xref: /openbmc/linux/net/sched/act_csum.c (revision 0bea2a65)
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 static const struct nla_policy csum_policy[TCA_CSUM_MAX + 1] = {
41 	[TCA_CSUM_PARMS] = { .len = sizeof(struct tc_csum), },
42 };
43 
44 static unsigned int csum_net_id;
45 static struct tc_action_ops act_csum_ops;
46 
47 static int tcf_csum_init(struct net *net, struct nlattr *nla,
48 			 struct nlattr *est, struct tc_action **a, int ovr,
49 			 int bind)
50 {
51 	struct tc_action_net *tn = net_generic(net, csum_net_id);
52 	struct nlattr *tb[TCA_CSUM_MAX + 1];
53 	struct tc_csum *parm;
54 	struct tcf_csum *p;
55 	int ret = 0, err;
56 
57 	if (nla == NULL)
58 		return -EINVAL;
59 
60 	err = nla_parse_nested(tb, TCA_CSUM_MAX, nla, csum_policy, NULL);
61 	if (err < 0)
62 		return err;
63 
64 	if (tb[TCA_CSUM_PARMS] == NULL)
65 		return -EINVAL;
66 	parm = nla_data(tb[TCA_CSUM_PARMS]);
67 
68 	if (!tcf_idr_check(tn, parm->index, a, bind)) {
69 		ret = tcf_idr_create(tn, parm->index, est, a,
70 				     &act_csum_ops, bind, false);
71 		if (ret)
72 			return ret;
73 		ret = ACT_P_CREATED;
74 	} else {
75 		if (bind)/* dont override defaults */
76 			return 0;
77 		tcf_idr_release(*a, bind);
78 		if (!ovr)
79 			return -EEXIST;
80 	}
81 
82 	p = to_tcf_csum(*a);
83 	spin_lock_bh(&p->tcf_lock);
84 	p->tcf_action = parm->action;
85 	p->update_flags = parm->update_flags;
86 	spin_unlock_bh(&p->tcf_lock);
87 
88 	if (ret == ACT_P_CREATED)
89 		tcf_idr_insert(tn, *a);
90 
91 	return ret;
92 }
93 
94 /**
95  * tcf_csum_skb_nextlayer - Get next layer pointer
96  * @skb: sk_buff to use
97  * @ihl: previous summed headers length
98  * @ipl: complete packet length
99  * @jhl: next header length
100  *
101  * Check the expected next layer availability in the specified sk_buff.
102  * Return the next layer pointer if pass, NULL otherwise.
103  */
104 static void *tcf_csum_skb_nextlayer(struct sk_buff *skb,
105 				    unsigned int ihl, unsigned int ipl,
106 				    unsigned int jhl)
107 {
108 	int ntkoff = skb_network_offset(skb);
109 	int hl = ihl + jhl;
110 
111 	if (!pskb_may_pull(skb, ipl + ntkoff) || (ipl < hl) ||
112 	    skb_try_make_writable(skb, hl + ntkoff))
113 		return NULL;
114 	else
115 		return (void *)(skb_network_header(skb) + ihl);
116 }
117 
118 static int tcf_csum_ipv4_icmp(struct sk_buff *skb, unsigned int ihl,
119 			      unsigned int ipl)
120 {
121 	struct icmphdr *icmph;
122 
123 	icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
124 	if (icmph == NULL)
125 		return 0;
126 
127 	icmph->checksum = 0;
128 	skb->csum = csum_partial(icmph, ipl - ihl, 0);
129 	icmph->checksum = csum_fold(skb->csum);
130 
131 	skb->ip_summed = CHECKSUM_NONE;
132 
133 	return 1;
134 }
135 
136 static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
137 			      unsigned int ihl, unsigned int ipl)
138 {
139 	struct igmphdr *igmph;
140 
141 	igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
142 	if (igmph == NULL)
143 		return 0;
144 
145 	igmph->csum = 0;
146 	skb->csum = csum_partial(igmph, ipl - ihl, 0);
147 	igmph->csum = csum_fold(skb->csum);
148 
149 	skb->ip_summed = CHECKSUM_NONE;
150 
151 	return 1;
152 }
153 
154 static int tcf_csum_ipv6_icmp(struct sk_buff *skb, unsigned int ihl,
155 			      unsigned int ipl)
156 {
157 	struct icmp6hdr *icmp6h;
158 	const struct ipv6hdr *ip6h;
159 
160 	icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
161 	if (icmp6h == NULL)
162 		return 0;
163 
164 	ip6h = ipv6_hdr(skb);
165 	icmp6h->icmp6_cksum = 0;
166 	skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
167 	icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
168 					      ipl - ihl, IPPROTO_ICMPV6,
169 					      skb->csum);
170 
171 	skb->ip_summed = CHECKSUM_NONE;
172 
173 	return 1;
174 }
175 
176 static int tcf_csum_ipv4_tcp(struct sk_buff *skb, unsigned int ihl,
177 			     unsigned int ipl)
178 {
179 	struct tcphdr *tcph;
180 	const struct iphdr *iph;
181 
182 	if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
183 		return 1;
184 
185 	tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
186 	if (tcph == NULL)
187 		return 0;
188 
189 	iph = ip_hdr(skb);
190 	tcph->check = 0;
191 	skb->csum = csum_partial(tcph, ipl - ihl, 0);
192 	tcph->check = tcp_v4_check(ipl - ihl,
193 				   iph->saddr, iph->daddr, skb->csum);
194 
195 	skb->ip_summed = CHECKSUM_NONE;
196 
197 	return 1;
198 }
199 
200 static int tcf_csum_ipv6_tcp(struct sk_buff *skb, unsigned int ihl,
201 			     unsigned int ipl)
202 {
203 	struct tcphdr *tcph;
204 	const struct ipv6hdr *ip6h;
205 
206 	if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
207 		return 1;
208 
209 	tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
210 	if (tcph == NULL)
211 		return 0;
212 
213 	ip6h = ipv6_hdr(skb);
214 	tcph->check = 0;
215 	skb->csum = csum_partial(tcph, ipl - ihl, 0);
216 	tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
217 				      ipl - ihl, IPPROTO_TCP,
218 				      skb->csum);
219 
220 	skb->ip_summed = CHECKSUM_NONE;
221 
222 	return 1;
223 }
224 
225 static int tcf_csum_ipv4_udp(struct sk_buff *skb, unsigned int ihl,
226 			     unsigned int ipl, int udplite)
227 {
228 	struct udphdr *udph;
229 	const struct iphdr *iph;
230 	u16 ul;
231 
232 	if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
233 		return 1;
234 
235 	/*
236 	 * Support both UDP and UDPLITE checksum algorithms, Don't use
237 	 * udph->len to get the real length without any protocol check,
238 	 * UDPLITE uses udph->len for another thing,
239 	 * Use iph->tot_len, or just ipl.
240 	 */
241 
242 	udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
243 	if (udph == NULL)
244 		return 0;
245 
246 	iph = ip_hdr(skb);
247 	ul = ntohs(udph->len);
248 
249 	if (udplite || udph->check) {
250 
251 		udph->check = 0;
252 
253 		if (udplite) {
254 			if (ul == 0)
255 				skb->csum = csum_partial(udph, ipl - ihl, 0);
256 			else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
257 				skb->csum = csum_partial(udph, ul, 0);
258 			else
259 				goto ignore_obscure_skb;
260 		} else {
261 			if (ul != ipl - ihl)
262 				goto ignore_obscure_skb;
263 
264 			skb->csum = csum_partial(udph, ul, 0);
265 		}
266 
267 		udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
268 						ul, iph->protocol,
269 						skb->csum);
270 
271 		if (!udph->check)
272 			udph->check = CSUM_MANGLED_0;
273 	}
274 
275 	skb->ip_summed = CHECKSUM_NONE;
276 
277 ignore_obscure_skb:
278 	return 1;
279 }
280 
281 static int tcf_csum_ipv6_udp(struct sk_buff *skb, unsigned int ihl,
282 			     unsigned int ipl, int udplite)
283 {
284 	struct udphdr *udph;
285 	const struct ipv6hdr *ip6h;
286 	u16 ul;
287 
288 	if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
289 		return 1;
290 
291 	/*
292 	 * Support both UDP and UDPLITE checksum algorithms, Don't use
293 	 * udph->len to get the real length without any protocol check,
294 	 * UDPLITE uses udph->len for another thing,
295 	 * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
296 	 */
297 
298 	udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
299 	if (udph == NULL)
300 		return 0;
301 
302 	ip6h = ipv6_hdr(skb);
303 	ul = ntohs(udph->len);
304 
305 	udph->check = 0;
306 
307 	if (udplite) {
308 		if (ul == 0)
309 			skb->csum = csum_partial(udph, ipl - ihl, 0);
310 
311 		else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
312 			skb->csum = csum_partial(udph, ul, 0);
313 
314 		else
315 			goto ignore_obscure_skb;
316 	} else {
317 		if (ul != ipl - ihl)
318 			goto ignore_obscure_skb;
319 
320 		skb->csum = csum_partial(udph, ul, 0);
321 	}
322 
323 	udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
324 				      udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
325 				      skb->csum);
326 
327 	if (!udph->check)
328 		udph->check = CSUM_MANGLED_0;
329 
330 	skb->ip_summed = CHECKSUM_NONE;
331 
332 ignore_obscure_skb:
333 	return 1;
334 }
335 
336 static int tcf_csum_sctp(struct sk_buff *skb, unsigned int ihl,
337 			 unsigned int ipl)
338 {
339 	struct sctphdr *sctph;
340 
341 	if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_SCTP)
342 		return 1;
343 
344 	sctph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*sctph));
345 	if (!sctph)
346 		return 0;
347 
348 	sctph->checksum = sctp_compute_cksum(skb,
349 					     skb_network_offset(skb) + ihl);
350 	skb->ip_summed = CHECKSUM_NONE;
351 	skb->csum_not_inet = 0;
352 
353 	return 1;
354 }
355 
356 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
357 {
358 	const struct iphdr *iph;
359 	int ntkoff;
360 
361 	ntkoff = skb_network_offset(skb);
362 
363 	if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
364 		goto fail;
365 
366 	iph = ip_hdr(skb);
367 
368 	switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
369 	case IPPROTO_ICMP:
370 		if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
371 			if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
372 						ntohs(iph->tot_len)))
373 				goto fail;
374 		break;
375 	case IPPROTO_IGMP:
376 		if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
377 			if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
378 						ntohs(iph->tot_len)))
379 				goto fail;
380 		break;
381 	case IPPROTO_TCP:
382 		if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
383 			if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
384 					       ntohs(iph->tot_len)))
385 				goto fail;
386 		break;
387 	case IPPROTO_UDP:
388 		if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
389 			if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
390 					       ntohs(iph->tot_len), 0))
391 				goto fail;
392 		break;
393 	case IPPROTO_UDPLITE:
394 		if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
395 			if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
396 					       ntohs(iph->tot_len), 1))
397 				goto fail;
398 		break;
399 	case IPPROTO_SCTP:
400 		if ((update_flags & TCA_CSUM_UPDATE_FLAG_SCTP) &&
401 		    !tcf_csum_sctp(skb, iph->ihl * 4, ntohs(iph->tot_len)))
402 			goto fail;
403 		break;
404 	}
405 
406 	if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
407 		if (skb_try_make_writable(skb, sizeof(*iph) + ntkoff))
408 			goto fail;
409 
410 		ip_send_check(ip_hdr(skb));
411 	}
412 
413 	return 1;
414 
415 fail:
416 	return 0;
417 }
418 
419 static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh, unsigned int ixhl,
420 				 unsigned int *pl)
421 {
422 	int off, len, optlen;
423 	unsigned char *xh = (void *)ip6xh;
424 
425 	off = sizeof(*ip6xh);
426 	len = ixhl - off;
427 
428 	while (len > 1) {
429 		switch (xh[off]) {
430 		case IPV6_TLV_PAD1:
431 			optlen = 1;
432 			break;
433 		case IPV6_TLV_JUMBO:
434 			optlen = xh[off + 1] + 2;
435 			if (optlen != 6 || len < 6 || (off & 3) != 2)
436 				/* wrong jumbo option length/alignment */
437 				return 0;
438 			*pl = ntohl(*(__be32 *)(xh + off + 2));
439 			goto done;
440 		default:
441 			optlen = xh[off + 1] + 2;
442 			if (optlen > len)
443 				/* ignore obscure options */
444 				goto done;
445 			break;
446 		}
447 		off += optlen;
448 		len -= optlen;
449 	}
450 
451 done:
452 	return 1;
453 }
454 
455 static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
456 {
457 	struct ipv6hdr *ip6h;
458 	struct ipv6_opt_hdr *ip6xh;
459 	unsigned int hl, ixhl;
460 	unsigned int pl;
461 	int ntkoff;
462 	u8 nexthdr;
463 
464 	ntkoff = skb_network_offset(skb);
465 
466 	hl = sizeof(*ip6h);
467 
468 	if (!pskb_may_pull(skb, hl + ntkoff))
469 		goto fail;
470 
471 	ip6h = ipv6_hdr(skb);
472 
473 	pl = ntohs(ip6h->payload_len);
474 	nexthdr = ip6h->nexthdr;
475 
476 	do {
477 		switch (nexthdr) {
478 		case NEXTHDR_FRAGMENT:
479 			goto ignore_skb;
480 		case NEXTHDR_ROUTING:
481 		case NEXTHDR_HOP:
482 		case NEXTHDR_DEST:
483 			if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
484 				goto fail;
485 			ip6xh = (void *)(skb_network_header(skb) + hl);
486 			ixhl = ipv6_optlen(ip6xh);
487 			if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
488 				goto fail;
489 			ip6xh = (void *)(skb_network_header(skb) + hl);
490 			if ((nexthdr == NEXTHDR_HOP) &&
491 			    !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
492 				goto fail;
493 			nexthdr = ip6xh->nexthdr;
494 			hl += ixhl;
495 			break;
496 		case IPPROTO_ICMPV6:
497 			if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
498 				if (!tcf_csum_ipv6_icmp(skb,
499 							hl, pl + sizeof(*ip6h)))
500 					goto fail;
501 			goto done;
502 		case IPPROTO_TCP:
503 			if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
504 				if (!tcf_csum_ipv6_tcp(skb,
505 						       hl, pl + sizeof(*ip6h)))
506 					goto fail;
507 			goto done;
508 		case IPPROTO_UDP:
509 			if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
510 				if (!tcf_csum_ipv6_udp(skb, hl,
511 						       pl + sizeof(*ip6h), 0))
512 					goto fail;
513 			goto done;
514 		case IPPROTO_UDPLITE:
515 			if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
516 				if (!tcf_csum_ipv6_udp(skb, hl,
517 						       pl + sizeof(*ip6h), 1))
518 					goto fail;
519 			goto done;
520 		case IPPROTO_SCTP:
521 			if ((update_flags & TCA_CSUM_UPDATE_FLAG_SCTP) &&
522 			    !tcf_csum_sctp(skb, hl, pl + sizeof(*ip6h)))
523 				goto fail;
524 			goto done;
525 		default:
526 			goto ignore_skb;
527 		}
528 	} while (pskb_may_pull(skb, hl + 1 + ntkoff));
529 
530 done:
531 ignore_skb:
532 	return 1;
533 
534 fail:
535 	return 0;
536 }
537 
538 static int tcf_csum(struct sk_buff *skb, const struct tc_action *a,
539 		    struct tcf_result *res)
540 {
541 	struct tcf_csum *p = to_tcf_csum(a);
542 	int action;
543 	u32 update_flags;
544 
545 	spin_lock(&p->tcf_lock);
546 	tcf_lastuse_update(&p->tcf_tm);
547 	bstats_update(&p->tcf_bstats, skb);
548 	action = p->tcf_action;
549 	update_flags = p->update_flags;
550 	spin_unlock(&p->tcf_lock);
551 
552 	if (unlikely(action == TC_ACT_SHOT))
553 		goto drop;
554 
555 	switch (tc_skb_protocol(skb)) {
556 	case cpu_to_be16(ETH_P_IP):
557 		if (!tcf_csum_ipv4(skb, update_flags))
558 			goto drop;
559 		break;
560 	case cpu_to_be16(ETH_P_IPV6):
561 		if (!tcf_csum_ipv6(skb, update_flags))
562 			goto drop;
563 		break;
564 	}
565 
566 	return action;
567 
568 drop:
569 	spin_lock(&p->tcf_lock);
570 	p->tcf_qstats.drops++;
571 	spin_unlock(&p->tcf_lock);
572 	return TC_ACT_SHOT;
573 }
574 
575 static int tcf_csum_dump(struct sk_buff *skb, struct tc_action *a, int bind,
576 			 int ref)
577 {
578 	unsigned char *b = skb_tail_pointer(skb);
579 	struct tcf_csum *p = to_tcf_csum(a);
580 	struct tc_csum opt = {
581 		.update_flags = p->update_flags,
582 		.index   = p->tcf_index,
583 		.action  = p->tcf_action,
584 		.refcnt  = p->tcf_refcnt - ref,
585 		.bindcnt = p->tcf_bindcnt - bind,
586 	};
587 	struct tcf_t t;
588 
589 	if (nla_put(skb, TCA_CSUM_PARMS, sizeof(opt), &opt))
590 		goto nla_put_failure;
591 
592 	tcf_tm_dump(&t, &p->tcf_tm);
593 	if (nla_put_64bit(skb, TCA_CSUM_TM, sizeof(t), &t, TCA_CSUM_PAD))
594 		goto nla_put_failure;
595 
596 	return skb->len;
597 
598 nla_put_failure:
599 	nlmsg_trim(skb, b);
600 	return -1;
601 }
602 
603 static int tcf_csum_walker(struct net *net, struct sk_buff *skb,
604 			   struct netlink_callback *cb, int type,
605 			   const struct tc_action_ops *ops)
606 {
607 	struct tc_action_net *tn = net_generic(net, csum_net_id);
608 
609 	return tcf_generic_walker(tn, skb, cb, type, ops);
610 }
611 
612 static int tcf_csum_search(struct net *net, struct tc_action **a, u32 index)
613 {
614 	struct tc_action_net *tn = net_generic(net, csum_net_id);
615 
616 	return tcf_idr_search(tn, a, index);
617 }
618 
619 static struct tc_action_ops act_csum_ops = {
620 	.kind		= "csum",
621 	.type		= TCA_ACT_CSUM,
622 	.owner		= THIS_MODULE,
623 	.act		= tcf_csum,
624 	.dump		= tcf_csum_dump,
625 	.init		= tcf_csum_init,
626 	.walk		= tcf_csum_walker,
627 	.lookup		= tcf_csum_search,
628 	.size		= sizeof(struct tcf_csum),
629 };
630 
631 static __net_init int csum_init_net(struct net *net)
632 {
633 	struct tc_action_net *tn = net_generic(net, csum_net_id);
634 
635 	return tc_action_net_init(tn, &act_csum_ops);
636 }
637 
638 static void __net_exit csum_exit_net(struct net *net)
639 {
640 	struct tc_action_net *tn = net_generic(net, csum_net_id);
641 
642 	tc_action_net_exit(tn);
643 }
644 
645 static struct pernet_operations csum_net_ops = {
646 	.init = csum_init_net,
647 	.exit = csum_exit_net,
648 	.id   = &csum_net_id,
649 	.size = sizeof(struct tc_action_net),
650 };
651 
652 MODULE_DESCRIPTION("Checksum updating actions");
653 MODULE_LICENSE("GPL");
654 
655 static int __init csum_init_module(void)
656 {
657 	return tcf_register_action(&act_csum_ops, &csum_net_ops);
658 }
659 
660 static void __exit csum_cleanup_module(void)
661 {
662 	tcf_unregister_action(&act_csum_ops, &csum_net_ops);
663 }
664 
665 module_init(csum_init_module);
666 module_exit(csum_cleanup_module);
667