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,
30 struct mlx5e_neigh *m_neigh,
31 struct net_device *neigh_dev)
33 struct mlx5e_rep_priv *rpriv = priv->ppriv;
34 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
35 struct mlx5_tun_entropy *tun_entropy = &uplink_priv->tun_entropy;
36 struct mlx5e_neigh_hash_entry *nhe;
39 err = mlx5_tun_entropy_refcount_inc(tun_entropy, e->reformat_type);
43 mutex_lock(&rpriv->neigh_update.encap_lock);
44 nhe = mlx5e_rep_neigh_entry_lookup(priv, m_neigh);
46 err = mlx5e_rep_neigh_entry_create(priv, m_neigh, neigh_dev, &nhe);
48 mutex_unlock(&rpriv->neigh_update.encap_lock);
49 mlx5_tun_entropy_refcount_dec(tun_entropy,
56 spin_lock(&nhe->encap_list_lock);
57 list_add_rcu(&e->encap_list, &nhe->encap_list);
58 spin_unlock(&nhe->encap_list_lock);
60 mutex_unlock(&rpriv->neigh_update.encap_lock);
65 void mlx5e_rep_encap_entry_detach(struct mlx5e_priv *priv,
66 struct mlx5e_encap_entry *e)
68 struct mlx5e_rep_priv *rpriv = priv->ppriv;
69 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
70 struct mlx5_tun_entropy *tun_entropy = &uplink_priv->tun_entropy;
75 spin_lock(&e->nhe->encap_list_lock);
76 list_del_rcu(&e->encap_list);
77 spin_unlock(&e->nhe->encap_list_lock);
79 mlx5e_rep_neigh_entry_release(e->nhe);
81 mlx5_tun_entropy_refcount_dec(tun_entropy, e->reformat_type);
84 void mlx5e_rep_update_flows(struct mlx5e_priv *priv,
85 struct mlx5e_encap_entry *e,
87 unsigned char ha[ETH_ALEN])
89 struct ethhdr *eth = (struct ethhdr *)e->encap_header;
90 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
96 /* wait for encap to be fully initialized */
97 wait_for_completion(&e->res_ready);
99 mutex_lock(&esw->offloads.encap_tbl_lock);
100 encap_connected = !!(e->flags & MLX5_ENCAP_ENTRY_VALID);
101 if (e->compl_result < 0 || (encap_connected == neigh_connected &&
102 ether_addr_equal(e->h_dest, ha)))
105 mlx5e_take_all_encap_flows(e, &flow_list);
107 if ((e->flags & MLX5_ENCAP_ENTRY_VALID) &&
108 (!neigh_connected || !ether_addr_equal(e->h_dest, ha)))
109 mlx5e_tc_encap_flows_del(priv, e, &flow_list);
111 if (neigh_connected && !(e->flags & MLX5_ENCAP_ENTRY_VALID)) {
112 struct net_device *route_dev;
114 ether_addr_copy(e->h_dest, ha);
115 ether_addr_copy(eth->h_dest, ha);
116 /* Update the encap source mac, in case that we delete
117 * the flows when encap source mac changed.
119 route_dev = __dev_get_by_index(dev_net(priv->netdev), e->route_dev_ifindex);
121 ether_addr_copy(eth->h_source, route_dev->dev_addr);
123 mlx5e_tc_encap_flows_add(priv, e, &flow_list);
126 mutex_unlock(&esw->offloads.encap_tbl_lock);
127 mlx5e_put_flow_list(priv, &flow_list);
131 mlx5e_rep_setup_tc_cls_flower(struct mlx5e_priv *priv,
132 struct flow_cls_offload *cls_flower, int flags)
134 switch (cls_flower->command) {
135 case FLOW_CLS_REPLACE:
136 return mlx5e_configure_flower(priv->netdev, priv, cls_flower,
138 case FLOW_CLS_DESTROY:
139 return mlx5e_delete_flower(priv->netdev, priv, cls_flower,
142 return mlx5e_stats_flower(priv->netdev, priv, cls_flower,
150 int mlx5e_rep_setup_tc_cls_matchall(struct mlx5e_priv *priv,
151 struct tc_cls_matchall_offload *ma)
153 switch (ma->command) {
154 case TC_CLSMATCHALL_REPLACE:
155 return mlx5e_tc_configure_matchall(priv, ma);
156 case TC_CLSMATCHALL_DESTROY:
157 return mlx5e_tc_delete_matchall(priv, ma);
158 case TC_CLSMATCHALL_STATS:
159 mlx5e_tc_stats_matchall(priv, ma);
166 static int mlx5e_rep_setup_tc_cb(enum tc_setup_type type, void *type_data,
169 unsigned long flags = MLX5_TC_FLAG(INGRESS) | MLX5_TC_FLAG(ESW_OFFLOAD);
170 struct mlx5e_priv *priv = cb_priv;
172 if (!priv->netdev || !netif_device_present(priv->netdev))
176 case TC_SETUP_CLSFLOWER:
177 return mlx5e_rep_setup_tc_cls_flower(priv, type_data, flags);
178 case TC_SETUP_CLSMATCHALL:
179 return mlx5e_rep_setup_tc_cls_matchall(priv, type_data);
185 static int mlx5e_rep_setup_ft_cb(enum tc_setup_type type, void *type_data,
188 struct flow_cls_offload tmp, *f = type_data;
189 struct mlx5e_priv *priv = cb_priv;
190 struct mlx5_eswitch *esw;
194 flags = MLX5_TC_FLAG(INGRESS) |
195 MLX5_TC_FLAG(ESW_OFFLOAD) |
196 MLX5_TC_FLAG(FT_OFFLOAD);
197 esw = priv->mdev->priv.eswitch;
200 case TC_SETUP_CLSFLOWER:
201 memcpy(&tmp, f, sizeof(*f));
203 if (!mlx5_chains_prios_supported(esw_chains(esw)))
206 /* Re-use tc offload path by moving the ft flow to the
209 * FT offload can use prio range [0, INT_MAX], so we normalize
210 * it to range [1, mlx5_esw_chains_get_prio_range(esw)]
211 * as with tc, where prio 0 isn't supported.
213 * We only support chain 0 of FT offload.
215 if (tmp.common.prio >= mlx5_chains_get_prio_range(esw_chains(esw)))
217 if (tmp.common.chain_index != 0)
220 tmp.common.chain_index = mlx5_chains_get_nf_ft_chain(esw_chains(esw));
222 err = mlx5e_rep_setup_tc_cls_flower(priv, &tmp, flags);
223 memcpy(&f->stats, &tmp.stats, sizeof(f->stats));
230 static LIST_HEAD(mlx5e_rep_block_tc_cb_list);
231 static LIST_HEAD(mlx5e_rep_block_ft_cb_list);
232 int mlx5e_rep_setup_tc(struct net_device *dev, enum tc_setup_type type,
235 struct mlx5e_priv *priv = netdev_priv(dev);
236 struct flow_block_offload *f = type_data;
238 f->unlocked_driver_cb = true;
242 return flow_block_cb_setup_simple(type_data,
243 &mlx5e_rep_block_tc_cb_list,
244 mlx5e_rep_setup_tc_cb,
247 return flow_block_cb_setup_simple(type_data,
248 &mlx5e_rep_block_ft_cb_list,
249 mlx5e_rep_setup_ft_cb,
256 int mlx5e_rep_tc_init(struct mlx5e_rep_priv *rpriv)
258 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
261 mutex_init(&uplink_priv->unready_flows_lock);
262 INIT_LIST_HEAD(&uplink_priv->unready_flows);
264 /* init shared tc flow table */
265 err = mlx5e_tc_esw_init(&uplink_priv->tc_ht);
269 void mlx5e_rep_tc_cleanup(struct mlx5e_rep_priv *rpriv)
271 /* delete shared tc flow table */
272 mlx5e_tc_esw_cleanup(&rpriv->uplink_priv.tc_ht);
273 mutex_destroy(&rpriv->uplink_priv.unready_flows_lock);
276 void mlx5e_rep_tc_enable(struct mlx5e_priv *priv)
278 struct mlx5e_rep_priv *rpriv = priv->ppriv;
280 INIT_WORK(&rpriv->uplink_priv.reoffload_flows_work,
281 mlx5e_tc_reoffload_flows_work);
284 void mlx5e_rep_tc_disable(struct mlx5e_priv *priv)
286 struct mlx5e_rep_priv *rpriv = priv->ppriv;
288 cancel_work_sync(&rpriv->uplink_priv.reoffload_flows_work);
291 int mlx5e_rep_tc_event_port_affinity(struct mlx5e_priv *priv)
293 struct mlx5e_rep_priv *rpriv = priv->ppriv;
295 queue_work(priv->wq, &rpriv->uplink_priv.reoffload_flows_work);
300 static struct mlx5e_rep_indr_block_priv *
301 mlx5e_rep_indr_block_priv_lookup(struct mlx5e_rep_priv *rpriv,
302 struct net_device *netdev)
304 struct mlx5e_rep_indr_block_priv *cb_priv;
306 /* All callback list access should be protected by RTNL. */
309 list_for_each_entry(cb_priv,
310 &rpriv->uplink_priv.tc_indr_block_priv_list,
312 if (cb_priv->netdev == netdev)
319 mlx5e_rep_indr_offload(struct net_device *netdev,
320 struct flow_cls_offload *flower,
321 struct mlx5e_rep_indr_block_priv *indr_priv,
324 struct mlx5e_priv *priv = netdev_priv(indr_priv->rpriv->netdev);
327 if (!netif_device_present(indr_priv->rpriv->netdev))
330 switch (flower->command) {
331 case FLOW_CLS_REPLACE:
332 err = mlx5e_configure_flower(netdev, priv, flower, flags);
334 case FLOW_CLS_DESTROY:
335 err = mlx5e_delete_flower(netdev, priv, flower, flags);
338 err = mlx5e_stats_flower(netdev, priv, flower, flags);
347 static int mlx5e_rep_indr_setup_tc_cb(enum tc_setup_type type,
348 void *type_data, void *indr_priv)
350 unsigned long flags = MLX5_TC_FLAG(EGRESS) | MLX5_TC_FLAG(ESW_OFFLOAD);
351 struct mlx5e_rep_indr_block_priv *priv = indr_priv;
354 case TC_SETUP_CLSFLOWER:
355 return mlx5e_rep_indr_offload(priv->netdev, type_data, priv,
362 static int mlx5e_rep_indr_setup_ft_cb(enum tc_setup_type type,
363 void *type_data, void *indr_priv)
365 struct mlx5e_rep_indr_block_priv *priv = indr_priv;
366 struct flow_cls_offload *f = type_data;
367 struct flow_cls_offload tmp;
368 struct mlx5e_priv *mpriv;
369 struct mlx5_eswitch *esw;
373 mpriv = netdev_priv(priv->rpriv->netdev);
374 esw = mpriv->mdev->priv.eswitch;
376 flags = MLX5_TC_FLAG(EGRESS) |
377 MLX5_TC_FLAG(ESW_OFFLOAD) |
378 MLX5_TC_FLAG(FT_OFFLOAD);
381 case TC_SETUP_CLSFLOWER:
382 memcpy(&tmp, f, sizeof(*f));
384 /* Re-use tc offload path by moving the ft flow to the
387 * FT offload can use prio range [0, INT_MAX], so we normalize
388 * it to range [1, mlx5_esw_chains_get_prio_range(esw)]
389 * as with tc, where prio 0 isn't supported.
391 * We only support chain 0 of FT offload.
393 if (!mlx5_chains_prios_supported(esw_chains(esw)) ||
394 tmp.common.prio >= mlx5_chains_get_prio_range(esw_chains(esw)) ||
395 tmp.common.chain_index)
398 tmp.common.chain_index = mlx5_chains_get_nf_ft_chain(esw_chains(esw));
400 err = mlx5e_rep_indr_offload(priv->netdev, &tmp, priv, flags);
401 memcpy(&f->stats, &tmp.stats, sizeof(f->stats));
408 static void mlx5e_rep_indr_block_unbind(void *cb_priv)
410 struct mlx5e_rep_indr_block_priv *indr_priv = cb_priv;
412 list_del(&indr_priv->list);
416 static LIST_HEAD(mlx5e_block_cb_list);
419 mlx5e_rep_indr_setup_block(struct net_device *netdev, struct Qdisc *sch,
420 struct mlx5e_rep_priv *rpriv,
421 struct flow_block_offload *f,
422 flow_setup_cb_t *setup_cb,
424 void (*cleanup)(struct flow_block_cb *block_cb))
426 struct mlx5e_priv *priv = netdev_priv(rpriv->netdev);
427 struct mlx5e_rep_indr_block_priv *indr_priv;
428 struct flow_block_cb *block_cb;
430 if (!mlx5e_tc_tun_device_to_offload(priv, netdev) &&
431 !(is_vlan_dev(netdev) && vlan_dev_real_dev(netdev) == rpriv->netdev))
434 if (f->binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
437 f->unlocked_driver_cb = true;
438 f->driver_block_list = &mlx5e_block_cb_list;
440 switch (f->command) {
441 case FLOW_BLOCK_BIND:
442 indr_priv = mlx5e_rep_indr_block_priv_lookup(rpriv, netdev);
446 indr_priv = kmalloc(sizeof(*indr_priv), GFP_KERNEL);
450 indr_priv->netdev = netdev;
451 indr_priv->rpriv = rpriv;
452 list_add(&indr_priv->list,
453 &rpriv->uplink_priv.tc_indr_block_priv_list);
455 block_cb = flow_indr_block_cb_alloc(setup_cb, indr_priv, indr_priv,
456 mlx5e_rep_indr_block_unbind,
457 f, netdev, sch, data, rpriv,
459 if (IS_ERR(block_cb)) {
460 list_del(&indr_priv->list);
462 return PTR_ERR(block_cb);
464 flow_block_cb_add(block_cb, f);
465 list_add_tail(&block_cb->driver_list, &mlx5e_block_cb_list);
468 case FLOW_BLOCK_UNBIND:
469 indr_priv = mlx5e_rep_indr_block_priv_lookup(rpriv, netdev);
473 block_cb = flow_block_cb_lookup(f->block, setup_cb, indr_priv);
477 flow_indr_block_cb_remove(block_cb, f);
478 list_del(&block_cb->driver_list);
487 int mlx5e_rep_indr_setup_cb(struct net_device *netdev, struct Qdisc *sch, void *cb_priv,
488 enum tc_setup_type type, void *type_data,
490 void (*cleanup)(struct flow_block_cb *block_cb))
494 return mlx5e_rep_indr_setup_block(netdev, sch, cb_priv, type_data,
495 mlx5e_rep_indr_setup_tc_cb,
498 return mlx5e_rep_indr_setup_block(netdev, sch, cb_priv, type_data,
499 mlx5e_rep_indr_setup_ft_cb,
506 int mlx5e_rep_tc_netdevice_event_register(struct mlx5e_rep_priv *rpriv)
508 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
510 /* init indirect block notifications */
511 INIT_LIST_HEAD(&uplink_priv->tc_indr_block_priv_list);
513 return flow_indr_dev_register(mlx5e_rep_indr_setup_cb, rpriv);
516 void mlx5e_rep_tc_netdevice_event_unregister(struct mlx5e_rep_priv *rpriv)
518 flow_indr_dev_unregister(mlx5e_rep_indr_setup_cb, rpriv,
519 mlx5e_rep_indr_block_unbind);
522 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
523 static bool mlx5e_restore_tunnel(struct mlx5e_priv *priv, struct sk_buff *skb,
524 struct mlx5e_tc_update_priv *tc_priv,
527 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
528 struct tunnel_match_enc_opts enc_opts = {};
529 struct mlx5_rep_uplink_priv *uplink_priv;
530 struct mlx5e_rep_priv *uplink_rpriv;
531 struct metadata_dst *tun_dst;
532 struct tunnel_match_key key;
533 u32 tun_id, enc_opts_id;
534 struct net_device *dev;
537 enc_opts_id = tunnel_id & ENC_OPTS_BITS_MASK;
538 tun_id = tunnel_id >> ENC_OPTS_BITS;
543 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
544 uplink_priv = &uplink_rpriv->uplink_priv;
546 err = mapping_find(uplink_priv->tunnel_mapping, tun_id, &key);
549 netdev_dbg(priv->netdev,
550 "Couldn't find tunnel for tun_id: %d, err: %d\n",
556 err = mapping_find(uplink_priv->tunnel_enc_opts_mapping,
557 enc_opts_id, &enc_opts);
559 netdev_dbg(priv->netdev,
560 "Couldn't find tunnel (opts) for tun_id: %d, err: %d\n",
566 if (key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
567 tun_dst = __ip_tun_set_dst(key.enc_ipv4.src, key.enc_ipv4.dst,
568 key.enc_ip.tos, key.enc_ip.ttl,
569 key.enc_tp.dst, TUNNEL_KEY,
570 key32_to_tunnel_id(key.enc_key_id.keyid),
572 } else if (key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
573 tun_dst = __ipv6_tun_set_dst(&key.enc_ipv6.src, &key.enc_ipv6.dst,
574 key.enc_ip.tos, key.enc_ip.ttl,
575 key.enc_tp.dst, 0, TUNNEL_KEY,
576 key32_to_tunnel_id(key.enc_key_id.keyid),
579 netdev_dbg(priv->netdev,
580 "Couldn't restore tunnel, unsupported addr_type: %d\n",
581 key.enc_control.addr_type);
586 netdev_dbg(priv->netdev, "Couldn't restore tunnel, no tun_dst\n");
590 tun_dst->u.tun_info.key.tp_src = key.enc_tp.src;
592 if (enc_opts.key.len)
593 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
596 enc_opts.key.dst_opt_type);
598 skb_dst_set(skb, (struct dst_entry *)tun_dst);
599 dev = dev_get_by_index(&init_net, key.filter_ifindex);
601 netdev_dbg(priv->netdev,
602 "Couldn't find tunnel device with ifindex: %d\n",
607 /* Set tun_dev so we do dev_put() after datapath */
608 tc_priv->tun_dev = dev;
614 #endif /* CONFIG_NET_TC_SKB_EXT */
616 bool mlx5e_rep_tc_update_skb(struct mlx5_cqe64 *cqe,
618 struct mlx5e_tc_update_priv *tc_priv)
620 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
621 u32 chain = 0, reg_c0, reg_c1, tunnel_id, zone_restore_id;
622 struct mlx5_rep_uplink_priv *uplink_priv;
623 struct mlx5e_rep_priv *uplink_rpriv;
624 struct tc_skb_ext *tc_skb_ext;
625 struct mlx5_eswitch *esw;
626 struct mlx5e_priv *priv;
629 reg_c0 = (be32_to_cpu(cqe->sop_drop_qpn) & MLX5E_TC_FLOW_ID_MASK);
630 if (!reg_c0 || reg_c0 == MLX5_FS_DEFAULT_FLOW_TAG)
633 /* If reg_c0 is not equal to the default flow tag then skb->mark
634 * is not supported and must be reset back to 0.
638 reg_c1 = be32_to_cpu(cqe->ft_metadata);
640 priv = netdev_priv(skb->dev);
641 esw = priv->mdev->priv.eswitch;
643 err = mlx5_get_chain_for_tag(esw_chains(esw), reg_c0, &chain);
645 netdev_dbg(priv->netdev,
646 "Couldn't find chain for chain tag: %d, err: %d\n",
652 tc_skb_ext = skb_ext_add(skb, TC_SKB_EXT);
658 tc_skb_ext->chain = chain;
660 zone_restore_id = reg_c1 & ESW_ZONE_ID_MASK;
662 uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
663 uplink_priv = &uplink_rpriv->uplink_priv;
664 if (!mlx5e_tc_ct_restore_flow(uplink_priv->ct_priv, skb,
669 tunnel_id = reg_c1 >> ESW_TUN_OFFSET;
670 return mlx5e_restore_tunnel(priv, skb, tc_priv, tunnel_id);
671 #endif /* CONFIG_NET_TC_SKB_EXT */
676 void mlx5_rep_tc_post_napi_receive(struct mlx5e_tc_update_priv *tc_priv)
678 if (tc_priv->tun_dev)
679 dev_put(tc_priv->tun_dev);