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 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
  35
  36 /* The list of all installed classifier types */
  37 static LIST_HEAD(tcf_proto_base);
  38
  39 /* Protects list of registered TC modules. It is pure SMP lock. */
  40 static DEFINE_RWLOCK(cls_mod_lock);
  41
  42 /* Find classifier type by string name */
  43
  44 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
  45 {
  46         const struct tcf_proto_ops *t, *res = NULL;
  47
  48         if (kind) {
  49                 read_lock(&cls_mod_lock);
  50                 list_for_each_entry(t, &tcf_proto_base, head) {
  51                         if (strcmp(kind, t->kind) == 0) {
  52                                 if (try_module_get(t->owner))
  53                                         res = t;
  54                                 break;
  55                         }
  56                 }
  57                 read_unlock(&cls_mod_lock);
  58         }
  59         return res;
  60 }
  61
  62 static const struct tcf_proto_ops *
  63 tcf_proto_lookup_ops(const char *kind, struct netlink_ext_ack *extack)
  64 {
  65         const struct tcf_proto_ops *ops;
  66
  67         ops = __tcf_proto_lookup_ops(kind);
  68         if (ops)
  69                 return ops;
  70 #ifdef CONFIG_MODULES
  71         rtnl_unlock();
  72         request_module("cls_%s", kind);
  73         rtnl_lock();
  74         ops = __tcf_proto_lookup_ops(kind);
  75         /* We dropped the RTNL semaphore in order to perform
  76          * the module load. So, even if we succeeded in loading
  77          * the module we have to replay the request. We indicate
  78          * this using -EAGAIN.
  79          */
  80         if (ops) {
  81                 module_put(ops->owner);
  82                 return ERR_PTR(-EAGAIN);
  83         }
  84 #endif
  85         NL_SET_ERR_MSG(extack, "TC classifier not found");
  86         return ERR_PTR(-ENOENT);
  87 }
  88
  89 /* Register(unregister) new classifier type */
  90
  91 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  92 {
  93         struct tcf_proto_ops *t;
  94         int rc = -EEXIST;
  95
  96         write_lock(&cls_mod_lock);
  97         list_for_each_entry(t, &tcf_proto_base, head)
  98                 if (!strcmp(ops->kind, t->kind))
  99                         goto out;
 100
 101         list_add_tail(&ops->head, &tcf_proto_base);
 102         rc = 0;
 103 out:
 104         write_unlock(&cls_mod_lock);
 105         return rc;
 106 }
 107 EXPORT_SYMBOL(register_tcf_proto_ops);
 108
 109 static struct workqueue_struct *tc_filter_wq;
 110
 111 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 112 {
 113         struct tcf_proto_ops *t;
 114         int rc = -ENOENT;
 115
 116         /* Wait for outstanding call_rcu()s, if any, from a
 117          * tcf_proto_ops's destroy() handler.
 118          */
 119         rcu_barrier();
 120         flush_workqueue(tc_filter_wq);
 121
 122         write_lock(&cls_mod_lock);
 123         list_for_each_entry(t, &tcf_proto_base, head) {
 124                 if (t == ops) {
 125                         list_del(&t->head);
 126                         rc = 0;
 127                         break;
 128                 }
 129         }
 130         write_unlock(&cls_mod_lock);
 131         return rc;
 132 }
 133 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 134
 135 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 136 {
 137         INIT_RCU_WORK(rwork, func);
 138         return queue_rcu_work(tc_filter_wq, rwork);
 139 }
 140 EXPORT_SYMBOL(tcf_queue_work);
 141
 142 /* Select new prio value from the range, managed by kernel. */
 143
 144 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 145 {
 146         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 147
 148         if (tp)
 149                 first = tp->prio - 1;
 150
 151         return TC_H_MAJ(first);
 152 }
 153
 154 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 155                                           u32 prio, struct tcf_chain *chain,
 156                                           struct netlink_ext_ack *extack)
 157 {
 158         struct tcf_proto *tp;
 159         int err;
 160
 161         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 162         if (!tp)
 163                 return ERR_PTR(-ENOBUFS);
 164
 165         tp->ops = tcf_proto_lookup_ops(kind, extack);
 166         if (IS_ERR(tp->ops)) {
 167                 err = PTR_ERR(tp->ops);
 168                 goto errout;
 169         }
 170         tp->classify = tp->ops->classify;
 171         tp->protocol = protocol;
 172         tp->prio = prio;
 173         tp->chain = chain;
 174
 175         err = tp->ops->init(tp);
 176         if (err) {
 177                 module_put(tp->ops->owner);
 178                 goto errout;
 179         }
 180         return tp;
 181
 182 errout:
 183         kfree(tp);
 184         return ERR_PTR(err);
 185 }
 186
 187 static void tcf_proto_destroy(struct tcf_proto *tp,
 188                               struct netlink_ext_ack *extack)
 189 {
 190         tp->ops->destroy(tp, extack);
 191         module_put(tp->ops->owner);
 192         kfree_rcu(tp, rcu);
 193 }
 194
 195 struct tcf_filter_chain_list_item {
 196         struct list_head list;
 197         tcf_chain_head_change_t *chain_head_change;
 198         void *chain_head_change_priv;
 199 };
 200
 201 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 202                                           u32 chain_index)
 203 {
 204         struct tcf_chain *chain;
 205
 206         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 207         if (!chain)
 208                 return NULL;
 209         list_add_tail(&chain->list, &block->chain_list);
 210         chain->block = block;
 211         chain->index = chain_index;
 212         chain->refcnt = 1;
 213         if (!chain->index)
 214                 block->chain0.chain = chain;
 215         return chain;
 216 }
 217
 218 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 219                                        struct tcf_proto *tp_head)
 220 {
 221         if (item->chain_head_change)
 222                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 223 }
 224
 225 static void tcf_chain0_head_change(struct tcf_chain *chain,
 226                                    struct tcf_proto *tp_head)
 227 {
 228         struct tcf_filter_chain_list_item *item;
 229         struct tcf_block *block = chain->block;
 230
 231         if (chain->index)
 232                 return;
 233         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 234                 tcf_chain_head_change_item(item, tp_head);
 235 }
 236
 237 static void tcf_chain_destroy(struct tcf_chain *chain)
 238 {
 239         struct tcf_block *block = chain->block;
 240
 241         list_del(&chain->list);
 242         if (!chain->index)
 243                 block->chain0.chain = NULL;
 244         kfree(chain);
 245         if (list_empty(&block->chain_list) && block->refcnt == 0)
 246                 kfree(block);
 247 }
 248
 249 static void tcf_chain_hold(struct tcf_chain *chain)
 250 {
 251         ++chain->refcnt;
 252 }
 253
 254 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 255 {
 256         /* In case all the references are action references, this
 257          * chain should not be shown to the user.
 258          */
 259         return chain->refcnt == chain->action_refcnt;
 260 }
 261
 262 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 263                                           u32 chain_index)
 264 {
 265         struct tcf_chain *chain;
 266
 267         list_for_each_entry(chain, &block->chain_list, list) {
 268                 if (chain->index == chain_index)
 269                         return chain;
 270         }
 271         return NULL;
 272 }
 273
 274 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 275                            u32 seq, u16 flags, int event, bool unicast);
 276
 277 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 278                                          u32 chain_index, bool create,
 279                                          bool by_act)
 280 {
 281         struct tcf_chain *chain = tcf_chain_lookup(block, chain_index);
 282
 283         if (chain) {
 284                 tcf_chain_hold(chain);
 285         } else {
 286                 if (!create)
 287                         return NULL;
 288                 chain = tcf_chain_create(block, chain_index);
 289                 if (!chain)
 290                         return NULL;
 291         }
 292
 293         if (by_act)
 294                 ++chain->action_refcnt;
 295
 296         /* Send notification only in case we got the first
 297          * non-action reference. Until then, the chain acts only as
 298          * a placeholder for actions pointing to it and user ought
 299          * not know about them.
 300          */
 301         if (chain->refcnt - chain->action_refcnt == 1 && !by_act)
 302                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 303                                 RTM_NEWCHAIN, false);
 304
 305         return chain;
 306 }
 307
 308 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 309                                        bool create)
 310 {
 311         return __tcf_chain_get(block, chain_index, create, false);
 312 }
 313
 314 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 315 {
 316         return __tcf_chain_get(block, chain_index, true, true);
 317 }
 318 EXPORT_SYMBOL(tcf_chain_get_by_act);
 319
 320 static void tc_chain_tmplt_del(struct tcf_chain *chain);
 321
 322 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act)
 323 {
 324         if (by_act)
 325                 chain->action_refcnt--;
 326         chain->refcnt--;
 327
 328         /* The last dropped non-action reference will trigger notification. */
 329         if (chain->refcnt - chain->action_refcnt == 0 && !by_act)
 330                 tc_chain_notify(chain, NULL, 0, 0, RTM_DELCHAIN, false);
 331
 332         if (chain->refcnt == 0) {
 333                 tc_chain_tmplt_del(chain);
 334                 tcf_chain_destroy(chain);
 335         }
 336 }
 337
 338 static void tcf_chain_put(struct tcf_chain *chain)
 339 {
 340         __tcf_chain_put(chain, false);
 341 }
 342
 343 void tcf_chain_put_by_act(struct tcf_chain *chain)
 344 {
 345         __tcf_chain_put(chain, true);
 346 }
 347 EXPORT_SYMBOL(tcf_chain_put_by_act);
 348
 349 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 350 {
 351         if (chain->explicitly_created)
 352                 tcf_chain_put(chain);
 353 }
 354
 355 static void tcf_chain_flush(struct tcf_chain *chain)
 356 {
 357         struct tcf_proto *tp = rtnl_dereference(chain->filter_chain);
 358
 359         tcf_chain0_head_change(chain, NULL);
 360         while (tp) {
 361                 RCU_INIT_POINTER(chain->filter_chain, tp->next);
 362                 tcf_proto_destroy(tp, NULL);
 363                 tp = rtnl_dereference(chain->filter_chain);
 364                 tcf_chain_put(chain);
 365         }
 366 }
 367
 368 static bool tcf_block_offload_in_use(struct tcf_block *block)
 369 {
 370         return block->offloadcnt;
 371 }
 372
 373 static int tcf_block_offload_cmd(struct tcf_block *block,
 374                                  struct net_device *dev,
 375                                  struct tcf_block_ext_info *ei,
 376                                  enum tc_block_command command,
 377                                  struct netlink_ext_ack *extack)
 378 {
 379         struct tc_block_offload bo = {};
 380
 381         bo.command = command;
 382         bo.binder_type = ei->binder_type;
 383         bo.block = block;
 384         bo.extack = extack;
 385         return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 386 }
 387
 388 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 389                                   struct tcf_block_ext_info *ei,
 390                                   struct netlink_ext_ack *extack)
 391 {
 392         struct net_device *dev = q->dev_queue->dev;
 393         int err;
 394
 395         if (!dev->netdev_ops->ndo_setup_tc)
 396                 goto no_offload_dev_inc;
 397
 398         /* If tc offload feature is disabled and the block we try to bind
 399          * to already has some offloaded filters, forbid to bind.
 400          */
 401         if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
 402                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 403                 return -EOPNOTSUPP;
 404         }
 405
 406         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_BIND, extack);
 407         if (err == -EOPNOTSUPP)
 408                 goto no_offload_dev_inc;
 409         return err;
 410
 411 no_offload_dev_inc:
 412         if (tcf_block_offload_in_use(block))
 413                 return -EOPNOTSUPP;
 414         block->nooffloaddevcnt++;
 415         return 0;
 416 }
 417
 418 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 419                                      struct tcf_block_ext_info *ei)
 420 {
 421         struct net_device *dev = q->dev_queue->dev;
 422         int err;
 423
 424         if (!dev->netdev_ops->ndo_setup_tc)
 425                 goto no_offload_dev_dec;
 426         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_UNBIND, NULL);
 427         if (err == -EOPNOTSUPP)
 428                 goto no_offload_dev_dec;
 429         return;
 430
 431 no_offload_dev_dec:
 432         WARN_ON(block->nooffloaddevcnt-- == 0);
 433 }
 434
 435 static int
 436 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 437                               struct tcf_block_ext_info *ei,
 438                               struct netlink_ext_ack *extack)
 439 {
 440         struct tcf_chain *chain0 = block->chain0.chain;
 441         struct tcf_filter_chain_list_item *item;
 442
 443         item = kmalloc(sizeof(*item), GFP_KERNEL);
 444         if (!item) {
 445                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 446                 return -ENOMEM;
 447         }
 448         item->chain_head_change = ei->chain_head_change;
 449         item->chain_head_change_priv = ei->chain_head_change_priv;
 450         if (chain0 && chain0->filter_chain)
 451                 tcf_chain_head_change_item(item, chain0->filter_chain);
 452         list_add(&item->list, &block->chain0.filter_chain_list);
 453         return 0;
 454 }
 455
 456 static void
 457 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 458                               struct tcf_block_ext_info *ei)
 459 {
 460         struct tcf_chain *chain0 = block->chain0.chain;
 461         struct tcf_filter_chain_list_item *item;
 462
 463         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 464                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 465                     (item->chain_head_change == ei->chain_head_change &&
 466                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 467                         if (chain0)
 468                                 tcf_chain_head_change_item(item, NULL);
 469                         list_del(&item->list);
 470                         kfree(item);
 471                         return;
 472                 }
 473         }
 474         WARN_ON(1);
 475 }
 476
 477 struct tcf_net {
 478         struct idr idr;
 479 };
 480
 481 static unsigned int tcf_net_id;
 482
 483 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 484                             struct netlink_ext_ack *extack)
 485 {
 486         struct tcf_net *tn = net_generic(net, tcf_net_id);
 487
 488         return idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 489                              GFP_KERNEL);
 490 }
 491
 492 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 493 {
 494         struct tcf_net *tn = net_generic(net, tcf_net_id);
 495
 496         idr_remove(&tn->idr, block->index);
 497 }
 498
 499 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 500                                           u32 block_index,
 501                                           struct netlink_ext_ack *extack)
 502 {
 503         struct tcf_block *block;
 504
 505         block = kzalloc(sizeof(*block), GFP_KERNEL);
 506         if (!block) {
 507                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 508                 return ERR_PTR(-ENOMEM);
 509         }
 510         INIT_LIST_HEAD(&block->chain_list);
 511         INIT_LIST_HEAD(&block->cb_list);
 512         INIT_LIST_HEAD(&block->owner_list);
 513         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 514
 515         block->refcnt = 1;
 516         block->net = net;
 517         block->index = block_index;
 518
 519         /* Don't store q pointer for blocks which are shared */
 520         if (!tcf_block_shared(block))
 521                 block->q = q;
 522         return block;
 523 }
 524
 525 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 526 {
 527         struct tcf_net *tn = net_generic(net, tcf_net_id);
 528
 529         return idr_find(&tn->idr, block_index);
 530 }
 531
 532 /* Find tcf block.
 533  * Set q, parent, cl when appropriate.
 534  */
 535
 536 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
 537                                         u32 *parent, unsigned long *cl,
 538                                         int ifindex, u32 block_index,
 539                                         struct netlink_ext_ack *extack)
 540 {
 541         struct tcf_block *block;
 542         int err = 0;
 543
 544         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
 545                 block = tcf_block_lookup(net, block_index);
 546                 if (!block) {
 547                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
 548                         return ERR_PTR(-EINVAL);
 549                 }
 550         } else {
 551                 const struct Qdisc_class_ops *cops;
 552                 struct net_device *dev;
 553
 554                 rcu_read_lock();
 555
 556                 /* Find link */
 557                 dev = dev_get_by_index_rcu(net, ifindex);
 558                 if (!dev) {
 559                         rcu_read_unlock();
 560                         return ERR_PTR(-ENODEV);
 561                 }
 562
 563                 /* Find qdisc */
 564                 if (!*parent) {
 565                         *q = dev->qdisc;
 566                         *parent = (*q)->handle;
 567                 } else {
 568                         *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
 569                         if (!*q) {
 570                                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
 571                                 err = -EINVAL;
 572                                 goto errout_rcu;
 573                         }
 574                 }
 575
 576                 *q = qdisc_refcount_inc_nz(*q);
 577                 if (!*q) {
 578                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
 579                         err = -EINVAL;
 580                         goto errout_rcu;
 581                 }
 582
 583                 /* Is it classful? */
 584                 cops = (*q)->ops->cl_ops;
 585                 if (!cops) {
 586                         NL_SET_ERR_MSG(extack, "Qdisc not classful");
 587                         err = -EINVAL;
 588                         goto errout_rcu;
 589                 }
 590
 591                 if (!cops->tcf_block) {
 592                         NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
 593                         err = -EOPNOTSUPP;
 594                         goto errout_rcu;
 595                 }
 596
 597                 /* At this point we know that qdisc is not noop_qdisc,
 598                  * which means that qdisc holds a reference to net_device
 599                  * and we hold a reference to qdisc, so it is safe to release
 600                  * rcu read lock.
 601                  */
 602                 rcu_read_unlock();
 603
 604                 /* Do we search for filter, attached to class? */
 605                 if (TC_H_MIN(*parent)) {
 606                         *cl = cops->find(*q, *parent);
 607                         if (*cl == 0) {
 608                                 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
 609                                 err = -ENOENT;
 610                                 goto errout_qdisc;
 611                         }
 612                 }
 613
 614                 /* And the last stroke */
 615                 block = cops->tcf_block(*q, *cl, extack);
 616                 if (!block) {
 617                         err = -EINVAL;
 618                         goto errout_qdisc;
 619                 }
 620                 if (tcf_block_shared(block)) {
 621                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
 622                         err = -EOPNOTSUPP;
 623                         goto errout_qdisc;
 624                 }
 625         }
 626
 627         return block;
 628
 629 errout_rcu:
 630         rcu_read_unlock();
 631 errout_qdisc:
 632         if (*q) {
 633                 qdisc_put(*q);
 634                 *q = NULL;
 635         }
 636         return ERR_PTR(err);
 637 }
 638
 639 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block)
 640 {
 641         if (q)
 642                 qdisc_put(q);
 643 }
 644
 645 struct tcf_block_owner_item {
 646         struct list_head list;
 647         struct Qdisc *q;
 648         enum tcf_block_binder_type binder_type;
 649 };
 650
 651 static void
 652 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
 653                                struct Qdisc *q,
 654                                enum tcf_block_binder_type binder_type)
 655 {
 656         if (block->keep_dst &&
 657             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
 658             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
 659                 netif_keep_dst(qdisc_dev(q));
 660 }
 661
 662 void tcf_block_netif_keep_dst(struct tcf_block *block)
 663 {
 664         struct tcf_block_owner_item *item;
 665
 666         block->keep_dst = true;
 667         list_for_each_entry(item, &block->owner_list, list)
 668                 tcf_block_owner_netif_keep_dst(block, item->q,
 669                                                item->binder_type);
 670 }
 671 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
 672
 673 static int tcf_block_owner_add(struct tcf_block *block,
 674                                struct Qdisc *q,
 675                                enum tcf_block_binder_type binder_type)
 676 {
 677         struct tcf_block_owner_item *item;
 678
 679         item = kmalloc(sizeof(*item), GFP_KERNEL);
 680         if (!item)
 681                 return -ENOMEM;
 682         item->q = q;
 683         item->binder_type = binder_type;
 684         list_add(&item->list, &block->owner_list);
 685         return 0;
 686 }
 687
 688 static void tcf_block_owner_del(struct tcf_block *block,
 689                                 struct Qdisc *q,
 690                                 enum tcf_block_binder_type binder_type)
 691 {
 692         struct tcf_block_owner_item *item;
 693
 694         list_for_each_entry(item, &block->owner_list, list) {
 695                 if (item->q == q && item->binder_type == binder_type) {
 696                         list_del(&item->list);
 697                         kfree(item);
 698                         return;
 699                 }
 700         }
 701         WARN_ON(1);
 702 }
 703
 704 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 705                       struct tcf_block_ext_info *ei,
 706                       struct netlink_ext_ack *extack)
 707 {
 708         struct net *net = qdisc_net(q);
 709         struct tcf_block *block = NULL;
 710         bool created = false;
 711         int err;
 712
 713         if (ei->block_index) {
 714                 /* block_index not 0 means the shared block is requested */
 715                 block = tcf_block_lookup(net, ei->block_index);
 716                 if (block)
 717                         block->refcnt++;
 718         }
 719
 720         if (!block) {
 721                 block = tcf_block_create(net, q, ei->block_index, extack);
 722                 if (IS_ERR(block))
 723                         return PTR_ERR(block);
 724                 created = true;
 725                 if (tcf_block_shared(block)) {
 726                         err = tcf_block_insert(block, net, extack);
 727                         if (err)
 728                                 goto err_block_insert;
 729                 }
 730         }
 731
 732         err = tcf_block_owner_add(block, q, ei->binder_type);
 733         if (err)
 734                 goto err_block_owner_add;
 735
 736         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
 737
 738         err = tcf_chain0_head_change_cb_add(block, ei, extack);
 739         if (err)
 740                 goto err_chain0_head_change_cb_add;
 741
 742         err = tcf_block_offload_bind(block, q, ei, extack);
 743         if (err)
 744                 goto err_block_offload_bind;
 745
 746         *p_block = block;
 747         return 0;
 748
 749 err_block_offload_bind:
 750         tcf_chain0_head_change_cb_del(block, ei);
 751 err_chain0_head_change_cb_add:
 752         tcf_block_owner_del(block, q, ei->binder_type);
 753 err_block_owner_add:
 754         if (created) {
 755                 if (tcf_block_shared(block))
 756                         tcf_block_remove(block, net);
 757 err_block_insert:
 758                 kfree(block);
 759         } else {
 760                 block->refcnt--;
 761         }
 762         return err;
 763 }
 764 EXPORT_SYMBOL(tcf_block_get_ext);
 765
 766 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
 767 {
 768         struct tcf_proto __rcu **p_filter_chain = priv;
 769
 770         rcu_assign_pointer(*p_filter_chain, tp_head);
 771 }
 772
 773 int tcf_block_get(struct tcf_block **p_block,
 774                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
 775                   struct netlink_ext_ack *extack)
 776 {
 777         struct tcf_block_ext_info ei = {
 778                 .chain_head_change = tcf_chain_head_change_dflt,
 779                 .chain_head_change_priv = p_filter_chain,
 780         };
 781
 782         WARN_ON(!p_filter_chain);
 783         return tcf_block_get_ext(p_block, q, &ei, extack);
 784 }
 785 EXPORT_SYMBOL(tcf_block_get);
 786
 787 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
 788  * actions should be all removed after flushing.
 789  */
 790 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 791                        struct tcf_block_ext_info *ei)
 792 {
 793         struct tcf_chain *chain, *tmp;
 794
 795         if (!block)
 796                 return;
 797         tcf_chain0_head_change_cb_del(block, ei);
 798         tcf_block_owner_del(block, q, ei->binder_type);
 799
 800         if (block->refcnt == 1) {
 801                 if (tcf_block_shared(block))
 802                         tcf_block_remove(block, block->net);
 803
 804                 /* Hold a refcnt for all chains, so that they don't disappear
 805                  * while we are iterating.
 806                  */
 807                 list_for_each_entry(chain, &block->chain_list, list)
 808                         tcf_chain_hold(chain);
 809
 810                 list_for_each_entry(chain, &block->chain_list, list)
 811                         tcf_chain_flush(chain);
 812         }
 813
 814         tcf_block_offload_unbind(block, q, ei);
 815
 816         if (block->refcnt == 1) {
 817                 /* At this point, all the chains should have refcnt >= 1. */
 818                 list_for_each_entry_safe(chain, tmp, &block->chain_list, list) {
 819                         tcf_chain_put_explicitly_created(chain);
 820                         tcf_chain_put(chain);
 821                 }
 822
 823                 block->refcnt--;
 824                 if (list_empty(&block->chain_list))
 825                         kfree(block);
 826         } else {
 827                 block->refcnt--;
 828         }
 829 }
 830 EXPORT_SYMBOL(tcf_block_put_ext);
 831
 832 void tcf_block_put(struct tcf_block *block)
 833 {
 834         struct tcf_block_ext_info ei = {0, };
 835
 836         if (!block)
 837                 return;
 838         tcf_block_put_ext(block, block->q, &ei);
 839 }
 840
 841 EXPORT_SYMBOL(tcf_block_put);
 842
 843 struct tcf_block_cb {
 844         struct list_head list;
 845         tc_setup_cb_t *cb;
 846         void *cb_ident;
 847         void *cb_priv;
 848         unsigned int refcnt;
 849 };
 850
 851 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
 852 {
 853         return block_cb->cb_priv;
 854 }
 855 EXPORT_SYMBOL(tcf_block_cb_priv);
 856
 857 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
 858                                          tc_setup_cb_t *cb, void *cb_ident)
 859 {       struct tcf_block_cb *block_cb;
 860
 861         list_for_each_entry(block_cb, &block->cb_list, list)
 862                 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
 863                         return block_cb;
 864         return NULL;
 865 }
 866 EXPORT_SYMBOL(tcf_block_cb_lookup);
 867
 868 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
 869 {
 870         block_cb->refcnt++;
 871 }
 872 EXPORT_SYMBOL(tcf_block_cb_incref);
 873
 874 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
 875 {
 876         return --block_cb->refcnt;
 877 }
 878 EXPORT_SYMBOL(tcf_block_cb_decref);
 879
 880 static int
 881 tcf_block_playback_offloads(struct tcf_block *block, tc_setup_cb_t *cb,
 882                             void *cb_priv, bool add, bool offload_in_use,
 883                             struct netlink_ext_ack *extack)
 884 {
 885         struct tcf_chain *chain;
 886         struct tcf_proto *tp;
 887         int err;
 888
 889         list_for_each_entry(chain, &block->chain_list, list) {
 890                 for (tp = rtnl_dereference(chain->filter_chain); tp;
 891                      tp = rtnl_dereference(tp->next)) {
 892                         if (tp->ops->reoffload) {
 893                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
 894                                                          extack);
 895                                 if (err && add)
 896                                         goto err_playback_remove;
 897                         } else if (add && offload_in_use) {
 898                                 err = -EOPNOTSUPP;
 899                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
 900                                 goto err_playback_remove;
 901                         }
 902                 }
 903         }
 904
 905         return 0;
 906
 907 err_playback_remove:
 908         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
 909                                     extack);
 910         return err;
 911 }
 912
 913 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
 914                                              tc_setup_cb_t *cb, void *cb_ident,
 915                                              void *cb_priv,
 916                                              struct netlink_ext_ack *extack)
 917 {
 918         struct tcf_block_cb *block_cb;
 919         int err;
 920
 921         /* Replay any already present rules */
 922         err = tcf_block_playback_offloads(block, cb, cb_priv, true,
 923                                           tcf_block_offload_in_use(block),
 924                                           extack);
 925         if (err)
 926                 return ERR_PTR(err);
 927
 928         block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
 929         if (!block_cb)
 930                 return ERR_PTR(-ENOMEM);
 931         block_cb->cb = cb;
 932         block_cb->cb_ident = cb_ident;
 933         block_cb->cb_priv = cb_priv;
 934         list_add(&block_cb->list, &block->cb_list);
 935         return block_cb;
 936 }
 937 EXPORT_SYMBOL(__tcf_block_cb_register);
 938
 939 int tcf_block_cb_register(struct tcf_block *block,
 940                           tc_setup_cb_t *cb, void *cb_ident,
 941                           void *cb_priv, struct netlink_ext_ack *extack)
 942 {
 943         struct tcf_block_cb *block_cb;
 944
 945         block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv,
 946                                            extack);
 947         return PTR_ERR_OR_ZERO(block_cb);
 948 }
 949 EXPORT_SYMBOL(tcf_block_cb_register);
 950
 951 void __tcf_block_cb_unregister(struct tcf_block *block,
 952                                struct tcf_block_cb *block_cb)
 953 {
 954         tcf_block_playback_offloads(block, block_cb->cb, block_cb->cb_priv,
 955                                     false, tcf_block_offload_in_use(block),
 956                                     NULL);
 957         list_del(&block_cb->list);
 958         kfree(block_cb);
 959 }
 960 EXPORT_SYMBOL(__tcf_block_cb_unregister);
 961
 962 void tcf_block_cb_unregister(struct tcf_block *block,
 963                              tc_setup_cb_t *cb, void *cb_ident)
 964 {
 965         struct tcf_block_cb *block_cb;
 966
 967         block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
 968         if (!block_cb)
 969                 return;
 970         __tcf_block_cb_unregister(block, block_cb);
 971 }
 972 EXPORT_SYMBOL(tcf_block_cb_unregister);
 973
 974 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type,
 975                              void *type_data, bool err_stop)
 976 {
 977         struct tcf_block_cb *block_cb;
 978         int ok_count = 0;
 979         int err;
 980
 981         /* Make sure all netdevs sharing this block are offload-capable. */
 982         if (block->nooffloaddevcnt && err_stop)
 983                 return -EOPNOTSUPP;
 984
 985         list_for_each_entry(block_cb, &block->cb_list, list) {
 986                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
 987                 if (err) {
 988                         if (err_stop)
 989                                 return err;
 990                 } else {
 991                         ok_count++;
 992                 }
 993         }
 994         return ok_count;
 995 }
 996
 997 /* Main classifier routine: scans classifier chain attached
 998  * to this qdisc, (optionally) tests for protocol and asks
 999  * specific classifiers.
1000  */
1001 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1002                  struct tcf_result *res, bool compat_mode)
1003 {
1004 #ifdef CONFIG_NET_CLS_ACT
1005         const int max_reclassify_loop = 4;
1006         const struct tcf_proto *orig_tp = tp;
1007         const struct tcf_proto *first_tp;
1008         int limit = 0;
1009
1010 reclassify:
1011 #endif
1012         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1013                 __be16 protocol = skb_protocol(skb, false);
1014                 int err;
1015
1016                 if (tp->protocol != protocol &&
1017                     tp->protocol != htons(ETH_P_ALL))
1018                         continue;
1019
1020                 err = tp->classify(skb, tp, res);
1021 #ifdef CONFIG_NET_CLS_ACT
1022                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1023                         first_tp = orig_tp;
1024                         goto reset;
1025                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1026                         first_tp = res->goto_tp;
1027                         goto reset;
1028                 }
1029 #endif
1030                 if (err >= 0)
1031                         return err;
1032         }
1033
1034         return TC_ACT_UNSPEC; /* signal: continue lookup */
1035 #ifdef CONFIG_NET_CLS_ACT
1036 reset:
1037         if (unlikely(limit++ >= max_reclassify_loop)) {
1038                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1039                                        tp->chain->block->index,
1040                                        tp->prio & 0xffff,
1041                                        ntohs(tp->protocol));
1042                 return TC_ACT_SHOT;
1043         }
1044
1045         tp = first_tp;
1046         goto reclassify;
1047 #endif
1048 }
1049 EXPORT_SYMBOL(tcf_classify);
1050
1051 struct tcf_chain_info {
1052         struct tcf_proto __rcu **pprev;
1053         struct tcf_proto __rcu *next;
1054 };
1055
1056 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
1057 {
1058         return rtnl_dereference(*chain_info->pprev);
1059 }
1060
1061 static void tcf_chain_tp_insert(struct tcf_chain *chain,
1062                                 struct tcf_chain_info *chain_info,
1063                                 struct tcf_proto *tp)
1064 {
1065         if (*chain_info->pprev == chain->filter_chain)
1066                 tcf_chain0_head_change(chain, tp);
1067         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
1068         rcu_assign_pointer(*chain_info->pprev, tp);
1069         tcf_chain_hold(chain);
1070 }
1071
1072 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1073                                 struct tcf_chain_info *chain_info,
1074                                 struct tcf_proto *tp)
1075 {
1076         struct tcf_proto *next = rtnl_dereference(chain_info->next);
1077
1078         if (tp == chain->filter_chain)
1079                 tcf_chain0_head_change(chain, next);
1080         RCU_INIT_POINTER(*chain_info->pprev, next);
1081         tcf_chain_put(chain);
1082 }
1083
1084 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1085                                            struct tcf_chain_info *chain_info,
1086                                            u32 protocol, u32 prio,
1087                                            bool prio_allocate)
1088 {
1089         struct tcf_proto **pprev;
1090         struct tcf_proto *tp;
1091
1092         /* Check the chain for existence of proto-tcf with this priority */
1093         for (pprev = &chain->filter_chain;
1094              (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
1095                 if (tp->prio >= prio) {
1096                         if (tp->prio == prio) {
1097                                 if (prio_allocate ||
1098                                     (tp->protocol != protocol && protocol))
1099                                         return ERR_PTR(-EINVAL);
1100                         } else {
1101                                 tp = NULL;
1102                         }
1103                         break;
1104                 }
1105         }
1106         chain_info->pprev = pprev;
1107         chain_info->next = tp ? tp->next : NULL;
1108         return tp;
1109 }
1110
1111 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1112                          struct tcf_proto *tp, struct tcf_block *block,
1113                          struct Qdisc *q, u32 parent, void *fh,
1114                          u32 portid, u32 seq, u16 flags, int event)
1115 {
1116         struct tcmsg *tcm;
1117         struct nlmsghdr  *nlh;
1118         unsigned char *b = skb_tail_pointer(skb);
1119
1120         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1121         if (!nlh)
1122                 goto out_nlmsg_trim;
1123         tcm = nlmsg_data(nlh);
1124         tcm->tcm_family = AF_UNSPEC;
1125         tcm->tcm__pad1 = 0;
1126         tcm->tcm__pad2 = 0;
1127         if (q) {
1128                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1129                 tcm->tcm_parent = parent;
1130         } else {
1131                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1132                 tcm->tcm_block_index = block->index;
1133         }
1134         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1135         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1136                 goto nla_put_failure;
1137         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1138                 goto nla_put_failure;
1139         if (!fh) {
1140                 tcm->tcm_handle = 0;
1141         } else {
1142                 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
1143                         goto nla_put_failure;
1144         }
1145         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1146         return skb->len;
1147
1148 out_nlmsg_trim:
1149 nla_put_failure:
1150         nlmsg_trim(skb, b);
1151         return -1;
1152 }
1153
1154 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1155                           struct nlmsghdr *n, struct tcf_proto *tp,
1156                           struct tcf_block *block, struct Qdisc *q,
1157                           u32 parent, void *fh, int event, bool unicast)
1158 {
1159         struct sk_buff *skb;
1160         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1161
1162         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1163         if (!skb)
1164                 return -ENOBUFS;
1165
1166         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1167                           n->nlmsg_seq, n->nlmsg_flags, event) <= 0) {
1168                 kfree_skb(skb);
1169                 return -EINVAL;
1170         }
1171
1172         if (unicast)
1173                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1174
1175         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1176                               n->nlmsg_flags & NLM_F_ECHO);
1177 }
1178
1179 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1180                               struct nlmsghdr *n, struct tcf_proto *tp,
1181                               struct tcf_block *block, struct Qdisc *q,
1182                               u32 parent, void *fh, bool unicast, bool *last,
1183                               struct netlink_ext_ack *extack)
1184 {
1185         struct sk_buff *skb;
1186         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1187         int err;
1188
1189         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1190         if (!skb)
1191                 return -ENOBUFS;
1192
1193         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1194                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
1195                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1196                 kfree_skb(skb);
1197                 return -EINVAL;
1198         }
1199
1200         err = tp->ops->delete(tp, fh, last, extack);
1201         if (err) {
1202                 kfree_skb(skb);
1203                 return err;
1204         }
1205
1206         if (unicast)
1207                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1208
1209         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1210                              n->nlmsg_flags & NLM_F_ECHO);
1211         if (err < 0)
1212                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1213         return err;
1214 }
1215
1216 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1217                                  struct tcf_block *block, struct Qdisc *q,
1218                                  u32 parent, struct nlmsghdr *n,
1219                                  struct tcf_chain *chain, int event)
1220 {
1221         struct tcf_proto *tp;
1222
1223         for (tp = rtnl_dereference(chain->filter_chain);
1224              tp; tp = rtnl_dereference(tp->next))
1225                 tfilter_notify(net, oskb, n, tp, block,
1226                                q, parent, NULL, event, false);
1227 }
1228
1229 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1230                           struct netlink_ext_ack *extack)
1231 {
1232         struct net *net = sock_net(skb->sk);
1233         struct nlattr *tca[TCA_MAX + 1];
1234         struct tcmsg *t;
1235         u32 protocol;
1236         u32 prio;
1237         bool prio_allocate;
1238         u32 parent;
1239         u32 chain_index;
1240         struct Qdisc *q = NULL;
1241         struct tcf_chain_info chain_info;
1242         struct tcf_chain *chain = NULL;
1243         struct tcf_block *block;
1244         struct tcf_proto *tp;
1245         unsigned long cl;
1246         void *fh;
1247         int err;
1248         int tp_created;
1249
1250         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1251                 return -EPERM;
1252
1253 replay:
1254         tp_created = 0;
1255
1256         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1257         if (err < 0)
1258                 return err;
1259
1260         t = nlmsg_data(n);
1261         protocol = TC_H_MIN(t->tcm_info);
1262         prio = TC_H_MAJ(t->tcm_info);
1263         prio_allocate = false;
1264         parent = t->tcm_parent;
1265         cl = 0;
1266
1267         if (prio == 0) {
1268                 /* If no priority is provided by the user,
1269                  * we allocate one.
1270                  */
1271                 if (n->nlmsg_flags & NLM_F_CREATE) {
1272                         prio = TC_H_MAKE(0x80000000U, 0U);
1273                         prio_allocate = true;
1274                 } else {
1275                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1276                         return -ENOENT;
1277                 }
1278         }
1279
1280         /* Find head of filter chain. */
1281
1282         block = tcf_block_find(net, &q, &parent, &cl,
1283                                t->tcm_ifindex, t->tcm_block_index, extack);
1284         if (IS_ERR(block)) {
1285                 err = PTR_ERR(block);
1286                 goto errout;
1287         }
1288
1289         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1290         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1291                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1292                 err = -EINVAL;
1293                 goto errout;
1294         }
1295         chain = tcf_chain_get(block, chain_index, true);
1296         if (!chain) {
1297                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
1298                 err = -ENOMEM;
1299                 goto errout;
1300         }
1301
1302         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1303                                prio, prio_allocate);
1304         if (IS_ERR(tp)) {
1305                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1306                 err = PTR_ERR(tp);
1307                 goto errout;
1308         }
1309
1310         if (tp == NULL) {
1311                 /* Proto-tcf does not exist, create new one */
1312
1313                 if (tca[TCA_KIND] == NULL || !protocol) {
1314                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
1315                         err = -EINVAL;
1316                         goto errout;
1317                 }
1318
1319                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1320                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1321                         err = -ENOENT;
1322                         goto errout;
1323                 }
1324
1325                 if (prio_allocate)
1326                         prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
1327
1328                 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
1329                                       protocol, prio, chain, extack);
1330                 if (IS_ERR(tp)) {
1331                         err = PTR_ERR(tp);
1332                         goto errout;
1333                 }
1334                 tp_created = 1;
1335         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1336                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1337                 err = -EINVAL;
1338                 goto errout;
1339         }
1340
1341         fh = tp->ops->get(tp, t->tcm_handle);
1342
1343         if (!fh) {
1344                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1345                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1346                         err = -ENOENT;
1347                         goto errout;
1348                 }
1349         } else if (n->nlmsg_flags & NLM_F_EXCL) {
1350                 NL_SET_ERR_MSG(extack, "Filter already exists");
1351                 err = -EEXIST;
1352                 goto errout;
1353         }
1354
1355         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
1356                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
1357                 err = -EINVAL;
1358                 goto errout;
1359         }
1360
1361         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
1362                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
1363                               extack);
1364         if (err == 0) {
1365                 if (tp_created)
1366                         tcf_chain_tp_insert(chain, &chain_info, tp);
1367                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1368                                RTM_NEWTFILTER, false);
1369                 /* q pointer is NULL for shared blocks */
1370                 if (q)
1371                         q->flags &= ~TCQ_F_CAN_BYPASS;
1372         } else {
1373                 if (tp_created)
1374                         tcf_proto_destroy(tp, NULL);
1375         }
1376
1377 errout:
1378         if (chain)
1379                 tcf_chain_put(chain);
1380         tcf_block_release(q, block);
1381         if (err == -EAGAIN)
1382                 /* Replay the request. */
1383                 goto replay;
1384         return err;
1385 }
1386
1387 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1388                           struct netlink_ext_ack *extack)
1389 {
1390         struct net *net = sock_net(skb->sk);
1391         struct nlattr *tca[TCA_MAX + 1];
1392         struct tcmsg *t;
1393         u32 protocol;
1394         u32 prio;
1395         u32 parent;
1396         u32 chain_index;
1397         struct Qdisc *q = NULL;
1398         struct tcf_chain_info chain_info;
1399         struct tcf_chain *chain = NULL;
1400         struct tcf_block *block;
1401         struct tcf_proto *tp = NULL;
1402         unsigned long cl = 0;
1403         void *fh = NULL;
1404         int err;
1405
1406         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1407                 return -EPERM;
1408
1409         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1410         if (err < 0)
1411                 return err;
1412
1413         t = nlmsg_data(n);
1414         protocol = TC_H_MIN(t->tcm_info);
1415         prio = TC_H_MAJ(t->tcm_info);
1416         parent = t->tcm_parent;
1417
1418         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
1419                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
1420                 return -ENOENT;
1421         }
1422
1423         /* Find head of filter chain. */
1424
1425         block = tcf_block_find(net, &q, &parent, &cl,
1426                                t->tcm_ifindex, t->tcm_block_index, extack);
1427         if (IS_ERR(block)) {
1428                 err = PTR_ERR(block);
1429                 goto errout;
1430         }
1431
1432         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1433         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1434                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1435                 err = -EINVAL;
1436                 goto errout;
1437         }
1438         chain = tcf_chain_get(block, chain_index, false);
1439         if (!chain) {
1440                 /* User requested flush on non-existent chain. Nothing to do,
1441                  * so just return success.
1442                  */
1443                 if (prio == 0) {
1444                         err = 0;
1445                         goto errout;
1446                 }
1447                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1448                 err = -ENOENT;
1449                 goto errout;
1450         }
1451
1452         if (prio == 0) {
1453                 tfilter_notify_chain(net, skb, block, q, parent, n,
1454                                      chain, RTM_DELTFILTER);
1455                 tcf_chain_flush(chain);
1456                 err = 0;
1457                 goto errout;
1458         }
1459
1460         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1461                                prio, false);
1462         if (!tp || IS_ERR(tp)) {
1463                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1464                 err = tp ? PTR_ERR(tp) : -ENOENT;
1465                 goto errout;
1466         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1467                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1468                 err = -EINVAL;
1469                 goto errout;
1470         }
1471
1472         fh = tp->ops->get(tp, t->tcm_handle);
1473
1474         if (!fh) {
1475                 if (t->tcm_handle == 0) {
1476                         tcf_chain_tp_remove(chain, &chain_info, tp);
1477                         tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1478                                        RTM_DELTFILTER, false);
1479                         tcf_proto_destroy(tp, extack);
1480                         err = 0;
1481                 } else {
1482                         NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1483                         err = -ENOENT;
1484                 }
1485         } else {
1486                 bool last;
1487
1488                 err = tfilter_del_notify(net, skb, n, tp, block,
1489                                          q, parent, fh, false, &last,
1490                                          extack);
1491                 if (err)
1492                         goto errout;
1493                 if (last) {
1494                         tcf_chain_tp_remove(chain, &chain_info, tp);
1495                         tcf_proto_destroy(tp, extack);
1496                 }
1497         }
1498
1499 errout:
1500         if (chain)
1501                 tcf_chain_put(chain);
1502         tcf_block_release(q, block);
1503         return err;
1504 }
1505
1506 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1507                           struct netlink_ext_ack *extack)
1508 {
1509         struct net *net = sock_net(skb->sk);
1510         struct nlattr *tca[TCA_MAX + 1];
1511         struct tcmsg *t;
1512         u32 protocol;
1513         u32 prio;
1514         u32 parent;
1515         u32 chain_index;
1516         struct Qdisc *q = NULL;
1517         struct tcf_chain_info chain_info;
1518         struct tcf_chain *chain = NULL;
1519         struct tcf_block *block;
1520         struct tcf_proto *tp = NULL;
1521         unsigned long cl = 0;
1522         void *fh = NULL;
1523         int err;
1524
1525         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1526         if (err < 0)
1527                 return err;
1528
1529         t = nlmsg_data(n);
1530         protocol = TC_H_MIN(t->tcm_info);
1531         prio = TC_H_MAJ(t->tcm_info);
1532         parent = t->tcm_parent;
1533
1534         if (prio == 0) {
1535                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1536                 return -ENOENT;
1537         }
1538
1539         /* Find head of filter chain. */
1540
1541         block = tcf_block_find(net, &q, &parent, &cl,
1542                                t->tcm_ifindex, t->tcm_block_index, extack);
1543         if (IS_ERR(block)) {
1544                 err = PTR_ERR(block);
1545                 goto errout;
1546         }
1547
1548         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1549         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1550                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1551                 err = -EINVAL;
1552                 goto errout;
1553         }
1554         chain = tcf_chain_get(block, chain_index, false);
1555         if (!chain) {
1556                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1557                 err = -EINVAL;
1558                 goto errout;
1559         }
1560
1561         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1562                                prio, false);
1563         if (!tp || IS_ERR(tp)) {
1564                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1565                 err = tp ? PTR_ERR(tp) : -ENOENT;
1566                 goto errout;
1567         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1568                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1569                 err = -EINVAL;
1570                 goto errout;
1571         }
1572
1573         fh = tp->ops->get(tp, t->tcm_handle);
1574
1575         if (!fh) {
1576                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1577                 err = -ENOENT;
1578         } else {
1579                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
1580                                      fh, RTM_NEWTFILTER, true);
1581                 if (err < 0)
1582                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
1583         }
1584
1585 errout:
1586         if (chain)
1587                 tcf_chain_put(chain);
1588         tcf_block_release(q, block);
1589         return err;
1590 }
1591
1592 struct tcf_dump_args {
1593         struct tcf_walker w;
1594         struct sk_buff *skb;
1595         struct netlink_callback *cb;
1596         struct tcf_block *block;
1597         struct Qdisc *q;
1598         u32 parent;
1599 };
1600
1601 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1602 {
1603         struct tcf_dump_args *a = (void *)arg;
1604         struct net *net = sock_net(a->skb->sk);
1605
1606         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
1607                              n, NETLINK_CB(a->cb->skb).portid,
1608                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1609                              RTM_NEWTFILTER);
1610 }
1611
1612 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
1613                            struct sk_buff *skb, struct netlink_callback *cb,
1614                            long index_start, long *p_index)
1615 {
1616         struct net *net = sock_net(skb->sk);
1617         struct tcf_block *block = chain->block;
1618         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1619         struct tcf_dump_args arg;
1620         struct tcf_proto *tp;
1621
1622         for (tp = rtnl_dereference(chain->filter_chain);
1623              tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
1624                 if (*p_index < index_start)
1625                         continue;
1626                 if (TC_H_MAJ(tcm->tcm_info) &&
1627                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
1628                         continue;
1629                 if (TC_H_MIN(tcm->tcm_info) &&
1630                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
1631                         continue;
1632                 if (*p_index > index_start)
1633                         memset(&cb->args[1], 0,
1634                                sizeof(cb->args) - sizeof(cb->args[0]));
1635                 if (cb->args[1] == 0) {
1636                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
1637                                           NETLINK_CB(cb->skb).portid,
1638                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
1639                                           RTM_NEWTFILTER) <= 0)
1640                                 return false;
1641
1642                         cb->args[1] = 1;
1643                 }
1644                 if (!tp->ops->walk)
1645                         continue;
1646                 arg.w.fn = tcf_node_dump;
1647                 arg.skb = skb;
1648                 arg.cb = cb;
1649                 arg.block = block;
1650                 arg.q = q;
1651                 arg.parent = parent;
1652                 arg.w.stop = 0;
1653                 arg.w.skip = cb->args[1] - 1;
1654                 arg.w.count = 0;
1655                 arg.w.cookie = cb->args[2];
1656                 tp->ops->walk(tp, &arg.w);
1657                 cb->args[2] = arg.w.cookie;
1658                 cb->args[1] = arg.w.count + 1;
1659                 if (arg.w.stop)
1660                         return false;
1661         }
1662         return true;
1663 }
1664
1665 /* called with RTNL */
1666 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
1667 {
1668         struct net *net = sock_net(skb->sk);
1669         struct nlattr *tca[TCA_MAX + 1];
1670         struct Qdisc *q = NULL;
1671         struct tcf_block *block;
1672         struct tcf_chain *chain;
1673         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1674         long index_start;
1675         long index;
1676         u32 parent;
1677         int err;
1678
1679         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1680                 return skb->len;
1681
1682         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1683         if (err)
1684                 return err;
1685
1686         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1687                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1688                 if (!block)
1689                         goto out;
1690                 /* If we work with block index, q is NULL and parent value
1691                  * will never be used in the following code. The check
1692                  * in tcf_fill_node prevents it. However, compiler does not
1693                  * see that far, so set parent to zero to silence the warning
1694                  * about parent being uninitialized.
1695                  */
1696                 parent = 0;
1697         } else {
1698                 const struct Qdisc_class_ops *cops;
1699                 struct net_device *dev;
1700                 unsigned long cl = 0;
1701
1702                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1703                 if (!dev)
1704                         return skb->len;
1705
1706                 parent = tcm->tcm_parent;
1707                 if (!parent) {
1708                         q = dev->qdisc;
1709                         parent = q->handle;
1710                 } else {
1711                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1712                 }
1713                 if (!q)
1714                         goto out;
1715                 cops = q->ops->cl_ops;
1716                 if (!cops)
1717                         goto out;
1718                 if (!cops->tcf_block)
1719                         goto out;
1720                 if (TC_H_MIN(tcm->tcm_parent)) {
1721                         cl = cops->find(q, tcm->tcm_parent);
1722                         if (cl == 0)
1723                                 goto out;
1724                 }
1725                 block = cops->tcf_block(q, cl, NULL);
1726                 if (!block)
1727                         goto out;
1728                 if (tcf_block_shared(block))
1729                         q = NULL;
1730         }
1731
1732         index_start = cb->args[0];
1733         index = 0;
1734
1735         list_for_each_entry(chain, &block->chain_list, list) {
1736                 if (tca[TCA_CHAIN] &&
1737                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1738                         continue;
1739                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
1740                                     index_start, &index)) {
1741                         err = -EMSGSIZE;
1742                         break;
1743                 }
1744         }
1745
1746         cb->args[0] = index;
1747
1748 out:
1749         /* If we did no progress, the error (EMSGSIZE) is real */
1750         if (skb->len == 0 && err)
1751                 return err;
1752         return skb->len;
1753 }
1754
1755 static int tc_chain_fill_node(struct tcf_chain *chain, struct net *net,
1756                               struct sk_buff *skb, struct tcf_block *block,
1757                               u32 portid, u32 seq, u16 flags, int event)
1758 {
1759         unsigned char *b = skb_tail_pointer(skb);
1760         const struct tcf_proto_ops *ops;
1761         struct nlmsghdr *nlh;
1762         struct tcmsg *tcm;
1763         void *priv;
1764
1765         ops = chain->tmplt_ops;
1766         priv = chain->tmplt_priv;
1767
1768         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1769         if (!nlh)
1770                 goto out_nlmsg_trim;
1771         tcm = nlmsg_data(nlh);
1772         tcm->tcm_family = AF_UNSPEC;
1773         tcm->tcm__pad1 = 0;
1774         tcm->tcm__pad2 = 0;
1775         tcm->tcm_handle = 0;
1776         if (block->q) {
1777                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
1778                 tcm->tcm_parent = block->q->handle;
1779         } else {
1780                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1781                 tcm->tcm_block_index = block->index;
1782         }
1783
1784         if (nla_put_u32(skb, TCA_CHAIN, chain->index))
1785                 goto nla_put_failure;
1786
1787         if (ops) {
1788                 if (nla_put_string(skb, TCA_KIND, ops->kind))
1789                         goto nla_put_failure;
1790                 if (ops->tmplt_dump(skb, net, priv) < 0)
1791                         goto nla_put_failure;
1792         }
1793
1794         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1795         return skb->len;
1796
1797 out_nlmsg_trim:
1798 nla_put_failure:
1799         nlmsg_trim(skb, b);
1800         return -EMSGSIZE;
1801 }
1802
1803 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
1804                            u32 seq, u16 flags, int event, bool unicast)
1805 {
1806         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1807         struct tcf_block *block = chain->block;
1808         struct net *net = block->net;
1809         struct sk_buff *skb;
1810
1811         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1812         if (!skb)
1813                 return -ENOBUFS;
1814
1815         if (tc_chain_fill_node(chain, net, skb, block, portid,
1816                                seq, flags, event) <= 0) {
1817                 kfree_skb(skb);
1818                 return -EINVAL;
1819         }
1820
1821         if (unicast)
1822                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1823
1824         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
1825 }
1826
1827 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
1828                               struct nlattr **tca,
1829                               struct netlink_ext_ack *extack)
1830 {
1831         const struct tcf_proto_ops *ops;
1832         void *tmplt_priv;
1833
1834         /* If kind is not set, user did not specify template. */
1835         if (!tca[TCA_KIND])
1836                 return 0;
1837
1838         ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), extack);
1839         if (IS_ERR(ops))
1840                 return PTR_ERR(ops);
1841         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
1842                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
1843                 return -EOPNOTSUPP;
1844         }
1845
1846         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
1847         if (IS_ERR(tmplt_priv)) {
1848                 module_put(ops->owner);
1849                 return PTR_ERR(tmplt_priv);
1850         }
1851         chain->tmplt_ops = ops;
1852         chain->tmplt_priv = tmplt_priv;
1853         return 0;
1854 }
1855
1856 static void tc_chain_tmplt_del(struct tcf_chain *chain)
1857 {
1858         const struct tcf_proto_ops *ops = chain->tmplt_ops;
1859
1860         /* If template ops are set, no work to do for us. */
1861         if (!ops)
1862                 return;
1863
1864         ops->tmplt_destroy(chain->tmplt_priv);
1865         module_put(ops->owner);
1866 }
1867
1868 /* Add/delete/get a chain */
1869
1870 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
1871                         struct netlink_ext_ack *extack)
1872 {
1873         struct net *net = sock_net(skb->sk);
1874         struct nlattr *tca[TCA_MAX + 1];
1875         struct tcmsg *t;
1876         u32 parent;
1877         u32 chain_index;
1878         struct Qdisc *q = NULL;
1879         struct tcf_chain *chain = NULL;
1880         struct tcf_block *block;
1881         unsigned long cl;
1882         int err;
1883
1884         if (n->nlmsg_type != RTM_GETCHAIN &&
1885             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1886                 return -EPERM;
1887
1888 replay:
1889         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1890         if (err < 0)
1891                 return err;
1892
1893         t = nlmsg_data(n);
1894         parent = t->tcm_parent;
1895         cl = 0;
1896
1897         block = tcf_block_find(net, &q, &parent, &cl,
1898                                t->tcm_ifindex, t->tcm_block_index, extack);
1899         if (IS_ERR(block))
1900                 return PTR_ERR(block);
1901
1902         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1903         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1904                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1905                 err = -EINVAL;
1906                 goto errout_block;
1907         }
1908         chain = tcf_chain_lookup(block, chain_index);
1909         if (n->nlmsg_type == RTM_NEWCHAIN) {
1910                 if (chain) {
1911                         if (tcf_chain_held_by_acts_only(chain)) {
1912                                 /* The chain exists only because there is
1913                                  * some action referencing it.
1914                                  */
1915                                 tcf_chain_hold(chain);
1916                         } else {
1917                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
1918                                 err = -EEXIST;
1919                                 goto errout_block;
1920                         }
1921                 } else {
1922                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1923                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
1924                                 err = -ENOENT;
1925                                 goto errout_block;
1926                         }
1927                         chain = tcf_chain_create(block, chain_index);
1928                         if (!chain) {
1929                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
1930                                 err = -ENOMEM;
1931                                 goto errout_block;
1932                         }
1933                 }
1934         } else {
1935                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
1936                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1937                         err = -EINVAL;
1938                         goto errout_block;
1939                 }
1940                 tcf_chain_hold(chain);
1941         }
1942
1943         switch (n->nlmsg_type) {
1944         case RTM_NEWCHAIN:
1945                 err = tc_chain_tmplt_add(chain, net, tca, extack);
1946                 if (err)
1947                         goto errout;
1948                 /* In case the chain was successfully added, take a reference
1949                  * to the chain. This ensures that an empty chain
1950                  * does not disappear at the end of this function.
1951                  */
1952                 tcf_chain_hold(chain);
1953                 chain->explicitly_created = true;
1954                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
1955                                 RTM_NEWCHAIN, false);
1956                 break;
1957         case RTM_DELCHAIN:
1958                 tfilter_notify_chain(net, skb, block, q, parent, n,
1959                                      chain, RTM_DELTFILTER);
1960                 /* Flush the chain first as the user requested chain removal. */
1961                 tcf_chain_flush(chain);
1962                 /* In case the chain was successfully deleted, put a reference
1963                  * to the chain previously taken during addition.
1964                  */
1965                 tcf_chain_put_explicitly_created(chain);
1966                 chain->explicitly_created = false;
1967                 break;
1968         case RTM_GETCHAIN:
1969                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
1970                                       n->nlmsg_flags, n->nlmsg_type, true);
1971                 if (err < 0)
1972                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
1973                 break;
1974         default:
1975                 err = -EOPNOTSUPP;
1976                 NL_SET_ERR_MSG(extack, "Unsupported message type");
1977                 goto errout;
1978         }
1979
1980 errout:
1981         tcf_chain_put(chain);
1982 errout_block:
1983         tcf_block_release(q, block);
1984         if (err == -EAGAIN)
1985                 /* Replay the request. */
1986                 goto replay;
1987         return err;
1988 }
1989
1990 /* called with RTNL */
1991 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
1992 {
1993         struct net *net = sock_net(skb->sk);
1994         struct nlattr *tca[TCA_MAX + 1];
1995         struct Qdisc *q = NULL;
1996         struct tcf_block *block;
1997         struct tcf_chain *chain;
1998         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1999         long index_start;
2000         long index;
2001         u32 parent;
2002         int err;
2003
2004         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2005                 return skb->len;
2006
2007         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
2008                           NULL);
2009         if (err)
2010                 return err;
2011
2012         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2013                 block = tcf_block_lookup(net, tcm->tcm_block_index);
2014                 if (!block)
2015                         goto out;
2016                 /* If we work with block index, q is NULL and parent value
2017                  * will never be used in the following code. The check
2018                  * in tcf_fill_node prevents it. However, compiler does not
2019                  * see that far, so set parent to zero to silence the warning
2020                  * about parent being uninitialized.
2021                  */
2022                 parent = 0;
2023         } else {
2024                 const struct Qdisc_class_ops *cops;
2025                 struct net_device *dev;
2026                 unsigned long cl = 0;
2027
2028                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2029                 if (!dev)
2030                         return skb->len;
2031
2032                 parent = tcm->tcm_parent;
2033                 if (!parent) {
2034                         q = dev->qdisc;
2035                         parent = q->handle;
2036                 } else {
2037                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2038                 }
2039                 if (!q)
2040                         goto out;
2041                 cops = q->ops->cl_ops;
2042                 if (!cops)
2043                         goto out;
2044                 if (!cops->tcf_block)
2045                         goto out;
2046                 if (TC_H_MIN(tcm->tcm_parent)) {
2047                         cl = cops->find(q, tcm->tcm_parent);
2048                         if (cl == 0)
2049                                 goto out;
2050                 }
2051                 block = cops->tcf_block(q, cl, NULL);
2052                 if (!block)
2053                         goto out;
2054                 if (tcf_block_shared(block))
2055                         q = NULL;
2056         }
2057
2058         index_start = cb->args[0];
2059         index = 0;
2060
2061         list_for_each_entry(chain, &block->chain_list, list) {
2062                 if ((tca[TCA_CHAIN] &&
2063                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2064                         continue;
2065                 if (index < index_start) {
2066                         index++;
2067                         continue;
2068                 }
2069                 if (tcf_chain_held_by_acts_only(chain))
2070                         continue;
2071                 err = tc_chain_fill_node(chain, net, skb, block,
2072                                          NETLINK_CB(cb->skb).portid,
2073                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
2074                                          RTM_NEWCHAIN);
2075                 if (err <= 0)
2076                         break;
2077                 index++;
2078         }
2079
2080         cb->args[0] = index;
2081
2082 out:
2083         /* If we did no progress, the error (EMSGSIZE) is real */
2084         if (skb->len == 0 && err)
2085                 return err;
2086         return skb->len;
2087 }
2088
2089 void tcf_exts_destroy(struct tcf_exts *exts)
2090 {
2091 #ifdef CONFIG_NET_CLS_ACT
2092         if (exts->actions) {
2093                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
2094                 kfree(exts->actions);
2095         }
2096         exts->nr_actions = 0;
2097 #endif
2098 }
2099 EXPORT_SYMBOL(tcf_exts_destroy);
2100
2101 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
2102                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
2103                       struct netlink_ext_ack *extack)
2104 {
2105 #ifdef CONFIG_NET_CLS_ACT
2106         {
2107                 struct tc_action *act;
2108                 size_t attr_size = 0;
2109
2110                 if (exts->police && tb[exts->police]) {
2111                         act = tcf_action_init_1(net, tp, tb[exts->police],
2112                                                 rate_tlv, "police", ovr,
2113                                                 TCA_ACT_BIND, true, extack);
2114                         if (IS_ERR(act))
2115                                 return PTR_ERR(act);
2116
2117                         act->type = exts->type = TCA_OLD_COMPAT;
2118                         exts->actions[0] = act;
2119                         exts->nr_actions = 1;
2120                 } else if (exts->action && tb[exts->action]) {
2121                         int err;
2122
2123                         err = tcf_action_init(net, tp, tb[exts->action],
2124                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
2125                                               exts->actions, &attr_size, true,
2126                                               extack);
2127                         if (err < 0)
2128                                 return err;
2129                         exts->nr_actions = err;
2130                 }
2131                 exts->net = net;
2132         }
2133 #else
2134         if ((exts->action && tb[exts->action]) ||
2135             (exts->police && tb[exts->police])) {
2136                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
2137                 return -EOPNOTSUPP;
2138         }
2139 #endif
2140
2141         return 0;
2142 }
2143 EXPORT_SYMBOL(tcf_exts_validate);
2144
2145 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
2146 {
2147 #ifdef CONFIG_NET_CLS_ACT
2148         struct tcf_exts old = *dst;
2149
2150         *dst = *src;
2151         tcf_exts_destroy(&old);
2152 #endif
2153 }
2154 EXPORT_SYMBOL(tcf_exts_change);
2155
2156 #ifdef CONFIG_NET_CLS_ACT
2157 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
2158 {
2159         if (exts->nr_actions == 0)
2160                 return NULL;
2161         else
2162                 return exts->actions[0];
2163 }
2164 #endif
2165
2166 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
2167 {
2168 #ifdef CONFIG_NET_CLS_ACT
2169         struct nlattr *nest;
2170
2171         if (exts->action && tcf_exts_has_actions(exts)) {
2172                 /*
2173                  * again for backward compatible mode - we want
2174                  * to work with both old and new modes of entering
2175                  * tc data even if iproute2  was newer - jhs
2176                  */
2177                 if (exts->type != TCA_OLD_COMPAT) {
2178                         nest = nla_nest_start(skb, exts->action);
2179                         if (nest == NULL)
2180                                 goto nla_put_failure;
2181
2182                         if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
2183                                 goto nla_put_failure;
2184                         nla_nest_end(skb, nest);
2185                 } else if (exts->police) {
2186                         struct tc_action *act = tcf_exts_first_act(exts);
2187                         nest = nla_nest_start(skb, exts->police);
2188                         if (nest == NULL || !act)
2189                                 goto nla_put_failure;
2190                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
2191                                 goto nla_put_failure;
2192                         nla_nest_end(skb, nest);
2193                 }
2194         }
2195         return 0;
2196
2197 nla_put_failure:
2198         nla_nest_cancel(skb, nest);
2199         return -1;
2200 #else
2201         return 0;
2202 #endif
2203 }
2204 EXPORT_SYMBOL(tcf_exts_dump);
2205
2206
2207 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
2208 {
2209 #ifdef CONFIG_NET_CLS_ACT
2210         struct tc_action *a = tcf_exts_first_act(exts);
2211         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
2212                 return -1;
2213 #endif
2214         return 0;
2215 }
2216 EXPORT_SYMBOL(tcf_exts_dump_stats);
2217
2218 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
2219                                        enum tc_setup_type type,
2220                                        void *type_data, bool err_stop)
2221 {
2222         int ok_count = 0;
2223 #ifdef CONFIG_NET_CLS_ACT
2224         const struct tc_action *a;
2225         struct net_device *dev;
2226         int i, ret;
2227
2228         if (!tcf_exts_has_actions(exts))
2229                 return 0;
2230
2231         for (i = 0; i < exts->nr_actions; i++) {
2232                 a = exts->actions[i];
2233                 if (!a->ops->get_dev)
2234                         continue;
2235                 dev = a->ops->get_dev(a);
2236                 if (!dev)
2237                         continue;
2238                 ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
2239                 a->ops->put_dev(dev);
2240                 if (ret < 0)
2241                         return ret;
2242                 ok_count += ret;
2243         }
2244 #endif
2245         return ok_count;
2246 }
2247
2248 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
2249                      enum tc_setup_type type, void *type_data, bool err_stop)
2250 {
2251         int ok_count;
2252         int ret;
2253
2254         ret = tcf_block_cb_call(block, type, type_data, err_stop);
2255         if (ret < 0)
2256                 return ret;
2257         ok_count = ret;
2258
2259         if (!exts || ok_count)
2260                 return ok_count;
2261         ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
2262         if (ret < 0)
2263                 return ret;
2264         ok_count += ret;
2265
2266         return ok_count;
2267 }
2268 EXPORT_SYMBOL(tc_setup_cb_call);
2269
2270 static __net_init int tcf_net_init(struct net *net)
2271 {
2272         struct tcf_net *tn = net_generic(net, tcf_net_id);
2273
2274         idr_init(&tn->idr);
2275         return 0;
2276 }
2277
2278 static void __net_exit tcf_net_exit(struct net *net)
2279 {
2280         struct tcf_net *tn = net_generic(net, tcf_net_id);
2281
2282         idr_destroy(&tn->idr);
2283 }
2284
2285 static struct pernet_operations tcf_net_ops = {
2286         .init = tcf_net_init,
2287         .exit = tcf_net_exit,
2288         .id   = &tcf_net_id,
2289         .size = sizeof(struct tcf_net),
2290 };
2291
2292 static int __init tc_filter_init(void)
2293 {
2294         int err;
2295
2296         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
2297         if (!tc_filter_wq)
2298                 return -ENOMEM;
2299
2300         err = register_pernet_subsys(&tcf_net_ops);
2301         if (err)
2302                 goto err_register_pernet_subsys;
2303
2304         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 0);
2305         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 0);
2306         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
2307                       tc_dump_tfilter, 0);
2308         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
2309         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
2310         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
2311                       tc_dump_chain, 0);
2312
2313         return 0;
2314
2315 err_register_pernet_subsys:
2316         destroy_workqueue(tc_filter_wq);
2317         return err;
2318 }
2319
2320 subsys_initcall(tc_filter_init);