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         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 179         if (tcph == NULL)
 180                 return 0;
 181
 182         iph = ip_hdr(skb);
 183         tcph->check = 0;
 184         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 185         tcph->check = tcp_v4_check(ipl - ihl,
 186                                    iph->saddr, iph->daddr, skb->csum);
 187
 188         skb->ip_summed = CHECKSUM_NONE;
 189
 190         return 1;
 191 }
 192
 193 static int tcf_csum_ipv6_tcp(struct sk_buff *skb,
 194                              unsigned int ihl, unsigned int ipl)
 195 {
 196         struct tcphdr *tcph;
 197         const struct ipv6hdr *ip6h;
 198
 199         tcph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*tcph));
 200         if (tcph == NULL)
 201                 return 0;
 202
 203         ip6h = ipv6_hdr(skb);
 204         tcph->check = 0;
 205         skb->csum = csum_partial(tcph, ipl - ihl, 0);
 206         tcph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 207                                       ipl - ihl, IPPROTO_TCP,
 208                                       skb->csum);
 209
 210         skb->ip_summed = CHECKSUM_NONE;
 211
 212         return 1;
 213 }
 214
 215 static int tcf_csum_ipv4_udp(struct sk_buff *skb,
 216                              unsigned int ihl, unsigned int ipl, int udplite)
 217 {
 218         struct udphdr *udph;
 219         const struct iphdr *iph;
 220         u16 ul;
 221
 222         /*
 223          * Support both UDP and UDPLITE checksum algorithms, Don't use
 224          * udph->len to get the real length without any protocol check,
 225          * UDPLITE uses udph->len for another thing,
 226          * Use iph->tot_len, or just ipl.
 227          */
 228
 229         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
 230         if (udph == NULL)
 231                 return 0;
 232
 233         iph = ip_hdr(skb);
 234         ul = ntohs(udph->len);
 235
 236         if (udplite || udph->check) {
 237
 238                 udph->check = 0;
 239
 240                 if (udplite) {
 241                         if (ul == 0)
 242                                 skb->csum = csum_partial(udph, ipl - ihl, 0);
 243                         else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 244                                 skb->csum = csum_partial(udph, ul, 0);
 245                         else
 246                                 goto ignore_obscure_skb;
 247                 } else {
 248                         if (ul != ipl - ihl)
 249                                 goto ignore_obscure_skb;
 250
 251                         skb->csum = csum_partial(udph, ul, 0);
 252                 }
 253
 254                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 255                                                 ul, iph->protocol,
 256                                                 skb->csum);
 257
 258                 if (!udph->check)
 259                         udph->check = CSUM_MANGLED_0;
 260         }
 261
 262         skb->ip_summed = CHECKSUM_NONE;
 263
 264 ignore_obscure_skb:
 265         return 1;
 266 }
 267
 268 static int tcf_csum_ipv6_udp(struct sk_buff *skb,
 269                              unsigned int ihl, unsigned int ipl, int udplite)
 270 {
 271         struct udphdr *udph;
 272         const struct ipv6hdr *ip6h;
 273         u16 ul;
 274
 275         /*
 276          * Support both UDP and UDPLITE checksum algorithms, Don't use
 277          * udph->len to get the real length without any protocol check,
 278          * UDPLITE uses udph->len for another thing,
 279          * Use ip6h->payload_len + sizeof(*ip6h) ... , or just ipl.
 280          */
 281
 282         udph = tcf_csum_skb_nextlayer(skb, ihl, ipl, sizeof(*udph));
 283         if (udph == NULL)
 284                 return 0;
 285
 286         ip6h = ipv6_hdr(skb);
 287         ul = ntohs(udph->len);
 288
 289         udph->check = 0;
 290
 291         if (udplite) {
 292                 if (ul == 0)
 293                         skb->csum = csum_partial(udph, ipl - ihl, 0);
 294
 295                 else if ((ul >= sizeof(*udph)) && (ul <= ipl - ihl))
 296                         skb->csum = csum_partial(udph, ul, 0);
 297
 298                 else
 299                         goto ignore_obscure_skb;
 300         } else {
 301                 if (ul != ipl - ihl)
 302                         goto ignore_obscure_skb;
 303
 304                 skb->csum = csum_partial(udph, ul, 0);
 305         }
 306
 307         udph->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr, ul,
 308                                       udplite ? IPPROTO_UDPLITE : IPPROTO_UDP,
 309                                       skb->csum);
 310
 311         if (!udph->check)
 312                 udph->check = CSUM_MANGLED_0;
 313
 314         skb->ip_summed = CHECKSUM_NONE;
 315
 316 ignore_obscure_skb:
 317         return 1;
 318 }
 319
 320 static int tcf_csum_ipv4(struct sk_buff *skb, u32 update_flags)
 321 {
 322         const struct iphdr *iph;
 323         int ntkoff;
 324
 325         ntkoff = skb_network_offset(skb);
 326
 327         if (!pskb_may_pull(skb, sizeof(*iph) + ntkoff))
 328                 goto fail;
 329
 330         iph = ip_hdr(skb);
 331
 332         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 333         case IPPROTO_ICMP:
 334                 if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
 335                         if (!tcf_csum_ipv4_icmp(skb, iph->ihl * 4,
 336                                                 ntohs(iph->tot_len)))
 337                                 goto fail;
 338                 break;
 339         case IPPROTO_IGMP:
 340                 if (update_flags & TCA_CSUM_UPDATE_FLAG_IGMP)
 341                         if (!tcf_csum_ipv4_igmp(skb, iph->ihl * 4,
 342                                                 ntohs(iph->tot_len)))
 343                                 goto fail;
 344                 break;
 345         case IPPROTO_TCP:
 346                 if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
 347                         if (!tcf_csum_ipv4_tcp(skb, iph->ihl * 4,
 348                                                ntohs(iph->tot_len)))
 349                                 goto fail;
 350                 break;
 351         case IPPROTO_UDP:
 352                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
 353                         if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
 354                                                ntohs(iph->tot_len), 0))
 355                                 goto fail;
 356                 break;
 357         case IPPROTO_UDPLITE:
 358                 if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
 359                         if (!tcf_csum_ipv4_udp(skb, iph->ihl * 4,
 360                                                ntohs(iph->tot_len), 1))
 361                                 goto fail;
 362                 break;
 363         }
 364
 365         if (update_flags & TCA_CSUM_UPDATE_FLAG_IPV4HDR) {
 366                 if (skb_try_make_writable(skb, sizeof(*iph) + ntkoff))
 367                         goto fail;
 368
 369                 ip_send_check(ip_hdr(skb));
 370         }
 371
 372         return 1;
 373
 374 fail:
 375         return 0;
 376 }
 377
 378 static int tcf_csum_ipv6_hopopts(struct ipv6_opt_hdr *ip6xh,
 379                                  unsigned int ixhl, unsigned int *pl)
 380 {
 381         int off, len, optlen;
 382         unsigned char *xh = (void *)ip6xh;
 383
 384         off = sizeof(*ip6xh);
 385         len = ixhl - off;
 386
 387         while (len > 1) {
 388                 switch (xh[off]) {
 389                 case IPV6_TLV_PAD1:
 390                         optlen = 1;
 391                         break;
 392                 case IPV6_TLV_JUMBO:
 393                         optlen = xh[off + 1] + 2;
 394                         if (optlen != 6 || len < 6 || (off & 3) != 2)
 395                                 /* wrong jumbo option length/alignment */
 396                                 return 0;
 397                         *pl = ntohl(*(__be32 *)(xh + off + 2));
 398                         goto done;
 399                 default:
 400                         optlen = xh[off + 1] + 2;
 401                         if (optlen > len)
 402                                 /* ignore obscure options */
 403                                 goto done;
 404                         break;
 405                 }
 406                 off += optlen;
 407                 len -= optlen;
 408         }
 409
 410 done:
 411         return 1;
 412 }
 413
 414 static int tcf_csum_ipv6(struct sk_buff *skb, u32 update_flags)
 415 {
 416         struct ipv6hdr *ip6h;
 417         struct ipv6_opt_hdr *ip6xh;
 418         unsigned int hl, ixhl;
 419         unsigned int pl;
 420         int ntkoff;
 421         u8 nexthdr;
 422
 423         ntkoff = skb_network_offset(skb);
 424
 425         hl = sizeof(*ip6h);
 426
 427         if (!pskb_may_pull(skb, hl + ntkoff))
 428                 goto fail;
 429
 430         ip6h = ipv6_hdr(skb);
 431
 432         pl = ntohs(ip6h->payload_len);
 433         nexthdr = ip6h->nexthdr;
 434
 435         do {
 436                 switch (nexthdr) {
 437                 case NEXTHDR_FRAGMENT:
 438                         goto ignore_skb;
 439                 case NEXTHDR_ROUTING:
 440                 case NEXTHDR_HOP:
 441                 case NEXTHDR_DEST:
 442                         if (!pskb_may_pull(skb, hl + sizeof(*ip6xh) + ntkoff))
 443                                 goto fail;
 444                         ip6xh = (void *)(skb_network_header(skb) + hl);
 445                         ixhl = ipv6_optlen(ip6xh);
 446                         if (!pskb_may_pull(skb, hl + ixhl + ntkoff))
 447                                 goto fail;
 448                         ip6xh = (void *)(skb_network_header(skb) + hl);
 449                         if ((nexthdr == NEXTHDR_HOP) &&
 450                             !(tcf_csum_ipv6_hopopts(ip6xh, ixhl, &pl)))
 451                                 goto fail;
 452                         nexthdr = ip6xh->nexthdr;
 453                         hl += ixhl;
 454                         break;
 455                 case IPPROTO_ICMPV6:
 456                         if (update_flags & TCA_CSUM_UPDATE_FLAG_ICMP)
 457                                 if (!tcf_csum_ipv6_icmp(skb,
 458                                                         hl, pl + sizeof(*ip6h)))
 459                                         goto fail;
 460                         goto done;
 461                 case IPPROTO_TCP:
 462                         if (update_flags & TCA_CSUM_UPDATE_FLAG_TCP)
 463                                 if (!tcf_csum_ipv6_tcp(skb,
 464                                                        hl, pl + sizeof(*ip6h)))
 465                                         goto fail;
 466                         goto done;
 467                 case IPPROTO_UDP:
 468                         if (update_flags & TCA_CSUM_UPDATE_FLAG_UDP)
 469                                 if (!tcf_csum_ipv6_udp(skb, hl,
 470                                                        pl + sizeof(*ip6h), 0))
 471                                         goto fail;
 472                         goto done;
 473                 case IPPROTO_UDPLITE:
 474                         if (update_flags & TCA_CSUM_UPDATE_FLAG_UDPLITE)
 475                                 if (!tcf_csum_ipv6_udp(skb, hl,
 476                                                        pl + sizeof(*ip6h), 1))
 477                                         goto fail;
 478                         goto done;
 479                 default:
 480                         goto ignore_skb;
 481                 }
 482         } while (pskb_may_pull(skb, hl + 1 + ntkoff));
 483
 484 done:
 485 ignore_skb:
 486         return 1;
 487
 488 fail:
 489         return 0;
 490 }
 491
 492 static int tcf_csum(struct sk_buff *skb,
 493                     const struct tc_action *a, struct tcf_result *res)
 494 {
 495         struct tcf_csum *p = a->priv;
 496         int action;
 497         u32 update_flags;
 498
 499         spin_lock(&p->tcf_lock);
 500         p->tcf_tm.lastuse = jiffies;
 501         bstats_update(&p->tcf_bstats, skb);
 502         action = p->tcf_action;
 503         update_flags = p->update_flags;
 504         spin_unlock(&p->tcf_lock);
 505
 506         if (unlikely(action == TC_ACT_SHOT))
 507                 goto drop;
 508
 509         switch (tc_skb_protocol(skb)) {
 510         case cpu_to_be16(ETH_P_IP):
 511                 if (!tcf_csum_ipv4(skb, update_flags))
 512                         goto drop;
 513                 break;
 514         case cpu_to_be16(ETH_P_IPV6):
 515                 if (!tcf_csum_ipv6(skb, update_flags))
 516                         goto drop;
 517                 break;
 518         }
 519
 520         return action;
 521
 522 drop:
 523         spin_lock(&p->tcf_lock);
 524         p->tcf_qstats.drops++;
 525         spin_unlock(&p->tcf_lock);
 526         return TC_ACT_SHOT;
 527 }
 528
 529 static int tcf_csum_dump(struct sk_buff *skb,
 530                          struct tc_action *a, int bind, int ref)
 531 {
 532         unsigned char *b = skb_tail_pointer(skb);
 533         struct tcf_csum *p = a->priv;
 534         struct tc_csum opt = {
 535                 .update_flags = p->update_flags,
 536                 .index   = p->tcf_index,
 537                 .action  = p->tcf_action,
 538                 .refcnt  = p->tcf_refcnt - ref,
 539                 .bindcnt = p->tcf_bindcnt - bind,
 540         };
 541         struct tcf_t t;
 542
 543         if (nla_put(skb, TCA_CSUM_PARMS, sizeof(opt), &opt))
 544                 goto nla_put_failure;
 545         t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
 546         t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
 547         t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
 548         if (nla_put(skb, TCA_CSUM_TM, sizeof(t), &t))
 549                 goto nla_put_failure;
 550
 551         return skb->len;
 552
 553 nla_put_failure:
 554         nlmsg_trim(skb, b);
 555         return -1;
 556 }
 557
 558 static struct tc_action_ops act_csum_ops = {
 559         .kind           = "csum",
 560         .type           = TCA_ACT_CSUM,
 561         .owner          = THIS_MODULE,
 562         .act            = tcf_csum,
 563         .dump           = tcf_csum_dump,
 564         .init           = tcf_csum_init,
 565 };
 566
 567 MODULE_DESCRIPTION("Checksum updating actions");
 568 MODULE_LICENSE("GPL");
 569
 570 static int __init csum_init_module(void)
 571 {
 572         return tcf_register_action(&act_csum_ops, CSUM_TAB_MASK);
 573 }
 574
 575 static void __exit csum_cleanup_module(void)
 576 {
 577         tcf_unregister_action(&act_csum_ops);
 578 }
 579
 580 module_init(csum_init_module);
 581 module_exit(csum_cleanup_module);