1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_PKT_CLS_H
3 #define __NET_PKT_CLS_H
5 #include <linux/pkt_cls.h>
6 #include <linux/workqueue.h>
7 #include <net/sch_generic.h>
8 #include <net/act_api.h>
9 #include <net/flow_offload.h>
10 #include <net/net_namespace.h>
12 /* TC action not accessible from user space */
13 #define TC_ACT_CONSUMED (TC_ACT_VALUE_MAX + 1)
15 /* Basic packet classifier frontend definitions. */
23 int (*fn)(struct tcf_proto *, void *node, struct tcf_walker *);
26 int register_tcf_proto_ops(struct tcf_proto_ops *ops);
27 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops);
29 struct tcf_block_ext_info {
30 enum flow_block_binder_type binder_type;
31 tcf_chain_head_change_t *chain_head_change;
32 void *chain_head_change_priv;
37 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func);
40 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block,
42 void tcf_chain_put_by_act(struct tcf_chain *chain);
43 struct tcf_chain *tcf_get_next_chain(struct tcf_block *block,
44 struct tcf_chain *chain);
45 struct tcf_proto *tcf_get_next_proto(struct tcf_chain *chain,
46 struct tcf_proto *tp, bool rtnl_held);
47 void tcf_block_netif_keep_dst(struct tcf_block *block);
48 int tcf_block_get(struct tcf_block **p_block,
49 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
50 struct netlink_ext_ack *extack);
51 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
52 struct tcf_block_ext_info *ei,
53 struct netlink_ext_ack *extack);
54 void tcf_block_put(struct tcf_block *block);
55 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
56 struct tcf_block_ext_info *ei);
58 static inline bool tcf_block_shared(struct tcf_block *block)
63 static inline bool tcf_block_non_null_shared(struct tcf_block *block)
65 return block && block->index;
68 static inline struct Qdisc *tcf_block_q(struct tcf_block *block)
70 WARN_ON(tcf_block_shared(block));
74 int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
75 tc_indr_block_bind_cb_t *cb, void *cb_ident);
76 int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
77 tc_indr_block_bind_cb_t *cb, void *cb_ident);
78 void __tc_indr_block_cb_unregister(struct net_device *dev,
79 tc_indr_block_bind_cb_t *cb, void *cb_ident);
80 void tc_indr_block_cb_unregister(struct net_device *dev,
81 tc_indr_block_bind_cb_t *cb, void *cb_ident);
83 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
84 struct tcf_result *res, bool compat_mode);
87 static inline bool tcf_block_shared(struct tcf_block *block)
92 static inline bool tcf_block_non_null_shared(struct tcf_block *block)
98 int tcf_block_get(struct tcf_block **p_block,
99 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
100 struct netlink_ext_ack *extack)
106 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
107 struct tcf_block_ext_info *ei,
108 struct netlink_ext_ack *extack)
113 static inline void tcf_block_put(struct tcf_block *block)
118 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
119 struct tcf_block_ext_info *ei)
123 static inline struct Qdisc *tcf_block_q(struct tcf_block *block)
129 int tc_setup_cb_block_register(struct tcf_block *block, tc_setup_cb_t *cb,
136 void tc_setup_cb_block_unregister(struct tcf_block *block, tc_setup_cb_t *cb,
142 int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
143 tc_indr_block_bind_cb_t *cb, void *cb_ident)
149 int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
150 tc_indr_block_bind_cb_t *cb, void *cb_ident)
156 void __tc_indr_block_cb_unregister(struct net_device *dev,
157 tc_indr_block_bind_cb_t *cb, void *cb_ident)
162 void tc_indr_block_cb_unregister(struct net_device *dev,
163 tc_indr_block_bind_cb_t *cb, void *cb_ident)
167 static inline int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
168 struct tcf_result *res, bool compat_mode)
170 return TC_ACT_UNSPEC;
174 static inline unsigned long
175 __cls_set_class(unsigned long *clp, unsigned long cl)
177 return xchg(clp, cl);
180 static inline unsigned long
181 cls_set_class(struct Qdisc *q, unsigned long *clp, unsigned long cl)
183 unsigned long old_cl;
186 old_cl = __cls_set_class(clp, cl);
192 tcf_bind_filter(struct tcf_proto *tp, struct tcf_result *r, unsigned long base)
194 struct Qdisc *q = tp->chain->block->q;
197 /* Check q as it is not set for shared blocks. In that case,
198 * setting class is not supported.
202 cl = q->ops->cl_ops->bind_tcf(q, base, r->classid);
203 cl = cls_set_class(q, &r->class, cl);
205 q->ops->cl_ops->unbind_tcf(q, cl);
209 tcf_unbind_filter(struct tcf_proto *tp, struct tcf_result *r)
211 struct Qdisc *q = tp->chain->block->q;
216 if ((cl = __cls_set_class(&r->class, 0)) != 0)
217 q->ops->cl_ops->unbind_tcf(q, cl);
221 #ifdef CONFIG_NET_CLS_ACT
222 __u32 type; /* for backward compat(TCA_OLD_COMPAT) */
224 struct tc_action **actions;
227 /* Map to export classifier specific extension TLV types to the
228 * generic extensions API. Unsupported extensions must be set to 0.
234 static inline int tcf_exts_init(struct tcf_exts *exts, struct net *net,
235 int action, int police)
237 #ifdef CONFIG_NET_CLS_ACT
239 exts->nr_actions = 0;
241 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
246 exts->action = action;
247 exts->police = police;
251 /* Return false if the netns is being destroyed in cleanup_net(). Callers
252 * need to do cleanup synchronously in this case, otherwise may race with
253 * tc_action_net_exit(). Return true for other cases.
255 static inline bool tcf_exts_get_net(struct tcf_exts *exts)
257 #ifdef CONFIG_NET_CLS_ACT
258 exts->net = maybe_get_net(exts->net);
259 return exts->net != NULL;
265 static inline void tcf_exts_put_net(struct tcf_exts *exts)
267 #ifdef CONFIG_NET_CLS_ACT
273 #ifdef CONFIG_NET_CLS_ACT
274 #define tcf_exts_for_each_action(i, a, exts) \
275 for (i = 0; i < TCA_ACT_MAX_PRIO && ((a) = (exts)->actions[i]); i++)
277 #define tcf_exts_for_each_action(i, a, exts) \
278 for (; 0; (void)(i), (void)(a), (void)(exts))
282 tcf_exts_stats_update(const struct tcf_exts *exts,
283 u64 bytes, u64 packets, u64 lastuse)
285 #ifdef CONFIG_NET_CLS_ACT
290 for (i = 0; i < exts->nr_actions; i++) {
291 struct tc_action *a = exts->actions[i];
293 tcf_action_stats_update(a, bytes, packets, lastuse, true);
301 * tcf_exts_has_actions - check if at least one action is present
302 * @exts: tc filter extensions handle
304 * Returns true if at least one action is present.
306 static inline bool tcf_exts_has_actions(struct tcf_exts *exts)
308 #ifdef CONFIG_NET_CLS_ACT
309 return exts->nr_actions;
316 * tcf_exts_exec - execute tc filter extensions
317 * @skb: socket buffer
318 * @exts: tc filter extensions handle
319 * @res: desired result
321 * Executes all configured extensions. Returns TC_ACT_OK on a normal execution,
322 * a negative number if the filter must be considered unmatched or
323 * a positive action code (TC_ACT_*) which must be returned to the
327 tcf_exts_exec(struct sk_buff *skb, struct tcf_exts *exts,
328 struct tcf_result *res)
330 #ifdef CONFIG_NET_CLS_ACT
331 return tcf_action_exec(skb, exts->actions, exts->nr_actions, res);
336 int tcf_exts_validate(struct net *net, struct tcf_proto *tp,
337 struct nlattr **tb, struct nlattr *rate_tlv,
338 struct tcf_exts *exts, bool ovr, bool rtnl_held,
339 struct netlink_ext_ack *extack);
340 void tcf_exts_destroy(struct tcf_exts *exts);
341 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src);
342 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts);
343 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts);
346 * struct tcf_pkt_info - packet information
348 struct tcf_pkt_info {
353 #ifdef CONFIG_NET_EMATCH
355 struct tcf_ematch_ops;
358 * struct tcf_ematch - extended match (ematch)
360 * @matchid: identifier to allow userspace to reidentify a match
361 * @flags: flags specifying attributes and the relation to other matches
362 * @ops: the operations lookup table of the corresponding ematch module
363 * @datalen: length of the ematch specific configuration data
364 * @data: ematch specific data
367 struct tcf_ematch_ops * ops;
369 unsigned int datalen;
375 static inline int tcf_em_is_container(struct tcf_ematch *em)
380 static inline int tcf_em_is_simple(struct tcf_ematch *em)
382 return em->flags & TCF_EM_SIMPLE;
385 static inline int tcf_em_is_inverted(struct tcf_ematch *em)
387 return em->flags & TCF_EM_INVERT;
390 static inline int tcf_em_last_match(struct tcf_ematch *em)
392 return (em->flags & TCF_EM_REL_MASK) == TCF_EM_REL_END;
395 static inline int tcf_em_early_end(struct tcf_ematch *em, int result)
397 if (tcf_em_last_match(em))
400 if (result == 0 && em->flags & TCF_EM_REL_AND)
403 if (result != 0 && em->flags & TCF_EM_REL_OR)
410 * struct tcf_ematch_tree - ematch tree handle
412 * @hdr: ematch tree header supplied by userspace
413 * @matches: array of ematches
415 struct tcf_ematch_tree {
416 struct tcf_ematch_tree_hdr hdr;
417 struct tcf_ematch * matches;
422 * struct tcf_ematch_ops - ematch module operations
424 * @kind: identifier (kind) of this ematch module
425 * @datalen: length of expected configuration data (optional)
426 * @change: called during validation (optional)
427 * @match: called during ematch tree evaluation, must return 1/0
428 * @destroy: called during destroyage (optional)
429 * @dump: called during dumping process (optional)
430 * @owner: owner, must be set to THIS_MODULE
431 * @link: link to previous/next ematch module (internal use)
433 struct tcf_ematch_ops {
436 int (*change)(struct net *net, void *,
437 int, struct tcf_ematch *);
438 int (*match)(struct sk_buff *, struct tcf_ematch *,
439 struct tcf_pkt_info *);
440 void (*destroy)(struct tcf_ematch *);
441 int (*dump)(struct sk_buff *, struct tcf_ematch *);
442 struct module *owner;
443 struct list_head link;
446 int tcf_em_register(struct tcf_ematch_ops *);
447 void tcf_em_unregister(struct tcf_ematch_ops *);
448 int tcf_em_tree_validate(struct tcf_proto *, struct nlattr *,
449 struct tcf_ematch_tree *);
450 void tcf_em_tree_destroy(struct tcf_ematch_tree *);
451 int tcf_em_tree_dump(struct sk_buff *, struct tcf_ematch_tree *, int);
452 int __tcf_em_tree_match(struct sk_buff *, struct tcf_ematch_tree *,
453 struct tcf_pkt_info *);
456 * tcf_em_tree_match - evaulate an ematch tree
458 * @skb: socket buffer of the packet in question
459 * @tree: ematch tree to be used for evaluation
460 * @info: packet information examined by classifier
462 * This function matches @skb against the ematch tree in @tree by going
463 * through all ematches respecting their logic relations returning
464 * as soon as the result is obvious.
466 * Returns 1 if the ematch tree as-one matches, no ematches are configured
467 * or ematch is not enabled in the kernel, otherwise 0 is returned.
469 static inline int tcf_em_tree_match(struct sk_buff *skb,
470 struct tcf_ematch_tree *tree,
471 struct tcf_pkt_info *info)
473 if (tree->hdr.nmatches)
474 return __tcf_em_tree_match(skb, tree, info);
479 #define MODULE_ALIAS_TCF_EMATCH(kind) MODULE_ALIAS("ematch-kind-" __stringify(kind))
481 #else /* CONFIG_NET_EMATCH */
483 struct tcf_ematch_tree {
486 #define tcf_em_tree_validate(tp, tb, t) ((void)(t), 0)
487 #define tcf_em_tree_destroy(t) do { (void)(t); } while(0)
488 #define tcf_em_tree_dump(skb, t, tlv) (0)
489 #define tcf_em_tree_match(skb, t, info) ((void)(info), 1)
491 #endif /* CONFIG_NET_EMATCH */
493 static inline unsigned char * tcf_get_base_ptr(struct sk_buff *skb, int layer)
497 return skb_mac_header(skb);
498 case TCF_LAYER_NETWORK:
499 return skb_network_header(skb);
500 case TCF_LAYER_TRANSPORT:
501 return skb_transport_header(skb);
507 static inline int tcf_valid_offset(const struct sk_buff *skb,
508 const unsigned char *ptr, const int len)
510 return likely((ptr + len) <= skb_tail_pointer(skb) &&
512 (ptr <= (ptr + len)));
516 tcf_change_indev(struct net *net, struct nlattr *indev_tlv,
517 struct netlink_ext_ack *extack)
519 char indev[IFNAMSIZ];
520 struct net_device *dev;
522 if (nla_strlcpy(indev, indev_tlv, IFNAMSIZ) >= IFNAMSIZ) {
523 NL_SET_ERR_MSG(extack, "Interface name too long");
526 dev = __dev_get_by_name(net, indev);
533 tcf_match_indev(struct sk_buff *skb, int ifindex)
539 return ifindex == skb->skb_iif;
542 int tc_setup_flow_action(struct flow_action *flow_action,
543 const struct tcf_exts *exts);
544 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
545 void *type_data, bool err_stop);
546 unsigned int tcf_exts_num_actions(struct tcf_exts *exts);
548 struct tc_cls_u32_knode {
549 struct tcf_exts *exts;
550 struct tcf_result *res;
551 struct tc_u32_sel *sel;
559 struct tc_cls_u32_hnode {
562 unsigned int divisor;
565 enum tc_clsu32_command {
567 TC_CLSU32_REPLACE_KNODE,
568 TC_CLSU32_DELETE_KNODE,
570 TC_CLSU32_REPLACE_HNODE,
571 TC_CLSU32_DELETE_HNODE,
574 struct tc_cls_u32_offload {
575 struct flow_cls_common_offload common;
577 enum tc_clsu32_command command;
579 struct tc_cls_u32_knode knode;
580 struct tc_cls_u32_hnode hnode;
584 static inline bool tc_can_offload(const struct net_device *dev)
586 return dev->features & NETIF_F_HW_TC;
589 static inline bool tc_can_offload_extack(const struct net_device *dev,
590 struct netlink_ext_ack *extack)
592 bool can = tc_can_offload(dev);
595 NL_SET_ERR_MSG(extack, "TC offload is disabled on net device");
601 tc_cls_can_offload_and_chain0(const struct net_device *dev,
602 struct flow_cls_common_offload *common)
604 if (!tc_can_offload_extack(dev, common->extack))
606 if (common->chain_index) {
607 NL_SET_ERR_MSG(common->extack,
608 "Driver supports only offload of chain 0");
614 static inline bool tc_skip_hw(u32 flags)
616 return (flags & TCA_CLS_FLAGS_SKIP_HW) ? true : false;
619 static inline bool tc_skip_sw(u32 flags)
621 return (flags & TCA_CLS_FLAGS_SKIP_SW) ? true : false;
624 /* SKIP_HW and SKIP_SW are mutually exclusive flags. */
625 static inline bool tc_flags_valid(u32 flags)
627 if (flags & ~(TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW |
628 TCA_CLS_FLAGS_VERBOSE))
631 flags &= TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW;
632 if (!(flags ^ (TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW)))
638 static inline bool tc_in_hw(u32 flags)
640 return (flags & TCA_CLS_FLAGS_IN_HW) ? true : false;
644 tc_cls_common_offload_init(struct flow_cls_common_offload *cls_common,
645 const struct tcf_proto *tp, u32 flags,
646 struct netlink_ext_ack *extack)
648 cls_common->chain_index = tp->chain->index;
649 cls_common->protocol = tp->protocol;
650 cls_common->prio = tp->prio;
651 if (tc_skip_sw(flags) || flags & TCA_CLS_FLAGS_VERBOSE)
652 cls_common->extack = extack;
655 enum tc_matchall_command {
656 TC_CLSMATCHALL_REPLACE,
657 TC_CLSMATCHALL_DESTROY,
658 TC_CLSMATCHALL_STATS,
661 struct tc_cls_matchall_offload {
662 struct flow_cls_common_offload common;
663 enum tc_matchall_command command;
664 struct flow_rule *rule;
665 struct flow_stats stats;
666 unsigned long cookie;
669 enum tc_clsbpf_command {
674 struct tc_cls_bpf_offload {
675 struct flow_cls_common_offload common;
676 enum tc_clsbpf_command command;
677 struct tcf_exts *exts;
678 struct bpf_prog *prog;
679 struct bpf_prog *oldprog;
681 bool exts_integrated;
684 struct tc_mqprio_qopt_offload {
685 /* struct tc_mqprio_qopt must always be the first element */
686 struct tc_mqprio_qopt qopt;
690 u64 min_rate[TC_QOPT_MAX_QUEUE];
691 u64 max_rate[TC_QOPT_MAX_QUEUE];
694 /* This structure holds cookie structure that is passed from user
695 * to the kernel for actions and classifiers
703 struct tc_qopt_offload_stats {
704 struct gnet_stats_basic_packed *bstats;
705 struct gnet_stats_queue *qstats;
715 struct tc_mq_opt_offload_graft_params {
720 struct tc_mq_qopt_offload {
721 enum tc_mq_command command;
724 struct tc_qopt_offload_stats stats;
725 struct tc_mq_opt_offload_graft_params graft_params;
729 enum tc_red_command {
737 struct tc_red_qopt_offload_params {
744 struct gnet_stats_queue *qstats;
747 struct tc_red_qopt_offload {
748 enum tc_red_command command;
752 struct tc_red_qopt_offload_params set;
753 struct tc_qopt_offload_stats stats;
754 struct red_stats *xstats;
759 enum tc_gred_command {
765 struct tc_gred_vq_qopt_offload_params {
774 /* Only need backlog, see struct tc_prio_qopt_offload_params */
778 struct tc_gred_qopt_offload_params {
783 struct gnet_stats_queue *qstats;
784 struct tc_gred_vq_qopt_offload_params tab[MAX_DPs];
787 struct tc_gred_qopt_offload_stats {
788 struct gnet_stats_basic_packed bstats[MAX_DPs];
789 struct gnet_stats_queue qstats[MAX_DPs];
790 struct red_stats *xstats[MAX_DPs];
793 struct tc_gred_qopt_offload {
794 enum tc_gred_command command;
798 struct tc_gred_qopt_offload_params set;
799 struct tc_gred_qopt_offload_stats stats;
803 enum tc_prio_command {
810 struct tc_prio_qopt_offload_params {
812 u8 priomap[TC_PRIO_MAX + 1];
813 /* In case that a prio qdisc is offloaded and now is changed to a
814 * non-offloadedable config, it needs to update the backlog & qlen
815 * values to negate the HW backlog & qlen values (and only them).
817 struct gnet_stats_queue *qstats;
820 struct tc_prio_qopt_offload_graft_params {
825 struct tc_prio_qopt_offload {
826 enum tc_prio_command command;
830 struct tc_prio_qopt_offload_params replace_params;
831 struct tc_qopt_offload_stats stats;
832 struct tc_prio_qopt_offload_graft_params graft_params;
836 enum tc_root_command {
840 struct tc_root_qopt_offload {
841 enum tc_root_command command;