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/jhash.h>
  24 #include <linux/rculist.h>
  25 #include <net/net_namespace.h>
  26 #include <net/sock.h>
  27 #include <net/netlink.h>
  28 #include <net/pkt_sched.h>
  29 #include <net/pkt_cls.h>
  30 #include <net/tc_act/tc_pedit.h>
  31 #include <net/tc_act/tc_mirred.h>
  32 #include <net/tc_act/tc_vlan.h>
  33 #include <net/tc_act/tc_tunnel_key.h>
  34 #include <net/tc_act/tc_csum.h>
  35 #include <net/tc_act/tc_gact.h>
  36 #include <net/tc_act/tc_police.h>
  37 #include <net/tc_act/tc_sample.h>
  38 #include <net/tc_act/tc_skbedit.h>
  39 #include <net/tc_act/tc_ct.h>
  40 #include <net/tc_act/tc_mpls.h>
  41 #include <net/tc_act/tc_gate.h>
  42 #include <net/flow_offload.h>
  43
  44 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
  45
  46 /* The list of all installed classifier types */
  47 static LIST_HEAD(tcf_proto_base);
  48
  49 /* Protects list of registered TC modules. It is pure SMP lock. */
  50 static DEFINE_RWLOCK(cls_mod_lock);
  51
  52 #ifdef CONFIG_NET_CLS_ACT
  53 DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
  54 EXPORT_SYMBOL(tc_skb_ext_tc);
  55
  56 void tc_skb_ext_tc_enable(void)
  57 {
  58         static_branch_inc(&tc_skb_ext_tc);
  59 }
  60 EXPORT_SYMBOL(tc_skb_ext_tc_enable);
  61
  62 void tc_skb_ext_tc_disable(void)
  63 {
  64         static_branch_dec(&tc_skb_ext_tc);
  65 }
  66 EXPORT_SYMBOL(tc_skb_ext_tc_disable);
  67 #endif
  68
  69 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
  70 {
  71         return jhash_3words(tp->chain->index, tp->prio,
  72                             (__force __u32)tp->protocol, 0);
  73 }
  74
  75 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
  76                                         struct tcf_proto *tp)
  77 {
  78         struct tcf_block *block = chain->block;
  79
  80         mutex_lock(&block->proto_destroy_lock);
  81         hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
  82                      destroy_obj_hashfn(tp));
  83         mutex_unlock(&block->proto_destroy_lock);
  84 }
  85
  86 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
  87                           const struct tcf_proto *tp2)
  88 {
  89         return tp1->chain->index == tp2->chain->index &&
  90                tp1->prio == tp2->prio &&
  91                tp1->protocol == tp2->protocol;
  92 }
  93
  94 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
  95                                         struct tcf_proto *tp)
  96 {
  97         u32 hash = destroy_obj_hashfn(tp);
  98         struct tcf_proto *iter;
  99         bool found = false;
 100
 101         rcu_read_lock();
 102         hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
 103                                    destroy_ht_node, hash) {
 104                 if (tcf_proto_cmp(tp, iter)) {
 105                         found = true;
 106                         break;
 107                 }
 108         }
 109         rcu_read_unlock();
 110
 111         return found;
 112 }
 113
 114 static void
 115 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
 116 {
 117         struct tcf_block *block = chain->block;
 118
 119         mutex_lock(&block->proto_destroy_lock);
 120         if (hash_hashed(&tp->destroy_ht_node))
 121                 hash_del_rcu(&tp->destroy_ht_node);
 122         mutex_unlock(&block->proto_destroy_lock);
 123 }
 124
 125 /* Find classifier type by string name */
 126
 127 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
 128 {
 129         const struct tcf_proto_ops *t, *res = NULL;
 130
 131         if (kind) {
 132                 read_lock(&cls_mod_lock);
 133                 list_for_each_entry(t, &tcf_proto_base, head) {
 134                         if (strcmp(kind, t->kind) == 0) {
 135                                 if (try_module_get(t->owner))
 136                                         res = t;
 137                                 break;
 138                         }
 139                 }
 140                 read_unlock(&cls_mod_lock);
 141         }
 142         return res;
 143 }
 144
 145 static const struct tcf_proto_ops *
 146 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
 147                      struct netlink_ext_ack *extack)
 148 {
 149         const struct tcf_proto_ops *ops;
 150
 151         ops = __tcf_proto_lookup_ops(kind);
 152         if (ops)
 153                 return ops;
 154 #ifdef CONFIG_MODULES
 155         if (rtnl_held)
 156                 rtnl_unlock();
 157         request_module("cls_%s", kind);
 158         if (rtnl_held)
 159                 rtnl_lock();
 160         ops = __tcf_proto_lookup_ops(kind);
 161         /* We dropped the RTNL semaphore in order to perform
 162          * the module load. So, even if we succeeded in loading
 163          * the module we have to replay the request. We indicate
 164          * this using -EAGAIN.
 165          */
 166         if (ops) {
 167                 module_put(ops->owner);
 168                 return ERR_PTR(-EAGAIN);
 169         }
 170 #endif
 171         NL_SET_ERR_MSG(extack, "TC classifier not found");
 172         return ERR_PTR(-ENOENT);
 173 }
 174
 175 /* Register(unregister) new classifier type */
 176
 177 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
 178 {
 179         struct tcf_proto_ops *t;
 180         int rc = -EEXIST;
 181
 182         write_lock(&cls_mod_lock);
 183         list_for_each_entry(t, &tcf_proto_base, head)
 184                 if (!strcmp(ops->kind, t->kind))
 185                         goto out;
 186
 187         list_add_tail(&ops->head, &tcf_proto_base);
 188         rc = 0;
 189 out:
 190         write_unlock(&cls_mod_lock);
 191         return rc;
 192 }
 193 EXPORT_SYMBOL(register_tcf_proto_ops);
 194
 195 static struct workqueue_struct *tc_filter_wq;
 196
 197 void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 198 {
 199         struct tcf_proto_ops *t;
 200         int rc = -ENOENT;
 201
 202         /* Wait for outstanding call_rcu()s, if any, from a
 203          * tcf_proto_ops's destroy() handler.
 204          */
 205         rcu_barrier();
 206         flush_workqueue(tc_filter_wq);
 207
 208         write_lock(&cls_mod_lock);
 209         list_for_each_entry(t, &tcf_proto_base, head) {
 210                 if (t == ops) {
 211                         list_del(&t->head);
 212                         rc = 0;
 213                         break;
 214                 }
 215         }
 216         write_unlock(&cls_mod_lock);
 217
 218         WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
 219 }
 220 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 221
 222 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 223 {
 224         INIT_RCU_WORK(rwork, func);
 225         return queue_rcu_work(tc_filter_wq, rwork);
 226 }
 227 EXPORT_SYMBOL(tcf_queue_work);
 228
 229 /* Select new prio value from the range, managed by kernel. */
 230
 231 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 232 {
 233         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 234
 235         if (tp)
 236                 first = tp->prio - 1;
 237
 238         return TC_H_MAJ(first);
 239 }
 240
 241 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
 242 {
 243         if (kind)
 244                 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
 245         memset(name, 0, IFNAMSIZ);
 246         return false;
 247 }
 248
 249 static bool tcf_proto_is_unlocked(const char *kind)
 250 {
 251         const struct tcf_proto_ops *ops;
 252         bool ret;
 253
 254         if (strlen(kind) == 0)
 255                 return false;
 256
 257         ops = tcf_proto_lookup_ops(kind, false, NULL);
 258         /* On error return false to take rtnl lock. Proto lookup/create
 259          * functions will perform lookup again and properly handle errors.
 260          */
 261         if (IS_ERR(ops))
 262                 return false;
 263
 264         ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
 265         module_put(ops->owner);
 266         return ret;
 267 }
 268
 269 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 270                                           u32 prio, struct tcf_chain *chain,
 271                                           bool rtnl_held,
 272                                           struct netlink_ext_ack *extack)
 273 {
 274         struct tcf_proto *tp;
 275         int err;
 276
 277         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 278         if (!tp)
 279                 return ERR_PTR(-ENOBUFS);
 280
 281         tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
 282         if (IS_ERR(tp->ops)) {
 283                 err = PTR_ERR(tp->ops);
 284                 goto errout;
 285         }
 286         tp->classify = tp->ops->classify;
 287         tp->protocol = protocol;
 288         tp->prio = prio;
 289         tp->chain = chain;
 290         spin_lock_init(&tp->lock);
 291         refcount_set(&tp->refcnt, 1);
 292
 293         err = tp->ops->init(tp);
 294         if (err) {
 295                 module_put(tp->ops->owner);
 296                 goto errout;
 297         }
 298         return tp;
 299
 300 errout:
 301         kfree(tp);
 302         return ERR_PTR(err);
 303 }
 304
 305 static void tcf_proto_get(struct tcf_proto *tp)
 306 {
 307         refcount_inc(&tp->refcnt);
 308 }
 309
 310 static void tcf_chain_put(struct tcf_chain *chain);
 311
 312 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
 313                               bool sig_destroy, struct netlink_ext_ack *extack)
 314 {
 315         tp->ops->destroy(tp, rtnl_held, extack);
 316         if (sig_destroy)
 317                 tcf_proto_signal_destroyed(tp->chain, tp);
 318         tcf_chain_put(tp->chain);
 319         module_put(tp->ops->owner);
 320         kfree_rcu(tp, rcu);
 321 }
 322
 323 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
 324                           struct netlink_ext_ack *extack)
 325 {
 326         if (refcount_dec_and_test(&tp->refcnt))
 327                 tcf_proto_destroy(tp, rtnl_held, true, extack);
 328 }
 329
 330 static bool tcf_proto_check_delete(struct tcf_proto *tp)
 331 {
 332         if (tp->ops->delete_empty)
 333                 return tp->ops->delete_empty(tp);
 334
 335         tp->deleting = true;
 336         return tp->deleting;
 337 }
 338
 339 static void tcf_proto_mark_delete(struct tcf_proto *tp)
 340 {
 341         spin_lock(&tp->lock);
 342         tp->deleting = true;
 343         spin_unlock(&tp->lock);
 344 }
 345
 346 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
 347 {
 348         bool deleting;
 349
 350         spin_lock(&tp->lock);
 351         deleting = tp->deleting;
 352         spin_unlock(&tp->lock);
 353
 354         return deleting;
 355 }
 356
 357 #define ASSERT_BLOCK_LOCKED(block)                                      \
 358         lockdep_assert_held(&(block)->lock)
 359
 360 struct tcf_filter_chain_list_item {
 361         struct list_head list;
 362         tcf_chain_head_change_t *chain_head_change;
 363         void *chain_head_change_priv;
 364 };
 365
 366 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 367                                           u32 chain_index)
 368 {
 369         struct tcf_chain *chain;
 370
 371         ASSERT_BLOCK_LOCKED(block);
 372
 373         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 374         if (!chain)
 375                 return NULL;
 376         list_add_tail_rcu(&chain->list, &block->chain_list);
 377         mutex_init(&chain->filter_chain_lock);
 378         chain->block = block;
 379         chain->index = chain_index;
 380         chain->refcnt = 1;
 381         if (!chain->index)
 382                 block->chain0.chain = chain;
 383         return chain;
 384 }
 385
 386 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 387                                        struct tcf_proto *tp_head)
 388 {
 389         if (item->chain_head_change)
 390                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 391 }
 392
 393 static void tcf_chain0_head_change(struct tcf_chain *chain,
 394                                    struct tcf_proto *tp_head)
 395 {
 396         struct tcf_filter_chain_list_item *item;
 397         struct tcf_block *block = chain->block;
 398
 399         if (chain->index)
 400                 return;
 401
 402         mutex_lock(&block->lock);
 403         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 404                 tcf_chain_head_change_item(item, tp_head);
 405         mutex_unlock(&block->lock);
 406 }
 407
 408 /* Returns true if block can be safely freed. */
 409
 410 static bool tcf_chain_detach(struct tcf_chain *chain)
 411 {
 412         struct tcf_block *block = chain->block;
 413
 414         ASSERT_BLOCK_LOCKED(block);
 415
 416         list_del_rcu(&chain->list);
 417         if (!chain->index)
 418                 block->chain0.chain = NULL;
 419
 420         if (list_empty(&block->chain_list) &&
 421             refcount_read(&block->refcnt) == 0)
 422                 return true;
 423
 424         return false;
 425 }
 426
 427 static void tcf_block_destroy(struct tcf_block *block)
 428 {
 429         mutex_destroy(&block->lock);
 430         mutex_destroy(&block->proto_destroy_lock);
 431         kfree_rcu(block, rcu);
 432 }
 433
 434 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
 435 {
 436         struct tcf_block *block = chain->block;
 437
 438         mutex_destroy(&chain->filter_chain_lock);
 439         kfree_rcu(chain, rcu);
 440         if (free_block)
 441                 tcf_block_destroy(block);
 442 }
 443
 444 static void tcf_chain_hold(struct tcf_chain *chain)
 445 {
 446         ASSERT_BLOCK_LOCKED(chain->block);
 447
 448         ++chain->refcnt;
 449 }
 450
 451 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 452 {
 453         ASSERT_BLOCK_LOCKED(chain->block);
 454
 455         /* In case all the references are action references, this
 456          * chain should not be shown to the user.
 457          */
 458         return chain->refcnt == chain->action_refcnt;
 459 }
 460
 461 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 462                                           u32 chain_index)
 463 {
 464         struct tcf_chain *chain;
 465
 466         ASSERT_BLOCK_LOCKED(block);
 467
 468         list_for_each_entry(chain, &block->chain_list, list) {
 469                 if (chain->index == chain_index)
 470                         return chain;
 471         }
 472         return NULL;
 473 }
 474
 475 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 476 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
 477                                               u32 chain_index)
 478 {
 479         struct tcf_chain *chain;
 480
 481         list_for_each_entry_rcu(chain, &block->chain_list, list) {
 482                 if (chain->index == chain_index)
 483                         return chain;
 484         }
 485         return NULL;
 486 }
 487 #endif
 488
 489 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 490                            u32 seq, u16 flags, int event, bool unicast);
 491
 492 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 493                                          u32 chain_index, bool create,
 494                                          bool by_act)
 495 {
 496         struct tcf_chain *chain = NULL;
 497         bool is_first_reference;
 498
 499         mutex_lock(&block->lock);
 500         chain = tcf_chain_lookup(block, chain_index);
 501         if (chain) {
 502                 tcf_chain_hold(chain);
 503         } else {
 504                 if (!create)
 505                         goto errout;
 506                 chain = tcf_chain_create(block, chain_index);
 507                 if (!chain)
 508                         goto errout;
 509         }
 510
 511         if (by_act)
 512                 ++chain->action_refcnt;
 513         is_first_reference = chain->refcnt - chain->action_refcnt == 1;
 514         mutex_unlock(&block->lock);
 515
 516         /* Send notification only in case we got the first
 517          * non-action reference. Until then, the chain acts only as
 518          * a placeholder for actions pointing to it and user ought
 519          * not know about them.
 520          */
 521         if (is_first_reference && !by_act)
 522                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 523                                 RTM_NEWCHAIN, false);
 524
 525         return chain;
 526
 527 errout:
 528         mutex_unlock(&block->lock);
 529         return chain;
 530 }
 531
 532 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 533                                        bool create)
 534 {
 535         return __tcf_chain_get(block, chain_index, create, false);
 536 }
 537
 538 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 539 {
 540         return __tcf_chain_get(block, chain_index, true, true);
 541 }
 542 EXPORT_SYMBOL(tcf_chain_get_by_act);
 543
 544 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
 545                                void *tmplt_priv);
 546 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
 547                                   void *tmplt_priv, u32 chain_index,
 548                                   struct tcf_block *block, struct sk_buff *oskb,
 549                                   u32 seq, u16 flags, bool unicast);
 550
 551 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
 552                             bool explicitly_created)
 553 {
 554         struct tcf_block *block = chain->block;
 555         const struct tcf_proto_ops *tmplt_ops;
 556         bool free_block = false;
 557         unsigned int refcnt;
 558         void *tmplt_priv;
 559
 560         mutex_lock(&block->lock);
 561         if (explicitly_created) {
 562                 if (!chain->explicitly_created) {
 563                         mutex_unlock(&block->lock);
 564                         return;
 565                 }
 566                 chain->explicitly_created = false;
 567         }
 568
 569         if (by_act)
 570                 chain->action_refcnt--;
 571
 572         /* tc_chain_notify_delete can't be called while holding block lock.
 573          * However, when block is unlocked chain can be changed concurrently, so
 574          * save these to temporary variables.
 575          */
 576         refcnt = --chain->refcnt;
 577         tmplt_ops = chain->tmplt_ops;
 578         tmplt_priv = chain->tmplt_priv;
 579
 580         /* The last dropped non-action reference will trigger notification. */
 581         if (refcnt - chain->action_refcnt == 0 && !by_act) {
 582                 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
 583                                        block, NULL, 0, 0, false);
 584                 /* Last reference to chain, no need to lock. */
 585                 chain->flushing = false;
 586         }
 587
 588         if (refcnt == 0)
 589                 free_block = tcf_chain_detach(chain);
 590         mutex_unlock(&block->lock);
 591
 592         if (refcnt == 0) {
 593                 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
 594                 tcf_chain_destroy(chain, free_block);
 595         }
 596 }
 597
 598 static void tcf_chain_put(struct tcf_chain *chain)
 599 {
 600         __tcf_chain_put(chain, false, false);
 601 }
 602
 603 void tcf_chain_put_by_act(struct tcf_chain *chain)
 604 {
 605         __tcf_chain_put(chain, true, false);
 606 }
 607 EXPORT_SYMBOL(tcf_chain_put_by_act);
 608
 609 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 610 {
 611         __tcf_chain_put(chain, false, true);
 612 }
 613
 614 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
 615 {
 616         struct tcf_proto *tp, *tp_next;
 617
 618         mutex_lock(&chain->filter_chain_lock);
 619         tp = tcf_chain_dereference(chain->filter_chain, chain);
 620         while (tp) {
 621                 tp_next = rcu_dereference_protected(tp->next, 1);
 622                 tcf_proto_signal_destroying(chain, tp);
 623                 tp = tp_next;
 624         }
 625         tp = tcf_chain_dereference(chain->filter_chain, chain);
 626         RCU_INIT_POINTER(chain->filter_chain, NULL);
 627         tcf_chain0_head_change(chain, NULL);
 628         chain->flushing = true;
 629         mutex_unlock(&chain->filter_chain_lock);
 630
 631         while (tp) {
 632                 tp_next = rcu_dereference_protected(tp->next, 1);
 633                 tcf_proto_put(tp, rtnl_held, NULL);
 634                 tp = tp_next;
 635         }
 636 }
 637
 638 static int tcf_block_setup(struct tcf_block *block,
 639                            struct flow_block_offload *bo);
 640
 641 static void tcf_block_offload_init(struct flow_block_offload *bo,
 642                                    struct net_device *dev, struct Qdisc *sch,
 643                                    enum flow_block_command command,
 644                                    enum flow_block_binder_type binder_type,
 645                                    struct flow_block *flow_block,
 646                                    bool shared, struct netlink_ext_ack *extack)
 647 {
 648         bo->net = dev_net(dev);
 649         bo->command = command;
 650         bo->binder_type = binder_type;
 651         bo->block = flow_block;
 652         bo->block_shared = shared;
 653         bo->extack = extack;
 654         bo->sch = sch;
 655         bo->cb_list_head = &flow_block->cb_list;
 656         INIT_LIST_HEAD(&bo->cb_list);
 657 }
 658
 659 static void tcf_block_unbind(struct tcf_block *block,
 660                              struct flow_block_offload *bo);
 661
 662 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
 663 {
 664         struct tcf_block *block = block_cb->indr.data;
 665         struct net_device *dev = block_cb->indr.dev;
 666         struct Qdisc *sch = block_cb->indr.sch;
 667         struct netlink_ext_ack extack = {};
 668         struct flow_block_offload bo = {};
 669
 670         tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
 671                                block_cb->indr.binder_type,
 672                                &block->flow_block, tcf_block_shared(block),
 673                                &extack);
 674         rtnl_lock();
 675         down_write(&block->cb_lock);
 676         list_del(&block_cb->driver_list);
 677         list_move(&block_cb->list, &bo.cb_list);
 678         tcf_block_unbind(block, &bo);
 679         up_write(&block->cb_lock);
 680         rtnl_unlock();
 681 }
 682
 683 static bool tcf_block_offload_in_use(struct tcf_block *block)
 684 {
 685         return atomic_read(&block->offloadcnt);
 686 }
 687
 688 static int tcf_block_offload_cmd(struct tcf_block *block,
 689                                  struct net_device *dev, struct Qdisc *sch,
 690                                  struct tcf_block_ext_info *ei,
 691                                  enum flow_block_command command,
 692                                  struct netlink_ext_ack *extack)
 693 {
 694         struct flow_block_offload bo = {};
 695
 696         tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
 697                                &block->flow_block, tcf_block_shared(block),
 698                                extack);
 699
 700         if (dev->netdev_ops->ndo_setup_tc) {
 701                 int err;
 702
 703                 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 704                 if (err < 0) {
 705                         if (err != -EOPNOTSUPP)
 706                                 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
 707                         return err;
 708                 }
 709
 710                 return tcf_block_setup(block, &bo);
 711         }
 712
 713         flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
 714                                     tc_block_indr_cleanup);
 715         tcf_block_setup(block, &bo);
 716
 717         return -EOPNOTSUPP;
 718 }
 719
 720 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 721                                   struct tcf_block_ext_info *ei,
 722                                   struct netlink_ext_ack *extack)
 723 {
 724         struct net_device *dev = q->dev_queue->dev;
 725         int err;
 726
 727         down_write(&block->cb_lock);
 728
 729         /* If tc offload feature is disabled and the block we try to bind
 730          * to already has some offloaded filters, forbid to bind.
 731          */
 732         if (dev->netdev_ops->ndo_setup_tc &&
 733             !tc_can_offload(dev) &&
 734             tcf_block_offload_in_use(block)) {
 735                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 736                 err = -EOPNOTSUPP;
 737                 goto err_unlock;
 738         }
 739
 740         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
 741         if (err == -EOPNOTSUPP)
 742                 goto no_offload_dev_inc;
 743         if (err)
 744                 goto err_unlock;
 745
 746         up_write(&block->cb_lock);
 747         return 0;
 748
 749 no_offload_dev_inc:
 750         if (tcf_block_offload_in_use(block))
 751                 goto err_unlock;
 752
 753         err = 0;
 754         block->nooffloaddevcnt++;
 755 err_unlock:
 756         up_write(&block->cb_lock);
 757         return err;
 758 }
 759
 760 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 761                                      struct tcf_block_ext_info *ei)
 762 {
 763         struct net_device *dev = q->dev_queue->dev;
 764         int err;
 765
 766         down_write(&block->cb_lock);
 767         err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
 768         if (err == -EOPNOTSUPP)
 769                 goto no_offload_dev_dec;
 770         up_write(&block->cb_lock);
 771         return;
 772
 773 no_offload_dev_dec:
 774         WARN_ON(block->nooffloaddevcnt-- == 0);
 775         up_write(&block->cb_lock);
 776 }
 777
 778 static int
 779 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 780                               struct tcf_block_ext_info *ei,
 781                               struct netlink_ext_ack *extack)
 782 {
 783         struct tcf_filter_chain_list_item *item;
 784         struct tcf_chain *chain0;
 785
 786         item = kmalloc(sizeof(*item), GFP_KERNEL);
 787         if (!item) {
 788                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 789                 return -ENOMEM;
 790         }
 791         item->chain_head_change = ei->chain_head_change;
 792         item->chain_head_change_priv = ei->chain_head_change_priv;
 793
 794         mutex_lock(&block->lock);
 795         chain0 = block->chain0.chain;
 796         if (chain0)
 797                 tcf_chain_hold(chain0);
 798         else
 799                 list_add(&item->list, &block->chain0.filter_chain_list);
 800         mutex_unlock(&block->lock);
 801
 802         if (chain0) {
 803                 struct tcf_proto *tp_head;
 804
 805                 mutex_lock(&chain0->filter_chain_lock);
 806
 807                 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
 808                 if (tp_head)
 809                         tcf_chain_head_change_item(item, tp_head);
 810
 811                 mutex_lock(&block->lock);
 812                 list_add(&item->list, &block->chain0.filter_chain_list);
 813                 mutex_unlock(&block->lock);
 814
 815                 mutex_unlock(&chain0->filter_chain_lock);
 816                 tcf_chain_put(chain0);
 817         }
 818
 819         return 0;
 820 }
 821
 822 static void
 823 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 824                               struct tcf_block_ext_info *ei)
 825 {
 826         struct tcf_filter_chain_list_item *item;
 827
 828         mutex_lock(&block->lock);
 829         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 830                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 831                     (item->chain_head_change == ei->chain_head_change &&
 832                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 833                         if (block->chain0.chain)
 834                                 tcf_chain_head_change_item(item, NULL);
 835                         list_del(&item->list);
 836                         mutex_unlock(&block->lock);
 837
 838                         kfree(item);
 839                         return;
 840                 }
 841         }
 842         mutex_unlock(&block->lock);
 843         WARN_ON(1);
 844 }
 845
 846 struct tcf_net {
 847         spinlock_t idr_lock; /* Protects idr */
 848         struct idr idr;
 849 };
 850
 851 static unsigned int tcf_net_id;
 852
 853 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 854                             struct netlink_ext_ack *extack)
 855 {
 856         struct tcf_net *tn = net_generic(net, tcf_net_id);
 857         int err;
 858
 859         idr_preload(GFP_KERNEL);
 860         spin_lock(&tn->idr_lock);
 861         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 862                             GFP_NOWAIT);
 863         spin_unlock(&tn->idr_lock);
 864         idr_preload_end();
 865
 866         return err;
 867 }
 868
 869 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 870 {
 871         struct tcf_net *tn = net_generic(net, tcf_net_id);
 872
 873         spin_lock(&tn->idr_lock);
 874         idr_remove(&tn->idr, block->index);
 875         spin_unlock(&tn->idr_lock);
 876 }
 877
 878 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 879                                           u32 block_index,
 880                                           struct netlink_ext_ack *extack)
 881 {
 882         struct tcf_block *block;
 883
 884         block = kzalloc(sizeof(*block), GFP_KERNEL);
 885         if (!block) {
 886                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 887                 return ERR_PTR(-ENOMEM);
 888         }
 889         mutex_init(&block->lock);
 890         mutex_init(&block->proto_destroy_lock);
 891         init_rwsem(&block->cb_lock);
 892         flow_block_init(&block->flow_block);
 893         INIT_LIST_HEAD(&block->chain_list);
 894         INIT_LIST_HEAD(&block->owner_list);
 895         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 896
 897         refcount_set(&block->refcnt, 1);
 898         block->net = net;
 899         block->index = block_index;
 900
 901         /* Don't store q pointer for blocks which are shared */
 902         if (!tcf_block_shared(block))
 903                 block->q = q;
 904         return block;
 905 }
 906
 907 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 908 {
 909         struct tcf_net *tn = net_generic(net, tcf_net_id);
 910
 911         return idr_find(&tn->idr, block_index);
 912 }
 913
 914 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
 915 {
 916         struct tcf_block *block;
 917
 918         rcu_read_lock();
 919         block = tcf_block_lookup(net, block_index);
 920         if (block && !refcount_inc_not_zero(&block->refcnt))
 921                 block = NULL;
 922         rcu_read_unlock();
 923
 924         return block;
 925 }
 926
 927 static struct tcf_chain *
 928 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
 929 {
 930         mutex_lock(&block->lock);
 931         if (chain)
 932                 chain = list_is_last(&chain->list, &block->chain_list) ?
 933                         NULL : list_next_entry(chain, list);
 934         else
 935                 chain = list_first_entry_or_null(&block->chain_list,
 936                                                  struct tcf_chain, list);
 937
 938         /* skip all action-only chains */
 939         while (chain && tcf_chain_held_by_acts_only(chain))
 940                 chain = list_is_last(&chain->list, &block->chain_list) ?
 941                         NULL : list_next_entry(chain, list);
 942
 943         if (chain)
 944                 tcf_chain_hold(chain);
 945         mutex_unlock(&block->lock);
 946
 947         return chain;
 948 }
 949
 950 /* Function to be used by all clients that want to iterate over all chains on
 951  * block. It properly obtains block->lock and takes reference to chain before
 952  * returning it. Users of this function must be tolerant to concurrent chain
 953  * insertion/deletion or ensure that no concurrent chain modification is
 954  * possible. Note that all netlink dump callbacks cannot guarantee to provide
 955  * consistent dump because rtnl lock is released each time skb is filled with
 956  * data and sent to user-space.
 957  */
 958
 959 struct tcf_chain *
 960 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
 961 {
 962         struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
 963
 964         if (chain)
 965                 tcf_chain_put(chain);
 966
 967         return chain_next;
 968 }
 969 EXPORT_SYMBOL(tcf_get_next_chain);
 970
 971 static struct tcf_proto *
 972 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
 973 {
 974         u32 prio = 0;
 975
 976         ASSERT_RTNL();
 977         mutex_lock(&chain->filter_chain_lock);
 978
 979         if (!tp) {
 980                 tp = tcf_chain_dereference(chain->filter_chain, chain);
 981         } else if (tcf_proto_is_deleting(tp)) {
 982                 /* 'deleting' flag is set and chain->filter_chain_lock was
 983                  * unlocked, which means next pointer could be invalid. Restart
 984                  * search.
 985                  */
 986                 prio = tp->prio + 1;
 987                 tp = tcf_chain_dereference(chain->filter_chain, chain);
 988
 989                 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
 990                         if (!tp->deleting && tp->prio >= prio)
 991                                 break;
 992         } else {
 993                 tp = tcf_chain_dereference(tp->next, chain);
 994         }
 995
 996         if (tp)
 997                 tcf_proto_get(tp);
 998
 999         mutex_unlock(&chain->filter_chain_lock);
1000
1001         return tp;
1002 }
1003
1004 /* Function to be used by all clients that want to iterate over all tp's on
1005  * chain. Users of this function must be tolerant to concurrent tp
1006  * insertion/deletion or ensure that no concurrent chain modification is
1007  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1008  * consistent dump because rtnl lock is released each time skb is filled with
1009  * data and sent to user-space.
1010  */
1011
1012 struct tcf_proto *
1013 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1014 {
1015         struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1016
1017         if (tp)
1018                 tcf_proto_put(tp, true, NULL);
1019
1020         return tp_next;
1021 }
1022 EXPORT_SYMBOL(tcf_get_next_proto);
1023
1024 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1025 {
1026         struct tcf_chain *chain;
1027
1028         /* Last reference to block. At this point chains cannot be added or
1029          * removed concurrently.
1030          */
1031         for (chain = tcf_get_next_chain(block, NULL);
1032              chain;
1033              chain = tcf_get_next_chain(block, chain)) {
1034                 tcf_chain_put_explicitly_created(chain);
1035                 tcf_chain_flush(chain, rtnl_held);
1036         }
1037 }
1038
1039 /* Lookup Qdisc and increments its reference counter.
1040  * Set parent, if necessary.
1041  */
1042
1043 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1044                             u32 *parent, int ifindex, bool rtnl_held,
1045                             struct netlink_ext_ack *extack)
1046 {
1047         const struct Qdisc_class_ops *cops;
1048         struct net_device *dev;
1049         int err = 0;
1050
1051         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1052                 return 0;
1053
1054         rcu_read_lock();
1055
1056         /* Find link */
1057         dev = dev_get_by_index_rcu(net, ifindex);
1058         if (!dev) {
1059                 rcu_read_unlock();
1060                 return -ENODEV;
1061         }
1062
1063         /* Find qdisc */
1064         if (!*parent) {
1065                 *q = rcu_dereference(dev->qdisc);
1066                 *parent = (*q)->handle;
1067         } else {
1068                 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1069                 if (!*q) {
1070                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1071                         err = -EINVAL;
1072                         goto errout_rcu;
1073                 }
1074         }
1075
1076         *q = qdisc_refcount_inc_nz(*q);
1077         if (!*q) {
1078                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1079                 err = -EINVAL;
1080                 goto errout_rcu;
1081         }
1082
1083         /* Is it classful? */
1084         cops = (*q)->ops->cl_ops;
1085         if (!cops) {
1086                 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1087                 err = -EINVAL;
1088                 goto errout_qdisc;
1089         }
1090
1091         if (!cops->tcf_block) {
1092                 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1093                 err = -EOPNOTSUPP;
1094                 goto errout_qdisc;
1095         }
1096
1097 errout_rcu:
1098         /* At this point we know that qdisc is not noop_qdisc,
1099          * which means that qdisc holds a reference to net_device
1100          * and we hold a reference to qdisc, so it is safe to release
1101          * rcu read lock.
1102          */
1103         rcu_read_unlock();
1104         return err;
1105
1106 errout_qdisc:
1107         rcu_read_unlock();
1108
1109         if (rtnl_held)
1110                 qdisc_put(*q);
1111         else
1112                 qdisc_put_unlocked(*q);
1113         *q = NULL;
1114
1115         return err;
1116 }
1117
1118 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1119                                int ifindex, struct netlink_ext_ack *extack)
1120 {
1121         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1122                 return 0;
1123
1124         /* Do we search for filter, attached to class? */
1125         if (TC_H_MIN(parent)) {
1126                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1127
1128                 *cl = cops->find(q, parent);
1129                 if (*cl == 0) {
1130                         NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1131                         return -ENOENT;
1132                 }
1133         }
1134
1135         return 0;
1136 }
1137
1138 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1139                                           unsigned long cl, int ifindex,
1140                                           u32 block_index,
1141                                           struct netlink_ext_ack *extack)
1142 {
1143         struct tcf_block *block;
1144
1145         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1146                 block = tcf_block_refcnt_get(net, block_index);
1147                 if (!block) {
1148                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
1149                         return ERR_PTR(-EINVAL);
1150                 }
1151         } else {
1152                 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1153
1154                 block = cops->tcf_block(q, cl, extack);
1155                 if (!block)
1156                         return ERR_PTR(-EINVAL);
1157
1158                 if (tcf_block_shared(block)) {
1159                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1160                         return ERR_PTR(-EOPNOTSUPP);
1161                 }
1162
1163                 /* Always take reference to block in order to support execution
1164                  * of rules update path of cls API without rtnl lock. Caller
1165                  * must release block when it is finished using it. 'if' block
1166                  * of this conditional obtain reference to block by calling
1167                  * tcf_block_refcnt_get().
1168                  */
1169                 refcount_inc(&block->refcnt);
1170         }
1171
1172         return block;
1173 }
1174
1175 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1176                             struct tcf_block_ext_info *ei, bool rtnl_held)
1177 {
1178         if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1179                 /* Flushing/putting all chains will cause the block to be
1180                  * deallocated when last chain is freed. However, if chain_list
1181                  * is empty, block has to be manually deallocated. After block
1182                  * reference counter reached 0, it is no longer possible to
1183                  * increment it or add new chains to block.
1184                  */
1185                 bool free_block = list_empty(&block->chain_list);
1186
1187                 mutex_unlock(&block->lock);
1188                 if (tcf_block_shared(block))
1189                         tcf_block_remove(block, block->net);
1190
1191                 if (q)
1192                         tcf_block_offload_unbind(block, q, ei);
1193
1194                 if (free_block)
1195                         tcf_block_destroy(block);
1196                 else
1197                         tcf_block_flush_all_chains(block, rtnl_held);
1198         } else if (q) {
1199                 tcf_block_offload_unbind(block, q, ei);
1200         }
1201 }
1202
1203 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1204 {
1205         __tcf_block_put(block, NULL, NULL, rtnl_held);
1206 }
1207
1208 /* Find tcf block.
1209  * Set q, parent, cl when appropriate.
1210  */
1211
1212 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1213                                         u32 *parent, unsigned long *cl,
1214                                         int ifindex, u32 block_index,
1215                                         struct netlink_ext_ack *extack)
1216 {
1217         struct tcf_block *block;
1218         int err = 0;
1219
1220         ASSERT_RTNL();
1221
1222         err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1223         if (err)
1224                 goto errout;
1225
1226         err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1227         if (err)
1228                 goto errout_qdisc;
1229
1230         block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1231         if (IS_ERR(block)) {
1232                 err = PTR_ERR(block);
1233                 goto errout_qdisc;
1234         }
1235
1236         return block;
1237
1238 errout_qdisc:
1239         if (*q)
1240                 qdisc_put(*q);
1241 errout:
1242         *q = NULL;
1243         return ERR_PTR(err);
1244 }
1245
1246 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1247                               bool rtnl_held)
1248 {
1249         if (!IS_ERR_OR_NULL(block))
1250                 tcf_block_refcnt_put(block, rtnl_held);
1251
1252         if (q) {
1253                 if (rtnl_held)
1254                         qdisc_put(q);
1255                 else
1256                         qdisc_put_unlocked(q);
1257         }
1258 }
1259
1260 struct tcf_block_owner_item {
1261         struct list_head list;
1262         struct Qdisc *q;
1263         enum flow_block_binder_type binder_type;
1264 };
1265
1266 static void
1267 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1268                                struct Qdisc *q,
1269                                enum flow_block_binder_type binder_type)
1270 {
1271         if (block->keep_dst &&
1272             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1273             binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1274                 netif_keep_dst(qdisc_dev(q));
1275 }
1276
1277 void tcf_block_netif_keep_dst(struct tcf_block *block)
1278 {
1279         struct tcf_block_owner_item *item;
1280
1281         block->keep_dst = true;
1282         list_for_each_entry(item, &block->owner_list, list)
1283                 tcf_block_owner_netif_keep_dst(block, item->q,
1284                                                item->binder_type);
1285 }
1286 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1287
1288 static int tcf_block_owner_add(struct tcf_block *block,
1289                                struct Qdisc *q,
1290                                enum flow_block_binder_type binder_type)
1291 {
1292         struct tcf_block_owner_item *item;
1293
1294         item = kmalloc(sizeof(*item), GFP_KERNEL);
1295         if (!item)
1296                 return -ENOMEM;
1297         item->q = q;
1298         item->binder_type = binder_type;
1299         list_add(&item->list, &block->owner_list);
1300         return 0;
1301 }
1302
1303 static void tcf_block_owner_del(struct tcf_block *block,
1304                                 struct Qdisc *q,
1305                                 enum flow_block_binder_type binder_type)
1306 {
1307         struct tcf_block_owner_item *item;
1308
1309         list_for_each_entry(item, &block->owner_list, list) {
1310                 if (item->q == q && item->binder_type == binder_type) {
1311                         list_del(&item->list);
1312                         kfree(item);
1313                         return;
1314                 }
1315         }
1316         WARN_ON(1);
1317 }
1318
1319 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1320                       struct tcf_block_ext_info *ei,
1321                       struct netlink_ext_ack *extack)
1322 {
1323         struct net *net = qdisc_net(q);
1324         struct tcf_block *block = NULL;
1325         int err;
1326
1327         if (ei->block_index)
1328                 /* block_index not 0 means the shared block is requested */
1329                 block = tcf_block_refcnt_get(net, ei->block_index);
1330
1331         if (!block) {
1332                 block = tcf_block_create(net, q, ei->block_index, extack);
1333                 if (IS_ERR(block))
1334                         return PTR_ERR(block);
1335                 if (tcf_block_shared(block)) {
1336                         err = tcf_block_insert(block, net, extack);
1337                         if (err)
1338                                 goto err_block_insert;
1339                 }
1340         }
1341
1342         err = tcf_block_owner_add(block, q, ei->binder_type);
1343         if (err)
1344                 goto err_block_owner_add;
1345
1346         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1347
1348         err = tcf_chain0_head_change_cb_add(block, ei, extack);
1349         if (err)
1350                 goto err_chain0_head_change_cb_add;
1351
1352         err = tcf_block_offload_bind(block, q, ei, extack);
1353         if (err)
1354                 goto err_block_offload_bind;
1355
1356         *p_block = block;
1357         return 0;
1358
1359 err_block_offload_bind:
1360         tcf_chain0_head_change_cb_del(block, ei);
1361 err_chain0_head_change_cb_add:
1362         tcf_block_owner_del(block, q, ei->binder_type);
1363 err_block_owner_add:
1364 err_block_insert:
1365         tcf_block_refcnt_put(block, true);
1366         return err;
1367 }
1368 EXPORT_SYMBOL(tcf_block_get_ext);
1369
1370 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1371 {
1372         struct tcf_proto __rcu **p_filter_chain = priv;
1373
1374         rcu_assign_pointer(*p_filter_chain, tp_head);
1375 }
1376
1377 int tcf_block_get(struct tcf_block **p_block,
1378                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1379                   struct netlink_ext_ack *extack)
1380 {
1381         struct tcf_block_ext_info ei = {
1382                 .chain_head_change = tcf_chain_head_change_dflt,
1383                 .chain_head_change_priv = p_filter_chain,
1384         };
1385
1386         WARN_ON(!p_filter_chain);
1387         return tcf_block_get_ext(p_block, q, &ei, extack);
1388 }
1389 EXPORT_SYMBOL(tcf_block_get);
1390
1391 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1392  * actions should be all removed after flushing.
1393  */
1394 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1395                        struct tcf_block_ext_info *ei)
1396 {
1397         if (!block)
1398                 return;
1399         tcf_chain0_head_change_cb_del(block, ei);
1400         tcf_block_owner_del(block, q, ei->binder_type);
1401
1402         __tcf_block_put(block, q, ei, true);
1403 }
1404 EXPORT_SYMBOL(tcf_block_put_ext);
1405
1406 void tcf_block_put(struct tcf_block *block)
1407 {
1408         struct tcf_block_ext_info ei = {0, };
1409
1410         if (!block)
1411                 return;
1412         tcf_block_put_ext(block, block->q, &ei);
1413 }
1414
1415 EXPORT_SYMBOL(tcf_block_put);
1416
1417 static int
1418 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1419                             void *cb_priv, bool add, bool offload_in_use,
1420                             struct netlink_ext_ack *extack)
1421 {
1422         struct tcf_chain *chain, *chain_prev;
1423         struct tcf_proto *tp, *tp_prev;
1424         int err;
1425
1426         lockdep_assert_held(&block->cb_lock);
1427
1428         for (chain = __tcf_get_next_chain(block, NULL);
1429              chain;
1430              chain_prev = chain,
1431                      chain = __tcf_get_next_chain(block, chain),
1432                      tcf_chain_put(chain_prev)) {
1433                 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1434                      tp_prev = tp,
1435                              tp = __tcf_get_next_proto(chain, tp),
1436                              tcf_proto_put(tp_prev, true, NULL)) {
1437                         if (tp->ops->reoffload) {
1438                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1439                                                          extack);
1440                                 if (err && add)
1441                                         goto err_playback_remove;
1442                         } else if (add && offload_in_use) {
1443                                 err = -EOPNOTSUPP;
1444                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1445                                 goto err_playback_remove;
1446                         }
1447                 }
1448         }
1449
1450         return 0;
1451
1452 err_playback_remove:
1453         tcf_proto_put(tp, true, NULL);
1454         tcf_chain_put(chain);
1455         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1456                                     extack);
1457         return err;
1458 }
1459
1460 static int tcf_block_bind(struct tcf_block *block,
1461                           struct flow_block_offload *bo)
1462 {
1463         struct flow_block_cb *block_cb, *next;
1464         int err, i = 0;
1465
1466         lockdep_assert_held(&block->cb_lock);
1467
1468         list_for_each_entry(block_cb, &bo->cb_list, list) {
1469                 err = tcf_block_playback_offloads(block, block_cb->cb,
1470                                                   block_cb->cb_priv, true,
1471                                                   tcf_block_offload_in_use(block),
1472                                                   bo->extack);
1473                 if (err)
1474                         goto err_unroll;
1475                 if (!bo->unlocked_driver_cb)
1476                         block->lockeddevcnt++;
1477
1478                 i++;
1479         }
1480         list_splice(&bo->cb_list, &block->flow_block.cb_list);
1481
1482         return 0;
1483
1484 err_unroll:
1485         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1486                 if (i-- > 0) {
1487                         list_del(&block_cb->list);
1488                         tcf_block_playback_offloads(block, block_cb->cb,
1489                                                     block_cb->cb_priv, false,
1490                                                     tcf_block_offload_in_use(block),
1491                                                     NULL);
1492                         if (!bo->unlocked_driver_cb)
1493                                 block->lockeddevcnt--;
1494                 }
1495                 flow_block_cb_free(block_cb);
1496         }
1497
1498         return err;
1499 }
1500
1501 static void tcf_block_unbind(struct tcf_block *block,
1502                              struct flow_block_offload *bo)
1503 {
1504         struct flow_block_cb *block_cb, *next;
1505
1506         lockdep_assert_held(&block->cb_lock);
1507
1508         list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1509                 tcf_block_playback_offloads(block, block_cb->cb,
1510                                             block_cb->cb_priv, false,
1511                                             tcf_block_offload_in_use(block),
1512                                             NULL);
1513                 list_del(&block_cb->list);
1514                 flow_block_cb_free(block_cb);
1515                 if (!bo->unlocked_driver_cb)
1516                         block->lockeddevcnt--;
1517         }
1518 }
1519
1520 static int tcf_block_setup(struct tcf_block *block,
1521                            struct flow_block_offload *bo)
1522 {
1523         int err;
1524
1525         switch (bo->command) {
1526         case FLOW_BLOCK_BIND:
1527                 err = tcf_block_bind(block, bo);
1528                 break;
1529         case FLOW_BLOCK_UNBIND:
1530                 err = 0;
1531                 tcf_block_unbind(block, bo);
1532                 break;
1533         default:
1534                 WARN_ON_ONCE(1);
1535                 err = -EOPNOTSUPP;
1536         }
1537
1538         return err;
1539 }
1540
1541 /* Main classifier routine: scans classifier chain attached
1542  * to this qdisc, (optionally) tests for protocol and asks
1543  * specific classifiers.
1544  */
1545 static inline int __tcf_classify(struct sk_buff *skb,
1546                                  const struct tcf_proto *tp,
1547                                  const struct tcf_proto *orig_tp,
1548                                  struct tcf_result *res,
1549                                  bool compat_mode,
1550                                  u32 *last_executed_chain)
1551 {
1552 #ifdef CONFIG_NET_CLS_ACT
1553         const int max_reclassify_loop = 16;
1554         const struct tcf_proto *first_tp;
1555         int limit = 0;
1556
1557 reclassify:
1558 #endif
1559         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1560                 __be16 protocol = skb_protocol(skb, false);
1561                 int err;
1562
1563                 if (tp->protocol != protocol &&
1564                     tp->protocol != htons(ETH_P_ALL))
1565                         continue;
1566
1567                 err = tp->classify(skb, tp, res);
1568 #ifdef CONFIG_NET_CLS_ACT
1569                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1570                         first_tp = orig_tp;
1571                         *last_executed_chain = first_tp->chain->index;
1572                         goto reset;
1573                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1574                         first_tp = res->goto_tp;
1575                         *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1576                         goto reset;
1577                 }
1578 #endif
1579                 if (err >= 0)
1580                         return err;
1581         }
1582
1583         return TC_ACT_UNSPEC; /* signal: continue lookup */
1584 #ifdef CONFIG_NET_CLS_ACT
1585 reset:
1586         if (unlikely(limit++ >= max_reclassify_loop)) {
1587                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1588                                        tp->chain->block->index,
1589                                        tp->prio & 0xffff,
1590                                        ntohs(tp->protocol));
1591                 return TC_ACT_SHOT;
1592         }
1593
1594         tp = first_tp;
1595         goto reclassify;
1596 #endif
1597 }
1598
1599 int tcf_classify(struct sk_buff *skb,
1600                  const struct tcf_block *block,
1601                  const struct tcf_proto *tp,
1602                  struct tcf_result *res, bool compat_mode)
1603 {
1604 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1605         u32 last_executed_chain = 0;
1606
1607         return __tcf_classify(skb, tp, tp, res, compat_mode,
1608                               &last_executed_chain);
1609 #else
1610         u32 last_executed_chain = tp ? tp->chain->index : 0;
1611         const struct tcf_proto *orig_tp = tp;
1612         struct tc_skb_ext *ext;
1613         int ret;
1614
1615         if (block) {
1616                 ext = skb_ext_find(skb, TC_SKB_EXT);
1617
1618                 if (ext && ext->chain) {
1619                         struct tcf_chain *fchain;
1620
1621                         fchain = tcf_chain_lookup_rcu(block, ext->chain);
1622                         if (!fchain)
1623                                 return TC_ACT_SHOT;
1624
1625                         /* Consume, so cloned/redirect skbs won't inherit ext */
1626                         skb_ext_del(skb, TC_SKB_EXT);
1627
1628                         tp = rcu_dereference_bh(fchain->filter_chain);
1629                         last_executed_chain = fchain->index;
1630                 }
1631         }
1632
1633         ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1634                              &last_executed_chain);
1635
1636         if (tc_skb_ext_tc_enabled()) {
1637                 /* If we missed on some chain */
1638                 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1639                         struct tc_skb_cb *cb = tc_skb_cb(skb);
1640
1641                         ext = tc_skb_ext_alloc(skb);
1642                         if (WARN_ON_ONCE(!ext))
1643                                 return TC_ACT_SHOT;
1644                         ext->chain = last_executed_chain;
1645                         ext->mru = cb->mru;
1646                         ext->post_ct = cb->post_ct;
1647                         ext->post_ct_snat = cb->post_ct_snat;
1648                         ext->post_ct_dnat = cb->post_ct_dnat;
1649                         ext->zone = cb->zone;
1650                 }
1651         }
1652
1653         return ret;
1654 #endif
1655 }
1656 EXPORT_SYMBOL(tcf_classify);
1657
1658 struct tcf_chain_info {
1659         struct tcf_proto __rcu **pprev;
1660         struct tcf_proto __rcu *next;
1661 };
1662
1663 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1664                                            struct tcf_chain_info *chain_info)
1665 {
1666         return tcf_chain_dereference(*chain_info->pprev, chain);
1667 }
1668
1669 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1670                                struct tcf_chain_info *chain_info,
1671                                struct tcf_proto *tp)
1672 {
1673         if (chain->flushing)
1674                 return -EAGAIN;
1675
1676         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1677         if (*chain_info->pprev == chain->filter_chain)
1678                 tcf_chain0_head_change(chain, tp);
1679         tcf_proto_get(tp);
1680         rcu_assign_pointer(*chain_info->pprev, tp);
1681
1682         return 0;
1683 }
1684
1685 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1686                                 struct tcf_chain_info *chain_info,
1687                                 struct tcf_proto *tp)
1688 {
1689         struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1690
1691         tcf_proto_mark_delete(tp);
1692         if (tp == chain->filter_chain)
1693                 tcf_chain0_head_change(chain, next);
1694         RCU_INIT_POINTER(*chain_info->pprev, next);
1695 }
1696
1697 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1698                                            struct tcf_chain_info *chain_info,
1699                                            u32 protocol, u32 prio,
1700                                            bool prio_allocate);
1701
1702 /* Try to insert new proto.
1703  * If proto with specified priority already exists, free new proto
1704  * and return existing one.
1705  */
1706
1707 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1708                                                     struct tcf_proto *tp_new,
1709                                                     u32 protocol, u32 prio,
1710                                                     bool rtnl_held)
1711 {
1712         struct tcf_chain_info chain_info;
1713         struct tcf_proto *tp;
1714         int err = 0;
1715
1716         mutex_lock(&chain->filter_chain_lock);
1717
1718         if (tcf_proto_exists_destroying(chain, tp_new)) {
1719                 mutex_unlock(&chain->filter_chain_lock);
1720                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1721                 return ERR_PTR(-EAGAIN);
1722         }
1723
1724         tp = tcf_chain_tp_find(chain, &chain_info,
1725                                protocol, prio, false);
1726         if (!tp)
1727                 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1728         mutex_unlock(&chain->filter_chain_lock);
1729
1730         if (tp) {
1731                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1732                 tp_new = tp;
1733         } else if (err) {
1734                 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1735                 tp_new = ERR_PTR(err);
1736         }
1737
1738         return tp_new;
1739 }
1740
1741 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1742                                       struct tcf_proto *tp, bool rtnl_held,
1743                                       struct netlink_ext_ack *extack)
1744 {
1745         struct tcf_chain_info chain_info;
1746         struct tcf_proto *tp_iter;
1747         struct tcf_proto **pprev;
1748         struct tcf_proto *next;
1749
1750         mutex_lock(&chain->filter_chain_lock);
1751
1752         /* Atomically find and remove tp from chain. */
1753         for (pprev = &chain->filter_chain;
1754              (tp_iter = tcf_chain_dereference(*pprev, chain));
1755              pprev = &tp_iter->next) {
1756                 if (tp_iter == tp) {
1757                         chain_info.pprev = pprev;
1758                         chain_info.next = tp_iter->next;
1759                         WARN_ON(tp_iter->deleting);
1760                         break;
1761                 }
1762         }
1763         /* Verify that tp still exists and no new filters were inserted
1764          * concurrently.
1765          * Mark tp for deletion if it is empty.
1766          */
1767         if (!tp_iter || !tcf_proto_check_delete(tp)) {
1768                 mutex_unlock(&chain->filter_chain_lock);
1769                 return;
1770         }
1771
1772         tcf_proto_signal_destroying(chain, tp);
1773         next = tcf_chain_dereference(chain_info.next, chain);
1774         if (tp == chain->filter_chain)
1775                 tcf_chain0_head_change(chain, next);
1776         RCU_INIT_POINTER(*chain_info.pprev, next);
1777         mutex_unlock(&chain->filter_chain_lock);
1778
1779         tcf_proto_put(tp, rtnl_held, extack);
1780 }
1781
1782 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1783                                            struct tcf_chain_info *chain_info,
1784                                            u32 protocol, u32 prio,
1785                                            bool prio_allocate)
1786 {
1787         struct tcf_proto **pprev;
1788         struct tcf_proto *tp;
1789
1790         /* Check the chain for existence of proto-tcf with this priority */
1791         for (pprev = &chain->filter_chain;
1792              (tp = tcf_chain_dereference(*pprev, chain));
1793              pprev = &tp->next) {
1794                 if (tp->prio >= prio) {
1795                         if (tp->prio == prio) {
1796                                 if (prio_allocate ||
1797                                     (tp->protocol != protocol && protocol))
1798                                         return ERR_PTR(-EINVAL);
1799                         } else {
1800                                 tp = NULL;
1801                         }
1802                         break;
1803                 }
1804         }
1805         chain_info->pprev = pprev;
1806         if (tp) {
1807                 chain_info->next = tp->next;
1808                 tcf_proto_get(tp);
1809         } else {
1810                 chain_info->next = NULL;
1811         }
1812         return tp;
1813 }
1814
1815 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1816                          struct tcf_proto *tp, struct tcf_block *block,
1817                          struct Qdisc *q, u32 parent, void *fh,
1818                          u32 portid, u32 seq, u16 flags, int event,
1819                          bool terse_dump, bool rtnl_held)
1820 {
1821         struct tcmsg *tcm;
1822         struct nlmsghdr  *nlh;
1823         unsigned char *b = skb_tail_pointer(skb);
1824
1825         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1826         if (!nlh)
1827                 goto out_nlmsg_trim;
1828         tcm = nlmsg_data(nlh);
1829         tcm->tcm_family = AF_UNSPEC;
1830         tcm->tcm__pad1 = 0;
1831         tcm->tcm__pad2 = 0;
1832         if (q) {
1833                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1834                 tcm->tcm_parent = parent;
1835         } else {
1836                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1837                 tcm->tcm_block_index = block->index;
1838         }
1839         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1840         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1841                 goto nla_put_failure;
1842         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1843                 goto nla_put_failure;
1844         if (!fh) {
1845                 tcm->tcm_handle = 0;
1846         } else if (terse_dump) {
1847                 if (tp->ops->terse_dump) {
1848                         if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1849                                                 rtnl_held) < 0)
1850                                 goto nla_put_failure;
1851                 } else {
1852                         goto cls_op_not_supp;
1853                 }
1854         } else {
1855                 if (tp->ops->dump &&
1856                     tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
1857                         goto nla_put_failure;
1858         }
1859         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1860         return skb->len;
1861
1862 out_nlmsg_trim:
1863 nla_put_failure:
1864 cls_op_not_supp:
1865         nlmsg_trim(skb, b);
1866         return -1;
1867 }
1868
1869 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1870                           struct nlmsghdr *n, struct tcf_proto *tp,
1871                           struct tcf_block *block, struct Qdisc *q,
1872                           u32 parent, void *fh, int event, bool unicast,
1873                           bool rtnl_held)
1874 {
1875         struct sk_buff *skb;
1876         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1877         int err = 0;
1878
1879         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1880         if (!skb)
1881                 return -ENOBUFS;
1882
1883         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1884                           n->nlmsg_seq, n->nlmsg_flags, event,
1885                           false, rtnl_held) <= 0) {
1886                 kfree_skb(skb);
1887                 return -EINVAL;
1888         }
1889
1890         if (unicast)
1891                 err = rtnl_unicast(skb, net, portid);
1892         else
1893                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1894                                      n->nlmsg_flags & NLM_F_ECHO);
1895         return err;
1896 }
1897
1898 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1899                               struct nlmsghdr *n, struct tcf_proto *tp,
1900                               struct tcf_block *block, struct Qdisc *q,
1901                               u32 parent, void *fh, bool unicast, bool *last,
1902                               bool rtnl_held, struct netlink_ext_ack *extack)
1903 {
1904         struct sk_buff *skb;
1905         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1906         int err;
1907
1908         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1909         if (!skb)
1910                 return -ENOBUFS;
1911
1912         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1913                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
1914                           false, rtnl_held) <= 0) {
1915                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1916                 kfree_skb(skb);
1917                 return -EINVAL;
1918         }
1919
1920         err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
1921         if (err) {
1922                 kfree_skb(skb);
1923                 return err;
1924         }
1925
1926         if (unicast)
1927                 err = rtnl_unicast(skb, net, portid);
1928         else
1929                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1930                                      n->nlmsg_flags & NLM_F_ECHO);
1931         if (err < 0)
1932                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1933
1934         return err;
1935 }
1936
1937 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1938                                  struct tcf_block *block, struct Qdisc *q,
1939                                  u32 parent, struct nlmsghdr *n,
1940                                  struct tcf_chain *chain, int event)
1941 {
1942         struct tcf_proto *tp;
1943
1944         for (tp = tcf_get_next_proto(chain, NULL);
1945              tp; tp = tcf_get_next_proto(chain, tp))
1946                 tfilter_notify(net, oskb, n, tp, block,
1947                                q, parent, NULL, event, false, true);
1948 }
1949
1950 static void tfilter_put(struct tcf_proto *tp, void *fh)
1951 {
1952         if (tp->ops->put && fh)
1953                 tp->ops->put(tp, fh);
1954 }
1955
1956 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1957                           struct netlink_ext_ack *extack)
1958 {
1959         struct net *net = sock_net(skb->sk);
1960         struct nlattr *tca[TCA_MAX + 1];
1961         char name[IFNAMSIZ];
1962         struct tcmsg *t;
1963         u32 protocol;
1964         u32 prio;
1965         bool prio_allocate;
1966         u32 parent;
1967         u32 chain_index;
1968         struct Qdisc *q;
1969         struct tcf_chain_info chain_info;
1970         struct tcf_chain *chain;
1971         struct tcf_block *block;
1972         struct tcf_proto *tp;
1973         unsigned long cl;
1974         void *fh;
1975         int err;
1976         int tp_created;
1977         bool rtnl_held = false;
1978         u32 flags;
1979
1980         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1981                 return -EPERM;
1982
1983 replay:
1984         tp_created = 0;
1985
1986         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1987                                      rtm_tca_policy, extack);
1988         if (err < 0)
1989                 return err;
1990
1991         t = nlmsg_data(n);
1992         protocol = TC_H_MIN(t->tcm_info);
1993         prio = TC_H_MAJ(t->tcm_info);
1994         prio_allocate = false;
1995         parent = t->tcm_parent;
1996         tp = NULL;
1997         cl = 0;
1998         block = NULL;
1999         q = NULL;
2000         chain = NULL;
2001         flags = 0;
2002
2003         if (prio == 0) {
2004                 /* If no priority is provided by the user,
2005                  * we allocate one.
2006                  */
2007                 if (n->nlmsg_flags & NLM_F_CREATE) {
2008                         prio = TC_H_MAKE(0x80000000U, 0U);
2009                         prio_allocate = true;
2010                 } else {
2011                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2012                         return -ENOENT;
2013                 }
2014         }
2015
2016         /* Find head of filter chain. */
2017
2018         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2019         if (err)
2020                 return err;
2021
2022         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2023                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2024                 err = -EINVAL;
2025                 goto errout;
2026         }
2027
2028         /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2029          * block is shared (no qdisc found), qdisc is not unlocked, classifier
2030          * type is not specified, classifier is not unlocked.
2031          */
2032         if (rtnl_held ||
2033             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2034             !tcf_proto_is_unlocked(name)) {
2035                 rtnl_held = true;
2036                 rtnl_lock();
2037         }
2038
2039         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2040         if (err)
2041                 goto errout;
2042
2043         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2044                                  extack);
2045         if (IS_ERR(block)) {
2046                 err = PTR_ERR(block);
2047                 goto errout;
2048         }
2049         block->classid = parent;
2050
2051         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2052         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2053                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2054                 err = -EINVAL;
2055                 goto errout;
2056         }
2057         chain = tcf_chain_get(block, chain_index, true);
2058         if (!chain) {
2059                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2060                 err = -ENOMEM;
2061                 goto errout;
2062         }
2063
2064         mutex_lock(&chain->filter_chain_lock);
2065         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2066                                prio, prio_allocate);
2067         if (IS_ERR(tp)) {
2068                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2069                 err = PTR_ERR(tp);
2070                 goto errout_locked;
2071         }
2072
2073         if (tp == NULL) {
2074                 struct tcf_proto *tp_new = NULL;
2075
2076                 if (chain->flushing) {
2077                         err = -EAGAIN;
2078                         goto errout_locked;
2079                 }
2080
2081                 /* Proto-tcf does not exist, create new one */
2082
2083                 if (tca[TCA_KIND] == NULL || !protocol) {
2084                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2085                         err = -EINVAL;
2086                         goto errout_locked;
2087                 }
2088
2089                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2090                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2091                         err = -ENOENT;
2092                         goto errout_locked;
2093                 }
2094
2095                 if (prio_allocate)
2096                         prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2097                                                                &chain_info));
2098
2099                 mutex_unlock(&chain->filter_chain_lock);
2100                 tp_new = tcf_proto_create(name, protocol, prio, chain,
2101                                           rtnl_held, extack);
2102                 if (IS_ERR(tp_new)) {
2103                         err = PTR_ERR(tp_new);
2104                         goto errout_tp;
2105                 }
2106
2107                 tp_created = 1;
2108                 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2109                                                 rtnl_held);
2110                 if (IS_ERR(tp)) {
2111                         err = PTR_ERR(tp);
2112                         goto errout_tp;
2113                 }
2114         } else {
2115                 mutex_unlock(&chain->filter_chain_lock);
2116         }
2117
2118         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2119                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2120                 err = -EINVAL;
2121                 goto errout;
2122         }
2123
2124         fh = tp->ops->get(tp, t->tcm_handle);
2125
2126         if (!fh) {
2127                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2128                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2129                         err = -ENOENT;
2130                         goto errout;
2131                 }
2132         } else if (n->nlmsg_flags & NLM_F_EXCL) {
2133                 tfilter_put(tp, fh);
2134                 NL_SET_ERR_MSG(extack, "Filter already exists");
2135                 err = -EEXIST;
2136                 goto errout;
2137         }
2138
2139         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2140                 tfilter_put(tp, fh);
2141                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2142                 err = -EINVAL;
2143                 goto errout;
2144         }
2145
2146         if (!(n->nlmsg_flags & NLM_F_CREATE))
2147                 flags |= TCA_ACT_FLAGS_REPLACE;
2148         if (!rtnl_held)
2149                 flags |= TCA_ACT_FLAGS_NO_RTNL;
2150         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2151                               flags, extack);
2152         if (err == 0) {
2153                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2154                                RTM_NEWTFILTER, false, rtnl_held);
2155                 tfilter_put(tp, fh);
2156                 /* q pointer is NULL for shared blocks */
2157                 if (q)
2158                         q->flags &= ~TCQ_F_CAN_BYPASS;
2159         }
2160
2161 errout:
2162         if (err && tp_created)
2163                 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2164 errout_tp:
2165         if (chain) {
2166                 if (tp && !IS_ERR(tp))
2167                         tcf_proto_put(tp, rtnl_held, NULL);
2168                 if (!tp_created)
2169                         tcf_chain_put(chain);
2170         }
2171         tcf_block_release(q, block, rtnl_held);
2172
2173         if (rtnl_held)
2174                 rtnl_unlock();
2175
2176         if (err == -EAGAIN) {
2177                 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2178                  * of target chain.
2179                  */
2180                 rtnl_held = true;
2181                 /* Replay the request. */
2182                 goto replay;
2183         }
2184         return err;
2185
2186 errout_locked:
2187         mutex_unlock(&chain->filter_chain_lock);
2188         goto errout;
2189 }
2190
2191 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2192                           struct netlink_ext_ack *extack)
2193 {
2194         struct net *net = sock_net(skb->sk);
2195         struct nlattr *tca[TCA_MAX + 1];
2196         char name[IFNAMSIZ];
2197         struct tcmsg *t;
2198         u32 protocol;
2199         u32 prio;
2200         u32 parent;
2201         u32 chain_index;
2202         struct Qdisc *q = NULL;
2203         struct tcf_chain_info chain_info;
2204         struct tcf_chain *chain = NULL;
2205         struct tcf_block *block = NULL;
2206         struct tcf_proto *tp = NULL;
2207         unsigned long cl = 0;
2208         void *fh = NULL;
2209         int err;
2210         bool rtnl_held = false;
2211
2212         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2213                 return -EPERM;
2214
2215         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2216                                      rtm_tca_policy, extack);
2217         if (err < 0)
2218                 return err;
2219
2220         t = nlmsg_data(n);
2221         protocol = TC_H_MIN(t->tcm_info);
2222         prio = TC_H_MAJ(t->tcm_info);
2223         parent = t->tcm_parent;
2224
2225         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2226                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2227                 return -ENOENT;
2228         }
2229
2230         /* Find head of filter chain. */
2231
2232         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2233         if (err)
2234                 return err;
2235
2236         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2237                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2238                 err = -EINVAL;
2239                 goto errout;
2240         }
2241         /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2242          * found), qdisc is not unlocked, classifier type is not specified,
2243          * classifier is not unlocked.
2244          */
2245         if (!prio ||
2246             (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2247             !tcf_proto_is_unlocked(name)) {
2248                 rtnl_held = true;
2249                 rtnl_lock();
2250         }
2251
2252         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2253         if (err)
2254                 goto errout;
2255
2256         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2257                                  extack);
2258         if (IS_ERR(block)) {
2259                 err = PTR_ERR(block);
2260                 goto errout;
2261         }
2262
2263         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2264         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2265                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2266                 err = -EINVAL;
2267                 goto errout;
2268         }
2269         chain = tcf_chain_get(block, chain_index, false);
2270         if (!chain) {
2271                 /* User requested flush on non-existent chain. Nothing to do,
2272                  * so just return success.
2273                  */
2274                 if (prio == 0) {
2275                         err = 0;
2276                         goto errout;
2277                 }
2278                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2279                 err = -ENOENT;
2280                 goto errout;
2281         }
2282
2283         if (prio == 0) {
2284                 tfilter_notify_chain(net, skb, block, q, parent, n,
2285                                      chain, RTM_DELTFILTER);
2286                 tcf_chain_flush(chain, rtnl_held);
2287                 err = 0;
2288                 goto errout;
2289         }
2290
2291         mutex_lock(&chain->filter_chain_lock);
2292         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2293                                prio, false);
2294         if (!tp || IS_ERR(tp)) {
2295                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2296                 err = tp ? PTR_ERR(tp) : -ENOENT;
2297                 goto errout_locked;
2298         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2299                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2300                 err = -EINVAL;
2301                 goto errout_locked;
2302         } else if (t->tcm_handle == 0) {
2303                 tcf_proto_signal_destroying(chain, tp);
2304                 tcf_chain_tp_remove(chain, &chain_info, tp);
2305                 mutex_unlock(&chain->filter_chain_lock);
2306
2307                 tcf_proto_put(tp, rtnl_held, NULL);
2308                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2309                                RTM_DELTFILTER, false, rtnl_held);
2310                 err = 0;
2311                 goto errout;
2312         }
2313         mutex_unlock(&chain->filter_chain_lock);
2314
2315         fh = tp->ops->get(tp, t->tcm_handle);
2316
2317         if (!fh) {
2318                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2319                 err = -ENOENT;
2320         } else {
2321                 bool last;
2322
2323                 err = tfilter_del_notify(net, skb, n, tp, block,
2324                                          q, parent, fh, false, &last,
2325                                          rtnl_held, extack);
2326
2327                 if (err)
2328                         goto errout;
2329                 if (last)
2330                         tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2331         }
2332
2333 errout:
2334         if (chain) {
2335                 if (tp && !IS_ERR(tp))
2336                         tcf_proto_put(tp, rtnl_held, NULL);
2337                 tcf_chain_put(chain);
2338         }
2339         tcf_block_release(q, block, rtnl_held);
2340
2341         if (rtnl_held)
2342                 rtnl_unlock();
2343
2344         return err;
2345
2346 errout_locked:
2347         mutex_unlock(&chain->filter_chain_lock);
2348         goto errout;
2349 }
2350
2351 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2352                           struct netlink_ext_ack *extack)
2353 {
2354         struct net *net = sock_net(skb->sk);
2355         struct nlattr *tca[TCA_MAX + 1];
2356         char name[IFNAMSIZ];
2357         struct tcmsg *t;
2358         u32 protocol;
2359         u32 prio;
2360         u32 parent;
2361         u32 chain_index;
2362         struct Qdisc *q = NULL;
2363         struct tcf_chain_info chain_info;
2364         struct tcf_chain *chain = NULL;
2365         struct tcf_block *block = NULL;
2366         struct tcf_proto *tp = NULL;
2367         unsigned long cl = 0;
2368         void *fh = NULL;
2369         int err;
2370         bool rtnl_held = false;
2371
2372         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2373                                      rtm_tca_policy, extack);
2374         if (err < 0)
2375                 return err;
2376
2377         t = nlmsg_data(n);
2378         protocol = TC_H_MIN(t->tcm_info);
2379         prio = TC_H_MAJ(t->tcm_info);
2380         parent = t->tcm_parent;
2381
2382         if (prio == 0) {
2383                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2384                 return -ENOENT;
2385         }
2386
2387         /* Find head of filter chain. */
2388
2389         err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2390         if (err)
2391                 return err;
2392
2393         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2394                 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2395                 err = -EINVAL;
2396                 goto errout;
2397         }
2398         /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2399          * unlocked, classifier type is not specified, classifier is not
2400          * unlocked.
2401          */
2402         if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2403             !tcf_proto_is_unlocked(name)) {
2404                 rtnl_held = true;
2405                 rtnl_lock();
2406         }
2407
2408         err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2409         if (err)
2410                 goto errout;
2411
2412         block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2413                                  extack);
2414         if (IS_ERR(block)) {
2415                 err = PTR_ERR(block);
2416                 goto errout;
2417         }
2418
2419         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2420         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2421                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2422                 err = -EINVAL;
2423                 goto errout;
2424         }
2425         chain = tcf_chain_get(block, chain_index, false);
2426         if (!chain) {
2427                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2428                 err = -EINVAL;
2429                 goto errout;
2430         }
2431
2432         mutex_lock(&chain->filter_chain_lock);
2433         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2434                                prio, false);
2435         mutex_unlock(&chain->filter_chain_lock);
2436         if (!tp || IS_ERR(tp)) {
2437                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2438                 err = tp ? PTR_ERR(tp) : -ENOENT;
2439                 goto errout;
2440         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2441                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2442                 err = -EINVAL;
2443                 goto errout;
2444         }
2445
2446         fh = tp->ops->get(tp, t->tcm_handle);
2447
2448         if (!fh) {
2449                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2450                 err = -ENOENT;
2451         } else {
2452                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
2453                                      fh, RTM_NEWTFILTER, true, rtnl_held);
2454                 if (err < 0)
2455                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2456         }
2457
2458         tfilter_put(tp, fh);
2459 errout:
2460         if (chain) {
2461                 if (tp && !IS_ERR(tp))
2462                         tcf_proto_put(tp, rtnl_held, NULL);
2463                 tcf_chain_put(chain);
2464         }
2465         tcf_block_release(q, block, rtnl_held);
2466
2467         if (rtnl_held)
2468                 rtnl_unlock();
2469
2470         return err;
2471 }
2472
2473 struct tcf_dump_args {
2474         struct tcf_walker w;
2475         struct sk_buff *skb;
2476         struct netlink_callback *cb;
2477         struct tcf_block *block;
2478         struct Qdisc *q;
2479         u32 parent;
2480         bool terse_dump;
2481 };
2482
2483 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2484 {
2485         struct tcf_dump_args *a = (void *)arg;
2486         struct net *net = sock_net(a->skb->sk);
2487
2488         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2489                              n, NETLINK_CB(a->cb->skb).portid,
2490                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2491                              RTM_NEWTFILTER, a->terse_dump, true);
2492 }
2493
2494 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2495                            struct sk_buff *skb, struct netlink_callback *cb,
2496                            long index_start, long *p_index, bool terse)
2497 {
2498         struct net *net = sock_net(skb->sk);
2499         struct tcf_block *block = chain->block;
2500         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2501         struct tcf_proto *tp, *tp_prev;
2502         struct tcf_dump_args arg;
2503
2504         for (tp = __tcf_get_next_proto(chain, NULL);
2505              tp;
2506              tp_prev = tp,
2507                      tp = __tcf_get_next_proto(chain, tp),
2508                      tcf_proto_put(tp_prev, true, NULL),
2509                      (*p_index)++) {
2510                 if (*p_index < index_start)
2511                         continue;
2512                 if (TC_H_MAJ(tcm->tcm_info) &&
2513                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
2514                         continue;
2515                 if (TC_H_MIN(tcm->tcm_info) &&
2516                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
2517                         continue;
2518                 if (*p_index > index_start)
2519                         memset(&cb->args[1], 0,
2520                                sizeof(cb->args) - sizeof(cb->args[0]));
2521                 if (cb->args[1] == 0) {
2522                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2523                                           NETLINK_CB(cb->skb).portid,
2524                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
2525                                           RTM_NEWTFILTER, false, true) <= 0)
2526                                 goto errout;
2527                         cb->args[1] = 1;
2528                 }
2529                 if (!tp->ops->walk)
2530                         continue;
2531                 arg.w.fn = tcf_node_dump;
2532                 arg.skb = skb;
2533                 arg.cb = cb;
2534                 arg.block = block;
2535                 arg.q = q;
2536                 arg.parent = parent;
2537                 arg.w.stop = 0;
2538                 arg.w.skip = cb->args[1] - 1;
2539                 arg.w.count = 0;
2540                 arg.w.cookie = cb->args[2];
2541                 arg.terse_dump = terse;
2542                 tp->ops->walk(tp, &arg.w, true);
2543                 cb->args[2] = arg.w.cookie;
2544                 cb->args[1] = arg.w.count + 1;
2545                 if (arg.w.stop)
2546                         goto errout;
2547         }
2548         return true;
2549
2550 errout:
2551         tcf_proto_put(tp, true, NULL);
2552         return false;
2553 }
2554
2555 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2556         [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2557 };
2558
2559 /* called with RTNL */
2560 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2561 {
2562         struct tcf_chain *chain, *chain_prev;
2563         struct net *net = sock_net(skb->sk);
2564         struct nlattr *tca[TCA_MAX + 1];
2565         struct Qdisc *q = NULL;
2566         struct tcf_block *block;
2567         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2568         bool terse_dump = false;
2569         long index_start;
2570         long index;
2571         u32 parent;
2572         int err;
2573
2574         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2575                 return skb->len;
2576
2577         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2578                                      tcf_tfilter_dump_policy, cb->extack);
2579         if (err)
2580                 return err;
2581
2582         if (tca[TCA_DUMP_FLAGS]) {
2583                 struct nla_bitfield32 flags =
2584                         nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2585
2586                 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2587         }
2588
2589         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2590                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2591                 if (!block)
2592                         goto out;
2593                 /* If we work with block index, q is NULL and parent value
2594                  * will never be used in the following code. The check
2595                  * in tcf_fill_node prevents it. However, compiler does not
2596                  * see that far, so set parent to zero to silence the warning
2597                  * about parent being uninitialized.
2598                  */
2599                 parent = 0;
2600         } else {
2601                 const struct Qdisc_class_ops *cops;
2602                 struct net_device *dev;
2603                 unsigned long cl = 0;
2604
2605                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2606                 if (!dev)
2607                         return skb->len;
2608
2609                 parent = tcm->tcm_parent;
2610                 if (!parent)
2611                         q = rtnl_dereference(dev->qdisc);
2612                 else
2613                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2614                 if (!q)
2615                         goto out;
2616                 cops = q->ops->cl_ops;
2617                 if (!cops)
2618                         goto out;
2619                 if (!cops->tcf_block)
2620                         goto out;
2621                 if (TC_H_MIN(tcm->tcm_parent)) {
2622                         cl = cops->find(q, tcm->tcm_parent);
2623                         if (cl == 0)
2624                                 goto out;
2625                 }
2626                 block = cops->tcf_block(q, cl, NULL);
2627                 if (!block)
2628                         goto out;
2629                 parent = block->classid;
2630                 if (tcf_block_shared(block))
2631                         q = NULL;
2632         }
2633
2634         index_start = cb->args[0];
2635         index = 0;
2636
2637         for (chain = __tcf_get_next_chain(block, NULL);
2638              chain;
2639              chain_prev = chain,
2640                      chain = __tcf_get_next_chain(block, chain),
2641                      tcf_chain_put(chain_prev)) {
2642                 if (tca[TCA_CHAIN] &&
2643                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2644                         continue;
2645                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
2646                                     index_start, &index, terse_dump)) {
2647                         tcf_chain_put(chain);
2648                         err = -EMSGSIZE;
2649                         break;
2650                 }
2651         }
2652
2653         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2654                 tcf_block_refcnt_put(block, true);
2655         cb->args[0] = index;
2656
2657 out:
2658         /* If we did no progress, the error (EMSGSIZE) is real */
2659         if (skb->len == 0 && err)
2660                 return err;
2661         return skb->len;
2662 }
2663
2664 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2665                               void *tmplt_priv, u32 chain_index,
2666                               struct net *net, struct sk_buff *skb,
2667                               struct tcf_block *block,
2668                               u32 portid, u32 seq, u16 flags, int event)
2669 {
2670         unsigned char *b = skb_tail_pointer(skb);
2671         const struct tcf_proto_ops *ops;
2672         struct nlmsghdr *nlh;
2673         struct tcmsg *tcm;
2674         void *priv;
2675
2676         ops = tmplt_ops;
2677         priv = tmplt_priv;
2678
2679         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2680         if (!nlh)
2681                 goto out_nlmsg_trim;
2682         tcm = nlmsg_data(nlh);
2683         tcm->tcm_family = AF_UNSPEC;
2684         tcm->tcm__pad1 = 0;
2685         tcm->tcm__pad2 = 0;
2686         tcm->tcm_handle = 0;
2687         if (block->q) {
2688                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2689                 tcm->tcm_parent = block->q->handle;
2690         } else {
2691                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2692                 tcm->tcm_block_index = block->index;
2693         }
2694
2695         if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2696                 goto nla_put_failure;
2697
2698         if (ops) {
2699                 if (nla_put_string(skb, TCA_KIND, ops->kind))
2700                         goto nla_put_failure;
2701                 if (ops->tmplt_dump(skb, net, priv) < 0)
2702                         goto nla_put_failure;
2703         }
2704
2705         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2706         return skb->len;
2707
2708 out_nlmsg_trim:
2709 nla_put_failure:
2710         nlmsg_trim(skb, b);
2711         return -EMSGSIZE;
2712 }
2713
2714 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2715                            u32 seq, u16 flags, int event, bool unicast)
2716 {
2717         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2718         struct tcf_block *block = chain->block;
2719         struct net *net = block->net;
2720         struct sk_buff *skb;
2721         int err = 0;
2722
2723         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2724         if (!skb)
2725                 return -ENOBUFS;
2726
2727         if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2728                                chain->index, net, skb, block, portid,
2729                                seq, flags, event) <= 0) {
2730                 kfree_skb(skb);
2731                 return -EINVAL;
2732         }
2733
2734         if (unicast)
2735                 err = rtnl_unicast(skb, net, portid);
2736         else
2737                 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2738                                      flags & NLM_F_ECHO);
2739
2740         return err;
2741 }
2742
2743 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2744                                   void *tmplt_priv, u32 chain_index,
2745                                   struct tcf_block *block, struct sk_buff *oskb,
2746                                   u32 seq, u16 flags, bool unicast)
2747 {
2748         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2749         struct net *net = block->net;
2750         struct sk_buff *skb;
2751
2752         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2753         if (!skb)
2754                 return -ENOBUFS;
2755
2756         if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2757                                block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2758                 kfree_skb(skb);
2759                 return -EINVAL;
2760         }
2761
2762         if (unicast)
2763                 return rtnl_unicast(skb, net, portid);
2764
2765         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2766 }
2767
2768 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2769                               struct nlattr **tca,
2770                               struct netlink_ext_ack *extack)
2771 {
2772         const struct tcf_proto_ops *ops;
2773         char name[IFNAMSIZ];
2774         void *tmplt_priv;
2775
2776         /* If kind is not set, user did not specify template. */
2777         if (!tca[TCA_KIND])
2778                 return 0;
2779
2780         if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2781                 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2782                 return -EINVAL;
2783         }
2784
2785         ops = tcf_proto_lookup_ops(name, true, extack);
2786         if (IS_ERR(ops))
2787                 return PTR_ERR(ops);
2788         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2789                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2790                 return -EOPNOTSUPP;
2791         }
2792
2793         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2794         if (IS_ERR(tmplt_priv)) {
2795                 module_put(ops->owner);
2796                 return PTR_ERR(tmplt_priv);
2797         }
2798         chain->tmplt_ops = ops;
2799         chain->tmplt_priv = tmplt_priv;
2800         return 0;
2801 }
2802
2803 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2804                                void *tmplt_priv)
2805 {
2806         /* If template ops are set, no work to do for us. */
2807         if (!tmplt_ops)
2808                 return;
2809
2810         tmplt_ops->tmplt_destroy(tmplt_priv);
2811         module_put(tmplt_ops->owner);
2812 }
2813
2814 /* Add/delete/get a chain */
2815
2816 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2817                         struct netlink_ext_ack *extack)
2818 {
2819         struct net *net = sock_net(skb->sk);
2820         struct nlattr *tca[TCA_MAX + 1];
2821         struct tcmsg *t;
2822         u32 parent;
2823         u32 chain_index;
2824         struct Qdisc *q;
2825         struct tcf_chain *chain;
2826         struct tcf_block *block;
2827         unsigned long cl;
2828         int err;
2829
2830         if (n->nlmsg_type != RTM_GETCHAIN &&
2831             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2832                 return -EPERM;
2833
2834 replay:
2835         q = NULL;
2836         err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2837                                      rtm_tca_policy, extack);
2838         if (err < 0)
2839                 return err;
2840
2841         t = nlmsg_data(n);
2842         parent = t->tcm_parent;
2843         cl = 0;
2844
2845         block = tcf_block_find(net, &q, &parent, &cl,
2846                                t->tcm_ifindex, t->tcm_block_index, extack);
2847         if (IS_ERR(block))
2848                 return PTR_ERR(block);
2849
2850         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2851         if (chain_index > TC_ACT_EXT_VAL_MASK) {
2852                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2853                 err = -EINVAL;
2854                 goto errout_block;
2855         }
2856
2857         mutex_lock(&block->lock);
2858         chain = tcf_chain_lookup(block, chain_index);
2859         if (n->nlmsg_type == RTM_NEWCHAIN) {
2860                 if (chain) {
2861                         if (tcf_chain_held_by_acts_only(chain)) {
2862                                 /* The chain exists only because there is
2863                                  * some action referencing it.
2864                                  */
2865                                 tcf_chain_hold(chain);
2866                         } else {
2867                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
2868                                 err = -EEXIST;
2869                                 goto errout_block_locked;
2870                         }
2871                 } else {
2872                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2873                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
2874                                 err = -ENOENT;
2875                                 goto errout_block_locked;
2876                         }
2877                         chain = tcf_chain_create(block, chain_index);
2878                         if (!chain) {
2879                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
2880                                 err = -ENOMEM;
2881                                 goto errout_block_locked;
2882                         }
2883                 }
2884         } else {
2885                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
2886                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2887                         err = -EINVAL;
2888                         goto errout_block_locked;
2889                 }
2890                 tcf_chain_hold(chain);
2891         }
2892
2893         if (n->nlmsg_type == RTM_NEWCHAIN) {
2894                 /* Modifying chain requires holding parent block lock. In case
2895                  * the chain was successfully added, take a reference to the
2896                  * chain. This ensures that an empty chain does not disappear at
2897                  * the end of this function.
2898                  */
2899                 tcf_chain_hold(chain);
2900                 chain->explicitly_created = true;
2901         }
2902         mutex_unlock(&block->lock);
2903
2904         switch (n->nlmsg_type) {
2905         case RTM_NEWCHAIN:
2906                 err = tc_chain_tmplt_add(chain, net, tca, extack);
2907                 if (err) {
2908                         tcf_chain_put_explicitly_created(chain);
2909                         goto errout;
2910                 }
2911
2912                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2913                                 RTM_NEWCHAIN, false);
2914                 break;
2915         case RTM_DELCHAIN:
2916                 tfilter_notify_chain(net, skb, block, q, parent, n,
2917                                      chain, RTM_DELTFILTER);
2918                 /* Flush the chain first as the user requested chain removal. */
2919                 tcf_chain_flush(chain, true);
2920                 /* In case the chain was successfully deleted, put a reference
2921                  * to the chain previously taken during addition.
2922                  */
2923                 tcf_chain_put_explicitly_created(chain);
2924                 break;
2925         case RTM_GETCHAIN:
2926                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2927                                       n->nlmsg_flags, n->nlmsg_type, true);
2928                 if (err < 0)
2929                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2930                 break;
2931         default:
2932                 err = -EOPNOTSUPP;
2933                 NL_SET_ERR_MSG(extack, "Unsupported message type");
2934                 goto errout;
2935         }
2936
2937 errout:
2938         tcf_chain_put(chain);
2939 errout_block:
2940         tcf_block_release(q, block, true);
2941         if (err == -EAGAIN)
2942                 /* Replay the request. */
2943                 goto replay;
2944         return err;
2945
2946 errout_block_locked:
2947         mutex_unlock(&block->lock);
2948         goto errout_block;
2949 }
2950
2951 /* called with RTNL */
2952 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2953 {
2954         struct net *net = sock_net(skb->sk);
2955         struct nlattr *tca[TCA_MAX + 1];
2956         struct Qdisc *q = NULL;
2957         struct tcf_block *block;
2958         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2959         struct tcf_chain *chain;
2960         long index_start;
2961         long index;
2962         int err;
2963
2964         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2965                 return skb->len;
2966
2967         err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2968                                      rtm_tca_policy, cb->extack);
2969         if (err)
2970                 return err;
2971
2972         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2973                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2974                 if (!block)
2975                         goto out;
2976         } else {
2977                 const struct Qdisc_class_ops *cops;
2978                 struct net_device *dev;
2979                 unsigned long cl = 0;
2980
2981                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2982                 if (!dev)
2983                         return skb->len;
2984
2985                 if (!tcm->tcm_parent)
2986                         q = rtnl_dereference(dev->qdisc);
2987                 else
2988                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2989
2990                 if (!q)
2991                         goto out;
2992                 cops = q->ops->cl_ops;
2993                 if (!cops)
2994                         goto out;
2995                 if (!cops->tcf_block)
2996                         goto out;
2997                 if (TC_H_MIN(tcm->tcm_parent)) {
2998                         cl = cops->find(q, tcm->tcm_parent);
2999                         if (cl == 0)
3000                                 goto out;
3001                 }
3002                 block = cops->tcf_block(q, cl, NULL);
3003                 if (!block)
3004                         goto out;
3005                 if (tcf_block_shared(block))
3006                         q = NULL;
3007         }
3008
3009         index_start = cb->args[0];
3010         index = 0;
3011
3012         mutex_lock(&block->lock);
3013         list_for_each_entry(chain, &block->chain_list, list) {
3014                 if ((tca[TCA_CHAIN] &&
3015                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3016                         continue;
3017                 if (index < index_start) {
3018                         index++;
3019                         continue;
3020                 }
3021                 if (tcf_chain_held_by_acts_only(chain))
3022                         continue;
3023                 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3024                                          chain->index, net, skb, block,
3025                                          NETLINK_CB(cb->skb).portid,
3026                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
3027                                          RTM_NEWCHAIN);
3028                 if (err <= 0)
3029                         break;
3030                 index++;
3031         }
3032         mutex_unlock(&block->lock);
3033
3034         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3035                 tcf_block_refcnt_put(block, true);
3036         cb->args[0] = index;
3037
3038 out:
3039         /* If we did no progress, the error (EMSGSIZE) is real */
3040         if (skb->len == 0 && err)
3041                 return err;
3042         return skb->len;
3043 }
3044
3045 void tcf_exts_destroy(struct tcf_exts *exts)
3046 {
3047 #ifdef CONFIG_NET_CLS_ACT
3048         if (exts->actions) {
3049                 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3050                 kfree(exts->actions);
3051         }
3052         exts->nr_actions = 0;
3053 #endif
3054 }
3055 EXPORT_SYMBOL(tcf_exts_destroy);
3056
3057 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3058                          struct nlattr *rate_tlv, struct tcf_exts *exts,
3059                          u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3060 {
3061 #ifdef CONFIG_NET_CLS_ACT
3062         {
3063                 int init_res[TCA_ACT_MAX_PRIO] = {};
3064                 struct tc_action *act;
3065                 size_t attr_size = 0;
3066
3067                 if (exts->police && tb[exts->police]) {
3068                         struct tc_action_ops *a_o;
3069
3070                         a_o = tc_action_load_ops(tb[exts->police], true,
3071                                                  !(flags & TCA_ACT_FLAGS_NO_RTNL),
3072                                                  extack);
3073                         if (IS_ERR(a_o))
3074                                 return PTR_ERR(a_o);
3075                         flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3076                         act = tcf_action_init_1(net, tp, tb[exts->police],
3077                                                 rate_tlv, a_o, init_res, flags,
3078                                                 extack);
3079                         module_put(a_o->owner);
3080                         if (IS_ERR(act))
3081                                 return PTR_ERR(act);
3082
3083                         act->type = exts->type = TCA_OLD_COMPAT;
3084                         exts->actions[0] = act;
3085                         exts->nr_actions = 1;
3086                         tcf_idr_insert_many(exts->actions);
3087                 } else if (exts->action && tb[exts->action]) {
3088                         int err;
3089
3090                         flags |= TCA_ACT_FLAGS_BIND;
3091                         err = tcf_action_init(net, tp, tb[exts->action],
3092                                               rate_tlv, exts->actions, init_res,
3093                                               &attr_size, flags, fl_flags,
3094                                               extack);
3095                         if (err < 0)
3096                                 return err;
3097                         exts->nr_actions = err;
3098                 }
3099         }
3100 #else
3101         if ((exts->action && tb[exts->action]) ||
3102             (exts->police && tb[exts->police])) {
3103                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3104                 return -EOPNOTSUPP;
3105         }
3106 #endif
3107
3108         return 0;
3109 }
3110 EXPORT_SYMBOL(tcf_exts_validate_ex);
3111
3112 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3113                       struct nlattr *rate_tlv, struct tcf_exts *exts,
3114                       u32 flags, struct netlink_ext_ack *extack)
3115 {
3116         return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3117                                     flags, 0, extack);
3118 }
3119 EXPORT_SYMBOL(tcf_exts_validate);
3120
3121 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3122 {
3123 #ifdef CONFIG_NET_CLS_ACT
3124         struct tcf_exts old = *dst;
3125
3126         *dst = *src;
3127         tcf_exts_destroy(&old);
3128 #endif
3129 }
3130 EXPORT_SYMBOL(tcf_exts_change);
3131
3132 #ifdef CONFIG_NET_CLS_ACT
3133 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3134 {
3135         if (exts->nr_actions == 0)
3136                 return NULL;
3137         else
3138                 return exts->actions[0];
3139 }
3140 #endif
3141
3142 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3143 {
3144 #ifdef CONFIG_NET_CLS_ACT
3145         struct nlattr *nest;
3146
3147         if (exts->action && tcf_exts_has_actions(exts)) {
3148                 /*
3149                  * again for backward compatible mode - we want
3150                  * to work with both old and new modes of entering
3151                  * tc data even if iproute2  was newer - jhs
3152                  */
3153                 if (exts->type != TCA_OLD_COMPAT) {
3154                         nest = nla_nest_start_noflag(skb, exts->action);
3155                         if (nest == NULL)
3156                                 goto nla_put_failure;
3157
3158                         if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3159                             < 0)
3160                                 goto nla_put_failure;
3161                         nla_nest_end(skb, nest);
3162                 } else if (exts->police) {
3163                         struct tc_action *act = tcf_exts_first_act(exts);
3164                         nest = nla_nest_start_noflag(skb, exts->police);
3165                         if (nest == NULL || !act)
3166                                 goto nla_put_failure;
3167                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3168                                 goto nla_put_failure;
3169                         nla_nest_end(skb, nest);
3170                 }
3171         }
3172         return 0;
3173
3174 nla_put_failure:
3175         nla_nest_cancel(skb, nest);
3176         return -1;
3177 #else
3178         return 0;
3179 #endif
3180 }
3181 EXPORT_SYMBOL(tcf_exts_dump);
3182
3183 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3184 {
3185 #ifdef CONFIG_NET_CLS_ACT
3186         struct nlattr *nest;
3187
3188         if (!exts->action || !tcf_exts_has_actions(exts))
3189                 return 0;
3190
3191         nest = nla_nest_start_noflag(skb, exts->action);
3192         if (!nest)
3193                 goto nla_put_failure;
3194
3195         if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3196                 goto nla_put_failure;
3197         nla_nest_end(skb, nest);
3198         return 0;
3199
3200 nla_put_failure:
3201         nla_nest_cancel(skb, nest);
3202         return -1;
3203 #else
3204         return 0;
3205 #endif
3206 }
3207 EXPORT_SYMBOL(tcf_exts_terse_dump);
3208
3209 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3210 {
3211 #ifdef CONFIG_NET_CLS_ACT
3212         struct tc_action *a = tcf_exts_first_act(exts);
3213         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3214                 return -1;
3215 #endif
3216         return 0;
3217 }
3218 EXPORT_SYMBOL(tcf_exts_dump_stats);
3219
3220 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3221 {
3222         if (*flags & TCA_CLS_FLAGS_IN_HW)
3223                 return;
3224         *flags |= TCA_CLS_FLAGS_IN_HW;
3225         atomic_inc(&block->offloadcnt);
3226 }
3227
3228 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3229 {
3230         if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3231                 return;
3232         *flags &= ~TCA_CLS_FLAGS_IN_HW;
3233         atomic_dec(&block->offloadcnt);
3234 }
3235
3236 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3237                                       struct tcf_proto *tp, u32 *cnt,
3238                                       u32 *flags, u32 diff, bool add)
3239 {
3240         lockdep_assert_held(&block->cb_lock);
3241
3242         spin_lock(&tp->lock);
3243         if (add) {
3244                 if (!*cnt)
3245                         tcf_block_offload_inc(block, flags);
3246                 *cnt += diff;
3247         } else {
3248                 *cnt -= diff;
3249                 if (!*cnt)
3250                         tcf_block_offload_dec(block, flags);
3251         }
3252         spin_unlock(&tp->lock);
3253 }
3254
3255 static void
3256 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3257                          u32 *cnt, u32 *flags)
3258 {
3259         lockdep_assert_held(&block->cb_lock);
3260
3261         spin_lock(&tp->lock);
3262         tcf_block_offload_dec(block, flags);
3263         *cnt = 0;
3264         spin_unlock(&tp->lock);
3265 }
3266
3267 static int
3268 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3269                    void *type_data, bool err_stop)
3270 {
3271         struct flow_block_cb *block_cb;
3272         int ok_count = 0;
3273         int err;
3274
3275         list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3276                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3277                 if (err) {
3278                         if (err_stop)
3279                                 return err;
3280                 } else {
3281                         ok_count++;
3282                 }
3283         }
3284         return ok_count;
3285 }
3286
3287 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3288                      void *type_data, bool err_stop, bool rtnl_held)
3289 {
3290         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3291         int ok_count;
3292
3293 retry:
3294         if (take_rtnl)
3295                 rtnl_lock();
3296         down_read(&block->cb_lock);
3297         /* Need to obtain rtnl lock if block is bound to devs that require it.
3298          * In block bind code cb_lock is obtained while holding rtnl, so we must
3299          * obtain the locks in same order here.
3300          */
3301         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3302                 up_read(&block->cb_lock);
3303                 take_rtnl = true;
3304                 goto retry;
3305         }
3306
3307         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3308
3309         up_read(&block->cb_lock);
3310         if (take_rtnl)
3311                 rtnl_unlock();
3312         return ok_count;
3313 }
3314 EXPORT_SYMBOL(tc_setup_cb_call);
3315
3316 /* Non-destructive filter add. If filter that wasn't already in hardware is
3317  * successfully offloaded, increment block offloads counter. On failure,
3318  * previously offloaded filter is considered to be intact and offloads counter
3319  * is not decremented.
3320  */
3321
3322 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3323                     enum tc_setup_type type, void *type_data, bool err_stop,
3324                     u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3325 {
3326         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3327         int ok_count;
3328
3329 retry:
3330         if (take_rtnl)
3331                 rtnl_lock();
3332         down_read(&block->cb_lock);
3333         /* Need to obtain rtnl lock if block is bound to devs that require it.
3334          * In block bind code cb_lock is obtained while holding rtnl, so we must
3335          * obtain the locks in same order here.
3336          */
3337         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3338                 up_read(&block->cb_lock);
3339                 take_rtnl = true;
3340                 goto retry;
3341         }
3342
3343         /* Make sure all netdevs sharing this block are offload-capable. */
3344         if (block->nooffloaddevcnt && err_stop) {
3345                 ok_count = -EOPNOTSUPP;
3346                 goto err_unlock;
3347         }
3348
3349         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3350         if (ok_count < 0)
3351                 goto err_unlock;
3352
3353         if (tp->ops->hw_add)
3354                 tp->ops->hw_add(tp, type_data);
3355         if (ok_count > 0)
3356                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3357                                           ok_count, true);
3358 err_unlock:
3359         up_read(&block->cb_lock);
3360         if (take_rtnl)
3361                 rtnl_unlock();
3362         return min(ok_count, 0);
3363 }
3364 EXPORT_SYMBOL(tc_setup_cb_add);
3365
3366 /* Destructive filter replace. If filter that wasn't already in hardware is
3367  * successfully offloaded, increment block offload counter. On failure,
3368  * previously offloaded filter is considered to be destroyed and offload counter
3369  * is decremented.
3370  */
3371
3372 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3373                         enum tc_setup_type type, void *type_data, bool err_stop,
3374                         u32 *old_flags, unsigned int *old_in_hw_count,
3375                         u32 *new_flags, unsigned int *new_in_hw_count,
3376                         bool rtnl_held)
3377 {
3378         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3379         int ok_count;
3380
3381 retry:
3382         if (take_rtnl)
3383                 rtnl_lock();
3384         down_read(&block->cb_lock);
3385         /* Need to obtain rtnl lock if block is bound to devs that require it.
3386          * In block bind code cb_lock is obtained while holding rtnl, so we must
3387          * obtain the locks in same order here.
3388          */
3389         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3390                 up_read(&block->cb_lock);
3391                 take_rtnl = true;
3392                 goto retry;
3393         }
3394
3395         /* Make sure all netdevs sharing this block are offload-capable. */
3396         if (block->nooffloaddevcnt && err_stop) {
3397                 ok_count = -EOPNOTSUPP;
3398                 goto err_unlock;
3399         }
3400
3401         tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3402         if (tp->ops->hw_del)
3403                 tp->ops->hw_del(tp, type_data);
3404
3405         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3406         if (ok_count < 0)
3407                 goto err_unlock;
3408
3409         if (tp->ops->hw_add)
3410                 tp->ops->hw_add(tp, type_data);
3411         if (ok_count > 0)
3412                 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3413                                           new_flags, ok_count, true);
3414 err_unlock:
3415         up_read(&block->cb_lock);
3416         if (take_rtnl)
3417                 rtnl_unlock();
3418         return min(ok_count, 0);
3419 }
3420 EXPORT_SYMBOL(tc_setup_cb_replace);
3421
3422 /* Destroy filter and decrement block offload counter, if filter was previously
3423  * offloaded.
3424  */
3425
3426 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3427                         enum tc_setup_type type, void *type_data, bool err_stop,
3428                         u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3429 {
3430         bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3431         int ok_count;
3432
3433 retry:
3434         if (take_rtnl)
3435                 rtnl_lock();
3436         down_read(&block->cb_lock);
3437         /* Need to obtain rtnl lock if block is bound to devs that require it.
3438          * In block bind code cb_lock is obtained while holding rtnl, so we must
3439          * obtain the locks in same order here.
3440          */
3441         if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3442                 up_read(&block->cb_lock);
3443                 take_rtnl = true;
3444                 goto retry;
3445         }
3446
3447         ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3448
3449         tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3450         if (tp->ops->hw_del)
3451                 tp->ops->hw_del(tp, type_data);
3452
3453         up_read(&block->cb_lock);
3454         if (take_rtnl)
3455                 rtnl_unlock();
3456         return min(ok_count, 0);
3457 }
3458 EXPORT_SYMBOL(tc_setup_cb_destroy);
3459
3460 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3461                           bool add, flow_setup_cb_t *cb,
3462                           enum tc_setup_type type, void *type_data,
3463                           void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3464 {
3465         int err = cb(type, type_data, cb_priv);
3466
3467         if (err) {
3468                 if (add && tc_skip_sw(*flags))
3469                         return err;
3470         } else {
3471                 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3472                                           add);
3473         }
3474
3475         return 0;
3476 }
3477 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3478
3479 static int tcf_act_get_cookie(struct flow_action_entry *entry,
3480                               const struct tc_action *act)
3481 {
3482         struct tc_cookie *cookie;
3483         int err = 0;
3484
3485         rcu_read_lock();
3486         cookie = rcu_dereference(act->act_cookie);
3487         if (cookie) {
3488                 entry->cookie = flow_action_cookie_create(cookie->data,
3489                                                           cookie->len,
3490                                                           GFP_ATOMIC);
3491                 if (!entry->cookie)
3492                         err = -ENOMEM;
3493         }
3494         rcu_read_unlock();
3495         return err;
3496 }
3497
3498 static void tcf_act_put_cookie(struct flow_action_entry *entry)
3499 {
3500         flow_action_cookie_destroy(entry->cookie);
3501 }
3502
3503 void tc_cleanup_offload_action(struct flow_action *flow_action)
3504 {
3505         struct flow_action_entry *entry;
3506         int i;
3507
3508         flow_action_for_each(i, entry, flow_action) {
3509                 tcf_act_put_cookie(entry);
3510                 if (entry->destructor)
3511                         entry->destructor(entry->destructor_priv);
3512         }
3513 }
3514 EXPORT_SYMBOL(tc_cleanup_offload_action);
3515
3516 static int tc_setup_offload_act(struct tc_action *act,
3517                                 struct flow_action_entry *entry,
3518                                 u32 *index_inc,
3519                                 struct netlink_ext_ack *extack)
3520 {
3521 #ifdef CONFIG_NET_CLS_ACT
3522         if (act->ops->offload_act_setup) {
3523                 return act->ops->offload_act_setup(act, entry, index_inc, true,
3524                                                    extack);
3525         } else {
3526                 NL_SET_ERR_MSG(extack, "Action does not support offload");
3527                 return -EOPNOTSUPP;
3528         }
3529 #else
3530         return 0;
3531 #endif
3532 }
3533
3534 int tc_setup_action(struct flow_action *flow_action,
3535                     struct tc_action *actions[],
3536                     struct netlink_ext_ack *extack)
3537 {
3538         int i, j, k, index, err = 0;
3539         struct tc_action *act;
3540
3541         BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3542         BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3543         BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3544
3545         if (!actions)
3546                 return 0;
3547
3548         j = 0;
3549         tcf_act_for_each_action(i, act, actions) {
3550                 struct flow_action_entry *entry;
3551
3552                 entry = &flow_action->entries[j];
3553                 spin_lock_bh(&act->tcfa_lock);
3554                 err = tcf_act_get_cookie(entry, act);
3555                 if (err)
3556                         goto err_out_locked;
3557
3558                 index = 0;
3559                 err = tc_setup_offload_act(act, entry, &index, extack);
3560                 if (err)
3561                         goto err_out_locked;
3562
3563                 for (k = 0; k < index ; k++) {
3564                         entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3565                         entry[k].hw_index = act->tcfa_index;
3566                 }
3567
3568                 j += index;
3569
3570                 spin_unlock_bh(&act->tcfa_lock);
3571         }
3572
3573 err_out:
3574         if (err)
3575                 tc_cleanup_offload_action(flow_action);
3576
3577         return err;
3578 err_out_locked:
3579         spin_unlock_bh(&act->tcfa_lock);
3580         goto err_out;
3581 }
3582
3583 int tc_setup_offload_action(struct flow_action *flow_action,
3584                             const struct tcf_exts *exts,
3585                             struct netlink_ext_ack *extack)
3586 {
3587 #ifdef CONFIG_NET_CLS_ACT
3588         if (!exts)
3589                 return 0;
3590
3591         return tc_setup_action(flow_action, exts->actions, extack);
3592 #else
3593         return 0;
3594 #endif
3595 }
3596 EXPORT_SYMBOL(tc_setup_offload_action);
3597
3598 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3599 {
3600         unsigned int num_acts = 0;
3601         struct tc_action *act;
3602         int i;
3603
3604         tcf_exts_for_each_action(i, act, exts) {
3605                 if (is_tcf_pedit(act))
3606                         num_acts += tcf_pedit_nkeys(act);
3607                 else
3608                         num_acts++;
3609         }
3610         return num_acts;
3611 }
3612 EXPORT_SYMBOL(tcf_exts_num_actions);
3613
3614 #ifdef CONFIG_NET_CLS_ACT
3615 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3616                                         u32 *p_block_index,
3617                                         struct netlink_ext_ack *extack)
3618 {
3619         *p_block_index = nla_get_u32(block_index_attr);
3620         if (!*p_block_index) {
3621                 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3622                 return -EINVAL;
3623         }
3624
3625         return 0;
3626 }
3627
3628 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3629                     enum flow_block_binder_type binder_type,
3630                     struct nlattr *block_index_attr,
3631                     struct netlink_ext_ack *extack)
3632 {
3633         u32 block_index;
3634         int err;
3635
3636         if (!block_index_attr)
3637                 return 0;
3638
3639         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3640         if (err)
3641                 return err;
3642
3643         if (!block_index)
3644                 return 0;
3645
3646         qe->info.binder_type = binder_type;
3647         qe->info.chain_head_change = tcf_chain_head_change_dflt;
3648         qe->info.chain_head_change_priv = &qe->filter_chain;
3649         qe->info.block_index = block_index;
3650
3651         return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3652 }
3653 EXPORT_SYMBOL(tcf_qevent_init);
3654
3655 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3656 {
3657         if (qe->info.block_index)
3658                 tcf_block_put_ext(qe->block, sch, &qe->info);
3659 }
3660 EXPORT_SYMBOL(tcf_qevent_destroy);
3661
3662 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3663                                struct netlink_ext_ack *extack)
3664 {
3665         u32 block_index;
3666         int err;
3667
3668         if (!block_index_attr)
3669                 return 0;
3670
3671         err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3672         if (err)
3673                 return err;
3674
3675         /* Bounce newly-configured block or change in block. */
3676         if (block_index != qe->info.block_index) {
3677                 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3678                 return -EINVAL;
3679         }
3680
3681         return 0;
3682 }
3683 EXPORT_SYMBOL(tcf_qevent_validate_change);
3684
3685 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3686                                   struct sk_buff **to_free, int *ret)
3687 {
3688         struct tcf_result cl_res;
3689         struct tcf_proto *fl;
3690
3691         if (!qe->info.block_index)
3692                 return skb;
3693
3694         fl = rcu_dereference_bh(qe->filter_chain);
3695
3696         switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3697         case TC_ACT_SHOT:
3698                 qdisc_qstats_drop(sch);
3699                 __qdisc_drop(skb, to_free);
3700                 *ret = __NET_XMIT_BYPASS;
3701                 return NULL;
3702         case TC_ACT_STOLEN:
3703         case TC_ACT_QUEUED:
3704         case TC_ACT_TRAP:
3705                 __qdisc_drop(skb, to_free);
3706                 *ret = __NET_XMIT_STOLEN;
3707                 return NULL;
3708         case TC_ACT_REDIRECT:
3709                 skb_do_redirect(skb);
3710                 *ret = __NET_XMIT_STOLEN;
3711                 return NULL;
3712         }
3713
3714         return skb;
3715 }
3716 EXPORT_SYMBOL(tcf_qevent_handle);
3717
3718 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3719 {
3720         if (!qe->info.block_index)
3721                 return 0;
3722         return nla_put_u32(skb, attr_name, qe->info.block_index);
3723 }
3724 EXPORT_SYMBOL(tcf_qevent_dump);
3725 #endif
3726
3727 static __net_init int tcf_net_init(struct net *net)
3728 {
3729         struct tcf_net *tn = net_generic(net, tcf_net_id);
3730
3731         spin_lock_init(&tn->idr_lock);
3732         idr_init(&tn->idr);
3733         return 0;
3734 }
3735
3736 static void __net_exit tcf_net_exit(struct net *net)
3737 {
3738         struct tcf_net *tn = net_generic(net, tcf_net_id);
3739
3740         idr_destroy(&tn->idr);
3741 }
3742
3743 static struct pernet_operations tcf_net_ops = {
3744         .init = tcf_net_init,
3745         .exit = tcf_net_exit,
3746         .id   = &tcf_net_id,
3747         .size = sizeof(struct tcf_net),
3748 };
3749
3750 static int __init tc_filter_init(void)
3751 {
3752         int err;
3753
3754         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3755         if (!tc_filter_wq)
3756                 return -ENOMEM;
3757
3758         err = register_pernet_subsys(&tcf_net_ops);
3759         if (err)
3760                 goto err_register_pernet_subsys;
3761
3762         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3763                       RTNL_FLAG_DOIT_UNLOCKED);
3764         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3765                       RTNL_FLAG_DOIT_UNLOCKED);
3766         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3767                       tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3768         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3769         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3770         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3771                       tc_dump_chain, 0);
3772
3773         return 0;
3774
3775 err_register_pernet_subsys:
3776         destroy_workqueue(tc_filter_wq);
3777         return err;
3778 }
3779
3780 subsys_initcall(tc_filter_init);