net/sched/cls_api.c

   1 /*
   2  * net/sched/cls_api.c  Packet classifier API.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *
  11  * Changes:
  12  *
  13  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
  14  *
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/errno.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/init.h>
  24 #include <linux/kmod.h>
  25 #include <linux/netlink.h>
  26 #include <linux/err.h>
  27 #include <net/net_namespace.h>
  28 #include <net/sock.h>
  29 #include <net/netlink.h>
  30 #include <net/pkt_sched.h>
  31 #include <net/pkt_cls.h>
  32
  33 /* The list of all installed classifier types */
  34
  35 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
  36
  37 /* Protects list of registered TC modules. It is pure SMP lock. */
  38 static DEFINE_RWLOCK(cls_mod_lock);
  39
  40 /* Find classifier type by string name */
  41
  42 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
  43 {
  44         struct tcf_proto_ops *t = NULL;
  45
  46         if (kind) {
  47                 read_lock(&cls_mod_lock);
  48                 for (t = tcf_proto_base; t; t = t->next) {
  49                         if (nla_strcmp(kind, t->kind) == 0) {
  50                                 if (!try_module_get(t->owner))
  51                                         t = NULL;
  52                                 break;
  53                         }
  54                 }
  55                 read_unlock(&cls_mod_lock);
  56         }
  57         return t;
  58 }
  59
  60 /* Register(unregister) new classifier type */
  61
  62 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  63 {
  64         struct tcf_proto_ops *t, **tp;
  65         int rc = -EEXIST;
  66
  67         write_lock(&cls_mod_lock);
  68         for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
  69                 if (!strcmp(ops->kind, t->kind))
  70                         goto out;
  71
  72         ops->next = NULL;
  73         *tp = ops;
  74         rc = 0;
  75 out:
  76         write_unlock(&cls_mod_lock);
  77         return rc;
  78 }
  79 EXPORT_SYMBOL(register_tcf_proto_ops);
  80
  81 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
  82 {
  83         struct tcf_proto_ops *t, **tp;
  84         int rc = -ENOENT;
  85
  86         write_lock(&cls_mod_lock);
  87         for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
  88                 if (t == ops)
  89                         break;
  90
  91         if (!t)
  92                 goto out;
  93         *tp = t->next;
  94         rc = 0;
  95 out:
  96         write_unlock(&cls_mod_lock);
  97         return rc;
  98 }
  99 EXPORT_SYMBOL(unregister_tcf_proto_ops);
 100
 101 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
 102                           struct tcf_proto *tp, unsigned long fh, int event);
 103
 104
 105 /* Select new prio value from the range, managed by kernel. */
 106
 107 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 108 {
 109         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 110
 111         if (tp)
 112                 first = tp->prio-1;
 113
 114         return first;
 115 }
 116
 117 /* Add/change/delete/get a filter node */
 118
 119 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
 120 {
 121         struct net *net = sock_net(skb->sk);
 122         struct nlattr *tca[TCA_MAX + 1];
 123         spinlock_t *root_lock;
 124         struct tcmsg *t;
 125         u32 protocol;
 126         u32 prio;
 127         u32 nprio;
 128         u32 parent;
 129         struct net_device *dev;
 130         struct Qdisc  *q;
 131         struct tcf_proto **back, **chain;
 132         struct tcf_proto *tp;
 133         struct tcf_proto_ops *tp_ops;
 134         const struct Qdisc_class_ops *cops;
 135         unsigned long cl;
 136         unsigned long fh;
 137         int err;
 138         int tp_created = 0;
 139
 140         if (net != &init_net)
 141                 return -EINVAL;
 142
 143 replay:
 144         t = NLMSG_DATA(n);
 145         protocol = TC_H_MIN(t->tcm_info);
 146         prio = TC_H_MAJ(t->tcm_info);
 147         nprio = prio;
 148         parent = t->tcm_parent;
 149         cl = 0;
 150
 151         if (prio == 0) {
 152                 /* If no priority is given, user wants we allocated it. */
 153                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
 154                         return -ENOENT;
 155                 prio = TC_H_MAKE(0x80000000U, 0U);
 156         }
 157
 158         /* Find head of filter chain. */
 159
 160         /* Find link */
 161         dev = __dev_get_by_index(&init_net, t->tcm_ifindex);
 162         if (dev == NULL)
 163                 return -ENODEV;
 164
 165         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
 166         if (err < 0)
 167                 return err;
 168
 169         /* Find qdisc */
 170         if (!parent) {
 171                 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, 0);
 172                 q = dev_queue->qdisc_sleeping;
 173                 parent = q->handle;
 174         } else {
 175                 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
 176                 if (q == NULL)
 177                         return -EINVAL;
 178         }
 179
 180         /* Is it classful? */
 181         if ((cops = q->ops->cl_ops) == NULL)
 182                 return -EINVAL;
 183
 184         /* Do we search for filter, attached to class? */
 185         if (TC_H_MIN(parent)) {
 186                 cl = cops->get(q, parent);
 187                 if (cl == 0)
 188                         return -ENOENT;
 189         }
 190
 191         /* And the last stroke */
 192         chain = cops->tcf_chain(q, cl);
 193         err = -EINVAL;
 194         if (chain == NULL)
 195                 goto errout;
 196
 197         /* Check the chain for existence of proto-tcf with this priority */
 198         for (back = chain; (tp=*back) != NULL; back = &tp->next) {
 199                 if (tp->prio >= prio) {
 200                         if (tp->prio == prio) {
 201                                 if (!nprio || (tp->protocol != protocol && protocol))
 202                                         goto errout;
 203                         } else
 204                                 tp = NULL;
 205                         break;
 206                 }
 207         }
 208
 209         root_lock = qdisc_root_sleeping_lock(q);
 210
 211         if (tp == NULL) {
 212                 /* Proto-tcf does not exist, create new one */
 213
 214                 if (tca[TCA_KIND] == NULL || !protocol)
 215                         goto errout;
 216
 217                 err = -ENOENT;
 218                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
 219                         goto errout;
 220
 221
 222                 /* Create new proto tcf */
 223
 224                 err = -ENOBUFS;
 225                 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 226                 if (tp == NULL)
 227                         goto errout;
 228                 err = -ENOENT;
 229                 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
 230                 if (tp_ops == NULL) {
 231 #ifdef CONFIG_MODULES
 232                         struct nlattr *kind = tca[TCA_KIND];
 233                         char name[IFNAMSIZ];
 234
 235                         if (kind != NULL &&
 236                             nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
 237                                 rtnl_unlock();
 238                                 request_module("cls_%s", name);
 239                                 rtnl_lock();
 240                                 tp_ops = tcf_proto_lookup_ops(kind);
 241                                 /* We dropped the RTNL semaphore in order to
 242                                  * perform the module load.  So, even if we
 243                                  * succeeded in loading the module we have to
 244                                  * replay the request.  We indicate this using
 245                                  * -EAGAIN.
 246                                  */
 247                                 if (tp_ops != NULL) {
 248                                         module_put(tp_ops->owner);
 249                                         err = -EAGAIN;
 250                                 }
 251                         }
 252 #endif
 253                         kfree(tp);
 254                         goto errout;
 255                 }
 256                 tp->ops = tp_ops;
 257                 tp->protocol = protocol;
 258                 tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
 259                 tp->q = q;
 260                 tp->classify = tp_ops->classify;
 261                 tp->classid = parent;
 262
 263                 err = tp_ops->init(tp);
 264                 if (err != 0) {
 265                         module_put(tp_ops->owner);
 266                         kfree(tp);
 267                         goto errout;
 268                 }
 269
 270                 tp_created = 1;
 271
 272         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
 273                 goto errout;
 274
 275         fh = tp->ops->get(tp, t->tcm_handle);
 276
 277         if (fh == 0) {
 278                 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
 279                         spin_lock_bh(root_lock);
 280                         *back = tp->next;
 281                         spin_unlock_bh(root_lock);
 282
 283                         tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
 284                         tcf_destroy(tp);
 285                         err = 0;
 286                         goto errout;
 287                 }
 288
 289                 err = -ENOENT;
 290                 if (n->nlmsg_type != RTM_NEWTFILTER ||
 291                     !(n->nlmsg_flags & NLM_F_CREATE))
 292                         goto errout;
 293         } else {
 294                 switch (n->nlmsg_type) {
 295                 case RTM_NEWTFILTER:
 296                         err = -EEXIST;
 297                         if (n->nlmsg_flags & NLM_F_EXCL) {
 298                                 if (tp_created)
 299                                         tcf_destroy(tp);
 300                                 goto errout;
 301                         }
 302                         break;
 303                 case RTM_DELTFILTER:
 304                         err = tp->ops->delete(tp, fh);
 305                         if (err == 0)
 306                                 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
 307                         goto errout;
 308                 case RTM_GETTFILTER:
 309                         err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
 310                         goto errout;
 311                 default:
 312                         err = -EINVAL;
 313                         goto errout;
 314                 }
 315         }
 316
 317         err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
 318         if (err == 0) {
 319                 if (tp_created) {
 320                         spin_lock_bh(root_lock);
 321                         tp->next = *back;
 322                         *back = tp;
 323                         spin_unlock_bh(root_lock);
 324                 }
 325                 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
 326         } else {
 327                 if (tp_created)
 328                         tcf_destroy(tp);
 329         }
 330
 331 errout:
 332         if (cl)
 333                 cops->put(q, cl);
 334         if (err == -EAGAIN)
 335                 /* Replay the request. */
 336                 goto replay;
 337         return err;
 338 }
 339
 340 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
 341                          unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
 342 {
 343         struct tcmsg *tcm;
 344         struct nlmsghdr  *nlh;
 345         unsigned char *b = skb_tail_pointer(skb);
 346
 347         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
 348         tcm = NLMSG_DATA(nlh);
 349         tcm->tcm_family = AF_UNSPEC;
 350         tcm->tcm__pad1 = 0;
 351         tcm->tcm__pad1 = 0;
 352         tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
 353         tcm->tcm_parent = tp->classid;
 354         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
 355         NLA_PUT_STRING(skb, TCA_KIND, tp->ops->kind);
 356         tcm->tcm_handle = fh;
 357         if (RTM_DELTFILTER != event) {
 358                 tcm->tcm_handle = 0;
 359                 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
 360                         goto nla_put_failure;
 361         }
 362         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 363         return skb->len;
 364
 365 nlmsg_failure:
 366 nla_put_failure:
 367         nlmsg_trim(skb, b);
 368         return -1;
 369 }
 370
 371 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
 372                           struct tcf_proto *tp, unsigned long fh, int event)
 373 {
 374         struct sk_buff *skb;
 375         u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
 376
 377         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 378         if (!skb)
 379                 return -ENOBUFS;
 380
 381         if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
 382                 kfree_skb(skb);
 383                 return -EINVAL;
 384         }
 385
 386         return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
 387                               n->nlmsg_flags & NLM_F_ECHO);
 388 }
 389
 390 struct tcf_dump_args {
 391         struct tcf_walker w;
 392         struct sk_buff *skb;
 393         struct netlink_callback *cb;
 394 };
 395
 396 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
 397                          struct tcf_walker *arg)
 398 {
 399         struct tcf_dump_args *a = (void *)arg;
 400
 401         return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
 402                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
 403 }
 404
 405 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
 406 {
 407         struct net *net = sock_net(skb->sk);
 408         struct netdev_queue *dev_queue;
 409         int t;
 410         int s_t;
 411         struct net_device *dev;
 412         struct Qdisc *q;
 413         struct tcf_proto *tp, **chain;
 414         struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
 415         unsigned long cl = 0;
 416         const struct Qdisc_class_ops *cops;
 417         struct tcf_dump_args arg;
 418
 419         if (net != &init_net)
 420                 return 0;
 421
 422         if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
 423                 return skb->len;
 424         if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
 425                 return skb->len;
 426
 427         dev_queue = netdev_get_tx_queue(dev, 0);
 428         if (!tcm->tcm_parent)
 429                 q = dev_queue->qdisc_sleeping;
 430         else
 431                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
 432         if (!q)
 433                 goto out;
 434         if ((cops = q->ops->cl_ops) == NULL)
 435                 goto errout;
 436         if (TC_H_MIN(tcm->tcm_parent)) {
 437                 cl = cops->get(q, tcm->tcm_parent);
 438                 if (cl == 0)
 439                         goto errout;
 440         }
 441         chain = cops->tcf_chain(q, cl);
 442         if (chain == NULL)
 443                 goto errout;
 444
 445         s_t = cb->args[0];
 446
 447         for (tp=*chain, t=0; tp; tp = tp->next, t++) {
 448                 if (t < s_t) continue;
 449                 if (TC_H_MAJ(tcm->tcm_info) &&
 450                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
 451                         continue;
 452                 if (TC_H_MIN(tcm->tcm_info) &&
 453                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
 454                         continue;
 455                 if (t > s_t)
 456                         memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
 457                 if (cb->args[1] == 0) {
 458                         if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
 459                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
 460                                           RTM_NEWTFILTER) <= 0)
 461                                 break;
 462
 463                         cb->args[1] = 1;
 464                 }
 465                 if (tp->ops->walk == NULL)
 466                         continue;
 467                 arg.w.fn = tcf_node_dump;
 468                 arg.skb = skb;
 469                 arg.cb = cb;
 470                 arg.w.stop = 0;
 471                 arg.w.skip = cb->args[1]-1;
 472                 arg.w.count = 0;
 473                 tp->ops->walk(tp, &arg.w);
 474                 cb->args[1] = arg.w.count+1;
 475                 if (arg.w.stop)
 476                         break;
 477         }
 478
 479         cb->args[0] = t;
 480
 481 errout:
 482         if (cl)
 483                 cops->put(q, cl);
 484 out:
 485         dev_put(dev);
 486         return skb->len;
 487 }
 488
 489 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
 490 {
 491 #ifdef CONFIG_NET_CLS_ACT
 492         if (exts->action) {
 493                 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
 494                 exts->action = NULL;
 495         }
 496 #endif
 497 }
 498 EXPORT_SYMBOL(tcf_exts_destroy);
 499
 500 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
 501                   struct nlattr *rate_tlv, struct tcf_exts *exts,
 502                   const struct tcf_ext_map *map)
 503 {
 504         memset(exts, 0, sizeof(*exts));
 505
 506 #ifdef CONFIG_NET_CLS_ACT
 507         {
 508                 struct tc_action *act;
 509
 510                 if (map->police && tb[map->police]) {
 511                         act = tcf_action_init_1(tb[map->police], rate_tlv,
 512                                                 "police", TCA_ACT_NOREPLACE,
 513                                                 TCA_ACT_BIND);
 514                         if (IS_ERR(act))
 515                                 return PTR_ERR(act);
 516
 517                         act->type = TCA_OLD_COMPAT;
 518                         exts->action = act;
 519                 } else if (map->action && tb[map->action]) {
 520                         act = tcf_action_init(tb[map->action], rate_tlv, NULL,
 521                                               TCA_ACT_NOREPLACE, TCA_ACT_BIND);
 522                         if (IS_ERR(act))
 523                                 return PTR_ERR(act);
 524
 525                         exts->action = act;
 526                 }
 527         }
 528 #else
 529         if ((map->action && tb[map->action]) ||
 530             (map->police && tb[map->police]))
 531                 return -EOPNOTSUPP;
 532 #endif
 533
 534         return 0;
 535 }
 536 EXPORT_SYMBOL(tcf_exts_validate);
 537
 538 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
 539                      struct tcf_exts *src)
 540 {
 541 #ifdef CONFIG_NET_CLS_ACT
 542         if (src->action) {
 543                 struct tc_action *act;
 544                 tcf_tree_lock(tp);
 545                 act = dst->action;
 546                 dst->action = src->action;
 547                 tcf_tree_unlock(tp);
 548                 if (act)
 549                         tcf_action_destroy(act, TCA_ACT_UNBIND);
 550         }
 551 #endif
 552 }
 553 EXPORT_SYMBOL(tcf_exts_change);
 554
 555 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
 556                   const struct tcf_ext_map *map)
 557 {
 558 #ifdef CONFIG_NET_CLS_ACT
 559         if (map->action && exts->action) {
 560                 /*
 561                  * again for backward compatible mode - we want
 562                  * to work with both old and new modes of entering
 563                  * tc data even if iproute2  was newer - jhs
 564                  */
 565                 struct nlattr *nest;
 566
 567                 if (exts->action->type != TCA_OLD_COMPAT) {
 568                         nest = nla_nest_start(skb, map->action);
 569                         if (nest == NULL)
 570                                 goto nla_put_failure;
 571                         if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
 572                                 goto nla_put_failure;
 573                         nla_nest_end(skb, nest);
 574                 } else if (map->police) {
 575                         nest = nla_nest_start(skb, map->police);
 576                         if (nest == NULL)
 577                                 goto nla_put_failure;
 578                         if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
 579                                 goto nla_put_failure;
 580                         nla_nest_end(skb, nest);
 581                 }
 582         }
 583 #endif
 584         return 0;
 585 nla_put_failure: __attribute__ ((unused))
 586         return -1;
 587 }
 588 EXPORT_SYMBOL(tcf_exts_dump);
 589
 590
 591 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
 592                         const struct tcf_ext_map *map)
 593 {
 594 #ifdef CONFIG_NET_CLS_ACT
 595         if (exts->action)
 596                 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
 597                         goto nla_put_failure;
 598 #endif
 599         return 0;
 600 nla_put_failure: __attribute__ ((unused))
 601         return -1;
 602 }
 603 EXPORT_SYMBOL(tcf_exts_dump_stats);
 604
 605 static int __init tc_filter_init(void)
 606 {
 607         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
 608         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
 609         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
 610                                                  tc_dump_tfilter);
 611
 612         return 0;
 613 }
 614
 615 subsys_initcall(tc_filter_init);