1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2017 6WIND S.A.
3 * Copyright 2017 Mellanox Technologies, Ltd
10 #include <sys/resource.h>
12 #include <rte_byteorder.h>
13 #include <rte_jhash.h>
14 #include <rte_malloc.h>
15 #include <rte_eth_tap.h>
17 #include <tap_autoconf.h>
18 #include <tap_tcmsgs.h>
21 #ifndef HAVE_TC_FLOWER
23 * For kernels < 4.2, this enum is not defined. Runtime checks will be made to
24 * avoid sending TC messages the kernel cannot understand.
31 TCA_FLOWER_KEY_ETH_DST
, /* ETH_ALEN */
32 TCA_FLOWER_KEY_ETH_DST_MASK
, /* ETH_ALEN */
33 TCA_FLOWER_KEY_ETH_SRC
, /* ETH_ALEN */
34 TCA_FLOWER_KEY_ETH_SRC_MASK
, /* ETH_ALEN */
35 TCA_FLOWER_KEY_ETH_TYPE
, /* be16 */
36 TCA_FLOWER_KEY_IP_PROTO
, /* u8 */
37 TCA_FLOWER_KEY_IPV4_SRC
, /* be32 */
38 TCA_FLOWER_KEY_IPV4_SRC_MASK
, /* be32 */
39 TCA_FLOWER_KEY_IPV4_DST
, /* be32 */
40 TCA_FLOWER_KEY_IPV4_DST_MASK
, /* be32 */
41 TCA_FLOWER_KEY_IPV6_SRC
, /* struct in6_addr */
42 TCA_FLOWER_KEY_IPV6_SRC_MASK
, /* struct in6_addr */
43 TCA_FLOWER_KEY_IPV6_DST
, /* struct in6_addr */
44 TCA_FLOWER_KEY_IPV6_DST_MASK
, /* struct in6_addr */
45 TCA_FLOWER_KEY_TCP_SRC
, /* be16 */
46 TCA_FLOWER_KEY_TCP_DST
, /* be16 */
47 TCA_FLOWER_KEY_UDP_SRC
, /* be16 */
48 TCA_FLOWER_KEY_UDP_DST
, /* be16 */
51 #ifndef HAVE_TC_VLAN_ID
53 /* TCA_FLOWER_FLAGS, */
54 TCA_FLOWER_KEY_VLAN_ID
= TCA_FLOWER_KEY_UDP_DST
+ 2, /* be16 */
55 TCA_FLOWER_KEY_VLAN_PRIO
, /* u8 */
56 TCA_FLOWER_KEY_VLAN_ETH_TYPE
, /* be16 */
60 * For kernels < 4.2 BPF related enums may not be defined.
61 * Runtime checks will be carried out to gracefully report on TC messages that
62 * are rejected by the kernel. Rejection reasons may be due to:
63 * 1. enum is not defined
64 * 2. enum is defined but kernel is not configured to support BPF system calls,
65 * BPF classifications or BPF actions.
77 #ifndef HAVE_TC_BPF_FD
79 TCA_BPF_FD
= TCA_BPF_OPS
+ 1,
83 #ifndef HAVE_TC_ACT_BPF
104 #ifndef HAVE_TC_ACT_BPF_FD
106 TCA_ACT_BPF_FD
= TCA_ACT_BPF_OPS
+ 1,
111 /* RSS key management */
125 #define ISOLATE_HANDLE 1
126 #define REMOTE_PROMISCUOUS_HANDLE 2
129 LIST_ENTRY(rte_flow
) next
; /* Pointer to the next rte_flow structure */
130 struct rte_flow
*remote_flow
; /* associated remote flow */
131 int bpf_fd
[SEC_MAX
]; /* list of bfs fds per ELF section */
132 uint32_t key_idx
; /* RSS rule key index into BPF map */
136 struct convert_data
{
140 struct rte_flow
*flow
;
144 struct rte_flow_attr attr
;
145 struct rte_flow_item items
[2];
146 struct rte_flow_action actions
[2];
155 struct tc_mirred mirred
;
157 struct tc_skbedit skbedit
;
161 struct tc_act_bpf bpf
;
163 const char *annotation
;
168 static int tap_flow_create_eth(const struct rte_flow_item
*item
, void *data
);
169 static int tap_flow_create_vlan(const struct rte_flow_item
*item
, void *data
);
170 static int tap_flow_create_ipv4(const struct rte_flow_item
*item
, void *data
);
171 static int tap_flow_create_ipv6(const struct rte_flow_item
*item
, void *data
);
172 static int tap_flow_create_udp(const struct rte_flow_item
*item
, void *data
);
173 static int tap_flow_create_tcp(const struct rte_flow_item
*item
, void *data
);
175 tap_flow_validate(struct rte_eth_dev
*dev
,
176 const struct rte_flow_attr
*attr
,
177 const struct rte_flow_item items
[],
178 const struct rte_flow_action actions
[],
179 struct rte_flow_error
*error
);
181 static struct rte_flow
*
182 tap_flow_create(struct rte_eth_dev
*dev
,
183 const struct rte_flow_attr
*attr
,
184 const struct rte_flow_item items
[],
185 const struct rte_flow_action actions
[],
186 struct rte_flow_error
*error
);
189 tap_flow_free(struct pmd_internals
*pmd
,
190 struct rte_flow
*flow
);
193 tap_flow_destroy(struct rte_eth_dev
*dev
,
194 struct rte_flow
*flow
,
195 struct rte_flow_error
*error
);
198 tap_flow_isolate(struct rte_eth_dev
*dev
,
200 struct rte_flow_error
*error
);
202 static int bpf_rss_key(enum bpf_rss_key_e cmd
, __u32
*key_idx
);
203 static int rss_enable(struct pmd_internals
*pmd
,
204 const struct rte_flow_attr
*attr
,
205 struct rte_flow_error
*error
);
206 static int rss_add_actions(struct rte_flow
*flow
, struct pmd_internals
*pmd
,
207 const struct rte_flow_action_rss
*rss
,
208 struct rte_flow_error
*error
);
210 static const struct rte_flow_ops tap_flow_ops
= {
211 .validate
= tap_flow_validate
,
212 .create
= tap_flow_create
,
213 .destroy
= tap_flow_destroy
,
214 .flush
= tap_flow_flush
,
215 .isolate
= tap_flow_isolate
,
218 /* Static initializer for items. */
220 (const enum rte_flow_item_type []){ \
221 __VA_ARGS__, RTE_FLOW_ITEM_TYPE_END, \
224 /* Structure to generate a simple graph of layers supported by the NIC. */
225 struct tap_flow_items
{
226 /* Bit-mask corresponding to what is supported for this item. */
228 const unsigned int mask_sz
; /* Bit-mask size in bytes. */
230 * Bit-mask corresponding to the default mask, if none is provided
231 * along with the item.
233 const void *default_mask
;
235 * Conversion function from rte_flow to netlink attributes.
238 * rte_flow item to convert.
240 * Internal structure to store the conversion.
243 * 0 on success, negative value otherwise.
245 int (*convert
)(const struct rte_flow_item
*item
, void *data
);
246 /** List of possible following items. */
247 const enum rte_flow_item_type
*const items
;
250 /* Graph of supported items and associated actions. */
251 static const struct tap_flow_items tap_flow_items
[] = {
252 [RTE_FLOW_ITEM_TYPE_END
] = {
253 .items
= ITEMS(RTE_FLOW_ITEM_TYPE_ETH
),
255 [RTE_FLOW_ITEM_TYPE_ETH
] = {
257 RTE_FLOW_ITEM_TYPE_VLAN
,
258 RTE_FLOW_ITEM_TYPE_IPV4
,
259 RTE_FLOW_ITEM_TYPE_IPV6
),
260 .mask
= &(const struct rte_flow_item_eth
){
261 .dst
.addr_bytes
= "\xff\xff\xff\xff\xff\xff",
262 .src
.addr_bytes
= "\xff\xff\xff\xff\xff\xff",
265 .mask_sz
= sizeof(struct rte_flow_item_eth
),
266 .default_mask
= &rte_flow_item_eth_mask
,
267 .convert
= tap_flow_create_eth
,
269 [RTE_FLOW_ITEM_TYPE_VLAN
] = {
270 .items
= ITEMS(RTE_FLOW_ITEM_TYPE_IPV4
,
271 RTE_FLOW_ITEM_TYPE_IPV6
),
272 .mask
= &(const struct rte_flow_item_vlan
){
273 /* DEI matching is not supported */
274 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
281 .mask_sz
= sizeof(struct rte_flow_item_vlan
),
282 .default_mask
= &rte_flow_item_vlan_mask
,
283 .convert
= tap_flow_create_vlan
,
285 [RTE_FLOW_ITEM_TYPE_IPV4
] = {
286 .items
= ITEMS(RTE_FLOW_ITEM_TYPE_UDP
,
287 RTE_FLOW_ITEM_TYPE_TCP
),
288 .mask
= &(const struct rte_flow_item_ipv4
){
295 .mask_sz
= sizeof(struct rte_flow_item_ipv4
),
296 .default_mask
= &rte_flow_item_ipv4_mask
,
297 .convert
= tap_flow_create_ipv4
,
299 [RTE_FLOW_ITEM_TYPE_IPV6
] = {
300 .items
= ITEMS(RTE_FLOW_ITEM_TYPE_UDP
,
301 RTE_FLOW_ITEM_TYPE_TCP
),
302 .mask
= &(const struct rte_flow_item_ipv6
){
305 "\xff\xff\xff\xff\xff\xff\xff\xff"
306 "\xff\xff\xff\xff\xff\xff\xff\xff",
309 "\xff\xff\xff\xff\xff\xff\xff\xff"
310 "\xff\xff\xff\xff\xff\xff\xff\xff",
315 .mask_sz
= sizeof(struct rte_flow_item_ipv6
),
316 .default_mask
= &rte_flow_item_ipv6_mask
,
317 .convert
= tap_flow_create_ipv6
,
319 [RTE_FLOW_ITEM_TYPE_UDP
] = {
320 .mask
= &(const struct rte_flow_item_udp
){
326 .mask_sz
= sizeof(struct rte_flow_item_udp
),
327 .default_mask
= &rte_flow_item_udp_mask
,
328 .convert
= tap_flow_create_udp
,
330 [RTE_FLOW_ITEM_TYPE_TCP
] = {
331 .mask
= &(const struct rte_flow_item_tcp
){
337 .mask_sz
= sizeof(struct rte_flow_item_tcp
),
338 .default_mask
= &rte_flow_item_tcp_mask
,
339 .convert
= tap_flow_create_tcp
,
344 * TC rules, by growing priority
346 * Remote netdevice Tap netdevice
347 * +-------------+-------------+ +-------------+-------------+
348 * | Ingress | Egress | | Ingress | Egress |
349 * |-------------|-------------| |-------------|-------------|
350 * | | \ / | | | REMOTE TX | prio 1
351 * | | \ / | | | \ / | prio 2
352 * | EXPLICIT | \ / | | EXPLICIT | \ / | .
353 * | | \ / | | | \ / | .
354 * | RULES | X | | RULES | X | .
355 * | . | / \ | | . | / \ | .
356 * | . | / \ | | . | / \ | .
357 * | . | / \ | | . | / \ | .
358 * | . | / \ | | . | / \ | .
360 * .... .... .... ....
362 * | . | \ / | | . | \ / | .
363 * | . | \ / | | . | \ / | .
364 * | | \ / | | | \ / |
365 * | LOCAL_MAC | \ / | | \ / | \ / | last prio - 5
366 * | PROMISC | X | | \ / | X | last prio - 4
367 * | ALLMULTI | / \ | | X | / \ | last prio - 3
368 * | BROADCAST | / \ | | / \ | / \ | last prio - 2
369 * | BROADCASTV6 | / \ | | / \ | / \ | last prio - 1
370 * | xx | / \ | | ISOLATE | / \ | last prio
371 * +-------------+-------------+ +-------------+-------------+
373 * The implicit flow rules are stored in a list in with mandatorily the last two
374 * being the ISOLATE and REMOTE_TX rules. e.g.:
376 * LOCAL_MAC -> BROADCAST -> BROADCASTV6 -> REMOTE_TX -> ISOLATE -> NULL
378 * That enables tap_flow_isolate() to remove implicit rules by popping the list
379 * head and remove it as long as it applies on the remote netdevice. The
380 * implicit rule for TX redirection is not removed, as isolate concerns only
384 static struct remote_rule implicit_rte_flows
[TAP_REMOTE_MAX_IDX
] = {
385 [TAP_REMOTE_LOCAL_MAC
] = {
388 .priority
= PRIORITY_MASK
- TAP_REMOTE_LOCAL_MAC
,
392 .type
= RTE_FLOW_ITEM_TYPE_ETH
,
393 .mask
= &(const struct rte_flow_item_eth
){
394 .dst
.addr_bytes
= "\xff\xff\xff\xff\xff\xff",
398 .type
= RTE_FLOW_ITEM_TYPE_END
,
400 .mirred
= TCA_EGRESS_REDIR
,
402 [TAP_REMOTE_BROADCAST
] = {
405 .priority
= PRIORITY_MASK
- TAP_REMOTE_BROADCAST
,
409 .type
= RTE_FLOW_ITEM_TYPE_ETH
,
410 .mask
= &(const struct rte_flow_item_eth
){
411 .dst
.addr_bytes
= "\xff\xff\xff\xff\xff\xff",
413 .spec
= &(const struct rte_flow_item_eth
){
414 .dst
.addr_bytes
= "\xff\xff\xff\xff\xff\xff",
418 .type
= RTE_FLOW_ITEM_TYPE_END
,
420 .mirred
= TCA_EGRESS_MIRROR
,
422 [TAP_REMOTE_BROADCASTV6
] = {
425 .priority
= PRIORITY_MASK
- TAP_REMOTE_BROADCASTV6
,
429 .type
= RTE_FLOW_ITEM_TYPE_ETH
,
430 .mask
= &(const struct rte_flow_item_eth
){
431 .dst
.addr_bytes
= "\x33\x33\x00\x00\x00\x00",
433 .spec
= &(const struct rte_flow_item_eth
){
434 .dst
.addr_bytes
= "\x33\x33\x00\x00\x00\x00",
438 .type
= RTE_FLOW_ITEM_TYPE_END
,
440 .mirred
= TCA_EGRESS_MIRROR
,
442 [TAP_REMOTE_PROMISC
] = {
445 .priority
= PRIORITY_MASK
- TAP_REMOTE_PROMISC
,
449 .type
= RTE_FLOW_ITEM_TYPE_VOID
,
452 .type
= RTE_FLOW_ITEM_TYPE_END
,
454 .mirred
= TCA_EGRESS_MIRROR
,
456 [TAP_REMOTE_ALLMULTI
] = {
459 .priority
= PRIORITY_MASK
- TAP_REMOTE_ALLMULTI
,
463 .type
= RTE_FLOW_ITEM_TYPE_ETH
,
464 .mask
= &(const struct rte_flow_item_eth
){
465 .dst
.addr_bytes
= "\x01\x00\x00\x00\x00\x00",
467 .spec
= &(const struct rte_flow_item_eth
){
468 .dst
.addr_bytes
= "\x01\x00\x00\x00\x00\x00",
472 .type
= RTE_FLOW_ITEM_TYPE_END
,
474 .mirred
= TCA_EGRESS_MIRROR
,
479 .priority
= TAP_REMOTE_TX
,
483 .type
= RTE_FLOW_ITEM_TYPE_VOID
,
486 .type
= RTE_FLOW_ITEM_TYPE_END
,
488 .mirred
= TCA_EGRESS_MIRROR
,
493 .priority
= PRIORITY_MASK
- TAP_ISOLATE
,
497 .type
= RTE_FLOW_ITEM_TYPE_VOID
,
500 .type
= RTE_FLOW_ITEM_TYPE_END
,
506 * Make as much checks as possible on an Ethernet item, and if a flow is
507 * provided, fill it appropriately with Ethernet info.
510 * Item specification.
511 * @param[in, out] data
512 * Additional data structure to tell next layers we've been here.
515 * 0 if checks are alright, -1 otherwise.
518 tap_flow_create_eth(const struct rte_flow_item
*item
, void *data
)
520 struct convert_data
*info
= (struct convert_data
*)data
;
521 const struct rte_flow_item_eth
*spec
= item
->spec
;
522 const struct rte_flow_item_eth
*mask
= item
->mask
;
523 struct rte_flow
*flow
= info
->flow
;
526 /* use default mask if none provided */
528 mask
= tap_flow_items
[RTE_FLOW_ITEM_TYPE_ETH
].default_mask
;
529 /* TC does not support eth_type masking. Only accept if exact match. */
530 if (mask
->type
&& mask
->type
!= 0xffff)
534 /* store eth_type for consistency if ipv4/6 pattern item comes next */
535 if (spec
->type
& mask
->type
)
536 info
->eth_type
= spec
->type
;
540 if (!rte_is_zero_ether_addr(&mask
->dst
)) {
541 tap_nlattr_add(&msg
->nh
, TCA_FLOWER_KEY_ETH_DST
,
543 &spec
->dst
.addr_bytes
);
544 tap_nlattr_add(&msg
->nh
,
545 TCA_FLOWER_KEY_ETH_DST_MASK
, RTE_ETHER_ADDR_LEN
,
546 &mask
->dst
.addr_bytes
);
548 if (!rte_is_zero_ether_addr(&mask
->src
)) {
549 tap_nlattr_add(&msg
->nh
, TCA_FLOWER_KEY_ETH_SRC
,
551 &spec
->src
.addr_bytes
);
552 tap_nlattr_add(&msg
->nh
,
553 TCA_FLOWER_KEY_ETH_SRC_MASK
, RTE_ETHER_ADDR_LEN
,
554 &mask
->src
.addr_bytes
);
560 * Make as much checks as possible on a VLAN item, and if a flow is provided,
561 * fill it appropriately with VLAN info.
564 * Item specification.
565 * @param[in, out] data
566 * Additional data structure to tell next layers we've been here.
569 * 0 if checks are alright, -1 otherwise.
572 tap_flow_create_vlan(const struct rte_flow_item
*item
, void *data
)
574 struct convert_data
*info
= (struct convert_data
*)data
;
575 const struct rte_flow_item_vlan
*spec
= item
->spec
;
576 const struct rte_flow_item_vlan
*mask
= item
->mask
;
577 struct rte_flow
*flow
= info
->flow
;
580 /* use default mask if none provided */
582 mask
= tap_flow_items
[RTE_FLOW_ITEM_TYPE_VLAN
].default_mask
;
583 /* Outer TPID cannot be matched. */
586 /* Double-tagging not supported. */
590 if (mask
->inner_type
) {
591 /* TC does not support partial eth_type masking */
592 if (mask
->inner_type
!= RTE_BE16(0xffff))
594 info
->eth_type
= spec
->inner_type
;
599 msg
->t
.tcm_info
= TC_H_MAKE(msg
->t
.tcm_info
, htons(ETH_P_8021Q
));
600 #define VLAN_PRIO(tci) ((tci) >> 13)
601 #define VLAN_ID(tci) ((tci) & 0xfff)
605 uint16_t tci
= ntohs(spec
->tci
) & mask
->tci
;
606 uint16_t prio
= VLAN_PRIO(tci
);
607 uint8_t vid
= VLAN_ID(tci
);
610 tap_nlattr_add8(&msg
->nh
,
611 TCA_FLOWER_KEY_VLAN_PRIO
, prio
);
613 tap_nlattr_add16(&msg
->nh
,
614 TCA_FLOWER_KEY_VLAN_ID
, vid
);
620 * Make as much checks as possible on an IPv4 item, and if a flow is provided,
621 * fill it appropriately with IPv4 info.
624 * Item specification.
625 * @param[in, out] data
626 * Additional data structure to tell next layers we've been here.
629 * 0 if checks are alright, -1 otherwise.
632 tap_flow_create_ipv4(const struct rte_flow_item
*item
, void *data
)
634 struct convert_data
*info
= (struct convert_data
*)data
;
635 const struct rte_flow_item_ipv4
*spec
= item
->spec
;
636 const struct rte_flow_item_ipv4
*mask
= item
->mask
;
637 struct rte_flow
*flow
= info
->flow
;
640 /* use default mask if none provided */
642 mask
= tap_flow_items
[RTE_FLOW_ITEM_TYPE_IPV4
].default_mask
;
643 /* check that previous eth type is compatible with ipv4 */
644 if (info
->eth_type
&& info
->eth_type
!= htons(ETH_P_IP
))
646 /* store ip_proto for consistency if udp/tcp pattern item comes next */
648 info
->ip_proto
= spec
->hdr
.next_proto_id
;
653 info
->eth_type
= htons(ETH_P_IP
);
656 if (mask
->hdr
.dst_addr
) {
657 tap_nlattr_add32(&msg
->nh
, TCA_FLOWER_KEY_IPV4_DST
,
659 tap_nlattr_add32(&msg
->nh
, TCA_FLOWER_KEY_IPV4_DST_MASK
,
662 if (mask
->hdr
.src_addr
) {
663 tap_nlattr_add32(&msg
->nh
, TCA_FLOWER_KEY_IPV4_SRC
,
665 tap_nlattr_add32(&msg
->nh
, TCA_FLOWER_KEY_IPV4_SRC_MASK
,
668 if (spec
->hdr
.next_proto_id
)
669 tap_nlattr_add8(&msg
->nh
, TCA_FLOWER_KEY_IP_PROTO
,
670 spec
->hdr
.next_proto_id
);
675 * Make as much checks as possible on an IPv6 item, and if a flow is provided,
676 * fill it appropriately with IPv6 info.
679 * Item specification.
680 * @param[in, out] data
681 * Additional data structure to tell next layers we've been here.
684 * 0 if checks are alright, -1 otherwise.
687 tap_flow_create_ipv6(const struct rte_flow_item
*item
, void *data
)
689 struct convert_data
*info
= (struct convert_data
*)data
;
690 const struct rte_flow_item_ipv6
*spec
= item
->spec
;
691 const struct rte_flow_item_ipv6
*mask
= item
->mask
;
692 struct rte_flow
*flow
= info
->flow
;
693 uint8_t empty_addr
[16] = { 0 };
696 /* use default mask if none provided */
698 mask
= tap_flow_items
[RTE_FLOW_ITEM_TYPE_IPV6
].default_mask
;
699 /* check that previous eth type is compatible with ipv6 */
700 if (info
->eth_type
&& info
->eth_type
!= htons(ETH_P_IPV6
))
702 /* store ip_proto for consistency if udp/tcp pattern item comes next */
704 info
->ip_proto
= spec
->hdr
.proto
;
709 info
->eth_type
= htons(ETH_P_IPV6
);
712 if (memcmp(mask
->hdr
.dst_addr
, empty_addr
, 16)) {
713 tap_nlattr_add(&msg
->nh
, TCA_FLOWER_KEY_IPV6_DST
,
714 sizeof(spec
->hdr
.dst_addr
), &spec
->hdr
.dst_addr
);
715 tap_nlattr_add(&msg
->nh
, TCA_FLOWER_KEY_IPV6_DST_MASK
,
716 sizeof(mask
->hdr
.dst_addr
), &mask
->hdr
.dst_addr
);
718 if (memcmp(mask
->hdr
.src_addr
, empty_addr
, 16)) {
719 tap_nlattr_add(&msg
->nh
, TCA_FLOWER_KEY_IPV6_SRC
,
720 sizeof(spec
->hdr
.src_addr
), &spec
->hdr
.src_addr
);
721 tap_nlattr_add(&msg
->nh
, TCA_FLOWER_KEY_IPV6_SRC_MASK
,
722 sizeof(mask
->hdr
.src_addr
), &mask
->hdr
.src_addr
);
725 tap_nlattr_add8(&msg
->nh
,
726 TCA_FLOWER_KEY_IP_PROTO
, spec
->hdr
.proto
);
731 * Make as much checks as possible on a UDP item, and if a flow is provided,
732 * fill it appropriately with UDP info.
735 * Item specification.
736 * @param[in, out] data
737 * Additional data structure to tell next layers we've been here.
740 * 0 if checks are alright, -1 otherwise.
743 tap_flow_create_udp(const struct rte_flow_item
*item
, void *data
)
745 struct convert_data
*info
= (struct convert_data
*)data
;
746 const struct rte_flow_item_udp
*spec
= item
->spec
;
747 const struct rte_flow_item_udp
*mask
= item
->mask
;
748 struct rte_flow
*flow
= info
->flow
;
751 /* use default mask if none provided */
753 mask
= tap_flow_items
[RTE_FLOW_ITEM_TYPE_UDP
].default_mask
;
754 /* check that previous ip_proto is compatible with udp */
755 if (info
->ip_proto
&& info
->ip_proto
!= IPPROTO_UDP
)
757 /* TC does not support UDP port masking. Only accept if exact match. */
758 if ((mask
->hdr
.src_port
&& mask
->hdr
.src_port
!= 0xffff) ||
759 (mask
->hdr
.dst_port
&& mask
->hdr
.dst_port
!= 0xffff))
764 tap_nlattr_add8(&msg
->nh
, TCA_FLOWER_KEY_IP_PROTO
, IPPROTO_UDP
);
767 if (mask
->hdr
.dst_port
)
768 tap_nlattr_add16(&msg
->nh
, TCA_FLOWER_KEY_UDP_DST
,
770 if (mask
->hdr
.src_port
)
771 tap_nlattr_add16(&msg
->nh
, TCA_FLOWER_KEY_UDP_SRC
,
777 * Make as much checks as possible on a TCP item, and if a flow is provided,
778 * fill it appropriately with TCP info.
781 * Item specification.
782 * @param[in, out] data
783 * Additional data structure to tell next layers we've been here.
786 * 0 if checks are alright, -1 otherwise.
789 tap_flow_create_tcp(const struct rte_flow_item
*item
, void *data
)
791 struct convert_data
*info
= (struct convert_data
*)data
;
792 const struct rte_flow_item_tcp
*spec
= item
->spec
;
793 const struct rte_flow_item_tcp
*mask
= item
->mask
;
794 struct rte_flow
*flow
= info
->flow
;
797 /* use default mask if none provided */
799 mask
= tap_flow_items
[RTE_FLOW_ITEM_TYPE_TCP
].default_mask
;
800 /* check that previous ip_proto is compatible with tcp */
801 if (info
->ip_proto
&& info
->ip_proto
!= IPPROTO_TCP
)
803 /* TC does not support TCP port masking. Only accept if exact match. */
804 if ((mask
->hdr
.src_port
&& mask
->hdr
.src_port
!= 0xffff) ||
805 (mask
->hdr
.dst_port
&& mask
->hdr
.dst_port
!= 0xffff))
810 tap_nlattr_add8(&msg
->nh
, TCA_FLOWER_KEY_IP_PROTO
, IPPROTO_TCP
);
813 if (mask
->hdr
.dst_port
)
814 tap_nlattr_add16(&msg
->nh
, TCA_FLOWER_KEY_TCP_DST
,
816 if (mask
->hdr
.src_port
)
817 tap_nlattr_add16(&msg
->nh
, TCA_FLOWER_KEY_TCP_SRC
,
823 * Check support for a given item.
826 * Item specification.
828 * Bit-Mask size in bytes.
829 * @param[in] supported_mask
830 * Bit-mask covering supported fields to compare with spec, last and mask in
832 * @param[in] default_mask
833 * Bit-mask default mask if none is provided in \item.
839 tap_flow_item_validate(const struct rte_flow_item
*item
,
841 const uint8_t *supported_mask
,
842 const uint8_t *default_mask
)
846 /* An empty layer is allowed, as long as all fields are NULL */
847 if (!item
->spec
&& (item
->mask
|| item
->last
))
849 /* Is the item spec compatible with what the NIC supports? */
850 if (item
->spec
&& !item
->mask
) {
852 const uint8_t *spec
= item
->spec
;
854 for (i
= 0; i
< size
; ++i
)
855 if ((spec
[i
] | supported_mask
[i
]) != supported_mask
[i
])
857 /* Is the default mask compatible with what the NIC supports? */
858 for (i
= 0; i
< size
; i
++)
859 if ((default_mask
[i
] | supported_mask
[i
]) !=
863 /* Is the item last compatible with what the NIC supports? */
864 if (item
->last
&& !item
->mask
) {
866 const uint8_t *spec
= item
->last
;
868 for (i
= 0; i
< size
; ++i
)
869 if ((spec
[i
] | supported_mask
[i
]) != supported_mask
[i
])
872 /* Is the item mask compatible with what the NIC supports? */
875 const uint8_t *spec
= item
->mask
;
877 for (i
= 0; i
< size
; ++i
)
878 if ((spec
[i
] | supported_mask
[i
]) != supported_mask
[i
])
882 * Once masked, Are item spec and item last equal?
883 * TC does not support range so anything else is invalid.
885 if (item
->spec
&& item
->last
) {
888 const uint8_t *apply
= default_mask
;
893 for (i
= 0; i
< size
; ++i
) {
894 spec
[i
] = ((const uint8_t *)item
->spec
)[i
] & apply
[i
];
895 last
[i
] = ((const uint8_t *)item
->last
)[i
] & apply
[i
];
897 ret
= memcmp(spec
, last
, size
);
903 * Configure the kernel with a TC action and its configured parameters
904 * Handled actions: "gact", "mirred", "skbedit", "bpf"
907 * Pointer to rte flow containing the netlink message
909 * @param[in, out] act_index
910 * Pointer to action sequence number in the TC command
913 * Pointer to struct holding the action parameters
916 * -1 on failure, 0 on success
919 add_action(struct rte_flow
*flow
, size_t *act_index
, struct action_data
*adata
)
921 struct nlmsg
*msg
= &flow
->msg
;
923 if (tap_nlattr_nested_start(msg
, (*act_index
)++) < 0)
926 tap_nlattr_add(&msg
->nh
, TCA_ACT_KIND
,
927 strlen(adata
->id
) + 1, adata
->id
);
928 if (tap_nlattr_nested_start(msg
, TCA_ACT_OPTIONS
) < 0)
930 if (strcmp("gact", adata
->id
) == 0) {
931 tap_nlattr_add(&msg
->nh
, TCA_GACT_PARMS
, sizeof(adata
->gact
),
933 } else if (strcmp("mirred", adata
->id
) == 0) {
934 if (adata
->mirred
.eaction
== TCA_EGRESS_MIRROR
)
935 adata
->mirred
.action
= TC_ACT_PIPE
;
937 adata
->mirred
.action
= TC_ACT_STOLEN
;
938 tap_nlattr_add(&msg
->nh
, TCA_MIRRED_PARMS
,
939 sizeof(adata
->mirred
),
941 } else if (strcmp("skbedit", adata
->id
) == 0) {
942 tap_nlattr_add(&msg
->nh
, TCA_SKBEDIT_PARMS
,
943 sizeof(adata
->skbedit
.skbedit
),
944 &adata
->skbedit
.skbedit
);
945 tap_nlattr_add16(&msg
->nh
, TCA_SKBEDIT_QUEUE_MAPPING
,
946 adata
->skbedit
.queue
);
947 } else if (strcmp("bpf", adata
->id
) == 0) {
948 tap_nlattr_add32(&msg
->nh
, TCA_ACT_BPF_FD
, adata
->bpf
.bpf_fd
);
949 tap_nlattr_add(&msg
->nh
, TCA_ACT_BPF_NAME
,
950 strlen(adata
->bpf
.annotation
) + 1,
951 adata
->bpf
.annotation
);
952 tap_nlattr_add(&msg
->nh
, TCA_ACT_BPF_PARMS
,
953 sizeof(adata
->bpf
.bpf
),
958 tap_nlattr_nested_finish(msg
); /* nested TCA_ACT_OPTIONS */
959 tap_nlattr_nested_finish(msg
); /* nested act_index */
964 * Helper function to send a serie of TC actions to the kernel
967 * Pointer to rte flow containing the netlink message
969 * @param[in] nb_actions
970 * Number of actions in an array of action structs
973 * Pointer to an array of action structs
975 * @param[in] classifier_actions
976 * The classifier on behave of which the actions are configured
979 * -1 on failure, 0 on success
982 add_actions(struct rte_flow
*flow
, int nb_actions
, struct action_data
*data
,
983 int classifier_action
)
985 struct nlmsg
*msg
= &flow
->msg
;
986 size_t act_index
= 1;
989 if (tap_nlattr_nested_start(msg
, classifier_action
) < 0)
991 for (i
= 0; i
< nb_actions
; i
++)
992 if (add_action(flow
, &act_index
, data
+ i
) < 0)
994 tap_nlattr_nested_finish(msg
); /* nested TCA_FLOWER_ACT */
999 * Validate a flow supported by TC.
1000 * If flow param is not NULL, then also fill the netlink message inside.
1003 * Pointer to private structure.
1005 * Flow rule attributes.
1006 * @param[in] pattern
1007 * Pattern specification (list terminated by the END pattern item).
1008 * @param[in] actions
1009 * Associated actions (list terminated by the END action).
1011 * Perform verbose error reporting if not NULL.
1012 * @param[in, out] flow
1013 * Flow structure to update.
1015 * If set to TCA_EGRESS_REDIR, provided actions will be replaced with a
1016 * redirection to the tap netdevice, and the TC rule will be configured
1017 * on the remote netdevice in pmd.
1018 * If set to TCA_EGRESS_MIRROR, provided actions will be replaced with a
1019 * mirroring to the tap netdevice, and the TC rule will be configured
1020 * on the remote netdevice in pmd. Matching packets will thus be duplicated.
1021 * If set to 0, the standard behavior is to be used: set correct actions for
1022 * the TC rule, and apply it on the tap netdevice.
1025 * 0 on success, a negative errno value otherwise and rte_errno is set.
1028 priv_flow_process(struct pmd_internals
*pmd
,
1029 const struct rte_flow_attr
*attr
,
1030 const struct rte_flow_item items
[],
1031 const struct rte_flow_action actions
[],
1032 struct rte_flow_error
*error
,
1033 struct rte_flow
*flow
,
1036 const struct tap_flow_items
*cur_item
= tap_flow_items
;
1037 struct convert_data data
= {
1042 int action
= 0; /* Only one action authorized for now */
1044 if (attr
->transfer
) {
1046 error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER
,
1047 NULL
, "transfer is not supported");
1050 if (attr
->group
> MAX_GROUP
) {
1052 error
, EINVAL
, RTE_FLOW_ERROR_TYPE_ATTR_GROUP
,
1053 NULL
, "group value too big: cannot exceed 15");
1056 if (attr
->priority
> MAX_PRIORITY
) {
1058 error
, EINVAL
, RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY
,
1059 NULL
, "priority value too big");
1062 uint16_t group
= attr
->group
<< GROUP_SHIFT
;
1063 uint16_t prio
= group
| (attr
->priority
+
1064 RSS_PRIORITY_OFFSET
+ PRIORITY_OFFSET
);
1065 flow
->msg
.t
.tcm_info
= TC_H_MAKE(prio
<< 16,
1066 flow
->msg
.t
.tcm_info
);
1071 * If attr->ingress, the rule applies on remote ingress
1072 * to match incoming packets
1073 * If attr->egress, the rule applies on tap ingress (as
1074 * seen from the kernel) to deal with packets going out
1075 * from the DPDK app.
1077 flow
->msg
.t
.tcm_parent
= TC_H_MAKE(TC_H_INGRESS
, 0);
1079 /* Standard rule on tap egress (kernel standpoint). */
1080 flow
->msg
.t
.tcm_parent
=
1081 TC_H_MAKE(MULTIQ_MAJOR_HANDLE
, 0);
1083 /* use flower filter type */
1084 tap_nlattr_add(&flow
->msg
.nh
, TCA_KIND
, sizeof("flower"), "flower");
1085 if (tap_nlattr_nested_start(&flow
->msg
, TCA_OPTIONS
) < 0)
1086 goto exit_item_not_supported
;
1088 for (; items
->type
!= RTE_FLOW_ITEM_TYPE_END
; ++items
) {
1089 const struct tap_flow_items
*token
= NULL
;
1093 if (items
->type
== RTE_FLOW_ITEM_TYPE_VOID
)
1097 cur_item
->items
[i
] != RTE_FLOW_ITEM_TYPE_END
;
1099 if (cur_item
->items
[i
] == items
->type
) {
1100 token
= &tap_flow_items
[items
->type
];
1105 goto exit_item_not_supported
;
1107 err
= tap_flow_item_validate(
1108 items
, cur_item
->mask_sz
,
1109 (const uint8_t *)cur_item
->mask
,
1110 (const uint8_t *)cur_item
->default_mask
);
1112 goto exit_item_not_supported
;
1113 if (flow
&& cur_item
->convert
) {
1114 err
= cur_item
->convert(items
, &data
);
1116 goto exit_item_not_supported
;
1121 tap_nlattr_add16(&flow
->msg
.nh
, TCA_FLOWER_KEY_ETH_TYPE
,
1122 htons(ETH_P_8021Q
));
1123 tap_nlattr_add16(&flow
->msg
.nh
,
1124 TCA_FLOWER_KEY_VLAN_ETH_TYPE
,
1126 data
.eth_type
: htons(ETH_P_ALL
));
1127 } else if (data
.eth_type
) {
1128 tap_nlattr_add16(&flow
->msg
.nh
, TCA_FLOWER_KEY_ETH_TYPE
,
1132 if (mirred
&& flow
) {
1133 struct action_data adata
= {
1141 * If attr->egress && mirred, then this is a special
1142 * case where the rule must be applied on the tap, to
1143 * redirect packets coming from the DPDK App, out
1144 * through the remote netdevice.
1146 adata
.mirred
.ifindex
= attr
->ingress
? pmd
->if_index
:
1147 pmd
->remote_if_index
;
1148 if (mirred
== TCA_EGRESS_MIRROR
)
1149 adata
.mirred
.action
= TC_ACT_PIPE
;
1151 adata
.mirred
.action
= TC_ACT_STOLEN
;
1152 if (add_actions(flow
, 1, &adata
, TCA_FLOWER_ACT
) < 0)
1153 goto exit_action_not_supported
;
1158 for (; actions
->type
!= RTE_FLOW_ACTION_TYPE_END
; ++actions
) {
1161 if (actions
->type
== RTE_FLOW_ACTION_TYPE_VOID
) {
1163 } else if (actions
->type
== RTE_FLOW_ACTION_TYPE_DROP
) {
1165 goto exit_action_not_supported
;
1168 struct action_data adata
= {
1171 .action
= TC_ACT_SHOT
,
1175 err
= add_actions(flow
, 1, &adata
,
1178 } else if (actions
->type
== RTE_FLOW_ACTION_TYPE_PASSTHRU
) {
1180 goto exit_action_not_supported
;
1183 struct action_data adata
= {
1187 .action
= TC_ACT_UNSPEC
,
1191 err
= add_actions(flow
, 1, &adata
,
1194 } else if (actions
->type
== RTE_FLOW_ACTION_TYPE_QUEUE
) {
1195 const struct rte_flow_action_queue
*queue
=
1196 (const struct rte_flow_action_queue
*)
1200 goto exit_action_not_supported
;
1203 (queue
->index
> pmd
->dev
->data
->nb_rx_queues
- 1))
1204 goto exit_action_not_supported
;
1206 struct action_data adata
= {
1210 .action
= TC_ACT_PIPE
,
1212 .queue
= queue
->index
,
1216 err
= add_actions(flow
, 1, &adata
,
1219 } else if (actions
->type
== RTE_FLOW_ACTION_TYPE_RSS
) {
1220 const struct rte_flow_action_rss
*rss
=
1221 (const struct rte_flow_action_rss
*)
1225 goto exit_action_not_supported
;
1227 if (!pmd
->rss_enabled
) {
1228 err
= rss_enable(pmd
, attr
, error
);
1230 goto exit_action_not_supported
;
1233 err
= rss_add_actions(flow
, pmd
, rss
, error
);
1235 goto exit_action_not_supported
;
1238 goto exit_action_not_supported
;
1240 /* When fate is unknown, drop traffic. */
1242 static const struct rte_flow_action drop
[] = {
1243 { .type
= RTE_FLOW_ACTION_TYPE_DROP
, },
1244 { .type
= RTE_FLOW_ACTION_TYPE_END
, },
1252 tap_nlattr_nested_finish(&flow
->msg
); /* nested TCA_OPTIONS */
1254 exit_item_not_supported
:
1255 rte_flow_error_set(error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_ITEM
,
1256 items
, "item not supported");
1258 exit_action_not_supported
:
1259 rte_flow_error_set(error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_ACTION
,
1260 actions
, "action not supported");
1269 * @see rte_flow_validate()
1273 tap_flow_validate(struct rte_eth_dev
*dev
,
1274 const struct rte_flow_attr
*attr
,
1275 const struct rte_flow_item items
[],
1276 const struct rte_flow_action actions
[],
1277 struct rte_flow_error
*error
)
1279 struct pmd_internals
*pmd
= dev
->data
->dev_private
;
1281 return priv_flow_process(pmd
, attr
, items
, actions
, error
, NULL
, 0);
1285 * Set a unique handle in a flow.
1287 * The kernel supports TC rules with equal priority, as long as they use the
1288 * same matching fields (e.g.: dst mac and ipv4) with different values (and
1289 * full mask to ensure no collision is possible).
1290 * In those rules, the handle (uint32_t) is the part that would identify
1291 * specifically each rule.
1293 * On 32-bit architectures, the handle can simply be the flow's pointer address.
1294 * On 64-bit architectures, we rely on jhash(flow) to find a (sufficiently)
1297 * @param[in, out] flow
1298 * The flow that needs its handle set.
1301 tap_flow_set_handle(struct rte_flow
*flow
)
1303 uint32_t handle
= 0;
1305 if (sizeof(flow
) > 4)
1306 handle
= rte_jhash(&flow
, sizeof(flow
), 1);
1308 handle
= (uintptr_t)flow
;
1309 /* must be at least 1 to avoid letting the kernel choose one for us */
1312 flow
->msg
.t
.tcm_handle
= handle
;
1316 * Free the flow opened file descriptors and allocated memory
1319 * Pointer to the flow to free
1323 tap_flow_free(struct pmd_internals
*pmd
, struct rte_flow
*flow
)
1330 if (pmd
->rss_enabled
) {
1331 /* Close flow BPF file descriptors */
1332 for (i
= 0; i
< SEC_MAX
; i
++)
1333 if (flow
->bpf_fd
[i
] != 0) {
1334 close(flow
->bpf_fd
[i
]);
1335 flow
->bpf_fd
[i
] = 0;
1338 /* Release the map key for this RSS rule */
1339 bpf_rss_key(KEY_CMD_RELEASE
, &flow
->key_idx
);
1343 /* Free flow allocated memory */
1350 * @see rte_flow_create()
1353 static struct rte_flow
*
1354 tap_flow_create(struct rte_eth_dev
*dev
,
1355 const struct rte_flow_attr
*attr
,
1356 const struct rte_flow_item items
[],
1357 const struct rte_flow_action actions
[],
1358 struct rte_flow_error
*error
)
1360 struct pmd_internals
*pmd
= dev
->data
->dev_private
;
1361 struct rte_flow
*remote_flow
= NULL
;
1362 struct rte_flow
*flow
= NULL
;
1363 struct nlmsg
*msg
= NULL
;
1366 if (!pmd
->if_index
) {
1367 rte_flow_error_set(error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1369 "can't create rule, ifindex not found");
1373 * No rules configured through standard rte_flow should be set on the
1374 * priorities used by implicit rules.
1376 if ((attr
->group
== MAX_GROUP
) &&
1377 attr
->priority
> (MAX_PRIORITY
- TAP_REMOTE_MAX_IDX
)) {
1379 error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY
,
1380 NULL
, "priority value too big");
1383 flow
= rte_zmalloc(__func__
, sizeof(struct rte_flow
), 0);
1385 rte_flow_error_set(error
, ENOMEM
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1386 NULL
, "cannot allocate memory for rte_flow");
1390 tc_init_msg(msg
, pmd
->if_index
, RTM_NEWTFILTER
,
1391 NLM_F_REQUEST
| NLM_F_ACK
| NLM_F_EXCL
| NLM_F_CREATE
);
1392 msg
->t
.tcm_info
= TC_H_MAKE(0, htons(ETH_P_ALL
));
1393 tap_flow_set_handle(flow
);
1394 if (priv_flow_process(pmd
, attr
, items
, actions
, error
, flow
, 0))
1396 err
= tap_nl_send(pmd
->nlsk_fd
, &msg
->nh
);
1398 rte_flow_error_set(error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1399 NULL
, "couldn't send request to kernel");
1402 err
= tap_nl_recv_ack(pmd
->nlsk_fd
);
1405 "Kernel refused TC filter rule creation (%d): %s",
1406 errno
, strerror(errno
));
1407 rte_flow_error_set(error
, EEXIST
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1409 "overlapping rules or Kernel too old for flower support");
1412 LIST_INSERT_HEAD(&pmd
->flows
, flow
, next
);
1414 * If a remote device is configured, a TC rule with identical items for
1415 * matching must be set on that device, with a single action: redirect
1416 * to the local pmd->if_index.
1418 if (pmd
->remote_if_index
) {
1419 remote_flow
= rte_zmalloc(__func__
, sizeof(struct rte_flow
), 0);
1422 error
, ENOMEM
, RTE_FLOW_ERROR_TYPE_HANDLE
, NULL
,
1423 "cannot allocate memory for rte_flow");
1426 msg
= &remote_flow
->msg
;
1427 /* set the rule if_index for the remote netdevice */
1429 msg
, pmd
->remote_if_index
, RTM_NEWTFILTER
,
1430 NLM_F_REQUEST
| NLM_F_ACK
| NLM_F_EXCL
| NLM_F_CREATE
);
1431 msg
->t
.tcm_info
= TC_H_MAKE(0, htons(ETH_P_ALL
));
1432 tap_flow_set_handle(remote_flow
);
1433 if (priv_flow_process(pmd
, attr
, items
, NULL
,
1434 error
, remote_flow
, TCA_EGRESS_REDIR
)) {
1436 error
, ENOMEM
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1437 NULL
, "rte flow rule validation failed");
1440 err
= tap_nl_send(pmd
->nlsk_fd
, &msg
->nh
);
1443 error
, ENOMEM
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1444 NULL
, "Failure sending nl request");
1447 err
= tap_nl_recv_ack(pmd
->nlsk_fd
);
1450 "Kernel refused TC filter rule creation (%d): %s",
1451 errno
, strerror(errno
));
1453 error
, ENOMEM
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1455 "overlapping rules or Kernel too old for flower support");
1458 flow
->remote_flow
= remote_flow
;
1463 rte_free(remote_flow
);
1465 tap_flow_free(pmd
, flow
);
1470 * Destroy a flow using pointer to pmd_internal.
1472 * @param[in, out] pmd
1473 * Pointer to private structure.
1475 * Pointer to the flow to destroy.
1476 * @param[in, out] error
1477 * Pointer to the flow error handler
1479 * @return 0 if the flow could be destroyed, -1 otherwise.
1482 tap_flow_destroy_pmd(struct pmd_internals
*pmd
,
1483 struct rte_flow
*flow
,
1484 struct rte_flow_error
*error
)
1486 struct rte_flow
*remote_flow
= flow
->remote_flow
;
1489 LIST_REMOVE(flow
, next
);
1490 flow
->msg
.nh
.nlmsg_flags
= NLM_F_REQUEST
| NLM_F_ACK
;
1491 flow
->msg
.nh
.nlmsg_type
= RTM_DELTFILTER
;
1493 ret
= tap_nl_send(pmd
->nlsk_fd
, &flow
->msg
.nh
);
1495 rte_flow_error_set(error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1496 NULL
, "couldn't send request to kernel");
1499 ret
= tap_nl_recv_ack(pmd
->nlsk_fd
);
1500 /* If errno is ENOENT, the rule is already no longer in the kernel. */
1501 if (ret
< 0 && errno
== ENOENT
)
1505 "Kernel refused TC filter rule deletion (%d): %s",
1506 errno
, strerror(errno
));
1508 error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_HANDLE
, NULL
,
1509 "couldn't receive kernel ack to our request");
1514 remote_flow
->msg
.nh
.nlmsg_flags
= NLM_F_REQUEST
| NLM_F_ACK
;
1515 remote_flow
->msg
.nh
.nlmsg_type
= RTM_DELTFILTER
;
1517 ret
= tap_nl_send(pmd
->nlsk_fd
, &remote_flow
->msg
.nh
);
1520 error
, ENOMEM
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1521 NULL
, "Failure sending nl request");
1524 ret
= tap_nl_recv_ack(pmd
->nlsk_fd
);
1525 if (ret
< 0 && errno
== ENOENT
)
1529 "Kernel refused TC filter rule deletion (%d): %s",
1530 errno
, strerror(errno
));
1532 error
, ENOMEM
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1533 NULL
, "Failure trying to receive nl ack");
1539 rte_free(remote_flow
);
1540 tap_flow_free(pmd
, flow
);
1547 * @see rte_flow_destroy()
1551 tap_flow_destroy(struct rte_eth_dev
*dev
,
1552 struct rte_flow
*flow
,
1553 struct rte_flow_error
*error
)
1555 struct pmd_internals
*pmd
= dev
->data
->dev_private
;
1557 return tap_flow_destroy_pmd(pmd
, flow
, error
);
1561 * Enable/disable flow isolation.
1563 * @see rte_flow_isolate()
1567 tap_flow_isolate(struct rte_eth_dev
*dev
,
1569 struct rte_flow_error
*error __rte_unused
)
1571 struct pmd_internals
*pmd
= dev
->data
->dev_private
;
1572 struct pmd_process_private
*process_private
= dev
->process_private
;
1574 /* normalize 'set' variable to contain 0 or 1 values */
1577 /* if already in the right isolation mode - nothing to do */
1578 if ((set
^ pmd
->flow_isolate
) == 0)
1580 /* mark the isolation mode for tap_flow_implicit_create() */
1581 pmd
->flow_isolate
= set
;
1583 * If netdevice is there, setup appropriate flow rules immediately.
1584 * Otherwise it will be set when bringing up the netdevice (tun_alloc).
1586 if (!process_private
->rxq_fds
[0])
1589 struct rte_flow
*remote_flow
;
1592 remote_flow
= LIST_FIRST(&pmd
->implicit_flows
);
1596 * Remove all implicit rules on the remote.
1597 * Keep the local rule to redirect packets on TX.
1598 * Keep also the last implicit local rule: ISOLATE.
1600 if (remote_flow
->msg
.t
.tcm_ifindex
== pmd
->if_index
)
1602 if (tap_flow_destroy_pmd(pmd
, remote_flow
, NULL
) < 0)
1605 /* Switch the TC rule according to pmd->flow_isolate */
1606 if (tap_flow_implicit_create(pmd
, TAP_ISOLATE
) == -1)
1609 /* Switch the TC rule according to pmd->flow_isolate */
1610 if (tap_flow_implicit_create(pmd
, TAP_ISOLATE
) == -1)
1612 if (!pmd
->remote_if_index
)
1614 if (tap_flow_implicit_create(pmd
, TAP_REMOTE_TX
) < 0)
1616 if (tap_flow_implicit_create(pmd
, TAP_REMOTE_LOCAL_MAC
) < 0)
1618 if (tap_flow_implicit_create(pmd
, TAP_REMOTE_BROADCAST
) < 0)
1620 if (tap_flow_implicit_create(pmd
, TAP_REMOTE_BROADCASTV6
) < 0)
1622 if (dev
->data
->promiscuous
&&
1623 tap_flow_implicit_create(pmd
, TAP_REMOTE_PROMISC
) < 0)
1625 if (dev
->data
->all_multicast
&&
1626 tap_flow_implicit_create(pmd
, TAP_REMOTE_ALLMULTI
) < 0)
1631 pmd
->flow_isolate
= 0;
1632 return rte_flow_error_set(
1633 error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_UNSPECIFIED
, NULL
,
1634 "TC rule creation failed");
1638 * Destroy all flows.
1640 * @see rte_flow_flush()
1644 tap_flow_flush(struct rte_eth_dev
*dev
, struct rte_flow_error
*error
)
1646 struct pmd_internals
*pmd
= dev
->data
->dev_private
;
1647 struct rte_flow
*flow
;
1649 while (!LIST_EMPTY(&pmd
->flows
)) {
1650 flow
= LIST_FIRST(&pmd
->flows
);
1651 if (tap_flow_destroy(dev
, flow
, error
) < 0)
1658 * Add an implicit flow rule on the remote device to make sure traffic gets to
1659 * the tap netdevice from there.
1662 * Pointer to private structure.
1664 * The idx in the implicit_rte_flows array specifying which rule to apply.
1666 * @return -1 if the rule couldn't be applied, 0 otherwise.
1668 int tap_flow_implicit_create(struct pmd_internals
*pmd
,
1669 enum implicit_rule_index idx
)
1671 uint16_t flags
= NLM_F_REQUEST
| NLM_F_ACK
| NLM_F_EXCL
| NLM_F_CREATE
;
1672 struct rte_flow_action
*actions
= implicit_rte_flows
[idx
].actions
;
1673 struct rte_flow_action isolate_actions
[2] = {
1675 .type
= RTE_FLOW_ACTION_TYPE_END
,
1678 struct rte_flow_item
*items
= implicit_rte_flows
[idx
].items
;
1679 struct rte_flow_attr
*attr
= &implicit_rte_flows
[idx
].attr
;
1680 struct rte_flow_item_eth eth_local
= { .type
= 0 };
1681 uint16_t if_index
= pmd
->remote_if_index
;
1682 struct rte_flow
*remote_flow
= NULL
;
1683 struct nlmsg
*msg
= NULL
;
1685 struct rte_flow_item items_local
[2] = {
1687 .type
= items
[0].type
,
1689 .mask
= items
[0].mask
,
1692 .type
= items
[1].type
,
1696 remote_flow
= rte_zmalloc(__func__
, sizeof(struct rte_flow
), 0);
1698 TAP_LOG(ERR
, "Cannot allocate memory for rte_flow");
1701 msg
= &remote_flow
->msg
;
1702 if (idx
== TAP_REMOTE_TX
) {
1703 if_index
= pmd
->if_index
;
1704 } else if (idx
== TAP_ISOLATE
) {
1705 if_index
= pmd
->if_index
;
1706 /* Don't be exclusive for this rule, it can be changed later. */
1707 flags
= NLM_F_REQUEST
| NLM_F_ACK
| NLM_F_CREATE
;
1708 isolate_actions
[0].type
= pmd
->flow_isolate
?
1709 RTE_FLOW_ACTION_TYPE_DROP
:
1710 RTE_FLOW_ACTION_TYPE_PASSTHRU
;
1711 actions
= isolate_actions
;
1712 } else if (idx
== TAP_REMOTE_LOCAL_MAC
) {
1714 * eth addr couldn't be set in implicit_rte_flows[] as it is not
1715 * known at compile time.
1717 memcpy(ð_local
.dst
, &pmd
->eth_addr
, sizeof(pmd
->eth_addr
));
1718 items
= items_local
;
1720 tc_init_msg(msg
, if_index
, RTM_NEWTFILTER
, flags
);
1721 msg
->t
.tcm_info
= TC_H_MAKE(0, htons(ETH_P_ALL
));
1723 * The ISOLATE rule is always present and must have a static handle, as
1724 * the action is changed whether the feature is enabled (DROP) or
1725 * disabled (PASSTHRU).
1726 * There is just one REMOTE_PROMISCUOUS rule in all cases. It should
1727 * have a static handle such that adding it twice will fail with EEXIST
1728 * with any kernel version. Remark: old kernels may falsely accept the
1729 * same REMOTE_PROMISCUOUS rules if they had different handles.
1731 if (idx
== TAP_ISOLATE
)
1732 remote_flow
->msg
.t
.tcm_handle
= ISOLATE_HANDLE
;
1733 else if (idx
== TAP_REMOTE_PROMISC
)
1734 remote_flow
->msg
.t
.tcm_handle
= REMOTE_PROMISCUOUS_HANDLE
;
1736 tap_flow_set_handle(remote_flow
);
1737 if (priv_flow_process(pmd
, attr
, items
, actions
, NULL
,
1738 remote_flow
, implicit_rte_flows
[idx
].mirred
)) {
1739 TAP_LOG(ERR
, "rte flow rule validation failed");
1742 err
= tap_nl_send(pmd
->nlsk_fd
, &msg
->nh
);
1744 TAP_LOG(ERR
, "Failure sending nl request");
1747 err
= tap_nl_recv_ack(pmd
->nlsk_fd
);
1749 /* Silently ignore re-entering existing rule */
1750 if (errno
== EEXIST
)
1753 "Kernel refused TC filter rule creation (%d): %s",
1754 errno
, strerror(errno
));
1757 LIST_INSERT_HEAD(&pmd
->implicit_flows
, remote_flow
, next
);
1762 rte_free(remote_flow
);
1767 * Remove specific implicit flow rule on the remote device.
1769 * @param[in, out] pmd
1770 * Pointer to private structure.
1772 * The idx in the implicit_rte_flows array specifying which rule to remove.
1774 * @return -1 if one of the implicit rules couldn't be created, 0 otherwise.
1776 int tap_flow_implicit_destroy(struct pmd_internals
*pmd
,
1777 enum implicit_rule_index idx
)
1779 struct rte_flow
*remote_flow
;
1781 int idx_prio
= implicit_rte_flows
[idx
].attr
.priority
+ PRIORITY_OFFSET
;
1783 for (remote_flow
= LIST_FIRST(&pmd
->implicit_flows
);
1785 remote_flow
= LIST_NEXT(remote_flow
, next
)) {
1786 cur_prio
= (remote_flow
->msg
.t
.tcm_info
>> 16) & PRIORITY_MASK
;
1787 if (cur_prio
!= idx_prio
)
1789 return tap_flow_destroy_pmd(pmd
, remote_flow
, NULL
);
1795 * Destroy all implicit flows.
1797 * @see rte_flow_flush()
1800 tap_flow_implicit_flush(struct pmd_internals
*pmd
, struct rte_flow_error
*error
)
1802 struct rte_flow
*remote_flow
;
1804 while (!LIST_EMPTY(&pmd
->implicit_flows
)) {
1805 remote_flow
= LIST_FIRST(&pmd
->implicit_flows
);
1806 if (tap_flow_destroy_pmd(pmd
, remote_flow
, error
) < 0)
1812 #define MAX_RSS_KEYS 256
1813 #define KEY_IDX_OFFSET (3 * MAX_RSS_KEYS)
1814 #define SEC_NAME_CLS_Q "cls_q"
1816 static const char *sec_name
[SEC_MAX
] = {
1817 [SEC_L3_L4
] = "l3_l4",
1821 * Enable RSS on tap: create TC rules for queuing.
1823 * @param[in, out] pmd
1824 * Pointer to private structure.
1827 * Pointer to rte_flow to get flow group
1830 * Pointer to error reporting if not NULL.
1832 * @return 0 on success, negative value on failure.
1834 static int rss_enable(struct pmd_internals
*pmd
,
1835 const struct rte_flow_attr
*attr
,
1836 struct rte_flow_error
*error
)
1838 struct rte_flow
*rss_flow
= NULL
;
1839 struct nlmsg
*msg
= NULL
;
1840 /* 4096 is the maximum number of instructions for a BPF program */
1841 char annotation
[64];
1845 /* unlimit locked memory */
1846 struct rlimit memlock_limit
= {
1847 .rlim_cur
= RLIM_INFINITY
,
1848 .rlim_max
= RLIM_INFINITY
,
1850 setrlimit(RLIMIT_MEMLOCK
, &memlock_limit
);
1852 /* Get a new map key for a new RSS rule */
1853 err
= bpf_rss_key(KEY_CMD_INIT
, NULL
);
1856 error
, EINVAL
, RTE_FLOW_ERROR_TYPE_HANDLE
, NULL
,
1857 "Failed to initialize BPF RSS keys");
1863 * Create BPF RSS MAP
1865 pmd
->map_fd
= tap_flow_bpf_rss_map_create(sizeof(__u32
), /* key size */
1866 sizeof(struct rss_key
),
1868 if (pmd
->map_fd
< 0) {
1870 "Failed to create BPF map (%d): %s",
1871 errno
, strerror(errno
));
1873 error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_HANDLE
, NULL
,
1874 "Kernel too old or not configured "
1875 "to support BPF maps");
1881 * Add a rule per queue to match reclassified packets and direct them to
1882 * the correct queue.
1884 for (i
= 0; i
< pmd
->dev
->data
->nb_rx_queues
; i
++) {
1885 pmd
->bpf_fd
[i
] = tap_flow_bpf_cls_q(i
);
1886 if (pmd
->bpf_fd
[i
] < 0) {
1888 "Failed to load BPF section %s for queue %d",
1891 error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_HANDLE
,
1893 "Kernel too old or not configured "
1894 "to support BPF programs loading");
1899 rss_flow
= rte_zmalloc(__func__
, sizeof(struct rte_flow
), 0);
1902 "Cannot allocate memory for rte_flow");
1905 msg
= &rss_flow
->msg
;
1906 tc_init_msg(msg
, pmd
->if_index
, RTM_NEWTFILTER
, NLM_F_REQUEST
|
1907 NLM_F_ACK
| NLM_F_EXCL
| NLM_F_CREATE
);
1908 msg
->t
.tcm_info
= TC_H_MAKE(0, htons(ETH_P_ALL
));
1909 tap_flow_set_handle(rss_flow
);
1910 uint16_t group
= attr
->group
<< GROUP_SHIFT
;
1911 uint16_t prio
= group
| (i
+ PRIORITY_OFFSET
);
1912 msg
->t
.tcm_info
= TC_H_MAKE(prio
<< 16, msg
->t
.tcm_info
);
1913 msg
->t
.tcm_parent
= TC_H_MAKE(MULTIQ_MAJOR_HANDLE
, 0);
1915 tap_nlattr_add(&msg
->nh
, TCA_KIND
, sizeof("bpf"), "bpf");
1916 if (tap_nlattr_nested_start(msg
, TCA_OPTIONS
) < 0)
1918 tap_nlattr_add32(&msg
->nh
, TCA_BPF_FD
, pmd
->bpf_fd
[i
]);
1919 snprintf(annotation
, sizeof(annotation
), "[%s%d]",
1921 tap_nlattr_add(&msg
->nh
, TCA_BPF_NAME
, strlen(annotation
) + 1,
1925 struct action_data adata
= {
1929 .action
= TC_ACT_PIPE
,
1934 if (add_actions(rss_flow
, 1, &adata
, TCA_BPF_ACT
) < 0)
1937 tap_nlattr_nested_finish(msg
); /* nested TCA_OPTIONS */
1939 /* Netlink message is now ready to be sent */
1940 if (tap_nl_send(pmd
->nlsk_fd
, &msg
->nh
) < 0)
1942 err
= tap_nl_recv_ack(pmd
->nlsk_fd
);
1945 "Kernel refused TC filter rule creation (%d): %s",
1946 errno
, strerror(errno
));
1949 LIST_INSERT_HEAD(&pmd
->rss_flows
, rss_flow
, next
);
1952 pmd
->rss_enabled
= 1;
1957 * Manage bpf RSS keys repository with operations: init, get, release
1960 * Command on RSS keys: init, get, release
1962 * @param[in, out] key_idx
1963 * Pointer to RSS Key index (out for get command, in for release command)
1965 * @return -1 if couldn't get, release or init the RSS keys, 0 otherwise.
1967 static int bpf_rss_key(enum bpf_rss_key_e cmd
, __u32
*key_idx
)
1971 static __u32 num_used_keys
;
1972 static __u32 rss_keys
[MAX_RSS_KEYS
] = {KEY_STAT_UNSPEC
};
1973 static __u32 rss_keys_initialized
;
1978 if (!rss_keys_initialized
) {
1983 if (num_used_keys
== RTE_DIM(rss_keys
)) {
1988 *key_idx
= num_used_keys
% RTE_DIM(rss_keys
);
1989 while (rss_keys
[*key_idx
] == KEY_STAT_USED
)
1990 *key_idx
= (*key_idx
+ 1) % RTE_DIM(rss_keys
);
1992 rss_keys
[*key_idx
] = KEY_STAT_USED
;
1995 * Add an offset to key_idx in order to handle a case of
1996 * RSS and non RSS flows mixture.
1997 * If a non RSS flow is destroyed it has an eBPF map
1998 * index 0 (initialized on flow creation) and might
1999 * unintentionally remove RSS entry 0 from eBPF map.
2000 * To avoid this issue, add an offset to the real index
2001 * during a KEY_CMD_GET operation and subtract this offset
2002 * during a KEY_CMD_RELEASE operation in order to restore
2005 *key_idx
+= KEY_IDX_OFFSET
;
2009 case KEY_CMD_RELEASE
:
2010 if (!rss_keys_initialized
)
2014 * Subtract offest to restore real key index
2015 * If a non RSS flow is falsely trying to release map
2016 * entry 0 - the offset subtraction will calculate the real
2017 * map index as an out-of-range value and the release operation
2018 * will be silently ignored.
2020 key
= *key_idx
- KEY_IDX_OFFSET
;
2021 if (key
>= RTE_DIM(rss_keys
))
2024 if (rss_keys
[key
] == KEY_STAT_USED
) {
2025 rss_keys
[key
] = KEY_STAT_AVAILABLE
;
2031 for (i
= 0; i
< RTE_DIM(rss_keys
); i
++)
2032 rss_keys
[i
] = KEY_STAT_AVAILABLE
;
2034 rss_keys_initialized
= 1;
2038 case KEY_CMD_DEINIT
:
2039 for (i
= 0; i
< RTE_DIM(rss_keys
); i
++)
2040 rss_keys
[i
] = KEY_STAT_UNSPEC
;
2042 rss_keys_initialized
= 0;
2054 * Add RSS hash calculations and queue selection
2056 * @param[in, out] pmd
2057 * Pointer to internal structure. Used to set/get RSS map fd
2060 * Pointer to RSS flow actions
2063 * Pointer to error reporting if not NULL.
2065 * @return 0 on success, negative value on failure
2067 static int rss_add_actions(struct rte_flow
*flow
, struct pmd_internals
*pmd
,
2068 const struct rte_flow_action_rss
*rss
,
2069 struct rte_flow_error
*error
)
2071 /* 4096 is the maximum number of instructions for a BPF program */
2074 struct rss_key rss_entry
= { .hash_fields
= 0,
2077 /* Check supported RSS features */
2078 if (rss
->func
!= RTE_ETH_HASH_FUNCTION_DEFAULT
)
2079 return rte_flow_error_set
2080 (error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_UNSPECIFIED
, NULL
,
2081 "non-default RSS hash functions are not supported");
2083 return rte_flow_error_set
2084 (error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_UNSPECIFIED
, NULL
,
2085 "a nonzero RSS encapsulation level is not supported");
2087 /* Get a new map key for a new RSS rule */
2088 err
= bpf_rss_key(KEY_CMD_GET
, &flow
->key_idx
);
2091 error
, EINVAL
, RTE_FLOW_ERROR_TYPE_HANDLE
, NULL
,
2092 "Failed to get BPF RSS key");
2097 /* Update RSS map entry with queues */
2098 rss_entry
.nb_queues
= rss
->queue_num
;
2099 for (i
= 0; i
< rss
->queue_num
; i
++)
2100 rss_entry
.queues
[i
] = rss
->queue
[i
];
2101 rss_entry
.hash_fields
=
2102 (1 << HASH_FIELD_IPV4_L3_L4
) | (1 << HASH_FIELD_IPV6_L3_L4
);
2104 /* Add this RSS entry to map */
2105 err
= tap_flow_bpf_update_rss_elem(pmd
->map_fd
,
2106 &flow
->key_idx
, &rss_entry
);
2110 "Failed to update BPF map entry #%u (%d): %s",
2111 flow
->key_idx
, errno
, strerror(errno
));
2113 error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_HANDLE
, NULL
,
2114 "Kernel too old or not configured "
2115 "to support BPF maps updates");
2122 * Load bpf rules to calculate hash for this key_idx
2125 flow
->bpf_fd
[SEC_L3_L4
] =
2126 tap_flow_bpf_calc_l3_l4_hash(flow
->key_idx
, pmd
->map_fd
);
2127 if (flow
->bpf_fd
[SEC_L3_L4
] < 0) {
2129 "Failed to load BPF section %s (%d): %s",
2130 sec_name
[SEC_L3_L4
], errno
, strerror(errno
));
2132 error
, ENOTSUP
, RTE_FLOW_ERROR_TYPE_HANDLE
, NULL
,
2133 "Kernel too old or not configured "
2134 "to support BPF program loading");
2141 struct action_data adata
[] = {
2145 .bpf_fd
= flow
->bpf_fd
[SEC_L3_L4
],
2146 .annotation
= sec_name
[SEC_L3_L4
],
2148 .action
= TC_ACT_PIPE
,
2154 if (add_actions(flow
, RTE_DIM(adata
), adata
,
2155 TCA_FLOWER_ACT
) < 0)
2163 * Manage filter operations.
2166 * Pointer to Ethernet device structure.
2167 * @param filter_type
2170 * Operation to perform.
2172 * Pointer to operation-specific structure.
2175 * 0 on success, negative errno value on failure.
2178 tap_dev_filter_ctrl(struct rte_eth_dev
*dev
,
2179 enum rte_filter_type filter_type
,
2180 enum rte_filter_op filter_op
,
2183 switch (filter_type
) {
2184 case RTE_ETH_FILTER_GENERIC
:
2185 if (filter_op
!= RTE_ETH_FILTER_GET
)
2187 *(const void **)arg
= &tap_flow_ops
;
2190 TAP_LOG(ERR
, "%p: filter type (%d) not supported",