xref: /openbmc/linux/drivers/infiniband/hw/mlx5/fs.c (revision c9933d494c54f72290831191c09bb8488bfd5905)
1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /*
3  * Copyright (c) 2018, Mellanox Technologies inc.  All rights reserved.
4  */
5 
6 #include <rdma/ib_user_verbs.h>
7 #include <rdma/ib_verbs.h>
8 #include <rdma/uverbs_types.h>
9 #include <rdma/uverbs_ioctl.h>
10 #include <rdma/uverbs_std_types.h>
11 #include <rdma/mlx5_user_ioctl_cmds.h>
12 #include <rdma/mlx5_user_ioctl_verbs.h>
13 #include <rdma/ib_hdrs.h>
14 #include <rdma/ib_umem.h>
15 #include <linux/mlx5/driver.h>
16 #include <linux/mlx5/fs.h>
17 #include <linux/mlx5/fs_helpers.h>
18 #include <linux/mlx5/eswitch.h>
19 #include <net/inet_ecn.h>
20 #include "mlx5_ib.h"
21 #include "counters.h"
22 #include "devx.h"
23 #include "fs.h"
24 
25 #define UVERBS_MODULE_NAME mlx5_ib
26 #include <rdma/uverbs_named_ioctl.h>
27 
28 enum {
29 	MATCH_CRITERIA_ENABLE_OUTER_BIT,
30 	MATCH_CRITERIA_ENABLE_MISC_BIT,
31 	MATCH_CRITERIA_ENABLE_INNER_BIT,
32 	MATCH_CRITERIA_ENABLE_MISC2_BIT
33 };
34 
35 #define HEADER_IS_ZERO(match_criteria, headers)			           \
36 	!(memchr_inv(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
37 		    0, MLX5_FLD_SZ_BYTES(fte_match_param, headers)))       \
38 
39 static u8 get_match_criteria_enable(u32 *match_criteria)
40 {
41 	u8 match_criteria_enable;
42 
43 	match_criteria_enable =
44 		(!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
45 		MATCH_CRITERIA_ENABLE_OUTER_BIT;
46 	match_criteria_enable |=
47 		(!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
48 		MATCH_CRITERIA_ENABLE_MISC_BIT;
49 	match_criteria_enable |=
50 		(!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
51 		MATCH_CRITERIA_ENABLE_INNER_BIT;
52 	match_criteria_enable |=
53 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
54 		MATCH_CRITERIA_ENABLE_MISC2_BIT;
55 
56 	return match_criteria_enable;
57 }
58 
59 static int set_proto(void *outer_c, void *outer_v, u8 mask, u8 val)
60 {
61 	u8 entry_mask;
62 	u8 entry_val;
63 	int err = 0;
64 
65 	if (!mask)
66 		goto out;
67 
68 	entry_mask = MLX5_GET(fte_match_set_lyr_2_4, outer_c,
69 			      ip_protocol);
70 	entry_val = MLX5_GET(fte_match_set_lyr_2_4, outer_v,
71 			     ip_protocol);
72 	if (!entry_mask) {
73 		MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_protocol, mask);
74 		MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_protocol, val);
75 		goto out;
76 	}
77 	/* Don't override existing ip protocol */
78 	if (mask != entry_mask || val != entry_val)
79 		err = -EINVAL;
80 out:
81 	return err;
82 }
83 
84 static void set_flow_label(void *misc_c, void *misc_v, u32 mask, u32 val,
85 			   bool inner)
86 {
87 	if (inner) {
88 		MLX5_SET(fte_match_set_misc,
89 			 misc_c, inner_ipv6_flow_label, mask);
90 		MLX5_SET(fte_match_set_misc,
91 			 misc_v, inner_ipv6_flow_label, val);
92 	} else {
93 		MLX5_SET(fte_match_set_misc,
94 			 misc_c, outer_ipv6_flow_label, mask);
95 		MLX5_SET(fte_match_set_misc,
96 			 misc_v, outer_ipv6_flow_label, val);
97 	}
98 }
99 
100 static void set_tos(void *outer_c, void *outer_v, u8 mask, u8 val)
101 {
102 	MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_ecn, mask);
103 	MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_ecn, val);
104 	MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_dscp, mask >> 2);
105 	MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_dscp, val >> 2);
106 }
107 
108 static int check_mpls_supp_fields(u32 field_support, const __be32 *set_mask)
109 {
110 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_label) &&
111 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_LABEL))
112 		return -EOPNOTSUPP;
113 
114 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_exp) &&
115 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_EXP))
116 		return -EOPNOTSUPP;
117 
118 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_s_bos) &&
119 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_S_BOS))
120 		return -EOPNOTSUPP;
121 
122 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_ttl) &&
123 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_TTL))
124 		return -EOPNOTSUPP;
125 
126 	return 0;
127 }
128 
129 #define LAST_ETH_FIELD vlan_tag
130 #define LAST_IB_FIELD sl
131 #define LAST_IPV4_FIELD tos
132 #define LAST_IPV6_FIELD traffic_class
133 #define LAST_TCP_UDP_FIELD src_port
134 #define LAST_TUNNEL_FIELD tunnel_id
135 #define LAST_FLOW_TAG_FIELD tag_id
136 #define LAST_DROP_FIELD size
137 #define LAST_COUNTERS_FIELD counters
138 
139 /* Field is the last supported field */
140 #define FIELDS_NOT_SUPPORTED(filter, field)                                    \
141 	memchr_inv((void *)&filter.field + sizeof(filter.field), 0,            \
142 		   sizeof(filter) - offsetofend(typeof(filter), field))
143 
144 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
145 			   bool is_egress,
146 			   struct mlx5_flow_act *action)
147 {
148 
149 	switch (maction->ib_action.type) {
150 	case IB_FLOW_ACTION_UNSPECIFIED:
151 		if (maction->flow_action_raw.sub_type ==
152 		    MLX5_IB_FLOW_ACTION_MODIFY_HEADER) {
153 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
154 				return -EINVAL;
155 			action->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
156 			action->modify_hdr =
157 				maction->flow_action_raw.modify_hdr;
158 			return 0;
159 		}
160 		if (maction->flow_action_raw.sub_type ==
161 		    MLX5_IB_FLOW_ACTION_DECAP) {
162 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_DECAP)
163 				return -EINVAL;
164 			action->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
165 			return 0;
166 		}
167 		if (maction->flow_action_raw.sub_type ==
168 		    MLX5_IB_FLOW_ACTION_PACKET_REFORMAT) {
169 			if (action->action &
170 			    MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT)
171 				return -EINVAL;
172 			action->action |=
173 				MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
174 			action->pkt_reformat =
175 				maction->flow_action_raw.pkt_reformat;
176 			return 0;
177 		}
178 		fallthrough;
179 	default:
180 		return -EOPNOTSUPP;
181 	}
182 }
183 
184 static int parse_flow_attr(struct mlx5_core_dev *mdev,
185 			   struct mlx5_flow_spec *spec,
186 			   const union ib_flow_spec *ib_spec,
187 			   const struct ib_flow_attr *flow_attr,
188 			   struct mlx5_flow_act *action, u32 prev_type)
189 {
190 	struct mlx5_flow_context *flow_context = &spec->flow_context;
191 	u32 *match_c = spec->match_criteria;
192 	u32 *match_v = spec->match_value;
193 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param, match_c,
194 					   misc_parameters);
195 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, match_v,
196 					   misc_parameters);
197 	void *misc_params2_c = MLX5_ADDR_OF(fte_match_param, match_c,
198 					    misc_parameters_2);
199 	void *misc_params2_v = MLX5_ADDR_OF(fte_match_param, match_v,
200 					    misc_parameters_2);
201 	void *headers_c;
202 	void *headers_v;
203 	int match_ipv;
204 	int ret;
205 
206 	if (ib_spec->type & IB_FLOW_SPEC_INNER) {
207 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
208 					 inner_headers);
209 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
210 					 inner_headers);
211 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
212 					ft_field_support.inner_ip_version);
213 	} else {
214 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
215 					 outer_headers);
216 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
217 					 outer_headers);
218 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
219 					ft_field_support.outer_ip_version);
220 	}
221 
222 	switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
223 	case IB_FLOW_SPEC_ETH:
224 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
225 			return -EOPNOTSUPP;
226 
227 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
228 					     dmac_47_16),
229 				ib_spec->eth.mask.dst_mac);
230 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
231 					     dmac_47_16),
232 				ib_spec->eth.val.dst_mac);
233 
234 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
235 					     smac_47_16),
236 				ib_spec->eth.mask.src_mac);
237 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
238 					     smac_47_16),
239 				ib_spec->eth.val.src_mac);
240 
241 		if (ib_spec->eth.mask.vlan_tag) {
242 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
243 				 cvlan_tag, 1);
244 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
245 				 cvlan_tag, 1);
246 
247 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
248 				 first_vid, ntohs(ib_spec->eth.mask.vlan_tag));
249 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
250 				 first_vid, ntohs(ib_spec->eth.val.vlan_tag));
251 
252 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
253 				 first_cfi,
254 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 12);
255 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
256 				 first_cfi,
257 				 ntohs(ib_spec->eth.val.vlan_tag) >> 12);
258 
259 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
260 				 first_prio,
261 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 13);
262 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
263 				 first_prio,
264 				 ntohs(ib_spec->eth.val.vlan_tag) >> 13);
265 		}
266 		MLX5_SET(fte_match_set_lyr_2_4, headers_c,
267 			 ethertype, ntohs(ib_spec->eth.mask.ether_type));
268 		MLX5_SET(fte_match_set_lyr_2_4, headers_v,
269 			 ethertype, ntohs(ib_spec->eth.val.ether_type));
270 		break;
271 	case IB_FLOW_SPEC_IPV4:
272 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
273 			return -EOPNOTSUPP;
274 
275 		if (match_ipv) {
276 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
277 				 ip_version, 0xf);
278 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
279 				 ip_version, MLX5_FS_IPV4_VERSION);
280 		} else {
281 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
282 				 ethertype, 0xffff);
283 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
284 				 ethertype, ETH_P_IP);
285 		}
286 
287 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
288 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
289 		       &ib_spec->ipv4.mask.src_ip,
290 		       sizeof(ib_spec->ipv4.mask.src_ip));
291 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
292 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
293 		       &ib_spec->ipv4.val.src_ip,
294 		       sizeof(ib_spec->ipv4.val.src_ip));
295 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
296 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
297 		       &ib_spec->ipv4.mask.dst_ip,
298 		       sizeof(ib_spec->ipv4.mask.dst_ip));
299 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
300 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
301 		       &ib_spec->ipv4.val.dst_ip,
302 		       sizeof(ib_spec->ipv4.val.dst_ip));
303 
304 		set_tos(headers_c, headers_v,
305 			ib_spec->ipv4.mask.tos, ib_spec->ipv4.val.tos);
306 
307 		if (set_proto(headers_c, headers_v,
308 			      ib_spec->ipv4.mask.proto,
309 			      ib_spec->ipv4.val.proto))
310 			return -EINVAL;
311 		break;
312 	case IB_FLOW_SPEC_IPV6:
313 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv6.mask, LAST_IPV6_FIELD))
314 			return -EOPNOTSUPP;
315 
316 		if (match_ipv) {
317 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
318 				 ip_version, 0xf);
319 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
320 				 ip_version, MLX5_FS_IPV6_VERSION);
321 		} else {
322 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
323 				 ethertype, 0xffff);
324 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
325 				 ethertype, ETH_P_IPV6);
326 		}
327 
328 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
329 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
330 		       &ib_spec->ipv6.mask.src_ip,
331 		       sizeof(ib_spec->ipv6.mask.src_ip));
332 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
333 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
334 		       &ib_spec->ipv6.val.src_ip,
335 		       sizeof(ib_spec->ipv6.val.src_ip));
336 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
337 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
338 		       &ib_spec->ipv6.mask.dst_ip,
339 		       sizeof(ib_spec->ipv6.mask.dst_ip));
340 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
341 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
342 		       &ib_spec->ipv6.val.dst_ip,
343 		       sizeof(ib_spec->ipv6.val.dst_ip));
344 
345 		set_tos(headers_c, headers_v,
346 			ib_spec->ipv6.mask.traffic_class,
347 			ib_spec->ipv6.val.traffic_class);
348 
349 		if (set_proto(headers_c, headers_v,
350 			      ib_spec->ipv6.mask.next_hdr,
351 			      ib_spec->ipv6.val.next_hdr))
352 			return -EINVAL;
353 
354 		set_flow_label(misc_params_c, misc_params_v,
355 			       ntohl(ib_spec->ipv6.mask.flow_label),
356 			       ntohl(ib_spec->ipv6.val.flow_label),
357 			       ib_spec->type & IB_FLOW_SPEC_INNER);
358 		break;
359 	case IB_FLOW_SPEC_ESP:
360 		return -EOPNOTSUPP;
361 	case IB_FLOW_SPEC_TCP:
362 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
363 					 LAST_TCP_UDP_FIELD))
364 			return -EOPNOTSUPP;
365 
366 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_TCP))
367 			return -EINVAL;
368 
369 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_sport,
370 			 ntohs(ib_spec->tcp_udp.mask.src_port));
371 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
372 			 ntohs(ib_spec->tcp_udp.val.src_port));
373 
374 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_dport,
375 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
376 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
377 			 ntohs(ib_spec->tcp_udp.val.dst_port));
378 		break;
379 	case IB_FLOW_SPEC_UDP:
380 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
381 					 LAST_TCP_UDP_FIELD))
382 			return -EOPNOTSUPP;
383 
384 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_UDP))
385 			return -EINVAL;
386 
387 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_sport,
388 			 ntohs(ib_spec->tcp_udp.mask.src_port));
389 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
390 			 ntohs(ib_spec->tcp_udp.val.src_port));
391 
392 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_dport,
393 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
394 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
395 			 ntohs(ib_spec->tcp_udp.val.dst_port));
396 		break;
397 	case IB_FLOW_SPEC_GRE:
398 		if (ib_spec->gre.mask.c_ks_res0_ver)
399 			return -EOPNOTSUPP;
400 
401 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_GRE))
402 			return -EINVAL;
403 
404 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
405 			 0xff);
406 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
407 			 IPPROTO_GRE);
408 
409 		MLX5_SET(fte_match_set_misc, misc_params_c, gre_protocol,
410 			 ntohs(ib_spec->gre.mask.protocol));
411 		MLX5_SET(fte_match_set_misc, misc_params_v, gre_protocol,
412 			 ntohs(ib_spec->gre.val.protocol));
413 
414 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_c,
415 				    gre_key.nvgre.hi),
416 		       &ib_spec->gre.mask.key,
417 		       sizeof(ib_spec->gre.mask.key));
418 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_v,
419 				    gre_key.nvgre.hi),
420 		       &ib_spec->gre.val.key,
421 		       sizeof(ib_spec->gre.val.key));
422 		break;
423 	case IB_FLOW_SPEC_MPLS:
424 		switch (prev_type) {
425 		case IB_FLOW_SPEC_UDP:
426 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
427 						   ft_field_support.outer_first_mpls_over_udp),
428 						   &ib_spec->mpls.mask.tag))
429 				return -EOPNOTSUPP;
430 
431 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
432 					    outer_first_mpls_over_udp),
433 			       &ib_spec->mpls.val.tag,
434 			       sizeof(ib_spec->mpls.val.tag));
435 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
436 					    outer_first_mpls_over_udp),
437 			       &ib_spec->mpls.mask.tag,
438 			       sizeof(ib_spec->mpls.mask.tag));
439 			break;
440 		case IB_FLOW_SPEC_GRE:
441 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
442 						   ft_field_support.outer_first_mpls_over_gre),
443 						   &ib_spec->mpls.mask.tag))
444 				return -EOPNOTSUPP;
445 
446 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
447 					    outer_first_mpls_over_gre),
448 			       &ib_spec->mpls.val.tag,
449 			       sizeof(ib_spec->mpls.val.tag));
450 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
451 					    outer_first_mpls_over_gre),
452 			       &ib_spec->mpls.mask.tag,
453 			       sizeof(ib_spec->mpls.mask.tag));
454 			break;
455 		default:
456 			if (ib_spec->type & IB_FLOW_SPEC_INNER) {
457 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
458 							   ft_field_support.inner_first_mpls),
459 							   &ib_spec->mpls.mask.tag))
460 					return -EOPNOTSUPP;
461 
462 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
463 						    inner_first_mpls),
464 				       &ib_spec->mpls.val.tag,
465 				       sizeof(ib_spec->mpls.val.tag));
466 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
467 						    inner_first_mpls),
468 				       &ib_spec->mpls.mask.tag,
469 				       sizeof(ib_spec->mpls.mask.tag));
470 			} else {
471 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
472 							   ft_field_support.outer_first_mpls),
473 							   &ib_spec->mpls.mask.tag))
474 					return -EOPNOTSUPP;
475 
476 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
477 						    outer_first_mpls),
478 				       &ib_spec->mpls.val.tag,
479 				       sizeof(ib_spec->mpls.val.tag));
480 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
481 						    outer_first_mpls),
482 				       &ib_spec->mpls.mask.tag,
483 				       sizeof(ib_spec->mpls.mask.tag));
484 			}
485 		}
486 		break;
487 	case IB_FLOW_SPEC_VXLAN_TUNNEL:
488 		if (FIELDS_NOT_SUPPORTED(ib_spec->tunnel.mask,
489 					 LAST_TUNNEL_FIELD))
490 			return -EOPNOTSUPP;
491 
492 		MLX5_SET(fte_match_set_misc, misc_params_c, vxlan_vni,
493 			 ntohl(ib_spec->tunnel.mask.tunnel_id));
494 		MLX5_SET(fte_match_set_misc, misc_params_v, vxlan_vni,
495 			 ntohl(ib_spec->tunnel.val.tunnel_id));
496 		break;
497 	case IB_FLOW_SPEC_ACTION_TAG:
498 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_tag,
499 					 LAST_FLOW_TAG_FIELD))
500 			return -EOPNOTSUPP;
501 		if (ib_spec->flow_tag.tag_id >= BIT(24))
502 			return -EINVAL;
503 
504 		flow_context->flow_tag = ib_spec->flow_tag.tag_id;
505 		flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
506 		break;
507 	case IB_FLOW_SPEC_ACTION_DROP:
508 		if (FIELDS_NOT_SUPPORTED(ib_spec->drop,
509 					 LAST_DROP_FIELD))
510 			return -EOPNOTSUPP;
511 		action->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
512 		break;
513 	case IB_FLOW_SPEC_ACTION_HANDLE:
514 		ret = parse_flow_flow_action(to_mflow_act(ib_spec->action.act),
515 			flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS, action);
516 		if (ret)
517 			return ret;
518 		break;
519 	case IB_FLOW_SPEC_ACTION_COUNT:
520 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_count,
521 					 LAST_COUNTERS_FIELD))
522 			return -EOPNOTSUPP;
523 
524 		/* for now support only one counters spec per flow */
525 		if (action->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
526 			return -EINVAL;
527 
528 		action->counters = ib_spec->flow_count.counters;
529 		action->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
530 		break;
531 	default:
532 		return -EINVAL;
533 	}
534 
535 	return 0;
536 }
537 
538 /* If a flow could catch both multicast and unicast packets,
539  * it won't fall into the multicast flow steering table and this rule
540  * could steal other multicast packets.
541  */
542 static bool flow_is_multicast_only(const struct ib_flow_attr *ib_attr)
543 {
544 	union ib_flow_spec *flow_spec;
545 
546 	if (ib_attr->type != IB_FLOW_ATTR_NORMAL ||
547 	    ib_attr->num_of_specs < 1)
548 		return false;
549 
550 	flow_spec = (union ib_flow_spec *)(ib_attr + 1);
551 	if (flow_spec->type == IB_FLOW_SPEC_IPV4) {
552 		struct ib_flow_spec_ipv4 *ipv4_spec;
553 
554 		ipv4_spec = (struct ib_flow_spec_ipv4 *)flow_spec;
555 		if (ipv4_is_multicast(ipv4_spec->val.dst_ip))
556 			return true;
557 
558 		return false;
559 	}
560 
561 	if (flow_spec->type == IB_FLOW_SPEC_ETH) {
562 		struct ib_flow_spec_eth *eth_spec;
563 
564 		eth_spec = (struct ib_flow_spec_eth *)flow_spec;
565 		return is_multicast_ether_addr(eth_spec->mask.dst_mac) &&
566 		       is_multicast_ether_addr(eth_spec->val.dst_mac);
567 	}
568 
569 	return false;
570 }
571 
572 static bool is_valid_ethertype(struct mlx5_core_dev *mdev,
573 			       const struct ib_flow_attr *flow_attr,
574 			       bool check_inner)
575 {
576 	union ib_flow_spec *ib_spec = (union ib_flow_spec *)(flow_attr + 1);
577 	int match_ipv = check_inner ?
578 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
579 					ft_field_support.inner_ip_version) :
580 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
581 					ft_field_support.outer_ip_version);
582 	int inner_bit = check_inner ? IB_FLOW_SPEC_INNER : 0;
583 	bool ipv4_spec_valid, ipv6_spec_valid;
584 	unsigned int ip_spec_type = 0;
585 	bool has_ethertype = false;
586 	unsigned int spec_index;
587 	bool mask_valid = true;
588 	u16 eth_type = 0;
589 	bool type_valid;
590 
591 	/* Validate that ethertype is correct */
592 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
593 		if ((ib_spec->type == (IB_FLOW_SPEC_ETH | inner_bit)) &&
594 		    ib_spec->eth.mask.ether_type) {
595 			mask_valid = (ib_spec->eth.mask.ether_type ==
596 				      htons(0xffff));
597 			has_ethertype = true;
598 			eth_type = ntohs(ib_spec->eth.val.ether_type);
599 		} else if ((ib_spec->type == (IB_FLOW_SPEC_IPV4 | inner_bit)) ||
600 			   (ib_spec->type == (IB_FLOW_SPEC_IPV6 | inner_bit))) {
601 			ip_spec_type = ib_spec->type;
602 		}
603 		ib_spec = (void *)ib_spec + ib_spec->size;
604 	}
605 
606 	type_valid = (!has_ethertype) || (!ip_spec_type);
607 	if (!type_valid && mask_valid) {
608 		ipv4_spec_valid = (eth_type == ETH_P_IP) &&
609 			(ip_spec_type == (IB_FLOW_SPEC_IPV4 | inner_bit));
610 		ipv6_spec_valid = (eth_type == ETH_P_IPV6) &&
611 			(ip_spec_type == (IB_FLOW_SPEC_IPV6 | inner_bit));
612 
613 		type_valid = (ipv4_spec_valid) || (ipv6_spec_valid) ||
614 			     (((eth_type == ETH_P_MPLS_UC) ||
615 			       (eth_type == ETH_P_MPLS_MC)) && match_ipv);
616 	}
617 
618 	return type_valid;
619 }
620 
621 static bool is_valid_attr(struct mlx5_core_dev *mdev,
622 			  const struct ib_flow_attr *flow_attr)
623 {
624 	return is_valid_ethertype(mdev, flow_attr, false) &&
625 	       is_valid_ethertype(mdev, flow_attr, true);
626 }
627 
628 static void put_flow_table(struct mlx5_ib_dev *dev,
629 			   struct mlx5_ib_flow_prio *prio, bool ft_added)
630 {
631 	prio->refcount -= !!ft_added;
632 	if (!prio->refcount) {
633 		mlx5_destroy_flow_table(prio->flow_table);
634 		prio->flow_table = NULL;
635 	}
636 }
637 
638 static int mlx5_ib_destroy_flow(struct ib_flow *flow_id)
639 {
640 	struct mlx5_ib_flow_handler *handler = container_of(flow_id,
641 							  struct mlx5_ib_flow_handler,
642 							  ibflow);
643 	struct mlx5_ib_flow_handler *iter, *tmp;
644 	struct mlx5_ib_dev *dev = handler->dev;
645 
646 	mutex_lock(&dev->flow_db->lock);
647 
648 	list_for_each_entry_safe(iter, tmp, &handler->list, list) {
649 		mlx5_del_flow_rules(iter->rule);
650 		put_flow_table(dev, iter->prio, true);
651 		list_del(&iter->list);
652 		kfree(iter);
653 	}
654 
655 	mlx5_del_flow_rules(handler->rule);
656 	put_flow_table(dev, handler->prio, true);
657 	mlx5_ib_counters_clear_description(handler->ibcounters);
658 	mutex_unlock(&dev->flow_db->lock);
659 	if (handler->flow_matcher)
660 		atomic_dec(&handler->flow_matcher->usecnt);
661 	kfree(handler);
662 
663 	return 0;
664 }
665 
666 static int ib_prio_to_core_prio(unsigned int priority, bool dont_trap)
667 {
668 	priority *= 2;
669 	if (!dont_trap)
670 		priority++;
671 	return priority;
672 }
673 
674 enum flow_table_type {
675 	MLX5_IB_FT_RX,
676 	MLX5_IB_FT_TX
677 };
678 
679 #define MLX5_FS_MAX_TYPES	 6
680 #define MLX5_FS_MAX_ENTRIES	 BIT(16)
681 
682 static struct mlx5_ib_flow_prio *_get_prio(struct mlx5_flow_namespace *ns,
683 					   struct mlx5_ib_flow_prio *prio,
684 					   int priority,
685 					   int num_entries, int num_groups,
686 					   u32 flags)
687 {
688 	struct mlx5_flow_table_attr ft_attr = {};
689 	struct mlx5_flow_table *ft;
690 
691 	ft_attr.prio = priority;
692 	ft_attr.max_fte = num_entries;
693 	ft_attr.flags = flags;
694 	ft_attr.autogroup.max_num_groups = num_groups;
695 	ft = mlx5_create_auto_grouped_flow_table(ns, &ft_attr);
696 	if (IS_ERR(ft))
697 		return ERR_CAST(ft);
698 
699 	prio->flow_table = ft;
700 	prio->refcount = 0;
701 	return prio;
702 }
703 
704 static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev,
705 						struct ib_flow_attr *flow_attr,
706 						enum flow_table_type ft_type)
707 {
708 	bool dont_trap = flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP;
709 	struct mlx5_flow_namespace *ns = NULL;
710 	enum mlx5_flow_namespace_type fn_type;
711 	struct mlx5_ib_flow_prio *prio;
712 	struct mlx5_flow_table *ft;
713 	int max_table_size;
714 	int num_entries;
715 	int num_groups;
716 	bool esw_encap;
717 	u32 flags = 0;
718 	int priority;
719 
720 	max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
721 						       log_max_ft_size));
722 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
723 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
724 	switch (flow_attr->type) {
725 	case IB_FLOW_ATTR_NORMAL:
726 		if (flow_is_multicast_only(flow_attr) && !dont_trap)
727 			priority = MLX5_IB_FLOW_MCAST_PRIO;
728 		else
729 			priority = ib_prio_to_core_prio(flow_attr->priority,
730 							dont_trap);
731 		if (ft_type == MLX5_IB_FT_RX) {
732 			fn_type = MLX5_FLOW_NAMESPACE_BYPASS;
733 			prio = &dev->flow_db->prios[priority];
734 			if (!dev->is_rep && !esw_encap &&
735 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap))
736 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
737 			if (!dev->is_rep && !esw_encap &&
738 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
739 						      reformat_l3_tunnel_to_l2))
740 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
741 		} else {
742 			max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_TX(
743 				dev->mdev, log_max_ft_size));
744 			fn_type = MLX5_FLOW_NAMESPACE_EGRESS;
745 			prio = &dev->flow_db->egress_prios[priority];
746 			if (!dev->is_rep && !esw_encap &&
747 			    MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat))
748 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
749 		}
750 		ns = mlx5_get_flow_namespace(dev->mdev, fn_type);
751 		num_entries = MLX5_FS_MAX_ENTRIES;
752 		num_groups = MLX5_FS_MAX_TYPES;
753 		break;
754 	case IB_FLOW_ATTR_ALL_DEFAULT:
755 	case IB_FLOW_ATTR_MC_DEFAULT:
756 		ns = mlx5_get_flow_namespace(dev->mdev,
757 					     MLX5_FLOW_NAMESPACE_LEFTOVERS);
758 		build_leftovers_ft_param(&priority, &num_entries, &num_groups);
759 		prio = &dev->flow_db->prios[MLX5_IB_FLOW_LEFTOVERS_PRIO];
760 		break;
761 	case IB_FLOW_ATTR_SNIFFER:
762 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
763 					allow_sniffer_and_nic_rx_shared_tir))
764 			return ERR_PTR(-EOPNOTSUPP);
765 
766 		ns = mlx5_get_flow_namespace(
767 			dev->mdev, ft_type == MLX5_IB_FT_RX ?
768 					   MLX5_FLOW_NAMESPACE_SNIFFER_RX :
769 					   MLX5_FLOW_NAMESPACE_SNIFFER_TX);
770 
771 		prio = &dev->flow_db->sniffer[ft_type];
772 		priority = 0;
773 		num_entries = 1;
774 		num_groups = 1;
775 		break;
776 	default:
777 		break;
778 	}
779 
780 	if (!ns)
781 		return ERR_PTR(-EOPNOTSUPP);
782 
783 	max_table_size = min_t(int, num_entries, max_table_size);
784 
785 	ft = prio->flow_table;
786 	if (!ft)
787 		return _get_prio(ns, prio, priority, max_table_size, num_groups,
788 				 flags);
789 
790 	return prio;
791 }
792 
793 enum {
794 	RDMA_RX_ECN_OPCOUNTER_PRIO,
795 	RDMA_RX_CNP_OPCOUNTER_PRIO,
796 };
797 
798 enum {
799 	RDMA_TX_CNP_OPCOUNTER_PRIO,
800 };
801 
802 static int set_vhca_port_spec(struct mlx5_ib_dev *dev, u32 port_num,
803 			      struct mlx5_flow_spec *spec)
804 {
805 	if (!MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev,
806 					ft_field_support.source_vhca_port) ||
807 	    !MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
808 					ft_field_support.source_vhca_port))
809 		return -EOPNOTSUPP;
810 
811 	MLX5_SET_TO_ONES(fte_match_param, &spec->match_criteria,
812 			 misc_parameters.source_vhca_port);
813 	MLX5_SET(fte_match_param, &spec->match_value,
814 		 misc_parameters.source_vhca_port, port_num);
815 
816 	return 0;
817 }
818 
819 static int set_ecn_ce_spec(struct mlx5_ib_dev *dev, u32 port_num,
820 			   struct mlx5_flow_spec *spec, int ipv)
821 {
822 	if (!MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev,
823 					ft_field_support.outer_ip_version))
824 		return -EOPNOTSUPP;
825 
826 	if (mlx5_core_mp_enabled(dev->mdev) &&
827 	    set_vhca_port_spec(dev, port_num, spec))
828 		return -EOPNOTSUPP;
829 
830 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
831 			 outer_headers.ip_ecn);
832 	MLX5_SET(fte_match_param, spec->match_value, outer_headers.ip_ecn,
833 		 INET_ECN_CE);
834 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
835 			 outer_headers.ip_version);
836 	MLX5_SET(fte_match_param, spec->match_value, outer_headers.ip_version,
837 		 ipv);
838 
839 	spec->match_criteria_enable =
840 		get_match_criteria_enable(spec->match_criteria);
841 
842 	return 0;
843 }
844 
845 static int set_cnp_spec(struct mlx5_ib_dev *dev, u32 port_num,
846 			struct mlx5_flow_spec *spec)
847 {
848 	if (mlx5_core_mp_enabled(dev->mdev) &&
849 	    set_vhca_port_spec(dev, port_num, spec))
850 		return -EOPNOTSUPP;
851 
852 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
853 			 misc_parameters.bth_opcode);
854 	MLX5_SET(fte_match_param, spec->match_value, misc_parameters.bth_opcode,
855 		 IB_BTH_OPCODE_CNP);
856 
857 	spec->match_criteria_enable =
858 		get_match_criteria_enable(spec->match_criteria);
859 
860 	return 0;
861 }
862 
863 int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num,
864 			 struct mlx5_ib_op_fc *opfc,
865 			 enum mlx5_ib_optional_counter_type type)
866 {
867 	enum mlx5_flow_namespace_type fn_type;
868 	int priority, i, err, spec_num;
869 	struct mlx5_flow_act flow_act = {};
870 	struct mlx5_flow_destination dst;
871 	struct mlx5_flow_namespace *ns;
872 	struct mlx5_ib_flow_prio *prio;
873 	struct mlx5_flow_spec *spec;
874 
875 	spec = kcalloc(MAX_OPFC_RULES, sizeof(*spec), GFP_KERNEL);
876 	if (!spec)
877 		return -ENOMEM;
878 
879 	switch (type) {
880 	case MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS:
881 		if (set_ecn_ce_spec(dev, port_num, &spec[0],
882 				    MLX5_FS_IPV4_VERSION) ||
883 		    set_ecn_ce_spec(dev, port_num, &spec[1],
884 				    MLX5_FS_IPV6_VERSION)) {
885 			err = -EOPNOTSUPP;
886 			goto free;
887 		}
888 		spec_num = 2;
889 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS;
890 		priority = RDMA_RX_ECN_OPCOUNTER_PRIO;
891 		break;
892 
893 	case MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS:
894 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
895 					ft_field_support_2_nic_receive_rdma.bth_opcode) ||
896 		    set_cnp_spec(dev, port_num, &spec[0])) {
897 			err = -EOPNOTSUPP;
898 			goto free;
899 		}
900 		spec_num = 1;
901 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS;
902 		priority = RDMA_RX_CNP_OPCOUNTER_PRIO;
903 		break;
904 
905 	case MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS:
906 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
907 					ft_field_support_2_nic_transmit_rdma.bth_opcode) ||
908 		    set_cnp_spec(dev, port_num, &spec[0])) {
909 			err = -EOPNOTSUPP;
910 			goto free;
911 		}
912 		spec_num = 1;
913 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS;
914 		priority = RDMA_TX_CNP_OPCOUNTER_PRIO;
915 		break;
916 
917 	default:
918 		err = -EOPNOTSUPP;
919 		goto free;
920 	}
921 
922 	ns = mlx5_get_flow_namespace(dev->mdev, fn_type);
923 	if (!ns) {
924 		err = -EOPNOTSUPP;
925 		goto free;
926 	}
927 
928 	prio = &dev->flow_db->opfcs[type];
929 	if (!prio->flow_table) {
930 		prio = _get_prio(ns, prio, priority,
931 				 dev->num_ports * MAX_OPFC_RULES, 1, 0);
932 		if (IS_ERR(prio)) {
933 			err = PTR_ERR(prio);
934 			goto free;
935 		}
936 	}
937 
938 	dst.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
939 	dst.counter_id = mlx5_fc_id(opfc->fc);
940 
941 	flow_act.action =
942 		MLX5_FLOW_CONTEXT_ACTION_COUNT | MLX5_FLOW_CONTEXT_ACTION_ALLOW;
943 
944 	for (i = 0; i < spec_num; i++) {
945 		opfc->rule[i] = mlx5_add_flow_rules(prio->flow_table, &spec[i],
946 						    &flow_act, &dst, 1);
947 		if (IS_ERR(opfc->rule[i])) {
948 			err = PTR_ERR(opfc->rule[i]);
949 			goto del_rules;
950 		}
951 	}
952 	prio->refcount += spec_num;
953 	kfree(spec);
954 
955 	return 0;
956 
957 del_rules:
958 	for (i -= 1; i >= 0; i--)
959 		mlx5_del_flow_rules(opfc->rule[i]);
960 	put_flow_table(dev, prio, false);
961 free:
962 	kfree(spec);
963 	return err;
964 }
965 
966 void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev,
967 			     struct mlx5_ib_op_fc *opfc,
968 			     enum mlx5_ib_optional_counter_type type)
969 {
970 	int i;
971 
972 	for (i = 0; i < MAX_OPFC_RULES && opfc->rule[i]; i++) {
973 		mlx5_del_flow_rules(opfc->rule[i]);
974 		put_flow_table(dev, &dev->flow_db->opfcs[type], true);
975 	}
976 }
977 
978 static void set_underlay_qp(struct mlx5_ib_dev *dev,
979 			    struct mlx5_flow_spec *spec,
980 			    u32 underlay_qpn)
981 {
982 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param,
983 					   spec->match_criteria,
984 					   misc_parameters);
985 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
986 					   misc_parameters);
987 
988 	if (underlay_qpn &&
989 	    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
990 				      ft_field_support.bth_dst_qp)) {
991 		MLX5_SET(fte_match_set_misc,
992 			 misc_params_v, bth_dst_qp, underlay_qpn);
993 		MLX5_SET(fte_match_set_misc,
994 			 misc_params_c, bth_dst_qp, 0xffffff);
995 	}
996 }
997 
998 static void mlx5_ib_set_rule_source_port(struct mlx5_ib_dev *dev,
999 					 struct mlx5_flow_spec *spec,
1000 					 struct mlx5_eswitch_rep *rep)
1001 {
1002 	struct mlx5_eswitch *esw = dev->mdev->priv.eswitch;
1003 	void *misc;
1004 
1005 	if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1006 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1007 				    misc_parameters_2);
1008 
1009 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1010 			 mlx5_eswitch_get_vport_metadata_for_match(rep->esw,
1011 								   rep->vport));
1012 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1013 				    misc_parameters_2);
1014 
1015 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1016 			 mlx5_eswitch_get_vport_metadata_mask());
1017 	} else {
1018 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1019 				    misc_parameters);
1020 
1021 		MLX5_SET(fte_match_set_misc, misc, source_port, rep->vport);
1022 
1023 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1024 				    misc_parameters);
1025 
1026 		MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
1027 	}
1028 }
1029 
1030 static struct mlx5_ib_flow_handler *_create_flow_rule(struct mlx5_ib_dev *dev,
1031 						      struct mlx5_ib_flow_prio *ft_prio,
1032 						      const struct ib_flow_attr *flow_attr,
1033 						      struct mlx5_flow_destination *dst,
1034 						      u32 underlay_qpn,
1035 						      struct mlx5_ib_create_flow *ucmd)
1036 {
1037 	struct mlx5_flow_table	*ft = ft_prio->flow_table;
1038 	struct mlx5_ib_flow_handler *handler;
1039 	struct mlx5_flow_act flow_act = {};
1040 	struct mlx5_flow_spec *spec;
1041 	struct mlx5_flow_destination dest_arr[2] = {};
1042 	struct mlx5_flow_destination *rule_dst = dest_arr;
1043 	const void *ib_flow = (const void *)flow_attr + sizeof(*flow_attr);
1044 	unsigned int spec_index;
1045 	u32 prev_type = 0;
1046 	int err = 0;
1047 	int dest_num = 0;
1048 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
1049 
1050 	if (!is_valid_attr(dev->mdev, flow_attr))
1051 		return ERR_PTR(-EINVAL);
1052 
1053 	if (dev->is_rep && is_egress)
1054 		return ERR_PTR(-EINVAL);
1055 
1056 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1057 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
1058 	if (!handler || !spec) {
1059 		err = -ENOMEM;
1060 		goto free;
1061 	}
1062 
1063 	INIT_LIST_HEAD(&handler->list);
1064 
1065 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
1066 		err = parse_flow_attr(dev->mdev, spec,
1067 				      ib_flow, flow_attr, &flow_act,
1068 				      prev_type);
1069 		if (err < 0)
1070 			goto free;
1071 
1072 		prev_type = ((union ib_flow_spec *)ib_flow)->type;
1073 		ib_flow += ((union ib_flow_spec *)ib_flow)->size;
1074 	}
1075 
1076 	if (dst && !(flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP)) {
1077 		memcpy(&dest_arr[0], dst, sizeof(*dst));
1078 		dest_num++;
1079 	}
1080 
1081 	if (!flow_is_multicast_only(flow_attr))
1082 		set_underlay_qp(dev, spec, underlay_qpn);
1083 
1084 	if (dev->is_rep && flow_attr->type != IB_FLOW_ATTR_SNIFFER) {
1085 		struct mlx5_eswitch_rep *rep;
1086 
1087 		rep = dev->port[flow_attr->port - 1].rep;
1088 		if (!rep) {
1089 			err = -EINVAL;
1090 			goto free;
1091 		}
1092 
1093 		mlx5_ib_set_rule_source_port(dev, spec, rep);
1094 	}
1095 
1096 	spec->match_criteria_enable = get_match_criteria_enable(spec->match_criteria);
1097 
1098 	if (is_egress) {
1099 		err = -EINVAL;
1100 		goto free;
1101 	}
1102 
1103 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1104 		struct mlx5_ib_mcounters *mcounters;
1105 
1106 		err = mlx5_ib_flow_counters_set_data(flow_act.counters, ucmd);
1107 		if (err)
1108 			goto free;
1109 
1110 		mcounters = to_mcounters(flow_act.counters);
1111 		handler->ibcounters = flow_act.counters;
1112 		dest_arr[dest_num].type =
1113 			MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1114 		dest_arr[dest_num].counter_id =
1115 			mlx5_fc_id(mcounters->hw_cntrs_hndl);
1116 		dest_num++;
1117 	}
1118 
1119 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP) {
1120 		if (!dest_num)
1121 			rule_dst = NULL;
1122 	} else {
1123 		if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1124 			flow_act.action |=
1125 				MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1126 		if (is_egress)
1127 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1128 		else if (dest_num)
1129 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1130 	}
1131 
1132 	if ((spec->flow_context.flags & FLOW_CONTEXT_HAS_TAG)  &&
1133 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1134 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1135 		mlx5_ib_warn(dev, "Flow tag %u and attribute type %x isn't allowed in leftovers\n",
1136 			     spec->flow_context.flow_tag, flow_attr->type);
1137 		err = -EINVAL;
1138 		goto free;
1139 	}
1140 	handler->rule = mlx5_add_flow_rules(ft, spec,
1141 					    &flow_act,
1142 					    rule_dst, dest_num);
1143 
1144 	if (IS_ERR(handler->rule)) {
1145 		err = PTR_ERR(handler->rule);
1146 		goto free;
1147 	}
1148 
1149 	ft_prio->refcount++;
1150 	handler->prio = ft_prio;
1151 	handler->dev = dev;
1152 
1153 	ft_prio->flow_table = ft;
1154 free:
1155 	if (err && handler) {
1156 		mlx5_ib_counters_clear_description(handler->ibcounters);
1157 		kfree(handler);
1158 	}
1159 	kvfree(spec);
1160 	return err ? ERR_PTR(err) : handler;
1161 }
1162 
1163 static struct mlx5_ib_flow_handler *create_flow_rule(struct mlx5_ib_dev *dev,
1164 						     struct mlx5_ib_flow_prio *ft_prio,
1165 						     const struct ib_flow_attr *flow_attr,
1166 						     struct mlx5_flow_destination *dst)
1167 {
1168 	return _create_flow_rule(dev, ft_prio, flow_attr, dst, 0, NULL);
1169 }
1170 
1171 enum {
1172 	LEFTOVERS_MC,
1173 	LEFTOVERS_UC,
1174 };
1175 
1176 static struct mlx5_ib_flow_handler *create_leftovers_rule(struct mlx5_ib_dev *dev,
1177 							  struct mlx5_ib_flow_prio *ft_prio,
1178 							  struct ib_flow_attr *flow_attr,
1179 							  struct mlx5_flow_destination *dst)
1180 {
1181 	struct mlx5_ib_flow_handler *handler_ucast = NULL;
1182 	struct mlx5_ib_flow_handler *handler = NULL;
1183 
1184 	static struct {
1185 		struct ib_flow_attr	flow_attr;
1186 		struct ib_flow_spec_eth eth_flow;
1187 	} leftovers_specs[] = {
1188 		[LEFTOVERS_MC] = {
1189 			.flow_attr = {
1190 				.num_of_specs = 1,
1191 				.size = sizeof(leftovers_specs[0])
1192 			},
1193 			.eth_flow = {
1194 				.type = IB_FLOW_SPEC_ETH,
1195 				.size = sizeof(struct ib_flow_spec_eth),
1196 				.mask = {.dst_mac = {0x1} },
1197 				.val =  {.dst_mac = {0x1} }
1198 			}
1199 		},
1200 		[LEFTOVERS_UC] = {
1201 			.flow_attr = {
1202 				.num_of_specs = 1,
1203 				.size = sizeof(leftovers_specs[0])
1204 			},
1205 			.eth_flow = {
1206 				.type = IB_FLOW_SPEC_ETH,
1207 				.size = sizeof(struct ib_flow_spec_eth),
1208 				.mask = {.dst_mac = {0x1} },
1209 				.val = {.dst_mac = {} }
1210 			}
1211 		}
1212 	};
1213 
1214 	handler = create_flow_rule(dev, ft_prio,
1215 				   &leftovers_specs[LEFTOVERS_MC].flow_attr,
1216 				   dst);
1217 	if (!IS_ERR(handler) &&
1218 	    flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT) {
1219 		handler_ucast = create_flow_rule(dev, ft_prio,
1220 						 &leftovers_specs[LEFTOVERS_UC].flow_attr,
1221 						 dst);
1222 		if (IS_ERR(handler_ucast)) {
1223 			mlx5_del_flow_rules(handler->rule);
1224 			ft_prio->refcount--;
1225 			kfree(handler);
1226 			handler = handler_ucast;
1227 		} else {
1228 			list_add(&handler_ucast->list, &handler->list);
1229 		}
1230 	}
1231 
1232 	return handler;
1233 }
1234 
1235 static struct mlx5_ib_flow_handler *create_sniffer_rule(struct mlx5_ib_dev *dev,
1236 							struct mlx5_ib_flow_prio *ft_rx,
1237 							struct mlx5_ib_flow_prio *ft_tx,
1238 							struct mlx5_flow_destination *dst)
1239 {
1240 	struct mlx5_ib_flow_handler *handler_rx;
1241 	struct mlx5_ib_flow_handler *handler_tx;
1242 	int err;
1243 	static const struct ib_flow_attr flow_attr  = {
1244 		.num_of_specs = 0,
1245 		.type = IB_FLOW_ATTR_SNIFFER,
1246 		.size = sizeof(flow_attr)
1247 	};
1248 
1249 	handler_rx = create_flow_rule(dev, ft_rx, &flow_attr, dst);
1250 	if (IS_ERR(handler_rx)) {
1251 		err = PTR_ERR(handler_rx);
1252 		goto err;
1253 	}
1254 
1255 	handler_tx = create_flow_rule(dev, ft_tx, &flow_attr, dst);
1256 	if (IS_ERR(handler_tx)) {
1257 		err = PTR_ERR(handler_tx);
1258 		goto err_tx;
1259 	}
1260 
1261 	list_add(&handler_tx->list, &handler_rx->list);
1262 
1263 	return handler_rx;
1264 
1265 err_tx:
1266 	mlx5_del_flow_rules(handler_rx->rule);
1267 	ft_rx->refcount--;
1268 	kfree(handler_rx);
1269 err:
1270 	return ERR_PTR(err);
1271 }
1272 
1273 static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
1274 					   struct ib_flow_attr *flow_attr,
1275 					   struct ib_udata *udata)
1276 {
1277 	struct mlx5_ib_dev *dev = to_mdev(qp->device);
1278 	struct mlx5_ib_qp *mqp = to_mqp(qp);
1279 	struct mlx5_ib_flow_handler *handler = NULL;
1280 	struct mlx5_flow_destination *dst = NULL;
1281 	struct mlx5_ib_flow_prio *ft_prio_tx = NULL;
1282 	struct mlx5_ib_flow_prio *ft_prio;
1283 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
1284 	struct mlx5_ib_create_flow *ucmd = NULL, ucmd_hdr;
1285 	size_t min_ucmd_sz, required_ucmd_sz;
1286 	int err;
1287 	int underlay_qpn;
1288 
1289 	if (udata && udata->inlen) {
1290 		min_ucmd_sz = offsetofend(struct mlx5_ib_create_flow, reserved);
1291 		if (udata->inlen < min_ucmd_sz)
1292 			return ERR_PTR(-EOPNOTSUPP);
1293 
1294 		err = ib_copy_from_udata(&ucmd_hdr, udata, min_ucmd_sz);
1295 		if (err)
1296 			return ERR_PTR(err);
1297 
1298 		/* currently supports only one counters data */
1299 		if (ucmd_hdr.ncounters_data > 1)
1300 			return ERR_PTR(-EINVAL);
1301 
1302 		required_ucmd_sz = min_ucmd_sz +
1303 			sizeof(struct mlx5_ib_flow_counters_data) *
1304 			ucmd_hdr.ncounters_data;
1305 		if (udata->inlen > required_ucmd_sz &&
1306 		    !ib_is_udata_cleared(udata, required_ucmd_sz,
1307 					 udata->inlen - required_ucmd_sz))
1308 			return ERR_PTR(-EOPNOTSUPP);
1309 
1310 		ucmd = kzalloc(required_ucmd_sz, GFP_KERNEL);
1311 		if (!ucmd)
1312 			return ERR_PTR(-ENOMEM);
1313 
1314 		err = ib_copy_from_udata(ucmd, udata, required_ucmd_sz);
1315 		if (err)
1316 			goto free_ucmd;
1317 	}
1318 
1319 	if (flow_attr->priority > MLX5_IB_FLOW_LAST_PRIO) {
1320 		err = -ENOMEM;
1321 		goto free_ucmd;
1322 	}
1323 
1324 	if (flow_attr->flags &
1325 	    ~(IB_FLOW_ATTR_FLAGS_DONT_TRAP | IB_FLOW_ATTR_FLAGS_EGRESS)) {
1326 		err = -EINVAL;
1327 		goto free_ucmd;
1328 	}
1329 
1330 	if (is_egress &&
1331 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1332 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1333 		err = -EINVAL;
1334 		goto free_ucmd;
1335 	}
1336 
1337 	dst = kzalloc(sizeof(*dst), GFP_KERNEL);
1338 	if (!dst) {
1339 		err = -ENOMEM;
1340 		goto free_ucmd;
1341 	}
1342 
1343 	mutex_lock(&dev->flow_db->lock);
1344 
1345 	ft_prio = get_flow_table(dev, flow_attr,
1346 				 is_egress ? MLX5_IB_FT_TX : MLX5_IB_FT_RX);
1347 	if (IS_ERR(ft_prio)) {
1348 		err = PTR_ERR(ft_prio);
1349 		goto unlock;
1350 	}
1351 	if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
1352 		ft_prio_tx = get_flow_table(dev, flow_attr, MLX5_IB_FT_TX);
1353 		if (IS_ERR(ft_prio_tx)) {
1354 			err = PTR_ERR(ft_prio_tx);
1355 			ft_prio_tx = NULL;
1356 			goto destroy_ft;
1357 		}
1358 	}
1359 
1360 	if (is_egress) {
1361 		dst->type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1362 	} else {
1363 		dst->type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1364 		if (mqp->is_rss)
1365 			dst->tir_num = mqp->rss_qp.tirn;
1366 		else
1367 			dst->tir_num = mqp->raw_packet_qp.rq.tirn;
1368 	}
1369 
1370 	switch (flow_attr->type) {
1371 	case IB_FLOW_ATTR_NORMAL:
1372 		underlay_qpn = (mqp->flags & IB_QP_CREATE_SOURCE_QPN) ?
1373 				       mqp->underlay_qpn :
1374 				       0;
1375 		handler = _create_flow_rule(dev, ft_prio, flow_attr, dst,
1376 					    underlay_qpn, ucmd);
1377 		break;
1378 	case IB_FLOW_ATTR_ALL_DEFAULT:
1379 	case IB_FLOW_ATTR_MC_DEFAULT:
1380 		handler = create_leftovers_rule(dev, ft_prio, flow_attr, dst);
1381 		break;
1382 	case IB_FLOW_ATTR_SNIFFER:
1383 		handler = create_sniffer_rule(dev, ft_prio, ft_prio_tx, dst);
1384 		break;
1385 	default:
1386 		err = -EINVAL;
1387 		goto destroy_ft;
1388 	}
1389 
1390 	if (IS_ERR(handler)) {
1391 		err = PTR_ERR(handler);
1392 		handler = NULL;
1393 		goto destroy_ft;
1394 	}
1395 
1396 	mutex_unlock(&dev->flow_db->lock);
1397 	kfree(dst);
1398 	kfree(ucmd);
1399 
1400 	return &handler->ibflow;
1401 
1402 destroy_ft:
1403 	put_flow_table(dev, ft_prio, false);
1404 	if (ft_prio_tx)
1405 		put_flow_table(dev, ft_prio_tx, false);
1406 unlock:
1407 	mutex_unlock(&dev->flow_db->lock);
1408 	kfree(dst);
1409 free_ucmd:
1410 	kfree(ucmd);
1411 	return ERR_PTR(err);
1412 }
1413 
1414 static struct mlx5_ib_flow_prio *
1415 _get_flow_table(struct mlx5_ib_dev *dev,
1416 		struct mlx5_ib_flow_matcher *fs_matcher,
1417 		bool mcast)
1418 {
1419 	struct mlx5_flow_namespace *ns = NULL;
1420 	struct mlx5_ib_flow_prio *prio = NULL;
1421 	int max_table_size = 0;
1422 	bool esw_encap;
1423 	u32 flags = 0;
1424 	int priority;
1425 
1426 	if (mcast)
1427 		priority = MLX5_IB_FLOW_MCAST_PRIO;
1428 	else
1429 		priority = ib_prio_to_core_prio(fs_matcher->priority, false);
1430 
1431 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
1432 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
1433 	switch (fs_matcher->ns_type) {
1434 	case MLX5_FLOW_NAMESPACE_BYPASS:
1435 		max_table_size = BIT(
1436 			MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, log_max_ft_size));
1437 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap) && !esw_encap)
1438 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1439 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
1440 					      reformat_l3_tunnel_to_l2) &&
1441 		    !esw_encap)
1442 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1443 		break;
1444 	case MLX5_FLOW_NAMESPACE_EGRESS:
1445 		max_table_size = BIT(
1446 			MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, log_max_ft_size));
1447 		if (MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat) &&
1448 		    !esw_encap)
1449 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1450 		break;
1451 	case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
1452 		max_table_size = BIT(
1453 			MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, log_max_ft_size));
1454 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, decap) && esw_encap)
1455 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1456 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev,
1457 					       reformat_l3_tunnel_to_l2) &&
1458 		    esw_encap)
1459 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1460 		priority = fs_matcher->priority;
1461 		break;
1462 	case MLX5_FLOW_NAMESPACE_RDMA_RX:
1463 		max_table_size = BIT(
1464 			MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev, log_max_ft_size));
1465 		priority = fs_matcher->priority;
1466 		break;
1467 	case MLX5_FLOW_NAMESPACE_RDMA_TX:
1468 		max_table_size = BIT(
1469 			MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev, log_max_ft_size));
1470 		priority = fs_matcher->priority;
1471 		break;
1472 	default:
1473 		break;
1474 	}
1475 
1476 	max_table_size = min_t(int, max_table_size, MLX5_FS_MAX_ENTRIES);
1477 
1478 	ns = mlx5_get_flow_namespace(dev->mdev, fs_matcher->ns_type);
1479 	if (!ns)
1480 		return ERR_PTR(-EOPNOTSUPP);
1481 
1482 	switch (fs_matcher->ns_type) {
1483 	case MLX5_FLOW_NAMESPACE_BYPASS:
1484 		prio = &dev->flow_db->prios[priority];
1485 		break;
1486 	case MLX5_FLOW_NAMESPACE_EGRESS:
1487 		prio = &dev->flow_db->egress_prios[priority];
1488 		break;
1489 	case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
1490 		prio = &dev->flow_db->fdb[priority];
1491 		break;
1492 	case MLX5_FLOW_NAMESPACE_RDMA_RX:
1493 		prio = &dev->flow_db->rdma_rx[priority];
1494 		break;
1495 	case MLX5_FLOW_NAMESPACE_RDMA_TX:
1496 		prio = &dev->flow_db->rdma_tx[priority];
1497 		break;
1498 	default: return ERR_PTR(-EINVAL);
1499 	}
1500 
1501 	if (!prio)
1502 		return ERR_PTR(-EINVAL);
1503 
1504 	if (prio->flow_table)
1505 		return prio;
1506 
1507 	return _get_prio(ns, prio, priority, max_table_size,
1508 			 MLX5_FS_MAX_TYPES, flags);
1509 }
1510 
1511 static struct mlx5_ib_flow_handler *
1512 _create_raw_flow_rule(struct mlx5_ib_dev *dev,
1513 		      struct mlx5_ib_flow_prio *ft_prio,
1514 		      struct mlx5_flow_destination *dst,
1515 		      struct mlx5_ib_flow_matcher  *fs_matcher,
1516 		      struct mlx5_flow_context *flow_context,
1517 		      struct mlx5_flow_act *flow_act,
1518 		      void *cmd_in, int inlen,
1519 		      int dst_num)
1520 {
1521 	struct mlx5_ib_flow_handler *handler;
1522 	struct mlx5_flow_spec *spec;
1523 	struct mlx5_flow_table *ft = ft_prio->flow_table;
1524 	int err = 0;
1525 
1526 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1527 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
1528 	if (!handler || !spec) {
1529 		err = -ENOMEM;
1530 		goto free;
1531 	}
1532 
1533 	INIT_LIST_HEAD(&handler->list);
1534 
1535 	memcpy(spec->match_value, cmd_in, inlen);
1536 	memcpy(spec->match_criteria, fs_matcher->matcher_mask.match_params,
1537 	       fs_matcher->mask_len);
1538 	spec->match_criteria_enable = fs_matcher->match_criteria_enable;
1539 	spec->flow_context = *flow_context;
1540 
1541 	handler->rule = mlx5_add_flow_rules(ft, spec,
1542 					    flow_act, dst, dst_num);
1543 
1544 	if (IS_ERR(handler->rule)) {
1545 		err = PTR_ERR(handler->rule);
1546 		goto free;
1547 	}
1548 
1549 	ft_prio->refcount++;
1550 	handler->prio = ft_prio;
1551 	handler->dev = dev;
1552 	ft_prio->flow_table = ft;
1553 
1554 free:
1555 	if (err)
1556 		kfree(handler);
1557 	kvfree(spec);
1558 	return err ? ERR_PTR(err) : handler;
1559 }
1560 
1561 static bool raw_fs_is_multicast(struct mlx5_ib_flow_matcher *fs_matcher,
1562 				void *match_v)
1563 {
1564 	void *match_c;
1565 	void *match_v_set_lyr_2_4, *match_c_set_lyr_2_4;
1566 	void *dmac, *dmac_mask;
1567 	void *ipv4, *ipv4_mask;
1568 
1569 	if (!(fs_matcher->match_criteria_enable &
1570 	      (1 << MATCH_CRITERIA_ENABLE_OUTER_BIT)))
1571 		return false;
1572 
1573 	match_c = fs_matcher->matcher_mask.match_params;
1574 	match_v_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_v,
1575 					   outer_headers);
1576 	match_c_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_c,
1577 					   outer_headers);
1578 
1579 	dmac = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1580 			    dmac_47_16);
1581 	dmac_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1582 				 dmac_47_16);
1583 
1584 	if (is_multicast_ether_addr(dmac) &&
1585 	    is_multicast_ether_addr(dmac_mask))
1586 		return true;
1587 
1588 	ipv4 = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1589 			    dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1590 
1591 	ipv4_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1592 				 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1593 
1594 	if (ipv4_is_multicast(*(__be32 *)(ipv4)) &&
1595 	    ipv4_is_multicast(*(__be32 *)(ipv4_mask)))
1596 		return true;
1597 
1598 	return false;
1599 }
1600 
1601 static struct mlx5_ib_flow_handler *raw_fs_rule_add(
1602 	struct mlx5_ib_dev *dev, struct mlx5_ib_flow_matcher *fs_matcher,
1603 	struct mlx5_flow_context *flow_context, struct mlx5_flow_act *flow_act,
1604 	u32 counter_id, void *cmd_in, int inlen, int dest_id, int dest_type)
1605 {
1606 	struct mlx5_flow_destination *dst;
1607 	struct mlx5_ib_flow_prio *ft_prio;
1608 	struct mlx5_ib_flow_handler *handler;
1609 	int dst_num = 0;
1610 	bool mcast;
1611 	int err;
1612 
1613 	if (fs_matcher->flow_type != MLX5_IB_FLOW_TYPE_NORMAL)
1614 		return ERR_PTR(-EOPNOTSUPP);
1615 
1616 	if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO)
1617 		return ERR_PTR(-ENOMEM);
1618 
1619 	dst = kcalloc(2, sizeof(*dst), GFP_KERNEL);
1620 	if (!dst)
1621 		return ERR_PTR(-ENOMEM);
1622 
1623 	mcast = raw_fs_is_multicast(fs_matcher, cmd_in);
1624 	mutex_lock(&dev->flow_db->lock);
1625 
1626 	ft_prio = _get_flow_table(dev, fs_matcher, mcast);
1627 	if (IS_ERR(ft_prio)) {
1628 		err = PTR_ERR(ft_prio);
1629 		goto unlock;
1630 	}
1631 
1632 	switch (dest_type) {
1633 	case MLX5_FLOW_DESTINATION_TYPE_TIR:
1634 		dst[dst_num].type = dest_type;
1635 		dst[dst_num++].tir_num = dest_id;
1636 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1637 		break;
1638 	case MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE:
1639 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM;
1640 		dst[dst_num++].ft_num = dest_id;
1641 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1642 		break;
1643 	case MLX5_FLOW_DESTINATION_TYPE_PORT:
1644 		dst[dst_num++].type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1645 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1646 		break;
1647 	default:
1648 		break;
1649 	}
1650 
1651 	if (flow_act->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1652 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1653 		dst[dst_num].counter_id = counter_id;
1654 		dst_num++;
1655 	}
1656 
1657 	handler = _create_raw_flow_rule(dev, ft_prio, dst_num ? dst : NULL,
1658 					fs_matcher, flow_context, flow_act,
1659 					cmd_in, inlen, dst_num);
1660 
1661 	if (IS_ERR(handler)) {
1662 		err = PTR_ERR(handler);
1663 		goto destroy_ft;
1664 	}
1665 
1666 	mutex_unlock(&dev->flow_db->lock);
1667 	atomic_inc(&fs_matcher->usecnt);
1668 	handler->flow_matcher = fs_matcher;
1669 
1670 	kfree(dst);
1671 
1672 	return handler;
1673 
1674 destroy_ft:
1675 	put_flow_table(dev, ft_prio, false);
1676 unlock:
1677 	mutex_unlock(&dev->flow_db->lock);
1678 	kfree(dst);
1679 
1680 	return ERR_PTR(err);
1681 }
1682 
1683 static void destroy_flow_action_raw(struct mlx5_ib_flow_action *maction)
1684 {
1685 	switch (maction->flow_action_raw.sub_type) {
1686 	case MLX5_IB_FLOW_ACTION_MODIFY_HEADER:
1687 		mlx5_modify_header_dealloc(maction->flow_action_raw.dev->mdev,
1688 					   maction->flow_action_raw.modify_hdr);
1689 		break;
1690 	case MLX5_IB_FLOW_ACTION_PACKET_REFORMAT:
1691 		mlx5_packet_reformat_dealloc(maction->flow_action_raw.dev->mdev,
1692 					     maction->flow_action_raw.pkt_reformat);
1693 		break;
1694 	case MLX5_IB_FLOW_ACTION_DECAP:
1695 		break;
1696 	default:
1697 		break;
1698 	}
1699 }
1700 
1701 static int mlx5_ib_destroy_flow_action(struct ib_flow_action *action)
1702 {
1703 	struct mlx5_ib_flow_action *maction = to_mflow_act(action);
1704 
1705 	switch (action->type) {
1706 	case IB_FLOW_ACTION_UNSPECIFIED:
1707 		destroy_flow_action_raw(maction);
1708 		break;
1709 	default:
1710 		WARN_ON(true);
1711 		break;
1712 	}
1713 
1714 	kfree(maction);
1715 	return 0;
1716 }
1717 
1718 static int
1719 mlx5_ib_ft_type_to_namespace(enum mlx5_ib_uapi_flow_table_type table_type,
1720 			     enum mlx5_flow_namespace_type *namespace)
1721 {
1722 	switch (table_type) {
1723 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX:
1724 		*namespace = MLX5_FLOW_NAMESPACE_BYPASS;
1725 		break;
1726 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX:
1727 		*namespace = MLX5_FLOW_NAMESPACE_EGRESS;
1728 		break;
1729 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_FDB:
1730 		*namespace = MLX5_FLOW_NAMESPACE_FDB_BYPASS;
1731 		break;
1732 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_RX:
1733 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_RX;
1734 		break;
1735 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_TX:
1736 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_TX;
1737 		break;
1738 	default:
1739 		return -EINVAL;
1740 	}
1741 
1742 	return 0;
1743 }
1744 
1745 static const struct uverbs_attr_spec mlx5_ib_flow_type[] = {
1746 	[MLX5_IB_FLOW_TYPE_NORMAL] = {
1747 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1748 		.u.ptr = {
1749 			.len = sizeof(u16), /* data is priority */
1750 			.min_len = sizeof(u16),
1751 		}
1752 	},
1753 	[MLX5_IB_FLOW_TYPE_SNIFFER] = {
1754 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1755 		UVERBS_ATTR_NO_DATA(),
1756 	},
1757 	[MLX5_IB_FLOW_TYPE_ALL_DEFAULT] = {
1758 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1759 		UVERBS_ATTR_NO_DATA(),
1760 	},
1761 	[MLX5_IB_FLOW_TYPE_MC_DEFAULT] = {
1762 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1763 		UVERBS_ATTR_NO_DATA(),
1764 	},
1765 };
1766 
1767 static bool is_flow_dest(void *obj, int *dest_id, int *dest_type)
1768 {
1769 	struct devx_obj *devx_obj = obj;
1770 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1771 
1772 	switch (opcode) {
1773 	case MLX5_CMD_OP_DESTROY_TIR:
1774 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1775 		*dest_id = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox,
1776 				    obj_id);
1777 		return true;
1778 
1779 	case MLX5_CMD_OP_DESTROY_FLOW_TABLE:
1780 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1781 		*dest_id = MLX5_GET(destroy_flow_table_in, devx_obj->dinbox,
1782 				    table_id);
1783 		return true;
1784 	default:
1785 		return false;
1786 	}
1787 }
1788 
1789 static int get_dests(struct uverbs_attr_bundle *attrs,
1790 		     struct mlx5_ib_flow_matcher *fs_matcher, int *dest_id,
1791 		     int *dest_type, struct ib_qp **qp, u32 *flags)
1792 {
1793 	bool dest_devx, dest_qp;
1794 	void *devx_obj;
1795 	int err;
1796 
1797 	dest_devx = uverbs_attr_is_valid(attrs,
1798 					 MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1799 	dest_qp = uverbs_attr_is_valid(attrs,
1800 				       MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1801 
1802 	*flags = 0;
1803 	err = uverbs_get_flags32(flags, attrs, MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
1804 				 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS |
1805 					 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP);
1806 	if (err)
1807 		return err;
1808 
1809 	/* Both flags are not allowed */
1810 	if (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS &&
1811 	    *flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1812 		return -EINVAL;
1813 
1814 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_BYPASS) {
1815 		if (dest_devx && (dest_qp || *flags))
1816 			return -EINVAL;
1817 		else if (dest_qp && *flags)
1818 			return -EINVAL;
1819 	}
1820 
1821 	/* Allow only DEVX object, drop as dest for FDB */
1822 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS &&
1823 	    !(dest_devx || (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)))
1824 		return -EINVAL;
1825 
1826 	/* Allow only DEVX object or QP as dest when inserting to RDMA_RX */
1827 	if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1828 	    ((!dest_devx && !dest_qp) || (dest_devx && dest_qp)))
1829 		return -EINVAL;
1830 
1831 	*qp = NULL;
1832 	if (dest_devx) {
1833 		devx_obj =
1834 			uverbs_attr_get_obj(attrs,
1835 					    MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1836 
1837 		/* Verify that the given DEVX object is a flow
1838 		 * steering destination.
1839 		 */
1840 		if (!is_flow_dest(devx_obj, dest_id, dest_type))
1841 			return -EINVAL;
1842 		/* Allow only flow table as dest when inserting to FDB or RDMA_RX */
1843 		if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS ||
1844 		     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1845 		    *dest_type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1846 			return -EINVAL;
1847 	} else if (dest_qp) {
1848 		struct mlx5_ib_qp *mqp;
1849 
1850 		*qp = uverbs_attr_get_obj(attrs,
1851 					  MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1852 		if (IS_ERR(*qp))
1853 			return PTR_ERR(*qp);
1854 
1855 		if ((*qp)->qp_type != IB_QPT_RAW_PACKET)
1856 			return -EINVAL;
1857 
1858 		mqp = to_mqp(*qp);
1859 		if (mqp->is_rss)
1860 			*dest_id = mqp->rss_qp.tirn;
1861 		else
1862 			*dest_id = mqp->raw_packet_qp.rq.tirn;
1863 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1864 	} else if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1865 		    fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX) &&
1866 		   !(*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)) {
1867 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1868 	}
1869 
1870 	if (*dest_type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1871 	    (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1872 	     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX))
1873 		return -EINVAL;
1874 
1875 	return 0;
1876 }
1877 
1878 static bool is_flow_counter(void *obj, u32 offset, u32 *counter_id)
1879 {
1880 	struct devx_obj *devx_obj = obj;
1881 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1882 
1883 	if (opcode == MLX5_CMD_OP_DEALLOC_FLOW_COUNTER) {
1884 
1885 		if (offset && offset >= devx_obj->flow_counter_bulk_size)
1886 			return false;
1887 
1888 		*counter_id = MLX5_GET(dealloc_flow_counter_in,
1889 				       devx_obj->dinbox,
1890 				       flow_counter_id);
1891 		*counter_id += offset;
1892 		return true;
1893 	}
1894 
1895 	return false;
1896 }
1897 
1898 #define MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS 2
1899 static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)(
1900 	struct uverbs_attr_bundle *attrs)
1901 {
1902 	struct mlx5_flow_context flow_context = {.flow_tag =
1903 		MLX5_FS_DEFAULT_FLOW_TAG};
1904 	u32 *offset_attr, offset = 0, counter_id = 0;
1905 	int dest_id, dest_type = -1, inlen, len, ret, i;
1906 	struct mlx5_ib_flow_handler *flow_handler;
1907 	struct mlx5_ib_flow_matcher *fs_matcher;
1908 	struct ib_uobject **arr_flow_actions;
1909 	struct ib_uflow_resources *uflow_res;
1910 	struct mlx5_flow_act flow_act = {};
1911 	struct ib_qp *qp = NULL;
1912 	void *devx_obj, *cmd_in;
1913 	struct ib_uobject *uobj;
1914 	struct mlx5_ib_dev *dev;
1915 	u32 flags;
1916 
1917 	if (!capable(CAP_NET_RAW))
1918 		return -EPERM;
1919 
1920 	fs_matcher = uverbs_attr_get_obj(attrs,
1921 					 MLX5_IB_ATTR_CREATE_FLOW_MATCHER);
1922 	uobj =  uverbs_attr_get_uobject(attrs, MLX5_IB_ATTR_CREATE_FLOW_HANDLE);
1923 	dev = mlx5_udata_to_mdev(&attrs->driver_udata);
1924 
1925 	if (get_dests(attrs, fs_matcher, &dest_id, &dest_type, &qp, &flags))
1926 		return -EINVAL;
1927 
1928 	if (flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS)
1929 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
1930 
1931 	if (flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1932 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1933 
1934 	len = uverbs_attr_get_uobjs_arr(attrs,
1935 		MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX, &arr_flow_actions);
1936 	if (len) {
1937 		devx_obj = arr_flow_actions[0]->object;
1938 
1939 		if (uverbs_attr_is_valid(attrs,
1940 					 MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET)) {
1941 
1942 			int num_offsets = uverbs_attr_ptr_get_array_size(
1943 				attrs,
1944 				MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
1945 				sizeof(u32));
1946 
1947 			if (num_offsets != 1)
1948 				return -EINVAL;
1949 
1950 			offset_attr = uverbs_attr_get_alloced_ptr(
1951 				attrs,
1952 				MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET);
1953 			offset = *offset_attr;
1954 		}
1955 
1956 		if (!is_flow_counter(devx_obj, offset, &counter_id))
1957 			return -EINVAL;
1958 
1959 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
1960 	}
1961 
1962 	cmd_in = uverbs_attr_get_alloced_ptr(
1963 		attrs, MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE);
1964 	inlen = uverbs_attr_get_len(attrs,
1965 				    MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE);
1966 
1967 	uflow_res = flow_resources_alloc(MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS);
1968 	if (!uflow_res)
1969 		return -ENOMEM;
1970 
1971 	len = uverbs_attr_get_uobjs_arr(attrs,
1972 		MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS, &arr_flow_actions);
1973 	for (i = 0; i < len; i++) {
1974 		struct mlx5_ib_flow_action *maction =
1975 			to_mflow_act(arr_flow_actions[i]->object);
1976 
1977 		ret = parse_flow_flow_action(maction, false, &flow_act);
1978 		if (ret)
1979 			goto err_out;
1980 		flow_resources_add(uflow_res, IB_FLOW_SPEC_ACTION_HANDLE,
1981 				   arr_flow_actions[i]->object);
1982 	}
1983 
1984 	ret = uverbs_copy_from(&flow_context.flow_tag, attrs,
1985 			       MLX5_IB_ATTR_CREATE_FLOW_TAG);
1986 	if (!ret) {
1987 		if (flow_context.flow_tag >= BIT(24)) {
1988 			ret = -EINVAL;
1989 			goto err_out;
1990 		}
1991 		flow_context.flags |= FLOW_CONTEXT_HAS_TAG;
1992 	}
1993 
1994 	flow_handler =
1995 		raw_fs_rule_add(dev, fs_matcher, &flow_context, &flow_act,
1996 				counter_id, cmd_in, inlen, dest_id, dest_type);
1997 	if (IS_ERR(flow_handler)) {
1998 		ret = PTR_ERR(flow_handler);
1999 		goto err_out;
2000 	}
2001 
2002 	ib_set_flow(uobj, &flow_handler->ibflow, qp, &dev->ib_dev, uflow_res);
2003 
2004 	return 0;
2005 err_out:
2006 	ib_uverbs_flow_resources_free(uflow_res);
2007 	return ret;
2008 }
2009 
2010 static int flow_matcher_cleanup(struct ib_uobject *uobject,
2011 				enum rdma_remove_reason why,
2012 				struct uverbs_attr_bundle *attrs)
2013 {
2014 	struct mlx5_ib_flow_matcher *obj = uobject->object;
2015 
2016 	if (atomic_read(&obj->usecnt))
2017 		return -EBUSY;
2018 
2019 	kfree(obj);
2020 	return 0;
2021 }
2022 
2023 static int mlx5_ib_matcher_ns(struct uverbs_attr_bundle *attrs,
2024 			      struct mlx5_ib_flow_matcher *obj)
2025 {
2026 	enum mlx5_ib_uapi_flow_table_type ft_type =
2027 		MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX;
2028 	u32 flags;
2029 	int err;
2030 
2031 	/* New users should use MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE and older
2032 	 * users should switch to it. We leave this to not break userspace
2033 	 */
2034 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE) &&
2035 	    uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS))
2036 		return -EINVAL;
2037 
2038 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE)) {
2039 		err = uverbs_get_const(&ft_type, attrs,
2040 				       MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE);
2041 		if (err)
2042 			return err;
2043 
2044 		err = mlx5_ib_ft_type_to_namespace(ft_type, &obj->ns_type);
2045 		if (err)
2046 			return err;
2047 
2048 		return 0;
2049 	}
2050 
2051 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS)) {
2052 		err = uverbs_get_flags32(&flags, attrs,
2053 					 MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2054 					 IB_FLOW_ATTR_FLAGS_EGRESS);
2055 		if (err)
2056 			return err;
2057 
2058 		if (flags) {
2059 			mlx5_ib_ft_type_to_namespace(
2060 				MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX,
2061 				&obj->ns_type);
2062 			return 0;
2063 		}
2064 	}
2065 
2066 	obj->ns_type = MLX5_FLOW_NAMESPACE_BYPASS;
2067 
2068 	return 0;
2069 }
2070 
2071 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_MATCHER_CREATE)(
2072 	struct uverbs_attr_bundle *attrs)
2073 {
2074 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2075 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE);
2076 	struct mlx5_ib_dev *dev = mlx5_udata_to_mdev(&attrs->driver_udata);
2077 	struct mlx5_ib_flow_matcher *obj;
2078 	int err;
2079 
2080 	obj = kzalloc(sizeof(struct mlx5_ib_flow_matcher), GFP_KERNEL);
2081 	if (!obj)
2082 		return -ENOMEM;
2083 
2084 	obj->mask_len = uverbs_attr_get_len(
2085 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2086 	err = uverbs_copy_from(&obj->matcher_mask,
2087 			       attrs,
2088 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2089 	if (err)
2090 		goto end;
2091 
2092 	obj->flow_type = uverbs_attr_get_enum_id(
2093 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2094 
2095 	if (obj->flow_type == MLX5_IB_FLOW_TYPE_NORMAL) {
2096 		err = uverbs_copy_from(&obj->priority,
2097 				       attrs,
2098 				       MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2099 		if (err)
2100 			goto end;
2101 	}
2102 
2103 	err = uverbs_copy_from(&obj->match_criteria_enable,
2104 			       attrs,
2105 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA);
2106 	if (err)
2107 		goto end;
2108 
2109 	err = mlx5_ib_matcher_ns(attrs, obj);
2110 	if (err)
2111 		goto end;
2112 
2113 	if (obj->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS &&
2114 	    mlx5_eswitch_mode(dev->mdev) != MLX5_ESWITCH_OFFLOADS) {
2115 		err = -EINVAL;
2116 		goto end;
2117 	}
2118 
2119 	uobj->object = obj;
2120 	obj->mdev = dev->mdev;
2121 	atomic_set(&obj->usecnt, 0);
2122 	return 0;
2123 
2124 end:
2125 	kfree(obj);
2126 	return err;
2127 }
2128 
2129 static struct ib_flow_action *
2130 mlx5_ib_create_modify_header(struct mlx5_ib_dev *dev,
2131 			     enum mlx5_ib_uapi_flow_table_type ft_type,
2132 			     u8 num_actions, void *in)
2133 {
2134 	enum mlx5_flow_namespace_type namespace;
2135 	struct mlx5_ib_flow_action *maction;
2136 	int ret;
2137 
2138 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2139 	if (ret)
2140 		return ERR_PTR(-EINVAL);
2141 
2142 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2143 	if (!maction)
2144 		return ERR_PTR(-ENOMEM);
2145 
2146 	maction->flow_action_raw.modify_hdr =
2147 		mlx5_modify_header_alloc(dev->mdev, namespace, num_actions, in);
2148 
2149 	if (IS_ERR(maction->flow_action_raw.modify_hdr)) {
2150 		ret = PTR_ERR(maction->flow_action_raw.modify_hdr);
2151 		kfree(maction);
2152 		return ERR_PTR(ret);
2153 	}
2154 	maction->flow_action_raw.sub_type =
2155 		MLX5_IB_FLOW_ACTION_MODIFY_HEADER;
2156 	maction->flow_action_raw.dev = dev;
2157 
2158 	return &maction->ib_action;
2159 }
2160 
2161 static bool mlx5_ib_modify_header_supported(struct mlx5_ib_dev *dev)
2162 {
2163 	return MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
2164 					 max_modify_header_actions) ||
2165 	       MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev,
2166 					 max_modify_header_actions) ||
2167 	       MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
2168 					 max_modify_header_actions);
2169 }
2170 
2171 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER)(
2172 	struct uverbs_attr_bundle *attrs)
2173 {
2174 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2175 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE);
2176 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2177 	enum mlx5_ib_uapi_flow_table_type ft_type;
2178 	struct ib_flow_action *action;
2179 	int num_actions;
2180 	void *in;
2181 	int ret;
2182 
2183 	if (!mlx5_ib_modify_header_supported(mdev))
2184 		return -EOPNOTSUPP;
2185 
2186 	in = uverbs_attr_get_alloced_ptr(attrs,
2187 		MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM);
2188 
2189 	num_actions = uverbs_attr_ptr_get_array_size(
2190 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2191 		MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto));
2192 	if (num_actions < 0)
2193 		return num_actions;
2194 
2195 	ret = uverbs_get_const(&ft_type, attrs,
2196 			       MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE);
2197 	if (ret)
2198 		return ret;
2199 	action = mlx5_ib_create_modify_header(mdev, ft_type, num_actions, in);
2200 	if (IS_ERR(action))
2201 		return PTR_ERR(action);
2202 
2203 	uverbs_flow_action_fill_action(action, uobj, &mdev->ib_dev,
2204 				       IB_FLOW_ACTION_UNSPECIFIED);
2205 
2206 	return 0;
2207 }
2208 
2209 static bool mlx5_ib_flow_action_packet_reformat_valid(struct mlx5_ib_dev *ibdev,
2210 						      u8 packet_reformat_type,
2211 						      u8 ft_type)
2212 {
2213 	switch (packet_reformat_type) {
2214 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2215 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2216 			return MLX5_CAP_FLOWTABLE(ibdev->mdev,
2217 						  encap_general_header);
2218 		break;
2219 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2220 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2221 			return MLX5_CAP_FLOWTABLE_NIC_TX(ibdev->mdev,
2222 				reformat_l2_to_l3_tunnel);
2223 		break;
2224 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2225 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2226 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev,
2227 				reformat_l3_tunnel_to_l2);
2228 		break;
2229 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2:
2230 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2231 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev, decap);
2232 		break;
2233 	default:
2234 		break;
2235 	}
2236 
2237 	return false;
2238 }
2239 
2240 static int mlx5_ib_dv_to_prm_packet_reforamt_type(u8 dv_prt, u8 *prm_prt)
2241 {
2242 	switch (dv_prt) {
2243 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2244 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L2_TUNNEL;
2245 		break;
2246 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2247 		*prm_prt = MLX5_REFORMAT_TYPE_L3_TUNNEL_TO_L2;
2248 		break;
2249 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2250 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
2251 		break;
2252 	default:
2253 		return -EINVAL;
2254 	}
2255 
2256 	return 0;
2257 }
2258 
2259 static int mlx5_ib_flow_action_create_packet_reformat_ctx(
2260 	struct mlx5_ib_dev *dev,
2261 	struct mlx5_ib_flow_action *maction,
2262 	u8 ft_type, u8 dv_prt,
2263 	void *in, size_t len)
2264 {
2265 	struct mlx5_pkt_reformat_params reformat_params;
2266 	enum mlx5_flow_namespace_type namespace;
2267 	u8 prm_prt;
2268 	int ret;
2269 
2270 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2271 	if (ret)
2272 		return ret;
2273 
2274 	ret = mlx5_ib_dv_to_prm_packet_reforamt_type(dv_prt, &prm_prt);
2275 	if (ret)
2276 		return ret;
2277 
2278 	memset(&reformat_params, 0, sizeof(reformat_params));
2279 	reformat_params.type = prm_prt;
2280 	reformat_params.size = len;
2281 	reformat_params.data = in;
2282 	maction->flow_action_raw.pkt_reformat =
2283 		mlx5_packet_reformat_alloc(dev->mdev, &reformat_params,
2284 					   namespace);
2285 	if (IS_ERR(maction->flow_action_raw.pkt_reformat)) {
2286 		ret = PTR_ERR(maction->flow_action_raw.pkt_reformat);
2287 		return ret;
2288 	}
2289 
2290 	maction->flow_action_raw.sub_type =
2291 		MLX5_IB_FLOW_ACTION_PACKET_REFORMAT;
2292 	maction->flow_action_raw.dev = dev;
2293 
2294 	return 0;
2295 }
2296 
2297 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT)(
2298 	struct uverbs_attr_bundle *attrs)
2299 {
2300 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs,
2301 		MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE);
2302 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2303 	enum mlx5_ib_uapi_flow_action_packet_reformat_type dv_prt;
2304 	enum mlx5_ib_uapi_flow_table_type ft_type;
2305 	struct mlx5_ib_flow_action *maction;
2306 	int ret;
2307 
2308 	ret = uverbs_get_const(&ft_type, attrs,
2309 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE);
2310 	if (ret)
2311 		return ret;
2312 
2313 	ret = uverbs_get_const(&dv_prt, attrs,
2314 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE);
2315 	if (ret)
2316 		return ret;
2317 
2318 	if (!mlx5_ib_flow_action_packet_reformat_valid(mdev, dv_prt, ft_type))
2319 		return -EOPNOTSUPP;
2320 
2321 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2322 	if (!maction)
2323 		return -ENOMEM;
2324 
2325 	if (dv_prt ==
2326 	    MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2) {
2327 		maction->flow_action_raw.sub_type =
2328 			MLX5_IB_FLOW_ACTION_DECAP;
2329 		maction->flow_action_raw.dev = mdev;
2330 	} else {
2331 		void *in;
2332 		int len;
2333 
2334 		in = uverbs_attr_get_alloced_ptr(attrs,
2335 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2336 		if (IS_ERR(in)) {
2337 			ret = PTR_ERR(in);
2338 			goto free_maction;
2339 		}
2340 
2341 		len = uverbs_attr_get_len(attrs,
2342 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2343 
2344 		ret = mlx5_ib_flow_action_create_packet_reformat_ctx(mdev,
2345 			maction, ft_type, dv_prt, in, len);
2346 		if (ret)
2347 			goto free_maction;
2348 	}
2349 
2350 	uverbs_flow_action_fill_action(&maction->ib_action, uobj, &mdev->ib_dev,
2351 				       IB_FLOW_ACTION_UNSPECIFIED);
2352 	return 0;
2353 
2354 free_maction:
2355 	kfree(maction);
2356 	return ret;
2357 }
2358 
2359 DECLARE_UVERBS_NAMED_METHOD(
2360 	MLX5_IB_METHOD_CREATE_FLOW,
2361 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2362 			UVERBS_OBJECT_FLOW,
2363 			UVERBS_ACCESS_NEW,
2364 			UA_MANDATORY),
2365 	UVERBS_ATTR_PTR_IN(
2366 		MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE,
2367 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2368 		UA_MANDATORY,
2369 		UA_ALLOC_AND_COPY),
2370 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_MATCHER,
2371 			MLX5_IB_OBJECT_FLOW_MATCHER,
2372 			UVERBS_ACCESS_READ,
2373 			UA_MANDATORY),
2374 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_QP,
2375 			UVERBS_OBJECT_QP,
2376 			UVERBS_ACCESS_READ),
2377 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX,
2378 			MLX5_IB_OBJECT_DEVX_OBJ,
2379 			UVERBS_ACCESS_READ),
2380 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS,
2381 			     UVERBS_OBJECT_FLOW_ACTION,
2382 			     UVERBS_ACCESS_READ, 1,
2383 			     MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS,
2384 			     UA_OPTIONAL),
2385 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_TAG,
2386 			   UVERBS_ATTR_TYPE(u32),
2387 			   UA_OPTIONAL),
2388 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX,
2389 			     MLX5_IB_OBJECT_DEVX_OBJ,
2390 			     UVERBS_ACCESS_READ, 1, 1,
2391 			     UA_OPTIONAL),
2392 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
2393 			   UVERBS_ATTR_MIN_SIZE(sizeof(u32)),
2394 			   UA_OPTIONAL,
2395 			   UA_ALLOC_AND_COPY),
2396 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
2397 			     enum mlx5_ib_create_flow_flags,
2398 			     UA_OPTIONAL));
2399 
2400 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2401 	MLX5_IB_METHOD_DESTROY_FLOW,
2402 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2403 			UVERBS_OBJECT_FLOW,
2404 			UVERBS_ACCESS_DESTROY,
2405 			UA_MANDATORY));
2406 
2407 ADD_UVERBS_METHODS(mlx5_ib_fs,
2408 		   UVERBS_OBJECT_FLOW,
2409 		   &UVERBS_METHOD(MLX5_IB_METHOD_CREATE_FLOW),
2410 		   &UVERBS_METHOD(MLX5_IB_METHOD_DESTROY_FLOW));
2411 
2412 DECLARE_UVERBS_NAMED_METHOD(
2413 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER,
2414 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE,
2415 			UVERBS_OBJECT_FLOW_ACTION,
2416 			UVERBS_ACCESS_NEW,
2417 			UA_MANDATORY),
2418 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2419 			   UVERBS_ATTR_MIN_SIZE(MLX5_UN_SZ_BYTES(
2420 				   set_add_copy_action_in_auto)),
2421 			   UA_MANDATORY,
2422 			   UA_ALLOC_AND_COPY),
2423 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE,
2424 			     enum mlx5_ib_uapi_flow_table_type,
2425 			     UA_MANDATORY));
2426 
2427 DECLARE_UVERBS_NAMED_METHOD(
2428 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT,
2429 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE,
2430 			UVERBS_OBJECT_FLOW_ACTION,
2431 			UVERBS_ACCESS_NEW,
2432 			UA_MANDATORY),
2433 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF,
2434 			   UVERBS_ATTR_MIN_SIZE(1),
2435 			   UA_ALLOC_AND_COPY,
2436 			   UA_OPTIONAL),
2437 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE,
2438 			     enum mlx5_ib_uapi_flow_action_packet_reformat_type,
2439 			     UA_MANDATORY),
2440 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE,
2441 			     enum mlx5_ib_uapi_flow_table_type,
2442 			     UA_MANDATORY));
2443 
2444 ADD_UVERBS_METHODS(
2445 	mlx5_ib_flow_actions,
2446 	UVERBS_OBJECT_FLOW_ACTION,
2447 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER),
2448 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT));
2449 
2450 DECLARE_UVERBS_NAMED_METHOD(
2451 	MLX5_IB_METHOD_FLOW_MATCHER_CREATE,
2452 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE,
2453 			MLX5_IB_OBJECT_FLOW_MATCHER,
2454 			UVERBS_ACCESS_NEW,
2455 			UA_MANDATORY),
2456 	UVERBS_ATTR_PTR_IN(
2457 		MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK,
2458 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2459 		UA_MANDATORY),
2460 	UVERBS_ATTR_ENUM_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE,
2461 			    mlx5_ib_flow_type,
2462 			    UA_MANDATORY),
2463 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA,
2464 			   UVERBS_ATTR_TYPE(u8),
2465 			   UA_MANDATORY),
2466 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2467 			     enum ib_flow_flags,
2468 			     UA_OPTIONAL),
2469 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE,
2470 			     enum mlx5_ib_uapi_flow_table_type,
2471 			     UA_OPTIONAL));
2472 
2473 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2474 	MLX5_IB_METHOD_FLOW_MATCHER_DESTROY,
2475 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_DESTROY_HANDLE,
2476 			MLX5_IB_OBJECT_FLOW_MATCHER,
2477 			UVERBS_ACCESS_DESTROY,
2478 			UA_MANDATORY));
2479 
2480 DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_FLOW_MATCHER,
2481 			    UVERBS_TYPE_ALLOC_IDR(flow_matcher_cleanup),
2482 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_CREATE),
2483 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_DESTROY));
2484 
2485 const struct uapi_definition mlx5_ib_flow_defs[] = {
2486 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
2487 		MLX5_IB_OBJECT_FLOW_MATCHER),
2488 	UAPI_DEF_CHAIN_OBJ_TREE(
2489 		UVERBS_OBJECT_FLOW,
2490 		&mlx5_ib_fs),
2491 	UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_FLOW_ACTION,
2492 				&mlx5_ib_flow_actions),
2493 	{},
2494 };
2495 
2496 static const struct ib_device_ops flow_ops = {
2497 	.create_flow = mlx5_ib_create_flow,
2498 	.destroy_flow = mlx5_ib_destroy_flow,
2499 	.destroy_flow_action = mlx5_ib_destroy_flow_action,
2500 };
2501 
2502 int mlx5_ib_fs_init(struct mlx5_ib_dev *dev)
2503 {
2504 	dev->flow_db = kzalloc(sizeof(*dev->flow_db), GFP_KERNEL);
2505 
2506 	if (!dev->flow_db)
2507 		return -ENOMEM;
2508 
2509 	mutex_init(&dev->flow_db->lock);
2510 
2511 	ib_set_device_ops(&dev->ib_dev, &flow_ops);
2512 	return 0;
2513 }
2514