SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 <net/net_namespace.h>
  27 #include <net/sock.h>
  28 #include <net/netlink.h>
  29 #include <net/pkt_sched.h>
  30 #include <net/pkt_cls.h>
  31
  32 /* The list of all installed classifier types */
  33
  34 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
  35
  36 /* Protects list of registered TC modules. It is pure SMP lock. */
  37 static DEFINE_RWLOCK(cls_mod_lock);
  38
  39 /* Find classifier type by string name */
  40
  41 static struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
  42 {
  43         struct tcf_proto_ops *t = NULL;
  44
  45         if (kind) {
  46                 read_lock(&cls_mod_lock);
  47                 for (t = tcf_proto_base; t; t = t->next) {
  48                         if (nla_strcmp(kind, t->kind) == 0) {
  49                                 if (!try_module_get(t->owner))
  50                                         t = NULL;
  51                                 break;
  52                         }
  53                 }
  54                 read_unlock(&cls_mod_lock);
  55         }
  56         return t;
  57 }
  58
  59 /* Register(unregister) new classifier type */
  60
  61 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
  62 {
  63         struct tcf_proto_ops *t, **tp;
  64         int rc = -EEXIST;
  65
  66         write_lock(&cls_mod_lock);
  67         for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
  68                 if (!strcmp(ops->kind, t->kind))
  69                         goto out;
  70
  71         ops->next = NULL;
  72         *tp = ops;
  73         rc = 0;
  74 out:
  75         write_unlock(&cls_mod_lock);
  76         return rc;
  77 }
  78 EXPORT_SYMBOL(register_tcf_proto_ops);
  79
  80 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
  81 {
  82         struct tcf_proto_ops *t, **tp;
  83         int rc = -ENOENT;
  84
  85         write_lock(&cls_mod_lock);
  86         for (tp = &tcf_proto_base; (t=*tp) != NULL; tp = &t->next)
  87                 if (t == ops)
  88                         break;
  89
  90         if (!t)
  91                 goto out;
  92         *tp = t->next;
  93         rc = 0;
  94 out:
  95         write_unlock(&cls_mod_lock);
  96         return rc;
  97 }
  98 EXPORT_SYMBOL(unregister_tcf_proto_ops);
  99
 100 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
 101                           struct tcf_proto *tp, unsigned long fh, int event);
 102
 103
 104 /* Select new prio value from the range, managed by kernel. */
 105
 106 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
 107 {
 108         u32 first = TC_H_MAKE(0xC0000000U, 0U);
 109
 110         if (tp)
 111                 first = tp->prio-1;
 112
 113         return first;
 114 }
 115
 116 /* Add/change/delete/get a filter node */
 117
 118 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
 119 {
 120         struct net *net = skb->sk->sk_net;
 121         struct nlattr *tca[TCA_MAX + 1];
 122         struct tcmsg *t;
 123         u32 protocol;
 124         u32 prio;
 125         u32 nprio;
 126         u32 parent;
 127         struct net_device *dev;
 128         struct Qdisc  *q;
 129         struct tcf_proto **back, **chain;
 130         struct tcf_proto *tp;
 131         struct tcf_proto_ops *tp_ops;
 132         const struct Qdisc_class_ops *cops;
 133         unsigned long cl;
 134         unsigned long fh;
 135         int err;
 136
 137         if (net != &init_net)
 138                 return -EINVAL;
 139
 140 replay:
 141         t = NLMSG_DATA(n);
 142         protocol = TC_H_MIN(t->tcm_info);
 143         prio = TC_H_MAJ(t->tcm_info);
 144         nprio = prio;
 145         parent = t->tcm_parent;
 146         cl = 0;
 147
 148         if (prio == 0) {
 149                 /* If no priority is given, user wants we allocated it. */
 150                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
 151                         return -ENOENT;
 152                 prio = TC_H_MAKE(0x80000000U, 0U);
 153         }
 154
 155         /* Find head of filter chain. */
 156
 157         /* Find link */
 158         dev = __dev_get_by_index(&init_net, t->tcm_ifindex);
 159         if (dev == NULL)
 160                 return -ENODEV;
 161
 162         err = nlmsg_parse(n, sizeof(*t), tca, TCA_MAX, NULL);
 163         if (err < 0)
 164                 return err;
 165
 166         /* Find qdisc */
 167         if (!parent) {
 168                 q = dev->qdisc_sleeping;
 169                 parent = q->handle;
 170         } else {
 171                 q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
 172                 if (q == NULL)
 173                         return -EINVAL;
 174         }
 175
 176         /* Is it classful? */
 177         if ((cops = q->ops->cl_ops) == NULL)
 178                 return -EINVAL;
 179
 180         /* Do we search for filter, attached to class? */
 181         if (TC_H_MIN(parent)) {
 182                 cl = cops->get(q, parent);
 183                 if (cl == 0)
 184                         return -ENOENT;
 185         }
 186
 187         /* And the last stroke */
 188         chain = cops->tcf_chain(q, cl);
 189         err = -EINVAL;
 190         if (chain == NULL)
 191                 goto errout;
 192
 193         /* Check the chain for existence of proto-tcf with this priority */
 194         for (back = chain; (tp=*back) != NULL; back = &tp->next) {
 195                 if (tp->prio >= prio) {
 196                         if (tp->prio == prio) {
 197                                 if (!nprio || (tp->protocol != protocol && protocol))
 198                                         goto errout;
 199                         } else
 200                                 tp = NULL;
 201                         break;
 202                 }
 203         }
 204
 205         if (tp == NULL) {
 206                 /* Proto-tcf does not exist, create new one */
 207
 208                 if (tca[TCA_KIND] == NULL || !protocol)
 209                         goto errout;
 210
 211                 err = -ENOENT;
 212                 if (n->nlmsg_type != RTM_NEWTFILTER || !(n->nlmsg_flags&NLM_F_CREATE))
 213                         goto errout;
 214
 215
 216                 /* Create new proto tcf */
 217
 218                 err = -ENOBUFS;
 219                 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
 220                 if (tp == NULL)
 221                         goto errout;
 222                 err = -EINVAL;
 223                 tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
 224                 if (tp_ops == NULL) {
 225 #ifdef CONFIG_KMOD
 226                         struct nlattr *kind = tca[TCA_KIND];
 227                         char name[IFNAMSIZ];
 228
 229                         if (kind != NULL &&
 230                             nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
 231                                 rtnl_unlock();
 232                                 request_module("cls_%s", name);
 233                                 rtnl_lock();
 234                                 tp_ops = tcf_proto_lookup_ops(kind);
 235                                 /* We dropped the RTNL semaphore in order to
 236                                  * perform the module load.  So, even if we
 237                                  * succeeded in loading the module we have to
 238                                  * replay the request.  We indicate this using
 239                                  * -EAGAIN.
 240                                  */
 241                                 if (tp_ops != NULL) {
 242                                         module_put(tp_ops->owner);
 243                                         err = -EAGAIN;
 244                                 }
 245                         }
 246 #endif
 247                         kfree(tp);
 248                         goto errout;
 249                 }
 250                 tp->ops = tp_ops;
 251                 tp->protocol = protocol;
 252                 tp->prio = nprio ? : tcf_auto_prio(*back);
 253                 tp->q = q;
 254                 tp->classify = tp_ops->classify;
 255                 tp->classid = parent;
 256
 257                 err = tp_ops->init(tp);
 258                 if (err != 0) {
 259                         module_put(tp_ops->owner);
 260                         kfree(tp);
 261                         goto errout;
 262                 }
 263
 264                 qdisc_lock_tree(dev);
 265                 tp->next = *back;
 266                 *back = tp;
 267                 qdisc_unlock_tree(dev);
 268
 269         } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
 270                 goto errout;
 271
 272         fh = tp->ops->get(tp, t->tcm_handle);
 273
 274         if (fh == 0) {
 275                 if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
 276                         qdisc_lock_tree(dev);
 277                         *back = tp->next;
 278                         qdisc_unlock_tree(dev);
 279
 280                         tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
 281                         tcf_destroy(tp);
 282                         err = 0;
 283                         goto errout;
 284                 }
 285
 286                 err = -ENOENT;
 287                 if (n->nlmsg_type != RTM_NEWTFILTER ||
 288                     !(n->nlmsg_flags & NLM_F_CREATE))
 289                         goto errout;
 290         } else {
 291                 switch (n->nlmsg_type) {
 292                 case RTM_NEWTFILTER:
 293                         err = -EEXIST;
 294                         if (n->nlmsg_flags & NLM_F_EXCL)
 295                                 goto errout;
 296                         break;
 297                 case RTM_DELTFILTER:
 298                         err = tp->ops->delete(tp, fh);
 299                         if (err == 0)
 300                                 tfilter_notify(skb, n, tp, fh, RTM_DELTFILTER);
 301                         goto errout;
 302                 case RTM_GETTFILTER:
 303                         err = tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
 304                         goto errout;
 305                 default:
 306                         err = -EINVAL;
 307                         goto errout;
 308                 }
 309         }
 310
 311         err = tp->ops->change(tp, cl, t->tcm_handle, tca, &fh);
 312         if (err == 0)
 313                 tfilter_notify(skb, n, tp, fh, RTM_NEWTFILTER);
 314
 315 errout:
 316         if (cl)
 317                 cops->put(q, cl);
 318         if (err == -EAGAIN)
 319                 /* Replay the request. */
 320                 goto replay;
 321         return err;
 322 }
 323
 324 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
 325                          unsigned long fh, u32 pid, u32 seq, u16 flags, int event)
 326 {
 327         struct tcmsg *tcm;
 328         struct nlmsghdr  *nlh;
 329         unsigned char *b = skb_tail_pointer(skb);
 330
 331         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
 332         tcm = NLMSG_DATA(nlh);
 333         tcm->tcm_family = AF_UNSPEC;
 334         tcm->tcm__pad1 = 0;
 335         tcm->tcm__pad1 = 0;
 336         tcm->tcm_ifindex = tp->q->dev->ifindex;
 337         tcm->tcm_parent = tp->classid;
 338         tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
 339         NLA_PUT(skb, TCA_KIND, IFNAMSIZ, tp->ops->kind);
 340         tcm->tcm_handle = fh;
 341         if (RTM_DELTFILTER != event) {
 342                 tcm->tcm_handle = 0;
 343                 if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
 344                         goto nla_put_failure;
 345         }
 346         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 347         return skb->len;
 348
 349 nlmsg_failure:
 350 nla_put_failure:
 351         nlmsg_trim(skb, b);
 352         return -1;
 353 }
 354
 355 static int tfilter_notify(struct sk_buff *oskb, struct nlmsghdr *n,
 356                           struct tcf_proto *tp, unsigned long fh, int event)
 357 {
 358         struct sk_buff *skb;
 359         u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
 360
 361         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 362         if (!skb)
 363                 return -ENOBUFS;
 364
 365         if (tcf_fill_node(skb, tp, fh, pid, n->nlmsg_seq, 0, event) <= 0) {
 366                 kfree_skb(skb);
 367                 return -EINVAL;
 368         }
 369
 370         return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
 371                               n->nlmsg_flags & NLM_F_ECHO);
 372 }
 373
 374 struct tcf_dump_args {
 375         struct tcf_walker w;
 376         struct sk_buff *skb;
 377         struct netlink_callback *cb;
 378 };
 379
 380 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
 381                          struct tcf_walker *arg)
 382 {
 383         struct tcf_dump_args *a = (void *)arg;
 384
 385         return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).pid,
 386                              a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
 387 }
 388
 389 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
 390 {
 391         struct net *net = skb->sk->sk_net;
 392         int t;
 393         int s_t;
 394         struct net_device *dev;
 395         struct Qdisc *q;
 396         struct tcf_proto *tp, **chain;
 397         struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
 398         unsigned long cl = 0;
 399         const struct Qdisc_class_ops *cops;
 400         struct tcf_dump_args arg;
 401
 402         if (net != &init_net)
 403                 return 0;
 404
 405         if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
 406                 return skb->len;
 407         if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
 408                 return skb->len;
 409
 410         if (!tcm->tcm_parent)
 411                 q = dev->qdisc_sleeping;
 412         else
 413                 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
 414         if (!q)
 415                 goto out;
 416         if ((cops = q->ops->cl_ops) == NULL)
 417                 goto errout;
 418         if (TC_H_MIN(tcm->tcm_parent)) {
 419                 cl = cops->get(q, tcm->tcm_parent);
 420                 if (cl == 0)
 421                         goto errout;
 422         }
 423         chain = cops->tcf_chain(q, cl);
 424         if (chain == NULL)
 425                 goto errout;
 426
 427         s_t = cb->args[0];
 428
 429         for (tp=*chain, t=0; tp; tp = tp->next, t++) {
 430                 if (t < s_t) continue;
 431                 if (TC_H_MAJ(tcm->tcm_info) &&
 432                     TC_H_MAJ(tcm->tcm_info) != tp->prio)
 433                         continue;
 434                 if (TC_H_MIN(tcm->tcm_info) &&
 435                     TC_H_MIN(tcm->tcm_info) != tp->protocol)
 436                         continue;
 437                 if (t > s_t)
 438                         memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
 439                 if (cb->args[1] == 0) {
 440                         if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).pid,
 441                                           cb->nlh->nlmsg_seq, NLM_F_MULTI,
 442                                           RTM_NEWTFILTER) <= 0)
 443                                 break;
 444
 445                         cb->args[1] = 1;
 446                 }
 447                 if (tp->ops->walk == NULL)
 448                         continue;
 449                 arg.w.fn = tcf_node_dump;
 450                 arg.skb = skb;
 451                 arg.cb = cb;
 452                 arg.w.stop = 0;
 453                 arg.w.skip = cb->args[1]-1;
 454                 arg.w.count = 0;
 455                 tp->ops->walk(tp, &arg.w);
 456                 cb->args[1] = arg.w.count+1;
 457                 if (arg.w.stop)
 458                         break;
 459         }
 460
 461         cb->args[0] = t;
 462
 463 errout:
 464         if (cl)
 465                 cops->put(q, cl);
 466 out:
 467         dev_put(dev);
 468         return skb->len;
 469 }
 470
 471 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
 472 {
 473 #ifdef CONFIG_NET_CLS_ACT
 474         if (exts->action) {
 475                 tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
 476                 exts->action = NULL;
 477         }
 478 #endif
 479 }
 480 EXPORT_SYMBOL(tcf_exts_destroy);
 481
 482 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
 483                   struct nlattr *rate_tlv, struct tcf_exts *exts,
 484                   struct tcf_ext_map *map)
 485 {
 486         memset(exts, 0, sizeof(*exts));
 487
 488 #ifdef CONFIG_NET_CLS_ACT
 489         {
 490                 int err;
 491                 struct tc_action *act;
 492
 493                 if (map->police && tb[map->police]) {
 494                         act = tcf_action_init_1((struct rtattr *)tb[map->police],
 495                                                 (struct rtattr *)rate_tlv,
 496                                                 "police", TCA_ACT_NOREPLACE,
 497                                                 TCA_ACT_BIND, &err);
 498                         if (act == NULL)
 499                                 return err;
 500
 501                         act->type = TCA_OLD_COMPAT;
 502                         exts->action = act;
 503                 } else if (map->action && tb[map->action]) {
 504                         act = tcf_action_init((struct rtattr *)tb[map->action],
 505                                               (struct rtattr *)rate_tlv, NULL,
 506                                 TCA_ACT_NOREPLACE, TCA_ACT_BIND, &err);
 507                         if (act == NULL)
 508                                 return err;
 509
 510                         exts->action = act;
 511                 }
 512         }
 513 #else
 514         if ((map->action && tb[map->action]) ||
 515             (map->police && tb[map->police]))
 516                 return -EOPNOTSUPP;
 517 #endif
 518
 519         return 0;
 520 }
 521 EXPORT_SYMBOL(tcf_exts_validate);
 522
 523 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
 524                      struct tcf_exts *src)
 525 {
 526 #ifdef CONFIG_NET_CLS_ACT
 527         if (src->action) {
 528                 struct tc_action *act;
 529                 tcf_tree_lock(tp);
 530                 act = xchg(&dst->action, src->action);
 531                 tcf_tree_unlock(tp);
 532                 if (act)
 533                         tcf_action_destroy(act, TCA_ACT_UNBIND);
 534         }
 535 #endif
 536 }
 537 EXPORT_SYMBOL(tcf_exts_change);
 538
 539 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
 540               struct tcf_ext_map *map)
 541 {
 542 #ifdef CONFIG_NET_CLS_ACT
 543         if (map->action && exts->action) {
 544                 /*
 545                  * again for backward compatible mode - we want
 546                  * to work with both old and new modes of entering
 547                  * tc data even if iproute2  was newer - jhs
 548                  */
 549                 struct nlattr *p_rta = (struct nlattr *)skb_tail_pointer(skb);
 550
 551                 if (exts->action->type != TCA_OLD_COMPAT) {
 552                         NLA_PUT(skb, map->action, 0, NULL);
 553                         if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
 554                                 goto nla_put_failure;
 555                         p_rta->nla_len = skb_tail_pointer(skb) - (u8 *)p_rta;
 556                 } else if (map->police) {
 557                         NLA_PUT(skb, map->police, 0, NULL);
 558                         if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
 559                                 goto nla_put_failure;
 560                         p_rta->nla_len = skb_tail_pointer(skb) - (u8 *)p_rta;
 561                 }
 562         }
 563 #endif
 564         return 0;
 565 nla_put_failure: __attribute__ ((unused))
 566         return -1;
 567 }
 568 EXPORT_SYMBOL(tcf_exts_dump);
 569
 570
 571 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
 572                         struct tcf_ext_map *map)
 573 {
 574 #ifdef CONFIG_NET_CLS_ACT
 575         if (exts->action)
 576                 if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
 577                         goto nla_put_failure;
 578 #endif
 579         return 0;
 580 nla_put_failure: __attribute__ ((unused))
 581         return -1;
 582 }
 583 EXPORT_SYMBOL(tcf_exts_dump_stats);
 584
 585 static int __init tc_filter_init(void)
 586 {
 587         rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL);
 588         rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL);
 589         rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
 590                                                  tc_dump_tfilter);
 591
 592         return 0;
 593 }
 594
 595 subsys_initcall(tc_filter_init);