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 HAVE_TCA_CHAIN
166 #ifndef HAVE_TCA_FLOWER_ACT
167 #define TCA_FLOWER_ACT 3
169 #ifndef HAVE_TCA_FLOWER_FLAGS
170 #define TCA_FLOWER_FLAGS 22
172 #ifndef HAVE_TCA_FLOWER_KEY_ETH_TYPE
173 #define TCA_FLOWER_KEY_ETH_TYPE 8
175 #ifndef HAVE_TCA_FLOWER_KEY_ETH_DST
176 #define TCA_FLOWER_KEY_ETH_DST 4
178 #ifndef HAVE_TCA_FLOWER_KEY_ETH_DST_MASK
179 #define TCA_FLOWER_KEY_ETH_DST_MASK 5
181 #ifndef HAVE_TCA_FLOWER_KEY_ETH_SRC
182 #define TCA_FLOWER_KEY_ETH_SRC 6
184 #ifndef HAVE_TCA_FLOWER_KEY_ETH_SRC_MASK
185 #define TCA_FLOWER_KEY_ETH_SRC_MASK 7
187 #ifndef HAVE_TCA_FLOWER_KEY_IP_PROTO
188 #define TCA_FLOWER_KEY_IP_PROTO 9
190 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_SRC
191 #define TCA_FLOWER_KEY_IPV4_SRC 10
193 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_SRC_MASK
194 #define TCA_FLOWER_KEY_IPV4_SRC_MASK 11
196 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_DST
197 #define TCA_FLOWER_KEY_IPV4_DST 12
199 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_DST_MASK
200 #define TCA_FLOWER_KEY_IPV4_DST_MASK 13
202 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_SRC
203 #define TCA_FLOWER_KEY_IPV6_SRC 14
205 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_SRC_MASK
206 #define TCA_FLOWER_KEY_IPV6_SRC_MASK 15
208 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_DST
209 #define TCA_FLOWER_KEY_IPV6_DST 16
211 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_DST_MASK
212 #define TCA_FLOWER_KEY_IPV6_DST_MASK 17
214 #ifndef HAVE_TCA_FLOWER_KEY_TCP_SRC
215 #define TCA_FLOWER_KEY_TCP_SRC 18
217 #ifndef HAVE_TCA_FLOWER_KEY_TCP_SRC_MASK
218 #define TCA_FLOWER_KEY_TCP_SRC_MASK 35
220 #ifndef HAVE_TCA_FLOWER_KEY_TCP_DST
221 #define TCA_FLOWER_KEY_TCP_DST 19
223 #ifndef HAVE_TCA_FLOWER_KEY_TCP_DST_MASK
224 #define TCA_FLOWER_KEY_TCP_DST_MASK 36
226 #ifndef HAVE_TCA_FLOWER_KEY_UDP_SRC
227 #define TCA_FLOWER_KEY_UDP_SRC 20
229 #ifndef HAVE_TCA_FLOWER_KEY_UDP_SRC_MASK
230 #define TCA_FLOWER_KEY_UDP_SRC_MASK 37
232 #ifndef HAVE_TCA_FLOWER_KEY_UDP_DST
233 #define TCA_FLOWER_KEY_UDP_DST 21
235 #ifndef HAVE_TCA_FLOWER_KEY_UDP_DST_MASK
236 #define TCA_FLOWER_KEY_UDP_DST_MASK 38
238 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_ID
239 #define TCA_FLOWER_KEY_VLAN_ID 23
241 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_PRIO
242 #define TCA_FLOWER_KEY_VLAN_PRIO 24
244 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_ETH_TYPE
245 #define TCA_FLOWER_KEY_VLAN_ETH_TYPE 25
247 #ifndef HAVE_TCA_FLOWER_KEY_ENC_KEY_ID
248 #define TCA_FLOWER_KEY_ENC_KEY_ID 26
250 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_SRC
251 #define TCA_FLOWER_KEY_ENC_IPV4_SRC 27
253 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK
254 #define TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK 28
256 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_DST
257 #define TCA_FLOWER_KEY_ENC_IPV4_DST 29
259 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_DST_MASK
260 #define TCA_FLOWER_KEY_ENC_IPV4_DST_MASK 30
262 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_SRC
263 #define TCA_FLOWER_KEY_ENC_IPV6_SRC 31
265 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK
266 #define TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK 32
268 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_DST
269 #define TCA_FLOWER_KEY_ENC_IPV6_DST 33
271 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_DST_MASK
272 #define TCA_FLOWER_KEY_ENC_IPV6_DST_MASK 34
274 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_SRC_PORT
275 #define TCA_FLOWER_KEY_ENC_UDP_SRC_PORT 43
277 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK
278 #define TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK 44
280 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_DST_PORT
281 #define TCA_FLOWER_KEY_ENC_UDP_DST_PORT 45
283 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK
284 #define TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK 46
286 #ifndef HAVE_TCA_FLOWER_KEY_TCP_FLAGS
287 #define TCA_FLOWER_KEY_TCP_FLAGS 71
289 #ifndef HAVE_TCA_FLOWER_KEY_TCP_FLAGS_MASK
290 #define TCA_FLOWER_KEY_TCP_FLAGS_MASK 72
292 #ifndef HAVE_TC_ACT_GOTO_CHAIN
293 #define TC_ACT_GOTO_CHAIN 0x20000000
296 #ifndef IPV6_ADDR_LEN
297 #define IPV6_ADDR_LEN 16
300 #ifndef IPV4_ADDR_LEN
301 #define IPV4_ADDR_LEN 4
305 #define TP_PORT_LEN 2 /* Transport Port (UDP/TCP) Length */
312 #ifndef TCA_ACT_MAX_PRIO
313 #define TCA_ACT_MAX_PRIO 32
316 /** UDP port range of VXLAN devices created by driver. */
317 #define MLX5_VXLAN_PORT_MIN 30000
318 #define MLX5_VXLAN_PORT_MAX 60000
319 #define MLX5_VXLAN_DEVICE_PFX "vmlx_"
321 /** Tunnel action type, used for @p type in header structure. */
322 enum flow_tcf_tunact_type {
323 FLOW_TCF_TUNACT_VXLAN_DECAP,
324 FLOW_TCF_TUNACT_VXLAN_ENCAP,
327 /** Flags used for @p mask in tunnel action encap descriptors. */
328 #define FLOW_TCF_ENCAP_ETH_SRC (1u << 0)
329 #define FLOW_TCF_ENCAP_ETH_DST (1u << 1)
330 #define FLOW_TCF_ENCAP_IPV4_SRC (1u << 2)
331 #define FLOW_TCF_ENCAP_IPV4_DST (1u << 3)
332 #define FLOW_TCF_ENCAP_IPV6_SRC (1u << 4)
333 #define FLOW_TCF_ENCAP_IPV6_DST (1u << 5)
334 #define FLOW_TCF_ENCAP_UDP_SRC (1u << 6)
335 #define FLOW_TCF_ENCAP_UDP_DST (1u << 7)
336 #define FLOW_TCF_ENCAP_VXLAN_VNI (1u << 8)
339 * Structure for holding netlink context.
340 * Note the size of the message buffer which is MNL_SOCKET_BUFFER_SIZE.
341 * Using this (8KB) buffer size ensures that netlink messages will never be
344 struct mlx5_flow_tcf_context {
345 struct mnl_socket *nl; /* NETLINK_ROUTE libmnl socket. */
346 uint32_t seq; /* Message sequence number. */
347 uint32_t buf_size; /* Message buffer size. */
348 uint8_t *buf; /* Message buffer. */
352 * Neigh rule structure. The neigh rule is applied via Netlink to
353 * outer tunnel iface in order to provide destination MAC address
354 * for the VXLAN encapsultion. The neigh rule is implicitly related
355 * to the Flow itself and can be shared by multiple Flows.
357 struct tcf_neigh_rule {
358 LIST_ENTRY(tcf_neigh_rule) next;
360 struct ether_addr eth;
367 uint8_t dst[IPV6_ADDR_LEN];
373 * Local rule structure. The local rule is applied via Netlink to
374 * outer tunnel iface in order to provide local and peer IP addresses
375 * of the VXLAN tunnel for encapsulation. The local rule is implicitly
376 * related to the Flow itself and can be shared by multiple Flows.
378 struct tcf_local_rule {
379 LIST_ENTRY(tcf_local_rule) next;
388 uint8_t dst[IPV6_ADDR_LEN];
389 uint8_t src[IPV6_ADDR_LEN];
394 /** VXLAN virtual netdev. */
396 LIST_ENTRY(tcf_vtep) next;
397 LIST_HEAD(, tcf_neigh_rule) neigh;
398 LIST_HEAD(, tcf_local_rule) local;
400 unsigned int ifindex; /**< Own interface index. */
401 unsigned int ifouter; /**< Index of device attached to. */
406 /** Tunnel descriptor header, common for all tunnel types. */
407 struct flow_tcf_tunnel_hdr {
408 uint32_t type; /**< Tunnel action type. */
409 struct tcf_vtep *vtep; /**< Virtual tunnel endpoint device. */
410 unsigned int ifindex_org; /**< Original dst/src interface */
411 unsigned int *ifindex_ptr; /**< Interface ptr in message. */
414 struct flow_tcf_vxlan_decap {
415 struct flow_tcf_tunnel_hdr hdr;
419 struct flow_tcf_vxlan_encap {
420 struct flow_tcf_tunnel_hdr hdr;
423 struct ether_addr dst;
424 struct ether_addr src;
432 uint8_t dst[IPV6_ADDR_LEN];
433 uint8_t src[IPV6_ADDR_LEN];
445 /** Structure used when extracting the values of a flow counters
446 * from a netlink message.
448 struct flow_tcf_stats_basic {
450 struct gnet_stats_basic counters;
453 /** Empty masks for known item types. */
455 struct rte_flow_item_port_id port_id;
456 struct rte_flow_item_eth eth;
457 struct rte_flow_item_vlan vlan;
458 struct rte_flow_item_ipv4 ipv4;
459 struct rte_flow_item_ipv6 ipv6;
460 struct rte_flow_item_tcp tcp;
461 struct rte_flow_item_udp udp;
462 struct rte_flow_item_vxlan vxlan;
463 } flow_tcf_mask_empty;
465 /** Supported masks for known item types. */
466 static const struct {
467 struct rte_flow_item_port_id port_id;
468 struct rte_flow_item_eth eth;
469 struct rte_flow_item_vlan vlan;
470 struct rte_flow_item_ipv4 ipv4;
471 struct rte_flow_item_ipv6 ipv6;
472 struct rte_flow_item_tcp tcp;
473 struct rte_flow_item_udp udp;
474 struct rte_flow_item_vxlan vxlan;
475 } flow_tcf_mask_supported = {
480 .type = RTE_BE16(0xffff),
481 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
482 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
485 /* PCP and VID only, no DEI. */
486 .tci = RTE_BE16(0xefff),
487 .inner_type = RTE_BE16(0xffff),
490 .next_proto_id = 0xff,
491 .src_addr = RTE_BE32(0xffffffff),
492 .dst_addr = RTE_BE32(0xffffffff),
497 "\xff\xff\xff\xff\xff\xff\xff\xff"
498 "\xff\xff\xff\xff\xff\xff\xff\xff",
500 "\xff\xff\xff\xff\xff\xff\xff\xff"
501 "\xff\xff\xff\xff\xff\xff\xff\xff",
504 .src_port = RTE_BE16(0xffff),
505 .dst_port = RTE_BE16(0xffff),
509 .src_port = RTE_BE16(0xffff),
510 .dst_port = RTE_BE16(0xffff),
513 .vni = "\xff\xff\xff",
517 #define SZ_NLATTR_HDR MNL_ALIGN(sizeof(struct nlattr))
518 #define SZ_NLATTR_NEST SZ_NLATTR_HDR
519 #define SZ_NLATTR_DATA_OF(len) MNL_ALIGN(SZ_NLATTR_HDR + (len))
520 #define SZ_NLATTR_TYPE_OF(typ) SZ_NLATTR_DATA_OF(sizeof(typ))
521 #define SZ_NLATTR_STRZ_OF(str) SZ_NLATTR_DATA_OF(strlen(str) + 1)
523 #define PTOI_TABLE_SZ_MAX(dev) (mlx5_dev_to_port_id((dev)->device, NULL, 0) + 2)
525 /** DPDK port to network interface index (ifindex) conversion. */
526 struct flow_tcf_ptoi {
527 uint16_t port_id; /**< DPDK port ID. */
528 unsigned int ifindex; /**< Network interface index. */
531 /* Due to a limitation on driver/FW. */
532 #define MLX5_TCF_GROUP_ID_MAX 3
535 * Due to a limitation on driver/FW, priority ranges from 1 to 16 in kernel.
536 * Priority in rte_flow attribute starts from 0 and is added by 1 in
537 * translation. This is subject to be changed to determine the max priority
538 * based on trial-and-error like Verbs driver once the restriction is lifted or
539 * the range is extended.
541 #define MLX5_TCF_GROUP_PRIORITY_MAX 15
543 #define MLX5_TCF_FATE_ACTIONS \
544 (MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_PORT_ID | \
545 MLX5_FLOW_ACTION_JUMP)
547 #define MLX5_TCF_VLAN_ACTIONS \
548 (MLX5_FLOW_ACTION_OF_POP_VLAN | MLX5_FLOW_ACTION_OF_PUSH_VLAN | \
549 MLX5_FLOW_ACTION_OF_SET_VLAN_VID | MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
551 #define MLX5_TCF_VXLAN_ACTIONS \
552 (MLX5_FLOW_ACTION_VXLAN_ENCAP | MLX5_FLOW_ACTION_VXLAN_DECAP)
554 #define MLX5_TCF_PEDIT_ACTIONS \
555 (MLX5_FLOW_ACTION_SET_IPV4_SRC | MLX5_FLOW_ACTION_SET_IPV4_DST | \
556 MLX5_FLOW_ACTION_SET_IPV6_SRC | MLX5_FLOW_ACTION_SET_IPV6_DST | \
557 MLX5_FLOW_ACTION_SET_TP_SRC | MLX5_FLOW_ACTION_SET_TP_DST | \
558 MLX5_FLOW_ACTION_SET_TTL | MLX5_FLOW_ACTION_DEC_TTL | \
559 MLX5_FLOW_ACTION_SET_MAC_SRC | MLX5_FLOW_ACTION_SET_MAC_DST)
561 #define MLX5_TCF_CONFIG_ACTIONS \
562 (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_JUMP | \
563 MLX5_FLOW_ACTION_OF_PUSH_VLAN | MLX5_FLOW_ACTION_OF_SET_VLAN_VID | \
564 MLX5_FLOW_ACTION_OF_SET_VLAN_PCP | \
565 (MLX5_TCF_PEDIT_ACTIONS & ~MLX5_FLOW_ACTION_DEC_TTL))
567 #define MAX_PEDIT_KEYS 128
568 #define SZ_PEDIT_KEY_VAL 4
570 #define NUM_OF_PEDIT_KEYS(sz) \
571 (((sz) / SZ_PEDIT_KEY_VAL) + (((sz) % SZ_PEDIT_KEY_VAL) ? 1 : 0))
573 struct pedit_key_ex {
574 enum pedit_header_type htype;
578 struct pedit_parser {
579 struct tc_pedit_sel sel;
580 struct tc_pedit_key keys[MAX_PEDIT_KEYS];
581 struct pedit_key_ex keys_ex[MAX_PEDIT_KEYS];
585 * Create space for using the implicitly created TC flow counter.
588 * Pointer to the Ethernet device structure.
591 * A pointer to the counter data structure, NULL otherwise and
594 static struct mlx5_flow_counter *
595 flow_tcf_counter_new(void)
597 struct mlx5_flow_counter *cnt;
600 * eswitch counter cannot be shared and its id is unknown.
601 * currently returning all with id 0.
602 * in the future maybe better to switch to unique numbers.
604 struct mlx5_flow_counter tmpl = {
607 cnt = rte_calloc(__func__, 1, sizeof(*cnt), 0);
613 /* Implicit counter, do not add to list. */
618 * Set pedit key of MAC address
621 * pointer to action specification
622 * @param[in,out] p_parser
623 * pointer to pedit_parser
626 flow_tcf_pedit_key_set_mac(const struct rte_flow_action *actions,
627 struct pedit_parser *p_parser)
629 int idx = p_parser->sel.nkeys;
630 uint32_t off = actions->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
631 offsetof(struct ether_hdr, s_addr) :
632 offsetof(struct ether_hdr, d_addr);
633 const struct rte_flow_action_set_mac *conf =
634 (const struct rte_flow_action_set_mac *)actions->conf;
636 p_parser->keys[idx].off = off;
637 p_parser->keys[idx].mask = ~UINT32_MAX;
638 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_ETH;
639 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
640 memcpy(&p_parser->keys[idx].val,
641 conf->mac_addr, SZ_PEDIT_KEY_VAL);
643 p_parser->keys[idx].off = off + SZ_PEDIT_KEY_VAL;
644 p_parser->keys[idx].mask = 0xFFFF0000;
645 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_ETH;
646 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
647 memcpy(&p_parser->keys[idx].val,
648 conf->mac_addr + SZ_PEDIT_KEY_VAL,
649 ETHER_ADDR_LEN - SZ_PEDIT_KEY_VAL);
650 p_parser->sel.nkeys = (++idx);
654 * Set pedit key of decrease/set ttl
657 * pointer to action specification
658 * @param[in,out] p_parser
659 * pointer to pedit_parser
660 * @param[in] item_flags
661 * flags of all items presented
664 flow_tcf_pedit_key_set_dec_ttl(const struct rte_flow_action *actions,
665 struct pedit_parser *p_parser,
668 int idx = p_parser->sel.nkeys;
670 p_parser->keys[idx].mask = 0xFFFFFF00;
671 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4) {
672 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP4;
673 p_parser->keys[idx].off =
674 offsetof(struct ipv4_hdr, time_to_live);
676 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6) {
677 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP6;
678 p_parser->keys[idx].off =
679 offsetof(struct ipv6_hdr, hop_limits);
681 if (actions->type == RTE_FLOW_ACTION_TYPE_DEC_TTL) {
682 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_ADD;
683 p_parser->keys[idx].val = 0x000000FF;
685 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
686 p_parser->keys[idx].val =
687 (__u32)((const struct rte_flow_action_set_ttl *)
688 actions->conf)->ttl_value;
690 p_parser->sel.nkeys = (++idx);
694 * Set pedit key of transport (TCP/UDP) port value
697 * pointer to action specification
698 * @param[in,out] p_parser
699 * pointer to pedit_parser
700 * @param[in] item_flags
701 * flags of all items presented
704 flow_tcf_pedit_key_set_tp_port(const struct rte_flow_action *actions,
705 struct pedit_parser *p_parser,
708 int idx = p_parser->sel.nkeys;
710 if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
711 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_UDP;
712 if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
713 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_TCP;
714 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
715 /* offset of src/dst port is same for TCP and UDP */
716 p_parser->keys[idx].off =
717 actions->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
718 offsetof(struct tcp_hdr, src_port) :
719 offsetof(struct tcp_hdr, dst_port);
720 p_parser->keys[idx].mask = 0xFFFF0000;
721 p_parser->keys[idx].val =
722 (__u32)((const struct rte_flow_action_set_tp *)
723 actions->conf)->port;
724 p_parser->sel.nkeys = (++idx);
728 * Set pedit key of ipv6 address
731 * pointer to action specification
732 * @param[in,out] p_parser
733 * pointer to pedit_parser
736 flow_tcf_pedit_key_set_ipv6_addr(const struct rte_flow_action *actions,
737 struct pedit_parser *p_parser)
739 int idx = p_parser->sel.nkeys;
740 int keys = NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
742 actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
743 offsetof(struct ipv6_hdr, src_addr) :
744 offsetof(struct ipv6_hdr, dst_addr);
745 const struct rte_flow_action_set_ipv6 *conf =
746 (const struct rte_flow_action_set_ipv6 *)actions->conf;
748 for (int i = 0; i < keys; i++, idx++) {
749 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP6;
750 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
751 p_parser->keys[idx].off = off_base + i * SZ_PEDIT_KEY_VAL;
752 p_parser->keys[idx].mask = ~UINT32_MAX;
753 memcpy(&p_parser->keys[idx].val,
754 conf->ipv6_addr + i * SZ_PEDIT_KEY_VAL,
757 p_parser->sel.nkeys += keys;
761 * Set pedit key of ipv4 address
764 * pointer to action specification
765 * @param[in,out] p_parser
766 * pointer to pedit_parser
769 flow_tcf_pedit_key_set_ipv4_addr(const struct rte_flow_action *actions,
770 struct pedit_parser *p_parser)
772 int idx = p_parser->sel.nkeys;
774 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP4;
775 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
776 p_parser->keys[idx].off =
777 actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
778 offsetof(struct ipv4_hdr, src_addr) :
779 offsetof(struct ipv4_hdr, dst_addr);
780 p_parser->keys[idx].mask = ~UINT32_MAX;
781 p_parser->keys[idx].val =
782 ((const struct rte_flow_action_set_ipv4 *)
783 actions->conf)->ipv4_addr;
784 p_parser->sel.nkeys = (++idx);
788 * Create the pedit's na attribute in netlink message
789 * on pre-allocate message buffer
792 * pointer to pre-allocated netlink message buffer
793 * @param[in,out] actions
794 * pointer to pointer of actions specification.
795 * @param[in,out] action_flags
796 * pointer to actions flags
797 * @param[in] item_flags
798 * flags of all item presented
801 flow_tcf_create_pedit_mnl_msg(struct nlmsghdr *nl,
802 const struct rte_flow_action **actions,
805 struct pedit_parser p_parser;
806 struct nlattr *na_act_options;
807 struct nlattr *na_pedit_keys;
809 memset(&p_parser, 0, sizeof(p_parser));
810 mnl_attr_put_strz(nl, TCA_ACT_KIND, "pedit");
811 na_act_options = mnl_attr_nest_start(nl, TCA_ACT_OPTIONS);
812 /* all modify header actions should be in one tc-pedit action */
813 for (; (*actions)->type != RTE_FLOW_ACTION_TYPE_END; (*actions)++) {
814 switch ((*actions)->type) {
815 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
816 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
817 flow_tcf_pedit_key_set_ipv4_addr(*actions, &p_parser);
819 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
820 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
821 flow_tcf_pedit_key_set_ipv6_addr(*actions, &p_parser);
823 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
824 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
825 flow_tcf_pedit_key_set_tp_port(*actions,
826 &p_parser, item_flags);
828 case RTE_FLOW_ACTION_TYPE_SET_TTL:
829 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
830 flow_tcf_pedit_key_set_dec_ttl(*actions,
831 &p_parser, item_flags);
833 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
834 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
835 flow_tcf_pedit_key_set_mac(*actions, &p_parser);
838 goto pedit_mnl_msg_done;
842 p_parser.sel.action = TC_ACT_PIPE;
843 mnl_attr_put(nl, TCA_PEDIT_PARMS_EX,
844 sizeof(p_parser.sel) +
845 p_parser.sel.nkeys * sizeof(struct tc_pedit_key),
848 mnl_attr_nest_start(nl, TCA_PEDIT_KEYS_EX | NLA_F_NESTED);
849 for (int i = 0; i < p_parser.sel.nkeys; i++) {
850 struct nlattr *na_pedit_key =
851 mnl_attr_nest_start(nl,
852 TCA_PEDIT_KEY_EX | NLA_F_NESTED);
853 mnl_attr_put_u16(nl, TCA_PEDIT_KEY_EX_HTYPE,
854 p_parser.keys_ex[i].htype);
855 mnl_attr_put_u16(nl, TCA_PEDIT_KEY_EX_CMD,
856 p_parser.keys_ex[i].cmd);
857 mnl_attr_nest_end(nl, na_pedit_key);
859 mnl_attr_nest_end(nl, na_pedit_keys);
860 mnl_attr_nest_end(nl, na_act_options);
865 * Calculate max memory size of one TC-pedit actions.
866 * One TC-pedit action can contain set of keys each defining
867 * a rewrite element (rte_flow action)
869 * @param[in,out] actions
870 * actions specification.
871 * @param[in,out] action_flags
873 * @param[in,out] size
876 * Max memory size of one TC-pedit action
879 flow_tcf_get_pedit_actions_size(const struct rte_flow_action **actions,
880 uint64_t *action_flags)
886 pedit_size += SZ_NLATTR_NEST + /* na_act_index. */
887 SZ_NLATTR_STRZ_OF("pedit") +
888 SZ_NLATTR_NEST; /* TCA_ACT_OPTIONS. */
889 for (; (*actions)->type != RTE_FLOW_ACTION_TYPE_END; (*actions)++) {
890 switch ((*actions)->type) {
891 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
892 keys += NUM_OF_PEDIT_KEYS(IPV4_ADDR_LEN);
893 flags |= MLX5_FLOW_ACTION_SET_IPV4_SRC;
895 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
896 keys += NUM_OF_PEDIT_KEYS(IPV4_ADDR_LEN);
897 flags |= MLX5_FLOW_ACTION_SET_IPV4_DST;
899 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
900 keys += NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
901 flags |= MLX5_FLOW_ACTION_SET_IPV6_SRC;
903 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
904 keys += NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
905 flags |= MLX5_FLOW_ACTION_SET_IPV6_DST;
907 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
908 /* TCP is as same as UDP */
909 keys += NUM_OF_PEDIT_KEYS(TP_PORT_LEN);
910 flags |= MLX5_FLOW_ACTION_SET_TP_SRC;
912 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
913 /* TCP is as same as UDP */
914 keys += NUM_OF_PEDIT_KEYS(TP_PORT_LEN);
915 flags |= MLX5_FLOW_ACTION_SET_TP_DST;
917 case RTE_FLOW_ACTION_TYPE_SET_TTL:
918 keys += NUM_OF_PEDIT_KEYS(TTL_LEN);
919 flags |= MLX5_FLOW_ACTION_SET_TTL;
921 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
922 keys += NUM_OF_PEDIT_KEYS(TTL_LEN);
923 flags |= MLX5_FLOW_ACTION_DEC_TTL;
925 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
926 keys += NUM_OF_PEDIT_KEYS(ETHER_ADDR_LEN);
927 flags |= MLX5_FLOW_ACTION_SET_MAC_SRC;
929 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
930 keys += NUM_OF_PEDIT_KEYS(ETHER_ADDR_LEN);
931 flags |= MLX5_FLOW_ACTION_SET_MAC_DST;
934 goto get_pedit_action_size_done;
937 get_pedit_action_size_done:
938 /* TCA_PEDIT_PARAMS_EX */
940 SZ_NLATTR_DATA_OF(sizeof(struct tc_pedit_sel) +
941 keys * sizeof(struct tc_pedit_key));
942 pedit_size += SZ_NLATTR_NEST; /* TCA_PEDIT_KEYS */
944 /* TCA_PEDIT_KEY_EX + HTYPE + CMD */
945 (SZ_NLATTR_NEST + SZ_NLATTR_DATA_OF(2) +
946 SZ_NLATTR_DATA_OF(2));
947 (*action_flags) |= flags;
953 * Retrieve mask for pattern item.
955 * This function does basic sanity checks on a pattern item in order to
956 * return the most appropriate mask for it.
959 * Item specification.
960 * @param[in] mask_default
961 * Default mask for pattern item as specified by the flow API.
962 * @param[in] mask_supported
963 * Mask fields supported by the implementation.
964 * @param[in] mask_empty
965 * Empty mask to return when there is no specification.
967 * Perform verbose error reporting if not NULL.
970 * Either @p item->mask or one of the mask parameters on success, NULL
971 * otherwise and rte_errno is set.
974 flow_tcf_item_mask(const struct rte_flow_item *item, const void *mask_default,
975 const void *mask_supported, const void *mask_empty,
976 size_t mask_size, struct rte_flow_error *error)
981 /* item->last and item->mask cannot exist without item->spec. */
982 if (!item->spec && (item->mask || item->last)) {
983 rte_flow_error_set(error, EINVAL,
984 RTE_FLOW_ERROR_TYPE_ITEM, item,
985 "\"mask\" or \"last\" field provided without"
986 " a corresponding \"spec\"");
989 /* No spec, no mask, no problem. */
992 mask = item->mask ? item->mask : mask_default;
995 * Single-pass check to make sure that:
996 * - Mask is supported, no bits are set outside mask_supported.
997 * - Both item->spec and item->last are included in mask.
999 for (i = 0; i != mask_size; ++i) {
1002 if ((mask[i] | ((const uint8_t *)mask_supported)[i]) !=
1003 ((const uint8_t *)mask_supported)[i]) {
1004 rte_flow_error_set(error, ENOTSUP,
1005 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1006 "unsupported field found"
1011 (((const uint8_t *)item->spec)[i] & mask[i]) !=
1012 (((const uint8_t *)item->last)[i] & mask[i])) {
1013 rte_flow_error_set(error, EINVAL,
1014 RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1016 "range between \"spec\" and \"last\""
1017 " not comprised in \"mask\"");
1025 * Build a conversion table between port ID and ifindex.
1028 * Pointer to Ethernet device.
1030 * Pointer to ptoi table.
1032 * Size of ptoi table provided.
1035 * Size of ptoi table filled.
1038 flow_tcf_build_ptoi_table(struct rte_eth_dev *dev, struct flow_tcf_ptoi *ptoi,
1041 unsigned int n = mlx5_dev_to_port_id(dev->device, NULL, 0);
1042 uint16_t port_id[n + 1];
1044 unsigned int own = 0;
1046 /* At least one port is needed when no switch domain is present. */
1049 port_id[0] = dev->data->port_id;
1051 n = RTE_MIN(mlx5_dev_to_port_id(dev->device, port_id, n), n);
1055 for (i = 0; i != n; ++i) {
1056 struct rte_eth_dev_info dev_info;
1058 rte_eth_dev_info_get(port_id[i], &dev_info);
1059 if (port_id[i] == dev->data->port_id)
1061 ptoi[i].port_id = port_id[i];
1062 ptoi[i].ifindex = dev_info.if_index;
1064 /* Ensure first entry of ptoi[] is the current device. */
1067 ptoi[0] = ptoi[own];
1068 ptoi[own] = ptoi[n];
1070 /* An entry with zero ifindex terminates ptoi[]. */
1071 ptoi[n].port_id = 0;
1072 ptoi[n].ifindex = 0;
1077 * Verify the @p attr will be correctly understood by the E-switch.
1080 * Pointer to flow attributes
1082 * Pointer to error structure.
1085 * 0 on success, a negative errno value otherwise and rte_errno is set.
1088 flow_tcf_validate_attributes(const struct rte_flow_attr *attr,
1089 struct rte_flow_error *error)
1092 * Supported attributes: groups, some priorities and ingress only.
1093 * group is supported only if kernel supports chain. Don't care about
1094 * transfer as it is the caller's problem.
1096 if (attr->group > MLX5_TCF_GROUP_ID_MAX)
1097 return rte_flow_error_set(error, ENOTSUP,
1098 RTE_FLOW_ERROR_TYPE_ATTR_GROUP, attr,
1099 "group ID larger than "
1100 RTE_STR(MLX5_TCF_GROUP_ID_MAX)
1101 " isn't supported");
1102 else if (attr->priority > MLX5_TCF_GROUP_PRIORITY_MAX)
1103 return rte_flow_error_set(error, ENOTSUP,
1104 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1106 "priority more than "
1107 RTE_STR(MLX5_TCF_GROUP_PRIORITY_MAX)
1108 " is not supported");
1110 return rte_flow_error_set(error, EINVAL,
1111 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1112 attr, "only ingress is supported");
1114 return rte_flow_error_set(error, ENOTSUP,
1115 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1116 attr, "egress is not supported");
1121 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_ETH item for E-Switch.
1122 * The routine checks the L2 fields to be used in encapsulation header.
1125 * Pointer to the item structure.
1127 * Pointer to the error structure.
1130 * 0 on success, a negative errno value otherwise and rte_errno is set.
1133 flow_tcf_validate_vxlan_encap_eth(const struct rte_flow_item *item,
1134 struct rte_flow_error *error)
1136 const struct rte_flow_item_eth *spec = item->spec;
1137 const struct rte_flow_item_eth *mask = item->mask;
1141 * Specification for L2 addresses can be empty
1142 * because these ones are optional and not
1143 * required directly by tc rule. Kernel tries
1144 * to resolve these ones on its own
1149 /* If mask is not specified use the default one. */
1150 mask = &rte_flow_item_eth_mask;
1152 if (memcmp(&mask->dst,
1153 &flow_tcf_mask_empty.eth.dst,
1154 sizeof(flow_tcf_mask_empty.eth.dst))) {
1155 if (memcmp(&mask->dst,
1156 &rte_flow_item_eth_mask.dst,
1157 sizeof(rte_flow_item_eth_mask.dst)))
1158 return rte_flow_error_set
1160 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1161 "no support for partial mask on"
1162 " \"eth.dst\" field");
1164 if (memcmp(&mask->src,
1165 &flow_tcf_mask_empty.eth.src,
1166 sizeof(flow_tcf_mask_empty.eth.src))) {
1167 if (memcmp(&mask->src,
1168 &rte_flow_item_eth_mask.src,
1169 sizeof(rte_flow_item_eth_mask.src)))
1170 return rte_flow_error_set
1172 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1173 "no support for partial mask on"
1174 " \"eth.src\" field");
1176 if (mask->type != RTE_BE16(0x0000)) {
1177 if (mask->type != RTE_BE16(0xffff))
1178 return rte_flow_error_set
1180 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1181 "no support for partial mask on"
1182 " \"eth.type\" field");
1184 "outer ethernet type field"
1185 " cannot be forced for vxlan"
1186 " encapsulation, parameter ignored");
1192 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_IPV4 item for E-Switch.
1193 * The routine checks the IPv4 fields to be used in encapsulation header.
1196 * Pointer to the item structure.
1198 * Pointer to the error structure.
1201 * 0 on success, a negative errno value otherwise and rte_errno is set.
1204 flow_tcf_validate_vxlan_encap_ipv4(const struct rte_flow_item *item,
1205 struct rte_flow_error *error)
1207 const struct rte_flow_item_ipv4 *spec = item->spec;
1208 const struct rte_flow_item_ipv4 *mask = item->mask;
1212 * Specification for IP addresses cannot be empty
1213 * because it is required by tunnel_key parameter.
1215 return rte_flow_error_set(error, EINVAL,
1216 RTE_FLOW_ERROR_TYPE_ITEM, item,
1217 "NULL outer ipv4 address"
1218 " specification for vxlan"
1222 mask = &rte_flow_item_ipv4_mask;
1223 if (mask->hdr.dst_addr != RTE_BE32(0x00000000)) {
1224 if (mask->hdr.dst_addr != RTE_BE32(0xffffffff))
1225 return rte_flow_error_set
1227 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1228 "no support for partial mask on"
1229 " \"ipv4.hdr.dst_addr\" field"
1230 " for vxlan encapsulation");
1231 /* More IPv4 address validations can be put here. */
1234 * Kernel uses the destination IP address to determine
1235 * the routing path and obtain the MAC destination
1236 * address, so IP destination address must be
1237 * specified in the tc rule.
1239 return rte_flow_error_set(error, EINVAL,
1240 RTE_FLOW_ERROR_TYPE_ITEM, item,
1241 "outer ipv4 destination address"
1242 " must be specified for"
1243 " vxlan encapsulation");
1245 if (mask->hdr.src_addr != RTE_BE32(0x00000000)) {
1246 if (mask->hdr.src_addr != RTE_BE32(0xffffffff))
1247 return rte_flow_error_set
1249 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1250 "no support for partial mask on"
1251 " \"ipv4.hdr.src_addr\" field"
1252 " for vxlan encapsulation");
1253 /* More IPv4 address validations can be put here. */
1256 * Kernel uses the source IP address to select the
1257 * interface for egress encapsulated traffic, so
1258 * it must be specified in the tc rule.
1260 return rte_flow_error_set(error, EINVAL,
1261 RTE_FLOW_ERROR_TYPE_ITEM, item,
1262 "outer ipv4 source address"
1263 " must be specified for"
1264 " vxlan encapsulation");
1270 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_IPV6 item for E-Switch.
1271 * The routine checks the IPv6 fields to be used in encapsulation header.
1274 * Pointer to the item structure.
1276 * Pointer to the error structure.
1279 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1282 flow_tcf_validate_vxlan_encap_ipv6(const struct rte_flow_item *item,
1283 struct rte_flow_error *error)
1285 const struct rte_flow_item_ipv6 *spec = item->spec;
1286 const struct rte_flow_item_ipv6 *mask = item->mask;
1290 * Specification for IP addresses cannot be empty
1291 * because it is required by tunnel_key parameter.
1293 return rte_flow_error_set(error, EINVAL,
1294 RTE_FLOW_ERROR_TYPE_ITEM, item,
1295 "NULL outer ipv6 address"
1296 " specification for"
1297 " vxlan encapsulation");
1300 mask = &rte_flow_item_ipv6_mask;
1301 if (memcmp(&mask->hdr.dst_addr,
1302 &flow_tcf_mask_empty.ipv6.hdr.dst_addr,
1304 if (memcmp(&mask->hdr.dst_addr,
1305 &rte_flow_item_ipv6_mask.hdr.dst_addr,
1307 return rte_flow_error_set
1309 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1310 "no support for partial mask on"
1311 " \"ipv6.hdr.dst_addr\" field"
1312 " for vxlan encapsulation");
1313 /* More IPv6 address validations can be put here. */
1316 * Kernel uses the destination IP address to determine
1317 * the routing path and obtain the MAC destination
1318 * address (heigh or gate), so IP destination address
1319 * must be specified within the tc rule.
1321 return rte_flow_error_set(error, EINVAL,
1322 RTE_FLOW_ERROR_TYPE_ITEM, item,
1323 "outer ipv6 destination address"
1324 " must be specified for"
1325 " vxlan encapsulation");
1327 if (memcmp(&mask->hdr.src_addr,
1328 &flow_tcf_mask_empty.ipv6.hdr.src_addr,
1330 if (memcmp(&mask->hdr.src_addr,
1331 &rte_flow_item_ipv6_mask.hdr.src_addr,
1333 return rte_flow_error_set
1335 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1336 "no support for partial mask on"
1337 " \"ipv6.hdr.src_addr\" field"
1338 " for vxlan encapsulation");
1339 /* More L3 address validation can be put here. */
1342 * Kernel uses the source IP address to select the
1343 * interface for egress encapsulated traffic, so
1344 * it must be specified in the tc rule.
1346 return rte_flow_error_set(error, EINVAL,
1347 RTE_FLOW_ERROR_TYPE_ITEM, item,
1348 "outer L3 source address"
1349 " must be specified for"
1350 " vxlan encapsulation");
1356 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_UDP item for E-Switch.
1357 * The routine checks the UDP fields to be used in encapsulation header.
1360 * Pointer to the item structure.
1362 * Pointer to the error structure.
1365 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1368 flow_tcf_validate_vxlan_encap_udp(const struct rte_flow_item *item,
1369 struct rte_flow_error *error)
1371 const struct rte_flow_item_udp *spec = item->spec;
1372 const struct rte_flow_item_udp *mask = item->mask;
1376 * Specification for UDP ports cannot be empty
1377 * because it is required by tunnel_key parameter.
1379 return rte_flow_error_set(error, EINVAL,
1380 RTE_FLOW_ERROR_TYPE_ITEM, item,
1381 "NULL UDP port specification "
1382 " for vxlan encapsulation");
1385 mask = &rte_flow_item_udp_mask;
1386 if (mask->hdr.dst_port != RTE_BE16(0x0000)) {
1387 if (mask->hdr.dst_port != RTE_BE16(0xffff))
1388 return rte_flow_error_set
1390 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1391 "no support for partial mask on"
1392 " \"udp.hdr.dst_port\" field"
1393 " for vxlan encapsulation");
1394 if (!spec->hdr.dst_port)
1395 return rte_flow_error_set
1397 RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 "outer UDP remote port cannot be"
1399 " 0 for vxlan encapsulation");
1401 return rte_flow_error_set(error, EINVAL,
1402 RTE_FLOW_ERROR_TYPE_ITEM, item,
1403 "outer UDP remote port"
1404 " must be specified for"
1405 " vxlan encapsulation");
1407 if (mask->hdr.src_port != RTE_BE16(0x0000)) {
1408 if (mask->hdr.src_port != RTE_BE16(0xffff))
1409 return rte_flow_error_set
1411 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1412 "no support for partial mask on"
1413 " \"udp.hdr.src_port\" field"
1414 " for vxlan encapsulation");
1416 "outer UDP source port cannot be"
1417 " forced for vxlan encapsulation,"
1418 " parameter ignored");
1424 * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_VXLAN item for E-Switch.
1425 * The routine checks the VNIP fields to be used in encapsulation header.
1428 * Pointer to the item structure.
1430 * Pointer to the error structure.
1433 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1436 flow_tcf_validate_vxlan_encap_vni(const struct rte_flow_item *item,
1437 struct rte_flow_error *error)
1439 const struct rte_flow_item_vxlan *spec = item->spec;
1440 const struct rte_flow_item_vxlan *mask = item->mask;
1443 /* Outer VNI is required by tunnel_key parameter. */
1444 return rte_flow_error_set(error, EINVAL,
1445 RTE_FLOW_ERROR_TYPE_ITEM, item,
1446 "NULL VNI specification"
1447 " for vxlan encapsulation");
1450 mask = &rte_flow_item_vxlan_mask;
1451 if (!mask->vni[0] && !mask->vni[1] && !mask->vni[2])
1452 return rte_flow_error_set(error, EINVAL,
1453 RTE_FLOW_ERROR_TYPE_ITEM, item,
1454 "outer VNI must be specified "
1455 "for vxlan encapsulation");
1456 if (mask->vni[0] != 0xff ||
1457 mask->vni[1] != 0xff ||
1458 mask->vni[2] != 0xff)
1459 return rte_flow_error_set(error, ENOTSUP,
1460 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1461 "no support for partial mask on"
1462 " \"vxlan.vni\" field");
1464 if (!spec->vni[0] && !spec->vni[1] && !spec->vni[2])
1465 return rte_flow_error_set(error, EINVAL,
1466 RTE_FLOW_ERROR_TYPE_ITEM, item,
1467 "vxlan vni cannot be 0");
1472 * Validate VXLAN_ENCAP action item list for E-Switch.
1473 * The routine checks items to be used in encapsulation header.
1476 * Pointer to the VXLAN_ENCAP action structure.
1478 * Pointer to the error structure.
1481 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1484 flow_tcf_validate_vxlan_encap(const struct rte_flow_action *action,
1485 struct rte_flow_error *error)
1487 const struct rte_flow_item *items;
1489 uint32_t item_flags = 0;
1492 return rte_flow_error_set(error, EINVAL,
1493 RTE_FLOW_ERROR_TYPE_ACTION, action,
1494 "Missing vxlan tunnel"
1495 " action configuration");
1496 items = ((const struct rte_flow_action_vxlan_encap *)
1497 action->conf)->definition;
1499 return rte_flow_error_set(error, EINVAL,
1500 RTE_FLOW_ERROR_TYPE_ACTION, action,
1501 "Missing vxlan tunnel"
1502 " encapsulation parameters");
1503 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1504 switch (items->type) {
1505 case RTE_FLOW_ITEM_TYPE_VOID:
1507 case RTE_FLOW_ITEM_TYPE_ETH:
1508 ret = mlx5_flow_validate_item_eth(items, item_flags,
1512 ret = flow_tcf_validate_vxlan_encap_eth(items, error);
1515 item_flags |= MLX5_FLOW_LAYER_OUTER_L2;
1518 case RTE_FLOW_ITEM_TYPE_IPV4:
1519 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
1523 ret = flow_tcf_validate_vxlan_encap_ipv4(items, error);
1526 item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1528 case RTE_FLOW_ITEM_TYPE_IPV6:
1529 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
1533 ret = flow_tcf_validate_vxlan_encap_ipv6(items, error);
1536 item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1538 case RTE_FLOW_ITEM_TYPE_UDP:
1539 ret = mlx5_flow_validate_item_udp(items, item_flags,
1543 ret = flow_tcf_validate_vxlan_encap_udp(items, error);
1546 item_flags |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
1548 case RTE_FLOW_ITEM_TYPE_VXLAN:
1549 ret = mlx5_flow_validate_item_vxlan(items,
1553 ret = flow_tcf_validate_vxlan_encap_vni(items, error);
1556 item_flags |= MLX5_FLOW_LAYER_VXLAN;
1559 return rte_flow_error_set
1561 RTE_FLOW_ERROR_TYPE_ITEM, items,
1562 "vxlan encap item not supported");
1565 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1566 return rte_flow_error_set(error, EINVAL,
1567 RTE_FLOW_ERROR_TYPE_ACTION, action,
1568 "no outer IP layer found"
1569 " for vxlan encapsulation");
1570 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1571 return rte_flow_error_set(error, EINVAL,
1572 RTE_FLOW_ERROR_TYPE_ACTION, action,
1573 "no outer UDP layer found"
1574 " for vxlan encapsulation");
1575 if (!(item_flags & MLX5_FLOW_LAYER_VXLAN))
1576 return rte_flow_error_set(error, EINVAL,
1577 RTE_FLOW_ERROR_TYPE_ACTION, action,
1578 "no VXLAN VNI found"
1579 " for vxlan encapsulation");
1584 * Validate RTE_FLOW_ITEM_TYPE_IPV4 item if VXLAN_DECAP action
1585 * is present in actions list.
1588 * Outer IPv4 address item (if any, NULL otherwise).
1590 * Pointer to the error structure.
1593 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1596 flow_tcf_validate_vxlan_decap_ipv4(const struct rte_flow_item *ipv4,
1597 struct rte_flow_error *error)
1599 const struct rte_flow_item_ipv4 *spec = ipv4->spec;
1600 const struct rte_flow_item_ipv4 *mask = ipv4->mask;
1604 * Specification for IP addresses cannot be empty
1605 * because it is required as decap parameter.
1607 return rte_flow_error_set(error, EINVAL,
1608 RTE_FLOW_ERROR_TYPE_ITEM, ipv4,
1609 "NULL outer ipv4 address"
1610 " specification for vxlan"
1611 " for vxlan decapsulation");
1614 mask = &rte_flow_item_ipv4_mask;
1615 if (mask->hdr.dst_addr != RTE_BE32(0x00000000)) {
1616 if (mask->hdr.dst_addr != RTE_BE32(0xffffffff))
1617 return rte_flow_error_set
1619 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1620 "no support for partial mask on"
1621 " \"ipv4.hdr.dst_addr\" field");
1622 /* More IP address validations can be put here. */
1625 * Kernel uses the destination IP address
1626 * to determine the ingress network interface
1627 * for traffic being decapsulated.
1629 return rte_flow_error_set(error, EINVAL,
1630 RTE_FLOW_ERROR_TYPE_ITEM, ipv4,
1631 "outer ipv4 destination address"
1632 " must be specified for"
1633 " vxlan decapsulation");
1635 /* Source IP address is optional for decap. */
1636 if (mask->hdr.src_addr != RTE_BE32(0x00000000) &&
1637 mask->hdr.src_addr != RTE_BE32(0xffffffff))
1638 return rte_flow_error_set(error, ENOTSUP,
1639 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1640 "no support for partial mask on"
1641 " \"ipv4.hdr.src_addr\" field");
1646 * Validate RTE_FLOW_ITEM_TYPE_IPV6 item if VXLAN_DECAP action
1647 * is present in actions list.
1650 * Outer IPv6 address item (if any, NULL otherwise).
1652 * Pointer to the error structure.
1655 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1658 flow_tcf_validate_vxlan_decap_ipv6(const struct rte_flow_item *ipv6,
1659 struct rte_flow_error *error)
1661 const struct rte_flow_item_ipv6 *spec = ipv6->spec;
1662 const struct rte_flow_item_ipv6 *mask = ipv6->mask;
1666 * Specification for IP addresses cannot be empty
1667 * because it is required as decap parameter.
1669 return rte_flow_error_set(error, EINVAL,
1670 RTE_FLOW_ERROR_TYPE_ITEM, ipv6,
1671 "NULL outer ipv6 address"
1672 " specification for vxlan"
1676 mask = &rte_flow_item_ipv6_mask;
1677 if (memcmp(&mask->hdr.dst_addr,
1678 &flow_tcf_mask_empty.ipv6.hdr.dst_addr,
1680 if (memcmp(&mask->hdr.dst_addr,
1681 &rte_flow_item_ipv6_mask.hdr.dst_addr,
1683 return rte_flow_error_set
1685 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1686 "no support for partial mask on"
1687 " \"ipv6.hdr.dst_addr\" field");
1688 /* More IP address validations can be put here. */
1691 * Kernel uses the destination IP address
1692 * to determine the ingress network interface
1693 * for traffic being decapsulated.
1695 return rte_flow_error_set(error, EINVAL,
1696 RTE_FLOW_ERROR_TYPE_ITEM, ipv6,
1697 "outer ipv6 destination address must be "
1698 "specified for vxlan decapsulation");
1700 /* Source IP address is optional for decap. */
1701 if (memcmp(&mask->hdr.src_addr,
1702 &flow_tcf_mask_empty.ipv6.hdr.src_addr,
1704 if (memcmp(&mask->hdr.src_addr,
1705 &rte_flow_item_ipv6_mask.hdr.src_addr,
1707 return rte_flow_error_set
1709 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1710 "no support for partial mask on"
1711 " \"ipv6.hdr.src_addr\" field");
1717 * Validate RTE_FLOW_ITEM_TYPE_UDP item if VXLAN_DECAP action
1718 * is present in actions list.
1721 * Outer UDP layer item (if any, NULL otherwise).
1723 * Pointer to the error structure.
1726 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1729 flow_tcf_validate_vxlan_decap_udp(const struct rte_flow_item *udp,
1730 struct rte_flow_error *error)
1732 const struct rte_flow_item_udp *spec = udp->spec;
1733 const struct rte_flow_item_udp *mask = udp->mask;
1737 * Specification for UDP ports cannot be empty
1738 * because it is required as decap parameter.
1740 return rte_flow_error_set(error, EINVAL,
1741 RTE_FLOW_ERROR_TYPE_ITEM, udp,
1742 "NULL UDP port specification"
1743 " for VXLAN decapsulation");
1745 mask = &rte_flow_item_udp_mask;
1746 if (mask->hdr.dst_port != RTE_BE16(0x0000)) {
1747 if (mask->hdr.dst_port != RTE_BE16(0xffff))
1748 return rte_flow_error_set
1750 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1751 "no support for partial mask on"
1752 " \"udp.hdr.dst_port\" field");
1753 if (!spec->hdr.dst_port)
1754 return rte_flow_error_set
1756 RTE_FLOW_ERROR_TYPE_ITEM, udp,
1757 "zero decap local UDP port");
1759 return rte_flow_error_set(error, EINVAL,
1760 RTE_FLOW_ERROR_TYPE_ITEM, udp,
1761 "outer UDP destination port must be "
1762 "specified for vxlan decapsulation");
1764 if (mask->hdr.src_port != RTE_BE16(0x0000)) {
1765 if (mask->hdr.src_port != RTE_BE16(0xffff))
1766 return rte_flow_error_set
1768 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1769 "no support for partial mask on"
1770 " \"udp.hdr.src_port\" field");
1772 "outer UDP local port cannot be "
1773 "forced for VXLAN encapsulation, "
1774 "parameter ignored");
1780 * Validate flow for E-Switch.
1783 * Pointer to the priv structure.
1785 * Pointer to the flow attributes.
1787 * Pointer to the list of items.
1788 * @param[in] actions
1789 * Pointer to the list of actions.
1791 * Pointer to the error structure.
1794 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1797 flow_tcf_validate(struct rte_eth_dev *dev,
1798 const struct rte_flow_attr *attr,
1799 const struct rte_flow_item items[],
1800 const struct rte_flow_action actions[],
1801 struct rte_flow_error *error)
1804 const struct rte_flow_item_port_id *port_id;
1805 const struct rte_flow_item_eth *eth;
1806 const struct rte_flow_item_vlan *vlan;
1807 const struct rte_flow_item_ipv4 *ipv4;
1808 const struct rte_flow_item_ipv6 *ipv6;
1809 const struct rte_flow_item_tcp *tcp;
1810 const struct rte_flow_item_udp *udp;
1811 const struct rte_flow_item_vxlan *vxlan;
1814 const struct rte_flow_action_port_id *port_id;
1815 const struct rte_flow_action_jump *jump;
1816 const struct rte_flow_action_of_push_vlan *of_push_vlan;
1817 const struct rte_flow_action_of_set_vlan_vid *
1819 const struct rte_flow_action_of_set_vlan_pcp *
1821 const struct rte_flow_action_vxlan_encap *vxlan_encap;
1822 const struct rte_flow_action_set_ipv4 *set_ipv4;
1823 const struct rte_flow_action_set_ipv6 *set_ipv6;
1825 uint64_t item_flags = 0;
1826 uint64_t action_flags = 0;
1827 uint8_t next_protocol = -1;
1828 unsigned int tcm_ifindex = 0;
1829 uint8_t pedit_validated = 0;
1830 struct flow_tcf_ptoi ptoi[PTOI_TABLE_SZ_MAX(dev)];
1831 struct rte_eth_dev *port_id_dev = NULL;
1832 bool in_port_id_set;
1835 claim_nonzero(flow_tcf_build_ptoi_table(dev, ptoi,
1836 PTOI_TABLE_SZ_MAX(dev)));
1837 ret = flow_tcf_validate_attributes(attr, error);
1840 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1842 uint64_t current_action_flag = 0;
1844 switch (actions->type) {
1845 case RTE_FLOW_ACTION_TYPE_VOID:
1847 case RTE_FLOW_ACTION_TYPE_PORT_ID:
1848 current_action_flag = MLX5_FLOW_ACTION_PORT_ID;
1851 conf.port_id = actions->conf;
1852 if (conf.port_id->original)
1855 for (i = 0; ptoi[i].ifindex; ++i)
1856 if (ptoi[i].port_id == conf.port_id->id)
1858 if (!ptoi[i].ifindex)
1859 return rte_flow_error_set
1861 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1863 "missing data to convert port ID to"
1865 port_id_dev = &rte_eth_devices[conf.port_id->id];
1867 case RTE_FLOW_ACTION_TYPE_JUMP:
1868 current_action_flag = MLX5_FLOW_ACTION_JUMP;
1871 conf.jump = actions->conf;
1872 if (attr->group >= conf.jump->group)
1873 return rte_flow_error_set
1875 RTE_FLOW_ERROR_TYPE_ACTION,
1877 "can jump only to a group forward");
1879 case RTE_FLOW_ACTION_TYPE_DROP:
1880 current_action_flag = MLX5_FLOW_ACTION_DROP;
1882 case RTE_FLOW_ACTION_TYPE_COUNT:
1884 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
1885 current_action_flag = MLX5_FLOW_ACTION_OF_POP_VLAN;
1887 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
1888 current_action_flag = MLX5_FLOW_ACTION_OF_PUSH_VLAN;
1890 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
1891 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
1892 return rte_flow_error_set
1894 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1895 "vlan modify is not supported,"
1896 " set action must follow push action");
1897 current_action_flag = MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
1899 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
1900 if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
1901 return rte_flow_error_set
1903 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1904 "vlan modify is not supported,"
1905 " set action must follow push action");
1906 current_action_flag = MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
1908 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
1909 current_action_flag = MLX5_FLOW_ACTION_VXLAN_DECAP;
1911 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
1912 ret = flow_tcf_validate_vxlan_encap(actions, error);
1915 current_action_flag = MLX5_FLOW_ACTION_VXLAN_ENCAP;
1917 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
1918 current_action_flag = MLX5_FLOW_ACTION_SET_IPV4_SRC;
1920 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
1921 current_action_flag = MLX5_FLOW_ACTION_SET_IPV4_DST;
1923 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
1924 current_action_flag = MLX5_FLOW_ACTION_SET_IPV6_SRC;
1926 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
1927 current_action_flag = MLX5_FLOW_ACTION_SET_IPV6_DST;
1929 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
1930 current_action_flag = MLX5_FLOW_ACTION_SET_TP_SRC;
1932 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
1933 current_action_flag = MLX5_FLOW_ACTION_SET_TP_DST;
1935 case RTE_FLOW_ACTION_TYPE_SET_TTL:
1936 current_action_flag = MLX5_FLOW_ACTION_SET_TTL;
1938 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
1939 current_action_flag = MLX5_FLOW_ACTION_DEC_TTL;
1941 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
1942 current_action_flag = MLX5_FLOW_ACTION_SET_MAC_SRC;
1944 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
1945 current_action_flag = MLX5_FLOW_ACTION_SET_MAC_DST;
1948 return rte_flow_error_set(error, ENOTSUP,
1949 RTE_FLOW_ERROR_TYPE_ACTION,
1951 "action not supported");
1953 if (current_action_flag & MLX5_TCF_CONFIG_ACTIONS) {
1955 return rte_flow_error_set
1957 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1959 "action configuration not set");
1961 if ((current_action_flag & MLX5_TCF_PEDIT_ACTIONS) &&
1963 return rte_flow_error_set(error, ENOTSUP,
1964 RTE_FLOW_ERROR_TYPE_ACTION,
1966 "set actions should be "
1967 "listed successively");
1968 if ((current_action_flag & ~MLX5_TCF_PEDIT_ACTIONS) &&
1969 (action_flags & MLX5_TCF_PEDIT_ACTIONS))
1970 pedit_validated = 1;
1971 if ((current_action_flag & MLX5_TCF_FATE_ACTIONS) &&
1972 (action_flags & MLX5_TCF_FATE_ACTIONS))
1973 return rte_flow_error_set(error, EINVAL,
1974 RTE_FLOW_ERROR_TYPE_ACTION,
1976 "can't have multiple fate"
1978 if ((current_action_flag & MLX5_TCF_VXLAN_ACTIONS) &&
1979 (action_flags & MLX5_TCF_VXLAN_ACTIONS))
1980 return rte_flow_error_set(error, EINVAL,
1981 RTE_FLOW_ERROR_TYPE_ACTION,
1983 "can't have multiple vxlan"
1985 if ((current_action_flag & MLX5_TCF_VXLAN_ACTIONS) &&
1986 (action_flags & MLX5_TCF_VLAN_ACTIONS))
1987 return rte_flow_error_set(error, ENOTSUP,
1988 RTE_FLOW_ERROR_TYPE_ACTION,
1990 "can't have vxlan and vlan"
1991 " actions in the same rule");
1992 action_flags |= current_action_flag;
1994 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1997 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1998 items->type != RTE_FLOW_ITEM_TYPE_ETH)
1999 return rte_flow_error_set(error, ENOTSUP,
2000 RTE_FLOW_ERROR_TYPE_ITEM,
2002 "only L2 inner item"
2004 switch (items->type) {
2005 case RTE_FLOW_ITEM_TYPE_VOID:
2007 case RTE_FLOW_ITEM_TYPE_PORT_ID:
2008 mask.port_id = flow_tcf_item_mask
2009 (items, &rte_flow_item_port_id_mask,
2010 &flow_tcf_mask_supported.port_id,
2011 &flow_tcf_mask_empty.port_id,
2012 sizeof(flow_tcf_mask_supported.port_id),
2016 if (mask.port_id == &flow_tcf_mask_empty.port_id) {
2020 spec.port_id = items->spec;
2021 if (mask.port_id->id && mask.port_id->id != 0xffffffff)
2022 return rte_flow_error_set
2024 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2026 "no support for partial mask on"
2028 if (!mask.port_id->id)
2031 for (i = 0; ptoi[i].ifindex; ++i)
2032 if (ptoi[i].port_id == spec.port_id->id)
2034 if (!ptoi[i].ifindex)
2035 return rte_flow_error_set
2037 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2039 "missing data to convert port ID to"
2041 if (in_port_id_set && ptoi[i].ifindex != tcm_ifindex)
2042 return rte_flow_error_set
2044 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2046 "cannot match traffic for"
2047 " several port IDs through"
2048 " a single flow rule");
2049 tcm_ifindex = ptoi[i].ifindex;
2052 case RTE_FLOW_ITEM_TYPE_ETH:
2053 ret = mlx5_flow_validate_item_eth(items, item_flags,
2057 item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
2058 MLX5_FLOW_LAYER_INNER_L2 :
2059 MLX5_FLOW_LAYER_OUTER_L2;
2061 * Redundant check due to different supported mask.
2062 * Same for the rest of items.
2064 mask.eth = flow_tcf_item_mask
2065 (items, &rte_flow_item_eth_mask,
2066 &flow_tcf_mask_supported.eth,
2067 &flow_tcf_mask_empty.eth,
2068 sizeof(flow_tcf_mask_supported.eth),
2072 if (mask.eth->type && mask.eth->type !=
2074 return rte_flow_error_set
2076 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2078 "no support for partial mask on"
2081 case RTE_FLOW_ITEM_TYPE_VLAN:
2082 ret = mlx5_flow_validate_item_vlan(items, item_flags,
2086 item_flags |= MLX5_FLOW_LAYER_OUTER_VLAN;
2087 mask.vlan = flow_tcf_item_mask
2088 (items, &rte_flow_item_vlan_mask,
2089 &flow_tcf_mask_supported.vlan,
2090 &flow_tcf_mask_empty.vlan,
2091 sizeof(flow_tcf_mask_supported.vlan),
2095 if ((mask.vlan->tci & RTE_BE16(0xe000) &&
2096 (mask.vlan->tci & RTE_BE16(0xe000)) !=
2097 RTE_BE16(0xe000)) ||
2098 (mask.vlan->tci & RTE_BE16(0x0fff) &&
2099 (mask.vlan->tci & RTE_BE16(0x0fff)) !=
2100 RTE_BE16(0x0fff)) ||
2101 (mask.vlan->inner_type &&
2102 mask.vlan->inner_type != RTE_BE16(0xffff)))
2103 return rte_flow_error_set
2105 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2107 "no support for partial masks on"
2108 " \"tci\" (PCP and VID parts) and"
2109 " \"inner_type\" fields");
2111 case RTE_FLOW_ITEM_TYPE_IPV4:
2112 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
2116 item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2117 mask.ipv4 = flow_tcf_item_mask
2118 (items, &rte_flow_item_ipv4_mask,
2119 &flow_tcf_mask_supported.ipv4,
2120 &flow_tcf_mask_empty.ipv4,
2121 sizeof(flow_tcf_mask_supported.ipv4),
2125 if (mask.ipv4->hdr.next_proto_id &&
2126 mask.ipv4->hdr.next_proto_id != 0xff)
2127 return rte_flow_error_set
2129 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2131 "no support for partial mask on"
2132 " \"hdr.next_proto_id\" field");
2133 else if (mask.ipv4->hdr.next_proto_id)
2135 ((const struct rte_flow_item_ipv4 *)
2136 (items->spec))->hdr.next_proto_id;
2137 if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2138 ret = flow_tcf_validate_vxlan_decap_ipv4
2144 case RTE_FLOW_ITEM_TYPE_IPV6:
2145 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
2149 item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2150 mask.ipv6 = flow_tcf_item_mask
2151 (items, &rte_flow_item_ipv6_mask,
2152 &flow_tcf_mask_supported.ipv6,
2153 &flow_tcf_mask_empty.ipv6,
2154 sizeof(flow_tcf_mask_supported.ipv6),
2158 if (mask.ipv6->hdr.proto &&
2159 mask.ipv6->hdr.proto != 0xff)
2160 return rte_flow_error_set
2162 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2164 "no support for partial mask on"
2165 " \"hdr.proto\" field");
2166 else if (mask.ipv6->hdr.proto)
2168 ((const struct rte_flow_item_ipv6 *)
2169 (items->spec))->hdr.proto;
2170 if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2171 ret = flow_tcf_validate_vxlan_decap_ipv6
2177 case RTE_FLOW_ITEM_TYPE_UDP:
2178 ret = mlx5_flow_validate_item_udp(items, item_flags,
2179 next_protocol, error);
2182 item_flags |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2183 mask.udp = flow_tcf_item_mask
2184 (items, &rte_flow_item_udp_mask,
2185 &flow_tcf_mask_supported.udp,
2186 &flow_tcf_mask_empty.udp,
2187 sizeof(flow_tcf_mask_supported.udp),
2191 if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2192 ret = flow_tcf_validate_vxlan_decap_udp
2198 case RTE_FLOW_ITEM_TYPE_TCP:
2199 ret = mlx5_flow_validate_item_tcp
2202 &flow_tcf_mask_supported.tcp,
2206 item_flags |= 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 (!(action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP))
2218 return rte_flow_error_set
2220 RTE_FLOW_ERROR_TYPE_ITEM,
2222 "vni pattern should be followed by"
2223 " vxlan decapsulation action");
2224 ret = mlx5_flow_validate_item_vxlan(items,
2228 item_flags |= MLX5_FLOW_LAYER_VXLAN;
2229 mask.vxlan = flow_tcf_item_mask
2230 (items, &rte_flow_item_vxlan_mask,
2231 &flow_tcf_mask_supported.vxlan,
2232 &flow_tcf_mask_empty.vxlan,
2233 sizeof(flow_tcf_mask_supported.vxlan), error);
2236 if (mask.vxlan->vni[0] != 0xff ||
2237 mask.vxlan->vni[1] != 0xff ||
2238 mask.vxlan->vni[2] != 0xff)
2239 return rte_flow_error_set
2241 RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2243 "no support for partial or "
2244 "empty mask on \"vxlan.vni\" field");
2247 return rte_flow_error_set(error, ENOTSUP,
2248 RTE_FLOW_ERROR_TYPE_ITEM,
2249 items, "item not supported");
2252 if ((action_flags & MLX5_TCF_PEDIT_ACTIONS) &&
2253 (action_flags & MLX5_FLOW_ACTION_DROP))
2254 return rte_flow_error_set(error, ENOTSUP,
2255 RTE_FLOW_ERROR_TYPE_ACTION,
2257 "set action is not compatible with "
2259 if ((action_flags & MLX5_TCF_PEDIT_ACTIONS) &&
2260 !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2261 return rte_flow_error_set(error, ENOTSUP,
2262 RTE_FLOW_ERROR_TYPE_ACTION,
2264 "set action must be followed by "
2267 (MLX5_FLOW_ACTION_SET_IPV4_SRC | MLX5_FLOW_ACTION_SET_IPV4_DST)) {
2268 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4))
2269 return rte_flow_error_set(error, EINVAL,
2270 RTE_FLOW_ERROR_TYPE_ACTION,
2272 "no ipv4 item found in"
2276 (MLX5_FLOW_ACTION_SET_IPV6_SRC | MLX5_FLOW_ACTION_SET_IPV6_DST)) {
2277 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6))
2278 return rte_flow_error_set(error, EINVAL,
2279 RTE_FLOW_ERROR_TYPE_ACTION,
2281 "no ipv6 item found in"
2285 (MLX5_FLOW_ACTION_SET_TP_SRC | MLX5_FLOW_ACTION_SET_TP_DST)) {
2287 (MLX5_FLOW_LAYER_OUTER_L4_UDP |
2288 MLX5_FLOW_LAYER_OUTER_L4_TCP)))
2289 return rte_flow_error_set(error, EINVAL,
2290 RTE_FLOW_ERROR_TYPE_ACTION,
2292 "no TCP/UDP item found in"
2296 * FW syndrome (0xA9C090):
2297 * set_flow_table_entry: push vlan action fte in fdb can ONLY be
2298 * forward to the uplink.
2300 if ((action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2301 (action_flags & MLX5_FLOW_ACTION_PORT_ID) &&
2302 ((struct priv *)port_id_dev->data->dev_private)->representor)
2303 return rte_flow_error_set(error, ENOTSUP,
2304 RTE_FLOW_ERROR_TYPE_ACTION, actions,
2305 "vlan push can only be applied"
2306 " when forwarding to uplink port");
2308 * FW syndrome (0x294609):
2309 * set_flow_table_entry: modify/pop/push actions in fdb flow table
2310 * are supported only while forwarding to vport.
2312 if ((action_flags & MLX5_TCF_VLAN_ACTIONS) &&
2313 !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2314 return rte_flow_error_set(error, ENOTSUP,
2315 RTE_FLOW_ERROR_TYPE_ACTION, actions,
2316 "vlan actions are supported"
2317 " only with port_id action");
2318 if ((action_flags & MLX5_TCF_VXLAN_ACTIONS) &&
2319 !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2320 return rte_flow_error_set(error, ENOTSUP,
2321 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2322 "vxlan actions are supported"
2323 " only with port_id action");
2324 if (!(action_flags & MLX5_TCF_FATE_ACTIONS))
2325 return rte_flow_error_set(error, EINVAL,
2326 RTE_FLOW_ERROR_TYPE_ACTION, actions,
2327 "no fate action is found");
2329 (MLX5_FLOW_ACTION_SET_TTL | MLX5_FLOW_ACTION_DEC_TTL)) {
2331 (MLX5_FLOW_LAYER_OUTER_L3_IPV4 |
2332 MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
2333 return rte_flow_error_set(error, EINVAL,
2334 RTE_FLOW_ERROR_TYPE_ACTION,
2336 "no IP found in pattern");
2339 (MLX5_FLOW_ACTION_SET_MAC_SRC | MLX5_FLOW_ACTION_SET_MAC_DST)) {
2340 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L2))
2341 return rte_flow_error_set(error, ENOTSUP,
2342 RTE_FLOW_ERROR_TYPE_ACTION,
2344 "no ethernet found in"
2347 if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2349 (MLX5_FLOW_LAYER_OUTER_L3_IPV4 |
2350 MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
2351 return rte_flow_error_set(error, EINVAL,
2352 RTE_FLOW_ERROR_TYPE_ACTION,
2354 "no outer IP pattern found"
2355 " for vxlan decap action");
2356 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2357 return rte_flow_error_set(error, EINVAL,
2358 RTE_FLOW_ERROR_TYPE_ACTION,
2360 "no outer UDP pattern found"
2361 " for vxlan decap action");
2362 if (!(item_flags & MLX5_FLOW_LAYER_VXLAN))
2363 return rte_flow_error_set(error, EINVAL,
2364 RTE_FLOW_ERROR_TYPE_ACTION,
2366 "no VNI pattern found"
2367 " for vxlan decap action");
2373 * Calculate maximum size of memory for flow items of Linux TC flower.
2376 * Pointer to the flow attributes.
2378 * Pointer to the list of items.
2381 * Maximum size of memory for items.
2384 flow_tcf_get_items_size(const struct rte_flow_attr *attr,
2385 const struct rte_flow_item items[])
2389 size += SZ_NLATTR_STRZ_OF("flower") +
2390 SZ_NLATTR_NEST + /* TCA_OPTIONS. */
2391 SZ_NLATTR_TYPE_OF(uint32_t); /* TCA_CLS_FLAGS_SKIP_SW. */
2392 if (attr->group > 0)
2393 size += SZ_NLATTR_TYPE_OF(uint32_t); /* TCA_CHAIN. */
2394 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2395 switch (items->type) {
2396 case RTE_FLOW_ITEM_TYPE_VOID:
2398 case RTE_FLOW_ITEM_TYPE_PORT_ID:
2400 case RTE_FLOW_ITEM_TYPE_ETH:
2401 size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2402 SZ_NLATTR_DATA_OF(ETHER_ADDR_LEN) * 4;
2403 /* dst/src MAC addr and mask. */
2405 case RTE_FLOW_ITEM_TYPE_VLAN:
2406 size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2407 SZ_NLATTR_TYPE_OF(uint16_t) +
2408 /* VLAN Ether type. */
2409 SZ_NLATTR_TYPE_OF(uint8_t) + /* VLAN prio. */
2410 SZ_NLATTR_TYPE_OF(uint16_t); /* VLAN ID. */
2412 case RTE_FLOW_ITEM_TYPE_IPV4:
2413 size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2414 SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2415 SZ_NLATTR_TYPE_OF(uint32_t) * 4;
2416 /* dst/src IP addr and mask. */
2418 case RTE_FLOW_ITEM_TYPE_IPV6:
2419 size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2420 SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2421 SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN) * 4;
2422 /* dst/src IP addr and mask. */
2424 case RTE_FLOW_ITEM_TYPE_UDP:
2425 size += SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2426 SZ_NLATTR_TYPE_OF(uint16_t) * 4;
2427 /* dst/src port and mask. */
2429 case RTE_FLOW_ITEM_TYPE_TCP:
2430 size += SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2431 SZ_NLATTR_TYPE_OF(uint16_t) * 4;
2432 /* dst/src port and mask. */
2434 case RTE_FLOW_ITEM_TYPE_VXLAN:
2435 size += SZ_NLATTR_TYPE_OF(uint32_t);
2439 "unsupported item %p type %d,"
2440 " items must be validated before flow creation",
2441 (const void *)items, items->type);
2449 * Calculate size of memory to store the VXLAN encapsultion
2450 * related items in the Netlink message buffer. Items list
2451 * is specified by RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action.
2452 * The item list should be validated.
2455 * RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action object.
2456 * List of pattern items to scan data from.
2459 * The size the part of Netlink message buffer to store the
2460 * VXLAN encapsulation item attributes.
2463 flow_tcf_vxlan_encap_size(const struct rte_flow_action *action)
2465 const struct rte_flow_item *items;
2468 assert(action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP);
2469 assert(action->conf);
2471 items = ((const struct rte_flow_action_vxlan_encap *)
2472 action->conf)->definition;
2474 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2475 switch (items->type) {
2476 case RTE_FLOW_ITEM_TYPE_VOID:
2478 case RTE_FLOW_ITEM_TYPE_ETH:
2479 /* This item does not require message buffer. */
2481 case RTE_FLOW_ITEM_TYPE_IPV4:
2482 size += SZ_NLATTR_DATA_OF(IPV4_ADDR_LEN) * 2;
2484 case RTE_FLOW_ITEM_TYPE_IPV6:
2485 size += SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN) * 2;
2487 case RTE_FLOW_ITEM_TYPE_UDP: {
2488 const struct rte_flow_item_udp *udp = items->mask;
2490 size += SZ_NLATTR_TYPE_OF(uint16_t);
2491 if (!udp || udp->hdr.src_port != RTE_BE16(0x0000))
2492 size += SZ_NLATTR_TYPE_OF(uint16_t);
2495 case RTE_FLOW_ITEM_TYPE_VXLAN:
2496 size += SZ_NLATTR_TYPE_OF(uint32_t);
2501 "unsupported item %p type %d,"
2502 " items must be validated"
2503 " before flow creation",
2504 (const void *)items, items->type);
2512 * Calculate maximum size of memory for flow actions of Linux TC flower and
2513 * extract specified actions.
2515 * @param[in] actions
2516 * Pointer to the list of actions.
2517 * @param[out] action_flags
2518 * Pointer to the detected actions.
2521 * Maximum size of memory for actions.
2524 flow_tcf_get_actions_and_size(const struct rte_flow_action actions[],
2525 uint64_t *action_flags)
2530 size += SZ_NLATTR_NEST; /* TCA_FLOWER_ACT. */
2531 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2532 switch (actions->type) {
2533 case RTE_FLOW_ACTION_TYPE_VOID:
2535 case RTE_FLOW_ACTION_TYPE_PORT_ID:
2536 size += SZ_NLATTR_NEST + /* na_act_index. */
2537 SZ_NLATTR_STRZ_OF("mirred") +
2538 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2539 SZ_NLATTR_TYPE_OF(struct tc_mirred);
2540 flags |= MLX5_FLOW_ACTION_PORT_ID;
2542 case RTE_FLOW_ACTION_TYPE_JUMP:
2543 size += SZ_NLATTR_NEST + /* na_act_index. */
2544 SZ_NLATTR_STRZ_OF("gact") +
2545 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2546 SZ_NLATTR_TYPE_OF(struct tc_gact);
2547 flags |= MLX5_FLOW_ACTION_JUMP;
2549 case RTE_FLOW_ACTION_TYPE_DROP:
2550 size += SZ_NLATTR_NEST + /* na_act_index. */
2551 SZ_NLATTR_STRZ_OF("gact") +
2552 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2553 SZ_NLATTR_TYPE_OF(struct tc_gact);
2554 flags |= MLX5_FLOW_ACTION_DROP;
2556 case RTE_FLOW_ACTION_TYPE_COUNT:
2558 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
2559 flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
2560 goto action_of_vlan;
2561 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
2562 flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
2563 goto action_of_vlan;
2564 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
2565 flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
2566 goto action_of_vlan;
2567 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
2568 flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
2569 goto action_of_vlan;
2571 size += SZ_NLATTR_NEST + /* na_act_index. */
2572 SZ_NLATTR_STRZ_OF("vlan") +
2573 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2574 SZ_NLATTR_TYPE_OF(struct tc_vlan) +
2575 SZ_NLATTR_TYPE_OF(uint16_t) +
2576 /* VLAN protocol. */
2577 SZ_NLATTR_TYPE_OF(uint16_t) + /* VLAN ID. */
2578 SZ_NLATTR_TYPE_OF(uint8_t); /* VLAN prio. */
2580 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2581 size += SZ_NLATTR_NEST + /* na_act_index. */
2582 SZ_NLATTR_STRZ_OF("tunnel_key") +
2583 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2584 SZ_NLATTR_TYPE_OF(uint8_t);
2585 size += SZ_NLATTR_TYPE_OF(struct tc_tunnel_key);
2586 size += flow_tcf_vxlan_encap_size(actions) +
2587 RTE_ALIGN_CEIL /* preceding encap params. */
2588 (sizeof(struct flow_tcf_vxlan_encap),
2590 flags |= MLX5_FLOW_ACTION_VXLAN_ENCAP;
2592 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
2593 size += SZ_NLATTR_NEST + /* na_act_index. */
2594 SZ_NLATTR_STRZ_OF("tunnel_key") +
2595 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2596 SZ_NLATTR_TYPE_OF(uint8_t);
2597 size += SZ_NLATTR_TYPE_OF(struct tc_tunnel_key);
2598 size += RTE_ALIGN_CEIL /* preceding decap params. */
2599 (sizeof(struct flow_tcf_vxlan_decap),
2601 flags |= MLX5_FLOW_ACTION_VXLAN_DECAP;
2603 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
2604 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
2605 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
2606 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
2607 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
2608 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
2609 case RTE_FLOW_ACTION_TYPE_SET_TTL:
2610 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
2611 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
2612 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
2613 size += flow_tcf_get_pedit_actions_size(&actions,
2618 "unsupported action %p type %d,"
2619 " items must be validated before flow creation",
2620 (const void *)actions, actions->type);
2624 *action_flags = flags;
2629 * Brand rtnetlink buffer with unique handle.
2631 * This handle should be unique for a given network interface to avoid
2635 * Pointer to Netlink message.
2637 * Unique 32-bit handle to use.
2640 flow_tcf_nl_brand(struct nlmsghdr *nlh, uint32_t handle)
2642 struct tcmsg *tcm = mnl_nlmsg_get_payload(nlh);
2644 tcm->tcm_handle = handle;
2645 DRV_LOG(DEBUG, "Netlink msg %p is branded with handle %x",
2646 (void *)nlh, handle);
2650 * Prepare a flow object for Linux TC flower. It calculates the maximum size of
2651 * memory required, allocates the memory, initializes Netlink message headers
2652 * and set unique TC message handle.
2655 * Pointer to the flow attributes.
2657 * Pointer to the list of items.
2658 * @param[in] actions
2659 * Pointer to the list of actions.
2661 * Pointer to the error structure.
2664 * Pointer to mlx5_flow object on success,
2665 * otherwise NULL and rte_ernno is set.
2667 static struct mlx5_flow *
2668 flow_tcf_prepare(const struct rte_flow_attr *attr,
2669 const struct rte_flow_item items[],
2670 const struct rte_flow_action actions[],
2671 struct rte_flow_error *error)
2673 size_t size = RTE_ALIGN_CEIL
2674 (sizeof(struct mlx5_flow),
2675 alignof(struct flow_tcf_tunnel_hdr)) +
2676 MNL_ALIGN(sizeof(struct nlmsghdr)) +
2677 MNL_ALIGN(sizeof(struct tcmsg));
2678 struct mlx5_flow *dev_flow;
2679 uint64_t action_flags = 0;
2680 struct nlmsghdr *nlh;
2682 uint8_t *sp, *tun = NULL;
2684 size += flow_tcf_get_items_size(attr, items);
2685 size += flow_tcf_get_actions_and_size(actions, &action_flags);
2686 dev_flow = rte_zmalloc(__func__, size, MNL_ALIGNTO);
2688 rte_flow_error_set(error, ENOMEM,
2689 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2690 "not enough memory to create E-Switch flow");
2693 sp = (uint8_t *)(dev_flow + 1);
2694 if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP) {
2696 (sp, alignof(struct flow_tcf_tunnel_hdr));
2698 sp += RTE_ALIGN_CEIL
2699 (sizeof(struct flow_tcf_vxlan_encap),
2702 size -= RTE_ALIGN_CEIL
2703 (sizeof(struct flow_tcf_vxlan_encap),
2706 } else if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2708 (sp, alignof(struct flow_tcf_tunnel_hdr));
2710 sp += RTE_ALIGN_CEIL
2711 (sizeof(struct flow_tcf_vxlan_decap),
2714 size -= RTE_ALIGN_CEIL
2715 (sizeof(struct flow_tcf_vxlan_decap),
2719 sp = RTE_PTR_ALIGN(sp, MNL_ALIGNTO);
2721 nlh = mnl_nlmsg_put_header(sp);
2722 tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
2723 *dev_flow = (struct mlx5_flow){
2724 .tcf = (struct mlx5_flow_tcf){
2726 .nlsize = size - RTE_ALIGN_CEIL
2727 (sizeof(struct mlx5_flow),
2728 alignof(struct flow_tcf_tunnel_hdr)),
2730 .tunnel = (struct flow_tcf_tunnel_hdr *)tun,
2735 if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP)
2736 dev_flow->tcf.tunnel->type = FLOW_TCF_TUNACT_VXLAN_DECAP;
2737 else if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP)
2738 dev_flow->tcf.tunnel->type = FLOW_TCF_TUNACT_VXLAN_ENCAP;
2740 * Generate a reasonably unique handle based on the address of the
2743 * This is straightforward on 32-bit systems where the flow pointer can
2744 * be used directly. Otherwise, its least significant part is taken
2745 * after shifting it by the previous power of two of the pointed buffer
2748 if (sizeof(dev_flow) <= 4)
2749 flow_tcf_nl_brand(nlh, (uintptr_t)dev_flow);
2751 flow_tcf_nl_brand(nlh, (uintptr_t)dev_flow >>
2752 rte_log2_u32(rte_align32prevpow2(size)));
2757 * Make adjustments for supporting count actions.
2760 * Pointer to the Ethernet device structure.
2761 * @param[in] dev_flow
2762 * Pointer to mlx5_flow.
2764 * Pointer to error structure.
2767 * 0 On success else a negative errno value is returned and rte_errno is set.
2770 flow_tcf_translate_action_count(struct rte_eth_dev *dev __rte_unused,
2771 struct mlx5_flow *dev_flow,
2772 struct rte_flow_error *error)
2774 struct rte_flow *flow = dev_flow->flow;
2776 if (!flow->counter) {
2777 flow->counter = flow_tcf_counter_new();
2779 return rte_flow_error_set(error, rte_errno,
2780 RTE_FLOW_ERROR_TYPE_ACTION,
2782 "cannot get counter"
2789 * Convert VXLAN VNI to 32-bit integer.
2792 * VXLAN VNI in 24-bit wire format.
2795 * VXLAN VNI as a 32-bit integer value in network endian.
2797 static inline rte_be32_t
2798 vxlan_vni_as_be32(const uint8_t vni[3])
2804 .vni = { 0, vni[0], vni[1], vni[2] },
2810 * Helper function to process RTE_FLOW_ITEM_TYPE_ETH entry in configuration
2811 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the MAC address fields
2812 * in the encapsulation parameters structure. The item must be prevalidated,
2813 * no any validation checks performed by function.
2816 * RTE_FLOW_ITEM_TYPE_ETH entry specification.
2818 * RTE_FLOW_ITEM_TYPE_ETH entry mask.
2820 * Structure to fill the gathered MAC address data.
2823 flow_tcf_parse_vxlan_encap_eth(const struct rte_flow_item_eth *spec,
2824 const struct rte_flow_item_eth *mask,
2825 struct flow_tcf_vxlan_encap *encap)
2827 /* Item must be validated before. No redundant checks. */
2829 if (!mask || !memcmp(&mask->dst,
2830 &rte_flow_item_eth_mask.dst,
2831 sizeof(rte_flow_item_eth_mask.dst))) {
2833 * Ethernet addresses are not supported by
2834 * tc as tunnel_key parameters. Destination
2835 * address is needed to form encap packet
2836 * header and retrieved by kernel from
2837 * implicit sources (ARP table, etc),
2838 * address masks are not supported at all.
2840 encap->eth.dst = spec->dst;
2841 encap->mask |= FLOW_TCF_ENCAP_ETH_DST;
2843 if (!mask || !memcmp(&mask->src,
2844 &rte_flow_item_eth_mask.src,
2845 sizeof(rte_flow_item_eth_mask.src))) {
2847 * Ethernet addresses are not supported by
2848 * tc as tunnel_key parameters. Source ethernet
2849 * address is ignored anyway.
2851 encap->eth.src = spec->src;
2852 encap->mask |= FLOW_TCF_ENCAP_ETH_SRC;
2857 * Helper function to process RTE_FLOW_ITEM_TYPE_IPV4 entry in configuration
2858 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the IPV4 address fields
2859 * in the encapsulation parameters structure. The item must be prevalidated,
2860 * no any validation checks performed by function.
2863 * RTE_FLOW_ITEM_TYPE_IPV4 entry specification.
2865 * Structure to fill the gathered IPV4 address data.
2868 flow_tcf_parse_vxlan_encap_ipv4(const struct rte_flow_item_ipv4 *spec,
2869 struct flow_tcf_vxlan_encap *encap)
2871 /* Item must be validated before. No redundant checks. */
2873 encap->ipv4.dst = spec->hdr.dst_addr;
2874 encap->ipv4.src = spec->hdr.src_addr;
2875 encap->mask |= FLOW_TCF_ENCAP_IPV4_SRC |
2876 FLOW_TCF_ENCAP_IPV4_DST;
2880 * Helper function to process RTE_FLOW_ITEM_TYPE_IPV6 entry in configuration
2881 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the IPV6 address fields
2882 * in the encapsulation parameters structure. The item must be prevalidated,
2883 * no any validation checks performed by function.
2886 * RTE_FLOW_ITEM_TYPE_IPV6 entry specification.
2888 * Structure to fill the gathered IPV6 address data.
2891 flow_tcf_parse_vxlan_encap_ipv6(const struct rte_flow_item_ipv6 *spec,
2892 struct flow_tcf_vxlan_encap *encap)
2894 /* Item must be validated before. No redundant checks. */
2896 memcpy(encap->ipv6.dst, spec->hdr.dst_addr, IPV6_ADDR_LEN);
2897 memcpy(encap->ipv6.src, spec->hdr.src_addr, IPV6_ADDR_LEN);
2898 encap->mask |= FLOW_TCF_ENCAP_IPV6_SRC |
2899 FLOW_TCF_ENCAP_IPV6_DST;
2903 * Helper function to process RTE_FLOW_ITEM_TYPE_UDP entry in configuration
2904 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the UDP port fields
2905 * in the encapsulation parameters structure. The item must be prevalidated,
2906 * no any validation checks performed by function.
2909 * RTE_FLOW_ITEM_TYPE_UDP entry specification.
2911 * RTE_FLOW_ITEM_TYPE_UDP entry mask.
2913 * Structure to fill the gathered UDP port data.
2916 flow_tcf_parse_vxlan_encap_udp(const struct rte_flow_item_udp *spec,
2917 const struct rte_flow_item_udp *mask,
2918 struct flow_tcf_vxlan_encap *encap)
2921 encap->udp.dst = spec->hdr.dst_port;
2922 encap->mask |= FLOW_TCF_ENCAP_UDP_DST;
2923 if (!mask || mask->hdr.src_port != RTE_BE16(0x0000)) {
2924 encap->udp.src = spec->hdr.src_port;
2925 encap->mask |= FLOW_TCF_ENCAP_IPV4_SRC;
2930 * Helper function to process RTE_FLOW_ITEM_TYPE_VXLAN entry in configuration
2931 * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the VNI fields
2932 * in the encapsulation parameters structure. The item must be prevalidated,
2933 * no any validation checks performed by function.
2936 * RTE_FLOW_ITEM_TYPE_VXLAN entry specification.
2938 * Structure to fill the gathered VNI address data.
2941 flow_tcf_parse_vxlan_encap_vni(const struct rte_flow_item_vxlan *spec,
2942 struct flow_tcf_vxlan_encap *encap)
2944 /* Item must be validated before. Do not redundant checks. */
2946 memcpy(encap->vxlan.vni, spec->vni, sizeof(encap->vxlan.vni));
2947 encap->mask |= FLOW_TCF_ENCAP_VXLAN_VNI;
2951 * Populate consolidated encapsulation object from list of pattern items.
2953 * Helper function to process configuration of action such as
2954 * RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. The item list should be
2955 * validated, there is no way to return an meaningful error.
2958 * RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action object.
2959 * List of pattern items to gather data from.
2961 * Structure to fill gathered data.
2964 flow_tcf_vxlan_encap_parse(const struct rte_flow_action *action,
2965 struct flow_tcf_vxlan_encap *encap)
2968 const struct rte_flow_item_eth *eth;
2969 const struct rte_flow_item_ipv4 *ipv4;
2970 const struct rte_flow_item_ipv6 *ipv6;
2971 const struct rte_flow_item_udp *udp;
2972 const struct rte_flow_item_vxlan *vxlan;
2974 const struct rte_flow_item *items;
2976 assert(action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP);
2977 assert(action->conf);
2979 items = ((const struct rte_flow_action_vxlan_encap *)
2980 action->conf)->definition;
2982 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2983 switch (items->type) {
2984 case RTE_FLOW_ITEM_TYPE_VOID:
2986 case RTE_FLOW_ITEM_TYPE_ETH:
2987 mask.eth = items->mask;
2988 spec.eth = items->spec;
2989 flow_tcf_parse_vxlan_encap_eth(spec.eth, mask.eth,
2992 case RTE_FLOW_ITEM_TYPE_IPV4:
2993 spec.ipv4 = items->spec;
2994 flow_tcf_parse_vxlan_encap_ipv4(spec.ipv4, encap);
2996 case RTE_FLOW_ITEM_TYPE_IPV6:
2997 spec.ipv6 = items->spec;
2998 flow_tcf_parse_vxlan_encap_ipv6(spec.ipv6, encap);
3000 case RTE_FLOW_ITEM_TYPE_UDP:
3001 mask.udp = items->mask;
3002 spec.udp = items->spec;
3003 flow_tcf_parse_vxlan_encap_udp(spec.udp, mask.udp,
3006 case RTE_FLOW_ITEM_TYPE_VXLAN:
3007 spec.vxlan = items->spec;
3008 flow_tcf_parse_vxlan_encap_vni(spec.vxlan, encap);
3013 "unsupported item %p type %d,"
3014 " items must be validated"
3015 " before flow creation",
3016 (const void *)items, items->type);
3024 * Translate flow for Linux TC flower and construct Netlink message.
3027 * Pointer to the priv structure.
3028 * @param[in, out] flow
3029 * Pointer to the sub flow.
3031 * Pointer to the flow attributes.
3033 * Pointer to the list of items.
3034 * @param[in] actions
3035 * Pointer to the list of actions.
3037 * Pointer to the error structure.
3040 * 0 on success, a negative errno value otherwise and rte_ernno is set.
3043 flow_tcf_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3044 const struct rte_flow_attr *attr,
3045 const struct rte_flow_item items[],
3046 const struct rte_flow_action actions[],
3047 struct rte_flow_error *error)
3050 const struct rte_flow_item_port_id *port_id;
3051 const struct rte_flow_item_eth *eth;
3052 const struct rte_flow_item_vlan *vlan;
3053 const struct rte_flow_item_ipv4 *ipv4;
3054 const struct rte_flow_item_ipv6 *ipv6;
3055 const struct rte_flow_item_tcp *tcp;
3056 const struct rte_flow_item_udp *udp;
3057 const struct rte_flow_item_vxlan *vxlan;
3060 const struct rte_flow_action_port_id *port_id;
3061 const struct rte_flow_action_jump *jump;
3062 const struct rte_flow_action_of_push_vlan *of_push_vlan;
3063 const struct rte_flow_action_of_set_vlan_vid *
3065 const struct rte_flow_action_of_set_vlan_pcp *
3069 struct flow_tcf_tunnel_hdr *hdr;
3070 struct flow_tcf_vxlan_decap *vxlan;
3075 struct flow_tcf_tunnel_hdr *hdr;
3076 struct flow_tcf_vxlan_encap *vxlan;
3080 struct flow_tcf_ptoi ptoi[PTOI_TABLE_SZ_MAX(dev)];
3081 struct nlmsghdr *nlh = dev_flow->tcf.nlh;
3082 struct tcmsg *tcm = dev_flow->tcf.tcm;
3083 uint32_t na_act_index_cur;
3084 bool eth_type_set = 0;
3085 bool vlan_present = 0;
3086 bool vlan_eth_type_set = 0;
3087 bool ip_proto_set = 0;
3088 struct nlattr *na_flower;
3089 struct nlattr *na_flower_act;
3090 struct nlattr *na_vlan_id = NULL;
3091 struct nlattr *na_vlan_priority = NULL;
3092 uint64_t item_flags = 0;
3095 claim_nonzero(flow_tcf_build_ptoi_table(dev, ptoi,
3096 PTOI_TABLE_SZ_MAX(dev)));
3097 if (dev_flow->tcf.tunnel) {
3098 switch (dev_flow->tcf.tunnel->type) {
3099 case FLOW_TCF_TUNACT_VXLAN_DECAP:
3100 decap.vxlan = dev_flow->tcf.vxlan_decap;
3102 case FLOW_TCF_TUNACT_VXLAN_ENCAP:
3103 encap.vxlan = dev_flow->tcf.vxlan_encap;
3105 /* New tunnel actions can be added here. */
3111 nlh = dev_flow->tcf.nlh;
3112 tcm = dev_flow->tcf.tcm;
3113 /* Prepare API must have been called beforehand. */
3114 assert(nlh != NULL && tcm != NULL);
3115 tcm->tcm_family = AF_UNSPEC;
3116 tcm->tcm_ifindex = ptoi[0].ifindex;
3117 tcm->tcm_parent = TC_H_MAKE(TC_H_INGRESS, TC_H_MIN_INGRESS);
3119 * Priority cannot be zero to prevent the kernel from picking one
3122 tcm->tcm_info = TC_H_MAKE((attr->priority + 1) << 16,
3123 RTE_BE16(ETH_P_ALL));
3124 if (attr->group > 0)
3125 mnl_attr_put_u32(nlh, TCA_CHAIN, attr->group);
3126 mnl_attr_put_strz(nlh, TCA_KIND, "flower");
3127 na_flower = mnl_attr_nest_start(nlh, TCA_OPTIONS);
3128 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3131 switch (items->type) {
3132 case RTE_FLOW_ITEM_TYPE_VOID:
3134 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3135 mask.port_id = flow_tcf_item_mask
3136 (items, &rte_flow_item_port_id_mask,
3137 &flow_tcf_mask_supported.port_id,
3138 &flow_tcf_mask_empty.port_id,
3139 sizeof(flow_tcf_mask_supported.port_id),
3141 assert(mask.port_id);
3142 if (mask.port_id == &flow_tcf_mask_empty.port_id)
3144 spec.port_id = items->spec;
3145 if (!mask.port_id->id)
3148 for (i = 0; ptoi[i].ifindex; ++i)
3149 if (ptoi[i].port_id == spec.port_id->id)
3151 assert(ptoi[i].ifindex);
3152 tcm->tcm_ifindex = ptoi[i].ifindex;
3154 case RTE_FLOW_ITEM_TYPE_ETH:
3155 item_flags |= (item_flags & MLX5_FLOW_LAYER_VXLAN) ?
3156 MLX5_FLOW_LAYER_INNER_L2 :
3157 MLX5_FLOW_LAYER_OUTER_L2;
3158 mask.eth = flow_tcf_item_mask
3159 (items, &rte_flow_item_eth_mask,
3160 &flow_tcf_mask_supported.eth,
3161 &flow_tcf_mask_empty.eth,
3162 sizeof(flow_tcf_mask_supported.eth),
3165 if (mask.eth == &flow_tcf_mask_empty.eth)
3167 spec.eth = items->spec;
3169 !(item_flags & MLX5_FLOW_LAYER_VXLAN)) {
3171 "outer L2 addresses cannot be forced"
3172 " for vxlan decapsulation, parameter"
3176 if (mask.eth->type) {
3177 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_ETH_TYPE,
3181 if (!is_zero_ether_addr(&mask.eth->dst)) {
3182 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_DST,
3184 spec.eth->dst.addr_bytes);
3185 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_DST_MASK,
3187 mask.eth->dst.addr_bytes);
3189 if (!is_zero_ether_addr(&mask.eth->src)) {
3190 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_SRC,
3192 spec.eth->src.addr_bytes);
3193 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_SRC_MASK,
3195 mask.eth->src.addr_bytes);
3197 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3199 case RTE_FLOW_ITEM_TYPE_VLAN:
3202 item_flags |= MLX5_FLOW_LAYER_OUTER_VLAN;
3203 mask.vlan = flow_tcf_item_mask
3204 (items, &rte_flow_item_vlan_mask,
3205 &flow_tcf_mask_supported.vlan,
3206 &flow_tcf_mask_empty.vlan,
3207 sizeof(flow_tcf_mask_supported.vlan),
3211 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_ETH_TYPE,
3212 RTE_BE16(ETH_P_8021Q));
3215 if (mask.vlan == &flow_tcf_mask_empty.vlan)
3217 spec.vlan = items->spec;
3218 if (mask.vlan->inner_type) {
3219 mnl_attr_put_u16(nlh,
3220 TCA_FLOWER_KEY_VLAN_ETH_TYPE,
3221 spec.vlan->inner_type);
3222 vlan_eth_type_set = 1;
3224 if (mask.vlan->tci & RTE_BE16(0xe000))
3225 mnl_attr_put_u8(nlh, TCA_FLOWER_KEY_VLAN_PRIO,
3227 (spec.vlan->tci) >> 13) & 0x7);
3228 if (mask.vlan->tci & RTE_BE16(0x0fff))
3229 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_VLAN_ID,
3233 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3235 case RTE_FLOW_ITEM_TYPE_IPV4:
3236 item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3237 mask.ipv4 = flow_tcf_item_mask
3238 (items, &rte_flow_item_ipv4_mask,
3239 &flow_tcf_mask_supported.ipv4,
3240 &flow_tcf_mask_empty.ipv4,
3241 sizeof(flow_tcf_mask_supported.ipv4),
3244 spec.ipv4 = items->spec;
3246 if (!eth_type_set && !vlan_eth_type_set)
3250 TCA_FLOWER_KEY_VLAN_ETH_TYPE :
3251 TCA_FLOWER_KEY_ETH_TYPE,
3252 RTE_BE16(ETH_P_IP));
3254 vlan_eth_type_set = 1;
3255 if (mask.ipv4 == &flow_tcf_mask_empty.ipv4)
3257 if (mask.ipv4->hdr.next_proto_id) {
3259 (nlh, TCA_FLOWER_KEY_IP_PROTO,
3260 spec.ipv4->hdr.next_proto_id);
3264 assert(mask.ipv4 != &flow_tcf_mask_empty.ipv4);
3266 if (mask.ipv4->hdr.src_addr) {
3269 TCA_FLOWER_KEY_ENC_IPV4_SRC :
3270 TCA_FLOWER_KEY_IPV4_SRC,
3271 spec.ipv4->hdr.src_addr);
3274 TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK :
3275 TCA_FLOWER_KEY_IPV4_SRC_MASK,
3276 mask.ipv4->hdr.src_addr);
3278 if (mask.ipv4->hdr.dst_addr) {
3281 TCA_FLOWER_KEY_ENC_IPV4_DST :
3282 TCA_FLOWER_KEY_IPV4_DST,
3283 spec.ipv4->hdr.dst_addr);
3286 TCA_FLOWER_KEY_ENC_IPV4_DST_MASK :
3287 TCA_FLOWER_KEY_IPV4_DST_MASK,
3288 mask.ipv4->hdr.dst_addr);
3290 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3292 case RTE_FLOW_ITEM_TYPE_IPV6:
3293 item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3294 mask.ipv6 = flow_tcf_item_mask
3295 (items, &rte_flow_item_ipv6_mask,
3296 &flow_tcf_mask_supported.ipv6,
3297 &flow_tcf_mask_empty.ipv6,
3298 sizeof(flow_tcf_mask_supported.ipv6),
3301 spec.ipv6 = items->spec;
3303 if (!eth_type_set || !vlan_eth_type_set) {
3307 TCA_FLOWER_KEY_VLAN_ETH_TYPE :
3308 TCA_FLOWER_KEY_ETH_TYPE,
3309 RTE_BE16(ETH_P_IPV6));
3312 vlan_eth_type_set = 1;
3313 if (mask.ipv6 == &flow_tcf_mask_empty.ipv6)
3315 if (mask.ipv6->hdr.proto) {
3317 (nlh, TCA_FLOWER_KEY_IP_PROTO,
3318 spec.ipv6->hdr.proto);
3322 assert(mask.ipv6 != &flow_tcf_mask_empty.ipv6);
3324 if (!IN6_IS_ADDR_UNSPECIFIED(mask.ipv6->hdr.src_addr)) {
3325 mnl_attr_put(nlh, decap.vxlan ?
3326 TCA_FLOWER_KEY_ENC_IPV6_SRC :
3327 TCA_FLOWER_KEY_IPV6_SRC,
3329 spec.ipv6->hdr.src_addr);
3330 mnl_attr_put(nlh, decap.vxlan ?
3331 TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK :
3332 TCA_FLOWER_KEY_IPV6_SRC_MASK,
3334 mask.ipv6->hdr.src_addr);
3336 if (!IN6_IS_ADDR_UNSPECIFIED(mask.ipv6->hdr.dst_addr)) {
3337 mnl_attr_put(nlh, decap.vxlan ?
3338 TCA_FLOWER_KEY_ENC_IPV6_DST :
3339 TCA_FLOWER_KEY_IPV6_DST,
3341 spec.ipv6->hdr.dst_addr);
3342 mnl_attr_put(nlh, decap.vxlan ?
3343 TCA_FLOWER_KEY_ENC_IPV6_DST_MASK :
3344 TCA_FLOWER_KEY_IPV6_DST_MASK,
3346 mask.ipv6->hdr.dst_addr);
3348 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3350 case RTE_FLOW_ITEM_TYPE_UDP:
3351 item_flags |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3352 mask.udp = flow_tcf_item_mask
3353 (items, &rte_flow_item_udp_mask,
3354 &flow_tcf_mask_supported.udp,
3355 &flow_tcf_mask_empty.udp,
3356 sizeof(flow_tcf_mask_supported.udp),
3359 spec.udp = items->spec;
3363 (nlh, TCA_FLOWER_KEY_IP_PROTO,
3365 if (mask.udp == &flow_tcf_mask_empty.udp)
3368 assert(mask.udp != &flow_tcf_mask_empty.udp);
3369 decap.vxlan->udp_port =
3371 (spec.udp->hdr.dst_port);
3373 if (mask.udp->hdr.src_port) {
3376 TCA_FLOWER_KEY_ENC_UDP_SRC_PORT :
3377 TCA_FLOWER_KEY_UDP_SRC,
3378 spec.udp->hdr.src_port);
3381 TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK :
3382 TCA_FLOWER_KEY_UDP_SRC_MASK,
3383 mask.udp->hdr.src_port);
3385 if (mask.udp->hdr.dst_port) {
3388 TCA_FLOWER_KEY_ENC_UDP_DST_PORT :
3389 TCA_FLOWER_KEY_UDP_DST,
3390 spec.udp->hdr.dst_port);
3393 TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK :
3394 TCA_FLOWER_KEY_UDP_DST_MASK,
3395 mask.udp->hdr.dst_port);
3397 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3399 case RTE_FLOW_ITEM_TYPE_TCP:
3400 item_flags |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3401 mask.tcp = flow_tcf_item_mask
3402 (items, &rte_flow_item_tcp_mask,
3403 &flow_tcf_mask_supported.tcp,
3404 &flow_tcf_mask_empty.tcp,
3405 sizeof(flow_tcf_mask_supported.tcp),
3409 mnl_attr_put_u8(nlh, TCA_FLOWER_KEY_IP_PROTO,
3411 if (mask.tcp == &flow_tcf_mask_empty.tcp)
3413 spec.tcp = items->spec;
3414 if (mask.tcp->hdr.src_port) {
3415 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_TCP_SRC,
3416 spec.tcp->hdr.src_port);
3417 mnl_attr_put_u16(nlh,
3418 TCA_FLOWER_KEY_TCP_SRC_MASK,
3419 mask.tcp->hdr.src_port);
3421 if (mask.tcp->hdr.dst_port) {
3422 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_TCP_DST,
3423 spec.tcp->hdr.dst_port);
3424 mnl_attr_put_u16(nlh,
3425 TCA_FLOWER_KEY_TCP_DST_MASK,
3426 mask.tcp->hdr.dst_port);
3428 if (mask.tcp->hdr.tcp_flags) {
3431 TCA_FLOWER_KEY_TCP_FLAGS,
3433 (spec.tcp->hdr.tcp_flags));
3436 TCA_FLOWER_KEY_TCP_FLAGS_MASK,
3438 (mask.tcp->hdr.tcp_flags));
3440 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3442 case RTE_FLOW_ITEM_TYPE_VXLAN:
3443 assert(decap.vxlan);
3444 item_flags |= MLX5_FLOW_LAYER_VXLAN;
3445 spec.vxlan = items->spec;
3446 mnl_attr_put_u32(nlh,
3447 TCA_FLOWER_KEY_ENC_KEY_ID,
3448 vxlan_vni_as_be32(spec.vxlan->vni));
3449 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3452 return rte_flow_error_set(error, ENOTSUP,
3453 RTE_FLOW_ERROR_TYPE_ITEM,
3454 NULL, "item not supported");
3457 na_flower_act = mnl_attr_nest_start(nlh, TCA_FLOWER_ACT);
3458 na_act_index_cur = 1;
3459 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3460 struct nlattr *na_act_index;
3461 struct nlattr *na_act;
3462 unsigned int vlan_act;
3465 switch (actions->type) {
3466 case RTE_FLOW_ACTION_TYPE_VOID:
3468 case RTE_FLOW_ACTION_TYPE_PORT_ID:
3469 conf.port_id = actions->conf;
3470 if (conf.port_id->original)
3473 for (i = 0; ptoi[i].ifindex; ++i)
3474 if (ptoi[i].port_id == conf.port_id->id)
3476 assert(ptoi[i].ifindex);
3478 mnl_attr_nest_start(nlh, na_act_index_cur++);
3479 assert(na_act_index);
3480 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "mirred");
3481 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3484 assert(dev_flow->tcf.tunnel);
3485 dev_flow->tcf.tunnel->ifindex_ptr =
3486 &((struct tc_mirred *)
3487 mnl_attr_get_payload
3488 (mnl_nlmsg_get_payload_tail
3491 mnl_attr_put(nlh, TCA_MIRRED_PARMS,
3492 sizeof(struct tc_mirred),
3493 &(struct tc_mirred){
3494 .action = TC_ACT_STOLEN,
3495 .eaction = TCA_EGRESS_REDIR,
3496 .ifindex = ptoi[i].ifindex,
3498 mnl_attr_nest_end(nlh, na_act);
3499 mnl_attr_nest_end(nlh, na_act_index);
3501 case RTE_FLOW_ACTION_TYPE_JUMP:
3502 conf.jump = actions->conf;
3504 mnl_attr_nest_start(nlh, na_act_index_cur++);
3505 assert(na_act_index);
3506 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "gact");
3507 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3509 mnl_attr_put(nlh, TCA_GACT_PARMS,
3510 sizeof(struct tc_gact),
3512 .action = TC_ACT_GOTO_CHAIN |
3515 mnl_attr_nest_end(nlh, na_act);
3516 mnl_attr_nest_end(nlh, na_act_index);
3518 case RTE_FLOW_ACTION_TYPE_DROP:
3520 mnl_attr_nest_start(nlh, na_act_index_cur++);
3521 assert(na_act_index);
3522 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "gact");
3523 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3525 mnl_attr_put(nlh, TCA_GACT_PARMS,
3526 sizeof(struct tc_gact),
3528 .action = TC_ACT_SHOT,
3530 mnl_attr_nest_end(nlh, na_act);
3531 mnl_attr_nest_end(nlh, na_act_index);
3533 case RTE_FLOW_ACTION_TYPE_COUNT:
3535 * Driver adds the count action implicitly for
3536 * each rule it creates.
3538 ret = flow_tcf_translate_action_count(dev,
3543 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3544 conf.of_push_vlan = NULL;
3545 vlan_act = TCA_VLAN_ACT_POP;
3546 goto action_of_vlan;
3547 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3548 conf.of_push_vlan = actions->conf;
3549 vlan_act = TCA_VLAN_ACT_PUSH;
3550 goto action_of_vlan;
3551 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3552 conf.of_set_vlan_vid = actions->conf;
3554 goto override_na_vlan_id;
3555 vlan_act = TCA_VLAN_ACT_MODIFY;
3556 goto action_of_vlan;
3557 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3558 conf.of_set_vlan_pcp = actions->conf;
3559 if (na_vlan_priority)
3560 goto override_na_vlan_priority;
3561 vlan_act = TCA_VLAN_ACT_MODIFY;
3562 goto action_of_vlan;
3565 mnl_attr_nest_start(nlh, na_act_index_cur++);
3566 assert(na_act_index);
3567 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "vlan");
3568 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3570 mnl_attr_put(nlh, TCA_VLAN_PARMS,
3571 sizeof(struct tc_vlan),
3573 .action = TC_ACT_PIPE,
3574 .v_action = vlan_act,
3576 if (vlan_act == TCA_VLAN_ACT_POP) {
3577 mnl_attr_nest_end(nlh, na_act);
3578 mnl_attr_nest_end(nlh, na_act_index);
3581 if (vlan_act == TCA_VLAN_ACT_PUSH)
3582 mnl_attr_put_u16(nlh,
3583 TCA_VLAN_PUSH_VLAN_PROTOCOL,
3584 conf.of_push_vlan->ethertype);
3585 na_vlan_id = mnl_nlmsg_get_payload_tail(nlh);
3586 mnl_attr_put_u16(nlh, TCA_VLAN_PAD, 0);
3587 na_vlan_priority = mnl_nlmsg_get_payload_tail(nlh);
3588 mnl_attr_put_u8(nlh, TCA_VLAN_PAD, 0);
3589 mnl_attr_nest_end(nlh, na_act);
3590 mnl_attr_nest_end(nlh, na_act_index);
3591 if (actions->type ==
3592 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
3593 override_na_vlan_id:
3594 na_vlan_id->nla_type = TCA_VLAN_PUSH_VLAN_ID;
3595 *(uint16_t *)mnl_attr_get_payload(na_vlan_id) =
3597 (conf.of_set_vlan_vid->vlan_vid);
3598 } else if (actions->type ==
3599 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
3600 override_na_vlan_priority:
3601 na_vlan_priority->nla_type =
3602 TCA_VLAN_PUSH_VLAN_PRIORITY;
3603 *(uint8_t *)mnl_attr_get_payload
3604 (na_vlan_priority) =
3605 conf.of_set_vlan_pcp->vlan_pcp;
3608 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3609 assert(decap.vxlan);
3610 assert(dev_flow->tcf.tunnel);
3611 dev_flow->tcf.tunnel->ifindex_ptr =
3612 (unsigned int *)&tcm->tcm_ifindex;
3614 mnl_attr_nest_start(nlh, na_act_index_cur++);
3615 assert(na_act_index);
3616 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "tunnel_key");
3617 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3619 mnl_attr_put(nlh, TCA_TUNNEL_KEY_PARMS,
3620 sizeof(struct tc_tunnel_key),
3621 &(struct tc_tunnel_key){
3622 .action = TC_ACT_PIPE,
3623 .t_action = TCA_TUNNEL_KEY_ACT_RELEASE,
3625 mnl_attr_nest_end(nlh, na_act);
3626 mnl_attr_nest_end(nlh, na_act_index);
3627 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3629 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3630 assert(encap.vxlan);
3631 flow_tcf_vxlan_encap_parse(actions, encap.vxlan);
3633 mnl_attr_nest_start(nlh, na_act_index_cur++);
3634 assert(na_act_index);
3635 mnl_attr_put_strz(nlh, TCA_ACT_KIND, "tunnel_key");
3636 na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3638 mnl_attr_put(nlh, TCA_TUNNEL_KEY_PARMS,
3639 sizeof(struct tc_tunnel_key),
3640 &(struct tc_tunnel_key){
3641 .action = TC_ACT_PIPE,
3642 .t_action = TCA_TUNNEL_KEY_ACT_SET,
3644 if (encap.vxlan->mask & FLOW_TCF_ENCAP_UDP_DST)
3645 mnl_attr_put_u16(nlh,
3646 TCA_TUNNEL_KEY_ENC_DST_PORT,
3647 encap.vxlan->udp.dst);
3648 if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV4_SRC)
3649 mnl_attr_put_u32(nlh,
3650 TCA_TUNNEL_KEY_ENC_IPV4_SRC,
3651 encap.vxlan->ipv4.src);
3652 if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV4_DST)
3653 mnl_attr_put_u32(nlh,
3654 TCA_TUNNEL_KEY_ENC_IPV4_DST,
3655 encap.vxlan->ipv4.dst);
3656 if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV6_SRC)
3658 TCA_TUNNEL_KEY_ENC_IPV6_SRC,
3659 sizeof(encap.vxlan->ipv6.src),
3660 &encap.vxlan->ipv6.src);
3661 if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV6_DST)
3663 TCA_TUNNEL_KEY_ENC_IPV6_DST,
3664 sizeof(encap.vxlan->ipv6.dst),
3665 &encap.vxlan->ipv6.dst);
3666 if (encap.vxlan->mask & FLOW_TCF_ENCAP_VXLAN_VNI)
3667 mnl_attr_put_u32(nlh,
3668 TCA_TUNNEL_KEY_ENC_KEY_ID,
3670 (encap.vxlan->vxlan.vni));
3671 mnl_attr_put_u8(nlh, TCA_TUNNEL_KEY_NO_CSUM, 0);
3672 mnl_attr_nest_end(nlh, na_act);
3673 mnl_attr_nest_end(nlh, na_act_index);
3674 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3676 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3677 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3678 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3679 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3680 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3681 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3682 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3683 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3684 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3685 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3687 mnl_attr_nest_start(nlh, na_act_index_cur++);
3688 flow_tcf_create_pedit_mnl_msg(nlh,
3689 &actions, item_flags);
3690 mnl_attr_nest_end(nlh, na_act_index);
3693 return rte_flow_error_set(error, ENOTSUP,
3694 RTE_FLOW_ERROR_TYPE_ACTION,
3696 "action not supported");
3700 assert(na_flower_act);
3701 mnl_attr_nest_end(nlh, na_flower_act);
3702 mnl_attr_put_u32(nlh, TCA_FLOWER_FLAGS, decap.vxlan ?
3703 0 : TCA_CLS_FLAGS_SKIP_SW);
3704 mnl_attr_nest_end(nlh, na_flower);
3705 if (dev_flow->tcf.tunnel && dev_flow->tcf.tunnel->ifindex_ptr)
3706 dev_flow->tcf.tunnel->ifindex_org =
3707 *dev_flow->tcf.tunnel->ifindex_ptr;
3708 assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3713 * Send Netlink message with acknowledgment.
3716 * Flow context to use.
3718 * Message to send. This function always raises the NLM_F_ACK flag before
3721 * Message length. Message buffer may contain multiple commands and
3722 * nlmsg_len field not always corresponds to actual message length.
3723 * If 0 specified the nlmsg_len field in header is used as message length.
3725 * Callback handler for received message.
3727 * Context pointer for callback handler.
3730 * 0 on success, a negative errno value otherwise and rte_errno is set.
3733 flow_tcf_nl_ack(struct mlx5_flow_tcf_context *tcf,
3734 struct nlmsghdr *nlh,
3736 mnl_cb_t cb, void *arg)
3738 unsigned int portid = mnl_socket_get_portid(tcf->nl);
3739 uint32_t seq = tcf->seq++;
3745 /* seq 0 is reserved for kernel event-driven notifications. */
3747 nlh->nlmsg_seq = seq;
3749 msglen = nlh->nlmsg_len;
3750 nlh->nlmsg_flags |= NLM_F_ACK;
3752 ret = mnl_socket_sendto(tcf->nl, nlh, msglen);
3753 err = (ret <= 0) ? errno : 0;
3754 nlh = (struct nlmsghdr *)(tcf->buf);
3756 * The following loop postpones non-fatal errors until multipart
3757 * messages are complete.
3761 ret = mnl_socket_recvfrom(tcf->nl, tcf->buf,
3769 ret = mnl_cb_run(nlh, ret, seq, portid,
3776 /* Will receive till end of multipart message */
3777 if (!(nlh->nlmsg_flags & NLM_F_MULTI) ||
3778 nlh->nlmsg_type == NLMSG_DONE)
3787 #define MNL_BUF_EXTRA_SPACE 16
3788 #define MNL_REQUEST_SIZE_MIN 256
3789 #define MNL_REQUEST_SIZE_MAX 2048
3790 #define MNL_REQUEST_SIZE RTE_MIN(RTE_MAX(sysconf(_SC_PAGESIZE), \
3791 MNL_REQUEST_SIZE_MIN), MNL_REQUEST_SIZE_MAX)
3793 /* Data structures used by flow_tcf_xxx_cb() routines. */
3794 struct tcf_nlcb_buf {
3795 LIST_ENTRY(tcf_nlcb_buf) next;
3797 alignas(struct nlmsghdr)
3798 uint8_t msg[]; /**< Netlink message data. */
3801 struct tcf_nlcb_context {
3802 unsigned int ifindex; /**< Base interface index. */
3804 LIST_HEAD(, tcf_nlcb_buf) nlbuf;
3808 * Allocate space for netlink command in buffer list
3810 * @param[in, out] ctx
3811 * Pointer to callback context with command buffers list.
3813 * Required size of data buffer to be allocated.
3816 * Pointer to allocated memory, aligned as message header.
3817 * NULL if some error occurred.
3819 static struct nlmsghdr *
3820 flow_tcf_alloc_nlcmd(struct tcf_nlcb_context *ctx, uint32_t size)
3822 struct tcf_nlcb_buf *buf;
3823 struct nlmsghdr *nlh;
3825 size = NLMSG_ALIGN(size);
3826 buf = LIST_FIRST(&ctx->nlbuf);
3827 if (buf && (buf->size + size) <= ctx->bufsize) {
3828 nlh = (struct nlmsghdr *)&buf->msg[buf->size];
3832 if (size > ctx->bufsize) {
3833 DRV_LOG(WARNING, "netlink: too long command buffer requested");
3836 buf = rte_malloc(__func__,
3837 ctx->bufsize + sizeof(struct tcf_nlcb_buf),
3838 alignof(struct tcf_nlcb_buf));
3840 DRV_LOG(WARNING, "netlink: no memory for command buffer");
3843 LIST_INSERT_HEAD(&ctx->nlbuf, buf, next);
3845 nlh = (struct nlmsghdr *)&buf->msg[0];
3850 * Set NLM_F_ACK flags in the last netlink command in buffer.
3851 * Only last command in the buffer will be acked by system.
3853 * @param[in, out] buf
3854 * Pointer to buffer with netlink commands.
3857 flow_tcf_setack_nlcmd(struct tcf_nlcb_buf *buf)
3859 struct nlmsghdr *nlh;
3864 nlh = (struct nlmsghdr *)&buf->msg[size];
3865 size += NLMSG_ALIGN(nlh->nlmsg_len);
3866 if (size >= buf->size) {
3867 nlh->nlmsg_flags |= NLM_F_ACK;
3874 * Send the buffers with prepared netlink commands. Scans the list and
3875 * sends all found buffers. Buffers are sent and freed anyway in order
3876 * to prevent memory leakage if some every message in received packet.
3879 * Context object initialized by mlx5_flow_tcf_context_create().
3880 * @param[in, out] ctx
3881 * Pointer to callback context with command buffers list.
3884 * Zero value on success, negative errno value otherwise
3885 * and rte_errno is set.
3888 flow_tcf_send_nlcmd(struct mlx5_flow_tcf_context *tcf,
3889 struct tcf_nlcb_context *ctx)
3891 struct tcf_nlcb_buf *bc, *bn;
3892 struct nlmsghdr *nlh;
3895 bc = LIST_FIRST(&ctx->nlbuf);
3899 bn = LIST_NEXT(bc, next);
3901 flow_tcf_setack_nlcmd(bc);
3902 nlh = (struct nlmsghdr *)&bc->msg;
3903 rc = flow_tcf_nl_ack(tcf, nlh, bc->size, NULL, NULL);
3910 LIST_INIT(&ctx->nlbuf);
3915 * Collect local IP address rules with scope link attribute on specified
3916 * network device. This is callback routine called by libmnl mnl_cb_run()
3917 * in loop for every message in received packet.
3920 * Pointer to reply header.
3921 * @param[in, out] arg
3922 * Opaque data pointer for this callback.
3925 * A positive, nonzero value on success, negative errno value otherwise
3926 * and rte_errno is set.
3929 flow_tcf_collect_local_cb(const struct nlmsghdr *nlh, void *arg)
3931 struct tcf_nlcb_context *ctx = arg;
3932 struct nlmsghdr *cmd;
3933 struct ifaddrmsg *ifa;
3935 struct nlattr *na_local = NULL;
3936 struct nlattr *na_peer = NULL;
3937 unsigned char family;
3940 if (nlh->nlmsg_type != RTM_NEWADDR) {
3944 ifa = mnl_nlmsg_get_payload(nlh);
3945 family = ifa->ifa_family;
3946 if (ifa->ifa_index != ctx->ifindex ||
3947 ifa->ifa_scope != RT_SCOPE_LINK ||
3948 !(ifa->ifa_flags & IFA_F_PERMANENT) ||
3949 (family != AF_INET && family != AF_INET6))
3951 mnl_attr_for_each(na, nlh, sizeof(*ifa)) {
3952 switch (mnl_attr_get_type(na)) {
3960 if (na_local && na_peer)
3963 if (!na_local || !na_peer)
3965 /* Local rule found with scope link, permanent and assigned peer. */
3966 size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
3967 MNL_ALIGN(sizeof(struct ifaddrmsg)) +
3968 (family == AF_INET6 ? 2 * SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN)
3969 : 2 * SZ_NLATTR_TYPE_OF(uint32_t));
3970 cmd = flow_tcf_alloc_nlcmd(ctx, size);
3975 cmd = mnl_nlmsg_put_header(cmd);
3976 cmd->nlmsg_type = RTM_DELADDR;
3977 cmd->nlmsg_flags = NLM_F_REQUEST;
3978 ifa = mnl_nlmsg_put_extra_header(cmd, sizeof(*ifa));
3979 ifa->ifa_flags = IFA_F_PERMANENT;
3980 ifa->ifa_scope = RT_SCOPE_LINK;
3981 ifa->ifa_index = ctx->ifindex;
3982 if (family == AF_INET) {
3983 ifa->ifa_family = AF_INET;
3984 ifa->ifa_prefixlen = 32;
3985 mnl_attr_put_u32(cmd, IFA_LOCAL, mnl_attr_get_u32(na_local));
3986 mnl_attr_put_u32(cmd, IFA_ADDRESS, mnl_attr_get_u32(na_peer));
3988 ifa->ifa_family = AF_INET6;
3989 ifa->ifa_prefixlen = 128;
3990 mnl_attr_put(cmd, IFA_LOCAL, IPV6_ADDR_LEN,
3991 mnl_attr_get_payload(na_local));
3992 mnl_attr_put(cmd, IFA_ADDRESS, IPV6_ADDR_LEN,
3993 mnl_attr_get_payload(na_peer));
3995 assert(size == cmd->nlmsg_len);
4000 * Cleanup the local IP addresses on outer interface.
4003 * Context object initialized by mlx5_flow_tcf_context_create().
4004 * @param[in] ifindex
4005 * Network inferface index to perform cleanup.
4008 flow_tcf_encap_local_cleanup(struct mlx5_flow_tcf_context *tcf,
4009 unsigned int ifindex)
4011 struct nlmsghdr *nlh;
4012 struct ifaddrmsg *ifa;
4013 struct tcf_nlcb_context ctx = {
4015 .bufsize = MNL_REQUEST_SIZE,
4016 .nlbuf = LIST_HEAD_INITIALIZER(),
4022 * Seek and destroy leftovers of local IP addresses with
4023 * matching properties "scope link".
4025 nlh = mnl_nlmsg_put_header(tcf->buf);
4026 nlh->nlmsg_type = RTM_GETADDR;
4027 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4028 ifa = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifa));
4029 ifa->ifa_family = AF_UNSPEC;
4030 ifa->ifa_index = ifindex;
4031 ifa->ifa_scope = RT_SCOPE_LINK;
4032 ret = flow_tcf_nl_ack(tcf, nlh, 0, flow_tcf_collect_local_cb, &ctx);
4034 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4035 ret = flow_tcf_send_nlcmd(tcf, &ctx);
4037 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4041 * Collect neigh permament rules on specified network device.
4042 * This is callback routine called by libmnl mnl_cb_run() in loop for
4043 * every message in received packet.
4046 * Pointer to reply header.
4047 * @param[in, out] arg
4048 * Opaque data pointer for this callback.
4051 * A positive, nonzero value on success, negative errno value otherwise
4052 * and rte_errno is set.
4055 flow_tcf_collect_neigh_cb(const struct nlmsghdr *nlh, void *arg)
4057 struct tcf_nlcb_context *ctx = arg;
4058 struct nlmsghdr *cmd;
4061 struct nlattr *na_ip = NULL;
4062 struct nlattr *na_mac = NULL;
4063 unsigned char family;
4066 if (nlh->nlmsg_type != RTM_NEWNEIGH) {
4070 ndm = mnl_nlmsg_get_payload(nlh);
4071 family = ndm->ndm_family;
4072 if (ndm->ndm_ifindex != (int)ctx->ifindex ||
4073 !(ndm->ndm_state & NUD_PERMANENT) ||
4074 (family != AF_INET && family != AF_INET6))
4076 mnl_attr_for_each(na, nlh, sizeof(*ndm)) {
4077 switch (mnl_attr_get_type(na)) {
4085 if (na_mac && na_ip)
4088 if (!na_mac || !na_ip)
4090 /* Neigh rule with permenent attribute found. */
4091 size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
4092 MNL_ALIGN(sizeof(struct ndmsg)) +
4093 SZ_NLATTR_DATA_OF(ETHER_ADDR_LEN) +
4094 (family == AF_INET6 ? SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN)
4095 : SZ_NLATTR_TYPE_OF(uint32_t));
4096 cmd = flow_tcf_alloc_nlcmd(ctx, size);
4101 cmd = mnl_nlmsg_put_header(cmd);
4102 cmd->nlmsg_type = RTM_DELNEIGH;
4103 cmd->nlmsg_flags = NLM_F_REQUEST;
4104 ndm = mnl_nlmsg_put_extra_header(cmd, sizeof(*ndm));
4105 ndm->ndm_ifindex = ctx->ifindex;
4106 ndm->ndm_state = NUD_PERMANENT;
4109 if (family == AF_INET) {
4110 ndm->ndm_family = AF_INET;
4111 mnl_attr_put_u32(cmd, NDA_DST, mnl_attr_get_u32(na_ip));
4113 ndm->ndm_family = AF_INET6;
4114 mnl_attr_put(cmd, NDA_DST, IPV6_ADDR_LEN,
4115 mnl_attr_get_payload(na_ip));
4117 mnl_attr_put(cmd, NDA_LLADDR, ETHER_ADDR_LEN,
4118 mnl_attr_get_payload(na_mac));
4119 assert(size == cmd->nlmsg_len);
4124 * Cleanup the neigh rules on outer interface.
4127 * Context object initialized by mlx5_flow_tcf_context_create().
4128 * @param[in] ifindex
4129 * Network inferface index to perform cleanup.
4132 flow_tcf_encap_neigh_cleanup(struct mlx5_flow_tcf_context *tcf,
4133 unsigned int ifindex)
4135 struct nlmsghdr *nlh;
4137 struct tcf_nlcb_context ctx = {
4139 .bufsize = MNL_REQUEST_SIZE,
4140 .nlbuf = LIST_HEAD_INITIALIZER(),
4145 /* Seek and destroy leftovers of neigh rules. */
4146 nlh = mnl_nlmsg_put_header(tcf->buf);
4147 nlh->nlmsg_type = RTM_GETNEIGH;
4148 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4149 ndm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ndm));
4150 ndm->ndm_family = AF_UNSPEC;
4151 ndm->ndm_ifindex = ifindex;
4152 ndm->ndm_state = NUD_PERMANENT;
4153 ret = flow_tcf_nl_ack(tcf, nlh, 0, flow_tcf_collect_neigh_cb, &ctx);
4155 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4156 ret = flow_tcf_send_nlcmd(tcf, &ctx);
4158 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4162 * Collect indices of VXLAN encap/decap interfaces associated with device.
4163 * This is callback routine called by libmnl mnl_cb_run() in loop for
4164 * every message in received packet.
4167 * Pointer to reply header.
4168 * @param[in, out] arg
4169 * Opaque data pointer for this callback.
4172 * A positive, nonzero value on success, negative errno value otherwise
4173 * and rte_errno is set.
4176 flow_tcf_collect_vxlan_cb(const struct nlmsghdr *nlh, void *arg)
4178 struct tcf_nlcb_context *ctx = arg;
4179 struct nlmsghdr *cmd;
4180 struct ifinfomsg *ifm;
4182 struct nlattr *na_info = NULL;
4183 struct nlattr *na_vxlan = NULL;
4185 unsigned int vxindex;
4188 if (nlh->nlmsg_type != RTM_NEWLINK) {
4192 ifm = mnl_nlmsg_get_payload(nlh);
4193 if (!ifm->ifi_index) {
4197 mnl_attr_for_each(na, nlh, sizeof(*ifm))
4198 if (mnl_attr_get_type(na) == IFLA_LINKINFO) {
4204 mnl_attr_for_each_nested(na, na_info) {
4205 switch (mnl_attr_get_type(na)) {
4206 case IFLA_INFO_KIND:
4207 if (!strncmp("vxlan", mnl_attr_get_str(na),
4208 mnl_attr_get_len(na)))
4211 case IFLA_INFO_DATA:
4215 if (found && na_vxlan)
4218 if (!found || !na_vxlan)
4221 mnl_attr_for_each_nested(na, na_vxlan) {
4222 if (mnl_attr_get_type(na) == IFLA_VXLAN_LINK &&
4223 mnl_attr_get_u32(na) == ctx->ifindex) {
4230 /* Attached VXLAN device found, store the command to delete. */
4231 vxindex = ifm->ifi_index;
4232 size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
4233 MNL_ALIGN(sizeof(struct ifinfomsg));
4234 cmd = flow_tcf_alloc_nlcmd(ctx, size);
4239 cmd = mnl_nlmsg_put_header(cmd);
4240 cmd->nlmsg_type = RTM_DELLINK;
4241 cmd->nlmsg_flags = NLM_F_REQUEST;
4242 ifm = mnl_nlmsg_put_extra_header(cmd, sizeof(*ifm));
4243 ifm->ifi_family = AF_UNSPEC;
4244 ifm->ifi_index = vxindex;
4245 assert(size == cmd->nlmsg_len);
4250 * Cleanup the outer interface. Removes all found vxlan devices
4251 * attached to specified index, flushes the meigh and local IP
4255 * Context object initialized by mlx5_flow_tcf_context_create().
4256 * @param[in] ifindex
4257 * Network inferface index to perform cleanup.
4260 flow_tcf_encap_iface_cleanup(struct mlx5_flow_tcf_context *tcf,
4261 unsigned int ifindex)
4263 struct nlmsghdr *nlh;
4264 struct ifinfomsg *ifm;
4265 struct tcf_nlcb_context ctx = {
4267 .bufsize = MNL_REQUEST_SIZE,
4268 .nlbuf = LIST_HEAD_INITIALIZER(),
4274 * Seek and destroy leftover VXLAN encap/decap interfaces with
4275 * matching properties.
4277 nlh = mnl_nlmsg_put_header(tcf->buf);
4278 nlh->nlmsg_type = RTM_GETLINK;
4279 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4280 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4281 ifm->ifi_family = AF_UNSPEC;
4282 ret = flow_tcf_nl_ack(tcf, nlh, 0, flow_tcf_collect_vxlan_cb, &ctx);
4284 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4285 ret = flow_tcf_send_nlcmd(tcf, &ctx);
4287 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4291 * Emit Netlink message to add/remove local address to the outer device.
4292 * The address being added is visible within the link only (scope link).
4294 * Note that an implicit route is maintained by the kernel due to the
4295 * presence of a peer address (IFA_ADDRESS).
4297 * These rules are used for encapsultion only and allow to assign
4298 * the outer tunnel source IP address.
4301 * Libmnl socket context object.
4303 * Encapsulation properties (source address and its peer).
4304 * @param[in] ifindex
4305 * Network interface to apply rule.
4307 * Toggle between add and remove.
4309 * Perform verbose error reporting if not NULL.
4312 * 0 on success, a negative errno value otherwise and rte_errno is set.
4315 flow_tcf_rule_local(struct mlx5_flow_tcf_context *tcf,
4316 const struct flow_tcf_vxlan_encap *encap,
4317 unsigned int ifindex,
4319 struct rte_flow_error *error)
4321 struct nlmsghdr *nlh;
4322 struct ifaddrmsg *ifa;
4323 alignas(struct nlmsghdr)
4324 uint8_t buf[mnl_nlmsg_size(sizeof(*ifa) + 128)];
4326 nlh = mnl_nlmsg_put_header(buf);
4327 nlh->nlmsg_type = enable ? RTM_NEWADDR : RTM_DELADDR;
4329 NLM_F_REQUEST | (enable ? NLM_F_CREATE | NLM_F_REPLACE : 0);
4331 ifa = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifa));
4332 ifa->ifa_flags = IFA_F_PERMANENT;
4333 ifa->ifa_scope = RT_SCOPE_LINK;
4334 ifa->ifa_index = ifindex;
4335 if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4336 ifa->ifa_family = AF_INET;
4337 ifa->ifa_prefixlen = 32;
4338 mnl_attr_put_u32(nlh, IFA_LOCAL, encap->ipv4.src);
4339 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST)
4340 mnl_attr_put_u32(nlh, IFA_ADDRESS,
4343 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4344 ifa->ifa_family = AF_INET6;
4345 ifa->ifa_prefixlen = 128;
4346 mnl_attr_put(nlh, IFA_LOCAL,
4347 sizeof(encap->ipv6.src),
4349 if (encap->mask & FLOW_TCF_ENCAP_IPV6_DST)
4350 mnl_attr_put(nlh, IFA_ADDRESS,
4351 sizeof(encap->ipv6.dst),
4354 if (!flow_tcf_nl_ack(tcf, nlh, 0, NULL, NULL))
4356 return rte_flow_error_set(error, rte_errno,
4357 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4358 "netlink: cannot complete IFA request"
4363 * Emit Netlink message to add/remove neighbor.
4366 * Libmnl socket context object.
4368 * Encapsulation properties (destination address).
4369 * @param[in] ifindex
4370 * Network interface.
4372 * Toggle between add and remove.
4374 * Perform verbose error reporting if not NULL.
4377 * 0 on success, a negative errno value otherwise and rte_errno is set.
4380 flow_tcf_rule_neigh(struct mlx5_flow_tcf_context *tcf,
4381 const struct flow_tcf_vxlan_encap *encap,
4382 unsigned int ifindex,
4384 struct rte_flow_error *error)
4386 struct nlmsghdr *nlh;
4388 alignas(struct nlmsghdr)
4389 uint8_t buf[mnl_nlmsg_size(sizeof(*ndm) + 128)];
4391 nlh = mnl_nlmsg_put_header(buf);
4392 nlh->nlmsg_type = enable ? RTM_NEWNEIGH : RTM_DELNEIGH;
4394 NLM_F_REQUEST | (enable ? NLM_F_CREATE | NLM_F_REPLACE : 0);
4396 ndm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ndm));
4397 ndm->ndm_ifindex = ifindex;
4398 ndm->ndm_state = NUD_PERMANENT;
4401 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4402 ndm->ndm_family = AF_INET;
4403 mnl_attr_put_u32(nlh, NDA_DST, encap->ipv4.dst);
4405 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4406 ndm->ndm_family = AF_INET6;
4407 mnl_attr_put(nlh, NDA_DST, sizeof(encap->ipv6.dst),
4410 if (encap->mask & FLOW_TCF_ENCAP_ETH_SRC && enable)
4412 "outer ethernet source address cannot be "
4413 "forced for VXLAN encapsulation");
4414 if (encap->mask & FLOW_TCF_ENCAP_ETH_DST)
4415 mnl_attr_put(nlh, NDA_LLADDR, sizeof(encap->eth.dst),
4417 if (!flow_tcf_nl_ack(tcf, nlh, 0, NULL, NULL))
4419 return rte_flow_error_set(error, rte_errno,
4420 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4421 "netlink: cannot complete ND request"
4426 * Manage the local IP addresses and their peers IP addresses on the
4427 * outer interface for encapsulation purposes. The kernel searches the
4428 * appropriate device for tunnel egress traffic using the outer source
4429 * IP, this IP should be assigned to the outer network device, otherwise
4430 * kernel rejects the rule.
4432 * Adds or removes the addresses using the Netlink command like this:
4433 * ip addr add <src_ip> peer <dst_ip> scope link dev <ifouter>
4435 * The addresses are local to the netdev ("scope link"), this reduces
4436 * the risk of conflicts. Note that an implicit route is maintained by
4437 * the kernel due to the presence of a peer address (IFA_ADDRESS).
4440 * Libmnl socket context object.
4442 * VTEP object, contains rule database and ifouter index.
4443 * @param[in] dev_flow
4444 * Flow object, contains the tunnel parameters (for encap only).
4446 * Toggle between add and remove.
4448 * Perform verbose error reporting if not NULL.
4451 * 0 on success, a negative errno value otherwise and rte_errno is set.
4454 flow_tcf_encap_local(struct mlx5_flow_tcf_context *tcf,
4455 struct tcf_vtep *vtep,
4456 struct mlx5_flow *dev_flow,
4458 struct rte_flow_error *error)
4460 const struct flow_tcf_vxlan_encap *encap = dev_flow->tcf.vxlan_encap;
4461 struct tcf_local_rule *rule;
4466 assert(encap->hdr.type == FLOW_TCF_TUNACT_VXLAN_ENCAP);
4467 if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4468 assert(encap->mask & FLOW_TCF_ENCAP_IPV4_DST);
4469 LIST_FOREACH(rule, &vtep->local, next) {
4470 if (rule->mask & FLOW_TCF_ENCAP_IPV4_SRC &&
4471 encap->ipv4.src == rule->ipv4.src &&
4472 encap->ipv4.dst == rule->ipv4.dst) {
4478 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4479 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4480 LIST_FOREACH(rule, &vtep->local, next) {
4481 if (rule->mask & FLOW_TCF_ENCAP_IPV6_SRC &&
4482 !memcmp(&encap->ipv6.src, &rule->ipv6.src,
4483 sizeof(encap->ipv6.src)) &&
4484 !memcmp(&encap->ipv6.dst, &rule->ipv6.dst,
4485 sizeof(encap->ipv6.dst))) {
4496 if (!rule->refcnt || !--rule->refcnt) {
4497 LIST_REMOVE(rule, next);
4498 return flow_tcf_rule_local(tcf, encap,
4499 vtep->ifouter, false, error);
4504 DRV_LOG(WARNING, "disabling not existing local rule");
4505 rte_flow_error_set(error, ENOENT,
4506 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4507 "disabling not existing local rule");
4510 rule = rte_zmalloc(__func__, sizeof(struct tcf_local_rule),
4511 alignof(struct tcf_local_rule));
4513 rte_flow_error_set(error, ENOMEM,
4514 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4515 "unable to allocate memory for local rule");
4518 *rule = (struct tcf_local_rule){.refcnt = 0,
4521 if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4522 rule->mask = FLOW_TCF_ENCAP_IPV4_SRC
4523 | FLOW_TCF_ENCAP_IPV4_DST;
4524 rule->ipv4.src = encap->ipv4.src;
4525 rule->ipv4.dst = encap->ipv4.dst;
4527 rule->mask = FLOW_TCF_ENCAP_IPV6_SRC
4528 | FLOW_TCF_ENCAP_IPV6_DST;
4529 memcpy(&rule->ipv6.src, &encap->ipv6.src, IPV6_ADDR_LEN);
4530 memcpy(&rule->ipv6.dst, &encap->ipv6.dst, IPV6_ADDR_LEN);
4532 ret = flow_tcf_rule_local(tcf, encap, vtep->ifouter, true, error);
4538 LIST_INSERT_HEAD(&vtep->local, rule, next);
4543 * Manage the destination MAC/IP addresses neigh database, kernel uses
4544 * this one to determine the destination MAC address within encapsulation
4545 * header. Adds or removes the entries using the Netlink command like this:
4546 * ip neigh add dev <ifouter> lladdr <dst_mac> to <dst_ip> nud permanent
4549 * Libmnl socket context object.
4551 * VTEP object, contains rule database and ifouter index.
4552 * @param[in] dev_flow
4553 * Flow object, contains the tunnel parameters (for encap only).
4555 * Toggle between add and remove.
4557 * Perform verbose error reporting if not NULL.
4560 * 0 on success, a negative errno value otherwise and rte_errno is set.
4563 flow_tcf_encap_neigh(struct mlx5_flow_tcf_context *tcf,
4564 struct tcf_vtep *vtep,
4565 struct mlx5_flow *dev_flow,
4567 struct rte_flow_error *error)
4569 const struct flow_tcf_vxlan_encap *encap = dev_flow->tcf.vxlan_encap;
4570 struct tcf_neigh_rule *rule;
4575 assert(encap->hdr.type == FLOW_TCF_TUNACT_VXLAN_ENCAP);
4576 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4577 assert(encap->mask & FLOW_TCF_ENCAP_IPV4_SRC);
4578 LIST_FOREACH(rule, &vtep->neigh, next) {
4579 if (rule->mask & FLOW_TCF_ENCAP_IPV4_DST &&
4580 encap->ipv4.dst == rule->ipv4.dst) {
4586 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4587 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4588 LIST_FOREACH(rule, &vtep->neigh, next) {
4589 if (rule->mask & FLOW_TCF_ENCAP_IPV6_DST &&
4590 !memcmp(&encap->ipv6.dst, &rule->ipv6.dst,
4591 sizeof(encap->ipv6.dst))) {
4598 if (memcmp(&encap->eth.dst, &rule->eth,
4599 sizeof(encap->eth.dst))) {
4600 DRV_LOG(WARNING, "Destination MAC differs"
4602 rte_flow_error_set(error, EEXIST,
4603 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4604 NULL, "Different MAC address"
4605 " neigh rule for the same"
4613 if (!rule->refcnt || !--rule->refcnt) {
4614 LIST_REMOVE(rule, next);
4615 return flow_tcf_rule_neigh(tcf, encap,
4622 DRV_LOG(WARNING, "Disabling not existing neigh rule");
4623 rte_flow_error_set(error, ENOENT,
4624 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4625 "unable to allocate memory for neigh rule");
4628 rule = rte_zmalloc(__func__, sizeof(struct tcf_neigh_rule),
4629 alignof(struct tcf_neigh_rule));
4631 rte_flow_error_set(error, ENOMEM,
4632 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4633 "unable to allocate memory for neigh rule");
4636 *rule = (struct tcf_neigh_rule){.refcnt = 0,
4639 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4640 rule->mask = FLOW_TCF_ENCAP_IPV4_DST;
4641 rule->ipv4.dst = encap->ipv4.dst;
4643 rule->mask = FLOW_TCF_ENCAP_IPV6_DST;
4644 memcpy(&rule->ipv6.dst, &encap->ipv6.dst, IPV6_ADDR_LEN);
4646 memcpy(&rule->eth, &encap->eth.dst, sizeof(rule->eth));
4647 ret = flow_tcf_rule_neigh(tcf, encap, vtep->ifouter, true, error);
4653 LIST_INSERT_HEAD(&vtep->neigh, rule, next);
4657 /* VTEP device list is shared between PMD port instances. */
4658 static LIST_HEAD(, tcf_vtep) vtep_list_vxlan = LIST_HEAD_INITIALIZER();
4659 static pthread_mutex_t vtep_list_mutex = PTHREAD_MUTEX_INITIALIZER;
4662 * Deletes VTEP network device.
4665 * Context object initialized by mlx5_flow_tcf_context_create().
4667 * Object represinting the network device to delete. Memory
4668 * allocated for this object is freed by routine.
4671 flow_tcf_vtep_delete(struct mlx5_flow_tcf_context *tcf,
4672 struct tcf_vtep *vtep)
4674 struct nlmsghdr *nlh;
4675 struct ifinfomsg *ifm;
4676 alignas(struct nlmsghdr)
4677 uint8_t buf[mnl_nlmsg_size(MNL_ALIGN(sizeof(*ifm))) +
4678 MNL_BUF_EXTRA_SPACE];
4681 assert(!vtep->refcnt);
4682 /* Delete only ifaces those we actually created. */
4683 if (vtep->created && vtep->ifindex) {
4684 DRV_LOG(INFO, "VTEP delete (%d)", vtep->ifindex);
4685 nlh = mnl_nlmsg_put_header(buf);
4686 nlh->nlmsg_type = RTM_DELLINK;
4687 nlh->nlmsg_flags = NLM_F_REQUEST;
4688 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4689 ifm->ifi_family = AF_UNSPEC;
4690 ifm->ifi_index = vtep->ifindex;
4691 assert(sizeof(buf) >= nlh->nlmsg_len);
4692 ret = flow_tcf_nl_ack(tcf, nlh, 0, NULL, NULL);
4694 DRV_LOG(WARNING, "netlink: error deleting vxlan"
4695 " encap/decap ifindex %u",
4702 * Creates VTEP network device.
4705 * Context object initialized by mlx5_flow_tcf_context_create().
4706 * @param[in] ifouter
4707 * Outer interface to attach new-created VXLAN device
4708 * If zero the VXLAN device will not be attached to any device.
4709 * These VTEPs are used for decapsulation and can be precreated
4710 * and shared between processes.
4712 * UDP port of created VTEP device.
4714 * Perform verbose error reporting if not NULL.
4717 * Pointer to created device structure on success,
4718 * NULL otherwise and rte_errno is set.
4720 #ifdef HAVE_IFLA_VXLAN_COLLECT_METADATA
4721 static struct tcf_vtep*
4722 flow_tcf_vtep_create(struct mlx5_flow_tcf_context *tcf,
4723 unsigned int ifouter,
4724 uint16_t port, struct rte_flow_error *error)
4726 struct tcf_vtep *vtep;
4727 struct nlmsghdr *nlh;
4728 struct ifinfomsg *ifm;
4729 char name[sizeof(MLX5_VXLAN_DEVICE_PFX) + 24];
4730 alignas(struct nlmsghdr)
4731 uint8_t buf[mnl_nlmsg_size(sizeof(*ifm)) +
4732 SZ_NLATTR_DATA_OF(sizeof(name)) +
4733 SZ_NLATTR_NEST * 2 +
4734 SZ_NLATTR_STRZ_OF("vxlan") +
4735 SZ_NLATTR_DATA_OF(sizeof(uint32_t)) +
4736 SZ_NLATTR_DATA_OF(sizeof(uint16_t)) +
4737 SZ_NLATTR_DATA_OF(sizeof(uint8_t)) * 3 +
4738 MNL_BUF_EXTRA_SPACE];
4739 struct nlattr *na_info;
4740 struct nlattr *na_vxlan;
4741 rte_be16_t vxlan_port = rte_cpu_to_be_16(port);
4744 vtep = rte_zmalloc(__func__, sizeof(*vtep), alignof(struct tcf_vtep));
4746 rte_flow_error_set(error, ENOMEM,
4747 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4748 "unable to allocate memory for VTEP");
4751 *vtep = (struct tcf_vtep){
4753 .local = LIST_HEAD_INITIALIZER(),
4754 .neigh = LIST_HEAD_INITIALIZER(),
4756 memset(buf, 0, sizeof(buf));
4757 nlh = mnl_nlmsg_put_header(buf);
4758 nlh->nlmsg_type = RTM_NEWLINK;
4759 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
4760 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4761 ifm->ifi_family = AF_UNSPEC;
4764 ifm->ifi_flags = IFF_UP;
4765 ifm->ifi_change = 0xffffffff;
4766 snprintf(name, sizeof(name), "%s%u", MLX5_VXLAN_DEVICE_PFX, port);
4767 mnl_attr_put_strz(nlh, IFLA_IFNAME, name);
4768 na_info = mnl_attr_nest_start(nlh, IFLA_LINKINFO);
4770 mnl_attr_put_strz(nlh, IFLA_INFO_KIND, "vxlan");
4771 na_vxlan = mnl_attr_nest_start(nlh, IFLA_INFO_DATA);
4773 mnl_attr_put_u32(nlh, IFLA_VXLAN_LINK, ifouter);
4775 mnl_attr_put_u8(nlh, IFLA_VXLAN_COLLECT_METADATA, 1);
4776 mnl_attr_put_u8(nlh, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 1);
4777 mnl_attr_put_u8(nlh, IFLA_VXLAN_LEARNING, 0);
4778 mnl_attr_put_u16(nlh, IFLA_VXLAN_PORT, vxlan_port);
4779 mnl_attr_nest_end(nlh, na_vxlan);
4780 mnl_attr_nest_end(nlh, na_info);
4781 assert(sizeof(buf) >= nlh->nlmsg_len);
4782 ret = flow_tcf_nl_ack(tcf, nlh, 0, NULL, NULL);
4785 "netlink: VTEP %s create failure (%d)",
4787 if (rte_errno != EEXIST || ifouter)
4789 * Some unhandled error occurred or device is
4790 * for encapsulation and cannot be shared.
4795 * Mark device we actually created.
4796 * We should explicitly delete
4797 * when we do not need it anymore.
4801 /* Try to get ifindex of created of pre-existing device. */
4802 ret = if_nametoindex(name);
4805 "VTEP %s failed to get index (%d)", name, errno);
4808 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4809 "netlink: failed to retrieve VTEP ifindex");
4812 vtep->ifindex = ret;
4813 vtep->ifouter = ifouter;
4814 memset(buf, 0, sizeof(buf));
4815 nlh = mnl_nlmsg_put_header(buf);
4816 nlh->nlmsg_type = RTM_NEWLINK;
4817 nlh->nlmsg_flags = NLM_F_REQUEST;
4818 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4819 ifm->ifi_family = AF_UNSPEC;
4821 ifm->ifi_index = vtep->ifindex;
4822 ifm->ifi_flags = IFF_UP;
4823 ifm->ifi_change = IFF_UP;
4824 ret = flow_tcf_nl_ack(tcf, nlh, 0, NULL, NULL);
4826 rte_flow_error_set(error, -errno,
4827 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4828 "netlink: failed to set VTEP link up");
4829 DRV_LOG(WARNING, "netlink: VTEP %s set link up failure (%d)",
4833 ret = mlx5_flow_tcf_init(tcf, vtep->ifindex, error);
4835 DRV_LOG(WARNING, "VTEP %s init failure (%d)", name, rte_errno);
4838 DRV_LOG(INFO, "VTEP create (%d, %d)", vtep->port, vtep->ifindex);
4842 flow_tcf_vtep_delete(tcf, vtep);
4849 static struct tcf_vtep*
4850 flow_tcf_vtep_create(struct mlx5_flow_tcf_context *tcf __rte_unused,
4851 unsigned int ifouter __rte_unused,
4852 uint16_t port __rte_unused,
4853 struct rte_flow_error *error)
4855 rte_flow_error_set(error, ENOTSUP,
4856 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4857 "netlink: failed to create VTEP, "
4858 "vxlan metadata are not supported by kernel");
4861 #endif /* HAVE_IFLA_VXLAN_COLLECT_METADATA */
4864 * Acquire target interface index for VXLAN tunneling decapsulation.
4865 * In order to share the UDP port within the other interfaces the
4866 * VXLAN device created as not attached to any interface (if created).
4869 * Context object initialized by mlx5_flow_tcf_context_create().
4870 * @param[in] dev_flow
4871 * Flow tcf object with tunnel structure pointer set.
4873 * Perform verbose error reporting if not NULL.
4875 * Interface descriptor pointer on success,
4876 * NULL otherwise and rte_errno is set.
4878 static struct tcf_vtep*
4879 flow_tcf_decap_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
4880 struct mlx5_flow *dev_flow,
4881 struct rte_flow_error *error)
4883 struct tcf_vtep *vtep;
4884 uint16_t port = dev_flow->tcf.vxlan_decap->udp_port;
4886 LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
4887 if (vtep->port == port)
4890 if (vtep && vtep->ifouter) {
4891 rte_flow_error_set(error, -errno,
4892 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4893 "Failed to create decap VTEP with specified"
4894 " UDP port, atatched device exists");
4898 /* Device exists, just increment the reference counter. */
4900 assert(vtep->ifindex);
4903 /* No decapsulation device exists, try to create the new one. */
4904 vtep = flow_tcf_vtep_create(tcf, 0, port, error);
4906 LIST_INSERT_HEAD(&vtep_list_vxlan, vtep, next);
4911 * Aqcuire target interface index for VXLAN tunneling encapsulation.
4914 * Context object initialized by mlx5_flow_tcf_context_create().
4915 * @param[in] ifouter
4916 * Network interface index to attach VXLAN encap device to.
4917 * @param[in] dev_flow
4918 * Flow tcf object with tunnel structure pointer set.
4920 * Perform verbose error reporting if not NULL.
4922 * Interface descriptor pointer on success,
4923 * NULL otherwise and rte_errno is set.
4925 static struct tcf_vtep*
4926 flow_tcf_encap_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
4927 unsigned int ifouter,
4928 struct mlx5_flow *dev_flow __rte_unused,
4929 struct rte_flow_error *error)
4931 static uint16_t encap_port = MLX5_VXLAN_PORT_MIN - 1;
4932 struct tcf_vtep *vtep;
4936 /* Look whether the attached VTEP for encap is created. */
4937 LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
4938 if (vtep->ifouter == ifouter)
4942 /* VTEP already exists, just increment the reference. */
4947 /* Not found, we should create the new attached VTEP. */
4948 flow_tcf_encap_iface_cleanup(tcf, ifouter);
4949 flow_tcf_encap_local_cleanup(tcf, ifouter);
4950 flow_tcf_encap_neigh_cleanup(tcf, ifouter);
4951 for (pcnt = 0; pcnt <= (MLX5_VXLAN_PORT_MAX
4952 - MLX5_VXLAN_PORT_MIN); pcnt++) {
4954 /* Wraparound the UDP port index. */
4955 if (encap_port < MLX5_VXLAN_PORT_MIN ||
4956 encap_port > MLX5_VXLAN_PORT_MAX)
4957 encap_port = MLX5_VXLAN_PORT_MIN;
4958 /* Check whether UDP port is in already in use. */
4959 LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
4960 if (vtep->port == encap_port)
4964 /* Port is in use, try the next one. */
4968 vtep = flow_tcf_vtep_create(tcf, ifouter,
4971 LIST_INSERT_HEAD(&vtep_list_vxlan, vtep, next);
4974 if (rte_errno != EEXIST)
4980 assert(vtep->ifouter == ifouter);
4981 assert(vtep->ifindex);
4982 /* Create local ipaddr with peer to specify the outer IPs. */
4983 ret = flow_tcf_encap_local(tcf, vtep, dev_flow, true, error);
4985 /* Create neigh rule to specify outer destination MAC. */
4986 ret = flow_tcf_encap_neigh(tcf, vtep, dev_flow, true, error);
4988 flow_tcf_encap_local(tcf, vtep,
4989 dev_flow, false, error);
4992 if (--vtep->refcnt == 0)
4993 flow_tcf_vtep_delete(tcf, vtep);
5000 * Acquires target interface index for tunneling of any type.
5001 * Creates the new VTEP if needed.
5004 * Context object initialized by mlx5_flow_tcf_context_create().
5005 * @param[in] ifouter
5006 * Network interface index to attach VXLAN encap device to.
5007 * @param[in] dev_flow
5008 * Flow tcf object with tunnel structure pointer set.
5010 * Perform verbose error reporting if not NULL.
5012 * Interface descriptor pointer on success,
5013 * NULL otherwise and rte_errno is set.
5015 static struct tcf_vtep*
5016 flow_tcf_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
5017 unsigned int ifouter,
5018 struct mlx5_flow *dev_flow,
5019 struct rte_flow_error *error)
5021 struct tcf_vtep *vtep = NULL;
5023 assert(dev_flow->tcf.tunnel);
5024 pthread_mutex_lock(&vtep_list_mutex);
5025 switch (dev_flow->tcf.tunnel->type) {
5026 case FLOW_TCF_TUNACT_VXLAN_ENCAP:
5027 vtep = flow_tcf_encap_vtep_acquire(tcf, ifouter,
5030 case FLOW_TCF_TUNACT_VXLAN_DECAP:
5031 vtep = flow_tcf_decap_vtep_acquire(tcf, dev_flow, error);
5034 rte_flow_error_set(error, ENOTSUP,
5035 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5036 "unsupported tunnel type");
5039 pthread_mutex_unlock(&vtep_list_mutex);
5044 * Release tunneling interface by ifindex. Decrements reference
5045 * counter and actually removes the device if counter is zero.
5048 * Context object initialized by mlx5_flow_tcf_context_create().
5050 * VTEP device descriptor structure.
5051 * @param[in] dev_flow
5052 * Flow tcf object with tunnel structure pointer set.
5055 flow_tcf_vtep_release(struct mlx5_flow_tcf_context *tcf,
5056 struct tcf_vtep *vtep,
5057 struct mlx5_flow *dev_flow)
5059 assert(dev_flow->tcf.tunnel);
5060 pthread_mutex_lock(&vtep_list_mutex);
5061 switch (dev_flow->tcf.tunnel->type) {
5062 case FLOW_TCF_TUNACT_VXLAN_DECAP:
5064 case FLOW_TCF_TUNACT_VXLAN_ENCAP:
5065 /* Remove the encap ancillary rules first. */
5066 flow_tcf_encap_neigh(tcf, vtep, dev_flow, false, NULL);
5067 flow_tcf_encap_local(tcf, vtep, dev_flow, false, NULL);
5071 DRV_LOG(WARNING, "Unsupported tunnel type");
5074 assert(vtep->refcnt);
5075 if (--vtep->refcnt == 0) {
5076 LIST_REMOVE(vtep, next);
5077 flow_tcf_vtep_delete(tcf, vtep);
5079 pthread_mutex_unlock(&vtep_list_mutex);
5084 * Apply flow to E-Switch by sending Netlink message.
5087 * Pointer to Ethernet device.
5088 * @param[in, out] flow
5089 * Pointer to the sub flow.
5091 * Pointer to the error structure.
5094 * 0 on success, a negative errno value otherwise and rte_ernno is set.
5097 flow_tcf_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
5098 struct rte_flow_error *error)
5100 struct priv *priv = dev->data->dev_private;
5101 struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5102 struct mlx5_flow *dev_flow;
5103 struct nlmsghdr *nlh;
5105 dev_flow = LIST_FIRST(&flow->dev_flows);
5106 /* E-Switch flow can't be expanded. */
5107 assert(!LIST_NEXT(dev_flow, next));
5108 if (dev_flow->tcf.applied)
5110 nlh = dev_flow->tcf.nlh;
5111 nlh->nlmsg_type = RTM_NEWTFILTER;
5112 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
5113 if (dev_flow->tcf.tunnel) {
5115 * Replace the interface index, target for
5116 * encapsulation, source for decapsulation.
5118 assert(!dev_flow->tcf.tunnel->vtep);
5119 assert(dev_flow->tcf.tunnel->ifindex_ptr);
5120 /* Acquire actual VTEP device when rule is being applied. */
5121 dev_flow->tcf.tunnel->vtep =
5122 flow_tcf_vtep_acquire(ctx,
5123 dev_flow->tcf.tunnel->ifindex_org,
5125 if (!dev_flow->tcf.tunnel->vtep)
5127 DRV_LOG(INFO, "Replace ifindex: %d->%d",
5128 dev_flow->tcf.tunnel->vtep->ifindex,
5129 dev_flow->tcf.tunnel->ifindex_org);
5130 *dev_flow->tcf.tunnel->ifindex_ptr =
5131 dev_flow->tcf.tunnel->vtep->ifindex;
5133 if (!flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL)) {
5134 dev_flow->tcf.applied = 1;
5137 return rte_flow_error_set(error, rte_errno,
5138 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5139 "netlink: failed to create TC flow rule");
5143 * Remove flow from E-Switch by sending Netlink message.
5146 * Pointer to Ethernet device.
5147 * @param[in, out] flow
5148 * Pointer to the sub flow.
5151 flow_tcf_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
5153 struct priv *priv = dev->data->dev_private;
5154 struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5155 struct mlx5_flow *dev_flow;
5156 struct nlmsghdr *nlh;
5160 dev_flow = LIST_FIRST(&flow->dev_flows);
5163 /* E-Switch flow can't be expanded. */
5164 assert(!LIST_NEXT(dev_flow, next));
5165 if (dev_flow->tcf.applied) {
5166 nlh = dev_flow->tcf.nlh;
5167 nlh->nlmsg_type = RTM_DELTFILTER;
5168 nlh->nlmsg_flags = NLM_F_REQUEST;
5169 flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL);
5170 if (dev_flow->tcf.tunnel) {
5171 assert(dev_flow->tcf.tunnel->vtep);
5172 flow_tcf_vtep_release(ctx,
5173 dev_flow->tcf.tunnel->vtep,
5175 dev_flow->tcf.tunnel->vtep = NULL;
5177 dev_flow->tcf.applied = 0;
5182 * Remove flow from E-Switch and release resources of the device flow.
5185 * Pointer to Ethernet device.
5186 * @param[in, out] flow
5187 * Pointer to the sub flow.
5190 flow_tcf_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
5192 struct mlx5_flow *dev_flow;
5196 flow_tcf_remove(dev, flow);
5197 if (flow->counter) {
5198 if (--flow->counter->ref_cnt == 0) {
5199 rte_free(flow->counter);
5200 flow->counter = NULL;
5203 dev_flow = LIST_FIRST(&flow->dev_flows);
5206 /* E-Switch flow can't be expanded. */
5207 assert(!LIST_NEXT(dev_flow, next));
5208 LIST_REMOVE(dev_flow, next);
5213 * Helper routine for figuring the space size required for a parse buffer.
5216 * array of values to use.
5218 * Current location in array.
5220 * Value to compare with.
5223 * The maximum between the given value and the array value on index.
5226 flow_tcf_arr_val_max(uint16_t array[], int idx, uint16_t value)
5228 return idx < 0 ? (value) : RTE_MAX((array)[idx], value);
5232 * Parse rtnetlink message attributes filling the attribute table with the info
5236 * Attribute table to be filled.
5238 * Maxinum entry in the attribute table.
5240 * The attributes section in the message to be parsed.
5242 * The length of the attributes section in the message.
5245 flow_tcf_nl_parse_rtattr(struct rtattr *tb[], int max,
5246 struct rtattr *rta, int len)
5248 unsigned short type;
5249 memset(tb, 0, sizeof(struct rtattr *) * (max + 1));
5250 while (RTA_OK(rta, len)) {
5251 type = rta->rta_type;
5252 if (type <= max && !tb[type])
5254 rta = RTA_NEXT(rta, len);
5259 * Extract flow counters from flower action.
5262 * flower action stats properties in the Netlink message received.
5264 * The backward sequence of rta_types, as written in the attribute table,
5265 * we need to traverse in order to get to the requested object.
5267 * Current location in rta_type table.
5269 * data holding the count statistics of the rte_flow retrieved from
5273 * 0 if data was found and retrieved, -1 otherwise.
5276 flow_tcf_nl_action_stats_parse_and_get(struct rtattr *rta,
5277 uint16_t rta_type[], int idx,
5278 struct gnet_stats_basic *data)
5280 int tca_stats_max = flow_tcf_arr_val_max(rta_type, idx,
5282 struct rtattr *tbs[tca_stats_max + 1];
5284 if (rta == NULL || idx < 0)
5286 flow_tcf_nl_parse_rtattr(tbs, tca_stats_max,
5287 RTA_DATA(rta), RTA_PAYLOAD(rta));
5288 switch (rta_type[idx]) {
5289 case TCA_STATS_BASIC:
5290 if (tbs[TCA_STATS_BASIC]) {
5291 memcpy(data, RTA_DATA(tbs[TCA_STATS_BASIC]),
5292 RTE_MIN(RTA_PAYLOAD(tbs[TCA_STATS_BASIC]),
5304 * Parse flower single action retrieving the requested action attribute,
5308 * flower action properties in the Netlink message received.
5310 * The backward sequence of rta_types, as written in the attribute table,
5311 * we need to traverse in order to get to the requested object.
5313 * Current location in rta_type table.
5315 * Count statistics retrieved from the message query.
5318 * 0 if data was found and retrieved, -1 otherwise.
5321 flow_tcf_nl_parse_one_action_and_get(struct rtattr *arg,
5322 uint16_t rta_type[], int idx, void *data)
5324 int tca_act_max = flow_tcf_arr_val_max(rta_type, idx, TCA_ACT_STATS);
5325 struct rtattr *tb[tca_act_max + 1];
5327 if (arg == NULL || idx < 0)
5329 flow_tcf_nl_parse_rtattr(tb, tca_act_max,
5330 RTA_DATA(arg), RTA_PAYLOAD(arg));
5331 if (tb[TCA_ACT_KIND] == NULL)
5333 switch (rta_type[idx]) {
5335 if (tb[TCA_ACT_STATS])
5336 return flow_tcf_nl_action_stats_parse_and_get
5339 (struct gnet_stats_basic *)data);
5348 * Parse flower action section in the message retrieving the requested
5349 * attribute from the first action that provides it.
5352 * flower section in the Netlink message received.
5354 * The backward sequence of rta_types, as written in the attribute table,
5355 * we need to traverse in order to get to the requested object.
5357 * Current location in rta_type table.
5359 * data retrieved from the message query.
5362 * 0 if data was found and retrieved, -1 otherwise.
5365 flow_tcf_nl_action_parse_and_get(struct rtattr *arg,
5366 uint16_t rta_type[], int idx, void *data)
5368 struct rtattr *tb[TCA_ACT_MAX_PRIO + 1];
5371 if (arg == NULL || idx < 0)
5373 flow_tcf_nl_parse_rtattr(tb, TCA_ACT_MAX_PRIO,
5374 RTA_DATA(arg), RTA_PAYLOAD(arg));
5375 switch (rta_type[idx]) {
5377 * flow counters are stored in the actions defined by the flow
5378 * and not in the flow itself, therefore we need to traverse the
5379 * flower chain of actions in search for them.
5381 * Note that the index is not decremented here.
5384 for (i = 0; i <= TCA_ACT_MAX_PRIO; i++) {
5386 !flow_tcf_nl_parse_one_action_and_get(tb[i],
5399 * Parse flower classifier options in the message, retrieving the requested
5400 * attribute if found.
5403 * flower section in the Netlink message received.
5405 * The backward sequence of rta_types, as written in the attribute table,
5406 * we need to traverse in order to get to the requested object.
5408 * Current location in rta_type table.
5410 * data retrieved from the message query.
5413 * 0 if data was found and retrieved, -1 otherwise.
5416 flow_tcf_nl_opts_parse_and_get(struct rtattr *opt,
5417 uint16_t rta_type[], int idx, void *data)
5419 int tca_flower_max = flow_tcf_arr_val_max(rta_type, idx,
5421 struct rtattr *tb[tca_flower_max + 1];
5423 if (!opt || idx < 0)
5425 flow_tcf_nl_parse_rtattr(tb, tca_flower_max,
5426 RTA_DATA(opt), RTA_PAYLOAD(opt));
5427 switch (rta_type[idx]) {
5428 case TCA_FLOWER_ACT:
5429 if (tb[TCA_FLOWER_ACT])
5430 return flow_tcf_nl_action_parse_and_get
5431 (tb[TCA_FLOWER_ACT],
5432 rta_type, --idx, data);
5441 * Parse Netlink reply on filter query, retrieving the flow counters.
5444 * Message received from Netlink.
5446 * The backward sequence of rta_types, as written in the attribute table,
5447 * we need to traverse in order to get to the requested object.
5449 * Current location in rta_type table.
5451 * data retrieved from the message query.
5454 * 0 if data was found and retrieved, -1 otherwise.
5457 flow_tcf_nl_filter_parse_and_get(struct nlmsghdr *cnlh,
5458 uint16_t rta_type[], int idx, void *data)
5460 struct nlmsghdr *nlh = cnlh;
5461 struct tcmsg *t = NLMSG_DATA(nlh);
5462 int len = nlh->nlmsg_len;
5463 int tca_max = flow_tcf_arr_val_max(rta_type, idx, TCA_OPTIONS);
5464 struct rtattr *tb[tca_max + 1];
5468 if (nlh->nlmsg_type != RTM_NEWTFILTER &&
5469 nlh->nlmsg_type != RTM_GETTFILTER &&
5470 nlh->nlmsg_type != RTM_DELTFILTER)
5472 len -= NLMSG_LENGTH(sizeof(*t));
5475 flow_tcf_nl_parse_rtattr(tb, tca_max, TCA_RTA(t), len);
5476 /* Not a TC flower flow - bail out */
5477 if (!tb[TCA_KIND] ||
5478 strcmp(RTA_DATA(tb[TCA_KIND]), "flower"))
5480 switch (rta_type[idx]) {
5482 if (tb[TCA_OPTIONS])
5483 return flow_tcf_nl_opts_parse_and_get(tb[TCA_OPTIONS],
5494 * A callback to parse Netlink reply on TC flower query.
5497 * Message received from Netlink.
5499 * Pointer to data area to be filled by the parsing routine.
5500 * assumed to be a pinter to struct flow_tcf_stats_basic.
5506 flow_tcf_nl_message_get_stats_basic(const struct nlmsghdr *nlh, void *data)
5509 * The backward sequence of rta_types to pass in order to get
5512 uint16_t rta_type[] = { TCA_STATS_BASIC, TCA_ACT_STATS,
5513 TCA_FLOWER_ACT, TCA_OPTIONS };
5514 struct flow_tcf_stats_basic *sb_data = data;
5516 const struct nlmsghdr *c;
5517 struct nlmsghdr *nc;
5518 } tnlh = { .c = nlh };
5520 if (!flow_tcf_nl_filter_parse_and_get(tnlh.nc, rta_type,
5521 RTE_DIM(rta_type) - 1,
5522 (void *)&sb_data->counters))
5523 sb_data->valid = true;
5528 * Query a TC flower rule for its statistics via netlink.
5531 * Pointer to Ethernet device.
5533 * Pointer to the sub flow.
5535 * data retrieved by the query.
5537 * Perform verbose error reporting if not NULL.
5540 * 0 on success, a negative errno value otherwise and rte_errno is set.
5543 flow_tcf_query_count(struct rte_eth_dev *dev,
5544 struct rte_flow *flow,
5546 struct rte_flow_error *error)
5548 struct flow_tcf_stats_basic sb_data = { 0 };
5549 struct rte_flow_query_count *qc = data;
5550 struct priv *priv = dev->data->dev_private;
5551 struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5552 struct mnl_socket *nl = ctx->nl;
5553 struct mlx5_flow *dev_flow;
5554 struct nlmsghdr *nlh;
5555 uint32_t seq = priv->tcf_context->seq++;
5559 dev_flow = LIST_FIRST(&flow->dev_flows);
5560 /* E-Switch flow can't be expanded. */
5561 assert(!LIST_NEXT(dev_flow, next));
5562 if (!dev_flow->flow->counter)
5564 nlh = dev_flow->tcf.nlh;
5565 nlh->nlmsg_type = RTM_GETTFILTER;
5566 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ECHO;
5567 nlh->nlmsg_seq = seq;
5568 if (mnl_socket_sendto(nl, nlh, nlh->nlmsg_len) == -1)
5571 ret = mnl_socket_recvfrom(nl, ctx->buf, ctx->buf_size);
5574 ret = mnl_cb_run(ctx->buf, ret, seq,
5575 mnl_socket_get_portid(nl),
5576 flow_tcf_nl_message_get_stats_basic,
5579 /* Return the delta from last reset. */
5580 if (sb_data.valid) {
5581 /* Return the delta from last reset. */
5584 qc->hits = sb_data.counters.packets - flow->counter->hits;
5585 qc->bytes = sb_data.counters.bytes - flow->counter->bytes;
5587 flow->counter->hits = sb_data.counters.packets;
5588 flow->counter->bytes = sb_data.counters.bytes;
5592 return rte_flow_error_set(error, EINVAL,
5593 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5595 "flow does not have counter");
5597 return rte_flow_error_set
5598 (error, errno, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5599 NULL, "netlink: failed to read flow rule counters");
5601 return rte_flow_error_set
5602 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5603 NULL, "counters are not available.");
5609 * @see rte_flow_query()
5613 flow_tcf_query(struct rte_eth_dev *dev,
5614 struct rte_flow *flow,
5615 const struct rte_flow_action *actions,
5617 struct rte_flow_error *error)
5621 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5622 switch (actions->type) {
5623 case RTE_FLOW_ACTION_TYPE_VOID:
5625 case RTE_FLOW_ACTION_TYPE_COUNT:
5626 ret = flow_tcf_query_count(dev, flow, data, error);
5629 return rte_flow_error_set(error, ENOTSUP,
5630 RTE_FLOW_ERROR_TYPE_ACTION,
5632 "action not supported");
5638 const struct mlx5_flow_driver_ops mlx5_flow_tcf_drv_ops = {
5639 .validate = flow_tcf_validate,
5640 .prepare = flow_tcf_prepare,
5641 .translate = flow_tcf_translate,
5642 .apply = flow_tcf_apply,
5643 .remove = flow_tcf_remove,
5644 .destroy = flow_tcf_destroy,
5645 .query = flow_tcf_query,
5649 * Create and configure a libmnl socket for Netlink flow rules.
5652 * A valid libmnl socket object pointer on success, NULL otherwise and
5655 static struct mnl_socket *
5656 flow_tcf_mnl_socket_create(void)
5658 struct mnl_socket *nl = mnl_socket_open(NETLINK_ROUTE);
5661 mnl_socket_setsockopt(nl, NETLINK_CAP_ACK, &(int){ 1 },
5663 if (!mnl_socket_bind(nl, 0, MNL_SOCKET_AUTOPID))
5668 mnl_socket_close(nl);
5673 * Destroy a libmnl socket.
5676 * Libmnl socket of the @p NETLINK_ROUTE kind.
5679 flow_tcf_mnl_socket_destroy(struct mnl_socket *nl)
5682 mnl_socket_close(nl);
5686 * Initialize ingress qdisc of a given network interface.
5689 * Pointer to tc-flower context to use.
5691 * Index of network interface to initialize.
5693 * Perform verbose error reporting if not NULL.
5696 * 0 on success, a negative errno value otherwise and rte_errno is set.
5699 mlx5_flow_tcf_init(struct mlx5_flow_tcf_context *ctx,
5700 unsigned int ifindex, struct rte_flow_error *error)
5702 struct nlmsghdr *nlh;
5704 alignas(struct nlmsghdr)
5705 uint8_t buf[mnl_nlmsg_size(sizeof(*tcm)) +
5706 SZ_NLATTR_STRZ_OF("ingress") +
5707 MNL_BUF_EXTRA_SPACE];
5709 /* Destroy existing ingress qdisc and everything attached to it. */
5710 nlh = mnl_nlmsg_put_header(buf);
5711 nlh->nlmsg_type = RTM_DELQDISC;
5712 nlh->nlmsg_flags = NLM_F_REQUEST;
5713 tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
5714 tcm->tcm_family = AF_UNSPEC;
5715 tcm->tcm_ifindex = ifindex;
5716 tcm->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
5717 tcm->tcm_parent = TC_H_INGRESS;
5718 assert(sizeof(buf) >= nlh->nlmsg_len);
5719 /* Ignore errors when qdisc is already absent. */
5720 if (flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL) &&
5721 rte_errno != EINVAL && rte_errno != ENOENT)
5722 return rte_flow_error_set(error, rte_errno,
5723 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5724 "netlink: failed to remove ingress"
5726 /* Create fresh ingress qdisc. */
5727 nlh = mnl_nlmsg_put_header(buf);
5728 nlh->nlmsg_type = RTM_NEWQDISC;
5729 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
5730 tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
5731 tcm->tcm_family = AF_UNSPEC;
5732 tcm->tcm_ifindex = ifindex;
5733 tcm->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
5734 tcm->tcm_parent = TC_H_INGRESS;
5735 mnl_attr_put_strz_check(nlh, sizeof(buf), TCA_KIND, "ingress");
5736 assert(sizeof(buf) >= nlh->nlmsg_len);
5737 if (flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL))
5738 return rte_flow_error_set(error, rte_errno,
5739 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5740 "netlink: failed to create ingress"
5746 * Create libmnl context for Netlink flow rules.
5749 * A valid libmnl socket object pointer on success, NULL otherwise and
5752 struct mlx5_flow_tcf_context *
5753 mlx5_flow_tcf_context_create(void)
5755 struct mlx5_flow_tcf_context *ctx = rte_zmalloc(__func__,
5760 ctx->nl = flow_tcf_mnl_socket_create();
5763 ctx->buf_size = MNL_SOCKET_BUFFER_SIZE;
5764 ctx->buf = rte_zmalloc(__func__,
5765 ctx->buf_size, sizeof(uint32_t));
5768 ctx->seq = random();
5771 mlx5_flow_tcf_context_destroy(ctx);
5776 * Destroy a libmnl context.
5779 * Libmnl socket of the @p NETLINK_ROUTE kind.
5782 mlx5_flow_tcf_context_destroy(struct mlx5_flow_tcf_context *ctx)
5786 flow_tcf_mnl_socket_destroy(ctx->nl);