xref: /openbmc/linux/drivers/infiniband/hw/mlx5/fs.c (revision c4e78957)
1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /*
3  * Copyright (c) 2018, Mellanox Technologies inc.  All rights reserved.
4  */
5 
6 #include <rdma/ib_user_verbs.h>
7 #include <rdma/ib_verbs.h>
8 #include <rdma/uverbs_types.h>
9 #include <rdma/uverbs_ioctl.h>
10 #include <rdma/uverbs_std_types.h>
11 #include <rdma/mlx5_user_ioctl_cmds.h>
12 #include <rdma/mlx5_user_ioctl_verbs.h>
13 #include <rdma/ib_hdrs.h>
14 #include <rdma/ib_umem.h>
15 #include <linux/mlx5/driver.h>
16 #include <linux/mlx5/fs.h>
17 #include <linux/mlx5/fs_helpers.h>
18 #include <linux/mlx5/eswitch.h>
19 #include <net/inet_ecn.h>
20 #include "mlx5_ib.h"
21 #include "counters.h"
22 #include "devx.h"
23 #include "fs.h"
24 
25 #define UVERBS_MODULE_NAME mlx5_ib
26 #include <rdma/uverbs_named_ioctl.h>
27 
28 enum {
29 	MATCH_CRITERIA_ENABLE_OUTER_BIT,
30 	MATCH_CRITERIA_ENABLE_MISC_BIT,
31 	MATCH_CRITERIA_ENABLE_INNER_BIT,
32 	MATCH_CRITERIA_ENABLE_MISC2_BIT
33 };
34 
35 #define HEADER_IS_ZERO(match_criteria, headers)			           \
36 	!(memchr_inv(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
37 		    0, MLX5_FLD_SZ_BYTES(fte_match_param, headers)))       \
38 
39 static u8 get_match_criteria_enable(u32 *match_criteria)
40 {
41 	u8 match_criteria_enable;
42 
43 	match_criteria_enable =
44 		(!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
45 		MATCH_CRITERIA_ENABLE_OUTER_BIT;
46 	match_criteria_enable |=
47 		(!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
48 		MATCH_CRITERIA_ENABLE_MISC_BIT;
49 	match_criteria_enable |=
50 		(!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
51 		MATCH_CRITERIA_ENABLE_INNER_BIT;
52 	match_criteria_enable |=
53 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
54 		MATCH_CRITERIA_ENABLE_MISC2_BIT;
55 
56 	return match_criteria_enable;
57 }
58 
59 static int set_proto(void *outer_c, void *outer_v, u8 mask, u8 val)
60 {
61 	u8 entry_mask;
62 	u8 entry_val;
63 	int err = 0;
64 
65 	if (!mask)
66 		goto out;
67 
68 	entry_mask = MLX5_GET(fte_match_set_lyr_2_4, outer_c,
69 			      ip_protocol);
70 	entry_val = MLX5_GET(fte_match_set_lyr_2_4, outer_v,
71 			     ip_protocol);
72 	if (!entry_mask) {
73 		MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_protocol, mask);
74 		MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_protocol, val);
75 		goto out;
76 	}
77 	/* Don't override existing ip protocol */
78 	if (mask != entry_mask || val != entry_val)
79 		err = -EINVAL;
80 out:
81 	return err;
82 }
83 
84 static void set_flow_label(void *misc_c, void *misc_v, u32 mask, u32 val,
85 			   bool inner)
86 {
87 	if (inner) {
88 		MLX5_SET(fte_match_set_misc,
89 			 misc_c, inner_ipv6_flow_label, mask);
90 		MLX5_SET(fte_match_set_misc,
91 			 misc_v, inner_ipv6_flow_label, val);
92 	} else {
93 		MLX5_SET(fte_match_set_misc,
94 			 misc_c, outer_ipv6_flow_label, mask);
95 		MLX5_SET(fte_match_set_misc,
96 			 misc_v, outer_ipv6_flow_label, val);
97 	}
98 }
99 
100 static void set_tos(void *outer_c, void *outer_v, u8 mask, u8 val)
101 {
102 	MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_ecn, mask);
103 	MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_ecn, val);
104 	MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_dscp, mask >> 2);
105 	MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_dscp, val >> 2);
106 }
107 
108 static int check_mpls_supp_fields(u32 field_support, const __be32 *set_mask)
109 {
110 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_label) &&
111 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_LABEL))
112 		return -EOPNOTSUPP;
113 
114 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_exp) &&
115 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_EXP))
116 		return -EOPNOTSUPP;
117 
118 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_s_bos) &&
119 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_S_BOS))
120 		return -EOPNOTSUPP;
121 
122 	if (MLX5_GET(fte_match_mpls, set_mask, mpls_ttl) &&
123 	    !(field_support & MLX5_FIELD_SUPPORT_MPLS_TTL))
124 		return -EOPNOTSUPP;
125 
126 	return 0;
127 }
128 
129 #define LAST_ETH_FIELD vlan_tag
130 #define LAST_IB_FIELD sl
131 #define LAST_IPV4_FIELD tos
132 #define LAST_IPV6_FIELD traffic_class
133 #define LAST_TCP_UDP_FIELD src_port
134 #define LAST_TUNNEL_FIELD tunnel_id
135 #define LAST_FLOW_TAG_FIELD tag_id
136 #define LAST_DROP_FIELD size
137 #define LAST_COUNTERS_FIELD counters
138 
139 /* Field is the last supported field */
140 #define FIELDS_NOT_SUPPORTED(filter, field)                                    \
141 	memchr_inv((void *)&filter.field + sizeof(filter.field), 0,            \
142 		   sizeof(filter) - offsetofend(typeof(filter), field))
143 
144 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
145 			   bool is_egress,
146 			   struct mlx5_flow_act *action)
147 {
148 
149 	switch (maction->ib_action.type) {
150 	case IB_FLOW_ACTION_UNSPECIFIED:
151 		if (maction->flow_action_raw.sub_type ==
152 		    MLX5_IB_FLOW_ACTION_MODIFY_HEADER) {
153 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
154 				return -EINVAL;
155 			action->action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
156 			action->modify_hdr =
157 				maction->flow_action_raw.modify_hdr;
158 			return 0;
159 		}
160 		if (maction->flow_action_raw.sub_type ==
161 		    MLX5_IB_FLOW_ACTION_DECAP) {
162 			if (action->action & MLX5_FLOW_CONTEXT_ACTION_DECAP)
163 				return -EINVAL;
164 			action->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
165 			return 0;
166 		}
167 		if (maction->flow_action_raw.sub_type ==
168 		    MLX5_IB_FLOW_ACTION_PACKET_REFORMAT) {
169 			if (action->action &
170 			    MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT)
171 				return -EINVAL;
172 			action->action |=
173 				MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
174 			action->pkt_reformat =
175 				maction->flow_action_raw.pkt_reformat;
176 			return 0;
177 		}
178 		fallthrough;
179 	default:
180 		return -EOPNOTSUPP;
181 	}
182 }
183 
184 static int parse_flow_attr(struct mlx5_core_dev *mdev,
185 			   struct mlx5_flow_spec *spec,
186 			   const union ib_flow_spec *ib_spec,
187 			   const struct ib_flow_attr *flow_attr,
188 			   struct mlx5_flow_act *action, u32 prev_type)
189 {
190 	struct mlx5_flow_context *flow_context = &spec->flow_context;
191 	u32 *match_c = spec->match_criteria;
192 	u32 *match_v = spec->match_value;
193 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param, match_c,
194 					   misc_parameters);
195 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, match_v,
196 					   misc_parameters);
197 	void *misc_params2_c = MLX5_ADDR_OF(fte_match_param, match_c,
198 					    misc_parameters_2);
199 	void *misc_params2_v = MLX5_ADDR_OF(fte_match_param, match_v,
200 					    misc_parameters_2);
201 	void *headers_c;
202 	void *headers_v;
203 	int match_ipv;
204 	int ret;
205 
206 	if (ib_spec->type & IB_FLOW_SPEC_INNER) {
207 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
208 					 inner_headers);
209 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
210 					 inner_headers);
211 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
212 					ft_field_support.inner_ip_version);
213 	} else {
214 		headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
215 					 outer_headers);
216 		headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
217 					 outer_headers);
218 		match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
219 					ft_field_support.outer_ip_version);
220 	}
221 
222 	switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
223 	case IB_FLOW_SPEC_ETH:
224 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
225 			return -EOPNOTSUPP;
226 
227 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
228 					     dmac_47_16),
229 				ib_spec->eth.mask.dst_mac);
230 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
231 					     dmac_47_16),
232 				ib_spec->eth.val.dst_mac);
233 
234 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
235 					     smac_47_16),
236 				ib_spec->eth.mask.src_mac);
237 		ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
238 					     smac_47_16),
239 				ib_spec->eth.val.src_mac);
240 
241 		if (ib_spec->eth.mask.vlan_tag) {
242 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
243 				 cvlan_tag, 1);
244 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
245 				 cvlan_tag, 1);
246 
247 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
248 				 first_vid, ntohs(ib_spec->eth.mask.vlan_tag));
249 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
250 				 first_vid, ntohs(ib_spec->eth.val.vlan_tag));
251 
252 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
253 				 first_cfi,
254 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 12);
255 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
256 				 first_cfi,
257 				 ntohs(ib_spec->eth.val.vlan_tag) >> 12);
258 
259 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
260 				 first_prio,
261 				 ntohs(ib_spec->eth.mask.vlan_tag) >> 13);
262 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
263 				 first_prio,
264 				 ntohs(ib_spec->eth.val.vlan_tag) >> 13);
265 		}
266 		MLX5_SET(fte_match_set_lyr_2_4, headers_c,
267 			 ethertype, ntohs(ib_spec->eth.mask.ether_type));
268 		MLX5_SET(fte_match_set_lyr_2_4, headers_v,
269 			 ethertype, ntohs(ib_spec->eth.val.ether_type));
270 		break;
271 	case IB_FLOW_SPEC_IPV4:
272 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
273 			return -EOPNOTSUPP;
274 
275 		if (match_ipv) {
276 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
277 				 ip_version, 0xf);
278 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
279 				 ip_version, MLX5_FS_IPV4_VERSION);
280 		} else {
281 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
282 				 ethertype, 0xffff);
283 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
284 				 ethertype, ETH_P_IP);
285 		}
286 
287 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
288 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
289 		       &ib_spec->ipv4.mask.src_ip,
290 		       sizeof(ib_spec->ipv4.mask.src_ip));
291 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
292 				    src_ipv4_src_ipv6.ipv4_layout.ipv4),
293 		       &ib_spec->ipv4.val.src_ip,
294 		       sizeof(ib_spec->ipv4.val.src_ip));
295 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
296 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
297 		       &ib_spec->ipv4.mask.dst_ip,
298 		       sizeof(ib_spec->ipv4.mask.dst_ip));
299 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
300 				    dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
301 		       &ib_spec->ipv4.val.dst_ip,
302 		       sizeof(ib_spec->ipv4.val.dst_ip));
303 
304 		set_tos(headers_c, headers_v,
305 			ib_spec->ipv4.mask.tos, ib_spec->ipv4.val.tos);
306 
307 		if (set_proto(headers_c, headers_v,
308 			      ib_spec->ipv4.mask.proto,
309 			      ib_spec->ipv4.val.proto))
310 			return -EINVAL;
311 		break;
312 	case IB_FLOW_SPEC_IPV6:
313 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv6.mask, LAST_IPV6_FIELD))
314 			return -EOPNOTSUPP;
315 
316 		if (match_ipv) {
317 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
318 				 ip_version, 0xf);
319 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
320 				 ip_version, MLX5_FS_IPV6_VERSION);
321 		} else {
322 			MLX5_SET(fte_match_set_lyr_2_4, headers_c,
323 				 ethertype, 0xffff);
324 			MLX5_SET(fte_match_set_lyr_2_4, headers_v,
325 				 ethertype, ETH_P_IPV6);
326 		}
327 
328 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
329 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
330 		       &ib_spec->ipv6.mask.src_ip,
331 		       sizeof(ib_spec->ipv6.mask.src_ip));
332 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
333 				    src_ipv4_src_ipv6.ipv6_layout.ipv6),
334 		       &ib_spec->ipv6.val.src_ip,
335 		       sizeof(ib_spec->ipv6.val.src_ip));
336 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
337 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
338 		       &ib_spec->ipv6.mask.dst_ip,
339 		       sizeof(ib_spec->ipv6.mask.dst_ip));
340 		memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
341 				    dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
342 		       &ib_spec->ipv6.val.dst_ip,
343 		       sizeof(ib_spec->ipv6.val.dst_ip));
344 
345 		set_tos(headers_c, headers_v,
346 			ib_spec->ipv6.mask.traffic_class,
347 			ib_spec->ipv6.val.traffic_class);
348 
349 		if (set_proto(headers_c, headers_v,
350 			      ib_spec->ipv6.mask.next_hdr,
351 			      ib_spec->ipv6.val.next_hdr))
352 			return -EINVAL;
353 
354 		set_flow_label(misc_params_c, misc_params_v,
355 			       ntohl(ib_spec->ipv6.mask.flow_label),
356 			       ntohl(ib_spec->ipv6.val.flow_label),
357 			       ib_spec->type & IB_FLOW_SPEC_INNER);
358 		break;
359 	case IB_FLOW_SPEC_ESP:
360 		return -EOPNOTSUPP;
361 	case IB_FLOW_SPEC_TCP:
362 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
363 					 LAST_TCP_UDP_FIELD))
364 			return -EOPNOTSUPP;
365 
366 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_TCP))
367 			return -EINVAL;
368 
369 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_sport,
370 			 ntohs(ib_spec->tcp_udp.mask.src_port));
371 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
372 			 ntohs(ib_spec->tcp_udp.val.src_port));
373 
374 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_dport,
375 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
376 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
377 			 ntohs(ib_spec->tcp_udp.val.dst_port));
378 		break;
379 	case IB_FLOW_SPEC_UDP:
380 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
381 					 LAST_TCP_UDP_FIELD))
382 			return -EOPNOTSUPP;
383 
384 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_UDP))
385 			return -EINVAL;
386 
387 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_sport,
388 			 ntohs(ib_spec->tcp_udp.mask.src_port));
389 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
390 			 ntohs(ib_spec->tcp_udp.val.src_port));
391 
392 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_dport,
393 			 ntohs(ib_spec->tcp_udp.mask.dst_port));
394 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
395 			 ntohs(ib_spec->tcp_udp.val.dst_port));
396 		break;
397 	case IB_FLOW_SPEC_GRE:
398 		if (ib_spec->gre.mask.c_ks_res0_ver)
399 			return -EOPNOTSUPP;
400 
401 		if (set_proto(headers_c, headers_v, 0xff, IPPROTO_GRE))
402 			return -EINVAL;
403 
404 		MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
405 			 0xff);
406 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
407 			 IPPROTO_GRE);
408 
409 		MLX5_SET(fte_match_set_misc, misc_params_c, gre_protocol,
410 			 ntohs(ib_spec->gre.mask.protocol));
411 		MLX5_SET(fte_match_set_misc, misc_params_v, gre_protocol,
412 			 ntohs(ib_spec->gre.val.protocol));
413 
414 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_c,
415 				    gre_key.nvgre.hi),
416 		       &ib_spec->gre.mask.key,
417 		       sizeof(ib_spec->gre.mask.key));
418 		memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_v,
419 				    gre_key.nvgre.hi),
420 		       &ib_spec->gre.val.key,
421 		       sizeof(ib_spec->gre.val.key));
422 		break;
423 	case IB_FLOW_SPEC_MPLS:
424 		switch (prev_type) {
425 		case IB_FLOW_SPEC_UDP:
426 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
427 						   ft_field_support.outer_first_mpls_over_udp),
428 						   &ib_spec->mpls.mask.tag))
429 				return -EOPNOTSUPP;
430 
431 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
432 					    outer_first_mpls_over_udp),
433 			       &ib_spec->mpls.val.tag,
434 			       sizeof(ib_spec->mpls.val.tag));
435 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
436 					    outer_first_mpls_over_udp),
437 			       &ib_spec->mpls.mask.tag,
438 			       sizeof(ib_spec->mpls.mask.tag));
439 			break;
440 		case IB_FLOW_SPEC_GRE:
441 			if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
442 						   ft_field_support.outer_first_mpls_over_gre),
443 						   &ib_spec->mpls.mask.tag))
444 				return -EOPNOTSUPP;
445 
446 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
447 					    outer_first_mpls_over_gre),
448 			       &ib_spec->mpls.val.tag,
449 			       sizeof(ib_spec->mpls.val.tag));
450 			memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
451 					    outer_first_mpls_over_gre),
452 			       &ib_spec->mpls.mask.tag,
453 			       sizeof(ib_spec->mpls.mask.tag));
454 			break;
455 		default:
456 			if (ib_spec->type & IB_FLOW_SPEC_INNER) {
457 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
458 							   ft_field_support.inner_first_mpls),
459 							   &ib_spec->mpls.mask.tag))
460 					return -EOPNOTSUPP;
461 
462 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
463 						    inner_first_mpls),
464 				       &ib_spec->mpls.val.tag,
465 				       sizeof(ib_spec->mpls.val.tag));
466 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
467 						    inner_first_mpls),
468 				       &ib_spec->mpls.mask.tag,
469 				       sizeof(ib_spec->mpls.mask.tag));
470 			} else {
471 				if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
472 							   ft_field_support.outer_first_mpls),
473 							   &ib_spec->mpls.mask.tag))
474 					return -EOPNOTSUPP;
475 
476 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
477 						    outer_first_mpls),
478 				       &ib_spec->mpls.val.tag,
479 				       sizeof(ib_spec->mpls.val.tag));
480 				memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
481 						    outer_first_mpls),
482 				       &ib_spec->mpls.mask.tag,
483 				       sizeof(ib_spec->mpls.mask.tag));
484 			}
485 		}
486 		break;
487 	case IB_FLOW_SPEC_VXLAN_TUNNEL:
488 		if (FIELDS_NOT_SUPPORTED(ib_spec->tunnel.mask,
489 					 LAST_TUNNEL_FIELD))
490 			return -EOPNOTSUPP;
491 
492 		MLX5_SET(fte_match_set_misc, misc_params_c, vxlan_vni,
493 			 ntohl(ib_spec->tunnel.mask.tunnel_id));
494 		MLX5_SET(fte_match_set_misc, misc_params_v, vxlan_vni,
495 			 ntohl(ib_spec->tunnel.val.tunnel_id));
496 		break;
497 	case IB_FLOW_SPEC_ACTION_TAG:
498 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_tag,
499 					 LAST_FLOW_TAG_FIELD))
500 			return -EOPNOTSUPP;
501 		if (ib_spec->flow_tag.tag_id >= BIT(24))
502 			return -EINVAL;
503 
504 		flow_context->flow_tag = ib_spec->flow_tag.tag_id;
505 		flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
506 		break;
507 	case IB_FLOW_SPEC_ACTION_DROP:
508 		if (FIELDS_NOT_SUPPORTED(ib_spec->drop,
509 					 LAST_DROP_FIELD))
510 			return -EOPNOTSUPP;
511 		action->action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
512 		break;
513 	case IB_FLOW_SPEC_ACTION_HANDLE:
514 		ret = parse_flow_flow_action(to_mflow_act(ib_spec->action.act),
515 			flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS, action);
516 		if (ret)
517 			return ret;
518 		break;
519 	case IB_FLOW_SPEC_ACTION_COUNT:
520 		if (FIELDS_NOT_SUPPORTED(ib_spec->flow_count,
521 					 LAST_COUNTERS_FIELD))
522 			return -EOPNOTSUPP;
523 
524 		/* for now support only one counters spec per flow */
525 		if (action->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
526 			return -EINVAL;
527 
528 		action->counters = ib_spec->flow_count.counters;
529 		action->action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
530 		break;
531 	default:
532 		return -EINVAL;
533 	}
534 
535 	return 0;
536 }
537 
538 /* If a flow could catch both multicast and unicast packets,
539  * it won't fall into the multicast flow steering table and this rule
540  * could steal other multicast packets.
541  */
542 static bool flow_is_multicast_only(const struct ib_flow_attr *ib_attr)
543 {
544 	union ib_flow_spec *flow_spec;
545 
546 	if (ib_attr->type != IB_FLOW_ATTR_NORMAL ||
547 	    ib_attr->num_of_specs < 1)
548 		return false;
549 
550 	flow_spec = (union ib_flow_spec *)(ib_attr + 1);
551 	if (flow_spec->type == IB_FLOW_SPEC_IPV4) {
552 		struct ib_flow_spec_ipv4 *ipv4_spec;
553 
554 		ipv4_spec = (struct ib_flow_spec_ipv4 *)flow_spec;
555 		if (ipv4_is_multicast(ipv4_spec->val.dst_ip))
556 			return true;
557 
558 		return false;
559 	}
560 
561 	if (flow_spec->type == IB_FLOW_SPEC_ETH) {
562 		struct ib_flow_spec_eth *eth_spec;
563 
564 		eth_spec = (struct ib_flow_spec_eth *)flow_spec;
565 		return is_multicast_ether_addr(eth_spec->mask.dst_mac) &&
566 		       is_multicast_ether_addr(eth_spec->val.dst_mac);
567 	}
568 
569 	return false;
570 }
571 
572 static bool is_valid_ethertype(struct mlx5_core_dev *mdev,
573 			       const struct ib_flow_attr *flow_attr,
574 			       bool check_inner)
575 {
576 	union ib_flow_spec *ib_spec = (union ib_flow_spec *)(flow_attr + 1);
577 	int match_ipv = check_inner ?
578 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
579 					ft_field_support.inner_ip_version) :
580 			MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
581 					ft_field_support.outer_ip_version);
582 	int inner_bit = check_inner ? IB_FLOW_SPEC_INNER : 0;
583 	bool ipv4_spec_valid, ipv6_spec_valid;
584 	unsigned int ip_spec_type = 0;
585 	bool has_ethertype = false;
586 	unsigned int spec_index;
587 	bool mask_valid = true;
588 	u16 eth_type = 0;
589 	bool type_valid;
590 
591 	/* Validate that ethertype is correct */
592 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
593 		if ((ib_spec->type == (IB_FLOW_SPEC_ETH | inner_bit)) &&
594 		    ib_spec->eth.mask.ether_type) {
595 			mask_valid = (ib_spec->eth.mask.ether_type ==
596 				      htons(0xffff));
597 			has_ethertype = true;
598 			eth_type = ntohs(ib_spec->eth.val.ether_type);
599 		} else if ((ib_spec->type == (IB_FLOW_SPEC_IPV4 | inner_bit)) ||
600 			   (ib_spec->type == (IB_FLOW_SPEC_IPV6 | inner_bit))) {
601 			ip_spec_type = ib_spec->type;
602 		}
603 		ib_spec = (void *)ib_spec + ib_spec->size;
604 	}
605 
606 	type_valid = (!has_ethertype) || (!ip_spec_type);
607 	if (!type_valid && mask_valid) {
608 		ipv4_spec_valid = (eth_type == ETH_P_IP) &&
609 			(ip_spec_type == (IB_FLOW_SPEC_IPV4 | inner_bit));
610 		ipv6_spec_valid = (eth_type == ETH_P_IPV6) &&
611 			(ip_spec_type == (IB_FLOW_SPEC_IPV6 | inner_bit));
612 
613 		type_valid = (ipv4_spec_valid) || (ipv6_spec_valid) ||
614 			     (((eth_type == ETH_P_MPLS_UC) ||
615 			       (eth_type == ETH_P_MPLS_MC)) && match_ipv);
616 	}
617 
618 	return type_valid;
619 }
620 
621 static bool is_valid_attr(struct mlx5_core_dev *mdev,
622 			  const struct ib_flow_attr *flow_attr)
623 {
624 	return is_valid_ethertype(mdev, flow_attr, false) &&
625 	       is_valid_ethertype(mdev, flow_attr, true);
626 }
627 
628 static void put_flow_table(struct mlx5_ib_dev *dev,
629 			   struct mlx5_ib_flow_prio *prio, bool ft_added)
630 {
631 	prio->refcount -= !!ft_added;
632 	if (!prio->refcount) {
633 		mlx5_destroy_flow_table(prio->flow_table);
634 		prio->flow_table = NULL;
635 	}
636 }
637 
638 static int mlx5_ib_destroy_flow(struct ib_flow *flow_id)
639 {
640 	struct mlx5_ib_flow_handler *handler = container_of(flow_id,
641 							  struct mlx5_ib_flow_handler,
642 							  ibflow);
643 	struct mlx5_ib_flow_handler *iter, *tmp;
644 	struct mlx5_ib_dev *dev = handler->dev;
645 
646 	mutex_lock(&dev->flow_db->lock);
647 
648 	list_for_each_entry_safe(iter, tmp, &handler->list, list) {
649 		mlx5_del_flow_rules(iter->rule);
650 		put_flow_table(dev, iter->prio, true);
651 		list_del(&iter->list);
652 		kfree(iter);
653 	}
654 
655 	mlx5_del_flow_rules(handler->rule);
656 	put_flow_table(dev, handler->prio, true);
657 	mlx5_ib_counters_clear_description(handler->ibcounters);
658 	mutex_unlock(&dev->flow_db->lock);
659 	if (handler->flow_matcher)
660 		atomic_dec(&handler->flow_matcher->usecnt);
661 	kfree(handler);
662 
663 	return 0;
664 }
665 
666 static int ib_prio_to_core_prio(unsigned int priority, bool dont_trap)
667 {
668 	priority *= 2;
669 	if (!dont_trap)
670 		priority++;
671 	return priority;
672 }
673 
674 enum flow_table_type {
675 	MLX5_IB_FT_RX,
676 	MLX5_IB_FT_TX
677 };
678 
679 #define MLX5_FS_MAX_TYPES	 6
680 #define MLX5_FS_MAX_ENTRIES	 BIT(16)
681 
682 static struct mlx5_ib_flow_prio *_get_prio(struct mlx5_flow_namespace *ns,
683 					   struct mlx5_ib_flow_prio *prio,
684 					   int priority,
685 					   int num_entries, int num_groups,
686 					   u32 flags)
687 {
688 	struct mlx5_flow_table_attr ft_attr = {};
689 	struct mlx5_flow_table *ft;
690 
691 	ft_attr.prio = priority;
692 	ft_attr.max_fte = num_entries;
693 	ft_attr.flags = flags;
694 	ft_attr.autogroup.max_num_groups = num_groups;
695 	ft = mlx5_create_auto_grouped_flow_table(ns, &ft_attr);
696 	if (IS_ERR(ft))
697 		return ERR_CAST(ft);
698 
699 	prio->flow_table = ft;
700 	prio->refcount = 0;
701 	return prio;
702 }
703 
704 static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev,
705 						struct ib_flow_attr *flow_attr,
706 						enum flow_table_type ft_type)
707 {
708 	bool dont_trap = flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP;
709 	struct mlx5_flow_namespace *ns = NULL;
710 	enum mlx5_flow_namespace_type fn_type;
711 	struct mlx5_ib_flow_prio *prio;
712 	struct mlx5_flow_table *ft;
713 	int max_table_size;
714 	int num_entries;
715 	int num_groups;
716 	bool esw_encap;
717 	u32 flags = 0;
718 	int priority;
719 
720 	max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
721 						       log_max_ft_size));
722 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
723 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
724 	switch (flow_attr->type) {
725 	case IB_FLOW_ATTR_NORMAL:
726 		if (flow_is_multicast_only(flow_attr) && !dont_trap)
727 			priority = MLX5_IB_FLOW_MCAST_PRIO;
728 		else
729 			priority = ib_prio_to_core_prio(flow_attr->priority,
730 							dont_trap);
731 		if (ft_type == MLX5_IB_FT_RX) {
732 			fn_type = MLX5_FLOW_NAMESPACE_BYPASS;
733 			prio = &dev->flow_db->prios[priority];
734 			if (!dev->is_rep && !esw_encap &&
735 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap))
736 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
737 			if (!dev->is_rep && !esw_encap &&
738 			    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
739 						      reformat_l3_tunnel_to_l2))
740 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
741 		} else {
742 			max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_TX(
743 				dev->mdev, log_max_ft_size));
744 			fn_type = MLX5_FLOW_NAMESPACE_EGRESS;
745 			prio = &dev->flow_db->egress_prios[priority];
746 			if (!dev->is_rep && !esw_encap &&
747 			    MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat))
748 				flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
749 		}
750 		ns = mlx5_get_flow_namespace(dev->mdev, fn_type);
751 		num_entries = MLX5_FS_MAX_ENTRIES;
752 		num_groups = MLX5_FS_MAX_TYPES;
753 		break;
754 	case IB_FLOW_ATTR_ALL_DEFAULT:
755 	case IB_FLOW_ATTR_MC_DEFAULT:
756 		ns = mlx5_get_flow_namespace(dev->mdev,
757 					     MLX5_FLOW_NAMESPACE_LEFTOVERS);
758 		build_leftovers_ft_param(&priority, &num_entries, &num_groups);
759 		prio = &dev->flow_db->prios[MLX5_IB_FLOW_LEFTOVERS_PRIO];
760 		break;
761 	case IB_FLOW_ATTR_SNIFFER:
762 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
763 					allow_sniffer_and_nic_rx_shared_tir))
764 			return ERR_PTR(-EOPNOTSUPP);
765 
766 		ns = mlx5_get_flow_namespace(
767 			dev->mdev, ft_type == MLX5_IB_FT_RX ?
768 					   MLX5_FLOW_NAMESPACE_SNIFFER_RX :
769 					   MLX5_FLOW_NAMESPACE_SNIFFER_TX);
770 
771 		prio = &dev->flow_db->sniffer[ft_type];
772 		priority = 0;
773 		num_entries = 1;
774 		num_groups = 1;
775 		break;
776 	default:
777 		break;
778 	}
779 
780 	if (!ns)
781 		return ERR_PTR(-EOPNOTSUPP);
782 
783 	max_table_size = min_t(int, num_entries, max_table_size);
784 
785 	ft = prio->flow_table;
786 	if (!ft)
787 		return _get_prio(ns, prio, priority, max_table_size, num_groups,
788 				 flags);
789 
790 	return prio;
791 }
792 
793 enum {
794 	RDMA_RX_ECN_OPCOUNTER_PRIO,
795 	RDMA_RX_CNP_OPCOUNTER_PRIO,
796 };
797 
798 enum {
799 	RDMA_TX_CNP_OPCOUNTER_PRIO,
800 };
801 
802 static int set_vhca_port_spec(struct mlx5_ib_dev *dev, u32 port_num,
803 			      struct mlx5_flow_spec *spec)
804 {
805 	if (!MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev,
806 					ft_field_support.source_vhca_port) ||
807 	    !MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
808 					ft_field_support.source_vhca_port))
809 		return -EOPNOTSUPP;
810 
811 	MLX5_SET_TO_ONES(fte_match_param, &spec->match_criteria,
812 			 misc_parameters.source_vhca_port);
813 	MLX5_SET(fte_match_param, &spec->match_value,
814 		 misc_parameters.source_vhca_port, port_num);
815 
816 	return 0;
817 }
818 
819 static int set_ecn_ce_spec(struct mlx5_ib_dev *dev, u32 port_num,
820 			   struct mlx5_flow_spec *spec, int ipv)
821 {
822 	if (!MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev,
823 					ft_field_support.outer_ip_version))
824 		return -EOPNOTSUPP;
825 
826 	if (mlx5_core_mp_enabled(dev->mdev) &&
827 	    set_vhca_port_spec(dev, port_num, spec))
828 		return -EOPNOTSUPP;
829 
830 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
831 			 outer_headers.ip_ecn);
832 	MLX5_SET(fte_match_param, spec->match_value, outer_headers.ip_ecn,
833 		 INET_ECN_CE);
834 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
835 			 outer_headers.ip_version);
836 	MLX5_SET(fte_match_param, spec->match_value, outer_headers.ip_version,
837 		 ipv);
838 
839 	spec->match_criteria_enable =
840 		get_match_criteria_enable(spec->match_criteria);
841 
842 	return 0;
843 }
844 
845 static int set_cnp_spec(struct mlx5_ib_dev *dev, u32 port_num,
846 			struct mlx5_flow_spec *spec)
847 {
848 	if (mlx5_core_mp_enabled(dev->mdev) &&
849 	    set_vhca_port_spec(dev, port_num, spec))
850 		return -EOPNOTSUPP;
851 
852 	MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria,
853 			 misc_parameters.bth_opcode);
854 	MLX5_SET(fte_match_param, spec->match_value, misc_parameters.bth_opcode,
855 		 IB_BTH_OPCODE_CNP);
856 
857 	spec->match_criteria_enable =
858 		get_match_criteria_enable(spec->match_criteria);
859 
860 	return 0;
861 }
862 
863 int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num,
864 			 struct mlx5_ib_op_fc *opfc,
865 			 enum mlx5_ib_optional_counter_type type)
866 {
867 	enum mlx5_flow_namespace_type fn_type;
868 	int priority, i, err, spec_num;
869 	struct mlx5_flow_act flow_act = {};
870 	struct mlx5_flow_destination dst;
871 	struct mlx5_flow_namespace *ns;
872 	struct mlx5_ib_flow_prio *prio;
873 	struct mlx5_flow_spec *spec;
874 
875 	spec = kcalloc(MAX_OPFC_RULES, sizeof(*spec), GFP_KERNEL);
876 	if (!spec)
877 		return -ENOMEM;
878 
879 	switch (type) {
880 	case MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS:
881 		if (set_ecn_ce_spec(dev, port_num, &spec[0],
882 				    MLX5_FS_IPV4_VERSION) ||
883 		    set_ecn_ce_spec(dev, port_num, &spec[1],
884 				    MLX5_FS_IPV6_VERSION)) {
885 			err = -EOPNOTSUPP;
886 			goto free;
887 		}
888 		spec_num = 2;
889 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS;
890 		priority = RDMA_RX_ECN_OPCOUNTER_PRIO;
891 		break;
892 
893 	case MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS:
894 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
895 					ft_field_support_2_nic_receive_rdma.bth_opcode) ||
896 		    set_cnp_spec(dev, port_num, &spec[0])) {
897 			err = -EOPNOTSUPP;
898 			goto free;
899 		}
900 		spec_num = 1;
901 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS;
902 		priority = RDMA_RX_CNP_OPCOUNTER_PRIO;
903 		break;
904 
905 	case MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS:
906 		if (!MLX5_CAP_FLOWTABLE(dev->mdev,
907 					ft_field_support_2_nic_transmit_rdma.bth_opcode) ||
908 		    set_cnp_spec(dev, port_num, &spec[0])) {
909 			err = -EOPNOTSUPP;
910 			goto free;
911 		}
912 		spec_num = 1;
913 		fn_type = MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS;
914 		priority = RDMA_TX_CNP_OPCOUNTER_PRIO;
915 		break;
916 
917 	default:
918 		err = -EOPNOTSUPP;
919 		goto free;
920 	}
921 
922 	ns = mlx5_get_flow_namespace(dev->mdev, fn_type);
923 	if (!ns) {
924 		err = -EOPNOTSUPP;
925 		goto free;
926 	}
927 
928 	prio = &dev->flow_db->opfcs[type];
929 	if (!prio->flow_table) {
930 		prio = _get_prio(ns, prio, priority,
931 				 dev->num_ports * MAX_OPFC_RULES, 1, 0);
932 		if (IS_ERR(prio)) {
933 			err = PTR_ERR(prio);
934 			goto free;
935 		}
936 	}
937 
938 	dst.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
939 	dst.counter_id = mlx5_fc_id(opfc->fc);
940 
941 	flow_act.action =
942 		MLX5_FLOW_CONTEXT_ACTION_COUNT | MLX5_FLOW_CONTEXT_ACTION_ALLOW;
943 
944 	for (i = 0; i < spec_num; i++) {
945 		opfc->rule[i] = mlx5_add_flow_rules(prio->flow_table, &spec[i],
946 						    &flow_act, &dst, 1);
947 		if (IS_ERR(opfc->rule[i])) {
948 			err = PTR_ERR(opfc->rule[i]);
949 			goto del_rules;
950 		}
951 	}
952 	prio->refcount += spec_num;
953 	kfree(spec);
954 
955 	return 0;
956 
957 del_rules:
958 	for (i -= 1; i >= 0; i--)
959 		mlx5_del_flow_rules(opfc->rule[i]);
960 	put_flow_table(dev, prio, false);
961 free:
962 	kfree(spec);
963 	return err;
964 }
965 
966 void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev,
967 			     struct mlx5_ib_op_fc *opfc,
968 			     enum mlx5_ib_optional_counter_type type)
969 {
970 	int i;
971 
972 	for (i = 0; i < MAX_OPFC_RULES && opfc->rule[i]; i++) {
973 		mlx5_del_flow_rules(opfc->rule[i]);
974 		put_flow_table(dev, &dev->flow_db->opfcs[type], true);
975 	}
976 }
977 
978 static void set_underlay_qp(struct mlx5_ib_dev *dev,
979 			    struct mlx5_flow_spec *spec,
980 			    u32 underlay_qpn)
981 {
982 	void *misc_params_c = MLX5_ADDR_OF(fte_match_param,
983 					   spec->match_criteria,
984 					   misc_parameters);
985 	void *misc_params_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
986 					   misc_parameters);
987 
988 	if (underlay_qpn &&
989 	    MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
990 				      ft_field_support.bth_dst_qp)) {
991 		MLX5_SET(fte_match_set_misc,
992 			 misc_params_v, bth_dst_qp, underlay_qpn);
993 		MLX5_SET(fte_match_set_misc,
994 			 misc_params_c, bth_dst_qp, 0xffffff);
995 	}
996 }
997 
998 static void mlx5_ib_set_rule_source_port(struct mlx5_ib_dev *dev,
999 					 struct mlx5_flow_spec *spec,
1000 					 struct mlx5_eswitch_rep *rep)
1001 {
1002 	struct mlx5_eswitch *esw = dev->mdev->priv.eswitch;
1003 	void *misc;
1004 
1005 	if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1006 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1007 				    misc_parameters_2);
1008 
1009 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1010 			 mlx5_eswitch_get_vport_metadata_for_match(rep->esw,
1011 								   rep->vport));
1012 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1013 				    misc_parameters_2);
1014 
1015 		MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1016 			 mlx5_eswitch_get_vport_metadata_mask());
1017 	} else {
1018 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1019 				    misc_parameters);
1020 
1021 		MLX5_SET(fte_match_set_misc, misc, source_port, rep->vport);
1022 
1023 		misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1024 				    misc_parameters);
1025 
1026 		MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
1027 	}
1028 }
1029 
1030 static struct mlx5_ib_flow_handler *_create_flow_rule(struct mlx5_ib_dev *dev,
1031 						      struct mlx5_ib_flow_prio *ft_prio,
1032 						      const struct ib_flow_attr *flow_attr,
1033 						      struct mlx5_flow_destination *dst,
1034 						      u32 underlay_qpn,
1035 						      struct mlx5_ib_create_flow *ucmd)
1036 {
1037 	struct mlx5_flow_table	*ft = ft_prio->flow_table;
1038 	struct mlx5_ib_flow_handler *handler;
1039 	struct mlx5_flow_act flow_act = {};
1040 	struct mlx5_flow_spec *spec;
1041 	struct mlx5_flow_destination dest_arr[2] = {};
1042 	struct mlx5_flow_destination *rule_dst = dest_arr;
1043 	const void *ib_flow = (const void *)flow_attr + sizeof(*flow_attr);
1044 	unsigned int spec_index;
1045 	u32 prev_type = 0;
1046 	int err = 0;
1047 	int dest_num = 0;
1048 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
1049 
1050 	if (!is_valid_attr(dev->mdev, flow_attr))
1051 		return ERR_PTR(-EINVAL);
1052 
1053 	if (dev->is_rep && is_egress)
1054 		return ERR_PTR(-EINVAL);
1055 
1056 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1057 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
1058 	if (!handler || !spec) {
1059 		err = -ENOMEM;
1060 		goto free;
1061 	}
1062 
1063 	INIT_LIST_HEAD(&handler->list);
1064 
1065 	for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
1066 		err = parse_flow_attr(dev->mdev, spec,
1067 				      ib_flow, flow_attr, &flow_act,
1068 				      prev_type);
1069 		if (err < 0)
1070 			goto free;
1071 
1072 		prev_type = ((union ib_flow_spec *)ib_flow)->type;
1073 		ib_flow += ((union ib_flow_spec *)ib_flow)->size;
1074 	}
1075 
1076 	if (dst && !(flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP)) {
1077 		memcpy(&dest_arr[0], dst, sizeof(*dst));
1078 		dest_num++;
1079 	}
1080 
1081 	if (!flow_is_multicast_only(flow_attr))
1082 		set_underlay_qp(dev, spec, underlay_qpn);
1083 
1084 	if (dev->is_rep && flow_attr->type != IB_FLOW_ATTR_SNIFFER) {
1085 		struct mlx5_eswitch_rep *rep;
1086 
1087 		rep = dev->port[flow_attr->port - 1].rep;
1088 		if (!rep) {
1089 			err = -EINVAL;
1090 			goto free;
1091 		}
1092 
1093 		mlx5_ib_set_rule_source_port(dev, spec, rep);
1094 	}
1095 
1096 	spec->match_criteria_enable = get_match_criteria_enable(spec->match_criteria);
1097 
1098 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1099 		struct mlx5_ib_mcounters *mcounters;
1100 
1101 		err = mlx5_ib_flow_counters_set_data(flow_act.counters, ucmd);
1102 		if (err)
1103 			goto free;
1104 
1105 		mcounters = to_mcounters(flow_act.counters);
1106 		handler->ibcounters = flow_act.counters;
1107 		dest_arr[dest_num].type =
1108 			MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1109 		dest_arr[dest_num].counter_id =
1110 			mlx5_fc_id(mcounters->hw_cntrs_hndl);
1111 		dest_num++;
1112 	}
1113 
1114 	if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP) {
1115 		if (!dest_num)
1116 			rule_dst = NULL;
1117 	} else {
1118 		if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1119 			flow_act.action |=
1120 				MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1121 		if (is_egress)
1122 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1123 		else if (dest_num)
1124 			flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1125 	}
1126 
1127 	if ((spec->flow_context.flags & FLOW_CONTEXT_HAS_TAG)  &&
1128 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1129 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1130 		mlx5_ib_warn(dev, "Flow tag %u and attribute type %x isn't allowed in leftovers\n",
1131 			     spec->flow_context.flow_tag, flow_attr->type);
1132 		err = -EINVAL;
1133 		goto free;
1134 	}
1135 	handler->rule = mlx5_add_flow_rules(ft, spec,
1136 					    &flow_act,
1137 					    rule_dst, dest_num);
1138 
1139 	if (IS_ERR(handler->rule)) {
1140 		err = PTR_ERR(handler->rule);
1141 		goto free;
1142 	}
1143 
1144 	ft_prio->refcount++;
1145 	handler->prio = ft_prio;
1146 	handler->dev = dev;
1147 
1148 	ft_prio->flow_table = ft;
1149 free:
1150 	if (err && handler) {
1151 		mlx5_ib_counters_clear_description(handler->ibcounters);
1152 		kfree(handler);
1153 	}
1154 	kvfree(spec);
1155 	return err ? ERR_PTR(err) : handler;
1156 }
1157 
1158 static struct mlx5_ib_flow_handler *create_flow_rule(struct mlx5_ib_dev *dev,
1159 						     struct mlx5_ib_flow_prio *ft_prio,
1160 						     const struct ib_flow_attr *flow_attr,
1161 						     struct mlx5_flow_destination *dst)
1162 {
1163 	return _create_flow_rule(dev, ft_prio, flow_attr, dst, 0, NULL);
1164 }
1165 
1166 enum {
1167 	LEFTOVERS_MC,
1168 	LEFTOVERS_UC,
1169 };
1170 
1171 static struct mlx5_ib_flow_handler *create_leftovers_rule(struct mlx5_ib_dev *dev,
1172 							  struct mlx5_ib_flow_prio *ft_prio,
1173 							  struct ib_flow_attr *flow_attr,
1174 							  struct mlx5_flow_destination *dst)
1175 {
1176 	struct mlx5_ib_flow_handler *handler_ucast = NULL;
1177 	struct mlx5_ib_flow_handler *handler = NULL;
1178 
1179 	static struct {
1180 		struct ib_flow_attr	flow_attr;
1181 		struct ib_flow_spec_eth eth_flow;
1182 	} leftovers_specs[] = {
1183 		[LEFTOVERS_MC] = {
1184 			.flow_attr = {
1185 				.num_of_specs = 1,
1186 				.size = sizeof(leftovers_specs[0])
1187 			},
1188 			.eth_flow = {
1189 				.type = IB_FLOW_SPEC_ETH,
1190 				.size = sizeof(struct ib_flow_spec_eth),
1191 				.mask = {.dst_mac = {0x1} },
1192 				.val =  {.dst_mac = {0x1} }
1193 			}
1194 		},
1195 		[LEFTOVERS_UC] = {
1196 			.flow_attr = {
1197 				.num_of_specs = 1,
1198 				.size = sizeof(leftovers_specs[0])
1199 			},
1200 			.eth_flow = {
1201 				.type = IB_FLOW_SPEC_ETH,
1202 				.size = sizeof(struct ib_flow_spec_eth),
1203 				.mask = {.dst_mac = {0x1} },
1204 				.val = {.dst_mac = {} }
1205 			}
1206 		}
1207 	};
1208 
1209 	handler = create_flow_rule(dev, ft_prio,
1210 				   &leftovers_specs[LEFTOVERS_MC].flow_attr,
1211 				   dst);
1212 	if (!IS_ERR(handler) &&
1213 	    flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT) {
1214 		handler_ucast = create_flow_rule(dev, ft_prio,
1215 						 &leftovers_specs[LEFTOVERS_UC].flow_attr,
1216 						 dst);
1217 		if (IS_ERR(handler_ucast)) {
1218 			mlx5_del_flow_rules(handler->rule);
1219 			ft_prio->refcount--;
1220 			kfree(handler);
1221 			handler = handler_ucast;
1222 		} else {
1223 			list_add(&handler_ucast->list, &handler->list);
1224 		}
1225 	}
1226 
1227 	return handler;
1228 }
1229 
1230 static struct mlx5_ib_flow_handler *create_sniffer_rule(struct mlx5_ib_dev *dev,
1231 							struct mlx5_ib_flow_prio *ft_rx,
1232 							struct mlx5_ib_flow_prio *ft_tx,
1233 							struct mlx5_flow_destination *dst)
1234 {
1235 	struct mlx5_ib_flow_handler *handler_rx;
1236 	struct mlx5_ib_flow_handler *handler_tx;
1237 	int err;
1238 	static const struct ib_flow_attr flow_attr  = {
1239 		.num_of_specs = 0,
1240 		.type = IB_FLOW_ATTR_SNIFFER,
1241 		.size = sizeof(flow_attr)
1242 	};
1243 
1244 	handler_rx = create_flow_rule(dev, ft_rx, &flow_attr, dst);
1245 	if (IS_ERR(handler_rx)) {
1246 		err = PTR_ERR(handler_rx);
1247 		goto err;
1248 	}
1249 
1250 	handler_tx = create_flow_rule(dev, ft_tx, &flow_attr, dst);
1251 	if (IS_ERR(handler_tx)) {
1252 		err = PTR_ERR(handler_tx);
1253 		goto err_tx;
1254 	}
1255 
1256 	list_add(&handler_tx->list, &handler_rx->list);
1257 
1258 	return handler_rx;
1259 
1260 err_tx:
1261 	mlx5_del_flow_rules(handler_rx->rule);
1262 	ft_rx->refcount--;
1263 	kfree(handler_rx);
1264 err:
1265 	return ERR_PTR(err);
1266 }
1267 
1268 static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
1269 					   struct ib_flow_attr *flow_attr,
1270 					   struct ib_udata *udata)
1271 {
1272 	struct mlx5_ib_dev *dev = to_mdev(qp->device);
1273 	struct mlx5_ib_qp *mqp = to_mqp(qp);
1274 	struct mlx5_ib_flow_handler *handler = NULL;
1275 	struct mlx5_flow_destination *dst = NULL;
1276 	struct mlx5_ib_flow_prio *ft_prio_tx = NULL;
1277 	struct mlx5_ib_flow_prio *ft_prio;
1278 	bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
1279 	struct mlx5_ib_create_flow *ucmd = NULL, ucmd_hdr;
1280 	size_t min_ucmd_sz, required_ucmd_sz;
1281 	int err;
1282 	int underlay_qpn;
1283 
1284 	if (udata && udata->inlen) {
1285 		min_ucmd_sz = offsetofend(struct mlx5_ib_create_flow, reserved);
1286 		if (udata->inlen < min_ucmd_sz)
1287 			return ERR_PTR(-EOPNOTSUPP);
1288 
1289 		err = ib_copy_from_udata(&ucmd_hdr, udata, min_ucmd_sz);
1290 		if (err)
1291 			return ERR_PTR(err);
1292 
1293 		/* currently supports only one counters data */
1294 		if (ucmd_hdr.ncounters_data > 1)
1295 			return ERR_PTR(-EINVAL);
1296 
1297 		required_ucmd_sz = min_ucmd_sz +
1298 			sizeof(struct mlx5_ib_flow_counters_data) *
1299 			ucmd_hdr.ncounters_data;
1300 		if (udata->inlen > required_ucmd_sz &&
1301 		    !ib_is_udata_cleared(udata, required_ucmd_sz,
1302 					 udata->inlen - required_ucmd_sz))
1303 			return ERR_PTR(-EOPNOTSUPP);
1304 
1305 		ucmd = kzalloc(required_ucmd_sz, GFP_KERNEL);
1306 		if (!ucmd)
1307 			return ERR_PTR(-ENOMEM);
1308 
1309 		err = ib_copy_from_udata(ucmd, udata, required_ucmd_sz);
1310 		if (err)
1311 			goto free_ucmd;
1312 	}
1313 
1314 	if (flow_attr->priority > MLX5_IB_FLOW_LAST_PRIO) {
1315 		err = -ENOMEM;
1316 		goto free_ucmd;
1317 	}
1318 
1319 	if (flow_attr->flags &
1320 	    ~(IB_FLOW_ATTR_FLAGS_DONT_TRAP | IB_FLOW_ATTR_FLAGS_EGRESS)) {
1321 		err = -EINVAL;
1322 		goto free_ucmd;
1323 	}
1324 
1325 	if (is_egress &&
1326 	    (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT ||
1327 	     flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
1328 		err = -EINVAL;
1329 		goto free_ucmd;
1330 	}
1331 
1332 	dst = kzalloc(sizeof(*dst), GFP_KERNEL);
1333 	if (!dst) {
1334 		err = -ENOMEM;
1335 		goto free_ucmd;
1336 	}
1337 
1338 	mutex_lock(&dev->flow_db->lock);
1339 
1340 	ft_prio = get_flow_table(dev, flow_attr,
1341 				 is_egress ? MLX5_IB_FT_TX : MLX5_IB_FT_RX);
1342 	if (IS_ERR(ft_prio)) {
1343 		err = PTR_ERR(ft_prio);
1344 		goto unlock;
1345 	}
1346 	if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
1347 		ft_prio_tx = get_flow_table(dev, flow_attr, MLX5_IB_FT_TX);
1348 		if (IS_ERR(ft_prio_tx)) {
1349 			err = PTR_ERR(ft_prio_tx);
1350 			ft_prio_tx = NULL;
1351 			goto destroy_ft;
1352 		}
1353 	}
1354 
1355 	if (is_egress) {
1356 		dst->type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1357 	} else {
1358 		dst->type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1359 		if (mqp->is_rss)
1360 			dst->tir_num = mqp->rss_qp.tirn;
1361 		else
1362 			dst->tir_num = mqp->raw_packet_qp.rq.tirn;
1363 	}
1364 
1365 	switch (flow_attr->type) {
1366 	case IB_FLOW_ATTR_NORMAL:
1367 		underlay_qpn = (mqp->flags & IB_QP_CREATE_SOURCE_QPN) ?
1368 				       mqp->underlay_qpn :
1369 				       0;
1370 		handler = _create_flow_rule(dev, ft_prio, flow_attr, dst,
1371 					    underlay_qpn, ucmd);
1372 		break;
1373 	case IB_FLOW_ATTR_ALL_DEFAULT:
1374 	case IB_FLOW_ATTR_MC_DEFAULT:
1375 		handler = create_leftovers_rule(dev, ft_prio, flow_attr, dst);
1376 		break;
1377 	case IB_FLOW_ATTR_SNIFFER:
1378 		handler = create_sniffer_rule(dev, ft_prio, ft_prio_tx, dst);
1379 		break;
1380 	default:
1381 		err = -EINVAL;
1382 		goto destroy_ft;
1383 	}
1384 
1385 	if (IS_ERR(handler)) {
1386 		err = PTR_ERR(handler);
1387 		handler = NULL;
1388 		goto destroy_ft;
1389 	}
1390 
1391 	mutex_unlock(&dev->flow_db->lock);
1392 	kfree(dst);
1393 	kfree(ucmd);
1394 
1395 	return &handler->ibflow;
1396 
1397 destroy_ft:
1398 	put_flow_table(dev, ft_prio, false);
1399 	if (ft_prio_tx)
1400 		put_flow_table(dev, ft_prio_tx, false);
1401 unlock:
1402 	mutex_unlock(&dev->flow_db->lock);
1403 	kfree(dst);
1404 free_ucmd:
1405 	kfree(ucmd);
1406 	return ERR_PTR(err);
1407 }
1408 
1409 static struct mlx5_ib_flow_prio *
1410 _get_flow_table(struct mlx5_ib_dev *dev,
1411 		struct mlx5_ib_flow_matcher *fs_matcher,
1412 		bool mcast)
1413 {
1414 	struct mlx5_flow_namespace *ns = NULL;
1415 	struct mlx5_ib_flow_prio *prio = NULL;
1416 	int max_table_size = 0;
1417 	bool esw_encap;
1418 	u32 flags = 0;
1419 	int priority;
1420 
1421 	if (mcast)
1422 		priority = MLX5_IB_FLOW_MCAST_PRIO;
1423 	else
1424 		priority = ib_prio_to_core_prio(fs_matcher->priority, false);
1425 
1426 	esw_encap = mlx5_eswitch_get_encap_mode(dev->mdev) !=
1427 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
1428 	switch (fs_matcher->ns_type) {
1429 	case MLX5_FLOW_NAMESPACE_BYPASS:
1430 		max_table_size = BIT(
1431 			MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, log_max_ft_size));
1432 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, decap) && !esw_encap)
1433 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1434 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
1435 					      reformat_l3_tunnel_to_l2) &&
1436 		    !esw_encap)
1437 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1438 		break;
1439 	case MLX5_FLOW_NAMESPACE_EGRESS:
1440 		max_table_size = BIT(
1441 			MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, log_max_ft_size));
1442 		if (MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, reformat) &&
1443 		    !esw_encap)
1444 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1445 		break;
1446 	case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
1447 		max_table_size = BIT(
1448 			MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, log_max_ft_size));
1449 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev, decap) && esw_encap)
1450 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_DECAP;
1451 		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev->mdev,
1452 					       reformat_l3_tunnel_to_l2) &&
1453 		    esw_encap)
1454 			flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT;
1455 		priority = fs_matcher->priority;
1456 		break;
1457 	case MLX5_FLOW_NAMESPACE_RDMA_RX:
1458 		max_table_size = BIT(
1459 			MLX5_CAP_FLOWTABLE_RDMA_RX(dev->mdev, log_max_ft_size));
1460 		priority = fs_matcher->priority;
1461 		break;
1462 	case MLX5_FLOW_NAMESPACE_RDMA_TX:
1463 		max_table_size = BIT(
1464 			MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev, log_max_ft_size));
1465 		priority = fs_matcher->priority;
1466 		break;
1467 	default:
1468 		break;
1469 	}
1470 
1471 	max_table_size = min_t(int, max_table_size, MLX5_FS_MAX_ENTRIES);
1472 
1473 	ns = mlx5_get_flow_namespace(dev->mdev, fs_matcher->ns_type);
1474 	if (!ns)
1475 		return ERR_PTR(-EOPNOTSUPP);
1476 
1477 	switch (fs_matcher->ns_type) {
1478 	case MLX5_FLOW_NAMESPACE_BYPASS:
1479 		prio = &dev->flow_db->prios[priority];
1480 		break;
1481 	case MLX5_FLOW_NAMESPACE_EGRESS:
1482 		prio = &dev->flow_db->egress_prios[priority];
1483 		break;
1484 	case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
1485 		prio = &dev->flow_db->fdb[priority];
1486 		break;
1487 	case MLX5_FLOW_NAMESPACE_RDMA_RX:
1488 		prio = &dev->flow_db->rdma_rx[priority];
1489 		break;
1490 	case MLX5_FLOW_NAMESPACE_RDMA_TX:
1491 		prio = &dev->flow_db->rdma_tx[priority];
1492 		break;
1493 	default: return ERR_PTR(-EINVAL);
1494 	}
1495 
1496 	if (!prio)
1497 		return ERR_PTR(-EINVAL);
1498 
1499 	if (prio->flow_table)
1500 		return prio;
1501 
1502 	return _get_prio(ns, prio, priority, max_table_size,
1503 			 MLX5_FS_MAX_TYPES, flags);
1504 }
1505 
1506 static struct mlx5_ib_flow_handler *
1507 _create_raw_flow_rule(struct mlx5_ib_dev *dev,
1508 		      struct mlx5_ib_flow_prio *ft_prio,
1509 		      struct mlx5_flow_destination *dst,
1510 		      struct mlx5_ib_flow_matcher  *fs_matcher,
1511 		      struct mlx5_flow_context *flow_context,
1512 		      struct mlx5_flow_act *flow_act,
1513 		      void *cmd_in, int inlen,
1514 		      int dst_num)
1515 {
1516 	struct mlx5_ib_flow_handler *handler;
1517 	struct mlx5_flow_spec *spec;
1518 	struct mlx5_flow_table *ft = ft_prio->flow_table;
1519 	int err = 0;
1520 
1521 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1522 	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
1523 	if (!handler || !spec) {
1524 		err = -ENOMEM;
1525 		goto free;
1526 	}
1527 
1528 	INIT_LIST_HEAD(&handler->list);
1529 
1530 	memcpy(spec->match_value, cmd_in, inlen);
1531 	memcpy(spec->match_criteria, fs_matcher->matcher_mask.match_params,
1532 	       fs_matcher->mask_len);
1533 	spec->match_criteria_enable = fs_matcher->match_criteria_enable;
1534 	spec->flow_context = *flow_context;
1535 
1536 	handler->rule = mlx5_add_flow_rules(ft, spec,
1537 					    flow_act, dst, dst_num);
1538 
1539 	if (IS_ERR(handler->rule)) {
1540 		err = PTR_ERR(handler->rule);
1541 		goto free;
1542 	}
1543 
1544 	ft_prio->refcount++;
1545 	handler->prio = ft_prio;
1546 	handler->dev = dev;
1547 	ft_prio->flow_table = ft;
1548 
1549 free:
1550 	if (err)
1551 		kfree(handler);
1552 	kvfree(spec);
1553 	return err ? ERR_PTR(err) : handler;
1554 }
1555 
1556 static bool raw_fs_is_multicast(struct mlx5_ib_flow_matcher *fs_matcher,
1557 				void *match_v)
1558 {
1559 	void *match_c;
1560 	void *match_v_set_lyr_2_4, *match_c_set_lyr_2_4;
1561 	void *dmac, *dmac_mask;
1562 	void *ipv4, *ipv4_mask;
1563 
1564 	if (!(fs_matcher->match_criteria_enable &
1565 	      (1 << MATCH_CRITERIA_ENABLE_OUTER_BIT)))
1566 		return false;
1567 
1568 	match_c = fs_matcher->matcher_mask.match_params;
1569 	match_v_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_v,
1570 					   outer_headers);
1571 	match_c_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_c,
1572 					   outer_headers);
1573 
1574 	dmac = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1575 			    dmac_47_16);
1576 	dmac_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1577 				 dmac_47_16);
1578 
1579 	if (is_multicast_ether_addr(dmac) &&
1580 	    is_multicast_ether_addr(dmac_mask))
1581 		return true;
1582 
1583 	ipv4 = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
1584 			    dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1585 
1586 	ipv4_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
1587 				 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
1588 
1589 	if (ipv4_is_multicast(*(__be32 *)(ipv4)) &&
1590 	    ipv4_is_multicast(*(__be32 *)(ipv4_mask)))
1591 		return true;
1592 
1593 	return false;
1594 }
1595 
1596 static struct mlx5_ib_flow_handler *raw_fs_rule_add(
1597 	struct mlx5_ib_dev *dev, struct mlx5_ib_flow_matcher *fs_matcher,
1598 	struct mlx5_flow_context *flow_context, struct mlx5_flow_act *flow_act,
1599 	u32 counter_id, void *cmd_in, int inlen, int dest_id, int dest_type)
1600 {
1601 	struct mlx5_flow_destination *dst;
1602 	struct mlx5_ib_flow_prio *ft_prio;
1603 	struct mlx5_ib_flow_handler *handler;
1604 	int dst_num = 0;
1605 	bool mcast;
1606 	int err;
1607 
1608 	if (fs_matcher->flow_type != MLX5_IB_FLOW_TYPE_NORMAL)
1609 		return ERR_PTR(-EOPNOTSUPP);
1610 
1611 	if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO)
1612 		return ERR_PTR(-ENOMEM);
1613 
1614 	dst = kcalloc(2, sizeof(*dst), GFP_KERNEL);
1615 	if (!dst)
1616 		return ERR_PTR(-ENOMEM);
1617 
1618 	mcast = raw_fs_is_multicast(fs_matcher, cmd_in);
1619 	mutex_lock(&dev->flow_db->lock);
1620 
1621 	ft_prio = _get_flow_table(dev, fs_matcher, mcast);
1622 	if (IS_ERR(ft_prio)) {
1623 		err = PTR_ERR(ft_prio);
1624 		goto unlock;
1625 	}
1626 
1627 	switch (dest_type) {
1628 	case MLX5_FLOW_DESTINATION_TYPE_TIR:
1629 		dst[dst_num].type = dest_type;
1630 		dst[dst_num++].tir_num = dest_id;
1631 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1632 		break;
1633 	case MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE:
1634 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM;
1635 		dst[dst_num++].ft_num = dest_id;
1636 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1637 		break;
1638 	case MLX5_FLOW_DESTINATION_TYPE_PORT:
1639 		dst[dst_num++].type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1640 		flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1641 		break;
1642 	default:
1643 		break;
1644 	}
1645 
1646 	if (flow_act->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
1647 		dst[dst_num].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1648 		dst[dst_num].counter_id = counter_id;
1649 		dst_num++;
1650 	}
1651 
1652 	handler = _create_raw_flow_rule(dev, ft_prio, dst_num ? dst : NULL,
1653 					fs_matcher, flow_context, flow_act,
1654 					cmd_in, inlen, dst_num);
1655 
1656 	if (IS_ERR(handler)) {
1657 		err = PTR_ERR(handler);
1658 		goto destroy_ft;
1659 	}
1660 
1661 	mutex_unlock(&dev->flow_db->lock);
1662 	atomic_inc(&fs_matcher->usecnt);
1663 	handler->flow_matcher = fs_matcher;
1664 
1665 	kfree(dst);
1666 
1667 	return handler;
1668 
1669 destroy_ft:
1670 	put_flow_table(dev, ft_prio, false);
1671 unlock:
1672 	mutex_unlock(&dev->flow_db->lock);
1673 	kfree(dst);
1674 
1675 	return ERR_PTR(err);
1676 }
1677 
1678 static void destroy_flow_action_raw(struct mlx5_ib_flow_action *maction)
1679 {
1680 	switch (maction->flow_action_raw.sub_type) {
1681 	case MLX5_IB_FLOW_ACTION_MODIFY_HEADER:
1682 		mlx5_modify_header_dealloc(maction->flow_action_raw.dev->mdev,
1683 					   maction->flow_action_raw.modify_hdr);
1684 		break;
1685 	case MLX5_IB_FLOW_ACTION_PACKET_REFORMAT:
1686 		mlx5_packet_reformat_dealloc(maction->flow_action_raw.dev->mdev,
1687 					     maction->flow_action_raw.pkt_reformat);
1688 		break;
1689 	case MLX5_IB_FLOW_ACTION_DECAP:
1690 		break;
1691 	default:
1692 		break;
1693 	}
1694 }
1695 
1696 static int mlx5_ib_destroy_flow_action(struct ib_flow_action *action)
1697 {
1698 	struct mlx5_ib_flow_action *maction = to_mflow_act(action);
1699 
1700 	switch (action->type) {
1701 	case IB_FLOW_ACTION_UNSPECIFIED:
1702 		destroy_flow_action_raw(maction);
1703 		break;
1704 	default:
1705 		WARN_ON(true);
1706 		break;
1707 	}
1708 
1709 	kfree(maction);
1710 	return 0;
1711 }
1712 
1713 static int
1714 mlx5_ib_ft_type_to_namespace(enum mlx5_ib_uapi_flow_table_type table_type,
1715 			     enum mlx5_flow_namespace_type *namespace)
1716 {
1717 	switch (table_type) {
1718 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX:
1719 		*namespace = MLX5_FLOW_NAMESPACE_BYPASS;
1720 		break;
1721 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX:
1722 		*namespace = MLX5_FLOW_NAMESPACE_EGRESS;
1723 		break;
1724 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_FDB:
1725 		*namespace = MLX5_FLOW_NAMESPACE_FDB_BYPASS;
1726 		break;
1727 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_RX:
1728 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_RX;
1729 		break;
1730 	case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_TX:
1731 		*namespace = MLX5_FLOW_NAMESPACE_RDMA_TX;
1732 		break;
1733 	default:
1734 		return -EINVAL;
1735 	}
1736 
1737 	return 0;
1738 }
1739 
1740 static const struct uverbs_attr_spec mlx5_ib_flow_type[] = {
1741 	[MLX5_IB_FLOW_TYPE_NORMAL] = {
1742 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1743 		.u.ptr = {
1744 			.len = sizeof(u16), /* data is priority */
1745 			.min_len = sizeof(u16),
1746 		}
1747 	},
1748 	[MLX5_IB_FLOW_TYPE_SNIFFER] = {
1749 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1750 		UVERBS_ATTR_NO_DATA(),
1751 	},
1752 	[MLX5_IB_FLOW_TYPE_ALL_DEFAULT] = {
1753 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1754 		UVERBS_ATTR_NO_DATA(),
1755 	},
1756 	[MLX5_IB_FLOW_TYPE_MC_DEFAULT] = {
1757 		.type = UVERBS_ATTR_TYPE_PTR_IN,
1758 		UVERBS_ATTR_NO_DATA(),
1759 	},
1760 };
1761 
1762 static bool is_flow_dest(void *obj, int *dest_id, int *dest_type)
1763 {
1764 	struct devx_obj *devx_obj = obj;
1765 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1766 
1767 	switch (opcode) {
1768 	case MLX5_CMD_OP_DESTROY_TIR:
1769 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1770 		*dest_id = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox,
1771 				    obj_id);
1772 		return true;
1773 
1774 	case MLX5_CMD_OP_DESTROY_FLOW_TABLE:
1775 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1776 		*dest_id = MLX5_GET(destroy_flow_table_in, devx_obj->dinbox,
1777 				    table_id);
1778 		return true;
1779 	default:
1780 		return false;
1781 	}
1782 }
1783 
1784 static int get_dests(struct uverbs_attr_bundle *attrs,
1785 		     struct mlx5_ib_flow_matcher *fs_matcher, int *dest_id,
1786 		     int *dest_type, struct ib_qp **qp, u32 *flags)
1787 {
1788 	bool dest_devx, dest_qp;
1789 	void *devx_obj;
1790 	int err;
1791 
1792 	dest_devx = uverbs_attr_is_valid(attrs,
1793 					 MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1794 	dest_qp = uverbs_attr_is_valid(attrs,
1795 				       MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1796 
1797 	*flags = 0;
1798 	err = uverbs_get_flags32(flags, attrs, MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
1799 				 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS |
1800 					 MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP);
1801 	if (err)
1802 		return err;
1803 
1804 	/* Both flags are not allowed */
1805 	if (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS &&
1806 	    *flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1807 		return -EINVAL;
1808 
1809 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_BYPASS) {
1810 		if (dest_devx && (dest_qp || *flags))
1811 			return -EINVAL;
1812 		else if (dest_qp && *flags)
1813 			return -EINVAL;
1814 	}
1815 
1816 	/* Allow only DEVX object, drop as dest for FDB */
1817 	if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS &&
1818 	    !(dest_devx || (*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)))
1819 		return -EINVAL;
1820 
1821 	/* Allow only DEVX object or QP as dest when inserting to RDMA_RX */
1822 	if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1823 	    ((!dest_devx && !dest_qp) || (dest_devx && dest_qp)))
1824 		return -EINVAL;
1825 
1826 	*qp = NULL;
1827 	if (dest_devx) {
1828 		devx_obj =
1829 			uverbs_attr_get_obj(attrs,
1830 					    MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX);
1831 
1832 		/* Verify that the given DEVX object is a flow
1833 		 * steering destination.
1834 		 */
1835 		if (!is_flow_dest(devx_obj, dest_id, dest_type))
1836 			return -EINVAL;
1837 		/* Allow only flow table as dest when inserting to FDB or RDMA_RX */
1838 		if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS ||
1839 		     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) &&
1840 		    *dest_type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1841 			return -EINVAL;
1842 	} else if (dest_qp) {
1843 		struct mlx5_ib_qp *mqp;
1844 
1845 		*qp = uverbs_attr_get_obj(attrs,
1846 					  MLX5_IB_ATTR_CREATE_FLOW_DEST_QP);
1847 		if (IS_ERR(*qp))
1848 			return PTR_ERR(*qp);
1849 
1850 		if ((*qp)->qp_type != IB_QPT_RAW_PACKET)
1851 			return -EINVAL;
1852 
1853 		mqp = to_mqp(*qp);
1854 		if (mqp->is_rss)
1855 			*dest_id = mqp->rss_qp.tirn;
1856 		else
1857 			*dest_id = mqp->raw_packet_qp.rq.tirn;
1858 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
1859 	} else if ((fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1860 		    fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX) &&
1861 		   !(*flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)) {
1862 		*dest_type = MLX5_FLOW_DESTINATION_TYPE_PORT;
1863 	}
1864 
1865 	if (*dest_type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1866 	    (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_EGRESS ||
1867 	     fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_TX))
1868 		return -EINVAL;
1869 
1870 	return 0;
1871 }
1872 
1873 static bool is_flow_counter(void *obj, u32 offset, u32 *counter_id)
1874 {
1875 	struct devx_obj *devx_obj = obj;
1876 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
1877 
1878 	if (opcode == MLX5_CMD_OP_DEALLOC_FLOW_COUNTER) {
1879 
1880 		if (offset && offset >= devx_obj->flow_counter_bulk_size)
1881 			return false;
1882 
1883 		*counter_id = MLX5_GET(dealloc_flow_counter_in,
1884 				       devx_obj->dinbox,
1885 				       flow_counter_id);
1886 		*counter_id += offset;
1887 		return true;
1888 	}
1889 
1890 	return false;
1891 }
1892 
1893 #define MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS 2
1894 static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)(
1895 	struct uverbs_attr_bundle *attrs)
1896 {
1897 	struct mlx5_flow_context flow_context = {.flow_tag =
1898 		MLX5_FS_DEFAULT_FLOW_TAG};
1899 	u32 *offset_attr, offset = 0, counter_id = 0;
1900 	int dest_id, dest_type = -1, inlen, len, ret, i;
1901 	struct mlx5_ib_flow_handler *flow_handler;
1902 	struct mlx5_ib_flow_matcher *fs_matcher;
1903 	struct ib_uobject **arr_flow_actions;
1904 	struct ib_uflow_resources *uflow_res;
1905 	struct mlx5_flow_act flow_act = {};
1906 	struct ib_qp *qp = NULL;
1907 	void *devx_obj, *cmd_in;
1908 	struct ib_uobject *uobj;
1909 	struct mlx5_ib_dev *dev;
1910 	u32 flags;
1911 
1912 	if (!capable(CAP_NET_RAW))
1913 		return -EPERM;
1914 
1915 	fs_matcher = uverbs_attr_get_obj(attrs,
1916 					 MLX5_IB_ATTR_CREATE_FLOW_MATCHER);
1917 	uobj =  uverbs_attr_get_uobject(attrs, MLX5_IB_ATTR_CREATE_FLOW_HANDLE);
1918 	dev = mlx5_udata_to_mdev(&attrs->driver_udata);
1919 
1920 	if (get_dests(attrs, fs_matcher, &dest_id, &dest_type, &qp, &flags))
1921 		return -EINVAL;
1922 
1923 	if (flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DEFAULT_MISS)
1924 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
1925 
1926 	if (flags & MLX5_IB_ATTR_CREATE_FLOW_FLAGS_DROP)
1927 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
1928 
1929 	len = uverbs_attr_get_uobjs_arr(attrs,
1930 		MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX, &arr_flow_actions);
1931 	if (len) {
1932 		devx_obj = arr_flow_actions[0]->object;
1933 
1934 		if (uverbs_attr_is_valid(attrs,
1935 					 MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET)) {
1936 
1937 			int num_offsets = uverbs_attr_ptr_get_array_size(
1938 				attrs,
1939 				MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
1940 				sizeof(u32));
1941 
1942 			if (num_offsets != 1)
1943 				return -EINVAL;
1944 
1945 			offset_attr = uverbs_attr_get_alloced_ptr(
1946 				attrs,
1947 				MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET);
1948 			offset = *offset_attr;
1949 		}
1950 
1951 		if (!is_flow_counter(devx_obj, offset, &counter_id))
1952 			return -EINVAL;
1953 
1954 		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
1955 	}
1956 
1957 	cmd_in = uverbs_attr_get_alloced_ptr(
1958 		attrs, MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE);
1959 	inlen = uverbs_attr_get_len(attrs,
1960 				    MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE);
1961 
1962 	uflow_res = flow_resources_alloc(MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS);
1963 	if (!uflow_res)
1964 		return -ENOMEM;
1965 
1966 	len = uverbs_attr_get_uobjs_arr(attrs,
1967 		MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS, &arr_flow_actions);
1968 	for (i = 0; i < len; i++) {
1969 		struct mlx5_ib_flow_action *maction =
1970 			to_mflow_act(arr_flow_actions[i]->object);
1971 
1972 		ret = parse_flow_flow_action(maction, false, &flow_act);
1973 		if (ret)
1974 			goto err_out;
1975 		flow_resources_add(uflow_res, IB_FLOW_SPEC_ACTION_HANDLE,
1976 				   arr_flow_actions[i]->object);
1977 	}
1978 
1979 	ret = uverbs_copy_from(&flow_context.flow_tag, attrs,
1980 			       MLX5_IB_ATTR_CREATE_FLOW_TAG);
1981 	if (!ret) {
1982 		if (flow_context.flow_tag >= BIT(24)) {
1983 			ret = -EINVAL;
1984 			goto err_out;
1985 		}
1986 		flow_context.flags |= FLOW_CONTEXT_HAS_TAG;
1987 	}
1988 
1989 	flow_handler =
1990 		raw_fs_rule_add(dev, fs_matcher, &flow_context, &flow_act,
1991 				counter_id, cmd_in, inlen, dest_id, dest_type);
1992 	if (IS_ERR(flow_handler)) {
1993 		ret = PTR_ERR(flow_handler);
1994 		goto err_out;
1995 	}
1996 
1997 	ib_set_flow(uobj, &flow_handler->ibflow, qp, &dev->ib_dev, uflow_res);
1998 
1999 	return 0;
2000 err_out:
2001 	ib_uverbs_flow_resources_free(uflow_res);
2002 	return ret;
2003 }
2004 
2005 static int flow_matcher_cleanup(struct ib_uobject *uobject,
2006 				enum rdma_remove_reason why,
2007 				struct uverbs_attr_bundle *attrs)
2008 {
2009 	struct mlx5_ib_flow_matcher *obj = uobject->object;
2010 
2011 	if (atomic_read(&obj->usecnt))
2012 		return -EBUSY;
2013 
2014 	kfree(obj);
2015 	return 0;
2016 }
2017 
2018 static int mlx5_ib_matcher_ns(struct uverbs_attr_bundle *attrs,
2019 			      struct mlx5_ib_flow_matcher *obj)
2020 {
2021 	enum mlx5_ib_uapi_flow_table_type ft_type =
2022 		MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX;
2023 	u32 flags;
2024 	int err;
2025 
2026 	/* New users should use MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE and older
2027 	 * users should switch to it. We leave this to not break userspace
2028 	 */
2029 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE) &&
2030 	    uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS))
2031 		return -EINVAL;
2032 
2033 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE)) {
2034 		err = uverbs_get_const(&ft_type, attrs,
2035 				       MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE);
2036 		if (err)
2037 			return err;
2038 
2039 		err = mlx5_ib_ft_type_to_namespace(ft_type, &obj->ns_type);
2040 		if (err)
2041 			return err;
2042 
2043 		return 0;
2044 	}
2045 
2046 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS)) {
2047 		err = uverbs_get_flags32(&flags, attrs,
2048 					 MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2049 					 IB_FLOW_ATTR_FLAGS_EGRESS);
2050 		if (err)
2051 			return err;
2052 
2053 		if (flags) {
2054 			mlx5_ib_ft_type_to_namespace(
2055 				MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX,
2056 				&obj->ns_type);
2057 			return 0;
2058 		}
2059 	}
2060 
2061 	obj->ns_type = MLX5_FLOW_NAMESPACE_BYPASS;
2062 
2063 	return 0;
2064 }
2065 
2066 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_MATCHER_CREATE)(
2067 	struct uverbs_attr_bundle *attrs)
2068 {
2069 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2070 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE);
2071 	struct mlx5_ib_dev *dev = mlx5_udata_to_mdev(&attrs->driver_udata);
2072 	struct mlx5_ib_flow_matcher *obj;
2073 	int err;
2074 
2075 	obj = kzalloc(sizeof(struct mlx5_ib_flow_matcher), GFP_KERNEL);
2076 	if (!obj)
2077 		return -ENOMEM;
2078 
2079 	obj->mask_len = uverbs_attr_get_len(
2080 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2081 	err = uverbs_copy_from(&obj->matcher_mask,
2082 			       attrs,
2083 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK);
2084 	if (err)
2085 		goto end;
2086 
2087 	obj->flow_type = uverbs_attr_get_enum_id(
2088 		attrs, MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2089 
2090 	if (obj->flow_type == MLX5_IB_FLOW_TYPE_NORMAL) {
2091 		err = uverbs_copy_from(&obj->priority,
2092 				       attrs,
2093 				       MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE);
2094 		if (err)
2095 			goto end;
2096 	}
2097 
2098 	err = uverbs_copy_from(&obj->match_criteria_enable,
2099 			       attrs,
2100 			       MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA);
2101 	if (err)
2102 		goto end;
2103 
2104 	err = mlx5_ib_matcher_ns(attrs, obj);
2105 	if (err)
2106 		goto end;
2107 
2108 	if (obj->ns_type == MLX5_FLOW_NAMESPACE_FDB_BYPASS &&
2109 	    mlx5_eswitch_mode(dev->mdev) != MLX5_ESWITCH_OFFLOADS) {
2110 		err = -EINVAL;
2111 		goto end;
2112 	}
2113 
2114 	uobj->object = obj;
2115 	obj->mdev = dev->mdev;
2116 	atomic_set(&obj->usecnt, 0);
2117 	return 0;
2118 
2119 end:
2120 	kfree(obj);
2121 	return err;
2122 }
2123 
2124 static struct ib_flow_action *
2125 mlx5_ib_create_modify_header(struct mlx5_ib_dev *dev,
2126 			     enum mlx5_ib_uapi_flow_table_type ft_type,
2127 			     u8 num_actions, void *in)
2128 {
2129 	enum mlx5_flow_namespace_type namespace;
2130 	struct mlx5_ib_flow_action *maction;
2131 	int ret;
2132 
2133 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2134 	if (ret)
2135 		return ERR_PTR(-EINVAL);
2136 
2137 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2138 	if (!maction)
2139 		return ERR_PTR(-ENOMEM);
2140 
2141 	maction->flow_action_raw.modify_hdr =
2142 		mlx5_modify_header_alloc(dev->mdev, namespace, num_actions, in);
2143 
2144 	if (IS_ERR(maction->flow_action_raw.modify_hdr)) {
2145 		ret = PTR_ERR(maction->flow_action_raw.modify_hdr);
2146 		kfree(maction);
2147 		return ERR_PTR(ret);
2148 	}
2149 	maction->flow_action_raw.sub_type =
2150 		MLX5_IB_FLOW_ACTION_MODIFY_HEADER;
2151 	maction->flow_action_raw.dev = dev;
2152 
2153 	return &maction->ib_action;
2154 }
2155 
2156 static bool mlx5_ib_modify_header_supported(struct mlx5_ib_dev *dev)
2157 {
2158 	return MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
2159 					 max_modify_header_actions) ||
2160 	       MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev,
2161 					 max_modify_header_actions) ||
2162 	       MLX5_CAP_FLOWTABLE_RDMA_TX(dev->mdev,
2163 					 max_modify_header_actions);
2164 }
2165 
2166 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER)(
2167 	struct uverbs_attr_bundle *attrs)
2168 {
2169 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2170 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE);
2171 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2172 	enum mlx5_ib_uapi_flow_table_type ft_type;
2173 	struct ib_flow_action *action;
2174 	int num_actions;
2175 	void *in;
2176 	int ret;
2177 
2178 	if (!mlx5_ib_modify_header_supported(mdev))
2179 		return -EOPNOTSUPP;
2180 
2181 	in = uverbs_attr_get_alloced_ptr(attrs,
2182 		MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM);
2183 
2184 	num_actions = uverbs_attr_ptr_get_array_size(
2185 		attrs, MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2186 		MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto));
2187 	if (num_actions < 0)
2188 		return num_actions;
2189 
2190 	ret = uverbs_get_const(&ft_type, attrs,
2191 			       MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE);
2192 	if (ret)
2193 		return ret;
2194 	action = mlx5_ib_create_modify_header(mdev, ft_type, num_actions, in);
2195 	if (IS_ERR(action))
2196 		return PTR_ERR(action);
2197 
2198 	uverbs_flow_action_fill_action(action, uobj, &mdev->ib_dev,
2199 				       IB_FLOW_ACTION_UNSPECIFIED);
2200 
2201 	return 0;
2202 }
2203 
2204 static bool mlx5_ib_flow_action_packet_reformat_valid(struct mlx5_ib_dev *ibdev,
2205 						      u8 packet_reformat_type,
2206 						      u8 ft_type)
2207 {
2208 	switch (packet_reformat_type) {
2209 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2210 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2211 			return MLX5_CAP_FLOWTABLE(ibdev->mdev,
2212 						  encap_general_header);
2213 		break;
2214 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2215 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX)
2216 			return MLX5_CAP_FLOWTABLE_NIC_TX(ibdev->mdev,
2217 				reformat_l2_to_l3_tunnel);
2218 		break;
2219 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2220 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2221 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev,
2222 				reformat_l3_tunnel_to_l2);
2223 		break;
2224 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2:
2225 		if (ft_type == MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX)
2226 			return MLX5_CAP_FLOWTABLE_NIC_RX(ibdev->mdev, decap);
2227 		break;
2228 	default:
2229 		break;
2230 	}
2231 
2232 	return false;
2233 }
2234 
2235 static int mlx5_ib_dv_to_prm_packet_reforamt_type(u8 dv_prt, u8 *prm_prt)
2236 {
2237 	switch (dv_prt) {
2238 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL:
2239 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L2_TUNNEL;
2240 		break;
2241 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2:
2242 		*prm_prt = MLX5_REFORMAT_TYPE_L3_TUNNEL_TO_L2;
2243 		break;
2244 	case MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL:
2245 		*prm_prt = MLX5_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
2246 		break;
2247 	default:
2248 		return -EINVAL;
2249 	}
2250 
2251 	return 0;
2252 }
2253 
2254 static int mlx5_ib_flow_action_create_packet_reformat_ctx(
2255 	struct mlx5_ib_dev *dev,
2256 	struct mlx5_ib_flow_action *maction,
2257 	u8 ft_type, u8 dv_prt,
2258 	void *in, size_t len)
2259 {
2260 	struct mlx5_pkt_reformat_params reformat_params;
2261 	enum mlx5_flow_namespace_type namespace;
2262 	u8 prm_prt;
2263 	int ret;
2264 
2265 	ret = mlx5_ib_ft_type_to_namespace(ft_type, &namespace);
2266 	if (ret)
2267 		return ret;
2268 
2269 	ret = mlx5_ib_dv_to_prm_packet_reforamt_type(dv_prt, &prm_prt);
2270 	if (ret)
2271 		return ret;
2272 
2273 	memset(&reformat_params, 0, sizeof(reformat_params));
2274 	reformat_params.type = prm_prt;
2275 	reformat_params.size = len;
2276 	reformat_params.data = in;
2277 	maction->flow_action_raw.pkt_reformat =
2278 		mlx5_packet_reformat_alloc(dev->mdev, &reformat_params,
2279 					   namespace);
2280 	if (IS_ERR(maction->flow_action_raw.pkt_reformat)) {
2281 		ret = PTR_ERR(maction->flow_action_raw.pkt_reformat);
2282 		return ret;
2283 	}
2284 
2285 	maction->flow_action_raw.sub_type =
2286 		MLX5_IB_FLOW_ACTION_PACKET_REFORMAT;
2287 	maction->flow_action_raw.dev = dev;
2288 
2289 	return 0;
2290 }
2291 
2292 static int UVERBS_HANDLER(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT)(
2293 	struct uverbs_attr_bundle *attrs)
2294 {
2295 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs,
2296 		MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE);
2297 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
2298 	enum mlx5_ib_uapi_flow_action_packet_reformat_type dv_prt;
2299 	enum mlx5_ib_uapi_flow_table_type ft_type;
2300 	struct mlx5_ib_flow_action *maction;
2301 	int ret;
2302 
2303 	ret = uverbs_get_const(&ft_type, attrs,
2304 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE);
2305 	if (ret)
2306 		return ret;
2307 
2308 	ret = uverbs_get_const(&dv_prt, attrs,
2309 			       MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE);
2310 	if (ret)
2311 		return ret;
2312 
2313 	if (!mlx5_ib_flow_action_packet_reformat_valid(mdev, dv_prt, ft_type))
2314 		return -EOPNOTSUPP;
2315 
2316 	maction = kzalloc(sizeof(*maction), GFP_KERNEL);
2317 	if (!maction)
2318 		return -ENOMEM;
2319 
2320 	if (dv_prt ==
2321 	    MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2) {
2322 		maction->flow_action_raw.sub_type =
2323 			MLX5_IB_FLOW_ACTION_DECAP;
2324 		maction->flow_action_raw.dev = mdev;
2325 	} else {
2326 		void *in;
2327 		int len;
2328 
2329 		in = uverbs_attr_get_alloced_ptr(attrs,
2330 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2331 		if (IS_ERR(in)) {
2332 			ret = PTR_ERR(in);
2333 			goto free_maction;
2334 		}
2335 
2336 		len = uverbs_attr_get_len(attrs,
2337 			MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF);
2338 
2339 		ret = mlx5_ib_flow_action_create_packet_reformat_ctx(mdev,
2340 			maction, ft_type, dv_prt, in, len);
2341 		if (ret)
2342 			goto free_maction;
2343 	}
2344 
2345 	uverbs_flow_action_fill_action(&maction->ib_action, uobj, &mdev->ib_dev,
2346 				       IB_FLOW_ACTION_UNSPECIFIED);
2347 	return 0;
2348 
2349 free_maction:
2350 	kfree(maction);
2351 	return ret;
2352 }
2353 
2354 DECLARE_UVERBS_NAMED_METHOD(
2355 	MLX5_IB_METHOD_CREATE_FLOW,
2356 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2357 			UVERBS_OBJECT_FLOW,
2358 			UVERBS_ACCESS_NEW,
2359 			UA_MANDATORY),
2360 	UVERBS_ATTR_PTR_IN(
2361 		MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE,
2362 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2363 		UA_MANDATORY,
2364 		UA_ALLOC_AND_COPY),
2365 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_MATCHER,
2366 			MLX5_IB_OBJECT_FLOW_MATCHER,
2367 			UVERBS_ACCESS_READ,
2368 			UA_MANDATORY),
2369 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_QP,
2370 			UVERBS_OBJECT_QP,
2371 			UVERBS_ACCESS_READ),
2372 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX,
2373 			MLX5_IB_OBJECT_DEVX_OBJ,
2374 			UVERBS_ACCESS_READ),
2375 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_FLOW_ACTIONS,
2376 			     UVERBS_OBJECT_FLOW_ACTION,
2377 			     UVERBS_ACCESS_READ, 1,
2378 			     MLX5_IB_CREATE_FLOW_MAX_FLOW_ACTIONS,
2379 			     UA_OPTIONAL),
2380 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_TAG,
2381 			   UVERBS_ATTR_TYPE(u32),
2382 			   UA_OPTIONAL),
2383 	UVERBS_ATTR_IDRS_ARR(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX,
2384 			     MLX5_IB_OBJECT_DEVX_OBJ,
2385 			     UVERBS_ACCESS_READ, 1, 1,
2386 			     UA_OPTIONAL),
2387 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_ARR_COUNTERS_DEVX_OFFSET,
2388 			   UVERBS_ATTR_MIN_SIZE(sizeof(u32)),
2389 			   UA_OPTIONAL,
2390 			   UA_ALLOC_AND_COPY),
2391 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_CREATE_FLOW_FLAGS,
2392 			     enum mlx5_ib_create_flow_flags,
2393 			     UA_OPTIONAL));
2394 
2395 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2396 	MLX5_IB_METHOD_DESTROY_FLOW,
2397 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_FLOW_HANDLE,
2398 			UVERBS_OBJECT_FLOW,
2399 			UVERBS_ACCESS_DESTROY,
2400 			UA_MANDATORY));
2401 
2402 ADD_UVERBS_METHODS(mlx5_ib_fs,
2403 		   UVERBS_OBJECT_FLOW,
2404 		   &UVERBS_METHOD(MLX5_IB_METHOD_CREATE_FLOW),
2405 		   &UVERBS_METHOD(MLX5_IB_METHOD_DESTROY_FLOW));
2406 
2407 DECLARE_UVERBS_NAMED_METHOD(
2408 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER,
2409 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_HANDLE,
2410 			UVERBS_OBJECT_FLOW_ACTION,
2411 			UVERBS_ACCESS_NEW,
2412 			UA_MANDATORY),
2413 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_ACTIONS_PRM,
2414 			   UVERBS_ATTR_MIN_SIZE(MLX5_UN_SZ_BYTES(
2415 				   set_add_copy_action_in_auto)),
2416 			   UA_MANDATORY,
2417 			   UA_ALLOC_AND_COPY),
2418 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_MODIFY_HEADER_FT_TYPE,
2419 			     enum mlx5_ib_uapi_flow_table_type,
2420 			     UA_MANDATORY));
2421 
2422 DECLARE_UVERBS_NAMED_METHOD(
2423 	MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT,
2424 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_HANDLE,
2425 			UVERBS_OBJECT_FLOW_ACTION,
2426 			UVERBS_ACCESS_NEW,
2427 			UA_MANDATORY),
2428 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_DATA_BUF,
2429 			   UVERBS_ATTR_MIN_SIZE(1),
2430 			   UA_ALLOC_AND_COPY,
2431 			   UA_OPTIONAL),
2432 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_TYPE,
2433 			     enum mlx5_ib_uapi_flow_action_packet_reformat_type,
2434 			     UA_MANDATORY),
2435 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_CREATE_PACKET_REFORMAT_FT_TYPE,
2436 			     enum mlx5_ib_uapi_flow_table_type,
2437 			     UA_MANDATORY));
2438 
2439 ADD_UVERBS_METHODS(
2440 	mlx5_ib_flow_actions,
2441 	UVERBS_OBJECT_FLOW_ACTION,
2442 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_MODIFY_HEADER),
2443 	&UVERBS_METHOD(MLX5_IB_METHOD_FLOW_ACTION_CREATE_PACKET_REFORMAT));
2444 
2445 DECLARE_UVERBS_NAMED_METHOD(
2446 	MLX5_IB_METHOD_FLOW_MATCHER_CREATE,
2447 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE,
2448 			MLX5_IB_OBJECT_FLOW_MATCHER,
2449 			UVERBS_ACCESS_NEW,
2450 			UA_MANDATORY),
2451 	UVERBS_ATTR_PTR_IN(
2452 		MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK,
2453 		UVERBS_ATTR_SIZE(1, sizeof(struct mlx5_ib_match_params)),
2454 		UA_MANDATORY),
2455 	UVERBS_ATTR_ENUM_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE,
2456 			    mlx5_ib_flow_type,
2457 			    UA_MANDATORY),
2458 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA,
2459 			   UVERBS_ATTR_TYPE(u8),
2460 			   UA_MANDATORY),
2461 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_FLOW_MATCHER_FLOW_FLAGS,
2462 			     enum ib_flow_flags,
2463 			     UA_OPTIONAL),
2464 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_FLOW_MATCHER_FT_TYPE,
2465 			     enum mlx5_ib_uapi_flow_table_type,
2466 			     UA_OPTIONAL));
2467 
2468 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2469 	MLX5_IB_METHOD_FLOW_MATCHER_DESTROY,
2470 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_FLOW_MATCHER_DESTROY_HANDLE,
2471 			MLX5_IB_OBJECT_FLOW_MATCHER,
2472 			UVERBS_ACCESS_DESTROY,
2473 			UA_MANDATORY));
2474 
2475 DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_FLOW_MATCHER,
2476 			    UVERBS_TYPE_ALLOC_IDR(flow_matcher_cleanup),
2477 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_CREATE),
2478 			    &UVERBS_METHOD(MLX5_IB_METHOD_FLOW_MATCHER_DESTROY));
2479 
2480 const struct uapi_definition mlx5_ib_flow_defs[] = {
2481 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
2482 		MLX5_IB_OBJECT_FLOW_MATCHER),
2483 	UAPI_DEF_CHAIN_OBJ_TREE(
2484 		UVERBS_OBJECT_FLOW,
2485 		&mlx5_ib_fs),
2486 	UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_FLOW_ACTION,
2487 				&mlx5_ib_flow_actions),
2488 	{},
2489 };
2490 
2491 static const struct ib_device_ops flow_ops = {
2492 	.create_flow = mlx5_ib_create_flow,
2493 	.destroy_flow = mlx5_ib_destroy_flow,
2494 	.destroy_flow_action = mlx5_ib_destroy_flow_action,
2495 };
2496 
2497 int mlx5_ib_fs_init(struct mlx5_ib_dev *dev)
2498 {
2499 	dev->flow_db = kzalloc(sizeof(*dev->flow_db), GFP_KERNEL);
2500 
2501 	if (!dev->flow_db)
2502 		return -ENOMEM;
2503 
2504 	mutex_init(&dev->flow_db->lock);
2505 
2506 	ib_set_device_ops(&dev->ib_dev, &flow_ops);
2507 	return 0;
2508 }
2509