xref: /linux/net/ipv6/reassembly.c (revision 91a4855d6c03e770e42f17c798a36a3c46e63de2) 
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 			  int *refs);
73 
74 static void ip6_frag_expire(struct timer_list *t)
75 {
76 	struct inet_frag_queue *frag = timer_container_of(frag, t, timer);
77 	struct frag_queue *fq;
78 
79 	fq = container_of(frag, struct frag_queue, q);
80 
81 	ip6frag_expire_frag_queue(fq->q.fqdir->net, fq);
82 }
83 
84 static struct frag_queue *
85 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
86 {
87 	struct frag_v6_compare_key key = {
88 		.id = id,
89 		.saddr = hdr->saddr,
90 		.daddr = hdr->daddr,
91 		.user = IP6_DEFRAG_LOCAL_DELIVER,
92 		.iif = iif,
93 	};
94 	struct inet_frag_queue *q;
95 
96 	if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
97 					    IPV6_ADDR_LINKLOCAL)))
98 		key.iif = 0;
99 
100 	q = inet_frag_find(net->ipv6.fqdir, &key);
101 	if (!q)
102 		return NULL;
103 
104 	return container_of(q, struct frag_queue, q);
105 }
106 
107 static int ip6_frag_queue(struct net *net,
108 			  struct frag_queue *fq, struct sk_buff *skb,
109 			  struct frag_hdr *fhdr, int nhoff,
110 			  u32 *prob_offset, int *refs)
111 {
112 	int offset, end, fragsize;
113 	struct sk_buff *prev_tail;
114 	struct net_device *dev;
115 	int err = -ENOENT;
116 	SKB_DR(reason);
117 	u8 ecn;
118 
119 	/* If reassembly is already done, @skb must be a duplicate frag. */
120 	if (fq->q.flags & INET_FRAG_COMPLETE) {
121 		SKB_DR_SET(reason, DUP_FRAG);
122 		goto err;
123 	}
124 
125 	err = -EINVAL;
126 	offset = ntohs(fhdr->frag_off) & ~0x7;
127 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
128 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
129 
130 	if ((unsigned int)end > IPV6_MAXPLEN) {
131 		*prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
132 		/* note that if prob_offset is set, the skb is freed elsewhere,
133 		 * we do not free it here.
134 		 */
135 		inet_frag_kill(&fq->q, refs);
136 		__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
137 				IPSTATS_MIB_REASMFAILS);
138 		return -1;
139 	}
140 
141 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
142 
143 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
144 		const unsigned char *nh = skb_network_header(skb);
145 		skb->csum = csum_sub(skb->csum,
146 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
147 						  0));
148 	}
149 
150 	/* Is this the final fragment? */
151 	if (!(fhdr->frag_off & htons(IP6_MF))) {
152 		/* If we already have some bits beyond end
153 		 * or have different end, the segment is corrupted.
154 		 */
155 		if (end < fq->q.len ||
156 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
157 			goto discard_fq;
158 		fq->q.flags |= INET_FRAG_LAST_IN;
159 		fq->q.len = end;
160 	} else {
161 		/* Check if the fragment is rounded to 8 bytes.
162 		 * Required by the RFC.
163 		 */
164 		if (end & 0x7) {
165 			/* RFC2460 says always send parameter problem in
166 			 * this case. -DaveM
167 			 */
168 			*prob_offset = offsetof(struct ipv6hdr, payload_len);
169 			inet_frag_kill(&fq->q, refs);
170 			__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
171 					IPSTATS_MIB_REASMFAILS);
172 			return -1;
173 		}
174 		if (end > fq->q.len) {
175 			/* Some bits beyond end -> corruption. */
176 			if (fq->q.flags & INET_FRAG_LAST_IN)
177 				goto discard_fq;
178 			fq->q.len = end;
179 		}
180 	}
181 
182 	if (end == offset)
183 		goto discard_fq;
184 
185 	err = -ENOMEM;
186 	/* Point into the IP datagram 'data' part. */
187 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
188 		goto discard_fq;
189 
190 	err = pskb_trim_rcsum(skb, end - offset);
191 	if (err)
192 		goto discard_fq;
193 
194 	/* Note : skb->rbnode and skb->dev share the same location. */
195 	dev = skb->dev;
196 	/* Makes sure compiler wont do silly aliasing games */
197 	barrier();
198 
199 	prev_tail = fq->q.fragments_tail;
200 	err = inet_frag_queue_insert(&fq->q, skb, offset, end);
201 	if (err)
202 		goto insert_error;
203 
204 	if (dev)
205 		fq->iif = dev->ifindex;
206 
207 	fq->q.stamp = skb->tstamp;
208 	fq->q.tstamp_type = skb->tstamp_type;
209 	fq->q.meat += skb->len;
210 	fq->ecn |= ecn;
211 	add_frag_mem_limit(fq->q.fqdir, skb->truesize);
212 
213 	fragsize = -skb_network_offset(skb) + skb->len;
214 	if (fragsize > fq->q.max_size)
215 		fq->q.max_size = fragsize;
216 
217 	/* The first fragment.
218 	 * nhoffset is obtained from the first fragment, of course.
219 	 */
220 	if (offset == 0) {
221 		fq->nhoffset = nhoff;
222 		fq->q.flags |= INET_FRAG_FIRST_IN;
223 	}
224 
225 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
226 	    fq->q.meat == fq->q.len) {
227 		unsigned long orefdst = skb->_skb_refdst;
228 
229 		skb->_skb_refdst = 0UL;
230 		err = ip6_frag_reasm(fq, skb, prev_tail, dev, refs);
231 		skb->_skb_refdst = orefdst;
232 		return err;
233 	}
234 
235 	skb_dst_drop(skb);
236 	return -EINPROGRESS;
237 
238 insert_error:
239 	if (err == IPFRAG_DUP) {
240 		SKB_DR_SET(reason, DUP_FRAG);
241 		err = -EINVAL;
242 		goto err;
243 	}
244 	err = -EINVAL;
245 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
246 			IPSTATS_MIB_REASM_OVERLAPS);
247 discard_fq:
248 	inet_frag_kill(&fq->q, refs);
249 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
250 			IPSTATS_MIB_REASMFAILS);
251 err:
252 	kfree_skb_reason(skb, reason);
253 	return err;
254 }
255 
256 /*
257  *	Check if this packet is complete.
258  *
259  *	It is called with locked fq, and caller must check that
260  *	queue is eligible for reassembly i.e. it is not COMPLETE,
261  *	the last and the first frames arrived and all the bits are here.
262  */
263 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
264 			  struct sk_buff *prev_tail, struct net_device *dev,
265 			  int *refs)
266 {
267 	struct net *net = fq->q.fqdir->net;
268 	unsigned int nhoff;
269 	void *reasm_data;
270 	int payload_len;
271 	u8 ecn;
272 
273 	inet_frag_kill(&fq->q, refs);
274 
275 	ecn = ip_frag_ecn_table[fq->ecn];
276 	if (unlikely(ecn == 0xff))
277 		goto out_fail;
278 
279 	reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
280 	if (!reasm_data)
281 		goto out_oom;
282 
283 	payload_len = -skb_network_offset(skb) -
284 		       sizeof(struct ipv6hdr) + fq->q.len -
285 		       sizeof(struct frag_hdr);
286 	if (payload_len > IPV6_MAXPLEN)
287 		goto out_oversize;
288 
289 	/* We have to remove fragment header from datagram and to relocate
290 	 * header in order to calculate ICV correctly. */
291 	nhoff = fq->nhoffset;
292 	skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
293 	memmove(skb->head + sizeof(struct frag_hdr), skb->head,
294 		(skb->data - skb->head) - sizeof(struct frag_hdr));
295 	if (skb_mac_header_was_set(skb))
296 		skb->mac_header += sizeof(struct frag_hdr);
297 	skb->network_header += sizeof(struct frag_hdr);
298 
299 	skb_reset_transport_header(skb);
300 
301 	inet_frag_reasm_finish(&fq->q, skb, reasm_data, true);
302 
303 	skb->dev = dev;
304 	ipv6_hdr(skb)->payload_len = htons(payload_len);
305 	ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
306 	IP6CB(skb)->nhoff = nhoff;
307 	IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
308 	IP6CB(skb)->frag_max_size = fq->q.max_size;
309 
310 	/* Yes, and fold redundant checksum back. 8) */
311 	skb_postpush_rcsum(skb, skb_network_header(skb),
312 			   skb_network_header_len(skb));
313 
314 	__IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS);
315 	fq->q.rb_fragments = RB_ROOT;
316 	fq->q.fragments_tail = NULL;
317 	fq->q.last_run_head = NULL;
318 	return 1;
319 
320 out_oversize:
321 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
322 	goto out_fail;
323 out_oom:
324 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
325 out_fail:
326 	__IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS);
327 	inet_frag_kill(&fq->q, refs);
328 	return -1;
329 }
330 
331 static int ipv6_frag_rcv(struct sk_buff *skb)
332 {
333 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
334 	struct net *net = skb_dst_dev_net(skb);
335 	struct frag_hdr *fhdr;
336 	struct frag_queue *fq;
337 	u8 nexthdr;
338 	int iif;
339 
340 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
341 		goto fail_hdr;
342 
343 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
344 
345 	/* Jumbo payload inhibits frag. header */
346 	if (hdr->payload_len == 0)
347 		goto fail_hdr;
348 
349 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
350 				 sizeof(struct frag_hdr))))
351 		goto fail_hdr;
352 
353 	hdr = ipv6_hdr(skb);
354 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
355 
356 	if (!(fhdr->frag_off & htons(IP6_OFFSET | IP6_MF))) {
357 		/* It is not a fragmented frame */
358 		skb->transport_header += sizeof(struct frag_hdr);
359 		__IP6_INC_STATS(net,
360 				ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
361 
362 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
363 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
364 		IP6CB(skb)->frag_max_size = ntohs(hdr->payload_len) +
365 					    sizeof(struct ipv6hdr);
366 		return 1;
367 	}
368 
369 	/* RFC 8200, Section 4.5 Fragment Header:
370 	 * If the first fragment does not include all headers through an
371 	 * Upper-Layer header, then that fragment should be discarded and
372 	 * an ICMP Parameter Problem, Code 3, message should be sent to
373 	 * the source of the fragment, with the Pointer field set to zero.
374 	 */
375 	nexthdr = hdr->nexthdr;
376 	if (ipv6frag_thdr_truncated(skb, skb_network_offset(skb) + sizeof(struct ipv6hdr), &nexthdr)) {
377 		__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
378 				IPSTATS_MIB_INHDRERRORS);
379 		icmpv6_param_prob(skb, ICMPV6_HDR_INCOMP, 0);
380 		return -1;
381 	}
382 
383 	iif = skb->dev ? skb->dev->ifindex : 0;
384 	rcu_read_lock();
385 	fq = fq_find(net, fhdr->identification, hdr, iif);
386 	if (fq) {
387 		u32 prob_offset = 0;
388 		int ret, refs = 0;
389 
390 		spin_lock(&fq->q.lock);
391 
392 		fq->iif = iif;
393 		ret = ip6_frag_queue(net, fq, skb, fhdr, IP6CB(skb)->nhoff,
394 				     &prob_offset, &refs);
395 
396 		spin_unlock(&fq->q.lock);
397 		rcu_read_unlock();
398 		inet_frag_putn(&fq->q, refs);
399 		if (prob_offset) {
400 			__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
401 					IPSTATS_MIB_INHDRERRORS);
402 			/* icmpv6_param_prob() calls kfree_skb(skb) */
403 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
404 		}
405 		return ret;
406 	}
407 	rcu_read_unlock();
408 
409 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
410 	kfree_skb(skb);
411 	return -1;
412 
413 fail_hdr:
414 	__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
415 			IPSTATS_MIB_INHDRERRORS);
416 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
417 	return -1;
418 }
419 
420 static const struct inet6_protocol frag_protocol = {
421 	.handler	=	ipv6_frag_rcv,
422 	.flags		=	INET6_PROTO_NOPOLICY,
423 };
424 
425 #ifdef CONFIG_SYSCTL
426 
427 static struct ctl_table ip6_frags_ns_ctl_table[] = {
428 	{
429 		.procname	= "ip6frag_high_thresh",
430 		.maxlen		= sizeof(unsigned long),
431 		.mode		= 0644,
432 		.proc_handler	= proc_doulongvec_minmax,
433 	},
434 	{
435 		.procname	= "ip6frag_low_thresh",
436 		.maxlen		= sizeof(unsigned long),
437 		.mode		= 0644,
438 		.proc_handler	= proc_doulongvec_minmax,
439 	},
440 	{
441 		.procname	= "ip6frag_time",
442 		.maxlen		= sizeof(int),
443 		.mode		= 0644,
444 		.proc_handler	= proc_dointvec_jiffies,
445 	},
446 };
447 
448 /* secret interval has been deprecated */
449 static int ip6_frags_secret_interval_unused;
450 static struct ctl_table ip6_frags_ctl_table[] = {
451 	{
452 		.procname	= "ip6frag_secret_interval",
453 		.data		= &ip6_frags_secret_interval_unused,
454 		.maxlen		= sizeof(int),
455 		.mode		= 0644,
456 		.proc_handler	= proc_dointvec_jiffies,
457 	},
458 };
459 
460 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
461 {
462 	struct ctl_table *table;
463 	struct ctl_table_header *hdr;
464 
465 	table = ip6_frags_ns_ctl_table;
466 	if (!net_eq(net, &init_net)) {
467 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
468 		if (!table)
469 			goto err_alloc;
470 
471 	}
472 	table[0].data	= &net->ipv6.fqdir->high_thresh;
473 	table[0].extra1	= &net->ipv6.fqdir->low_thresh;
474 	table[1].data	= &net->ipv6.fqdir->low_thresh;
475 	table[1].extra2	= &net->ipv6.fqdir->high_thresh;
476 	table[2].data	= &net->ipv6.fqdir->timeout;
477 
478 	hdr = register_net_sysctl_sz(net, "net/ipv6", table,
479 				     ARRAY_SIZE(ip6_frags_ns_ctl_table));
480 	if (!hdr)
481 		goto err_reg;
482 
483 	net->ipv6.sysctl.frags_hdr = hdr;
484 	return 0;
485 
486 err_reg:
487 	if (!net_eq(net, &init_net))
488 		kfree(table);
489 err_alloc:
490 	return -ENOMEM;
491 }
492 
493 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
494 {
495 	const struct ctl_table *table;
496 
497 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
498 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
499 	if (!net_eq(net, &init_net))
500 		kfree(table);
501 }
502 
503 static struct ctl_table_header *ip6_ctl_header;
504 
505 static int ip6_frags_sysctl_register(void)
506 {
507 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
508 			ip6_frags_ctl_table);
509 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
510 }
511 
512 static void ip6_frags_sysctl_unregister(void)
513 {
514 	unregister_net_sysctl_table(ip6_ctl_header);
515 }
516 #else
517 static int ip6_frags_ns_sysctl_register(struct net *net)
518 {
519 	return 0;
520 }
521 
522 static void ip6_frags_ns_sysctl_unregister(struct net *net)
523 {
524 }
525 
526 static int ip6_frags_sysctl_register(void)
527 {
528 	return 0;
529 }
530 
531 static void ip6_frags_sysctl_unregister(void)
532 {
533 }
534 #endif
535 
536 static int __net_init ipv6_frags_init_net(struct net *net)
537 {
538 	int res;
539 
540 	res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
541 	if (res < 0)
542 		return res;
543 
544 	net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
545 	net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
546 	net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
547 
548 	res = ip6_frags_ns_sysctl_register(net);
549 	if (res < 0)
550 		fqdir_exit(net->ipv6.fqdir);
551 	return res;
552 }
553 
554 static void __net_exit ipv6_frags_pre_exit_net(struct net *net)
555 {
556 	fqdir_pre_exit(net->ipv6.fqdir);
557 }
558 
559 static void __net_exit ipv6_frags_exit_net(struct net *net)
560 {
561 	ip6_frags_ns_sysctl_unregister(net);
562 	fqdir_exit(net->ipv6.fqdir);
563 }
564 
565 static struct pernet_operations ip6_frags_ops = {
566 	.init		= ipv6_frags_init_net,
567 	.pre_exit	= ipv6_frags_pre_exit_net,
568 	.exit		= ipv6_frags_exit_net,
569 };
570 
571 static const struct rhashtable_params ip6_rhash_params = {
572 	.head_offset		= offsetof(struct inet_frag_queue, node),
573 	.hashfn			= ip6frag_key_hashfn,
574 	.obj_hashfn		= ip6frag_obj_hashfn,
575 	.obj_cmpfn		= ip6frag_obj_cmpfn,
576 	.automatic_shrinking	= true,
577 };
578 
579 int __init ipv6_frag_init(void)
580 {
581 	int ret;
582 
583 	ip6_frags.constructor = ip6frag_init;
584 	ip6_frags.destructor = NULL;
585 	ip6_frags.qsize = sizeof(struct frag_queue);
586 	ip6_frags.frag_expire = ip6_frag_expire;
587 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
588 	ip6_frags.rhash_params = ip6_rhash_params;
589 	ret = inet_frags_init(&ip6_frags);
590 	if (ret)
591 		goto out;
592 
593 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
594 	if (ret)
595 		goto err_protocol;
596 
597 	ret = ip6_frags_sysctl_register();
598 	if (ret)
599 		goto err_sysctl;
600 
601 	ret = register_pernet_subsys(&ip6_frags_ops);
602 	if (ret)
603 		goto err_pernet;
604 
605 out:
606 	return ret;
607 
608 err_pernet:
609 	ip6_frags_sysctl_unregister();
610 err_sysctl:
611 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
612 err_protocol:
613 	inet_frags_fini(&ip6_frags);
614 	goto out;
615 }
616 
617 void ipv6_frag_exit(void)
618 {
619 	ip6_frags_sysctl_unregister();
620 	unregister_pernet_subsys(&ip6_frags_ops);
621 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
622 	inet_frags_fini(&ip6_frags);
623 }
624