Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
cls_api.c
Go to the documentation of this file.
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, <[email protected]>
10  *
11  * Changes:
12  *
13  * Eduardo J. Blanco <[email protected]> :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 <linux/slab.h>
28 #include <net/net_namespace.h>
29 #include <net/sock.h>
30 #include <net/netlink.h>
31 #include <net/pkt_sched.h>
32 #include <net/pkt_cls.h>
33 
34 /* The list of all installed classifier types */
35 
36 static struct tcf_proto_ops *tcf_proto_base __read_mostly;
37 
38 /* Protects list of registered TC modules. It is pure SMP lock. */
39 static DEFINE_RWLOCK(cls_mod_lock);
40 
41 /* Find classifier type by string name */
42 
43 static const struct tcf_proto_ops *tcf_proto_lookup_ops(struct nlattr *kind)
44 {
45  const struct tcf_proto_ops *t = NULL;
46 
47  if (kind) {
48  read_lock(&cls_mod_lock);
49  for (t = tcf_proto_base; t; t = t->next) {
50  if (nla_strcmp(kind, t->kind) == 0) {
51  if (!try_module_get(t->owner))
52  t = NULL;
53  break;
54  }
55  }
56  read_unlock(&cls_mod_lock);
57  }
58  return t;
59 }
60 
61 /* Register(unregister) new classifier type */
62 
64 {
65  struct tcf_proto_ops *t, **tp;
66  int rc = -EEXIST;
67 
68  write_lock(&cls_mod_lock);
69  for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
70  if (!strcmp(ops->kind, t->kind))
71  goto out;
72 
73  ops->next = NULL;
74  *tp = ops;
75  rc = 0;
76 out:
77  write_unlock(&cls_mod_lock);
78  return rc;
79 }
81 
83 {
84  struct tcf_proto_ops *t, **tp;
85  int rc = -ENOENT;
86 
87  write_lock(&cls_mod_lock);
88  for (tp = &tcf_proto_base; (t = *tp) != NULL; tp = &t->next)
89  if (t == ops)
90  break;
91 
92  if (!t)
93  goto out;
94  *tp = t->next;
95  rc = 0;
96 out:
97  write_unlock(&cls_mod_lock);
98  return rc;
99 }
101 
102 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
103  struct nlmsghdr *n, struct tcf_proto *tp,
104  unsigned long fh, int event);
105 
106 
107 /* Select new prio value from the range, managed by kernel. */
108 
109 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
110 {
111  u32 first = TC_H_MAKE(0xC0000000U, 0U);
112 
113  if (tp)
114  first = tp->prio - 1;
115 
116  return first;
117 }
118 
119 /* Add/change/delete/get a filter node */
120 
121 static int tc_ctl_tfilter(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
122 {
123  struct net *net = sock_net(skb->sk);
124  struct nlattr *tca[TCA_MAX + 1];
125  spinlock_t *root_lock;
126  struct tcmsg *t;
127  u32 protocol;
128  u32 prio;
129  u32 nprio;
130  u32 parent;
131  struct net_device *dev;
132  struct Qdisc *q;
133  struct tcf_proto **back, **chain;
134  struct tcf_proto *tp;
135  const struct tcf_proto_ops *tp_ops;
136  const struct Qdisc_class_ops *cops;
137  unsigned long cl;
138  unsigned long fh;
139  int err;
140  int tp_created = 0;
141 
142 replay:
143  t = nlmsg_data(n);
144  protocol = TC_H_MIN(t->tcm_info);
145  prio = TC_H_MAJ(t->tcm_info);
146  nprio = prio;
147  parent = t->tcm_parent;
148  cl = 0;
149 
150  if (prio == 0) {
151  /* If no priority is given, user wants we allocated it. */
152  if (n->nlmsg_type != RTM_NEWTFILTER ||
153  !(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(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  q = dev->qdisc;
172  parent = q->handle;
173  } else {
174  q = qdisc_lookup(dev, TC_H_MAJ(t->tcm_parent));
175  if (q == NULL)
176  return -EINVAL;
177  }
178 
179  /* Is it classful? */
180  cops = q->ops->cl_ops;
181  if (!cops)
182  return -EINVAL;
183 
184  if (cops->tcf_chain == NULL)
185  return -EOPNOTSUPP;
186 
187  /* Do we search for filter, attached to class? */
188  if (TC_H_MIN(parent)) {
189  cl = cops->get(q, parent);
190  if (cl == 0)
191  return -ENOENT;
192  }
193 
194  /* And the last stroke */
195  chain = cops->tcf_chain(q, cl);
196  err = -EINVAL;
197  if (chain == NULL)
198  goto errout;
199 
200  /* Check the chain for existence of proto-tcf with this priority */
201  for (back = chain; (tp = *back) != NULL; back = &tp->next) {
202  if (tp->prio >= prio) {
203  if (tp->prio == prio) {
204  if (!nprio ||
205  (tp->protocol != protocol && protocol))
206  goto errout;
207  } else
208  tp = NULL;
209  break;
210  }
211  }
212 
213  root_lock = qdisc_root_sleeping_lock(q);
214 
215  if (tp == NULL) {
216  /* Proto-tcf does not exist, create new one */
217 
218  if (tca[TCA_KIND] == NULL || !protocol)
219  goto errout;
220 
221  err = -ENOENT;
222  if (n->nlmsg_type != RTM_NEWTFILTER ||
223  !(n->nlmsg_flags & NLM_F_CREATE))
224  goto errout;
225 
226 
227  /* Create new proto tcf */
228 
229  err = -ENOBUFS;
230  tp = kzalloc(sizeof(*tp), GFP_KERNEL);
231  if (tp == NULL)
232  goto errout;
233  err = -ENOENT;
234  tp_ops = tcf_proto_lookup_ops(tca[TCA_KIND]);
235  if (tp_ops == NULL) {
236 #ifdef CONFIG_MODULES
237  struct nlattr *kind = tca[TCA_KIND];
238  char name[IFNAMSIZ];
239 
240  if (kind != NULL &&
241  nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
242  rtnl_unlock();
243  request_module("cls_%s", name);
244  rtnl_lock();
245  tp_ops = tcf_proto_lookup_ops(kind);
246  /* We dropped the RTNL semaphore in order to
247  * perform the module load. So, even if we
248  * succeeded in loading the module we have to
249  * replay the request. We indicate this using
250  * -EAGAIN.
251  */
252  if (tp_ops != NULL) {
253  module_put(tp_ops->owner);
254  err = -EAGAIN;
255  }
256  }
257 #endif
258  kfree(tp);
259  goto errout;
260  }
261  tp->ops = tp_ops;
262  tp->protocol = protocol;
263  tp->prio = nprio ? : TC_H_MAJ(tcf_auto_prio(*back));
264  tp->q = q;
265  tp->classify = tp_ops->classify;
266  tp->classid = parent;
267 
268  err = tp_ops->init(tp);
269  if (err != 0) {
270  module_put(tp_ops->owner);
271  kfree(tp);
272  goto errout;
273  }
274 
275  tp_created = 1;
276 
277  } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind))
278  goto errout;
279 
280  fh = tp->ops->get(tp, t->tcm_handle);
281 
282  if (fh == 0) {
283  if (n->nlmsg_type == RTM_DELTFILTER && t->tcm_handle == 0) {
284  spin_lock_bh(root_lock);
285  *back = tp->next;
286  spin_unlock_bh(root_lock);
287 
288  tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
289  tcf_destroy(tp);
290  err = 0;
291  goto errout;
292  }
293 
294  err = -ENOENT;
295  if (n->nlmsg_type != RTM_NEWTFILTER ||
296  !(n->nlmsg_flags & NLM_F_CREATE))
297  goto errout;
298  } else {
299  switch (n->nlmsg_type) {
300  case RTM_NEWTFILTER:
301  err = -EEXIST;
302  if (n->nlmsg_flags & NLM_F_EXCL) {
303  if (tp_created)
304  tcf_destroy(tp);
305  goto errout;
306  }
307  break;
308  case RTM_DELTFILTER:
309  err = tp->ops->delete(tp, fh);
310  if (err == 0)
311  tfilter_notify(net, skb, n, tp, fh, RTM_DELTFILTER);
312  goto errout;
313  case RTM_GETTFILTER:
314  err = tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
315  goto errout;
316  default:
317  err = -EINVAL;
318  goto errout;
319  }
320  }
321 
322  err = tp->ops->change(skb, tp, cl, t->tcm_handle, tca, &fh);
323  if (err == 0) {
324  if (tp_created) {
325  spin_lock_bh(root_lock);
326  tp->next = *back;
327  *back = tp;
328  spin_unlock_bh(root_lock);
329  }
330  tfilter_notify(net, skb, n, tp, fh, RTM_NEWTFILTER);
331  } else {
332  if (tp_created)
333  tcf_destroy(tp);
334  }
335 
336 errout:
337  if (cl)
338  cops->put(q, cl);
339  if (err == -EAGAIN)
340  /* Replay the request. */
341  goto replay;
342  return err;
343 }
344 
345 static int tcf_fill_node(struct sk_buff *skb, struct tcf_proto *tp,
346  unsigned long fh, u32 portid, u32 seq, u16 flags, int event)
347 {
348  struct tcmsg *tcm;
349  struct nlmsghdr *nlh;
350  unsigned char *b = skb_tail_pointer(skb);
351 
352  nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
353  if (!nlh)
354  goto out_nlmsg_trim;
355  tcm = nlmsg_data(nlh);
356  tcm->tcm_family = AF_UNSPEC;
357  tcm->tcm__pad1 = 0;
358  tcm->tcm__pad2 = 0;
359  tcm->tcm_ifindex = qdisc_dev(tp->q)->ifindex;
360  tcm->tcm_parent = tp->classid;
361  tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
362  if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
363  goto nla_put_failure;
364  tcm->tcm_handle = fh;
365  if (RTM_DELTFILTER != event) {
366  tcm->tcm_handle = 0;
367  if (tp->ops->dump && tp->ops->dump(tp, fh, skb, tcm) < 0)
368  goto nla_put_failure;
369  }
370  nlh->nlmsg_len = skb_tail_pointer(skb) - b;
371  return skb->len;
372 
373 out_nlmsg_trim:
374 nla_put_failure:
375  nlmsg_trim(skb, b);
376  return -1;
377 }
378 
379 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
380  struct nlmsghdr *n, struct tcf_proto *tp,
381  unsigned long fh, int event)
382 {
383  struct sk_buff *skb;
384  u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
385 
386  skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
387  if (!skb)
388  return -ENOBUFS;
389 
390  if (tcf_fill_node(skb, tp, fh, portid, n->nlmsg_seq, 0, event) <= 0) {
391  kfree_skb(skb);
392  return -EINVAL;
393  }
394 
395  return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
396  n->nlmsg_flags & NLM_F_ECHO);
397 }
398 
400  struct tcf_walker w;
401  struct sk_buff *skb;
403 };
404 
405 static int tcf_node_dump(struct tcf_proto *tp, unsigned long n,
406  struct tcf_walker *arg)
407 {
408  struct tcf_dump_args *a = (void *)arg;
409 
410  return tcf_fill_node(a->skb, tp, n, NETLINK_CB(a->cb->skb).portid,
411  a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTFILTER);
412 }
413 
414 /* called with RTNL */
415 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
416 {
417  struct net *net = sock_net(skb->sk);
418  int t;
419  int s_t;
420  struct net_device *dev;
421  struct Qdisc *q;
422  struct tcf_proto *tp, **chain;
423  struct tcmsg *tcm = nlmsg_data(cb->nlh);
424  unsigned long cl = 0;
425  const struct Qdisc_class_ops *cops;
426  struct tcf_dump_args arg;
427 
428  if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
429  return skb->len;
430  dev = __dev_get_by_index(net, tcm->tcm_ifindex);
431  if (!dev)
432  return skb->len;
433 
434  if (!tcm->tcm_parent)
435  q = dev->qdisc;
436  else
437  q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
438  if (!q)
439  goto out;
440  cops = q->ops->cl_ops;
441  if (!cops)
442  goto errout;
443  if (cops->tcf_chain == NULL)
444  goto errout;
445  if (TC_H_MIN(tcm->tcm_parent)) {
446  cl = cops->get(q, tcm->tcm_parent);
447  if (cl == 0)
448  goto errout;
449  }
450  chain = cops->tcf_chain(q, cl);
451  if (chain == NULL)
452  goto errout;
453 
454  s_t = cb->args[0];
455 
456  for (tp = *chain, t = 0; tp; tp = tp->next, t++) {
457  if (t < s_t)
458  continue;
459  if (TC_H_MAJ(tcm->tcm_info) &&
460  TC_H_MAJ(tcm->tcm_info) != tp->prio)
461  continue;
462  if (TC_H_MIN(tcm->tcm_info) &&
463  TC_H_MIN(tcm->tcm_info) != tp->protocol)
464  continue;
465  if (t > s_t)
466  memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
467  if (cb->args[1] == 0) {
468  if (tcf_fill_node(skb, tp, 0, NETLINK_CB(cb->skb).portid,
469  cb->nlh->nlmsg_seq, NLM_F_MULTI,
470  RTM_NEWTFILTER) <= 0)
471  break;
472 
473  cb->args[1] = 1;
474  }
475  if (tp->ops->walk == NULL)
476  continue;
477  arg.w.fn = tcf_node_dump;
478  arg.skb = skb;
479  arg.cb = cb;
480  arg.w.stop = 0;
481  arg.w.skip = cb->args[1] - 1;
482  arg.w.count = 0;
483  tp->ops->walk(tp, &arg.w);
484  cb->args[1] = arg.w.count + 1;
485  if (arg.w.stop)
486  break;
487  }
488 
489  cb->args[0] = t;
490 
491 errout:
492  if (cl)
493  cops->put(q, cl);
494 out:
495  return skb->len;
496 }
497 
498 void tcf_exts_destroy(struct tcf_proto *tp, struct tcf_exts *exts)
499 {
500 #ifdef CONFIG_NET_CLS_ACT
501  if (exts->action) {
502  tcf_action_destroy(exts->action, TCA_ACT_UNBIND);
503  exts->action = NULL;
504  }
505 #endif
506 }
508 
509 int tcf_exts_validate(struct tcf_proto *tp, struct nlattr **tb,
510  struct nlattr *rate_tlv, struct tcf_exts *exts,
511  const struct tcf_ext_map *map)
512 {
513  memset(exts, 0, sizeof(*exts));
514 
515 #ifdef CONFIG_NET_CLS_ACT
516  {
517  struct tc_action *act;
518 
519  if (map->police && tb[map->police]) {
520  act = tcf_action_init_1(tb[map->police], rate_tlv,
521  "police", TCA_ACT_NOREPLACE,
522  TCA_ACT_BIND);
523  if (IS_ERR(act))
524  return PTR_ERR(act);
525 
526  act->type = TCA_OLD_COMPAT;
527  exts->action = act;
528  } else if (map->action && tb[map->action]) {
529  act = tcf_action_init(tb[map->action], rate_tlv, NULL,
531  if (IS_ERR(act))
532  return PTR_ERR(act);
533 
534  exts->action = act;
535  }
536  }
537 #else
538  if ((map->action && tb[map->action]) ||
539  (map->police && tb[map->police]))
540  return -EOPNOTSUPP;
541 #endif
542 
543  return 0;
544 }
546 
547 void tcf_exts_change(struct tcf_proto *tp, struct tcf_exts *dst,
548  struct tcf_exts *src)
549 {
550 #ifdef CONFIG_NET_CLS_ACT
551  if (src->action) {
552  struct tc_action *act;
553  tcf_tree_lock(tp);
554  act = dst->action;
555  dst->action = src->action;
556  tcf_tree_unlock(tp);
557  if (act)
559  }
560 #endif
561 }
563 
564 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts,
565  const struct tcf_ext_map *map)
566 {
567 #ifdef CONFIG_NET_CLS_ACT
568  if (map->action && exts->action) {
569  /*
570  * again for backward compatible mode - we want
571  * to work with both old and new modes of entering
572  * tc data even if iproute2 was newer - jhs
573  */
574  struct nlattr *nest;
575 
576  if (exts->action->type != TCA_OLD_COMPAT) {
577  nest = nla_nest_start(skb, map->action);
578  if (nest == NULL)
579  goto nla_put_failure;
580  if (tcf_action_dump(skb, exts->action, 0, 0) < 0)
581  goto nla_put_failure;
582  nla_nest_end(skb, nest);
583  } else if (map->police) {
584  nest = nla_nest_start(skb, map->police);
585  if (nest == NULL)
586  goto nla_put_failure;
587  if (tcf_action_dump_old(skb, exts->action, 0, 0) < 0)
588  goto nla_put_failure;
589  nla_nest_end(skb, nest);
590  }
591  }
592 #endif
593  return 0;
594 nla_put_failure: __attribute__ ((unused))
595  return -1;
596 }
598 
599 
600 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts,
601  const struct tcf_ext_map *map)
602 {
603 #ifdef CONFIG_NET_CLS_ACT
604  if (exts->action)
605  if (tcf_action_copy_stats(skb, exts->action, 1) < 0)
606  goto nla_put_failure;
607 #endif
608  return 0;
609 nla_put_failure: __attribute__ ((unused))
610  return -1;
611 }
613 
614 static int __init tc_filter_init(void)
615 {
616  rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_ctl_tfilter, NULL, NULL);
617  rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_ctl_tfilter, NULL, NULL);
618  rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_ctl_tfilter,
619  tc_dump_tfilter, NULL);
620 
621  return 0;
622 }
623 
624 subsys_initcall(tc_filter_init);