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 csum_net_id;
  46
  47 static int tcf_csum_init(struct net *net, struct nlattr *nla,
  48                          struct nlattr *est, struct tc_action *a, int ovr,
  49                          int bind)
  50 {
  51         struct tc_action_net *tn = net_generic(net, csum_net_id);
  52         struct nlattr *tb[TCA_CSUM_MAX + 1];
  53         struct tc_csum *parm;
  54         struct tcf_csum *p;
  55         int ret = 0, err;
  56
  57         if (nla == NULL)
  58                 return -EINVAL;
  59
  60         err = nla_parse_nested(tb, TCA_CSUM_MAX, nla, csum_policy);
  61         if (err < 0)
  62                 return err;
  63
  64         if (tb[TCA_CSUM_PARMS] == NULL)
  65                 return -EINVAL;
  66         parm = nla_data(tb[TCA_CSUM_PARMS]);
  67
  68         if (!tcf_hash_check(tn, parm->index, a, bind)) {
  69                 ret = tcf_hash_create(tn, parm->index, est, a,
  70                                       sizeof(*p), bind, false);
  71                 if (ret)
  72                         return ret;
  73                 ret = ACT_P_CREATED;
  74         } else {
  75                 if (bind)/* dont override defaults */
  76                         return 0;
  77                 tcf_hash_release(a, bind);
  78                 if (!ovr)
  79                         return -EEXIST;
  80         }
  81
  82         p = to_tcf_csum(a);
  83         spin_lock_bh(&p->tcf_lock);
  84         p->tcf_action = parm->action;
  85         p->update_flags = parm->update_flags;
  86         spin_unlock_bh(&p->tcf_lock);
  87
  88         if (ret == ACT_P_CREATED)
  89                 tcf_hash_insert(tn, a);
  90
  91         return ret;
  92 }
  93
  94 /**
  95  * tcf_csum_skb_nextlayer - Get next layer pointer
  96  * @skb: sk_buff to use
  97  * @ihl: previous summed headers length
  98  * @ipl: complete packet length
  99  * @jhl: next header length
 100  *
 101  * Check the expected next layer availability in the specified sk_buff.
 102  * Return the next layer pointer if pass, NULL otherwise.
 103  */
 104 static void *tcf_csum_skb_nextlayer(struct sk_buff *skb,
 105                                     unsigned int ihl, unsigned int ipl,
 106                                     unsigned int jhl)
 107 {
 108         int ntkoff = skb_network_offset(skb);
 109         int hl = ihl + jhl;
 110
 111         if (!pskb_may_pull(skb, ipl + ntkoff) || (ipl < hl) ||
 112             skb_try_make_writable(skb, hl + ntkoff))
 113                 return NULL;
 114         else
 115                 return (void *)(skb_network_header(skb) + ihl);
 116 }
 117
 118 static int tcf_csum_ipv4_icmp(struct sk_buff *skb,
 119                               unsigned int ihl, unsigned int ipl)
 120 {
 121         struct icmphdr *icmph;
 122
 123         icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
 124         if (icmph == NULL)
 125                 return 0;
 126
 127         icmph->checksum = 0;
 128         skb->csum = csum_partial(icmph, ipl - ihl, 0);
 129         icmph->checksum = csum_fold(skb->csum);
 130
 131         skb->ip_summed = CHECKSUM_NONE;
 132
 133         return 1;
 134 }
 135
 136 static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
 137                               unsigned int ihl, unsigned int ipl)
 138 {
 139         struct igmphdr *igmph;
 140
 141         igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
 142         if (igmph == NULL)
 143                 return 0;
 144
 145         igmph->csum = 0;
 146         skb->csum = csum_partial(igmph, ipl - ihl, 0);
 147         igmph->csum = csum_fold(skb->csum);
 148
 149         skb->ip_summed = CHECKSUM_NONE;
 150
 151         return 1;
 152 }
 153
 154 static int tcf_csum_ipv6_icmp(struct sk_buff *skb,
 155                               unsigned int ihl, unsigned int ipl)
 156 {
 157         struct icmp6hdr *icmp6h;
 158         const struct ipv6hdr *ip6h;
 159
 160         icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
 161         if (icmp6h == NULL)
 162                 return 0;
 163
 164         ip6h = ipv6_hdr(skb);
 165         icmp6h->icmp6_cksum = 0;
 166         skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
 167         icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 168                                               ipl - ihl, IPPROTO_ICMPV6,
 169                                               skb->csum);
 170
 171         skb->ip_summed = CHECKSUM_NONE;
 172
 173         return 1;
 174 }
 175
 176 static int tcf_csum_ipv4_tcp(struct sk_buff *skb,
 177                              unsigned int ihl, unsigned int ipl)
 178 {
 179         struct tcphdr *tcph;
 180         const struct iphdr *iph;
 181
 182         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 183         if (tcph == NULL)
 184                 return 0;
 185
 186         iph = ip_hdr(skb);
 187         tcph->check = 0;
 188         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 189         tcph->check = tcp_v4_check(ipl - ihl,
 190                                    iph->saddr, iph->daddr, skb->csum);
 191
 192         skb->ip_summed = CHECKSUM_NONE;
 193
 194         return 1;
 195 }
 196
 197 static int tcf_csum_ipv6_tcp(struct sk_buff *skb,
 198                              unsigned int ihl, unsigned int ipl)
 199 {
 200         struct tcphdr *tcph;
 201         const struct ipv6hdr *ip6h;
 202
 203         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 204         if (tcph == NULL)
 205                 return 0;
 206
 207         ip6h = ipv6_hdr(skb);
 208         tcph->check = 0;
 209         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 210         tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 211                                       ipl - ihl, IPPROTO_TCP,
 212                                       skb->csum);
 213
 214         skb->ip_summed = CHECKSUM_NONE;
 215
 216         return 1;
 217 }
 218
 219 static int tcf_csum_ipv4_udp(struct sk_buff *skb,
 220                              unsigned int ihl, unsigned int ipl, int udplite)
 221 {
 222         struct udphdr *udph;
 223         const struct iphdr *iph;
 224         u16 ul;
 225
 226         /*
 227          * Support both UDP and UDPLITE checksum algorithms, Don't use
 228          * udph->len to get the real length without any protocol check,
 229          * UDPLITE uses udph->len for another thing,
 230          * Use iph->tot_len, or just ipl.
 231          */
 232
 233         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
 234         if (udph == NULL)
 235                 return 0;
 236
 237         iph = ip_hdr(skb);
 238         ul = ntohs(udph->len);
 239
 240         if (udplite || udph->check) {
 241
 242                 udph->check = 0;
 243
 244                 if (udplite) {
 245                         if (ul == 0)
 246                                 skb->csum = csum_partial(udph, ipl - ihl, 0);
 247                         else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 248                                 skb->csum = csum_partial(udph, ul, 0);
 249                         else
 250                                 goto ignore_obscure_skb;
 251                 } else {
 252                         if (ul != ipl - ihl)
 253                                 goto ignore_obscure_skb;
 254
 255                         skb->csum = csum_partial(udph, ul, 0);
 256                 }
 257
 258                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 259                                                 ul, iph->protocol,
 260                                                 skb->csum);
 261
 262                 if (!udph->check)
 263                         udph->check = CSUM_MANGLED_0;
 264         }
 265
 266         skb->ip_summed = CHECKSUM_NONE;
 267
 268 ignore_obscure_skb:
 269         return 1;
 270 }
 271
 272 static int tcf_csum_ipv6_udp(struct sk_buff *skb,
 273                              unsigned int ihl, unsigned int ipl, int udplite)
 274 {
 275         struct udphdr *udph;
 276         const struct ipv6hdr *ip6h;
 277         u16 ul;
 278
 279         /*
 280          * Support both UDP and UDPLITE checksum algorithms, Don't use
 281          * udph->len to get the real length without any protocol check,
 282          * UDPLITE uses udph->len for another thing,
 283          * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
 284          */
 285
 286         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
 287         if (udph == NULL)
 288                 return 0;
 289
 290         ip6h = ipv6_hdr(skb);
 291         ul = ntohs(udph->len);
 292
 293         udph->check = 0;
 294
 295         if (udplite) {
 296                 if (ul == 0)
 297                         skb->csum = csum_partial(udph, ipl - ihl, 0);
 298
 299                 else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 300                         skb->csum = csum_partial(udph, ul, 0);
 301
 302                 else
 303                         goto ignore_obscure_skb;
 304         } else {
 305                 if (ul != ipl - ihl)
 306                         goto ignore_obscure_skb;
 307
 308                 skb->csum = csum_partial(udph, ul, 0);
 309         }
 310
 311         udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
 312                                       udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
 313                                       skb->csum);
 314
 315         if (!udph->check)
 316                 udph->check = CSUM_MANGLED_0;
 317
 318         skb->ip_summed = CHECKSUM_NONE;
 319
 320 ignore_obscure_skb:
 321         return 1;
 322 }
 323
 324 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
 325 {
 326         const struct iphdr *iph;
 327         int ntkoff;
 328
 329         ntkoff = skb_network_offset(skb);
 330
 331         if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
 332                 goto fail;
 333
 334         iph = ip_hdr(skb);
 335
 336         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 337         case IPPROTO_ICMP:
 338                 if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
 339                         if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
 340                                                 ntohs(iph->tot_len)))
 341                                 goto fail;
 342                 break;
 343         case IPPROTO_IGMP:
 344                 if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
 345                         if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
 346                                                 ntohs(iph->tot_len)))
 347                                 goto fail;
 348                 break;
 349         case IPPROTO_TCP:
 350                 if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
 351                         if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
 352                                                ntohs(iph->tot_len)))
 353                                 goto fail;
 354                 break;
 355         case IPPROTO_UDP:
 356                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
 357                         if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
 358                                                ntohs(iph->tot_len), 0))
 359                                 goto fail;
 360                 break;
 361         case IPPROTO_UDPLITE:
 362                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
 363                         if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
 364                                                ntohs(iph->tot_len), 1))
 365                                 goto fail;
 366                 break;
 367         }
 368
 369         if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
 370                 if (skb_try_make_writable(skb, sizeof(*iph) + ntkoff))
 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 int tcf_csum_walker(struct net *net, struct sk_buff *skb,
 563                            struct netlink_callback *cb, int type,
 564                            struct tc_action *a)
 565 {
 566         struct tc_action_net *tn = net_generic(net, csum_net_id);
 567
 568         return tcf_generic_walker(tn, skb, cb, type, a);
 569 }
 570
 571 static int tcf_csum_search(struct net *net, struct tc_action *a, u32 index)
 572 {
 573         struct tc_action_net *tn = net_generic(net, csum_net_id);
 574
 575         return tcf_hash_search(tn, a, index);
 576 }
 577
 578 static struct tc_action_ops act_csum_ops = {
 579         .kind           = "csum",
 580         .type           = TCA_ACT_CSUM,
 581         .owner          = THIS_MODULE,
 582         .act            = tcf_csum,
 583         .dump           = tcf_csum_dump,
 584         .init           = tcf_csum_init,
 585         .walk           = tcf_csum_walker,
 586         .lookup         = tcf_csum_search,
 587 };
 588
 589 static __net_init int csum_init_net(struct net *net)
 590 {
 591         struct tc_action_net *tn = net_generic(net, csum_net_id);
 592
 593         return tc_action_net_init(tn, &act_csum_ops, CSUM_TAB_MASK);
 594 }
 595
 596 static void __net_exit csum_exit_net(struct net *net)
 597 {
 598         struct tc_action_net *tn = net_generic(net, csum_net_id);
 599
 600         tc_action_net_exit(tn);
 601 }
 602
 603 static struct pernet_operations csum_net_ops = {
 604         .init = csum_init_net,
 605         .exit = csum_exit_net,
 606         .id   = &csum_net_id,
 607         .size = sizeof(struct tc_action_net),
 608 };
 609
 610 MODULE_DESCRIPTION("Checksum updating actions");
 611 MODULE_LICENSE("GPL");
 612
 613 static int __init csum_init_module(void)
 614 {
 615         return tcf_register_action(&act_csum_ops, &csum_net_ops);
 616 }
 617
 618 static void __exit csum_cleanup_module(void)
 619 {
 620         tcf_unregister_action(&act_csum_ops, &csum_net_ops);
 621 }
 622
 623 module_init(csum_init_module);
 624 module_exit(csum_cleanup_module);