xref: /linux/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_flower.c (revision d53b8e36925256097a08d7cb749198d85cbf9b2b)
1 /*
2  * This file is part of the Chelsio T4/T5/T6 Ethernet driver for Linux.
3  *
4  * Copyright (c) 2017 Chelsio Communications, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include <net/tc_act/tc_mirred.h>
36 #include <net/tc_act/tc_pedit.h>
37 #include <net/tc_act/tc_gact.h>
38 #include <net/tc_act/tc_vlan.h>
39 
40 #include "cxgb4.h"
41 #include "cxgb4_filter.h"
42 #include "cxgb4_tc_flower.h"
43 
44 #define STATS_CHECK_PERIOD (HZ / 2)
45 
46 static struct ch_tc_pedit_fields pedits[] = {
47 	PEDIT_FIELDS(ETH_, DMAC_31_0, 4, dmac, 0),
48 	PEDIT_FIELDS(ETH_, DMAC_47_32, 2, dmac, 4),
49 	PEDIT_FIELDS(ETH_, SMAC_15_0, 2, smac, 0),
50 	PEDIT_FIELDS(ETH_, SMAC_47_16, 4, smac, 2),
51 	PEDIT_FIELDS(IP4_, SRC, 4, nat_fip, 0),
52 	PEDIT_FIELDS(IP4_, DST, 4, nat_lip, 0),
53 	PEDIT_FIELDS(IP6_, SRC_31_0, 4, nat_fip, 0),
54 	PEDIT_FIELDS(IP6_, SRC_63_32, 4, nat_fip, 4),
55 	PEDIT_FIELDS(IP6_, SRC_95_64, 4, nat_fip, 8),
56 	PEDIT_FIELDS(IP6_, SRC_127_96, 4, nat_fip, 12),
57 	PEDIT_FIELDS(IP6_, DST_31_0, 4, nat_lip, 0),
58 	PEDIT_FIELDS(IP6_, DST_63_32, 4, nat_lip, 4),
59 	PEDIT_FIELDS(IP6_, DST_95_64, 4, nat_lip, 8),
60 	PEDIT_FIELDS(IP6_, DST_127_96, 4, nat_lip, 12),
61 };
62 
63 static const struct cxgb4_natmode_config cxgb4_natmode_config_array[] = {
64 	/* Default supported NAT modes */
65 	{
66 		.chip = CHELSIO_T5,
67 		.flags = CXGB4_ACTION_NATMODE_NONE,
68 		.natmode = NAT_MODE_NONE,
69 	},
70 	{
71 		.chip = CHELSIO_T5,
72 		.flags = CXGB4_ACTION_NATMODE_DIP,
73 		.natmode = NAT_MODE_DIP,
74 	},
75 	{
76 		.chip = CHELSIO_T5,
77 		.flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_DPORT,
78 		.natmode = NAT_MODE_DIP_DP,
79 	},
80 	{
81 		.chip = CHELSIO_T5,
82 		.flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_DPORT |
83 			 CXGB4_ACTION_NATMODE_SIP,
84 		.natmode = NAT_MODE_DIP_DP_SIP,
85 	},
86 	{
87 		.chip = CHELSIO_T5,
88 		.flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_DPORT |
89 			 CXGB4_ACTION_NATMODE_SPORT,
90 		.natmode = NAT_MODE_DIP_DP_SP,
91 	},
92 	{
93 		.chip = CHELSIO_T5,
94 		.flags = CXGB4_ACTION_NATMODE_SIP | CXGB4_ACTION_NATMODE_SPORT,
95 		.natmode = NAT_MODE_SIP_SP,
96 	},
97 	{
98 		.chip = CHELSIO_T5,
99 		.flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_SIP |
100 			 CXGB4_ACTION_NATMODE_SPORT,
101 		.natmode = NAT_MODE_DIP_SIP_SP,
102 	},
103 	{
104 		.chip = CHELSIO_T5,
105 		.flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_SIP |
106 			 CXGB4_ACTION_NATMODE_DPORT |
107 			 CXGB4_ACTION_NATMODE_SPORT,
108 		.natmode = NAT_MODE_ALL,
109 	},
110 	/* T6+ can ignore L4 ports when they're disabled. */
111 	{
112 		.chip = CHELSIO_T6,
113 		.flags = CXGB4_ACTION_NATMODE_SIP,
114 		.natmode = NAT_MODE_SIP_SP,
115 	},
116 	{
117 		.chip = CHELSIO_T6,
118 		.flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_SPORT,
119 		.natmode = NAT_MODE_DIP_DP_SP,
120 	},
121 	{
122 		.chip = CHELSIO_T6,
123 		.flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_SIP,
124 		.natmode = NAT_MODE_ALL,
125 	},
126 };
127 
128 static void cxgb4_action_natmode_tweak(struct ch_filter_specification *fs,
129 				       u8 natmode_flags)
130 {
131 	u8 i = 0;
132 
133 	/* Translate the enabled NAT 4-tuple fields to one of the
134 	 * hardware supported NAT mode configurations. This ensures
135 	 * that we pick a valid combination, where the disabled fields
136 	 * do not get overwritten to 0.
137 	 */
138 	for (i = 0; i < ARRAY_SIZE(cxgb4_natmode_config_array); i++) {
139 		if (cxgb4_natmode_config_array[i].flags == natmode_flags) {
140 			fs->nat_mode = cxgb4_natmode_config_array[i].natmode;
141 			return;
142 		}
143 	}
144 }
145 
146 static struct ch_tc_flower_entry *allocate_flower_entry(void)
147 {
148 	struct ch_tc_flower_entry *new = kzalloc(sizeof(*new), GFP_KERNEL);
149 	if (new)
150 		spin_lock_init(&new->lock);
151 	return new;
152 }
153 
154 /* Must be called with either RTNL or rcu_read_lock */
155 static struct ch_tc_flower_entry *ch_flower_lookup(struct adapter *adap,
156 						   unsigned long flower_cookie)
157 {
158 	return rhashtable_lookup_fast(&adap->flower_tbl, &flower_cookie,
159 				      adap->flower_ht_params);
160 }
161 
162 static void cxgb4_process_flow_match(struct net_device *dev,
163 				     struct flow_rule *rule,
164 				     struct ch_filter_specification *fs)
165 {
166 	u16 addr_type = 0;
167 
168 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
169 		struct flow_match_control match;
170 
171 		flow_rule_match_control(rule, &match);
172 		addr_type = match.key->addr_type;
173 	} else if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV4_ADDRS)) {
174 		addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
175 	} else if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV6_ADDRS)) {
176 		addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
177 	}
178 
179 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
180 		struct flow_match_basic match;
181 		u16 ethtype_key, ethtype_mask;
182 
183 		flow_rule_match_basic(rule, &match);
184 		ethtype_key = ntohs(match.key->n_proto);
185 		ethtype_mask = ntohs(match.mask->n_proto);
186 
187 		if (ethtype_key == ETH_P_ALL) {
188 			ethtype_key = 0;
189 			ethtype_mask = 0;
190 		}
191 
192 		if (ethtype_key == ETH_P_IPV6)
193 			fs->type = 1;
194 
195 		fs->val.ethtype = ethtype_key;
196 		fs->mask.ethtype = ethtype_mask;
197 		fs->val.proto = match.key->ip_proto;
198 		fs->mask.proto = match.mask->ip_proto;
199 	}
200 
201 	if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
202 		struct flow_match_ipv4_addrs match;
203 
204 		flow_rule_match_ipv4_addrs(rule, &match);
205 		fs->type = 0;
206 		memcpy(&fs->val.lip[0], &match.key->dst, sizeof(match.key->dst));
207 		memcpy(&fs->val.fip[0], &match.key->src, sizeof(match.key->src));
208 		memcpy(&fs->mask.lip[0], &match.mask->dst, sizeof(match.mask->dst));
209 		memcpy(&fs->mask.fip[0], &match.mask->src, sizeof(match.mask->src));
210 
211 		/* also initialize nat_lip/fip to same values */
212 		memcpy(&fs->nat_lip[0], &match.key->dst, sizeof(match.key->dst));
213 		memcpy(&fs->nat_fip[0], &match.key->src, sizeof(match.key->src));
214 	}
215 
216 	if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
217 		struct flow_match_ipv6_addrs match;
218 
219 		flow_rule_match_ipv6_addrs(rule, &match);
220 		fs->type = 1;
221 		memcpy(&fs->val.lip[0], match.key->dst.s6_addr,
222 		       sizeof(match.key->dst));
223 		memcpy(&fs->val.fip[0], match.key->src.s6_addr,
224 		       sizeof(match.key->src));
225 		memcpy(&fs->mask.lip[0], match.mask->dst.s6_addr,
226 		       sizeof(match.mask->dst));
227 		memcpy(&fs->mask.fip[0], match.mask->src.s6_addr,
228 		       sizeof(match.mask->src));
229 
230 		/* also initialize nat_lip/fip to same values */
231 		memcpy(&fs->nat_lip[0], match.key->dst.s6_addr,
232 		       sizeof(match.key->dst));
233 		memcpy(&fs->nat_fip[0], match.key->src.s6_addr,
234 		       sizeof(match.key->src));
235 	}
236 
237 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
238 		struct flow_match_ports match;
239 
240 		flow_rule_match_ports(rule, &match);
241 		fs->val.lport = be16_to_cpu(match.key->dst);
242 		fs->mask.lport = be16_to_cpu(match.mask->dst);
243 		fs->val.fport = be16_to_cpu(match.key->src);
244 		fs->mask.fport = be16_to_cpu(match.mask->src);
245 
246 		/* also initialize nat_lport/fport to same values */
247 		fs->nat_lport = fs->val.lport;
248 		fs->nat_fport = fs->val.fport;
249 	}
250 
251 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
252 		struct flow_match_ip match;
253 
254 		flow_rule_match_ip(rule, &match);
255 		fs->val.tos = match.key->tos;
256 		fs->mask.tos = match.mask->tos;
257 	}
258 
259 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
260 		struct flow_match_enc_keyid match;
261 
262 		flow_rule_match_enc_keyid(rule, &match);
263 		fs->val.vni = be32_to_cpu(match.key->keyid);
264 		fs->mask.vni = be32_to_cpu(match.mask->keyid);
265 		if (fs->mask.vni) {
266 			fs->val.encap_vld = 1;
267 			fs->mask.encap_vld = 1;
268 		}
269 	}
270 
271 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
272 		struct flow_match_vlan match;
273 		u16 vlan_tci, vlan_tci_mask;
274 
275 		flow_rule_match_vlan(rule, &match);
276 		vlan_tci = match.key->vlan_id | (match.key->vlan_priority <<
277 					       VLAN_PRIO_SHIFT);
278 		vlan_tci_mask = match.mask->vlan_id | (match.mask->vlan_priority <<
279 						     VLAN_PRIO_SHIFT);
280 		fs->val.ivlan = vlan_tci;
281 		fs->mask.ivlan = vlan_tci_mask;
282 
283 		fs->val.ivlan_vld = 1;
284 		fs->mask.ivlan_vld = 1;
285 
286 		/* Chelsio adapters use ivlan_vld bit to match vlan packets
287 		 * as 802.1Q. Also, when vlan tag is present in packets,
288 		 * ethtype match is used then to match on ethtype of inner
289 		 * header ie. the header following the vlan header.
290 		 * So, set the ivlan_vld based on ethtype info supplied by
291 		 * TC for vlan packets if its 802.1Q. And then reset the
292 		 * ethtype value else, hw will try to match the supplied
293 		 * ethtype value with ethtype of inner header.
294 		 */
295 		if (fs->val.ethtype == ETH_P_8021Q) {
296 			fs->val.ethtype = 0;
297 			fs->mask.ethtype = 0;
298 		}
299 	}
300 
301 	/* Match only packets coming from the ingress port where this
302 	 * filter will be created.
303 	 */
304 	fs->val.iport = netdev2pinfo(dev)->port_id;
305 	fs->mask.iport = ~0;
306 }
307 
308 static int cxgb4_validate_flow_match(struct netlink_ext_ack *extack,
309 				     struct flow_rule *rule)
310 {
311 	struct flow_dissector *dissector = rule->match.dissector;
312 	u16 ethtype_mask = 0;
313 	u16 ethtype_key = 0;
314 
315 	if (dissector->used_keys &
316 	    ~(BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) |
317 	      BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) |
318 	      BIT_ULL(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
319 	      BIT_ULL(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
320 	      BIT_ULL(FLOW_DISSECTOR_KEY_PORTS) |
321 	      BIT_ULL(FLOW_DISSECTOR_KEY_ENC_KEYID) |
322 	      BIT_ULL(FLOW_DISSECTOR_KEY_VLAN) |
323 	      BIT_ULL(FLOW_DISSECTOR_KEY_IP))) {
324 		NL_SET_ERR_MSG_FMT_MOD(extack,
325 				       "Unsupported key used: 0x%llx",
326 				       dissector->used_keys);
327 		return -EOPNOTSUPP;
328 	}
329 
330 	if (flow_rule_match_has_control_flags(rule, extack))
331 		return -EOPNOTSUPP;
332 
333 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
334 		struct flow_match_basic match;
335 
336 		flow_rule_match_basic(rule, &match);
337 		ethtype_key = ntohs(match.key->n_proto);
338 		ethtype_mask = ntohs(match.mask->n_proto);
339 	}
340 
341 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
342 		u16 eth_ip_type = ethtype_key & ethtype_mask;
343 		struct flow_match_ip match;
344 
345 		if (eth_ip_type != ETH_P_IP && eth_ip_type != ETH_P_IPV6) {
346 			NL_SET_ERR_MSG_MOD(extack,
347 					   "IP Key supported only with IPv4/v6");
348 			return -EINVAL;
349 		}
350 
351 		flow_rule_match_ip(rule, &match);
352 		if (match.mask->ttl) {
353 			NL_SET_ERR_MSG_MOD(extack,
354 					   "ttl match unsupported for offload");
355 			return -EOPNOTSUPP;
356 		}
357 	}
358 
359 	return 0;
360 }
361 
362 static void offload_pedit(struct ch_filter_specification *fs, u32 val, u32 mask,
363 			  u8 field)
364 {
365 	u32 set_val = val & ~mask;
366 	u32 offset = 0;
367 	u8 size = 1;
368 	int i;
369 
370 	for (i = 0; i < ARRAY_SIZE(pedits); i++) {
371 		if (pedits[i].field == field) {
372 			offset = pedits[i].offset;
373 			size = pedits[i].size;
374 			break;
375 		}
376 	}
377 	memcpy((u8 *)fs + offset, &set_val, size);
378 }
379 
380 static void process_pedit_field(struct ch_filter_specification *fs, u32 val,
381 				u32 mask, u32 offset, u8 htype,
382 				u8 *natmode_flags)
383 {
384 	switch (htype) {
385 	case FLOW_ACT_MANGLE_HDR_TYPE_ETH:
386 		switch (offset) {
387 		case PEDIT_ETH_DMAC_31_0:
388 			fs->newdmac = 1;
389 			offload_pedit(fs, val, mask, ETH_DMAC_31_0);
390 			break;
391 		case PEDIT_ETH_DMAC_47_32_SMAC_15_0:
392 			if (~mask & PEDIT_ETH_DMAC_MASK)
393 				offload_pedit(fs, val, mask, ETH_DMAC_47_32);
394 			else
395 				offload_pedit(fs, val >> 16, mask >> 16,
396 					      ETH_SMAC_15_0);
397 			break;
398 		case PEDIT_ETH_SMAC_47_16:
399 			fs->newsmac = 1;
400 			offload_pedit(fs, val, mask, ETH_SMAC_47_16);
401 		}
402 		break;
403 	case FLOW_ACT_MANGLE_HDR_TYPE_IP4:
404 		switch (offset) {
405 		case PEDIT_IP4_SRC:
406 			offload_pedit(fs, val, mask, IP4_SRC);
407 			*natmode_flags |= CXGB4_ACTION_NATMODE_SIP;
408 			break;
409 		case PEDIT_IP4_DST:
410 			offload_pedit(fs, val, mask, IP4_DST);
411 			*natmode_flags |= CXGB4_ACTION_NATMODE_DIP;
412 		}
413 		break;
414 	case FLOW_ACT_MANGLE_HDR_TYPE_IP6:
415 		switch (offset) {
416 		case PEDIT_IP6_SRC_31_0:
417 			offload_pedit(fs, val, mask, IP6_SRC_31_0);
418 			*natmode_flags |= CXGB4_ACTION_NATMODE_SIP;
419 			break;
420 		case PEDIT_IP6_SRC_63_32:
421 			offload_pedit(fs, val, mask, IP6_SRC_63_32);
422 			*natmode_flags |=  CXGB4_ACTION_NATMODE_SIP;
423 			break;
424 		case PEDIT_IP6_SRC_95_64:
425 			offload_pedit(fs, val, mask, IP6_SRC_95_64);
426 			*natmode_flags |= CXGB4_ACTION_NATMODE_SIP;
427 			break;
428 		case PEDIT_IP6_SRC_127_96:
429 			offload_pedit(fs, val, mask, IP6_SRC_127_96);
430 			*natmode_flags |= CXGB4_ACTION_NATMODE_SIP;
431 			break;
432 		case PEDIT_IP6_DST_31_0:
433 			offload_pedit(fs, val, mask, IP6_DST_31_0);
434 			*natmode_flags |= CXGB4_ACTION_NATMODE_DIP;
435 			break;
436 		case PEDIT_IP6_DST_63_32:
437 			offload_pedit(fs, val, mask, IP6_DST_63_32);
438 			*natmode_flags |= CXGB4_ACTION_NATMODE_DIP;
439 			break;
440 		case PEDIT_IP6_DST_95_64:
441 			offload_pedit(fs, val, mask, IP6_DST_95_64);
442 			*natmode_flags |= CXGB4_ACTION_NATMODE_DIP;
443 			break;
444 		case PEDIT_IP6_DST_127_96:
445 			offload_pedit(fs, val, mask, IP6_DST_127_96);
446 			*natmode_flags |= CXGB4_ACTION_NATMODE_DIP;
447 		}
448 		break;
449 	case FLOW_ACT_MANGLE_HDR_TYPE_TCP:
450 		switch (offset) {
451 		case PEDIT_TCP_SPORT_DPORT:
452 			if (~mask & PEDIT_TCP_UDP_SPORT_MASK) {
453 				fs->nat_fport = val;
454 				*natmode_flags |= CXGB4_ACTION_NATMODE_SPORT;
455 			} else {
456 				fs->nat_lport = val >> 16;
457 				*natmode_flags |= CXGB4_ACTION_NATMODE_DPORT;
458 			}
459 		}
460 		break;
461 	case FLOW_ACT_MANGLE_HDR_TYPE_UDP:
462 		switch (offset) {
463 		case PEDIT_UDP_SPORT_DPORT:
464 			if (~mask & PEDIT_TCP_UDP_SPORT_MASK) {
465 				fs->nat_fport = val;
466 				*natmode_flags |= CXGB4_ACTION_NATMODE_SPORT;
467 			} else {
468 				fs->nat_lport = val >> 16;
469 				*natmode_flags |= CXGB4_ACTION_NATMODE_DPORT;
470 			}
471 		}
472 		break;
473 	}
474 }
475 
476 static int cxgb4_action_natmode_validate(struct adapter *adap, u8 natmode_flags,
477 					 struct netlink_ext_ack *extack)
478 {
479 	u8 i = 0;
480 
481 	/* Extract the NAT mode to enable based on what 4-tuple fields
482 	 * are enabled to be overwritten. This ensures that the
483 	 * disabled fields don't get overwritten to 0.
484 	 */
485 	for (i = 0; i < ARRAY_SIZE(cxgb4_natmode_config_array); i++) {
486 		const struct cxgb4_natmode_config *c;
487 
488 		c = &cxgb4_natmode_config_array[i];
489 		if (CHELSIO_CHIP_VERSION(adap->params.chip) >= c->chip &&
490 		    natmode_flags == c->flags)
491 			return 0;
492 	}
493 	NL_SET_ERR_MSG_MOD(extack, "Unsupported NAT mode 4-tuple combination");
494 	return -EOPNOTSUPP;
495 }
496 
497 void cxgb4_process_flow_actions(struct net_device *in,
498 				struct flow_action *actions,
499 				struct ch_filter_specification *fs)
500 {
501 	struct flow_action_entry *act;
502 	u8 natmode_flags = 0;
503 	int i;
504 
505 	flow_action_for_each(i, act, actions) {
506 		switch (act->id) {
507 		case FLOW_ACTION_ACCEPT:
508 			fs->action = FILTER_PASS;
509 			break;
510 		case FLOW_ACTION_DROP:
511 			fs->action = FILTER_DROP;
512 			break;
513 		case FLOW_ACTION_MIRRED:
514 		case FLOW_ACTION_REDIRECT: {
515 			struct net_device *out = act->dev;
516 			struct port_info *pi = netdev_priv(out);
517 
518 			fs->action = FILTER_SWITCH;
519 			fs->eport = pi->port_id;
520 			}
521 			break;
522 		case FLOW_ACTION_VLAN_POP:
523 		case FLOW_ACTION_VLAN_PUSH:
524 		case FLOW_ACTION_VLAN_MANGLE: {
525 			u8 prio = act->vlan.prio;
526 			u16 vid = act->vlan.vid;
527 			u16 vlan_tci = (prio << VLAN_PRIO_SHIFT) | vid;
528 			switch (act->id) {
529 			case FLOW_ACTION_VLAN_POP:
530 				fs->newvlan |= VLAN_REMOVE;
531 				break;
532 			case FLOW_ACTION_VLAN_PUSH:
533 				fs->newvlan |= VLAN_INSERT;
534 				fs->vlan = vlan_tci;
535 				break;
536 			case FLOW_ACTION_VLAN_MANGLE:
537 				fs->newvlan |= VLAN_REWRITE;
538 				fs->vlan = vlan_tci;
539 				break;
540 			default:
541 				break;
542 			}
543 			}
544 			break;
545 		case FLOW_ACTION_MANGLE: {
546 			u32 mask, val, offset;
547 			u8 htype;
548 
549 			htype = act->mangle.htype;
550 			mask = act->mangle.mask;
551 			val = act->mangle.val;
552 			offset = act->mangle.offset;
553 
554 			process_pedit_field(fs, val, mask, offset, htype,
555 					    &natmode_flags);
556 			}
557 			break;
558 		case FLOW_ACTION_QUEUE:
559 			fs->action = FILTER_PASS;
560 			fs->dirsteer = 1;
561 			fs->iq = act->queue.index;
562 			break;
563 		default:
564 			break;
565 		}
566 	}
567 	if (natmode_flags)
568 		cxgb4_action_natmode_tweak(fs, natmode_flags);
569 
570 }
571 
572 static bool valid_l4_mask(u32 mask)
573 {
574 	u16 hi, lo;
575 
576 	/* Either the upper 16-bits (SPORT) OR the lower
577 	 * 16-bits (DPORT) can be set, but NOT BOTH.
578 	 */
579 	hi = (mask >> 16) & 0xFFFF;
580 	lo = mask & 0xFFFF;
581 
582 	return hi && lo ? false : true;
583 }
584 
585 static bool valid_pedit_action(struct netlink_ext_ack *extack,
586 			       const struct flow_action_entry *act,
587 			       u8 *natmode_flags)
588 {
589 	u32 mask, offset;
590 	u8 htype;
591 
592 	htype = act->mangle.htype;
593 	mask = act->mangle.mask;
594 	offset = act->mangle.offset;
595 
596 	switch (htype) {
597 	case FLOW_ACT_MANGLE_HDR_TYPE_ETH:
598 		switch (offset) {
599 		case PEDIT_ETH_DMAC_31_0:
600 		case PEDIT_ETH_DMAC_47_32_SMAC_15_0:
601 		case PEDIT_ETH_SMAC_47_16:
602 			break;
603 		default:
604 			NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field");
605 			return false;
606 		}
607 		break;
608 	case FLOW_ACT_MANGLE_HDR_TYPE_IP4:
609 		switch (offset) {
610 		case PEDIT_IP4_SRC:
611 			*natmode_flags |= CXGB4_ACTION_NATMODE_SIP;
612 			break;
613 		case PEDIT_IP4_DST:
614 			*natmode_flags |= CXGB4_ACTION_NATMODE_DIP;
615 			break;
616 		default:
617 			NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field");
618 			return false;
619 		}
620 		break;
621 	case FLOW_ACT_MANGLE_HDR_TYPE_IP6:
622 		switch (offset) {
623 		case PEDIT_IP6_SRC_31_0:
624 		case PEDIT_IP6_SRC_63_32:
625 		case PEDIT_IP6_SRC_95_64:
626 		case PEDIT_IP6_SRC_127_96:
627 			*natmode_flags |= CXGB4_ACTION_NATMODE_SIP;
628 			break;
629 		case PEDIT_IP6_DST_31_0:
630 		case PEDIT_IP6_DST_63_32:
631 		case PEDIT_IP6_DST_95_64:
632 		case PEDIT_IP6_DST_127_96:
633 			*natmode_flags |= CXGB4_ACTION_NATMODE_DIP;
634 			break;
635 		default:
636 			NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field");
637 			return false;
638 		}
639 		break;
640 	case FLOW_ACT_MANGLE_HDR_TYPE_TCP:
641 		switch (offset) {
642 		case PEDIT_TCP_SPORT_DPORT:
643 			if (!valid_l4_mask(~mask)) {
644 				NL_SET_ERR_MSG_MOD(extack,
645 						   "Unsupported mask for TCP L4 ports");
646 				return false;
647 			}
648 			if (~mask & PEDIT_TCP_UDP_SPORT_MASK)
649 				*natmode_flags |= CXGB4_ACTION_NATMODE_SPORT;
650 			else
651 				*natmode_flags |= CXGB4_ACTION_NATMODE_DPORT;
652 			break;
653 		default:
654 			NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field");
655 			return false;
656 		}
657 		break;
658 	case FLOW_ACT_MANGLE_HDR_TYPE_UDP:
659 		switch (offset) {
660 		case PEDIT_UDP_SPORT_DPORT:
661 			if (!valid_l4_mask(~mask)) {
662 				NL_SET_ERR_MSG_MOD(extack,
663 						   "Unsupported mask for UDP L4 ports");
664 				return false;
665 			}
666 			if (~mask & PEDIT_TCP_UDP_SPORT_MASK)
667 				*natmode_flags |= CXGB4_ACTION_NATMODE_SPORT;
668 			else
669 				*natmode_flags |= CXGB4_ACTION_NATMODE_DPORT;
670 			break;
671 		default:
672 			NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field");
673 			return false;
674 		}
675 		break;
676 	default:
677 		NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit type");
678 		return false;
679 	}
680 	return true;
681 }
682 
683 int cxgb4_validate_flow_actions(struct net_device *dev,
684 				struct flow_action *actions,
685 				struct netlink_ext_ack *extack,
686 				u8 matchall_filter)
687 {
688 	struct adapter *adap = netdev2adap(dev);
689 	struct flow_action_entry *act;
690 	bool act_redir = false;
691 	bool act_pedit = false;
692 	bool act_vlan = false;
693 	u8 natmode_flags = 0;
694 	int i;
695 
696 	if (!flow_action_basic_hw_stats_check(actions, extack))
697 		return -EOPNOTSUPP;
698 
699 	flow_action_for_each(i, act, actions) {
700 		switch (act->id) {
701 		case FLOW_ACTION_ACCEPT:
702 		case FLOW_ACTION_DROP:
703 			/* Do nothing */
704 			break;
705 		case FLOW_ACTION_MIRRED:
706 		case FLOW_ACTION_REDIRECT: {
707 			struct net_device *n_dev, *target_dev;
708 			bool found = false;
709 			unsigned int i;
710 
711 			if (act->id == FLOW_ACTION_MIRRED &&
712 			    !matchall_filter) {
713 				NL_SET_ERR_MSG_MOD(extack,
714 						   "Egress mirror action is only supported for tc-matchall");
715 				return -EOPNOTSUPP;
716 			}
717 
718 			target_dev = act->dev;
719 			for_each_port(adap, i) {
720 				n_dev = adap->port[i];
721 				if (target_dev == n_dev) {
722 					found = true;
723 					break;
724 				}
725 			}
726 
727 			/* If interface doesn't belong to our hw, then
728 			 * the provided output port is not valid
729 			 */
730 			if (!found) {
731 				NL_SET_ERR_MSG_MOD(extack, "Out port invalid");
732 				return -EINVAL;
733 			}
734 			act_redir = true;
735 			}
736 			break;
737 		case FLOW_ACTION_VLAN_POP:
738 		case FLOW_ACTION_VLAN_PUSH:
739 		case FLOW_ACTION_VLAN_MANGLE: {
740 			u16 proto = be16_to_cpu(act->vlan.proto);
741 
742 			switch (act->id) {
743 			case FLOW_ACTION_VLAN_POP:
744 				break;
745 			case FLOW_ACTION_VLAN_PUSH:
746 			case FLOW_ACTION_VLAN_MANGLE:
747 				if (proto != ETH_P_8021Q) {
748 					NL_SET_ERR_MSG_MOD(extack,
749 							   "Unsupported vlan proto");
750 					return -EOPNOTSUPP;
751 				}
752 				break;
753 			default:
754 				NL_SET_ERR_MSG_MOD(extack,
755 						   "Unsupported vlan action");
756 				return -EOPNOTSUPP;
757 			}
758 			act_vlan = true;
759 			}
760 			break;
761 		case FLOW_ACTION_MANGLE: {
762 			bool pedit_valid = valid_pedit_action(extack, act,
763 							      &natmode_flags);
764 
765 			if (!pedit_valid)
766 				return -EOPNOTSUPP;
767 			act_pedit = true;
768 			}
769 			break;
770 		case FLOW_ACTION_QUEUE:
771 			/* Do nothing. cxgb4_set_filter will validate */
772 			break;
773 		default:
774 			NL_SET_ERR_MSG_MOD(extack, "Unsupported action");
775 			return -EOPNOTSUPP;
776 		}
777 	}
778 
779 	if ((act_pedit || act_vlan) && !act_redir) {
780 		NL_SET_ERR_MSG_MOD(extack,
781 				   "pedit/vlan rewrite invalid without egress redirect");
782 		return -EINVAL;
783 	}
784 
785 	if (act_pedit) {
786 		int ret;
787 
788 		ret = cxgb4_action_natmode_validate(adap, natmode_flags,
789 						    extack);
790 		if (ret)
791 			return ret;
792 	}
793 
794 	return 0;
795 }
796 
797 static void cxgb4_tc_flower_hash_prio_add(struct adapter *adap, u32 tc_prio)
798 {
799 	spin_lock_bh(&adap->tids.ftid_lock);
800 	if (adap->tids.tc_hash_tids_max_prio < tc_prio)
801 		adap->tids.tc_hash_tids_max_prio = tc_prio;
802 	spin_unlock_bh(&adap->tids.ftid_lock);
803 }
804 
805 static void cxgb4_tc_flower_hash_prio_del(struct adapter *adap, u32 tc_prio)
806 {
807 	struct tid_info *t = &adap->tids;
808 	struct ch_tc_flower_entry *fe;
809 	struct rhashtable_iter iter;
810 	u32 found = 0;
811 
812 	spin_lock_bh(&t->ftid_lock);
813 	/* Bail if the current rule is not the one with the max
814 	 * prio.
815 	 */
816 	if (t->tc_hash_tids_max_prio != tc_prio)
817 		goto out_unlock;
818 
819 	/* Search for the next rule having the same or next lower
820 	 * max prio.
821 	 */
822 	rhashtable_walk_enter(&adap->flower_tbl, &iter);
823 	do {
824 		rhashtable_walk_start(&iter);
825 
826 		fe = rhashtable_walk_next(&iter);
827 		while (!IS_ERR_OR_NULL(fe)) {
828 			if (fe->fs.hash &&
829 			    fe->fs.tc_prio <= t->tc_hash_tids_max_prio) {
830 				t->tc_hash_tids_max_prio = fe->fs.tc_prio;
831 				found++;
832 
833 				/* Bail if we found another rule
834 				 * having the same prio as the
835 				 * current max one.
836 				 */
837 				if (fe->fs.tc_prio == tc_prio)
838 					break;
839 			}
840 
841 			fe = rhashtable_walk_next(&iter);
842 		}
843 
844 		rhashtable_walk_stop(&iter);
845 	} while (fe == ERR_PTR(-EAGAIN));
846 	rhashtable_walk_exit(&iter);
847 
848 	if (!found)
849 		t->tc_hash_tids_max_prio = 0;
850 
851 out_unlock:
852 	spin_unlock_bh(&t->ftid_lock);
853 }
854 
855 int cxgb4_flow_rule_replace(struct net_device *dev, struct flow_rule *rule,
856 			    u32 tc_prio, struct netlink_ext_ack *extack,
857 			    struct ch_filter_specification *fs, u32 *tid)
858 {
859 	struct adapter *adap = netdev2adap(dev);
860 	struct filter_ctx ctx;
861 	u8 inet_family;
862 	int fidx, ret;
863 
864 	if (cxgb4_validate_flow_actions(dev, &rule->action, extack, 0))
865 		return -EOPNOTSUPP;
866 
867 	if (cxgb4_validate_flow_match(extack, rule))
868 		return -EOPNOTSUPP;
869 
870 	cxgb4_process_flow_match(dev, rule, fs);
871 	cxgb4_process_flow_actions(dev, &rule->action, fs);
872 
873 	fs->hash = is_filter_exact_match(adap, fs);
874 	inet_family = fs->type ? PF_INET6 : PF_INET;
875 
876 	/* Get a free filter entry TID, where we can insert this new
877 	 * rule. Only insert rule if its prio doesn't conflict with
878 	 * existing rules.
879 	 */
880 	fidx = cxgb4_get_free_ftid(dev, inet_family, fs->hash,
881 				   tc_prio);
882 	if (fidx < 0) {
883 		NL_SET_ERR_MSG_MOD(extack,
884 				   "No free LETCAM index available");
885 		return -ENOMEM;
886 	}
887 
888 	if (fidx < adap->tids.nhpftids) {
889 		fs->prio = 1;
890 		fs->hash = 0;
891 	}
892 
893 	/* If the rule can be inserted into HASH region, then ignore
894 	 * the index to normal FILTER region.
895 	 */
896 	if (fs->hash)
897 		fidx = 0;
898 
899 	fs->tc_prio = tc_prio;
900 
901 	init_completion(&ctx.completion);
902 	ret = __cxgb4_set_filter(dev, fidx, fs, &ctx);
903 	if (ret) {
904 		NL_SET_ERR_MSG_FMT_MOD(extack, "filter creation err %d", ret);
905 		return ret;
906 	}
907 
908 	/* Wait for reply */
909 	ret = wait_for_completion_timeout(&ctx.completion, 10 * HZ);
910 	if (!ret)
911 		return -ETIMEDOUT;
912 
913 	/* Check if hw returned error for filter creation */
914 	if (ctx.result)
915 		return ctx.result;
916 
917 	*tid = ctx.tid;
918 
919 	if (fs->hash)
920 		cxgb4_tc_flower_hash_prio_add(adap, tc_prio);
921 
922 	return 0;
923 }
924 
925 int cxgb4_tc_flower_replace(struct net_device *dev,
926 			    struct flow_cls_offload *cls)
927 {
928 	struct flow_rule *rule = flow_cls_offload_flow_rule(cls);
929 	struct netlink_ext_ack *extack = cls->common.extack;
930 	struct adapter *adap = netdev2adap(dev);
931 	struct ch_tc_flower_entry *ch_flower;
932 	struct ch_filter_specification *fs;
933 	int ret;
934 
935 	ch_flower = allocate_flower_entry();
936 	if (!ch_flower) {
937 		netdev_err(dev, "%s: ch_flower alloc failed.\n", __func__);
938 		return -ENOMEM;
939 	}
940 
941 	fs = &ch_flower->fs;
942 	fs->hitcnts = 1;
943 	fs->tc_cookie = cls->cookie;
944 
945 	ret = cxgb4_flow_rule_replace(dev, rule, cls->common.prio, extack, fs,
946 				      &ch_flower->filter_id);
947 	if (ret)
948 		goto free_entry;
949 
950 	ch_flower->tc_flower_cookie = cls->cookie;
951 	ret = rhashtable_insert_fast(&adap->flower_tbl, &ch_flower->node,
952 				     adap->flower_ht_params);
953 	if (ret)
954 		goto del_filter;
955 
956 	return 0;
957 
958 del_filter:
959 	if (fs->hash)
960 		cxgb4_tc_flower_hash_prio_del(adap, cls->common.prio);
961 
962 	cxgb4_del_filter(dev, ch_flower->filter_id, &ch_flower->fs);
963 
964 free_entry:
965 	kfree(ch_flower);
966 	return ret;
967 }
968 
969 int cxgb4_flow_rule_destroy(struct net_device *dev, u32 tc_prio,
970 			    struct ch_filter_specification *fs, int tid)
971 {
972 	struct adapter *adap = netdev2adap(dev);
973 	u8 hash;
974 	int ret;
975 
976 	hash = fs->hash;
977 
978 	ret = cxgb4_del_filter(dev, tid, fs);
979 	if (ret)
980 		return ret;
981 
982 	if (hash)
983 		cxgb4_tc_flower_hash_prio_del(adap, tc_prio);
984 
985 	return ret;
986 }
987 
988 int cxgb4_tc_flower_destroy(struct net_device *dev,
989 			    struct flow_cls_offload *cls)
990 {
991 	struct adapter *adap = netdev2adap(dev);
992 	struct ch_tc_flower_entry *ch_flower;
993 	int ret;
994 
995 	ch_flower = ch_flower_lookup(adap, cls->cookie);
996 	if (!ch_flower)
997 		return -ENOENT;
998 
999 	rhashtable_remove_fast(&adap->flower_tbl, &ch_flower->node,
1000 			       adap->flower_ht_params);
1001 
1002 	ret = cxgb4_flow_rule_destroy(dev, ch_flower->fs.tc_prio,
1003 				      &ch_flower->fs, ch_flower->filter_id);
1004 	if (ret)
1005 		netdev_err(dev, "Flow rule destroy failed for tid: %u, ret: %d",
1006 			   ch_flower->filter_id, ret);
1007 
1008 	kfree_rcu(ch_flower, rcu);
1009 	return ret;
1010 }
1011 
1012 static void ch_flower_stats_handler(struct work_struct *work)
1013 {
1014 	struct adapter *adap = container_of(work, struct adapter,
1015 					    flower_stats_work);
1016 	struct ch_tc_flower_entry *flower_entry;
1017 	struct ch_tc_flower_stats *ofld_stats;
1018 	struct rhashtable_iter iter;
1019 	u64 packets;
1020 	u64 bytes;
1021 	int ret;
1022 
1023 	rhashtable_walk_enter(&adap->flower_tbl, &iter);
1024 	do {
1025 		rhashtable_walk_start(&iter);
1026 
1027 		while ((flower_entry = rhashtable_walk_next(&iter)) &&
1028 		       !IS_ERR(flower_entry)) {
1029 			ret = cxgb4_get_filter_counters(adap->port[0],
1030 							flower_entry->filter_id,
1031 							&packets, &bytes,
1032 							flower_entry->fs.hash);
1033 			if (!ret) {
1034 				spin_lock(&flower_entry->lock);
1035 				ofld_stats = &flower_entry->stats;
1036 
1037 				if (ofld_stats->prev_packet_count != packets) {
1038 					ofld_stats->prev_packet_count = packets;
1039 					ofld_stats->last_used = jiffies;
1040 				}
1041 				spin_unlock(&flower_entry->lock);
1042 			}
1043 		}
1044 
1045 		rhashtable_walk_stop(&iter);
1046 
1047 	} while (flower_entry == ERR_PTR(-EAGAIN));
1048 	rhashtable_walk_exit(&iter);
1049 	mod_timer(&adap->flower_stats_timer, jiffies + STATS_CHECK_PERIOD);
1050 }
1051 
1052 static void ch_flower_stats_cb(struct timer_list *t)
1053 {
1054 	struct adapter *adap = from_timer(adap, t, flower_stats_timer);
1055 
1056 	schedule_work(&adap->flower_stats_work);
1057 }
1058 
1059 int cxgb4_tc_flower_stats(struct net_device *dev,
1060 			  struct flow_cls_offload *cls)
1061 {
1062 	struct adapter *adap = netdev2adap(dev);
1063 	struct ch_tc_flower_stats *ofld_stats;
1064 	struct ch_tc_flower_entry *ch_flower;
1065 	u64 packets;
1066 	u64 bytes;
1067 	int ret;
1068 
1069 	ch_flower = ch_flower_lookup(adap, cls->cookie);
1070 	if (!ch_flower) {
1071 		ret = -ENOENT;
1072 		goto err;
1073 	}
1074 
1075 	ret = cxgb4_get_filter_counters(dev, ch_flower->filter_id,
1076 					&packets, &bytes,
1077 					ch_flower->fs.hash);
1078 	if (ret < 0)
1079 		goto err;
1080 
1081 	spin_lock_bh(&ch_flower->lock);
1082 	ofld_stats = &ch_flower->stats;
1083 	if (ofld_stats->packet_count != packets) {
1084 		if (ofld_stats->prev_packet_count != packets)
1085 			ofld_stats->last_used = jiffies;
1086 		flow_stats_update(&cls->stats, bytes - ofld_stats->byte_count,
1087 				  packets - ofld_stats->packet_count, 0,
1088 				  ofld_stats->last_used,
1089 				  FLOW_ACTION_HW_STATS_IMMEDIATE);
1090 
1091 		ofld_stats->packet_count = packets;
1092 		ofld_stats->byte_count = bytes;
1093 		ofld_stats->prev_packet_count = packets;
1094 	}
1095 	spin_unlock_bh(&ch_flower->lock);
1096 	return 0;
1097 
1098 err:
1099 	return ret;
1100 }
1101 
1102 static const struct rhashtable_params cxgb4_tc_flower_ht_params = {
1103 	.nelem_hint = 384,
1104 	.head_offset = offsetof(struct ch_tc_flower_entry, node),
1105 	.key_offset = offsetof(struct ch_tc_flower_entry, tc_flower_cookie),
1106 	.key_len = sizeof(((struct ch_tc_flower_entry *)0)->tc_flower_cookie),
1107 	.max_size = 524288,
1108 	.min_size = 512,
1109 	.automatic_shrinking = true
1110 };
1111 
1112 int cxgb4_init_tc_flower(struct adapter *adap)
1113 {
1114 	int ret;
1115 
1116 	if (adap->tc_flower_initialized)
1117 		return -EEXIST;
1118 
1119 	adap->flower_ht_params = cxgb4_tc_flower_ht_params;
1120 	ret = rhashtable_init(&adap->flower_tbl, &adap->flower_ht_params);
1121 	if (ret)
1122 		return ret;
1123 
1124 	INIT_WORK(&adap->flower_stats_work, ch_flower_stats_handler);
1125 	timer_setup(&adap->flower_stats_timer, ch_flower_stats_cb, 0);
1126 	mod_timer(&adap->flower_stats_timer, jiffies + STATS_CHECK_PERIOD);
1127 	adap->tc_flower_initialized = true;
1128 	return 0;
1129 }
1130 
1131 void cxgb4_cleanup_tc_flower(struct adapter *adap)
1132 {
1133 	if (!adap->tc_flower_initialized)
1134 		return;
1135 
1136 	if (adap->flower_stats_timer.function)
1137 		timer_shutdown_sync(&adap->flower_stats_timer);
1138 	cancel_work_sync(&adap->flower_stats_work);
1139 	rhashtable_destroy(&adap->flower_tbl);
1140 	adap->tc_flower_initialized = false;
1141 }
1142