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 <net/net_namespace.h>
  25 #include <net/sock.h>
  26 #include <net/sch_generic.h>
  27 #include <net/pkt_cls.h>
  28 #include <net/act_api.h>
  29 #include <net/netlink.h>
  30
  31 static int tcf_action_goto_chain_init(struct tc_action *a, struct tcf_proto *tp)
  32 {
  33         u32 chain_index = a->tcfa_action & TC_ACT_EXT_VAL_MASK;
  34
  35         if (!tp)
  36                 return -EINVAL;
  37         a->goto_chain = tcf_chain_get(tp->chain->block, chain_index, true);
  38         if (!a->goto_chain)
  39                 return -ENOMEM;
  40         return 0;
  41 }
  42
  43 static void tcf_action_goto_chain_fini(struct tc_action *a)
  44 {
  45         tcf_chain_put(a->goto_chain);
  46 }
  47
  48 static void tcf_action_goto_chain_exec(const struct tc_action *a,
  49                                        struct tcf_result *res)
  50 {
  51         const struct tcf_chain *chain = a->goto_chain;
  52
  53         res->goto_tp = rcu_dereference_bh(chain->filter_chain);
  54 }
  55
  56 /* XXX: For standalone actions, we don't need a RCU grace period either, because
  57  * actions are always connected to filters and filters are already destroyed in
  58  * RCU callbacks, so after a RCU grace period actions are already disconnected
  59  * from filters. Readers later can not find us.
  60  */
  61 static void free_tcf(struct tc_action *p)
  62 {
  63         free_percpu(p->cpu_bstats);
  64         free_percpu(p->cpu_qstats);
  65
  66         if (p->act_cookie) {
  67                 kfree(p->act_cookie->data);
  68                 kfree(p->act_cookie);
  69         }
  70         if (p->goto_chain)
  71                 tcf_action_goto_chain_fini(p);
  72
  73         kfree(p);
  74 }
  75
  76 static void tcf_idr_remove(struct tcf_idrinfo *idrinfo, struct tc_action *p)
  77 {
  78         spin_lock_bh(&idrinfo->lock);
  79         idr_remove_ext(&idrinfo->action_idr, p->tcfa_index);
  80         spin_unlock_bh(&idrinfo->lock);
  81         gen_kill_estimator(&p->tcfa_rate_est);
  82         free_tcf(p);
  83 }
  84
  85 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
  86 {
  87         int ret = 0;
  88
  89         ASSERT_RTNL();
  90
  91         if (p) {
  92                 if (bind)
  93                         p->tcfa_bindcnt--;
  94                 else if (strict && p->tcfa_bindcnt > 0)
  95                         return -EPERM;
  96
  97                 p->tcfa_refcnt--;
  98                 if (p->tcfa_bindcnt <= 0 && p->tcfa_refcnt <= 0) {
  99                         if (p->ops->cleanup)
 100                                 p->ops->cleanup(p, bind);
 101                         tcf_idr_remove(p->idrinfo, p);
 102                         ret = ACT_P_DELETED;
 103                 }
 104         }
 105
 106         return ret;
 107 }
 108 EXPORT_SYMBOL(__tcf_idr_release);
 109
 110 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 111                            struct netlink_callback *cb)
 112 {
 113         int err = 0, index = -1, s_i = 0, n_i = 0;
 114         u32 act_flags = cb->args[2];
 115         unsigned long jiffy_since = cb->args[3];
 116         struct nlattr *nest;
 117         struct idr *idr = &idrinfo->action_idr;
 118         struct tc_action *p;
 119         unsigned long id = 1;
 120
 121         spin_lock_bh(&idrinfo->lock);
 122
 123         s_i = cb->args[0];
 124
 125         idr_for_each_entry_ext(idr, p, id) {
 126                 index++;
 127                 if (index < s_i)
 128                         continue;
 129
 130                 if (jiffy_since &&
 131                     time_after(jiffy_since,
 132                                (unsigned long)p->tcfa_tm.lastuse))
 133                         continue;
 134
 135                 nest = nla_nest_start(skb, n_i);
 136                 if (!nest) {
 137                         index--;
 138                         goto nla_put_failure;
 139                 }
 140                 err = tcf_action_dump_1(skb, p, 0, 0);
 141                 if (err < 0) {
 142                         index--;
 143                         nlmsg_trim(skb, nest);
 144                         goto done;
 145                 }
 146                 nla_nest_end(skb, nest);
 147                 n_i++;
 148                 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
 149                     n_i >= TCA_ACT_MAX_PRIO)
 150                         goto done;
 151         }
 152 done:
 153         if (index >= 0)
 154                 cb->args[0] = index + 1;
 155
 156         spin_unlock_bh(&idrinfo->lock);
 157         if (n_i) {
 158                 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
 159                         cb->args[1] = n_i;
 160         }
 161         return n_i;
 162
 163 nla_put_failure:
 164         nla_nest_cancel(skb, nest);
 165         goto done;
 166 }
 167
 168 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
 169                           const struct tc_action_ops *ops)
 170 {
 171         struct nlattr *nest;
 172         int n_i = 0;
 173         int ret = -EINVAL;
 174         struct idr *idr = &idrinfo->action_idr;
 175         struct tc_action *p;
 176         unsigned long id = 1;
 177
 178         nest = nla_nest_start(skb, 0);
 179         if (nest == NULL)
 180                 goto nla_put_failure;
 181         if (nla_put_string(skb, TCA_KIND, ops->kind))
 182                 goto nla_put_failure;
 183
 184         idr_for_each_entry_ext(idr, p, id) {
 185                 ret = __tcf_idr_release(p, false, true);
 186                 if (ret == ACT_P_DELETED) {
 187                         module_put(ops->owner);
 188                         n_i++;
 189                 } else if (ret < 0) {
 190                         goto nla_put_failure;
 191                 }
 192         }
 193         if (nla_put_u32(skb, TCA_FCNT, n_i))
 194                 goto nla_put_failure;
 195         nla_nest_end(skb, nest);
 196
 197         return n_i;
 198 nla_put_failure:
 199         nla_nest_cancel(skb, nest);
 200         return ret;
 201 }
 202
 203 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
 204                        struct netlink_callback *cb, int type,
 205                        const struct tc_action_ops *ops)
 206 {
 207         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 208
 209         if (type == RTM_DELACTION) {
 210                 return tcf_del_walker(idrinfo, skb, ops);
 211         } else if (type == RTM_GETACTION) {
 212                 return tcf_dump_walker(idrinfo, skb, cb);
 213         } else {
 214                 WARN(1, "tcf_generic_walker: unknown action %d\n", type);
 215                 return -EINVAL;
 216         }
 217 }
 218 EXPORT_SYMBOL(tcf_generic_walker);
 219
 220 static struct tc_action *tcf_idr_lookup(u32 index, struct tcf_idrinfo *idrinfo)
 221 {
 222         struct tc_action *p = NULL;
 223
 224         spin_lock_bh(&idrinfo->lock);
 225         p = idr_find_ext(&idrinfo->action_idr, index);
 226         spin_unlock_bh(&idrinfo->lock);
 227
 228         return p;
 229 }
 230
 231 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
 232 {
 233         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 234         struct tc_action *p = tcf_idr_lookup(index, idrinfo);
 235
 236         if (p) {
 237                 *a = p;
 238                 return 1;
 239         }
 240         return 0;
 241 }
 242 EXPORT_SYMBOL(tcf_idr_search);
 243
 244 bool tcf_idr_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
 245                    int bind)
 246 {
 247         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 248         struct tc_action *p = tcf_idr_lookup(index, idrinfo);
 249
 250         if (index && p) {
 251                 if (bind)
 252                         p->tcfa_bindcnt++;
 253                 p->tcfa_refcnt++;
 254                 *a = p;
 255                 return true;
 256         }
 257         return false;
 258 }
 259 EXPORT_SYMBOL(tcf_idr_check);
 260
 261 void tcf_idr_cleanup(struct tc_action *a, struct nlattr *est)
 262 {
 263         if (est)
 264                 gen_kill_estimator(&a->tcfa_rate_est);
 265         free_tcf(a);
 266 }
 267 EXPORT_SYMBOL(tcf_idr_cleanup);
 268
 269 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
 270                    struct tc_action **a, const struct tc_action_ops *ops,
 271                    int bind, bool cpustats)
 272 {
 273         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
 274         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 275         struct idr *idr = &idrinfo->action_idr;
 276         int err = -ENOMEM;
 277         unsigned long idr_index;
 278
 279         if (unlikely(!p))
 280                 return -ENOMEM;
 281         p->tcfa_refcnt = 1;
 282         if (bind)
 283                 p->tcfa_bindcnt = 1;
 284
 285         if (cpustats) {
 286                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 287                 if (!p->cpu_bstats) {
 288 err1:
 289                         kfree(p);
 290                         return err;
 291                 }
 292                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 293                 if (!p->cpu_qstats) {
 294 err2:
 295                         free_percpu(p->cpu_bstats);
 296                         goto err1;
 297                 }
 298         }
 299         spin_lock_init(&p->tcfa_lock);
 300         /* user doesn't specify an index */
 301         if (!index) {
 302                 idr_preload(GFP_KERNEL);
 303                 spin_lock_bh(&idrinfo->lock);
 304                 err = idr_alloc_ext(idr, NULL, &idr_index, 1, 0,
 305                                     GFP_ATOMIC);
 306                 spin_unlock_bh(&idrinfo->lock);
 307                 idr_preload_end();
 308                 if (err) {
 309 err3:
 310                         free_percpu(p->cpu_qstats);
 311                         goto err2;
 312                 }
 313                 p->tcfa_index = idr_index;
 314         } else {
 315                 idr_preload(GFP_KERNEL);
 316                 spin_lock_bh(&idrinfo->lock);
 317                 err = idr_alloc_ext(idr, NULL, NULL, index, index + 1,
 318                                     GFP_ATOMIC);
 319                 spin_unlock_bh(&idrinfo->lock);
 320                 idr_preload_end();
 321                 if (err)
 322                         goto err3;
 323                 p->tcfa_index = index;
 324         }
 325
 326         p->tcfa_tm.install = jiffies;
 327         p->tcfa_tm.lastuse = jiffies;
 328         p->tcfa_tm.firstuse = 0;
 329         if (est) {
 330                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
 331                                         &p->tcfa_rate_est,
 332                                         &p->tcfa_lock, NULL, est);
 333                 if (err) {
 334                         goto err3;
 335                 }
 336         }
 337
 338         p->idrinfo = idrinfo;
 339         p->ops = ops;
 340         INIT_LIST_HEAD(&p->list);
 341         *a = p;
 342         return 0;
 343 }
 344 EXPORT_SYMBOL(tcf_idr_create);
 345
 346 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
 347 {
 348         struct tcf_idrinfo *idrinfo = tn->idrinfo;
 349
 350         spin_lock_bh(&idrinfo->lock);
 351         idr_replace_ext(&idrinfo->action_idr, a, a->tcfa_index);
 352         spin_unlock_bh(&idrinfo->lock);
 353 }
 354 EXPORT_SYMBOL(tcf_idr_insert);
 355
 356 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
 357                          struct tcf_idrinfo *idrinfo)
 358 {
 359         struct idr *idr = &idrinfo->action_idr;
 360         struct tc_action *p;
 361         int ret;
 362         unsigned long id = 1;
 363
 364         idr_for_each_entry_ext(idr, p, id) {
 365                 ret = __tcf_idr_release(p, false, true);
 366                 if (ret == ACT_P_DELETED)
 367                         module_put(ops->owner);
 368                 else if (ret < 0)
 369                         return;
 370         }
 371         idr_destroy(&idrinfo->action_idr);
 372 }
 373 EXPORT_SYMBOL(tcf_idrinfo_destroy);
 374
 375 static LIST_HEAD(act_base);
 376 static DEFINE_RWLOCK(act_mod_lock);
 377
 378 int tcf_register_action(struct tc_action_ops *act,
 379                         struct pernet_operations *ops)
 380 {
 381         struct tc_action_ops *a;
 382         int ret;
 383
 384         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
 385                 return -EINVAL;
 386
 387         /* We have to register pernet ops before making the action ops visible,
 388          * otherwise tcf_action_init_1() could get a partially initialized
 389          * netns.
 390          */
 391         ret = register_pernet_subsys(ops);
 392         if (ret)
 393                 return ret;
 394
 395         write_lock(&act_mod_lock);
 396         list_for_each_entry(a, &act_base, head) {
 397                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
 398                         write_unlock(&act_mod_lock);
 399                         unregister_pernet_subsys(ops);
 400                         return -EEXIST;
 401                 }
 402         }
 403         list_add_tail(&act->head, &act_base);
 404         write_unlock(&act_mod_lock);
 405
 406         return 0;
 407 }
 408 EXPORT_SYMBOL(tcf_register_action);
 409
 410 int tcf_unregister_action(struct tc_action_ops *act,
 411                           struct pernet_operations *ops)
 412 {
 413         struct tc_action_ops *a;
 414         int err = -ENOENT;
 415
 416         write_lock(&act_mod_lock);
 417         list_for_each_entry(a, &act_base, head) {
 418                 if (a == act) {
 419                         list_del(&act->head);
 420                         err = 0;
 421                         break;
 422                 }
 423         }
 424         write_unlock(&act_mod_lock);
 425         if (!err)
 426                 unregister_pernet_subsys(ops);
 427         return err;
 428 }
 429 EXPORT_SYMBOL(tcf_unregister_action);
 430
 431 /* lookup by name */
 432 static struct tc_action_ops *tc_lookup_action_n(char *kind)
 433 {
 434         struct tc_action_ops *a, *res = NULL;
 435
 436         if (kind) {
 437                 read_lock(&act_mod_lock);
 438                 list_for_each_entry(a, &act_base, head) {
 439                         if (strcmp(kind, a->kind) == 0) {
 440                                 if (try_module_get(a->owner))
 441                                         res = a;
 442                                 break;
 443                         }
 444                 }
 445                 read_unlock(&act_mod_lock);
 446         }
 447         return res;
 448 }
 449
 450 /* lookup by nlattr */
 451 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
 452 {
 453         struct tc_action_ops *a, *res = NULL;
 454
 455         if (kind) {
 456                 read_lock(&act_mod_lock);
 457                 list_for_each_entry(a, &act_base, head) {
 458                         if (nla_strcmp(kind, a->kind) == 0) {
 459                                 if (try_module_get(a->owner))
 460                                         res = a;
 461                                 break;
 462                         }
 463                 }
 464                 read_unlock(&act_mod_lock);
 465         }
 466         return res;
 467 }
 468
 469 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
 470 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
 471 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 472                     int nr_actions, struct tcf_result *res)
 473 {
 474         u32 jmp_prgcnt = 0;
 475         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
 476         int i;
 477         int ret = TC_ACT_OK;
 478
 479         if (skb_skip_tc_classify(skb))
 480                 return TC_ACT_OK;
 481
 482 restart_act_graph:
 483         for (i = 0; i < nr_actions; i++) {
 484                 const struct tc_action *a = actions[i];
 485
 486                 if (jmp_prgcnt > 0) {
 487                         jmp_prgcnt -= 1;
 488                         continue;
 489                 }
 490 repeat:
 491                 ret = a->ops->act(skb, a, res);
 492                 if (ret == TC_ACT_REPEAT)
 493                         goto repeat;    /* we need a ttl - JHS */
 494
 495                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
 496                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
 497                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
 498                                 /* faulty opcode, stop pipeline */
 499                                 return TC_ACT_OK;
 500                         } else {
 501                                 jmp_ttl -= 1;
 502                                 if (jmp_ttl > 0)
 503                                         goto restart_act_graph;
 504                                 else /* faulty graph, stop pipeline */
 505                                         return TC_ACT_OK;
 506                         }
 507                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
 508                         tcf_action_goto_chain_exec(a, res);
 509                 }
 510
 511                 if (ret != TC_ACT_PIPE)
 512                         break;
 513         }
 514
 515         return ret;
 516 }
 517 EXPORT_SYMBOL(tcf_action_exec);
 518
 519 int tcf_action_destroy(struct list_head *actions, int bind)
 520 {
 521         const struct tc_action_ops *ops;
 522         struct tc_action *a, *tmp;
 523         int ret = 0;
 524
 525         list_for_each_entry_safe(a, tmp, actions, list) {
 526                 ops = a->ops;
 527                 ret = __tcf_idr_release(a, bind, true);
 528                 if (ret == ACT_P_DELETED)
 529                         module_put(ops->owner);
 530                 else if (ret < 0)
 531                         return ret;
 532         }
 533         return ret;
 534 }
 535
 536 int
 537 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 538 {
 539         return a->ops->dump(skb, a, bind, ref);
 540 }
 541
 542 int
 543 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 544 {
 545         int err = -EINVAL;
 546         unsigned char *b = skb_tail_pointer(skb);
 547         struct nlattr *nest;
 548
 549         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
 550                 goto nla_put_failure;
 551         if (tcf_action_copy_stats(skb, a, 0))
 552                 goto nla_put_failure;
 553         if (a->act_cookie) {
 554                 if (nla_put(skb, TCA_ACT_COOKIE, a->act_cookie->len,
 555                             a->act_cookie->data))
 556                         goto nla_put_failure;
 557         }
 558
 559         nest = nla_nest_start(skb, TCA_OPTIONS);
 560         if (nest == NULL)
 561                 goto nla_put_failure;
 562         err = tcf_action_dump_old(skb, a, bind, ref);
 563         if (err > 0) {
 564                 nla_nest_end(skb, nest);
 565                 return err;
 566         }
 567
 568 nla_put_failure:
 569         nlmsg_trim(skb, b);
 570         return -1;
 571 }
 572 EXPORT_SYMBOL(tcf_action_dump_1);
 573
 574 int tcf_action_dump(struct sk_buff *skb, struct list_head *actions,
 575                     int bind, int ref)
 576 {
 577         struct tc_action *a;
 578         int err = -EINVAL;
 579         struct nlattr *nest;
 580
 581         list_for_each_entry(a, actions, list) {
 582                 nest = nla_nest_start(skb, a->order);
 583                 if (nest == NULL)
 584                         goto nla_put_failure;
 585                 err = tcf_action_dump_1(skb, a, bind, ref);
 586                 if (err < 0)
 587                         goto errout;
 588                 nla_nest_end(skb, nest);
 589         }
 590
 591         return 0;
 592
 593 nla_put_failure:
 594         err = -EINVAL;
 595 errout:
 596         nla_nest_cancel(skb, nest);
 597         return err;
 598 }
 599
 600 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
 601 {
 602         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
 603         if (!c)
 604                 return NULL;
 605
 606         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
 607         if (!c->data) {
 608                 kfree(c);
 609                 return NULL;
 610         }
 611         c->len = nla_len(tb[TCA_ACT_COOKIE]);
 612
 613         return c;
 614 }
 615
 616 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
 617                                     struct nlattr *nla, struct nlattr *est,
 618                                     char *name, int ovr, int bind)
 619 {
 620         struct tc_action *a;
 621         struct tc_action_ops *a_o;
 622         struct tc_cookie *cookie = NULL;
 623         char act_name[IFNAMSIZ];
 624         struct nlattr *tb[TCA_ACT_MAX + 1];
 625         struct nlattr *kind;
 626         int err;
 627
 628         if (name == NULL) {
 629                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
 630                 if (err < 0)
 631                         goto err_out;
 632                 err = -EINVAL;
 633                 kind = tb[TCA_ACT_KIND];
 634                 if (kind == NULL)
 635                         goto err_out;
 636                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
 637                         goto err_out;
 638                 if (tb[TCA_ACT_COOKIE]) {
 639                         int cklen = nla_len(tb[TCA_ACT_COOKIE]);
 640
 641                         if (cklen > TC_COOKIE_MAX_SIZE)
 642                                 goto err_out;
 643
 644                         cookie = nla_memdup_cookie(tb);
 645                         if (!cookie) {
 646                                 err = -ENOMEM;
 647                                 goto err_out;
 648                         }
 649                 }
 650         } else {
 651                 err = -EINVAL;
 652                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
 653                         goto err_out;
 654         }
 655
 656         a_o = tc_lookup_action_n(act_name);
 657         if (a_o == NULL) {
 658 #ifdef CONFIG_MODULES
 659                 rtnl_unlock();
 660                 request_module("act_%s", act_name);
 661                 rtnl_lock();
 662
 663                 a_o = tc_lookup_action_n(act_name);
 664
 665                 /* We dropped the RTNL semaphore in order to
 666                  * perform the module load.  So, even if we
 667                  * succeeded in loading the module we have to
 668                  * tell the caller to replay the request.  We
 669                  * indicate this using -EAGAIN.
 670                  */
 671                 if (a_o != NULL) {
 672                         err = -EAGAIN;
 673                         goto err_mod;
 674                 }
 675 #endif
 676                 err = -ENOENT;
 677                 goto err_out;
 678         }
 679
 680         /* backward compatibility for policer */
 681         if (name == NULL)
 682                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind);
 683         else
 684                 err = a_o->init(net, nla, est, &a, ovr, bind);
 685         if (err < 0)
 686                 goto err_mod;
 687
 688         if (name == NULL && tb[TCA_ACT_COOKIE]) {
 689                 if (a->act_cookie) {
 690                         kfree(a->act_cookie->data);
 691                         kfree(a->act_cookie);
 692                 }
 693                 a->act_cookie = cookie;
 694         }
 695
 696         /* module count goes up only when brand new policy is created
 697          * if it exists and is only bound to in a_o->init() then
 698          * ACT_P_CREATED is not returned (a zero is).
 699          */
 700         if (err != ACT_P_CREATED)
 701                 module_put(a_o->owner);
 702
 703         if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
 704                 err = tcf_action_goto_chain_init(a, tp);
 705                 if (err) {
 706                         LIST_HEAD(actions);
 707
 708                         list_add_tail(&a->list, &actions);
 709                         tcf_action_destroy(&actions, bind);
 710                         return ERR_PTR(err);
 711                 }
 712         }
 713
 714         return a;
 715
 716 err_mod:
 717         module_put(a_o->owner);
 718 err_out:
 719         if (cookie) {
 720                 kfree(cookie->data);
 721                 kfree(cookie);
 722         }
 723         return ERR_PTR(err);
 724 }
 725
 726 static void cleanup_a(struct list_head *actions, int ovr)
 727 {
 728         struct tc_action *a;
 729
 730         if (!ovr)
 731                 return;
 732
 733         list_for_each_entry(a, actions, list)
 734                 a->tcfa_refcnt--;
 735 }
 736
 737 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
 738                     struct nlattr *est, char *name, int ovr, int bind,
 739                     struct list_head *actions)
 740 {
 741         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
 742         struct tc_action *act;
 743         int err;
 744         int i;
 745
 746         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, NULL);
 747         if (err < 0)
 748                 return err;
 749
 750         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
 751                 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind);
 752                 if (IS_ERR(act)) {
 753                         err = PTR_ERR(act);
 754                         goto err;
 755                 }
 756                 act->order = i;
 757                 if (ovr)
 758                         act->tcfa_refcnt++;
 759                 list_add_tail(&act->list, actions);
 760         }
 761
 762         /* Remove the temp refcnt which was necessary to protect against
 763          * destroying an existing action which was being replaced
 764          */
 765         cleanup_a(actions, ovr);
 766         return 0;
 767
 768 err:
 769         tcf_action_destroy(actions, bind);
 770         return err;
 771 }
 772
 773 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
 774                           int compat_mode)
 775 {
 776         int err = 0;
 777         struct gnet_dump d;
 778
 779         if (p == NULL)
 780                 goto errout;
 781
 782         /* compat_mode being true specifies a call that is supposed
 783          * to add additional backward compatibility statistic TLVs.
 784          */
 785         if (compat_mode) {
 786                 if (p->type == TCA_OLD_COMPAT)
 787                         err = gnet_stats_start_copy_compat(skb, 0,
 788                                                            TCA_STATS,
 789                                                            TCA_XSTATS,
 790                                                            &p->tcfa_lock, &d,
 791                                                            TCA_PAD);
 792                 else
 793                         return 0;
 794         } else
 795                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
 796                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
 797
 798         if (err < 0)
 799                 goto errout;
 800
 801         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
 802             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
 803             gnet_stats_copy_queue(&d, p->cpu_qstats,
 804                                   &p->tcfa_qstats,
 805                                   p->tcfa_qstats.qlen) < 0)
 806                 goto errout;
 807
 808         if (gnet_stats_finish_copy(&d) < 0)
 809                 goto errout;
 810
 811         return 0;
 812
 813 errout:
 814         return -1;
 815 }
 816
 817 static int tca_get_fill(struct sk_buff *skb, struct list_head *actions,
 818                         u32 portid, u32 seq, u16 flags, int event, int bind,
 819                         int ref)
 820 {
 821         struct tcamsg *t;
 822         struct nlmsghdr *nlh;
 823         unsigned char *b = skb_tail_pointer(skb);
 824         struct nlattr *nest;
 825
 826         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
 827         if (!nlh)
 828                 goto out_nlmsg_trim;
 829         t = nlmsg_data(nlh);
 830         t->tca_family = AF_UNSPEC;
 831         t->tca__pad1 = 0;
 832         t->tca__pad2 = 0;
 833
 834         nest = nla_nest_start(skb, TCA_ACT_TAB);
 835         if (nest == NULL)
 836                 goto out_nlmsg_trim;
 837
 838         if (tcf_action_dump(skb, actions, bind, ref) < 0)
 839                 goto out_nlmsg_trim;
 840
 841         nla_nest_end(skb, nest);
 842
 843         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 844         return skb->len;
 845
 846 out_nlmsg_trim:
 847         nlmsg_trim(skb, b);
 848         return -1;
 849 }
 850
 851 static int
 852 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
 853                struct list_head *actions, int event)
 854 {
 855         struct sk_buff *skb;
 856
 857         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 858         if (!skb)
 859                 return -ENOBUFS;
 860         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
 861                          0, 0) <= 0) {
 862                 kfree_skb(skb);
 863                 return -EINVAL;
 864         }
 865
 866         return rtnl_unicast(skb, net, portid);
 867 }
 868
 869 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
 870                                           struct nlmsghdr *n, u32 portid)
 871 {
 872         struct nlattr *tb[TCA_ACT_MAX + 1];
 873         const struct tc_action_ops *ops;
 874         struct tc_action *a;
 875         int index;
 876         int err;
 877
 878         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
 879         if (err < 0)
 880                 goto err_out;
 881
 882         err = -EINVAL;
 883         if (tb[TCA_ACT_INDEX] == NULL ||
 884             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
 885                 goto err_out;
 886         index = nla_get_u32(tb[TCA_ACT_INDEX]);
 887
 888         err = -EINVAL;
 889         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
 890         if (!ops) /* could happen in batch of actions */
 891                 goto err_out;
 892         err = -ENOENT;
 893         if (ops->lookup(net, &a, index) == 0)
 894                 goto err_mod;
 895
 896         module_put(ops->owner);
 897         return a;
 898
 899 err_mod:
 900         module_put(ops->owner);
 901 err_out:
 902         return ERR_PTR(err);
 903 }
 904
 905 static int tca_action_flush(struct net *net, struct nlattr *nla,
 906                             struct nlmsghdr *n, u32 portid)
 907 {
 908         struct sk_buff *skb;
 909         unsigned char *b;
 910         struct nlmsghdr *nlh;
 911         struct tcamsg *t;
 912         struct netlink_callback dcb;
 913         struct nlattr *nest;
 914         struct nlattr *tb[TCA_ACT_MAX + 1];
 915         const struct tc_action_ops *ops;
 916         struct nlattr *kind;
 917         int err = -ENOMEM;
 918
 919         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 920         if (!skb) {
 921                 pr_debug("tca_action_flush: failed skb alloc\n");
 922                 return err;
 923         }
 924
 925         b = skb_tail_pointer(skb);
 926
 927         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, NULL);
 928         if (err < 0)
 929                 goto err_out;
 930
 931         err = -EINVAL;
 932         kind = tb[TCA_ACT_KIND];
 933         ops = tc_lookup_action(kind);
 934         if (!ops) /*some idjot trying to flush unknown action */
 935                 goto err_out;
 936
 937         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
 938                         sizeof(*t), 0);
 939         if (!nlh)
 940                 goto out_module_put;
 941         t = nlmsg_data(nlh);
 942         t->tca_family = AF_UNSPEC;
 943         t->tca__pad1 = 0;
 944         t->tca__pad2 = 0;
 945
 946         nest = nla_nest_start(skb, TCA_ACT_TAB);
 947         if (nest == NULL)
 948                 goto out_module_put;
 949
 950         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops);
 951         if (err <= 0)
 952                 goto out_module_put;
 953
 954         nla_nest_end(skb, nest);
 955
 956         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 957         nlh->nlmsg_flags |= NLM_F_ROOT;
 958         module_put(ops->owner);
 959         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 960                              n->nlmsg_flags & NLM_F_ECHO);
 961         if (err > 0)
 962                 return 0;
 963
 964         return err;
 965
 966 out_module_put:
 967         module_put(ops->owner);
 968 err_out:
 969         kfree_skb(skb);
 970         return err;
 971 }
 972
 973 static int
 974 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
 975                u32 portid)
 976 {
 977         int ret;
 978         struct sk_buff *skb;
 979
 980         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 981         if (!skb)
 982                 return -ENOBUFS;
 983
 984         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
 985                          0, 1) <= 0) {
 986                 kfree_skb(skb);
 987                 return -EINVAL;
 988         }
 989
 990         /* now do the delete */
 991         ret = tcf_action_destroy(actions, 0);
 992         if (ret < 0) {
 993                 kfree_skb(skb);
 994                 return ret;
 995         }
 996
 997         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
 998                              n->nlmsg_flags & NLM_F_ECHO);
 999         if (ret > 0)
1000                 return 0;
1001         return ret;
1002 }
1003
1004 static int
1005 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1006               u32 portid, int event)
1007 {
1008         int i, ret;
1009         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1010         struct tc_action *act;
1011         LIST_HEAD(actions);
1012
1013         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, NULL);
1014         if (ret < 0)
1015                 return ret;
1016
1017         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1018                 if (tb[1] != NULL)
1019                         return tca_action_flush(net, tb[1], n, portid);
1020                 else
1021                         return -EINVAL;
1022         }
1023
1024         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1025                 act = tcf_action_get_1(net, tb[i], n, portid);
1026                 if (IS_ERR(act)) {
1027                         ret = PTR_ERR(act);
1028                         goto err;
1029                 }
1030                 act->order = i;
1031                 list_add_tail(&act->list, &actions);
1032         }
1033
1034         if (event == RTM_GETACTION)
1035                 ret = tcf_get_notify(net, portid, n, &actions, event);
1036         else { /* delete */
1037                 ret = tcf_del_notify(net, n, &actions, portid);
1038                 if (ret)
1039                         goto err;
1040                 return ret;
1041         }
1042 err:
1043         if (event != RTM_GETACTION)
1044                 tcf_action_destroy(&actions, 0);
1045         return ret;
1046 }
1047
1048 static int
1049 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct list_head *actions,
1050                u32 portid)
1051 {
1052         struct sk_buff *skb;
1053         int err = 0;
1054
1055         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1056         if (!skb)
1057                 return -ENOBUFS;
1058
1059         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1060                          RTM_NEWACTION, 0, 0) <= 0) {
1061                 kfree_skb(skb);
1062                 return -EINVAL;
1063         }
1064
1065         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1066                              n->nlmsg_flags & NLM_F_ECHO);
1067         if (err > 0)
1068                 err = 0;
1069         return err;
1070 }
1071
1072 static int tcf_action_add(struct net *net, struct nlattr *nla,
1073                           struct nlmsghdr *n, u32 portid, int ovr)
1074 {
1075         int loop, ret;
1076         LIST_HEAD(actions);
1077
1078         for (loop = 0; loop < 10; loop++) {
1079                 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0,
1080                                       &actions);
1081                 if (ret != -EAGAIN)
1082                         break;
1083         }
1084
1085         if (ret)
1086                 return ret;
1087
1088         return tcf_add_notify(net, n, &actions, portid);
1089 }
1090
1091 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1092 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1093         [TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1094                              .validation_data = &tcaa_root_flags_allowed },
1095         [TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1096 };
1097
1098 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1099                          struct netlink_ext_ack *extack)
1100 {
1101         struct net *net = sock_net(skb->sk);
1102         struct nlattr *tca[TCA_ROOT_MAX + 1];
1103         u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1104         int ret = 0, ovr = 0;
1105
1106         if ((n->nlmsg_type != RTM_GETACTION) &&
1107             !netlink_capable(skb, CAP_NET_ADMIN))
1108                 return -EPERM;
1109
1110         ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1111                           extack);
1112         if (ret < 0)
1113                 return ret;
1114
1115         if (tca[TCA_ACT_TAB] == NULL) {
1116                 pr_notice("tc_ctl_action: received NO action attribs\n");
1117                 return -EINVAL;
1118         }
1119
1120         /* n->nlmsg_flags & NLM_F_CREATE */
1121         switch (n->nlmsg_type) {
1122         case RTM_NEWACTION:
1123                 /* we are going to assume all other flags
1124                  * imply create only if it doesn't exist
1125                  * Note that CREATE | EXCL implies that
1126                  * but since we want avoid ambiguity (eg when flags
1127                  * is zero) then just set this
1128                  */
1129                 if (n->nlmsg_flags & NLM_F_REPLACE)
1130                         ovr = 1;
1131                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr);
1132                 break;
1133         case RTM_DELACTION:
1134                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1135                                     portid, RTM_DELACTION);
1136                 break;
1137         case RTM_GETACTION:
1138                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1139                                     portid, RTM_GETACTION);
1140                 break;
1141         default:
1142                 BUG();
1143         }
1144
1145         return ret;
1146 }
1147
1148 static struct nlattr *find_dump_kind(struct nlattr **nla)
1149 {
1150         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1151         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1152         struct nlattr *kind;
1153
1154         tb1 = nla[TCA_ACT_TAB];
1155         if (tb1 == NULL)
1156                 return NULL;
1157
1158         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1159                       NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1160                 return NULL;
1161
1162         if (tb[1] == NULL)
1163                 return NULL;
1164         if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1165                 return NULL;
1166         kind = tb2[TCA_ACT_KIND];
1167
1168         return kind;
1169 }
1170
1171 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1172 {
1173         struct net *net = sock_net(skb->sk);
1174         struct nlmsghdr *nlh;
1175         unsigned char *b = skb_tail_pointer(skb);
1176         struct nlattr *nest;
1177         struct tc_action_ops *a_o;
1178         int ret = 0;
1179         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1180         struct nlattr *tb[TCA_ROOT_MAX + 1];
1181         struct nlattr *count_attr = NULL;
1182         unsigned long jiffy_since = 0;
1183         struct nlattr *kind = NULL;
1184         struct nla_bitfield32 bf;
1185         u32 msecs_since = 0;
1186         u32 act_count = 0;
1187
1188         ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1189                           tcaa_policy, NULL);
1190         if (ret < 0)
1191                 return ret;
1192
1193         kind = find_dump_kind(tb);
1194         if (kind == NULL) {
1195                 pr_info("tc_dump_action: action bad kind\n");
1196                 return 0;
1197         }
1198
1199         a_o = tc_lookup_action(kind);
1200         if (a_o == NULL)
1201                 return 0;
1202
1203         cb->args[2] = 0;
1204         if (tb[TCA_ROOT_FLAGS]) {
1205                 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1206                 cb->args[2] = bf.value;
1207         }
1208
1209         if (tb[TCA_ROOT_TIME_DELTA]) {
1210                 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1211         }
1212
1213         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1214                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1215         if (!nlh)
1216                 goto out_module_put;
1217
1218         if (msecs_since)
1219                 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1220
1221         t = nlmsg_data(nlh);
1222         t->tca_family = AF_UNSPEC;
1223         t->tca__pad1 = 0;
1224         t->tca__pad2 = 0;
1225         cb->args[3] = jiffy_since;
1226         count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1227         if (!count_attr)
1228                 goto out_module_put;
1229
1230         nest = nla_nest_start(skb, TCA_ACT_TAB);
1231         if (nest == NULL)
1232                 goto out_module_put;
1233
1234         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o);
1235         if (ret < 0)
1236                 goto out_module_put;
1237
1238         if (ret > 0) {
1239                 nla_nest_end(skb, nest);
1240                 ret = skb->len;
1241                 act_count = cb->args[1];
1242                 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1243                 cb->args[1] = 0;
1244         } else
1245                 nlmsg_trim(skb, b);
1246
1247         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1248         if (NETLINK_CB(cb->skb).portid && ret)
1249                 nlh->nlmsg_flags |= NLM_F_MULTI;
1250         module_put(a_o->owner);
1251         return skb->len;
1252
1253 out_module_put:
1254         module_put(a_o->owner);
1255         nlmsg_trim(skb, b);
1256         return skb->len;
1257 }
1258
1259 static int __init tc_action_init(void)
1260 {
1261         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1262         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1263         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1264                       0);
1265
1266         return 0;
1267 }
1268
1269 subsys_initcall(tc_action_init);