1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */
4 #include <linux/skbuff.h>
5 #include <net/devlink.h>
6 #include <net/pkt_cls.h>
10 #include "conntrack.h"
11 #include "../nfpcore/nfp_cpp.h"
12 #include "../nfpcore/nfp_nsp.h"
13 #include "../nfp_app.h"
14 #include "../nfp_main.h"
15 #include "../nfp_net.h"
16 #include "../nfp_port.h"
18 #define NFP_FLOWER_SUPPORTED_TCPFLAGS \
19 (TCPHDR_FIN | TCPHDR_SYN | TCPHDR_RST | \
20 TCPHDR_PSH | TCPHDR_URG)
22 #define NFP_FLOWER_SUPPORTED_CTLFLAGS \
23 (FLOW_DIS_IS_FRAGMENT | \
26 #define NFP_FLOWER_WHITELIST_DISSECTOR \
27 (BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) | \
28 BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) | \
29 BIT_ULL(FLOW_DISSECTOR_KEY_IPV4_ADDRS) | \
30 BIT_ULL(FLOW_DISSECTOR_KEY_IPV6_ADDRS) | \
31 BIT_ULL(FLOW_DISSECTOR_KEY_TCP) | \
32 BIT_ULL(FLOW_DISSECTOR_KEY_PORTS) | \
33 BIT_ULL(FLOW_DISSECTOR_KEY_ETH_ADDRS) | \
34 BIT_ULL(FLOW_DISSECTOR_KEY_VLAN) | \
35 BIT_ULL(FLOW_DISSECTOR_KEY_CVLAN) | \
36 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_KEYID) | \
37 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) | \
38 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) | \
39 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_CONTROL) | \
40 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_PORTS) | \
41 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_OPTS) | \
42 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IP) | \
43 BIT_ULL(FLOW_DISSECTOR_KEY_MPLS) | \
44 BIT_ULL(FLOW_DISSECTOR_KEY_CT) | \
45 BIT_ULL(FLOW_DISSECTOR_KEY_META) | \
46 BIT_ULL(FLOW_DISSECTOR_KEY_IP))
48 #define NFP_FLOWER_WHITELIST_TUN_DISSECTOR \
49 (BIT_ULL(FLOW_DISSECTOR_KEY_ENC_CONTROL) | \
50 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_KEYID) | \
51 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) | \
52 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) | \
53 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_OPTS) | \
54 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_PORTS) | \
55 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IP))
57 #define NFP_FLOWER_WHITELIST_TUN_DISSECTOR_R \
58 (BIT_ULL(FLOW_DISSECTOR_KEY_ENC_CONTROL) | \
59 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS))
61 #define NFP_FLOWER_WHITELIST_TUN_DISSECTOR_V6_R \
62 (BIT_ULL(FLOW_DISSECTOR_KEY_ENC_CONTROL) | \
63 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS))
65 #define NFP_FLOWER_MERGE_FIELDS \
66 (NFP_FLOWER_LAYER_PORT | \
67 NFP_FLOWER_LAYER_MAC | \
68 NFP_FLOWER_LAYER_TP | \
69 NFP_FLOWER_LAYER_IPV4 | \
70 NFP_FLOWER_LAYER_IPV6)
72 #define NFP_FLOWER_PRE_TUN_RULE_FIELDS \
73 (NFP_FLOWER_LAYER_EXT_META | \
74 NFP_FLOWER_LAYER_PORT | \
75 NFP_FLOWER_LAYER_MAC | \
76 NFP_FLOWER_LAYER_IPV4 | \
77 NFP_FLOWER_LAYER_IPV6)
79 struct nfp_flower_merge_check {
83 struct nfp_flower_mac_mpls l2;
84 struct nfp_flower_tp_ports l4;
86 struct nfp_flower_ipv4 ipv4;
87 struct nfp_flower_ipv6 ipv6;
90 unsigned long vals[8];
95 nfp_flower_xmit_flow(struct nfp_app *app, struct nfp_fl_payload *nfp_flow,
98 u32 meta_len, key_len, mask_len, act_len, tot_len;
102 meta_len = sizeof(struct nfp_fl_rule_metadata);
103 key_len = nfp_flow->meta.key_len;
104 mask_len = nfp_flow->meta.mask_len;
105 act_len = nfp_flow->meta.act_len;
107 tot_len = meta_len + key_len + mask_len + act_len;
109 /* Convert to long words as firmware expects
110 * lengths in units of NFP_FL_LW_SIZ.
112 nfp_flow->meta.key_len >>= NFP_FL_LW_SIZ;
113 nfp_flow->meta.mask_len >>= NFP_FL_LW_SIZ;
114 nfp_flow->meta.act_len >>= NFP_FL_LW_SIZ;
116 skb = nfp_flower_cmsg_alloc(app, tot_len, mtype, GFP_KERNEL);
120 msg = nfp_flower_cmsg_get_data(skb);
121 memcpy(msg, &nfp_flow->meta, meta_len);
122 memcpy(&msg[meta_len], nfp_flow->unmasked_data, key_len);
123 memcpy(&msg[meta_len + key_len], nfp_flow->mask_data, mask_len);
124 memcpy(&msg[meta_len + key_len + mask_len],
125 nfp_flow->action_data, act_len);
127 /* Convert back to bytes as software expects
128 * lengths in units of bytes.
130 nfp_flow->meta.key_len <<= NFP_FL_LW_SIZ;
131 nfp_flow->meta.mask_len <<= NFP_FL_LW_SIZ;
132 nfp_flow->meta.act_len <<= NFP_FL_LW_SIZ;
134 nfp_ctrl_tx(app->ctrl, skb);
139 static bool nfp_flower_check_higher_than_mac(struct flow_rule *rule)
141 return flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV4_ADDRS) ||
142 flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV6_ADDRS) ||
143 flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS) ||
144 flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ICMP);
147 static bool nfp_flower_check_higher_than_l3(struct flow_rule *rule)
149 return flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS) ||
150 flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ICMP);
154 nfp_flower_calc_opt_layer(struct flow_dissector_key_enc_opts *enc_opts,
155 u32 *key_layer_two, int *key_size, bool ipv6,
156 struct netlink_ext_ack *extack)
158 if (enc_opts->len > NFP_FL_MAX_GENEVE_OPT_KEY ||
159 (ipv6 && enc_opts->len > NFP_FL_MAX_GENEVE_OPT_KEY_V6)) {
160 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: geneve options exceed maximum length");
164 if (enc_opts->len > 0) {
165 *key_layer_two |= NFP_FLOWER_LAYER2_GENEVE_OP;
166 *key_size += sizeof(struct nfp_flower_geneve_options);
173 nfp_flower_calc_udp_tun_layer(struct flow_dissector_key_ports *enc_ports,
174 struct flow_dissector_key_enc_opts *enc_op,
175 u32 *key_layer_two, u8 *key_layer, int *key_size,
176 struct nfp_flower_priv *priv,
177 enum nfp_flower_tun_type *tun_type, bool ipv6,
178 struct netlink_ext_ack *extack)
182 switch (enc_ports->dst) {
183 case htons(IANA_VXLAN_UDP_PORT):
184 *tun_type = NFP_FL_TUNNEL_VXLAN;
185 *key_layer |= NFP_FLOWER_LAYER_VXLAN;
188 *key_layer |= NFP_FLOWER_LAYER_EXT_META;
189 *key_size += sizeof(struct nfp_flower_ext_meta);
190 *key_layer_two |= NFP_FLOWER_LAYER2_TUN_IPV6;
191 *key_size += sizeof(struct nfp_flower_ipv6_udp_tun);
193 *key_size += sizeof(struct nfp_flower_ipv4_udp_tun);
197 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: encap options not supported on vxlan tunnels");
201 case htons(GENEVE_UDP_PORT):
202 if (!(priv->flower_ext_feats & NFP_FL_FEATS_GENEVE)) {
203 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: loaded firmware does not support geneve offload");
206 *tun_type = NFP_FL_TUNNEL_GENEVE;
207 *key_layer |= NFP_FLOWER_LAYER_EXT_META;
208 *key_size += sizeof(struct nfp_flower_ext_meta);
209 *key_layer_two |= NFP_FLOWER_LAYER2_GENEVE;
212 *key_layer_two |= NFP_FLOWER_LAYER2_TUN_IPV6;
213 *key_size += sizeof(struct nfp_flower_ipv6_udp_tun);
215 *key_size += sizeof(struct nfp_flower_ipv4_udp_tun);
220 if (!(priv->flower_ext_feats & NFP_FL_FEATS_GENEVE_OPT)) {
221 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: loaded firmware does not support geneve option offload");
224 err = nfp_flower_calc_opt_layer(enc_op, key_layer_two, key_size,
230 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: tunnel type unknown");
238 nfp_flower_calculate_key_layers(struct nfp_app *app,
239 struct net_device *netdev,
240 struct nfp_fl_key_ls *ret_key_ls,
241 struct flow_rule *rule,
242 enum nfp_flower_tun_type *tun_type,
243 struct netlink_ext_ack *extack)
245 struct flow_dissector *dissector = rule->match.dissector;
246 struct flow_match_basic basic = { NULL, NULL};
247 struct nfp_flower_priv *priv = app->priv;
253 if (dissector->used_keys & ~NFP_FLOWER_WHITELIST_DISSECTOR) {
254 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: match not supported");
258 /* If any tun dissector is used then the required set must be used. */
259 if (dissector->used_keys & NFP_FLOWER_WHITELIST_TUN_DISSECTOR &&
260 (dissector->used_keys & NFP_FLOWER_WHITELIST_TUN_DISSECTOR_V6_R)
261 != NFP_FLOWER_WHITELIST_TUN_DISSECTOR_V6_R &&
262 (dissector->used_keys & NFP_FLOWER_WHITELIST_TUN_DISSECTOR_R)
263 != NFP_FLOWER_WHITELIST_TUN_DISSECTOR_R) {
264 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: tunnel match not supported");
269 key_layer = NFP_FLOWER_LAYER_PORT;
270 key_size = sizeof(struct nfp_flower_meta_tci) +
271 sizeof(struct nfp_flower_in_port);
273 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS) ||
274 flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_MPLS)) {
275 key_layer |= NFP_FLOWER_LAYER_MAC;
276 key_size += sizeof(struct nfp_flower_mac_mpls);
279 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
280 struct flow_match_vlan vlan;
282 flow_rule_match_vlan(rule, &vlan);
283 if (!(priv->flower_ext_feats & NFP_FL_FEATS_VLAN_PCP) &&
284 vlan.key->vlan_priority) {
285 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: loaded firmware does not support VLAN PCP offload");
288 if (priv->flower_ext_feats & NFP_FL_FEATS_VLAN_QINQ &&
289 !(key_layer_two & NFP_FLOWER_LAYER2_QINQ)) {
290 key_layer |= NFP_FLOWER_LAYER_EXT_META;
291 key_size += sizeof(struct nfp_flower_ext_meta);
292 key_size += sizeof(struct nfp_flower_vlan);
293 key_layer_two |= NFP_FLOWER_LAYER2_QINQ;
297 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
298 struct flow_match_vlan cvlan;
300 if (!(priv->flower_ext_feats & NFP_FL_FEATS_VLAN_QINQ)) {
301 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: loaded firmware does not support VLAN QinQ offload");
305 flow_rule_match_vlan(rule, &cvlan);
306 if (!(key_layer_two & NFP_FLOWER_LAYER2_QINQ)) {
307 key_layer |= NFP_FLOWER_LAYER_EXT_META;
308 key_size += sizeof(struct nfp_flower_ext_meta);
309 key_size += sizeof(struct nfp_flower_vlan);
310 key_layer_two |= NFP_FLOWER_LAYER2_QINQ;
314 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
315 struct flow_match_enc_opts enc_op = { NULL, NULL };
316 struct flow_match_ipv4_addrs ipv4_addrs;
317 struct flow_match_ipv6_addrs ipv6_addrs;
318 struct flow_match_control enc_ctl;
319 struct flow_match_ports enc_ports;
320 bool ipv6_tun = false;
322 flow_rule_match_enc_control(rule, &enc_ctl);
324 if (enc_ctl.mask->addr_type != 0xffff) {
325 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: wildcarded protocols on tunnels are not supported");
329 ipv6_tun = enc_ctl.key->addr_type ==
330 FLOW_DISSECTOR_KEY_IPV6_ADDRS;
332 !(priv->flower_ext_feats & NFP_FL_FEATS_IPV6_TUN)) {
333 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: firmware does not support IPv6 tunnels");
338 enc_ctl.key->addr_type != FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
339 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: tunnel address type not IPv4 or IPv6");
344 flow_rule_match_enc_ipv6_addrs(rule, &ipv6_addrs);
345 if (memchr_inv(&ipv6_addrs.mask->dst, 0xff,
346 sizeof(ipv6_addrs.mask->dst))) {
347 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: only an exact match IPv6 destination address is supported");
351 flow_rule_match_enc_ipv4_addrs(rule, &ipv4_addrs);
352 if (ipv4_addrs.mask->dst != cpu_to_be32(~0)) {
353 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: only an exact match IPv4 destination address is supported");
358 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_OPTS))
359 flow_rule_match_enc_opts(rule, &enc_op);
361 if (!flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
362 /* Check if GRE, which has no enc_ports */
363 if (!netif_is_gretap(netdev) && !netif_is_ip6gretap(netdev)) {
364 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: an exact match on L4 destination port is required for non-GRE tunnels");
368 *tun_type = NFP_FL_TUNNEL_GRE;
369 key_layer |= NFP_FLOWER_LAYER_EXT_META;
370 key_size += sizeof(struct nfp_flower_ext_meta);
371 key_layer_two |= NFP_FLOWER_LAYER2_GRE;
374 key_layer_two |= NFP_FLOWER_LAYER2_TUN_IPV6;
376 sizeof(struct nfp_flower_ipv6_gre_tun);
379 sizeof(struct nfp_flower_ipv4_gre_tun);
383 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: encap options not supported on GRE tunnels");
387 flow_rule_match_enc_ports(rule, &enc_ports);
388 if (enc_ports.mask->dst != cpu_to_be16(~0)) {
389 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: only an exact match L4 destination port is supported");
393 err = nfp_flower_calc_udp_tun_layer(enc_ports.key,
403 /* Ensure the ingress netdev matches the expected
406 if (!nfp_fl_netdev_is_tunnel_type(netdev, *tun_type)) {
407 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: ingress netdev does not match the expected tunnel type");
413 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC))
414 flow_rule_match_basic(rule, &basic);
416 if (basic.mask && basic.mask->n_proto) {
417 /* Ethernet type is present in the key. */
418 switch (basic.key->n_proto) {
419 case cpu_to_be16(ETH_P_IP):
420 key_layer |= NFP_FLOWER_LAYER_IPV4;
421 key_size += sizeof(struct nfp_flower_ipv4);
424 case cpu_to_be16(ETH_P_IPV6):
425 key_layer |= NFP_FLOWER_LAYER_IPV6;
426 key_size += sizeof(struct nfp_flower_ipv6);
429 /* Currently we do not offload ARP
430 * because we rely on it to get to the host.
432 case cpu_to_be16(ETH_P_ARP):
433 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: ARP not supported");
436 case cpu_to_be16(ETH_P_MPLS_UC):
437 case cpu_to_be16(ETH_P_MPLS_MC):
438 if (!(key_layer & NFP_FLOWER_LAYER_MAC)) {
439 key_layer |= NFP_FLOWER_LAYER_MAC;
440 key_size += sizeof(struct nfp_flower_mac_mpls);
444 /* Will be included in layer 2. */
445 case cpu_to_be16(ETH_P_8021Q):
449 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: match on given EtherType is not supported");
452 } else if (nfp_flower_check_higher_than_mac(rule)) {
453 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: cannot match above L2 without specified EtherType");
457 if (basic.mask && basic.mask->ip_proto) {
458 switch (basic.key->ip_proto) {
464 key_layer |= NFP_FLOWER_LAYER_TP;
465 key_size += sizeof(struct nfp_flower_tp_ports);
470 if (!(key_layer & NFP_FLOWER_LAYER_TP) &&
471 nfp_flower_check_higher_than_l3(rule)) {
472 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: cannot match on L4 information without specified IP protocol type");
476 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_TCP)) {
477 struct flow_match_tcp tcp;
480 flow_rule_match_tcp(rule, &tcp);
481 tcp_flags = be16_to_cpu(tcp.key->flags);
483 if (tcp_flags & ~NFP_FLOWER_SUPPORTED_TCPFLAGS) {
484 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: no match support for selected TCP flags");
488 /* We only support PSH and URG flags when either
489 * FIN, SYN or RST is present as well.
491 if ((tcp_flags & (TCPHDR_PSH | TCPHDR_URG)) &&
492 !(tcp_flags & (TCPHDR_FIN | TCPHDR_SYN | TCPHDR_RST))) {
493 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: PSH and URG is only supported when used with FIN, SYN or RST");
497 /* We need to store TCP flags in the either the IPv4 or IPv6 key
498 * space, thus we need to ensure we include a IPv4/IPv6 key
499 * layer if we have not done so already.
502 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: match on TCP flags requires a match on L3 protocol");
506 if (!(key_layer & NFP_FLOWER_LAYER_IPV4) &&
507 !(key_layer & NFP_FLOWER_LAYER_IPV6)) {
508 switch (basic.key->n_proto) {
509 case cpu_to_be16(ETH_P_IP):
510 key_layer |= NFP_FLOWER_LAYER_IPV4;
511 key_size += sizeof(struct nfp_flower_ipv4);
514 case cpu_to_be16(ETH_P_IPV6):
515 key_layer |= NFP_FLOWER_LAYER_IPV6;
516 key_size += sizeof(struct nfp_flower_ipv6);
520 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: match on TCP flags requires a match on IPv4/IPv6");
526 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
527 struct flow_match_control ctl;
529 flow_rule_match_control(rule, &ctl);
530 if (ctl.key->flags & ~NFP_FLOWER_SUPPORTED_CTLFLAGS) {
531 NL_SET_ERR_MSG_MOD(extack, "unsupported offload: match on unknown control flag");
536 ret_key_ls->key_layer = key_layer;
537 ret_key_ls->key_layer_two = key_layer_two;
538 ret_key_ls->key_size = key_size;
543 struct nfp_fl_payload *
544 nfp_flower_allocate_new(struct nfp_fl_key_ls *key_layer)
546 struct nfp_fl_payload *flow_pay;
548 flow_pay = kmalloc(sizeof(*flow_pay), GFP_KERNEL);
552 flow_pay->meta.key_len = key_layer->key_size;
553 flow_pay->unmasked_data = kmalloc(key_layer->key_size, GFP_KERNEL);
554 if (!flow_pay->unmasked_data)
557 flow_pay->meta.mask_len = key_layer->key_size;
558 flow_pay->mask_data = kmalloc(key_layer->key_size, GFP_KERNEL);
559 if (!flow_pay->mask_data)
560 goto err_free_unmasked;
562 flow_pay->action_data = kmalloc(NFP_FL_MAX_A_SIZ, GFP_KERNEL);
563 if (!flow_pay->action_data)
566 flow_pay->nfp_tun_ipv4_addr = 0;
567 flow_pay->nfp_tun_ipv6 = NULL;
568 flow_pay->meta.flags = 0;
569 INIT_LIST_HEAD(&flow_pay->linked_flows);
570 flow_pay->in_hw = false;
571 flow_pay->pre_tun_rule.dev = NULL;
576 kfree(flow_pay->mask_data);
578 kfree(flow_pay->unmasked_data);
585 nfp_flower_update_merge_with_actions(struct nfp_fl_payload *flow,
586 struct nfp_flower_merge_check *merge,
587 u8 *last_act_id, int *act_out)
589 struct nfp_fl_set_ipv6_tc_hl_fl *ipv6_tc_hl_fl;
590 struct nfp_fl_set_ip4_ttl_tos *ipv4_ttl_tos;
591 struct nfp_fl_set_ip4_addrs *ipv4_add;
592 struct nfp_fl_set_ipv6_addr *ipv6_add;
593 struct nfp_fl_push_vlan *push_vlan;
594 struct nfp_fl_pre_tunnel *pre_tun;
595 struct nfp_fl_set_tport *tport;
596 struct nfp_fl_set_eth *eth;
597 struct nfp_fl_act_head *a;
598 unsigned int act_off = 0;
599 bool ipv6_tun = false;
604 while (act_off < flow->meta.act_len) {
605 a = (struct nfp_fl_act_head *)&flow->action_data[act_off];
609 case NFP_FL_ACTION_OPCODE_OUTPUT:
613 case NFP_FL_ACTION_OPCODE_PUSH_VLAN:
614 push_vlan = (struct nfp_fl_push_vlan *)a;
615 if (push_vlan->vlan_tci)
616 merge->tci = cpu_to_be16(0xffff);
618 case NFP_FL_ACTION_OPCODE_POP_VLAN:
619 merge->tci = cpu_to_be16(0);
621 case NFP_FL_ACTION_OPCODE_SET_TUNNEL:
622 /* New tunnel header means l2 to l4 can be matched. */
623 eth_broadcast_addr(&merge->l2.mac_dst[0]);
624 eth_broadcast_addr(&merge->l2.mac_src[0]);
625 memset(&merge->l4, 0xff,
626 sizeof(struct nfp_flower_tp_ports));
628 memset(&merge->ipv6, 0xff,
629 sizeof(struct nfp_flower_ipv6));
631 memset(&merge->ipv4, 0xff,
632 sizeof(struct nfp_flower_ipv4));
634 case NFP_FL_ACTION_OPCODE_SET_ETHERNET:
635 eth = (struct nfp_fl_set_eth *)a;
636 for (i = 0; i < ETH_ALEN; i++)
637 merge->l2.mac_dst[i] |= eth->eth_addr_mask[i];
638 for (i = 0; i < ETH_ALEN; i++)
639 merge->l2.mac_src[i] |=
640 eth->eth_addr_mask[ETH_ALEN + i];
642 case NFP_FL_ACTION_OPCODE_SET_IPV4_ADDRS:
643 ipv4_add = (struct nfp_fl_set_ip4_addrs *)a;
644 merge->ipv4.ipv4_src |= ipv4_add->ipv4_src_mask;
645 merge->ipv4.ipv4_dst |= ipv4_add->ipv4_dst_mask;
647 case NFP_FL_ACTION_OPCODE_SET_IPV4_TTL_TOS:
648 ipv4_ttl_tos = (struct nfp_fl_set_ip4_ttl_tos *)a;
649 merge->ipv4.ip_ext.ttl |= ipv4_ttl_tos->ipv4_ttl_mask;
650 merge->ipv4.ip_ext.tos |= ipv4_ttl_tos->ipv4_tos_mask;
652 case NFP_FL_ACTION_OPCODE_SET_IPV6_SRC:
653 ipv6_add = (struct nfp_fl_set_ipv6_addr *)a;
654 for (i = 0; i < 4; i++)
655 merge->ipv6.ipv6_src.in6_u.u6_addr32[i] |=
656 ipv6_add->ipv6[i].mask;
658 case NFP_FL_ACTION_OPCODE_SET_IPV6_DST:
659 ipv6_add = (struct nfp_fl_set_ipv6_addr *)a;
660 for (i = 0; i < 4; i++)
661 merge->ipv6.ipv6_dst.in6_u.u6_addr32[i] |=
662 ipv6_add->ipv6[i].mask;
664 case NFP_FL_ACTION_OPCODE_SET_IPV6_TC_HL_FL:
665 ipv6_tc_hl_fl = (struct nfp_fl_set_ipv6_tc_hl_fl *)a;
666 merge->ipv6.ip_ext.ttl |=
667 ipv6_tc_hl_fl->ipv6_hop_limit_mask;
668 merge->ipv6.ip_ext.tos |= ipv6_tc_hl_fl->ipv6_tc_mask;
669 merge->ipv6.ipv6_flow_label_exthdr |=
670 ipv6_tc_hl_fl->ipv6_label_mask;
672 case NFP_FL_ACTION_OPCODE_SET_UDP:
673 case NFP_FL_ACTION_OPCODE_SET_TCP:
674 tport = (struct nfp_fl_set_tport *)a;
675 ports = (u8 *)&merge->l4.port_src;
676 for (i = 0; i < 4; i++)
677 ports[i] |= tport->tp_port_mask[i];
679 case NFP_FL_ACTION_OPCODE_PRE_TUNNEL:
680 pre_tun = (struct nfp_fl_pre_tunnel *)a;
681 ipv6_tun = be16_to_cpu(pre_tun->flags) &
684 case NFP_FL_ACTION_OPCODE_PRE_LAG:
685 case NFP_FL_ACTION_OPCODE_PUSH_GENEVE:
691 act_off += a->len_lw << NFP_FL_LW_SIZ;
695 *last_act_id = act_id;
701 nfp_flower_populate_merge_match(struct nfp_fl_payload *flow,
702 struct nfp_flower_merge_check *merge,
705 struct nfp_flower_meta_tci *meta_tci;
706 u8 *mask = flow->mask_data;
707 u8 key_layer, match_size;
709 memset(merge, 0, sizeof(struct nfp_flower_merge_check));
711 meta_tci = (struct nfp_flower_meta_tci *)mask;
712 key_layer = meta_tci->nfp_flow_key_layer;
714 if (key_layer & ~NFP_FLOWER_MERGE_FIELDS && !extra_fields)
717 merge->tci = meta_tci->tci;
718 mask += sizeof(struct nfp_flower_meta_tci);
720 if (key_layer & NFP_FLOWER_LAYER_EXT_META)
721 mask += sizeof(struct nfp_flower_ext_meta);
723 mask += sizeof(struct nfp_flower_in_port);
725 if (key_layer & NFP_FLOWER_LAYER_MAC) {
726 match_size = sizeof(struct nfp_flower_mac_mpls);
727 memcpy(&merge->l2, mask, match_size);
731 if (key_layer & NFP_FLOWER_LAYER_TP) {
732 match_size = sizeof(struct nfp_flower_tp_ports);
733 memcpy(&merge->l4, mask, match_size);
737 if (key_layer & NFP_FLOWER_LAYER_IPV4) {
738 match_size = sizeof(struct nfp_flower_ipv4);
739 memcpy(&merge->ipv4, mask, match_size);
742 if (key_layer & NFP_FLOWER_LAYER_IPV6) {
743 match_size = sizeof(struct nfp_flower_ipv6);
744 memcpy(&merge->ipv6, mask, match_size);
751 nfp_flower_can_merge(struct nfp_fl_payload *sub_flow1,
752 struct nfp_fl_payload *sub_flow2)
754 /* Two flows can be merged if sub_flow2 only matches on bits that are
755 * either matched by sub_flow1 or set by a sub_flow1 action. This
756 * ensures that every packet that hits sub_flow1 and recirculates is
757 * guaranteed to hit sub_flow2.
759 struct nfp_flower_merge_check sub_flow1_merge, sub_flow2_merge;
760 int err, act_out = 0;
763 err = nfp_flower_populate_merge_match(sub_flow1, &sub_flow1_merge,
768 err = nfp_flower_populate_merge_match(sub_flow2, &sub_flow2_merge,
773 err = nfp_flower_update_merge_with_actions(sub_flow1, &sub_flow1_merge,
774 &last_act_id, &act_out);
778 /* Must only be 1 output action and it must be the last in sequence. */
779 if (act_out != 1 || last_act_id != NFP_FL_ACTION_OPCODE_OUTPUT)
782 /* Reject merge if sub_flow2 matches on something that is not matched
783 * on or set in an action by sub_flow1.
785 err = bitmap_andnot(sub_flow2_merge.vals, sub_flow2_merge.vals,
786 sub_flow1_merge.vals,
787 sizeof(struct nfp_flower_merge_check) * 8);
795 nfp_flower_copy_pre_actions(char *act_dst, char *act_src, int len,
798 unsigned int act_off = 0, act_len;
799 struct nfp_fl_act_head *a;
802 while (act_off < len) {
803 a = (struct nfp_fl_act_head *)&act_src[act_off];
804 act_len = a->len_lw << NFP_FL_LW_SIZ;
808 case NFP_FL_ACTION_OPCODE_PRE_TUNNEL:
812 case NFP_FL_ACTION_OPCODE_PRE_LAG:
813 memcpy(act_dst + act_off, act_src + act_off, act_len);
826 nfp_fl_verify_post_tun_acts(char *acts, int len, struct nfp_fl_push_vlan **vlan)
828 struct nfp_fl_act_head *a;
829 unsigned int act_off = 0;
831 while (act_off < len) {
832 a = (struct nfp_fl_act_head *)&acts[act_off];
834 if (a->jump_id == NFP_FL_ACTION_OPCODE_PUSH_VLAN && !act_off)
835 *vlan = (struct nfp_fl_push_vlan *)a;
836 else if (a->jump_id != NFP_FL_ACTION_OPCODE_OUTPUT)
839 act_off += a->len_lw << NFP_FL_LW_SIZ;
842 /* Ensure any VLAN push also has an egress action. */
843 if (*vlan && act_off <= sizeof(struct nfp_fl_push_vlan))
850 nfp_fl_push_vlan_after_tun(char *acts, int len, struct nfp_fl_push_vlan *vlan)
852 struct nfp_fl_set_tun *tun;
853 struct nfp_fl_act_head *a;
854 unsigned int act_off = 0;
856 while (act_off < len) {
857 a = (struct nfp_fl_act_head *)&acts[act_off];
859 if (a->jump_id == NFP_FL_ACTION_OPCODE_SET_TUNNEL) {
860 tun = (struct nfp_fl_set_tun *)a;
861 tun->outer_vlan_tpid = vlan->vlan_tpid;
862 tun->outer_vlan_tci = vlan->vlan_tci;
867 act_off += a->len_lw << NFP_FL_LW_SIZ;
870 /* Return error if no tunnel action is found. */
875 nfp_flower_merge_action(struct nfp_fl_payload *sub_flow1,
876 struct nfp_fl_payload *sub_flow2,
877 struct nfp_fl_payload *merge_flow)
879 unsigned int sub1_act_len, sub2_act_len, pre_off1, pre_off2;
880 struct nfp_fl_push_vlan *post_tun_push_vlan = NULL;
881 bool tunnel_act = false;
885 /* The last action of sub_flow1 must be output - do not merge this. */
886 sub1_act_len = sub_flow1->meta.act_len - sizeof(struct nfp_fl_output);
887 sub2_act_len = sub_flow2->meta.act_len;
892 if (sub1_act_len + sub2_act_len > NFP_FL_MAX_A_SIZ)
895 /* A shortcut can only be applied if there is a single action. */
897 merge_flow->meta.shortcut = cpu_to_be32(NFP_FL_SC_ACT_NULL);
899 merge_flow->meta.shortcut = sub_flow2->meta.shortcut;
901 merge_flow->meta.act_len = sub1_act_len + sub2_act_len;
902 merge_act = merge_flow->action_data;
904 /* Copy any pre-actions to the start of merge flow action list. */
905 pre_off1 = nfp_flower_copy_pre_actions(merge_act,
906 sub_flow1->action_data,
907 sub1_act_len, &tunnel_act);
908 merge_act += pre_off1;
909 sub1_act_len -= pre_off1;
910 pre_off2 = nfp_flower_copy_pre_actions(merge_act,
911 sub_flow2->action_data,
913 merge_act += pre_off2;
914 sub2_act_len -= pre_off2;
916 /* FW does a tunnel push when egressing, therefore, if sub_flow 1 pushes
917 * a tunnel, there are restrictions on what sub_flow 2 actions lead to a
921 char *post_tun_acts = &sub_flow2->action_data[pre_off2];
923 err = nfp_fl_verify_post_tun_acts(post_tun_acts, sub2_act_len,
924 &post_tun_push_vlan);
928 if (post_tun_push_vlan) {
929 pre_off2 += sizeof(*post_tun_push_vlan);
930 sub2_act_len -= sizeof(*post_tun_push_vlan);
934 /* Copy remaining actions from sub_flows 1 and 2. */
935 memcpy(merge_act, sub_flow1->action_data + pre_off1, sub1_act_len);
937 if (post_tun_push_vlan) {
938 /* Update tunnel action in merge to include VLAN push. */
939 err = nfp_fl_push_vlan_after_tun(merge_act, sub1_act_len,
944 merge_flow->meta.act_len -= sizeof(*post_tun_push_vlan);
947 merge_act += sub1_act_len;
948 memcpy(merge_act, sub_flow2->action_data + pre_off2, sub2_act_len);
953 /* Flow link code should only be accessed under RTNL. */
954 static void nfp_flower_unlink_flow(struct nfp_fl_payload_link *link)
956 list_del(&link->merge_flow.list);
957 list_del(&link->sub_flow.list);
961 static void nfp_flower_unlink_flows(struct nfp_fl_payload *merge_flow,
962 struct nfp_fl_payload *sub_flow)
964 struct nfp_fl_payload_link *link;
966 list_for_each_entry(link, &merge_flow->linked_flows, merge_flow.list)
967 if (link->sub_flow.flow == sub_flow) {
968 nfp_flower_unlink_flow(link);
973 static int nfp_flower_link_flows(struct nfp_fl_payload *merge_flow,
974 struct nfp_fl_payload *sub_flow)
976 struct nfp_fl_payload_link *link;
978 link = kmalloc(sizeof(*link), GFP_KERNEL);
982 link->merge_flow.flow = merge_flow;
983 list_add_tail(&link->merge_flow.list, &merge_flow->linked_flows);
984 link->sub_flow.flow = sub_flow;
985 list_add_tail(&link->sub_flow.list, &sub_flow->linked_flows);
991 * nfp_flower_merge_offloaded_flows() - Merge 2 existing flows to single flow.
992 * @app: Pointer to the APP handle
993 * @sub_flow1: Initial flow matched to produce merge hint
994 * @sub_flow2: Post recirculation flow matched in merge hint
996 * Combines 2 flows (if valid) to a single flow, removing the initial from hw
997 * and offloading the new, merged flow.
999 * Return: negative value on error, 0 in success.
1001 int nfp_flower_merge_offloaded_flows(struct nfp_app *app,
1002 struct nfp_fl_payload *sub_flow1,
1003 struct nfp_fl_payload *sub_flow2)
1005 struct nfp_flower_priv *priv = app->priv;
1006 struct nfp_fl_payload *merge_flow;
1007 struct nfp_fl_key_ls merge_key_ls;
1008 struct nfp_merge_info *merge_info;
1014 if (sub_flow1 == sub_flow2 ||
1015 nfp_flower_is_merge_flow(sub_flow1) ||
1016 nfp_flower_is_merge_flow(sub_flow2))
1019 /* Check if the two flows are already merged */
1020 parent_ctx = (u64)(be32_to_cpu(sub_flow1->meta.host_ctx_id)) << 32;
1021 parent_ctx |= (u64)(be32_to_cpu(sub_flow2->meta.host_ctx_id));
1022 if (rhashtable_lookup_fast(&priv->merge_table,
1023 &parent_ctx, merge_table_params)) {
1024 nfp_flower_cmsg_warn(app, "The two flows are already merged.\n");
1028 err = nfp_flower_can_merge(sub_flow1, sub_flow2);
1032 merge_key_ls.key_size = sub_flow1->meta.key_len;
1034 merge_flow = nfp_flower_allocate_new(&merge_key_ls);
1038 merge_flow->tc_flower_cookie = (unsigned long)merge_flow;
1039 merge_flow->ingress_dev = sub_flow1->ingress_dev;
1041 memcpy(merge_flow->unmasked_data, sub_flow1->unmasked_data,
1042 sub_flow1->meta.key_len);
1043 memcpy(merge_flow->mask_data, sub_flow1->mask_data,
1044 sub_flow1->meta.mask_len);
1046 err = nfp_flower_merge_action(sub_flow1, sub_flow2, merge_flow);
1048 goto err_destroy_merge_flow;
1050 err = nfp_flower_link_flows(merge_flow, sub_flow1);
1052 goto err_destroy_merge_flow;
1054 err = nfp_flower_link_flows(merge_flow, sub_flow2);
1056 goto err_unlink_sub_flow1;
1058 err = nfp_compile_flow_metadata(app, merge_flow->tc_flower_cookie, merge_flow,
1059 merge_flow->ingress_dev, NULL);
1061 goto err_unlink_sub_flow2;
1063 err = rhashtable_insert_fast(&priv->flow_table, &merge_flow->fl_node,
1064 nfp_flower_table_params);
1066 goto err_release_metadata;
1068 merge_info = kmalloc(sizeof(*merge_info), GFP_KERNEL);
1071 goto err_remove_rhash;
1073 merge_info->parent_ctx = parent_ctx;
1074 err = rhashtable_insert_fast(&priv->merge_table, &merge_info->ht_node,
1075 merge_table_params);
1077 goto err_destroy_merge_info;
1079 err = nfp_flower_xmit_flow(app, merge_flow,
1080 NFP_FLOWER_CMSG_TYPE_FLOW_MOD);
1082 goto err_remove_merge_info;
1084 merge_flow->in_hw = true;
1085 sub_flow1->in_hw = false;
1089 err_remove_merge_info:
1090 WARN_ON_ONCE(rhashtable_remove_fast(&priv->merge_table,
1091 &merge_info->ht_node,
1092 merge_table_params));
1093 err_destroy_merge_info:
1096 WARN_ON_ONCE(rhashtable_remove_fast(&priv->flow_table,
1097 &merge_flow->fl_node,
1098 nfp_flower_table_params));
1099 err_release_metadata:
1100 nfp_modify_flow_metadata(app, merge_flow);
1101 err_unlink_sub_flow2:
1102 nfp_flower_unlink_flows(merge_flow, sub_flow2);
1103 err_unlink_sub_flow1:
1104 nfp_flower_unlink_flows(merge_flow, sub_flow1);
1105 err_destroy_merge_flow:
1106 kfree(merge_flow->action_data);
1107 kfree(merge_flow->mask_data);
1108 kfree(merge_flow->unmasked_data);
1114 * nfp_flower_validate_pre_tun_rule()
1115 * @app: Pointer to the APP handle
1116 * @flow: Pointer to NFP flow representation of rule
1117 * @key_ls: Pointer to NFP key layers structure
1118 * @extack: Netlink extended ACK report
1120 * Verifies the flow as a pre-tunnel rule.
1122 * Return: negative value on error, 0 if verified.
1125 nfp_flower_validate_pre_tun_rule(struct nfp_app *app,
1126 struct nfp_fl_payload *flow,
1127 struct nfp_fl_key_ls *key_ls,
1128 struct netlink_ext_ack *extack)
1130 struct nfp_flower_priv *priv = app->priv;
1131 struct nfp_flower_meta_tci *meta_tci;
1132 struct nfp_flower_mac_mpls *mac;
1133 u8 *ext = flow->unmasked_data;
1134 struct nfp_fl_act_head *act;
1135 u8 *mask = flow->mask_data;
1140 meta_tci = (struct nfp_flower_meta_tci *)flow->unmasked_data;
1141 key_layer = key_ls->key_layer;
1142 if (!(priv->flower_ext_feats & NFP_FL_FEATS_VLAN_QINQ)) {
1143 if (meta_tci->tci & cpu_to_be16(NFP_FLOWER_MASK_VLAN_PRESENT)) {
1144 u16 vlan_tci = be16_to_cpu(meta_tci->tci);
1146 vlan_tci &= ~NFP_FLOWER_MASK_VLAN_PRESENT;
1147 flow->pre_tun_rule.vlan_tci = cpu_to_be16(vlan_tci);
1150 flow->pre_tun_rule.vlan_tci = cpu_to_be16(0xffff);
1154 if (key_layer & ~NFP_FLOWER_PRE_TUN_RULE_FIELDS) {
1155 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: too many match fields");
1157 } else if (key_ls->key_layer_two & ~NFP_FLOWER_LAYER2_QINQ) {
1158 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: non-vlan in extended match fields");
1162 if (!(key_layer & NFP_FLOWER_LAYER_MAC)) {
1163 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: MAC fields match required");
1167 if (!(key_layer & NFP_FLOWER_LAYER_IPV4) &&
1168 !(key_layer & NFP_FLOWER_LAYER_IPV6)) {
1169 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: match on ipv4/ipv6 eth_type must be present");
1173 if (key_layer & NFP_FLOWER_LAYER_IPV6)
1174 flow->pre_tun_rule.is_ipv6 = true;
1176 flow->pre_tun_rule.is_ipv6 = false;
1178 /* Skip fields known to exist. */
1179 mask += sizeof(struct nfp_flower_meta_tci);
1180 ext += sizeof(struct nfp_flower_meta_tci);
1181 if (key_ls->key_layer_two) {
1182 mask += sizeof(struct nfp_flower_ext_meta);
1183 ext += sizeof(struct nfp_flower_ext_meta);
1185 mask += sizeof(struct nfp_flower_in_port);
1186 ext += sizeof(struct nfp_flower_in_port);
1188 /* Ensure destination MAC address is fully matched. */
1189 mac = (struct nfp_flower_mac_mpls *)mask;
1190 if (!is_broadcast_ether_addr(&mac->mac_dst[0])) {
1191 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: dest MAC field must not be masked");
1195 /* Ensure source MAC address is fully matched. This is only needed
1196 * for firmware with the DECAP_V2 feature enabled. Don't do this
1197 * for firmware without this feature to keep old behaviour.
1199 if (priv->flower_ext_feats & NFP_FL_FEATS_DECAP_V2) {
1200 mac = (struct nfp_flower_mac_mpls *)mask;
1201 if (!is_broadcast_ether_addr(&mac->mac_src[0])) {
1202 NL_SET_ERR_MSG_MOD(extack,
1203 "unsupported pre-tunnel rule: source MAC field must not be masked");
1208 if (mac->mpls_lse) {
1209 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: MPLS not supported");
1213 /* Ensure destination MAC address matches pre_tun_dev. */
1214 mac = (struct nfp_flower_mac_mpls *)ext;
1215 if (memcmp(&mac->mac_dst[0], flow->pre_tun_rule.dev->dev_addr, 6)) {
1216 NL_SET_ERR_MSG_MOD(extack,
1217 "unsupported pre-tunnel rule: dest MAC must match output dev MAC");
1221 /* Save mac addresses in pre_tun_rule entry for later use */
1222 memcpy(&flow->pre_tun_rule.loc_mac, &mac->mac_dst[0], ETH_ALEN);
1223 memcpy(&flow->pre_tun_rule.rem_mac, &mac->mac_src[0], ETH_ALEN);
1225 mask += sizeof(struct nfp_flower_mac_mpls);
1226 ext += sizeof(struct nfp_flower_mac_mpls);
1227 if (key_layer & NFP_FLOWER_LAYER_IPV4 ||
1228 key_layer & NFP_FLOWER_LAYER_IPV6) {
1229 /* Flags and proto fields have same offset in IPv4 and IPv6. */
1230 int ip_flags = offsetof(struct nfp_flower_ipv4, ip_ext.flags);
1231 int ip_proto = offsetof(struct nfp_flower_ipv4, ip_ext.proto);
1235 size = key_layer & NFP_FLOWER_LAYER_IPV4 ?
1236 sizeof(struct nfp_flower_ipv4) :
1237 sizeof(struct nfp_flower_ipv6);
1240 /* Ensure proto and flags are the only IP layer fields. */
1241 for (i = 0; i < size; i++)
1242 if (mask[i] && i != ip_flags && i != ip_proto) {
1243 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: only flags and proto can be matched in ip header");
1250 if ((priv->flower_ext_feats & NFP_FL_FEATS_VLAN_QINQ)) {
1251 if (key_ls->key_layer_two & NFP_FLOWER_LAYER2_QINQ) {
1252 struct nfp_flower_vlan *vlan_tags;
1256 vlan_tags = (struct nfp_flower_vlan *)ext;
1258 vlan_tci = be16_to_cpu(vlan_tags->outer_tci);
1259 vlan_tpid = be16_to_cpu(vlan_tags->outer_tpid);
1261 vlan_tci &= ~NFP_FLOWER_MASK_VLAN_PRESENT;
1262 flow->pre_tun_rule.vlan_tci = cpu_to_be16(vlan_tci);
1263 flow->pre_tun_rule.vlan_tpid = cpu_to_be16(vlan_tpid);
1266 flow->pre_tun_rule.vlan_tci = cpu_to_be16(0xffff);
1267 flow->pre_tun_rule.vlan_tpid = cpu_to_be16(0xffff);
1271 /* Action must be a single egress or pop_vlan and egress. */
1273 act = (struct nfp_fl_act_head *)&flow->action_data[act_offset];
1275 if (act->jump_id != NFP_FL_ACTION_OPCODE_POP_VLAN) {
1276 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: match on VLAN must have VLAN pop as first action");
1280 act_offset += act->len_lw << NFP_FL_LW_SIZ;
1281 act = (struct nfp_fl_act_head *)&flow->action_data[act_offset];
1284 if (act->jump_id != NFP_FL_ACTION_OPCODE_OUTPUT) {
1285 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: non egress action detected where egress was expected");
1289 act_offset += act->len_lw << NFP_FL_LW_SIZ;
1291 /* Ensure there are no more actions after egress. */
1292 if (act_offset != flow->meta.act_len) {
1293 NL_SET_ERR_MSG_MOD(extack, "unsupported pre-tunnel rule: egress is not the last action");
1300 static bool offload_pre_check(struct flow_cls_offload *flow)
1302 struct flow_rule *rule = flow_cls_offload_flow_rule(flow);
1303 struct flow_dissector *dissector = rule->match.dissector;
1304 struct flow_match_ct ct;
1306 if (dissector->used_keys & BIT_ULL(FLOW_DISSECTOR_KEY_CT)) {
1307 flow_rule_match_ct(rule, &ct);
1308 /* Allow special case where CT match is all 0 */
1309 if (memchr_inv(ct.key, 0, sizeof(*ct.key)))
1313 if (flow->common.chain_index)
1320 * nfp_flower_add_offload() - Adds a new flow to hardware.
1321 * @app: Pointer to the APP handle
1322 * @netdev: netdev structure.
1323 * @flow: TC flower classifier offload structure.
1325 * Adds a new flow to the repeated hash structure and action payload.
1327 * Return: negative value on error, 0 if configured successfully.
1330 nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
1331 struct flow_cls_offload *flow)
1333 struct flow_rule *rule = flow_cls_offload_flow_rule(flow);
1334 enum nfp_flower_tun_type tun_type = NFP_FL_TUNNEL_NONE;
1335 struct nfp_flower_priv *priv = app->priv;
1336 struct netlink_ext_ack *extack = NULL;
1337 struct nfp_fl_payload *flow_pay;
1338 struct nfp_fl_key_ls *key_layer;
1339 struct nfp_port *port = NULL;
1342 extack = flow->common.extack;
1343 if (nfp_netdev_is_nfp_repr(netdev))
1344 port = nfp_port_from_netdev(netdev);
1346 if (is_pre_ct_flow(flow))
1347 return nfp_fl_ct_handle_pre_ct(priv, netdev, flow, extack, NULL);
1349 if (is_post_ct_flow(flow))
1350 return nfp_fl_ct_handle_post_ct(priv, netdev, flow, extack);
1352 if (!offload_pre_check(flow))
1355 key_layer = kmalloc(sizeof(*key_layer), GFP_KERNEL);
1359 err = nfp_flower_calculate_key_layers(app, netdev, key_layer, rule,
1362 goto err_free_key_ls;
1364 flow_pay = nfp_flower_allocate_new(key_layer);
1367 goto err_free_key_ls;
1370 err = nfp_flower_compile_flow_match(app, rule, key_layer, netdev,
1371 flow_pay, tun_type, extack);
1373 goto err_destroy_flow;
1375 err = nfp_flower_compile_action(app, rule, netdev, flow_pay, extack);
1377 goto err_destroy_flow;
1379 if (flow_pay->pre_tun_rule.dev) {
1380 err = nfp_flower_validate_pre_tun_rule(app, flow_pay, key_layer, extack);
1382 goto err_destroy_flow;
1385 err = nfp_compile_flow_metadata(app, flow->cookie, flow_pay, netdev, extack);
1387 goto err_destroy_flow;
1389 flow_pay->tc_flower_cookie = flow->cookie;
1390 err = rhashtable_insert_fast(&priv->flow_table, &flow_pay->fl_node,
1391 nfp_flower_table_params);
1393 NL_SET_ERR_MSG_MOD(extack, "invalid entry: cannot insert flow into tables for offloads");
1394 goto err_release_metadata;
1397 if (flow_pay->pre_tun_rule.dev) {
1398 if (priv->flower_ext_feats & NFP_FL_FEATS_DECAP_V2) {
1399 struct nfp_predt_entry *predt;
1401 predt = kzalloc(sizeof(*predt), GFP_KERNEL);
1404 goto err_remove_rhash;
1406 predt->flow_pay = flow_pay;
1407 INIT_LIST_HEAD(&predt->nn_list);
1408 spin_lock_bh(&priv->predt_lock);
1409 list_add(&predt->list_head, &priv->predt_list);
1410 flow_pay->pre_tun_rule.predt = predt;
1411 nfp_tun_link_and_update_nn_entries(app, predt);
1412 spin_unlock_bh(&priv->predt_lock);
1414 err = nfp_flower_xmit_pre_tun_flow(app, flow_pay);
1417 err = nfp_flower_xmit_flow(app, flow_pay,
1418 NFP_FLOWER_CMSG_TYPE_FLOW_ADD);
1422 goto err_remove_rhash;
1425 port->tc_offload_cnt++;
1427 flow_pay->in_hw = true;
1429 /* Deallocate flow payload when flower rule has been destroyed. */
1435 WARN_ON_ONCE(rhashtable_remove_fast(&priv->flow_table,
1437 nfp_flower_table_params));
1438 err_release_metadata:
1439 nfp_modify_flow_metadata(app, flow_pay);
1441 if (flow_pay->nfp_tun_ipv6)
1442 nfp_tunnel_put_ipv6_off(app, flow_pay->nfp_tun_ipv6);
1443 kfree(flow_pay->action_data);
1444 kfree(flow_pay->mask_data);
1445 kfree(flow_pay->unmasked_data);
1453 nfp_flower_remove_merge_flow(struct nfp_app *app,
1454 struct nfp_fl_payload *del_sub_flow,
1455 struct nfp_fl_payload *merge_flow)
1457 struct nfp_flower_priv *priv = app->priv;
1458 struct nfp_fl_payload_link *link, *temp;
1459 struct nfp_merge_info *merge_info;
1460 struct nfp_fl_payload *origin;
1465 link = list_first_entry(&merge_flow->linked_flows,
1466 struct nfp_fl_payload_link, merge_flow.list);
1467 origin = link->sub_flow.flow;
1469 /* Re-add rule the merge had overwritten if it has not been deleted. */
1470 if (origin != del_sub_flow)
1473 err = nfp_modify_flow_metadata(app, merge_flow);
1475 nfp_flower_cmsg_warn(app, "Metadata fail for merge flow delete.\n");
1476 goto err_free_links;
1480 err = nfp_flower_xmit_flow(app, merge_flow,
1481 NFP_FLOWER_CMSG_TYPE_FLOW_DEL);
1483 nfp_flower_cmsg_warn(app, "Failed to delete merged flow.\n");
1484 goto err_free_links;
1487 __nfp_modify_flow_metadata(priv, origin);
1488 err = nfp_flower_xmit_flow(app, origin,
1489 NFP_FLOWER_CMSG_TYPE_FLOW_MOD);
1491 nfp_flower_cmsg_warn(app, "Failed to revert merge flow.\n");
1492 origin->in_hw = true;
1496 /* Clean any links connected with the merged flow. */
1497 list_for_each_entry_safe(link, temp, &merge_flow->linked_flows,
1499 u32 ctx_id = be32_to_cpu(link->sub_flow.flow->meta.host_ctx_id);
1501 parent_ctx = (parent_ctx << 32) | (u64)(ctx_id);
1502 nfp_flower_unlink_flow(link);
1505 merge_info = rhashtable_lookup_fast(&priv->merge_table,
1507 merge_table_params);
1509 WARN_ON_ONCE(rhashtable_remove_fast(&priv->merge_table,
1510 &merge_info->ht_node,
1511 merge_table_params));
1515 kfree(merge_flow->action_data);
1516 kfree(merge_flow->mask_data);
1517 kfree(merge_flow->unmasked_data);
1518 WARN_ON_ONCE(rhashtable_remove_fast(&priv->flow_table,
1519 &merge_flow->fl_node,
1520 nfp_flower_table_params));
1521 kfree_rcu(merge_flow, rcu);
1525 nfp_flower_del_linked_merge_flows(struct nfp_app *app,
1526 struct nfp_fl_payload *sub_flow)
1528 struct nfp_fl_payload_link *link, *temp;
1530 /* Remove any merge flow formed from the deleted sub_flow. */
1531 list_for_each_entry_safe(link, temp, &sub_flow->linked_flows,
1533 nfp_flower_remove_merge_flow(app, sub_flow,
1534 link->merge_flow.flow);
1538 * nfp_flower_del_offload() - Removes a flow from hardware.
1539 * @app: Pointer to the APP handle
1540 * @netdev: netdev structure.
1541 * @flow: TC flower classifier offload structure
1543 * Removes a flow from the repeated hash structure and clears the
1544 * action payload. Any flows merged from this are also deleted.
1546 * Return: negative value on error, 0 if removed successfully.
1549 nfp_flower_del_offload(struct nfp_app *app, struct net_device *netdev,
1550 struct flow_cls_offload *flow)
1552 struct nfp_flower_priv *priv = app->priv;
1553 struct nfp_fl_ct_map_entry *ct_map_ent;
1554 struct netlink_ext_ack *extack = NULL;
1555 struct nfp_fl_payload *nfp_flow;
1556 struct nfp_port *port = NULL;
1559 extack = flow->common.extack;
1560 if (nfp_netdev_is_nfp_repr(netdev))
1561 port = nfp_port_from_netdev(netdev);
1563 /* Check ct_map_table */
1564 ct_map_ent = rhashtable_lookup_fast(&priv->ct_map_table, &flow->cookie,
1567 err = nfp_fl_ct_del_flow(ct_map_ent);
1571 nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, netdev);
1573 NL_SET_ERR_MSG_MOD(extack, "invalid entry: cannot remove flow that does not exist");
1577 err = nfp_modify_flow_metadata(app, nfp_flow);
1579 goto err_free_merge_flow;
1581 if (nfp_flow->nfp_tun_ipv4_addr)
1582 nfp_tunnel_del_ipv4_off(app, nfp_flow->nfp_tun_ipv4_addr);
1584 if (nfp_flow->nfp_tun_ipv6)
1585 nfp_tunnel_put_ipv6_off(app, nfp_flow->nfp_tun_ipv6);
1587 if (!nfp_flow->in_hw) {
1589 goto err_free_merge_flow;
1592 if (nfp_flow->pre_tun_rule.dev) {
1593 if (priv->flower_ext_feats & NFP_FL_FEATS_DECAP_V2) {
1594 struct nfp_predt_entry *predt;
1596 predt = nfp_flow->pre_tun_rule.predt;
1598 spin_lock_bh(&priv->predt_lock);
1599 nfp_tun_unlink_and_update_nn_entries(app, predt);
1600 list_del(&predt->list_head);
1601 spin_unlock_bh(&priv->predt_lock);
1605 err = nfp_flower_xmit_pre_tun_del_flow(app, nfp_flow);
1608 err = nfp_flower_xmit_flow(app, nfp_flow,
1609 NFP_FLOWER_CMSG_TYPE_FLOW_DEL);
1611 /* Fall through on error. */
1613 err_free_merge_flow:
1614 nfp_flower_del_linked_merge_flows(app, nfp_flow);
1616 port->tc_offload_cnt--;
1617 kfree(nfp_flow->action_data);
1618 kfree(nfp_flow->mask_data);
1619 kfree(nfp_flow->unmasked_data);
1620 WARN_ON_ONCE(rhashtable_remove_fast(&priv->flow_table,
1622 nfp_flower_table_params));
1623 kfree_rcu(nfp_flow, rcu);
1628 __nfp_flower_update_merge_stats(struct nfp_app *app,
1629 struct nfp_fl_payload *merge_flow)
1631 struct nfp_flower_priv *priv = app->priv;
1632 struct nfp_fl_payload_link *link;
1633 struct nfp_fl_payload *sub_flow;
1634 u64 pkts, bytes, used;
1637 ctx_id = be32_to_cpu(merge_flow->meta.host_ctx_id);
1638 pkts = priv->stats[ctx_id].pkts;
1639 /* Do not cycle subflows if no stats to distribute. */
1642 bytes = priv->stats[ctx_id].bytes;
1643 used = priv->stats[ctx_id].used;
1645 /* Reset stats for the merge flow. */
1646 priv->stats[ctx_id].pkts = 0;
1647 priv->stats[ctx_id].bytes = 0;
1649 /* The merge flow has received stats updates from firmware.
1650 * Distribute these stats to all subflows that form the merge.
1651 * The stats will collected from TC via the subflows.
1653 list_for_each_entry(link, &merge_flow->linked_flows, merge_flow.list) {
1654 sub_flow = link->sub_flow.flow;
1655 ctx_id = be32_to_cpu(sub_flow->meta.host_ctx_id);
1656 priv->stats[ctx_id].pkts += pkts;
1657 priv->stats[ctx_id].bytes += bytes;
1658 priv->stats[ctx_id].used = max_t(u64, used,
1659 priv->stats[ctx_id].used);
1664 nfp_flower_update_merge_stats(struct nfp_app *app,
1665 struct nfp_fl_payload *sub_flow)
1667 struct nfp_fl_payload_link *link;
1669 /* Get merge flows that the subflow forms to distribute their stats. */
1670 list_for_each_entry(link, &sub_flow->linked_flows, sub_flow.list)
1671 __nfp_flower_update_merge_stats(app, link->merge_flow.flow);
1675 * nfp_flower_get_stats() - Populates flow stats obtained from hardware.
1676 * @app: Pointer to the APP handle
1677 * @netdev: Netdev structure.
1678 * @flow: TC flower classifier offload structure
1680 * Populates a flow statistics structure which which corresponds to a
1683 * Return: negative value on error, 0 if stats populated successfully.
1686 nfp_flower_get_stats(struct nfp_app *app, struct net_device *netdev,
1687 struct flow_cls_offload *flow)
1689 struct nfp_flower_priv *priv = app->priv;
1690 struct nfp_fl_ct_map_entry *ct_map_ent;
1691 struct netlink_ext_ack *extack = NULL;
1692 struct nfp_fl_payload *nfp_flow;
1695 /* Check ct_map table first */
1696 ct_map_ent = rhashtable_lookup_fast(&priv->ct_map_table, &flow->cookie,
1699 return nfp_fl_ct_stats(flow, ct_map_ent);
1701 extack = flow->common.extack;
1702 nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, netdev);
1704 NL_SET_ERR_MSG_MOD(extack, "invalid entry: cannot dump stats for flow that does not exist");
1708 ctx_id = be32_to_cpu(nfp_flow->meta.host_ctx_id);
1710 spin_lock_bh(&priv->stats_lock);
1711 /* If request is for a sub_flow, update stats from merged flows. */
1712 if (!list_empty(&nfp_flow->linked_flows))
1713 nfp_flower_update_merge_stats(app, nfp_flow);
1715 flow_stats_update(&flow->stats, priv->stats[ctx_id].bytes,
1716 priv->stats[ctx_id].pkts, 0, priv->stats[ctx_id].used,
1717 FLOW_ACTION_HW_STATS_DELAYED);
1719 priv->stats[ctx_id].pkts = 0;
1720 priv->stats[ctx_id].bytes = 0;
1721 spin_unlock_bh(&priv->stats_lock);
1727 nfp_flower_repr_offload(struct nfp_app *app, struct net_device *netdev,
1728 struct flow_cls_offload *flower)
1730 if (!eth_proto_is_802_3(flower->common.protocol))
1733 switch (flower->command) {
1734 case FLOW_CLS_REPLACE:
1735 return nfp_flower_add_offload(app, netdev, flower);
1736 case FLOW_CLS_DESTROY:
1737 return nfp_flower_del_offload(app, netdev, flower);
1738 case FLOW_CLS_STATS:
1739 return nfp_flower_get_stats(app, netdev, flower);
1745 static int nfp_flower_setup_tc_block_cb(enum tc_setup_type type,
1746 void *type_data, void *cb_priv)
1748 struct flow_cls_common_offload *common = type_data;
1749 struct nfp_repr *repr = cb_priv;
1751 if (!tc_can_offload_extack(repr->netdev, common->extack))
1755 case TC_SETUP_CLSFLOWER:
1756 return nfp_flower_repr_offload(repr->app, repr->netdev,
1758 case TC_SETUP_CLSMATCHALL:
1759 return nfp_flower_setup_qos_offload(repr->app, repr->netdev,
1766 static LIST_HEAD(nfp_block_cb_list);
1768 static int nfp_flower_setup_tc_block(struct net_device *netdev,
1769 struct flow_block_offload *f)
1771 struct nfp_repr *repr = netdev_priv(netdev);
1772 struct nfp_flower_repr_priv *repr_priv;
1773 struct flow_block_cb *block_cb;
1775 if (f->binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
1778 repr_priv = repr->app_priv;
1779 repr_priv->block_shared = f->block_shared;
1780 f->driver_block_list = &nfp_block_cb_list;
1782 switch (f->command) {
1783 case FLOW_BLOCK_BIND:
1784 if (flow_block_cb_is_busy(nfp_flower_setup_tc_block_cb, repr,
1785 &nfp_block_cb_list))
1788 block_cb = flow_block_cb_alloc(nfp_flower_setup_tc_block_cb,
1790 if (IS_ERR(block_cb))
1791 return PTR_ERR(block_cb);
1793 flow_block_cb_add(block_cb, f);
1794 list_add_tail(&block_cb->driver_list, &nfp_block_cb_list);
1796 case FLOW_BLOCK_UNBIND:
1797 block_cb = flow_block_cb_lookup(f->block,
1798 nfp_flower_setup_tc_block_cb,
1803 flow_block_cb_remove(block_cb, f);
1804 list_del(&block_cb->driver_list);
1811 int nfp_flower_setup_tc(struct nfp_app *app, struct net_device *netdev,
1812 enum tc_setup_type type, void *type_data)
1815 case TC_SETUP_BLOCK:
1816 return nfp_flower_setup_tc_block(netdev, type_data);
1822 struct nfp_flower_indr_block_cb_priv {
1823 struct net_device *netdev;
1824 struct nfp_app *app;
1825 struct list_head list;
1828 static struct nfp_flower_indr_block_cb_priv *
1829 nfp_flower_indr_block_cb_priv_lookup(struct nfp_app *app,
1830 struct net_device *netdev)
1832 struct nfp_flower_indr_block_cb_priv *cb_priv;
1833 struct nfp_flower_priv *priv = app->priv;
1835 list_for_each_entry(cb_priv, &priv->indr_block_cb_priv, list)
1836 if (cb_priv->netdev == netdev)
1842 static int nfp_flower_setup_indr_block_cb(enum tc_setup_type type,
1843 void *type_data, void *cb_priv)
1845 struct nfp_flower_indr_block_cb_priv *priv = cb_priv;
1848 case TC_SETUP_CLSFLOWER:
1849 return nfp_flower_repr_offload(priv->app, priv->netdev,
1856 void nfp_flower_setup_indr_tc_release(void *cb_priv)
1858 struct nfp_flower_indr_block_cb_priv *priv = cb_priv;
1860 list_del(&priv->list);
1865 nfp_flower_setup_indr_tc_block(struct net_device *netdev, struct Qdisc *sch, struct nfp_app *app,
1866 struct flow_block_offload *f, void *data,
1867 void (*cleanup)(struct flow_block_cb *block_cb))
1869 struct nfp_flower_indr_block_cb_priv *cb_priv;
1870 struct nfp_flower_priv *priv = app->priv;
1871 struct flow_block_cb *block_cb;
1873 if ((f->binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1874 !nfp_flower_internal_port_can_offload(app, netdev)) ||
1875 (f->binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS &&
1876 nfp_flower_internal_port_can_offload(app, netdev)))
1879 switch (f->command) {
1880 case FLOW_BLOCK_BIND:
1881 cb_priv = nfp_flower_indr_block_cb_priv_lookup(app, netdev);
1883 flow_block_cb_is_busy(nfp_flower_setup_indr_block_cb,
1885 &nfp_block_cb_list))
1888 cb_priv = kmalloc(sizeof(*cb_priv), GFP_KERNEL);
1892 cb_priv->netdev = netdev;
1894 list_add(&cb_priv->list, &priv->indr_block_cb_priv);
1896 block_cb = flow_indr_block_cb_alloc(nfp_flower_setup_indr_block_cb,
1898 nfp_flower_setup_indr_tc_release,
1899 f, netdev, sch, data, app, cleanup);
1900 if (IS_ERR(block_cb)) {
1901 list_del(&cb_priv->list);
1903 return PTR_ERR(block_cb);
1906 flow_block_cb_add(block_cb, f);
1907 list_add_tail(&block_cb->driver_list, &nfp_block_cb_list);
1909 case FLOW_BLOCK_UNBIND:
1910 cb_priv = nfp_flower_indr_block_cb_priv_lookup(app, netdev);
1914 block_cb = flow_block_cb_lookup(f->block,
1915 nfp_flower_setup_indr_block_cb,
1920 flow_indr_block_cb_remove(block_cb, f);
1921 list_del(&block_cb->driver_list);
1930 nfp_setup_tc_no_dev(struct nfp_app *app, enum tc_setup_type type, void *data)
1937 return nfp_setup_tc_act_offload(app, data);
1944 nfp_flower_indr_setup_tc_cb(struct net_device *netdev, struct Qdisc *sch, void *cb_priv,
1945 enum tc_setup_type type, void *type_data,
1947 void (*cleanup)(struct flow_block_cb *block_cb))
1950 return nfp_setup_tc_no_dev(cb_priv, type, data);
1952 if (!nfp_fl_is_netdev_to_offload(netdev))
1956 case TC_SETUP_BLOCK:
1957 return nfp_flower_setup_indr_tc_block(netdev, sch, cb_priv,
1958 type_data, data, cleanup);