xref: /linux/drivers/net/ethernet/sfc/tc.c (revision 55f1b540d893da740a81200450014c45a8103f54)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2019 Solarflare Communications Inc.
5  * Copyright 2020-2022 Xilinx Inc.
6  *
7  * This program is free software; you can redistribute it and/or modify it
8  * under the terms of the GNU General Public License version 2 as published
9  * by the Free Software Foundation, incorporated herein by reference.
10  */
11 
12 #include <net/pkt_cls.h>
13 #include <net/vxlan.h>
14 #include <net/geneve.h>
15 #include <net/tc_act/tc_ct.h>
16 #include "tc.h"
17 #include "tc_bindings.h"
18 #include "tc_encap_actions.h"
19 #include "tc_conntrack.h"
20 #include "mae.h"
21 #include "ef100_rep.h"
22 #include "efx.h"
23 
24 enum efx_encap_type efx_tc_indr_netdev_type(struct net_device *net_dev)
25 {
26 	if (netif_is_vxlan(net_dev))
27 		return EFX_ENCAP_TYPE_VXLAN;
28 	if (netif_is_geneve(net_dev))
29 		return EFX_ENCAP_TYPE_GENEVE;
30 
31 	return EFX_ENCAP_TYPE_NONE;
32 }
33 
34 #define EFX_TC_HDR_TYPE_TTL_MASK ((u32)0xff)
35 /* Hoplimit is stored in the most significant byte in the pedit ipv6 header action */
36 #define EFX_TC_HDR_TYPE_HLIMIT_MASK ~((u32)0xff000000)
37 #define EFX_EFV_PF	NULL
38 /* Look up the representor information (efv) for a device.
39  * May return NULL for the PF (us), or an error pointer for a device that
40  * isn't supported as a TC offload endpoint
41  */
42 struct efx_rep *efx_tc_flower_lookup_efv(struct efx_nic *efx,
43 					 struct net_device *dev)
44 {
45 	struct efx_rep *efv;
46 
47 	if (!dev)
48 		return ERR_PTR(-EOPNOTSUPP);
49 	/* Is it us (the PF)? */
50 	if (dev == efx->net_dev)
51 		return EFX_EFV_PF;
52 	/* Is it an efx vfrep at all? */
53 	if (dev->netdev_ops != &efx_ef100_rep_netdev_ops)
54 		return ERR_PTR(-EOPNOTSUPP);
55 	/* Is it ours?  We don't support TC rules that include another
56 	 * EF100's netdevices (not even on another port of the same NIC).
57 	 */
58 	efv = netdev_priv(dev);
59 	if (efv->parent != efx)
60 		return ERR_PTR(-EOPNOTSUPP);
61 	return efv;
62 }
63 
64 /* Convert a driver-internal vport ID into an internal device (PF or VF) */
65 static s64 efx_tc_flower_internal_mport(struct efx_nic *efx, struct efx_rep *efv)
66 {
67 	u32 mport;
68 
69 	if (IS_ERR(efv))
70 		return PTR_ERR(efv);
71 	if (!efv) /* device is PF (us) */
72 		efx_mae_mport_uplink(efx, &mport);
73 	else /* device is repr */
74 		efx_mae_mport_mport(efx, efv->mport, &mport);
75 	return mport;
76 }
77 
78 /* Convert a driver-internal vport ID into an external device (wire or VF) */
79 s64 efx_tc_flower_external_mport(struct efx_nic *efx, struct efx_rep *efv)
80 {
81 	u32 mport;
82 
83 	if (IS_ERR(efv))
84 		return PTR_ERR(efv);
85 	if (!efv) /* device is PF (us) */
86 		efx_mae_mport_wire(efx, &mport);
87 	else /* device is repr */
88 		efx_mae_mport_mport(efx, efv->mport, &mport);
89 	return mport;
90 }
91 
92 static const struct rhashtable_params efx_tc_mac_ht_params = {
93 	.key_len	= offsetofend(struct efx_tc_mac_pedit_action, h_addr),
94 	.key_offset	= 0,
95 	.head_offset	= offsetof(struct efx_tc_mac_pedit_action, linkage),
96 };
97 
98 static const struct rhashtable_params efx_tc_encap_match_ht_params = {
99 	.key_len	= offsetof(struct efx_tc_encap_match, linkage),
100 	.key_offset	= 0,
101 	.head_offset	= offsetof(struct efx_tc_encap_match, linkage),
102 };
103 
104 static const struct rhashtable_params efx_tc_match_action_ht_params = {
105 	.key_len	= sizeof(unsigned long),
106 	.key_offset	= offsetof(struct efx_tc_flow_rule, cookie),
107 	.head_offset	= offsetof(struct efx_tc_flow_rule, linkage),
108 };
109 
110 static const struct rhashtable_params efx_tc_lhs_rule_ht_params = {
111 	.key_len	= sizeof(unsigned long),
112 	.key_offset	= offsetof(struct efx_tc_lhs_rule, cookie),
113 	.head_offset	= offsetof(struct efx_tc_lhs_rule, linkage),
114 };
115 
116 static const struct rhashtable_params efx_tc_recirc_ht_params = {
117 	.key_len	= offsetof(struct efx_tc_recirc_id, linkage),
118 	.key_offset	= 0,
119 	.head_offset	= offsetof(struct efx_tc_recirc_id, linkage),
120 };
121 
122 static struct efx_tc_mac_pedit_action *efx_tc_flower_get_mac(struct efx_nic *efx,
123 							     unsigned char h_addr[ETH_ALEN],
124 							     struct netlink_ext_ack *extack)
125 {
126 	struct efx_tc_mac_pedit_action *ped, *old;
127 	int rc;
128 
129 	ped = kzalloc(sizeof(*ped), GFP_USER);
130 	if (!ped)
131 		return ERR_PTR(-ENOMEM);
132 	memcpy(ped->h_addr, h_addr, ETH_ALEN);
133 	old = rhashtable_lookup_get_insert_fast(&efx->tc->mac_ht,
134 						&ped->linkage,
135 						efx_tc_mac_ht_params);
136 	if (old) {
137 		/* don't need our new entry */
138 		kfree(ped);
139 		if (IS_ERR(old)) /* oh dear, it's actually an error */
140 			return ERR_CAST(old);
141 		if (!refcount_inc_not_zero(&old->ref))
142 			return ERR_PTR(-EAGAIN);
143 		/* existing entry found, ref taken */
144 		return old;
145 	}
146 
147 	rc = efx_mae_allocate_pedit_mac(efx, ped);
148 	if (rc < 0) {
149 		NL_SET_ERR_MSG_MOD(extack, "Failed to store pedit MAC address in hw");
150 		goto out_remove;
151 	}
152 
153 	/* ref and return */
154 	refcount_set(&ped->ref, 1);
155 	return ped;
156 out_remove:
157 	rhashtable_remove_fast(&efx->tc->mac_ht, &ped->linkage,
158 			       efx_tc_mac_ht_params);
159 	kfree(ped);
160 	return ERR_PTR(rc);
161 }
162 
163 static void efx_tc_flower_put_mac(struct efx_nic *efx,
164 				  struct efx_tc_mac_pedit_action *ped)
165 {
166 	if (!refcount_dec_and_test(&ped->ref))
167 		return; /* still in use */
168 	rhashtable_remove_fast(&efx->tc->mac_ht, &ped->linkage,
169 			       efx_tc_mac_ht_params);
170 	efx_mae_free_pedit_mac(efx, ped);
171 	kfree(ped);
172 }
173 
174 static void efx_tc_free_action_set(struct efx_nic *efx,
175 				   struct efx_tc_action_set *act, bool in_hw)
176 {
177 	/* Failure paths calling this on the 'cursor' action set in_hw=false,
178 	 * because if the alloc had succeeded we'd've put it in acts.list and
179 	 * not still have it in act.
180 	 */
181 	if (in_hw) {
182 		efx_mae_free_action_set(efx, act->fw_id);
183 		/* in_hw is true iff we are on an acts.list; make sure to
184 		 * remove ourselves from that list before we are freed.
185 		 */
186 		list_del(&act->list);
187 	}
188 	if (act->count) {
189 		spin_lock_bh(&act->count->cnt->lock);
190 		if (!list_empty(&act->count_user))
191 			list_del(&act->count_user);
192 		spin_unlock_bh(&act->count->cnt->lock);
193 		efx_tc_flower_put_counter_index(efx, act->count);
194 	}
195 	if (act->encap_md) {
196 		list_del(&act->encap_user);
197 		efx_tc_flower_release_encap_md(efx, act->encap_md);
198 	}
199 	if (act->src_mac)
200 		efx_tc_flower_put_mac(efx, act->src_mac);
201 	if (act->dst_mac)
202 		efx_tc_flower_put_mac(efx, act->dst_mac);
203 	kfree(act);
204 }
205 
206 static void efx_tc_free_action_set_list(struct efx_nic *efx,
207 					struct efx_tc_action_set_list *acts,
208 					bool in_hw)
209 {
210 	struct efx_tc_action_set *act, *next;
211 
212 	/* Failure paths set in_hw=false, because usually the acts didn't get
213 	 * to efx_mae_alloc_action_set_list(); if they did, the failure tree
214 	 * has a separate efx_mae_free_action_set_list() before calling us.
215 	 */
216 	if (in_hw)
217 		efx_mae_free_action_set_list(efx, acts);
218 	/* Any act that's on the list will be in_hw even if the list isn't */
219 	list_for_each_entry_safe(act, next, &acts->list, list)
220 		efx_tc_free_action_set(efx, act, true);
221 	/* Don't kfree, as acts is embedded inside a struct efx_tc_flow_rule */
222 }
223 
224 /* Boilerplate for the simple 'copy a field' cases */
225 #define _MAP_KEY_AND_MASK(_name, _type, _tcget, _tcfield, _field)	\
226 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_##_name)) {		\
227 	struct flow_match_##_type fm;					\
228 									\
229 	flow_rule_match_##_tcget(rule, &fm);				\
230 	match->value._field = fm.key->_tcfield;				\
231 	match->mask._field = fm.mask->_tcfield;				\
232 }
233 #define MAP_KEY_AND_MASK(_name, _type, _tcfield, _field)	\
234 	_MAP_KEY_AND_MASK(_name, _type, _type, _tcfield, _field)
235 #define MAP_ENC_KEY_AND_MASK(_name, _type, _tcget, _tcfield, _field)	\
236 	_MAP_KEY_AND_MASK(ENC_##_name, _type, _tcget, _tcfield, _field)
237 
238 static int efx_tc_flower_parse_match(struct efx_nic *efx,
239 				     struct flow_rule *rule,
240 				     struct efx_tc_match *match,
241 				     struct netlink_ext_ack *extack)
242 {
243 	struct flow_dissector *dissector = rule->match.dissector;
244 	unsigned char ipv = 0;
245 
246 	/* Owing to internal TC infelicities, the IPV6_ADDRS key might be set
247 	 * even on IPv4 filters; so rather than relying on dissector->used_keys
248 	 * we check the addr_type in the CONTROL key.  If we don't find it (or
249 	 * it's masked, which should never happen), we treat both IPV4_ADDRS
250 	 * and IPV6_ADDRS as absent.
251 	 */
252 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
253 		struct flow_match_control fm;
254 
255 		flow_rule_match_control(rule, &fm);
256 		if (IS_ALL_ONES(fm.mask->addr_type))
257 			switch (fm.key->addr_type) {
258 			case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
259 				ipv = 4;
260 				break;
261 			case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
262 				ipv = 6;
263 				break;
264 			default:
265 				break;
266 			}
267 
268 		if (fm.mask->flags & FLOW_DIS_IS_FRAGMENT) {
269 			match->value.ip_frag = fm.key->flags & FLOW_DIS_IS_FRAGMENT;
270 			match->mask.ip_frag = true;
271 		}
272 		if (fm.mask->flags & FLOW_DIS_FIRST_FRAG) {
273 			match->value.ip_firstfrag = fm.key->flags & FLOW_DIS_FIRST_FRAG;
274 			match->mask.ip_firstfrag = true;
275 		}
276 		if (!flow_rule_is_supp_control_flags(FLOW_DIS_IS_FRAGMENT |
277 						     FLOW_DIS_FIRST_FRAG,
278 						     fm.mask->flags, extack))
279 			return -EOPNOTSUPP;
280 	}
281 	if (dissector->used_keys &
282 	    ~(BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) |
283 	      BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) |
284 	      BIT_ULL(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
285 	      BIT_ULL(FLOW_DISSECTOR_KEY_VLAN) |
286 	      BIT_ULL(FLOW_DISSECTOR_KEY_CVLAN) |
287 	      BIT_ULL(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
288 	      BIT_ULL(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
289 	      BIT_ULL(FLOW_DISSECTOR_KEY_PORTS) |
290 	      BIT_ULL(FLOW_DISSECTOR_KEY_ENC_KEYID) |
291 	      BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
292 	      BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
293 	      BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IP) |
294 	      BIT_ULL(FLOW_DISSECTOR_KEY_ENC_PORTS) |
295 	      BIT_ULL(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
296 	      BIT_ULL(FLOW_DISSECTOR_KEY_CT) |
297 	      BIT_ULL(FLOW_DISSECTOR_KEY_TCP) |
298 	      BIT_ULL(FLOW_DISSECTOR_KEY_IP))) {
299 		NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported flower keys %#llx",
300 				       dissector->used_keys);
301 		return -EOPNOTSUPP;
302 	}
303 
304 	MAP_KEY_AND_MASK(BASIC, basic, n_proto, eth_proto);
305 	/* Make sure we're IP if any L3/L4 keys used. */
306 	if (!IS_ALL_ONES(match->mask.eth_proto) ||
307 	    !(match->value.eth_proto == htons(ETH_P_IP) ||
308 	      match->value.eth_proto == htons(ETH_P_IPV6)))
309 		if (dissector->used_keys &
310 		    (BIT_ULL(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
311 		     BIT_ULL(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
312 		     BIT_ULL(FLOW_DISSECTOR_KEY_PORTS) |
313 		     BIT_ULL(FLOW_DISSECTOR_KEY_IP) |
314 		     BIT_ULL(FLOW_DISSECTOR_KEY_TCP))) {
315 			NL_SET_ERR_MSG_FMT_MOD(extack,
316 					       "L3/L4 flower keys %#llx require protocol ipv[46]",
317 					       dissector->used_keys);
318 			return -EINVAL;
319 		}
320 
321 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
322 		struct flow_match_vlan fm;
323 
324 		flow_rule_match_vlan(rule, &fm);
325 		if (fm.mask->vlan_id || fm.mask->vlan_priority || fm.mask->vlan_tpid) {
326 			match->value.vlan_proto[0] = fm.key->vlan_tpid;
327 			match->mask.vlan_proto[0] = fm.mask->vlan_tpid;
328 			match->value.vlan_tci[0] = cpu_to_be16(fm.key->vlan_priority << 13 |
329 							       fm.key->vlan_id);
330 			match->mask.vlan_tci[0] = cpu_to_be16(fm.mask->vlan_priority << 13 |
331 							      fm.mask->vlan_id);
332 		}
333 	}
334 
335 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
336 		struct flow_match_vlan fm;
337 
338 		flow_rule_match_cvlan(rule, &fm);
339 		if (fm.mask->vlan_id || fm.mask->vlan_priority || fm.mask->vlan_tpid) {
340 			match->value.vlan_proto[1] = fm.key->vlan_tpid;
341 			match->mask.vlan_proto[1] = fm.mask->vlan_tpid;
342 			match->value.vlan_tci[1] = cpu_to_be16(fm.key->vlan_priority << 13 |
343 							       fm.key->vlan_id);
344 			match->mask.vlan_tci[1] = cpu_to_be16(fm.mask->vlan_priority << 13 |
345 							      fm.mask->vlan_id);
346 		}
347 	}
348 
349 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
350 		struct flow_match_eth_addrs fm;
351 
352 		flow_rule_match_eth_addrs(rule, &fm);
353 		ether_addr_copy(match->value.eth_saddr, fm.key->src);
354 		ether_addr_copy(match->value.eth_daddr, fm.key->dst);
355 		ether_addr_copy(match->mask.eth_saddr, fm.mask->src);
356 		ether_addr_copy(match->mask.eth_daddr, fm.mask->dst);
357 	}
358 
359 	MAP_KEY_AND_MASK(BASIC, basic, ip_proto, ip_proto);
360 	/* Make sure we're TCP/UDP if any L4 keys used. */
361 	if ((match->value.ip_proto != IPPROTO_UDP &&
362 	     match->value.ip_proto != IPPROTO_TCP) || !IS_ALL_ONES(match->mask.ip_proto))
363 		if (dissector->used_keys &
364 		    (BIT_ULL(FLOW_DISSECTOR_KEY_PORTS) |
365 		     BIT_ULL(FLOW_DISSECTOR_KEY_TCP))) {
366 			NL_SET_ERR_MSG_FMT_MOD(extack,
367 					       "L4 flower keys %#llx require ipproto udp or tcp",
368 					       dissector->used_keys);
369 			return -EINVAL;
370 		}
371 	MAP_KEY_AND_MASK(IP, ip, tos, ip_tos);
372 	MAP_KEY_AND_MASK(IP, ip, ttl, ip_ttl);
373 	if (ipv == 4) {
374 		MAP_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, src, src_ip);
375 		MAP_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, dst, dst_ip);
376 	}
377 #ifdef CONFIG_IPV6
378 	else if (ipv == 6) {
379 		MAP_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, src, src_ip6);
380 		MAP_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, dst, dst_ip6);
381 	}
382 #endif
383 	MAP_KEY_AND_MASK(PORTS, ports, src, l4_sport);
384 	MAP_KEY_AND_MASK(PORTS, ports, dst, l4_dport);
385 	MAP_KEY_AND_MASK(TCP, tcp, flags, tcp_flags);
386 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
387 		struct flow_match_control fm;
388 
389 		flow_rule_match_enc_control(rule, &fm);
390 		if (flow_rule_has_enc_control_flags(fm.mask->flags, extack))
391 			return -EOPNOTSUPP;
392 		if (!IS_ALL_ONES(fm.mask->addr_type)) {
393 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported enc addr_type mask %u (key %u)",
394 					       fm.mask->addr_type,
395 					       fm.key->addr_type);
396 			return -EOPNOTSUPP;
397 		}
398 		switch (fm.key->addr_type) {
399 		case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
400 			MAP_ENC_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, enc_ipv4_addrs,
401 					     src, enc_src_ip);
402 			MAP_ENC_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, enc_ipv4_addrs,
403 					     dst, enc_dst_ip);
404 			break;
405 #ifdef CONFIG_IPV6
406 		case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
407 			MAP_ENC_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, enc_ipv6_addrs,
408 					     src, enc_src_ip6);
409 			MAP_ENC_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, enc_ipv6_addrs,
410 					     dst, enc_dst_ip6);
411 			break;
412 #endif
413 		default:
414 			NL_SET_ERR_MSG_FMT_MOD(extack,
415 					       "Unsupported enc addr_type %u (supported are IPv4, IPv6)",
416 					       fm.key->addr_type);
417 			return -EOPNOTSUPP;
418 		}
419 		MAP_ENC_KEY_AND_MASK(IP, ip, enc_ip, tos, enc_ip_tos);
420 		MAP_ENC_KEY_AND_MASK(IP, ip, enc_ip, ttl, enc_ip_ttl);
421 		MAP_ENC_KEY_AND_MASK(PORTS, ports, enc_ports, src, enc_sport);
422 		MAP_ENC_KEY_AND_MASK(PORTS, ports, enc_ports, dst, enc_dport);
423 		MAP_ENC_KEY_AND_MASK(KEYID, enc_keyid, enc_keyid, keyid, enc_keyid);
424 	} else if (dissector->used_keys &
425 		   (BIT_ULL(FLOW_DISSECTOR_KEY_ENC_KEYID) |
426 		    BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
427 		    BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
428 		    BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IP) |
429 		    BIT_ULL(FLOW_DISSECTOR_KEY_ENC_PORTS))) {
430 		NL_SET_ERR_MSG_FMT_MOD(extack,
431 				       "Flower enc keys require enc_control (keys: %#llx)",
432 				       dissector->used_keys);
433 		return -EOPNOTSUPP;
434 	}
435 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CT)) {
436 		struct flow_match_ct fm;
437 
438 		flow_rule_match_ct(rule, &fm);
439 		match->value.ct_state_trk = !!(fm.key->ct_state & TCA_FLOWER_KEY_CT_FLAGS_TRACKED);
440 		match->mask.ct_state_trk = !!(fm.mask->ct_state & TCA_FLOWER_KEY_CT_FLAGS_TRACKED);
441 		match->value.ct_state_est = !!(fm.key->ct_state & TCA_FLOWER_KEY_CT_FLAGS_ESTABLISHED);
442 		match->mask.ct_state_est = !!(fm.mask->ct_state & TCA_FLOWER_KEY_CT_FLAGS_ESTABLISHED);
443 		if (fm.mask->ct_state & ~(TCA_FLOWER_KEY_CT_FLAGS_TRACKED |
444 					  TCA_FLOWER_KEY_CT_FLAGS_ESTABLISHED)) {
445 			NL_SET_ERR_MSG_FMT_MOD(extack,
446 					       "Unsupported ct_state match %#x",
447 					       fm.mask->ct_state);
448 			return -EOPNOTSUPP;
449 		}
450 		match->value.ct_mark = fm.key->ct_mark;
451 		match->mask.ct_mark = fm.mask->ct_mark;
452 		match->value.ct_zone = fm.key->ct_zone;
453 		match->mask.ct_zone = fm.mask->ct_zone;
454 
455 		if (memchr_inv(fm.mask->ct_labels, 0, sizeof(fm.mask->ct_labels))) {
456 			NL_SET_ERR_MSG_MOD(extack, "Matching on ct_label not supported");
457 			return -EOPNOTSUPP;
458 		}
459 	}
460 
461 	return 0;
462 }
463 
464 static void efx_tc_flower_release_encap_match(struct efx_nic *efx,
465 					      struct efx_tc_encap_match *encap)
466 {
467 	int rc;
468 
469 	if (!refcount_dec_and_test(&encap->ref))
470 		return; /* still in use */
471 
472 	if (encap->type == EFX_TC_EM_DIRECT) {
473 		rc = efx_mae_unregister_encap_match(efx, encap);
474 		if (rc)
475 			/* Display message but carry on and remove entry from our
476 			 * SW tables, because there's not much we can do about it.
477 			 */
478 			netif_err(efx, drv, efx->net_dev,
479 				  "Failed to release encap match %#x, rc %d\n",
480 				  encap->fw_id, rc);
481 	}
482 	rhashtable_remove_fast(&efx->tc->encap_match_ht, &encap->linkage,
483 			       efx_tc_encap_match_ht_params);
484 	if (encap->pseudo)
485 		efx_tc_flower_release_encap_match(efx, encap->pseudo);
486 	kfree(encap);
487 }
488 
489 static int efx_tc_flower_record_encap_match(struct efx_nic *efx,
490 					    struct efx_tc_match *match,
491 					    enum efx_encap_type type,
492 					    enum efx_tc_em_pseudo_type em_type,
493 					    u8 child_ip_tos_mask,
494 					    __be16 child_udp_sport_mask,
495 					    struct netlink_ext_ack *extack)
496 {
497 	struct efx_tc_encap_match *encap, *old, *pseudo = NULL;
498 	bool ipv6 = false;
499 	int rc;
500 
501 	/* We require that the socket-defining fields (IP addrs and UDP dest
502 	 * port) are present and exact-match.  Other fields may only be used
503 	 * if the field-set (and any masks) are the same for all encap
504 	 * matches on the same <sip,dip,dport> tuple; this is enforced by
505 	 * pseudo encap matches.
506 	 */
507 	if (match->mask.enc_dst_ip | match->mask.enc_src_ip) {
508 		if (!IS_ALL_ONES(match->mask.enc_dst_ip)) {
509 			NL_SET_ERR_MSG_MOD(extack,
510 					   "Egress encap match is not exact on dst IP address");
511 			return -EOPNOTSUPP;
512 		}
513 		if (!IS_ALL_ONES(match->mask.enc_src_ip)) {
514 			NL_SET_ERR_MSG_MOD(extack,
515 					   "Egress encap match is not exact on src IP address");
516 			return -EOPNOTSUPP;
517 		}
518 #ifdef CONFIG_IPV6
519 		if (!ipv6_addr_any(&match->mask.enc_dst_ip6) ||
520 		    !ipv6_addr_any(&match->mask.enc_src_ip6)) {
521 			NL_SET_ERR_MSG_MOD(extack,
522 					   "Egress encap match on both IPv4 and IPv6, don't understand");
523 			return -EOPNOTSUPP;
524 		}
525 	} else {
526 		ipv6 = true;
527 		if (!efx_ipv6_addr_all_ones(&match->mask.enc_dst_ip6)) {
528 			NL_SET_ERR_MSG_MOD(extack,
529 					   "Egress encap match is not exact on dst IP address");
530 			return -EOPNOTSUPP;
531 		}
532 		if (!efx_ipv6_addr_all_ones(&match->mask.enc_src_ip6)) {
533 			NL_SET_ERR_MSG_MOD(extack,
534 					   "Egress encap match is not exact on src IP address");
535 			return -EOPNOTSUPP;
536 		}
537 #endif
538 	}
539 	if (!IS_ALL_ONES(match->mask.enc_dport)) {
540 		NL_SET_ERR_MSG_MOD(extack, "Egress encap match is not exact on dst UDP port");
541 		return -EOPNOTSUPP;
542 	}
543 	if (match->mask.enc_sport || match->mask.enc_ip_tos) {
544 		struct efx_tc_match pmatch = *match;
545 
546 		if (em_type == EFX_TC_EM_PSEUDO_MASK) { /* can't happen */
547 			NL_SET_ERR_MSG_MOD(extack, "Bad recursion in egress encap match handler");
548 			return -EOPNOTSUPP;
549 		}
550 		pmatch.value.enc_ip_tos = 0;
551 		pmatch.mask.enc_ip_tos = 0;
552 		pmatch.value.enc_sport = 0;
553 		pmatch.mask.enc_sport = 0;
554 		rc = efx_tc_flower_record_encap_match(efx, &pmatch, type,
555 						      EFX_TC_EM_PSEUDO_MASK,
556 						      match->mask.enc_ip_tos,
557 						      match->mask.enc_sport,
558 						      extack);
559 		if (rc)
560 			return rc;
561 		pseudo = pmatch.encap;
562 	}
563 	if (match->mask.enc_ip_ttl) {
564 		NL_SET_ERR_MSG_MOD(extack, "Egress encap match on IP TTL not supported");
565 		rc = -EOPNOTSUPP;
566 		goto fail_pseudo;
567 	}
568 
569 	rc = efx_mae_check_encap_match_caps(efx, ipv6, match->mask.enc_ip_tos,
570 					    match->mask.enc_sport, extack);
571 	if (rc)
572 		goto fail_pseudo;
573 
574 	encap = kzalloc(sizeof(*encap), GFP_USER);
575 	if (!encap) {
576 		rc = -ENOMEM;
577 		goto fail_pseudo;
578 	}
579 	encap->src_ip = match->value.enc_src_ip;
580 	encap->dst_ip = match->value.enc_dst_ip;
581 #ifdef CONFIG_IPV6
582 	encap->src_ip6 = match->value.enc_src_ip6;
583 	encap->dst_ip6 = match->value.enc_dst_ip6;
584 #endif
585 	encap->udp_dport = match->value.enc_dport;
586 	encap->tun_type = type;
587 	encap->ip_tos = match->value.enc_ip_tos;
588 	encap->ip_tos_mask = match->mask.enc_ip_tos;
589 	encap->child_ip_tos_mask = child_ip_tos_mask;
590 	encap->udp_sport = match->value.enc_sport;
591 	encap->udp_sport_mask = match->mask.enc_sport;
592 	encap->child_udp_sport_mask = child_udp_sport_mask;
593 	encap->type = em_type;
594 	encap->pseudo = pseudo;
595 	old = rhashtable_lookup_get_insert_fast(&efx->tc->encap_match_ht,
596 						&encap->linkage,
597 						efx_tc_encap_match_ht_params);
598 	if (old) {
599 		/* don't need our new entry */
600 		kfree(encap);
601 		if (pseudo) /* don't need our new pseudo either */
602 			efx_tc_flower_release_encap_match(efx, pseudo);
603 		if (IS_ERR(old)) /* oh dear, it's actually an error */
604 			return PTR_ERR(old);
605 		/* check old and new em_types are compatible */
606 		switch (old->type) {
607 		case EFX_TC_EM_DIRECT:
608 			/* old EM is in hardware, so mustn't overlap with a
609 			 * pseudo, but may be shared with another direct EM
610 			 */
611 			if (em_type == EFX_TC_EM_DIRECT)
612 				break;
613 			NL_SET_ERR_MSG_MOD(extack, "Pseudo encap match conflicts with existing direct entry");
614 			return -EEXIST;
615 		case EFX_TC_EM_PSEUDO_MASK:
616 			/* old EM is protecting a ToS- or src port-qualified
617 			 * filter, so may only be shared with another pseudo
618 			 * for the same ToS and src port masks.
619 			 */
620 			if (em_type != EFX_TC_EM_PSEUDO_MASK) {
621 				NL_SET_ERR_MSG_FMT_MOD(extack,
622 						       "%s encap match conflicts with existing pseudo(MASK) entry",
623 						       em_type ? "Pseudo" : "Direct");
624 				return -EEXIST;
625 			}
626 			if (child_ip_tos_mask != old->child_ip_tos_mask) {
627 				NL_SET_ERR_MSG_FMT_MOD(extack,
628 						       "Pseudo encap match for TOS mask %#04x conflicts with existing mask %#04x",
629 						       child_ip_tos_mask,
630 						       old->child_ip_tos_mask);
631 				return -EEXIST;
632 			}
633 			if (child_udp_sport_mask != old->child_udp_sport_mask) {
634 				NL_SET_ERR_MSG_FMT_MOD(extack,
635 						       "Pseudo encap match for UDP src port mask %#x conflicts with existing mask %#x",
636 						       child_udp_sport_mask,
637 						       old->child_udp_sport_mask);
638 				return -EEXIST;
639 			}
640 			break;
641 		case EFX_TC_EM_PSEUDO_OR:
642 			/* old EM corresponds to an OR that has to be unique
643 			 * (it must not overlap with any other OR, whether
644 			 * direct-EM or pseudo).
645 			 */
646 			NL_SET_ERR_MSG_FMT_MOD(extack,
647 					       "%s encap match conflicts with existing pseudo(OR) entry",
648 					       em_type ? "Pseudo" : "Direct");
649 			return -EEXIST;
650 		default: /* Unrecognised pseudo-type.  Just say no */
651 			NL_SET_ERR_MSG_FMT_MOD(extack,
652 					       "%s encap match conflicts with existing pseudo(%d) entry",
653 					       em_type ? "Pseudo" : "Direct",
654 					       old->type);
655 			return -EEXIST;
656 		}
657 		/* check old and new tun_types are compatible */
658 		if (old->tun_type != type) {
659 			NL_SET_ERR_MSG_FMT_MOD(extack,
660 					       "Egress encap match with conflicting tun_type %u != %u",
661 					       old->tun_type, type);
662 			return -EEXIST;
663 		}
664 		if (!refcount_inc_not_zero(&old->ref))
665 			return -EAGAIN;
666 		/* existing entry found */
667 		encap = old;
668 	} else {
669 		if (em_type == EFX_TC_EM_DIRECT) {
670 			rc = efx_mae_register_encap_match(efx, encap);
671 			if (rc) {
672 				NL_SET_ERR_MSG_MOD(extack, "Failed to record egress encap match in HW");
673 				goto fail;
674 			}
675 		}
676 		refcount_set(&encap->ref, 1);
677 	}
678 	match->encap = encap;
679 	return 0;
680 fail:
681 	rhashtable_remove_fast(&efx->tc->encap_match_ht, &encap->linkage,
682 			       efx_tc_encap_match_ht_params);
683 	kfree(encap);
684 fail_pseudo:
685 	if (pseudo)
686 		efx_tc_flower_release_encap_match(efx, pseudo);
687 	return rc;
688 }
689 
690 static struct efx_tc_recirc_id *efx_tc_get_recirc_id(struct efx_nic *efx,
691 						     u32 chain_index,
692 						     struct net_device *net_dev)
693 {
694 	struct efx_tc_recirc_id *rid, *old;
695 	int rc;
696 
697 	rid = kzalloc(sizeof(*rid), GFP_USER);
698 	if (!rid)
699 		return ERR_PTR(-ENOMEM);
700 	rid->chain_index = chain_index;
701 	/* We don't take a reference here, because it's implied - if there's
702 	 * a rule on the net_dev that's been offloaded to us, then the net_dev
703 	 * can't go away until the rule has been deoffloaded.
704 	 */
705 	rid->net_dev = net_dev;
706 	old = rhashtable_lookup_get_insert_fast(&efx->tc->recirc_ht,
707 						&rid->linkage,
708 						efx_tc_recirc_ht_params);
709 	if (old) {
710 		/* don't need our new entry */
711 		kfree(rid);
712 		if (IS_ERR(old)) /* oh dear, it's actually an error */
713 			return ERR_CAST(old);
714 		if (!refcount_inc_not_zero(&old->ref))
715 			return ERR_PTR(-EAGAIN);
716 		/* existing entry found */
717 		rid = old;
718 	} else {
719 		rc = ida_alloc_range(&efx->tc->recirc_ida, 1, U8_MAX, GFP_USER);
720 		if (rc < 0) {
721 			rhashtable_remove_fast(&efx->tc->recirc_ht,
722 					       &rid->linkage,
723 					       efx_tc_recirc_ht_params);
724 			kfree(rid);
725 			return ERR_PTR(rc);
726 		}
727 		rid->fw_id = rc;
728 		refcount_set(&rid->ref, 1);
729 	}
730 	return rid;
731 }
732 
733 static void efx_tc_put_recirc_id(struct efx_nic *efx, struct efx_tc_recirc_id *rid)
734 {
735 	if (!refcount_dec_and_test(&rid->ref))
736 		return; /* still in use */
737 	rhashtable_remove_fast(&efx->tc->recirc_ht, &rid->linkage,
738 			       efx_tc_recirc_ht_params);
739 	ida_free(&efx->tc->recirc_ida, rid->fw_id);
740 	kfree(rid);
741 }
742 
743 static void efx_tc_delete_rule(struct efx_nic *efx, struct efx_tc_flow_rule *rule)
744 {
745 	efx_mae_delete_rule(efx, rule->fw_id);
746 
747 	/* Release entries in subsidiary tables */
748 	efx_tc_free_action_set_list(efx, &rule->acts, true);
749 	if (rule->match.rid)
750 		efx_tc_put_recirc_id(efx, rule->match.rid);
751 	if (rule->match.encap)
752 		efx_tc_flower_release_encap_match(efx, rule->match.encap);
753 	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
754 }
755 
756 static const char *efx_tc_encap_type_name(enum efx_encap_type typ)
757 {
758 	switch (typ) {
759 	case EFX_ENCAP_TYPE_NONE:
760 		return "none";
761 	case EFX_ENCAP_TYPE_VXLAN:
762 		return "vxlan";
763 	case EFX_ENCAP_TYPE_GENEVE:
764 		return "geneve";
765 	default:
766 		pr_warn_once("Unknown efx_encap_type %d encountered\n", typ);
767 		return "unknown";
768 	}
769 }
770 
771 /* For details of action order constraints refer to SF-123102-TC-1§12.6.1 */
772 enum efx_tc_action_order {
773 	EFX_TC_AO_DECAP,
774 	EFX_TC_AO_DEC_TTL,
775 	EFX_TC_AO_PEDIT_MAC_ADDRS,
776 	EFX_TC_AO_VLAN_POP,
777 	EFX_TC_AO_VLAN_PUSH,
778 	EFX_TC_AO_COUNT,
779 	EFX_TC_AO_ENCAP,
780 	EFX_TC_AO_DELIVER
781 };
782 /* Determine whether we can add @new action without violating order */
783 static bool efx_tc_flower_action_order_ok(const struct efx_tc_action_set *act,
784 					  enum efx_tc_action_order new)
785 {
786 	switch (new) {
787 	case EFX_TC_AO_DECAP:
788 		if (act->decap)
789 			return false;
790 		/* PEDIT_MAC_ADDRS must not happen before DECAP, though it
791 		 * can wait until much later
792 		 */
793 		if (act->dst_mac || act->src_mac)
794 			return false;
795 
796 		/* Decrementing ttl must not happen before DECAP */
797 		if (act->do_ttl_dec)
798 			return false;
799 		fallthrough;
800 	case EFX_TC_AO_VLAN_POP:
801 		if (act->vlan_pop >= 2)
802 			return false;
803 		/* If we've already pushed a VLAN, we can't then pop it;
804 		 * the hardware would instead try to pop an existing VLAN
805 		 * before pushing the new one.
806 		 */
807 		if (act->vlan_push)
808 			return false;
809 		fallthrough;
810 	case EFX_TC_AO_VLAN_PUSH:
811 		if (act->vlan_push >= 2)
812 			return false;
813 		fallthrough;
814 	case EFX_TC_AO_COUNT:
815 		if (act->count)
816 			return false;
817 		fallthrough;
818 	case EFX_TC_AO_PEDIT_MAC_ADDRS:
819 	case EFX_TC_AO_ENCAP:
820 		if (act->encap_md)
821 			return false;
822 		fallthrough;
823 	case EFX_TC_AO_DELIVER:
824 		return !act->deliver;
825 	case EFX_TC_AO_DEC_TTL:
826 		if (act->encap_md)
827 			return false;
828 		return !act->do_ttl_dec;
829 	default:
830 		/* Bad caller.  Whatever they wanted to do, say they can't. */
831 		WARN_ON_ONCE(1);
832 		return false;
833 	}
834 }
835 
836 /**
837  * DOC: TC conntrack sequences
838  *
839  * The MAE hardware can handle at most two rounds of action rule matching,
840  * consequently we support conntrack through the notion of a "left-hand side
841  * rule".  This is a rule which typically contains only the actions "ct" and
842  * "goto chain N", and corresponds to one or more "right-hand side rules" in
843  * chain N, which typically match on +trk+est, and may perform ct(nat) actions.
844  * RHS rules go in the Action Rule table as normal but with a nonzero recirc_id
845  * (the hardware equivalent of chain_index), while LHS rules may go in either
846  * the Action Rule or the Outer Rule table, the latter being preferred for
847  * performance reasons, and set both DO_CT and a recirc_id in their response.
848  *
849  * Besides the RHS rules, there are often also similar rules matching on
850  * +trk+new which perform the ct(commit) action.  These are not offloaded.
851  */
852 
853 static bool efx_tc_rule_is_lhs_rule(struct flow_rule *fr,
854 				    struct efx_tc_match *match)
855 {
856 	const struct flow_action_entry *fa;
857 	int i;
858 
859 	flow_action_for_each(i, fa, &fr->action) {
860 		switch (fa->id) {
861 		case FLOW_ACTION_GOTO:
862 			return true;
863 		case FLOW_ACTION_CT:
864 			/* If rule is -trk, or doesn't mention trk at all, then
865 			 * a CT action implies a conntrack lookup (hence it's an
866 			 * LHS rule).  If rule is +trk, then a CT action could
867 			 * just be ct(nat) or even ct(commit) (though the latter
868 			 * can't be offloaded).
869 			 */
870 			if (!match->mask.ct_state_trk || !match->value.ct_state_trk)
871 				return true;
872 			break;
873 		default:
874 			break;
875 		}
876 	}
877 	return false;
878 }
879 
880 /* A foreign LHS rule has matches on enc_ keys at the TC layer (including an
881  * implied match on enc_ip_proto UDP).  Translate these into non-enc_ keys,
882  * so that we can use the same MAE machinery as local LHS rules (and so that
883  * the lhs_rules entries have uniform semantics).  It may seem odd to do it
884  * this way round, given that the corresponding fields in the MAE MCDIs are
885  * all ENC_, but (a) we don't have enc_L2 or enc_ip_proto in struct
886  * efx_tc_match_fields and (b) semantically an LHS rule doesn't have inner
887  * fields so it's just matching on *the* header rather than the outer header.
888  * Make sure that the non-enc_ keys were not already being matched on, as that
889  * would imply a rule that needed a triple lookup.  (Hardware can do that,
890  * with OR-AR-CT-AR, but it halves packet rate so we avoid it where possible;
891  * see efx_tc_flower_flhs_needs_ar().)
892  */
893 static int efx_tc_flower_translate_flhs_match(struct efx_tc_match *match)
894 {
895 	int rc = 0;
896 
897 #define COPY_MASK_AND_VALUE(_key, _ekey)	({	\
898 	if (match->mask._key) {				\
899 		rc = -EOPNOTSUPP;			\
900 	} else {					\
901 		match->mask._key = match->mask._ekey;	\
902 		match->mask._ekey = 0;			\
903 		match->value._key = match->value._ekey;	\
904 		match->value._ekey = 0;			\
905 	}						\
906 	rc;						\
907 })
908 #define COPY_FROM_ENC(_key)	COPY_MASK_AND_VALUE(_key, enc_##_key)
909 	if (match->mask.ip_proto)
910 		return -EOPNOTSUPP;
911 	match->mask.ip_proto = ~0;
912 	match->value.ip_proto = IPPROTO_UDP;
913 	if (COPY_FROM_ENC(src_ip) || COPY_FROM_ENC(dst_ip))
914 		return rc;
915 #ifdef CONFIG_IPV6
916 	if (!ipv6_addr_any(&match->mask.src_ip6))
917 		return -EOPNOTSUPP;
918 	match->mask.src_ip6 = match->mask.enc_src_ip6;
919 	memset(&match->mask.enc_src_ip6, 0, sizeof(struct in6_addr));
920 	if (!ipv6_addr_any(&match->mask.dst_ip6))
921 		return -EOPNOTSUPP;
922 	match->mask.dst_ip6 = match->mask.enc_dst_ip6;
923 	memset(&match->mask.enc_dst_ip6, 0, sizeof(struct in6_addr));
924 #endif
925 	if (COPY_FROM_ENC(ip_tos) || COPY_FROM_ENC(ip_ttl))
926 		return rc;
927 	/* should really copy enc_ip_frag but we don't have that in
928 	 * parse_match yet
929 	 */
930 	if (COPY_MASK_AND_VALUE(l4_sport, enc_sport) ||
931 	    COPY_MASK_AND_VALUE(l4_dport, enc_dport))
932 		return rc;
933 	return 0;
934 #undef COPY_FROM_ENC
935 #undef COPY_MASK_AND_VALUE
936 }
937 
938 /* If a foreign LHS rule wants to match on keys that are only available after
939  * encap header identification and parsing, then it can't be done in the Outer
940  * Rule lookup, because that lookup determines the encap type used to parse
941  * beyond the outer headers.  Thus, such rules must use the OR-AR-CT-AR lookup
942  * sequence, with an EM (struct efx_tc_encap_match) in the OR step.
943  * Return true iff the passed match requires this.
944  */
945 static bool efx_tc_flower_flhs_needs_ar(struct efx_tc_match *match)
946 {
947 	/* matches on inner-header keys can't be done in OR */
948 	return match->mask.eth_proto ||
949 	       match->mask.vlan_tci[0] || match->mask.vlan_tci[1] ||
950 	       match->mask.vlan_proto[0] || match->mask.vlan_proto[1] ||
951 	       memchr_inv(match->mask.eth_saddr, 0, ETH_ALEN) ||
952 	       memchr_inv(match->mask.eth_daddr, 0, ETH_ALEN) ||
953 	       match->mask.ip_proto ||
954 	       match->mask.ip_tos || match->mask.ip_ttl ||
955 	       match->mask.src_ip || match->mask.dst_ip ||
956 #ifdef CONFIG_IPV6
957 	       !ipv6_addr_any(&match->mask.src_ip6) ||
958 	       !ipv6_addr_any(&match->mask.dst_ip6) ||
959 #endif
960 	       match->mask.ip_frag || match->mask.ip_firstfrag ||
961 	       match->mask.l4_sport || match->mask.l4_dport ||
962 	       match->mask.tcp_flags ||
963 	/* nor can VNI */
964 	       match->mask.enc_keyid;
965 }
966 
967 static int efx_tc_flower_handle_lhs_actions(struct efx_nic *efx,
968 					    struct flow_cls_offload *tc,
969 					    struct flow_rule *fr,
970 					    struct net_device *net_dev,
971 					    struct efx_tc_lhs_rule *rule)
972 
973 {
974 	struct netlink_ext_ack *extack = tc->common.extack;
975 	struct efx_tc_lhs_action *act = &rule->lhs_act;
976 	const struct flow_action_entry *fa;
977 	enum efx_tc_counter_type ctype;
978 	bool pipe = true;
979 	int i;
980 
981 	ctype = rule->is_ar ? EFX_TC_COUNTER_TYPE_AR : EFX_TC_COUNTER_TYPE_OR;
982 
983 	flow_action_for_each(i, fa, &fr->action) {
984 		struct efx_tc_ct_zone *ct_zone;
985 		struct efx_tc_recirc_id *rid;
986 
987 		if (!pipe) {
988 			/* more actions after a non-pipe action */
989 			NL_SET_ERR_MSG_MOD(extack, "Action follows non-pipe action");
990 			return -EINVAL;
991 		}
992 		switch (fa->id) {
993 		case FLOW_ACTION_GOTO:
994 			if (!fa->chain_index) {
995 				NL_SET_ERR_MSG_MOD(extack, "Can't goto chain 0, no looping in hw");
996 				return -EOPNOTSUPP;
997 			}
998 			rid = efx_tc_get_recirc_id(efx, fa->chain_index,
999 						   net_dev);
1000 			if (IS_ERR(rid)) {
1001 				NL_SET_ERR_MSG_MOD(extack, "Failed to allocate a hardware recirculation ID for this chain_index");
1002 				return PTR_ERR(rid);
1003 			}
1004 			act->rid = rid;
1005 			if (fa->hw_stats) {
1006 				struct efx_tc_counter_index *cnt;
1007 
1008 				if (!(fa->hw_stats & FLOW_ACTION_HW_STATS_DELAYED)) {
1009 					NL_SET_ERR_MSG_FMT_MOD(extack,
1010 							       "hw_stats_type %u not supported (only 'delayed')",
1011 							       fa->hw_stats);
1012 					return -EOPNOTSUPP;
1013 				}
1014 				cnt = efx_tc_flower_get_counter_index(efx, tc->cookie,
1015 								      ctype);
1016 				if (IS_ERR(cnt)) {
1017 					NL_SET_ERR_MSG_MOD(extack, "Failed to obtain a counter");
1018 					return PTR_ERR(cnt);
1019 				}
1020 				WARN_ON(act->count); /* can't happen */
1021 				act->count = cnt;
1022 			}
1023 			pipe = false;
1024 			break;
1025 		case FLOW_ACTION_CT:
1026 			if (act->zone) {
1027 				NL_SET_ERR_MSG_MOD(extack, "Can't offload multiple ct actions");
1028 				return -EOPNOTSUPP;
1029 			}
1030 			if (fa->ct.action & (TCA_CT_ACT_COMMIT |
1031 					     TCA_CT_ACT_FORCE)) {
1032 				NL_SET_ERR_MSG_MOD(extack, "Can't offload ct commit/force");
1033 				return -EOPNOTSUPP;
1034 			}
1035 			if (fa->ct.action & TCA_CT_ACT_CLEAR) {
1036 				NL_SET_ERR_MSG_MOD(extack, "Can't clear ct in LHS rule");
1037 				return -EOPNOTSUPP;
1038 			}
1039 			if (fa->ct.action & (TCA_CT_ACT_NAT |
1040 					     TCA_CT_ACT_NAT_SRC |
1041 					     TCA_CT_ACT_NAT_DST)) {
1042 				NL_SET_ERR_MSG_MOD(extack, "Can't perform NAT in LHS rule - packet isn't conntracked yet");
1043 				return -EOPNOTSUPP;
1044 			}
1045 			if (fa->ct.action) {
1046 				NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled ct.action %u for LHS rule\n",
1047 						       fa->ct.action);
1048 				return -EOPNOTSUPP;
1049 			}
1050 			ct_zone = efx_tc_ct_register_zone(efx, fa->ct.zone,
1051 							  fa->ct.flow_table);
1052 			if (IS_ERR(ct_zone)) {
1053 				NL_SET_ERR_MSG_MOD(extack, "Failed to register for CT updates");
1054 				return PTR_ERR(ct_zone);
1055 			}
1056 			act->zone = ct_zone;
1057 			break;
1058 		default:
1059 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled action %u for LHS rule\n",
1060 					       fa->id);
1061 			return -EOPNOTSUPP;
1062 		}
1063 	}
1064 
1065 	if (pipe) {
1066 		NL_SET_ERR_MSG_MOD(extack, "Missing goto chain in LHS rule");
1067 		return -EOPNOTSUPP;
1068 	}
1069 	return 0;
1070 }
1071 
1072 static void efx_tc_flower_release_lhs_actions(struct efx_nic *efx,
1073 					      struct efx_tc_lhs_action *act)
1074 {
1075 	if (act->rid)
1076 		efx_tc_put_recirc_id(efx, act->rid);
1077 	if (act->zone)
1078 		efx_tc_ct_unregister_zone(efx, act->zone);
1079 	if (act->count)
1080 		efx_tc_flower_put_counter_index(efx, act->count);
1081 }
1082 
1083 /**
1084  * struct efx_tc_mangler_state - accumulates 32-bit pedits into fields
1085  *
1086  * @dst_mac_32:	dst_mac[0:3] has been populated
1087  * @dst_mac_16:	dst_mac[4:5] has been populated
1088  * @src_mac_16:	src_mac[0:1] has been populated
1089  * @src_mac_32:	src_mac[2:5] has been populated
1090  * @dst_mac:	h_dest field of ethhdr
1091  * @src_mac:	h_source field of ethhdr
1092  *
1093  * Since FLOW_ACTION_MANGLE comes in 32-bit chunks that do not
1094  * necessarily equate to whole fields of the packet header, this
1095  * structure is used to hold the cumulative effect of the partial
1096  * field pedits that have been processed so far.
1097  */
1098 struct efx_tc_mangler_state {
1099 	u8 dst_mac_32:1; /* eth->h_dest[0:3] */
1100 	u8 dst_mac_16:1; /* eth->h_dest[4:5] */
1101 	u8 src_mac_16:1; /* eth->h_source[0:1] */
1102 	u8 src_mac_32:1; /* eth->h_source[2:5] */
1103 	unsigned char dst_mac[ETH_ALEN];
1104 	unsigned char src_mac[ETH_ALEN];
1105 };
1106 
1107 /** efx_tc_complete_mac_mangle() - pull complete field pedits out of @mung
1108  * @efx:	NIC we're installing a flow rule on
1109  * @act:	action set (cursor) to update
1110  * @mung:	accumulated partial mangles
1111  * @extack:	netlink extended ack for reporting errors
1112  *
1113  * Check @mung to find any combinations of partial mangles that can be
1114  * combined into a complete packet field edit, add that edit to @act,
1115  * and consume the partial mangles from @mung.
1116  */
1117 
1118 static int efx_tc_complete_mac_mangle(struct efx_nic *efx,
1119 				      struct efx_tc_action_set *act,
1120 				      struct efx_tc_mangler_state *mung,
1121 				      struct netlink_ext_ack *extack)
1122 {
1123 	struct efx_tc_mac_pedit_action *ped;
1124 
1125 	if (mung->dst_mac_32 && mung->dst_mac_16) {
1126 		ped = efx_tc_flower_get_mac(efx, mung->dst_mac, extack);
1127 		if (IS_ERR(ped))
1128 			return PTR_ERR(ped);
1129 
1130 		/* Check that we have not already populated dst_mac */
1131 		if (act->dst_mac)
1132 			efx_tc_flower_put_mac(efx, act->dst_mac);
1133 
1134 		act->dst_mac = ped;
1135 
1136 		/* consume the incomplete state */
1137 		mung->dst_mac_32 = 0;
1138 		mung->dst_mac_16 = 0;
1139 	}
1140 	if (mung->src_mac_16 && mung->src_mac_32) {
1141 		ped = efx_tc_flower_get_mac(efx, mung->src_mac, extack);
1142 		if (IS_ERR(ped))
1143 			return PTR_ERR(ped);
1144 
1145 		/* Check that we have not already populated src_mac */
1146 		if (act->src_mac)
1147 			efx_tc_flower_put_mac(efx, act->src_mac);
1148 
1149 		act->src_mac = ped;
1150 
1151 		/* consume the incomplete state */
1152 		mung->src_mac_32 = 0;
1153 		mung->src_mac_16 = 0;
1154 	}
1155 	return 0;
1156 }
1157 
1158 static int efx_tc_pedit_add(struct efx_nic *efx, struct efx_tc_action_set *act,
1159 			    const struct flow_action_entry *fa,
1160 			    struct netlink_ext_ack *extack)
1161 {
1162 	switch (fa->mangle.htype) {
1163 	case FLOW_ACT_MANGLE_HDR_TYPE_IP4:
1164 		switch (fa->mangle.offset) {
1165 		case offsetof(struct iphdr, ttl):
1166 			/* check that pedit applies to ttl only */
1167 			if (fa->mangle.mask != ~EFX_TC_HDR_TYPE_TTL_MASK)
1168 				break;
1169 
1170 			/* Adding 0xff is equivalent to decrementing the ttl.
1171 			 * Other added values are not supported.
1172 			 */
1173 			if ((fa->mangle.val & EFX_TC_HDR_TYPE_TTL_MASK) != U8_MAX)
1174 				break;
1175 
1176 			/* check that we do not decrement ttl twice */
1177 			if (!efx_tc_flower_action_order_ok(act,
1178 							   EFX_TC_AO_DEC_TTL)) {
1179 				NL_SET_ERR_MSG_MOD(extack, "multiple dec ttl are not supported");
1180 				return -EOPNOTSUPP;
1181 			}
1182 			act->do_ttl_dec = 1;
1183 			return 0;
1184 		default:
1185 			break;
1186 		}
1187 		break;
1188 	case FLOW_ACT_MANGLE_HDR_TYPE_IP6:
1189 		switch (fa->mangle.offset) {
1190 		case round_down(offsetof(struct ipv6hdr, hop_limit), 4):
1191 			/* check that pedit applies to hoplimit only */
1192 			if (fa->mangle.mask != EFX_TC_HDR_TYPE_HLIMIT_MASK)
1193 				break;
1194 
1195 			/* Adding 0xff is equivalent to decrementing the hoplimit.
1196 			 * Other added values are not supported.
1197 			 */
1198 			if ((fa->mangle.val >> 24) != U8_MAX)
1199 				break;
1200 
1201 			/* check that we do not decrement hoplimit twice */
1202 			if (!efx_tc_flower_action_order_ok(act,
1203 							   EFX_TC_AO_DEC_TTL)) {
1204 				NL_SET_ERR_MSG_MOD(extack, "multiple dec ttl are not supported");
1205 				return -EOPNOTSUPP;
1206 			}
1207 			act->do_ttl_dec = 1;
1208 			return 0;
1209 		default:
1210 			break;
1211 		}
1212 		break;
1213 	default:
1214 		break;
1215 	}
1216 
1217 	NL_SET_ERR_MSG_FMT_MOD(extack,
1218 			       "ttl add action type %x %x %x/%x is not supported",
1219 			       fa->mangle.htype, fa->mangle.offset,
1220 			       fa->mangle.val, fa->mangle.mask);
1221 	return -EOPNOTSUPP;
1222 }
1223 
1224 /**
1225  * efx_tc_mangle() - handle a single 32-bit (or less) pedit
1226  * @efx:	NIC we're installing a flow rule on
1227  * @act:	action set (cursor) to update
1228  * @fa:		FLOW_ACTION_MANGLE action metadata
1229  * @mung:	accumulator for partial mangles
1230  * @extack:	netlink extended ack for reporting errors
1231  * @match:	original match used along with the mangle action
1232  *
1233  * Identify the fields written by a FLOW_ACTION_MANGLE, and record
1234  * the partial mangle state in @mung.  If this mangle completes an
1235  * earlier partial mangle, consume and apply to @act by calling
1236  * efx_tc_complete_mac_mangle().
1237  */
1238 
1239 static int efx_tc_mangle(struct efx_nic *efx, struct efx_tc_action_set *act,
1240 			 const struct flow_action_entry *fa,
1241 			 struct efx_tc_mangler_state *mung,
1242 			 struct netlink_ext_ack *extack,
1243 			 struct efx_tc_match *match)
1244 {
1245 	__le32 mac32;
1246 	__le16 mac16;
1247 	u8 tr_ttl;
1248 
1249 	switch (fa->mangle.htype) {
1250 	case FLOW_ACT_MANGLE_HDR_TYPE_ETH:
1251 		BUILD_BUG_ON(offsetof(struct ethhdr, h_dest) != 0);
1252 		BUILD_BUG_ON(offsetof(struct ethhdr, h_source) != 6);
1253 		if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_PEDIT_MAC_ADDRS)) {
1254 			NL_SET_ERR_MSG_MOD(extack,
1255 					   "Pedit mangle mac action violates action order");
1256 			return -EOPNOTSUPP;
1257 		}
1258 		switch (fa->mangle.offset) {
1259 		case 0:
1260 			if (fa->mangle.mask) {
1261 				NL_SET_ERR_MSG_FMT_MOD(extack,
1262 						       "mask (%#x) of eth.dst32 mangle is not supported",
1263 						       fa->mangle.mask);
1264 				return -EOPNOTSUPP;
1265 			}
1266 			/* Ethernet address is little-endian */
1267 			mac32 = cpu_to_le32(fa->mangle.val);
1268 			memcpy(mung->dst_mac, &mac32, sizeof(mac32));
1269 			mung->dst_mac_32 = 1;
1270 			return efx_tc_complete_mac_mangle(efx, act, mung, extack);
1271 		case 4:
1272 			if (fa->mangle.mask == 0xffff) {
1273 				mac16 = cpu_to_le16(fa->mangle.val >> 16);
1274 				memcpy(mung->src_mac, &mac16, sizeof(mac16));
1275 				mung->src_mac_16 = 1;
1276 			} else if (fa->mangle.mask == 0xffff0000) {
1277 				mac16 = cpu_to_le16((u16)fa->mangle.val);
1278 				memcpy(mung->dst_mac + 4, &mac16, sizeof(mac16));
1279 				mung->dst_mac_16 = 1;
1280 			} else {
1281 				NL_SET_ERR_MSG_FMT_MOD(extack,
1282 						       "mask (%#x) of eth+4 mangle is not high or low 16b",
1283 						       fa->mangle.mask);
1284 				return -EOPNOTSUPP;
1285 			}
1286 			return efx_tc_complete_mac_mangle(efx, act, mung, extack);
1287 		case 8:
1288 			if (fa->mangle.mask) {
1289 				NL_SET_ERR_MSG_FMT_MOD(extack,
1290 						       "mask (%#x) of eth.src32 mangle is not supported",
1291 						       fa->mangle.mask);
1292 				return -EOPNOTSUPP;
1293 			}
1294 			mac32 = cpu_to_le32(fa->mangle.val);
1295 			memcpy(mung->src_mac + 2, &mac32, sizeof(mac32));
1296 			mung->src_mac_32 = 1;
1297 			return efx_tc_complete_mac_mangle(efx, act, mung, extack);
1298 		default:
1299 			NL_SET_ERR_MSG_FMT_MOD(extack, "mangle eth+%u %x/%x is not supported",
1300 					       fa->mangle.offset, fa->mangle.val, fa->mangle.mask);
1301 			return -EOPNOTSUPP;
1302 		}
1303 		break;
1304 	case FLOW_ACT_MANGLE_HDR_TYPE_IP4:
1305 		switch (fa->mangle.offset) {
1306 		case offsetof(struct iphdr, ttl):
1307 			/* we currently only support pedit IP4 when it applies
1308 			 * to TTL and then only when it can be achieved with a
1309 			 * decrement ttl action
1310 			 */
1311 
1312 			/* check that pedit applies to ttl only */
1313 			if (fa->mangle.mask != ~EFX_TC_HDR_TYPE_TTL_MASK) {
1314 				NL_SET_ERR_MSG_FMT_MOD(extack,
1315 						       "mask (%#x) out of range, only support mangle action on ipv4.ttl",
1316 						       fa->mangle.mask);
1317 				return -EOPNOTSUPP;
1318 			}
1319 
1320 			/* we can only convert to a dec ttl when we have an
1321 			 * exact match on the ttl field
1322 			 */
1323 			if (match->mask.ip_ttl != U8_MAX) {
1324 				NL_SET_ERR_MSG_FMT_MOD(extack,
1325 						       "only support mangle ttl when we have an exact match, current mask (%#x)",
1326 						       match->mask.ip_ttl);
1327 				return -EOPNOTSUPP;
1328 			}
1329 
1330 			/* check that we don't try to decrement 0, which equates
1331 			 * to setting the ttl to 0xff
1332 			 */
1333 			if (match->value.ip_ttl == 0) {
1334 				NL_SET_ERR_MSG_MOD(extack,
1335 						   "decrement ttl past 0 is not supported");
1336 				return -EOPNOTSUPP;
1337 			}
1338 
1339 			/* check that we do not decrement ttl twice */
1340 			if (!efx_tc_flower_action_order_ok(act,
1341 							   EFX_TC_AO_DEC_TTL)) {
1342 				NL_SET_ERR_MSG_MOD(extack,
1343 						   "multiple dec ttl is not supported");
1344 				return -EOPNOTSUPP;
1345 			}
1346 
1347 			/* check pedit can be achieved with decrement action */
1348 			tr_ttl = match->value.ip_ttl - 1;
1349 			if ((fa->mangle.val & EFX_TC_HDR_TYPE_TTL_MASK) == tr_ttl) {
1350 				act->do_ttl_dec = 1;
1351 				return 0;
1352 			}
1353 
1354 			fallthrough;
1355 		default:
1356 			NL_SET_ERR_MSG_FMT_MOD(extack,
1357 					       "only support mangle on the ttl field (offset is %u)",
1358 					       fa->mangle.offset);
1359 			return -EOPNOTSUPP;
1360 		}
1361 		break;
1362 	case FLOW_ACT_MANGLE_HDR_TYPE_IP6:
1363 		switch (fa->mangle.offset) {
1364 		case round_down(offsetof(struct ipv6hdr, hop_limit), 4):
1365 			/* we currently only support pedit IP6 when it applies
1366 			 * to the hoplimit and then only when it can be achieved
1367 			 * with a decrement hoplimit action
1368 			 */
1369 
1370 			/* check that pedit applies to ttl only */
1371 			if (fa->mangle.mask != EFX_TC_HDR_TYPE_HLIMIT_MASK) {
1372 				NL_SET_ERR_MSG_FMT_MOD(extack,
1373 						       "mask (%#x) out of range, only support mangle action on ipv6.hop_limit",
1374 						       fa->mangle.mask);
1375 
1376 				return -EOPNOTSUPP;
1377 			}
1378 
1379 			/* we can only convert to a dec ttl when we have an
1380 			 * exact match on the ttl field
1381 			 */
1382 			if (match->mask.ip_ttl != U8_MAX) {
1383 				NL_SET_ERR_MSG_FMT_MOD(extack,
1384 						       "only support hop_limit when we have an exact match, current mask (%#x)",
1385 						       match->mask.ip_ttl);
1386 				return -EOPNOTSUPP;
1387 			}
1388 
1389 			/* check that we don't try to decrement 0, which equates
1390 			 * to setting the ttl to 0xff
1391 			 */
1392 			if (match->value.ip_ttl == 0) {
1393 				NL_SET_ERR_MSG_MOD(extack,
1394 						   "decrementing hop_limit past 0 is not supported");
1395 				return -EOPNOTSUPP;
1396 			}
1397 
1398 			/* check that we do not decrement hoplimit twice */
1399 			if (!efx_tc_flower_action_order_ok(act,
1400 							   EFX_TC_AO_DEC_TTL)) {
1401 				NL_SET_ERR_MSG_MOD(extack,
1402 						   "multiple dec ttl is not supported");
1403 				return -EOPNOTSUPP;
1404 			}
1405 
1406 			/* check pedit can be achieved with decrement action */
1407 			tr_ttl = match->value.ip_ttl - 1;
1408 			if ((fa->mangle.val >> 24) == tr_ttl) {
1409 				act->do_ttl_dec = 1;
1410 				return 0;
1411 			}
1412 
1413 			fallthrough;
1414 		default:
1415 			NL_SET_ERR_MSG_FMT_MOD(extack,
1416 					       "only support mangle on the hop_limit field");
1417 			return -EOPNOTSUPP;
1418 		}
1419 	default:
1420 		NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled mangle htype %u for action rule",
1421 				       fa->mangle.htype);
1422 		return -EOPNOTSUPP;
1423 	}
1424 	return 0;
1425 }
1426 
1427 /**
1428  * efx_tc_incomplete_mangle() - check for leftover partial pedits
1429  * @mung:	accumulator for partial mangles
1430  * @extack:	netlink extended ack for reporting errors
1431  *
1432  * Since the MAE can only overwrite whole fields, any partial
1433  * field mangle left over on reaching packet delivery (mirred or
1434  * end of TC actions) cannot be offloaded.  Check for any such
1435  * and reject them with -%EOPNOTSUPP.
1436  */
1437 
1438 static int efx_tc_incomplete_mangle(struct efx_tc_mangler_state *mung,
1439 				    struct netlink_ext_ack *extack)
1440 {
1441 	if (mung->dst_mac_32 || mung->dst_mac_16) {
1442 		NL_SET_ERR_MSG_MOD(extack, "Incomplete pedit of destination MAC address");
1443 		return -EOPNOTSUPP;
1444 	}
1445 	if (mung->src_mac_16 || mung->src_mac_32) {
1446 		NL_SET_ERR_MSG_MOD(extack, "Incomplete pedit of source MAC address");
1447 		return -EOPNOTSUPP;
1448 	}
1449 	return 0;
1450 }
1451 
1452 static int efx_tc_flower_replace_foreign_lhs_ar(struct efx_nic *efx,
1453 						struct flow_cls_offload *tc,
1454 						struct flow_rule *fr,
1455 						struct efx_tc_match *match,
1456 						struct net_device *net_dev)
1457 {
1458 	struct netlink_ext_ack *extack = tc->common.extack;
1459 	struct efx_tc_lhs_rule *rule, *old;
1460 	enum efx_encap_type type;
1461 	int rc;
1462 
1463 	type = efx_tc_indr_netdev_type(net_dev);
1464 	if (type == EFX_ENCAP_TYPE_NONE) {
1465 		NL_SET_ERR_MSG_MOD(extack, "Egress encap match on unsupported tunnel device");
1466 		return -EOPNOTSUPP;
1467 	}
1468 
1469 	rc = efx_mae_check_encap_type_supported(efx, type);
1470 	if (rc) {
1471 		NL_SET_ERR_MSG_FMT_MOD(extack,
1472 				       "Firmware reports no support for %s encap match",
1473 				       efx_tc_encap_type_name(type));
1474 		return rc;
1475 	}
1476 	/* This is an Action Rule, so it needs a separate Encap Match in the
1477 	 * Outer Rule table.  Insert that now.
1478 	 */
1479 	rc = efx_tc_flower_record_encap_match(efx, match, type,
1480 					      EFX_TC_EM_DIRECT, 0, 0, extack);
1481 	if (rc)
1482 		return rc;
1483 
1484 	match->mask.recirc_id = 0xff;
1485 	if (match->mask.ct_state_trk && match->value.ct_state_trk) {
1486 		NL_SET_ERR_MSG_MOD(extack, "LHS rule can never match +trk");
1487 		rc = -EOPNOTSUPP;
1488 		goto release_encap_match;
1489 	}
1490 	/* LHS rules are always -trk, so we don't need to match on that */
1491 	match->mask.ct_state_trk = 0;
1492 	match->value.ct_state_trk = 0;
1493 	/* We must inhibit match on TCP SYN/FIN/RST, so that SW can see
1494 	 * the packet and update the conntrack table.
1495 	 * Outer Rules will do that with CT_TCP_FLAGS_INHIBIT, but Action
1496 	 * Rules don't have that; instead they support matching on
1497 	 * TCP_SYN_FIN_RST (aka TCP_INTERESTING_FLAGS), so use that.
1498 	 * This is only strictly needed if there will be a DO_CT action,
1499 	 * which we don't know yet, but typically there will be and it's
1500 	 * simpler not to bother checking here.
1501 	 */
1502 	match->mask.tcp_syn_fin_rst = true;
1503 
1504 	rc = efx_mae_match_check_caps(efx, &match->mask, extack);
1505 	if (rc)
1506 		goto release_encap_match;
1507 
1508 	rule = kzalloc(sizeof(*rule), GFP_USER);
1509 	if (!rule) {
1510 		rc = -ENOMEM;
1511 		goto release_encap_match;
1512 	}
1513 	rule->cookie = tc->cookie;
1514 	rule->is_ar = true;
1515 	old = rhashtable_lookup_get_insert_fast(&efx->tc->lhs_rule_ht,
1516 						&rule->linkage,
1517 						efx_tc_lhs_rule_ht_params);
1518 	if (old) {
1519 		netif_dbg(efx, drv, efx->net_dev,
1520 			  "Already offloaded rule (cookie %lx)\n", tc->cookie);
1521 		rc = -EEXIST;
1522 		NL_SET_ERR_MSG_MOD(extack, "Rule already offloaded");
1523 		goto release;
1524 	}
1525 
1526 	/* Parse actions */
1527 	rc = efx_tc_flower_handle_lhs_actions(efx, tc, fr, net_dev, rule);
1528 	if (rc)
1529 		goto release;
1530 
1531 	rule->match = *match;
1532 	rule->lhs_act.tun_type = type;
1533 
1534 	rc = efx_mae_insert_lhs_rule(efx, rule, EFX_TC_PRIO_TC);
1535 	if (rc) {
1536 		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
1537 		goto release;
1538 	}
1539 	netif_dbg(efx, drv, efx->net_dev,
1540 		  "Successfully parsed lhs rule (cookie %lx)\n",
1541 		  tc->cookie);
1542 	return 0;
1543 
1544 release:
1545 	efx_tc_flower_release_lhs_actions(efx, &rule->lhs_act);
1546 	if (!old)
1547 		rhashtable_remove_fast(&efx->tc->lhs_rule_ht, &rule->linkage,
1548 				       efx_tc_lhs_rule_ht_params);
1549 	kfree(rule);
1550 release_encap_match:
1551 	if (match->encap)
1552 		efx_tc_flower_release_encap_match(efx, match->encap);
1553 	return rc;
1554 }
1555 
1556 static int efx_tc_flower_replace_foreign_lhs(struct efx_nic *efx,
1557 					     struct flow_cls_offload *tc,
1558 					     struct flow_rule *fr,
1559 					     struct efx_tc_match *match,
1560 					     struct net_device *net_dev)
1561 {
1562 	struct netlink_ext_ack *extack = tc->common.extack;
1563 	struct efx_tc_lhs_rule *rule, *old;
1564 	enum efx_encap_type type;
1565 	int rc;
1566 
1567 	if (tc->common.chain_index) {
1568 		NL_SET_ERR_MSG_MOD(extack, "LHS rule only allowed in chain 0");
1569 		return -EOPNOTSUPP;
1570 	}
1571 
1572 	if (!efx_tc_match_is_encap(&match->mask)) {
1573 		/* This is not a tunnel decap rule, ignore it */
1574 		netif_dbg(efx, drv, efx->net_dev, "Ignoring foreign LHS filter without encap match\n");
1575 		return -EOPNOTSUPP;
1576 	}
1577 
1578 	if (efx_tc_flower_flhs_needs_ar(match))
1579 		return efx_tc_flower_replace_foreign_lhs_ar(efx, tc, fr, match,
1580 							    net_dev);
1581 
1582 	type = efx_tc_indr_netdev_type(net_dev);
1583 	if (type == EFX_ENCAP_TYPE_NONE) {
1584 		NL_SET_ERR_MSG_MOD(extack, "Egress encap match on unsupported tunnel device\n");
1585 		return -EOPNOTSUPP;
1586 	}
1587 
1588 	rc = efx_mae_check_encap_type_supported(efx, type);
1589 	if (rc) {
1590 		NL_SET_ERR_MSG_FMT_MOD(extack,
1591 				       "Firmware reports no support for %s encap match",
1592 				       efx_tc_encap_type_name(type));
1593 		return rc;
1594 	}
1595 	/* Reserve the outer tuple with a pseudo Encap Match */
1596 	rc = efx_tc_flower_record_encap_match(efx, match, type,
1597 					      EFX_TC_EM_PSEUDO_OR, 0, 0,
1598 					      extack);
1599 	if (rc)
1600 		return rc;
1601 
1602 	if (match->mask.ct_state_trk && match->value.ct_state_trk) {
1603 		NL_SET_ERR_MSG_MOD(extack, "LHS rule can never match +trk");
1604 		rc = -EOPNOTSUPP;
1605 		goto release_encap_match;
1606 	}
1607 	/* LHS rules are always -trk, so we don't need to match on that */
1608 	match->mask.ct_state_trk = 0;
1609 	match->value.ct_state_trk = 0;
1610 
1611 	rc = efx_tc_flower_translate_flhs_match(match);
1612 	if (rc) {
1613 		NL_SET_ERR_MSG_MOD(extack, "LHS rule cannot match on inner fields");
1614 		goto release_encap_match;
1615 	}
1616 
1617 	rc = efx_mae_match_check_caps_lhs(efx, &match->mask, extack);
1618 	if (rc)
1619 		goto release_encap_match;
1620 
1621 	rule = kzalloc(sizeof(*rule), GFP_USER);
1622 	if (!rule) {
1623 		rc = -ENOMEM;
1624 		goto release_encap_match;
1625 	}
1626 	rule->cookie = tc->cookie;
1627 	old = rhashtable_lookup_get_insert_fast(&efx->tc->lhs_rule_ht,
1628 						&rule->linkage,
1629 						efx_tc_lhs_rule_ht_params);
1630 	if (old) {
1631 		netif_dbg(efx, drv, efx->net_dev,
1632 			  "Already offloaded rule (cookie %lx)\n", tc->cookie);
1633 		rc = -EEXIST;
1634 		NL_SET_ERR_MSG_MOD(extack, "Rule already offloaded");
1635 		goto release;
1636 	}
1637 
1638 	/* Parse actions */
1639 	rc = efx_tc_flower_handle_lhs_actions(efx, tc, fr, net_dev, rule);
1640 	if (rc)
1641 		goto release;
1642 
1643 	rule->match = *match;
1644 	rule->lhs_act.tun_type = type;
1645 
1646 	rc = efx_mae_insert_lhs_rule(efx, rule, EFX_TC_PRIO_TC);
1647 	if (rc) {
1648 		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
1649 		goto release;
1650 	}
1651 	netif_dbg(efx, drv, efx->net_dev,
1652 		  "Successfully parsed lhs rule (cookie %lx)\n",
1653 		  tc->cookie);
1654 	return 0;
1655 
1656 release:
1657 	efx_tc_flower_release_lhs_actions(efx, &rule->lhs_act);
1658 	if (!old)
1659 		rhashtable_remove_fast(&efx->tc->lhs_rule_ht, &rule->linkage,
1660 				       efx_tc_lhs_rule_ht_params);
1661 	kfree(rule);
1662 release_encap_match:
1663 	if (match->encap)
1664 		efx_tc_flower_release_encap_match(efx, match->encap);
1665 	return rc;
1666 }
1667 
1668 static int efx_tc_flower_replace_foreign(struct efx_nic *efx,
1669 					 struct net_device *net_dev,
1670 					 struct flow_cls_offload *tc)
1671 {
1672 	struct flow_rule *fr = flow_cls_offload_flow_rule(tc);
1673 	struct netlink_ext_ack *extack = tc->common.extack;
1674 	struct efx_tc_flow_rule *rule = NULL, *old = NULL;
1675 	struct efx_tc_action_set *act = NULL;
1676 	bool found = false, uplinked = false;
1677 	const struct flow_action_entry *fa;
1678 	struct efx_tc_match match;
1679 	struct efx_rep *to_efv;
1680 	s64 rc;
1681 	int i;
1682 
1683 	/* Parse match */
1684 	memset(&match, 0, sizeof(match));
1685 	rc = efx_tc_flower_parse_match(efx, fr, &match, extack);
1686 	if (rc)
1687 		return rc;
1688 	/* The rule as given to us doesn't specify a source netdevice.
1689 	 * But, determining whether packets from a VF should match it is
1690 	 * complicated, so leave those to the software slowpath: qualify
1691 	 * the filter with source m-port == wire.
1692 	 */
1693 	rc = efx_tc_flower_external_mport(efx, EFX_EFV_PF);
1694 	if (rc < 0) {
1695 		NL_SET_ERR_MSG_MOD(extack, "Failed to identify ingress m-port for foreign filter");
1696 		return rc;
1697 	}
1698 	match.value.ingress_port = rc;
1699 	match.mask.ingress_port = ~0;
1700 
1701 	if (efx_tc_rule_is_lhs_rule(fr, &match))
1702 		return efx_tc_flower_replace_foreign_lhs(efx, tc, fr, &match,
1703 							 net_dev);
1704 
1705 	if (tc->common.chain_index) {
1706 		struct efx_tc_recirc_id *rid;
1707 
1708 		rid = efx_tc_get_recirc_id(efx, tc->common.chain_index, net_dev);
1709 		if (IS_ERR(rid)) {
1710 			NL_SET_ERR_MSG_FMT_MOD(extack,
1711 					       "Failed to allocate a hardware recirculation ID for chain_index %u",
1712 					       tc->common.chain_index);
1713 			return PTR_ERR(rid);
1714 		}
1715 		match.rid = rid;
1716 		match.value.recirc_id = rid->fw_id;
1717 	}
1718 	match.mask.recirc_id = 0xff;
1719 
1720 	/* AR table can't match on DO_CT (+trk).  But a commonly used pattern is
1721 	 * +trk+est, which is strictly implied by +est, so rewrite it to that.
1722 	 */
1723 	if (match.mask.ct_state_trk && match.value.ct_state_trk &&
1724 	    match.mask.ct_state_est && match.value.ct_state_est)
1725 		match.mask.ct_state_trk = 0;
1726 	/* Thanks to CT_TCP_FLAGS_INHIBIT, packets with interesting flags could
1727 	 * match +trk-est (CT_HIT=0) despite being on an established connection.
1728 	 * So make -est imply -tcp_syn_fin_rst match to ensure these packets
1729 	 * still hit the software path.
1730 	 */
1731 	if (match.mask.ct_state_est && !match.value.ct_state_est) {
1732 		if (match.value.tcp_syn_fin_rst) {
1733 			/* Can't offload this combination */
1734 			NL_SET_ERR_MSG_MOD(extack, "TCP flags and -est conflict for offload");
1735 			rc = -EOPNOTSUPP;
1736 			goto release;
1737 		}
1738 		match.mask.tcp_syn_fin_rst = true;
1739 	}
1740 
1741 	flow_action_for_each(i, fa, &fr->action) {
1742 		switch (fa->id) {
1743 		case FLOW_ACTION_REDIRECT:
1744 		case FLOW_ACTION_MIRRED: /* mirred means mirror here */
1745 			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
1746 			if (IS_ERR(to_efv))
1747 				continue;
1748 			found = true;
1749 			break;
1750 		default:
1751 			break;
1752 		}
1753 	}
1754 	if (!found) { /* We don't care. */
1755 		netif_dbg(efx, drv, efx->net_dev,
1756 			  "Ignoring foreign filter that doesn't egdev us\n");
1757 		rc = -EOPNOTSUPP;
1758 		goto release;
1759 	}
1760 
1761 	rc = efx_mae_match_check_caps(efx, &match.mask, extack);
1762 	if (rc)
1763 		goto release;
1764 
1765 	if (efx_tc_match_is_encap(&match.mask)) {
1766 		enum efx_encap_type type;
1767 
1768 		type = efx_tc_indr_netdev_type(net_dev);
1769 		if (type == EFX_ENCAP_TYPE_NONE) {
1770 			NL_SET_ERR_MSG_MOD(extack,
1771 					   "Egress encap match on unsupported tunnel device");
1772 			rc = -EOPNOTSUPP;
1773 			goto release;
1774 		}
1775 
1776 		rc = efx_mae_check_encap_type_supported(efx, type);
1777 		if (rc) {
1778 			NL_SET_ERR_MSG_FMT_MOD(extack,
1779 					       "Firmware reports no support for %s encap match",
1780 					       efx_tc_encap_type_name(type));
1781 			goto release;
1782 		}
1783 
1784 		rc = efx_tc_flower_record_encap_match(efx, &match, type,
1785 						      EFX_TC_EM_DIRECT, 0, 0,
1786 						      extack);
1787 		if (rc)
1788 			goto release;
1789 	} else if (!tc->common.chain_index) {
1790 		/* This is not a tunnel decap rule, ignore it */
1791 		netif_dbg(efx, drv, efx->net_dev,
1792 			  "Ignoring foreign filter without encap match\n");
1793 		rc = -EOPNOTSUPP;
1794 		goto release;
1795 	}
1796 
1797 	rule = kzalloc(sizeof(*rule), GFP_USER);
1798 	if (!rule) {
1799 		rc = -ENOMEM;
1800 		goto release;
1801 	}
1802 	INIT_LIST_HEAD(&rule->acts.list);
1803 	rule->cookie = tc->cookie;
1804 	old = rhashtable_lookup_get_insert_fast(&efx->tc->match_action_ht,
1805 						&rule->linkage,
1806 						efx_tc_match_action_ht_params);
1807 	if (IS_ERR(old)) {
1808 		rc = PTR_ERR(old);
1809 		goto release;
1810 	} else if (old) {
1811 		netif_dbg(efx, drv, efx->net_dev,
1812 			  "Ignoring already-offloaded rule (cookie %lx)\n",
1813 			  tc->cookie);
1814 		rc = -EEXIST;
1815 		goto release;
1816 	}
1817 
1818 	act = kzalloc(sizeof(*act), GFP_USER);
1819 	if (!act) {
1820 		rc = -ENOMEM;
1821 		goto release;
1822 	}
1823 
1824 	/* Parse actions.  For foreign rules we only support decap & redirect.
1825 	 * See corresponding code in efx_tc_flower_replace() for theory of
1826 	 * operation & how 'act' cursor is used.
1827 	 */
1828 	flow_action_for_each(i, fa, &fr->action) {
1829 		struct efx_tc_action_set save;
1830 
1831 		switch (fa->id) {
1832 		case FLOW_ACTION_REDIRECT:
1833 		case FLOW_ACTION_MIRRED:
1834 			/* See corresponding code in efx_tc_flower_replace() for
1835 			 * long explanations of what's going on here.
1836 			 */
1837 			save = *act;
1838 			if (fa->hw_stats) {
1839 				struct efx_tc_counter_index *ctr;
1840 
1841 				if (!(fa->hw_stats & FLOW_ACTION_HW_STATS_DELAYED)) {
1842 					NL_SET_ERR_MSG_FMT_MOD(extack,
1843 							       "hw_stats_type %u not supported (only 'delayed')",
1844 							       fa->hw_stats);
1845 					rc = -EOPNOTSUPP;
1846 					goto release;
1847 				}
1848 				if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_COUNT)) {
1849 					NL_SET_ERR_MSG_MOD(extack, "Count action violates action order (can't happen)");
1850 					rc = -EOPNOTSUPP;
1851 					goto release;
1852 				}
1853 
1854 				ctr = efx_tc_flower_get_counter_index(efx,
1855 								      tc->cookie,
1856 								      EFX_TC_COUNTER_TYPE_AR);
1857 				if (IS_ERR(ctr)) {
1858 					rc = PTR_ERR(ctr);
1859 					NL_SET_ERR_MSG_MOD(extack, "Failed to obtain a counter");
1860 					goto release;
1861 				}
1862 				act->count = ctr;
1863 				INIT_LIST_HEAD(&act->count_user);
1864 			}
1865 
1866 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DELIVER)) {
1867 				/* can't happen */
1868 				rc = -EOPNOTSUPP;
1869 				NL_SET_ERR_MSG_MOD(extack,
1870 						   "Deliver action violates action order (can't happen)");
1871 				goto release;
1872 			}
1873 			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
1874 			/* PF implies egdev is us, in which case we really
1875 			 * want to deliver to the uplink (because this is an
1876 			 * ingress filter).  If we don't recognise the egdev
1877 			 * at all, then we'd better trap so SW can handle it.
1878 			 */
1879 			if (IS_ERR(to_efv))
1880 				to_efv = EFX_EFV_PF;
1881 			if (to_efv == EFX_EFV_PF) {
1882 				if (uplinked)
1883 					break;
1884 				uplinked = true;
1885 			}
1886 			rc = efx_tc_flower_internal_mport(efx, to_efv);
1887 			if (rc < 0) {
1888 				NL_SET_ERR_MSG_MOD(extack, "Failed to identify egress m-port");
1889 				goto release;
1890 			}
1891 			act->dest_mport = rc;
1892 			act->deliver = 1;
1893 			rc = efx_mae_alloc_action_set(efx, act);
1894 			if (rc) {
1895 				NL_SET_ERR_MSG_MOD(extack,
1896 						   "Failed to write action set to hw (mirred)");
1897 				goto release;
1898 			}
1899 			list_add_tail(&act->list, &rule->acts.list);
1900 			act = NULL;
1901 			if (fa->id == FLOW_ACTION_REDIRECT)
1902 				break; /* end of the line */
1903 			/* Mirror, so continue on with saved act */
1904 			act = kzalloc(sizeof(*act), GFP_USER);
1905 			if (!act) {
1906 				rc = -ENOMEM;
1907 				goto release;
1908 			}
1909 			*act = save;
1910 			break;
1911 		case FLOW_ACTION_TUNNEL_DECAP:
1912 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DECAP)) {
1913 				rc = -EINVAL;
1914 				NL_SET_ERR_MSG_MOD(extack, "Decap action violates action order");
1915 				goto release;
1916 			}
1917 			act->decap = 1;
1918 			/* If we previously delivered/trapped to uplink, now
1919 			 * that we've decapped we'll want another copy if we
1920 			 * try to deliver/trap to uplink again.
1921 			 */
1922 			uplinked = false;
1923 			break;
1924 		default:
1925 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled action %u",
1926 					       fa->id);
1927 			rc = -EOPNOTSUPP;
1928 			goto release;
1929 		}
1930 	}
1931 
1932 	if (act) {
1933 		if (!uplinked) {
1934 			/* Not shot/redirected, so deliver to default dest (which is
1935 			 * the uplink, as this is an ingress filter)
1936 			 */
1937 			efx_mae_mport_uplink(efx, &act->dest_mport);
1938 			act->deliver = 1;
1939 		}
1940 		rc = efx_mae_alloc_action_set(efx, act);
1941 		if (rc) {
1942 			NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (deliver)");
1943 			goto release;
1944 		}
1945 		list_add_tail(&act->list, &rule->acts.list);
1946 		act = NULL; /* Prevent double-free in error path */
1947 	}
1948 
1949 	rule->match = match;
1950 
1951 	netif_dbg(efx, drv, efx->net_dev,
1952 		  "Successfully parsed foreign filter (cookie %lx)\n",
1953 		  tc->cookie);
1954 
1955 	rc = efx_mae_alloc_action_set_list(efx, &rule->acts);
1956 	if (rc) {
1957 		NL_SET_ERR_MSG_MOD(extack, "Failed to write action set list to hw");
1958 		goto release;
1959 	}
1960 	rc = efx_mae_insert_rule(efx, &rule->match, EFX_TC_PRIO_TC,
1961 				 rule->acts.fw_id, &rule->fw_id);
1962 	if (rc) {
1963 		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
1964 		goto release_acts;
1965 	}
1966 	return 0;
1967 
1968 release_acts:
1969 	efx_mae_free_action_set_list(efx, &rule->acts);
1970 release:
1971 	/* We failed to insert the rule, so free up any entries we created in
1972 	 * subsidiary tables.
1973 	 */
1974 	if (match.rid)
1975 		efx_tc_put_recirc_id(efx, match.rid);
1976 	if (act)
1977 		efx_tc_free_action_set(efx, act, false);
1978 	if (rule) {
1979 		if (!old)
1980 			rhashtable_remove_fast(&efx->tc->match_action_ht,
1981 					       &rule->linkage,
1982 					       efx_tc_match_action_ht_params);
1983 		efx_tc_free_action_set_list(efx, &rule->acts, false);
1984 	}
1985 	kfree(rule);
1986 	if (match.encap)
1987 		efx_tc_flower_release_encap_match(efx, match.encap);
1988 	return rc;
1989 }
1990 
1991 static int efx_tc_flower_replace_lhs(struct efx_nic *efx,
1992 				     struct flow_cls_offload *tc,
1993 				     struct flow_rule *fr,
1994 				     struct efx_tc_match *match,
1995 				     struct efx_rep *efv,
1996 				     struct net_device *net_dev)
1997 {
1998 	struct netlink_ext_ack *extack = tc->common.extack;
1999 	struct efx_tc_lhs_rule *rule, *old;
2000 	int rc;
2001 
2002 	if (tc->common.chain_index) {
2003 		NL_SET_ERR_MSG_MOD(extack, "LHS rule only allowed in chain 0");
2004 		return -EOPNOTSUPP;
2005 	}
2006 
2007 	if (match->mask.ct_state_trk && match->value.ct_state_trk) {
2008 		NL_SET_ERR_MSG_MOD(extack, "LHS rule can never match +trk");
2009 		return -EOPNOTSUPP;
2010 	}
2011 	/* LHS rules are always -trk, so we don't need to match on that */
2012 	match->mask.ct_state_trk = 0;
2013 	match->value.ct_state_trk = 0;
2014 
2015 	rc = efx_mae_match_check_caps_lhs(efx, &match->mask, extack);
2016 	if (rc)
2017 		return rc;
2018 
2019 	rule = kzalloc(sizeof(*rule), GFP_USER);
2020 	if (!rule)
2021 		return -ENOMEM;
2022 	rule->cookie = tc->cookie;
2023 	old = rhashtable_lookup_get_insert_fast(&efx->tc->lhs_rule_ht,
2024 						&rule->linkage,
2025 						efx_tc_lhs_rule_ht_params);
2026 	if (IS_ERR(old)) {
2027 		rc = PTR_ERR(old);
2028 		goto release;
2029 	} else if (old) {
2030 		netif_dbg(efx, drv, efx->net_dev,
2031 			  "Already offloaded rule (cookie %lx)\n", tc->cookie);
2032 		rc = -EEXIST;
2033 		NL_SET_ERR_MSG_MOD(extack, "Rule already offloaded");
2034 		goto release;
2035 	}
2036 
2037 	/* Parse actions */
2038 	/* See note in efx_tc_flower_replace() regarding passed net_dev
2039 	 * (used for efx_tc_get_recirc_id()).
2040 	 */
2041 	rc = efx_tc_flower_handle_lhs_actions(efx, tc, fr, efx->net_dev, rule);
2042 	if (rc)
2043 		goto release;
2044 
2045 	rule->match = *match;
2046 
2047 	rc = efx_mae_insert_lhs_rule(efx, rule, EFX_TC_PRIO_TC);
2048 	if (rc) {
2049 		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
2050 		goto release;
2051 	}
2052 	netif_dbg(efx, drv, efx->net_dev,
2053 		  "Successfully parsed lhs rule (cookie %lx)\n",
2054 		  tc->cookie);
2055 	return 0;
2056 
2057 release:
2058 	efx_tc_flower_release_lhs_actions(efx, &rule->lhs_act);
2059 	if (!old)
2060 		rhashtable_remove_fast(&efx->tc->lhs_rule_ht, &rule->linkage,
2061 				       efx_tc_lhs_rule_ht_params);
2062 	kfree(rule);
2063 	return rc;
2064 }
2065 
2066 static int efx_tc_flower_replace(struct efx_nic *efx,
2067 				 struct net_device *net_dev,
2068 				 struct flow_cls_offload *tc,
2069 				 struct efx_rep *efv)
2070 {
2071 	struct flow_rule *fr = flow_cls_offload_flow_rule(tc);
2072 	struct netlink_ext_ack *extack = tc->common.extack;
2073 	const struct ip_tunnel_info *encap_info = NULL;
2074 	struct efx_tc_flow_rule *rule = NULL, *old;
2075 	struct efx_tc_mangler_state mung = {};
2076 	struct efx_tc_action_set *act = NULL;
2077 	const struct flow_action_entry *fa;
2078 	struct efx_rep *from_efv, *to_efv;
2079 	struct efx_tc_match match;
2080 	u32 acts_id;
2081 	s64 rc;
2082 	int i;
2083 
2084 	if (!tc_can_offload_extack(efx->net_dev, extack))
2085 		return -EOPNOTSUPP;
2086 	if (WARN_ON(!efx->tc))
2087 		return -ENETDOWN;
2088 	if (WARN_ON(!efx->tc->up))
2089 		return -ENETDOWN;
2090 
2091 	from_efv = efx_tc_flower_lookup_efv(efx, net_dev);
2092 	if (IS_ERR(from_efv)) {
2093 		/* Not from our PF or representors, so probably a tunnel dev */
2094 		return efx_tc_flower_replace_foreign(efx, net_dev, tc);
2095 	}
2096 
2097 	if (efv != from_efv) {
2098 		/* can't happen */
2099 		NL_SET_ERR_MSG_FMT_MOD(extack, "for %s efv is %snull but from_efv is %snull (can't happen)",
2100 				       netdev_name(net_dev), efv ? "non-" : "",
2101 				       from_efv ? "non-" : "");
2102 		return -EINVAL;
2103 	}
2104 
2105 	/* Parse match */
2106 	memset(&match, 0, sizeof(match));
2107 	rc = efx_tc_flower_external_mport(efx, from_efv);
2108 	if (rc < 0) {
2109 		NL_SET_ERR_MSG_MOD(extack, "Failed to identify ingress m-port");
2110 		return rc;
2111 	}
2112 	match.value.ingress_port = rc;
2113 	match.mask.ingress_port = ~0;
2114 	rc = efx_tc_flower_parse_match(efx, fr, &match, extack);
2115 	if (rc)
2116 		return rc;
2117 	if (efx_tc_match_is_encap(&match.mask)) {
2118 		NL_SET_ERR_MSG_MOD(extack, "Ingress enc_key matches not supported");
2119 		return -EOPNOTSUPP;
2120 	}
2121 
2122 	if (efx_tc_rule_is_lhs_rule(fr, &match))
2123 		return efx_tc_flower_replace_lhs(efx, tc, fr, &match, efv,
2124 						 net_dev);
2125 
2126 	/* chain_index 0 is always recirc_id 0 (and does not appear in recirc_ht).
2127 	 * Conveniently, match.rid == NULL and match.value.recirc_id == 0 owing
2128 	 * to the initial memset(), so we don't need to do anything in that case.
2129 	 */
2130 	if (tc->common.chain_index) {
2131 		struct efx_tc_recirc_id *rid;
2132 
2133 		/* Note regarding passed net_dev:
2134 		 * VFreps and PF can share chain namespace, as they have
2135 		 * distinct ingress_mports.  So we don't need to burn an
2136 		 * extra recirc_id if both use the same chain_index.
2137 		 * (Strictly speaking, we could give each VFrep its own
2138 		 * recirc_id namespace that doesn't take IDs away from the
2139 		 * PF, but that would require a bunch of additional IDAs -
2140 		 * one for each representor - and that's not likely to be
2141 		 * the main cause of recirc_id exhaustion anyway.)
2142 		 */
2143 		rid = efx_tc_get_recirc_id(efx, tc->common.chain_index,
2144 					   efx->net_dev);
2145 		if (IS_ERR(rid)) {
2146 			NL_SET_ERR_MSG_FMT_MOD(extack,
2147 					       "Failed to allocate a hardware recirculation ID for chain_index %u",
2148 					       tc->common.chain_index);
2149 			return PTR_ERR(rid);
2150 		}
2151 		match.rid = rid;
2152 		match.value.recirc_id = rid->fw_id;
2153 	}
2154 	match.mask.recirc_id = 0xff;
2155 
2156 	/* AR table can't match on DO_CT (+trk).  But a commonly used pattern is
2157 	 * +trk+est, which is strictly implied by +est, so rewrite it to that.
2158 	 */
2159 	if (match.mask.ct_state_trk && match.value.ct_state_trk &&
2160 	    match.mask.ct_state_est && match.value.ct_state_est)
2161 		match.mask.ct_state_trk = 0;
2162 	/* Thanks to CT_TCP_FLAGS_INHIBIT, packets with interesting flags could
2163 	 * match +trk-est (CT_HIT=0) despite being on an established connection.
2164 	 * So make -est imply -tcp_syn_fin_rst match to ensure these packets
2165 	 * still hit the software path.
2166 	 */
2167 	if (match.mask.ct_state_est && !match.value.ct_state_est) {
2168 		if (match.value.tcp_syn_fin_rst) {
2169 			/* Can't offload this combination */
2170 			rc = -EOPNOTSUPP;
2171 			goto release;
2172 		}
2173 		match.mask.tcp_syn_fin_rst = true;
2174 	}
2175 
2176 	rc = efx_mae_match_check_caps(efx, &match.mask, extack);
2177 	if (rc)
2178 		goto release;
2179 
2180 	rule = kzalloc(sizeof(*rule), GFP_USER);
2181 	if (!rule) {
2182 		rc = -ENOMEM;
2183 		goto release;
2184 	}
2185 	INIT_LIST_HEAD(&rule->acts.list);
2186 	rule->cookie = tc->cookie;
2187 	old = rhashtable_lookup_get_insert_fast(&efx->tc->match_action_ht,
2188 						&rule->linkage,
2189 						efx_tc_match_action_ht_params);
2190 	if (IS_ERR(old)) {
2191 		rc = PTR_ERR(old);
2192 		goto release;
2193 	} else if (old) {
2194 		netif_dbg(efx, drv, efx->net_dev,
2195 			  "Already offloaded rule (cookie %lx)\n", tc->cookie);
2196 		NL_SET_ERR_MSG_MOD(extack, "Rule already offloaded");
2197 		rc = -EEXIST;
2198 		goto release;
2199 	}
2200 
2201 	/* Parse actions */
2202 	act = kzalloc(sizeof(*act), GFP_USER);
2203 	if (!act) {
2204 		rc = -ENOMEM;
2205 		goto release;
2206 	}
2207 
2208 	/**
2209 	 * DOC: TC action translation
2210 	 *
2211 	 * Actions in TC are sequential and cumulative, with delivery actions
2212 	 * potentially anywhere in the order.  The EF100 MAE, however, takes
2213 	 * an 'action set list' consisting of 'action sets', each of which is
2214 	 * applied to the _original_ packet, and consists of a set of optional
2215 	 * actions in a fixed order with delivery at the end.
2216 	 * To translate between these two models, we maintain a 'cursor', @act,
2217 	 * which describes the cumulative effect of all the packet-mutating
2218 	 * actions encountered so far; on handling a delivery (mirred or drop)
2219 	 * action, once the action-set has been inserted into hardware, we
2220 	 * append @act to the action-set list (@rule->acts); if this is a pipe
2221 	 * action (mirred mirror) we then allocate a new @act with a copy of
2222 	 * the cursor state _before_ the delivery action, otherwise we set @act
2223 	 * to %NULL.
2224 	 * This ensures that every allocated action-set is either attached to
2225 	 * @rule->acts or pointed to by @act (and never both), and that only
2226 	 * those action-sets in @rule->acts exist in hardware.  Consequently,
2227 	 * in the failure path, @act only needs to be freed in memory, whereas
2228 	 * for @rule->acts we remove each action-set from hardware before
2229 	 * freeing it (efx_tc_free_action_set_list()), even if the action-set
2230 	 * list itself is not in hardware.
2231 	 */
2232 	flow_action_for_each(i, fa, &fr->action) {
2233 		struct efx_tc_action_set save;
2234 		u16 tci;
2235 
2236 		if (!act) {
2237 			/* more actions after a non-pipe action */
2238 			NL_SET_ERR_MSG_MOD(extack, "Action follows non-pipe action");
2239 			rc = -EINVAL;
2240 			goto release;
2241 		}
2242 
2243 		if ((fa->id == FLOW_ACTION_REDIRECT ||
2244 		     fa->id == FLOW_ACTION_MIRRED ||
2245 		     fa->id == FLOW_ACTION_DROP) && fa->hw_stats) {
2246 			struct efx_tc_counter_index *ctr;
2247 
2248 			/* Currently the only actions that want stats are
2249 			 * mirred and gact (ok, shot, trap, goto-chain), which
2250 			 * means we want stats just before delivery.  Also,
2251 			 * note that tunnel_key set shouldn't change the length
2252 			 * — it's only the subsequent mirred that does that,
2253 			 * and the stats are taken _before_ the mirred action
2254 			 * happens.
2255 			 */
2256 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_COUNT)) {
2257 				/* All supported actions that count either steal
2258 				 * (gact shot, mirred redirect) or clone act
2259 				 * (mirred mirror), so we should never get two
2260 				 * count actions on one action_set.
2261 				 */
2262 				NL_SET_ERR_MSG_MOD(extack, "Count-action conflict (can't happen)");
2263 				rc = -EOPNOTSUPP;
2264 				goto release;
2265 			}
2266 
2267 			if (!(fa->hw_stats & FLOW_ACTION_HW_STATS_DELAYED)) {
2268 				NL_SET_ERR_MSG_FMT_MOD(extack, "hw_stats_type %u not supported (only 'delayed')",
2269 						       fa->hw_stats);
2270 				rc = -EOPNOTSUPP;
2271 				goto release;
2272 			}
2273 
2274 			ctr = efx_tc_flower_get_counter_index(efx, tc->cookie,
2275 							      EFX_TC_COUNTER_TYPE_AR);
2276 			if (IS_ERR(ctr)) {
2277 				rc = PTR_ERR(ctr);
2278 				NL_SET_ERR_MSG_MOD(extack, "Failed to obtain a counter");
2279 				goto release;
2280 			}
2281 			act->count = ctr;
2282 			INIT_LIST_HEAD(&act->count_user);
2283 		}
2284 
2285 		switch (fa->id) {
2286 		case FLOW_ACTION_DROP:
2287 			rc = efx_mae_alloc_action_set(efx, act);
2288 			if (rc) {
2289 				NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (drop)");
2290 				goto release;
2291 			}
2292 			list_add_tail(&act->list, &rule->acts.list);
2293 			act = NULL; /* end of the line */
2294 			break;
2295 		case FLOW_ACTION_REDIRECT:
2296 		case FLOW_ACTION_MIRRED:
2297 			save = *act;
2298 
2299 			if (encap_info) {
2300 				struct efx_tc_encap_action *encap;
2301 
2302 				if (!efx_tc_flower_action_order_ok(act,
2303 								   EFX_TC_AO_ENCAP)) {
2304 					rc = -EOPNOTSUPP;
2305 					NL_SET_ERR_MSG_MOD(extack, "Encap action violates action order");
2306 					goto release;
2307 				}
2308 				encap = efx_tc_flower_create_encap_md(
2309 						efx, encap_info, fa->dev, extack);
2310 				if (IS_ERR_OR_NULL(encap)) {
2311 					rc = PTR_ERR(encap);
2312 					if (!rc)
2313 						rc = -EIO; /* arbitrary */
2314 					goto release;
2315 				}
2316 				act->encap_md = encap;
2317 				list_add_tail(&act->encap_user, &encap->users);
2318 				act->dest_mport = encap->dest_mport;
2319 				act->deliver = 1;
2320 				if (act->count && !WARN_ON(!act->count->cnt)) {
2321 					/* This counter is used by an encap
2322 					 * action, which needs a reference back
2323 					 * so it can prod neighbouring whenever
2324 					 * traffic is seen.
2325 					 */
2326 					spin_lock_bh(&act->count->cnt->lock);
2327 					list_add_tail(&act->count_user,
2328 						      &act->count->cnt->users);
2329 					spin_unlock_bh(&act->count->cnt->lock);
2330 				}
2331 				rc = efx_mae_alloc_action_set(efx, act);
2332 				if (rc) {
2333 					NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (encap)");
2334 					goto release;
2335 				}
2336 				list_add_tail(&act->list, &rule->acts.list);
2337 				act->user = &rule->acts;
2338 				act = NULL;
2339 				if (fa->id == FLOW_ACTION_REDIRECT)
2340 					break; /* end of the line */
2341 				/* Mirror, so continue on with saved act */
2342 				save.count = NULL;
2343 				act = kzalloc(sizeof(*act), GFP_USER);
2344 				if (!act) {
2345 					rc = -ENOMEM;
2346 					goto release;
2347 				}
2348 				*act = save;
2349 				break;
2350 			}
2351 
2352 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DELIVER)) {
2353 				/* can't happen */
2354 				rc = -EOPNOTSUPP;
2355 				NL_SET_ERR_MSG_MOD(extack, "Deliver action violates action order (can't happen)");
2356 				goto release;
2357 			}
2358 
2359 			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
2360 			if (IS_ERR(to_efv)) {
2361 				NL_SET_ERR_MSG_MOD(extack, "Mirred egress device not on switch");
2362 				rc = PTR_ERR(to_efv);
2363 				goto release;
2364 			}
2365 			rc = efx_tc_flower_external_mport(efx, to_efv);
2366 			if (rc < 0) {
2367 				NL_SET_ERR_MSG_MOD(extack, "Failed to identify egress m-port");
2368 				goto release;
2369 			}
2370 			act->dest_mport = rc;
2371 			act->deliver = 1;
2372 			rc = efx_mae_alloc_action_set(efx, act);
2373 			if (rc) {
2374 				NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (mirred)");
2375 				goto release;
2376 			}
2377 			list_add_tail(&act->list, &rule->acts.list);
2378 			act = NULL;
2379 			if (fa->id == FLOW_ACTION_REDIRECT)
2380 				break; /* end of the line */
2381 			/* Mirror, so continue on with saved act */
2382 			save.count = NULL;
2383 			act = kzalloc(sizeof(*act), GFP_USER);
2384 			if (!act) {
2385 				rc = -ENOMEM;
2386 				goto release;
2387 			}
2388 			*act = save;
2389 			break;
2390 		case FLOW_ACTION_VLAN_POP:
2391 			if (act->vlan_push) {
2392 				act->vlan_push--;
2393 			} else if (efx_tc_flower_action_order_ok(act, EFX_TC_AO_VLAN_POP)) {
2394 				act->vlan_pop++;
2395 			} else {
2396 				NL_SET_ERR_MSG_MOD(extack,
2397 						   "More than two VLAN pops, or action order violated");
2398 				rc = -EINVAL;
2399 				goto release;
2400 			}
2401 			break;
2402 		case FLOW_ACTION_VLAN_PUSH:
2403 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_VLAN_PUSH)) {
2404 				rc = -EINVAL;
2405 				NL_SET_ERR_MSG_MOD(extack,
2406 						   "More than two VLAN pushes, or action order violated");
2407 				goto release;
2408 			}
2409 			tci = fa->vlan.vid & VLAN_VID_MASK;
2410 			tci |= fa->vlan.prio << VLAN_PRIO_SHIFT;
2411 			act->vlan_tci[act->vlan_push] = cpu_to_be16(tci);
2412 			act->vlan_proto[act->vlan_push] = fa->vlan.proto;
2413 			act->vlan_push++;
2414 			break;
2415 		case FLOW_ACTION_ADD:
2416 			rc = efx_tc_pedit_add(efx, act, fa, extack);
2417 			if (rc < 0)
2418 				goto release;
2419 			break;
2420 		case FLOW_ACTION_MANGLE:
2421 			rc = efx_tc_mangle(efx, act, fa, &mung, extack, &match);
2422 			if (rc < 0)
2423 				goto release;
2424 			break;
2425 		case FLOW_ACTION_TUNNEL_ENCAP:
2426 			if (encap_info) {
2427 				/* Can't specify encap multiple times.
2428 				 * If you want to overwrite an existing
2429 				 * encap_info, use an intervening
2430 				 * FLOW_ACTION_TUNNEL_DECAP to clear it.
2431 				 */
2432 				NL_SET_ERR_MSG_MOD(extack, "Tunnel key set when already set");
2433 				rc = -EINVAL;
2434 				goto release;
2435 			}
2436 			if (!fa->tunnel) {
2437 				NL_SET_ERR_MSG_MOD(extack, "Tunnel key set is missing key");
2438 				rc = -EOPNOTSUPP;
2439 				goto release;
2440 			}
2441 			encap_info = fa->tunnel;
2442 			break;
2443 		case FLOW_ACTION_TUNNEL_DECAP:
2444 			if (encap_info) {
2445 				encap_info = NULL;
2446 				break;
2447 			}
2448 			/* Since we don't support enc_key matches on ingress
2449 			 * (and if we did there'd be no tunnel-device to give
2450 			 * us a type), we can't offload a decap that's not
2451 			 * just undoing a previous encap action.
2452 			 */
2453 			NL_SET_ERR_MSG_MOD(extack, "Cannot offload tunnel decap action without tunnel device");
2454 			rc = -EOPNOTSUPP;
2455 			goto release;
2456 		case FLOW_ACTION_CT:
2457 			if (fa->ct.action != TCA_CT_ACT_NAT) {
2458 				rc = -EOPNOTSUPP;
2459 				NL_SET_ERR_MSG_FMT_MOD(extack, "Can only offload CT 'nat' action in RHS rules, not %d", fa->ct.action);
2460 				goto release;
2461 			}
2462 			act->do_nat = 1;
2463 			break;
2464 		default:
2465 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled action %u",
2466 					       fa->id);
2467 			rc = -EOPNOTSUPP;
2468 			goto release;
2469 		}
2470 	}
2471 
2472 	rc = efx_tc_incomplete_mangle(&mung, extack);
2473 	if (rc < 0)
2474 		goto release;
2475 	if (act) {
2476 		/* Not shot/redirected, so deliver to default dest */
2477 		if (from_efv == EFX_EFV_PF)
2478 			/* Rule applies to traffic from the wire,
2479 			 * and default dest is thus the PF
2480 			 */
2481 			efx_mae_mport_uplink(efx, &act->dest_mport);
2482 		else
2483 			/* Representor, so rule applies to traffic from
2484 			 * representee, and default dest is thus the rep.
2485 			 * All reps use the same mport for delivery
2486 			 */
2487 			efx_mae_mport_mport(efx, efx->tc->reps_mport_id,
2488 					    &act->dest_mport);
2489 		act->deliver = 1;
2490 		rc = efx_mae_alloc_action_set(efx, act);
2491 		if (rc) {
2492 			NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (deliver)");
2493 			goto release;
2494 		}
2495 		list_add_tail(&act->list, &rule->acts.list);
2496 		act = NULL; /* Prevent double-free in error path */
2497 	}
2498 
2499 	netif_dbg(efx, drv, efx->net_dev,
2500 		  "Successfully parsed filter (cookie %lx)\n",
2501 		  tc->cookie);
2502 
2503 	rule->match = match;
2504 
2505 	rc = efx_mae_alloc_action_set_list(efx, &rule->acts);
2506 	if (rc) {
2507 		NL_SET_ERR_MSG_MOD(extack, "Failed to write action set list to hw");
2508 		goto release;
2509 	}
2510 	if (from_efv == EFX_EFV_PF)
2511 		/* PF netdev, so rule applies to traffic from wire */
2512 		rule->fallback = &efx->tc->facts.pf;
2513 	else
2514 		/* repdev, so rule applies to traffic from representee */
2515 		rule->fallback = &efx->tc->facts.reps;
2516 	if (!efx_tc_check_ready(efx, rule)) {
2517 		netif_dbg(efx, drv, efx->net_dev, "action not ready for hw\n");
2518 		acts_id = rule->fallback->fw_id;
2519 	} else {
2520 		netif_dbg(efx, drv, efx->net_dev, "ready for hw\n");
2521 		acts_id = rule->acts.fw_id;
2522 	}
2523 	rc = efx_mae_insert_rule(efx, &rule->match, EFX_TC_PRIO_TC,
2524 				 acts_id, &rule->fw_id);
2525 	if (rc) {
2526 		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
2527 		goto release_acts;
2528 	}
2529 	return 0;
2530 
2531 release_acts:
2532 	efx_mae_free_action_set_list(efx, &rule->acts);
2533 release:
2534 	/* We failed to insert the rule, so free up any entries we created in
2535 	 * subsidiary tables.
2536 	 */
2537 	if (match.rid)
2538 		efx_tc_put_recirc_id(efx, match.rid);
2539 	if (act)
2540 		efx_tc_free_action_set(efx, act, false);
2541 	if (rule) {
2542 		if (!old)
2543 			rhashtable_remove_fast(&efx->tc->match_action_ht,
2544 					       &rule->linkage,
2545 					       efx_tc_match_action_ht_params);
2546 		efx_tc_free_action_set_list(efx, &rule->acts, false);
2547 	}
2548 	kfree(rule);
2549 	return rc;
2550 }
2551 
2552 static int efx_tc_flower_destroy(struct efx_nic *efx,
2553 				 struct net_device *net_dev,
2554 				 struct flow_cls_offload *tc)
2555 {
2556 	struct netlink_ext_ack *extack = tc->common.extack;
2557 	struct efx_tc_lhs_rule *lhs_rule;
2558 	struct efx_tc_flow_rule *rule;
2559 
2560 	lhs_rule = rhashtable_lookup_fast(&efx->tc->lhs_rule_ht, &tc->cookie,
2561 					  efx_tc_lhs_rule_ht_params);
2562 	if (lhs_rule) {
2563 		/* Remove it from HW */
2564 		efx_mae_remove_lhs_rule(efx, lhs_rule);
2565 		/* Delete it from SW */
2566 		efx_tc_flower_release_lhs_actions(efx, &lhs_rule->lhs_act);
2567 		rhashtable_remove_fast(&efx->tc->lhs_rule_ht, &lhs_rule->linkage,
2568 				       efx_tc_lhs_rule_ht_params);
2569 		if (lhs_rule->match.encap)
2570 			efx_tc_flower_release_encap_match(efx, lhs_rule->match.encap);
2571 		netif_dbg(efx, drv, efx->net_dev, "Removed (lhs) filter %lx\n",
2572 			  lhs_rule->cookie);
2573 		kfree(lhs_rule);
2574 		return 0;
2575 	}
2576 
2577 	rule = rhashtable_lookup_fast(&efx->tc->match_action_ht, &tc->cookie,
2578 				      efx_tc_match_action_ht_params);
2579 	if (!rule) {
2580 		/* Only log a message if we're the ingress device.  Otherwise
2581 		 * it's a foreign filter and we might just not have been
2582 		 * interested (e.g. we might not have been the egress device
2583 		 * either).
2584 		 */
2585 		if (!IS_ERR(efx_tc_flower_lookup_efv(efx, net_dev)))
2586 			netif_warn(efx, drv, efx->net_dev,
2587 				   "Filter %lx not found to remove\n", tc->cookie);
2588 		NL_SET_ERR_MSG_MOD(extack, "Flow cookie not found in offloaded rules");
2589 		return -ENOENT;
2590 	}
2591 
2592 	/* Remove it from HW */
2593 	efx_tc_delete_rule(efx, rule);
2594 	/* Delete it from SW */
2595 	rhashtable_remove_fast(&efx->tc->match_action_ht, &rule->linkage,
2596 			       efx_tc_match_action_ht_params);
2597 	netif_dbg(efx, drv, efx->net_dev, "Removed filter %lx\n", rule->cookie);
2598 	kfree(rule);
2599 	return 0;
2600 }
2601 
2602 static int efx_tc_flower_stats(struct efx_nic *efx, struct net_device *net_dev,
2603 			       struct flow_cls_offload *tc)
2604 {
2605 	struct netlink_ext_ack *extack = tc->common.extack;
2606 	struct efx_tc_counter_index *ctr;
2607 	struct efx_tc_counter *cnt;
2608 	u64 packets, bytes;
2609 
2610 	ctr = efx_tc_flower_find_counter_index(efx, tc->cookie);
2611 	if (!ctr) {
2612 		/* See comment in efx_tc_flower_destroy() */
2613 		if (!IS_ERR(efx_tc_flower_lookup_efv(efx, net_dev)))
2614 			if (net_ratelimit())
2615 				netif_warn(efx, drv, efx->net_dev,
2616 					   "Filter %lx not found for stats\n",
2617 					   tc->cookie);
2618 		NL_SET_ERR_MSG_MOD(extack, "Flow cookie not found in offloaded rules");
2619 		return -ENOENT;
2620 	}
2621 	if (WARN_ON(!ctr->cnt)) /* can't happen */
2622 		return -EIO;
2623 	cnt = ctr->cnt;
2624 
2625 	spin_lock_bh(&cnt->lock);
2626 	/* Report only new pkts/bytes since last time TC asked */
2627 	packets = cnt->packets;
2628 	bytes = cnt->bytes;
2629 	flow_stats_update(&tc->stats, bytes - cnt->old_bytes,
2630 			  packets - cnt->old_packets, 0, cnt->touched,
2631 			  FLOW_ACTION_HW_STATS_DELAYED);
2632 	cnt->old_packets = packets;
2633 	cnt->old_bytes = bytes;
2634 	spin_unlock_bh(&cnt->lock);
2635 	return 0;
2636 }
2637 
2638 int efx_tc_flower(struct efx_nic *efx, struct net_device *net_dev,
2639 		  struct flow_cls_offload *tc, struct efx_rep *efv)
2640 {
2641 	int rc;
2642 
2643 	if (!efx->tc)
2644 		return -EOPNOTSUPP;
2645 
2646 	mutex_lock(&efx->tc->mutex);
2647 	switch (tc->command) {
2648 	case FLOW_CLS_REPLACE:
2649 		rc = efx_tc_flower_replace(efx, net_dev, tc, efv);
2650 		break;
2651 	case FLOW_CLS_DESTROY:
2652 		rc = efx_tc_flower_destroy(efx, net_dev, tc);
2653 		break;
2654 	case FLOW_CLS_STATS:
2655 		rc = efx_tc_flower_stats(efx, net_dev, tc);
2656 		break;
2657 	default:
2658 		rc = -EOPNOTSUPP;
2659 		break;
2660 	}
2661 	mutex_unlock(&efx->tc->mutex);
2662 	return rc;
2663 }
2664 
2665 static int efx_tc_configure_default_rule(struct efx_nic *efx, u32 ing_port,
2666 					 u32 eg_port, struct efx_tc_flow_rule *rule)
2667 {
2668 	struct efx_tc_action_set_list *acts = &rule->acts;
2669 	struct efx_tc_match *match = &rule->match;
2670 	struct efx_tc_action_set *act;
2671 	int rc;
2672 
2673 	match->value.ingress_port = ing_port;
2674 	match->mask.ingress_port = ~0;
2675 	act = kzalloc(sizeof(*act), GFP_KERNEL);
2676 	if (!act)
2677 		return -ENOMEM;
2678 	act->deliver = 1;
2679 	act->dest_mport = eg_port;
2680 	rc = efx_mae_alloc_action_set(efx, act);
2681 	if (rc)
2682 		goto fail1;
2683 	EFX_WARN_ON_PARANOID(!list_empty(&acts->list));
2684 	list_add_tail(&act->list, &acts->list);
2685 	rc = efx_mae_alloc_action_set_list(efx, acts);
2686 	if (rc)
2687 		goto fail2;
2688 	rc = efx_mae_insert_rule(efx, match, EFX_TC_PRIO_DFLT,
2689 				 acts->fw_id, &rule->fw_id);
2690 	if (rc)
2691 		goto fail3;
2692 	return 0;
2693 fail3:
2694 	efx_mae_free_action_set_list(efx, acts);
2695 fail2:
2696 	list_del(&act->list);
2697 	efx_mae_free_action_set(efx, act->fw_id);
2698 fail1:
2699 	kfree(act);
2700 	return rc;
2701 }
2702 
2703 static int efx_tc_configure_default_rule_pf(struct efx_nic *efx)
2704 {
2705 	struct efx_tc_flow_rule *rule = &efx->tc->dflt.pf;
2706 	u32 ing_port, eg_port;
2707 
2708 	efx_mae_mport_uplink(efx, &ing_port);
2709 	efx_mae_mport_wire(efx, &eg_port);
2710 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
2711 }
2712 
2713 static int efx_tc_configure_default_rule_wire(struct efx_nic *efx)
2714 {
2715 	struct efx_tc_flow_rule *rule = &efx->tc->dflt.wire;
2716 	u32 ing_port, eg_port;
2717 
2718 	efx_mae_mport_wire(efx, &ing_port);
2719 	efx_mae_mport_uplink(efx, &eg_port);
2720 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
2721 }
2722 
2723 int efx_tc_configure_default_rule_rep(struct efx_rep *efv)
2724 {
2725 	struct efx_tc_flow_rule *rule = &efv->dflt;
2726 	struct efx_nic *efx = efv->parent;
2727 	u32 ing_port, eg_port;
2728 
2729 	efx_mae_mport_mport(efx, efv->mport, &ing_port);
2730 	efx_mae_mport_mport(efx, efx->tc->reps_mport_id, &eg_port);
2731 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
2732 }
2733 
2734 void efx_tc_deconfigure_default_rule(struct efx_nic *efx,
2735 				     struct efx_tc_flow_rule *rule)
2736 {
2737 	if (rule->fw_id != MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL)
2738 		efx_tc_delete_rule(efx, rule);
2739 	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
2740 }
2741 
2742 static int efx_tc_configure_fallback_acts(struct efx_nic *efx, u32 eg_port,
2743 					  struct efx_tc_action_set_list *acts)
2744 {
2745 	struct efx_tc_action_set *act;
2746 	int rc;
2747 
2748 	act = kzalloc(sizeof(*act), GFP_KERNEL);
2749 	if (!act)
2750 		return -ENOMEM;
2751 	act->deliver = 1;
2752 	act->dest_mport = eg_port;
2753 	rc = efx_mae_alloc_action_set(efx, act);
2754 	if (rc)
2755 		goto fail1;
2756 	EFX_WARN_ON_PARANOID(!list_empty(&acts->list));
2757 	list_add_tail(&act->list, &acts->list);
2758 	rc = efx_mae_alloc_action_set_list(efx, acts);
2759 	if (rc)
2760 		goto fail2;
2761 	return 0;
2762 fail2:
2763 	list_del(&act->list);
2764 	efx_mae_free_action_set(efx, act->fw_id);
2765 fail1:
2766 	kfree(act);
2767 	return rc;
2768 }
2769 
2770 static int efx_tc_configure_fallback_acts_pf(struct efx_nic *efx)
2771 {
2772 	struct efx_tc_action_set_list *acts = &efx->tc->facts.pf;
2773 	u32 eg_port;
2774 
2775 	efx_mae_mport_uplink(efx, &eg_port);
2776 	return efx_tc_configure_fallback_acts(efx, eg_port, acts);
2777 }
2778 
2779 static int efx_tc_configure_fallback_acts_reps(struct efx_nic *efx)
2780 {
2781 	struct efx_tc_action_set_list *acts = &efx->tc->facts.reps;
2782 	u32 eg_port;
2783 
2784 	efx_mae_mport_mport(efx, efx->tc->reps_mport_id, &eg_port);
2785 	return efx_tc_configure_fallback_acts(efx, eg_port, acts);
2786 }
2787 
2788 static void efx_tc_deconfigure_fallback_acts(struct efx_nic *efx,
2789 					     struct efx_tc_action_set_list *acts)
2790 {
2791 	efx_tc_free_action_set_list(efx, acts, true);
2792 }
2793 
2794 static int efx_tc_configure_rep_mport(struct efx_nic *efx)
2795 {
2796 	u32 rep_mport_label;
2797 	int rc;
2798 
2799 	rc = efx_mae_allocate_mport(efx, &efx->tc->reps_mport_id, &rep_mport_label);
2800 	if (rc)
2801 		return rc;
2802 	pci_dbg(efx->pci_dev, "created rep mport 0x%08x (0x%04x)\n",
2803 		efx->tc->reps_mport_id, rep_mport_label);
2804 	/* Use mport *selector* as vport ID */
2805 	efx_mae_mport_mport(efx, efx->tc->reps_mport_id,
2806 			    &efx->tc->reps_mport_vport_id);
2807 	return 0;
2808 }
2809 
2810 static void efx_tc_deconfigure_rep_mport(struct efx_nic *efx)
2811 {
2812 	efx_mae_free_mport(efx, efx->tc->reps_mport_id);
2813 	efx->tc->reps_mport_id = MAE_MPORT_SELECTOR_NULL;
2814 }
2815 
2816 int efx_tc_insert_rep_filters(struct efx_nic *efx)
2817 {
2818 	struct efx_filter_spec promisc, allmulti;
2819 	int rc;
2820 
2821 	if (efx->type->is_vf)
2822 		return 0;
2823 	if (!efx->tc)
2824 		return 0;
2825 	efx_filter_init_rx(&promisc, EFX_FILTER_PRI_REQUIRED, 0, 0);
2826 	efx_filter_set_uc_def(&promisc);
2827 	efx_filter_set_vport_id(&promisc, efx->tc->reps_mport_vport_id);
2828 	rc = efx_filter_insert_filter(efx, &promisc, false);
2829 	if (rc < 0)
2830 		return rc;
2831 	efx->tc->reps_filter_uc = rc;
2832 	efx_filter_init_rx(&allmulti, EFX_FILTER_PRI_REQUIRED, 0, 0);
2833 	efx_filter_set_mc_def(&allmulti);
2834 	efx_filter_set_vport_id(&allmulti, efx->tc->reps_mport_vport_id);
2835 	rc = efx_filter_insert_filter(efx, &allmulti, false);
2836 	if (rc < 0)
2837 		return rc;
2838 	efx->tc->reps_filter_mc = rc;
2839 	return 0;
2840 }
2841 
2842 void efx_tc_remove_rep_filters(struct efx_nic *efx)
2843 {
2844 	if (efx->type->is_vf)
2845 		return;
2846 	if (!efx->tc)
2847 		return;
2848 	if (efx->tc->reps_filter_mc >= 0)
2849 		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED, efx->tc->reps_filter_mc);
2850 	efx->tc->reps_filter_mc = -1;
2851 	if (efx->tc->reps_filter_uc >= 0)
2852 		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED, efx->tc->reps_filter_uc);
2853 	efx->tc->reps_filter_uc = -1;
2854 }
2855 
2856 int efx_init_tc(struct efx_nic *efx)
2857 {
2858 	int rc;
2859 
2860 	rc = efx_mae_get_caps(efx, efx->tc->caps);
2861 	if (rc)
2862 		return rc;
2863 	if (efx->tc->caps->match_field_count > MAE_NUM_FIELDS)
2864 		/* Firmware supports some match fields the driver doesn't know
2865 		 * about.  Not fatal, unless any of those fields are required
2866 		 * (MAE_FIELD_SUPPORTED_MATCH_ALWAYS) but if so we don't know.
2867 		 */
2868 		netif_warn(efx, probe, efx->net_dev,
2869 			   "FW reports additional match fields %u\n",
2870 			   efx->tc->caps->match_field_count);
2871 	if (efx->tc->caps->action_prios < EFX_TC_PRIO__NUM) {
2872 		netif_err(efx, probe, efx->net_dev,
2873 			  "Too few action prios supported (have %u, need %u)\n",
2874 			  efx->tc->caps->action_prios, EFX_TC_PRIO__NUM);
2875 		return -EIO;
2876 	}
2877 	rc = efx_tc_configure_default_rule_pf(efx);
2878 	if (rc)
2879 		return rc;
2880 	rc = efx_tc_configure_default_rule_wire(efx);
2881 	if (rc)
2882 		return rc;
2883 	rc = efx_tc_configure_rep_mport(efx);
2884 	if (rc)
2885 		return rc;
2886 	rc = efx_tc_configure_fallback_acts_pf(efx);
2887 	if (rc)
2888 		return rc;
2889 	rc = efx_tc_configure_fallback_acts_reps(efx);
2890 	if (rc)
2891 		return rc;
2892 	rc = efx_mae_get_tables(efx);
2893 	if (rc)
2894 		return rc;
2895 	rc = flow_indr_dev_register(efx_tc_indr_setup_cb, efx);
2896 	if (rc)
2897 		goto out_free;
2898 	efx->tc->up = true;
2899 	return 0;
2900 out_free:
2901 	efx_mae_free_tables(efx);
2902 	return rc;
2903 }
2904 
2905 void efx_fini_tc(struct efx_nic *efx)
2906 {
2907 	/* We can get called even if efx_init_struct_tc() failed */
2908 	if (!efx->tc)
2909 		return;
2910 	if (efx->tc->up)
2911 		flow_indr_dev_unregister(efx_tc_indr_setup_cb, efx, efx_tc_block_unbind);
2912 	efx_tc_deconfigure_rep_mport(efx);
2913 	efx_tc_deconfigure_default_rule(efx, &efx->tc->dflt.pf);
2914 	efx_tc_deconfigure_default_rule(efx, &efx->tc->dflt.wire);
2915 	efx_tc_deconfigure_fallback_acts(efx, &efx->tc->facts.pf);
2916 	efx_tc_deconfigure_fallback_acts(efx, &efx->tc->facts.reps);
2917 	efx->tc->up = false;
2918 	efx_mae_free_tables(efx);
2919 }
2920 
2921 /* At teardown time, all TC filter rules (and thus all resources they created)
2922  * should already have been removed.  If we find any in our hashtables, make a
2923  * cursory attempt to clean up the software side.
2924  */
2925 static void efx_tc_encap_match_free(void *ptr, void *__unused)
2926 {
2927 	struct efx_tc_encap_match *encap = ptr;
2928 
2929 	WARN_ON(refcount_read(&encap->ref));
2930 	kfree(encap);
2931 }
2932 
2933 static void efx_tc_recirc_free(void *ptr, void *arg)
2934 {
2935 	struct efx_tc_recirc_id *rid = ptr;
2936 	struct efx_nic *efx = arg;
2937 
2938 	WARN_ON(refcount_read(&rid->ref));
2939 	ida_free(&efx->tc->recirc_ida, rid->fw_id);
2940 	kfree(rid);
2941 }
2942 
2943 static void efx_tc_lhs_free(void *ptr, void *arg)
2944 {
2945 	struct efx_tc_lhs_rule *rule = ptr;
2946 	struct efx_nic *efx = arg;
2947 
2948 	netif_err(efx, drv, efx->net_dev,
2949 		  "tc lhs_rule %lx still present at teardown, removing\n",
2950 		  rule->cookie);
2951 
2952 	if (rule->lhs_act.zone)
2953 		efx_tc_ct_unregister_zone(efx, rule->lhs_act.zone);
2954 	if (rule->lhs_act.count)
2955 		efx_tc_flower_put_counter_index(efx, rule->lhs_act.count);
2956 	efx_mae_remove_lhs_rule(efx, rule);
2957 
2958 	kfree(rule);
2959 }
2960 
2961 static void efx_tc_mac_free(void *ptr, void *__unused)
2962 {
2963 	struct efx_tc_mac_pedit_action *ped = ptr;
2964 
2965 	WARN_ON(refcount_read(&ped->ref));
2966 	kfree(ped);
2967 }
2968 
2969 static void efx_tc_flow_free(void *ptr, void *arg)
2970 {
2971 	struct efx_tc_flow_rule *rule = ptr;
2972 	struct efx_nic *efx = arg;
2973 
2974 	netif_err(efx, drv, efx->net_dev,
2975 		  "tc rule %lx still present at teardown, removing\n",
2976 		  rule->cookie);
2977 
2978 	/* Also releases entries in subsidiary tables */
2979 	efx_tc_delete_rule(efx, rule);
2980 
2981 	kfree(rule);
2982 }
2983 
2984 int efx_init_struct_tc(struct efx_nic *efx)
2985 {
2986 	int rc;
2987 
2988 	if (efx->type->is_vf)
2989 		return 0;
2990 
2991 	efx->tc = kzalloc(sizeof(*efx->tc), GFP_KERNEL);
2992 	if (!efx->tc)
2993 		return -ENOMEM;
2994 	efx->tc->caps = kzalloc(sizeof(struct mae_caps), GFP_KERNEL);
2995 	if (!efx->tc->caps) {
2996 		rc = -ENOMEM;
2997 		goto fail_alloc_caps;
2998 	}
2999 	INIT_LIST_HEAD(&efx->tc->block_list);
3000 
3001 	mutex_init(&efx->tc->mutex);
3002 	init_waitqueue_head(&efx->tc->flush_wq);
3003 	rc = efx_tc_init_encap_actions(efx);
3004 	if (rc < 0)
3005 		goto fail_encap_actions;
3006 	rc = efx_tc_init_counters(efx);
3007 	if (rc < 0)
3008 		goto fail_counters;
3009 	rc = rhashtable_init(&efx->tc->mac_ht, &efx_tc_mac_ht_params);
3010 	if (rc < 0)
3011 		goto fail_mac_ht;
3012 	rc = rhashtable_init(&efx->tc->encap_match_ht, &efx_tc_encap_match_ht_params);
3013 	if (rc < 0)
3014 		goto fail_encap_match_ht;
3015 	rc = rhashtable_init(&efx->tc->match_action_ht, &efx_tc_match_action_ht_params);
3016 	if (rc < 0)
3017 		goto fail_match_action_ht;
3018 	rc = rhashtable_init(&efx->tc->lhs_rule_ht, &efx_tc_lhs_rule_ht_params);
3019 	if (rc < 0)
3020 		goto fail_lhs_rule_ht;
3021 	rc = efx_tc_init_conntrack(efx);
3022 	if (rc < 0)
3023 		goto fail_conntrack;
3024 	rc = rhashtable_init(&efx->tc->recirc_ht, &efx_tc_recirc_ht_params);
3025 	if (rc < 0)
3026 		goto fail_recirc_ht;
3027 	ida_init(&efx->tc->recirc_ida);
3028 	efx->tc->reps_filter_uc = -1;
3029 	efx->tc->reps_filter_mc = -1;
3030 	INIT_LIST_HEAD(&efx->tc->dflt.pf.acts.list);
3031 	efx->tc->dflt.pf.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
3032 	INIT_LIST_HEAD(&efx->tc->dflt.wire.acts.list);
3033 	efx->tc->dflt.wire.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
3034 	INIT_LIST_HEAD(&efx->tc->facts.pf.list);
3035 	efx->tc->facts.pf.fw_id = MC_CMD_MAE_ACTION_SET_ALLOC_OUT_ACTION_SET_ID_NULL;
3036 	INIT_LIST_HEAD(&efx->tc->facts.reps.list);
3037 	efx->tc->facts.reps.fw_id = MC_CMD_MAE_ACTION_SET_ALLOC_OUT_ACTION_SET_ID_NULL;
3038 	efx->extra_channel_type[EFX_EXTRA_CHANNEL_TC] = &efx_tc_channel_type;
3039 	return 0;
3040 fail_recirc_ht:
3041 	efx_tc_destroy_conntrack(efx);
3042 fail_conntrack:
3043 	rhashtable_destroy(&efx->tc->lhs_rule_ht);
3044 fail_lhs_rule_ht:
3045 	rhashtable_destroy(&efx->tc->match_action_ht);
3046 fail_match_action_ht:
3047 	rhashtable_destroy(&efx->tc->encap_match_ht);
3048 fail_encap_match_ht:
3049 	rhashtable_destroy(&efx->tc->mac_ht);
3050 fail_mac_ht:
3051 	efx_tc_destroy_counters(efx);
3052 fail_counters:
3053 	efx_tc_destroy_encap_actions(efx);
3054 fail_encap_actions:
3055 	mutex_destroy(&efx->tc->mutex);
3056 	kfree(efx->tc->caps);
3057 fail_alloc_caps:
3058 	kfree(efx->tc);
3059 	efx->tc = NULL;
3060 	return rc;
3061 }
3062 
3063 void efx_fini_struct_tc(struct efx_nic *efx)
3064 {
3065 	if (!efx->tc)
3066 		return;
3067 
3068 	mutex_lock(&efx->tc->mutex);
3069 	EFX_WARN_ON_PARANOID(efx->tc->dflt.pf.fw_id !=
3070 			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
3071 	EFX_WARN_ON_PARANOID(efx->tc->dflt.wire.fw_id !=
3072 			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
3073 	EFX_WARN_ON_PARANOID(efx->tc->facts.pf.fw_id !=
3074 			     MC_CMD_MAE_ACTION_SET_LIST_ALLOC_OUT_ACTION_SET_LIST_ID_NULL);
3075 	EFX_WARN_ON_PARANOID(efx->tc->facts.reps.fw_id !=
3076 			     MC_CMD_MAE_ACTION_SET_LIST_ALLOC_OUT_ACTION_SET_LIST_ID_NULL);
3077 	rhashtable_free_and_destroy(&efx->tc->lhs_rule_ht, efx_tc_lhs_free, efx);
3078 	rhashtable_free_and_destroy(&efx->tc->match_action_ht, efx_tc_flow_free,
3079 				    efx);
3080 	rhashtable_free_and_destroy(&efx->tc->encap_match_ht,
3081 				    efx_tc_encap_match_free, NULL);
3082 	efx_tc_fini_conntrack(efx);
3083 	rhashtable_free_and_destroy(&efx->tc->recirc_ht, efx_tc_recirc_free, efx);
3084 	WARN_ON(!ida_is_empty(&efx->tc->recirc_ida));
3085 	ida_destroy(&efx->tc->recirc_ida);
3086 	rhashtable_free_and_destroy(&efx->tc->mac_ht, efx_tc_mac_free, NULL);
3087 	efx_tc_fini_counters(efx);
3088 	efx_tc_fini_encap_actions(efx);
3089 	mutex_unlock(&efx->tc->mutex);
3090 	mutex_destroy(&efx->tc->mutex);
3091 	kfree(efx->tc->caps);
3092 	kfree(efx->tc);
3093 	efx->tc = NULL;
3094 }
3095