_unregister();
596 err_sysctl:
597 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
598 err_protocol:
599 	inet_frags_fini(&ip6_frags);
600 	goto out;
601 }
602 
603 void ipv6_frag_exit(void)
604 {
605 	ip6_frags_sysctl_unregister();
606 	unregister_pernet_subsys(&ip6_frags_ops);
607 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
608 	inet_frags_fini(&ip6_frags);
609 }
610