net/sched/cls_api.c

   1 /*
   2  * net/sched/cls_api.c  Packet classifier API.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *
  11  * Changes:
  12  *
  13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  14  *
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/init.h>
  25 #include <linux/kmod.h>
  26 #include <linux/slab.h>
  27 #include <linux/idr.h>
  28 #include <net/net_namespace.h>
  29 #include <net/sock.h>
  30 #include <net/netlink.h>
  31 #include <net/pkt_sched.h>
  32 #include <net/pkt_cls.h>
  33
  34 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
  35
  36 /* The list of all installed classifier types */
  37 static LIST_HEAD(tcf_proto_base);
  38
  39 /* Protects list of registered TC modules. It is pure SMP lock. */
  40 static DEFINE_RWLOCK(cls_mod_lock);
  41
  42 /* Find classifier type by string name */
  43
  44 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
  45 {
  46         const struct tcf_proto_ops *t, *res = NULL;
  47
  48         if (kind) {
  49                 read_lock(&cls_mod_lock);
  50                 list_for_each_entry(t, &tcf_proto_base, head) {
  51                         if (strcmp(kind, t->kind) == 0) {
  52                                 if (try_module_get(t->owner))
  53                                         res = t;
  54                                 break;
  55                         }
  56                 }
  57                 read_unlock(&cls_mod_lock);
  58         }
  59         return res;
  60 }
  61
  62 static const struct tcf_proto_ops *
  63 tcf_proto_lookup_ops(const char *kind, struct netlink_ext_ack *extack)
  64 {
  65         const struct tcf_proto_ops *ops;
  66
  67         ops = __tcf_proto_lookup_ops(kind);
  68         if (ops)
  69                 return ops;
  70 #ifdef CONFIG_MODULES
  71         rtnl_unlock();
  72         request_module("cls_%s", kind);
  73         rtnl_lock();
  74         ops = __tcf_proto_lookup_ops(kind);
  75         /* We dropped the RTNL semaphore in order to perform
  76          * the module load. So, even if we succeeded in loading
  77          * the module we have to replay the request. We indicate
  78          * this using -EAGAIN.
  79          */
  80         if (ops) {
  81                 module_put(ops->owner);
  82                 return ERR_PTR(-EAGAIN);
  83         }
  84 #endif
  85         NL_SET_ERR_MSG(extack, "TC classifier not found");
  86         return ERR_PTR(-ENOENT);
  87 }
  88
  89 /* Register(unregister) new classifier type */
  90
  91 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  92 {
  93         struct tcf_proto_ops *t;
  94         int rc = -EEXIST;
  95
  96         write_lock(&cls_mod_lock);
  97         list_for_each_entry(t, &tcf_proto_base, head)
  98                 if (!strcmp(ops->kind, t->kind))
  99                         goto out;
 100
 101         list_add_tail(&ops->head, &tcf_proto_base);
 102         rc = 0;
 103 out:
 104         write_unlock(&cls_mod_lock);
 105         return rc;
 106 }
 107 EXPORT_SYMBOL(register_tcf_proto_ops);
 108
 109 static struct workqueue_struct *tc_filter_wq;
 110
 111 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
 112 {
 113         struct tcf_proto_ops *t;
 114         int rc = -ENOENT;
 115
 116         /* Wait for outstanding call_rcu()s, if any, from a
 117          * tcf_proto_ops's destroy() handler.
 118          */
 119         rcu_barrier();
 120         flush_workqueue(tc_filter_wq);
 121
 122         write_lock(&cls_mod_lock);
 123         list_for_each_entry(t, &tcf_proto_base, head) {
 124                 if (t == ops) {
 125                         list_del(&t->head);
 126                         rc = 0;
 127                         break;
 128                 }
 129         }
 130         write_unlock(&cls_mod_lock);
 131         return rc;
 132 }
 133 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 134
 135 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
 136 {
 137         INIT_RCU_WORK(rwork, func);
 138         return queue_rcu_work(tc_filter_wq, rwork);
 139 }
 140 EXPORT_SYMBOL(tcf_queue_work);
 141
 142 /* Select new prio value from the range, managed by kernel. */
 143
 144 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 145 {
 146         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 147
 148         if (tp)
 149                 first = tp->prio - 1;
 150
 151         return TC_H_MAJ(first);
 152 }
 153
 154 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
 155                                           u32 prio, struct tcf_chain *chain,
 156                                           struct netlink_ext_ack *extack)
 157 {
 158         struct tcf_proto *tp;
 159         int err;
 160
 161         tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 162         if (!tp)
 163                 return ERR_PTR(-ENOBUFS);
 164
 165         tp->ops = tcf_proto_lookup_ops(kind, extack);
 166         if (IS_ERR(tp->ops)) {
 167                 err = PTR_ERR(tp->ops);
 168                 goto errout;
 169         }
 170         tp->classify = tp->ops->classify;
 171         tp->protocol = protocol;
 172         tp->prio = prio;
 173         tp->chain = chain;
 174
 175         err = tp->ops->init(tp);
 176         if (err) {
 177                 module_put(tp->ops->owner);
 178                 goto errout;
 179         }
 180         return tp;
 181
 182 errout:
 183         kfree(tp);
 184         return ERR_PTR(err);
 185 }
 186
 187 static void tcf_proto_destroy(struct tcf_proto *tp,
 188                               struct netlink_ext_ack *extack)
 189 {
 190         tp->ops->destroy(tp, extack);
 191         module_put(tp->ops->owner);
 192         kfree_rcu(tp, rcu);
 193 }
 194
 195 struct tcf_filter_chain_list_item {
 196         struct list_head list;
 197         tcf_chain_head_change_t *chain_head_change;
 198         void *chain_head_change_priv;
 199 };
 200
 201 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
 202                                           u32 chain_index)
 203 {
 204         struct tcf_chain *chain;
 205
 206         chain = kzalloc(sizeof(*chain), GFP_KERNEL);
 207         if (!chain)
 208                 return NULL;
 209         list_add_tail(&chain->list, &block->chain_list);
 210         chain->block = block;
 211         chain->index = chain_index;
 212         chain->refcnt = 1;
 213         if (!chain->index)
 214                 block->chain0.chain = chain;
 215         return chain;
 216 }
 217
 218 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
 219                                        struct tcf_proto *tp_head)
 220 {
 221         if (item->chain_head_change)
 222                 item->chain_head_change(tp_head, item->chain_head_change_priv);
 223 }
 224
 225 static void tcf_chain0_head_change(struct tcf_chain *chain,
 226                                    struct tcf_proto *tp_head)
 227 {
 228         struct tcf_filter_chain_list_item *item;
 229         struct tcf_block *block = chain->block;
 230
 231         if (chain->index)
 232                 return;
 233         list_for_each_entry(item, &block->chain0.filter_chain_list, list)
 234                 tcf_chain_head_change_item(item, tp_head);
 235 }
 236
 237 static void tcf_chain_destroy(struct tcf_chain *chain)
 238 {
 239         struct tcf_block *block = chain->block;
 240
 241         list_del(&chain->list);
 242         if (!chain->index)
 243                 block->chain0.chain = NULL;
 244         kfree(chain);
 245         if (list_empty(&block->chain_list) && !refcount_read(&block->refcnt))
 246                 kfree_rcu(block, rcu);
 247 }
 248
 249 static void tcf_chain_hold(struct tcf_chain *chain)
 250 {
 251         ++chain->refcnt;
 252 }
 253
 254 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
 255 {
 256         /* In case all the references are action references, this
 257          * chain should not be shown to the user.
 258          */
 259         return chain->refcnt == chain->action_refcnt;
 260 }
 261
 262 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
 263                                           u32 chain_index)
 264 {
 265         struct tcf_chain *chain;
 266
 267         list_for_each_entry(chain, &block->chain_list, list) {
 268                 if (chain->index == chain_index)
 269                         return chain;
 270         }
 271         return NULL;
 272 }
 273
 274 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
 275                            u32 seq, u16 flags, int event, bool unicast);
 276
 277 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
 278                                          u32 chain_index, bool create,
 279                                          bool by_act)
 280 {
 281         struct tcf_chain *chain = tcf_chain_lookup(block, chain_index);
 282
 283         if (chain) {
 284                 tcf_chain_hold(chain);
 285         } else {
 286                 if (!create)
 287                         return NULL;
 288                 chain = tcf_chain_create(block, chain_index);
 289                 if (!chain)
 290                         return NULL;
 291         }
 292
 293         if (by_act)
 294                 ++chain->action_refcnt;
 295
 296         /* Send notification only in case we got the first
 297          * non-action reference. Until then, the chain acts only as
 298          * a placeholder for actions pointing to it and user ought
 299          * not know about them.
 300          */
 301         if (chain->refcnt - chain->action_refcnt == 1 && !by_act)
 302                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
 303                                 RTM_NEWCHAIN, false);
 304
 305         return chain;
 306 }
 307
 308 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
 309                                        bool create)
 310 {
 311         return __tcf_chain_get(block, chain_index, create, false);
 312 }
 313
 314 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
 315 {
 316         return __tcf_chain_get(block, chain_index, true, true);
 317 }
 318 EXPORT_SYMBOL(tcf_chain_get_by_act);
 319
 320 static void tc_chain_tmplt_del(struct tcf_chain *chain);
 321
 322 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act)
 323 {
 324         if (by_act)
 325                 chain->action_refcnt--;
 326         chain->refcnt--;
 327
 328         /* The last dropped non-action reference will trigger notification. */
 329         if (chain->refcnt - chain->action_refcnt == 0 && !by_act)
 330                 tc_chain_notify(chain, NULL, 0, 0, RTM_DELCHAIN, false);
 331
 332         if (chain->refcnt == 0) {
 333                 tc_chain_tmplt_del(chain);
 334                 tcf_chain_destroy(chain);
 335         }
 336 }
 337
 338 static void tcf_chain_put(struct tcf_chain *chain)
 339 {
 340         __tcf_chain_put(chain, false);
 341 }
 342
 343 void tcf_chain_put_by_act(struct tcf_chain *chain)
 344 {
 345         __tcf_chain_put(chain, true);
 346 }
 347 EXPORT_SYMBOL(tcf_chain_put_by_act);
 348
 349 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
 350 {
 351         if (chain->explicitly_created)
 352                 tcf_chain_put(chain);
 353 }
 354
 355 static void tcf_chain_flush(struct tcf_chain *chain)
 356 {
 357         struct tcf_proto *tp = rtnl_dereference(chain->filter_chain);
 358
 359         tcf_chain0_head_change(chain, NULL);
 360         while (tp) {
 361                 RCU_INIT_POINTER(chain->filter_chain, tp->next);
 362                 tcf_proto_destroy(tp, NULL);
 363                 tp = rtnl_dereference(chain->filter_chain);
 364                 tcf_chain_put(chain);
 365         }
 366 }
 367
 368 static bool tcf_block_offload_in_use(struct tcf_block *block)
 369 {
 370         return block->offloadcnt;
 371 }
 372
 373 static int tcf_block_offload_cmd(struct tcf_block *block,
 374                                  struct net_device *dev,
 375                                  struct tcf_block_ext_info *ei,
 376                                  enum tc_block_command command,
 377                                  struct netlink_ext_ack *extack)
 378 {
 379         struct tc_block_offload bo = {};
 380
 381         bo.command = command;
 382         bo.binder_type = ei->binder_type;
 383         bo.block = block;
 384         bo.extack = extack;
 385         return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
 386 }
 387
 388 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 389                                   struct tcf_block_ext_info *ei,
 390                                   struct netlink_ext_ack *extack)
 391 {
 392         struct net_device *dev = q->dev_queue->dev;
 393         int err;
 394
 395         if (!dev->netdev_ops->ndo_setup_tc)
 396                 goto no_offload_dev_inc;
 397
 398         /* If tc offload feature is disabled and the block we try to bind
 399          * to already has some offloaded filters, forbid to bind.
 400          */
 401         if (!tc_can_offload(dev) && tcf_block_offload_in_use(block)) {
 402                 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
 403                 return -EOPNOTSUPP;
 404         }
 405
 406         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_BIND, extack);
 407         if (err == -EOPNOTSUPP)
 408                 goto no_offload_dev_inc;
 409         return err;
 410
 411 no_offload_dev_inc:
 412         if (tcf_block_offload_in_use(block))
 413                 return -EOPNOTSUPP;
 414         block->nooffloaddevcnt++;
 415         return 0;
 416 }
 417
 418 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 419                                      struct tcf_block_ext_info *ei)
 420 {
 421         struct net_device *dev = q->dev_queue->dev;
 422         int err;
 423
 424         if (!dev->netdev_ops->ndo_setup_tc)
 425                 goto no_offload_dev_dec;
 426         err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_UNBIND, NULL);
 427         if (err == -EOPNOTSUPP)
 428                 goto no_offload_dev_dec;
 429         return;
 430
 431 no_offload_dev_dec:
 432         WARN_ON(block->nooffloaddevcnt-- == 0);
 433 }
 434
 435 static int
 436 tcf_chain0_head_change_cb_add(struct tcf_block *block,
 437                               struct tcf_block_ext_info *ei,
 438                               struct netlink_ext_ack *extack)
 439 {
 440         struct tcf_chain *chain0 = block->chain0.chain;
 441         struct tcf_filter_chain_list_item *item;
 442
 443         item = kmalloc(sizeof(*item), GFP_KERNEL);
 444         if (!item) {
 445                 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
 446                 return -ENOMEM;
 447         }
 448         item->chain_head_change = ei->chain_head_change;
 449         item->chain_head_change_priv = ei->chain_head_change_priv;
 450         if (chain0 && chain0->filter_chain)
 451                 tcf_chain_head_change_item(item, chain0->filter_chain);
 452         list_add(&item->list, &block->chain0.filter_chain_list);
 453         return 0;
 454 }
 455
 456 static void
 457 tcf_chain0_head_change_cb_del(struct tcf_block *block,
 458                               struct tcf_block_ext_info *ei)
 459 {
 460         struct tcf_chain *chain0 = block->chain0.chain;
 461         struct tcf_filter_chain_list_item *item;
 462
 463         list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
 464                 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
 465                     (item->chain_head_change == ei->chain_head_change &&
 466                      item->chain_head_change_priv == ei->chain_head_change_priv)) {
 467                         if (chain0)
 468                                 tcf_chain_head_change_item(item, NULL);
 469                         list_del(&item->list);
 470                         kfree(item);
 471                         return;
 472                 }
 473         }
 474         WARN_ON(1);
 475 }
 476
 477 struct tcf_net {
 478         spinlock_t idr_lock; /* Protects idr */
 479         struct idr idr;
 480 };
 481
 482 static unsigned int tcf_net_id;
 483
 484 static int tcf_block_insert(struct tcf_block *block, struct net *net,
 485                             struct netlink_ext_ack *extack)
 486 {
 487         struct tcf_net *tn = net_generic(net, tcf_net_id);
 488         int err;
 489
 490         idr_preload(GFP_KERNEL);
 491         spin_lock(&tn->idr_lock);
 492         err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
 493                             GFP_NOWAIT);
 494         spin_unlock(&tn->idr_lock);
 495         idr_preload_end();
 496
 497         return err;
 498 }
 499
 500 static void tcf_block_remove(struct tcf_block *block, struct net *net)
 501 {
 502         struct tcf_net *tn = net_generic(net, tcf_net_id);
 503
 504         spin_lock(&tn->idr_lock);
 505         idr_remove(&tn->idr, block->index);
 506         spin_unlock(&tn->idr_lock);
 507 }
 508
 509 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
 510                                           u32 block_index,
 511                                           struct netlink_ext_ack *extack)
 512 {
 513         struct tcf_block *block;
 514
 515         block = kzalloc(sizeof(*block), GFP_KERNEL);
 516         if (!block) {
 517                 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
 518                 return ERR_PTR(-ENOMEM);
 519         }
 520         INIT_LIST_HEAD(&block->chain_list);
 521         INIT_LIST_HEAD(&block->cb_list);
 522         INIT_LIST_HEAD(&block->owner_list);
 523         INIT_LIST_HEAD(&block->chain0.filter_chain_list);
 524
 525         refcount_set(&block->refcnt, 1);
 526         block->net = net;
 527         block->index = block_index;
 528
 529         /* Don't store q pointer for blocks which are shared */
 530         if (!tcf_block_shared(block))
 531                 block->q = q;
 532         return block;
 533 }
 534
 535 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
 536 {
 537         struct tcf_net *tn = net_generic(net, tcf_net_id);
 538
 539         return idr_find(&tn->idr, block_index);
 540 }
 541
 542 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
 543 {
 544         struct tcf_block *block;
 545
 546         rcu_read_lock();
 547         block = tcf_block_lookup(net, block_index);
 548         if (block && !refcount_inc_not_zero(&block->refcnt))
 549                 block = NULL;
 550         rcu_read_unlock();
 551
 552         return block;
 553 }
 554
 555 static void tcf_block_flush_all_chains(struct tcf_block *block)
 556 {
 557         struct tcf_chain *chain;
 558
 559         /* Hold a refcnt for all chains, so that they don't disappear
 560          * while we are iterating.
 561          */
 562         list_for_each_entry(chain, &block->chain_list, list)
 563                 tcf_chain_hold(chain);
 564
 565         list_for_each_entry(chain, &block->chain_list, list)
 566                 tcf_chain_flush(chain);
 567 }
 568
 569 static void tcf_block_put_all_chains(struct tcf_block *block)
 570 {
 571         struct tcf_chain *chain, *tmp;
 572
 573         /* At this point, all the chains should have refcnt >= 1. */
 574         list_for_each_entry_safe(chain, tmp, &block->chain_list, list) {
 575                 tcf_chain_put_explicitly_created(chain);
 576                 tcf_chain_put(chain);
 577         }
 578 }
 579
 580 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
 581                             struct tcf_block_ext_info *ei)
 582 {
 583         if (refcount_dec_and_test(&block->refcnt)) {
 584                 /* Flushing/putting all chains will cause the block to be
 585                  * deallocated when last chain is freed. However, if chain_list
 586                  * is empty, block has to be manually deallocated. After block
 587                  * reference counter reached 0, it is no longer possible to
 588                  * increment it or add new chains to block.
 589                  */
 590                 bool free_block = list_empty(&block->chain_list);
 591
 592                 if (tcf_block_shared(block))
 593                         tcf_block_remove(block, block->net);
 594                 if (!free_block)
 595                         tcf_block_flush_all_chains(block);
 596
 597                 if (q)
 598                         tcf_block_offload_unbind(block, q, ei);
 599
 600                 if (free_block)
 601                         kfree_rcu(block, rcu);
 602                 else
 603                         tcf_block_put_all_chains(block);
 604         } else if (q) {
 605                 tcf_block_offload_unbind(block, q, ei);
 606         }
 607 }
 608
 609 static void tcf_block_refcnt_put(struct tcf_block *block)
 610 {
 611         __tcf_block_put(block, NULL, NULL);
 612 }
 613
 614 /* Find tcf block.
 615  * Set q, parent, cl when appropriate.
 616  */
 617
 618 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
 619                                         u32 *parent, unsigned long *cl,
 620                                         int ifindex, u32 block_index,
 621                                         struct netlink_ext_ack *extack)
 622 {
 623         struct tcf_block *block;
 624         int err = 0;
 625
 626         if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
 627                 block = tcf_block_refcnt_get(net, block_index);
 628                 if (!block) {
 629                         NL_SET_ERR_MSG(extack, "Block of given index was not found");
 630                         return ERR_PTR(-EINVAL);
 631                 }
 632         } else {
 633                 const struct Qdisc_class_ops *cops;
 634                 struct net_device *dev;
 635
 636                 rcu_read_lock();
 637
 638                 /* Find link */
 639                 dev = dev_get_by_index_rcu(net, ifindex);
 640                 if (!dev) {
 641                         rcu_read_unlock();
 642                         return ERR_PTR(-ENODEV);
 643                 }
 644
 645                 /* Find qdisc */
 646                 if (!*parent) {
 647                         *q = dev->qdisc;
 648                         *parent = (*q)->handle;
 649                 } else {
 650                         *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
 651                         if (!*q) {
 652                                 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
 653                                 err = -EINVAL;
 654                                 goto errout_rcu;
 655                         }
 656                 }
 657
 658                 *q = qdisc_refcount_inc_nz(*q);
 659                 if (!*q) {
 660                         NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
 661                         err = -EINVAL;
 662                         goto errout_rcu;
 663                 }
 664
 665                 /* Is it classful? */
 666                 cops = (*q)->ops->cl_ops;
 667                 if (!cops) {
 668                         NL_SET_ERR_MSG(extack, "Qdisc not classful");
 669                         err = -EINVAL;
 670                         goto errout_rcu;
 671                 }
 672
 673                 if (!cops->tcf_block) {
 674                         NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
 675                         err = -EOPNOTSUPP;
 676                         goto errout_rcu;
 677                 }
 678
 679                 /* At this point we know that qdisc is not noop_qdisc,
 680                  * which means that qdisc holds a reference to net_device
 681                  * and we hold a reference to qdisc, so it is safe to release
 682                  * rcu read lock.
 683                  */
 684                 rcu_read_unlock();
 685
 686                 /* Do we search for filter, attached to class? */
 687                 if (TC_H_MIN(*parent)) {
 688                         *cl = cops->find(*q, *parent);
 689                         if (*cl == 0) {
 690                                 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
 691                                 err = -ENOENT;
 692                                 goto errout_qdisc;
 693                         }
 694                 }
 695
 696                 /* And the last stroke */
 697                 block = cops->tcf_block(*q, *cl, extack);
 698                 if (!block) {
 699                         err = -EINVAL;
 700                         goto errout_qdisc;
 701                 }
 702                 if (tcf_block_shared(block)) {
 703                         NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
 704                         err = -EOPNOTSUPP;
 705                         goto errout_qdisc;
 706                 }
 707
 708                 /* Always take reference to block in order to support execution
 709                  * of rules update path of cls API without rtnl lock. Caller
 710                  * must release block when it is finished using it. 'if' block
 711                  * of this conditional obtain reference to block by calling
 712                  * tcf_block_refcnt_get().
 713                  */
 714                 refcount_inc(&block->refcnt);
 715         }
 716
 717         return block;
 718
 719 errout_rcu:
 720         rcu_read_unlock();
 721 errout_qdisc:
 722         if (*q) {
 723                 qdisc_put(*q);
 724                 *q = NULL;
 725         }
 726         return ERR_PTR(err);
 727 }
 728
 729 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block)
 730 {
 731         if (!IS_ERR_OR_NULL(block))
 732                 tcf_block_refcnt_put(block);
 733
 734         if (q)
 735                 qdisc_put(q);
 736 }
 737
 738 struct tcf_block_owner_item {
 739         struct list_head list;
 740         struct Qdisc *q;
 741         enum tcf_block_binder_type binder_type;
 742 };
 743
 744 static void
 745 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
 746                                struct Qdisc *q,
 747                                enum tcf_block_binder_type binder_type)
 748 {
 749         if (block->keep_dst &&
 750             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
 751             binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
 752                 netif_keep_dst(qdisc_dev(q));
 753 }
 754
 755 void tcf_block_netif_keep_dst(struct tcf_block *block)
 756 {
 757         struct tcf_block_owner_item *item;
 758
 759         block->keep_dst = true;
 760         list_for_each_entry(item, &block->owner_list, list)
 761                 tcf_block_owner_netif_keep_dst(block, item->q,
 762                                                item->binder_type);
 763 }
 764 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
 765
 766 static int tcf_block_owner_add(struct tcf_block *block,
 767                                struct Qdisc *q,
 768                                enum tcf_block_binder_type binder_type)
 769 {
 770         struct tcf_block_owner_item *item;
 771
 772         item = kmalloc(sizeof(*item), GFP_KERNEL);
 773         if (!item)
 774                 return -ENOMEM;
 775         item->q = q;
 776         item->binder_type = binder_type;
 777         list_add(&item->list, &block->owner_list);
 778         return 0;
 779 }
 780
 781 static void tcf_block_owner_del(struct tcf_block *block,
 782                                 struct Qdisc *q,
 783                                 enum tcf_block_binder_type binder_type)
 784 {
 785         struct tcf_block_owner_item *item;
 786
 787         list_for_each_entry(item, &block->owner_list, list) {
 788                 if (item->q == q && item->binder_type == binder_type) {
 789                         list_del(&item->list);
 790                         kfree(item);
 791                         return;
 792                 }
 793         }
 794         WARN_ON(1);
 795 }
 796
 797 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 798                       struct tcf_block_ext_info *ei,
 799                       struct netlink_ext_ack *extack)
 800 {
 801         struct net *net = qdisc_net(q);
 802         struct tcf_block *block = NULL;
 803         int err;
 804
 805         if (ei->block_index)
 806                 /* block_index not 0 means the shared block is requested */
 807                 block = tcf_block_refcnt_get(net, ei->block_index);
 808
 809         if (!block) {
 810                 block = tcf_block_create(net, q, ei->block_index, extack);
 811                 if (IS_ERR(block))
 812                         return PTR_ERR(block);
 813                 if (tcf_block_shared(block)) {
 814                         err = tcf_block_insert(block, net, extack);
 815                         if (err)
 816                                 goto err_block_insert;
 817                 }
 818         }
 819
 820         err = tcf_block_owner_add(block, q, ei->binder_type);
 821         if (err)
 822                 goto err_block_owner_add;
 823
 824         tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
 825
 826         err = tcf_chain0_head_change_cb_add(block, ei, extack);
 827         if (err)
 828                 goto err_chain0_head_change_cb_add;
 829
 830         err = tcf_block_offload_bind(block, q, ei, extack);
 831         if (err)
 832                 goto err_block_offload_bind;
 833
 834         *p_block = block;
 835         return 0;
 836
 837 err_block_offload_bind:
 838         tcf_chain0_head_change_cb_del(block, ei);
 839 err_chain0_head_change_cb_add:
 840         tcf_block_owner_del(block, q, ei->binder_type);
 841 err_block_owner_add:
 842 err_block_insert:
 843         tcf_block_refcnt_put(block);
 844         return err;
 845 }
 846 EXPORT_SYMBOL(tcf_block_get_ext);
 847
 848 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
 849 {
 850         struct tcf_proto __rcu **p_filter_chain = priv;
 851
 852         rcu_assign_pointer(*p_filter_chain, tp_head);
 853 }
 854
 855 int tcf_block_get(struct tcf_block **p_block,
 856                   struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
 857                   struct netlink_ext_ack *extack)
 858 {
 859         struct tcf_block_ext_info ei = {
 860                 .chain_head_change = tcf_chain_head_change_dflt,
 861                 .chain_head_change_priv = p_filter_chain,
 862         };
 863
 864         WARN_ON(!p_filter_chain);
 865         return tcf_block_get_ext(p_block, q, &ei, extack);
 866 }
 867 EXPORT_SYMBOL(tcf_block_get);
 868
 869 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
 870  * actions should be all removed after flushing.
 871  */
 872 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 873                        struct tcf_block_ext_info *ei)
 874 {
 875         if (!block)
 876                 return;
 877         tcf_chain0_head_change_cb_del(block, ei);
 878         tcf_block_owner_del(block, q, ei->binder_type);
 879
 880         __tcf_block_put(block, q, ei);
 881 }
 882 EXPORT_SYMBOL(tcf_block_put_ext);
 883
 884 void tcf_block_put(struct tcf_block *block)
 885 {
 886         struct tcf_block_ext_info ei = {0, };
 887
 888         if (!block)
 889                 return;
 890         tcf_block_put_ext(block, block->q, &ei);
 891 }
 892
 893 EXPORT_SYMBOL(tcf_block_put);
 894
 895 struct tcf_block_cb {
 896         struct list_head list;
 897         tc_setup_cb_t *cb;
 898         void *cb_ident;
 899         void *cb_priv;
 900         unsigned int refcnt;
 901 };
 902
 903 void *tcf_block_cb_priv(struct tcf_block_cb *block_cb)
 904 {
 905         return block_cb->cb_priv;
 906 }
 907 EXPORT_SYMBOL(tcf_block_cb_priv);
 908
 909 struct tcf_block_cb *tcf_block_cb_lookup(struct tcf_block *block,
 910                                          tc_setup_cb_t *cb, void *cb_ident)
 911 {       struct tcf_block_cb *block_cb;
 912
 913         list_for_each_entry(block_cb, &block->cb_list, list)
 914                 if (block_cb->cb == cb && block_cb->cb_ident == cb_ident)
 915                         return block_cb;
 916         return NULL;
 917 }
 918 EXPORT_SYMBOL(tcf_block_cb_lookup);
 919
 920 void tcf_block_cb_incref(struct tcf_block_cb *block_cb)
 921 {
 922         block_cb->refcnt++;
 923 }
 924 EXPORT_SYMBOL(tcf_block_cb_incref);
 925
 926 unsigned int tcf_block_cb_decref(struct tcf_block_cb *block_cb)
 927 {
 928         return --block_cb->refcnt;
 929 }
 930 EXPORT_SYMBOL(tcf_block_cb_decref);
 931
 932 static int
 933 tcf_block_playback_offloads(struct tcf_block *block, tc_setup_cb_t *cb,
 934                             void *cb_priv, bool add, bool offload_in_use,
 935                             struct netlink_ext_ack *extack)
 936 {
 937         struct tcf_chain *chain;
 938         struct tcf_proto *tp;
 939         int err;
 940
 941         list_for_each_entry(chain, &block->chain_list, list) {
 942                 for (tp = rtnl_dereference(chain->filter_chain); tp;
 943                      tp = rtnl_dereference(tp->next)) {
 944                         if (tp->ops->reoffload) {
 945                                 err = tp->ops->reoffload(tp, add, cb, cb_priv,
 946                                                          extack);
 947                                 if (err && add)
 948                                         goto err_playback_remove;
 949                         } else if (add && offload_in_use) {
 950                                 err = -EOPNOTSUPP;
 951                                 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
 952                                 goto err_playback_remove;
 953                         }
 954                 }
 955         }
 956
 957         return 0;
 958
 959 err_playback_remove:
 960         tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
 961                                     extack);
 962         return err;
 963 }
 964
 965 struct tcf_block_cb *__tcf_block_cb_register(struct tcf_block *block,
 966                                              tc_setup_cb_t *cb, void *cb_ident,
 967                                              void *cb_priv,
 968                                              struct netlink_ext_ack *extack)
 969 {
 970         struct tcf_block_cb *block_cb;
 971         int err;
 972
 973         /* Replay any already present rules */
 974         err = tcf_block_playback_offloads(block, cb, cb_priv, true,
 975                                           tcf_block_offload_in_use(block),
 976                                           extack);
 977         if (err)
 978                 return ERR_PTR(err);
 979
 980         block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
 981         if (!block_cb)
 982                 return ERR_PTR(-ENOMEM);
 983         block_cb->cb = cb;
 984         block_cb->cb_ident = cb_ident;
 985         block_cb->cb_priv = cb_priv;
 986         list_add(&block_cb->list, &block->cb_list);
 987         return block_cb;
 988 }
 989 EXPORT_SYMBOL(__tcf_block_cb_register);
 990
 991 int tcf_block_cb_register(struct tcf_block *block,
 992                           tc_setup_cb_t *cb, void *cb_ident,
 993                           void *cb_priv, struct netlink_ext_ack *extack)
 994 {
 995         struct tcf_block_cb *block_cb;
 996
 997         block_cb = __tcf_block_cb_register(block, cb, cb_ident, cb_priv,
 998                                            extack);
 999         return PTR_ERR_OR_ZERO(block_cb);
1000 }
1001 EXPORT_SYMBOL(tcf_block_cb_register);
1002
1003 void __tcf_block_cb_unregister(struct tcf_block *block,
1004                                struct tcf_block_cb *block_cb)
1005 {
1006         tcf_block_playback_offloads(block, block_cb->cb, block_cb->cb_priv,
1007                                     false, tcf_block_offload_in_use(block),
1008                                     NULL);
1009         list_del(&block_cb->list);
1010         kfree(block_cb);
1011 }
1012 EXPORT_SYMBOL(__tcf_block_cb_unregister);
1013
1014 void tcf_block_cb_unregister(struct tcf_block *block,
1015                              tc_setup_cb_t *cb, void *cb_ident)
1016 {
1017         struct tcf_block_cb *block_cb;
1018
1019         block_cb = tcf_block_cb_lookup(block, cb, cb_ident);
1020         if (!block_cb)
1021                 return;
1022         __tcf_block_cb_unregister(block, block_cb);
1023 }
1024 EXPORT_SYMBOL(tcf_block_cb_unregister);
1025
1026 static int tcf_block_cb_call(struct tcf_block *block, enum tc_setup_type type,
1027                              void *type_data, bool err_stop)
1028 {
1029         struct tcf_block_cb *block_cb;
1030         int ok_count = 0;
1031         int err;
1032
1033         /* Make sure all netdevs sharing this block are offload-capable. */
1034         if (block->nooffloaddevcnt && err_stop)
1035                 return -EOPNOTSUPP;
1036
1037         list_for_each_entry(block_cb, &block->cb_list, list) {
1038                 err = block_cb->cb(type, type_data, block_cb->cb_priv);
1039                 if (err) {
1040                         if (err_stop)
1041                                 return err;
1042                 } else {
1043                         ok_count++;
1044                 }
1045         }
1046         return ok_count;
1047 }
1048
1049 /* Main classifier routine: scans classifier chain attached
1050  * to this qdisc, (optionally) tests for protocol and asks
1051  * specific classifiers.
1052  */
1053 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1054                  struct tcf_result *res, bool compat_mode)
1055 {
1056         __be16 protocol = tc_skb_protocol(skb);
1057 #ifdef CONFIG_NET_CLS_ACT
1058         const int max_reclassify_loop = 4;
1059         const struct tcf_proto *orig_tp = tp;
1060         const struct tcf_proto *first_tp;
1061         int limit = 0;
1062
1063 reclassify:
1064 #endif
1065         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1066                 int err;
1067
1068                 if (tp->protocol != protocol &&
1069                     tp->protocol != htons(ETH_P_ALL))
1070                         continue;
1071
1072                 err = tp->classify(skb, tp, res);
1073 #ifdef CONFIG_NET_CLS_ACT
1074                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1075                         first_tp = orig_tp;
1076                         goto reset;
1077                 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1078                         first_tp = res->goto_tp;
1079                         goto reset;
1080                 }
1081 #endif
1082                 if (err >= 0)
1083                         return err;
1084         }
1085
1086         return TC_ACT_UNSPEC; /* signal: continue lookup */
1087 #ifdef CONFIG_NET_CLS_ACT
1088 reset:
1089         if (unlikely(limit++ >= max_reclassify_loop)) {
1090                 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1091                                        tp->chain->block->index,
1092                                        tp->prio & 0xffff,
1093                                        ntohs(tp->protocol));
1094                 return TC_ACT_SHOT;
1095         }
1096
1097         tp = first_tp;
1098         protocol = tc_skb_protocol(skb);
1099         goto reclassify;
1100 #endif
1101 }
1102 EXPORT_SYMBOL(tcf_classify);
1103
1104 struct tcf_chain_info {
1105         struct tcf_proto __rcu **pprev;
1106         struct tcf_proto __rcu *next;
1107 };
1108
1109 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain_info *chain_info)
1110 {
1111         return rtnl_dereference(*chain_info->pprev);
1112 }
1113
1114 static void tcf_chain_tp_insert(struct tcf_chain *chain,
1115                                 struct tcf_chain_info *chain_info,
1116                                 struct tcf_proto *tp)
1117 {
1118         if (*chain_info->pprev == chain->filter_chain)
1119                 tcf_chain0_head_change(chain, tp);
1120         RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain_info));
1121         rcu_assign_pointer(*chain_info->pprev, tp);
1122         tcf_chain_hold(chain);
1123 }
1124
1125 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1126                                 struct tcf_chain_info *chain_info,
1127                                 struct tcf_proto *tp)
1128 {
1129         struct tcf_proto *next = rtnl_dereference(chain_info->next);
1130
1131         if (tp == chain->filter_chain)
1132                 tcf_chain0_head_change(chain, next);
1133         RCU_INIT_POINTER(*chain_info->pprev, next);
1134         tcf_chain_put(chain);
1135 }
1136
1137 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1138                                            struct tcf_chain_info *chain_info,
1139                                            u32 protocol, u32 prio,
1140                                            bool prio_allocate)
1141 {
1142         struct tcf_proto **pprev;
1143         struct tcf_proto *tp;
1144
1145         /* Check the chain for existence of proto-tcf with this priority */
1146         for (pprev = &chain->filter_chain;
1147              (tp = rtnl_dereference(*pprev)); pprev = &tp->next) {
1148                 if (tp->prio >= prio) {
1149                         if (tp->prio == prio) {
1150                                 if (prio_allocate ||
1151                                     (tp->protocol != protocol && protocol))
1152                                         return ERR_PTR(-EINVAL);
1153                         } else {
1154                                 tp = NULL;
1155                         }
1156                         break;
1157                 }
1158         }
1159         chain_info->pprev = pprev;
1160         chain_info->next = tp ? tp->next : NULL;
1161         return tp;
1162 }
1163
1164 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1165                          struct tcf_proto *tp, struct tcf_block *block,
1166                          struct Qdisc *q, u32 parent, void *fh,
1167                          u32 portid, u32 seq, u16 flags, int event)
1168 {
1169         struct tcmsg *tcm;
1170         struct nlmsghdr  *nlh;
1171         unsigned char *b = skb_tail_pointer(skb);
1172
1173         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1174         if (!nlh)
1175                 goto out_nlmsg_trim;
1176         tcm = nlmsg_data(nlh);
1177         tcm->tcm_family = AF_UNSPEC;
1178         tcm->tcm__pad1 = 0;
1179         tcm->tcm__pad2 = 0;
1180         if (q) {
1181                 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1182                 tcm->tcm_parent = parent;
1183         } else {
1184                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1185                 tcm->tcm_block_index = block->index;
1186         }
1187         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1188         if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1189                 goto nla_put_failure;
1190         if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1191                 goto nla_put_failure;
1192         if (!fh) {
1193                 tcm->tcm_handle = 0;
1194         } else {
1195                 if (tp->ops->dump && tp->ops->dump(net, tp, fh, skb, tcm) < 0)
1196                         goto nla_put_failure;
1197         }
1198         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1199         return skb->len;
1200
1201 out_nlmsg_trim:
1202 nla_put_failure:
1203         nlmsg_trim(skb, b);
1204         return -1;
1205 }
1206
1207 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1208                           struct nlmsghdr *n, struct tcf_proto *tp,
1209                           struct tcf_block *block, struct Qdisc *q,
1210                           u32 parent, void *fh, int event, bool unicast)
1211 {
1212         struct sk_buff *skb;
1213         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1214
1215         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1216         if (!skb)
1217                 return -ENOBUFS;
1218
1219         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1220                           n->nlmsg_seq, n->nlmsg_flags, event) <= 0) {
1221                 kfree_skb(skb);
1222                 return -EINVAL;
1223         }
1224
1225         if (unicast)
1226                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1227
1228         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1229                               n->nlmsg_flags & NLM_F_ECHO);
1230 }
1231
1232 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1233                               struct nlmsghdr *n, struct tcf_proto *tp,
1234                               struct tcf_block *block, struct Qdisc *q,
1235                               u32 parent, void *fh, bool unicast, bool *last,
1236                               struct netlink_ext_ack *extack)
1237 {
1238         struct sk_buff *skb;
1239         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1240         int err;
1241
1242         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1243         if (!skb)
1244                 return -ENOBUFS;
1245
1246         if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
1247                           n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER) <= 0) {
1248                 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
1249                 kfree_skb(skb);
1250                 return -EINVAL;
1251         }
1252
1253         err = tp->ops->delete(tp, fh, last, extack);
1254         if (err) {
1255                 kfree_skb(skb);
1256                 return err;
1257         }
1258
1259         if (unicast)
1260                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1261
1262         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1263                              n->nlmsg_flags & NLM_F_ECHO);
1264         if (err < 0)
1265                 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
1266         return err;
1267 }
1268
1269 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
1270                                  struct tcf_block *block, struct Qdisc *q,
1271                                  u32 parent, struct nlmsghdr *n,
1272                                  struct tcf_chain *chain, int event)
1273 {
1274         struct tcf_proto *tp;
1275
1276         for (tp = rtnl_dereference(chain->filter_chain);
1277              tp; tp = rtnl_dereference(tp->next))
1278                 tfilter_notify(net, oskb, n, tp, block,
1279                                q, parent, NULL, event, false);
1280 }
1281
1282 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1283                           struct netlink_ext_ack *extack)
1284 {
1285         struct net *net = sock_net(skb->sk);
1286         struct nlattr *tca[TCA_MAX + 1];
1287         struct tcmsg *t;
1288         u32 protocol;
1289         u32 prio;
1290         bool prio_allocate;
1291         u32 parent;
1292         u32 chain_index;
1293         struct Qdisc *q = NULL;
1294         struct tcf_chain_info chain_info;
1295         struct tcf_chain *chain = NULL;
1296         struct tcf_block *block;
1297         struct tcf_proto *tp;
1298         unsigned long cl;
1299         void *fh;
1300         int err;
1301         int tp_created;
1302
1303         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1304                 return -EPERM;
1305
1306 replay:
1307         tp_created = 0;
1308
1309         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1310         if (err < 0)
1311                 return err;
1312
1313         t = nlmsg_data(n);
1314         protocol = TC_H_MIN(t->tcm_info);
1315         prio = TC_H_MAJ(t->tcm_info);
1316         prio_allocate = false;
1317         parent = t->tcm_parent;
1318         cl = 0;
1319
1320         if (prio == 0) {
1321                 /* If no priority is provided by the user,
1322                  * we allocate one.
1323                  */
1324                 if (n->nlmsg_flags & NLM_F_CREATE) {
1325                         prio = TC_H_MAKE(0x80000000U, 0U);
1326                         prio_allocate = true;
1327                 } else {
1328                         NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1329                         return -ENOENT;
1330                 }
1331         }
1332
1333         /* Find head of filter chain. */
1334
1335         block = tcf_block_find(net, &q, &parent, &cl,
1336                                t->tcm_ifindex, t->tcm_block_index, extack);
1337         if (IS_ERR(block)) {
1338                 err = PTR_ERR(block);
1339                 goto errout;
1340         }
1341
1342         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1343         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1344                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1345                 err = -EINVAL;
1346                 goto errout;
1347         }
1348         chain = tcf_chain_get(block, chain_index, true);
1349         if (!chain) {
1350                 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
1351                 err = -ENOMEM;
1352                 goto errout;
1353         }
1354
1355         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1356                                prio, prio_allocate);
1357         if (IS_ERR(tp)) {
1358                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1359                 err = PTR_ERR(tp);
1360                 goto errout;
1361         }
1362
1363         if (tp == NULL) {
1364                 /* Proto-tcf does not exist, create new one */
1365
1366                 if (tca[TCA_KIND] == NULL || !protocol) {
1367                         NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
1368                         err = -EINVAL;
1369                         goto errout;
1370                 }
1371
1372                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1373                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1374                         err = -ENOENT;
1375                         goto errout;
1376                 }
1377
1378                 if (prio_allocate)
1379                         prio = tcf_auto_prio(tcf_chain_tp_prev(&chain_info));
1380
1381                 tp = tcf_proto_create(nla_data(tca[TCA_KIND]),
1382                                       protocol, prio, chain, extack);
1383                 if (IS_ERR(tp)) {
1384                         err = PTR_ERR(tp);
1385                         goto errout;
1386                 }
1387                 tp_created = 1;
1388         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1389                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1390                 err = -EINVAL;
1391                 goto errout;
1392         }
1393
1394         fh = tp->ops->get(tp, t->tcm_handle);
1395
1396         if (!fh) {
1397                 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1398                         NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
1399                         err = -ENOENT;
1400                         goto errout;
1401                 }
1402         } else if (n->nlmsg_flags & NLM_F_EXCL) {
1403                 NL_SET_ERR_MSG(extack, "Filter already exists");
1404                 err = -EEXIST;
1405                 goto errout;
1406         }
1407
1408         if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
1409                 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
1410                 err = -EINVAL;
1411                 goto errout;
1412         }
1413
1414         err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
1415                               n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
1416                               extack);
1417         if (err == 0) {
1418                 if (tp_created)
1419                         tcf_chain_tp_insert(chain, &chain_info, tp);
1420                 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1421                                RTM_NEWTFILTER, false);
1422         } else {
1423                 if (tp_created)
1424                         tcf_proto_destroy(tp, NULL);
1425         }
1426
1427 errout:
1428         if (chain)
1429                 tcf_chain_put(chain);
1430         tcf_block_release(q, block);
1431         if (err == -EAGAIN)
1432                 /* Replay the request. */
1433                 goto replay;
1434         return err;
1435 }
1436
1437 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1438                           struct netlink_ext_ack *extack)
1439 {
1440         struct net *net = sock_net(skb->sk);
1441         struct nlattr *tca[TCA_MAX + 1];
1442         struct tcmsg *t;
1443         u32 protocol;
1444         u32 prio;
1445         u32 parent;
1446         u32 chain_index;
1447         struct Qdisc *q = NULL;
1448         struct tcf_chain_info chain_info;
1449         struct tcf_chain *chain = NULL;
1450         struct tcf_block *block;
1451         struct tcf_proto *tp = NULL;
1452         unsigned long cl = 0;
1453         void *fh = NULL;
1454         int err;
1455
1456         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1457                 return -EPERM;
1458
1459         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1460         if (err < 0)
1461                 return err;
1462
1463         t = nlmsg_data(n);
1464         protocol = TC_H_MIN(t->tcm_info);
1465         prio = TC_H_MAJ(t->tcm_info);
1466         parent = t->tcm_parent;
1467
1468         if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
1469                 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
1470                 return -ENOENT;
1471         }
1472
1473         /* Find head of filter chain. */
1474
1475         block = tcf_block_find(net, &q, &parent, &cl,
1476                                t->tcm_ifindex, t->tcm_block_index, extack);
1477         if (IS_ERR(block)) {
1478                 err = PTR_ERR(block);
1479                 goto errout;
1480         }
1481
1482         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1483         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1484                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1485                 err = -EINVAL;
1486                 goto errout;
1487         }
1488         chain = tcf_chain_get(block, chain_index, false);
1489         if (!chain) {
1490                 /* User requested flush on non-existent chain. Nothing to do,
1491                  * so just return success.
1492                  */
1493                 if (prio == 0) {
1494                         err = 0;
1495                         goto errout;
1496                 }
1497                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1498                 err = -ENOENT;
1499                 goto errout;
1500         }
1501
1502         if (prio == 0) {
1503                 tfilter_notify_chain(net, skb, block, q, parent, n,
1504                                      chain, RTM_DELTFILTER);
1505                 tcf_chain_flush(chain);
1506                 err = 0;
1507                 goto errout;
1508         }
1509
1510         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1511                                prio, false);
1512         if (!tp || IS_ERR(tp)) {
1513                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1514                 err = tp ? PTR_ERR(tp) : -ENOENT;
1515                 goto errout;
1516         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1517                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1518                 err = -EINVAL;
1519                 goto errout;
1520         }
1521
1522         fh = tp->ops->get(tp, t->tcm_handle);
1523
1524         if (!fh) {
1525                 if (t->tcm_handle == 0) {
1526                         tcf_chain_tp_remove(chain, &chain_info, tp);
1527                         tfilter_notify(net, skb, n, tp, block, q, parent, fh,
1528                                        RTM_DELTFILTER, false);
1529                         tcf_proto_destroy(tp, extack);
1530                         err = 0;
1531                 } else {
1532                         NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1533                         err = -ENOENT;
1534                 }
1535         } else {
1536                 bool last;
1537
1538                 err = tfilter_del_notify(net, skb, n, tp, block,
1539                                          q, parent, fh, false, &last,
1540                                          extack);
1541                 if (err)
1542                         goto errout;
1543                 if (last) {
1544                         tcf_chain_tp_remove(chain, &chain_info, tp);
1545                         tcf_proto_destroy(tp, extack);
1546                 }
1547         }
1548
1549 errout:
1550         if (chain)
1551                 tcf_chain_put(chain);
1552         tcf_block_release(q, block);
1553         return err;
1554 }
1555
1556 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
1557                           struct netlink_ext_ack *extack)
1558 {
1559         struct net *net = sock_net(skb->sk);
1560         struct nlattr *tca[TCA_MAX + 1];
1561         struct tcmsg *t;
1562         u32 protocol;
1563         u32 prio;
1564         u32 parent;
1565         u32 chain_index;
1566         struct Qdisc *q = NULL;
1567         struct tcf_chain_info chain_info;
1568         struct tcf_chain *chain = NULL;
1569         struct tcf_block *block;
1570         struct tcf_proto *tp = NULL;
1571         unsigned long cl = 0;
1572         void *fh = NULL;
1573         int err;
1574
1575         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1576         if (err < 0)
1577                 return err;
1578
1579         t = nlmsg_data(n);
1580         protocol = TC_H_MIN(t->tcm_info);
1581         prio = TC_H_MAJ(t->tcm_info);
1582         parent = t->tcm_parent;
1583
1584         if (prio == 0) {
1585                 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1586                 return -ENOENT;
1587         }
1588
1589         /* Find head of filter chain. */
1590
1591         block = tcf_block_find(net, &q, &parent, &cl,
1592                                t->tcm_ifindex, t->tcm_block_index, extack);
1593         if (IS_ERR(block)) {
1594                 err = PTR_ERR(block);
1595                 goto errout;
1596         }
1597
1598         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1599         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1600                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1601                 err = -EINVAL;
1602                 goto errout;
1603         }
1604         chain = tcf_chain_get(block, chain_index, false);
1605         if (!chain) {
1606                 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1607                 err = -EINVAL;
1608                 goto errout;
1609         }
1610
1611         tp = tcf_chain_tp_find(chain, &chain_info, protocol,
1612                                prio, false);
1613         if (!tp || IS_ERR(tp)) {
1614                 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
1615                 err = tp ? PTR_ERR(tp) : -ENOENT;
1616                 goto errout;
1617         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
1618                 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
1619                 err = -EINVAL;
1620                 goto errout;
1621         }
1622
1623         fh = tp->ops->get(tp, t->tcm_handle);
1624
1625         if (!fh) {
1626                 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
1627                 err = -ENOENT;
1628         } else {
1629                 err = tfilter_notify(net, skb, n, tp, block, q, parent,
1630                                      fh, RTM_NEWTFILTER, true);
1631                 if (err < 0)
1632                         NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
1633         }
1634
1635 errout:
1636         if (chain)
1637                 tcf_chain_put(chain);
1638         tcf_block_release(q, block);
1639         return err;
1640 }
1641
1642 struct tcf_dump_args {
1643         struct tcf_walker w;
1644         struct sk_buff *skb;
1645         struct netlink_callback *cb;
1646         struct tcf_block *block;
1647         struct Qdisc *q;
1648         u32 parent;
1649 };
1650
1651 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1652 {
1653         struct tcf_dump_args *a = (void *)arg;
1654         struct net *net = sock_net(a->skb->sk);
1655
1656         return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
1657                              n, NETLINK_CB(a->cb->skb).portid,
1658                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1659                              RTM_NEWTFILTER);
1660 }
1661
1662 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
1663                            struct sk_buff *skb, struct netlink_callback *cb,
1664                            long index_start, long *p_index)
1665 {
1666         struct net *net = sock_net(skb->sk);
1667         struct tcf_block *block = chain->block;
1668         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1669         struct tcf_dump_args arg;
1670         struct tcf_proto *tp;
1671
1672         for (tp = rtnl_dereference(chain->filter_chain);
1673              tp; tp = rtnl_dereference(tp->next), (*p_index)++) {
1674                 if (*p_index < index_start)
1675                         continue;
1676                 if (TC_H_MAJ(tcm->tcm_info) &&
1677                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
1678                         continue;
1679                 if (TC_H_MIN(tcm->tcm_info) &&
1680                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
1681                         continue;
1682                 if (*p_index > index_start)
1683                         memset(&cb->args[1], 0,
1684                                sizeof(cb->args) - sizeof(cb->args[0]));
1685                 if (cb->args[1] == 0) {
1686                         if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
1687                                           NETLINK_CB(cb->skb).portid,
1688                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
1689                                           RTM_NEWTFILTER) <= 0)
1690                                 return false;
1691
1692                         cb->args[1] = 1;
1693                 }
1694                 if (!tp->ops->walk)
1695                         continue;
1696                 arg.w.fn = tcf_node_dump;
1697                 arg.skb = skb;
1698                 arg.cb = cb;
1699                 arg.block = block;
1700                 arg.q = q;
1701                 arg.parent = parent;
1702                 arg.w.stop = 0;
1703                 arg.w.skip = cb->args[1] - 1;
1704                 arg.w.count = 0;
1705                 arg.w.cookie = cb->args[2];
1706                 tp->ops->walk(tp, &arg.w);
1707                 cb->args[2] = arg.w.cookie;
1708                 cb->args[1] = arg.w.count + 1;
1709                 if (arg.w.stop)
1710                         return false;
1711         }
1712         return true;
1713 }
1714
1715 /* called with RTNL */
1716 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
1717 {
1718         struct net *net = sock_net(skb->sk);
1719         struct nlattr *tca[TCA_MAX + 1];
1720         struct Qdisc *q = NULL;
1721         struct tcf_block *block;
1722         struct tcf_chain *chain;
1723         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1724         long index_start;
1725         long index;
1726         u32 parent;
1727         int err;
1728
1729         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1730                 return skb->len;
1731
1732         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, NULL,
1733                           cb->extack);
1734         if (err)
1735                 return err;
1736
1737         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1738                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
1739                 if (!block)
1740                         goto out;
1741                 /* If we work with block index, q is NULL and parent value
1742                  * will never be used in the following code. The check
1743                  * in tcf_fill_node prevents it. However, compiler does not
1744                  * see that far, so set parent to zero to silence the warning
1745                  * about parent being uninitialized.
1746                  */
1747                 parent = 0;
1748         } else {
1749                 const struct Qdisc_class_ops *cops;
1750                 struct net_device *dev;
1751                 unsigned long cl = 0;
1752
1753                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1754                 if (!dev)
1755                         return skb->len;
1756
1757                 parent = tcm->tcm_parent;
1758                 if (!parent) {
1759                         q = dev->qdisc;
1760                         parent = q->handle;
1761                 } else {
1762                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
1763                 }
1764                 if (!q)
1765                         goto out;
1766                 cops = q->ops->cl_ops;
1767                 if (!cops)
1768                         goto out;
1769                 if (!cops->tcf_block)
1770                         goto out;
1771                 if (TC_H_MIN(tcm->tcm_parent)) {
1772                         cl = cops->find(q, tcm->tcm_parent);
1773                         if (cl == 0)
1774                                 goto out;
1775                 }
1776                 block = cops->tcf_block(q, cl, NULL);
1777                 if (!block)
1778                         goto out;
1779                 if (tcf_block_shared(block))
1780                         q = NULL;
1781         }
1782
1783         index_start = cb->args[0];
1784         index = 0;
1785
1786         list_for_each_entry(chain, &block->chain_list, list) {
1787                 if (tca[TCA_CHAIN] &&
1788                     nla_get_u32(tca[TCA_CHAIN]) != chain->index)
1789                         continue;
1790                 if (!tcf_chain_dump(chain, q, parent, skb, cb,
1791                                     index_start, &index)) {
1792                         err = -EMSGSIZE;
1793                         break;
1794                 }
1795         }
1796
1797         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1798                 tcf_block_refcnt_put(block);
1799         cb->args[0] = index;
1800
1801 out:
1802         /* If we did no progress, the error (EMSGSIZE) is real */
1803         if (skb->len == 0 && err)
1804                 return err;
1805         return skb->len;
1806 }
1807
1808 static int tc_chain_fill_node(struct tcf_chain *chain, struct net *net,
1809                               struct sk_buff *skb, struct tcf_block *block,
1810                               u32 portid, u32 seq, u16 flags, int event)
1811 {
1812         unsigned char *b = skb_tail_pointer(skb);
1813         const struct tcf_proto_ops *ops;
1814         struct nlmsghdr *nlh;
1815         struct tcmsg *tcm;
1816         void *priv;
1817
1818         ops = chain->tmplt_ops;
1819         priv = chain->tmplt_priv;
1820
1821         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1822         if (!nlh)
1823                 goto out_nlmsg_trim;
1824         tcm = nlmsg_data(nlh);
1825         tcm->tcm_family = AF_UNSPEC;
1826         tcm->tcm__pad1 = 0;
1827         tcm->tcm__pad2 = 0;
1828         tcm->tcm_handle = 0;
1829         if (block->q) {
1830                 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
1831                 tcm->tcm_parent = block->q->handle;
1832         } else {
1833                 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1834                 tcm->tcm_block_index = block->index;
1835         }
1836
1837         if (nla_put_u32(skb, TCA_CHAIN, chain->index))
1838                 goto nla_put_failure;
1839
1840         if (ops) {
1841                 if (nla_put_string(skb, TCA_KIND, ops->kind))
1842                         goto nla_put_failure;
1843                 if (ops->tmplt_dump(skb, net, priv) < 0)
1844                         goto nla_put_failure;
1845         }
1846
1847         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1848         return skb->len;
1849
1850 out_nlmsg_trim:
1851 nla_put_failure:
1852         nlmsg_trim(skb, b);
1853         return -EMSGSIZE;
1854 }
1855
1856 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
1857                            u32 seq, u16 flags, int event, bool unicast)
1858 {
1859         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1860         struct tcf_block *block = chain->block;
1861         struct net *net = block->net;
1862         struct sk_buff *skb;
1863
1864         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1865         if (!skb)
1866                 return -ENOBUFS;
1867
1868         if (tc_chain_fill_node(chain, net, skb, block, portid,
1869                                seq, flags, event) <= 0) {
1870                 kfree_skb(skb);
1871                 return -EINVAL;
1872         }
1873
1874         if (unicast)
1875                 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1876
1877         return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
1878 }
1879
1880 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
1881                               struct nlattr **tca,
1882                               struct netlink_ext_ack *extack)
1883 {
1884         const struct tcf_proto_ops *ops;
1885         void *tmplt_priv;
1886
1887         /* If kind is not set, user did not specify template. */
1888         if (!tca[TCA_KIND])
1889                 return 0;
1890
1891         ops = tcf_proto_lookup_ops(nla_data(tca[TCA_KIND]), extack);
1892         if (IS_ERR(ops))
1893                 return PTR_ERR(ops);
1894         if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
1895                 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
1896                 return -EOPNOTSUPP;
1897         }
1898
1899         tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
1900         if (IS_ERR(tmplt_priv)) {
1901                 module_put(ops->owner);
1902                 return PTR_ERR(tmplt_priv);
1903         }
1904         chain->tmplt_ops = ops;
1905         chain->tmplt_priv = tmplt_priv;
1906         return 0;
1907 }
1908
1909 static void tc_chain_tmplt_del(struct tcf_chain *chain)
1910 {
1911         const struct tcf_proto_ops *ops = chain->tmplt_ops;
1912
1913         /* If template ops are set, no work to do for us. */
1914         if (!ops)
1915                 return;
1916
1917         ops->tmplt_destroy(chain->tmplt_priv);
1918         module_put(ops->owner);
1919 }
1920
1921 /* Add/delete/get a chain */
1922
1923 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
1924                         struct netlink_ext_ack *extack)
1925 {
1926         struct net *net = sock_net(skb->sk);
1927         struct nlattr *tca[TCA_MAX + 1];
1928         struct tcmsg *t;
1929         u32 parent;
1930         u32 chain_index;
1931         struct Qdisc *q = NULL;
1932         struct tcf_chain *chain = NULL;
1933         struct tcf_block *block;
1934         unsigned long cl;
1935         int err;
1936
1937         if (n->nlmsg_type != RTM_GETCHAIN &&
1938             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1939                 return -EPERM;
1940
1941 replay:
1942         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, rtm_tca_policy, extack);
1943         if (err < 0)
1944                 return err;
1945
1946         t = nlmsg_data(n);
1947         parent = t->tcm_parent;
1948         cl = 0;
1949
1950         block = tcf_block_find(net, &q, &parent, &cl,
1951                                t->tcm_ifindex, t->tcm_block_index, extack);
1952         if (IS_ERR(block))
1953                 return PTR_ERR(block);
1954
1955         chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
1956         if (chain_index > TC_ACT_EXT_VAL_MASK) {
1957                 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
1958                 err = -EINVAL;
1959                 goto errout_block;
1960         }
1961         chain = tcf_chain_lookup(block, chain_index);
1962         if (n->nlmsg_type == RTM_NEWCHAIN) {
1963                 if (chain) {
1964                         if (tcf_chain_held_by_acts_only(chain)) {
1965                                 /* The chain exists only because there is
1966                                  * some action referencing it.
1967                                  */
1968                                 tcf_chain_hold(chain);
1969                         } else {
1970                                 NL_SET_ERR_MSG(extack, "Filter chain already exists");
1971                                 err = -EEXIST;
1972                                 goto errout_block;
1973                         }
1974                 } else {
1975                         if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1976                                 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
1977                                 err = -ENOENT;
1978                                 goto errout_block;
1979                         }
1980                         chain = tcf_chain_create(block, chain_index);
1981                         if (!chain) {
1982                                 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
1983                                 err = -ENOMEM;
1984                                 goto errout_block;
1985                         }
1986                 }
1987         } else {
1988                 if (!chain || tcf_chain_held_by_acts_only(chain)) {
1989                         NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
1990                         err = -EINVAL;
1991                         goto errout_block;
1992                 }
1993                 tcf_chain_hold(chain);
1994         }
1995
1996         switch (n->nlmsg_type) {
1997         case RTM_NEWCHAIN:
1998                 err = tc_chain_tmplt_add(chain, net, tca, extack);
1999                 if (err)
2000                         goto errout;
2001                 /* In case the chain was successfully added, take a reference
2002                  * to the chain. This ensures that an empty chain
2003                  * does not disappear at the end of this function.
2004                  */
2005                 tcf_chain_hold(chain);
2006                 chain->explicitly_created = true;
2007                 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2008                                 RTM_NEWCHAIN, false);
2009                 break;
2010         case RTM_DELCHAIN:
2011                 tfilter_notify_chain(net, skb, block, q, parent, n,
2012                                      chain, RTM_DELTFILTER);
2013                 /* Flush the chain first as the user requested chain removal. */
2014                 tcf_chain_flush(chain);
2015                 /* In case the chain was successfully deleted, put a reference
2016                  * to the chain previously taken during addition.
2017                  */
2018                 tcf_chain_put_explicitly_created(chain);
2019                 chain->explicitly_created = false;
2020                 break;
2021         case RTM_GETCHAIN:
2022                 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2023                                       n->nlmsg_seq, n->nlmsg_type, true);
2024                 if (err < 0)
2025                         NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2026                 break;
2027         default:
2028                 err = -EOPNOTSUPP;
2029                 NL_SET_ERR_MSG(extack, "Unsupported message type");
2030                 goto errout;
2031         }
2032
2033 errout:
2034         tcf_chain_put(chain);
2035 errout_block:
2036         tcf_block_release(q, block);
2037         if (err == -EAGAIN)
2038                 /* Replay the request. */
2039                 goto replay;
2040         return err;
2041 }
2042
2043 /* called with RTNL */
2044 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2045 {
2046         struct net *net = sock_net(skb->sk);
2047         struct nlattr *tca[TCA_MAX + 1];
2048         struct Qdisc *q = NULL;
2049         struct tcf_block *block;
2050         struct tcf_chain *chain;
2051         struct tcmsg *tcm = nlmsg_data(cb->nlh);
2052         long index_start;
2053         long index;
2054         u32 parent;
2055         int err;
2056
2057         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2058                 return skb->len;
2059
2060         err = nlmsg_parse(cb->nlh, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
2061                           cb->extack);
2062         if (err)
2063                 return err;
2064
2065         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2066                 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2067                 if (!block)
2068                         goto out;
2069                 /* If we work with block index, q is NULL and parent value
2070                  * will never be used in the following code. The check
2071                  * in tcf_fill_node prevents it. However, compiler does not
2072                  * see that far, so set parent to zero to silence the warning
2073                  * about parent being uninitialized.
2074                  */
2075                 parent = 0;
2076         } else {
2077                 const struct Qdisc_class_ops *cops;
2078                 struct net_device *dev;
2079                 unsigned long cl = 0;
2080
2081                 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2082                 if (!dev)
2083                         return skb->len;
2084
2085                 parent = tcm->tcm_parent;
2086                 if (!parent) {
2087                         q = dev->qdisc;
2088                         parent = q->handle;
2089                 } else {
2090                         q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2091                 }
2092                 if (!q)
2093                         goto out;
2094                 cops = q->ops->cl_ops;
2095                 if (!cops)
2096                         goto out;
2097                 if (!cops->tcf_block)
2098                         goto out;
2099                 if (TC_H_MIN(tcm->tcm_parent)) {
2100                         cl = cops->find(q, tcm->tcm_parent);
2101                         if (cl == 0)
2102                                 goto out;
2103                 }
2104                 block = cops->tcf_block(q, cl, NULL);
2105                 if (!block)
2106                         goto out;
2107                 if (tcf_block_shared(block))
2108                         q = NULL;
2109         }
2110
2111         index_start = cb->args[0];
2112         index = 0;
2113
2114         list_for_each_entry(chain, &block->chain_list, list) {
2115                 if ((tca[TCA_CHAIN] &&
2116                      nla_get_u32(tca[TCA_CHAIN]) != chain->index))
2117                         continue;
2118                 if (index < index_start) {
2119                         index++;
2120                         continue;
2121                 }
2122                 if (tcf_chain_held_by_acts_only(chain))
2123                         continue;
2124                 err = tc_chain_fill_node(chain, net, skb, block,
2125                                          NETLINK_CB(cb->skb).portid,
2126                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
2127                                          RTM_NEWCHAIN);
2128                 if (err <= 0)
2129                         break;
2130                 index++;
2131         }
2132
2133         if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2134                 tcf_block_refcnt_put(block);
2135         cb->args[0] = index;
2136
2137 out:
2138         /* If we did no progress, the error (EMSGSIZE) is real */
2139         if (skb->len == 0 && err)
2140                 return err;
2141         return skb->len;
2142 }
2143
2144 void tcf_exts_destroy(struct tcf_exts *exts)
2145 {
2146 #ifdef CONFIG_NET_CLS_ACT
2147         tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
2148         kfree(exts->actions);
2149         exts->nr_actions = 0;
2150 #endif
2151 }
2152 EXPORT_SYMBOL(tcf_exts_destroy);
2153
2154 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
2155                       struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
2156                       struct netlink_ext_ack *extack)
2157 {
2158 #ifdef CONFIG_NET_CLS_ACT
2159         {
2160                 struct tc_action *act;
2161                 size_t attr_size = 0;
2162
2163                 if (exts->police && tb[exts->police]) {
2164                         act = tcf_action_init_1(net, tp, tb[exts->police],
2165                                                 rate_tlv, "police", ovr,
2166                                                 TCA_ACT_BIND, true, extack);
2167                         if (IS_ERR(act))
2168                                 return PTR_ERR(act);
2169
2170                         act->type = exts->type = TCA_OLD_COMPAT;
2171                         exts->actions[0] = act;
2172                         exts->nr_actions = 1;
2173                 } else if (exts->action && tb[exts->action]) {
2174                         int err;
2175
2176                         err = tcf_action_init(net, tp, tb[exts->action],
2177                                               rate_tlv, NULL, ovr, TCA_ACT_BIND,
2178                                               exts->actions, &attr_size, true,
2179                                               extack);
2180                         if (err < 0)
2181                                 return err;
2182                         exts->nr_actions = err;
2183                 }
2184                 exts->net = net;
2185         }
2186 #else
2187         if ((exts->action && tb[exts->action]) ||
2188             (exts->police && tb[exts->police])) {
2189                 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
2190                 return -EOPNOTSUPP;
2191         }
2192 #endif
2193
2194         return 0;
2195 }
2196 EXPORT_SYMBOL(tcf_exts_validate);
2197
2198 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
2199 {
2200 #ifdef CONFIG_NET_CLS_ACT
2201         struct tcf_exts old = *dst;
2202
2203         *dst = *src;
2204         tcf_exts_destroy(&old);
2205 #endif
2206 }
2207 EXPORT_SYMBOL(tcf_exts_change);
2208
2209 #ifdef CONFIG_NET_CLS_ACT
2210 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
2211 {
2212         if (exts->nr_actions == 0)
2213                 return NULL;
2214         else
2215                 return exts->actions[0];
2216 }
2217 #endif
2218
2219 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
2220 {
2221 #ifdef CONFIG_NET_CLS_ACT
2222         struct nlattr *nest;
2223
2224         if (exts->action && tcf_exts_has_actions(exts)) {
2225                 /*
2226                  * again for backward compatible mode - we want
2227                  * to work with both old and new modes of entering
2228                  * tc data even if iproute2  was newer - jhs
2229                  */
2230                 if (exts->type != TCA_OLD_COMPAT) {
2231                         nest = nla_nest_start(skb, exts->action);
2232                         if (nest == NULL)
2233                                 goto nla_put_failure;
2234
2235                         if (tcf_action_dump(skb, exts->actions, 0, 0) < 0)
2236                                 goto nla_put_failure;
2237                         nla_nest_end(skb, nest);
2238                 } else if (exts->police) {
2239                         struct tc_action *act = tcf_exts_first_act(exts);
2240                         nest = nla_nest_start(skb, exts->police);
2241                         if (nest == NULL || !act)
2242                                 goto nla_put_failure;
2243                         if (tcf_action_dump_old(skb, act, 0, 0) < 0)
2244                                 goto nla_put_failure;
2245                         nla_nest_end(skb, nest);
2246                 }
2247         }
2248         return 0;
2249
2250 nla_put_failure:
2251         nla_nest_cancel(skb, nest);
2252         return -1;
2253 #else
2254         return 0;
2255 #endif
2256 }
2257 EXPORT_SYMBOL(tcf_exts_dump);
2258
2259
2260 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
2261 {
2262 #ifdef CONFIG_NET_CLS_ACT
2263         struct tc_action *a = tcf_exts_first_act(exts);
2264         if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
2265                 return -1;
2266 #endif
2267         return 0;
2268 }
2269 EXPORT_SYMBOL(tcf_exts_dump_stats);
2270
2271 static int tc_exts_setup_cb_egdev_call(struct tcf_exts *exts,
2272                                        enum tc_setup_type type,
2273                                        void *type_data, bool err_stop)
2274 {
2275         int ok_count = 0;
2276 #ifdef CONFIG_NET_CLS_ACT
2277         const struct tc_action *a;
2278         struct net_device *dev;
2279         int i, ret;
2280
2281         if (!tcf_exts_has_actions(exts))
2282                 return 0;
2283
2284         for (i = 0; i < exts->nr_actions; i++) {
2285                 a = exts->actions[i];
2286                 if (!a->ops->get_dev)
2287                         continue;
2288                 dev = a->ops->get_dev(a);
2289                 if (!dev)
2290                         continue;
2291                 ret = tc_setup_cb_egdev_call(dev, type, type_data, err_stop);
2292                 a->ops->put_dev(dev);
2293                 if (ret < 0)
2294                         return ret;
2295                 ok_count += ret;
2296         }
2297 #endif
2298         return ok_count;
2299 }
2300
2301 int tc_setup_cb_call(struct tcf_block *block, struct tcf_exts *exts,
2302                      enum tc_setup_type type, void *type_data, bool err_stop)
2303 {
2304         int ok_count;
2305         int ret;
2306
2307         ret = tcf_block_cb_call(block, type, type_data, err_stop);
2308         if (ret < 0)
2309                 return ret;
2310         ok_count = ret;
2311
2312         if (!exts || ok_count)
2313                 return ok_count;
2314         ret = tc_exts_setup_cb_egdev_call(exts, type, type_data, err_stop);
2315         if (ret < 0)
2316                 return ret;
2317         ok_count += ret;
2318
2319         return ok_count;
2320 }
2321 EXPORT_SYMBOL(tc_setup_cb_call);
2322
2323 static __net_init int tcf_net_init(struct net *net)
2324 {
2325         struct tcf_net *tn = net_generic(net, tcf_net_id);
2326
2327         spin_lock_init(&tn->idr_lock);
2328         idr_init(&tn->idr);
2329         return 0;
2330 }
2331
2332 static void __net_exit tcf_net_exit(struct net *net)
2333 {
2334         struct tcf_net *tn = net_generic(net, tcf_net_id);
2335
2336         idr_destroy(&tn->idr);
2337 }
2338
2339 static struct pernet_operations tcf_net_ops = {
2340         .init = tcf_net_init,
2341         .exit = tcf_net_exit,
2342         .id   = &tcf_net_id,
2343         .size = sizeof(struct tcf_net),
2344 };
2345
2346 static int __init tc_filter_init(void)
2347 {
2348         int err;
2349
2350         tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
2351         if (!tc_filter_wq)
2352                 return -ENOMEM;
2353
2354         err = register_pernet_subsys(&tcf_net_ops);
2355         if (err)
2356                 goto err_register_pernet_subsys;
2357
2358         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 0);
2359         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 0);
2360         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
2361                       tc_dump_tfilter, 0);
2362         rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
2363         rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
2364         rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
2365                       tc_dump_chain, 0);
2366
2367         return 0;
2368
2369 err_register_pernet_subsys:
2370         destroy_workqueue(tc_filter_wq);
2371         return err;
2372 }
2373
2374 subsys_initcall(tc_filter_init);