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 static const struct tcf_proto_ops *
  61 tcf_proto_lookup_ops(const char *kind, struct netlink_ext_ack *extack)
  62 {
  63         const struct tcf_proto_ops *ops;
  64
  65         ops = __tcf_proto_lookup_ops(kind);
  66         if (ops)
  67                 return ops;
  68 #ifdef CONFIG_MODULES
  69         rtnl_unlock();
  70         request_module("cls_%s", kind);
  71         rtnl_lock();
  72         ops = __tcf_proto_lookup_ops(kind);
  73         /* We dropped the RTNL semaphore in order to perform
  74          * the module load. So, even if we succeeded in loading
  75          * the module we have to replay the request. We indicate
  76          * this using -EAGAIN.
  77          */
  78         if (ops) {
  79                 module_put(ops->owner);
  80                 return ERR_PTR(-EAGAIN);
  81         }
  82 #endif
  83         NL_SET_ERR_MSG(extack, "TC classifier not found");
  84         return ERR_PTR(-ENOENT);
  85 }
  86
  87 /* Register(unregister) new classifier type */
  88
  89 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  90 {
  91         struct tcf_proto_ops *t;
  92         int rc = -EEXIST;
  93
  94         write_lock(&cls_mod_lock);
  95         list_for_each_entry(t, &tcf_proto_base, head)
  96                 if (!strcmp(ops->kind, t->kind))
  97                         goto out;
  98
  99         list_add_tail(&ops->head, &tcf_proto_base);
 100         rc = 0;
 101 out:
 102         write_unlock(&cls_mod_lock);
 103         return rc;
 104 }
 105 EXPORT_SYMBOL(register_tcf_proto_ops);
 106
 107 static struct workqueue_struct *tc_filter_wq;
 108
 109 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 110 {
 111         struct tcf_proto_ops *t;
 112         int rc = -ENOENT;
 113
 114         /* Wait for outstanding call_rcu()s, if any, from a
 115          * tcf_proto_ops's destroy() handler.
 116          */
 117         rcu_barrier();
 118         flush_workqueue(tc_filter_wq);
 119
 120         write_lock(&cls_mod_lock);
 121         list_for_each_entry(t, &tcf_proto_base, head) {
 122                 if (t == ops) {
 123                         list_del(&t->head);
 124                         rc = 0;
 125                         break;
 126                 }
 127         }
 128         write_unlock(&cls_mod_lock);
 129         return rc;
 130 }
 131 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 132
 133 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 134 {
 135         INIT_RCU_WORK(rwork, func);
 136         return queue_rcu_work(tc_filter_wq, rwork);
 137 }
 138 EXPORT_SYMBOL(tcf_queue_work);
 139
 140 /* Select new prio value from the range, managed by kernel. */
 141
 142 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 143 {
 144         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 145
 146         if (tp)
 147                 first = tp->prio - 1;
 148
 149         return TC_H_MAJ(first);
 150 }
 151
 152 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 153                                           u32 prio, struct tcf_chain *chain,
 154                                           struct netlink_ext_ack *extack)
 155 {
 156         struct tcf_proto *tp;
 157         int err;
 158
 159         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 160         if (!tp)
 161                 return ERR_PTR(-ENOBUFS);
 162
 163         tp->ops = tcf_proto_lookup_ops(kind, extack);
 164         if (IS_ERR(tp->ops)) {
 165                 err = PTR_ERR(tp->ops);
 166                 goto errout;
 167         }
 168         tp->classify = tp->ops->classify;
 169         tp->protocol = protocol;
 170         tp->prio = prio;
 171         tp->chain = chain;
 172
 173         err = tp->ops->init(tp);
 174         if (err) {
 175                 module_put(tp->ops->owner);
 176                 goto errout;
 177         }
 178         return tp;
 179
 180 errout:
 181         kfree(tp);
 182         return ERR_PTR(err);
 183 }
 184
 185 static void tcf_proto_destroy(struct tcf_proto *tp,
 186                               struct netlink_ext_ack *extack)
 187 {
 188         tp->ops->destroy(tp, extack);
 189         module_put(tp->ops->owner);
 190         kfree_rcu(tp, rcu);
 191 }
 192
 193 struct tcf_filter_chain_list_item {
 194         struct list_head list;
 195         tcf_chain_head_change_t *chain_head_change;
 196         void *chain_head_change_priv;
 197 };
 198
 199 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 200                                           u32 chain_index)
 201 {
 202         struct tcf_chain *chain;
 203
 204         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 205         if (!chain)
 206                 return NULL;
 207         list_add_tail(&chain->list, &block->chain_list);
 208         chain->block = block;
 209         chain->index = chain_index;
 210         chain->refcnt = 1;
 211         if (!chain->index)
 212                 block->chain0.chain = chain;
 213         return chain;
 214 }
 215
 216 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 217                                        struct tcf_proto *tp_head)
 218 {
 219         if (item->chain_head_change)
 220                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 221 }
 222
 223 static void tcf_chain0_head_change(struct tcf_chain *chain,
 224                                    struct tcf_proto *tp_head)
 225 {
 226         struct tcf_filter_chain_list_item *item;
 227         struct tcf_block *block = chain->block;
 228
 229         if (chain->index)
 230                 return;
 231         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 232                 tcf_chain_head_change_item(item, tp_head);
 233 }
 234
 235 static void tcf_chain_flush(struct tcf_chain *chain)
 236 {
 237         struct tcf_proto *tp = rtnl_dereference(chain->filter_chain);
 238
 239         tcf_chain0_head_change(chain, NULL);
 240         while (tp) {
 241                 RCU_INIT_POINTER(chain->filter_chain, tp->next);
 242                 tcf_proto_destroy(tp, NULL);
 243                 tp = rtnl_dereference(chain->filter_chain);
 244                 tcf_chain_put(chain);
 245         }
 246 }
 247
 248 static void tcf_chain_destroy(struct tcf_chain *chain)
 249 {
 250         struct tcf_block *block = chain->block;
 251
 252         list_del(&chain->list);
 253         if (!chain->index)
 254                 block->chain0.chain = NULL;
 255         kfree(chain);
 256         if (list_empty(&block->chain_list) && block->refcnt == 0)
 257                 kfree(block);
 258 }
 259
 260 static void tcf_chain_hold(struct tcf_chain *chain)
 261 {
 262         ++chain->refcnt;
 263 }
 264
 265 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 266                                           u32 chain_index)
 267 {
 268         struct tcf_chain *chain;
 269
 270         list_for_each_entry(chain, &block->chain_list, list) {
 271                 if (chain->index == chain_index)
 272                         return chain;
 273         }
 274         return NULL;
 275 }
 276
 277 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 278                            u32 seq, u16 flags, int event, bool unicast);
 279
 280 struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 281                                 bool create)
 282 {
 283         struct tcf_chain *chain = tcf_chain_lookup(block, chain_index);
 284
 285         if (chain) {
 286                 tcf_chain_hold(chain);
 287                 return chain;
 288         }
 289
 290         if (!create)
 291                 return NULL;
 292         chain = tcf_chain_create(block, chain_index);
 293         if (!chain)
 294                 return NULL;
 295         tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 296                         RTM_NEWCHAIN, false);
 297         return chain;
 298 }
 299 EXPORT_SYMBOL(tcf_chain_get);
 300
 301 static void tc_chain_tmplt_del(struct tcf_chain *chain);
 302
 303 void tcf_chain_put(struct tcf_chain *chain)
 304 {
 305         if (--chain->refcnt == 0) {
 306                 tc_chain_notify(chain, NULL, 0, 0, RTM_DELCHAIN, false);
 307                 tc_chain_tmplt_del(chain);
 308                 tcf_chain_destroy(chain);
 309         }
 310 }
 311 EXPORT_SYMBOL(tcf_chain_put);
 312
 313 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 314 {
 315         if (chain->explicitly_created)
 316                 tcf_chain_put(chain);
 317 }
 318
 319 static bool tcf_block_offload_in_use(struct tcf_block *block)
 320 {
 321         return block->offloadcnt;
 322 }
 323
 324 static int tcf_block_offload_cmd(struct tcf_block *block,
 325                                  struct net_device *dev,
 326                                  struct tcf_block_ext_info *ei,
 327                                  enum tc_block_command command,
 328                                  struct netlink_ext_ack *extack)
 329 {
 330         struct tc_block_offload bo = {};
 331
 332         bo.command = command;
 333         bo.binder_type = ei->binder_type;
 334         bo.block = block;
 335         bo.extack = extack;
 336         return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 337 }
 338
 339 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 340                                   struct tcf_block_ext_info *ei,
 341                                   struct netlink_ext_ack *extack)
 342 {
 343         struct net_device *dev = q->dev_queue->dev;
 344         int err;
 345
 346         if (!dev->netdev_ops->ndo_setup_tc)
 347                 goto no_offload_dev_inc;
 348
 349         /* If tc offload feature is disabled and the block we try to bind
 350          * to already has some offloaded filters, forbid to bind.
 351          */
 352         if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
 353                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 354                 return -EOPNOTSUPP;
 355         }
 356
 357         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_BIND, extack);
 358         if (err == -EOPNOTSUPP)
 359                 goto no_offload_dev_inc;
 360         return err;
 361
 362 no_offload_dev_inc:
 363         if (tcf_block_offload_in_use(block))
 364                 return -EOPNOTSUPP;
 365         block->nooffloaddevcnt++;
 366         return 0;
 367 }
 368
 369 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 370                                      struct tcf_block_ext_info *ei)
 371 {
 372         struct net_device *dev = q->dev_queue->dev;
 373         int err;
 374
 375         if (!dev->netdev_ops->ndo_setup_tc)
 376                 goto no_offload_dev_dec;
 377         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_UNBIND, NULL);
 378         if (err == -EOPNOTSUPP)
 379                 goto no_offload_dev_dec;
 380         return;
 381
 382 no_offload_dev_dec:
 383         WARN_ON(block->nooffloaddevcnt-- == 0);
 384 }
 385
 386 static int
 387 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 388                               struct tcf_block_ext_info *ei,
 389                               struct netlink_ext_ack *extack)
 390 {
 391         struct tcf_chain *chain0 = block->chain0.chain;
 392         struct tcf_filter_chain_list_item *item;
 393
 394         item = kmalloc(sizeof(*item), GFP_KERNEL);
 395         if (!item) {
 396                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 397                 return -ENOMEM;
 398         }
 399         item->chain_head_change = ei->chain_head_change;
 400         item->chain_head_change_priv = ei->chain_head_change_priv;
 401         if (chain0 && chain0->filter_chain)
 402                 tcf_chain_head_change_item(item, chain0->filter_chain);
 403         list_add(&item->list, &block->chain0.filter_chain_list);
 404         return 0;
 405 }
 406
 407 static void
 408 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 409                               struct tcf_block_ext_info *ei)
 410 {
 411         struct tcf_chain *chain0 = block->chain0.chain;
 412         struct tcf_filter_chain_list_item *item;
 413
 414         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 415                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 416                     (item->chain_head_change == ei->chain_head_change &&
 417                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 418                         if (chain0)
 419                                 tcf_chain_head_change_item(item, NULL);
 420                         list_del(&item->list);
 421                         kfree(item);
 422                         return;
 423                 }
 424         }
 425         WARN_ON(1);
 426 }
 427
 428 struct tcf_net {
 429         struct idr idr;
 430 };
 431
 432 static unsigned int tcf_net_id;
 433
 434 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 435                             struct netlink_ext_ack *extack)
 436 {
 437         struct tcf_net *tn = net_generic(net, tcf_net_id);
 438
 439         return idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 440                              GFP_KERNEL);
 441 }
 442
 443 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 444 {
 445         struct tcf_net *tn = net_generic(net, tcf_net_id);
 446
 447         idr_remove(&tn->idr, block->index);
 448 }
 449
 450 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 451                                           u32 block_index,
 452                                           struct netlink_ext_ack *extack)
 453 {
 454         struct tcf_block *block;
 455
 456         block = kzalloc(sizeof(*block), GFP_KERNEL);
 457         if (!block) {
 458                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 459                 return ERR_PTR(-ENOMEM);
 460         }
 461         INIT_LIST_HEAD(&block->chain_list);
 462         INIT_LIST_HEAD(&block->cb_list);
 463         INIT_LIST_HEAD(&block->owner_list);
 464         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 465
 466         block->refcnt = 1;
 467         block->net = net;
 468         block->index = block_index;
 469
 470         /* Don't store q pointer for blocks which are shared */
 471         if (!tcf_block_shared(block))
 472                 block->q = q;
 473         return block;
 474 }
 475
 476 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 477 {
 478         struct tcf_net *tn = net_generic(net, tcf_net_id);
 479
 480         return idr_find(&tn->idr, block_index);
 481 }
 482
 483 /* Find tcf block.
 484  * Set q, parent, cl when appropriate.
 485  */
 486
 487 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
 488                                         u32 *parent, unsigned long *cl,
 489                                         int ifindex, u32 block_index,
 490                                         struct netlink_ext_ack *extack)
 491 {
 492         struct tcf_block *block;
 493
 494         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
 495                 block = tcf_block_lookup(net, block_index);
 496                 if (!block) {
 497                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
 498                         return ERR_PTR(-EINVAL);
 499                 }
 500         } else {
 501                 const struct Qdisc_class_ops *cops;
 502                 struct net_device *dev;
 503
 504                 /* Find link */
 505                 dev = __dev_get_by_index(net, ifindex);
 506                 if (!dev)
 507                         return ERR_PTR(-ENODEV);
 508
 509                 /* Find qdisc */
 510                 if (!*parent) {
 511                         *q = dev->qdisc;
 512                         *parent = (*q)->handle;
 513                 } else {
 514                         *q = qdisc_lookup(dev, TC_H_MAJ(*parent));
 515                         if (!*q) {
 516                                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
 517                                 return ERR_PTR(-EINVAL);
 518                         }
 519                 }
 520
 521                 /* Is it classful? */
 522                 cops = (*q)->ops->cl_ops;
 523                 if (!cops) {
 524                         NL_SET_ERR_MSG(extack, "Qdisc not classful");
 525                         return ERR_PTR(-EINVAL);
 526                 }
 527
 528                 if (!cops->tcf_block) {
 529                         NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
 530                         return ERR_PTR(-EOPNOTSUPP);
 531                 }
 532
 533                 /* Do we search for filter, attached to class? */
 534                 if (TC_H_MIN(*parent)) {
 535                         *cl = cops->find(*q, *parent);
 536                         if (*cl == 0) {
 537                                 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
 538                                 return ERR_PTR(-ENOENT);
 539                         }
 540                 }
 541
 542                 /* And the last stroke */
 543                 block = cops->tcf_block(*q, *cl, extack);
 544                 if (!block)
 545                         return ERR_PTR(-EINVAL);
 546                 if (tcf_block_shared(block)) {
 547                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
 548                         return ERR_PTR(-EOPNOTSUPP);
 549                 }
 550         }
 551
 552         return block;
 553 }
 554
 555 struct tcf_block_owner_item {
 556         struct list_head list;
 557         struct Qdisc *q;
 558         enum tcf_block_binder_type binder_type;
 559 };
 560
 561 static void
 562 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
 563                                struct Qdisc *q,
 564                                enum tcf_block_binder_type binder_type)
 565 {
 566         if (block->keep_dst &&
 567             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
 568             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
 569                 netif_keep_dst(qdisc_dev(q));
 570 }
 571
 572 void tcf_block_netif_keep_dst(struct tcf_block *block)
 573 {
 574         struct tcf_block_owner_item *item;
 575
 576         block->keep_dst = true;
 577         list_for_each_entry(item, &block->owner_list, list)
 578                 tcf_block_owner_netif_keep_dst(block, item->q,
 579                                                item->binder_type);
 580 }
 581 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
 582
 583 static int tcf_block_owner_add(struct tcf_block *block,
 584                                struct Qdisc *q,
 585                                enum tcf_block_binder_type binder_type)
 586 {
 587         struct tcf_block_owner_item *item;
 588
 589         item = kmalloc(sizeof(*item), GFP_KERNEL);
 590         if (!item)
 591                 return -ENOMEM;
 592         item->q = q;
 593         item->binder_type = binder_type;
 594         list_add(&item->list, &block->owner_list);
 595         return 0;
 596 }
 597
 598 static void tcf_block_owner_del(struct tcf_block *block,
 599                                 struct Qdisc *q,
 600                                 enum tcf_block_binder_type binder_type)
 601 {
 602         struct tcf_block_owner_item *item;
 603
 604         list_for_each_entry(item, &block->owner_list, list) {
 605                 if (item->q == q && item->binder_type == binder_type) {
 606                         list_del(&item->list);
 607                         kfree(item);
 608                         return;
 609                 }
 610         }
 611         WARN_ON(1);
 612 }
 613
 614 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 615                       struct tcf_block_ext_info *ei,
 616                       struct netlink_ext_ack *extack)
 617 {
 618         struct net *net = qdisc_net(q);
 619         struct tcf_block *block = NULL;
 620         bool created = false;
 621         int err;
 622
 623         if (ei->block_index) {
 624                 /* block_index not 0 means the shared block is requested */
 625                 block = tcf_block_lookup(net, ei->block_index);
 626                 if (block)
 627                         block->refcnt++;
 628         }
 629
 630         if (!block) {
 631                 block = tcf_block_create(net, q, ei->block_index, extack);
 632                 if (IS_ERR(block))
 633                         return PTR_ERR(block);
 634                 created = true;
 635                 if (tcf_block_shared(block)) {
 636                         err = tcf_block_insert(block, net, extack);
 637                         if (err)
 638                                 goto err_block_insert;
 639                 }
 640         }
 641
 642         err = tcf_block_owner_add(block, q, ei->binder_type);
 643         if (err)
 644                 goto err_block_owner_add;
 645
 646         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
 647
 648         err = tcf_chain0_head_change_cb_add(block, ei, extack);
 649         if (err)
 650                 goto err_chain0_head_change_cb_add;
 651
 652         err = tcf_block_offload_bind(block, q, ei, extack);
 653         if (err)
 654                 goto err_block_offload_bind;
 655
 656         *p_block = block;
 657         return 0;
 658
 659 err_block_offload_bind:
 660         tcf_chain0_head_change_cb_del(block, ei);
 661 err_chain0_head_change_cb_add:
 662         tcf_block_owner_del(block, q, ei->binder_type);
 663 err_block_owner_add:
 664         if (created) {
 665                 if (tcf_block_shared(block))
 666                         tcf_block_remove(block, net);
 667 err_block_insert:
 668                 kfree(block);
 669         } else {
 670                 block->refcnt--;
 671         }
 672         return err;
 673 }
 674 EXPORT_SYMBOL(tcf_block_get_ext);
 675
 676 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
 677 {
 678         struct tcf_proto __rcu **p_filter_chain = priv;
 679
 680         rcu_assign_pointer(*p_filter_chain, tp_head);
 681 }
 682
 683 int tcf_block_get(struct tcf_block **p_block,
 684                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
 685                   struct netlink_ext_ack *extack)
 686 {
 687         struct tcf_block_ext_info ei = {
 688                 .chain_head_change = tcf_chain_head_change_dflt,
 689                 .chain_head_change_priv = p_filter_chain,
 690         };
 691
 692         WARN_ON(!p_filter_chain);
 693         return tcf_block_get_ext(p_block, q, &ei, extack);
 694 }
 695 EXPORT_SYMBOL(tcf_block_get);
 696
 697 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
 698  * actions should be all removed after flushing.
 699  */
 700 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 701                        struct tcf_block_ext_info *ei)
 702 {
 703         struct tcf_chain *chain, *tmp;
 704
 705         if (!block)
 706                 return;
 707         tcf_chain0_head_change_cb_del(block, ei);
 708         tcf_block_owner_del(block, q, ei->binder_type);
 709
 710         if (block->refcnt == 1) {
 711                 if (tcf_block_shared(block))
 712                         tcf_block_remove(block, block->net);
 713
 714                 /* Hold a refcnt for all chains, so that they don't disappear
 715                  * while we are iterating.
 716                  */
 717                 list_for_each_entry(chain, &block->chain_list, list)
 718                         tcf_chain_hold(chain);
 719
 720                 list_for_each_entry(chain, &block->chain_list, list)
 721                         tcf_chain_flush(chain);
 722         }
 723
 724         tcf_block_offload_unbind(block, q, ei);
 725
 726         if (block->refcnt == 1) {
 727                 /* At this point, all the chains should have refcnt >= 1. */
 728                 list_for_each_entry_safe(chain, tmp, &block->chain_list, list) {
 729                         tcf_chain_put_explicitly_created(chain);
 730                         tcf_chain_put(chain);
 731                 }
 732
 733                 block->refcnt--;
 734                 if (list_empty(&block->chain_list))
 735                         kfree(block);
 736         }
 737 }
 738 EXPORT_SYMBOL(tcf_block_put_ext);
 739
 740 void tcf_block_put(struct tcf_block *block)
 741 {
 742         struct tcf_block_ext_info ei = {0, };
 743
 744         if (!block)
 745                 return;
 746         tcf_block_put_ext(block, block->q, &ei);
 747 }
 748
 749 EXPORT_SYMBOL(tcf_block_put);
 750
 751 struct tcf_block_cb {
 752         struct list_head list;
 753         tc_setup_cb_t *cb;
 754         void *cb_ident;
 755         void *cb_priv;
 756         unsigned int refcnt;
 757 };
 758
 759 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
 760 {
 761         return block_cb->cb_priv;
 762 }
 763 EXPORT_SYMBOL(tcf_block_cb_priv);
 764
 765 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
 766                                          tc_setup_cb_t *cb, void *cb_ident)
 767 {       struct tcf_block_cb *block_cb;
 768
 769         list_for_each_entry(block_cb, &block->cb_list, list)
 770                 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
 771                         return block_cb;
 772         return NULL;
 773 }
 774 EXPORT_SYMBOL(tcf_block_cb_lookup);
 775
 776 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
 777 {
 778         block_cb->refcnt++;
 779 }
 780 EXPORT_SYMBOL(tcf_block_cb_incref);
 781
 782 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
 783 {
 784         return --block_cb->refcnt;
 785 }
 786 EXPORT_SYMBOL(tcf_block_cb_decref);
 787
 788 static int
 789 tcf_block_playback_offloads(struct tcf_block *block, tc_setup_cb_t *cb,
 790                             void *cb_priv, bool add, bool offload_in_use,
 791                             struct netlink_ext_ack *extack)
 792 {
 793         struct tcf_chain *chain;
 794         struct tcf_proto *tp;
 795         int err;
 796
 797         list_for_each_entry(chain, &block->chain_list, list) {
 798                 for (tp = rtnl_dereference(chain->filter_chain); tp;
 799                      tp = rtnl_dereference(tp->next)) {
 800                         if (tp->ops->reoffload) {
 801                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
 802                                                          extack);
 803                                 if (err && add)
 804                                         goto err_playback_remove;
 805                         } else if (add && offload_in_use) {
 806                                 err = -EOPNOTSUPP;
 807                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
 808                                 goto err_playback_remove;
 809                         }
 810                 }
 811         }
 812
 813         return 0;
 814
 815 err_playback_remove:
 816         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
 817                                     extack);
 818         return err;
 819 }
 820
 821 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
 822                                              tc_setup_cb_t *cb, void *cb_ident,
 823                                              void *cb_priv,
 824                                              struct netlink_ext_ack *extack)
 825 {
 826         struct tcf_block_cb *block_cb;
 827         int err;
 828
 829         /* Replay any already present rules */
 830         err = tcf_block_playback_offloads(block, cb, cb_priv, true,
 831                                           tcf_block_offload_in_use(block),
 832                                           extack);
 833         if (err)
 834                 return ERR_PTR(err);
 835
 836         block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
 837         if (!block_cb)
 838                 return ERR_PTR(-ENOMEM);
 839         block_cb->cb = cb;
 840         block_cb->cb_ident = cb_ident;
 841         block_cb->cb_priv = cb_priv;
 842         list_add(&block_cb->list, &block->cb_list);
 843         return block_cb;
 844 }
 845 EXPORT_SYMBOL(__tcf_block_cb_register);
 846
 847 int tcf_block_cb_register(struct tcf_block *block,
 848                           tc_setup_cb_t *cb, void *cb_ident,
 849                           void *cb_priv, struct netlink_ext_ack *extack)
 850 {
 851         struct tcf_block_cb *block_cb;
 852
 853         block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv,
 854                                            extack);
 855         return PTR_ERR_OR_ZERO(block_cb);
 856 }
 857 EXPORT_SYMBOL(tcf_block_cb_register);
 858
 859 void __tcf_block_cb_unregister(struct tcf_block *block,
 860                                struct tcf_block_cb *block_cb)
 861 {
 862         tcf_block_playback_offloads(block, block_cb->cb, block_cb->cb_priv,
 863                                     false, tcf_block_offload_in_use(block),
 864                                     NULL);
 865         list_del(&block_cb->list);
 866         kfree(block_cb);
 867 }
 868 EXPORT_SYMBOL(__tcf_block_cb_unregister);
 869
 870 void tcf_block_cb_unregister(struct tcf_block *block,
 871                              tc_setup_cb_t *cb, void *cb_ident)
 872 {
 873         struct tcf_block_cb *block_cb;
 874
 875         block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
 876         if (!block_cb)
 877                 return;
 878         __tcf_block_cb_unregister(block, block_cb);
 879 }
 880 EXPORT_SYMBOL(tcf_block_cb_unregister);
 881
 882 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type,
 883                              void *type_data, bool err_stop)
 884 {
 885         struct tcf_block_cb *block_cb;
 886         int ok_count = 0;
 887         int err;
 888
 889         /* Make sure all netdevs sharing this block are offload-capable. */
 890         if (block->nooffloaddevcnt && err_stop)
 891                 return -EOPNOTSUPP;
 892
 893         list_for_each_entry(block_cb, &block->cb_list, list) {
 894                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
 895                 if (err) {
 896                         if (err_stop)
 897                                 return err;
 898                 } else {
 899                         ok_count++;
 900                 }
 901         }
 902         return ok_count;
 903 }
 904
 905 /* Main classifier routine: scans classifier chain attached
 906  * to this qdisc, (optionally) tests for protocol and asks
 907  * specific classifiers.
 908  */
 909 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 910                  struct tcf_result *res, bool compat_mode)
 911 {
 912         __be16 protocol = tc_skb_protocol(skb);
 913 #ifdef CONFIG_NET_CLS_ACT
 914         const int max_reclassify_loop = 4;
 915         const struct tcf_proto *orig_tp = tp;
 916         const struct tcf_proto *first_tp;
 917         int limit = 0;
 918
 919 reclassify:
 920 #endif
 921         for (; tp; tp = rcu_dereference_bh(tp->next)) {
 922                 int err;
 923
 924                 if (tp->protocol != protocol &&
 925                     tp->protocol != htons(ETH_P_ALL))
 926                         continue;
 927
 928                 err = tp->classify(skb, tp, res);
 929 #ifdef CONFIG_NET_CLS_ACT
 930                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
 931                         first_tp = orig_tp;
 932                         goto reset;
 933                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
 934                         first_tp = res->goto_tp;
 935                         goto reset;
 936                 }
 937 #endif
 938                 if (err >= 0)
 939                         return err;
 940         }
 941
 942         return TC_ACT_UNSPEC; /* signal: continue lookup */
 943 #ifdef CONFIG_NET_CLS_ACT
 944 reset:
 945         if (unlikely(limit++ >= max_reclassify_loop)) {
 946                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
 947                                        tp->chain->block->index,
 948                                        tp->prio & 0xffff,
 949                                        ntohs(tp->protocol));
 950                 return TC_ACT_SHOT;
 951         }
 952
 953         tp = first_tp;
 954         protocol = tc_skb_protocol(skb);
 955         goto reclassify;
 956 #endif
 957 }
 958 EXPORT_SYMBOL(tcf_classify);
 959
 960 struct tcf_chain_info {
 961         struct tcf_proto __rcu **pprev;
 962         struct tcf_proto __rcu *next;
 963 };
 964
 965 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
 966 {
 967         return rtnl_dereference(*chain_info->pprev);
 968 }
 969
 970 static void tcf_chain_tp_insert(struct tcf_chain *chain,
 971                                 struct tcf_chain_info *chain_info,
 972                                 struct tcf_proto *tp)
 973 {
 974         if (*chain_info->pprev == chain->filter_chain)
 975                 tcf_chain0_head_change(chain, tp);
 976         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
 977         rcu_assign_pointer(*chain_info->pprev, tp);
 978         tcf_chain_hold(chain);
 979 }
 980
 981 static void tcf_chain_tp_remove(struct tcf_chain *chain,
 982                                 struct tcf_chain_info *chain_info,
 983                                 struct tcf_proto *tp)
 984 {
 985         struct tcf_proto *next = rtnl_dereference(chain_info->next);
 986
 987         if (tp == chain->filter_chain)
 988                 tcf_chain0_head_change(chain, next);
 989         RCU_INIT_POINTER(*chain_info->pprev, next);
 990         tcf_chain_put(chain);
 991 }
 992
 993 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
 994                                            struct tcf_chain_info *chain_info,
 995                                            u32 protocol, u32 prio,
 996                                            bool prio_allocate)
 997 {
 998         struct tcf_proto **pprev;
 999         struct tcf_proto *tp;
1000
1001         /* Check the chain for existence of proto-tcf with this priority */
1002         for (pprev = &chain->filter_chain;
1003              (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
1004                 if (tp->prio >= prio) {
1005                         if (tp->prio == prio) {
1006                                 if (prio_allocate ||
1007                                     (tp->protocol != protocol && protocol))
1008                                         return ERR_PTR(-EINVAL);
1009                         } else {
1010                                 tp = NULL;
1011                         }
1012                         break;
1013                 }
1014         }
1015         chain_info->pprev = pprev;
1016         chain_info->next = tp ? tp->next : NULL;
1017         return tp;
1018 }
1019
1020 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1021                          struct tcf_proto *tp, struct tcf_block *block,
1022                          struct Qdisc *q, u32 parent, void *fh,
1023                          u32 portid, u32 seq, u16 flags, int event)
1024 {
1025         struct tcmsg *tcm;
1026         struct nlmsghdr  *nlh;
1027         unsigned char *b = skb_tail_pointer(skb);
1028
1029         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1030         if (!nlh)
1031                 goto out_nlmsg_trim;
1032         tcm = nlmsg_data(nlh);
1033         tcm->tcm_family = AF_UNSPEC;
1034         tcm->tcm__pad1 = 0;
1035         tcm->tcm__pad2 = 0;
1036         if (q) {
1037                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1038                 tcm->tcm_parent = parent;
1039         } else {
1040                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1041                 tcm->tcm_block_index = block->index;
1042         }
1043         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1044         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1045                 goto nla_put_failure;
1046         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1047                 goto nla_put_failure;
1048         if (!fh) {
1049                 tcm->tcm_handle = 0;
1050         } else {
1051                 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
1052                         goto nla_put_failure;
1053         }
1054         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1055         return skb->len;
1056
1057 out_nlmsg_trim:
1058 nla_put_failure:
1059         nlmsg_trim(skb, b);
1060         return -1;
1061 }
1062
1063 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1064                           struct nlmsghdr *n, struct tcf_proto *tp,
1065                           struct tcf_block *block, struct Qdisc *q,
1066                           u32 parent, void *fh, int event, bool unicast)
1067 {
1068         struct sk_buff *skb;
1069         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1070
1071         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1072         if (!skb)
1073                 return -ENOBUFS;
1074
1075         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1076                           n->nlmsg_seq, n->nlmsg_flags, event) <= 0) {
1077                 kfree_skb(skb);
1078                 return -EINVAL;
1079         }
1080
1081         if (unicast)
1082                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1083
1084         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1085                               n->nlmsg_flags & NLM_F_ECHO);
1086 }
1087
1088 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1089                               struct nlmsghdr *n, struct tcf_proto *tp,
1090                               struct tcf_block *block, struct Qdisc *q,
1091                               u32 parent, void *fh, bool unicast, bool *last,
1092                               struct netlink_ext_ack *extack)
1093 {
1094         struct sk_buff *skb;
1095         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1096         int err;
1097
1098         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1099         if (!skb)
1100                 return -ENOBUFS;
1101
1102         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1103                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
1104                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1105                 kfree_skb(skb);
1106                 return -EINVAL;
1107         }
1108
1109         err = tp->ops->delete(tp, fh, last, extack);
1110         if (err) {
1111                 kfree_skb(skb);
1112                 return err;
1113         }
1114
1115         if (unicast)
1116                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1117
1118         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1119                              n->nlmsg_flags & NLM_F_ECHO);
1120         if (err < 0)
1121                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1122         return err;
1123 }
1124
1125 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1126                                  struct tcf_block *block, struct Qdisc *q,
1127                                  u32 parent, struct nlmsghdr *n,
1128                                  struct tcf_chain *chain, int event)
1129 {
1130         struct tcf_proto *tp;
1131
1132         for (tp = rtnl_dereference(chain->filter_chain);
1133              tp; tp = rtnl_dereference(tp->next))
1134                 tfilter_notify(net, oskb, n, tp, block,
1135                                q, parent, NULL, event, false);
1136 }
1137
1138 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1139                           struct netlink_ext_ack *extack)
1140 {
1141         struct net *net = sock_net(skb->sk);
1142         struct nlattr *tca[TCA_MAX + 1];
1143         struct tcmsg *t;
1144         u32 protocol;
1145         u32 prio;
1146         bool prio_allocate;
1147         u32 parent;
1148         u32 chain_index;
1149         struct Qdisc *q = NULL;
1150         struct tcf_chain_info chain_info;
1151         struct tcf_chain *chain = NULL;
1152         struct tcf_block *block;
1153         struct tcf_proto *tp;
1154         unsigned long cl;
1155         void *fh;
1156         int err;
1157         int tp_created;
1158
1159         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1160                 return -EPERM;
1161
1162 replay:
1163         tp_created = 0;
1164
1165         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1166         if (err < 0)
1167                 return err;
1168
1169         t = nlmsg_data(n);
1170         protocol = TC_H_MIN(t->tcm_info);
1171         prio = TC_H_MAJ(t->tcm_info);
1172         prio_allocate = false;
1173         parent = t->tcm_parent;
1174         cl = 0;
1175
1176         if (prio == 0) {
1177                 /* If no priority is provided by the user,
1178                  * we allocate one.
1179                  */
1180                 if (n->nlmsg_flags & NLM_F_CREATE) {
1181                         prio = TC_H_MAKE(0x80000000U, 0U);
1182                         prio_allocate = true;
1183                 } else {
1184                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1185                         return -ENOENT;
1186                 }
1187         }
1188
1189         /* Find head of filter chain. */
1190
1191         block = tcf_block_find(net, &q, &parent, &cl,
1192                                t->tcm_ifindex, t->tcm_block_index, extack);
1193         if (IS_ERR(block)) {
1194                 err = PTR_ERR(block);
1195                 goto errout;
1196         }
1197
1198         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1199         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1200                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1201                 err = -EINVAL;
1202                 goto errout;
1203         }
1204         chain = tcf_chain_get(block, chain_index, true);
1205         if (!chain) {
1206                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1207                 err = -ENOMEM;
1208                 goto errout;
1209         }
1210
1211         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1212                                prio, prio_allocate);
1213         if (IS_ERR(tp)) {
1214                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1215                 err = PTR_ERR(tp);
1216                 goto errout;
1217         }
1218
1219         if (tp == NULL) {
1220                 /* Proto-tcf does not exist, create new one */
1221
1222                 if (tca[TCA_KIND] == NULL || !protocol) {
1223                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
1224                         err = -EINVAL;
1225                         goto errout;
1226                 }
1227
1228                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1229                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1230                         err = -ENOENT;
1231                         goto errout;
1232                 }
1233
1234                 if (prio_allocate)
1235                         prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
1236
1237                 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
1238                                       protocol, prio, chain, extack);
1239                 if (IS_ERR(tp)) {
1240                         err = PTR_ERR(tp);
1241                         goto errout;
1242                 }
1243                 tp_created = 1;
1244         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1245                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1246                 err = -EINVAL;
1247                 goto errout;
1248         }
1249
1250         fh = tp->ops->get(tp, t->tcm_handle);
1251
1252         if (!fh) {
1253                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1254                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1255                         err = -ENOENT;
1256                         goto errout;
1257                 }
1258         } else if (n->nlmsg_flags & NLM_F_EXCL) {
1259                 NL_SET_ERR_MSG(extack, "Filter already exists");
1260                 err = -EEXIST;
1261                 goto errout;
1262         }
1263
1264         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
1265                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
1266                 err = -EINVAL;
1267                 goto errout;
1268         }
1269
1270         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
1271                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
1272                               extack);
1273         if (err == 0) {
1274                 if (tp_created)
1275                         tcf_chain_tp_insert(chain, &chain_info, tp);
1276                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1277                                RTM_NEWTFILTER, false);
1278         } else {
1279                 if (tp_created)
1280                         tcf_proto_destroy(tp, NULL);
1281         }
1282
1283 errout:
1284         if (chain)
1285                 tcf_chain_put(chain);
1286         if (err == -EAGAIN)
1287                 /* Replay the request. */
1288                 goto replay;
1289         return err;
1290 }
1291
1292 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1293                           struct netlink_ext_ack *extack)
1294 {
1295         struct net *net = sock_net(skb->sk);
1296         struct nlattr *tca[TCA_MAX + 1];
1297         struct tcmsg *t;
1298         u32 protocol;
1299         u32 prio;
1300         u32 parent;
1301         u32 chain_index;
1302         struct Qdisc *q = NULL;
1303         struct tcf_chain_info chain_info;
1304         struct tcf_chain *chain = NULL;
1305         struct tcf_block *block;
1306         struct tcf_proto *tp = NULL;
1307         unsigned long cl = 0;
1308         void *fh = NULL;
1309         int err;
1310
1311         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1312                 return -EPERM;
1313
1314         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1315         if (err < 0)
1316                 return err;
1317
1318         t = nlmsg_data(n);
1319         protocol = TC_H_MIN(t->tcm_info);
1320         prio = TC_H_MAJ(t->tcm_info);
1321         parent = t->tcm_parent;
1322
1323         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
1324                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
1325                 return -ENOENT;
1326         }
1327
1328         /* Find head of filter chain. */
1329
1330         block = tcf_block_find(net, &q, &parent, &cl,
1331                                t->tcm_ifindex, t->tcm_block_index, extack);
1332         if (IS_ERR(block)) {
1333                 err = PTR_ERR(block);
1334                 goto errout;
1335         }
1336
1337         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1338         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1339                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1340                 err = -EINVAL;
1341                 goto errout;
1342         }
1343         chain = tcf_chain_get(block, chain_index, false);
1344         if (!chain) {
1345                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1346                 err = -EINVAL;
1347                 goto errout;
1348         }
1349
1350         if (prio == 0) {
1351                 tfilter_notify_chain(net, skb, block, q, parent, n,
1352                                      chain, RTM_DELTFILTER);
1353                 tcf_chain_flush(chain);
1354                 err = 0;
1355                 goto errout;
1356         }
1357
1358         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1359                                prio, false);
1360         if (!tp || IS_ERR(tp)) {
1361                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1362                 err = tp ? PTR_ERR(tp) : -ENOENT;
1363                 goto errout;
1364         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1365                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1366                 err = -EINVAL;
1367                 goto errout;
1368         }
1369
1370         fh = tp->ops->get(tp, t->tcm_handle);
1371
1372         if (!fh) {
1373                 if (t->tcm_handle == 0) {
1374                         tcf_chain_tp_remove(chain, &chain_info, tp);
1375                         tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1376                                        RTM_DELTFILTER, false);
1377                         tcf_proto_destroy(tp, extack);
1378                         err = 0;
1379                 } else {
1380                         NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1381                         err = -ENOENT;
1382                 }
1383         } else {
1384                 bool last;
1385
1386                 err = tfilter_del_notify(net, skb, n, tp, block,
1387                                          q, parent, fh, false, &last,
1388                                          extack);
1389                 if (err)
1390                         goto errout;
1391                 if (last) {
1392                         tcf_chain_tp_remove(chain, &chain_info, tp);
1393                         tcf_proto_destroy(tp, extack);
1394                 }
1395         }
1396
1397 errout:
1398         if (chain)
1399                 tcf_chain_put(chain);
1400         return err;
1401 }
1402
1403 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1404                           struct netlink_ext_ack *extack)
1405 {
1406         struct net *net = sock_net(skb->sk);
1407         struct nlattr *tca[TCA_MAX + 1];
1408         struct tcmsg *t;
1409         u32 protocol;
1410         u32 prio;
1411         u32 parent;
1412         u32 chain_index;
1413         struct Qdisc *q = NULL;
1414         struct tcf_chain_info chain_info;
1415         struct tcf_chain *chain = NULL;
1416         struct tcf_block *block;
1417         struct tcf_proto *tp = NULL;
1418         unsigned long cl = 0;
1419         void *fh = NULL;
1420         int err;
1421
1422         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1423         if (err < 0)
1424                 return err;
1425
1426         t = nlmsg_data(n);
1427         protocol = TC_H_MIN(t->tcm_info);
1428         prio = TC_H_MAJ(t->tcm_info);
1429         parent = t->tcm_parent;
1430
1431         if (prio == 0) {
1432                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1433                 return -ENOENT;
1434         }
1435
1436         /* Find head of filter chain. */
1437
1438         block = tcf_block_find(net, &q, &parent, &cl,
1439                                t->tcm_ifindex, t->tcm_block_index, extack);
1440         if (IS_ERR(block)) {
1441                 err = PTR_ERR(block);
1442                 goto errout;
1443         }
1444
1445         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1446         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1447                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1448                 err = -EINVAL;
1449                 goto errout;
1450         }
1451         chain = tcf_chain_get(block, chain_index, false);
1452         if (!chain) {
1453                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1454                 err = -EINVAL;
1455                 goto errout;
1456         }
1457
1458         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1459                                prio, false);
1460         if (!tp || IS_ERR(tp)) {
1461                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1462                 err = tp ? PTR_ERR(tp) : -ENOENT;
1463                 goto errout;
1464         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1465                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1466                 err = -EINVAL;
1467                 goto errout;
1468         }
1469
1470         fh = tp->ops->get(tp, t->tcm_handle);
1471
1472         if (!fh) {
1473                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1474                 err = -ENOENT;
1475         } else {
1476                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
1477                                      fh, RTM_NEWTFILTER, true);
1478                 if (err < 0)
1479                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
1480         }
1481
1482 errout:
1483         if (chain)
1484                 tcf_chain_put(chain);
1485         return err;
1486 }
1487
1488 struct tcf_dump_args {
1489         struct tcf_walker w;
1490         struct sk_buff *skb;
1491         struct netlink_callback *cb;
1492         struct tcf_block *block;
1493         struct Qdisc *q;
1494         u32 parent;
1495 };
1496
1497 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1498 {
1499         struct tcf_dump_args *a = (void *)arg;
1500         struct net *net = sock_net(a->skb->sk);
1501
1502         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
1503                              n, NETLINK_CB(a->cb->skb).portid,
1504                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1505                              RTM_NEWTFILTER);
1506 }
1507
1508 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
1509                            struct sk_buff *skb, struct netlink_callback *cb,
1510                            long index_start, long *p_index)
1511 {
1512         struct net *net = sock_net(skb->sk);
1513         struct tcf_block *block = chain->block;
1514         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1515         struct tcf_dump_args arg;
1516         struct tcf_proto *tp;
1517
1518         for (tp = rtnl_dereference(chain->filter_chain);
1519              tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
1520                 if (*p_index < index_start)
1521                         continue;
1522                 if (TC_H_MAJ(tcm->tcm_info) &&
1523                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
1524                         continue;
1525                 if (TC_H_MIN(tcm->tcm_info) &&
1526                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
1527                         continue;
1528                 if (*p_index > index_start)
1529                         memset(&cb->args[1], 0,
1530                                sizeof(cb->args) - sizeof(cb->args[0]));
1531                 if (cb->args[1] == 0) {
1532                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
1533                                           NETLINK_CB(cb->skb).portid,
1534                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
1535                                           RTM_NEWTFILTER) <= 0)
1536                                 return false;
1537
1538                         cb->args[1] = 1;
1539                 }
1540                 if (!tp->ops->walk)
1541                         continue;
1542                 arg.w.fn = tcf_node_dump;
1543                 arg.skb = skb;
1544                 arg.cb = cb;
1545                 arg.block = block;
1546                 arg.q = q;
1547                 arg.parent = parent;
1548                 arg.w.stop = 0;
1549                 arg.w.skip = cb->args[1] - 1;
1550                 arg.w.count = 0;
1551                 arg.w.cookie = cb->args[2];
1552                 tp->ops->walk(tp, &arg.w);
1553                 cb->args[2] = arg.w.cookie;
1554                 cb->args[1] = arg.w.count + 1;
1555                 if (arg.w.stop)
1556                         return false;
1557         }
1558         return true;
1559 }
1560
1561 /* called with RTNL */
1562 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
1563 {
1564         struct net *net = sock_net(skb->sk);
1565         struct nlattr *tca[TCA_MAX + 1];
1566         struct Qdisc *q = NULL;
1567         struct tcf_block *block;
1568         struct tcf_chain *chain;
1569         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1570         long index_start;
1571         long index;
1572         u32 parent;
1573         int err;
1574
1575         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1576                 return skb->len;
1577
1578         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1579         if (err)
1580                 return err;
1581
1582         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1583                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1584                 if (!block)
1585                         goto out;
1586                 /* If we work with block index, q is NULL and parent value
1587                  * will never be used in the following code. The check
1588                  * in tcf_fill_node prevents it. However, compiler does not
1589                  * see that far, so set parent to zero to silence the warning
1590                  * about parent being uninitialized.
1591                  */
1592                 parent = 0;
1593         } else {
1594                 const struct Qdisc_class_ops *cops;
1595                 struct net_device *dev;
1596                 unsigned long cl = 0;
1597
1598                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1599                 if (!dev)
1600                         return skb->len;
1601
1602                 parent = tcm->tcm_parent;
1603                 if (!parent) {
1604                         q = dev->qdisc;
1605                         parent = q->handle;
1606                 } else {
1607                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1608                 }
1609                 if (!q)
1610                         goto out;
1611                 cops = q->ops->cl_ops;
1612                 if (!cops)
1613                         goto out;
1614                 if (!cops->tcf_block)
1615                         goto out;
1616                 if (TC_H_MIN(tcm->tcm_parent)) {
1617                         cl = cops->find(q, tcm->tcm_parent);
1618                         if (cl == 0)
1619                                 goto out;
1620                 }
1621                 block = cops->tcf_block(q, cl, NULL);
1622                 if (!block)
1623                         goto out;
1624                 if (tcf_block_shared(block))
1625                         q = NULL;
1626         }
1627
1628         index_start = cb->args[0];
1629         index = 0;
1630
1631         list_for_each_entry(chain, &block->chain_list, list) {
1632                 if (tca[TCA_CHAIN] &&
1633                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1634                         continue;
1635                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
1636                                     index_start, &index)) {
1637                         err = -EMSGSIZE;
1638                         break;
1639                 }
1640         }
1641
1642         cb->args[0] = index;
1643
1644 out:
1645         /* If we did no progress, the error (EMSGSIZE) is real */
1646         if (skb->len == 0 && err)
1647                 return err;
1648         return skb->len;
1649 }
1650
1651 static int tc_chain_fill_node(struct tcf_chain *chain, struct net *net,
1652                               struct sk_buff *skb, struct tcf_block *block,
1653                               u32 portid, u32 seq, u16 flags, int event)
1654 {
1655         unsigned char *b = skb_tail_pointer(skb);
1656         const struct tcf_proto_ops *ops;
1657         struct nlmsghdr *nlh;
1658         struct tcmsg *tcm;
1659         void *priv;
1660
1661         ops = chain->tmplt_ops;
1662         priv = chain->tmplt_priv;
1663
1664         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1665         if (!nlh)
1666                 goto out_nlmsg_trim;
1667         tcm = nlmsg_data(nlh);
1668         tcm->tcm_family = AF_UNSPEC;
1669         tcm->tcm__pad1 = 0;
1670         tcm->tcm__pad2 = 0;
1671         tcm->tcm_handle = 0;
1672         if (block->q) {
1673                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
1674                 tcm->tcm_parent = block->q->handle;
1675         } else {
1676                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1677                 tcm->tcm_block_index = block->index;
1678         }
1679
1680         if (nla_put_u32(skb, TCA_CHAIN, chain->index))
1681                 goto nla_put_failure;
1682
1683         if (ops) {
1684                 if (nla_put_string(skb, TCA_KIND, ops->kind))
1685                         goto nla_put_failure;
1686                 if (ops->tmplt_dump(skb, net, priv) < 0)
1687                         goto nla_put_failure;
1688         }
1689
1690         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1691         return skb->len;
1692
1693 out_nlmsg_trim:
1694 nla_put_failure:
1695         nlmsg_trim(skb, b);
1696         return -EMSGSIZE;
1697 }
1698
1699 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
1700                            u32 seq, u16 flags, int event, bool unicast)
1701 {
1702         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1703         struct tcf_block *block = chain->block;
1704         struct net *net = block->net;
1705         struct sk_buff *skb;
1706
1707         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1708         if (!skb)
1709                 return -ENOBUFS;
1710
1711         if (tc_chain_fill_node(chain, net, skb, block, portid,
1712                                seq, flags, event) <= 0) {
1713                 kfree_skb(skb);
1714                 return -EINVAL;
1715         }
1716
1717         if (unicast)
1718                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1719
1720         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
1721 }
1722
1723 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
1724                               struct nlattr **tca,
1725                               struct netlink_ext_ack *extack)
1726 {
1727         const struct tcf_proto_ops *ops;
1728         void *tmplt_priv;
1729
1730         /* If kind is not set, user did not specify template. */
1731         if (!tca[TCA_KIND])
1732                 return 0;
1733
1734         ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), extack);
1735         if (IS_ERR(ops))
1736                 return PTR_ERR(ops);
1737         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
1738                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
1739                 return -EOPNOTSUPP;
1740         }
1741
1742         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
1743         if (IS_ERR(tmplt_priv)) {
1744                 module_put(ops->owner);
1745                 return PTR_ERR(tmplt_priv);
1746         }
1747         chain->tmplt_ops = ops;
1748         chain->tmplt_priv = tmplt_priv;
1749         return 0;
1750 }
1751
1752 static void tc_chain_tmplt_del(struct tcf_chain *chain)
1753 {
1754         const struct tcf_proto_ops *ops = chain->tmplt_ops;
1755
1756         /* If template ops are set, no work to do for us. */
1757         if (!ops)
1758                 return;
1759
1760         ops->tmplt_destroy(chain->tmplt_priv);
1761         module_put(ops->owner);
1762 }
1763
1764 /* Add/delete/get a chain */
1765
1766 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
1767                         struct netlink_ext_ack *extack)
1768 {
1769         struct net *net = sock_net(skb->sk);
1770         struct nlattr *tca[TCA_MAX + 1];
1771         struct tcmsg *t;
1772         u32 parent;
1773         u32 chain_index;
1774         struct Qdisc *q = NULL;
1775         struct tcf_chain *chain = NULL;
1776         struct tcf_block *block;
1777         unsigned long cl;
1778         int err;
1779
1780         if (n->nlmsg_type != RTM_GETCHAIN &&
1781             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1782                 return -EPERM;
1783
1784 replay:
1785         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1786         if (err < 0)
1787                 return err;
1788
1789         t = nlmsg_data(n);
1790         parent = t->tcm_parent;
1791         cl = 0;
1792
1793         block = tcf_block_find(net, &q, &parent, &cl,
1794                                t->tcm_ifindex, t->tcm_block_index, extack);
1795         if (IS_ERR(block))
1796                 return PTR_ERR(block);
1797
1798         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1799         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1800                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1801                 return -EINVAL;
1802         }
1803         chain = tcf_chain_lookup(block, chain_index);
1804         if (n->nlmsg_type == RTM_NEWCHAIN) {
1805                 if (chain) {
1806                         NL_SET_ERR_MSG(extack, "Filter chain already exists");
1807                         return -EEXIST;
1808                 }
1809                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1810                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
1811                         return -ENOENT;
1812                 }
1813                 chain = tcf_chain_create(block, chain_index);
1814                 if (!chain) {
1815                         NL_SET_ERR_MSG(extack, "Failed to create filter chain");
1816                         return -ENOMEM;
1817                 }
1818         } else {
1819                 if (!chain) {
1820                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1821                         return -EINVAL;
1822                 }
1823                 tcf_chain_hold(chain);
1824         }
1825
1826         switch (n->nlmsg_type) {
1827         case RTM_NEWCHAIN:
1828                 err = tc_chain_tmplt_add(chain, net, tca, extack);
1829                 if (err)
1830                         goto errout;
1831                 /* In case the chain was successfully added, take a reference
1832                  * to the chain. This ensures that an empty chain
1833                  * does not disappear at the end of this function.
1834                  */
1835                 tcf_chain_hold(chain);
1836                 chain->explicitly_created = true;
1837                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
1838                                 RTM_NEWCHAIN, false);
1839                 break;
1840         case RTM_DELCHAIN:
1841                 /* Flush the chain first as the user requested chain removal. */
1842                 tcf_chain_flush(chain);
1843                 /* In case the chain was successfully deleted, put a reference
1844                  * to the chain previously taken during addition.
1845                  */
1846                 tcf_chain_put_explicitly_created(chain);
1847                 chain->explicitly_created = false;
1848                 break;
1849         case RTM_GETCHAIN:
1850                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
1851                                       n->nlmsg_seq, n->nlmsg_type, true);
1852                 if (err < 0)
1853                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
1854                 break;
1855         default:
1856                 err = -EOPNOTSUPP;
1857                 NL_SET_ERR_MSG(extack, "Unsupported message type");
1858                 goto errout;
1859         }
1860
1861 errout:
1862         tcf_chain_put(chain);
1863         if (err == -EAGAIN)
1864                 /* Replay the request. */
1865                 goto replay;
1866         return err;
1867 }
1868
1869 /* called with RTNL */
1870 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
1871 {
1872         struct net *net = sock_net(skb->sk);
1873         struct nlattr *tca[TCA_MAX + 1];
1874         struct Qdisc *q = NULL;
1875         struct tcf_block *block;
1876         struct tcf_chain *chain;
1877         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1878         long index_start;
1879         long index;
1880         u32 parent;
1881         int err;
1882
1883         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1884                 return skb->len;
1885
1886         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1887         if (err)
1888                 return err;
1889
1890         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1891                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1892                 if (!block)
1893                         goto out;
1894                 /* If we work with block index, q is NULL and parent value
1895                  * will never be used in the following code. The check
1896                  * in tcf_fill_node prevents it. However, compiler does not
1897                  * see that far, so set parent to zero to silence the warning
1898                  * about parent being uninitialized.
1899                  */
1900                 parent = 0;
1901         } else {
1902                 const struct Qdisc_class_ops *cops;
1903                 struct net_device *dev;
1904                 unsigned long cl = 0;
1905
1906                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1907                 if (!dev)
1908                         return skb->len;
1909
1910                 parent = tcm->tcm_parent;
1911                 if (!parent) {
1912                         q = dev->qdisc;
1913                         parent = q->handle;
1914                 } else {
1915                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1916                 }
1917                 if (!q)
1918                         goto out;
1919                 cops = q->ops->cl_ops;
1920                 if (!cops)
1921                         goto out;
1922                 if (!cops->tcf_block)
1923                         goto out;
1924                 if (TC_H_MIN(tcm->tcm_parent)) {
1925                         cl = cops->find(q, tcm->tcm_parent);
1926                         if (cl == 0)
1927                                 goto out;
1928                 }
1929                 block = cops->tcf_block(q, cl, NULL);
1930                 if (!block)
1931                         goto out;
1932                 if (tcf_block_shared(block))
1933                         q = NULL;
1934         }
1935
1936         index_start = cb->args[0];
1937         index = 0;
1938
1939         list_for_each_entry(chain, &block->chain_list, list) {
1940                 if ((tca[TCA_CHAIN] &&
1941                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
1942                         continue;
1943                 if (index < index_start) {
1944                         index++;
1945                         continue;
1946                 }
1947                 err = tc_chain_fill_node(chain, net, skb, block,
1948                                          NETLINK_CB(cb->skb).portid,
1949                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1950                                          RTM_NEWCHAIN);
1951                 if (err <= 0)
1952                         break;
1953                 index++;
1954         }
1955
1956         cb->args[0] = index;
1957
1958 out:
1959         /* If we did no progress, the error (EMSGSIZE) is real */
1960         if (skb->len == 0 && err)
1961                 return err;
1962         return skb->len;
1963 }
1964
1965 void tcf_exts_destroy(struct tcf_exts *exts)
1966 {
1967 #ifdef CONFIG_NET_CLS_ACT
1968         tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
1969         kfree(exts->actions);
1970         exts->nr_actions = 0;
1971 #endif
1972 }
1973 EXPORT_SYMBOL(tcf_exts_destroy);
1974
1975 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
1976                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
1977                       struct netlink_ext_ack *extack)
1978 {
1979 #ifdef CONFIG_NET_CLS_ACT
1980         {
1981                 struct tc_action *act;
1982                 size_t attr_size = 0;
1983
1984                 if (exts->police && tb[exts->police]) {
1985                         act = tcf_action_init_1(net, tp, tb[exts->police],
1986                                                 rate_tlv, "police", ovr,
1987                                                 TCA_ACT_BIND, true, extack);
1988                         if (IS_ERR(act))
1989                                 return PTR_ERR(act);
1990
1991                         act->type = exts->type = TCA_OLD_COMPAT;
1992                         exts->actions[0] = act;
1993                         exts->nr_actions = 1;
1994                 } else if (exts->action && tb[exts->action]) {
1995                         int err;
1996
1997                         err = tcf_action_init(net, tp, tb[exts->action],
1998                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
1999                                               exts->actions, &attr_size, true,
2000                                               extack);
2001                         if (err < 0)
2002                                 return err;
2003                         exts->nr_actions = err;
2004                 }
2005                 exts->net = net;
2006         }
2007 #else
2008         if ((exts->action && tb[exts->action]) ||
2009             (exts->police && tb[exts->police])) {
2010                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
2011                 return -EOPNOTSUPP;
2012         }
2013 #endif
2014
2015         return 0;
2016 }
2017 EXPORT_SYMBOL(tcf_exts_validate);
2018
2019 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
2020 {
2021 #ifdef CONFIG_NET_CLS_ACT
2022         struct tcf_exts old = *dst;
2023
2024         *dst = *src;
2025         tcf_exts_destroy(&old);
2026 #endif
2027 }
2028 EXPORT_SYMBOL(tcf_exts_change);
2029
2030 #ifdef CONFIG_NET_CLS_ACT
2031 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
2032 {
2033         if (exts->nr_actions == 0)
2034                 return NULL;
2035         else
2036                 return exts->actions[0];
2037 }
2038 #endif
2039
2040 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
2041 {
2042 #ifdef CONFIG_NET_CLS_ACT
2043         struct nlattr *nest;
2044
2045         if (exts->action && tcf_exts_has_actions(exts)) {
2046                 /*
2047                  * again for backward compatible mode - we want
2048                  * to work with both old and new modes of entering
2049                  * tc data even if iproute2  was newer - jhs
2050                  */
2051                 if (exts->type != TCA_OLD_COMPAT) {
2052                         nest = nla_nest_start(skb, exts->action);
2053                         if (nest == NULL)
2054                                 goto nla_put_failure;
2055
2056                         if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
2057                                 goto nla_put_failure;
2058                         nla_nest_end(skb, nest);
2059                 } else if (exts->police) {
2060                         struct tc_action *act = tcf_exts_first_act(exts);
2061                         nest = nla_nest_start(skb, exts->police);
2062                         if (nest == NULL || !act)
2063                                 goto nla_put_failure;
2064                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
2065                                 goto nla_put_failure;
2066                         nla_nest_end(skb, nest);
2067                 }
2068         }
2069         return 0;
2070
2071 nla_put_failure:
2072         nla_nest_cancel(skb, nest);
2073         return -1;
2074 #else
2075         return 0;
2076 #endif
2077 }
2078 EXPORT_SYMBOL(tcf_exts_dump);
2079
2080
2081 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
2082 {
2083 #ifdef CONFIG_NET_CLS_ACT
2084         struct tc_action *a = tcf_exts_first_act(exts);
2085         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
2086                 return -1;
2087 #endif
2088         return 0;
2089 }
2090 EXPORT_SYMBOL(tcf_exts_dump_stats);
2091
2092 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
2093                                        enum tc_setup_type type,
2094                                        void *type_data, bool err_stop)
2095 {
2096         int ok_count = 0;
2097 #ifdef CONFIG_NET_CLS_ACT
2098         const struct tc_action *a;
2099         struct net_device *dev;
2100         int i, ret;
2101
2102         if (!tcf_exts_has_actions(exts))
2103                 return 0;
2104
2105         for (i = 0; i < exts->nr_actions; i++) {
2106                 a = exts->actions[i];
2107                 if (!a->ops->get_dev)
2108                         continue;
2109                 dev = a->ops->get_dev(a);
2110                 if (!dev)
2111                         continue;
2112                 ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
2113                 if (ret < 0)
2114                         return ret;
2115                 ok_count += ret;
2116         }
2117 #endif
2118         return ok_count;
2119 }
2120
2121 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
2122                      enum tc_setup_type type, void *type_data, bool err_stop)
2123 {
2124         int ok_count;
2125         int ret;
2126
2127         ret = tcf_block_cb_call(block, type, type_data, err_stop);
2128         if (ret < 0)
2129                 return ret;
2130         ok_count = ret;
2131
2132         if (!exts || ok_count)
2133                 return ok_count;
2134         ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
2135         if (ret < 0)
2136                 return ret;
2137         ok_count += ret;
2138
2139         return ok_count;
2140 }
2141 EXPORT_SYMBOL(tc_setup_cb_call);
2142
2143 static __net_init int tcf_net_init(struct net *net)
2144 {
2145         struct tcf_net *tn = net_generic(net, tcf_net_id);
2146
2147         idr_init(&tn->idr);
2148         return 0;
2149 }
2150
2151 static void __net_exit tcf_net_exit(struct net *net)
2152 {
2153         struct tcf_net *tn = net_generic(net, tcf_net_id);
2154
2155         idr_destroy(&tn->idr);
2156 }
2157
2158 static struct pernet_operations tcf_net_ops = {
2159         .init = tcf_net_init,
2160         .exit = tcf_net_exit,
2161         .id   = &tcf_net_id,
2162         .size = sizeof(struct tcf_net),
2163 };
2164
2165 static int __init tc_filter_init(void)
2166 {
2167         int err;
2168
2169         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
2170         if (!tc_filter_wq)
2171                 return -ENOMEM;
2172
2173         err = register_pernet_subsys(&tcf_net_ops);
2174         if (err)
2175                 goto err_register_pernet_subsys;
2176
2177         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 0);
2178         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 0);
2179         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
2180                       tc_dump_tfilter, 0);
2181         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
2182         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
2183         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
2184                       tc_dump_chain, 0);
2185
2186         return 0;
2187
2188 err_register_pernet_subsys:
2189         destroy_workqueue(tc_filter_wq);
2190         return err;
2191 }
2192
2193 subsys_initcall(tc_filter_init);