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_try_make_writable(skb, hl + ntkoff))
 126                 return NULL;
 127         else
 128                 return (void *)(skb_network_header(skb) + ihl);
 129 }
 130
 131 static int tcf_csum_ipv4_icmp(struct sk_buff *skb,
 132                               unsigned int ihl, unsigned int ipl)
 133 {
 134         struct icmphdr *icmph;
 135
 136         icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
 137         if (icmph == NULL)
 138                 return 0;
 139
 140         icmph->checksum = 0;
 141         skb->csum = csum_partial(icmph, ipl - ihl, 0);
 142         icmph->checksum = csum_fold(skb->csum);
 143
 144         skb->ip_summed = CHECKSUM_NONE;
 145
 146         return 1;
 147 }
 148
 149 static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
 150                               unsigned int ihl, unsigned int ipl)
 151 {
 152         struct igmphdr *igmph;
 153
 154         igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
 155         if (igmph == NULL)
 156                 return 0;
 157
 158         igmph->csum = 0;
 159         skb->csum = csum_partial(igmph, ipl - ihl, 0);
 160         igmph->csum = csum_fold(skb->csum);
 161
 162         skb->ip_summed = CHECKSUM_NONE;
 163
 164         return 1;
 165 }
 166
 167 static int tcf_csum_ipv6_icmp(struct sk_buff *skb, struct ipv6hdr *ip6h,
 168                               unsigned int ihl, unsigned int ipl)
 169 {
 170         struct icmp6hdr *icmp6h;
 171
 172         icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
 173         if (icmp6h == NULL)
 174                 return 0;
 175
 176         icmp6h->icmp6_cksum = 0;
 177         skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
 178         icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 179                                               ipl - ihl, IPPROTO_ICMPV6,
 180                                               skb->csum);
 181
 182         skb->ip_summed = CHECKSUM_NONE;
 183
 184         return 1;
 185 }
 186
 187 static int tcf_csum_ipv4_tcp(struct sk_buff *skb, struct iphdr *iph,
 188                              unsigned int ihl, unsigned int ipl)
 189 {
 190         struct tcphdr *tcph;
 191
 192         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 193         if (tcph == NULL)
 194                 return 0;
 195
 196         tcph->check = 0;
 197         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 198         tcph->check = tcp_v4_check(ipl - ihl,
 199                                    iph->saddr, iph->daddr, skb->csum);
 200
 201         skb->ip_summed = CHECKSUM_NONE;
 202
 203         return 1;
 204 }
 205
 206 static int tcf_csum_ipv6_tcp(struct sk_buff *skb, struct ipv6hdr *ip6h,
 207                              unsigned int ihl, unsigned int ipl)
 208 {
 209         struct tcphdr *tcph;
 210
 211         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 212         if (tcph == NULL)
 213                 return 0;
 214
 215         tcph->check = 0;
 216         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 217         tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 218                                       ipl - ihl, IPPROTO_TCP,
 219                                       skb->csum);
 220
 221         skb->ip_summed = CHECKSUM_NONE;
 222
 223         return 1;
 224 }
 225
 226 static int tcf_csum_ipv4_udp(struct sk_buff *skb, struct iphdr *iph,
 227                              unsigned int ihl, unsigned int ipl, int udplite)
 228 {
 229         struct udphdr *udph;
 230         u16 ul;
 231
 232         /*
 233          * Support both UDP and UDPLITE checksum algorithms, Don't use
 234          * udph->len to get the real length without any protocol check,
 235          * UDPLITE uses udph->len for another thing,
 236          * Use iph->tot_len, or just ipl.
 237          */
 238
 239         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
 240         if (udph == NULL)
 241                 return 0;
 242
 243         ul = ntohs(udph->len);
 244
 245         if (udplite || udph->check) {
 246
 247                 udph->check = 0;
 248
 249                 if (udplite) {
 250                         if (ul == 0)
 251                                 skb->csum = csum_partial(udph, ipl - ihl, 0);
 252                         else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 253                                 skb->csum = csum_partial(udph, ul, 0);
 254                         else
 255                                 goto ignore_obscure_skb;
 256                 } else {
 257                         if (ul != ipl - ihl)
 258                                 goto ignore_obscure_skb;
 259
 260                         skb->csum = csum_partial(udph, ul, 0);
 261                 }
 262
 263                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 264                                                 ul, iph->protocol,
 265                                                 skb->csum);
 266
 267                 if (!udph->check)
 268                         udph->check = CSUM_MANGLED_0;
 269         }
 270
 271         skb->ip_summed = CHECKSUM_NONE;
 272
 273 ignore_obscure_skb:
 274         return 1;
 275 }
 276
 277 static int tcf_csum_ipv6_udp(struct sk_buff *skb, struct ipv6hdr *ip6h,
 278                              unsigned int ihl, unsigned int ipl, int udplite)
 279 {
 280         struct udphdr *udph;
 281         u16 ul;
 282
 283         /*
 284          * Support both UDP and UDPLITE checksum algorithms, Don't use
 285          * udph->len to get the real length without any protocol check,
 286          * UDPLITE uses udph->len for another thing,
 287          * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
 288          */
 289
 290         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
 291         if (udph == NULL)
 292                 return 0;
 293
 294         ul = ntohs(udph->len);
 295
 296         udph->check = 0;
 297
 298         if (udplite) {
 299                 if (ul == 0)
 300                         skb->csum = csum_partial(udph, ipl - ihl, 0);
 301
 302                 else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 303                         skb->csum = csum_partial(udph, ul, 0);
 304
 305                 else
 306                         goto ignore_obscure_skb;
 307         } else {
 308                 if (ul != ipl - ihl)
 309                         goto ignore_obscure_skb;
 310
 311                 skb->csum = csum_partial(udph, ul, 0);
 312         }
 313
 314         udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
 315                                       udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
 316                                       skb->csum);
 317
 318         if (!udph->check)
 319                 udph->check = CSUM_MANGLED_0;
 320
 321         skb->ip_summed = CHECKSUM_NONE;
 322
 323 ignore_obscure_skb:
 324         return 1;
 325 }
 326
 327 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
 328 {
 329         struct iphdr *iph;
 330         int ntkoff;
 331
 332         ntkoff = skb_network_offset(skb);
 333
 334         if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
 335                 goto fail;
 336
 337         iph = ip_hdr(skb);
 338
 339         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 340         case IPPROTO_ICMP:
 341                 if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
 342                         if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
 343                                                 ntohs(iph->tot_len)))
 344                                 goto fail;
 345                 break;
 346         case IPPROTO_IGMP:
 347                 if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
 348                         if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
 349                                                 ntohs(iph->tot_len)))
 350                                 goto fail;
 351                 break;
 352         case IPPROTO_TCP:
 353                 if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
 354                         if (!tcf_csum_ipv4_tcp(skb, iph, iph->ihl * 4,
 355                                                ntohs(iph->tot_len)))
 356                                 goto fail;
 357                 break;
 358         case IPPROTO_UDP:
 359                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
 360                         if (!tcf_csum_ipv4_udp(skb, iph, iph->ihl * 4,
 361                                                ntohs(iph->tot_len), 0))
 362                                 goto fail;
 363                 break;
 364         case IPPROTO_UDPLITE:
 365                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
 366                         if (!tcf_csum_ipv4_udp(skb, iph, iph->ihl * 4,
 367                                                ntohs(iph->tot_len), 1))
 368                                 goto fail;
 369                 break;
 370         }
 371
 372         if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
 373                 if (skb_try_make_writable(skb, sizeof(*iph) + ntkoff))
 374                         goto fail;
 375
 376                 ip_send_check(iph);
 377         }
 378
 379         return 1;
 380
 381 fail:
 382         return 0;
 383 }
 384
 385 static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh,
 386                                  unsigned int ixhl, unsigned int *pl)
 387 {
 388         int off, len, optlen;
 389         unsigned char *xh = (void *)ip6xh;
 390
 391         off = sizeof(*ip6xh);
 392         len = ixhl - off;
 393
 394         while (len > 1) {
 395                 switch (xh[off]) {
 396                 case IPV6_TLV_PAD0:
 397                         optlen = 1;
 398                         break;
 399                 case IPV6_TLV_JUMBO:
 400                         optlen = xh[off + 1] + 2;
 401                         if (optlen != 6 || len < 6 || (off & 3) != 2)
 402                                 /* wrong jumbo option length/alignment */
 403                                 return 0;
 404                         *pl = ntohl(*(__be32 *)(xh + off + 2));
 405                         goto done;
 406                 default:
 407                         optlen = xh[off + 1] + 2;
 408                         if (optlen > len)
 409                                 /* ignore obscure options */
 410                                 goto done;
 411                         break;
 412                 }
 413                 off += optlen;
 414                 len -= optlen;
 415         }
 416
 417 done:
 418         return 1;
 419 }
 420
 421 static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
 422 {
 423         struct ipv6hdr *ip6h;
 424         struct ipv6_opt_hdr *ip6xh;
 425         unsigned int hl, ixhl;
 426         unsigned int pl;
 427         int ntkoff;
 428         u8 nexthdr;
 429
 430         ntkoff = skb_network_offset(skb);
 431
 432         hl = sizeof(*ip6h);
 433
 434         if (!pskb_may_pull(skb, hl + ntkoff))
 435                 goto fail;
 436
 437         ip6h = ipv6_hdr(skb);
 438
 439         pl = ntohs(ip6h->payload_len);
 440         nexthdr = ip6h->nexthdr;
 441
 442         do {
 443                 switch (nexthdr) {
 444                 case NEXTHDR_FRAGMENT:
 445                         goto ignore_skb;
 446                 case NEXTHDR_ROUTING:
 447                 case NEXTHDR_HOP:
 448                 case NEXTHDR_DEST:
 449                         if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
 450                                 goto fail;
 451                         ip6xh = (void *)(skb_network_header(skb) + hl);
 452                         ixhl = ipv6_optlen(ip6xh);
 453                         if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
 454                                 goto fail;
 455                         if ((nexthdr == NEXTHDR_HOP) &&
 456                             !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
 457                                 goto fail;
 458                         nexthdr = ip6xh->nexthdr;
 459                         hl += ixhl;
 460                         break;
 461                 case IPPROTO_ICMPV6:
 462                         if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
 463                                 if (!tcf_csum_ipv6_icmp(skb, ip6h,
 464                                                         hl, pl + sizeof(*ip6h)))
 465                                         goto fail;
 466                         goto done;
 467                 case IPPROTO_TCP:
 468                         if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
 469                                 if (!tcf_csum_ipv6_tcp(skb, ip6h,
 470                                                        hl, pl + sizeof(*ip6h)))
 471                                         goto fail;
 472                         goto done;
 473                 case IPPROTO_UDP:
 474                         if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
 475                                 if (!tcf_csum_ipv6_udp(skb, ip6h, hl,
 476                                                        pl + sizeof(*ip6h), 0))
 477                                         goto fail;
 478                         goto done;
 479                 case IPPROTO_UDPLITE:
 480                         if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
 481                                 if (!tcf_csum_ipv6_udp(skb, ip6h, hl,
 482                                                        pl + sizeof(*ip6h), 1))
 483                                         goto fail;
 484                         goto done;
 485                 default:
 486                         goto ignore_skb;
 487                 }
 488         } while (pskb_may_pull(skb, hl + 1 + ntkoff));
 489
 490 done:
 491 ignore_skb:
 492         return 1;
 493
 494 fail:
 495         return 0;
 496 }
 497
 498 static int tcf_csum(struct sk_buff *skb,
 499                     const struct tc_action *a, struct tcf_result *res)
 500 {
 501         struct tcf_csum *p = a->priv;
 502         int action;
 503         u32 update_flags;
 504
 505         spin_lock(&p->tcf_lock);
 506         p->tcf_tm.lastuse = jiffies;
 507         bstats_update(&p->tcf_bstats, skb);
 508         action = p->tcf_action;
 509         update_flags = p->update_flags;
 510         spin_unlock(&p->tcf_lock);
 511
 512         if (unlikely(action == TC_ACT_SHOT))
 513                 goto drop;
 514
 515         switch (skb->protocol) {
 516         case cpu_to_be16(ETH_P_IP):
 517                 if (!tcf_csum_ipv4(skb, update_flags))
 518                         goto drop;
 519                 break;
 520         case cpu_to_be16(ETH_P_IPV6):
 521                 if (!tcf_csum_ipv6(skb, update_flags))
 522                         goto drop;
 523                 break;
 524         }
 525
 526         return action;
 527
 528 drop:
 529         spin_lock(&p->tcf_lock);
 530         p->tcf_qstats.drops++;
 531         spin_unlock(&p->tcf_lock);
 532         return TC_ACT_SHOT;
 533 }
 534
 535 static int tcf_csum_dump(struct sk_buff *skb,
 536                          struct tc_action *a, int bind, int ref)
 537 {
 538         unsigned char *b = skb_tail_pointer(skb);
 539         struct tcf_csum *p = a->priv;
 540         struct tc_csum opt = {
 541                 .update_flags = p->update_flags,
 542                 .index   = p->tcf_index,
 543                 .action  = p->tcf_action,
 544                 .refcnt  = p->tcf_refcnt - ref,
 545                 .bindcnt = p->tcf_bindcnt - bind,
 546         };
 547         struct tcf_t t;
 548
 549         NLA_PUT(skb, TCA_CSUM_PARMS, sizeof(opt), &opt);
 550         t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
 551         t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
 552         t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
 553         NLA_PUT(skb, TCA_CSUM_TM, sizeof(t), &t);
 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         .hinfo          = &csum_hash_info,
 565         .type           = TCA_ACT_CSUM,
 566         .capab          = TCA_CAP_NONE,
 567         .owner          = THIS_MODULE,
 568         .act            = tcf_csum,
 569         .dump           = tcf_csum_dump,
 570         .cleanup        = tcf_csum_cleanup,
 571         .lookup         = tcf_hash_search,
 572         .init           = tcf_csum_init,
 573         .walk           = tcf_generic_walker
 574 };
 575
 576 MODULE_DESCRIPTION("Checksum updating actions");
 577 MODULE_LICENSE("GPL");
 578
 579 static int __init csum_init_module(void)
 580 {
 581         return tcf_register_action(&act_csum_ops);
 582 }
 583
 584 static void __exit csum_cleanup_module(void)
 585 {
 586         tcf_unregister_action(&act_csum_ops);
 587 }
 588
 589 module_init(csum_init_module);
 590 module_exit(csum_cleanup_module);