xref: /openbmc/linux/drivers/infiniband/hw/mlx5/fs.c (revision 4eb5928d)
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_umem.h>
14 #include <linux/mlx5/driver.h>
15 #include <linux/mlx5/fs.h>
16 #include <linux/mlx5/fs_helpers.h>
17 #include <linux/mlx5/accel.h>
18 #include <linux/mlx5/eswitch.h>
19 #include "mlx5_ib.h"
20 #include "counters.h"
21 #include "devx.h"
22 #include "fs.h"
23 
24 #define UVERBS_MODULE_NAME mlx5_ib
25 #include <rdma/uverbs_named_ioctl.h>
26 
27 enum {
28 	MATCH_CRITERIA_ENABLE_OUTER_BIT,
29 	MATCH_CRITERIA_ENABLE_MISC_BIT,
30 	MATCH_CRITERIA_ENABLE_INNER_BIT,
31 	MATCH_CRITERIA_ENABLE_MISC2_BIT
32 };
33 
34 #define HEADER_IS_ZERO(match_criteria, headers)			           \
35 	!(memchr_inv(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
36 		    0, MLX5_FLD_SZ_BYTES(fte_match_param, headers)))       \
37 
38 static u8 get_match_criteria_enable(u32 *match_criteria)
39 {
40 	u8 match_criteria_enable;
41 
42 	match_criteria_enable =
43 		(!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
44 		MATCH_CRITERIA_ENABLE_OUTER_BIT;
45 	match_criteria_enable |=
46 		(!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
47 		MATCH_CRITERIA_ENABLE_MISC_BIT;
48 	match_criteria_enable |=
49 		(!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
50 		MATCH_CRITERIA_ENABLE_INNER_BIT;
51 	match_criteria_enable |=
52 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
53 		MATCH_CRITERIA_ENABLE_MISC2_BIT;
54 
55 	return match_criteria_enable;
56 }
57 
58 static int set_proto(void *outer_c, void *outer_v, u8 mask, u8 val)
59 {
60 	u8 entry_mask;
61 	u8 entry_val;
62 	int err = 0;
63 
64 	if (!mask)
65 		goto out;
66 
67 	entry_mask = MLX5_GET(fte_match_set_lyr_2_4, outer_c,
68 			      ip_protocol);
69 	entry_val = MLX5_GET(fte_match_set_lyr_2_4, outer_v,
70 			     ip_protocol);
71 	if (!entry_mask) {
72 		MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_protocol, mask);
73 		MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_protocol, val);
74 		goto out;
75 	}
76 	/* Don't override existing ip protocol */
77 	if (mask != entry_mask || val != entry_val)
78 		err = -EINVAL;
79 out:
80 	return err;
81 }
82 
83 static void set_flow_label(void *misc_c, void *misc_v, u32 mask, u32 val,
84 			   bool inner)
85 {
86 	if (inner) {
87 		MLX5_SET(fte_match_set_misc,
88 			 misc_c, inner_ipv6_flow_label, mask);
89 		MLX5_SET(fte_match_set_misc,
90 			 misc_v, inner_ipv6_flow_label, val);
91 	} else {
92 		MLX5_SET(fte_match_set_misc,
93 			 misc_c, outer_ipv6_flow_label, mask);
94 		MLX5_SET(fte_match_set_misc,
95 			 misc_v, outer_ipv6_flow_label, val);
96 	}
97 }
98 
99 static void set_tos(void *outer_c, void *outer_v, u8 mask, u8 val)
100 {
101 	MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_ecn, mask);
102 	MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_ecn, val);
103 	MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_dscp, mask >> 2);
104 	MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_dscp, val >> 2);
105 }
106 
107 static int check_mpls_supp_fields(u32 field_support, const __be32 *set_mask)
108 {
109 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_label) &&
110 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_LABEL))
111 		return -EOPNOTSUPP;
112 
113 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_exp) &&
114 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_EXP))
115 		return -EOPNOTSUPP;
116 
117 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_s_bos) &&
118 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_S_BOS))
119 		return -EOPNOTSUPP;
120 
121 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_ttl) &&
122 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_TTL))
123 		return -EOPNOTSUPP;
124 
125 	return 0;
126 }
127 
128 #define LAST_ETH_FIELD vlan_tag
129 #define LAST_IB_FIELD sl
130 #define LAST_IPV4_FIELD tos
131 #define LAST_IPV6_FIELD traffic_class
132 #define LAST_TCP_UDP_FIELD src_port
133 #define LAST_TUNNEL_FIELD tunnel_id
134 #define LAST_FLOW_TAG_FIELD tag_id
135 #define LAST_DROP_FIELD size
136 #define LAST_COUNTERS_FIELD counters
137 
138 /* Field is the last supported field */
139 #define FIELDS_NOT_SUPPORTED(filter, field)\
140 	memchr_inv((void *)&filter.field  +\
141 		   sizeof(filter.field), 0,\
142 		   sizeof(filter) -\
143 		   offsetof(typeof(filter), field) -\
144 		   sizeof(filter.field))
145 
146 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
147 			   bool is_egress,
148 			   struct mlx5_flow_act *action)
149 {
150 
151 	switch (maction->ib_action.type) {
152 	case IB_FLOW_ACTION_ESP:
153 		if (action->action & (MLX5_FLOW_CONTEXT_ACTION_ENCRYPT |
154 				      MLX5_FLOW_CONTEXT_ACTION_DECRYPT))
155 			return -EINVAL;
156 		/* Currently only AES_GCM keymat is supported by the driver */
157 		action->esp_id = (uintptr_t)maction->esp_aes_gcm.ctx;
158 		action->action |= is_egress ?
159 			MLX5_FLOW_CONTEXT_ACTION_ENCRYPT :
160 			MLX5_FLOW_CONTEXT_ACTION_DECRYPT;
161 		return 0;
162 	case IB_FLOW_ACTION_UNSPECIFIED:
163 		if (maction->flow_action_raw.sub_type ==
164 		    MLX5_IB_FLOW_ACTION_MODIFY_HEADER) {
165 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
166 				return -EINVAL;
167 			action->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
168 			action->modify_hdr =
169 				maction->flow_action_raw.modify_hdr;
170 			return 0;
171 		}
172 		if (maction->flow_action_raw.sub_type ==
173 		    MLX5_IB_FLOW_ACTION_DECAP) {
174 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_DECAP)
175 				return -EINVAL;
176 			action->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
177 			return 0;
178 		}
179 		if (maction->flow_action_raw.sub_type ==
180 		    MLX5_IB_FLOW_ACTION_PACKET_REFORMAT) {
181 			if (action->action &
182 			    MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT)
183 				return -EINVAL;
184 			action->action |=
185 				MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
186 			action->pkt_reformat =
187 				maction->flow_action_raw.pkt_reformat;
188 			return 0;
189 		}
190 		fallthrough;
191 	default:
192 		return -EOPNOTSUPP;
193 	}
194 }
195 
196 static int parse_flow_attr(struct mlx5_core_dev *mdev,
197 			   struct mlx5_flow_spec *spec,
198 			   const union ib_flow_spec *ib_spec,
199 			   const struct ib_flow_attr *flow_attr,
200 			   struct mlx5_flow_act *action, u32 prev_type)
201 {
202 	struct mlx5_flow_context *flow_context = &spec->flow_context;
203 	u32 *match_c = spec->match_criteria;
204 	u32 *match_v = spec->match_value;
205 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param, match_c,
206 					   misc_parameters);
207 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, match_v,
208 					   misc_parameters);
209 	void *misc_params2_c = MLX5_ADDR_OF(fte_match_param, match_c,
210 					    misc_parameters_2);
211 	void *misc_params2_v = MLX5_ADDR_OF(fte_match_param, match_v,
212 					    misc_parameters_2);
213 	void *headers_c;
214 	void *headers_v;
215 	int match_ipv;
216 	int ret;
217 
218 	if (ib_spec->type & IB_FLOW_SPEC_INNER) {
219 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
220 					 inner_headers);
221 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
222 					 inner_headers);
223 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
224 					ft_field_support.inner_ip_version);
225 	} else {
226 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
227 					 outer_headers);
228 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
229 					 outer_headers);
230 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
231 					ft_field_support.outer_ip_version);
232 	}
233 
234 	switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
235 	case IB_FLOW_SPEC_ETH:
236 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
237 			return -EOPNOTSUPP;
238 
239 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
240 					     dmac_47_16),
241 				ib_spec->eth.mask.dst_mac);
242 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
243 					     dmac_47_16),
244 				ib_spec->eth.val.dst_mac);
245 
246 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
247 					     smac_47_16),
248 				ib_spec->eth.mask.src_mac);
249 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
250 					     smac_47_16),
251 				ib_spec->eth.val.src_mac);
252 
253 		if (ib_spec->eth.mask.vlan_tag) {
254 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
255 				 cvlan_tag, 1);
256 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
257 				 cvlan_tag, 1);
258 
259 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
260 				 first_vid, ntohs(ib_spec->eth.mask.vlan_tag));
261 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
262 				 first_vid, ntohs(ib_spec->eth.val.vlan_tag));
263 
264 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
265 				 first_cfi,
266 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 12);
267 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
268 				 first_cfi,
269 				 ntohs(ib_spec->eth.val.vlan_tag) >> 12);
270 
271 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
272 				 first_prio,
273 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 13);
274 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
275 				 first_prio,
276 				 ntohs(ib_spec->eth.val.vlan_tag) >> 13);
277 		}
278 		MLX5_SET(fte_match_set_lyr_2_4, headers_c,
279 			 ethertype, ntohs(ib_spec->eth.mask.ether_type));
280 		MLX5_SET(fte_match_set_lyr_2_4, headers_v,
281 			 ethertype, ntohs(ib_spec->eth.val.ether_type));
282 		break;
283 	case IB_FLOW_SPEC_IPV4:
284 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
285 			return -EOPNOTSUPP;
286 
287 		if (match_ipv) {
288 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
289 				 ip_version, 0xf);
290 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
291 				 ip_version, MLX5_FS_IPV4_VERSION);
292 		} else {
293 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
294 				 ethertype, 0xffff);
295 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
296 				 ethertype, ETH_P_IP);
297 		}
298 
299 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
300 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
301 		       &ib_spec->ipv4.mask.src_ip,
302 		       sizeof(ib_spec->ipv4.mask.src_ip));
303 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
304 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
305 		       &ib_spec->ipv4.val.src_ip,
306 		       sizeof(ib_spec->ipv4.val.src_ip));
307 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
308 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
309 		       &ib_spec->ipv4.mask.dst_ip,
310 		       sizeof(ib_spec->ipv4.mask.dst_ip));
311 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
312 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
313 		       &ib_spec->ipv4.val.dst_ip,
314 		       sizeof(ib_spec->ipv4.val.dst_ip));
315 
316 		set_tos(headers_c, headers_v,
317 			ib_spec->ipv4.mask.tos, ib_spec->ipv4.val.tos);
318 
319 		if (set_proto(headers_c, headers_v,
320 			      ib_spec->ipv4.mask.proto,
321 			      ib_spec->ipv4.val.proto))
322 			return -EINVAL;
323 		break;
324 	case IB_FLOW_SPEC_IPV6:
325 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv6.mask, LAST_IPV6_FIELD))
326 			return -EOPNOTSUPP;
327 
328 		if (match_ipv) {
329 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
330 				 ip_version, 0xf);
331 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
332 				 ip_version, MLX5_FS_IPV6_VERSION);
333 		} else {
334 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
335 				 ethertype, 0xffff);
336 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
337 				 ethertype, ETH_P_IPV6);
338 		}
339 
340 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
341 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
342 		       &ib_spec->ipv6.mask.src_ip,
343 		       sizeof(ib_spec->ipv6.mask.src_ip));
344 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
345 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
346 		       &ib_spec->ipv6.val.src_ip,
347 		       sizeof(ib_spec->ipv6.val.src_ip));
348 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
349 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
350 		       &ib_spec->ipv6.mask.dst_ip,
351 		       sizeof(ib_spec->ipv6.mask.dst_ip));
352 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
353 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
354 		       &ib_spec->ipv6.val.dst_ip,
355 		       sizeof(ib_spec->ipv6.val.dst_ip));
356 
357 		set_tos(headers_c, headers_v,
358 			ib_spec->ipv6.mask.traffic_class,
359 			ib_spec->ipv6.val.traffic_class);
360 
361 		if (set_proto(headers_c, headers_v,
362 			      ib_spec->ipv6.mask.next_hdr,
363 			      ib_spec->ipv6.val.next_hdr))
364 			return -EINVAL;
365 
366 		set_flow_label(misc_params_c, misc_params_v,
367 			       ntohl(ib_spec->ipv6.mask.flow_label),
368 			       ntohl(ib_spec->ipv6.val.flow_label),
369 			       ib_spec->type & IB_FLOW_SPEC_INNER);
370 		break;
371 	case IB_FLOW_SPEC_ESP:
372 		if (ib_spec->esp.mask.seq)
373 			return -EOPNOTSUPP;
374 
375 		MLX5_SET(fte_match_set_misc, misc_params_c, outer_esp_spi,
376 			 ntohl(ib_spec->esp.mask.spi));
377 		MLX5_SET(fte_match_set_misc, misc_params_v, outer_esp_spi,
378 			 ntohl(ib_spec->esp.val.spi));
379 		break;
380 	case IB_FLOW_SPEC_TCP:
381 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
382 					 LAST_TCP_UDP_FIELD))
383 			return -EOPNOTSUPP;
384 
385 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_TCP))
386 			return -EINVAL;
387 
388 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_sport,
389 			 ntohs(ib_spec->tcp_udp.mask.src_port));
390 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
391 			 ntohs(ib_spec->tcp_udp.val.src_port));
392 
393 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_dport,
394 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
395 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
396 			 ntohs(ib_spec->tcp_udp.val.dst_port));
397 		break;
398 	case IB_FLOW_SPEC_UDP:
399 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
400 					 LAST_TCP_UDP_FIELD))
401 			return -EOPNOTSUPP;
402 
403 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_UDP))
404 			return -EINVAL;
405 
406 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_sport,
407 			 ntohs(ib_spec->tcp_udp.mask.src_port));
408 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
409 			 ntohs(ib_spec->tcp_udp.val.src_port));
410 
411 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_dport,
412 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
413 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
414 			 ntohs(ib_spec->tcp_udp.val.dst_port));
415 		break;
416 	case IB_FLOW_SPEC_GRE:
417 		if (ib_spec->gre.mask.c_ks_res0_ver)
418 			return -EOPNOTSUPP;
419 
420 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_GRE))
421 			return -EINVAL;
422 
423 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
424 			 0xff);
425 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
426 			 IPPROTO_GRE);
427 
428 		MLX5_SET(fte_match_set_misc, misc_params_c, gre_protocol,
429 			 ntohs(ib_spec->gre.mask.protocol));
430 		MLX5_SET(fte_match_set_misc, misc_params_v, gre_protocol,
431 			 ntohs(ib_spec->gre.val.protocol));
432 
433 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_c,
434 				    gre_key.nvgre.hi),
435 		       &ib_spec->gre.mask.key,
436 		       sizeof(ib_spec->gre.mask.key));
437 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_v,
438 				    gre_key.nvgre.hi),
439 		       &ib_spec->gre.val.key,
440 		       sizeof(ib_spec->gre.val.key));
441 		break;
442 	case IB_FLOW_SPEC_MPLS:
443 		switch (prev_type) {
444 		case IB_FLOW_SPEC_UDP:
445 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
446 						   ft_field_support.outer_first_mpls_over_udp),
447 						   &ib_spec->mpls.mask.tag))
448 				return -EOPNOTSUPP;
449 
450 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
451 					    outer_first_mpls_over_udp),
452 			       &ib_spec->mpls.val.tag,
453 			       sizeof(ib_spec->mpls.val.tag));
454 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
455 					    outer_first_mpls_over_udp),
456 			       &ib_spec->mpls.mask.tag,
457 			       sizeof(ib_spec->mpls.mask.tag));
458 			break;
459 		case IB_FLOW_SPEC_GRE:
460 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
461 						   ft_field_support.outer_first_mpls_over_gre),
462 						   &ib_spec->mpls.mask.tag))
463 				return -EOPNOTSUPP;
464 
465 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
466 					    outer_first_mpls_over_gre),
467 			       &ib_spec->mpls.val.tag,
468 			       sizeof(ib_spec->mpls.val.tag));
469 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
470 					    outer_first_mpls_over_gre),
471 			       &ib_spec->mpls.mask.tag,
472 			       sizeof(ib_spec->mpls.mask.tag));
473 			break;
474 		default:
475 			if (ib_spec->type & IB_FLOW_SPEC_INNER) {
476 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
477 							   ft_field_support.inner_first_mpls),
478 							   &ib_spec->mpls.mask.tag))
479 					return -EOPNOTSUPP;
480 
481 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
482 						    inner_first_mpls),
483 				       &ib_spec->mpls.val.tag,
484 				       sizeof(ib_spec->mpls.val.tag));
485 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
486 						    inner_first_mpls),
487 				       &ib_spec->mpls.mask.tag,
488 				       sizeof(ib_spec->mpls.mask.tag));
489 			} else {
490 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
491 							   ft_field_support.outer_first_mpls),
492 							   &ib_spec->mpls.mask.tag))
493 					return -EOPNOTSUPP;
494 
495 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
496 						    outer_first_mpls),
497 				       &ib_spec->mpls.val.tag,
498 				       sizeof(ib_spec->mpls.val.tag));
499 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
500 						    outer_first_mpls),
501 				       &ib_spec->mpls.mask.tag,
502 				       sizeof(ib_spec->mpls.mask.tag));
503 			}
504 		}
505 		break;
506 	case IB_FLOW_SPEC_VXLAN_TUNNEL:
507 		if (FIELDS_NOT_SUPPORTED(ib_spec->tunnel.mask,
508 					 LAST_TUNNEL_FIELD))
509 			return -EOPNOTSUPP;
510 
511 		MLX5_SET(fte_match_set_misc, misc_params_c, vxlan_vni,
512 			 ntohl(ib_spec->tunnel.mask.tunnel_id));
513 		MLX5_SET(fte_match_set_misc, misc_params_v, vxlan_vni,
514 			 ntohl(ib_spec->tunnel.val.tunnel_id));
515 		break;
516 	case IB_FLOW_SPEC_ACTION_TAG:
517 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_tag,
518 					 LAST_FLOW_TAG_FIELD))
519 			return -EOPNOTSUPP;
520 		if (ib_spec->flow_tag.tag_id >= BIT(24))
521 			return -EINVAL;
522 
523 		flow_context->flow_tag = ib_spec->flow_tag.tag_id;
524 		flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
525 		break;
526 	case IB_FLOW_SPEC_ACTION_DROP:
527 		if (FIELDS_NOT_SUPPORTED(ib_spec->drop,
528 					 LAST_DROP_FIELD))
529 			return -EOPNOTSUPP;
530 		action->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
531 		break;
532 	case IB_FLOW_SPEC_ACTION_HANDLE:
533 		ret = parse_flow_flow_action(to_mflow_act(ib_spec->action.act),
534 			flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS, action);
535 		if (ret)
536 			return ret;
537 		break;
538 	case IB_FLOW_SPEC_ACTION_COUNT:
539 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_count,
540 					 LAST_COUNTERS_FIELD))
541 			return -EOPNOTSUPP;
542 
543 		/* for now support only one counters spec per flow */
544 		if (action->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
545 			return -EINVAL;
546 
547 		action->counters = ib_spec->flow_count.counters;
548 		action->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
549 		break;
550 	default:
551 		return -EINVAL;
552 	}
553 
554 	return 0;
555 }
556 
557 /* If a flow could catch both multicast and unicast packets,
558  * it won't fall into the multicast flow steering table and this rule
559  * could steal other multicast packets.
560  */
561 static bool flow_is_multicast_only(const struct ib_flow_attr *ib_attr)
562 {
563 	union ib_flow_spec *flow_spec;
564 
565 	if (ib_attr->type != IB_FLOW_ATTR_NORMAL ||
566 	    ib_attr->num_of_specs < 1)
567 		return false;
568 
569 	flow_spec = (union ib_flow_spec *)(ib_attr + 1);
570 	if (flow_spec->type == IB_FLOW_SPEC_IPV4) {
571 		struct ib_flow_spec_ipv4 *ipv4_spec;
572 
573 		ipv4_spec = (struct ib_flow_spec_ipv4 *)flow_spec;
574 		if (ipv4_is_multicast(ipv4_spec->val.dst_ip))
575 			return true;
576 
577 		return false;
578 	}
579 
580 	if (flow_spec->type == IB_FLOW_SPEC_ETH) {
581 		struct ib_flow_spec_eth *eth_spec;
582 
583 		eth_spec = (struct ib_flow_spec_eth *)flow_spec;
584 		return is_multicast_ether_addr(eth_spec->mask.dst_mac) &&
585 		       is_multicast_ether_addr(eth_spec->val.dst_mac);
586 	}
587 
588 	return false;
589 }
590 
591 enum valid_spec {
592 	VALID_SPEC_INVALID,
593 	VALID_SPEC_VALID,
594 	VALID_SPEC_NA,
595 };
596 
597 static enum valid_spec
598 is_valid_esp_aes_gcm(struct mlx5_core_dev *mdev,
599 		     const struct mlx5_flow_spec *spec,
600 		     const struct mlx5_flow_act *flow_act,
601 		     bool egress)
602 {
603 	const u32 *match_c = spec->match_criteria;
604 	bool is_crypto =
605 		(flow_act->action & (MLX5_FLOW_CONTEXT_ACTION_ENCRYPT |
606 				     MLX5_FLOW_CONTEXT_ACTION_DECRYPT));
607 	bool is_ipsec = mlx5_fs_is_ipsec_flow(match_c);
608 	bool is_drop = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_DROP;
609 
610 	/*
611 	 * Currently only crypto is supported in egress, when regular egress
612 	 * rules would be supported, always return VALID_SPEC_NA.
613 	 */
614 	if (!is_crypto)
615 		return VALID_SPEC_NA;
616 
617 	return is_crypto && is_ipsec &&
618 		(!egress || (!is_drop &&
619 			     !(spec->flow_context.flags & FLOW_CONTEXT_HAS_TAG))) ?
620 		VALID_SPEC_VALID : VALID_SPEC_INVALID;
621 }
622 
623 static bool is_valid_spec(struct mlx5_core_dev *mdev,
624 			  const struct mlx5_flow_spec *spec,
625 			  const struct mlx5_flow_act *flow_act,
626 			  bool egress)
627 {
628 	/* We curretly only support ipsec egress flow */
629 	return is_valid_esp_aes_gcm(mdev, spec, flow_act, egress) != VALID_SPEC_INVALID;
630 }
631 
632 static bool is_valid_ethertype(struct mlx5_core_dev *mdev,
633 			       const struct ib_flow_attr *flow_attr,
634 			       bool check_inner)
635 {
636 	union ib_flow_spec *ib_spec = (union ib_flow_spec *)(flow_attr + 1);
637 	int match_ipv = check_inner ?
638 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
639 					ft_field_support.inner_ip_version) :
640 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
641 					ft_field_support.outer_ip_version);
642 	int inner_bit = check_inner ? IB_FLOW_SPEC_INNER : 0;
643 	bool ipv4_spec_valid, ipv6_spec_valid;
644 	unsigned int ip_spec_type = 0;
645 	bool has_ethertype = false;
646 	unsigned int spec_index;
647 	bool mask_valid = true;
648 	u16 eth_type = 0;
649 	bool type_valid;
650 
651 	/* Validate that ethertype is correct */
652 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
653 		if ((ib_spec->type == (IB_FLOW_SPEC_ETH | inner_bit)) &&
654 		    ib_spec->eth.mask.ether_type) {
655 			mask_valid = (ib_spec->eth.mask.ether_type ==
656 				      htons(0xffff));
657 			has_ethertype = true;
658 			eth_type = ntohs(ib_spec->eth.val.ether_type);
659 		} else if ((ib_spec->type == (IB_FLOW_SPEC_IPV4 | inner_bit)) ||
660 			   (ib_spec->type == (IB_FLOW_SPEC_IPV6 | inner_bit))) {
661 			ip_spec_type = ib_spec->type;
662 		}
663 		ib_spec = (void *)ib_spec + ib_spec->size;
664 	}
665 
666 	type_valid = (!has_ethertype) || (!ip_spec_type);
667 	if (!type_valid && mask_valid) {
668 		ipv4_spec_valid = (eth_type == ETH_P_IP) &&
669 			(ip_spec_type == (IB_FLOW_SPEC_IPV4 | inner_bit));
670 		ipv6_spec_valid = (eth_type == ETH_P_IPV6) &&
671 			(ip_spec_type == (IB_FLOW_SPEC_IPV6 | inner_bit));
672 
673 		type_valid = (ipv4_spec_valid) || (ipv6_spec_valid) ||
674 			     (((eth_type == ETH_P_MPLS_UC) ||
675 			       (eth_type == ETH_P_MPLS_MC)) && match_ipv);
676 	}
677 
678 	return type_valid;
679 }
680 
681 static bool is_valid_attr(struct mlx5_core_dev *mdev,
682 			  const struct ib_flow_attr *flow_attr)
683 {
684 	return is_valid_ethertype(mdev, flow_attr, false) &&
685 	       is_valid_ethertype(mdev, flow_attr, true);
686 }
687 
688 static void put_flow_table(struct mlx5_ib_dev *dev,
689 			   struct mlx5_ib_flow_prio *prio, bool ft_added)
690 {
691 	prio->refcount -= !!ft_added;
692 	if (!prio->refcount) {
693 		mlx5_destroy_flow_table(prio->flow_table);
694 		prio->flow_table = NULL;
695 	}
696 }
697 
698 static int mlx5_ib_destroy_flow(struct ib_flow *flow_id)
699 {
700 	struct mlx5_ib_flow_handler *handler = container_of(flow_id,
701 							  struct mlx5_ib_flow_handler,
702 							  ibflow);
703 	struct mlx5_ib_flow_handler *iter, *tmp;
704 	struct mlx5_ib_dev *dev = handler->dev;
705 
706 	mutex_lock(&dev->flow_db->lock);
707 
708 	list_for_each_entry_safe(iter, tmp, &handler->list, list) {
709 		mlx5_del_flow_rules(iter->rule);
710 		put_flow_table(dev, iter->prio, true);
711 		list_del(&iter->list);
712 		kfree(iter);
713 	}
714 
715 	mlx5_del_flow_rules(handler->rule);
716 	put_flow_table(dev, handler->prio, true);
717 	mlx5_ib_counters_clear_description(handler->ibcounters);
718 	mutex_unlock(&dev->flow_db->lock);
719 	if (handler->flow_matcher)
720 		atomic_dec(&handler->flow_matcher->usecnt);
721 	kfree(handler);
722 
723 	return 0;
724 }
725 
726 static int ib_prio_to_core_prio(unsigned int priority, bool dont_trap)
727 {
728 	priority *= 2;
729 	if (!dont_trap)
730 		priority++;
731 	return priority;
732 }
733 
734 enum flow_table_type {
735 	MLX5_IB_FT_RX,
736 	MLX5_IB_FT_TX
737 };
738 
739 #define MLX5_FS_MAX_TYPES	 6
740 #define MLX5_FS_MAX_ENTRIES	 BIT(16)
741 
742 static struct mlx5_ib_flow_prio *_get_prio(struct mlx5_flow_namespace *ns,
743 					   struct mlx5_ib_flow_prio *prio,
744 					   int priority,
745 					   int num_entries, int num_groups,
746 					   u32 flags)
747 {
748 	struct mlx5_flow_table_attr ft_attr = {};
749 	struct mlx5_flow_table *ft;
750 
751 	ft_attr.prio = priority;
752 	ft_attr.max_fte = num_entries;
753 	ft_attr.flags = flags;
754 	ft_attr.autogroup.max_num_groups = num_groups;
755 	ft = mlx5_create_auto_grouped_flow_table(ns, &ft_attr);
756 	if (IS_ERR(ft))
757 		return ERR_CAST(ft);
758 
759 	prio->flow_table = ft;
760 	prio->refcount = 0;
761 	return prio;
762 }
763 
764 static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev,
765 						struct ib_flow_attr *flow_attr,
766 						enum flow_table_type ft_type)
767 {
768 	bool dont_trap = flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP;
769 	struct mlx5_flow_namespace *ns = NULL;
770 	struct mlx5_ib_flow_prio *prio;
771 	struct mlx5_flow_table *ft;
772 	int max_table_size;
773 	int num_entries;
774 	int num_groups;
775 	bool esw_encap;
776 	u32 flags = 0;
777 	int priority;
778 
779 	max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
780 						       log_max_ft_size));
781 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
782 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
783 	if (flow_attr->type == IB_FLOW_ATTR_NORMAL) {
784 		enum mlx5_flow_namespace_type fn_type;
785 
786 		if (flow_is_multicast_only(flow_attr) &&
787 		    !dont_trap)
788 			priority = MLX5_IB_FLOW_MCAST_PRIO;
789 		else
790 			priority = ib_prio_to_core_prio(flow_attr->priority,
791 							dont_trap);
792 		if (ft_type == MLX5_IB_FT_RX) {
793 			fn_type = MLX5_FLOW_NAMESPACE_BYPASS;
794 			prio = &dev->flow_db->prios[priority];
795 			if (!dev->is_rep && !esw_encap &&
796 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap))
797 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
798 			if (!dev->is_rep && !esw_encap &&
799 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
800 					reformat_l3_tunnel_to_l2))
801 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
802 		} else {
803 			max_table_size =
804 				BIT(MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev,
805 							      log_max_ft_size));
806 			fn_type = MLX5_FLOW_NAMESPACE_EGRESS;
807 			prio = &dev->flow_db->egress_prios[priority];
808 			if (!dev->is_rep && !esw_encap &&
809 			    MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat))
810 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
811 		}
812 		ns = mlx5_get_flow_namespace(dev->mdev, fn_type);
813 		num_entries = MLX5_FS_MAX_ENTRIES;
814 		num_groups = MLX5_FS_MAX_TYPES;
815 	} else if (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
816 		   flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT) {
817 		ns = mlx5_get_flow_namespace(dev->mdev,
818 					     MLX5_FLOW_NAMESPACE_LEFTOVERS);
819 		build_leftovers_ft_param(&priority,
820 					 &num_entries,
821 					 &num_groups);
822 		prio = &dev->flow_db->prios[MLX5_IB_FLOW_LEFTOVERS_PRIO];
823 	} else if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
824 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
825 					allow_sniffer_and_nic_rx_shared_tir))
826 			return ERR_PTR(-EOPNOTSUPP);
827 
828 		ns = mlx5_get_flow_namespace(dev->mdev, ft_type == MLX5_IB_FT_RX ?
829 					     MLX5_FLOW_NAMESPACE_SNIFFER_RX :
830 					     MLX5_FLOW_NAMESPACE_SNIFFER_TX);
831 
832 		prio = &dev->flow_db->sniffer[ft_type];
833 		priority = 0;
834 		num_entries = 1;
835 		num_groups = 1;
836 	}
837 
838 	if (!ns)
839 		return ERR_PTR(-EOPNOTSUPP);
840 
841 	max_table_size = min_t(int, num_entries, max_table_size);
842 
843 	ft = prio->flow_table;
844 	if (!ft)
845 		return _get_prio(ns, prio, priority, max_table_size, num_groups,
846 				 flags);
847 
848 	return prio;
849 }
850 
851 static void set_underlay_qp(struct mlx5_ib_dev *dev,
852 			    struct mlx5_flow_spec *spec,
853 			    u32 underlay_qpn)
854 {
855 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param,
856 					   spec->match_criteria,
857 					   misc_parameters);
858 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
859 					   misc_parameters);
860 
861 	if (underlay_qpn &&
862 	    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
863 				      ft_field_support.bth_dst_qp)) {
864 		MLX5_SET(fte_match_set_misc,
865 			 misc_params_v, bth_dst_qp, underlay_qpn);
866 		MLX5_SET(fte_match_set_misc,
867 			 misc_params_c, bth_dst_qp, 0xffffff);
868 	}
869 }
870 
871 static void mlx5_ib_set_rule_source_port(struct mlx5_ib_dev *dev,
872 					 struct mlx5_flow_spec *spec,
873 					 struct mlx5_eswitch_rep *rep)
874 {
875 	struct mlx5_eswitch *esw = dev->mdev->priv.eswitch;
876 	void *misc;
877 
878 	if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
879 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
880 				    misc_parameters_2);
881 
882 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
883 			 mlx5_eswitch_get_vport_metadata_for_match(esw,
884 								   rep->vport));
885 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
886 				    misc_parameters_2);
887 
888 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
889 			 mlx5_eswitch_get_vport_metadata_mask());
890 	} else {
891 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
892 				    misc_parameters);
893 
894 		MLX5_SET(fte_match_set_misc, misc, source_port, rep->vport);
895 
896 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
897 				    misc_parameters);
898 
899 		MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
900 	}
901 }
902 
903 static struct mlx5_ib_flow_handler *_create_flow_rule(struct mlx5_ib_dev *dev,
904 						      struct mlx5_ib_flow_prio *ft_prio,
905 						      const struct ib_flow_attr *flow_attr,
906 						      struct mlx5_flow_destination *dst,
907 						      u32 underlay_qpn,
908 						      struct mlx5_ib_create_flow *ucmd)
909 {
910 	struct mlx5_flow_table	*ft = ft_prio->flow_table;
911 	struct mlx5_ib_flow_handler *handler;
912 	struct mlx5_flow_act flow_act = {};
913 	struct mlx5_flow_spec *spec;
914 	struct mlx5_flow_destination dest_arr[2] = {};
915 	struct mlx5_flow_destination *rule_dst = dest_arr;
916 	const void *ib_flow = (const void *)flow_attr + sizeof(*flow_attr);
917 	unsigned int spec_index;
918 	u32 prev_type = 0;
919 	int err = 0;
920 	int dest_num = 0;
921 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
922 
923 	if (!is_valid_attr(dev->mdev, flow_attr))
924 		return ERR_PTR(-EINVAL);
925 
926 	if (dev->is_rep && is_egress)
927 		return ERR_PTR(-EINVAL);
928 
929 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
930 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
931 	if (!handler || !spec) {
932 		err = -ENOMEM;
933 		goto free;
934 	}
935 
936 	INIT_LIST_HEAD(&handler->list);
937 
938 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
939 		err = parse_flow_attr(dev->mdev, spec,
940 				      ib_flow, flow_attr, &flow_act,
941 				      prev_type);
942 		if (err < 0)
943 			goto free;
944 
945 		prev_type = ((union ib_flow_spec *)ib_flow)->type;
946 		ib_flow += ((union ib_flow_spec *)ib_flow)->size;
947 	}
948 
949 	if (dst && !(flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP)) {
950 		memcpy(&dest_arr[0], dst, sizeof(*dst));
951 		dest_num++;
952 	}
953 
954 	if (!flow_is_multicast_only(flow_attr))
955 		set_underlay_qp(dev, spec, underlay_qpn);
956 
957 	if (dev->is_rep) {
958 		struct mlx5_eswitch_rep *rep;
959 
960 		rep = dev->port[flow_attr->port - 1].rep;
961 		if (!rep) {
962 			err = -EINVAL;
963 			goto free;
964 		}
965 
966 		mlx5_ib_set_rule_source_port(dev, spec, rep);
967 	}
968 
969 	spec->match_criteria_enable = get_match_criteria_enable(spec->match_criteria);
970 
971 	if (is_egress &&
972 	    !is_valid_spec(dev->mdev, spec, &flow_act, is_egress)) {
973 		err = -EINVAL;
974 		goto free;
975 	}
976 
977 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
978 		struct mlx5_ib_mcounters *mcounters;
979 
980 		err = mlx5_ib_flow_counters_set_data(flow_act.counters, ucmd);
981 		if (err)
982 			goto free;
983 
984 		mcounters = to_mcounters(flow_act.counters);
985 		handler->ibcounters = flow_act.counters;
986 		dest_arr[dest_num].type =
987 			MLX5_FLOW_DESTINATION_TYPE_COUNTER;
988 		dest_arr[dest_num].counter_id =
989 			mlx5_fc_id(mcounters->hw_cntrs_hndl);
990 		dest_num++;
991 	}
992 
993 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP) {
994 		if (!dest_num)
995 			rule_dst = NULL;
996 	} else {
997 		if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)
998 			flow_act.action |=
999 				MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1000 		if (is_egress)
1001 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1002 		else if (dest_num)
1003 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1004 	}
1005 
1006 	if ((spec->flow_context.flags & FLOW_CONTEXT_HAS_TAG)  &&
1007 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1008 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1009 		mlx5_ib_warn(dev, "Flow tag %u and attribute type %x isn't allowed in leftovers\n",
1010 			     spec->flow_context.flow_tag, flow_attr->type);
1011 		err = -EINVAL;
1012 		goto free;
1013 	}
1014 	handler->rule = mlx5_add_flow_rules(ft, spec,
1015 					    &flow_act,
1016 					    rule_dst, dest_num);
1017 
1018 	if (IS_ERR(handler->rule)) {
1019 		err = PTR_ERR(handler->rule);
1020 		goto free;
1021 	}
1022 
1023 	ft_prio->refcount++;
1024 	handler->prio = ft_prio;
1025 	handler->dev = dev;
1026 
1027 	ft_prio->flow_table = ft;
1028 free:
1029 	if (err && handler) {
1030 		mlx5_ib_counters_clear_description(handler->ibcounters);
1031 		kfree(handler);
1032 	}
1033 	kvfree(spec);
1034 	return err ? ERR_PTR(err) : handler;
1035 }
1036 
1037 static struct mlx5_ib_flow_handler *create_flow_rule(struct mlx5_ib_dev *dev,
1038 						     struct mlx5_ib_flow_prio *ft_prio,
1039 						     const struct ib_flow_attr *flow_attr,
1040 						     struct mlx5_flow_destination *dst)
1041 {
1042 	return _create_flow_rule(dev, ft_prio, flow_attr, dst, 0, NULL);
1043 }
1044 
1045 enum {
1046 	LEFTOVERS_MC,
1047 	LEFTOVERS_UC,
1048 };
1049 
1050 static struct mlx5_ib_flow_handler *create_leftovers_rule(struct mlx5_ib_dev *dev,
1051 							  struct mlx5_ib_flow_prio *ft_prio,
1052 							  struct ib_flow_attr *flow_attr,
1053 							  struct mlx5_flow_destination *dst)
1054 {
1055 	struct mlx5_ib_flow_handler *handler_ucast = NULL;
1056 	struct mlx5_ib_flow_handler *handler = NULL;
1057 
1058 	static struct {
1059 		struct ib_flow_attr	flow_attr;
1060 		struct ib_flow_spec_eth eth_flow;
1061 	} leftovers_specs[] = {
1062 		[LEFTOVERS_MC] = {
1063 			.flow_attr = {
1064 				.num_of_specs = 1,
1065 				.size = sizeof(leftovers_specs[0])
1066 			},
1067 			.eth_flow = {
1068 				.type = IB_FLOW_SPEC_ETH,
1069 				.size = sizeof(struct ib_flow_spec_eth),
1070 				.mask = {.dst_mac = {0x1} },
1071 				.val =  {.dst_mac = {0x1} }
1072 			}
1073 		},
1074 		[LEFTOVERS_UC] = {
1075 			.flow_attr = {
1076 				.num_of_specs = 1,
1077 				.size = sizeof(leftovers_specs[0])
1078 			},
1079 			.eth_flow = {
1080 				.type = IB_FLOW_SPEC_ETH,
1081 				.size = sizeof(struct ib_flow_spec_eth),
1082 				.mask = {.dst_mac = {0x1} },
1083 				.val = {.dst_mac = {} }
1084 			}
1085 		}
1086 	};
1087 
1088 	handler = create_flow_rule(dev, ft_prio,
1089 				   &leftovers_specs[LEFTOVERS_MC].flow_attr,
1090 				   dst);
1091 	if (!IS_ERR(handler) &&
1092 	    flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT) {
1093 		handler_ucast = create_flow_rule(dev, ft_prio,
1094 						 &leftovers_specs[LEFTOVERS_UC].flow_attr,
1095 						 dst);
1096 		if (IS_ERR(handler_ucast)) {
1097 			mlx5_del_flow_rules(handler->rule);
1098 			ft_prio->refcount--;
1099 			kfree(handler);
1100 			handler = handler_ucast;
1101 		} else {
1102 			list_add(&handler_ucast->list, &handler->list);
1103 		}
1104 	}
1105 
1106 	return handler;
1107 }
1108 
1109 static struct mlx5_ib_flow_handler *create_sniffer_rule(struct mlx5_ib_dev *dev,
1110 							struct mlx5_ib_flow_prio *ft_rx,
1111 							struct mlx5_ib_flow_prio *ft_tx,
1112 							struct mlx5_flow_destination *dst)
1113 {
1114 	struct mlx5_ib_flow_handler *handler_rx;
1115 	struct mlx5_ib_flow_handler *handler_tx;
1116 	int err;
1117 	static const struct ib_flow_attr flow_attr  = {
1118 		.num_of_specs = 0,
1119 		.size = sizeof(flow_attr)
1120 	};
1121 
1122 	handler_rx = create_flow_rule(dev, ft_rx, &flow_attr, dst);
1123 	if (IS_ERR(handler_rx)) {
1124 		err = PTR_ERR(handler_rx);
1125 		goto err;
1126 	}
1127 
1128 	handler_tx = create_flow_rule(dev, ft_tx, &flow_attr, dst);
1129 	if (IS_ERR(handler_tx)) {
1130 		err = PTR_ERR(handler_tx);
1131 		goto err_tx;
1132 	}
1133 
1134 	list_add(&handler_tx->list, &handler_rx->list);
1135 
1136 	return handler_rx;
1137 
1138 err_tx:
1139 	mlx5_del_flow_rules(handler_rx->rule);
1140 	ft_rx->refcount--;
1141 	kfree(handler_rx);
1142 err:
1143 	return ERR_PTR(err);
1144 }
1145 
1146 
1147 static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
1148 					   struct ib_flow_attr *flow_attr,
1149 					   int domain,
1150 					   struct ib_udata *udata)
1151 {
1152 	struct mlx5_ib_dev *dev = to_mdev(qp->device);
1153 	struct mlx5_ib_qp *mqp = to_mqp(qp);
1154 	struct mlx5_ib_flow_handler *handler = NULL;
1155 	struct mlx5_flow_destination *dst = NULL;
1156 	struct mlx5_ib_flow_prio *ft_prio_tx = NULL;
1157 	struct mlx5_ib_flow_prio *ft_prio;
1158 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
1159 	struct mlx5_ib_create_flow *ucmd = NULL, ucmd_hdr;
1160 	size_t min_ucmd_sz, required_ucmd_sz;
1161 	int err;
1162 	int underlay_qpn;
1163 
1164 	if (udata && udata->inlen) {
1165 		min_ucmd_sz = offsetof(typeof(ucmd_hdr), reserved) +
1166 				sizeof(ucmd_hdr.reserved);
1167 		if (udata->inlen < min_ucmd_sz)
1168 			return ERR_PTR(-EOPNOTSUPP);
1169 
1170 		err = ib_copy_from_udata(&ucmd_hdr, udata, min_ucmd_sz);
1171 		if (err)
1172 			return ERR_PTR(err);
1173 
1174 		/* currently supports only one counters data */
1175 		if (ucmd_hdr.ncounters_data > 1)
1176 			return ERR_PTR(-EINVAL);
1177 
1178 		required_ucmd_sz = min_ucmd_sz +
1179 			sizeof(struct mlx5_ib_flow_counters_data) *
1180 			ucmd_hdr.ncounters_data;
1181 		if (udata->inlen > required_ucmd_sz &&
1182 		    !ib_is_udata_cleared(udata, required_ucmd_sz,
1183 					 udata->inlen - required_ucmd_sz))
1184 			return ERR_PTR(-EOPNOTSUPP);
1185 
1186 		ucmd = kzalloc(required_ucmd_sz, GFP_KERNEL);
1187 		if (!ucmd)
1188 			return ERR_PTR(-ENOMEM);
1189 
1190 		err = ib_copy_from_udata(ucmd, udata, required_ucmd_sz);
1191 		if (err)
1192 			goto free_ucmd;
1193 	}
1194 
1195 	if (flow_attr->priority > MLX5_IB_FLOW_LAST_PRIO) {
1196 		err = -ENOMEM;
1197 		goto free_ucmd;
1198 	}
1199 
1200 	if (domain != IB_FLOW_DOMAIN_USER ||
1201 	    flow_attr->port > dev->num_ports ||
1202 	    (flow_attr->flags & ~(IB_FLOW_ATTR_FLAGS_DONT_TRAP |
1203 				  IB_FLOW_ATTR_FLAGS_EGRESS))) {
1204 		err = -EINVAL;
1205 		goto free_ucmd;
1206 	}
1207 
1208 	if (is_egress &&
1209 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1210 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1211 		err = -EINVAL;
1212 		goto free_ucmd;
1213 	}
1214 
1215 	dst = kzalloc(sizeof(*dst), GFP_KERNEL);
1216 	if (!dst) {
1217 		err = -ENOMEM;
1218 		goto free_ucmd;
1219 	}
1220 
1221 	mutex_lock(&dev->flow_db->lock);
1222 
1223 	ft_prio = get_flow_table(dev, flow_attr,
1224 				 is_egress ? MLX5_IB_FT_TX : MLX5_IB_FT_RX);
1225 	if (IS_ERR(ft_prio)) {
1226 		err = PTR_ERR(ft_prio);
1227 		goto unlock;
1228 	}
1229 	if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
1230 		ft_prio_tx = get_flow_table(dev, flow_attr, MLX5_IB_FT_TX);
1231 		if (IS_ERR(ft_prio_tx)) {
1232 			err = PTR_ERR(ft_prio_tx);
1233 			ft_prio_tx = NULL;
1234 			goto destroy_ft;
1235 		}
1236 	}
1237 
1238 	if (is_egress) {
1239 		dst->type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1240 	} else {
1241 		dst->type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1242 		if (mqp->is_rss)
1243 			dst->tir_num = mqp->rss_qp.tirn;
1244 		else
1245 			dst->tir_num = mqp->raw_packet_qp.rq.tirn;
1246 	}
1247 
1248 	if (flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1249 		underlay_qpn = (mqp->flags & IB_QP_CREATE_SOURCE_QPN) ?
1250 				       mqp->underlay_qpn :
1251 				       0;
1252 		handler = _create_flow_rule(dev, ft_prio, flow_attr, dst,
1253 					    underlay_qpn, ucmd);
1254 	} else if (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1255 		   flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT) {
1256 		handler = create_leftovers_rule(dev, ft_prio, flow_attr,
1257 						dst);
1258 	} else if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
1259 		handler = create_sniffer_rule(dev, ft_prio, ft_prio_tx, dst);
1260 	} else {
1261 		err = -EINVAL;
1262 		goto destroy_ft;
1263 	}
1264 
1265 	if (IS_ERR(handler)) {
1266 		err = PTR_ERR(handler);
1267 		handler = NULL;
1268 		goto destroy_ft;
1269 	}
1270 
1271 	mutex_unlock(&dev->flow_db->lock);
1272 	kfree(dst);
1273 	kfree(ucmd);
1274 
1275 	return &handler->ibflow;
1276 
1277 destroy_ft:
1278 	put_flow_table(dev, ft_prio, false);
1279 	if (ft_prio_tx)
1280 		put_flow_table(dev, ft_prio_tx, false);
1281 unlock:
1282 	mutex_unlock(&dev->flow_db->lock);
1283 	kfree(dst);
1284 free_ucmd:
1285 	kfree(ucmd);
1286 	return ERR_PTR(err);
1287 }
1288 
1289 static struct mlx5_ib_flow_prio *
1290 _get_flow_table(struct mlx5_ib_dev *dev,
1291 		struct mlx5_ib_flow_matcher *fs_matcher,
1292 		bool mcast)
1293 {
1294 	struct mlx5_flow_namespace *ns = NULL;
1295 	struct mlx5_ib_flow_prio *prio = NULL;
1296 	int max_table_size = 0;
1297 	bool esw_encap;
1298 	u32 flags = 0;
1299 	int priority;
1300 
1301 	if (mcast)
1302 		priority = MLX5_IB_FLOW_MCAST_PRIO;
1303 	else
1304 		priority = ib_prio_to_core_prio(fs_matcher->priority, false);
1305 
1306 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
1307 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
1308 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_BYPASS) {
1309 		max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
1310 					log_max_ft_size));
1311 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap) && !esw_encap)
1312 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1313 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
1314 					      reformat_l3_tunnel_to_l2) &&
1315 		    !esw_encap)
1316 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1317 	} else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS) {
1318 		max_table_size = BIT(
1319 			MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, log_max_ft_size));
1320 		if (MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat) && !esw_encap)
1321 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1322 	} else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB) {
1323 		max_table_size = BIT(
1324 			MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, log_max_ft_size));
1325 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, decap) && esw_encap)
1326 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1327 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, reformat_l3_tunnel_to_l2) &&
1328 		    esw_encap)
1329 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1330 		priority = FDB_BYPASS_PATH;
1331 	} else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
1332 		max_table_size =
1333 			BIT(MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev,
1334 						       log_max_ft_size));
1335 		priority = fs_matcher->priority;
1336 	} else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
1337 		max_table_size =
1338 			BIT(MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
1339 						       log_max_ft_size));
1340 		priority = fs_matcher->priority;
1341 	}
1342 
1343 	max_table_size = min_t(int, max_table_size, MLX5_FS_MAX_ENTRIES);
1344 
1345 	ns = mlx5_get_flow_namespace(dev->mdev, fs_matcher->ns_type);
1346 	if (!ns)
1347 		return ERR_PTR(-EOPNOTSUPP);
1348 
1349 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_BYPASS)
1350 		prio = &dev->flow_db->prios[priority];
1351 	else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS)
1352 		prio = &dev->flow_db->egress_prios[priority];
1353 	else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB)
1354 		prio = &dev->flow_db->fdb;
1355 	else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX)
1356 		prio = &dev->flow_db->rdma_rx[priority];
1357 	else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX)
1358 		prio = &dev->flow_db->rdma_tx[priority];
1359 
1360 	if (!prio)
1361 		return ERR_PTR(-EINVAL);
1362 
1363 	if (prio->flow_table)
1364 		return prio;
1365 
1366 	return _get_prio(ns, prio, priority, max_table_size,
1367 			 MLX5_FS_MAX_TYPES, flags);
1368 }
1369 
1370 static struct mlx5_ib_flow_handler *
1371 _create_raw_flow_rule(struct mlx5_ib_dev *dev,
1372 		      struct mlx5_ib_flow_prio *ft_prio,
1373 		      struct mlx5_flow_destination *dst,
1374 		      struct mlx5_ib_flow_matcher  *fs_matcher,
1375 		      struct mlx5_flow_context *flow_context,
1376 		      struct mlx5_flow_act *flow_act,
1377 		      void *cmd_in, int inlen,
1378 		      int dst_num)
1379 {
1380 	struct mlx5_ib_flow_handler *handler;
1381 	struct mlx5_flow_spec *spec;
1382 	struct mlx5_flow_table *ft = ft_prio->flow_table;
1383 	int err = 0;
1384 
1385 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1386 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
1387 	if (!handler || !spec) {
1388 		err = -ENOMEM;
1389 		goto free;
1390 	}
1391 
1392 	INIT_LIST_HEAD(&handler->list);
1393 
1394 	memcpy(spec->match_value, cmd_in, inlen);
1395 	memcpy(spec->match_criteria, fs_matcher->matcher_mask.match_params,
1396 	       fs_matcher->mask_len);
1397 	spec->match_criteria_enable = fs_matcher->match_criteria_enable;
1398 	spec->flow_context = *flow_context;
1399 
1400 	handler->rule = mlx5_add_flow_rules(ft, spec,
1401 					    flow_act, dst, dst_num);
1402 
1403 	if (IS_ERR(handler->rule)) {
1404 		err = PTR_ERR(handler->rule);
1405 		goto free;
1406 	}
1407 
1408 	ft_prio->refcount++;
1409 	handler->prio = ft_prio;
1410 	handler->dev = dev;
1411 	ft_prio->flow_table = ft;
1412 
1413 free:
1414 	if (err)
1415 		kfree(handler);
1416 	kvfree(spec);
1417 	return err ? ERR_PTR(err) : handler;
1418 }
1419 
1420 static bool raw_fs_is_multicast(struct mlx5_ib_flow_matcher *fs_matcher,
1421 				void *match_v)
1422 {
1423 	void *match_c;
1424 	void *match_v_set_lyr_2_4, *match_c_set_lyr_2_4;
1425 	void *dmac, *dmac_mask;
1426 	void *ipv4, *ipv4_mask;
1427 
1428 	if (!(fs_matcher->match_criteria_enable &
1429 	      (1 << MATCH_CRITERIA_ENABLE_OUTER_BIT)))
1430 		return false;
1431 
1432 	match_c = fs_matcher->matcher_mask.match_params;
1433 	match_v_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_v,
1434 					   outer_headers);
1435 	match_c_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_c,
1436 					   outer_headers);
1437 
1438 	dmac = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1439 			    dmac_47_16);
1440 	dmac_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1441 				 dmac_47_16);
1442 
1443 	if (is_multicast_ether_addr(dmac) &&
1444 	    is_multicast_ether_addr(dmac_mask))
1445 		return true;
1446 
1447 	ipv4 = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1448 			    dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1449 
1450 	ipv4_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1451 				 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1452 
1453 	if (ipv4_is_multicast(*(__be32 *)(ipv4)) &&
1454 	    ipv4_is_multicast(*(__be32 *)(ipv4_mask)))
1455 		return true;
1456 
1457 	return false;
1458 }
1459 
1460 static struct mlx5_ib_flow_handler *raw_fs_rule_add(
1461 	struct mlx5_ib_dev *dev, struct mlx5_ib_flow_matcher *fs_matcher,
1462 	struct mlx5_flow_context *flow_context, struct mlx5_flow_act *flow_act,
1463 	u32 counter_id, void *cmd_in, int inlen, int dest_id, int dest_type)
1464 {
1465 	struct mlx5_flow_destination *dst;
1466 	struct mlx5_ib_flow_prio *ft_prio;
1467 	struct mlx5_ib_flow_handler *handler;
1468 	int dst_num = 0;
1469 	bool mcast;
1470 	int err;
1471 
1472 	if (fs_matcher->flow_type != MLX5_IB_FLOW_TYPE_NORMAL)
1473 		return ERR_PTR(-EOPNOTSUPP);
1474 
1475 	if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO)
1476 		return ERR_PTR(-ENOMEM);
1477 
1478 	dst = kcalloc(2, sizeof(*dst), GFP_KERNEL);
1479 	if (!dst)
1480 		return ERR_PTR(-ENOMEM);
1481 
1482 	mcast = raw_fs_is_multicast(fs_matcher, cmd_in);
1483 	mutex_lock(&dev->flow_db->lock);
1484 
1485 	ft_prio = _get_flow_table(dev, fs_matcher, mcast);
1486 	if (IS_ERR(ft_prio)) {
1487 		err = PTR_ERR(ft_prio);
1488 		goto unlock;
1489 	}
1490 
1491 	if (dest_type == MLX5_FLOW_DESTINATION_TYPE_TIR) {
1492 		dst[dst_num].type = dest_type;
1493 		dst[dst_num++].tir_num = dest_id;
1494 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1495 	} else if (dest_type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) {
1496 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM;
1497 		dst[dst_num++].ft_num = dest_id;
1498 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1499 	} else  if (dest_type == MLX5_FLOW_DESTINATION_TYPE_PORT) {
1500 		dst[dst_num++].type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1501 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1502 	}
1503 
1504 
1505 	if (flow_act->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1506 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1507 		dst[dst_num].counter_id = counter_id;
1508 		dst_num++;
1509 	}
1510 
1511 	handler = _create_raw_flow_rule(dev, ft_prio, dst, fs_matcher,
1512 					flow_context, flow_act,
1513 					cmd_in, inlen, dst_num);
1514 
1515 	if (IS_ERR(handler)) {
1516 		err = PTR_ERR(handler);
1517 		goto destroy_ft;
1518 	}
1519 
1520 	mutex_unlock(&dev->flow_db->lock);
1521 	atomic_inc(&fs_matcher->usecnt);
1522 	handler->flow_matcher = fs_matcher;
1523 
1524 	kfree(dst);
1525 
1526 	return handler;
1527 
1528 destroy_ft:
1529 	put_flow_table(dev, ft_prio, false);
1530 unlock:
1531 	mutex_unlock(&dev->flow_db->lock);
1532 	kfree(dst);
1533 
1534 	return ERR_PTR(err);
1535 }
1536 
1537 static u32 mlx5_ib_flow_action_flags_to_accel_xfrm_flags(u32 mlx5_flags)
1538 {
1539 	u32 flags = 0;
1540 
1541 	if (mlx5_flags & MLX5_IB_UAPI_FLOW_ACTION_FLAGS_REQUIRE_METADATA)
1542 		flags |= MLX5_ACCEL_XFRM_FLAG_REQUIRE_METADATA;
1543 
1544 	return flags;
1545 }
1546 
1547 #define MLX5_FLOW_ACTION_ESP_CREATE_LAST_SUPPORTED                             \
1548 	MLX5_IB_UAPI_FLOW_ACTION_FLAGS_REQUIRE_METADATA
1549 static struct ib_flow_action *
1550 mlx5_ib_create_flow_action_esp(struct ib_device *device,
1551 			       const struct ib_flow_action_attrs_esp *attr,
1552 			       struct uverbs_attr_bundle *attrs)
1553 {
1554 	struct mlx5_ib_dev *mdev = to_mdev(device);
1555 	struct ib_uverbs_flow_action_esp_keymat_aes_gcm *aes_gcm;
1556 	struct mlx5_accel_esp_xfrm_attrs accel_attrs = {};
1557 	struct mlx5_ib_flow_action *action;
1558 	u64 action_flags;
1559 	u64 flags;
1560 	int err = 0;
1561 
1562 	err = uverbs_get_flags64(
1563 		&action_flags, attrs, MLX5_IB_ATTR_CREATE_FLOW_ACTION_FLAGS,
1564 		((MLX5_FLOW_ACTION_ESP_CREATE_LAST_SUPPORTED << 1) - 1));
1565 	if (err)
1566 		return ERR_PTR(err);
1567 
1568 	flags = mlx5_ib_flow_action_flags_to_accel_xfrm_flags(action_flags);
1569 
1570 	/* We current only support a subset of the standard features. Only a
1571 	 * keymat of type AES_GCM, with icv_len == 16, iv_algo == SEQ and esn
1572 	 * (with overlap). Full offload mode isn't supported.
1573 	 */
1574 	if (!attr->keymat || attr->replay || attr->encap ||
1575 	    attr->spi || attr->seq || attr->tfc_pad ||
1576 	    attr->hard_limit_pkts ||
1577 	    (attr->flags & ~(IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED |
1578 			     IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ENCRYPT)))
1579 		return ERR_PTR(-EOPNOTSUPP);
1580 
1581 	if (attr->keymat->protocol !=
1582 	    IB_UVERBS_FLOW_ACTION_ESP_KEYMAT_AES_GCM)
1583 		return ERR_PTR(-EOPNOTSUPP);
1584 
1585 	aes_gcm = &attr->keymat->keymat.aes_gcm;
1586 
1587 	if (aes_gcm->icv_len != 16 ||
1588 	    aes_gcm->iv_algo != IB_UVERBS_FLOW_ACTION_IV_ALGO_SEQ)
1589 		return ERR_PTR(-EOPNOTSUPP);
1590 
1591 	action = kmalloc(sizeof(*action), GFP_KERNEL);
1592 	if (!action)
1593 		return ERR_PTR(-ENOMEM);
1594 
1595 	action->esp_aes_gcm.ib_flags = attr->flags;
1596 	memcpy(&accel_attrs.keymat.aes_gcm.aes_key, &aes_gcm->aes_key,
1597 	       sizeof(accel_attrs.keymat.aes_gcm.aes_key));
1598 	accel_attrs.keymat.aes_gcm.key_len = aes_gcm->key_len * 8;
1599 	memcpy(&accel_attrs.keymat.aes_gcm.salt, &aes_gcm->salt,
1600 	       sizeof(accel_attrs.keymat.aes_gcm.salt));
1601 	memcpy(&accel_attrs.keymat.aes_gcm.seq_iv, &aes_gcm->iv,
1602 	       sizeof(accel_attrs.keymat.aes_gcm.seq_iv));
1603 	accel_attrs.keymat.aes_gcm.icv_len = aes_gcm->icv_len * 8;
1604 	accel_attrs.keymat.aes_gcm.iv_algo = MLX5_ACCEL_ESP_AES_GCM_IV_ALGO_SEQ;
1605 	accel_attrs.keymat_type = MLX5_ACCEL_ESP_KEYMAT_AES_GCM;
1606 
1607 	accel_attrs.esn = attr->esn;
1608 	if (attr->flags & IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED)
1609 		accel_attrs.flags |= MLX5_ACCEL_ESP_FLAGS_ESN_TRIGGERED;
1610 	if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)
1611 		accel_attrs.flags |= MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
1612 
1613 	if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ENCRYPT)
1614 		accel_attrs.action |= MLX5_ACCEL_ESP_ACTION_ENCRYPT;
1615 
1616 	action->esp_aes_gcm.ctx =
1617 		mlx5_accel_esp_create_xfrm(mdev->mdev, &accel_attrs, flags);
1618 	if (IS_ERR(action->esp_aes_gcm.ctx)) {
1619 		err = PTR_ERR(action->esp_aes_gcm.ctx);
1620 		goto err_parse;
1621 	}
1622 
1623 	action->esp_aes_gcm.ib_flags = attr->flags;
1624 
1625 	return &action->ib_action;
1626 
1627 err_parse:
1628 	kfree(action);
1629 	return ERR_PTR(err);
1630 }
1631 
1632 static int
1633 mlx5_ib_modify_flow_action_esp(struct ib_flow_action *action,
1634 			       const struct ib_flow_action_attrs_esp *attr,
1635 			       struct uverbs_attr_bundle *attrs)
1636 {
1637 	struct mlx5_ib_flow_action *maction = to_mflow_act(action);
1638 	struct mlx5_accel_esp_xfrm_attrs accel_attrs;
1639 	int err = 0;
1640 
1641 	if (attr->keymat || attr->replay || attr->encap ||
1642 	    attr->spi || attr->seq || attr->tfc_pad ||
1643 	    attr->hard_limit_pkts ||
1644 	    (attr->flags & ~(IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED |
1645 			     IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS |
1646 			     IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)))
1647 		return -EOPNOTSUPP;
1648 
1649 	/* Only the ESN value or the MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP can
1650 	 * be modified.
1651 	 */
1652 	if (!(maction->esp_aes_gcm.ib_flags &
1653 	      IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED) &&
1654 	    attr->flags & (IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED |
1655 			   IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW))
1656 		return -EINVAL;
1657 
1658 	memcpy(&accel_attrs, &maction->esp_aes_gcm.ctx->attrs,
1659 	       sizeof(accel_attrs));
1660 
1661 	accel_attrs.esn = attr->esn;
1662 	if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)
1663 		accel_attrs.flags |= MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
1664 	else
1665 		accel_attrs.flags &= ~MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
1666 
1667 	err = mlx5_accel_esp_modify_xfrm(maction->esp_aes_gcm.ctx,
1668 					 &accel_attrs);
1669 	if (err)
1670 		return err;
1671 
1672 	maction->esp_aes_gcm.ib_flags &=
1673 		~IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW;
1674 	maction->esp_aes_gcm.ib_flags |=
1675 		attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW;
1676 
1677 	return 0;
1678 }
1679 
1680 static void destroy_flow_action_raw(struct mlx5_ib_flow_action *maction)
1681 {
1682 	switch (maction->flow_action_raw.sub_type) {
1683 	case MLX5_IB_FLOW_ACTION_MODIFY_HEADER:
1684 		mlx5_modify_header_dealloc(maction->flow_action_raw.dev->mdev,
1685 					   maction->flow_action_raw.modify_hdr);
1686 		break;
1687 	case MLX5_IB_FLOW_ACTION_PACKET_REFORMAT:
1688 		mlx5_packet_reformat_dealloc(maction->flow_action_raw.dev->mdev,
1689 					     maction->flow_action_raw.pkt_reformat);
1690 		break;
1691 	case MLX5_IB_FLOW_ACTION_DECAP:
1692 		break;
1693 	default:
1694 		break;
1695 	}
1696 }
1697 
1698 static int mlx5_ib_destroy_flow_action(struct ib_flow_action *action)
1699 {
1700 	struct mlx5_ib_flow_action *maction = to_mflow_act(action);
1701 
1702 	switch (action->type) {
1703 	case IB_FLOW_ACTION_ESP:
1704 		/*
1705 		 * We only support aes_gcm by now, so we implicitly know this is
1706 		 * the underline crypto.
1707 		 */
1708 		mlx5_accel_esp_destroy_xfrm(maction->esp_aes_gcm.ctx);
1709 		break;
1710 	case IB_FLOW_ACTION_UNSPECIFIED:
1711 		destroy_flow_action_raw(maction);
1712 		break;
1713 	default:
1714 		WARN_ON(true);
1715 		break;
1716 	}
1717 
1718 	kfree(maction);
1719 	return 0;
1720 }
1721 
1722 static int
1723 mlx5_ib_ft_type_to_namespace(enum mlx5_ib_uapi_flow_table_type table_type,
1724 			     enum mlx5_flow_namespace_type *namespace)
1725 {
1726 	switch (table_type) {
1727 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX:
1728 		*namespace = MLX5_FLOW_NAMESPACE_BYPASS;
1729 		break;
1730 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX:
1731 		*namespace = MLX5_FLOW_NAMESPACE_EGRESS;
1732 		break;
1733 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_FDB:
1734 		*namespace = MLX5_FLOW_NAMESPACE_FDB;
1735 		break;
1736 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_RX:
1737 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_RX;
1738 		break;
1739 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_TX:
1740 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_TX;
1741 		break;
1742 	default:
1743 		return -EINVAL;
1744 	}
1745 
1746 	return 0;
1747 }
1748 
1749 static const struct uverbs_attr_spec mlx5_ib_flow_type[] = {
1750 	[MLX5_IB_FLOW_TYPE_NORMAL] = {
1751 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1752 		.u.ptr = {
1753 			.len = sizeof(u16), /* data is priority */
1754 			.min_len = sizeof(u16),
1755 		}
1756 	},
1757 	[MLX5_IB_FLOW_TYPE_SNIFFER] = {
1758 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1759 		UVERBS_ATTR_NO_DATA(),
1760 	},
1761 	[MLX5_IB_FLOW_TYPE_ALL_DEFAULT] = {
1762 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1763 		UVERBS_ATTR_NO_DATA(),
1764 	},
1765 	[MLX5_IB_FLOW_TYPE_MC_DEFAULT] = {
1766 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1767 		UVERBS_ATTR_NO_DATA(),
1768 	},
1769 };
1770 
1771 static bool is_flow_dest(void *obj, int *dest_id, int *dest_type)
1772 {
1773 	struct devx_obj *devx_obj = obj;
1774 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1775 
1776 	switch (opcode) {
1777 	case MLX5_CMD_OP_DESTROY_TIR:
1778 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1779 		*dest_id = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox,
1780 				    obj_id);
1781 		return true;
1782 
1783 	case MLX5_CMD_OP_DESTROY_FLOW_TABLE:
1784 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1785 		*dest_id = MLX5_GET(destroy_flow_table_in, devx_obj->dinbox,
1786 				    table_id);
1787 		return true;
1788 	default:
1789 		return false;
1790 	}
1791 }
1792 
1793 static int get_dests(struct uverbs_attr_bundle *attrs,
1794 		     struct mlx5_ib_flow_matcher *fs_matcher, int *dest_id,
1795 		     int *dest_type, struct ib_qp **qp, u32 *flags)
1796 {
1797 	bool dest_devx, dest_qp;
1798 	void *devx_obj;
1799 	int err;
1800 
1801 	dest_devx = uverbs_attr_is_valid(attrs,
1802 					 MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1803 	dest_qp = uverbs_attr_is_valid(attrs,
1804 				       MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1805 
1806 	*flags = 0;
1807 	err = uverbs_get_flags32(flags, attrs, MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
1808 				 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS |
1809 					 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP);
1810 	if (err)
1811 		return err;
1812 
1813 	/* Both flags are not allowed */
1814 	if (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS &&
1815 	    *flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1816 		return -EINVAL;
1817 
1818 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_BYPASS) {
1819 		if (dest_devx && (dest_qp || *flags))
1820 			return -EINVAL;
1821 		else if (dest_qp && *flags)
1822 			return -EINVAL;
1823 	}
1824 
1825 	/* Allow only DEVX object, drop as dest for FDB */
1826 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB && !(dest_devx ||
1827 	     (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)))
1828 		return -EINVAL;
1829 
1830 	/* Allow only DEVX object or QP as dest when inserting to RDMA_RX */
1831 	if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1832 	    ((!dest_devx && !dest_qp) || (dest_devx && dest_qp)))
1833 		return -EINVAL;
1834 
1835 	*qp = NULL;
1836 	if (dest_devx) {
1837 		devx_obj =
1838 			uverbs_attr_get_obj(attrs,
1839 					    MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1840 
1841 		/* Verify that the given DEVX object is a flow
1842 		 * steering destination.
1843 		 */
1844 		if (!is_flow_dest(devx_obj, dest_id, dest_type))
1845 			return -EINVAL;
1846 		/* Allow only flow table as dest when inserting to FDB or RDMA_RX */
1847 		if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB ||
1848 		     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1849 		    *dest_type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1850 			return -EINVAL;
1851 	} else if (dest_qp) {
1852 		struct mlx5_ib_qp *mqp;
1853 
1854 		*qp = uverbs_attr_get_obj(attrs,
1855 					  MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1856 		if (IS_ERR(*qp))
1857 			return PTR_ERR(*qp);
1858 
1859 		if ((*qp)->qp_type != IB_QPT_RAW_PACKET)
1860 			return -EINVAL;
1861 
1862 		mqp = to_mqp(*qp);
1863 		if (mqp->is_rss)
1864 			*dest_id = mqp->rss_qp.tirn;
1865 		else
1866 			*dest_id = mqp->raw_packet_qp.rq.tirn;
1867 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1868 	} else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1869 		   fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
1870 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1871 	}
1872 
1873 	if (*dest_type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1874 	    (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1875 	     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX))
1876 		return -EINVAL;
1877 
1878 	return 0;
1879 }
1880 
1881 static bool is_flow_counter(void *obj, u32 offset, u32 *counter_id)
1882 {
1883 	struct devx_obj *devx_obj = obj;
1884 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1885 
1886 	if (opcode == MLX5_CMD_OP_DEALLOC_FLOW_COUNTER) {
1887 
1888 		if (offset && offset >= devx_obj->flow_counter_bulk_size)
1889 			return false;
1890 
1891 		*counter_id = MLX5_GET(dealloc_flow_counter_in,
1892 				       devx_obj->dinbox,
1893 				       flow_counter_id);
1894 		*counter_id += offset;
1895 		return true;
1896 	}
1897 
1898 	return false;
1899 }
1900 
1901 #define MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS 2
1902 static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)(
1903 	struct uverbs_attr_bundle *attrs)
1904 {
1905 	struct mlx5_flow_context flow_context = {.flow_tag =
1906 		MLX5_FS_DEFAULT_FLOW_TAG};
1907 	u32 *offset_attr, offset = 0, counter_id = 0;
1908 	int dest_id, dest_type = -1, inlen, len, ret, i;
1909 	struct mlx5_ib_flow_handler *flow_handler;
1910 	struct mlx5_ib_flow_matcher *fs_matcher;
1911 	struct ib_uobject **arr_flow_actions;
1912 	struct ib_uflow_resources *uflow_res;
1913 	struct mlx5_flow_act flow_act = {};
1914 	struct ib_qp *qp = NULL;
1915 	void *devx_obj, *cmd_in;
1916 	struct ib_uobject *uobj;
1917 	struct mlx5_ib_dev *dev;
1918 	u32 flags;
1919 
1920 	if (!capable(CAP_NET_RAW))
1921 		return -EPERM;
1922 
1923 	fs_matcher = uverbs_attr_get_obj(attrs,
1924 					 MLX5_IB_ATTR_CREATE_FLOW_MATCHER);
1925 	uobj =  uverbs_attr_get_uobject(attrs, MLX5_IB_ATTR_CREATE_FLOW_HANDLE);
1926 	dev = mlx5_udata_to_mdev(&attrs->driver_udata);
1927 
1928 	if (get_dests(attrs, fs_matcher, &dest_id, &dest_type, &qp, &flags))
1929 		return -EINVAL;
1930 
1931 	if (flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS)
1932 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
1933 
1934 	if (flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1935 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1936 
1937 	len = uverbs_attr_get_uobjs_arr(attrs,
1938 		MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX, &arr_flow_actions);
1939 	if (len) {
1940 		devx_obj = arr_flow_actions[0]->object;
1941 
1942 		if (uverbs_attr_is_valid(attrs,
1943 					 MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET)) {
1944 
1945 			int num_offsets = uverbs_attr_ptr_get_array_size(
1946 				attrs,
1947 				MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
1948 				sizeof(u32));
1949 
1950 			if (num_offsets != 1)
1951 				return -EINVAL;
1952 
1953 			offset_attr = uverbs_attr_get_alloced_ptr(
1954 				attrs,
1955 				MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET);
1956 			offset = *offset_attr;
1957 		}
1958 
1959 		if (!is_flow_counter(devx_obj, offset, &counter_id))
1960 			return -EINVAL;
1961 
1962 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
1963 	}
1964 
1965 	cmd_in = uverbs_attr_get_alloced_ptr(
1966 		attrs, MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE);
1967 	inlen = uverbs_attr_get_len(attrs,
1968 				    MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE);
1969 
1970 	uflow_res = flow_resources_alloc(MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS);
1971 	if (!uflow_res)
1972 		return -ENOMEM;
1973 
1974 	len = uverbs_attr_get_uobjs_arr(attrs,
1975 		MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS, &arr_flow_actions);
1976 	for (i = 0; i < len; i++) {
1977 		struct mlx5_ib_flow_action *maction =
1978 			to_mflow_act(arr_flow_actions[i]->object);
1979 
1980 		ret = parse_flow_flow_action(maction, false, &flow_act);
1981 		if (ret)
1982 			goto err_out;
1983 		flow_resources_add(uflow_res, IB_FLOW_SPEC_ACTION_HANDLE,
1984 				   arr_flow_actions[i]->object);
1985 	}
1986 
1987 	ret = uverbs_copy_from(&flow_context.flow_tag, attrs,
1988 			       MLX5_IB_ATTR_CREATE_FLOW_TAG);
1989 	if (!ret) {
1990 		if (flow_context.flow_tag >= BIT(24)) {
1991 			ret = -EINVAL;
1992 			goto err_out;
1993 		}
1994 		flow_context.flags |= FLOW_CONTEXT_HAS_TAG;
1995 	}
1996 
1997 	flow_handler =
1998 		raw_fs_rule_add(dev, fs_matcher, &flow_context, &flow_act,
1999 				counter_id, cmd_in, inlen, dest_id, dest_type);
2000 	if (IS_ERR(flow_handler)) {
2001 		ret = PTR_ERR(flow_handler);
2002 		goto err_out;
2003 	}
2004 
2005 	ib_set_flow(uobj, &flow_handler->ibflow, qp, &dev->ib_dev, uflow_res);
2006 
2007 	return 0;
2008 err_out:
2009 	ib_uverbs_flow_resources_free(uflow_res);
2010 	return ret;
2011 }
2012 
2013 static int flow_matcher_cleanup(struct ib_uobject *uobject,
2014 				enum rdma_remove_reason why,
2015 				struct uverbs_attr_bundle *attrs)
2016 {
2017 	struct mlx5_ib_flow_matcher *obj = uobject->object;
2018 	int ret;
2019 
2020 	ret = ib_destroy_usecnt(&obj->usecnt, why, uobject);
2021 	if (ret)
2022 		return ret;
2023 
2024 	kfree(obj);
2025 	return 0;
2026 }
2027 
2028 static int mlx5_ib_matcher_ns(struct uverbs_attr_bundle *attrs,
2029 			      struct mlx5_ib_flow_matcher *obj)
2030 {
2031 	enum mlx5_ib_uapi_flow_table_type ft_type =
2032 		MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX;
2033 	u32 flags;
2034 	int err;
2035 
2036 	/* New users should use MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE and older
2037 	 * users should switch to it. We leave this to not break userspace
2038 	 */
2039 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE) &&
2040 	    uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS))
2041 		return -EINVAL;
2042 
2043 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE)) {
2044 		err = uverbs_get_const(&ft_type, attrs,
2045 				       MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE);
2046 		if (err)
2047 			return err;
2048 
2049 		err = mlx5_ib_ft_type_to_namespace(ft_type, &obj->ns_type);
2050 		if (err)
2051 			return err;
2052 
2053 		return 0;
2054 	}
2055 
2056 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS)) {
2057 		err = uverbs_get_flags32(&flags, attrs,
2058 					 MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2059 					 IB_FLOW_ATTR_FLAGS_EGRESS);
2060 		if (err)
2061 			return err;
2062 
2063 		if (flags) {
2064 			mlx5_ib_ft_type_to_namespace(
2065 				MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX,
2066 				&obj->ns_type);
2067 			return 0;
2068 		}
2069 	}
2070 
2071 	obj->ns_type = MLX5_FLOW_NAMESPACE_BYPASS;
2072 
2073 	return 0;
2074 }
2075 
2076 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_MATCHER_CREATE)(
2077 	struct uverbs_attr_bundle *attrs)
2078 {
2079 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2080 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE);
2081 	struct mlx5_ib_dev *dev = mlx5_udata_to_mdev(&attrs->driver_udata);
2082 	struct mlx5_ib_flow_matcher *obj;
2083 	int err;
2084 
2085 	obj = kzalloc(sizeof(struct mlx5_ib_flow_matcher), GFP_KERNEL);
2086 	if (!obj)
2087 		return -ENOMEM;
2088 
2089 	obj->mask_len = uverbs_attr_get_len(
2090 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2091 	err = uverbs_copy_from(&obj->matcher_mask,
2092 			       attrs,
2093 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2094 	if (err)
2095 		goto end;
2096 
2097 	obj->flow_type = uverbs_attr_get_enum_id(
2098 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2099 
2100 	if (obj->flow_type == MLX5_IB_FLOW_TYPE_NORMAL) {
2101 		err = uverbs_copy_from(&obj->priority,
2102 				       attrs,
2103 				       MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2104 		if (err)
2105 			goto end;
2106 	}
2107 
2108 	err = uverbs_copy_from(&obj->match_criteria_enable,
2109 			       attrs,
2110 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA);
2111 	if (err)
2112 		goto end;
2113 
2114 	err = mlx5_ib_matcher_ns(attrs, obj);
2115 	if (err)
2116 		goto end;
2117 
2118 	uobj->object = obj;
2119 	obj->mdev = dev->mdev;
2120 	atomic_set(&obj->usecnt, 0);
2121 	return 0;
2122 
2123 end:
2124 	kfree(obj);
2125 	return err;
2126 }
2127 
2128 static struct ib_flow_action *
2129 mlx5_ib_create_modify_header(struct mlx5_ib_dev *dev,
2130 			     enum mlx5_ib_uapi_flow_table_type ft_type,
2131 			     u8 num_actions, void *in)
2132 {
2133 	enum mlx5_flow_namespace_type namespace;
2134 	struct mlx5_ib_flow_action *maction;
2135 	int ret;
2136 
2137 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2138 	if (ret)
2139 		return ERR_PTR(-EINVAL);
2140 
2141 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2142 	if (!maction)
2143 		return ERR_PTR(-ENOMEM);
2144 
2145 	maction->flow_action_raw.modify_hdr =
2146 		mlx5_modify_header_alloc(dev->mdev, namespace, num_actions, in);
2147 
2148 	if (IS_ERR(maction->flow_action_raw.modify_hdr)) {
2149 		ret = PTR_ERR(maction->flow_action_raw.modify_hdr);
2150 		kfree(maction);
2151 		return ERR_PTR(ret);
2152 	}
2153 	maction->flow_action_raw.sub_type =
2154 		MLX5_IB_FLOW_ACTION_MODIFY_HEADER;
2155 	maction->flow_action_raw.dev = dev;
2156 
2157 	return &maction->ib_action;
2158 }
2159 
2160 static bool mlx5_ib_modify_header_supported(struct mlx5_ib_dev *dev)
2161 {
2162 	return MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
2163 					 max_modify_header_actions) ||
2164 	       MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev,
2165 					 max_modify_header_actions) ||
2166 	       MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
2167 					 max_modify_header_actions);
2168 }
2169 
2170 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER)(
2171 	struct uverbs_attr_bundle *attrs)
2172 {
2173 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2174 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE);
2175 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2176 	enum mlx5_ib_uapi_flow_table_type ft_type;
2177 	struct ib_flow_action *action;
2178 	int num_actions;
2179 	void *in;
2180 	int ret;
2181 
2182 	if (!mlx5_ib_modify_header_supported(mdev))
2183 		return -EOPNOTSUPP;
2184 
2185 	in = uverbs_attr_get_alloced_ptr(attrs,
2186 		MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM);
2187 
2188 	num_actions = uverbs_attr_ptr_get_array_size(
2189 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2190 		MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto));
2191 	if (num_actions < 0)
2192 		return num_actions;
2193 
2194 	ret = uverbs_get_const(&ft_type, attrs,
2195 			       MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE);
2196 	if (ret)
2197 		return ret;
2198 	action = mlx5_ib_create_modify_header(mdev, ft_type, num_actions, in);
2199 	if (IS_ERR(action))
2200 		return PTR_ERR(action);
2201 
2202 	uverbs_flow_action_fill_action(action, uobj, &mdev->ib_dev,
2203 				       IB_FLOW_ACTION_UNSPECIFIED);
2204 
2205 	return 0;
2206 }
2207 
2208 static bool mlx5_ib_flow_action_packet_reformat_valid(struct mlx5_ib_dev *ibdev,
2209 						      u8 packet_reformat_type,
2210 						      u8 ft_type)
2211 {
2212 	switch (packet_reformat_type) {
2213 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2214 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2215 			return MLX5_CAP_FLOWTABLE(ibdev->mdev,
2216 						  encap_general_header);
2217 		break;
2218 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2219 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2220 			return MLX5_CAP_FLOWTABLE_NIC_TX(ibdev->mdev,
2221 				reformat_l2_to_l3_tunnel);
2222 		break;
2223 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2224 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2225 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev,
2226 				reformat_l3_tunnel_to_l2);
2227 		break;
2228 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2:
2229 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2230 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev, decap);
2231 		break;
2232 	default:
2233 		break;
2234 	}
2235 
2236 	return false;
2237 }
2238 
2239 static int mlx5_ib_dv_to_prm_packet_reforamt_type(u8 dv_prt, u8 *prm_prt)
2240 {
2241 	switch (dv_prt) {
2242 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2243 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L2_TUNNEL;
2244 		break;
2245 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2246 		*prm_prt = MLX5_REFORMAT_TYPE_L3_TUNNEL_TO_L2;
2247 		break;
2248 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2249 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
2250 		break;
2251 	default:
2252 		return -EINVAL;
2253 	}
2254 
2255 	return 0;
2256 }
2257 
2258 static int mlx5_ib_flow_action_create_packet_reformat_ctx(
2259 	struct mlx5_ib_dev *dev,
2260 	struct mlx5_ib_flow_action *maction,
2261 	u8 ft_type, u8 dv_prt,
2262 	void *in, size_t len)
2263 {
2264 	enum mlx5_flow_namespace_type namespace;
2265 	u8 prm_prt;
2266 	int ret;
2267 
2268 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2269 	if (ret)
2270 		return ret;
2271 
2272 	ret = mlx5_ib_dv_to_prm_packet_reforamt_type(dv_prt, &prm_prt);
2273 	if (ret)
2274 		return ret;
2275 
2276 	maction->flow_action_raw.pkt_reformat =
2277 		mlx5_packet_reformat_alloc(dev->mdev, prm_prt, len,
2278 					   in, namespace);
2279 	if (IS_ERR(maction->flow_action_raw.pkt_reformat)) {
2280 		ret = PTR_ERR(maction->flow_action_raw.pkt_reformat);
2281 		return ret;
2282 	}
2283 
2284 	maction->flow_action_raw.sub_type =
2285 		MLX5_IB_FLOW_ACTION_PACKET_REFORMAT;
2286 	maction->flow_action_raw.dev = dev;
2287 
2288 	return 0;
2289 }
2290 
2291 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT)(
2292 	struct uverbs_attr_bundle *attrs)
2293 {
2294 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs,
2295 		MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE);
2296 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2297 	enum mlx5_ib_uapi_flow_action_packet_reformat_type dv_prt;
2298 	enum mlx5_ib_uapi_flow_table_type ft_type;
2299 	struct mlx5_ib_flow_action *maction;
2300 	int ret;
2301 
2302 	ret = uverbs_get_const(&ft_type, attrs,
2303 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE);
2304 	if (ret)
2305 		return ret;
2306 
2307 	ret = uverbs_get_const(&dv_prt, attrs,
2308 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE);
2309 	if (ret)
2310 		return ret;
2311 
2312 	if (!mlx5_ib_flow_action_packet_reformat_valid(mdev, dv_prt, ft_type))
2313 		return -EOPNOTSUPP;
2314 
2315 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2316 	if (!maction)
2317 		return -ENOMEM;
2318 
2319 	if (dv_prt ==
2320 	    MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2) {
2321 		maction->flow_action_raw.sub_type =
2322 			MLX5_IB_FLOW_ACTION_DECAP;
2323 		maction->flow_action_raw.dev = mdev;
2324 	} else {
2325 		void *in;
2326 		int len;
2327 
2328 		in = uverbs_attr_get_alloced_ptr(attrs,
2329 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2330 		if (IS_ERR(in)) {
2331 			ret = PTR_ERR(in);
2332 			goto free_maction;
2333 		}
2334 
2335 		len = uverbs_attr_get_len(attrs,
2336 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2337 
2338 		ret = mlx5_ib_flow_action_create_packet_reformat_ctx(mdev,
2339 			maction, ft_type, dv_prt, in, len);
2340 		if (ret)
2341 			goto free_maction;
2342 	}
2343 
2344 	uverbs_flow_action_fill_action(&maction->ib_action, uobj, &mdev->ib_dev,
2345 				       IB_FLOW_ACTION_UNSPECIFIED);
2346 	return 0;
2347 
2348 free_maction:
2349 	kfree(maction);
2350 	return ret;
2351 }
2352 
2353 DECLARE_UVERBS_NAMED_METHOD(
2354 	MLX5_IB_METHOD_CREATE_FLOW,
2355 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2356 			UVERBS_OBJECT_FLOW,
2357 			UVERBS_ACCESS_NEW,
2358 			UA_MANDATORY),
2359 	UVERBS_ATTR_PTR_IN(
2360 		MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE,
2361 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2362 		UA_MANDATORY,
2363 		UA_ALLOC_AND_COPY),
2364 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_MATCHER,
2365 			MLX5_IB_OBJECT_FLOW_MATCHER,
2366 			UVERBS_ACCESS_READ,
2367 			UA_MANDATORY),
2368 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_QP,
2369 			UVERBS_OBJECT_QP,
2370 			UVERBS_ACCESS_READ),
2371 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX,
2372 			MLX5_IB_OBJECT_DEVX_OBJ,
2373 			UVERBS_ACCESS_READ),
2374 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS,
2375 			     UVERBS_OBJECT_FLOW_ACTION,
2376 			     UVERBS_ACCESS_READ, 1,
2377 			     MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS,
2378 			     UA_OPTIONAL),
2379 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_TAG,
2380 			   UVERBS_ATTR_TYPE(u32),
2381 			   UA_OPTIONAL),
2382 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX,
2383 			     MLX5_IB_OBJECT_DEVX_OBJ,
2384 			     UVERBS_ACCESS_READ, 1, 1,
2385 			     UA_OPTIONAL),
2386 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
2387 			   UVERBS_ATTR_MIN_SIZE(sizeof(u32)),
2388 			   UA_OPTIONAL,
2389 			   UA_ALLOC_AND_COPY),
2390 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
2391 			     enum mlx5_ib_create_flow_flags,
2392 			     UA_OPTIONAL));
2393 
2394 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2395 	MLX5_IB_METHOD_DESTROY_FLOW,
2396 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2397 			UVERBS_OBJECT_FLOW,
2398 			UVERBS_ACCESS_DESTROY,
2399 			UA_MANDATORY));
2400 
2401 ADD_UVERBS_METHODS(mlx5_ib_fs,
2402 		   UVERBS_OBJECT_FLOW,
2403 		   &UVERBS_METHOD(MLX5_IB_METHOD_CREATE_FLOW),
2404 		   &UVERBS_METHOD(MLX5_IB_METHOD_DESTROY_FLOW));
2405 
2406 DECLARE_UVERBS_NAMED_METHOD(
2407 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER,
2408 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE,
2409 			UVERBS_OBJECT_FLOW_ACTION,
2410 			UVERBS_ACCESS_NEW,
2411 			UA_MANDATORY),
2412 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2413 			   UVERBS_ATTR_MIN_SIZE(MLX5_UN_SZ_BYTES(
2414 				   set_add_copy_action_in_auto)),
2415 			   UA_MANDATORY,
2416 			   UA_ALLOC_AND_COPY),
2417 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE,
2418 			     enum mlx5_ib_uapi_flow_table_type,
2419 			     UA_MANDATORY));
2420 
2421 DECLARE_UVERBS_NAMED_METHOD(
2422 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT,
2423 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE,
2424 			UVERBS_OBJECT_FLOW_ACTION,
2425 			UVERBS_ACCESS_NEW,
2426 			UA_MANDATORY),
2427 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF,
2428 			   UVERBS_ATTR_MIN_SIZE(1),
2429 			   UA_ALLOC_AND_COPY,
2430 			   UA_OPTIONAL),
2431 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE,
2432 			     enum mlx5_ib_uapi_flow_action_packet_reformat_type,
2433 			     UA_MANDATORY),
2434 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE,
2435 			     enum mlx5_ib_uapi_flow_table_type,
2436 			     UA_MANDATORY));
2437 
2438 ADD_UVERBS_METHODS(
2439 	mlx5_ib_flow_actions,
2440 	UVERBS_OBJECT_FLOW_ACTION,
2441 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER),
2442 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT));
2443 
2444 DECLARE_UVERBS_NAMED_METHOD(
2445 	MLX5_IB_METHOD_FLOW_MATCHER_CREATE,
2446 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE,
2447 			MLX5_IB_OBJECT_FLOW_MATCHER,
2448 			UVERBS_ACCESS_NEW,
2449 			UA_MANDATORY),
2450 	UVERBS_ATTR_PTR_IN(
2451 		MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK,
2452 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2453 		UA_MANDATORY),
2454 	UVERBS_ATTR_ENUM_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE,
2455 			    mlx5_ib_flow_type,
2456 			    UA_MANDATORY),
2457 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA,
2458 			   UVERBS_ATTR_TYPE(u8),
2459 			   UA_MANDATORY),
2460 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2461 			     enum ib_flow_flags,
2462 			     UA_OPTIONAL),
2463 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE,
2464 			     enum mlx5_ib_uapi_flow_table_type,
2465 			     UA_OPTIONAL));
2466 
2467 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2468 	MLX5_IB_METHOD_FLOW_MATCHER_DESTROY,
2469 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_DESTROY_HANDLE,
2470 			MLX5_IB_OBJECT_FLOW_MATCHER,
2471 			UVERBS_ACCESS_DESTROY,
2472 			UA_MANDATORY));
2473 
2474 DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_FLOW_MATCHER,
2475 			    UVERBS_TYPE_ALLOC_IDR(flow_matcher_cleanup),
2476 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_CREATE),
2477 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_DESTROY));
2478 
2479 const struct uapi_definition mlx5_ib_flow_defs[] = {
2480 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
2481 		MLX5_IB_OBJECT_FLOW_MATCHER),
2482 	UAPI_DEF_CHAIN_OBJ_TREE(
2483 		UVERBS_OBJECT_FLOW,
2484 		&mlx5_ib_fs),
2485 	UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_FLOW_ACTION,
2486 				&mlx5_ib_flow_actions),
2487 	{},
2488 };
2489 
2490 static const struct ib_device_ops flow_ops = {
2491 	.create_flow = mlx5_ib_create_flow,
2492 	.destroy_flow = mlx5_ib_destroy_flow,
2493 	.destroy_flow_action = mlx5_ib_destroy_flow_action,
2494 };
2495 
2496 static const struct ib_device_ops flow_ipsec_ops = {
2497 	.create_flow_action_esp = mlx5_ib_create_flow_action_esp,
2498 	.modify_flow_action_esp = mlx5_ib_modify_flow_action_esp,
2499 };
2500 
2501 int mlx5_ib_fs_init(struct mlx5_ib_dev *dev)
2502 {
2503 	dev->flow_db = kzalloc(sizeof(*dev->flow_db), GFP_KERNEL);
2504 
2505 	if (!dev->flow_db)
2506 		return -ENOMEM;
2507 
2508 	mutex_init(&dev->flow_db->lock);
2509 
2510 	ib_set_device_ops(&dev->ib_dev, &flow_ops);
2511 	if (mlx5_accel_ipsec_device_caps(dev->mdev) &
2512 	    MLX5_ACCEL_IPSEC_CAP_DEVICE)
2513 		ib_set_device_ops(&dev->ib_dev, &flow_ipsec_ops);
2514 
2515 	return 0;
2516 }
2517