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
 543         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
 544                 block = tcf_block_lookup(net, block_index);
 545                 if (!block) {
 546                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
 547                         return ERR_PTR(-EINVAL);
 548                 }
 549         } else {
 550                 const struct Qdisc_class_ops *cops;
 551                 struct net_device *dev;
 552
 553                 /* Find link */
 554                 dev = __dev_get_by_index(net, ifindex);
 555                 if (!dev)
 556                         return ERR_PTR(-ENODEV);
 557
 558                 /* Find qdisc */
 559                 if (!*parent) {
 560                         *q = dev->qdisc;
 561                         *parent = (*q)->handle;
 562                 } else {
 563                         *q = qdisc_lookup(dev, TC_H_MAJ(*parent));
 564                         if (!*q) {
 565                                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
 566                                 return ERR_PTR(-EINVAL);
 567                         }
 568                 }
 569
 570                 /* Is it classful? */
 571                 cops = (*q)->ops->cl_ops;
 572                 if (!cops) {
 573                         NL_SET_ERR_MSG(extack, "Qdisc not classful");
 574                         return ERR_PTR(-EINVAL);
 575                 }
 576
 577                 if (!cops->tcf_block) {
 578                         NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
 579                         return ERR_PTR(-EOPNOTSUPP);
 580                 }
 581
 582                 /* Do we search for filter, attached to class? */
 583                 if (TC_H_MIN(*parent)) {
 584                         *cl = cops->find(*q, *parent);
 585                         if (*cl == 0) {
 586                                 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
 587                                 return ERR_PTR(-ENOENT);
 588                         }
 589                 }
 590
 591                 /* And the last stroke */
 592                 block = cops->tcf_block(*q, *cl, extack);
 593                 if (!block)
 594                         return ERR_PTR(-EINVAL);
 595                 if (tcf_block_shared(block)) {
 596                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
 597                         return ERR_PTR(-EOPNOTSUPP);
 598                 }
 599         }
 600
 601         return block;
 602 }
 603
 604 struct tcf_block_owner_item {
 605         struct list_head list;
 606         struct Qdisc *q;
 607         enum tcf_block_binder_type binder_type;
 608 };
 609
 610 static void
 611 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
 612                                struct Qdisc *q,
 613                                enum tcf_block_binder_type binder_type)
 614 {
 615         if (block->keep_dst &&
 616             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
 617             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
 618                 netif_keep_dst(qdisc_dev(q));
 619 }
 620
 621 void tcf_block_netif_keep_dst(struct tcf_block *block)
 622 {
 623         struct tcf_block_owner_item *item;
 624
 625         block->keep_dst = true;
 626         list_for_each_entry(item, &block->owner_list, list)
 627                 tcf_block_owner_netif_keep_dst(block, item->q,
 628                                                item->binder_type);
 629 }
 630 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
 631
 632 static int tcf_block_owner_add(struct tcf_block *block,
 633                                struct Qdisc *q,
 634                                enum tcf_block_binder_type binder_type)
 635 {
 636         struct tcf_block_owner_item *item;
 637
 638         item = kmalloc(sizeof(*item), GFP_KERNEL);
 639         if (!item)
 640                 return -ENOMEM;
 641         item->q = q;
 642         item->binder_type = binder_type;
 643         list_add(&item->list, &block->owner_list);
 644         return 0;
 645 }
 646
 647 static void tcf_block_owner_del(struct tcf_block *block,
 648                                 struct Qdisc *q,
 649                                 enum tcf_block_binder_type binder_type)
 650 {
 651         struct tcf_block_owner_item *item;
 652
 653         list_for_each_entry(item, &block->owner_list, list) {
 654                 if (item->q == q && item->binder_type == binder_type) {
 655                         list_del(&item->list);
 656                         kfree(item);
 657                         return;
 658                 }
 659         }
 660         WARN_ON(1);
 661 }
 662
 663 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 664                       struct tcf_block_ext_info *ei,
 665                       struct netlink_ext_ack *extack)
 666 {
 667         struct net *net = qdisc_net(q);
 668         struct tcf_block *block = NULL;
 669         bool created = false;
 670         int err;
 671
 672         if (ei->block_index) {
 673                 /* block_index not 0 means the shared block is requested */
 674                 block = tcf_block_lookup(net, ei->block_index);
 675                 if (block)
 676                         block->refcnt++;
 677         }
 678
 679         if (!block) {
 680                 block = tcf_block_create(net, q, ei->block_index, extack);
 681                 if (IS_ERR(block))
 682                         return PTR_ERR(block);
 683                 created = true;
 684                 if (tcf_block_shared(block)) {
 685                         err = tcf_block_insert(block, net, extack);
 686                         if (err)
 687                                 goto err_block_insert;
 688                 }
 689         }
 690
 691         err = tcf_block_owner_add(block, q, ei->binder_type);
 692         if (err)
 693                 goto err_block_owner_add;
 694
 695         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
 696
 697         err = tcf_chain0_head_change_cb_add(block, ei, extack);
 698         if (err)
 699                 goto err_chain0_head_change_cb_add;
 700
 701         err = tcf_block_offload_bind(block, q, ei, extack);
 702         if (err)
 703                 goto err_block_offload_bind;
 704
 705         *p_block = block;
 706         return 0;
 707
 708 err_block_offload_bind:
 709         tcf_chain0_head_change_cb_del(block, ei);
 710 err_chain0_head_change_cb_add:
 711         tcf_block_owner_del(block, q, ei->binder_type);
 712 err_block_owner_add:
 713         if (created) {
 714                 if (tcf_block_shared(block))
 715                         tcf_block_remove(block, net);
 716 err_block_insert:
 717                 kfree(block);
 718         } else {
 719                 block->refcnt--;
 720         }
 721         return err;
 722 }
 723 EXPORT_SYMBOL(tcf_block_get_ext);
 724
 725 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
 726 {
 727         struct tcf_proto __rcu **p_filter_chain = priv;
 728
 729         rcu_assign_pointer(*p_filter_chain, tp_head);
 730 }
 731
 732 int tcf_block_get(struct tcf_block **p_block,
 733                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
 734                   struct netlink_ext_ack *extack)
 735 {
 736         struct tcf_block_ext_info ei = {
 737                 .chain_head_change = tcf_chain_head_change_dflt,
 738                 .chain_head_change_priv = p_filter_chain,
 739         };
 740
 741         WARN_ON(!p_filter_chain);
 742         return tcf_block_get_ext(p_block, q, &ei, extack);
 743 }
 744 EXPORT_SYMBOL(tcf_block_get);
 745
 746 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
 747  * actions should be all removed after flushing.
 748  */
 749 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 750                        struct tcf_block_ext_info *ei)
 751 {
 752         struct tcf_chain *chain, *tmp;
 753
 754         if (!block)
 755                 return;
 756         tcf_chain0_head_change_cb_del(block, ei);
 757         tcf_block_owner_del(block, q, ei->binder_type);
 758
 759         if (block->refcnt == 1) {
 760                 if (tcf_block_shared(block))
 761                         tcf_block_remove(block, block->net);
 762
 763                 /* Hold a refcnt for all chains, so that they don't disappear
 764                  * while we are iterating.
 765                  */
 766                 list_for_each_entry(chain, &block->chain_list, list)
 767                         tcf_chain_hold(chain);
 768
 769                 list_for_each_entry(chain, &block->chain_list, list)
 770                         tcf_chain_flush(chain);
 771         }
 772
 773         tcf_block_offload_unbind(block, q, ei);
 774
 775         if (block->refcnt == 1) {
 776                 /* At this point, all the chains should have refcnt >= 1. */
 777                 list_for_each_entry_safe(chain, tmp, &block->chain_list, list) {
 778                         tcf_chain_put_explicitly_created(chain);
 779                         tcf_chain_put(chain);
 780                 }
 781
 782                 block->refcnt--;
 783                 if (list_empty(&block->chain_list))
 784                         kfree(block);
 785         } else {
 786                 block->refcnt--;
 787         }
 788 }
 789 EXPORT_SYMBOL(tcf_block_put_ext);
 790
 791 void tcf_block_put(struct tcf_block *block)
 792 {
 793         struct tcf_block_ext_info ei = {0, };
 794
 795         if (!block)
 796                 return;
 797         tcf_block_put_ext(block, block->q, &ei);
 798 }
 799
 800 EXPORT_SYMBOL(tcf_block_put);
 801
 802 struct tcf_block_cb {
 803         struct list_head list;
 804         tc_setup_cb_t *cb;
 805         void *cb_ident;
 806         void *cb_priv;
 807         unsigned int refcnt;
 808 };
 809
 810 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
 811 {
 812         return block_cb->cb_priv;
 813 }
 814 EXPORT_SYMBOL(tcf_block_cb_priv);
 815
 816 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
 817                                          tc_setup_cb_t *cb, void *cb_ident)
 818 {       struct tcf_block_cb *block_cb;
 819
 820         list_for_each_entry(block_cb, &block->cb_list, list)
 821                 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
 822                         return block_cb;
 823         return NULL;
 824 }
 825 EXPORT_SYMBOL(tcf_block_cb_lookup);
 826
 827 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
 828 {
 829         block_cb->refcnt++;
 830 }
 831 EXPORT_SYMBOL(tcf_block_cb_incref);
 832
 833 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
 834 {
 835         return --block_cb->refcnt;
 836 }
 837 EXPORT_SYMBOL(tcf_block_cb_decref);
 838
 839 static int
 840 tcf_block_playback_offloads(struct tcf_block *block, tc_setup_cb_t *cb,
 841                             void *cb_priv, bool add, bool offload_in_use,
 842                             struct netlink_ext_ack *extack)
 843 {
 844         struct tcf_chain *chain;
 845         struct tcf_proto *tp;
 846         int err;
 847
 848         list_for_each_entry(chain, &block->chain_list, list) {
 849                 for (tp = rtnl_dereference(chain->filter_chain); tp;
 850                      tp = rtnl_dereference(tp->next)) {
 851                         if (tp->ops->reoffload) {
 852                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
 853                                                          extack);
 854                                 if (err && add)
 855                                         goto err_playback_remove;
 856                         } else if (add && offload_in_use) {
 857                                 err = -EOPNOTSUPP;
 858                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
 859                                 goto err_playback_remove;
 860                         }
 861                 }
 862         }
 863
 864         return 0;
 865
 866 err_playback_remove:
 867         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
 868                                     extack);
 869         return err;
 870 }
 871
 872 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
 873                                              tc_setup_cb_t *cb, void *cb_ident,
 874                                              void *cb_priv,
 875                                              struct netlink_ext_ack *extack)
 876 {
 877         struct tcf_block_cb *block_cb;
 878         int err;
 879
 880         /* Replay any already present rules */
 881         err = tcf_block_playback_offloads(block, cb, cb_priv, true,
 882                                           tcf_block_offload_in_use(block),
 883                                           extack);
 884         if (err)
 885                 return ERR_PTR(err);
 886
 887         block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
 888         if (!block_cb)
 889                 return ERR_PTR(-ENOMEM);
 890         block_cb->cb = cb;
 891         block_cb->cb_ident = cb_ident;
 892         block_cb->cb_priv = cb_priv;
 893         list_add(&block_cb->list, &block->cb_list);
 894         return block_cb;
 895 }
 896 EXPORT_SYMBOL(__tcf_block_cb_register);
 897
 898 int tcf_block_cb_register(struct tcf_block *block,
 899                           tc_setup_cb_t *cb, void *cb_ident,
 900                           void *cb_priv, struct netlink_ext_ack *extack)
 901 {
 902         struct tcf_block_cb *block_cb;
 903
 904         block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv,
 905                                            extack);
 906         return PTR_ERR_OR_ZERO(block_cb);
 907 }
 908 EXPORT_SYMBOL(tcf_block_cb_register);
 909
 910 void __tcf_block_cb_unregister(struct tcf_block *block,
 911                                struct tcf_block_cb *block_cb)
 912 {
 913         tcf_block_playback_offloads(block, block_cb->cb, block_cb->cb_priv,
 914                                     false, tcf_block_offload_in_use(block),
 915                                     NULL);
 916         list_del(&block_cb->list);
 917         kfree(block_cb);
 918 }
 919 EXPORT_SYMBOL(__tcf_block_cb_unregister);
 920
 921 void tcf_block_cb_unregister(struct tcf_block *block,
 922                              tc_setup_cb_t *cb, void *cb_ident)
 923 {
 924         struct tcf_block_cb *block_cb;
 925
 926         block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
 927         if (!block_cb)
 928                 return;
 929         __tcf_block_cb_unregister(block, block_cb);
 930 }
 931 EXPORT_SYMBOL(tcf_block_cb_unregister);
 932
 933 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type,
 934                              void *type_data, bool err_stop)
 935 {
 936         struct tcf_block_cb *block_cb;
 937         int ok_count = 0;
 938         int err;
 939
 940         /* Make sure all netdevs sharing this block are offload-capable. */
 941         if (block->nooffloaddevcnt && err_stop)
 942                 return -EOPNOTSUPP;
 943
 944         list_for_each_entry(block_cb, &block->cb_list, list) {
 945                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
 946                 if (err) {
 947                         if (err_stop)
 948                                 return err;
 949                 } else {
 950                         ok_count++;
 951                 }
 952         }
 953         return ok_count;
 954 }
 955
 956 /* Main classifier routine: scans classifier chain attached
 957  * to this qdisc, (optionally) tests for protocol and asks
 958  * specific classifiers.
 959  */
 960 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 961                  struct tcf_result *res, bool compat_mode)
 962 {
 963 #ifdef CONFIG_NET_CLS_ACT
 964         const int max_reclassify_loop = 4;
 965         const struct tcf_proto *orig_tp = tp;
 966         const struct tcf_proto *first_tp;
 967         int limit = 0;
 968
 969 reclassify:
 970 #endif
 971         for (; tp; tp = rcu_dereference_bh(tp->next)) {
 972                 __be16 protocol = skb_protocol(skb, false);
 973                 int err;
 974
 975                 if (tp->protocol != protocol &&
 976                     tp->protocol != htons(ETH_P_ALL))
 977                         continue;
 978
 979                 err = tp->classify(skb, tp, res);
 980 #ifdef CONFIG_NET_CLS_ACT
 981                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
 982                         first_tp = orig_tp;
 983                         goto reset;
 984                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
 985                         first_tp = res->goto_tp;
 986                         goto reset;
 987                 }
 988 #endif
 989                 if (err >= 0)
 990                         return err;
 991         }
 992
 993         return TC_ACT_UNSPEC; /* signal: continue lookup */
 994 #ifdef CONFIG_NET_CLS_ACT
 995 reset:
 996         if (unlikely(limit++ >= max_reclassify_loop)) {
 997                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
 998                                        tp->chain->block->index,
 999                                        tp->prio & 0xffff,
1000                                        ntohs(tp->protocol));
1001                 return TC_ACT_SHOT;
1002         }
1003
1004         tp = first_tp;
1005         goto reclassify;
1006 #endif
1007 }
1008 EXPORT_SYMBOL(tcf_classify);
1009
1010 struct tcf_chain_info {
1011         struct tcf_proto __rcu **pprev;
1012         struct tcf_proto __rcu *next;
1013 };
1014
1015 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
1016 {
1017         return rtnl_dereference(*chain_info->pprev);
1018 }
1019
1020 static void tcf_chain_tp_insert(struct tcf_chain *chain,
1021                                 struct tcf_chain_info *chain_info,
1022                                 struct tcf_proto *tp)
1023 {
1024         if (*chain_info->pprev == chain->filter_chain)
1025                 tcf_chain0_head_change(chain, tp);
1026         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
1027         rcu_assign_pointer(*chain_info->pprev, tp);
1028         tcf_chain_hold(chain);
1029 }
1030
1031 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1032                                 struct tcf_chain_info *chain_info,
1033                                 struct tcf_proto *tp)
1034 {
1035         struct tcf_proto *next = rtnl_dereference(chain_info->next);
1036
1037         if (tp == chain->filter_chain)
1038                 tcf_chain0_head_change(chain, next);
1039         RCU_INIT_POINTER(*chain_info->pprev, next);
1040         tcf_chain_put(chain);
1041 }
1042
1043 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1044                                            struct tcf_chain_info *chain_info,
1045                                            u32 protocol, u32 prio,
1046                                            bool prio_allocate)
1047 {
1048         struct tcf_proto **pprev;
1049         struct tcf_proto *tp;
1050
1051         /* Check the chain for existence of proto-tcf with this priority */
1052         for (pprev = &chain->filter_chain;
1053              (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
1054                 if (tp->prio >= prio) {
1055                         if (tp->prio == prio) {
1056                                 if (prio_allocate ||
1057                                     (tp->protocol != protocol && protocol))
1058                                         return ERR_PTR(-EINVAL);
1059                         } else {
1060                                 tp = NULL;
1061                         }
1062                         break;
1063                 }
1064         }
1065         chain_info->pprev = pprev;
1066         chain_info->next = tp ? tp->next : NULL;
1067         return tp;
1068 }
1069
1070 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1071                          struct tcf_proto *tp, struct tcf_block *block,
1072                          struct Qdisc *q, u32 parent, void *fh,
1073                          u32 portid, u32 seq, u16 flags, int event)
1074 {
1075         struct tcmsg *tcm;
1076         struct nlmsghdr  *nlh;
1077         unsigned char *b = skb_tail_pointer(skb);
1078
1079         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1080         if (!nlh)
1081                 goto out_nlmsg_trim;
1082         tcm = nlmsg_data(nlh);
1083         tcm->tcm_family = AF_UNSPEC;
1084         tcm->tcm__pad1 = 0;
1085         tcm->tcm__pad2 = 0;
1086         if (q) {
1087                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1088                 tcm->tcm_parent = parent;
1089         } else {
1090                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1091                 tcm->tcm_block_index = block->index;
1092         }
1093         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1094         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1095                 goto nla_put_failure;
1096         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1097                 goto nla_put_failure;
1098         if (!fh) {
1099                 tcm->tcm_handle = 0;
1100         } else {
1101                 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
1102                         goto nla_put_failure;
1103         }
1104         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1105         return skb->len;
1106
1107 out_nlmsg_trim:
1108 nla_put_failure:
1109         nlmsg_trim(skb, b);
1110         return -1;
1111 }
1112
1113 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1114                           struct nlmsghdr *n, struct tcf_proto *tp,
1115                           struct tcf_block *block, struct Qdisc *q,
1116                           u32 parent, void *fh, int event, bool unicast)
1117 {
1118         struct sk_buff *skb;
1119         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1120
1121         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1122         if (!skb)
1123                 return -ENOBUFS;
1124
1125         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1126                           n->nlmsg_seq, n->nlmsg_flags, event) <= 0) {
1127                 kfree_skb(skb);
1128                 return -EINVAL;
1129         }
1130
1131         if (unicast)
1132                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1133
1134         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1135                               n->nlmsg_flags & NLM_F_ECHO);
1136 }
1137
1138 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1139                               struct nlmsghdr *n, struct tcf_proto *tp,
1140                               struct tcf_block *block, struct Qdisc *q,
1141                               u32 parent, void *fh, bool unicast, bool *last,
1142                               struct netlink_ext_ack *extack)
1143 {
1144         struct sk_buff *skb;
1145         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1146         int err;
1147
1148         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1149         if (!skb)
1150                 return -ENOBUFS;
1151
1152         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1153                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
1154                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1155                 kfree_skb(skb);
1156                 return -EINVAL;
1157         }
1158
1159         err = tp->ops->delete(tp, fh, last, extack);
1160         if (err) {
1161                 kfree_skb(skb);
1162                 return err;
1163         }
1164
1165         if (unicast)
1166                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1167
1168         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1169                              n->nlmsg_flags & NLM_F_ECHO);
1170         if (err < 0)
1171                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1172         return err;
1173 }
1174
1175 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1176                                  struct tcf_block *block, struct Qdisc *q,
1177                                  u32 parent, struct nlmsghdr *n,
1178                                  struct tcf_chain *chain, int event)
1179 {
1180         struct tcf_proto *tp;
1181
1182         for (tp = rtnl_dereference(chain->filter_chain);
1183              tp; tp = rtnl_dereference(tp->next))
1184                 tfilter_notify(net, oskb, n, tp, block,
1185                                q, parent, NULL, event, false);
1186 }
1187
1188 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1189                           struct netlink_ext_ack *extack)
1190 {
1191         struct net *net = sock_net(skb->sk);
1192         struct nlattr *tca[TCA_MAX + 1];
1193         struct tcmsg *t;
1194         u32 protocol;
1195         u32 prio;
1196         bool prio_allocate;
1197         u32 parent;
1198         u32 chain_index;
1199         struct Qdisc *q = NULL;
1200         struct tcf_chain_info chain_info;
1201         struct tcf_chain *chain = NULL;
1202         struct tcf_block *block;
1203         struct tcf_proto *tp;
1204         unsigned long cl;
1205         void *fh;
1206         int err;
1207         int tp_created;
1208
1209         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1210                 return -EPERM;
1211
1212 replay:
1213         tp_created = 0;
1214
1215         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1216         if (err < 0)
1217                 return err;
1218
1219         t = nlmsg_data(n);
1220         protocol = TC_H_MIN(t->tcm_info);
1221         prio = TC_H_MAJ(t->tcm_info);
1222         prio_allocate = false;
1223         parent = t->tcm_parent;
1224         cl = 0;
1225
1226         if (prio == 0) {
1227                 /* If no priority is provided by the user,
1228                  * we allocate one.
1229                  */
1230                 if (n->nlmsg_flags & NLM_F_CREATE) {
1231                         prio = TC_H_MAKE(0x80000000U, 0U);
1232                         prio_allocate = true;
1233                 } else {
1234                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1235                         return -ENOENT;
1236                 }
1237         }
1238
1239         /* Find head of filter chain. */
1240
1241         block = tcf_block_find(net, &q, &parent, &cl,
1242                                t->tcm_ifindex, t->tcm_block_index, extack);
1243         if (IS_ERR(block)) {
1244                 err = PTR_ERR(block);
1245                 goto errout;
1246         }
1247
1248         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1249         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1250                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1251                 err = -EINVAL;
1252                 goto errout;
1253         }
1254         chain = tcf_chain_get(block, chain_index, true);
1255         if (!chain) {
1256                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
1257                 err = -ENOMEM;
1258                 goto errout;
1259         }
1260
1261         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1262                                prio, prio_allocate);
1263         if (IS_ERR(tp)) {
1264                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1265                 err = PTR_ERR(tp);
1266                 goto errout;
1267         }
1268
1269         if (tp == NULL) {
1270                 /* Proto-tcf does not exist, create new one */
1271
1272                 if (tca[TCA_KIND] == NULL || !protocol) {
1273                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
1274                         err = -EINVAL;
1275                         goto errout;
1276                 }
1277
1278                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1279                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1280                         err = -ENOENT;
1281                         goto errout;
1282                 }
1283
1284                 if (prio_allocate)
1285                         prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
1286
1287                 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
1288                                       protocol, prio, chain, extack);
1289                 if (IS_ERR(tp)) {
1290                         err = PTR_ERR(tp);
1291                         goto errout;
1292                 }
1293                 tp_created = 1;
1294         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1295                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1296                 err = -EINVAL;
1297                 goto errout;
1298         }
1299
1300         fh = tp->ops->get(tp, t->tcm_handle);
1301
1302         if (!fh) {
1303                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1304                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1305                         err = -ENOENT;
1306                         goto errout;
1307                 }
1308         } else if (n->nlmsg_flags & NLM_F_EXCL) {
1309                 NL_SET_ERR_MSG(extack, "Filter already exists");
1310                 err = -EEXIST;
1311                 goto errout;
1312         }
1313
1314         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
1315                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
1316                 err = -EINVAL;
1317                 goto errout;
1318         }
1319
1320         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
1321                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
1322                               extack);
1323         if (err == 0) {
1324                 if (tp_created)
1325                         tcf_chain_tp_insert(chain, &chain_info, tp);
1326                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1327                                RTM_NEWTFILTER, false);
1328                 /* q pointer is NULL for shared blocks */
1329                 if (q)
1330                         q->flags &= ~TCQ_F_CAN_BYPASS;
1331         } else {
1332                 if (tp_created)
1333                         tcf_proto_destroy(tp, NULL);
1334         }
1335
1336 errout:
1337         if (chain)
1338                 tcf_chain_put(chain);
1339         if (err == -EAGAIN)
1340                 /* Replay the request. */
1341                 goto replay;
1342         return err;
1343 }
1344
1345 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1346                           struct netlink_ext_ack *extack)
1347 {
1348         struct net *net = sock_net(skb->sk);
1349         struct nlattr *tca[TCA_MAX + 1];
1350         struct tcmsg *t;
1351         u32 protocol;
1352         u32 prio;
1353         u32 parent;
1354         u32 chain_index;
1355         struct Qdisc *q = NULL;
1356         struct tcf_chain_info chain_info;
1357         struct tcf_chain *chain = NULL;
1358         struct tcf_block *block;
1359         struct tcf_proto *tp = NULL;
1360         unsigned long cl = 0;
1361         void *fh = NULL;
1362         int err;
1363
1364         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1365                 return -EPERM;
1366
1367         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1368         if (err < 0)
1369                 return err;
1370
1371         t = nlmsg_data(n);
1372         protocol = TC_H_MIN(t->tcm_info);
1373         prio = TC_H_MAJ(t->tcm_info);
1374         parent = t->tcm_parent;
1375
1376         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
1377                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
1378                 return -ENOENT;
1379         }
1380
1381         /* Find head of filter chain. */
1382
1383         block = tcf_block_find(net, &q, &parent, &cl,
1384                                t->tcm_ifindex, t->tcm_block_index, extack);
1385         if (IS_ERR(block)) {
1386                 err = PTR_ERR(block);
1387                 goto errout;
1388         }
1389
1390         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1391         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1392                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1393                 err = -EINVAL;
1394                 goto errout;
1395         }
1396         chain = tcf_chain_get(block, chain_index, false);
1397         if (!chain) {
1398                 /* User requested flush on non-existent chain. Nothing to do,
1399                  * so just return success.
1400                  */
1401                 if (prio == 0) {
1402                         err = 0;
1403                         goto errout;
1404                 }
1405                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1406                 err = -ENOENT;
1407                 goto errout;
1408         }
1409
1410         if (prio == 0) {
1411                 tfilter_notify_chain(net, skb, block, q, parent, n,
1412                                      chain, RTM_DELTFILTER);
1413                 tcf_chain_flush(chain);
1414                 err = 0;
1415                 goto errout;
1416         }
1417
1418         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1419                                prio, false);
1420         if (!tp || IS_ERR(tp)) {
1421                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1422                 err = tp ? PTR_ERR(tp) : -ENOENT;
1423                 goto errout;
1424         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1425                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1426                 err = -EINVAL;
1427                 goto errout;
1428         }
1429
1430         fh = tp->ops->get(tp, t->tcm_handle);
1431
1432         if (!fh) {
1433                 if (t->tcm_handle == 0) {
1434                         tcf_chain_tp_remove(chain, &chain_info, tp);
1435                         tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1436                                        RTM_DELTFILTER, false);
1437                         tcf_proto_destroy(tp, extack);
1438                         err = 0;
1439                 } else {
1440                         NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1441                         err = -ENOENT;
1442                 }
1443         } else {
1444                 bool last;
1445
1446                 err = tfilter_del_notify(net, skb, n, tp, block,
1447                                          q, parent, fh, false, &last,
1448                                          extack);
1449                 if (err)
1450                         goto errout;
1451                 if (last) {
1452                         tcf_chain_tp_remove(chain, &chain_info, tp);
1453                         tcf_proto_destroy(tp, extack);
1454                 }
1455         }
1456
1457 errout:
1458         if (chain)
1459                 tcf_chain_put(chain);
1460         return err;
1461 }
1462
1463 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1464                           struct netlink_ext_ack *extack)
1465 {
1466         struct net *net = sock_net(skb->sk);
1467         struct nlattr *tca[TCA_MAX + 1];
1468         struct tcmsg *t;
1469         u32 protocol;
1470         u32 prio;
1471         u32 parent;
1472         u32 chain_index;
1473         struct Qdisc *q = NULL;
1474         struct tcf_chain_info chain_info;
1475         struct tcf_chain *chain = NULL;
1476         struct tcf_block *block;
1477         struct tcf_proto *tp = NULL;
1478         unsigned long cl = 0;
1479         void *fh = NULL;
1480         int err;
1481
1482         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1483         if (err < 0)
1484                 return err;
1485
1486         t = nlmsg_data(n);
1487         protocol = TC_H_MIN(t->tcm_info);
1488         prio = TC_H_MAJ(t->tcm_info);
1489         parent = t->tcm_parent;
1490
1491         if (prio == 0) {
1492                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1493                 return -ENOENT;
1494         }
1495
1496         /* Find head of filter chain. */
1497
1498         block = tcf_block_find(net, &q, &parent, &cl,
1499                                t->tcm_ifindex, t->tcm_block_index, extack);
1500         if (IS_ERR(block)) {
1501                 err = PTR_ERR(block);
1502                 goto errout;
1503         }
1504
1505         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1506         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1507                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1508                 err = -EINVAL;
1509                 goto errout;
1510         }
1511         chain = tcf_chain_get(block, chain_index, false);
1512         if (!chain) {
1513                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1514                 err = -EINVAL;
1515                 goto errout;
1516         }
1517
1518         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1519                                prio, false);
1520         if (!tp || IS_ERR(tp)) {
1521                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1522                 err = tp ? PTR_ERR(tp) : -ENOENT;
1523                 goto errout;
1524         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1525                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1526                 err = -EINVAL;
1527                 goto errout;
1528         }
1529
1530         fh = tp->ops->get(tp, t->tcm_handle);
1531
1532         if (!fh) {
1533                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1534                 err = -ENOENT;
1535         } else {
1536                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
1537                                      fh, RTM_NEWTFILTER, true);
1538                 if (err < 0)
1539                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
1540         }
1541
1542 errout:
1543         if (chain)
1544                 tcf_chain_put(chain);
1545         return err;
1546 }
1547
1548 struct tcf_dump_args {
1549         struct tcf_walker w;
1550         struct sk_buff *skb;
1551         struct netlink_callback *cb;
1552         struct tcf_block *block;
1553         struct Qdisc *q;
1554         u32 parent;
1555 };
1556
1557 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1558 {
1559         struct tcf_dump_args *a = (void *)arg;
1560         struct net *net = sock_net(a->skb->sk);
1561
1562         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
1563                              n, NETLINK_CB(a->cb->skb).portid,
1564                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1565                              RTM_NEWTFILTER);
1566 }
1567
1568 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
1569                            struct sk_buff *skb, struct netlink_callback *cb,
1570                            long index_start, long *p_index)
1571 {
1572         struct net *net = sock_net(skb->sk);
1573         struct tcf_block *block = chain->block;
1574         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1575         struct tcf_dump_args arg;
1576         struct tcf_proto *tp;
1577
1578         for (tp = rtnl_dereference(chain->filter_chain);
1579              tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
1580                 if (*p_index < index_start)
1581                         continue;
1582                 if (TC_H_MAJ(tcm->tcm_info) &&
1583                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
1584                         continue;
1585                 if (TC_H_MIN(tcm->tcm_info) &&
1586                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
1587                         continue;
1588                 if (*p_index > index_start)
1589                         memset(&cb->args[1], 0,
1590                                sizeof(cb->args) - sizeof(cb->args[0]));
1591                 if (cb->args[1] == 0) {
1592                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
1593                                           NETLINK_CB(cb->skb).portid,
1594                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
1595                                           RTM_NEWTFILTER) <= 0)
1596                                 return false;
1597
1598                         cb->args[1] = 1;
1599                 }
1600                 if (!tp->ops->walk)
1601                         continue;
1602                 arg.w.fn = tcf_node_dump;
1603                 arg.skb = skb;
1604                 arg.cb = cb;
1605                 arg.block = block;
1606                 arg.q = q;
1607                 arg.parent = parent;
1608                 arg.w.stop = 0;
1609                 arg.w.skip = cb->args[1] - 1;
1610                 arg.w.count = 0;
1611                 arg.w.cookie = cb->args[2];
1612                 tp->ops->walk(tp, &arg.w);
1613                 cb->args[2] = arg.w.cookie;
1614                 cb->args[1] = arg.w.count + 1;
1615                 if (arg.w.stop)
1616                         return false;
1617         }
1618         return true;
1619 }
1620
1621 /* called with RTNL */
1622 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
1623 {
1624         struct net *net = sock_net(skb->sk);
1625         struct nlattr *tca[TCA_MAX + 1];
1626         struct Qdisc *q = NULL;
1627         struct tcf_block *block;
1628         struct tcf_chain *chain;
1629         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1630         long index_start;
1631         long index;
1632         u32 parent;
1633         int err;
1634
1635         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1636                 return skb->len;
1637
1638         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1639         if (err)
1640                 return err;
1641
1642         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1643                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1644                 if (!block)
1645                         goto out;
1646                 /* If we work with block index, q is NULL and parent value
1647                  * will never be used in the following code. The check
1648                  * in tcf_fill_node prevents it. However, compiler does not
1649                  * see that far, so set parent to zero to silence the warning
1650                  * about parent being uninitialized.
1651                  */
1652                 parent = 0;
1653         } else {
1654                 const struct Qdisc_class_ops *cops;
1655                 struct net_device *dev;
1656                 unsigned long cl = 0;
1657
1658                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1659                 if (!dev)
1660                         return skb->len;
1661
1662                 parent = tcm->tcm_parent;
1663                 if (!parent) {
1664                         q = dev->qdisc;
1665                         parent = q->handle;
1666                 } else {
1667                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1668                 }
1669                 if (!q)
1670                         goto out;
1671                 cops = q->ops->cl_ops;
1672                 if (!cops)
1673                         goto out;
1674                 if (!cops->tcf_block)
1675                         goto out;
1676                 if (TC_H_MIN(tcm->tcm_parent)) {
1677                         cl = cops->find(q, tcm->tcm_parent);
1678                         if (cl == 0)
1679                                 goto out;
1680                 }
1681                 block = cops->tcf_block(q, cl, NULL);
1682                 if (!block)
1683                         goto out;
1684                 if (tcf_block_shared(block))
1685                         q = NULL;
1686         }
1687
1688         index_start = cb->args[0];
1689         index = 0;
1690
1691         list_for_each_entry(chain, &block->chain_list, list) {
1692                 if (tca[TCA_CHAIN] &&
1693                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1694                         continue;
1695                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
1696                                     index_start, &index)) {
1697                         err = -EMSGSIZE;
1698                         break;
1699                 }
1700         }
1701
1702         cb->args[0] = index;
1703
1704 out:
1705         /* If we did no progress, the error (EMSGSIZE) is real */
1706         if (skb->len == 0 && err)
1707                 return err;
1708         return skb->len;
1709 }
1710
1711 static int tc_chain_fill_node(struct tcf_chain *chain, struct net *net,
1712                               struct sk_buff *skb, struct tcf_block *block,
1713                               u32 portid, u32 seq, u16 flags, int event)
1714 {
1715         unsigned char *b = skb_tail_pointer(skb);
1716         const struct tcf_proto_ops *ops;
1717         struct nlmsghdr *nlh;
1718         struct tcmsg *tcm;
1719         void *priv;
1720
1721         ops = chain->tmplt_ops;
1722         priv = chain->tmplt_priv;
1723
1724         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1725         if (!nlh)
1726                 goto out_nlmsg_trim;
1727         tcm = nlmsg_data(nlh);
1728         tcm->tcm_family = AF_UNSPEC;
1729         tcm->tcm__pad1 = 0;
1730         tcm->tcm__pad2 = 0;
1731         tcm->tcm_handle = 0;
1732         if (block->q) {
1733                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
1734                 tcm->tcm_parent = block->q->handle;
1735         } else {
1736                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1737                 tcm->tcm_block_index = block->index;
1738         }
1739
1740         if (nla_put_u32(skb, TCA_CHAIN, chain->index))
1741                 goto nla_put_failure;
1742
1743         if (ops) {
1744                 if (nla_put_string(skb, TCA_KIND, ops->kind))
1745                         goto nla_put_failure;
1746                 if (ops->tmplt_dump(skb, net, priv) < 0)
1747                         goto nla_put_failure;
1748         }
1749
1750         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1751         return skb->len;
1752
1753 out_nlmsg_trim:
1754 nla_put_failure:
1755         nlmsg_trim(skb, b);
1756         return -EMSGSIZE;
1757 }
1758
1759 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
1760                            u32 seq, u16 flags, int event, bool unicast)
1761 {
1762         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1763         struct tcf_block *block = chain->block;
1764         struct net *net = block->net;
1765         struct sk_buff *skb;
1766
1767         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1768         if (!skb)
1769                 return -ENOBUFS;
1770
1771         if (tc_chain_fill_node(chain, net, skb, block, portid,
1772                                seq, flags, event) <= 0) {
1773                 kfree_skb(skb);
1774                 return -EINVAL;
1775         }
1776
1777         if (unicast)
1778                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1779
1780         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
1781 }
1782
1783 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
1784                               struct nlattr **tca,
1785                               struct netlink_ext_ack *extack)
1786 {
1787         const struct tcf_proto_ops *ops;
1788         void *tmplt_priv;
1789
1790         /* If kind is not set, user did not specify template. */
1791         if (!tca[TCA_KIND])
1792                 return 0;
1793
1794         ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), extack);
1795         if (IS_ERR(ops))
1796                 return PTR_ERR(ops);
1797         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
1798                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
1799                 return -EOPNOTSUPP;
1800         }
1801
1802         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
1803         if (IS_ERR(tmplt_priv)) {
1804                 module_put(ops->owner);
1805                 return PTR_ERR(tmplt_priv);
1806         }
1807         chain->tmplt_ops = ops;
1808         chain->tmplt_priv = tmplt_priv;
1809         return 0;
1810 }
1811
1812 static void tc_chain_tmplt_del(struct tcf_chain *chain)
1813 {
1814         const struct tcf_proto_ops *ops = chain->tmplt_ops;
1815
1816         /* If template ops are set, no work to do for us. */
1817         if (!ops)
1818                 return;
1819
1820         ops->tmplt_destroy(chain->tmplt_priv);
1821         module_put(ops->owner);
1822 }
1823
1824 /* Add/delete/get a chain */
1825
1826 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
1827                         struct netlink_ext_ack *extack)
1828 {
1829         struct net *net = sock_net(skb->sk);
1830         struct nlattr *tca[TCA_MAX + 1];
1831         struct tcmsg *t;
1832         u32 parent;
1833         u32 chain_index;
1834         struct Qdisc *q = NULL;
1835         struct tcf_chain *chain = NULL;
1836         struct tcf_block *block;
1837         unsigned long cl;
1838         int err;
1839
1840         if (n->nlmsg_type != RTM_GETCHAIN &&
1841             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1842                 return -EPERM;
1843
1844 replay:
1845         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1846         if (err < 0)
1847                 return err;
1848
1849         t = nlmsg_data(n);
1850         parent = t->tcm_parent;
1851         cl = 0;
1852
1853         block = tcf_block_find(net, &q, &parent, &cl,
1854                                t->tcm_ifindex, t->tcm_block_index, extack);
1855         if (IS_ERR(block))
1856                 return PTR_ERR(block);
1857
1858         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1859         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1860                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1861                 return -EINVAL;
1862         }
1863         chain = tcf_chain_lookup(block, chain_index);
1864         if (n->nlmsg_type == RTM_NEWCHAIN) {
1865                 if (chain) {
1866                         if (tcf_chain_held_by_acts_only(chain)) {
1867                                 /* The chain exists only because there is
1868                                  * some action referencing it.
1869                                  */
1870                                 tcf_chain_hold(chain);
1871                         } else {
1872                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
1873                                 return -EEXIST;
1874                         }
1875                 } else {
1876                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1877                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
1878                                 return -ENOENT;
1879                         }
1880                         chain = tcf_chain_create(block, chain_index);
1881                         if (!chain) {
1882                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
1883                                 return -ENOMEM;
1884                         }
1885                 }
1886         } else {
1887                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
1888                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1889                         return -EINVAL;
1890                 }
1891                 tcf_chain_hold(chain);
1892         }
1893
1894         switch (n->nlmsg_type) {
1895         case RTM_NEWCHAIN:
1896                 err = tc_chain_tmplt_add(chain, net, tca, extack);
1897                 if (err)
1898                         goto errout;
1899                 /* In case the chain was successfully added, take a reference
1900                  * to the chain. This ensures that an empty chain
1901                  * does not disappear at the end of this function.
1902                  */
1903                 tcf_chain_hold(chain);
1904                 chain->explicitly_created = true;
1905                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
1906                                 RTM_NEWCHAIN, false);
1907                 break;
1908         case RTM_DELCHAIN:
1909                 tfilter_notify_chain(net, skb, block, q, parent, n,
1910                                      chain, RTM_DELTFILTER);
1911                 /* Flush the chain first as the user requested chain removal. */
1912                 tcf_chain_flush(chain);
1913                 /* In case the chain was successfully deleted, put a reference
1914                  * to the chain previously taken during addition.
1915                  */
1916                 tcf_chain_put_explicitly_created(chain);
1917                 chain->explicitly_created = false;
1918                 break;
1919         case RTM_GETCHAIN:
1920                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
1921                                       n->nlmsg_flags, n->nlmsg_type, true);
1922                 if (err < 0)
1923                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
1924                 break;
1925         default:
1926                 err = -EOPNOTSUPP;
1927                 NL_SET_ERR_MSG(extack, "Unsupported message type");
1928                 goto errout;
1929         }
1930
1931 errout:
1932         tcf_chain_put(chain);
1933         if (err == -EAGAIN)
1934                 /* Replay the request. */
1935                 goto replay;
1936         return err;
1937 }
1938
1939 /* called with RTNL */
1940 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
1941 {
1942         struct net *net = sock_net(skb->sk);
1943         struct nlattr *tca[TCA_MAX + 1];
1944         struct Qdisc *q = NULL;
1945         struct tcf_block *block;
1946         struct tcf_chain *chain;
1947         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1948         long index_start;
1949         long index;
1950         u32 parent;
1951         int err;
1952
1953         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1954                 return skb->len;
1955
1956         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
1957                           NULL);
1958         if (err)
1959                 return err;
1960
1961         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1962                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1963                 if (!block)
1964                         goto out;
1965                 /* If we work with block index, q is NULL and parent value
1966                  * will never be used in the following code. The check
1967                  * in tcf_fill_node prevents it. However, compiler does not
1968                  * see that far, so set parent to zero to silence the warning
1969                  * about parent being uninitialized.
1970                  */
1971                 parent = 0;
1972         } else {
1973                 const struct Qdisc_class_ops *cops;
1974                 struct net_device *dev;
1975                 unsigned long cl = 0;
1976
1977                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1978                 if (!dev)
1979                         return skb->len;
1980
1981                 parent = tcm->tcm_parent;
1982                 if (!parent) {
1983                         q = dev->qdisc;
1984                         parent = q->handle;
1985                 } else {
1986                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1987                 }
1988                 if (!q)
1989                         goto out;
1990                 cops = q->ops->cl_ops;
1991                 if (!cops)
1992                         goto out;
1993                 if (!cops->tcf_block)
1994                         goto out;
1995                 if (TC_H_MIN(tcm->tcm_parent)) {
1996                         cl = cops->find(q, tcm->tcm_parent);
1997                         if (cl == 0)
1998                                 goto out;
1999                 }
2000                 block = cops->tcf_block(q, cl, NULL);
2001                 if (!block)
2002                         goto out;
2003                 if (tcf_block_shared(block))
2004                         q = NULL;
2005         }
2006
2007         index_start = cb->args[0];
2008         index = 0;
2009
2010         list_for_each_entry(chain, &block->chain_list, list) {
2011                 if ((tca[TCA_CHAIN] &&
2012                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2013                         continue;
2014                 if (index < index_start) {
2015                         index++;
2016                         continue;
2017                 }
2018                 if (tcf_chain_held_by_acts_only(chain))
2019                         continue;
2020                 err = tc_chain_fill_node(chain, net, skb, block,
2021                                          NETLINK_CB(cb->skb).portid,
2022                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
2023                                          RTM_NEWCHAIN);
2024                 if (err <= 0)
2025                         break;
2026                 index++;
2027         }
2028
2029         cb->args[0] = index;
2030
2031 out:
2032         /* If we did no progress, the error (EMSGSIZE) is real */
2033         if (skb->len == 0 && err)
2034                 return err;
2035         return skb->len;
2036 }
2037
2038 void tcf_exts_destroy(struct tcf_exts *exts)
2039 {
2040 #ifdef CONFIG_NET_CLS_ACT
2041         if (exts->actions) {
2042                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
2043                 kfree(exts->actions);
2044         }
2045         exts->nr_actions = 0;
2046 #endif
2047 }
2048 EXPORT_SYMBOL(tcf_exts_destroy);
2049
2050 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
2051                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
2052                       struct netlink_ext_ack *extack)
2053 {
2054 #ifdef CONFIG_NET_CLS_ACT
2055         {
2056                 struct tc_action *act;
2057                 size_t attr_size = 0;
2058
2059                 if (exts->police && tb[exts->police]) {
2060                         act = tcf_action_init_1(net, tp, tb[exts->police],
2061                                                 rate_tlv, "police", ovr,
2062                                                 TCA_ACT_BIND, true, extack);
2063                         if (IS_ERR(act))
2064                                 return PTR_ERR(act);
2065
2066                         act->type = exts->type = TCA_OLD_COMPAT;
2067                         exts->actions[0] = act;
2068                         exts->nr_actions = 1;
2069                 } else if (exts->action && tb[exts->action]) {
2070                         int err;
2071
2072                         err = tcf_action_init(net, tp, tb[exts->action],
2073                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
2074                                               exts->actions, &attr_size, true,
2075                                               extack);
2076                         if (err < 0)
2077                                 return err;
2078                         exts->nr_actions = err;
2079                 }
2080                 exts->net = net;
2081         }
2082 #else
2083         if ((exts->action && tb[exts->action]) ||
2084             (exts->police && tb[exts->police])) {
2085                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
2086                 return -EOPNOTSUPP;
2087         }
2088 #endif
2089
2090         return 0;
2091 }
2092 EXPORT_SYMBOL(tcf_exts_validate);
2093
2094 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
2095 {
2096 #ifdef CONFIG_NET_CLS_ACT
2097         struct tcf_exts old = *dst;
2098
2099         *dst = *src;
2100         tcf_exts_destroy(&old);
2101 #endif
2102 }
2103 EXPORT_SYMBOL(tcf_exts_change);
2104
2105 #ifdef CONFIG_NET_CLS_ACT
2106 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
2107 {
2108         if (exts->nr_actions == 0)
2109                 return NULL;
2110         else
2111                 return exts->actions[0];
2112 }
2113 #endif
2114
2115 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
2116 {
2117 #ifdef CONFIG_NET_CLS_ACT
2118         struct nlattr *nest;
2119
2120         if (exts->action && tcf_exts_has_actions(exts)) {
2121                 /*
2122                  * again for backward compatible mode - we want
2123                  * to work with both old and new modes of entering
2124                  * tc data even if iproute2  was newer - jhs
2125                  */
2126                 if (exts->type != TCA_OLD_COMPAT) {
2127                         nest = nla_nest_start(skb, exts->action);
2128                         if (nest == NULL)
2129                                 goto nla_put_failure;
2130
2131                         if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
2132                                 goto nla_put_failure;
2133                         nla_nest_end(skb, nest);
2134                 } else if (exts->police) {
2135                         struct tc_action *act = tcf_exts_first_act(exts);
2136                         nest = nla_nest_start(skb, exts->police);
2137                         if (nest == NULL || !act)
2138                                 goto nla_put_failure;
2139                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
2140                                 goto nla_put_failure;
2141                         nla_nest_end(skb, nest);
2142                 }
2143         }
2144         return 0;
2145
2146 nla_put_failure:
2147         nla_nest_cancel(skb, nest);
2148         return -1;
2149 #else
2150         return 0;
2151 #endif
2152 }
2153 EXPORT_SYMBOL(tcf_exts_dump);
2154
2155
2156 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
2157 {
2158 #ifdef CONFIG_NET_CLS_ACT
2159         struct tc_action *a = tcf_exts_first_act(exts);
2160         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
2161                 return -1;
2162 #endif
2163         return 0;
2164 }
2165 EXPORT_SYMBOL(tcf_exts_dump_stats);
2166
2167 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
2168                                        enum tc_setup_type type,
2169                                        void *type_data, bool err_stop)
2170 {
2171         int ok_count = 0;
2172 #ifdef CONFIG_NET_CLS_ACT
2173         const struct tc_action *a;
2174         struct net_device *dev;
2175         int i, ret;
2176
2177         if (!tcf_exts_has_actions(exts))
2178                 return 0;
2179
2180         for (i = 0; i < exts->nr_actions; i++) {
2181                 a = exts->actions[i];
2182                 if (!a->ops->get_dev)
2183                         continue;
2184                 dev = a->ops->get_dev(a);
2185                 if (!dev)
2186                         continue;
2187                 ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
2188                 a->ops->put_dev(dev);
2189                 if (ret < 0)
2190                         return ret;
2191                 ok_count += ret;
2192         }
2193 #endif
2194         return ok_count;
2195 }
2196
2197 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
2198                      enum tc_setup_type type, void *type_data, bool err_stop)
2199 {
2200         int ok_count;
2201         int ret;
2202
2203         ret = tcf_block_cb_call(block, type, type_data, err_stop);
2204         if (ret < 0)
2205                 return ret;
2206         ok_count = ret;
2207
2208         if (!exts || ok_count)
2209                 return ok_count;
2210         ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
2211         if (ret < 0)
2212                 return ret;
2213         ok_count += ret;
2214
2215         return ok_count;
2216 }
2217 EXPORT_SYMBOL(tc_setup_cb_call);
2218
2219 static __net_init int tcf_net_init(struct net *net)
2220 {
2221         struct tcf_net *tn = net_generic(net, tcf_net_id);
2222
2223         idr_init(&tn->idr);
2224         return 0;
2225 }
2226
2227 static void __net_exit tcf_net_exit(struct net *net)
2228 {
2229         struct tcf_net *tn = net_generic(net, tcf_net_id);
2230
2231         idr_destroy(&tn->idr);
2232 }
2233
2234 static struct pernet_operations tcf_net_ops = {
2235         .init = tcf_net_init,
2236         .exit = tcf_net_exit,
2237         .id   = &tcf_net_id,
2238         .size = sizeof(struct tcf_net),
2239 };
2240
2241 static int __init tc_filter_init(void)
2242 {
2243         int err;
2244
2245         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
2246         if (!tc_filter_wq)
2247                 return -ENOMEM;
2248
2249         err = register_pernet_subsys(&tcf_net_ops);
2250         if (err)
2251                 goto err_register_pernet_subsys;
2252
2253         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 0);
2254         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 0);
2255         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
2256                       tc_dump_tfilter, 0);
2257         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
2258         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
2259         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
2260                       tc_dump_chain, 0);
2261
2262         return 0;
2263
2264 err_register_pernet_subsys:
2265         destroy_workqueue(tc_filter_wq);
2266         return err;
2267 }
2268
2269 subsys_initcall(tc_filter_init);