xref: /openbmc/linux/net/ipv6/reassembly.c (revision 3a83e4e6) 
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 
46 #include <net/sock.h>
47 #include <net/snmp.h>
48 
49 #include <net/ipv6.h>
50 #include <net/ip6_route.h>
51 #include <net/protocol.h>
52 #include <net/transp_v6.h>
53 #include <net/rawv6.h>
54 #include <net/ndisc.h>
55 #include <net/addrconf.h>
56 #include <net/ipv6_frag.h>
57 #include <net/inet_ecn.h>
58 
59 static const char ip6_frag_cache_name[] = "ip6-frags";
60 
61 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
62 {
63 	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
64 }
65 
66 static struct inet_frags ip6_frags;
67 
68 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
69 			  struct sk_buff *prev_tail, struct net_device *dev);
70 
71 static void ip6_frag_expire(struct timer_list *t)
72 {
73 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
74 	struct frag_queue *fq;
75 
76 	fq = container_of(frag, struct frag_queue, q);
77 
78 	ip6frag_expire_frag_queue(fq->q.fqdir->net, fq);
79 }
80 
81 static struct frag_queue *
82 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
83 {
84 	struct frag_v6_compare_key key = {
85 		.id = id,
86 		.saddr = hdr->saddr,
87 		.daddr = hdr->daddr,
88 		.user = IP6_DEFRAG_LOCAL_DELIVER,
89 		.iif = iif,
90 	};
91 	struct inet_frag_queue *q;
92 
93 	if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
94 					    IPV6_ADDR_LINKLOCAL)))
95 		key.iif = 0;
96 
97 	q = inet_frag_find(net->ipv6.fqdir, &key);
98 	if (!q)
99 		return NULL;
100 
101 	return container_of(q, struct frag_queue, q);
102 }
103 
104 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
105 			  struct frag_hdr *fhdr, int nhoff,
106 			  u32 *prob_offset)
107 {
108 	struct net *net = dev_net(skb_dst(skb)->dev);
109 	int offset, end, fragsize;
110 	struct sk_buff *prev_tail;
111 	struct net_device *dev;
112 	int err = -ENOENT;
113 	u8 ecn;
114 
115 	if (fq->q.flags & INET_FRAG_COMPLETE)
116 		goto err;
117 
118 	err = -EINVAL;
119 	offset = ntohs(fhdr->frag_off) & ~0x7;
120 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
121 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
122 
123 	if ((unsigned int)end > IPV6_MAXPLEN) {
124 		*prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
125 		/* note that if prob_offset is set, the skb is freed elsewhere,
126 		 * we do not free it here.
127 		 */
128 		return -1;
129 	}
130 
131 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
132 
133 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
134 		const unsigned char *nh = skb_network_header(skb);
135 		skb->csum = csum_sub(skb->csum,
136 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
137 						  0));
138 	}
139 
140 	/* Is this the final fragment? */
141 	if (!(fhdr->frag_off & htons(IP6_MF))) {
142 		/* If we already have some bits beyond end
143 		 * or have different end, the segment is corrupted.
144 		 */
145 		if (end < fq->q.len ||
146 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
147 			goto discard_fq;
148 		fq->q.flags |= INET_FRAG_LAST_IN;
149 		fq->q.len = end;
150 	} else {
151 		/* Check if the fragment is rounded to 8 bytes.
152 		 * Required by the RFC.
153 		 */
154 		if (end & 0x7) {
155 			/* RFC2460 says always send parameter problem in
156 			 * this case. -DaveM
157 			 */
158 			*prob_offset = offsetof(struct ipv6hdr, payload_len);
159 			return -1;
160 		}
161 		if (end > fq->q.len) {
162 			/* Some bits beyond end -> corruption. */
163 			if (fq->q.flags & INET_FRAG_LAST_IN)
164 				goto discard_fq;
165 			fq->q.len = end;
166 		}
167 	}
168 
169 	if (end == offset)
170 		goto discard_fq;
171 
172 	err = -ENOMEM;
173 	/* Point into the IP datagram 'data' part. */
174 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
175 		goto discard_fq;
176 
177 	err = pskb_trim_rcsum(skb, end - offset);
178 	if (err)
179 		goto discard_fq;
180 
181 	/* Note : skb->rbnode and skb->dev share the same location. */
182 	dev = skb->dev;
183 	/* Makes sure compiler wont do silly aliasing games */
184 	barrier();
185 
186 	prev_tail = fq->q.fragments_tail;
187 	err = inet_frag_queue_insert(&fq->q, skb, offset, end);
188 	if (err)
189 		goto insert_error;
190 
191 	if (dev)
192 		fq->iif = dev->ifindex;
193 
194 	fq->q.stamp = skb->tstamp;
195 	fq->q.meat += skb->len;
196 	fq->ecn |= ecn;
197 	add_frag_mem_limit(fq->q.fqdir, skb->truesize);
198 
199 	fragsize = -skb_network_offset(skb) + skb->len;
200 	if (fragsize > fq->q.max_size)
201 		fq->q.max_size = fragsize;
202 
203 	/* The first fragment.
204 	 * nhoffset is obtained from the first fragment, of course.
205 	 */
206 	if (offset == 0) {
207 		fq->nhoffset = nhoff;
208 		fq->q.flags |= INET_FRAG_FIRST_IN;
209 	}
210 
211 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
212 	    fq->q.meat == fq->q.len) {
213 		unsigned long orefdst = skb->_skb_refdst;
214 
215 		skb->_skb_refdst = 0UL;
216 		err = ip6_frag_reasm(fq, skb, prev_tail, dev);
217 		skb->_skb_refdst = orefdst;
218 		return err;
219 	}
220 
221 	skb_dst_drop(skb);
222 	return -EINPROGRESS;
223 
224 insert_error:
225 	if (err == IPFRAG_DUP) {
226 		kfree_skb(skb);
227 		return -EINVAL;
228 	}
229 	err = -EINVAL;
230 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
231 			IPSTATS_MIB_REASM_OVERLAPS);
232 discard_fq:
233 	inet_frag_kill(&fq->q);
234 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
235 			IPSTATS_MIB_REASMFAILS);
236 err:
237 	kfree_skb(skb);
238 	return err;
239 }
240 
241 /*
242  *	Check if this packet is complete.
243  *
244  *	It is called with locked fq, and caller must check that
245  *	queue is eligible for reassembly i.e. it is not COMPLETE,
246  *	the last and the first frames arrived and all the bits are here.
247  */
248 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
249 			  struct sk_buff *prev_tail, struct net_device *dev)
250 {
251 	struct net *net = fq->q.fqdir->net;
252 	unsigned int nhoff;
253 	void *reasm_data;
254 	int payload_len;
255 	u8 ecn;
256 
257 	inet_frag_kill(&fq->q);
258 
259 	ecn = ip_frag_ecn_table[fq->ecn];
260 	if (unlikely(ecn == 0xff))
261 		goto out_fail;
262 
263 	reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
264 	if (!reasm_data)
265 		goto out_oom;
266 
267 	payload_len = ((skb->data - skb_network_header(skb)) -
268 		       sizeof(struct ipv6hdr) + fq->q.len -
269 		       sizeof(struct frag_hdr));
270 	if (payload_len > IPV6_MAXPLEN)
271 		goto out_oversize;
272 
273 	/* We have to remove fragment header from datagram and to relocate
274 	 * header in order to calculate ICV correctly. */
275 	nhoff = fq->nhoffset;
276 	skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
277 	memmove(skb->head + sizeof(struct frag_hdr), skb->head,
278 		(skb->data - skb->head) - sizeof(struct frag_hdr));
279 	if (skb_mac_header_was_set(skb))
280 		skb->mac_header += sizeof(struct frag_hdr);
281 	skb->network_header += sizeof(struct frag_hdr);
282 
283 	skb_reset_transport_header(skb);
284 
285 	inet_frag_reasm_finish(&fq->q, skb, reasm_data, true);
286 
287 	skb->dev = dev;
288 	ipv6_hdr(skb)->payload_len = htons(payload_len);
289 	ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
290 	IP6CB(skb)->nhoff = nhoff;
291 	IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
292 	IP6CB(skb)->frag_max_size = fq->q.max_size;
293 
294 	/* Yes, and fold redundant checksum back. 8) */
295 	skb_postpush_rcsum(skb, skb_network_header(skb),
296 			   skb_network_header_len(skb));
297 
298 	rcu_read_lock();
299 	__IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS);
300 	rcu_read_unlock();
301 	fq->q.rb_fragments = RB_ROOT;
302 	fq->q.fragments_tail = NULL;
303 	fq->q.last_run_head = NULL;
304 	return 1;
305 
306 out_oversize:
307 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
308 	goto out_fail;
309 out_oom:
310 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
311 out_fail:
312 	rcu_read_lock();
313 	__IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS);
314 	rcu_read_unlock();
315 	inet_frag_kill(&fq->q);
316 	return -1;
317 }
318 
319 static int ipv6_frag_rcv(struct sk_buff *skb)
320 {
321 	struct frag_hdr *fhdr;
322 	struct frag_queue *fq;
323 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
324 	struct net *net = dev_net(skb_dst(skb)->dev);
325 	int iif;
326 
327 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
328 		goto fail_hdr;
329 
330 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
331 
332 	/* Jumbo payload inhibits frag. header */
333 	if (hdr->payload_len == 0)
334 		goto fail_hdr;
335 
336 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
337 				 sizeof(struct frag_hdr))))
338 		goto fail_hdr;
339 
340 	hdr = ipv6_hdr(skb);
341 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
342 
343 	if (!(fhdr->frag_off & htons(0xFFF9))) {
344 		/* It is not a fragmented frame */
345 		skb->transport_header += sizeof(struct frag_hdr);
346 		__IP6_INC_STATS(net,
347 				ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
348 
349 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
350 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
351 		return 1;
352 	}
353 
354 	iif = skb->dev ? skb->dev->ifindex : 0;
355 	fq = fq_find(net, fhdr->identification, hdr, iif);
356 	if (fq) {
357 		u32 prob_offset = 0;
358 		int ret;
359 
360 		spin_lock(&fq->q.lock);
361 
362 		fq->iif = iif;
363 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
364 				     &prob_offset);
365 
366 		spin_unlock(&fq->q.lock);
367 		inet_frag_put(&fq->q);
368 		if (prob_offset) {
369 			__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
370 					IPSTATS_MIB_INHDRERRORS);
371 			/* icmpv6_param_prob() calls kfree_skb(skb) */
372 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
373 		}
374 		return ret;
375 	}
376 
377 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
378 	kfree_skb(skb);
379 	return -1;
380 
381 fail_hdr:
382 	__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
383 			IPSTATS_MIB_INHDRERRORS);
384 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
385 	return -1;
386 }
387 
388 static const struct inet6_protocol frag_protocol = {
389 	.handler	=	ipv6_frag_rcv,
390 	.flags		=	INET6_PROTO_NOPOLICY,
391 };
392 
393 #ifdef CONFIG_SYSCTL
394 
395 static struct ctl_table ip6_frags_ns_ctl_table[] = {
396 	{
397 		.procname	= "ip6frag_high_thresh",
398 		.maxlen		= sizeof(unsigned long),
399 		.mode		= 0644,
400 		.proc_handler	= proc_doulongvec_minmax,
401 	},
402 	{
403 		.procname	= "ip6frag_low_thresh",
404 		.maxlen		= sizeof(unsigned long),
405 		.mode		= 0644,
406 		.proc_handler	= proc_doulongvec_minmax,
407 	},
408 	{
409 		.procname	= "ip6frag_time",
410 		.maxlen		= sizeof(int),
411 		.mode		= 0644,
412 		.proc_handler	= proc_dointvec_jiffies,
413 	},
414 	{ }
415 };
416 
417 /* secret interval has been deprecated */
418 static int ip6_frags_secret_interval_unused;
419 static struct ctl_table ip6_frags_ctl_table[] = {
420 	{
421 		.procname	= "ip6frag_secret_interval",
422 		.data		= &ip6_frags_secret_interval_unused,
423 		.maxlen		= sizeof(int),
424 		.mode		= 0644,
425 		.proc_handler	= proc_dointvec_jiffies,
426 	},
427 	{ }
428 };
429 
430 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
431 {
432 	struct ctl_table *table;
433 	struct ctl_table_header *hdr;
434 
435 	table = ip6_frags_ns_ctl_table;
436 	if (!net_eq(net, &init_net)) {
437 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
438 		if (!table)
439 			goto err_alloc;
440 
441 	}
442 	table[0].data	= &net->ipv6.fqdir->high_thresh;
443 	table[0].extra1	= &net->ipv6.fqdir->low_thresh;
444 	table[1].data	= &net->ipv6.fqdir->low_thresh;
445 	table[1].extra2	= &net->ipv6.fqdir->high_thresh;
446 	table[2].data	= &net->ipv6.fqdir->timeout;
447 
448 	hdr = register_net_sysctl(net, "net/ipv6", table);
449 	if (!hdr)
450 		goto err_reg;
451 
452 	net->ipv6.sysctl.frags_hdr = hdr;
453 	return 0;
454 
455 err_reg:
456 	if (!net_eq(net, &init_net))
457 		kfree(table);
458 err_alloc:
459 	return -ENOMEM;
460 }
461 
462 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
463 {
464 	struct ctl_table *table;
465 
466 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
467 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
468 	if (!net_eq(net, &init_net))
469 		kfree(table);
470 }
471 
472 static struct ctl_table_header *ip6_ctl_header;
473 
474 static int ip6_frags_sysctl_register(void)
475 {
476 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
477 			ip6_frags_ctl_table);
478 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
479 }
480 
481 static void ip6_frags_sysctl_unregister(void)
482 {
483 	unregister_net_sysctl_table(ip6_ctl_header);
484 }
485 #else
486 static int ip6_frags_ns_sysctl_register(struct net *net)
487 {
488 	return 0;
489 }
490 
491 static void ip6_frags_ns_sysctl_unregister(struct net *net)
492 {
493 }
494 
495 static int ip6_frags_sysctl_register(void)
496 {
497 	return 0;
498 }
499 
500 static void ip6_frags_sysctl_unregister(void)
501 {
502 }
503 #endif
504 
505 static int __net_init ipv6_frags_init_net(struct net *net)
506 {
507 	int res;
508 
509 	res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
510 	if (res < 0)
511 		return res;
512 
513 	net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
514 	net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
515 	net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
516 
517 	res = ip6_frags_ns_sysctl_register(net);
518 	if (res < 0)
519 		fqdir_exit(net->ipv6.fqdir);
520 	return res;
521 }
522 
523 static void __net_exit ipv6_frags_pre_exit_net(struct net *net)
524 {
525 	fqdir_pre_exit(net->ipv6.fqdir);
526 }
527 
528 static void __net_exit ipv6_frags_exit_net(struct net *net)
529 {
530 	ip6_frags_ns_sysctl_unregister(net);
531 	fqdir_exit(net->ipv6.fqdir);
532 }
533 
534 static struct pernet_operations ip6_frags_ops = {
535 	.init		= ipv6_frags_init_net,
536 	.pre_exit	= ipv6_frags_pre_exit_net,
537 	.exit		= ipv6_frags_exit_net,
538 };
539 
540 static const struct rhashtable_params ip6_rhash_params = {
541 	.head_offset		= offsetof(struct inet_frag_queue, node),
542 	.hashfn			= ip6frag_key_hashfn,
543 	.obj_hashfn		= ip6frag_obj_hashfn,
544 	.obj_cmpfn		= ip6frag_obj_cmpfn,
545 	.automatic_shrinking	= true,
546 };
547 
548 int __init ipv6_frag_init(void)
549 {
550 	int ret;
551 
552 	ip6_frags.constructor = ip6frag_init;
553 	ip6_frags.destructor = NULL;
554 	ip6_frags.qsize = sizeof(struct frag_queue);
555 	ip6_frags.frag_expire = ip6_frag_expire;
556 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
557 	ip6_frags.rhash_params = ip6_rhash_params;
558 	ret = inet_frags_init(&ip6_frags);
559 	if (ret)
560 		goto out;
561 
562 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
563 	if (ret)
564 		goto err_protocol;
565 
566 	ret = ip6_frags_sysctl_register();
567 	if (ret)
568 		goto err_sysctl;
569 
570 	ret = register_pernet_subsys(&ip6_frags_ops);
571 	if (ret)
572 		goto err_pernet;
573 
574 out:
575 	return ret;
576 
577 err_pernet:
578 	ip6_frags_sysctl_unregister();
579 err_sysctl:
580 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
581 err_protocol:
582 	inet_frags_fini(&ip6_frags);
583 	goto out;
584 }
585 
586 void ipv6_frag_exit(void)
587 {
588 	ip6_frags_sysctl_unregister();
589 	unregister_pernet_subsys(&ip6_frags_ops);
590 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
591 	inet_frags_fini(&ip6_frags);
592 }
593