projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
[linux-block.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
35d39fec
PB		 52#ifdef CONFIG_NET_CLS_ACT
53DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
54EXPORT_SYMBOL(tc_skb_ext_tc);
55
56void tc_skb_ext_tc_enable(void)
57{
58 static_branch_inc(&tc_skb_ext_tc);
59}
60EXPORT_SYMBOL(tc_skb_ext_tc_enable);
61
62void tc_skb_ext_tc_disable(void)
63{
64 static_branch_dec(&tc_skb_ext_tc);
65}
66EXPORT_SYMBOL(tc_skb_ext_tc_disable);
67#endif
68
59eb87cb
JH		 69static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
70{
71 return jhash_3words(tp->chain->index, tp->prio,
72 (__force __u32)tp->protocol, 0);
73}
74
75static void tcf_proto_signal_destroying(struct tcf_chain *chain,
76 struct tcf_proto *tp)
77{
78 struct tcf_block *block = chain->block;
79
80 mutex_lock(&block->proto_destroy_lock);
81 hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
82 destroy_obj_hashfn(tp));
83 mutex_unlock(&block->proto_destroy_lock);
84}
85
86static bool tcf_proto_cmp(const struct tcf_proto *tp1,
87 const struct tcf_proto *tp2)
88{
89 return tp1->chain->index == tp2->chain->index &&
90 tp1->prio == tp2->prio &&
91 tp1->protocol == tp2->protocol;
92}
93
94static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
95 struct tcf_proto *tp)
96{
97 u32 hash = destroy_obj_hashfn(tp);
98 struct tcf_proto *iter;
99 bool found = false;
100
101 rcu_read_lock();
102 hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
103 destroy_ht_node, hash) {
104 if (tcf_proto_cmp(tp, iter)) {
105 found = true;
106 break;
107 }
108 }
109 rcu_read_unlock();
110
111 return found;
112}
113
114static void
115tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
116{
117 struct tcf_block *block = chain->block;
118
119 mutex_lock(&block->proto_destroy_lock);
120 if (hash_hashed(&tp->destroy_ht_node))
121 hash_del_rcu(&tp->destroy_ht_node);
122 mutex_unlock(&block->proto_destroy_lock);
123}
124
1da177e4
LT		 125/* Find classifier type by string name */
126
f34e8bff 127static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
1da177e4 128{
dcd76081 129 const struct tcf_proto_ops *t, *res = NULL;
1da177e4
LT		 130
131 if (kind) {
132 read_lock(&cls_mod_lock);
36272874 133 list_for_each_entry(t, &tcf_proto_base, head) {
33a48927 134 if (strcmp(kind, t->kind) == 0) {
dcd76081
ED		 135 if (try_module_get(t->owner))
136 res = t;
1da177e4
LT		 137 break;
138 }
139 }
140 read_unlock(&cls_mod_lock);
141 }
dcd76081 142 return res;
1da177e4
LT		 143}
144
f34e8bff 145static const struct tcf_proto_ops *
12db03b6
VB		 146tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
147 struct netlink_ext_ack *extack)
f34e8bff
JP		 148{
149 const struct tcf_proto_ops *ops;
150
151 ops = __tcf_proto_lookup_ops(kind);
152 if (ops)
153 return ops;
154#ifdef CONFIG_MODULES
12db03b6
VB		 155 if (rtnl_held)
156 rtnl_unlock();
f34e8bff 157 request_module("cls_%s", kind);
12db03b6
VB		 158 if (rtnl_held)
159 rtnl_lock();
f34e8bff
JP		 160 ops = __tcf_proto_lookup_ops(kind);
161 /* We dropped the RTNL semaphore in order to perform
162 * the module load. So, even if we succeeded in loading
163 * the module we have to replay the request. We indicate
164 * this using -EAGAIN.
165 */
166 if (ops) {
167 module_put(ops->owner);
168 return ERR_PTR(-EAGAIN);
169 }
170#endif
171 NL_SET_ERR_MSG(extack, "TC classifier not found");
172 return ERR_PTR(-ENOENT);
173}
174
1da177e4
LT		 175/* Register(unregister) new classifier type */
176
177int register_tcf_proto_ops(struct tcf_proto_ops *ops)
178{
36272874 179 struct tcf_proto_ops *t;
1da177e4
LT		 180 int rc = -EEXIST;
181
182 write_lock(&cls_mod_lock);
36272874 183 list_for_each_entry(t, &tcf_proto_base, head)
1da177e4
LT		 184 if (!strcmp(ops->kind, t->kind))
185 goto out;
186
36272874 187 list_add_tail(&ops->head, &tcf_proto_base);
1da177e4
LT		 188 rc = 0;
189out:
190 write_unlock(&cls_mod_lock);
191 return rc;
192}
aa767bfe 193EXPORT_SYMBOL(register_tcf_proto_ops);
1da177e4 194
7aa0045d
CW		 195static struct workqueue_struct *tc_filter_wq;
196
bc5c8260 197void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
1da177e4 198{
36272874 199 struct tcf_proto_ops *t;
1da177e4
LT		 200 int rc = -ENOENT;
201
c78e1746
DB		 202 /* Wait for outstanding call_rcu()s, if any, from a
203 * tcf_proto_ops's destroy() handler.
204 */
205 rcu_barrier();
7aa0045d 206 flush_workqueue(tc_filter_wq);
c78e1746 207
1da177e4 208 write_lock(&cls_mod_lock);
dcd76081
ED		 209 list_for_each_entry(t, &tcf_proto_base, head) {
210 if (t == ops) {
211 list_del(&t->head);
212 rc = 0;
1da177e4 213 break;
dcd76081
ED		 214 }
215 }
1da177e4 216 write_unlock(&cls_mod_lock);
bc5c8260
ZS		 217
218 WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
1da177e4 219}
aa767bfe 220EXPORT_SYMBOL(unregister_tcf_proto_ops);
1da177e4 221
aaa908ff 222bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
7aa0045d 223{
aaa908ff
CW		 224 INIT_RCU_WORK(rwork, func);
225 return queue_rcu_work(tc_filter_wq, rwork);
7aa0045d
CW		 226}
227EXPORT_SYMBOL(tcf_queue_work);
228
1da177e4
LT		 229/* Select new prio value from the range, managed by kernel. */
230
aa767bfe 231static inline u32 tcf_auto_prio(struct tcf_proto *tp)
1da177e4 232{
aa767bfe 233 u32 first = TC_H_MAKE(0xC0000000U, 0U);
1da177e4
LT		 234
235 if (tp)
cc7ec456 236 first = tp->prio - 1;
1da177e4 237
7961973a 238 return TC_H_MAJ(first);
1da177e4
LT		 239}
240
6f96c3c6
CW		 241static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
242{
243 if (kind)
872f6903 244 return nla_strscpy(name, kind, IFNAMSIZ) < 0;
6f96c3c6
CW		 245 memset(name, 0, IFNAMSIZ);
246 return false;
247}
248
470502de
VB		 249static bool tcf_proto_is_unlocked(const char *kind)
250{
251 const struct tcf_proto_ops *ops;
252 bool ret;
253
6f96c3c6
CW		 254 if (strlen(kind) == 0)
255 return false;
256
470502de
VB		 257 ops = tcf_proto_lookup_ops(kind, false, NULL);
258 /* On error return false to take rtnl lock. Proto lookup/create
259 * functions will perform lookup again and properly handle errors.
260 */
261 if (IS_ERR(ops))
262 return false;
263
264 ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
265 module_put(ops->owner);
266 return ret;
267}
268
33a48927 269static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
c35a4acc 270 u32 prio, struct tcf_chain *chain,
12db03b6 271 bool rtnl_held,
c35a4acc 272 struct netlink_ext_ack *extack)
33a48927
JP		 273{
274 struct tcf_proto *tp;
275 int err;
276
277 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
278 if (!tp)
279 return ERR_PTR(-ENOBUFS);
280
12db03b6 281 tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
f34e8bff
JP		 282 if (IS_ERR(tp->ops)) {
283 err = PTR_ERR(tp->ops);
d68d75fd 284 goto errout;
33a48927
JP		 285 }
286 tp->classify = tp->ops->classify;
287 tp->protocol = protocol;
288 tp->prio = prio;
5bc17018 289 tp->chain = chain;
8b64678e 290 spin_lock_init(&tp->lock);
4dbfa766 291 refcount_set(&tp->refcnt, 1);
33a48927
JP		 292
293 err = tp->ops->init(tp);
294 if (err) {
295 module_put(tp->ops->owner);
296 goto errout;
297 }
298 return tp;
299
300errout:
301 kfree(tp);
302 return ERR_PTR(err);
303}
304
4dbfa766
VB		 305static void tcf_proto_get(struct tcf_proto *tp)
306{
307 refcount_inc(&tp->refcnt);
308}
309
310static void tcf_chain_put(struct tcf_chain *chain);
311
12db03b6 312static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
59eb87cb 313 bool sig_destroy, struct netlink_ext_ack *extack)
cf1facda 314{
12db03b6 315 tp->ops->destroy(tp, rtnl_held, extack);
59eb87cb
JH		 316 if (sig_destroy)
317 tcf_proto_signal_destroyed(tp->chain, tp);
4dbfa766 318 tcf_chain_put(tp->chain);
763dbf63
WC		 319 module_put(tp->ops->owner);
320 kfree_rcu(tp, rcu);
cf1facda
JP		 321}
322
12db03b6 323static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
4dbfa766
VB		 324 struct netlink_ext_ack *extack)
325{
326 if (refcount_dec_and_test(&tp->refcnt))
59eb87cb 327 tcf_proto_destroy(tp, rtnl_held, true, extack);
4dbfa766
VB		 328}
329
a5b72a08 330static bool tcf_proto_check_delete(struct tcf_proto *tp)
8b64678e 331{
a5b72a08
DC		 332 if (tp->ops->delete_empty)
333 return tp->ops->delete_empty(tp);
8b64678e 334
a5b72a08 335 tp->deleting = true;
8b64678e
VB		 336 return tp->deleting;
337}
338
339static void tcf_proto_mark_delete(struct tcf_proto *tp)
340{
341 spin_lock(&tp->lock);
342 tp->deleting = true;
343 spin_unlock(&tp->lock);
344}
345
346static bool tcf_proto_is_deleting(struct tcf_proto *tp)
347{
348 bool deleting;
349
350 spin_lock(&tp->lock);
351 deleting = tp->deleting;
352 spin_unlock(&tp->lock);
353
354 return deleting;
355}
356
c266f64d
VB		 357#define ASSERT_BLOCK_LOCKED(block) \
358 lockdep_assert_held(&(block)->lock)
359
a9b19443
JP		 360struct tcf_filter_chain_list_item {
361 struct list_head list;
362 tcf_chain_head_change_t *chain_head_change;
363 void *chain_head_change_priv;
364};
365
5bc17018
JP		 366static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
367 u32 chain_index)
2190d1d0 368{
5bc17018
JP		 369 struct tcf_chain *chain;
370
c266f64d
VB		 371 ASSERT_BLOCK_LOCKED(block);
372
5bc17018
JP		 373 chain = kzalloc(sizeof(*chain), GFP_KERNEL);
374 if (!chain)
375 return NULL;
43719298 376 list_add_tail_rcu(&chain->list, &block->chain_list);
ed76f5ed 377 mutex_init(&chain->filter_chain_lock);
5bc17018
JP		 378 chain->block = block;
379 chain->index = chain_index;
e2ef7544 380 chain->refcnt = 1;
f71e0ca4
JP		 381 if (!chain->index)
382 block->chain0.chain = chain;
5bc17018 383 return chain;
2190d1d0
JP		 384}
385
a9b19443
JP		 386static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
387 struct tcf_proto *tp_head)
388{
389 if (item->chain_head_change)
390 item->chain_head_change(tp_head, item->chain_head_change_priv);
391}
f71e0ca4
JP		 392
393static void tcf_chain0_head_change(struct tcf_chain *chain,
394 struct tcf_proto *tp_head)
c7eb7d72 395{
a9b19443 396 struct tcf_filter_chain_list_item *item;
f71e0ca4 397 struct tcf_block *block = chain->block;
a9b19443 398
f71e0ca4
JP		 399 if (chain->index)
400 return;
165f0135
VB		 401
402 mutex_lock(&block->lock);
f71e0ca4 403 list_for_each_entry(item, &block->chain0.filter_chain_list, list)
a9b19443 404 tcf_chain_head_change_item(item, tp_head);
165f0135 405 mutex_unlock(&block->lock);
c7eb7d72
JP		 406}
407
c266f64d
VB		 408/* Returns true if block can be safely freed. */
409
410static bool tcf_chain_detach(struct tcf_chain *chain)
f93e1cdc 411{
efbf7897
CW		 412 struct tcf_block *block = chain->block;
413
c266f64d
VB		 414 ASSERT_BLOCK_LOCKED(block);
415
43719298 416 list_del_rcu(&chain->list);
f71e0ca4
JP		 417 if (!chain->index)
418 block->chain0.chain = NULL;
c266f64d
VB		 419
420 if (list_empty(&block->chain_list) &&
421 refcount_read(&block->refcnt) == 0)
422 return true;
423
424 return false;
425}
426
427static void tcf_block_destroy(struct tcf_block *block)
428{
429 mutex_destroy(&block->lock);
59eb87cb 430 mutex_destroy(&block->proto_destroy_lock);
c266f64d
VB		 431 kfree_rcu(block, rcu);
432}
433
434static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
435{
436 struct tcf_block *block = chain->block;
437
ed76f5ed 438 mutex_destroy(&chain->filter_chain_lock);
ee3bbfe8 439 kfree_rcu(chain, rcu);
c266f64d
VB		 440 if (free_block)
441 tcf_block_destroy(block);
e2ef7544 442}
744a4cf6 443
e2ef7544
CW		 444static void tcf_chain_hold(struct tcf_chain *chain)
445{
c266f64d
VB		 446 ASSERT_BLOCK_LOCKED(chain->block);
447
e2ef7544 448 ++chain->refcnt;
2190d1d0
JP		 449}
450
3d32f4c5 451static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
1f3ed383 452{
c266f64d
VB		 453 ASSERT_BLOCK_LOCKED(chain->block);
454
1f3ed383 455 /* In case all the references are action references, this
3d32f4c5 456 * chain should not be shown to the user.
1f3ed383
JP		 457 */
458 return chain->refcnt == chain->action_refcnt;
459}
460
32a4f5ec
JP		 461static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
462 u32 chain_index)
5bc17018
JP		 463{
464 struct tcf_chain *chain;
465
c266f64d
VB		 466 ASSERT_BLOCK_LOCKED(block);
467
5bc17018 468 list_for_each_entry(chain, &block->chain_list, list) {
32a4f5ec 469 if (chain->index == chain_index)
e2ef7544 470 return chain;
32a4f5ec
JP		 471 }
472 return NULL;
473}
474
af699626
PB		 475#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
476static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
477 u32 chain_index)
478{
479 struct tcf_chain *chain;
480
481 list_for_each_entry_rcu(chain, &block->chain_list, list) {
482 if (chain->index == chain_index)
483 return chain;
484 }
485 return NULL;
486}
487#endif
488
32a4f5ec
JP		 489static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
490 u32 seq, u16 flags, int event, bool unicast);
491
53681407
JP		 492static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
493 u32 chain_index, bool create,
494 bool by_act)
32a4f5ec 495{
c266f64d
VB		 496 struct tcf_chain *chain = NULL;
497 bool is_first_reference;
32a4f5ec 498
c266f64d
VB		 499 mutex_lock(&block->lock);
500 chain = tcf_chain_lookup(block, chain_index);
32a4f5ec
JP		 501 if (chain) {
502 tcf_chain_hold(chain);
53681407
JP		 503 } else {
504 if (!create)
c266f64d 505 goto errout;
53681407
JP		 506 chain = tcf_chain_create(block, chain_index);
507 if (!chain)
c266f64d 508 goto errout;
5bc17018 509 }
80532384 510
53681407
JP		 511 if (by_act)
512 ++chain->action_refcnt;
c266f64d
VB		 513 is_first_reference = chain->refcnt - chain->action_refcnt == 1;
514 mutex_unlock(&block->lock);
53681407
JP		 515
516 /* Send notification only in case we got the first
517 * non-action reference. Until then, the chain acts only as
518 * a placeholder for actions pointing to it and user ought
519 * not know about them.
520 */
c266f64d 521 if (is_first_reference && !by_act)
53681407
JP		 522 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
523 RTM_NEWCHAIN, false);
524
32a4f5ec 525 return chain;
c266f64d
VB		 526
527errout:
528 mutex_unlock(&block->lock);
529 return chain;
5bc17018 530}
53681407 531
290b1c8b
JP		 532static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
533 bool create)
53681407
JP		 534{
535 return __tcf_chain_get(block, chain_index, create, false);
536}
5bc17018 537
1f3ed383
JP		 538struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
539{
53681407 540 return __tcf_chain_get(block, chain_index, true, true);
1f3ed383
JP		 541}
542EXPORT_SYMBOL(tcf_chain_get_by_act);
543
a5654820
VB		 544static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
545 void *tmplt_priv);
546static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
547 void *tmplt_priv, u32 chain_index,
548 struct tcf_block *block, struct sk_buff *oskb,
549 u32 seq, u16 flags, bool unicast);
9f407f17 550
91052fa1
VB		 551static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
552 bool explicitly_created)
5bc17018 553{
c266f64d 554 struct tcf_block *block = chain->block;
a5654820 555 const struct tcf_proto_ops *tmplt_ops;
b62989fc 556 bool free_block = false;
c266f64d 557 unsigned int refcnt;
a5654820 558 void *tmplt_priv;
c266f64d
VB		 559
560 mutex_lock(&block->lock);
91052fa1
VB		 561 if (explicitly_created) {
562 if (!chain->explicitly_created) {
563 mutex_unlock(&block->lock);
564 return;
565 }
566 chain->explicitly_created = false;
567 }
568
53681407
JP		 569 if (by_act)
570 chain->action_refcnt--;
c266f64d
VB		 571
572 /* tc_chain_notify_delete can't be called while holding block lock.
573 * However, when block is unlocked chain can be changed concurrently, so
574 * save these to temporary variables.
575 */
576 refcnt = --chain->refcnt;
a5654820
VB		 577 tmplt_ops = chain->tmplt_ops;
578 tmplt_priv = chain->tmplt_priv;
53681407
JP		 579
580 /* The last dropped non-action reference will trigger notification. */
b62989fc
VB		 581 if (refcnt - chain->action_refcnt == 0 && !by_act) {
582 tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
a5654820 583 block, NULL, 0, 0, false);
726d0612
VB		 584 /* Last reference to chain, no need to lock. */
585 chain->flushing = false;
586 }
53681407 587
b62989fc
VB		 588 if (refcnt == 0)
589 free_block = tcf_chain_detach(chain);
590 mutex_unlock(&block->lock);
591
c266f64d 592 if (refcnt == 0) {
a5654820 593 tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
c266f64d 594 tcf_chain_destroy(chain, free_block);
32a4f5ec 595 }
5bc17018 596}
53681407 597
290b1c8b 598static void tcf_chain_put(struct tcf_chain *chain)
53681407 599{
91052fa1 600 __tcf_chain_put(chain, false, false);
53681407 601}
5bc17018 602
1f3ed383
JP		 603void tcf_chain_put_by_act(struct tcf_chain *chain)
604{
91052fa1 605 __tcf_chain_put(chain, true, false);
1f3ed383
JP		 606}
607EXPORT_SYMBOL(tcf_chain_put_by_act);
608
32a4f5ec
JP		 609static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
610{
91052fa1 611 __tcf_chain_put(chain, false, true);
32a4f5ec
JP		 612}
613
12db03b6 614static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
290b1c8b 615{
4dbfa766 616 struct tcf_proto *tp, *tp_next;
290b1c8b 617
ed76f5ed
VB		 618 mutex_lock(&chain->filter_chain_lock);
619 tp = tcf_chain_dereference(chain->filter_chain, chain);
59eb87cb
JH		 620 while (tp) {
621 tp_next = rcu_dereference_protected(tp->next, 1);
622 tcf_proto_signal_destroying(chain, tp);
623 tp = tp_next;
624 }
625 tp = tcf_chain_dereference(chain->filter_chain, chain);
4dbfa766 626 RCU_INIT_POINTER(chain->filter_chain, NULL);
290b1c8b 627 tcf_chain0_head_change(chain, NULL);
726d0612 628 chain->flushing = true;
ed76f5ed
VB		 629 mutex_unlock(&chain->filter_chain_lock);
630
290b1c8b 631 while (tp) {
4dbfa766 632 tp_next = rcu_dereference_protected(tp->next, 1);
12db03b6 633 tcf_proto_put(tp, rtnl_held, NULL);
4dbfa766 634 tp = tp_next;
290b1c8b
JP		 635 }
636}
637
59094b1e
PNA		 638static int tcf_block_setup(struct tcf_block *block,
639 struct flow_block_offload *bo);
640
324a823b 641static void tcf_block_offload_init(struct flow_block_offload *bo,
c40f4e50 642 struct net_device *dev, struct Qdisc *sch,
324a823b
PNA		 643 enum flow_block_command command,
644 enum flow_block_binder_type binder_type,
645 struct flow_block *flow_block,
646 bool shared, struct netlink_ext_ack *extack)
647{
648 bo->net = dev_net(dev);
649 bo->command = command;
650 bo->binder_type = binder_type;
651 bo->block = flow_block;
652 bo->block_shared = shared;
653 bo->extack = extack;
c40f4e50 654 bo->sch = sch;
74fc4f82 655 bo->cb_list_head = &flow_block->cb_list;
324a823b
PNA		 656 INIT_LIST_HEAD(&bo->cb_list);
657}
658
0fdcf78d
PNA		 659static void tcf_block_unbind(struct tcf_block *block,
660 struct flow_block_offload *bo);
661
662static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
7f76fa36 663{
0fdcf78d
PNA		 664 struct tcf_block *block = block_cb->indr.data;
665 struct net_device *dev = block_cb->indr.dev;
c40f4e50 666 struct Qdisc *sch = block_cb->indr.sch;
0fdcf78d 667 struct netlink_ext_ack extack = {};
990b03b0 668 struct flow_block_offload bo = {};
7f76fa36 669
c40f4e50 670 tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
0fdcf78d
PNA		 671 block_cb->indr.binder_type,
672 &block->flow_block, tcf_block_shared(block),
673 &extack);
d6535dca 674 rtnl_lock();
0fdcf78d 675 down_write(&block->cb_lock);
a1db2178 676 list_del(&block_cb->driver_list);
0fdcf78d 677 list_move(&block_cb->list, &bo.cb_list);
0fdcf78d 678 tcf_block_unbind(block, &bo);
d6535dca 679 up_write(&block->cb_lock);
0fdcf78d 680 rtnl_unlock();
7f76fa36
JH		 681}
682
caa72601
JP		 683static bool tcf_block_offload_in_use(struct tcf_block *block)
684{
97394bef 685 return atomic_read(&block->offloadcnt);
caa72601
JP		 686}
687
688static int tcf_block_offload_cmd(struct tcf_block *block,
c40f4e50 689 struct net_device *dev, struct Qdisc *sch,
caa72601 690 struct tcf_block_ext_info *ei,
9c0e189e 691 enum flow_block_command command,
60513bd8 692 struct netlink_ext_ack *extack)
8c4083b3 693{
955bcb6e 694 struct flow_block_offload bo = {};
8c4083b3 695
c40f4e50 696 tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
324a823b
PNA		 697 &block->flow_block, tcf_block_shared(block),
698 extack);
59094b1e 699
3c005110 700 if (dev->netdev_ops->ndo_setup_tc) {
701 int err;
702
0fdcf78d 703 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
3c005110 704 if (err < 0) {
705 if (err != -EOPNOTSUPP)
706 NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
707 return err;
708 }
709
710 return tcf_block_setup(block, &bo);
b70ba69e 711 }
59094b1e 712
c40f4e50 713 flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
3c005110 714 tc_block_indr_cleanup);
715 tcf_block_setup(block, &bo);
716
717 return -EOPNOTSUPP;
8c4083b3
JP		 718}
719
caa72601 720static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
60513bd8
JH		 721 struct tcf_block_ext_info *ei,
722 struct netlink_ext_ack *extack)
8c4083b3 723{
caa72601
JP		 724 struct net_device *dev = q->dev_queue->dev;
725 int err;
726
4f8116c8 727 down_write(&block->cb_lock);
caa72601
JP		 728
729 /* If tc offload feature is disabled and the block we try to bind
730 * to already has some offloaded filters, forbid to bind.
731 */
0fdcf78d
PNA		 732 if (dev->netdev_ops->ndo_setup_tc &&
733 !tc_can_offload(dev) &&
734 tcf_block_offload_in_use(block)) {
60513bd8 735 NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
4f8116c8
VB		 736 err = -EOPNOTSUPP;
737 goto err_unlock;
60513bd8 738 }
caa72601 739
c40f4e50 740 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
caa72601
JP		 741 if (err == -EOPNOTSUPP)
742 goto no_offload_dev_inc;
7f76fa36 743 if (err)
4f8116c8 744 goto err_unlock;
7f76fa36 745
4f8116c8 746 up_write(&block->cb_lock);
7f76fa36 747 return 0;
caa72601
JP		 748
749no_offload_dev_inc:
0fdcf78d 750 if (tcf_block_offload_in_use(block))
4f8116c8 751 goto err_unlock;
0fdcf78d 752
4f8116c8 753 err = 0;
caa72601 754 block->nooffloaddevcnt++;
4f8116c8
VB		 755err_unlock:
756 up_write(&block->cb_lock);
757 return err;
8c4083b3
JP		 758}
759
760static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
761 struct tcf_block_ext_info *ei)
762{
caa72601
JP		 763 struct net_device *dev = q->dev_queue->dev;
764 int err;
765
4f8116c8 766 down_write(&block->cb_lock);
c40f4e50 767 err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
caa72601
JP		 768 if (err == -EOPNOTSUPP)
769 goto no_offload_dev_dec;
4f8116c8 770 up_write(&block->cb_lock);
caa72601
JP		 771 return;
772
773no_offload_dev_dec:
774 WARN_ON(block->nooffloaddevcnt-- == 0);
4f8116c8 775 up_write(&block->cb_lock);
8c4083b3
JP		 776}
777
a9b19443 778static int
f71e0ca4
JP		 779tcf_chain0_head_change_cb_add(struct tcf_block *block,
780 struct tcf_block_ext_info *ei,
781 struct netlink_ext_ack *extack)
a9b19443
JP		 782{
783 struct tcf_filter_chain_list_item *item;
165f0135 784 struct tcf_chain *chain0;
a9b19443
JP		 785
786 item = kmalloc(sizeof(*item), GFP_KERNEL);
787 if (!item) {
788 NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
789 return -ENOMEM;
790 }
791 item->chain_head_change = ei->chain_head_change;
792 item->chain_head_change_priv = ei->chain_head_change_priv;
165f0135
VB		 793
794 mutex_lock(&block->lock);
795 chain0 = block->chain0.chain;
ed76f5ed
VB		 796 if (chain0)
797 tcf_chain_hold(chain0);
798 else
799 list_add(&item->list, &block->chain0.filter_chain_list);
165f0135
VB		 800 mutex_unlock(&block->lock);
801
ed76f5ed
VB		 802 if (chain0) {
803 struct tcf_proto *tp_head;
804
805 mutex_lock(&chain0->filter_chain_lock);
806
807 tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
808 if (tp_head)
809 tcf_chain_head_change_item(item, tp_head);
810
811 mutex_lock(&block->lock);
812 list_add(&item->list, &block->chain0.filter_chain_list);
813 mutex_unlock(&block->lock);
814
815 mutex_unlock(&chain0->filter_chain_lock);
816 tcf_chain_put(chain0);
817 }
818
a9b19443
JP		 819 return 0;
820}
821
822static void
f71e0ca4
JP		 823tcf_chain0_head_change_cb_del(struct tcf_block *block,
824 struct tcf_block_ext_info *ei)
a9b19443
JP		 825{
826 struct tcf_filter_chain_list_item *item;
827
165f0135 828 mutex_lock(&block->lock);
f71e0ca4 829 list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
a9b19443
JP		 830 if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
831 (item->chain_head_change == ei->chain_head_change &&
832 item->chain_head_change_priv == ei->chain_head_change_priv)) {
165f0135 833 if (block->chain0.chain)
f71e0ca4 834 tcf_chain_head_change_item(item, NULL);
a9b19443 835 list_del(&item->list);
165f0135
VB		 836 mutex_unlock(&block->lock);
837
a9b19443
JP		 838 kfree(item);
839 return;
840 }
841 }
165f0135 842 mutex_unlock(&block->lock);
a9b19443
JP		 843 WARN_ON(1);
844}
845
48617387 846struct tcf_net {
ab281629 847 spinlock_t idr_lock; /* Protects idr */
48617387
JP		 848 struct idr idr;
849};
850
851static unsigned int tcf_net_id;
852
853static int tcf_block_insert(struct tcf_block *block, struct net *net,
bb047ddd 854 struct netlink_ext_ack *extack)
a9b19443 855{
48617387 856 struct tcf_net *tn = net_generic(net, tcf_net_id);
ab281629
VB		 857 int err;
858
859 idr_preload(GFP_KERNEL);
860 spin_lock(&tn->idr_lock);
861 err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
862 GFP_NOWAIT);
863 spin_unlock(&tn->idr_lock);
864 idr_preload_end();
48617387 865
ab281629 866 return err;
a9b19443
JP		 867}
868
48617387
JP		 869static void tcf_block_remove(struct tcf_block *block, struct net *net)
870{
871 struct tcf_net *tn = net_generic(net, tcf_net_id);
872
ab281629 873 spin_lock(&tn->idr_lock);
9c160941 874 idr_remove(&tn->idr, block->index);
ab281629 875 spin_unlock(&tn->idr_lock);
48617387
JP		 876}
877
878static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
bb047ddd 879 u32 block_index,
48617387 880 struct netlink_ext_ack *extack)
6529eaba 881{
48617387 882 struct tcf_block *block;
6529eaba 883
48617387 884 block = kzalloc(sizeof(*block), GFP_KERNEL);
8d1a77f9
AA		 885 if (!block) {
886 NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
48617387 887 return ERR_PTR(-ENOMEM);
8d1a77f9 888 }
c266f64d 889 mutex_init(&block->lock);
59eb87cb 890 mutex_init(&block->proto_destroy_lock);
4f8116c8 891 init_rwsem(&block->cb_lock);
14bfb13f 892 flow_block_init(&block->flow_block);
5bc17018 893 INIT_LIST_HEAD(&block->chain_list);
f36fe1c4 894 INIT_LIST_HEAD(&block->owner_list);
f71e0ca4 895 INIT_LIST_HEAD(&block->chain0.filter_chain_list);
acb67442 896
cfebd7e2 897 refcount_set(&block->refcnt, 1);
48617387 898 block->net = net;
bb047ddd
JP		 899 block->index = block_index;
900
901 /* Don't store q pointer for blocks which are shared */
902 if (!tcf_block_shared(block))
903 block->q = q;
48617387 904 return block;
48617387
JP		 905}
906
907static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
908{
909 struct tcf_net *tn = net_generic(net, tcf_net_id);
910
322d884b 911 return idr_find(&tn->idr, block_index);
48617387
JP		 912}
913
0607e439
VB		 914static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
915{
916 struct tcf_block *block;
917
918 rcu_read_lock();
919 block = tcf_block_lookup(net, block_index);
920 if (block && !refcount_inc_not_zero(&block->refcnt))
921 block = NULL;
922 rcu_read_unlock();
923
924 return block;
925}
926
bbf73830
VB		 927static struct tcf_chain *
928__tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
f0023436 929{
bbf73830
VB		 930 mutex_lock(&block->lock);
931 if (chain)
932 chain = list_is_last(&chain->list, &block->chain_list) ?
933 NULL : list_next_entry(chain, list);
934 else
935 chain = list_first_entry_or_null(&block->chain_list,
936 struct tcf_chain, list);
f0023436 937
bbf73830
VB		 938 /* skip all action-only chains */
939 while (chain && tcf_chain_held_by_acts_only(chain))
940 chain = list_is_last(&chain->list, &block->chain_list) ?
941 NULL : list_next_entry(chain, list);
942
943 if (chain)
f0023436 944 tcf_chain_hold(chain);
bbf73830 945 mutex_unlock(&block->lock);
f0023436 946
bbf73830 947 return chain;
f0023436
VB		 948}
949
bbf73830
VB		 950/* Function to be used by all clients that want to iterate over all chains on
951 * block. It properly obtains block->lock and takes reference to chain before
952 * returning it. Users of this function must be tolerant to concurrent chain
953 * insertion/deletion or ensure that no concurrent chain modification is
954 * possible. Note that all netlink dump callbacks cannot guarantee to provide
955 * consistent dump because rtnl lock is released each time skb is filled with
956 * data and sent to user-space.
957 */
958
959struct tcf_chain *
960tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
f0023436 961{
bbf73830 962 struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
f0023436 963
bbf73830 964 if (chain)
f0023436 965 tcf_chain_put(chain);
bbf73830
VB		 966
967 return chain_next;
968}
969EXPORT_SYMBOL(tcf_get_next_chain);
970
fe2923af
VB		 971static struct tcf_proto *
972__tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
973{
8b64678e
VB		 974 u32 prio = 0;
975
fe2923af
VB		 976 ASSERT_RTNL();
977 mutex_lock(&chain->filter_chain_lock);
978
8b64678e 979 if (!tp) {
fe2923af 980 tp = tcf_chain_dereference(chain->filter_chain, chain);
8b64678e
VB		 981 } else if (tcf_proto_is_deleting(tp)) {
982 /* 'deleting' flag is set and chain->filter_chain_lock was
983 * unlocked, which means next pointer could be invalid. Restart
984 * search.
985 */
986 prio = tp->prio + 1;
987 tp = tcf_chain_dereference(chain->filter_chain, chain);
988
989 for (; tp; tp = tcf_chain_dereference(tp->next, chain))
990 if (!tp->deleting && tp->prio >= prio)
991 break;
992 } else {
fe2923af 993 tp = tcf_chain_dereference(tp->next, chain);
8b64678e 994 }
fe2923af
VB		 995
996 if (tp)
997 tcf_proto_get(tp);
998
999 mutex_unlock(&chain->filter_chain_lock);
1000
1001 return tp;
1002}
1003
1004/* Function to be used by all clients that want to iterate over all tp's on
1005 * chain. Users of this function must be tolerant to concurrent tp
1006 * insertion/deletion or ensure that no concurrent chain modification is
1007 * possible. Note that all netlink dump callbacks cannot guarantee to provide
1008 * consistent dump because rtnl lock is released each time skb is filled with
1009 * data and sent to user-space.
1010 */
1011
1012struct tcf_proto *
0fca55ed 1013tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
fe2923af
VB		 1014{
1015 struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1016
1017 if (tp)
0fca55ed 1018 tcf_proto_put(tp, true, NULL);
fe2923af
VB		 1019
1020 return tp_next;
1021}
1022EXPORT_SYMBOL(tcf_get_next_proto);
1023
12db03b6 1024static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
bbf73830
VB		 1025{
1026 struct tcf_chain *chain;
1027
1028 /* Last reference to block. At this point chains cannot be added or
1029 * removed concurrently.
1030 */
1031 for (chain = tcf_get_next_chain(block, NULL);
1032 chain;
1033 chain = tcf_get_next_chain(block, chain)) {
1034 tcf_chain_put_explicitly_created(chain);
12db03b6 1035 tcf_chain_flush(chain, rtnl_held);
f0023436
VB		 1036 }
1037}
1038
18d3eefb
VB		 1039/* Lookup Qdisc and increments its reference counter.
1040 * Set parent, if necessary.
1041 */
1042
1043static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1044 u32 *parent, int ifindex, bool rtnl_held,
1045 struct netlink_ext_ack *extack)
1046{
1047 const struct Qdisc_class_ops *cops;
1048 struct net_device *dev;
1049 int err = 0;
1050
1051 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1052 return 0;
1053
1054 rcu_read_lock();
1055
1056 /* Find link */
1057 dev = dev_get_by_index_rcu(net, ifindex);
1058 if (!dev) {
1059 rcu_read_unlock();
1060 return -ENODEV;
1061 }
1062
1063 /* Find qdisc */
1064 if (!*parent) {
5891cd5e 1065 *q = rcu_dereference(dev->qdisc);
18d3eefb
VB		 1066 *parent = (*q)->handle;
1067 } else {
1068 *q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1069 if (!*q) {
1070 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1071 err = -EINVAL;
1072 goto errout_rcu;
1073 }
1074 }
1075
1076 *q = qdisc_refcount_inc_nz(*q);
1077 if (!*q) {
1078 NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1079 err = -EINVAL;
1080 goto errout_rcu;
1081 }
1082
1083 /* Is it classful? */
1084 cops = (*q)->ops->cl_ops;
1085 if (!cops) {
1086 NL_SET_ERR_MSG(extack, "Qdisc not classful");
1087 err = -EINVAL;
1088 goto errout_qdisc;
1089 }
1090
1091 if (!cops->tcf_block) {
1092 NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1093 err = -EOPNOTSUPP;
1094 goto errout_qdisc;
1095 }
1096
1097errout_rcu:
1098 /* At this point we know that qdisc is not noop_qdisc,
1099 * which means that qdisc holds a reference to net_device
1100 * and we hold a reference to qdisc, so it is safe to release
1101 * rcu read lock.
1102 */
1103 rcu_read_unlock();
1104 return err;
1105
1106errout_qdisc:
1107 rcu_read_unlock();
1108
1109 if (rtnl_held)
1110 qdisc_put(*q);
1111 else
1112 qdisc_put_unlocked(*q);
1113 *q = NULL;
1114
1115 return err;
1116}
1117
1118static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1119 int ifindex, struct netlink_ext_ack *extack)
1120{
1121 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1122 return 0;
1123
1124 /* Do we search for filter, attached to class? */
1125 if (TC_H_MIN(parent)) {
1126 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1127
1128 *cl = cops->find(q, parent);
1129 if (*cl == 0) {
1130 NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1131 return -ENOENT;
1132 }
1133 }
1134
1135 return 0;
1136}
1137
1138static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1139 unsigned long cl, int ifindex,
1140 u32 block_index,
1141 struct netlink_ext_ack *extack)
1142{
1143 struct tcf_block *block;
1144
1145 if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1146 block = tcf_block_refcnt_get(net, block_index);
1147 if (!block) {
1148 NL_SET_ERR_MSG(extack, "Block of given index was not found");
1149 return ERR_PTR(-EINVAL);
1150 }
1151 } else {
1152 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1153
1154 block = cops->tcf_block(q, cl, extack);
1155 if (!block)
1156 return ERR_PTR(-EINVAL);
1157
1158 if (tcf_block_shared(block)) {
1159 NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1160 return ERR_PTR(-EOPNOTSUPP);
1161 }
1162
1163 /* Always take reference to block in order to support execution
1164 * of rules update path of cls API without rtnl lock. Caller
1165 * must release block when it is finished using it. 'if' block
1166 * of this conditional obtain reference to block by calling
1167 * tcf_block_refcnt_get().
1168 */
1169 refcount_inc(&block->refcnt);
1170 }
1171
1172 return block;
1173}
1174
0607e439 1175static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
12db03b6 1176 struct tcf_block_ext_info *ei, bool rtnl_held)
0607e439 1177{
c266f64d 1178 if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
0607e439
VB		 1179 /* Flushing/putting all chains will cause the block to be
1180 * deallocated when last chain is freed. However, if chain_list
1181 * is empty, block has to be manually deallocated. After block
1182 * reference counter reached 0, it is no longer possible to
1183 * increment it or add new chains to block.
1184 */
1185 bool free_block = list_empty(&block->chain_list);
1186
c266f64d 1187 mutex_unlock(&block->lock);
0607e439
VB		 1188 if (tcf_block_shared(block))
1189 tcf_block_remove(block, block->net);
0607e439
VB		 1190
1191 if (q)
1192 tcf_block_offload_unbind(block, q, ei);
1193
1194 if (free_block)
c266f64d 1195 tcf_block_destroy(block);
0607e439 1196 else
12db03b6 1197 tcf_block_flush_all_chains(block, rtnl_held);
0607e439
VB		 1198 } else if (q) {
1199 tcf_block_offload_unbind(block, q, ei);
1200 }
1201}
1202
12db03b6 1203static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
0607e439 1204{
12db03b6 1205 __tcf_block_put(block, NULL, NULL, rtnl_held);
0607e439
VB		 1206}
1207
c431f89b
VB		 1208/* Find tcf block.
1209 * Set q, parent, cl when appropriate.
1210 */
1211
1212static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1213 u32 *parent, unsigned long *cl,
1214 int ifindex, u32 block_index,
1215 struct netlink_ext_ack *extack)
1216{
1217 struct tcf_block *block;
e368fdb6 1218 int err = 0;
c431f89b 1219
18d3eefb 1220 ASSERT_RTNL();
e368fdb6 1221
18d3eefb
VB		 1222 err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1223 if (err)
1224 goto errout;
c431f89b 1225
18d3eefb
VB		 1226 err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1227 if (err)
1228 goto errout_qdisc;
787ce6d0 1229
18d3eefb 1230 block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
af736bf0
DC		 1231 if (IS_ERR(block)) {
1232 err = PTR_ERR(block);
18d3eefb 1233 goto errout_qdisc;
af736bf0 1234 }
c431f89b
VB		 1235
1236 return block;
e368fdb6 1237
e368fdb6 1238errout_qdisc:
18d3eefb 1239 if (*q)
e368fdb6 1240 qdisc_put(*q);
18d3eefb
VB		 1241errout:
1242 *q = NULL;
e368fdb6
VB		 1243 return ERR_PTR(err);
1244}
1245
12db03b6
VB		 1246static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1247 bool rtnl_held)
e368fdb6 1248{
787ce6d0 1249 if (!IS_ERR_OR_NULL(block))
12db03b6 1250 tcf_block_refcnt_put(block, rtnl_held);
787ce6d0 1251
470502de
VB		 1252 if (q) {
1253 if (rtnl_held)
1254 qdisc_put(q);
1255 else
1256 qdisc_put_unlocked(q);
1257 }
c431f89b
VB		 1258}
1259
f36fe1c4
JP		 1260struct tcf_block_owner_item {
1261 struct list_head list;
1262 struct Qdisc *q;
32f8c409 1263 enum flow_block_binder_type binder_type;
f36fe1c4
JP		 1264};
1265
1266static void
1267tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1268 struct Qdisc *q,
32f8c409 1269 enum flow_block_binder_type binder_type)
f36fe1c4
JP		 1270{
1271 if (block->keep_dst &&
32f8c409
PNA		 1272 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1273 binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
f36fe1c4
JP		 1274 netif_keep_dst(qdisc_dev(q));
1275}
1276
1277void tcf_block_netif_keep_dst(struct tcf_block *block)
1278{
1279 struct tcf_block_owner_item *item;
1280
1281 block->keep_dst = true;
1282 list_for_each_entry(item, &block->owner_list, list)
1283 tcf_block_owner_netif_keep_dst(block, item->q,
1284 item->binder_type);
1285}
1286EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1287
1288static int tcf_block_owner_add(struct tcf_block *block,
1289 struct Qdisc *q,
32f8c409 1290 enum flow_block_binder_type binder_type)
f36fe1c4
JP		 1291{
1292 struct tcf_block_owner_item *item;
1293
1294 item = kmalloc(sizeof(*item), GFP_KERNEL);
1295 if (!item)
1296 return -ENOMEM;
1297 item->q = q;
1298 item->binder_type = binder_type;
1299 list_add(&item->list, &block->owner_list);
1300 return 0;
1301}
1302
1303static void tcf_block_owner_del(struct tcf_block *block,
1304 struct Qdisc *q,
32f8c409 1305 enum flow_block_binder_type binder_type)
f36fe1c4
JP		 1306{
1307 struct tcf_block_owner_item *item;
1308
1309 list_for_each_entry(item, &block->owner_list, list) {
1310 if (item->q == q && item->binder_type == binder_type) {
1311 list_del(&item->list);
1312 kfree(item);
1313 return;
1314 }
1315 }
1316 WARN_ON(1);
1317}
1318
48617387
JP		 1319int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1320 struct tcf_block_ext_info *ei,
1321 struct netlink_ext_ack *extack)
1322{
1323 struct net *net = qdisc_net(q);
1324 struct tcf_block *block = NULL;
48617387
JP		 1325 int err;
1326
787ce6d0 1327 if (ei->block_index)
48617387 1328 /* block_index not 0 means the shared block is requested */
787ce6d0 1329 block = tcf_block_refcnt_get(net, ei->block_index);
48617387
JP		 1330
1331 if (!block) {
bb047ddd 1332 block = tcf_block_create(net, q, ei->block_index, extack);
48617387
JP		 1333 if (IS_ERR(block))
1334 return PTR_ERR(block);
bb047ddd
JP		 1335 if (tcf_block_shared(block)) {
1336 err = tcf_block_insert(block, net, extack);
48617387
JP		 1337 if (err)
1338 goto err_block_insert;
1339 }
1340 }
1341
f36fe1c4
JP		 1342 err = tcf_block_owner_add(block, q, ei->binder_type);
1343 if (err)
1344 goto err_block_owner_add;
1345
1346 tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1347
f71e0ca4 1348 err = tcf_chain0_head_change_cb_add(block, ei, extack);
a9b19443 1349 if (err)
f71e0ca4 1350 goto err_chain0_head_change_cb_add;
caa72601 1351
60513bd8 1352 err = tcf_block_offload_bind(block, q, ei, extack);
caa72601
JP		 1353 if (err)
1354 goto err_block_offload_bind;
1355
6529eaba
JP		 1356 *p_block = block;
1357 return 0;
2190d1d0 1358
caa72601 1359err_block_offload_bind:
f71e0ca4
JP		 1360 tcf_chain0_head_change_cb_del(block, ei);
1361err_chain0_head_change_cb_add:
f36fe1c4
JP		 1362 tcf_block_owner_del(block, q, ei->binder_type);
1363err_block_owner_add:
48617387 1364err_block_insert:
12db03b6 1365 tcf_block_refcnt_put(block, true);
2190d1d0 1366 return err;
6529eaba 1367}
8c4083b3
JP		 1368EXPORT_SYMBOL(tcf_block_get_ext);
1369
c7eb7d72
JP		 1370static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1371{
1372 struct tcf_proto __rcu **p_filter_chain = priv;
1373
1374 rcu_assign_pointer(*p_filter_chain, tp_head);
1375}
1376
8c4083b3 1377int tcf_block_get(struct tcf_block **p_block,
8d1a77f9
AA		 1378 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1379 struct netlink_ext_ack *extack)
8c4083b3 1380{
c7eb7d72
JP		 1381 struct tcf_block_ext_info ei = {
1382 .chain_head_change = tcf_chain_head_change_dflt,
1383 .chain_head_change_priv = p_filter_chain,
1384 };
8c4083b3 1385
c7eb7d72 1386 WARN_ON(!p_filter_chain);
8d1a77f9 1387 return tcf_block_get_ext(p_block, q, &ei, extack);
8c4083b3 1388}
6529eaba
JP		 1389EXPORT_SYMBOL(tcf_block_get);
1390
7aa0045d 1391/* XXX: Standalone actions are not allowed to jump to any chain, and bound
a60b3f51 1392 * actions should be all removed after flushing.
7aa0045d 1393 */
c7eb7d72 1394void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
e1ea2f98 1395 struct tcf_block_ext_info *ei)
7aa0045d 1396{
c30abd5e
DM		 1397 if (!block)
1398 return;
f71e0ca4 1399 tcf_chain0_head_change_cb_del(block, ei);
f36fe1c4 1400 tcf_block_owner_del(block, q, ei->binder_type);
a60b3f51 1401
12db03b6 1402 __tcf_block_put(block, q, ei, true);
6529eaba 1403}
8c4083b3
JP		 1404EXPORT_SYMBOL(tcf_block_put_ext);
1405
1406void tcf_block_put(struct tcf_block *block)
1407{
1408 struct tcf_block_ext_info ei = {0, };
1409
4853f128
JP		 1410 if (!block)
1411 return;
c7eb7d72 1412 tcf_block_put_ext(block, block->q, &ei);
8c4083b3 1413}
e1ea2f98 1414
6529eaba 1415EXPORT_SYMBOL(tcf_block_put);
cf1facda 1416
32636742 1417static int
a7323311 1418tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
32636742
JH		 1419 void *cb_priv, bool add, bool offload_in_use,
1420 struct netlink_ext_ack *extack)
1421{
bbf73830 1422 struct tcf_chain *chain, *chain_prev;
fe2923af 1423 struct tcf_proto *tp, *tp_prev;
32636742
JH		 1424 int err;
1425
4f8116c8
VB		 1426 lockdep_assert_held(&block->cb_lock);
1427
bbf73830
VB		 1428 for (chain = __tcf_get_next_chain(block, NULL);
1429 chain;
1430 chain_prev = chain,
1431 chain = __tcf_get_next_chain(block, chain),
1432 tcf_chain_put(chain_prev)) {
fe2923af
VB		 1433 for (tp = __tcf_get_next_proto(chain, NULL); tp;
1434 tp_prev = tp,
1435 tp = __tcf_get_next_proto(chain, tp),
12db03b6 1436 tcf_proto_put(tp_prev, true, NULL)) {
32636742
JH		 1437 if (tp->ops->reoffload) {
1438 err = tp->ops->reoffload(tp, add, cb, cb_priv,
1439 extack);
1440 if (err && add)
1441 goto err_playback_remove;
1442 } else if (add && offload_in_use) {
1443 err = -EOPNOTSUPP;
1444 NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1445 goto err_playback_remove;
1446 }
1447 }
1448 }
1449
1450 return 0;
1451
1452err_playback_remove:
12db03b6 1453 tcf_proto_put(tp, true, NULL);
bbf73830 1454 tcf_chain_put(chain);
32636742
JH		 1455 tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1456 extack);
1457 return err;
1458}
1459
59094b1e
PNA		 1460static int tcf_block_bind(struct tcf_block *block,
1461 struct flow_block_offload *bo)
1462{
1463 struct flow_block_cb *block_cb, *next;
1464 int err, i = 0;
1465
4f8116c8
VB		 1466 lockdep_assert_held(&block->cb_lock);
1467
59094b1e
PNA		 1468 list_for_each_entry(block_cb, &bo->cb_list, list) {
1469 err = tcf_block_playback_offloads(block, block_cb->cb,
1470 block_cb->cb_priv, true,
1471 tcf_block_offload_in_use(block),
1472 bo->extack);
1473 if (err)
1474 goto err_unroll;
c9f14470
VB		 1475 if (!bo->unlocked_driver_cb)
1476 block->lockeddevcnt++;
59094b1e
PNA		 1477
1478 i++;
1479 }
14bfb13f 1480 list_splice(&bo->cb_list, &block->flow_block.cb_list);
59094b1e
PNA		 1481
1482 return 0;
1483
1484err_unroll:
1485 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1486 if (i-- > 0) {
1487 list_del(&block_cb->list);
1488 tcf_block_playback_offloads(block, block_cb->cb,
1489 block_cb->cb_priv, false,
1490 tcf_block_offload_in_use(block),
1491 NULL);
c9f14470
VB		 1492 if (!bo->unlocked_driver_cb)
1493 block->lockeddevcnt--;
59094b1e
PNA		 1494 }
1495 flow_block_cb_free(block_cb);
1496 }
1497
1498 return err;
1499}
1500
1501static void tcf_block_unbind(struct tcf_block *block,
1502 struct flow_block_offload *bo)
1503{
1504 struct flow_block_cb *block_cb, *next;
1505
4f8116c8
VB		 1506 lockdep_assert_held(&block->cb_lock);
1507
59094b1e
PNA		 1508 list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1509 tcf_block_playback_offloads(block, block_cb->cb,
1510 block_cb->cb_priv, false,
1511 tcf_block_offload_in_use(block),
1512 NULL);
1513 list_del(&block_cb->list);
1514 flow_block_cb_free(block_cb);
c9f14470
VB		 1515 if (!bo->unlocked_driver_cb)
1516 block->lockeddevcnt--;
59094b1e
PNA		 1517 }
1518}
1519
1520static int tcf_block_setup(struct tcf_block *block,
1521 struct flow_block_offload *bo)
1522{
1523 int err;
1524
1525 switch (bo->command) {
1526 case FLOW_BLOCK_BIND:
1527 err = tcf_block_bind(block, bo);
1528 break;
1529 case FLOW_BLOCK_UNBIND:
1530 err = 0;
1531 tcf_block_unbind(block, bo);
1532 break;
1533 default:
1534 WARN_ON_ONCE(1);
1535 err = -EOPNOTSUPP;
1536 }
1537
1538 return err;
1539}
1540
87d83093
JP		 1541/* Main classifier routine: scans classifier chain attached
1542 * to this qdisc, (optionally) tests for protocol and asks
1543 * specific classifiers.
1544 */
9410c940
PB		 1545static inline int __tcf_classify(struct sk_buff *skb,
1546 const struct tcf_proto *tp,
af699626 1547 const struct tcf_proto *orig_tp,
9410c940
PB		 1548 struct tcf_result *res,
1549 bool compat_mode,
1550 u32 *last_executed_chain)
87d83093 1551{
87d83093 1552#ifdef CONFIG_NET_CLS_ACT
05ff8435 1553 const int max_reclassify_loop = 16;
ee538dce 1554 const struct tcf_proto *first_tp;
87d83093
JP		 1555 int limit = 0;
1556
1557reclassify:
1558#endif
1559 for (; tp; tp = rcu_dereference_bh(tp->next)) {
d7bf2ebe 1560 __be16 protocol = skb_protocol(skb, false);
87d83093
JP		 1561 int err;
1562
1563 if (tp->protocol != protocol &&
1564 tp->protocol != htons(ETH_P_ALL))
1565 continue;
1566
1567 err = tp->classify(skb, tp, res);
1568#ifdef CONFIG_NET_CLS_ACT
db50514f 1569 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
ee538dce 1570 first_tp = orig_tp;
9410c940 1571 *last_executed_chain = first_tp->chain->index;
87d83093 1572 goto reset;
db50514f 1573 } else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
ee538dce 1574 first_tp = res->goto_tp;
9410c940 1575 *last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
db50514f
JP		 1576 goto reset;
1577 }
87d83093
JP		 1578#endif
1579 if (err >= 0)
1580 return err;
1581 }
1582
1583 return TC_ACT_UNSPEC; /* signal: continue lookup */
1584#ifdef CONFIG_NET_CLS_ACT
1585reset:
1586 if (unlikely(limit++ >= max_reclassify_loop)) {
9d3aaff3
JP		 1587 net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1588 tp->chain->block->index,
1589 tp->prio & 0xffff,
87d83093
JP		 1590 ntohs(tp->protocol));
1591 return TC_ACT_SHOT;
1592 }
1593
ee538dce 1594 tp = first_tp;
87d83093
JP		 1595 goto reclassify;
1596#endif
1597}
9410c940 1598
3aa26055
DC		 1599int tcf_classify(struct sk_buff *skb,
1600 const struct tcf_block *block,
1601 const struct tcf_proto *tp,
9410c940 1602 struct tcf_result *res, bool compat_mode)
9410c940
PB		 1603{
1604#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1605 u32 last_executed_chain = 0;
1606
af699626 1607 return __tcf_classify(skb, tp, tp, res, compat_mode,
9410c940
PB		 1608 &last_executed_chain);
1609#else
1610 u32 last_executed_chain = tp ? tp->chain->index : 0;
af699626 1611 const struct tcf_proto *orig_tp = tp;
9410c940
PB		 1612 struct tc_skb_ext *ext;
1613 int ret;
1614
3aa26055
DC		 1615 if (block) {
1616 ext = skb_ext_find(skb, TC_SKB_EXT);
af699626 1617
3aa26055
DC		 1618 if (ext && ext->chain) {
1619 struct tcf_chain *fchain;
af699626 1620
3aa26055
DC		 1621 fchain = tcf_chain_lookup_rcu(block, ext->chain);
1622 if (!fchain)
1623 return TC_ACT_SHOT;
af699626 1624
3aa26055
DC		 1625 /* Consume, so cloned/redirect skbs won't inherit ext */
1626 skb_ext_del(skb, TC_SKB_EXT);
af699626 1627
3aa26055
DC		 1628 tp = rcu_dereference_bh(fchain->filter_chain);
1629 last_executed_chain = fchain->index;
1630 }
af699626
PB		 1631 }
1632
1633 ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode,
1634 &last_executed_chain);
9410c940 1635
35d39fec
PB		 1636 if (tc_skb_ext_tc_enabled()) {
1637 /* If we missed on some chain */
1638 if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1639 struct tc_skb_cb *cb = tc_skb_cb(skb);
1640
1641 ext = tc_skb_ext_alloc(skb);
1642 if (WARN_ON_ONCE(!ext))
1643 return TC_ACT_SHOT;
1644 ext->chain = last_executed_chain;
1645 ext->mru = cb->mru;
1646 ext->post_ct = cb->post_ct;
1647 ext->post_ct_snat = cb->post_ct_snat;
1648 ext->post_ct_dnat = cb->post_ct_dnat;
1649 ext->zone = cb->zone;
1650 }
9410c940
PB		 1651 }
1652
1653 return ret;
1654#endif
1655}
3aa26055 1656EXPORT_SYMBOL(tcf_classify);
9410c940 1657
2190d1d0
JP		 1658struct tcf_chain_info {
1659 struct tcf_proto __rcu **pprev;
1660 struct tcf_proto __rcu *next;
1661};
1662
ed76f5ed
VB		 1663static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1664 struct tcf_chain_info *chain_info)
2190d1d0 1665{
ed76f5ed 1666 return tcf_chain_dereference(*chain_info->pprev, chain);
2190d1d0
JP		 1667}
1668
726d0612
VB		 1669static int tcf_chain_tp_insert(struct tcf_chain *chain,
1670 struct tcf_chain_info *chain_info,
1671 struct tcf_proto *tp)
2190d1d0 1672{
726d0612
VB		 1673 if (chain->flushing)
1674 return -EAGAIN;
1675
e65812fd 1676 RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
c7eb7d72 1677 if (*chain_info->pprev == chain->filter_chain)
f71e0ca4 1678 tcf_chain0_head_change(chain, tp);
4dbfa766 1679 tcf_proto_get(tp);
2190d1d0 1680 rcu_assign_pointer(*chain_info->pprev, tp);
726d0612
VB		 1681
1682 return 0;
2190d1d0
JP		 1683}
1684
1685static void tcf_chain_tp_remove(struct tcf_chain *chain,
1686 struct tcf_chain_info *chain_info,
1687 struct tcf_proto *tp)
1688{
ed76f5ed 1689 struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
2190d1d0 1690
8b64678e 1691 tcf_proto_mark_delete(tp);
c7eb7d72 1692 if (tp == chain->filter_chain)
f71e0ca4 1693 tcf_chain0_head_change(chain, next);
2190d1d0
JP		 1694 RCU_INIT_POINTER(*chain_info->pprev, next);
1695}
1696
8b64678e
VB		 1697static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1698 struct tcf_chain_info *chain_info,
1699 u32 protocol, u32 prio,
1700 bool prio_allocate);
1701
1702/* Try to insert new proto.
1703 * If proto with specified priority already exists, free new proto
1704 * and return existing one.
1705 */
1706
1707static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1708 struct tcf_proto *tp_new,
12db03b6
VB		 1709 u32 protocol, u32 prio,
1710 bool rtnl_held)
8b64678e
VB		 1711{
1712 struct tcf_chain_info chain_info;
1713 struct tcf_proto *tp;
726d0612 1714 int err = 0;
8b64678e
VB		 1715
1716 mutex_lock(&chain->filter_chain_lock);
1717
59eb87cb
JH		 1718 if (tcf_proto_exists_destroying(chain, tp_new)) {
1719 mutex_unlock(&chain->filter_chain_lock);
1720 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1721 return ERR_PTR(-EAGAIN);
1722 }
1723
8b64678e
VB		 1724 tp = tcf_chain_tp_find(chain, &chain_info,
1725 protocol, prio, false);
1726 if (!tp)
726d0612 1727 err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
8b64678e
VB		 1728 mutex_unlock(&chain->filter_chain_lock);
1729
1730 if (tp) {
59eb87cb 1731 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
8b64678e 1732 tp_new = tp;
726d0612 1733 } else if (err) {
59eb87cb 1734 tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
726d0612 1735 tp_new = ERR_PTR(err);
8b64678e
VB		 1736 }
1737
1738 return tp_new;
1739}
1740
1741static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
12db03b6 1742 struct tcf_proto *tp, bool rtnl_held,
8b64678e
VB		 1743 struct netlink_ext_ack *extack)
1744{
1745 struct tcf_chain_info chain_info;
1746 struct tcf_proto *tp_iter;
1747 struct tcf_proto **pprev;
1748 struct tcf_proto *next;
1749
1750 mutex_lock(&chain->filter_chain_lock);
1751
1752 /* Atomically find and remove tp from chain. */
1753 for (pprev = &chain->filter_chain;
1754 (tp_iter = tcf_chain_dereference(*pprev, chain));
1755 pprev = &tp_iter->next) {
1756 if (tp_iter == tp) {
1757 chain_info.pprev = pprev;
1758 chain_info.next = tp_iter->next;
1759 WARN_ON(tp_iter->deleting);
1760 break;
1761 }
1762 }
1763 /* Verify that tp still exists and no new filters were inserted
1764 * concurrently.
1765 * Mark tp for deletion if it is empty.
1766 */
a5b72a08 1767 if (!tp_iter || !tcf_proto_check_delete(tp)) {
8b64678e
VB		 1768 mutex_unlock(&chain->filter_chain_lock);
1769 return;
1770 }
1771
59eb87cb 1772 tcf_proto_signal_destroying(chain, tp);
8b64678e
VB		 1773 next = tcf_chain_dereference(chain_info.next, chain);
1774 if (tp == chain->filter_chain)
1775 tcf_chain0_head_change(chain, next);
1776 RCU_INIT_POINTER(*chain_info.pprev, next);
1777 mutex_unlock(&chain->filter_chain_lock);
1778
12db03b6 1779 tcf_proto_put(tp, rtnl_held, extack);
8b64678e
VB		 1780}
1781
2190d1d0
JP		 1782static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1783 struct tcf_chain_info *chain_info,
1784 u32 protocol, u32 prio,
1785 bool prio_allocate)
1786{
1787 struct tcf_proto **pprev;
1788 struct tcf_proto *tp;
1789
1790 /* Check the chain for existence of proto-tcf with this priority */
1791 for (pprev = &chain->filter_chain;
ed76f5ed
VB		 1792 (tp = tcf_chain_dereference(*pprev, chain));
1793 pprev = &tp->next) {
2190d1d0
JP		 1794 if (tp->prio >= prio) {
1795 if (tp->prio == prio) {
1796 if (prio_allocate ||
1797 (tp->protocol != protocol && protocol))
1798 return ERR_PTR(-EINVAL);
1799 } else {
1800 tp = NULL;
1801 }
1802 break;
1803 }
1804 }
1805 chain_info->pprev = pprev;
4dbfa766
VB		 1806 if (tp) {
1807 chain_info->next = tp->next;
1808 tcf_proto_get(tp);
1809 } else {
1810 chain_info->next = NULL;
1811 }
2190d1d0
JP		 1812 return tp;
1813}
1814
7120371c 1815static int tcf_fill_node(struct net *net, struct sk_buff *skb,
7960d1da
JP		 1816 struct tcf_proto *tp, struct tcf_block *block,
1817 struct Qdisc *q, u32 parent, void *fh,
12db03b6 1818 u32 portid, u32 seq, u16 flags, int event,
f8ab1807 1819 bool terse_dump, bool rtnl_held)
7120371c
WC		 1820{
1821 struct tcmsg *tcm;
1822 struct nlmsghdr *nlh;
1823 unsigned char *b = skb_tail_pointer(skb);
1824
1825 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1826 if (!nlh)
1827 goto out_nlmsg_trim;
1828 tcm = nlmsg_data(nlh);
1829 tcm->tcm_family = AF_UNSPEC;
1830 tcm->tcm__pad1 = 0;
1831 tcm->tcm__pad2 = 0;
7960d1da
JP		 1832 if (q) {
1833 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1834 tcm->tcm_parent = parent;
1835 } else {
1836 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1837 tcm->tcm_block_index = block->index;
1838 }
7120371c
WC		 1839 tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1840 if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1841 goto nla_put_failure;
1842 if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1843 goto nla_put_failure;
1844 if (!fh) {
1845 tcm->tcm_handle = 0;
f8ab1807
VB		 1846 } else if (terse_dump) {
1847 if (tp->ops->terse_dump) {
1848 if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1849 rtnl_held) < 0)
1850 goto nla_put_failure;
1851 } else {
1852 goto cls_op_not_supp;
1853 }
7120371c 1854 } else {
12db03b6
VB		 1855 if (tp->ops->dump &&
1856 tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
7120371c
WC		 1857 goto nla_put_failure;
1858 }
1859 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1860 return skb->len;
1861
1862out_nlmsg_trim:
1863nla_put_failure:
f8ab1807 1864cls_op_not_supp:
7120371c
WC		 1865 nlmsg_trim(skb, b);
1866 return -1;
1867}
1868
1869static int tfilter_notify(struct net *net, struct sk_buff *oskb,
1870 struct nlmsghdr *n, struct tcf_proto *tp,
7960d1da 1871 struct tcf_block *block, struct Qdisc *q,
12db03b6
VB		 1872 u32 parent, void *fh, int event, bool unicast,
1873 bool rtnl_held)
7120371c
WC		 1874{
1875 struct sk_buff *skb;
1876 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
5b5f99b1 1877 int err = 0;
7120371c
WC		 1878
1879 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1880 if (!skb)
1881 return -ENOBUFS;
1882
7960d1da 1883 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
12db03b6 1884 n->nlmsg_seq, n->nlmsg_flags, event,
f8ab1807 1885 false, rtnl_held) <= 0) {
7120371c
WC		 1886 kfree_skb(skb);
1887 return -EINVAL;
1888 }
1889
1890 if (unicast)
f79a3bcb 1891 err = rtnl_unicast(skb, net, portid);
5b5f99b1
ZW		 1892 else
1893 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1894 n->nlmsg_flags & NLM_F_ECHO);
5b5f99b1 1895 return err;
7120371c
WC		 1896}
1897
1898static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
1899 struct nlmsghdr *n, struct tcf_proto *tp,
7960d1da 1900 struct tcf_block *block, struct Qdisc *q,
c35a4acc 1901 u32 parent, void *fh, bool unicast, bool *last,
12db03b6 1902 bool rtnl_held, struct netlink_ext_ack *extack)
7120371c
WC		 1903{
1904 struct sk_buff *skb;
1905 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1906 int err;
1907
1908 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1909 if (!skb)
1910 return -ENOBUFS;
1911
7960d1da 1912 if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
12db03b6 1913 n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
f8ab1807 1914 false, rtnl_held) <= 0) {
c35a4acc 1915 NL_SET_ERR_MSG(extack, "Failed to build del event notification");
7120371c
WC		 1916 kfree_skb(skb);
1917 return -EINVAL;
1918 }
1919
12db03b6 1920 err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
7120371c
WC		 1921 if (err) {
1922 kfree_skb(skb);
1923 return err;
1924 }
1925
1926 if (unicast)
f79a3bcb 1927 err = rtnl_unicast(skb, net, portid);
5b5f99b1
ZW		 1928 else
1929 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1930 n->nlmsg_flags & NLM_F_ECHO);
c35a4acc
AA		 1931 if (err < 0)
1932 NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
5b5f99b1 1933
c35a4acc 1934 return err;
7120371c
WC		 1935}
1936
1937static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
7960d1da
JP		 1938 struct tcf_block *block, struct Qdisc *q,
1939 u32 parent, struct nlmsghdr *n,
0fca55ed 1940 struct tcf_chain *chain, int event)
7120371c
WC		 1941{
1942 struct tcf_proto *tp;
1943
0fca55ed
VB		 1944 for (tp = tcf_get_next_proto(chain, NULL);
1945 tp; tp = tcf_get_next_proto(chain, tp))
7960d1da 1946 tfilter_notify(net, oskb, n, tp, block,
0fca55ed 1947 q, parent, NULL, event, false, true);
7120371c
WC		 1948}
1949
7d5509fa
VB		 1950static void tfilter_put(struct tcf_proto *tp, void *fh)
1951{
1952 if (tp->ops->put && fh)
1953 tp->ops->put(tp, fh);
1954}
1955
c431f89b 1956static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
c21ef3e3 1957 struct netlink_ext_ack *extack)
1da177e4 1958{
3b1e0a65 1959 struct net *net = sock_net(skb->sk);
add93b61 1960 struct nlattr *tca[TCA_MAX + 1];
6f96c3c6 1961 char name[IFNAMSIZ];
1da177e4
LT		 1962 struct tcmsg *t;
1963 u32 protocol;
1964 u32 prio;
9d36d9e5 1965 bool prio_allocate;
1da177e4 1966 u32 parent;
5bc17018 1967 u32 chain_index;
04c2a47f 1968 struct Qdisc *q;
2190d1d0 1969 struct tcf_chain_info chain_info;
04c2a47f 1970 struct tcf_chain *chain;
6529eaba 1971 struct tcf_block *block;
1da177e4 1972 struct tcf_proto *tp;
1da177e4 1973 unsigned long cl;
8113c095 1974 void *fh;
1da177e4 1975 int err;
628185cf 1976 int tp_created;
470502de 1977 bool rtnl_held = false;
a5397d68 1978 u32 flags;
1da177e4 1979
c431f89b 1980 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
dfc47ef8 1981 return -EPERM;
de179c8c 1982
1da177e4 1983replay:
628185cf
DB		 1984 tp_created = 0;
1985
8cb08174
JB		 1986 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
1987 rtm_tca_policy, extack);
de179c8c
H		 1988 if (err < 0)
1989 return err;
1990
942b8165 1991 t = nlmsg_data(n);
1da177e4
LT		 1992 protocol = TC_H_MIN(t->tcm_info);
1993 prio = TC_H_MAJ(t->tcm_info);
9d36d9e5 1994 prio_allocate = false;
1da177e4 1995 parent = t->tcm_parent;
4dbfa766 1996 tp = NULL;
1da177e4 1997 cl = 0;
470502de 1998 block = NULL;
04c2a47f
ED		 1999 q = NULL;
2000 chain = NULL;
a5397d68 2001 flags = 0;
1da177e4
LT		 2002
2003 if (prio == 0) {
c431f89b
VB		 2004 /* If no priority is provided by the user,
2005 * we allocate one.
2006 */
2007 if (n->nlmsg_flags & NLM_F_CREATE) {
2008 prio = TC_H_MAKE(0x80000000U, 0U);
2009 prio_allocate = true;
2010 } else {
c35a4acc 2011 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
1da177e4 2012 return -ENOENT;
ea7f8277 2013 }
1da177e4
LT		 2014 }
2015
2016 /* Find head of filter chain. */
2017
470502de
VB		 2018 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2019 if (err)
2020 return err;
2021
6f96c3c6
CW		 2022 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2023 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2024 err = -EINVAL;
2025 goto errout;
2026 }
2027
470502de
VB		 2028 /* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2029 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2030 * type is not specified, classifier is not unlocked.
2031 */
2032 if (rtnl_held ||
2033 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
6f96c3c6 2034 !tcf_proto_is_unlocked(name)) {
470502de
VB		 2035 rtnl_held = true;
2036 rtnl_lock();
2037 }
2038
2039 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2040 if (err)
2041 goto errout;
2042
2043 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2044 extack);
c431f89b
VB		 2045 if (IS_ERR(block)) {
2046 err = PTR_ERR(block);
2047 goto errout;
6bb16e7a 2048 }
a7df4870 2049 block->classid = parent;
5bc17018
JP		 2050
2051 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2052 if (chain_index > TC_ACT_EXT_VAL_MASK) {
c35a4acc 2053 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
5bc17018
JP		 2054 err = -EINVAL;
2055 goto errout;
2056 }
c431f89b 2057 chain = tcf_chain_get(block, chain_index, true);
5bc17018 2058 if (!chain) {
d5ed72a5 2059 NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
c431f89b 2060 err = -ENOMEM;
ea7f8277
DB		 2061 goto errout;
2062 }
1da177e4 2063
ed76f5ed 2064 mutex_lock(&chain->filter_chain_lock);
2190d1d0
JP		 2065 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2066 prio, prio_allocate);
2067 if (IS_ERR(tp)) {
c35a4acc 2068 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2190d1d0 2069 err = PTR_ERR(tp);
ed76f5ed 2070 goto errout_locked;
1da177e4
LT		 2071 }
2072
2073 if (tp == NULL) {
8b64678e
VB		 2074 struct tcf_proto *tp_new = NULL;
2075
726d0612
VB		 2076 if (chain->flushing) {
2077 err = -EAGAIN;
2078 goto errout_locked;
2079 }
2080
1da177e4
LT		 2081 /* Proto-tcf does not exist, create new one */
2082
6bb16e7a 2083 if (tca[TCA_KIND] == NULL || !protocol) {
c35a4acc 2084 NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
6bb16e7a 2085 err = -EINVAL;
ed76f5ed 2086 goto errout_locked;
6bb16e7a 2087 }
1da177e4 2088
c431f89b 2089 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
c35a4acc 2090 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
6bb16e7a 2091 err = -ENOENT;
ed76f5ed 2092 goto errout_locked;
6bb16e7a 2093 }
1da177e4 2094
9d36d9e5 2095 if (prio_allocate)
ed76f5ed
VB		 2096 prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2097 &chain_info));
1da177e4 2098
ed76f5ed 2099 mutex_unlock(&chain->filter_chain_lock);
36d79af7
ED		 2100 tp_new = tcf_proto_create(name, protocol, prio, chain,
2101 rtnl_held, extack);
8b64678e
VB		 2102 if (IS_ERR(tp_new)) {
2103 err = PTR_ERR(tp_new);
726d0612 2104 goto errout_tp;
1da177e4 2105 }
ed76f5ed 2106
12186be7 2107 tp_created = 1;
12db03b6
VB		 2108 tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2109 rtnl_held);
726d0612
VB		 2110 if (IS_ERR(tp)) {
2111 err = PTR_ERR(tp);
2112 goto errout_tp;
2113 }
ed76f5ed
VB		 2114 } else {
2115 mutex_unlock(&chain->filter_chain_lock);
6bb16e7a 2116 }
1da177e4 2117
8b64678e
VB		 2118 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2119 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2120 err = -EINVAL;
2121 goto errout;
2122 }
2123
1da177e4
LT		 2124 fh = tp->ops->get(tp, t->tcm_handle);
2125
8113c095 2126 if (!fh) {
c431f89b 2127 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
c35a4acc 2128 NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
6bb16e7a 2129 err = -ENOENT;
1da177e4 2130 goto errout;
6bb16e7a 2131 }
c431f89b 2132 } else if (n->nlmsg_flags & NLM_F_EXCL) {
7d5509fa 2133 tfilter_put(tp, fh);
c431f89b
VB		 2134 NL_SET_ERR_MSG(extack, "Filter already exists");
2135 err = -EEXIST;
2136 goto errout;
1da177e4
LT		 2137 }
2138
9f407f17 2139 if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
c2e1cfef 2140 tfilter_put(tp, fh);
9f407f17
JP		 2141 NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2142 err = -EINVAL;
2143 goto errout;
2144 }
2145
695176bf
CW		 2146 if (!(n->nlmsg_flags & NLM_F_CREATE))
2147 flags |= TCA_ACT_FLAGS_REPLACE;
2148 if (!rtnl_held)
2149 flags |= TCA_ACT_FLAGS_NO_RTNL;
2f7ef2f8 2150 err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
695176bf 2151 flags, extack);
7d5509fa 2152 if (err == 0) {
7960d1da 2153 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
12db03b6 2154 RTM_NEWTFILTER, false, rtnl_held);
7d5509fa 2155 tfilter_put(tp, fh);
503d81d4
VB		 2156 /* q pointer is NULL for shared blocks */
2157 if (q)
2158 q->flags &= ~TCQ_F_CAN_BYPASS;
7d5509fa 2159 }
1da177e4
LT		 2160
2161errout:
8b64678e 2162 if (err && tp_created)
12db03b6 2163 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
726d0612 2164errout_tp:
4dbfa766
VB		 2165 if (chain) {
2166 if (tp && !IS_ERR(tp))
12db03b6 2167 tcf_proto_put(tp, rtnl_held, NULL);
4dbfa766
VB		 2168 if (!tp_created)
2169 tcf_chain_put(chain);
2170 }
12db03b6 2171 tcf_block_release(q, block, rtnl_held);
470502de
VB		 2172
2173 if (rtnl_held)
2174 rtnl_unlock();
2175
2176 if (err == -EAGAIN) {
2177 /* Take rtnl lock in case EAGAIN is caused by concurrent flush
2178 * of target chain.
2179 */
2180 rtnl_held = true;
1da177e4
LT		 2181 /* Replay the request. */
2182 goto replay;
470502de 2183 }
1da177e4 2184 return err;
ed76f5ed
VB		 2185
2186errout_locked:
2187 mutex_unlock(&chain->filter_chain_lock);
2188 goto errout;
1da177e4
LT		 2189}
2190
c431f89b
VB		 2191static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2192 struct netlink_ext_ack *extack)
2193{
2194 struct net *net = sock_net(skb->sk);
2195 struct nlattr *tca[TCA_MAX + 1];
6f96c3c6 2196 char name[IFNAMSIZ];
c431f89b
VB		 2197 struct tcmsg *t;
2198 u32 protocol;
2199 u32 prio;
2200 u32 parent;
2201 u32 chain_index;
2202 struct Qdisc *q = NULL;
2203 struct tcf_chain_info chain_info;
2204 struct tcf_chain *chain = NULL;
470502de 2205 struct tcf_block *block = NULL;
c431f89b
VB		 2206 struct tcf_proto *tp = NULL;
2207 unsigned long cl = 0;
2208 void *fh = NULL;
2209 int err;
470502de 2210 bool rtnl_held = false;
c431f89b
VB		 2211
2212 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2213 return -EPERM;
2214
8cb08174
JB		 2215 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2216 rtm_tca_policy, extack);
c431f89b
VB		 2217 if (err < 0)
2218 return err;
2219
2220 t = nlmsg_data(n);
2221 protocol = TC_H_MIN(t->tcm_info);
2222 prio = TC_H_MAJ(t->tcm_info);
2223 parent = t->tcm_parent;
2224
2225 if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2226 NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2227 return -ENOENT;
2228 }
2229
2230 /* Find head of filter chain. */
2231
470502de
VB		 2232 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2233 if (err)
2234 return err;
2235
6f96c3c6
CW		 2236 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2237 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2238 err = -EINVAL;
2239 goto errout;
2240 }
470502de
VB		 2241 /* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2242 * found), qdisc is not unlocked, classifier type is not specified,
2243 * classifier is not unlocked.
2244 */
2245 if (!prio ||
2246 (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
6f96c3c6 2247 !tcf_proto_is_unlocked(name)) {
470502de
VB		 2248 rtnl_held = true;
2249 rtnl_lock();
2250 }
2251
2252 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2253 if (err)
2254 goto errout;
2255
2256 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2257 extack);
c431f89b
VB		 2258 if (IS_ERR(block)) {
2259 err = PTR_ERR(block);
2260 goto errout;
2261 }
2262
2263 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2264 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2265 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2266 err = -EINVAL;
2267 goto errout;
2268 }
2269 chain = tcf_chain_get(block, chain_index, false);
2270 if (!chain) {
5ca8a25c
JP		 2271 /* User requested flush on non-existent chain. Nothing to do,
2272 * so just return success.
2273 */
2274 if (prio == 0) {
2275 err = 0;
2276 goto errout;
2277 }
c431f89b 2278 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
b7b4247d 2279 err = -ENOENT;
c431f89b
VB		 2280 goto errout;
2281 }
2282
2283 if (prio == 0) {
2284 tfilter_notify_chain(net, skb, block, q, parent, n,
0fca55ed 2285 chain, RTM_DELTFILTER);
12db03b6 2286 tcf_chain_flush(chain, rtnl_held);
c431f89b
VB		 2287 err = 0;
2288 goto errout;
2289 }
2290
ed76f5ed 2291 mutex_lock(&chain->filter_chain_lock);
c431f89b
VB		 2292 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2293 prio, false);
2294 if (!tp || IS_ERR(tp)) {
2295 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
0e399035 2296 err = tp ? PTR_ERR(tp) : -ENOENT;
ed76f5ed 2297 goto errout_locked;
c431f89b
VB		 2298 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2299 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2300 err = -EINVAL;
ed76f5ed
VB		 2301 goto errout_locked;
2302 } else if (t->tcm_handle == 0) {
59eb87cb 2303 tcf_proto_signal_destroying(chain, tp);
ed76f5ed
VB		 2304 tcf_chain_tp_remove(chain, &chain_info, tp);
2305 mutex_unlock(&chain->filter_chain_lock);
2306
12db03b6 2307 tcf_proto_put(tp, rtnl_held, NULL);
ed76f5ed 2308 tfilter_notify(net, skb, n, tp, block, q, parent, fh,
12db03b6 2309 RTM_DELTFILTER, false, rtnl_held);
ed76f5ed 2310 err = 0;
c431f89b
VB		 2311 goto errout;
2312 }
ed76f5ed 2313 mutex_unlock(&chain->filter_chain_lock);
c431f89b
VB		 2314
2315 fh = tp->ops->get(tp, t->tcm_handle);
2316
2317 if (!fh) {
ed76f5ed
VB		 2318 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2319 err = -ENOENT;
c431f89b
VB		 2320 } else {
2321 bool last;
2322
2323 err = tfilter_del_notify(net, skb, n, tp, block,
2324 q, parent, fh, false, &last,
12db03b6
VB		 2325 rtnl_held, extack);
2326
c431f89b
VB		 2327 if (err)
2328 goto errout;
8b64678e 2329 if (last)
12db03b6 2330 tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
c431f89b
VB		 2331 }
2332
2333errout:
4dbfa766
VB		 2334 if (chain) {
2335 if (tp && !IS_ERR(tp))
12db03b6 2336 tcf_proto_put(tp, rtnl_held, NULL);
c431f89b 2337 tcf_chain_put(chain);
4dbfa766 2338 }
12db03b6 2339 tcf_block_release(q, block, rtnl_held);
470502de
VB		 2340
2341 if (rtnl_held)
2342 rtnl_unlock();
2343
c431f89b 2344 return err;
ed76f5ed
VB		 2345
2346errout_locked:
2347 mutex_unlock(&chain->filter_chain_lock);
2348 goto errout;
c431f89b
VB		 2349}
2350
2351static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2352 struct netlink_ext_ack *extack)
2353{
2354 struct net *net = sock_net(skb->sk);
2355 struct nlattr *tca[TCA_MAX + 1];
6f96c3c6 2356 char name[IFNAMSIZ];
c431f89b
VB		 2357 struct tcmsg *t;
2358 u32 protocol;
2359 u32 prio;
2360 u32 parent;
2361 u32 chain_index;
2362 struct Qdisc *q = NULL;
2363 struct tcf_chain_info chain_info;
2364 struct tcf_chain *chain = NULL;
470502de 2365 struct tcf_block *block = NULL;
c431f89b
VB		 2366 struct tcf_proto *tp = NULL;
2367 unsigned long cl = 0;
2368 void *fh = NULL;
2369 int err;
470502de 2370 bool rtnl_held = false;
c431f89b 2371
8cb08174
JB		 2372 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2373 rtm_tca_policy, extack);
c431f89b
VB		 2374 if (err < 0)
2375 return err;
2376
2377 t = nlmsg_data(n);
2378 protocol = TC_H_MIN(t->tcm_info);
2379 prio = TC_H_MAJ(t->tcm_info);
2380 parent = t->tcm_parent;
2381
2382 if (prio == 0) {
2383 NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2384 return -ENOENT;
2385 }
2386
2387 /* Find head of filter chain. */
2388
470502de
VB		 2389 err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2390 if (err)
2391 return err;
2392
6f96c3c6
CW		 2393 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2394 NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2395 err = -EINVAL;
2396 goto errout;
2397 }
470502de
VB		 2398 /* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2399 * unlocked, classifier type is not specified, classifier is not
2400 * unlocked.
2401 */
2402 if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
6f96c3c6 2403 !tcf_proto_is_unlocked(name)) {
470502de
VB		 2404 rtnl_held = true;
2405 rtnl_lock();
2406 }
2407
2408 err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2409 if (err)
2410 goto errout;
2411
2412 block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2413 extack);
c431f89b
VB		 2414 if (IS_ERR(block)) {
2415 err = PTR_ERR(block);
2416 goto errout;
2417 }
2418
2419 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2420 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2421 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2422 err = -EINVAL;
2423 goto errout;
2424 }
2425 chain = tcf_chain_get(block, chain_index, false);
2426 if (!chain) {
2427 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2428 err = -EINVAL;
2429 goto errout;
2430 }
2431
ed76f5ed 2432 mutex_lock(&chain->filter_chain_lock);
c431f89b
VB		 2433 tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2434 prio, false);
ed76f5ed 2435 mutex_unlock(&chain->filter_chain_lock);
c431f89b
VB		 2436 if (!tp || IS_ERR(tp)) {
2437 NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
0e399035 2438 err = tp ? PTR_ERR(tp) : -ENOENT;
c431f89b
VB		 2439 goto errout;
2440 } else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2441 NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2442 err = -EINVAL;
2443 goto errout;
2444 }
2445
2446 fh = tp->ops->get(tp, t->tcm_handle);
2447
2448 if (!fh) {
2449 NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2450 err = -ENOENT;
2451 } else {
2452 err = tfilter_notify(net, skb, n, tp, block, q, parent,
12db03b6 2453 fh, RTM_NEWTFILTER, true, rtnl_held);
c431f89b
VB		 2454 if (err < 0)
2455 NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2456 }
2457
7d5509fa 2458 tfilter_put(tp, fh);
c431f89b 2459errout:
4dbfa766
VB		 2460 if (chain) {
2461 if (tp && !IS_ERR(tp))
12db03b6 2462 tcf_proto_put(tp, rtnl_held, NULL);
c431f89b 2463 tcf_chain_put(chain);
4dbfa766 2464 }
12db03b6 2465 tcf_block_release(q, block, rtnl_held);
470502de
VB		 2466
2467 if (rtnl_held)
2468 rtnl_unlock();
2469
c431f89b
VB		 2470 return err;
2471}
2472
aa767bfe 2473struct tcf_dump_args {
1da177e4
LT		 2474 struct tcf_walker w;
2475 struct sk_buff *skb;
2476 struct netlink_callback *cb;
7960d1da 2477 struct tcf_block *block;
a10fa201
JP		 2478 struct Qdisc *q;
2479 u32 parent;
f8ab1807 2480 bool terse_dump;
1da177e4
LT		 2481};
2482
8113c095 2483static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1da177e4 2484{
aa767bfe 2485 struct tcf_dump_args *a = (void *)arg;
832d1d5b 2486 struct net *net = sock_net(a->skb->sk);
1da177e4 2487
7960d1da 2488 return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
a10fa201 2489 n, NETLINK_CB(a->cb->skb).portid,
5a7a5555 2490 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
f8ab1807 2491 RTM_NEWTFILTER, a->terse_dump, true);
1da177e4
LT		 2492}
2493
a10fa201
JP		 2494static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2495 struct sk_buff *skb, struct netlink_callback *cb,
f8ab1807 2496 long index_start, long *p_index, bool terse)
acb31fae
JP		 2497{
2498 struct net *net = sock_net(skb->sk);
7960d1da 2499 struct tcf_block *block = chain->block;
acb31fae 2500 struct tcmsg *tcm = nlmsg_data(cb->nlh);
fe2923af 2501 struct tcf_proto *tp, *tp_prev;
acb31fae 2502 struct tcf_dump_args arg;
acb31fae 2503
fe2923af
VB		 2504 for (tp = __tcf_get_next_proto(chain, NULL);
2505 tp;
2506 tp_prev = tp,
2507 tp = __tcf_get_next_proto(chain, tp),
12db03b6 2508 tcf_proto_put(tp_prev, true, NULL),
fe2923af 2509 (*p_index)++) {
acb31fae
JP		 2510 if (*p_index < index_start)
2511 continue;
2512 if (TC_H_MAJ(tcm->tcm_info) &&
2513 TC_H_MAJ(tcm->tcm_info) != tp->prio)
2514 continue;
2515 if (TC_H_MIN(tcm->tcm_info) &&
2516 TC_H_MIN(tcm->tcm_info) != tp->protocol)
2517 continue;
2518 if (*p_index > index_start)
2519 memset(&cb->args[1], 0,
2520 sizeof(cb->args) - sizeof(cb->args[0]));
2521 if (cb->args[1] == 0) {
53189183 2522 if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
acb31fae
JP		 2523 NETLINK_CB(cb->skb).portid,
2524 cb->nlh->nlmsg_seq, NLM_F_MULTI,
f8ab1807 2525 RTM_NEWTFILTER, false, true) <= 0)
fe2923af 2526 goto errout;
acb31fae
JP		 2527 cb->args[1] = 1;
2528 }
2529 if (!tp->ops->walk)
2530 continue;
2531 arg.w.fn = tcf_node_dump;
2532 arg.skb = skb;
2533 arg.cb = cb;
7960d1da 2534 arg.block = block;
a10fa201
JP		 2535 arg.q = q;
2536 arg.parent = parent;
acb31fae
JP		 2537 arg.w.stop = 0;
2538 arg.w.skip = cb->args[1] - 1;
2539 arg.w.count = 0;
01683a14 2540 arg.w.cookie = cb->args[2];
f8ab1807 2541 arg.terse_dump = terse;
12db03b6 2542 tp->ops->walk(tp, &arg.w, true);
01683a14 2543 cb->args[2] = arg.w.cookie;
acb31fae
JP		 2544 cb->args[1] = arg.w.count + 1;
2545 if (arg.w.stop)
fe2923af 2546 goto errout;
acb31fae 2547 }
5bc17018 2548 return true;
fe2923af
VB		 2549
2550errout:
12db03b6 2551 tcf_proto_put(tp, true, NULL);
fe2923af 2552 return false;
acb31fae
JP		 2553}
2554
f8ab1807
VB		 2555static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2556 [TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2557};
2558
bd27a875 2559/* called with RTNL */
1da177e4
LT		 2560static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2561{
bbf73830 2562 struct tcf_chain *chain, *chain_prev;
3b1e0a65 2563 struct net *net = sock_net(skb->sk);
5bc17018 2564 struct nlattr *tca[TCA_MAX + 1];
7960d1da 2565 struct Qdisc *q = NULL;
6529eaba 2566 struct tcf_block *block;
942b8165 2567 struct tcmsg *tcm = nlmsg_data(cb->nlh);
f8ab1807 2568 bool terse_dump = false;
acb31fae
JP		 2569 long index_start;
2570 long index;
a10fa201 2571 u32 parent;
5bc17018 2572 int err;
1da177e4 2573
573ce260 2574 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1da177e4 2575 return skb->len;
5bc17018 2576
8cb08174 2577 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
f8ab1807 2578 tcf_tfilter_dump_policy, cb->extack);
5bc17018
JP		 2579 if (err)
2580 return err;
2581
f8ab1807
VB		 2582 if (tca[TCA_DUMP_FLAGS]) {
2583 struct nla_bitfield32 flags =
2584 nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2585
2586 terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2587 }
2588
7960d1da 2589 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
787ce6d0 2590 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
7960d1da
JP		 2591 if (!block)
2592 goto out;
d680b352
JP		 2593 /* If we work with block index, q is NULL and parent value
2594 * will never be used in the following code. The check
2595 * in tcf_fill_node prevents it. However, compiler does not
2596 * see that far, so set parent to zero to silence the warning
2597 * about parent being uninitialized.
2598 */
2599 parent = 0;
a10fa201 2600 } else {
7960d1da
JP		 2601 const struct Qdisc_class_ops *cops;
2602 struct net_device *dev;
2603 unsigned long cl = 0;
2604
2605 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2606 if (!dev)
2607 return skb->len;
2608
2609 parent = tcm->tcm_parent;
a7df4870 2610 if (!parent)
5891cd5e 2611 q = rtnl_dereference(dev->qdisc);
a7df4870 2612 else
7960d1da 2613 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
7960d1da
JP		 2614 if (!q)
2615 goto out;
2616 cops = q->ops->cl_ops;
2617 if (!cops)
143976ce 2618 goto out;
7960d1da
JP		 2619 if (!cops->tcf_block)
2620 goto out;
2621 if (TC_H_MIN(tcm->tcm_parent)) {
2622 cl = cops->find(q, tcm->tcm_parent);
2623 if (cl == 0)
2624 goto out;
2625 }
2626 block = cops->tcf_block(q, cl, NULL);
2627 if (!block)
2628 goto out;
a7df4870 2629 parent = block->classid;
7960d1da
JP		 2630 if (tcf_block_shared(block))
2631 q = NULL;
1da177e4 2632 }
1da177e4 2633
acb31fae
JP		 2634 index_start = cb->args[0];
2635 index = 0;
5bc17018 2636
bbf73830
VB		 2637 for (chain = __tcf_get_next_chain(block, NULL);
2638 chain;
2639 chain_prev = chain,
2640 chain = __tcf_get_next_chain(block, chain),
2641 tcf_chain_put(chain_prev)) {
5bc17018
JP		 2642 if (tca[TCA_CHAIN] &&
2643 nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2644 continue;
a10fa201 2645 if (!tcf_chain_dump(chain, q, parent, skb, cb,
f8ab1807 2646 index_start, &index, terse_dump)) {
bbf73830 2647 tcf_chain_put(chain);
5ae437ad 2648 err = -EMSGSIZE;
5bc17018 2649 break;
5ae437ad 2650 }
5bc17018
JP		 2651 }
2652
787ce6d0 2653 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
12db03b6 2654 tcf_block_refcnt_put(block, true);
acb31fae 2655 cb->args[0] = index;
1da177e4 2656
1da177e4 2657out:
5ae437ad
RK		 2658 /* If we did no progress, the error (EMSGSIZE) is real */
2659 if (skb->len == 0 && err)
2660 return err;
1da177e4
LT		 2661 return skb->len;
2662}
2663
a5654820
VB		 2664static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2665 void *tmplt_priv, u32 chain_index,
2666 struct net *net, struct sk_buff *skb,
2667 struct tcf_block *block,
32a4f5ec
JP		 2668 u32 portid, u32 seq, u16 flags, int event)
2669{
2670 unsigned char *b = skb_tail_pointer(skb);
9f407f17 2671 const struct tcf_proto_ops *ops;
32a4f5ec
JP		 2672 struct nlmsghdr *nlh;
2673 struct tcmsg *tcm;
9f407f17
JP		 2674 void *priv;
2675
a5654820
VB		 2676 ops = tmplt_ops;
2677 priv = tmplt_priv;
32a4f5ec
JP		 2678
2679 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2680 if (!nlh)
2681 goto out_nlmsg_trim;
2682 tcm = nlmsg_data(nlh);
2683 tcm->tcm_family = AF_UNSPEC;
2684 tcm->tcm__pad1 = 0;
2685 tcm->tcm__pad2 = 0;
2686 tcm->tcm_handle = 0;
2687 if (block->q) {
2688 tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2689 tcm->tcm_parent = block->q->handle;
2690 } else {
2691 tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2692 tcm->tcm_block_index = block->index;
2693 }
2694
a5654820 2695 if (nla_put_u32(skb, TCA_CHAIN, chain_index))
32a4f5ec
JP		 2696 goto nla_put_failure;
2697
9f407f17
JP		 2698 if (ops) {
2699 if (nla_put_string(skb, TCA_KIND, ops->kind))
2700 goto nla_put_failure;
2701 if (ops->tmplt_dump(skb, net, priv) < 0)
2702 goto nla_put_failure;
2703 }
2704
32a4f5ec
JP		 2705 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2706 return skb->len;
2707
2708out_nlmsg_trim:
2709nla_put_failure:
2710 nlmsg_trim(skb, b);
2711 return -EMSGSIZE;
2712}
2713
2714static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2715 u32 seq, u16 flags, int event, bool unicast)
2716{
2717 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2718 struct tcf_block *block = chain->block;
2719 struct net *net = block->net;
2720 struct sk_buff *skb;
5b5f99b1 2721 int err = 0;
32a4f5ec
JP		 2722
2723 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2724 if (!skb)
2725 return -ENOBUFS;
2726
a5654820
VB		 2727 if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2728 chain->index, net, skb, block, portid,
32a4f5ec
JP		 2729 seq, flags, event) <= 0) {
2730 kfree_skb(skb);
2731 return -EINVAL;
2732 }
2733
2734 if (unicast)
f79a3bcb 2735 err = rtnl_unicast(skb, net, portid);
5b5f99b1
ZW		 2736 else
2737 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2738 flags & NLM_F_ECHO);
32a4f5ec 2739
5b5f99b1 2740 return err;
32a4f5ec
JP		 2741}
2742
a5654820
VB		 2743static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2744 void *tmplt_priv, u32 chain_index,
2745 struct tcf_block *block, struct sk_buff *oskb,
2746 u32 seq, u16 flags, bool unicast)
2747{
2748 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2749 struct net *net = block->net;
2750 struct sk_buff *skb;
2751
2752 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2753 if (!skb)
2754 return -ENOBUFS;
2755
2756 if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2757 block, portid, seq, flags, RTM_DELCHAIN) <= 0) {
2758 kfree_skb(skb);
2759 return -EINVAL;
2760 }
2761
2762 if (unicast)
f79a3bcb 2763 return rtnl_unicast(skb, net, portid);
a5654820
VB		 2764
2765 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2766}
2767
9f407f17
JP		 2768static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2769 struct nlattr **tca,
2770 struct netlink_ext_ack *extack)
2771{
2772 const struct tcf_proto_ops *ops;
2dd5616e 2773 char name[IFNAMSIZ];
9f407f17
JP		 2774 void *tmplt_priv;
2775
2776 /* If kind is not set, user did not specify template. */
2777 if (!tca[TCA_KIND])
2778 return 0;
2779
2dd5616e
ED		 2780 if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2781 NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2782 return -EINVAL;
2783 }
2784
2785 ops = tcf_proto_lookup_ops(name, true, extack);
9f407f17
JP		 2786 if (IS_ERR(ops))
2787 return PTR_ERR(ops);
2788 if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2789 NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2790 return -EOPNOTSUPP;
2791 }
2792
2793 tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2794 if (IS_ERR(tmplt_priv)) {
2795 module_put(ops->owner);
2796 return PTR_ERR(tmplt_priv);
2797 }
2798 chain->tmplt_ops = ops;
2799 chain->tmplt_priv = tmplt_priv;
2800 return 0;
2801}
2802
a5654820
VB		 2803static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2804 void *tmplt_priv)
9f407f17 2805{
9f407f17 2806 /* If template ops are set, no work to do for us. */
a5654820 2807 if (!tmplt_ops)
9f407f17
JP		 2808 return;
2809
a5654820
VB		 2810 tmplt_ops->tmplt_destroy(tmplt_priv);
2811 module_put(tmplt_ops->owner);
9f407f17
JP		 2812}
2813
32a4f5ec
JP		 2814/* Add/delete/get a chain */
2815
2816static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2817 struct netlink_ext_ack *extack)
2818{
2819 struct net *net = sock_net(skb->sk);
2820 struct nlattr *tca[TCA_MAX + 1];
2821 struct tcmsg *t;
2822 u32 parent;
2823 u32 chain_index;
04c2a47f
ED		 2824 struct Qdisc *q;
2825 struct tcf_chain *chain;
32a4f5ec
JP		 2826 struct tcf_block *block;
2827 unsigned long cl;
2828 int err;
2829
2830 if (n->nlmsg_type != RTM_GETCHAIN &&
2831 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2832 return -EPERM;
2833
2834replay:
04c2a47f 2835 q = NULL;
8cb08174
JB		 2836 err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2837 rtm_tca_policy, extack);
32a4f5ec
JP		 2838 if (err < 0)
2839 return err;
2840
2841 t = nlmsg_data(n);
2842 parent = t->tcm_parent;
2843 cl = 0;
2844
2845 block = tcf_block_find(net, &q, &parent, &cl,
2846 t->tcm_ifindex, t->tcm_block_index, extack);
2847 if (IS_ERR(block))
2848 return PTR_ERR(block);
2849
2850 chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2851 if (chain_index > TC_ACT_EXT_VAL_MASK) {
2852 NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
e368fdb6
VB		 2853 err = -EINVAL;
2854 goto errout_block;
32a4f5ec 2855 }
2cbfab07
VB		 2856
2857 mutex_lock(&block->lock);
32a4f5ec
JP		 2858 chain = tcf_chain_lookup(block, chain_index);
2859 if (n->nlmsg_type == RTM_NEWCHAIN) {
2860 if (chain) {
3d32f4c5 2861 if (tcf_chain_held_by_acts_only(chain)) {
1f3ed383 2862 /* The chain exists only because there is
3d32f4c5 2863 * some action referencing it.
1f3ed383
JP		 2864 */
2865 tcf_chain_hold(chain);
2866 } else {
2867 NL_SET_ERR_MSG(extack, "Filter chain already exists");
e368fdb6 2868 err = -EEXIST;
2cbfab07 2869 goto errout_block_locked;
1f3ed383
JP		 2870 }
2871 } else {
2872 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2873 NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
e368fdb6 2874 err = -ENOENT;
2cbfab07 2875 goto errout_block_locked;
1f3ed383
JP		 2876 }
2877 chain = tcf_chain_create(block, chain_index);
2878 if (!chain) {
2879 NL_SET_ERR_MSG(extack, "Failed to create filter chain");
e368fdb6 2880 err = -ENOMEM;
2cbfab07 2881 goto errout_block_locked;
1f3ed383 2882 }
32a4f5ec
JP		 2883 }
2884 } else {
3d32f4c5 2885 if (!chain || tcf_chain_held_by_acts_only(chain)) {
32a4f5ec 2886 NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
e368fdb6 2887 err = -EINVAL;
2cbfab07 2888 goto errout_block_locked;
32a4f5ec
JP		 2889 }
2890 tcf_chain_hold(chain);
2891 }
2892
2cbfab07
VB		 2893 if (n->nlmsg_type == RTM_NEWCHAIN) {
2894 /* Modifying chain requires holding parent block lock. In case
2895 * the chain was successfully added, take a reference to the
2896 * chain. This ensures that an empty chain does not disappear at
2897 * the end of this function.
2898 */
2899 tcf_chain_hold(chain);
2900 chain->explicitly_created = true;
2901 }
2902 mutex_unlock(&block->lock);
2903
32a4f5ec
JP		 2904 switch (n->nlmsg_type) {
2905 case RTM_NEWCHAIN:
9f407f17 2906 err = tc_chain_tmplt_add(chain, net, tca, extack);
2cbfab07
VB		 2907 if (err) {
2908 tcf_chain_put_explicitly_created(chain);
9f407f17 2909 goto errout;
2cbfab07
VB		 2910 }
2911
32a4f5ec
JP		 2912 tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
2913 RTM_NEWCHAIN, false);
2914 break;
2915 case RTM_DELCHAIN:
f5b9bac7 2916 tfilter_notify_chain(net, skb, block, q, parent, n,
0fca55ed 2917 chain, RTM_DELTFILTER);
32a4f5ec 2918 /* Flush the chain first as the user requested chain removal. */
12db03b6 2919 tcf_chain_flush(chain, true);
32a4f5ec
JP		 2920 /* In case the chain was successfully deleted, put a reference
2921 * to the chain previously taken during addition.
2922 */
2923 tcf_chain_put_explicitly_created(chain);
2924 break;
2925 case RTM_GETCHAIN:
32a4f5ec 2926 err = tc_chain_notify(chain, skb, n->nlmsg_seq,
9d85a6f4 2927 n->nlmsg_flags, n->nlmsg_type, true);
32a4f5ec
JP		 2928 if (err < 0)
2929 NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
2930 break;
2931 default:
2932 err = -EOPNOTSUPP;
2933 NL_SET_ERR_MSG(extack, "Unsupported message type");
2934 goto errout;
2935 }
2936
2937errout:
2938 tcf_chain_put(chain);
e368fdb6 2939errout_block:
12db03b6 2940 tcf_block_release(q, block, true);
32a4f5ec
JP		 2941 if (err == -EAGAIN)
2942 /* Replay the request. */
2943 goto replay;
2944 return err;
2cbfab07
VB		 2945
2946errout_block_locked:
2947 mutex_unlock(&block->lock);
2948 goto errout_block;
32a4f5ec
JP		 2949}
2950
2951/* called with RTNL */
2952static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
2953{
2954 struct net *net = sock_net(skb->sk);
2955 struct nlattr *tca[TCA_MAX + 1];
2956 struct Qdisc *q = NULL;
2957 struct tcf_block *block;
32a4f5ec 2958 struct tcmsg *tcm = nlmsg_data(cb->nlh);
ace4a267 2959 struct tcf_chain *chain;
32a4f5ec
JP		 2960 long index_start;
2961 long index;
32a4f5ec
JP		 2962 int err;
2963
2964 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2965 return skb->len;
2966
8cb08174
JB		 2967 err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2968 rtm_tca_policy, cb->extack);
32a4f5ec
JP		 2969 if (err)
2970 return err;
2971
2972 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
787ce6d0 2973 block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
32a4f5ec
JP		 2974 if (!block)
2975 goto out;
32a4f5ec
JP		 2976 } else {
2977 const struct Qdisc_class_ops *cops;
2978 struct net_device *dev;
2979 unsigned long cl = 0;
2980
2981 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2982 if (!dev)
2983 return skb->len;
2984
0ad41b24 2985 if (!tcm->tcm_parent)
5891cd5e 2986 q = rtnl_dereference(dev->qdisc);
0ad41b24 2987 else
32a4f5ec 2988 q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
0ad41b24 2989
32a4f5ec
JP		 2990 if (!q)
2991 goto out;
2992 cops = q->ops->cl_ops;
2993 if (!cops)
2994 goto out;
2995 if (!cops->tcf_block)
2996 goto out;
2997 if (TC_H_MIN(tcm->tcm_parent)) {
2998 cl = cops->find(q, tcm->tcm_parent);
2999 if (cl == 0)
3000 goto out;
3001 }
3002 block = cops->tcf_block(q, cl, NULL);
3003 if (!block)
3004 goto out;
3005 if (tcf_block_shared(block))
3006 q = NULL;
3007 }
3008
3009 index_start = cb->args[0];
3010 index = 0;
3011
ace4a267
VB		 3012 mutex_lock(&block->lock);
3013 list_for_each_entry(chain, &block->chain_list, list) {
32a4f5ec
JP		 3014 if ((tca[TCA_CHAIN] &&
3015 nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3016 continue;
3017 if (index < index_start) {
3018 index++;
3019 continue;
3020 }
ace4a267
VB		 3021 if (tcf_chain_held_by_acts_only(chain))
3022 continue;
a5654820
VB		 3023 err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3024 chain->index, net, skb, block,
32a4f5ec
JP		 3025 NETLINK_CB(cb->skb).portid,
3026 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3027 RTM_NEWCHAIN);
ace4a267 3028 if (err <= 0)
32a4f5ec
JP		 3029 break;
3030 index++;
3031 }
ace4a267 3032 mutex_unlock(&block->lock);
32a4f5ec 3033
787ce6d0 3034 if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
12db03b6 3035 tcf_block_refcnt_put(block, true);
32a4f5ec
JP		 3036 cb->args[0] = index;
3037
3038out:
3039 /* If we did no progress, the error (EMSGSIZE) is real */
3040 if (skb->len == 0 && err)
3041 return err;
3042 return skb->len;
3043}
3044
18d0264f 3045void tcf_exts_destroy(struct tcf_exts *exts)
1da177e4
LT		 3046{
3047#ifdef CONFIG_NET_CLS_ACT
3d66b89c
ED		 3048 if (exts->actions) {
3049 tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3050 kfree(exts->actions);
3051 }
22dc13c8 3052 exts->nr_actions = 0;
1da177e4
LT		 3053#endif
3054}
aa767bfe 3055EXPORT_SYMBOL(tcf_exts_destroy);
1da177e4 3056
c86e0209
BZ		 3057int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3058 struct nlattr *rate_tlv, struct tcf_exts *exts,
3059 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
1da177e4 3060{
1da177e4
LT		 3061#ifdef CONFIG_NET_CLS_ACT
3062 {
87c750e8 3063 int init_res[TCA_ACT_MAX_PRIO] = {};
1da177e4 3064 struct tc_action *act;
d04e6990 3065 size_t attr_size = 0;
1da177e4 3066
5da57f42 3067 if (exts->police && tb[exts->police]) {
d349f997
CW		 3068 struct tc_action_ops *a_o;
3069
695176bf
CW		 3070 a_o = tc_action_load_ops(tb[exts->police], true,
3071 !(flags & TCA_ACT_FLAGS_NO_RTNL),
3072 extack);
d349f997
CW		 3073 if (IS_ERR(a_o))
3074 return PTR_ERR(a_o);
695176bf 3075 flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
9fb9f251 3076 act = tcf_action_init_1(net, tp, tb[exts->police],
695176bf
CW		 3077 rate_tlv, a_o, init_res, flags,
3078 extack);
b3650bf7
VB		 3079 module_put(a_o->owner);
3080 if (IS_ERR(act))
ab27cfb8 3081 return PTR_ERR(act);
1da177e4 3082
33be6271 3083 act->type = exts->type = TCA_OLD_COMPAT;
22dc13c8
WC		 3084 exts->actions[0] = act;
3085 exts->nr_actions = 1;
396d7f23 3086 tcf_idr_insert_many(exts->actions);
5da57f42 3087 } else if (exts->action && tb[exts->action]) {
90b73b77 3088 int err;
22dc13c8 3089
695176bf 3090 flags |= TCA_ACT_FLAGS_BIND;
9fb9f251 3091 err = tcf_action_init(net, tp, tb[exts->action],
695176bf 3092 rate_tlv, exts->actions, init_res,
c86e0209
BZ		 3093 &attr_size, flags, fl_flags,
3094 extack);
90b73b77 3095 if (err < 0)
33be6271 3096 return err;
90b73b77 3097 exts->nr_actions = err;
1da177e4
LT		 3098 }
3099 }
1da177e4 3100#else
5da57f42 3101 if ((exts->action && tb[exts->action]) ||
50a56190
AA		 3102 (exts->police && tb[exts->police])) {
3103 NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
1da177e4 3104 return -EOPNOTSUPP;
50a56190 3105 }
1da177e4
LT		 3106#endif
3107
3108 return 0;
3109}
c86e0209
BZ		 3110EXPORT_SYMBOL(tcf_exts_validate_ex);
3111
3112int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3113 struct nlattr *rate_tlv, struct tcf_exts *exts,
3114 u32 flags, struct netlink_ext_ack *extack)
3115{
3116 return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3117 flags, 0, extack);
3118}
aa767bfe 3119EXPORT_SYMBOL(tcf_exts_validate);
1da177e4 3120
9b0d4446 3121void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
1da177e4
LT		 3122{
3123#ifdef CONFIG_NET_CLS_ACT
22dc13c8
WC		 3124 struct tcf_exts old = *dst;
3125
9b0d4446 3126 *dst = *src;
22dc13c8 3127 tcf_exts_destroy(&old);
1da177e4
LT		 3128#endif
3129}
aa767bfe 3130EXPORT_SYMBOL(tcf_exts_change);
1da177e4 3131
22dc13c8
WC		 3132#ifdef CONFIG_NET_CLS_ACT
3133static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3134{
3135 if (exts->nr_actions == 0)
3136 return NULL;
3137 else
3138 return exts->actions[0];
3139}
3140#endif
33be6271 3141
5da57f42 3142int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
1da177e4
LT		 3143{
3144#ifdef CONFIG_NET_CLS_ACT
9cc63db5
CW		 3145 struct nlattr *nest;
3146
978dfd8d 3147 if (exts->action && tcf_exts_has_actions(exts)) {
1da177e4
LT		 3148 /*
3149 * again for backward compatible mode - we want
3150 * to work with both old and new modes of entering
3151 * tc data even if iproute2 was newer - jhs
3152 */
33be6271 3153 if (exts->type != TCA_OLD_COMPAT) {
ae0be8de 3154 nest = nla_nest_start_noflag(skb, exts->action);
4b3550ef
PM		 3155 if (nest == NULL)
3156 goto nla_put_failure;
22dc13c8 3157
ca44b738
VB		 3158 if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3159 < 0)
add93b61 3160 goto nla_put_failure;
4b3550ef 3161 nla_nest_end(skb, nest);
5da57f42 3162 } else if (exts->police) {
33be6271 3163 struct tc_action *act = tcf_exts_first_act(exts);
ae0be8de 3164 nest = nla_nest_start_noflag(skb, exts->police);
63acd680 3165 if (nest == NULL || !act)
4b3550ef 3166 goto nla_put_failure;
33be6271 3167 if (tcf_action_dump_old(skb, act, 0, 0) < 0)
add93b61 3168 goto nla_put_failure;
4b3550ef 3169 nla_nest_end(skb, nest);
1da177e4
LT		 3170 }
3171 }
1da177e4 3172 return 0;
9cc63db5
CW		 3173
3174nla_put_failure:
3175 nla_nest_cancel(skb, nest);
1da177e4 3176 return -1;
9cc63db5
CW		 3177#else
3178 return 0;
3179#endif
1da177e4 3180}
aa767bfe 3181EXPORT_SYMBOL(tcf_exts_dump);
1da177e4 3182
ca44b738
VB		 3183int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3184{
3185#ifdef CONFIG_NET_CLS_ACT
3186 struct nlattr *nest;
3187
3188 if (!exts->action || !tcf_exts_has_actions(exts))
3189 return 0;
3190
3191 nest = nla_nest_start_noflag(skb, exts->action);
3192 if (!nest)
3193 goto nla_put_failure;
3194
3195 if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3196 goto nla_put_failure;
3197 nla_nest_end(skb, nest);
3198 return 0;
3199
3200nla_put_failure:
3201 nla_nest_cancel(skb, nest);
3202 return -1;
3203#else
3204 return 0;
3205#endif
3206}
3207EXPORT_SYMBOL(tcf_exts_terse_dump);
aa767bfe 3208
5da57f42 3209int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
1da177e4
LT		 3210{
3211#ifdef CONFIG_NET_CLS_ACT
33be6271 3212 struct tc_action *a = tcf_exts_first_act(exts);
b057df24 3213 if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
33be6271 3214 return -1;
1da177e4
LT		 3215#endif
3216 return 0;
1da177e4 3217}
aa767bfe 3218EXPORT_SYMBOL(tcf_exts_dump_stats);
1da177e4 3219
40119211
VB		 3220static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3221{
3222 if (*flags & TCA_CLS_FLAGS_IN_HW)
3223 return;
3224 *flags |= TCA_CLS_FLAGS_IN_HW;
3225 atomic_inc(&block->offloadcnt);
3226}
3227
3228static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3229{
3230 if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3231 return;
3232 *flags &= ~TCA_CLS_FLAGS_IN_HW;
3233 atomic_dec(&block->offloadcnt);
3234}
3235
3236static void tc_cls_offload_cnt_update(struct tcf_block *block,
3237 struct tcf_proto *tp, u32 *cnt,
3238 u32 *flags, u32 diff, bool add)
3239{
3240 lockdep_assert_held(&block->cb_lock);
3241
3242 spin_lock(&tp->lock);
3243 if (add) {
3244 if (!*cnt)
3245 tcf_block_offload_inc(block, flags);
3246 *cnt += diff;
3247 } else {
3248 *cnt -= diff;
3249 if (!*cnt)
3250 tcf_block_offload_dec(block, flags);
3251 }
3252 spin_unlock(&tp->lock);
3253}
3254
3255static void
3256tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3257 u32 *cnt, u32 *flags)
3258{
3259 lockdep_assert_held(&block->cb_lock);
3260
3261 spin_lock(&tp->lock);
3262 tcf_block_offload_dec(block, flags);
3263 *cnt = 0;
3264 spin_unlock(&tp->lock);
3265}
3266
3267static int
3268__tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3269 void *type_data, bool err_stop)
717503b9 3270{
955bcb6e 3271 struct flow_block_cb *block_cb;
aeb3fecd
CW		 3272 int ok_count = 0;
3273 int err;
3274
14bfb13f 3275 list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
aeb3fecd
CW		 3276 err = block_cb->cb(type, type_data, block_cb->cb_priv);
3277 if (err) {
40119211
VB		 3278 if (err_stop)
3279 return err;
aeb3fecd
CW		 3280 } else {
3281 ok_count++;
3282 }
3283 }
40119211
VB		 3284 return ok_count;
3285}
3286
3287int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3288 void *type_data, bool err_stop, bool rtnl_held)
3289{
11bd634d 3290 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
40119211
VB		 3291 int ok_count;
3292
11bd634d
VB		 3293retry:
3294 if (take_rtnl)
3295 rtnl_lock();
40119211 3296 down_read(&block->cb_lock);
11bd634d
VB		 3297 /* Need to obtain rtnl lock if block is bound to devs that require it.
3298 * In block bind code cb_lock is obtained while holding rtnl, so we must
3299 * obtain the locks in same order here.
3300 */
3301 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3302 up_read(&block->cb_lock);
3303 take_rtnl = true;
3304 goto retry;
3305 }
3306
40119211 3307 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
11bd634d 3308
4f8116c8 3309 up_read(&block->cb_lock);
11bd634d
VB		 3310 if (take_rtnl)
3311 rtnl_unlock();
aeb3fecd 3312 return ok_count;
717503b9
JP		 3313}
3314EXPORT_SYMBOL(tc_setup_cb_call);
b3f55bdd 3315
40119211
VB		 3316/* Non-destructive filter add. If filter that wasn't already in hardware is
3317 * successfully offloaded, increment block offloads counter. On failure,
3318 * previously offloaded filter is considered to be intact and offloads counter
3319 * is not decremented.
3320 */
3321
3322int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3323 enum tc_setup_type type, void *type_data, bool err_stop,
3324 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3325{
11bd634d 3326 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
40119211
VB		 3327 int ok_count;
3328
11bd634d
VB		 3329retry:
3330 if (take_rtnl)
3331 rtnl_lock();
40119211 3332 down_read(&block->cb_lock);
11bd634d
VB		 3333 /* Need to obtain rtnl lock if block is bound to devs that require it.
3334 * In block bind code cb_lock is obtained while holding rtnl, so we must
3335 * obtain the locks in same order here.
3336 */
3337 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3338 up_read(&block->cb_lock);
3339 take_rtnl = true;
3340 goto retry;
3341 }
3342
40119211
VB		 3343 /* Make sure all netdevs sharing this block are offload-capable. */
3344 if (block->nooffloaddevcnt && err_stop) {
3345 ok_count = -EOPNOTSUPP;
3346 goto err_unlock;
3347 }
3348
3349 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
a449a3e7
VB		 3350 if (ok_count < 0)
3351 goto err_unlock;
3352
3353 if (tp->ops->hw_add)
3354 tp->ops->hw_add(tp, type_data);
40119211
VB		 3355 if (ok_count > 0)
3356 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3357 ok_count, true);
3358err_unlock:
3359 up_read(&block->cb_lock);
11bd634d
VB		 3360 if (take_rtnl)
3361 rtnl_unlock();
c48c94b0 3362 return min(ok_count, 0);
40119211
VB		 3363}
3364EXPORT_SYMBOL(tc_setup_cb_add);
3365
3366/* Destructive filter replace. If filter that wasn't already in hardware is
3367 * successfully offloaded, increment block offload counter. On failure,
3368 * previously offloaded filter is considered to be destroyed and offload counter
3369 * is decremented.
3370 */
3371
3372int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3373 enum tc_setup_type type, void *type_data, bool err_stop,
3374 u32 *old_flags, unsigned int *old_in_hw_count,
3375 u32 *new_flags, unsigned int *new_in_hw_count,
3376 bool rtnl_held)
3377{
11bd634d 3378 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
40119211
VB		 3379 int ok_count;
3380
11bd634d
VB		 3381retry:
3382 if (take_rtnl)
3383 rtnl_lock();
40119211 3384 down_read(&block->cb_lock);
11bd634d
VB		 3385 /* Need to obtain rtnl lock if block is bound to devs that require it.
3386 * In block bind code cb_lock is obtained while holding rtnl, so we must
3387 * obtain the locks in same order here.
3388 */
3389 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3390 up_read(&block->cb_lock);
3391 take_rtnl = true;
3392 goto retry;
3393 }
3394
40119211
VB		 3395 /* Make sure all netdevs sharing this block are offload-capable. */
3396 if (block->nooffloaddevcnt && err_stop) {
3397 ok_count = -EOPNOTSUPP;
3398 goto err_unlock;
3399 }
3400
3401 tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
a449a3e7
VB		 3402 if (tp->ops->hw_del)
3403 tp->ops->hw_del(tp, type_data);
40119211
VB		 3404
3405 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
a449a3e7
VB		 3406 if (ok_count < 0)
3407 goto err_unlock;
3408
3409 if (tp->ops->hw_add)
3410 tp->ops->hw_add(tp, type_data);
40119211 3411 if (ok_count > 0)
a449a3e7
VB		 3412 tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3413 new_flags, ok_count, true);
40119211
VB		 3414err_unlock:
3415 up_read(&block->cb_lock);
11bd634d
VB		 3416 if (take_rtnl)
3417 rtnl_unlock();
c48c94b0 3418 return min(ok_count, 0);
40119211
VB		 3419}
3420EXPORT_SYMBOL(tc_setup_cb_replace);
3421
3422/* Destroy filter and decrement block offload counter, if filter was previously
3423 * offloaded.
3424 */
3425
3426int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3427 enum tc_setup_type type, void *type_data, bool err_stop,
3428 u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3429{
11bd634d 3430 bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
40119211
VB		 3431 int ok_count;
3432
11bd634d
VB		 3433retry:
3434 if (take_rtnl)
3435 rtnl_lock();
40119211 3436 down_read(&block->cb_lock);
11bd634d
VB		 3437 /* Need to obtain rtnl lock if block is bound to devs that require it.
3438 * In block bind code cb_lock is obtained while holding rtnl, so we must
3439 * obtain the locks in same order here.
3440 */
3441 if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3442 up_read(&block->cb_lock);
3443 take_rtnl = true;
3444 goto retry;
3445 }
3446
40119211
VB		 3447 ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3448
3449 tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
a449a3e7
VB		 3450 if (tp->ops->hw_del)
3451 tp->ops->hw_del(tp, type_data);
3452
40119211 3453 up_read(&block->cb_lock);
11bd634d
VB		 3454 if (take_rtnl)
3455 rtnl_unlock();
c48c94b0 3456 return min(ok_count, 0);
40119211
VB		 3457}
3458EXPORT_SYMBOL(tc_setup_cb_destroy);
3459
3460int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3461 bool add, flow_setup_cb_t *cb,
3462 enum tc_setup_type type, void *type_data,
3463 void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3464{
3465 int err = cb(type, type_data, cb_priv);
3466
3467 if (err) {
3468 if (add && tc_skip_sw(*flags))
3469 return err;
3470 } else {
3471 tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3472 add);
3473 }
3474
3475 return 0;
3476}
3477EXPORT_SYMBOL(tc_setup_cb_reoffload);
3478
2008495d
JP		 3479static int tcf_act_get_cookie(struct flow_action_entry *entry,
3480 const struct tc_action *act)
3481{
3482 struct tc_cookie *cookie;
3483 int err = 0;
3484
3485 rcu_read_lock();
3486 cookie = rcu_dereference(act->act_cookie);
3487 if (cookie) {
3488 entry->cookie = flow_action_cookie_create(cookie->data,
3489 cookie->len,
3490 GFP_ATOMIC);
3491 if (!entry->cookie)
3492 err = -ENOMEM;
3493 }
3494 rcu_read_unlock();
3495 return err;
3496}
3497
3498static void tcf_act_put_cookie(struct flow_action_entry *entry)
3499{
3500 flow_action_cookie_destroy(entry->cookie);
3501}
3502
9c1c0e12 3503void tc_cleanup_offload_action(struct flow_action *flow_action)
5a6ff4b1
VB		 3504{
3505 struct flow_action_entry *entry;
3506 int i;
3507
2008495d
JP		 3508 flow_action_for_each(i, entry, flow_action) {
3509 tcf_act_put_cookie(entry);
1158958a
VB		 3510 if (entry->destructor)
3511 entry->destructor(entry->destructor_priv);
2008495d 3512 }
5a6ff4b1 3513}
9c1c0e12 3514EXPORT_SYMBOL(tc_cleanup_offload_action);
5a6ff4b1 3515
c54e1d92
BZ		 3516static int tc_setup_offload_act(struct tc_action *act,
3517 struct flow_action_entry *entry,
c2ccf84e
IS		 3518 u32 *index_inc,
3519 struct netlink_ext_ack *extack)
1158958a 3520{
470d5060 3521#ifdef CONFIG_NET_CLS_ACT
c440615f 3522 if (act->ops->offload_act_setup) {
c2ccf84e
IS		 3523 return act->ops->offload_act_setup(act, entry, index_inc, true,
3524 extack);
c440615f
IS		 3525 } else {
3526 NL_SET_ERR_MSG(extack, "Action does not support offload");
c54e1d92 3527 return -EOPNOTSUPP;
c440615f 3528 }
c54e1d92 3529#else
1158958a 3530 return 0;
4a5da47d
VB		 3531#endif
3532}
3533
8cbfe939 3534int tc_setup_action(struct flow_action *flow_action,
c2ccf84e
IS		 3535 struct tc_action *actions[],
3536 struct netlink_ext_ack *extack)
3a7b6861 3537{
c0f47c28 3538 int i, j, k, index, err = 0;
7a472814 3539 struct tc_action *act;
3a7b6861 3540
0dfb2d82
JK		 3541 BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3542 BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3543 BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
44f86580 3544
8cbfe939 3545 if (!actions)
3a7b6861
PNA		 3546 return 0;
3547
3548 j = 0;
8cbfe939 3549 tcf_act_for_each_action(i, act, actions) {
3a7b6861
PNA		 3550 struct flow_action_entry *entry;
3551
3552 entry = &flow_action->entries[j];
7a472814 3553 spin_lock_bh(&act->tcfa_lock);
2008495d
JP		 3554 err = tcf_act_get_cookie(entry, act);
3555 if (err)
3556 goto err_out_locked;
44f86580 3557
c54e1d92 3558 index = 0;
c2ccf84e 3559 err = tc_setup_offload_act(act, entry, &index, extack);
c0f47c28 3560 if (err)
7a472814 3561 goto err_out_locked;
c0f47c28
OS		 3562
3563 for (k = 0; k < index ; k++) {
3564 entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3565 entry[k].hw_index = act->tcfa_index;
3566 }
3567
3568 j += index;
3569
7a472814 3570 spin_unlock_bh(&act->tcfa_lock);
3a7b6861 3571 }
9838b20a 3572
3a7b6861 3573err_out:
5a6ff4b1 3574 if (err)
9c1c0e12 3575 tc_cleanup_offload_action(flow_action);
5a6ff4b1 3576
9838b20a 3577 return err;
7a472814
VB		 3578err_out_locked:
3579 spin_unlock_bh(&act->tcfa_lock);
3580 goto err_out;
3a7b6861 3581}
8cbfe939
BZ		 3582
3583int tc_setup_offload_action(struct flow_action *flow_action,
c2ccf84e
IS		 3584 const struct tcf_exts *exts,
3585 struct netlink_ext_ack *extack)
8cbfe939
BZ		 3586{
3587#ifdef CONFIG_NET_CLS_ACT
3588 if (!exts)
3589 return 0;
3590
c2ccf84e 3591 return tc_setup_action(flow_action, exts->actions, extack);
8cbfe939
BZ		 3592#else
3593 return 0;
3594#endif
3595}
9c1c0e12 3596EXPORT_SYMBOL(tc_setup_offload_action);
3a7b6861 3597
e3ab786b
PNA		 3598unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3599{
3600 unsigned int num_acts = 0;
3601 struct tc_action *act;
3602 int i;
3603
3604 tcf_exts_for_each_action(i, act, exts) {
3605 if (is_tcf_pedit(act))
3606 num_acts += tcf_pedit_nkeys(act);
3607 else
3608 num_acts++;
3609 }
3610 return num_acts;
3611}
3612EXPORT_SYMBOL(tcf_exts_num_actions);
3613
3625750f
PM		 3614#ifdef CONFIG_NET_CLS_ACT
3615static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3616 u32 *p_block_index,
3617 struct netlink_ext_ack *extack)
3618{
3619 *p_block_index = nla_get_u32(block_index_attr);
3620 if (!*p_block_index) {
3621 NL_SET_ERR_MSG(extack, "Block number may not be zero");
3622 return -EINVAL;
3623 }
3624
3625 return 0;
3626}
3627
3628int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3629 enum flow_block_binder_type binder_type,
3630 struct nlattr *block_index_attr,
3631 struct netlink_ext_ack *extack)
3632{
3633 u32 block_index;
3634 int err;
3635
3636 if (!block_index_attr)
3637 return 0;
3638
3639 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3640 if (err)
3641 return err;
3642
3643 if (!block_index)
3644 return 0;
3645
3646 qe->info.binder_type = binder_type;
3647 qe->info.chain_head_change = tcf_chain_head_change_dflt;
3648 qe->info.chain_head_change_priv = &qe->filter_chain;
3649 qe->info.block_index = block_index;
3650
3651 return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3652}
3653EXPORT_SYMBOL(tcf_qevent_init);
3654
3655void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3656{
3657 if (qe->info.block_index)
3658 tcf_block_put_ext(qe->block, sch, &qe->info);
3659}
3660EXPORT_SYMBOL(tcf_qevent_destroy);
3661
3662int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3663 struct netlink_ext_ack *extack)
3664{
3665 u32 block_index;
3666 int err;
3667
3668 if (!block_index_attr)
3669 return 0;
3670
3671 err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3672 if (err)
3673 return err;
3674
3675 /* Bounce newly-configured block or change in block. */
3676 if (block_index != qe->info.block_index) {
3677 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3678 return -EINVAL;
3679 }
3680
3681 return 0;
3682}
3683EXPORT_SYMBOL(tcf_qevent_validate_change);
3684
3685struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
55f656cd 3686 struct sk_buff **to_free, int *ret)
3625750f
PM		 3687{
3688 struct tcf_result cl_res;
3689 struct tcf_proto *fl;
3690
3691 if (!qe->info.block_index)
3692 return skb;
3693
3694 fl = rcu_dereference_bh(qe->filter_chain);
3695
3aa26055 3696 switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3625750f
PM		 3697 case TC_ACT_SHOT:
3698 qdisc_qstats_drop(sch);
3699 __qdisc_drop(skb, to_free);
3700 *ret = __NET_XMIT_BYPASS;
3701 return NULL;
3702 case TC_ACT_STOLEN:
3703 case TC_ACT_QUEUED:
3704 case TC_ACT_TRAP:
3705 __qdisc_drop(skb, to_free);
3706 *ret = __NET_XMIT_STOLEN;
3707 return NULL;
3708 case TC_ACT_REDIRECT:
3709 skb_do_redirect(skb);
3710 *ret = __NET_XMIT_STOLEN;
3711 return NULL;
3712 }
3713
3625750f
PM		 3714 return skb;
3715}
3716EXPORT_SYMBOL(tcf_qevent_handle);
3717
3718int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3719{
3720 if (!qe->info.block_index)
3721 return 0;
3722 return nla_put_u32(skb, attr_name, qe->info.block_index);
3723}
3724EXPORT_SYMBOL(tcf_qevent_dump);
3725#endif
3726
48617387
JP		 3727static __net_init int tcf_net_init(struct net *net)
3728{
3729 struct tcf_net *tn = net_generic(net, tcf_net_id);
3730
ab281629 3731 spin_lock_init(&tn->idr_lock);
48617387
JP		 3732 idr_init(&tn->idr);
3733 return 0;
3734}
3735
3736static void __net_exit tcf_net_exit(struct net *net)
3737{
3738 struct tcf_net *tn = net_generic(net, tcf_net_id);
3739
3740 idr_destroy(&tn->idr);
3741}
3742
3743static struct pernet_operations tcf_net_ops = {
3744 .init = tcf_net_init,
3745 .exit = tcf_net_exit,
3746 .id = &tcf_net_id,
3747 .size = sizeof(struct tcf_net),
3748};
3749
1da177e4
LT		 3750static int __init tc_filter_init(void)
3751{
48617387
JP		 3752 int err;
3753
7aa0045d
CW		 3754 tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3755 if (!tc_filter_wq)
3756 return -ENOMEM;
3757
48617387
JP		 3758 err = register_pernet_subsys(&tcf_net_ops);
3759 if (err)
3760 goto err_register_pernet_subsys;
3761
470502de
VB		 3762 rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3763 RTNL_FLAG_DOIT_UNLOCKED);
3764 rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3765 RTNL_FLAG_DOIT_UNLOCKED);
c431f89b 3766 rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
470502de 3767 tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
32a4f5ec
JP		 3768 rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3769 rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3770 rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3771 tc_dump_chain, 0);
1da177e4 3772
1da177e4 3773 return 0;
48617387
JP		 3774
3775err_register_pernet_subsys:
3776 destroy_workqueue(tc_filter_wq);
3777 return err;
1da177e4
LT		 3778}
3779
3780subsys_initcall(tc_filter_init);
Linux 6.x block layer and io_uring tree(s)