1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * IPv6 fragment reassembly for connection tracking
4  *
5  * Copyright (C)2004 USAGI/WIDE Project
6  *
7  * Author:
8  *	Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
9  *
10  * Based on: net/ipv6/reassembly.c
11  */
12 
13 #define pr_fmt(fmt) "IPv6-nf: " fmt
14 
15 #include <linux/errno.h>
16 #include <linux/types.h>
17 #include <linux/string.h>
18 #include <linux/socket.h>
19 #include <linux/sockios.h>
20 #include <linux/jiffies.h>
21 #include <linux/net.h>
22 #include <linux/list.h>
23 #include <linux/netdevice.h>
24 #include <linux/in6.h>
25 #include <linux/ipv6.h>
26 #include <linux/icmpv6.h>
27 #include <linux/random.h>
28 #include <linux/slab.h>
29 
30 #include <net/sock.h>
31 #include <net/snmp.h>
32 #include <net/ipv6_frag.h>
33 
34 #include <net/protocol.h>
35 #include <net/transp_v6.h>
36 #include <net/rawv6.h>
37 #include <net/ndisc.h>
38 #include <net/addrconf.h>
39 #include <net/inet_ecn.h>
40 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
41 #include <linux/sysctl.h>
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
44 #include <linux/kernel.h>
45 #include <linux/module.h>
46 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
47 
48 static const char nf_frags_cache_name[] = "nf-frags";
49 
50 static struct inet_frags nf_frags;
51 
52 #ifdef CONFIG_SYSCTL
53 
54 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
55 	{
56 		.procname	= "nf_conntrack_frag6_timeout",
57 		.data		= &init_net.nf_frag.frags.timeout,
58 		.maxlen		= sizeof(unsigned int),
59 		.mode		= 0644,
60 		.proc_handler	= proc_dointvec_jiffies,
61 	},
62 	{
63 		.procname	= "nf_conntrack_frag6_low_thresh",
64 		.data		= &init_net.nf_frag.frags.low_thresh,
65 		.maxlen		= sizeof(unsigned long),
66 		.mode		= 0644,
67 		.proc_handler	= proc_doulongvec_minmax,
68 		.extra2		= &init_net.nf_frag.frags.high_thresh
69 	},
70 	{
71 		.procname	= "nf_conntrack_frag6_high_thresh",
72 		.data		= &init_net.nf_frag.frags.high_thresh,
73 		.maxlen		= sizeof(unsigned long),
74 		.mode		= 0644,
75 		.proc_handler	= proc_doulongvec_minmax,
76 		.extra1		= &init_net.nf_frag.frags.low_thresh
77 	},
78 	{ }
79 };
80 
81 static int nf_ct_frag6_sysctl_register(struct net *net)
82 {
83 	struct ctl_table *table;
84 	struct ctl_table_header *hdr;
85 
86 	table = nf_ct_frag6_sysctl_table;
87 	if (!net_eq(net, &init_net)) {
88 		table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
89 				GFP_KERNEL);
90 		if (table == NULL)
91 			goto err_alloc;
92 
93 		table[0].data = &net->nf_frag.frags.timeout;
94 		table[1].data = &net->nf_frag.frags.low_thresh;
95 		table[1].extra2 = &net->nf_frag.frags.high_thresh;
96 		table[2].data = &net->nf_frag.frags.high_thresh;
97 		table[2].extra1 = &net->nf_frag.frags.low_thresh;
98 		table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
99 	}
100 
101 	hdr = register_net_sysctl(net, "net/netfilter", table);
102 	if (hdr == NULL)
103 		goto err_reg;
104 
105 	net->nf_frag_frags_hdr = hdr;
106 	return 0;
107 
108 err_reg:
109 	if (!net_eq(net, &init_net))
110 		kfree(table);
111 err_alloc:
112 	return -ENOMEM;
113 }
114 
115 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
116 {
117 	struct ctl_table *table;
118 
119 	table = net->nf_frag_frags_hdr->ctl_table_arg;
120 	unregister_net_sysctl_table(net->nf_frag_frags_hdr);
121 	if (!net_eq(net, &init_net))
122 		kfree(table);
123 }
124 
125 #else
126 static int nf_ct_frag6_sysctl_register(struct net *net)
127 {
128 	return 0;
129 }
130 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
131 {
132 }
133 #endif
134 
135 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
136 			     struct sk_buff *prev_tail, struct net_device *dev);
137 
138 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
139 {
140 	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
141 }
142 
143 static void nf_ct_frag6_expire(struct timer_list *t)
144 {
145 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
146 	struct frag_queue *fq;
147 	struct net *net;
148 
149 	fq = container_of(frag, struct frag_queue, q);
150 	net = container_of(fq->q.net, struct net, nf_frag.frags);
151 
152 	ip6frag_expire_frag_queue(net, fq);
153 }
154 
155 /* Creation primitives. */
156 static struct frag_queue *fq_find(struct net *net, __be32 id, u32 user,
157 				  const struct ipv6hdr *hdr, int iif)
158 {
159 	struct frag_v6_compare_key key = {
160 		.id = id,
161 		.saddr = hdr->saddr,
162 		.daddr = hdr->daddr,
163 		.user = user,
164 		.iif = iif,
165 	};
166 	struct inet_frag_queue *q;
167 
168 	q = inet_frag_find(&net->nf_frag.frags, &key);
169 	if (!q)
170 		return NULL;
171 
172 	return container_of(q, struct frag_queue, q);
173 }
174 
175 
176 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
177 			     const struct frag_hdr *fhdr, int nhoff)
178 {
179 	unsigned int payload_len;
180 	struct net_device *dev;
181 	struct sk_buff *prev;
182 	int offset, end, err;
183 	u8 ecn;
184 
185 	if (fq->q.flags & INET_FRAG_COMPLETE) {
186 		pr_debug("Already completed\n");
187 		goto err;
188 	}
189 
190 	payload_len = ntohs(ipv6_hdr(skb)->payload_len);
191 
192 	offset = ntohs(fhdr->frag_off) & ~0x7;
193 	end = offset + (payload_len -
194 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
195 
196 	if ((unsigned int)end > IPV6_MAXPLEN) {
197 		pr_debug("offset is too large.\n");
198 		return -EINVAL;
199 	}
200 
201 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
202 
203 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
204 		const unsigned char *nh = skb_network_header(skb);
205 		skb->csum = csum_sub(skb->csum,
206 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
207 						  0));
208 	}
209 
210 	/* Is this the final fragment? */
211 	if (!(fhdr->frag_off & htons(IP6_MF))) {
212 		/* If we already have some bits beyond end
213 		 * or have different end, the segment is corrupted.
214 		 */
215 		if (end < fq->q.len ||
216 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
217 			pr_debug("already received last fragment\n");
218 			goto err;
219 		}
220 		fq->q.flags |= INET_FRAG_LAST_IN;
221 		fq->q.len = end;
222 	} else {
223 		/* Check if the fragment is rounded to 8 bytes.
224 		 * Required by the RFC.
225 		 */
226 		if (end & 0x7) {
227 			/* RFC2460 says always send parameter problem in
228 			 * this case. -DaveM
229 			 */
230 			pr_debug("end of fragment not rounded to 8 bytes.\n");
231 			inet_frag_kill(&fq->q);
232 			return -EPROTO;
233 		}
234 		if (end > fq->q.len) {
235 			/* Some bits beyond end -> corruption. */
236 			if (fq->q.flags & INET_FRAG_LAST_IN) {
237 				pr_debug("last packet already reached.\n");
238 				goto err;
239 			}
240 			fq->q.len = end;
241 		}
242 	}
243 
244 	if (end == offset)
245 		goto err;
246 
247 	/* Point into the IP datagram 'data' part. */
248 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
249 		pr_debug("queue: message is too short.\n");
250 		goto err;
251 	}
252 	if (pskb_trim_rcsum(skb, end - offset)) {
253 		pr_debug("Can't trim\n");
254 		goto err;
255 	}
256 
257 	/* Note : skb->rbnode and skb->dev share the same location. */
258 	dev = skb->dev;
259 	/* Makes sure compiler wont do silly aliasing games */
260 	barrier();
261 
262 	prev = fq->q.fragments_tail;
263 	err = inet_frag_queue_insert(&fq->q, skb, offset, end);
264 	if (err)
265 		goto insert_error;
266 
267 	if (dev)
268 		fq->iif = dev->ifindex;
269 
270 	fq->q.stamp = skb->tstamp;
271 	fq->q.meat += skb->len;
272 	fq->ecn |= ecn;
273 	if (payload_len > fq->q.max_size)
274 		fq->q.max_size = payload_len;
275 	add_frag_mem_limit(fq->q.net, skb->truesize);
276 
277 	/* The first fragment.
278 	 * nhoffset is obtained from the first fragment, of course.
279 	 */
280 	if (offset == 0) {
281 		fq->nhoffset = nhoff;
282 		fq->q.flags |= INET_FRAG_FIRST_IN;
283 	}
284 
285 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
286 	    fq->q.meat == fq->q.len) {
287 		unsigned long orefdst = skb->_skb_refdst;
288 
289 		skb->_skb_refdst = 0UL;
290 		err = nf_ct_frag6_reasm(fq, skb, prev, dev);
291 		skb->_skb_refdst = orefdst;
292 		return err;
293 	}
294 
295 	skb_dst_drop(skb);
296 	return -EINPROGRESS;
297 
298 insert_error:
299 	if (err == IPFRAG_DUP)
300 		goto err;
301 	inet_frag_kill(&fq->q);
302 err:
303 	skb_dst_drop(skb);
304 	return -EINVAL;
305 }
306 
307 /*
308  *	Check if this packet is complete.
309  *
310  *	It is called with locked fq, and caller must check that
311  *	queue is eligible for reassembly i.e. it is not COMPLETE,
312  *	the last and the first frames arrived and all the bits are here.
313  */
314 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
315 			     struct sk_buff *prev_tail, struct net_device *dev)
316 {
317 	void *reasm_data;
318 	int payload_len;
319 	u8 ecn;
320 
321 	inet_frag_kill(&fq->q);
322 
323 	ecn = ip_frag_ecn_table[fq->ecn];
324 	if (unlikely(ecn == 0xff))
325 		goto err;
326 
327 	reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
328 	if (!reasm_data)
329 		goto err;
330 
331 	payload_len = ((skb->data - skb_network_header(skb)) -
332 		       sizeof(struct ipv6hdr) + fq->q.len -
333 		       sizeof(struct frag_hdr));
334 	if (payload_len > IPV6_MAXPLEN) {
335 		net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
336 				    payload_len);
337 		goto err;
338 	}
339 
340 	/* We have to remove fragment header from datagram and to relocate
341 	 * header in order to calculate ICV correctly. */
342 	skb_network_header(skb)[fq->nhoffset] = skb_transport_header(skb)[0];
343 	memmove(skb->head + sizeof(struct frag_hdr), skb->head,
344 		(skb->data - skb->head) - sizeof(struct frag_hdr));
345 	skb->mac_header += sizeof(struct frag_hdr);
346 	skb->network_header += sizeof(struct frag_hdr);
347 
348 	skb_reset_transport_header(skb);
349 
350 	inet_frag_reasm_finish(&fq->q, skb, reasm_data);
351 
352 	skb->ignore_df = 1;
353 	skb->dev = dev;
354 	ipv6_hdr(skb)->payload_len = htons(payload_len);
355 	ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
356 	IP6CB(skb)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
357 
358 	/* Yes, and fold redundant checksum back. 8) */
359 	if (skb->ip_summed == CHECKSUM_COMPLETE)
360 		skb->csum = csum_partial(skb_network_header(skb),
361 					 skb_network_header_len(skb),
362 					 skb->csum);
363 
364 	fq->q.rb_fragments = RB_ROOT;
365 	fq->q.fragments_tail = NULL;
366 	fq->q.last_run_head = NULL;
367 
368 	return 0;
369 
370 err:
371 	inet_frag_kill(&fq->q);
372 	return -EINVAL;
373 }
374 
375 /*
376  * find the header just before Fragment Header.
377  *
378  * if success return 0 and set ...
379  * (*prevhdrp): the value of "Next Header Field" in the header
380  *		just before Fragment Header.
381  * (*prevhoff): the offset of "Next Header Field" in the header
382  *		just before Fragment Header.
383  * (*fhoff)   : the offset of Fragment Header.
384  *
385  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
386  *
387  */
388 static int
389 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
390 {
391 	u8 nexthdr = ipv6_hdr(skb)->nexthdr;
392 	const int netoff = skb_network_offset(skb);
393 	u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
394 	int start = netoff + sizeof(struct ipv6hdr);
395 	int len = skb->len - start;
396 	u8 prevhdr = NEXTHDR_IPV6;
397 
398 	while (nexthdr != NEXTHDR_FRAGMENT) {
399 		struct ipv6_opt_hdr hdr;
400 		int hdrlen;
401 
402 		if (!ipv6_ext_hdr(nexthdr)) {
403 			return -1;
404 		}
405 		if (nexthdr == NEXTHDR_NONE) {
406 			pr_debug("next header is none\n");
407 			return -1;
408 		}
409 		if (len < (int)sizeof(struct ipv6_opt_hdr)) {
410 			pr_debug("too short\n");
411 			return -1;
412 		}
413 		if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
414 			BUG();
415 		if (nexthdr == NEXTHDR_AUTH)
416 			hdrlen = (hdr.hdrlen+2)<<2;
417 		else
418 			hdrlen = ipv6_optlen(&hdr);
419 
420 		prevhdr = nexthdr;
421 		prev_nhoff = start;
422 
423 		nexthdr = hdr.nexthdr;
424 		len -= hdrlen;
425 		start += hdrlen;
426 	}
427 
428 	if (len < 0)
429 		return -1;
430 
431 	*prevhdrp = prevhdr;
432 	*prevhoff = prev_nhoff;
433 	*fhoff = start;
434 
435 	return 0;
436 }
437 
438 int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
439 {
440 	u16 savethdr = skb->transport_header;
441 	int fhoff, nhoff, ret;
442 	struct frag_hdr *fhdr;
443 	struct frag_queue *fq;
444 	struct ipv6hdr *hdr;
445 	u8 prevhdr;
446 
447 	/* Jumbo payload inhibits frag. header */
448 	if (ipv6_hdr(skb)->payload_len == 0) {
449 		pr_debug("payload len = 0\n");
450 		return 0;
451 	}
452 
453 	if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
454 		return 0;
455 
456 	if (!pskb_may_pull(skb, fhoff + sizeof(*fhdr)))
457 		return -ENOMEM;
458 
459 	skb_set_transport_header(skb, fhoff);
460 	hdr = ipv6_hdr(skb);
461 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
462 
463 	skb_orphan(skb);
464 	fq = fq_find(net, fhdr->identification, user, hdr,
465 		     skb->dev ? skb->dev->ifindex : 0);
466 	if (fq == NULL) {
467 		pr_debug("Can't find and can't create new queue\n");
468 		return -ENOMEM;
469 	}
470 
471 	spin_lock_bh(&fq->q.lock);
472 
473 	ret = nf_ct_frag6_queue(fq, skb, fhdr, nhoff);
474 	if (ret == -EPROTO) {
475 		skb->transport_header = savethdr;
476 		ret = 0;
477 	}
478 
479 	/* after queue has assumed skb ownership, only 0 or -EINPROGRESS
480 	 * must be returned.
481 	 */
482 	if (ret)
483 		ret = -EINPROGRESS;
484 
485 	spin_unlock_bh(&fq->q.lock);
486 	inet_frag_put(&fq->q);
487 	return ret;
488 }
489 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
490 
491 static int nf_ct_net_init(struct net *net)
492 {
493 	int res;
494 
495 	net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
496 	net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
497 	net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
498 	net->nf_frag.frags.f = &nf_frags;
499 
500 	res = inet_frags_init_net(&net->nf_frag.frags);
501 	if (res < 0)
502 		return res;
503 	res = nf_ct_frag6_sysctl_register(net);
504 	if (res < 0)
505 		inet_frags_exit_net(&net->nf_frag.frags);
506 	return res;
507 }
508 
509 static void nf_ct_net_exit(struct net *net)
510 {
511 	nf_ct_frags6_sysctl_unregister(net);
512 	inet_frags_exit_net(&net->nf_frag.frags);
513 }
514 
515 static struct pernet_operations nf_ct_net_ops = {
516 	.init = nf_ct_net_init,
517 	.exit = nf_ct_net_exit,
518 };
519 
520 static const struct rhashtable_params nfct_rhash_params = {
521 	.head_offset		= offsetof(struct inet_frag_queue, node),
522 	.hashfn			= ip6frag_key_hashfn,
523 	.obj_hashfn		= ip6frag_obj_hashfn,
524 	.obj_cmpfn		= ip6frag_obj_cmpfn,
525 	.automatic_shrinking	= true,
526 };
527 
528 int nf_ct_frag6_init(void)
529 {
530 	int ret = 0;
531 
532 	nf_frags.constructor = ip6frag_init;
533 	nf_frags.destructor = NULL;
534 	nf_frags.qsize = sizeof(struct frag_queue);
535 	nf_frags.frag_expire = nf_ct_frag6_expire;
536 	nf_frags.frags_cache_name = nf_frags_cache_name;
537 	nf_frags.rhash_params = nfct_rhash_params;
538 	ret = inet_frags_init(&nf_frags);
539 	if (ret)
540 		goto out;
541 	ret = register_pernet_subsys(&nf_ct_net_ops);
542 	if (ret)
543 		inet_frags_fini(&nf_frags);
544 
545 out:
546 	return ret;
547 }
548 
549 void nf_ct_frag6_cleanup(void)
550 {
551 	unregister_pernet_subsys(&nf_ct_net_ops);
552 	inet_frags_fini(&nf_frags);
553 }
554