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