1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* Copyright (c) 2020 Mellanox Technologies. */
4 #include <net/dst_metadata.h>
5 #include <linux/netdevice.h>
6 #include <linux/list.h>
7 #include <linux/rculist.h>
8 #include <linux/rtnetlink.h>
9 #include <linux/workqueue.h>
10 #include <linux/spinlock.h>
15 #include "lib/fs_chains.h"
17 #include "en/mapping.h"
18 #include "en/tc_tun.h"
19 #include "lib/port_tun.h"
21 struct mlx5e_rep_indr_block_priv {
22 struct net_device *netdev;
23 struct mlx5e_rep_priv *rpriv;
25 struct list_head list;
28 int mlx5e_rep_encap_entry_attach(struct mlx5e_priv *priv,
29 struct mlx5e_encap_entry *e)
31 struct mlx5e_rep_priv *rpriv = priv->ppriv;
32 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
33 struct mlx5_tun_entropy *tun_entropy = &uplink_priv->tun_entropy;
34 struct mlx5e_neigh_hash_entry *nhe;
37 err = mlx5_tun_entropy_refcount_inc(tun_entropy, e->reformat_type);
41 mutex_lock(&rpriv->neigh_update.encap_lock);
42 nhe = mlx5e_rep_neigh_entry_lookup(priv, &e->m_neigh);
44 err = mlx5e_rep_neigh_entry_create(priv, e, &nhe);
46 mutex_unlock(&rpriv->neigh_update.encap_lock);
47 mlx5_tun_entropy_refcount_dec(tun_entropy,
54 spin_lock(&nhe->encap_list_lock);
55 list_add_rcu(&e->encap_list, &nhe->encap_list);
56 spin_unlock(&nhe->encap_list_lock);
58 mutex_unlock(&rpriv->neigh_update.encap_lock);
63 void mlx5e_rep_encap_entry_detach(struct mlx5e_priv *priv,
64 struct mlx5e_encap_entry *e)
66 struct mlx5e_rep_priv *rpriv = priv->ppriv;
67 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
68 struct mlx5_tun_entropy *tun_entropy = &uplink_priv->tun_entropy;
73 spin_lock(&e->nhe->encap_list_lock);
74 list_del_rcu(&e->encap_list);
75 spin_unlock(&e->nhe->encap_list_lock);
77 mlx5e_rep_neigh_entry_release(e->nhe);
79 mlx5_tun_entropy_refcount_dec(tun_entropy, e->reformat_type);
82 void mlx5e_rep_update_flows(struct mlx5e_priv *priv,
83 struct mlx5e_encap_entry *e,
85 unsigned char ha[ETH_ALEN])
87 struct ethhdr *eth = (struct ethhdr *)e->encap_header;
88 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
94 /* wait for encap to be fully initialized */
95 wait_for_completion(&e->res_ready);
97 mutex_lock(&esw->offloads.encap_tbl_lock);
98 encap_connected = !!(e->flags & MLX5_ENCAP_ENTRY_VALID);
99 if (e->compl_result < 0 || (encap_connected == neigh_connected &&
100 ether_addr_equal(e->h_dest, ha)))
103 mlx5e_take_all_encap_flows(e, &flow_list);
105 if ((e->flags & MLX5_ENCAP_ENTRY_VALID) &&
106 (!neigh_connected || !ether_addr_equal(e->h_dest, ha)))
107 mlx5e_tc_encap_flows_del(priv, e, &flow_list);
109 if (neigh_connected && !(e->flags & MLX5_ENCAP_ENTRY_VALID)) {
110 struct net_device *route_dev;
112 ether_addr_copy(e->h_dest, ha);
113 ether_addr_copy(eth->h_dest, ha);
114 /* Update the encap source mac, in case that we delete
115 * the flows when encap source mac changed.
117 route_dev = __dev_get_by_index(dev_net(priv->netdev), e->route_dev_ifindex);
119 ether_addr_copy(eth->h_source, route_dev->dev_addr);
121 mlx5e_tc_encap_flows_add(priv, e, &flow_list);
124 mutex_unlock(&esw->offloads.encap_tbl_lock);
125 mlx5e_put_encap_flow_list(priv, &flow_list);
129 mlx5e_rep_setup_tc_cls_flower(struct mlx5e_priv *priv,
130 struct flow_cls_offload *cls_flower, int flags)
132 switch (cls_flower->command) {
133 case FLOW_CLS_REPLACE:
134 return mlx5e_configure_flower(priv->netdev, priv, cls_flower,
136 case FLOW_CLS_DESTROY:
137 return mlx5e_delete_flower(priv->netdev, priv, cls_flower,
140 return mlx5e_stats_flower(priv->netdev, priv, cls_flower,
148 int mlx5e_rep_setup_tc_cls_matchall(struct mlx5e_priv *priv,
149 struct tc_cls_matchall_offload *ma)
151 switch (ma->command) {
152 case TC_CLSMATCHALL_REPLACE:
153 return mlx5e_tc_configure_matchall(priv, ma);
154 case TC_CLSMATCHALL_DESTROY:
155 return mlx5e_tc_delete_matchall(priv, ma);
156 case TC_CLSMATCHALL_STATS:
157 mlx5e_tc_stats_matchall(priv, ma);
164 static int mlx5e_rep_setup_tc_cb(enum tc_setup_type type, void *type_data,
167 unsigned long flags = MLX5_TC_FLAG(INGRESS) | MLX5_TC_FLAG(ESW_OFFLOAD);
168 struct mlx5e_priv *priv = cb_priv;
171 case TC_SETUP_CLSFLOWER:
172 return mlx5e_rep_setup_tc_cls_flower(priv, type_data, flags);
173 case TC_SETUP_CLSMATCHALL:
174 return mlx5e_rep_setup_tc_cls_matchall(priv, type_data);
180 static int mlx5e_rep_setup_ft_cb(enum tc_setup_type type, void *type_data,
183 struct flow_cls_offload tmp, *f = type_data;
184 struct mlx5e_priv *priv = cb_priv;
185 struct mlx5_eswitch *esw;
189 flags = MLX5_TC_FLAG(INGRESS) |
190 MLX5_TC_FLAG(ESW_OFFLOAD) |
191 MLX5_TC_FLAG(FT_OFFLOAD);
192 esw = priv->mdev->priv.eswitch;
195 case TC_SETUP_CLSFLOWER:
196 memcpy(&tmp, f, sizeof(*f));
198 if (!mlx5_chains_prios_supported(esw_chains(esw)))
201 /* Re-use tc offload path by moving the ft flow to the
204 * FT offload can use prio range [0, INT_MAX], so we normalize
205 * it to range [1, mlx5_esw_chains_get_prio_range(esw)]
206 * as with tc, where prio 0 isn't supported.
208 * We only support chain 0 of FT offload.
210 if (tmp.common.prio >= mlx5_chains_get_prio_range(esw_chains(esw)))
212 if (tmp.common.chain_index != 0)
215 tmp.common.chain_index = mlx5_chains_get_nf_ft_chain(esw_chains(esw));
217 err = mlx5e_rep_setup_tc_cls_flower(priv, &tmp, flags);
218 memcpy(&f->stats, &tmp.stats, sizeof(f->stats));
225 static LIST_HEAD(mlx5e_rep_block_tc_cb_list);
226 static LIST_HEAD(mlx5e_rep_block_ft_cb_list);
227 int mlx5e_rep_setup_tc(struct net_device *dev, enum tc_setup_type type,
230 struct mlx5e_priv *priv = netdev_priv(dev);
231 struct flow_block_offload *f = type_data;
233 f->unlocked_driver_cb = true;
237 return flow_block_cb_setup_simple(type_data,
238 &mlx5e_rep_block_tc_cb_list,
239 mlx5e_rep_setup_tc_cb,
242 return flow_block_cb_setup_simple(type_data,
243 &mlx5e_rep_block_ft_cb_list,
244 mlx5e_rep_setup_ft_cb,
251 int mlx5e_rep_tc_init(struct mlx5e_rep_priv *rpriv)
253 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
256 mutex_init(&uplink_priv->unready_flows_lock);
257 INIT_LIST_HEAD(&uplink_priv->unready_flows);
259 /* init shared tc flow table */
260 err = mlx5e_tc_esw_init(&uplink_priv->tc_ht);
264 void mlx5e_rep_tc_cleanup(struct mlx5e_rep_priv *rpriv)
266 /* delete shared tc flow table */
267 mlx5e_tc_esw_cleanup(&rpriv->uplink_priv.tc_ht);
268 mutex_destroy(&rpriv->uplink_priv.unready_flows_lock);
271 void mlx5e_rep_tc_enable(struct mlx5e_priv *priv)
273 struct mlx5e_rep_priv *rpriv = priv->ppriv;
275 INIT_WORK(&rpriv->uplink_priv.reoffload_flows_work,
276 mlx5e_tc_reoffload_flows_work);
279 void mlx5e_rep_tc_disable(struct mlx5e_priv *priv)
281 struct mlx5e_rep_priv *rpriv = priv->ppriv;
283 cancel_work_sync(&rpriv->uplink_priv.reoffload_flows_work);
286 int mlx5e_rep_tc_event_port_affinity(struct mlx5e_priv *priv)
288 struct mlx5e_rep_priv *rpriv = priv->ppriv;
290 queue_work(priv->wq, &rpriv->uplink_priv.reoffload_flows_work);
295 static struct mlx5e_rep_indr_block_priv *
296 mlx5e_rep_indr_block_priv_lookup(struct mlx5e_rep_priv *rpriv,
297 struct net_device *netdev)
299 struct mlx5e_rep_indr_block_priv *cb_priv;
301 /* All callback list access should be protected by RTNL. */
304 list_for_each_entry(cb_priv,
305 &rpriv->uplink_priv.tc_indr_block_priv_list,
307 if (cb_priv->netdev == netdev)
314 mlx5e_rep_indr_offload(struct net_device *netdev,
315 struct flow_cls_offload *flower,
316 struct mlx5e_rep_indr_block_priv *indr_priv,
319 struct mlx5e_priv *priv = netdev_priv(indr_priv->rpriv->netdev);
322 switch (flower->command) {
323 case FLOW_CLS_REPLACE:
324 err = mlx5e_configure_flower(netdev, priv, flower, flags);
326 case FLOW_CLS_DESTROY:
327 err = mlx5e_delete_flower(netdev, priv, flower, flags);
330 err = mlx5e_stats_flower(netdev, priv, flower, flags);
339 static int mlx5e_rep_indr_setup_tc_cb(enum tc_setup_type type,
340 void *type_data, void *indr_priv)
342 unsigned long flags = MLX5_TC_FLAG(EGRESS) | MLX5_TC_FLAG(ESW_OFFLOAD);
343 struct mlx5e_rep_indr_block_priv *priv = indr_priv;
346 case TC_SETUP_CLSFLOWER:
347 return mlx5e_rep_indr_offload(priv->netdev, type_data, priv,
354 static int mlx5e_rep_indr_setup_ft_cb(enum tc_setup_type type,
355 void *type_data, void *indr_priv)
357 struct mlx5e_rep_indr_block_priv *priv = indr_priv;
358 struct flow_cls_offload *f = type_data;
359 struct flow_cls_offload tmp;
360 struct mlx5e_priv *mpriv;
361 struct mlx5_eswitch *esw;
365 mpriv = netdev_priv(priv->rpriv->netdev);
366 esw = mpriv->mdev->priv.eswitch;
368 flags = MLX5_TC_FLAG(EGRESS) |
369 MLX5_TC_FLAG(ESW_OFFLOAD) |
370 MLX5_TC_FLAG(FT_OFFLOAD);
373 case TC_SETUP_CLSFLOWER:
374 memcpy(&tmp, f, sizeof(*f));
376 /* Re-use tc offload path by moving the ft flow to the
379 * FT offload can use prio range [0, INT_MAX], so we normalize
380 * it to range [1, mlx5_esw_chains_get_prio_range(esw)]
381 * as with tc, where prio 0 isn't supported.
383 * We only support chain 0 of FT offload.
385 if (!mlx5_chains_prios_supported(esw_chains(esw)) ||
386 tmp.common.prio >= mlx5_chains_get_prio_range(esw_chains(esw)) ||
387 tmp.common.chain_index)
390 tmp.common.chain_index = mlx5_chains_get_nf_ft_chain(esw_chains(esw));
392 err = mlx5e_rep_indr_offload(priv->netdev, &tmp, priv, flags);
393 memcpy(&f->stats, &tmp.stats, sizeof(f->stats));
400 static void mlx5e_rep_indr_block_unbind(void *cb_priv)
402 struct mlx5e_rep_indr_block_priv *indr_priv = cb_priv;
404 list_del(&indr_priv->list);
408 static LIST_HEAD(mlx5e_block_cb_list);
411 mlx5e_rep_indr_setup_block(struct net_device *netdev, struct Qdisc *sch,
412 struct mlx5e_rep_priv *rpriv,
413 struct flow_block_offload *f,
414 flow_setup_cb_t *setup_cb,
416 void (*cleanup)(struct flow_block_cb *block_cb))
418 struct mlx5e_priv *priv = netdev_priv(rpriv->netdev);
419 struct mlx5e_rep_indr_block_priv *indr_priv;
420 struct flow_block_cb *block_cb;
422 if (!mlx5e_tc_tun_device_to_offload(priv, netdev) &&
423 !(is_vlan_dev(netdev) && vlan_dev_real_dev(netdev) == rpriv->netdev))
426 if (f->binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
429 f->unlocked_driver_cb = true;
430 f->driver_block_list = &mlx5e_block_cb_list;
432 switch (f->command) {
433 case FLOW_BLOCK_BIND:
434 indr_priv = mlx5e_rep_indr_block_priv_lookup(rpriv, netdev);
438 indr_priv = kmalloc(sizeof(*indr_priv), GFP_KERNEL);
442 indr_priv->netdev = netdev;
443 indr_priv->rpriv = rpriv;
444 list_add(&indr_priv->list,
445 &rpriv->uplink_priv.tc_indr_block_priv_list);
447 block_cb = flow_indr_block_cb_alloc(setup_cb, indr_priv, indr_priv,
448 mlx5e_rep_indr_block_unbind,
449 f, netdev, sch, data, rpriv,
451 if (IS_ERR(block_cb)) {
452 list_del(&indr_priv->list);
454 return PTR_ERR(block_cb);
456 flow_block_cb_add(block_cb, f);
457 list_add_tail(&block_cb->driver_list, &mlx5e_block_cb_list);
460 case FLOW_BLOCK_UNBIND:
461 indr_priv = mlx5e_rep_indr_block_priv_lookup(rpriv, netdev);
465 block_cb = flow_block_cb_lookup(f->block, setup_cb, indr_priv);
469 flow_indr_block_cb_remove(block_cb, f);
470 list_del(&block_cb->driver_list);
479 int mlx5e_rep_indr_setup_cb(struct net_device *netdev, struct Qdisc *sch, void *cb_priv,
480 enum tc_setup_type type, void *type_data,
482 void (*cleanup)(struct flow_block_cb *block_cb))
486 return mlx5e_rep_indr_setup_block(netdev, sch, cb_priv, type_data,
487 mlx5e_rep_indr_setup_tc_cb,
490 return mlx5e_rep_indr_setup_block(netdev, sch, cb_priv, type_data,
491 mlx5e_rep_indr_setup_ft_cb,
498 int mlx5e_rep_tc_netdevice_event_register(struct mlx5e_rep_priv *rpriv)
500 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
502 /* init indirect block notifications */
503 INIT_LIST_HEAD(&uplink_priv->tc_indr_block_priv_list);
505 return flow_indr_dev_register(mlx5e_rep_indr_setup_cb, rpriv);
508 void mlx5e_rep_tc_netdevice_event_unregister(struct mlx5e_rep_priv *rpriv)
510 flow_indr_dev_unregister(mlx5e_rep_indr_setup_cb, rpriv,
511 mlx5e_rep_indr_block_unbind);
514 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
515 static bool mlx5e_restore_tunnel(struct mlx5e_priv *priv, struct sk_buff *skb,
516 struct mlx5e_tc_update_priv *tc_priv,
519 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
520 struct tunnel_match_enc_opts enc_opts = {};
521 struct mlx5_rep_uplink_priv *uplink_priv;
522 struct mlx5e_rep_priv *uplink_rpriv;
523 struct metadata_dst *tun_dst;
524 struct tunnel_match_key key;
525 u32 tun_id, enc_opts_id;
526 struct net_device *dev;
529 enc_opts_id = tunnel_id & ENC_OPTS_BITS_MASK;
530 tun_id = tunnel_id >> ENC_OPTS_BITS;
535 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
536 uplink_priv = &uplink_rpriv->uplink_priv;
538 err = mapping_find(uplink_priv->tunnel_mapping, tun_id, &key);
541 netdev_dbg(priv->netdev,
542 "Couldn't find tunnel for tun_id: %d, err: %d\n",
548 err = mapping_find(uplink_priv->tunnel_enc_opts_mapping,
549 enc_opts_id, &enc_opts);
551 netdev_dbg(priv->netdev,
552 "Couldn't find tunnel (opts) for tun_id: %d, err: %d\n",
558 if (key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
559 tun_dst = __ip_tun_set_dst(key.enc_ipv4.src, key.enc_ipv4.dst,
560 key.enc_ip.tos, key.enc_ip.ttl,
561 key.enc_tp.dst, TUNNEL_KEY,
562 key32_to_tunnel_id(key.enc_key_id.keyid),
564 } else if (key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
565 tun_dst = __ipv6_tun_set_dst(&key.enc_ipv6.src, &key.enc_ipv6.dst,
566 key.enc_ip.tos, key.enc_ip.ttl,
567 key.enc_tp.dst, 0, TUNNEL_KEY,
568 key32_to_tunnel_id(key.enc_key_id.keyid),
571 netdev_dbg(priv->netdev,
572 "Couldn't restore tunnel, unsupported addr_type: %d\n",
573 key.enc_control.addr_type);
578 netdev_dbg(priv->netdev, "Couldn't restore tunnel, no tun_dst\n");
582 tun_dst->u.tun_info.key.tp_src = key.enc_tp.src;
584 if (enc_opts.key.len)
585 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
588 enc_opts.key.dst_opt_type);
590 skb_dst_set(skb, (struct dst_entry *)tun_dst);
591 dev = dev_get_by_index(&init_net, key.filter_ifindex);
593 netdev_dbg(priv->netdev,
594 "Couldn't find tunnel device with ifindex: %d\n",
599 /* Set tun_dev so we do dev_put() after datapath */
600 tc_priv->tun_dev = dev;
606 #endif /* CONFIG_NET_TC_SKB_EXT */
608 bool mlx5e_rep_tc_update_skb(struct mlx5_cqe64 *cqe,
610 struct mlx5e_tc_update_priv *tc_priv)
612 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
613 u32 chain = 0, reg_c0, reg_c1, tunnel_id, zone_restore_id;
614 struct mlx5_rep_uplink_priv *uplink_priv;
615 struct mlx5e_rep_priv *uplink_rpriv;
616 struct tc_skb_ext *tc_skb_ext;
617 struct mlx5_eswitch *esw;
618 struct mlx5e_priv *priv;
621 reg_c0 = (be32_to_cpu(cqe->sop_drop_qpn) & MLX5E_TC_FLOW_ID_MASK);
622 if (reg_c0 == MLX5_FS_DEFAULT_FLOW_TAG)
624 reg_c1 = be32_to_cpu(cqe->ft_metadata);
629 priv = netdev_priv(skb->dev);
630 esw = priv->mdev->priv.eswitch;
632 err = mlx5_get_chain_for_tag(esw_chains(esw), reg_c0, &chain);
634 netdev_dbg(priv->netdev,
635 "Couldn't find chain for chain tag: %d, err: %d\n",
641 tc_skb_ext = skb_ext_add(skb, TC_SKB_EXT);
647 tc_skb_ext->chain = chain;
649 zone_restore_id = reg_c1 & ZONE_RESTORE_MAX;
651 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
652 uplink_priv = &uplink_rpriv->uplink_priv;
653 if (!mlx5e_tc_ct_restore_flow(uplink_priv->ct_priv, skb,
658 tunnel_id = reg_c1 >> REG_MAPPING_SHIFT(TUNNEL_TO_REG);
659 return mlx5e_restore_tunnel(priv, skb, tc_priv, tunnel_id);
660 #endif /* CONFIG_NET_TC_SKB_EXT */
665 void mlx5_rep_tc_post_napi_receive(struct mlx5e_tc_update_priv *tc_priv)
667 if (tc_priv->tun_dev)
668 dev_put(tc_priv->tun_dev);