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);
  67         if (!pc) {
  68                 pc = tcf_hash_create(parm->index, est, a, sizeof(*p), bind);
  69                 if (IS_ERR(pc))
  70                         return PTR_ERR(pc);
  71                 ret = ACT_P_CREATED;
  72         } else {
  73                 if (bind)/* dont override defaults */
  74                         return 0;
  75                 tcf_hash_release(pc, bind, a->ops->hinfo);
  76                 if (!ovr)
  77                         return -EEXIST;
  78         }
  79
  80         p = to_tcf_csum(pc);
  81         spin_lock_bh(&p->tcf_lock);
  82         p->tcf_action = parm->action;
  83         p->update_flags = parm->update_flags;
  84         spin_unlock_bh(&p->tcf_lock);
  85
  86         if (ret == ACT_P_CREATED)
  87                 tcf_hash_insert(pc, a->ops->hinfo);
  88
  89         return ret;
  90 }
  91
  92 static int tcf_csum_cleanup(struct tc_action *a, int bind)
  93 {
  94         struct tcf_csum *p = a->priv;
  95         return tcf_hash_release(&p->common, bind, &csum_hash_info);
  96 }
  97
  98 /**
  99  * tcf_csum_skb_nextlayer - Get next layer pointer
 100  * @skb: sk_buff to use
 101  * @ihl: previous summed headers length
 102  * @ipl: complete packet length
 103  * @jhl: next header length
 104  *
 105  * Check the expected next layer availability in the specified sk_buff.
 106  * Return the next layer pointer if pass, NULL otherwise.
 107  */
 108 static void *tcf_csum_skb_nextlayer(struct sk_buff *skb,
 109                                     unsigned int ihl, unsigned int ipl,
 110                                     unsigned int jhl)
 111 {
 112         int ntkoff = skb_network_offset(skb);
 113         int hl = ihl + jhl;
 114
 115         if (!pskb_may_pull(skb, ipl + ntkoff) || (ipl < hl) ||
 116             (skb_cloned(skb) &&
 117              !skb_clone_writable(skb, hl + ntkoff) &&
 118              pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
 119                 return NULL;
 120         else
 121                 return (void *)(skb_network_header(skb) + ihl);
 122 }
 123
 124 static int tcf_csum_ipv4_icmp(struct sk_buff *skb,
 125                               unsigned int ihl, unsigned int ipl)
 126 {
 127         struct icmphdr *icmph;
 128
 129         icmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmph));
 130         if (icmph == NULL)
 131                 return 0;
 132
 133         icmph->checksum = 0;
 134         skb->csum = csum_partial(icmph, ipl - ihl, 0);
 135         icmph->checksum = csum_fold(skb->csum);
 136
 137         skb->ip_summed = CHECKSUM_NONE;
 138
 139         return 1;
 140 }
 141
 142 static int tcf_csum_ipv4_igmp(struct sk_buff *skb,
 143                               unsigned int ihl, unsigned int ipl)
 144 {
 145         struct igmphdr *igmph;
 146
 147         igmph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*igmph));
 148         if (igmph == NULL)
 149                 return 0;
 150
 151         igmph->csum = 0;
 152         skb->csum = csum_partial(igmph, ipl - ihl, 0);
 153         igmph->csum = csum_fold(skb->csum);
 154
 155         skb->ip_summed = CHECKSUM_NONE;
 156
 157         return 1;
 158 }
 159
 160 static int tcf_csum_ipv6_icmp(struct sk_buff *skb,
 161                               unsigned int ihl, unsigned int ipl)
 162 {
 163         struct icmp6hdr *icmp6h;
 164         const struct ipv6hdr *ip6h;
 165
 166         icmp6h = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*icmp6h));
 167         if (icmp6h == NULL)
 168                 return 0;
 169
 170         ip6h = ipv6_hdr(skb);
 171         icmp6h->icmp6_cksum = 0;
 172         skb->csum = csum_partial(icmp6h, ipl - ihl, 0);
 173         icmp6h->icmp6_cksum = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 174                                               ipl - ihl, IPPROTO_ICMPV6,
 175                                               skb->csum);
 176
 177         skb->ip_summed = CHECKSUM_NONE;
 178
 179         return 1;
 180 }
 181
 182 static int tcf_csum_ipv4_tcp(struct sk_buff *skb,
 183                              unsigned int ihl, unsigned int ipl)
 184 {
 185         struct tcphdr *tcph;
 186         const struct iphdr *iph;
 187
 188         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 189         if (tcph == NULL)
 190                 return 0;
 191
 192         iph = ip_hdr(skb);
 193         tcph->check = 0;
 194         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 195         tcph->check = tcp_v4_check(ipl - ihl,
 196                                    iph->saddr, iph->daddr, skb->csum);
 197
 198         skb->ip_summed = CHECKSUM_NONE;
 199
 200         return 1;
 201 }
 202
 203 static int tcf_csum_ipv6_tcp(struct sk_buff *skb,
 204                              unsigned int ihl, unsigned int ipl)
 205 {
 206         struct tcphdr *tcph;
 207         const struct ipv6hdr *ip6h;
 208
 209         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 210         if (tcph == NULL)
 211                 return 0;
 212
 213         ip6h = ipv6_hdr(skb);
 214         tcph->check = 0;
 215         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 216         tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 217                                       ipl - ihl, IPPROTO_TCP,
 218                                       skb->csum);
 219
 220         skb->ip_summed = CHECKSUM_NONE;
 221
 222         return 1;
 223 }
 224
 225 static int tcf_csum_ipv4_udp(struct sk_buff *skb,
 226                              unsigned int ihl, unsigned int ipl, int udplite)
 227 {
 228         struct udphdr *udph;
 229         const struct iphdr *iph;
 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         iph = ip_hdr(skb);
 244         ul = ntohs(udph->len);
 245
 246         if (udplite || udph->check) {
 247
 248                 udph->check = 0;
 249
 250                 if (udplite) {
 251                         if (ul == 0)
 252                                 skb->csum = csum_partial(udph, ipl - ihl, 0);
 253                         else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 254                                 skb->csum = csum_partial(udph, ul, 0);
 255                         else
 256                                 goto ignore_obscure_skb;
 257                 } else {
 258                         if (ul != ipl - ihl)
 259                                 goto ignore_obscure_skb;
 260
 261                         skb->csum = csum_partial(udph, ul, 0);
 262                 }
 263
 264                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 265                                                 ul, iph->protocol,
 266                                                 skb->csum);
 267
 268                 if (!udph->check)
 269                         udph->check = CSUM_MANGLED_0;
 270         }
 271
 272         skb->ip_summed = CHECKSUM_NONE;
 273
 274 ignore_obscure_skb:
 275         return 1;
 276 }
 277
 278 static int tcf_csum_ipv6_udp(struct sk_buff *skb,
 279                              unsigned int ihl, unsigned int ipl, int udplite)
 280 {
 281         struct udphdr *udph;
 282         const struct ipv6hdr *ip6h;
 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         ip6h = ipv6_hdr(skb);
 297         ul = ntohs(udph->len);
 298
 299         udph->check = 0;
 300
 301         if (udplite) {
 302                 if (ul == 0)
 303                         skb->csum = csum_partial(udph, ipl - ihl, 0);
 304
 305                 else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 306                         skb->csum = csum_partial(udph, ul, 0);
 307
 308                 else
 309                         goto ignore_obscure_skb;
 310         } else {
 311                 if (ul != ipl - ihl)
 312                         goto ignore_obscure_skb;
 313
 314                 skb->csum = csum_partial(udph, ul, 0);
 315         }
 316
 317         udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
 318                                       udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
 319                                       skb->csum);
 320
 321         if (!udph->check)
 322                 udph->check = CSUM_MANGLED_0;
 323
 324         skb->ip_summed = CHECKSUM_NONE;
 325
 326 ignore_obscure_skb:
 327         return 1;
 328 }
 329
 330 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
 331 {
 332         const struct iphdr *iph;
 333         int ntkoff;
 334
 335         ntkoff = skb_network_offset(skb);
 336
 337         if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
 338                 goto fail;
 339
 340         iph = ip_hdr(skb);
 341
 342         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 343         case IPPROTO_ICMP:
 344                 if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
 345                         if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
 346                                                 ntohs(iph->tot_len)))
 347                                 goto fail;
 348                 break;
 349         case IPPROTO_IGMP:
 350                 if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
 351                         if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
 352                                                 ntohs(iph->tot_len)))
 353                                 goto fail;
 354                 break;
 355         case IPPROTO_TCP:
 356                 if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
 357                         if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
 358                                                ntohs(iph->tot_len)))
 359                                 goto fail;
 360                 break;
 361         case IPPROTO_UDP:
 362                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
 363                         if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
 364                                                ntohs(iph->tot_len), 0))
 365                                 goto fail;
 366                 break;
 367         case IPPROTO_UDPLITE:
 368                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
 369                         if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
 370                                                ntohs(iph->tot_len), 1))
 371                                 goto fail;
 372                 break;
 373         }
 374
 375         if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
 376                 if (skb_cloned(skb) &&
 377                     !skb_clone_writable(skb, sizeof(*iph) + ntkoff) &&
 378                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
 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 (skb->protocol) {
 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         .hinfo          = &csum_hash_info,
 573         .type           = TCA_ACT_CSUM,
 574         .owner          = THIS_MODULE,
 575         .act            = tcf_csum,
 576         .dump           = tcf_csum_dump,
 577         .cleanup        = tcf_csum_cleanup,
 578         .init           = tcf_csum_init,
 579 };
 580
 581 MODULE_DESCRIPTION("Checksum updating actions");
 582 MODULE_LICENSE("GPL");
 583
 584 static int __init csum_init_module(void)
 585 {
 586         int err = tcf_hashinfo_init(&csum_hash_info, CSUM_TAB_MASK);
 587         if (err)
 588                 return err;
 589
 590         return tcf_register_action(&act_csum_ops);
 591 }
 592
 593 static void __exit csum_cleanup_module(void)
 594 {
 595         tcf_unregister_action(&act_csum_ops);
 596 }
 597
 598 module_init(csum_init_module);
 599 module_exit(csum_cleanup_module);