00bf7d2b0bdd1feb363ae157ea41b4ebd9ccb156
[linux-2.6-block.git] / net / sched / act_api.c
1 /*
2  * net/sched/act_api.c  Packet action API.
3  *
4  *              This program is free software; you can redistribute it and/or
5  *              modify it under the terms of the GNU General Public License
6  *              as published by the Free Software Foundation; either version
7  *              2 of the License, or (at your option) any later version.
8  *
9  * Author:      Jamal Hadi Salim
10  *
11  *
12  */
13
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/slab.h>
19 #include <linux/skbuff.h>
20 #include <linux/init.h>
21 #include <linux/kmod.h>
22 #include <linux/err.h>
23 #include <linux/module.h>
24 #include <linux/rhashtable.h>
25 #include <linux/list.h>
26 #include <net/net_namespace.h>
27 #include <net/sock.h>
28 #include <net/sch_generic.h>
29 #include <net/pkt_cls.h>
30 #include <net/act_api.h>
31 #include <net/netlink.h>
32
33 static int tcf_action_goto_chain_init(struct tc_action *a, struct tcf_proto *tp)
34 {
35         u32 chain_index = a->tcfa_action & TC_ACT_EXT_VAL_MASK;
36
37         if (!tp)
38                 return -EINVAL;
39         a->goto_chain = tcf_chain_get_by_act(tp->chain->block, chain_index);
40         if (!a->goto_chain)
41                 return -ENOMEM;
42         return 0;
43 }
44
45 static void tcf_action_goto_chain_fini(struct tc_action *a)
46 {
47         tcf_chain_put_by_act(a->goto_chain);
48 }
49
50 static void tcf_action_goto_chain_exec(const struct tc_action *a,
51                                        struct tcf_result *res)
52 {
53         const struct tcf_chain *chain = a->goto_chain;
54
55         res->goto_tp = rcu_dereference_bh(chain->filter_chain);
56 }
57
58 static void tcf_free_cookie_rcu(struct rcu_head *p)
59 {
60         struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
61
62         kfree(cookie->data);
63         kfree(cookie);
64 }
65
66 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
67                                   struct tc_cookie *new_cookie)
68 {
69         struct tc_cookie *old;
70
71         old = xchg((__force struct tc_cookie **)old_cookie, new_cookie);
72         if (old)
73                 call_rcu(&old->rcu, tcf_free_cookie_rcu);
74 }
75
76 /* XXX: For standalone actions, we don't need a RCU grace period either, because
77  * actions are always connected to filters and filters are already destroyed in
78  * RCU callbacks, so after a RCU grace period actions are already disconnected
79  * from filters. Readers later can not find us.
80  */
81 static void free_tcf(struct tc_action *p)
82 {
83         free_percpu(p->cpu_bstats);
84         free_percpu(p->cpu_qstats);
85
86         tcf_set_action_cookie(&p->act_cookie, NULL);
87         if (p->goto_chain)
88                 tcf_action_goto_chain_fini(p);
89
90         kfree(p);
91 }
92
93 static void tcf_action_cleanup(struct tc_action *p)
94 {
95         if (p->ops->cleanup)
96                 p->ops->cleanup(p);
97
98         gen_kill_estimator(&p->tcfa_rate_est);
99         free_tcf(p);
100 }
101
102 static int __tcf_action_put(struct tc_action *p, bool bind)
103 {
104         struct tcf_idrinfo *idrinfo = p->idrinfo;
105
106         if (refcount_dec_and_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
107                 if (bind)
108                         atomic_dec(&p->tcfa_bindcnt);
109                 idr_remove(&idrinfo->action_idr, p->tcfa_index);
110                 spin_unlock(&idrinfo->lock);
111
112                 tcf_action_cleanup(p);
113                 return 1;
114         }
115
116         if (bind)
117                 atomic_dec(&p->tcfa_bindcnt);
118
119         return 0;
120 }
121
122 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
123 {
124         int ret = 0;
125
126         /* Release with strict==1 and bind==0 is only called through act API
127          * interface (classifiers always bind). Only case when action with
128          * positive reference count and zero bind count can exist is when it was
129          * also created with act API (unbinding last classifier will destroy the
130          * action if it was created by classifier). So only case when bind count
131          * can be changed after initial check is when unbound action is
132          * destroyed by act API while classifier binds to action with same id
133          * concurrently. This result either creation of new action(same behavior
134          * as before), or reusing existing action if concurrent process
135          * increments reference count before action is deleted. Both scenarios
136          * are acceptable.
137          */
138         if (p) {
139                 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
140                         return -EPERM;
141
142                 if (__tcf_action_put(p, bind))
143                         ret = ACT_P_DELETED;
144         }
145
146         return ret;
147 }
148 EXPORT_SYMBOL(__tcf_idr_release);
149
150 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
151 {
152         struct tc_cookie *act_cookie;
153         u32 cookie_len = 0;
154
155         rcu_read_lock();
156         act_cookie = rcu_dereference(act->act_cookie);
157
158         if (act_cookie)
159                 cookie_len = nla_total_size(act_cookie->len);
160         rcu_read_unlock();
161
162         return  nla_total_size(0) /* action number nested */
163                 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
164                 + cookie_len /* TCA_ACT_COOKIE */
165                 + nla_total_size(0) /* TCA_ACT_STATS nested */
166                 /* TCA_STATS_BASIC */
167                 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
168                 /* TCA_STATS_QUEUE */
169                 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
170                 + nla_total_size(0) /* TCA_OPTIONS nested */
171                 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
172 }
173
174 static size_t tcf_action_full_attrs_size(size_t sz)
175 {
176         return NLMSG_HDRLEN                     /* struct nlmsghdr */
177                 + sizeof(struct tcamsg)
178                 + nla_total_size(0)             /* TCA_ACT_TAB nested */
179                 + sz;
180 }
181
182 static size_t tcf_action_fill_size(const struct tc_action *act)
183 {
184         size_t sz = tcf_action_shared_attrs_size(act);
185
186         if (act->ops->get_fill_size)
187                 return act->ops->get_fill_size(act) + sz;
188         return sz;
189 }
190
191 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
192                            struct netlink_callback *cb)
193 {
194         int err = 0, index = -1, s_i = 0, n_i = 0;
195         u32 act_flags = cb->args[2];
196         unsigned long jiffy_since = cb->args[3];
197         struct nlattr *nest;
198         struct idr *idr = &idrinfo->action_idr;
199         struct tc_action *p;
200         unsigned long id = 1;
201
202         spin_lock(&idrinfo->lock);
203
204         s_i = cb->args[0];
205
206         idr_for_each_entry_ul(idr, p, id) {
207                 index++;
208                 if (index < s_i)
209                         continue;
210
211                 if (jiffy_since &&
212                     time_after(jiffy_since,
213                                (unsigned long)p->tcfa_tm.lastuse))
214                         continue;
215
216                 nest = nla_nest_start(skb, n_i);
217                 if (!nest) {
218                         index--;
219                         goto nla_put_failure;
220                 }
221                 err = tcf_action_dump_1(skb, p, 0, 0);
222                 if (err < 0) {
223                         index--;
224                         nlmsg_trim(skb, nest);
225                         goto done;
226                 }
227                 nla_nest_end(skb, nest);
228                 n_i++;
229                 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
230                     n_i >= TCA_ACT_MAX_PRIO)
231                         goto done;
232         }
233 done:
234         if (index >= 0)
235                 cb->args[0] = index + 1;
236
237         spin_unlock(&idrinfo->lock);
238         if (n_i) {
239                 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
240                         cb->args[1] = n_i;
241         }
242         return n_i;
243
244 nla_put_failure:
245         nla_nest_cancel(skb, nest);
246         goto done;
247 }
248
249 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
250                           const struct tc_action_ops *ops)
251 {
252         struct nlattr *nest;
253         int n_i = 0;
254         int ret = -EINVAL;
255         struct idr *idr = &idrinfo->action_idr;
256         struct tc_action *p;
257         unsigned long id = 1;
258
259         nest = nla_nest_start(skb, 0);
260         if (nest == NULL)
261                 goto nla_put_failure;
262         if (nla_put_string(skb, TCA_KIND, ops->kind))
263                 goto nla_put_failure;
264
265         idr_for_each_entry_ul(idr, p, id) {
266                 ret = __tcf_idr_release(p, false, true);
267                 if (ret == ACT_P_DELETED) {
268                         module_put(ops->owner);
269                         n_i++;
270                 } else if (ret < 0) {
271                         goto nla_put_failure;
272                 }
273         }
274         if (nla_put_u32(skb, TCA_FCNT, n_i))
275                 goto nla_put_failure;
276         nla_nest_end(skb, nest);
277
278         return n_i;
279 nla_put_failure:
280         nla_nest_cancel(skb, nest);
281         return ret;
282 }
283
284 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
285                        struct netlink_callback *cb, int type,
286                        const struct tc_action_ops *ops,
287                        struct netlink_ext_ack *extack)
288 {
289         struct tcf_idrinfo *idrinfo = tn->idrinfo;
290
291         if (type == RTM_DELACTION) {
292                 return tcf_del_walker(idrinfo, skb, ops);
293         } else if (type == RTM_GETACTION) {
294                 return tcf_dump_walker(idrinfo, skb, cb);
295         } else {
296                 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
297                 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
298                 return -EINVAL;
299         }
300 }
301 EXPORT_SYMBOL(tcf_generic_walker);
302
303 static bool __tcf_idr_check(struct tc_action_net *tn, u32 index,
304                             struct tc_action **a, int bind)
305 {
306         struct tcf_idrinfo *idrinfo = tn->idrinfo;
307         struct tc_action *p;
308
309         spin_lock(&idrinfo->lock);
310         p = idr_find(&idrinfo->action_idr, index);
311         if (IS_ERR(p)) {
312                 p = NULL;
313         } else if (p) {
314                 refcount_inc(&p->tcfa_refcnt);
315                 if (bind)
316                         atomic_inc(&p->tcfa_bindcnt);
317         }
318         spin_unlock(&idrinfo->lock);
319
320         if (p) {
321                 *a = p;
322                 return true;
323         }
324         return false;
325 }
326
327 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
328 {
329         return __tcf_idr_check(tn, index, a, 0);
330 }
331 EXPORT_SYMBOL(tcf_idr_search);
332
333 bool tcf_idr_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
334                    int bind)
335 {
336         return __tcf_idr_check(tn, index, a, bind);
337 }
338 EXPORT_SYMBOL(tcf_idr_check);
339
340 static int tcf_idr_delete_index(struct tcf_idrinfo *idrinfo, u32 index)
341 {
342         struct tc_action *p;
343         int ret = 0;
344
345         spin_lock(&idrinfo->lock);
346         p = idr_find(&idrinfo->action_idr, index);
347         if (!p) {
348                 spin_unlock(&idrinfo->lock);
349                 return -ENOENT;
350         }
351
352         if (!atomic_read(&p->tcfa_bindcnt)) {
353                 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
354                         struct module *owner = p->ops->owner;
355
356                         WARN_ON(p != idr_remove(&idrinfo->action_idr,
357                                                 p->tcfa_index));
358                         spin_unlock(&idrinfo->lock);
359
360                         tcf_action_cleanup(p);
361                         module_put(owner);
362                         return 0;
363                 }
364                 ret = 0;
365         } else {
366                 ret = -EPERM;
367         }
368
369         spin_unlock(&idrinfo->lock);
370         return ret;
371 }
372
373 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
374                    struct tc_action **a, const struct tc_action_ops *ops,
375                    int bind, bool cpustats)
376 {
377         struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
378         struct tcf_idrinfo *idrinfo = tn->idrinfo;
379         int err = -ENOMEM;
380
381         if (unlikely(!p))
382                 return -ENOMEM;
383         refcount_set(&p->tcfa_refcnt, 1);
384         if (bind)
385                 atomic_set(&p->tcfa_bindcnt, 1);
386
387         if (cpustats) {
388                 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
389                 if (!p->cpu_bstats)
390                         goto err1;
391                 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
392                 if (!p->cpu_qstats)
393                         goto err2;
394         }
395         spin_lock_init(&p->tcfa_lock);
396         p->tcfa_index = index;
397         p->tcfa_tm.install = jiffies;
398         p->tcfa_tm.lastuse = jiffies;
399         p->tcfa_tm.firstuse = 0;
400         if (est) {
401                 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
402                                         &p->tcfa_rate_est,
403                                         &p->tcfa_lock, NULL, est);
404                 if (err)
405                         goto err3;
406         }
407
408         p->idrinfo = idrinfo;
409         p->ops = ops;
410         INIT_LIST_HEAD(&p->list);
411         *a = p;
412         return 0;
413 err3:
414         free_percpu(p->cpu_qstats);
415 err2:
416         free_percpu(p->cpu_bstats);
417 err1:
418         kfree(p);
419         return err;
420 }
421 EXPORT_SYMBOL(tcf_idr_create);
422
423 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
424 {
425         struct tcf_idrinfo *idrinfo = tn->idrinfo;
426
427         spin_lock(&idrinfo->lock);
428         /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
429         WARN_ON(!IS_ERR(idr_replace(&idrinfo->action_idr, a, a->tcfa_index)));
430         spin_unlock(&idrinfo->lock);
431 }
432 EXPORT_SYMBOL(tcf_idr_insert);
433
434 /* Cleanup idr index that was allocated but not initialized. */
435
436 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
437 {
438         struct tcf_idrinfo *idrinfo = tn->idrinfo;
439
440         spin_lock(&idrinfo->lock);
441         /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
442         WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
443         spin_unlock(&idrinfo->lock);
444 }
445 EXPORT_SYMBOL(tcf_idr_cleanup);
446
447 /* Check if action with specified index exists. If actions is found, increments
448  * its reference and bind counters, and return 1. Otherwise insert temporary
449  * error pointer (to prevent concurrent users from inserting actions with same
450  * index) and return 0.
451  */
452
453 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
454                         struct tc_action **a, int bind)
455 {
456         struct tcf_idrinfo *idrinfo = tn->idrinfo;
457         struct tc_action *p;
458         int ret;
459
460 again:
461         spin_lock(&idrinfo->lock);
462         if (*index) {
463                 p = idr_find(&idrinfo->action_idr, *index);
464                 if (IS_ERR(p)) {
465                         /* This means that another process allocated
466                          * index but did not assign the pointer yet.
467                          */
468                         spin_unlock(&idrinfo->lock);
469                         goto again;
470                 }
471
472                 if (p) {
473                         refcount_inc(&p->tcfa_refcnt);
474                         if (bind)
475                                 atomic_inc(&p->tcfa_bindcnt);
476                         *a = p;
477                         ret = 1;
478                 } else {
479                         *a = NULL;
480                         ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
481                                             *index, GFP_ATOMIC);
482                         if (!ret)
483                                 idr_replace(&idrinfo->action_idr,
484                                             ERR_PTR(-EBUSY), *index);
485                 }
486         } else {
487                 *index = 1;
488                 *a = NULL;
489                 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
490                                     UINT_MAX, GFP_ATOMIC);
491                 if (!ret)
492                         idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
493                                     *index);
494         }
495         spin_unlock(&idrinfo->lock);
496         return ret;
497 }
498 EXPORT_SYMBOL(tcf_idr_check_alloc);
499
500 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
501                          struct tcf_idrinfo *idrinfo)
502 {
503         struct idr *idr = &idrinfo->action_idr;
504         struct tc_action *p;
505         int ret;
506         unsigned long id = 1;
507
508         idr_for_each_entry_ul(idr, p, id) {
509                 ret = __tcf_idr_release(p, false, true);
510                 if (ret == ACT_P_DELETED)
511                         module_put(ops->owner);
512                 else if (ret < 0)
513                         return;
514         }
515         idr_destroy(&idrinfo->action_idr);
516 }
517 EXPORT_SYMBOL(tcf_idrinfo_destroy);
518
519 static LIST_HEAD(act_base);
520 static DEFINE_RWLOCK(act_mod_lock);
521
522 int tcf_register_action(struct tc_action_ops *act,
523                         struct pernet_operations *ops)
524 {
525         struct tc_action_ops *a;
526         int ret;
527
528         if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
529                 return -EINVAL;
530
531         /* We have to register pernet ops before making the action ops visible,
532          * otherwise tcf_action_init_1() could get a partially initialized
533          * netns.
534          */
535         ret = register_pernet_subsys(ops);
536         if (ret)
537                 return ret;
538
539         write_lock(&act_mod_lock);
540         list_for_each_entry(a, &act_base, head) {
541                 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
542                         write_unlock(&act_mod_lock);
543                         unregister_pernet_subsys(ops);
544                         return -EEXIST;
545                 }
546         }
547         list_add_tail(&act->head, &act_base);
548         write_unlock(&act_mod_lock);
549
550         return 0;
551 }
552 EXPORT_SYMBOL(tcf_register_action);
553
554 int tcf_unregister_action(struct tc_action_ops *act,
555                           struct pernet_operations *ops)
556 {
557         struct tc_action_ops *a;
558         int err = -ENOENT;
559
560         write_lock(&act_mod_lock);
561         list_for_each_entry(a, &act_base, head) {
562                 if (a == act) {
563                         list_del(&act->head);
564                         err = 0;
565                         break;
566                 }
567         }
568         write_unlock(&act_mod_lock);
569         if (!err)
570                 unregister_pernet_subsys(ops);
571         return err;
572 }
573 EXPORT_SYMBOL(tcf_unregister_action);
574
575 /* lookup by name */
576 static struct tc_action_ops *tc_lookup_action_n(char *kind)
577 {
578         struct tc_action_ops *a, *res = NULL;
579
580         if (kind) {
581                 read_lock(&act_mod_lock);
582                 list_for_each_entry(a, &act_base, head) {
583                         if (strcmp(kind, a->kind) == 0) {
584                                 if (try_module_get(a->owner))
585                                         res = a;
586                                 break;
587                         }
588                 }
589                 read_unlock(&act_mod_lock);
590         }
591         return res;
592 }
593
594 /* lookup by nlattr */
595 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
596 {
597         struct tc_action_ops *a, *res = NULL;
598
599         if (kind) {
600                 read_lock(&act_mod_lock);
601                 list_for_each_entry(a, &act_base, head) {
602                         if (nla_strcmp(kind, a->kind) == 0) {
603                                 if (try_module_get(a->owner))
604                                         res = a;
605                                 break;
606                         }
607                 }
608                 read_unlock(&act_mod_lock);
609         }
610         return res;
611 }
612
613 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
614 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
615 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
616                     int nr_actions, struct tcf_result *res)
617 {
618         u32 jmp_prgcnt = 0;
619         u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
620         int i;
621         int ret = TC_ACT_OK;
622
623         if (skb_skip_tc_classify(skb))
624                 return TC_ACT_OK;
625
626 restart_act_graph:
627         for (i = 0; i < nr_actions; i++) {
628                 const struct tc_action *a = actions[i];
629
630                 if (jmp_prgcnt > 0) {
631                         jmp_prgcnt -= 1;
632                         continue;
633                 }
634 repeat:
635                 ret = a->ops->act(skb, a, res);
636                 if (ret == TC_ACT_REPEAT)
637                         goto repeat;    /* we need a ttl - JHS */
638
639                 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
640                         jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
641                         if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
642                                 /* faulty opcode, stop pipeline */
643                                 return TC_ACT_OK;
644                         } else {
645                                 jmp_ttl -= 1;
646                                 if (jmp_ttl > 0)
647                                         goto restart_act_graph;
648                                 else /* faulty graph, stop pipeline */
649                                         return TC_ACT_OK;
650                         }
651                 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
652                         tcf_action_goto_chain_exec(a, res);
653                 }
654
655                 if (ret != TC_ACT_PIPE)
656                         break;
657         }
658
659         return ret;
660 }
661 EXPORT_SYMBOL(tcf_action_exec);
662
663 int tcf_action_destroy(struct tc_action *actions[], int bind)
664 {
665         const struct tc_action_ops *ops;
666         struct tc_action *a;
667         int ret = 0, i;
668
669         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
670                 a = actions[i];
671                 actions[i] = NULL;
672                 ops = a->ops;
673                 ret = __tcf_idr_release(a, bind, true);
674                 if (ret == ACT_P_DELETED)
675                         module_put(ops->owner);
676                 else if (ret < 0)
677                         return ret;
678         }
679         return ret;
680 }
681
682 static int tcf_action_put(struct tc_action *p)
683 {
684         return __tcf_action_put(p, false);
685 }
686
687 /* Put all actions in this array, skip those NULL's. */
688 static void tcf_action_put_many(struct tc_action *actions[])
689 {
690         int i;
691
692         for (i = 0; i < TCA_ACT_MAX_PRIO; i++) {
693                 struct tc_action *a = actions[i];
694                 const struct tc_action_ops *ops;
695
696                 if (!a)
697                         continue;
698                 ops = a->ops;
699                 if (tcf_action_put(a))
700                         module_put(ops->owner);
701         }
702 }
703
704 int
705 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
706 {
707         return a->ops->dump(skb, a, bind, ref);
708 }
709
710 int
711 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
712 {
713         int err = -EINVAL;
714         unsigned char *b = skb_tail_pointer(skb);
715         struct nlattr *nest;
716         struct tc_cookie *cookie;
717
718         if (nla_put_string(skb, TCA_KIND, a->ops->kind))
719                 goto nla_put_failure;
720         if (tcf_action_copy_stats(skb, a, 0))
721                 goto nla_put_failure;
722
723         rcu_read_lock();
724         cookie = rcu_dereference(a->act_cookie);
725         if (cookie) {
726                 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
727                         rcu_read_unlock();
728                         goto nla_put_failure;
729                 }
730         }
731         rcu_read_unlock();
732
733         nest = nla_nest_start(skb, TCA_OPTIONS);
734         if (nest == NULL)
735                 goto nla_put_failure;
736         err = tcf_action_dump_old(skb, a, bind, ref);
737         if (err > 0) {
738                 nla_nest_end(skb, nest);
739                 return err;
740         }
741
742 nla_put_failure:
743         nlmsg_trim(skb, b);
744         return -1;
745 }
746 EXPORT_SYMBOL(tcf_action_dump_1);
747
748 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
749                     int bind, int ref)
750 {
751         struct tc_action *a;
752         int err = -EINVAL, i;
753         struct nlattr *nest;
754
755         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
756                 a = actions[i];
757                 nest = nla_nest_start(skb, a->order);
758                 if (nest == NULL)
759                         goto nla_put_failure;
760                 err = tcf_action_dump_1(skb, a, bind, ref);
761                 if (err < 0)
762                         goto errout;
763                 nla_nest_end(skb, nest);
764         }
765
766         return 0;
767
768 nla_put_failure:
769         err = -EINVAL;
770 errout:
771         nla_nest_cancel(skb, nest);
772         return err;
773 }
774
775 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
776 {
777         struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
778         if (!c)
779                 return NULL;
780
781         c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
782         if (!c->data) {
783                 kfree(c);
784                 return NULL;
785         }
786         c->len = nla_len(tb[TCA_ACT_COOKIE]);
787
788         return c;
789 }
790
791 static bool tcf_action_valid(int action)
792 {
793         int opcode = TC_ACT_EXT_OPCODE(action);
794
795         if (!opcode)
796                 return action <= TC_ACT_VALUE_MAX;
797         return opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC;
798 }
799
800 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
801                                     struct nlattr *nla, struct nlattr *est,
802                                     char *name, int ovr, int bind,
803                                     bool rtnl_held,
804                                     struct netlink_ext_ack *extack)
805 {
806         struct tc_action *a;
807         struct tc_action_ops *a_o;
808         struct tc_cookie *cookie = NULL;
809         char act_name[IFNAMSIZ];
810         struct nlattr *tb[TCA_ACT_MAX + 1];
811         struct nlattr *kind;
812         int err;
813
814         if (name == NULL) {
815                 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
816                 if (err < 0)
817                         goto err_out;
818                 err = -EINVAL;
819                 kind = tb[TCA_ACT_KIND];
820                 if (!kind) {
821                         NL_SET_ERR_MSG(extack, "TC action kind must be specified");
822                         goto err_out;
823                 }
824                 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
825                         NL_SET_ERR_MSG(extack, "TC action name too long");
826                         goto err_out;
827                 }
828                 if (tb[TCA_ACT_COOKIE]) {
829                         int cklen = nla_len(tb[TCA_ACT_COOKIE]);
830
831                         if (cklen > TC_COOKIE_MAX_SIZE) {
832                                 NL_SET_ERR_MSG(extack, "TC cookie size above the maximum");
833                                 goto err_out;
834                         }
835
836                         cookie = nla_memdup_cookie(tb);
837                         if (!cookie) {
838                                 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
839                                 err = -ENOMEM;
840                                 goto err_out;
841                         }
842                 }
843         } else {
844                 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
845                         NL_SET_ERR_MSG(extack, "TC action name too long");
846                         err = -EINVAL;
847                         goto err_out;
848                 }
849         }
850
851         a_o = tc_lookup_action_n(act_name);
852         if (a_o == NULL) {
853 #ifdef CONFIG_MODULES
854                 if (rtnl_held)
855                         rtnl_unlock();
856                 request_module("act_%s", act_name);
857                 if (rtnl_held)
858                         rtnl_lock();
859
860                 a_o = tc_lookup_action_n(act_name);
861
862                 /* We dropped the RTNL semaphore in order to
863                  * perform the module load.  So, even if we
864                  * succeeded in loading the module we have to
865                  * tell the caller to replay the request.  We
866                  * indicate this using -EAGAIN.
867                  */
868                 if (a_o != NULL) {
869                         err = -EAGAIN;
870                         goto err_mod;
871                 }
872 #endif
873                 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
874                 err = -ENOENT;
875                 goto err_out;
876         }
877
878         /* backward compatibility for policer */
879         if (name == NULL)
880                 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
881                                 rtnl_held, extack);
882         else
883                 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
884                                 extack);
885         if (err < 0)
886                 goto err_mod;
887
888         if (!name && tb[TCA_ACT_COOKIE])
889                 tcf_set_action_cookie(&a->act_cookie, cookie);
890
891         /* module count goes up only when brand new policy is created
892          * if it exists and is only bound to in a_o->init() then
893          * ACT_P_CREATED is not returned (a zero is).
894          */
895         if (err != ACT_P_CREATED)
896                 module_put(a_o->owner);
897
898         if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
899                 err = tcf_action_goto_chain_init(a, tp);
900                 if (err) {
901                         struct tc_action *actions[] = { a, NULL };
902
903                         tcf_action_destroy(actions, bind);
904                         NL_SET_ERR_MSG(extack, "Failed to init TC action chain");
905                         return ERR_PTR(err);
906                 }
907         }
908
909         if (!tcf_action_valid(a->tcfa_action)) {
910                 NL_SET_ERR_MSG(extack, "invalid action value, using TC_ACT_UNSPEC instead");
911                 a->tcfa_action = TC_ACT_UNSPEC;
912         }
913
914         return a;
915
916 err_mod:
917         module_put(a_o->owner);
918 err_out:
919         if (cookie) {
920                 kfree(cookie->data);
921                 kfree(cookie);
922         }
923         return ERR_PTR(err);
924 }
925
926 /* Returns numbers of initialized actions or negative error. */
927
928 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
929                     struct nlattr *est, char *name, int ovr, int bind,
930                     struct tc_action *actions[], size_t *attr_size,
931                     bool rtnl_held, struct netlink_ext_ack *extack)
932 {
933         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
934         struct tc_action *act;
935         size_t sz = 0;
936         int err;
937         int i;
938
939         err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
940         if (err < 0)
941                 return err;
942
943         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
944                 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
945                                         rtnl_held, extack);
946                 if (IS_ERR(act)) {
947                         err = PTR_ERR(act);
948                         goto err;
949                 }
950                 act->order = i;
951                 sz += tcf_action_fill_size(act);
952                 /* Start from index 0 */
953                 actions[i - 1] = act;
954         }
955
956         *attr_size = tcf_action_full_attrs_size(sz);
957         return i - 1;
958
959 err:
960         tcf_action_destroy(actions, bind);
961         return err;
962 }
963
964 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
965                           int compat_mode)
966 {
967         int err = 0;
968         struct gnet_dump d;
969
970         if (p == NULL)
971                 goto errout;
972
973         /* compat_mode being true specifies a call that is supposed
974          * to add additional backward compatibility statistic TLVs.
975          */
976         if (compat_mode) {
977                 if (p->type == TCA_OLD_COMPAT)
978                         err = gnet_stats_start_copy_compat(skb, 0,
979                                                            TCA_STATS,
980                                                            TCA_XSTATS,
981                                                            &p->tcfa_lock, &d,
982                                                            TCA_PAD);
983                 else
984                         return 0;
985         } else
986                 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
987                                             &p->tcfa_lock, &d, TCA_ACT_PAD);
988
989         if (err < 0)
990                 goto errout;
991
992         if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
993             gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
994             gnet_stats_copy_queue(&d, p->cpu_qstats,
995                                   &p->tcfa_qstats,
996                                   p->tcfa_qstats.qlen) < 0)
997                 goto errout;
998
999         if (gnet_stats_finish_copy(&d) < 0)
1000                 goto errout;
1001
1002         return 0;
1003
1004 errout:
1005         return -1;
1006 }
1007
1008 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1009                         u32 portid, u32 seq, u16 flags, int event, int bind,
1010                         int ref)
1011 {
1012         struct tcamsg *t;
1013         struct nlmsghdr *nlh;
1014         unsigned char *b = skb_tail_pointer(skb);
1015         struct nlattr *nest;
1016
1017         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1018         if (!nlh)
1019                 goto out_nlmsg_trim;
1020         t = nlmsg_data(nlh);
1021         t->tca_family = AF_UNSPEC;
1022         t->tca__pad1 = 0;
1023         t->tca__pad2 = 0;
1024
1025         nest = nla_nest_start(skb, TCA_ACT_TAB);
1026         if (!nest)
1027                 goto out_nlmsg_trim;
1028
1029         if (tcf_action_dump(skb, actions, bind, ref) < 0)
1030                 goto out_nlmsg_trim;
1031
1032         nla_nest_end(skb, nest);
1033
1034         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1035         return skb->len;
1036
1037 out_nlmsg_trim:
1038         nlmsg_trim(skb, b);
1039         return -1;
1040 }
1041
1042 static int
1043 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1044                struct tc_action *actions[], int event,
1045                struct netlink_ext_ack *extack)
1046 {
1047         struct sk_buff *skb;
1048
1049         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1050         if (!skb)
1051                 return -ENOBUFS;
1052         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1053                          0, 1) <= 0) {
1054                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1055                 kfree_skb(skb);
1056                 return -EINVAL;
1057         }
1058
1059         return rtnl_unicast(skb, net, portid);
1060 }
1061
1062 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1063                                           struct nlmsghdr *n, u32 portid,
1064                                           struct netlink_ext_ack *extack)
1065 {
1066         struct nlattr *tb[TCA_ACT_MAX + 1];
1067         const struct tc_action_ops *ops;
1068         struct tc_action *a;
1069         int index;
1070         int err;
1071
1072         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1073         if (err < 0)
1074                 goto err_out;
1075
1076         err = -EINVAL;
1077         if (tb[TCA_ACT_INDEX] == NULL ||
1078             nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1079                 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1080                 goto err_out;
1081         }
1082         index = nla_get_u32(tb[TCA_ACT_INDEX]);
1083
1084         err = -EINVAL;
1085         ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1086         if (!ops) { /* could happen in batch of actions */
1087                 NL_SET_ERR_MSG(extack, "Specified TC action not found");
1088                 goto err_out;
1089         }
1090         err = -ENOENT;
1091         if (ops->lookup(net, &a, index, extack) == 0)
1092                 goto err_mod;
1093
1094         module_put(ops->owner);
1095         return a;
1096
1097 err_mod:
1098         module_put(ops->owner);
1099 err_out:
1100         return ERR_PTR(err);
1101 }
1102
1103 static int tca_action_flush(struct net *net, struct nlattr *nla,
1104                             struct nlmsghdr *n, u32 portid,
1105                             struct netlink_ext_ack *extack)
1106 {
1107         struct sk_buff *skb;
1108         unsigned char *b;
1109         struct nlmsghdr *nlh;
1110         struct tcamsg *t;
1111         struct netlink_callback dcb;
1112         struct nlattr *nest;
1113         struct nlattr *tb[TCA_ACT_MAX + 1];
1114         const struct tc_action_ops *ops;
1115         struct nlattr *kind;
1116         int err = -ENOMEM;
1117
1118         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1119         if (!skb)
1120                 return err;
1121
1122         b = skb_tail_pointer(skb);
1123
1124         err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1125         if (err < 0)
1126                 goto err_out;
1127
1128         err = -EINVAL;
1129         kind = tb[TCA_ACT_KIND];
1130         ops = tc_lookup_action(kind);
1131         if (!ops) { /*some idjot trying to flush unknown action */
1132                 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1133                 goto err_out;
1134         }
1135
1136         nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1137                         sizeof(*t), 0);
1138         if (!nlh) {
1139                 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1140                 goto out_module_put;
1141         }
1142         t = nlmsg_data(nlh);
1143         t->tca_family = AF_UNSPEC;
1144         t->tca__pad1 = 0;
1145         t->tca__pad2 = 0;
1146
1147         nest = nla_nest_start(skb, TCA_ACT_TAB);
1148         if (!nest) {
1149                 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1150                 goto out_module_put;
1151         }
1152
1153         err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1154         if (err <= 0) {
1155                 nla_nest_cancel(skb, nest);
1156                 goto out_module_put;
1157         }
1158
1159         nla_nest_end(skb, nest);
1160
1161         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1162         nlh->nlmsg_flags |= NLM_F_ROOT;
1163         module_put(ops->owner);
1164         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1165                              n->nlmsg_flags & NLM_F_ECHO);
1166         if (err > 0)
1167                 return 0;
1168         if (err < 0)
1169                 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1170
1171         return err;
1172
1173 out_module_put:
1174         module_put(ops->owner);
1175 err_out:
1176         kfree_skb(skb);
1177         return err;
1178 }
1179
1180 static int tcf_action_delete(struct net *net, struct tc_action *actions[],
1181                              struct netlink_ext_ack *extack)
1182 {
1183         int i;
1184
1185         for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1186                 struct tc_action *a = actions[i];
1187                 const struct tc_action_ops *ops = a->ops;
1188                 /* Actions can be deleted concurrently so we must save their
1189                  * type and id to search again after reference is released.
1190                  */
1191                 struct tcf_idrinfo *idrinfo = a->idrinfo;
1192                 u32 act_index = a->tcfa_index;
1193
1194                 if (tcf_action_put(a)) {
1195                         /* last reference, action was deleted concurrently */
1196                         module_put(ops->owner);
1197                 } else  {
1198                         int ret;
1199
1200                         /* now do the delete */
1201                         ret = tcf_idr_delete_index(idrinfo, act_index);
1202                         if (ret < 0)
1203                                 return ret;
1204                 }
1205                 actions[i] = NULL;
1206         }
1207         return 0;
1208 }
1209
1210 static int
1211 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1212                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1213 {
1214         int ret;
1215         struct sk_buff *skb;
1216
1217         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1218                         GFP_KERNEL);
1219         if (!skb)
1220                 return -ENOBUFS;
1221
1222         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1223                          0, 2) <= 0) {
1224                 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1225                 kfree_skb(skb);
1226                 return -EINVAL;
1227         }
1228
1229         /* now do the delete */
1230         ret = tcf_action_delete(net, actions, extack);
1231         if (ret < 0) {
1232                 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1233                 kfree_skb(skb);
1234                 return ret;
1235         }
1236
1237         ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1238                              n->nlmsg_flags & NLM_F_ECHO);
1239         if (ret > 0)
1240                 return 0;
1241         return ret;
1242 }
1243
1244 static int
1245 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1246               u32 portid, int event, struct netlink_ext_ack *extack)
1247 {
1248         int i, ret;
1249         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1250         struct tc_action *act;
1251         size_t attr_size = 0;
1252         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1253
1254         ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
1255         if (ret < 0)
1256                 return ret;
1257
1258         if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1259                 if (tb[1])
1260                         return tca_action_flush(net, tb[1], n, portid, extack);
1261
1262                 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1263                 return -EINVAL;
1264         }
1265
1266         for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1267                 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1268                 if (IS_ERR(act)) {
1269                         ret = PTR_ERR(act);
1270                         goto err;
1271                 }
1272                 act->order = i;
1273                 attr_size += tcf_action_fill_size(act);
1274                 actions[i - 1] = act;
1275         }
1276
1277         attr_size = tcf_action_full_attrs_size(attr_size);
1278
1279         if (event == RTM_GETACTION)
1280                 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1281         else { /* delete */
1282                 ret = tcf_del_notify(net, n, actions, portid, attr_size, extack);
1283                 if (ret)
1284                         goto err;
1285                 return 0;
1286         }
1287 err:
1288         tcf_action_put_many(actions);
1289         return ret;
1290 }
1291
1292 static int
1293 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1294                u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1295 {
1296         struct sk_buff *skb;
1297         int err = 0;
1298
1299         skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1300                         GFP_KERNEL);
1301         if (!skb)
1302                 return -ENOBUFS;
1303
1304         if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1305                          RTM_NEWACTION, 0, 0) <= 0) {
1306                 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1307                 kfree_skb(skb);
1308                 return -EINVAL;
1309         }
1310
1311         err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1312                              n->nlmsg_flags & NLM_F_ECHO);
1313         if (err > 0)
1314                 err = 0;
1315         return err;
1316 }
1317
1318 static int tcf_action_add(struct net *net, struct nlattr *nla,
1319                           struct nlmsghdr *n, u32 portid, int ovr,
1320                           struct netlink_ext_ack *extack)
1321 {
1322         size_t attr_size = 0;
1323         int ret = 0;
1324         struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1325
1326         ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0, actions,
1327                               &attr_size, true, extack);
1328         if (ret < 0)
1329                 return ret;
1330         ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1331         if (ovr)
1332                 tcf_action_put_many(actions);
1333
1334         return ret;
1335 }
1336
1337 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1338 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1339         [TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1340                              .validation_data = &tcaa_root_flags_allowed },
1341         [TCA_ROOT_TIME_DELTA]      = { .type = NLA_U32 },
1342 };
1343
1344 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1345                          struct netlink_ext_ack *extack)
1346 {
1347         struct net *net = sock_net(skb->sk);
1348         struct nlattr *tca[TCA_ROOT_MAX + 1];
1349         u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1350         int ret = 0, ovr = 0;
1351
1352         if ((n->nlmsg_type != RTM_GETACTION) &&
1353             !netlink_capable(skb, CAP_NET_ADMIN))
1354                 return -EPERM;
1355
1356         ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1357                           extack);
1358         if (ret < 0)
1359                 return ret;
1360
1361         if (tca[TCA_ACT_TAB] == NULL) {
1362                 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1363                 return -EINVAL;
1364         }
1365
1366         /* n->nlmsg_flags & NLM_F_CREATE */
1367         switch (n->nlmsg_type) {
1368         case RTM_NEWACTION:
1369                 /* we are going to assume all other flags
1370                  * imply create only if it doesn't exist
1371                  * Note that CREATE | EXCL implies that
1372                  * but since we want avoid ambiguity (eg when flags
1373                  * is zero) then just set this
1374                  */
1375                 if (n->nlmsg_flags & NLM_F_REPLACE)
1376                         ovr = 1;
1377 replay:
1378                 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1379                                      extack);
1380                 if (ret == -EAGAIN)
1381                         goto replay;
1382                 break;
1383         case RTM_DELACTION:
1384                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1385                                     portid, RTM_DELACTION, extack);
1386                 break;
1387         case RTM_GETACTION:
1388                 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1389                                     portid, RTM_GETACTION, extack);
1390                 break;
1391         default:
1392                 BUG();
1393         }
1394
1395         return ret;
1396 }
1397
1398 static struct nlattr *find_dump_kind(struct nlattr **nla)
1399 {
1400         struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1401         struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1402         struct nlattr *kind;
1403
1404         tb1 = nla[TCA_ACT_TAB];
1405         if (tb1 == NULL)
1406                 return NULL;
1407
1408         if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1409                       NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1410                 return NULL;
1411
1412         if (tb[1] == NULL)
1413                 return NULL;
1414         if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1415                 return NULL;
1416         kind = tb2[TCA_ACT_KIND];
1417
1418         return kind;
1419 }
1420
1421 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1422 {
1423         struct net *net = sock_net(skb->sk);
1424         struct nlmsghdr *nlh;
1425         unsigned char *b = skb_tail_pointer(skb);
1426         struct nlattr *nest;
1427         struct tc_action_ops *a_o;
1428         int ret = 0;
1429         struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1430         struct nlattr *tb[TCA_ROOT_MAX + 1];
1431         struct nlattr *count_attr = NULL;
1432         unsigned long jiffy_since = 0;
1433         struct nlattr *kind = NULL;
1434         struct nla_bitfield32 bf;
1435         u32 msecs_since = 0;
1436         u32 act_count = 0;
1437
1438         ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1439                           tcaa_policy, NULL);
1440         if (ret < 0)
1441                 return ret;
1442
1443         kind = find_dump_kind(tb);
1444         if (kind == NULL) {
1445                 pr_info("tc_dump_action: action bad kind\n");
1446                 return 0;
1447         }
1448
1449         a_o = tc_lookup_action(kind);
1450         if (a_o == NULL)
1451                 return 0;
1452
1453         cb->args[2] = 0;
1454         if (tb[TCA_ROOT_FLAGS]) {
1455                 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1456                 cb->args[2] = bf.value;
1457         }
1458
1459         if (tb[TCA_ROOT_TIME_DELTA]) {
1460                 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1461         }
1462
1463         nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1464                         cb->nlh->nlmsg_type, sizeof(*t), 0);
1465         if (!nlh)
1466                 goto out_module_put;
1467
1468         if (msecs_since)
1469                 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1470
1471         t = nlmsg_data(nlh);
1472         t->tca_family = AF_UNSPEC;
1473         t->tca__pad1 = 0;
1474         t->tca__pad2 = 0;
1475         cb->args[3] = jiffy_since;
1476         count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1477         if (!count_attr)
1478                 goto out_module_put;
1479
1480         nest = nla_nest_start(skb, TCA_ACT_TAB);
1481         if (nest == NULL)
1482                 goto out_module_put;
1483
1484         ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1485         if (ret < 0)
1486                 goto out_module_put;
1487
1488         if (ret > 0) {
1489                 nla_nest_end(skb, nest);
1490                 ret = skb->len;
1491                 act_count = cb->args[1];
1492                 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1493                 cb->args[1] = 0;
1494         } else
1495                 nlmsg_trim(skb, b);
1496
1497         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1498         if (NETLINK_CB(cb->skb).portid && ret)
1499                 nlh->nlmsg_flags |= NLM_F_MULTI;
1500         module_put(a_o->owner);
1501         return skb->len;
1502
1503 out_module_put:
1504         module_put(a_o->owner);
1505         nlmsg_trim(skb, b);
1506         return skb->len;
1507 }
1508
1509 struct tcf_action_net {
1510         struct rhashtable egdev_ht;
1511 };
1512
1513 static unsigned int tcf_action_net_id;
1514
1515 struct tcf_action_egdev_cb {
1516         struct list_head list;
1517         tc_setup_cb_t *cb;
1518         void *cb_priv;
1519 };
1520
1521 struct tcf_action_egdev {
1522         struct rhash_head ht_node;
1523         const struct net_device *dev;
1524         unsigned int refcnt;
1525         struct list_head cb_list;
1526 };
1527
1528 static const struct rhashtable_params tcf_action_egdev_ht_params = {
1529         .key_offset = offsetof(struct tcf_action_egdev, dev),
1530         .head_offset = offsetof(struct tcf_action_egdev, ht_node),
1531         .key_len = sizeof(const struct net_device *),
1532 };
1533
1534 static struct tcf_action_egdev *
1535 tcf_action_egdev_lookup(const struct net_device *dev)
1536 {
1537         struct net *net = dev_net(dev);
1538         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1539
1540         return rhashtable_lookup_fast(&tan->egdev_ht, &dev,
1541                                       tcf_action_egdev_ht_params);
1542 }
1543
1544 static struct tcf_action_egdev *
1545 tcf_action_egdev_get(const struct net_device *dev)
1546 {
1547         struct tcf_action_egdev *egdev;
1548         struct tcf_action_net *tan;
1549
1550         egdev = tcf_action_egdev_lookup(dev);
1551         if (egdev)
1552                 goto inc_ref;
1553
1554         egdev = kzalloc(sizeof(*egdev), GFP_KERNEL);
1555         if (!egdev)
1556                 return NULL;
1557         INIT_LIST_HEAD(&egdev->cb_list);
1558         egdev->dev = dev;
1559         tan = net_generic(dev_net(dev), tcf_action_net_id);
1560         rhashtable_insert_fast(&tan->egdev_ht, &egdev->ht_node,
1561                                tcf_action_egdev_ht_params);
1562
1563 inc_ref:
1564         egdev->refcnt++;
1565         return egdev;
1566 }
1567
1568 static void tcf_action_egdev_put(struct tcf_action_egdev *egdev)
1569 {
1570         struct tcf_action_net *tan;
1571
1572         if (--egdev->refcnt)
1573                 return;
1574         tan = net_generic(dev_net(egdev->dev), tcf_action_net_id);
1575         rhashtable_remove_fast(&tan->egdev_ht, &egdev->ht_node,
1576                                tcf_action_egdev_ht_params);
1577         kfree(egdev);
1578 }
1579
1580 static struct tcf_action_egdev_cb *
1581 tcf_action_egdev_cb_lookup(struct tcf_action_egdev *egdev,
1582                            tc_setup_cb_t *cb, void *cb_priv)
1583 {
1584         struct tcf_action_egdev_cb *egdev_cb;
1585
1586         list_for_each_entry(egdev_cb, &egdev->cb_list, list)
1587                 if (egdev_cb->cb == cb && egdev_cb->cb_priv == cb_priv)
1588                         return egdev_cb;
1589         return NULL;
1590 }
1591
1592 static int tcf_action_egdev_cb_call(struct tcf_action_egdev *egdev,
1593                                     enum tc_setup_type type,
1594                                     void *type_data, bool err_stop)
1595 {
1596         struct tcf_action_egdev_cb *egdev_cb;
1597         int ok_count = 0;
1598         int err;
1599
1600         list_for_each_entry(egdev_cb, &egdev->cb_list, list) {
1601                 err = egdev_cb->cb(type, type_data, egdev_cb->cb_priv);
1602                 if (err) {
1603                         if (err_stop)
1604                                 return err;
1605                 } else {
1606                         ok_count++;
1607                 }
1608         }
1609         return ok_count;
1610 }
1611
1612 static int tcf_action_egdev_cb_add(struct tcf_action_egdev *egdev,
1613                                    tc_setup_cb_t *cb, void *cb_priv)
1614 {
1615         struct tcf_action_egdev_cb *egdev_cb;
1616
1617         egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1618         if (WARN_ON(egdev_cb))
1619                 return -EEXIST;
1620         egdev_cb = kzalloc(sizeof(*egdev_cb), GFP_KERNEL);
1621         if (!egdev_cb)
1622                 return -ENOMEM;
1623         egdev_cb->cb = cb;
1624         egdev_cb->cb_priv = cb_priv;
1625         list_add(&egdev_cb->list, &egdev->cb_list);
1626         return 0;
1627 }
1628
1629 static void tcf_action_egdev_cb_del(struct tcf_action_egdev *egdev,
1630                                     tc_setup_cb_t *cb, void *cb_priv)
1631 {
1632         struct tcf_action_egdev_cb *egdev_cb;
1633
1634         egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1635         if (WARN_ON(!egdev_cb))
1636                 return;
1637         list_del(&egdev_cb->list);
1638         kfree(egdev_cb);
1639 }
1640
1641 static int __tc_setup_cb_egdev_register(const struct net_device *dev,
1642                                         tc_setup_cb_t *cb, void *cb_priv)
1643 {
1644         struct tcf_action_egdev *egdev = tcf_action_egdev_get(dev);
1645         int err;
1646
1647         if (!egdev)
1648                 return -ENOMEM;
1649         err = tcf_action_egdev_cb_add(egdev, cb, cb_priv);
1650         if (err)
1651                 goto err_cb_add;
1652         return 0;
1653
1654 err_cb_add:
1655         tcf_action_egdev_put(egdev);
1656         return err;
1657 }
1658 int tc_setup_cb_egdev_register(const struct net_device *dev,
1659                                tc_setup_cb_t *cb, void *cb_priv)
1660 {
1661         int err;
1662
1663         rtnl_lock();
1664         err = __tc_setup_cb_egdev_register(dev, cb, cb_priv);
1665         rtnl_unlock();
1666         return err;
1667 }
1668 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_register);
1669
1670 static void __tc_setup_cb_egdev_unregister(const struct net_device *dev,
1671                                            tc_setup_cb_t *cb, void *cb_priv)
1672 {
1673         struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1674
1675         if (WARN_ON(!egdev))
1676                 return;
1677         tcf_action_egdev_cb_del(egdev, cb, cb_priv);
1678         tcf_action_egdev_put(egdev);
1679 }
1680 void tc_setup_cb_egdev_unregister(const struct net_device *dev,
1681                                   tc_setup_cb_t *cb, void *cb_priv)
1682 {
1683         rtnl_lock();
1684         __tc_setup_cb_egdev_unregister(dev, cb, cb_priv);
1685         rtnl_unlock();
1686 }
1687 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_unregister);
1688
1689 int tc_setup_cb_egdev_call(const struct net_device *dev,
1690                            enum tc_setup_type type, void *type_data,
1691                            bool err_stop)
1692 {
1693         struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1694
1695         if (!egdev)
1696                 return 0;
1697         return tcf_action_egdev_cb_call(egdev, type, type_data, err_stop);
1698 }
1699 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_call);
1700
1701 static __net_init int tcf_action_net_init(struct net *net)
1702 {
1703         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1704
1705         return rhashtable_init(&tan->egdev_ht, &tcf_action_egdev_ht_params);
1706 }
1707
1708 static void __net_exit tcf_action_net_exit(struct net *net)
1709 {
1710         struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1711
1712         rhashtable_destroy(&tan->egdev_ht);
1713 }
1714
1715 static struct pernet_operations tcf_action_net_ops = {
1716         .init = tcf_action_net_init,
1717         .exit = tcf_action_net_exit,
1718         .id = &tcf_action_net_id,
1719         .size = sizeof(struct tcf_action_net),
1720 };
1721
1722 static int __init tc_action_init(void)
1723 {
1724         int err;
1725
1726         err = register_pernet_subsys(&tcf_action_net_ops);
1727         if (err)
1728                 return err;
1729
1730         rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1731         rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1732         rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1733                       0);
1734
1735         return 0;
1736 }
1737
1738 subsys_initcall(tc_action_init);