xref: /openbmc/linux/net/sched/act_csum.c (revision 4e1a33b1)
1 /*
2  * Checksum updating actions
3  *
4  * Copyright (c) 2010 Gregoire Baron <baronchon@n7mm.org>
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License as published by the Free
8  * Software Foundation; either version 2 of the License, or (at your option)
9  * any later version.
10  *
11  */
12 
13 #include <linux/types.h>
14 #include <linux/init.h>
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/spinlock.h>
18 
19 #include <linux/netlink.h>
20 #include <net/netlink.h>
21 #include <linux/rtnetlink.h>
22 
23 #include <linux/skbuff.h>
24 
25 #include <net/ip.h>
26 #include <net/ipv6.h>
27 #include <net/icmp.h>
28 #include <linux/icmpv6.h>
29 #include <linux/igmp.h>
30 #include <net/tcp.h>
31 #include <net/udp.h>
32 #include <net/ip6_checksum.h>
33 #include <net/sctp/checksum.h>
34 
35 #include <net/act_api.h>
36 
37 #include <linux/tc_act/tc_csum.h>
38 #include <net/tc_act/tc_csum.h>
39 
40 #define CSUM_TAB_MASK 15
41 
42 static const struct nla_policy csum_policy[TCA_CSUM_MAX + 1] = {
43 	[TCA_CSUM_PARMS] = { .len = sizeof(struct tc_csum), },
44 };
45 
46 static unsigned int csum_net_id;
47 static struct tc_action_ops act_csum_ops;
48 
49 static int tcf_csum_init(struct net *net, struct nlattr *nla,
50 			 struct nlattr *est, struct tc_action **a, int ovr,
51 			 int bind)
52 {
53 	struct tc_action_net *tn = net_generic(net, csum_net_id);
54 	struct nlattr *tb[TCA_CSUM_MAX + 1];
55 	struct tc_csum *parm;
56 	struct tcf_csum *p;
57 	int ret = 0, err;
58 
59 	if (nla == NULL)
60 		return -EINVAL;
61 
62 	err = nla_parse_nested(tb, TCA_CSUM_MAX, nla, csum_policy);
63 	if (err < 0)
64 		return err;
65 
66 	if (tb[TCA_CSUM_PARMS] == NULL)
67 		return -EINVAL;
68 	parm = nla_data(tb[TCA_CSUM_PARMS]);
69 
70 	if (!tcf_hash_check(tn, parm->index, a, bind)) {
71 		ret = tcf_hash_create(tn, parm->index, est, a,
72 				      &act_csum_ops, bind, false);
73 		if (ret)
74 			return ret;
75 		ret = ACT_P_CREATED;
76 	} else {
77 		if (bind)/* dont override defaults */
78 			return 0;
79 		tcf_hash_release(*a, bind);
80 		if (!ovr)
81 			return -EEXIST;
82 	}
83 
84 	p = to_tcf_csum(*a);
85 	spin_lock_bh(&p->tcf_lock);
86 	p->tcf_action = parm->action;
87 	p->update_flags = parm->update_flags;
88 	spin_unlock_bh(&p->tcf_lock);
89 
90 	if (ret == ACT_P_CREATED)
91 		tcf_hash_insert(tn, *a);
92 
93 	return ret;
94 }
95 
96 /**
97  * tcf_csum_skb_nextlayer - Get next layer pointer
98  * @skb: sk_buff to use
99  * @ihl: previous summed headers length
100  * @ipl: complete packet length
101  * @jhl: next header length
102  *
103  * Check the expected next layer availability in the specified sk_buff.
104  * Return the next layer pointer if pass, NULL otherwise.
105  */
106 static void *tcf_csum_skb_nextlayer(struct sk_buff *skb,
107 				    unsigned int ihl, unsigned int ipl,
108 				    unsigned int jhl)
109 {
110 	int ntkoff = skb_network_offset(skb);
111 	int hl = ihl + jhl;
112 
113 	if (!pskb_may_pull(skb, ipl + ntkoff) || (ipl < hl) ||
114 	    skb_try_make_writable(skb, hl + ntkoff))
115 		return NULL;
116 	else
117 		return (void *)(skb_network_header(skb) + ihl);
118 }
119 
120 static int tcf_csum_ipv4_icmp(struct sk_buff *skb, unsigned int ihl,
121 			      unsigned int ipl)
122 {
123 	struct icmphdr *icmph;
124 
125 	icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
126 	if (icmph == NULL)
127 		return 0;
128 
129 	icmph->checksum = 0;
130 	skb->csum = csum_partial(icmph, ipl - ihl, 0);
131 	icmph->checksum = csum_fold(skb->csum);
132 
133 	skb->ip_summed = CHECKSUM_NONE;
134 
135 	return 1;
136 }
137 
138 static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
139 			      unsigned int ihl, unsigned int ipl)
140 {
141 	struct igmphdr *igmph;
142 
143 	igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
144 	if (igmph == NULL)
145 		return 0;
146 
147 	igmph->csum = 0;
148 	skb->csum = csum_partial(igmph, ipl - ihl, 0);
149 	igmph->csum = csum_fold(skb->csum);
150 
151 	skb->ip_summed = CHECKSUM_NONE;
152 
153 	return 1;
154 }
155 
156 static int tcf_csum_ipv6_icmp(struct sk_buff *skb, unsigned int ihl,
157 			      unsigned int ipl)
158 {
159 	struct icmp6hdr *icmp6h;
160 	const struct ipv6hdr *ip6h;
161 
162 	icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
163 	if (icmp6h == NULL)
164 		return 0;
165 
166 	ip6h = ipv6_hdr(skb);
167 	icmp6h->icmp6_cksum = 0;
168 	skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
169 	icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
170 					      ipl - ihl, IPPROTO_ICMPV6,
171 					      skb->csum);
172 
173 	skb->ip_summed = CHECKSUM_NONE;
174 
175 	return 1;
176 }
177 
178 static int tcf_csum_ipv4_tcp(struct sk_buff *skb, unsigned int ihl,
179 			     unsigned int ipl)
180 {
181 	struct tcphdr *tcph;
182 	const struct iphdr *iph;
183 
184 	tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
185 	if (tcph == NULL)
186 		return 0;
187 
188 	iph = ip_hdr(skb);
189 	tcph->check = 0;
190 	skb->csum = csum_partial(tcph, ipl - ihl, 0);
191 	tcph->check = tcp_v4_check(ipl - ihl,
192 				   iph->saddr, iph->daddr, skb->csum);
193 
194 	skb->ip_summed = CHECKSUM_NONE;
195 
196 	return 1;
197 }
198 
199 static int tcf_csum_ipv6_tcp(struct sk_buff *skb, unsigned int ihl,
200 			     unsigned int ipl)
201 {
202 	struct tcphdr *tcph;
203 	const struct ipv6hdr *ip6h;
204 
205 	tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
206 	if (tcph == NULL)
207 		return 0;
208 
209 	ip6h = ipv6_hdr(skb);
210 	tcph->check = 0;
211 	skb->csum = csum_partial(tcph, ipl - ihl, 0);
212 	tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
213 				      ipl - ihl, IPPROTO_TCP,
214 				      skb->csum);
215 
216 	skb->ip_summed = CHECKSUM_NONE;
217 
218 	return 1;
219 }
220 
221 static int tcf_csum_ipv4_udp(struct sk_buff *skb, unsigned int ihl,
222 			     unsigned int ipl, int udplite)
223 {
224 	struct udphdr *udph;
225 	const struct iphdr *iph;
226 	u16 ul;
227 
228 	/*
229 	 * Support both UDP and UDPLITE checksum algorithms, Don't use
230 	 * udph->len to get the real length without any protocol check,
231 	 * UDPLITE uses udph->len for another thing,
232 	 * Use iph->tot_len, or just ipl.
233 	 */
234 
235 	udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
236 	if (udph == NULL)
237 		return 0;
238 
239 	iph = ip_hdr(skb);
240 	ul = ntohs(udph->len);
241 
242 	if (udplite || udph->check) {
243 
244 		udph->check = 0;
245 
246 		if (udplite) {
247 			if (ul == 0)
248 				skb->csum = csum_partial(udph, ipl - ihl, 0);
249 			else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
250 				skb->csum = csum_partial(udph, ul, 0);
251 			else
252 				goto ignore_obscure_skb;
253 		} else {
254 			if (ul != ipl - ihl)
255 				goto ignore_obscure_skb;
256 
257 			skb->csum = csum_partial(udph, ul, 0);
258 		}
259 
260 		udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
261 						ul, iph->protocol,
262 						skb->csum);
263 
264 		if (!udph->check)
265 			udph->check = CSUM_MANGLED_0;
266 	}
267 
268 	skb->ip_summed = CHECKSUM_NONE;
269 
270 ignore_obscure_skb:
271 	return 1;
272 }
273 
274 static int tcf_csum_ipv6_udp(struct sk_buff *skb, unsigned int ihl,
275 			     unsigned int ipl, int udplite)
276 {
277 	struct udphdr *udph;
278 	const struct ipv6hdr *ip6h;
279 	u16 ul;
280 
281 	/*
282 	 * Support both UDP and UDPLITE checksum algorithms, Don't use
283 	 * udph->len to get the real length without any protocol check,
284 	 * UDPLITE uses udph->len for another thing,
285 	 * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
286 	 */
287 
288 	udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
289 	if (udph == NULL)
290 		return 0;
291 
292 	ip6h = ipv6_hdr(skb);
293 	ul = ntohs(udph->len);
294 
295 	udph->check = 0;
296 
297 	if (udplite) {
298 		if (ul == 0)
299 			skb->csum = csum_partial(udph, ipl - ihl, 0);
300 
301 		else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
302 			skb->csum = csum_partial(udph, ul, 0);
303 
304 		else
305 			goto ignore_obscure_skb;
306 	} else {
307 		if (ul != ipl - ihl)
308 			goto ignore_obscure_skb;
309 
310 		skb->csum = csum_partial(udph, ul, 0);
311 	}
312 
313 	udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
314 				      udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
315 				      skb->csum);
316 
317 	if (!udph->check)
318 		udph->check = CSUM_MANGLED_0;
319 
320 	skb->ip_summed = CHECKSUM_NONE;
321 
322 ignore_obscure_skb:
323 	return 1;
324 }
325 
326 static int tcf_csum_sctp(struct sk_buff *skb, unsigned int ihl,
327 			 unsigned int ipl)
328 {
329 	struct sctphdr *sctph;
330 
331 	if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_SCTP)
332 		return 1;
333 
334 	sctph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*sctph));
335 	if (!sctph)
336 		return 0;
337 
338 	sctph->checksum = sctp_compute_cksum(skb,
339 					     skb_network_offset(skb) + ihl);
340 	skb->ip_summed = CHECKSUM_NONE;
341 
342 	return 1;
343 }
344 
345 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
346 {
347 	const struct iphdr *iph;
348 	int ntkoff;
349 
350 	ntkoff = skb_network_offset(skb);
351 
352 	if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
353 		goto fail;
354 
355 	iph = ip_hdr(skb);
356 
357 	switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
358 	case IPPROTO_ICMP:
359 		if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
360 			if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
361 						ntohs(iph->tot_len)))
362 				goto fail;
363 		break;
364 	case IPPROTO_IGMP:
365 		if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
366 			if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
367 						ntohs(iph->tot_len)))
368 				goto fail;
369 		break;
370 	case IPPROTO_TCP:
371 		if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
372 			if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
373 					       ntohs(iph->tot_len)))
374 				goto fail;
375 		break;
376 	case IPPROTO_UDP:
377 		if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
378 			if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
379 					       ntohs(iph->tot_len), 0))
380 				goto fail;
381 		break;
382 	case IPPROTO_UDPLITE:
383 		if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
384 			if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
385 					       ntohs(iph->tot_len), 1))
386 				goto fail;
387 		break;
388 	case IPPROTO_SCTP:
389 		if ((update_flags & TCA_CSUM_UPDATE_FLAG_SCTP) &&
390 		    !tcf_csum_sctp(skb, iph->ihl * 4, ntohs(iph->tot_len)))
391 			goto fail;
392 		break;
393 	}
394 
395 	if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
396 		if (skb_try_make_writable(skb, sizeof(*iph) + ntkoff))
397 			goto fail;
398 
399 		ip_send_check(ip_hdr(skb));
400 	}
401 
402 	return 1;
403 
404 fail:
405 	return 0;
406 }
407 
408 static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh, unsigned int ixhl,
409 				 unsigned int *pl)
410 {
411 	int off, len, optlen;
412 	unsigned char *xh = (void *)ip6xh;
413 
414 	off = sizeof(*ip6xh);
415 	len = ixhl - off;
416 
417 	while (len > 1) {
418 		switch (xh[off]) {
419 		case IPV6_TLV_PAD1:
420 			optlen = 1;
421 			break;
422 		case IPV6_TLV_JUMBO:
423 			optlen = xh[off + 1] + 2;
424 			if (optlen != 6 || len < 6 || (off & 3) != 2)
425 				/* wrong jumbo option length/alignment */
426 				return 0;
427 			*pl = ntohl(*(__be32 *)(xh + off + 2));
428 			goto done;
429 		default:
430 			optlen = xh[off + 1] + 2;
431 			if (optlen > len)
432 				/* ignore obscure options */
433 				goto done;
434 			break;
435 		}
436 		off += optlen;
437 		len -= optlen;
438 	}
439 
440 done:
441 	return 1;
442 }
443 
444 static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
445 {
446 	struct ipv6hdr *ip6h;
447 	struct ipv6_opt_hdr *ip6xh;
448 	unsigned int hl, ixhl;
449 	unsigned int pl;
450 	int ntkoff;
451 	u8 nexthdr;
452 
453 	ntkoff = skb_network_offset(skb);
454 
455 	hl = sizeof(*ip6h);
456 
457 	if (!pskb_may_pull(skb, hl + ntkoff))
458 		goto fail;
459 
460 	ip6h = ipv6_hdr(skb);
461 
462 	pl = ntohs(ip6h->payload_len);
463 	nexthdr = ip6h->nexthdr;
464 
465 	do {
466 		switch (nexthdr) {
467 		case NEXTHDR_FRAGMENT:
468 			goto ignore_skb;
469 		case NEXTHDR_ROUTING:
470 		case NEXTHDR_HOP:
471 		case NEXTHDR_DEST:
472 			if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
473 				goto fail;
474 			ip6xh = (void *)(skb_network_header(skb) + hl);
475 			ixhl = ipv6_optlen(ip6xh);
476 			if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
477 				goto fail;
478 			ip6xh = (void *)(skb_network_header(skb) + hl);
479 			if ((nexthdr == NEXTHDR_HOP) &&
480 			    !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
481 				goto fail;
482 			nexthdr = ip6xh->nexthdr;
483 			hl += ixhl;
484 			break;
485 		case IPPROTO_ICMPV6:
486 			if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
487 				if (!tcf_csum_ipv6_icmp(skb,
488 							hl, pl + sizeof(*ip6h)))
489 					goto fail;
490 			goto done;
491 		case IPPROTO_TCP:
492 			if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
493 				if (!tcf_csum_ipv6_tcp(skb,
494 						       hl, pl + sizeof(*ip6h)))
495 					goto fail;
496 			goto done;
497 		case IPPROTO_UDP:
498 			if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
499 				if (!tcf_csum_ipv6_udp(skb, hl,
500 						       pl + sizeof(*ip6h), 0))
501 					goto fail;
502 			goto done;
503 		case IPPROTO_UDPLITE:
504 			if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
505 				if (!tcf_csum_ipv6_udp(skb, hl,
506 						       pl + sizeof(*ip6h), 1))
507 					goto fail;
508 			goto done;
509 		case IPPROTO_SCTP:
510 			if ((update_flags & TCA_CSUM_UPDATE_FLAG_SCTP) &&
511 			    !tcf_csum_sctp(skb, hl, pl + sizeof(*ip6h)))
512 				goto fail;
513 			goto done;
514 		default:
515 			goto ignore_skb;
516 		}
517 	} while (pskb_may_pull(skb, hl + 1 + ntkoff));
518 
519 done:
520 ignore_skb:
521 	return 1;
522 
523 fail:
524 	return 0;
525 }
526 
527 static int tcf_csum(struct sk_buff *skb, const struct tc_action *a,
528 		    struct tcf_result *res)
529 {
530 	struct tcf_csum *p = to_tcf_csum(a);
531 	int action;
532 	u32 update_flags;
533 
534 	spin_lock(&p->tcf_lock);
535 	tcf_lastuse_update(&p->tcf_tm);
536 	bstats_update(&p->tcf_bstats, skb);
537 	action = p->tcf_action;
538 	update_flags = p->update_flags;
539 	spin_unlock(&p->tcf_lock);
540 
541 	if (unlikely(action == TC_ACT_SHOT))
542 		goto drop;
543 
544 	switch (tc_skb_protocol(skb)) {
545 	case cpu_to_be16(ETH_P_IP):
546 		if (!tcf_csum_ipv4(skb, update_flags))
547 			goto drop;
548 		break;
549 	case cpu_to_be16(ETH_P_IPV6):
550 		if (!tcf_csum_ipv6(skb, update_flags))
551 			goto drop;
552 		break;
553 	}
554 
555 	return action;
556 
557 drop:
558 	spin_lock(&p->tcf_lock);
559 	p->tcf_qstats.drops++;
560 	spin_unlock(&p->tcf_lock);
561 	return TC_ACT_SHOT;
562 }
563 
564 static int tcf_csum_dump(struct sk_buff *skb, struct tc_action *a, int bind,
565 			 int ref)
566 {
567 	unsigned char *b = skb_tail_pointer(skb);
568 	struct tcf_csum *p = to_tcf_csum(a);
569 	struct tc_csum opt = {
570 		.update_flags = p->update_flags,
571 		.index   = p->tcf_index,
572 		.action  = p->tcf_action,
573 		.refcnt  = p->tcf_refcnt - ref,
574 		.bindcnt = p->tcf_bindcnt - bind,
575 	};
576 	struct tcf_t t;
577 
578 	if (nla_put(skb, TCA_CSUM_PARMS, sizeof(opt), &opt))
579 		goto nla_put_failure;
580 
581 	tcf_tm_dump(&t, &p->tcf_tm);
582 	if (nla_put_64bit(skb, TCA_CSUM_TM, sizeof(t), &t, TCA_CSUM_PAD))
583 		goto nla_put_failure;
584 
585 	return skb->len;
586 
587 nla_put_failure:
588 	nlmsg_trim(skb, b);
589 	return -1;
590 }
591 
592 static int tcf_csum_walker(struct net *net, struct sk_buff *skb,
593 			   struct netlink_callback *cb, int type,
594 			   const struct tc_action_ops *ops)
595 {
596 	struct tc_action_net *tn = net_generic(net, csum_net_id);
597 
598 	return tcf_generic_walker(tn, skb, cb, type, ops);
599 }
600 
601 static int tcf_csum_search(struct net *net, struct tc_action **a, u32 index)
602 {
603 	struct tc_action_net *tn = net_generic(net, csum_net_id);
604 
605 	return tcf_hash_search(tn, a, index);
606 }
607 
608 static struct tc_action_ops act_csum_ops = {
609 	.kind		= "csum",
610 	.type		= TCA_ACT_CSUM,
611 	.owner		= THIS_MODULE,
612 	.act		= tcf_csum,
613 	.dump		= tcf_csum_dump,
614 	.init		= tcf_csum_init,
615 	.walk		= tcf_csum_walker,
616 	.lookup		= tcf_csum_search,
617 	.size		= sizeof(struct tcf_csum),
618 };
619 
620 static __net_init int csum_init_net(struct net *net)
621 {
622 	struct tc_action_net *tn = net_generic(net, csum_net_id);
623 
624 	return tc_action_net_init(tn, &act_csum_ops, CSUM_TAB_MASK);
625 }
626 
627 static void __net_exit csum_exit_net(struct net *net)
628 {
629 	struct tc_action_net *tn = net_generic(net, csum_net_id);
630 
631 	tc_action_net_exit(tn);
632 }
633 
634 static struct pernet_operations csum_net_ops = {
635 	.init = csum_init_net,
636 	.exit = csum_exit_net,
637 	.id   = &csum_net_id,
638 	.size = sizeof(struct tc_action_net),
639 };
640 
641 MODULE_DESCRIPTION("Checksum updating actions");
642 MODULE_LICENSE("GPL");
643 
644 static int __init csum_init_module(void)
645 {
646 	return tcf_register_action(&act_csum_ops, &csum_net_ops);
647 }
648 
649 static void __exit csum_cleanup_module(void)
650 {
651 	tcf_unregister_action(&act_csum_ops, &csum_net_ops);
652 }
653 
654 module_init(csum_init_module);
655 module_exit(csum_cleanup_module);
656