1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018 6WIND S.A.
3 * Copyright 2018 Mellanox Technologies, Ltd
8 #include <libmnl/libmnl.h>
9 #include <linux/gen_stats.h>
10 #include <linux/if_ether.h>
11 #include <linux/netlink.h>
12 #include <linux/pkt_cls.h>
13 #include <linux/pkt_sched.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/tc_act/tc_gact.h>
16 #include <linux/tc_act/tc_mirred.h>
17 #include <netinet/in.h>
23 #include <sys/socket.h>
25 #include <rte_byteorder.h>
26 #include <rte_errno.h>
27 #include <rte_ether.h>
29 #include <rte_malloc.h>
30 #include <rte_common.h>
33 #include "mlx5_flow.h"
34 #include "mlx5_autoconf.h"
36 #ifdef HAVE_TC_ACT_VLAN
38 #include <linux/tc_act/tc_vlan.h>
40 #else /* HAVE_TC_ACT_VLAN */
42 #define TCA_VLAN_ACT_POP 1
43 #define TCA_VLAN_ACT_PUSH 2
44 #define TCA_VLAN_ACT_MODIFY 3
45 #define TCA_VLAN_PARMS 2
46 #define TCA_VLAN_PUSH_VLAN_ID 3
47 #define TCA_VLAN_PUSH_VLAN_PROTOCOL 4
48 #define TCA_VLAN_PAD 5
49 #define TCA_VLAN_PUSH_VLAN_PRIORITY 6
56 #endif /* HAVE_TC_ACT_VLAN */
58 #ifdef HAVE_TC_ACT_PEDIT
60 #include <linux/tc_act/tc_pedit.h>
62 #else /* HAVE_TC_ACT_VLAN */
76 TCA_PEDIT_KEY_EX_HTYPE = 1,
77 TCA_PEDIT_KEY_EX_CMD = 2,
78 __TCA_PEDIT_KEY_EX_MAX
81 enum pedit_header_type {
82 TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK = 0,
83 TCA_PEDIT_KEY_EX_HDR_TYPE_ETH = 1,
84 TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 = 2,
85 TCA_PEDIT_KEY_EX_HDR_TYPE_IP6 = 3,
86 TCA_PEDIT_KEY_EX_HDR_TYPE_TCP = 4,
87 TCA_PEDIT_KEY_EX_HDR_TYPE_UDP = 5,
92 TCA_PEDIT_KEY_EX_CMD_SET = 0,
93 TCA_PEDIT_KEY_EX_CMD_ADD = 1,
100 __u32 off; /*offset */
107 struct tc_pedit_sel {
111 struct tc_pedit_key keys[0];
114 #endif /* HAVE_TC_ACT_VLAN */
116 #ifdef HAVE_TC_ACT_TUNNEL_KEY
118 #include <linux/tc_act/tc_tunnel_key.h>
120 #ifndef HAVE_TCA_TUNNEL_KEY_ENC_DST_PORT
121 #define TCA_TUNNEL_KEY_ENC_DST_PORT 9
124 #ifndef HAVE_TCA_TUNNEL_KEY_NO_CSUM
125 #define TCA_TUNNEL_KEY_NO_CSUM 10
128 #else /* HAVE_TC_ACT_TUNNEL_KEY */
130 #define TCA_ACT_TUNNEL_KEY 17
131 #define TCA_TUNNEL_KEY_ACT_SET 1
132 #define TCA_TUNNEL_KEY_ACT_RELEASE 2
133 #define TCA_TUNNEL_KEY_PARMS 2
134 #define TCA_TUNNEL_KEY_ENC_IPV4_SRC 3
135 #define TCA_TUNNEL_KEY_ENC_IPV4_DST 4
136 #define TCA_TUNNEL_KEY_ENC_IPV6_SRC 5
137 #define TCA_TUNNEL_KEY_ENC_IPV6_DST 6
138 #define TCA_TUNNEL_KEY_ENC_KEY_ID 7
139 #define TCA_TUNNEL_KEY_ENC_DST_PORT 9
140 #define TCA_TUNNEL_KEY_NO_CSUM 10
142 struct tc_tunnel_key {
147 #endif /* HAVE_TC_ACT_TUNNEL_KEY */
149 /* Normally found in linux/netlink.h. */
150 #ifndef NETLINK_CAP_ACK
151 #define NETLINK_CAP_ACK 10
154 /* Normally found in linux/pkt_sched.h. */
155 #ifndef TC_H_MIN_INGRESS
156 #define TC_H_MIN_INGRESS 0xfff2u
159 /* Normally found in linux/pkt_cls.h. */
160 #ifndef TCA_CLS_FLAGS_SKIP_SW
161 #define TCA_CLS_FLAGS_SKIP_SW (1 << 1)
163 #ifndef TCA_CLS_FLAGS_IN_HW
164 #define TCA_CLS_FLAGS_IN_HW (1 << 2)
166 #ifndef HAVE_TCA_CHAIN
169 #ifndef HAVE_TCA_FLOWER_ACT
170 #define TCA_FLOWER_ACT 3
172 #ifndef HAVE_TCA_FLOWER_FLAGS
173 #define TCA_FLOWER_FLAGS 22
175 #ifndef HAVE_TCA_FLOWER_KEY_ETH_TYPE
176 #define TCA_FLOWER_KEY_ETH_TYPE 8
178 #ifndef HAVE_TCA_FLOWER_KEY_ETH_DST
179 #define TCA_FLOWER_KEY_ETH_DST 4
181 #ifndef HAVE_TCA_FLOWER_KEY_ETH_DST_MASK
182 #define TCA_FLOWER_KEY_ETH_DST_MASK 5
184 #ifndef HAVE_TCA_FLOWER_KEY_ETH_SRC
185 #define TCA_FLOWER_KEY_ETH_SRC 6
187 #ifndef HAVE_TCA_FLOWER_KEY_ETH_SRC_MASK
188 #define TCA_FLOWER_KEY_ETH_SRC_MASK 7
190 #ifndef HAVE_TCA_FLOWER_KEY_IP_PROTO
191 #define TCA_FLOWER_KEY_IP_PROTO 9
193 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_SRC
194 #define TCA_FLOWER_KEY_IPV4_SRC 10
196 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_SRC_MASK
197 #define TCA_FLOWER_KEY_IPV4_SRC_MASK 11
199 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_DST
200 #define TCA_FLOWER_KEY_IPV4_DST 12
202 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_DST_MASK
203 #define TCA_FLOWER_KEY_IPV4_DST_MASK 13
205 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_SRC
206 #define TCA_FLOWER_KEY_IPV6_SRC 14
208 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_SRC_MASK
209 #define TCA_FLOWER_KEY_IPV6_SRC_MASK 15
211 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_DST
212 #define TCA_FLOWER_KEY_IPV6_DST 16
214 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_DST_MASK
215 #define TCA_FLOWER_KEY_IPV6_DST_MASK 17
217 #ifndef HAVE_TCA_FLOWER_KEY_TCP_SRC
218 #define TCA_FLOWER_KEY_TCP_SRC 18
220 #ifndef HAVE_TCA_FLOWER_KEY_TCP_SRC_MASK
221 #define TCA_FLOWER_KEY_TCP_SRC_MASK 35
223 #ifndef HAVE_TCA_FLOWER_KEY_TCP_DST
224 #define TCA_FLOWER_KEY_TCP_DST 19
226 #ifndef HAVE_TCA_FLOWER_KEY_TCP_DST_MASK
227 #define TCA_FLOWER_KEY_TCP_DST_MASK 36
229 #ifndef HAVE_TCA_FLOWER_KEY_UDP_SRC
230 #define TCA_FLOWER_KEY_UDP_SRC 20
232 #ifndef HAVE_TCA_FLOWER_KEY_UDP_SRC_MASK
233 #define TCA_FLOWER_KEY_UDP_SRC_MASK 37
235 #ifndef HAVE_TCA_FLOWER_KEY_UDP_DST
236 #define TCA_FLOWER_KEY_UDP_DST 21
238 #ifndef HAVE_TCA_FLOWER_KEY_UDP_DST_MASK
239 #define TCA_FLOWER_KEY_UDP_DST_MASK 38
241 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_ID
242 #define TCA_FLOWER_KEY_VLAN_ID 23
244 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_PRIO
245 #define TCA_FLOWER_KEY_VLAN_PRIO 24
247 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_ETH_TYPE
248 #define TCA_FLOWER_KEY_VLAN_ETH_TYPE 25
250 #ifndef HAVE_TCA_FLOWER_KEY_ENC_KEY_ID
251 #define TCA_FLOWER_KEY_ENC_KEY_ID 26
253 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_SRC
254 #define TCA_FLOWER_KEY_ENC_IPV4_SRC 27
256 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK
257 #define TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK 28
259 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_DST
260 #define TCA_FLOWER_KEY_ENC_IPV4_DST 29
262 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_DST_MASK
263 #define TCA_FLOWER_KEY_ENC_IPV4_DST_MASK 30
265 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_SRC
266 #define TCA_FLOWER_KEY_ENC_IPV6_SRC 31
268 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK
269 #define TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK 32
271 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_DST
272 #define TCA_FLOWER_KEY_ENC_IPV6_DST 33
274 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_DST_MASK
275 #define TCA_FLOWER_KEY_ENC_IPV6_DST_MASK 34
277 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_SRC_PORT
278 #define TCA_FLOWER_KEY_ENC_UDP_SRC_PORT 43
280 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK
281 #define TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK 44
283 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_DST_PORT
284 #define TCA_FLOWER_KEY_ENC_UDP_DST_PORT 45
286 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK
287 #define TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK 46
289 #ifndef HAVE_TCA_FLOWER_KEY_TCP_FLAGS
290 #define TCA_FLOWER_KEY_TCP_FLAGS 71
292 #ifndef HAVE_TCA_FLOWER_KEY_TCP_FLAGS_MASK
293 #define TCA_FLOWER_KEY_TCP_FLAGS_MASK 72
295 #ifndef HAVE_TC_ACT_GOTO_CHAIN
296 #define TC_ACT_GOTO_CHAIN 0x20000000
299 #ifndef IPV6_ADDR_LEN
300 #define IPV6_ADDR_LEN 16
303 #ifndef IPV4_ADDR_LEN
304 #define IPV4_ADDR_LEN 4
308 #define TP_PORT_LEN 2 /* Transport Port (UDP/TCP) Length */
315 #ifndef TCA_ACT_MAX_PRIO
316 #define TCA_ACT_MAX_PRIO 32
319 /** UDP port range of VXLAN devices created by driver. */
320 #define MLX5_VXLAN_PORT_MIN 30000
321 #define MLX5_VXLAN_PORT_MAX 60000
322 #define MLX5_VXLAN_DEVICE_PFX "vmlx_"
324 /** Tunnel action type, used for @p type in header structure. */
325 enum flow_tcf_tunact_type {
326 FLOW_TCF_TUNACT_VXLAN_DECAP,
327 FLOW_TCF_TUNACT_VXLAN_ENCAP,
330 /** Flags used for @p mask in tunnel action encap descriptors. */
331 #define FLOW_TCF_ENCAP_ETH_SRC (1u << 0)
332 #define FLOW_TCF_ENCAP_ETH_DST (1u << 1)
333 #define FLOW_TCF_ENCAP_IPV4_SRC (1u << 2)
334 #define FLOW_TCF_ENCAP_IPV4_DST (1u << 3)
335 #define FLOW_TCF_ENCAP_IPV6_SRC (1u << 4)
336 #define FLOW_TCF_ENCAP_IPV6_DST (1u << 5)
337 #define FLOW_TCF_ENCAP_UDP_SRC (1u << 6)
338 #define FLOW_TCF_ENCAP_UDP_DST (1u << 7)
339 #define FLOW_TCF_ENCAP_VXLAN_VNI (1u << 8)
342 * Structure for holding netlink context.
343 * Note the size of the message buffer which is MNL_SOCKET_BUFFER_SIZE.
344 * Using this (8KB) buffer size ensures that netlink messages will never be
347 struct mlx5_flow_tcf_context {
348 struct mnl_socket *nl; /* NETLINK_ROUTE libmnl socket. */
349 uint32_t seq; /* Message sequence number. */
350 uint32_t buf_size; /* Message buffer size. */
351 uint8_t *buf; /* Message buffer. */
355 * Neigh rule structure. The neigh rule is applied via Netlink to
356 * outer tunnel iface in order to provide destination MAC address
357 * for the VXLAN encapsultion. The neigh rule is implicitly related
358 * to the Flow itself and can be shared by multiple Flows.
360 struct tcf_neigh_rule {
361 LIST_ENTRY(tcf_neigh_rule) next;
363 struct ether_addr eth;
370 uint8_t dst[IPV6_ADDR_LEN];
376 * Local rule structure. The local rule is applied via Netlink to
377 * outer tunnel iface in order to provide local and peer IP addresses
378 * of the VXLAN tunnel for encapsulation. The local rule is implicitly
379 * related to the Flow itself and can be shared by multiple Flows.
381 struct tcf_local_rule {
382 LIST_ENTRY(tcf_local_rule) next;
391 uint8_t dst[IPV6_ADDR_LEN];
392 uint8_t src[IPV6_ADDR_LEN];
397 /** VXLAN virtual netdev. */
399 LIST_ENTRY(tcf_vtep) next;
400 LIST_HEAD(, tcf_neigh_rule) neigh;
401 LIST_HEAD(, tcf_local_rule) local;
403 unsigned int ifindex; /**< Own interface index. */
404 unsigned int ifouter; /**< Index of device attached to. */
409 /** Tunnel descriptor header, common for all tunnel types. */
410 struct flow_tcf_tunnel_hdr {
411 uint32_t type; /**< Tunnel action type. */
412 struct tcf_vtep *vtep; /**< Virtual tunnel endpoint device. */
413 unsigned int ifindex_org; /**< Original dst/src interface */
414 unsigned int *ifindex_ptr; /**< Interface ptr in message. */
417 struct flow_tcf_vxlan_decap {
418 struct flow_tcf_tunnel_hdr hdr;
422 struct flow_tcf_vxlan_encap {
423 struct flow_tcf_tunnel_hdr hdr;
426 struct ether_addr dst;
427 struct ether_addr src;
435 uint8_t dst[IPV6_ADDR_LEN];
436 uint8_t src[IPV6_ADDR_LEN];
448 /** Structure used when extracting the values of a flow counters
449 * from a netlink message.
451 struct flow_tcf_stats_basic {
453 struct gnet_stats_basic counters;
456 /** Empty masks for known item types. */
458 struct rte_flow_item_port_id port_id;
459 struct rte_flow_item_eth eth;
460 struct rte_flow_item_vlan vlan;
461 struct rte_flow_item_ipv4 ipv4;
462 struct rte_flow_item_ipv6 ipv6;
463 struct rte_flow_item_tcp tcp;
464 struct rte_flow_item_udp udp;
465 struct rte_flow_item_vxlan vxlan;
466 } flow_tcf_mask_empty = {
470 /** Supported masks for known item types. */
471 static const struct {
472 struct rte_flow_item_port_id port_id;
473 struct rte_flow_item_eth eth;
474 struct rte_flow_item_vlan vlan;
475 struct rte_flow_item_ipv4 ipv4;
476 struct rte_flow_item_ipv6 ipv6;
477 struct rte_flow_item_tcp tcp;
478 struct rte_flow_item_udp udp;
479 struct rte_flow_item_vxlan vxlan;
480 } flow_tcf_mask_supported = {
485 .type = RTE_BE16(0xffff),
486 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
487 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
490 /* PCP and VID only, no DEI. */
491 .tci = RTE_BE16(0xefff),
492 .inner_type = RTE_BE16(0xffff),
495 .next_proto_id = 0xff,
496 .src_addr = RTE_BE32(0xffffffff),
497 .dst_addr = RTE_BE32(0xffffffff),
502 "\xff\xff\xff\xff\xff\xff\xff\xff"
503 "\xff\xff\xff\xff\xff\xff\xff\xff",
505 "\xff\xff\xff\xff\xff\xff\xff\xff"
506 "\xff\xff\xff\xff\xff\xff\xff\xff",
509 .src_port = RTE_BE16(0xffff),
510 .dst_port = RTE_BE16(0xffff),
514 .src_port = RTE_BE16(0xffff),
515 .dst_port = RTE_BE16(0xffff),
518 .vni = "\xff\xff\xff",
522 #define SZ_NLATTR_HDR MNL_ALIGN(sizeof(struct nlattr))
523 #define SZ_NLATTR_NEST SZ_NLATTR_HDR
524 #define SZ_NLATTR_DATA_OF(len) MNL_ALIGN(SZ_NLATTR_HDR + (len))
525 #define SZ_NLATTR_TYPE_OF(typ) SZ_NLATTR_DATA_OF(sizeof(typ))
526 #define SZ_NLATTR_STRZ_OF(str) SZ_NLATTR_DATA_OF(strlen(str) + 1)
528 #define PTOI_TABLE_SZ_MAX(dev) (mlx5_dev_to_port_id((dev)->device, NULL, 0) + 2)
530 /** DPDK port to network interface index (ifindex) conversion. */
531 struct flow_tcf_ptoi {
532 uint16_t port_id; /**< DPDK port ID. */
533 unsigned int ifindex; /**< Network interface index. */
536 /* Due to a limitation on driver/FW. */
537 #define MLX5_TCF_GROUP_ID_MAX 3
540 * Due to a limitation on driver/FW, priority ranges from 1 to 16 in kernel.
541 * Priority in rte_flow attribute starts from 0 and is added by 1 in
542 * translation. This is subject to be changed to determine the max priority
543 * based on trial-and-error like Verbs driver once the restriction is lifted or
544 * the range is extended.
546 #define MLX5_TCF_GROUP_PRIORITY_MAX 15
548 #define MLX5_TCF_FATE_ACTIONS \
549 (MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_PORT_ID | \
550 MLX5_FLOW_ACTION_JUMP)
552 #define MLX5_TCF_VLAN_ACTIONS \
553 (MLX5_FLOW_ACTION_OF_POP_VLAN | MLX5_FLOW_ACTION_OF_PUSH_VLAN | \
554 MLX5_FLOW_ACTION_OF_SET_VLAN_VID | MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
556 #define MLX5_TCF_VXLAN_ACTIONS \
557 (MLX5_FLOW_ACTION_VXLAN_ENCAP | MLX5_FLOW_ACTION_VXLAN_DECAP)
559 #define MLX5_TCF_PEDIT_ACTIONS \
560 (MLX5_FLOW_ACTION_SET_IPV4_SRC | MLX5_FLOW_ACTION_SET_IPV4_DST | \
561 MLX5_FLOW_ACTION_SET_IPV6_SRC | MLX5_FLOW_ACTION_SET_IPV6_DST | \
562 MLX5_FLOW_ACTION_SET_TP_SRC | MLX5_FLOW_ACTION_SET_TP_DST | \
563 MLX5_FLOW_ACTION_SET_TTL | MLX5_FLOW_ACTION_DEC_TTL | \
564 MLX5_FLOW_ACTION_SET_MAC_SRC | MLX5_FLOW_ACTION_SET_MAC_DST)
566 #define MLX5_TCF_CONFIG_ACTIONS \
567 (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_JUMP | \
568 MLX5_FLOW_ACTION_OF_PUSH_VLAN | MLX5_FLOW_ACTION_OF_SET_VLAN_VID | \
569 MLX5_FLOW_ACTION_OF_SET_VLAN_PCP | \
570 (MLX5_TCF_PEDIT_ACTIONS & ~MLX5_FLOW_ACTION_DEC_TTL))
572 #define MAX_PEDIT_KEYS 128
573 #define SZ_PEDIT_KEY_VAL 4
575 #define NUM_OF_PEDIT_KEYS(sz) \
576 (((sz) / SZ_PEDIT_KEY_VAL) + (((sz) % SZ_PEDIT_KEY_VAL) ? 1 : 0))
578 struct pedit_key_ex {
579 enum pedit_header_type htype;
583 struct pedit_parser {
584 struct tc_pedit_sel sel;
585 struct tc_pedit_key keys[MAX_PEDIT_KEYS];
586 struct pedit_key_ex keys_ex[MAX_PEDIT_KEYS];
590 * Create space for using the implicitly created TC flow counter.
593 * Pointer to the Ethernet device structure.
596 * A pointer to the counter data structure, NULL otherwise and
599 static struct mlx5_flow_counter *
600 flow_tcf_counter_new(void)
602 struct mlx5_flow_counter *cnt;
605 * eswitch counter cannot be shared and its id is unknown.
606 * currently returning all with id 0.
607 * in the future maybe better to switch to unique numbers.
609 struct mlx5_flow_counter tmpl = {
612 cnt = rte_calloc(__func__, 1, sizeof(*cnt), 0);
618 /* Implicit counter, do not add to list. */
623 * Set pedit key of MAC address
626 * pointer to action specification
627 * @param[in,out] p_parser
628 * pointer to pedit_parser
631 flow_tcf_pedit_key_set_mac(const struct rte_flow_action *actions,
632 struct pedit_parser *p_parser)
634 int idx = p_parser->sel.nkeys;
635 uint32_t off = actions->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
636 offsetof(struct ether_hdr, s_addr) :
637 offsetof(struct ether_hdr, d_addr);
638 const struct rte_flow_action_set_mac *conf =
639 (const struct rte_flow_action_set_mac *)actions->conf;
641 p_parser->keys[idx].off = off;
642 p_parser->keys[idx].mask = ~UINT32_MAX;
643 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_ETH;
644 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
645 memcpy(&p_parser->keys[idx].val,
646 conf->mac_addr, SZ_PEDIT_KEY_VAL);
648 p_parser->keys[idx].off = off + SZ_PEDIT_KEY_VAL;
649 p_parser->keys[idx].mask = 0xFFFF0000;
650 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_ETH;
651 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
652 memcpy(&p_parser->keys[idx].val,
653 conf->mac_addr + SZ_PEDIT_KEY_VAL,
654 ETHER_ADDR_LEN - SZ_PEDIT_KEY_VAL);
655 p_parser->sel.nkeys = (++idx);
659 * Set pedit key of decrease/set ttl
662 * pointer to action specification
663 * @param[in,out] p_parser
664 * pointer to pedit_parser
665 * @param[in] item_flags
666 * flags of all items presented
669 flow_tcf_pedit_key_set_dec_ttl(const struct rte_flow_action *actions,
670 struct pedit_parser *p_parser,
673 int idx = p_parser->sel.nkeys;
675 p_parser->keys[idx].mask = 0xFFFFFF00;
676 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4) {
677 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP4;
678 p_parser->keys[idx].off =
679 offsetof(struct ipv4_hdr, time_to_live);
681 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6) {
682 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP6;
683 p_parser->keys[idx].off =
684 offsetof(struct ipv6_hdr, hop_limits);
686 if (actions->type == RTE_FLOW_ACTION_TYPE_DEC_TTL) {
687 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_ADD;
688 p_parser->keys[idx].val = 0x000000FF;
690 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
691 p_parser->keys[idx].val =
692 (__u32)((const struct rte_flow_action_set_ttl *)
693 actions->conf)->ttl_value;
695 p_parser->sel.nkeys = (++idx);
699 * Set pedit key of transport (TCP/UDP) port value
702 * pointer to action specification
703 * @param[in,out] p_parser
704 * pointer to pedit_parser
705 * @param[in] item_flags
706 * flags of all items presented
709 flow_tcf_pedit_key_set_tp_port(const struct rte_flow_action *actions,
710 struct pedit_parser *p_parser,
713 int idx = p_parser->sel.nkeys;
715 if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
716 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_UDP;
717 if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
718 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_TCP;
719 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
720 /* offset of src/dst port is same for TCP and UDP */
721 p_parser->keys[idx].off =
722 actions->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
723 offsetof(struct tcp_hdr, src_port) :
724 offsetof(struct tcp_hdr, dst_port);
725 p_parser->keys[idx].mask = 0xFFFF0000;
726 p_parser->keys[idx].val =
727 (__u32)((const struct rte_flow_action_set_tp *)
728 actions->conf)->port;
729 p_parser->sel.nkeys = (++idx);
733 * Set pedit key of ipv6 address
736 * pointer to action specification
737 * @param[in,out] p_parser
738 * pointer to pedit_parser
741 flow_tcf_pedit_key_set_ipv6_addr(const struct rte_flow_action *actions,
742 struct pedit_parser *p_parser)
744 int idx = p_parser->sel.nkeys;
745 int keys = NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
747 actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
748 offsetof(struct ipv6_hdr, src_addr) :
749 offsetof(struct ipv6_hdr, dst_addr);
750 const struct rte_flow_action_set_ipv6 *conf =
751 (const struct rte_flow_action_set_ipv6 *)actions->conf;
753 for (int i = 0; i < keys; i++, idx++) {
754 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP6;
755 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
756 p_parser->keys[idx].off = off_base + i * SZ_PEDIT_KEY_VAL;
757 p_parser->keys[idx].mask = ~UINT32_MAX;
758 memcpy(&p_parser->keys[idx].val,
759 conf->ipv6_addr + i * SZ_PEDIT_KEY_VAL,
762 p_parser->sel.nkeys += keys;
766 * Set pedit key of ipv4 address
769 * pointer to action specification
770 * @param[in,out] p_parser
771 * pointer to pedit_parser
774 flow_tcf_pedit_key_set_ipv4_addr(const struct rte_flow_action *actions,
775 struct pedit_parser *p_parser)
777 int idx = p_parser->sel.nkeys;
779 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP4;
780 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
781 p_parser->keys[idx].off =
782 actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
783 offsetof(struct ipv4_hdr, src_addr) :
784 offsetof(struct ipv4_hdr, dst_addr);
785 p_parser->keys[idx].mask = ~UINT32_MAX;
786 p_parser->keys[idx].val =
787 ((const struct rte_flow_action_set_ipv4 *)
788 actions->conf)->ipv4_addr;
789 p_parser->sel.nkeys = (++idx);
793 * Create the pedit's na attribute in netlink message
794 * on pre-allocate message buffer
797 * pointer to pre-allocated netlink message buffer
798 * @param[in,out] actions
799 * pointer to pointer of actions specification.
800 * @param[in,out] action_flags
801 * pointer to actions flags
802 * @param[in] item_flags
803 * flags of all item presented
806 flow_tcf_create_pedit_mnl_msg(struct nlmsghdr *nl,
807 const struct rte_flow_action **actions,
810 struct pedit_parser p_parser;
811 struct nlattr *na_act_options;
812 struct nlattr *na_pedit_keys;
814 memset(&p_parser, 0, sizeof(p_parser));
815 mnl_attr_put_strz(nl, TCA_ACT_KIND, "pedit");
816 na_act_options = mnl_attr_nest_start(nl, TCA_ACT_OPTIONS);
817 /* all modify header actions should be in one tc-pedit action */
818 for (; (*actions)->type != RTE_FLOW_ACTION_TYPE_END; (*actions)++) {
819 switch ((*actions)->type) {
820 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
821 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
822 flow_tcf_pedit_key_set_ipv4_addr(*actions, &p_parser);
824 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
825 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
826 flow_tcf_pedit_key_set_ipv6_addr(*actions, &p_parser);
828 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
829 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
830 flow_tcf_pedit_key_set_tp_port(*actions,
831 &p_parser, item_flags);
833 case RTE_FLOW_ACTION_TYPE_SET_TTL:
834 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
835 flow_tcf_pedit_key_set_dec_ttl(*actions,
836 &p_parser, item_flags);
838 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
839 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
840 flow_tcf_pedit_key_set_mac(*actions, &p_parser);
843 goto pedit_mnl_msg_done;
847 p_parser.sel.action = TC_ACT_PIPE;
848 mnl_attr_put(nl, TCA_PEDIT_PARMS_EX,
849 sizeof(p_parser.sel) +
850 p_parser.sel.nkeys * sizeof(struct tc_pedit_key),
853 mnl_attr_nest_start(nl, TCA_PEDIT_KEYS_EX | NLA_F_NESTED);
854 for (int i = 0; i < p_parser.sel.nkeys; i++) {
855 struct nlattr *na_pedit_key =
856 mnl_attr_nest_start(nl,
857 TCA_PEDIT_KEY_EX | NLA_F_NESTED);
858 mnl_attr_put_u16(nl, TCA_PEDIT_KEY_EX_HTYPE,
859 p_parser.keys_ex[i].htype);
860 mnl_attr_put_u16(nl, TCA_PEDIT_KEY_EX_CMD,
861 p_parser.keys_ex[i].cmd);
862 mnl_attr_nest_end(nl, na_pedit_key);
864 mnl_attr_nest_end(nl, na_pedit_keys);
865 mnl_attr_nest_end(nl, na_act_options);
870 * Calculate max memory size of one TC-pedit actions.
871 * One TC-pedit action can contain set of keys each defining
872 * a rewrite element (rte_flow action)
874 * @param[in,out] actions
875 * actions specification.
876 * @param[in,out] action_flags
878 * @param[in,out] size
881 * Max memory size of one TC-pedit action
884 flow_tcf_get_pedit_actions_size(const struct rte_flow_action **actions,
885 uint64_t *action_flags)
891 pedit_size += SZ_NLATTR_NEST + /* na_act_index. */
892 SZ_NLATTR_STRZ_OF("pedit") +
893 SZ_NLATTR_NEST; /* TCA_ACT_OPTIONS. */
894 for (; (*actions)->type != RTE_FLOW_ACTION_TYPE_END; (*actions)++) {
895 switch ((*actions)->type) {
896 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
897 keys += NUM_OF_PEDIT_KEYS(IPV4_ADDR_LEN);
898 flags |= MLX5_FLOW_ACTION_SET_IPV4_SRC;
900 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
901 keys += NUM_OF_PEDIT_KEYS(IPV4_ADDR_LEN);
902 flags |= MLX5_FLOW_ACTION_SET_IPV4_DST;
904 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
905 keys += NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
906 flags |= MLX5_FLOW_ACTION_SET_IPV6_SRC;
908 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
909 keys += NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
910 flags |= MLX5_FLOW_ACTION_SET_IPV6_DST;
912 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
913 /* TCP is as same as UDP */
914 keys += NUM_OF_PEDIT_KEYS(TP_PORT_LEN);
915 flags |= MLX5_FLOW_ACTION_SET_TP_SRC;
917 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
918 /* TCP is as same as UDP */
919 keys += NUM_OF_PEDIT_KEYS(TP_PORT_LEN);
920 flags |= MLX5_FLOW_ACTION_SET_TP_DST;
922 case RTE_FLOW_ACTION_TYPE_SET_TTL:
923 keys += NUM_OF_PEDIT_KEYS(TTL_LEN);
924 flags |= MLX5_FLOW_ACTION_SET_TTL;
926 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
927 keys += NUM_OF_PEDIT_KEYS(TTL_LEN);
928 flags |= MLX5_FLOW_ACTION_DEC_TTL;
930 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
931 keys += NUM_OF_PEDIT_KEYS(ETHER_ADDR_LEN);
932 flags |= MLX5_FLOW_ACTION_SET_MAC_SRC;
934 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
935 keys += NUM_OF_PEDIT_KEYS(ETHER_ADDR_LEN);
936 flags |= MLX5_FLOW_ACTION_SET_MAC_DST;
939 goto get_pedit_action_size_done;
942 get_pedit_action_size_done:
943 /* TCA_PEDIT_PARAMS_EX */
945 SZ_NLATTR_DATA_OF(sizeof(struct tc_pedit_sel) +
946 keys * sizeof(struct tc_pedit_key));
947 pedit_size += SZ_NLATTR_NEST; /* TCA_PEDIT_KEYS */
949 /* TCA_PEDIT_KEY_EX + HTYPE + CMD */
950 (SZ_NLATTR_NEST + SZ_NLATTR_DATA_OF(2) +
951 SZ_NLATTR_DATA_OF(2));
952 (*action_flags) |= flags;
958 * Retrieve mask for pattern item.
960 * This function does basic sanity checks on a pattern item in order to
961 * return the most appropriate mask for it.
964 * Item specification.
965 * @param[in] mask_default
966 * Default mask for pattern item as specified by the flow API.
967 * @param[in] mask_supported
968 * Mask fields supported by the implementation.
969 * @param[in] mask_empty
970 * Empty mask to return when there is no specification.
972 * Perform verbose error reporting if not NULL.
975 * Either @p item->mask or one of the mask parameters on success, NULL
976 * otherwise and rte_errno is set.
979 flow_tcf_item_mask(const struct rte_flow_item *item, const void *mask_default,
980 const void *mask_supported, const void *mask_empty,
981 size_t mask_size, struct rte_flow_error *error)
986 /* item->last and item->mask cannot exist without item->spec. */
987 if (!item->spec && (item->mask || item->last)) {
988 rte_flow_error_set(error, EINVAL,
989 RTE_FLOW_ERROR_TYPE_ITEM, item,
990 "\"mask\" or \"last\" field provided without"
991 " a corresponding \"spec\"");
994 /* No spec, no mask, no problem. */
997 mask = item->mask ? item->mask : mask_default;
1000 * Single-pass check to make sure that:
1001 * - Mask is supported, no bits are set outside mask_supported.
1002 * - Both item->spec and item->last are included in mask.
1004 for (i = 0; i != mask_size; ++i) {
1007 if ((mask[i] | ((const uint8_t *)mask_supported)[i]) !=
1008 ((const uint8_t *)mask_supported)[i]) {
1009 rte_flow_error_set(error, ENOTSUP,
1010 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1011 "unsupported field found"
1016 (((const uint8_t *)item->spec)[i] & mask[i]) !=
1017 (((const uint8_t *)item->last)[i] & mask[i])) {
1018 rte_flow_error_set(error, EINVAL,
1019 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1021 "range between \"spec\" and \"last\""
1022 " not comprised in \"mask\"");
1030 * Build a conversion table between port ID and ifindex.
1033 * Pointer to Ethernet device.
1035 * Pointer to ptoi table.
1037 * Size of ptoi table provided.
1040 * Size of ptoi table filled.
1043 flow_tcf_build_ptoi_table(struct rte_eth_dev *dev, struct flow_tcf_ptoi *ptoi,
1046 unsigned int n = mlx5_dev_to_port_id(dev->device, NULL, 0);
1047 uint16_t port_id[n + 1];
1049 unsigned int own = 0;
1051 /* At least one port is needed when no switch domain is present. */
1054 port_id[0] = dev->data->port_id;
1056 n = RTE_MIN(mlx5_dev_to_port_id(dev->device, port_id, n), n);
1060 for (i = 0; i != n; ++i) {
1061 struct rte_eth_dev_info dev_info;
1063 rte_eth_dev_info_get(port_id[i], &dev_info);
1064 if (port_id[i] == dev->data->port_id)
1066 ptoi[i].port_id = port_id[i];
1067 ptoi[i].ifindex = dev_info.if_index;
1069 /* Ensure first entry of ptoi[] is the current device. */
1072 ptoi[0] = ptoi[own];
1073 ptoi[own] = ptoi[n];
1075 /* An entry with zero ifindex terminates ptoi[]. */
1076 ptoi[n].port_id = 0;
1077 ptoi[n].ifindex = 0;
1082 * Verify the @p attr will be correctly understood by the E-switch.
1085 * Pointer to flow attributes
1087 * Pointer to error structure.
1090 * 0 on success, a negative errno value otherwise and rte_errno is set.
1093 flow_tcf_validate_attributes(const struct rte_flow_attr *attr,
1094 struct rte_flow_error *error)
1097 * Supported attributes: groups, some priorities and ingress only.
1098 * group is supported only if kernel supports chain. Don't care about
1099 * transfer as it is the caller's problem.
1101 if (attr->group > MLX5_TCF_GROUP_ID_MAX)
1102 return rte_flow_error_set(error, ENOTSUP,
1103 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, attr,
1104 "group ID larger than "
1105 RTE_STR(MLX5_TCF_GROUP_ID_MAX)
1106 " isn't supported");
1107 else if (attr->priority > MLX5_TCF_GROUP_PRIORITY_MAX)
1108 return rte_flow_error_set(error, ENOTSUP,
1109 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1111 "priority more than "
1112 RTE_STR(MLX5_TCF_GROUP_PRIORITY_MAX)
1113 " is not supported");
1115 return rte_flow_error_set(error, EINVAL,
1116 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1117 attr, "only ingress is supported");
1119 return rte_flow_error_set(error, ENOTSUP,
1120 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1121 attr, "egress is not supported");
1126 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_ETH item for E-Switch.
1127 * The routine checks the L2 fields to be used in encapsulation header.
1130 * Pointer to the item structure.
1132 * Pointer to the error structure.
1135 * 0 on success, a negative errno value otherwise and rte_errno is set.
1138 flow_tcf_validate_vxlan_encap_eth(const struct rte_flow_item *item,
1139 struct rte_flow_error *error)
1141 const struct rte_flow_item_eth *spec = item->spec;
1142 const struct rte_flow_item_eth *mask = item->mask;
1146 * Specification for L2 addresses can be empty
1147 * because these ones are optional and not
1148 * required directly by tc rule. Kernel tries
1149 * to resolve these ones on its own
1154 /* If mask is not specified use the default one. */
1155 mask = &rte_flow_item_eth_mask;
1157 if (memcmp(&mask->dst,
1158 &flow_tcf_mask_empty.eth.dst,
1159 sizeof(flow_tcf_mask_empty.eth.dst))) {
1160 if (memcmp(&mask->dst,
1161 &rte_flow_item_eth_mask.dst,
1162 sizeof(rte_flow_item_eth_mask.dst)))
1163 return rte_flow_error_set
1165 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1166 "no support for partial mask on"
1167 " \"eth.dst\" field");
1169 if (memcmp(&mask->src,
1170 &flow_tcf_mask_empty.eth.src,
1171 sizeof(flow_tcf_mask_empty.eth.src))) {
1172 if (memcmp(&mask->src,
1173 &rte_flow_item_eth_mask.src,
1174 sizeof(rte_flow_item_eth_mask.src)))
1175 return rte_flow_error_set
1177 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1178 "no support for partial mask on"
1179 " \"eth.src\" field");
1181 if (mask->type != RTE_BE16(0x0000)) {
1182 if (mask->type != RTE_BE16(0xffff))
1183 return rte_flow_error_set
1185 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1186 "no support for partial mask on"
1187 " \"eth.type\" field");
1189 "outer ethernet type field"
1190 " cannot be forced for vxlan"
1191 " encapsulation, parameter ignored");
1197 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_IPV4 item for E-Switch.
1198 * The routine checks the IPv4 fields to be used in encapsulation header.
1201 * Pointer to the item structure.
1203 * Pointer to the error structure.
1206 * 0 on success, a negative errno value otherwise and rte_errno is set.
1209 flow_tcf_validate_vxlan_encap_ipv4(const struct rte_flow_item *item,
1210 struct rte_flow_error *error)
1212 const struct rte_flow_item_ipv4 *spec = item->spec;
1213 const struct rte_flow_item_ipv4 *mask = item->mask;
1217 * Specification for IP addresses cannot be empty
1218 * because it is required by tunnel_key parameter.
1220 return rte_flow_error_set(error, EINVAL,
1221 RTE_FLOW_ERROR_TYPE_ITEM, item,
1222 "NULL outer ipv4 address"
1223 " specification for vxlan"
1227 mask = &rte_flow_item_ipv4_mask;
1228 if (mask->hdr.dst_addr != RTE_BE32(0x00000000)) {
1229 if (mask->hdr.dst_addr != RTE_BE32(0xffffffff))
1230 return rte_flow_error_set
1232 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1233 "no support for partial mask on"
1234 " \"ipv4.hdr.dst_addr\" field"
1235 " for vxlan encapsulation");
1236 /* More IPv4 address validations can be put here. */
1239 * Kernel uses the destination IP address to determine
1240 * the routing path and obtain the MAC destination
1241 * address, so IP destination address must be
1242 * specified in the tc rule.
1244 return rte_flow_error_set(error, EINVAL,
1245 RTE_FLOW_ERROR_TYPE_ITEM, item,
1246 "outer ipv4 destination address"
1247 " must be specified for"
1248 " vxlan encapsulation");
1250 if (mask->hdr.src_addr != RTE_BE32(0x00000000)) {
1251 if (mask->hdr.src_addr != RTE_BE32(0xffffffff))
1252 return rte_flow_error_set
1254 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1255 "no support for partial mask on"
1256 " \"ipv4.hdr.src_addr\" field"
1257 " for vxlan encapsulation");
1258 /* More IPv4 address validations can be put here. */
1261 * Kernel uses the source IP address to select the
1262 * interface for egress encapsulated traffic, so
1263 * it must be specified in the tc rule.
1265 return rte_flow_error_set(error, EINVAL,
1266 RTE_FLOW_ERROR_TYPE_ITEM, item,
1267 "outer ipv4 source address"
1268 " must be specified for"
1269 " vxlan encapsulation");
1275 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_IPV6 item for E-Switch.
1276 * The routine checks the IPv6 fields to be used in encapsulation header.
1279 * Pointer to the item structure.
1281 * Pointer to the error structure.
1284 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1287 flow_tcf_validate_vxlan_encap_ipv6(const struct rte_flow_item *item,
1288 struct rte_flow_error *error)
1290 const struct rte_flow_item_ipv6 *spec = item->spec;
1291 const struct rte_flow_item_ipv6 *mask = item->mask;
1295 * Specification for IP addresses cannot be empty
1296 * because it is required by tunnel_key parameter.
1298 return rte_flow_error_set(error, EINVAL,
1299 RTE_FLOW_ERROR_TYPE_ITEM, item,
1300 "NULL outer ipv6 address"
1301 " specification for"
1302 " vxlan encapsulation");
1305 mask = &rte_flow_item_ipv6_mask;
1306 if (memcmp(&mask->hdr.dst_addr,
1307 &flow_tcf_mask_empty.ipv6.hdr.dst_addr,
1309 if (memcmp(&mask->hdr.dst_addr,
1310 &rte_flow_item_ipv6_mask.hdr.dst_addr,
1312 return rte_flow_error_set
1314 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1315 "no support for partial mask on"
1316 " \"ipv6.hdr.dst_addr\" field"
1317 " for vxlan encapsulation");
1318 /* More IPv6 address validations can be put here. */
1321 * Kernel uses the destination IP address to determine
1322 * the routing path and obtain the MAC destination
1323 * address (heigh or gate), so IP destination address
1324 * must be specified within the tc rule.
1326 return rte_flow_error_set(error, EINVAL,
1327 RTE_FLOW_ERROR_TYPE_ITEM, item,
1328 "outer ipv6 destination address"
1329 " must be specified for"
1330 " vxlan encapsulation");
1332 if (memcmp(&mask->hdr.src_addr,
1333 &flow_tcf_mask_empty.ipv6.hdr.src_addr,
1335 if (memcmp(&mask->hdr.src_addr,
1336 &rte_flow_item_ipv6_mask.hdr.src_addr,
1338 return rte_flow_error_set
1340 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1341 "no support for partial mask on"
1342 " \"ipv6.hdr.src_addr\" field"
1343 " for vxlan encapsulation");
1344 /* More L3 address validation can be put here. */
1347 * Kernel uses the source IP address to select the
1348 * interface for egress encapsulated traffic, so
1349 * it must be specified in the tc rule.
1351 return rte_flow_error_set(error, EINVAL,
1352 RTE_FLOW_ERROR_TYPE_ITEM, item,
1353 "outer L3 source address"
1354 " must be specified for"
1355 " vxlan encapsulation");
1361 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_UDP item for E-Switch.
1362 * The routine checks the UDP fields to be used in encapsulation header.
1365 * Pointer to the item structure.
1367 * Pointer to the error structure.
1370 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1373 flow_tcf_validate_vxlan_encap_udp(const struct rte_flow_item *item,
1374 struct rte_flow_error *error)
1376 const struct rte_flow_item_udp *spec = item->spec;
1377 const struct rte_flow_item_udp *mask = item->mask;
1381 * Specification for UDP ports cannot be empty
1382 * because it is required by tunnel_key parameter.
1384 return rte_flow_error_set(error, EINVAL,
1385 RTE_FLOW_ERROR_TYPE_ITEM, item,
1386 "NULL UDP port specification "
1387 " for vxlan encapsulation");
1390 mask = &rte_flow_item_udp_mask;
1391 if (mask->hdr.dst_port != RTE_BE16(0x0000)) {
1392 if (mask->hdr.dst_port != RTE_BE16(0xffff))
1393 return rte_flow_error_set
1395 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1396 "no support for partial mask on"
1397 " \"udp.hdr.dst_port\" field"
1398 " for vxlan encapsulation");
1399 if (!spec->hdr.dst_port)
1400 return rte_flow_error_set
1402 RTE_FLOW_ERROR_TYPE_ITEM, item,
1403 "outer UDP remote port cannot be"
1404 " 0 for vxlan encapsulation");
1406 return rte_flow_error_set(error, EINVAL,
1407 RTE_FLOW_ERROR_TYPE_ITEM, item,
1408 "outer UDP remote port"
1409 " must be specified for"
1410 " vxlan encapsulation");
1412 if (mask->hdr.src_port != RTE_BE16(0x0000)) {
1413 if (mask->hdr.src_port != RTE_BE16(0xffff))
1414 return rte_flow_error_set
1416 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1417 "no support for partial mask on"
1418 " \"udp.hdr.src_port\" field"
1419 " for vxlan encapsulation");
1421 "outer UDP source port cannot be"
1422 " forced for vxlan encapsulation,"
1423 " parameter ignored");
1429 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_VXLAN item for E-Switch.
1430 * The routine checks the VNIP fields to be used in encapsulation header.
1433 * Pointer to the item structure.
1435 * Pointer to the error structure.
1438 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1441 flow_tcf_validate_vxlan_encap_vni(const struct rte_flow_item *item,
1442 struct rte_flow_error *error)
1444 const struct rte_flow_item_vxlan *spec = item->spec;
1445 const struct rte_flow_item_vxlan *mask = item->mask;
1448 /* Outer VNI is required by tunnel_key parameter. */
1449 return rte_flow_error_set(error, EINVAL,
1450 RTE_FLOW_ERROR_TYPE_ITEM, item,
1451 "NULL VNI specification"
1452 " for vxlan encapsulation");
1455 mask = &rte_flow_item_vxlan_mask;
1456 if (!mask->vni[0] && !mask->vni[1] && !mask->vni[2])
1457 return rte_flow_error_set(error, EINVAL,
1458 RTE_FLOW_ERROR_TYPE_ITEM, item,
1459 "outer VNI must be specified "
1460 "for vxlan encapsulation");
1461 if (mask->vni[0] != 0xff ||
1462 mask->vni[1] != 0xff ||
1463 mask->vni[2] != 0xff)
1464 return rte_flow_error_set(error, ENOTSUP,
1465 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1466 "no support for partial mask on"
1467 " \"vxlan.vni\" field");
1469 if (!spec->vni[0] && !spec->vni[1] && !spec->vni[2])
1470 return rte_flow_error_set(error, EINVAL,
1471 RTE_FLOW_ERROR_TYPE_ITEM, item,
1472 "vxlan vni cannot be 0");
1477 * Validate VXLAN_ENCAP action item list for E-Switch.
1478 * The routine checks items to be used in encapsulation header.
1481 * Pointer to the VXLAN_ENCAP action structure.
1483 * Pointer to the error structure.
1486 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1489 flow_tcf_validate_vxlan_encap(const struct rte_flow_action *action,
1490 struct rte_flow_error *error)
1492 const struct rte_flow_item *items;
1494 uint32_t item_flags = 0;
1497 return rte_flow_error_set(error, EINVAL,
1498 RTE_FLOW_ERROR_TYPE_ACTION, action,
1499 "Missing vxlan tunnel"
1500 " action configuration");
1501 items = ((const struct rte_flow_action_vxlan_encap *)
1502 action->conf)->definition;
1504 return rte_flow_error_set(error, EINVAL,
1505 RTE_FLOW_ERROR_TYPE_ACTION, action,
1506 "Missing vxlan tunnel"
1507 " encapsulation parameters");
1508 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1509 switch (items->type) {
1510 case RTE_FLOW_ITEM_TYPE_VOID:
1512 case RTE_FLOW_ITEM_TYPE_ETH:
1513 ret = mlx5_flow_validate_item_eth(items, item_flags,
1517 ret = flow_tcf_validate_vxlan_encap_eth(items, error);
1520 item_flags |= MLX5_FLOW_LAYER_OUTER_L2;
1523 case RTE_FLOW_ITEM_TYPE_IPV4:
1524 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
1528 ret = flow_tcf_validate_vxlan_encap_ipv4(items, error);
1531 item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1533 case RTE_FLOW_ITEM_TYPE_IPV6:
1534 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
1538 ret = flow_tcf_validate_vxlan_encap_ipv6(items, error);
1541 item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1543 case RTE_FLOW_ITEM_TYPE_UDP:
1544 ret = mlx5_flow_validate_item_udp(items, item_flags,
1548 ret = flow_tcf_validate_vxlan_encap_udp(items, error);
1551 item_flags |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
1553 case RTE_FLOW_ITEM_TYPE_VXLAN:
1554 ret = mlx5_flow_validate_item_vxlan(items,
1558 ret = flow_tcf_validate_vxlan_encap_vni(items, error);
1561 item_flags |= MLX5_FLOW_LAYER_VXLAN;
1564 return rte_flow_error_set
1566 RTE_FLOW_ERROR_TYPE_ITEM, items,
1567 "vxlan encap item not supported");
1570 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1571 return rte_flow_error_set(error, EINVAL,
1572 RTE_FLOW_ERROR_TYPE_ACTION, action,
1573 "no outer IP layer found"
1574 " for vxlan encapsulation");
1575 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1576 return rte_flow_error_set(error, EINVAL,
1577 RTE_FLOW_ERROR_TYPE_ACTION, action,
1578 "no outer UDP layer found"
1579 " for vxlan encapsulation");
1580 if (!(item_flags & MLX5_FLOW_LAYER_VXLAN))
1581 return rte_flow_error_set(error, EINVAL,
1582 RTE_FLOW_ERROR_TYPE_ACTION, action,
1583 "no VXLAN VNI found"
1584 " for vxlan encapsulation");
1589 * Validate outer RTE_FLOW_ITEM_TYPE_UDP item if tunnel item
1590 * RTE_FLOW_ITEM_TYPE_VXLAN is present in item list.
1593 * Outer UDP layer item (if any, NULL otherwise).
1595 * Pointer to the error structure.
1598 * 0 on success, a negative errno value otherwise and rte_errno is set.
1601 flow_tcf_validate_vxlan_decap_udp(const struct rte_flow_item *udp,
1602 struct rte_flow_error *error)
1604 const struct rte_flow_item_udp *spec = udp->spec;
1605 const struct rte_flow_item_udp *mask = udp->mask;
1609 * Specification for UDP ports cannot be empty
1610 * because it is required as decap parameter.
1612 return rte_flow_error_set(error, EINVAL,
1613 RTE_FLOW_ERROR_TYPE_ITEM, udp,
1614 "NULL UDP port specification"
1615 " for VXLAN decapsulation");
1617 mask = &rte_flow_item_udp_mask;
1618 if (mask->hdr.dst_port != RTE_BE16(0x0000)) {
1619 if (mask->hdr.dst_port != RTE_BE16(0xffff))
1620 return rte_flow_error_set
1622 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1623 "no support for partial mask on"
1624 " \"udp.hdr.dst_port\" field");
1625 if (!spec->hdr.dst_port)
1626 return rte_flow_error_set
1628 RTE_FLOW_ERROR_TYPE_ITEM, udp,
1629 "zero decap local UDP port");
1631 return rte_flow_error_set(error, EINVAL,
1632 RTE_FLOW_ERROR_TYPE_ITEM, udp,
1633 "outer UDP destination port must be "
1634 "specified for vxlan decapsulation");
1636 if (mask->hdr.src_port != RTE_BE16(0x0000)) {
1637 if (mask->hdr.src_port != RTE_BE16(0xffff))
1638 return rte_flow_error_set
1640 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1641 "no support for partial mask on"
1642 " \"udp.hdr.src_port\" field");
1644 "outer UDP local port cannot be "
1645 "forced for VXLAN encapsulation, "
1646 "parameter ignored");
1652 * Validate flow for E-Switch.
1655 * Pointer to the priv structure.
1657 * Pointer to the flow attributes.
1659 * Pointer to the list of items.
1660 * @param[in] actions
1661 * Pointer to the list of actions.
1663 * Pointer to the error structure.
1666 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1669 flow_tcf_validate(struct rte_eth_dev *dev,
1670 const struct rte_flow_attr *attr,
1671 const struct rte_flow_item items[],
1672 const struct rte_flow_action actions[],
1673 struct rte_flow_error *error)
1676 const struct rte_flow_item_port_id *port_id;
1677 const struct rte_flow_item_eth *eth;
1678 const struct rte_flow_item_vlan *vlan;
1679 const struct rte_flow_item_ipv4 *ipv4;
1680 const struct rte_flow_item_ipv6 *ipv6;
1681 const struct rte_flow_item_tcp *tcp;
1682 const struct rte_flow_item_udp *udp;
1683 const struct rte_flow_item_vxlan *vxlan;
1686 const struct rte_flow_action_port_id *port_id;
1687 const struct rte_flow_action_jump *jump;
1688 const struct rte_flow_action_of_push_vlan *of_push_vlan;
1689 const struct rte_flow_action_of_set_vlan_vid *
1691 const struct rte_flow_action_of_set_vlan_pcp *
1693 const struct rte_flow_action_vxlan_encap *vxlan_encap;
1694 const struct rte_flow_action_set_ipv4 *set_ipv4;
1695 const struct rte_flow_action_set_ipv6 *set_ipv6;
1697 const struct rte_flow_item *outer_udp = NULL;
1698 rte_be16_t inner_etype = RTE_BE16(ETH_P_ALL);
1699 rte_be16_t outer_etype = RTE_BE16(ETH_P_ALL);
1700 rte_be16_t vlan_etype = RTE_BE16(ETH_P_ALL);
1701 uint64_t item_flags = 0;
1702 uint64_t action_flags = 0;
1703 uint8_t next_protocol = 0xff;
1704 unsigned int tcm_ifindex = 0;
1705 uint8_t pedit_validated = 0;
1706 struct flow_tcf_ptoi ptoi[PTOI_TABLE_SZ_MAX(dev)];
1707 struct rte_eth_dev *port_id_dev = NULL;
1708 bool in_port_id_set;
1711 claim_nonzero(flow_tcf_build_ptoi_table(dev, ptoi,
1712 PTOI_TABLE_SZ_MAX(dev)));
1713 ret = flow_tcf_validate_attributes(attr, error);
1716 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1718 uint64_t current_action_flag = 0;
1720 switch (actions->type) {
1721 case RTE_FLOW_ACTION_TYPE_VOID:
1723 case RTE_FLOW_ACTION_TYPE_PORT_ID:
1724 current_action_flag = MLX5_FLOW_ACTION_PORT_ID;
1727 conf.port_id = actions->conf;
1728 if (conf.port_id->original)
1731 for (i = 0; ptoi[i].ifindex; ++i)
1732 if (ptoi[i].port_id == conf.port_id->id)
1734 if (!ptoi[i].ifindex)
1735 return rte_flow_error_set
1737 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1739 "missing data to convert port ID to"
1741 port_id_dev = &rte_eth_devices[conf.port_id->id];
1743 case RTE_FLOW_ACTION_TYPE_JUMP:
1744 current_action_flag = MLX5_FLOW_ACTION_JUMP;
1747 conf.jump = actions->conf;
1748 if (attr->group >= conf.jump->group)
1749 return rte_flow_error_set
1751 RTE_FLOW_ERROR_TYPE_ACTION,
1753 "can jump only to a group forward");
1755 case RTE_FLOW_ACTION_TYPE_DROP:
1756 current_action_flag = MLX5_FLOW_ACTION_DROP;
1758 case RTE_FLOW_ACTION_TYPE_COUNT:
1760 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
1761 current_action_flag = MLX5_FLOW_ACTION_OF_POP_VLAN;
1763 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN: {
1764 rte_be16_t ethertype;
1766 current_action_flag = MLX5_FLOW_ACTION_OF_PUSH_VLAN;
1769 conf.of_push_vlan = actions->conf;
1770 ethertype = conf.of_push_vlan->ethertype;
1771 if (ethertype != RTE_BE16(ETH_P_8021Q) &&
1772 ethertype != RTE_BE16(ETH_P_8021AD))
1773 return rte_flow_error_set
1775 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1776 "vlan push TPID must be "
1777 "802.1Q or 802.1AD");
1780 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
1781 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
1782 return rte_flow_error_set
1784 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1785 "vlan modify is not supported,"
1786 " set action must follow push action");
1787 current_action_flag = MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
1789 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
1790 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
1791 return rte_flow_error_set
1793 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1794 "vlan modify is not supported,"
1795 " set action must follow push action");
1796 current_action_flag = MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
1798 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
1799 current_action_flag = MLX5_FLOW_ACTION_VXLAN_DECAP;
1801 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
1802 ret = flow_tcf_validate_vxlan_encap(actions, error);
1805 current_action_flag = MLX5_FLOW_ACTION_VXLAN_ENCAP;
1807 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
1808 current_action_flag = MLX5_FLOW_ACTION_SET_IPV4_SRC;
1810 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
1811 current_action_flag = MLX5_FLOW_ACTION_SET_IPV4_DST;
1813 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
1814 current_action_flag = MLX5_FLOW_ACTION_SET_IPV6_SRC;
1816 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
1817 current_action_flag = MLX5_FLOW_ACTION_SET_IPV6_DST;
1819 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
1820 current_action_flag = MLX5_FLOW_ACTION_SET_TP_SRC;
1822 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
1823 current_action_flag = MLX5_FLOW_ACTION_SET_TP_DST;
1825 case RTE_FLOW_ACTION_TYPE_SET_TTL:
1826 current_action_flag = MLX5_FLOW_ACTION_SET_TTL;
1828 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
1829 current_action_flag = MLX5_FLOW_ACTION_DEC_TTL;
1831 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
1832 current_action_flag = MLX5_FLOW_ACTION_SET_MAC_SRC;
1834 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
1835 current_action_flag = MLX5_FLOW_ACTION_SET_MAC_DST;
1838 return rte_flow_error_set(error, ENOTSUP,
1839 RTE_FLOW_ERROR_TYPE_ACTION,
1841 "action not supported");
1843 if (current_action_flag & MLX5_TCF_CONFIG_ACTIONS) {
1845 return rte_flow_error_set
1847 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1849 "action configuration not set");
1851 if ((current_action_flag & MLX5_TCF_PEDIT_ACTIONS) &&
1853 return rte_flow_error_set(error, ENOTSUP,
1854 RTE_FLOW_ERROR_TYPE_ACTION,
1856 "set actions should be "
1857 "listed successively");
1858 if ((current_action_flag & ~MLX5_TCF_PEDIT_ACTIONS) &&
1859 (action_flags & MLX5_TCF_PEDIT_ACTIONS))
1860 pedit_validated = 1;
1861 if ((current_action_flag & MLX5_TCF_FATE_ACTIONS) &&
1862 (action_flags & MLX5_TCF_FATE_ACTIONS))
1863 return rte_flow_error_set(error, EINVAL,
1864 RTE_FLOW_ERROR_TYPE_ACTION,
1866 "can't have multiple fate"
1868 if ((current_action_flag & MLX5_TCF_VXLAN_ACTIONS) &&
1869 (action_flags & MLX5_TCF_VXLAN_ACTIONS))
1870 return rte_flow_error_set(error, EINVAL,
1871 RTE_FLOW_ERROR_TYPE_ACTION,
1873 "can't have multiple vxlan"
1875 if ((current_action_flag & MLX5_TCF_VXLAN_ACTIONS) &&
1876 (action_flags & MLX5_TCF_VLAN_ACTIONS))
1877 return rte_flow_error_set(error, ENOTSUP,
1878 RTE_FLOW_ERROR_TYPE_ACTION,
1880 "can't have vxlan and vlan"
1881 " actions in the same rule");
1882 action_flags |= current_action_flag;
1884 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1887 switch (items->type) {
1888 case RTE_FLOW_ITEM_TYPE_VOID:
1890 case RTE_FLOW_ITEM_TYPE_PORT_ID:
1891 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1892 return rte_flow_error_set
1894 RTE_FLOW_ERROR_TYPE_ITEM, items,
1895 "inner tunnel port id"
1896 " item is not supported");
1897 mask.port_id = flow_tcf_item_mask
1898 (items, &rte_flow_item_port_id_mask,
1899 &flow_tcf_mask_supported.port_id,
1900 &flow_tcf_mask_empty.port_id,
1901 sizeof(flow_tcf_mask_supported.port_id),
1905 if (mask.port_id == &flow_tcf_mask_empty.port_id) {
1909 spec.port_id = items->spec;
1910 if (mask.port_id->id && mask.port_id->id != 0xffffffff)
1911 return rte_flow_error_set
1913 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1915 "no support for partial mask on"
1917 if (!mask.port_id->id)
1920 for (i = 0; ptoi[i].ifindex; ++i)
1921 if (ptoi[i].port_id == spec.port_id->id)
1923 if (!ptoi[i].ifindex)
1924 return rte_flow_error_set
1926 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1928 "missing data to convert port ID to"
1930 if (in_port_id_set && ptoi[i].ifindex != tcm_ifindex)
1931 return rte_flow_error_set
1933 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1935 "cannot match traffic for"
1936 " several port IDs through"
1937 " a single flow rule");
1938 tcm_ifindex = ptoi[i].ifindex;
1941 case RTE_FLOW_ITEM_TYPE_ETH:
1942 ret = mlx5_flow_validate_item_eth(items, item_flags,
1946 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
1947 MLX5_FLOW_LAYER_INNER_L2 :
1948 MLX5_FLOW_LAYER_OUTER_L2;
1950 * Redundant check due to different supported mask.
1951 * Same for the rest of items.
1953 mask.eth = flow_tcf_item_mask
1954 (items, &rte_flow_item_eth_mask,
1955 &flow_tcf_mask_supported.eth,
1956 &flow_tcf_mask_empty.eth,
1957 sizeof(flow_tcf_mask_supported.eth),
1961 if (mask.eth->type && mask.eth->type !=
1963 return rte_flow_error_set
1965 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1967 "no support for partial mask on"
1969 assert(items->spec);
1970 spec.eth = items->spec;
1971 if (mask.eth->type &&
1972 (item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1973 inner_etype != RTE_BE16(ETH_P_ALL) &&
1974 inner_etype != spec.eth->type)
1975 return rte_flow_error_set
1977 RTE_FLOW_ERROR_TYPE_ITEM,
1979 "inner eth_type conflict");
1980 if (mask.eth->type &&
1981 !(item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1982 outer_etype != RTE_BE16(ETH_P_ALL) &&
1983 outer_etype != spec.eth->type)
1984 return rte_flow_error_set
1986 RTE_FLOW_ERROR_TYPE_ITEM,
1988 "outer eth_type conflict");
1989 if (mask.eth->type) {
1990 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1991 inner_etype = spec.eth->type;
1993 outer_etype = spec.eth->type;
1996 case RTE_FLOW_ITEM_TYPE_VLAN:
1997 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1998 return rte_flow_error_set
2000 RTE_FLOW_ERROR_TYPE_ITEM, items,
2002 " is not supported");
2003 ret = mlx5_flow_validate_item_vlan(items, item_flags,
2007 item_flags |= MLX5_FLOW_LAYER_OUTER_VLAN;
2008 mask.vlan = flow_tcf_item_mask
2009 (items, &rte_flow_item_vlan_mask,
2010 &flow_tcf_mask_supported.vlan,
2011 &flow_tcf_mask_empty.vlan,
2012 sizeof(flow_tcf_mask_supported.vlan),
2016 if ((mask.vlan->tci & RTE_BE16(0xe000) &&
2017 (mask.vlan->tci & RTE_BE16(0xe000)) !=
2018 RTE_BE16(0xe000)) ||
2019 (mask.vlan->tci & RTE_BE16(0x0fff) &&
2020 (mask.vlan->tci & RTE_BE16(0x0fff)) !=
2021 RTE_BE16(0x0fff)) ||
2022 (mask.vlan->inner_type &&
2023 mask.vlan->inner_type != RTE_BE16(0xffff)))
2024 return rte_flow_error_set
2026 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2028 "no support for partial masks on"
2029 " \"tci\" (PCP and VID parts) and"
2030 " \"inner_type\" fields");
2031 if (outer_etype != RTE_BE16(ETH_P_ALL) &&
2032 outer_etype != RTE_BE16(ETH_P_8021Q))
2033 return rte_flow_error_set
2035 RTE_FLOW_ERROR_TYPE_ITEM,
2037 "outer eth_type conflict,"
2039 outer_etype = RTE_BE16(ETH_P_8021Q);
2040 assert(items->spec);
2041 spec.vlan = items->spec;
2042 if (mask.vlan->inner_type &&
2043 vlan_etype != RTE_BE16(ETH_P_ALL) &&
2044 vlan_etype != spec.vlan->inner_type)
2045 return rte_flow_error_set
2047 RTE_FLOW_ERROR_TYPE_ITEM,
2049 "vlan eth_type conflict");
2050 if (mask.vlan->inner_type)
2051 vlan_etype = spec.vlan->inner_type;
2053 case RTE_FLOW_ITEM_TYPE_IPV4:
2054 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
2058 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
2059 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2060 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2061 mask.ipv4 = flow_tcf_item_mask
2062 (items, &rte_flow_item_ipv4_mask,
2063 &flow_tcf_mask_supported.ipv4,
2064 &flow_tcf_mask_empty.ipv4,
2065 sizeof(flow_tcf_mask_supported.ipv4),
2069 if (mask.ipv4->hdr.next_proto_id &&
2070 mask.ipv4->hdr.next_proto_id != 0xff)
2071 return rte_flow_error_set
2073 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2075 "no support for partial mask on"
2076 " \"hdr.next_proto_id\" field");
2077 else if (mask.ipv4->hdr.next_proto_id)
2079 ((const struct rte_flow_item_ipv4 *)
2080 (items->spec))->hdr.next_proto_id;
2081 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2082 if (inner_etype != RTE_BE16(ETH_P_ALL) &&
2083 inner_etype != RTE_BE16(ETH_P_IP))
2084 return rte_flow_error_set
2086 RTE_FLOW_ERROR_TYPE_ITEM,
2088 "inner eth_type conflict,"
2089 " IPv4 is required");
2090 inner_etype = RTE_BE16(ETH_P_IP);
2091 } else if (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN) {
2092 if (vlan_etype != RTE_BE16(ETH_P_ALL) &&
2093 vlan_etype != RTE_BE16(ETH_P_IP))
2094 return rte_flow_error_set
2096 RTE_FLOW_ERROR_TYPE_ITEM,
2098 "vlan eth_type conflict,"
2099 " IPv4 is required");
2100 vlan_etype = RTE_BE16(ETH_P_IP);
2102 if (outer_etype != RTE_BE16(ETH_P_ALL) &&
2103 outer_etype != RTE_BE16(ETH_P_IP))
2104 return rte_flow_error_set
2106 RTE_FLOW_ERROR_TYPE_ITEM,
2108 "eth_type conflict,"
2109 " IPv4 is required");
2110 outer_etype = RTE_BE16(ETH_P_IP);
2113 case RTE_FLOW_ITEM_TYPE_IPV6:
2114 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
2118 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
2119 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2120 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2121 mask.ipv6 = flow_tcf_item_mask
2122 (items, &rte_flow_item_ipv6_mask,
2123 &flow_tcf_mask_supported.ipv6,
2124 &flow_tcf_mask_empty.ipv6,
2125 sizeof(flow_tcf_mask_supported.ipv6),
2129 if (mask.ipv6->hdr.proto &&
2130 mask.ipv6->hdr.proto != 0xff)
2131 return rte_flow_error_set
2133 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2135 "no support for partial mask on"
2136 " \"hdr.proto\" field");
2137 else if (mask.ipv6->hdr.proto)
2139 ((const struct rte_flow_item_ipv6 *)
2140 (items->spec))->hdr.proto;
2141 if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2142 if (inner_etype != RTE_BE16(ETH_P_ALL) &&
2143 inner_etype != RTE_BE16(ETH_P_IPV6))
2144 return rte_flow_error_set
2146 RTE_FLOW_ERROR_TYPE_ITEM,
2148 "inner eth_type conflict,"
2149 " IPv6 is required");
2150 inner_etype = RTE_BE16(ETH_P_IPV6);
2151 } else if (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN) {
2152 if (vlan_etype != RTE_BE16(ETH_P_ALL) &&
2153 vlan_etype != RTE_BE16(ETH_P_IPV6))
2154 return rte_flow_error_set
2156 RTE_FLOW_ERROR_TYPE_ITEM,
2158 "vlan eth_type conflict,"
2159 " IPv6 is required");
2160 vlan_etype = RTE_BE16(ETH_P_IPV6);
2162 if (outer_etype != RTE_BE16(ETH_P_ALL) &&
2163 outer_etype != RTE_BE16(ETH_P_IPV6))
2164 return rte_flow_error_set
2166 RTE_FLOW_ERROR_TYPE_ITEM,
2168 "eth_type conflict,"
2169 " IPv6 is required");
2170 outer_etype = RTE_BE16(ETH_P_IPV6);
2173 case RTE_FLOW_ITEM_TYPE_UDP:
2174 ret = mlx5_flow_validate_item_udp(items, item_flags,
2175 next_protocol, error);
2178 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
2179 MLX5_FLOW_LAYER_INNER_L4_UDP :
2180 MLX5_FLOW_LAYER_OUTER_L4_UDP;
2181 mask.udp = flow_tcf_item_mask
2182 (items, &rte_flow_item_udp_mask,
2183 &flow_tcf_mask_supported.udp,
2184 &flow_tcf_mask_empty.udp,
2185 sizeof(flow_tcf_mask_supported.udp),
2190 * Save the presumed outer UDP item for extra check
2191 * if the tunnel item will be found later in the list.
2193 if (!(item_flags & MLX5_FLOW_LAYER_TUNNEL))
2196 case RTE_FLOW_ITEM_TYPE_TCP:
2197 ret = mlx5_flow_validate_item_tcp
2200 &flow_tcf_mask_supported.tcp,
2204 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
2205 MLX5_FLOW_LAYER_INNER_L4_TCP :
2206 MLX5_FLOW_LAYER_OUTER_L4_TCP;
2207 mask.tcp = flow_tcf_item_mask
2208 (items, &rte_flow_item_tcp_mask,
2209 &flow_tcf_mask_supported.tcp,
2210 &flow_tcf_mask_empty.tcp,
2211 sizeof(flow_tcf_mask_supported.tcp),
2216 case RTE_FLOW_ITEM_TYPE_VXLAN:
2217 if (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)
2218 return rte_flow_error_set
2220 RTE_FLOW_ERROR_TYPE_ITEM, items,
2221 "vxlan tunnel over vlan"
2222 " is not supported");
2223 ret = mlx5_flow_validate_item_vxlan(items,
2227 item_flags |= MLX5_FLOW_LAYER_VXLAN;
2228 mask.vxlan = flow_tcf_item_mask
2229 (items, &rte_flow_item_vxlan_mask,
2230 &flow_tcf_mask_supported.vxlan,
2231 &flow_tcf_mask_empty.vxlan,
2232 sizeof(flow_tcf_mask_supported.vxlan), error);
2235 if (mask.vxlan->vni[0] != 0xff ||
2236 mask.vxlan->vni[1] != 0xff ||
2237 mask.vxlan->vni[2] != 0xff)
2238 return rte_flow_error_set
2240 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2242 "no support for partial or "
2243 "empty mask on \"vxlan.vni\" field");
2245 * The VNI item assumes the VXLAN tunnel, it requires
2246 * at least the outer destination UDP port must be
2247 * specified without wildcards to allow kernel select
2248 * the virtual VXLAN device by port. Also outer IPv4
2249 * or IPv6 item must be specified (wilcards or even
2250 * zero mask are allowed) to let driver know the tunnel
2251 * IP version and process UDP traffic correctly.
2254 (MLX5_FLOW_LAYER_OUTER_L3_IPV4 |
2255 MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
2256 return rte_flow_error_set
2258 RTE_FLOW_ERROR_TYPE_ACTION,
2260 "no outer IP pattern found"
2261 " for vxlan tunnel");
2262 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2263 return rte_flow_error_set
2265 RTE_FLOW_ERROR_TYPE_ACTION,
2267 "no outer UDP pattern found"
2268 " for vxlan tunnel");
2270 * All items preceding the tunnel item become outer
2271 * ones and we should do extra validation for them
2272 * due to tc limitations for tunnel outer parameters.
2273 * Currently only outer UDP item requres extra check,
2274 * use the saved pointer instead of item list rescan.
2277 ret = flow_tcf_validate_vxlan_decap_udp
2281 /* Reset L4 protocol for inner parameters. */
2282 next_protocol = 0xff;
2285 return rte_flow_error_set(error, ENOTSUP,
2286 RTE_FLOW_ERROR_TYPE_ITEM,
2287 items, "item not supported");
2290 if ((action_flags & MLX5_TCF_PEDIT_ACTIONS) &&
2291 (action_flags & MLX5_FLOW_ACTION_DROP))
2292 return rte_flow_error_set(error, ENOTSUP,
2293 RTE_FLOW_ERROR_TYPE_ACTION,
2295 "set action is not compatible with "
2297 if ((action_flags & MLX5_TCF_PEDIT_ACTIONS) &&
2298 !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2299 return rte_flow_error_set(error, ENOTSUP,
2300 RTE_FLOW_ERROR_TYPE_ACTION,
2302 "set action must be followed by "
2305 (MLX5_FLOW_ACTION_SET_IPV4_SRC | MLX5_FLOW_ACTION_SET_IPV4_DST)) {
2306 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4))
2307 return rte_flow_error_set(error, EINVAL,
2308 RTE_FLOW_ERROR_TYPE_ACTION,
2310 "no ipv4 item found in"
2314 (MLX5_FLOW_ACTION_SET_IPV6_SRC | MLX5_FLOW_ACTION_SET_IPV6_DST)) {
2315 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6))
2316 return rte_flow_error_set(error, EINVAL,
2317 RTE_FLOW_ERROR_TYPE_ACTION,
2319 "no ipv6 item found in"
2323 (MLX5_FLOW_ACTION_SET_TP_SRC | MLX5_FLOW_ACTION_SET_TP_DST)) {
2325 (MLX5_FLOW_LAYER_OUTER_L4_UDP |
2326 MLX5_FLOW_LAYER_OUTER_L4_TCP)))
2327 return rte_flow_error_set(error, EINVAL,
2328 RTE_FLOW_ERROR_TYPE_ACTION,
2330 "no TCP/UDP item found in"
2334 * FW syndrome (0xA9C090):
2335 * set_flow_table_entry: push vlan action fte in fdb can ONLY be
2336 * forward to the uplink.
2338 if ((action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2339 (action_flags & MLX5_FLOW_ACTION_PORT_ID) &&
2340 ((struct priv *)port_id_dev->data->dev_private)->representor)
2341 return rte_flow_error_set(error, ENOTSUP,
2342 RTE_FLOW_ERROR_TYPE_ACTION, actions,
2343 "vlan push can only be applied"
2344 " when forwarding to uplink port");
2346 * FW syndrome (0x294609):
2347 * set_flow_table_entry: modify/pop/push actions in fdb flow table
2348 * are supported only while forwarding to vport.
2350 if ((action_flags & MLX5_TCF_VLAN_ACTIONS) &&
2351 !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2352 return rte_flow_error_set(error, ENOTSUP,
2353 RTE_FLOW_ERROR_TYPE_ACTION, actions,
2354 "vlan actions are supported"
2355 " only with port_id action");
2356 if ((action_flags & MLX5_TCF_VXLAN_ACTIONS) &&
2357 !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2358 return rte_flow_error_set(error, ENOTSUP,
2359 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2360 "vxlan actions are supported"
2361 " only with port_id action");
2362 if (!(action_flags & MLX5_TCF_FATE_ACTIONS))
2363 return rte_flow_error_set(error, EINVAL,
2364 RTE_FLOW_ERROR_TYPE_ACTION, actions,
2365 "no fate action is found");
2367 (MLX5_FLOW_ACTION_SET_TTL | MLX5_FLOW_ACTION_DEC_TTL)) {
2369 (MLX5_FLOW_LAYER_OUTER_L3_IPV4 |
2370 MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
2371 return rte_flow_error_set(error, EINVAL,
2372 RTE_FLOW_ERROR_TYPE_ACTION,
2374 "no IP found in pattern");
2377 (MLX5_FLOW_ACTION_SET_MAC_SRC | MLX5_FLOW_ACTION_SET_MAC_DST)) {
2378 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L2))
2379 return rte_flow_error_set(error, ENOTSUP,
2380 RTE_FLOW_ERROR_TYPE_ACTION,
2382 "no ethernet found in"
2385 if ((action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) &&
2386 !(item_flags & MLX5_FLOW_LAYER_VXLAN))
2387 return rte_flow_error_set(error, EINVAL,
2388 RTE_FLOW_ERROR_TYPE_ACTION,
2390 "no VNI pattern found"
2391 " for vxlan decap action");
2392 if ((action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP) &&
2393 (item_flags & MLX5_FLOW_LAYER_TUNNEL))
2394 return rte_flow_error_set(error, EINVAL,
2395 RTE_FLOW_ERROR_TYPE_ACTION,
2397 "vxlan encap not supported"
2398 " for tunneled traffic");
2403 * Calculate maximum size of memory for flow items of Linux TC flower.
2406 * Pointer to the flow attributes.
2408 * Pointer to the list of items.
2409 * @param[out] action_flags
2410 * Pointer to the detected actions.
2413 * Maximum size of memory for items.
2416 flow_tcf_get_items_size(const struct rte_flow_attr *attr,
2417 const struct rte_flow_item items[],
2418 uint64_t *action_flags)
2422 size += SZ_NLATTR_STRZ_OF("flower") +
2423 SZ_NLATTR_NEST + /* TCA_OPTIONS. */
2424 SZ_NLATTR_TYPE_OF(uint32_t); /* TCA_CLS_FLAGS_SKIP_SW. */
2425 if (attr->group > 0)
2426 size += SZ_NLATTR_TYPE_OF(uint32_t); /* TCA_CHAIN. */
2427 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2428 switch (items->type) {
2429 case RTE_FLOW_ITEM_TYPE_VOID:
2431 case RTE_FLOW_ITEM_TYPE_PORT_ID:
2433 case RTE_FLOW_ITEM_TYPE_ETH:
2434 size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2435 SZ_NLATTR_DATA_OF(ETHER_ADDR_LEN) * 4;
2436 /* dst/src MAC addr and mask. */
2438 case RTE_FLOW_ITEM_TYPE_VLAN:
2439 size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2440 SZ_NLATTR_TYPE_OF(uint16_t) +
2441 /* VLAN Ether type. */
2442 SZ_NLATTR_TYPE_OF(uint8_t) + /* VLAN prio. */
2443 SZ_NLATTR_TYPE_OF(uint16_t); /* VLAN ID. */
2445 case RTE_FLOW_ITEM_TYPE_IPV4:
2446 size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2447 SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2448 SZ_NLATTR_TYPE_OF(uint32_t) * 4;
2449 /* dst/src IP addr and mask. */
2451 case RTE_FLOW_ITEM_TYPE_IPV6:
2452 size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2453 SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2454 SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN) * 4;
2455 /* dst/src IP addr and mask. */
2457 case RTE_FLOW_ITEM_TYPE_UDP:
2458 size += SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2459 SZ_NLATTR_TYPE_OF(uint16_t) * 4;
2460 /* dst/src port and mask. */
2462 case RTE_FLOW_ITEM_TYPE_TCP:
2463 size += SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2464 SZ_NLATTR_TYPE_OF(uint16_t) * 4;
2465 /* dst/src port and mask. */
2467 case RTE_FLOW_ITEM_TYPE_VXLAN:
2468 size += SZ_NLATTR_TYPE_OF(uint32_t);
2470 * There might be no VXLAN decap action in the action
2471 * list, nonetheless the VXLAN tunnel flow requires
2472 * the decap structure to be correctly applied to
2473 * VXLAN device, set the flag to create the structure.
2474 * Translation routine will not put the decap action
2475 * in tne Netlink message if there is no actual action
2478 *action_flags |= MLX5_FLOW_ACTION_VXLAN_DECAP;
2482 "unsupported item %p type %d,"
2483 " items must be validated before flow creation",
2484 (const void *)items, items->type);
2492 * Calculate size of memory to store the VXLAN encapsultion
2493 * related items in the Netlink message buffer. Items list
2494 * is specified by RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action.
2495 * The item list should be validated.
2498 * RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action object.
2499 * List of pattern items to scan data from.
2502 * The size the part of Netlink message buffer to store the
2503 * VXLAN encapsulation item attributes.
2506 flow_tcf_vxlan_encap_size(const struct rte_flow_action *action)
2508 const struct rte_flow_item *items;
2511 assert(action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP);
2512 assert(action->conf);
2514 items = ((const struct rte_flow_action_vxlan_encap *)
2515 action->conf)->definition;
2517 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2518 switch (items->type) {
2519 case RTE_FLOW_ITEM_TYPE_VOID:
2521 case RTE_FLOW_ITEM_TYPE_ETH:
2522 /* This item does not require message buffer. */
2524 case RTE_FLOW_ITEM_TYPE_IPV4:
2525 size += SZ_NLATTR_DATA_OF(IPV4_ADDR_LEN) * 2;
2527 case RTE_FLOW_ITEM_TYPE_IPV6:
2528 size += SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN) * 2;
2530 case RTE_FLOW_ITEM_TYPE_UDP: {
2531 const struct rte_flow_item_udp *udp = items->mask;
2533 size += SZ_NLATTR_TYPE_OF(uint16_t);
2534 if (!udp || udp->hdr.src_port != RTE_BE16(0x0000))
2535 size += SZ_NLATTR_TYPE_OF(uint16_t);
2538 case RTE_FLOW_ITEM_TYPE_VXLAN:
2539 size += SZ_NLATTR_TYPE_OF(uint32_t);
2544 "unsupported item %p type %d,"
2545 " items must be validated"
2546 " before flow creation",
2547 (const void *)items, items->type);
2555 * Calculate maximum size of memory for flow actions of Linux TC flower and
2556 * extract specified actions.
2558 * @param[in] actions
2559 * Pointer to the list of actions.
2560 * @param[out] action_flags
2561 * Pointer to the detected actions.
2564 * Maximum size of memory for actions.
2567 flow_tcf_get_actions_and_size(const struct rte_flow_action actions[],
2568 uint64_t *action_flags)
2573 size += SZ_NLATTR_NEST; /* TCA_FLOWER_ACT. */
2574 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2575 switch (actions->type) {
2576 case RTE_FLOW_ACTION_TYPE_VOID:
2578 case RTE_FLOW_ACTION_TYPE_PORT_ID:
2579 size += SZ_NLATTR_NEST + /* na_act_index. */
2580 SZ_NLATTR_STRZ_OF("mirred") +
2581 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2582 SZ_NLATTR_TYPE_OF(struct tc_mirred);
2583 flags |= MLX5_FLOW_ACTION_PORT_ID;
2585 case RTE_FLOW_ACTION_TYPE_JUMP:
2586 size += SZ_NLATTR_NEST + /* na_act_index. */
2587 SZ_NLATTR_STRZ_OF("gact") +
2588 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2589 SZ_NLATTR_TYPE_OF(struct tc_gact);
2590 flags |= MLX5_FLOW_ACTION_JUMP;
2592 case RTE_FLOW_ACTION_TYPE_DROP:
2593 size += SZ_NLATTR_NEST + /* na_act_index. */
2594 SZ_NLATTR_STRZ_OF("gact") +
2595 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2596 SZ_NLATTR_TYPE_OF(struct tc_gact);
2597 flags |= MLX5_FLOW_ACTION_DROP;
2599 case RTE_FLOW_ACTION_TYPE_COUNT:
2601 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
2602 flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
2603 goto action_of_vlan;
2604 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
2605 flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
2606 goto action_of_vlan;
2607 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
2608 flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
2609 goto action_of_vlan;
2610 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
2611 flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
2612 goto action_of_vlan;
2614 size += SZ_NLATTR_NEST + /* na_act_index. */
2615 SZ_NLATTR_STRZ_OF("vlan") +
2616 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2617 SZ_NLATTR_TYPE_OF(struct tc_vlan) +
2618 SZ_NLATTR_TYPE_OF(uint16_t) +
2619 /* VLAN protocol. */
2620 SZ_NLATTR_TYPE_OF(uint16_t) + /* VLAN ID. */
2621 SZ_NLATTR_TYPE_OF(uint8_t); /* VLAN prio. */
2623 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2624 size += SZ_NLATTR_NEST + /* na_act_index. */
2625 SZ_NLATTR_STRZ_OF("tunnel_key") +
2626 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2627 SZ_NLATTR_TYPE_OF(uint8_t);
2628 size += SZ_NLATTR_TYPE_OF(struct tc_tunnel_key);
2629 size += flow_tcf_vxlan_encap_size(actions) +
2630 RTE_ALIGN_CEIL /* preceding encap params. */
2631 (sizeof(struct flow_tcf_vxlan_encap),
2633 flags |= MLX5_FLOW_ACTION_VXLAN_ENCAP;
2635 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
2636 size += SZ_NLATTR_NEST + /* na_act_index. */
2637 SZ_NLATTR_STRZ_OF("tunnel_key") +
2638 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2639 SZ_NLATTR_TYPE_OF(uint8_t);
2640 size += SZ_NLATTR_TYPE_OF(struct tc_tunnel_key);
2641 size += RTE_ALIGN_CEIL /* preceding decap params. */
2642 (sizeof(struct flow_tcf_vxlan_decap),
2644 flags |= MLX5_FLOW_ACTION_VXLAN_DECAP;
2646 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
2647 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
2648 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
2649 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
2650 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
2651 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
2652 case RTE_FLOW_ACTION_TYPE_SET_TTL:
2653 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
2654 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
2655 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
2656 size += flow_tcf_get_pedit_actions_size(&actions,
2661 "unsupported action %p type %d,"
2662 " items must be validated before flow creation",
2663 (const void *)actions, actions->type);
2667 *action_flags = flags;
2672 * Brand rtnetlink buffer with unique handle.
2674 * This handle should be unique for a given network interface to avoid
2678 * Pointer to Netlink message.
2680 * Unique 32-bit handle to use.
2683 flow_tcf_nl_brand(struct nlmsghdr *nlh, uint32_t handle)
2685 struct tcmsg *tcm = mnl_nlmsg_get_payload(nlh);
2687 tcm->tcm_handle = handle;
2688 DRV_LOG(DEBUG, "Netlink msg %p is branded with handle %x",
2689 (void *)nlh, handle);
2693 * Prepare a flow object for Linux TC flower. It calculates the maximum size of
2694 * memory required, allocates the memory, initializes Netlink message headers
2695 * and set unique TC message handle.
2698 * Pointer to the flow attributes.
2700 * Pointer to the list of items.
2701 * @param[in] actions
2702 * Pointer to the list of actions.
2704 * Pointer to the error structure.
2707 * Pointer to mlx5_flow object on success,
2708 * otherwise NULL and rte_ernno is set.
2710 static struct mlx5_flow *
2711 flow_tcf_prepare(const struct rte_flow_attr *attr,
2712 const struct rte_flow_item items[],
2713 const struct rte_flow_action actions[],
2714 struct rte_flow_error *error)
2716 size_t size = RTE_ALIGN_CEIL
2717 (sizeof(struct mlx5_flow),
2718 alignof(struct flow_tcf_tunnel_hdr)) +
2719 MNL_ALIGN(sizeof(struct nlmsghdr)) +
2720 MNL_ALIGN(sizeof(struct tcmsg));
2721 struct mlx5_flow *dev_flow;
2722 uint64_t action_flags = 0;
2723 struct nlmsghdr *nlh;
2725 uint8_t *sp, *tun = NULL;
2727 size += flow_tcf_get_items_size(attr, items, &action_flags);
2728 size += flow_tcf_get_actions_and_size(actions, &action_flags);
2729 dev_flow = rte_zmalloc(__func__, size, MNL_ALIGNTO);
2731 rte_flow_error_set(error, ENOMEM,
2732 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2733 "not enough memory to create E-Switch flow");
2736 sp = (uint8_t *)(dev_flow + 1);
2737 if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP) {
2739 (sp, alignof(struct flow_tcf_tunnel_hdr));
2741 sp += RTE_ALIGN_CEIL
2742 (sizeof(struct flow_tcf_vxlan_encap),
2745 size -= RTE_ALIGN_CEIL
2746 (sizeof(struct flow_tcf_vxlan_encap),
2749 } else if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2751 (sp, alignof(struct flow_tcf_tunnel_hdr));
2753 sp += RTE_ALIGN_CEIL
2754 (sizeof(struct flow_tcf_vxlan_decap),
2757 size -= RTE_ALIGN_CEIL
2758 (sizeof(struct flow_tcf_vxlan_decap),
2762 sp = RTE_PTR_ALIGN(sp, MNL_ALIGNTO);
2764 nlh = mnl_nlmsg_put_header(sp);
2765 tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
2766 *dev_flow = (struct mlx5_flow){
2767 .tcf = (struct mlx5_flow_tcf){
2769 .nlsize = size - RTE_ALIGN_CEIL
2770 (sizeof(struct mlx5_flow),
2771 alignof(struct flow_tcf_tunnel_hdr)),
2773 .tunnel = (struct flow_tcf_tunnel_hdr *)tun,
2778 if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP)
2779 dev_flow->tcf.tunnel->type = FLOW_TCF_TUNACT_VXLAN_DECAP;
2780 else if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP)
2781 dev_flow->tcf.tunnel->type = FLOW_TCF_TUNACT_VXLAN_ENCAP;
2783 * Generate a reasonably unique handle based on the address of the
2786 * This is straightforward on 32-bit systems where the flow pointer can
2787 * be used directly. Otherwise, its least significant part is taken
2788 * after shifting it by the previous power of two of the pointed buffer
2791 if (sizeof(dev_flow) <= 4)
2792 flow_tcf_nl_brand(nlh, (uintptr_t)dev_flow);
2794 flow_tcf_nl_brand(nlh, (uintptr_t)dev_flow >>
2795 rte_log2_u32(rte_align32prevpow2(size)));
2800 * Make adjustments for supporting count actions.
2803 * Pointer to the Ethernet device structure.
2804 * @param[in] dev_flow
2805 * Pointer to mlx5_flow.
2807 * Pointer to error structure.
2810 * 0 On success else a negative errno value is returned and rte_errno is set.
2813 flow_tcf_translate_action_count(struct rte_eth_dev *dev __rte_unused,
2814 struct mlx5_flow *dev_flow,
2815 struct rte_flow_error *error)
2817 struct rte_flow *flow = dev_flow->flow;
2819 if (!flow->counter) {
2820 flow->counter = flow_tcf_counter_new();
2822 return rte_flow_error_set(error, rte_errno,
2823 RTE_FLOW_ERROR_TYPE_ACTION,
2825 "cannot get counter"
2832 * Convert VXLAN VNI to 32-bit integer.
2835 * VXLAN VNI in 24-bit wire format.
2838 * VXLAN VNI as a 32-bit integer value in network endian.
2840 static inline rte_be32_t
2841 vxlan_vni_as_be32(const uint8_t vni[3])
2847 .vni = { 0, vni[0], vni[1], vni[2] },
2853 * Helper function to process RTE_FLOW_ITEM_TYPE_ETH entry in configuration
2854 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the MAC address fields
2855 * in the encapsulation parameters structure. The item must be prevalidated,
2856 * no any validation checks performed by function.
2859 * RTE_FLOW_ITEM_TYPE_ETH entry specification.
2861 * RTE_FLOW_ITEM_TYPE_ETH entry mask.
2863 * Structure to fill the gathered MAC address data.
2866 flow_tcf_parse_vxlan_encap_eth(const struct rte_flow_item_eth *spec,
2867 const struct rte_flow_item_eth *mask,
2868 struct flow_tcf_vxlan_encap *encap)
2870 /* Item must be validated before. No redundant checks. */
2872 if (!mask || !memcmp(&mask->dst,
2873 &rte_flow_item_eth_mask.dst,
2874 sizeof(rte_flow_item_eth_mask.dst))) {
2876 * Ethernet addresses are not supported by
2877 * tc as tunnel_key parameters. Destination
2878 * address is needed to form encap packet
2879 * header and retrieved by kernel from
2880 * implicit sources (ARP table, etc),
2881 * address masks are not supported at all.
2883 encap->eth.dst = spec->dst;
2884 encap->mask |= FLOW_TCF_ENCAP_ETH_DST;
2886 if (!mask || !memcmp(&mask->src,
2887 &rte_flow_item_eth_mask.src,
2888 sizeof(rte_flow_item_eth_mask.src))) {
2890 * Ethernet addresses are not supported by
2891 * tc as tunnel_key parameters. Source ethernet
2892 * address is ignored anyway.
2894 encap->eth.src = spec->src;
2895 encap->mask |= FLOW_TCF_ENCAP_ETH_SRC;
2900 * Helper function to process RTE_FLOW_ITEM_TYPE_IPV4 entry in configuration
2901 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the IPV4 address fields
2902 * in the encapsulation parameters structure. The item must be prevalidated,
2903 * no any validation checks performed by function.
2906 * RTE_FLOW_ITEM_TYPE_IPV4 entry specification.
2908 * Structure to fill the gathered IPV4 address data.
2911 flow_tcf_parse_vxlan_encap_ipv4(const struct rte_flow_item_ipv4 *spec,
2912 struct flow_tcf_vxlan_encap *encap)
2914 /* Item must be validated before. No redundant checks. */
2916 encap->ipv4.dst = spec->hdr.dst_addr;
2917 encap->ipv4.src = spec->hdr.src_addr;
2918 encap->mask |= FLOW_TCF_ENCAP_IPV4_SRC |
2919 FLOW_TCF_ENCAP_IPV4_DST;
2923 * Helper function to process RTE_FLOW_ITEM_TYPE_IPV6 entry in configuration
2924 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the IPV6 address fields
2925 * in the encapsulation parameters structure. The item must be prevalidated,
2926 * no any validation checks performed by function.
2929 * RTE_FLOW_ITEM_TYPE_IPV6 entry specification.
2931 * Structure to fill the gathered IPV6 address data.
2934 flow_tcf_parse_vxlan_encap_ipv6(const struct rte_flow_item_ipv6 *spec,
2935 struct flow_tcf_vxlan_encap *encap)
2937 /* Item must be validated before. No redundant checks. */
2939 memcpy(encap->ipv6.dst, spec->hdr.dst_addr, IPV6_ADDR_LEN);
2940 memcpy(encap->ipv6.src, spec->hdr.src_addr, IPV6_ADDR_LEN);
2941 encap->mask |= FLOW_TCF_ENCAP_IPV6_SRC |
2942 FLOW_TCF_ENCAP_IPV6_DST;
2946 * Helper function to process RTE_FLOW_ITEM_TYPE_UDP entry in configuration
2947 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the UDP port fields
2948 * in the encapsulation parameters structure. The item must be prevalidated,
2949 * no any validation checks performed by function.
2952 * RTE_FLOW_ITEM_TYPE_UDP entry specification.
2954 * RTE_FLOW_ITEM_TYPE_UDP entry mask.
2956 * Structure to fill the gathered UDP port data.
2959 flow_tcf_parse_vxlan_encap_udp(const struct rte_flow_item_udp *spec,
2960 const struct rte_flow_item_udp *mask,
2961 struct flow_tcf_vxlan_encap *encap)
2964 encap->udp.dst = spec->hdr.dst_port;
2965 encap->mask |= FLOW_TCF_ENCAP_UDP_DST;
2966 if (!mask || mask->hdr.src_port != RTE_BE16(0x0000)) {
2967 encap->udp.src = spec->hdr.src_port;
2968 encap->mask |= FLOW_TCF_ENCAP_IPV4_SRC;
2973 * Helper function to process RTE_FLOW_ITEM_TYPE_VXLAN entry in configuration
2974 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the VNI fields
2975 * in the encapsulation parameters structure. The item must be prevalidated,
2976 * no any validation checks performed by function.
2979 * RTE_FLOW_ITEM_TYPE_VXLAN entry specification.
2981 * Structure to fill the gathered VNI address data.
2984 flow_tcf_parse_vxlan_encap_vni(const struct rte_flow_item_vxlan *spec,
2985 struct flow_tcf_vxlan_encap *encap)
2987 /* Item must be validated before. Do not redundant checks. */
2989 memcpy(encap->vxlan.vni, spec->vni, sizeof(encap->vxlan.vni));
2990 encap->mask |= FLOW_TCF_ENCAP_VXLAN_VNI;
2994 * Populate consolidated encapsulation object from list of pattern items.
2996 * Helper function to process configuration of action such as
2997 * RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. The item list should be
2998 * validated, there is no way to return an meaningful error.
3001 * RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action object.
3002 * List of pattern items to gather data from.
3004 * Structure to fill gathered data.
3007 flow_tcf_vxlan_encap_parse(const struct rte_flow_action *action,
3008 struct flow_tcf_vxlan_encap *encap)
3011 const struct rte_flow_item_eth *eth;
3012 const struct rte_flow_item_ipv4 *ipv4;
3013 const struct rte_flow_item_ipv6 *ipv6;
3014 const struct rte_flow_item_udp *udp;
3015 const struct rte_flow_item_vxlan *vxlan;
3017 const struct rte_flow_item *items;
3019 assert(action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP);
3020 assert(action->conf);
3022 items = ((const struct rte_flow_action_vxlan_encap *)
3023 action->conf)->definition;
3025 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3026 switch (items->type) {
3027 case RTE_FLOW_ITEM_TYPE_VOID:
3029 case RTE_FLOW_ITEM_TYPE_ETH:
3030 mask.eth = items->mask;
3031 spec.eth = items->spec;
3032 flow_tcf_parse_vxlan_encap_eth(spec.eth, mask.eth,
3035 case RTE_FLOW_ITEM_TYPE_IPV4:
3036 spec.ipv4 = items->spec;
3037 flow_tcf_parse_vxlan_encap_ipv4(spec.ipv4, encap);
3039 case RTE_FLOW_ITEM_TYPE_IPV6:
3040 spec.ipv6 = items->spec;
3041 flow_tcf_parse_vxlan_encap_ipv6(spec.ipv6, encap);
3043 case RTE_FLOW_ITEM_TYPE_UDP:
3044 mask.udp = items->mask;
3045 spec.udp = items->spec;
3046 flow_tcf_parse_vxlan_encap_udp(spec.udp, mask.udp,
3049 case RTE_FLOW_ITEM_TYPE_VXLAN:
3050 spec.vxlan = items->spec;
3051 flow_tcf_parse_vxlan_encap_vni(spec.vxlan, encap);
3056 "unsupported item %p type %d,"
3057 " items must be validated"
3058 " before flow creation",
3059 (const void *)items, items->type);
3067 * Translate flow for Linux TC flower and construct Netlink message.
3070 * Pointer to the priv structure.
3071 * @param[in, out] flow
3072 * Pointer to the sub flow.
3074 * Pointer to the flow attributes.
3076 * Pointer to the list of items.
3077 * @param[in] actions
3078 * Pointer to the list of actions.
3080 * Pointer to the error structure.
3083 * 0 on success, a negative errno value otherwise and rte_ernno is set.
3086 flow_tcf_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3087 const struct rte_flow_attr *attr,
3088 const struct rte_flow_item items[],
3089 const struct rte_flow_action actions[],
3090 struct rte_flow_error *error)
3093 const struct rte_flow_item_port_id *port_id;
3094 const struct rte_flow_item_eth *eth;
3095 const struct rte_flow_item_vlan *vlan;
3096 const struct rte_flow_item_ipv4 *ipv4;
3097 const struct rte_flow_item_ipv6 *ipv6;
3098 const struct rte_flow_item_tcp *tcp;
3099 const struct rte_flow_item_udp *udp;
3100 const struct rte_flow_item_vxlan *vxlan;
3103 const struct rte_flow_action_port_id *port_id;
3104 const struct rte_flow_action_jump *jump;
3105 const struct rte_flow_action_of_push_vlan *of_push_vlan;
3106 const struct rte_flow_action_of_set_vlan_vid *
3108 const struct rte_flow_action_of_set_vlan_pcp *
3112 struct flow_tcf_tunnel_hdr *hdr;
3113 struct flow_tcf_vxlan_decap *vxlan;
3118 struct flow_tcf_tunnel_hdr *hdr;
3119 struct flow_tcf_vxlan_encap *vxlan;
3123 struct flow_tcf_ptoi ptoi[PTOI_TABLE_SZ_MAX(dev)];
3124 struct nlmsghdr *nlh = dev_flow->tcf.nlh;
3125 struct tcmsg *tcm = dev_flow->tcf.tcm;
3126 uint32_t na_act_index_cur;
3127 bool eth_type_set = 0;
3128 bool vlan_present = 0;
3129 bool vlan_eth_type_set = 0;
3130 bool ip_proto_set = 0;
3131 bool tunnel_outer = 0;
3132 struct nlattr *na_flower;
3133 struct nlattr *na_flower_act;
3134 struct nlattr *na_vlan_id = NULL;
3135 struct nlattr *na_vlan_priority = NULL;
3136 uint64_t item_flags = 0;
3139 claim_nonzero(flow_tcf_build_ptoi_table(dev, ptoi,
3140 PTOI_TABLE_SZ_MAX(dev)));
3141 if (dev_flow->tcf.tunnel) {
3142 switch (dev_flow->tcf.tunnel->type) {
3143 case FLOW_TCF_TUNACT_VXLAN_DECAP:
3144 decap.vxlan = dev_flow->tcf.vxlan_decap;
3147 case FLOW_TCF_TUNACT_VXLAN_ENCAP:
3148 encap.vxlan = dev_flow->tcf.vxlan_encap;
3150 /* New tunnel actions can be added here. */
3156 nlh = dev_flow->tcf.nlh;
3157 tcm = dev_flow->tcf.tcm;
3158 /* Prepare API must have been called beforehand. */
3159 assert(nlh != NULL && tcm != NULL);
3160 tcm->tcm_family = AF_UNSPEC;
3161 tcm->tcm_ifindex = ptoi[0].ifindex;
3162 tcm->tcm_parent = TC_H_MAKE(TC_H_INGRESS, TC_H_MIN_INGRESS);
3164 * Priority cannot be zero to prevent the kernel from picking one
3167 tcm->tcm_info = TC_H_MAKE((attr->priority + 1) << 16,
3168 RTE_BE16(ETH_P_ALL));
3169 if (attr->group > 0)
3170 mnl_attr_put_u32(nlh, TCA_CHAIN, attr->group);
3171 mnl_attr_put_strz(nlh, TCA_KIND, "flower");
3172 na_flower = mnl_attr_nest_start(nlh, TCA_OPTIONS);
3173 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3176 switch (items->type) {
3177 case RTE_FLOW_ITEM_TYPE_VOID:
3179 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3180 mask.port_id = flow_tcf_item_mask
3181 (items, &rte_flow_item_port_id_mask,
3182 &flow_tcf_mask_supported.port_id,
3183 &flow_tcf_mask_empty.port_id,
3184 sizeof(flow_tcf_mask_supported.port_id),
3186 assert(mask.port_id);
3187 if (mask.port_id == &flow_tcf_mask_empty.port_id)
3189 spec.port_id = items->spec;
3190 if (!mask.port_id->id)
3193 for (i = 0; ptoi[i].ifindex; ++i)
3194 if (ptoi[i].port_id == spec.port_id->id)
3196 assert(ptoi[i].ifindex);
3197 tcm->tcm_ifindex = ptoi[i].ifindex;
3199 case RTE_FLOW_ITEM_TYPE_ETH:
3200 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
3201 MLX5_FLOW_LAYER_INNER_L2 :
3202 MLX5_FLOW_LAYER_OUTER_L2;
3203 mask.eth = flow_tcf_item_mask
3204 (items, &rte_flow_item_eth_mask,
3205 &flow_tcf_mask_supported.eth,
3206 &flow_tcf_mask_empty.eth,
3207 sizeof(flow_tcf_mask_supported.eth),
3210 if (mask.eth == &flow_tcf_mask_empty.eth)
3212 spec.eth = items->spec;
3215 "outer L2 addresses cannot be"
3216 " forced is outer ones for tunnel,"
3217 " parameter is ignored");
3220 if (mask.eth->type) {
3221 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_ETH_TYPE,
3225 if (!is_zero_ether_addr(&mask.eth->dst)) {
3226 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_DST,
3228 spec.eth->dst.addr_bytes);
3229 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_DST_MASK,
3231 mask.eth->dst.addr_bytes);
3233 if (!is_zero_ether_addr(&mask.eth->src)) {
3234 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_SRC,
3236 spec.eth->src.addr_bytes);
3237 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_SRC_MASK,
3239 mask.eth->src.addr_bytes);
3241 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3243 case RTE_FLOW_ITEM_TYPE_VLAN:
3246 assert(!tunnel_outer);
3247 item_flags |= MLX5_FLOW_LAYER_OUTER_VLAN;
3248 mask.vlan = flow_tcf_item_mask
3249 (items, &rte_flow_item_vlan_mask,
3250 &flow_tcf_mask_supported.vlan,
3251 &flow_tcf_mask_empty.vlan,
3252 sizeof(flow_tcf_mask_supported.vlan),
3256 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_ETH_TYPE,
3257 RTE_BE16(ETH_P_8021Q));
3260 if (mask.vlan == &flow_tcf_mask_empty.vlan)
3262 spec.vlan = items->spec;
3263 if (mask.vlan->inner_type) {
3264 mnl_attr_put_u16(nlh,
3265 TCA_FLOWER_KEY_VLAN_ETH_TYPE,
3266 spec.vlan->inner_type);
3267 vlan_eth_type_set = 1;
3269 if (mask.vlan->tci & RTE_BE16(0xe000))
3270 mnl_attr_put_u8(nlh, TCA_FLOWER_KEY_VLAN_PRIO,
3272 (spec.vlan->tci) >> 13) & 0x7);
3273 if (mask.vlan->tci & RTE_BE16(0x0fff))
3274 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_VLAN_ID,
3278 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3280 case RTE_FLOW_ITEM_TYPE_IPV4:
3281 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
3282 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3283 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3284 mask.ipv4 = flow_tcf_item_mask
3285 (items, &rte_flow_item_ipv4_mask,
3286 &flow_tcf_mask_supported.ipv4,
3287 &flow_tcf_mask_empty.ipv4,
3288 sizeof(flow_tcf_mask_supported.ipv4),
3291 spec.ipv4 = items->spec;
3292 if (!tunnel_outer) {
3293 if (!eth_type_set ||
3294 (!vlan_eth_type_set && vlan_present))
3298 TCA_FLOWER_KEY_VLAN_ETH_TYPE :
3299 TCA_FLOWER_KEY_ETH_TYPE,
3300 RTE_BE16(ETH_P_IP));
3302 vlan_eth_type_set = 1;
3304 if (!tunnel_outer && mask.ipv4->hdr.next_proto_id) {
3306 * No way to set IP protocol for outer tunnel
3307 * layers. Usually it is fixed, for example,
3308 * to UDP for VXLAN/GPE.
3310 assert(spec.ipv4); /* Mask is not empty. */
3311 mnl_attr_put_u8(nlh, TCA_FLOWER_KEY_IP_PROTO,
3312 spec.ipv4->hdr.next_proto_id);
3315 if (mask.ipv4 == &flow_tcf_mask_empty.ipv4 ||
3316 (!mask.ipv4->hdr.src_addr &&
3317 !mask.ipv4->hdr.dst_addr)) {
3321 * For tunnel outer we must set outer IP key
3322 * anyway, even if the specification/mask is
3323 * empty. There is no another way to tell
3324 * kernel about he outer layer protocol.
3327 (nlh, TCA_FLOWER_KEY_ENC_IPV4_SRC,
3328 mask.ipv4->hdr.src_addr);
3330 (nlh, TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK,
3331 mask.ipv4->hdr.src_addr);
3332 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3335 if (mask.ipv4->hdr.src_addr) {
3337 (nlh, tunnel_outer ?
3338 TCA_FLOWER_KEY_ENC_IPV4_SRC :
3339 TCA_FLOWER_KEY_IPV4_SRC,
3340 spec.ipv4->hdr.src_addr);
3342 (nlh, tunnel_outer ?
3343 TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK :
3344 TCA_FLOWER_KEY_IPV4_SRC_MASK,
3345 mask.ipv4->hdr.src_addr);
3347 if (mask.ipv4->hdr.dst_addr) {
3349 (nlh, tunnel_outer ?
3350 TCA_FLOWER_KEY_ENC_IPV4_DST :
3351 TCA_FLOWER_KEY_IPV4_DST,
3352 spec.ipv4->hdr.dst_addr);
3354 (nlh, tunnel_outer ?
3355 TCA_FLOWER_KEY_ENC_IPV4_DST_MASK :
3356 TCA_FLOWER_KEY_IPV4_DST_MASK,
3357 mask.ipv4->hdr.dst_addr);
3359 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3361 case RTE_FLOW_ITEM_TYPE_IPV6: {
3362 bool ipv6_src, ipv6_dst;
3364 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
3365 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3366 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3367 mask.ipv6 = flow_tcf_item_mask
3368 (items, &rte_flow_item_ipv6_mask,
3369 &flow_tcf_mask_supported.ipv6,
3370 &flow_tcf_mask_empty.ipv6,
3371 sizeof(flow_tcf_mask_supported.ipv6),
3374 spec.ipv6 = items->spec;
3375 if (!tunnel_outer) {
3376 if (!eth_type_set ||
3377 (!vlan_eth_type_set && vlan_present))
3381 TCA_FLOWER_KEY_VLAN_ETH_TYPE :
3382 TCA_FLOWER_KEY_ETH_TYPE,
3383 RTE_BE16(ETH_P_IPV6));
3385 vlan_eth_type_set = 1;
3387 if (!tunnel_outer && mask.ipv6->hdr.proto) {
3389 * No way to set IP protocol for outer tunnel
3390 * layers. Usually it is fixed, for example,
3391 * to UDP for VXLAN/GPE.
3393 assert(spec.ipv6); /* Mask is not empty. */
3394 mnl_attr_put_u8(nlh, TCA_FLOWER_KEY_IP_PROTO,
3395 spec.ipv6->hdr.proto);
3398 ipv6_dst = !IN6_IS_ADDR_UNSPECIFIED
3399 (mask.ipv6->hdr.dst_addr);
3400 ipv6_src = !IN6_IS_ADDR_UNSPECIFIED
3401 (mask.ipv6->hdr.src_addr);
3402 if (mask.ipv6 == &flow_tcf_mask_empty.ipv6 ||
3403 (!ipv6_dst && !ipv6_src)) {
3407 * For tunnel outer we must set outer IP key
3408 * anyway, even if the specification/mask is
3409 * empty. There is no another way to tell
3410 * kernel about he outer layer protocol.
3413 TCA_FLOWER_KEY_ENC_IPV6_SRC,
3415 mask.ipv6->hdr.src_addr);
3417 TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK,
3419 mask.ipv6->hdr.src_addr);
3420 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3424 mnl_attr_put(nlh, tunnel_outer ?
3425 TCA_FLOWER_KEY_ENC_IPV6_SRC :
3426 TCA_FLOWER_KEY_IPV6_SRC,
3428 spec.ipv6->hdr.src_addr);
3429 mnl_attr_put(nlh, tunnel_outer ?
3430 TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK :
3431 TCA_FLOWER_KEY_IPV6_SRC_MASK,
3433 mask.ipv6->hdr.src_addr);
3436 mnl_attr_put(nlh, tunnel_outer ?
3437 TCA_FLOWER_KEY_ENC_IPV6_DST :
3438 TCA_FLOWER_KEY_IPV6_DST,
3440 spec.ipv6->hdr.dst_addr);
3441 mnl_attr_put(nlh, tunnel_outer ?
3442 TCA_FLOWER_KEY_ENC_IPV6_DST_MASK :
3443 TCA_FLOWER_KEY_IPV6_DST_MASK,
3445 mask.ipv6->hdr.dst_addr);
3447 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3450 case RTE_FLOW_ITEM_TYPE_UDP:
3451 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
3452 MLX5_FLOW_LAYER_INNER_L4_UDP :
3453 MLX5_FLOW_LAYER_OUTER_L4_UDP;
3454 mask.udp = flow_tcf_item_mask
3455 (items, &rte_flow_item_udp_mask,
3456 &flow_tcf_mask_supported.udp,
3457 &flow_tcf_mask_empty.udp,
3458 sizeof(flow_tcf_mask_supported.udp),
3461 spec.udp = items->spec;
3462 if (!tunnel_outer) {
3465 (nlh, TCA_FLOWER_KEY_IP_PROTO,
3467 if (mask.udp == &flow_tcf_mask_empty.udp)
3470 assert(mask.udp != &flow_tcf_mask_empty.udp);
3471 decap.vxlan->udp_port =
3473 (spec.udp->hdr.dst_port);
3475 if (mask.udp->hdr.src_port) {
3477 (nlh, tunnel_outer ?
3478 TCA_FLOWER_KEY_ENC_UDP_SRC_PORT :
3479 TCA_FLOWER_KEY_UDP_SRC,
3480 spec.udp->hdr.src_port);
3482 (nlh, tunnel_outer ?
3483 TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK :
3484 TCA_FLOWER_KEY_UDP_SRC_MASK,
3485 mask.udp->hdr.src_port);
3487 if (mask.udp->hdr.dst_port) {
3489 (nlh, tunnel_outer ?
3490 TCA_FLOWER_KEY_ENC_UDP_DST_PORT :
3491 TCA_FLOWER_KEY_UDP_DST,
3492 spec.udp->hdr.dst_port);
3494 (nlh, tunnel_outer ?
3495 TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK :
3496 TCA_FLOWER_KEY_UDP_DST_MASK,
3497 mask.udp->hdr.dst_port);
3499 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3501 case RTE_FLOW_ITEM_TYPE_TCP:
3502 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
3503 MLX5_FLOW_LAYER_INNER_L4_TCP :
3504 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3505 mask.tcp = flow_tcf_item_mask
3506 (items, &rte_flow_item_tcp_mask,
3507 &flow_tcf_mask_supported.tcp,
3508 &flow_tcf_mask_empty.tcp,
3509 sizeof(flow_tcf_mask_supported.tcp),
3513 mnl_attr_put_u8(nlh, TCA_FLOWER_KEY_IP_PROTO,
3515 if (mask.tcp == &flow_tcf_mask_empty.tcp)
3517 spec.tcp = items->spec;
3518 if (mask.tcp->hdr.src_port) {
3519 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_TCP_SRC,
3520 spec.tcp->hdr.src_port);
3521 mnl_attr_put_u16(nlh,
3522 TCA_FLOWER_KEY_TCP_SRC_MASK,
3523 mask.tcp->hdr.src_port);
3525 if (mask.tcp->hdr.dst_port) {
3526 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_TCP_DST,
3527 spec.tcp->hdr.dst_port);
3528 mnl_attr_put_u16(nlh,
3529 TCA_FLOWER_KEY_TCP_DST_MASK,
3530 mask.tcp->hdr.dst_port);
3532 if (mask.tcp->hdr.tcp_flags) {
3535 TCA_FLOWER_KEY_TCP_FLAGS,
3537 (spec.tcp->hdr.tcp_flags));
3540 TCA_FLOWER_KEY_TCP_FLAGS_MASK,
3542 (mask.tcp->hdr.tcp_flags));
3544 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3546 case RTE_FLOW_ITEM_TYPE_VXLAN:
3547 assert(decap.vxlan);
3549 item_flags |= MLX5_FLOW_LAYER_VXLAN;
3550 spec.vxlan = items->spec;
3551 mnl_attr_put_u32(nlh,
3552 TCA_FLOWER_KEY_ENC_KEY_ID,
3553 vxlan_vni_as_be32(spec.vxlan->vni));
3554 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3557 return rte_flow_error_set(error, ENOTSUP,
3558 RTE_FLOW_ERROR_TYPE_ITEM,
3559 NULL, "item not supported");
3562 na_flower_act = mnl_attr_nest_start(nlh, TCA_FLOWER_ACT);
3563 na_act_index_cur = 1;
3564 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3565 struct nlattr *na_act_index;
3566 struct nlattr *na_act;
3567 unsigned int vlan_act;
3570 switch (actions->type) {
3571 case RTE_FLOW_ACTION_TYPE_VOID:
3573 case RTE_FLOW_ACTION_TYPE_PORT_ID:
3574 conf.port_id = actions->conf;
3575 if (conf.port_id->original)
3578 for (i = 0; ptoi[i].ifindex; ++i)
3579 if (ptoi[i].port_id == conf.port_id->id)
3581 assert(ptoi[i].ifindex);
3583 mnl_attr_nest_start(nlh, na_act_index_cur++);
3584 assert(na_act_index);
3585 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "mirred");
3586 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3589 assert(dev_flow->tcf.tunnel);
3590 dev_flow->tcf.tunnel->ifindex_ptr =
3591 &((struct tc_mirred *)
3592 mnl_attr_get_payload
3593 (mnl_nlmsg_get_payload_tail
3596 mnl_attr_put(nlh, TCA_MIRRED_PARMS,
3597 sizeof(struct tc_mirred),
3598 &(struct tc_mirred){
3599 .action = TC_ACT_STOLEN,
3600 .eaction = TCA_EGRESS_REDIR,
3601 .ifindex = ptoi[i].ifindex,
3603 mnl_attr_nest_end(nlh, na_act);
3604 mnl_attr_nest_end(nlh, na_act_index);
3606 case RTE_FLOW_ACTION_TYPE_JUMP:
3607 conf.jump = actions->conf;
3609 mnl_attr_nest_start(nlh, na_act_index_cur++);
3610 assert(na_act_index);
3611 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "gact");
3612 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3614 mnl_attr_put(nlh, TCA_GACT_PARMS,
3615 sizeof(struct tc_gact),
3617 .action = TC_ACT_GOTO_CHAIN |
3620 mnl_attr_nest_end(nlh, na_act);
3621 mnl_attr_nest_end(nlh, na_act_index);
3623 case RTE_FLOW_ACTION_TYPE_DROP:
3625 mnl_attr_nest_start(nlh, na_act_index_cur++);
3626 assert(na_act_index);
3627 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "gact");
3628 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3630 mnl_attr_put(nlh, TCA_GACT_PARMS,
3631 sizeof(struct tc_gact),
3633 .action = TC_ACT_SHOT,
3635 mnl_attr_nest_end(nlh, na_act);
3636 mnl_attr_nest_end(nlh, na_act_index);
3638 case RTE_FLOW_ACTION_TYPE_COUNT:
3640 * Driver adds the count action implicitly for
3641 * each rule it creates.
3643 ret = flow_tcf_translate_action_count(dev,
3648 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3649 conf.of_push_vlan = NULL;
3650 vlan_act = TCA_VLAN_ACT_POP;
3651 goto action_of_vlan;
3652 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3653 conf.of_push_vlan = actions->conf;
3654 vlan_act = TCA_VLAN_ACT_PUSH;
3655 goto action_of_vlan;
3656 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3657 conf.of_set_vlan_vid = actions->conf;
3659 goto override_na_vlan_id;
3660 vlan_act = TCA_VLAN_ACT_MODIFY;
3661 goto action_of_vlan;
3662 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3663 conf.of_set_vlan_pcp = actions->conf;
3664 if (na_vlan_priority)
3665 goto override_na_vlan_priority;
3666 vlan_act = TCA_VLAN_ACT_MODIFY;
3667 goto action_of_vlan;
3670 mnl_attr_nest_start(nlh, na_act_index_cur++);
3671 assert(na_act_index);
3672 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "vlan");
3673 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3675 mnl_attr_put(nlh, TCA_VLAN_PARMS,
3676 sizeof(struct tc_vlan),
3678 .action = TC_ACT_PIPE,
3679 .v_action = vlan_act,
3681 if (vlan_act == TCA_VLAN_ACT_POP) {
3682 mnl_attr_nest_end(nlh, na_act);
3683 mnl_attr_nest_end(nlh, na_act_index);
3686 if (vlan_act == TCA_VLAN_ACT_PUSH)
3687 mnl_attr_put_u16(nlh,
3688 TCA_VLAN_PUSH_VLAN_PROTOCOL,
3689 conf.of_push_vlan->ethertype);
3690 na_vlan_id = mnl_nlmsg_get_payload_tail(nlh);
3691 mnl_attr_put_u16(nlh, TCA_VLAN_PAD, 0);
3692 na_vlan_priority = mnl_nlmsg_get_payload_tail(nlh);
3693 mnl_attr_put_u8(nlh, TCA_VLAN_PAD, 0);
3694 mnl_attr_nest_end(nlh, na_act);
3695 mnl_attr_nest_end(nlh, na_act_index);
3696 if (actions->type ==
3697 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
3698 override_na_vlan_id:
3699 na_vlan_id->nla_type = TCA_VLAN_PUSH_VLAN_ID;
3700 *(uint16_t *)mnl_attr_get_payload(na_vlan_id) =
3702 (conf.of_set_vlan_vid->vlan_vid);
3703 } else if (actions->type ==
3704 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
3705 override_na_vlan_priority:
3706 na_vlan_priority->nla_type =
3707 TCA_VLAN_PUSH_VLAN_PRIORITY;
3708 *(uint8_t *)mnl_attr_get_payload
3709 (na_vlan_priority) =
3710 conf.of_set_vlan_pcp->vlan_pcp;
3713 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3714 assert(decap.vxlan);
3715 assert(dev_flow->tcf.tunnel);
3716 dev_flow->tcf.tunnel->ifindex_ptr =
3717 (unsigned int *)&tcm->tcm_ifindex;
3719 mnl_attr_nest_start(nlh, na_act_index_cur++);
3720 assert(na_act_index);
3721 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "tunnel_key");
3722 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3724 mnl_attr_put(nlh, TCA_TUNNEL_KEY_PARMS,
3725 sizeof(struct tc_tunnel_key),
3726 &(struct tc_tunnel_key){
3727 .action = TC_ACT_PIPE,
3728 .t_action = TCA_TUNNEL_KEY_ACT_RELEASE,
3730 mnl_attr_nest_end(nlh, na_act);
3731 mnl_attr_nest_end(nlh, na_act_index);
3732 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3734 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3735 assert(encap.vxlan);
3736 flow_tcf_vxlan_encap_parse(actions, encap.vxlan);
3738 mnl_attr_nest_start(nlh, na_act_index_cur++);
3739 assert(na_act_index);
3740 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "tunnel_key");
3741 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3743 mnl_attr_put(nlh, TCA_TUNNEL_KEY_PARMS,
3744 sizeof(struct tc_tunnel_key),
3745 &(struct tc_tunnel_key){
3746 .action = TC_ACT_PIPE,
3747 .t_action = TCA_TUNNEL_KEY_ACT_SET,
3749 if (encap.vxlan->mask & FLOW_TCF_ENCAP_UDP_DST)
3750 mnl_attr_put_u16(nlh,
3751 TCA_TUNNEL_KEY_ENC_DST_PORT,
3752 encap.vxlan->udp.dst);
3753 if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV4_SRC)
3754 mnl_attr_put_u32(nlh,
3755 TCA_TUNNEL_KEY_ENC_IPV4_SRC,
3756 encap.vxlan->ipv4.src);
3757 if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV4_DST)
3758 mnl_attr_put_u32(nlh,
3759 TCA_TUNNEL_KEY_ENC_IPV4_DST,
3760 encap.vxlan->ipv4.dst);
3761 if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV6_SRC)
3763 TCA_TUNNEL_KEY_ENC_IPV6_SRC,
3764 sizeof(encap.vxlan->ipv6.src),
3765 &encap.vxlan->ipv6.src);
3766 if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV6_DST)
3768 TCA_TUNNEL_KEY_ENC_IPV6_DST,
3769 sizeof(encap.vxlan->ipv6.dst),
3770 &encap.vxlan->ipv6.dst);
3771 if (encap.vxlan->mask & FLOW_TCF_ENCAP_VXLAN_VNI)
3772 mnl_attr_put_u32(nlh,
3773 TCA_TUNNEL_KEY_ENC_KEY_ID,
3775 (encap.vxlan->vxlan.vni));
3776 mnl_attr_put_u8(nlh, TCA_TUNNEL_KEY_NO_CSUM, 0);
3777 mnl_attr_nest_end(nlh, na_act);
3778 mnl_attr_nest_end(nlh, na_act_index);
3779 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3781 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3782 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3783 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3784 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3785 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3786 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3787 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3788 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3789 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3790 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3792 mnl_attr_nest_start(nlh, na_act_index_cur++);
3793 flow_tcf_create_pedit_mnl_msg(nlh,
3794 &actions, item_flags);
3795 mnl_attr_nest_end(nlh, na_act_index);
3798 return rte_flow_error_set(error, ENOTSUP,
3799 RTE_FLOW_ERROR_TYPE_ACTION,
3801 "action not supported");
3805 assert(na_flower_act);
3806 mnl_attr_nest_end(nlh, na_flower_act);
3807 dev_flow->tcf.ptc_flags = mnl_attr_get_payload
3808 (mnl_nlmsg_get_payload_tail(nlh));
3809 mnl_attr_put_u32(nlh, TCA_FLOWER_FLAGS, decap.vxlan ?
3810 0 : TCA_CLS_FLAGS_SKIP_SW);
3811 mnl_attr_nest_end(nlh, na_flower);
3812 if (dev_flow->tcf.tunnel && dev_flow->tcf.tunnel->ifindex_ptr)
3813 dev_flow->tcf.tunnel->ifindex_org =
3814 *dev_flow->tcf.tunnel->ifindex_ptr;
3815 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3820 * Send Netlink message with acknowledgment.
3823 * Flow context to use.
3825 * Message to send. This function always raises the NLM_F_ACK flag before
3828 * Callback handler for received message.
3830 * Context pointer for callback handler.
3833 * 0 on success, a negative errno value otherwise and rte_errno is set.
3836 flow_tcf_nl_ack(struct mlx5_flow_tcf_context *tcf,
3837 struct nlmsghdr *nlh,
3838 mnl_cb_t cb, void *arg)
3840 unsigned int portid = mnl_socket_get_portid(tcf->nl);
3841 uint32_t seq = tcf->seq++;
3847 /* seq 0 is reserved for kernel event-driven notifications. */
3850 nlh->nlmsg_seq = seq;
3851 nlh->nlmsg_flags |= NLM_F_ACK;
3852 ret = mnl_socket_sendto(tcf->nl, nlh, nlh->nlmsg_len);
3854 /* Message send error occurres. */
3858 nlh = (struct nlmsghdr *)(tcf->buf);
3860 * The following loop postpones non-fatal errors until multipart
3861 * messages are complete.
3864 ret = mnl_socket_recvfrom(tcf->nl, tcf->buf, tcf->buf_size);
3868 * In case of overflow Will receive till
3869 * end of multipart message. We may lost part
3870 * of reply messages but mark and return an error.
3872 if (err != ENOSPC ||
3873 !(nlh->nlmsg_flags & NLM_F_MULTI) ||
3874 nlh->nlmsg_type == NLMSG_DONE)
3877 ret = mnl_cb_run(nlh, ret, seq, portid, cb, arg);
3880 * libmnl returns 0 if DONE or
3881 * success ACK message found.
3887 * ACK message with error found
3888 * or some error occurred.
3893 /* We should continue receiving. */
3902 #define MNL_BUF_EXTRA_SPACE 16
3903 #define MNL_REQUEST_SIZE_MIN 256
3904 #define MNL_REQUEST_SIZE_MAX 2048
3905 #define MNL_REQUEST_SIZE RTE_MIN(RTE_MAX(sysconf(_SC_PAGESIZE), \
3906 MNL_REQUEST_SIZE_MIN), MNL_REQUEST_SIZE_MAX)
3908 /* Data structures used by flow_tcf_xxx_cb() routines. */
3909 struct tcf_nlcb_buf {
3910 LIST_ENTRY(tcf_nlcb_buf) next;
3912 alignas(struct nlmsghdr)
3913 uint8_t msg[]; /**< Netlink message data. */
3916 struct tcf_nlcb_context {
3917 unsigned int ifindex; /**< Base interface index. */
3919 LIST_HEAD(, tcf_nlcb_buf) nlbuf;
3923 * Allocate space for netlink command in buffer list
3925 * @param[in, out] ctx
3926 * Pointer to callback context with command buffers list.
3928 * Required size of data buffer to be allocated.
3931 * Pointer to allocated memory, aligned as message header.
3932 * NULL if some error occurred.
3934 static struct nlmsghdr *
3935 flow_tcf_alloc_nlcmd(struct tcf_nlcb_context *ctx, uint32_t size)
3937 struct tcf_nlcb_buf *buf;
3938 struct nlmsghdr *nlh;
3940 size = NLMSG_ALIGN(size);
3941 buf = LIST_FIRST(&ctx->nlbuf);
3942 if (buf && (buf->size + size) <= ctx->bufsize) {
3943 nlh = (struct nlmsghdr *)&buf->msg[buf->size];
3947 if (size > ctx->bufsize) {
3948 DRV_LOG(WARNING, "netlink: too long command buffer requested");
3951 buf = rte_malloc(__func__,
3952 ctx->bufsize + sizeof(struct tcf_nlcb_buf),
3953 alignof(struct tcf_nlcb_buf));
3955 DRV_LOG(WARNING, "netlink: no memory for command buffer");
3958 LIST_INSERT_HEAD(&ctx->nlbuf, buf, next);
3960 nlh = (struct nlmsghdr *)&buf->msg[0];
3965 * Send the buffers with prepared netlink commands. Scans the list and
3966 * sends all found buffers. Buffers are sent and freed anyway in order
3967 * to prevent memory leakage if some every message in received packet.
3970 * Context object initialized by mlx5_flow_tcf_context_create().
3971 * @param[in, out] ctx
3972 * Pointer to callback context with command buffers list.
3975 * Zero value on success, negative errno value otherwise
3976 * and rte_errno is set.
3979 flow_tcf_send_nlcmd(struct mlx5_flow_tcf_context *tcf,
3980 struct tcf_nlcb_context *ctx)
3982 struct tcf_nlcb_buf *bc = LIST_FIRST(&ctx->nlbuf);
3986 struct tcf_nlcb_buf *bn = LIST_NEXT(bc, next);
3987 struct nlmsghdr *nlh;
3991 while (msg < bc->size) {
3993 * Send Netlink commands from buffer in one by one
3994 * fashion. If we send multiple rule deletion commands
3995 * in one Netlink message and some error occurs it may
3996 * cause multiple ACK error messages and break sequence
3997 * numbers of Netlink communication, because we expect
3998 * the only one ACK reply.
4000 assert((bc->size - msg) >= sizeof(struct nlmsghdr));
4001 nlh = (struct nlmsghdr *)&bc->msg[msg];
4002 assert((bc->size - msg) >= nlh->nlmsg_len);
4003 msg += nlh->nlmsg_len;
4004 rc = flow_tcf_nl_ack(tcf, nlh, NULL, NULL);
4007 "netlink: cleanup error %d", rc);
4015 LIST_INIT(&ctx->nlbuf);
4020 * Collect local IP address rules with scope link attribute on specified
4021 * network device. This is callback routine called by libmnl mnl_cb_run()
4022 * in loop for every message in received packet.
4025 * Pointer to reply header.
4026 * @param[in, out] arg
4027 * Opaque data pointer for this callback.
4030 * A positive, nonzero value on success, negative errno value otherwise
4031 * and rte_errno is set.
4034 flow_tcf_collect_local_cb(const struct nlmsghdr *nlh, void *arg)
4036 struct tcf_nlcb_context *ctx = arg;
4037 struct nlmsghdr *cmd;
4038 struct ifaddrmsg *ifa;
4040 struct nlattr *na_local = NULL;
4041 struct nlattr *na_peer = NULL;
4042 unsigned char family;
4045 if (nlh->nlmsg_type != RTM_NEWADDR) {
4049 ifa = mnl_nlmsg_get_payload(nlh);
4050 family = ifa->ifa_family;
4051 if (ifa->ifa_index != ctx->ifindex ||
4052 ifa->ifa_scope != RT_SCOPE_LINK ||
4053 !(ifa->ifa_flags & IFA_F_PERMANENT) ||
4054 (family != AF_INET && family != AF_INET6))
4056 mnl_attr_for_each(na, nlh, sizeof(*ifa)) {
4057 switch (mnl_attr_get_type(na)) {
4065 if (na_local && na_peer)
4068 if (!na_local || !na_peer)
4070 /* Local rule found with scope link, permanent and assigned peer. */
4071 size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
4072 MNL_ALIGN(sizeof(struct ifaddrmsg)) +
4073 (family == AF_INET6 ? 2 * SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN)
4074 : 2 * SZ_NLATTR_TYPE_OF(uint32_t));
4075 cmd = flow_tcf_alloc_nlcmd(ctx, size);
4080 cmd = mnl_nlmsg_put_header(cmd);
4081 cmd->nlmsg_type = RTM_DELADDR;
4082 cmd->nlmsg_flags = NLM_F_REQUEST;
4083 ifa = mnl_nlmsg_put_extra_header(cmd, sizeof(*ifa));
4084 ifa->ifa_flags = IFA_F_PERMANENT;
4085 ifa->ifa_scope = RT_SCOPE_LINK;
4086 ifa->ifa_index = ctx->ifindex;
4087 if (family == AF_INET) {
4088 ifa->ifa_family = AF_INET;
4089 ifa->ifa_prefixlen = 32;
4090 mnl_attr_put_u32(cmd, IFA_LOCAL, mnl_attr_get_u32(na_local));
4091 mnl_attr_put_u32(cmd, IFA_ADDRESS, mnl_attr_get_u32(na_peer));
4093 ifa->ifa_family = AF_INET6;
4094 ifa->ifa_prefixlen = 128;
4095 mnl_attr_put(cmd, IFA_LOCAL, IPV6_ADDR_LEN,
4096 mnl_attr_get_payload(na_local));
4097 mnl_attr_put(cmd, IFA_ADDRESS, IPV6_ADDR_LEN,
4098 mnl_attr_get_payload(na_peer));
4100 assert(size == cmd->nlmsg_len);
4105 * Cleanup the local IP addresses on outer interface.
4108 * Context object initialized by mlx5_flow_tcf_context_create().
4109 * @param[in] ifindex
4110 * Network inferface index to perform cleanup.
4113 flow_tcf_encap_local_cleanup(struct mlx5_flow_tcf_context *tcf,
4114 unsigned int ifindex)
4116 struct nlmsghdr *nlh;
4117 struct ifaddrmsg *ifa;
4118 struct tcf_nlcb_context ctx = {
4120 .bufsize = MNL_REQUEST_SIZE,
4121 .nlbuf = LIST_HEAD_INITIALIZER(),
4127 * Seek and destroy leftovers of local IP addresses with
4128 * matching properties "scope link".
4130 nlh = mnl_nlmsg_put_header(tcf->buf);
4131 nlh->nlmsg_type = RTM_GETADDR;
4132 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4133 ifa = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifa));
4134 ifa->ifa_family = AF_UNSPEC;
4135 ifa->ifa_index = ifindex;
4136 ifa->ifa_scope = RT_SCOPE_LINK;
4137 ret = flow_tcf_nl_ack(tcf, nlh, flow_tcf_collect_local_cb, &ctx);
4139 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4140 ret = flow_tcf_send_nlcmd(tcf, &ctx);
4142 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4146 * Collect neigh permament rules on specified network device.
4147 * This is callback routine called by libmnl mnl_cb_run() in loop for
4148 * every message in received packet.
4151 * Pointer to reply header.
4152 * @param[in, out] arg
4153 * Opaque data pointer for this callback.
4156 * A positive, nonzero value on success, negative errno value otherwise
4157 * and rte_errno is set.
4160 flow_tcf_collect_neigh_cb(const struct nlmsghdr *nlh, void *arg)
4162 struct tcf_nlcb_context *ctx = arg;
4163 struct nlmsghdr *cmd;
4166 struct nlattr *na_ip = NULL;
4167 struct nlattr *na_mac = NULL;
4168 unsigned char family;
4171 if (nlh->nlmsg_type != RTM_NEWNEIGH) {
4175 ndm = mnl_nlmsg_get_payload(nlh);
4176 family = ndm->ndm_family;
4177 if (ndm->ndm_ifindex != (int)ctx->ifindex ||
4178 !(ndm->ndm_state & NUD_PERMANENT) ||
4179 (family != AF_INET && family != AF_INET6))
4181 mnl_attr_for_each(na, nlh, sizeof(*ndm)) {
4182 switch (mnl_attr_get_type(na)) {
4190 if (na_mac && na_ip)
4193 if (!na_mac || !na_ip)
4195 /* Neigh rule with permenent attribute found. */
4196 size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
4197 MNL_ALIGN(sizeof(struct ndmsg)) +
4198 SZ_NLATTR_DATA_OF(ETHER_ADDR_LEN) +
4199 (family == AF_INET6 ? SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN)
4200 : SZ_NLATTR_TYPE_OF(uint32_t));
4201 cmd = flow_tcf_alloc_nlcmd(ctx, size);
4206 cmd = mnl_nlmsg_put_header(cmd);
4207 cmd->nlmsg_type = RTM_DELNEIGH;
4208 cmd->nlmsg_flags = NLM_F_REQUEST;
4209 ndm = mnl_nlmsg_put_extra_header(cmd, sizeof(*ndm));
4210 ndm->ndm_ifindex = ctx->ifindex;
4211 ndm->ndm_state = NUD_PERMANENT;
4214 if (family == AF_INET) {
4215 ndm->ndm_family = AF_INET;
4216 mnl_attr_put_u32(cmd, NDA_DST, mnl_attr_get_u32(na_ip));
4218 ndm->ndm_family = AF_INET6;
4219 mnl_attr_put(cmd, NDA_DST, IPV6_ADDR_LEN,
4220 mnl_attr_get_payload(na_ip));
4222 mnl_attr_put(cmd, NDA_LLADDR, ETHER_ADDR_LEN,
4223 mnl_attr_get_payload(na_mac));
4224 assert(size == cmd->nlmsg_len);
4229 * Cleanup the neigh rules on outer interface.
4232 * Context object initialized by mlx5_flow_tcf_context_create().
4233 * @param[in] ifindex
4234 * Network inferface index to perform cleanup.
4237 flow_tcf_encap_neigh_cleanup(struct mlx5_flow_tcf_context *tcf,
4238 unsigned int ifindex)
4240 struct nlmsghdr *nlh;
4242 struct tcf_nlcb_context ctx = {
4244 .bufsize = MNL_REQUEST_SIZE,
4245 .nlbuf = LIST_HEAD_INITIALIZER(),
4250 /* Seek and destroy leftovers of neigh rules. */
4251 nlh = mnl_nlmsg_put_header(tcf->buf);
4252 nlh->nlmsg_type = RTM_GETNEIGH;
4253 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4254 ndm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ndm));
4255 ndm->ndm_family = AF_UNSPEC;
4256 ndm->ndm_ifindex = ifindex;
4257 ndm->ndm_state = NUD_PERMANENT;
4258 ret = flow_tcf_nl_ack(tcf, nlh, flow_tcf_collect_neigh_cb, &ctx);
4260 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4261 ret = flow_tcf_send_nlcmd(tcf, &ctx);
4263 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4267 * Collect indices of VXLAN encap/decap interfaces associated with device.
4268 * This is callback routine called by libmnl mnl_cb_run() in loop for
4269 * every message in received packet.
4272 * Pointer to reply header.
4273 * @param[in, out] arg
4274 * Opaque data pointer for this callback.
4277 * A positive, nonzero value on success, negative errno value otherwise
4278 * and rte_errno is set.
4281 flow_tcf_collect_vxlan_cb(const struct nlmsghdr *nlh, void *arg)
4283 struct tcf_nlcb_context *ctx = arg;
4284 struct nlmsghdr *cmd;
4285 struct ifinfomsg *ifm;
4287 struct nlattr *na_info = NULL;
4288 struct nlattr *na_vxlan = NULL;
4290 unsigned int vxindex;
4293 if (nlh->nlmsg_type != RTM_NEWLINK) {
4297 ifm = mnl_nlmsg_get_payload(nlh);
4298 if (!ifm->ifi_index) {
4302 mnl_attr_for_each(na, nlh, sizeof(*ifm))
4303 if (mnl_attr_get_type(na) == IFLA_LINKINFO) {
4309 mnl_attr_for_each_nested(na, na_info) {
4310 switch (mnl_attr_get_type(na)) {
4311 case IFLA_INFO_KIND:
4312 if (!strncmp("vxlan", mnl_attr_get_str(na),
4313 mnl_attr_get_len(na)))
4316 case IFLA_INFO_DATA:
4320 if (found && na_vxlan)
4323 if (!found || !na_vxlan)
4326 mnl_attr_for_each_nested(na, na_vxlan) {
4327 if (mnl_attr_get_type(na) == IFLA_VXLAN_LINK &&
4328 mnl_attr_get_u32(na) == ctx->ifindex) {
4335 /* Attached VXLAN device found, store the command to delete. */
4336 vxindex = ifm->ifi_index;
4337 size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
4338 MNL_ALIGN(sizeof(struct ifinfomsg));
4339 cmd = flow_tcf_alloc_nlcmd(ctx, size);
4344 cmd = mnl_nlmsg_put_header(cmd);
4345 cmd->nlmsg_type = RTM_DELLINK;
4346 cmd->nlmsg_flags = NLM_F_REQUEST;
4347 ifm = mnl_nlmsg_put_extra_header(cmd, sizeof(*ifm));
4348 ifm->ifi_family = AF_UNSPEC;
4349 ifm->ifi_index = vxindex;
4350 assert(size == cmd->nlmsg_len);
4355 * Cleanup the outer interface. Removes all found vxlan devices
4356 * attached to specified index, flushes the meigh and local IP
4360 * Context object initialized by mlx5_flow_tcf_context_create().
4361 * @param[in] ifindex
4362 * Network inferface index to perform cleanup.
4365 flow_tcf_encap_iface_cleanup(struct mlx5_flow_tcf_context *tcf,
4366 unsigned int ifindex)
4368 struct nlmsghdr *nlh;
4369 struct ifinfomsg *ifm;
4370 struct tcf_nlcb_context ctx = {
4372 .bufsize = MNL_REQUEST_SIZE,
4373 .nlbuf = LIST_HEAD_INITIALIZER(),
4379 * Seek and destroy leftover VXLAN encap/decap interfaces with
4380 * matching properties.
4382 nlh = mnl_nlmsg_put_header(tcf->buf);
4383 nlh->nlmsg_type = RTM_GETLINK;
4384 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4385 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4386 ifm->ifi_family = AF_UNSPEC;
4387 ret = flow_tcf_nl_ack(tcf, nlh, flow_tcf_collect_vxlan_cb, &ctx);
4389 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4390 ret = flow_tcf_send_nlcmd(tcf, &ctx);
4392 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4396 * Emit Netlink message to add/remove local address to the outer device.
4397 * The address being added is visible within the link only (scope link).
4399 * Note that an implicit route is maintained by the kernel due to the
4400 * presence of a peer address (IFA_ADDRESS).
4402 * These rules are used for encapsultion only and allow to assign
4403 * the outer tunnel source IP address.
4406 * Libmnl socket context object.
4408 * Encapsulation properties (source address and its peer).
4409 * @param[in] ifindex
4410 * Network interface to apply rule.
4412 * Toggle between add and remove.
4414 * Perform verbose error reporting if not NULL.
4417 * 0 on success, a negative errno value otherwise and rte_errno is set.
4420 flow_tcf_rule_local(struct mlx5_flow_tcf_context *tcf,
4421 const struct flow_tcf_vxlan_encap *encap,
4422 unsigned int ifindex,
4424 struct rte_flow_error *error)
4426 struct nlmsghdr *nlh;
4427 struct ifaddrmsg *ifa;
4428 alignas(struct nlmsghdr)
4429 uint8_t buf[mnl_nlmsg_size(sizeof(*ifa) + 128)];
4431 nlh = mnl_nlmsg_put_header(buf);
4432 nlh->nlmsg_type = enable ? RTM_NEWADDR : RTM_DELADDR;
4434 NLM_F_REQUEST | (enable ? NLM_F_CREATE | NLM_F_REPLACE : 0);
4436 ifa = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifa));
4437 ifa->ifa_flags = IFA_F_PERMANENT;
4438 ifa->ifa_scope = RT_SCOPE_LINK;
4439 ifa->ifa_index = ifindex;
4440 if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4441 ifa->ifa_family = AF_INET;
4442 ifa->ifa_prefixlen = 32;
4443 mnl_attr_put_u32(nlh, IFA_LOCAL, encap->ipv4.src);
4444 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST)
4445 mnl_attr_put_u32(nlh, IFA_ADDRESS,
4448 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4449 ifa->ifa_family = AF_INET6;
4450 ifa->ifa_prefixlen = 128;
4451 mnl_attr_put(nlh, IFA_LOCAL,
4452 sizeof(encap->ipv6.src),
4454 if (encap->mask & FLOW_TCF_ENCAP_IPV6_DST)
4455 mnl_attr_put(nlh, IFA_ADDRESS,
4456 sizeof(encap->ipv6.dst),
4459 if (!flow_tcf_nl_ack(tcf, nlh, NULL, NULL))
4461 return rte_flow_error_set(error, rte_errno,
4462 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4463 "netlink: cannot complete IFA request"
4468 * Emit Netlink message to add/remove neighbor.
4471 * Libmnl socket context object.
4473 * Encapsulation properties (destination address).
4474 * @param[in] ifindex
4475 * Network interface.
4477 * Toggle between add and remove.
4479 * Perform verbose error reporting if not NULL.
4482 * 0 on success, a negative errno value otherwise and rte_errno is set.
4485 flow_tcf_rule_neigh(struct mlx5_flow_tcf_context *tcf,
4486 const struct flow_tcf_vxlan_encap *encap,
4487 unsigned int ifindex,
4489 struct rte_flow_error *error)
4491 struct nlmsghdr *nlh;
4493 alignas(struct nlmsghdr)
4494 uint8_t buf[mnl_nlmsg_size(sizeof(*ndm) + 128)];
4496 nlh = mnl_nlmsg_put_header(buf);
4497 nlh->nlmsg_type = enable ? RTM_NEWNEIGH : RTM_DELNEIGH;
4499 NLM_F_REQUEST | (enable ? NLM_F_CREATE | NLM_F_REPLACE : 0);
4501 ndm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ndm));
4502 ndm->ndm_ifindex = ifindex;
4503 ndm->ndm_state = NUD_PERMANENT;
4506 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4507 ndm->ndm_family = AF_INET;
4508 mnl_attr_put_u32(nlh, NDA_DST, encap->ipv4.dst);
4510 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4511 ndm->ndm_family = AF_INET6;
4512 mnl_attr_put(nlh, NDA_DST, sizeof(encap->ipv6.dst),
4515 if (encap->mask & FLOW_TCF_ENCAP_ETH_SRC && enable)
4517 "outer ethernet source address cannot be "
4518 "forced for VXLAN encapsulation");
4519 if (encap->mask & FLOW_TCF_ENCAP_ETH_DST)
4520 mnl_attr_put(nlh, NDA_LLADDR, sizeof(encap->eth.dst),
4522 if (!flow_tcf_nl_ack(tcf, nlh, NULL, NULL))
4524 return rte_flow_error_set(error, rte_errno,
4525 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4526 "netlink: cannot complete ND request"
4531 * Manage the local IP addresses and their peers IP addresses on the
4532 * outer interface for encapsulation purposes. The kernel searches the
4533 * appropriate device for tunnel egress traffic using the outer source
4534 * IP, this IP should be assigned to the outer network device, otherwise
4535 * kernel rejects the rule.
4537 * Adds or removes the addresses using the Netlink command like this:
4538 * ip addr add <src_ip> peer <dst_ip> scope link dev <ifouter>
4540 * The addresses are local to the netdev ("scope link"), this reduces
4541 * the risk of conflicts. Note that an implicit route is maintained by
4542 * the kernel due to the presence of a peer address (IFA_ADDRESS).
4545 * Libmnl socket context object.
4547 * VTEP object, contains rule database and ifouter index.
4548 * @param[in] dev_flow
4549 * Flow object, contains the tunnel parameters (for encap only).
4551 * Toggle between add and remove.
4553 * Perform verbose error reporting if not NULL.
4556 * 0 on success, a negative errno value otherwise and rte_errno is set.
4559 flow_tcf_encap_local(struct mlx5_flow_tcf_context *tcf,
4560 struct tcf_vtep *vtep,
4561 struct mlx5_flow *dev_flow,
4563 struct rte_flow_error *error)
4565 const struct flow_tcf_vxlan_encap *encap = dev_flow->tcf.vxlan_encap;
4566 struct tcf_local_rule *rule;
4571 assert(encap->hdr.type == FLOW_TCF_TUNACT_VXLAN_ENCAP);
4572 if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4573 assert(encap->mask & FLOW_TCF_ENCAP_IPV4_DST);
4574 LIST_FOREACH(rule, &vtep->local, next) {
4575 if (rule->mask & FLOW_TCF_ENCAP_IPV4_SRC &&
4576 encap->ipv4.src == rule->ipv4.src &&
4577 encap->ipv4.dst == rule->ipv4.dst) {
4583 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4584 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4585 LIST_FOREACH(rule, &vtep->local, next) {
4586 if (rule->mask & FLOW_TCF_ENCAP_IPV6_SRC &&
4587 !memcmp(&encap->ipv6.src, &rule->ipv6.src,
4588 sizeof(encap->ipv6.src)) &&
4589 !memcmp(&encap->ipv6.dst, &rule->ipv6.dst,
4590 sizeof(encap->ipv6.dst))) {
4601 if (!rule->refcnt || !--rule->refcnt) {
4602 LIST_REMOVE(rule, next);
4603 return flow_tcf_rule_local(tcf, encap,
4604 vtep->ifouter, false, error);
4609 DRV_LOG(WARNING, "disabling not existing local rule");
4610 rte_flow_error_set(error, ENOENT,
4611 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4612 "disabling not existing local rule");
4615 rule = rte_zmalloc(__func__, sizeof(struct tcf_local_rule),
4616 alignof(struct tcf_local_rule));
4618 rte_flow_error_set(error, ENOMEM,
4619 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4620 "unable to allocate memory for local rule");
4623 *rule = (struct tcf_local_rule){.refcnt = 0,
4626 if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4627 rule->mask = FLOW_TCF_ENCAP_IPV4_SRC
4628 | FLOW_TCF_ENCAP_IPV4_DST;
4629 rule->ipv4.src = encap->ipv4.src;
4630 rule->ipv4.dst = encap->ipv4.dst;
4632 rule->mask = FLOW_TCF_ENCAP_IPV6_SRC
4633 | FLOW_TCF_ENCAP_IPV6_DST;
4634 memcpy(&rule->ipv6.src, &encap->ipv6.src, IPV6_ADDR_LEN);
4635 memcpy(&rule->ipv6.dst, &encap->ipv6.dst, IPV6_ADDR_LEN);
4637 ret = flow_tcf_rule_local(tcf, encap, vtep->ifouter, true, error);
4643 LIST_INSERT_HEAD(&vtep->local, rule, next);
4648 * Manage the destination MAC/IP addresses neigh database, kernel uses
4649 * this one to determine the destination MAC address within encapsulation
4650 * header. Adds or removes the entries using the Netlink command like this:
4651 * ip neigh add dev <ifouter> lladdr <dst_mac> to <dst_ip> nud permanent
4654 * Libmnl socket context object.
4656 * VTEP object, contains rule database and ifouter index.
4657 * @param[in] dev_flow
4658 * Flow object, contains the tunnel parameters (for encap only).
4660 * Toggle between add and remove.
4662 * Perform verbose error reporting if not NULL.
4665 * 0 on success, a negative errno value otherwise and rte_errno is set.
4668 flow_tcf_encap_neigh(struct mlx5_flow_tcf_context *tcf,
4669 struct tcf_vtep *vtep,
4670 struct mlx5_flow *dev_flow,
4672 struct rte_flow_error *error)
4674 const struct flow_tcf_vxlan_encap *encap = dev_flow->tcf.vxlan_encap;
4675 struct tcf_neigh_rule *rule;
4680 assert(encap->hdr.type == FLOW_TCF_TUNACT_VXLAN_ENCAP);
4681 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4682 assert(encap->mask & FLOW_TCF_ENCAP_IPV4_SRC);
4683 LIST_FOREACH(rule, &vtep->neigh, next) {
4684 if (rule->mask & FLOW_TCF_ENCAP_IPV4_DST &&
4685 encap->ipv4.dst == rule->ipv4.dst) {
4691 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4692 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4693 LIST_FOREACH(rule, &vtep->neigh, next) {
4694 if (rule->mask & FLOW_TCF_ENCAP_IPV6_DST &&
4695 !memcmp(&encap->ipv6.dst, &rule->ipv6.dst,
4696 sizeof(encap->ipv6.dst))) {
4703 if (memcmp(&encap->eth.dst, &rule->eth,
4704 sizeof(encap->eth.dst))) {
4705 DRV_LOG(WARNING, "Destination MAC differs"
4707 rte_flow_error_set(error, EEXIST,
4708 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4709 NULL, "Different MAC address"
4710 " neigh rule for the same"
4718 if (!rule->refcnt || !--rule->refcnt) {
4719 LIST_REMOVE(rule, next);
4720 return flow_tcf_rule_neigh(tcf, encap,
4727 DRV_LOG(WARNING, "Disabling not existing neigh rule");
4728 rte_flow_error_set(error, ENOENT,
4729 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4730 "unable to allocate memory for neigh rule");
4733 rule = rte_zmalloc(__func__, sizeof(struct tcf_neigh_rule),
4734 alignof(struct tcf_neigh_rule));
4736 rte_flow_error_set(error, ENOMEM,
4737 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4738 "unable to allocate memory for neigh rule");
4741 *rule = (struct tcf_neigh_rule){.refcnt = 0,
4744 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4745 rule->mask = FLOW_TCF_ENCAP_IPV4_DST;
4746 rule->ipv4.dst = encap->ipv4.dst;
4748 rule->mask = FLOW_TCF_ENCAP_IPV6_DST;
4749 memcpy(&rule->ipv6.dst, &encap->ipv6.dst, IPV6_ADDR_LEN);
4751 memcpy(&rule->eth, &encap->eth.dst, sizeof(rule->eth));
4752 ret = flow_tcf_rule_neigh(tcf, encap, vtep->ifouter, true, error);
4758 LIST_INSERT_HEAD(&vtep->neigh, rule, next);
4762 /* VTEP device list is shared between PMD port instances. */
4763 static LIST_HEAD(, tcf_vtep) vtep_list_vxlan = LIST_HEAD_INITIALIZER();
4764 static pthread_mutex_t vtep_list_mutex = PTHREAD_MUTEX_INITIALIZER;
4767 * Deletes VTEP network device.
4770 * Context object initialized by mlx5_flow_tcf_context_create().
4772 * Object represinting the network device to delete. Memory
4773 * allocated for this object is freed by routine.
4776 flow_tcf_vtep_delete(struct mlx5_flow_tcf_context *tcf,
4777 struct tcf_vtep *vtep)
4779 struct nlmsghdr *nlh;
4780 struct ifinfomsg *ifm;
4781 alignas(struct nlmsghdr)
4782 uint8_t buf[mnl_nlmsg_size(MNL_ALIGN(sizeof(*ifm))) +
4783 MNL_BUF_EXTRA_SPACE];
4786 assert(!vtep->refcnt);
4787 /* Delete only ifaces those we actually created. */
4788 if (vtep->created && vtep->ifindex) {
4789 DRV_LOG(INFO, "VTEP delete (%d)", vtep->ifindex);
4790 nlh = mnl_nlmsg_put_header(buf);
4791 nlh->nlmsg_type = RTM_DELLINK;
4792 nlh->nlmsg_flags = NLM_F_REQUEST;
4793 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4794 ifm->ifi_family = AF_UNSPEC;
4795 ifm->ifi_index = vtep->ifindex;
4796 assert(sizeof(buf) >= nlh->nlmsg_len);
4797 ret = flow_tcf_nl_ack(tcf, nlh, NULL, NULL);
4799 DRV_LOG(WARNING, "netlink: error deleting vxlan"
4800 " encap/decap ifindex %u",
4807 * Creates VTEP network device.
4810 * Context object initialized by mlx5_flow_tcf_context_create().
4811 * @param[in] ifouter
4812 * Outer interface to attach new-created VXLAN device
4813 * If zero the VXLAN device will not be attached to any device.
4814 * These VTEPs are used for decapsulation and can be precreated
4815 * and shared between processes.
4817 * UDP port of created VTEP device.
4819 * Perform verbose error reporting if not NULL.
4822 * Pointer to created device structure on success,
4823 * NULL otherwise and rte_errno is set.
4825 #ifdef HAVE_IFLA_VXLAN_COLLECT_METADATA
4826 static struct tcf_vtep*
4827 flow_tcf_vtep_create(struct mlx5_flow_tcf_context *tcf,
4828 unsigned int ifouter,
4829 uint16_t port, struct rte_flow_error *error)
4831 struct tcf_vtep *vtep;
4832 struct nlmsghdr *nlh;
4833 struct ifinfomsg *ifm;
4834 char name[sizeof(MLX5_VXLAN_DEVICE_PFX) + 24];
4835 alignas(struct nlmsghdr)
4836 uint8_t buf[mnl_nlmsg_size(sizeof(*ifm)) +
4837 SZ_NLATTR_DATA_OF(sizeof(name)) +
4838 SZ_NLATTR_NEST * 2 +
4839 SZ_NLATTR_STRZ_OF("vxlan") +
4840 SZ_NLATTR_DATA_OF(sizeof(uint32_t)) +
4841 SZ_NLATTR_DATA_OF(sizeof(uint16_t)) +
4842 SZ_NLATTR_DATA_OF(sizeof(uint8_t)) * 3 +
4843 MNL_BUF_EXTRA_SPACE];
4844 struct nlattr *na_info;
4845 struct nlattr *na_vxlan;
4846 rte_be16_t vxlan_port = rte_cpu_to_be_16(port);
4849 vtep = rte_zmalloc(__func__, sizeof(*vtep), alignof(struct tcf_vtep));
4851 rte_flow_error_set(error, ENOMEM,
4852 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4853 "unable to allocate memory for VTEP");
4856 *vtep = (struct tcf_vtep){
4858 .local = LIST_HEAD_INITIALIZER(),
4859 .neigh = LIST_HEAD_INITIALIZER(),
4861 memset(buf, 0, sizeof(buf));
4862 nlh = mnl_nlmsg_put_header(buf);
4863 nlh->nlmsg_type = RTM_NEWLINK;
4864 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
4865 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4866 ifm->ifi_family = AF_UNSPEC;
4869 ifm->ifi_flags = IFF_UP;
4870 ifm->ifi_change = 0xffffffff;
4871 snprintf(name, sizeof(name), "%s%u", MLX5_VXLAN_DEVICE_PFX, port);
4872 mnl_attr_put_strz(nlh, IFLA_IFNAME, name);
4873 na_info = mnl_attr_nest_start(nlh, IFLA_LINKINFO);
4875 mnl_attr_put_strz(nlh, IFLA_INFO_KIND, "vxlan");
4876 na_vxlan = mnl_attr_nest_start(nlh, IFLA_INFO_DATA);
4878 mnl_attr_put_u32(nlh, IFLA_VXLAN_LINK, ifouter);
4880 mnl_attr_put_u8(nlh, IFLA_VXLAN_COLLECT_METADATA, 1);
4881 mnl_attr_put_u8(nlh, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 1);
4882 mnl_attr_put_u8(nlh, IFLA_VXLAN_LEARNING, 0);
4883 mnl_attr_put_u16(nlh, IFLA_VXLAN_PORT, vxlan_port);
4884 mnl_attr_nest_end(nlh, na_vxlan);
4885 mnl_attr_nest_end(nlh, na_info);
4886 assert(sizeof(buf) >= nlh->nlmsg_len);
4887 ret = flow_tcf_nl_ack(tcf, nlh, NULL, NULL);
4890 "netlink: VTEP %s create failure (%d)",
4892 if (rte_errno != EEXIST || ifouter)
4894 * Some unhandled error occurred or device is
4895 * for encapsulation and cannot be shared.
4900 * Mark device we actually created.
4901 * We should explicitly delete
4902 * when we do not need it anymore.
4906 /* Try to get ifindex of created of pre-existing device. */
4907 ret = if_nametoindex(name);
4910 "VTEP %s failed to get index (%d)", name, errno);
4913 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4914 "netlink: failed to retrieve VTEP ifindex");
4917 vtep->ifindex = ret;
4918 vtep->ifouter = ifouter;
4919 memset(buf, 0, sizeof(buf));
4920 nlh = mnl_nlmsg_put_header(buf);
4921 nlh->nlmsg_type = RTM_NEWLINK;
4922 nlh->nlmsg_flags = NLM_F_REQUEST;
4923 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4924 ifm->ifi_family = AF_UNSPEC;
4926 ifm->ifi_index = vtep->ifindex;
4927 ifm->ifi_flags = IFF_UP;
4928 ifm->ifi_change = IFF_UP;
4929 ret = flow_tcf_nl_ack(tcf, nlh, NULL, NULL);
4931 rte_flow_error_set(error, -errno,
4932 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4933 "netlink: failed to set VTEP link up");
4934 DRV_LOG(WARNING, "netlink: VTEP %s set link up failure (%d)",
4938 ret = mlx5_flow_tcf_init(tcf, vtep->ifindex, error);
4940 DRV_LOG(WARNING, "VTEP %s init failure (%d)", name, rte_errno);
4943 DRV_LOG(INFO, "VTEP create (%d, %d)", vtep->port, vtep->ifindex);
4947 flow_tcf_vtep_delete(tcf, vtep);
4954 static struct tcf_vtep*
4955 flow_tcf_vtep_create(struct mlx5_flow_tcf_context *tcf __rte_unused,
4956 unsigned int ifouter __rte_unused,
4957 uint16_t port __rte_unused,
4958 struct rte_flow_error *error)
4960 rte_flow_error_set(error, ENOTSUP,
4961 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4962 "netlink: failed to create VTEP, "
4963 "vxlan metadata are not supported by kernel");
4966 #endif /* HAVE_IFLA_VXLAN_COLLECT_METADATA */
4969 * Acquire target interface index for VXLAN tunneling decapsulation.
4970 * In order to share the UDP port within the other interfaces the
4971 * VXLAN device created as not attached to any interface (if created).
4974 * Context object initialized by mlx5_flow_tcf_context_create().
4975 * @param[in] dev_flow
4976 * Flow tcf object with tunnel structure pointer set.
4978 * Perform verbose error reporting if not NULL.
4980 * Interface descriptor pointer on success,
4981 * NULL otherwise and rte_errno is set.
4983 static struct tcf_vtep*
4984 flow_tcf_decap_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
4985 struct mlx5_flow *dev_flow,
4986 struct rte_flow_error *error)
4988 struct tcf_vtep *vtep;
4989 uint16_t port = dev_flow->tcf.vxlan_decap->udp_port;
4991 LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
4992 if (vtep->port == port)
4995 if (vtep && vtep->ifouter) {
4996 rte_flow_error_set(error, -errno,
4997 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4998 "Failed to create decap VTEP with specified"
4999 " UDP port, atatched device exists");
5003 /* Device exists, just increment the reference counter. */
5005 assert(vtep->ifindex);
5008 /* No decapsulation device exists, try to create the new one. */
5009 vtep = flow_tcf_vtep_create(tcf, 0, port, error);
5011 LIST_INSERT_HEAD(&vtep_list_vxlan, vtep, next);
5016 * Aqcuire target interface index for VXLAN tunneling encapsulation.
5019 * Context object initialized by mlx5_flow_tcf_context_create().
5020 * @param[in] ifouter
5021 * Network interface index to attach VXLAN encap device to.
5022 * @param[in] dev_flow
5023 * Flow tcf object with tunnel structure pointer set.
5025 * Perform verbose error reporting if not NULL.
5027 * Interface descriptor pointer on success,
5028 * NULL otherwise and rte_errno is set.
5030 static struct tcf_vtep*
5031 flow_tcf_encap_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
5032 unsigned int ifouter,
5033 struct mlx5_flow *dev_flow __rte_unused,
5034 struct rte_flow_error *error)
5036 static uint16_t encap_port = MLX5_VXLAN_PORT_MIN - 1;
5037 struct tcf_vtep *vtep;
5041 /* Look whether the attached VTEP for encap is created. */
5042 LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
5043 if (vtep->ifouter == ifouter)
5047 /* VTEP already exists, just increment the reference. */
5052 /* Not found, we should create the new attached VTEP. */
5053 flow_tcf_encap_iface_cleanup(tcf, ifouter);
5054 flow_tcf_encap_local_cleanup(tcf, ifouter);
5055 flow_tcf_encap_neigh_cleanup(tcf, ifouter);
5056 for (pcnt = 0; pcnt <= (MLX5_VXLAN_PORT_MAX
5057 - MLX5_VXLAN_PORT_MIN); pcnt++) {
5059 /* Wraparound the UDP port index. */
5060 if (encap_port < MLX5_VXLAN_PORT_MIN ||
5061 encap_port > MLX5_VXLAN_PORT_MAX)
5062 encap_port = MLX5_VXLAN_PORT_MIN;
5063 /* Check whether UDP port is in already in use. */
5064 LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
5065 if (vtep->port == encap_port)
5069 /* Port is in use, try the next one. */
5073 vtep = flow_tcf_vtep_create(tcf, ifouter,
5076 LIST_INSERT_HEAD(&vtep_list_vxlan, vtep, next);
5079 if (rte_errno != EEXIST)
5085 assert(vtep->ifouter == ifouter);
5086 assert(vtep->ifindex);
5087 /* Create local ipaddr with peer to specify the outer IPs. */
5088 ret = flow_tcf_encap_local(tcf, vtep, dev_flow, true, error);
5090 /* Create neigh rule to specify outer destination MAC. */
5091 ret = flow_tcf_encap_neigh(tcf, vtep, dev_flow, true, error);
5093 flow_tcf_encap_local(tcf, vtep,
5094 dev_flow, false, error);
5097 if (--vtep->refcnt == 0)
5098 flow_tcf_vtep_delete(tcf, vtep);
5105 * Acquires target interface index for tunneling of any type.
5106 * Creates the new VTEP if needed.
5109 * Context object initialized by mlx5_flow_tcf_context_create().
5110 * @param[in] ifouter
5111 * Network interface index to attach VXLAN encap device to.
5112 * @param[in] dev_flow
5113 * Flow tcf object with tunnel structure pointer set.
5115 * Perform verbose error reporting if not NULL.
5117 * Interface descriptor pointer on success,
5118 * NULL otherwise and rte_errno is set.
5120 static struct tcf_vtep*
5121 flow_tcf_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
5122 unsigned int ifouter,
5123 struct mlx5_flow *dev_flow,
5124 struct rte_flow_error *error)
5126 struct tcf_vtep *vtep = NULL;
5128 assert(dev_flow->tcf.tunnel);
5129 pthread_mutex_lock(&vtep_list_mutex);
5130 switch (dev_flow->tcf.tunnel->type) {
5131 case FLOW_TCF_TUNACT_VXLAN_ENCAP:
5132 vtep = flow_tcf_encap_vtep_acquire(tcf, ifouter,
5135 case FLOW_TCF_TUNACT_VXLAN_DECAP:
5136 vtep = flow_tcf_decap_vtep_acquire(tcf, dev_flow, error);
5139 rte_flow_error_set(error, ENOTSUP,
5140 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5141 "unsupported tunnel type");
5144 pthread_mutex_unlock(&vtep_list_mutex);
5149 * Release tunneling interface by ifindex. Decrements reference
5150 * counter and actually removes the device if counter is zero.
5153 * Context object initialized by mlx5_flow_tcf_context_create().
5155 * VTEP device descriptor structure.
5156 * @param[in] dev_flow
5157 * Flow tcf object with tunnel structure pointer set.
5160 flow_tcf_vtep_release(struct mlx5_flow_tcf_context *tcf,
5161 struct tcf_vtep *vtep,
5162 struct mlx5_flow *dev_flow)
5164 assert(dev_flow->tcf.tunnel);
5165 pthread_mutex_lock(&vtep_list_mutex);
5166 switch (dev_flow->tcf.tunnel->type) {
5167 case FLOW_TCF_TUNACT_VXLAN_DECAP:
5169 case FLOW_TCF_TUNACT_VXLAN_ENCAP:
5170 /* Remove the encap ancillary rules first. */
5171 flow_tcf_encap_neigh(tcf, vtep, dev_flow, false, NULL);
5172 flow_tcf_encap_local(tcf, vtep, dev_flow, false, NULL);
5176 DRV_LOG(WARNING, "Unsupported tunnel type");
5179 assert(vtep->refcnt);
5180 if (--vtep->refcnt == 0) {
5181 LIST_REMOVE(vtep, next);
5182 flow_tcf_vtep_delete(tcf, vtep);
5184 pthread_mutex_unlock(&vtep_list_mutex);
5187 struct tcf_nlcb_query {
5190 uint32_t flags_valid:1;
5194 * Collect queried rule attributes. This is callback routine called by
5195 * libmnl mnl_cb_run() in loop for every message in received packet.
5196 * Current implementation collects the flower flags only.
5199 * Pointer to reply header.
5200 * @param[in, out] arg
5201 * Context pointer for this callback.
5204 * A positive, nonzero value on success (required by libmnl
5205 * to continue messages processing).
5208 flow_tcf_collect_query_cb(const struct nlmsghdr *nlh, void *arg)
5210 struct tcf_nlcb_query *query = arg;
5211 struct tcmsg *tcm = mnl_nlmsg_get_payload(nlh);
5212 struct nlattr *na, *na_opt;
5213 bool flower = false;
5215 if (nlh->nlmsg_type != RTM_NEWTFILTER ||
5216 tcm->tcm_handle != query->handle)
5218 mnl_attr_for_each(na, nlh, sizeof(*tcm)) {
5219 switch (mnl_attr_get_type(na)) {
5221 if (strcmp(mnl_attr_get_payload(na), "flower")) {
5222 /* Not flower filter, drop entire message. */
5229 /* Not flower options, drop entire message. */
5232 /* Check nested flower options. */
5233 mnl_attr_for_each_nested(na_opt, na) {
5234 switch (mnl_attr_get_type(na_opt)) {
5235 case TCA_FLOWER_FLAGS:
5236 query->flags_valid = 1;
5238 mnl_attr_get_u32(na_opt);
5249 * Query a TC flower rule flags via netlink.
5252 * Context object initialized by mlx5_flow_tcf_context_create().
5253 * @param[in] dev_flow
5254 * Pointer to the flow.
5255 * @param[out] pflags
5256 * pointer to the data retrieved by the query.
5259 * 0 on success, a negative errno value otherwise.
5262 flow_tcf_query_flags(struct mlx5_flow_tcf_context *tcf,
5263 struct mlx5_flow *dev_flow,
5266 struct nlmsghdr *nlh;
5268 struct tcf_nlcb_query query = {
5269 .handle = dev_flow->tcf.tcm->tcm_handle,
5272 nlh = mnl_nlmsg_put_header(tcf->buf);
5273 nlh->nlmsg_type = RTM_GETTFILTER;
5274 nlh->nlmsg_flags = NLM_F_REQUEST;
5275 tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
5276 memcpy(tcm, dev_flow->tcf.tcm, sizeof(*tcm));
5278 * Ignore Netlink error for filter query operations.
5279 * The reply length is sent by kernel as errno.
5280 * Just check we got the flags option.
5282 flow_tcf_nl_ack(tcf, nlh, flow_tcf_collect_query_cb, &query);
5283 if (!query.flags_valid) {
5287 *pflags = query.tc_flags;
5292 * Query and check the in_hw set for specified rule.
5295 * Context object initialized by mlx5_flow_tcf_context_create().
5296 * @param[in] dev_flow
5297 * Pointer to the flow to check.
5300 * 0 on success, a negative errno value otherwise.
5303 flow_tcf_check_inhw(struct mlx5_flow_tcf_context *tcf,
5304 struct mlx5_flow *dev_flow)
5309 ret = flow_tcf_query_flags(tcf, dev_flow, &flags);
5312 return (flags & TCA_CLS_FLAGS_IN_HW) ? 0 : -ENOENT;
5316 * Remove flow from E-Switch by sending Netlink message.
5319 * Pointer to Ethernet device.
5320 * @param[in, out] flow
5321 * Pointer to the sub flow.
5324 flow_tcf_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
5326 struct priv *priv = dev->data->dev_private;
5327 struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5328 struct mlx5_flow *dev_flow;
5329 struct nlmsghdr *nlh;
5333 dev_flow = LIST_FIRST(&flow->dev_flows);
5336 /* E-Switch flow can't be expanded. */
5337 assert(!LIST_NEXT(dev_flow, next));
5338 if (dev_flow->tcf.applied) {
5339 nlh = dev_flow->tcf.nlh;
5340 nlh->nlmsg_type = RTM_DELTFILTER;
5341 nlh->nlmsg_flags = NLM_F_REQUEST;
5342 flow_tcf_nl_ack(ctx, nlh, NULL, NULL);
5343 if (dev_flow->tcf.tunnel) {
5344 assert(dev_flow->tcf.tunnel->vtep);
5345 flow_tcf_vtep_release(ctx,
5346 dev_flow->tcf.tunnel->vtep,
5348 dev_flow->tcf.tunnel->vtep = NULL;
5350 dev_flow->tcf.applied = 0;
5355 * Apply flow to E-Switch by sending Netlink message.
5358 * Pointer to Ethernet device.
5359 * @param[in, out] flow
5360 * Pointer to the sub flow.
5362 * Pointer to the error structure.
5365 * 0 on success, a negative errno value otherwise and rte_ernno is set.
5368 flow_tcf_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
5369 struct rte_flow_error *error)
5371 struct priv *priv = dev->data->dev_private;
5372 struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5373 struct mlx5_flow *dev_flow;
5374 struct nlmsghdr *nlh;
5376 dev_flow = LIST_FIRST(&flow->dev_flows);
5377 /* E-Switch flow can't be expanded. */
5378 assert(!LIST_NEXT(dev_flow, next));
5379 if (dev_flow->tcf.applied)
5381 nlh = dev_flow->tcf.nlh;
5382 nlh->nlmsg_type = RTM_NEWTFILTER;
5383 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
5384 if (dev_flow->tcf.tunnel) {
5386 * Replace the interface index, target for
5387 * encapsulation, source for decapsulation.
5389 assert(!dev_flow->tcf.tunnel->vtep);
5390 assert(dev_flow->tcf.tunnel->ifindex_ptr);
5391 /* Acquire actual VTEP device when rule is being applied. */
5392 dev_flow->tcf.tunnel->vtep =
5393 flow_tcf_vtep_acquire(ctx,
5394 dev_flow->tcf.tunnel->ifindex_org,
5396 if (!dev_flow->tcf.tunnel->vtep)
5398 DRV_LOG(INFO, "Replace ifindex: %d->%d",
5399 dev_flow->tcf.tunnel->vtep->ifindex,
5400 dev_flow->tcf.tunnel->ifindex_org);
5401 *dev_flow->tcf.tunnel->ifindex_ptr =
5402 dev_flow->tcf.tunnel->vtep->ifindex;
5404 if (!flow_tcf_nl_ack(ctx, nlh, NULL, NULL)) {
5405 dev_flow->tcf.applied = 1;
5406 if (*dev_flow->tcf.ptc_flags & TCA_CLS_FLAGS_SKIP_SW)
5409 * Rule was applied without skip_sw flag set.
5410 * We should check whether the rule was acctually
5411 * accepted by hardware (have look at in_hw flag).
5413 if (flow_tcf_check_inhw(ctx, dev_flow)) {
5414 flow_tcf_remove(dev, flow);
5415 return rte_flow_error_set
5417 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5418 "netlink: rule has no in_hw flag set");
5422 if (dev_flow->tcf.tunnel) {
5423 /* Rollback the VTEP configuration if rule apply failed. */
5424 assert(dev_flow->tcf.tunnel->vtep);
5425 flow_tcf_vtep_release(ctx, dev_flow->tcf.tunnel->vtep,
5427 dev_flow->tcf.tunnel->vtep = NULL;
5429 return rte_flow_error_set(error, rte_errno,
5430 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5431 "netlink: failed to create TC flow rule");
5435 * Remove flow from E-Switch and release resources of the device flow.
5438 * Pointer to Ethernet device.
5439 * @param[in, out] flow
5440 * Pointer to the sub flow.
5443 flow_tcf_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
5445 struct mlx5_flow *dev_flow;
5449 flow_tcf_remove(dev, flow);
5450 if (flow->counter) {
5451 if (--flow->counter->ref_cnt == 0) {
5452 rte_free(flow->counter);
5453 flow->counter = NULL;
5456 dev_flow = LIST_FIRST(&flow->dev_flows);
5459 /* E-Switch flow can't be expanded. */
5460 assert(!LIST_NEXT(dev_flow, next));
5461 LIST_REMOVE(dev_flow, next);
5466 * Helper routine for figuring the space size required for a parse buffer.
5469 * array of values to use.
5471 * Current location in array.
5473 * Value to compare with.
5476 * The maximum between the given value and the array value on index.
5479 flow_tcf_arr_val_max(uint16_t array[], int idx, uint16_t value)
5481 return idx < 0 ? (value) : RTE_MAX((array)[idx], value);
5485 * Parse rtnetlink message attributes filling the attribute table with the info
5489 * Attribute table to be filled.
5491 * Maxinum entry in the attribute table.
5493 * The attributes section in the message to be parsed.
5495 * The length of the attributes section in the message.
5498 flow_tcf_nl_parse_rtattr(struct rtattr *tb[], int max,
5499 struct rtattr *rta, int len)
5501 unsigned short type;
5502 memset(tb, 0, sizeof(struct rtattr *) * (max + 1));
5503 while (RTA_OK(rta, len)) {
5504 type = rta->rta_type;
5505 if (type <= max && !tb[type])
5507 rta = RTA_NEXT(rta, len);
5512 * Extract flow counters from flower action.
5515 * flower action stats properties in the Netlink message received.
5517 * The backward sequence of rta_types, as written in the attribute table,
5518 * we need to traverse in order to get to the requested object.
5520 * Current location in rta_type table.
5522 * data holding the count statistics of the rte_flow retrieved from
5526 * 0 if data was found and retrieved, -1 otherwise.
5529 flow_tcf_nl_action_stats_parse_and_get(struct rtattr *rta,
5530 uint16_t rta_type[], int idx,
5531 struct gnet_stats_basic *data)
5533 int tca_stats_max = flow_tcf_arr_val_max(rta_type, idx,
5535 struct rtattr *tbs[tca_stats_max + 1];
5537 if (rta == NULL || idx < 0)
5539 flow_tcf_nl_parse_rtattr(tbs, tca_stats_max,
5540 RTA_DATA(rta), RTA_PAYLOAD(rta));
5541 switch (rta_type[idx]) {
5542 case TCA_STATS_BASIC:
5543 if (tbs[TCA_STATS_BASIC]) {
5544 memcpy(data, RTA_DATA(tbs[TCA_STATS_BASIC]),
5545 RTE_MIN(RTA_PAYLOAD(tbs[TCA_STATS_BASIC]),
5557 * Parse flower single action retrieving the requested action attribute,
5561 * flower action properties in the Netlink message received.
5563 * The backward sequence of rta_types, as written in the attribute table,
5564 * we need to traverse in order to get to the requested object.
5566 * Current location in rta_type table.
5568 * Count statistics retrieved from the message query.
5571 * 0 if data was found and retrieved, -1 otherwise.
5574 flow_tcf_nl_parse_one_action_and_get(struct rtattr *arg,
5575 uint16_t rta_type[], int idx, void *data)
5577 int tca_act_max = flow_tcf_arr_val_max(rta_type, idx, TCA_ACT_STATS);
5578 struct rtattr *tb[tca_act_max + 1];
5580 if (arg == NULL || idx < 0)
5582 flow_tcf_nl_parse_rtattr(tb, tca_act_max,
5583 RTA_DATA(arg), RTA_PAYLOAD(arg));
5584 if (tb[TCA_ACT_KIND] == NULL)
5586 switch (rta_type[idx]) {
5588 if (tb[TCA_ACT_STATS])
5589 return flow_tcf_nl_action_stats_parse_and_get
5592 (struct gnet_stats_basic *)data);
5601 * Parse flower action section in the message retrieving the requested
5602 * attribute from the first action that provides it.
5605 * flower section in the Netlink message received.
5607 * The backward sequence of rta_types, as written in the attribute table,
5608 * we need to traverse in order to get to the requested object.
5610 * Current location in rta_type table.
5612 * data retrieved from the message query.
5615 * 0 if data was found and retrieved, -1 otherwise.
5618 flow_tcf_nl_action_parse_and_get(struct rtattr *arg,
5619 uint16_t rta_type[], int idx, void *data)
5621 struct rtattr *tb[TCA_ACT_MAX_PRIO + 1];
5624 if (arg == NULL || idx < 0)
5626 flow_tcf_nl_parse_rtattr(tb, TCA_ACT_MAX_PRIO,
5627 RTA_DATA(arg), RTA_PAYLOAD(arg));
5628 switch (rta_type[idx]) {
5630 * flow counters are stored in the actions defined by the flow
5631 * and not in the flow itself, therefore we need to traverse the
5632 * flower chain of actions in search for them.
5634 * Note that the index is not decremented here.
5637 for (i = 0; i <= TCA_ACT_MAX_PRIO; i++) {
5639 !flow_tcf_nl_parse_one_action_and_get(tb[i],
5652 * Parse flower classifier options in the message, retrieving the requested
5653 * attribute if found.
5656 * flower section in the Netlink message received.
5658 * The backward sequence of rta_types, as written in the attribute table,
5659 * we need to traverse in order to get to the requested object.
5661 * Current location in rta_type table.
5663 * data retrieved from the message query.
5666 * 0 if data was found and retrieved, -1 otherwise.
5669 flow_tcf_nl_opts_parse_and_get(struct rtattr *opt,
5670 uint16_t rta_type[], int idx, void *data)
5672 int tca_flower_max = flow_tcf_arr_val_max(rta_type, idx,
5674 struct rtattr *tb[tca_flower_max + 1];
5676 if (!opt || idx < 0)
5678 flow_tcf_nl_parse_rtattr(tb, tca_flower_max,
5679 RTA_DATA(opt), RTA_PAYLOAD(opt));
5680 switch (rta_type[idx]) {
5681 case TCA_FLOWER_ACT:
5682 if (tb[TCA_FLOWER_ACT])
5683 return flow_tcf_nl_action_parse_and_get
5684 (tb[TCA_FLOWER_ACT],
5685 rta_type, --idx, data);
5694 * Parse Netlink reply on filter query, retrieving the flow counters.
5697 * Message received from Netlink.
5699 * The backward sequence of rta_types, as written in the attribute table,
5700 * we need to traverse in order to get to the requested object.
5702 * Current location in rta_type table.
5704 * data retrieved from the message query.
5707 * 0 if data was found and retrieved, -1 otherwise.
5710 flow_tcf_nl_filter_parse_and_get(struct nlmsghdr *cnlh,
5711 uint16_t rta_type[], int idx, void *data)
5713 struct nlmsghdr *nlh = cnlh;
5714 struct tcmsg *t = NLMSG_DATA(nlh);
5715 int len = nlh->nlmsg_len;
5716 int tca_max = flow_tcf_arr_val_max(rta_type, idx, TCA_OPTIONS);
5717 struct rtattr *tb[tca_max + 1];
5721 if (nlh->nlmsg_type != RTM_NEWTFILTER &&
5722 nlh->nlmsg_type != RTM_GETTFILTER &&
5723 nlh->nlmsg_type != RTM_DELTFILTER)
5725 len -= NLMSG_LENGTH(sizeof(*t));
5728 flow_tcf_nl_parse_rtattr(tb, tca_max, TCA_RTA(t), len);
5729 /* Not a TC flower flow - bail out */
5730 if (!tb[TCA_KIND] ||
5731 strcmp(RTA_DATA(tb[TCA_KIND]), "flower"))
5733 switch (rta_type[idx]) {
5735 if (tb[TCA_OPTIONS])
5736 return flow_tcf_nl_opts_parse_and_get(tb[TCA_OPTIONS],
5747 * A callback to parse Netlink reply on TC flower query.
5750 * Message received from Netlink.
5752 * Pointer to data area to be filled by the parsing routine.
5753 * assumed to be a pointer to struct flow_tcf_stats_basic.
5759 flow_tcf_nl_message_get_stats_basic(const struct nlmsghdr *nlh, void *data)
5762 * The backward sequence of rta_types to pass in order to get
5765 uint16_t rta_type[] = { TCA_STATS_BASIC, TCA_ACT_STATS,
5766 TCA_FLOWER_ACT, TCA_OPTIONS };
5767 struct flow_tcf_stats_basic *sb_data = data;
5769 const struct nlmsghdr *c;
5770 struct nlmsghdr *nc;
5771 } tnlh = { .c = nlh };
5773 if (!flow_tcf_nl_filter_parse_and_get(tnlh.nc, rta_type,
5774 RTE_DIM(rta_type) - 1,
5775 (void *)&sb_data->counters))
5776 sb_data->valid = true;
5781 * Query a TC flower rule for its statistics via netlink.
5784 * Pointer to Ethernet device.
5786 * Pointer to the sub flow.
5788 * data retrieved by the query.
5790 * Perform verbose error reporting if not NULL.
5793 * 0 on success, a negative errno value otherwise and rte_errno is set.
5796 flow_tcf_query_count(struct rte_eth_dev *dev,
5797 struct rte_flow *flow,
5799 struct rte_flow_error *error)
5801 struct flow_tcf_stats_basic sb_data;
5802 struct rte_flow_query_count *qc = data;
5803 struct priv *priv = dev->data->dev_private;
5804 struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5805 struct mnl_socket *nl = ctx->nl;
5806 struct mlx5_flow *dev_flow;
5807 struct nlmsghdr *nlh;
5808 uint32_t seq = priv->tcf_context->seq++;
5812 memset(&sb_data, 0, sizeof(sb_data));
5813 dev_flow = LIST_FIRST(&flow->dev_flows);
5814 /* E-Switch flow can't be expanded. */
5815 assert(!LIST_NEXT(dev_flow, next));
5816 if (!dev_flow->flow->counter)
5818 nlh = dev_flow->tcf.nlh;
5819 nlh->nlmsg_type = RTM_GETTFILTER;
5820 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ECHO;
5821 nlh->nlmsg_seq = seq;
5822 if (mnl_socket_sendto(nl, nlh, nlh->nlmsg_len) == -1)
5825 ret = mnl_socket_recvfrom(nl, ctx->buf, ctx->buf_size);
5828 ret = mnl_cb_run(ctx->buf, ret, seq,
5829 mnl_socket_get_portid(nl),
5830 flow_tcf_nl_message_get_stats_basic,
5833 /* Return the delta from last reset. */
5834 if (sb_data.valid) {
5835 /* Return the delta from last reset. */
5838 qc->hits = sb_data.counters.packets - flow->counter->hits;
5839 qc->bytes = sb_data.counters.bytes - flow->counter->bytes;
5841 flow->counter->hits = sb_data.counters.packets;
5842 flow->counter->bytes = sb_data.counters.bytes;
5846 return rte_flow_error_set(error, EINVAL,
5847 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5849 "flow does not have counter");
5851 return rte_flow_error_set
5852 (error, errno, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5853 NULL, "netlink: failed to read flow rule counters");
5855 return rte_flow_error_set
5856 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5857 NULL, "counters are not available.");
5863 * @see rte_flow_query()
5867 flow_tcf_query(struct rte_eth_dev *dev,
5868 struct rte_flow *flow,
5869 const struct rte_flow_action *actions,
5871 struct rte_flow_error *error)
5875 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5876 switch (actions->type) {
5877 case RTE_FLOW_ACTION_TYPE_VOID:
5879 case RTE_FLOW_ACTION_TYPE_COUNT:
5880 ret = flow_tcf_query_count(dev, flow, data, error);
5883 return rte_flow_error_set(error, ENOTSUP,
5884 RTE_FLOW_ERROR_TYPE_ACTION,
5886 "action not supported");
5892 const struct mlx5_flow_driver_ops mlx5_flow_tcf_drv_ops = {
5893 .validate = flow_tcf_validate,
5894 .prepare = flow_tcf_prepare,
5895 .translate = flow_tcf_translate,
5896 .apply = flow_tcf_apply,
5897 .remove = flow_tcf_remove,
5898 .destroy = flow_tcf_destroy,
5899 .query = flow_tcf_query,
5903 * Create and configure a libmnl socket for Netlink flow rules.
5906 * A valid libmnl socket object pointer on success, NULL otherwise and
5909 static struct mnl_socket *
5910 flow_tcf_mnl_socket_create(void)
5912 struct mnl_socket *nl = mnl_socket_open(NETLINK_ROUTE);
5915 mnl_socket_setsockopt(nl, NETLINK_CAP_ACK, &(int){ 1 },
5917 if (!mnl_socket_bind(nl, 0, MNL_SOCKET_AUTOPID))
5922 mnl_socket_close(nl);
5927 * Destroy a libmnl socket.
5930 * Libmnl socket of the @p NETLINK_ROUTE kind.
5933 flow_tcf_mnl_socket_destroy(struct mnl_socket *nl)
5936 mnl_socket_close(nl);
5940 * Initialize ingress qdisc of a given network interface.
5943 * Pointer to tc-flower context to use.
5945 * Index of network interface to initialize.
5947 * Perform verbose error reporting if not NULL.
5950 * 0 on success, a negative errno value otherwise and rte_errno is set.
5953 mlx5_flow_tcf_init(struct mlx5_flow_tcf_context *ctx,
5954 unsigned int ifindex, struct rte_flow_error *error)
5956 struct nlmsghdr *nlh;
5958 alignas(struct nlmsghdr)
5959 uint8_t buf[mnl_nlmsg_size(sizeof(*tcm)) +
5960 SZ_NLATTR_STRZ_OF("ingress") +
5961 MNL_BUF_EXTRA_SPACE];
5963 /* Destroy existing ingress qdisc and everything attached to it. */
5964 nlh = mnl_nlmsg_put_header(buf);
5965 nlh->nlmsg_type = RTM_DELQDISC;
5966 nlh->nlmsg_flags = NLM_F_REQUEST;
5967 tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
5968 tcm->tcm_family = AF_UNSPEC;
5969 tcm->tcm_ifindex = ifindex;
5970 tcm->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
5971 tcm->tcm_parent = TC_H_INGRESS;
5972 assert(sizeof(buf) >= nlh->nlmsg_len);
5973 /* Ignore errors when qdisc is already absent. */
5974 if (flow_tcf_nl_ack(ctx, nlh, NULL, NULL) &&
5975 rte_errno != EINVAL && rte_errno != ENOENT)
5976 return rte_flow_error_set(error, rte_errno,
5977 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5978 "netlink: failed to remove ingress"
5980 /* Create fresh ingress qdisc. */
5981 nlh = mnl_nlmsg_put_header(buf);
5982 nlh->nlmsg_type = RTM_NEWQDISC;
5983 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
5984 tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
5985 tcm->tcm_family = AF_UNSPEC;
5986 tcm->tcm_ifindex = ifindex;
5987 tcm->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
5988 tcm->tcm_parent = TC_H_INGRESS;
5989 mnl_attr_put_strz_check(nlh, sizeof(buf), TCA_KIND, "ingress");
5990 assert(sizeof(buf) >= nlh->nlmsg_len);
5991 if (flow_tcf_nl_ack(ctx, nlh, NULL, NULL))
5992 return rte_flow_error_set(error, rte_errno,
5993 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5994 "netlink: failed to create ingress"
6000 * Create libmnl context for Netlink flow rules.
6003 * A valid libmnl socket object pointer on success, NULL otherwise and
6006 struct mlx5_flow_tcf_context *
6007 mlx5_flow_tcf_context_create(void)
6009 struct mlx5_flow_tcf_context *ctx = rte_zmalloc(__func__,
6014 ctx->nl = flow_tcf_mnl_socket_create();
6017 ctx->buf_size = MNL_SOCKET_BUFFER_SIZE;
6018 ctx->buf = rte_zmalloc(__func__,
6019 ctx->buf_size, sizeof(uint32_t));
6022 ctx->seq = random();
6025 mlx5_flow_tcf_context_destroy(ctx);
6030 * Destroy a libmnl context.
6033 * Libmnl socket of the @p NETLINK_ROUTE kind.
6036 mlx5_flow_tcf_context_destroy(struct mlx5_flow_tcf_context *ctx)
6040 flow_tcf_mnl_socket_destroy(ctx->nl);