xref: /openbmc/linux/net/sched/act_csum.c (revision 9cfc5c90)
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 
34 #include <net/act_api.h>
35 
36 #include <linux/tc_act/tc_csum.h>
37 #include <net/tc_act/tc_csum.h>
38 
39 #define CSUM_TAB_MASK 15
40 
41 static const struct nla_policy csum_policy[TCA_CSUM_MAX + 1] = {
42 	[TCA_CSUM_PARMS] = { .len = sizeof(struct tc_csum), },
43 };
44 
45 static int tcf_csum_init(struct net *n, struct nlattr *nla, struct nlattr *est,
46 			 struct tc_action *a, int ovr, int bind)
47 {
48 	struct nlattr *tb[TCA_CSUM_MAX + 1];
49 	struct tc_csum *parm;
50 	struct tcf_csum *p;
51 	int ret = 0, err;
52 
53 	if (nla == NULL)
54 		return -EINVAL;
55 
56 	err = nla_parse_nested(tb, TCA_CSUM_MAX, nla, csum_policy);
57 	if (err < 0)
58 		return err;
59 
60 	if (tb[TCA_CSUM_PARMS] == NULL)
61 		return -EINVAL;
62 	parm = nla_data(tb[TCA_CSUM_PARMS]);
63 
64 	if (!tcf_hash_check(parm->index, a, bind)) {
65 		ret = tcf_hash_create(parm->index, est, a, sizeof(*p),
66 				      bind, false);
67 		if (ret)
68 			return ret;
69 		ret = ACT_P_CREATED;
70 	} else {
71 		if (bind)/* dont override defaults */
72 			return 0;
73 		tcf_hash_release(a, bind);
74 		if (!ovr)
75 			return -EEXIST;
76 	}
77 
78 	p = to_tcf_csum(a);
79 	spin_lock_bh(&p->tcf_lock);
80 	p->tcf_action = parm->action;
81 	p->update_flags = parm->update_flags;
82 	spin_unlock_bh(&p->tcf_lock);
83 
84 	if (ret == ACT_P_CREATED)
85 		tcf_hash_insert(a);
86 
87 	return ret;
88 }
89 
90 /**
91  * tcf_csum_skb_nextlayer - Get next layer pointer
92  * @skb: sk_buff to use
93  * @ihl: previous summed headers length
94  * @ipl: complete packet length
95  * @jhl: next header length
96  *
97  * Check the expected next layer availability in the specified sk_buff.
98  * Return the next layer pointer if pass, NULL otherwise.
99  */
100 static void *tcf_csum_skb_nextlayer(struct sk_buff *skb,
101 				    unsigned int ihl, unsigned int ipl,
102 				    unsigned int jhl)
103 {
104 	int ntkoff = skb_network_offset(skb);
105 	int hl = ihl + jhl;
106 
107 	if (!pskb_may_pull(skb, ipl + ntkoff) || (ipl < hl) ||
108 	    (skb_cloned(skb) &&
109 	     !skb_clone_writable(skb, hl + ntkoff) &&
110 	     pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
111 		return NULL;
112 	else
113 		return (void *)(skb_network_header(skb) + ihl);
114 }
115 
116 static int tcf_csum_ipv4_icmp(struct sk_buff *skb,
117 			      unsigned int ihl, unsigned int ipl)
118 {
119 	struct icmphdr *icmph;
120 
121 	icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
122 	if (icmph == NULL)
123 		return 0;
124 
125 	icmph->checksum = 0;
126 	skb->csum = csum_partial(icmph, ipl - ihl, 0);
127 	icmph->checksum = csum_fold(skb->csum);
128 
129 	skb->ip_summed = CHECKSUM_NONE;
130 
131 	return 1;
132 }
133 
134 static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
135 			      unsigned int ihl, unsigned int ipl)
136 {
137 	struct igmphdr *igmph;
138 
139 	igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
140 	if (igmph == NULL)
141 		return 0;
142 
143 	igmph->csum = 0;
144 	skb->csum = csum_partial(igmph, ipl - ihl, 0);
145 	igmph->csum = csum_fold(skb->csum);
146 
147 	skb->ip_summed = CHECKSUM_NONE;
148 
149 	return 1;
150 }
151 
152 static int tcf_csum_ipv6_icmp(struct sk_buff *skb,
153 			      unsigned int ihl, unsigned int ipl)
154 {
155 	struct icmp6hdr *icmp6h;
156 	const struct ipv6hdr *ip6h;
157 
158 	icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
159 	if (icmp6h == NULL)
160 		return 0;
161 
162 	ip6h = ipv6_hdr(skb);
163 	icmp6h->icmp6_cksum = 0;
164 	skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
165 	icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
166 					      ipl - ihl, IPPROTO_ICMPV6,
167 					      skb->csum);
168 
169 	skb->ip_summed = CHECKSUM_NONE;
170 
171 	return 1;
172 }
173 
174 static int tcf_csum_ipv4_tcp(struct sk_buff *skb,
175 			     unsigned int ihl, unsigned int ipl)
176 {
177 	struct tcphdr *tcph;
178 	const struct iphdr *iph;
179 
180 	tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
181 	if (tcph == NULL)
182 		return 0;
183 
184 	iph = ip_hdr(skb);
185 	tcph->check = 0;
186 	skb->csum = csum_partial(tcph, ipl - ihl, 0);
187 	tcph->check = tcp_v4_check(ipl - ihl,
188 				   iph->saddr, iph->daddr, skb->csum);
189 
190 	skb->ip_summed = CHECKSUM_NONE;
191 
192 	return 1;
193 }
194 
195 static int tcf_csum_ipv6_tcp(struct sk_buff *skb,
196 			     unsigned int ihl, unsigned int ipl)
197 {
198 	struct tcphdr *tcph;
199 	const struct ipv6hdr *ip6h;
200 
201 	tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
202 	if (tcph == NULL)
203 		return 0;
204 
205 	ip6h = ipv6_hdr(skb);
206 	tcph->check = 0;
207 	skb->csum = csum_partial(tcph, ipl - ihl, 0);
208 	tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
209 				      ipl - ihl, IPPROTO_TCP,
210 				      skb->csum);
211 
212 	skb->ip_summed = CHECKSUM_NONE;
213 
214 	return 1;
215 }
216 
217 static int tcf_csum_ipv4_udp(struct sk_buff *skb,
218 			     unsigned int ihl, unsigned int ipl, int udplite)
219 {
220 	struct udphdr *udph;
221 	const struct iphdr *iph;
222 	u16 ul;
223 
224 	/*
225 	 * Support both UDP and UDPLITE checksum algorithms, Don't use
226 	 * udph->len to get the real length without any protocol check,
227 	 * UDPLITE uses udph->len for another thing,
228 	 * Use iph->tot_len, or just ipl.
229 	 */
230 
231 	udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
232 	if (udph == NULL)
233 		return 0;
234 
235 	iph = ip_hdr(skb);
236 	ul = ntohs(udph->len);
237 
238 	if (udplite || udph->check) {
239 
240 		udph->check = 0;
241 
242 		if (udplite) {
243 			if (ul == 0)
244 				skb->csum = csum_partial(udph, ipl - ihl, 0);
245 			else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
246 				skb->csum = csum_partial(udph, ul, 0);
247 			else
248 				goto ignore_obscure_skb;
249 		} else {
250 			if (ul != ipl - ihl)
251 				goto ignore_obscure_skb;
252 
253 			skb->csum = csum_partial(udph, ul, 0);
254 		}
255 
256 		udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
257 						ul, iph->protocol,
258 						skb->csum);
259 
260 		if (!udph->check)
261 			udph->check = CSUM_MANGLED_0;
262 	}
263 
264 	skb->ip_summed = CHECKSUM_NONE;
265 
266 ignore_obscure_skb:
267 	return 1;
268 }
269 
270 static int tcf_csum_ipv6_udp(struct sk_buff *skb,
271 			     unsigned int ihl, unsigned int ipl, int udplite)
272 {
273 	struct udphdr *udph;
274 	const struct ipv6hdr *ip6h;
275 	u16 ul;
276 
277 	/*
278 	 * Support both UDP and UDPLITE checksum algorithms, Don't use
279 	 * udph->len to get the real length without any protocol check,
280 	 * UDPLITE uses udph->len for another thing,
281 	 * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
282 	 */
283 
284 	udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
285 	if (udph == NULL)
286 		return 0;
287 
288 	ip6h = ipv6_hdr(skb);
289 	ul = ntohs(udph->len);
290 
291 	udph->check = 0;
292 
293 	if (udplite) {
294 		if (ul == 0)
295 			skb->csum = csum_partial(udph, ipl - ihl, 0);
296 
297 		else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
298 			skb->csum = csum_partial(udph, ul, 0);
299 
300 		else
301 			goto ignore_obscure_skb;
302 	} else {
303 		if (ul != ipl - ihl)
304 			goto ignore_obscure_skb;
305 
306 		skb->csum = csum_partial(udph, ul, 0);
307 	}
308 
309 	udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
310 				      udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
311 				      skb->csum);
312 
313 	if (!udph->check)
314 		udph->check = CSUM_MANGLED_0;
315 
316 	skb->ip_summed = CHECKSUM_NONE;
317 
318 ignore_obscure_skb:
319 	return 1;
320 }
321 
322 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
323 {
324 	const struct iphdr *iph;
325 	int ntkoff;
326 
327 	ntkoff = skb_network_offset(skb);
328 
329 	if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
330 		goto fail;
331 
332 	iph = ip_hdr(skb);
333 
334 	switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
335 	case IPPROTO_ICMP:
336 		if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
337 			if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
338 						ntohs(iph->tot_len)))
339 				goto fail;
340 		break;
341 	case IPPROTO_IGMP:
342 		if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
343 			if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
344 						ntohs(iph->tot_len)))
345 				goto fail;
346 		break;
347 	case IPPROTO_TCP:
348 		if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
349 			if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
350 					       ntohs(iph->tot_len)))
351 				goto fail;
352 		break;
353 	case IPPROTO_UDP:
354 		if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
355 			if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
356 					       ntohs(iph->tot_len), 0))
357 				goto fail;
358 		break;
359 	case IPPROTO_UDPLITE:
360 		if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
361 			if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
362 					       ntohs(iph->tot_len), 1))
363 				goto fail;
364 		break;
365 	}
366 
367 	if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
368 		if (skb_cloned(skb) &&
369 		    !skb_clone_writable(skb, sizeof(*iph) + ntkoff) &&
370 		    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
371 			goto fail;
372 
373 		ip_send_check(ip_hdr(skb));
374 	}
375 
376 	return 1;
377 
378 fail:
379 	return 0;
380 }
381 
382 static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh,
383 				 unsigned int ixhl, unsigned int *pl)
384 {
385 	int off, len, optlen;
386 	unsigned char *xh = (void *)ip6xh;
387 
388 	off = sizeof(*ip6xh);
389 	len = ixhl - off;
390 
391 	while (len > 1) {
392 		switch (xh[off]) {
393 		case IPV6_TLV_PAD1:
394 			optlen = 1;
395 			break;
396 		case IPV6_TLV_JUMBO:
397 			optlen = xh[off + 1] + 2;
398 			if (optlen != 6 || len < 6 || (off & 3) != 2)
399 				/* wrong jumbo option length/alignment */
400 				return 0;
401 			*pl = ntohl(*(__be32 *)(xh + off + 2));
402 			goto done;
403 		default:
404 			optlen = xh[off + 1] + 2;
405 			if (optlen > len)
406 				/* ignore obscure options */
407 				goto done;
408 			break;
409 		}
410 		off += optlen;
411 		len -= optlen;
412 	}
413 
414 done:
415 	return 1;
416 }
417 
418 static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
419 {
420 	struct ipv6hdr *ip6h;
421 	struct ipv6_opt_hdr *ip6xh;
422 	unsigned int hl, ixhl;
423 	unsigned int pl;
424 	int ntkoff;
425 	u8 nexthdr;
426 
427 	ntkoff = skb_network_offset(skb);
428 
429 	hl = sizeof(*ip6h);
430 
431 	if (!pskb_may_pull(skb, hl + ntkoff))
432 		goto fail;
433 
434 	ip6h = ipv6_hdr(skb);
435 
436 	pl = ntohs(ip6h->payload_len);
437 	nexthdr = ip6h->nexthdr;
438 
439 	do {
440 		switch (nexthdr) {
441 		case NEXTHDR_FRAGMENT:
442 			goto ignore_skb;
443 		case NEXTHDR_ROUTING:
444 		case NEXTHDR_HOP:
445 		case NEXTHDR_DEST:
446 			if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
447 				goto fail;
448 			ip6xh = (void *)(skb_network_header(skb) + hl);
449 			ixhl = ipv6_optlen(ip6xh);
450 			if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
451 				goto fail;
452 			ip6xh = (void *)(skb_network_header(skb) + hl);
453 			if ((nexthdr == NEXTHDR_HOP) &&
454 			    !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
455 				goto fail;
456 			nexthdr = ip6xh->nexthdr;
457 			hl += ixhl;
458 			break;
459 		case IPPROTO_ICMPV6:
460 			if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
461 				if (!tcf_csum_ipv6_icmp(skb,
462 							hl, pl + sizeof(*ip6h)))
463 					goto fail;
464 			goto done;
465 		case IPPROTO_TCP:
466 			if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
467 				if (!tcf_csum_ipv6_tcp(skb,
468 						       hl, pl + sizeof(*ip6h)))
469 					goto fail;
470 			goto done;
471 		case IPPROTO_UDP:
472 			if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
473 				if (!tcf_csum_ipv6_udp(skb, hl,
474 						       pl + sizeof(*ip6h), 0))
475 					goto fail;
476 			goto done;
477 		case IPPROTO_UDPLITE:
478 			if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
479 				if (!tcf_csum_ipv6_udp(skb, hl,
480 						       pl + sizeof(*ip6h), 1))
481 					goto fail;
482 			goto done;
483 		default:
484 			goto ignore_skb;
485 		}
486 	} while (pskb_may_pull(skb, hl + 1 + ntkoff));
487 
488 done:
489 ignore_skb:
490 	return 1;
491 
492 fail:
493 	return 0;
494 }
495 
496 static int tcf_csum(struct sk_buff *skb,
497 		    const struct tc_action *a, struct tcf_result *res)
498 {
499 	struct tcf_csum *p = a->priv;
500 	int action;
501 	u32 update_flags;
502 
503 	spin_lock(&p->tcf_lock);
504 	p->tcf_tm.lastuse = jiffies;
505 	bstats_update(&p->tcf_bstats, skb);
506 	action = p->tcf_action;
507 	update_flags = p->update_flags;
508 	spin_unlock(&p->tcf_lock);
509 
510 	if (unlikely(action == TC_ACT_SHOT))
511 		goto drop;
512 
513 	switch (tc_skb_protocol(skb)) {
514 	case cpu_to_be16(ETH_P_IP):
515 		if (!tcf_csum_ipv4(skb, update_flags))
516 			goto drop;
517 		break;
518 	case cpu_to_be16(ETH_P_IPV6):
519 		if (!tcf_csum_ipv6(skb, update_flags))
520 			goto drop;
521 		break;
522 	}
523 
524 	return action;
525 
526 drop:
527 	spin_lock(&p->tcf_lock);
528 	p->tcf_qstats.drops++;
529 	spin_unlock(&p->tcf_lock);
530 	return TC_ACT_SHOT;
531 }
532 
533 static int tcf_csum_dump(struct sk_buff *skb,
534 			 struct tc_action *a, int bind, int ref)
535 {
536 	unsigned char *b = skb_tail_pointer(skb);
537 	struct tcf_csum *p = a->priv;
538 	struct tc_csum opt = {
539 		.update_flags = p->update_flags,
540 		.index   = p->tcf_index,
541 		.action  = p->tcf_action,
542 		.refcnt  = p->tcf_refcnt - ref,
543 		.bindcnt = p->tcf_bindcnt - bind,
544 	};
545 	struct tcf_t t;
546 
547 	if (nla_put(skb, TCA_CSUM_PARMS, sizeof(opt), &opt))
548 		goto nla_put_failure;
549 	t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
550 	t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
551 	t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
552 	if (nla_put(skb, TCA_CSUM_TM, sizeof(t), &t))
553 		goto nla_put_failure;
554 
555 	return skb->len;
556 
557 nla_put_failure:
558 	nlmsg_trim(skb, b);
559 	return -1;
560 }
561 
562 static struct tc_action_ops act_csum_ops = {
563 	.kind		= "csum",
564 	.type		= TCA_ACT_CSUM,
565 	.owner		= THIS_MODULE,
566 	.act		= tcf_csum,
567 	.dump		= tcf_csum_dump,
568 	.init		= tcf_csum_init,
569 };
570 
571 MODULE_DESCRIPTION("Checksum updating actions");
572 MODULE_LICENSE("GPL");
573 
574 static int __init csum_init_module(void)
575 {
576 	return tcf_register_action(&act_csum_ops, CSUM_TAB_MASK);
577 }
578 
579 static void __exit csum_cleanup_module(void)
580 {
581 	tcf_unregister_action(&act_csum_ops);
582 }
583 
584 module_init(csum_init_module);
585 module_exit(csum_cleanup_module);
586