net/sched/act_api.c

   1 /*
   2  * net/sched/act_api.c  Packet action 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  * Author:      Jamal Hadi Salim
  10  *
  11  *
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/kernel.h>
  16 #include <linux/string.h>
  17 #include <linux/errno.h>
  18 #include <linux/slab.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/init.h>
  21 #include <linux/kmod.h>
  22 #include <linux/err.h>
  23 #include <linux/module.h>
  24 #include <linux/rhashtable.h>
  25 #include <linux/list.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28 #include <net/sch_generic.h>
  29 #include <net/pkt_cls.h>
  30 #include <net/act_api.h>
  31 #include <net/netlink.h>
  32
  33 static int tcf_action_goto_chain_init(struct tc_action *a, struct tcf_proto *tp)
  34 {
  35         u32 chain_index = a->tcfa_action & TC_ACT_EXT_VAL_MASK;
  36
  37         if (!tp)
  38                 return -EINVAL;
  39         a->goto_chain = tcf_chain_get_by_act(tp->chain->block, chain_index);
  40         if (!a->goto_chain)
  41                 return -ENOMEM;
  42         return 0;
  43 }
  44
  45 static void tcf_action_goto_chain_fini(struct tc_action *a)
  46 {
  47         tcf_chain_put_by_act(a->goto_chain);
  48 }
  49
  50 static void tcf_action_goto_chain_exec(const struct tc_action *a,
  51                                        struct tcf_result *res)
  52 {
  53         const struct tcf_chain *chain = a->goto_chain;
  54
  55         res->goto_tp = rcu_dereference_bh(chain->filter_chain);
  56 }
  57
  58 static void tcf_free_cookie_rcu(struct rcu_head *p)
  59 {
  60         struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
  61
  62         kfree(cookie->data);
  63         kfree(cookie);
  64 }
  65
  66 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
  67                                   struct tc_cookie *new_cookie)
  68 {
  69         struct tc_cookie *old;
  70
  71         old = xchg((__force struct tc_cookie **)old_cookie, new_cookie);
  72         if (old)
  73                 call_rcu(&old->rcu, tcf_free_cookie_rcu);
  74 }
  75
  76 /* XXX: For standalone actions, we don't need a RCU grace period either, because
  77  * actions are always connected to filters and filters are already destroyed in
  78  * RCU callbacks, so after a RCU grace period actions are already disconnected
  79  * from filters. Readers later can not find us.
  80  */
  81 static void free_tcf(struct tc_action *p)
  82 {
  83         free_percpu(p->cpu_bstats);
  84         free_percpu(p->cpu_bstats_hw);
  85         free_percpu(p->cpu_qstats);
  86
  87         tcf_set_action_cookie(&p->act_cookie, NULL);
  88         if (p->goto_chain)
  89                 tcf_action_goto_chain_fini(p);
  90
  91         kfree(p);
  92 }
  93
  94 static void tcf_action_cleanup(struct tc_action *p)
  95 {
  96         if (p->ops->cleanup)
  97                 p->ops->cleanup(p);
  98
  99         gen_kill_estimator(&p->tcfa_rate_est);
 100         free_tcf(p);
 101 }
 102
 103 static int __tcf_action_put(struct tc_action *p, bool bind)
 104 {
 105         struct tcf_idrinfo *idrinfo = p->idrinfo;
 106
 107         if (refcount_dec_and_mutex_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
 108                 if (bind)
 109                         atomic_dec(&p->tcfa_bindcnt);
 110                 idr_remove(&idrinfo->action_idr, p->tcfa_index);
 111                 mutex_unlock(&idrinfo->lock);
 112
 113                 tcf_action_cleanup(p);
 114                 return 1;
 115         }
 116
 117         if (bind)
 118                 atomic_dec(&p->tcfa_bindcnt);
 119
 120         return 0;
 121 }
 122
 123 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
 124 {
 125         int ret = 0;
 126
 127         /* Release with strict==1 and bind==0 is only called through act API
 128          * interface (classifiers always bind). Only case when action with
 129          * positive reference count and zero bind count can exist is when it was
 130          * also created with act API (unbinding last classifier will destroy the
 131          * action if it was created by classifier). So only case when bind count
 132          * can be changed after initial check is when unbound action is
 133          * destroyed by act API while classifier binds to action with same id
 134          * concurrently. This result either creation of new action(same behavior
 135          * as before), or reusing existing action if concurrent process
 136          * increments reference count before action is deleted. Both scenarios
 137          * are acceptable.
 138          */
 139         if (p) {
 140                 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
 141                         return -EPERM;
 142
 143                 if (__tcf_action_put(p, bind))
 144                         ret = ACT_P_DELETED;
 145         }
 146
 147         return ret;
 148 }
 149 EXPORT_SYMBOL(__tcf_idr_release);
 150
 151 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
 152 {
 153         struct tc_cookie *act_cookie;
 154         u32 cookie_len = 0;
 155
 156         rcu_read_lock();
 157         act_cookie = rcu_dereference(act->act_cookie);
 158
 159         if (act_cookie)
 160                 cookie_len = nla_total_size(act_cookie->len);
 161         rcu_read_unlock();
 162
 163         return  nla_total_size(0) /* action number nested */
 164                 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
 165                 + cookie_len /* TCA_ACT_COOKIE */
 166                 + nla_total_size(0) /* TCA_ACT_STATS nested */
 167                 /* TCA_STATS_BASIC */
 168                 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
 169                 /* TCA_STATS_QUEUE */
 170                 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
 171                 + nla_total_size(0) /* TCA_OPTIONS nested */
 172                 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
 173 }
 174
 175 static size_t tcf_action_full_attrs_size(size_t sz)
 176 {
 177         return NLMSG_HDRLEN                     /* struct nlmsghdr */
 178                 + sizeof(struct tcamsg)
 179                 + nla_total_size(0)             /* TCA_ACT_TAB nested */
 180                 + sz;
 181 }
 182
 183 static size_t tcf_action_fill_size(const struct tc_action *act)
 184 {
 185         size_t sz = tcf_action_shared_attrs_size(act);
 186
 187         if (act->ops->get_fill_size)
 188                 return act->ops->get_fill_size(act) + sz;
 189         return sz;
 190 }
 191
 192 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 193                            struct netlink_callback *cb)
 194 {
 195         int err = 0, index = -1, s_i = 0, n_i = 0;
 196         u32 act_flags = cb->args[2];
 197         unsigned long jiffy_since = cb->args[3];
 198         struct nlattr *nest;
 199         struct idr *idr = &idrinfo->action_idr;
 200         struct tc_action *p;
 201         unsigned long id = 1;
 202
 203         mutex_lock(&idrinfo->lock);
 204
 205         s_i = cb->args[0];
 206
 207         idr_for_each_entry_ul(idr, p, id) {
 208                 index++;
 209                 if (index < s_i)
 210                         continue;
 211
 212                 if (jiffy_since &&
 213                     time_after(jiffy_since,
 214                                (unsigned long)p->tcfa_tm.lastuse))
 215                         continue;
 216
 217                 nest = nla_nest_start(skb, n_i);
 218                 if (!nest) {
 219                         index--;
 220                         goto nla_put_failure;
 221                 }
 222                 err = tcf_action_dump_1(skb, p, 0, 0);
 223                 if (err < 0) {
 224                         index--;
 225                         nlmsg_trim(skb, nest);
 226                         goto done;
 227                 }
 228                 nla_nest_end(skb, nest);
 229                 n_i++;
 230                 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
 231                     n_i >= TCA_ACT_MAX_PRIO)
 232                         goto done;
 233         }
 234 done:
 235         if (index >= 0)
 236                 cb->args[0] = index + 1;
 237
 238         mutex_unlock(&idrinfo->lock);
 239         if (n_i) {
 240                 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
 241                         cb->args[1] = n_i;
 242         }
 243         return n_i;
 244
 245 nla_put_failure:
 246         nla_nest_cancel(skb, nest);
 247         goto done;
 248 }
 249
 250 static int tcf_idr_release_unsafe(struct tc_action *p)
 251 {
 252         if (atomic_read(&p->tcfa_bindcnt) > 0)
 253                 return -EPERM;
 254
 255         if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 256                 idr_remove(&p->idrinfo->action_idr, p->tcfa_index);
 257                 tcf_action_cleanup(p);
 258                 return ACT_P_DELETED;
 259         }
 260
 261         return 0;
 262 }
 263
 264 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 265                           const struct tc_action_ops *ops)
 266 {
 267         struct nlattr *nest;
 268         int n_i = 0;
 269         int ret = -EINVAL;
 270         struct idr *idr = &idrinfo->action_idr;
 271         struct tc_action *p;
 272         unsigned long id = 1;
 273
 274         nest = nla_nest_start(skb, 0);
 275         if (nest == NULL)
 276                 goto nla_put_failure;
 277         if (nla_put_string(skb, TCA_KIND, ops->kind))
 278                 goto nla_put_failure;
 279
 280         mutex_lock(&idrinfo->lock);
 281         idr_for_each_entry_ul(idr, p, id) {
 282                 ret = tcf_idr_release_unsafe(p);
 283                 if (ret == ACT_P_DELETED) {
 284                         module_put(ops->owner);
 285                         n_i++;
 286                 } else if (ret < 0) {
 287                         mutex_unlock(&idrinfo->lock);
 288                         goto nla_put_failure;
 289                 }
 290         }
 291         mutex_unlock(&idrinfo->lock);
 292
 293         if (nla_put_u32(skb, TCA_FCNT, n_i))
 294                 goto nla_put_failure;
 295         nla_nest_end(skb, nest);
 296
 297         return n_i;
 298 nla_put_failure:
 299         nla_nest_cancel(skb, nest);
 300         return ret;
 301 }
 302
 303 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 304                        struct netlink_callback *cb, int type,
 305                        const struct tc_action_ops *ops,
 306                        struct netlink_ext_ack *extack)
 307 {
 308         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 309
 310         if (type == RTM_DELACTION) {
 311                 return tcf_del_walker(idrinfo, skb, ops);
 312         } else if (type == RTM_GETACTION) {
 313                 return tcf_dump_walker(idrinfo, skb, cb);
 314         } else {
 315                 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
 316                 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
 317                 return -EINVAL;
 318         }
 319 }
 320 EXPORT_SYMBOL(tcf_generic_walker);
 321
 322 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 323 {
 324         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 325         struct tc_action *p;
 326
 327         mutex_lock(&idrinfo->lock);
 328         p = idr_find(&idrinfo->action_idr, index);
 329         if (IS_ERR(p))
 330                 p = NULL;
 331         else if (p)
 332                 refcount_inc(&p->tcfa_refcnt);
 333         mutex_unlock(&idrinfo->lock);
 334
 335         if (p) {
 336                 *a = p;
 337                 return true;
 338         }
 339         return false;
 340 }
 341 EXPORT_SYMBOL(tcf_idr_search);
 342
 343 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
 344 {
 345         struct tc_action *p;
 346         int ret = 0;
 347
 348         mutex_lock(&idrinfo->lock);
 349         p = idr_find(&idrinfo->action_idr, index);
 350         if (!p) {
 351                 mutex_unlock(&idrinfo->lock);
 352                 return -ENOENT;
 353         }
 354
 355         if (!atomic_read(&p->tcfa_bindcnt)) {
 356                 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 357                         struct module *owner = p->ops->owner;
 358
 359                         WARN_ON(p != idr_remove(&idrinfo->action_idr,
 360                                                 p->tcfa_index));
 361                         mutex_unlock(&idrinfo->lock);
 362
 363                         tcf_action_cleanup(p);
 364                         module_put(owner);
 365                         return 0;
 366                 }
 367                 ret = 0;
 368         } else {
 369                 ret = -EPERM;
 370         }
 371
 372         mutex_unlock(&idrinfo->lock);
 373         return ret;
 374 }
 375
 376 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 377                    struct tc_action **a, const struct tc_action_ops *ops,
 378                    int bind, bool cpustats)
 379 {
 380         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 381         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 382         int err = -ENOMEM;
 383
 384         if (unlikely(!p))
 385                 return -ENOMEM;
 386         refcount_set(&p->tcfa_refcnt, 1);
 387         if (bind)
 388                 atomic_set(&p->tcfa_bindcnt, 1);
 389
 390         if (cpustats) {
 391                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 392                 if (!p->cpu_bstats)
 393                         goto err1;
 394                 p->cpu_bstats_hw = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 395                 if (!p->cpu_bstats_hw)
 396                         goto err2;
 397                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 398                 if (!p->cpu_qstats)
 399                         goto err3;
 400         }
 401         spin_lock_init(&p->tcfa_lock);
 402         p->tcfa_index = index;
 403         p->tcfa_tm.install = jiffies;
 404         p->tcfa_tm.lastuse = jiffies;
 405         p->tcfa_tm.firstuse = 0;
 406         if (est) {
 407                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 408                                         &p->tcfa_rate_est,
 409                                         &p->tcfa_lock, NULL, est);
 410                 if (err)
 411                         goto err4;
 412         }
 413
 414         p->idrinfo = idrinfo;
 415         p->ops = ops;
 416         *a = p;
 417         return 0;
 418 err4:
 419         free_percpu(p->cpu_qstats);
 420 err3:
 421         free_percpu(p->cpu_bstats_hw);
 422 err2:
 423         free_percpu(p->cpu_bstats);
 424 err1:
 425         kfree(p);
 426         return err;
 427 }
 428 EXPORT_SYMBOL(tcf_idr_create);
 429
 430 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
 431 {
 432         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 433
 434         mutex_lock(&idrinfo->lock);
 435         /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
 436         WARN_ON(!IS_ERR(idr_replace(&idrinfo->action_idr, a, a->tcfa_index)));
 437         mutex_unlock(&idrinfo->lock);
 438 }
 439 EXPORT_SYMBOL(tcf_idr_insert);
 440
 441 /* Cleanup idr index that was allocated but not initialized. */
 442
 443 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
 444 {
 445         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 446
 447         mutex_lock(&idrinfo->lock);
 448         /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
 449         WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
 450         mutex_unlock(&idrinfo->lock);
 451 }
 452 EXPORT_SYMBOL(tcf_idr_cleanup);
 453
 454 /* Check if action with specified index exists. If actions is found, increments
 455  * its reference and bind counters, and return 1. Otherwise insert temporary
 456  * error pointer (to prevent concurrent users from inserting actions with same
 457  * index) and return 0.
 458  */
 459
 460 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
 461                         struct tc_action **a, int bind)
 462 {
 463         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 464         struct tc_action *p;
 465         int ret;
 466
 467 again:
 468         mutex_lock(&idrinfo->lock);
 469         if (*index) {
 470                 p = idr_find(&idrinfo->action_idr, *index);
 471                 if (IS_ERR(p)) {
 472                         /* This means that another process allocated
 473                          * index but did not assign the pointer yet.
 474                          */
 475                         mutex_unlock(&idrinfo->lock);
 476                         goto again;
 477                 }
 478
 479                 if (p) {
 480                         refcount_inc(&p->tcfa_refcnt);
 481                         if (bind)
 482                                 atomic_inc(&p->tcfa_bindcnt);
 483                         *a = p;
 484                         ret = 1;
 485                 } else {
 486                         *a = NULL;
 487                         ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 488                                             *index, GFP_KERNEL);
 489                         if (!ret)
 490                                 idr_replace(&idrinfo->action_idr,
 491                                             ERR_PTR(-EBUSY), *index);
 492                 }
 493         } else {
 494                 *index = 1;
 495                 *a = NULL;
 496                 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 497                                     UINT_MAX, GFP_KERNEL);
 498                 if (!ret)
 499                         idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
 500                                     *index);
 501         }
 502         mutex_unlock(&idrinfo->lock);
 503         return ret;
 504 }
 505 EXPORT_SYMBOL(tcf_idr_check_alloc);
 506
 507 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
 508                          struct tcf_idrinfo *idrinfo)
 509 {
 510         struct idr *idr = &idrinfo->action_idr;
 511         struct tc_action *p;
 512         int ret;
 513         unsigned long id = 1;
 514
 515         idr_for_each_entry_ul(idr, p, id) {
 516                 ret = __tcf_idr_release(p, false, true);
 517                 if (ret == ACT_P_DELETED)
 518                         module_put(ops->owner);
 519                 else if (ret < 0)
 520                         return;
 521         }
 522         idr_destroy(&idrinfo->action_idr);
 523 }
 524 EXPORT_SYMBOL(tcf_idrinfo_destroy);
 525
 526 static LIST_HEAD(act_base);
 527 static DEFINE_RWLOCK(act_mod_lock);
 528
 529 int tcf_register_action(struct tc_action_ops *act,
 530                         struct pernet_operations *ops)
 531 {
 532         struct tc_action_ops *a;
 533         int ret;
 534
 535         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 536                 return -EINVAL;
 537
 538         /* We have to register pernet ops before making the action ops visible,
 539          * otherwise tcf_action_init_1() could get a partially initialized
 540          * netns.
 541          */
 542         ret = register_pernet_subsys(ops);
 543         if (ret)
 544                 return ret;
 545
 546         write_lock(&act_mod_lock);
 547         list_for_each_entry(a, &act_base, head) {
 548                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
 549                         write_unlock(&act_mod_lock);
 550                         unregister_pernet_subsys(ops);
 551                         return -EEXIST;
 552                 }
 553         }
 554         list_add_tail(&act->head, &act_base);
 555         write_unlock(&act_mod_lock);
 556
 557         return 0;
 558 }
 559 EXPORT_SYMBOL(tcf_register_action);
 560
 561 int tcf_unregister_action(struct tc_action_ops *act,
 562                           struct pernet_operations *ops)
 563 {
 564         struct tc_action_ops *a;
 565         int err = -ENOENT;
 566
 567         write_lock(&act_mod_lock);
 568         list_for_each_entry(a, &act_base, head) {
 569                 if (a == act) {
 570                         list_del(&act->head);
 571                         err = 0;
 572                         break;
 573                 }
 574         }
 575         write_unlock(&act_mod_lock);
 576         if (!err)
 577                 unregister_pernet_subsys(ops);
 578         return err;
 579 }
 580 EXPORT_SYMBOL(tcf_unregister_action);
 581
 582 /* lookup by name */
 583 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 584 {
 585         struct tc_action_ops *a, *res = NULL;
 586
 587         if (kind) {
 588                 read_lock(&act_mod_lock);
 589                 list_for_each_entry(a, &act_base, head) {
 590                         if (strcmp(kind, a->kind) == 0) {
 591                                 if (try_module_get(a->owner))
 592                                         res = a;
 593                                 break;
 594                         }
 595                 }
 596                 read_unlock(&act_mod_lock);
 597         }
 598         return res;
 599 }
 600
 601 /* lookup by nlattr */
 602 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 603 {
 604         struct tc_action_ops *a, *res = NULL;
 605
 606         if (kind) {
 607                 read_lock(&act_mod_lock);
 608                 list_for_each_entry(a, &act_base, head) {
 609                         if (nla_strcmp(kind, a->kind) == 0) {
 610                                 if (try_module_get(a->owner))
 611                                         res = a;
 612                                 break;
 613                         }
 614                 }
 615                 read_unlock(&act_mod_lock);
 616         }
 617         return res;
 618 }
 619
 620 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
 621 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
 622 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 623                     int nr_actions, struct tcf_result *res)
 624 {
 625         u32 jmp_prgcnt = 0;
 626         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
 627         int i;
 628         int ret = TC_ACT_OK;
 629
 630         if (skb_skip_tc_classify(skb))
 631                 return TC_ACT_OK;
 632
 633 restart_act_graph:
 634         for (i = 0; i < nr_actions; i++) {
 635                 const struct tc_action *a = actions[i];
 636
 637                 if (jmp_prgcnt > 0) {
 638                         jmp_prgcnt -= 1;
 639                         continue;
 640                 }
 641 repeat:
 642                 ret = a->ops->act(skb, a, res);
 643                 if (ret == TC_ACT_REPEAT)
 644                         goto repeat;    /* we need a ttl - JHS */
 645
 646                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
 647                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
 648                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
 649                                 /* faulty opcode, stop pipeline */
 650                                 return TC_ACT_OK;
 651                         } else {
 652                                 jmp_ttl -= 1;
 653                                 if (jmp_ttl > 0)
 654                                         goto restart_act_graph;
 655                                 else /* faulty graph, stop pipeline */
 656                                         return TC_ACT_OK;
 657                         }
 658                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
 659                         tcf_action_goto_chain_exec(a, res);
 660                 }
 661
 662                 if (ret != TC_ACT_PIPE)
 663                         break;
 664         }
 665
 666         return ret;
 667 }
 668 EXPORT_SYMBOL(tcf_action_exec);
 669
 670 int tcf_action_destroy(struct tc_action *actions[], int bind)
 671 {
 672         const struct tc_action_ops *ops;
 673         struct tc_action *a;
 674         int ret = 0, i;
 675
 676         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 677                 a = actions[i];
 678                 actions[i] = NULL;
 679                 ops = a->ops;
 680                 ret = __tcf_idr_release(a, bind, true);
 681                 if (ret == ACT_P_DELETED)
 682                         module_put(ops->owner);
 683                 else if (ret < 0)
 684                         return ret;
 685         }
 686         return ret;
 687 }
 688
 689 static int tcf_action_destroy_1(struct tc_action *a, int bind)
 690 {
 691         struct tc_action *actions[] = { a, NULL };
 692
 693         return tcf_action_destroy(actions, bind);
 694 }
 695
 696 static int tcf_action_put(struct tc_action *p)
 697 {
 698         return __tcf_action_put(p, false);
 699 }
 700
 701 /* Put all actions in this array, skip those NULL's. */
 702 static void tcf_action_put_many(struct tc_action *actions[])
 703 {
 704         int i;
 705
 706         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 707                 struct tc_action *a = actions[i];
 708                 const struct tc_action_ops *ops;
 709
 710                 if (!a)
 711                         continue;
 712                 ops = a->ops;
 713                 if (tcf_action_put(a))
 714                         module_put(ops->owner);
 715         }
 716 }
 717
 718 int
 719 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 720 {
 721         return a->ops->dump(skb, a, bind, ref);
 722 }
 723
 724 int
 725 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 726 {
 727         int err = -EINVAL;
 728         unsigned char *b = skb_tail_pointer(skb);
 729         struct nlattr *nest;
 730         struct tc_cookie *cookie;
 731
 732         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 733                 goto nla_put_failure;
 734         if (tcf_action_copy_stats(skb, a, 0))
 735                 goto nla_put_failure;
 736
 737         rcu_read_lock();
 738         cookie = rcu_dereference(a->act_cookie);
 739         if (cookie) {
 740                 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
 741                         rcu_read_unlock();
 742                         goto nla_put_failure;
 743                 }
 744         }
 745         rcu_read_unlock();
 746
 747         nest = nla_nest_start(skb, TCA_OPTIONS);
 748         if (nest == NULL)
 749                 goto nla_put_failure;
 750         err = tcf_action_dump_old(skb, a, bind, ref);
 751         if (err > 0) {
 752                 nla_nest_end(skb, nest);
 753                 return err;
 754         }
 755
 756 nla_put_failure:
 757         nlmsg_trim(skb, b);
 758         return -1;
 759 }
 760 EXPORT_SYMBOL(tcf_action_dump_1);
 761
 762 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
 763                     int bind, int ref)
 764 {
 765         struct tc_action *a;
 766         int err = -EINVAL, i;
 767         struct nlattr *nest;
 768
 769         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 770                 a = actions[i];
 771                 nest = nla_nest_start(skb, a->order);
 772                 if (nest == NULL)
 773                         goto nla_put_failure;
 774                 err = tcf_action_dump_1(skb, a, bind, ref);
 775                 if (err < 0)
 776                         goto errout;
 777                 nla_nest_end(skb, nest);
 778         }
 779
 780         return 0;
 781
 782 nla_put_failure:
 783         err = -EINVAL;
 784 errout:
 785         nla_nest_cancel(skb, nest);
 786         return err;
 787 }
 788
 789 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
 790 {
 791         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
 792         if (!c)
 793                 return NULL;
 794
 795         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
 796         if (!c->data) {
 797                 kfree(c);
 798                 return NULL;
 799         }
 800         c->len = nla_len(tb[TCA_ACT_COOKIE]);
 801
 802         return c;
 803 }
 804
 805 static bool tcf_action_valid(int action)
 806 {
 807         int opcode = TC_ACT_EXT_OPCODE(action);
 808
 809         if (!opcode)
 810                 return action <= TC_ACT_VALUE_MAX;
 811         return opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC;
 812 }
 813
 814 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
 815                                     struct nlattr *nla, struct nlattr *est,
 816                                     char *name, int ovr, int bind,
 817                                     bool rtnl_held,
 818                                     struct netlink_ext_ack *extack)
 819 {
 820         struct tc_action *a;
 821         struct tc_action_ops *a_o;
 822         struct tc_cookie *cookie = NULL;
 823         char act_name[IFNAMSIZ];
 824         struct nlattr *tb[TCA_ACT_MAX + 1];
 825         struct nlattr *kind;
 826         int err;
 827
 828         if (name == NULL) {
 829                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
 830                 if (err < 0)
 831                         goto err_out;
 832                 err = -EINVAL;
 833                 kind = tb[TCA_ACT_KIND];
 834                 if (!kind) {
 835                         NL_SET_ERR_MSG(extack, "TC action kind must be specified");
 836                         goto err_out;
 837                 }
 838                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
 839                         NL_SET_ERR_MSG(extack, "TC action name too long");
 840                         goto err_out;
 841                 }
 842                 if (tb[TCA_ACT_COOKIE]) {
 843                         int cklen = nla_len(tb[TCA_ACT_COOKIE]);
 844
 845                         if (cklen > TC_COOKIE_MAX_SIZE) {
 846                                 NL_SET_ERR_MSG(extack, "TC cookie size above the maximum");
 847                                 goto err_out;
 848                         }
 849
 850                         cookie = nla_memdup_cookie(tb);
 851                         if (!cookie) {
 852                                 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
 853                                 err = -ENOMEM;
 854                                 goto err_out;
 855                         }
 856                 }
 857         } else {
 858                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
 859                         NL_SET_ERR_MSG(extack, "TC action name too long");
 860                         err = -EINVAL;
 861                         goto err_out;
 862                 }
 863         }
 864
 865         a_o = tc_lookup_action_n(act_name);
 866         if (a_o == NULL) {
 867 #ifdef CONFIG_MODULES
 868                 if (rtnl_held)
 869                         rtnl_unlock();
 870                 request_module("act_%s", act_name);
 871                 if (rtnl_held)
 872                         rtnl_lock();
 873
 874                 a_o = tc_lookup_action_n(act_name);
 875
 876                 /* We dropped the RTNL semaphore in order to
 877                  * perform the module load.  So, even if we
 878                  * succeeded in loading the module we have to
 879                  * tell the caller to replay the request.  We
 880                  * indicate this using -EAGAIN.
 881                  */
 882                 if (a_o != NULL) {
 883                         err = -EAGAIN;
 884                         goto err_mod;
 885                 }
 886 #endif
 887                 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
 888                 err = -ENOENT;
 889                 goto err_out;
 890         }
 891
 892         /* backward compatibility for policer */
 893         if (name == NULL)
 894                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
 895                                 rtnl_held, extack);
 896         else
 897                 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
 898                                 extack);
 899         if (err < 0)
 900                 goto err_mod;
 901
 902         if (!name && tb[TCA_ACT_COOKIE])
 903                 tcf_set_action_cookie(&a->act_cookie, cookie);
 904
 905         /* module count goes up only when brand new policy is created
 906          * if it exists and is only bound to in a_o->init() then
 907          * ACT_P_CREATED is not returned (a zero is).
 908          */
 909         if (err != ACT_P_CREATED)
 910                 module_put(a_o->owner);
 911
 912         if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
 913                 err = tcf_action_goto_chain_init(a, tp);
 914                 if (err) {
 915                         tcf_action_destroy_1(a, bind);
 916                         NL_SET_ERR_MSG(extack, "Failed to init TC action chain");
 917                         return ERR_PTR(err);
 918                 }
 919         }
 920
 921         if (!tcf_action_valid(a->tcfa_action)) {
 922                 tcf_action_destroy_1(a, bind);
 923                 NL_SET_ERR_MSG(extack, "Invalid control action value");
 924                 return ERR_PTR(-EINVAL);
 925         }
 926
 927         return a;
 928
 929 err_mod:
 930         module_put(a_o->owner);
 931 err_out:
 932         if (cookie) {
 933                 kfree(cookie->data);
 934                 kfree(cookie);
 935         }
 936         return ERR_PTR(err);
 937 }
 938
 939 /* Returns numbers of initialized actions or negative error. */
 940
 941 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
 942                     struct nlattr *est, char *name, int ovr, int bind,
 943                     struct tc_action *actions[], size_t *attr_size,
 944                     bool rtnl_held, struct netlink_ext_ack *extack)
 945 {
 946         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 947         struct tc_action *act;
 948         size_t sz = 0;
 949         int err;
 950         int i;
 951
 952         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
 953         if (err < 0)
 954                 return err;
 955
 956         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 957                 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
 958                                         rtnl_held, extack);
 959                 if (IS_ERR(act)) {
 960                         err = PTR_ERR(act);
 961                         goto err;
 962                 }
 963                 act->order = i;
 964                 sz += tcf_action_fill_size(act);
 965                 /* Start from index 0 */
 966                 actions[i - 1] = act;
 967         }
 968
 969         *attr_size = tcf_action_full_attrs_size(sz);
 970         return i - 1;
 971
 972 err:
 973         tcf_action_destroy(actions, bind);
 974         return err;
 975 }
 976
 977 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
 978                           int compat_mode)
 979 {
 980         int err = 0;
 981         struct gnet_dump d;
 982
 983         if (p == NULL)
 984                 goto errout;
 985
 986         /* compat_mode being true specifies a call that is supposed
 987          * to add additional backward compatibility statistic TLVs.
 988          */
 989         if (compat_mode) {
 990                 if (p->type == TCA_OLD_COMPAT)
 991                         err = gnet_stats_start_copy_compat(skb, 0,
 992                                                            TCA_STATS,
 993                                                            TCA_XSTATS,
 994                                                            &p->tcfa_lock, &d,
 995                                                            TCA_PAD);
 996                 else
 997                         return 0;
 998         } else
 999                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
1000                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
1001
1002         if (err < 0)
1003                 goto errout;
1004
1005         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
1006             gnet_stats_copy_basic_hw(NULL, &d, p->cpu_bstats_hw,
1007                                      &p->tcfa_bstats_hw) < 0 ||
1008             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
1009             gnet_stats_copy_queue(&d, p->cpu_qstats,
1010                                   &p->tcfa_qstats,
1011                                   p->tcfa_qstats.qlen) < 0)
1012                 goto errout;
1013
1014         if (gnet_stats_finish_copy(&d) < 0)
1015                 goto errout;
1016
1017         return 0;
1018
1019 errout:
1020         return -1;
1021 }
1022
1023 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1024                         u32 portid, u32 seq, u16 flags, int event, int bind,
1025                         int ref)
1026 {
1027         struct tcamsg *t;
1028         struct nlmsghdr *nlh;
1029         unsigned char *b = skb_tail_pointer(skb);
1030         struct nlattr *nest;
1031
1032         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1033         if (!nlh)
1034                 goto out_nlmsg_trim;
1035         t = nlmsg_data(nlh);
1036         t->tca_family = AF_UNSPEC;
1037         t->tca__pad1 = 0;
1038         t->tca__pad2 = 0;
1039
1040         nest = nla_nest_start(skb, TCA_ACT_TAB);
1041         if (!nest)
1042                 goto out_nlmsg_trim;
1043
1044         if (tcf_action_dump(skb, actions, bind, ref) < 0)
1045                 goto out_nlmsg_trim;
1046
1047         nla_nest_end(skb, nest);
1048
1049         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1050         return skb->len;
1051
1052 out_nlmsg_trim:
1053         nlmsg_trim(skb, b);
1054         return -1;
1055 }
1056
1057 static int
1058 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1059                struct tc_action *actions[], int event,
1060                struct netlink_ext_ack *extack)
1061 {
1062         struct sk_buff *skb;
1063
1064         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1065         if (!skb)
1066                 return -ENOBUFS;
1067         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1068                          0, 1) <= 0) {
1069                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1070                 kfree_skb(skb);
1071                 return -EINVAL;
1072         }
1073
1074         return rtnl_unicast(skb, net, portid);
1075 }
1076
1077 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1078                                           struct nlmsghdr *n, u32 portid,
1079                                           struct netlink_ext_ack *extack)
1080 {
1081         struct nlattr *tb[TCA_ACT_MAX + 1];
1082         const struct tc_action_ops *ops;
1083         struct tc_action *a;
1084         int index;
1085         int err;
1086
1087         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1088         if (err < 0)
1089                 goto err_out;
1090
1091         err = -EINVAL;
1092         if (tb[TCA_ACT_INDEX] == NULL ||
1093             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1094                 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1095                 goto err_out;
1096         }
1097         index = nla_get_u32(tb[TCA_ACT_INDEX]);
1098
1099         err = -EINVAL;
1100         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1101         if (!ops) { /* could happen in batch of actions */
1102                 NL_SET_ERR_MSG(extack, "Specified TC action kind not found");
1103                 goto err_out;
1104         }
1105         err = -ENOENT;
1106         if (ops->lookup(net, &a, index) == 0) {
1107                 NL_SET_ERR_MSG(extack, "TC action with specified index not found");
1108                 goto err_mod;
1109         }
1110
1111         module_put(ops->owner);
1112         return a;
1113
1114 err_mod:
1115         module_put(ops->owner);
1116 err_out:
1117         return ERR_PTR(err);
1118 }
1119
1120 static int tca_action_flush(struct net *net, struct nlattr *nla,
1121                             struct nlmsghdr *n, u32 portid,
1122                             struct netlink_ext_ack *extack)
1123 {
1124         struct sk_buff *skb;
1125         unsigned char *b;
1126         struct nlmsghdr *nlh;
1127         struct tcamsg *t;
1128         struct netlink_callback dcb;
1129         struct nlattr *nest;
1130         struct nlattr *tb[TCA_ACT_MAX + 1];
1131         const struct tc_action_ops *ops;
1132         struct nlattr *kind;
1133         int err = -ENOMEM;
1134
1135         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1136         if (!skb)
1137                 return err;
1138
1139         b = skb_tail_pointer(skb);
1140
1141         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1142         if (err < 0)
1143                 goto err_out;
1144
1145         err = -EINVAL;
1146         kind = tb[TCA_ACT_KIND];
1147         ops = tc_lookup_action(kind);
1148         if (!ops) { /*some idjot trying to flush unknown action */
1149                 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1150                 goto err_out;
1151         }
1152
1153         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1154                         sizeof(*t), 0);
1155         if (!nlh) {
1156                 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1157                 goto out_module_put;
1158         }
1159         t = nlmsg_data(nlh);
1160         t->tca_family = AF_UNSPEC;
1161         t->tca__pad1 = 0;
1162         t->tca__pad2 = 0;
1163
1164         nest = nla_nest_start(skb, TCA_ACT_TAB);
1165         if (!nest) {
1166                 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1167                 goto out_module_put;
1168         }
1169
1170         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1171         if (err <= 0) {
1172                 nla_nest_cancel(skb, nest);
1173                 goto out_module_put;
1174         }
1175
1176         nla_nest_end(skb, nest);
1177
1178         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1179         nlh->nlmsg_flags |= NLM_F_ROOT;
1180         module_put(ops->owner);
1181         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1182                              n->nlmsg_flags & NLM_F_ECHO);
1183         if (err > 0)
1184                 return 0;
1185         if (err < 0)
1186                 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1187
1188         return err;
1189
1190 out_module_put:
1191         module_put(ops->owner);
1192 err_out:
1193         kfree_skb(skb);
1194         return err;
1195 }
1196
1197 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1198 {
1199         int i;
1200
1201         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1202                 struct tc_action *a = actions[i];
1203                 const struct tc_action_ops *ops = a->ops;
1204                 /* Actions can be deleted concurrently so we must save their
1205                  * type and id to search again after reference is released.
1206                  */
1207                 struct tcf_idrinfo *idrinfo = a->idrinfo;
1208                 u32 act_index = a->tcfa_index;
1209
1210                 actions[i] = NULL;
1211                 if (tcf_action_put(a)) {
1212                         /* last reference, action was deleted concurrently */
1213                         module_put(ops->owner);
1214                 } else  {
1215                         int ret;
1216
1217                         /* now do the delete */
1218                         ret = tcf_idr_delete_index(idrinfo, act_index);
1219                         if (ret < 0)
1220                                 return ret;
1221                 }
1222         }
1223         return 0;
1224 }
1225
1226 static int
1227 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1228                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1229 {
1230         int ret;
1231         struct sk_buff *skb;
1232
1233         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1234                         GFP_KERNEL);
1235         if (!skb)
1236                 return -ENOBUFS;
1237
1238         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1239                          0, 2) <= 0) {
1240                 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1241                 kfree_skb(skb);
1242                 return -EINVAL;
1243         }
1244
1245         /* now do the delete */
1246         ret = tcf_action_delete(net, actions);
1247         if (ret < 0) {
1248                 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1249                 kfree_skb(skb);
1250                 return ret;
1251         }
1252
1253         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1254                              n->nlmsg_flags & NLM_F_ECHO);
1255         if (ret > 0)
1256                 return 0;
1257         return ret;
1258 }
1259
1260 static int
1261 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1262               u32 portid, int event, struct netlink_ext_ack *extack)
1263 {
1264         int i, ret;
1265         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1266         struct tc_action *act;
1267         size_t attr_size = 0;
1268         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1269
1270         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
1271         if (ret < 0)
1272                 return ret;
1273
1274         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1275                 if (tb[1])
1276                         return tca_action_flush(net, tb[1], n, portid, extack);
1277
1278                 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1279                 return -EINVAL;
1280         }
1281
1282         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1283                 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1284                 if (IS_ERR(act)) {
1285                         ret = PTR_ERR(act);
1286                         goto err;
1287                 }
1288                 act->order = i;
1289                 attr_size += tcf_action_fill_size(act);
1290                 actions[i - 1] = act;
1291         }
1292
1293         attr_size = tcf_action_full_attrs_size(attr_size);
1294
1295         if (event == RTM_GETACTION)
1296                 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1297         else { /* delete */
1298                 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1299                 if (ret)
1300                         goto err;
1301                 return 0;
1302         }
1303 err:
1304         tcf_action_put_many(actions);
1305         return ret;
1306 }
1307
1308 static int
1309 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1310                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1311 {
1312         struct sk_buff *skb;
1313         int err = 0;
1314
1315         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1316                         GFP_KERNEL);
1317         if (!skb)
1318                 return -ENOBUFS;
1319
1320         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1321                          RTM_NEWACTION, 0, 0) <= 0) {
1322                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1323                 kfree_skb(skb);
1324                 return -EINVAL;
1325         }
1326
1327         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1328                              n->nlmsg_flags & NLM_F_ECHO);
1329         if (err > 0)
1330                 err = 0;
1331         return err;
1332 }
1333
1334 static int tcf_action_add(struct net *net, struct nlattr *nla,
1335                           struct nlmsghdr *n, u32 portid, int ovr,
1336                           struct netlink_ext_ack *extack)
1337 {
1338         size_t attr_size = 0;
1339         int ret = 0;
1340         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1341
1342         ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0, actions,
1343                               &attr_size, true, extack);
1344         if (ret < 0)
1345                 return ret;
1346         ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1347         if (ovr)
1348                 tcf_action_put_many(actions);
1349
1350         return ret;
1351 }
1352
1353 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1354 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1355         [TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1356                              .validation_data = &tcaa_root_flags_allowed },
1357         [TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1358 };
1359
1360 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1361                          struct netlink_ext_ack *extack)
1362 {
1363         struct net *net = sock_net(skb->sk);
1364         struct nlattr *tca[TCA_ROOT_MAX + 1];
1365         u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1366         int ret = 0, ovr = 0;
1367
1368         if ((n->nlmsg_type != RTM_GETACTION) &&
1369             !netlink_capable(skb, CAP_NET_ADMIN))
1370                 return -EPERM;
1371
1372         ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1373                           extack);
1374         if (ret < 0)
1375                 return ret;
1376
1377         if (tca[TCA_ACT_TAB] == NULL) {
1378                 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1379                 return -EINVAL;
1380         }
1381
1382         /* n->nlmsg_flags & NLM_F_CREATE */
1383         switch (n->nlmsg_type) {
1384         case RTM_NEWACTION:
1385                 /* we are going to assume all other flags
1386                  * imply create only if it doesn't exist
1387                  * Note that CREATE | EXCL implies that
1388                  * but since we want avoid ambiguity (eg when flags
1389                  * is zero) then just set this
1390                  */
1391                 if (n->nlmsg_flags & NLM_F_REPLACE)
1392                         ovr = 1;
1393 replay:
1394                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1395                                      extack);
1396                 if (ret == -EAGAIN)
1397                         goto replay;
1398                 break;
1399         case RTM_DELACTION:
1400                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1401                                     portid, RTM_DELACTION, extack);
1402                 break;
1403         case RTM_GETACTION:
1404                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1405                                     portid, RTM_GETACTION, extack);
1406                 break;
1407         default:
1408                 BUG();
1409         }
1410
1411         return ret;
1412 }
1413
1414 static struct nlattr *find_dump_kind(struct nlattr **nla)
1415 {
1416         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1417         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1418         struct nlattr *kind;
1419
1420         tb1 = nla[TCA_ACT_TAB];
1421         if (tb1 == NULL)
1422                 return NULL;
1423
1424         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1425                       NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1426                 return NULL;
1427
1428         if (tb[1] == NULL)
1429                 return NULL;
1430         if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1431                 return NULL;
1432         kind = tb2[TCA_ACT_KIND];
1433
1434         return kind;
1435 }
1436
1437 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1438 {
1439         struct net *net = sock_net(skb->sk);
1440         struct nlmsghdr *nlh;
1441         unsigned char *b = skb_tail_pointer(skb);
1442         struct nlattr *nest;
1443         struct tc_action_ops *a_o;
1444         int ret = 0;
1445         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1446         struct nlattr *tb[TCA_ROOT_MAX + 1];
1447         struct nlattr *count_attr = NULL;
1448         unsigned long jiffy_since = 0;
1449         struct nlattr *kind = NULL;
1450         struct nla_bitfield32 bf;
1451         u32 msecs_since = 0;
1452         u32 act_count = 0;
1453
1454         ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1455                           tcaa_policy, cb->extack);
1456         if (ret < 0)
1457                 return ret;
1458
1459         kind = find_dump_kind(tb);
1460         if (kind == NULL) {
1461                 pr_info("tc_dump_action: action bad kind\n");
1462                 return 0;
1463         }
1464
1465         a_o = tc_lookup_action(kind);
1466         if (a_o == NULL)
1467                 return 0;
1468
1469         cb->args[2] = 0;
1470         if (tb[TCA_ROOT_FLAGS]) {
1471                 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1472                 cb->args[2] = bf.value;
1473         }
1474
1475         if (tb[TCA_ROOT_TIME_DELTA]) {
1476                 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1477         }
1478
1479         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1480                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1481         if (!nlh)
1482                 goto out_module_put;
1483
1484         if (msecs_since)
1485                 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1486
1487         t = nlmsg_data(nlh);
1488         t->tca_family = AF_UNSPEC;
1489         t->tca__pad1 = 0;
1490         t->tca__pad2 = 0;
1491         cb->args[3] = jiffy_since;
1492         count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1493         if (!count_attr)
1494                 goto out_module_put;
1495
1496         nest = nla_nest_start(skb, TCA_ACT_TAB);
1497         if (nest == NULL)
1498                 goto out_module_put;
1499
1500         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1501         if (ret < 0)
1502                 goto out_module_put;
1503
1504         if (ret > 0) {
1505                 nla_nest_end(skb, nest);
1506                 ret = skb->len;
1507                 act_count = cb->args[1];
1508                 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1509                 cb->args[1] = 0;
1510         } else
1511                 nlmsg_trim(skb, b);
1512
1513         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1514         if (NETLINK_CB(cb->skb).portid && ret)
1515                 nlh->nlmsg_flags |= NLM_F_MULTI;
1516         module_put(a_o->owner);
1517         return skb->len;
1518
1519 out_module_put:
1520         module_put(a_o->owner);
1521         nlmsg_trim(skb, b);
1522         return skb->len;
1523 }
1524
1525 struct tcf_action_net {
1526         struct rhashtable egdev_ht;
1527 };
1528
1529 static unsigned int tcf_action_net_id;
1530
1531 struct tcf_action_egdev_cb {
1532         struct list_head list;
1533         tc_setup_cb_t *cb;
1534         void *cb_priv;
1535 };
1536
1537 struct tcf_action_egdev {
1538         struct rhash_head ht_node;
1539         const struct net_device *dev;
1540         unsigned int refcnt;
1541         struct list_head cb_list;
1542 };
1543
1544 static const struct rhashtable_params tcf_action_egdev_ht_params = {
1545         .key_offset = offsetof(struct tcf_action_egdev, dev),
1546         .head_offset = offsetof(struct tcf_action_egdev, ht_node),
1547         .key_len = sizeof(const struct net_device *),
1548 };
1549
1550 static struct tcf_action_egdev *
1551 tcf_action_egdev_lookup(const struct net_device *dev)
1552 {
1553         struct net *net = dev_net(dev);
1554         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1555
1556         return rhashtable_lookup_fast(&tan->egdev_ht, &dev,
1557                                       tcf_action_egdev_ht_params);
1558 }
1559
1560 static struct tcf_action_egdev *
1561 tcf_action_egdev_get(const struct net_device *dev)
1562 {
1563         struct tcf_action_egdev *egdev;
1564         struct tcf_action_net *tan;
1565
1566         egdev = tcf_action_egdev_lookup(dev);
1567         if (egdev)
1568                 goto inc_ref;
1569
1570         egdev = kzalloc(sizeof(*egdev), GFP_KERNEL);
1571         if (!egdev)
1572                 return NULL;
1573         INIT_LIST_HEAD(&egdev->cb_list);
1574         egdev->dev = dev;
1575         tan = net_generic(dev_net(dev), tcf_action_net_id);
1576         rhashtable_insert_fast(&tan->egdev_ht, &egdev->ht_node,
1577                                tcf_action_egdev_ht_params);
1578
1579 inc_ref:
1580         egdev->refcnt++;
1581         return egdev;
1582 }
1583
1584 static void tcf_action_egdev_put(struct tcf_action_egdev *egdev)
1585 {
1586         struct tcf_action_net *tan;
1587
1588         if (--egdev->refcnt)
1589                 return;
1590         tan = net_generic(dev_net(egdev->dev), tcf_action_net_id);
1591         rhashtable_remove_fast(&tan->egdev_ht, &egdev->ht_node,
1592                                tcf_action_egdev_ht_params);
1593         kfree(egdev);
1594 }
1595
1596 static struct tcf_action_egdev_cb *
1597 tcf_action_egdev_cb_lookup(struct tcf_action_egdev *egdev,
1598                            tc_setup_cb_t *cb, void *cb_priv)
1599 {
1600         struct tcf_action_egdev_cb *egdev_cb;
1601
1602         list_for_each_entry(egdev_cb, &egdev->cb_list, list)
1603                 if (egdev_cb->cb == cb && egdev_cb->cb_priv == cb_priv)
1604                         return egdev_cb;
1605         return NULL;
1606 }
1607
1608 static int tcf_action_egdev_cb_call(struct tcf_action_egdev *egdev,
1609                                     enum tc_setup_type type,
1610                                     void *type_data, bool err_stop)
1611 {
1612         struct tcf_action_egdev_cb *egdev_cb;
1613         int ok_count = 0;
1614         int err;
1615
1616         list_for_each_entry(egdev_cb, &egdev->cb_list, list) {
1617                 err = egdev_cb->cb(type, type_data, egdev_cb->cb_priv);
1618                 if (err) {
1619                         if (err_stop)
1620                                 return err;
1621                 } else {
1622                         ok_count++;
1623                 }
1624         }
1625         return ok_count;
1626 }
1627
1628 static int tcf_action_egdev_cb_add(struct tcf_action_egdev *egdev,
1629                                    tc_setup_cb_t *cb, void *cb_priv)
1630 {
1631         struct tcf_action_egdev_cb *egdev_cb;
1632
1633         egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1634         if (WARN_ON(egdev_cb))
1635                 return -EEXIST;
1636         egdev_cb = kzalloc(sizeof(*egdev_cb), GFP_KERNEL);
1637         if (!egdev_cb)
1638                 return -ENOMEM;
1639         egdev_cb->cb = cb;
1640         egdev_cb->cb_priv = cb_priv;
1641         list_add(&egdev_cb->list, &egdev->cb_list);
1642         return 0;
1643 }
1644
1645 static void tcf_action_egdev_cb_del(struct tcf_action_egdev *egdev,
1646                                     tc_setup_cb_t *cb, void *cb_priv)
1647 {
1648         struct tcf_action_egdev_cb *egdev_cb;
1649
1650         egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1651         if (WARN_ON(!egdev_cb))
1652                 return;
1653         list_del(&egdev_cb->list);
1654         kfree(egdev_cb);
1655 }
1656
1657 static int __tc_setup_cb_egdev_register(const struct net_device *dev,
1658                                         tc_setup_cb_t *cb, void *cb_priv)
1659 {
1660         struct tcf_action_egdev *egdev = tcf_action_egdev_get(dev);
1661         int err;
1662
1663         if (!egdev)
1664                 return -ENOMEM;
1665         err = tcf_action_egdev_cb_add(egdev, cb, cb_priv);
1666         if (err)
1667                 goto err_cb_add;
1668         return 0;
1669
1670 err_cb_add:
1671         tcf_action_egdev_put(egdev);
1672         return err;
1673 }
1674 int tc_setup_cb_egdev_register(const struct net_device *dev,
1675                                tc_setup_cb_t *cb, void *cb_priv)
1676 {
1677         int err;
1678
1679         rtnl_lock();
1680         err = __tc_setup_cb_egdev_register(dev, cb, cb_priv);
1681         rtnl_unlock();
1682         return err;
1683 }
1684 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_register);
1685
1686 static void __tc_setup_cb_egdev_unregister(const struct net_device *dev,
1687                                            tc_setup_cb_t *cb, void *cb_priv)
1688 {
1689         struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1690
1691         if (WARN_ON(!egdev))
1692                 return;
1693         tcf_action_egdev_cb_del(egdev, cb, cb_priv);
1694         tcf_action_egdev_put(egdev);
1695 }
1696 void tc_setup_cb_egdev_unregister(const struct net_device *dev,
1697                                   tc_setup_cb_t *cb, void *cb_priv)
1698 {
1699         rtnl_lock();
1700         __tc_setup_cb_egdev_unregister(dev, cb, cb_priv);
1701         rtnl_unlock();
1702 }
1703 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_unregister);
1704
1705 int tc_setup_cb_egdev_call(const struct net_device *dev,
1706                            enum tc_setup_type type, void *type_data,
1707                            bool err_stop)
1708 {
1709         struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1710
1711         if (!egdev)
1712                 return 0;
1713         return tcf_action_egdev_cb_call(egdev, type, type_data, err_stop);
1714 }
1715 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_call);
1716
1717 static __net_init int tcf_action_net_init(struct net *net)
1718 {
1719         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1720
1721         return rhashtable_init(&tan->egdev_ht, &tcf_action_egdev_ht_params);
1722 }
1723
1724 static void __net_exit tcf_action_net_exit(struct net *net)
1725 {
1726         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1727
1728         rhashtable_destroy(&tan->egdev_ht);
1729 }
1730
1731 static struct pernet_operations tcf_action_net_ops = {
1732         .init = tcf_action_net_init,
1733         .exit = tcf_action_net_exit,
1734         .id = &tcf_action_net_id,
1735         .size = sizeof(struct tcf_action_net),
1736 };
1737
1738 static int __init tc_action_init(void)
1739 {
1740         int err;
1741
1742         err = register_pernet_subsys(&tcf_action_net_ops);
1743         if (err)
1744                 return err;
1745
1746         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1747         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1748         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1749                       0);
1750
1751         return 0;
1752 }
1753
1754 subsys_initcall(tc_action_init);