xref: /openbmc/linux/net/ipv6/reassembly.c (revision 6d425d7c) 
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(IP6_OFFSET | IP6_MF))) {
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 		IP6CB(skb)->frag_max_size = ntohs(hdr->payload_len) +
355 					    sizeof(struct ipv6hdr);
356 		return 1;
357 	}
358 
359 	/* RFC 8200, Section 4.5 Fragment Header:
360 	 * If the first fragment does not include all headers through an
361 	 * Upper-Layer header, then that fragment should be discarded and
362 	 * an ICMP Parameter Problem, Code 3, message should be sent to
363 	 * the source of the fragment, with the Pointer field set to zero.
364 	 */
365 	nexthdr = hdr->nexthdr;
366 	if (ipv6frag_thdr_truncated(skb, skb_transport_offset(skb), &nexthdr)) {
367 		__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
368 				IPSTATS_MIB_INHDRERRORS);
369 		icmpv6_param_prob(skb, ICMPV6_HDR_INCOMP, 0);
370 		return -1;
371 	}
372 
373 	iif = skb->dev ? skb->dev->ifindex : 0;
374 	fq = fq_find(net, fhdr->identification, hdr, iif);
375 	if (fq) {
376 		u32 prob_offset = 0;
377 		int ret;
378 
379 		spin_lock(&fq->q.lock);
380 
381 		fq->iif = iif;
382 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
383 				     &prob_offset);
384 
385 		spin_unlock(&fq->q.lock);
386 		inet_frag_put(&fq->q);
387 		if (prob_offset) {
388 			__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
389 					IPSTATS_MIB_INHDRERRORS);
390 			/* icmpv6_param_prob() calls kfree_skb(skb) */
391 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
392 		}
393 		return ret;
394 	}
395 
396 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
397 	kfree_skb(skb);
398 	return -1;
399 
400 fail_hdr:
401 	__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
402 			IPSTATS_MIB_INHDRERRORS);
403 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
404 	return -1;
405 }
406 
407 static const struct inet6_protocol frag_protocol = {
408 	.handler	=	ipv6_frag_rcv,
409 	.flags		=	INET6_PROTO_NOPOLICY,
410 };
411 
412 #ifdef CONFIG_SYSCTL
413 
414 static struct ctl_table ip6_frags_ns_ctl_table[] = {
415 	{
416 		.procname	= "ip6frag_high_thresh",
417 		.maxlen		= sizeof(unsigned long),
418 		.mode		= 0644,
419 		.proc_handler	= proc_doulongvec_minmax,
420 	},
421 	{
422 		.procname	= "ip6frag_low_thresh",
423 		.maxlen		= sizeof(unsigned long),
424 		.mode		= 0644,
425 		.proc_handler	= proc_doulongvec_minmax,
426 	},
427 	{
428 		.procname	= "ip6frag_time",
429 		.maxlen		= sizeof(int),
430 		.mode		= 0644,
431 		.proc_handler	= proc_dointvec_jiffies,
432 	},
433 	{ }
434 };
435 
436 /* secret interval has been deprecated */
437 static int ip6_frags_secret_interval_unused;
438 static struct ctl_table ip6_frags_ctl_table[] = {
439 	{
440 		.procname	= "ip6frag_secret_interval",
441 		.data		= &ip6_frags_secret_interval_unused,
442 		.maxlen		= sizeof(int),
443 		.mode		= 0644,
444 		.proc_handler	= proc_dointvec_jiffies,
445 	},
446 	{ }
447 };
448 
449 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
450 {
451 	struct ctl_table *table;
452 	struct ctl_table_header *hdr;
453 
454 	table = ip6_frags_ns_ctl_table;
455 	if (!net_eq(net, &init_net)) {
456 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
457 		if (!table)
458 			goto err_alloc;
459 
460 	}
461 	table[0].data	= &net->ipv6.fqdir->high_thresh;
462 	table[0].extra1	= &net->ipv6.fqdir->low_thresh;
463 	table[1].data	= &net->ipv6.fqdir->low_thresh;
464 	table[1].extra2	= &net->ipv6.fqdir->high_thresh;
465 	table[2].data	= &net->ipv6.fqdir->timeout;
466 
467 	hdr = register_net_sysctl(net, "net/ipv6", table);
468 	if (!hdr)
469 		goto err_reg;
470 
471 	net->ipv6.sysctl.frags_hdr = hdr;
472 	return 0;
473 
474 err_reg:
475 	if (!net_eq(net, &init_net))
476 		kfree(table);
477 err_alloc:
478 	return -ENOMEM;
479 }
480 
481 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
482 {
483 	struct ctl_table *table;
484 
485 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
486 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
487 	if (!net_eq(net, &init_net))
488 		kfree(table);
489 }
490 
491 static struct ctl_table_header *ip6_ctl_header;
492 
493 static int ip6_frags_sysctl_register(void)
494 {
495 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
496 			ip6_frags_ctl_table);
497 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
498 }
499 
500 static void ip6_frags_sysctl_unregister(void)
501 {
502 	unregister_net_sysctl_table(ip6_ctl_header);
503 }
504 #else
505 static int ip6_frags_ns_sysctl_register(struct net *net)
506 {
507 	return 0;
508 }
509 
510 static void ip6_frags_ns_sysctl_unregister(struct net *net)
511 {
512 }
513 
514 static int ip6_frags_sysctl_register(void)
515 {
516 	return 0;
517 }
518 
519 static void ip6_frags_sysctl_unregister(void)
520 {
521 }
522 #endif
523 
524 static int __net_init ipv6_frags_init_net(struct net *net)
525 {
526 	int res;
527 
528 	res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
529 	if (res < 0)
530 		return res;
531 
532 	net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
533 	net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
534 	net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
535 
536 	res = ip6_frags_ns_sysctl_register(net);
537 	if (res < 0)
538 		fqdir_exit(net->ipv6.fqdir);
539 	return res;
540 }
541 
542 static void __net_exit ipv6_frags_pre_exit_net(struct net *net)
543 {
544 	fqdir_pre_exit(net->ipv6.fqdir);
545 }
546 
547 static void __net_exit ipv6_frags_exit_net(struct net *net)
548 {
549 	ip6_frags_ns_sysctl_unregister(net);
550 	fqdir_exit(net->ipv6.fqdir);
551 }
552 
553 static struct pernet_operations ip6_frags_ops = {
554 	.init		= ipv6_frags_init_net,
555 	.pre_exit	= ipv6_frags_pre_exit_net,
556 	.exit		= ipv6_frags_exit_net,
557 };
558 
559 static const struct rhashtable_params ip6_rhash_params = {
560 	.head_offset		= offsetof(struct inet_frag_queue, node),
561 	.hashfn			= ip6frag_key_hashfn,
562 	.obj_hashfn		= ip6frag_obj_hashfn,
563 	.obj_cmpfn		= ip6frag_obj_cmpfn,
564 	.automatic_shrinking	= true,
565 };
566 
567 int __init ipv6_frag_init(void)
568 {
569 	int ret;
570 
571 	ip6_frags.constructor = ip6frag_init;
572 	ip6_frags.destructor = NULL;
573 	ip6_frags.qsize = sizeof(struct frag_queue);
574 	ip6_frags.frag_expire = ip6_frag_expire;
575 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
576 	ip6_frags.rhash_params = ip6_rhash_params;
577 	ret = inet_frags_init(&ip6_frags);
578 	if (ret)
579 		goto out;
580 
581 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
582 	if (ret)
583 		goto err_protocol;
584 
585 	ret = ip6_frags_sysctl_register();
586 	if (ret)
587 		goto err_sysctl;
588 
589 	ret = register_pernet_subsys(&ip6_frags_ops);
590 	if (ret)
591 		goto err_pernet;
592 
593 out:
594 	return ret;
595 
596 err_pernet:
597 	ip6_frags_sysctl_unregister();
598 err_sysctl:
599 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
600 err_protocol:
601 	inet_frags_fini(&ip6_frags);
602 	goto out;
603 }
604 
605 void ipv6_frag_exit(void)
606 {
607 	ip6_frags_sysctl_unregister();
608 	unregister_pernet_subsys(&ip6_frags_ops);
609 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
610 	inet_frags_fini(&ip6_frags);
611 }
612