xref: /openbmc/linux/drivers/infiniband/hw/mlx5/fs.c (revision 9dbbc3b9)
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 + sizeof(filter.field), 0,            \
141 		   sizeof(filter) - offsetofend(typeof(filter), field))
142 
143 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
144 			   bool is_egress,
145 			   struct mlx5_flow_act *action)
146 {
147 
148 	switch (maction->ib_action.type) {
149 	case IB_FLOW_ACTION_ESP:
150 		if (action->action & (MLX5_FLOW_CONTEXT_ACTION_ENCRYPT |
151 				      MLX5_FLOW_CONTEXT_ACTION_DECRYPT))
152 			return -EINVAL;
153 		/* Currently only AES_GCM keymat is supported by the driver */
154 		action->esp_id = (uintptr_t)maction->esp_aes_gcm.ctx;
155 		action->action |= is_egress ?
156 			MLX5_FLOW_CONTEXT_ACTION_ENCRYPT :
157 			MLX5_FLOW_CONTEXT_ACTION_DECRYPT;
158 		return 0;
159 	case IB_FLOW_ACTION_UNSPECIFIED:
160 		if (maction->flow_action_raw.sub_type ==
161 		    MLX5_IB_FLOW_ACTION_MODIFY_HEADER) {
162 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
163 				return -EINVAL;
164 			action->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
165 			action->modify_hdr =
166 				maction->flow_action_raw.modify_hdr;
167 			return 0;
168 		}
169 		if (maction->flow_action_raw.sub_type ==
170 		    MLX5_IB_FLOW_ACTION_DECAP) {
171 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_DECAP)
172 				return -EINVAL;
173 			action->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
174 			return 0;
175 		}
176 		if (maction->flow_action_raw.sub_type ==
177 		    MLX5_IB_FLOW_ACTION_PACKET_REFORMAT) {
178 			if (action->action &
179 			    MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT)
180 				return -EINVAL;
181 			action->action |=
182 				MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
183 			action->pkt_reformat =
184 				maction->flow_action_raw.pkt_reformat;
185 			return 0;
186 		}
187 		fallthrough;
188 	default:
189 		return -EOPNOTSUPP;
190 	}
191 }
192 
193 static int parse_flow_attr(struct mlx5_core_dev *mdev,
194 			   struct mlx5_flow_spec *spec,
195 			   const union ib_flow_spec *ib_spec,
196 			   const struct ib_flow_attr *flow_attr,
197 			   struct mlx5_flow_act *action, u32 prev_type)
198 {
199 	struct mlx5_flow_context *flow_context = &spec->flow_context;
200 	u32 *match_c = spec->match_criteria;
201 	u32 *match_v = spec->match_value;
202 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param, match_c,
203 					   misc_parameters);
204 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, match_v,
205 					   misc_parameters);
206 	void *misc_params2_c = MLX5_ADDR_OF(fte_match_param, match_c,
207 					    misc_parameters_2);
208 	void *misc_params2_v = MLX5_ADDR_OF(fte_match_param, match_v,
209 					    misc_parameters_2);
210 	void *headers_c;
211 	void *headers_v;
212 	int match_ipv;
213 	int ret;
214 
215 	if (ib_spec->type & IB_FLOW_SPEC_INNER) {
216 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
217 					 inner_headers);
218 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
219 					 inner_headers);
220 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
221 					ft_field_support.inner_ip_version);
222 	} else {
223 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
224 					 outer_headers);
225 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
226 					 outer_headers);
227 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
228 					ft_field_support.outer_ip_version);
229 	}
230 
231 	switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
232 	case IB_FLOW_SPEC_ETH:
233 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
234 			return -EOPNOTSUPP;
235 
236 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
237 					     dmac_47_16),
238 				ib_spec->eth.mask.dst_mac);
239 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
240 					     dmac_47_16),
241 				ib_spec->eth.val.dst_mac);
242 
243 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
244 					     smac_47_16),
245 				ib_spec->eth.mask.src_mac);
246 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
247 					     smac_47_16),
248 				ib_spec->eth.val.src_mac);
249 
250 		if (ib_spec->eth.mask.vlan_tag) {
251 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
252 				 cvlan_tag, 1);
253 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
254 				 cvlan_tag, 1);
255 
256 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
257 				 first_vid, ntohs(ib_spec->eth.mask.vlan_tag));
258 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
259 				 first_vid, ntohs(ib_spec->eth.val.vlan_tag));
260 
261 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
262 				 first_cfi,
263 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 12);
264 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
265 				 first_cfi,
266 				 ntohs(ib_spec->eth.val.vlan_tag) >> 12);
267 
268 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
269 				 first_prio,
270 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 13);
271 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
272 				 first_prio,
273 				 ntohs(ib_spec->eth.val.vlan_tag) >> 13);
274 		}
275 		MLX5_SET(fte_match_set_lyr_2_4, headers_c,
276 			 ethertype, ntohs(ib_spec->eth.mask.ether_type));
277 		MLX5_SET(fte_match_set_lyr_2_4, headers_v,
278 			 ethertype, ntohs(ib_spec->eth.val.ether_type));
279 		break;
280 	case IB_FLOW_SPEC_IPV4:
281 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
282 			return -EOPNOTSUPP;
283 
284 		if (match_ipv) {
285 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
286 				 ip_version, 0xf);
287 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
288 				 ip_version, MLX5_FS_IPV4_VERSION);
289 		} else {
290 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
291 				 ethertype, 0xffff);
292 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
293 				 ethertype, ETH_P_IP);
294 		}
295 
296 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
297 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
298 		       &ib_spec->ipv4.mask.src_ip,
299 		       sizeof(ib_spec->ipv4.mask.src_ip));
300 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
301 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
302 		       &ib_spec->ipv4.val.src_ip,
303 		       sizeof(ib_spec->ipv4.val.src_ip));
304 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
305 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
306 		       &ib_spec->ipv4.mask.dst_ip,
307 		       sizeof(ib_spec->ipv4.mask.dst_ip));
308 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
309 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
310 		       &ib_spec->ipv4.val.dst_ip,
311 		       sizeof(ib_spec->ipv4.val.dst_ip));
312 
313 		set_tos(headers_c, headers_v,
314 			ib_spec->ipv4.mask.tos, ib_spec->ipv4.val.tos);
315 
316 		if (set_proto(headers_c, headers_v,
317 			      ib_spec->ipv4.mask.proto,
318 			      ib_spec->ipv4.val.proto))
319 			return -EINVAL;
320 		break;
321 	case IB_FLOW_SPEC_IPV6:
322 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv6.mask, LAST_IPV6_FIELD))
323 			return -EOPNOTSUPP;
324 
325 		if (match_ipv) {
326 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
327 				 ip_version, 0xf);
328 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
329 				 ip_version, MLX5_FS_IPV6_VERSION);
330 		} else {
331 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
332 				 ethertype, 0xffff);
333 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
334 				 ethertype, ETH_P_IPV6);
335 		}
336 
337 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
338 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
339 		       &ib_spec->ipv6.mask.src_ip,
340 		       sizeof(ib_spec->ipv6.mask.src_ip));
341 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
342 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
343 		       &ib_spec->ipv6.val.src_ip,
344 		       sizeof(ib_spec->ipv6.val.src_ip));
345 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
346 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
347 		       &ib_spec->ipv6.mask.dst_ip,
348 		       sizeof(ib_spec->ipv6.mask.dst_ip));
349 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
350 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
351 		       &ib_spec->ipv6.val.dst_ip,
352 		       sizeof(ib_spec->ipv6.val.dst_ip));
353 
354 		set_tos(headers_c, headers_v,
355 			ib_spec->ipv6.mask.traffic_class,
356 			ib_spec->ipv6.val.traffic_class);
357 
358 		if (set_proto(headers_c, headers_v,
359 			      ib_spec->ipv6.mask.next_hdr,
360 			      ib_spec->ipv6.val.next_hdr))
361 			return -EINVAL;
362 
363 		set_flow_label(misc_params_c, misc_params_v,
364 			       ntohl(ib_spec->ipv6.mask.flow_label),
365 			       ntohl(ib_spec->ipv6.val.flow_label),
366 			       ib_spec->type & IB_FLOW_SPEC_INNER);
367 		break;
368 	case IB_FLOW_SPEC_ESP:
369 		if (ib_spec->esp.mask.seq)
370 			return -EOPNOTSUPP;
371 
372 		MLX5_SET(fte_match_set_misc, misc_params_c, outer_esp_spi,
373 			 ntohl(ib_spec->esp.mask.spi));
374 		MLX5_SET(fte_match_set_misc, misc_params_v, outer_esp_spi,
375 			 ntohl(ib_spec->esp.val.spi));
376 		break;
377 	case IB_FLOW_SPEC_TCP:
378 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
379 					 LAST_TCP_UDP_FIELD))
380 			return -EOPNOTSUPP;
381 
382 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_TCP))
383 			return -EINVAL;
384 
385 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_sport,
386 			 ntohs(ib_spec->tcp_udp.mask.src_port));
387 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
388 			 ntohs(ib_spec->tcp_udp.val.src_port));
389 
390 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_dport,
391 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
392 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
393 			 ntohs(ib_spec->tcp_udp.val.dst_port));
394 		break;
395 	case IB_FLOW_SPEC_UDP:
396 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
397 					 LAST_TCP_UDP_FIELD))
398 			return -EOPNOTSUPP;
399 
400 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_UDP))
401 			return -EINVAL;
402 
403 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_sport,
404 			 ntohs(ib_spec->tcp_udp.mask.src_port));
405 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
406 			 ntohs(ib_spec->tcp_udp.val.src_port));
407 
408 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_dport,
409 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
410 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
411 			 ntohs(ib_spec->tcp_udp.val.dst_port));
412 		break;
413 	case IB_FLOW_SPEC_GRE:
414 		if (ib_spec->gre.mask.c_ks_res0_ver)
415 			return -EOPNOTSUPP;
416 
417 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_GRE))
418 			return -EINVAL;
419 
420 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
421 			 0xff);
422 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
423 			 IPPROTO_GRE);
424 
425 		MLX5_SET(fte_match_set_misc, misc_params_c, gre_protocol,
426 			 ntohs(ib_spec->gre.mask.protocol));
427 		MLX5_SET(fte_match_set_misc, misc_params_v, gre_protocol,
428 			 ntohs(ib_spec->gre.val.protocol));
429 
430 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_c,
431 				    gre_key.nvgre.hi),
432 		       &ib_spec->gre.mask.key,
433 		       sizeof(ib_spec->gre.mask.key));
434 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_v,
435 				    gre_key.nvgre.hi),
436 		       &ib_spec->gre.val.key,
437 		       sizeof(ib_spec->gre.val.key));
438 		break;
439 	case IB_FLOW_SPEC_MPLS:
440 		switch (prev_type) {
441 		case IB_FLOW_SPEC_UDP:
442 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
443 						   ft_field_support.outer_first_mpls_over_udp),
444 						   &ib_spec->mpls.mask.tag))
445 				return -EOPNOTSUPP;
446 
447 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
448 					    outer_first_mpls_over_udp),
449 			       &ib_spec->mpls.val.tag,
450 			       sizeof(ib_spec->mpls.val.tag));
451 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
452 					    outer_first_mpls_over_udp),
453 			       &ib_spec->mpls.mask.tag,
454 			       sizeof(ib_spec->mpls.mask.tag));
455 			break;
456 		case IB_FLOW_SPEC_GRE:
457 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
458 						   ft_field_support.outer_first_mpls_over_gre),
459 						   &ib_spec->mpls.mask.tag))
460 				return -EOPNOTSUPP;
461 
462 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
463 					    outer_first_mpls_over_gre),
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 					    outer_first_mpls_over_gre),
468 			       &ib_spec->mpls.mask.tag,
469 			       sizeof(ib_spec->mpls.mask.tag));
470 			break;
471 		default:
472 			if (ib_spec->type & IB_FLOW_SPEC_INNER) {
473 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
474 							   ft_field_support.inner_first_mpls),
475 							   &ib_spec->mpls.mask.tag))
476 					return -EOPNOTSUPP;
477 
478 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
479 						    inner_first_mpls),
480 				       &ib_spec->mpls.val.tag,
481 				       sizeof(ib_spec->mpls.val.tag));
482 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
483 						    inner_first_mpls),
484 				       &ib_spec->mpls.mask.tag,
485 				       sizeof(ib_spec->mpls.mask.tag));
486 			} else {
487 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
488 							   ft_field_support.outer_first_mpls),
489 							   &ib_spec->mpls.mask.tag))
490 					return -EOPNOTSUPP;
491 
492 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
493 						    outer_first_mpls),
494 				       &ib_spec->mpls.val.tag,
495 				       sizeof(ib_spec->mpls.val.tag));
496 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
497 						    outer_first_mpls),
498 				       &ib_spec->mpls.mask.tag,
499 				       sizeof(ib_spec->mpls.mask.tag));
500 			}
501 		}
502 		break;
503 	case IB_FLOW_SPEC_VXLAN_TUNNEL:
504 		if (FIELDS_NOT_SUPPORTED(ib_spec->tunnel.mask,
505 					 LAST_TUNNEL_FIELD))
506 			return -EOPNOTSUPP;
507 
508 		MLX5_SET(fte_match_set_misc, misc_params_c, vxlan_vni,
509 			 ntohl(ib_spec->tunnel.mask.tunnel_id));
510 		MLX5_SET(fte_match_set_misc, misc_params_v, vxlan_vni,
511 			 ntohl(ib_spec->tunnel.val.tunnel_id));
512 		break;
513 	case IB_FLOW_SPEC_ACTION_TAG:
514 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_tag,
515 					 LAST_FLOW_TAG_FIELD))
516 			return -EOPNOTSUPP;
517 		if (ib_spec->flow_tag.tag_id >= BIT(24))
518 			return -EINVAL;
519 
520 		flow_context->flow_tag = ib_spec->flow_tag.tag_id;
521 		flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
522 		break;
523 	case IB_FLOW_SPEC_ACTION_DROP:
524 		if (FIELDS_NOT_SUPPORTED(ib_spec->drop,
525 					 LAST_DROP_FIELD))
526 			return -EOPNOTSUPP;
527 		action->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
528 		break;
529 	case IB_FLOW_SPEC_ACTION_HANDLE:
530 		ret = parse_flow_flow_action(to_mflow_act(ib_spec->action.act),
531 			flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS, action);
532 		if (ret)
533 			return ret;
534 		break;
535 	case IB_FLOW_SPEC_ACTION_COUNT:
536 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_count,
537 					 LAST_COUNTERS_FIELD))
538 			return -EOPNOTSUPP;
539 
540 		/* for now support only one counters spec per flow */
541 		if (action->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
542 			return -EINVAL;
543 
544 		action->counters = ib_spec->flow_count.counters;
545 		action->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
546 		break;
547 	default:
548 		return -EINVAL;
549 	}
550 
551 	return 0;
552 }
553 
554 /* If a flow could catch both multicast and unicast packets,
555  * it won't fall into the multicast flow steering table and this rule
556  * could steal other multicast packets.
557  */
558 static bool flow_is_multicast_only(const struct ib_flow_attr *ib_attr)
559 {
560 	union ib_flow_spec *flow_spec;
561 
562 	if (ib_attr->type != IB_FLOW_ATTR_NORMAL ||
563 	    ib_attr->num_of_specs < 1)
564 		return false;
565 
566 	flow_spec = (union ib_flow_spec *)(ib_attr + 1);
567 	if (flow_spec->type == IB_FLOW_SPEC_IPV4) {
568 		struct ib_flow_spec_ipv4 *ipv4_spec;
569 
570 		ipv4_spec = (struct ib_flow_spec_ipv4 *)flow_spec;
571 		if (ipv4_is_multicast(ipv4_spec->val.dst_ip))
572 			return true;
573 
574 		return false;
575 	}
576 
577 	if (flow_spec->type == IB_FLOW_SPEC_ETH) {
578 		struct ib_flow_spec_eth *eth_spec;
579 
580 		eth_spec = (struct ib_flow_spec_eth *)flow_spec;
581 		return is_multicast_ether_addr(eth_spec->mask.dst_mac) &&
582 		       is_multicast_ether_addr(eth_spec->val.dst_mac);
583 	}
584 
585 	return false;
586 }
587 
588 enum valid_spec {
589 	VALID_SPEC_INVALID,
590 	VALID_SPEC_VALID,
591 	VALID_SPEC_NA,
592 };
593 
594 static enum valid_spec
595 is_valid_esp_aes_gcm(struct mlx5_core_dev *mdev,
596 		     const struct mlx5_flow_spec *spec,
597 		     const struct mlx5_flow_act *flow_act,
598 		     bool egress)
599 {
600 	const u32 *match_c = spec->match_criteria;
601 	bool is_crypto =
602 		(flow_act->action & (MLX5_FLOW_CONTEXT_ACTION_ENCRYPT |
603 				     MLX5_FLOW_CONTEXT_ACTION_DECRYPT));
604 	bool is_ipsec = mlx5_fs_is_ipsec_flow(match_c);
605 	bool is_drop = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_DROP;
606 
607 	/*
608 	 * Currently only crypto is supported in egress, when regular egress
609 	 * rules would be supported, always return VALID_SPEC_NA.
610 	 */
611 	if (!is_crypto)
612 		return VALID_SPEC_NA;
613 
614 	return is_crypto && is_ipsec &&
615 		(!egress || (!is_drop &&
616 			     !(spec->flow_context.flags & FLOW_CONTEXT_HAS_TAG))) ?
617 		VALID_SPEC_VALID : VALID_SPEC_INVALID;
618 }
619 
620 static bool is_valid_spec(struct mlx5_core_dev *mdev,
621 			  const struct mlx5_flow_spec *spec,
622 			  const struct mlx5_flow_act *flow_act,
623 			  bool egress)
624 {
625 	/* We curretly only support ipsec egress flow */
626 	return is_valid_esp_aes_gcm(mdev, spec, flow_act, egress) != VALID_SPEC_INVALID;
627 }
628 
629 static bool is_valid_ethertype(struct mlx5_core_dev *mdev,
630 			       const struct ib_flow_attr *flow_attr,
631 			       bool check_inner)
632 {
633 	union ib_flow_spec *ib_spec = (union ib_flow_spec *)(flow_attr + 1);
634 	int match_ipv = check_inner ?
635 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
636 					ft_field_support.inner_ip_version) :
637 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
638 					ft_field_support.outer_ip_version);
639 	int inner_bit = check_inner ? IB_FLOW_SPEC_INNER : 0;
640 	bool ipv4_spec_valid, ipv6_spec_valid;
641 	unsigned int ip_spec_type = 0;
642 	bool has_ethertype = false;
643 	unsigned int spec_index;
644 	bool mask_valid = true;
645 	u16 eth_type = 0;
646 	bool type_valid;
647 
648 	/* Validate that ethertype is correct */
649 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
650 		if ((ib_spec->type == (IB_FLOW_SPEC_ETH | inner_bit)) &&
651 		    ib_spec->eth.mask.ether_type) {
652 			mask_valid = (ib_spec->eth.mask.ether_type ==
653 				      htons(0xffff));
654 			has_ethertype = true;
655 			eth_type = ntohs(ib_spec->eth.val.ether_type);
656 		} else if ((ib_spec->type == (IB_FLOW_SPEC_IPV4 | inner_bit)) ||
657 			   (ib_spec->type == (IB_FLOW_SPEC_IPV6 | inner_bit))) {
658 			ip_spec_type = ib_spec->type;
659 		}
660 		ib_spec = (void *)ib_spec + ib_spec->size;
661 	}
662 
663 	type_valid = (!has_ethertype) || (!ip_spec_type);
664 	if (!type_valid && mask_valid) {
665 		ipv4_spec_valid = (eth_type == ETH_P_IP) &&
666 			(ip_spec_type == (IB_FLOW_SPEC_IPV4 | inner_bit));
667 		ipv6_spec_valid = (eth_type == ETH_P_IPV6) &&
668 			(ip_spec_type == (IB_FLOW_SPEC_IPV6 | inner_bit));
669 
670 		type_valid = (ipv4_spec_valid) || (ipv6_spec_valid) ||
671 			     (((eth_type == ETH_P_MPLS_UC) ||
672 			       (eth_type == ETH_P_MPLS_MC)) && match_ipv);
673 	}
674 
675 	return type_valid;
676 }
677 
678 static bool is_valid_attr(struct mlx5_core_dev *mdev,
679 			  const struct ib_flow_attr *flow_attr)
680 {
681 	return is_valid_ethertype(mdev, flow_attr, false) &&
682 	       is_valid_ethertype(mdev, flow_attr, true);
683 }
684 
685 static void put_flow_table(struct mlx5_ib_dev *dev,
686 			   struct mlx5_ib_flow_prio *prio, bool ft_added)
687 {
688 	prio->refcount -= !!ft_added;
689 	if (!prio->refcount) {
690 		mlx5_destroy_flow_table(prio->flow_table);
691 		prio->flow_table = NULL;
692 	}
693 }
694 
695 static int mlx5_ib_destroy_flow(struct ib_flow *flow_id)
696 {
697 	struct mlx5_ib_flow_handler *handler = container_of(flow_id,
698 							  struct mlx5_ib_flow_handler,
699 							  ibflow);
700 	struct mlx5_ib_flow_handler *iter, *tmp;
701 	struct mlx5_ib_dev *dev = handler->dev;
702 
703 	mutex_lock(&dev->flow_db->lock);
704 
705 	list_for_each_entry_safe(iter, tmp, &handler->list, list) {
706 		mlx5_del_flow_rules(iter->rule);
707 		put_flow_table(dev, iter->prio, true);
708 		list_del(&iter->list);
709 		kfree(iter);
710 	}
711 
712 	mlx5_del_flow_rules(handler->rule);
713 	put_flow_table(dev, handler->prio, true);
714 	mlx5_ib_counters_clear_description(handler->ibcounters);
715 	mutex_unlock(&dev->flow_db->lock);
716 	if (handler->flow_matcher)
717 		atomic_dec(&handler->flow_matcher->usecnt);
718 	kfree(handler);
719 
720 	return 0;
721 }
722 
723 static int ib_prio_to_core_prio(unsigned int priority, bool dont_trap)
724 {
725 	priority *= 2;
726 	if (!dont_trap)
727 		priority++;
728 	return priority;
729 }
730 
731 enum flow_table_type {
732 	MLX5_IB_FT_RX,
733 	MLX5_IB_FT_TX
734 };
735 
736 #define MLX5_FS_MAX_TYPES	 6
737 #define MLX5_FS_MAX_ENTRIES	 BIT(16)
738 
739 static struct mlx5_ib_flow_prio *_get_prio(struct mlx5_flow_namespace *ns,
740 					   struct mlx5_ib_flow_prio *prio,
741 					   int priority,
742 					   int num_entries, int num_groups,
743 					   u32 flags)
744 {
745 	struct mlx5_flow_table_attr ft_attr = {};
746 	struct mlx5_flow_table *ft;
747 
748 	ft_attr.prio = priority;
749 	ft_attr.max_fte = num_entries;
750 	ft_attr.flags = flags;
751 	ft_attr.autogroup.max_num_groups = num_groups;
752 	ft = mlx5_create_auto_grouped_flow_table(ns, &ft_attr);
753 	if (IS_ERR(ft))
754 		return ERR_CAST(ft);
755 
756 	prio->flow_table = ft;
757 	prio->refcount = 0;
758 	return prio;
759 }
760 
761 static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev,
762 						struct ib_flow_attr *flow_attr,
763 						enum flow_table_type ft_type)
764 {
765 	bool dont_trap = flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP;
766 	struct mlx5_flow_namespace *ns = NULL;
767 	enum mlx5_flow_namespace_type fn_type;
768 	struct mlx5_ib_flow_prio *prio;
769 	struct mlx5_flow_table *ft;
770 	int max_table_size;
771 	int num_entries;
772 	int num_groups;
773 	bool esw_encap;
774 	u32 flags = 0;
775 	int priority;
776 
777 	max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
778 						       log_max_ft_size));
779 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
780 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
781 	switch (flow_attr->type) {
782 	case IB_FLOW_ATTR_NORMAL:
783 		if (flow_is_multicast_only(flow_attr) && !dont_trap)
784 			priority = MLX5_IB_FLOW_MCAST_PRIO;
785 		else
786 			priority = ib_prio_to_core_prio(flow_attr->priority,
787 							dont_trap);
788 		if (ft_type == MLX5_IB_FT_RX) {
789 			fn_type = MLX5_FLOW_NAMESPACE_BYPASS;
790 			prio = &dev->flow_db->prios[priority];
791 			if (!dev->is_rep && !esw_encap &&
792 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap))
793 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
794 			if (!dev->is_rep && !esw_encap &&
795 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
796 						      reformat_l3_tunnel_to_l2))
797 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
798 		} else {
799 			max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_TX(
800 				dev->mdev, log_max_ft_size));
801 			fn_type = MLX5_FLOW_NAMESPACE_EGRESS;
802 			prio = &dev->flow_db->egress_prios[priority];
803 			if (!dev->is_rep && !esw_encap &&
804 			    MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat))
805 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
806 		}
807 		ns = mlx5_get_flow_namespace(dev->mdev, fn_type);
808 		num_entries = MLX5_FS_MAX_ENTRIES;
809 		num_groups = MLX5_FS_MAX_TYPES;
810 		break;
811 	case IB_FLOW_ATTR_ALL_DEFAULT:
812 	case IB_FLOW_ATTR_MC_DEFAULT:
813 		ns = mlx5_get_flow_namespace(dev->mdev,
814 					     MLX5_FLOW_NAMESPACE_LEFTOVERS);
815 		build_leftovers_ft_param(&priority, &num_entries, &num_groups);
816 		prio = &dev->flow_db->prios[MLX5_IB_FLOW_LEFTOVERS_PRIO];
817 		break;
818 	case IB_FLOW_ATTR_SNIFFER:
819 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
820 					allow_sniffer_and_nic_rx_shared_tir))
821 			return ERR_PTR(-EOPNOTSUPP);
822 
823 		ns = mlx5_get_flow_namespace(
824 			dev->mdev, ft_type == MLX5_IB_FT_RX ?
825 					   MLX5_FLOW_NAMESPACE_SNIFFER_RX :
826 					   MLX5_FLOW_NAMESPACE_SNIFFER_TX);
827 
828 		prio = &dev->flow_db->sniffer[ft_type];
829 		priority = 0;
830 		num_entries = 1;
831 		num_groups = 1;
832 		break;
833 	default:
834 		break;
835 	}
836 
837 	if (!ns)
838 		return ERR_PTR(-EOPNOTSUPP);
839 
840 	max_table_size = min_t(int, num_entries, max_table_size);
841 
842 	ft = prio->flow_table;
843 	if (!ft)
844 		return _get_prio(ns, prio, priority, max_table_size, num_groups,
845 				 flags);
846 
847 	return prio;
848 }
849 
850 static void set_underlay_qp(struct mlx5_ib_dev *dev,
851 			    struct mlx5_flow_spec *spec,
852 			    u32 underlay_qpn)
853 {
854 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param,
855 					   spec->match_criteria,
856 					   misc_parameters);
857 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
858 					   misc_parameters);
859 
860 	if (underlay_qpn &&
861 	    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
862 				      ft_field_support.bth_dst_qp)) {
863 		MLX5_SET(fte_match_set_misc,
864 			 misc_params_v, bth_dst_qp, underlay_qpn);
865 		MLX5_SET(fte_match_set_misc,
866 			 misc_params_c, bth_dst_qp, 0xffffff);
867 	}
868 }
869 
870 static void mlx5_ib_set_rule_source_port(struct mlx5_ib_dev *dev,
871 					 struct mlx5_flow_spec *spec,
872 					 struct mlx5_eswitch_rep *rep)
873 {
874 	struct mlx5_eswitch *esw = dev->mdev->priv.eswitch;
875 	void *misc;
876 
877 	if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
878 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
879 				    misc_parameters_2);
880 
881 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
882 			 mlx5_eswitch_get_vport_metadata_for_match(rep->esw,
883 								   rep->vport));
884 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
885 				    misc_parameters_2);
886 
887 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
888 			 mlx5_eswitch_get_vport_metadata_mask());
889 	} else {
890 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
891 				    misc_parameters);
892 
893 		MLX5_SET(fte_match_set_misc, misc, source_port, rep->vport);
894 
895 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
896 				    misc_parameters);
897 
898 		MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
899 	}
900 }
901 
902 static struct mlx5_ib_flow_handler *_create_flow_rule(struct mlx5_ib_dev *dev,
903 						      struct mlx5_ib_flow_prio *ft_prio,
904 						      const struct ib_flow_attr *flow_attr,
905 						      struct mlx5_flow_destination *dst,
906 						      u32 underlay_qpn,
907 						      struct mlx5_ib_create_flow *ucmd)
908 {
909 	struct mlx5_flow_table	*ft = ft_prio->flow_table;
910 	struct mlx5_ib_flow_handler *handler;
911 	struct mlx5_flow_act flow_act = {};
912 	struct mlx5_flow_spec *spec;
913 	struct mlx5_flow_destination dest_arr[2] = {};
914 	struct mlx5_flow_destination *rule_dst = dest_arr;
915 	const void *ib_flow = (const void *)flow_attr + sizeof(*flow_attr);
916 	unsigned int spec_index;
917 	u32 prev_type = 0;
918 	int err = 0;
919 	int dest_num = 0;
920 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
921 
922 	if (!is_valid_attr(dev->mdev, flow_attr))
923 		return ERR_PTR(-EINVAL);
924 
925 	if (dev->is_rep && is_egress)
926 		return ERR_PTR(-EINVAL);
927 
928 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
929 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
930 	if (!handler || !spec) {
931 		err = -ENOMEM;
932 		goto free;
933 	}
934 
935 	INIT_LIST_HEAD(&handler->list);
936 
937 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
938 		err = parse_flow_attr(dev->mdev, spec,
939 				      ib_flow, flow_attr, &flow_act,
940 				      prev_type);
941 		if (err < 0)
942 			goto free;
943 
944 		prev_type = ((union ib_flow_spec *)ib_flow)->type;
945 		ib_flow += ((union ib_flow_spec *)ib_flow)->size;
946 	}
947 
948 	if (dst && !(flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP)) {
949 		memcpy(&dest_arr[0], dst, sizeof(*dst));
950 		dest_num++;
951 	}
952 
953 	if (!flow_is_multicast_only(flow_attr))
954 		set_underlay_qp(dev, spec, underlay_qpn);
955 
956 	if (dev->is_rep && flow_attr->type != IB_FLOW_ATTR_SNIFFER) {
957 		struct mlx5_eswitch_rep *rep;
958 
959 		rep = dev->port[flow_attr->port - 1].rep;
960 		if (!rep) {
961 			err = -EINVAL;
962 			goto free;
963 		}
964 
965 		mlx5_ib_set_rule_source_port(dev, spec, rep);
966 	}
967 
968 	spec->match_criteria_enable = get_match_criteria_enable(spec->match_criteria);
969 
970 	if (is_egress &&
971 	    !is_valid_spec(dev->mdev, spec, &flow_act, is_egress)) {
972 		err = -EINVAL;
973 		goto free;
974 	}
975 
976 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
977 		struct mlx5_ib_mcounters *mcounters;
978 
979 		err = mlx5_ib_flow_counters_set_data(flow_act.counters, ucmd);
980 		if (err)
981 			goto free;
982 
983 		mcounters = to_mcounters(flow_act.counters);
984 		handler->ibcounters = flow_act.counters;
985 		dest_arr[dest_num].type =
986 			MLX5_FLOW_DESTINATION_TYPE_COUNTER;
987 		dest_arr[dest_num].counter_id =
988 			mlx5_fc_id(mcounters->hw_cntrs_hndl);
989 		dest_num++;
990 	}
991 
992 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP) {
993 		if (!dest_num)
994 			rule_dst = NULL;
995 	} else {
996 		if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)
997 			flow_act.action |=
998 				MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
999 		if (is_egress)
1000 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1001 		else if (dest_num)
1002 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1003 	}
1004 
1005 	if ((spec->flow_context.flags & FLOW_CONTEXT_HAS_TAG)  &&
1006 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1007 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1008 		mlx5_ib_warn(dev, "Flow tag %u and attribute type %x isn't allowed in leftovers\n",
1009 			     spec->flow_context.flow_tag, flow_attr->type);
1010 		err = -EINVAL;
1011 		goto free;
1012 	}
1013 	handler->rule = mlx5_add_flow_rules(ft, spec,
1014 					    &flow_act,
1015 					    rule_dst, dest_num);
1016 
1017 	if (IS_ERR(handler->rule)) {
1018 		err = PTR_ERR(handler->rule);
1019 		goto free;
1020 	}
1021 
1022 	ft_prio->refcount++;
1023 	handler->prio = ft_prio;
1024 	handler->dev = dev;
1025 
1026 	ft_prio->flow_table = ft;
1027 free:
1028 	if (err && handler) {
1029 		mlx5_ib_counters_clear_description(handler->ibcounters);
1030 		kfree(handler);
1031 	}
1032 	kvfree(spec);
1033 	return err ? ERR_PTR(err) : handler;
1034 }
1035 
1036 static struct mlx5_ib_flow_handler *create_flow_rule(struct mlx5_ib_dev *dev,
1037 						     struct mlx5_ib_flow_prio *ft_prio,
1038 						     const struct ib_flow_attr *flow_attr,
1039 						     struct mlx5_flow_destination *dst)
1040 {
1041 	return _create_flow_rule(dev, ft_prio, flow_attr, dst, 0, NULL);
1042 }
1043 
1044 enum {
1045 	LEFTOVERS_MC,
1046 	LEFTOVERS_UC,
1047 };
1048 
1049 static struct mlx5_ib_flow_handler *create_leftovers_rule(struct mlx5_ib_dev *dev,
1050 							  struct mlx5_ib_flow_prio *ft_prio,
1051 							  struct ib_flow_attr *flow_attr,
1052 							  struct mlx5_flow_destination *dst)
1053 {
1054 	struct mlx5_ib_flow_handler *handler_ucast = NULL;
1055 	struct mlx5_ib_flow_handler *handler = NULL;
1056 
1057 	static struct {
1058 		struct ib_flow_attr	flow_attr;
1059 		struct ib_flow_spec_eth eth_flow;
1060 	} leftovers_specs[] = {
1061 		[LEFTOVERS_MC] = {
1062 			.flow_attr = {
1063 				.num_of_specs = 1,
1064 				.size = sizeof(leftovers_specs[0])
1065 			},
1066 			.eth_flow = {
1067 				.type = IB_FLOW_SPEC_ETH,
1068 				.size = sizeof(struct ib_flow_spec_eth),
1069 				.mask = {.dst_mac = {0x1} },
1070 				.val =  {.dst_mac = {0x1} }
1071 			}
1072 		},
1073 		[LEFTOVERS_UC] = {
1074 			.flow_attr = {
1075 				.num_of_specs = 1,
1076 				.size = sizeof(leftovers_specs[0])
1077 			},
1078 			.eth_flow = {
1079 				.type = IB_FLOW_SPEC_ETH,
1080 				.size = sizeof(struct ib_flow_spec_eth),
1081 				.mask = {.dst_mac = {0x1} },
1082 				.val = {.dst_mac = {} }
1083 			}
1084 		}
1085 	};
1086 
1087 	handler = create_flow_rule(dev, ft_prio,
1088 				   &leftovers_specs[LEFTOVERS_MC].flow_attr,
1089 				   dst);
1090 	if (!IS_ERR(handler) &&
1091 	    flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT) {
1092 		handler_ucast = create_flow_rule(dev, ft_prio,
1093 						 &leftovers_specs[LEFTOVERS_UC].flow_attr,
1094 						 dst);
1095 		if (IS_ERR(handler_ucast)) {
1096 			mlx5_del_flow_rules(handler->rule);
1097 			ft_prio->refcount--;
1098 			kfree(handler);
1099 			handler = handler_ucast;
1100 		} else {
1101 			list_add(&handler_ucast->list, &handler->list);
1102 		}
1103 	}
1104 
1105 	return handler;
1106 }
1107 
1108 static struct mlx5_ib_flow_handler *create_sniffer_rule(struct mlx5_ib_dev *dev,
1109 							struct mlx5_ib_flow_prio *ft_rx,
1110 							struct mlx5_ib_flow_prio *ft_tx,
1111 							struct mlx5_flow_destination *dst)
1112 {
1113 	struct mlx5_ib_flow_handler *handler_rx;
1114 	struct mlx5_ib_flow_handler *handler_tx;
1115 	int err;
1116 	static const struct ib_flow_attr flow_attr  = {
1117 		.num_of_specs = 0,
1118 		.type = IB_FLOW_ATTR_SNIFFER,
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 static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
1147 					   struct ib_flow_attr *flow_attr,
1148 					   struct ib_udata *udata)
1149 {
1150 	struct mlx5_ib_dev *dev = to_mdev(qp->device);
1151 	struct mlx5_ib_qp *mqp = to_mqp(qp);
1152 	struct mlx5_ib_flow_handler *handler = NULL;
1153 	struct mlx5_flow_destination *dst = NULL;
1154 	struct mlx5_ib_flow_prio *ft_prio_tx = NULL;
1155 	struct mlx5_ib_flow_prio *ft_prio;
1156 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
1157 	struct mlx5_ib_create_flow *ucmd = NULL, ucmd_hdr;
1158 	size_t min_ucmd_sz, required_ucmd_sz;
1159 	int err;
1160 	int underlay_qpn;
1161 
1162 	if (udata && udata->inlen) {
1163 		min_ucmd_sz = offsetofend(struct mlx5_ib_create_flow, reserved);
1164 		if (udata->inlen < min_ucmd_sz)
1165 			return ERR_PTR(-EOPNOTSUPP);
1166 
1167 		err = ib_copy_from_udata(&ucmd_hdr, udata, min_ucmd_sz);
1168 		if (err)
1169 			return ERR_PTR(err);
1170 
1171 		/* currently supports only one counters data */
1172 		if (ucmd_hdr.ncounters_data > 1)
1173 			return ERR_PTR(-EINVAL);
1174 
1175 		required_ucmd_sz = min_ucmd_sz +
1176 			sizeof(struct mlx5_ib_flow_counters_data) *
1177 			ucmd_hdr.ncounters_data;
1178 		if (udata->inlen > required_ucmd_sz &&
1179 		    !ib_is_udata_cleared(udata, required_ucmd_sz,
1180 					 udata->inlen - required_ucmd_sz))
1181 			return ERR_PTR(-EOPNOTSUPP);
1182 
1183 		ucmd = kzalloc(required_ucmd_sz, GFP_KERNEL);
1184 		if (!ucmd)
1185 			return ERR_PTR(-ENOMEM);
1186 
1187 		err = ib_copy_from_udata(ucmd, udata, required_ucmd_sz);
1188 		if (err)
1189 			goto free_ucmd;
1190 	}
1191 
1192 	if (flow_attr->priority > MLX5_IB_FLOW_LAST_PRIO) {
1193 		err = -ENOMEM;
1194 		goto free_ucmd;
1195 	}
1196 
1197 	if (flow_attr->flags &
1198 	    ~(IB_FLOW_ATTR_FLAGS_DONT_TRAP | IB_FLOW_ATTR_FLAGS_EGRESS)) {
1199 		err = -EINVAL;
1200 		goto free_ucmd;
1201 	}
1202 
1203 	if (is_egress &&
1204 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1205 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1206 		err = -EINVAL;
1207 		goto free_ucmd;
1208 	}
1209 
1210 	dst = kzalloc(sizeof(*dst), GFP_KERNEL);
1211 	if (!dst) {
1212 		err = -ENOMEM;
1213 		goto free_ucmd;
1214 	}
1215 
1216 	mutex_lock(&dev->flow_db->lock);
1217 
1218 	ft_prio = get_flow_table(dev, flow_attr,
1219 				 is_egress ? MLX5_IB_FT_TX : MLX5_IB_FT_RX);
1220 	if (IS_ERR(ft_prio)) {
1221 		err = PTR_ERR(ft_prio);
1222 		goto unlock;
1223 	}
1224 	if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
1225 		ft_prio_tx = get_flow_table(dev, flow_attr, MLX5_IB_FT_TX);
1226 		if (IS_ERR(ft_prio_tx)) {
1227 			err = PTR_ERR(ft_prio_tx);
1228 			ft_prio_tx = NULL;
1229 			goto destroy_ft;
1230 		}
1231 	}
1232 
1233 	if (is_egress) {
1234 		dst->type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1235 	} else {
1236 		dst->type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1237 		if (mqp->is_rss)
1238 			dst->tir_num = mqp->rss_qp.tirn;
1239 		else
1240 			dst->tir_num = mqp->raw_packet_qp.rq.tirn;
1241 	}
1242 
1243 	switch (flow_attr->type) {
1244 	case IB_FLOW_ATTR_NORMAL:
1245 		underlay_qpn = (mqp->flags & IB_QP_CREATE_SOURCE_QPN) ?
1246 				       mqp->underlay_qpn :
1247 				       0;
1248 		handler = _create_flow_rule(dev, ft_prio, flow_attr, dst,
1249 					    underlay_qpn, ucmd);
1250 		break;
1251 	case IB_FLOW_ATTR_ALL_DEFAULT:
1252 	case IB_FLOW_ATTR_MC_DEFAULT:
1253 		handler = create_leftovers_rule(dev, ft_prio, flow_attr, dst);
1254 		break;
1255 	case IB_FLOW_ATTR_SNIFFER:
1256 		handler = create_sniffer_rule(dev, ft_prio, ft_prio_tx, dst);
1257 		break;
1258 	default:
1259 		err = -EINVAL;
1260 		goto destroy_ft;
1261 	}
1262 
1263 	if (IS_ERR(handler)) {
1264 		err = PTR_ERR(handler);
1265 		handler = NULL;
1266 		goto destroy_ft;
1267 	}
1268 
1269 	mutex_unlock(&dev->flow_db->lock);
1270 	kfree(dst);
1271 	kfree(ucmd);
1272 
1273 	return &handler->ibflow;
1274 
1275 destroy_ft:
1276 	put_flow_table(dev, ft_prio, false);
1277 	if (ft_prio_tx)
1278 		put_flow_table(dev, ft_prio_tx, false);
1279 unlock:
1280 	mutex_unlock(&dev->flow_db->lock);
1281 	kfree(dst);
1282 free_ucmd:
1283 	kfree(ucmd);
1284 	return ERR_PTR(err);
1285 }
1286 
1287 static struct mlx5_ib_flow_prio *
1288 _get_flow_table(struct mlx5_ib_dev *dev,
1289 		struct mlx5_ib_flow_matcher *fs_matcher,
1290 		bool mcast)
1291 {
1292 	struct mlx5_flow_namespace *ns = NULL;
1293 	struct mlx5_ib_flow_prio *prio = NULL;
1294 	int max_table_size = 0;
1295 	bool esw_encap;
1296 	u32 flags = 0;
1297 	int priority;
1298 
1299 	if (mcast)
1300 		priority = MLX5_IB_FLOW_MCAST_PRIO;
1301 	else
1302 		priority = ib_prio_to_core_prio(fs_matcher->priority, false);
1303 
1304 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
1305 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
1306 	switch (fs_matcher->ns_type) {
1307 	case MLX5_FLOW_NAMESPACE_BYPASS:
1308 		max_table_size = BIT(
1309 			MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, log_max_ft_size));
1310 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap) && !esw_encap)
1311 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1312 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
1313 					      reformat_l3_tunnel_to_l2) &&
1314 		    !esw_encap)
1315 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1316 		break;
1317 	case 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) &&
1321 		    !esw_encap)
1322 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1323 		break;
1324 	case MLX5_FLOW_NAMESPACE_FDB:
1325 		max_table_size = BIT(
1326 			MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, log_max_ft_size));
1327 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, decap) && esw_encap)
1328 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1329 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev,
1330 					       reformat_l3_tunnel_to_l2) &&
1331 		    esw_encap)
1332 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1333 		priority = FDB_BYPASS_PATH;
1334 		break;
1335 	case MLX5_FLOW_NAMESPACE_RDMA_RX:
1336 		max_table_size = BIT(
1337 			MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev, log_max_ft_size));
1338 		priority = fs_matcher->priority;
1339 		break;
1340 	case MLX5_FLOW_NAMESPACE_RDMA_TX:
1341 		max_table_size = BIT(
1342 			MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev, log_max_ft_size));
1343 		priority = fs_matcher->priority;
1344 		break;
1345 	default:
1346 		break;
1347 	}
1348 
1349 	max_table_size = min_t(int, max_table_size, MLX5_FS_MAX_ENTRIES);
1350 
1351 	ns = mlx5_get_flow_namespace(dev->mdev, fs_matcher->ns_type);
1352 	if (!ns)
1353 		return ERR_PTR(-EOPNOTSUPP);
1354 
1355 	switch (fs_matcher->ns_type) {
1356 	case MLX5_FLOW_NAMESPACE_BYPASS:
1357 		prio = &dev->flow_db->prios[priority];
1358 		break;
1359 	case MLX5_FLOW_NAMESPACE_EGRESS:
1360 		prio = &dev->flow_db->egress_prios[priority];
1361 		break;
1362 	case MLX5_FLOW_NAMESPACE_FDB:
1363 		prio = &dev->flow_db->fdb;
1364 		break;
1365 	case MLX5_FLOW_NAMESPACE_RDMA_RX:
1366 		prio = &dev->flow_db->rdma_rx[priority];
1367 		break;
1368 	case MLX5_FLOW_NAMESPACE_RDMA_TX:
1369 		prio = &dev->flow_db->rdma_tx[priority];
1370 		break;
1371 	default: return ERR_PTR(-EINVAL);
1372 	}
1373 
1374 	if (!prio)
1375 		return ERR_PTR(-EINVAL);
1376 
1377 	if (prio->flow_table)
1378 		return prio;
1379 
1380 	return _get_prio(ns, prio, priority, max_table_size,
1381 			 MLX5_FS_MAX_TYPES, flags);
1382 }
1383 
1384 static struct mlx5_ib_flow_handler *
1385 _create_raw_flow_rule(struct mlx5_ib_dev *dev,
1386 		      struct mlx5_ib_flow_prio *ft_prio,
1387 		      struct mlx5_flow_destination *dst,
1388 		      struct mlx5_ib_flow_matcher  *fs_matcher,
1389 		      struct mlx5_flow_context *flow_context,
1390 		      struct mlx5_flow_act *flow_act,
1391 		      void *cmd_in, int inlen,
1392 		      int dst_num)
1393 {
1394 	struct mlx5_ib_flow_handler *handler;
1395 	struct mlx5_flow_spec *spec;
1396 	struct mlx5_flow_table *ft = ft_prio->flow_table;
1397 	int err = 0;
1398 
1399 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1400 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
1401 	if (!handler || !spec) {
1402 		err = -ENOMEM;
1403 		goto free;
1404 	}
1405 
1406 	INIT_LIST_HEAD(&handler->list);
1407 
1408 	memcpy(spec->match_value, cmd_in, inlen);
1409 	memcpy(spec->match_criteria, fs_matcher->matcher_mask.match_params,
1410 	       fs_matcher->mask_len);
1411 	spec->match_criteria_enable = fs_matcher->match_criteria_enable;
1412 	spec->flow_context = *flow_context;
1413 
1414 	handler->rule = mlx5_add_flow_rules(ft, spec,
1415 					    flow_act, dst, dst_num);
1416 
1417 	if (IS_ERR(handler->rule)) {
1418 		err = PTR_ERR(handler->rule);
1419 		goto free;
1420 	}
1421 
1422 	ft_prio->refcount++;
1423 	handler->prio = ft_prio;
1424 	handler->dev = dev;
1425 	ft_prio->flow_table = ft;
1426 
1427 free:
1428 	if (err)
1429 		kfree(handler);
1430 	kvfree(spec);
1431 	return err ? ERR_PTR(err) : handler;
1432 }
1433 
1434 static bool raw_fs_is_multicast(struct mlx5_ib_flow_matcher *fs_matcher,
1435 				void *match_v)
1436 {
1437 	void *match_c;
1438 	void *match_v_set_lyr_2_4, *match_c_set_lyr_2_4;
1439 	void *dmac, *dmac_mask;
1440 	void *ipv4, *ipv4_mask;
1441 
1442 	if (!(fs_matcher->match_criteria_enable &
1443 	      (1 << MATCH_CRITERIA_ENABLE_OUTER_BIT)))
1444 		return false;
1445 
1446 	match_c = fs_matcher->matcher_mask.match_params;
1447 	match_v_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_v,
1448 					   outer_headers);
1449 	match_c_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_c,
1450 					   outer_headers);
1451 
1452 	dmac = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1453 			    dmac_47_16);
1454 	dmac_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1455 				 dmac_47_16);
1456 
1457 	if (is_multicast_ether_addr(dmac) &&
1458 	    is_multicast_ether_addr(dmac_mask))
1459 		return true;
1460 
1461 	ipv4 = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1462 			    dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1463 
1464 	ipv4_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1465 				 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1466 
1467 	if (ipv4_is_multicast(*(__be32 *)(ipv4)) &&
1468 	    ipv4_is_multicast(*(__be32 *)(ipv4_mask)))
1469 		return true;
1470 
1471 	return false;
1472 }
1473 
1474 static struct mlx5_ib_flow_handler *raw_fs_rule_add(
1475 	struct mlx5_ib_dev *dev, struct mlx5_ib_flow_matcher *fs_matcher,
1476 	struct mlx5_flow_context *flow_context, struct mlx5_flow_act *flow_act,
1477 	u32 counter_id, void *cmd_in, int inlen, int dest_id, int dest_type)
1478 {
1479 	struct mlx5_flow_destination *dst;
1480 	struct mlx5_ib_flow_prio *ft_prio;
1481 	struct mlx5_ib_flow_handler *handler;
1482 	int dst_num = 0;
1483 	bool mcast;
1484 	int err;
1485 
1486 	if (fs_matcher->flow_type != MLX5_IB_FLOW_TYPE_NORMAL)
1487 		return ERR_PTR(-EOPNOTSUPP);
1488 
1489 	if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO)
1490 		return ERR_PTR(-ENOMEM);
1491 
1492 	dst = kcalloc(2, sizeof(*dst), GFP_KERNEL);
1493 	if (!dst)
1494 		return ERR_PTR(-ENOMEM);
1495 
1496 	mcast = raw_fs_is_multicast(fs_matcher, cmd_in);
1497 	mutex_lock(&dev->flow_db->lock);
1498 
1499 	ft_prio = _get_flow_table(dev, fs_matcher, mcast);
1500 	if (IS_ERR(ft_prio)) {
1501 		err = PTR_ERR(ft_prio);
1502 		goto unlock;
1503 	}
1504 
1505 	switch (dest_type) {
1506 	case MLX5_FLOW_DESTINATION_TYPE_TIR:
1507 		dst[dst_num].type = dest_type;
1508 		dst[dst_num++].tir_num = dest_id;
1509 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1510 		break;
1511 	case MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE:
1512 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM;
1513 		dst[dst_num++].ft_num = dest_id;
1514 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1515 		break;
1516 	case MLX5_FLOW_DESTINATION_TYPE_PORT:
1517 		dst[dst_num++].type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1518 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1519 		break;
1520 	default:
1521 		break;
1522 	}
1523 
1524 	if (flow_act->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1525 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1526 		dst[dst_num].counter_id = counter_id;
1527 		dst_num++;
1528 	}
1529 
1530 	handler = _create_raw_flow_rule(dev, ft_prio, dst_num ? dst : NULL,
1531 					fs_matcher, flow_context, flow_act,
1532 					cmd_in, inlen, dst_num);
1533 
1534 	if (IS_ERR(handler)) {
1535 		err = PTR_ERR(handler);
1536 		goto destroy_ft;
1537 	}
1538 
1539 	mutex_unlock(&dev->flow_db->lock);
1540 	atomic_inc(&fs_matcher->usecnt);
1541 	handler->flow_matcher = fs_matcher;
1542 
1543 	kfree(dst);
1544 
1545 	return handler;
1546 
1547 destroy_ft:
1548 	put_flow_table(dev, ft_prio, false);
1549 unlock:
1550 	mutex_unlock(&dev->flow_db->lock);
1551 	kfree(dst);
1552 
1553 	return ERR_PTR(err);
1554 }
1555 
1556 static u32 mlx5_ib_flow_action_flags_to_accel_xfrm_flags(u32 mlx5_flags)
1557 {
1558 	u32 flags = 0;
1559 
1560 	if (mlx5_flags & MLX5_IB_UAPI_FLOW_ACTION_FLAGS_REQUIRE_METADATA)
1561 		flags |= MLX5_ACCEL_XFRM_FLAG_REQUIRE_METADATA;
1562 
1563 	return flags;
1564 }
1565 
1566 #define MLX5_FLOW_ACTION_ESP_CREATE_LAST_SUPPORTED                             \
1567 	MLX5_IB_UAPI_FLOW_ACTION_FLAGS_REQUIRE_METADATA
1568 static struct ib_flow_action *
1569 mlx5_ib_create_flow_action_esp(struct ib_device *device,
1570 			       const struct ib_flow_action_attrs_esp *attr,
1571 			       struct uverbs_attr_bundle *attrs)
1572 {
1573 	struct mlx5_ib_dev *mdev = to_mdev(device);
1574 	struct ib_uverbs_flow_action_esp_keymat_aes_gcm *aes_gcm;
1575 	struct mlx5_accel_esp_xfrm_attrs accel_attrs = {};
1576 	struct mlx5_ib_flow_action *action;
1577 	u64 action_flags;
1578 	u64 flags;
1579 	int err = 0;
1580 
1581 	err = uverbs_get_flags64(
1582 		&action_flags, attrs, MLX5_IB_ATTR_CREATE_FLOW_ACTION_FLAGS,
1583 		((MLX5_FLOW_ACTION_ESP_CREATE_LAST_SUPPORTED << 1) - 1));
1584 	if (err)
1585 		return ERR_PTR(err);
1586 
1587 	flags = mlx5_ib_flow_action_flags_to_accel_xfrm_flags(action_flags);
1588 
1589 	/* We current only support a subset of the standard features. Only a
1590 	 * keymat of type AES_GCM, with icv_len == 16, iv_algo == SEQ and esn
1591 	 * (with overlap). Full offload mode isn't supported.
1592 	 */
1593 	if (!attr->keymat || attr->replay || attr->encap ||
1594 	    attr->spi || attr->seq || attr->tfc_pad ||
1595 	    attr->hard_limit_pkts ||
1596 	    (attr->flags & ~(IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED |
1597 			     IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ENCRYPT)))
1598 		return ERR_PTR(-EOPNOTSUPP);
1599 
1600 	if (attr->keymat->protocol !=
1601 	    IB_UVERBS_FLOW_ACTION_ESP_KEYMAT_AES_GCM)
1602 		return ERR_PTR(-EOPNOTSUPP);
1603 
1604 	aes_gcm = &attr->keymat->keymat.aes_gcm;
1605 
1606 	if (aes_gcm->icv_len != 16 ||
1607 	    aes_gcm->iv_algo != IB_UVERBS_FLOW_ACTION_IV_ALGO_SEQ)
1608 		return ERR_PTR(-EOPNOTSUPP);
1609 
1610 	action = kmalloc(sizeof(*action), GFP_KERNEL);
1611 	if (!action)
1612 		return ERR_PTR(-ENOMEM);
1613 
1614 	action->esp_aes_gcm.ib_flags = attr->flags;
1615 	memcpy(&accel_attrs.keymat.aes_gcm.aes_key, &aes_gcm->aes_key,
1616 	       sizeof(accel_attrs.keymat.aes_gcm.aes_key));
1617 	accel_attrs.keymat.aes_gcm.key_len = aes_gcm->key_len * 8;
1618 	memcpy(&accel_attrs.keymat.aes_gcm.salt, &aes_gcm->salt,
1619 	       sizeof(accel_attrs.keymat.aes_gcm.salt));
1620 	memcpy(&accel_attrs.keymat.aes_gcm.seq_iv, &aes_gcm->iv,
1621 	       sizeof(accel_attrs.keymat.aes_gcm.seq_iv));
1622 	accel_attrs.keymat.aes_gcm.icv_len = aes_gcm->icv_len * 8;
1623 	accel_attrs.keymat.aes_gcm.iv_algo = MLX5_ACCEL_ESP_AES_GCM_IV_ALGO_SEQ;
1624 	accel_attrs.keymat_type = MLX5_ACCEL_ESP_KEYMAT_AES_GCM;
1625 
1626 	accel_attrs.esn = attr->esn;
1627 	if (attr->flags & IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED)
1628 		accel_attrs.flags |= MLX5_ACCEL_ESP_FLAGS_ESN_TRIGGERED;
1629 	if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)
1630 		accel_attrs.flags |= MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
1631 
1632 	if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ENCRYPT)
1633 		accel_attrs.action |= MLX5_ACCEL_ESP_ACTION_ENCRYPT;
1634 
1635 	action->esp_aes_gcm.ctx =
1636 		mlx5_accel_esp_create_xfrm(mdev->mdev, &accel_attrs, flags);
1637 	if (IS_ERR(action->esp_aes_gcm.ctx)) {
1638 		err = PTR_ERR(action->esp_aes_gcm.ctx);
1639 		goto err_parse;
1640 	}
1641 
1642 	action->esp_aes_gcm.ib_flags = attr->flags;
1643 
1644 	return &action->ib_action;
1645 
1646 err_parse:
1647 	kfree(action);
1648 	return ERR_PTR(err);
1649 }
1650 
1651 static int
1652 mlx5_ib_modify_flow_action_esp(struct ib_flow_action *action,
1653 			       const struct ib_flow_action_attrs_esp *attr,
1654 			       struct uverbs_attr_bundle *attrs)
1655 {
1656 	struct mlx5_ib_flow_action *maction = to_mflow_act(action);
1657 	struct mlx5_accel_esp_xfrm_attrs accel_attrs;
1658 	int err = 0;
1659 
1660 	if (attr->keymat || attr->replay || attr->encap ||
1661 	    attr->spi || attr->seq || attr->tfc_pad ||
1662 	    attr->hard_limit_pkts ||
1663 	    (attr->flags & ~(IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED |
1664 			     IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS |
1665 			     IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)))
1666 		return -EOPNOTSUPP;
1667 
1668 	/* Only the ESN value or the MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP can
1669 	 * be modified.
1670 	 */
1671 	if (!(maction->esp_aes_gcm.ib_flags &
1672 	      IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED) &&
1673 	    attr->flags & (IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED |
1674 			   IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW))
1675 		return -EINVAL;
1676 
1677 	memcpy(&accel_attrs, &maction->esp_aes_gcm.ctx->attrs,
1678 	       sizeof(accel_attrs));
1679 
1680 	accel_attrs.esn = attr->esn;
1681 	if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)
1682 		accel_attrs.flags |= MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
1683 	else
1684 		accel_attrs.flags &= ~MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
1685 
1686 	err = mlx5_accel_esp_modify_xfrm(maction->esp_aes_gcm.ctx,
1687 					 &accel_attrs);
1688 	if (err)
1689 		return err;
1690 
1691 	maction->esp_aes_gcm.ib_flags &=
1692 		~IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW;
1693 	maction->esp_aes_gcm.ib_flags |=
1694 		attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW;
1695 
1696 	return 0;
1697 }
1698 
1699 static void destroy_flow_action_raw(struct mlx5_ib_flow_action *maction)
1700 {
1701 	switch (maction->flow_action_raw.sub_type) {
1702 	case MLX5_IB_FLOW_ACTION_MODIFY_HEADER:
1703 		mlx5_modify_header_dealloc(maction->flow_action_raw.dev->mdev,
1704 					   maction->flow_action_raw.modify_hdr);
1705 		break;
1706 	case MLX5_IB_FLOW_ACTION_PACKET_REFORMAT:
1707 		mlx5_packet_reformat_dealloc(maction->flow_action_raw.dev->mdev,
1708 					     maction->flow_action_raw.pkt_reformat);
1709 		break;
1710 	case MLX5_IB_FLOW_ACTION_DECAP:
1711 		break;
1712 	default:
1713 		break;
1714 	}
1715 }
1716 
1717 static int mlx5_ib_destroy_flow_action(struct ib_flow_action *action)
1718 {
1719 	struct mlx5_ib_flow_action *maction = to_mflow_act(action);
1720 
1721 	switch (action->type) {
1722 	case IB_FLOW_ACTION_ESP:
1723 		/*
1724 		 * We only support aes_gcm by now, so we implicitly know this is
1725 		 * the underline crypto.
1726 		 */
1727 		mlx5_accel_esp_destroy_xfrm(maction->esp_aes_gcm.ctx);
1728 		break;
1729 	case IB_FLOW_ACTION_UNSPECIFIED:
1730 		destroy_flow_action_raw(maction);
1731 		break;
1732 	default:
1733 		WARN_ON(true);
1734 		break;
1735 	}
1736 
1737 	kfree(maction);
1738 	return 0;
1739 }
1740 
1741 static int
1742 mlx5_ib_ft_type_to_namespace(enum mlx5_ib_uapi_flow_table_type table_type,
1743 			     enum mlx5_flow_namespace_type *namespace)
1744 {
1745 	switch (table_type) {
1746 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX:
1747 		*namespace = MLX5_FLOW_NAMESPACE_BYPASS;
1748 		break;
1749 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX:
1750 		*namespace = MLX5_FLOW_NAMESPACE_EGRESS;
1751 		break;
1752 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_FDB:
1753 		*namespace = MLX5_FLOW_NAMESPACE_FDB;
1754 		break;
1755 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_RX:
1756 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_RX;
1757 		break;
1758 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_TX:
1759 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_TX;
1760 		break;
1761 	default:
1762 		return -EINVAL;
1763 	}
1764 
1765 	return 0;
1766 }
1767 
1768 static const struct uverbs_attr_spec mlx5_ib_flow_type[] = {
1769 	[MLX5_IB_FLOW_TYPE_NORMAL] = {
1770 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1771 		.u.ptr = {
1772 			.len = sizeof(u16), /* data is priority */
1773 			.min_len = sizeof(u16),
1774 		}
1775 	},
1776 	[MLX5_IB_FLOW_TYPE_SNIFFER] = {
1777 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1778 		UVERBS_ATTR_NO_DATA(),
1779 	},
1780 	[MLX5_IB_FLOW_TYPE_ALL_DEFAULT] = {
1781 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1782 		UVERBS_ATTR_NO_DATA(),
1783 	},
1784 	[MLX5_IB_FLOW_TYPE_MC_DEFAULT] = {
1785 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1786 		UVERBS_ATTR_NO_DATA(),
1787 	},
1788 };
1789 
1790 static bool is_flow_dest(void *obj, int *dest_id, int *dest_type)
1791 {
1792 	struct devx_obj *devx_obj = obj;
1793 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1794 
1795 	switch (opcode) {
1796 	case MLX5_CMD_OP_DESTROY_TIR:
1797 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1798 		*dest_id = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox,
1799 				    obj_id);
1800 		return true;
1801 
1802 	case MLX5_CMD_OP_DESTROY_FLOW_TABLE:
1803 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1804 		*dest_id = MLX5_GET(destroy_flow_table_in, devx_obj->dinbox,
1805 				    table_id);
1806 		return true;
1807 	default:
1808 		return false;
1809 	}
1810 }
1811 
1812 static int get_dests(struct uverbs_attr_bundle *attrs,
1813 		     struct mlx5_ib_flow_matcher *fs_matcher, int *dest_id,
1814 		     int *dest_type, struct ib_qp **qp, u32 *flags)
1815 {
1816 	bool dest_devx, dest_qp;
1817 	void *devx_obj;
1818 	int err;
1819 
1820 	dest_devx = uverbs_attr_is_valid(attrs,
1821 					 MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1822 	dest_qp = uverbs_attr_is_valid(attrs,
1823 				       MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1824 
1825 	*flags = 0;
1826 	err = uverbs_get_flags32(flags, attrs, MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
1827 				 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS |
1828 					 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP);
1829 	if (err)
1830 		return err;
1831 
1832 	/* Both flags are not allowed */
1833 	if (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS &&
1834 	    *flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1835 		return -EINVAL;
1836 
1837 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_BYPASS) {
1838 		if (dest_devx && (dest_qp || *flags))
1839 			return -EINVAL;
1840 		else if (dest_qp && *flags)
1841 			return -EINVAL;
1842 	}
1843 
1844 	/* Allow only DEVX object, drop as dest for FDB */
1845 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB && !(dest_devx ||
1846 	     (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)))
1847 		return -EINVAL;
1848 
1849 	/* Allow only DEVX object or QP as dest when inserting to RDMA_RX */
1850 	if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1851 	    ((!dest_devx && !dest_qp) || (dest_devx && dest_qp)))
1852 		return -EINVAL;
1853 
1854 	*qp = NULL;
1855 	if (dest_devx) {
1856 		devx_obj =
1857 			uverbs_attr_get_obj(attrs,
1858 					    MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1859 
1860 		/* Verify that the given DEVX object is a flow
1861 		 * steering destination.
1862 		 */
1863 		if (!is_flow_dest(devx_obj, dest_id, dest_type))
1864 			return -EINVAL;
1865 		/* Allow only flow table as dest when inserting to FDB or RDMA_RX */
1866 		if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB ||
1867 		     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1868 		    *dest_type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1869 			return -EINVAL;
1870 	} else if (dest_qp) {
1871 		struct mlx5_ib_qp *mqp;
1872 
1873 		*qp = uverbs_attr_get_obj(attrs,
1874 					  MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1875 		if (IS_ERR(*qp))
1876 			return PTR_ERR(*qp);
1877 
1878 		if ((*qp)->qp_type != IB_QPT_RAW_PACKET)
1879 			return -EINVAL;
1880 
1881 		mqp = to_mqp(*qp);
1882 		if (mqp->is_rss)
1883 			*dest_id = mqp->rss_qp.tirn;
1884 		else
1885 			*dest_id = mqp->raw_packet_qp.rq.tirn;
1886 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1887 	} else if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1888 		    fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX) &&
1889 		   !(*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)) {
1890 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1891 	}
1892 
1893 	if (*dest_type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1894 	    (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1895 	     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX))
1896 		return -EINVAL;
1897 
1898 	return 0;
1899 }
1900 
1901 static bool is_flow_counter(void *obj, u32 offset, u32 *counter_id)
1902 {
1903 	struct devx_obj *devx_obj = obj;
1904 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1905 
1906 	if (opcode == MLX5_CMD_OP_DEALLOC_FLOW_COUNTER) {
1907 
1908 		if (offset && offset >= devx_obj->flow_counter_bulk_size)
1909 			return false;
1910 
1911 		*counter_id = MLX5_GET(dealloc_flow_counter_in,
1912 				       devx_obj->dinbox,
1913 				       flow_counter_id);
1914 		*counter_id += offset;
1915 		return true;
1916 	}
1917 
1918 	return false;
1919 }
1920 
1921 #define MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS 2
1922 static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)(
1923 	struct uverbs_attr_bundle *attrs)
1924 {
1925 	struct mlx5_flow_context flow_context = {.flow_tag =
1926 		MLX5_FS_DEFAULT_FLOW_TAG};
1927 	u32 *offset_attr, offset = 0, counter_id = 0;
1928 	int dest_id, dest_type = -1, inlen, len, ret, i;
1929 	struct mlx5_ib_flow_handler *flow_handler;
1930 	struct mlx5_ib_flow_matcher *fs_matcher;
1931 	struct ib_uobject **arr_flow_actions;
1932 	struct ib_uflow_resources *uflow_res;
1933 	struct mlx5_flow_act flow_act = {};
1934 	struct ib_qp *qp = NULL;
1935 	void *devx_obj, *cmd_in;
1936 	struct ib_uobject *uobj;
1937 	struct mlx5_ib_dev *dev;
1938 	u32 flags;
1939 
1940 	if (!capable(CAP_NET_RAW))
1941 		return -EPERM;
1942 
1943 	fs_matcher = uverbs_attr_get_obj(attrs,
1944 					 MLX5_IB_ATTR_CREATE_FLOW_MATCHER);
1945 	uobj =  uverbs_attr_get_uobject(attrs, MLX5_IB_ATTR_CREATE_FLOW_HANDLE);
1946 	dev = mlx5_udata_to_mdev(&attrs->driver_udata);
1947 
1948 	if (get_dests(attrs, fs_matcher, &dest_id, &dest_type, &qp, &flags))
1949 		return -EINVAL;
1950 
1951 	if (flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS)
1952 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
1953 
1954 	if (flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1955 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1956 
1957 	len = uverbs_attr_get_uobjs_arr(attrs,
1958 		MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX, &arr_flow_actions);
1959 	if (len) {
1960 		devx_obj = arr_flow_actions[0]->object;
1961 
1962 		if (uverbs_attr_is_valid(attrs,
1963 					 MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET)) {
1964 
1965 			int num_offsets = uverbs_attr_ptr_get_array_size(
1966 				attrs,
1967 				MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
1968 				sizeof(u32));
1969 
1970 			if (num_offsets != 1)
1971 				return -EINVAL;
1972 
1973 			offset_attr = uverbs_attr_get_alloced_ptr(
1974 				attrs,
1975 				MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET);
1976 			offset = *offset_attr;
1977 		}
1978 
1979 		if (!is_flow_counter(devx_obj, offset, &counter_id))
1980 			return -EINVAL;
1981 
1982 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
1983 	}
1984 
1985 	cmd_in = uverbs_attr_get_alloced_ptr(
1986 		attrs, MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE);
1987 	inlen = uverbs_attr_get_len(attrs,
1988 				    MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE);
1989 
1990 	uflow_res = flow_resources_alloc(MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS);
1991 	if (!uflow_res)
1992 		return -ENOMEM;
1993 
1994 	len = uverbs_attr_get_uobjs_arr(attrs,
1995 		MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS, &arr_flow_actions);
1996 	for (i = 0; i < len; i++) {
1997 		struct mlx5_ib_flow_action *maction =
1998 			to_mflow_act(arr_flow_actions[i]->object);
1999 
2000 		ret = parse_flow_flow_action(maction, false, &flow_act);
2001 		if (ret)
2002 			goto err_out;
2003 		flow_resources_add(uflow_res, IB_FLOW_SPEC_ACTION_HANDLE,
2004 				   arr_flow_actions[i]->object);
2005 	}
2006 
2007 	ret = uverbs_copy_from(&flow_context.flow_tag, attrs,
2008 			       MLX5_IB_ATTR_CREATE_FLOW_TAG);
2009 	if (!ret) {
2010 		if (flow_context.flow_tag >= BIT(24)) {
2011 			ret = -EINVAL;
2012 			goto err_out;
2013 		}
2014 		flow_context.flags |= FLOW_CONTEXT_HAS_TAG;
2015 	}
2016 
2017 	flow_handler =
2018 		raw_fs_rule_add(dev, fs_matcher, &flow_context, &flow_act,
2019 				counter_id, cmd_in, inlen, dest_id, dest_type);
2020 	if (IS_ERR(flow_handler)) {
2021 		ret = PTR_ERR(flow_handler);
2022 		goto err_out;
2023 	}
2024 
2025 	ib_set_flow(uobj, &flow_handler->ibflow, qp, &dev->ib_dev, uflow_res);
2026 
2027 	return 0;
2028 err_out:
2029 	ib_uverbs_flow_resources_free(uflow_res);
2030 	return ret;
2031 }
2032 
2033 static int flow_matcher_cleanup(struct ib_uobject *uobject,
2034 				enum rdma_remove_reason why,
2035 				struct uverbs_attr_bundle *attrs)
2036 {
2037 	struct mlx5_ib_flow_matcher *obj = uobject->object;
2038 
2039 	if (atomic_read(&obj->usecnt))
2040 		return -EBUSY;
2041 
2042 	kfree(obj);
2043 	return 0;
2044 }
2045 
2046 static int mlx5_ib_matcher_ns(struct uverbs_attr_bundle *attrs,
2047 			      struct mlx5_ib_flow_matcher *obj)
2048 {
2049 	enum mlx5_ib_uapi_flow_table_type ft_type =
2050 		MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX;
2051 	u32 flags;
2052 	int err;
2053 
2054 	/* New users should use MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE and older
2055 	 * users should switch to it. We leave this to not break userspace
2056 	 */
2057 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE) &&
2058 	    uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS))
2059 		return -EINVAL;
2060 
2061 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE)) {
2062 		err = uverbs_get_const(&ft_type, attrs,
2063 				       MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE);
2064 		if (err)
2065 			return err;
2066 
2067 		err = mlx5_ib_ft_type_to_namespace(ft_type, &obj->ns_type);
2068 		if (err)
2069 			return err;
2070 
2071 		return 0;
2072 	}
2073 
2074 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS)) {
2075 		err = uverbs_get_flags32(&flags, attrs,
2076 					 MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2077 					 IB_FLOW_ATTR_FLAGS_EGRESS);
2078 		if (err)
2079 			return err;
2080 
2081 		if (flags) {
2082 			mlx5_ib_ft_type_to_namespace(
2083 				MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX,
2084 				&obj->ns_type);
2085 			return 0;
2086 		}
2087 	}
2088 
2089 	obj->ns_type = MLX5_FLOW_NAMESPACE_BYPASS;
2090 
2091 	return 0;
2092 }
2093 
2094 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_MATCHER_CREATE)(
2095 	struct uverbs_attr_bundle *attrs)
2096 {
2097 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2098 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE);
2099 	struct mlx5_ib_dev *dev = mlx5_udata_to_mdev(&attrs->driver_udata);
2100 	struct mlx5_ib_flow_matcher *obj;
2101 	int err;
2102 
2103 	obj = kzalloc(sizeof(struct mlx5_ib_flow_matcher), GFP_KERNEL);
2104 	if (!obj)
2105 		return -ENOMEM;
2106 
2107 	obj->mask_len = uverbs_attr_get_len(
2108 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2109 	err = uverbs_copy_from(&obj->matcher_mask,
2110 			       attrs,
2111 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2112 	if (err)
2113 		goto end;
2114 
2115 	obj->flow_type = uverbs_attr_get_enum_id(
2116 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2117 
2118 	if (obj->flow_type == MLX5_IB_FLOW_TYPE_NORMAL) {
2119 		err = uverbs_copy_from(&obj->priority,
2120 				       attrs,
2121 				       MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2122 		if (err)
2123 			goto end;
2124 	}
2125 
2126 	err = uverbs_copy_from(&obj->match_criteria_enable,
2127 			       attrs,
2128 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA);
2129 	if (err)
2130 		goto end;
2131 
2132 	err = mlx5_ib_matcher_ns(attrs, obj);
2133 	if (err)
2134 		goto end;
2135 
2136 	if (obj->ns_type == MLX5_FLOW_NAMESPACE_FDB &&
2137 	    mlx5_eswitch_mode(dev->mdev) != MLX5_ESWITCH_OFFLOADS) {
2138 		err = -EINVAL;
2139 		goto end;
2140 	}
2141 
2142 	uobj->object = obj;
2143 	obj->mdev = dev->mdev;
2144 	atomic_set(&obj->usecnt, 0);
2145 	return 0;
2146 
2147 end:
2148 	kfree(obj);
2149 	return err;
2150 }
2151 
2152 static struct ib_flow_action *
2153 mlx5_ib_create_modify_header(struct mlx5_ib_dev *dev,
2154 			     enum mlx5_ib_uapi_flow_table_type ft_type,
2155 			     u8 num_actions, void *in)
2156 {
2157 	enum mlx5_flow_namespace_type namespace;
2158 	struct mlx5_ib_flow_action *maction;
2159 	int ret;
2160 
2161 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2162 	if (ret)
2163 		return ERR_PTR(-EINVAL);
2164 
2165 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2166 	if (!maction)
2167 		return ERR_PTR(-ENOMEM);
2168 
2169 	maction->flow_action_raw.modify_hdr =
2170 		mlx5_modify_header_alloc(dev->mdev, namespace, num_actions, in);
2171 
2172 	if (IS_ERR(maction->flow_action_raw.modify_hdr)) {
2173 		ret = PTR_ERR(maction->flow_action_raw.modify_hdr);
2174 		kfree(maction);
2175 		return ERR_PTR(ret);
2176 	}
2177 	maction->flow_action_raw.sub_type =
2178 		MLX5_IB_FLOW_ACTION_MODIFY_HEADER;
2179 	maction->flow_action_raw.dev = dev;
2180 
2181 	return &maction->ib_action;
2182 }
2183 
2184 static bool mlx5_ib_modify_header_supported(struct mlx5_ib_dev *dev)
2185 {
2186 	return MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
2187 					 max_modify_header_actions) ||
2188 	       MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev,
2189 					 max_modify_header_actions) ||
2190 	       MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
2191 					 max_modify_header_actions);
2192 }
2193 
2194 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER)(
2195 	struct uverbs_attr_bundle *attrs)
2196 {
2197 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2198 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE);
2199 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2200 	enum mlx5_ib_uapi_flow_table_type ft_type;
2201 	struct ib_flow_action *action;
2202 	int num_actions;
2203 	void *in;
2204 	int ret;
2205 
2206 	if (!mlx5_ib_modify_header_supported(mdev))
2207 		return -EOPNOTSUPP;
2208 
2209 	in = uverbs_attr_get_alloced_ptr(attrs,
2210 		MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM);
2211 
2212 	num_actions = uverbs_attr_ptr_get_array_size(
2213 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2214 		MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto));
2215 	if (num_actions < 0)
2216 		return num_actions;
2217 
2218 	ret = uverbs_get_const(&ft_type, attrs,
2219 			       MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE);
2220 	if (ret)
2221 		return ret;
2222 	action = mlx5_ib_create_modify_header(mdev, ft_type, num_actions, in);
2223 	if (IS_ERR(action))
2224 		return PTR_ERR(action);
2225 
2226 	uverbs_flow_action_fill_action(action, uobj, &mdev->ib_dev,
2227 				       IB_FLOW_ACTION_UNSPECIFIED);
2228 
2229 	return 0;
2230 }
2231 
2232 static bool mlx5_ib_flow_action_packet_reformat_valid(struct mlx5_ib_dev *ibdev,
2233 						      u8 packet_reformat_type,
2234 						      u8 ft_type)
2235 {
2236 	switch (packet_reformat_type) {
2237 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2238 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2239 			return MLX5_CAP_FLOWTABLE(ibdev->mdev,
2240 						  encap_general_header);
2241 		break;
2242 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2243 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2244 			return MLX5_CAP_FLOWTABLE_NIC_TX(ibdev->mdev,
2245 				reformat_l2_to_l3_tunnel);
2246 		break;
2247 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2248 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2249 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev,
2250 				reformat_l3_tunnel_to_l2);
2251 		break;
2252 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2:
2253 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2254 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev, decap);
2255 		break;
2256 	default:
2257 		break;
2258 	}
2259 
2260 	return false;
2261 }
2262 
2263 static int mlx5_ib_dv_to_prm_packet_reforamt_type(u8 dv_prt, u8 *prm_prt)
2264 {
2265 	switch (dv_prt) {
2266 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2267 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L2_TUNNEL;
2268 		break;
2269 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2270 		*prm_prt = MLX5_REFORMAT_TYPE_L3_TUNNEL_TO_L2;
2271 		break;
2272 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2273 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
2274 		break;
2275 	default:
2276 		return -EINVAL;
2277 	}
2278 
2279 	return 0;
2280 }
2281 
2282 static int mlx5_ib_flow_action_create_packet_reformat_ctx(
2283 	struct mlx5_ib_dev *dev,
2284 	struct mlx5_ib_flow_action *maction,
2285 	u8 ft_type, u8 dv_prt,
2286 	void *in, size_t len)
2287 {
2288 	struct mlx5_pkt_reformat_params reformat_params;
2289 	enum mlx5_flow_namespace_type namespace;
2290 	u8 prm_prt;
2291 	int ret;
2292 
2293 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2294 	if (ret)
2295 		return ret;
2296 
2297 	ret = mlx5_ib_dv_to_prm_packet_reforamt_type(dv_prt, &prm_prt);
2298 	if (ret)
2299 		return ret;
2300 
2301 	memset(&reformat_params, 0, sizeof(reformat_params));
2302 	reformat_params.type = prm_prt;
2303 	reformat_params.size = len;
2304 	reformat_params.data = in;
2305 	maction->flow_action_raw.pkt_reformat =
2306 		mlx5_packet_reformat_alloc(dev->mdev, &reformat_params,
2307 					   namespace);
2308 	if (IS_ERR(maction->flow_action_raw.pkt_reformat)) {
2309 		ret = PTR_ERR(maction->flow_action_raw.pkt_reformat);
2310 		return ret;
2311 	}
2312 
2313 	maction->flow_action_raw.sub_type =
2314 		MLX5_IB_FLOW_ACTION_PACKET_REFORMAT;
2315 	maction->flow_action_raw.dev = dev;
2316 
2317 	return 0;
2318 }
2319 
2320 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT)(
2321 	struct uverbs_attr_bundle *attrs)
2322 {
2323 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs,
2324 		MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE);
2325 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2326 	enum mlx5_ib_uapi_flow_action_packet_reformat_type dv_prt;
2327 	enum mlx5_ib_uapi_flow_table_type ft_type;
2328 	struct mlx5_ib_flow_action *maction;
2329 	int ret;
2330 
2331 	ret = uverbs_get_const(&ft_type, attrs,
2332 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE);
2333 	if (ret)
2334 		return ret;
2335 
2336 	ret = uverbs_get_const(&dv_prt, attrs,
2337 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE);
2338 	if (ret)
2339 		return ret;
2340 
2341 	if (!mlx5_ib_flow_action_packet_reformat_valid(mdev, dv_prt, ft_type))
2342 		return -EOPNOTSUPP;
2343 
2344 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2345 	if (!maction)
2346 		return -ENOMEM;
2347 
2348 	if (dv_prt ==
2349 	    MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2) {
2350 		maction->flow_action_raw.sub_type =
2351 			MLX5_IB_FLOW_ACTION_DECAP;
2352 		maction->flow_action_raw.dev = mdev;
2353 	} else {
2354 		void *in;
2355 		int len;
2356 
2357 		in = uverbs_attr_get_alloced_ptr(attrs,
2358 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2359 		if (IS_ERR(in)) {
2360 			ret = PTR_ERR(in);
2361 			goto free_maction;
2362 		}
2363 
2364 		len = uverbs_attr_get_len(attrs,
2365 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2366 
2367 		ret = mlx5_ib_flow_action_create_packet_reformat_ctx(mdev,
2368 			maction, ft_type, dv_prt, in, len);
2369 		if (ret)
2370 			goto free_maction;
2371 	}
2372 
2373 	uverbs_flow_action_fill_action(&maction->ib_action, uobj, &mdev->ib_dev,
2374 				       IB_FLOW_ACTION_UNSPECIFIED);
2375 	return 0;
2376 
2377 free_maction:
2378 	kfree(maction);
2379 	return ret;
2380 }
2381 
2382 DECLARE_UVERBS_NAMED_METHOD(
2383 	MLX5_IB_METHOD_CREATE_FLOW,
2384 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2385 			UVERBS_OBJECT_FLOW,
2386 			UVERBS_ACCESS_NEW,
2387 			UA_MANDATORY),
2388 	UVERBS_ATTR_PTR_IN(
2389 		MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE,
2390 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2391 		UA_MANDATORY,
2392 		UA_ALLOC_AND_COPY),
2393 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_MATCHER,
2394 			MLX5_IB_OBJECT_FLOW_MATCHER,
2395 			UVERBS_ACCESS_READ,
2396 			UA_MANDATORY),
2397 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_QP,
2398 			UVERBS_OBJECT_QP,
2399 			UVERBS_ACCESS_READ),
2400 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX,
2401 			MLX5_IB_OBJECT_DEVX_OBJ,
2402 			UVERBS_ACCESS_READ),
2403 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS,
2404 			     UVERBS_OBJECT_FLOW_ACTION,
2405 			     UVERBS_ACCESS_READ, 1,
2406 			     MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS,
2407 			     UA_OPTIONAL),
2408 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_TAG,
2409 			   UVERBS_ATTR_TYPE(u32),
2410 			   UA_OPTIONAL),
2411 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX,
2412 			     MLX5_IB_OBJECT_DEVX_OBJ,
2413 			     UVERBS_ACCESS_READ, 1, 1,
2414 			     UA_OPTIONAL),
2415 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
2416 			   UVERBS_ATTR_MIN_SIZE(sizeof(u32)),
2417 			   UA_OPTIONAL,
2418 			   UA_ALLOC_AND_COPY),
2419 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
2420 			     enum mlx5_ib_create_flow_flags,
2421 			     UA_OPTIONAL));
2422 
2423 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2424 	MLX5_IB_METHOD_DESTROY_FLOW,
2425 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2426 			UVERBS_OBJECT_FLOW,
2427 			UVERBS_ACCESS_DESTROY,
2428 			UA_MANDATORY));
2429 
2430 ADD_UVERBS_METHODS(mlx5_ib_fs,
2431 		   UVERBS_OBJECT_FLOW,
2432 		   &UVERBS_METHOD(MLX5_IB_METHOD_CREATE_FLOW),
2433 		   &UVERBS_METHOD(MLX5_IB_METHOD_DESTROY_FLOW));
2434 
2435 DECLARE_UVERBS_NAMED_METHOD(
2436 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER,
2437 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE,
2438 			UVERBS_OBJECT_FLOW_ACTION,
2439 			UVERBS_ACCESS_NEW,
2440 			UA_MANDATORY),
2441 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2442 			   UVERBS_ATTR_MIN_SIZE(MLX5_UN_SZ_BYTES(
2443 				   set_add_copy_action_in_auto)),
2444 			   UA_MANDATORY,
2445 			   UA_ALLOC_AND_COPY),
2446 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE,
2447 			     enum mlx5_ib_uapi_flow_table_type,
2448 			     UA_MANDATORY));
2449 
2450 DECLARE_UVERBS_NAMED_METHOD(
2451 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT,
2452 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE,
2453 			UVERBS_OBJECT_FLOW_ACTION,
2454 			UVERBS_ACCESS_NEW,
2455 			UA_MANDATORY),
2456 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF,
2457 			   UVERBS_ATTR_MIN_SIZE(1),
2458 			   UA_ALLOC_AND_COPY,
2459 			   UA_OPTIONAL),
2460 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE,
2461 			     enum mlx5_ib_uapi_flow_action_packet_reformat_type,
2462 			     UA_MANDATORY),
2463 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE,
2464 			     enum mlx5_ib_uapi_flow_table_type,
2465 			     UA_MANDATORY));
2466 
2467 ADD_UVERBS_METHODS(
2468 	mlx5_ib_flow_actions,
2469 	UVERBS_OBJECT_FLOW_ACTION,
2470 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER),
2471 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT));
2472 
2473 DECLARE_UVERBS_NAMED_METHOD(
2474 	MLX5_IB_METHOD_FLOW_MATCHER_CREATE,
2475 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE,
2476 			MLX5_IB_OBJECT_FLOW_MATCHER,
2477 			UVERBS_ACCESS_NEW,
2478 			UA_MANDATORY),
2479 	UVERBS_ATTR_PTR_IN(
2480 		MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK,
2481 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2482 		UA_MANDATORY),
2483 	UVERBS_ATTR_ENUM_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE,
2484 			    mlx5_ib_flow_type,
2485 			    UA_MANDATORY),
2486 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA,
2487 			   UVERBS_ATTR_TYPE(u8),
2488 			   UA_MANDATORY),
2489 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2490 			     enum ib_flow_flags,
2491 			     UA_OPTIONAL),
2492 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE,
2493 			     enum mlx5_ib_uapi_flow_table_type,
2494 			     UA_OPTIONAL));
2495 
2496 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2497 	MLX5_IB_METHOD_FLOW_MATCHER_DESTROY,
2498 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_DESTROY_HANDLE,
2499 			MLX5_IB_OBJECT_FLOW_MATCHER,
2500 			UVERBS_ACCESS_DESTROY,
2501 			UA_MANDATORY));
2502 
2503 DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_FLOW_MATCHER,
2504 			    UVERBS_TYPE_ALLOC_IDR(flow_matcher_cleanup),
2505 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_CREATE),
2506 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_DESTROY));
2507 
2508 const struct uapi_definition mlx5_ib_flow_defs[] = {
2509 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
2510 		MLX5_IB_OBJECT_FLOW_MATCHER),
2511 	UAPI_DEF_CHAIN_OBJ_TREE(
2512 		UVERBS_OBJECT_FLOW,
2513 		&mlx5_ib_fs),
2514 	UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_FLOW_ACTION,
2515 				&mlx5_ib_flow_actions),
2516 	{},
2517 };
2518 
2519 static const struct ib_device_ops flow_ops = {
2520 	.create_flow = mlx5_ib_create_flow,
2521 	.destroy_flow = mlx5_ib_destroy_flow,
2522 	.destroy_flow_action = mlx5_ib_destroy_flow_action,
2523 };
2524 
2525 static const struct ib_device_ops flow_ipsec_ops = {
2526 	.create_flow_action_esp = mlx5_ib_create_flow_action_esp,
2527 	.modify_flow_action_esp = mlx5_ib_modify_flow_action_esp,
2528 };
2529 
2530 int mlx5_ib_fs_init(struct mlx5_ib_dev *dev)
2531 {
2532 	dev->flow_db = kzalloc(sizeof(*dev->flow_db), GFP_KERNEL);
2533 
2534 	if (!dev->flow_db)
2535 		return -ENOMEM;
2536 
2537 	mutex_init(&dev->flow_db->lock);
2538 
2539 	ib_set_device_ops(&dev->ib_dev, &flow_ops);
2540 	if (mlx5_accel_ipsec_device_caps(dev->mdev) &
2541 	    MLX5_ACCEL_IPSEC_CAP_DEVICE)
2542 		ib_set_device_ops(&dev->ib_dev, &flow_ipsec_ops);
2543 
2544 	return 0;
2545 }
2546