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
  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_try_make_writable(skb, hl + ntkoff))
 109                 return NULL;
 110         else
 111                 return (void *)(skb_network_header(skb) + ihl);
 112 }
 113
 114 static int tcf_csum_ipv4_icmp(struct sk_buff *skb,
 115                               unsigned int ihl, unsigned int ipl)
 116 {
 117         struct icmphdr *icmph;
 118
 119         icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
 120         if (icmph == NULL)
 121                 return 0;
 122
 123         icmph->checksum = 0;
 124         skb->csum = csum_partial(icmph, ipl - ihl, 0);
 125         icmph->checksum = csum_fold(skb->csum);
 126
 127         skb->ip_summed = CHECKSUM_NONE;
 128
 129         return 1;
 130 }
 131
 132 static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
 133                               unsigned int ihl, unsigned int ipl)
 134 {
 135         struct igmphdr *igmph;
 136
 137         igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
 138         if (igmph == NULL)
 139                 return 0;
 140
 141         igmph->csum = 0;
 142         skb->csum = csum_partial(igmph, ipl - ihl, 0);
 143         igmph->csum = csum_fold(skb->csum);
 144
 145         skb->ip_summed = CHECKSUM_NONE;
 146
 147         return 1;
 148 }
 149
 150 static int tcf_csum_ipv6_icmp(struct sk_buff *skb,
 151                               unsigned int ihl, unsigned int ipl)
 152 {
 153         struct icmp6hdr *icmp6h;
 154         const struct ipv6hdr *ip6h;
 155
 156         icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
 157         if (icmp6h == NULL)
 158                 return 0;
 159
 160         ip6h = ipv6_hdr(skb);
 161         icmp6h->icmp6_cksum = 0;
 162         skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
 163         icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 164                                               ipl - ihl, IPPROTO_ICMPV6,
 165                                               skb->csum);
 166
 167         skb->ip_summed = CHECKSUM_NONE;
 168
 169         return 1;
 170 }
 171
 172 static int tcf_csum_ipv4_tcp(struct sk_buff *skb,
 173                              unsigned int ihl, unsigned int ipl)
 174 {
 175         struct tcphdr *tcph;
 176         const struct iphdr *iph;
 177
 178         if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 179                 return 1;
 180
 181         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 182         if (tcph == NULL)
 183                 return 0;
 184
 185         iph = ip_hdr(skb);
 186         tcph->check = 0;
 187         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 188         tcph->check = tcp_v4_check(ipl - ihl,
 189                                    iph->saddr, iph->daddr, skb->csum);
 190
 191         skb->ip_summed = CHECKSUM_NONE;
 192
 193         return 1;
 194 }
 195
 196 static int tcf_csum_ipv6_tcp(struct sk_buff *skb,
 197                              unsigned int ihl, unsigned int ipl)
 198 {
 199         struct tcphdr *tcph;
 200         const struct ipv6hdr *ip6h;
 201
 202         if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 203                 return 1;
 204
 205         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 206         if (tcph == NULL)
 207                 return 0;
 208
 209         ip6h = ipv6_hdr(skb);
 210         tcph->check = 0;
 211         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 212         tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 213                                       ipl - ihl, IPPROTO_TCP,
 214                                       skb->csum);
 215
 216         skb->ip_summed = CHECKSUM_NONE;
 217
 218         return 1;
 219 }
 220
 221 static int tcf_csum_ipv4_udp(struct sk_buff *skb,
 222                              unsigned int ihl, unsigned int ipl, int udplite)
 223 {
 224         struct udphdr *udph;
 225         const struct iphdr *iph;
 226         u16 ul;
 227
 228         if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 229                 return 1;
 230
 231         /*
 232          * Support both UDP and UDPLITE checksum algorithms, Don't use
 233          * udph->len to get the real length without any protocol check,
 234          * UDPLITE uses udph->len for another thing,
 235          * Use iph->tot_len, or just ipl.
 236          */
 237
 238         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
 239         if (udph == NULL)
 240                 return 0;
 241
 242         iph = ip_hdr(skb);
 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,
 278                              unsigned int ihl, unsigned int ipl, int udplite)
 279 {
 280         struct udphdr *udph;
 281         const struct ipv6hdr *ip6h;
 282         u16 ul;
 283
 284         if (skb_is_gso(skb) && skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 285                 return 1;
 286
 287         /*
 288          * Support both UDP and UDPLITE checksum algorithms, Don't use
 289          * udph->len to get the real length without any protocol check,
 290          * UDPLITE uses udph->len for another thing,
 291          * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
 292          */
 293
 294         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
 295         if (udph == NULL)
 296                 return 0;
 297
 298         ip6h = ipv6_hdr(skb);
 299         ul = ntohs(udph->len);
 300
 301         udph->check = 0;
 302
 303         if (udplite) {
 304                 if (ul == 0)
 305                         skb->csum = csum_partial(udph, ipl - ihl, 0);
 306
 307                 else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 308                         skb->csum = csum_partial(udph, ul, 0);
 309
 310                 else
 311                         goto ignore_obscure_skb;
 312         } else {
 313                 if (ul != ipl - ihl)
 314                         goto ignore_obscure_skb;
 315
 316                 skb->csum = csum_partial(udph, ul, 0);
 317         }
 318
 319         udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
 320                                       udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
 321                                       skb->csum);
 322
 323         if (!udph->check)
 324                 udph->check = CSUM_MANGLED_0;
 325
 326         skb->ip_summed = CHECKSUM_NONE;
 327
 328 ignore_obscure_skb:
 329         return 1;
 330 }
 331
 332 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
 333 {
 334         const struct iphdr *iph;
 335         int ntkoff;
 336
 337         ntkoff = skb_network_offset(skb);
 338
 339         if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
 340                 goto fail;
 341
 342         iph = ip_hdr(skb);
 343
 344         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 345         case IPPROTO_ICMP:
 346                 if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
 347                         if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
 348                                                 ntohs(iph->tot_len)))
 349                                 goto fail;
 350                 break;
 351         case IPPROTO_IGMP:
 352                 if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
 353                         if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
 354                                                 ntohs(iph->tot_len)))
 355                                 goto fail;
 356                 break;
 357         case IPPROTO_TCP:
 358                 if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
 359                         if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
 360                                                ntohs(iph->tot_len)))
 361                                 goto fail;
 362                 break;
 363         case IPPROTO_UDP:
 364                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
 365                         if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
 366                                                ntohs(iph->tot_len), 0))
 367                                 goto fail;
 368                 break;
 369         case IPPROTO_UDPLITE:
 370                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
 371                         if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
 372                                                ntohs(iph->tot_len), 1))
 373                                 goto fail;
 374                 break;
 375         }
 376
 377         if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
 378                 if (skb_try_make_writable(skb, sizeof(*iph) + ntkoff))
 379                         goto fail;
 380
 381                 ip_send_check(ip_hdr(skb));
 382         }
 383
 384         return 1;
 385
 386 fail:
 387         return 0;
 388 }
 389
 390 static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh,
 391                                  unsigned int ixhl, unsigned int *pl)
 392 {
 393         int off, len, optlen;
 394         unsigned char *xh = (void *)ip6xh;
 395
 396         off = sizeof(*ip6xh);
 397         len = ixhl - off;
 398
 399         while (len > 1) {
 400                 switch (xh[off]) {
 401                 case IPV6_TLV_PAD1:
 402                         optlen = 1;
 403                         break;
 404                 case IPV6_TLV_JUMBO:
 405                         optlen = xh[off + 1] + 2;
 406                         if (optlen != 6 || len < 6 || (off & 3) != 2)
 407                                 /* wrong jumbo option length/alignment */
 408                                 return 0;
 409                         *pl = ntohl(*(__be32 *)(xh + off + 2));
 410                         goto done;
 411                 default:
 412                         optlen = xh[off + 1] + 2;
 413                         if (optlen > len)
 414                                 /* ignore obscure options */
 415                                 goto done;
 416                         break;
 417                 }
 418                 off += optlen;
 419                 len -= optlen;
 420         }
 421
 422 done:
 423         return 1;
 424 }
 425
 426 static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
 427 {
 428         struct ipv6hdr *ip6h;
 429         struct ipv6_opt_hdr *ip6xh;
 430         unsigned int hl, ixhl;
 431         unsigned int pl;
 432         int ntkoff;
 433         u8 nexthdr;
 434
 435         ntkoff = skb_network_offset(skb);
 436
 437         hl = sizeof(*ip6h);
 438
 439         if (!pskb_may_pull(skb, hl + ntkoff))
 440                 goto fail;
 441
 442         ip6h = ipv6_hdr(skb);
 443
 444         pl = ntohs(ip6h->payload_len);
 445         nexthdr = ip6h->nexthdr;
 446
 447         do {
 448                 switch (nexthdr) {
 449                 case NEXTHDR_FRAGMENT:
 450                         goto ignore_skb;
 451                 case NEXTHDR_ROUTING:
 452                 case NEXTHDR_HOP:
 453                 case NEXTHDR_DEST:
 454                         if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
 455                                 goto fail;
 456                         ip6xh = (void *)(skb_network_header(skb) + hl);
 457                         ixhl = ipv6_optlen(ip6xh);
 458                         if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
 459                                 goto fail;
 460                         ip6xh = (void *)(skb_network_header(skb) + hl);
 461                         if ((nexthdr == NEXTHDR_HOP) &&
 462                             !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
 463                                 goto fail;
 464                         nexthdr = ip6xh->nexthdr;
 465                         hl += ixhl;
 466                         break;
 467                 case IPPROTO_ICMPV6:
 468                         if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
 469                                 if (!tcf_csum_ipv6_icmp(skb,
 470                                                         hl, pl + sizeof(*ip6h)))
 471                                         goto fail;
 472                         goto done;
 473                 case IPPROTO_TCP:
 474                         if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
 475                                 if (!tcf_csum_ipv6_tcp(skb,
 476                                                        hl, pl + sizeof(*ip6h)))
 477                                         goto fail;
 478                         goto done;
 479                 case IPPROTO_UDP:
 480                         if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
 481                                 if (!tcf_csum_ipv6_udp(skb, hl,
 482                                                        pl + sizeof(*ip6h), 0))
 483                                         goto fail;
 484                         goto done;
 485                 case IPPROTO_UDPLITE:
 486                         if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
 487                                 if (!tcf_csum_ipv6_udp(skb, hl,
 488                                                        pl + sizeof(*ip6h), 1))
 489                                         goto fail;
 490                         goto done;
 491                 default:
 492                         goto ignore_skb;
 493                 }
 494         } while (pskb_may_pull(skb, hl + 1 + ntkoff));
 495
 496 done:
 497 ignore_skb:
 498         return 1;
 499
 500 fail:
 501         return 0;
 502 }
 503
 504 static int tcf_csum(struct sk_buff *skb,
 505                     const struct tc_action *a, struct tcf_result *res)
 506 {
 507         struct tcf_csum *p = a->priv;
 508         int action;
 509         u32 update_flags;
 510
 511         spin_lock(&p->tcf_lock);
 512         p->tcf_tm.lastuse = jiffies;
 513         bstats_update(&p->tcf_bstats, skb);
 514         action = p->tcf_action;
 515         update_flags = p->update_flags;
 516         spin_unlock(&p->tcf_lock);
 517
 518         if (unlikely(action == TC_ACT_SHOT))
 519                 goto drop;
 520
 521         switch (tc_skb_protocol(skb)) {
 522         case cpu_to_be16(ETH_P_IP):
 523                 if (!tcf_csum_ipv4(skb, update_flags))
 524                         goto drop;
 525                 break;
 526         case cpu_to_be16(ETH_P_IPV6):
 527                 if (!tcf_csum_ipv6(skb, update_flags))
 528                         goto drop;
 529                 break;
 530         }
 531
 532         return action;
 533
 534 drop:
 535         spin_lock(&p->tcf_lock);
 536         p->tcf_qstats.drops++;
 537         spin_unlock(&p->tcf_lock);
 538         return TC_ACT_SHOT;
 539 }
 540
 541 static int tcf_csum_dump(struct sk_buff *skb,
 542                          struct tc_action *a, int bind, int ref)
 543 {
 544         unsigned char *b = skb_tail_pointer(skb);
 545         struct tcf_csum *p = a->priv;
 546         struct tc_csum opt = {
 547                 .update_flags = p->update_flags,
 548                 .index   = p->tcf_index,
 549                 .action  = p->tcf_action,
 550                 .refcnt  = p->tcf_refcnt - ref,
 551                 .bindcnt = p->tcf_bindcnt - bind,
 552         };
 553         struct tcf_t t;
 554
 555         if (nla_put(skb, TCA_CSUM_PARMS, sizeof(opt), &opt))
 556                 goto nla_put_failure;
 557         t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
 558         t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
 559         t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
 560         if (nla_put(skb, TCA_CSUM_TM, sizeof(t), &t))
 561                 goto nla_put_failure;
 562
 563         return skb->len;
 564
 565 nla_put_failure:
 566         nlmsg_trim(skb, b);
 567         return -1;
 568 }
 569
 570 static struct tc_action_ops act_csum_ops = {
 571         .kind           = "csum",
 572         .type           = TCA_ACT_CSUM,
 573         .owner          = THIS_MODULE,
 574         .act            = tcf_csum,
 575         .dump           = tcf_csum_dump,
 576         .init           = tcf_csum_init,
 577 };
 578
 579 MODULE_DESCRIPTION("Checksum updating actions");
 580 MODULE_LICENSE("GPL");
 581
 582 static int __init csum_init_module(void)
 583 {
 584         return tcf_register_action(&act_csum_ops, CSUM_TAB_MASK);
 585 }
 586
 587 static void __exit csum_cleanup_module(void)
 588 {
 589         tcf_unregister_action(&act_csum_ops);
 590 }
 591
 592 module_init(csum_init_module);
 593 module_exit(csum_cleanup_module);