Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes...
[linux-2.6] / 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/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/kmod.h>
21 #include <linux/err.h>
22 #include <net/net_namespace.h>
23 #include <net/sock.h>
24 #include <net/sch_generic.h>
25 #include <net/act_api.h>
26 #include <net/netlink.h>
27
28 void tcf_hash_destroy(struct tcf_common *p, struct tcf_hashinfo *hinfo)
29 {
30         unsigned int h = tcf_hash(p->tcfc_index, hinfo->hmask);
31         struct tcf_common **p1p;
32
33         for (p1p = &hinfo->htab[h]; *p1p; p1p = &(*p1p)->tcfc_next) {
34                 if (*p1p == p) {
35                         write_lock_bh(hinfo->lock);
36                         *p1p = p->tcfc_next;
37                         write_unlock_bh(hinfo->lock);
38                         gen_kill_estimator(&p->tcfc_bstats,
39                                            &p->tcfc_rate_est);
40                         kfree(p);
41                         return;
42                 }
43         }
44         WARN_ON(1);
45 }
46 EXPORT_SYMBOL(tcf_hash_destroy);
47
48 int tcf_hash_release(struct tcf_common *p, int bind,
49                      struct tcf_hashinfo *hinfo)
50 {
51         int ret = 0;
52
53         if (p) {
54                 if (bind)
55                         p->tcfc_bindcnt--;
56
57                 p->tcfc_refcnt--;
58                 if (p->tcfc_bindcnt <= 0 && p->tcfc_refcnt <= 0) {
59                         tcf_hash_destroy(p, hinfo);
60                         ret = 1;
61                 }
62         }
63         return ret;
64 }
65 EXPORT_SYMBOL(tcf_hash_release);
66
67 static int tcf_dump_walker(struct sk_buff *skb, struct netlink_callback *cb,
68                            struct tc_action *a, struct tcf_hashinfo *hinfo)
69 {
70         struct tcf_common *p;
71         int err = 0, index = -1,i = 0, s_i = 0, n_i = 0;
72         struct nlattr *nest;
73
74         read_lock_bh(hinfo->lock);
75
76         s_i = cb->args[0];
77
78         for (i = 0; i < (hinfo->hmask + 1); i++) {
79                 p = hinfo->htab[tcf_hash(i, hinfo->hmask)];
80
81                 for (; p; p = p->tcfc_next) {
82                         index++;
83                         if (index < s_i)
84                                 continue;
85                         a->priv = p;
86                         a->order = n_i;
87
88                         nest = nla_nest_start(skb, a->order);
89                         if (nest == NULL)
90                                 goto nla_put_failure;
91                         err = tcf_action_dump_1(skb, a, 0, 0);
92                         if (err < 0) {
93                                 index--;
94                                 nlmsg_trim(skb, nest);
95                                 goto done;
96                         }
97                         nla_nest_end(skb, nest);
98                         n_i++;
99                         if (n_i >= TCA_ACT_MAX_PRIO)
100                                 goto done;
101                 }
102         }
103 done:
104         read_unlock_bh(hinfo->lock);
105         if (n_i)
106                 cb->args[0] += n_i;
107         return n_i;
108
109 nla_put_failure:
110         nla_nest_cancel(skb, nest);
111         goto done;
112 }
113
114 static int tcf_del_walker(struct sk_buff *skb, struct tc_action *a,
115                           struct tcf_hashinfo *hinfo)
116 {
117         struct tcf_common *p, *s_p;
118         struct nlattr *nest;
119         int i= 0, n_i = 0;
120
121         nest = nla_nest_start(skb, a->order);
122         if (nest == NULL)
123                 goto nla_put_failure;
124         NLA_PUT_STRING(skb, TCA_KIND, a->ops->kind);
125         for (i = 0; i < (hinfo->hmask + 1); i++) {
126                 p = hinfo->htab[tcf_hash(i, hinfo->hmask)];
127
128                 while (p != NULL) {
129                         s_p = p->tcfc_next;
130                         if (ACT_P_DELETED == tcf_hash_release(p, 0, hinfo))
131                                  module_put(a->ops->owner);
132                         n_i++;
133                         p = s_p;
134                 }
135         }
136         NLA_PUT_U32(skb, TCA_FCNT, n_i);
137         nla_nest_end(skb, nest);
138
139         return n_i;
140 nla_put_failure:
141         nla_nest_cancel(skb, nest);
142         return -EINVAL;
143 }
144
145 int tcf_generic_walker(struct sk_buff *skb, struct netlink_callback *cb,
146                        int type, struct tc_action *a)
147 {
148         struct tcf_hashinfo *hinfo = a->ops->hinfo;
149
150         if (type == RTM_DELACTION) {
151                 return tcf_del_walker(skb, a, hinfo);
152         } else if (type == RTM_GETACTION) {
153                 return tcf_dump_walker(skb, cb, a, hinfo);
154         } else {
155                 printk("tcf_generic_walker: unknown action %d\n", type);
156                 return -EINVAL;
157         }
158 }
159 EXPORT_SYMBOL(tcf_generic_walker);
160
161 struct tcf_common *tcf_hash_lookup(u32 index, struct tcf_hashinfo *hinfo)
162 {
163         struct tcf_common *p;
164
165         read_lock_bh(hinfo->lock);
166         for (p = hinfo->htab[tcf_hash(index, hinfo->hmask)]; p;
167              p = p->tcfc_next) {
168                 if (p->tcfc_index == index)
169                         break;
170         }
171         read_unlock_bh(hinfo->lock);
172
173         return p;
174 }
175 EXPORT_SYMBOL(tcf_hash_lookup);
176
177 u32 tcf_hash_new_index(u32 *idx_gen, struct tcf_hashinfo *hinfo)
178 {
179         u32 val = *idx_gen;
180
181         do {
182                 if (++val == 0)
183                         val = 1;
184         } while (tcf_hash_lookup(val, hinfo));
185
186         return (*idx_gen = val);
187 }
188 EXPORT_SYMBOL(tcf_hash_new_index);
189
190 int tcf_hash_search(struct tc_action *a, u32 index)
191 {
192         struct tcf_hashinfo *hinfo = a->ops->hinfo;
193         struct tcf_common *p = tcf_hash_lookup(index, hinfo);
194
195         if (p) {
196                 a->priv = p;
197                 return 1;
198         }
199         return 0;
200 }
201 EXPORT_SYMBOL(tcf_hash_search);
202
203 struct tcf_common *tcf_hash_check(u32 index, struct tc_action *a, int bind,
204                                   struct tcf_hashinfo *hinfo)
205 {
206         struct tcf_common *p = NULL;
207         if (index && (p = tcf_hash_lookup(index, hinfo)) != NULL) {
208                 if (bind)
209                         p->tcfc_bindcnt++;
210                 p->tcfc_refcnt++;
211                 a->priv = p;
212         }
213         return p;
214 }
215 EXPORT_SYMBOL(tcf_hash_check);
216
217 struct tcf_common *tcf_hash_create(u32 index, struct nlattr *est,
218                                    struct tc_action *a, int size, int bind,
219                                    u32 *idx_gen, struct tcf_hashinfo *hinfo)
220 {
221         struct tcf_common *p = kzalloc(size, GFP_KERNEL);
222
223         if (unlikely(!p))
224                 return ERR_PTR(-ENOMEM);
225         p->tcfc_refcnt = 1;
226         if (bind)
227                 p->tcfc_bindcnt = 1;
228
229         spin_lock_init(&p->tcfc_lock);
230         p->tcfc_index = index ? index : tcf_hash_new_index(idx_gen, hinfo);
231         p->tcfc_tm.install = jiffies;
232         p->tcfc_tm.lastuse = jiffies;
233         if (est) {
234                 int err = gen_new_estimator(&p->tcfc_bstats, &p->tcfc_rate_est,
235                                             &p->tcfc_lock, est);
236                 if (err) {
237                         kfree(p);
238                         return ERR_PTR(err);
239                 }
240         }
241
242         a->priv = (void *) p;
243         return p;
244 }
245 EXPORT_SYMBOL(tcf_hash_create);
246
247 void tcf_hash_insert(struct tcf_common *p, struct tcf_hashinfo *hinfo)
248 {
249         unsigned int h = tcf_hash(p->tcfc_index, hinfo->hmask);
250
251         write_lock_bh(hinfo->lock);
252         p->tcfc_next = hinfo->htab[h];
253         hinfo->htab[h] = p;
254         write_unlock_bh(hinfo->lock);
255 }
256 EXPORT_SYMBOL(tcf_hash_insert);
257
258 static struct tc_action_ops *act_base = NULL;
259 static DEFINE_RWLOCK(act_mod_lock);
260
261 int tcf_register_action(struct tc_action_ops *act)
262 {
263         struct tc_action_ops *a, **ap;
264
265         write_lock(&act_mod_lock);
266         for (ap = &act_base; (a = *ap) != NULL; ap = &a->next) {
267                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
268                         write_unlock(&act_mod_lock);
269                         return -EEXIST;
270                 }
271         }
272         act->next = NULL;
273         *ap = act;
274         write_unlock(&act_mod_lock);
275         return 0;
276 }
277 EXPORT_SYMBOL(tcf_register_action);
278
279 int tcf_unregister_action(struct tc_action_ops *act)
280 {
281         struct tc_action_ops *a, **ap;
282         int err = -ENOENT;
283
284         write_lock(&act_mod_lock);
285         for (ap = &act_base; (a = *ap) != NULL; ap = &a->next)
286                 if (a == act)
287                         break;
288         if (a) {
289                 *ap = a->next;
290                 a->next = NULL;
291                 err = 0;
292         }
293         write_unlock(&act_mod_lock);
294         return err;
295 }
296 EXPORT_SYMBOL(tcf_unregister_action);
297
298 /* lookup by name */
299 static struct tc_action_ops *tc_lookup_action_n(char *kind)
300 {
301         struct tc_action_ops *a = NULL;
302
303         if (kind) {
304                 read_lock(&act_mod_lock);
305                 for (a = act_base; a; a = a->next) {
306                         if (strcmp(kind, a->kind) == 0) {
307                                 if (!try_module_get(a->owner)) {
308                                         read_unlock(&act_mod_lock);
309                                         return NULL;
310                                 }
311                                 break;
312                         }
313                 }
314                 read_unlock(&act_mod_lock);
315         }
316         return a;
317 }
318
319 /* lookup by nlattr */
320 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
321 {
322         struct tc_action_ops *a = NULL;
323
324         if (kind) {
325                 read_lock(&act_mod_lock);
326                 for (a = act_base; a; a = a->next) {
327                         if (nla_strcmp(kind, a->kind) == 0) {
328                                 if (!try_module_get(a->owner)) {
329                                         read_unlock(&act_mod_lock);
330                                         return NULL;
331                                 }
332                                 break;
333                         }
334                 }
335                 read_unlock(&act_mod_lock);
336         }
337         return a;
338 }
339
340 #if 0
341 /* lookup by id */
342 static struct tc_action_ops *tc_lookup_action_id(u32 type)
343 {
344         struct tc_action_ops *a = NULL;
345
346         if (type) {
347                 read_lock(&act_mod_lock);
348                 for (a = act_base; a; a = a->next) {
349                         if (a->type == type) {
350                                 if (!try_module_get(a->owner)) {
351                                         read_unlock(&act_mod_lock);
352                                         return NULL;
353                                 }
354                                 break;
355                         }
356                 }
357                 read_unlock(&act_mod_lock);
358         }
359         return a;
360 }
361 #endif
362
363 int tcf_action_exec(struct sk_buff *skb, struct tc_action *act,
364                     struct tcf_result *res)
365 {
366         struct tc_action *a;
367         int ret = -1;
368
369         if (skb->tc_verd & TC_NCLS) {
370                 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
371                 ret = TC_ACT_OK;
372                 goto exec_done;
373         }
374         while ((a = act) != NULL) {
375 repeat:
376                 if (a->ops && a->ops->act) {
377                         ret = a->ops->act(skb, a, res);
378                         if (TC_MUNGED & skb->tc_verd) {
379                                 /* copied already, allow trampling */
380                                 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
381                                 skb->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
382                         }
383                         if (ret == TC_ACT_REPEAT)
384                                 goto repeat;    /* we need a ttl - JHS */
385                         if (ret != TC_ACT_PIPE)
386                                 goto exec_done;
387                 }
388                 act = a->next;
389         }
390 exec_done:
391         return ret;
392 }
393 EXPORT_SYMBOL(tcf_action_exec);
394
395 void tcf_action_destroy(struct tc_action *act, int bind)
396 {
397         struct tc_action *a;
398
399         for (a = act; a; a = act) {
400                 if (a->ops && a->ops->cleanup) {
401                         if (a->ops->cleanup(a, bind) == ACT_P_DELETED)
402                                 module_put(a->ops->owner);
403                         act = act->next;
404                         kfree(a);
405                 } else { /*FIXME: Remove later - catch insertion bugs*/
406                         printk("tcf_action_destroy: BUG? destroying NULL ops\n");
407                         act = act->next;
408                         kfree(a);
409                 }
410         }
411 }
412
413 int
414 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
415 {
416         int err = -EINVAL;
417
418         if (a->ops == NULL || a->ops->dump == NULL)
419                 return err;
420         return a->ops->dump(skb, a, bind, ref);
421 }
422
423 int
424 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
425 {
426         int err = -EINVAL;
427         unsigned char *b = skb_tail_pointer(skb);
428         struct nlattr *nest;
429
430         if (a->ops == NULL || a->ops->dump == NULL)
431                 return err;
432
433         NLA_PUT_STRING(skb, TCA_KIND, a->ops->kind);
434         if (tcf_action_copy_stats(skb, a, 0))
435                 goto nla_put_failure;
436         nest = nla_nest_start(skb, TCA_OPTIONS);
437         if (nest == NULL)
438                 goto nla_put_failure;
439         if ((err = tcf_action_dump_old(skb, a, bind, ref)) > 0) {
440                 nla_nest_end(skb, nest);
441                 return err;
442         }
443
444 nla_put_failure:
445         nlmsg_trim(skb, b);
446         return -1;
447 }
448 EXPORT_SYMBOL(tcf_action_dump_1);
449
450 int
451 tcf_action_dump(struct sk_buff *skb, struct tc_action *act, int bind, int ref)
452 {
453         struct tc_action *a;
454         int err = -EINVAL;
455         struct nlattr *nest;
456
457         while ((a = act) != NULL) {
458                 act = a->next;
459                 nest = nla_nest_start(skb, a->order);
460                 if (nest == NULL)
461                         goto nla_put_failure;
462                 err = tcf_action_dump_1(skb, a, bind, ref);
463                 if (err < 0)
464                         goto errout;
465                 nla_nest_end(skb, nest);
466         }
467
468         return 0;
469
470 nla_put_failure:
471         err = -EINVAL;
472 errout:
473         nla_nest_cancel(skb, nest);
474         return err;
475 }
476
477 struct tc_action *tcf_action_init_1(struct nlattr *nla, struct nlattr *est,
478                                     char *name, int ovr, int bind)
479 {
480         struct tc_action *a;
481         struct tc_action_ops *a_o;
482         char act_name[IFNAMSIZ];
483         struct nlattr *tb[TCA_ACT_MAX+1];
484         struct nlattr *kind;
485         int err;
486
487         if (name == NULL) {
488                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
489                 if (err < 0)
490                         goto err_out;
491                 err = -EINVAL;
492                 kind = tb[TCA_ACT_KIND];
493                 if (kind == NULL)
494                         goto err_out;
495                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ)
496                         goto err_out;
497         } else {
498                 err = -EINVAL;
499                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ)
500                         goto err_out;
501         }
502
503         a_o = tc_lookup_action_n(act_name);
504         if (a_o == NULL) {
505 #ifdef CONFIG_MODULES
506                 rtnl_unlock();
507                 request_module("act_%s", act_name);
508                 rtnl_lock();
509
510                 a_o = tc_lookup_action_n(act_name);
511
512                 /* We dropped the RTNL semaphore in order to
513                  * perform the module load.  So, even if we
514                  * succeeded in loading the module we have to
515                  * tell the caller to replay the request.  We
516                  * indicate this using -EAGAIN.
517                  */
518                 if (a_o != NULL) {
519                         err = -EAGAIN;
520                         goto err_mod;
521                 }
522 #endif
523                 err = -ENOENT;
524                 goto err_out;
525         }
526
527         err = -ENOMEM;
528         a = kzalloc(sizeof(*a), GFP_KERNEL);
529         if (a == NULL)
530                 goto err_mod;
531
532         /* backward compatibility for policer */
533         if (name == NULL)
534                 err = a_o->init(tb[TCA_ACT_OPTIONS], est, a, ovr, bind);
535         else
536                 err = a_o->init(nla, est, a, ovr, bind);
537         if (err < 0)
538                 goto err_free;
539
540         /* module count goes up only when brand new policy is created
541            if it exists and is only bound to in a_o->init() then
542            ACT_P_CREATED is not returned (a zero is).
543         */
544         if (err != ACT_P_CREATED)
545                 module_put(a_o->owner);
546         a->ops = a_o;
547
548         return a;
549
550 err_free:
551         kfree(a);
552 err_mod:
553         module_put(a_o->owner);
554 err_out:
555         return ERR_PTR(err);
556 }
557
558 struct tc_action *tcf_action_init(struct nlattr *nla, struct nlattr *est,
559                                   char *name, int ovr, int bind)
560 {
561         struct nlattr *tb[TCA_ACT_MAX_PRIO+1];
562         struct tc_action *head = NULL, *act, *act_prev = NULL;
563         int err;
564         int i;
565
566         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
567         if (err < 0)
568                 return ERR_PTR(err);
569
570         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
571                 act = tcf_action_init_1(tb[i], est, name, ovr, bind);
572                 if (IS_ERR(act))
573                         goto err;
574                 act->order = i;
575
576                 if (head == NULL)
577                         head = act;
578                 else
579                         act_prev->next = act;
580                 act_prev = act;
581         }
582         return head;
583
584 err:
585         if (head != NULL)
586                 tcf_action_destroy(head, bind);
587         return act;
588 }
589
590 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *a,
591                           int compat_mode)
592 {
593         int err = 0;
594         struct gnet_dump d;
595         struct tcf_act_hdr *h = a->priv;
596
597         if (h == NULL)
598                 goto errout;
599
600         /* compat_mode being true specifies a call that is supposed
601          * to add additional backward compatiblity statistic TLVs.
602          */
603         if (compat_mode) {
604                 if (a->type == TCA_OLD_COMPAT)
605                         err = gnet_stats_start_copy_compat(skb, 0,
606                                 TCA_STATS, TCA_XSTATS, &h->tcf_lock, &d);
607                 else
608                         return 0;
609         } else
610                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
611                                             &h->tcf_lock, &d);
612
613         if (err < 0)
614                 goto errout;
615
616         if (a->ops != NULL && a->ops->get_stats != NULL)
617                 if (a->ops->get_stats(skb, a) < 0)
618                         goto errout;
619
620         if (gnet_stats_copy_basic(&d, &h->tcf_bstats) < 0 ||
621             gnet_stats_copy_rate_est(&d, &h->tcf_rate_est) < 0 ||
622             gnet_stats_copy_queue(&d, &h->tcf_qstats) < 0)
623                 goto errout;
624
625         if (gnet_stats_finish_copy(&d) < 0)
626                 goto errout;
627
628         return 0;
629
630 errout:
631         return -1;
632 }
633
634 static int
635 tca_get_fill(struct sk_buff *skb, struct tc_action *a, u32 pid, u32 seq,
636              u16 flags, int event, int bind, int ref)
637 {
638         struct tcamsg *t;
639         struct nlmsghdr *nlh;
640         unsigned char *b = skb_tail_pointer(skb);
641         struct nlattr *nest;
642
643         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*t), flags);
644
645         t = NLMSG_DATA(nlh);
646         t->tca_family = AF_UNSPEC;
647         t->tca__pad1 = 0;
648         t->tca__pad2 = 0;
649
650         nest = nla_nest_start(skb, TCA_ACT_TAB);
651         if (nest == NULL)
652                 goto nla_put_failure;
653
654         if (tcf_action_dump(skb, a, bind, ref) < 0)
655                 goto nla_put_failure;
656
657         nla_nest_end(skb, nest);
658
659         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
660         return skb->len;
661
662 nla_put_failure:
663 nlmsg_failure:
664         nlmsg_trim(skb, b);
665         return -1;
666 }
667
668 static int
669 act_get_notify(u32 pid, struct nlmsghdr *n, struct tc_action *a, int event)
670 {
671         struct sk_buff *skb;
672
673         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
674         if (!skb)
675                 return -ENOBUFS;
676         if (tca_get_fill(skb, a, pid, n->nlmsg_seq, 0, event, 0, 0) <= 0) {
677                 kfree_skb(skb);
678                 return -EINVAL;
679         }
680
681         return rtnl_unicast(skb, &init_net, pid);
682 }
683
684 static struct tc_action *
685 tcf_action_get_1(struct nlattr *nla, struct nlmsghdr *n, u32 pid)
686 {
687         struct nlattr *tb[TCA_ACT_MAX+1];
688         struct tc_action *a;
689         int index;
690         int err;
691
692         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
693         if (err < 0)
694                 goto err_out;
695
696         err = -EINVAL;
697         if (tb[TCA_ACT_INDEX] == NULL ||
698             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index))
699                 goto err_out;
700         index = nla_get_u32(tb[TCA_ACT_INDEX]);
701
702         err = -ENOMEM;
703         a = kzalloc(sizeof(struct tc_action), GFP_KERNEL);
704         if (a == NULL)
705                 goto err_out;
706
707         err = -EINVAL;
708         a->ops = tc_lookup_action(tb[TCA_ACT_KIND]);
709         if (a->ops == NULL)
710                 goto err_free;
711         if (a->ops->lookup == NULL)
712                 goto err_mod;
713         err = -ENOENT;
714         if (a->ops->lookup(a, index) == 0)
715                 goto err_mod;
716
717         module_put(a->ops->owner);
718         return a;
719
720 err_mod:
721         module_put(a->ops->owner);
722 err_free:
723         kfree(a);
724 err_out:
725         return ERR_PTR(err);
726 }
727
728 static void cleanup_a(struct tc_action *act)
729 {
730         struct tc_action *a;
731
732         for (a = act; a; a = act) {
733                 act = a->next;
734                 kfree(a);
735         }
736 }
737
738 static struct tc_action *create_a(int i)
739 {
740         struct tc_action *act;
741
742         act = kzalloc(sizeof(*act), GFP_KERNEL);
743         if (act == NULL) {
744                 printk("create_a: failed to alloc!\n");
745                 return NULL;
746         }
747         act->order = i;
748         return act;
749 }
750
751 static int tca_action_flush(struct nlattr *nla, struct nlmsghdr *n, u32 pid)
752 {
753         struct sk_buff *skb;
754         unsigned char *b;
755         struct nlmsghdr *nlh;
756         struct tcamsg *t;
757         struct netlink_callback dcb;
758         struct nlattr *nest;
759         struct nlattr *tb[TCA_ACT_MAX+1];
760         struct nlattr *kind;
761         struct tc_action *a = create_a(0);
762         int err = -ENOMEM;
763
764         if (a == NULL) {
765                 printk("tca_action_flush: couldnt create tc_action\n");
766                 return err;
767         }
768
769         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
770         if (!skb) {
771                 printk("tca_action_flush: failed skb alloc\n");
772                 kfree(a);
773                 return err;
774         }
775
776         b = skb_tail_pointer(skb);
777
778         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL);
779         if (err < 0)
780                 goto err_out;
781
782         err = -EINVAL;
783         kind = tb[TCA_ACT_KIND];
784         a->ops = tc_lookup_action(kind);
785         if (a->ops == NULL)
786                 goto err_out;
787
788         nlh = NLMSG_PUT(skb, pid, n->nlmsg_seq, RTM_DELACTION, sizeof(*t));
789         t = NLMSG_DATA(nlh);
790         t->tca_family = AF_UNSPEC;
791         t->tca__pad1 = 0;
792         t->tca__pad2 = 0;
793
794         nest = nla_nest_start(skb, TCA_ACT_TAB);
795         if (nest == NULL)
796                 goto nla_put_failure;
797
798         err = a->ops->walk(skb, &dcb, RTM_DELACTION, a);
799         if (err < 0)
800                 goto nla_put_failure;
801         if (err == 0)
802                 goto noflush_out;
803
804         nla_nest_end(skb, nest);
805
806         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
807         nlh->nlmsg_flags |= NLM_F_ROOT;
808         module_put(a->ops->owner);
809         kfree(a);
810         err = rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
811         if (err > 0)
812                 return 0;
813
814         return err;
815
816 nla_put_failure:
817 nlmsg_failure:
818         module_put(a->ops->owner);
819 err_out:
820 noflush_out:
821         kfree_skb(skb);
822         kfree(a);
823         return err;
824 }
825
826 static int
827 tca_action_gd(struct nlattr *nla, struct nlmsghdr *n, u32 pid, int event)
828 {
829         int i, ret;
830         struct nlattr *tb[TCA_ACT_MAX_PRIO+1];
831         struct tc_action *head = NULL, *act, *act_prev = NULL;
832
833         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL);
834         if (ret < 0)
835                 return ret;
836
837         if (event == RTM_DELACTION && n->nlmsg_flags&NLM_F_ROOT) {
838                 if (tb[1] != NULL)
839                         return tca_action_flush(tb[1], n, pid);
840                 else
841                         return -EINVAL;
842         }
843
844         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
845                 act = tcf_action_get_1(tb[i], n, pid);
846                 if (IS_ERR(act)) {
847                         ret = PTR_ERR(act);
848                         goto err;
849                 }
850                 act->order = i;
851
852                 if (head == NULL)
853                         head = act;
854                 else
855                         act_prev->next = act;
856                 act_prev = act;
857         }
858
859         if (event == RTM_GETACTION)
860                 ret = act_get_notify(pid, n, head, event);
861         else { /* delete */
862                 struct sk_buff *skb;
863
864                 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
865                 if (!skb) {
866                         ret = -ENOBUFS;
867                         goto err;
868                 }
869
870                 if (tca_get_fill(skb, head, pid, n->nlmsg_seq, 0, event,
871                                  0, 1) <= 0) {
872                         kfree_skb(skb);
873                         ret = -EINVAL;
874                         goto err;
875                 }
876
877                 /* now do the delete */
878                 tcf_action_destroy(head, 0);
879                 ret = rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC,
880                                      n->nlmsg_flags&NLM_F_ECHO);
881                 if (ret > 0)
882                         return 0;
883                 return ret;
884         }
885 err:
886         cleanup_a(head);
887         return ret;
888 }
889
890 static int tcf_add_notify(struct tc_action *a, u32 pid, u32 seq, int event,
891                           u16 flags)
892 {
893         struct tcamsg *t;
894         struct nlmsghdr *nlh;
895         struct sk_buff *skb;
896         struct nlattr *nest;
897         unsigned char *b;
898         int err = 0;
899
900         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
901         if (!skb)
902                 return -ENOBUFS;
903
904         b = skb_tail_pointer(skb);
905
906         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*t), flags);
907         t = NLMSG_DATA(nlh);
908         t->tca_family = AF_UNSPEC;
909         t->tca__pad1 = 0;
910         t->tca__pad2 = 0;
911
912         nest = nla_nest_start(skb, TCA_ACT_TAB);
913         if (nest == NULL)
914                 goto nla_put_failure;
915
916         if (tcf_action_dump(skb, a, 0, 0) < 0)
917                 goto nla_put_failure;
918
919         nla_nest_end(skb, nest);
920
921         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
922         NETLINK_CB(skb).dst_group = RTNLGRP_TC;
923
924         err = rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, flags&NLM_F_ECHO);
925         if (err > 0)
926                 err = 0;
927         return err;
928
929 nla_put_failure:
930 nlmsg_failure:
931         kfree_skb(skb);
932         return -1;
933 }
934
935
936 static int
937 tcf_action_add(struct nlattr *nla, struct nlmsghdr *n, u32 pid, int ovr)
938 {
939         int ret = 0;
940         struct tc_action *act;
941         struct tc_action *a;
942         u32 seq = n->nlmsg_seq;
943
944         act = tcf_action_init(nla, NULL, NULL, ovr, 0);
945         if (act == NULL)
946                 goto done;
947         if (IS_ERR(act)) {
948                 ret = PTR_ERR(act);
949                 goto done;
950         }
951
952         /* dump then free all the actions after update; inserted policy
953          * stays intact
954          * */
955         ret = tcf_add_notify(act, pid, seq, RTM_NEWACTION, n->nlmsg_flags);
956         for (a = act; a; a = act) {
957                 act = a->next;
958                 kfree(a);
959         }
960 done:
961         return ret;
962 }
963
964 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
965 {
966         struct net *net = sock_net(skb->sk);
967         struct nlattr *tca[TCA_ACT_MAX + 1];
968         u32 pid = skb ? NETLINK_CB(skb).pid : 0;
969         int ret = 0, ovr = 0;
970
971         if (net != &init_net)
972                 return -EINVAL;
973
974         ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ACT_MAX, NULL);
975         if (ret < 0)
976                 return ret;
977
978         if (tca[TCA_ACT_TAB] == NULL) {
979                 printk("tc_ctl_action: received NO action attribs\n");
980                 return -EINVAL;
981         }
982
983         /* n->nlmsg_flags&NLM_F_CREATE
984          * */
985         switch (n->nlmsg_type) {
986         case RTM_NEWACTION:
987                 /* we are going to assume all other flags
988                  * imply create only if it doesnt exist
989                  * Note that CREATE | EXCL implies that
990                  * but since we want avoid ambiguity (eg when flags
991                  * is zero) then just set this
992                  */
993                 if (n->nlmsg_flags&NLM_F_REPLACE)
994                         ovr = 1;
995 replay:
996                 ret = tcf_action_add(tca[TCA_ACT_TAB], n, pid, ovr);
997                 if (ret == -EAGAIN)
998                         goto replay;
999                 break;
1000         case RTM_DELACTION:
1001                 ret = tca_action_gd(tca[TCA_ACT_TAB], n, pid, RTM_DELACTION);
1002                 break;
1003         case RTM_GETACTION:
1004                 ret = tca_action_gd(tca[TCA_ACT_TAB], n, pid, RTM_GETACTION);
1005                 break;
1006         default:
1007                 BUG();
1008         }
1009
1010         return ret;
1011 }
1012
1013 static struct nlattr *
1014 find_dump_kind(struct nlmsghdr *n)
1015 {
1016         struct nlattr *tb1, *tb2[TCA_ACT_MAX+1];
1017         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1018         struct nlattr *nla[TCAA_MAX + 1];
1019         struct nlattr *kind;
1020
1021         if (nlmsg_parse(n, sizeof(struct tcamsg), nla, TCAA_MAX, NULL) < 0)
1022                 return NULL;
1023         tb1 = nla[TCA_ACT_TAB];
1024         if (tb1 == NULL)
1025                 return NULL;
1026
1027         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1028                       NLMSG_ALIGN(nla_len(tb1)), NULL) < 0)
1029                 return NULL;
1030
1031         if (tb[1] == NULL)
1032                 return NULL;
1033         if (nla_parse(tb2, TCA_ACT_MAX, nla_data(tb[1]),
1034                       nla_len(tb[1]), NULL) < 0)
1035                 return NULL;
1036         kind = tb2[TCA_ACT_KIND];
1037
1038         return kind;
1039 }
1040
1041 static int
1042 tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1043 {
1044         struct net *net = sock_net(skb->sk);
1045         struct nlmsghdr *nlh;
1046         unsigned char *b = skb_tail_pointer(skb);
1047         struct nlattr *nest;
1048         struct tc_action_ops *a_o;
1049         struct tc_action a;
1050         int ret = 0;
1051         struct tcamsg *t = (struct tcamsg *) NLMSG_DATA(cb->nlh);
1052         struct nlattr *kind = find_dump_kind(cb->nlh);
1053
1054         if (net != &init_net)
1055                 return 0;
1056
1057         if (kind == NULL) {
1058                 printk("tc_dump_action: action bad kind\n");
1059                 return 0;
1060         }
1061
1062         a_o = tc_lookup_action(kind);
1063         if (a_o == NULL) {
1064                 return 0;
1065         }
1066
1067         memset(&a, 0, sizeof(struct tc_action));
1068         a.ops = a_o;
1069
1070         if (a_o->walk == NULL) {
1071                 printk("tc_dump_action: %s !capable of dumping table\n", a_o->kind);
1072                 goto nla_put_failure;
1073         }
1074
1075         nlh = NLMSG_PUT(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq,
1076                         cb->nlh->nlmsg_type, sizeof(*t));
1077         t = NLMSG_DATA(nlh);
1078         t->tca_family = AF_UNSPEC;
1079         t->tca__pad1 = 0;
1080         t->tca__pad2 = 0;
1081
1082         nest = nla_nest_start(skb, TCA_ACT_TAB);
1083         if (nest == NULL)
1084                 goto nla_put_failure;
1085
1086         ret = a_o->walk(skb, cb, RTM_GETACTION, &a);
1087         if (ret < 0)
1088                 goto nla_put_failure;
1089
1090         if (ret > 0) {
1091                 nla_nest_end(skb, nest);
1092                 ret = skb->len;
1093         } else
1094                 nla_nest_cancel(skb, nest);
1095
1096         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1097         if (NETLINK_CB(cb->skb).pid && ret)
1098                 nlh->nlmsg_flags |= NLM_F_MULTI;
1099         module_put(a_o->owner);
1100         return skb->len;
1101
1102 nla_put_failure:
1103 nlmsg_failure:
1104         module_put(a_o->owner);
1105         nlmsg_trim(skb, b);
1106         return skb->len;
1107 }
1108
1109 static int __init tc_action_init(void)
1110 {
1111         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL);
1112         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL);
1113         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action);
1114
1115         return 0;
1116 }
1117
1118 subsys_initcall(tc_action_init);