net/sched/cls_api.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/cls_api.c  Packet classifier API.
   4  *
   5  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   6  *
   7  * Changes:
   8  *
   9  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/types.h>
  14 #include <linux/kernel.h>
  15 #include <linux/string.h>
  16 #include <linux/errno.h>
  17 #include <linux/err.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/init.h>
  20 #include <linux/kmod.h>
  21 #include <linux/slab.h>
  22 #include <linux/idr.h>
  23 #include <linux/rhashtable.h>
  24 #include <net/net_namespace.h>
  25 #include <net/sock.h>
  26 #include <net/netlink.h>
  27 #include <net/pkt_sched.h>
  28 #include <net/pkt_cls.h>
  29 #include <net/tc_act/tc_pedit.h>
  30 #include <net/tc_act/tc_mirred.h>
  31 #include <net/tc_act/tc_vlan.h>
  32 #include <net/tc_act/tc_tunnel_key.h>
  33 #include <net/tc_act/tc_csum.h>
  34 #include <net/tc_act/tc_gact.h>
  35 #include <net/tc_act/tc_police.h>
  36 #include <net/tc_act/tc_sample.h>
  37 #include <net/tc_act/tc_skbedit.h>
  38 #include <net/tc_act/tc_ct.h>
  39 #include <net/tc_act/tc_mpls.h>
  40 #include <net/flow_offload.h>
  41
  42 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
  43
  44 /* The list of all installed classifier types */
  45 static LIST_HEAD(tcf_proto_base);
  46
  47 /* Protects list of registered TC modules. It is pure SMP lock. */
  48 static DEFINE_RWLOCK(cls_mod_lock);
  49
  50 /* Find classifier type by string name */
  51
  52 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
  53 {
  54         const struct tcf_proto_ops *t, *res = NULL;
  55
  56         if (kind) {
  57                 read_lock(&cls_mod_lock);
  58                 list_for_each_entry(t, &tcf_proto_base, head) {
  59                         if (strcmp(kind, t->kind) == 0) {
  60                                 if (try_module_get(t->owner))
  61                                         res = t;
  62                                 break;
  63                         }
  64                 }
  65                 read_unlock(&cls_mod_lock);
  66         }
  67         return res;
  68 }
  69
  70 static const struct tcf_proto_ops *
  71 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
  72                      struct netlink_ext_ack *extack)
  73 {
  74         const struct tcf_proto_ops *ops;
  75
  76         ops = __tcf_proto_lookup_ops(kind);
  77         if (ops)
  78                 return ops;
  79 #ifdef CONFIG_MODULES
  80         if (rtnl_held)
  81                 rtnl_unlock();
  82         request_module("cls_%s", kind);
  83         if (rtnl_held)
  84                 rtnl_lock();
  85         ops = __tcf_proto_lookup_ops(kind);
  86         /* We dropped the RTNL semaphore in order to perform
  87          * the module load. So, even if we succeeded in loading
  88          * the module we have to replay the request. We indicate
  89          * this using -EAGAIN.
  90          */
  91         if (ops) {
  92                 module_put(ops->owner);
  93                 return ERR_PTR(-EAGAIN);
  94         }
  95 #endif
  96         NL_SET_ERR_MSG(extack, "TC classifier not found");
  97         return ERR_PTR(-ENOENT);
  98 }
  99
 100 /* Register(unregister) new classifier type */
 101
 102 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
 103 {
 104         struct tcf_proto_ops *t;
 105         int rc = -EEXIST;
 106
 107         write_lock(&cls_mod_lock);
 108         list_for_each_entry(t, &tcf_proto_base, head)
 109                 if (!strcmp(ops->kind, t->kind))
 110                         goto out;
 111
 112         list_add_tail(&ops->head, &tcf_proto_base);
 113         rc = 0;
 114 out:
 115         write_unlock(&cls_mod_lock);
 116         return rc;
 117 }
 118 EXPORT_SYMBOL(register_tcf_proto_ops);
 119
 120 static struct workqueue_struct *tc_filter_wq;
 121
 122 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 123 {
 124         struct tcf_proto_ops *t;
 125         int rc = -ENOENT;
 126
 127         /* Wait for outstanding call_rcu()s, if any, from a
 128          * tcf_proto_ops's destroy() handler.
 129          */
 130         rcu_barrier();
 131         flush_workqueue(tc_filter_wq);
 132
 133         write_lock(&cls_mod_lock);
 134         list_for_each_entry(t, &tcf_proto_base, head) {
 135                 if (t == ops) {
 136                         list_del(&t->head);
 137                         rc = 0;
 138                         break;
 139                 }
 140         }
 141         write_unlock(&cls_mod_lock);
 142         return rc;
 143 }
 144 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 145
 146 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 147 {
 148         INIT_RCU_WORK(rwork, func);
 149         return queue_rcu_work(tc_filter_wq, rwork);
 150 }
 151 EXPORT_SYMBOL(tcf_queue_work);
 152
 153 /* Select new prio value from the range, managed by kernel. */
 154
 155 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 156 {
 157         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 158
 159         if (tp)
 160                 first = tp->prio - 1;
 161
 162         return TC_H_MAJ(first);
 163 }
 164
 165 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
 166 {
 167         if (kind)
 168                 return nla_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
 169         memset(name, 0, IFNAMSIZ);
 170         return false;
 171 }
 172
 173 static bool tcf_proto_is_unlocked(const char *kind)
 174 {
 175         const struct tcf_proto_ops *ops;
 176         bool ret;
 177
 178         if (strlen(kind) == 0)
 179                 return false;
 180
 181         ops = tcf_proto_lookup_ops(kind, false, NULL);
 182         /* On error return false to take rtnl lock. Proto lookup/create
 183          * functions will perform lookup again and properly handle errors.
 184          */
 185         if (IS_ERR(ops))
 186                 return false;
 187
 188         ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
 189         module_put(ops->owner);
 190         return ret;
 191 }
 192
 193 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 194                                           u32 prio, struct tcf_chain *chain,
 195                                           bool rtnl_held,
 196                                           struct netlink_ext_ack *extack)
 197 {
 198         struct tcf_proto *tp;
 199         int err;
 200
 201         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 202         if (!tp)
 203                 return ERR_PTR(-ENOBUFS);
 204
 205         tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
 206         if (IS_ERR(tp->ops)) {
 207                 err = PTR_ERR(tp->ops);
 208                 goto errout;
 209         }
 210         tp->classify = tp->ops->classify;
 211         tp->protocol = protocol;
 212         tp->prio = prio;
 213         tp->chain = chain;
 214         spin_lock_init(&tp->lock);
 215         refcount_set(&tp->refcnt, 1);
 216
 217         err = tp->ops->init(tp);
 218         if (err) {
 219                 module_put(tp->ops->owner);
 220                 goto errout;
 221         }
 222         return tp;
 223
 224 errout:
 225         kfree(tp);
 226         return ERR_PTR(err);
 227 }
 228
 229 static void tcf_proto_get(struct tcf_proto *tp)
 230 {
 231         refcount_inc(&tp->refcnt);
 232 }
 233
 234 static void tcf_chain_put(struct tcf_chain *chain);
 235
 236 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 237                               struct netlink_ext_ack *extack)
 238 {
 239         tp->ops->destroy(tp, rtnl_held, extack);
 240         tcf_chain_put(tp->chain);
 241         module_put(tp->ops->owner);
 242         kfree_rcu(tp, rcu);
 243 }
 244
 245 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 246                           struct netlink_ext_ack *extack)
 247 {
 248         if (refcount_dec_and_test(&tp->refcnt))
 249                 tcf_proto_destroy(tp, rtnl_held, extack);
 250 }
 251
 252 static int walker_check_empty(struct tcf_proto *tp, void *fh,
 253                               struct tcf_walker *arg)
 254 {
 255         if (fh) {
 256                 arg->nonempty = true;
 257                 return -1;
 258         }
 259         return 0;
 260 }
 261
 262 static bool tcf_proto_is_empty(struct tcf_proto *tp, bool rtnl_held)
 263 {
 264         struct tcf_walker walker = { .fn = walker_check_empty, };
 265
 266         if (tp->ops->walk) {
 267                 tp->ops->walk(tp, &walker, rtnl_held);
 268                 return !walker.nonempty;
 269         }
 270         return true;
 271 }
 272
 273 static bool tcf_proto_check_delete(struct tcf_proto *tp, bool rtnl_held)
 274 {
 275         spin_lock(&tp->lock);
 276         if (tcf_proto_is_empty(tp, rtnl_held))
 277                 tp->deleting = true;
 278         spin_unlock(&tp->lock);
 279         return tp->deleting;
 280 }
 281
 282 static void tcf_proto_mark_delete(struct tcf_proto *tp)
 283 {
 284         spin_lock(&tp->lock);
 285         tp->deleting = true;
 286         spin_unlock(&tp->lock);
 287 }
 288
 289 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 290 {
 291         bool deleting;
 292
 293         spin_lock(&tp->lock);
 294         deleting = tp->deleting;
 295         spin_unlock(&tp->lock);
 296
 297         return deleting;
 298 }
 299
 300 #define ASSERT_BLOCK_LOCKED(block)                                      \
 301         lockdep_assert_held(&(block)->lock)
 302
 303 struct tcf_filter_chain_list_item {
 304         struct list_head list;
 305         tcf_chain_head_change_t *chain_head_change;
 306         void *chain_head_change_priv;
 307 };
 308
 309 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 310                                           u32 chain_index)
 311 {
 312         struct tcf_chain *chain;
 313
 314         ASSERT_BLOCK_LOCKED(block);
 315
 316         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 317         if (!chain)
 318                 return NULL;
 319         list_add_tail(&chain->list, &block->chain_list);
 320         mutex_init(&chain->filter_chain_lock);
 321         chain->block = block;
 322         chain->index = chain_index;
 323         chain->refcnt = 1;
 324         if (!chain->index)
 325                 block->chain0.chain = chain;
 326         return chain;
 327 }
 328
 329 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 330                                        struct tcf_proto *tp_head)
 331 {
 332         if (item->chain_head_change)
 333                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 334 }
 335
 336 static void tcf_chain0_head_change(struct tcf_chain *chain,
 337                                    struct tcf_proto *tp_head)
 338 {
 339         struct tcf_filter_chain_list_item *item;
 340         struct tcf_block *block = chain->block;
 341
 342         if (chain->index)
 343                 return;
 344
 345         mutex_lock(&block->lock);
 346         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 347                 tcf_chain_head_change_item(item, tp_head);
 348         mutex_unlock(&block->lock);
 349 }
 350
 351 /* Returns true if block can be safely freed. */
 352
 353 static bool tcf_chain_detach(struct tcf_chain *chain)
 354 {
 355         struct tcf_block *block = chain->block;
 356
 357         ASSERT_BLOCK_LOCKED(block);
 358
 359         list_del(&chain->list);
 360         if (!chain->index)
 361                 block->chain0.chain = NULL;
 362
 363         if (list_empty(&block->chain_list) &&
 364             refcount_read(&block->refcnt) == 0)
 365                 return true;
 366
 367         return false;
 368 }
 369
 370 static void tcf_block_destroy(struct tcf_block *block)
 371 {
 372         mutex_destroy(&block->lock);
 373         kfree_rcu(block, rcu);
 374 }
 375
 376 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 377 {
 378         struct tcf_block *block = chain->block;
 379
 380         mutex_destroy(&chain->filter_chain_lock);
 381         kfree_rcu(chain, rcu);
 382         if (free_block)
 383                 tcf_block_destroy(block);
 384 }
 385
 386 static void tcf_chain_hold(struct tcf_chain *chain)
 387 {
 388         ASSERT_BLOCK_LOCKED(chain->block);
 389
 390         ++chain->refcnt;
 391 }
 392
 393 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 394 {
 395         ASSERT_BLOCK_LOCKED(chain->block);
 396
 397         /* In case all the references are action references, this
 398          * chain should not be shown to the user.
 399          */
 400         return chain->refcnt == chain->action_refcnt;
 401 }
 402
 403 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 404                                           u32 chain_index)
 405 {
 406         struct tcf_chain *chain;
 407
 408         ASSERT_BLOCK_LOCKED(block);
 409
 410         list_for_each_entry(chain, &block->chain_list, list) {
 411                 if (chain->index == chain_index)
 412                         return chain;
 413         }
 414         return NULL;
 415 }
 416
 417 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 418                            u32 seq, u16 flags, int event, bool unicast);
 419
 420 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 421                                          u32 chain_index, bool create,
 422                                          bool by_act)
 423 {
 424         struct tcf_chain *chain = NULL;
 425         bool is_first_reference;
 426
 427         mutex_lock(&block->lock);
 428         chain = tcf_chain_lookup(block, chain_index);
 429         if (chain) {
 430                 tcf_chain_hold(chain);
 431         } else {
 432                 if (!create)
 433                         goto errout;
 434                 chain = tcf_chain_create(block, chain_index);
 435                 if (!chain)
 436                         goto errout;
 437         }
 438
 439         if (by_act)
 440                 ++chain->action_refcnt;
 441         is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 442         mutex_unlock(&block->lock);
 443
 444         /* Send notification only in case we got the first
 445          * non-action reference. Until then, the chain acts only as
 446          * a placeholder for actions pointing to it and user ought
 447          * not know about them.
 448          */
 449         if (is_first_reference && !by_act)
 450                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 451                                 RTM_NEWCHAIN, false);
 452
 453         return chain;
 454
 455 errout:
 456         mutex_unlock(&block->lock);
 457         return chain;
 458 }
 459
 460 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 461                                        bool create)
 462 {
 463         return __tcf_chain_get(block, chain_index, create, false);
 464 }
 465
 466 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 467 {
 468         return __tcf_chain_get(block, chain_index, true, true);
 469 }
 470 EXPORT_SYMBOL(tcf_chain_get_by_act);
 471
 472 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 473                                void *tmplt_priv);
 474 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 475                                   void *tmplt_priv, u32 chain_index,
 476                                   struct tcf_block *block, struct sk_buff *oskb,
 477                                   u32 seq, u16 flags, bool unicast);
 478
 479 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 480                             bool explicitly_created)
 481 {
 482         struct tcf_block *block = chain->block;
 483         const struct tcf_proto_ops *tmplt_ops;
 484         bool free_block = false;
 485         unsigned int refcnt;
 486         void *tmplt_priv;
 487
 488         mutex_lock(&block->lock);
 489         if (explicitly_created) {
 490                 if (!chain->explicitly_created) {
 491                         mutex_unlock(&block->lock);
 492                         return;
 493                 }
 494                 chain->explicitly_created = false;
 495         }
 496
 497         if (by_act)
 498                 chain->action_refcnt--;
 499
 500         /* tc_chain_notify_delete can't be called while holding block lock.
 501          * However, when block is unlocked chain can be changed concurrently, so
 502          * save these to temporary variables.
 503          */
 504         refcnt = --chain->refcnt;
 505         tmplt_ops = chain->tmplt_ops;
 506         tmplt_priv = chain->tmplt_priv;
 507
 508         /* The last dropped non-action reference will trigger notification. */
 509         if (refcnt - chain->action_refcnt == 0 && !by_act) {
 510                 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
 511                                        block, NULL, 0, 0, false);
 512                 /* Last reference to chain, no need to lock. */
 513                 chain->flushing = false;
 514         }
 515
 516         if (refcnt == 0)
 517                 free_block = tcf_chain_detach(chain);
 518         mutex_unlock(&block->lock);
 519
 520         if (refcnt == 0) {
 521                 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 522                 tcf_chain_destroy(chain, free_block);
 523         }
 524 }
 525
 526 static void tcf_chain_put(struct tcf_chain *chain)
 527 {
 528         __tcf_chain_put(chain, false, false);
 529 }
 530
 531 void tcf_chain_put_by_act(struct tcf_chain *chain)
 532 {
 533         __tcf_chain_put(chain, true, false);
 534 }
 535 EXPORT_SYMBOL(tcf_chain_put_by_act);
 536
 537 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 538 {
 539         __tcf_chain_put(chain, false, true);
 540 }
 541
 542 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 543 {
 544         struct tcf_proto *tp, *tp_next;
 545
 546         mutex_lock(&chain->filter_chain_lock);
 547         tp = tcf_chain_dereference(chain->filter_chain, chain);
 548         RCU_INIT_POINTER(chain->filter_chain, NULL);
 549         tcf_chain0_head_change(chain, NULL);
 550         chain->flushing = true;
 551         mutex_unlock(&chain->filter_chain_lock);
 552
 553         while (tp) {
 554                 tp_next = rcu_dereference_protected(tp->next, 1);
 555                 tcf_proto_put(tp, rtnl_held, NULL);
 556                 tp = tp_next;
 557         }
 558 }
 559
 560 static int tcf_block_setup(struct tcf_block *block,
 561                            struct flow_block_offload *bo);
 562
 563 static void tc_indr_block_ing_cmd(struct net_device *dev,
 564                                   struct tcf_block *block,
 565                                   flow_indr_block_bind_cb_t *cb,
 566                                   void *cb_priv,
 567                                   enum flow_block_command command)
 568 {
 569         struct flow_block_offload bo = {
 570                 .command        = command,
 571                 .binder_type    = FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS,
 572                 .net            = dev_net(dev),
 573                 .block_shared   = tcf_block_non_null_shared(block),
 574         };
 575         INIT_LIST_HEAD(&bo.cb_list);
 576
 577         if (!block)
 578                 return;
 579
 580         bo.block = &block->flow_block;
 581
 582         down_write(&block->cb_lock);
 583         cb(dev, cb_priv, TC_SETUP_BLOCK, &bo);
 584
 585         tcf_block_setup(block, &bo);
 586         up_write(&block->cb_lock);
 587 }
 588
 589 static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
 590 {
 591         const struct Qdisc_class_ops *cops;
 592         struct Qdisc *qdisc;
 593
 594         if (!dev_ingress_queue(dev))
 595                 return NULL;
 596
 597         qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
 598         if (!qdisc)
 599                 return NULL;
 600
 601         cops = qdisc->ops->cl_ops;
 602         if (!cops)
 603                 return NULL;
 604
 605         if (!cops->tcf_block)
 606                 return NULL;
 607
 608         return cops->tcf_block(qdisc, TC_H_MIN_INGRESS, NULL);
 609 }
 610
 611 static void tc_indr_block_get_and_ing_cmd(struct net_device *dev,
 612                                           flow_indr_block_bind_cb_t *cb,
 613                                           void *cb_priv,
 614                                           enum flow_block_command command)
 615 {
 616         struct tcf_block *block = tc_dev_ingress_block(dev);
 617
 618         tc_indr_block_ing_cmd(dev, block, cb, cb_priv, command);
 619 }
 620
 621 static void tc_indr_block_call(struct tcf_block *block,
 622                                struct net_device *dev,
 623                                struct tcf_block_ext_info *ei,
 624                                enum flow_block_command command,
 625                                struct netlink_ext_ack *extack)
 626 {
 627         struct flow_block_offload bo = {
 628                 .command        = command,
 629                 .binder_type    = ei->binder_type,
 630                 .net            = dev_net(dev),
 631                 .block          = &block->flow_block,
 632                 .block_shared   = tcf_block_shared(block),
 633                 .extack         = extack,
 634         };
 635         INIT_LIST_HEAD(&bo.cb_list);
 636
 637         flow_indr_block_call(dev, &bo, command);
 638         tcf_block_setup(block, &bo);
 639 }
 640
 641 static bool tcf_block_offload_in_use(struct tcf_block *block)
 642 {
 643         return atomic_read(&block->offloadcnt);
 644 }
 645
 646 static int tcf_block_offload_cmd(struct tcf_block *block,
 647                                  struct net_device *dev,
 648                                  struct tcf_block_ext_info *ei,
 649                                  enum flow_block_command command,
 650                                  struct netlink_ext_ack *extack)
 651 {
 652         struct flow_block_offload bo = {};
 653         int err;
 654
 655         bo.net = dev_net(dev);
 656         bo.command = command;
 657         bo.binder_type = ei->binder_type;
 658         bo.block = &block->flow_block;
 659         bo.block_shared = tcf_block_shared(block);
 660         bo.extack = extack;
 661         INIT_LIST_HEAD(&bo.cb_list);
 662
 663         err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 664         if (err < 0)
 665                 return err;
 666
 667         return tcf_block_setup(block, &bo);
 668 }
 669
 670 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 671                                   struct tcf_block_ext_info *ei,
 672                                   struct netlink_ext_ack *extack)
 673 {
 674         struct net_device *dev = q->dev_queue->dev;
 675         int err;
 676
 677         down_write(&block->cb_lock);
 678         if (!dev->netdev_ops->ndo_setup_tc)
 679                 goto no_offload_dev_inc;
 680
 681         /* If tc offload feature is disabled and the block we try to bind
 682          * to already has some offloaded filters, forbid to bind.
 683          */
 684         if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
 685                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 686                 err = -EOPNOTSUPP;
 687                 goto err_unlock;
 688         }
 689
 690         err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_BIND, extack);
 691         if (err == -EOPNOTSUPP)
 692                 goto no_offload_dev_inc;
 693         if (err)
 694                 goto err_unlock;
 695
 696         tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
 697         up_write(&block->cb_lock);
 698         return 0;
 699
 700 no_offload_dev_inc:
 701         if (tcf_block_offload_in_use(block)) {
 702                 err = -EOPNOTSUPP;
 703                 goto err_unlock;
 704         }
 705         err = 0;
 706         block->nooffloaddevcnt++;
 707         tc_indr_block_call(block, dev, ei, FLOW_BLOCK_BIND, extack);
 708 err_unlock:
 709         up_write(&block->cb_lock);
 710         return err;
 711 }
 712
 713 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 714                                      struct tcf_block_ext_info *ei)
 715 {
 716         struct net_device *dev = q->dev_queue->dev;
 717         int err;
 718
 719         down_write(&block->cb_lock);
 720         tc_indr_block_call(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
 721
 722         if (!dev->netdev_ops->ndo_setup_tc)
 723                 goto no_offload_dev_dec;
 724         err = tcf_block_offload_cmd(block, dev, ei, FLOW_BLOCK_UNBIND, NULL);
 725         if (err == -EOPNOTSUPP)
 726                 goto no_offload_dev_dec;
 727         up_write(&block->cb_lock);
 728         return;
 729
 730 no_offload_dev_dec:
 731         WARN_ON(block->nooffloaddevcnt-- == 0);
 732         up_write(&block->cb_lock);
 733 }
 734
 735 static int
 736 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 737                               struct tcf_block_ext_info *ei,
 738                               struct netlink_ext_ack *extack)
 739 {
 740         struct tcf_filter_chain_list_item *item;
 741         struct tcf_chain *chain0;
 742
 743         item = kmalloc(sizeof(*item), GFP_KERNEL);
 744         if (!item) {
 745                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 746                 return -ENOMEM;
 747         }
 748         item->chain_head_change = ei->chain_head_change;
 749         item->chain_head_change_priv = ei->chain_head_change_priv;
 750
 751         mutex_lock(&block->lock);
 752         chain0 = block->chain0.chain;
 753         if (chain0)
 754                 tcf_chain_hold(chain0);
 755         else
 756                 list_add(&item->list, &block->chain0.filter_chain_list);
 757         mutex_unlock(&block->lock);
 758
 759         if (chain0) {
 760                 struct tcf_proto *tp_head;
 761
 762                 mutex_lock(&chain0->filter_chain_lock);
 763
 764                 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 765                 if (tp_head)
 766                         tcf_chain_head_change_item(item, tp_head);
 767
 768                 mutex_lock(&block->lock);
 769                 list_add(&item->list, &block->chain0.filter_chain_list);
 770                 mutex_unlock(&block->lock);
 771
 772                 mutex_unlock(&chain0->filter_chain_lock);
 773                 tcf_chain_put(chain0);
 774         }
 775
 776         return 0;
 777 }
 778
 779 static void
 780 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 781                               struct tcf_block_ext_info *ei)
 782 {
 783         struct tcf_filter_chain_list_item *item;
 784
 785         mutex_lock(&block->lock);
 786         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 787                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 788                     (item->chain_head_change == ei->chain_head_change &&
 789                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 790                         if (block->chain0.chain)
 791                                 tcf_chain_head_change_item(item, NULL);
 792                         list_del(&item->list);
 793                         mutex_unlock(&block->lock);
 794
 795                         kfree(item);
 796                         return;
 797                 }
 798         }
 799         mutex_unlock(&block->lock);
 800         WARN_ON(1);
 801 }
 802
 803 struct tcf_net {
 804         spinlock_t idr_lock; /* Protects idr */
 805         struct idr idr;
 806 };
 807
 808 static unsigned int tcf_net_id;
 809
 810 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 811                             struct netlink_ext_ack *extack)
 812 {
 813         struct tcf_net *tn = net_generic(net, tcf_net_id);
 814         int err;
 815
 816         idr_preload(GFP_KERNEL);
 817         spin_lock(&tn->idr_lock);
 818         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 819                             GFP_NOWAIT);
 820         spin_unlock(&tn->idr_lock);
 821         idr_preload_end();
 822
 823         return err;
 824 }
 825
 826 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 827 {
 828         struct tcf_net *tn = net_generic(net, tcf_net_id);
 829
 830         spin_lock(&tn->idr_lock);
 831         idr_remove(&tn->idr, block->index);
 832         spin_unlock(&tn->idr_lock);
 833 }
 834
 835 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 836                                           u32 block_index,
 837                                           struct netlink_ext_ack *extack)
 838 {
 839         struct tcf_block *block;
 840
 841         block = kzalloc(sizeof(*block), GFP_KERNEL);
 842         if (!block) {
 843                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 844                 return ERR_PTR(-ENOMEM);
 845         }
 846         mutex_init(&block->lock);
 847         init_rwsem(&block->cb_lock);
 848         flow_block_init(&block->flow_block);
 849         INIT_LIST_HEAD(&block->chain_list);
 850         INIT_LIST_HEAD(&block->owner_list);
 851         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 852
 853         refcount_set(&block->refcnt, 1);
 854         block->net = net;
 855         block->index = block_index;
 856
 857         /* Don't store q pointer for blocks which are shared */
 858         if (!tcf_block_shared(block))
 859                 block->q = q;
 860         return block;
 861 }
 862
 863 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 864 {
 865         struct tcf_net *tn = net_generic(net, tcf_net_id);
 866
 867         return idr_find(&tn->idr, block_index);
 868 }
 869
 870 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
 871 {
 872         struct tcf_block *block;
 873
 874         rcu_read_lock();
 875         block = tcf_block_lookup(net, block_index);
 876         if (block && !refcount_inc_not_zero(&block->refcnt))
 877                 block = NULL;
 878         rcu_read_unlock();
 879
 880         return block;
 881 }
 882
 883 static struct tcf_chain *
 884 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
 885 {
 886         mutex_lock(&block->lock);
 887         if (chain)
 888                 chain = list_is_last(&chain->list, &block->chain_list) ?
 889                         NULL : list_next_entry(chain, list);
 890         else
 891                 chain = list_first_entry_or_null(&block->chain_list,
 892                                                  struct tcf_chain, list);
 893
 894         /* skip all action-only chains */
 895         while (chain && tcf_chain_held_by_acts_only(chain))
 896                 chain = list_is_last(&chain->list, &block->chain_list) ?
 897                         NULL : list_next_entry(chain, list);
 898
 899         if (chain)
 900                 tcf_chain_hold(chain);
 901         mutex_unlock(&block->lock);
 902
 903         return chain;
 904 }
 905
 906 /* Function to be used by all clients that want to iterate over all chains on
 907  * block. It properly obtains block->lock and takes reference to chain before
 908  * returning it. Users of this function must be tolerant to concurrent chain
 909  * insertion/deletion or ensure that no concurrent chain modification is
 910  * possible. Note that all netlink dump callbacks cannot guarantee to provide
 911  * consistent dump because rtnl lock is released each time skb is filled with
 912  * data and sent to user-space.
 913  */
 914
 915 struct tcf_chain *
 916 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
 917 {
 918         struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
 919
 920         if (chain)
 921                 tcf_chain_put(chain);
 922
 923         return chain_next;
 924 }
 925 EXPORT_SYMBOL(tcf_get_next_chain);
 926
 927 static struct tcf_proto *
 928 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
 929 {
 930         u32 prio = 0;
 931
 932         ASSERT_RTNL();
 933         mutex_lock(&chain->filter_chain_lock);
 934
 935         if (!tp) {
 936                 tp = tcf_chain_dereference(chain->filter_chain, chain);
 937         } else if (tcf_proto_is_deleting(tp)) {
 938                 /* 'deleting' flag is set and chain->filter_chain_lock was
 939                  * unlocked, which means next pointer could be invalid. Restart
 940                  * search.
 941                  */
 942                 prio = tp->prio + 1;
 943                 tp = tcf_chain_dereference(chain->filter_chain, chain);
 944
 945                 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
 946                         if (!tp->deleting && tp->prio >= prio)
 947                                 break;
 948         } else {
 949                 tp = tcf_chain_dereference(tp->next, chain);
 950         }
 951
 952         if (tp)
 953                 tcf_proto_get(tp);
 954
 955         mutex_unlock(&chain->filter_chain_lock);
 956
 957         return tp;
 958 }
 959
 960 /* Function to be used by all clients that want to iterate over all tp's on
 961  * chain. Users of this function must be tolerant to concurrent tp
 962  * insertion/deletion or ensure that no concurrent chain modification is
 963  * possible. Note that all netlink dump callbacks cannot guarantee to provide
 964  * consistent dump because rtnl lock is released each time skb is filled with
 965  * data and sent to user-space.
 966  */
 967
 968 struct tcf_proto *
 969 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
 970                    bool rtnl_held)
 971 {
 972         struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
 973
 974         if (tp)
 975                 tcf_proto_put(tp, rtnl_held, NULL);
 976
 977         return tp_next;
 978 }
 979 EXPORT_SYMBOL(tcf_get_next_proto);
 980
 981 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
 982 {
 983         struct tcf_chain *chain;
 984
 985         /* Last reference to block. At this point chains cannot be added or
 986          * removed concurrently.
 987          */
 988         for (chain = tcf_get_next_chain(block, NULL);
 989              chain;
 990              chain = tcf_get_next_chain(block, chain)) {
 991                 tcf_chain_put_explicitly_created(chain);
 992                 tcf_chain_flush(chain, rtnl_held);
 993         }
 994 }
 995
 996 /* Lookup Qdisc and increments its reference counter.
 997  * Set parent, if necessary.
 998  */
 999
1000 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1001                             u32 *parent, int ifindex, bool rtnl_held,
1002                             struct netlink_ext_ack *extack)
1003 {
1004         const struct Qdisc_class_ops *cops;
1005         struct net_device *dev;
1006         int err = 0;
1007
1008         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1009                 return 0;
1010
1011         rcu_read_lock();
1012
1013         /* Find link */
1014         dev = dev_get_by_index_rcu(net, ifindex);
1015         if (!dev) {
1016                 rcu_read_unlock();
1017                 return -ENODEV;
1018         }
1019
1020         /* Find qdisc */
1021         if (!*parent) {
1022                 *q = dev->qdisc;
1023                 *parent = (*q)->handle;
1024         } else {
1025                 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1026                 if (!*q) {
1027                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1028                         err = -EINVAL;
1029                         goto errout_rcu;
1030                 }
1031         }
1032
1033         *q = qdisc_refcount_inc_nz(*q);
1034         if (!*q) {
1035                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1036                 err = -EINVAL;
1037                 goto errout_rcu;
1038         }
1039
1040         /* Is it classful? */
1041         cops = (*q)->ops->cl_ops;
1042         if (!cops) {
1043                 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1044                 err = -EINVAL;
1045                 goto errout_qdisc;
1046         }
1047
1048         if (!cops->tcf_block) {
1049                 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1050                 err = -EOPNOTSUPP;
1051                 goto errout_qdisc;
1052         }
1053
1054 errout_rcu:
1055         /* At this point we know that qdisc is not noop_qdisc,
1056          * which means that qdisc holds a reference to net_device
1057          * and we hold a reference to qdisc, so it is safe to release
1058          * rcu read lock.
1059          */
1060         rcu_read_unlock();
1061         return err;
1062
1063 errout_qdisc:
1064         rcu_read_unlock();
1065
1066         if (rtnl_held)
1067                 qdisc_put(*q);
1068         else
1069                 qdisc_put_unlocked(*q);
1070         *q = NULL;
1071
1072         return err;
1073 }
1074
1075 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1076                                int ifindex, struct netlink_ext_ack *extack)
1077 {
1078         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1079                 return 0;
1080
1081         /* Do we search for filter, attached to class? */
1082         if (TC_H_MIN(parent)) {
1083                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1084
1085                 *cl = cops->find(q, parent);
1086                 if (*cl == 0) {
1087                         NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1088                         return -ENOENT;
1089                 }
1090         }
1091
1092         return 0;
1093 }
1094
1095 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1096                                           unsigned long cl, int ifindex,
1097                                           u32 block_index,
1098                                           struct netlink_ext_ack *extack)
1099 {
1100         struct tcf_block *block;
1101
1102         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1103                 block = tcf_block_refcnt_get(net, block_index);
1104                 if (!block) {
1105                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
1106                         return ERR_PTR(-EINVAL);
1107                 }
1108         } else {
1109                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1110
1111                 block = cops->tcf_block(q, cl, extack);
1112                 if (!block)
1113                         return ERR_PTR(-EINVAL);
1114
1115                 if (tcf_block_shared(block)) {
1116                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1117                         return ERR_PTR(-EOPNOTSUPP);
1118                 }
1119
1120                 /* Always take reference to block in order to support execution
1121                  * of rules update path of cls API without rtnl lock. Caller
1122                  * must release block when it is finished using it. 'if' block
1123                  * of this conditional obtain reference to block by calling
1124                  * tcf_block_refcnt_get().
1125                  */
1126                 refcount_inc(&block->refcnt);
1127         }
1128
1129         return block;
1130 }
1131
1132 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1133                             struct tcf_block_ext_info *ei, bool rtnl_held)
1134 {
1135         if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1136                 /* Flushing/putting all chains will cause the block to be
1137                  * deallocated when last chain is freed. However, if chain_list
1138                  * is empty, block has to be manually deallocated. After block
1139                  * reference counter reached 0, it is no longer possible to
1140                  * increment it or add new chains to block.
1141                  */
1142                 bool free_block = list_empty(&block->chain_list);
1143
1144                 mutex_unlock(&block->lock);
1145                 if (tcf_block_shared(block))
1146                         tcf_block_remove(block, block->net);
1147
1148                 if (q)
1149                         tcf_block_offload_unbind(block, q, ei);
1150
1151                 if (free_block)
1152                         tcf_block_destroy(block);
1153                 else
1154                         tcf_block_flush_all_chains(block, rtnl_held);
1155         } else if (q) {
1156                 tcf_block_offload_unbind(block, q, ei);
1157         }
1158 }
1159
1160 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1161 {
1162         __tcf_block_put(block, NULL, NULL, rtnl_held);
1163 }
1164
1165 /* Find tcf block.
1166  * Set q, parent, cl when appropriate.
1167  */
1168
1169 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1170                                         u32 *parent, unsigned long *cl,
1171                                         int ifindex, u32 block_index,
1172                                         struct netlink_ext_ack *extack)
1173 {
1174         struct tcf_block *block;
1175         int err = 0;
1176
1177         ASSERT_RTNL();
1178
1179         err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1180         if (err)
1181                 goto errout;
1182
1183         err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1184         if (err)
1185                 goto errout_qdisc;
1186
1187         block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1188         if (IS_ERR(block)) {
1189                 err = PTR_ERR(block);
1190                 goto errout_qdisc;
1191         }
1192
1193         return block;
1194
1195 errout_qdisc:
1196         if (*q)
1197                 qdisc_put(*q);
1198 errout:
1199         *q = NULL;
1200         return ERR_PTR(err);
1201 }
1202
1203 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1204                               bool rtnl_held)
1205 {
1206         if (!IS_ERR_OR_NULL(block))
1207                 tcf_block_refcnt_put(block, rtnl_held);
1208
1209         if (q) {
1210                 if (rtnl_held)
1211                         qdisc_put(q);
1212                 else
1213                         qdisc_put_unlocked(q);
1214         }
1215 }
1216
1217 struct tcf_block_owner_item {
1218         struct list_head list;
1219         struct Qdisc *q;
1220         enum flow_block_binder_type binder_type;
1221 };
1222
1223 static void
1224 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1225                                struct Qdisc *q,
1226                                enum flow_block_binder_type binder_type)
1227 {
1228         if (block->keep_dst &&
1229             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1230             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1231                 netif_keep_dst(qdisc_dev(q));
1232 }
1233
1234 void tcf_block_netif_keep_dst(struct tcf_block *block)
1235 {
1236         struct tcf_block_owner_item *item;
1237
1238         block->keep_dst = true;
1239         list_for_each_entry(item, &block->owner_list, list)
1240                 tcf_block_owner_netif_keep_dst(block, item->q,
1241                                                item->binder_type);
1242 }
1243 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1244
1245 static int tcf_block_owner_add(struct tcf_block *block,
1246                                struct Qdisc *q,
1247                                enum flow_block_binder_type binder_type)
1248 {
1249         struct tcf_block_owner_item *item;
1250
1251         item = kmalloc(sizeof(*item), GFP_KERNEL);
1252         if (!item)
1253                 return -ENOMEM;
1254         item->q = q;
1255         item->binder_type = binder_type;
1256         list_add(&item->list, &block->owner_list);
1257         return 0;
1258 }
1259
1260 static void tcf_block_owner_del(struct tcf_block *block,
1261                                 struct Qdisc *q,
1262                                 enum flow_block_binder_type binder_type)
1263 {
1264         struct tcf_block_owner_item *item;
1265
1266         list_for_each_entry(item, &block->owner_list, list) {
1267                 if (item->q == q && item->binder_type == binder_type) {
1268                         list_del(&item->list);
1269                         kfree(item);
1270                         return;
1271                 }
1272         }
1273         WARN_ON(1);
1274 }
1275
1276 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1277                       struct tcf_block_ext_info *ei,
1278                       struct netlink_ext_ack *extack)
1279 {
1280         struct net *net = qdisc_net(q);
1281         struct tcf_block *block = NULL;
1282         int err;
1283
1284         if (ei->block_index)
1285                 /* block_index not 0 means the shared block is requested */
1286                 block = tcf_block_refcnt_get(net, ei->block_index);
1287
1288         if (!block) {
1289                 block = tcf_block_create(net, q, ei->block_index, extack);
1290                 if (IS_ERR(block))
1291                         return PTR_ERR(block);
1292                 if (tcf_block_shared(block)) {
1293                         err = tcf_block_insert(block, net, extack);
1294                         if (err)
1295                                 goto err_block_insert;
1296                 }
1297         }
1298
1299         err = tcf_block_owner_add(block, q, ei->binder_type);
1300         if (err)
1301                 goto err_block_owner_add;
1302
1303         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1304
1305         err = tcf_chain0_head_change_cb_add(block, ei, extack);
1306         if (err)
1307                 goto err_chain0_head_change_cb_add;
1308
1309         err = tcf_block_offload_bind(block, q, ei, extack);
1310         if (err)
1311                 goto err_block_offload_bind;
1312
1313         *p_block = block;
1314         return 0;
1315
1316 err_block_offload_bind:
1317         tcf_chain0_head_change_cb_del(block, ei);
1318 err_chain0_head_change_cb_add:
1319         tcf_block_owner_del(block, q, ei->binder_type);
1320 err_block_owner_add:
1321 err_block_insert:
1322         tcf_block_refcnt_put(block, true);
1323         return err;
1324 }
1325 EXPORT_SYMBOL(tcf_block_get_ext);
1326
1327 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1328 {
1329         struct tcf_proto __rcu **p_filter_chain = priv;
1330
1331         rcu_assign_pointer(*p_filter_chain, tp_head);
1332 }
1333
1334 int tcf_block_get(struct tcf_block **p_block,
1335                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1336                   struct netlink_ext_ack *extack)
1337 {
1338         struct tcf_block_ext_info ei = {
1339                 .chain_head_change = tcf_chain_head_change_dflt,
1340                 .chain_head_change_priv = p_filter_chain,
1341         };
1342
1343         WARN_ON(!p_filter_chain);
1344         return tcf_block_get_ext(p_block, q, &ei, extack);
1345 }
1346 EXPORT_SYMBOL(tcf_block_get);
1347
1348 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1349  * actions should be all removed after flushing.
1350  */
1351 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1352                        struct tcf_block_ext_info *ei)
1353 {
1354         if (!block)
1355                 return;
1356         tcf_chain0_head_change_cb_del(block, ei);
1357         tcf_block_owner_del(block, q, ei->binder_type);
1358
1359         __tcf_block_put(block, q, ei, true);
1360 }
1361 EXPORT_SYMBOL(tcf_block_put_ext);
1362
1363 void tcf_block_put(struct tcf_block *block)
1364 {
1365         struct tcf_block_ext_info ei = {0, };
1366
1367         if (!block)
1368                 return;
1369         tcf_block_put_ext(block, block->q, &ei);
1370 }
1371
1372 EXPORT_SYMBOL(tcf_block_put);
1373
1374 static int
1375 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1376                             void *cb_priv, bool add, bool offload_in_use,
1377                             struct netlink_ext_ack *extack)
1378 {
1379         struct tcf_chain *chain, *chain_prev;
1380         struct tcf_proto *tp, *tp_prev;
1381         int err;
1382
1383         lockdep_assert_held(&block->cb_lock);
1384
1385         for (chain = __tcf_get_next_chain(block, NULL);
1386              chain;
1387              chain_prev = chain,
1388                      chain = __tcf_get_next_chain(block, chain),
1389                      tcf_chain_put(chain_prev)) {
1390                 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1391                      tp_prev = tp,
1392                              tp = __tcf_get_next_proto(chain, tp),
1393                              tcf_proto_put(tp_prev, true, NULL)) {
1394                         if (tp->ops->reoffload) {
1395                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1396                                                          extack);
1397                                 if (err && add)
1398                                         goto err_playback_remove;
1399                         } else if (add && offload_in_use) {
1400                                 err = -EOPNOTSUPP;
1401                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1402                                 goto err_playback_remove;
1403                         }
1404                 }
1405         }
1406
1407         return 0;
1408
1409 err_playback_remove:
1410         tcf_proto_put(tp, true, NULL);
1411         tcf_chain_put(chain);
1412         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1413                                     extack);
1414         return err;
1415 }
1416
1417 static int tcf_block_bind(struct tcf_block *block,
1418                           struct flow_block_offload *bo)
1419 {
1420         struct flow_block_cb *block_cb, *next;
1421         int err, i = 0;
1422
1423         lockdep_assert_held(&block->cb_lock);
1424
1425         list_for_each_entry(block_cb, &bo->cb_list, list) {
1426                 err = tcf_block_playback_offloads(block, block_cb->cb,
1427                                                   block_cb->cb_priv, true,
1428                                                   tcf_block_offload_in_use(block),
1429                                                   bo->extack);
1430                 if (err)
1431                         goto err_unroll;
1432                 if (!bo->unlocked_driver_cb)
1433                         block->lockeddevcnt++;
1434
1435                 i++;
1436         }
1437         list_splice(&bo->cb_list, &block->flow_block.cb_list);
1438
1439         return 0;
1440
1441 err_unroll:
1442         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1443                 if (i-- > 0) {
1444                         list_del(&block_cb->list);
1445                         tcf_block_playback_offloads(block, block_cb->cb,
1446                                                     block_cb->cb_priv, false,
1447                                                     tcf_block_offload_in_use(block),
1448                                                     NULL);
1449                         if (!bo->unlocked_driver_cb)
1450                                 block->lockeddevcnt--;
1451                 }
1452                 flow_block_cb_free(block_cb);
1453         }
1454
1455         return err;
1456 }
1457
1458 static void tcf_block_unbind(struct tcf_block *block,
1459                              struct flow_block_offload *bo)
1460 {
1461         struct flow_block_cb *block_cb, *next;
1462
1463         lockdep_assert_held(&block->cb_lock);
1464
1465         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1466                 tcf_block_playback_offloads(block, block_cb->cb,
1467                                             block_cb->cb_priv, false,
1468                                             tcf_block_offload_in_use(block),
1469                                             NULL);
1470                 list_del(&block_cb->list);
1471                 flow_block_cb_free(block_cb);
1472                 if (!bo->unlocked_driver_cb)
1473                         block->lockeddevcnt--;
1474         }
1475 }
1476
1477 static int tcf_block_setup(struct tcf_block *block,
1478                            struct flow_block_offload *bo)
1479 {
1480         int err;
1481
1482         switch (bo->command) {
1483         case FLOW_BLOCK_BIND:
1484                 err = tcf_block_bind(block, bo);
1485                 break;
1486         case FLOW_BLOCK_UNBIND:
1487                 err = 0;
1488                 tcf_block_unbind(block, bo);
1489                 break;
1490         default:
1491                 WARN_ON_ONCE(1);
1492                 err = -EOPNOTSUPP;
1493         }
1494
1495         return err;
1496 }
1497
1498 /* Main classifier routine: scans classifier chain attached
1499  * to this qdisc, (optionally) tests for protocol and asks
1500  * specific classifiers.
1501  */
1502 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1503                  struct tcf_result *res, bool compat_mode)
1504 {
1505 #ifdef CONFIG_NET_CLS_ACT
1506         const int max_reclassify_loop = 4;
1507         const struct tcf_proto *orig_tp = tp;
1508         const struct tcf_proto *first_tp;
1509         int limit = 0;
1510
1511 reclassify:
1512 #endif
1513         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1514                 __be16 protocol = tc_skb_protocol(skb);
1515                 int err;
1516
1517                 if (tp->protocol != protocol &&
1518                     tp->protocol != htons(ETH_P_ALL))
1519                         continue;
1520
1521                 err = tp->classify(skb, tp, res);
1522 #ifdef CONFIG_NET_CLS_ACT
1523                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1524                         first_tp = orig_tp;
1525                         goto reset;
1526                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1527                         first_tp = res->goto_tp;
1528
1529 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1530                         {
1531                                 struct tc_skb_ext *ext;
1532
1533                                 ext = skb_ext_add(skb, TC_SKB_EXT);
1534                                 if (WARN_ON_ONCE(!ext))
1535                                         return TC_ACT_SHOT;
1536
1537                                 ext->chain = err & TC_ACT_EXT_VAL_MASK;
1538                         }
1539 #endif
1540                         goto reset;
1541                 }
1542 #endif
1543                 if (err >= 0)
1544                         return err;
1545         }
1546
1547         return TC_ACT_UNSPEC; /* signal: continue lookup */
1548 #ifdef CONFIG_NET_CLS_ACT
1549 reset:
1550         if (unlikely(limit++ >= max_reclassify_loop)) {
1551                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1552                                        tp->chain->block->index,
1553                                        tp->prio & 0xffff,
1554                                        ntohs(tp->protocol));
1555                 return TC_ACT_SHOT;
1556         }
1557
1558         tp = first_tp;
1559         goto reclassify;
1560 #endif
1561 }
1562 EXPORT_SYMBOL(tcf_classify);
1563
1564 struct tcf_chain_info {
1565         struct tcf_proto __rcu **pprev;
1566         struct tcf_proto __rcu *next;
1567 };
1568
1569 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1570                                            struct tcf_chain_info *chain_info)
1571 {
1572         return tcf_chain_dereference(*chain_info->pprev, chain);
1573 }
1574
1575 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1576                                struct tcf_chain_info *chain_info,
1577                                struct tcf_proto *tp)
1578 {
1579         if (chain->flushing)
1580                 return -EAGAIN;
1581
1582         if (*chain_info->pprev == chain->filter_chain)
1583                 tcf_chain0_head_change(chain, tp);
1584         tcf_proto_get(tp);
1585         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1586         rcu_assign_pointer(*chain_info->pprev, tp);
1587
1588         return 0;
1589 }
1590
1591 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1592                                 struct tcf_chain_info *chain_info,
1593                                 struct tcf_proto *tp)
1594 {
1595         struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1596
1597         tcf_proto_mark_delete(tp);
1598         if (tp == chain->filter_chain)
1599                 tcf_chain0_head_change(chain, next);
1600         RCU_INIT_POINTER(*chain_info->pprev, next);
1601 }
1602
1603 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1604                                            struct tcf_chain_info *chain_info,
1605                                            u32 protocol, u32 prio,
1606                                            bool prio_allocate);
1607
1608 /* Try to insert new proto.
1609  * If proto with specified priority already exists, free new proto
1610  * and return existing one.
1611  */
1612
1613 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1614                                                     struct tcf_proto *tp_new,
1615                                                     u32 protocol, u32 prio,
1616                                                     bool rtnl_held)
1617 {
1618         struct tcf_chain_info chain_info;
1619         struct tcf_proto *tp;
1620         int err = 0;
1621
1622         mutex_lock(&chain->filter_chain_lock);
1623
1624         tp = tcf_chain_tp_find(chain, &chain_info,
1625                                protocol, prio, false);
1626         if (!tp)
1627                 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1628         mutex_unlock(&chain->filter_chain_lock);
1629
1630         if (tp) {
1631                 tcf_proto_destroy(tp_new, rtnl_held, NULL);
1632                 tp_new = tp;
1633         } else if (err) {
1634                 tcf_proto_destroy(tp_new, rtnl_held, NULL);
1635                 tp_new = ERR_PTR(err);
1636         }
1637
1638         return tp_new;
1639 }
1640
1641 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1642                                       struct tcf_proto *tp, bool rtnl_held,
1643                                       struct netlink_ext_ack *extack)
1644 {
1645         struct tcf_chain_info chain_info;
1646         struct tcf_proto *tp_iter;
1647         struct tcf_proto **pprev;
1648         struct tcf_proto *next;
1649
1650         mutex_lock(&chain->filter_chain_lock);
1651
1652         /* Atomically find and remove tp from chain. */
1653         for (pprev = &chain->filter_chain;
1654              (tp_iter = tcf_chain_dereference(*pprev, chain));
1655              pprev = &tp_iter->next) {
1656                 if (tp_iter == tp) {
1657                         chain_info.pprev = pprev;
1658                         chain_info.next = tp_iter->next;
1659                         WARN_ON(tp_iter->deleting);
1660                         break;
1661                 }
1662         }
1663         /* Verify that tp still exists and no new filters were inserted
1664          * concurrently.
1665          * Mark tp for deletion if it is empty.
1666          */
1667         if (!tp_iter || !tcf_proto_check_delete(tp, rtnl_held)) {
1668                 mutex_unlock(&chain->filter_chain_lock);
1669                 return;
1670         }
1671
1672         next = tcf_chain_dereference(chain_info.next, chain);
1673         if (tp == chain->filter_chain)
1674                 tcf_chain0_head_change(chain, next);
1675         RCU_INIT_POINTER(*chain_info.pprev, next);
1676         mutex_unlock(&chain->filter_chain_lock);
1677
1678         tcf_proto_put(tp, rtnl_held, extack);
1679 }
1680
1681 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1682                                            struct tcf_chain_info *chain_info,
1683                                            u32 protocol, u32 prio,
1684                                            bool prio_allocate)
1685 {
1686         struct tcf_proto **pprev;
1687         struct tcf_proto *tp;
1688
1689         /* Check the chain for existence of proto-tcf with this priority */
1690         for (pprev = &chain->filter_chain;
1691              (tp = tcf_chain_dereference(*pprev, chain));
1692              pprev = &tp->next) {
1693                 if (tp->prio >= prio) {
1694                         if (tp->prio == prio) {
1695                                 if (prio_allocate ||
1696                                     (tp->protocol != protocol && protocol))
1697                                         return ERR_PTR(-EINVAL);
1698                         } else {
1699                                 tp = NULL;
1700                         }
1701                         break;
1702                 }
1703         }
1704         chain_info->pprev = pprev;
1705         if (tp) {
1706                 chain_info->next = tp->next;
1707                 tcf_proto_get(tp);
1708         } else {
1709                 chain_info->next = NULL;
1710         }
1711         return tp;
1712 }
1713
1714 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1715                          struct tcf_proto *tp, struct tcf_block *block,
1716                          struct Qdisc *q, u32 parent, void *fh,
1717                          u32 portid, u32 seq, u16 flags, int event,
1718                          bool rtnl_held)
1719 {
1720         struct tcmsg *tcm;
1721         struct nlmsghdr  *nlh;
1722         unsigned char *b = skb_tail_pointer(skb);
1723
1724         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1725         if (!nlh)
1726                 goto out_nlmsg_trim;
1727         tcm = nlmsg_data(nlh);
1728         tcm->tcm_family = AF_UNSPEC;
1729         tcm->tcm__pad1 = 0;
1730         tcm->tcm__pad2 = 0;
1731         if (q) {
1732                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1733                 tcm->tcm_parent = parent;
1734         } else {
1735                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1736                 tcm->tcm_block_index = block->index;
1737         }
1738         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1739         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1740                 goto nla_put_failure;
1741         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1742                 goto nla_put_failure;
1743         if (!fh) {
1744                 tcm->tcm_handle = 0;
1745         } else {
1746                 if (tp->ops->dump &&
1747                     tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1748                         goto nla_put_failure;
1749         }
1750         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1751         return skb->len;
1752
1753 out_nlmsg_trim:
1754 nla_put_failure:
1755         nlmsg_trim(skb, b);
1756         return -1;
1757 }
1758
1759 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1760                           struct nlmsghdr *n, struct tcf_proto *tp,
1761                           struct tcf_block *block, struct Qdisc *q,
1762                           u32 parent, void *fh, int event, bool unicast,
1763                           bool rtnl_held)
1764 {
1765         struct sk_buff *skb;
1766         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1767         int err = 0;
1768
1769         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1770         if (!skb)
1771                 return -ENOBUFS;
1772
1773         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1774                           n->nlmsg_seq, n->nlmsg_flags, event,
1775                           rtnl_held) <= 0) {
1776                 kfree_skb(skb);
1777                 return -EINVAL;
1778         }
1779
1780         if (unicast)
1781                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1782         else
1783                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1784                                      n->nlmsg_flags & NLM_F_ECHO);
1785
1786         if (err > 0)
1787                 err = 0;
1788         return err;
1789 }
1790
1791 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1792                               struct nlmsghdr *n, struct tcf_proto *tp,
1793                               struct tcf_block *block, struct Qdisc *q,
1794                               u32 parent, void *fh, bool unicast, bool *last,
1795                               bool rtnl_held, struct netlink_ext_ack *extack)
1796 {
1797         struct sk_buff *skb;
1798         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1799         int err;
1800
1801         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1802         if (!skb)
1803                 return -ENOBUFS;
1804
1805         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1806                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1807                           rtnl_held) <= 0) {
1808                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1809                 kfree_skb(skb);
1810                 return -EINVAL;
1811         }
1812
1813         err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1814         if (err) {
1815                 kfree_skb(skb);
1816                 return err;
1817         }
1818
1819         if (unicast)
1820                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1821         else
1822                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1823                                      n->nlmsg_flags & NLM_F_ECHO);
1824         if (err < 0)
1825                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1826
1827         if (err > 0)
1828                 err = 0;
1829         return err;
1830 }
1831
1832 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1833                                  struct tcf_block *block, struct Qdisc *q,
1834                                  u32 parent, struct nlmsghdr *n,
1835                                  struct tcf_chain *chain, int event,
1836                                  bool rtnl_held)
1837 {
1838         struct tcf_proto *tp;
1839
1840         for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1841              tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
1842                 tfilter_notify(net, oskb, n, tp, block,
1843                                q, parent, NULL, event, false, rtnl_held);
1844 }
1845
1846 static void tfilter_put(struct tcf_proto *tp, void *fh)
1847 {
1848         if (tp->ops->put && fh)
1849                 tp->ops->put(tp, fh);
1850 }
1851
1852 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1853                           struct netlink_ext_ack *extack)
1854 {
1855         struct net *net = sock_net(skb->sk);
1856         struct nlattr *tca[TCA_MAX + 1];
1857         char name[IFNAMSIZ];
1858         struct tcmsg *t;
1859         u32 protocol;
1860         u32 prio;
1861         bool prio_allocate;
1862         u32 parent;
1863         u32 chain_index;
1864         struct Qdisc *q = NULL;
1865         struct tcf_chain_info chain_info;
1866         struct tcf_chain *chain = NULL;
1867         struct tcf_block *block;
1868         struct tcf_proto *tp;
1869         unsigned long cl;
1870         void *fh;
1871         int err;
1872         int tp_created;
1873         bool rtnl_held = false;
1874
1875         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1876                 return -EPERM;
1877
1878 replay:
1879         tp_created = 0;
1880
1881         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1882                                      rtm_tca_policy, extack);
1883         if (err < 0)
1884                 return err;
1885
1886         t = nlmsg_data(n);
1887         protocol = TC_H_MIN(t->tcm_info);
1888         prio = TC_H_MAJ(t->tcm_info);
1889         prio_allocate = false;
1890         parent = t->tcm_parent;
1891         tp = NULL;
1892         cl = 0;
1893         block = NULL;
1894
1895         if (prio == 0) {
1896                 /* If no priority is provided by the user,
1897                  * we allocate one.
1898                  */
1899                 if (n->nlmsg_flags & NLM_F_CREATE) {
1900                         prio = TC_H_MAKE(0x80000000U, 0U);
1901                         prio_allocate = true;
1902                 } else {
1903                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1904                         return -ENOENT;
1905                 }
1906         }
1907
1908         /* Find head of filter chain. */
1909
1910         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
1911         if (err)
1912                 return err;
1913
1914         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
1915                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
1916                 err = -EINVAL;
1917                 goto errout;
1918         }
1919
1920         /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
1921          * block is shared (no qdisc found), qdisc is not unlocked, classifier
1922          * type is not specified, classifier is not unlocked.
1923          */
1924         if (rtnl_held ||
1925             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
1926             !tcf_proto_is_unlocked(name)) {
1927                 rtnl_held = true;
1928                 rtnl_lock();
1929         }
1930
1931         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
1932         if (err)
1933                 goto errout;
1934
1935         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
1936                                  extack);
1937         if (IS_ERR(block)) {
1938                 err = PTR_ERR(block);
1939                 goto errout;
1940         }
1941
1942         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1943         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1944                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1945                 err = -EINVAL;
1946                 goto errout;
1947         }
1948         chain = tcf_chain_get(block, chain_index, true);
1949         if (!chain) {
1950                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
1951                 err = -ENOMEM;
1952                 goto errout;
1953         }
1954
1955         mutex_lock(&chain->filter_chain_lock);
1956         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1957                                prio, prio_allocate);
1958         if (IS_ERR(tp)) {
1959                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1960                 err = PTR_ERR(tp);
1961                 goto errout_locked;
1962         }
1963
1964         if (tp == NULL) {
1965                 struct tcf_proto *tp_new = NULL;
1966
1967                 if (chain->flushing) {
1968                         err = -EAGAIN;
1969                         goto errout_locked;
1970                 }
1971
1972                 /* Proto-tcf does not exist, create new one */
1973
1974                 if (tca[TCA_KIND] == NULL || !protocol) {
1975                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
1976                         err = -EINVAL;
1977                         goto errout_locked;
1978                 }
1979
1980                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1981                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1982                         err = -ENOENT;
1983                         goto errout_locked;
1984                 }
1985
1986                 if (prio_allocate)
1987                         prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
1988                                                                &chain_info));
1989
1990                 mutex_unlock(&chain->filter_chain_lock);
1991                 tp_new = tcf_proto_create(nla_data(tca[TCA_KIND]),
1992                                           protocol, prio, chain, rtnl_held,
1993                                           extack);
1994                 if (IS_ERR(tp_new)) {
1995                         err = PTR_ERR(tp_new);
1996                         goto errout_tp;
1997                 }
1998
1999                 tp_created = 1;
2000                 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2001                                                 rtnl_held);
2002                 if (IS_ERR(tp)) {
2003                         err = PTR_ERR(tp);
2004                         goto errout_tp;
2005                 }
2006         } else {
2007                 mutex_unlock(&chain->filter_chain_lock);
2008         }
2009
2010         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2011                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2012                 err = -EINVAL;
2013                 goto errout;
2014         }
2015
2016         fh = tp->ops->get(tp, t->tcm_handle);
2017
2018         if (!fh) {
2019                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2020                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2021                         err = -ENOENT;
2022                         goto errout;
2023                 }
2024         } else if (n->nlmsg_flags & NLM_F_EXCL) {
2025                 tfilter_put(tp, fh);
2026                 NL_SET_ERR_MSG(extack, "Filter already exists");
2027                 err = -EEXIST;
2028                 goto errout;
2029         }
2030
2031         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2032                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2033                 err = -EINVAL;
2034                 goto errout;
2035         }
2036
2037         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2038                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
2039                               rtnl_held, extack);
2040         if (err == 0) {
2041                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2042                                RTM_NEWTFILTER, false, rtnl_held);
2043                 tfilter_put(tp, fh);
2044                 /* q pointer is NULL for shared blocks */
2045                 if (q)
2046                         q->flags &= ~TCQ_F_CAN_BYPASS;
2047         }
2048
2049 errout:
2050         if (err && tp_created)
2051                 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2052 errout_tp:
2053         if (chain) {
2054                 if (tp && !IS_ERR(tp))
2055                         tcf_proto_put(tp, rtnl_held, NULL);
2056                 if (!tp_created)
2057                         tcf_chain_put(chain);
2058         }
2059         tcf_block_release(q, block, rtnl_held);
2060
2061         if (rtnl_held)
2062                 rtnl_unlock();
2063
2064         if (err == -EAGAIN) {
2065                 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2066                  * of target chain.
2067                  */
2068                 rtnl_held = true;
2069                 /* Replay the request. */
2070                 goto replay;
2071         }
2072         return err;
2073
2074 errout_locked:
2075         mutex_unlock(&chain->filter_chain_lock);
2076         goto errout;
2077 }
2078
2079 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2080                           struct netlink_ext_ack *extack)
2081 {
2082         struct net *net = sock_net(skb->sk);
2083         struct nlattr *tca[TCA_MAX + 1];
2084         char name[IFNAMSIZ];
2085         struct tcmsg *t;
2086         u32 protocol;
2087         u32 prio;
2088         u32 parent;
2089         u32 chain_index;
2090         struct Qdisc *q = NULL;
2091         struct tcf_chain_info chain_info;
2092         struct tcf_chain *chain = NULL;
2093         struct tcf_block *block = NULL;
2094         struct tcf_proto *tp = NULL;
2095         unsigned long cl = 0;
2096         void *fh = NULL;
2097         int err;
2098         bool rtnl_held = false;
2099
2100         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2101                 return -EPERM;
2102
2103         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2104                                      rtm_tca_policy, extack);
2105         if (err < 0)
2106                 return err;
2107
2108         t = nlmsg_data(n);
2109         protocol = TC_H_MIN(t->tcm_info);
2110         prio = TC_H_MAJ(t->tcm_info);
2111         parent = t->tcm_parent;
2112
2113         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2114                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2115                 return -ENOENT;
2116         }
2117
2118         /* Find head of filter chain. */
2119
2120         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2121         if (err)
2122                 return err;
2123
2124         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2125                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2126                 err = -EINVAL;
2127                 goto errout;
2128         }
2129         /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2130          * found), qdisc is not unlocked, classifier type is not specified,
2131          * classifier is not unlocked.
2132          */
2133         if (!prio ||
2134             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2135             !tcf_proto_is_unlocked(name)) {
2136                 rtnl_held = true;
2137                 rtnl_lock();
2138         }
2139
2140         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2141         if (err)
2142                 goto errout;
2143
2144         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2145                                  extack);
2146         if (IS_ERR(block)) {
2147                 err = PTR_ERR(block);
2148                 goto errout;
2149         }
2150
2151         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2152         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2153                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2154                 err = -EINVAL;
2155                 goto errout;
2156         }
2157         chain = tcf_chain_get(block, chain_index, false);
2158         if (!chain) {
2159                 /* User requested flush on non-existent chain. Nothing to do,
2160                  * so just return success.
2161                  */
2162                 if (prio == 0) {
2163                         err = 0;
2164                         goto errout;
2165                 }
2166                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2167                 err = -ENOENT;
2168                 goto errout;
2169         }
2170
2171         if (prio == 0) {
2172                 tfilter_notify_chain(net, skb, block, q, parent, n,
2173                                      chain, RTM_DELTFILTER, rtnl_held);
2174                 tcf_chain_flush(chain, rtnl_held);
2175                 err = 0;
2176                 goto errout;
2177         }
2178
2179         mutex_lock(&chain->filter_chain_lock);
2180         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2181                                prio, false);
2182         if (!tp || IS_ERR(tp)) {
2183                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2184                 err = tp ? PTR_ERR(tp) : -ENOENT;
2185                 goto errout_locked;
2186         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2187                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2188                 err = -EINVAL;
2189                 goto errout_locked;
2190         } else if (t->tcm_handle == 0) {
2191                 tcf_chain_tp_remove(chain, &chain_info, tp);
2192                 mutex_unlock(&chain->filter_chain_lock);
2193
2194                 tcf_proto_put(tp, rtnl_held, NULL);
2195                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2196                                RTM_DELTFILTER, false, rtnl_held);
2197                 err = 0;
2198                 goto errout;
2199         }
2200         mutex_unlock(&chain->filter_chain_lock);
2201
2202         fh = tp->ops->get(tp, t->tcm_handle);
2203
2204         if (!fh) {
2205                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2206                 err = -ENOENT;
2207         } else {
2208                 bool last;
2209
2210                 err = tfilter_del_notify(net, skb, n, tp, block,
2211                                          q, parent, fh, false, &last,
2212                                          rtnl_held, extack);
2213
2214                 if (err)
2215                         goto errout;
2216                 if (last)
2217                         tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2218         }
2219
2220 errout:
2221         if (chain) {
2222                 if (tp && !IS_ERR(tp))
2223                         tcf_proto_put(tp, rtnl_held, NULL);
2224                 tcf_chain_put(chain);
2225         }
2226         tcf_block_release(q, block, rtnl_held);
2227
2228         if (rtnl_held)
2229                 rtnl_unlock();
2230
2231         return err;
2232
2233 errout_locked:
2234         mutex_unlock(&chain->filter_chain_lock);
2235         goto errout;
2236 }
2237
2238 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2239                           struct netlink_ext_ack *extack)
2240 {
2241         struct net *net = sock_net(skb->sk);
2242         struct nlattr *tca[TCA_MAX + 1];
2243         char name[IFNAMSIZ];
2244         struct tcmsg *t;
2245         u32 protocol;
2246         u32 prio;
2247         u32 parent;
2248         u32 chain_index;
2249         struct Qdisc *q = NULL;
2250         struct tcf_chain_info chain_info;
2251         struct tcf_chain *chain = NULL;
2252         struct tcf_block *block = NULL;
2253         struct tcf_proto *tp = NULL;
2254         unsigned long cl = 0;
2255         void *fh = NULL;
2256         int err;
2257         bool rtnl_held = false;
2258
2259         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2260                                      rtm_tca_policy, extack);
2261         if (err < 0)
2262                 return err;
2263
2264         t = nlmsg_data(n);
2265         protocol = TC_H_MIN(t->tcm_info);
2266         prio = TC_H_MAJ(t->tcm_info);
2267         parent = t->tcm_parent;
2268
2269         if (prio == 0) {
2270                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2271                 return -ENOENT;
2272         }
2273
2274         /* Find head of filter chain. */
2275
2276         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2277         if (err)
2278                 return err;
2279
2280         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2281                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2282                 err = -EINVAL;
2283                 goto errout;
2284         }
2285         /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2286          * unlocked, classifier type is not specified, classifier is not
2287          * unlocked.
2288          */
2289         if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2290             !tcf_proto_is_unlocked(name)) {
2291                 rtnl_held = true;
2292                 rtnl_lock();
2293         }
2294
2295         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2296         if (err)
2297                 goto errout;
2298
2299         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2300                                  extack);
2301         if (IS_ERR(block)) {
2302                 err = PTR_ERR(block);
2303                 goto errout;
2304         }
2305
2306         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2307         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2308                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2309                 err = -EINVAL;
2310                 goto errout;
2311         }
2312         chain = tcf_chain_get(block, chain_index, false);
2313         if (!chain) {
2314                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2315                 err = -EINVAL;
2316                 goto errout;
2317         }
2318
2319         mutex_lock(&chain->filter_chain_lock);
2320         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2321                                prio, false);
2322         mutex_unlock(&chain->filter_chain_lock);
2323         if (!tp || IS_ERR(tp)) {
2324                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2325                 err = tp ? PTR_ERR(tp) : -ENOENT;
2326                 goto errout;
2327         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2328                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2329                 err = -EINVAL;
2330                 goto errout;
2331         }
2332
2333         fh = tp->ops->get(tp, t->tcm_handle);
2334
2335         if (!fh) {
2336                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2337                 err = -ENOENT;
2338         } else {
2339                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2340                                      fh, RTM_NEWTFILTER, true, rtnl_held);
2341                 if (err < 0)
2342                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2343         }
2344
2345         tfilter_put(tp, fh);
2346 errout:
2347         if (chain) {
2348                 if (tp && !IS_ERR(tp))
2349                         tcf_proto_put(tp, rtnl_held, NULL);
2350                 tcf_chain_put(chain);
2351         }
2352         tcf_block_release(q, block, rtnl_held);
2353
2354         if (rtnl_held)
2355                 rtnl_unlock();
2356
2357         return err;
2358 }
2359
2360 struct tcf_dump_args {
2361         struct tcf_walker w;
2362         struct sk_buff *skb;
2363         struct netlink_callback *cb;
2364         struct tcf_block *block;
2365         struct Qdisc *q;
2366         u32 parent;
2367 };
2368
2369 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2370 {
2371         struct tcf_dump_args *a = (void *)arg;
2372         struct net *net = sock_net(a->skb->sk);
2373
2374         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2375                              n, NETLINK_CB(a->cb->skb).portid,
2376                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2377                              RTM_NEWTFILTER, true);
2378 }
2379
2380 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2381                            struct sk_buff *skb, struct netlink_callback *cb,
2382                            long index_start, long *p_index)
2383 {
2384         struct net *net = sock_net(skb->sk);
2385         struct tcf_block *block = chain->block;
2386         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2387         struct tcf_proto *tp, *tp_prev;
2388         struct tcf_dump_args arg;
2389
2390         for (tp = __tcf_get_next_proto(chain, NULL);
2391              tp;
2392              tp_prev = tp,
2393                      tp = __tcf_get_next_proto(chain, tp),
2394                      tcf_proto_put(tp_prev, true, NULL),
2395                      (*p_index)++) {
2396                 if (*p_index < index_start)
2397                         continue;
2398                 if (TC_H_MAJ(tcm->tcm_info) &&
2399                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
2400                         continue;
2401                 if (TC_H_MIN(tcm->tcm_info) &&
2402                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
2403                         continue;
2404                 if (*p_index > index_start)
2405                         memset(&cb->args[1], 0,
2406                                sizeof(cb->args) - sizeof(cb->args[0]));
2407                 if (cb->args[1] == 0) {
2408                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2409                                           NETLINK_CB(cb->skb).portid,
2410                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
2411                                           RTM_NEWTFILTER, true) <= 0)
2412                                 goto errout;
2413                         cb->args[1] = 1;
2414                 }
2415                 if (!tp->ops->walk)
2416                         continue;
2417                 arg.w.fn = tcf_node_dump;
2418                 arg.skb = skb;
2419                 arg.cb = cb;
2420                 arg.block = block;
2421                 arg.q = q;
2422                 arg.parent = parent;
2423                 arg.w.stop = 0;
2424                 arg.w.skip = cb->args[1] - 1;
2425                 arg.w.count = 0;
2426                 arg.w.cookie = cb->args[2];
2427                 tp->ops->walk(tp, &arg.w, true);
2428                 cb->args[2] = arg.w.cookie;
2429                 cb->args[1] = arg.w.count + 1;
2430                 if (arg.w.stop)
2431                         goto errout;
2432         }
2433         return true;
2434
2435 errout:
2436         tcf_proto_put(tp, true, NULL);
2437         return false;
2438 }
2439
2440 /* called with RTNL */
2441 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2442 {
2443         struct tcf_chain *chain, *chain_prev;
2444         struct net *net = sock_net(skb->sk);
2445         struct nlattr *tca[TCA_MAX + 1];
2446         struct Qdisc *q = NULL;
2447         struct tcf_block *block;
2448         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2449         long index_start;
2450         long index;
2451         u32 parent;
2452         int err;
2453
2454         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2455                 return skb->len;
2456
2457         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2458                                      NULL, cb->extack);
2459         if (err)
2460                 return err;
2461
2462         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2463                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2464                 if (!block)
2465                         goto out;
2466                 /* If we work with block index, q is NULL and parent value
2467                  * will never be used in the following code. The check
2468                  * in tcf_fill_node prevents it. However, compiler does not
2469                  * see that far, so set parent to zero to silence the warning
2470                  * about parent being uninitialized.
2471                  */
2472                 parent = 0;
2473         } else {
2474                 const struct Qdisc_class_ops *cops;
2475                 struct net_device *dev;
2476                 unsigned long cl = 0;
2477
2478                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2479                 if (!dev)
2480                         return skb->len;
2481
2482                 parent = tcm->tcm_parent;
2483                 if (!parent) {
2484                         q = dev->qdisc;
2485                         parent = q->handle;
2486                 } else {
2487                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2488                 }
2489                 if (!q)
2490                         goto out;
2491                 cops = q->ops->cl_ops;
2492                 if (!cops)
2493                         goto out;
2494                 if (!cops->tcf_block)
2495                         goto out;
2496                 if (TC_H_MIN(tcm->tcm_parent)) {
2497                         cl = cops->find(q, tcm->tcm_parent);
2498                         if (cl == 0)
2499                                 goto out;
2500                 }
2501                 block = cops->tcf_block(q, cl, NULL);
2502                 if (!block)
2503                         goto out;
2504                 if (tcf_block_shared(block))
2505                         q = NULL;
2506         }
2507
2508         index_start = cb->args[0];
2509         index = 0;
2510
2511         for (chain = __tcf_get_next_chain(block, NULL);
2512              chain;
2513              chain_prev = chain,
2514                      chain = __tcf_get_next_chain(block, chain),
2515                      tcf_chain_put(chain_prev)) {
2516                 if (tca[TCA_CHAIN] &&
2517                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2518                         continue;
2519                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2520                                     index_start, &index)) {
2521                         tcf_chain_put(chain);
2522                         err = -EMSGSIZE;
2523                         break;
2524                 }
2525         }
2526
2527         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2528                 tcf_block_refcnt_put(block, true);
2529         cb->args[0] = index;
2530
2531 out:
2532         /* If we did no progress, the error (EMSGSIZE) is real */
2533         if (skb->len == 0 && err)
2534                 return err;
2535         return skb->len;
2536 }
2537
2538 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2539                               void *tmplt_priv, u32 chain_index,
2540                               struct net *net, struct sk_buff *skb,
2541                               struct tcf_block *block,
2542                               u32 portid, u32 seq, u16 flags, int event)
2543 {
2544         unsigned char *b = skb_tail_pointer(skb);
2545         const struct tcf_proto_ops *ops;
2546         struct nlmsghdr *nlh;
2547         struct tcmsg *tcm;
2548         void *priv;
2549
2550         ops = tmplt_ops;
2551         priv = tmplt_priv;
2552
2553         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2554         if (!nlh)
2555                 goto out_nlmsg_trim;
2556         tcm = nlmsg_data(nlh);
2557         tcm->tcm_family = AF_UNSPEC;
2558         tcm->tcm__pad1 = 0;
2559         tcm->tcm__pad2 = 0;
2560         tcm->tcm_handle = 0;
2561         if (block->q) {
2562                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2563                 tcm->tcm_parent = block->q->handle;
2564         } else {
2565                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2566                 tcm->tcm_block_index = block->index;
2567         }
2568
2569         if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2570                 goto nla_put_failure;
2571
2572         if (ops) {
2573                 if (nla_put_string(skb, TCA_KIND, ops->kind))
2574                         goto nla_put_failure;
2575                 if (ops->tmplt_dump(skb, net, priv) < 0)
2576                         goto nla_put_failure;
2577         }
2578
2579         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2580         return skb->len;
2581
2582 out_nlmsg_trim:
2583 nla_put_failure:
2584         nlmsg_trim(skb, b);
2585         return -EMSGSIZE;
2586 }
2587
2588 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2589                            u32 seq, u16 flags, int event, bool unicast)
2590 {
2591         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2592         struct tcf_block *block = chain->block;
2593         struct net *net = block->net;
2594         struct sk_buff *skb;
2595         int err = 0;
2596
2597         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2598         if (!skb)
2599                 return -ENOBUFS;
2600
2601         if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2602                                chain->index, net, skb, block, portid,
2603                                seq, flags, event) <= 0) {
2604                 kfree_skb(skb);
2605                 return -EINVAL;
2606         }
2607
2608         if (unicast)
2609                 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2610         else
2611                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2612                                      flags & NLM_F_ECHO);
2613
2614         if (err > 0)
2615                 err = 0;
2616         return err;
2617 }
2618
2619 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2620                                   void *tmplt_priv, u32 chain_index,
2621                                   struct tcf_block *block, struct sk_buff *oskb,
2622                                   u32 seq, u16 flags, bool unicast)
2623 {
2624         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2625         struct net *net = block->net;
2626         struct sk_buff *skb;
2627
2628         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2629         if (!skb)
2630                 return -ENOBUFS;
2631
2632         if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2633                                block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2634                 kfree_skb(skb);
2635                 return -EINVAL;
2636         }
2637
2638         if (unicast)
2639                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2640
2641         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2642 }
2643
2644 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2645                               struct nlattr **tca,
2646                               struct netlink_ext_ack *extack)
2647 {
2648         const struct tcf_proto_ops *ops;
2649         void *tmplt_priv;
2650
2651         /* If kind is not set, user did not specify template. */
2652         if (!tca[TCA_KIND])
2653                 return 0;
2654
2655         ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), true, extack);
2656         if (IS_ERR(ops))
2657                 return PTR_ERR(ops);
2658         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2659                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2660                 return -EOPNOTSUPP;
2661         }
2662
2663         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2664         if (IS_ERR(tmplt_priv)) {
2665                 module_put(ops->owner);
2666                 return PTR_ERR(tmplt_priv);
2667         }
2668         chain->tmplt_ops = ops;
2669         chain->tmplt_priv = tmplt_priv;
2670         return 0;
2671 }
2672
2673 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2674                                void *tmplt_priv)
2675 {
2676         /* If template ops are set, no work to do for us. */
2677         if (!tmplt_ops)
2678                 return;
2679
2680         tmplt_ops->tmplt_destroy(tmplt_priv);
2681         module_put(tmplt_ops->owner);
2682 }
2683
2684 /* Add/delete/get a chain */
2685
2686 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2687                         struct netlink_ext_ack *extack)
2688 {
2689         struct net *net = sock_net(skb->sk);
2690         struct nlattr *tca[TCA_MAX + 1];
2691         struct tcmsg *t;
2692         u32 parent;
2693         u32 chain_index;
2694         struct Qdisc *q = NULL;
2695         struct tcf_chain *chain = NULL;
2696         struct tcf_block *block;
2697         unsigned long cl;
2698         int err;
2699
2700         if (n->nlmsg_type != RTM_GETCHAIN &&
2701             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2702                 return -EPERM;
2703
2704 replay:
2705         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2706                                      rtm_tca_policy, extack);
2707         if (err < 0)
2708                 return err;
2709
2710         t = nlmsg_data(n);
2711         parent = t->tcm_parent;
2712         cl = 0;
2713
2714         block = tcf_block_find(net, &q, &parent, &cl,
2715                                t->tcm_ifindex, t->tcm_block_index, extack);
2716         if (IS_ERR(block))
2717                 return PTR_ERR(block);
2718
2719         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2720         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2721                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2722                 err = -EINVAL;
2723                 goto errout_block;
2724         }
2725
2726         mutex_lock(&block->lock);
2727         chain = tcf_chain_lookup(block, chain_index);
2728         if (n->nlmsg_type == RTM_NEWCHAIN) {
2729                 if (chain) {
2730                         if (tcf_chain_held_by_acts_only(chain)) {
2731                                 /* The chain exists only because there is
2732                                  * some action referencing it.
2733                                  */
2734                                 tcf_chain_hold(chain);
2735                         } else {
2736                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2737                                 err = -EEXIST;
2738                                 goto errout_block_locked;
2739                         }
2740                 } else {
2741                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2742                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2743                                 err = -ENOENT;
2744                                 goto errout_block_locked;
2745                         }
2746                         chain = tcf_chain_create(block, chain_index);
2747                         if (!chain) {
2748                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2749                                 err = -ENOMEM;
2750                                 goto errout_block_locked;
2751                         }
2752                 }
2753         } else {
2754                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2755                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2756                         err = -EINVAL;
2757                         goto errout_block_locked;
2758                 }
2759                 tcf_chain_hold(chain);
2760         }
2761
2762         if (n->nlmsg_type == RTM_NEWCHAIN) {
2763                 /* Modifying chain requires holding parent block lock. In case
2764                  * the chain was successfully added, take a reference to the
2765                  * chain. This ensures that an empty chain does not disappear at
2766                  * the end of this function.
2767                  */
2768                 tcf_chain_hold(chain);
2769                 chain->explicitly_created = true;
2770         }
2771         mutex_unlock(&block->lock);
2772
2773         switch (n->nlmsg_type) {
2774         case RTM_NEWCHAIN:
2775                 err = tc_chain_tmplt_add(chain, net, tca, extack);
2776                 if (err) {
2777                         tcf_chain_put_explicitly_created(chain);
2778                         goto errout;
2779                 }
2780
2781                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2782                                 RTM_NEWCHAIN, false);
2783                 break;
2784         case RTM_DELCHAIN:
2785                 tfilter_notify_chain(net, skb, block, q, parent, n,
2786                                      chain, RTM_DELTFILTER, true);
2787                 /* Flush the chain first as the user requested chain removal. */
2788                 tcf_chain_flush(chain, true);
2789                 /* In case the chain was successfully deleted, put a reference
2790                  * to the chain previously taken during addition.
2791                  */
2792                 tcf_chain_put_explicitly_created(chain);
2793                 break;
2794         case RTM_GETCHAIN:
2795                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2796                                       n->nlmsg_seq, n->nlmsg_type, true);
2797                 if (err < 0)
2798                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2799                 break;
2800         default:
2801                 err = -EOPNOTSUPP;
2802                 NL_SET_ERR_MSG(extack, "Unsupported message type");
2803                 goto errout;
2804         }
2805
2806 errout:
2807         tcf_chain_put(chain);
2808 errout_block:
2809         tcf_block_release(q, block, true);
2810         if (err == -EAGAIN)
2811                 /* Replay the request. */
2812                 goto replay;
2813         return err;
2814
2815 errout_block_locked:
2816         mutex_unlock(&block->lock);
2817         goto errout_block;
2818 }
2819
2820 /* called with RTNL */
2821 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2822 {
2823         struct net *net = sock_net(skb->sk);
2824         struct nlattr *tca[TCA_MAX + 1];
2825         struct Qdisc *q = NULL;
2826         struct tcf_block *block;
2827         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2828         struct tcf_chain *chain;
2829         long index_start;
2830         long index;
2831         u32 parent;
2832         int err;
2833
2834         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2835                 return skb->len;
2836
2837         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2838                                      rtm_tca_policy, cb->extack);
2839         if (err)
2840                 return err;
2841
2842         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2843                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2844                 if (!block)
2845                         goto out;
2846                 /* If we work with block index, q is NULL and parent value
2847                  * will never be used in the following code. The check
2848                  * in tcf_fill_node prevents it. However, compiler does not
2849                  * see that far, so set parent to zero to silence the warning
2850                  * about parent being uninitialized.
2851                  */
2852                 parent = 0;
2853         } else {
2854                 const struct Qdisc_class_ops *cops;
2855                 struct net_device *dev;
2856                 unsigned long cl = 0;
2857
2858                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2859                 if (!dev)
2860                         return skb->len;
2861
2862                 parent = tcm->tcm_parent;
2863                 if (!parent) {
2864                         q = dev->qdisc;
2865                         parent = q->handle;
2866                 } else {
2867                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2868                 }
2869                 if (!q)
2870                         goto out;
2871                 cops = q->ops->cl_ops;
2872                 if (!cops)
2873                         goto out;
2874                 if (!cops->tcf_block)
2875                         goto out;
2876                 if (TC_H_MIN(tcm->tcm_parent)) {
2877                         cl = cops->find(q, tcm->tcm_parent);
2878                         if (cl == 0)
2879                                 goto out;
2880                 }
2881                 block = cops->tcf_block(q, cl, NULL);
2882                 if (!block)
2883                         goto out;
2884                 if (tcf_block_shared(block))
2885                         q = NULL;
2886         }
2887
2888         index_start = cb->args[0];
2889         index = 0;
2890
2891         mutex_lock(&block->lock);
2892         list_for_each_entry(chain, &block->chain_list, list) {
2893                 if ((tca[TCA_CHAIN] &&
2894                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2895                         continue;
2896                 if (index < index_start) {
2897                         index++;
2898                         continue;
2899                 }
2900                 if (tcf_chain_held_by_acts_only(chain))
2901                         continue;
2902                 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2903                                          chain->index, net, skb, block,
2904                                          NETLINK_CB(cb->skb).portid,
2905                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
2906                                          RTM_NEWCHAIN);
2907                 if (err <= 0)
2908                         break;
2909                 index++;
2910         }
2911         mutex_unlock(&block->lock);
2912
2913         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2914                 tcf_block_refcnt_put(block, true);
2915         cb->args[0] = index;
2916
2917 out:
2918         /* If we did no progress, the error (EMSGSIZE) is real */
2919         if (skb->len == 0 && err)
2920                 return err;
2921         return skb->len;
2922 }
2923
2924 void tcf_exts_destroy(struct tcf_exts *exts)
2925 {
2926 #ifdef CONFIG_NET_CLS_ACT
2927         if (exts->actions) {
2928                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
2929                 kfree(exts->actions);
2930         }
2931         exts->nr_actions = 0;
2932 #endif
2933 }
2934 EXPORT_SYMBOL(tcf_exts_destroy);
2935
2936 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
2937                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
2938                       bool rtnl_held, struct netlink_ext_ack *extack)
2939 {
2940 #ifdef CONFIG_NET_CLS_ACT
2941         {
2942                 struct tc_action *act;
2943                 size_t attr_size = 0;
2944
2945                 if (exts->police && tb[exts->police]) {
2946                         act = tcf_action_init_1(net, tp, tb[exts->police],
2947                                                 rate_tlv, "police", ovr,
2948                                                 TCA_ACT_BIND, rtnl_held,
2949                                                 extack);
2950                         if (IS_ERR(act))
2951                                 return PTR_ERR(act);
2952
2953                         act->type = exts->type = TCA_OLD_COMPAT;
2954                         exts->actions[0] = act;
2955                         exts->nr_actions = 1;
2956                 } else if (exts->action && tb[exts->action]) {
2957                         int err;
2958
2959                         err = tcf_action_init(net, tp, tb[exts->action],
2960                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
2961                                               exts->actions, &attr_size,
2962                                               rtnl_held, extack);
2963                         if (err < 0)
2964                                 return err;
2965                         exts->nr_actions = err;
2966                 }
2967         }
2968 #else
2969         if ((exts->action && tb[exts->action]) ||
2970             (exts->police && tb[exts->police])) {
2971                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
2972                 return -EOPNOTSUPP;
2973         }
2974 #endif
2975
2976         return 0;
2977 }
2978 EXPORT_SYMBOL(tcf_exts_validate);
2979
2980 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
2981 {
2982 #ifdef CONFIG_NET_CLS_ACT
2983         struct tcf_exts old = *dst;
2984
2985         *dst = *src;
2986         tcf_exts_destroy(&old);
2987 #endif
2988 }
2989 EXPORT_SYMBOL(tcf_exts_change);
2990
2991 #ifdef CONFIG_NET_CLS_ACT
2992 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
2993 {
2994         if (exts->nr_actions == 0)
2995                 return NULL;
2996         else
2997                 return exts->actions[0];
2998 }
2999 #endif
3000
3001 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3002 {
3003 #ifdef CONFIG_NET_CLS_ACT
3004         struct nlattr *nest;
3005
3006         if (exts->action && tcf_exts_has_actions(exts)) {
3007                 /*
3008                  * again for backward compatible mode - we want
3009                  * to work with both old and new modes of entering
3010                  * tc data even if iproute2  was newer - jhs
3011                  */
3012                 if (exts->type != TCA_OLD_COMPAT) {
3013                         nest = nla_nest_start_noflag(skb, exts->action);
3014                         if (nest == NULL)
3015                                 goto nla_put_failure;
3016
3017                         if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
3018                                 goto nla_put_failure;
3019                         nla_nest_end(skb, nest);
3020                 } else if (exts->police) {
3021                         struct tc_action *act = tcf_exts_first_act(exts);
3022                         nest = nla_nest_start_noflag(skb, exts->police);
3023                         if (nest == NULL || !act)
3024                                 goto nla_put_failure;
3025                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3026                                 goto nla_put_failure;
3027                         nla_nest_end(skb, nest);
3028                 }
3029         }
3030         return 0;
3031
3032 nla_put_failure:
3033         nla_nest_cancel(skb, nest);
3034         return -1;
3035 #else
3036         return 0;
3037 #endif
3038 }
3039 EXPORT_SYMBOL(tcf_exts_dump);
3040
3041
3042 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3043 {
3044 #ifdef CONFIG_NET_CLS_ACT
3045         struct tc_action *a = tcf_exts_first_act(exts);
3046         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3047                 return -1;
3048 #endif
3049         return 0;
3050 }
3051 EXPORT_SYMBOL(tcf_exts_dump_stats);
3052
3053 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3054 {
3055         if (*flags & TCA_CLS_FLAGS_IN_HW)
3056                 return;
3057         *flags |= TCA_CLS_FLAGS_IN_HW;
3058         atomic_inc(&block->offloadcnt);
3059 }
3060
3061 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3062 {
3063         if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3064                 return;
3065         *flags &= ~TCA_CLS_FLAGS_IN_HW;
3066         atomic_dec(&block->offloadcnt);
3067 }
3068
3069 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3070                                       struct tcf_proto *tp, u32 *cnt,
3071                                       u32 *flags, u32 diff, bool add)
3072 {
3073         lockdep_assert_held(&block->cb_lock);
3074
3075         spin_lock(&tp->lock);
3076         if (add) {
3077                 if (!*cnt)
3078                         tcf_block_offload_inc(block, flags);
3079                 *cnt += diff;
3080         } else {
3081                 *cnt -= diff;
3082                 if (!*cnt)
3083                         tcf_block_offload_dec(block, flags);
3084         }
3085         spin_unlock(&tp->lock);
3086 }
3087
3088 static void
3089 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3090                          u32 *cnt, u32 *flags)
3091 {
3092         lockdep_assert_held(&block->cb_lock);
3093
3094         spin_lock(&tp->lock);
3095         tcf_block_offload_dec(block, flags);
3096         *cnt = 0;
3097         spin_unlock(&tp->lock);
3098 }
3099
3100 static int
3101 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3102                    void *type_data, bool err_stop)
3103 {
3104         struct flow_block_cb *block_cb;
3105         int ok_count = 0;
3106         int err;
3107
3108         list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3109                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3110                 if (err) {
3111                         if (err_stop)
3112                                 return err;
3113                 } else {
3114                         ok_count++;
3115                 }
3116         }
3117         return ok_count;
3118 }
3119
3120 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3121                      void *type_data, bool err_stop, bool rtnl_held)
3122 {
3123         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3124         int ok_count;
3125
3126 retry:
3127         if (take_rtnl)
3128                 rtnl_lock();
3129         down_read(&block->cb_lock);
3130         /* Need to obtain rtnl lock if block is bound to devs that require it.
3131          * In block bind code cb_lock is obtained while holding rtnl, so we must
3132          * obtain the locks in same order here.
3133          */
3134         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3135                 up_read(&block->cb_lock);
3136                 take_rtnl = true;
3137                 goto retry;
3138         }
3139
3140         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3141
3142         up_read(&block->cb_lock);
3143         if (take_rtnl)
3144                 rtnl_unlock();
3145         return ok_count;
3146 }
3147 EXPORT_SYMBOL(tc_setup_cb_call);
3148
3149 /* Non-destructive filter add. If filter that wasn't already in hardware is
3150  * successfully offloaded, increment block offloads counter. On failure,
3151  * previously offloaded filter is considered to be intact and offloads counter
3152  * is not decremented.
3153  */
3154
3155 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3156                     enum tc_setup_type type, void *type_data, bool err_stop,
3157                     u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3158 {
3159         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3160         int ok_count;
3161
3162 retry:
3163         if (take_rtnl)
3164                 rtnl_lock();
3165         down_read(&block->cb_lock);
3166         /* Need to obtain rtnl lock if block is bound to devs that require it.
3167          * In block bind code cb_lock is obtained while holding rtnl, so we must
3168          * obtain the locks in same order here.
3169          */
3170         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3171                 up_read(&block->cb_lock);
3172                 take_rtnl = true;
3173                 goto retry;
3174         }
3175
3176         /* Make sure all netdevs sharing this block are offload-capable. */
3177         if (block->nooffloaddevcnt && err_stop) {
3178                 ok_count = -EOPNOTSUPP;
3179                 goto err_unlock;
3180         }
3181
3182         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3183         if (ok_count < 0)
3184                 goto err_unlock;
3185
3186         if (tp->ops->hw_add)
3187                 tp->ops->hw_add(tp, type_data);
3188         if (ok_count > 0)
3189                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3190                                           ok_count, true);
3191 err_unlock:
3192         up_read(&block->cb_lock);
3193         if (take_rtnl)
3194                 rtnl_unlock();
3195         return ok_count < 0 ? ok_count : 0;
3196 }
3197 EXPORT_SYMBOL(tc_setup_cb_add);
3198
3199 /* Destructive filter replace. If filter that wasn't already in hardware is
3200  * successfully offloaded, increment block offload counter. On failure,
3201  * previously offloaded filter is considered to be destroyed and offload counter
3202  * is decremented.
3203  */
3204
3205 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3206                         enum tc_setup_type type, void *type_data, bool err_stop,
3207                         u32 *old_flags, unsigned int *old_in_hw_count,
3208                         u32 *new_flags, unsigned int *new_in_hw_count,
3209                         bool rtnl_held)
3210 {
3211         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3212         int ok_count;
3213
3214 retry:
3215         if (take_rtnl)
3216                 rtnl_lock();
3217         down_read(&block->cb_lock);
3218         /* Need to obtain rtnl lock if block is bound to devs that require it.
3219          * In block bind code cb_lock is obtained while holding rtnl, so we must
3220          * obtain the locks in same order here.
3221          */
3222         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3223                 up_read(&block->cb_lock);
3224                 take_rtnl = true;
3225                 goto retry;
3226         }
3227
3228         /* Make sure all netdevs sharing this block are offload-capable. */
3229         if (block->nooffloaddevcnt && err_stop) {
3230                 ok_count = -EOPNOTSUPP;
3231                 goto err_unlock;
3232         }
3233
3234         tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3235         if (tp->ops->hw_del)
3236                 tp->ops->hw_del(tp, type_data);
3237
3238         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3239         if (ok_count < 0)
3240                 goto err_unlock;
3241
3242         if (tp->ops->hw_add)
3243                 tp->ops->hw_add(tp, type_data);
3244         if (ok_count > 0)
3245                 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3246                                           new_flags, ok_count, true);
3247 err_unlock:
3248         up_read(&block->cb_lock);
3249         if (take_rtnl)
3250                 rtnl_unlock();
3251         return ok_count < 0 ? ok_count : 0;
3252 }
3253 EXPORT_SYMBOL(tc_setup_cb_replace);
3254
3255 /* Destroy filter and decrement block offload counter, if filter was previously
3256  * offloaded.
3257  */
3258
3259 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3260                         enum tc_setup_type type, void *type_data, bool err_stop,
3261                         u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3262 {
3263         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3264         int ok_count;
3265
3266 retry:
3267         if (take_rtnl)
3268                 rtnl_lock();
3269         down_read(&block->cb_lock);
3270         /* Need to obtain rtnl lock if block is bound to devs that require it.
3271          * In block bind code cb_lock is obtained while holding rtnl, so we must
3272          * obtain the locks in same order here.
3273          */
3274         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3275                 up_read(&block->cb_lock);
3276                 take_rtnl = true;
3277                 goto retry;
3278         }
3279
3280         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3281
3282         tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3283         if (tp->ops->hw_del)
3284                 tp->ops->hw_del(tp, type_data);
3285
3286         up_read(&block->cb_lock);
3287         if (take_rtnl)
3288                 rtnl_unlock();
3289         return ok_count < 0 ? ok_count : 0;
3290 }
3291 EXPORT_SYMBOL(tc_setup_cb_destroy);
3292
3293 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3294                           bool add, flow_setup_cb_t *cb,
3295                           enum tc_setup_type type, void *type_data,
3296                           void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3297 {
3298         int err = cb(type, type_data, cb_priv);
3299
3300         if (err) {
3301                 if (add && tc_skip_sw(*flags))
3302                         return err;
3303         } else {
3304                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3305                                           add);
3306         }
3307
3308         return 0;
3309 }
3310 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3311
3312 void tc_cleanup_flow_action(struct flow_action *flow_action)
3313 {
3314         struct flow_action_entry *entry;
3315         int i;
3316
3317         flow_action_for_each(i, entry, flow_action)
3318                 if (entry->destructor)
3319                         entry->destructor(entry->destructor_priv);
3320 }
3321 EXPORT_SYMBOL(tc_cleanup_flow_action);
3322
3323 static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3324                                const struct tc_action *act)
3325 {
3326 #ifdef CONFIG_NET_CLS_ACT
3327         entry->dev = act->ops->get_dev(act, &entry->destructor);
3328         if (!entry->dev)
3329                 return;
3330         entry->destructor_priv = entry->dev;
3331 #endif
3332 }
3333
3334 static void tcf_tunnel_encap_put_tunnel(void *priv)
3335 {
3336         struct ip_tunnel_info *tunnel = priv;
3337
3338         kfree(tunnel);
3339 }
3340
3341 static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3342                                        const struct tc_action *act)
3343 {
3344         entry->tunnel = tcf_tunnel_info_copy(act);
3345         if (!entry->tunnel)
3346                 return -ENOMEM;
3347         entry->destructor = tcf_tunnel_encap_put_tunnel;
3348         entry->destructor_priv = entry->tunnel;
3349         return 0;
3350 }
3351
3352 static void tcf_sample_get_group(struct flow_action_entry *entry,
3353                                  const struct tc_action *act)
3354 {
3355 #ifdef CONFIG_NET_CLS_ACT
3356         entry->sample.psample_group =
3357                 act->ops->get_psample_group(act, &entry->destructor);
3358         entry->destructor_priv = entry->sample.psample_group;
3359 #endif
3360 }
3361
3362 int tc_setup_flow_action(struct flow_action *flow_action,
3363                          const struct tcf_exts *exts, bool rtnl_held)
3364 {
3365         const struct tc_action *act;
3366         int i, j, k, err = 0;
3367
3368         if (!exts)
3369                 return 0;
3370
3371         if (!rtnl_held)
3372                 rtnl_lock();
3373
3374         j = 0;
3375         tcf_exts_for_each_action(i, act, exts) {
3376                 struct flow_action_entry *entry;
3377
3378                 entry = &flow_action->entries[j];
3379                 if (is_tcf_gact_ok(act)) {
3380                         entry->id = FLOW_ACTION_ACCEPT;
3381                 } else if (is_tcf_gact_shot(act)) {
3382                         entry->id = FLOW_ACTION_DROP;
3383                 } else if (is_tcf_gact_trap(act)) {
3384                         entry->id = FLOW_ACTION_TRAP;
3385                 } else if (is_tcf_gact_goto_chain(act)) {
3386                         entry->id = FLOW_ACTION_GOTO;
3387                         entry->chain_index = tcf_gact_goto_chain_index(act);
3388                 } else if (is_tcf_mirred_egress_redirect(act)) {
3389                         entry->id = FLOW_ACTION_REDIRECT;
3390                         tcf_mirred_get_dev(entry, act);
3391                 } else if (is_tcf_mirred_egress_mirror(act)) {
3392                         entry->id = FLOW_ACTION_MIRRED;
3393                         tcf_mirred_get_dev(entry, act);
3394                 } else if (is_tcf_mirred_ingress_redirect(act)) {
3395                         entry->id = FLOW_ACTION_REDIRECT_INGRESS;
3396                         tcf_mirred_get_dev(entry, act);
3397                 } else if (is_tcf_mirred_ingress_mirror(act)) {
3398                         entry->id = FLOW_ACTION_MIRRED_INGRESS;
3399                         tcf_mirred_get_dev(entry, act);
3400                 } else if (is_tcf_vlan(act)) {
3401                         switch (tcf_vlan_action(act)) {
3402                         case TCA_VLAN_ACT_PUSH:
3403                                 entry->id = FLOW_ACTION_VLAN_PUSH;
3404                                 entry->vlan.vid = tcf_vlan_push_vid(act);
3405                                 entry->vlan.proto = tcf_vlan_push_proto(act);
3406                                 entry->vlan.prio = tcf_vlan_push_prio(act);
3407                                 break;
3408                         case TCA_VLAN_ACT_POP:
3409                                 entry->id = FLOW_ACTION_VLAN_POP;
3410                                 break;
3411                         case TCA_VLAN_ACT_MODIFY:
3412                                 entry->id = FLOW_ACTION_VLAN_MANGLE;
3413                                 entry->vlan.vid = tcf_vlan_push_vid(act);
3414                                 entry->vlan.proto = tcf_vlan_push_proto(act);
3415                                 entry->vlan.prio = tcf_vlan_push_prio(act);
3416                                 break;
3417                         default:
3418                                 err = -EOPNOTSUPP;
3419                                 goto err_out;
3420                         }
3421                 } else if (is_tcf_tunnel_set(act)) {
3422                         entry->id = FLOW_ACTION_TUNNEL_ENCAP;
3423                         err = tcf_tunnel_encap_get_tunnel(entry, act);
3424                         if (err)
3425                                 goto err_out;
3426                 } else if (is_tcf_tunnel_release(act)) {
3427                         entry->id = FLOW_ACTION_TUNNEL_DECAP;
3428                 } else if (is_tcf_pedit(act)) {
3429                         for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3430                                 switch (tcf_pedit_cmd(act, k)) {
3431                                 case TCA_PEDIT_KEY_EX_CMD_SET:
3432                                         entry->id = FLOW_ACTION_MANGLE;
3433                                         break;
3434                                 case TCA_PEDIT_KEY_EX_CMD_ADD:
3435                                         entry->id = FLOW_ACTION_ADD;
3436                                         break;
3437                                 default:
3438                                         err = -EOPNOTSUPP;
3439                                         goto err_out;
3440                                 }
3441                                 entry->mangle.htype = tcf_pedit_htype(act, k);
3442                                 entry->mangle.mask = tcf_pedit_mask(act, k);
3443                                 entry->mangle.val = tcf_pedit_val(act, k);
3444                                 entry->mangle.offset = tcf_pedit_offset(act, k);
3445                                 entry = &flow_action->entries[++j];
3446                         }
3447                 } else if (is_tcf_csum(act)) {
3448                         entry->id = FLOW_ACTION_CSUM;
3449                         entry->csum_flags = tcf_csum_update_flags(act);
3450                 } else if (is_tcf_skbedit_mark(act)) {
3451                         entry->id = FLOW_ACTION_MARK;
3452                         entry->mark = tcf_skbedit_mark(act);
3453                 } else if (is_tcf_sample(act)) {
3454                         entry->id = FLOW_ACTION_SAMPLE;
3455                         entry->sample.trunc_size = tcf_sample_trunc_size(act);
3456                         entry->sample.truncate = tcf_sample_truncate(act);
3457                         entry->sample.rate = tcf_sample_rate(act);
3458                         tcf_sample_get_group(entry, act);
3459                 } else if (is_tcf_police(act)) {
3460                         entry->id = FLOW_ACTION_POLICE;
3461                         entry->police.burst = tcf_police_tcfp_burst(act);
3462                         entry->police.rate_bytes_ps =
3463                                 tcf_police_rate_bytes_ps(act);
3464                 } else if (is_tcf_ct(act)) {
3465                         entry->id = FLOW_ACTION_CT;
3466                         entry->ct.action = tcf_ct_action(act);
3467                         entry->ct.zone = tcf_ct_zone(act);
3468                 } else if (is_tcf_mpls(act)) {
3469                         switch (tcf_mpls_action(act)) {
3470                         case TCA_MPLS_ACT_PUSH:
3471                                 entry->id = FLOW_ACTION_MPLS_PUSH;
3472                                 entry->mpls_push.proto = tcf_mpls_proto(act);
3473                                 entry->mpls_push.label = tcf_mpls_label(act);
3474                                 entry->mpls_push.tc = tcf_mpls_tc(act);
3475                                 entry->mpls_push.bos = tcf_mpls_bos(act);
3476                                 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3477                                 break;
3478                         case TCA_MPLS_ACT_POP:
3479                                 entry->id = FLOW_ACTION_MPLS_POP;
3480                                 entry->mpls_pop.proto = tcf_mpls_proto(act);
3481                                 break;
3482                         case TCA_MPLS_ACT_MODIFY:
3483                                 entry->id = FLOW_ACTION_MPLS_MANGLE;
3484                                 entry->mpls_mangle.label = tcf_mpls_label(act);
3485                                 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3486                                 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3487                                 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3488                                 break;
3489                         default:
3490                                 goto err_out;
3491                         }
3492                 } else if (is_tcf_skbedit_ptype(act)) {
3493                         entry->id = FLOW_ACTION_PTYPE;
3494                         entry->ptype = tcf_skbedit_ptype(act);
3495                 } else {
3496                         err = -EOPNOTSUPP;
3497                         goto err_out;
3498                 }
3499
3500                 if (!is_tcf_pedit(act))
3501                         j++;
3502         }
3503
3504 err_out:
3505         if (!rtnl_held)
3506                 rtnl_unlock();
3507
3508         if (err)
3509                 tc_cleanup_flow_action(flow_action);
3510
3511         return err;
3512 }
3513 EXPORT_SYMBOL(tc_setup_flow_action);
3514
3515 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3516 {
3517         unsigned int num_acts = 0;
3518         struct tc_action *act;
3519         int i;
3520
3521         tcf_exts_for_each_action(i, act, exts) {
3522                 if (is_tcf_pedit(act))
3523                         num_acts += tcf_pedit_nkeys(act);
3524                 else
3525                         num_acts++;
3526         }
3527         return num_acts;
3528 }
3529 EXPORT_SYMBOL(tcf_exts_num_actions);
3530
3531 static __net_init int tcf_net_init(struct net *net)
3532 {
3533         struct tcf_net *tn = net_generic(net, tcf_net_id);
3534
3535         spin_lock_init(&tn->idr_lock);
3536         idr_init(&tn->idr);
3537         return 0;
3538 }
3539
3540 static void __net_exit tcf_net_exit(struct net *net)
3541 {
3542         struct tcf_net *tn = net_generic(net, tcf_net_id);
3543
3544         idr_destroy(&tn->idr);
3545 }
3546
3547 static struct pernet_operations tcf_net_ops = {
3548         .init = tcf_net_init,
3549         .exit = tcf_net_exit,
3550         .id   = &tcf_net_id,
3551         .size = sizeof(struct tcf_net),
3552 };
3553
3554 static struct flow_indr_block_ing_entry block_ing_entry = {
3555         .cb = tc_indr_block_get_and_ing_cmd,
3556         .list = LIST_HEAD_INIT(block_ing_entry.list),
3557 };
3558
3559 static int __init tc_filter_init(void)
3560 {
3561         int err;
3562
3563         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3564         if (!tc_filter_wq)
3565                 return -ENOMEM;
3566
3567         err = register_pernet_subsys(&tcf_net_ops);
3568         if (err)
3569                 goto err_register_pernet_subsys;
3570
3571         flow_indr_add_block_ing_cb(&block_ing_entry);
3572
3573         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3574                       RTNL_FLAG_DOIT_UNLOCKED);
3575         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3576                       RTNL_FLAG_DOIT_UNLOCKED);
3577         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3578                       tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3579         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3580         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3581         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3582                       tc_dump_chain, 0);
3583
3584         return 0;
3585
3586 err_register_pernet_subsys:
3587         destroy_workqueue(tc_filter_wq);
3588         return err;
3589 }
3590
3591 subsys_initcall(tc_filter_init);