xref: /openbmc/linux/drivers/infiniband/hw/mlx5/fs.c (revision 9134211f)
1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /*
3  * Copyright (c) 2018, Mellanox Technologies inc.  All rights reserved.
4  */
5 
6 #include <rdma/ib_user_verbs.h>
7 #include <rdma/ib_verbs.h>
8 #include <rdma/uverbs_types.h>
9 #include <rdma/uverbs_ioctl.h>
10 #include <rdma/uverbs_std_types.h>
11 #include <rdma/mlx5_user_ioctl_cmds.h>
12 #include <rdma/mlx5_user_ioctl_verbs.h>
13 #include <rdma/ib_hdrs.h>
14 #include <rdma/ib_umem.h>
15 #include <linux/mlx5/driver.h>
16 #include <linux/mlx5/fs.h>
17 #include <linux/mlx5/fs_helpers.h>
18 #include <linux/mlx5/eswitch.h>
19 #include <net/inet_ecn.h>
20 #include "mlx5_ib.h"
21 #include "counters.h"
22 #include "devx.h"
23 #include "fs.h"
24 
25 #define UVERBS_MODULE_NAME mlx5_ib
26 #include <rdma/uverbs_named_ioctl.h>
27 
28 enum {
29 	MATCH_CRITERIA_ENABLE_OUTER_BIT,
30 	MATCH_CRITERIA_ENABLE_MISC_BIT,
31 	MATCH_CRITERIA_ENABLE_INNER_BIT,
32 	MATCH_CRITERIA_ENABLE_MISC2_BIT
33 };
34 
35 #define HEADER_IS_ZERO(match_criteria, headers)			           \
36 	!(memchr_inv(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
37 		    0, MLX5_FLD_SZ_BYTES(fte_match_param, headers)))       \
38 
39 static u8 get_match_criteria_enable(u32 *match_criteria)
40 {
41 	u8 match_criteria_enable;
42 
43 	match_criteria_enable =
44 		(!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
45 		MATCH_CRITERIA_ENABLE_OUTER_BIT;
46 	match_criteria_enable |=
47 		(!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
48 		MATCH_CRITERIA_ENABLE_MISC_BIT;
49 	match_criteria_enable |=
50 		(!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
51 		MATCH_CRITERIA_ENABLE_INNER_BIT;
52 	match_criteria_enable |=
53 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
54 		MATCH_CRITERIA_ENABLE_MISC2_BIT;
55 
56 	return match_criteria_enable;
57 }
58 
59 static int set_proto(void *outer_c, void *outer_v, u8 mask, u8 val)
60 {
61 	u8 entry_mask;
62 	u8 entry_val;
63 	int err = 0;
64 
65 	if (!mask)
66 		goto out;
67 
68 	entry_mask = MLX5_GET(fte_match_set_lyr_2_4, outer_c,
69 			      ip_protocol);
70 	entry_val = MLX5_GET(fte_match_set_lyr_2_4, outer_v,
71 			     ip_protocol);
72 	if (!entry_mask) {
73 		MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_protocol, mask);
74 		MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_protocol, val);
75 		goto out;
76 	}
77 	/* Don't override existing ip protocol */
78 	if (mask != entry_mask || val != entry_val)
79 		err = -EINVAL;
80 out:
81 	return err;
82 }
83 
84 static void set_flow_label(void *misc_c, void *misc_v, u32 mask, u32 val,
85 			   bool inner)
86 {
87 	if (inner) {
88 		MLX5_SET(fte_match_set_misc,
89 			 misc_c, inner_ipv6_flow_label, mask);
90 		MLX5_SET(fte_match_set_misc,
91 			 misc_v, inner_ipv6_flow_label, val);
92 	} else {
93 		MLX5_SET(fte_match_set_misc,
94 			 misc_c, outer_ipv6_flow_label, mask);
95 		MLX5_SET(fte_match_set_misc,
96 			 misc_v, outer_ipv6_flow_label, val);
97 	}
98 }
99 
100 static void set_tos(void *outer_c, void *outer_v, u8 mask, u8 val)
101 {
102 	MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_ecn, mask);
103 	MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_ecn, val);
104 	MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_dscp, mask >> 2);
105 	MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_dscp, val >> 2);
106 }
107 
108 static int check_mpls_supp_fields(u32 field_support, const __be32 *set_mask)
109 {
110 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_label) &&
111 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_LABEL))
112 		return -EOPNOTSUPP;
113 
114 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_exp) &&
115 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_EXP))
116 		return -EOPNOTSUPP;
117 
118 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_s_bos) &&
119 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_S_BOS))
120 		return -EOPNOTSUPP;
121 
122 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_ttl) &&
123 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_TTL))
124 		return -EOPNOTSUPP;
125 
126 	return 0;
127 }
128 
129 #define LAST_ETH_FIELD vlan_tag
130 #define LAST_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_UNSPECIFIED:
150 		if (maction->flow_action_raw.sub_type ==
151 		    MLX5_IB_FLOW_ACTION_MODIFY_HEADER) {
152 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
153 				return -EINVAL;
154 			action->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
155 			action->modify_hdr =
156 				maction->flow_action_raw.modify_hdr;
157 			return 0;
158 		}
159 		if (maction->flow_action_raw.sub_type ==
160 		    MLX5_IB_FLOW_ACTION_DECAP) {
161 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_DECAP)
162 				return -EINVAL;
163 			action->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
164 			return 0;
165 		}
166 		if (maction->flow_action_raw.sub_type ==
167 		    MLX5_IB_FLOW_ACTION_PACKET_REFORMAT) {
168 			if (action->action &
169 			    MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT)
170 				return -EINVAL;
171 			action->action |=
172 				MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
173 			action->pkt_reformat =
174 				maction->flow_action_raw.pkt_reformat;
175 			return 0;
176 		}
177 		fallthrough;
178 	default:
179 		return -EOPNOTSUPP;
180 	}
181 }
182 
183 static int parse_flow_attr(struct mlx5_core_dev *mdev,
184 			   struct mlx5_flow_spec *spec,
185 			   const union ib_flow_spec *ib_spec,
186 			   const struct ib_flow_attr *flow_attr,
187 			   struct mlx5_flow_act *action, u32 prev_type)
188 {
189 	struct mlx5_flow_context *flow_context = &spec->flow_context;
190 	u32 *match_c = spec->match_criteria;
191 	u32 *match_v = spec->match_value;
192 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param, match_c,
193 					   misc_parameters);
194 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, match_v,
195 					   misc_parameters);
196 	void *misc_params2_c = MLX5_ADDR_OF(fte_match_param, match_c,
197 					    misc_parameters_2);
198 	void *misc_params2_v = MLX5_ADDR_OF(fte_match_param, match_v,
199 					    misc_parameters_2);
200 	void *headers_c;
201 	void *headers_v;
202 	int match_ipv;
203 	int ret;
204 
205 	if (ib_spec->type & IB_FLOW_SPEC_INNER) {
206 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
207 					 inner_headers);
208 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
209 					 inner_headers);
210 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
211 					ft_field_support.inner_ip_version);
212 	} else {
213 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
214 					 outer_headers);
215 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
216 					 outer_headers);
217 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
218 					ft_field_support.outer_ip_version);
219 	}
220 
221 	switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
222 	case IB_FLOW_SPEC_ETH:
223 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
224 			return -EOPNOTSUPP;
225 
226 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
227 					     dmac_47_16),
228 				ib_spec->eth.mask.dst_mac);
229 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
230 					     dmac_47_16),
231 				ib_spec->eth.val.dst_mac);
232 
233 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
234 					     smac_47_16),
235 				ib_spec->eth.mask.src_mac);
236 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
237 					     smac_47_16),
238 				ib_spec->eth.val.src_mac);
239 
240 		if (ib_spec->eth.mask.vlan_tag) {
241 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
242 				 cvlan_tag, 1);
243 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
244 				 cvlan_tag, 1);
245 
246 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
247 				 first_vid, ntohs(ib_spec->eth.mask.vlan_tag));
248 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
249 				 first_vid, ntohs(ib_spec->eth.val.vlan_tag));
250 
251 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
252 				 first_cfi,
253 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 12);
254 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
255 				 first_cfi,
256 				 ntohs(ib_spec->eth.val.vlan_tag) >> 12);
257 
258 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
259 				 first_prio,
260 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 13);
261 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
262 				 first_prio,
263 				 ntohs(ib_spec->eth.val.vlan_tag) >> 13);
264 		}
265 		MLX5_SET(fte_match_set_lyr_2_4, headers_c,
266 			 ethertype, ntohs(ib_spec->eth.mask.ether_type));
267 		MLX5_SET(fte_match_set_lyr_2_4, headers_v,
268 			 ethertype, ntohs(ib_spec->eth.val.ether_type));
269 		break;
270 	case IB_FLOW_SPEC_IPV4:
271 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
272 			return -EOPNOTSUPP;
273 
274 		if (match_ipv) {
275 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
276 				 ip_version, 0xf);
277 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
278 				 ip_version, MLX5_FS_IPV4_VERSION);
279 		} else {
280 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
281 				 ethertype, 0xffff);
282 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
283 				 ethertype, ETH_P_IP);
284 		}
285 
286 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
287 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
288 		       &ib_spec->ipv4.mask.src_ip,
289 		       sizeof(ib_spec->ipv4.mask.src_ip));
290 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
291 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
292 		       &ib_spec->ipv4.val.src_ip,
293 		       sizeof(ib_spec->ipv4.val.src_ip));
294 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
295 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
296 		       &ib_spec->ipv4.mask.dst_ip,
297 		       sizeof(ib_spec->ipv4.mask.dst_ip));
298 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
299 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
300 		       &ib_spec->ipv4.val.dst_ip,
301 		       sizeof(ib_spec->ipv4.val.dst_ip));
302 
303 		set_tos(headers_c, headers_v,
304 			ib_spec->ipv4.mask.tos, ib_spec->ipv4.val.tos);
305 
306 		if (set_proto(headers_c, headers_v,
307 			      ib_spec->ipv4.mask.proto,
308 			      ib_spec->ipv4.val.proto))
309 			return -EINVAL;
310 		break;
311 	case IB_FLOW_SPEC_IPV6:
312 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv6.mask, LAST_IPV6_FIELD))
313 			return -EOPNOTSUPP;
314 
315 		if (match_ipv) {
316 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
317 				 ip_version, 0xf);
318 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
319 				 ip_version, MLX5_FS_IPV6_VERSION);
320 		} else {
321 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
322 				 ethertype, 0xffff);
323 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
324 				 ethertype, ETH_P_IPV6);
325 		}
326 
327 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
328 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
329 		       &ib_spec->ipv6.mask.src_ip,
330 		       sizeof(ib_spec->ipv6.mask.src_ip));
331 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
332 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
333 		       &ib_spec->ipv6.val.src_ip,
334 		       sizeof(ib_spec->ipv6.val.src_ip));
335 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
336 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
337 		       &ib_spec->ipv6.mask.dst_ip,
338 		       sizeof(ib_spec->ipv6.mask.dst_ip));
339 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
340 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
341 		       &ib_spec->ipv6.val.dst_ip,
342 		       sizeof(ib_spec->ipv6.val.dst_ip));
343 
344 		set_tos(headers_c, headers_v,
345 			ib_spec->ipv6.mask.traffic_class,
346 			ib_spec->ipv6.val.traffic_class);
347 
348 		if (set_proto(headers_c, headers_v,
349 			      ib_spec->ipv6.mask.next_hdr,
350 			      ib_spec->ipv6.val.next_hdr))
351 			return -EINVAL;
352 
353 		set_flow_label(misc_params_c, misc_params_v,
354 			       ntohl(ib_spec->ipv6.mask.flow_label),
355 			       ntohl(ib_spec->ipv6.val.flow_label),
356 			       ib_spec->type & IB_FLOW_SPEC_INNER);
357 		break;
358 	case IB_FLOW_SPEC_ESP:
359 		return -EOPNOTSUPP;
360 	case IB_FLOW_SPEC_TCP:
361 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
362 					 LAST_TCP_UDP_FIELD))
363 			return -EOPNOTSUPP;
364 
365 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_TCP))
366 			return -EINVAL;
367 
368 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_sport,
369 			 ntohs(ib_spec->tcp_udp.mask.src_port));
370 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
371 			 ntohs(ib_spec->tcp_udp.val.src_port));
372 
373 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_dport,
374 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
375 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
376 			 ntohs(ib_spec->tcp_udp.val.dst_port));
377 		break;
378 	case IB_FLOW_SPEC_UDP:
379 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
380 					 LAST_TCP_UDP_FIELD))
381 			return -EOPNOTSUPP;
382 
383 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_UDP))
384 			return -EINVAL;
385 
386 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_sport,
387 			 ntohs(ib_spec->tcp_udp.mask.src_port));
388 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
389 			 ntohs(ib_spec->tcp_udp.val.src_port));
390 
391 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_dport,
392 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
393 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
394 			 ntohs(ib_spec->tcp_udp.val.dst_port));
395 		break;
396 	case IB_FLOW_SPEC_GRE:
397 		if (ib_spec->gre.mask.c_ks_res0_ver)
398 			return -EOPNOTSUPP;
399 
400 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_GRE))
401 			return -EINVAL;
402 
403 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
404 			 0xff);
405 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
406 			 IPPROTO_GRE);
407 
408 		MLX5_SET(fte_match_set_misc, misc_params_c, gre_protocol,
409 			 ntohs(ib_spec->gre.mask.protocol));
410 		MLX5_SET(fte_match_set_misc, misc_params_v, gre_protocol,
411 			 ntohs(ib_spec->gre.val.protocol));
412 
413 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_c,
414 				    gre_key.nvgre.hi),
415 		       &ib_spec->gre.mask.key,
416 		       sizeof(ib_spec->gre.mask.key));
417 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_v,
418 				    gre_key.nvgre.hi),
419 		       &ib_spec->gre.val.key,
420 		       sizeof(ib_spec->gre.val.key));
421 		break;
422 	case IB_FLOW_SPEC_MPLS:
423 		switch (prev_type) {
424 		case IB_FLOW_SPEC_UDP:
425 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
426 						   ft_field_support.outer_first_mpls_over_udp),
427 						   &ib_spec->mpls.mask.tag))
428 				return -EOPNOTSUPP;
429 
430 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
431 					    outer_first_mpls_over_udp),
432 			       &ib_spec->mpls.val.tag,
433 			       sizeof(ib_spec->mpls.val.tag));
434 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
435 					    outer_first_mpls_over_udp),
436 			       &ib_spec->mpls.mask.tag,
437 			       sizeof(ib_spec->mpls.mask.tag));
438 			break;
439 		case IB_FLOW_SPEC_GRE:
440 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
441 						   ft_field_support.outer_first_mpls_over_gre),
442 						   &ib_spec->mpls.mask.tag))
443 				return -EOPNOTSUPP;
444 
445 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
446 					    outer_first_mpls_over_gre),
447 			       &ib_spec->mpls.val.tag,
448 			       sizeof(ib_spec->mpls.val.tag));
449 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
450 					    outer_first_mpls_over_gre),
451 			       &ib_spec->mpls.mask.tag,
452 			       sizeof(ib_spec->mpls.mask.tag));
453 			break;
454 		default:
455 			if (ib_spec->type & IB_FLOW_SPEC_INNER) {
456 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
457 							   ft_field_support.inner_first_mpls),
458 							   &ib_spec->mpls.mask.tag))
459 					return -EOPNOTSUPP;
460 
461 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
462 						    inner_first_mpls),
463 				       &ib_spec->mpls.val.tag,
464 				       sizeof(ib_spec->mpls.val.tag));
465 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
466 						    inner_first_mpls),
467 				       &ib_spec->mpls.mask.tag,
468 				       sizeof(ib_spec->mpls.mask.tag));
469 			} else {
470 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
471 							   ft_field_support.outer_first_mpls),
472 							   &ib_spec->mpls.mask.tag))
473 					return -EOPNOTSUPP;
474 
475 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
476 						    outer_first_mpls),
477 				       &ib_spec->mpls.val.tag,
478 				       sizeof(ib_spec->mpls.val.tag));
479 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
480 						    outer_first_mpls),
481 				       &ib_spec->mpls.mask.tag,
482 				       sizeof(ib_spec->mpls.mask.tag));
483 			}
484 		}
485 		break;
486 	case IB_FLOW_SPEC_VXLAN_TUNNEL:
487 		if (FIELDS_NOT_SUPPORTED(ib_spec->tunnel.mask,
488 					 LAST_TUNNEL_FIELD))
489 			return -EOPNOTSUPP;
490 
491 		MLX5_SET(fte_match_set_misc, misc_params_c, vxlan_vni,
492 			 ntohl(ib_spec->tunnel.mask.tunnel_id));
493 		MLX5_SET(fte_match_set_misc, misc_params_v, vxlan_vni,
494 			 ntohl(ib_spec->tunnel.val.tunnel_id));
495 		break;
496 	case IB_FLOW_SPEC_ACTION_TAG:
497 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_tag,
498 					 LAST_FLOW_TAG_FIELD))
499 			return -EOPNOTSUPP;
500 		if (ib_spec->flow_tag.tag_id >= BIT(24))
501 			return -EINVAL;
502 
503 		flow_context->flow_tag = ib_spec->flow_tag.tag_id;
504 		flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
505 		break;
506 	case IB_FLOW_SPEC_ACTION_DROP:
507 		if (FIELDS_NOT_SUPPORTED(ib_spec->drop,
508 					 LAST_DROP_FIELD))
509 			return -EOPNOTSUPP;
510 		action->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
511 		break;
512 	case IB_FLOW_SPEC_ACTION_HANDLE:
513 		ret = parse_flow_flow_action(to_mflow_act(ib_spec->action.act),
514 			flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS, action);
515 		if (ret)
516 			return ret;
517 		break;
518 	case IB_FLOW_SPEC_ACTION_COUNT:
519 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_count,
520 					 LAST_COUNTERS_FIELD))
521 			return -EOPNOTSUPP;
522 
523 		/* for now support only one counters spec per flow */
524 		if (action->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
525 			return -EINVAL;
526 
527 		action->counters = ib_spec->flow_count.counters;
528 		action->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
529 		break;
530 	default:
531 		return -EINVAL;
532 	}
533 
534 	return 0;
535 }
536 
537 /* If a flow could catch both multicast and unicast packets,
538  * it won't fall into the multicast flow steering table and this rule
539  * could steal other multicast packets.
540  */
541 static bool flow_is_multicast_only(const struct ib_flow_attr *ib_attr)
542 {
543 	union ib_flow_spec *flow_spec;
544 
545 	if (ib_attr->type != IB_FLOW_ATTR_NORMAL ||
546 	    ib_attr->num_of_specs < 1)
547 		return false;
548 
549 	flow_spec = (union ib_flow_spec *)(ib_attr + 1);
550 	if (flow_spec->type == IB_FLOW_SPEC_IPV4) {
551 		struct ib_flow_spec_ipv4 *ipv4_spec;
552 
553 		ipv4_spec = (struct ib_flow_spec_ipv4 *)flow_spec;
554 		if (ipv4_is_multicast(ipv4_spec->val.dst_ip))
555 			return true;
556 
557 		return false;
558 	}
559 
560 	if (flow_spec->type == IB_FLOW_SPEC_ETH) {
561 		struct ib_flow_spec_eth *eth_spec;
562 
563 		eth_spec = (struct ib_flow_spec_eth *)flow_spec;
564 		return is_multicast_ether_addr(eth_spec->mask.dst_mac) &&
565 		       is_multicast_ether_addr(eth_spec->val.dst_mac);
566 	}
567 
568 	return false;
569 }
570 
571 static bool is_valid_ethertype(struct mlx5_core_dev *mdev,
572 			       const struct ib_flow_attr *flow_attr,
573 			       bool check_inner)
574 {
575 	union ib_flow_spec *ib_spec = (union ib_flow_spec *)(flow_attr + 1);
576 	int match_ipv = check_inner ?
577 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
578 					ft_field_support.inner_ip_version) :
579 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
580 					ft_field_support.outer_ip_version);
581 	int inner_bit = check_inner ? IB_FLOW_SPEC_INNER : 0;
582 	bool ipv4_spec_valid, ipv6_spec_valid;
583 	unsigned int ip_spec_type = 0;
584 	bool has_ethertype = false;
585 	unsigned int spec_index;
586 	bool mask_valid = true;
587 	u16 eth_type = 0;
588 	bool type_valid;
589 
590 	/* Validate that ethertype is correct */
591 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
592 		if ((ib_spec->type == (IB_FLOW_SPEC_ETH | inner_bit)) &&
593 		    ib_spec->eth.mask.ether_type) {
594 			mask_valid = (ib_spec->eth.mask.ether_type ==
595 				      htons(0xffff));
596 			has_ethertype = true;
597 			eth_type = ntohs(ib_spec->eth.val.ether_type);
598 		} else if ((ib_spec->type == (IB_FLOW_SPEC_IPV4 | inner_bit)) ||
599 			   (ib_spec->type == (IB_FLOW_SPEC_IPV6 | inner_bit))) {
600 			ip_spec_type = ib_spec->type;
601 		}
602 		ib_spec = (void *)ib_spec + ib_spec->size;
603 	}
604 
605 	type_valid = (!has_ethertype) || (!ip_spec_type);
606 	if (!type_valid && mask_valid) {
607 		ipv4_spec_valid = (eth_type == ETH_P_IP) &&
608 			(ip_spec_type == (IB_FLOW_SPEC_IPV4 | inner_bit));
609 		ipv6_spec_valid = (eth_type == ETH_P_IPV6) &&
610 			(ip_spec_type == (IB_FLOW_SPEC_IPV6 | inner_bit));
611 
612 		type_valid = (ipv4_spec_valid) || (ipv6_spec_valid) ||
613 			     (((eth_type == ETH_P_MPLS_UC) ||
614 			       (eth_type == ETH_P_MPLS_MC)) && match_ipv);
615 	}
616 
617 	return type_valid;
618 }
619 
620 static bool is_valid_attr(struct mlx5_core_dev *mdev,
621 			  const struct ib_flow_attr *flow_attr)
622 {
623 	return is_valid_ethertype(mdev, flow_attr, false) &&
624 	       is_valid_ethertype(mdev, flow_attr, true);
625 }
626 
627 static void put_flow_table(struct mlx5_ib_dev *dev,
628 			   struct mlx5_ib_flow_prio *prio, bool ft_added)
629 {
630 	prio->refcount -= !!ft_added;
631 	if (!prio->refcount) {
632 		mlx5_destroy_flow_table(prio->flow_table);
633 		prio->flow_table = NULL;
634 	}
635 }
636 
637 static int mlx5_ib_destroy_flow(struct ib_flow *flow_id)
638 {
639 	struct mlx5_ib_flow_handler *handler = container_of(flow_id,
640 							  struct mlx5_ib_flow_handler,
641 							  ibflow);
642 	struct mlx5_ib_flow_handler *iter, *tmp;
643 	struct mlx5_ib_dev *dev = handler->dev;
644 
645 	mutex_lock(&dev->flow_db->lock);
646 
647 	list_for_each_entry_safe(iter, tmp, &handler->list, list) {
648 		mlx5_del_flow_rules(iter->rule);
649 		put_flow_table(dev, iter->prio, true);
650 		list_del(&iter->list);
651 		kfree(iter);
652 	}
653 
654 	mlx5_del_flow_rules(handler->rule);
655 	put_flow_table(dev, handler->prio, true);
656 	mlx5_ib_counters_clear_description(handler->ibcounters);
657 	mutex_unlock(&dev->flow_db->lock);
658 	if (handler->flow_matcher)
659 		atomic_dec(&handler->flow_matcher->usecnt);
660 	kfree(handler);
661 
662 	return 0;
663 }
664 
665 static int ib_prio_to_core_prio(unsigned int priority, bool dont_trap)
666 {
667 	priority *= 2;
668 	if (!dont_trap)
669 		priority++;
670 	return priority;
671 }
672 
673 enum flow_table_type {
674 	MLX5_IB_FT_RX,
675 	MLX5_IB_FT_TX
676 };
677 
678 #define MLX5_FS_MAX_TYPES	 6
679 #define MLX5_FS_MAX_ENTRIES	 BIT(16)
680 
681 static bool mlx5_ib_shared_ft_allowed(struct ib_device *device)
682 {
683 	struct mlx5_ib_dev *dev = to_mdev(device);
684 
685 	return MLX5_CAP_GEN(dev->mdev, shared_object_to_user_object_allowed);
686 }
687 
688 static struct mlx5_ib_flow_prio *_get_prio(struct mlx5_ib_dev *dev,
689 					   struct mlx5_flow_namespace *ns,
690 					   struct mlx5_ib_flow_prio *prio,
691 					   int priority,
692 					   int num_entries, int num_groups,
693 					   u32 flags)
694 {
695 	struct mlx5_flow_table_attr ft_attr = {};
696 	struct mlx5_flow_table *ft;
697 
698 	ft_attr.prio = priority;
699 	ft_attr.max_fte = num_entries;
700 	ft_attr.flags = flags;
701 	ft_attr.autogroup.max_num_groups = num_groups;
702 	ft = mlx5_create_auto_grouped_flow_table(ns, &ft_attr);
703 	if (IS_ERR(ft))
704 		return ERR_CAST(ft);
705 
706 	prio->flow_table = ft;
707 	prio->refcount = 0;
708 	return prio;
709 }
710 
711 static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev,
712 						struct ib_flow_attr *flow_attr,
713 						enum flow_table_type ft_type)
714 {
715 	bool dont_trap = flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP;
716 	struct mlx5_flow_namespace *ns = NULL;
717 	enum mlx5_flow_namespace_type fn_type;
718 	struct mlx5_ib_flow_prio *prio;
719 	struct mlx5_flow_table *ft;
720 	int max_table_size;
721 	int num_entries;
722 	int num_groups;
723 	bool esw_encap;
724 	u32 flags = 0;
725 	int priority;
726 
727 	max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
728 						       log_max_ft_size));
729 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
730 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
731 	switch (flow_attr->type) {
732 	case IB_FLOW_ATTR_NORMAL:
733 		if (flow_is_multicast_only(flow_attr) && !dont_trap)
734 			priority = MLX5_IB_FLOW_MCAST_PRIO;
735 		else
736 			priority = ib_prio_to_core_prio(flow_attr->priority,
737 							dont_trap);
738 		if (ft_type == MLX5_IB_FT_RX) {
739 			fn_type = MLX5_FLOW_NAMESPACE_BYPASS;
740 			prio = &dev->flow_db->prios[priority];
741 			if (!dev->is_rep && !esw_encap &&
742 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap))
743 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
744 			if (!dev->is_rep && !esw_encap &&
745 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
746 						      reformat_l3_tunnel_to_l2))
747 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
748 		} else {
749 			max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_TX(
750 				dev->mdev, log_max_ft_size));
751 			fn_type = MLX5_FLOW_NAMESPACE_EGRESS;
752 			prio = &dev->flow_db->egress_prios[priority];
753 			if (!dev->is_rep && !esw_encap &&
754 			    MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat))
755 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
756 		}
757 		ns = mlx5_get_flow_namespace(dev->mdev, fn_type);
758 		num_entries = MLX5_FS_MAX_ENTRIES;
759 		num_groups = MLX5_FS_MAX_TYPES;
760 		break;
761 	case IB_FLOW_ATTR_ALL_DEFAULT:
762 	case IB_FLOW_ATTR_MC_DEFAULT:
763 		ns = mlx5_get_flow_namespace(dev->mdev,
764 					     MLX5_FLOW_NAMESPACE_LEFTOVERS);
765 		build_leftovers_ft_param(&priority, &num_entries, &num_groups);
766 		prio = &dev->flow_db->prios[MLX5_IB_FLOW_LEFTOVERS_PRIO];
767 		break;
768 	case IB_FLOW_ATTR_SNIFFER:
769 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
770 					allow_sniffer_and_nic_rx_shared_tir))
771 			return ERR_PTR(-EOPNOTSUPP);
772 
773 		ns = mlx5_get_flow_namespace(
774 			dev->mdev, ft_type == MLX5_IB_FT_RX ?
775 					   MLX5_FLOW_NAMESPACE_SNIFFER_RX :
776 					   MLX5_FLOW_NAMESPACE_SNIFFER_TX);
777 
778 		prio = &dev->flow_db->sniffer[ft_type];
779 		priority = 0;
780 		num_entries = 1;
781 		num_groups = 1;
782 		break;
783 	default:
784 		break;
785 	}
786 
787 	if (!ns)
788 		return ERR_PTR(-EOPNOTSUPP);
789 
790 	max_table_size = min_t(int, num_entries, max_table_size);
791 
792 	ft = prio->flow_table;
793 	if (!ft)
794 		return _get_prio(dev, ns, prio, priority, max_table_size,
795 				 num_groups, flags);
796 
797 	return prio;
798 }
799 
800 enum {
801 	RDMA_RX_ECN_OPCOUNTER_PRIO,
802 	RDMA_RX_CNP_OPCOUNTER_PRIO,
803 };
804 
805 enum {
806 	RDMA_TX_CNP_OPCOUNTER_PRIO,
807 };
808 
809 static int set_vhca_port_spec(struct mlx5_ib_dev *dev, u32 port_num,
810 			      struct mlx5_flow_spec *spec)
811 {
812 	if (!MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev,
813 					ft_field_support.source_vhca_port) ||
814 	    !MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
815 					ft_field_support.source_vhca_port))
816 		return -EOPNOTSUPP;
817 
818 	MLX5_SET_TO_ONES(fte_match_param, &spec->match_criteria,
819 			 misc_parameters.source_vhca_port);
820 	MLX5_SET(fte_match_param, &spec->match_value,
821 		 misc_parameters.source_vhca_port, port_num);
822 
823 	return 0;
824 }
825 
826 static int set_ecn_ce_spec(struct mlx5_ib_dev *dev, u32 port_num,
827 			   struct mlx5_flow_spec *spec, int ipv)
828 {
829 	if (!MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev,
830 					ft_field_support.outer_ip_version))
831 		return -EOPNOTSUPP;
832 
833 	if (mlx5_core_mp_enabled(dev->mdev) &&
834 	    set_vhca_port_spec(dev, port_num, spec))
835 		return -EOPNOTSUPP;
836 
837 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
838 			 outer_headers.ip_ecn);
839 	MLX5_SET(fte_match_param, spec->match_value, outer_headers.ip_ecn,
840 		 INET_ECN_CE);
841 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
842 			 outer_headers.ip_version);
843 	MLX5_SET(fte_match_param, spec->match_value, outer_headers.ip_version,
844 		 ipv);
845 
846 	spec->match_criteria_enable =
847 		get_match_criteria_enable(spec->match_criteria);
848 
849 	return 0;
850 }
851 
852 static int set_cnp_spec(struct mlx5_ib_dev *dev, u32 port_num,
853 			struct mlx5_flow_spec *spec)
854 {
855 	if (mlx5_core_mp_enabled(dev->mdev) &&
856 	    set_vhca_port_spec(dev, port_num, spec))
857 		return -EOPNOTSUPP;
858 
859 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
860 			 misc_parameters.bth_opcode);
861 	MLX5_SET(fte_match_param, spec->match_value, misc_parameters.bth_opcode,
862 		 IB_BTH_OPCODE_CNP);
863 
864 	spec->match_criteria_enable =
865 		get_match_criteria_enable(spec->match_criteria);
866 
867 	return 0;
868 }
869 
870 int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num,
871 			 struct mlx5_ib_op_fc *opfc,
872 			 enum mlx5_ib_optional_counter_type type)
873 {
874 	enum mlx5_flow_namespace_type fn_type;
875 	int priority, i, err, spec_num;
876 	struct mlx5_flow_act flow_act = {};
877 	struct mlx5_flow_destination dst;
878 	struct mlx5_flow_namespace *ns;
879 	struct mlx5_ib_flow_prio *prio;
880 	struct mlx5_flow_spec *spec;
881 
882 	spec = kcalloc(MAX_OPFC_RULES, sizeof(*spec), GFP_KERNEL);
883 	if (!spec)
884 		return -ENOMEM;
885 
886 	switch (type) {
887 	case MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS:
888 		if (set_ecn_ce_spec(dev, port_num, &spec[0],
889 				    MLX5_FS_IPV4_VERSION) ||
890 		    set_ecn_ce_spec(dev, port_num, &spec[1],
891 				    MLX5_FS_IPV6_VERSION)) {
892 			err = -EOPNOTSUPP;
893 			goto free;
894 		}
895 		spec_num = 2;
896 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS;
897 		priority = RDMA_RX_ECN_OPCOUNTER_PRIO;
898 		break;
899 
900 	case MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS:
901 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
902 					ft_field_support_2_nic_receive_rdma.bth_opcode) ||
903 		    set_cnp_spec(dev, port_num, &spec[0])) {
904 			err = -EOPNOTSUPP;
905 			goto free;
906 		}
907 		spec_num = 1;
908 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS;
909 		priority = RDMA_RX_CNP_OPCOUNTER_PRIO;
910 		break;
911 
912 	case MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS:
913 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
914 					ft_field_support_2_nic_transmit_rdma.bth_opcode) ||
915 		    set_cnp_spec(dev, port_num, &spec[0])) {
916 			err = -EOPNOTSUPP;
917 			goto free;
918 		}
919 		spec_num = 1;
920 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS;
921 		priority = RDMA_TX_CNP_OPCOUNTER_PRIO;
922 		break;
923 
924 	default:
925 		err = -EOPNOTSUPP;
926 		goto free;
927 	}
928 
929 	ns = mlx5_get_flow_namespace(dev->mdev, fn_type);
930 	if (!ns) {
931 		err = -EOPNOTSUPP;
932 		goto free;
933 	}
934 
935 	prio = &dev->flow_db->opfcs[type];
936 	if (!prio->flow_table) {
937 		prio = _get_prio(dev, ns, prio, priority,
938 				 dev->num_ports * MAX_OPFC_RULES, 1, 0);
939 		if (IS_ERR(prio)) {
940 			err = PTR_ERR(prio);
941 			goto free;
942 		}
943 	}
944 
945 	dst.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
946 	dst.counter_id = mlx5_fc_id(opfc->fc);
947 
948 	flow_act.action =
949 		MLX5_FLOW_CONTEXT_ACTION_COUNT | MLX5_FLOW_CONTEXT_ACTION_ALLOW;
950 
951 	for (i = 0; i < spec_num; i++) {
952 		opfc->rule[i] = mlx5_add_flow_rules(prio->flow_table, &spec[i],
953 						    &flow_act, &dst, 1);
954 		if (IS_ERR(opfc->rule[i])) {
955 			err = PTR_ERR(opfc->rule[i]);
956 			goto del_rules;
957 		}
958 	}
959 	prio->refcount += spec_num;
960 	kfree(spec);
961 
962 	return 0;
963 
964 del_rules:
965 	for (i -= 1; i >= 0; i--)
966 		mlx5_del_flow_rules(opfc->rule[i]);
967 	put_flow_table(dev, prio, false);
968 free:
969 	kfree(spec);
970 	return err;
971 }
972 
973 void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev,
974 			     struct mlx5_ib_op_fc *opfc,
975 			     enum mlx5_ib_optional_counter_type type)
976 {
977 	int i;
978 
979 	for (i = 0; i < MAX_OPFC_RULES && opfc->rule[i]; i++) {
980 		mlx5_del_flow_rules(opfc->rule[i]);
981 		put_flow_table(dev, &dev->flow_db->opfcs[type], true);
982 	}
983 }
984 
985 static void set_underlay_qp(struct mlx5_ib_dev *dev,
986 			    struct mlx5_flow_spec *spec,
987 			    u32 underlay_qpn)
988 {
989 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param,
990 					   spec->match_criteria,
991 					   misc_parameters);
992 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
993 					   misc_parameters);
994 
995 	if (underlay_qpn &&
996 	    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
997 				      ft_field_support.bth_dst_qp)) {
998 		MLX5_SET(fte_match_set_misc,
999 			 misc_params_v, bth_dst_qp, underlay_qpn);
1000 		MLX5_SET(fte_match_set_misc,
1001 			 misc_params_c, bth_dst_qp, 0xffffff);
1002 	}
1003 }
1004 
1005 static void mlx5_ib_set_rule_source_port(struct mlx5_ib_dev *dev,
1006 					 struct mlx5_flow_spec *spec,
1007 					 struct mlx5_eswitch_rep *rep)
1008 {
1009 	struct mlx5_eswitch *esw = dev->mdev->priv.eswitch;
1010 	void *misc;
1011 
1012 	if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1013 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1014 				    misc_parameters_2);
1015 
1016 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1017 			 mlx5_eswitch_get_vport_metadata_for_match(rep->esw,
1018 								   rep->vport));
1019 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1020 				    misc_parameters_2);
1021 
1022 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1023 			 mlx5_eswitch_get_vport_metadata_mask());
1024 	} else {
1025 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1026 				    misc_parameters);
1027 
1028 		MLX5_SET(fte_match_set_misc, misc, source_port, rep->vport);
1029 
1030 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1031 				    misc_parameters);
1032 
1033 		MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
1034 	}
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 						      u32 underlay_qpn,
1042 						      struct mlx5_ib_create_flow *ucmd)
1043 {
1044 	struct mlx5_flow_table	*ft = ft_prio->flow_table;
1045 	struct mlx5_ib_flow_handler *handler;
1046 	struct mlx5_flow_act flow_act = {};
1047 	struct mlx5_flow_spec *spec;
1048 	struct mlx5_flow_destination dest_arr[2] = {};
1049 	struct mlx5_flow_destination *rule_dst = dest_arr;
1050 	const void *ib_flow = (const void *)flow_attr + sizeof(*flow_attr);
1051 	unsigned int spec_index;
1052 	u32 prev_type = 0;
1053 	int err = 0;
1054 	int dest_num = 0;
1055 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
1056 
1057 	if (!is_valid_attr(dev->mdev, flow_attr))
1058 		return ERR_PTR(-EINVAL);
1059 
1060 	if (dev->is_rep && is_egress)
1061 		return ERR_PTR(-EINVAL);
1062 
1063 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1064 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
1065 	if (!handler || !spec) {
1066 		err = -ENOMEM;
1067 		goto free;
1068 	}
1069 
1070 	INIT_LIST_HEAD(&handler->list);
1071 
1072 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
1073 		err = parse_flow_attr(dev->mdev, spec,
1074 				      ib_flow, flow_attr, &flow_act,
1075 				      prev_type);
1076 		if (err < 0)
1077 			goto free;
1078 
1079 		prev_type = ((union ib_flow_spec *)ib_flow)->type;
1080 		ib_flow += ((union ib_flow_spec *)ib_flow)->size;
1081 	}
1082 
1083 	if (dst && !(flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP)) {
1084 		memcpy(&dest_arr[0], dst, sizeof(*dst));
1085 		dest_num++;
1086 	}
1087 
1088 	if (!flow_is_multicast_only(flow_attr))
1089 		set_underlay_qp(dev, spec, underlay_qpn);
1090 
1091 	if (dev->is_rep && flow_attr->type != IB_FLOW_ATTR_SNIFFER) {
1092 		struct mlx5_eswitch_rep *rep;
1093 
1094 		rep = dev->port[flow_attr->port - 1].rep;
1095 		if (!rep) {
1096 			err = -EINVAL;
1097 			goto free;
1098 		}
1099 
1100 		mlx5_ib_set_rule_source_port(dev, spec, rep);
1101 	}
1102 
1103 	spec->match_criteria_enable = get_match_criteria_enable(spec->match_criteria);
1104 
1105 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1106 		struct mlx5_ib_mcounters *mcounters;
1107 
1108 		err = mlx5_ib_flow_counters_set_data(flow_act.counters, ucmd);
1109 		if (err)
1110 			goto free;
1111 
1112 		mcounters = to_mcounters(flow_act.counters);
1113 		handler->ibcounters = flow_act.counters;
1114 		dest_arr[dest_num].type =
1115 			MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1116 		dest_arr[dest_num].counter_id =
1117 			mlx5_fc_id(mcounters->hw_cntrs_hndl);
1118 		dest_num++;
1119 	}
1120 
1121 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP) {
1122 		if (!dest_num)
1123 			rule_dst = NULL;
1124 	} else {
1125 		if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1126 			flow_act.action |=
1127 				MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1128 		if (is_egress)
1129 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1130 		else if (dest_num)
1131 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1132 	}
1133 
1134 	if ((spec->flow_context.flags & FLOW_CONTEXT_HAS_TAG)  &&
1135 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1136 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1137 		mlx5_ib_warn(dev, "Flow tag %u and attribute type %x isn't allowed in leftovers\n",
1138 			     spec->flow_context.flow_tag, flow_attr->type);
1139 		err = -EINVAL;
1140 		goto free;
1141 	}
1142 	handler->rule = mlx5_add_flow_rules(ft, spec,
1143 					    &flow_act,
1144 					    rule_dst, dest_num);
1145 
1146 	if (IS_ERR(handler->rule)) {
1147 		err = PTR_ERR(handler->rule);
1148 		goto free;
1149 	}
1150 
1151 	ft_prio->refcount++;
1152 	handler->prio = ft_prio;
1153 	handler->dev = dev;
1154 
1155 	ft_prio->flow_table = ft;
1156 free:
1157 	if (err && handler) {
1158 		mlx5_ib_counters_clear_description(handler->ibcounters);
1159 		kfree(handler);
1160 	}
1161 	kvfree(spec);
1162 	return err ? ERR_PTR(err) : handler;
1163 }
1164 
1165 static struct mlx5_ib_flow_handler *create_flow_rule(struct mlx5_ib_dev *dev,
1166 						     struct mlx5_ib_flow_prio *ft_prio,
1167 						     const struct ib_flow_attr *flow_attr,
1168 						     struct mlx5_flow_destination *dst)
1169 {
1170 	return _create_flow_rule(dev, ft_prio, flow_attr, dst, 0, NULL);
1171 }
1172 
1173 enum {
1174 	LEFTOVERS_MC,
1175 	LEFTOVERS_UC,
1176 };
1177 
1178 static struct mlx5_ib_flow_handler *create_leftovers_rule(struct mlx5_ib_dev *dev,
1179 							  struct mlx5_ib_flow_prio *ft_prio,
1180 							  struct ib_flow_attr *flow_attr,
1181 							  struct mlx5_flow_destination *dst)
1182 {
1183 	struct mlx5_ib_flow_handler *handler_ucast = NULL;
1184 	struct mlx5_ib_flow_handler *handler = NULL;
1185 
1186 	static struct {
1187 		struct ib_flow_attr	flow_attr;
1188 		struct ib_flow_spec_eth eth_flow;
1189 	} leftovers_specs[] = {
1190 		[LEFTOVERS_MC] = {
1191 			.flow_attr = {
1192 				.num_of_specs = 1,
1193 				.size = sizeof(leftovers_specs[0])
1194 			},
1195 			.eth_flow = {
1196 				.type = IB_FLOW_SPEC_ETH,
1197 				.size = sizeof(struct ib_flow_spec_eth),
1198 				.mask = {.dst_mac = {0x1} },
1199 				.val =  {.dst_mac = {0x1} }
1200 			}
1201 		},
1202 		[LEFTOVERS_UC] = {
1203 			.flow_attr = {
1204 				.num_of_specs = 1,
1205 				.size = sizeof(leftovers_specs[0])
1206 			},
1207 			.eth_flow = {
1208 				.type = IB_FLOW_SPEC_ETH,
1209 				.size = sizeof(struct ib_flow_spec_eth),
1210 				.mask = {.dst_mac = {0x1} },
1211 				.val = {.dst_mac = {} }
1212 			}
1213 		}
1214 	};
1215 
1216 	handler = create_flow_rule(dev, ft_prio,
1217 				   &leftovers_specs[LEFTOVERS_MC].flow_attr,
1218 				   dst);
1219 	if (!IS_ERR(handler) &&
1220 	    flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT) {
1221 		handler_ucast = create_flow_rule(dev, ft_prio,
1222 						 &leftovers_specs[LEFTOVERS_UC].flow_attr,
1223 						 dst);
1224 		if (IS_ERR(handler_ucast)) {
1225 			mlx5_del_flow_rules(handler->rule);
1226 			ft_prio->refcount--;
1227 			kfree(handler);
1228 			handler = handler_ucast;
1229 		} else {
1230 			list_add(&handler_ucast->list, &handler->list);
1231 		}
1232 	}
1233 
1234 	return handler;
1235 }
1236 
1237 static struct mlx5_ib_flow_handler *create_sniffer_rule(struct mlx5_ib_dev *dev,
1238 							struct mlx5_ib_flow_prio *ft_rx,
1239 							struct mlx5_ib_flow_prio *ft_tx,
1240 							struct mlx5_flow_destination *dst)
1241 {
1242 	struct mlx5_ib_flow_handler *handler_rx;
1243 	struct mlx5_ib_flow_handler *handler_tx;
1244 	int err;
1245 	static const struct ib_flow_attr flow_attr  = {
1246 		.num_of_specs = 0,
1247 		.type = IB_FLOW_ATTR_SNIFFER,
1248 		.size = sizeof(flow_attr)
1249 	};
1250 
1251 	handler_rx = create_flow_rule(dev, ft_rx, &flow_attr, dst);
1252 	if (IS_ERR(handler_rx)) {
1253 		err = PTR_ERR(handler_rx);
1254 		goto err;
1255 	}
1256 
1257 	handler_tx = create_flow_rule(dev, ft_tx, &flow_attr, dst);
1258 	if (IS_ERR(handler_tx)) {
1259 		err = PTR_ERR(handler_tx);
1260 		goto err_tx;
1261 	}
1262 
1263 	list_add(&handler_tx->list, &handler_rx->list);
1264 
1265 	return handler_rx;
1266 
1267 err_tx:
1268 	mlx5_del_flow_rules(handler_rx->rule);
1269 	ft_rx->refcount--;
1270 	kfree(handler_rx);
1271 err:
1272 	return ERR_PTR(err);
1273 }
1274 
1275 static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
1276 					   struct ib_flow_attr *flow_attr,
1277 					   struct ib_udata *udata)
1278 {
1279 	struct mlx5_ib_dev *dev = to_mdev(qp->device);
1280 	struct mlx5_ib_qp *mqp = to_mqp(qp);
1281 	struct mlx5_ib_flow_handler *handler = NULL;
1282 	struct mlx5_flow_destination *dst = NULL;
1283 	struct mlx5_ib_flow_prio *ft_prio_tx = NULL;
1284 	struct mlx5_ib_flow_prio *ft_prio;
1285 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
1286 	struct mlx5_ib_create_flow *ucmd = NULL, ucmd_hdr;
1287 	size_t min_ucmd_sz, required_ucmd_sz;
1288 	int err;
1289 	int underlay_qpn;
1290 
1291 	if (udata && udata->inlen) {
1292 		min_ucmd_sz = offsetofend(struct mlx5_ib_create_flow, reserved);
1293 		if (udata->inlen < min_ucmd_sz)
1294 			return ERR_PTR(-EOPNOTSUPP);
1295 
1296 		err = ib_copy_from_udata(&ucmd_hdr, udata, min_ucmd_sz);
1297 		if (err)
1298 			return ERR_PTR(err);
1299 
1300 		/* currently supports only one counters data */
1301 		if (ucmd_hdr.ncounters_data > 1)
1302 			return ERR_PTR(-EINVAL);
1303 
1304 		required_ucmd_sz = min_ucmd_sz +
1305 			sizeof(struct mlx5_ib_flow_counters_data) *
1306 			ucmd_hdr.ncounters_data;
1307 		if (udata->inlen > required_ucmd_sz &&
1308 		    !ib_is_udata_cleared(udata, required_ucmd_sz,
1309 					 udata->inlen - required_ucmd_sz))
1310 			return ERR_PTR(-EOPNOTSUPP);
1311 
1312 		ucmd = kzalloc(required_ucmd_sz, GFP_KERNEL);
1313 		if (!ucmd)
1314 			return ERR_PTR(-ENOMEM);
1315 
1316 		err = ib_copy_from_udata(ucmd, udata, required_ucmd_sz);
1317 		if (err)
1318 			goto free_ucmd;
1319 	}
1320 
1321 	if (flow_attr->priority > MLX5_IB_FLOW_LAST_PRIO) {
1322 		err = -ENOMEM;
1323 		goto free_ucmd;
1324 	}
1325 
1326 	if (flow_attr->flags &
1327 	    ~(IB_FLOW_ATTR_FLAGS_DONT_TRAP | IB_FLOW_ATTR_FLAGS_EGRESS)) {
1328 		err = -EINVAL;
1329 		goto free_ucmd;
1330 	}
1331 
1332 	if (is_egress &&
1333 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1334 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1335 		err = -EINVAL;
1336 		goto free_ucmd;
1337 	}
1338 
1339 	dst = kzalloc(sizeof(*dst), GFP_KERNEL);
1340 	if (!dst) {
1341 		err = -ENOMEM;
1342 		goto free_ucmd;
1343 	}
1344 
1345 	mutex_lock(&dev->flow_db->lock);
1346 
1347 	ft_prio = get_flow_table(dev, flow_attr,
1348 				 is_egress ? MLX5_IB_FT_TX : MLX5_IB_FT_RX);
1349 	if (IS_ERR(ft_prio)) {
1350 		err = PTR_ERR(ft_prio);
1351 		goto unlock;
1352 	}
1353 	if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
1354 		ft_prio_tx = get_flow_table(dev, flow_attr, MLX5_IB_FT_TX);
1355 		if (IS_ERR(ft_prio_tx)) {
1356 			err = PTR_ERR(ft_prio_tx);
1357 			ft_prio_tx = NULL;
1358 			goto destroy_ft;
1359 		}
1360 	}
1361 
1362 	if (is_egress) {
1363 		dst->type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1364 	} else {
1365 		dst->type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1366 		if (mqp->is_rss)
1367 			dst->tir_num = mqp->rss_qp.tirn;
1368 		else
1369 			dst->tir_num = mqp->raw_packet_qp.rq.tirn;
1370 	}
1371 
1372 	switch (flow_attr->type) {
1373 	case IB_FLOW_ATTR_NORMAL:
1374 		underlay_qpn = (mqp->flags & IB_QP_CREATE_SOURCE_QPN) ?
1375 				       mqp->underlay_qpn :
1376 				       0;
1377 		handler = _create_flow_rule(dev, ft_prio, flow_attr, dst,
1378 					    underlay_qpn, ucmd);
1379 		break;
1380 	case IB_FLOW_ATTR_ALL_DEFAULT:
1381 	case IB_FLOW_ATTR_MC_DEFAULT:
1382 		handler = create_leftovers_rule(dev, ft_prio, flow_attr, dst);
1383 		break;
1384 	case IB_FLOW_ATTR_SNIFFER:
1385 		handler = create_sniffer_rule(dev, ft_prio, ft_prio_tx, dst);
1386 		break;
1387 	default:
1388 		err = -EINVAL;
1389 		goto destroy_ft;
1390 	}
1391 
1392 	if (IS_ERR(handler)) {
1393 		err = PTR_ERR(handler);
1394 		handler = NULL;
1395 		goto destroy_ft;
1396 	}
1397 
1398 	mutex_unlock(&dev->flow_db->lock);
1399 	kfree(dst);
1400 	kfree(ucmd);
1401 
1402 	return &handler->ibflow;
1403 
1404 destroy_ft:
1405 	put_flow_table(dev, ft_prio, false);
1406 	if (ft_prio_tx)
1407 		put_flow_table(dev, ft_prio_tx, false);
1408 unlock:
1409 	mutex_unlock(&dev->flow_db->lock);
1410 	kfree(dst);
1411 free_ucmd:
1412 	kfree(ucmd);
1413 	return ERR_PTR(err);
1414 }
1415 
1416 static struct mlx5_ib_flow_prio *
1417 _get_flow_table(struct mlx5_ib_dev *dev, u16 user_priority,
1418 		enum mlx5_flow_namespace_type ns_type,
1419 		bool mcast)
1420 {
1421 	struct mlx5_flow_namespace *ns = NULL;
1422 	struct mlx5_ib_flow_prio *prio = NULL;
1423 	int max_table_size = 0;
1424 	bool esw_encap;
1425 	u32 flags = 0;
1426 	int priority;
1427 
1428 	if (mcast)
1429 		priority = MLX5_IB_FLOW_MCAST_PRIO;
1430 	else
1431 		priority = ib_prio_to_core_prio(user_priority, false);
1432 
1433 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
1434 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
1435 	switch (ns_type) {
1436 	case MLX5_FLOW_NAMESPACE_BYPASS:
1437 		max_table_size = BIT(
1438 			MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, log_max_ft_size));
1439 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap) && !esw_encap)
1440 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1441 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
1442 					      reformat_l3_tunnel_to_l2) &&
1443 		    !esw_encap)
1444 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1445 		break;
1446 	case MLX5_FLOW_NAMESPACE_EGRESS:
1447 		max_table_size = BIT(
1448 			MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, log_max_ft_size));
1449 		if (MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat) &&
1450 		    !esw_encap)
1451 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1452 		break;
1453 	case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
1454 		max_table_size = BIT(
1455 			MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, log_max_ft_size));
1456 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, decap) && esw_encap)
1457 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1458 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev,
1459 					       reformat_l3_tunnel_to_l2) &&
1460 		    esw_encap)
1461 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1462 		priority = user_priority;
1463 		break;
1464 	case MLX5_FLOW_NAMESPACE_RDMA_RX:
1465 		max_table_size = BIT(
1466 			MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev, log_max_ft_size));
1467 		priority = user_priority;
1468 		break;
1469 	case MLX5_FLOW_NAMESPACE_RDMA_TX:
1470 		max_table_size = BIT(
1471 			MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev, log_max_ft_size));
1472 		priority = user_priority;
1473 		break;
1474 	default:
1475 		break;
1476 	}
1477 
1478 	max_table_size = min_t(int, max_table_size, MLX5_FS_MAX_ENTRIES);
1479 
1480 	ns = mlx5_get_flow_namespace(dev->mdev, ns_type);
1481 	if (!ns)
1482 		return ERR_PTR(-EOPNOTSUPP);
1483 
1484 	switch (ns_type) {
1485 	case MLX5_FLOW_NAMESPACE_BYPASS:
1486 		prio = &dev->flow_db->prios[priority];
1487 		break;
1488 	case MLX5_FLOW_NAMESPACE_EGRESS:
1489 		prio = &dev->flow_db->egress_prios[priority];
1490 		break;
1491 	case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
1492 		prio = &dev->flow_db->fdb[priority];
1493 		break;
1494 	case MLX5_FLOW_NAMESPACE_RDMA_RX:
1495 		prio = &dev->flow_db->rdma_rx[priority];
1496 		break;
1497 	case MLX5_FLOW_NAMESPACE_RDMA_TX:
1498 		prio = &dev->flow_db->rdma_tx[priority];
1499 		break;
1500 	default: return ERR_PTR(-EINVAL);
1501 	}
1502 
1503 	if (!prio)
1504 		return ERR_PTR(-EINVAL);
1505 
1506 	if (prio->flow_table)
1507 		return prio;
1508 
1509 	return _get_prio(dev, ns, prio, priority, max_table_size,
1510 			 MLX5_FS_MAX_TYPES, flags);
1511 }
1512 
1513 static struct mlx5_ib_flow_handler *
1514 _create_raw_flow_rule(struct mlx5_ib_dev *dev,
1515 		      struct mlx5_ib_flow_prio *ft_prio,
1516 		      struct mlx5_flow_destination *dst,
1517 		      struct mlx5_ib_flow_matcher  *fs_matcher,
1518 		      struct mlx5_flow_context *flow_context,
1519 		      struct mlx5_flow_act *flow_act,
1520 		      void *cmd_in, int inlen,
1521 		      int dst_num)
1522 {
1523 	struct mlx5_ib_flow_handler *handler;
1524 	struct mlx5_flow_spec *spec;
1525 	struct mlx5_flow_table *ft = ft_prio->flow_table;
1526 	int err = 0;
1527 
1528 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1529 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
1530 	if (!handler || !spec) {
1531 		err = -ENOMEM;
1532 		goto free;
1533 	}
1534 
1535 	INIT_LIST_HEAD(&handler->list);
1536 
1537 	memcpy(spec->match_value, cmd_in, inlen);
1538 	memcpy(spec->match_criteria, fs_matcher->matcher_mask.match_params,
1539 	       fs_matcher->mask_len);
1540 	spec->match_criteria_enable = fs_matcher->match_criteria_enable;
1541 	spec->flow_context = *flow_context;
1542 
1543 	handler->rule = mlx5_add_flow_rules(ft, spec,
1544 					    flow_act, dst, dst_num);
1545 
1546 	if (IS_ERR(handler->rule)) {
1547 		err = PTR_ERR(handler->rule);
1548 		goto free;
1549 	}
1550 
1551 	ft_prio->refcount++;
1552 	handler->prio = ft_prio;
1553 	handler->dev = dev;
1554 	ft_prio->flow_table = ft;
1555 
1556 free:
1557 	if (err)
1558 		kfree(handler);
1559 	kvfree(spec);
1560 	return err ? ERR_PTR(err) : handler;
1561 }
1562 
1563 static bool raw_fs_is_multicast(struct mlx5_ib_flow_matcher *fs_matcher,
1564 				void *match_v)
1565 {
1566 	void *match_c;
1567 	void *match_v_set_lyr_2_4, *match_c_set_lyr_2_4;
1568 	void *dmac, *dmac_mask;
1569 	void *ipv4, *ipv4_mask;
1570 
1571 	if (!(fs_matcher->match_criteria_enable &
1572 	      (1 << MATCH_CRITERIA_ENABLE_OUTER_BIT)))
1573 		return false;
1574 
1575 	match_c = fs_matcher->matcher_mask.match_params;
1576 	match_v_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_v,
1577 					   outer_headers);
1578 	match_c_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_c,
1579 					   outer_headers);
1580 
1581 	dmac = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1582 			    dmac_47_16);
1583 	dmac_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1584 				 dmac_47_16);
1585 
1586 	if (is_multicast_ether_addr(dmac) &&
1587 	    is_multicast_ether_addr(dmac_mask))
1588 		return true;
1589 
1590 	ipv4 = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1591 			    dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1592 
1593 	ipv4_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1594 				 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1595 
1596 	if (ipv4_is_multicast(*(__be32 *)(ipv4)) &&
1597 	    ipv4_is_multicast(*(__be32 *)(ipv4_mask)))
1598 		return true;
1599 
1600 	return false;
1601 }
1602 
1603 static struct mlx5_ib_flow_handler *raw_fs_rule_add(
1604 	struct mlx5_ib_dev *dev, struct mlx5_ib_flow_matcher *fs_matcher,
1605 	struct mlx5_flow_context *flow_context, struct mlx5_flow_act *flow_act,
1606 	u32 counter_id, void *cmd_in, int inlen, int dest_id, int dest_type)
1607 {
1608 	struct mlx5_flow_destination *dst;
1609 	struct mlx5_ib_flow_prio *ft_prio;
1610 	struct mlx5_ib_flow_handler *handler;
1611 	int dst_num = 0;
1612 	bool mcast;
1613 	int err;
1614 
1615 	if (fs_matcher->flow_type != MLX5_IB_FLOW_TYPE_NORMAL)
1616 		return ERR_PTR(-EOPNOTSUPP);
1617 
1618 	if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO)
1619 		return ERR_PTR(-ENOMEM);
1620 
1621 	dst = kcalloc(2, sizeof(*dst), GFP_KERNEL);
1622 	if (!dst)
1623 		return ERR_PTR(-ENOMEM);
1624 
1625 	mcast = raw_fs_is_multicast(fs_matcher, cmd_in);
1626 	mutex_lock(&dev->flow_db->lock);
1627 
1628 	ft_prio = _get_flow_table(dev, fs_matcher->priority,
1629 				  fs_matcher->ns_type, mcast);
1630 	if (IS_ERR(ft_prio)) {
1631 		err = PTR_ERR(ft_prio);
1632 		goto unlock;
1633 	}
1634 
1635 	switch (dest_type) {
1636 	case MLX5_FLOW_DESTINATION_TYPE_TIR:
1637 		dst[dst_num].type = dest_type;
1638 		dst[dst_num++].tir_num = dest_id;
1639 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1640 		break;
1641 	case MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE:
1642 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM;
1643 		dst[dst_num++].ft_num = dest_id;
1644 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1645 		break;
1646 	case MLX5_FLOW_DESTINATION_TYPE_PORT:
1647 		dst[dst_num++].type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1648 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1649 		break;
1650 	default:
1651 		break;
1652 	}
1653 
1654 	if (flow_act->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1655 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1656 		dst[dst_num].counter_id = counter_id;
1657 		dst_num++;
1658 	}
1659 
1660 	handler = _create_raw_flow_rule(dev, ft_prio, dst_num ? dst : NULL,
1661 					fs_matcher, flow_context, flow_act,
1662 					cmd_in, inlen, dst_num);
1663 
1664 	if (IS_ERR(handler)) {
1665 		err = PTR_ERR(handler);
1666 		goto destroy_ft;
1667 	}
1668 
1669 	mutex_unlock(&dev->flow_db->lock);
1670 	atomic_inc(&fs_matcher->usecnt);
1671 	handler->flow_matcher = fs_matcher;
1672 
1673 	kfree(dst);
1674 
1675 	return handler;
1676 
1677 destroy_ft:
1678 	put_flow_table(dev, ft_prio, false);
1679 unlock:
1680 	mutex_unlock(&dev->flow_db->lock);
1681 	kfree(dst);
1682 
1683 	return ERR_PTR(err);
1684 }
1685 
1686 static void destroy_flow_action_raw(struct mlx5_ib_flow_action *maction)
1687 {
1688 	switch (maction->flow_action_raw.sub_type) {
1689 	case MLX5_IB_FLOW_ACTION_MODIFY_HEADER:
1690 		mlx5_modify_header_dealloc(maction->flow_action_raw.dev->mdev,
1691 					   maction->flow_action_raw.modify_hdr);
1692 		break;
1693 	case MLX5_IB_FLOW_ACTION_PACKET_REFORMAT:
1694 		mlx5_packet_reformat_dealloc(maction->flow_action_raw.dev->mdev,
1695 					     maction->flow_action_raw.pkt_reformat);
1696 		break;
1697 	case MLX5_IB_FLOW_ACTION_DECAP:
1698 		break;
1699 	default:
1700 		break;
1701 	}
1702 }
1703 
1704 static int mlx5_ib_destroy_flow_action(struct ib_flow_action *action)
1705 {
1706 	struct mlx5_ib_flow_action *maction = to_mflow_act(action);
1707 
1708 	switch (action->type) {
1709 	case IB_FLOW_ACTION_UNSPECIFIED:
1710 		destroy_flow_action_raw(maction);
1711 		break;
1712 	default:
1713 		WARN_ON(true);
1714 		break;
1715 	}
1716 
1717 	kfree(maction);
1718 	return 0;
1719 }
1720 
1721 static int
1722 mlx5_ib_ft_type_to_namespace(enum mlx5_ib_uapi_flow_table_type table_type,
1723 			     enum mlx5_flow_namespace_type *namespace)
1724 {
1725 	switch (table_type) {
1726 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX:
1727 		*namespace = MLX5_FLOW_NAMESPACE_BYPASS;
1728 		break;
1729 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX:
1730 		*namespace = MLX5_FLOW_NAMESPACE_EGRESS;
1731 		break;
1732 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_FDB:
1733 		*namespace = MLX5_FLOW_NAMESPACE_FDB_BYPASS;
1734 		break;
1735 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_RX:
1736 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_RX;
1737 		break;
1738 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_TX:
1739 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_TX;
1740 		break;
1741 	default:
1742 		return -EINVAL;
1743 	}
1744 
1745 	return 0;
1746 }
1747 
1748 static const struct uverbs_attr_spec mlx5_ib_flow_type[] = {
1749 	[MLX5_IB_FLOW_TYPE_NORMAL] = {
1750 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1751 		.u.ptr = {
1752 			.len = sizeof(u16), /* data is priority */
1753 			.min_len = sizeof(u16),
1754 		}
1755 	},
1756 	[MLX5_IB_FLOW_TYPE_SNIFFER] = {
1757 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1758 		UVERBS_ATTR_NO_DATA(),
1759 	},
1760 	[MLX5_IB_FLOW_TYPE_ALL_DEFAULT] = {
1761 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1762 		UVERBS_ATTR_NO_DATA(),
1763 	},
1764 	[MLX5_IB_FLOW_TYPE_MC_DEFAULT] = {
1765 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1766 		UVERBS_ATTR_NO_DATA(),
1767 	},
1768 };
1769 
1770 static bool is_flow_dest(void *obj, int *dest_id, int *dest_type)
1771 {
1772 	struct devx_obj *devx_obj = obj;
1773 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1774 
1775 	switch (opcode) {
1776 	case MLX5_CMD_OP_DESTROY_TIR:
1777 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1778 		*dest_id = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox,
1779 				    obj_id);
1780 		return true;
1781 
1782 	case MLX5_CMD_OP_DESTROY_FLOW_TABLE:
1783 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1784 		*dest_id = MLX5_GET(destroy_flow_table_in, devx_obj->dinbox,
1785 				    table_id);
1786 		return true;
1787 	default:
1788 		return false;
1789 	}
1790 }
1791 
1792 static int get_dests(struct uverbs_attr_bundle *attrs,
1793 		     struct mlx5_ib_flow_matcher *fs_matcher, int *dest_id,
1794 		     int *dest_type, struct ib_qp **qp, u32 *flags)
1795 {
1796 	bool dest_devx, dest_qp;
1797 	void *devx_obj;
1798 	int err;
1799 
1800 	dest_devx = uverbs_attr_is_valid(attrs,
1801 					 MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1802 	dest_qp = uverbs_attr_is_valid(attrs,
1803 				       MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1804 
1805 	*flags = 0;
1806 	err = uverbs_get_flags32(flags, attrs, MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
1807 				 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS |
1808 					 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP);
1809 	if (err)
1810 		return err;
1811 
1812 	/* Both flags are not allowed */
1813 	if (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS &&
1814 	    *flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1815 		return -EINVAL;
1816 
1817 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_BYPASS) {
1818 		if (dest_devx && (dest_qp || *flags))
1819 			return -EINVAL;
1820 		else if (dest_qp && *flags)
1821 			return -EINVAL;
1822 	}
1823 
1824 	/* Allow only DEVX object, drop as dest for FDB */
1825 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS &&
1826 	    !(dest_devx || (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)))
1827 		return -EINVAL;
1828 
1829 	/* Allow only DEVX object or QP as dest when inserting to RDMA_RX */
1830 	if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1831 	    ((!dest_devx && !dest_qp) || (dest_devx && dest_qp)))
1832 		return -EINVAL;
1833 
1834 	*qp = NULL;
1835 	if (dest_devx) {
1836 		devx_obj =
1837 			uverbs_attr_get_obj(attrs,
1838 					    MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1839 
1840 		/* Verify that the given DEVX object is a flow
1841 		 * steering destination.
1842 		 */
1843 		if (!is_flow_dest(devx_obj, dest_id, dest_type))
1844 			return -EINVAL;
1845 		/* Allow only flow table as dest when inserting to FDB or RDMA_RX */
1846 		if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS ||
1847 		     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1848 		    *dest_type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1849 			return -EINVAL;
1850 	} else if (dest_qp) {
1851 		struct mlx5_ib_qp *mqp;
1852 
1853 		*qp = uverbs_attr_get_obj(attrs,
1854 					  MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1855 		if (IS_ERR(*qp))
1856 			return PTR_ERR(*qp);
1857 
1858 		if ((*qp)->qp_type != IB_QPT_RAW_PACKET)
1859 			return -EINVAL;
1860 
1861 		mqp = to_mqp(*qp);
1862 		if (mqp->is_rss)
1863 			*dest_id = mqp->rss_qp.tirn;
1864 		else
1865 			*dest_id = mqp->raw_packet_qp.rq.tirn;
1866 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1867 	} else if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1868 		    fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX) &&
1869 		   !(*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)) {
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 
2019 	if (atomic_read(&obj->usecnt))
2020 		return -EBUSY;
2021 
2022 	kfree(obj);
2023 	return 0;
2024 }
2025 
2026 static int steering_anchor_create_ft(struct mlx5_ib_dev *dev,
2027 				     struct mlx5_ib_flow_prio *ft_prio,
2028 				     enum mlx5_flow_namespace_type ns_type)
2029 {
2030 	struct mlx5_flow_table_attr ft_attr = {};
2031 	struct mlx5_flow_namespace *ns;
2032 	struct mlx5_flow_table *ft;
2033 
2034 	if (ft_prio->anchor.ft)
2035 		return 0;
2036 
2037 	ns = mlx5_get_flow_namespace(dev->mdev, ns_type);
2038 	if (!ns)
2039 		return -EOPNOTSUPP;
2040 
2041 	ft_attr.flags = MLX5_FLOW_TABLE_UNMANAGED;
2042 	ft_attr.uid = MLX5_SHARED_RESOURCE_UID;
2043 	ft_attr.prio = 0;
2044 	ft_attr.max_fte = 2;
2045 	ft_attr.level = 1;
2046 
2047 	ft = mlx5_create_flow_table(ns, &ft_attr);
2048 	if (IS_ERR(ft))
2049 		return PTR_ERR(ft);
2050 
2051 	ft_prio->anchor.ft = ft;
2052 
2053 	return 0;
2054 }
2055 
2056 static void steering_anchor_destroy_ft(struct mlx5_ib_flow_prio *ft_prio)
2057 {
2058 	if (ft_prio->anchor.ft) {
2059 		mlx5_destroy_flow_table(ft_prio->anchor.ft);
2060 		ft_prio->anchor.ft = NULL;
2061 	}
2062 }
2063 
2064 static int
2065 steering_anchor_create_fg_drop(struct mlx5_ib_flow_prio *ft_prio)
2066 {
2067 	int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
2068 	struct mlx5_flow_group *fg;
2069 	void *flow_group_in;
2070 	int err = 0;
2071 
2072 	if (ft_prio->anchor.fg_drop)
2073 		return 0;
2074 
2075 	flow_group_in = kvzalloc(inlen, GFP_KERNEL);
2076 	if (!flow_group_in)
2077 		return -ENOMEM;
2078 
2079 	MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, 1);
2080 	MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index, 1);
2081 
2082 	fg = mlx5_create_flow_group(ft_prio->anchor.ft, flow_group_in);
2083 	if (IS_ERR(fg)) {
2084 		err = PTR_ERR(fg);
2085 		goto out;
2086 	}
2087 
2088 	ft_prio->anchor.fg_drop = fg;
2089 
2090 out:
2091 	kvfree(flow_group_in);
2092 
2093 	return err;
2094 }
2095 
2096 static void
2097 steering_anchor_destroy_fg_drop(struct mlx5_ib_flow_prio *ft_prio)
2098 {
2099 	if (ft_prio->anchor.fg_drop) {
2100 		mlx5_destroy_flow_group(ft_prio->anchor.fg_drop);
2101 		ft_prio->anchor.fg_drop = NULL;
2102 	}
2103 }
2104 
2105 static int
2106 steering_anchor_create_fg_goto_table(struct mlx5_ib_flow_prio *ft_prio)
2107 {
2108 	int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
2109 	struct mlx5_flow_group *fg;
2110 	void *flow_group_in;
2111 	int err = 0;
2112 
2113 	if (ft_prio->anchor.fg_goto_table)
2114 		return 0;
2115 
2116 	flow_group_in = kvzalloc(inlen, GFP_KERNEL);
2117 	if (!flow_group_in)
2118 		return -ENOMEM;
2119 
2120 	fg = mlx5_create_flow_group(ft_prio->anchor.ft, flow_group_in);
2121 	if (IS_ERR(fg)) {
2122 		err = PTR_ERR(fg);
2123 		goto out;
2124 	}
2125 	ft_prio->anchor.fg_goto_table = fg;
2126 
2127 out:
2128 	kvfree(flow_group_in);
2129 
2130 	return err;
2131 }
2132 
2133 static void
2134 steering_anchor_destroy_fg_goto_table(struct mlx5_ib_flow_prio *ft_prio)
2135 {
2136 	if (ft_prio->anchor.fg_goto_table) {
2137 		mlx5_destroy_flow_group(ft_prio->anchor.fg_goto_table);
2138 		ft_prio->anchor.fg_goto_table = NULL;
2139 	}
2140 }
2141 
2142 static int
2143 steering_anchor_create_rule_drop(struct mlx5_ib_flow_prio *ft_prio)
2144 {
2145 	struct mlx5_flow_act flow_act = {};
2146 	struct mlx5_flow_handle *handle;
2147 
2148 	if (ft_prio->anchor.rule_drop)
2149 		return 0;
2150 
2151 	flow_act.fg = ft_prio->anchor.fg_drop;
2152 	flow_act.action = MLX5_FLOW_CONTEXT_ACTION_DROP;
2153 
2154 	handle = mlx5_add_flow_rules(ft_prio->anchor.ft, NULL, &flow_act,
2155 				     NULL, 0);
2156 	if (IS_ERR(handle))
2157 		return PTR_ERR(handle);
2158 
2159 	ft_prio->anchor.rule_drop = handle;
2160 
2161 	return 0;
2162 }
2163 
2164 static void steering_anchor_destroy_rule_drop(struct mlx5_ib_flow_prio *ft_prio)
2165 {
2166 	if (ft_prio->anchor.rule_drop) {
2167 		mlx5_del_flow_rules(ft_prio->anchor.rule_drop);
2168 		ft_prio->anchor.rule_drop = NULL;
2169 	}
2170 }
2171 
2172 static int
2173 steering_anchor_create_rule_goto_table(struct mlx5_ib_flow_prio *ft_prio)
2174 {
2175 	struct mlx5_flow_destination dest = {};
2176 	struct mlx5_flow_act flow_act = {};
2177 	struct mlx5_flow_handle *handle;
2178 
2179 	if (ft_prio->anchor.rule_goto_table)
2180 		return 0;
2181 
2182 	flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2183 	flow_act.flags |= FLOW_ACT_IGNORE_FLOW_LEVEL;
2184 	flow_act.fg = ft_prio->anchor.fg_goto_table;
2185 
2186 	dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
2187 	dest.ft = ft_prio->flow_table;
2188 
2189 	handle = mlx5_add_flow_rules(ft_prio->anchor.ft, NULL, &flow_act,
2190 				     &dest, 1);
2191 	if (IS_ERR(handle))
2192 		return PTR_ERR(handle);
2193 
2194 	ft_prio->anchor.rule_goto_table = handle;
2195 
2196 	return 0;
2197 }
2198 
2199 static void
2200 steering_anchor_destroy_rule_goto_table(struct mlx5_ib_flow_prio *ft_prio)
2201 {
2202 	if (ft_prio->anchor.rule_goto_table) {
2203 		mlx5_del_flow_rules(ft_prio->anchor.rule_goto_table);
2204 		ft_prio->anchor.rule_goto_table = NULL;
2205 	}
2206 }
2207 
2208 static int steering_anchor_create_res(struct mlx5_ib_dev *dev,
2209 				      struct mlx5_ib_flow_prio *ft_prio,
2210 				      enum mlx5_flow_namespace_type ns_type)
2211 {
2212 	int err;
2213 
2214 	err = steering_anchor_create_ft(dev, ft_prio, ns_type);
2215 	if (err)
2216 		return err;
2217 
2218 	err = steering_anchor_create_fg_drop(ft_prio);
2219 	if (err)
2220 		goto destroy_ft;
2221 
2222 	err = steering_anchor_create_fg_goto_table(ft_prio);
2223 	if (err)
2224 		goto destroy_fg_drop;
2225 
2226 	err = steering_anchor_create_rule_drop(ft_prio);
2227 	if (err)
2228 		goto destroy_fg_goto_table;
2229 
2230 	err = steering_anchor_create_rule_goto_table(ft_prio);
2231 	if (err)
2232 		goto destroy_rule_drop;
2233 
2234 	return 0;
2235 
2236 destroy_rule_drop:
2237 	steering_anchor_destroy_rule_drop(ft_prio);
2238 destroy_fg_goto_table:
2239 	steering_anchor_destroy_fg_goto_table(ft_prio);
2240 destroy_fg_drop:
2241 	steering_anchor_destroy_fg_drop(ft_prio);
2242 destroy_ft:
2243 	steering_anchor_destroy_ft(ft_prio);
2244 
2245 	return err;
2246 }
2247 
2248 static void mlx5_steering_anchor_destroy_res(struct mlx5_ib_flow_prio *ft_prio)
2249 {
2250 	steering_anchor_destroy_rule_goto_table(ft_prio);
2251 	steering_anchor_destroy_rule_drop(ft_prio);
2252 	steering_anchor_destroy_fg_goto_table(ft_prio);
2253 	steering_anchor_destroy_fg_drop(ft_prio);
2254 	steering_anchor_destroy_ft(ft_prio);
2255 }
2256 
2257 static int steering_anchor_cleanup(struct ib_uobject *uobject,
2258 				   enum rdma_remove_reason why,
2259 				   struct uverbs_attr_bundle *attrs)
2260 {
2261 	struct mlx5_ib_steering_anchor *obj = uobject->object;
2262 
2263 	if (atomic_read(&obj->usecnt))
2264 		return -EBUSY;
2265 
2266 	mutex_lock(&obj->dev->flow_db->lock);
2267 	if (!--obj->ft_prio->anchor.rule_goto_table_ref)
2268 		steering_anchor_destroy_rule_goto_table(obj->ft_prio);
2269 
2270 	put_flow_table(obj->dev, obj->ft_prio, true);
2271 	mutex_unlock(&obj->dev->flow_db->lock);
2272 
2273 	kfree(obj);
2274 	return 0;
2275 }
2276 
2277 static void fs_cleanup_anchor(struct mlx5_ib_flow_prio *prio,
2278 			      int count)
2279 {
2280 	while (count--)
2281 		mlx5_steering_anchor_destroy_res(&prio[count]);
2282 }
2283 
2284 void mlx5_ib_fs_cleanup_anchor(struct mlx5_ib_dev *dev)
2285 {
2286 	fs_cleanup_anchor(dev->flow_db->prios, MLX5_IB_NUM_FLOW_FT);
2287 	fs_cleanup_anchor(dev->flow_db->egress_prios, MLX5_IB_NUM_FLOW_FT);
2288 	fs_cleanup_anchor(dev->flow_db->sniffer, MLX5_IB_NUM_SNIFFER_FTS);
2289 	fs_cleanup_anchor(dev->flow_db->egress, MLX5_IB_NUM_EGRESS_FTS);
2290 	fs_cleanup_anchor(dev->flow_db->fdb, MLX5_IB_NUM_FDB_FTS);
2291 	fs_cleanup_anchor(dev->flow_db->rdma_rx, MLX5_IB_NUM_FLOW_FT);
2292 	fs_cleanup_anchor(dev->flow_db->rdma_tx, MLX5_IB_NUM_FLOW_FT);
2293 }
2294 
2295 static int mlx5_ib_matcher_ns(struct uverbs_attr_bundle *attrs,
2296 			      struct mlx5_ib_flow_matcher *obj)
2297 {
2298 	enum mlx5_ib_uapi_flow_table_type ft_type =
2299 		MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX;
2300 	u32 flags;
2301 	int err;
2302 
2303 	/* New users should use MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE and older
2304 	 * users should switch to it. We leave this to not break userspace
2305 	 */
2306 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE) &&
2307 	    uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS))
2308 		return -EINVAL;
2309 
2310 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE)) {
2311 		err = uverbs_get_const(&ft_type, attrs,
2312 				       MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE);
2313 		if (err)
2314 			return err;
2315 
2316 		err = mlx5_ib_ft_type_to_namespace(ft_type, &obj->ns_type);
2317 		if (err)
2318 			return err;
2319 
2320 		return 0;
2321 	}
2322 
2323 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS)) {
2324 		err = uverbs_get_flags32(&flags, attrs,
2325 					 MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2326 					 IB_FLOW_ATTR_FLAGS_EGRESS);
2327 		if (err)
2328 			return err;
2329 
2330 		if (flags)
2331 			return mlx5_ib_ft_type_to_namespace(
2332 				MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX,
2333 				&obj->ns_type);
2334 	}
2335 
2336 	obj->ns_type = MLX5_FLOW_NAMESPACE_BYPASS;
2337 
2338 	return 0;
2339 }
2340 
2341 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_MATCHER_CREATE)(
2342 	struct uverbs_attr_bundle *attrs)
2343 {
2344 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2345 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE);
2346 	struct mlx5_ib_dev *dev = mlx5_udata_to_mdev(&attrs->driver_udata);
2347 	struct mlx5_ib_flow_matcher *obj;
2348 	int err;
2349 
2350 	obj = kzalloc(sizeof(struct mlx5_ib_flow_matcher), GFP_KERNEL);
2351 	if (!obj)
2352 		return -ENOMEM;
2353 
2354 	obj->mask_len = uverbs_attr_get_len(
2355 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2356 	err = uverbs_copy_from(&obj->matcher_mask,
2357 			       attrs,
2358 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2359 	if (err)
2360 		goto end;
2361 
2362 	obj->flow_type = uverbs_attr_get_enum_id(
2363 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2364 
2365 	if (obj->flow_type == MLX5_IB_FLOW_TYPE_NORMAL) {
2366 		err = uverbs_copy_from(&obj->priority,
2367 				       attrs,
2368 				       MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2369 		if (err)
2370 			goto end;
2371 	}
2372 
2373 	err = uverbs_copy_from(&obj->match_criteria_enable,
2374 			       attrs,
2375 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA);
2376 	if (err)
2377 		goto end;
2378 
2379 	err = mlx5_ib_matcher_ns(attrs, obj);
2380 	if (err)
2381 		goto end;
2382 
2383 	if (obj->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS &&
2384 	    mlx5_eswitch_mode(dev->mdev) != MLX5_ESWITCH_OFFLOADS) {
2385 		err = -EINVAL;
2386 		goto end;
2387 	}
2388 
2389 	uobj->object = obj;
2390 	obj->mdev = dev->mdev;
2391 	atomic_set(&obj->usecnt, 0);
2392 	return 0;
2393 
2394 end:
2395 	kfree(obj);
2396 	return err;
2397 }
2398 
2399 static int UVERBS_HANDLER(MLX5_IB_METHOD_STEERING_ANCHOR_CREATE)(
2400 	struct uverbs_attr_bundle *attrs)
2401 {
2402 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2403 		attrs, MLX5_IB_ATTR_STEERING_ANCHOR_CREATE_HANDLE);
2404 	struct mlx5_ib_dev *dev = mlx5_udata_to_mdev(&attrs->driver_udata);
2405 	enum mlx5_ib_uapi_flow_table_type ib_uapi_ft_type;
2406 	enum mlx5_flow_namespace_type ns_type;
2407 	struct mlx5_ib_steering_anchor *obj;
2408 	struct mlx5_ib_flow_prio *ft_prio;
2409 	u16 priority;
2410 	u32 ft_id;
2411 	int err;
2412 
2413 	if (!capable(CAP_NET_RAW))
2414 		return -EPERM;
2415 
2416 	err = uverbs_get_const(&ib_uapi_ft_type, attrs,
2417 			       MLX5_IB_ATTR_STEERING_ANCHOR_FT_TYPE);
2418 	if (err)
2419 		return err;
2420 
2421 	err = mlx5_ib_ft_type_to_namespace(ib_uapi_ft_type, &ns_type);
2422 	if (err)
2423 		return err;
2424 
2425 	err = uverbs_copy_from(&priority, attrs,
2426 			       MLX5_IB_ATTR_STEERING_ANCHOR_PRIORITY);
2427 	if (err)
2428 		return err;
2429 
2430 	obj = kzalloc(sizeof(*obj), GFP_KERNEL);
2431 	if (!obj)
2432 		return -ENOMEM;
2433 
2434 	mutex_lock(&dev->flow_db->lock);
2435 
2436 	ft_prio = _get_flow_table(dev, priority, ns_type, 0);
2437 	if (IS_ERR(ft_prio)) {
2438 		err = PTR_ERR(ft_prio);
2439 		goto free_obj;
2440 	}
2441 
2442 	ft_prio->refcount++;
2443 
2444 	if (!ft_prio->anchor.rule_goto_table_ref) {
2445 		err = steering_anchor_create_res(dev, ft_prio, ns_type);
2446 		if (err)
2447 			goto put_flow_table;
2448 	}
2449 
2450 	ft_prio->anchor.rule_goto_table_ref++;
2451 
2452 	ft_id = mlx5_flow_table_id(ft_prio->anchor.ft);
2453 
2454 	err = uverbs_copy_to(attrs, MLX5_IB_ATTR_STEERING_ANCHOR_FT_ID,
2455 			     &ft_id, sizeof(ft_id));
2456 	if (err)
2457 		goto destroy_res;
2458 
2459 	mutex_unlock(&dev->flow_db->lock);
2460 
2461 	uobj->object = obj;
2462 	obj->dev = dev;
2463 	obj->ft_prio = ft_prio;
2464 	atomic_set(&obj->usecnt, 0);
2465 
2466 	return 0;
2467 
2468 destroy_res:
2469 	--ft_prio->anchor.rule_goto_table_ref;
2470 	mlx5_steering_anchor_destroy_res(ft_prio);
2471 put_flow_table:
2472 	put_flow_table(dev, ft_prio, true);
2473 	mutex_unlock(&dev->flow_db->lock);
2474 free_obj:
2475 	kfree(obj);
2476 
2477 	return err;
2478 }
2479 
2480 static struct ib_flow_action *
2481 mlx5_ib_create_modify_header(struct mlx5_ib_dev *dev,
2482 			     enum mlx5_ib_uapi_flow_table_type ft_type,
2483 			     u8 num_actions, void *in)
2484 {
2485 	enum mlx5_flow_namespace_type namespace;
2486 	struct mlx5_ib_flow_action *maction;
2487 	int ret;
2488 
2489 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2490 	if (ret)
2491 		return ERR_PTR(-EINVAL);
2492 
2493 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2494 	if (!maction)
2495 		return ERR_PTR(-ENOMEM);
2496 
2497 	maction->flow_action_raw.modify_hdr =
2498 		mlx5_modify_header_alloc(dev->mdev, namespace, num_actions, in);
2499 
2500 	if (IS_ERR(maction->flow_action_raw.modify_hdr)) {
2501 		ret = PTR_ERR(maction->flow_action_raw.modify_hdr);
2502 		kfree(maction);
2503 		return ERR_PTR(ret);
2504 	}
2505 	maction->flow_action_raw.sub_type =
2506 		MLX5_IB_FLOW_ACTION_MODIFY_HEADER;
2507 	maction->flow_action_raw.dev = dev;
2508 
2509 	return &maction->ib_action;
2510 }
2511 
2512 static bool mlx5_ib_modify_header_supported(struct mlx5_ib_dev *dev)
2513 {
2514 	return MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
2515 					 max_modify_header_actions) ||
2516 	       MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev,
2517 					 max_modify_header_actions) ||
2518 	       MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
2519 					 max_modify_header_actions);
2520 }
2521 
2522 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER)(
2523 	struct uverbs_attr_bundle *attrs)
2524 {
2525 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2526 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE);
2527 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2528 	enum mlx5_ib_uapi_flow_table_type ft_type;
2529 	struct ib_flow_action *action;
2530 	int num_actions;
2531 	void *in;
2532 	int ret;
2533 
2534 	if (!mlx5_ib_modify_header_supported(mdev))
2535 		return -EOPNOTSUPP;
2536 
2537 	in = uverbs_attr_get_alloced_ptr(attrs,
2538 		MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM);
2539 
2540 	num_actions = uverbs_attr_ptr_get_array_size(
2541 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2542 		MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto));
2543 	if (num_actions < 0)
2544 		return num_actions;
2545 
2546 	ret = uverbs_get_const(&ft_type, attrs,
2547 			       MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE);
2548 	if (ret)
2549 		return ret;
2550 	action = mlx5_ib_create_modify_header(mdev, ft_type, num_actions, in);
2551 	if (IS_ERR(action))
2552 		return PTR_ERR(action);
2553 
2554 	uverbs_flow_action_fill_action(action, uobj, &mdev->ib_dev,
2555 				       IB_FLOW_ACTION_UNSPECIFIED);
2556 
2557 	return 0;
2558 }
2559 
2560 static bool mlx5_ib_flow_action_packet_reformat_valid(struct mlx5_ib_dev *ibdev,
2561 						      u8 packet_reformat_type,
2562 						      u8 ft_type)
2563 {
2564 	switch (packet_reformat_type) {
2565 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2566 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2567 			return MLX5_CAP_FLOWTABLE(ibdev->mdev,
2568 						  encap_general_header);
2569 		break;
2570 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2571 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2572 			return MLX5_CAP_FLOWTABLE_NIC_TX(ibdev->mdev,
2573 				reformat_l2_to_l3_tunnel);
2574 		break;
2575 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2576 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2577 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev,
2578 				reformat_l3_tunnel_to_l2);
2579 		break;
2580 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2:
2581 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2582 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev, decap);
2583 		break;
2584 	default:
2585 		break;
2586 	}
2587 
2588 	return false;
2589 }
2590 
2591 static int mlx5_ib_dv_to_prm_packet_reforamt_type(u8 dv_prt, u8 *prm_prt)
2592 {
2593 	switch (dv_prt) {
2594 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2595 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L2_TUNNEL;
2596 		break;
2597 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2598 		*prm_prt = MLX5_REFORMAT_TYPE_L3_TUNNEL_TO_L2;
2599 		break;
2600 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2601 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
2602 		break;
2603 	default:
2604 		return -EINVAL;
2605 	}
2606 
2607 	return 0;
2608 }
2609 
2610 static int mlx5_ib_flow_action_create_packet_reformat_ctx(
2611 	struct mlx5_ib_dev *dev,
2612 	struct mlx5_ib_flow_action *maction,
2613 	u8 ft_type, u8 dv_prt,
2614 	void *in, size_t len)
2615 {
2616 	struct mlx5_pkt_reformat_params reformat_params;
2617 	enum mlx5_flow_namespace_type namespace;
2618 	u8 prm_prt;
2619 	int ret;
2620 
2621 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2622 	if (ret)
2623 		return ret;
2624 
2625 	ret = mlx5_ib_dv_to_prm_packet_reforamt_type(dv_prt, &prm_prt);
2626 	if (ret)
2627 		return ret;
2628 
2629 	memset(&reformat_params, 0, sizeof(reformat_params));
2630 	reformat_params.type = prm_prt;
2631 	reformat_params.size = len;
2632 	reformat_params.data = in;
2633 	maction->flow_action_raw.pkt_reformat =
2634 		mlx5_packet_reformat_alloc(dev->mdev, &reformat_params,
2635 					   namespace);
2636 	if (IS_ERR(maction->flow_action_raw.pkt_reformat)) {
2637 		ret = PTR_ERR(maction->flow_action_raw.pkt_reformat);
2638 		return ret;
2639 	}
2640 
2641 	maction->flow_action_raw.sub_type =
2642 		MLX5_IB_FLOW_ACTION_PACKET_REFORMAT;
2643 	maction->flow_action_raw.dev = dev;
2644 
2645 	return 0;
2646 }
2647 
2648 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT)(
2649 	struct uverbs_attr_bundle *attrs)
2650 {
2651 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs,
2652 		MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE);
2653 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2654 	enum mlx5_ib_uapi_flow_action_packet_reformat_type dv_prt;
2655 	enum mlx5_ib_uapi_flow_table_type ft_type;
2656 	struct mlx5_ib_flow_action *maction;
2657 	int ret;
2658 
2659 	ret = uverbs_get_const(&ft_type, attrs,
2660 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE);
2661 	if (ret)
2662 		return ret;
2663 
2664 	ret = uverbs_get_const(&dv_prt, attrs,
2665 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE);
2666 	if (ret)
2667 		return ret;
2668 
2669 	if (!mlx5_ib_flow_action_packet_reformat_valid(mdev, dv_prt, ft_type))
2670 		return -EOPNOTSUPP;
2671 
2672 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2673 	if (!maction)
2674 		return -ENOMEM;
2675 
2676 	if (dv_prt ==
2677 	    MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2) {
2678 		maction->flow_action_raw.sub_type =
2679 			MLX5_IB_FLOW_ACTION_DECAP;
2680 		maction->flow_action_raw.dev = mdev;
2681 	} else {
2682 		void *in;
2683 		int len;
2684 
2685 		in = uverbs_attr_get_alloced_ptr(attrs,
2686 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2687 		if (IS_ERR(in)) {
2688 			ret = PTR_ERR(in);
2689 			goto free_maction;
2690 		}
2691 
2692 		len = uverbs_attr_get_len(attrs,
2693 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2694 
2695 		ret = mlx5_ib_flow_action_create_packet_reformat_ctx(mdev,
2696 			maction, ft_type, dv_prt, in, len);
2697 		if (ret)
2698 			goto free_maction;
2699 	}
2700 
2701 	uverbs_flow_action_fill_action(&maction->ib_action, uobj, &mdev->ib_dev,
2702 				       IB_FLOW_ACTION_UNSPECIFIED);
2703 	return 0;
2704 
2705 free_maction:
2706 	kfree(maction);
2707 	return ret;
2708 }
2709 
2710 DECLARE_UVERBS_NAMED_METHOD(
2711 	MLX5_IB_METHOD_CREATE_FLOW,
2712 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2713 			UVERBS_OBJECT_FLOW,
2714 			UVERBS_ACCESS_NEW,
2715 			UA_MANDATORY),
2716 	UVERBS_ATTR_PTR_IN(
2717 		MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE,
2718 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2719 		UA_MANDATORY,
2720 		UA_ALLOC_AND_COPY),
2721 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_MATCHER,
2722 			MLX5_IB_OBJECT_FLOW_MATCHER,
2723 			UVERBS_ACCESS_READ,
2724 			UA_MANDATORY),
2725 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_QP,
2726 			UVERBS_OBJECT_QP,
2727 			UVERBS_ACCESS_READ),
2728 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX,
2729 			MLX5_IB_OBJECT_DEVX_OBJ,
2730 			UVERBS_ACCESS_READ),
2731 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS,
2732 			     UVERBS_OBJECT_FLOW_ACTION,
2733 			     UVERBS_ACCESS_READ, 1,
2734 			     MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS,
2735 			     UA_OPTIONAL),
2736 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_TAG,
2737 			   UVERBS_ATTR_TYPE(u32),
2738 			   UA_OPTIONAL),
2739 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX,
2740 			     MLX5_IB_OBJECT_DEVX_OBJ,
2741 			     UVERBS_ACCESS_READ, 1, 1,
2742 			     UA_OPTIONAL),
2743 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
2744 			   UVERBS_ATTR_MIN_SIZE(sizeof(u32)),
2745 			   UA_OPTIONAL,
2746 			   UA_ALLOC_AND_COPY),
2747 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
2748 			     enum mlx5_ib_create_flow_flags,
2749 			     UA_OPTIONAL));
2750 
2751 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2752 	MLX5_IB_METHOD_DESTROY_FLOW,
2753 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2754 			UVERBS_OBJECT_FLOW,
2755 			UVERBS_ACCESS_DESTROY,
2756 			UA_MANDATORY));
2757 
2758 ADD_UVERBS_METHODS(mlx5_ib_fs,
2759 		   UVERBS_OBJECT_FLOW,
2760 		   &UVERBS_METHOD(MLX5_IB_METHOD_CREATE_FLOW),
2761 		   &UVERBS_METHOD(MLX5_IB_METHOD_DESTROY_FLOW));
2762 
2763 DECLARE_UVERBS_NAMED_METHOD(
2764 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER,
2765 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE,
2766 			UVERBS_OBJECT_FLOW_ACTION,
2767 			UVERBS_ACCESS_NEW,
2768 			UA_MANDATORY),
2769 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2770 			   UVERBS_ATTR_MIN_SIZE(MLX5_UN_SZ_BYTES(
2771 				   set_add_copy_action_in_auto)),
2772 			   UA_MANDATORY,
2773 			   UA_ALLOC_AND_COPY),
2774 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE,
2775 			     enum mlx5_ib_uapi_flow_table_type,
2776 			     UA_MANDATORY));
2777 
2778 DECLARE_UVERBS_NAMED_METHOD(
2779 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT,
2780 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE,
2781 			UVERBS_OBJECT_FLOW_ACTION,
2782 			UVERBS_ACCESS_NEW,
2783 			UA_MANDATORY),
2784 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF,
2785 			   UVERBS_ATTR_MIN_SIZE(1),
2786 			   UA_ALLOC_AND_COPY,
2787 			   UA_OPTIONAL),
2788 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE,
2789 			     enum mlx5_ib_uapi_flow_action_packet_reformat_type,
2790 			     UA_MANDATORY),
2791 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE,
2792 			     enum mlx5_ib_uapi_flow_table_type,
2793 			     UA_MANDATORY));
2794 
2795 ADD_UVERBS_METHODS(
2796 	mlx5_ib_flow_actions,
2797 	UVERBS_OBJECT_FLOW_ACTION,
2798 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER),
2799 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT));
2800 
2801 DECLARE_UVERBS_NAMED_METHOD(
2802 	MLX5_IB_METHOD_FLOW_MATCHER_CREATE,
2803 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE,
2804 			MLX5_IB_OBJECT_FLOW_MATCHER,
2805 			UVERBS_ACCESS_NEW,
2806 			UA_MANDATORY),
2807 	UVERBS_ATTR_PTR_IN(
2808 		MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK,
2809 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2810 		UA_MANDATORY),
2811 	UVERBS_ATTR_ENUM_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE,
2812 			    mlx5_ib_flow_type,
2813 			    UA_MANDATORY),
2814 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA,
2815 			   UVERBS_ATTR_TYPE(u8),
2816 			   UA_MANDATORY),
2817 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2818 			     enum ib_flow_flags,
2819 			     UA_OPTIONAL),
2820 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE,
2821 			     enum mlx5_ib_uapi_flow_table_type,
2822 			     UA_OPTIONAL));
2823 
2824 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2825 	MLX5_IB_METHOD_FLOW_MATCHER_DESTROY,
2826 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_DESTROY_HANDLE,
2827 			MLX5_IB_OBJECT_FLOW_MATCHER,
2828 			UVERBS_ACCESS_DESTROY,
2829 			UA_MANDATORY));
2830 
2831 DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_FLOW_MATCHER,
2832 			    UVERBS_TYPE_ALLOC_IDR(flow_matcher_cleanup),
2833 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_CREATE),
2834 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_DESTROY));
2835 
2836 DECLARE_UVERBS_NAMED_METHOD(
2837 	MLX5_IB_METHOD_STEERING_ANCHOR_CREATE,
2838 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_STEERING_ANCHOR_CREATE_HANDLE,
2839 			MLX5_IB_OBJECT_STEERING_ANCHOR,
2840 			UVERBS_ACCESS_NEW,
2841 			UA_MANDATORY),
2842 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_STEERING_ANCHOR_FT_TYPE,
2843 			     enum mlx5_ib_uapi_flow_table_type,
2844 			     UA_MANDATORY),
2845 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_STEERING_ANCHOR_PRIORITY,
2846 			   UVERBS_ATTR_TYPE(u16),
2847 			   UA_MANDATORY),
2848 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_STEERING_ANCHOR_FT_ID,
2849 			   UVERBS_ATTR_TYPE(u32),
2850 			   UA_MANDATORY));
2851 
2852 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2853 	MLX5_IB_METHOD_STEERING_ANCHOR_DESTROY,
2854 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_STEERING_ANCHOR_DESTROY_HANDLE,
2855 			MLX5_IB_OBJECT_STEERING_ANCHOR,
2856 			UVERBS_ACCESS_DESTROY,
2857 			UA_MANDATORY));
2858 
2859 DECLARE_UVERBS_NAMED_OBJECT(
2860 	MLX5_IB_OBJECT_STEERING_ANCHOR,
2861 	UVERBS_TYPE_ALLOC_IDR(steering_anchor_cleanup),
2862 	&UVERBS_METHOD(MLX5_IB_METHOD_STEERING_ANCHOR_CREATE),
2863 	&UVERBS_METHOD(MLX5_IB_METHOD_STEERING_ANCHOR_DESTROY));
2864 
2865 const struct uapi_definition mlx5_ib_flow_defs[] = {
2866 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
2867 		MLX5_IB_OBJECT_FLOW_MATCHER),
2868 	UAPI_DEF_CHAIN_OBJ_TREE(
2869 		UVERBS_OBJECT_FLOW,
2870 		&mlx5_ib_fs),
2871 	UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_FLOW_ACTION,
2872 				&mlx5_ib_flow_actions),
2873 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
2874 		MLX5_IB_OBJECT_STEERING_ANCHOR,
2875 		UAPI_DEF_IS_OBJ_SUPPORTED(mlx5_ib_shared_ft_allowed)),
2876 	{},
2877 };
2878 
2879 static const struct ib_device_ops flow_ops = {
2880 	.create_flow = mlx5_ib_create_flow,
2881 	.destroy_flow = mlx5_ib_destroy_flow,
2882 	.destroy_flow_action = mlx5_ib_destroy_flow_action,
2883 };
2884 
2885 int mlx5_ib_fs_init(struct mlx5_ib_dev *dev)
2886 {
2887 	dev->flow_db = kzalloc(sizeof(*dev->flow_db), GFP_KERNEL);
2888 
2889 	if (!dev->flow_db)
2890 		return -ENOMEM;
2891 
2892 	mutex_init(&dev->flow_db->lock);
2893 
2894 	ib_set_device_ops(&dev->ib_dev, &flow_ops);
2895 	return 0;
2896 }
2897