net/sched/cls_api.c

   1 /*
   2  * net/sched/cls_api.c  Packet classifier API.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *
  11  * Changes:
  12  *
  13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  14  *
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/init.h>
  25 #include <linux/kmod.h>
  26 #include <linux/slab.h>
  27 #include <linux/idr.h>
  28 #include <net/net_namespace.h>
  29 #include <net/sock.h>
  30 #include <net/netlink.h>
  31 #include <net/pkt_sched.h>
  32 #include <net/pkt_cls.h>
  33
  34 /* The list of all installed classifier types */
  35 static LIST_HEAD(tcf_proto_base);
  36
  37 /* Protects list of registered TC modules. It is pure SMP lock. */
  38 static DEFINE_RWLOCK(cls_mod_lock);
  39
  40 /* Find classifier type by string name */
  41
  42 static const struct tcf_proto_ops *tcf_proto_lookup_ops(const char *kind)
  43 {
  44         const struct tcf_proto_ops *t, *res = NULL;
  45
  46         if (kind) {
  47                 read_lock(&cls_mod_lock);
  48                 list_for_each_entry(t, &tcf_proto_base, head) {
  49                         if (strcmp(kind, t->kind) == 0) {
  50                                 if (try_module_get(t->owner))
  51                                         res = t;
  52                                 break;
  53                         }
  54                 }
  55                 read_unlock(&cls_mod_lock);
  56         }
  57         return res;
  58 }
  59
  60 /* Register(unregister) new classifier type */
  61
  62 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  63 {
  64         struct tcf_proto_ops *t;
  65         int rc = -EEXIST;
  66
  67         write_lock(&cls_mod_lock);
  68         list_for_each_entry(t, &tcf_proto_base, head)
  69                 if (!strcmp(ops->kind, t->kind))
  70                         goto out;
  71
  72         list_add_tail(&ops->head, &tcf_proto_base);
  73         rc = 0;
  74 out:
  75         write_unlock(&cls_mod_lock);
  76         return rc;
  77 }
  78 EXPORT_SYMBOL(register_tcf_proto_ops);
  79
  80 static struct workqueue_struct *tc_filter_wq;
  81
  82 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
  83 {
  84         struct tcf_proto_ops *t;
  85         int rc = -ENOENT;
  86
  87         /* Wait for outstanding call_rcu()s, if any, from a
  88          * tcf_proto_ops's destroy() handler.
  89          */
  90         rcu_barrier();
  91         flush_workqueue(tc_filter_wq);
  92
  93         write_lock(&cls_mod_lock);
  94         list_for_each_entry(t, &tcf_proto_base, head) {
  95                 if (t == ops) {
  96                         list_del(&t->head);
  97                         rc = 0;
  98                         break;
  99                 }
 100         }
 101         write_unlock(&cls_mod_lock);
 102         return rc;
 103 }
 104 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 105
 106 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 107 {
 108         INIT_RCU_WORK(rwork, func);
 109         return queue_rcu_work(tc_filter_wq, rwork);
 110 }
 111 EXPORT_SYMBOL(tcf_queue_work);
 112
 113 /* Select new prio value from the range, managed by kernel. */
 114
 115 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 116 {
 117         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 118
 119         if (tp)
 120                 first = tp->prio - 1;
 121
 122         return TC_H_MAJ(first);
 123 }
 124
 125 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 126                                           u32 prio, struct tcf_chain *chain,
 127                                           struct netlink_ext_ack *extack)
 128 {
 129         struct tcf_proto *tp;
 130         int err;
 131
 132         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 133         if (!tp)
 134                 return ERR_PTR(-ENOBUFS);
 135
 136         err = -ENOENT;
 137         tp->ops = tcf_proto_lookup_ops(kind);
 138         if (!tp->ops) {
 139 #ifdef CONFIG_MODULES
 140                 rtnl_unlock();
 141                 request_module("cls_%s", kind);
 142                 rtnl_lock();
 143                 tp->ops = tcf_proto_lookup_ops(kind);
 144                 /* We dropped the RTNL semaphore in order to perform
 145                  * the module load. So, even if we succeeded in loading
 146                  * the module we have to replay the request. We indicate
 147                  * this using -EAGAIN.
 148                  */
 149                 if (tp->ops) {
 150                         module_put(tp->ops->owner);
 151                         err = -EAGAIN;
 152                 } else {
 153                         NL_SET_ERR_MSG(extack, "TC classifier not found");
 154                         err = -ENOENT;
 155                 }
 156 #endif
 157                 goto errout;
 158         }
 159         tp->classify = tp->ops->classify;
 160         tp->protocol = protocol;
 161         tp->prio = prio;
 162         tp->chain = chain;
 163
 164         err = tp->ops->init(tp);
 165         if (err) {
 166                 module_put(tp->ops->owner);
 167                 goto errout;
 168         }
 169         return tp;
 170
 171 errout:
 172         kfree(tp);
 173         return ERR_PTR(err);
 174 }
 175
 176 static void tcf_proto_destroy(struct tcf_proto *tp,
 177                               struct netlink_ext_ack *extack)
 178 {
 179         tp->ops->destroy(tp, extack);
 180         module_put(tp->ops->owner);
 181         kfree_rcu(tp, rcu);
 182 }
 183
 184 struct tcf_filter_chain_list_item {
 185         struct list_head list;
 186         tcf_chain_head_change_t *chain_head_change;
 187         void *chain_head_change_priv;
 188 };
 189
 190 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 191                                           u32 chain_index)
 192 {
 193         struct tcf_chain *chain;
 194
 195         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 196         if (!chain)
 197                 return NULL;
 198         INIT_LIST_HEAD(&chain->filter_chain_list);
 199         list_add_tail(&chain->list, &block->chain_list);
 200         chain->block = block;
 201         chain->index = chain_index;
 202         chain->refcnt = 1;
 203         return chain;
 204 }
 205
 206 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 207                                        struct tcf_proto *tp_head)
 208 {
 209         if (item->chain_head_change)
 210                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 211 }
 212 static void tcf_chain_head_change(struct tcf_chain *chain,
 213                                   struct tcf_proto *tp_head)
 214 {
 215         struct tcf_filter_chain_list_item *item;
 216
 217         list_for_each_entry(item, &chain->filter_chain_list, list)
 218                 tcf_chain_head_change_item(item, tp_head);
 219 }
 220
 221 static void tcf_chain_flush(struct tcf_chain *chain)
 222 {
 223         struct tcf_proto *tp = rtnl_dereference(chain->filter_chain);
 224
 225         tcf_chain_head_change(chain, NULL);
 226         while (tp) {
 227                 RCU_INIT_POINTER(chain->filter_chain, tp->next);
 228                 tcf_proto_destroy(tp, NULL);
 229                 tp = rtnl_dereference(chain->filter_chain);
 230                 tcf_chain_put(chain);
 231         }
 232 }
 233
 234 static void tcf_chain_destroy(struct tcf_chain *chain)
 235 {
 236         struct tcf_block *block = chain->block;
 237
 238         list_del(&chain->list);
 239         kfree(chain);
 240         if (list_empty(&block->chain_list))
 241                 kfree(block);
 242 }
 243
 244 static void tcf_chain_hold(struct tcf_chain *chain)
 245 {
 246         ++chain->refcnt;
 247 }
 248
 249 struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 250                                 bool create)
 251 {
 252         struct tcf_chain *chain;
 253
 254         list_for_each_entry(chain, &block->chain_list, list) {
 255                 if (chain->index == chain_index) {
 256                         tcf_chain_hold(chain);
 257                         return chain;
 258                 }
 259         }
 260
 261         return create ? tcf_chain_create(block, chain_index) : NULL;
 262 }
 263 EXPORT_SYMBOL(tcf_chain_get);
 264
 265 void tcf_chain_put(struct tcf_chain *chain)
 266 {
 267         if (--chain->refcnt == 0)
 268                 tcf_chain_destroy(chain);
 269 }
 270 EXPORT_SYMBOL(tcf_chain_put);
 271
 272 static bool tcf_block_offload_in_use(struct tcf_block *block)
 273 {
 274         return block->offloadcnt;
 275 }
 276
 277 static int tcf_block_offload_cmd(struct tcf_block *block,
 278                                  struct net_device *dev,
 279                                  struct tcf_block_ext_info *ei,
 280                                  enum tc_block_command command)
 281 {
 282         struct tc_block_offload bo = {};
 283
 284         bo.command = command;
 285         bo.binder_type = ei->binder_type;
 286         bo.block = block;
 287         return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 288 }
 289
 290 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 291                                   struct tcf_block_ext_info *ei)
 292 {
 293         struct net_device *dev = q->dev_queue->dev;
 294         int err;
 295
 296         if (!dev->netdev_ops->ndo_setup_tc)
 297                 goto no_offload_dev_inc;
 298
 299         /* If tc offload feature is disabled and the block we try to bind
 300          * to already has some offloaded filters, forbid to bind.
 301          */
 302         if (!tc_can_offload(dev) && tcf_block_offload_in_use(block))
 303                 return -EOPNOTSUPP;
 304
 305         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_BIND);
 306         if (err == -EOPNOTSUPP)
 307                 goto no_offload_dev_inc;
 308         return err;
 309
 310 no_offload_dev_inc:
 311         if (tcf_block_offload_in_use(block))
 312                 return -EOPNOTSUPP;
 313         block->nooffloaddevcnt++;
 314         return 0;
 315 }
 316
 317 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 318                                      struct tcf_block_ext_info *ei)
 319 {
 320         struct net_device *dev = q->dev_queue->dev;
 321         int err;
 322
 323         if (!dev->netdev_ops->ndo_setup_tc)
 324                 goto no_offload_dev_dec;
 325         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_UNBIND);
 326         if (err == -EOPNOTSUPP)
 327                 goto no_offload_dev_dec;
 328         return;
 329
 330 no_offload_dev_dec:
 331         WARN_ON(block->nooffloaddevcnt-- == 0);
 332 }
 333
 334 static int
 335 tcf_chain_head_change_cb_add(struct tcf_chain *chain,
 336                              struct tcf_block_ext_info *ei,
 337                              struct netlink_ext_ack *extack)
 338 {
 339         struct tcf_filter_chain_list_item *item;
 340
 341         item = kmalloc(sizeof(*item), GFP_KERNEL);
 342         if (!item) {
 343                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 344                 return -ENOMEM;
 345         }
 346         item->chain_head_change = ei->chain_head_change;
 347         item->chain_head_change_priv = ei->chain_head_change_priv;
 348         if (chain->filter_chain)
 349                 tcf_chain_head_change_item(item, chain->filter_chain);
 350         list_add(&item->list, &chain->filter_chain_list);
 351         return 0;
 352 }
 353
 354 static void
 355 tcf_chain_head_change_cb_del(struct tcf_chain *chain,
 356                              struct tcf_block_ext_info *ei)
 357 {
 358         struct tcf_filter_chain_list_item *item;
 359
 360         list_for_each_entry(item, &chain->filter_chain_list, list) {
 361                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 362                     (item->chain_head_change == ei->chain_head_change &&
 363                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 364                         tcf_chain_head_change_item(item, NULL);
 365                         list_del(&item->list);
 366                         kfree(item);
 367                         return;
 368                 }
 369         }
 370         WARN_ON(1);
 371 }
 372
 373 struct tcf_net {
 374         struct idr idr;
 375 };
 376
 377 static unsigned int tcf_net_id;
 378
 379 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 380                             struct netlink_ext_ack *extack)
 381 {
 382         struct tcf_net *tn = net_generic(net, tcf_net_id);
 383
 384         return idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 385                              GFP_KERNEL);
 386 }
 387
 388 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 389 {
 390         struct tcf_net *tn = net_generic(net, tcf_net_id);
 391
 392         idr_remove(&tn->idr, block->index);
 393 }
 394
 395 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 396                                           u32 block_index,
 397                                           struct netlink_ext_ack *extack)
 398 {
 399         struct tcf_block *block;
 400         struct tcf_chain *chain;
 401         int err;
 402
 403         block = kzalloc(sizeof(*block), GFP_KERNEL);
 404         if (!block) {
 405                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 406                 return ERR_PTR(-ENOMEM);
 407         }
 408         INIT_LIST_HEAD(&block->chain_list);
 409         INIT_LIST_HEAD(&block->cb_list);
 410         INIT_LIST_HEAD(&block->owner_list);
 411
 412         /* Create chain 0 by default, it has to be always present. */
 413         chain = tcf_chain_create(block, 0);
 414         if (!chain) {
 415                 NL_SET_ERR_MSG(extack, "Failed to create new tcf chain");
 416                 err = -ENOMEM;
 417                 goto err_chain_create;
 418         }
 419         block->refcnt = 1;
 420         block->net = net;
 421         block->index = block_index;
 422
 423         /* Don't store q pointer for blocks which are shared */
 424         if (!tcf_block_shared(block))
 425                 block->q = q;
 426         return block;
 427
 428 err_chain_create:
 429         kfree(block);
 430         return ERR_PTR(err);
 431 }
 432
 433 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 434 {
 435         struct tcf_net *tn = net_generic(net, tcf_net_id);
 436
 437         return idr_find(&tn->idr, block_index);
 438 }
 439
 440 /* Find tcf block.
 441  * Set q, parent, cl when appropriate.
 442  */
 443
 444 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
 445                                         u32 *parent, unsigned long *cl,
 446                                         int ifindex, u32 block_index,
 447                                         struct netlink_ext_ack *extack)
 448 {
 449         struct tcf_block *block;
 450
 451         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
 452                 block = tcf_block_lookup(net, block_index);
 453                 if (!block) {
 454                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
 455                         return ERR_PTR(-EINVAL);
 456                 }
 457         } else {
 458                 const struct Qdisc_class_ops *cops;
 459                 struct net_device *dev;
 460
 461                 /* Find link */
 462                 dev = __dev_get_by_index(net, ifindex);
 463                 if (!dev)
 464                         return ERR_PTR(-ENODEV);
 465
 466                 /* Find qdisc */
 467                 if (!*parent) {
 468                         *q = dev->qdisc;
 469                         *parent = (*q)->handle;
 470                 } else {
 471                         *q = qdisc_lookup(dev, TC_H_MAJ(*parent));
 472                         if (!*q) {
 473                                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
 474                                 return ERR_PTR(-EINVAL);
 475                         }
 476                 }
 477
 478                 /* Is it classful? */
 479                 cops = (*q)->ops->cl_ops;
 480                 if (!cops) {
 481                         NL_SET_ERR_MSG(extack, "Qdisc not classful");
 482                         return ERR_PTR(-EINVAL);
 483                 }
 484
 485                 if (!cops->tcf_block) {
 486                         NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
 487                         return ERR_PTR(-EOPNOTSUPP);
 488                 }
 489
 490                 /* Do we search for filter, attached to class? */
 491                 if (TC_H_MIN(*parent)) {
 492                         *cl = cops->find(*q, *parent);
 493                         if (*cl == 0) {
 494                                 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
 495                                 return ERR_PTR(-ENOENT);
 496                         }
 497                 }
 498
 499                 /* And the last stroke */
 500                 block = cops->tcf_block(*q, *cl, extack);
 501                 if (!block)
 502                         return ERR_PTR(-EINVAL);
 503                 if (tcf_block_shared(block)) {
 504                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
 505                         return ERR_PTR(-EOPNOTSUPP);
 506                 }
 507         }
 508
 509         return block;
 510 }
 511
 512 static struct tcf_chain *tcf_block_chain_zero(struct tcf_block *block)
 513 {
 514         return list_first_entry(&block->chain_list, struct tcf_chain, list);
 515 }
 516
 517 struct tcf_block_owner_item {
 518         struct list_head list;
 519         struct Qdisc *q;
 520         enum tcf_block_binder_type binder_type;
 521 };
 522
 523 static void
 524 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
 525                                struct Qdisc *q,
 526                                enum tcf_block_binder_type binder_type)
 527 {
 528         if (block->keep_dst &&
 529             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
 530             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
 531                 netif_keep_dst(qdisc_dev(q));
 532 }
 533
 534 void tcf_block_netif_keep_dst(struct tcf_block *block)
 535 {
 536         struct tcf_block_owner_item *item;
 537
 538         block->keep_dst = true;
 539         list_for_each_entry(item, &block->owner_list, list)
 540                 tcf_block_owner_netif_keep_dst(block, item->q,
 541                                                item->binder_type);
 542 }
 543 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
 544
 545 static int tcf_block_owner_add(struct tcf_block *block,
 546                                struct Qdisc *q,
 547                                enum tcf_block_binder_type binder_type)
 548 {
 549         struct tcf_block_owner_item *item;
 550
 551         item = kmalloc(sizeof(*item), GFP_KERNEL);
 552         if (!item)
 553                 return -ENOMEM;
 554         item->q = q;
 555         item->binder_type = binder_type;
 556         list_add(&item->list, &block->owner_list);
 557         return 0;
 558 }
 559
 560 static void tcf_block_owner_del(struct tcf_block *block,
 561                                 struct Qdisc *q,
 562                                 enum tcf_block_binder_type binder_type)
 563 {
 564         struct tcf_block_owner_item *item;
 565
 566         list_for_each_entry(item, &block->owner_list, list) {
 567                 if (item->q == q && item->binder_type == binder_type) {
 568                         list_del(&item->list);
 569                         kfree(item);
 570                         return;
 571                 }
 572         }
 573         WARN_ON(1);
 574 }
 575
 576 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 577                       struct tcf_block_ext_info *ei,
 578                       struct netlink_ext_ack *extack)
 579 {
 580         struct net *net = qdisc_net(q);
 581         struct tcf_block *block = NULL;
 582         bool created = false;
 583         int err;
 584
 585         if (ei->block_index) {
 586                 /* block_index not 0 means the shared block is requested */
 587                 block = tcf_block_lookup(net, ei->block_index);
 588                 if (block)
 589                         block->refcnt++;
 590         }
 591
 592         if (!block) {
 593                 block = tcf_block_create(net, q, ei->block_index, extack);
 594                 if (IS_ERR(block))
 595                         return PTR_ERR(block);
 596                 created = true;
 597                 if (tcf_block_shared(block)) {
 598                         err = tcf_block_insert(block, net, extack);
 599                         if (err)
 600                                 goto err_block_insert;
 601                 }
 602         }
 603
 604         err = tcf_block_owner_add(block, q, ei->binder_type);
 605         if (err)
 606                 goto err_block_owner_add;
 607
 608         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
 609
 610         err = tcf_chain_head_change_cb_add(tcf_block_chain_zero(block),
 611                                            ei, extack);
 612         if (err)
 613                 goto err_chain_head_change_cb_add;
 614
 615         err = tcf_block_offload_bind(block, q, ei);
 616         if (err)
 617                 goto err_block_offload_bind;
 618
 619         *p_block = block;
 620         return 0;
 621
 622 err_block_offload_bind:
 623         tcf_chain_head_change_cb_del(tcf_block_chain_zero(block), ei);
 624 err_chain_head_change_cb_add:
 625         tcf_block_owner_del(block, q, ei->binder_type);
 626 err_block_owner_add:
 627         if (created) {
 628                 if (tcf_block_shared(block))
 629                         tcf_block_remove(block, net);
 630 err_block_insert:
 631                 kfree(tcf_block_chain_zero(block));
 632                 kfree(block);
 633         } else {
 634                 block->refcnt--;
 635         }
 636         return err;
 637 }
 638 EXPORT_SYMBOL(tcf_block_get_ext);
 639
 640 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
 641 {
 642         struct tcf_proto __rcu **p_filter_chain = priv;
 643
 644         rcu_assign_pointer(*p_filter_chain, tp_head);
 645 }
 646
 647 int tcf_block_get(struct tcf_block **p_block,
 648                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
 649                   struct netlink_ext_ack *extack)
 650 {
 651         struct tcf_block_ext_info ei = {
 652                 .chain_head_change = tcf_chain_head_change_dflt,
 653                 .chain_head_change_priv = p_filter_chain,
 654         };
 655
 656         WARN_ON(!p_filter_chain);
 657         return tcf_block_get_ext(p_block, q, &ei, extack);
 658 }
 659 EXPORT_SYMBOL(tcf_block_get);
 660
 661 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
 662  * actions should be all removed after flushing.
 663  */
 664 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 665                        struct tcf_block_ext_info *ei)
 666 {
 667         struct tcf_chain *chain, *tmp;
 668
 669         if (!block)
 670                 return;
 671         tcf_chain_head_change_cb_del(tcf_block_chain_zero(block), ei);
 672         tcf_block_owner_del(block, q, ei->binder_type);
 673
 674         if (--block->refcnt == 0) {
 675                 if (tcf_block_shared(block))
 676                         tcf_block_remove(block, block->net);
 677
 678                 /* Hold a refcnt for all chains, so that they don't disappear
 679                  * while we are iterating.
 680                  */
 681                 list_for_each_entry(chain, &block->chain_list, list)
 682                         tcf_chain_hold(chain);
 683
 684                 list_for_each_entry(chain, &block->chain_list, list)
 685                         tcf_chain_flush(chain);
 686         }
 687
 688         tcf_block_offload_unbind(block, q, ei);
 689
 690         if (block->refcnt == 0) {
 691                 /* At this point, all the chains should have refcnt >= 1. */
 692                 list_for_each_entry_safe(chain, tmp, &block->chain_list, list)
 693                         tcf_chain_put(chain);
 694
 695                 /* Finally, put chain 0 and allow block to be freed. */
 696                 tcf_chain_put(tcf_block_chain_zero(block));
 697         }
 698 }
 699 EXPORT_SYMBOL(tcf_block_put_ext);
 700
 701 void tcf_block_put(struct tcf_block *block)
 702 {
 703         struct tcf_block_ext_info ei = {0, };
 704
 705         if (!block)
 706                 return;
 707         tcf_block_put_ext(block, block->q, &ei);
 708 }
 709
 710 EXPORT_SYMBOL(tcf_block_put);
 711
 712 struct tcf_block_cb {
 713         struct list_head list;
 714         tc_setup_cb_t *cb;
 715         void *cb_ident;
 716         void *cb_priv;
 717         unsigned int refcnt;
 718 };
 719
 720 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
 721 {
 722         return block_cb->cb_priv;
 723 }
 724 EXPORT_SYMBOL(tcf_block_cb_priv);
 725
 726 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
 727                                          tc_setup_cb_t *cb, void *cb_ident)
 728 {       struct tcf_block_cb *block_cb;
 729
 730         list_for_each_entry(block_cb, &block->cb_list, list)
 731                 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
 732                         return block_cb;
 733         return NULL;
 734 }
 735 EXPORT_SYMBOL(tcf_block_cb_lookup);
 736
 737 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
 738 {
 739         block_cb->refcnt++;
 740 }
 741 EXPORT_SYMBOL(tcf_block_cb_incref);
 742
 743 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
 744 {
 745         return --block_cb->refcnt;
 746 }
 747 EXPORT_SYMBOL(tcf_block_cb_decref);
 748
 749 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
 750                                              tc_setup_cb_t *cb, void *cb_ident,
 751                                              void *cb_priv)
 752 {
 753         struct tcf_block_cb *block_cb;
 754
 755         /* At this point, playback of previous block cb calls is not supported,
 756          * so forbid to register to block which already has some offloaded
 757          * filters present.
 758          */
 759         if (tcf_block_offload_in_use(block))
 760                 return ERR_PTR(-EOPNOTSUPP);
 761
 762         block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
 763         if (!block_cb)
 764                 return ERR_PTR(-ENOMEM);
 765         block_cb->cb = cb;
 766         block_cb->cb_ident = cb_ident;
 767         block_cb->cb_priv = cb_priv;
 768         list_add(&block_cb->list, &block->cb_list);
 769         return block_cb;
 770 }
 771 EXPORT_SYMBOL(__tcf_block_cb_register);
 772
 773 int tcf_block_cb_register(struct tcf_block *block,
 774                           tc_setup_cb_t *cb, void *cb_ident,
 775                           void *cb_priv)
 776 {
 777         struct tcf_block_cb *block_cb;
 778
 779         block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv);
 780         return IS_ERR(block_cb) ? PTR_ERR(block_cb) : 0;
 781 }
 782 EXPORT_SYMBOL(tcf_block_cb_register);
 783
 784 void __tcf_block_cb_unregister(struct tcf_block_cb *block_cb)
 785 {
 786         list_del(&block_cb->list);
 787         kfree(block_cb);
 788 }
 789 EXPORT_SYMBOL(__tcf_block_cb_unregister);
 790
 791 void tcf_block_cb_unregister(struct tcf_block *block,
 792                              tc_setup_cb_t *cb, void *cb_ident)
 793 {
 794         struct tcf_block_cb *block_cb;
 795
 796         block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
 797         if (!block_cb)
 798                 return;
 799         __tcf_block_cb_unregister(block_cb);
 800 }
 801 EXPORT_SYMBOL(tcf_block_cb_unregister);
 802
 803 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type,
 804                              void *type_data, bool err_stop)
 805 {
 806         struct tcf_block_cb *block_cb;
 807         int ok_count = 0;
 808         int err;
 809
 810         list_for_each_entry(block_cb, &block->cb_list, list) {
 811                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
 812                 if (err) {
 813                         if (err_stop)
 814                                 return err;
 815                 } else {
 816                         ok_count++;
 817                 }
 818         }
 819         return ok_count;
 820 }
 821
 822 /* Main classifier routine: scans classifier chain attached
 823  * to this qdisc, (optionally) tests for protocol and asks
 824  * specific classifiers.
 825  */
 826 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 827                  struct tcf_result *res, bool compat_mode)
 828 {
 829         __be16 protocol = tc_skb_protocol(skb);
 830 #ifdef CONFIG_NET_CLS_ACT
 831         const int max_reclassify_loop = 4;
 832         const struct tcf_proto *orig_tp = tp;
 833         const struct tcf_proto *first_tp;
 834         int limit = 0;
 835
 836 reclassify:
 837 #endif
 838         for (; tp; tp = rcu_dereference_bh(tp->next)) {
 839                 int err;
 840
 841                 if (tp->protocol != protocol &&
 842                     tp->protocol != htons(ETH_P_ALL))
 843                         continue;
 844
 845                 err = tp->classify(skb, tp, res);
 846 #ifdef CONFIG_NET_CLS_ACT
 847                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
 848                         first_tp = orig_tp;
 849                         goto reset;
 850                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
 851                         first_tp = res->goto_tp;
 852                         goto reset;
 853                 }
 854 #endif
 855                 if (err >= 0)
 856                         return err;
 857         }
 858
 859         return TC_ACT_UNSPEC; /* signal: continue lookup */
 860 #ifdef CONFIG_NET_CLS_ACT
 861 reset:
 862         if (unlikely(limit++ >= max_reclassify_loop)) {
 863                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
 864                                        tp->chain->block->index,
 865                                        tp->prio & 0xffff,
 866                                        ntohs(tp->protocol));
 867                 return TC_ACT_SHOT;
 868         }
 869
 870         tp = first_tp;
 871         protocol = tc_skb_protocol(skb);
 872         goto reclassify;
 873 #endif
 874 }
 875 EXPORT_SYMBOL(tcf_classify);
 876
 877 struct tcf_chain_info {
 878         struct tcf_proto __rcu **pprev;
 879         struct tcf_proto __rcu *next;
 880 };
 881
 882 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
 883 {
 884         return rtnl_dereference(*chain_info->pprev);
 885 }
 886
 887 static void tcf_chain_tp_insert(struct tcf_chain *chain,
 888                                 struct tcf_chain_info *chain_info,
 889                                 struct tcf_proto *tp)
 890 {
 891         if (*chain_info->pprev == chain->filter_chain)
 892                 tcf_chain_head_change(chain, tp);
 893         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
 894         rcu_assign_pointer(*chain_info->pprev, tp);
 895         tcf_chain_hold(chain);
 896 }
 897
 898 static void tcf_chain_tp_remove(struct tcf_chain *chain,
 899                                 struct tcf_chain_info *chain_info,
 900                                 struct tcf_proto *tp)
 901 {
 902         struct tcf_proto *next = rtnl_dereference(chain_info->next);
 903
 904         if (tp == chain->filter_chain)
 905                 tcf_chain_head_change(chain, next);
 906         RCU_INIT_POINTER(*chain_info->pprev, next);
 907         tcf_chain_put(chain);
 908 }
 909
 910 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
 911                                            struct tcf_chain_info *chain_info,
 912                                            u32 protocol, u32 prio,
 913                                            bool prio_allocate)
 914 {
 915         struct tcf_proto **pprev;
 916         struct tcf_proto *tp;
 917
 918         /* Check the chain for existence of proto-tcf with this priority */
 919         for (pprev = &chain->filter_chain;
 920              (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
 921                 if (tp->prio >= prio) {
 922                         if (tp->prio == prio) {
 923                                 if (prio_allocate ||
 924                                     (tp->protocol != protocol && protocol))
 925                                         return ERR_PTR(-EINVAL);
 926                         } else {
 927                                 tp = NULL;
 928                         }
 929                         break;
 930                 }
 931         }
 932         chain_info->pprev = pprev;
 933         chain_info->next = tp ? tp->next : NULL;
 934         return tp;
 935 }
 936
 937 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
 938                          struct tcf_proto *tp, struct tcf_block *block,
 939                          struct Qdisc *q, u32 parent, void *fh,
 940                          u32 portid, u32 seq, u16 flags, int event)
 941 {
 942         struct tcmsg *tcm;
 943         struct nlmsghdr  *nlh;
 944         unsigned char *b = skb_tail_pointer(skb);
 945
 946         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
 947         if (!nlh)
 948                 goto out_nlmsg_trim;
 949         tcm = nlmsg_data(nlh);
 950         tcm->tcm_family = AF_UNSPEC;
 951         tcm->tcm__pad1 = 0;
 952         tcm->tcm__pad2 = 0;
 953         if (q) {
 954                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
 955                 tcm->tcm_parent = parent;
 956         } else {
 957                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
 958                 tcm->tcm_block_index = block->index;
 959         }
 960         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
 961         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
 962                 goto nla_put_failure;
 963         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
 964                 goto nla_put_failure;
 965         if (!fh) {
 966                 tcm->tcm_handle = 0;
 967         } else {
 968                 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
 969                         goto nla_put_failure;
 970         }
 971         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 972         return skb->len;
 973
 974 out_nlmsg_trim:
 975 nla_put_failure:
 976         nlmsg_trim(skb, b);
 977         return -1;
 978 }
 979
 980 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
 981                           struct nlmsghdr *n, struct tcf_proto *tp,
 982                           struct tcf_block *block, struct Qdisc *q,
 983                           u32 parent, void *fh, int event, bool unicast)
 984 {
 985         struct sk_buff *skb;
 986         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
 987
 988         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 989         if (!skb)
 990                 return -ENOBUFS;
 991
 992         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
 993                           n->nlmsg_seq, n->nlmsg_flags, event) <= 0) {
 994                 kfree_skb(skb);
 995                 return -EINVAL;
 996         }
 997
 998         if (unicast)
 999                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1000
1001         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1002                               n->nlmsg_flags & NLM_F_ECHO);
1003 }
1004
1005 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1006                               struct nlmsghdr *n, struct tcf_proto *tp,
1007                               struct tcf_block *block, struct Qdisc *q,
1008                               u32 parent, void *fh, bool unicast, bool *last,
1009                               struct netlink_ext_ack *extack)
1010 {
1011         struct sk_buff *skb;
1012         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1013         int err;
1014
1015         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1016         if (!skb)
1017                 return -ENOBUFS;
1018
1019         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1020                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
1021                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1022                 kfree_skb(skb);
1023                 return -EINVAL;
1024         }
1025
1026         err = tp->ops->delete(tp, fh, last, extack);
1027         if (err) {
1028                 kfree_skb(skb);
1029                 return err;
1030         }
1031
1032         if (unicast)
1033                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1034
1035         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1036                              n->nlmsg_flags & NLM_F_ECHO);
1037         if (err < 0)
1038                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1039         return err;
1040 }
1041
1042 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1043                                  struct tcf_block *block, struct Qdisc *q,
1044                                  u32 parent, struct nlmsghdr *n,
1045                                  struct tcf_chain *chain, int event)
1046 {
1047         struct tcf_proto *tp;
1048
1049         for (tp = rtnl_dereference(chain->filter_chain);
1050              tp; tp = rtnl_dereference(tp->next))
1051                 tfilter_notify(net, oskb, n, tp, block,
1052                                q, parent, 0, event, false);
1053 }
1054
1055 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1056                           struct netlink_ext_ack *extack)
1057 {
1058         struct net *net = sock_net(skb->sk);
1059         struct nlattr *tca[TCA_MAX + 1];
1060         struct tcmsg *t;
1061         u32 protocol;
1062         u32 prio;
1063         bool prio_allocate;
1064         u32 parent;
1065         u32 chain_index;
1066         struct Qdisc *q = NULL;
1067         struct tcf_chain_info chain_info;
1068         struct tcf_chain *chain = NULL;
1069         struct tcf_block *block;
1070         struct tcf_proto *tp;
1071         unsigned long cl;
1072         void *fh;
1073         int err;
1074         int tp_created;
1075
1076         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1077                 return -EPERM;
1078
1079 replay:
1080         tp_created = 0;
1081
1082         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1083         if (err < 0)
1084                 return err;
1085
1086         t = nlmsg_data(n);
1087         protocol = TC_H_MIN(t->tcm_info);
1088         prio = TC_H_MAJ(t->tcm_info);
1089         prio_allocate = false;
1090         parent = t->tcm_parent;
1091         cl = 0;
1092
1093         if (prio == 0) {
1094                 /* If no priority is provided by the user,
1095                  * we allocate one.
1096                  */
1097                 if (n->nlmsg_flags & NLM_F_CREATE) {
1098                         prio = TC_H_MAKE(0x80000000U, 0U);
1099                         prio_allocate = true;
1100                 } else {
1101                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1102                         return -ENOENT;
1103                 }
1104         }
1105
1106         /* Find head of filter chain. */
1107
1108         block = tcf_block_find(net, &q, &parent, &cl,
1109                                t->tcm_ifindex, t->tcm_block_index, extack);
1110         if (IS_ERR(block)) {
1111                 err = PTR_ERR(block);
1112                 goto errout;
1113         }
1114
1115         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1116         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1117                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1118                 err = -EINVAL;
1119                 goto errout;
1120         }
1121         chain = tcf_chain_get(block, chain_index, true);
1122         if (!chain) {
1123                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1124                 err = -ENOMEM;
1125                 goto errout;
1126         }
1127
1128         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1129                                prio, prio_allocate);
1130         if (IS_ERR(tp)) {
1131                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1132                 err = PTR_ERR(tp);
1133                 goto errout;
1134         }
1135
1136         if (tp == NULL) {
1137                 /* Proto-tcf does not exist, create new one */
1138
1139                 if (tca[TCA_KIND] == NULL || !protocol) {
1140                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
1141                         err = -EINVAL;
1142                         goto errout;
1143                 }
1144
1145                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1146                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1147                         err = -ENOENT;
1148                         goto errout;
1149                 }
1150
1151                 if (prio_allocate)
1152                         prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
1153
1154                 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
1155                                       protocol, prio, chain, extack);
1156                 if (IS_ERR(tp)) {
1157                         err = PTR_ERR(tp);
1158                         goto errout;
1159                 }
1160                 tp_created = 1;
1161         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1162                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1163                 err = -EINVAL;
1164                 goto errout;
1165         }
1166
1167         fh = tp->ops->get(tp, t->tcm_handle);
1168
1169         if (!fh) {
1170                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1171                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1172                         err = -ENOENT;
1173                         goto errout;
1174                 }
1175         } else if (n->nlmsg_flags & NLM_F_EXCL) {
1176                 NL_SET_ERR_MSG(extack, "Filter already exists");
1177                 err = -EEXIST;
1178                 goto errout;
1179         }
1180
1181         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
1182                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
1183                               extack);
1184         if (err == 0) {
1185                 if (tp_created)
1186                         tcf_chain_tp_insert(chain, &chain_info, tp);
1187                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1188                                RTM_NEWTFILTER, false);
1189         } else {
1190                 if (tp_created)
1191                         tcf_proto_destroy(tp, NULL);
1192         }
1193
1194 errout:
1195         if (chain)
1196                 tcf_chain_put(chain);
1197         if (err == -EAGAIN)
1198                 /* Replay the request. */
1199                 goto replay;
1200         return err;
1201 }
1202
1203 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1204                           struct netlink_ext_ack *extack)
1205 {
1206         struct net *net = sock_net(skb->sk);
1207         struct nlattr *tca[TCA_MAX + 1];
1208         struct tcmsg *t;
1209         u32 protocol;
1210         u32 prio;
1211         u32 parent;
1212         u32 chain_index;
1213         struct Qdisc *q = NULL;
1214         struct tcf_chain_info chain_info;
1215         struct tcf_chain *chain = NULL;
1216         struct tcf_block *block;
1217         struct tcf_proto *tp = NULL;
1218         unsigned long cl = 0;
1219         void *fh = NULL;
1220         int err;
1221
1222         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1223                 return -EPERM;
1224
1225         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1226         if (err < 0)
1227                 return err;
1228
1229         t = nlmsg_data(n);
1230         protocol = TC_H_MIN(t->tcm_info);
1231         prio = TC_H_MAJ(t->tcm_info);
1232         parent = t->tcm_parent;
1233
1234         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
1235                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
1236                 return -ENOENT;
1237         }
1238
1239         /* Find head of filter chain. */
1240
1241         block = tcf_block_find(net, &q, &parent, &cl,
1242                                t->tcm_ifindex, t->tcm_block_index, extack);
1243         if (IS_ERR(block)) {
1244                 err = PTR_ERR(block);
1245                 goto errout;
1246         }
1247
1248         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1249         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1250                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1251                 err = -EINVAL;
1252                 goto errout;
1253         }
1254         chain = tcf_chain_get(block, chain_index, false);
1255         if (!chain) {
1256                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1257                 err = -EINVAL;
1258                 goto errout;
1259         }
1260
1261         if (prio == 0) {
1262                 tfilter_notify_chain(net, skb, block, q, parent, n,
1263                                      chain, RTM_DELTFILTER);
1264                 tcf_chain_flush(chain);
1265                 err = 0;
1266                 goto errout;
1267         }
1268
1269         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1270                                prio, false);
1271         if (!tp || IS_ERR(tp)) {
1272                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1273                 err = tp ? PTR_ERR(tp) : -ENOENT;
1274                 goto errout;
1275         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1276                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1277                 err = -EINVAL;
1278                 goto errout;
1279         }
1280
1281         fh = tp->ops->get(tp, t->tcm_handle);
1282
1283         if (!fh) {
1284                 if (t->tcm_handle == 0) {
1285                         tcf_chain_tp_remove(chain, &chain_info, tp);
1286                         tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1287                                        RTM_DELTFILTER, false);
1288                         tcf_proto_destroy(tp, extack);
1289                         err = 0;
1290                 } else {
1291                         NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1292                         err = -ENOENT;
1293                 }
1294         } else {
1295                 bool last;
1296
1297                 err = tfilter_del_notify(net, skb, n, tp, block,
1298                                          q, parent, fh, false, &last,
1299                                          extack);
1300                 if (err)
1301                         goto errout;
1302                 if (last) {
1303                         tcf_chain_tp_remove(chain, &chain_info, tp);
1304                         tcf_proto_destroy(tp, extack);
1305                 }
1306         }
1307
1308 errout:
1309         if (chain)
1310                 tcf_chain_put(chain);
1311         return err;
1312 }
1313
1314 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1315                           struct netlink_ext_ack *extack)
1316 {
1317         struct net *net = sock_net(skb->sk);
1318         struct nlattr *tca[TCA_MAX + 1];
1319         struct tcmsg *t;
1320         u32 protocol;
1321         u32 prio;
1322         u32 parent;
1323         u32 chain_index;
1324         struct Qdisc *q = NULL;
1325         struct tcf_chain_info chain_info;
1326         struct tcf_chain *chain = NULL;
1327         struct tcf_block *block;
1328         struct tcf_proto *tp = NULL;
1329         unsigned long cl = 0;
1330         void *fh = NULL;
1331         int err;
1332
1333         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1334         if (err < 0)
1335                 return err;
1336
1337         t = nlmsg_data(n);
1338         protocol = TC_H_MIN(t->tcm_info);
1339         prio = TC_H_MAJ(t->tcm_info);
1340         parent = t->tcm_parent;
1341
1342         if (prio == 0) {
1343                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1344                 return -ENOENT;
1345         }
1346
1347         /* Find head of filter chain. */
1348
1349         block = tcf_block_find(net, &q, &parent, &cl,
1350                                t->tcm_ifindex, t->tcm_block_index, extack);
1351         if (IS_ERR(block)) {
1352                 err = PTR_ERR(block);
1353                 goto errout;
1354         }
1355
1356         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1357         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1358                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1359                 err = -EINVAL;
1360                 goto errout;
1361         }
1362         chain = tcf_chain_get(block, chain_index, false);
1363         if (!chain) {
1364                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1365                 err = -EINVAL;
1366                 goto errout;
1367         }
1368
1369         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1370                                prio, false);
1371         if (!tp || IS_ERR(tp)) {
1372                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1373                 err = tp ? PTR_ERR(tp) : -ENOENT;
1374                 goto errout;
1375         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1376                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1377                 err = -EINVAL;
1378                 goto errout;
1379         }
1380
1381         fh = tp->ops->get(tp, t->tcm_handle);
1382
1383         if (!fh) {
1384                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1385                 err = -ENOENT;
1386         } else {
1387                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
1388                                      fh, RTM_NEWTFILTER, true);
1389                 if (err < 0)
1390                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
1391         }
1392
1393 errout:
1394         if (chain)
1395                 tcf_chain_put(chain);
1396         return err;
1397 }
1398
1399 struct tcf_dump_args {
1400         struct tcf_walker w;
1401         struct sk_buff *skb;
1402         struct netlink_callback *cb;
1403         struct tcf_block *block;
1404         struct Qdisc *q;
1405         u32 parent;
1406 };
1407
1408 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1409 {
1410         struct tcf_dump_args *a = (void *)arg;
1411         struct net *net = sock_net(a->skb->sk);
1412
1413         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
1414                              n, NETLINK_CB(a->cb->skb).portid,
1415                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1416                              RTM_NEWTFILTER);
1417 }
1418
1419 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
1420                            struct sk_buff *skb, struct netlink_callback *cb,
1421                            long index_start, long *p_index)
1422 {
1423         struct net *net = sock_net(skb->sk);
1424         struct tcf_block *block = chain->block;
1425         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1426         struct tcf_dump_args arg;
1427         struct tcf_proto *tp;
1428
1429         for (tp = rtnl_dereference(chain->filter_chain);
1430              tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
1431                 if (*p_index < index_start)
1432                         continue;
1433                 if (TC_H_MAJ(tcm->tcm_info) &&
1434                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
1435                         continue;
1436                 if (TC_H_MIN(tcm->tcm_info) &&
1437                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
1438                         continue;
1439                 if (*p_index > index_start)
1440                         memset(&cb->args[1], 0,
1441                                sizeof(cb->args) - sizeof(cb->args[0]));
1442                 if (cb->args[1] == 0) {
1443                         if (tcf_fill_node(net, skb, tp, block, q, parent, 0,
1444                                           NETLINK_CB(cb->skb).portid,
1445                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
1446                                           RTM_NEWTFILTER) <= 0)
1447                                 return false;
1448
1449                         cb->args[1] = 1;
1450                 }
1451                 if (!tp->ops->walk)
1452                         continue;
1453                 arg.w.fn = tcf_node_dump;
1454                 arg.skb = skb;
1455                 arg.cb = cb;
1456                 arg.block = block;
1457                 arg.q = q;
1458                 arg.parent = parent;
1459                 arg.w.stop = 0;
1460                 arg.w.skip = cb->args[1] - 1;
1461                 arg.w.count = 0;
1462                 tp->ops->walk(tp, &arg.w);
1463                 cb->args[1] = arg.w.count + 1;
1464                 if (arg.w.stop)
1465                         return false;
1466         }
1467         return true;
1468 }
1469
1470 /* called with RTNL */
1471 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
1472 {
1473         struct net *net = sock_net(skb->sk);
1474         struct nlattr *tca[TCA_MAX + 1];
1475         struct Qdisc *q = NULL;
1476         struct tcf_block *block;
1477         struct tcf_chain *chain;
1478         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1479         long index_start;
1480         long index;
1481         u32 parent;
1482         int err;
1483
1484         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1485                 return skb->len;
1486
1487         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1488         if (err)
1489                 return err;
1490
1491         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1492                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1493                 if (!block)
1494                         goto out;
1495                 /* If we work with block index, q is NULL and parent value
1496                  * will never be used in the following code. The check
1497                  * in tcf_fill_node prevents it. However, compiler does not
1498                  * see that far, so set parent to zero to silence the warning
1499                  * about parent being uninitialized.
1500                  */
1501                 parent = 0;
1502         } else {
1503                 const struct Qdisc_class_ops *cops;
1504                 struct net_device *dev;
1505                 unsigned long cl = 0;
1506
1507                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1508                 if (!dev)
1509                         return skb->len;
1510
1511                 parent = tcm->tcm_parent;
1512                 if (!parent) {
1513                         q = dev->qdisc;
1514                         parent = q->handle;
1515                 } else {
1516                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1517                 }
1518                 if (!q)
1519                         goto out;
1520                 cops = q->ops->cl_ops;
1521                 if (!cops)
1522                         goto out;
1523                 if (!cops->tcf_block)
1524                         goto out;
1525                 if (TC_H_MIN(tcm->tcm_parent)) {
1526                         cl = cops->find(q, tcm->tcm_parent);
1527                         if (cl == 0)
1528                                 goto out;
1529                 }
1530                 block = cops->tcf_block(q, cl, NULL);
1531                 if (!block)
1532                         goto out;
1533                 if (tcf_block_shared(block))
1534                         q = NULL;
1535         }
1536
1537         index_start = cb->args[0];
1538         index = 0;
1539
1540         list_for_each_entry(chain, &block->chain_list, list) {
1541                 if (tca[TCA_CHAIN] &&
1542                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1543                         continue;
1544                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
1545                                     index_start, &index)) {
1546                         err = -EMSGSIZE;
1547                         break;
1548                 }
1549         }
1550
1551         cb->args[0] = index;
1552
1553 out:
1554         /* If we did no progress, the error (EMSGSIZE) is real */
1555         if (skb->len == 0 && err)
1556                 return err;
1557         return skb->len;
1558 }
1559
1560 void tcf_exts_destroy(struct tcf_exts *exts)
1561 {
1562 #ifdef CONFIG_NET_CLS_ACT
1563         LIST_HEAD(actions);
1564
1565         ASSERT_RTNL();
1566         tcf_exts_to_list(exts, &actions);
1567         tcf_action_destroy(&actions, TCA_ACT_UNBIND);
1568         kfree(exts->actions);
1569         exts->nr_actions = 0;
1570 #endif
1571 }
1572 EXPORT_SYMBOL(tcf_exts_destroy);
1573
1574 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
1575                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
1576                       struct netlink_ext_ack *extack)
1577 {
1578 #ifdef CONFIG_NET_CLS_ACT
1579         {
1580                 struct tc_action *act;
1581                 size_t attr_size = 0;
1582
1583                 if (exts->police && tb[exts->police]) {
1584                         act = tcf_action_init_1(net, tp, tb[exts->police],
1585                                                 rate_tlv, "police", ovr,
1586                                                 TCA_ACT_BIND, extack);
1587                         if (IS_ERR(act))
1588                                 return PTR_ERR(act);
1589
1590                         act->type = exts->type = TCA_OLD_COMPAT;
1591                         exts->actions[0] = act;
1592                         exts->nr_actions = 1;
1593                 } else if (exts->action && tb[exts->action]) {
1594                         LIST_HEAD(actions);
1595                         int err, i = 0;
1596
1597                         err = tcf_action_init(net, tp, tb[exts->action],
1598                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
1599                                               &actions, &attr_size, extack);
1600                         if (err)
1601                                 return err;
1602                         list_for_each_entry(act, &actions, list)
1603                                 exts->actions[i++] = act;
1604                         exts->nr_actions = i;
1605                 }
1606                 exts->net = net;
1607         }
1608 #else
1609         if ((exts->action && tb[exts->action]) ||
1610             (exts->police && tb[exts->police])) {
1611                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
1612                 return -EOPNOTSUPP;
1613         }
1614 #endif
1615
1616         return 0;
1617 }
1618 EXPORT_SYMBOL(tcf_exts_validate);
1619
1620 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
1621 {
1622 #ifdef CONFIG_NET_CLS_ACT
1623         struct tcf_exts old = *dst;
1624
1625         *dst = *src;
1626         tcf_exts_destroy(&old);
1627 #endif
1628 }
1629 EXPORT_SYMBOL(tcf_exts_change);
1630
1631 #ifdef CONFIG_NET_CLS_ACT
1632 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
1633 {
1634         if (exts->nr_actions == 0)
1635                 return NULL;
1636         else
1637                 return exts->actions[0];
1638 }
1639 #endif
1640
1641 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
1642 {
1643 #ifdef CONFIG_NET_CLS_ACT
1644         struct nlattr *nest;
1645
1646         if (exts->action && tcf_exts_has_actions(exts)) {
1647                 /*
1648                  * again for backward compatible mode - we want
1649                  * to work with both old and new modes of entering
1650                  * tc data even if iproute2  was newer - jhs
1651                  */
1652                 if (exts->type != TCA_OLD_COMPAT) {
1653                         LIST_HEAD(actions);
1654
1655                         nest = nla_nest_start(skb, exts->action);
1656                         if (nest == NULL)
1657                                 goto nla_put_failure;
1658
1659                         tcf_exts_to_list(exts, &actions);
1660                         if (tcf_action_dump(skb, &actions, 0, 0) < 0)
1661                                 goto nla_put_failure;
1662                         nla_nest_end(skb, nest);
1663                 } else if (exts->police) {
1664                         struct tc_action *act = tcf_exts_first_act(exts);
1665                         nest = nla_nest_start(skb, exts->police);
1666                         if (nest == NULL || !act)
1667                                 goto nla_put_failure;
1668                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
1669                                 goto nla_put_failure;
1670                         nla_nest_end(skb, nest);
1671                 }
1672         }
1673         return 0;
1674
1675 nla_put_failure:
1676         nla_nest_cancel(skb, nest);
1677         return -1;
1678 #else
1679         return 0;
1680 #endif
1681 }
1682 EXPORT_SYMBOL(tcf_exts_dump);
1683
1684
1685 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1686 {
1687 #ifdef CONFIG_NET_CLS_ACT
1688         struct tc_action *a = tcf_exts_first_act(exts);
1689         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
1690                 return -1;
1691 #endif
1692         return 0;
1693 }
1694 EXPORT_SYMBOL(tcf_exts_dump_stats);
1695
1696 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
1697                                        enum tc_setup_type type,
1698                                        void *type_data, bool err_stop)
1699 {
1700         int ok_count = 0;
1701 #ifdef CONFIG_NET_CLS_ACT
1702         const struct tc_action *a;
1703         struct net_device *dev;
1704         int i, ret;
1705
1706         if (!tcf_exts_has_actions(exts))
1707                 return 0;
1708
1709         for (i = 0; i < exts->nr_actions; i++) {
1710                 a = exts->actions[i];
1711                 if (!a->ops->get_dev)
1712                         continue;
1713                 dev = a->ops->get_dev(a);
1714                 if (!dev)
1715                         continue;
1716                 ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
1717                 if (ret < 0)
1718                         return ret;
1719                 ok_count += ret;
1720         }
1721 #endif
1722         return ok_count;
1723 }
1724
1725 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
1726                      enum tc_setup_type type, void *type_data, bool err_stop)
1727 {
1728         int ok_count = 0;
1729         int ret;
1730
1731         if (!block->nooffloaddevcnt) {
1732                 ret = tcf_block_cb_call(block, type, type_data, err_stop);
1733                 if (ret < 0)
1734                         return ret;
1735                 ok_count = ret;
1736         }
1737
1738         if (!exts || ok_count)
1739                 goto skip_egress;
1740
1741         ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
1742         if (ret < 0)
1743                 return ret;
1744         ok_count += ret;
1745
1746 skip_egress:
1747         /* if one of the netdevs sharing this block are not offload-capable
1748          * make sure we succeeded in egress instead.
1749          */
1750         if (block->nooffloaddevcnt && !ok_count && err_stop)
1751                 return -EOPNOTSUPP;
1752
1753         return ok_count;
1754 }
1755 EXPORT_SYMBOL(tc_setup_cb_call);
1756
1757 static __net_init int tcf_net_init(struct net *net)
1758 {
1759         struct tcf_net *tn = net_generic(net, tcf_net_id);
1760
1761         idr_init(&tn->idr);
1762         return 0;
1763 }
1764
1765 static void __net_exit tcf_net_exit(struct net *net)
1766 {
1767         struct tcf_net *tn = net_generic(net, tcf_net_id);
1768
1769         idr_destroy(&tn->idr);
1770 }
1771
1772 static struct pernet_operations tcf_net_ops = {
1773         .init = tcf_net_init,
1774         .exit = tcf_net_exit,
1775         .id   = &tcf_net_id,
1776         .size = sizeof(struct tcf_net),
1777 };
1778
1779 static int __init tc_filter_init(void)
1780 {
1781         int err;
1782
1783         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
1784         if (!tc_filter_wq)
1785                 return -ENOMEM;
1786
1787         err = register_pernet_subsys(&tcf_net_ops);
1788         if (err)
1789                 goto err_register_pernet_subsys;
1790
1791         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 0);
1792         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 0);
1793         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
1794                       tc_dump_tfilter, 0);
1795
1796         return 0;
1797
1798 err_register_pernet_subsys:
1799         destroy_workqueue(tc_filter_wq);
1800         return err;
1801 }
1802
1803 subsys_initcall(tc_filter_init);