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
40static struct tcf_common *tcf_csum_ht[CSUM_TAB_MASK + 1];
41static u32 csum_idx_gen;
42static DEFINE_RWLOCK(csum_lock);
43
44static struct tcf_hashinfo csum_hash_info = {
45	.htab	= tcf_csum_ht,
46	.hmask	= CSUM_TAB_MASK,
47	.lock	= &csum_lock,
48};
49
50static const struct nla_policy csum_policy[TCA_CSUM_MAX + 1] = {
51	[TCA_CSUM_PARMS] = { .len = sizeof(struct tc_csum), },
52};
53
54static int tcf_csum_init(struct net *n, struct nlattr *nla, struct nlattr *est,
55			 struct tc_action *a, int ovr, int bind)
56{
57	struct nlattr *tb[TCA_CSUM_MAX + 1];
58	struct tc_csum *parm;
59	struct tcf_common *pc;
60	struct tcf_csum *p;
61	int ret = 0, err;
62
63	if (nla == NULL)
64		return -EINVAL;
65
66	err = nla_parse_nested(tb, TCA_CSUM_MAX, nla, csum_policy);
67	if (err < 0)
68		return err;
69
70	if (tb[TCA_CSUM_PARMS] == NULL)
71		return -EINVAL;
72	parm = nla_data(tb[TCA_CSUM_PARMS]);
73
74	pc = tcf_hash_check(parm->index, a, bind, &csum_hash_info);
75	if (!pc) {
76		pc = tcf_hash_create(parm->index, est, a, sizeof(*p), bind,
77				     &csum_idx_gen, &csum_hash_info);
78		if (IS_ERR(pc))
79			return PTR_ERR(pc);
80		p = to_tcf_csum(pc);
81		ret = ACT_P_CREATED;
82	} else {
83		p = to_tcf_csum(pc);
84		if (!ovr) {
85			tcf_hash_release(pc, bind, &csum_hash_info);
86			return -EEXIST;
87		}
88	}
89
90	spin_lock_bh(&p->tcf_lock);
91	p->tcf_action = parm->action;
92	p->update_flags = parm->update_flags;
93	spin_unlock_bh(&p->tcf_lock);
94
95	if (ret == ACT_P_CREATED)
96		tcf_hash_insert(pc, &csum_hash_info);
97
98	return ret;
99}
100
101static int tcf_csum_cleanup(struct tc_action *a, int bind)
102{
103	struct tcf_csum *p = a->priv;
104	return tcf_hash_release(&p->common, bind, &csum_hash_info);
105}
106
107/**
108 * tcf_csum_skb_nextlayer - Get next layer pointer
109 * @skb: sk_buff to use
110 * @ihl: previous summed headers length
111 * @ipl: complete packet length
112 * @jhl: next header length
113 *
114 * Check the expected next layer availability in the specified sk_buff.
115 * Return the next layer pointer if pass, NULL otherwise.
116 */
117static void *tcf_csum_skb_nextlayer(struct sk_buff *skb,
118				    unsigned int ihl, unsigned int ipl,
119				    unsigned int jhl)
120{
121	int ntkoff = skb_network_offset(skb);
122	int hl = ihl + jhl;
123
124	if (!pskb_may_pull(skb, ipl + ntkoff) || (ipl < hl) ||
125	    (skb_cloned(skb) &&
126	     !skb_clone_writable(skb, hl + ntkoff) &&
127	     pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
128		return NULL;
129	else
130		return (void *)(skb_network_header(skb) + ihl);
131}
132
133static int tcf_csum_ipv4_icmp(struct sk_buff *skb,
134			      unsigned int ihl, unsigned int ipl)
135{
136	struct icmphdr *icmph;
137
138	icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
139	if (icmph == NULL)
140		return 0;
141
142	icmph->checksum = 0;
143	skb->csum = csum_partial(icmph, ipl - ihl, 0);
144	icmph->checksum = csum_fold(skb->csum);
145
146	skb->ip_summed = CHECKSUM_NONE;
147
148	return 1;
149}
150
151static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
152			      unsigned int ihl, unsigned int ipl)
153{
154	struct igmphdr *igmph;
155
156	igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
157	if (igmph == NULL)
158		return 0;
159
160	igmph->csum = 0;
161	skb->csum = csum_partial(igmph, ipl - ihl, 0);
162	igmph->csum = csum_fold(skb->csum);
163
164	skb->ip_summed = CHECKSUM_NONE;
165
166	return 1;
167}
168
169static int tcf_csum_ipv6_icmp(struct sk_buff *skb,
170			      unsigned int ihl, unsigned int ipl)
171{
172	struct icmp6hdr *icmp6h;
173	const struct ipv6hdr *ip6h;
174
175	icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
176	if (icmp6h == NULL)
177		return 0;
178
179	ip6h = ipv6_hdr(skb);
180	icmp6h->icmp6_cksum = 0;
181	skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
182	icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
183					      ipl - ihl, IPPROTO_ICMPV6,
184					      skb->csum);
185
186	skb->ip_summed = CHECKSUM_NONE;
187
188	return 1;
189}
190
191static int tcf_csum_ipv4_tcp(struct sk_buff *skb,
192			     unsigned int ihl, unsigned int ipl)
193{
194	struct tcphdr *tcph;
195	const struct iphdr *iph;
196
197	tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
198	if (tcph == NULL)
199		return 0;
200
201	iph = ip_hdr(skb);
202	tcph->check = 0;
203	skb->csum = csum_partial(tcph, ipl - ihl, 0);
204	tcph->check = tcp_v4_check(ipl - ihl,
205				   iph->saddr, iph->daddr, skb->csum);
206
207	skb->ip_summed = CHECKSUM_NONE;
208
209	return 1;
210}
211
212static int tcf_csum_ipv6_tcp(struct sk_buff *skb,
213			     unsigned int ihl, unsigned int ipl)
214{
215	struct tcphdr *tcph;
216	const struct ipv6hdr *ip6h;
217
218	tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
219	if (tcph == NULL)
220		return 0;
221
222	ip6h = ipv6_hdr(skb);
223	tcph->check = 0;
224	skb->csum = csum_partial(tcph, ipl - ihl, 0);
225	tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
226				      ipl - ihl, IPPROTO_TCP,
227				      skb->csum);
228
229	skb->ip_summed = CHECKSUM_NONE;
230
231	return 1;
232}
233
234static int tcf_csum_ipv4_udp(struct sk_buff *skb,
235			     unsigned int ihl, unsigned int ipl, int udplite)
236{
237	struct udphdr *udph;
238	const struct iphdr *iph;
239	u16 ul;
240
241	/*
242	 * Support both UDP and UDPLITE checksum algorithms, Don't use
243	 * udph->len to get the real length without any protocol check,
244	 * UDPLITE uses udph->len for another thing,
245	 * Use iph->tot_len, or just ipl.
246	 */
247
248	udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
249	if (udph == NULL)
250		return 0;
251
252	iph = ip_hdr(skb);
253	ul = ntohs(udph->len);
254
255	if (udplite || udph->check) {
256
257		udph->check = 0;
258
259		if (udplite) {
260			if (ul == 0)
261				skb->csum = csum_partial(udph, ipl - ihl, 0);
262			else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
263				skb->csum = csum_partial(udph, ul, 0);
264			else
265				goto ignore_obscure_skb;
266		} else {
267			if (ul != ipl - ihl)
268				goto ignore_obscure_skb;
269
270			skb->csum = csum_partial(udph, ul, 0);
271		}
272
273		udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
274						ul, iph->protocol,
275						skb->csum);
276
277		if (!udph->check)
278			udph->check = CSUM_MANGLED_0;
279	}
280
281	skb->ip_summed = CHECKSUM_NONE;
282
283ignore_obscure_skb:
284	return 1;
285}
286
287static int tcf_csum_ipv6_udp(struct sk_buff *skb,
288			     unsigned int ihl, unsigned int ipl, int udplite)
289{
290	struct udphdr *udph;
291	const struct ipv6hdr *ip6h;
292	u16 ul;
293
294	/*
295	 * Support both UDP and UDPLITE checksum algorithms, Don't use
296	 * udph->len to get the real length without any protocol check,
297	 * UDPLITE uses udph->len for another thing,
298	 * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
299	 */
300
301	udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
302	if (udph == NULL)
303		return 0;
304
305	ip6h = ipv6_hdr(skb);
306	ul = ntohs(udph->len);
307
308	udph->check = 0;
309
310	if (udplite) {
311		if (ul == 0)
312			skb->csum = csum_partial(udph, ipl - ihl, 0);
313
314		else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
315			skb->csum = csum_partial(udph, ul, 0);
316
317		else
318			goto ignore_obscure_skb;
319	} else {
320		if (ul != ipl - ihl)
321			goto ignore_obscure_skb;
322
323		skb->csum = csum_partial(udph, ul, 0);
324	}
325
326	udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
327				      udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
328				      skb->csum);
329
330	if (!udph->check)
331		udph->check = CSUM_MANGLED_0;
332
333	skb->ip_summed = CHECKSUM_NONE;
334
335ignore_obscure_skb:
336	return 1;
337}
338
339static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
340{
341	const struct iphdr *iph;
342	int ntkoff;
343
344	ntkoff = skb_network_offset(skb);
345
346	if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
347		goto fail;
348
349	iph = ip_hdr(skb);
350
351	switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
352	case IPPROTO_ICMP:
353		if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
354			if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
355						ntohs(iph->tot_len)))
356				goto fail;
357		break;
358	case IPPROTO_IGMP:
359		if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
360			if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
361						ntohs(iph->tot_len)))
362				goto fail;
363		break;
364	case IPPROTO_TCP:
365		if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
366			if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
367					       ntohs(iph->tot_len)))
368				goto fail;
369		break;
370	case IPPROTO_UDP:
371		if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
372			if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
373					       ntohs(iph->tot_len), 0))
374				goto fail;
375		break;
376	case IPPROTO_UDPLITE:
377		if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
378			if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
379					       ntohs(iph->tot_len), 1))
380				goto fail;
381		break;
382	}
383
384	if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
385		if (skb_cloned(skb) &&
386		    !skb_clone_writable(skb, sizeof(*iph) + ntkoff) &&
387		    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
388			goto fail;
389
390		ip_send_check(ip_hdr(skb));
391	}
392
393	return 1;
394
395fail:
396	return 0;
397}
398
399static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh,
400				 unsigned int ixhl, unsigned int *pl)
401{
402	int off, len, optlen;
403	unsigned char *xh = (void *)ip6xh;
404
405	off = sizeof(*ip6xh);
406	len = ixhl - off;
407
408	while (len > 1) {
409		switch (xh[off]) {
410		case IPV6_TLV_PAD1:
411			optlen = 1;
412			break;
413		case IPV6_TLV_JUMBO:
414			optlen = xh[off + 1] + 2;
415			if (optlen != 6 || len < 6 || (off & 3) != 2)
416				/* wrong jumbo option length/alignment */
417				return 0;
418			*pl = ntohl(*(__be32 *)(xh + off + 2));
419			goto done;
420		default:
421			optlen = xh[off + 1] + 2;
422			if (optlen > len)
423				/* ignore obscure options */
424				goto done;
425			break;
426		}
427		off += optlen;
428		len -= optlen;
429	}
430
431done:
432	return 1;
433}
434
435static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
436{
437	struct ipv6hdr *ip6h;
438	struct ipv6_opt_hdr *ip6xh;
439	unsigned int hl, ixhl;
440	unsigned int pl;
441	int ntkoff;
442	u8 nexthdr;
443
444	ntkoff = skb_network_offset(skb);
445
446	hl = sizeof(*ip6h);
447
448	if (!pskb_may_pull(skb, hl + ntkoff))
449		goto fail;
450
451	ip6h = ipv6_hdr(skb);
452
453	pl = ntohs(ip6h->payload_len);
454	nexthdr = ip6h->nexthdr;
455
456	do {
457		switch (nexthdr) {
458		case NEXTHDR_FRAGMENT:
459			goto ignore_skb;
460		case NEXTHDR_ROUTING:
461		case NEXTHDR_HOP:
462		case NEXTHDR_DEST:
463			if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
464				goto fail;
465			ip6xh = (void *)(skb_network_header(skb) + hl);
466			ixhl = ipv6_optlen(ip6xh);
467			if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
468				goto fail;
469			ip6xh = (void *)(skb_network_header(skb) + hl);
470			if ((nexthdr == NEXTHDR_HOP) &&
471			    !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
472				goto fail;
473			nexthdr = ip6xh->nexthdr;
474			hl += ixhl;
475			break;
476		case IPPROTO_ICMPV6:
477			if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
478				if (!tcf_csum_ipv6_icmp(skb,
479							hl, pl + sizeof(*ip6h)))
480					goto fail;
481			goto done;
482		case IPPROTO_TCP:
483			if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
484				if (!tcf_csum_ipv6_tcp(skb,
485						       hl, pl + sizeof(*ip6h)))
486					goto fail;
487			goto done;
488		case IPPROTO_UDP:
489			if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
490				if (!tcf_csum_ipv6_udp(skb, hl,
491						       pl + sizeof(*ip6h), 0))
492					goto fail;
493			goto done;
494		case IPPROTO_UDPLITE:
495			if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
496				if (!tcf_csum_ipv6_udp(skb, hl,
497						       pl + sizeof(*ip6h), 1))
498					goto fail;
499			goto done;
500		default:
501			goto ignore_skb;
502		}
503	} while (pskb_may_pull(skb, hl + 1 + ntkoff));
504
505done:
506ignore_skb:
507	return 1;
508
509fail:
510	return 0;
511}
512
513static int tcf_csum(struct sk_buff *skb,
514		    const struct tc_action *a, struct tcf_result *res)
515{
516	struct tcf_csum *p = a->priv;
517	int action;
518	u32 update_flags;
519
520	spin_lock(&p->tcf_lock);
521	p->tcf_tm.lastuse = jiffies;
522	bstats_update(&p->tcf_bstats, skb);
523	action = p->tcf_action;
524	update_flags = p->update_flags;
525	spin_unlock(&p->tcf_lock);
526
527	if (unlikely(action == TC_ACT_SHOT))
528		goto drop;
529
530	switch (skb->protocol) {
531	case cpu_to_be16(ETH_P_IP):
532		if (!tcf_csum_ipv4(skb, update_flags))
533			goto drop;
534		break;
535	case cpu_to_be16(ETH_P_IPV6):
536		if (!tcf_csum_ipv6(skb, update_flags))
537			goto drop;
538		break;
539	}
540
541	return action;
542
543drop:
544	spin_lock(&p->tcf_lock);
545	p->tcf_qstats.drops++;
546	spin_unlock(&p->tcf_lock);
547	return TC_ACT_SHOT;
548}
549
550static int tcf_csum_dump(struct sk_buff *skb,
551			 struct tc_action *a, int bind, int ref)
552{
553	unsigned char *b = skb_tail_pointer(skb);
554	struct tcf_csum *p = a->priv;
555	struct tc_csum opt = {
556		.update_flags = p->update_flags,
557		.index   = p->tcf_index,
558		.action  = p->tcf_action,
559		.refcnt  = p->tcf_refcnt - ref,
560		.bindcnt = p->tcf_bindcnt - bind,
561	};
562	struct tcf_t t;
563
564	if (nla_put(skb, TCA_CSUM_PARMS, sizeof(opt), &opt))
565		goto nla_put_failure;
566	t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
567	t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
568	t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
569	if (nla_put(skb, TCA_CSUM_TM, sizeof(t), &t))
570		goto nla_put_failure;
571
572	return skb->len;
573
574nla_put_failure:
575	nlmsg_trim(skb, b);
576	return -1;
577}
578
579static struct tc_action_ops act_csum_ops = {
580	.kind		= "csum",
581	.hinfo		= &csum_hash_info,
582	.type		= TCA_ACT_CSUM,
583	.capab		= TCA_CAP_NONE,
584	.owner		= THIS_MODULE,
585	.act		= tcf_csum,
586	.dump		= tcf_csum_dump,
587	.cleanup	= tcf_csum_cleanup,
588	.lookup		= tcf_hash_search,
589	.init		= tcf_csum_init,
590	.walk		= tcf_generic_walker
591};
592
593MODULE_DESCRIPTION("Checksum updating actions");
594MODULE_LICENSE("GPL");
595
596static int __init csum_init_module(void)
597{
598	return tcf_register_action(&act_csum_ops);
599}
600
601static void __exit csum_cleanup_module(void)
602{
603	tcf_unregister_action(&act_csum_ops);
604}
605
606module_init(csum_init_module);
607module_exit(csum_cleanup_module);
608