net/sched/act_ife.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/ife.c      Inter-FE action based on ForCES WG InterFE LFB
   4  *
   5  *              Refer to:
   6  *              draft-ietf-forces-interfelfb-03
   7  *              and
   8  *              netdev01 paper:
   9  *              "Distributing Linux Traffic Control Classifier-Action
  10  *              Subsystem"
  11  *              Authors: Jamal Hadi Salim and Damascene M. Joachimpillai
  12  *
  13  * copyright Jamal Hadi Salim (2015)
  14 */
  15
  16 #include <linux/types.h>
  17 #include <linux/kernel.h>
  18 #include <linux/string.h>
  19 #include <linux/errno.h>
  20 #include <linux/skbuff.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/module.h>
  23 #include <linux/init.h>
  24 #include <net/net_namespace.h>
  25 #include <net/netlink.h>
  26 #include <net/pkt_sched.h>
  27 #include <net/pkt_cls.h>
  28 #include <uapi/linux/tc_act/tc_ife.h>
  29 #include <net/tc_act/tc_ife.h>
  30 #include <linux/etherdevice.h>
  31 #include <net/ife.h>
  32
  33 static unsigned int ife_net_id;
  34 static int max_metacnt = IFE_META_MAX + 1;
  35 static struct tc_action_ops act_ife_ops;
  36
  37 static const struct nla_policy ife_policy[TCA_IFE_MAX + 1] = {
  38         [TCA_IFE_PARMS] = { .len = sizeof(struct tc_ife)},
  39         [TCA_IFE_DMAC] = { .len = ETH_ALEN},
  40         [TCA_IFE_SMAC] = { .len = ETH_ALEN},
  41         [TCA_IFE_TYPE] = { .type = NLA_U16},
  42 };
  43
  44 int ife_encode_meta_u16(u16 metaval, void *skbdata, struct tcf_meta_info *mi)
  45 {
  46         u16 edata = 0;
  47
  48         if (mi->metaval)
  49                 edata = *(u16 *)mi->metaval;
  50         else if (metaval)
  51                 edata = metaval;
  52
  53         if (!edata) /* will not encode */
  54                 return 0;
  55
  56         edata = htons(edata);
  57         return ife_tlv_meta_encode(skbdata, mi->metaid, 2, &edata);
  58 }
  59 EXPORT_SYMBOL_GPL(ife_encode_meta_u16);
  60
  61 int ife_get_meta_u32(struct sk_buff *skb, struct tcf_meta_info *mi)
  62 {
  63         if (mi->metaval)
  64                 return nla_put_u32(skb, mi->metaid, *(u32 *)mi->metaval);
  65         else
  66                 return nla_put(skb, mi->metaid, 0, NULL);
  67 }
  68 EXPORT_SYMBOL_GPL(ife_get_meta_u32);
  69
  70 int ife_check_meta_u32(u32 metaval, struct tcf_meta_info *mi)
  71 {
  72         if (metaval || mi->metaval)
  73                 return 8; /* T+L+V == 2+2+4 */
  74
  75         return 0;
  76 }
  77 EXPORT_SYMBOL_GPL(ife_check_meta_u32);
  78
  79 int ife_check_meta_u16(u16 metaval, struct tcf_meta_info *mi)
  80 {
  81         if (metaval || mi->metaval)
  82                 return 8; /* T+L+(V) == 2+2+(2+2bytepad) */
  83
  84         return 0;
  85 }
  86 EXPORT_SYMBOL_GPL(ife_check_meta_u16);
  87
  88 int ife_encode_meta_u32(u32 metaval, void *skbdata, struct tcf_meta_info *mi)
  89 {
  90         u32 edata = metaval;
  91
  92         if (mi->metaval)
  93                 edata = *(u32 *)mi->metaval;
  94         else if (metaval)
  95                 edata = metaval;
  96
  97         if (!edata) /* will not encode */
  98                 return 0;
  99
 100         edata = htonl(edata);
 101         return ife_tlv_meta_encode(skbdata, mi->metaid, 4, &edata);
 102 }
 103 EXPORT_SYMBOL_GPL(ife_encode_meta_u32);
 104
 105 int ife_get_meta_u16(struct sk_buff *skb, struct tcf_meta_info *mi)
 106 {
 107         if (mi->metaval)
 108                 return nla_put_u16(skb, mi->metaid, *(u16 *)mi->metaval);
 109         else
 110                 return nla_put(skb, mi->metaid, 0, NULL);
 111 }
 112 EXPORT_SYMBOL_GPL(ife_get_meta_u16);
 113
 114 int ife_alloc_meta_u32(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
 115 {
 116         mi->metaval = kmemdup(metaval, sizeof(u32), gfp);
 117         if (!mi->metaval)
 118                 return -ENOMEM;
 119
 120         return 0;
 121 }
 122 EXPORT_SYMBOL_GPL(ife_alloc_meta_u32);
 123
 124 int ife_alloc_meta_u16(struct tcf_meta_info *mi, void *metaval, gfp_t gfp)
 125 {
 126         mi->metaval = kmemdup(metaval, sizeof(u16), gfp);
 127         if (!mi->metaval)
 128                 return -ENOMEM;
 129
 130         return 0;
 131 }
 132 EXPORT_SYMBOL_GPL(ife_alloc_meta_u16);
 133
 134 void ife_release_meta_gen(struct tcf_meta_info *mi)
 135 {
 136         kfree(mi->metaval);
 137 }
 138 EXPORT_SYMBOL_GPL(ife_release_meta_gen);
 139
 140 int ife_validate_meta_u32(void *val, int len)
 141 {
 142         if (len == sizeof(u32))
 143                 return 0;
 144
 145         return -EINVAL;
 146 }
 147 EXPORT_SYMBOL_GPL(ife_validate_meta_u32);
 148
 149 int ife_validate_meta_u16(void *val, int len)
 150 {
 151         /* length will not include padding */
 152         if (len == sizeof(u16))
 153                 return 0;
 154
 155         return -EINVAL;
 156 }
 157 EXPORT_SYMBOL_GPL(ife_validate_meta_u16);
 158
 159 static LIST_HEAD(ifeoplist);
 160 static DEFINE_RWLOCK(ife_mod_lock);
 161
 162 static struct tcf_meta_ops *find_ife_oplist(u16 metaid)
 163 {
 164         struct tcf_meta_ops *o;
 165
 166         read_lock(&ife_mod_lock);
 167         list_for_each_entry(o, &ifeoplist, list) {
 168                 if (o->metaid == metaid) {
 169                         if (!try_module_get(o->owner))
 170                                 o = NULL;
 171                         read_unlock(&ife_mod_lock);
 172                         return o;
 173                 }
 174         }
 175         read_unlock(&ife_mod_lock);
 176
 177         return NULL;
 178 }
 179
 180 int register_ife_op(struct tcf_meta_ops *mops)
 181 {
 182         struct tcf_meta_ops *m;
 183
 184         if (!mops->metaid || !mops->metatype || !mops->name ||
 185             !mops->check_presence || !mops->encode || !mops->decode ||
 186             !mops->get || !mops->alloc)
 187                 return -EINVAL;
 188
 189         write_lock(&ife_mod_lock);
 190
 191         list_for_each_entry(m, &ifeoplist, list) {
 192                 if (m->metaid == mops->metaid ||
 193                     (strcmp(mops->name, m->name) == 0)) {
 194                         write_unlock(&ife_mod_lock);
 195                         return -EEXIST;
 196                 }
 197         }
 198
 199         if (!mops->release)
 200                 mops->release = ife_release_meta_gen;
 201
 202         list_add_tail(&mops->list, &ifeoplist);
 203         write_unlock(&ife_mod_lock);
 204         return 0;
 205 }
 206 EXPORT_SYMBOL_GPL(unregister_ife_op);
 207
 208 int unregister_ife_op(struct tcf_meta_ops *mops)
 209 {
 210         struct tcf_meta_ops *m;
 211         int err = -ENOENT;
 212
 213         write_lock(&ife_mod_lock);
 214         list_for_each_entry(m, &ifeoplist, list) {
 215                 if (m->metaid == mops->metaid) {
 216                         list_del(&mops->list);
 217                         err = 0;
 218                         break;
 219                 }
 220         }
 221         write_unlock(&ife_mod_lock);
 222
 223         return err;
 224 }
 225 EXPORT_SYMBOL_GPL(register_ife_op);
 226
 227 static int ife_validate_metatype(struct tcf_meta_ops *ops, void *val, int len)
 228 {
 229         int ret = 0;
 230         /* XXX: unfortunately cant use nla_policy at this point
 231         * because a length of 0 is valid in the case of
 232         * "allow". "use" semantics do enforce for proper
 233         * length and i couldve use nla_policy but it makes it hard
 234         * to use it just for that..
 235         */
 236         if (ops->validate)
 237                 return ops->validate(val, len);
 238
 239         if (ops->metatype == NLA_U32)
 240                 ret = ife_validate_meta_u32(val, len);
 241         else if (ops->metatype == NLA_U16)
 242                 ret = ife_validate_meta_u16(val, len);
 243
 244         return ret;
 245 }
 246
 247 #ifdef CONFIG_MODULES
 248 static const char *ife_meta_id2name(u32 metaid)
 249 {
 250         switch (metaid) {
 251         case IFE_META_SKBMARK:
 252                 return "skbmark";
 253         case IFE_META_PRIO:
 254                 return "skbprio";
 255         case IFE_META_TCINDEX:
 256                 return "tcindex";
 257         default:
 258                 return "unknown";
 259         }
 260 }
 261 #endif
 262
 263 /* called when adding new meta information
 264 */
 265 static int load_metaops_and_vet(u32 metaid, void *val, int len, bool rtnl_held)
 266 {
 267         struct tcf_meta_ops *ops = find_ife_oplist(metaid);
 268         int ret = 0;
 269
 270         if (!ops) {
 271                 ret = -ENOENT;
 272 #ifdef CONFIG_MODULES
 273                 if (rtnl_held)
 274                         rtnl_unlock();
 275                 request_module("ife-meta-%s", ife_meta_id2name(metaid));
 276                 if (rtnl_held)
 277                         rtnl_lock();
 278                 ops = find_ife_oplist(metaid);
 279 #endif
 280         }
 281
 282         if (ops) {
 283                 ret = 0;
 284                 if (len)
 285                         ret = ife_validate_metatype(ops, val, len);
 286
 287                 module_put(ops->owner);
 288         }
 289
 290         return ret;
 291 }
 292
 293 /* called when adding new meta information
 294 */
 295 static int __add_metainfo(const struct tcf_meta_ops *ops,
 296                           struct tcf_ife_info *ife, u32 metaid, void *metaval,
 297                           int len, bool atomic, bool exists)
 298 {
 299         struct tcf_meta_info *mi = NULL;
 300         int ret = 0;
 301
 302         mi = kzalloc(sizeof(*mi), atomic ? GFP_ATOMIC : GFP_KERNEL);
 303         if (!mi)
 304                 return -ENOMEM;
 305
 306         mi->metaid = metaid;
 307         mi->ops = ops;
 308         if (len > 0) {
 309                 ret = ops->alloc(mi, metaval, atomic ? GFP_ATOMIC : GFP_KERNEL);
 310                 if (ret != 0) {
 311                         kfree(mi);
 312                         return ret;
 313                 }
 314         }
 315
 316         if (exists)
 317                 spin_lock_bh(&ife->tcf_lock);
 318         list_add_tail(&mi->metalist, &ife->metalist);
 319         if (exists)
 320                 spin_unlock_bh(&ife->tcf_lock);
 321
 322         return ret;
 323 }
 324
 325 static int add_metainfo_and_get_ops(const struct tcf_meta_ops *ops,
 326                                     struct tcf_ife_info *ife, u32 metaid,
 327                                     bool exists)
 328 {
 329         int ret;
 330
 331         if (!try_module_get(ops->owner))
 332                 return -ENOENT;
 333         ret = __add_metainfo(ops, ife, metaid, NULL, 0, true, exists);
 334         if (ret)
 335                 module_put(ops->owner);
 336         return ret;
 337 }
 338
 339 static int add_metainfo(struct tcf_ife_info *ife, u32 metaid, void *metaval,
 340                         int len, bool exists)
 341 {
 342         const struct tcf_meta_ops *ops = find_ife_oplist(metaid);
 343         int ret;
 344
 345         if (!ops)
 346                 return -ENOENT;
 347         ret = __add_metainfo(ops, ife, metaid, metaval, len, false, exists);
 348         if (ret)
 349                 /*put back what find_ife_oplist took */
 350                 module_put(ops->owner);
 351         return ret;
 352 }
 353
 354 static int use_all_metadata(struct tcf_ife_info *ife, bool exists)
 355 {
 356         struct tcf_meta_ops *o;
 357         int rc = 0;
 358         int installed = 0;
 359
 360         read_lock(&ife_mod_lock);
 361         list_for_each_entry(o, &ifeoplist, list) {
 362                 rc = add_metainfo_and_get_ops(o, ife, o->metaid, exists);
 363                 if (rc == 0)
 364                         installed += 1;
 365         }
 366         read_unlock(&ife_mod_lock);
 367
 368         if (installed)
 369                 return 0;
 370         else
 371                 return -EINVAL;
 372 }
 373
 374 static int dump_metalist(struct sk_buff *skb, struct tcf_ife_info *ife)
 375 {
 376         struct tcf_meta_info *e;
 377         struct nlattr *nest;
 378         unsigned char *b = skb_tail_pointer(skb);
 379         int total_encoded = 0;
 380
 381         /*can only happen on decode */
 382         if (list_empty(&ife->metalist))
 383                 return 0;
 384
 385         nest = nla_nest_start_noflag(skb, TCA_IFE_METALST);
 386         if (!nest)
 387                 goto out_nlmsg_trim;
 388
 389         list_for_each_entry(e, &ife->metalist, metalist) {
 390                 if (!e->ops->get(skb, e))
 391                         total_encoded += 1;
 392         }
 393
 394         if (!total_encoded)
 395                 goto out_nlmsg_trim;
 396
 397         nla_nest_end(skb, nest);
 398
 399         return 0;
 400
 401 out_nlmsg_trim:
 402         nlmsg_trim(skb, b);
 403         return -1;
 404 }
 405
 406 /* under ife->tcf_lock */
 407 static void _tcf_ife_cleanup(struct tc_action *a)
 408 {
 409         struct tcf_ife_info *ife = to_ife(a);
 410         struct tcf_meta_info *e, *n;
 411
 412         list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
 413                 list_del(&e->metalist);
 414                 if (e->metaval) {
 415                         if (e->ops->release)
 416                                 e->ops->release(e);
 417                         else
 418                                 kfree(e->metaval);
 419                 }
 420                 module_put(e->ops->owner);
 421                 kfree(e);
 422         }
 423 }
 424
 425 static void tcf_ife_cleanup(struct tc_action *a)
 426 {
 427         struct tcf_ife_info *ife = to_ife(a);
 428         struct tcf_ife_params *p;
 429
 430         spin_lock_bh(&ife->tcf_lock);
 431         _tcf_ife_cleanup(a);
 432         spin_unlock_bh(&ife->tcf_lock);
 433
 434         p = rcu_dereference_protected(ife->params, 1);
 435         if (p)
 436                 kfree_rcu(p, rcu);
 437 }
 438
 439 static int load_metalist(struct nlattr **tb, bool rtnl_held)
 440 {
 441         int i;
 442
 443         for (i = 1; i < max_metacnt; i++) {
 444                 if (tb[i]) {
 445                         void *val = nla_data(tb[i]);
 446                         int len = nla_len(tb[i]);
 447                         int rc;
 448
 449                         rc = load_metaops_and_vet(i, val, len, rtnl_held);
 450                         if (rc != 0)
 451                                 return rc;
 452                 }
 453         }
 454
 455         return 0;
 456 }
 457
 458 static int populate_metalist(struct tcf_ife_info *ife, struct nlattr **tb,
 459                              bool exists, bool rtnl_held)
 460 {
 461         int len = 0;
 462         int rc = 0;
 463         int i = 0;
 464         void *val;
 465
 466         for (i = 1; i < max_metacnt; i++) {
 467                 if (tb[i]) {
 468                         val = nla_data(tb[i]);
 469                         len = nla_len(tb[i]);
 470
 471                         rc = add_metainfo(ife, i, val, len, exists);
 472                         if (rc)
 473                                 return rc;
 474                 }
 475         }
 476
 477         return rc;
 478 }
 479
 480 static int tcf_ife_init(struct net *net, struct nlattr *nla,
 481                         struct nlattr *est, struct tc_action **a,
 482                         int ovr, int bind, bool rtnl_held,
 483                         struct tcf_proto *tp, struct netlink_ext_ack *extack)
 484 {
 485         struct tc_action_net *tn = net_generic(net, ife_net_id);
 486         struct nlattr *tb[TCA_IFE_MAX + 1];
 487         struct nlattr *tb2[IFE_META_MAX + 1];
 488         struct tcf_chain *goto_ch = NULL;
 489         struct tcf_ife_params *p;
 490         struct tcf_ife_info *ife;
 491         u16 ife_type = ETH_P_IFE;
 492         struct tc_ife *parm;
 493         u8 *daddr = NULL;
 494         u8 *saddr = NULL;
 495         bool exists = false;
 496         int ret = 0;
 497         u32 index;
 498         int err;
 499
 500         if (!nla) {
 501                 NL_SET_ERR_MSG_MOD(extack, "IFE requires attributes to be passed");
 502                 return -EINVAL;
 503         }
 504
 505         err = nla_parse_nested_deprecated(tb, TCA_IFE_MAX, nla, ife_policy,
 506                                           NULL);
 507         if (err < 0)
 508                 return err;
 509
 510         if (!tb[TCA_IFE_PARMS])
 511                 return -EINVAL;
 512
 513         parm = nla_data(tb[TCA_IFE_PARMS]);
 514
 515         /* IFE_DECODE is 0 and indicates the opposite of IFE_ENCODE because
 516          * they cannot run as the same time. Check on all other values which
 517          * are not supported right now.
 518          */
 519         if (parm->flags & ~IFE_ENCODE)
 520                 return -EINVAL;
 521
 522         p = kzalloc(sizeof(*p), GFP_KERNEL);
 523         if (!p)
 524                 return -ENOMEM;
 525
 526         if (tb[TCA_IFE_METALST]) {
 527                 err = nla_parse_nested_deprecated(tb2, IFE_META_MAX,
 528                                                   tb[TCA_IFE_METALST], NULL,
 529                                                   NULL);
 530                 if (err) {
 531                         kfree(p);
 532                         return err;
 533                 }
 534                 err = load_metalist(tb2, rtnl_held);
 535                 if (err) {
 536                         kfree(p);
 537                         return err;
 538                 }
 539         }
 540
 541         index = parm->index;
 542         err = tcf_idr_check_alloc(tn, &index, a, bind);
 543         if (err < 0) {
 544                 kfree(p);
 545                 return err;
 546         }
 547         exists = err;
 548         if (exists && bind) {
 549                 kfree(p);
 550                 return 0;
 551         }
 552
 553         if (!exists) {
 554                 ret = tcf_idr_create(tn, index, est, a, &act_ife_ops,
 555                                      bind, true);
 556                 if (ret) {
 557                         tcf_idr_cleanup(tn, index);
 558                         kfree(p);
 559                         return ret;
 560                 }
 561                 ret = ACT_P_CREATED;
 562         } else if (!ovr) {
 563                 tcf_idr_release(*a, bind);
 564                 kfree(p);
 565                 return -EEXIST;
 566         }
 567
 568         ife = to_ife(*a);
 569         if (ret == ACT_P_CREATED)
 570                 INIT_LIST_HEAD(&ife->metalist);
 571
 572         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
 573         if (err < 0)
 574                 goto release_idr;
 575
 576         p->flags = parm->flags;
 577
 578         if (parm->flags & IFE_ENCODE) {
 579                 if (tb[TCA_IFE_TYPE])
 580                         ife_type = nla_get_u16(tb[TCA_IFE_TYPE]);
 581                 if (tb[TCA_IFE_DMAC])
 582                         daddr = nla_data(tb[TCA_IFE_DMAC]);
 583                 if (tb[TCA_IFE_SMAC])
 584                         saddr = nla_data(tb[TCA_IFE_SMAC]);
 585         }
 586
 587         if (parm->flags & IFE_ENCODE) {
 588                 if (daddr)
 589                         ether_addr_copy(p->eth_dst, daddr);
 590                 else
 591                         eth_zero_addr(p->eth_dst);
 592
 593                 if (saddr)
 594                         ether_addr_copy(p->eth_src, saddr);
 595                 else
 596                         eth_zero_addr(p->eth_src);
 597
 598                 p->eth_type = ife_type;
 599         }
 600
 601         if (tb[TCA_IFE_METALST]) {
 602                 err = populate_metalist(ife, tb2, exists, rtnl_held);
 603                 if (err)
 604                         goto metadata_parse_err;
 605         } else {
 606                 /* if no passed metadata allow list or passed allow-all
 607                  * then here we process by adding as many supported metadatum
 608                  * as we can. You better have at least one else we are
 609                  * going to bail out
 610                  */
 611                 err = use_all_metadata(ife, exists);
 612                 if (err)
 613                         goto metadata_parse_err;
 614         }
 615
 616         if (exists)
 617                 spin_lock_bh(&ife->tcf_lock);
 618         /* protected by tcf_lock when modifying existing action */
 619         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
 620         rcu_swap_protected(ife->params, p, 1);
 621
 622         if (exists)
 623                 spin_unlock_bh(&ife->tcf_lock);
 624         if (goto_ch)
 625                 tcf_chain_put_by_act(goto_ch);
 626         if (p)
 627                 kfree_rcu(p, rcu);
 628
 629         return ret;
 630 metadata_parse_err:
 631         if (goto_ch)
 632                 tcf_chain_put_by_act(goto_ch);
 633 release_idr:
 634         kfree(p);
 635         tcf_idr_release(*a, bind);
 636         return err;
 637 }
 638
 639 static int tcf_ife_dump(struct sk_buff *skb, struct tc_action *a, int bind,
 640                         int ref)
 641 {
 642         unsigned char *b = skb_tail_pointer(skb);
 643         struct tcf_ife_info *ife = to_ife(a);
 644         struct tcf_ife_params *p;
 645         struct tc_ife opt = {
 646                 .index = ife->tcf_index,
 647                 .refcnt = refcount_read(&ife->tcf_refcnt) - ref,
 648                 .bindcnt = atomic_read(&ife->tcf_bindcnt) - bind,
 649         };
 650         struct tcf_t t;
 651
 652         spin_lock_bh(&ife->tcf_lock);
 653         opt.action = ife->tcf_action;
 654         p = rcu_dereference_protected(ife->params,
 655                                       lockdep_is_held(&ife->tcf_lock));
 656         opt.flags = p->flags;
 657
 658         if (nla_put(skb, TCA_IFE_PARMS, sizeof(opt), &opt))
 659                 goto nla_put_failure;
 660
 661         tcf_tm_dump(&t, &ife->tcf_tm);
 662         if (nla_put_64bit(skb, TCA_IFE_TM, sizeof(t), &t, TCA_IFE_PAD))
 663                 goto nla_put_failure;
 664
 665         if (!is_zero_ether_addr(p->eth_dst)) {
 666                 if (nla_put(skb, TCA_IFE_DMAC, ETH_ALEN, p->eth_dst))
 667                         goto nla_put_failure;
 668         }
 669
 670         if (!is_zero_ether_addr(p->eth_src)) {
 671                 if (nla_put(skb, TCA_IFE_SMAC, ETH_ALEN, p->eth_src))
 672                         goto nla_put_failure;
 673         }
 674
 675         if (nla_put(skb, TCA_IFE_TYPE, 2, &p->eth_type))
 676                 goto nla_put_failure;
 677
 678         if (dump_metalist(skb, ife)) {
 679                 /*ignore failure to dump metalist */
 680                 pr_info("Failed to dump metalist\n");
 681         }
 682
 683         spin_unlock_bh(&ife->tcf_lock);
 684         return skb->len;
 685
 686 nla_put_failure:
 687         spin_unlock_bh(&ife->tcf_lock);
 688         nlmsg_trim(skb, b);
 689         return -1;
 690 }
 691
 692 static int find_decode_metaid(struct sk_buff *skb, struct tcf_ife_info *ife,
 693                               u16 metaid, u16 mlen, void *mdata)
 694 {
 695         struct tcf_meta_info *e;
 696
 697         /* XXX: use hash to speed up */
 698         list_for_each_entry(e, &ife->metalist, metalist) {
 699                 if (metaid == e->metaid) {
 700                         if (e->ops) {
 701                                 /* We check for decode presence already */
 702                                 return e->ops->decode(skb, mdata, mlen);
 703                         }
 704                 }
 705         }
 706
 707         return -ENOENT;
 708 }
 709
 710 static int tcf_ife_decode(struct sk_buff *skb, const struct tc_action *a,
 711                           struct tcf_result *res)
 712 {
 713         struct tcf_ife_info *ife = to_ife(a);
 714         int action = ife->tcf_action;
 715         u8 *ifehdr_end;
 716         u8 *tlv_data;
 717         u16 metalen;
 718
 719         bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
 720         tcf_lastuse_update(&ife->tcf_tm);
 721
 722         if (skb_at_tc_ingress(skb))
 723                 skb_push(skb, skb->dev->hard_header_len);
 724
 725         tlv_data = ife_decode(skb, &metalen);
 726         if (unlikely(!tlv_data)) {
 727                 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
 728                 return TC_ACT_SHOT;
 729         }
 730
 731         ifehdr_end = tlv_data + metalen;
 732         for (; tlv_data < ifehdr_end; tlv_data = ife_tlv_meta_next(tlv_data)) {
 733                 u8 *curr_data;
 734                 u16 mtype;
 735                 u16 dlen;
 736
 737                 curr_data = ife_tlv_meta_decode(tlv_data, ifehdr_end, &mtype,
 738                                                 &dlen, NULL);
 739                 if (!curr_data) {
 740                         qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
 741                         return TC_ACT_SHOT;
 742                 }
 743
 744                 if (find_decode_metaid(skb, ife, mtype, dlen, curr_data)) {
 745                         /* abuse overlimits to count when we receive metadata
 746                          * but dont have an ops for it
 747                          */
 748                         pr_info_ratelimited("Unknown metaid %d dlen %d\n",
 749                                             mtype, dlen);
 750                         qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
 751                 }
 752         }
 753
 754         if (WARN_ON(tlv_data != ifehdr_end)) {
 755                 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
 756                 return TC_ACT_SHOT;
 757         }
 758
 759         skb->protocol = eth_type_trans(skb, skb->dev);
 760         skb_reset_network_header(skb);
 761
 762         return action;
 763 }
 764
 765 /*XXX: check if we can do this at install time instead of current
 766  * send data path
 767 **/
 768 static int ife_get_sz(struct sk_buff *skb, struct tcf_ife_info *ife)
 769 {
 770         struct tcf_meta_info *e, *n;
 771         int tot_run_sz = 0, run_sz = 0;
 772
 773         list_for_each_entry_safe(e, n, &ife->metalist, metalist) {
 774                 if (e->ops->check_presence) {
 775                         run_sz = e->ops->check_presence(skb, e);
 776                         tot_run_sz += run_sz;
 777                 }
 778         }
 779
 780         return tot_run_sz;
 781 }
 782
 783 static int tcf_ife_encode(struct sk_buff *skb, const struct tc_action *a,
 784                           struct tcf_result *res, struct tcf_ife_params *p)
 785 {
 786         struct tcf_ife_info *ife = to_ife(a);
 787         int action = ife->tcf_action;
 788         struct ethhdr *oethh;   /* outer ether header */
 789         struct tcf_meta_info *e;
 790         /*
 791            OUTERHDR:TOTMETALEN:{TLVHDR:Metadatum:TLVHDR..}:ORIGDATA
 792            where ORIGDATA = original ethernet header ...
 793          */
 794         u16 metalen = ife_get_sz(skb, ife);
 795         int hdrm = metalen + skb->dev->hard_header_len + IFE_METAHDRLEN;
 796         unsigned int skboff = 0;
 797         int new_len = skb->len + hdrm;
 798         bool exceed_mtu = false;
 799         void *ife_meta;
 800         int err = 0;
 801
 802         if (!skb_at_tc_ingress(skb)) {
 803                 if (new_len > skb->dev->mtu)
 804                         exceed_mtu = true;
 805         }
 806
 807         bstats_cpu_update(this_cpu_ptr(ife->common.cpu_bstats), skb);
 808         tcf_lastuse_update(&ife->tcf_tm);
 809
 810         if (!metalen) {         /* no metadata to send */
 811                 /* abuse overlimits to count when we allow packet
 812                  * with no metadata
 813                  */
 814                 qstats_overlimit_inc(this_cpu_ptr(ife->common.cpu_qstats));
 815                 return action;
 816         }
 817         /* could be stupid policy setup or mtu config
 818          * so lets be conservative.. */
 819         if ((action == TC_ACT_SHOT) || exceed_mtu) {
 820                 qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
 821                 return TC_ACT_SHOT;
 822         }
 823
 824         if (skb_at_tc_ingress(skb))
 825                 skb_push(skb, skb->dev->hard_header_len);
 826
 827         ife_meta = ife_encode(skb, metalen);
 828
 829         spin_lock(&ife->tcf_lock);
 830
 831         /* XXX: we dont have a clever way of telling encode to
 832          * not repeat some of the computations that are done by
 833          * ops->presence_check...
 834          */
 835         list_for_each_entry(e, &ife->metalist, metalist) {
 836                 if (e->ops->encode) {
 837                         err = e->ops->encode(skb, (void *)(ife_meta + skboff),
 838                                              e);
 839                 }
 840                 if (err < 0) {
 841                         /* too corrupt to keep around if overwritten */
 842                         spin_unlock(&ife->tcf_lock);
 843                         qstats_drop_inc(this_cpu_ptr(ife->common.cpu_qstats));
 844                         return TC_ACT_SHOT;
 845                 }
 846                 skboff += err;
 847         }
 848         spin_unlock(&ife->tcf_lock);
 849         oethh = (struct ethhdr *)skb->data;
 850
 851         if (!is_zero_ether_addr(p->eth_src))
 852                 ether_addr_copy(oethh->h_source, p->eth_src);
 853         if (!is_zero_ether_addr(p->eth_dst))
 854                 ether_addr_copy(oethh->h_dest, p->eth_dst);
 855         oethh->h_proto = htons(p->eth_type);
 856
 857         if (skb_at_tc_ingress(skb))
 858                 skb_pull(skb, skb->dev->hard_header_len);
 859
 860         return action;
 861 }
 862
 863 static int tcf_ife_act(struct sk_buff *skb, const struct tc_action *a,
 864                        struct tcf_result *res)
 865 {
 866         struct tcf_ife_info *ife = to_ife(a);
 867         struct tcf_ife_params *p;
 868         int ret;
 869
 870         p = rcu_dereference_bh(ife->params);
 871         if (p->flags & IFE_ENCODE) {
 872                 ret = tcf_ife_encode(skb, a, res, p);
 873                 return ret;
 874         }
 875
 876         return tcf_ife_decode(skb, a, res);
 877 }
 878
 879 static int tcf_ife_walker(struct net *net, struct sk_buff *skb,
 880                           struct netlink_callback *cb, int type,
 881                           const struct tc_action_ops *ops,
 882                           struct netlink_ext_ack *extack)
 883 {
 884         struct tc_action_net *tn = net_generic(net, ife_net_id);
 885
 886         return tcf_generic_walker(tn, skb, cb, type, ops, extack);
 887 }
 888
 889 static int tcf_ife_search(struct net *net, struct tc_action **a, u32 index)
 890 {
 891         struct tc_action_net *tn = net_generic(net, ife_net_id);
 892
 893         return tcf_idr_search(tn, a, index);
 894 }
 895
 896 static struct tc_action_ops act_ife_ops = {
 897         .kind = "ife",
 898         .id = TCA_ID_IFE,
 899         .owner = THIS_MODULE,
 900         .act = tcf_ife_act,
 901         .dump = tcf_ife_dump,
 902         .cleanup = tcf_ife_cleanup,
 903         .init = tcf_ife_init,
 904         .walk = tcf_ife_walker,
 905         .lookup = tcf_ife_search,
 906         .size = sizeof(struct tcf_ife_info),
 907 };
 908
 909 static __net_init int ife_init_net(struct net *net)
 910 {
 911         struct tc_action_net *tn = net_generic(net, ife_net_id);
 912
 913         return tc_action_net_init(net, tn, &act_ife_ops);
 914 }
 915
 916 static void __net_exit ife_exit_net(struct list_head *net_list)
 917 {
 918         tc_action_net_exit(net_list, ife_net_id);
 919 }
 920
 921 static struct pernet_operations ife_net_ops = {
 922         .init = ife_init_net,
 923         .exit_batch = ife_exit_net,
 924         .id   = &ife_net_id,
 925         .size = sizeof(struct tc_action_net),
 926 };
 927
 928 static int __init ife_init_module(void)
 929 {
 930         return tcf_register_action(&act_ife_ops, &ife_net_ops);
 931 }
 932
 933 static void __exit ife_cleanup_module(void)
 934 {
 935         tcf_unregister_action(&act_ife_ops, &ife_net_ops);
 936 }
 937
 938 module_init(ife_init_module);
 939 module_exit(ife_cleanup_module);
 940
 941 MODULE_AUTHOR("Jamal Hadi Salim(2015)");
 942 MODULE_DESCRIPTION("Inter-FE LFB action");
 943 MODULE_LICENSE("GPL");