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         /* Make sure all netdevs sharing this block are offload-capable. */
 811         if (block->nooffloaddevcnt && err_stop)
 812                 return -EOPNOTSUPP;
 813
 814         list_for_each_entry(block_cb, &block->cb_list, list) {
 815                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
 816                 if (err) {
 817                         if (err_stop)
 818                                 return err;
 819                 } else {
 820                         ok_count++;
 821                 }
 822         }
 823         return ok_count;
 824 }
 825
 826 /* Main classifier routine: scans classifier chain attached
 827  * to this qdisc, (optionally) tests for protocol and asks
 828  * specific classifiers.
 829  */
 830 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 831                  struct tcf_result *res, bool compat_mode)
 832 {
 833         __be16 protocol = tc_skb_protocol(skb);
 834 #ifdef CONFIG_NET_CLS_ACT
 835         const int max_reclassify_loop = 4;
 836         const struct tcf_proto *orig_tp = tp;
 837         const struct tcf_proto *first_tp;
 838         int limit = 0;
 839
 840 reclassify:
 841 #endif
 842         for (; tp; tp = rcu_dereference_bh(tp->next)) {
 843                 int err;
 844
 845                 if (tp->protocol != protocol &&
 846                     tp->protocol != htons(ETH_P_ALL))
 847                         continue;
 848
 849                 err = tp->classify(skb, tp, res);
 850 #ifdef CONFIG_NET_CLS_ACT
 851                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
 852                         first_tp = orig_tp;
 853                         goto reset;
 854                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
 855                         first_tp = res->goto_tp;
 856                         goto reset;
 857                 }
 858 #endif
 859                 if (err >= 0)
 860                         return err;
 861         }
 862
 863         return TC_ACT_UNSPEC; /* signal: continue lookup */
 864 #ifdef CONFIG_NET_CLS_ACT
 865 reset:
 866         if (unlikely(limit++ >= max_reclassify_loop)) {
 867                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
 868                                        tp->chain->block->index,
 869                                        tp->prio & 0xffff,
 870                                        ntohs(tp->protocol));
 871                 return TC_ACT_SHOT;
 872         }
 873
 874         tp = first_tp;
 875         protocol = tc_skb_protocol(skb);
 876         goto reclassify;
 877 #endif
 878 }
 879 EXPORT_SYMBOL(tcf_classify);
 880
 881 struct tcf_chain_info {
 882         struct tcf_proto __rcu **pprev;
 883         struct tcf_proto __rcu *next;
 884 };
 885
 886 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
 887 {
 888         return rtnl_dereference(*chain_info->pprev);
 889 }
 890
 891 static void tcf_chain_tp_insert(struct tcf_chain *chain,
 892                                 struct tcf_chain_info *chain_info,
 893                                 struct tcf_proto *tp)
 894 {
 895         if (*chain_info->pprev == chain->filter_chain)
 896                 tcf_chain_head_change(chain, tp);
 897         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
 898         rcu_assign_pointer(*chain_info->pprev, tp);
 899         tcf_chain_hold(chain);
 900 }
 901
 902 static void tcf_chain_tp_remove(struct tcf_chain *chain,
 903                                 struct tcf_chain_info *chain_info,
 904                                 struct tcf_proto *tp)
 905 {
 906         struct tcf_proto *next = rtnl_dereference(chain_info->next);
 907
 908         if (tp == chain->filter_chain)
 909                 tcf_chain_head_change(chain, next);
 910         RCU_INIT_POINTER(*chain_info->pprev, next);
 911         tcf_chain_put(chain);
 912 }
 913
 914 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
 915                                            struct tcf_chain_info *chain_info,
 916                                            u32 protocol, u32 prio,
 917                                            bool prio_allocate)
 918 {
 919         struct tcf_proto **pprev;
 920         struct tcf_proto *tp;
 921
 922         /* Check the chain for existence of proto-tcf with this priority */
 923         for (pprev = &chain->filter_chain;
 924              (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
 925                 if (tp->prio >= prio) {
 926                         if (tp->prio == prio) {
 927                                 if (prio_allocate ||
 928                                     (tp->protocol != protocol && protocol))
 929                                         return ERR_PTR(-EINVAL);
 930                         } else {
 931                                 tp = NULL;
 932                         }
 933                         break;
 934                 }
 935         }
 936         chain_info->pprev = pprev;
 937         chain_info->next = tp ? tp->next : NULL;
 938         return tp;
 939 }
 940
 941 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
 942                          struct tcf_proto *tp, struct tcf_block *block,
 943                          struct Qdisc *q, u32 parent, void *fh,
 944                          u32 portid, u32 seq, u16 flags, int event)
 945 {
 946         struct tcmsg *tcm;
 947         struct nlmsghdr  *nlh;
 948         unsigned char *b = skb_tail_pointer(skb);
 949
 950         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
 951         if (!nlh)
 952                 goto out_nlmsg_trim;
 953         tcm = nlmsg_data(nlh);
 954         tcm->tcm_family = AF_UNSPEC;
 955         tcm->tcm__pad1 = 0;
 956         tcm->tcm__pad2 = 0;
 957         if (q) {
 958                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
 959                 tcm->tcm_parent = parent;
 960         } else {
 961                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
 962                 tcm->tcm_block_index = block->index;
 963         }
 964         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
 965         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
 966                 goto nla_put_failure;
 967         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
 968                 goto nla_put_failure;
 969         if (!fh) {
 970                 tcm->tcm_handle = 0;
 971         } else {
 972                 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
 973                         goto nla_put_failure;
 974         }
 975         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 976         return skb->len;
 977
 978 out_nlmsg_trim:
 979 nla_put_failure:
 980         nlmsg_trim(skb, b);
 981         return -1;
 982 }
 983
 984 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
 985                           struct nlmsghdr *n, struct tcf_proto *tp,
 986                           struct tcf_block *block, struct Qdisc *q,
 987                           u32 parent, void *fh, int event, bool unicast)
 988 {
 989         struct sk_buff *skb;
 990         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
 991
 992         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 993         if (!skb)
 994                 return -ENOBUFS;
 995
 996         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
 997                           n->nlmsg_seq, n->nlmsg_flags, event) <= 0) {
 998                 kfree_skb(skb);
 999                 return -EINVAL;
1000         }
1001
1002         if (unicast)
1003                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1004
1005         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1006                               n->nlmsg_flags & NLM_F_ECHO);
1007 }
1008
1009 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1010                               struct nlmsghdr *n, struct tcf_proto *tp,
1011                               struct tcf_block *block, struct Qdisc *q,
1012                               u32 parent, void *fh, bool unicast, bool *last,
1013                               struct netlink_ext_ack *extack)
1014 {
1015         struct sk_buff *skb;
1016         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1017         int err;
1018
1019         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1020         if (!skb)
1021                 return -ENOBUFS;
1022
1023         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1024                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
1025                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1026                 kfree_skb(skb);
1027                 return -EINVAL;
1028         }
1029
1030         err = tp->ops->delete(tp, fh, last, extack);
1031         if (err) {
1032                 kfree_skb(skb);
1033                 return err;
1034         }
1035
1036         if (unicast)
1037                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1038
1039         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1040                              n->nlmsg_flags & NLM_F_ECHO);
1041         if (err < 0)
1042                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1043         return err;
1044 }
1045
1046 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1047                                  struct tcf_block *block, struct Qdisc *q,
1048                                  u32 parent, struct nlmsghdr *n,
1049                                  struct tcf_chain *chain, int event)
1050 {
1051         struct tcf_proto *tp;
1052
1053         for (tp = rtnl_dereference(chain->filter_chain);
1054              tp; tp = rtnl_dereference(tp->next))
1055                 tfilter_notify(net, oskb, n, tp, block,
1056                                q, parent, NULL, event, false);
1057 }
1058
1059 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1060                           struct netlink_ext_ack *extack)
1061 {
1062         struct net *net = sock_net(skb->sk);
1063         struct nlattr *tca[TCA_MAX + 1];
1064         struct tcmsg *t;
1065         u32 protocol;
1066         u32 prio;
1067         bool prio_allocate;
1068         u32 parent;
1069         u32 chain_index;
1070         struct Qdisc *q = NULL;
1071         struct tcf_chain_info chain_info;
1072         struct tcf_chain *chain = NULL;
1073         struct tcf_block *block;
1074         struct tcf_proto *tp;
1075         unsigned long cl;
1076         void *fh;
1077         int err;
1078         int tp_created;
1079
1080         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1081                 return -EPERM;
1082
1083 replay:
1084         tp_created = 0;
1085
1086         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1087         if (err < 0)
1088                 return err;
1089
1090         t = nlmsg_data(n);
1091         protocol = TC_H_MIN(t->tcm_info);
1092         prio = TC_H_MAJ(t->tcm_info);
1093         prio_allocate = false;
1094         parent = t->tcm_parent;
1095         cl = 0;
1096
1097         if (prio == 0) {
1098                 /* If no priority is provided by the user,
1099                  * we allocate one.
1100                  */
1101                 if (n->nlmsg_flags & NLM_F_CREATE) {
1102                         prio = TC_H_MAKE(0x80000000U, 0U);
1103                         prio_allocate = true;
1104                 } else {
1105                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1106                         return -ENOENT;
1107                 }
1108         }
1109
1110         /* Find head of filter chain. */
1111
1112         block = tcf_block_find(net, &q, &parent, &cl,
1113                                t->tcm_ifindex, t->tcm_block_index, extack);
1114         if (IS_ERR(block)) {
1115                 err = PTR_ERR(block);
1116                 goto errout;
1117         }
1118
1119         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1120         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1121                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1122                 err = -EINVAL;
1123                 goto errout;
1124         }
1125         chain = tcf_chain_get(block, chain_index, true);
1126         if (!chain) {
1127                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1128                 err = -ENOMEM;
1129                 goto errout;
1130         }
1131
1132         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1133                                prio, prio_allocate);
1134         if (IS_ERR(tp)) {
1135                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1136                 err = PTR_ERR(tp);
1137                 goto errout;
1138         }
1139
1140         if (tp == NULL) {
1141                 /* Proto-tcf does not exist, create new one */
1142
1143                 if (tca[TCA_KIND] == NULL || !protocol) {
1144                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
1145                         err = -EINVAL;
1146                         goto errout;
1147                 }
1148
1149                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1150                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1151                         err = -ENOENT;
1152                         goto errout;
1153                 }
1154
1155                 if (prio_allocate)
1156                         prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
1157
1158                 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
1159                                       protocol, prio, chain, extack);
1160                 if (IS_ERR(tp)) {
1161                         err = PTR_ERR(tp);
1162                         goto errout;
1163                 }
1164                 tp_created = 1;
1165         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1166                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1167                 err = -EINVAL;
1168                 goto errout;
1169         }
1170
1171         fh = tp->ops->get(tp, t->tcm_handle);
1172
1173         if (!fh) {
1174                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1175                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1176                         err = -ENOENT;
1177                         goto errout;
1178                 }
1179         } else if (n->nlmsg_flags & NLM_F_EXCL) {
1180                 NL_SET_ERR_MSG(extack, "Filter already exists");
1181                 err = -EEXIST;
1182                 goto errout;
1183         }
1184
1185         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
1186                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
1187                               extack);
1188         if (err == 0) {
1189                 if (tp_created)
1190                         tcf_chain_tp_insert(chain, &chain_info, tp);
1191                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1192                                RTM_NEWTFILTER, false);
1193         } else {
1194                 if (tp_created)
1195                         tcf_proto_destroy(tp, NULL);
1196         }
1197
1198 errout:
1199         if (chain)
1200                 tcf_chain_put(chain);
1201         if (err == -EAGAIN)
1202                 /* Replay the request. */
1203                 goto replay;
1204         return err;
1205 }
1206
1207 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1208                           struct netlink_ext_ack *extack)
1209 {
1210         struct net *net = sock_net(skb->sk);
1211         struct nlattr *tca[TCA_MAX + 1];
1212         struct tcmsg *t;
1213         u32 protocol;
1214         u32 prio;
1215         u32 parent;
1216         u32 chain_index;
1217         struct Qdisc *q = NULL;
1218         struct tcf_chain_info chain_info;
1219         struct tcf_chain *chain = NULL;
1220         struct tcf_block *block;
1221         struct tcf_proto *tp = NULL;
1222         unsigned long cl = 0;
1223         void *fh = NULL;
1224         int err;
1225
1226         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1227                 return -EPERM;
1228
1229         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1230         if (err < 0)
1231                 return err;
1232
1233         t = nlmsg_data(n);
1234         protocol = TC_H_MIN(t->tcm_info);
1235         prio = TC_H_MAJ(t->tcm_info);
1236         parent = t->tcm_parent;
1237
1238         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
1239                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
1240                 return -ENOENT;
1241         }
1242
1243         /* Find head of filter chain. */
1244
1245         block = tcf_block_find(net, &q, &parent, &cl,
1246                                t->tcm_ifindex, t->tcm_block_index, extack);
1247         if (IS_ERR(block)) {
1248                 err = PTR_ERR(block);
1249                 goto errout;
1250         }
1251
1252         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1253         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1254                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1255                 err = -EINVAL;
1256                 goto errout;
1257         }
1258         chain = tcf_chain_get(block, chain_index, false);
1259         if (!chain) {
1260                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1261                 err = -EINVAL;
1262                 goto errout;
1263         }
1264
1265         if (prio == 0) {
1266                 tfilter_notify_chain(net, skb, block, q, parent, n,
1267                                      chain, RTM_DELTFILTER);
1268                 tcf_chain_flush(chain);
1269                 err = 0;
1270                 goto errout;
1271         }
1272
1273         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1274                                prio, false);
1275         if (!tp || IS_ERR(tp)) {
1276                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1277                 err = tp ? PTR_ERR(tp) : -ENOENT;
1278                 goto errout;
1279         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1280                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1281                 err = -EINVAL;
1282                 goto errout;
1283         }
1284
1285         fh = tp->ops->get(tp, t->tcm_handle);
1286
1287         if (!fh) {
1288                 if (t->tcm_handle == 0) {
1289                         tcf_chain_tp_remove(chain, &chain_info, tp);
1290                         tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1291                                        RTM_DELTFILTER, false);
1292                         tcf_proto_destroy(tp, extack);
1293                         err = 0;
1294                 } else {
1295                         NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1296                         err = -ENOENT;
1297                 }
1298         } else {
1299                 bool last;
1300
1301                 err = tfilter_del_notify(net, skb, n, tp, block,
1302                                          q, parent, fh, false, &last,
1303                                          extack);
1304                 if (err)
1305                         goto errout;
1306                 if (last) {
1307                         tcf_chain_tp_remove(chain, &chain_info, tp);
1308                         tcf_proto_destroy(tp, extack);
1309                 }
1310         }
1311
1312 errout:
1313         if (chain)
1314                 tcf_chain_put(chain);
1315         return err;
1316 }
1317
1318 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1319                           struct netlink_ext_ack *extack)
1320 {
1321         struct net *net = sock_net(skb->sk);
1322         struct nlattr *tca[TCA_MAX + 1];
1323         struct tcmsg *t;
1324         u32 protocol;
1325         u32 prio;
1326         u32 parent;
1327         u32 chain_index;
1328         struct Qdisc *q = NULL;
1329         struct tcf_chain_info chain_info;
1330         struct tcf_chain *chain = NULL;
1331         struct tcf_block *block;
1332         struct tcf_proto *tp = NULL;
1333         unsigned long cl = 0;
1334         void *fh = NULL;
1335         int err;
1336
1337         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1338         if (err < 0)
1339                 return err;
1340
1341         t = nlmsg_data(n);
1342         protocol = TC_H_MIN(t->tcm_info);
1343         prio = TC_H_MAJ(t->tcm_info);
1344         parent = t->tcm_parent;
1345
1346         if (prio == 0) {
1347                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1348                 return -ENOENT;
1349         }
1350
1351         /* Find head of filter chain. */
1352
1353         block = tcf_block_find(net, &q, &parent, &cl,
1354                                t->tcm_ifindex, t->tcm_block_index, extack);
1355         if (IS_ERR(block)) {
1356                 err = PTR_ERR(block);
1357                 goto errout;
1358         }
1359
1360         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1361         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1362                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1363                 err = -EINVAL;
1364                 goto errout;
1365         }
1366         chain = tcf_chain_get(block, chain_index, false);
1367         if (!chain) {
1368                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1369                 err = -EINVAL;
1370                 goto errout;
1371         }
1372
1373         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1374                                prio, false);
1375         if (!tp || IS_ERR(tp)) {
1376                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1377                 err = tp ? PTR_ERR(tp) : -ENOENT;
1378                 goto errout;
1379         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1380                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1381                 err = -EINVAL;
1382                 goto errout;
1383         }
1384
1385         fh = tp->ops->get(tp, t->tcm_handle);
1386
1387         if (!fh) {
1388                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1389                 err = -ENOENT;
1390         } else {
1391                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
1392                                      fh, RTM_NEWTFILTER, true);
1393                 if (err < 0)
1394                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
1395         }
1396
1397 errout:
1398         if (chain)
1399                 tcf_chain_put(chain);
1400         return err;
1401 }
1402
1403 struct tcf_dump_args {
1404         struct tcf_walker w;
1405         struct sk_buff *skb;
1406         struct netlink_callback *cb;
1407         struct tcf_block *block;
1408         struct Qdisc *q;
1409         u32 parent;
1410 };
1411
1412 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1413 {
1414         struct tcf_dump_args *a = (void *)arg;
1415         struct net *net = sock_net(a->skb->sk);
1416
1417         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
1418                              n, NETLINK_CB(a->cb->skb).portid,
1419                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1420                              RTM_NEWTFILTER);
1421 }
1422
1423 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
1424                            struct sk_buff *skb, struct netlink_callback *cb,
1425                            long index_start, long *p_index)
1426 {
1427         struct net *net = sock_net(skb->sk);
1428         struct tcf_block *block = chain->block;
1429         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1430         struct tcf_dump_args arg;
1431         struct tcf_proto *tp;
1432
1433         for (tp = rtnl_dereference(chain->filter_chain);
1434              tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
1435                 if (*p_index < index_start)
1436                         continue;
1437                 if (TC_H_MAJ(tcm->tcm_info) &&
1438                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
1439                         continue;
1440                 if (TC_H_MIN(tcm->tcm_info) &&
1441                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
1442                         continue;
1443                 if (*p_index > index_start)
1444                         memset(&cb->args[1], 0,
1445                                sizeof(cb->args) - sizeof(cb->args[0]));
1446                 if (cb->args[1] == 0) {
1447                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
1448                                           NETLINK_CB(cb->skb).portid,
1449                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
1450                                           RTM_NEWTFILTER) <= 0)
1451                                 return false;
1452
1453                         cb->args[1] = 1;
1454                 }
1455                 if (!tp->ops->walk)
1456                         continue;
1457                 arg.w.fn = tcf_node_dump;
1458                 arg.skb = skb;
1459                 arg.cb = cb;
1460                 arg.block = block;
1461                 arg.q = q;
1462                 arg.parent = parent;
1463                 arg.w.stop = 0;
1464                 arg.w.skip = cb->args[1] - 1;
1465                 arg.w.count = 0;
1466                 tp->ops->walk(tp, &arg.w);
1467                 cb->args[1] = arg.w.count + 1;
1468                 if (arg.w.stop)
1469                         return false;
1470         }
1471         return true;
1472 }
1473
1474 /* called with RTNL */
1475 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
1476 {
1477         struct net *net = sock_net(skb->sk);
1478         struct nlattr *tca[TCA_MAX + 1];
1479         struct Qdisc *q = NULL;
1480         struct tcf_block *block;
1481         struct tcf_chain *chain;
1482         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1483         long index_start;
1484         long index;
1485         u32 parent;
1486         int err;
1487
1488         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1489                 return skb->len;
1490
1491         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1492         if (err)
1493                 return err;
1494
1495         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1496                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1497                 if (!block)
1498                         goto out;
1499                 /* If we work with block index, q is NULL and parent value
1500                  * will never be used in the following code. The check
1501                  * in tcf_fill_node prevents it. However, compiler does not
1502                  * see that far, so set parent to zero to silence the warning
1503                  * about parent being uninitialized.
1504                  */
1505                 parent = 0;
1506         } else {
1507                 const struct Qdisc_class_ops *cops;
1508                 struct net_device *dev;
1509                 unsigned long cl = 0;
1510
1511                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1512                 if (!dev)
1513                         return skb->len;
1514
1515                 parent = tcm->tcm_parent;
1516                 if (!parent) {
1517                         q = dev->qdisc;
1518                         parent = q->handle;
1519                 } else {
1520                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1521                 }
1522                 if (!q)
1523                         goto out;
1524                 cops = q->ops->cl_ops;
1525                 if (!cops)
1526                         goto out;
1527                 if (!cops->tcf_block)
1528                         goto out;
1529                 if (TC_H_MIN(tcm->tcm_parent)) {
1530                         cl = cops->find(q, tcm->tcm_parent);
1531                         if (cl == 0)
1532                                 goto out;
1533                 }
1534                 block = cops->tcf_block(q, cl, NULL);
1535                 if (!block)
1536                         goto out;
1537                 if (tcf_block_shared(block))
1538                         q = NULL;
1539         }
1540
1541         index_start = cb->args[0];
1542         index = 0;
1543
1544         list_for_each_entry(chain, &block->chain_list, list) {
1545                 if (tca[TCA_CHAIN] &&
1546                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1547                         continue;
1548                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
1549                                     index_start, &index)) {
1550                         err = -EMSGSIZE;
1551                         break;
1552                 }
1553         }
1554
1555         cb->args[0] = index;
1556
1557 out:
1558         /* If we did no progress, the error (EMSGSIZE) is real */
1559         if (skb->len == 0 && err)
1560                 return err;
1561         return skb->len;
1562 }
1563
1564 void tcf_exts_destroy(struct tcf_exts *exts)
1565 {
1566 #ifdef CONFIG_NET_CLS_ACT
1567         LIST_HEAD(actions);
1568
1569         ASSERT_RTNL();
1570         tcf_exts_to_list(exts, &actions);
1571         tcf_action_destroy(&actions, TCA_ACT_UNBIND);
1572         kfree(exts->actions);
1573         exts->nr_actions = 0;
1574 #endif
1575 }
1576 EXPORT_SYMBOL(tcf_exts_destroy);
1577
1578 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
1579                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
1580                       struct netlink_ext_ack *extack)
1581 {
1582 #ifdef CONFIG_NET_CLS_ACT
1583         {
1584                 struct tc_action *act;
1585                 size_t attr_size = 0;
1586
1587                 if (exts->police && tb[exts->police]) {
1588                         act = tcf_action_init_1(net, tp, tb[exts->police],
1589                                                 rate_tlv, "police", ovr,
1590                                                 TCA_ACT_BIND, extack);
1591                         if (IS_ERR(act))
1592                                 return PTR_ERR(act);
1593
1594                         act->type = exts->type = TCA_OLD_COMPAT;
1595                         exts->actions[0] = act;
1596                         exts->nr_actions = 1;
1597                 } else if (exts->action && tb[exts->action]) {
1598                         LIST_HEAD(actions);
1599                         int err, i = 0;
1600
1601                         err = tcf_action_init(net, tp, tb[exts->action],
1602                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
1603                                               &actions, &attr_size, extack);
1604                         if (err)
1605                                 return err;
1606                         list_for_each_entry(act, &actions, list)
1607                                 exts->actions[i++] = act;
1608                         exts->nr_actions = i;
1609                 }
1610                 exts->net = net;
1611         }
1612 #else
1613         if ((exts->action && tb[exts->action]) ||
1614             (exts->police && tb[exts->police])) {
1615                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
1616                 return -EOPNOTSUPP;
1617         }
1618 #endif
1619
1620         return 0;
1621 }
1622 EXPORT_SYMBOL(tcf_exts_validate);
1623
1624 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
1625 {
1626 #ifdef CONFIG_NET_CLS_ACT
1627         struct tcf_exts old = *dst;
1628
1629         *dst = *src;
1630         tcf_exts_destroy(&old);
1631 #endif
1632 }
1633 EXPORT_SYMBOL(tcf_exts_change);
1634
1635 #ifdef CONFIG_NET_CLS_ACT
1636 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
1637 {
1638         if (exts->nr_actions == 0)
1639                 return NULL;
1640         else
1641                 return exts->actions[0];
1642 }
1643 #endif
1644
1645 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
1646 {
1647 #ifdef CONFIG_NET_CLS_ACT
1648         struct nlattr *nest;
1649
1650         if (exts->action && tcf_exts_has_actions(exts)) {
1651                 /*
1652                  * again for backward compatible mode - we want
1653                  * to work with both old and new modes of entering
1654                  * tc data even if iproute2  was newer - jhs
1655                  */
1656                 if (exts->type != TCA_OLD_COMPAT) {
1657                         LIST_HEAD(actions);
1658
1659                         nest = nla_nest_start(skb, exts->action);
1660                         if (nest == NULL)
1661                                 goto nla_put_failure;
1662
1663                         tcf_exts_to_list(exts, &actions);
1664                         if (tcf_action_dump(skb, &actions, 0, 0) < 0)
1665                                 goto nla_put_failure;
1666                         nla_nest_end(skb, nest);
1667                 } else if (exts->police) {
1668                         struct tc_action *act = tcf_exts_first_act(exts);
1669                         nest = nla_nest_start(skb, exts->police);
1670                         if (nest == NULL || !act)
1671                                 goto nla_put_failure;
1672                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
1673                                 goto nla_put_failure;
1674                         nla_nest_end(skb, nest);
1675                 }
1676         }
1677         return 0;
1678
1679 nla_put_failure:
1680         nla_nest_cancel(skb, nest);
1681         return -1;
1682 #else
1683         return 0;
1684 #endif
1685 }
1686 EXPORT_SYMBOL(tcf_exts_dump);
1687
1688
1689 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1690 {
1691 #ifdef CONFIG_NET_CLS_ACT
1692         struct tc_action *a = tcf_exts_first_act(exts);
1693         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
1694                 return -1;
1695 #endif
1696         return 0;
1697 }
1698 EXPORT_SYMBOL(tcf_exts_dump_stats);
1699
1700 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
1701                                        enum tc_setup_type type,
1702                                        void *type_data, bool err_stop)
1703 {
1704         int ok_count = 0;
1705 #ifdef CONFIG_NET_CLS_ACT
1706         const struct tc_action *a;
1707         struct net_device *dev;
1708         int i, ret;
1709
1710         if (!tcf_exts_has_actions(exts))
1711                 return 0;
1712
1713         for (i = 0; i < exts->nr_actions; i++) {
1714                 a = exts->actions[i];
1715                 if (!a->ops->get_dev)
1716                         continue;
1717                 dev = a->ops->get_dev(a);
1718                 if (!dev)
1719                         continue;
1720                 ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
1721                 if (ret < 0)
1722                         return ret;
1723                 ok_count += ret;
1724         }
1725 #endif
1726         return ok_count;
1727 }
1728
1729 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
1730                      enum tc_setup_type type, void *type_data, bool err_stop)
1731 {
1732         int ok_count;
1733         int ret;
1734
1735         ret = tcf_block_cb_call(block, type, type_data, err_stop);
1736         if (ret < 0)
1737                 return ret;
1738         ok_count = ret;
1739
1740         if (!exts || ok_count)
1741                 return ok_count;
1742         ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
1743         if (ret < 0)
1744                 return ret;
1745         ok_count += ret;
1746
1747         return ok_count;
1748 }
1749 EXPORT_SYMBOL(tc_setup_cb_call);
1750
1751 static __net_init int tcf_net_init(struct net *net)
1752 {
1753         struct tcf_net *tn = net_generic(net, tcf_net_id);
1754
1755         idr_init(&tn->idr);
1756         return 0;
1757 }
1758
1759 static void __net_exit tcf_net_exit(struct net *net)
1760 {
1761         struct tcf_net *tn = net_generic(net, tcf_net_id);
1762
1763         idr_destroy(&tn->idr);
1764 }
1765
1766 static struct pernet_operations tcf_net_ops = {
1767         .init = tcf_net_init,
1768         .exit = tcf_net_exit,
1769         .id   = &tcf_net_id,
1770         .size = sizeof(struct tcf_net),
1771 };
1772
1773 static int __init tc_filter_init(void)
1774 {
1775         int err;
1776
1777         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
1778         if (!tc_filter_wq)
1779                 return -ENOMEM;
1780
1781         err = register_pernet_subsys(&tcf_net_ops);
1782         if (err)
1783                 goto err_register_pernet_subsys;
1784
1785         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 0);
1786         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 0);
1787         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
1788                       tc_dump_tfilter, 0);
1789
1790         return 0;
1791
1792 err_register_pernet_subsys:
1793         destroy_workqueue(tc_filter_wq);
1794         return err;
1795 }
1796
1797 subsys_initcall(tc_filter_init);