net/sched/act_api.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/act_api.c  Packet action API.
   4  *
   5  * Author:      Jamal Hadi Salim
   6  */
   7
   8 #include <linux/types.h>
   9 #include <linux/kernel.h>
  10 #include <linux/string.h>
  11 #include <linux/errno.h>
  12 #include <linux/slab.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/init.h>
  15 #include <linux/kmod.h>
  16 #include <linux/err.h>
  17 #include <linux/module.h>
  18 #include <net/net_namespace.h>
  19 #include <net/sock.h>
  20 #include <net/sch_generic.h>
  21 #include <net/pkt_cls.h>
  22 #include <net/act_api.h>
  23 #include <net/netlink.h>
  24
  25 static void tcf_action_goto_chain_exec(const struct tc_action *a,
  26                                        struct tcf_result *res)
  27 {
  28         const struct tcf_chain *chain = rcu_dereference_bh(a->goto_chain);
  29
  30         res->goto_tp = rcu_dereference_bh(chain->filter_chain);
  31 }
  32
  33 static void tcf_free_cookie_rcu(struct rcu_head *p)
  34 {
  35         struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
  36
  37         kfree(cookie->data);
  38         kfree(cookie);
  39 }
  40
  41 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
  42                                   struct tc_cookie *new_cookie)
  43 {
  44         struct tc_cookie *old;
  45
  46         old = xchg((__force struct tc_cookie **)old_cookie, new_cookie);
  47         if (old)
  48                 call_rcu(&old->rcu, tcf_free_cookie_rcu);
  49 }
  50
  51 int tcf_action_check_ctrlact(int action, struct tcf_proto *tp,
  52                              struct tcf_chain **newchain,
  53                              struct netlink_ext_ack *extack)
  54 {
  55         int opcode = TC_ACT_EXT_OPCODE(action), ret = -EINVAL;
  56         u32 chain_index;
  57
  58         if (!opcode)
  59                 ret = action > TC_ACT_VALUE_MAX ? -EINVAL : 0;
  60         else if (opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC)
  61                 ret = 0;
  62         if (ret) {
  63                 NL_SET_ERR_MSG(extack, "invalid control action");
  64                 goto end;
  65         }
  66
  67         if (TC_ACT_EXT_CMP(action, TC_ACT_GOTO_CHAIN)) {
  68                 chain_index = action & TC_ACT_EXT_VAL_MASK;
  69                 if (!tp || !newchain) {
  70                         ret = -EINVAL;
  71                         NL_SET_ERR_MSG(extack,
  72                                        "can't goto NULL proto/chain");
  73                         goto end;
  74                 }
  75                 *newchain = tcf_chain_get_by_act(tp->chain->block, chain_index);
  76                 if (!*newchain) {
  77                         ret = -ENOMEM;
  78                         NL_SET_ERR_MSG(extack,
  79                                        "can't allocate goto_chain");
  80                 }
  81         }
  82 end:
  83         return ret;
  84 }
  85 EXPORT_SYMBOL(tcf_action_check_ctrlact);
  86
  87 struct tcf_chain *tcf_action_set_ctrlact(struct tc_action *a, int action,
  88                                          struct tcf_chain *goto_chain)
  89 {
  90         a->tcfa_action = action;
  91         goto_chain = rcu_replace_pointer(a->goto_chain, goto_chain, 1);
  92         return goto_chain;
  93 }
  94 EXPORT_SYMBOL(tcf_action_set_ctrlact);
  95
  96 /* XXX: For standalone actions, we don't need a RCU grace period either, because
  97  * actions are always connected to filters and filters are already destroyed in
  98  * RCU callbacks, so after a RCU grace period actions are already disconnected
  99  * from filters. Readers later can not find us.
 100  */
 101 static void free_tcf(struct tc_action *p)
 102 {
 103         struct tcf_chain *chain = rcu_dereference_protected(p->goto_chain, 1);
 104
 105         free_percpu(p->cpu_bstats);
 106         free_percpu(p->cpu_bstats_hw);
 107         free_percpu(p->cpu_qstats);
 108
 109         tcf_set_action_cookie(&p->act_cookie, NULL);
 110         if (chain)
 111                 tcf_chain_put_by_act(chain);
 112
 113         kfree(p);
 114 }
 115
 116 static void tcf_action_cleanup(struct tc_action *p)
 117 {
 118         if (p->ops->cleanup)
 119                 p->ops->cleanup(p);
 120
 121         gen_kill_estimator(&p->tcfa_rate_est);
 122         free_tcf(p);
 123 }
 124
 125 static int __tcf_action_put(struct tc_action *p, bool bind)
 126 {
 127         struct tcf_idrinfo *idrinfo = p->idrinfo;
 128
 129         if (refcount_dec_and_mutex_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
 130                 if (bind)
 131                         atomic_dec(&p->tcfa_bindcnt);
 132                 idr_remove(&idrinfo->action_idr, p->tcfa_index);
 133                 mutex_unlock(&idrinfo->lock);
 134
 135                 tcf_action_cleanup(p);
 136                 return 1;
 137         }
 138
 139         if (bind)
 140                 atomic_dec(&p->tcfa_bindcnt);
 141
 142         return 0;
 143 }
 144
 145 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
 146 {
 147         int ret = 0;
 148
 149         /* Release with strict==1 and bind==0 is only called through act API
 150          * interface (classifiers always bind). Only case when action with
 151          * positive reference count and zero bind count can exist is when it was
 152          * also created with act API (unbinding last classifier will destroy the
 153          * action if it was created by classifier). So only case when bind count
 154          * can be changed after initial check is when unbound action is
 155          * destroyed by act API while classifier binds to action with same id
 156          * concurrently. This result either creation of new action(same behavior
 157          * as before), or reusing existing action if concurrent process
 158          * increments reference count before action is deleted. Both scenarios
 159          * are acceptable.
 160          */
 161         if (p) {
 162                 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
 163                         return -EPERM;
 164
 165                 if (__tcf_action_put(p, bind))
 166                         ret = ACT_P_DELETED;
 167         }
 168
 169         return ret;
 170 }
 171 EXPORT_SYMBOL(__tcf_idr_release);
 172
 173 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
 174 {
 175         struct tc_cookie *act_cookie;
 176         u32 cookie_len = 0;
 177
 178         rcu_read_lock();
 179         act_cookie = rcu_dereference(act->act_cookie);
 180
 181         if (act_cookie)
 182                 cookie_len = nla_total_size(act_cookie->len);
 183         rcu_read_unlock();
 184
 185         return  nla_total_size(0) /* action number nested */
 186                 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
 187                 + cookie_len /* TCA_ACT_COOKIE */
 188                 + nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_HW_STATS */
 189                 + nla_total_size(0) /* TCA_ACT_STATS nested */
 190                 + nla_total_size(sizeof(struct nla_bitfield32)) /* TCA_ACT_FLAGS */
 191                 /* TCA_STATS_BASIC */
 192                 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
 193                 /* TCA_STATS_PKT64 */
 194                 + nla_total_size_64bit(sizeof(u64))
 195                 /* TCA_STATS_QUEUE */
 196                 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
 197                 + nla_total_size(0) /* TCA_OPTIONS nested */
 198                 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
 199 }
 200
 201 static size_t tcf_action_full_attrs_size(size_t sz)
 202 {
 203         return NLMSG_HDRLEN                     /* struct nlmsghdr */
 204                 + sizeof(struct tcamsg)
 205                 + nla_total_size(0)             /* TCA_ACT_TAB nested */
 206                 + sz;
 207 }
 208
 209 static size_t tcf_action_fill_size(const struct tc_action *act)
 210 {
 211         size_t sz = tcf_action_shared_attrs_size(act);
 212
 213         if (act->ops->get_fill_size)
 214                 return act->ops->get_fill_size(act) + sz;
 215         return sz;
 216 }
 217
 218 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 219                            struct netlink_callback *cb)
 220 {
 221         int err = 0, index = -1, s_i = 0, n_i = 0;
 222         u32 act_flags = cb->args[2];
 223         unsigned long jiffy_since = cb->args[3];
 224         struct nlattr *nest;
 225         struct idr *idr = &idrinfo->action_idr;
 226         struct tc_action *p;
 227         unsigned long id = 1;
 228         unsigned long tmp;
 229
 230         mutex_lock(&idrinfo->lock);
 231
 232         s_i = cb->args[0];
 233
 234         idr_for_each_entry_ul(idr, p, tmp, id) {
 235                 index++;
 236                 if (index < s_i)
 237                         continue;
 238                 if (IS_ERR(p))
 239                         continue;
 240
 241                 if (jiffy_since &&
 242                     time_after(jiffy_since,
 243                                (unsigned long)p->tcfa_tm.lastuse))
 244                         continue;
 245
 246                 nest = nla_nest_start_noflag(skb, n_i);
 247                 if (!nest) {
 248                         index--;
 249                         goto nla_put_failure;
 250                 }
 251                 err = tcf_action_dump_1(skb, p, 0, 0);
 252                 if (err < 0) {
 253                         index--;
 254                         nlmsg_trim(skb, nest);
 255                         goto done;
 256                 }
 257                 nla_nest_end(skb, nest);
 258                 n_i++;
 259                 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
 260                     n_i >= TCA_ACT_MAX_PRIO)
 261                         goto done;
 262         }
 263 done:
 264         if (index >= 0)
 265                 cb->args[0] = index + 1;
 266
 267         mutex_unlock(&idrinfo->lock);
 268         if (n_i) {
 269                 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
 270                         cb->args[1] = n_i;
 271         }
 272         return n_i;
 273
 274 nla_put_failure:
 275         nla_nest_cancel(skb, nest);
 276         goto done;
 277 }
 278
 279 static int tcf_idr_release_unsafe(struct tc_action *p)
 280 {
 281         if (atomic_read(&p->tcfa_bindcnt) > 0)
 282                 return -EPERM;
 283
 284         if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 285                 idr_remove(&p->idrinfo->action_idr, p->tcfa_index);
 286                 tcf_action_cleanup(p);
 287                 return ACT_P_DELETED;
 288         }
 289
 290         return 0;
 291 }
 292
 293 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 294                           const struct tc_action_ops *ops)
 295 {
 296         struct nlattr *nest;
 297         int n_i = 0;
 298         int ret = -EINVAL;
 299         struct idr *idr = &idrinfo->action_idr;
 300         struct tc_action *p;
 301         unsigned long id = 1;
 302         unsigned long tmp;
 303
 304         nest = nla_nest_start_noflag(skb, 0);
 305         if (nest == NULL)
 306                 goto nla_put_failure;
 307         if (nla_put_string(skb, TCA_KIND, ops->kind))
 308                 goto nla_put_failure;
 309
 310         mutex_lock(&idrinfo->lock);
 311         idr_for_each_entry_ul(idr, p, tmp, id) {
 312                 if (IS_ERR(p))
 313                         continue;
 314                 ret = tcf_idr_release_unsafe(p);
 315                 if (ret == ACT_P_DELETED) {
 316                         module_put(ops->owner);
 317                         n_i++;
 318                 } else if (ret < 0) {
 319                         mutex_unlock(&idrinfo->lock);
 320                         goto nla_put_failure;
 321                 }
 322         }
 323         mutex_unlock(&idrinfo->lock);
 324
 325         if (nla_put_u32(skb, TCA_FCNT, n_i))
 326                 goto nla_put_failure;
 327         nla_nest_end(skb, nest);
 328
 329         return n_i;
 330 nla_put_failure:
 331         nla_nest_cancel(skb, nest);
 332         return ret;
 333 }
 334
 335 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 336                        struct netlink_callback *cb, int type,
 337                        const struct tc_action_ops *ops,
 338                        struct netlink_ext_ack *extack)
 339 {
 340         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 341
 342         if (type == RTM_DELACTION) {
 343                 return tcf_del_walker(idrinfo, skb, ops);
 344         } else if (type == RTM_GETACTION) {
 345                 return tcf_dump_walker(idrinfo, skb, cb);
 346         } else {
 347                 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
 348                 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
 349                 return -EINVAL;
 350         }
 351 }
 352 EXPORT_SYMBOL(tcf_generic_walker);
 353
 354 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 355 {
 356         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 357         struct tc_action *p;
 358
 359         mutex_lock(&idrinfo->lock);
 360         p = idr_find(&idrinfo->action_idr, index);
 361         if (IS_ERR(p))
 362                 p = NULL;
 363         else if (p)
 364                 refcount_inc(&p->tcfa_refcnt);
 365         mutex_unlock(&idrinfo->lock);
 366
 367         if (p) {
 368                 *a = p;
 369                 return true;
 370         }
 371         return false;
 372 }
 373 EXPORT_SYMBOL(tcf_idr_search);
 374
 375 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
 376 {
 377         struct tc_action *p;
 378         int ret = 0;
 379
 380         mutex_lock(&idrinfo->lock);
 381         p = idr_find(&idrinfo->action_idr, index);
 382         if (!p) {
 383                 mutex_unlock(&idrinfo->lock);
 384                 return -ENOENT;
 385         }
 386
 387         if (!atomic_read(&p->tcfa_bindcnt)) {
 388                 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
 389                         struct module *owner = p->ops->owner;
 390
 391                         WARN_ON(p != idr_remove(&idrinfo->action_idr,
 392                                                 p->tcfa_index));
 393                         mutex_unlock(&idrinfo->lock);
 394
 395                         tcf_action_cleanup(p);
 396                         module_put(owner);
 397                         return 0;
 398                 }
 399                 ret = 0;
 400         } else {
 401                 ret = -EPERM;
 402         }
 403
 404         mutex_unlock(&idrinfo->lock);
 405         return ret;
 406 }
 407
 408 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 409                    struct tc_action **a, const struct tc_action_ops *ops,
 410                    int bind, bool cpustats, u32 flags)
 411 {
 412         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 413         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 414         int err = -ENOMEM;
 415
 416         if (unlikely(!p))
 417                 return -ENOMEM;
 418         refcount_set(&p->tcfa_refcnt, 1);
 419         if (bind)
 420                 atomic_set(&p->tcfa_bindcnt, 1);
 421
 422         if (cpustats) {
 423                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 424                 if (!p->cpu_bstats)
 425                         goto err1;
 426                 p->cpu_bstats_hw = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 427                 if (!p->cpu_bstats_hw)
 428                         goto err2;
 429                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 430                 if (!p->cpu_qstats)
 431                         goto err3;
 432         }
 433         spin_lock_init(&p->tcfa_lock);
 434         p->tcfa_index = index;
 435         p->tcfa_tm.install = jiffies;
 436         p->tcfa_tm.lastuse = jiffies;
 437         p->tcfa_tm.firstuse = 0;
 438         p->tcfa_flags = flags;
 439         if (est) {
 440                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 441                                         &p->tcfa_rate_est,
 442                                         &p->tcfa_lock, NULL, est);
 443                 if (err)
 444                         goto err4;
 445         }
 446
 447         p->idrinfo = idrinfo;
 448         p->ops = ops;
 449         *a = p;
 450         return 0;
 451 err4:
 452         free_percpu(p->cpu_qstats);
 453 err3:
 454         free_percpu(p->cpu_bstats_hw);
 455 err2:
 456         free_percpu(p->cpu_bstats);
 457 err1:
 458         kfree(p);
 459         return err;
 460 }
 461 EXPORT_SYMBOL(tcf_idr_create);
 462
 463 int tcf_idr_create_from_flags(struct tc_action_net *tn, u32 index,
 464                               struct nlattr *est, struct tc_action **a,
 465                               const struct tc_action_ops *ops, int bind,
 466                               u32 flags)
 467 {
 468         /* Set cpustats according to actions flags. */
 469         return tcf_idr_create(tn, index, est, a, ops, bind,
 470                               !(flags & TCA_ACT_FLAGS_NO_PERCPU_STATS), flags);
 471 }
 472 EXPORT_SYMBOL(tcf_idr_create_from_flags);
 473
 474 /* Cleanup idr index that was allocated but not initialized. */
 475
 476 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
 477 {
 478         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 479
 480         mutex_lock(&idrinfo->lock);
 481         /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
 482         WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
 483         mutex_unlock(&idrinfo->lock);
 484 }
 485 EXPORT_SYMBOL(tcf_idr_cleanup);
 486
 487 /* Check if action with specified index exists. If actions is found, increments
 488  * its reference and bind counters, and return 1. Otherwise insert temporary
 489  * error pointer (to prevent concurrent users from inserting actions with same
 490  * index) and return 0.
 491  */
 492
 493 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
 494                         struct tc_action **a, int bind)
 495 {
 496         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 497         struct tc_action *p;
 498         int ret;
 499
 500 again:
 501         mutex_lock(&idrinfo->lock);
 502         if (*index) {
 503                 p = idr_find(&idrinfo->action_idr, *index);
 504                 if (IS_ERR(p)) {
 505                         /* This means that another process allocated
 506                          * index but did not assign the pointer yet.
 507                          */
 508                         mutex_unlock(&idrinfo->lock);
 509                         goto again;
 510                 }
 511
 512                 if (p) {
 513                         refcount_inc(&p->tcfa_refcnt);
 514                         if (bind)
 515                                 atomic_inc(&p->tcfa_bindcnt);
 516                         *a = p;
 517                         ret = 1;
 518                 } else {
 519                         *a = NULL;
 520                         ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 521                                             *index, GFP_KERNEL);
 522                         if (!ret)
 523                                 idr_replace(&idrinfo->action_idr,
 524                                             ERR_PTR(-EBUSY), *index);
 525                 }
 526         } else {
 527                 *index = 1;
 528                 *a = NULL;
 529                 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
 530                                     UINT_MAX, GFP_KERNEL);
 531                 if (!ret)
 532                         idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
 533                                     *index);
 534         }
 535         mutex_unlock(&idrinfo->lock);
 536         return ret;
 537 }
 538 EXPORT_SYMBOL(tcf_idr_check_alloc);
 539
 540 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
 541                          struct tcf_idrinfo *idrinfo)
 542 {
 543         struct idr *idr = &idrinfo->action_idr;
 544         struct tc_action *p;
 545         int ret;
 546         unsigned long id = 1;
 547         unsigned long tmp;
 548
 549         idr_for_each_entry_ul(idr, p, tmp, id) {
 550                 ret = __tcf_idr_release(p, false, true);
 551                 if (ret == ACT_P_DELETED)
 552                         module_put(ops->owner);
 553                 else if (ret < 0)
 554                         return;
 555         }
 556         idr_destroy(&idrinfo->action_idr);
 557 }
 558 EXPORT_SYMBOL(tcf_idrinfo_destroy);
 559
 560 static LIST_HEAD(act_base);
 561 static DEFINE_RWLOCK(act_mod_lock);
 562
 563 int tcf_register_action(struct tc_action_ops *act,
 564                         struct pernet_operations *ops)
 565 {
 566         struct tc_action_ops *a;
 567         int ret;
 568
 569         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 570                 return -EINVAL;
 571
 572         /* We have to register pernet ops before making the action ops visible,
 573          * otherwise tcf_action_init_1() could get a partially initialized
 574          * netns.
 575          */
 576         ret = register_pernet_subsys(ops);
 577         if (ret)
 578                 return ret;
 579
 580         write_lock(&act_mod_lock);
 581         list_for_each_entry(a, &act_base, head) {
 582                 if (act->id == a->id || (strcmp(act->kind, a->kind) == 0)) {
 583                         write_unlock(&act_mod_lock);
 584                         unregister_pernet_subsys(ops);
 585                         return -EEXIST;
 586                 }
 587         }
 588         list_add_tail(&act->head, &act_base);
 589         write_unlock(&act_mod_lock);
 590
 591         return 0;
 592 }
 593 EXPORT_SYMBOL(tcf_register_action);
 594
 595 int tcf_unregister_action(struct tc_action_ops *act,
 596                           struct pernet_operations *ops)
 597 {
 598         struct tc_action_ops *a;
 599         int err = -ENOENT;
 600
 601         write_lock(&act_mod_lock);
 602         list_for_each_entry(a, &act_base, head) {
 603                 if (a == act) {
 604                         list_del(&act->head);
 605                         err = 0;
 606                         break;
 607                 }
 608         }
 609         write_unlock(&act_mod_lock);
 610         if (!err)
 611                 unregister_pernet_subsys(ops);
 612         return err;
 613 }
 614 EXPORT_SYMBOL(tcf_unregister_action);
 615
 616 /* lookup by name */
 617 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 618 {
 619         struct tc_action_ops *a, *res = NULL;
 620
 621         if (kind) {
 622                 read_lock(&act_mod_lock);
 623                 list_for_each_entry(a, &act_base, head) {
 624                         if (strcmp(kind, a->kind) == 0) {
 625                                 if (try_module_get(a->owner))
 626                                         res = a;
 627                                 break;
 628                         }
 629                 }
 630                 read_unlock(&act_mod_lock);
 631         }
 632         return res;
 633 }
 634
 635 /* lookup by nlattr */
 636 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 637 {
 638         struct tc_action_ops *a, *res = NULL;
 639
 640         if (kind) {
 641                 read_lock(&act_mod_lock);
 642                 list_for_each_entry(a, &act_base, head) {
 643                         if (nla_strcmp(kind, a->kind) == 0) {
 644                                 if (try_module_get(a->owner))
 645                                         res = a;
 646                                 break;
 647                         }
 648                 }
 649                 read_unlock(&act_mod_lock);
 650         }
 651         return res;
 652 }
 653
 654 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
 655 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
 656 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 657                     int nr_actions, struct tcf_result *res)
 658 {
 659         u32 jmp_prgcnt = 0;
 660         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
 661         int i;
 662         int ret = TC_ACT_OK;
 663
 664         if (skb_skip_tc_classify(skb))
 665                 return TC_ACT_OK;
 666
 667 restart_act_graph:
 668         for (i = 0; i < nr_actions; i++) {
 669                 const struct tc_action *a = actions[i];
 670
 671                 if (jmp_prgcnt > 0) {
 672                         jmp_prgcnt -= 1;
 673                         continue;
 674                 }
 675 repeat:
 676                 ret = a->ops->act(skb, a, res);
 677                 if (ret == TC_ACT_REPEAT)
 678                         goto repeat;    /* we need a ttl - JHS */
 679
 680                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
 681                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
 682                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
 683                                 /* faulty opcode, stop pipeline */
 684                                 return TC_ACT_OK;
 685                         } else {
 686                                 jmp_ttl -= 1;
 687                                 if (jmp_ttl > 0)
 688                                         goto restart_act_graph;
 689                                 else /* faulty graph, stop pipeline */
 690                                         return TC_ACT_OK;
 691                         }
 692                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
 693                         if (unlikely(!rcu_access_pointer(a->goto_chain))) {
 694                                 net_warn_ratelimited("can't go to NULL chain!\n");
 695                                 return TC_ACT_SHOT;
 696                         }
 697                         tcf_action_goto_chain_exec(a, res);
 698                 }
 699
 700                 if (ret != TC_ACT_PIPE)
 701                         break;
 702         }
 703
 704         return ret;
 705 }
 706 EXPORT_SYMBOL(tcf_action_exec);
 707
 708 int tcf_action_destroy(struct tc_action *actions[], int bind)
 709 {
 710         const struct tc_action_ops *ops;
 711         struct tc_action *a;
 712         int ret = 0, i;
 713
 714         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 715                 a = actions[i];
 716                 actions[i] = NULL;
 717                 ops = a->ops;
 718                 ret = __tcf_idr_release(a, bind, true);
 719                 if (ret == ACT_P_DELETED)
 720                         module_put(ops->owner);
 721                 else if (ret < 0)
 722                         return ret;
 723         }
 724         return ret;
 725 }
 726
 727 static int tcf_action_put(struct tc_action *p)
 728 {
 729         return __tcf_action_put(p, false);
 730 }
 731
 732 /* Put all actions in this array, skip those NULL's. */
 733 static void tcf_action_put_many(struct tc_action *actions[])
 734 {
 735         int i;
 736
 737         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 738                 struct tc_action *a = actions[i];
 739                 const struct tc_action_ops *ops;
 740
 741                 if (!a)
 742                         continue;
 743                 ops = a->ops;
 744                 if (tcf_action_put(a))
 745                         module_put(ops->owner);
 746         }
 747 }
 748
 749 int
 750 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 751 {
 752         return a->ops->dump(skb, a, bind, ref);
 753 }
 754
 755 static int
 756 tcf_action_dump_terse(struct sk_buff *skb, struct tc_action *a)
 757 {
 758         unsigned char *b = skb_tail_pointer(skb);
 759         struct tc_cookie *cookie;
 760
 761         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 762                 goto nla_put_failure;
 763         if (tcf_action_copy_stats(skb, a, 0))
 764                 goto nla_put_failure;
 765
 766         rcu_read_lock();
 767         cookie = rcu_dereference(a->act_cookie);
 768         if (cookie) {
 769                 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
 770                         rcu_read_unlock();
 771                         goto nla_put_failure;
 772                 }
 773         }
 774         rcu_read_unlock();
 775
 776         return 0;
 777
 778 nla_put_failure:
 779         nlmsg_trim(skb, b);
 780         return -1;
 781 }
 782
 783 int
 784 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 785 {
 786         int err = -EINVAL;
 787         unsigned char *b = skb_tail_pointer(skb);
 788         struct nlattr *nest;
 789
 790         if (tcf_action_dump_terse(skb, a))
 791                 goto nla_put_failure;
 792
 793         if (a->hw_stats != TCA_ACT_HW_STATS_ANY &&
 794             nla_put_bitfield32(skb, TCA_ACT_HW_STATS,
 795                                a->hw_stats, TCA_ACT_HW_STATS_ANY))
 796                 goto nla_put_failure;
 797
 798         if (a->used_hw_stats_valid &&
 799             nla_put_bitfield32(skb, TCA_ACT_USED_HW_STATS,
 800                                a->used_hw_stats, TCA_ACT_HW_STATS_ANY))
 801                 goto nla_put_failure;
 802
 803         if (a->tcfa_flags &&
 804             nla_put_bitfield32(skb, TCA_ACT_FLAGS,
 805                                a->tcfa_flags, a->tcfa_flags))
 806                 goto nla_put_failure;
 807
 808         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
 809         if (nest == NULL)
 810                 goto nla_put_failure;
 811         err = tcf_action_dump_old(skb, a, bind, ref);
 812         if (err > 0) {
 813                 nla_nest_end(skb, nest);
 814                 return err;
 815         }
 816
 817 nla_put_failure:
 818         nlmsg_trim(skb, b);
 819         return -1;
 820 }
 821 EXPORT_SYMBOL(tcf_action_dump_1);
 822
 823 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
 824                     int bind, int ref, bool terse)
 825 {
 826         struct tc_action *a;
 827         int err = -EINVAL, i;
 828         struct nlattr *nest;
 829
 830         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
 831                 a = actions[i];
 832                 nest = nla_nest_start_noflag(skb, i + 1);
 833                 if (nest == NULL)
 834                         goto nla_put_failure;
 835                 err = terse ? tcf_action_dump_terse(skb, a) :
 836                         tcf_action_dump_1(skb, a, bind, ref);
 837                 if (err < 0)
 838                         goto errout;
 839                 nla_nest_end(skb, nest);
 840         }
 841
 842         return 0;
 843
 844 nla_put_failure:
 845         err = -EINVAL;
 846 errout:
 847         nla_nest_cancel(skb, nest);
 848         return err;
 849 }
 850
 851 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
 852 {
 853         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
 854         if (!c)
 855                 return NULL;
 856
 857         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
 858         if (!c->data) {
 859                 kfree(c);
 860                 return NULL;
 861         }
 862         c->len = nla_len(tb[TCA_ACT_COOKIE]);
 863
 864         return c;
 865 }
 866
 867 static u8 tcf_action_hw_stats_get(struct nlattr *hw_stats_attr)
 868 {
 869         struct nla_bitfield32 hw_stats_bf;
 870
 871         /* If the user did not pass the attr, that means he does
 872          * not care about the type. Return "any" in that case
 873          * which is setting on all supported types.
 874          */
 875         if (!hw_stats_attr)
 876                 return TCA_ACT_HW_STATS_ANY;
 877         hw_stats_bf = nla_get_bitfield32(hw_stats_attr);
 878         return hw_stats_bf.value;
 879 }
 880
 881 static const struct nla_policy tcf_action_policy[TCA_ACT_MAX + 1] = {
 882         [TCA_ACT_KIND]          = { .type = NLA_STRING },
 883         [TCA_ACT_INDEX]         = { .type = NLA_U32 },
 884         [TCA_ACT_COOKIE]        = { .type = NLA_BINARY,
 885                                     .len = TC_COOKIE_MAX_SIZE },
 886         [TCA_ACT_OPTIONS]       = { .type = NLA_NESTED },
 887         [TCA_ACT_FLAGS]         = NLA_POLICY_BITFIELD32(TCA_ACT_FLAGS_NO_PERCPU_STATS),
 888         [TCA_ACT_HW_STATS]      = NLA_POLICY_BITFIELD32(TCA_ACT_HW_STATS_ANY),
 889 };
 890
 891 static void tcf_idr_insert_many(struct tc_action *actions[])
 892 {
 893         int i;
 894
 895         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
 896                 struct tc_action *a = actions[i];
 897                 struct tcf_idrinfo *idrinfo;
 898
 899                 if (!a)
 900                         continue;
 901                 idrinfo = a->idrinfo;
 902                 mutex_lock(&idrinfo->lock);
 903                 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc if
 904                  * it is just created, otherwise this is just a nop.
 905                  */
 906                 idr_replace(&idrinfo->action_idr, a, a->tcfa_index);
 907                 mutex_unlock(&idrinfo->lock);
 908         }
 909 }
 910
 911 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
 912                                     struct nlattr *nla, struct nlattr *est,
 913                                     char *name, int ovr, int bind,
 914                                     bool rtnl_held,
 915                                     struct netlink_ext_ack *extack)
 916 {
 917         struct nla_bitfield32 flags = { 0, 0 };
 918         u8 hw_stats = TCA_ACT_HW_STATS_ANY;
 919         struct tc_action *a;
 920         struct tc_action_ops *a_o;
 921         struct tc_cookie *cookie = NULL;
 922         char act_name[IFNAMSIZ];
 923         struct nlattr *tb[TCA_ACT_MAX + 1];
 924         struct nlattr *kind;
 925         int err;
 926
 927         if (name == NULL) {
 928                 err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
 929                                                   tcf_action_policy, extack);
 930                 if (err < 0)
 931                         goto err_out;
 932                 err = -EINVAL;
 933                 kind = tb[TCA_ACT_KIND];
 934                 if (!kind) {
 935                         NL_SET_ERR_MSG(extack, "TC action kind must be specified");
 936                         goto err_out;
 937                 }
 938                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
 939                         NL_SET_ERR_MSG(extack, "TC action name too long");
 940                         goto err_out;
 941                 }
 942                 if (tb[TCA_ACT_COOKIE]) {
 943                         cookie = nla_memdup_cookie(tb);
 944                         if (!cookie) {
 945                                 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
 946                                 err = -ENOMEM;
 947                                 goto err_out;
 948                         }
 949                 }
 950                 hw_stats = tcf_action_hw_stats_get(tb[TCA_ACT_HW_STATS]);
 951                 if (tb[TCA_ACT_FLAGS])
 952                         flags = nla_get_bitfield32(tb[TCA_ACT_FLAGS]);
 953         } else {
 954                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
 955                         NL_SET_ERR_MSG(extack, "TC action name too long");
 956                         err = -EINVAL;
 957                         goto err_out;
 958                 }
 959         }
 960
 961         a_o = tc_lookup_action_n(act_name);
 962         if (a_o == NULL) {
 963 #ifdef CONFIG_MODULES
 964                 if (rtnl_held)
 965                         rtnl_unlock();
 966                 request_module("act_%s", act_name);
 967                 if (rtnl_held)
 968                         rtnl_lock();
 969
 970                 a_o = tc_lookup_action_n(act_name);
 971
 972                 /* We dropped the RTNL semaphore in order to
 973                  * perform the module load.  So, even if we
 974                  * succeeded in loading the module we have to
 975                  * tell the caller to replay the request.  We
 976                  * indicate this using -EAGAIN.
 977                  */
 978                 if (a_o != NULL) {
 979                         err = -EAGAIN;
 980                         goto err_mod;
 981                 }
 982 #endif
 983                 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
 984                 err = -ENOENT;
 985                 goto err_free;
 986         }
 987
 988         /* backward compatibility for policer */
 989         if (name == NULL)
 990                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
 991                                 rtnl_held, tp, flags.value, extack);
 992         else
 993                 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
 994                                 tp, flags.value, extack);
 995         if (err < 0)
 996                 goto err_mod;
 997
 998         if (!name && tb[TCA_ACT_COOKIE])
 999                 tcf_set_action_cookie(&a->act_cookie, cookie);
1000
1001         if (!name)
1002                 a->hw_stats = hw_stats;
1003
1004         /* module count goes up only when brand new policy is created
1005          * if it exists and is only bound to in a_o->init() then
1006          * ACT_P_CREATED is not returned (a zero is).
1007          */
1008         if (err != ACT_P_CREATED)
1009                 module_put(a_o->owner);
1010
1011         return a;
1012
1013 err_mod:
1014         module_put(a_o->owner);
1015 err_free:
1016         if (cookie) {
1017                 kfree(cookie->data);
1018                 kfree(cookie);
1019         }
1020 err_out:
1021         return ERR_PTR(err);
1022 }
1023
1024 /* Returns numbers of initialized actions or negative error. */
1025
1026 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
1027                     struct nlattr *est, char *name, int ovr, int bind,
1028                     struct tc_action *actions[], size_t *attr_size,
1029                     bool rtnl_held, struct netlink_ext_ack *extack)
1030 {
1031         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1032         struct tc_action *act;
1033         size_t sz = 0;
1034         int err;
1035         int i;
1036
1037         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1038                                           extack);
1039         if (err < 0)
1040                 return err;
1041
1042         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1043                 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
1044                                         rtnl_held, extack);
1045                 if (IS_ERR(act)) {
1046                         err = PTR_ERR(act);
1047                         goto err;
1048                 }
1049                 sz += tcf_action_fill_size(act);
1050                 /* Start from index 0 */
1051                 actions[i - 1] = act;
1052         }
1053
1054         /* We have to commit them all together, because if any error happened in
1055          * between, we could not handle the failure gracefully.
1056          */
1057         tcf_idr_insert_many(actions);
1058
1059         *attr_size = tcf_action_full_attrs_size(sz);
1060         return i - 1;
1061
1062 err:
1063         tcf_action_destroy(actions, bind);
1064         return err;
1065 }
1066
1067 void tcf_action_update_stats(struct tc_action *a, u64 bytes, u64 packets,
1068                              u64 drops, bool hw)
1069 {
1070         if (a->cpu_bstats) {
1071                 _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), bytes, packets);
1072
1073                 this_cpu_ptr(a->cpu_qstats)->drops += drops;
1074
1075                 if (hw)
1076                         _bstats_cpu_update(this_cpu_ptr(a->cpu_bstats_hw),
1077                                            bytes, packets);
1078                 return;
1079         }
1080
1081         _bstats_update(&a->tcfa_bstats, bytes, packets);
1082         a->tcfa_qstats.drops += drops;
1083         if (hw)
1084                 _bstats_update(&a->tcfa_bstats_hw, bytes, packets);
1085 }
1086 EXPORT_SYMBOL(tcf_action_update_stats);
1087
1088 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
1089                           int compat_mode)
1090 {
1091         int err = 0;
1092         struct gnet_dump d;
1093
1094         if (p == NULL)
1095                 goto errout;
1096
1097         /* compat_mode being true specifies a call that is supposed
1098          * to add additional backward compatibility statistic TLVs.
1099          */
1100         if (compat_mode) {
1101                 if (p->type == TCA_OLD_COMPAT)
1102                         err = gnet_stats_start_copy_compat(skb, 0,
1103                                                            TCA_STATS,
1104                                                            TCA_XSTATS,
1105                                                            &p->tcfa_lock, &d,
1106                                                            TCA_PAD);
1107                 else
1108                         return 0;
1109         } else
1110                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
1111                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
1112
1113         if (err < 0)
1114                 goto errout;
1115
1116         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
1117             gnet_stats_copy_basic_hw(NULL, &d, p->cpu_bstats_hw,
1118                                      &p->tcfa_bstats_hw) < 0 ||
1119             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
1120             gnet_stats_copy_queue(&d, p->cpu_qstats,
1121                                   &p->tcfa_qstats,
1122                                   p->tcfa_qstats.qlen) < 0)
1123                 goto errout;
1124
1125         if (gnet_stats_finish_copy(&d) < 0)
1126                 goto errout;
1127
1128         return 0;
1129
1130 errout:
1131         return -1;
1132 }
1133
1134 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1135                         u32 portid, u32 seq, u16 flags, int event, int bind,
1136                         int ref)
1137 {
1138         struct tcamsg *t;
1139         struct nlmsghdr *nlh;
1140         unsigned char *b = skb_tail_pointer(skb);
1141         struct nlattr *nest;
1142
1143         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1144         if (!nlh)
1145                 goto out_nlmsg_trim;
1146         t = nlmsg_data(nlh);
1147         t->tca_family = AF_UNSPEC;
1148         t->tca__pad1 = 0;
1149         t->tca__pad2 = 0;
1150
1151         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1152         if (!nest)
1153                 goto out_nlmsg_trim;
1154
1155         if (tcf_action_dump(skb, actions, bind, ref, false) < 0)
1156                 goto out_nlmsg_trim;
1157
1158         nla_nest_end(skb, nest);
1159
1160         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1161         return skb->len;
1162
1163 out_nlmsg_trim:
1164         nlmsg_trim(skb, b);
1165         return -1;
1166 }
1167
1168 static int
1169 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1170                struct tc_action *actions[], int event,
1171                struct netlink_ext_ack *extack)
1172 {
1173         struct sk_buff *skb;
1174
1175         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1176         if (!skb)
1177                 return -ENOBUFS;
1178         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1179                          0, 1) <= 0) {
1180                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1181                 kfree_skb(skb);
1182                 return -EINVAL;
1183         }
1184
1185         return rtnl_unicast(skb, net, portid);
1186 }
1187
1188 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1189                                           struct nlmsghdr *n, u32 portid,
1190                                           struct netlink_ext_ack *extack)
1191 {
1192         struct nlattr *tb[TCA_ACT_MAX + 1];
1193         const struct tc_action_ops *ops;
1194         struct tc_action *a;
1195         int index;
1196         int err;
1197
1198         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1199                                           tcf_action_policy, extack);
1200         if (err < 0)
1201                 goto err_out;
1202
1203         err = -EINVAL;
1204         if (tb[TCA_ACT_INDEX] == NULL ||
1205             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1206                 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1207                 goto err_out;
1208         }
1209         index = nla_get_u32(tb[TCA_ACT_INDEX]);
1210
1211         err = -EINVAL;
1212         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1213         if (!ops) { /* could happen in batch of actions */
1214                 NL_SET_ERR_MSG(extack, "Specified TC action kind not found");
1215                 goto err_out;
1216         }
1217         err = -ENOENT;
1218         if (ops->lookup(net, &a, index) == 0) {
1219                 NL_SET_ERR_MSG(extack, "TC action with specified index not found");
1220                 goto err_mod;
1221         }
1222
1223         module_put(ops->owner);
1224         return a;
1225
1226 err_mod:
1227         module_put(ops->owner);
1228 err_out:
1229         return ERR_PTR(err);
1230 }
1231
1232 static int tca_action_flush(struct net *net, struct nlattr *nla,
1233                             struct nlmsghdr *n, u32 portid,
1234                             struct netlink_ext_ack *extack)
1235 {
1236         struct sk_buff *skb;
1237         unsigned char *b;
1238         struct nlmsghdr *nlh;
1239         struct tcamsg *t;
1240         struct netlink_callback dcb;
1241         struct nlattr *nest;
1242         struct nlattr *tb[TCA_ACT_MAX + 1];
1243         const struct tc_action_ops *ops;
1244         struct nlattr *kind;
1245         int err = -ENOMEM;
1246
1247         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1248         if (!skb)
1249                 return err;
1250
1251         b = skb_tail_pointer(skb);
1252
1253         err = nla_parse_nested_deprecated(tb, TCA_ACT_MAX, nla,
1254                                           tcf_action_policy, extack);
1255         if (err < 0)
1256                 goto err_out;
1257
1258         err = -EINVAL;
1259         kind = tb[TCA_ACT_KIND];
1260         ops = tc_lookup_action(kind);
1261         if (!ops) { /*some idjot trying to flush unknown action */
1262                 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1263                 goto err_out;
1264         }
1265
1266         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1267                         sizeof(*t), 0);
1268         if (!nlh) {
1269                 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1270                 goto out_module_put;
1271         }
1272         t = nlmsg_data(nlh);
1273         t->tca_family = AF_UNSPEC;
1274         t->tca__pad1 = 0;
1275         t->tca__pad2 = 0;
1276
1277         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1278         if (!nest) {
1279                 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1280                 goto out_module_put;
1281         }
1282
1283         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1284         if (err <= 0) {
1285                 nla_nest_cancel(skb, nest);
1286                 goto out_module_put;
1287         }
1288
1289         nla_nest_end(skb, nest);
1290
1291         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1292         nlh->nlmsg_flags |= NLM_F_ROOT;
1293         module_put(ops->owner);
1294         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1295                              n->nlmsg_flags & NLM_F_ECHO);
1296         if (err > 0)
1297                 return 0;
1298         if (err < 0)
1299                 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1300
1301         return err;
1302
1303 out_module_put:
1304         module_put(ops->owner);
1305 err_out:
1306         kfree_skb(skb);
1307         return err;
1308 }
1309
1310 static int tcf_action_delete(struct net *net, struct tc_action *actions[])
1311 {
1312         int i;
1313
1314         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1315                 struct tc_action *a = actions[i];
1316                 const struct tc_action_ops *ops = a->ops;
1317                 /* Actions can be deleted concurrently so we must save their
1318                  * type and id to search again after reference is released.
1319                  */
1320                 struct tcf_idrinfo *idrinfo = a->idrinfo;
1321                 u32 act_index = a->tcfa_index;
1322
1323                 actions[i] = NULL;
1324                 if (tcf_action_put(a)) {
1325                         /* last reference, action was deleted concurrently */
1326                         module_put(ops->owner);
1327                 } else  {
1328                         int ret;
1329
1330                         /* now do the delete */
1331                         ret = tcf_idr_delete_index(idrinfo, act_index);
1332                         if (ret < 0)
1333                                 return ret;
1334                 }
1335         }
1336         return 0;
1337 }
1338
1339 static int
1340 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1341                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1342 {
1343         int ret;
1344         struct sk_buff *skb;
1345
1346         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1347                         GFP_KERNEL);
1348         if (!skb)
1349                 return -ENOBUFS;
1350
1351         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1352                          0, 2) <= 0) {
1353                 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1354                 kfree_skb(skb);
1355                 return -EINVAL;
1356         }
1357
1358         /* now do the delete */
1359         ret = tcf_action_delete(net, actions);
1360         if (ret < 0) {
1361                 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1362                 kfree_skb(skb);
1363                 return ret;
1364         }
1365
1366         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1367                              n->nlmsg_flags & NLM_F_ECHO);
1368         if (ret > 0)
1369                 return 0;
1370         return ret;
1371 }
1372
1373 static int
1374 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1375               u32 portid, int event, struct netlink_ext_ack *extack)
1376 {
1377         int i, ret;
1378         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1379         struct tc_action *act;
1380         size_t attr_size = 0;
1381         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1382
1383         ret = nla_parse_nested_deprecated(tb, TCA_ACT_MAX_PRIO, nla, NULL,
1384                                           extack);
1385         if (ret < 0)
1386                 return ret;
1387
1388         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1389                 if (tb[1])
1390                         return tca_action_flush(net, tb[1], n, portid, extack);
1391
1392                 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1393                 return -EINVAL;
1394         }
1395
1396         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1397                 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1398                 if (IS_ERR(act)) {
1399                         ret = PTR_ERR(act);
1400                         goto err;
1401                 }
1402                 attr_size += tcf_action_fill_size(act);
1403                 actions[i - 1] = act;
1404         }
1405
1406         attr_size = tcf_action_full_attrs_size(attr_size);
1407
1408         if (event == RTM_GETACTION)
1409                 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1410         else { /* delete */
1411                 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1412                 if (ret)
1413                         goto err;
1414                 return 0;
1415         }
1416 err:
1417         tcf_action_put_many(actions);
1418         return ret;
1419 }
1420
1421 static int
1422 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1423                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1424 {
1425         struct sk_buff *skb;
1426         int err = 0;
1427
1428         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1429                         GFP_KERNEL);
1430         if (!skb)
1431                 return -ENOBUFS;
1432
1433         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1434                          RTM_NEWACTION, 0, 0) <= 0) {
1435                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1436                 kfree_skb(skb);
1437                 return -EINVAL;
1438         }
1439
1440         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1441                              n->nlmsg_flags & NLM_F_ECHO);
1442         if (err > 0)
1443                 err = 0;
1444         return err;
1445 }
1446
1447 static int tcf_action_add(struct net *net, struct nlattr *nla,
1448                           struct nlmsghdr *n, u32 portid, int ovr,
1449                           struct netlink_ext_ack *extack)
1450 {
1451         size_t attr_size = 0;
1452         int loop, ret;
1453         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1454
1455         for (loop = 0; loop < 10; loop++) {
1456                 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0,
1457                                       actions, &attr_size, true, extack);
1458                 if (ret != -EAGAIN)
1459                         break;
1460         }
1461
1462         if (ret < 0)
1463                 return ret;
1464         ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1465         if (ovr)
1466                 tcf_action_put_many(actions);
1467
1468         return ret;
1469 }
1470
1471 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1472         [TCA_ROOT_FLAGS] = NLA_POLICY_BITFIELD32(TCA_FLAG_LARGE_DUMP_ON),
1473         [TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1474 };
1475
1476 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1477                          struct netlink_ext_ack *extack)
1478 {
1479         struct net *net = sock_net(skb->sk);
1480         struct nlattr *tca[TCA_ROOT_MAX + 1];
1481         u32 portid = NETLINK_CB(skb).portid;
1482         int ret = 0, ovr = 0;
1483
1484         if ((n->nlmsg_type != RTM_GETACTION) &&
1485             !netlink_capable(skb, CAP_NET_ADMIN))
1486                 return -EPERM;
1487
1488         ret = nlmsg_parse_deprecated(n, sizeof(struct tcamsg), tca,
1489                                      TCA_ROOT_MAX, NULL, extack);
1490         if (ret < 0)
1491                 return ret;
1492
1493         if (tca[TCA_ACT_TAB] == NULL) {
1494                 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1495                 return -EINVAL;
1496         }
1497
1498         /* n->nlmsg_flags & NLM_F_CREATE */
1499         switch (n->nlmsg_type) {
1500         case RTM_NEWACTION:
1501                 /* we are going to assume all other flags
1502                  * imply create only if it doesn't exist
1503                  * Note that CREATE | EXCL implies that
1504                  * but since we want avoid ambiguity (eg when flags
1505                  * is zero) then just set this
1506                  */
1507                 if (n->nlmsg_flags & NLM_F_REPLACE)
1508                         ovr = 1;
1509                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1510                                      extack);
1511                 break;
1512         case RTM_DELACTION:
1513                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1514                                     portid, RTM_DELACTION, extack);
1515                 break;
1516         case RTM_GETACTION:
1517                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1518                                     portid, RTM_GETACTION, extack);
1519                 break;
1520         default:
1521                 BUG();
1522         }
1523
1524         return ret;
1525 }
1526
1527 static struct nlattr *find_dump_kind(struct nlattr **nla)
1528 {
1529         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1530         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1531         struct nlattr *kind;
1532
1533         tb1 = nla[TCA_ACT_TAB];
1534         if (tb1 == NULL)
1535                 return NULL;
1536
1537         if (nla_parse_deprecated(tb, TCA_ACT_MAX_PRIO, nla_data(tb1), NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1538                 return NULL;
1539
1540         if (tb[1] == NULL)
1541                 return NULL;
1542         if (nla_parse_nested_deprecated(tb2, TCA_ACT_MAX, tb[1], tcf_action_policy, NULL) < 0)
1543                 return NULL;
1544         kind = tb2[TCA_ACT_KIND];
1545
1546         return kind;
1547 }
1548
1549 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1550 {
1551         struct net *net = sock_net(skb->sk);
1552         struct nlmsghdr *nlh;
1553         unsigned char *b = skb_tail_pointer(skb);
1554         struct nlattr *nest;
1555         struct tc_action_ops *a_o;
1556         int ret = 0;
1557         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1558         struct nlattr *tb[TCA_ROOT_MAX + 1];
1559         struct nlattr *count_attr = NULL;
1560         unsigned long jiffy_since = 0;
1561         struct nlattr *kind = NULL;
1562         struct nla_bitfield32 bf;
1563         u32 msecs_since = 0;
1564         u32 act_count = 0;
1565
1566         ret = nlmsg_parse_deprecated(cb->nlh, sizeof(struct tcamsg), tb,
1567                                      TCA_ROOT_MAX, tcaa_policy, cb->extack);
1568         if (ret < 0)
1569                 return ret;
1570
1571         kind = find_dump_kind(tb);
1572         if (kind == NULL) {
1573                 pr_info("tc_dump_action: action bad kind\n");
1574                 return 0;
1575         }
1576
1577         a_o = tc_lookup_action(kind);
1578         if (a_o == NULL)
1579                 return 0;
1580
1581         cb->args[2] = 0;
1582         if (tb[TCA_ROOT_FLAGS]) {
1583                 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1584                 cb->args[2] = bf.value;
1585         }
1586
1587         if (tb[TCA_ROOT_TIME_DELTA]) {
1588                 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1589         }
1590
1591         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1592                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1593         if (!nlh)
1594                 goto out_module_put;
1595
1596         if (msecs_since)
1597                 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1598
1599         t = nlmsg_data(nlh);
1600         t->tca_family = AF_UNSPEC;
1601         t->tca__pad1 = 0;
1602         t->tca__pad2 = 0;
1603         cb->args[3] = jiffy_since;
1604         count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1605         if (!count_attr)
1606                 goto out_module_put;
1607
1608         nest = nla_nest_start_noflag(skb, TCA_ACT_TAB);
1609         if (nest == NULL)
1610                 goto out_module_put;
1611
1612         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1613         if (ret < 0)
1614                 goto out_module_put;
1615
1616         if (ret > 0) {
1617                 nla_nest_end(skb, nest);
1618                 ret = skb->len;
1619                 act_count = cb->args[1];
1620                 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1621                 cb->args[1] = 0;
1622         } else
1623                 nlmsg_trim(skb, b);
1624
1625         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1626         if (NETLINK_CB(cb->skb).portid && ret)
1627                 nlh->nlmsg_flags |= NLM_F_MULTI;
1628         module_put(a_o->owner);
1629         return skb->len;
1630
1631 out_module_put:
1632         module_put(a_o->owner);
1633         nlmsg_trim(skb, b);
1634         return skb->len;
1635 }
1636
1637 static int __init tc_action_init(void)
1638 {
1639         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1640         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1641         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1642                       0);
1643
1644         return 0;
1645 }
1646
1647 subsys_initcall(tc_action_init);