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/net_namespace.h>
11 /* TC action not accessible from user space */
12 #define TC_ACT_CONSUMED (TC_ACT_VALUE_MAX + 1)
14 /* Basic packet classifier frontend definitions. */
22 int (*fn)(struct tcf_proto *, void *node, struct tcf_walker *);
25 int register_tcf_proto_ops(struct tcf_proto_ops *ops);
26 int unregister_tcf_proto_ops(struct tcf_proto_ops *ops);
28 struct tcf_block_ext_info {
29 enum flow_block_binder_type binder_type;
30 tcf_chain_head_change_t *chain_head_change;
31 void *chain_head_change_priv;
36 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func);
39 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block,
41 void tcf_chain_put_by_act(struct tcf_chain *chain);
42 struct tcf_chain *tcf_get_next_chain(struct tcf_block *block,
43 struct tcf_chain *chain);
44 struct tcf_proto *tcf_get_next_proto(struct tcf_chain *chain,
45 struct tcf_proto *tp, bool rtnl_held);
46 void tcf_block_netif_keep_dst(struct tcf_block *block);
47 int tcf_block_get(struct tcf_block **p_block,
48 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
49 struct netlink_ext_ack *extack);
50 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
51 struct tcf_block_ext_info *ei,
52 struct netlink_ext_ack *extack);
53 void tcf_block_put(struct tcf_block *block);
54 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
55 struct tcf_block_ext_info *ei);
57 static inline bool tcf_block_shared(struct tcf_block *block)
62 static inline bool tcf_block_non_null_shared(struct tcf_block *block)
64 return block && block->index;
67 static inline struct Qdisc *tcf_block_q(struct tcf_block *block)
69 WARN_ON(tcf_block_shared(block));
73 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
74 struct tcf_result *res, bool compat_mode);
77 static inline bool tcf_block_shared(struct tcf_block *block)
82 static inline bool tcf_block_non_null_shared(struct tcf_block *block)
88 int tcf_block_get(struct tcf_block **p_block,
89 struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
90 struct netlink_ext_ack *extack)
96 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
97 struct tcf_block_ext_info *ei,
98 struct netlink_ext_ack *extack)
103 static inline void tcf_block_put(struct tcf_block *block)
108 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
109 struct tcf_block_ext_info *ei)
113 static inline struct Qdisc *tcf_block_q(struct tcf_block *block)
119 int tc_setup_cb_block_register(struct tcf_block *block, flow_setup_cb_t *cb,
126 void tc_setup_cb_block_unregister(struct tcf_block *block, flow_setup_cb_t *cb,
131 static inline int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
132 struct tcf_result *res, bool compat_mode)
134 return TC_ACT_UNSPEC;
138 static inline unsigned long
139 __cls_set_class(unsigned long *clp, unsigned long cl)
141 return xchg(clp, cl);
144 static inline unsigned long
145 cls_set_class(struct Qdisc *q, unsigned long *clp, unsigned long cl)
147 unsigned long old_cl;
150 old_cl = __cls_set_class(clp, cl);
156 tcf_bind_filter(struct tcf_proto *tp, struct tcf_result *r, unsigned long base)
158 struct Qdisc *q = tp->chain->block->q;
161 /* Check q as it is not set for shared blocks. In that case,
162 * setting class is not supported.
166 cl = q->ops->cl_ops->bind_tcf(q, base, r->classid);
167 cl = cls_set_class(q, &r->class, cl);
169 q->ops->cl_ops->unbind_tcf(q, cl);
173 tcf_unbind_filter(struct tcf_proto *tp, struct tcf_result *r)
175 struct Qdisc *q = tp->chain->block->q;
180 if ((cl = __cls_set_class(&r->class, 0)) != 0)
181 q->ops->cl_ops->unbind_tcf(q, cl);
185 #ifdef CONFIG_NET_CLS_ACT
186 __u32 type; /* for backward compat(TCA_OLD_COMPAT) */
188 struct tc_action **actions;
191 /* Map to export classifier specific extension TLV types to the
192 * generic extensions API. Unsupported extensions must be set to 0.
198 static inline int tcf_exts_init(struct tcf_exts *exts, struct net *net,
199 int action, int police)
201 #ifdef CONFIG_NET_CLS_ACT
203 exts->nr_actions = 0;
205 exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
210 exts->action = action;
211 exts->police = police;
215 /* Return false if the netns is being destroyed in cleanup_net(). Callers
216 * need to do cleanup synchronously in this case, otherwise may race with
217 * tc_action_net_exit(). Return true for other cases.
219 static inline bool tcf_exts_get_net(struct tcf_exts *exts)
221 #ifdef CONFIG_NET_CLS_ACT
222 exts->net = maybe_get_net(exts->net);
223 return exts->net != NULL;
229 static inline void tcf_exts_put_net(struct tcf_exts *exts)
231 #ifdef CONFIG_NET_CLS_ACT
237 #ifdef CONFIG_NET_CLS_ACT
238 #define tcf_exts_for_each_action(i, a, exts) \
239 for (i = 0; i < TCA_ACT_MAX_PRIO && ((a) = (exts)->actions[i]); i++)
241 #define tcf_exts_for_each_action(i, a, exts) \
242 for (; 0; (void)(i), (void)(a), (void)(exts))
246 tcf_exts_stats_update(const struct tcf_exts *exts,
247 u64 bytes, u64 packets, u64 lastuse)
249 #ifdef CONFIG_NET_CLS_ACT
254 for (i = 0; i < exts->nr_actions; i++) {
255 struct tc_action *a = exts->actions[i];
257 tcf_action_stats_update(a, bytes, packets, lastuse, true);
265 * tcf_exts_has_actions - check if at least one action is present
266 * @exts: tc filter extensions handle
268 * Returns true if at least one action is present.
270 static inline bool tcf_exts_has_actions(struct tcf_exts *exts)
272 #ifdef CONFIG_NET_CLS_ACT
273 return exts->nr_actions;
280 * tcf_exts_exec - execute tc filter extensions
281 * @skb: socket buffer
282 * @exts: tc filter extensions handle
283 * @res: desired result
285 * Executes all configured extensions. Returns TC_ACT_OK on a normal execution,
286 * a negative number if the filter must be considered unmatched or
287 * a positive action code (TC_ACT_*) which must be returned to the
291 tcf_exts_exec(struct sk_buff *skb, struct tcf_exts *exts,
292 struct tcf_result *res)
294 #ifdef CONFIG_NET_CLS_ACT
295 return tcf_action_exec(skb, exts->actions, exts->nr_actions, res);
300 int tcf_exts_validate(struct net *net, struct tcf_proto *tp,
301 struct nlattr **tb, struct nlattr *rate_tlv,
302 struct tcf_exts *exts, bool ovr, bool rtnl_held,
303 struct netlink_ext_ack *extack);
304 void tcf_exts_destroy(struct tcf_exts *exts);
305 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src);
306 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts);
307 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts);
310 * struct tcf_pkt_info - packet information
312 struct tcf_pkt_info {
317 #ifdef CONFIG_NET_EMATCH
319 struct tcf_ematch_ops;
322 * struct tcf_ematch - extended match (ematch)
324 * @matchid: identifier to allow userspace to reidentify a match
325 * @flags: flags specifying attributes and the relation to other matches
326 * @ops: the operations lookup table of the corresponding ematch module
327 * @datalen: length of the ematch specific configuration data
328 * @data: ematch specific data
331 struct tcf_ematch_ops * ops;
333 unsigned int datalen;
339 static inline int tcf_em_is_container(struct tcf_ematch *em)
344 static inline int tcf_em_is_simple(struct tcf_ematch *em)
346 return em->flags & TCF_EM_SIMPLE;
349 static inline int tcf_em_is_inverted(struct tcf_ematch *em)
351 return em->flags & TCF_EM_INVERT;
354 static inline int tcf_em_last_match(struct tcf_ematch *em)
356 return (em->flags & TCF_EM_REL_MASK) == TCF_EM_REL_END;
359 static inline int tcf_em_early_end(struct tcf_ematch *em, int result)
361 if (tcf_em_last_match(em))
364 if (result == 0 && em->flags & TCF_EM_REL_AND)
367 if (result != 0 && em->flags & TCF_EM_REL_OR)
374 * struct tcf_ematch_tree - ematch tree handle
376 * @hdr: ematch tree header supplied by userspace
377 * @matches: array of ematches
379 struct tcf_ematch_tree {
380 struct tcf_ematch_tree_hdr hdr;
381 struct tcf_ematch * matches;
386 * struct tcf_ematch_ops - ematch module operations
388 * @kind: identifier (kind) of this ematch module
389 * @datalen: length of expected configuration data (optional)
390 * @change: called during validation (optional)
391 * @match: called during ematch tree evaluation, must return 1/0
392 * @destroy: called during destroyage (optional)
393 * @dump: called during dumping process (optional)
394 * @owner: owner, must be set to THIS_MODULE
395 * @link: link to previous/next ematch module (internal use)
397 struct tcf_ematch_ops {
400 int (*change)(struct net *net, void *,
401 int, struct tcf_ematch *);
402 int (*match)(struct sk_buff *, struct tcf_ematch *,
403 struct tcf_pkt_info *);
404 void (*destroy)(struct tcf_ematch *);
405 int (*dump)(struct sk_buff *, struct tcf_ematch *);
406 struct module *owner;
407 struct list_head link;
410 int tcf_em_register(struct tcf_ematch_ops *);
411 void tcf_em_unregister(struct tcf_ematch_ops *);
412 int tcf_em_tree_validate(struct tcf_proto *, struct nlattr *,
413 struct tcf_ematch_tree *);
414 void tcf_em_tree_destroy(struct tcf_ematch_tree *);
415 int tcf_em_tree_dump(struct sk_buff *, struct tcf_ematch_tree *, int);
416 int __tcf_em_tree_match(struct sk_buff *, struct tcf_ematch_tree *,
417 struct tcf_pkt_info *);
420 * tcf_em_tree_match - evaulate an ematch tree
422 * @skb: socket buffer of the packet in question
423 * @tree: ematch tree to be used for evaluation
424 * @info: packet information examined by classifier
426 * This function matches @skb against the ematch tree in @tree by going
427 * through all ematches respecting their logic relations returning
428 * as soon as the result is obvious.
430 * Returns 1 if the ematch tree as-one matches, no ematches are configured
431 * or ematch is not enabled in the kernel, otherwise 0 is returned.
433 static inline int tcf_em_tree_match(struct sk_buff *skb,
434 struct tcf_ematch_tree *tree,
435 struct tcf_pkt_info *info)
437 if (tree->hdr.nmatches)
438 return __tcf_em_tree_match(skb, tree, info);
443 #define MODULE_ALIAS_TCF_EMATCH(kind) MODULE_ALIAS("ematch-kind-" __stringify(kind))
445 #else /* CONFIG_NET_EMATCH */
447 struct tcf_ematch_tree {
450 #define tcf_em_tree_validate(tp, tb, t) ((void)(t), 0)
451 #define tcf_em_tree_destroy(t) do { (void)(t); } while(0)
452 #define tcf_em_tree_dump(skb, t, tlv) (0)
453 #define tcf_em_tree_match(skb, t, info) ((void)(info), 1)
455 #endif /* CONFIG_NET_EMATCH */
457 static inline unsigned char * tcf_get_base_ptr(struct sk_buff *skb, int layer)
461 return skb_mac_header(skb);
462 case TCF_LAYER_NETWORK:
463 return skb_network_header(skb);
464 case TCF_LAYER_TRANSPORT:
465 return skb_transport_header(skb);
471 static inline int tcf_valid_offset(const struct sk_buff *skb,
472 const unsigned char *ptr, const int len)
474 return likely((ptr + len) <= skb_tail_pointer(skb) &&
476 (ptr <= (ptr + len)));
480 tcf_change_indev(struct net *net, struct nlattr *indev_tlv,
481 struct netlink_ext_ack *extack)
483 char indev[IFNAMSIZ];
484 struct net_device *dev;
486 if (nla_strlcpy(indev, indev_tlv, IFNAMSIZ) >= IFNAMSIZ) {
487 NL_SET_ERR_MSG(extack, "Interface name too long");
490 dev = __dev_get_by_name(net, indev);
497 tcf_match_indev(struct sk_buff *skb, int ifindex)
503 return ifindex == skb->skb_iif;
506 int tc_setup_flow_action(struct flow_action *flow_action,
507 const struct tcf_exts *exts, bool rtnl_held);
508 void tc_cleanup_flow_action(struct flow_action *flow_action);
510 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
511 void *type_data, bool err_stop, bool rtnl_held);
512 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
513 enum tc_setup_type type, void *type_data, bool err_stop,
514 u32 *flags, unsigned int *in_hw_count, bool rtnl_held);
515 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
516 enum tc_setup_type type, void *type_data, bool err_stop,
517 u32 *old_flags, unsigned int *old_in_hw_count,
518 u32 *new_flags, unsigned int *new_in_hw_count,
520 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
521 enum tc_setup_type type, void *type_data, bool err_stop,
522 u32 *flags, unsigned int *in_hw_count, bool rtnl_held);
523 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
524 bool add, flow_setup_cb_t *cb,
525 enum tc_setup_type type, void *type_data,
526 void *cb_priv, u32 *flags, unsigned int *in_hw_count);
527 unsigned int tcf_exts_num_actions(struct tcf_exts *exts);
529 struct tc_cls_u32_knode {
530 struct tcf_exts *exts;
531 struct tcf_result *res;
532 struct tc_u32_sel *sel;
540 struct tc_cls_u32_hnode {
543 unsigned int divisor;
546 enum tc_clsu32_command {
548 TC_CLSU32_REPLACE_KNODE,
549 TC_CLSU32_DELETE_KNODE,
551 TC_CLSU32_REPLACE_HNODE,
552 TC_CLSU32_DELETE_HNODE,
555 struct tc_cls_u32_offload {
556 struct flow_cls_common_offload common;
558 enum tc_clsu32_command command;
560 struct tc_cls_u32_knode knode;
561 struct tc_cls_u32_hnode hnode;
565 static inline bool tc_can_offload(const struct net_device *dev)
567 return dev->features & NETIF_F_HW_TC;
570 static inline bool tc_can_offload_extack(const struct net_device *dev,
571 struct netlink_ext_ack *extack)
573 bool can = tc_can_offload(dev);
576 NL_SET_ERR_MSG(extack, "TC offload is disabled on net device");
582 tc_cls_can_offload_and_chain0(const struct net_device *dev,
583 struct flow_cls_common_offload *common)
585 if (!tc_can_offload_extack(dev, common->extack))
587 if (common->chain_index) {
588 NL_SET_ERR_MSG(common->extack,
589 "Driver supports only offload of chain 0");
595 static inline bool tc_skip_hw(u32 flags)
597 return (flags & TCA_CLS_FLAGS_SKIP_HW) ? true : false;
600 static inline bool tc_skip_sw(u32 flags)
602 return (flags & TCA_CLS_FLAGS_SKIP_SW) ? true : false;
605 /* SKIP_HW and SKIP_SW are mutually exclusive flags. */
606 static inline bool tc_flags_valid(u32 flags)
608 if (flags & ~(TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW |
609 TCA_CLS_FLAGS_VERBOSE))
612 flags &= TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW;
613 if (!(flags ^ (TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW)))
619 static inline bool tc_in_hw(u32 flags)
621 return (flags & TCA_CLS_FLAGS_IN_HW) ? true : false;
625 tc_cls_common_offload_init(struct flow_cls_common_offload *cls_common,
626 const struct tcf_proto *tp, u32 flags,
627 struct netlink_ext_ack *extack)
629 cls_common->chain_index = tp->chain->index;
630 cls_common->protocol = tp->protocol;
631 cls_common->prio = tp->prio >> 16;
632 if (tc_skip_sw(flags) || flags & TCA_CLS_FLAGS_VERBOSE)
633 cls_common->extack = extack;
636 enum tc_matchall_command {
637 TC_CLSMATCHALL_REPLACE,
638 TC_CLSMATCHALL_DESTROY,
639 TC_CLSMATCHALL_STATS,
642 struct tc_cls_matchall_offload {
643 struct flow_cls_common_offload common;
644 enum tc_matchall_command command;
645 struct flow_rule *rule;
646 struct flow_stats stats;
647 unsigned long cookie;
650 enum tc_clsbpf_command {
655 struct tc_cls_bpf_offload {
656 struct flow_cls_common_offload common;
657 enum tc_clsbpf_command command;
658 struct tcf_exts *exts;
659 struct bpf_prog *prog;
660 struct bpf_prog *oldprog;
662 bool exts_integrated;
665 struct tc_mqprio_qopt_offload {
666 /* struct tc_mqprio_qopt must always be the first element */
667 struct tc_mqprio_qopt qopt;
671 u64 min_rate[TC_QOPT_MAX_QUEUE];
672 u64 max_rate[TC_QOPT_MAX_QUEUE];
675 /* This structure holds cookie structure that is passed from user
676 * to the kernel for actions and classifiers
684 struct tc_qopt_offload_stats {
685 struct gnet_stats_basic_packed *bstats;
686 struct gnet_stats_queue *qstats;
696 struct tc_mq_opt_offload_graft_params {
701 struct tc_mq_qopt_offload {
702 enum tc_mq_command command;
705 struct tc_qopt_offload_stats stats;
706 struct tc_mq_opt_offload_graft_params graft_params;
710 enum tc_red_command {
718 struct tc_red_qopt_offload_params {
725 struct gnet_stats_queue *qstats;
728 struct tc_red_qopt_offload {
729 enum tc_red_command command;
733 struct tc_red_qopt_offload_params set;
734 struct tc_qopt_offload_stats stats;
735 struct red_stats *xstats;
740 enum tc_gred_command {
746 struct tc_gred_vq_qopt_offload_params {
755 /* Only need backlog, see struct tc_prio_qopt_offload_params */
759 struct tc_gred_qopt_offload_params {
764 struct gnet_stats_queue *qstats;
765 struct tc_gred_vq_qopt_offload_params tab[MAX_DPs];
768 struct tc_gred_qopt_offload_stats {
769 struct gnet_stats_basic_packed bstats[MAX_DPs];
770 struct gnet_stats_queue qstats[MAX_DPs];
771 struct red_stats *xstats[MAX_DPs];
774 struct tc_gred_qopt_offload {
775 enum tc_gred_command command;
779 struct tc_gred_qopt_offload_params set;
780 struct tc_gred_qopt_offload_stats stats;
784 enum tc_prio_command {
791 struct tc_prio_qopt_offload_params {
793 u8 priomap[TC_PRIO_MAX + 1];
794 /* In case that a prio qdisc is offloaded and now is changed to a
795 * non-offloadedable config, it needs to update the backlog & qlen
796 * values to negate the HW backlog & qlen values (and only them).
798 struct gnet_stats_queue *qstats;
801 struct tc_prio_qopt_offload_graft_params {
806 struct tc_prio_qopt_offload {
807 enum tc_prio_command command;
811 struct tc_prio_qopt_offload_params replace_params;
812 struct tc_qopt_offload_stats stats;
813 struct tc_prio_qopt_offload_graft_params graft_params;
817 enum tc_root_command {
821 struct tc_root_qopt_offload {
822 enum tc_root_command command;