net/sched/cls_api.c

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