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