xref: /dpdk/drivers/net/sfc/sfc_flow.c (revision 76e9a55b5b8248f5b48d3819a7c878fa11d68726)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2017-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <rte_byteorder.h>
11 #include <rte_tailq.h>
12 #include <rte_common.h>
13 #include <rte_ethdev_driver.h>
14 #include <rte_eth_ctrl.h>
15 #include <rte_ether.h>
16 #include <rte_flow.h>
17 #include <rte_flow_driver.h>
18 
19 #include "efx.h"
20 
21 #include "sfc.h"
22 #include "sfc_rx.h"
23 #include "sfc_filter.h"
24 #include "sfc_flow.h"
25 #include "sfc_log.h"
26 
27 /*
28  * At now flow API is implemented in such a manner that each
29  * flow rule is converted to one or more hardware filters.
30  * All elements of flow rule (attributes, pattern items, actions)
31  * correspond to one or more fields in the efx_filter_spec_s structure
32  * that is responsible for the hardware filter.
33  * If some required field is unset in the flow rule, then a handful
34  * of filter copies will be created to cover all possible values
35  * of such a field.
36  */
37 
38 enum sfc_flow_item_layers {
39 	SFC_FLOW_ITEM_ANY_LAYER,
40 	SFC_FLOW_ITEM_START_LAYER,
41 	SFC_FLOW_ITEM_L2,
42 	SFC_FLOW_ITEM_L3,
43 	SFC_FLOW_ITEM_L4,
44 };
45 
46 typedef int (sfc_flow_item_parse)(const struct rte_flow_item *item,
47 				  efx_filter_spec_t *spec,
48 				  struct rte_flow_error *error);
49 
50 struct sfc_flow_item {
51 	enum rte_flow_item_type type;		/* Type of item */
52 	enum sfc_flow_item_layers layer;	/* Layer of item */
53 	enum sfc_flow_item_layers prev_layer;	/* Previous layer of item */
54 	sfc_flow_item_parse *parse;		/* Parsing function */
55 };
56 
57 static sfc_flow_item_parse sfc_flow_parse_void;
58 static sfc_flow_item_parse sfc_flow_parse_eth;
59 static sfc_flow_item_parse sfc_flow_parse_vlan;
60 static sfc_flow_item_parse sfc_flow_parse_ipv4;
61 static sfc_flow_item_parse sfc_flow_parse_ipv6;
62 static sfc_flow_item_parse sfc_flow_parse_tcp;
63 static sfc_flow_item_parse sfc_flow_parse_udp;
64 static sfc_flow_item_parse sfc_flow_parse_vxlan;
65 static sfc_flow_item_parse sfc_flow_parse_geneve;
66 static sfc_flow_item_parse sfc_flow_parse_nvgre;
67 
68 typedef int (sfc_flow_spec_set_vals)(struct sfc_flow_spec *spec,
69 				     unsigned int filters_count_for_one_val,
70 				     struct rte_flow_error *error);
71 
72 typedef boolean_t (sfc_flow_spec_check)(efx_filter_match_flags_t match,
73 					efx_filter_spec_t *spec,
74 					struct sfc_filter *filter);
75 
76 struct sfc_flow_copy_flag {
77 	/* EFX filter specification match flag */
78 	efx_filter_match_flags_t flag;
79 	/* Number of values of corresponding field */
80 	unsigned int vals_count;
81 	/* Function to set values in specifications */
82 	sfc_flow_spec_set_vals *set_vals;
83 	/*
84 	 * Function to check that the specification is suitable
85 	 * for adding this match flag
86 	 */
87 	sfc_flow_spec_check *spec_check;
88 };
89 
90 static sfc_flow_spec_set_vals sfc_flow_set_unknown_dst_flags;
91 static sfc_flow_spec_check sfc_flow_check_unknown_dst_flags;
92 static sfc_flow_spec_set_vals sfc_flow_set_ethertypes;
93 static sfc_flow_spec_set_vals sfc_flow_set_ifrm_unknown_dst_flags;
94 static sfc_flow_spec_check sfc_flow_check_ifrm_unknown_dst_flags;
95 
96 static boolean_t
97 sfc_flow_is_zero(const uint8_t *buf, unsigned int size)
98 {
99 	uint8_t sum = 0;
100 	unsigned int i;
101 
102 	for (i = 0; i < size; i++)
103 		sum |= buf[i];
104 
105 	return (sum == 0) ? B_TRUE : B_FALSE;
106 }
107 
108 /*
109  * Validate item and prepare structures spec and mask for parsing
110  */
111 static int
112 sfc_flow_parse_init(const struct rte_flow_item *item,
113 		    const void **spec_ptr,
114 		    const void **mask_ptr,
115 		    const void *supp_mask,
116 		    const void *def_mask,
117 		    unsigned int size,
118 		    struct rte_flow_error *error)
119 {
120 	const uint8_t *spec;
121 	const uint8_t *mask;
122 	const uint8_t *last;
123 	uint8_t match;
124 	uint8_t supp;
125 	unsigned int i;
126 
127 	if (item == NULL) {
128 		rte_flow_error_set(error, EINVAL,
129 				   RTE_FLOW_ERROR_TYPE_ITEM, NULL,
130 				   "NULL item");
131 		return -rte_errno;
132 	}
133 
134 	if ((item->last != NULL || item->mask != NULL) && item->spec == NULL) {
135 		rte_flow_error_set(error, EINVAL,
136 				   RTE_FLOW_ERROR_TYPE_ITEM, item,
137 				   "Mask or last is set without spec");
138 		return -rte_errno;
139 	}
140 
141 	/*
142 	 * If "mask" is not set, default mask is used,
143 	 * but if default mask is NULL, "mask" should be set
144 	 */
145 	if (item->mask == NULL) {
146 		if (def_mask == NULL) {
147 			rte_flow_error_set(error, EINVAL,
148 				RTE_FLOW_ERROR_TYPE_ITEM, NULL,
149 				"Mask should be specified");
150 			return -rte_errno;
151 		}
152 
153 		mask = def_mask;
154 	} else {
155 		mask = item->mask;
156 	}
157 
158 	spec = item->spec;
159 	last = item->last;
160 
161 	if (spec == NULL)
162 		goto exit;
163 
164 	/*
165 	 * If field values in "last" are either 0 or equal to the corresponding
166 	 * values in "spec" then they are ignored
167 	 */
168 	if (last != NULL &&
169 	    !sfc_flow_is_zero(last, size) &&
170 	    memcmp(last, spec, size) != 0) {
171 		rte_flow_error_set(error, ENOTSUP,
172 				   RTE_FLOW_ERROR_TYPE_ITEM, item,
173 				   "Ranging is not supported");
174 		return -rte_errno;
175 	}
176 
177 	if (supp_mask == NULL) {
178 		rte_flow_error_set(error, EINVAL,
179 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
180 			"Supported mask for item should be specified");
181 		return -rte_errno;
182 	}
183 
184 	/* Check that mask and spec not asks for more match than supp_mask */
185 	for (i = 0; i < size; i++) {
186 		match = spec[i] | mask[i];
187 		supp = ((const uint8_t *)supp_mask)[i];
188 
189 		if ((match | supp) != supp) {
190 			rte_flow_error_set(error, ENOTSUP,
191 					   RTE_FLOW_ERROR_TYPE_ITEM, item,
192 					   "Item's field is not supported");
193 			return -rte_errno;
194 		}
195 	}
196 
197 exit:
198 	*spec_ptr = spec;
199 	*mask_ptr = mask;
200 	return 0;
201 }
202 
203 /*
204  * Protocol parsers.
205  * Masking is not supported, so masks in items should be either
206  * full or empty (zeroed) and set only for supported fields which
207  * are specified in the supp_mask.
208  */
209 
210 static int
211 sfc_flow_parse_void(__rte_unused const struct rte_flow_item *item,
212 		    __rte_unused efx_filter_spec_t *efx_spec,
213 		    __rte_unused struct rte_flow_error *error)
214 {
215 	return 0;
216 }
217 
218 /**
219  * Convert Ethernet item to EFX filter specification.
220  *
221  * @param item[in]
222  *   Item specification. Outer frame specification may only comprise
223  *   source/destination addresses and Ethertype field.
224  *   Inner frame specification may contain destination address only.
225  *   There is support for individual/group mask as well as for empty and full.
226  *   If the mask is NULL, default mask will be used. Ranging is not supported.
227  * @param efx_spec[in, out]
228  *   EFX filter specification to update.
229  * @param[out] error
230  *   Perform verbose error reporting if not NULL.
231  */
232 static int
233 sfc_flow_parse_eth(const struct rte_flow_item *item,
234 		   efx_filter_spec_t *efx_spec,
235 		   struct rte_flow_error *error)
236 {
237 	int rc;
238 	const struct rte_flow_item_eth *spec = NULL;
239 	const struct rte_flow_item_eth *mask = NULL;
240 	const struct rte_flow_item_eth supp_mask = {
241 		.dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
242 		.src.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
243 		.type = 0xffff,
244 	};
245 	const struct rte_flow_item_eth ifrm_supp_mask = {
246 		.dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
247 	};
248 	const uint8_t ig_mask[EFX_MAC_ADDR_LEN] = {
249 		0x01, 0x00, 0x00, 0x00, 0x00, 0x00
250 	};
251 	const struct rte_flow_item_eth *supp_mask_p;
252 	const struct rte_flow_item_eth *def_mask_p;
253 	uint8_t *loc_mac = NULL;
254 	boolean_t is_ifrm = (efx_spec->efs_encap_type !=
255 		EFX_TUNNEL_PROTOCOL_NONE);
256 
257 	if (is_ifrm) {
258 		supp_mask_p = &ifrm_supp_mask;
259 		def_mask_p = &ifrm_supp_mask;
260 		loc_mac = efx_spec->efs_ifrm_loc_mac;
261 	} else {
262 		supp_mask_p = &supp_mask;
263 		def_mask_p = &rte_flow_item_eth_mask;
264 		loc_mac = efx_spec->efs_loc_mac;
265 	}
266 
267 	rc = sfc_flow_parse_init(item,
268 				 (const void **)&spec,
269 				 (const void **)&mask,
270 				 supp_mask_p, def_mask_p,
271 				 sizeof(struct rte_flow_item_eth),
272 				 error);
273 	if (rc != 0)
274 		return rc;
275 
276 	/* If "spec" is not set, could be any Ethernet */
277 	if (spec == NULL)
278 		return 0;
279 
280 	if (is_same_ether_addr(&mask->dst, &supp_mask.dst)) {
281 		efx_spec->efs_match_flags |= is_ifrm ?
282 			EFX_FILTER_MATCH_IFRM_LOC_MAC :
283 			EFX_FILTER_MATCH_LOC_MAC;
284 		rte_memcpy(loc_mac, spec->dst.addr_bytes,
285 			   EFX_MAC_ADDR_LEN);
286 	} else if (memcmp(mask->dst.addr_bytes, ig_mask,
287 			  EFX_MAC_ADDR_LEN) == 0) {
288 		if (is_unicast_ether_addr(&spec->dst))
289 			efx_spec->efs_match_flags |= is_ifrm ?
290 				EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST :
291 				EFX_FILTER_MATCH_UNKNOWN_UCAST_DST;
292 		else
293 			efx_spec->efs_match_flags |= is_ifrm ?
294 				EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST :
295 				EFX_FILTER_MATCH_UNKNOWN_MCAST_DST;
296 	} else if (!is_zero_ether_addr(&mask->dst)) {
297 		goto fail_bad_mask;
298 	}
299 
300 	/*
301 	 * ifrm_supp_mask ensures that the source address and
302 	 * ethertype masks are equal to zero in inner frame,
303 	 * so these fields are filled in only for the outer frame
304 	 */
305 	if (is_same_ether_addr(&mask->src, &supp_mask.src)) {
306 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_MAC;
307 		rte_memcpy(efx_spec->efs_rem_mac, spec->src.addr_bytes,
308 			   EFX_MAC_ADDR_LEN);
309 	} else if (!is_zero_ether_addr(&mask->src)) {
310 		goto fail_bad_mask;
311 	}
312 
313 	/*
314 	 * Ether type is in big-endian byte order in item and
315 	 * in little-endian in efx_spec, so byte swap is used
316 	 */
317 	if (mask->type == supp_mask.type) {
318 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
319 		efx_spec->efs_ether_type = rte_bswap16(spec->type);
320 	} else if (mask->type != 0) {
321 		goto fail_bad_mask;
322 	}
323 
324 	return 0;
325 
326 fail_bad_mask:
327 	rte_flow_error_set(error, EINVAL,
328 			   RTE_FLOW_ERROR_TYPE_ITEM, item,
329 			   "Bad mask in the ETH pattern item");
330 	return -rte_errno;
331 }
332 
333 /**
334  * Convert VLAN item to EFX filter specification.
335  *
336  * @param item[in]
337  *   Item specification. Only VID field is supported.
338  *   The mask can not be NULL. Ranging is not supported.
339  * @param efx_spec[in, out]
340  *   EFX filter specification to update.
341  * @param[out] error
342  *   Perform verbose error reporting if not NULL.
343  */
344 static int
345 sfc_flow_parse_vlan(const struct rte_flow_item *item,
346 		    efx_filter_spec_t *efx_spec,
347 		    struct rte_flow_error *error)
348 {
349 	int rc;
350 	uint16_t vid;
351 	const struct rte_flow_item_vlan *spec = NULL;
352 	const struct rte_flow_item_vlan *mask = NULL;
353 	const struct rte_flow_item_vlan supp_mask = {
354 		.tci = rte_cpu_to_be_16(ETH_VLAN_ID_MAX),
355 		.inner_type = RTE_BE16(0xffff),
356 	};
357 
358 	rc = sfc_flow_parse_init(item,
359 				 (const void **)&spec,
360 				 (const void **)&mask,
361 				 &supp_mask,
362 				 NULL,
363 				 sizeof(struct rte_flow_item_vlan),
364 				 error);
365 	if (rc != 0)
366 		return rc;
367 
368 	/*
369 	 * VID is in big-endian byte order in item and
370 	 * in little-endian in efx_spec, so byte swap is used.
371 	 * If two VLAN items are included, the first matches
372 	 * the outer tag and the next matches the inner tag.
373 	 */
374 	if (mask->tci == supp_mask.tci) {
375 		vid = rte_bswap16(spec->tci);
376 
377 		if (!(efx_spec->efs_match_flags &
378 		      EFX_FILTER_MATCH_OUTER_VID)) {
379 			efx_spec->efs_match_flags |= EFX_FILTER_MATCH_OUTER_VID;
380 			efx_spec->efs_outer_vid = vid;
381 		} else if (!(efx_spec->efs_match_flags &
382 			     EFX_FILTER_MATCH_INNER_VID)) {
383 			efx_spec->efs_match_flags |= EFX_FILTER_MATCH_INNER_VID;
384 			efx_spec->efs_inner_vid = vid;
385 		} else {
386 			rte_flow_error_set(error, EINVAL,
387 					   RTE_FLOW_ERROR_TYPE_ITEM, item,
388 					   "More than two VLAN items");
389 			return -rte_errno;
390 		}
391 	} else {
392 		rte_flow_error_set(error, EINVAL,
393 				   RTE_FLOW_ERROR_TYPE_ITEM, item,
394 				   "VLAN ID in TCI match is required");
395 		return -rte_errno;
396 	}
397 
398 	if (efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE) {
399 		rte_flow_error_set(error, EINVAL,
400 				   RTE_FLOW_ERROR_TYPE_ITEM, item,
401 				   "VLAN TPID matching is not supported");
402 		return -rte_errno;
403 	}
404 	if (mask->inner_type == supp_mask.inner_type) {
405 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
406 		efx_spec->efs_ether_type = rte_bswap16(spec->inner_type);
407 	} else if (mask->inner_type) {
408 		rte_flow_error_set(error, EINVAL,
409 				   RTE_FLOW_ERROR_TYPE_ITEM, item,
410 				   "Bad mask for VLAN inner_type");
411 		return -rte_errno;
412 	}
413 
414 	return 0;
415 }
416 
417 /**
418  * Convert IPv4 item to EFX filter specification.
419  *
420  * @param item[in]
421  *   Item specification. Only source and destination addresses and
422  *   protocol fields are supported. If the mask is NULL, default
423  *   mask will be used. Ranging is not supported.
424  * @param efx_spec[in, out]
425  *   EFX filter specification to update.
426  * @param[out] error
427  *   Perform verbose error reporting if not NULL.
428  */
429 static int
430 sfc_flow_parse_ipv4(const struct rte_flow_item *item,
431 		    efx_filter_spec_t *efx_spec,
432 		    struct rte_flow_error *error)
433 {
434 	int rc;
435 	const struct rte_flow_item_ipv4 *spec = NULL;
436 	const struct rte_flow_item_ipv4 *mask = NULL;
437 	const uint16_t ether_type_ipv4 = rte_cpu_to_le_16(EFX_ETHER_TYPE_IPV4);
438 	const struct rte_flow_item_ipv4 supp_mask = {
439 		.hdr = {
440 			.src_addr = 0xffffffff,
441 			.dst_addr = 0xffffffff,
442 			.next_proto_id = 0xff,
443 		}
444 	};
445 
446 	rc = sfc_flow_parse_init(item,
447 				 (const void **)&spec,
448 				 (const void **)&mask,
449 				 &supp_mask,
450 				 &rte_flow_item_ipv4_mask,
451 				 sizeof(struct rte_flow_item_ipv4),
452 				 error);
453 	if (rc != 0)
454 		return rc;
455 
456 	/*
457 	 * Filtering by IPv4 source and destination addresses requires
458 	 * the appropriate ETHER_TYPE in hardware filters
459 	 */
460 	if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE)) {
461 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
462 		efx_spec->efs_ether_type = ether_type_ipv4;
463 	} else if (efx_spec->efs_ether_type != ether_type_ipv4) {
464 		rte_flow_error_set(error, EINVAL,
465 			RTE_FLOW_ERROR_TYPE_ITEM, item,
466 			"Ethertype in pattern with IPV4 item should be appropriate");
467 		return -rte_errno;
468 	}
469 
470 	if (spec == NULL)
471 		return 0;
472 
473 	/*
474 	 * IPv4 addresses are in big-endian byte order in item and in
475 	 * efx_spec
476 	 */
477 	if (mask->hdr.src_addr == supp_mask.hdr.src_addr) {
478 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_HOST;
479 		efx_spec->efs_rem_host.eo_u32[0] = spec->hdr.src_addr;
480 	} else if (mask->hdr.src_addr != 0) {
481 		goto fail_bad_mask;
482 	}
483 
484 	if (mask->hdr.dst_addr == supp_mask.hdr.dst_addr) {
485 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_HOST;
486 		efx_spec->efs_loc_host.eo_u32[0] = spec->hdr.dst_addr;
487 	} else if (mask->hdr.dst_addr != 0) {
488 		goto fail_bad_mask;
489 	}
490 
491 	if (mask->hdr.next_proto_id == supp_mask.hdr.next_proto_id) {
492 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
493 		efx_spec->efs_ip_proto = spec->hdr.next_proto_id;
494 	} else if (mask->hdr.next_proto_id != 0) {
495 		goto fail_bad_mask;
496 	}
497 
498 	return 0;
499 
500 fail_bad_mask:
501 	rte_flow_error_set(error, EINVAL,
502 			   RTE_FLOW_ERROR_TYPE_ITEM, item,
503 			   "Bad mask in the IPV4 pattern item");
504 	return -rte_errno;
505 }
506 
507 /**
508  * Convert IPv6 item to EFX filter specification.
509  *
510  * @param item[in]
511  *   Item specification. Only source and destination addresses and
512  *   next header fields are supported. If the mask is NULL, default
513  *   mask will be used. Ranging is not supported.
514  * @param efx_spec[in, out]
515  *   EFX filter specification to update.
516  * @param[out] error
517  *   Perform verbose error reporting if not NULL.
518  */
519 static int
520 sfc_flow_parse_ipv6(const struct rte_flow_item *item,
521 		    efx_filter_spec_t *efx_spec,
522 		    struct rte_flow_error *error)
523 {
524 	int rc;
525 	const struct rte_flow_item_ipv6 *spec = NULL;
526 	const struct rte_flow_item_ipv6 *mask = NULL;
527 	const uint16_t ether_type_ipv6 = rte_cpu_to_le_16(EFX_ETHER_TYPE_IPV6);
528 	const struct rte_flow_item_ipv6 supp_mask = {
529 		.hdr = {
530 			.src_addr = { 0xff, 0xff, 0xff, 0xff,
531 				      0xff, 0xff, 0xff, 0xff,
532 				      0xff, 0xff, 0xff, 0xff,
533 				      0xff, 0xff, 0xff, 0xff },
534 			.dst_addr = { 0xff, 0xff, 0xff, 0xff,
535 				      0xff, 0xff, 0xff, 0xff,
536 				      0xff, 0xff, 0xff, 0xff,
537 				      0xff, 0xff, 0xff, 0xff },
538 			.proto = 0xff,
539 		}
540 	};
541 
542 	rc = sfc_flow_parse_init(item,
543 				 (const void **)&spec,
544 				 (const void **)&mask,
545 				 &supp_mask,
546 				 &rte_flow_item_ipv6_mask,
547 				 sizeof(struct rte_flow_item_ipv6),
548 				 error);
549 	if (rc != 0)
550 		return rc;
551 
552 	/*
553 	 * Filtering by IPv6 source and destination addresses requires
554 	 * the appropriate ETHER_TYPE in hardware filters
555 	 */
556 	if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE)) {
557 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
558 		efx_spec->efs_ether_type = ether_type_ipv6;
559 	} else if (efx_spec->efs_ether_type != ether_type_ipv6) {
560 		rte_flow_error_set(error, EINVAL,
561 			RTE_FLOW_ERROR_TYPE_ITEM, item,
562 			"Ethertype in pattern with IPV6 item should be appropriate");
563 		return -rte_errno;
564 	}
565 
566 	if (spec == NULL)
567 		return 0;
568 
569 	/*
570 	 * IPv6 addresses are in big-endian byte order in item and in
571 	 * efx_spec
572 	 */
573 	if (memcmp(mask->hdr.src_addr, supp_mask.hdr.src_addr,
574 		   sizeof(mask->hdr.src_addr)) == 0) {
575 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_HOST;
576 
577 		RTE_BUILD_BUG_ON(sizeof(efx_spec->efs_rem_host) !=
578 				 sizeof(spec->hdr.src_addr));
579 		rte_memcpy(&efx_spec->efs_rem_host, spec->hdr.src_addr,
580 			   sizeof(efx_spec->efs_rem_host));
581 	} else if (!sfc_flow_is_zero(mask->hdr.src_addr,
582 				     sizeof(mask->hdr.src_addr))) {
583 		goto fail_bad_mask;
584 	}
585 
586 	if (memcmp(mask->hdr.dst_addr, supp_mask.hdr.dst_addr,
587 		   sizeof(mask->hdr.dst_addr)) == 0) {
588 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_HOST;
589 
590 		RTE_BUILD_BUG_ON(sizeof(efx_spec->efs_loc_host) !=
591 				 sizeof(spec->hdr.dst_addr));
592 		rte_memcpy(&efx_spec->efs_loc_host, spec->hdr.dst_addr,
593 			   sizeof(efx_spec->efs_loc_host));
594 	} else if (!sfc_flow_is_zero(mask->hdr.dst_addr,
595 				     sizeof(mask->hdr.dst_addr))) {
596 		goto fail_bad_mask;
597 	}
598 
599 	if (mask->hdr.proto == supp_mask.hdr.proto) {
600 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
601 		efx_spec->efs_ip_proto = spec->hdr.proto;
602 	} else if (mask->hdr.proto != 0) {
603 		goto fail_bad_mask;
604 	}
605 
606 	return 0;
607 
608 fail_bad_mask:
609 	rte_flow_error_set(error, EINVAL,
610 			   RTE_FLOW_ERROR_TYPE_ITEM, item,
611 			   "Bad mask in the IPV6 pattern item");
612 	return -rte_errno;
613 }
614 
615 /**
616  * Convert TCP item to EFX filter specification.
617  *
618  * @param item[in]
619  *   Item specification. Only source and destination ports fields
620  *   are supported. If the mask is NULL, default mask will be used.
621  *   Ranging is not supported.
622  * @param efx_spec[in, out]
623  *   EFX filter specification to update.
624  * @param[out] error
625  *   Perform verbose error reporting if not NULL.
626  */
627 static int
628 sfc_flow_parse_tcp(const struct rte_flow_item *item,
629 		   efx_filter_spec_t *efx_spec,
630 		   struct rte_flow_error *error)
631 {
632 	int rc;
633 	const struct rte_flow_item_tcp *spec = NULL;
634 	const struct rte_flow_item_tcp *mask = NULL;
635 	const struct rte_flow_item_tcp supp_mask = {
636 		.hdr = {
637 			.src_port = 0xffff,
638 			.dst_port = 0xffff,
639 		}
640 	};
641 
642 	rc = sfc_flow_parse_init(item,
643 				 (const void **)&spec,
644 				 (const void **)&mask,
645 				 &supp_mask,
646 				 &rte_flow_item_tcp_mask,
647 				 sizeof(struct rte_flow_item_tcp),
648 				 error);
649 	if (rc != 0)
650 		return rc;
651 
652 	/*
653 	 * Filtering by TCP source and destination ports requires
654 	 * the appropriate IP_PROTO in hardware filters
655 	 */
656 	if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
657 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
658 		efx_spec->efs_ip_proto = EFX_IPPROTO_TCP;
659 	} else if (efx_spec->efs_ip_proto != EFX_IPPROTO_TCP) {
660 		rte_flow_error_set(error, EINVAL,
661 			RTE_FLOW_ERROR_TYPE_ITEM, item,
662 			"IP proto in pattern with TCP item should be appropriate");
663 		return -rte_errno;
664 	}
665 
666 	if (spec == NULL)
667 		return 0;
668 
669 	/*
670 	 * Source and destination ports are in big-endian byte order in item and
671 	 * in little-endian in efx_spec, so byte swap is used
672 	 */
673 	if (mask->hdr.src_port == supp_mask.hdr.src_port) {
674 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_PORT;
675 		efx_spec->efs_rem_port = rte_bswap16(spec->hdr.src_port);
676 	} else if (mask->hdr.src_port != 0) {
677 		goto fail_bad_mask;
678 	}
679 
680 	if (mask->hdr.dst_port == supp_mask.hdr.dst_port) {
681 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_PORT;
682 		efx_spec->efs_loc_port = rte_bswap16(spec->hdr.dst_port);
683 	} else if (mask->hdr.dst_port != 0) {
684 		goto fail_bad_mask;
685 	}
686 
687 	return 0;
688 
689 fail_bad_mask:
690 	rte_flow_error_set(error, EINVAL,
691 			   RTE_FLOW_ERROR_TYPE_ITEM, item,
692 			   "Bad mask in the TCP pattern item");
693 	return -rte_errno;
694 }
695 
696 /**
697  * Convert UDP item to EFX filter specification.
698  *
699  * @param item[in]
700  *   Item specification. Only source and destination ports fields
701  *   are supported. If the mask is NULL, default mask will be used.
702  *   Ranging is not supported.
703  * @param efx_spec[in, out]
704  *   EFX filter specification to update.
705  * @param[out] error
706  *   Perform verbose error reporting if not NULL.
707  */
708 static int
709 sfc_flow_parse_udp(const struct rte_flow_item *item,
710 		   efx_filter_spec_t *efx_spec,
711 		   struct rte_flow_error *error)
712 {
713 	int rc;
714 	const struct rte_flow_item_udp *spec = NULL;
715 	const struct rte_flow_item_udp *mask = NULL;
716 	const struct rte_flow_item_udp supp_mask = {
717 		.hdr = {
718 			.src_port = 0xffff,
719 			.dst_port = 0xffff,
720 		}
721 	};
722 
723 	rc = sfc_flow_parse_init(item,
724 				 (const void **)&spec,
725 				 (const void **)&mask,
726 				 &supp_mask,
727 				 &rte_flow_item_udp_mask,
728 				 sizeof(struct rte_flow_item_udp),
729 				 error);
730 	if (rc != 0)
731 		return rc;
732 
733 	/*
734 	 * Filtering by UDP source and destination ports requires
735 	 * the appropriate IP_PROTO in hardware filters
736 	 */
737 	if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
738 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
739 		efx_spec->efs_ip_proto = EFX_IPPROTO_UDP;
740 	} else if (efx_spec->efs_ip_proto != EFX_IPPROTO_UDP) {
741 		rte_flow_error_set(error, EINVAL,
742 			RTE_FLOW_ERROR_TYPE_ITEM, item,
743 			"IP proto in pattern with UDP item should be appropriate");
744 		return -rte_errno;
745 	}
746 
747 	if (spec == NULL)
748 		return 0;
749 
750 	/*
751 	 * Source and destination ports are in big-endian byte order in item and
752 	 * in little-endian in efx_spec, so byte swap is used
753 	 */
754 	if (mask->hdr.src_port == supp_mask.hdr.src_port) {
755 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_PORT;
756 		efx_spec->efs_rem_port = rte_bswap16(spec->hdr.src_port);
757 	} else if (mask->hdr.src_port != 0) {
758 		goto fail_bad_mask;
759 	}
760 
761 	if (mask->hdr.dst_port == supp_mask.hdr.dst_port) {
762 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_PORT;
763 		efx_spec->efs_loc_port = rte_bswap16(spec->hdr.dst_port);
764 	} else if (mask->hdr.dst_port != 0) {
765 		goto fail_bad_mask;
766 	}
767 
768 	return 0;
769 
770 fail_bad_mask:
771 	rte_flow_error_set(error, EINVAL,
772 			   RTE_FLOW_ERROR_TYPE_ITEM, item,
773 			   "Bad mask in the UDP pattern item");
774 	return -rte_errno;
775 }
776 
777 /*
778  * Filters for encapsulated packets match based on the EtherType and IP
779  * protocol in the outer frame.
780  */
781 static int
782 sfc_flow_set_match_flags_for_encap_pkts(const struct rte_flow_item *item,
783 					efx_filter_spec_t *efx_spec,
784 					uint8_t ip_proto,
785 					struct rte_flow_error *error)
786 {
787 	if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
788 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
789 		efx_spec->efs_ip_proto = ip_proto;
790 	} else if (efx_spec->efs_ip_proto != ip_proto) {
791 		switch (ip_proto) {
792 		case EFX_IPPROTO_UDP:
793 			rte_flow_error_set(error, EINVAL,
794 				RTE_FLOW_ERROR_TYPE_ITEM, item,
795 				"Outer IP header protocol must be UDP "
796 				"in VxLAN/GENEVE pattern");
797 			return -rte_errno;
798 
799 		case EFX_IPPROTO_GRE:
800 			rte_flow_error_set(error, EINVAL,
801 				RTE_FLOW_ERROR_TYPE_ITEM, item,
802 				"Outer IP header protocol must be GRE "
803 				"in NVGRE pattern");
804 			return -rte_errno;
805 
806 		default:
807 			rte_flow_error_set(error, EINVAL,
808 				RTE_FLOW_ERROR_TYPE_ITEM, item,
809 				"Only VxLAN/GENEVE/NVGRE tunneling patterns "
810 				"are supported");
811 			return -rte_errno;
812 		}
813 	}
814 
815 	if (efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE &&
816 	    efx_spec->efs_ether_type != EFX_ETHER_TYPE_IPV4 &&
817 	    efx_spec->efs_ether_type != EFX_ETHER_TYPE_IPV6) {
818 		rte_flow_error_set(error, EINVAL,
819 			RTE_FLOW_ERROR_TYPE_ITEM, item,
820 			"Outer frame EtherType in pattern with tunneling "
821 			"must be IPv4 or IPv6");
822 		return -rte_errno;
823 	}
824 
825 	return 0;
826 }
827 
828 static int
829 sfc_flow_set_efx_spec_vni_or_vsid(efx_filter_spec_t *efx_spec,
830 				  const uint8_t *vni_or_vsid_val,
831 				  const uint8_t *vni_or_vsid_mask,
832 				  const struct rte_flow_item *item,
833 				  struct rte_flow_error *error)
834 {
835 	const uint8_t vni_or_vsid_full_mask[EFX_VNI_OR_VSID_LEN] = {
836 		0xff, 0xff, 0xff
837 	};
838 
839 	if (memcmp(vni_or_vsid_mask, vni_or_vsid_full_mask,
840 		   EFX_VNI_OR_VSID_LEN) == 0) {
841 		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_VNI_OR_VSID;
842 		rte_memcpy(efx_spec->efs_vni_or_vsid, vni_or_vsid_val,
843 			   EFX_VNI_OR_VSID_LEN);
844 	} else if (!sfc_flow_is_zero(vni_or_vsid_mask, EFX_VNI_OR_VSID_LEN)) {
845 		rte_flow_error_set(error, EINVAL,
846 				   RTE_FLOW_ERROR_TYPE_ITEM, item,
847 				   "Unsupported VNI/VSID mask");
848 		return -rte_errno;
849 	}
850 
851 	return 0;
852 }
853 
854 /**
855  * Convert VXLAN item to EFX filter specification.
856  *
857  * @param item[in]
858  *   Item specification. Only VXLAN network identifier field is supported.
859  *   If the mask is NULL, default mask will be used.
860  *   Ranging is not supported.
861  * @param efx_spec[in, out]
862  *   EFX filter specification to update.
863  * @param[out] error
864  *   Perform verbose error reporting if not NULL.
865  */
866 static int
867 sfc_flow_parse_vxlan(const struct rte_flow_item *item,
868 		     efx_filter_spec_t *efx_spec,
869 		     struct rte_flow_error *error)
870 {
871 	int rc;
872 	const struct rte_flow_item_vxlan *spec = NULL;
873 	const struct rte_flow_item_vxlan *mask = NULL;
874 	const struct rte_flow_item_vxlan supp_mask = {
875 		.vni = { 0xff, 0xff, 0xff }
876 	};
877 
878 	rc = sfc_flow_parse_init(item,
879 				 (const void **)&spec,
880 				 (const void **)&mask,
881 				 &supp_mask,
882 				 &rte_flow_item_vxlan_mask,
883 				 sizeof(struct rte_flow_item_vxlan),
884 				 error);
885 	if (rc != 0)
886 		return rc;
887 
888 	rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
889 						     EFX_IPPROTO_UDP, error);
890 	if (rc != 0)
891 		return rc;
892 
893 	efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_VXLAN;
894 	efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
895 
896 	if (spec == NULL)
897 		return 0;
898 
899 	rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->vni,
900 					       mask->vni, item, error);
901 
902 	return rc;
903 }
904 
905 /**
906  * Convert GENEVE item to EFX filter specification.
907  *
908  * @param item[in]
909  *   Item specification. Only Virtual Network Identifier and protocol type
910  *   fields are supported. But protocol type can be only Ethernet (0x6558).
911  *   If the mask is NULL, default mask will be used.
912  *   Ranging is not supported.
913  * @param efx_spec[in, out]
914  *   EFX filter specification to update.
915  * @param[out] error
916  *   Perform verbose error reporting if not NULL.
917  */
918 static int
919 sfc_flow_parse_geneve(const struct rte_flow_item *item,
920 		      efx_filter_spec_t *efx_spec,
921 		      struct rte_flow_error *error)
922 {
923 	int rc;
924 	const struct rte_flow_item_geneve *spec = NULL;
925 	const struct rte_flow_item_geneve *mask = NULL;
926 	const struct rte_flow_item_geneve supp_mask = {
927 		.protocol = RTE_BE16(0xffff),
928 		.vni = { 0xff, 0xff, 0xff }
929 	};
930 
931 	rc = sfc_flow_parse_init(item,
932 				 (const void **)&spec,
933 				 (const void **)&mask,
934 				 &supp_mask,
935 				 &rte_flow_item_geneve_mask,
936 				 sizeof(struct rte_flow_item_geneve),
937 				 error);
938 	if (rc != 0)
939 		return rc;
940 
941 	rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
942 						     EFX_IPPROTO_UDP, error);
943 	if (rc != 0)
944 		return rc;
945 
946 	efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_GENEVE;
947 	efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
948 
949 	if (spec == NULL)
950 		return 0;
951 
952 	if (mask->protocol == supp_mask.protocol) {
953 		if (spec->protocol != rte_cpu_to_be_16(ETHER_TYPE_TEB)) {
954 			rte_flow_error_set(error, EINVAL,
955 				RTE_FLOW_ERROR_TYPE_ITEM, item,
956 				"GENEVE encap. protocol must be Ethernet "
957 				"(0x6558) in the GENEVE pattern item");
958 			return -rte_errno;
959 		}
960 	} else if (mask->protocol != 0) {
961 		rte_flow_error_set(error, EINVAL,
962 			RTE_FLOW_ERROR_TYPE_ITEM, item,
963 			"Unsupported mask for GENEVE encap. protocol");
964 		return -rte_errno;
965 	}
966 
967 	rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->vni,
968 					       mask->vni, item, error);
969 
970 	return rc;
971 }
972 
973 /**
974  * Convert NVGRE item to EFX filter specification.
975  *
976  * @param item[in]
977  *   Item specification. Only virtual subnet ID field is supported.
978  *   If the mask is NULL, default mask will be used.
979  *   Ranging is not supported.
980  * @param efx_spec[in, out]
981  *   EFX filter specification to update.
982  * @param[out] error
983  *   Perform verbose error reporting if not NULL.
984  */
985 static int
986 sfc_flow_parse_nvgre(const struct rte_flow_item *item,
987 		     efx_filter_spec_t *efx_spec,
988 		     struct rte_flow_error *error)
989 {
990 	int rc;
991 	const struct rte_flow_item_nvgre *spec = NULL;
992 	const struct rte_flow_item_nvgre *mask = NULL;
993 	const struct rte_flow_item_nvgre supp_mask = {
994 		.tni = { 0xff, 0xff, 0xff }
995 	};
996 
997 	rc = sfc_flow_parse_init(item,
998 				 (const void **)&spec,
999 				 (const void **)&mask,
1000 				 &supp_mask,
1001 				 &rte_flow_item_nvgre_mask,
1002 				 sizeof(struct rte_flow_item_nvgre),
1003 				 error);
1004 	if (rc != 0)
1005 		return rc;
1006 
1007 	rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
1008 						     EFX_IPPROTO_GRE, error);
1009 	if (rc != 0)
1010 		return rc;
1011 
1012 	efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_NVGRE;
1013 	efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
1014 
1015 	if (spec == NULL)
1016 		return 0;
1017 
1018 	rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->tni,
1019 					       mask->tni, item, error);
1020 
1021 	return rc;
1022 }
1023 
1024 static const struct sfc_flow_item sfc_flow_items[] = {
1025 	{
1026 		.type = RTE_FLOW_ITEM_TYPE_VOID,
1027 		.prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
1028 		.layer = SFC_FLOW_ITEM_ANY_LAYER,
1029 		.parse = sfc_flow_parse_void,
1030 	},
1031 	{
1032 		.type = RTE_FLOW_ITEM_TYPE_ETH,
1033 		.prev_layer = SFC_FLOW_ITEM_START_LAYER,
1034 		.layer = SFC_FLOW_ITEM_L2,
1035 		.parse = sfc_flow_parse_eth,
1036 	},
1037 	{
1038 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
1039 		.prev_layer = SFC_FLOW_ITEM_L2,
1040 		.layer = SFC_FLOW_ITEM_L2,
1041 		.parse = sfc_flow_parse_vlan,
1042 	},
1043 	{
1044 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
1045 		.prev_layer = SFC_FLOW_ITEM_L2,
1046 		.layer = SFC_FLOW_ITEM_L3,
1047 		.parse = sfc_flow_parse_ipv4,
1048 	},
1049 	{
1050 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
1051 		.prev_layer = SFC_FLOW_ITEM_L2,
1052 		.layer = SFC_FLOW_ITEM_L3,
1053 		.parse = sfc_flow_parse_ipv6,
1054 	},
1055 	{
1056 		.type = RTE_FLOW_ITEM_TYPE_TCP,
1057 		.prev_layer = SFC_FLOW_ITEM_L3,
1058 		.layer = SFC_FLOW_ITEM_L4,
1059 		.parse = sfc_flow_parse_tcp,
1060 	},
1061 	{
1062 		.type = RTE_FLOW_ITEM_TYPE_UDP,
1063 		.prev_layer = SFC_FLOW_ITEM_L3,
1064 		.layer = SFC_FLOW_ITEM_L4,
1065 		.parse = sfc_flow_parse_udp,
1066 	},
1067 	{
1068 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
1069 		.prev_layer = SFC_FLOW_ITEM_L4,
1070 		.layer = SFC_FLOW_ITEM_START_LAYER,
1071 		.parse = sfc_flow_parse_vxlan,
1072 	},
1073 	{
1074 		.type = RTE_FLOW_ITEM_TYPE_GENEVE,
1075 		.prev_layer = SFC_FLOW_ITEM_L4,
1076 		.layer = SFC_FLOW_ITEM_START_LAYER,
1077 		.parse = sfc_flow_parse_geneve,
1078 	},
1079 	{
1080 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
1081 		.prev_layer = SFC_FLOW_ITEM_L3,
1082 		.layer = SFC_FLOW_ITEM_START_LAYER,
1083 		.parse = sfc_flow_parse_nvgre,
1084 	},
1085 };
1086 
1087 /*
1088  * Protocol-independent flow API support
1089  */
1090 static int
1091 sfc_flow_parse_attr(const struct rte_flow_attr *attr,
1092 		    struct rte_flow *flow,
1093 		    struct rte_flow_error *error)
1094 {
1095 	if (attr == NULL) {
1096 		rte_flow_error_set(error, EINVAL,
1097 				   RTE_FLOW_ERROR_TYPE_ATTR, NULL,
1098 				   "NULL attribute");
1099 		return -rte_errno;
1100 	}
1101 	if (attr->group != 0) {
1102 		rte_flow_error_set(error, ENOTSUP,
1103 				   RTE_FLOW_ERROR_TYPE_ATTR_GROUP, attr,
1104 				   "Groups are not supported");
1105 		return -rte_errno;
1106 	}
1107 	if (attr->priority != 0) {
1108 		rte_flow_error_set(error, ENOTSUP,
1109 				   RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY, attr,
1110 				   "Priorities are not supported");
1111 		return -rte_errno;
1112 	}
1113 	if (attr->egress != 0) {
1114 		rte_flow_error_set(error, ENOTSUP,
1115 				   RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attr,
1116 				   "Egress is not supported");
1117 		return -rte_errno;
1118 	}
1119 	if (attr->transfer != 0) {
1120 		rte_flow_error_set(error, ENOTSUP,
1121 				   RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER, attr,
1122 				   "Transfer is not supported");
1123 		return -rte_errno;
1124 	}
1125 	if (attr->ingress == 0) {
1126 		rte_flow_error_set(error, ENOTSUP,
1127 				   RTE_FLOW_ERROR_TYPE_ATTR_INGRESS, attr,
1128 				   "Only ingress is supported");
1129 		return -rte_errno;
1130 	}
1131 
1132 	flow->spec.template.efs_flags |= EFX_FILTER_FLAG_RX;
1133 	flow->spec.template.efs_rss_context = EFX_RSS_CONTEXT_DEFAULT;
1134 
1135 	return 0;
1136 }
1137 
1138 /* Get item from array sfc_flow_items */
1139 static const struct sfc_flow_item *
1140 sfc_flow_get_item(enum rte_flow_item_type type)
1141 {
1142 	unsigned int i;
1143 
1144 	for (i = 0; i < RTE_DIM(sfc_flow_items); i++)
1145 		if (sfc_flow_items[i].type == type)
1146 			return &sfc_flow_items[i];
1147 
1148 	return NULL;
1149 }
1150 
1151 static int
1152 sfc_flow_parse_pattern(const struct rte_flow_item pattern[],
1153 		       struct rte_flow *flow,
1154 		       struct rte_flow_error *error)
1155 {
1156 	int rc;
1157 	unsigned int prev_layer = SFC_FLOW_ITEM_ANY_LAYER;
1158 	boolean_t is_ifrm = B_FALSE;
1159 	const struct sfc_flow_item *item;
1160 
1161 	if (pattern == NULL) {
1162 		rte_flow_error_set(error, EINVAL,
1163 				   RTE_FLOW_ERROR_TYPE_ITEM_NUM, NULL,
1164 				   "NULL pattern");
1165 		return -rte_errno;
1166 	}
1167 
1168 	for (; pattern->type != RTE_FLOW_ITEM_TYPE_END; pattern++) {
1169 		item = sfc_flow_get_item(pattern->type);
1170 		if (item == NULL) {
1171 			rte_flow_error_set(error, ENOTSUP,
1172 					   RTE_FLOW_ERROR_TYPE_ITEM, pattern,
1173 					   "Unsupported pattern item");
1174 			return -rte_errno;
1175 		}
1176 
1177 		/*
1178 		 * Omitting one or several protocol layers at the beginning
1179 		 * of pattern is supported
1180 		 */
1181 		if (item->prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
1182 		    prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
1183 		    item->prev_layer != prev_layer) {
1184 			rte_flow_error_set(error, ENOTSUP,
1185 					   RTE_FLOW_ERROR_TYPE_ITEM, pattern,
1186 					   "Unexpected sequence of pattern items");
1187 			return -rte_errno;
1188 		}
1189 
1190 		/*
1191 		 * Allow only VOID and ETH pattern items in the inner frame.
1192 		 * Also check that there is only one tunneling protocol.
1193 		 */
1194 		switch (item->type) {
1195 		case RTE_FLOW_ITEM_TYPE_VOID:
1196 		case RTE_FLOW_ITEM_TYPE_ETH:
1197 			break;
1198 
1199 		case RTE_FLOW_ITEM_TYPE_VXLAN:
1200 		case RTE_FLOW_ITEM_TYPE_GENEVE:
1201 		case RTE_FLOW_ITEM_TYPE_NVGRE:
1202 			if (is_ifrm) {
1203 				rte_flow_error_set(error, EINVAL,
1204 					RTE_FLOW_ERROR_TYPE_ITEM,
1205 					pattern,
1206 					"More than one tunneling protocol");
1207 				return -rte_errno;
1208 			}
1209 			is_ifrm = B_TRUE;
1210 			break;
1211 
1212 		default:
1213 			if (is_ifrm) {
1214 				rte_flow_error_set(error, EINVAL,
1215 					RTE_FLOW_ERROR_TYPE_ITEM,
1216 					pattern,
1217 					"There is an unsupported pattern item "
1218 					"in the inner frame");
1219 				return -rte_errno;
1220 			}
1221 			break;
1222 		}
1223 
1224 		rc = item->parse(pattern, &flow->spec.template, error);
1225 		if (rc != 0)
1226 			return rc;
1227 
1228 		if (item->layer != SFC_FLOW_ITEM_ANY_LAYER)
1229 			prev_layer = item->layer;
1230 	}
1231 
1232 	return 0;
1233 }
1234 
1235 static int
1236 sfc_flow_parse_queue(struct sfc_adapter *sa,
1237 		     const struct rte_flow_action_queue *queue,
1238 		     struct rte_flow *flow)
1239 {
1240 	struct sfc_rxq *rxq;
1241 
1242 	if (queue->index >= sa->rxq_count)
1243 		return -EINVAL;
1244 
1245 	rxq = sa->rxq_info[queue->index].rxq;
1246 	flow->spec.template.efs_dmaq_id = (uint16_t)rxq->hw_index;
1247 
1248 	return 0;
1249 }
1250 
1251 #if EFSYS_OPT_RX_SCALE
1252 static int
1253 sfc_flow_parse_rss(struct sfc_adapter *sa,
1254 		   const struct rte_flow_action_rss *rss,
1255 		   struct rte_flow *flow)
1256 {
1257 	unsigned int rxq_sw_index;
1258 	struct sfc_rxq *rxq;
1259 	unsigned int rxq_hw_index_min;
1260 	unsigned int rxq_hw_index_max;
1261 	const uint8_t *rss_key;
1262 	struct sfc_flow_rss *sfc_rss_conf = &flow->rss_conf;
1263 	unsigned int i;
1264 
1265 	if (rss->queue_num == 0)
1266 		return -EINVAL;
1267 
1268 	rxq_sw_index = sa->rxq_count - 1;
1269 	rxq = sa->rxq_info[rxq_sw_index].rxq;
1270 	rxq_hw_index_min = rxq->hw_index;
1271 	rxq_hw_index_max = 0;
1272 
1273 	for (i = 0; i < rss->queue_num; ++i) {
1274 		rxq_sw_index = rss->queue[i];
1275 
1276 		if (rxq_sw_index >= sa->rxq_count)
1277 			return -EINVAL;
1278 
1279 		rxq = sa->rxq_info[rxq_sw_index].rxq;
1280 
1281 		if (rxq->hw_index < rxq_hw_index_min)
1282 			rxq_hw_index_min = rxq->hw_index;
1283 
1284 		if (rxq->hw_index > rxq_hw_index_max)
1285 			rxq_hw_index_max = rxq->hw_index;
1286 	}
1287 
1288 	switch (rss->func) {
1289 	case RTE_ETH_HASH_FUNCTION_DEFAULT:
1290 	case RTE_ETH_HASH_FUNCTION_TOEPLITZ:
1291 		break;
1292 	default:
1293 		return -EINVAL;
1294 	}
1295 
1296 	if (rss->level)
1297 		return -EINVAL;
1298 
1299 	if ((rss->types & ~SFC_RSS_OFFLOADS) != 0)
1300 		return -EINVAL;
1301 
1302 	if (rss->key_len) {
1303 		if (rss->key_len != sizeof(sa->rss_key))
1304 			return -EINVAL;
1305 
1306 		rss_key = rss->key;
1307 	} else {
1308 		rss_key = sa->rss_key;
1309 	}
1310 
1311 	flow->rss = B_TRUE;
1312 
1313 	sfc_rss_conf->rxq_hw_index_min = rxq_hw_index_min;
1314 	sfc_rss_conf->rxq_hw_index_max = rxq_hw_index_max;
1315 	sfc_rss_conf->rss_hash_types = sfc_rte_to_efx_hash_type(rss->types);
1316 	rte_memcpy(sfc_rss_conf->rss_key, rss_key, sizeof(sa->rss_key));
1317 
1318 	for (i = 0; i < RTE_DIM(sfc_rss_conf->rss_tbl); ++i) {
1319 		unsigned int rxq_sw_index = rss->queue[i % rss->queue_num];
1320 		struct sfc_rxq *rxq = sa->rxq_info[rxq_sw_index].rxq;
1321 
1322 		sfc_rss_conf->rss_tbl[i] = rxq->hw_index - rxq_hw_index_min;
1323 	}
1324 
1325 	return 0;
1326 }
1327 #endif /* EFSYS_OPT_RX_SCALE */
1328 
1329 static int
1330 sfc_flow_spec_flush(struct sfc_adapter *sa, struct sfc_flow_spec *spec,
1331 		    unsigned int filters_count)
1332 {
1333 	unsigned int i;
1334 	int ret = 0;
1335 
1336 	for (i = 0; i < filters_count; i++) {
1337 		int rc;
1338 
1339 		rc = efx_filter_remove(sa->nic, &spec->filters[i]);
1340 		if (ret == 0 && rc != 0) {
1341 			sfc_err(sa, "failed to remove filter specification "
1342 				"(rc = %d)", rc);
1343 			ret = rc;
1344 		}
1345 	}
1346 
1347 	return ret;
1348 }
1349 
1350 static int
1351 sfc_flow_spec_insert(struct sfc_adapter *sa, struct sfc_flow_spec *spec)
1352 {
1353 	unsigned int i;
1354 	int rc = 0;
1355 
1356 	for (i = 0; i < spec->count; i++) {
1357 		rc = efx_filter_insert(sa->nic, &spec->filters[i]);
1358 		if (rc != 0) {
1359 			sfc_flow_spec_flush(sa, spec, i);
1360 			break;
1361 		}
1362 	}
1363 
1364 	return rc;
1365 }
1366 
1367 static int
1368 sfc_flow_spec_remove(struct sfc_adapter *sa, struct sfc_flow_spec *spec)
1369 {
1370 	return sfc_flow_spec_flush(sa, spec, spec->count);
1371 }
1372 
1373 static int
1374 sfc_flow_filter_insert(struct sfc_adapter *sa,
1375 		       struct rte_flow *flow)
1376 {
1377 #if EFSYS_OPT_RX_SCALE
1378 	struct sfc_flow_rss *rss = &flow->rss_conf;
1379 	uint32_t efs_rss_context = EFX_RSS_CONTEXT_DEFAULT;
1380 	unsigned int i;
1381 	int rc = 0;
1382 
1383 	if (flow->rss) {
1384 		unsigned int rss_spread = MIN(rss->rxq_hw_index_max -
1385 					      rss->rxq_hw_index_min + 1,
1386 					      EFX_MAXRSS);
1387 
1388 		rc = efx_rx_scale_context_alloc(sa->nic,
1389 						EFX_RX_SCALE_EXCLUSIVE,
1390 						rss_spread,
1391 						&efs_rss_context);
1392 		if (rc != 0)
1393 			goto fail_scale_context_alloc;
1394 
1395 		rc = efx_rx_scale_mode_set(sa->nic, efs_rss_context,
1396 					   EFX_RX_HASHALG_TOEPLITZ,
1397 					   rss->rss_hash_types, B_TRUE);
1398 		if (rc != 0)
1399 			goto fail_scale_mode_set;
1400 
1401 		rc = efx_rx_scale_key_set(sa->nic, efs_rss_context,
1402 					  rss->rss_key,
1403 					  sizeof(sa->rss_key));
1404 		if (rc != 0)
1405 			goto fail_scale_key_set;
1406 
1407 		/*
1408 		 * At this point, fully elaborated filter specifications
1409 		 * have been produced from the template. To make sure that
1410 		 * RSS behaviour is consistent between them, set the same
1411 		 * RSS context value everywhere.
1412 		 */
1413 		for (i = 0; i < flow->spec.count; i++) {
1414 			efx_filter_spec_t *spec = &flow->spec.filters[i];
1415 
1416 			spec->efs_rss_context = efs_rss_context;
1417 			spec->efs_dmaq_id = rss->rxq_hw_index_min;
1418 			spec->efs_flags |= EFX_FILTER_FLAG_RX_RSS;
1419 		}
1420 	}
1421 
1422 	rc = sfc_flow_spec_insert(sa, &flow->spec);
1423 	if (rc != 0)
1424 		goto fail_filter_insert;
1425 
1426 	if (flow->rss) {
1427 		/*
1428 		 * Scale table is set after filter insertion because
1429 		 * the table entries are relative to the base RxQ ID
1430 		 * and the latter is submitted to the HW by means of
1431 		 * inserting a filter, so by the time of the request
1432 		 * the HW knows all the information needed to verify
1433 		 * the table entries, and the operation will succeed
1434 		 */
1435 		rc = efx_rx_scale_tbl_set(sa->nic, efs_rss_context,
1436 					  rss->rss_tbl, RTE_DIM(rss->rss_tbl));
1437 		if (rc != 0)
1438 			goto fail_scale_tbl_set;
1439 	}
1440 
1441 	return 0;
1442 
1443 fail_scale_tbl_set:
1444 	sfc_flow_spec_remove(sa, &flow->spec);
1445 
1446 fail_filter_insert:
1447 fail_scale_key_set:
1448 fail_scale_mode_set:
1449 	if (efs_rss_context != EFX_RSS_CONTEXT_DEFAULT)
1450 		efx_rx_scale_context_free(sa->nic, efs_rss_context);
1451 
1452 fail_scale_context_alloc:
1453 	return rc;
1454 #else /* !EFSYS_OPT_RX_SCALE */
1455 	return sfc_flow_spec_insert(sa, &flow->spec);
1456 #endif /* EFSYS_OPT_RX_SCALE */
1457 }
1458 
1459 static int
1460 sfc_flow_filter_remove(struct sfc_adapter *sa,
1461 		       struct rte_flow *flow)
1462 {
1463 	int rc = 0;
1464 
1465 	rc = sfc_flow_spec_remove(sa, &flow->spec);
1466 	if (rc != 0)
1467 		return rc;
1468 
1469 #if EFSYS_OPT_RX_SCALE
1470 	if (flow->rss) {
1471 		/*
1472 		 * All specifications for a given flow rule have the same RSS
1473 		 * context, so that RSS context value is taken from the first
1474 		 * filter specification
1475 		 */
1476 		efx_filter_spec_t *spec = &flow->spec.filters[0];
1477 
1478 		rc = efx_rx_scale_context_free(sa->nic, spec->efs_rss_context);
1479 	}
1480 #endif /* EFSYS_OPT_RX_SCALE */
1481 
1482 	return rc;
1483 }
1484 
1485 static int
1486 sfc_flow_parse_actions(struct sfc_adapter *sa,
1487 		       const struct rte_flow_action actions[],
1488 		       struct rte_flow *flow,
1489 		       struct rte_flow_error *error)
1490 {
1491 	int rc;
1492 	boolean_t is_specified = B_FALSE;
1493 
1494 	if (actions == NULL) {
1495 		rte_flow_error_set(error, EINVAL,
1496 				   RTE_FLOW_ERROR_TYPE_ACTION_NUM, NULL,
1497 				   "NULL actions");
1498 		return -rte_errno;
1499 	}
1500 
1501 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1502 		/* This one may appear anywhere multiple times. */
1503 		if (actions->type == RTE_FLOW_ACTION_TYPE_VOID)
1504 			continue;
1505 		/* Fate-deciding actions may appear exactly once. */
1506 		if (is_specified) {
1507 			rte_flow_error_set
1508 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
1509 				 actions,
1510 				 "Cannot combine several fate-deciding actions,"
1511 				 "choose between QUEUE, RSS or DROP");
1512 			return -rte_errno;
1513 		}
1514 		switch (actions->type) {
1515 		case RTE_FLOW_ACTION_TYPE_QUEUE:
1516 			rc = sfc_flow_parse_queue(sa, actions->conf, flow);
1517 			if (rc != 0) {
1518 				rte_flow_error_set(error, EINVAL,
1519 					RTE_FLOW_ERROR_TYPE_ACTION, actions,
1520 					"Bad QUEUE action");
1521 				return -rte_errno;
1522 			}
1523 
1524 			is_specified = B_TRUE;
1525 			break;
1526 
1527 #if EFSYS_OPT_RX_SCALE
1528 		case RTE_FLOW_ACTION_TYPE_RSS:
1529 			rc = sfc_flow_parse_rss(sa, actions->conf, flow);
1530 			if (rc != 0) {
1531 				rte_flow_error_set(error, rc,
1532 					RTE_FLOW_ERROR_TYPE_ACTION, actions,
1533 					"Bad RSS action");
1534 				return -rte_errno;
1535 			}
1536 
1537 			is_specified = B_TRUE;
1538 			break;
1539 #endif /* EFSYS_OPT_RX_SCALE */
1540 
1541 		case RTE_FLOW_ACTION_TYPE_DROP:
1542 			flow->spec.template.efs_dmaq_id =
1543 				EFX_FILTER_SPEC_RX_DMAQ_ID_DROP;
1544 
1545 			is_specified = B_TRUE;
1546 			break;
1547 
1548 		default:
1549 			rte_flow_error_set(error, ENOTSUP,
1550 					   RTE_FLOW_ERROR_TYPE_ACTION, actions,
1551 					   "Action is not supported");
1552 			return -rte_errno;
1553 		}
1554 	}
1555 
1556 	/* When fate is unknown, drop traffic. */
1557 	if (!is_specified) {
1558 		flow->spec.template.efs_dmaq_id =
1559 			EFX_FILTER_SPEC_RX_DMAQ_ID_DROP;
1560 	}
1561 
1562 	return 0;
1563 }
1564 
1565 /**
1566  * Set the EFX_FILTER_MATCH_UNKNOWN_UCAST_DST
1567  * and EFX_FILTER_MATCH_UNKNOWN_MCAST_DST match flags in the same
1568  * specifications after copying.
1569  *
1570  * @param spec[in, out]
1571  *   SFC flow specification to update.
1572  * @param filters_count_for_one_val[in]
1573  *   How many specifications should have the same match flag, what is the
1574  *   number of specifications before copying.
1575  * @param error[out]
1576  *   Perform verbose error reporting if not NULL.
1577  */
1578 static int
1579 sfc_flow_set_unknown_dst_flags(struct sfc_flow_spec *spec,
1580 			       unsigned int filters_count_for_one_val,
1581 			       struct rte_flow_error *error)
1582 {
1583 	unsigned int i;
1584 	static const efx_filter_match_flags_t vals[] = {
1585 		EFX_FILTER_MATCH_UNKNOWN_UCAST_DST,
1586 		EFX_FILTER_MATCH_UNKNOWN_MCAST_DST
1587 	};
1588 
1589 	if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1590 		rte_flow_error_set(error, EINVAL,
1591 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1592 			"Number of specifications is incorrect while copying "
1593 			"by unknown destination flags");
1594 		return -rte_errno;
1595 	}
1596 
1597 	for (i = 0; i < spec->count; i++) {
1598 		/* The check above ensures that divisor can't be zero here */
1599 		spec->filters[i].efs_match_flags |=
1600 			vals[i / filters_count_for_one_val];
1601 	}
1602 
1603 	return 0;
1604 }
1605 
1606 /**
1607  * Check that the following conditions are met:
1608  * - the list of supported filters has a filter
1609  *   with EFX_FILTER_MATCH_UNKNOWN_MCAST_DST flag instead of
1610  *   EFX_FILTER_MATCH_UNKNOWN_UCAST_DST, since this filter will also
1611  *   be inserted.
1612  *
1613  * @param match[in]
1614  *   The match flags of filter.
1615  * @param spec[in]
1616  *   Specification to be supplemented.
1617  * @param filter[in]
1618  *   SFC filter with list of supported filters.
1619  */
1620 static boolean_t
1621 sfc_flow_check_unknown_dst_flags(efx_filter_match_flags_t match,
1622 				 __rte_unused efx_filter_spec_t *spec,
1623 				 struct sfc_filter *filter)
1624 {
1625 	unsigned int i;
1626 	efx_filter_match_flags_t match_mcast_dst;
1627 
1628 	match_mcast_dst =
1629 		(match & ~EFX_FILTER_MATCH_UNKNOWN_UCAST_DST) |
1630 		EFX_FILTER_MATCH_UNKNOWN_MCAST_DST;
1631 	for (i = 0; i < filter->supported_match_num; i++) {
1632 		if (match_mcast_dst == filter->supported_match[i])
1633 			return B_TRUE;
1634 	}
1635 
1636 	return B_FALSE;
1637 }
1638 
1639 /**
1640  * Set the EFX_FILTER_MATCH_ETHER_TYPE match flag and EFX_ETHER_TYPE_IPV4 and
1641  * EFX_ETHER_TYPE_IPV6 values of the corresponding field in the same
1642  * specifications after copying.
1643  *
1644  * @param spec[in, out]
1645  *   SFC flow specification to update.
1646  * @param filters_count_for_one_val[in]
1647  *   How many specifications should have the same EtherType value, what is the
1648  *   number of specifications before copying.
1649  * @param error[out]
1650  *   Perform verbose error reporting if not NULL.
1651  */
1652 static int
1653 sfc_flow_set_ethertypes(struct sfc_flow_spec *spec,
1654 			unsigned int filters_count_for_one_val,
1655 			struct rte_flow_error *error)
1656 {
1657 	unsigned int i;
1658 	static const uint16_t vals[] = {
1659 		EFX_ETHER_TYPE_IPV4, EFX_ETHER_TYPE_IPV6
1660 	};
1661 
1662 	if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1663 		rte_flow_error_set(error, EINVAL,
1664 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1665 			"Number of specifications is incorrect "
1666 			"while copying by Ethertype");
1667 		return -rte_errno;
1668 	}
1669 
1670 	for (i = 0; i < spec->count; i++) {
1671 		spec->filters[i].efs_match_flags |=
1672 			EFX_FILTER_MATCH_ETHER_TYPE;
1673 
1674 		/*
1675 		 * The check above ensures that
1676 		 * filters_count_for_one_val is not 0
1677 		 */
1678 		spec->filters[i].efs_ether_type =
1679 			vals[i / filters_count_for_one_val];
1680 	}
1681 
1682 	return 0;
1683 }
1684 
1685 /**
1686  * Set the EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST and
1687  * EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST match flags in the same
1688  * specifications after copying.
1689  *
1690  * @param spec[in, out]
1691  *   SFC flow specification to update.
1692  * @param filters_count_for_one_val[in]
1693  *   How many specifications should have the same match flag, what is the
1694  *   number of specifications before copying.
1695  * @param error[out]
1696  *   Perform verbose error reporting if not NULL.
1697  */
1698 static int
1699 sfc_flow_set_ifrm_unknown_dst_flags(struct sfc_flow_spec *spec,
1700 				    unsigned int filters_count_for_one_val,
1701 				    struct rte_flow_error *error)
1702 {
1703 	unsigned int i;
1704 	static const efx_filter_match_flags_t vals[] = {
1705 		EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST,
1706 		EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST
1707 	};
1708 
1709 	if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1710 		rte_flow_error_set(error, EINVAL,
1711 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1712 			"Number of specifications is incorrect while copying "
1713 			"by inner frame unknown destination flags");
1714 		return -rte_errno;
1715 	}
1716 
1717 	for (i = 0; i < spec->count; i++) {
1718 		/* The check above ensures that divisor can't be zero here */
1719 		spec->filters[i].efs_match_flags |=
1720 			vals[i / filters_count_for_one_val];
1721 	}
1722 
1723 	return 0;
1724 }
1725 
1726 /**
1727  * Check that the following conditions are met:
1728  * - the specification corresponds to a filter for encapsulated traffic
1729  * - the list of supported filters has a filter
1730  *   with EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST flag instead of
1731  *   EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST, since this filter will also
1732  *   be inserted.
1733  *
1734  * @param match[in]
1735  *   The match flags of filter.
1736  * @param spec[in]
1737  *   Specification to be supplemented.
1738  * @param filter[in]
1739  *   SFC filter with list of supported filters.
1740  */
1741 static boolean_t
1742 sfc_flow_check_ifrm_unknown_dst_flags(efx_filter_match_flags_t match,
1743 				      efx_filter_spec_t *spec,
1744 				      struct sfc_filter *filter)
1745 {
1746 	unsigned int i;
1747 	efx_tunnel_protocol_t encap_type = spec->efs_encap_type;
1748 	efx_filter_match_flags_t match_mcast_dst;
1749 
1750 	if (encap_type == EFX_TUNNEL_PROTOCOL_NONE)
1751 		return B_FALSE;
1752 
1753 	match_mcast_dst =
1754 		(match & ~EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST) |
1755 		EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST;
1756 	for (i = 0; i < filter->supported_match_num; i++) {
1757 		if (match_mcast_dst == filter->supported_match[i])
1758 			return B_TRUE;
1759 	}
1760 
1761 	return B_FALSE;
1762 }
1763 
1764 /*
1765  * Match flags that can be automatically added to filters.
1766  * Selecting the last minimum when searching for the copy flag ensures that the
1767  * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST flag has a higher priority than
1768  * EFX_FILTER_MATCH_ETHER_TYPE. This is because the filter
1769  * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST is at the end of the list of supported
1770  * filters.
1771  */
1772 static const struct sfc_flow_copy_flag sfc_flow_copy_flags[] = {
1773 	{
1774 		.flag = EFX_FILTER_MATCH_UNKNOWN_UCAST_DST,
1775 		.vals_count = 2,
1776 		.set_vals = sfc_flow_set_unknown_dst_flags,
1777 		.spec_check = sfc_flow_check_unknown_dst_flags,
1778 	},
1779 	{
1780 		.flag = EFX_FILTER_MATCH_ETHER_TYPE,
1781 		.vals_count = 2,
1782 		.set_vals = sfc_flow_set_ethertypes,
1783 		.spec_check = NULL,
1784 	},
1785 	{
1786 		.flag = EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST,
1787 		.vals_count = 2,
1788 		.set_vals = sfc_flow_set_ifrm_unknown_dst_flags,
1789 		.spec_check = sfc_flow_check_ifrm_unknown_dst_flags,
1790 	},
1791 };
1792 
1793 /* Get item from array sfc_flow_copy_flags */
1794 static const struct sfc_flow_copy_flag *
1795 sfc_flow_get_copy_flag(efx_filter_match_flags_t flag)
1796 {
1797 	unsigned int i;
1798 
1799 	for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1800 		if (sfc_flow_copy_flags[i].flag == flag)
1801 			return &sfc_flow_copy_flags[i];
1802 	}
1803 
1804 	return NULL;
1805 }
1806 
1807 /**
1808  * Make copies of the specifications, set match flag and values
1809  * of the field that corresponds to it.
1810  *
1811  * @param spec[in, out]
1812  *   SFC flow specification to update.
1813  * @param flag[in]
1814  *   The match flag to add.
1815  * @param error[out]
1816  *   Perform verbose error reporting if not NULL.
1817  */
1818 static int
1819 sfc_flow_spec_add_match_flag(struct sfc_flow_spec *spec,
1820 			     efx_filter_match_flags_t flag,
1821 			     struct rte_flow_error *error)
1822 {
1823 	unsigned int i;
1824 	unsigned int new_filters_count;
1825 	unsigned int filters_count_for_one_val;
1826 	const struct sfc_flow_copy_flag *copy_flag;
1827 	int rc;
1828 
1829 	copy_flag = sfc_flow_get_copy_flag(flag);
1830 	if (copy_flag == NULL) {
1831 		rte_flow_error_set(error, ENOTSUP,
1832 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1833 				   "Unsupported spec field for copying");
1834 		return -rte_errno;
1835 	}
1836 
1837 	new_filters_count = spec->count * copy_flag->vals_count;
1838 	if (new_filters_count > SF_FLOW_SPEC_NB_FILTERS_MAX) {
1839 		rte_flow_error_set(error, EINVAL,
1840 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1841 			"Too much EFX specifications in the flow rule");
1842 		return -rte_errno;
1843 	}
1844 
1845 	/* Copy filters specifications */
1846 	for (i = spec->count; i < new_filters_count; i++)
1847 		spec->filters[i] = spec->filters[i - spec->count];
1848 
1849 	filters_count_for_one_val = spec->count;
1850 	spec->count = new_filters_count;
1851 
1852 	rc = copy_flag->set_vals(spec, filters_count_for_one_val, error);
1853 	if (rc != 0)
1854 		return rc;
1855 
1856 	return 0;
1857 }
1858 
1859 /**
1860  * Check that the given set of match flags missing in the original filter spec
1861  * could be covered by adding spec copies which specify the corresponding
1862  * flags and packet field values to match.
1863  *
1864  * @param miss_flags[in]
1865  *   Flags that are missing until the supported filter.
1866  * @param spec[in]
1867  *   Specification to be supplemented.
1868  * @param filter[in]
1869  *   SFC filter.
1870  *
1871  * @return
1872  *   Number of specifications after copy or 0, if the flags can not be added.
1873  */
1874 static unsigned int
1875 sfc_flow_check_missing_flags(efx_filter_match_flags_t miss_flags,
1876 			     efx_filter_spec_t *spec,
1877 			     struct sfc_filter *filter)
1878 {
1879 	unsigned int i;
1880 	efx_filter_match_flags_t copy_flags = 0;
1881 	efx_filter_match_flags_t flag;
1882 	efx_filter_match_flags_t match = spec->efs_match_flags | miss_flags;
1883 	sfc_flow_spec_check *check;
1884 	unsigned int multiplier = 1;
1885 
1886 	for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1887 		flag = sfc_flow_copy_flags[i].flag;
1888 		check = sfc_flow_copy_flags[i].spec_check;
1889 		if ((flag & miss_flags) == flag) {
1890 			if (check != NULL && (!check(match, spec, filter)))
1891 				continue;
1892 
1893 			copy_flags |= flag;
1894 			multiplier *= sfc_flow_copy_flags[i].vals_count;
1895 		}
1896 	}
1897 
1898 	if (copy_flags == miss_flags)
1899 		return multiplier;
1900 
1901 	return 0;
1902 }
1903 
1904 /**
1905  * Attempt to supplement the specification template to the minimally
1906  * supported set of match flags. To do this, it is necessary to copy
1907  * the specifications, filling them with the values of fields that
1908  * correspond to the missing flags.
1909  * The necessary and sufficient filter is built from the fewest number
1910  * of copies which could be made to cover the minimally required set
1911  * of flags.
1912  *
1913  * @param sa[in]
1914  *   SFC adapter.
1915  * @param spec[in, out]
1916  *   SFC flow specification to update.
1917  * @param error[out]
1918  *   Perform verbose error reporting if not NULL.
1919  */
1920 static int
1921 sfc_flow_spec_filters_complete(struct sfc_adapter *sa,
1922 			       struct sfc_flow_spec *spec,
1923 			       struct rte_flow_error *error)
1924 {
1925 	struct sfc_filter *filter = &sa->filter;
1926 	efx_filter_match_flags_t miss_flags;
1927 	efx_filter_match_flags_t min_miss_flags = 0;
1928 	efx_filter_match_flags_t match;
1929 	unsigned int min_multiplier = UINT_MAX;
1930 	unsigned int multiplier;
1931 	unsigned int i;
1932 	int rc;
1933 
1934 	match = spec->template.efs_match_flags;
1935 	for (i = 0; i < filter->supported_match_num; i++) {
1936 		if ((match & filter->supported_match[i]) == match) {
1937 			miss_flags = filter->supported_match[i] & (~match);
1938 			multiplier = sfc_flow_check_missing_flags(miss_flags,
1939 				&spec->template, filter);
1940 			if (multiplier > 0) {
1941 				if (multiplier <= min_multiplier) {
1942 					min_multiplier = multiplier;
1943 					min_miss_flags = miss_flags;
1944 				}
1945 			}
1946 		}
1947 	}
1948 
1949 	if (min_multiplier == UINT_MAX) {
1950 		rte_flow_error_set(error, ENOTSUP,
1951 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1952 				   "Flow rule pattern is not supported");
1953 		return -rte_errno;
1954 	}
1955 
1956 	for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1957 		efx_filter_match_flags_t flag = sfc_flow_copy_flags[i].flag;
1958 
1959 		if ((flag & min_miss_flags) == flag) {
1960 			rc = sfc_flow_spec_add_match_flag(spec, flag, error);
1961 			if (rc != 0)
1962 				return rc;
1963 		}
1964 	}
1965 
1966 	return 0;
1967 }
1968 
1969 /**
1970  * Check that set of match flags is referred to by a filter. Filter is
1971  * described by match flags with the ability to add OUTER_VID and INNER_VID
1972  * flags.
1973  *
1974  * @param match_flags[in]
1975  *   Set of match flags.
1976  * @param flags_pattern[in]
1977  *   Pattern of filter match flags.
1978  */
1979 static boolean_t
1980 sfc_flow_is_match_with_vids(efx_filter_match_flags_t match_flags,
1981 			    efx_filter_match_flags_t flags_pattern)
1982 {
1983 	if ((match_flags & flags_pattern) != flags_pattern)
1984 		return B_FALSE;
1985 
1986 	switch (match_flags & ~flags_pattern) {
1987 	case 0:
1988 	case EFX_FILTER_MATCH_OUTER_VID:
1989 	case EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_INNER_VID:
1990 		return B_TRUE;
1991 	default:
1992 		return B_FALSE;
1993 	}
1994 }
1995 
1996 /**
1997  * Check whether the spec maps to a hardware filter which is known to be
1998  * ineffective despite being valid.
1999  *
2000  * @param spec[in]
2001  *   SFC flow specification.
2002  */
2003 static boolean_t
2004 sfc_flow_is_match_flags_exception(struct sfc_flow_spec *spec)
2005 {
2006 	unsigned int i;
2007 	uint16_t ether_type;
2008 	uint8_t ip_proto;
2009 	efx_filter_match_flags_t match_flags;
2010 
2011 	for (i = 0; i < spec->count; i++) {
2012 		match_flags = spec->filters[i].efs_match_flags;
2013 
2014 		if (sfc_flow_is_match_with_vids(match_flags,
2015 						EFX_FILTER_MATCH_ETHER_TYPE) ||
2016 		    sfc_flow_is_match_with_vids(match_flags,
2017 						EFX_FILTER_MATCH_ETHER_TYPE |
2018 						EFX_FILTER_MATCH_LOC_MAC)) {
2019 			ether_type = spec->filters[i].efs_ether_type;
2020 			if (ether_type == EFX_ETHER_TYPE_IPV4 ||
2021 			    ether_type == EFX_ETHER_TYPE_IPV6)
2022 				return B_TRUE;
2023 		} else if (sfc_flow_is_match_with_vids(match_flags,
2024 				EFX_FILTER_MATCH_ETHER_TYPE |
2025 				EFX_FILTER_MATCH_IP_PROTO) ||
2026 			   sfc_flow_is_match_with_vids(match_flags,
2027 				EFX_FILTER_MATCH_ETHER_TYPE |
2028 				EFX_FILTER_MATCH_IP_PROTO |
2029 				EFX_FILTER_MATCH_LOC_MAC)) {
2030 			ip_proto = spec->filters[i].efs_ip_proto;
2031 			if (ip_proto == EFX_IPPROTO_TCP ||
2032 			    ip_proto == EFX_IPPROTO_UDP)
2033 				return B_TRUE;
2034 		}
2035 	}
2036 
2037 	return B_FALSE;
2038 }
2039 
2040 static int
2041 sfc_flow_validate_match_flags(struct sfc_adapter *sa,
2042 			      struct rte_flow *flow,
2043 			      struct rte_flow_error *error)
2044 {
2045 	efx_filter_spec_t *spec_tmpl = &flow->spec.template;
2046 	efx_filter_match_flags_t match_flags = spec_tmpl->efs_match_flags;
2047 	int rc;
2048 
2049 	/* Initialize the first filter spec with template */
2050 	flow->spec.filters[0] = *spec_tmpl;
2051 	flow->spec.count = 1;
2052 
2053 	if (!sfc_filter_is_match_supported(sa, match_flags)) {
2054 		rc = sfc_flow_spec_filters_complete(sa, &flow->spec, error);
2055 		if (rc != 0)
2056 			return rc;
2057 	}
2058 
2059 	if (sfc_flow_is_match_flags_exception(&flow->spec)) {
2060 		rte_flow_error_set(error, ENOTSUP,
2061 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2062 			"The flow rule pattern is unsupported");
2063 		return -rte_errno;
2064 	}
2065 
2066 	return 0;
2067 }
2068 
2069 static int
2070 sfc_flow_parse(struct rte_eth_dev *dev,
2071 	       const struct rte_flow_attr *attr,
2072 	       const struct rte_flow_item pattern[],
2073 	       const struct rte_flow_action actions[],
2074 	       struct rte_flow *flow,
2075 	       struct rte_flow_error *error)
2076 {
2077 	struct sfc_adapter *sa = dev->data->dev_private;
2078 	int rc;
2079 
2080 	rc = sfc_flow_parse_attr(attr, flow, error);
2081 	if (rc != 0)
2082 		goto fail_bad_value;
2083 
2084 	rc = sfc_flow_parse_pattern(pattern, flow, error);
2085 	if (rc != 0)
2086 		goto fail_bad_value;
2087 
2088 	rc = sfc_flow_parse_actions(sa, actions, flow, error);
2089 	if (rc != 0)
2090 		goto fail_bad_value;
2091 
2092 	rc = sfc_flow_validate_match_flags(sa, flow, error);
2093 	if (rc != 0)
2094 		goto fail_bad_value;
2095 
2096 	return 0;
2097 
2098 fail_bad_value:
2099 	return rc;
2100 }
2101 
2102 static int
2103 sfc_flow_validate(struct rte_eth_dev *dev,
2104 		  const struct rte_flow_attr *attr,
2105 		  const struct rte_flow_item pattern[],
2106 		  const struct rte_flow_action actions[],
2107 		  struct rte_flow_error *error)
2108 {
2109 	struct rte_flow flow;
2110 
2111 	memset(&flow, 0, sizeof(flow));
2112 
2113 	return sfc_flow_parse(dev, attr, pattern, actions, &flow, error);
2114 }
2115 
2116 static struct rte_flow *
2117 sfc_flow_create(struct rte_eth_dev *dev,
2118 		const struct rte_flow_attr *attr,
2119 		const struct rte_flow_item pattern[],
2120 		const struct rte_flow_action actions[],
2121 		struct rte_flow_error *error)
2122 {
2123 	struct sfc_adapter *sa = dev->data->dev_private;
2124 	struct rte_flow *flow = NULL;
2125 	int rc;
2126 
2127 	flow = rte_zmalloc("sfc_rte_flow", sizeof(*flow), 0);
2128 	if (flow == NULL) {
2129 		rte_flow_error_set(error, ENOMEM,
2130 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2131 				   "Failed to allocate memory");
2132 		goto fail_no_mem;
2133 	}
2134 
2135 	rc = sfc_flow_parse(dev, attr, pattern, actions, flow, error);
2136 	if (rc != 0)
2137 		goto fail_bad_value;
2138 
2139 	TAILQ_INSERT_TAIL(&sa->filter.flow_list, flow, entries);
2140 
2141 	sfc_adapter_lock(sa);
2142 
2143 	if (sa->state == SFC_ADAPTER_STARTED) {
2144 		rc = sfc_flow_filter_insert(sa, flow);
2145 		if (rc != 0) {
2146 			rte_flow_error_set(error, rc,
2147 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2148 				"Failed to insert filter");
2149 			goto fail_filter_insert;
2150 		}
2151 	}
2152 
2153 	sfc_adapter_unlock(sa);
2154 
2155 	return flow;
2156 
2157 fail_filter_insert:
2158 	TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2159 
2160 fail_bad_value:
2161 	rte_free(flow);
2162 	sfc_adapter_unlock(sa);
2163 
2164 fail_no_mem:
2165 	return NULL;
2166 }
2167 
2168 static int
2169 sfc_flow_remove(struct sfc_adapter *sa,
2170 		struct rte_flow *flow,
2171 		struct rte_flow_error *error)
2172 {
2173 	int rc = 0;
2174 
2175 	SFC_ASSERT(sfc_adapter_is_locked(sa));
2176 
2177 	if (sa->state == SFC_ADAPTER_STARTED) {
2178 		rc = sfc_flow_filter_remove(sa, flow);
2179 		if (rc != 0)
2180 			rte_flow_error_set(error, rc,
2181 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2182 				"Failed to destroy flow rule");
2183 	}
2184 
2185 	TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2186 	rte_free(flow);
2187 
2188 	return rc;
2189 }
2190 
2191 static int
2192 sfc_flow_destroy(struct rte_eth_dev *dev,
2193 		 struct rte_flow *flow,
2194 		 struct rte_flow_error *error)
2195 {
2196 	struct sfc_adapter *sa = dev->data->dev_private;
2197 	struct rte_flow *flow_ptr;
2198 	int rc = EINVAL;
2199 
2200 	sfc_adapter_lock(sa);
2201 
2202 	TAILQ_FOREACH(flow_ptr, &sa->filter.flow_list, entries) {
2203 		if (flow_ptr == flow)
2204 			rc = 0;
2205 	}
2206 	if (rc != 0) {
2207 		rte_flow_error_set(error, rc,
2208 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2209 				   "Failed to find flow rule to destroy");
2210 		goto fail_bad_value;
2211 	}
2212 
2213 	rc = sfc_flow_remove(sa, flow, error);
2214 
2215 fail_bad_value:
2216 	sfc_adapter_unlock(sa);
2217 
2218 	return -rc;
2219 }
2220 
2221 static int
2222 sfc_flow_flush(struct rte_eth_dev *dev,
2223 	       struct rte_flow_error *error)
2224 {
2225 	struct sfc_adapter *sa = dev->data->dev_private;
2226 	struct rte_flow *flow;
2227 	int rc = 0;
2228 	int ret = 0;
2229 
2230 	sfc_adapter_lock(sa);
2231 
2232 	while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
2233 		rc = sfc_flow_remove(sa, flow, error);
2234 		if (rc != 0)
2235 			ret = rc;
2236 	}
2237 
2238 	sfc_adapter_unlock(sa);
2239 
2240 	return -ret;
2241 }
2242 
2243 static int
2244 sfc_flow_isolate(struct rte_eth_dev *dev, int enable,
2245 		 struct rte_flow_error *error)
2246 {
2247 	struct sfc_adapter *sa = dev->data->dev_private;
2248 	struct sfc_port *port = &sa->port;
2249 	int ret = 0;
2250 
2251 	sfc_adapter_lock(sa);
2252 	if (sa->state != SFC_ADAPTER_INITIALIZED) {
2253 		rte_flow_error_set(error, EBUSY,
2254 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2255 				   NULL, "please close the port first");
2256 		ret = -rte_errno;
2257 	} else {
2258 		port->isolated = (enable) ? B_TRUE : B_FALSE;
2259 	}
2260 	sfc_adapter_unlock(sa);
2261 
2262 	return ret;
2263 }
2264 
2265 const struct rte_flow_ops sfc_flow_ops = {
2266 	.validate = sfc_flow_validate,
2267 	.create = sfc_flow_create,
2268 	.destroy = sfc_flow_destroy,
2269 	.flush = sfc_flow_flush,
2270 	.query = NULL,
2271 	.isolate = sfc_flow_isolate,
2272 };
2273 
2274 void
2275 sfc_flow_init(struct sfc_adapter *sa)
2276 {
2277 	SFC_ASSERT(sfc_adapter_is_locked(sa));
2278 
2279 	TAILQ_INIT(&sa->filter.flow_list);
2280 }
2281 
2282 void
2283 sfc_flow_fini(struct sfc_adapter *sa)
2284 {
2285 	struct rte_flow *flow;
2286 
2287 	SFC_ASSERT(sfc_adapter_is_locked(sa));
2288 
2289 	while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
2290 		TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2291 		rte_free(flow);
2292 	}
2293 }
2294 
2295 void
2296 sfc_flow_stop(struct sfc_adapter *sa)
2297 {
2298 	struct rte_flow *flow;
2299 
2300 	SFC_ASSERT(sfc_adapter_is_locked(sa));
2301 
2302 	TAILQ_FOREACH(flow, &sa->filter.flow_list, entries)
2303 		sfc_flow_filter_remove(sa, flow);
2304 }
2305 
2306 int
2307 sfc_flow_start(struct sfc_adapter *sa)
2308 {
2309 	struct rte_flow *flow;
2310 	int rc = 0;
2311 
2312 	sfc_log_init(sa, "entry");
2313 
2314 	SFC_ASSERT(sfc_adapter_is_locked(sa));
2315 
2316 	TAILQ_FOREACH(flow, &sa->filter.flow_list, entries) {
2317 		rc = sfc_flow_filter_insert(sa, flow);
2318 		if (rc != 0)
2319 			goto fail_bad_flow;
2320 	}
2321 
2322 	sfc_log_init(sa, "done");
2323 
2324 fail_bad_flow:
2325 	return rc;
2326 }
2327