net_sched: factorize qdisc stats handling
[linux-2.6.git] / net / sched / act_csum.c
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 static struct tcf_common *tcf_csum_ht[CSUM_TAB_MASK + 1];
41 static u32 csum_idx_gen;
42 static DEFINE_RWLOCK(csum_lock);
43
44 static struct tcf_hashinfo csum_hash_info = {
45         .htab   = tcf_csum_ht,
46         .hmask  = CSUM_TAB_MASK,
47         .lock   = &csum_lock,
48 };
49
50 static const struct nla_policy csum_policy[TCA_CSUM_MAX + 1] = {
51         [TCA_CSUM_PARMS] = { .len = sizeof(struct tc_csum), },
52 };
53
54 static int tcf_csum_init(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
101 static 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  */
117 static 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
133 static 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
151 static 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
169 static int tcf_csum_ipv6_icmp(struct sk_buff *skb, struct ipv6hdr *ip6h,
170                               unsigned int ihl, unsigned int ipl)
171 {
172         struct icmp6hdr *icmp6h;
173
174         icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
175         if (icmp6h == NULL)
176                 return 0;
177
178         icmp6h->icmp6_cksum = 0;
179         skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
180         icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
181                                               ipl - ihl, IPPROTO_ICMPV6,
182                                               skb->csum);
183
184         skb->ip_summed = CHECKSUM_NONE;
185
186         return 1;
187 }
188
189 static int tcf_csum_ipv4_tcp(struct sk_buff *skb, struct iphdr *iph,
190                              unsigned int ihl, unsigned int ipl)
191 {
192         struct tcphdr *tcph;
193
194         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
195         if (tcph == NULL)
196                 return 0;
197
198         tcph->check = 0;
199         skb->csum = csum_partial(tcph, ipl - ihl, 0);
200         tcph->check = tcp_v4_check(ipl - ihl,
201                                    iph->saddr, iph->daddr, skb->csum);
202
203         skb->ip_summed = CHECKSUM_NONE;
204
205         return 1;
206 }
207
208 static int tcf_csum_ipv6_tcp(struct sk_buff *skb, struct ipv6hdr *ip6h,
209                              unsigned int ihl, unsigned int ipl)
210 {
211         struct tcphdr *tcph;
212
213         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
214         if (tcph == NULL)
215                 return 0;
216
217         tcph->check = 0;
218         skb->csum = csum_partial(tcph, ipl - ihl, 0);
219         tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
220                                       ipl - ihl, IPPROTO_TCP,
221                                       skb->csum);
222
223         skb->ip_summed = CHECKSUM_NONE;
224
225         return 1;
226 }
227
228 static int tcf_csum_ipv4_udp(struct sk_buff *skb, struct iphdr *iph,
229                              unsigned int ihl, unsigned int ipl, int udplite)
230 {
231         struct udphdr *udph;
232         u16 ul;
233
234         /*
235          * Support both UDP and UDPLITE checksum algorithms, Don't use
236          * udph->len to get the real length without any protocol check,
237          * UDPLITE uses udph->len for another thing,
238          * Use iph->tot_len, or just ipl.
239          */
240
241         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
242         if (udph == NULL)
243                 return 0;
244
245         ul = ntohs(udph->len);
246
247         if (udplite || udph->check) {
248
249                 udph->check = 0;
250
251                 if (udplite) {
252                         if (ul == 0)
253                                 skb->csum = csum_partial(udph, ipl - ihl, 0);
254                         else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
255                                 skb->csum = csum_partial(udph, ul, 0);
256                         else
257                                 goto ignore_obscure_skb;
258                 } else {
259                         if (ul != ipl - ihl)
260                                 goto ignore_obscure_skb;
261
262                         skb->csum = csum_partial(udph, ul, 0);
263                 }
264
265                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
266                                                 ul, iph->protocol,
267                                                 skb->csum);
268
269                 if (!udph->check)
270                         udph->check = CSUM_MANGLED_0;
271         }
272
273         skb->ip_summed = CHECKSUM_NONE;
274
275 ignore_obscure_skb:
276         return 1;
277 }
278
279 static int tcf_csum_ipv6_udp(struct sk_buff *skb, struct ipv6hdr *ip6h,
280                              unsigned int ihl, unsigned int ipl, int udplite)
281 {
282         struct udphdr *udph;
283         u16 ul;
284
285         /*
286          * Support both UDP and UDPLITE checksum algorithms, Don't use
287          * udph->len to get the real length without any protocol check,
288          * UDPLITE uses udph->len for another thing,
289          * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
290          */
291
292         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
293         if (udph == NULL)
294                 return 0;
295
296         ul = ntohs(udph->len);
297
298         udph->check = 0;
299
300         if (udplite) {
301                 if (ul == 0)
302                         skb->csum = csum_partial(udph, ipl - ihl, 0);
303
304                 else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
305                         skb->csum = csum_partial(udph, ul, 0);
306
307                 else
308                         goto ignore_obscure_skb;
309         } else {
310                 if (ul != ipl - ihl)
311                         goto ignore_obscure_skb;
312
313                 skb->csum = csum_partial(udph, ul, 0);
314         }
315
316         udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
317                                       udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
318                                       skb->csum);
319
320         if (!udph->check)
321                 udph->check = CSUM_MANGLED_0;
322
323         skb->ip_summed = CHECKSUM_NONE;
324
325 ignore_obscure_skb:
326         return 1;
327 }
328
329 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
330 {
331         struct iphdr *iph;
332         int ntkoff;
333
334         ntkoff = skb_network_offset(skb);
335
336         if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
337                 goto fail;
338
339         iph = ip_hdr(skb);
340
341         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
342         case IPPROTO_ICMP:
343                 if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
344                         if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
345                                                 ntohs(iph->tot_len)))
346                                 goto fail;
347                 break;
348         case IPPROTO_IGMP:
349                 if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
350                         if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
351                                                 ntohs(iph->tot_len)))
352                                 goto fail;
353                 break;
354         case IPPROTO_TCP:
355                 if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
356                         if (!tcf_csum_ipv4_tcp(skb, iph, iph->ihl * 4,
357                                                ntohs(iph->tot_len)))
358                                 goto fail;
359                 break;
360         case IPPROTO_UDP:
361                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
362                         if (!tcf_csum_ipv4_udp(skb, iph, iph->ihl * 4,
363                                                ntohs(iph->tot_len), 0))
364                                 goto fail;
365                 break;
366         case IPPROTO_UDPLITE:
367                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
368                         if (!tcf_csum_ipv4_udp(skb, iph, iph->ihl * 4,
369                                                ntohs(iph->tot_len), 1))
370                                 goto fail;
371                 break;
372         }
373
374         if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
375                 if (skb_cloned(skb) &&
376                     !skb_clone_writable(skb, sizeof(*iph) + ntkoff) &&
377                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
378                         goto fail;
379
380                 ip_send_check(iph);
381         }
382
383         return 1;
384
385 fail:
386         return 0;
387 }
388
389 static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh,
390                                  unsigned int ixhl, unsigned int *pl)
391 {
392         int off, len, optlen;
393         unsigned char *xh = (void *)ip6xh;
394
395         off = sizeof(*ip6xh);
396         len = ixhl - off;
397
398         while (len > 1) {
399                 switch (xh[off]) {
400                 case IPV6_TLV_PAD0:
401                         optlen = 1;
402                         break;
403                 case IPV6_TLV_JUMBO:
404                         optlen = xh[off + 1] + 2;
405                         if (optlen != 6 || len < 6 || (off & 3) != 2)
406                                 /* wrong jumbo option length/alignment */
407                                 return 0;
408                         *pl = ntohl(*(__be32 *)(xh + off + 2));
409                         goto done;
410                 default:
411                         optlen = xh[off + 1] + 2;
412                         if (optlen > len)
413                                 /* ignore obscure options */
414                                 goto done;
415                         break;
416                 }
417                 off += optlen;
418                 len -= optlen;
419         }
420
421 done:
422         return 1;
423 }
424
425 static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
426 {
427         struct ipv6hdr *ip6h;
428         struct ipv6_opt_hdr *ip6xh;
429         unsigned int hl, ixhl;
430         unsigned int pl;
431         int ntkoff;
432         u8 nexthdr;
433
434         ntkoff = skb_network_offset(skb);
435
436         hl = sizeof(*ip6h);
437
438         if (!pskb_may_pull(skb, hl + ntkoff))
439                 goto fail;
440
441         ip6h = ipv6_hdr(skb);
442
443         pl = ntohs(ip6h->payload_len);
444         nexthdr = ip6h->nexthdr;
445
446         do {
447                 switch (nexthdr) {
448                 case NEXTHDR_FRAGMENT:
449                         goto ignore_skb;
450                 case NEXTHDR_ROUTING:
451                 case NEXTHDR_HOP:
452                 case NEXTHDR_DEST:
453                         if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
454                                 goto fail;
455                         ip6xh = (void *)(skb_network_header(skb) + hl);
456                         ixhl = ipv6_optlen(ip6xh);
457                         if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
458                                 goto fail;
459                         if ((nexthdr == NEXTHDR_HOP) &&
460                             !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
461                                 goto fail;
462                         nexthdr = ip6xh->nexthdr;
463                         hl += ixhl;
464                         break;
465                 case IPPROTO_ICMPV6:
466                         if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
467                                 if (!tcf_csum_ipv6_icmp(skb, ip6h,
468                                                         hl, pl + sizeof(*ip6h)))
469                                         goto fail;
470                         goto done;
471                 case IPPROTO_TCP:
472                         if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
473                                 if (!tcf_csum_ipv6_tcp(skb, ip6h,
474                                                        hl, pl + sizeof(*ip6h)))
475                                         goto fail;
476                         goto done;
477                 case IPPROTO_UDP:
478                         if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
479                                 if (!tcf_csum_ipv6_udp(skb, ip6h, hl,
480                                                        pl + sizeof(*ip6h), 0))
481                                         goto fail;
482                         goto done;
483                 case IPPROTO_UDPLITE:
484                         if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
485                                 if (!tcf_csum_ipv6_udp(skb, ip6h, hl,
486                                                        pl + sizeof(*ip6h), 1))
487                                         goto fail;
488                         goto done;
489                 default:
490                         goto ignore_skb;
491                 }
492         } while (pskb_may_pull(skb, hl + 1 + ntkoff));
493
494 done:
495 ignore_skb:
496         return 1;
497
498 fail:
499         return 0;
500 }
501
502 static int tcf_csum(struct sk_buff *skb,
503                     struct tc_action *a, struct tcf_result *res)
504 {
505         struct tcf_csum *p = a->priv;
506         int action;
507         u32 update_flags;
508
509         spin_lock(&p->tcf_lock);
510         p->tcf_tm.lastuse = jiffies;
511         bstats_update(&p->tcf_bstats, skb);
512         action = p->tcf_action;
513         update_flags = p->update_flags;
514         spin_unlock(&p->tcf_lock);
515
516         if (unlikely(action == TC_ACT_SHOT))
517                 goto drop;
518
519         switch (skb->protocol) {
520         case cpu_to_be16(ETH_P_IP):
521                 if (!tcf_csum_ipv4(skb, update_flags))
522                         goto drop;
523                 break;
524         case cpu_to_be16(ETH_P_IPV6):
525                 if (!tcf_csum_ipv6(skb, update_flags))
526                         goto drop;
527                 break;
528         }
529
530         return action;
531
532 drop:
533         spin_lock(&p->tcf_lock);
534         p->tcf_qstats.drops++;
535         spin_unlock(&p->tcf_lock);
536         return TC_ACT_SHOT;
537 }
538
539 static int tcf_csum_dump(struct sk_buff *skb,
540                          struct tc_action *a, int bind, int ref)
541 {
542         unsigned char *b = skb_tail_pointer(skb);
543         struct tcf_csum *p = a->priv;
544         struct tc_csum opt = {
545                 .update_flags = p->update_flags,
546                 .index   = p->tcf_index,
547                 .action  = p->tcf_action,
548                 .refcnt  = p->tcf_refcnt - ref,
549                 .bindcnt = p->tcf_bindcnt - bind,
550         };
551         struct tcf_t t;
552
553         NLA_PUT(skb, TCA_CSUM_PARMS, sizeof(opt), &opt);
554         t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
555         t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
556         t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
557         NLA_PUT(skb, TCA_CSUM_TM, sizeof(t), &t);
558
559         return skb->len;
560
561 nla_put_failure:
562         nlmsg_trim(skb, b);
563         return -1;
564 }
565
566 static struct tc_action_ops act_csum_ops = {
567         .kind           = "csum",
568         .hinfo          = &csum_hash_info,
569         .type           = TCA_ACT_CSUM,
570         .capab          = TCA_CAP_NONE,
571         .owner          = THIS_MODULE,
572         .act            = tcf_csum,
573         .dump           = tcf_csum_dump,
574         .cleanup        = tcf_csum_cleanup,
575         .lookup         = tcf_hash_search,
576         .init           = tcf_csum_init,
577         .walk           = tcf_generic_walker
578 };
579
580 MODULE_DESCRIPTION("Checksum updating actions");
581 MODULE_LICENSE("GPL");
582
583 static int __init csum_init_module(void)
584 {
585         return tcf_register_action(&act_csum_ops);
586 }
587
588 static void __exit csum_cleanup_module(void)
589 {
590         tcf_unregister_action(&act_csum_ops);
591 }
592
593 module_init(csum_init_module);
594 module_exit(csum_cleanup_module);