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"
20 #include "en/tc/sample.h"
22 struct mlx5e_rep_indr_block_priv {
23 struct net_device *netdev;
24 struct mlx5e_rep_priv *rpriv;
26 struct list_head list;
29 int mlx5e_rep_encap_entry_attach(struct mlx5e_priv *priv,
30 struct mlx5e_encap_entry *e,
31 struct mlx5e_neigh *m_neigh,
32 struct net_device *neigh_dev)
34 struct mlx5e_rep_priv *rpriv = priv->ppriv;
35 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
36 struct mlx5_tun_entropy *tun_entropy = &uplink_priv->tun_entropy;
37 struct mlx5e_neigh_hash_entry *nhe;
40 err = mlx5_tun_entropy_refcount_inc(tun_entropy, e->reformat_type);
44 mutex_lock(&rpriv->neigh_update.encap_lock);
45 nhe = mlx5e_rep_neigh_entry_lookup(priv, m_neigh);
47 err = mlx5e_rep_neigh_entry_create(priv, m_neigh, neigh_dev, &nhe);
49 mutex_unlock(&rpriv->neigh_update.encap_lock);
50 mlx5_tun_entropy_refcount_dec(tun_entropy,
57 spin_lock(&nhe->encap_list_lock);
58 list_add_rcu(&e->encap_list, &nhe->encap_list);
59 spin_unlock(&nhe->encap_list_lock);
61 mutex_unlock(&rpriv->neigh_update.encap_lock);
66 void mlx5e_rep_encap_entry_detach(struct mlx5e_priv *priv,
67 struct mlx5e_encap_entry *e)
69 struct mlx5e_rep_priv *rpriv = priv->ppriv;
70 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
71 struct mlx5_tun_entropy *tun_entropy = &uplink_priv->tun_entropy;
76 spin_lock(&e->nhe->encap_list_lock);
77 list_del_rcu(&e->encap_list);
78 spin_unlock(&e->nhe->encap_list_lock);
80 mlx5e_rep_neigh_entry_release(e->nhe);
82 mlx5_tun_entropy_refcount_dec(tun_entropy, e->reformat_type);
85 void mlx5e_rep_update_flows(struct mlx5e_priv *priv,
86 struct mlx5e_encap_entry *e,
88 unsigned char ha[ETH_ALEN])
90 struct ethhdr *eth = (struct ethhdr *)e->encap_header;
91 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
97 mutex_lock(&esw->offloads.encap_tbl_lock);
98 encap_connected = !!(e->flags & MLX5_ENCAP_ENTRY_VALID);
99 if (encap_connected == neigh_connected && ether_addr_equal(e->h_dest, ha))
102 mlx5e_take_all_encap_flows(e, &flow_list);
104 if ((e->flags & MLX5_ENCAP_ENTRY_VALID) &&
105 (!neigh_connected || !ether_addr_equal(e->h_dest, ha)))
106 mlx5e_tc_encap_flows_del(priv, e, &flow_list);
108 if (neigh_connected && !(e->flags & MLX5_ENCAP_ENTRY_VALID)) {
109 struct net_device *route_dev;
111 ether_addr_copy(e->h_dest, ha);
112 ether_addr_copy(eth->h_dest, ha);
113 /* Update the encap source mac, in case that we delete
114 * the flows when encap source mac changed.
116 route_dev = __dev_get_by_index(dev_net(priv->netdev), e->route_dev_ifindex);
118 ether_addr_copy(eth->h_source, route_dev->dev_addr);
120 mlx5e_tc_encap_flows_add(priv, e, &flow_list);
123 mutex_unlock(&esw->offloads.encap_tbl_lock);
124 mlx5e_put_flow_list(priv, &flow_list);
128 mlx5e_rep_setup_tc_cls_flower(struct mlx5e_priv *priv,
129 struct flow_cls_offload *cls_flower, int flags)
131 switch (cls_flower->command) {
132 case FLOW_CLS_REPLACE:
133 return mlx5e_configure_flower(priv->netdev, priv, cls_flower,
135 case FLOW_CLS_DESTROY:
136 return mlx5e_delete_flower(priv->netdev, priv, cls_flower,
139 return mlx5e_stats_flower(priv->netdev, priv, cls_flower,
147 int mlx5e_rep_setup_tc_cls_matchall(struct mlx5e_priv *priv,
148 struct tc_cls_matchall_offload *ma)
150 switch (ma->command) {
151 case TC_CLSMATCHALL_REPLACE:
152 return mlx5e_tc_configure_matchall(priv, ma);
153 case TC_CLSMATCHALL_DESTROY:
154 return mlx5e_tc_delete_matchall(priv, ma);
155 case TC_CLSMATCHALL_STATS:
156 mlx5e_tc_stats_matchall(priv, ma);
163 static int mlx5e_rep_setup_tc_cb(enum tc_setup_type type, void *type_data,
166 unsigned long flags = MLX5_TC_FLAG(INGRESS) | MLX5_TC_FLAG(ESW_OFFLOAD);
167 struct mlx5e_priv *priv = cb_priv;
169 if (!priv->netdev || !netif_device_present(priv->netdev))
173 case TC_SETUP_CLSFLOWER:
174 return mlx5e_rep_setup_tc_cls_flower(priv, type_data, flags);
175 case TC_SETUP_CLSMATCHALL:
176 return mlx5e_rep_setup_tc_cls_matchall(priv, type_data);
182 static int mlx5e_rep_setup_ft_cb(enum tc_setup_type type, void *type_data,
185 struct flow_cls_offload tmp, *f = type_data;
186 struct mlx5e_priv *priv = cb_priv;
187 struct mlx5_eswitch *esw;
191 flags = MLX5_TC_FLAG(INGRESS) |
192 MLX5_TC_FLAG(ESW_OFFLOAD) |
193 MLX5_TC_FLAG(FT_OFFLOAD);
194 esw = priv->mdev->priv.eswitch;
197 case TC_SETUP_CLSFLOWER:
198 memcpy(&tmp, f, sizeof(*f));
200 if (!mlx5_chains_prios_supported(esw_chains(esw)))
203 /* Re-use tc offload path by moving the ft flow to the
206 * FT offload can use prio range [0, INT_MAX], so we normalize
207 * it to range [1, mlx5_esw_chains_get_prio_range(esw)]
208 * as with tc, where prio 0 isn't supported.
210 * We only support chain 0 of FT offload.
212 if (tmp.common.prio >= mlx5_chains_get_prio_range(esw_chains(esw)))
214 if (tmp.common.chain_index != 0)
217 tmp.common.chain_index = mlx5_chains_get_nf_ft_chain(esw_chains(esw));
219 err = mlx5e_rep_setup_tc_cls_flower(priv, &tmp, flags);
220 memcpy(&f->stats, &tmp.stats, sizeof(f->stats));
227 static LIST_HEAD(mlx5e_rep_block_tc_cb_list);
228 static LIST_HEAD(mlx5e_rep_block_ft_cb_list);
229 int mlx5e_rep_setup_tc(struct net_device *dev, enum tc_setup_type type,
232 struct mlx5e_priv *priv = netdev_priv(dev);
233 struct flow_block_offload *f = type_data;
235 f->unlocked_driver_cb = true;
239 return flow_block_cb_setup_simple(type_data,
240 &mlx5e_rep_block_tc_cb_list,
241 mlx5e_rep_setup_tc_cb,
244 return flow_block_cb_setup_simple(type_data,
245 &mlx5e_rep_block_ft_cb_list,
246 mlx5e_rep_setup_ft_cb,
253 int mlx5e_rep_tc_init(struct mlx5e_rep_priv *rpriv)
255 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
258 mutex_init(&uplink_priv->unready_flows_lock);
259 INIT_LIST_HEAD(&uplink_priv->unready_flows);
261 /* init shared tc flow table */
262 err = mlx5e_tc_esw_init(&uplink_priv->tc_ht);
266 void mlx5e_rep_tc_cleanup(struct mlx5e_rep_priv *rpriv)
268 /* delete shared tc flow table */
269 mlx5e_tc_esw_cleanup(&rpriv->uplink_priv.tc_ht);
270 mutex_destroy(&rpriv->uplink_priv.unready_flows_lock);
273 void mlx5e_rep_tc_enable(struct mlx5e_priv *priv)
275 struct mlx5e_rep_priv *rpriv = priv->ppriv;
277 INIT_WORK(&rpriv->uplink_priv.reoffload_flows_work,
278 mlx5e_tc_reoffload_flows_work);
281 void mlx5e_rep_tc_disable(struct mlx5e_priv *priv)
283 struct mlx5e_rep_priv *rpriv = priv->ppriv;
285 cancel_work_sync(&rpriv->uplink_priv.reoffload_flows_work);
288 int mlx5e_rep_tc_event_port_affinity(struct mlx5e_priv *priv)
290 struct mlx5e_rep_priv *rpriv = priv->ppriv;
292 queue_work(priv->wq, &rpriv->uplink_priv.reoffload_flows_work);
297 static struct mlx5e_rep_indr_block_priv *
298 mlx5e_rep_indr_block_priv_lookup(struct mlx5e_rep_priv *rpriv,
299 struct net_device *netdev)
301 struct mlx5e_rep_indr_block_priv *cb_priv;
303 /* All callback list access should be protected by RTNL. */
306 list_for_each_entry(cb_priv,
307 &rpriv->uplink_priv.tc_indr_block_priv_list,
309 if (cb_priv->netdev == netdev)
316 mlx5e_rep_indr_offload(struct net_device *netdev,
317 struct flow_cls_offload *flower,
318 struct mlx5e_rep_indr_block_priv *indr_priv,
321 struct mlx5e_priv *priv = netdev_priv(indr_priv->rpriv->netdev);
324 if (!netif_device_present(indr_priv->rpriv->netdev))
327 switch (flower->command) {
328 case FLOW_CLS_REPLACE:
329 err = mlx5e_configure_flower(netdev, priv, flower, flags);
331 case FLOW_CLS_DESTROY:
332 err = mlx5e_delete_flower(netdev, priv, flower, flags);
335 err = mlx5e_stats_flower(netdev, priv, flower, flags);
344 static int mlx5e_rep_indr_setup_tc_cb(enum tc_setup_type type,
345 void *type_data, void *indr_priv)
347 unsigned long flags = MLX5_TC_FLAG(EGRESS) | MLX5_TC_FLAG(ESW_OFFLOAD);
348 struct mlx5e_rep_indr_block_priv *priv = indr_priv;
351 case TC_SETUP_CLSFLOWER:
352 return mlx5e_rep_indr_offload(priv->netdev, type_data, priv,
359 static int mlx5e_rep_indr_setup_ft_cb(enum tc_setup_type type,
360 void *type_data, void *indr_priv)
362 struct mlx5e_rep_indr_block_priv *priv = indr_priv;
363 struct flow_cls_offload *f = type_data;
364 struct flow_cls_offload tmp;
365 struct mlx5e_priv *mpriv;
366 struct mlx5_eswitch *esw;
370 mpriv = netdev_priv(priv->rpriv->netdev);
371 esw = mpriv->mdev->priv.eswitch;
373 flags = MLX5_TC_FLAG(EGRESS) |
374 MLX5_TC_FLAG(ESW_OFFLOAD) |
375 MLX5_TC_FLAG(FT_OFFLOAD);
378 case TC_SETUP_CLSFLOWER:
379 memcpy(&tmp, f, sizeof(*f));
381 /* Re-use tc offload path by moving the ft flow to the
384 * FT offload can use prio range [0, INT_MAX], so we normalize
385 * it to range [1, mlx5_esw_chains_get_prio_range(esw)]
386 * as with tc, where prio 0 isn't supported.
388 * We only support chain 0 of FT offload.
390 if (!mlx5_chains_prios_supported(esw_chains(esw)) ||
391 tmp.common.prio >= mlx5_chains_get_prio_range(esw_chains(esw)) ||
392 tmp.common.chain_index)
395 tmp.common.chain_index = mlx5_chains_get_nf_ft_chain(esw_chains(esw));
397 err = mlx5e_rep_indr_offload(priv->netdev, &tmp, priv, flags);
398 memcpy(&f->stats, &tmp.stats, sizeof(f->stats));
405 static void mlx5e_rep_indr_block_unbind(void *cb_priv)
407 struct mlx5e_rep_indr_block_priv *indr_priv = cb_priv;
409 list_del(&indr_priv->list);
413 static LIST_HEAD(mlx5e_block_cb_list);
416 mlx5e_rep_indr_setup_block(struct net_device *netdev, struct Qdisc *sch,
417 struct mlx5e_rep_priv *rpriv,
418 struct flow_block_offload *f,
419 flow_setup_cb_t *setup_cb,
421 void (*cleanup)(struct flow_block_cb *block_cb))
423 struct mlx5e_priv *priv = netdev_priv(rpriv->netdev);
424 struct mlx5e_rep_indr_block_priv *indr_priv;
425 struct flow_block_cb *block_cb;
427 if (!mlx5e_tc_tun_device_to_offload(priv, netdev) &&
428 !(is_vlan_dev(netdev) && vlan_dev_real_dev(netdev) == rpriv->netdev))
431 if (f->binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
434 f->unlocked_driver_cb = true;
435 f->driver_block_list = &mlx5e_block_cb_list;
437 switch (f->command) {
438 case FLOW_BLOCK_BIND:
439 indr_priv = mlx5e_rep_indr_block_priv_lookup(rpriv, netdev);
443 indr_priv = kmalloc(sizeof(*indr_priv), GFP_KERNEL);
447 indr_priv->netdev = netdev;
448 indr_priv->rpriv = rpriv;
449 list_add(&indr_priv->list,
450 &rpriv->uplink_priv.tc_indr_block_priv_list);
452 block_cb = flow_indr_block_cb_alloc(setup_cb, indr_priv, indr_priv,
453 mlx5e_rep_indr_block_unbind,
454 f, netdev, sch, data, rpriv,
456 if (IS_ERR(block_cb)) {
457 list_del(&indr_priv->list);
459 return PTR_ERR(block_cb);
461 flow_block_cb_add(block_cb, f);
462 list_add_tail(&block_cb->driver_list, &mlx5e_block_cb_list);
465 case FLOW_BLOCK_UNBIND:
466 indr_priv = mlx5e_rep_indr_block_priv_lookup(rpriv, netdev);
470 block_cb = flow_block_cb_lookup(f->block, setup_cb, indr_priv);
474 flow_indr_block_cb_remove(block_cb, f);
475 list_del(&block_cb->driver_list);
484 int mlx5e_rep_indr_setup_cb(struct net_device *netdev, struct Qdisc *sch, void *cb_priv,
485 enum tc_setup_type type, void *type_data,
487 void (*cleanup)(struct flow_block_cb *block_cb))
491 return mlx5e_rep_indr_setup_block(netdev, sch, cb_priv, type_data,
492 mlx5e_rep_indr_setup_tc_cb,
495 return mlx5e_rep_indr_setup_block(netdev, sch, cb_priv, type_data,
496 mlx5e_rep_indr_setup_ft_cb,
503 int mlx5e_rep_tc_netdevice_event_register(struct mlx5e_rep_priv *rpriv)
505 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
507 /* init indirect block notifications */
508 INIT_LIST_HEAD(&uplink_priv->tc_indr_block_priv_list);
510 return flow_indr_dev_register(mlx5e_rep_indr_setup_cb, rpriv);
513 void mlx5e_rep_tc_netdevice_event_unregister(struct mlx5e_rep_priv *rpriv)
515 flow_indr_dev_unregister(mlx5e_rep_indr_setup_cb, rpriv,
516 mlx5e_rep_indr_block_unbind);
519 static bool mlx5e_restore_tunnel(struct mlx5e_priv *priv, struct sk_buff *skb,
520 struct mlx5e_tc_update_priv *tc_priv,
523 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
524 struct tunnel_match_enc_opts enc_opts = {};
525 struct mlx5_rep_uplink_priv *uplink_priv;
526 struct mlx5e_rep_priv *uplink_rpriv;
527 struct metadata_dst *tun_dst;
528 struct tunnel_match_key key;
529 u32 tun_id, enc_opts_id;
530 struct net_device *dev;
533 enc_opts_id = tunnel_id & ENC_OPTS_BITS_MASK;
534 tun_id = tunnel_id >> ENC_OPTS_BITS;
539 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
540 uplink_priv = &uplink_rpriv->uplink_priv;
542 err = mapping_find(uplink_priv->tunnel_mapping, tun_id, &key);
545 netdev_dbg(priv->netdev,
546 "Couldn't find tunnel for tun_id: %d, err: %d\n",
552 err = mapping_find(uplink_priv->tunnel_enc_opts_mapping,
553 enc_opts_id, &enc_opts);
555 netdev_dbg(priv->netdev,
556 "Couldn't find tunnel (opts) for tun_id: %d, err: %d\n",
562 if (key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
563 tun_dst = __ip_tun_set_dst(key.enc_ipv4.src, key.enc_ipv4.dst,
564 key.enc_ip.tos, key.enc_ip.ttl,
565 key.enc_tp.dst, TUNNEL_KEY,
566 key32_to_tunnel_id(key.enc_key_id.keyid),
568 } else if (key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
569 tun_dst = __ipv6_tun_set_dst(&key.enc_ipv6.src, &key.enc_ipv6.dst,
570 key.enc_ip.tos, key.enc_ip.ttl,
571 key.enc_tp.dst, 0, TUNNEL_KEY,
572 key32_to_tunnel_id(key.enc_key_id.keyid),
575 netdev_dbg(priv->netdev,
576 "Couldn't restore tunnel, unsupported addr_type: %d\n",
577 key.enc_control.addr_type);
582 netdev_dbg(priv->netdev, "Couldn't restore tunnel, no tun_dst\n");
586 tun_dst->u.tun_info.key.tp_src = key.enc_tp.src;
588 if (enc_opts.key.len)
589 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
592 enc_opts.key.dst_opt_type);
594 skb_dst_set(skb, (struct dst_entry *)tun_dst);
595 dev = dev_get_by_index(&init_net, key.filter_ifindex);
597 netdev_dbg(priv->netdev,
598 "Couldn't find tunnel device with ifindex: %d\n",
603 /* Set tun_dev so we do dev_put() after datapath */
604 tc_priv->tun_dev = dev;
611 static bool mlx5e_restore_skb_chain(struct sk_buff *skb, u32 chain, u32 reg_c1,
612 struct mlx5e_tc_update_priv *tc_priv)
614 struct mlx5e_priv *priv = netdev_priv(skb->dev);
615 u32 tunnel_id = (reg_c1 >> ESW_TUN_OFFSET) & TUNNEL_ID_MASK;
617 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
619 struct mlx5_rep_uplink_priv *uplink_priv;
620 struct mlx5e_rep_priv *uplink_rpriv;
621 struct tc_skb_ext *tc_skb_ext;
622 struct mlx5_eswitch *esw;
625 tc_skb_ext = tc_skb_ext_alloc(skb);
630 tc_skb_ext->chain = chain;
631 zone_restore_id = reg_c1 & ESW_ZONE_ID_MASK;
632 esw = priv->mdev->priv.eswitch;
633 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
634 uplink_priv = &uplink_rpriv->uplink_priv;
635 if (!mlx5e_tc_ct_restore_flow(uplink_priv->ct_priv, skb,
639 #endif /* CONFIG_NET_TC_SKB_EXT */
641 return mlx5e_restore_tunnel(priv, skb, tc_priv, tunnel_id);
644 static void mlx5e_restore_skb_sample(struct mlx5e_priv *priv, struct sk_buff *skb,
645 struct mlx5_mapped_obj *mapped_obj,
646 struct mlx5e_tc_update_priv *tc_priv)
648 if (!mlx5e_restore_tunnel(priv, skb, tc_priv, mapped_obj->sample.tunnel_id)) {
649 netdev_dbg(priv->netdev,
650 "Failed to restore tunnel info for sampled packet\n");
653 #if IS_ENABLED(CONFIG_MLX5_TC_SAMPLE)
654 mlx5e_tc_sample_skb(skb, mapped_obj);
655 #endif /* CONFIG_MLX5_TC_SAMPLE */
656 mlx5_rep_tc_post_napi_receive(tc_priv);
659 bool mlx5e_rep_tc_update_skb(struct mlx5_cqe64 *cqe,
661 struct mlx5e_tc_update_priv *tc_priv)
663 struct mlx5_mapped_obj mapped_obj;
664 struct mlx5_eswitch *esw;
665 struct mlx5e_priv *priv;
669 reg_c0 = (be32_to_cpu(cqe->sop_drop_qpn) & MLX5E_TC_FLOW_ID_MASK);
670 if (!reg_c0 || reg_c0 == MLX5_FS_DEFAULT_FLOW_TAG)
673 /* If reg_c0 is not equal to the default flow tag then skb->mark
674 * is not supported and must be reset back to 0.
678 priv = netdev_priv(skb->dev);
679 esw = priv->mdev->priv.eswitch;
680 err = mapping_find(esw->offloads.reg_c0_obj_pool, reg_c0, &mapped_obj);
682 netdev_dbg(priv->netdev,
683 "Couldn't find mapped object for reg_c0: %d, err: %d\n",
688 if (mapped_obj.type == MLX5_MAPPED_OBJ_CHAIN) {
689 u32 reg_c1 = be32_to_cpu(cqe->ft_metadata);
691 return mlx5e_restore_skb_chain(skb, mapped_obj.chain, reg_c1, tc_priv);
692 } else if (mapped_obj.type == MLX5_MAPPED_OBJ_SAMPLE) {
693 mlx5e_restore_skb_sample(priv, skb, &mapped_obj, tc_priv);
696 netdev_dbg(priv->netdev, "Invalid mapped object type: %d\n", mapped_obj.type);
703 void mlx5_rep_tc_post_napi_receive(struct mlx5e_tc_update_priv *tc_priv)
705 if (tc_priv->tun_dev)
706 dev_put(tc_priv->tun_dev);