net/sched/cls_api.c

   1 /*
   2  * net/sched/cls_api.c  Packet classifier API.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *
  11  * Changes:
  12  *
  13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  14  *
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/init.h>
  25 #include <linux/kmod.h>
  26 #include <linux/slab.h>
  27 #include <linux/idr.h>
  28 #include <net/net_namespace.h>
  29 #include <net/sock.h>
  30 #include <net/netlink.h>
  31 #include <net/pkt_sched.h>
  32 #include <net/pkt_cls.h>
  33
  34 /* The list of all installed classifier types */
  35 static LIST_HEAD(tcf_proto_base);
  36
  37 /* Protects list of registered TC modules. It is pure SMP lock. */
  38 static DEFINE_RWLOCK(cls_mod_lock);
  39
  40 /* Find classifier type by string name */
  41
  42 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
  43 {
  44         const struct tcf_proto_ops *t, *res = NULL;
  45
  46         if (kind) {
  47                 read_lock(&cls_mod_lock);
  48                 list_for_each_entry(t, &tcf_proto_base, head) {
  49                         if (strcmp(kind, t->kind) == 0) {
  50                                 if (try_module_get(t->owner))
  51                                         res = t;
  52                                 break;
  53                         }
  54                 }
  55                 read_unlock(&cls_mod_lock);
  56         }
  57         return res;
  58 }
  59
  60 static const struct tcf_proto_ops *
  61 tcf_proto_lookup_ops(const char *kind, struct netlink_ext_ack *extack)
  62 {
  63         const struct tcf_proto_ops *ops;
  64
  65         ops = __tcf_proto_lookup_ops(kind);
  66         if (ops)
  67                 return ops;
  68 #ifdef CONFIG_MODULES
  69         rtnl_unlock();
  70         request_module("cls_%s", kind);
  71         rtnl_lock();
  72         ops = __tcf_proto_lookup_ops(kind);
  73         /* We dropped the RTNL semaphore in order to perform
  74          * the module load. So, even if we succeeded in loading
  75          * the module we have to replay the request. We indicate
  76          * this using -EAGAIN.
  77          */
  78         if (ops) {
  79                 module_put(ops->owner);
  80                 return ERR_PTR(-EAGAIN);
  81         }
  82 #endif
  83         NL_SET_ERR_MSG(extack, "TC classifier not found");
  84         return ERR_PTR(-ENOENT);
  85 }
  86
  87 /* Register(unregister) new classifier type */
  88
  89 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  90 {
  91         struct tcf_proto_ops *t;
  92         int rc = -EEXIST;
  93
  94         write_lock(&cls_mod_lock);
  95         list_for_each_entry(t, &tcf_proto_base, head)
  96                 if (!strcmp(ops->kind, t->kind))
  97                         goto out;
  98
  99         list_add_tail(&ops->head, &tcf_proto_base);
 100         rc = 0;
 101 out:
 102         write_unlock(&cls_mod_lock);
 103         return rc;
 104 }
 105 EXPORT_SYMBOL(register_tcf_proto_ops);
 106
 107 static struct workqueue_struct *tc_filter_wq;
 108
 109 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 110 {
 111         struct tcf_proto_ops *t;
 112         int rc = -ENOENT;
 113
 114         /* Wait for outstanding call_rcu()s, if any, from a
 115          * tcf_proto_ops's destroy() handler.
 116          */
 117         rcu_barrier();
 118         flush_workqueue(tc_filter_wq);
 119
 120         write_lock(&cls_mod_lock);
 121         list_for_each_entry(t, &tcf_proto_base, head) {
 122                 if (t == ops) {
 123                         list_del(&t->head);
 124                         rc = 0;
 125                         break;
 126                 }
 127         }
 128         write_unlock(&cls_mod_lock);
 129         return rc;
 130 }
 131 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 132
 133 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 134 {
 135         INIT_RCU_WORK(rwork, func);
 136         return queue_rcu_work(tc_filter_wq, rwork);
 137 }
 138 EXPORT_SYMBOL(tcf_queue_work);
 139
 140 /* Select new prio value from the range, managed by kernel. */
 141
 142 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 143 {
 144         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 145
 146         if (tp)
 147                 first = tp->prio - 1;
 148
 149         return TC_H_MAJ(first);
 150 }
 151
 152 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 153                                           u32 prio, struct tcf_chain *chain,
 154                                           struct netlink_ext_ack *extack)
 155 {
 156         struct tcf_proto *tp;
 157         int err;
 158
 159         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 160         if (!tp)
 161                 return ERR_PTR(-ENOBUFS);
 162
 163         tp->ops = tcf_proto_lookup_ops(kind, extack);
 164         if (IS_ERR(tp->ops)) {
 165                 err = PTR_ERR(tp->ops);
 166                 goto errout;
 167         }
 168         tp->classify = tp->ops->classify;
 169         tp->protocol = protocol;
 170         tp->prio = prio;
 171         tp->chain = chain;
 172
 173         err = tp->ops->init(tp);
 174         if (err) {
 175                 module_put(tp->ops->owner);
 176                 goto errout;
 177         }
 178         return tp;
 179
 180 errout:
 181         kfree(tp);
 182         return ERR_PTR(err);
 183 }
 184
 185 static void tcf_proto_destroy(struct tcf_proto *tp,
 186                               struct netlink_ext_ack *extack)
 187 {
 188         tp->ops->destroy(tp, extack);
 189         module_put(tp->ops->owner);
 190         kfree_rcu(tp, rcu);
 191 }
 192
 193 struct tcf_filter_chain_list_item {
 194         struct list_head list;
 195         tcf_chain_head_change_t *chain_head_change;
 196         void *chain_head_change_priv;
 197 };
 198
 199 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 200                                           u32 chain_index)
 201 {
 202         struct tcf_chain *chain;
 203
 204         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 205         if (!chain)
 206                 return NULL;
 207         list_add_tail(&chain->list, &block->chain_list);
 208         chain->block = block;
 209         chain->index = chain_index;
 210         chain->refcnt = 1;
 211         if (!chain->index)
 212                 block->chain0.chain = chain;
 213         return chain;
 214 }
 215
 216 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 217                                        struct tcf_proto *tp_head)
 218 {
 219         if (item->chain_head_change)
 220                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 221 }
 222
 223 static void tcf_chain0_head_change(struct tcf_chain *chain,
 224                                    struct tcf_proto *tp_head)
 225 {
 226         struct tcf_filter_chain_list_item *item;
 227         struct tcf_block *block = chain->block;
 228
 229         if (chain->index)
 230                 return;
 231         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 232                 tcf_chain_head_change_item(item, tp_head);
 233 }
 234
 235 static void tcf_chain_destroy(struct tcf_chain *chain)
 236 {
 237         struct tcf_block *block = chain->block;
 238
 239         list_del(&chain->list);
 240         if (!chain->index)
 241                 block->chain0.chain = NULL;
 242         kfree(chain);
 243         if (list_empty(&block->chain_list) && block->refcnt == 0)
 244                 kfree(block);
 245 }
 246
 247 static void tcf_chain_hold(struct tcf_chain *chain)
 248 {
 249         ++chain->refcnt;
 250 }
 251
 252 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 253 {
 254         /* In case all the references are action references, this
 255          * chain should not be shown to the user.
 256          */
 257         return chain->refcnt == chain->action_refcnt;
 258 }
 259
 260 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 261                                           u32 chain_index)
 262 {
 263         struct tcf_chain *chain;
 264
 265         list_for_each_entry(chain, &block->chain_list, list) {
 266                 if (chain->index == chain_index)
 267                         return chain;
 268         }
 269         return NULL;
 270 }
 271
 272 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 273                            u32 seq, u16 flags, int event, bool unicast);
 274
 275 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 276                                          u32 chain_index, bool create,
 277                                          bool by_act)
 278 {
 279         struct tcf_chain *chain = tcf_chain_lookup(block, chain_index);
 280
 281         if (chain) {
 282                 tcf_chain_hold(chain);
 283         } else {
 284                 if (!create)
 285                         return NULL;
 286                 chain = tcf_chain_create(block, chain_index);
 287                 if (!chain)
 288                         return NULL;
 289         }
 290
 291         if (by_act)
 292                 ++chain->action_refcnt;
 293
 294         /* Send notification only in case we got the first
 295          * non-action reference. Until then, the chain acts only as
 296          * a placeholder for actions pointing to it and user ought
 297          * not know about them.
 298          */
 299         if (chain->refcnt - chain->action_refcnt == 1 && !by_act)
 300                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 301                                 RTM_NEWCHAIN, false);
 302
 303         return chain;
 304 }
 305
 306 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 307                                        bool create)
 308 {
 309         return __tcf_chain_get(block, chain_index, create, false);
 310 }
 311
 312 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 313 {
 314         return __tcf_chain_get(block, chain_index, true, true);
 315 }
 316 EXPORT_SYMBOL(tcf_chain_get_by_act);
 317
 318 static void tc_chain_tmplt_del(struct tcf_chain *chain);
 319
 320 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act)
 321 {
 322         if (by_act)
 323                 chain->action_refcnt--;
 324         chain->refcnt--;
 325
 326         /* The last dropped non-action reference will trigger notification. */
 327         if (chain->refcnt - chain->action_refcnt == 0 && !by_act)
 328                 tc_chain_notify(chain, NULL, 0, 0, RTM_DELCHAIN, false);
 329
 330         if (chain->refcnt == 0) {
 331                 tc_chain_tmplt_del(chain);
 332                 tcf_chain_destroy(chain);
 333         }
 334 }
 335
 336 static void tcf_chain_put(struct tcf_chain *chain)
 337 {
 338         __tcf_chain_put(chain, false);
 339 }
 340
 341 void tcf_chain_put_by_act(struct tcf_chain *chain)
 342 {
 343         __tcf_chain_put(chain, true);
 344 }
 345 EXPORT_SYMBOL(tcf_chain_put_by_act);
 346
 347 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 348 {
 349         if (chain->explicitly_created)
 350                 tcf_chain_put(chain);
 351 }
 352
 353 static void tcf_chain_flush(struct tcf_chain *chain)
 354 {
 355         struct tcf_proto *tp = rtnl_dereference(chain->filter_chain);
 356
 357         tcf_chain0_head_change(chain, NULL);
 358         while (tp) {
 359                 RCU_INIT_POINTER(chain->filter_chain, tp->next);
 360                 tcf_proto_destroy(tp, NULL);
 361                 tp = rtnl_dereference(chain->filter_chain);
 362                 tcf_chain_put(chain);
 363         }
 364 }
 365
 366 static bool tcf_block_offload_in_use(struct tcf_block *block)
 367 {
 368         return block->offloadcnt;
 369 }
 370
 371 static int tcf_block_offload_cmd(struct tcf_block *block,
 372                                  struct net_device *dev,
 373                                  struct tcf_block_ext_info *ei,
 374                                  enum tc_block_command command,
 375                                  struct netlink_ext_ack *extack)
 376 {
 377         struct tc_block_offload bo = {};
 378
 379         bo.command = command;
 380         bo.binder_type = ei->binder_type;
 381         bo.block = block;
 382         bo.extack = extack;
 383         return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 384 }
 385
 386 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 387                                   struct tcf_block_ext_info *ei,
 388                                   struct netlink_ext_ack *extack)
 389 {
 390         struct net_device *dev = q->dev_queue->dev;
 391         int err;
 392
 393         if (!dev->netdev_ops->ndo_setup_tc)
 394                 goto no_offload_dev_inc;
 395
 396         /* If tc offload feature is disabled and the block we try to bind
 397          * to already has some offloaded filters, forbid to bind.
 398          */
 399         if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
 400                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 401                 return -EOPNOTSUPP;
 402         }
 403
 404         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_BIND, extack);
 405         if (err == -EOPNOTSUPP)
 406                 goto no_offload_dev_inc;
 407         return err;
 408
 409 no_offload_dev_inc:
 410         if (tcf_block_offload_in_use(block))
 411                 return -EOPNOTSUPP;
 412         block->nooffloaddevcnt++;
 413         return 0;
 414 }
 415
 416 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 417                                      struct tcf_block_ext_info *ei)
 418 {
 419         struct net_device *dev = q->dev_queue->dev;
 420         int err;
 421
 422         if (!dev->netdev_ops->ndo_setup_tc)
 423                 goto no_offload_dev_dec;
 424         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_UNBIND, NULL);
 425         if (err == -EOPNOTSUPP)
 426                 goto no_offload_dev_dec;
 427         return;
 428
 429 no_offload_dev_dec:
 430         WARN_ON(block->nooffloaddevcnt-- == 0);
 431 }
 432
 433 static int
 434 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 435                               struct tcf_block_ext_info *ei,
 436                               struct netlink_ext_ack *extack)
 437 {
 438         struct tcf_chain *chain0 = block->chain0.chain;
 439         struct tcf_filter_chain_list_item *item;
 440
 441         item = kmalloc(sizeof(*item), GFP_KERNEL);
 442         if (!item) {
 443                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 444                 return -ENOMEM;
 445         }
 446         item->chain_head_change = ei->chain_head_change;
 447         item->chain_head_change_priv = ei->chain_head_change_priv;
 448         if (chain0 && chain0->filter_chain)
 449                 tcf_chain_head_change_item(item, chain0->filter_chain);
 450         list_add(&item->list, &block->chain0.filter_chain_list);
 451         return 0;
 452 }
 453
 454 static void
 455 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 456                               struct tcf_block_ext_info *ei)
 457 {
 458         struct tcf_chain *chain0 = block->chain0.chain;
 459         struct tcf_filter_chain_list_item *item;
 460
 461         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 462                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 463                     (item->chain_head_change == ei->chain_head_change &&
 464                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 465                         if (chain0)
 466                                 tcf_chain_head_change_item(item, NULL);
 467                         list_del(&item->list);
 468                         kfree(item);
 469                         return;
 470                 }
 471         }
 472         WARN_ON(1);
 473 }
 474
 475 struct tcf_net {
 476         struct idr idr;
 477 };
 478
 479 static unsigned int tcf_net_id;
 480
 481 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 482                             struct netlink_ext_ack *extack)
 483 {
 484         struct tcf_net *tn = net_generic(net, tcf_net_id);
 485
 486         return idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 487                              GFP_KERNEL);
 488 }
 489
 490 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 491 {
 492         struct tcf_net *tn = net_generic(net, tcf_net_id);
 493
 494         idr_remove(&tn->idr, block->index);
 495 }
 496
 497 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 498                                           u32 block_index,
 499                                           struct netlink_ext_ack *extack)
 500 {
 501         struct tcf_block *block;
 502
 503         block = kzalloc(sizeof(*block), GFP_KERNEL);
 504         if (!block) {
 505                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 506                 return ERR_PTR(-ENOMEM);
 507         }
 508         INIT_LIST_HEAD(&block->chain_list);
 509         INIT_LIST_HEAD(&block->cb_list);
 510         INIT_LIST_HEAD(&block->owner_list);
 511         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 512
 513         block->refcnt = 1;
 514         block->net = net;
 515         block->index = block_index;
 516
 517         /* Don't store q pointer for blocks which are shared */
 518         if (!tcf_block_shared(block))
 519                 block->q = q;
 520         return block;
 521 }
 522
 523 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 524 {
 525         struct tcf_net *tn = net_generic(net, tcf_net_id);
 526
 527         return idr_find(&tn->idr, block_index);
 528 }
 529
 530 /* Find tcf block.
 531  * Set q, parent, cl when appropriate.
 532  */
 533
 534 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
 535                                         u32 *parent, unsigned long *cl,
 536                                         int ifindex, u32 block_index,
 537                                         struct netlink_ext_ack *extack)
 538 {
 539         struct tcf_block *block;
 540
 541         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
 542                 block = tcf_block_lookup(net, block_index);
 543                 if (!block) {
 544                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
 545                         return ERR_PTR(-EINVAL);
 546                 }
 547         } else {
 548                 const struct Qdisc_class_ops *cops;
 549                 struct net_device *dev;
 550
 551                 /* Find link */
 552                 dev = __dev_get_by_index(net, ifindex);
 553                 if (!dev)
 554                         return ERR_PTR(-ENODEV);
 555
 556                 /* Find qdisc */
 557                 if (!*parent) {
 558                         *q = dev->qdisc;
 559                         *parent = (*q)->handle;
 560                 } else {
 561                         *q = qdisc_lookup(dev, TC_H_MAJ(*parent));
 562                         if (!*q) {
 563                                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
 564                                 return ERR_PTR(-EINVAL);
 565                         }
 566                 }
 567
 568                 /* Is it classful? */
 569                 cops = (*q)->ops->cl_ops;
 570                 if (!cops) {
 571                         NL_SET_ERR_MSG(extack, "Qdisc not classful");
 572                         return ERR_PTR(-EINVAL);
 573                 }
 574
 575                 if (!cops->tcf_block) {
 576                         NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
 577                         return ERR_PTR(-EOPNOTSUPP);
 578                 }
 579
 580                 /* Do we search for filter, attached to class? */
 581                 if (TC_H_MIN(*parent)) {
 582                         *cl = cops->find(*q, *parent);
 583                         if (*cl == 0) {
 584                                 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
 585                                 return ERR_PTR(-ENOENT);
 586                         }
 587                 }
 588
 589                 /* And the last stroke */
 590                 block = cops->tcf_block(*q, *cl, extack);
 591                 if (!block)
 592                         return ERR_PTR(-EINVAL);
 593                 if (tcf_block_shared(block)) {
 594                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
 595                         return ERR_PTR(-EOPNOTSUPP);
 596                 }
 597         }
 598
 599         return block;
 600 }
 601
 602 struct tcf_block_owner_item {
 603         struct list_head list;
 604         struct Qdisc *q;
 605         enum tcf_block_binder_type binder_type;
 606 };
 607
 608 static void
 609 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
 610                                struct Qdisc *q,
 611                                enum tcf_block_binder_type binder_type)
 612 {
 613         if (block->keep_dst &&
 614             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
 615             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
 616                 netif_keep_dst(qdisc_dev(q));
 617 }
 618
 619 void tcf_block_netif_keep_dst(struct tcf_block *block)
 620 {
 621         struct tcf_block_owner_item *item;
 622
 623         block->keep_dst = true;
 624         list_for_each_entry(item, &block->owner_list, list)
 625                 tcf_block_owner_netif_keep_dst(block, item->q,
 626                                                item->binder_type);
 627 }
 628 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
 629
 630 static int tcf_block_owner_add(struct tcf_block *block,
 631                                struct Qdisc *q,
 632                                enum tcf_block_binder_type binder_type)
 633 {
 634         struct tcf_block_owner_item *item;
 635
 636         item = kmalloc(sizeof(*item), GFP_KERNEL);
 637         if (!item)
 638                 return -ENOMEM;
 639         item->q = q;
 640         item->binder_type = binder_type;
 641         list_add(&item->list, &block->owner_list);
 642         return 0;
 643 }
 644
 645 static void tcf_block_owner_del(struct tcf_block *block,
 646                                 struct Qdisc *q,
 647                                 enum tcf_block_binder_type binder_type)
 648 {
 649         struct tcf_block_owner_item *item;
 650
 651         list_for_each_entry(item, &block->owner_list, list) {
 652                 if (item->q == q && item->binder_type == binder_type) {
 653                         list_del(&item->list);
 654                         kfree(item);
 655                         return;
 656                 }
 657         }
 658         WARN_ON(1);
 659 }
 660
 661 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 662                       struct tcf_block_ext_info *ei,
 663                       struct netlink_ext_ack *extack)
 664 {
 665         struct net *net = qdisc_net(q);
 666         struct tcf_block *block = NULL;
 667         bool created = false;
 668         int err;
 669
 670         if (ei->block_index) {
 671                 /* block_index not 0 means the shared block is requested */
 672                 block = tcf_block_lookup(net, ei->block_index);
 673                 if (block)
 674                         block->refcnt++;
 675         }
 676
 677         if (!block) {
 678                 block = tcf_block_create(net, q, ei->block_index, extack);
 679                 if (IS_ERR(block))
 680                         return PTR_ERR(block);
 681                 created = true;
 682                 if (tcf_block_shared(block)) {
 683                         err = tcf_block_insert(block, net, extack);
 684                         if (err)
 685                                 goto err_block_insert;
 686                 }
 687         }
 688
 689         err = tcf_block_owner_add(block, q, ei->binder_type);
 690         if (err)
 691                 goto err_block_owner_add;
 692
 693         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
 694
 695         err = tcf_chain0_head_change_cb_add(block, ei, extack);
 696         if (err)
 697                 goto err_chain0_head_change_cb_add;
 698
 699         err = tcf_block_offload_bind(block, q, ei, extack);
 700         if (err)
 701                 goto err_block_offload_bind;
 702
 703         *p_block = block;
 704         return 0;
 705
 706 err_block_offload_bind:
 707         tcf_chain0_head_change_cb_del(block, ei);
 708 err_chain0_head_change_cb_add:
 709         tcf_block_owner_del(block, q, ei->binder_type);
 710 err_block_owner_add:
 711         if (created) {
 712                 if (tcf_block_shared(block))
 713                         tcf_block_remove(block, net);
 714 err_block_insert:
 715                 kfree(block);
 716         } else {
 717                 block->refcnt--;
 718         }
 719         return err;
 720 }
 721 EXPORT_SYMBOL(tcf_block_get_ext);
 722
 723 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
 724 {
 725         struct tcf_proto __rcu **p_filter_chain = priv;
 726
 727         rcu_assign_pointer(*p_filter_chain, tp_head);
 728 }
 729
 730 int tcf_block_get(struct tcf_block **p_block,
 731                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
 732                   struct netlink_ext_ack *extack)
 733 {
 734         struct tcf_block_ext_info ei = {
 735                 .chain_head_change = tcf_chain_head_change_dflt,
 736                 .chain_head_change_priv = p_filter_chain,
 737         };
 738
 739         WARN_ON(!p_filter_chain);
 740         return tcf_block_get_ext(p_block, q, &ei, extack);
 741 }
 742 EXPORT_SYMBOL(tcf_block_get);
 743
 744 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
 745  * actions should be all removed after flushing.
 746  */
 747 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 748                        struct tcf_block_ext_info *ei)
 749 {
 750         struct tcf_chain *chain, *tmp;
 751
 752         if (!block)
 753                 return;
 754         tcf_chain0_head_change_cb_del(block, ei);
 755         tcf_block_owner_del(block, q, ei->binder_type);
 756
 757         if (block->refcnt == 1) {
 758                 if (tcf_block_shared(block))
 759                         tcf_block_remove(block, block->net);
 760
 761                 /* Hold a refcnt for all chains, so that they don't disappear
 762                  * while we are iterating.
 763                  */
 764                 list_for_each_entry(chain, &block->chain_list, list)
 765                         tcf_chain_hold(chain);
 766
 767                 list_for_each_entry(chain, &block->chain_list, list)
 768                         tcf_chain_flush(chain);
 769         }
 770
 771         tcf_block_offload_unbind(block, q, ei);
 772
 773         if (block->refcnt == 1) {
 774                 /* At this point, all the chains should have refcnt >= 1. */
 775                 list_for_each_entry_safe(chain, tmp, &block->chain_list, list) {
 776                         tcf_chain_put_explicitly_created(chain);
 777                         tcf_chain_put(chain);
 778                 }
 779
 780                 block->refcnt--;
 781                 if (list_empty(&block->chain_list))
 782                         kfree(block);
 783         } else {
 784                 block->refcnt--;
 785         }
 786 }
 787 EXPORT_SYMBOL(tcf_block_put_ext);
 788
 789 void tcf_block_put(struct tcf_block *block)
 790 {
 791         struct tcf_block_ext_info ei = {0, };
 792
 793         if (!block)
 794                 return;
 795         tcf_block_put_ext(block, block->q, &ei);
 796 }
 797
 798 EXPORT_SYMBOL(tcf_block_put);
 799
 800 struct tcf_block_cb {
 801         struct list_head list;
 802         tc_setup_cb_t *cb;
 803         void *cb_ident;
 804         void *cb_priv;
 805         unsigned int refcnt;
 806 };
 807
 808 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
 809 {
 810         return block_cb->cb_priv;
 811 }
 812 EXPORT_SYMBOL(tcf_block_cb_priv);
 813
 814 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
 815                                          tc_setup_cb_t *cb, void *cb_ident)
 816 {       struct tcf_block_cb *block_cb;
 817
 818         list_for_each_entry(block_cb, &block->cb_list, list)
 819                 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
 820                         return block_cb;
 821         return NULL;
 822 }
 823 EXPORT_SYMBOL(tcf_block_cb_lookup);
 824
 825 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
 826 {
 827         block_cb->refcnt++;
 828 }
 829 EXPORT_SYMBOL(tcf_block_cb_incref);
 830
 831 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
 832 {
 833         return --block_cb->refcnt;
 834 }
 835 EXPORT_SYMBOL(tcf_block_cb_decref);
 836
 837 static int
 838 tcf_block_playback_offloads(struct tcf_block *block, tc_setup_cb_t *cb,
 839                             void *cb_priv, bool add, bool offload_in_use,
 840                             struct netlink_ext_ack *extack)
 841 {
 842         struct tcf_chain *chain;
 843         struct tcf_proto *tp;
 844         int err;
 845
 846         list_for_each_entry(chain, &block->chain_list, list) {
 847                 for (tp = rtnl_dereference(chain->filter_chain); tp;
 848                      tp = rtnl_dereference(tp->next)) {
 849                         if (tp->ops->reoffload) {
 850                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
 851                                                          extack);
 852                                 if (err && add)
 853                                         goto err_playback_remove;
 854                         } else if (add && offload_in_use) {
 855                                 err = -EOPNOTSUPP;
 856                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
 857                                 goto err_playback_remove;
 858                         }
 859                 }
 860         }
 861
 862         return 0;
 863
 864 err_playback_remove:
 865         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
 866                                     extack);
 867         return err;
 868 }
 869
 870 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
 871                                              tc_setup_cb_t *cb, void *cb_ident,
 872                                              void *cb_priv,
 873                                              struct netlink_ext_ack *extack)
 874 {
 875         struct tcf_block_cb *block_cb;
 876         int err;
 877
 878         /* Replay any already present rules */
 879         err = tcf_block_playback_offloads(block, cb, cb_priv, true,
 880                                           tcf_block_offload_in_use(block),
 881                                           extack);
 882         if (err)
 883                 return ERR_PTR(err);
 884
 885         block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
 886         if (!block_cb)
 887                 return ERR_PTR(-ENOMEM);
 888         block_cb->cb = cb;
 889         block_cb->cb_ident = cb_ident;
 890         block_cb->cb_priv = cb_priv;
 891         list_add(&block_cb->list, &block->cb_list);
 892         return block_cb;
 893 }
 894 EXPORT_SYMBOL(__tcf_block_cb_register);
 895
 896 int tcf_block_cb_register(struct tcf_block *block,
 897                           tc_setup_cb_t *cb, void *cb_ident,
 898                           void *cb_priv, struct netlink_ext_ack *extack)
 899 {
 900         struct tcf_block_cb *block_cb;
 901
 902         block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv,
 903                                            extack);
 904         return PTR_ERR_OR_ZERO(block_cb);
 905 }
 906 EXPORT_SYMBOL(tcf_block_cb_register);
 907
 908 void __tcf_block_cb_unregister(struct tcf_block *block,
 909                                struct tcf_block_cb *block_cb)
 910 {
 911         tcf_block_playback_offloads(block, block_cb->cb, block_cb->cb_priv,
 912                                     false, tcf_block_offload_in_use(block),
 913                                     NULL);
 914         list_del(&block_cb->list);
 915         kfree(block_cb);
 916 }
 917 EXPORT_SYMBOL(__tcf_block_cb_unregister);
 918
 919 void tcf_block_cb_unregister(struct tcf_block *block,
 920                              tc_setup_cb_t *cb, void *cb_ident)
 921 {
 922         struct tcf_block_cb *block_cb;
 923
 924         block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
 925         if (!block_cb)
 926                 return;
 927         __tcf_block_cb_unregister(block, block_cb);
 928 }
 929 EXPORT_SYMBOL(tcf_block_cb_unregister);
 930
 931 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type,
 932                              void *type_data, bool err_stop)
 933 {
 934         struct tcf_block_cb *block_cb;
 935         int ok_count = 0;
 936         int err;
 937
 938         /* Make sure all netdevs sharing this block are offload-capable. */
 939         if (block->nooffloaddevcnt && err_stop)
 940                 return -EOPNOTSUPP;
 941
 942         list_for_each_entry(block_cb, &block->cb_list, list) {
 943                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
 944                 if (err) {
 945                         if (err_stop)
 946                                 return err;
 947                 } else {
 948                         ok_count++;
 949                 }
 950         }
 951         return ok_count;
 952 }
 953
 954 /* Main classifier routine: scans classifier chain attached
 955  * to this qdisc, (optionally) tests for protocol and asks
 956  * specific classifiers.
 957  */
 958 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 959                  struct tcf_result *res, bool compat_mode)
 960 {
 961         __be16 protocol = tc_skb_protocol(skb);
 962 #ifdef CONFIG_NET_CLS_ACT
 963         const int max_reclassify_loop = 4;
 964         const struct tcf_proto *orig_tp = tp;
 965         const struct tcf_proto *first_tp;
 966         int limit = 0;
 967
 968 reclassify:
 969 #endif
 970         for (; tp; tp = rcu_dereference_bh(tp->next)) {
 971                 int err;
 972
 973                 if (tp->protocol != protocol &&
 974                     tp->protocol != htons(ETH_P_ALL))
 975                         continue;
 976
 977                 err = tp->classify(skb, tp, res);
 978 #ifdef CONFIG_NET_CLS_ACT
 979                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
 980                         first_tp = orig_tp;
 981                         goto reset;
 982                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
 983                         first_tp = res->goto_tp;
 984                         goto reset;
 985                 }
 986 #endif
 987                 if (err >= 0)
 988                         return err;
 989         }
 990
 991         return TC_ACT_UNSPEC; /* signal: continue lookup */
 992 #ifdef CONFIG_NET_CLS_ACT
 993 reset:
 994         if (unlikely(limit++ >= max_reclassify_loop)) {
 995                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
 996                                        tp->chain->block->index,
 997                                        tp->prio & 0xffff,
 998                                        ntohs(tp->protocol));
 999                 return TC_ACT_SHOT;
1000         }
1001
1002         tp = first_tp;
1003         protocol = tc_skb_protocol(skb);
1004         goto reclassify;
1005 #endif
1006 }
1007 EXPORT_SYMBOL(tcf_classify);
1008
1009 struct tcf_chain_info {
1010         struct tcf_proto __rcu **pprev;
1011         struct tcf_proto __rcu *next;
1012 };
1013
1014 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
1015 {
1016         return rtnl_dereference(*chain_info->pprev);
1017 }
1018
1019 static void tcf_chain_tp_insert(struct tcf_chain *chain,
1020                                 struct tcf_chain_info *chain_info,
1021                                 struct tcf_proto *tp)
1022 {
1023         if (*chain_info->pprev == chain->filter_chain)
1024                 tcf_chain0_head_change(chain, tp);
1025         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
1026         rcu_assign_pointer(*chain_info->pprev, tp);
1027         tcf_chain_hold(chain);
1028 }
1029
1030 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1031                                 struct tcf_chain_info *chain_info,
1032                                 struct tcf_proto *tp)
1033 {
1034         struct tcf_proto *next = rtnl_dereference(chain_info->next);
1035
1036         if (tp == chain->filter_chain)
1037                 tcf_chain0_head_change(chain, next);
1038         RCU_INIT_POINTER(*chain_info->pprev, next);
1039         tcf_chain_put(chain);
1040 }
1041
1042 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1043                                            struct tcf_chain_info *chain_info,
1044                                            u32 protocol, u32 prio,
1045                                            bool prio_allocate)
1046 {
1047         struct tcf_proto **pprev;
1048         struct tcf_proto *tp;
1049
1050         /* Check the chain for existence of proto-tcf with this priority */
1051         for (pprev = &chain->filter_chain;
1052              (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
1053                 if (tp->prio >= prio) {
1054                         if (tp->prio == prio) {
1055                                 if (prio_allocate ||
1056                                     (tp->protocol != protocol && protocol))
1057                                         return ERR_PTR(-EINVAL);
1058                         } else {
1059                                 tp = NULL;
1060                         }
1061                         break;
1062                 }
1063         }
1064         chain_info->pprev = pprev;
1065         chain_info->next = tp ? tp->next : NULL;
1066         return tp;
1067 }
1068
1069 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1070                          struct tcf_proto *tp, struct tcf_block *block,
1071                          struct Qdisc *q, u32 parent, void *fh,
1072                          u32 portid, u32 seq, u16 flags, int event)
1073 {
1074         struct tcmsg *tcm;
1075         struct nlmsghdr  *nlh;
1076         unsigned char *b = skb_tail_pointer(skb);
1077
1078         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1079         if (!nlh)
1080                 goto out_nlmsg_trim;
1081         tcm = nlmsg_data(nlh);
1082         tcm->tcm_family = AF_UNSPEC;
1083         tcm->tcm__pad1 = 0;
1084         tcm->tcm__pad2 = 0;
1085         if (q) {
1086                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1087                 tcm->tcm_parent = parent;
1088         } else {
1089                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1090                 tcm->tcm_block_index = block->index;
1091         }
1092         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1093         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1094                 goto nla_put_failure;
1095         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1096                 goto nla_put_failure;
1097         if (!fh) {
1098                 tcm->tcm_handle = 0;
1099         } else {
1100                 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
1101                         goto nla_put_failure;
1102         }
1103         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1104         return skb->len;
1105
1106 out_nlmsg_trim:
1107 nla_put_failure:
1108         nlmsg_trim(skb, b);
1109         return -1;
1110 }
1111
1112 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1113                           struct nlmsghdr *n, struct tcf_proto *tp,
1114                           struct tcf_block *block, struct Qdisc *q,
1115                           u32 parent, void *fh, int event, bool unicast)
1116 {
1117         struct sk_buff *skb;
1118         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1119
1120         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1121         if (!skb)
1122                 return -ENOBUFS;
1123
1124         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1125                           n->nlmsg_seq, n->nlmsg_flags, event) <= 0) {
1126                 kfree_skb(skb);
1127                 return -EINVAL;
1128         }
1129
1130         if (unicast)
1131                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1132
1133         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1134                               n->nlmsg_flags & NLM_F_ECHO);
1135 }
1136
1137 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1138                               struct nlmsghdr *n, struct tcf_proto *tp,
1139                               struct tcf_block *block, struct Qdisc *q,
1140                               u32 parent, void *fh, bool unicast, bool *last,
1141                               struct netlink_ext_ack *extack)
1142 {
1143         struct sk_buff *skb;
1144         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1145         int err;
1146
1147         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1148         if (!skb)
1149                 return -ENOBUFS;
1150
1151         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1152                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
1153                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1154                 kfree_skb(skb);
1155                 return -EINVAL;
1156         }
1157
1158         err = tp->ops->delete(tp, fh, last, extack);
1159         if (err) {
1160                 kfree_skb(skb);
1161                 return err;
1162         }
1163
1164         if (unicast)
1165                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1166
1167         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1168                              n->nlmsg_flags & NLM_F_ECHO);
1169         if (err < 0)
1170                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1171         return err;
1172 }
1173
1174 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1175                                  struct tcf_block *block, struct Qdisc *q,
1176                                  u32 parent, struct nlmsghdr *n,
1177                                  struct tcf_chain *chain, int event)
1178 {
1179         struct tcf_proto *tp;
1180
1181         for (tp = rtnl_dereference(chain->filter_chain);
1182              tp; tp = rtnl_dereference(tp->next))
1183                 tfilter_notify(net, oskb, n, tp, block,
1184                                q, parent, NULL, event, false);
1185 }
1186
1187 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1188                           struct netlink_ext_ack *extack)
1189 {
1190         struct net *net = sock_net(skb->sk);
1191         struct nlattr *tca[TCA_MAX + 1];
1192         struct tcmsg *t;
1193         u32 protocol;
1194         u32 prio;
1195         bool prio_allocate;
1196         u32 parent;
1197         u32 chain_index;
1198         struct Qdisc *q = NULL;
1199         struct tcf_chain_info chain_info;
1200         struct tcf_chain *chain = NULL;
1201         struct tcf_block *block;
1202         struct tcf_proto *tp;
1203         unsigned long cl;
1204         void *fh;
1205         int err;
1206         int tp_created;
1207
1208         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1209                 return -EPERM;
1210
1211 replay:
1212         tp_created = 0;
1213
1214         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1215         if (err < 0)
1216                 return err;
1217
1218         t = nlmsg_data(n);
1219         protocol = TC_H_MIN(t->tcm_info);
1220         prio = TC_H_MAJ(t->tcm_info);
1221         prio_allocate = false;
1222         parent = t->tcm_parent;
1223         cl = 0;
1224
1225         if (prio == 0) {
1226                 /* If no priority is provided by the user,
1227                  * we allocate one.
1228                  */
1229                 if (n->nlmsg_flags & NLM_F_CREATE) {
1230                         prio = TC_H_MAKE(0x80000000U, 0U);
1231                         prio_allocate = true;
1232                 } else {
1233                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1234                         return -ENOENT;
1235                 }
1236         }
1237
1238         /* Find head of filter chain. */
1239
1240         block = tcf_block_find(net, &q, &parent, &cl,
1241                                t->tcm_ifindex, t->tcm_block_index, extack);
1242         if (IS_ERR(block)) {
1243                 err = PTR_ERR(block);
1244                 goto errout;
1245         }
1246
1247         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1248         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1249                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1250                 err = -EINVAL;
1251                 goto errout;
1252         }
1253         chain = tcf_chain_get(block, chain_index, true);
1254         if (!chain) {
1255                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
1256                 err = -ENOMEM;
1257                 goto errout;
1258         }
1259
1260         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1261                                prio, prio_allocate);
1262         if (IS_ERR(tp)) {
1263                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1264                 err = PTR_ERR(tp);
1265                 goto errout;
1266         }
1267
1268         if (tp == NULL) {
1269                 /* Proto-tcf does not exist, create new one */
1270
1271                 if (tca[TCA_KIND] == NULL || !protocol) {
1272                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
1273                         err = -EINVAL;
1274                         goto errout;
1275                 }
1276
1277                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1278                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1279                         err = -ENOENT;
1280                         goto errout;
1281                 }
1282
1283                 if (prio_allocate)
1284                         prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
1285
1286                 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
1287                                       protocol, prio, chain, extack);
1288                 if (IS_ERR(tp)) {
1289                         err = PTR_ERR(tp);
1290                         goto errout;
1291                 }
1292                 tp_created = 1;
1293         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1294                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1295                 err = -EINVAL;
1296                 goto errout;
1297         }
1298
1299         fh = tp->ops->get(tp, t->tcm_handle);
1300
1301         if (!fh) {
1302                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1303                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1304                         err = -ENOENT;
1305                         goto errout;
1306                 }
1307         } else if (n->nlmsg_flags & NLM_F_EXCL) {
1308                 NL_SET_ERR_MSG(extack, "Filter already exists");
1309                 err = -EEXIST;
1310                 goto errout;
1311         }
1312
1313         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
1314                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
1315                 err = -EINVAL;
1316                 goto errout;
1317         }
1318
1319         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
1320                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
1321                               extack);
1322         if (err == 0) {
1323                 if (tp_created)
1324                         tcf_chain_tp_insert(chain, &chain_info, tp);
1325                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1326                                RTM_NEWTFILTER, false);
1327         } else {
1328                 if (tp_created)
1329                         tcf_proto_destroy(tp, NULL);
1330         }
1331
1332 errout:
1333         if (chain)
1334                 tcf_chain_put(chain);
1335         if (err == -EAGAIN)
1336                 /* Replay the request. */
1337                 goto replay;
1338         return err;
1339 }
1340
1341 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1342                           struct netlink_ext_ack *extack)
1343 {
1344         struct net *net = sock_net(skb->sk);
1345         struct nlattr *tca[TCA_MAX + 1];
1346         struct tcmsg *t;
1347         u32 protocol;
1348         u32 prio;
1349         u32 parent;
1350         u32 chain_index;
1351         struct Qdisc *q = NULL;
1352         struct tcf_chain_info chain_info;
1353         struct tcf_chain *chain = NULL;
1354         struct tcf_block *block;
1355         struct tcf_proto *tp = NULL;
1356         unsigned long cl = 0;
1357         void *fh = NULL;
1358         int err;
1359
1360         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1361                 return -EPERM;
1362
1363         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1364         if (err < 0)
1365                 return err;
1366
1367         t = nlmsg_data(n);
1368         protocol = TC_H_MIN(t->tcm_info);
1369         prio = TC_H_MAJ(t->tcm_info);
1370         parent = t->tcm_parent;
1371
1372         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
1373                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
1374                 return -ENOENT;
1375         }
1376
1377         /* Find head of filter chain. */
1378
1379         block = tcf_block_find(net, &q, &parent, &cl,
1380                                t->tcm_ifindex, t->tcm_block_index, extack);
1381         if (IS_ERR(block)) {
1382                 err = PTR_ERR(block);
1383                 goto errout;
1384         }
1385
1386         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1387         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1388                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1389                 err = -EINVAL;
1390                 goto errout;
1391         }
1392         chain = tcf_chain_get(block, chain_index, false);
1393         if (!chain) {
1394                 /* User requested flush on non-existent chain. Nothing to do,
1395                  * so just return success.
1396                  */
1397                 if (prio == 0) {
1398                         err = 0;
1399                         goto errout;
1400                 }
1401                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1402                 err = -ENOENT;
1403                 goto errout;
1404         }
1405
1406         if (prio == 0) {
1407                 tfilter_notify_chain(net, skb, block, q, parent, n,
1408                                      chain, RTM_DELTFILTER);
1409                 tcf_chain_flush(chain);
1410                 err = 0;
1411                 goto errout;
1412         }
1413
1414         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1415                                prio, false);
1416         if (!tp || IS_ERR(tp)) {
1417                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1418                 err = tp ? PTR_ERR(tp) : -ENOENT;
1419                 goto errout;
1420         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1421                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1422                 err = -EINVAL;
1423                 goto errout;
1424         }
1425
1426         fh = tp->ops->get(tp, t->tcm_handle);
1427
1428         if (!fh) {
1429                 if (t->tcm_handle == 0) {
1430                         tcf_chain_tp_remove(chain, &chain_info, tp);
1431                         tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1432                                        RTM_DELTFILTER, false);
1433                         tcf_proto_destroy(tp, extack);
1434                         err = 0;
1435                 } else {
1436                         NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1437                         err = -ENOENT;
1438                 }
1439         } else {
1440                 bool last;
1441
1442                 err = tfilter_del_notify(net, skb, n, tp, block,
1443                                          q, parent, fh, false, &last,
1444                                          extack);
1445                 if (err)
1446                         goto errout;
1447                 if (last) {
1448                         tcf_chain_tp_remove(chain, &chain_info, tp);
1449                         tcf_proto_destroy(tp, extack);
1450                 }
1451         }
1452
1453 errout:
1454         if (chain)
1455                 tcf_chain_put(chain);
1456         return err;
1457 }
1458
1459 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1460                           struct netlink_ext_ack *extack)
1461 {
1462         struct net *net = sock_net(skb->sk);
1463         struct nlattr *tca[TCA_MAX + 1];
1464         struct tcmsg *t;
1465         u32 protocol;
1466         u32 prio;
1467         u32 parent;
1468         u32 chain_index;
1469         struct Qdisc *q = NULL;
1470         struct tcf_chain_info chain_info;
1471         struct tcf_chain *chain = NULL;
1472         struct tcf_block *block;
1473         struct tcf_proto *tp = NULL;
1474         unsigned long cl = 0;
1475         void *fh = NULL;
1476         int err;
1477
1478         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1479         if (err < 0)
1480                 return err;
1481
1482         t = nlmsg_data(n);
1483         protocol = TC_H_MIN(t->tcm_info);
1484         prio = TC_H_MAJ(t->tcm_info);
1485         parent = t->tcm_parent;
1486
1487         if (prio == 0) {
1488                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1489                 return -ENOENT;
1490         }
1491
1492         /* Find head of filter chain. */
1493
1494         block = tcf_block_find(net, &q, &parent, &cl,
1495                                t->tcm_ifindex, t->tcm_block_index, extack);
1496         if (IS_ERR(block)) {
1497                 err = PTR_ERR(block);
1498                 goto errout;
1499         }
1500
1501         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1502         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1503                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1504                 err = -EINVAL;
1505                 goto errout;
1506         }
1507         chain = tcf_chain_get(block, chain_index, false);
1508         if (!chain) {
1509                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1510                 err = -EINVAL;
1511                 goto errout;
1512         }
1513
1514         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1515                                prio, false);
1516         if (!tp || IS_ERR(tp)) {
1517                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1518                 err = tp ? PTR_ERR(tp) : -ENOENT;
1519                 goto errout;
1520         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1521                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1522                 err = -EINVAL;
1523                 goto errout;
1524         }
1525
1526         fh = tp->ops->get(tp, t->tcm_handle);
1527
1528         if (!fh) {
1529                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1530                 err = -ENOENT;
1531         } else {
1532                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
1533                                      fh, RTM_NEWTFILTER, true);
1534                 if (err < 0)
1535                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
1536         }
1537
1538 errout:
1539         if (chain)
1540                 tcf_chain_put(chain);
1541         return err;
1542 }
1543
1544 struct tcf_dump_args {
1545         struct tcf_walker w;
1546         struct sk_buff *skb;
1547         struct netlink_callback *cb;
1548         struct tcf_block *block;
1549         struct Qdisc *q;
1550         u32 parent;
1551 };
1552
1553 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1554 {
1555         struct tcf_dump_args *a = (void *)arg;
1556         struct net *net = sock_net(a->skb->sk);
1557
1558         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
1559                              n, NETLINK_CB(a->cb->skb).portid,
1560                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1561                              RTM_NEWTFILTER);
1562 }
1563
1564 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
1565                            struct sk_buff *skb, struct netlink_callback *cb,
1566                            long index_start, long *p_index)
1567 {
1568         struct net *net = sock_net(skb->sk);
1569         struct tcf_block *block = chain->block;
1570         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1571         struct tcf_dump_args arg;
1572         struct tcf_proto *tp;
1573
1574         for (tp = rtnl_dereference(chain->filter_chain);
1575              tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
1576                 if (*p_index < index_start)
1577                         continue;
1578                 if (TC_H_MAJ(tcm->tcm_info) &&
1579                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
1580                         continue;
1581                 if (TC_H_MIN(tcm->tcm_info) &&
1582                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
1583                         continue;
1584                 if (*p_index > index_start)
1585                         memset(&cb->args[1], 0,
1586                                sizeof(cb->args) - sizeof(cb->args[0]));
1587                 if (cb->args[1] == 0) {
1588                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
1589                                           NETLINK_CB(cb->skb).portid,
1590                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
1591                                           RTM_NEWTFILTER) <= 0)
1592                                 return false;
1593
1594                         cb->args[1] = 1;
1595                 }
1596                 if (!tp->ops->walk)
1597                         continue;
1598                 arg.w.fn = tcf_node_dump;
1599                 arg.skb = skb;
1600                 arg.cb = cb;
1601                 arg.block = block;
1602                 arg.q = q;
1603                 arg.parent = parent;
1604                 arg.w.stop = 0;
1605                 arg.w.skip = cb->args[1] - 1;
1606                 arg.w.count = 0;
1607                 arg.w.cookie = cb->args[2];
1608                 tp->ops->walk(tp, &arg.w);
1609                 cb->args[2] = arg.w.cookie;
1610                 cb->args[1] = arg.w.count + 1;
1611                 if (arg.w.stop)
1612                         return false;
1613         }
1614         return true;
1615 }
1616
1617 /* called with RTNL */
1618 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
1619 {
1620         struct net *net = sock_net(skb->sk);
1621         struct nlattr *tca[TCA_MAX + 1];
1622         struct Qdisc *q = NULL;
1623         struct tcf_block *block;
1624         struct tcf_chain *chain;
1625         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1626         long index_start;
1627         long index;
1628         u32 parent;
1629         int err;
1630
1631         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1632                 return skb->len;
1633
1634         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1635         if (err)
1636                 return err;
1637
1638         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1639                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1640                 if (!block)
1641                         goto out;
1642                 /* If we work with block index, q is NULL and parent value
1643                  * will never be used in the following code. The check
1644                  * in tcf_fill_node prevents it. However, compiler does not
1645                  * see that far, so set parent to zero to silence the warning
1646                  * about parent being uninitialized.
1647                  */
1648                 parent = 0;
1649         } else {
1650                 const struct Qdisc_class_ops *cops;
1651                 struct net_device *dev;
1652                 unsigned long cl = 0;
1653
1654                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1655                 if (!dev)
1656                         return skb->len;
1657
1658                 parent = tcm->tcm_parent;
1659                 if (!parent) {
1660                         q = dev->qdisc;
1661                         parent = q->handle;
1662                 } else {
1663                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1664                 }
1665                 if (!q)
1666                         goto out;
1667                 cops = q->ops->cl_ops;
1668                 if (!cops)
1669                         goto out;
1670                 if (!cops->tcf_block)
1671                         goto out;
1672                 if (TC_H_MIN(tcm->tcm_parent)) {
1673                         cl = cops->find(q, tcm->tcm_parent);
1674                         if (cl == 0)
1675                                 goto out;
1676                 }
1677                 block = cops->tcf_block(q, cl, NULL);
1678                 if (!block)
1679                         goto out;
1680                 if (tcf_block_shared(block))
1681                         q = NULL;
1682         }
1683
1684         index_start = cb->args[0];
1685         index = 0;
1686
1687         list_for_each_entry(chain, &block->chain_list, list) {
1688                 if (tca[TCA_CHAIN] &&
1689                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1690                         continue;
1691                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
1692                                     index_start, &index)) {
1693                         err = -EMSGSIZE;
1694                         break;
1695                 }
1696         }
1697
1698         cb->args[0] = index;
1699
1700 out:
1701         /* If we did no progress, the error (EMSGSIZE) is real */
1702         if (skb->len == 0 && err)
1703                 return err;
1704         return skb->len;
1705 }
1706
1707 static int tc_chain_fill_node(struct tcf_chain *chain, struct net *net,
1708                               struct sk_buff *skb, struct tcf_block *block,
1709                               u32 portid, u32 seq, u16 flags, int event)
1710 {
1711         unsigned char *b = skb_tail_pointer(skb);
1712         const struct tcf_proto_ops *ops;
1713         struct nlmsghdr *nlh;
1714         struct tcmsg *tcm;
1715         void *priv;
1716
1717         ops = chain->tmplt_ops;
1718         priv = chain->tmplt_priv;
1719
1720         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1721         if (!nlh)
1722                 goto out_nlmsg_trim;
1723         tcm = nlmsg_data(nlh);
1724         tcm->tcm_family = AF_UNSPEC;
1725         tcm->tcm__pad1 = 0;
1726         tcm->tcm__pad2 = 0;
1727         tcm->tcm_handle = 0;
1728         if (block->q) {
1729                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
1730                 tcm->tcm_parent = block->q->handle;
1731         } else {
1732                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1733                 tcm->tcm_block_index = block->index;
1734         }
1735
1736         if (nla_put_u32(skb, TCA_CHAIN, chain->index))
1737                 goto nla_put_failure;
1738
1739         if (ops) {
1740                 if (nla_put_string(skb, TCA_KIND, ops->kind))
1741                         goto nla_put_failure;
1742                 if (ops->tmplt_dump(skb, net, priv) < 0)
1743                         goto nla_put_failure;
1744         }
1745
1746         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1747         return skb->len;
1748
1749 out_nlmsg_trim:
1750 nla_put_failure:
1751         nlmsg_trim(skb, b);
1752         return -EMSGSIZE;
1753 }
1754
1755 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
1756                            u32 seq, u16 flags, int event, bool unicast)
1757 {
1758         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1759         struct tcf_block *block = chain->block;
1760         struct net *net = block->net;
1761         struct sk_buff *skb;
1762
1763         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1764         if (!skb)
1765                 return -ENOBUFS;
1766
1767         if (tc_chain_fill_node(chain, net, skb, block, portid,
1768                                seq, flags, event) <= 0) {
1769                 kfree_skb(skb);
1770                 return -EINVAL;
1771         }
1772
1773         if (unicast)
1774                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1775
1776         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
1777 }
1778
1779 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
1780                               struct nlattr **tca,
1781                               struct netlink_ext_ack *extack)
1782 {
1783         const struct tcf_proto_ops *ops;
1784         void *tmplt_priv;
1785
1786         /* If kind is not set, user did not specify template. */
1787         if (!tca[TCA_KIND])
1788                 return 0;
1789
1790         ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), extack);
1791         if (IS_ERR(ops))
1792                 return PTR_ERR(ops);
1793         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
1794                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
1795                 return -EOPNOTSUPP;
1796         }
1797
1798         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
1799         if (IS_ERR(tmplt_priv)) {
1800                 module_put(ops->owner);
1801                 return PTR_ERR(tmplt_priv);
1802         }
1803         chain->tmplt_ops = ops;
1804         chain->tmplt_priv = tmplt_priv;
1805         return 0;
1806 }
1807
1808 static void tc_chain_tmplt_del(struct tcf_chain *chain)
1809 {
1810         const struct tcf_proto_ops *ops = chain->tmplt_ops;
1811
1812         /* If template ops are set, no work to do for us. */
1813         if (!ops)
1814                 return;
1815
1816         ops->tmplt_destroy(chain->tmplt_priv);
1817         module_put(ops->owner);
1818 }
1819
1820 /* Add/delete/get a chain */
1821
1822 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
1823                         struct netlink_ext_ack *extack)
1824 {
1825         struct net *net = sock_net(skb->sk);
1826         struct nlattr *tca[TCA_MAX + 1];
1827         struct tcmsg *t;
1828         u32 parent;
1829         u32 chain_index;
1830         struct Qdisc *q = NULL;
1831         struct tcf_chain *chain = NULL;
1832         struct tcf_block *block;
1833         unsigned long cl;
1834         int err;
1835
1836         if (n->nlmsg_type != RTM_GETCHAIN &&
1837             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1838                 return -EPERM;
1839
1840 replay:
1841         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL, extack);
1842         if (err < 0)
1843                 return err;
1844
1845         t = nlmsg_data(n);
1846         parent = t->tcm_parent;
1847         cl = 0;
1848
1849         block = tcf_block_find(net, &q, &parent, &cl,
1850                                t->tcm_ifindex, t->tcm_block_index, extack);
1851         if (IS_ERR(block))
1852                 return PTR_ERR(block);
1853
1854         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1855         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1856                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1857                 return -EINVAL;
1858         }
1859         chain = tcf_chain_lookup(block, chain_index);
1860         if (n->nlmsg_type == RTM_NEWCHAIN) {
1861                 if (chain) {
1862                         if (tcf_chain_held_by_acts_only(chain)) {
1863                                 /* The chain exists only because there is
1864                                  * some action referencing it.
1865                                  */
1866                                 tcf_chain_hold(chain);
1867                         } else {
1868                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
1869                                 return -EEXIST;
1870                         }
1871                 } else {
1872                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1873                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
1874                                 return -ENOENT;
1875                         }
1876                         chain = tcf_chain_create(block, chain_index);
1877                         if (!chain) {
1878                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
1879                                 return -ENOMEM;
1880                         }
1881                 }
1882         } else {
1883                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
1884                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1885                         return -EINVAL;
1886                 }
1887                 tcf_chain_hold(chain);
1888         }
1889
1890         switch (n->nlmsg_type) {
1891         case RTM_NEWCHAIN:
1892                 err = tc_chain_tmplt_add(chain, net, tca, extack);
1893                 if (err)
1894                         goto errout;
1895                 /* In case the chain was successfully added, take a reference
1896                  * to the chain. This ensures that an empty chain
1897                  * does not disappear at the end of this function.
1898                  */
1899                 tcf_chain_hold(chain);
1900                 chain->explicitly_created = true;
1901                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
1902                                 RTM_NEWCHAIN, false);
1903                 break;
1904         case RTM_DELCHAIN:
1905                 tfilter_notify_chain(net, skb, block, q, parent, n,
1906                                      chain, RTM_DELTFILTER);
1907                 /* Flush the chain first as the user requested chain removal. */
1908                 tcf_chain_flush(chain);
1909                 /* In case the chain was successfully deleted, put a reference
1910                  * to the chain previously taken during addition.
1911                  */
1912                 tcf_chain_put_explicitly_created(chain);
1913                 chain->explicitly_created = false;
1914                 break;
1915         case RTM_GETCHAIN:
1916                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
1917                                       n->nlmsg_seq, n->nlmsg_type, true);
1918                 if (err < 0)
1919                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
1920                 break;
1921         default:
1922                 err = -EOPNOTSUPP;
1923                 NL_SET_ERR_MSG(extack, "Unsupported message type");
1924                 goto errout;
1925         }
1926
1927 errout:
1928         tcf_chain_put(chain);
1929         if (err == -EAGAIN)
1930                 /* Replay the request. */
1931                 goto replay;
1932         return err;
1933 }
1934
1935 /* called with RTNL */
1936 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
1937 {
1938         struct net *net = sock_net(skb->sk);
1939         struct nlattr *tca[TCA_MAX + 1];
1940         struct Qdisc *q = NULL;
1941         struct tcf_block *block;
1942         struct tcf_chain *chain;
1943         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1944         long index_start;
1945         long index;
1946         u32 parent;
1947         int err;
1948
1949         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1950                 return skb->len;
1951
1952         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL, NULL);
1953         if (err)
1954                 return err;
1955
1956         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1957                 block = tcf_block_lookup(net, tcm->tcm_block_index);
1958                 if (!block)
1959                         goto out;
1960                 /* If we work with block index, q is NULL and parent value
1961                  * will never be used in the following code. The check
1962                  * in tcf_fill_node prevents it. However, compiler does not
1963                  * see that far, so set parent to zero to silence the warning
1964                  * about parent being uninitialized.
1965                  */
1966                 parent = 0;
1967         } else {
1968                 const struct Qdisc_class_ops *cops;
1969                 struct net_device *dev;
1970                 unsigned long cl = 0;
1971
1972                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1973                 if (!dev)
1974                         return skb->len;
1975
1976                 parent = tcm->tcm_parent;
1977                 if (!parent) {
1978                         q = dev->qdisc;
1979                         parent = q->handle;
1980                 } else {
1981                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1982                 }
1983                 if (!q)
1984                         goto out;
1985                 cops = q->ops->cl_ops;
1986                 if (!cops)
1987                         goto out;
1988                 if (!cops->tcf_block)
1989                         goto out;
1990                 if (TC_H_MIN(tcm->tcm_parent)) {
1991                         cl = cops->find(q, tcm->tcm_parent);
1992                         if (cl == 0)
1993                                 goto out;
1994                 }
1995                 block = cops->tcf_block(q, cl, NULL);
1996                 if (!block)
1997                         goto out;
1998                 if (tcf_block_shared(block))
1999                         q = NULL;
2000         }
2001
2002         index_start = cb->args[0];
2003         index = 0;
2004
2005         list_for_each_entry(chain, &block->chain_list, list) {
2006                 if ((tca[TCA_CHAIN] &&
2007                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2008                         continue;
2009                 if (index < index_start) {
2010                         index++;
2011                         continue;
2012                 }
2013                 if (tcf_chain_held_by_acts_only(chain))
2014                         continue;
2015                 err = tc_chain_fill_node(chain, net, skb, block,
2016                                          NETLINK_CB(cb->skb).portid,
2017                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
2018                                          RTM_NEWCHAIN);
2019                 if (err <= 0)
2020                         break;
2021                 index++;
2022         }
2023
2024         cb->args[0] = index;
2025
2026 out:
2027         /* If we did no progress, the error (EMSGSIZE) is real */
2028         if (skb->len == 0 && err)
2029                 return err;
2030         return skb->len;
2031 }
2032
2033 void tcf_exts_destroy(struct tcf_exts *exts)
2034 {
2035 #ifdef CONFIG_NET_CLS_ACT
2036         tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
2037         kfree(exts->actions);
2038         exts->nr_actions = 0;
2039 #endif
2040 }
2041 EXPORT_SYMBOL(tcf_exts_destroy);
2042
2043 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
2044                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
2045                       struct netlink_ext_ack *extack)
2046 {
2047 #ifdef CONFIG_NET_CLS_ACT
2048         {
2049                 struct tc_action *act;
2050                 size_t attr_size = 0;
2051
2052                 if (exts->police && tb[exts->police]) {
2053                         act = tcf_action_init_1(net, tp, tb[exts->police],
2054                                                 rate_tlv, "police", ovr,
2055                                                 TCA_ACT_BIND, true, extack);
2056                         if (IS_ERR(act))
2057                                 return PTR_ERR(act);
2058
2059                         act->type = exts->type = TCA_OLD_COMPAT;
2060                         exts->actions[0] = act;
2061                         exts->nr_actions = 1;
2062                 } else if (exts->action && tb[exts->action]) {
2063                         int err;
2064
2065                         err = tcf_action_init(net, tp, tb[exts->action],
2066                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
2067                                               exts->actions, &attr_size, true,
2068                                               extack);
2069                         if (err < 0)
2070                                 return err;
2071                         exts->nr_actions = err;
2072                 }
2073                 exts->net = net;
2074         }
2075 #else
2076         if ((exts->action && tb[exts->action]) ||
2077             (exts->police && tb[exts->police])) {
2078                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
2079                 return -EOPNOTSUPP;
2080         }
2081 #endif
2082
2083         return 0;
2084 }
2085 EXPORT_SYMBOL(tcf_exts_validate);
2086
2087 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
2088 {
2089 #ifdef CONFIG_NET_CLS_ACT
2090         struct tcf_exts old = *dst;
2091
2092         *dst = *src;
2093         tcf_exts_destroy(&old);
2094 #endif
2095 }
2096 EXPORT_SYMBOL(tcf_exts_change);
2097
2098 #ifdef CONFIG_NET_CLS_ACT
2099 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
2100 {
2101         if (exts->nr_actions == 0)
2102                 return NULL;
2103         else
2104                 return exts->actions[0];
2105 }
2106 #endif
2107
2108 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
2109 {
2110 #ifdef CONFIG_NET_CLS_ACT
2111         struct nlattr *nest;
2112
2113         if (exts->action && tcf_exts_has_actions(exts)) {
2114                 /*
2115                  * again for backward compatible mode - we want
2116                  * to work with both old and new modes of entering
2117                  * tc data even if iproute2  was newer - jhs
2118                  */
2119                 if (exts->type != TCA_OLD_COMPAT) {
2120                         nest = nla_nest_start(skb, exts->action);
2121                         if (nest == NULL)
2122                                 goto nla_put_failure;
2123
2124                         if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
2125                                 goto nla_put_failure;
2126                         nla_nest_end(skb, nest);
2127                 } else if (exts->police) {
2128                         struct tc_action *act = tcf_exts_first_act(exts);
2129                         nest = nla_nest_start(skb, exts->police);
2130                         if (nest == NULL || !act)
2131                                 goto nla_put_failure;
2132                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
2133                                 goto nla_put_failure;
2134                         nla_nest_end(skb, nest);
2135                 }
2136         }
2137         return 0;
2138
2139 nla_put_failure:
2140         nla_nest_cancel(skb, nest);
2141         return -1;
2142 #else
2143         return 0;
2144 #endif
2145 }
2146 EXPORT_SYMBOL(tcf_exts_dump);
2147
2148
2149 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
2150 {
2151 #ifdef CONFIG_NET_CLS_ACT
2152         struct tc_action *a = tcf_exts_first_act(exts);
2153         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
2154                 return -1;
2155 #endif
2156         return 0;
2157 }
2158 EXPORT_SYMBOL(tcf_exts_dump_stats);
2159
2160 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
2161                                        enum tc_setup_type type,
2162                                        void *type_data, bool err_stop)
2163 {
2164         int ok_count = 0;
2165 #ifdef CONFIG_NET_CLS_ACT
2166         const struct tc_action *a;
2167         struct net_device *dev;
2168         int i, ret;
2169
2170         if (!tcf_exts_has_actions(exts))
2171                 return 0;
2172
2173         for (i = 0; i < exts->nr_actions; i++) {
2174                 a = exts->actions[i];
2175                 if (!a->ops->get_dev)
2176                         continue;
2177                 dev = a->ops->get_dev(a);
2178                 if (!dev)
2179                         continue;
2180                 ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
2181                 a->ops->put_dev(dev);
2182                 if (ret < 0)
2183                         return ret;
2184                 ok_count += ret;
2185         }
2186 #endif
2187         return ok_count;
2188 }
2189
2190 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
2191                      enum tc_setup_type type, void *type_data, bool err_stop)
2192 {
2193         int ok_count;
2194         int ret;
2195
2196         ret = tcf_block_cb_call(block, type, type_data, err_stop);
2197         if (ret < 0)
2198                 return ret;
2199         ok_count = ret;
2200
2201         if (!exts || ok_count)
2202                 return ok_count;
2203         ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
2204         if (ret < 0)
2205                 return ret;
2206         ok_count += ret;
2207
2208         return ok_count;
2209 }
2210 EXPORT_SYMBOL(tc_setup_cb_call);
2211
2212 static __net_init int tcf_net_init(struct net *net)
2213 {
2214         struct tcf_net *tn = net_generic(net, tcf_net_id);
2215
2216         idr_init(&tn->idr);
2217         return 0;
2218 }
2219
2220 static void __net_exit tcf_net_exit(struct net *net)
2221 {
2222         struct tcf_net *tn = net_generic(net, tcf_net_id);
2223
2224         idr_destroy(&tn->idr);
2225 }
2226
2227 static struct pernet_operations tcf_net_ops = {
2228         .init = tcf_net_init,
2229         .exit = tcf_net_exit,
2230         .id   = &tcf_net_id,
2231         .size = sizeof(struct tcf_net),
2232 };
2233
2234 static int __init tc_filter_init(void)
2235 {
2236         int err;
2237
2238         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
2239         if (!tc_filter_wq)
2240                 return -ENOMEM;
2241
2242         err = register_pernet_subsys(&tcf_net_ops);
2243         if (err)
2244                 goto err_register_pernet_subsys;
2245
2246         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 0);
2247         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 0);
2248         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
2249                       tc_dump_tfilter, 0);
2250         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
2251         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
2252         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
2253                       tc_dump_chain, 0);
2254
2255         return 0;
2256
2257 err_register_pernet_subsys:
2258         destroy_workqueue(tc_filter_wq);
2259         return err;
2260 }
2261
2262 subsys_initcall(tc_filter_init);