xref: /openbmc/linux/net/ipv6/reassembly.c (revision c4e78957) 
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.mono_delivery_time = skb->mono_delivery_time;
198 	fq->q.meat += skb->len;
199 	fq->ecn |= ecn;
200 	add_frag_mem_limit(fq->q.fqdir, skb->truesize);
201 
202 	fragsize = -skb_network_offset(skb) + skb->len;
203 	if (fragsize > fq->q.max_size)
204 		fq->q.max_size = fragsize;
205 
206 	/* The first fragment.
207 	 * nhoffset is obtained from the first fragment, of course.
208 	 */
209 	if (offset == 0) {
210 		fq->nhoffset = nhoff;
211 		fq->q.flags |= INET_FRAG_FIRST_IN;
212 	}
213 
214 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
215 	    fq->q.meat == fq->q.len) {
216 		unsigned long orefdst = skb->_skb_refdst;
217 
218 		skb->_skb_refdst = 0UL;
219 		err = ip6_frag_reasm(fq, skb, prev_tail, dev);
220 		skb->_skb_refdst = orefdst;
221 		return err;
222 	}
223 
224 	skb_dst_drop(skb);
225 	return -EINPROGRESS;
226 
227 insert_error:
228 	if (err == IPFRAG_DUP) {
229 		kfree_skb(skb);
230 		return -EINVAL;
231 	}
232 	err = -EINVAL;
233 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
234 			IPSTATS_MIB_REASM_OVERLAPS);
235 discard_fq:
236 	inet_frag_kill(&fq->q);
237 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
238 			IPSTATS_MIB_REASMFAILS);
239 err:
240 	kfree_skb(skb);
241 	return err;
242 }
243 
244 /*
245  *	Check if this packet is complete.
246  *
247  *	It is called with locked fq, and caller must check that
248  *	queue is eligible for reassembly i.e. it is not COMPLETE,
249  *	the last and the first frames arrived and all the bits are here.
250  */
251 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
252 			  struct sk_buff *prev_tail, struct net_device *dev)
253 {
254 	struct net *net = fq->q.fqdir->net;
255 	unsigned int nhoff;
256 	void *reasm_data;
257 	int payload_len;
258 	u8 ecn;
259 
260 	inet_frag_kill(&fq->q);
261 
262 	ecn = ip_frag_ecn_table[fq->ecn];
263 	if (unlikely(ecn == 0xff))
264 		goto out_fail;
265 
266 	reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
267 	if (!reasm_data)
268 		goto out_oom;
269 
270 	payload_len = ((skb->data - skb_network_header(skb)) -
271 		       sizeof(struct ipv6hdr) + fq->q.len -
272 		       sizeof(struct frag_hdr));
273 	if (payload_len > IPV6_MAXPLEN)
274 		goto out_oversize;
275 
276 	/* We have to remove fragment header from datagram and to relocate
277 	 * header in order to calculate ICV correctly. */
278 	nhoff = fq->nhoffset;
279 	skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
280 	memmove(skb->head + sizeof(struct frag_hdr), skb->head,
281 		(skb->data - skb->head) - sizeof(struct frag_hdr));
282 	if (skb_mac_header_was_set(skb))
283 		skb->mac_header += sizeof(struct frag_hdr);
284 	skb->network_header += sizeof(struct frag_hdr);
285 
286 	skb_reset_transport_header(skb);
287 
288 	inet_frag_reasm_finish(&fq->q, skb, reasm_data, true);
289 
290 	skb->dev = dev;
291 	ipv6_hdr(skb)->payload_len = htons(payload_len);
292 	ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
293 	IP6CB(skb)->nhoff = nhoff;
294 	IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
295 	IP6CB(skb)->frag_max_size = fq->q.max_size;
296 
297 	/* Yes, and fold redundant checksum back. 8) */
298 	skb_postpush_rcsum(skb, skb_network_header(skb),
299 			   skb_network_header_len(skb));
300 
301 	rcu_read_lock();
302 	__IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS);
303 	rcu_read_unlock();
304 	fq->q.rb_fragments = RB_ROOT;
305 	fq->q.fragments_tail = NULL;
306 	fq->q.last_run_head = NULL;
307 	return 1;
308 
309 out_oversize:
310 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
311 	goto out_fail;
312 out_oom:
313 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
314 out_fail:
315 	rcu_read_lock();
316 	__IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS);
317 	rcu_read_unlock();
318 	inet_frag_kill(&fq->q);
319 	return -1;
320 }
321 
322 static int ipv6_frag_rcv(struct sk_buff *skb)
323 {
324 	struct frag_hdr *fhdr;
325 	struct frag_queue *fq;
326 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
327 	struct net *net = dev_net(skb_dst(skb)->dev);
328 	u8 nexthdr;
329 	int iif;
330 
331 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
332 		goto fail_hdr;
333 
334 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
335 
336 	/* Jumbo payload inhibits frag. header */
337 	if (hdr->payload_len == 0)
338 		goto fail_hdr;
339 
340 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
341 				 sizeof(struct frag_hdr))))
342 		goto fail_hdr;
343 
344 	hdr = ipv6_hdr(skb);
345 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
346 
347 	if (!(fhdr->frag_off & htons(IP6_OFFSET | IP6_MF))) {
348 		/* It is not a fragmented frame */
349 		skb->transport_header += sizeof(struct frag_hdr);
350 		__IP6_INC_STATS(net,
351 				ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
352 
353 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
354 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
355 		IP6CB(skb)->frag_max_size = ntohs(hdr->payload_len) +
356 					    sizeof(struct ipv6hdr);
357 		return 1;
358 	}
359 
360 	/* RFC 8200, Section 4.5 Fragment Header:
361 	 * If the first fragment does not include all headers through an
362 	 * Upper-Layer header, then that fragment should be discarded and
363 	 * an ICMP Parameter Problem, Code 3, message should be sent to
364 	 * the source of the fragment, with the Pointer field set to zero.
365 	 */
366 	nexthdr = hdr->nexthdr;
367 	if (ipv6frag_thdr_truncated(skb, skb_transport_offset(skb), &nexthdr)) {
368 		__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
369 				IPSTATS_MIB_INHDRERRORS);
370 		icmpv6_param_prob(skb, ICMPV6_HDR_INCOMP, 0);
371 		return -1;
372 	}
373 
374 	iif = skb->dev ? skb->dev->ifindex : 0;
375 	fq = fq_find(net, fhdr->identification, hdr, iif);
376 	if (fq) {
377 		u32 prob_offset = 0;
378 		int ret;
379 
380 		spin_lock(&fq->q.lock);
381 
382 		fq->iif = iif;
383 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
384 				     &prob_offset);
385 
386 		spin_unlock(&fq->q.lock);
387 		inet_frag_put(&fq->q);
388 		if (prob_offset) {
389 			__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
390 					IPSTATS_MIB_INHDRERRORS);
391 			/* icmpv6_param_prob() calls kfree_skb(skb) */
392 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
393 		}
394 		return ret;
395 	}
396 
397 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
398 	kfree_skb(skb);
399 	return -1;
400 
401 fail_hdr:
402 	__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
403 			IPSTATS_MIB_INHDRERRORS);
404 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
405 	return -1;
406 }
407 
408 static const struct inet6_protocol frag_protocol = {
409 	.handler	=	ipv6_frag_rcv,
410 	.flags		=	INET6_PROTO_NOPOLICY,
411 };
412 
413 #ifdef CONFIG_SYSCTL
414 
415 static struct ctl_table ip6_frags_ns_ctl_table[] = {
416 	{
417 		.procname	= "ip6frag_high_thresh",
418 		.maxlen		= sizeof(unsigned long),
419 		.mode		= 0644,
420 		.proc_handler	= proc_doulongvec_minmax,
421 	},
422 	{
423 		.procname	= "ip6frag_low_thresh",
424 		.maxlen		= sizeof(unsigned long),
425 		.mode		= 0644,
426 		.proc_handler	= proc_doulongvec_minmax,
427 	},
428 	{
429 		.procname	= "ip6frag_time",
430 		.maxlen		= sizeof(int),
431 		.mode		= 0644,
432 		.proc_handler	= proc_dointvec_jiffies,
433 	},
434 	{ }
435 };
436 
437 /* secret interval has been deprecated */
438 static int ip6_frags_secret_interval_unused;
439 static struct ctl_table ip6_frags_ctl_table[] = {
440 	{
441 		.procname	= "ip6frag_secret_interval",
442 		.data		= &ip6_frags_secret_interval_unused,
443 		.maxlen		= sizeof(int),
444 		.mode		= 0644,
445 		.proc_handler	= proc_dointvec_jiffies,
446 	},
447 	{ }
448 };
449 
450 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
451 {
452 	struct ctl_table *table;
453 	struct ctl_table_header *hdr;
454 
455 	table = ip6_frags_ns_ctl_table;
456 	if (!net_eq(net, &init_net)) {
457 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
458 		if (!table)
459 			goto err_alloc;
460 
461 	}
462 	table[0].data	= &net->ipv6.fqdir->high_thresh;
463 	table[0].extra1	= &net->ipv6.fqdir->low_thresh;
464 	table[1].data	= &net->ipv6.fqdir->low_thresh;
465 	table[1].extra2	= &net->ipv6.fqdir->high_thresh;
466 	table[2].data	= &net->ipv6.fqdir->timeout;
467 
468 	hdr = register_net_sysctl(net, "net/ipv6", table);
469 	if (!hdr)
470 		goto err_reg;
471 
472 	net->ipv6.sysctl.frags_hdr = hdr;
473 	return 0;
474 
475 err_reg:
476 	if (!net_eq(net, &init_net))
477 		kfree(table);
478 err_alloc:
479 	return -ENOMEM;
480 }
481 
482 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
483 {
484 	struct ctl_table *table;
485 
486 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
487 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
488 	if (!net_eq(net, &init_net))
489 		kfree(table);
490 }
491 
492 static struct ctl_table_header *ip6_ctl_header;
493 
494 static int ip6_frags_sysctl_register(void)
495 {
496 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
497 			ip6_frags_ctl_table);
498 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
499 }
500 
501 static void ip6_frags_sysctl_unregister(void)
502 {
503 	unregister_net_sysctl_table(ip6_ctl_header);
504 }
505 #else
506 static int ip6_frags_ns_sysctl_register(struct net *net)
507 {
508 	return 0;
509 }
510 
511 static void ip6_frags_ns_sysctl_unregister(struct net *net)
512 {
513 }
514 
515 static int ip6_frags_sysctl_register(void)
516 {
517 	return 0;
518 }
519 
520 static void ip6_frags_sysctl_unregister(void)
521 {
522 }
523 #endif
524 
525 static int __net_init ipv6_frags_init_net(struct net *net)
526 {
527 	int res;
528 
529 	res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
530 	if (res < 0)
531 		return res;
532 
533 	net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
534 	net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
535 	net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
536 
537 	res = ip6_frags_ns_sysctl_register(net);
538 	if (res < 0)
539 		fqdir_exit(net->ipv6.fqdir);
540 	return res;
541 }
542 
543 static void __net_exit ipv6_frags_pre_exit_net(struct net *net)
544 {
545 	fqdir_pre_exit(net->ipv6.fqdir);
546 }
547 
548 static void __net_exit ipv6_frags_exit_net(struct net *net)
549 {
550 	ip6_frags_ns_sysctl_unregister(net);
551 	fqdir_exit(net->ipv6.fqdir);
552 }
553 
554 static struct pernet_operations ip6_frags_ops = {
555 	.init		= ipv6_frags_init_net,
556 	.pre_exit	= ipv6_frags_pre_exit_net,
557 	.exit		= ipv6_frags_exit_net,
558 };
559 
560 static const struct rhashtable_params ip6_rhash_params = {
561 	.head_offset		= offsetof(struct inet_frag_queue, node),
562 	.hashfn			= ip6frag_key_hashfn,
563 	.obj_hashfn		= ip6frag_obj_hashfn,
564 	.obj_cmpfn		= ip6frag_obj_cmpfn,
565 	.automatic_shrinking	= true,
566 };
567 
568 int __init ipv6_frag_init(void)
569 {
570 	int ret;
571 
572 	ip6_frags.constructor = ip6frag_init;
573 	ip6_frags.destructor = NULL;
574 	ip6_frags.qsize = sizeof(struct frag_queue);
575 	ip6_frags.frag_expire = ip6_frag_expire;
576 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
577 	ip6_frags.rhash_params = ip6_rhash_params;
578 	ret = inet_frags_init(&ip6_frags);
579 	if (ret)
580 		goto out;
581 
582 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
583 	if (ret)
584 		goto err_protocol;
585 
586 	ret = ip6_frags_sysctl_register();
587 	if (ret)
588 		goto err_sysctl;
589 
590 	ret = register_pernet_subsys(&ip6_frags_ops);
591 	if (ret)
592 		goto err_pernet;
593 
594 out:
595 	return ret;
596 
597 err_pernet:
598 	ip6_frags_sysctl_unregister();
599 err_sysctl:
600 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
601 err_protocol:
602 	inet_frags_fini(&ip6_frags);
603 	goto out;
604 }
605 
606 void ipv6_frag_exit(void)
607 {
608 	ip6_frags_sysctl_unregister();
609 	unregister_pernet_subsys(&ip6_frags_ops);
610 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
611 	inet_frags_fini(&ip6_frags);
612 }
613