1 /* SPDX-License-Identifier: BSD-3-Clause
3 * Copyright (c) 2017-2018 Solarflare Communications Inc.
6 * This software was jointly developed between OKTET Labs (under contract
7 * for Solarflare) and Solarflare Communications, Inc.
10 #include <rte_tailq.h>
11 #include <rte_common.h>
12 #include <rte_ethdev_driver.h>
13 #include <rte_eth_ctrl.h>
14 #include <rte_ether.h>
16 #include <rte_flow_driver.h>
22 #include "sfc_filter.h"
27 * At now flow API is implemented in such a manner that each
28 * flow rule is converted to one or more hardware filters.
29 * All elements of flow rule (attributes, pattern items, actions)
30 * correspond to one or more fields in the efx_filter_spec_s structure
31 * that is responsible for the hardware filter.
32 * If some required field is unset in the flow rule, then a handful
33 * of filter copies will be created to cover all possible values
37 enum sfc_flow_item_layers {
38 SFC_FLOW_ITEM_ANY_LAYER,
39 SFC_FLOW_ITEM_START_LAYER,
45 typedef int (sfc_flow_item_parse)(const struct rte_flow_item *item,
46 efx_filter_spec_t *spec,
47 struct rte_flow_error *error);
49 struct sfc_flow_item {
50 enum rte_flow_item_type type; /* Type of item */
51 enum sfc_flow_item_layers layer; /* Layer of item */
52 enum sfc_flow_item_layers prev_layer; /* Previous layer of item */
53 sfc_flow_item_parse *parse; /* Parsing function */
56 static sfc_flow_item_parse sfc_flow_parse_void;
57 static sfc_flow_item_parse sfc_flow_parse_eth;
58 static sfc_flow_item_parse sfc_flow_parse_vlan;
59 static sfc_flow_item_parse sfc_flow_parse_ipv4;
60 static sfc_flow_item_parse sfc_flow_parse_ipv6;
61 static sfc_flow_item_parse sfc_flow_parse_tcp;
62 static sfc_flow_item_parse sfc_flow_parse_udp;
63 static sfc_flow_item_parse sfc_flow_parse_vxlan;
64 static sfc_flow_item_parse sfc_flow_parse_geneve;
65 static sfc_flow_item_parse sfc_flow_parse_nvgre;
67 typedef int (sfc_flow_spec_set_vals)(struct sfc_flow_spec *spec,
68 unsigned int filters_count_for_one_val,
69 struct rte_flow_error *error);
71 typedef boolean_t (sfc_flow_spec_check)(efx_filter_match_flags_t match,
72 efx_filter_spec_t *spec,
73 struct sfc_filter *filter);
75 struct sfc_flow_copy_flag {
76 /* EFX filter specification match flag */
77 efx_filter_match_flags_t flag;
78 /* Number of values of corresponding field */
79 unsigned int vals_count;
80 /* Function to set values in specifications */
81 sfc_flow_spec_set_vals *set_vals;
83 * Function to check that the specification is suitable
84 * for adding this match flag
86 sfc_flow_spec_check *spec_check;
89 static sfc_flow_spec_set_vals sfc_flow_set_unknown_dst_flags;
90 static sfc_flow_spec_check sfc_flow_check_unknown_dst_flags;
91 static sfc_flow_spec_set_vals sfc_flow_set_ethertypes;
92 static sfc_flow_spec_set_vals sfc_flow_set_ifrm_unknown_dst_flags;
93 static sfc_flow_spec_check sfc_flow_check_ifrm_unknown_dst_flags;
96 sfc_flow_is_zero(const uint8_t *buf, unsigned int size)
101 for (i = 0; i < size; i++)
104 return (sum == 0) ? B_TRUE : B_FALSE;
108 * Validate item and prepare structures spec and mask for parsing
111 sfc_flow_parse_init(const struct rte_flow_item *item,
112 const void **spec_ptr,
113 const void **mask_ptr,
114 const void *supp_mask,
115 const void *def_mask,
117 struct rte_flow_error *error)
127 rte_flow_error_set(error, EINVAL,
128 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
133 if ((item->last != NULL || item->mask != NULL) && item->spec == NULL) {
134 rte_flow_error_set(error, EINVAL,
135 RTE_FLOW_ERROR_TYPE_ITEM, item,
136 "Mask or last is set without spec");
141 * If "mask" is not set, default mask is used,
142 * but if default mask is NULL, "mask" should be set
144 if (item->mask == NULL) {
145 if (def_mask == NULL) {
146 rte_flow_error_set(error, EINVAL,
147 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
148 "Mask should be specified");
164 * If field values in "last" are either 0 or equal to the corresponding
165 * values in "spec" then they are ignored
168 !sfc_flow_is_zero(last, size) &&
169 memcmp(last, spec, size) != 0) {
170 rte_flow_error_set(error, ENOTSUP,
171 RTE_FLOW_ERROR_TYPE_ITEM, item,
172 "Ranging is not supported");
176 if (supp_mask == NULL) {
177 rte_flow_error_set(error, EINVAL,
178 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
179 "Supported mask for item should be specified");
183 /* Check that mask and spec not asks for more match than supp_mask */
184 for (i = 0; i < size; i++) {
185 match = spec[i] | mask[i];
186 supp = ((const uint8_t *)supp_mask)[i];
188 if ((match | supp) != supp) {
189 rte_flow_error_set(error, ENOTSUP,
190 RTE_FLOW_ERROR_TYPE_ITEM, item,
191 "Item's field is not supported");
204 * Masking is not supported, so masks in items should be either
205 * full or empty (zeroed) and set only for supported fields which
206 * are specified in the supp_mask.
210 sfc_flow_parse_void(__rte_unused const struct rte_flow_item *item,
211 __rte_unused efx_filter_spec_t *efx_spec,
212 __rte_unused struct rte_flow_error *error)
218 * Convert Ethernet item to EFX filter specification.
221 * Item specification. Outer frame specification may only comprise
222 * source/destination addresses and Ethertype field.
223 * Inner frame specification may contain destination address only.
224 * There is support for individual/group mask as well as for empty and full.
225 * If the mask is NULL, default mask will be used. Ranging is not supported.
226 * @param efx_spec[in, out]
227 * EFX filter specification to update.
229 * Perform verbose error reporting if not NULL.
232 sfc_flow_parse_eth(const struct rte_flow_item *item,
233 efx_filter_spec_t *efx_spec,
234 struct rte_flow_error *error)
237 const struct rte_flow_item_eth *spec = NULL;
238 const struct rte_flow_item_eth *mask = NULL;
239 const struct rte_flow_item_eth supp_mask = {
240 .dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
241 .src.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
244 const struct rte_flow_item_eth ifrm_supp_mask = {
245 .dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
247 const uint8_t ig_mask[EFX_MAC_ADDR_LEN] = {
248 0x01, 0x00, 0x00, 0x00, 0x00, 0x00
250 const struct rte_flow_item_eth *supp_mask_p;
251 const struct rte_flow_item_eth *def_mask_p;
252 uint8_t *loc_mac = NULL;
253 boolean_t is_ifrm = (efx_spec->efs_encap_type !=
254 EFX_TUNNEL_PROTOCOL_NONE);
257 supp_mask_p = &ifrm_supp_mask;
258 def_mask_p = &ifrm_supp_mask;
259 loc_mac = efx_spec->efs_ifrm_loc_mac;
261 supp_mask_p = &supp_mask;
262 def_mask_p = &rte_flow_item_eth_mask;
263 loc_mac = efx_spec->efs_loc_mac;
266 rc = sfc_flow_parse_init(item,
267 (const void **)&spec,
268 (const void **)&mask,
269 supp_mask_p, def_mask_p,
270 sizeof(struct rte_flow_item_eth),
275 /* If "spec" is not set, could be any Ethernet */
279 if (is_same_ether_addr(&mask->dst, &supp_mask.dst)) {
280 efx_spec->efs_match_flags |= is_ifrm ?
281 EFX_FILTER_MATCH_IFRM_LOC_MAC :
282 EFX_FILTER_MATCH_LOC_MAC;
283 rte_memcpy(loc_mac, spec->dst.addr_bytes,
285 } else if (memcmp(mask->dst.addr_bytes, ig_mask,
286 EFX_MAC_ADDR_LEN) == 0) {
287 if (is_unicast_ether_addr(&spec->dst))
288 efx_spec->efs_match_flags |= is_ifrm ?
289 EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST :
290 EFX_FILTER_MATCH_UNKNOWN_UCAST_DST;
292 efx_spec->efs_match_flags |= is_ifrm ?
293 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST :
294 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST;
295 } else if (!is_zero_ether_addr(&mask->dst)) {
300 * ifrm_supp_mask ensures that the source address and
301 * ethertype masks are equal to zero in inner frame,
302 * so these fields are filled in only for the outer frame
304 if (is_same_ether_addr(&mask->src, &supp_mask.src)) {
305 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_MAC;
306 rte_memcpy(efx_spec->efs_rem_mac, spec->src.addr_bytes,
308 } else if (!is_zero_ether_addr(&mask->src)) {
313 * Ether type is in big-endian byte order in item and
314 * in little-endian in efx_spec, so byte swap is used
316 if (mask->type == supp_mask.type) {
317 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
318 efx_spec->efs_ether_type = rte_bswap16(spec->type);
319 } else if (mask->type != 0) {
326 rte_flow_error_set(error, EINVAL,
327 RTE_FLOW_ERROR_TYPE_ITEM, item,
328 "Bad mask in the ETH pattern item");
333 * Convert VLAN item to EFX filter specification.
336 * Item specification. Only VID field is supported.
337 * The mask can not be NULL. Ranging is not supported.
338 * @param efx_spec[in, out]
339 * EFX filter specification to update.
341 * Perform verbose error reporting if not NULL.
344 sfc_flow_parse_vlan(const struct rte_flow_item *item,
345 efx_filter_spec_t *efx_spec,
346 struct rte_flow_error *error)
350 const struct rte_flow_item_vlan *spec = NULL;
351 const struct rte_flow_item_vlan *mask = NULL;
352 const struct rte_flow_item_vlan supp_mask = {
353 .tci = rte_cpu_to_be_16(ETH_VLAN_ID_MAX),
356 rc = sfc_flow_parse_init(item,
357 (const void **)&spec,
358 (const void **)&mask,
361 sizeof(struct rte_flow_item_vlan),
367 * VID is in big-endian byte order in item and
368 * in little-endian in efx_spec, so byte swap is used.
369 * If two VLAN items are included, the first matches
370 * the outer tag and the next matches the inner tag.
372 if (mask->tci == supp_mask.tci) {
373 vid = rte_bswap16(spec->tci);
375 if (!(efx_spec->efs_match_flags &
376 EFX_FILTER_MATCH_OUTER_VID)) {
377 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_OUTER_VID;
378 efx_spec->efs_outer_vid = vid;
379 } else if (!(efx_spec->efs_match_flags &
380 EFX_FILTER_MATCH_INNER_VID)) {
381 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_INNER_VID;
382 efx_spec->efs_inner_vid = vid;
384 rte_flow_error_set(error, EINVAL,
385 RTE_FLOW_ERROR_TYPE_ITEM, item,
386 "More than two VLAN items");
390 rte_flow_error_set(error, EINVAL,
391 RTE_FLOW_ERROR_TYPE_ITEM, item,
392 "VLAN ID in TCI match is required");
400 * Convert IPv4 item to EFX filter specification.
403 * Item specification. Only source and destination addresses and
404 * protocol fields are supported. If the mask is NULL, default
405 * mask will be used. Ranging is not supported.
406 * @param efx_spec[in, out]
407 * EFX filter specification to update.
409 * Perform verbose error reporting if not NULL.
412 sfc_flow_parse_ipv4(const struct rte_flow_item *item,
413 efx_filter_spec_t *efx_spec,
414 struct rte_flow_error *error)
417 const struct rte_flow_item_ipv4 *spec = NULL;
418 const struct rte_flow_item_ipv4 *mask = NULL;
419 const uint16_t ether_type_ipv4 = rte_cpu_to_le_16(EFX_ETHER_TYPE_IPV4);
420 const struct rte_flow_item_ipv4 supp_mask = {
422 .src_addr = 0xffffffff,
423 .dst_addr = 0xffffffff,
424 .next_proto_id = 0xff,
428 rc = sfc_flow_parse_init(item,
429 (const void **)&spec,
430 (const void **)&mask,
432 &rte_flow_item_ipv4_mask,
433 sizeof(struct rte_flow_item_ipv4),
439 * Filtering by IPv4 source and destination addresses requires
440 * the appropriate ETHER_TYPE in hardware filters
442 if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE)) {
443 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
444 efx_spec->efs_ether_type = ether_type_ipv4;
445 } else if (efx_spec->efs_ether_type != ether_type_ipv4) {
446 rte_flow_error_set(error, EINVAL,
447 RTE_FLOW_ERROR_TYPE_ITEM, item,
448 "Ethertype in pattern with IPV4 item should be appropriate");
456 * IPv4 addresses are in big-endian byte order in item and in
459 if (mask->hdr.src_addr == supp_mask.hdr.src_addr) {
460 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_HOST;
461 efx_spec->efs_rem_host.eo_u32[0] = spec->hdr.src_addr;
462 } else if (mask->hdr.src_addr != 0) {
466 if (mask->hdr.dst_addr == supp_mask.hdr.dst_addr) {
467 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_HOST;
468 efx_spec->efs_loc_host.eo_u32[0] = spec->hdr.dst_addr;
469 } else if (mask->hdr.dst_addr != 0) {
473 if (mask->hdr.next_proto_id == supp_mask.hdr.next_proto_id) {
474 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
475 efx_spec->efs_ip_proto = spec->hdr.next_proto_id;
476 } else if (mask->hdr.next_proto_id != 0) {
483 rte_flow_error_set(error, EINVAL,
484 RTE_FLOW_ERROR_TYPE_ITEM, item,
485 "Bad mask in the IPV4 pattern item");
490 * Convert IPv6 item to EFX filter specification.
493 * Item specification. Only source and destination addresses and
494 * next header fields are supported. If the mask is NULL, default
495 * mask will be used. Ranging is not supported.
496 * @param efx_spec[in, out]
497 * EFX filter specification to update.
499 * Perform verbose error reporting if not NULL.
502 sfc_flow_parse_ipv6(const struct rte_flow_item *item,
503 efx_filter_spec_t *efx_spec,
504 struct rte_flow_error *error)
507 const struct rte_flow_item_ipv6 *spec = NULL;
508 const struct rte_flow_item_ipv6 *mask = NULL;
509 const uint16_t ether_type_ipv6 = rte_cpu_to_le_16(EFX_ETHER_TYPE_IPV6);
510 const struct rte_flow_item_ipv6 supp_mask = {
512 .src_addr = { 0xff, 0xff, 0xff, 0xff,
513 0xff, 0xff, 0xff, 0xff,
514 0xff, 0xff, 0xff, 0xff,
515 0xff, 0xff, 0xff, 0xff },
516 .dst_addr = { 0xff, 0xff, 0xff, 0xff,
517 0xff, 0xff, 0xff, 0xff,
518 0xff, 0xff, 0xff, 0xff,
519 0xff, 0xff, 0xff, 0xff },
524 rc = sfc_flow_parse_init(item,
525 (const void **)&spec,
526 (const void **)&mask,
528 &rte_flow_item_ipv6_mask,
529 sizeof(struct rte_flow_item_ipv6),
535 * Filtering by IPv6 source and destination addresses requires
536 * the appropriate ETHER_TYPE in hardware filters
538 if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE)) {
539 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
540 efx_spec->efs_ether_type = ether_type_ipv6;
541 } else if (efx_spec->efs_ether_type != ether_type_ipv6) {
542 rte_flow_error_set(error, EINVAL,
543 RTE_FLOW_ERROR_TYPE_ITEM, item,
544 "Ethertype in pattern with IPV6 item should be appropriate");
552 * IPv6 addresses are in big-endian byte order in item and in
555 if (memcmp(mask->hdr.src_addr, supp_mask.hdr.src_addr,
556 sizeof(mask->hdr.src_addr)) == 0) {
557 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_HOST;
559 RTE_BUILD_BUG_ON(sizeof(efx_spec->efs_rem_host) !=
560 sizeof(spec->hdr.src_addr));
561 rte_memcpy(&efx_spec->efs_rem_host, spec->hdr.src_addr,
562 sizeof(efx_spec->efs_rem_host));
563 } else if (!sfc_flow_is_zero(mask->hdr.src_addr,
564 sizeof(mask->hdr.src_addr))) {
568 if (memcmp(mask->hdr.dst_addr, supp_mask.hdr.dst_addr,
569 sizeof(mask->hdr.dst_addr)) == 0) {
570 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_HOST;
572 RTE_BUILD_BUG_ON(sizeof(efx_spec->efs_loc_host) !=
573 sizeof(spec->hdr.dst_addr));
574 rte_memcpy(&efx_spec->efs_loc_host, spec->hdr.dst_addr,
575 sizeof(efx_spec->efs_loc_host));
576 } else if (!sfc_flow_is_zero(mask->hdr.dst_addr,
577 sizeof(mask->hdr.dst_addr))) {
581 if (mask->hdr.proto == supp_mask.hdr.proto) {
582 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
583 efx_spec->efs_ip_proto = spec->hdr.proto;
584 } else if (mask->hdr.proto != 0) {
591 rte_flow_error_set(error, EINVAL,
592 RTE_FLOW_ERROR_TYPE_ITEM, item,
593 "Bad mask in the IPV6 pattern item");
598 * Convert TCP item to EFX filter specification.
601 * Item specification. Only source and destination ports fields
602 * are supported. If the mask is NULL, default mask will be used.
603 * Ranging is not supported.
604 * @param efx_spec[in, out]
605 * EFX filter specification to update.
607 * Perform verbose error reporting if not NULL.
610 sfc_flow_parse_tcp(const struct rte_flow_item *item,
611 efx_filter_spec_t *efx_spec,
612 struct rte_flow_error *error)
615 const struct rte_flow_item_tcp *spec = NULL;
616 const struct rte_flow_item_tcp *mask = NULL;
617 const struct rte_flow_item_tcp supp_mask = {
624 rc = sfc_flow_parse_init(item,
625 (const void **)&spec,
626 (const void **)&mask,
628 &rte_flow_item_tcp_mask,
629 sizeof(struct rte_flow_item_tcp),
635 * Filtering by TCP source and destination ports requires
636 * the appropriate IP_PROTO in hardware filters
638 if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
639 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
640 efx_spec->efs_ip_proto = EFX_IPPROTO_TCP;
641 } else if (efx_spec->efs_ip_proto != EFX_IPPROTO_TCP) {
642 rte_flow_error_set(error, EINVAL,
643 RTE_FLOW_ERROR_TYPE_ITEM, item,
644 "IP proto in pattern with TCP item should be appropriate");
652 * Source and destination ports are in big-endian byte order in item and
653 * in little-endian in efx_spec, so byte swap is used
655 if (mask->hdr.src_port == supp_mask.hdr.src_port) {
656 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_PORT;
657 efx_spec->efs_rem_port = rte_bswap16(spec->hdr.src_port);
658 } else if (mask->hdr.src_port != 0) {
662 if (mask->hdr.dst_port == supp_mask.hdr.dst_port) {
663 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_PORT;
664 efx_spec->efs_loc_port = rte_bswap16(spec->hdr.dst_port);
665 } else if (mask->hdr.dst_port != 0) {
672 rte_flow_error_set(error, EINVAL,
673 RTE_FLOW_ERROR_TYPE_ITEM, item,
674 "Bad mask in the TCP pattern item");
679 * Convert UDP item to EFX filter specification.
682 * Item specification. Only source and destination ports fields
683 * are supported. If the mask is NULL, default mask will be used.
684 * Ranging is not supported.
685 * @param efx_spec[in, out]
686 * EFX filter specification to update.
688 * Perform verbose error reporting if not NULL.
691 sfc_flow_parse_udp(const struct rte_flow_item *item,
692 efx_filter_spec_t *efx_spec,
693 struct rte_flow_error *error)
696 const struct rte_flow_item_udp *spec = NULL;
697 const struct rte_flow_item_udp *mask = NULL;
698 const struct rte_flow_item_udp supp_mask = {
705 rc = sfc_flow_parse_init(item,
706 (const void **)&spec,
707 (const void **)&mask,
709 &rte_flow_item_udp_mask,
710 sizeof(struct rte_flow_item_udp),
716 * Filtering by UDP source and destination ports requires
717 * the appropriate IP_PROTO in hardware filters
719 if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
720 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
721 efx_spec->efs_ip_proto = EFX_IPPROTO_UDP;
722 } else if (efx_spec->efs_ip_proto != EFX_IPPROTO_UDP) {
723 rte_flow_error_set(error, EINVAL,
724 RTE_FLOW_ERROR_TYPE_ITEM, item,
725 "IP proto in pattern with UDP item should be appropriate");
733 * Source and destination ports are in big-endian byte order in item and
734 * in little-endian in efx_spec, so byte swap is used
736 if (mask->hdr.src_port == supp_mask.hdr.src_port) {
737 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_PORT;
738 efx_spec->efs_rem_port = rte_bswap16(spec->hdr.src_port);
739 } else if (mask->hdr.src_port != 0) {
743 if (mask->hdr.dst_port == supp_mask.hdr.dst_port) {
744 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_PORT;
745 efx_spec->efs_loc_port = rte_bswap16(spec->hdr.dst_port);
746 } else if (mask->hdr.dst_port != 0) {
753 rte_flow_error_set(error, EINVAL,
754 RTE_FLOW_ERROR_TYPE_ITEM, item,
755 "Bad mask in the UDP pattern item");
760 * Filters for encapsulated packets match based on the EtherType and IP
761 * protocol in the outer frame.
764 sfc_flow_set_match_flags_for_encap_pkts(const struct rte_flow_item *item,
765 efx_filter_spec_t *efx_spec,
767 struct rte_flow_error *error)
769 if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
770 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
771 efx_spec->efs_ip_proto = ip_proto;
772 } else if (efx_spec->efs_ip_proto != ip_proto) {
774 case EFX_IPPROTO_UDP:
775 rte_flow_error_set(error, EINVAL,
776 RTE_FLOW_ERROR_TYPE_ITEM, item,
777 "Outer IP header protocol must be UDP "
778 "in VxLAN/GENEVE pattern");
781 case EFX_IPPROTO_GRE:
782 rte_flow_error_set(error, EINVAL,
783 RTE_FLOW_ERROR_TYPE_ITEM, item,
784 "Outer IP header protocol must be GRE "
789 rte_flow_error_set(error, EINVAL,
790 RTE_FLOW_ERROR_TYPE_ITEM, item,
791 "Only VxLAN/GENEVE/NVGRE tunneling patterns "
797 if (efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE &&
798 efx_spec->efs_ether_type != EFX_ETHER_TYPE_IPV4 &&
799 efx_spec->efs_ether_type != EFX_ETHER_TYPE_IPV6) {
800 rte_flow_error_set(error, EINVAL,
801 RTE_FLOW_ERROR_TYPE_ITEM, item,
802 "Outer frame EtherType in pattern with tunneling "
803 "must be IPv4 or IPv6");
811 sfc_flow_set_efx_spec_vni_or_vsid(efx_filter_spec_t *efx_spec,
812 const uint8_t *vni_or_vsid_val,
813 const uint8_t *vni_or_vsid_mask,
814 const struct rte_flow_item *item,
815 struct rte_flow_error *error)
817 const uint8_t vni_or_vsid_full_mask[EFX_VNI_OR_VSID_LEN] = {
821 if (memcmp(vni_or_vsid_mask, vni_or_vsid_full_mask,
822 EFX_VNI_OR_VSID_LEN) == 0) {
823 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_VNI_OR_VSID;
824 rte_memcpy(efx_spec->efs_vni_or_vsid, vni_or_vsid_val,
825 EFX_VNI_OR_VSID_LEN);
826 } else if (!sfc_flow_is_zero(vni_or_vsid_mask, EFX_VNI_OR_VSID_LEN)) {
827 rte_flow_error_set(error, EINVAL,
828 RTE_FLOW_ERROR_TYPE_ITEM, item,
829 "Unsupported VNI/VSID mask");
837 * Convert VXLAN item to EFX filter specification.
840 * Item specification. Only VXLAN network identifier field is supported.
841 * If the mask is NULL, default mask will be used.
842 * Ranging is not supported.
843 * @param efx_spec[in, out]
844 * EFX filter specification to update.
846 * Perform verbose error reporting if not NULL.
849 sfc_flow_parse_vxlan(const struct rte_flow_item *item,
850 efx_filter_spec_t *efx_spec,
851 struct rte_flow_error *error)
854 const struct rte_flow_item_vxlan *spec = NULL;
855 const struct rte_flow_item_vxlan *mask = NULL;
856 const struct rte_flow_item_vxlan supp_mask = {
857 .vni = { 0xff, 0xff, 0xff }
860 rc = sfc_flow_parse_init(item,
861 (const void **)&spec,
862 (const void **)&mask,
864 &rte_flow_item_vxlan_mask,
865 sizeof(struct rte_flow_item_vxlan),
870 rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
871 EFX_IPPROTO_UDP, error);
875 efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_VXLAN;
876 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
881 rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->vni,
882 mask->vni, item, error);
888 * Convert GENEVE item to EFX filter specification.
891 * Item specification. Only Virtual Network Identifier and protocol type
892 * fields are supported. But protocol type can be only Ethernet (0x6558).
893 * If the mask is NULL, default mask will be used.
894 * Ranging is not supported.
895 * @param efx_spec[in, out]
896 * EFX filter specification to update.
898 * Perform verbose error reporting if not NULL.
901 sfc_flow_parse_geneve(const struct rte_flow_item *item,
902 efx_filter_spec_t *efx_spec,
903 struct rte_flow_error *error)
906 const struct rte_flow_item_geneve *spec = NULL;
907 const struct rte_flow_item_geneve *mask = NULL;
908 const struct rte_flow_item_geneve supp_mask = {
909 .protocol = RTE_BE16(0xffff),
910 .vni = { 0xff, 0xff, 0xff }
913 rc = sfc_flow_parse_init(item,
914 (const void **)&spec,
915 (const void **)&mask,
917 &rte_flow_item_geneve_mask,
918 sizeof(struct rte_flow_item_geneve),
923 rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
924 EFX_IPPROTO_UDP, error);
928 efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_GENEVE;
929 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
934 if (mask->protocol == supp_mask.protocol) {
935 if (spec->protocol != rte_cpu_to_be_16(ETHER_TYPE_TEB)) {
936 rte_flow_error_set(error, EINVAL,
937 RTE_FLOW_ERROR_TYPE_ITEM, item,
938 "GENEVE encap. protocol must be Ethernet "
939 "(0x6558) in the GENEVE pattern item");
942 } else if (mask->protocol != 0) {
943 rte_flow_error_set(error, EINVAL,
944 RTE_FLOW_ERROR_TYPE_ITEM, item,
945 "Unsupported mask for GENEVE encap. protocol");
949 rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->vni,
950 mask->vni, item, error);
956 * Convert NVGRE item to EFX filter specification.
959 * Item specification. Only virtual subnet ID field is supported.
960 * If the mask is NULL, default mask will be used.
961 * Ranging is not supported.
962 * @param efx_spec[in, out]
963 * EFX filter specification to update.
965 * Perform verbose error reporting if not NULL.
968 sfc_flow_parse_nvgre(const struct rte_flow_item *item,
969 efx_filter_spec_t *efx_spec,
970 struct rte_flow_error *error)
973 const struct rte_flow_item_nvgre *spec = NULL;
974 const struct rte_flow_item_nvgre *mask = NULL;
975 const struct rte_flow_item_nvgre supp_mask = {
976 .tni = { 0xff, 0xff, 0xff }
979 rc = sfc_flow_parse_init(item,
980 (const void **)&spec,
981 (const void **)&mask,
983 &rte_flow_item_nvgre_mask,
984 sizeof(struct rte_flow_item_nvgre),
989 rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
990 EFX_IPPROTO_GRE, error);
994 efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_NVGRE;
995 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
1000 rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->tni,
1001 mask->tni, item, error);
1006 static const struct sfc_flow_item sfc_flow_items[] = {
1008 .type = RTE_FLOW_ITEM_TYPE_VOID,
1009 .prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
1010 .layer = SFC_FLOW_ITEM_ANY_LAYER,
1011 .parse = sfc_flow_parse_void,
1014 .type = RTE_FLOW_ITEM_TYPE_ETH,
1015 .prev_layer = SFC_FLOW_ITEM_START_LAYER,
1016 .layer = SFC_FLOW_ITEM_L2,
1017 .parse = sfc_flow_parse_eth,
1020 .type = RTE_FLOW_ITEM_TYPE_VLAN,
1021 .prev_layer = SFC_FLOW_ITEM_L2,
1022 .layer = SFC_FLOW_ITEM_L2,
1023 .parse = sfc_flow_parse_vlan,
1026 .type = RTE_FLOW_ITEM_TYPE_IPV4,
1027 .prev_layer = SFC_FLOW_ITEM_L2,
1028 .layer = SFC_FLOW_ITEM_L3,
1029 .parse = sfc_flow_parse_ipv4,
1032 .type = RTE_FLOW_ITEM_TYPE_IPV6,
1033 .prev_layer = SFC_FLOW_ITEM_L2,
1034 .layer = SFC_FLOW_ITEM_L3,
1035 .parse = sfc_flow_parse_ipv6,
1038 .type = RTE_FLOW_ITEM_TYPE_TCP,
1039 .prev_layer = SFC_FLOW_ITEM_L3,
1040 .layer = SFC_FLOW_ITEM_L4,
1041 .parse = sfc_flow_parse_tcp,
1044 .type = RTE_FLOW_ITEM_TYPE_UDP,
1045 .prev_layer = SFC_FLOW_ITEM_L3,
1046 .layer = SFC_FLOW_ITEM_L4,
1047 .parse = sfc_flow_parse_udp,
1050 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
1051 .prev_layer = SFC_FLOW_ITEM_L4,
1052 .layer = SFC_FLOW_ITEM_START_LAYER,
1053 .parse = sfc_flow_parse_vxlan,
1056 .type = RTE_FLOW_ITEM_TYPE_GENEVE,
1057 .prev_layer = SFC_FLOW_ITEM_L4,
1058 .layer = SFC_FLOW_ITEM_START_LAYER,
1059 .parse = sfc_flow_parse_geneve,
1062 .type = RTE_FLOW_ITEM_TYPE_NVGRE,
1063 .prev_layer = SFC_FLOW_ITEM_L3,
1064 .layer = SFC_FLOW_ITEM_START_LAYER,
1065 .parse = sfc_flow_parse_nvgre,
1070 * Protocol-independent flow API support
1073 sfc_flow_parse_attr(const struct rte_flow_attr *attr,
1074 struct rte_flow *flow,
1075 struct rte_flow_error *error)
1078 rte_flow_error_set(error, EINVAL,
1079 RTE_FLOW_ERROR_TYPE_ATTR, NULL,
1083 if (attr->group != 0) {
1084 rte_flow_error_set(error, ENOTSUP,
1085 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, attr,
1086 "Groups are not supported");
1089 if (attr->priority != 0) {
1090 rte_flow_error_set(error, ENOTSUP,
1091 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY, attr,
1092 "Priorities are not supported");
1095 if (attr->egress != 0) {
1096 rte_flow_error_set(error, ENOTSUP,
1097 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attr,
1098 "Egress is not supported");
1101 if (attr->ingress == 0) {
1102 rte_flow_error_set(error, ENOTSUP,
1103 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS, attr,
1104 "Only ingress is supported");
1108 flow->spec.template.efs_flags |= EFX_FILTER_FLAG_RX;
1109 flow->spec.template.efs_rss_context = EFX_RSS_CONTEXT_DEFAULT;
1114 /* Get item from array sfc_flow_items */
1115 static const struct sfc_flow_item *
1116 sfc_flow_get_item(enum rte_flow_item_type type)
1120 for (i = 0; i < RTE_DIM(sfc_flow_items); i++)
1121 if (sfc_flow_items[i].type == type)
1122 return &sfc_flow_items[i];
1128 sfc_flow_parse_pattern(const struct rte_flow_item pattern[],
1129 struct rte_flow *flow,
1130 struct rte_flow_error *error)
1133 unsigned int prev_layer = SFC_FLOW_ITEM_ANY_LAYER;
1134 boolean_t is_ifrm = B_FALSE;
1135 const struct sfc_flow_item *item;
1137 if (pattern == NULL) {
1138 rte_flow_error_set(error, EINVAL,
1139 RTE_FLOW_ERROR_TYPE_ITEM_NUM, NULL,
1144 for (; pattern->type != RTE_FLOW_ITEM_TYPE_END; pattern++) {
1145 item = sfc_flow_get_item(pattern->type);
1147 rte_flow_error_set(error, ENOTSUP,
1148 RTE_FLOW_ERROR_TYPE_ITEM, pattern,
1149 "Unsupported pattern item");
1154 * Omitting one or several protocol layers at the beginning
1155 * of pattern is supported
1157 if (item->prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
1158 prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
1159 item->prev_layer != prev_layer) {
1160 rte_flow_error_set(error, ENOTSUP,
1161 RTE_FLOW_ERROR_TYPE_ITEM, pattern,
1162 "Unexpected sequence of pattern items");
1167 * Allow only VOID and ETH pattern items in the inner frame.
1168 * Also check that there is only one tunneling protocol.
1170 switch (item->type) {
1171 case RTE_FLOW_ITEM_TYPE_VOID:
1172 case RTE_FLOW_ITEM_TYPE_ETH:
1175 case RTE_FLOW_ITEM_TYPE_VXLAN:
1176 case RTE_FLOW_ITEM_TYPE_GENEVE:
1177 case RTE_FLOW_ITEM_TYPE_NVGRE:
1179 rte_flow_error_set(error, EINVAL,
1180 RTE_FLOW_ERROR_TYPE_ITEM,
1182 "More than one tunneling protocol");
1190 rte_flow_error_set(error, EINVAL,
1191 RTE_FLOW_ERROR_TYPE_ITEM,
1193 "There is an unsupported pattern item "
1194 "in the inner frame");
1200 rc = item->parse(pattern, &flow->spec.template, error);
1204 if (item->layer != SFC_FLOW_ITEM_ANY_LAYER)
1205 prev_layer = item->layer;
1212 sfc_flow_parse_queue(struct sfc_adapter *sa,
1213 const struct rte_flow_action_queue *queue,
1214 struct rte_flow *flow)
1216 struct sfc_rxq *rxq;
1218 if (queue->index >= sa->rxq_count)
1221 rxq = sa->rxq_info[queue->index].rxq;
1222 flow->spec.template.efs_dmaq_id = (uint16_t)rxq->hw_index;
1227 #if EFSYS_OPT_RX_SCALE
1229 sfc_flow_parse_rss(struct sfc_adapter *sa,
1230 const struct rte_flow_action_rss *rss,
1231 struct rte_flow *flow)
1233 unsigned int rxq_sw_index;
1234 struct sfc_rxq *rxq;
1235 unsigned int rxq_hw_index_min;
1236 unsigned int rxq_hw_index_max;
1237 const uint8_t *rss_key;
1238 struct sfc_flow_rss *sfc_rss_conf = &flow->rss_conf;
1241 if (rss->queue_num == 0)
1244 rxq_sw_index = sa->rxq_count - 1;
1245 rxq = sa->rxq_info[rxq_sw_index].rxq;
1246 rxq_hw_index_min = rxq->hw_index;
1247 rxq_hw_index_max = 0;
1249 for (i = 0; i < rss->queue_num; ++i) {
1250 rxq_sw_index = rss->queue[i];
1252 if (rxq_sw_index >= sa->rxq_count)
1255 rxq = sa->rxq_info[rxq_sw_index].rxq;
1257 if (rxq->hw_index < rxq_hw_index_min)
1258 rxq_hw_index_min = rxq->hw_index;
1260 if (rxq->hw_index > rxq_hw_index_max)
1261 rxq_hw_index_max = rxq->hw_index;
1264 switch (rss->func) {
1265 case RTE_ETH_HASH_FUNCTION_DEFAULT:
1266 case RTE_ETH_HASH_FUNCTION_TOEPLITZ:
1272 if ((rss->types & ~SFC_RSS_OFFLOADS) != 0)
1276 if (rss->key_len != sizeof(sa->rss_key))
1281 rss_key = sa->rss_key;
1286 sfc_rss_conf->rxq_hw_index_min = rxq_hw_index_min;
1287 sfc_rss_conf->rxq_hw_index_max = rxq_hw_index_max;
1288 sfc_rss_conf->rss_hash_types = sfc_rte_to_efx_hash_type(rss->types);
1289 rte_memcpy(sfc_rss_conf->rss_key, rss_key, sizeof(sa->rss_key));
1291 for (i = 0; i < RTE_DIM(sfc_rss_conf->rss_tbl); ++i) {
1292 unsigned int rxq_sw_index = rss->queue[i % rss->queue_num];
1293 struct sfc_rxq *rxq = sa->rxq_info[rxq_sw_index].rxq;
1295 sfc_rss_conf->rss_tbl[i] = rxq->hw_index - rxq_hw_index_min;
1300 #endif /* EFSYS_OPT_RX_SCALE */
1303 sfc_flow_spec_flush(struct sfc_adapter *sa, struct sfc_flow_spec *spec,
1304 unsigned int filters_count)
1309 for (i = 0; i < filters_count; i++) {
1312 rc = efx_filter_remove(sa->nic, &spec->filters[i]);
1313 if (ret == 0 && rc != 0) {
1314 sfc_err(sa, "failed to remove filter specification "
1324 sfc_flow_spec_insert(struct sfc_adapter *sa, struct sfc_flow_spec *spec)
1329 for (i = 0; i < spec->count; i++) {
1330 rc = efx_filter_insert(sa->nic, &spec->filters[i]);
1332 sfc_flow_spec_flush(sa, spec, i);
1341 sfc_flow_spec_remove(struct sfc_adapter *sa, struct sfc_flow_spec *spec)
1343 return sfc_flow_spec_flush(sa, spec, spec->count);
1347 sfc_flow_filter_insert(struct sfc_adapter *sa,
1348 struct rte_flow *flow)
1350 #if EFSYS_OPT_RX_SCALE
1351 struct sfc_flow_rss *rss = &flow->rss_conf;
1352 uint32_t efs_rss_context = EFX_RSS_CONTEXT_DEFAULT;
1357 unsigned int rss_spread = MIN(rss->rxq_hw_index_max -
1358 rss->rxq_hw_index_min + 1,
1361 rc = efx_rx_scale_context_alloc(sa->nic,
1362 EFX_RX_SCALE_EXCLUSIVE,
1366 goto fail_scale_context_alloc;
1368 rc = efx_rx_scale_mode_set(sa->nic, efs_rss_context,
1369 EFX_RX_HASHALG_TOEPLITZ,
1370 rss->rss_hash_types, B_TRUE);
1372 goto fail_scale_mode_set;
1374 rc = efx_rx_scale_key_set(sa->nic, efs_rss_context,
1376 sizeof(sa->rss_key));
1378 goto fail_scale_key_set;
1381 * At this point, fully elaborated filter specifications
1382 * have been produced from the template. To make sure that
1383 * RSS behaviour is consistent between them, set the same
1384 * RSS context value everywhere.
1386 for (i = 0; i < flow->spec.count; i++) {
1387 efx_filter_spec_t *spec = &flow->spec.filters[i];
1389 spec->efs_rss_context = efs_rss_context;
1390 spec->efs_dmaq_id = rss->rxq_hw_index_min;
1391 spec->efs_flags |= EFX_FILTER_FLAG_RX_RSS;
1395 rc = sfc_flow_spec_insert(sa, &flow->spec);
1397 goto fail_filter_insert;
1401 * Scale table is set after filter insertion because
1402 * the table entries are relative to the base RxQ ID
1403 * and the latter is submitted to the HW by means of
1404 * inserting a filter, so by the time of the request
1405 * the HW knows all the information needed to verify
1406 * the table entries, and the operation will succeed
1408 rc = efx_rx_scale_tbl_set(sa->nic, efs_rss_context,
1409 rss->rss_tbl, RTE_DIM(rss->rss_tbl));
1411 goto fail_scale_tbl_set;
1417 sfc_flow_spec_remove(sa, &flow->spec);
1421 fail_scale_mode_set:
1422 if (efs_rss_context != EFX_RSS_CONTEXT_DEFAULT)
1423 efx_rx_scale_context_free(sa->nic, efs_rss_context);
1425 fail_scale_context_alloc:
1427 #else /* !EFSYS_OPT_RX_SCALE */
1428 return sfc_flow_spec_insert(sa, &flow->spec);
1429 #endif /* EFSYS_OPT_RX_SCALE */
1433 sfc_flow_filter_remove(struct sfc_adapter *sa,
1434 struct rte_flow *flow)
1438 rc = sfc_flow_spec_remove(sa, &flow->spec);
1442 #if EFSYS_OPT_RX_SCALE
1445 * All specifications for a given flow rule have the same RSS
1446 * context, so that RSS context value is taken from the first
1447 * filter specification
1449 efx_filter_spec_t *spec = &flow->spec.filters[0];
1451 rc = efx_rx_scale_context_free(sa->nic, spec->efs_rss_context);
1453 #endif /* EFSYS_OPT_RX_SCALE */
1459 sfc_flow_parse_actions(struct sfc_adapter *sa,
1460 const struct rte_flow_action actions[],
1461 struct rte_flow *flow,
1462 struct rte_flow_error *error)
1465 boolean_t is_specified = B_FALSE;
1467 if (actions == NULL) {
1468 rte_flow_error_set(error, EINVAL,
1469 RTE_FLOW_ERROR_TYPE_ACTION_NUM, NULL,
1474 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1475 /* This one may appear anywhere multiple times. */
1476 if (actions->type == RTE_FLOW_ACTION_TYPE_VOID)
1478 /* Fate-deciding actions may appear exactly once. */
1481 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
1483 "Cannot combine several fate-deciding actions,"
1484 "choose between QUEUE, RSS or DROP");
1487 switch (actions->type) {
1488 case RTE_FLOW_ACTION_TYPE_QUEUE:
1489 rc = sfc_flow_parse_queue(sa, actions->conf, flow);
1491 rte_flow_error_set(error, EINVAL,
1492 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1493 "Bad QUEUE action");
1497 is_specified = B_TRUE;
1500 #if EFSYS_OPT_RX_SCALE
1501 case RTE_FLOW_ACTION_TYPE_RSS:
1502 rc = sfc_flow_parse_rss(sa, actions->conf, flow);
1504 rte_flow_error_set(error, rc,
1505 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1510 is_specified = B_TRUE;
1512 #endif /* EFSYS_OPT_RX_SCALE */
1514 case RTE_FLOW_ACTION_TYPE_DROP:
1515 flow->spec.template.efs_dmaq_id =
1516 EFX_FILTER_SPEC_RX_DMAQ_ID_DROP;
1518 is_specified = B_TRUE;
1522 rte_flow_error_set(error, ENOTSUP,
1523 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1524 "Action is not supported");
1529 /* When fate is unknown, drop traffic. */
1530 if (!is_specified) {
1531 flow->spec.template.efs_dmaq_id =
1532 EFX_FILTER_SPEC_RX_DMAQ_ID_DROP;
1539 * Set the EFX_FILTER_MATCH_UNKNOWN_UCAST_DST
1540 * and EFX_FILTER_MATCH_UNKNOWN_MCAST_DST match flags in the same
1541 * specifications after copying.
1543 * @param spec[in, out]
1544 * SFC flow specification to update.
1545 * @param filters_count_for_one_val[in]
1546 * How many specifications should have the same match flag, what is the
1547 * number of specifications before copying.
1549 * Perform verbose error reporting if not NULL.
1552 sfc_flow_set_unknown_dst_flags(struct sfc_flow_spec *spec,
1553 unsigned int filters_count_for_one_val,
1554 struct rte_flow_error *error)
1557 static const efx_filter_match_flags_t vals[] = {
1558 EFX_FILTER_MATCH_UNKNOWN_UCAST_DST,
1559 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST
1562 if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1563 rte_flow_error_set(error, EINVAL,
1564 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1565 "Number of specifications is incorrect while copying "
1566 "by unknown destination flags");
1570 for (i = 0; i < spec->count; i++) {
1571 /* The check above ensures that divisor can't be zero here */
1572 spec->filters[i].efs_match_flags |=
1573 vals[i / filters_count_for_one_val];
1580 * Check that the following conditions are met:
1581 * - the list of supported filters has a filter
1582 * with EFX_FILTER_MATCH_UNKNOWN_MCAST_DST flag instead of
1583 * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST, since this filter will also
1587 * The match flags of filter.
1589 * Specification to be supplemented.
1591 * SFC filter with list of supported filters.
1594 sfc_flow_check_unknown_dst_flags(efx_filter_match_flags_t match,
1595 __rte_unused efx_filter_spec_t *spec,
1596 struct sfc_filter *filter)
1599 efx_filter_match_flags_t match_mcast_dst;
1602 (match & ~EFX_FILTER_MATCH_UNKNOWN_UCAST_DST) |
1603 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST;
1604 for (i = 0; i < filter->supported_match_num; i++) {
1605 if (match_mcast_dst == filter->supported_match[i])
1613 * Set the EFX_FILTER_MATCH_ETHER_TYPE match flag and EFX_ETHER_TYPE_IPV4 and
1614 * EFX_ETHER_TYPE_IPV6 values of the corresponding field in the same
1615 * specifications after copying.
1617 * @param spec[in, out]
1618 * SFC flow specification to update.
1619 * @param filters_count_for_one_val[in]
1620 * How many specifications should have the same EtherType value, what is the
1621 * number of specifications before copying.
1623 * Perform verbose error reporting if not NULL.
1626 sfc_flow_set_ethertypes(struct sfc_flow_spec *spec,
1627 unsigned int filters_count_for_one_val,
1628 struct rte_flow_error *error)
1631 static const uint16_t vals[] = {
1632 EFX_ETHER_TYPE_IPV4, EFX_ETHER_TYPE_IPV6
1635 if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1636 rte_flow_error_set(error, EINVAL,
1637 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1638 "Number of specifications is incorrect "
1639 "while copying by Ethertype");
1643 for (i = 0; i < spec->count; i++) {
1644 spec->filters[i].efs_match_flags |=
1645 EFX_FILTER_MATCH_ETHER_TYPE;
1648 * The check above ensures that
1649 * filters_count_for_one_val is not 0
1651 spec->filters[i].efs_ether_type =
1652 vals[i / filters_count_for_one_val];
1659 * Set the EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST and
1660 * EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST match flags in the same
1661 * specifications after copying.
1663 * @param spec[in, out]
1664 * SFC flow specification to update.
1665 * @param filters_count_for_one_val[in]
1666 * How many specifications should have the same match flag, what is the
1667 * number of specifications before copying.
1669 * Perform verbose error reporting if not NULL.
1672 sfc_flow_set_ifrm_unknown_dst_flags(struct sfc_flow_spec *spec,
1673 unsigned int filters_count_for_one_val,
1674 struct rte_flow_error *error)
1677 static const efx_filter_match_flags_t vals[] = {
1678 EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST,
1679 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST
1682 if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1683 rte_flow_error_set(error, EINVAL,
1684 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1685 "Number of specifications is incorrect while copying "
1686 "by inner frame unknown destination flags");
1690 for (i = 0; i < spec->count; i++) {
1691 /* The check above ensures that divisor can't be zero here */
1692 spec->filters[i].efs_match_flags |=
1693 vals[i / filters_count_for_one_val];
1700 * Check that the following conditions are met:
1701 * - the specification corresponds to a filter for encapsulated traffic
1702 * - the list of supported filters has a filter
1703 * with EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST flag instead of
1704 * EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST, since this filter will also
1708 * The match flags of filter.
1710 * Specification to be supplemented.
1712 * SFC filter with list of supported filters.
1715 sfc_flow_check_ifrm_unknown_dst_flags(efx_filter_match_flags_t match,
1716 efx_filter_spec_t *spec,
1717 struct sfc_filter *filter)
1720 efx_tunnel_protocol_t encap_type = spec->efs_encap_type;
1721 efx_filter_match_flags_t match_mcast_dst;
1723 if (encap_type == EFX_TUNNEL_PROTOCOL_NONE)
1727 (match & ~EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST) |
1728 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST;
1729 for (i = 0; i < filter->supported_match_num; i++) {
1730 if (match_mcast_dst == filter->supported_match[i])
1738 * Match flags that can be automatically added to filters.
1739 * Selecting the last minimum when searching for the copy flag ensures that the
1740 * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST flag has a higher priority than
1741 * EFX_FILTER_MATCH_ETHER_TYPE. This is because the filter
1742 * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST is at the end of the list of supported
1745 static const struct sfc_flow_copy_flag sfc_flow_copy_flags[] = {
1747 .flag = EFX_FILTER_MATCH_UNKNOWN_UCAST_DST,
1749 .set_vals = sfc_flow_set_unknown_dst_flags,
1750 .spec_check = sfc_flow_check_unknown_dst_flags,
1753 .flag = EFX_FILTER_MATCH_ETHER_TYPE,
1755 .set_vals = sfc_flow_set_ethertypes,
1759 .flag = EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST,
1761 .set_vals = sfc_flow_set_ifrm_unknown_dst_flags,
1762 .spec_check = sfc_flow_check_ifrm_unknown_dst_flags,
1766 /* Get item from array sfc_flow_copy_flags */
1767 static const struct sfc_flow_copy_flag *
1768 sfc_flow_get_copy_flag(efx_filter_match_flags_t flag)
1772 for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1773 if (sfc_flow_copy_flags[i].flag == flag)
1774 return &sfc_flow_copy_flags[i];
1781 * Make copies of the specifications, set match flag and values
1782 * of the field that corresponds to it.
1784 * @param spec[in, out]
1785 * SFC flow specification to update.
1787 * The match flag to add.
1789 * Perform verbose error reporting if not NULL.
1792 sfc_flow_spec_add_match_flag(struct sfc_flow_spec *spec,
1793 efx_filter_match_flags_t flag,
1794 struct rte_flow_error *error)
1797 unsigned int new_filters_count;
1798 unsigned int filters_count_for_one_val;
1799 const struct sfc_flow_copy_flag *copy_flag;
1802 copy_flag = sfc_flow_get_copy_flag(flag);
1803 if (copy_flag == NULL) {
1804 rte_flow_error_set(error, ENOTSUP,
1805 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1806 "Unsupported spec field for copying");
1810 new_filters_count = spec->count * copy_flag->vals_count;
1811 if (new_filters_count > SF_FLOW_SPEC_NB_FILTERS_MAX) {
1812 rte_flow_error_set(error, EINVAL,
1813 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1814 "Too much EFX specifications in the flow rule");
1818 /* Copy filters specifications */
1819 for (i = spec->count; i < new_filters_count; i++)
1820 spec->filters[i] = spec->filters[i - spec->count];
1822 filters_count_for_one_val = spec->count;
1823 spec->count = new_filters_count;
1825 rc = copy_flag->set_vals(spec, filters_count_for_one_val, error);
1833 * Check that the given set of match flags missing in the original filter spec
1834 * could be covered by adding spec copies which specify the corresponding
1835 * flags and packet field values to match.
1837 * @param miss_flags[in]
1838 * Flags that are missing until the supported filter.
1840 * Specification to be supplemented.
1845 * Number of specifications after copy or 0, if the flags can not be added.
1848 sfc_flow_check_missing_flags(efx_filter_match_flags_t miss_flags,
1849 efx_filter_spec_t *spec,
1850 struct sfc_filter *filter)
1853 efx_filter_match_flags_t copy_flags = 0;
1854 efx_filter_match_flags_t flag;
1855 efx_filter_match_flags_t match = spec->efs_match_flags | miss_flags;
1856 sfc_flow_spec_check *check;
1857 unsigned int multiplier = 1;
1859 for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1860 flag = sfc_flow_copy_flags[i].flag;
1861 check = sfc_flow_copy_flags[i].spec_check;
1862 if ((flag & miss_flags) == flag) {
1863 if (check != NULL && (!check(match, spec, filter)))
1867 multiplier *= sfc_flow_copy_flags[i].vals_count;
1871 if (copy_flags == miss_flags)
1878 * Attempt to supplement the specification template to the minimally
1879 * supported set of match flags. To do this, it is necessary to copy
1880 * the specifications, filling them with the values of fields that
1881 * correspond to the missing flags.
1882 * The necessary and sufficient filter is built from the fewest number
1883 * of copies which could be made to cover the minimally required set
1888 * @param spec[in, out]
1889 * SFC flow specification to update.
1891 * Perform verbose error reporting if not NULL.
1894 sfc_flow_spec_filters_complete(struct sfc_adapter *sa,
1895 struct sfc_flow_spec *spec,
1896 struct rte_flow_error *error)
1898 struct sfc_filter *filter = &sa->filter;
1899 efx_filter_match_flags_t miss_flags;
1900 efx_filter_match_flags_t min_miss_flags = 0;
1901 efx_filter_match_flags_t match;
1902 unsigned int min_multiplier = UINT_MAX;
1903 unsigned int multiplier;
1907 match = spec->template.efs_match_flags;
1908 for (i = 0; i < filter->supported_match_num; i++) {
1909 if ((match & filter->supported_match[i]) == match) {
1910 miss_flags = filter->supported_match[i] & (~match);
1911 multiplier = sfc_flow_check_missing_flags(miss_flags,
1912 &spec->template, filter);
1913 if (multiplier > 0) {
1914 if (multiplier <= min_multiplier) {
1915 min_multiplier = multiplier;
1916 min_miss_flags = miss_flags;
1922 if (min_multiplier == UINT_MAX) {
1923 rte_flow_error_set(error, ENOTSUP,
1924 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1925 "Flow rule pattern is not supported");
1929 for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1930 efx_filter_match_flags_t flag = sfc_flow_copy_flags[i].flag;
1932 if ((flag & min_miss_flags) == flag) {
1933 rc = sfc_flow_spec_add_match_flag(spec, flag, error);
1943 * Check that set of match flags is referred to by a filter. Filter is
1944 * described by match flags with the ability to add OUTER_VID and INNER_VID
1947 * @param match_flags[in]
1948 * Set of match flags.
1949 * @param flags_pattern[in]
1950 * Pattern of filter match flags.
1953 sfc_flow_is_match_with_vids(efx_filter_match_flags_t match_flags,
1954 efx_filter_match_flags_t flags_pattern)
1956 if ((match_flags & flags_pattern) != flags_pattern)
1959 switch (match_flags & ~flags_pattern) {
1961 case EFX_FILTER_MATCH_OUTER_VID:
1962 case EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_INNER_VID:
1970 * Check whether the spec maps to a hardware filter which is known to be
1971 * ineffective despite being valid.
1974 * SFC flow specification.
1977 sfc_flow_is_match_flags_exception(struct sfc_flow_spec *spec)
1980 uint16_t ether_type;
1982 efx_filter_match_flags_t match_flags;
1984 for (i = 0; i < spec->count; i++) {
1985 match_flags = spec->filters[i].efs_match_flags;
1987 if (sfc_flow_is_match_with_vids(match_flags,
1988 EFX_FILTER_MATCH_ETHER_TYPE) ||
1989 sfc_flow_is_match_with_vids(match_flags,
1990 EFX_FILTER_MATCH_ETHER_TYPE |
1991 EFX_FILTER_MATCH_LOC_MAC)) {
1992 ether_type = spec->filters[i].efs_ether_type;
1993 if (ether_type == EFX_ETHER_TYPE_IPV4 ||
1994 ether_type == EFX_ETHER_TYPE_IPV6)
1996 } else if (sfc_flow_is_match_with_vids(match_flags,
1997 EFX_FILTER_MATCH_ETHER_TYPE |
1998 EFX_FILTER_MATCH_IP_PROTO) ||
1999 sfc_flow_is_match_with_vids(match_flags,
2000 EFX_FILTER_MATCH_ETHER_TYPE |
2001 EFX_FILTER_MATCH_IP_PROTO |
2002 EFX_FILTER_MATCH_LOC_MAC)) {
2003 ip_proto = spec->filters[i].efs_ip_proto;
2004 if (ip_proto == EFX_IPPROTO_TCP ||
2005 ip_proto == EFX_IPPROTO_UDP)
2014 sfc_flow_validate_match_flags(struct sfc_adapter *sa,
2015 struct rte_flow *flow,
2016 struct rte_flow_error *error)
2018 efx_filter_spec_t *spec_tmpl = &flow->spec.template;
2019 efx_filter_match_flags_t match_flags = spec_tmpl->efs_match_flags;
2022 /* Initialize the first filter spec with template */
2023 flow->spec.filters[0] = *spec_tmpl;
2024 flow->spec.count = 1;
2026 if (!sfc_filter_is_match_supported(sa, match_flags)) {
2027 rc = sfc_flow_spec_filters_complete(sa, &flow->spec, error);
2032 if (sfc_flow_is_match_flags_exception(&flow->spec)) {
2033 rte_flow_error_set(error, ENOTSUP,
2034 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2035 "The flow rule pattern is unsupported");
2043 sfc_flow_parse(struct rte_eth_dev *dev,
2044 const struct rte_flow_attr *attr,
2045 const struct rte_flow_item pattern[],
2046 const struct rte_flow_action actions[],
2047 struct rte_flow *flow,
2048 struct rte_flow_error *error)
2050 struct sfc_adapter *sa = dev->data->dev_private;
2053 rc = sfc_flow_parse_attr(attr, flow, error);
2055 goto fail_bad_value;
2057 rc = sfc_flow_parse_pattern(pattern, flow, error);
2059 goto fail_bad_value;
2061 rc = sfc_flow_parse_actions(sa, actions, flow, error);
2063 goto fail_bad_value;
2065 rc = sfc_flow_validate_match_flags(sa, flow, error);
2067 goto fail_bad_value;
2076 sfc_flow_validate(struct rte_eth_dev *dev,
2077 const struct rte_flow_attr *attr,
2078 const struct rte_flow_item pattern[],
2079 const struct rte_flow_action actions[],
2080 struct rte_flow_error *error)
2082 struct rte_flow flow;
2084 memset(&flow, 0, sizeof(flow));
2086 return sfc_flow_parse(dev, attr, pattern, actions, &flow, error);
2089 static struct rte_flow *
2090 sfc_flow_create(struct rte_eth_dev *dev,
2091 const struct rte_flow_attr *attr,
2092 const struct rte_flow_item pattern[],
2093 const struct rte_flow_action actions[],
2094 struct rte_flow_error *error)
2096 struct sfc_adapter *sa = dev->data->dev_private;
2097 struct rte_flow *flow = NULL;
2100 flow = rte_zmalloc("sfc_rte_flow", sizeof(*flow), 0);
2102 rte_flow_error_set(error, ENOMEM,
2103 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2104 "Failed to allocate memory");
2108 rc = sfc_flow_parse(dev, attr, pattern, actions, flow, error);
2110 goto fail_bad_value;
2112 TAILQ_INSERT_TAIL(&sa->filter.flow_list, flow, entries);
2114 sfc_adapter_lock(sa);
2116 if (sa->state == SFC_ADAPTER_STARTED) {
2117 rc = sfc_flow_filter_insert(sa, flow);
2119 rte_flow_error_set(error, rc,
2120 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2121 "Failed to insert filter");
2122 goto fail_filter_insert;
2126 sfc_adapter_unlock(sa);
2131 TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2135 sfc_adapter_unlock(sa);
2142 sfc_flow_remove(struct sfc_adapter *sa,
2143 struct rte_flow *flow,
2144 struct rte_flow_error *error)
2148 SFC_ASSERT(sfc_adapter_is_locked(sa));
2150 if (sa->state == SFC_ADAPTER_STARTED) {
2151 rc = sfc_flow_filter_remove(sa, flow);
2153 rte_flow_error_set(error, rc,
2154 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2155 "Failed to destroy flow rule");
2158 TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2165 sfc_flow_destroy(struct rte_eth_dev *dev,
2166 struct rte_flow *flow,
2167 struct rte_flow_error *error)
2169 struct sfc_adapter *sa = dev->data->dev_private;
2170 struct rte_flow *flow_ptr;
2173 sfc_adapter_lock(sa);
2175 TAILQ_FOREACH(flow_ptr, &sa->filter.flow_list, entries) {
2176 if (flow_ptr == flow)
2180 rte_flow_error_set(error, rc,
2181 RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2182 "Failed to find flow rule to destroy");
2183 goto fail_bad_value;
2186 rc = sfc_flow_remove(sa, flow, error);
2189 sfc_adapter_unlock(sa);
2195 sfc_flow_flush(struct rte_eth_dev *dev,
2196 struct rte_flow_error *error)
2198 struct sfc_adapter *sa = dev->data->dev_private;
2199 struct rte_flow *flow;
2203 sfc_adapter_lock(sa);
2205 while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
2206 rc = sfc_flow_remove(sa, flow, error);
2211 sfc_adapter_unlock(sa);
2217 sfc_flow_isolate(struct rte_eth_dev *dev, int enable,
2218 struct rte_flow_error *error)
2220 struct sfc_adapter *sa = dev->data->dev_private;
2221 struct sfc_port *port = &sa->port;
2224 sfc_adapter_lock(sa);
2225 if (sa->state != SFC_ADAPTER_INITIALIZED) {
2226 rte_flow_error_set(error, EBUSY,
2227 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2228 NULL, "please close the port first");
2231 port->isolated = (enable) ? B_TRUE : B_FALSE;
2233 sfc_adapter_unlock(sa);
2238 const struct rte_flow_ops sfc_flow_ops = {
2239 .validate = sfc_flow_validate,
2240 .create = sfc_flow_create,
2241 .destroy = sfc_flow_destroy,
2242 .flush = sfc_flow_flush,
2244 .isolate = sfc_flow_isolate,
2248 sfc_flow_init(struct sfc_adapter *sa)
2250 SFC_ASSERT(sfc_adapter_is_locked(sa));
2252 TAILQ_INIT(&sa->filter.flow_list);
2256 sfc_flow_fini(struct sfc_adapter *sa)
2258 struct rte_flow *flow;
2260 SFC_ASSERT(sfc_adapter_is_locked(sa));
2262 while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
2263 TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2269 sfc_flow_stop(struct sfc_adapter *sa)
2271 struct rte_flow *flow;
2273 SFC_ASSERT(sfc_adapter_is_locked(sa));
2275 TAILQ_FOREACH(flow, &sa->filter.flow_list, entries)
2276 sfc_flow_filter_remove(sa, flow);
2280 sfc_flow_start(struct sfc_adapter *sa)
2282 struct rte_flow *flow;
2285 sfc_log_init(sa, "entry");
2287 SFC_ASSERT(sfc_adapter_is_locked(sa));
2289 TAILQ_FOREACH(flow, &sa->filter.flow_list, entries) {
2290 rc = sfc_flow_filter_insert(sa, flow);
2295 sfc_log_init(sa, "done");