]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - net/sched/cls_api.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'io_uring-5.9-2020-09-25' of git://git.kernel.dk/linux-block
[mirror_ubuntu-jammy-kernel.git] / net / sched / cls_api.c
CommitLineData
2874c5fd 1// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4
LT		 2/*
3 * net/sched/cls_api.c Packet classifier API.
4 *
1da177e4
LT		 5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Changes:
8 *
9 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
1da177e4
LT		 10 */
11
1da177e4
LT		 12#include <linux/module.h>
13#include <linux/types.h>
14#include <linux/kernel.h>
1da177e4 15#include <linux/string.h>
1da177e4 16#include <linux/errno.h>
33a48927 17#include <linux/err.h>
1da177e4 18#include <linux/skbuff.h>
1da177e4
LT		 19#include <linux/init.h>
20#include <linux/kmod.h>
5a0e3ad6 21#include <linux/slab.h>
48617387 22#include <linux/idr.h>
59eb87cb 23#include <linux/jhash.h>
43719298 24#include <linux/rculist.h>
b854272b
DL		 25#include <net/net_namespace.h>
26#include <net/sock.h>
dc5fc579 27#include <net/netlink.h>
1da177e4
LT		 28#include <net/pkt_sched.h>
29#include <net/pkt_cls.h>
e3ab786b 30#include <net/tc_act/tc_pedit.h>
3a7b6861
PNA		 31#include <net/tc_act/tc_mirred.h>
32#include <net/tc_act/tc_vlan.h>
33#include <net/tc_act/tc_tunnel_key.h>
34#include <net/tc_act/tc_csum.h>
35#include <net/tc_act/tc_gact.h>
8c8cfc6e 36#include <net/tc_act/tc_police.h>
a7a7be60 37#include <net/tc_act/tc_sample.h>
3a7b6861 38#include <net/tc_act/tc_skbedit.h>
b57dc7c1 39#include <net/tc_act/tc_ct.h>
6749d590 40#include <net/tc_act/tc_mpls.h>
d29bdd69 41#include <net/tc_act/tc_gate.h>
4e481908 42#include <net/flow_offload.h>
1da177e4 43
e331473f
DC		 44extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
45
1da177e4 46/* The list of all installed classifier types */
36272874 47static LIST_HEAD(tcf_proto_base);
1da177e4
LT		 48
49/* Protects list of registered TC modules. It is pure SMP lock. */
50static DEFINE_RWLOCK(cls_mod_lock);
51
59eb87cb
JH		 52static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
53{
54 return jhash_3words(tp->chain->index, tp->prio,
55 (__force __u32)tp->protocol, 0);
56}
57
58static void tcf_proto_signal_destroying(struct tcf_chain *chain,
59 struct tcf_proto *tp)
60{
61 struct tcf_block *block = chain->block;
62
63 mutex_lock(&block->proto_destroy_lock);
64 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
65 destroy_obj_hashfn(tp));
66 mutex_unlock(&block->proto_destroy_lock);
67}
68
69static bool tcf_proto_cmp(const struct tcf_proto *tp1,
70 const struct tcf_proto *tp2)
71{
72 return tp1->chain->index == tp2->chain->index &&
73 tp1->prio == tp2->prio &&
74 tp1->protocol == tp2->protocol;
75}
76
77static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
78 struct tcf_proto *tp)
79{
80 u32 hash = destroy_obj_hashfn(tp);
81 struct tcf_proto *iter;
82 bool found = false;
83
84 rcu_read_lock();
85 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
86 destroy_ht_node, hash) {
87 if (tcf_proto_cmp(tp, iter)) {
88 found = true;
89 break;
90 }
91 }
92 rcu_read_unlock();
93
94 return found;
95}
96
97static void
98tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
99{
100 struct tcf_block *block = chain->block;
101
102 mutex_lock(&block->proto_destroy_lock);
103 if (hash_hashed(&tp->destroy_ht_node))
104 hash_del_rcu(&tp->destroy_ht_node);
105 mutex_unlock(&block->proto_destroy_lock);
106}
107
1da177e4
LT		 108/* Find classifier type by string name */
109
f34e8bff 110static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
1da177e4 111{
dcd76081 112 const struct tcf_proto_ops *t, *res = NULL;
1da177e4
LT		 113
114 if (kind) {
115 read_lock(&cls_mod_lock);
36272874 116 list_for_each_entry(t, &tcf_proto_base, head) {
33a48927 117 if (strcmp(kind, t->kind) == 0) {
dcd76081
ED		 118 if (try_module_get(t->owner))
119 res = t;
1da177e4
LT		 120 break;
121 }
122 }
123 read_unlock(&cls_mod_lock);
124 }
dcd76081 125 return res;
1da177e4
LT		 126}
127
f34e8bff 128static const struct tcf_proto_ops *
12db03b6
VB		 129tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
130 struct netlink_ext_ack *extack)
f34e8bff
JP		 131{
132 const struct tcf_proto_ops *ops;
133
134 ops = __tcf_proto_lookup_ops(kind);
135 if (ops)
136 return ops;
137#ifdef CONFIG_MODULES
12db03b6
VB		 138 if (rtnl_held)
139 rtnl_unlock();
f34e8bff 140 request_module("cls_%s", kind);
12db03b6
VB		 141 if (rtnl_held)
142 rtnl_lock();
f34e8bff
JP		 143 ops = __tcf_proto_lookup_ops(kind);
144 /* We dropped the RTNL semaphore in order to perform
145 * the module load. So, even if we succeeded in loading
146 * the module we have to replay the request. We indicate
147 * this using -EAGAIN.
148 */
149 if (ops) {
150 module_put(ops->owner);
151 return ERR_PTR(-EAGAIN);
152 }
153#endif
154 NL_SET_ERR_MSG(extack, "TC classifier not found");
155 return ERR_PTR(-ENOENT);
156}
157
1da177e4
LT		 158/* Register(unregister) new classifier type */
159
160int register_tcf_proto_ops(struct tcf_proto_ops *ops)
161{
36272874 162 struct tcf_proto_ops *t;
1da177e4
LT		 163 int rc = -EEXIST;
164
165 write_lock(&cls_mod_lock);
36272874 166 list_for_each_entry(t, &tcf_proto_base, head)
1da177e4
LT		 167 if (!strcmp(ops->kind, t->kind))
168 goto out;
169
36272874 170 list_add_tail(&ops->head, &tcf_proto_base);
1da177e4
LT		 171 rc = 0;
172out:
173 write_unlock(&cls_mod_lock);
174 return rc;
175}
aa767bfe 176EXPORT_SYMBOL(register_tcf_proto_ops);
1da177e4 177
7aa0045d
CW		 178static struct workqueue_struct *tc_filter_wq;
179
1da177e4
LT		 180int unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
181{
36272874 182 struct tcf_proto_ops *t;
1da177e4
LT		 183 int rc = -ENOENT;
184
c78e1746
DB		 185 /* Wait for outstanding call_rcu()s, if any, from a
186 * tcf_proto_ops's destroy() handler.
187 */
188 rcu_barrier();
7aa0045d 189 flush_workqueue(tc_filter_wq);
c78e1746 190
1da177e4 191 write_lock(&cls_mod_lock);
dcd76081
ED		 192 list_for_each_entry(t, &tcf_proto_base, head) {
193 if (t == ops) {
194 list_del(&t->head);
195 rc = 0;
1da177e4 196 break;
dcd76081
ED		 197 }
198 }
1da177e4
LT		 199 write_unlock(&cls_mod_lock);
200 return rc;
201}
aa767bfe 202EXPORT_SYMBOL(unregister_tcf_proto_ops);
1da177e4 203
aaa908ff 204bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
7aa0045d 205{
aaa908ff
CW		 206 INIT_RCU_WORK(rwork, func);
207 return queue_rcu_work(tc_filter_wq, rwork);
7aa0045d
CW		 208}
209EXPORT_SYMBOL(tcf_queue_work);
210
1da177e4
LT		 211/* Select new prio value from the range, managed by kernel. */
212
aa767bfe 213static inline u32 tcf_auto_prio(struct tcf_proto *tp)
1da177e4 214{
aa767bfe 215 u32 first = TC_H_MAKE(0xC0000000U, 0U);
1da177e4
LT		 216
217 if (tp)
cc7ec456 218 first = tp->prio - 1;
1da177e4 219
7961973a 220 return TC_H_MAJ(first);
1da177e4
LT		 221}
222
6f96c3c6
CW		 223static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
224{
225 if (kind)
226 return nla_strlcpy(name, kind, IFNAMSIZ) >= IFNAMSIZ;
227 memset(name, 0, IFNAMSIZ);
228 return false;
229}
230
470502de
VB		 231static bool tcf_proto_is_unlocked(const char *kind)
232{
233 const struct tcf_proto_ops *ops;
234 bool ret;
235
6f96c3c6
CW		 236 if (strlen(kind) == 0)
237 return false;
238
470502de
VB		 239 ops = tcf_proto_lookup_ops(kind, false, NULL);
240 /* On error return false to take rtnl lock. Proto lookup/create
241 * functions will perform lookup again and properly handle errors.
242 */
243 if (IS_ERR(ops))
244 return false;
245
246 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
247 module_put(ops->owner);
248 return ret;
249}
250
33a48927 251static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
c35a4acc 252 u32 prio, struct tcf_chain *chain,
12db03b6 253 bool rtnl_held,
c35a4acc 254 struct netlink_ext_ack *extack)
33a48927
JP		 255{
256 struct tcf_proto *tp;
257 int err;
258
259 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
260 if (!tp)
261 return ERR_PTR(-ENOBUFS);
262
12db03b6 263 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
f34e8bff
JP		 264 if (IS_ERR(tp->ops)) {
265 err = PTR_ERR(tp->ops);
d68d75fd 266 goto errout;
33a48927
JP		 267 }
268 tp->classify = tp->ops->classify;
269 tp->protocol = protocol;
270 tp->prio = prio;
5bc17018 271 tp->chain = chain;
8b64678e 272 spin_lock_init(&tp->lock);
4dbfa766 273 refcount_set(&tp->refcnt, 1);
33a48927
JP		 274
275 err = tp->ops->init(tp);
276 if (err) {
277 module_put(tp->ops->owner);
278 goto errout;
279 }
280 return tp;
281
282errout:
283 kfree(tp);
284 return ERR_PTR(err);
285}
286
4dbfa766
VB		 287static void tcf_proto_get(struct tcf_proto *tp)
288{
289 refcount_inc(&tp->refcnt);
290}
291
292static void tcf_chain_put(struct tcf_chain *chain);
293
12db03b6 294static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
59eb87cb 295 bool sig_destroy, struct netlink_ext_ack *extack)
cf1facda 296{
12db03b6 297 tp->ops->destroy(tp, rtnl_held, extack);
59eb87cb
JH		 298 if (sig_destroy)
299 tcf_proto_signal_destroyed(tp->chain, tp);
4dbfa766 300 tcf_chain_put(tp->chain);
763dbf63
WC		 301 module_put(tp->ops->owner);
302 kfree_rcu(tp, rcu);
cf1facda
JP		 303}
304
12db03b6 305static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
4dbfa766
VB		 306 struct netlink_ext_ack *extack)
307{
308 if (refcount_dec_and_test(&tp->refcnt))
59eb87cb 309 tcf_proto_destroy(tp, rtnl_held, true, extack);
4dbfa766
VB		 310}
311
a5b72a08 312static bool tcf_proto_check_delete(struct tcf_proto *tp)
8b64678e 313{
a5b72a08
DC		 314 if (tp->ops->delete_empty)
315 return tp->ops->delete_empty(tp);
8b64678e 316
a5b72a08 317 tp->deleting = true;
8b64678e
VB		 318 return tp->deleting;
319}
320
321static void tcf_proto_mark_delete(struct tcf_proto *tp)
322{
323 spin_lock(&tp->lock);
324 tp->deleting = true;
325 spin_unlock(&tp->lock);
326}
327
328static bool tcf_proto_is_deleting(struct tcf_proto *tp)
329{
330 bool deleting;
331
332 spin_lock(&tp->lock);
333 deleting = tp->deleting;
334 spin_unlock(&tp->lock);
335
336 return deleting;
337}
338
c266f64d
VB		 339#define ASSERT_BLOCK_LOCKED(block) \
340 lockdep_assert_held(&(block)->lock)
341
a9b19443
JP		 342struct tcf_filter_chain_list_item {
343 struct list_head list;
344 tcf_chain_head_change_t *chain_head_change;
345 void *chain_head_change_priv;
346};
347
5bc17018
JP		 348static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
349 u32 chain_index)
2190d1d0 350{
5bc17018
JP		 351 struct tcf_chain *chain;
352
c266f64d
VB		 353 ASSERT_BLOCK_LOCKED(block);
354
5bc17018
JP		 355 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
356 if (!chain)
357 return NULL;
43719298 358 list_add_tail_rcu(&chain->list, &block->chain_list);
ed76f5ed 359 mutex_init(&chain->filter_chain_lock);
5bc17018
JP		 360 chain->block = block;
361 chain->index = chain_index;
e2ef7544 362 chain->refcnt = 1;
f71e0ca4
JP		 363 if (!chain->index)
364 block->chain0.chain = chain;
5bc17018 365 return chain;
2190d1d0
JP		 366}
367
a9b19443
JP		 368static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
369 struct tcf_proto *tp_head)
370{
371 if (item->chain_head_change)
372 item->chain_head_change(tp_head, item->chain_head_change_priv);
373}
f71e0ca4
JP		 374
375static void tcf_chain0_head_change(struct tcf_chain *chain,
376 struct tcf_proto *tp_head)
c7eb7d72 377{
a9b19443 378 struct tcf_filter_chain_list_item *item;
f71e0ca4 379 struct tcf_block *block = chain->block;
a9b19443 380
f71e0ca4
JP		 381 if (chain->index)
382 return;
165f0135
VB		 383
384 mutex_lock(&block->lock);
f71e0ca4 385 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
a9b19443 386 tcf_chain_head_change_item(item, tp_head);
165f0135 387 mutex_unlock(&block->lock);
c7eb7d72
JP		 388}
389
c266f64d
VB		 390/* Returns true if block can be safely freed. */
391
392static bool tcf_chain_detach(struct tcf_chain *chain)
f93e1cdc 393{
efbf7897
CW		 394 struct tcf_block *block = chain->block;
395
c266f64d
VB		 396 ASSERT_BLOCK_LOCKED(block);
397
43719298 398 list_del_rcu(&chain->list);
f71e0ca4
JP		 399 if (!chain->index)
400 block->chain0.chain = NULL;
c266f64d
VB		 401
402 if (list_empty(&block->chain_list) &&
403 refcount_read(&block->refcnt) == 0)
404 return true;
405
406 return false;
407}
408
409static void tcf_block_destroy(struct tcf_block *block)
410{
411 mutex_destroy(&block->lock);
59eb87cb 412 mutex_destroy(&block->proto_destroy_lock);
c266f64d
VB		 413 kfree_rcu(block, rcu);
414}
415
416static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
417{
418 struct tcf_block *block = chain->block;
419
ed76f5ed 420 mutex_destroy(&chain->filter_chain_lock);
ee3bbfe8 421 kfree_rcu(chain, rcu);
c266f64d
VB		 422 if (free_block)
423 tcf_block_destroy(block);
e2ef7544 424}
744a4cf6 425
e2ef7544
CW		 426static void tcf_chain_hold(struct tcf_chain *chain)
427{
c266f64d
VB		 428 ASSERT_BLOCK_LOCKED(chain->block);
429
e2ef7544 430 ++chain->refcnt;
2190d1d0
JP		 431}
432
3d32f4c5 433static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
1f3ed383 434{
c266f64d
VB		 435 ASSERT_BLOCK_LOCKED(chain->block);
436
1f3ed383 437 /* In case all the references are action references, this
3d32f4c5 438 * chain should not be shown to the user.
1f3ed383
JP		 439 */
440 return chain->refcnt == chain->action_refcnt;
441}
442
32a4f5ec
JP		 443static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
444 u32 chain_index)
5bc17018
JP		 445{
446 struct tcf_chain *chain;
447
c266f64d
VB		 448 ASSERT_BLOCK_LOCKED(block);
449
5bc17018 450 list_for_each_entry(chain, &block->chain_list, list) {
32a4f5ec 451 if (chain->index == chain_index)
e2ef7544 452 return chain;
32a4f5ec
JP		 453 }
454 return NULL;
455}
456
af699626
PB		 457#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
458static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
459 u32 chain_index)
460{
461 struct tcf_chain *chain;
462
463 list_for_each_entry_rcu(chain, &block->chain_list, list) {
464 if (chain->index == chain_index)
465 return chain;
466 }
467 return NULL;
468}
469#endif
470
32a4f5ec
JP		 471static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
472 u32 seq, u16 flags, int event, bool unicast);
473
53681407
JP		 474static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
475 u32 chain_index, bool create,
476 bool by_act)
32a4f5ec 477{
c266f64d
VB		 478 struct tcf_chain *chain = NULL;
479 bool is_first_reference;
32a4f5ec 480
c266f64d
VB		 481 mutex_lock(&block->lock);
482 chain = tcf_chain_lookup(block, chain_index);
32a4f5ec
JP		 483 if (chain) {
484 tcf_chain_hold(chain);
53681407
JP		 485 } else {
486 if (!create)
c266f64d 487 goto errout;
53681407
JP		 488 chain = tcf_chain_create(block, chain_index);
489 if (!chain)
c266f64d 490 goto errout;
5bc17018 491 }
80532384 492
53681407
JP		 493 if (by_act)
494 ++chain->action_refcnt;
c266f64d
VB		 495 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
496 mutex_unlock(&block->lock);
53681407
JP		 497
498 /* Send notification only in case we got the first
499 * non-action reference. Until then, the chain acts only as
500 * a placeholder for actions pointing to it and user ought
501 * not know about them.
502 */
c266f64d 503 if (is_first_reference && !by_act)
53681407
JP		 504 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
505 RTM_NEWCHAIN, false);
506
32a4f5ec 507 return chain;
c266f64d
VB		 508
509errout:
510 mutex_unlock(&block->lock);
511 return chain;
5bc17018 512}
53681407 513
290b1c8b
JP		 514static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
515 bool create)
53681407
JP		 516{
517 return __tcf_chain_get(block, chain_index, create, false);
518}
5bc17018 519
1f3ed383
JP		 520struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
521{
53681407 522 return __tcf_chain_get(block, chain_index, true, true);
1f3ed383
JP		 523}
524EXPORT_SYMBOL(tcf_chain_get_by_act);
525
a5654820
VB		 526static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
527 void *tmplt_priv);
528static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
529 void *tmplt_priv, u32 chain_index,
530 struct tcf_block *block, struct sk_buff *oskb,
531 u32 seq, u16 flags, bool unicast);
9f407f17 532
91052fa1
VB		 533static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
534 bool explicitly_created)
5bc17018 535{
c266f64d 536 struct tcf_block *block = chain->block;
a5654820 537 const struct tcf_proto_ops *tmplt_ops;
b62989fc 538 bool free_block = false;
c266f64d 539 unsigned int refcnt;
a5654820 540 void *tmplt_priv;
c266f64d
VB		 541
542 mutex_lock(&block->lock);
91052fa1
VB		 543 if (explicitly_created) {
544 if (!chain->explicitly_created) {
545 mutex_unlock(&block->lock);
546 return;
547 }
548 chain->explicitly_created = false;
549 }
550
53681407
JP		 551 if (by_act)
552 chain->action_refcnt--;
c266f64d
VB		 553
554 /* tc_chain_notify_delete can't be called while holding block lock.
555 * However, when block is unlocked chain can be changed concurrently, so
556 * save these to temporary variables.
557 */
558 refcnt = --chain->refcnt;
a5654820
VB		 559 tmplt_ops = chain->tmplt_ops;
560 tmplt_priv = chain->tmplt_priv;
53681407
JP		 561
562 /* The last dropped non-action reference will trigger notification. */
b62989fc
VB		 563 if (refcnt - chain->action_refcnt == 0 && !by_act) {
564 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
a5654820 565 block, NULL, 0, 0, false);
726d0612
VB		 566 /* Last reference to chain, no need to lock. */
567 chain->flushing = false;
568 }
53681407 569
b62989fc
VB		 570 if (refcnt == 0)
571 free_block = tcf_chain_detach(chain);
572 mutex_unlock(&block->lock);
573
c266f64d 574 if (refcnt == 0) {
a5654820 575 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
c266f64d 576 tcf_chain_destroy(chain, free_block);
32a4f5ec 577 }
5bc17018 578}
53681407 579
290b1c8b 580static void tcf_chain_put(struct tcf_chain *chain)
53681407 581{
91052fa1 582 __tcf_chain_put(chain, false, false);
53681407 583}
5bc17018 584
1f3ed383
JP		 585void tcf_chain_put_by_act(struct tcf_chain *chain)
586{
91052fa1 587 __tcf_chain_put(chain, true, false);
1f3ed383
JP		 588}
589EXPORT_SYMBOL(tcf_chain_put_by_act);
590
32a4f5ec
JP		 591static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
592{
91052fa1 593 __tcf_chain_put(chain, false, true);
32a4f5ec
JP		 594}
595
12db03b6 596static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
290b1c8b 597{
4dbfa766 598 struct tcf_proto *tp, *tp_next;
290b1c8b 599
ed76f5ed
VB		 600 mutex_lock(&chain->filter_chain_lock);
601 tp = tcf_chain_dereference(chain->filter_chain, chain);
59eb87cb
JH		 602 while (tp) {
603 tp_next = rcu_dereference_protected(tp->next, 1);
604 tcf_proto_signal_destroying(chain, tp);
605 tp = tp_next;
606 }
607 tp = tcf_chain_dereference(chain->filter_chain, chain);
4dbfa766 608 RCU_INIT_POINTER(chain->filter_chain, NULL);
290b1c8b 609 tcf_chain0_head_change(chain, NULL);
726d0612 610 chain->flushing = true;
ed76f5ed
VB		 611 mutex_unlock(&chain->filter_chain_lock);
612
290b1c8b 613 while (tp) {
4dbfa766 614 tp_next = rcu_dereference_protected(tp->next, 1);
12db03b6 615 tcf_proto_put(tp, rtnl_held, NULL);
4dbfa766 616 tp = tp_next;
290b1c8b
JP		 617 }
618}
619
59094b1e
PNA		 620static int tcf_block_setup(struct tcf_block *block,
621 struct flow_block_offload *bo);
622
324a823b 623static void tcf_block_offload_init(struct flow_block_offload *bo,
c40f4e50 624 struct net_device *dev, struct Qdisc *sch,
324a823b
PNA		 625 enum flow_block_command command,
626 enum flow_block_binder_type binder_type,
627 struct flow_block *flow_block,
628 bool shared, struct netlink_ext_ack *extack)
629{
630 bo->net = dev_net(dev);
631 bo->command = command;
632 bo->binder_type = binder_type;
633 bo->block = flow_block;
634 bo->block_shared = shared;
635 bo->extack = extack;
c40f4e50 636 bo->sch = sch;
324a823b
PNA		 637 INIT_LIST_HEAD(&bo->cb_list);
638}
639
0fdcf78d
PNA		 640static void tcf_block_unbind(struct tcf_block *block,
641 struct flow_block_offload *bo);
642
643static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
7f76fa36 644{
0fdcf78d
PNA		 645 struct tcf_block *block = block_cb->indr.data;
646 struct net_device *dev = block_cb->indr.dev;
c40f4e50 647 struct Qdisc *sch = block_cb->indr.sch;
0fdcf78d
PNA		 648 struct netlink_ext_ack extack = {};
649 struct flow_block_offload bo;
7f76fa36 650
c40f4e50 651 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
0fdcf78d
PNA		 652 block_cb->indr.binder_type,
653 &block->flow_block, tcf_block_shared(block),
654 &extack);
655 down_write(&block->cb_lock);
a1db2178 656 list_del(&block_cb->driver_list);
0fdcf78d
PNA		 657 list_move(&block_cb->list, &bo.cb_list);
658 up_write(&block->cb_lock);
659 rtnl_lock();
660 tcf_block_unbind(block, &bo);
661 rtnl_unlock();
7f76fa36
JH		 662}
663
caa72601
JP		 664static bool tcf_block_offload_in_use(struct tcf_block *block)
665{
97394bef 666 return atomic_read(&block->offloadcnt);
caa72601
JP		 667}
668
669static int tcf_block_offload_cmd(struct tcf_block *block,
c40f4e50 670 struct net_device *dev, struct Qdisc *sch,
caa72601 671 struct tcf_block_ext_info *ei,
9c0e189e 672 enum flow_block_command command,
60513bd8 673 struct netlink_ext_ack *extack)
8c4083b3 674{
955bcb6e 675 struct flow_block_offload bo = {};
8c4083b3 676
c40f4e50 677 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
324a823b
PNA		 678 &block->flow_block, tcf_block_shared(block),
679 extack);
59094b1e 680
3c005110 681 if (dev->netdev_ops->ndo_setup_tc) {
682 int err;
683
0fdcf78d 684 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
3c005110 685 if (err < 0) {
686 if (err != -EOPNOTSUPP)
687 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
688 return err;
689 }
690
691 return tcf_block_setup(block, &bo);
b70ba69e 692 }
59094b1e 693
c40f4e50 694 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
3c005110 695 tc_block_indr_cleanup);
696 tcf_block_setup(block, &bo);
697
698 return -EOPNOTSUPP;
8c4083b3
JP		 699}
700
caa72601 701static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
60513bd8
JH		 702 struct tcf_block_ext_info *ei,
703 struct netlink_ext_ack *extack)
8c4083b3 704{
caa72601
JP		 705 struct net_device *dev = q->dev_queue->dev;
706 int err;
707
4f8116c8 708 down_write(&block->cb_lock);
caa72601
JP		 709
710 /* If tc offload feature is disabled and the block we try to bind
711 * to already has some offloaded filters, forbid to bind.
712 */
0fdcf78d
PNA		 713 if (dev->netdev_ops->ndo_setup_tc &&
714 !tc_can_offload(dev) &&
715 tcf_block_offload_in_use(block)) {
60513bd8 716 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
4f8116c8
VB		 717 err = -EOPNOTSUPP;
718 goto err_unlock;
60513bd8 719 }
caa72601 720
c40f4e50 721 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
caa72601
JP		 722 if (err == -EOPNOTSUPP)
723 goto no_offload_dev_inc;
7f76fa36 724 if (err)
4f8116c8 725 goto err_unlock;
7f76fa36 726
4f8116c8 727 up_write(&block->cb_lock);
7f76fa36 728 return 0;
caa72601
JP		 729
730no_offload_dev_inc:
0fdcf78d 731 if (tcf_block_offload_in_use(block))
4f8116c8 732 goto err_unlock;
0fdcf78d 733
4f8116c8 734 err = 0;
caa72601 735 block->nooffloaddevcnt++;
4f8116c8
VB		 736err_unlock:
737 up_write(&block->cb_lock);
738 return err;
8c4083b3
JP		 739}
740
741static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
742 struct tcf_block_ext_info *ei)
743{
caa72601
JP		 744 struct net_device *dev = q->dev_queue->dev;
745 int err;
746
4f8116c8 747 down_write(&block->cb_lock);
c40f4e50 748 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
caa72601
JP		 749 if (err == -EOPNOTSUPP)
750 goto no_offload_dev_dec;
4f8116c8 751 up_write(&block->cb_lock);
caa72601
JP		 752 return;
753
754no_offload_dev_dec:
755 WARN_ON(block->nooffloaddevcnt-- == 0);
4f8116c8 756 up_write(&block->cb_lock);
8c4083b3
JP		 757}
758
a9b19443 759static int
f71e0ca4
JP		 760tcf_chain0_head_change_cb_add(struct tcf_block *block,
761 struct tcf_block_ext_info *ei,
762 struct netlink_ext_ack *extack)
a9b19443
JP		 763{
764 struct tcf_filter_chain_list_item *item;
165f0135 765 struct tcf_chain *chain0;
a9b19443
JP		 766
767 item = kmalloc(sizeof(*item), GFP_KERNEL);
768 if (!item) {
769 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
770 return -ENOMEM;
771 }
772 item->chain_head_change = ei->chain_head_change;
773 item->chain_head_change_priv = ei->chain_head_change_priv;
165f0135
VB		 774
775 mutex_lock(&block->lock);
776 chain0 = block->chain0.chain;
ed76f5ed
VB		 777 if (chain0)
778 tcf_chain_hold(chain0);
779 else
780 list_add(&item->list, &block->chain0.filter_chain_list);
165f0135
VB		 781 mutex_unlock(&block->lock);
782
ed76f5ed
VB		 783 if (chain0) {
784 struct tcf_proto *tp_head;
785
786 mutex_lock(&chain0->filter_chain_lock);
787
788 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
789 if (tp_head)
790 tcf_chain_head_change_item(item, tp_head);
791
792 mutex_lock(&block->lock);
793 list_add(&item->list, &block->chain0.filter_chain_list);
794 mutex_unlock(&block->lock);
795
796 mutex_unlock(&chain0->filter_chain_lock);
797 tcf_chain_put(chain0);
798 }
799
a9b19443
JP		 800 return 0;
801}
802
803static void
f71e0ca4
JP		 804tcf_chain0_head_change_cb_del(struct tcf_block *block,
805 struct tcf_block_ext_info *ei)
a9b19443
JP		 806{
807 struct tcf_filter_chain_list_item *item;
808
165f0135 809 mutex_lock(&block->lock);
f71e0ca4 810 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
a9b19443
JP		 811 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
812 (item->chain_head_change == ei->chain_head_change &&
813 item->chain_head_change_priv == ei->chain_head_change_priv)) {
165f0135 814 if (block->chain0.chain)
f71e0ca4 815 tcf_chain_head_change_item(item, NULL);
a9b19443 816 list_del(&item->list);
165f0135
VB		 817 mutex_unlock(&block->lock);
818
a9b19443
JP		 819 kfree(item);
820 return;
821 }
822 }
165f0135 823 mutex_unlock(&block->lock);
a9b19443
JP		 824 WARN_ON(1);
825}
826
48617387 827struct tcf_net {
ab281629 828 spinlock_t idr_lock; /* Protects idr */
48617387
JP		 829 struct idr idr;
830};
831
832static unsigned int tcf_net_id;
833
834static int tcf_block_insert(struct tcf_block *block, struct net *net,
bb047ddd 835 struct netlink_ext_ack *extack)
a9b19443 836{
48617387 837 struct tcf_net *tn = net_generic(net, tcf_net_id);
ab281629
VB		 838 int err;
839
840 idr_preload(GFP_KERNEL);
841 spin_lock(&tn->idr_lock);
842 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
843 GFP_NOWAIT);
844 spin_unlock(&tn->idr_lock);
845 idr_preload_end();
48617387 846
ab281629 847 return err;
a9b19443
JP		 848}
849
48617387
JP		 850static void tcf_block_remove(struct tcf_block *block, struct net *net)
851{
852 struct tcf_net *tn = net_generic(net, tcf_net_id);
853
ab281629 854 spin_lock(&tn->idr_lock);
9c160941 855 idr_remove(&tn->idr, block->index);
ab281629 856 spin_unlock(&tn->idr_lock);
48617387
JP		 857}
858
859static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
bb047ddd 860 u32 block_index,
48617387 861 struct netlink_ext_ack *extack)
6529eaba 862{
48617387 863 struct tcf_block *block;
6529eaba 864
48617387 865 block = kzalloc(sizeof(*block), GFP_KERNEL);
8d1a77f9
AA		 866 if (!block) {
867 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
48617387 868 return ERR_PTR(-ENOMEM);
8d1a77f9 869 }
c266f64d 870 mutex_init(&block->lock);
59eb87cb 871 mutex_init(&block->proto_destroy_lock);
4f8116c8 872 init_rwsem(&block->cb_lock);
14bfb13f 873 flow_block_init(&block->flow_block);
5bc17018 874 INIT_LIST_HEAD(&block->chain_list);
f36fe1c4 875 INIT_LIST_HEAD(&block->owner_list);
f71e0ca4 876 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
acb67442 877
cfebd7e2 878 refcount_set(&block->refcnt, 1);
48617387 879 block->net = net;
bb047ddd
JP		 880 block->index = block_index;
881
882 /* Don't store q pointer for blocks which are shared */
883 if (!tcf_block_shared(block))
884 block->q = q;
48617387 885 return block;
48617387
JP		 886}
887
888static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
889{
890 struct tcf_net *tn = net_generic(net, tcf_net_id);
891
322d884b 892 return idr_find(&tn->idr, block_index);
48617387
JP		 893}
894
0607e439
VB		 895static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
896{
897 struct tcf_block *block;
898
899 rcu_read_lock();
900 block = tcf_block_lookup(net, block_index);
901 if (block && !refcount_inc_not_zero(&block->refcnt))
902 block = NULL;
903 rcu_read_unlock();
904
905 return block;
906}
907
bbf73830
VB		 908static struct tcf_chain *
909__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
f0023436 910{
bbf73830
VB		 911 mutex_lock(&block->lock);
912 if (chain)
913 chain = list_is_last(&chain->list, &block->chain_list) ?
914 NULL : list_next_entry(chain, list);
915 else
916 chain = list_first_entry_or_null(&block->chain_list,
917 struct tcf_chain, list);
f0023436 918
bbf73830
VB		 919 /* skip all action-only chains */
920 while (chain && tcf_chain_held_by_acts_only(chain))
921 chain = list_is_last(&chain->list, &block->chain_list) ?
922 NULL : list_next_entry(chain, list);
923
924 if (chain)
f0023436 925 tcf_chain_hold(chain);
bbf73830 926 mutex_unlock(&block->lock);
f0023436 927
bbf73830 928 return chain;
f0023436
VB		 929}
930
bbf73830
VB		 931/* Function to be used by all clients that want to iterate over all chains on
932 * block. It properly obtains block->lock and takes reference to chain before
933 * returning it. Users of this function must be tolerant to concurrent chain
934 * insertion/deletion or ensure that no concurrent chain modification is
935 * possible. Note that all netlink dump callbacks cannot guarantee to provide
936 * consistent dump because rtnl lock is released each time skb is filled with
937 * data and sent to user-space.
938 */
939
940struct tcf_chain *
941tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
f0023436 942{
bbf73830 943 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
f0023436 944
bbf73830 945 if (chain)
f0023436 946 tcf_chain_put(chain);
bbf73830
VB		 947
948 return chain_next;
949}
950EXPORT_SYMBOL(tcf_get_next_chain);
951
fe2923af
VB		 952static struct tcf_proto *
953__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
954{
8b64678e
VB		 955 u32 prio = 0;
956
fe2923af
VB		 957 ASSERT_RTNL();
958 mutex_lock(&chain->filter_chain_lock);
959
8b64678e 960 if (!tp) {
fe2923af 961 tp = tcf_chain_dereference(chain->filter_chain, chain);
8b64678e
VB		 962 } else if (tcf_proto_is_deleting(tp)) {
963 /* 'deleting' flag is set and chain->filter_chain_lock was
964 * unlocked, which means next pointer could be invalid. Restart
965 * search.
966 */
967 prio = tp->prio + 1;
968 tp = tcf_chain_dereference(chain->filter_chain, chain);
969
970 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
971 if (!tp->deleting && tp->prio >= prio)
972 break;
973 } else {
fe2923af 974 tp = tcf_chain_dereference(tp->next, chain);
8b64678e 975 }
fe2923af
VB		 976
977 if (tp)
978 tcf_proto_get(tp);
979
980 mutex_unlock(&chain->filter_chain_lock);
981
982 return tp;
983}
984
985/* Function to be used by all clients that want to iterate over all tp's on
986 * chain. Users of this function must be tolerant to concurrent tp
987 * insertion/deletion or ensure that no concurrent chain modification is
988 * possible. Note that all netlink dump callbacks cannot guarantee to provide
989 * consistent dump because rtnl lock is released each time skb is filled with
990 * data and sent to user-space.
991 */
992
993struct tcf_proto *
12db03b6
VB		 994tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp,
995 bool rtnl_held)
fe2923af
VB		 996{
997 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
998
999 if (tp)
12db03b6 1000 tcf_proto_put(tp, rtnl_held, NULL);
fe2923af
VB		 1001
1002 return tp_next;
1003}
1004EXPORT_SYMBOL(tcf_get_next_proto);
1005
12db03b6 1006static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
bbf73830
VB		 1007{
1008 struct tcf_chain *chain;
1009
1010 /* Last reference to block. At this point chains cannot be added or
1011 * removed concurrently.
1012 */
1013 for (chain = tcf_get_next_chain(block, NULL);
1014 chain;
1015 chain = tcf_get_next_chain(block, chain)) {
1016 tcf_chain_put_explicitly_created(chain);
12db03b6 1017 tcf_chain_flush(chain, rtnl_held);
f0023436
VB		 1018 }
1019}
1020
18d3eefb
VB		 1021/* Lookup Qdisc and increments its reference counter.
1022 * Set parent, if necessary.
1023 */
1024
1025static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1026 u32 *parent, int ifindex, bool rtnl_held,
1027 struct netlink_ext_ack *extack)
1028{
1029 const struct Qdisc_class_ops *cops;
1030 struct net_device *dev;
1031 int err = 0;
1032
1033 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1034 return 0;
1035
1036 rcu_read_lock();
1037
1038 /* Find link */
1039 dev = dev_get_by_index_rcu(net, ifindex);
1040 if (!dev) {
1041 rcu_read_unlock();
1042 return -ENODEV;
1043 }
1044
1045 /* Find qdisc */
1046 if (!*parent) {
1047 *q = dev->qdisc;
1048 *parent = (*q)->handle;
1049 } else {
1050 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1051 if (!*q) {
1052 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1053 err = -EINVAL;
1054 goto errout_rcu;
1055 }
1056 }
1057
1058 *q = qdisc_refcount_inc_nz(*q);
1059 if (!*q) {
1060 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1061 err = -EINVAL;
1062 goto errout_rcu;
1063 }
1064
1065 /* Is it classful? */
1066 cops = (*q)->ops->cl_ops;
1067 if (!cops) {
1068 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1069 err = -EINVAL;
1070 goto errout_qdisc;
1071 }
1072
1073 if (!cops->tcf_block) {
1074 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1075 err = -EOPNOTSUPP;
1076 goto errout_qdisc;
1077 }
1078
1079errout_rcu:
1080 /* At this point we know that qdisc is not noop_qdisc,
1081 * which means that qdisc holds a reference to net_device
1082 * and we hold a reference to qdisc, so it is safe to release
1083 * rcu read lock.
1084 */
1085 rcu_read_unlock();
1086 return err;
1087
1088errout_qdisc:
1089 rcu_read_unlock();
1090
1091 if (rtnl_held)
1092 qdisc_put(*q);
1093 else
1094 qdisc_put_unlocked(*q);
1095 *q = NULL;
1096
1097 return err;
1098}
1099
1100static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1101 int ifindex, struct netlink_ext_ack *extack)
1102{
1103 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1104 return 0;
1105
1106 /* Do we search for filter, attached to class? */
1107 if (TC_H_MIN(parent)) {
1108 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1109
1110 *cl = cops->find(q, parent);
1111 if (*cl == 0) {
1112 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1113 return -ENOENT;
1114 }
1115 }
1116
1117 return 0;
1118}
1119
1120static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1121 unsigned long cl, int ifindex,
1122 u32 block_index,
1123 struct netlink_ext_ack *extack)
1124{
1125 struct tcf_block *block;
1126
1127 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1128 block = tcf_block_refcnt_get(net, block_index);
1129 if (!block) {
1130 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1131 return ERR_PTR(-EINVAL);
1132 }
1133 } else {
1134 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1135
1136 block = cops->tcf_block(q, cl, extack);
1137 if (!block)
1138 return ERR_PTR(-EINVAL);
1139
1140 if (tcf_block_shared(block)) {
1141 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1142 return ERR_PTR(-EOPNOTSUPP);
1143 }
1144
1145 /* Always take reference to block in order to support execution
1146 * of rules update path of cls API without rtnl lock. Caller
1147 * must release block when it is finished using it. 'if' block
1148 * of this conditional obtain reference to block by calling
1149 * tcf_block_refcnt_get().
1150 */
1151 refcount_inc(&block->refcnt);
1152 }
1153
1154 return block;
1155}
1156
0607e439 1157static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
12db03b6 1158 struct tcf_block_ext_info *ei, bool rtnl_held)
0607e439 1159{
c266f64d 1160 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
0607e439
VB		 1161 /* Flushing/putting all chains will cause the block to be
1162 * deallocated when last chain is freed. However, if chain_list
1163 * is empty, block has to be manually deallocated. After block
1164 * reference counter reached 0, it is no longer possible to
1165 * increment it or add new chains to block.
1166 */
1167 bool free_block = list_empty(&block->chain_list);
1168
c266f64d 1169 mutex_unlock(&block->lock);
0607e439
VB		 1170 if (tcf_block_shared(block))
1171 tcf_block_remove(block, block->net);
0607e439
VB		 1172
1173 if (q)
1174 tcf_block_offload_unbind(block, q, ei);
1175
1176 if (free_block)
c266f64d 1177 tcf_block_destroy(block);
0607e439 1178 else
12db03b6 1179 tcf_block_flush_all_chains(block, rtnl_held);
0607e439
VB		 1180 } else if (q) {
1181 tcf_block_offload_unbind(block, q, ei);
1182 }
1183}
1184
12db03b6 1185static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
0607e439 1186{
12db03b6 1187 __tcf_block_put(block, NULL, NULL, rtnl_held);
0607e439
VB		 1188}
1189
c431f89b
VB		 1190/* Find tcf block.
1191 * Set q, parent, cl when appropriate.
1192 */
1193
1194static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1195 u32 *parent, unsigned long *cl,
1196 int ifindex, u32 block_index,
1197 struct netlink_ext_ack *extack)
1198{
1199 struct tcf_block *block;
e368fdb6 1200 int err = 0;
c431f89b 1201
18d3eefb 1202 ASSERT_RTNL();
e368fdb6 1203
18d3eefb
VB		 1204 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1205 if (err)
1206 goto errout;
c431f89b 1207
18d3eefb
VB		 1208 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1209 if (err)
1210 goto errout_qdisc;
787ce6d0 1211
18d3eefb 1212 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
af736bf0
DC		 1213 if (IS_ERR(block)) {
1214 err = PTR_ERR(block);
18d3eefb 1215 goto errout_qdisc;
af736bf0 1216 }
c431f89b
VB		 1217
1218 return block;
e368fdb6 1219
e368fdb6 1220errout_qdisc:
18d3eefb 1221 if (*q)
e368fdb6 1222 qdisc_put(*q);
18d3eefb
VB		 1223errout:
1224 *q = NULL;
e368fdb6
VB		 1225 return ERR_PTR(err);
1226}
1227
12db03b6
VB		 1228static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1229 bool rtnl_held)
e368fdb6 1230{
787ce6d0 1231 if (!IS_ERR_OR_NULL(block))
12db03b6 1232 tcf_block_refcnt_put(block, rtnl_held);
787ce6d0 1233
470502de
VB		 1234 if (q) {
1235 if (rtnl_held)
1236 qdisc_put(q);
1237 else
1238 qdisc_put_unlocked(q);
1239 }
c431f89b
VB		 1240}
1241
f36fe1c4
JP		 1242struct tcf_block_owner_item {
1243 struct list_head list;
1244 struct Qdisc *q;
32f8c409 1245 enum flow_block_binder_type binder_type;
f36fe1c4
JP		 1246};
1247
1248static void
1249tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1250 struct Qdisc *q,
32f8c409 1251 enum flow_block_binder_type binder_type)
f36fe1c4
JP		 1252{
1253 if (block->keep_dst &&
32f8c409
PNA		 1254 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1255 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
f36fe1c4
JP		 1256 netif_keep_dst(qdisc_dev(q));
1257}
1258
1259void tcf_block_netif_keep_dst(struct tcf_block *block)
1260{
1261 struct tcf_block_owner_item *item;
1262
1263 block->keep_dst = true;
1264 list_for_each_entry(item, &block->owner_list, list)
1265 tcf_block_owner_netif_keep_dst(block, item->q,
1266 item->binder_type);
1267}
1268EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1269
1270static int tcf_block_owner_add(struct tcf_block *block,
1271 struct Qdisc *q,
32f8c409 1272 enum flow_block_binder_type binder_type)
f36fe1c4
JP		 1273{
1274 struct tcf_block_owner_item *item;
1275
1276 item = kmalloc(sizeof(*item), GFP_KERNEL);
1277 if (!item)
1278 return -ENOMEM;
1279 item->q = q;
1280 item->binder_type = binder_type;
1281 list_add(&item->list, &block->owner_list);
1282 return 0;
1283}
1284
1285static void tcf_block_owner_del(struct tcf_block *block,
1286 struct Qdisc *q,
32f8c409 1287 enum flow_block_binder_type binder_type)
f36fe1c4
JP		 1288{
1289 struct tcf_block_owner_item *item;
1290
1291 list_for_each_entry(item, &block->owner_list, list) {
1292 if (item->q == q && item->binder_type == binder_type) {
1293 list_del(&item->list);
1294 kfree(item);
1295 return;
1296 }
1297 }
1298 WARN_ON(1);
1299}
1300
48617387
JP		 1301int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1302 struct tcf_block_ext_info *ei,
1303 struct netlink_ext_ack *extack)
1304{
1305 struct net *net = qdisc_net(q);
1306 struct tcf_block *block = NULL;
48617387
JP		 1307 int err;
1308
787ce6d0 1309 if (ei->block_index)
48617387 1310 /* block_index not 0 means the shared block is requested */
787ce6d0 1311 block = tcf_block_refcnt_get(net, ei->block_index);
48617387
JP		 1312
1313 if (!block) {
bb047ddd 1314 block = tcf_block_create(net, q, ei->block_index, extack);
48617387
JP		 1315 if (IS_ERR(block))
1316 return PTR_ERR(block);
bb047ddd
JP		 1317 if (tcf_block_shared(block)) {
1318 err = tcf_block_insert(block, net, extack);
48617387
JP		 1319 if (err)
1320 goto err_block_insert;
1321 }
1322 }
1323
f36fe1c4
JP		 1324 err = tcf_block_owner_add(block, q, ei->binder_type);
1325 if (err)
1326 goto err_block_owner_add;
1327
1328 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1329
f71e0ca4 1330 err = tcf_chain0_head_change_cb_add(block, ei, extack);
a9b19443 1331 if (err)
f71e0ca4 1332 goto err_chain0_head_change_cb_add;
caa72601 1333
60513bd8 1334 err = tcf_block_offload_bind(block, q, ei, extack);
caa72601
JP		 1335 if (err)
1336 goto err_block_offload_bind;
1337
6529eaba
JP		 1338 *p_block = block;
1339 return 0;
2190d1d0 1340
caa72601 1341err_block_offload_bind:
f71e0ca4
JP		 1342 tcf_chain0_head_change_cb_del(block, ei);
1343err_chain0_head_change_cb_add:
f36fe1c4
JP		 1344 tcf_block_owner_del(block, q, ei->binder_type);
1345err_block_owner_add:
48617387 1346err_block_insert:
12db03b6 1347 tcf_block_refcnt_put(block, true);
2190d1d0 1348 return err;
6529eaba 1349}
8c4083b3
JP		 1350EXPORT_SYMBOL(tcf_block_get_ext);
1351
c7eb7d72
JP		 1352static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1353{
1354 struct tcf_proto __rcu **p_filter_chain = priv;
1355
1356 rcu_assign_pointer(*p_filter_chain, tp_head);
1357}
1358
8c4083b3 1359int tcf_block_get(struct tcf_block **p_block,
8d1a77f9
AA		 1360 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1361 struct netlink_ext_ack *extack)
8c4083b3 1362{
c7eb7d72
JP		 1363 struct tcf_block_ext_info ei = {
1364 .chain_head_change = tcf_chain_head_change_dflt,
1365 .chain_head_change_priv = p_filter_chain,
1366 };
8c4083b3 1367
c7eb7d72 1368 WARN_ON(!p_filter_chain);
8d1a77f9 1369 return tcf_block_get_ext(p_block, q, &ei, extack);
8c4083b3 1370}
6529eaba
JP		 1371EXPORT_SYMBOL(tcf_block_get);
1372
7aa0045d 1373/* XXX: Standalone actions are not allowed to jump to any chain, and bound
a60b3f51 1374 * actions should be all removed after flushing.
7aa0045d 1375 */
c7eb7d72 1376void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
e1ea2f98 1377 struct tcf_block_ext_info *ei)
7aa0045d 1378{
c30abd5e
DM		 1379 if (!block)
1380 return;
f71e0ca4 1381 tcf_chain0_head_change_cb_del(block, ei);
f36fe1c4 1382 tcf_block_owner_del(block, q, ei->binder_type);
a60b3f51 1383
12db03b6 1384 __tcf_block_put(block, q, ei, true);
6529eaba 1385}
8c4083b3
JP		 1386EXPORT_SYMBOL(tcf_block_put_ext);
1387
1388void tcf_block_put(struct tcf_block *block)
1389{
1390 struct tcf_block_ext_info ei = {0, };
1391
4853f128
JP		 1392 if (!block)
1393 return;
c7eb7d72 1394 tcf_block_put_ext(block, block->q, &ei);
8c4083b3 1395}
e1ea2f98 1396
6529eaba 1397EXPORT_SYMBOL(tcf_block_put);
cf1facda 1398
32636742 1399static int
a7323311 1400tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
32636742
JH		 1401 void *cb_priv, bool add, bool offload_in_use,
1402 struct netlink_ext_ack *extack)
1403{
bbf73830 1404 struct tcf_chain *chain, *chain_prev;
fe2923af 1405 struct tcf_proto *tp, *tp_prev;
32636742
JH		 1406 int err;
1407
4f8116c8
VB		 1408 lockdep_assert_held(&block->cb_lock);
1409
bbf73830
VB		 1410 for (chain = __tcf_get_next_chain(block, NULL);
1411 chain;
1412 chain_prev = chain,
1413 chain = __tcf_get_next_chain(block, chain),
1414 tcf_chain_put(chain_prev)) {
fe2923af
VB		 1415 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1416 tp_prev = tp,
1417 tp = __tcf_get_next_proto(chain, tp),
12db03b6 1418 tcf_proto_put(tp_prev, true, NULL)) {
32636742
JH		 1419 if (tp->ops->reoffload) {
1420 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1421 extack);
1422 if (err && add)
1423 goto err_playback_remove;
1424 } else if (add && offload_in_use) {
1425 err = -EOPNOTSUPP;
1426 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1427 goto err_playback_remove;
1428 }
1429 }
1430 }
1431
1432 return 0;
1433
1434err_playback_remove:
12db03b6 1435 tcf_proto_put(tp, true, NULL);
bbf73830 1436 tcf_chain_put(chain);
32636742
JH		 1437 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1438 extack);
1439 return err;
1440}
1441
59094b1e
PNA		 1442static int tcf_block_bind(struct tcf_block *block,
1443 struct flow_block_offload *bo)
1444{
1445 struct flow_block_cb *block_cb, *next;
1446 int err, i = 0;
1447
4f8116c8
VB		 1448 lockdep_assert_held(&block->cb_lock);
1449
59094b1e
PNA		 1450 list_for_each_entry(block_cb, &bo->cb_list, list) {
1451 err = tcf_block_playback_offloads(block, block_cb->cb,
1452 block_cb->cb_priv, true,
1453 tcf_block_offload_in_use(block),
1454 bo->extack);
1455 if (err)
1456 goto err_unroll;
c9f14470
VB		 1457 if (!bo->unlocked_driver_cb)
1458 block->lockeddevcnt++;
59094b1e
PNA		 1459
1460 i++;
1461 }
14bfb13f 1462 list_splice(&bo->cb_list, &block->flow_block.cb_list);
59094b1e
PNA		 1463
1464 return 0;
1465
1466err_unroll:
1467 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1468 if (i-- > 0) {
1469 list_del(&block_cb->list);
1470 tcf_block_playback_offloads(block, block_cb->cb,
1471 block_cb->cb_priv, false,
1472 tcf_block_offload_in_use(block),
1473 NULL);
c9f14470
VB		 1474 if (!bo->unlocked_driver_cb)
1475 block->lockeddevcnt--;
59094b1e
PNA		 1476 }
1477 flow_block_cb_free(block_cb);
1478 }
1479
1480 return err;
1481}
1482
1483static void tcf_block_unbind(struct tcf_block *block,
1484 struct flow_block_offload *bo)
1485{
1486 struct flow_block_cb *block_cb, *next;
1487
4f8116c8
VB		 1488 lockdep_assert_held(&block->cb_lock);
1489
59094b1e
PNA		 1490 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1491 tcf_block_playback_offloads(block, block_cb->cb,
1492 block_cb->cb_priv, false,
1493 tcf_block_offload_in_use(block),
1494 NULL);
1495 list_del(&block_cb->list);
1496 flow_block_cb_free(block_cb);
c9f14470
VB		 1497 if (!bo->unlocked_driver_cb)
1498 block->lockeddevcnt--;
59094b1e
PNA		 1499 }
1500}
1501
1502static int tcf_block_setup(struct tcf_block *block,
1503 struct flow_block_offload *bo)
1504{
1505 int err;
1506
1507 switch (bo->command) {
1508 case FLOW_BLOCK_BIND:
1509 err = tcf_block_bind(block, bo);
1510 break;
1511 case FLOW_BLOCK_UNBIND:
1512 err = 0;
1513 tcf_block_unbind(block, bo);
1514 break;
1515 default:
1516 WARN_ON_ONCE(1);
1517 err = -EOPNOTSUPP;
1518 }
1519
1520 return err;
1521}
1522
87d83093
JP		 1523/* Main classifier routine: scans classifier chain attached
1524 * to this qdisc, (optionally) tests for protocol and asks
1525 * specific classifiers.
1526 */
9410c940
PB		 1527static inline int __tcf_classify(struct sk_buff *skb,
1528 const struct tcf_proto *tp,
af699626 1529 const struct tcf_proto *orig_tp,
9410c940
PB		 1530 struct tcf_result *res,
1531 bool compat_mode,
1532 u32 *last_executed_chain)
87d83093 1533{
87d83093
JP		 1534#ifdef CONFIG_NET_CLS_ACT
1535 const int max_reclassify_loop = 4;
ee538dce 1536 const struct tcf_proto *first_tp;
87d83093
JP		 1537 int limit = 0;
1538
1539reclassify:
1540#endif
1541 for (; tp; tp = rcu_dereference_bh(tp->next)) {
d7bf2ebe 1542 __be16 protocol = skb_protocol(skb, false);
87d83093
JP		 1543 int err;
1544
1545 if (tp->protocol != protocol &&
1546 tp->protocol != htons(ETH_P_ALL))
1547 continue;
1548
1549 err = tp->classify(skb, tp, res);
1550#ifdef CONFIG_NET_CLS_ACT
db50514f 1551 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
ee538dce 1552 first_tp = orig_tp;
9410c940 1553 *last_executed_chain = first_tp->chain->index;
87d83093 1554 goto reset;
db50514f 1555 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
ee538dce 1556 first_tp = res->goto_tp;
9410c940 1557 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
db50514f
JP		 1558 goto reset;
1559 }
87d83093
JP		 1560#endif
1561 if (err >= 0)
1562 return err;
1563 }
1564
1565 return TC_ACT_UNSPEC; /* signal: continue lookup */
1566#ifdef CONFIG_NET_CLS_ACT
1567reset:
1568 if (unlikely(limit++ >= max_reclassify_loop)) {
9d3aaff3
JP		 1569 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1570 tp->chain->block->index,
1571 tp->prio & 0xffff,
87d83093
JP		 1572 ntohs(tp->protocol));
1573 return TC_ACT_SHOT;
1574 }
1575
ee538dce 1576 tp = first_tp;
87d83093
JP		 1577 goto reclassify;
1578#endif
1579}
9410c940
PB		 1580
1581int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1582 struct tcf_result *res, bool compat_mode)
1583{
1584 u32 last_executed_chain = 0;
1585
af699626 1586 return __tcf_classify(skb, tp, tp, res, compat_mode,
9410c940
PB		 1587 &last_executed_chain);
1588}
87d83093
JP		 1589EXPORT_SYMBOL(tcf_classify);
1590
7d17c544
PB		 1591int tcf_classify_ingress(struct sk_buff *skb,
1592 const struct tcf_block *ingress_block,
1593 const struct tcf_proto *tp,
9410c940
PB		 1594 struct tcf_result *res, bool compat_mode)
1595{
1596#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1597 u32 last_executed_chain = 0;
1598
af699626 1599 return __tcf_classify(skb, tp, tp, res, compat_mode,
9410c940
PB		 1600 &last_executed_chain);
1601#else
1602 u32 last_executed_chain = tp ? tp->chain->index : 0;
af699626 1603 const struct tcf_proto *orig_tp = tp;
9410c940
PB		 1604 struct tc_skb_ext *ext;
1605 int ret;
1606
af699626
PB		 1607 ext = skb_ext_find(skb, TC_SKB_EXT);
1608
1609 if (ext && ext->chain) {
1610 struct tcf_chain *fchain;
1611
1612 fchain = tcf_chain_lookup_rcu(ingress_block, ext->chain);
1613 if (!fchain)
1614 return TC_ACT_SHOT;
1615
1616 /* Consume, so cloned/redirect skbs won't inherit ext */
1617 skb_ext_del(skb, TC_SKB_EXT);
1618
1619 tp = rcu_dereference_bh(fchain->filter_chain);
a080da6a 1620 last_executed_chain = fchain->index;
af699626
PB		 1621 }
1622
1623 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1624 &last_executed_chain);
9410c940
PB		 1625
1626 /* If we missed on some chain */
1627 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1628 ext = skb_ext_add(skb, TC_SKB_EXT);
1629 if (WARN_ON_ONCE(!ext))
1630 return TC_ACT_SHOT;
1631 ext->chain = last_executed_chain;
038ebb1a 1632 ext->mru = qdisc_skb_cb(skb)->mru;
9410c940
PB		 1633 }
1634
1635 return ret;
1636#endif
1637}
1638EXPORT_SYMBOL(tcf_classify_ingress);
1639
2190d1d0
JP		 1640struct tcf_chain_info {
1641 struct tcf_proto __rcu **pprev;
1642 struct tcf_proto __rcu *next;
1643};
1644
ed76f5ed
VB		 1645static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1646 struct tcf_chain_info *chain_info)
2190d1d0 1647{
ed76f5ed 1648 return tcf_chain_dereference(*chain_info->pprev, chain);
2190d1d0
JP		 1649}
1650
726d0612
VB		 1651static int tcf_chain_tp_insert(struct tcf_chain *chain,
1652 struct tcf_chain_info *chain_info,
1653 struct tcf_proto *tp)
2190d1d0 1654{
726d0612
VB		 1655 if (chain->flushing)
1656 return -EAGAIN;
1657
c7eb7d72 1658 if (*chain_info->pprev == chain->filter_chain)
f71e0ca4 1659 tcf_chain0_head_change(chain, tp);
4dbfa766 1660 tcf_proto_get(tp);
ed76f5ed 1661 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
2190d1d0 1662 rcu_assign_pointer(*chain_info->pprev, tp);
726d0612
VB		 1663
1664 return 0;
2190d1d0
JP		 1665}
1666
1667static void tcf_chain_tp_remove(struct tcf_chain *chain,
1668 struct tcf_chain_info *chain_info,
1669 struct tcf_proto *tp)
1670{
ed76f5ed 1671 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
2190d1d0 1672
8b64678e 1673 tcf_proto_mark_delete(tp);
c7eb7d72 1674 if (tp == chain->filter_chain)
f71e0ca4 1675 tcf_chain0_head_change(chain, next);
2190d1d0
JP		 1676 RCU_INIT_POINTER(*chain_info->pprev, next);
1677}
1678
8b64678e
VB		 1679static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1680 struct tcf_chain_info *chain_info,
1681 u32 protocol, u32 prio,
1682 bool prio_allocate);
1683
1684/* Try to insert new proto.
1685 * If proto with specified priority already exists, free new proto
1686 * and return existing one.
1687 */
1688
1689static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1690 struct tcf_proto *tp_new,
12db03b6
VB		 1691 u32 protocol, u32 prio,
1692 bool rtnl_held)
8b64678e
VB		 1693{
1694 struct tcf_chain_info chain_info;
1695 struct tcf_proto *tp;
726d0612 1696 int err = 0;
8b64678e
VB		 1697
1698 mutex_lock(&chain->filter_chain_lock);
1699
59eb87cb
JH		 1700 if (tcf_proto_exists_destroying(chain, tp_new)) {
1701 mutex_unlock(&chain->filter_chain_lock);
1702 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1703 return ERR_PTR(-EAGAIN);
1704 }
1705
8b64678e
VB		 1706 tp = tcf_chain_tp_find(chain, &chain_info,
1707 protocol, prio, false);
1708 if (!tp)
726d0612 1709 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
8b64678e
VB		 1710 mutex_unlock(&chain->filter_chain_lock);
1711
1712 if (tp) {
59eb87cb 1713 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
8b64678e 1714 tp_new = tp;
726d0612 1715 } else if (err) {
59eb87cb 1716 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
726d0612 1717 tp_new = ERR_PTR(err);
8b64678e
VB		 1718 }
1719
1720 return tp_new;
1721}
1722
1723static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
12db03b6 1724 struct tcf_proto *tp, bool rtnl_held,
8b64678e
VB		 1725 struct netlink_ext_ack *extack)
1726{
1727 struct tcf_chain_info chain_info;
1728 struct tcf_proto *tp_iter;
1729 struct tcf_proto **pprev;
1730 struct tcf_proto *next;
1731
1732 mutex_lock(&chain->filter_chain_lock);
1733
1734 /* Atomically find and remove tp from chain. */
1735 for (pprev = &chain->filter_chain;
1736 (tp_iter = tcf_chain_dereference(*pprev, chain));
1737 pprev = &tp_iter->next) {
1738 if (tp_iter == tp) {
1739 chain_info.pprev = pprev;
1740 chain_info.next = tp_iter->next;
1741 WARN_ON(tp_iter->deleting);
1742 break;
1743 }
1744 }
1745 /* Verify that tp still exists and no new filters were inserted
1746 * concurrently.
1747 * Mark tp for deletion if it is empty.
1748 */
a5b72a08 1749 if (!tp_iter || !tcf_proto_check_delete(tp)) {
8b64678e
VB		 1750 mutex_unlock(&chain->filter_chain_lock);
1751 return;
1752 }
1753
59eb87cb 1754 tcf_proto_signal_destroying(chain, tp);
8b64678e
VB		 1755 next = tcf_chain_dereference(chain_info.next, chain);
1756 if (tp == chain->filter_chain)
1757 tcf_chain0_head_change(chain, next);
1758 RCU_INIT_POINTER(*chain_info.pprev, next);
1759 mutex_unlock(&chain->filter_chain_lock);
1760
12db03b6 1761 tcf_proto_put(tp, rtnl_held, extack);
8b64678e
VB		 1762}
1763
2190d1d0
JP		 1764static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1765 struct tcf_chain_info *chain_info,
1766 u32 protocol, u32 prio,
1767 bool prio_allocate)
1768{
1769 struct tcf_proto **pprev;
1770 struct tcf_proto *tp;
1771
1772 /* Check the chain for existence of proto-tcf with this priority */
1773 for (pprev = &chain->filter_chain;
ed76f5ed
VB		 1774 (tp = tcf_chain_dereference(*pprev, chain));
1775 pprev = &tp->next) {
2190d1d0
JP		 1776 if (tp->prio >= prio) {
1777 if (tp->prio == prio) {
1778 if (prio_allocate ||
1779 (tp->protocol != protocol && protocol))
1780 return ERR_PTR(-EINVAL);
1781 } else {
1782 tp = NULL;
1783 }
1784 break;
1785 }
1786 }
1787 chain_info->pprev = pprev;
4dbfa766
VB		 1788 if (tp) {
1789 chain_info->next = tp->next;
1790 tcf_proto_get(tp);
1791 } else {
1792 chain_info->next = NULL;
1793 }
2190d1d0
JP		 1794 return tp;
1795}
1796
7120371c 1797static int tcf_fill_node(struct net *net, struct sk_buff *skb,
7960d1da
JP		 1798 struct tcf_proto *tp, struct tcf_block *block,
1799 struct Qdisc *q, u32 parent, void *fh,
12db03b6 1800 u32 portid, u32 seq, u16 flags, int event,
f8ab1807 1801 bool terse_dump, bool rtnl_held)
7120371c
WC		 1802{
1803 struct tcmsg *tcm;
1804 struct nlmsghdr *nlh;
1805 unsigned char *b = skb_tail_pointer(skb);
1806
1807 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1808 if (!nlh)
1809 goto out_nlmsg_trim;
1810 tcm = nlmsg_data(nlh);
1811 tcm->tcm_family = AF_UNSPEC;
1812 tcm->tcm__pad1 = 0;
1813 tcm->tcm__pad2 = 0;
7960d1da
JP		 1814 if (q) {
1815 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1816 tcm->tcm_parent = parent;
1817 } else {
1818 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1819 tcm->tcm_block_index = block->index;
1820 }
7120371c
WC		 1821 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1822 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1823 goto nla_put_failure;
1824 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1825 goto nla_put_failure;
1826 if (!fh) {
1827 tcm->tcm_handle = 0;
f8ab1807
VB		 1828 } else if (terse_dump) {
1829 if (tp->ops->terse_dump) {
1830 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1831 rtnl_held) < 0)
1832 goto nla_put_failure;
1833 } else {
1834 goto cls_op_not_supp;
1835 }
7120371c 1836 } else {
12db03b6
VB		 1837 if (tp->ops->dump &&
1838 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
7120371c
WC		 1839 goto nla_put_failure;
1840 }
1841 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1842 return skb->len;
1843
1844out_nlmsg_trim:
1845nla_put_failure:
f8ab1807 1846cls_op_not_supp:
7120371c
WC		 1847 nlmsg_trim(skb, b);
1848 return -1;
1849}
1850
1851static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1852 struct nlmsghdr *n, struct tcf_proto *tp,
7960d1da 1853 struct tcf_block *block, struct Qdisc *q,
12db03b6
VB		 1854 u32 parent, void *fh, int event, bool unicast,
1855 bool rtnl_held)
7120371c
WC		 1856{
1857 struct sk_buff *skb;
1858 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
5b5f99b1 1859 int err = 0;
7120371c
WC		 1860
1861 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1862 if (!skb)
1863 return -ENOBUFS;
1864
7960d1da 1865 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
12db03b6 1866 n->nlmsg_seq, n->nlmsg_flags, event,
f8ab1807 1867 false, rtnl_held) <= 0) {
7120371c
WC		 1868 kfree_skb(skb);
1869 return -EINVAL;
1870 }
1871
1872 if (unicast)
5b5f99b1
ZW		 1873 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1874 else
1875 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1876 n->nlmsg_flags & NLM_F_ECHO);
7120371c 1877
5b5f99b1
ZW		 1878 if (err > 0)
1879 err = 0;
1880 return err;
7120371c
WC		 1881}
1882
1883static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1884 struct nlmsghdr *n, struct tcf_proto *tp,
7960d1da 1885 struct tcf_block *block, struct Qdisc *q,
c35a4acc 1886 u32 parent, void *fh, bool unicast, bool *last,
12db03b6 1887 bool rtnl_held, struct netlink_ext_ack *extack)
7120371c
WC		 1888{
1889 struct sk_buff *skb;
1890 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1891 int err;
1892
1893 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1894 if (!skb)
1895 return -ENOBUFS;
1896
7960d1da 1897 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
12db03b6 1898 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
f8ab1807 1899 false, rtnl_held) <= 0) {
c35a4acc 1900 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
7120371c
WC		 1901 kfree_skb(skb);
1902 return -EINVAL;
1903 }
1904
12db03b6 1905 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
7120371c
WC		 1906 if (err) {
1907 kfree_skb(skb);
1908 return err;
1909 }
1910
1911 if (unicast)
5b5f99b1
ZW		 1912 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
1913 else
1914 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1915 n->nlmsg_flags & NLM_F_ECHO);
c35a4acc
AA		 1916 if (err < 0)
1917 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
5b5f99b1
ZW		 1918
1919 if (err > 0)
1920 err = 0;
c35a4acc 1921 return err;
7120371c
WC		 1922}
1923
1924static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
7960d1da
JP		 1925 struct tcf_block *block, struct Qdisc *q,
1926 u32 parent, struct nlmsghdr *n,
12db03b6
VB		 1927 struct tcf_chain *chain, int event,
1928 bool rtnl_held)
7120371c
WC		 1929{
1930 struct tcf_proto *tp;
1931
12db03b6
VB		 1932 for (tp = tcf_get_next_proto(chain, NULL, rtnl_held);
1933 tp; tp = tcf_get_next_proto(chain, tp, rtnl_held))
7960d1da 1934 tfilter_notify(net, oskb, n, tp, block,
12db03b6 1935 q, parent, NULL, event, false, rtnl_held);
7120371c
WC		 1936}
1937
7d5509fa
VB		 1938static void tfilter_put(struct tcf_proto *tp, void *fh)
1939{
1940 if (tp->ops->put && fh)
1941 tp->ops->put(tp, fh);
1942}
1943
c431f89b 1944static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
c21ef3e3 1945 struct netlink_ext_ack *extack)
1da177e4 1946{
3b1e0a65 1947 struct net *net = sock_net(skb->sk);
add93b61 1948 struct nlattr *tca[TCA_MAX + 1];
6f96c3c6 1949 char name[IFNAMSIZ];
1da177e4
LT		 1950 struct tcmsg *t;
1951 u32 protocol;
1952 u32 prio;
9d36d9e5 1953 bool prio_allocate;
1da177e4 1954 u32 parent;
5bc17018 1955 u32 chain_index;
7960d1da 1956 struct Qdisc *q = NULL;
2190d1d0 1957 struct tcf_chain_info chain_info;
5bc17018 1958 struct tcf_chain *chain = NULL;
6529eaba 1959 struct tcf_block *block;
1da177e4 1960 struct tcf_proto *tp;
1da177e4 1961 unsigned long cl;
8113c095 1962 void *fh;
1da177e4 1963 int err;
628185cf 1964 int tp_created;
470502de 1965 bool rtnl_held = false;
1da177e4 1966
c431f89b 1967 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
dfc47ef8 1968 return -EPERM;
de179c8c 1969
1da177e4 1970replay:
628185cf
DB		 1971 tp_created = 0;
1972
8cb08174
JB		 1973 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1974 rtm_tca_policy, extack);
de179c8c
H		 1975 if (err < 0)
1976 return err;
1977
942b8165 1978 t = nlmsg_data(n);
1da177e4
LT		 1979 protocol = TC_H_MIN(t->tcm_info);
1980 prio = TC_H_MAJ(t->tcm_info);
9d36d9e5 1981 prio_allocate = false;
1da177e4 1982 parent = t->tcm_parent;
4dbfa766 1983 tp = NULL;
1da177e4 1984 cl = 0;
470502de 1985 block = NULL;
1da177e4
LT		 1986
1987 if (prio == 0) {
c431f89b
VB		 1988 /* If no priority is provided by the user,
1989 * we allocate one.
1990 */
1991 if (n->nlmsg_flags & NLM_F_CREATE) {
1992 prio = TC_H_MAKE(0x80000000U, 0U);
1993 prio_allocate = true;
1994 } else {
c35a4acc 1995 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1da177e4 1996 return -ENOENT;
ea7f8277 1997 }
1da177e4
LT		 1998 }
1999
2000 /* Find head of filter chain. */
2001
470502de
VB		 2002 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2003 if (err)
2004 return err;
2005
6f96c3c6
CW		 2006 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2007 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2008 err = -EINVAL;
2009 goto errout;
2010 }
2011
470502de
VB		 2012 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2013 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2014 * type is not specified, classifier is not unlocked.
2015 */
2016 if (rtnl_held ||
2017 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
6f96c3c6 2018 !tcf_proto_is_unlocked(name)) {
470502de
VB		 2019 rtnl_held = true;
2020 rtnl_lock();
2021 }
2022
2023 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2024 if (err)
2025 goto errout;
2026
2027 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2028 extack);
c431f89b
VB		 2029 if (IS_ERR(block)) {
2030 err = PTR_ERR(block);
2031 goto errout;
6bb16e7a 2032 }
a7df4870 2033 block->classid = parent;
5bc17018
JP		 2034
2035 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2036 if (chain_index > TC_ACT_EXT_VAL_MASK) {
c35a4acc 2037 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
5bc17018
JP		 2038 err = -EINVAL;
2039 goto errout;
2040 }
c431f89b 2041 chain = tcf_chain_get(block, chain_index, true);
5bc17018 2042 if (!chain) {
d5ed72a5 2043 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
c431f89b 2044 err = -ENOMEM;
ea7f8277
DB		 2045 goto errout;
2046 }
1da177e4 2047
ed76f5ed 2048 mutex_lock(&chain->filter_chain_lock);
2190d1d0
JP		 2049 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2050 prio, prio_allocate);
2051 if (IS_ERR(tp)) {
c35a4acc 2052 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2190d1d0 2053 err = PTR_ERR(tp);
ed76f5ed 2054 goto errout_locked;
1da177e4
LT		 2055 }
2056
2057 if (tp == NULL) {
8b64678e
VB		 2058 struct tcf_proto *tp_new = NULL;
2059
726d0612
VB		 2060 if (chain->flushing) {
2061 err = -EAGAIN;
2062 goto errout_locked;
2063 }
2064
1da177e4
LT		 2065 /* Proto-tcf does not exist, create new one */
2066
6bb16e7a 2067 if (tca[TCA_KIND] == NULL || !protocol) {
c35a4acc 2068 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
6bb16e7a 2069 err = -EINVAL;
ed76f5ed 2070 goto errout_locked;
6bb16e7a 2071 }
1da177e4 2072
c431f89b 2073 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
c35a4acc 2074 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
6bb16e7a 2075 err = -ENOENT;
ed76f5ed 2076 goto errout_locked;
6bb16e7a 2077 }
1da177e4 2078
9d36d9e5 2079 if (prio_allocate)
ed76f5ed
VB		 2080 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2081 &chain_info));
1da177e4 2082
ed76f5ed 2083 mutex_unlock(&chain->filter_chain_lock);
36d79af7
ED		 2084 tp_new = tcf_proto_create(name, protocol, prio, chain,
2085 rtnl_held, extack);
8b64678e
VB		 2086 if (IS_ERR(tp_new)) {
2087 err = PTR_ERR(tp_new);
726d0612 2088 goto errout_tp;
1da177e4 2089 }
ed76f5ed 2090
12186be7 2091 tp_created = 1;
12db03b6
VB		 2092 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2093 rtnl_held);
726d0612
VB		 2094 if (IS_ERR(tp)) {
2095 err = PTR_ERR(tp);
2096 goto errout_tp;
2097 }
ed76f5ed
VB		 2098 } else {
2099 mutex_unlock(&chain->filter_chain_lock);
6bb16e7a 2100 }
1da177e4 2101
8b64678e
VB		 2102 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2103 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2104 err = -EINVAL;
2105 goto errout;
2106 }
2107
1da177e4
LT		 2108 fh = tp->ops->get(tp, t->tcm_handle);
2109
8113c095 2110 if (!fh) {
c431f89b 2111 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
c35a4acc 2112 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
6bb16e7a 2113 err = -ENOENT;
1da177e4 2114 goto errout;
6bb16e7a 2115 }
c431f89b 2116 } else if (n->nlmsg_flags & NLM_F_EXCL) {
7d5509fa 2117 tfilter_put(tp, fh);
c431f89b
VB		 2118 NL_SET_ERR_MSG(extack, "Filter already exists");
2119 err = -EEXIST;
2120 goto errout;
1da177e4
LT		 2121 }
2122
9f407f17
JP		 2123 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2124 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2125 err = -EINVAL;
2126 goto errout;
2127 }
2128
2f7ef2f8 2129 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
7306db38 2130 n->nlmsg_flags & NLM_F_CREATE ? TCA_ACT_NOREPLACE : TCA_ACT_REPLACE,
12db03b6 2131 rtnl_held, extack);
7d5509fa 2132 if (err == 0) {
7960d1da 2133 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
12db03b6 2134 RTM_NEWTFILTER, false, rtnl_held);
7d5509fa 2135 tfilter_put(tp, fh);
503d81d4
VB		 2136 /* q pointer is NULL for shared blocks */
2137 if (q)
2138 q->flags &= ~TCQ_F_CAN_BYPASS;
7d5509fa 2139 }
1da177e4
LT		 2140
2141errout:
8b64678e 2142 if (err && tp_created)
12db03b6 2143 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
726d0612 2144errout_tp:
4dbfa766
VB		 2145 if (chain) {
2146 if (tp && !IS_ERR(tp))
12db03b6 2147 tcf_proto_put(tp, rtnl_held, NULL);
4dbfa766
VB		 2148 if (!tp_created)
2149 tcf_chain_put(chain);
2150 }
12db03b6 2151 tcf_block_release(q, block, rtnl_held);
470502de
VB		 2152
2153 if (rtnl_held)
2154 rtnl_unlock();
2155
2156 if (err == -EAGAIN) {
2157 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2158 * of target chain.
2159 */
2160 rtnl_held = true;
1da177e4
LT		 2161 /* Replay the request. */
2162 goto replay;
470502de 2163 }
1da177e4 2164 return err;
ed76f5ed
VB		 2165
2166errout_locked:
2167 mutex_unlock(&chain->filter_chain_lock);
2168 goto errout;
1da177e4
LT		 2169}
2170
c431f89b
VB		 2171static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2172 struct netlink_ext_ack *extack)
2173{
2174 struct net *net = sock_net(skb->sk);
2175 struct nlattr *tca[TCA_MAX + 1];
6f96c3c6 2176 char name[IFNAMSIZ];
c431f89b
VB		 2177 struct tcmsg *t;
2178 u32 protocol;
2179 u32 prio;
2180 u32 parent;
2181 u32 chain_index;
2182 struct Qdisc *q = NULL;
2183 struct tcf_chain_info chain_info;
2184 struct tcf_chain *chain = NULL;
470502de 2185 struct tcf_block *block = NULL;
c431f89b
VB		 2186 struct tcf_proto *tp = NULL;
2187 unsigned long cl = 0;
2188 void *fh = NULL;
2189 int err;
470502de 2190 bool rtnl_held = false;
c431f89b
VB		 2191
2192 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2193 return -EPERM;
2194
8cb08174
JB		 2195 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2196 rtm_tca_policy, extack);
c431f89b
VB		 2197 if (err < 0)
2198 return err;
2199
2200 t = nlmsg_data(n);
2201 protocol = TC_H_MIN(t->tcm_info);
2202 prio = TC_H_MAJ(t->tcm_info);
2203 parent = t->tcm_parent;
2204
2205 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2206 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2207 return -ENOENT;
2208 }
2209
2210 /* Find head of filter chain. */
2211
470502de
VB		 2212 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2213 if (err)
2214 return err;
2215
6f96c3c6
CW		 2216 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2217 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2218 err = -EINVAL;
2219 goto errout;
2220 }
470502de
VB		 2221 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2222 * found), qdisc is not unlocked, classifier type is not specified,
2223 * classifier is not unlocked.
2224 */
2225 if (!prio ||
2226 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
6f96c3c6 2227 !tcf_proto_is_unlocked(name)) {
470502de
VB		 2228 rtnl_held = true;
2229 rtnl_lock();
2230 }
2231
2232 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2233 if (err)
2234 goto errout;
2235
2236 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2237 extack);
c431f89b
VB		 2238 if (IS_ERR(block)) {
2239 err = PTR_ERR(block);
2240 goto errout;
2241 }
2242
2243 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2244 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2245 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2246 err = -EINVAL;
2247 goto errout;
2248 }
2249 chain = tcf_chain_get(block, chain_index, false);
2250 if (!chain) {
5ca8a25c
JP		 2251 /* User requested flush on non-existent chain. Nothing to do,
2252 * so just return success.
2253 */
2254 if (prio == 0) {
2255 err = 0;
2256 goto errout;
2257 }
c431f89b 2258 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
b7b4247d 2259 err = -ENOENT;
c431f89b
VB		 2260 goto errout;
2261 }
2262
2263 if (prio == 0) {
2264 tfilter_notify_chain(net, skb, block, q, parent, n,
12db03b6
VB		 2265 chain, RTM_DELTFILTER, rtnl_held);
2266 tcf_chain_flush(chain, rtnl_held);
c431f89b
VB		 2267 err = 0;
2268 goto errout;
2269 }
2270
ed76f5ed 2271 mutex_lock(&chain->filter_chain_lock);
c431f89b
VB		 2272 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2273 prio, false);
2274 if (!tp || IS_ERR(tp)) {
2275 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
0e399035 2276 err = tp ? PTR_ERR(tp) : -ENOENT;
ed76f5ed 2277 goto errout_locked;
c431f89b
VB		 2278 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2279 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2280 err = -EINVAL;
ed76f5ed
VB		 2281 goto errout_locked;
2282 } else if (t->tcm_handle == 0) {
59eb87cb 2283 tcf_proto_signal_destroying(chain, tp);
ed76f5ed
VB		 2284 tcf_chain_tp_remove(chain, &chain_info, tp);
2285 mutex_unlock(&chain->filter_chain_lock);
2286
12db03b6 2287 tcf_proto_put(tp, rtnl_held, NULL);
ed76f5ed 2288 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
12db03b6 2289 RTM_DELTFILTER, false, rtnl_held);
ed76f5ed 2290 err = 0;
c431f89b
VB		 2291 goto errout;
2292 }
ed76f5ed 2293 mutex_unlock(&chain->filter_chain_lock);
c431f89b
VB		 2294
2295 fh = tp->ops->get(tp, t->tcm_handle);
2296
2297 if (!fh) {
ed76f5ed
VB		 2298 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2299 err = -ENOENT;
c431f89b
VB		 2300 } else {
2301 bool last;
2302
2303 err = tfilter_del_notify(net, skb, n, tp, block,
2304 q, parent, fh, false, &last,
12db03b6
VB		 2305 rtnl_held, extack);
2306
c431f89b
VB		 2307 if (err)
2308 goto errout;
8b64678e 2309 if (last)
12db03b6 2310 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
c431f89b
VB		 2311 }
2312
2313errout:
4dbfa766
VB		 2314 if (chain) {
2315 if (tp && !IS_ERR(tp))
12db03b6 2316 tcf_proto_put(tp, rtnl_held, NULL);
c431f89b 2317 tcf_chain_put(chain);
4dbfa766 2318 }
12db03b6 2319 tcf_block_release(q, block, rtnl_held);
470502de
VB		 2320
2321 if (rtnl_held)
2322 rtnl_unlock();
2323
c431f89b 2324 return err;
ed76f5ed
VB		 2325
2326errout_locked:
2327 mutex_unlock(&chain->filter_chain_lock);
2328 goto errout;
c431f89b
VB		 2329}
2330
2331static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2332 struct netlink_ext_ack *extack)
2333{
2334 struct net *net = sock_net(skb->sk);
2335 struct nlattr *tca[TCA_MAX + 1];
6f96c3c6 2336 char name[IFNAMSIZ];
c431f89b
VB		 2337 struct tcmsg *t;
2338 u32 protocol;
2339 u32 prio;
2340 u32 parent;
2341 u32 chain_index;
2342 struct Qdisc *q = NULL;
2343 struct tcf_chain_info chain_info;
2344 struct tcf_chain *chain = NULL;
470502de 2345 struct tcf_block *block = NULL;
c431f89b
VB		 2346 struct tcf_proto *tp = NULL;
2347 unsigned long cl = 0;
2348 void *fh = NULL;
2349 int err;
470502de 2350 bool rtnl_held = false;
c431f89b 2351
8cb08174
JB		 2352 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2353 rtm_tca_policy, extack);
c431f89b
VB		 2354 if (err < 0)
2355 return err;
2356
2357 t = nlmsg_data(n);
2358 protocol = TC_H_MIN(t->tcm_info);
2359 prio = TC_H_MAJ(t->tcm_info);
2360 parent = t->tcm_parent;
2361
2362 if (prio == 0) {
2363 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2364 return -ENOENT;
2365 }
2366
2367 /* Find head of filter chain. */
2368
470502de
VB		 2369 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2370 if (err)
2371 return err;
2372
6f96c3c6
CW		 2373 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2374 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2375 err = -EINVAL;
2376 goto errout;
2377 }
470502de
VB		 2378 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2379 * unlocked, classifier type is not specified, classifier is not
2380 * unlocked.
2381 */
2382 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
6f96c3c6 2383 !tcf_proto_is_unlocked(name)) {
470502de
VB		 2384 rtnl_held = true;
2385 rtnl_lock();
2386 }
2387
2388 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2389 if (err)
2390 goto errout;
2391
2392 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2393 extack);
c431f89b
VB		 2394 if (IS_ERR(block)) {
2395 err = PTR_ERR(block);
2396 goto errout;
2397 }
2398
2399 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2400 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2401 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2402 err = -EINVAL;
2403 goto errout;
2404 }
2405 chain = tcf_chain_get(block, chain_index, false);
2406 if (!chain) {
2407 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2408 err = -EINVAL;
2409 goto errout;
2410 }
2411
ed76f5ed 2412 mutex_lock(&chain->filter_chain_lock);
c431f89b
VB		 2413 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2414 prio, false);
ed76f5ed 2415 mutex_unlock(&chain->filter_chain_lock);
c431f89b
VB		 2416 if (!tp || IS_ERR(tp)) {
2417 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
0e399035 2418 err = tp ? PTR_ERR(tp) : -ENOENT;
c431f89b
VB		 2419 goto errout;
2420 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2421 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2422 err = -EINVAL;
2423 goto errout;
2424 }
2425
2426 fh = tp->ops->get(tp, t->tcm_handle);
2427
2428 if (!fh) {
2429 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2430 err = -ENOENT;
2431 } else {
2432 err = tfilter_notify(net, skb, n, tp, block, q, parent,
12db03b6 2433 fh, RTM_NEWTFILTER, true, rtnl_held);
c431f89b
VB		 2434 if (err < 0)
2435 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2436 }
2437
7d5509fa 2438 tfilter_put(tp, fh);
c431f89b 2439errout:
4dbfa766
VB		 2440 if (chain) {
2441 if (tp && !IS_ERR(tp))
12db03b6 2442 tcf_proto_put(tp, rtnl_held, NULL);
c431f89b 2443 tcf_chain_put(chain);
4dbfa766 2444 }
12db03b6 2445 tcf_block_release(q, block, rtnl_held);
470502de
VB		 2446
2447 if (rtnl_held)
2448 rtnl_unlock();
2449
c431f89b
VB		 2450 return err;
2451}
2452
aa767bfe 2453struct tcf_dump_args {
1da177e4
LT		 2454 struct tcf_walker w;
2455 struct sk_buff *skb;
2456 struct netlink_callback *cb;
7960d1da 2457 struct tcf_block *block;
a10fa201
JP		 2458 struct Qdisc *q;
2459 u32 parent;
f8ab1807 2460 bool terse_dump;
1da177e4
LT		 2461};
2462
8113c095 2463static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1da177e4 2464{
aa767bfe 2465 struct tcf_dump_args *a = (void *)arg;
832d1d5b 2466 struct net *net = sock_net(a->skb->sk);
1da177e4 2467
7960d1da 2468 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
a10fa201 2469 n, NETLINK_CB(a->cb->skb).portid,
5a7a5555 2470 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
f8ab1807 2471 RTM_NEWTFILTER, a->terse_dump, true);
1da177e4
LT		 2472}
2473
a10fa201
JP		 2474static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2475 struct sk_buff *skb, struct netlink_callback *cb,
f8ab1807 2476 long index_start, long *p_index, bool terse)
acb31fae
JP		 2477{
2478 struct net *net = sock_net(skb->sk);
7960d1da 2479 struct tcf_block *block = chain->block;
acb31fae 2480 struct tcmsg *tcm = nlmsg_data(cb->nlh);
fe2923af 2481 struct tcf_proto *tp, *tp_prev;
acb31fae 2482 struct tcf_dump_args arg;
acb31fae 2483
fe2923af
VB		 2484 for (tp = __tcf_get_next_proto(chain, NULL);
2485 tp;
2486 tp_prev = tp,
2487 tp = __tcf_get_next_proto(chain, tp),
12db03b6 2488 tcf_proto_put(tp_prev, true, NULL),
fe2923af 2489 (*p_index)++) {
acb31fae
JP		 2490 if (*p_index < index_start)
2491 continue;
2492 if (TC_H_MAJ(tcm->tcm_info) &&
2493 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2494 continue;
2495 if (TC_H_MIN(tcm->tcm_info) &&
2496 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2497 continue;
2498 if (*p_index > index_start)
2499 memset(&cb->args[1], 0,
2500 sizeof(cb->args) - sizeof(cb->args[0]));
2501 if (cb->args[1] == 0) {
53189183 2502 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
acb31fae
JP		 2503 NETLINK_CB(cb->skb).portid,
2504 cb->nlh->nlmsg_seq, NLM_F_MULTI,
f8ab1807 2505 RTM_NEWTFILTER, false, true) <= 0)
fe2923af 2506 goto errout;
acb31fae
JP		 2507 cb->args[1] = 1;
2508 }
2509 if (!tp->ops->walk)
2510 continue;
2511 arg.w.fn = tcf_node_dump;
2512 arg.skb = skb;
2513 arg.cb = cb;
7960d1da 2514 arg.block = block;
a10fa201
JP		 2515 arg.q = q;
2516 arg.parent = parent;
acb31fae
JP		 2517 arg.w.stop = 0;
2518 arg.w.skip = cb->args[1] - 1;
2519 arg.w.count = 0;
01683a14 2520 arg.w.cookie = cb->args[2];
f8ab1807 2521 arg.terse_dump = terse;
12db03b6 2522 tp->ops->walk(tp, &arg.w, true);
01683a14 2523 cb->args[2] = arg.w.cookie;
acb31fae
JP		 2524 cb->args[1] = arg.w.count + 1;
2525 if (arg.w.stop)
fe2923af 2526 goto errout;
acb31fae 2527 }
5bc17018 2528 return true;
fe2923af
VB		 2529
2530errout:
12db03b6 2531 tcf_proto_put(tp, true, NULL);
fe2923af 2532 return false;
acb31fae
JP		 2533}
2534
f8ab1807
VB		 2535static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2536 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2537};
2538
bd27a875 2539/* called with RTNL */
1da177e4
LT		 2540static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2541{
bbf73830 2542 struct tcf_chain *chain, *chain_prev;
3b1e0a65 2543 struct net *net = sock_net(skb->sk);
5bc17018 2544 struct nlattr *tca[TCA_MAX + 1];
7960d1da 2545 struct Qdisc *q = NULL;
6529eaba 2546 struct tcf_block *block;
942b8165 2547 struct tcmsg *tcm = nlmsg_data(cb->nlh);
f8ab1807 2548 bool terse_dump = false;
acb31fae
JP		 2549 long index_start;
2550 long index;
a10fa201 2551 u32 parent;
5bc17018 2552 int err;
1da177e4 2553
573ce260 2554 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1da177e4 2555 return skb->len;
5bc17018 2556
8cb08174 2557 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
f8ab1807 2558 tcf_tfilter_dump_policy, cb->extack);
5bc17018
JP		 2559 if (err)
2560 return err;
2561
f8ab1807
VB		 2562 if (tca[TCA_DUMP_FLAGS]) {
2563 struct nla_bitfield32 flags =
2564 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2565
2566 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2567 }
2568
7960d1da 2569 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
787ce6d0 2570 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
7960d1da
JP		 2571 if (!block)
2572 goto out;
d680b352
JP		 2573 /* If we work with block index, q is NULL and parent value
2574 * will never be used in the following code. The check
2575 * in tcf_fill_node prevents it. However, compiler does not
2576 * see that far, so set parent to zero to silence the warning
2577 * about parent being uninitialized.
2578 */
2579 parent = 0;
a10fa201 2580 } else {
7960d1da
JP		 2581 const struct Qdisc_class_ops *cops;
2582 struct net_device *dev;
2583 unsigned long cl = 0;
2584
2585 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2586 if (!dev)
2587 return skb->len;
2588
2589 parent = tcm->tcm_parent;
a7df4870 2590 if (!parent)
7960d1da 2591 q = dev->qdisc;
a7df4870 2592 else
7960d1da 2593 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
7960d1da
JP		 2594 if (!q)
2595 goto out;
2596 cops = q->ops->cl_ops;
2597 if (!cops)
143976ce 2598 goto out;
7960d1da
JP		 2599 if (!cops->tcf_block)
2600 goto out;
2601 if (TC_H_MIN(tcm->tcm_parent)) {
2602 cl = cops->find(q, tcm->tcm_parent);
2603 if (cl == 0)
2604 goto out;
2605 }
2606 block = cops->tcf_block(q, cl, NULL);
2607 if (!block)
2608 goto out;
a7df4870 2609 parent = block->classid;
7960d1da
JP		 2610 if (tcf_block_shared(block))
2611 q = NULL;
1da177e4 2612 }
1da177e4 2613
acb31fae
JP		 2614 index_start = cb->args[0];
2615 index = 0;
5bc17018 2616
bbf73830
VB		 2617 for (chain = __tcf_get_next_chain(block, NULL);
2618 chain;
2619 chain_prev = chain,
2620 chain = __tcf_get_next_chain(block, chain),
2621 tcf_chain_put(chain_prev)) {
5bc17018
JP		 2622 if (tca[TCA_CHAIN] &&
2623 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2624 continue;
a10fa201 2625 if (!tcf_chain_dump(chain, q, parent, skb, cb,
f8ab1807 2626 index_start, &index, terse_dump)) {
bbf73830 2627 tcf_chain_put(chain);
5ae437ad 2628 err = -EMSGSIZE;
5bc17018 2629 break;
5ae437ad 2630 }
5bc17018
JP		 2631 }
2632
787ce6d0 2633 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
12db03b6 2634 tcf_block_refcnt_put(block, true);
acb31fae 2635 cb->args[0] = index;
1da177e4 2636
1da177e4 2637out:
5ae437ad
RK		 2638 /* If we did no progress, the error (EMSGSIZE) is real */
2639 if (skb->len == 0 && err)
2640 return err;
1da177e4
LT		 2641 return skb->len;
2642}
2643
a5654820
VB		 2644static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2645 void *tmplt_priv, u32 chain_index,
2646 struct net *net, struct sk_buff *skb,
2647 struct tcf_block *block,
32a4f5ec
JP		 2648 u32 portid, u32 seq, u16 flags, int event)
2649{
2650 unsigned char *b = skb_tail_pointer(skb);
9f407f17 2651 const struct tcf_proto_ops *ops;
32a4f5ec
JP		 2652 struct nlmsghdr *nlh;
2653 struct tcmsg *tcm;
9f407f17
JP		 2654 void *priv;
2655
a5654820
VB		 2656 ops = tmplt_ops;
2657 priv = tmplt_priv;
32a4f5ec
JP		 2658
2659 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2660 if (!nlh)
2661 goto out_nlmsg_trim;
2662 tcm = nlmsg_data(nlh);
2663 tcm->tcm_family = AF_UNSPEC;
2664 tcm->tcm__pad1 = 0;
2665 tcm->tcm__pad2 = 0;
2666 tcm->tcm_handle = 0;
2667 if (block->q) {
2668 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2669 tcm->tcm_parent = block->q->handle;
2670 } else {
2671 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2672 tcm->tcm_block_index = block->index;
2673 }
2674
a5654820 2675 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
32a4f5ec
JP		 2676 goto nla_put_failure;
2677
9f407f17
JP		 2678 if (ops) {
2679 if (nla_put_string(skb, TCA_KIND, ops->kind))
2680 goto nla_put_failure;
2681 if (ops->tmplt_dump(skb, net, priv) < 0)
2682 goto nla_put_failure;
2683 }
2684
32a4f5ec
JP		 2685 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2686 return skb->len;
2687
2688out_nlmsg_trim:
2689nla_put_failure:
2690 nlmsg_trim(skb, b);
2691 return -EMSGSIZE;
2692}
2693
2694static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2695 u32 seq, u16 flags, int event, bool unicast)
2696{
2697 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2698 struct tcf_block *block = chain->block;
2699 struct net *net = block->net;
2700 struct sk_buff *skb;
5b5f99b1 2701 int err = 0;
32a4f5ec
JP		 2702
2703 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2704 if (!skb)
2705 return -ENOBUFS;
2706
a5654820
VB		 2707 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2708 chain->index, net, skb, block, portid,
32a4f5ec
JP		 2709 seq, flags, event) <= 0) {
2710 kfree_skb(skb);
2711 return -EINVAL;
2712 }
2713
2714 if (unicast)
5b5f99b1
ZW		 2715 err = netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2716 else
2717 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2718 flags & NLM_F_ECHO);
32a4f5ec 2719
5b5f99b1
ZW		 2720 if (err > 0)
2721 err = 0;
2722 return err;
32a4f5ec
JP		 2723}
2724
a5654820
VB		 2725static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2726 void *tmplt_priv, u32 chain_index,
2727 struct tcf_block *block, struct sk_buff *oskb,
2728 u32 seq, u16 flags, bool unicast)
2729{
2730 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2731 struct net *net = block->net;
2732 struct sk_buff *skb;
2733
2734 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2735 if (!skb)
2736 return -ENOBUFS;
2737
2738 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2739 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2740 kfree_skb(skb);
2741 return -EINVAL;
2742 }
2743
2744 if (unicast)
2745 return netlink_unicast(net->rtnl, skb, portid, MSG_DONTWAIT);
2746
2747 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2748}
2749
9f407f17
JP		 2750static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2751 struct nlattr **tca,
2752 struct netlink_ext_ack *extack)
2753{
2754 const struct tcf_proto_ops *ops;
2dd5616e 2755 char name[IFNAMSIZ];
9f407f17
JP		 2756 void *tmplt_priv;
2757
2758 /* If kind is not set, user did not specify template. */
2759 if (!tca[TCA_KIND])
2760 return 0;
2761
2dd5616e
ED		 2762 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2763 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2764 return -EINVAL;
2765 }
2766
2767 ops = tcf_proto_lookup_ops(name, true, extack);
9f407f17
JP		 2768 if (IS_ERR(ops))
2769 return PTR_ERR(ops);
2770 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2771 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2772 return -EOPNOTSUPP;
2773 }
2774
2775 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2776 if (IS_ERR(tmplt_priv)) {
2777 module_put(ops->owner);
2778 return PTR_ERR(tmplt_priv);
2779 }
2780 chain->tmplt_ops = ops;
2781 chain->tmplt_priv = tmplt_priv;
2782 return 0;
2783}
2784
a5654820
VB		 2785static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2786 void *tmplt_priv)
9f407f17 2787{
9f407f17 2788 /* If template ops are set, no work to do for us. */
a5654820 2789 if (!tmplt_ops)
9f407f17
JP		 2790 return;
2791
a5654820
VB		 2792 tmplt_ops->tmplt_destroy(tmplt_priv);
2793 module_put(tmplt_ops->owner);
9f407f17
JP		 2794}
2795
32a4f5ec
JP		 2796/* Add/delete/get a chain */
2797
2798static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2799 struct netlink_ext_ack *extack)
2800{
2801 struct net *net = sock_net(skb->sk);
2802 struct nlattr *tca[TCA_MAX + 1];
2803 struct tcmsg *t;
2804 u32 parent;
2805 u32 chain_index;
2806 struct Qdisc *q = NULL;
2807 struct tcf_chain *chain = NULL;
2808 struct tcf_block *block;
2809 unsigned long cl;
2810 int err;
2811
2812 if (n->nlmsg_type != RTM_GETCHAIN &&
2813 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2814 return -EPERM;
2815
2816replay:
8cb08174
JB		 2817 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2818 rtm_tca_policy, extack);
32a4f5ec
JP		 2819 if (err < 0)
2820 return err;
2821
2822 t = nlmsg_data(n);
2823 parent = t->tcm_parent;
2824 cl = 0;
2825
2826 block = tcf_block_find(net, &q, &parent, &cl,
2827 t->tcm_ifindex, t->tcm_block_index, extack);
2828 if (IS_ERR(block))
2829 return PTR_ERR(block);
2830
2831 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2832 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2833 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
e368fdb6
VB		 2834 err = -EINVAL;
2835 goto errout_block;
32a4f5ec 2836 }
2cbfab07
VB		 2837
2838 mutex_lock(&block->lock);
32a4f5ec
JP		 2839 chain = tcf_chain_lookup(block, chain_index);
2840 if (n->nlmsg_type == RTM_NEWCHAIN) {
2841 if (chain) {
3d32f4c5 2842 if (tcf_chain_held_by_acts_only(chain)) {
1f3ed383 2843 /* The chain exists only because there is
3d32f4c5 2844 * some action referencing it.
1f3ed383
JP		 2845 */
2846 tcf_chain_hold(chain);
2847 } else {
2848 NL_SET_ERR_MSG(extack, "Filter chain already exists");
e368fdb6 2849 err = -EEXIST;
2cbfab07 2850 goto errout_block_locked;
1f3ed383
JP		 2851 }
2852 } else {
2853 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2854 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
e368fdb6 2855 err = -ENOENT;
2cbfab07 2856 goto errout_block_locked;
1f3ed383
JP		 2857 }
2858 chain = tcf_chain_create(block, chain_index);
2859 if (!chain) {
2860 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
e368fdb6 2861 err = -ENOMEM;
2cbfab07 2862 goto errout_block_locked;
1f3ed383 2863 }
32a4f5ec
JP		 2864 }
2865 } else {
3d32f4c5 2866 if (!chain || tcf_chain_held_by_acts_only(chain)) {
32a4f5ec 2867 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
e368fdb6 2868 err = -EINVAL;
2cbfab07 2869 goto errout_block_locked;
32a4f5ec
JP		 2870 }
2871 tcf_chain_hold(chain);
2872 }
2873
2cbfab07
VB		 2874 if (n->nlmsg_type == RTM_NEWCHAIN) {
2875 /* Modifying chain requires holding parent block lock. In case
2876 * the chain was successfully added, take a reference to the
2877 * chain. This ensures that an empty chain does not disappear at
2878 * the end of this function.
2879 */
2880 tcf_chain_hold(chain);
2881 chain->explicitly_created = true;
2882 }
2883 mutex_unlock(&block->lock);
2884
32a4f5ec
JP		 2885 switch (n->nlmsg_type) {
2886 case RTM_NEWCHAIN:
9f407f17 2887 err = tc_chain_tmplt_add(chain, net, tca, extack);
2cbfab07
VB		 2888 if (err) {
2889 tcf_chain_put_explicitly_created(chain);
9f407f17 2890 goto errout;
2cbfab07
VB		 2891 }
2892
32a4f5ec
JP		 2893 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2894 RTM_NEWCHAIN, false);
2895 break;
2896 case RTM_DELCHAIN:
f5b9bac7 2897 tfilter_notify_chain(net, skb, block, q, parent, n,
12db03b6 2898 chain, RTM_DELTFILTER, true);
32a4f5ec 2899 /* Flush the chain first as the user requested chain removal. */
12db03b6 2900 tcf_chain_flush(chain, true);
32a4f5ec
JP		 2901 /* In case the chain was successfully deleted, put a reference
2902 * to the chain previously taken during addition.
2903 */
2904 tcf_chain_put_explicitly_created(chain);
2905 break;
2906 case RTM_GETCHAIN:
32a4f5ec
JP		 2907 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
2908 n->nlmsg_seq, n->nlmsg_type, true);
2909 if (err < 0)
2910 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2911 break;
2912 default:
2913 err = -EOPNOTSUPP;
2914 NL_SET_ERR_MSG(extack, "Unsupported message type");
2915 goto errout;
2916 }
2917
2918errout:
2919 tcf_chain_put(chain);
e368fdb6 2920errout_block:
12db03b6 2921 tcf_block_release(q, block, true);
32a4f5ec
JP		 2922 if (err == -EAGAIN)
2923 /* Replay the request. */
2924 goto replay;
2925 return err;
2cbfab07
VB		 2926
2927errout_block_locked:
2928 mutex_unlock(&block->lock);
2929 goto errout_block;
32a4f5ec
JP		 2930}
2931
2932/* called with RTNL */
2933static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2934{
2935 struct net *net = sock_net(skb->sk);
2936 struct nlattr *tca[TCA_MAX + 1];
2937 struct Qdisc *q = NULL;
2938 struct tcf_block *block;
32a4f5ec 2939 struct tcmsg *tcm = nlmsg_data(cb->nlh);
ace4a267 2940 struct tcf_chain *chain;
32a4f5ec
JP		 2941 long index_start;
2942 long index;
2943 u32 parent;
2944 int err;
2945
2946 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2947 return skb->len;
2948
8cb08174
JB		 2949 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2950 rtm_tca_policy, cb->extack);
32a4f5ec
JP		 2951 if (err)
2952 return err;
2953
2954 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
787ce6d0 2955 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
32a4f5ec
JP		 2956 if (!block)
2957 goto out;
2958 /* If we work with block index, q is NULL and parent value
2959 * will never be used in the following code. The check
2960 * in tcf_fill_node prevents it. However, compiler does not
2961 * see that far, so set parent to zero to silence the warning
2962 * about parent being uninitialized.
2963 */
2964 parent = 0;
2965 } else {
2966 const struct Qdisc_class_ops *cops;
2967 struct net_device *dev;
2968 unsigned long cl = 0;
2969
2970 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2971 if (!dev)
2972 return skb->len;
2973
2974 parent = tcm->tcm_parent;
2975 if (!parent) {
2976 q = dev->qdisc;
2977 parent = q->handle;
2978 } else {
2979 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2980 }
2981 if (!q)
2982 goto out;
2983 cops = q->ops->cl_ops;
2984 if (!cops)
2985 goto out;
2986 if (!cops->tcf_block)
2987 goto out;
2988 if (TC_H_MIN(tcm->tcm_parent)) {
2989 cl = cops->find(q, tcm->tcm_parent);
2990 if (cl == 0)
2991 goto out;
2992 }
2993 block = cops->tcf_block(q, cl, NULL);
2994 if (!block)
2995 goto out;
2996 if (tcf_block_shared(block))
2997 q = NULL;
2998 }
2999
3000 index_start = cb->args[0];
3001 index = 0;
3002
ace4a267
VB		 3003 mutex_lock(&block->lock);
3004 list_for_each_entry(chain, &block->chain_list, list) {
32a4f5ec
JP		 3005 if ((tca[TCA_CHAIN] &&
3006 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3007 continue;
3008 if (index < index_start) {
3009 index++;
3010 continue;
3011 }
ace4a267
VB		 3012 if (tcf_chain_held_by_acts_only(chain))
3013 continue;
a5654820
VB		 3014 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3015 chain->index, net, skb, block,
32a4f5ec
JP		 3016 NETLINK_CB(cb->skb).portid,
3017 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3018 RTM_NEWCHAIN);
ace4a267 3019 if (err <= 0)
32a4f5ec
JP		 3020 break;
3021 index++;
3022 }
ace4a267 3023 mutex_unlock(&block->lock);
32a4f5ec 3024
787ce6d0 3025 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
12db03b6 3026 tcf_block_refcnt_put(block, true);
32a4f5ec
JP		 3027 cb->args[0] = index;
3028
3029out:
3030 /* If we did no progress, the error (EMSGSIZE) is real */
3031 if (skb->len == 0 && err)
3032 return err;
3033 return skb->len;
3034}
3035
18d0264f 3036void tcf_exts_destroy(struct tcf_exts *exts)
1da177e4
LT		 3037{
3038#ifdef CONFIG_NET_CLS_ACT
3d66b89c
ED		 3039 if (exts->actions) {
3040 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3041 kfree(exts->actions);
3042 }
22dc13c8 3043 exts->nr_actions = 0;
1da177e4
LT		 3044#endif
3045}
aa767bfe 3046EXPORT_SYMBOL(tcf_exts_destroy);
1da177e4 3047
c1b52739 3048int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
50a56190 3049 struct nlattr *rate_tlv, struct tcf_exts *exts, bool ovr,
ec6743a1 3050 bool rtnl_held, struct netlink_ext_ack *extack)
1da177e4 3051{
1da177e4
LT		 3052#ifdef CONFIG_NET_CLS_ACT
3053 {
1da177e4 3054 struct tc_action *act;
d04e6990 3055 size_t attr_size = 0;
1da177e4 3056
5da57f42 3057 if (exts->police && tb[exts->police]) {
9fb9f251
JP		 3058 act = tcf_action_init_1(net, tp, tb[exts->police],
3059 rate_tlv, "police", ovr,
ec6743a1
VB		 3060 TCA_ACT_BIND, rtnl_held,
3061 extack);
ab27cfb8
PM		 3062 if (IS_ERR(act))
3063 return PTR_ERR(act);
1da177e4 3064
33be6271 3065 act->type = exts->type = TCA_OLD_COMPAT;
22dc13c8
WC		 3066 exts->actions[0] = act;
3067 exts->nr_actions = 1;
5da57f42 3068 } else if (exts->action && tb[exts->action]) {
90b73b77 3069 int err;
22dc13c8 3070
9fb9f251
JP		 3071 err = tcf_action_init(net, tp, tb[exts->action],
3072 rate_tlv, NULL, ovr, TCA_ACT_BIND,
ec6743a1
VB		 3073 exts->actions, &attr_size,
3074 rtnl_held, extack);
90b73b77 3075 if (err < 0)
33be6271 3076 return err;
90b73b77 3077 exts->nr_actions = err;
1da177e4
LT		 3078 }
3079 }
1da177e4 3080#else
5da57f42 3081 if ((exts->action && tb[exts->action]) ||
50a56190
AA		 3082 (exts->police && tb[exts->police])) {
3083 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
1da177e4 3084 return -EOPNOTSUPP;
50a56190 3085 }
1da177e4
LT		 3086#endif
3087
3088 return 0;
3089}
aa767bfe 3090EXPORT_SYMBOL(tcf_exts_validate);
1da177e4 3091
9b0d4446 3092void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
1da177e4
LT		 3093{
3094#ifdef CONFIG_NET_CLS_ACT
22dc13c8
WC		 3095 struct tcf_exts old = *dst;
3096
9b0d4446 3097 *dst = *src;
22dc13c8 3098 tcf_exts_destroy(&old);
1da177e4
LT		 3099#endif
3100}
aa767bfe 3101EXPORT_SYMBOL(tcf_exts_change);
1da177e4 3102
22dc13c8
WC		 3103#ifdef CONFIG_NET_CLS_ACT
3104static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3105{
3106 if (exts->nr_actions == 0)
3107 return NULL;
3108 else
3109 return exts->actions[0];
3110}
3111#endif
33be6271 3112
5da57f42 3113int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
1da177e4
LT		 3114{
3115#ifdef CONFIG_NET_CLS_ACT
9cc63db5
CW		 3116 struct nlattr *nest;
3117
978dfd8d 3118 if (exts->action && tcf_exts_has_actions(exts)) {
1da177e4
LT		 3119 /*
3120 * again for backward compatible mode - we want
3121 * to work with both old and new modes of entering
3122 * tc data even if iproute2 was newer - jhs
3123 */
33be6271 3124 if (exts->type != TCA_OLD_COMPAT) {
ae0be8de 3125 nest = nla_nest_start_noflag(skb, exts->action);
4b3550ef
PM		 3126 if (nest == NULL)
3127 goto nla_put_failure;
22dc13c8 3128
ca44b738
VB		 3129 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3130 < 0)
add93b61 3131 goto nla_put_failure;
4b3550ef 3132 nla_nest_end(skb, nest);
5da57f42 3133 } else if (exts->police) {
33be6271 3134 struct tc_action *act = tcf_exts_first_act(exts);
ae0be8de 3135 nest = nla_nest_start_noflag(skb, exts->police);
63acd680 3136 if (nest == NULL || !act)
4b3550ef 3137 goto nla_put_failure;
33be6271 3138 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
add93b61 3139 goto nla_put_failure;
4b3550ef 3140 nla_nest_end(skb, nest);
1da177e4
LT		 3141 }
3142 }
1da177e4 3143 return 0;
9cc63db5
CW		 3144
3145nla_put_failure:
3146 nla_nest_cancel(skb, nest);
1da177e4 3147 return -1;
9cc63db5
CW		 3148#else
3149 return 0;
3150#endif
1da177e4 3151}
aa767bfe 3152EXPORT_SYMBOL(tcf_exts_dump);
1da177e4 3153
ca44b738
VB		 3154int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3155{
3156#ifdef CONFIG_NET_CLS_ACT
3157 struct nlattr *nest;
3158
3159 if (!exts->action || !tcf_exts_has_actions(exts))
3160 return 0;
3161
3162 nest = nla_nest_start_noflag(skb, exts->action);
3163 if (!nest)
3164 goto nla_put_failure;
3165
3166 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3167 goto nla_put_failure;
3168 nla_nest_end(skb, nest);
3169 return 0;
3170
3171nla_put_failure:
3172 nla_nest_cancel(skb, nest);
3173 return -1;
3174#else
3175 return 0;
3176#endif
3177}
3178EXPORT_SYMBOL(tcf_exts_terse_dump);
aa767bfe 3179
5da57f42 3180int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1da177e4
LT		 3181{
3182#ifdef CONFIG_NET_CLS_ACT
33be6271 3183 struct tc_action *a = tcf_exts_first_act(exts);
b057df24 3184 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
33be6271 3185 return -1;
1da177e4
LT		 3186#endif
3187 return 0;
1da177e4 3188}
aa767bfe 3189EXPORT_SYMBOL(tcf_exts_dump_stats);
1da177e4 3190
40119211
VB		 3191static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3192{
3193 if (*flags & TCA_CLS_FLAGS_IN_HW)
3194 return;
3195 *flags |= TCA_CLS_FLAGS_IN_HW;
3196 atomic_inc(&block->offloadcnt);
3197}
3198
3199static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3200{
3201 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3202 return;
3203 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3204 atomic_dec(&block->offloadcnt);
3205}
3206
3207static void tc_cls_offload_cnt_update(struct tcf_block *block,
3208 struct tcf_proto *tp, u32 *cnt,
3209 u32 *flags, u32 diff, bool add)
3210{
3211 lockdep_assert_held(&block->cb_lock);
3212
3213 spin_lock(&tp->lock);
3214 if (add) {
3215 if (!*cnt)
3216 tcf_block_offload_inc(block, flags);
3217 *cnt += diff;
3218 } else {
3219 *cnt -= diff;
3220 if (!*cnt)
3221 tcf_block_offload_dec(block, flags);
3222 }
3223 spin_unlock(&tp->lock);
3224}
3225
3226static void
3227tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3228 u32 *cnt, u32 *flags)
3229{
3230 lockdep_assert_held(&block->cb_lock);
3231
3232 spin_lock(&tp->lock);
3233 tcf_block_offload_dec(block, flags);
3234 *cnt = 0;
3235 spin_unlock(&tp->lock);
3236}
3237
3238static int
3239__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3240 void *type_data, bool err_stop)
717503b9 3241{
955bcb6e 3242 struct flow_block_cb *block_cb;
aeb3fecd
CW		 3243 int ok_count = 0;
3244 int err;
3245
14bfb13f 3246 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
aeb3fecd
CW		 3247 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3248 if (err) {
40119211
VB		 3249 if (err_stop)
3250 return err;
aeb3fecd
CW		 3251 } else {
3252 ok_count++;
3253 }
3254 }
40119211
VB		 3255 return ok_count;
3256}
3257
3258int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3259 void *type_data, bool err_stop, bool rtnl_held)
3260{
11bd634d 3261 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
40119211
VB		 3262 int ok_count;
3263
11bd634d
VB		 3264retry:
3265 if (take_rtnl)
3266 rtnl_lock();
40119211 3267 down_read(&block->cb_lock);
11bd634d
VB		 3268 /* Need to obtain rtnl lock if block is bound to devs that require it.
3269 * In block bind code cb_lock is obtained while holding rtnl, so we must
3270 * obtain the locks in same order here.
3271 */
3272 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3273 up_read(&block->cb_lock);
3274 take_rtnl = true;
3275 goto retry;
3276 }
3277
40119211 3278 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
11bd634d 3279
4f8116c8 3280 up_read(&block->cb_lock);
11bd634d
VB		 3281 if (take_rtnl)
3282 rtnl_unlock();
aeb3fecd 3283 return ok_count;
717503b9
JP		 3284}
3285EXPORT_SYMBOL(tc_setup_cb_call);
b3f55bdd 3286
40119211
VB		 3287/* Non-destructive filter add. If filter that wasn't already in hardware is
3288 * successfully offloaded, increment block offloads counter. On failure,
3289 * previously offloaded filter is considered to be intact and offloads counter
3290 * is not decremented.
3291 */
3292
3293int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3294 enum tc_setup_type type, void *type_data, bool err_stop,
3295 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3296{
11bd634d 3297 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
40119211
VB		 3298 int ok_count;
3299
11bd634d
VB		 3300retry:
3301 if (take_rtnl)
3302 rtnl_lock();
40119211 3303 down_read(&block->cb_lock);
11bd634d
VB		 3304 /* Need to obtain rtnl lock if block is bound to devs that require it.
3305 * In block bind code cb_lock is obtained while holding rtnl, so we must
3306 * obtain the locks in same order here.
3307 */
3308 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3309 up_read(&block->cb_lock);
3310 take_rtnl = true;
3311 goto retry;
3312 }
3313
40119211
VB		 3314 /* Make sure all netdevs sharing this block are offload-capable. */
3315 if (block->nooffloaddevcnt && err_stop) {
3316 ok_count = -EOPNOTSUPP;
3317 goto err_unlock;
3318 }
3319
3320 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
a449a3e7
VB		 3321 if (ok_count < 0)
3322 goto err_unlock;
3323
3324 if (tp->ops->hw_add)
3325 tp->ops->hw_add(tp, type_data);
40119211
VB		 3326 if (ok_count > 0)
3327 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3328 ok_count, true);
3329err_unlock:
3330 up_read(&block->cb_lock);
11bd634d
VB		 3331 if (take_rtnl)
3332 rtnl_unlock();
40119211
VB		 3333 return ok_count < 0 ? ok_count : 0;
3334}
3335EXPORT_SYMBOL(tc_setup_cb_add);
3336
3337/* Destructive filter replace. If filter that wasn't already in hardware is
3338 * successfully offloaded, increment block offload counter. On failure,
3339 * previously offloaded filter is considered to be destroyed and offload counter
3340 * is decremented.
3341 */
3342
3343int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3344 enum tc_setup_type type, void *type_data, bool err_stop,
3345 u32 *old_flags, unsigned int *old_in_hw_count,
3346 u32 *new_flags, unsigned int *new_in_hw_count,
3347 bool rtnl_held)
3348{
11bd634d 3349 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
40119211
VB		 3350 int ok_count;
3351
11bd634d
VB		 3352retry:
3353 if (take_rtnl)
3354 rtnl_lock();
40119211 3355 down_read(&block->cb_lock);
11bd634d
VB		 3356 /* Need to obtain rtnl lock if block is bound to devs that require it.
3357 * In block bind code cb_lock is obtained while holding rtnl, so we must
3358 * obtain the locks in same order here.
3359 */
3360 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3361 up_read(&block->cb_lock);
3362 take_rtnl = true;
3363 goto retry;
3364 }
3365
40119211
VB		 3366 /* Make sure all netdevs sharing this block are offload-capable. */
3367 if (block->nooffloaddevcnt && err_stop) {
3368 ok_count = -EOPNOTSUPP;
3369 goto err_unlock;
3370 }
3371
3372 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
a449a3e7
VB		 3373 if (tp->ops->hw_del)
3374 tp->ops->hw_del(tp, type_data);
40119211
VB		 3375
3376 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
a449a3e7
VB		 3377 if (ok_count < 0)
3378 goto err_unlock;
3379
3380 if (tp->ops->hw_add)
3381 tp->ops->hw_add(tp, type_data);
40119211 3382 if (ok_count > 0)
a449a3e7
VB		 3383 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3384 new_flags, ok_count, true);
40119211
VB		 3385err_unlock:
3386 up_read(&block->cb_lock);
11bd634d
VB		 3387 if (take_rtnl)
3388 rtnl_unlock();
40119211
VB		 3389 return ok_count < 0 ? ok_count : 0;
3390}
3391EXPORT_SYMBOL(tc_setup_cb_replace);
3392
3393/* Destroy filter and decrement block offload counter, if filter was previously
3394 * offloaded.
3395 */
3396
3397int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3398 enum tc_setup_type type, void *type_data, bool err_stop,
3399 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3400{
11bd634d 3401 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
40119211
VB		 3402 int ok_count;
3403
11bd634d
VB		 3404retry:
3405 if (take_rtnl)
3406 rtnl_lock();
40119211 3407 down_read(&block->cb_lock);
11bd634d
VB		 3408 /* Need to obtain rtnl lock if block is bound to devs that require it.
3409 * In block bind code cb_lock is obtained while holding rtnl, so we must
3410 * obtain the locks in same order here.
3411 */
3412 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3413 up_read(&block->cb_lock);
3414 take_rtnl = true;
3415 goto retry;
3416 }
3417
40119211
VB		 3418 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3419
3420 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
a449a3e7
VB		 3421 if (tp->ops->hw_del)
3422 tp->ops->hw_del(tp, type_data);
3423
40119211 3424 up_read(&block->cb_lock);
11bd634d
VB		 3425 if (take_rtnl)
3426 rtnl_unlock();
40119211
VB		 3427 return ok_count < 0 ? ok_count : 0;
3428}
3429EXPORT_SYMBOL(tc_setup_cb_destroy);
3430
3431int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3432 bool add, flow_setup_cb_t *cb,
3433 enum tc_setup_type type, void *type_data,
3434 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3435{
3436 int err = cb(type, type_data, cb_priv);
3437
3438 if (err) {
3439 if (add && tc_skip_sw(*flags))
3440 return err;
3441 } else {
3442 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3443 add);
3444 }
3445
3446 return 0;
3447}
3448EXPORT_SYMBOL(tc_setup_cb_reoffload);
3449
2008495d
JP		 3450static int tcf_act_get_cookie(struct flow_action_entry *entry,
3451 const struct tc_action *act)
3452{
3453 struct tc_cookie *cookie;
3454 int err = 0;
3455
3456 rcu_read_lock();
3457 cookie = rcu_dereference(act->act_cookie);
3458 if (cookie) {
3459 entry->cookie = flow_action_cookie_create(cookie->data,
3460 cookie->len,
3461 GFP_ATOMIC);
3462 if (!entry->cookie)
3463 err = -ENOMEM;
3464 }
3465 rcu_read_unlock();
3466 return err;
3467}
3468
3469static void tcf_act_put_cookie(struct flow_action_entry *entry)
3470{
3471 flow_action_cookie_destroy(entry->cookie);
3472}
3473
5a6ff4b1
VB		 3474void tc_cleanup_flow_action(struct flow_action *flow_action)
3475{
3476 struct flow_action_entry *entry;
3477 int i;
3478
2008495d
JP		 3479 flow_action_for_each(i, entry, flow_action) {
3480 tcf_act_put_cookie(entry);
1158958a
VB		 3481 if (entry->destructor)
3482 entry->destructor(entry->destructor_priv);
2008495d 3483 }
5a6ff4b1
VB		 3484}
3485EXPORT_SYMBOL(tc_cleanup_flow_action);
3486
1158958a
VB		 3487static void tcf_mirred_get_dev(struct flow_action_entry *entry,
3488 const struct tc_action *act)
3489{
470d5060
VB		 3490#ifdef CONFIG_NET_CLS_ACT
3491 entry->dev = act->ops->get_dev(act, &entry->destructor);
1158958a
VB		 3492 if (!entry->dev)
3493 return;
1158958a 3494 entry->destructor_priv = entry->dev;
470d5060 3495#endif
1158958a
VB		 3496}
3497
3498static void tcf_tunnel_encap_put_tunnel(void *priv)
3499{
3500 struct ip_tunnel_info *tunnel = priv;
3501
3502 kfree(tunnel);
3503}
3504
3505static int tcf_tunnel_encap_get_tunnel(struct flow_action_entry *entry,
3506 const struct tc_action *act)
3507{
3508 entry->tunnel = tcf_tunnel_info_copy(act);
3509 if (!entry->tunnel)
3510 return -ENOMEM;
3511 entry->destructor = tcf_tunnel_encap_put_tunnel;
3512 entry->destructor_priv = entry->tunnel;
3513 return 0;
3514}
3515
4a5da47d
VB		 3516static void tcf_sample_get_group(struct flow_action_entry *entry,
3517 const struct tc_action *act)
3518{
3519#ifdef CONFIG_NET_CLS_ACT
3520 entry->sample.psample_group =
3521 act->ops->get_psample_group(act, &entry->destructor);
3522 entry->destructor_priv = entry->sample.psample_group;
3523#endif
3524}
3525
d29bdd69
PL		 3526static void tcf_gate_entry_destructor(void *priv)
3527{
3528 struct action_gate_entry *oe = priv;
3529
3530 kfree(oe);
3531}
3532
3533static int tcf_gate_get_entries(struct flow_action_entry *entry,
3534 const struct tc_action *act)
3535{
3536 entry->gate.entries = tcf_gate_get_list(act);
3537
3538 if (!entry->gate.entries)
3539 return -EINVAL;
3540
3541 entry->destructor = tcf_gate_entry_destructor;
3542 entry->destructor_priv = entry->gate.entries;
3543
3544 return 0;
3545}
3546
16f80360
PNA		 3547static enum flow_action_hw_stats tc_act_hw_stats(u8 hw_stats)
3548{
3549 if (WARN_ON_ONCE(hw_stats > TCA_ACT_HW_STATS_ANY))
3550 return FLOW_ACTION_HW_STATS_DONT_CARE;
3551 else if (!hw_stats)
3552 return FLOW_ACTION_HW_STATS_DISABLED;
3553
3554 return hw_stats;
3555}
3556
3a7b6861 3557int tc_setup_flow_action(struct flow_action *flow_action,
b15e7a6e 3558 const struct tcf_exts *exts)
3a7b6861 3559{
7a472814 3560 struct tc_action *act;
9838b20a 3561 int i, j, k, err = 0;
3a7b6861 3562
0dfb2d82
JK		 3563 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3564 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3565 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
44f86580 3566
3a7b6861
PNA		 3567 if (!exts)
3568 return 0;
3569
3570 j = 0;
3571 tcf_exts_for_each_action(i, act, exts) {
3572 struct flow_action_entry *entry;
3573
3574 entry = &flow_action->entries[j];
7a472814 3575 spin_lock_bh(&act->tcfa_lock);
2008495d
JP		 3576 err = tcf_act_get_cookie(entry, act);
3577 if (err)
3578 goto err_out_locked;
44f86580 3579
16f80360 3580 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
44f86580 3581
3a7b6861
PNA		 3582 if (is_tcf_gact_ok(act)) {
3583 entry->id = FLOW_ACTION_ACCEPT;
3584 } else if (is_tcf_gact_shot(act)) {
3585 entry->id = FLOW_ACTION_DROP;
3586 } else if (is_tcf_gact_trap(act)) {
3587 entry->id = FLOW_ACTION_TRAP;
3588 } else if (is_tcf_gact_goto_chain(act)) {
3589 entry->id = FLOW_ACTION_GOTO;
3590 entry->chain_index = tcf_gact_goto_chain_index(act);
3591 } else if (is_tcf_mirred_egress_redirect(act)) {
3592 entry->id = FLOW_ACTION_REDIRECT;
1158958a 3593 tcf_mirred_get_dev(entry, act);
3a7b6861
PNA		 3594 } else if (is_tcf_mirred_egress_mirror(act)) {
3595 entry->id = FLOW_ACTION_MIRRED;
1158958a 3596 tcf_mirred_get_dev(entry, act);
48e584ac
JH		 3597 } else if (is_tcf_mirred_ingress_redirect(act)) {
3598 entry->id = FLOW_ACTION_REDIRECT_INGRESS;
1158958a 3599 tcf_mirred_get_dev(entry, act);
48e584ac
JH		 3600 } else if (is_tcf_mirred_ingress_mirror(act)) {
3601 entry->id = FLOW_ACTION_MIRRED_INGRESS;
1158958a 3602 tcf_mirred_get_dev(entry, act);
3a7b6861
PNA		 3603 } else if (is_tcf_vlan(act)) {
3604 switch (tcf_vlan_action(act)) {
3605 case TCA_VLAN_ACT_PUSH:
3606 entry->id = FLOW_ACTION_VLAN_PUSH;
3607 entry->vlan.vid = tcf_vlan_push_vid(act);
3608 entry->vlan.proto = tcf_vlan_push_proto(act);
3609 entry->vlan.prio = tcf_vlan_push_prio(act);
3610 break;
3611 case TCA_VLAN_ACT_POP:
3612 entry->id = FLOW_ACTION_VLAN_POP;
3613 break;
3614 case TCA_VLAN_ACT_MODIFY:
3615 entry->id = FLOW_ACTION_VLAN_MANGLE;
3616 entry->vlan.vid = tcf_vlan_push_vid(act);
3617 entry->vlan.proto = tcf_vlan_push_proto(act);
3618 entry->vlan.prio = tcf_vlan_push_prio(act);
3619 break;
3620 default:
9838b20a 3621 err = -EOPNOTSUPP;
7a472814 3622 goto err_out_locked;
3a7b6861
PNA		 3623 }
3624 } else if (is_tcf_tunnel_set(act)) {
3625 entry->id = FLOW_ACTION_TUNNEL_ENCAP;
1158958a
VB		 3626 err = tcf_tunnel_encap_get_tunnel(entry, act);
3627 if (err)
7a472814 3628 goto err_out_locked;
3a7b6861
PNA		 3629 } else if (is_tcf_tunnel_release(act)) {
3630 entry->id = FLOW_ACTION_TUNNEL_DECAP;
3a7b6861
PNA		 3631 } else if (is_tcf_pedit(act)) {
3632 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
3633 switch (tcf_pedit_cmd(act, k)) {
3634 case TCA_PEDIT_KEY_EX_CMD_SET:
3635 entry->id = FLOW_ACTION_MANGLE;
3636 break;
3637 case TCA_PEDIT_KEY_EX_CMD_ADD:
3638 entry->id = FLOW_ACTION_ADD;
3639 break;
3640 default:
9838b20a 3641 err = -EOPNOTSUPP;
7a472814 3642 goto err_out_locked;
3a7b6861
PNA		 3643 }
3644 entry->mangle.htype = tcf_pedit_htype(act, k);
3645 entry->mangle.mask = tcf_pedit_mask(act, k);
3646 entry->mangle.val = tcf_pedit_val(act, k);
3647 entry->mangle.offset = tcf_pedit_offset(act, k);
16f80360 3648 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
2c4b58dc 3649 entry = &flow_action->entries[++j];
3a7b6861
PNA		 3650 }
3651 } else if (is_tcf_csum(act)) {
3652 entry->id = FLOW_ACTION_CSUM;
3653 entry->csum_flags = tcf_csum_update_flags(act);
3654 } else if (is_tcf_skbedit_mark(act)) {
3655 entry->id = FLOW_ACTION_MARK;
3656 entry->mark = tcf_skbedit_mark(act);
a7a7be60
PJV		 3657 } else if (is_tcf_sample(act)) {
3658 entry->id = FLOW_ACTION_SAMPLE;
a7a7be60
PJV		 3659 entry->sample.trunc_size = tcf_sample_trunc_size(act);
3660 entry->sample.truncate = tcf_sample_truncate(act);
3661 entry->sample.rate = tcf_sample_rate(act);
4a5da47d 3662 tcf_sample_get_group(entry, act);
8c8cfc6e
PJV		 3663 } else if (is_tcf_police(act)) {
3664 entry->id = FLOW_ACTION_POLICE;
5f035af7 3665 entry->police.burst = tcf_police_burst(act);
8c8cfc6e
PJV		 3666 entry->police.rate_bytes_ps =
3667 tcf_police_rate_bytes_ps(act);
19e528dc 3668 entry->police.mtu = tcf_police_tcfp_mtu(act);
627e39b1 3669 entry->police.index = act->tcfa_index;
b57dc7c1
PB		 3670 } else if (is_tcf_ct(act)) {
3671 entry->id = FLOW_ACTION_CT;
3672 entry->ct.action = tcf_ct_action(act);
3673 entry->ct.zone = tcf_ct_zone(act);
edd5861e 3674 entry->ct.flow_table = tcf_ct_ft(act);
6749d590
JH		 3675 } else if (is_tcf_mpls(act)) {
3676 switch (tcf_mpls_action(act)) {
3677 case TCA_MPLS_ACT_PUSH:
3678 entry->id = FLOW_ACTION_MPLS_PUSH;
3679 entry->mpls_push.proto = tcf_mpls_proto(act);
3680 entry->mpls_push.label = tcf_mpls_label(act);
3681 entry->mpls_push.tc = tcf_mpls_tc(act);
3682 entry->mpls_push.bos = tcf_mpls_bos(act);
3683 entry->mpls_push.ttl = tcf_mpls_ttl(act);
3684 break;
3685 case TCA_MPLS_ACT_POP:
3686 entry->id = FLOW_ACTION_MPLS_POP;
3687 entry->mpls_pop.proto = tcf_mpls_proto(act);
3688 break;
3689 case TCA_MPLS_ACT_MODIFY:
3690 entry->id = FLOW_ACTION_MPLS_MANGLE;
3691 entry->mpls_mangle.label = tcf_mpls_label(act);
3692 entry->mpls_mangle.tc = tcf_mpls_tc(act);
3693 entry->mpls_mangle.bos = tcf_mpls_bos(act);
3694 entry->mpls_mangle.ttl = tcf_mpls_ttl(act);
3695 break;
3696 default:
7a472814 3697 goto err_out_locked;
6749d590 3698 }
fb1b775a
JH		 3699 } else if (is_tcf_skbedit_ptype(act)) {
3700 entry->id = FLOW_ACTION_PTYPE;
3701 entry->ptype = tcf_skbedit_ptype(act);
2ce12410
PM		 3702 } else if (is_tcf_skbedit_priority(act)) {
3703 entry->id = FLOW_ACTION_PRIORITY;
3704 entry->priority = tcf_skbedit_priority(act);
d29bdd69
PL		 3705 } else if (is_tcf_gate(act)) {
3706 entry->id = FLOW_ACTION_GATE;
3707 entry->gate.index = tcf_gate_index(act);
3708 entry->gate.prio = tcf_gate_prio(act);
3709 entry->gate.basetime = tcf_gate_basetime(act);
3710 entry->gate.cycletime = tcf_gate_cycletime(act);
3711 entry->gate.cycletimeext = tcf_gate_cycletimeext(act);
3712 entry->gate.num_entries = tcf_gate_num_entries(act);
3713 err = tcf_gate_get_entries(entry, act);
3714 if (err)
3715 goto err_out;
3a7b6861 3716 } else {
9838b20a 3717 err = -EOPNOTSUPP;
7a472814 3718 goto err_out_locked;
3a7b6861 3719 }
7a472814 3720 spin_unlock_bh(&act->tcfa_lock);
3a7b6861
PNA		 3721
3722 if (!is_tcf_pedit(act))
3723 j++;
3724 }
9838b20a 3725
3a7b6861 3726err_out:
5a6ff4b1
VB		 3727 if (err)
3728 tc_cleanup_flow_action(flow_action);
3729
9838b20a 3730 return err;
7a472814
VB		 3731err_out_locked:
3732 spin_unlock_bh(&act->tcfa_lock);
3733 goto err_out;
3a7b6861
PNA		 3734}
3735EXPORT_SYMBOL(tc_setup_flow_action);
3736
e3ab786b
PNA		 3737unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3738{
3739 unsigned int num_acts = 0;
3740 struct tc_action *act;
3741 int i;
3742
3743 tcf_exts_for_each_action(i, act, exts) {
3744 if (is_tcf_pedit(act))
3745 num_acts += tcf_pedit_nkeys(act);
3746 else
3747 num_acts++;
3748 }
3749 return num_acts;
3750}
3751EXPORT_SYMBOL(tcf_exts_num_actions);
3752
3625750f
PM		 3753#ifdef CONFIG_NET_CLS_ACT
3754static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3755 u32 *p_block_index,
3756 struct netlink_ext_ack *extack)
3757{
3758 *p_block_index = nla_get_u32(block_index_attr);
3759 if (!*p_block_index) {
3760 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3761 return -EINVAL;
3762 }
3763
3764 return 0;
3765}
3766
3767int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3768 enum flow_block_binder_type binder_type,
3769 struct nlattr *block_index_attr,
3770 struct netlink_ext_ack *extack)
3771{
3772 u32 block_index;
3773 int err;
3774
3775 if (!block_index_attr)
3776 return 0;
3777
3778 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3779 if (err)
3780 return err;
3781
3782 if (!block_index)
3783 return 0;
3784
3785 qe->info.binder_type = binder_type;
3786 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3787 qe->info.chain_head_change_priv = &qe->filter_chain;
3788 qe->info.block_index = block_index;
3789
3790 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3791}
3792EXPORT_SYMBOL(tcf_qevent_init);
3793
3794void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3795{
3796 if (qe->info.block_index)
3797 tcf_block_put_ext(qe->block, sch, &qe->info);
3798}
3799EXPORT_SYMBOL(tcf_qevent_destroy);
3800
3801int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3802 struct netlink_ext_ack *extack)
3803{
3804 u32 block_index;
3805 int err;
3806
3807 if (!block_index_attr)
3808 return 0;
3809
3810 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3811 if (err)
3812 return err;
3813
3814 /* Bounce newly-configured block or change in block. */
3815 if (block_index != qe->info.block_index) {
3816 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3817 return -EINVAL;
3818 }
3819
3820 return 0;
3821}
3822EXPORT_SYMBOL(tcf_qevent_validate_change);
3823
3824struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
55f656cd 3825 struct sk_buff **to_free, int *ret)
3625750f
PM		 3826{
3827 struct tcf_result cl_res;
3828 struct tcf_proto *fl;
3829
3830 if (!qe->info.block_index)
3831 return skb;
3832
3833 fl = rcu_dereference_bh(qe->filter_chain);
3834
3625750f
PM		 3835 switch (tcf_classify(skb, fl, &cl_res, false)) {
3836 case TC_ACT_SHOT:
3837 qdisc_qstats_drop(sch);
3838 __qdisc_drop(skb, to_free);
3839 *ret = __NET_XMIT_BYPASS;
3840 return NULL;
3841 case TC_ACT_STOLEN:
3842 case TC_ACT_QUEUED:
3843 case TC_ACT_TRAP:
3844 __qdisc_drop(skb, to_free);
3845 *ret = __NET_XMIT_STOLEN;
3846 return NULL;
3847 case TC_ACT_REDIRECT:
3848 skb_do_redirect(skb);
3849 *ret = __NET_XMIT_STOLEN;
3850 return NULL;
3851 }
3852
3625750f
PM		 3853 return skb;
3854}
3855EXPORT_SYMBOL(tcf_qevent_handle);
3856
3857int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3858{
3859 if (!qe->info.block_index)
3860 return 0;
3861 return nla_put_u32(skb, attr_name, qe->info.block_index);
3862}
3863EXPORT_SYMBOL(tcf_qevent_dump);
3864#endif
3865
48617387
JP		 3866static __net_init int tcf_net_init(struct net *net)
3867{
3868 struct tcf_net *tn = net_generic(net, tcf_net_id);
3869
ab281629 3870 spin_lock_init(&tn->idr_lock);
48617387
JP		 3871 idr_init(&tn->idr);
3872 return 0;
3873}
3874
3875static void __net_exit tcf_net_exit(struct net *net)
3876{
3877 struct tcf_net *tn = net_generic(net, tcf_net_id);
3878
3879 idr_destroy(&tn->idr);
3880}
3881
3882static struct pernet_operations tcf_net_ops = {
3883 .init = tcf_net_init,
3884 .exit = tcf_net_exit,
3885 .id = &tcf_net_id,
3886 .size = sizeof(struct tcf_net),
3887};
3888
1da177e4
LT		 3889static int __init tc_filter_init(void)
3890{
48617387
JP		 3891 int err;
3892
7aa0045d
CW		 3893 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3894 if (!tc_filter_wq)
3895 return -ENOMEM;
3896
48617387
JP		 3897 err = register_pernet_subsys(&tcf_net_ops);
3898 if (err)
3899 goto err_register_pernet_subsys;
3900
470502de
VB		 3901 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3902 RTNL_FLAG_DOIT_UNLOCKED);
3903 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3904 RTNL_FLAG_DOIT_UNLOCKED);
c431f89b 3905 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
470502de 3906 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
32a4f5ec
JP		 3907 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3908 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3909 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3910 tc_dump_chain, 0);
1da177e4 3911
1da177e4 3912 return 0;
48617387
JP		 3913
3914err_register_pernet_subsys:
3915 destroy_workqueue(tc_filter_wq);
3916 return err;
1da177e4
LT		 3917}
3918
3919subsys_initcall(tc_filter_init);
Ubuntu Jammy Linux kernel mirror