1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
2 /* Copyright (c) 2016-2018 Mellanox Technologies. All rights reserved */
4 #include <linux/kernel.h>
5 #include <linux/types.h>
6 #include <linux/rhashtable.h>
7 #include <linux/bitops.h>
9 #include <linux/notifier.h>
10 #include <linux/inetdevice.h>
11 #include <linux/netdevice.h>
12 #include <linux/if_bridge.h>
13 #include <linux/socket.h>
14 #include <linux/route.h>
15 #include <linux/gcd.h>
16 #include <linux/if_macvlan.h>
17 #include <linux/refcount.h>
18 #include <linux/jhash.h>
19 #include <linux/net_namespace.h>
20 #include <linux/mutex.h>
21 #include <net/netevent.h>
22 #include <net/neighbour.h>
24 #include <net/ip_fib.h>
25 #include <net/ip6_fib.h>
26 #include <net/nexthop.h>
27 #include <net/fib_rules.h>
28 #include <net/ip_tunnels.h>
29 #include <net/l3mdev.h>
30 #include <net/addrconf.h>
31 #include <net/ndisc.h>
33 #include <net/fib_notifier.h>
34 #include <net/switchdev.h>
39 #include "spectrum_cnt.h"
40 #include "spectrum_dpipe.h"
41 #include "spectrum_ipip.h"
42 #include "spectrum_mr.h"
43 #include "spectrum_mr_tcam.h"
44 #include "spectrum_router.h"
45 #include "spectrum_span.h"
49 struct mlxsw_sp_lpm_tree;
50 struct mlxsw_sp_rif_ops;
53 struct list_head nexthop_list;
54 struct list_head neigh_list;
55 struct net_device *dev; /* NULL for underlay RIF */
56 struct mlxsw_sp_fid *fid;
57 unsigned char addr[ETH_ALEN];
61 const struct mlxsw_sp_rif_ops *ops;
62 struct mlxsw_sp *mlxsw_sp;
64 unsigned int counter_ingress;
65 bool counter_ingress_valid;
66 unsigned int counter_egress;
67 bool counter_egress_valid;
70 struct mlxsw_sp_rif_params {
71 struct net_device *dev;
80 struct mlxsw_sp_rif_subport {
81 struct mlxsw_sp_rif common;
91 struct mlxsw_sp_rif_ipip_lb {
92 struct mlxsw_sp_rif common;
93 struct mlxsw_sp_rif_ipip_lb_config lb_config;
94 u16 ul_vr_id; /* Reserved for Spectrum-2. */
95 u16 ul_rif_id; /* Reserved for Spectrum. */
98 struct mlxsw_sp_rif_params_ipip_lb {
99 struct mlxsw_sp_rif_params common;
100 struct mlxsw_sp_rif_ipip_lb_config lb_config;
103 struct mlxsw_sp_rif_ops {
104 enum mlxsw_sp_rif_type type;
107 void (*setup)(struct mlxsw_sp_rif *rif,
108 const struct mlxsw_sp_rif_params *params);
109 int (*configure)(struct mlxsw_sp_rif *rif);
110 void (*deconfigure)(struct mlxsw_sp_rif *rif);
111 struct mlxsw_sp_fid * (*fid_get)(struct mlxsw_sp_rif *rif,
112 struct netlink_ext_ack *extack);
113 void (*fdb_del)(struct mlxsw_sp_rif *rif, const char *mac);
116 static struct mlxsw_sp_rif *
117 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
118 const struct net_device *dev);
119 static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif);
120 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree);
121 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
122 struct mlxsw_sp_lpm_tree *lpm_tree);
123 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
124 const struct mlxsw_sp_fib *fib,
126 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
127 const struct mlxsw_sp_fib *fib);
129 static unsigned int *
130 mlxsw_sp_rif_p_counter_get(struct mlxsw_sp_rif *rif,
131 enum mlxsw_sp_rif_counter_dir dir)
134 case MLXSW_SP_RIF_COUNTER_EGRESS:
135 return &rif->counter_egress;
136 case MLXSW_SP_RIF_COUNTER_INGRESS:
137 return &rif->counter_ingress;
143 mlxsw_sp_rif_counter_valid_get(struct mlxsw_sp_rif *rif,
144 enum mlxsw_sp_rif_counter_dir dir)
147 case MLXSW_SP_RIF_COUNTER_EGRESS:
148 return rif->counter_egress_valid;
149 case MLXSW_SP_RIF_COUNTER_INGRESS:
150 return rif->counter_ingress_valid;
156 mlxsw_sp_rif_counter_valid_set(struct mlxsw_sp_rif *rif,
157 enum mlxsw_sp_rif_counter_dir dir,
161 case MLXSW_SP_RIF_COUNTER_EGRESS:
162 rif->counter_egress_valid = valid;
164 case MLXSW_SP_RIF_COUNTER_INGRESS:
165 rif->counter_ingress_valid = valid;
170 static int mlxsw_sp_rif_counter_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
171 unsigned int counter_index, bool enable,
172 enum mlxsw_sp_rif_counter_dir dir)
174 char ritr_pl[MLXSW_REG_RITR_LEN];
175 bool is_egress = false;
178 if (dir == MLXSW_SP_RIF_COUNTER_EGRESS)
180 mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
181 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
185 mlxsw_reg_ritr_counter_pack(ritr_pl, counter_index, enable,
187 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
190 int mlxsw_sp_rif_counter_value_get(struct mlxsw_sp *mlxsw_sp,
191 struct mlxsw_sp_rif *rif,
192 enum mlxsw_sp_rif_counter_dir dir, u64 *cnt)
194 char ricnt_pl[MLXSW_REG_RICNT_LEN];
195 unsigned int *p_counter_index;
199 valid = mlxsw_sp_rif_counter_valid_get(rif, dir);
203 p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
204 if (!p_counter_index)
206 mlxsw_reg_ricnt_pack(ricnt_pl, *p_counter_index,
207 MLXSW_REG_RICNT_OPCODE_NOP);
208 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
211 *cnt = mlxsw_reg_ricnt_good_unicast_packets_get(ricnt_pl);
215 static int mlxsw_sp_rif_counter_clear(struct mlxsw_sp *mlxsw_sp,
216 unsigned int counter_index)
218 char ricnt_pl[MLXSW_REG_RICNT_LEN];
220 mlxsw_reg_ricnt_pack(ricnt_pl, counter_index,
221 MLXSW_REG_RICNT_OPCODE_CLEAR);
222 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
225 int mlxsw_sp_rif_counter_alloc(struct mlxsw_sp *mlxsw_sp,
226 struct mlxsw_sp_rif *rif,
227 enum mlxsw_sp_rif_counter_dir dir)
229 unsigned int *p_counter_index;
232 p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
233 if (!p_counter_index)
235 err = mlxsw_sp_counter_alloc(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
240 err = mlxsw_sp_rif_counter_clear(mlxsw_sp, *p_counter_index);
242 goto err_counter_clear;
244 err = mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
245 *p_counter_index, true, dir);
247 goto err_counter_edit;
248 mlxsw_sp_rif_counter_valid_set(rif, dir, true);
253 mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
258 void mlxsw_sp_rif_counter_free(struct mlxsw_sp *mlxsw_sp,
259 struct mlxsw_sp_rif *rif,
260 enum mlxsw_sp_rif_counter_dir dir)
262 unsigned int *p_counter_index;
264 if (!mlxsw_sp_rif_counter_valid_get(rif, dir))
267 p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
268 if (WARN_ON(!p_counter_index))
270 mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
271 *p_counter_index, false, dir);
272 mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
274 mlxsw_sp_rif_counter_valid_set(rif, dir, false);
277 static void mlxsw_sp_rif_counters_alloc(struct mlxsw_sp_rif *rif)
279 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
280 struct devlink *devlink;
282 devlink = priv_to_devlink(mlxsw_sp->core);
283 if (!devlink_dpipe_table_counter_enabled(devlink,
284 MLXSW_SP_DPIPE_TABLE_NAME_ERIF))
286 mlxsw_sp_rif_counter_alloc(mlxsw_sp, rif, MLXSW_SP_RIF_COUNTER_EGRESS);
289 static void mlxsw_sp_rif_counters_free(struct mlxsw_sp_rif *rif)
291 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
293 mlxsw_sp_rif_counter_free(mlxsw_sp, rif, MLXSW_SP_RIF_COUNTER_EGRESS);
296 #define MLXSW_SP_PREFIX_COUNT (sizeof(struct in6_addr) * BITS_PER_BYTE + 1)
298 struct mlxsw_sp_prefix_usage {
299 DECLARE_BITMAP(b, MLXSW_SP_PREFIX_COUNT);
302 #define mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) \
303 for_each_set_bit(prefix, (prefix_usage)->b, MLXSW_SP_PREFIX_COUNT)
306 mlxsw_sp_prefix_usage_eq(struct mlxsw_sp_prefix_usage *prefix_usage1,
307 struct mlxsw_sp_prefix_usage *prefix_usage2)
309 return !memcmp(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
313 mlxsw_sp_prefix_usage_cpy(struct mlxsw_sp_prefix_usage *prefix_usage1,
314 struct mlxsw_sp_prefix_usage *prefix_usage2)
316 memcpy(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
320 mlxsw_sp_prefix_usage_set(struct mlxsw_sp_prefix_usage *prefix_usage,
321 unsigned char prefix_len)
323 set_bit(prefix_len, prefix_usage->b);
327 mlxsw_sp_prefix_usage_clear(struct mlxsw_sp_prefix_usage *prefix_usage,
328 unsigned char prefix_len)
330 clear_bit(prefix_len, prefix_usage->b);
333 struct mlxsw_sp_fib_key {
334 unsigned char addr[sizeof(struct in6_addr)];
335 unsigned char prefix_len;
338 enum mlxsw_sp_fib_entry_type {
339 MLXSW_SP_FIB_ENTRY_TYPE_REMOTE,
340 MLXSW_SP_FIB_ENTRY_TYPE_LOCAL,
341 MLXSW_SP_FIB_ENTRY_TYPE_TRAP,
342 MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE,
343 MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE,
345 /* This is a special case of local delivery, where a packet should be
346 * decapsulated on reception. Note that there is no corresponding ENCAP,
347 * because that's a type of next hop, not of FIB entry. (There can be
348 * several next hops in a REMOTE entry, and some of them may be
349 * encapsulating entries.)
351 MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP,
352 MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP,
355 struct mlxsw_sp_nexthop_group_info;
356 struct mlxsw_sp_nexthop_group;
357 struct mlxsw_sp_fib_entry;
359 struct mlxsw_sp_fib_node {
360 struct mlxsw_sp_fib_entry *fib_entry;
361 struct list_head list;
362 struct rhash_head ht_node;
363 struct mlxsw_sp_fib *fib;
364 struct mlxsw_sp_fib_key key;
367 struct mlxsw_sp_fib_entry_decap {
368 struct mlxsw_sp_ipip_entry *ipip_entry;
372 static struct mlxsw_sp_fib_entry_priv *
373 mlxsw_sp_fib_entry_priv_create(const struct mlxsw_sp_router_ll_ops *ll_ops)
375 struct mlxsw_sp_fib_entry_priv *priv;
377 if (!ll_ops->fib_entry_priv_size)
378 /* No need to have priv */
381 priv = kzalloc(sizeof(*priv) + ll_ops->fib_entry_priv_size, GFP_KERNEL);
383 return ERR_PTR(-ENOMEM);
384 refcount_set(&priv->refcnt, 1);
389 mlxsw_sp_fib_entry_priv_destroy(struct mlxsw_sp_fib_entry_priv *priv)
394 static void mlxsw_sp_fib_entry_priv_hold(struct mlxsw_sp_fib_entry_priv *priv)
396 refcount_inc(&priv->refcnt);
399 static void mlxsw_sp_fib_entry_priv_put(struct mlxsw_sp_fib_entry_priv *priv)
401 if (!priv || !refcount_dec_and_test(&priv->refcnt))
403 mlxsw_sp_fib_entry_priv_destroy(priv);
406 static void mlxsw_sp_fib_entry_op_ctx_priv_hold(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
407 struct mlxsw_sp_fib_entry_priv *priv)
411 mlxsw_sp_fib_entry_priv_hold(priv);
412 list_add(&priv->list, &op_ctx->fib_entry_priv_list);
415 static void mlxsw_sp_fib_entry_op_ctx_priv_put_all(struct mlxsw_sp_fib_entry_op_ctx *op_ctx)
417 struct mlxsw_sp_fib_entry_priv *priv, *tmp;
419 list_for_each_entry_safe(priv, tmp, &op_ctx->fib_entry_priv_list, list)
420 mlxsw_sp_fib_entry_priv_put(priv);
421 INIT_LIST_HEAD(&op_ctx->fib_entry_priv_list);
424 struct mlxsw_sp_fib_entry {
425 struct mlxsw_sp_fib_node *fib_node;
426 enum mlxsw_sp_fib_entry_type type;
427 struct list_head nexthop_group_node;
428 struct mlxsw_sp_nexthop_group *nh_group;
429 struct mlxsw_sp_fib_entry_decap decap; /* Valid for decap entries. */
430 struct mlxsw_sp_fib_entry_priv *priv;
433 struct mlxsw_sp_fib4_entry {
434 struct mlxsw_sp_fib_entry common;
441 struct mlxsw_sp_fib6_entry {
442 struct mlxsw_sp_fib_entry common;
443 struct list_head rt6_list;
447 struct mlxsw_sp_rt6 {
448 struct list_head list;
449 struct fib6_info *rt;
452 struct mlxsw_sp_lpm_tree {
454 unsigned int ref_count;
455 enum mlxsw_sp_l3proto proto;
456 unsigned long prefix_ref_count[MLXSW_SP_PREFIX_COUNT];
457 struct mlxsw_sp_prefix_usage prefix_usage;
460 struct mlxsw_sp_fib {
461 struct rhashtable ht;
462 struct list_head node_list;
463 struct mlxsw_sp_vr *vr;
464 struct mlxsw_sp_lpm_tree *lpm_tree;
465 enum mlxsw_sp_l3proto proto;
466 const struct mlxsw_sp_router_ll_ops *ll_ops;
470 u16 id; /* virtual router ID */
471 u32 tb_id; /* kernel fib table id */
472 unsigned int rif_count;
473 struct mlxsw_sp_fib *fib4;
474 struct mlxsw_sp_fib *fib6;
475 struct mlxsw_sp_mr_table *mr_table[MLXSW_SP_L3_PROTO_MAX];
476 struct mlxsw_sp_rif *ul_rif;
477 refcount_t ul_rif_refcnt;
480 static int mlxsw_sp_router_ll_basic_ralta_write(struct mlxsw_sp *mlxsw_sp, char *xralta_pl)
482 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta),
483 xralta_pl + MLXSW_REG_XRALTA_RALTA_OFFSET);
486 static int mlxsw_sp_router_ll_basic_ralst_write(struct mlxsw_sp *mlxsw_sp, char *xralst_pl)
488 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst),
489 xralst_pl + MLXSW_REG_XRALST_RALST_OFFSET);
492 static int mlxsw_sp_router_ll_basic_raltb_write(struct mlxsw_sp *mlxsw_sp, char *xraltb_pl)
494 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb),
495 xraltb_pl + MLXSW_REG_XRALTB_RALTB_OFFSET);
498 static const struct rhashtable_params mlxsw_sp_fib_ht_params;
500 static struct mlxsw_sp_fib *mlxsw_sp_fib_create(struct mlxsw_sp *mlxsw_sp,
501 struct mlxsw_sp_vr *vr,
502 enum mlxsw_sp_l3proto proto)
504 const struct mlxsw_sp_router_ll_ops *ll_ops = mlxsw_sp->router->proto_ll_ops[proto];
505 struct mlxsw_sp_lpm_tree *lpm_tree;
506 struct mlxsw_sp_fib *fib;
509 lpm_tree = mlxsw_sp->router->lpm.proto_trees[proto];
510 fib = kzalloc(sizeof(*fib), GFP_KERNEL);
512 return ERR_PTR(-ENOMEM);
513 err = rhashtable_init(&fib->ht, &mlxsw_sp_fib_ht_params);
515 goto err_rhashtable_init;
516 INIT_LIST_HEAD(&fib->node_list);
519 fib->lpm_tree = lpm_tree;
520 fib->ll_ops = ll_ops;
521 mlxsw_sp_lpm_tree_hold(lpm_tree);
522 err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, lpm_tree->id);
524 goto err_lpm_tree_bind;
528 mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
534 static void mlxsw_sp_fib_destroy(struct mlxsw_sp *mlxsw_sp,
535 struct mlxsw_sp_fib *fib)
537 mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, fib);
538 mlxsw_sp_lpm_tree_put(mlxsw_sp, fib->lpm_tree);
539 WARN_ON(!list_empty(&fib->node_list));
540 rhashtable_destroy(&fib->ht);
544 static struct mlxsw_sp_lpm_tree *
545 mlxsw_sp_lpm_tree_find_unused(struct mlxsw_sp *mlxsw_sp)
547 static struct mlxsw_sp_lpm_tree *lpm_tree;
550 for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
551 lpm_tree = &mlxsw_sp->router->lpm.trees[i];
552 if (lpm_tree->ref_count == 0)
558 static int mlxsw_sp_lpm_tree_alloc(struct mlxsw_sp *mlxsw_sp,
559 const struct mlxsw_sp_router_ll_ops *ll_ops,
560 struct mlxsw_sp_lpm_tree *lpm_tree)
562 char xralta_pl[MLXSW_REG_XRALTA_LEN];
564 mlxsw_reg_xralta_pack(xralta_pl, true,
565 (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
567 return ll_ops->ralta_write(mlxsw_sp, xralta_pl);
570 static void mlxsw_sp_lpm_tree_free(struct mlxsw_sp *mlxsw_sp,
571 const struct mlxsw_sp_router_ll_ops *ll_ops,
572 struct mlxsw_sp_lpm_tree *lpm_tree)
574 char xralta_pl[MLXSW_REG_XRALTA_LEN];
576 mlxsw_reg_xralta_pack(xralta_pl, false,
577 (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
579 ll_ops->ralta_write(mlxsw_sp, xralta_pl);
583 mlxsw_sp_lpm_tree_left_struct_set(struct mlxsw_sp *mlxsw_sp,
584 const struct mlxsw_sp_router_ll_ops *ll_ops,
585 struct mlxsw_sp_prefix_usage *prefix_usage,
586 struct mlxsw_sp_lpm_tree *lpm_tree)
588 char xralst_pl[MLXSW_REG_XRALST_LEN];
591 u8 last_prefix = MLXSW_REG_RALST_BIN_NO_CHILD;
593 mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage)
596 mlxsw_reg_xralst_pack(xralst_pl, root_bin, lpm_tree->id);
597 mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) {
600 mlxsw_reg_xralst_bin_pack(xralst_pl, prefix, last_prefix,
601 MLXSW_REG_RALST_BIN_NO_CHILD);
602 last_prefix = prefix;
604 return ll_ops->ralst_write(mlxsw_sp, xralst_pl);
607 static struct mlxsw_sp_lpm_tree *
608 mlxsw_sp_lpm_tree_create(struct mlxsw_sp *mlxsw_sp,
609 const struct mlxsw_sp_router_ll_ops *ll_ops,
610 struct mlxsw_sp_prefix_usage *prefix_usage,
611 enum mlxsw_sp_l3proto proto)
613 struct mlxsw_sp_lpm_tree *lpm_tree;
616 lpm_tree = mlxsw_sp_lpm_tree_find_unused(mlxsw_sp);
618 return ERR_PTR(-EBUSY);
619 lpm_tree->proto = proto;
620 err = mlxsw_sp_lpm_tree_alloc(mlxsw_sp, ll_ops, lpm_tree);
624 err = mlxsw_sp_lpm_tree_left_struct_set(mlxsw_sp, ll_ops, prefix_usage, lpm_tree);
626 goto err_left_struct_set;
627 memcpy(&lpm_tree->prefix_usage, prefix_usage,
628 sizeof(lpm_tree->prefix_usage));
629 memset(&lpm_tree->prefix_ref_count, 0,
630 sizeof(lpm_tree->prefix_ref_count));
631 lpm_tree->ref_count = 1;
635 mlxsw_sp_lpm_tree_free(mlxsw_sp, ll_ops, lpm_tree);
639 static void mlxsw_sp_lpm_tree_destroy(struct mlxsw_sp *mlxsw_sp,
640 const struct mlxsw_sp_router_ll_ops *ll_ops,
641 struct mlxsw_sp_lpm_tree *lpm_tree)
643 mlxsw_sp_lpm_tree_free(mlxsw_sp, ll_ops, lpm_tree);
646 static struct mlxsw_sp_lpm_tree *
647 mlxsw_sp_lpm_tree_get(struct mlxsw_sp *mlxsw_sp,
648 struct mlxsw_sp_prefix_usage *prefix_usage,
649 enum mlxsw_sp_l3proto proto)
651 const struct mlxsw_sp_router_ll_ops *ll_ops = mlxsw_sp->router->proto_ll_ops[proto];
652 struct mlxsw_sp_lpm_tree *lpm_tree;
655 for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
656 lpm_tree = &mlxsw_sp->router->lpm.trees[i];
657 if (lpm_tree->ref_count != 0 &&
658 lpm_tree->proto == proto &&
659 mlxsw_sp_prefix_usage_eq(&lpm_tree->prefix_usage,
661 mlxsw_sp_lpm_tree_hold(lpm_tree);
665 return mlxsw_sp_lpm_tree_create(mlxsw_sp, ll_ops, prefix_usage, proto);
668 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree)
670 lpm_tree->ref_count++;
673 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
674 struct mlxsw_sp_lpm_tree *lpm_tree)
676 const struct mlxsw_sp_router_ll_ops *ll_ops =
677 mlxsw_sp->router->proto_ll_ops[lpm_tree->proto];
679 if (--lpm_tree->ref_count == 0)
680 mlxsw_sp_lpm_tree_destroy(mlxsw_sp, ll_ops, lpm_tree);
683 #define MLXSW_SP_LPM_TREE_MIN 1 /* tree 0 is reserved */
685 static int mlxsw_sp_lpm_init(struct mlxsw_sp *mlxsw_sp)
687 struct mlxsw_sp_prefix_usage req_prefix_usage = {{ 0 } };
688 struct mlxsw_sp_lpm_tree *lpm_tree;
692 if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_LPM_TREES))
695 max_trees = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_LPM_TREES);
696 mlxsw_sp->router->lpm.tree_count = max_trees - MLXSW_SP_LPM_TREE_MIN;
697 mlxsw_sp->router->lpm.trees = kcalloc(mlxsw_sp->router->lpm.tree_count,
698 sizeof(struct mlxsw_sp_lpm_tree),
700 if (!mlxsw_sp->router->lpm.trees)
703 for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
704 lpm_tree = &mlxsw_sp->router->lpm.trees[i];
705 lpm_tree->id = i + MLXSW_SP_LPM_TREE_MIN;
708 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
709 MLXSW_SP_L3_PROTO_IPV4);
710 if (IS_ERR(lpm_tree)) {
711 err = PTR_ERR(lpm_tree);
712 goto err_ipv4_tree_get;
714 mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4] = lpm_tree;
716 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
717 MLXSW_SP_L3_PROTO_IPV6);
718 if (IS_ERR(lpm_tree)) {
719 err = PTR_ERR(lpm_tree);
720 goto err_ipv6_tree_get;
722 mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV6] = lpm_tree;
727 lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4];
728 mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
730 kfree(mlxsw_sp->router->lpm.trees);
734 static void mlxsw_sp_lpm_fini(struct mlxsw_sp *mlxsw_sp)
736 struct mlxsw_sp_lpm_tree *lpm_tree;
738 lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV6];
739 mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
741 lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4];
742 mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
744 kfree(mlxsw_sp->router->lpm.trees);
747 static bool mlxsw_sp_vr_is_used(const struct mlxsw_sp_vr *vr)
749 return !!vr->fib4 || !!vr->fib6 ||
750 !!vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] ||
751 !!vr->mr_table[MLXSW_SP_L3_PROTO_IPV6];
754 static struct mlxsw_sp_vr *mlxsw_sp_vr_find_unused(struct mlxsw_sp *mlxsw_sp)
756 struct mlxsw_sp_vr *vr;
759 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
760 vr = &mlxsw_sp->router->vrs[i];
761 if (!mlxsw_sp_vr_is_used(vr))
767 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
768 const struct mlxsw_sp_fib *fib, u8 tree_id)
770 char xraltb_pl[MLXSW_REG_XRALTB_LEN];
772 mlxsw_reg_xraltb_pack(xraltb_pl, fib->vr->id,
773 (enum mlxsw_reg_ralxx_protocol) fib->proto,
775 return fib->ll_ops->raltb_write(mlxsw_sp, xraltb_pl);
778 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
779 const struct mlxsw_sp_fib *fib)
781 char xraltb_pl[MLXSW_REG_XRALTB_LEN];
783 /* Bind to tree 0 which is default */
784 mlxsw_reg_xraltb_pack(xraltb_pl, fib->vr->id,
785 (enum mlxsw_reg_ralxx_protocol) fib->proto, 0);
786 return fib->ll_ops->raltb_write(mlxsw_sp, xraltb_pl);
789 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
791 /* For our purpose, squash main, default and local tables into one */
792 if (tb_id == RT_TABLE_LOCAL || tb_id == RT_TABLE_DEFAULT)
793 tb_id = RT_TABLE_MAIN;
797 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
800 struct mlxsw_sp_vr *vr;
803 tb_id = mlxsw_sp_fix_tb_id(tb_id);
805 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
806 vr = &mlxsw_sp->router->vrs[i];
807 if (mlxsw_sp_vr_is_used(vr) && vr->tb_id == tb_id)
813 int mlxsw_sp_router_tb_id_vr_id(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
816 struct mlxsw_sp_vr *vr;
819 mutex_lock(&mlxsw_sp->router->lock);
820 vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
827 mutex_unlock(&mlxsw_sp->router->lock);
831 static struct mlxsw_sp_fib *mlxsw_sp_vr_fib(const struct mlxsw_sp_vr *vr,
832 enum mlxsw_sp_l3proto proto)
835 case MLXSW_SP_L3_PROTO_IPV4:
837 case MLXSW_SP_L3_PROTO_IPV6:
843 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
845 struct netlink_ext_ack *extack)
847 struct mlxsw_sp_mr_table *mr4_table, *mr6_table;
848 struct mlxsw_sp_fib *fib4;
849 struct mlxsw_sp_fib *fib6;
850 struct mlxsw_sp_vr *vr;
853 vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
855 NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported virtual routers");
856 return ERR_PTR(-EBUSY);
858 fib4 = mlxsw_sp_fib_create(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
860 return ERR_CAST(fib4);
861 fib6 = mlxsw_sp_fib_create(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
864 goto err_fib6_create;
866 mr4_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
867 MLXSW_SP_L3_PROTO_IPV4);
868 if (IS_ERR(mr4_table)) {
869 err = PTR_ERR(mr4_table);
870 goto err_mr4_table_create;
872 mr6_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
873 MLXSW_SP_L3_PROTO_IPV6);
874 if (IS_ERR(mr6_table)) {
875 err = PTR_ERR(mr6_table);
876 goto err_mr6_table_create;
881 vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] = mr4_table;
882 vr->mr_table[MLXSW_SP_L3_PROTO_IPV6] = mr6_table;
886 err_mr6_table_create:
887 mlxsw_sp_mr_table_destroy(mr4_table);
888 err_mr4_table_create:
889 mlxsw_sp_fib_destroy(mlxsw_sp, fib6);
891 mlxsw_sp_fib_destroy(mlxsw_sp, fib4);
895 static void mlxsw_sp_vr_destroy(struct mlxsw_sp *mlxsw_sp,
896 struct mlxsw_sp_vr *vr)
898 mlxsw_sp_mr_table_destroy(vr->mr_table[MLXSW_SP_L3_PROTO_IPV6]);
899 vr->mr_table[MLXSW_SP_L3_PROTO_IPV6] = NULL;
900 mlxsw_sp_mr_table_destroy(vr->mr_table[MLXSW_SP_L3_PROTO_IPV4]);
901 vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] = NULL;
902 mlxsw_sp_fib_destroy(mlxsw_sp, vr->fib6);
904 mlxsw_sp_fib_destroy(mlxsw_sp, vr->fib4);
908 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
909 struct netlink_ext_ack *extack)
911 struct mlxsw_sp_vr *vr;
913 tb_id = mlxsw_sp_fix_tb_id(tb_id);
914 vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
916 vr = mlxsw_sp_vr_create(mlxsw_sp, tb_id, extack);
920 static void mlxsw_sp_vr_put(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr)
922 if (!vr->rif_count && list_empty(&vr->fib4->node_list) &&
923 list_empty(&vr->fib6->node_list) &&
924 mlxsw_sp_mr_table_empty(vr->mr_table[MLXSW_SP_L3_PROTO_IPV4]) &&
925 mlxsw_sp_mr_table_empty(vr->mr_table[MLXSW_SP_L3_PROTO_IPV6]))
926 mlxsw_sp_vr_destroy(mlxsw_sp, vr);
930 mlxsw_sp_vr_lpm_tree_should_replace(struct mlxsw_sp_vr *vr,
931 enum mlxsw_sp_l3proto proto, u8 tree_id)
933 struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
935 if (!mlxsw_sp_vr_is_used(vr))
937 if (fib->lpm_tree->id == tree_id)
942 static int mlxsw_sp_vr_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
943 struct mlxsw_sp_fib *fib,
944 struct mlxsw_sp_lpm_tree *new_tree)
946 struct mlxsw_sp_lpm_tree *old_tree = fib->lpm_tree;
949 fib->lpm_tree = new_tree;
950 mlxsw_sp_lpm_tree_hold(new_tree);
951 err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, new_tree->id);
954 mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
958 mlxsw_sp_lpm_tree_put(mlxsw_sp, new_tree);
959 fib->lpm_tree = old_tree;
963 static int mlxsw_sp_vrs_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
964 struct mlxsw_sp_fib *fib,
965 struct mlxsw_sp_lpm_tree *new_tree)
967 enum mlxsw_sp_l3proto proto = fib->proto;
968 struct mlxsw_sp_lpm_tree *old_tree;
969 u8 old_id, new_id = new_tree->id;
970 struct mlxsw_sp_vr *vr;
973 old_tree = mlxsw_sp->router->lpm.proto_trees[proto];
974 old_id = old_tree->id;
976 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
977 vr = &mlxsw_sp->router->vrs[i];
978 if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, old_id))
980 err = mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
981 mlxsw_sp_vr_fib(vr, proto),
984 goto err_tree_replace;
987 memcpy(new_tree->prefix_ref_count, old_tree->prefix_ref_count,
988 sizeof(new_tree->prefix_ref_count));
989 mlxsw_sp->router->lpm.proto_trees[proto] = new_tree;
990 mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
995 for (i--; i >= 0; i--) {
996 if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, new_id))
998 mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
999 mlxsw_sp_vr_fib(vr, proto),
1005 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
1007 struct mlxsw_sp_vr *vr;
1011 if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_VRS))
1014 max_vrs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS);
1015 mlxsw_sp->router->vrs = kcalloc(max_vrs, sizeof(struct mlxsw_sp_vr),
1017 if (!mlxsw_sp->router->vrs)
1020 for (i = 0; i < max_vrs; i++) {
1021 vr = &mlxsw_sp->router->vrs[i];
1028 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
1030 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
1032 /* At this stage we're guaranteed not to have new incoming
1033 * FIB notifications and the work queue is free from FIBs
1034 * sitting on top of mlxsw netdevs. However, we can still
1035 * have other FIBs queued. Flush the queue before flushing
1036 * the device's tables. No need for locks, as we're the only
1039 mlxsw_core_flush_owq();
1040 mlxsw_sp_router_fib_flush(mlxsw_sp);
1041 kfree(mlxsw_sp->router->vrs);
1044 static struct net_device *
1045 __mlxsw_sp_ipip_netdev_ul_dev_get(const struct net_device *ol_dev)
1047 struct ip_tunnel *tun = netdev_priv(ol_dev);
1048 struct net *net = dev_net(ol_dev);
1050 return dev_get_by_index_rcu(net, tun->parms.link);
1053 u32 mlxsw_sp_ipip_dev_ul_tb_id(const struct net_device *ol_dev)
1055 struct net_device *d;
1059 d = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1061 tb_id = l3mdev_fib_table(d) ? : RT_TABLE_MAIN;
1063 tb_id = RT_TABLE_MAIN;
1069 static struct mlxsw_sp_rif *
1070 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
1071 const struct mlxsw_sp_rif_params *params,
1072 struct netlink_ext_ack *extack);
1074 static struct mlxsw_sp_rif_ipip_lb *
1075 mlxsw_sp_ipip_ol_ipip_lb_create(struct mlxsw_sp *mlxsw_sp,
1076 enum mlxsw_sp_ipip_type ipipt,
1077 struct net_device *ol_dev,
1078 struct netlink_ext_ack *extack)
1080 struct mlxsw_sp_rif_params_ipip_lb lb_params;
1081 const struct mlxsw_sp_ipip_ops *ipip_ops;
1082 struct mlxsw_sp_rif *rif;
1084 ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1085 lb_params = (struct mlxsw_sp_rif_params_ipip_lb) {
1086 .common.dev = ol_dev,
1087 .common.lag = false,
1088 .lb_config = ipip_ops->ol_loopback_config(mlxsw_sp, ol_dev),
1091 rif = mlxsw_sp_rif_create(mlxsw_sp, &lb_params.common, extack);
1093 return ERR_CAST(rif);
1094 return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
1097 static struct mlxsw_sp_ipip_entry *
1098 mlxsw_sp_ipip_entry_alloc(struct mlxsw_sp *mlxsw_sp,
1099 enum mlxsw_sp_ipip_type ipipt,
1100 struct net_device *ol_dev)
1102 const struct mlxsw_sp_ipip_ops *ipip_ops;
1103 struct mlxsw_sp_ipip_entry *ipip_entry;
1104 struct mlxsw_sp_ipip_entry *ret = NULL;
1106 ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1107 ipip_entry = kzalloc(sizeof(*ipip_entry), GFP_KERNEL);
1109 return ERR_PTR(-ENOMEM);
1111 ipip_entry->ol_lb = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp, ipipt,
1113 if (IS_ERR(ipip_entry->ol_lb)) {
1114 ret = ERR_CAST(ipip_entry->ol_lb);
1115 goto err_ol_ipip_lb_create;
1118 ipip_entry->ipipt = ipipt;
1119 ipip_entry->ol_dev = ol_dev;
1121 switch (ipip_ops->ul_proto) {
1122 case MLXSW_SP_L3_PROTO_IPV4:
1123 ipip_entry->parms4 = mlxsw_sp_ipip_netdev_parms4(ol_dev);
1125 case MLXSW_SP_L3_PROTO_IPV6:
1132 err_ol_ipip_lb_create:
1138 mlxsw_sp_ipip_entry_dealloc(struct mlxsw_sp_ipip_entry *ipip_entry)
1140 mlxsw_sp_rif_destroy(&ipip_entry->ol_lb->common);
1145 mlxsw_sp_ipip_entry_saddr_matches(struct mlxsw_sp *mlxsw_sp,
1146 const enum mlxsw_sp_l3proto ul_proto,
1147 union mlxsw_sp_l3addr saddr,
1149 struct mlxsw_sp_ipip_entry *ipip_entry)
1151 u32 tun_ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1152 enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1153 union mlxsw_sp_l3addr tun_saddr;
1155 if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1158 tun_saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1159 return tun_ul_tb_id == ul_tb_id &&
1160 mlxsw_sp_l3addr_eq(&tun_saddr, &saddr);
1164 mlxsw_sp_fib_entry_decap_init(struct mlxsw_sp *mlxsw_sp,
1165 struct mlxsw_sp_fib_entry *fib_entry,
1166 struct mlxsw_sp_ipip_entry *ipip_entry)
1171 err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
1176 ipip_entry->decap_fib_entry = fib_entry;
1177 fib_entry->decap.ipip_entry = ipip_entry;
1178 fib_entry->decap.tunnel_index = tunnel_index;
1182 static void mlxsw_sp_fib_entry_decap_fini(struct mlxsw_sp *mlxsw_sp,
1183 struct mlxsw_sp_fib_entry *fib_entry)
1185 /* Unlink this node from the IPIP entry that it's the decap entry of. */
1186 fib_entry->decap.ipip_entry->decap_fib_entry = NULL;
1187 fib_entry->decap.ipip_entry = NULL;
1188 mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
1189 1, fib_entry->decap.tunnel_index);
1192 static struct mlxsw_sp_fib_node *
1193 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
1194 size_t addr_len, unsigned char prefix_len);
1195 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1196 struct mlxsw_sp_fib_entry *fib_entry);
1199 mlxsw_sp_ipip_entry_demote_decap(struct mlxsw_sp *mlxsw_sp,
1200 struct mlxsw_sp_ipip_entry *ipip_entry)
1202 struct mlxsw_sp_fib_entry *fib_entry = ipip_entry->decap_fib_entry;
1204 mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
1205 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1207 mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1211 mlxsw_sp_ipip_entry_promote_decap(struct mlxsw_sp *mlxsw_sp,
1212 struct mlxsw_sp_ipip_entry *ipip_entry,
1213 struct mlxsw_sp_fib_entry *decap_fib_entry)
1215 if (mlxsw_sp_fib_entry_decap_init(mlxsw_sp, decap_fib_entry,
1218 decap_fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
1220 if (mlxsw_sp_fib_entry_update(mlxsw_sp, decap_fib_entry))
1221 mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1224 static struct mlxsw_sp_fib_entry *
1225 mlxsw_sp_router_ip2me_fib_entry_find(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
1226 enum mlxsw_sp_l3proto proto,
1227 const union mlxsw_sp_l3addr *addr,
1228 enum mlxsw_sp_fib_entry_type type)
1230 struct mlxsw_sp_fib_node *fib_node;
1231 unsigned char addr_prefix_len;
1232 struct mlxsw_sp_fib *fib;
1233 struct mlxsw_sp_vr *vr;
1238 vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
1241 fib = mlxsw_sp_vr_fib(vr, proto);
1244 case MLXSW_SP_L3_PROTO_IPV4:
1245 addr4 = be32_to_cpu(addr->addr4);
1248 addr_prefix_len = 32;
1250 case MLXSW_SP_L3_PROTO_IPV6:
1256 fib_node = mlxsw_sp_fib_node_lookup(fib, addrp, addr_len,
1258 if (!fib_node || fib_node->fib_entry->type != type)
1261 return fib_node->fib_entry;
1264 /* Given an IPIP entry, find the corresponding decap route. */
1265 static struct mlxsw_sp_fib_entry *
1266 mlxsw_sp_ipip_entry_find_decap(struct mlxsw_sp *mlxsw_sp,
1267 struct mlxsw_sp_ipip_entry *ipip_entry)
1269 static struct mlxsw_sp_fib_node *fib_node;
1270 const struct mlxsw_sp_ipip_ops *ipip_ops;
1271 unsigned char saddr_prefix_len;
1272 union mlxsw_sp_l3addr saddr;
1273 struct mlxsw_sp_fib *ul_fib;
1274 struct mlxsw_sp_vr *ul_vr;
1280 ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1282 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1283 ul_vr = mlxsw_sp_vr_find(mlxsw_sp, ul_tb_id);
1287 ul_fib = mlxsw_sp_vr_fib(ul_vr, ipip_ops->ul_proto);
1288 saddr = mlxsw_sp_ipip_netdev_saddr(ipip_ops->ul_proto,
1289 ipip_entry->ol_dev);
1291 switch (ipip_ops->ul_proto) {
1292 case MLXSW_SP_L3_PROTO_IPV4:
1293 saddr4 = be32_to_cpu(saddr.addr4);
1296 saddr_prefix_len = 32;
1303 fib_node = mlxsw_sp_fib_node_lookup(ul_fib, saddrp, saddr_len,
1306 fib_node->fib_entry->type != MLXSW_SP_FIB_ENTRY_TYPE_TRAP)
1309 return fib_node->fib_entry;
1312 static struct mlxsw_sp_ipip_entry *
1313 mlxsw_sp_ipip_entry_create(struct mlxsw_sp *mlxsw_sp,
1314 enum mlxsw_sp_ipip_type ipipt,
1315 struct net_device *ol_dev)
1317 struct mlxsw_sp_ipip_entry *ipip_entry;
1319 ipip_entry = mlxsw_sp_ipip_entry_alloc(mlxsw_sp, ipipt, ol_dev);
1320 if (IS_ERR(ipip_entry))
1323 list_add_tail(&ipip_entry->ipip_list_node,
1324 &mlxsw_sp->router->ipip_list);
1330 mlxsw_sp_ipip_entry_destroy(struct mlxsw_sp *mlxsw_sp,
1331 struct mlxsw_sp_ipip_entry *ipip_entry)
1333 list_del(&ipip_entry->ipip_list_node);
1334 mlxsw_sp_ipip_entry_dealloc(ipip_entry);
1338 mlxsw_sp_ipip_entry_matches_decap(struct mlxsw_sp *mlxsw_sp,
1339 const struct net_device *ul_dev,
1340 enum mlxsw_sp_l3proto ul_proto,
1341 union mlxsw_sp_l3addr ul_dip,
1342 struct mlxsw_sp_ipip_entry *ipip_entry)
1344 u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1345 enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1347 if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1350 return mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, ul_dip,
1351 ul_tb_id, ipip_entry);
1354 /* Given decap parameters, find the corresponding IPIP entry. */
1355 static struct mlxsw_sp_ipip_entry *
1356 mlxsw_sp_ipip_entry_find_by_decap(struct mlxsw_sp *mlxsw_sp, int ul_dev_ifindex,
1357 enum mlxsw_sp_l3proto ul_proto,
1358 union mlxsw_sp_l3addr ul_dip)
1360 struct mlxsw_sp_ipip_entry *ipip_entry = NULL;
1361 struct net_device *ul_dev;
1365 ul_dev = dev_get_by_index_rcu(mlxsw_sp_net(mlxsw_sp), ul_dev_ifindex);
1369 list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1371 if (mlxsw_sp_ipip_entry_matches_decap(mlxsw_sp, ul_dev,
1385 static bool mlxsw_sp_netdev_ipip_type(const struct mlxsw_sp *mlxsw_sp,
1386 const struct net_device *dev,
1387 enum mlxsw_sp_ipip_type *p_type)
1389 struct mlxsw_sp_router *router = mlxsw_sp->router;
1390 const struct mlxsw_sp_ipip_ops *ipip_ops;
1391 enum mlxsw_sp_ipip_type ipipt;
1393 for (ipipt = 0; ipipt < MLXSW_SP_IPIP_TYPE_MAX; ++ipipt) {
1394 ipip_ops = router->ipip_ops_arr[ipipt];
1395 if (dev->type == ipip_ops->dev_type) {
1404 bool mlxsw_sp_netdev_is_ipip_ol(const struct mlxsw_sp *mlxsw_sp,
1405 const struct net_device *dev)
1407 return mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL);
1410 static struct mlxsw_sp_ipip_entry *
1411 mlxsw_sp_ipip_entry_find_by_ol_dev(struct mlxsw_sp *mlxsw_sp,
1412 const struct net_device *ol_dev)
1414 struct mlxsw_sp_ipip_entry *ipip_entry;
1416 list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1418 if (ipip_entry->ol_dev == ol_dev)
1424 static struct mlxsw_sp_ipip_entry *
1425 mlxsw_sp_ipip_entry_find_by_ul_dev(const struct mlxsw_sp *mlxsw_sp,
1426 const struct net_device *ul_dev,
1427 struct mlxsw_sp_ipip_entry *start)
1429 struct mlxsw_sp_ipip_entry *ipip_entry;
1431 ipip_entry = list_prepare_entry(start, &mlxsw_sp->router->ipip_list,
1433 list_for_each_entry_continue(ipip_entry, &mlxsw_sp->router->ipip_list,
1435 struct net_device *ol_dev = ipip_entry->ol_dev;
1436 struct net_device *ipip_ul_dev;
1439 ipip_ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1442 if (ipip_ul_dev == ul_dev)
1449 bool mlxsw_sp_netdev_is_ipip_ul(struct mlxsw_sp *mlxsw_sp,
1450 const struct net_device *dev)
1454 mutex_lock(&mlxsw_sp->router->lock);
1455 is_ipip_ul = mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp, dev, NULL);
1456 mutex_unlock(&mlxsw_sp->router->lock);
1461 static bool mlxsw_sp_netdevice_ipip_can_offload(struct mlxsw_sp *mlxsw_sp,
1462 const struct net_device *ol_dev,
1463 enum mlxsw_sp_ipip_type ipipt)
1465 const struct mlxsw_sp_ipip_ops *ops
1466 = mlxsw_sp->router->ipip_ops_arr[ipipt];
1468 return ops->can_offload(mlxsw_sp, ol_dev);
1471 static int mlxsw_sp_netdevice_ipip_ol_reg_event(struct mlxsw_sp *mlxsw_sp,
1472 struct net_device *ol_dev)
1474 enum mlxsw_sp_ipip_type ipipt = MLXSW_SP_IPIP_TYPE_MAX;
1475 struct mlxsw_sp_ipip_entry *ipip_entry;
1476 enum mlxsw_sp_l3proto ul_proto;
1477 union mlxsw_sp_l3addr saddr;
1480 mlxsw_sp_netdev_ipip_type(mlxsw_sp, ol_dev, &ipipt);
1481 if (mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev, ipipt)) {
1482 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ol_dev);
1483 ul_proto = mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto;
1484 saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ol_dev);
1485 if (!mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1488 ipip_entry = mlxsw_sp_ipip_entry_create(mlxsw_sp, ipipt,
1490 if (IS_ERR(ipip_entry))
1491 return PTR_ERR(ipip_entry);
1498 static void mlxsw_sp_netdevice_ipip_ol_unreg_event(struct mlxsw_sp *mlxsw_sp,
1499 struct net_device *ol_dev)
1501 struct mlxsw_sp_ipip_entry *ipip_entry;
1503 ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1505 mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1509 mlxsw_sp_ipip_entry_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1510 struct mlxsw_sp_ipip_entry *ipip_entry)
1512 struct mlxsw_sp_fib_entry *decap_fib_entry;
1514 decap_fib_entry = mlxsw_sp_ipip_entry_find_decap(mlxsw_sp, ipip_entry);
1515 if (decap_fib_entry)
1516 mlxsw_sp_ipip_entry_promote_decap(mlxsw_sp, ipip_entry,
1521 mlxsw_sp_rif_ipip_lb_op(struct mlxsw_sp_rif_ipip_lb *lb_rif, u16 ul_vr_id,
1522 u16 ul_rif_id, bool enable)
1524 struct mlxsw_sp_rif_ipip_lb_config lb_cf = lb_rif->lb_config;
1525 struct mlxsw_sp_rif *rif = &lb_rif->common;
1526 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
1527 char ritr_pl[MLXSW_REG_RITR_LEN];
1530 switch (lb_cf.ul_protocol) {
1531 case MLXSW_SP_L3_PROTO_IPV4:
1532 saddr4 = be32_to_cpu(lb_cf.saddr.addr4);
1533 mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
1534 rif->rif_index, rif->vr_id, rif->dev->mtu);
1535 mlxsw_reg_ritr_loopback_ipip4_pack(ritr_pl, lb_cf.lb_ipipt,
1536 MLXSW_REG_RITR_LOOPBACK_IPIP_OPTIONS_GRE_KEY_PRESET,
1537 ul_vr_id, ul_rif_id, saddr4, lb_cf.okey);
1540 case MLXSW_SP_L3_PROTO_IPV6:
1541 return -EAFNOSUPPORT;
1544 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
1547 static int mlxsw_sp_netdevice_ipip_ol_update_mtu(struct mlxsw_sp *mlxsw_sp,
1548 struct net_device *ol_dev)
1550 struct mlxsw_sp_ipip_entry *ipip_entry;
1551 struct mlxsw_sp_rif_ipip_lb *lb_rif;
1554 ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1556 lb_rif = ipip_entry->ol_lb;
1557 err = mlxsw_sp_rif_ipip_lb_op(lb_rif, lb_rif->ul_vr_id,
1558 lb_rif->ul_rif_id, true);
1561 lb_rif->common.mtu = ol_dev->mtu;
1568 static void mlxsw_sp_netdevice_ipip_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1569 struct net_device *ol_dev)
1571 struct mlxsw_sp_ipip_entry *ipip_entry;
1573 ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1575 mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1579 mlxsw_sp_ipip_entry_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1580 struct mlxsw_sp_ipip_entry *ipip_entry)
1582 if (ipip_entry->decap_fib_entry)
1583 mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1586 static void mlxsw_sp_netdevice_ipip_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1587 struct net_device *ol_dev)
1589 struct mlxsw_sp_ipip_entry *ipip_entry;
1591 ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1593 mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1596 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
1597 struct mlxsw_sp_rif *old_rif,
1598 struct mlxsw_sp_rif *new_rif);
1600 mlxsw_sp_ipip_entry_ol_lb_update(struct mlxsw_sp *mlxsw_sp,
1601 struct mlxsw_sp_ipip_entry *ipip_entry,
1603 struct netlink_ext_ack *extack)
1605 struct mlxsw_sp_rif_ipip_lb *old_lb_rif = ipip_entry->ol_lb;
1606 struct mlxsw_sp_rif_ipip_lb *new_lb_rif;
1608 new_lb_rif = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp,
1612 if (IS_ERR(new_lb_rif))
1613 return PTR_ERR(new_lb_rif);
1614 ipip_entry->ol_lb = new_lb_rif;
1617 mlxsw_sp_nexthop_rif_migrate(mlxsw_sp, &old_lb_rif->common,
1618 &new_lb_rif->common);
1620 mlxsw_sp_rif_destroy(&old_lb_rif->common);
1625 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
1626 struct mlxsw_sp_rif *rif);
1629 * __mlxsw_sp_ipip_entry_update_tunnel - Update offload related to IPIP entry.
1630 * @mlxsw_sp: mlxsw_sp.
1631 * @ipip_entry: IPIP entry.
1632 * @recreate_loopback: Recreates the associated loopback RIF.
1633 * @keep_encap: Updates next hops that use the tunnel netdevice. This is only
1634 * relevant when recreate_loopback is true.
1635 * @update_nexthops: Updates next hops, keeping the current loopback RIF. This
1636 * is only relevant when recreate_loopback is false.
1639 * Return: Non-zero value on failure.
1641 int __mlxsw_sp_ipip_entry_update_tunnel(struct mlxsw_sp *mlxsw_sp,
1642 struct mlxsw_sp_ipip_entry *ipip_entry,
1643 bool recreate_loopback,
1645 bool update_nexthops,
1646 struct netlink_ext_ack *extack)
1650 /* RIFs can't be edited, so to update loopback, we need to destroy and
1651 * recreate it. That creates a window of opportunity where RALUE and
1652 * RATR registers end up referencing a RIF that's already gone. RATRs
1653 * are handled in mlxsw_sp_ipip_entry_ol_lb_update(), and to take care
1654 * of RALUE, demote the decap route back.
1656 if (ipip_entry->decap_fib_entry)
1657 mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1659 if (recreate_loopback) {
1660 err = mlxsw_sp_ipip_entry_ol_lb_update(mlxsw_sp, ipip_entry,
1661 keep_encap, extack);
1664 } else if (update_nexthops) {
1665 mlxsw_sp_nexthop_rif_update(mlxsw_sp,
1666 &ipip_entry->ol_lb->common);
1669 if (ipip_entry->ol_dev->flags & IFF_UP)
1670 mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1675 static int mlxsw_sp_netdevice_ipip_ol_vrf_event(struct mlxsw_sp *mlxsw_sp,
1676 struct net_device *ol_dev,
1677 struct netlink_ext_ack *extack)
1679 struct mlxsw_sp_ipip_entry *ipip_entry =
1680 mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1685 return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1686 true, false, false, extack);
1690 mlxsw_sp_netdevice_ipip_ul_vrf_event(struct mlxsw_sp *mlxsw_sp,
1691 struct mlxsw_sp_ipip_entry *ipip_entry,
1692 struct net_device *ul_dev,
1694 struct netlink_ext_ack *extack)
1696 u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1697 enum mlxsw_sp_l3proto ul_proto;
1698 union mlxsw_sp_l3addr saddr;
1700 /* Moving underlay to a different VRF might cause local address
1701 * conflict, and the conflicting tunnels need to be demoted.
1703 ul_proto = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt]->ul_proto;
1704 saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1705 if (mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1708 *demote_this = true;
1712 return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1713 true, true, false, extack);
1717 mlxsw_sp_netdevice_ipip_ul_up_event(struct mlxsw_sp *mlxsw_sp,
1718 struct mlxsw_sp_ipip_entry *ipip_entry,
1719 struct net_device *ul_dev)
1721 return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1722 false, false, true, NULL);
1726 mlxsw_sp_netdevice_ipip_ul_down_event(struct mlxsw_sp *mlxsw_sp,
1727 struct mlxsw_sp_ipip_entry *ipip_entry,
1728 struct net_device *ul_dev)
1730 /* A down underlay device causes encapsulated packets to not be
1731 * forwarded, but decap still works. So refresh next hops without
1732 * touching anything else.
1734 return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1735 false, false, true, NULL);
1739 mlxsw_sp_netdevice_ipip_ol_change_event(struct mlxsw_sp *mlxsw_sp,
1740 struct net_device *ol_dev,
1741 struct netlink_ext_ack *extack)
1743 const struct mlxsw_sp_ipip_ops *ipip_ops;
1744 struct mlxsw_sp_ipip_entry *ipip_entry;
1747 ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1749 /* A change might make a tunnel eligible for offloading, but
1750 * that is currently not implemented. What falls to slow path
1755 /* A change might make a tunnel not eligible for offloading. */
1756 if (!mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev,
1757 ipip_entry->ipipt)) {
1758 mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1762 ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1763 err = ipip_ops->ol_netdev_change(mlxsw_sp, ipip_entry, extack);
1767 void mlxsw_sp_ipip_entry_demote_tunnel(struct mlxsw_sp *mlxsw_sp,
1768 struct mlxsw_sp_ipip_entry *ipip_entry)
1770 struct net_device *ol_dev = ipip_entry->ol_dev;
1772 if (ol_dev->flags & IFF_UP)
1773 mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1774 mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1777 /* The configuration where several tunnels have the same local address in the
1778 * same underlay table needs special treatment in the HW. That is currently not
1779 * implemented in the driver. This function finds and demotes the first tunnel
1780 * with a given source address, except the one passed in in the argument
1784 mlxsw_sp_ipip_demote_tunnel_by_saddr(struct mlxsw_sp *mlxsw_sp,
1785 enum mlxsw_sp_l3proto ul_proto,
1786 union mlxsw_sp_l3addr saddr,
1788 const struct mlxsw_sp_ipip_entry *except)
1790 struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1792 list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1794 if (ipip_entry != except &&
1795 mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, saddr,
1796 ul_tb_id, ipip_entry)) {
1797 mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1805 static void mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(struct mlxsw_sp *mlxsw_sp,
1806 struct net_device *ul_dev)
1808 struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1810 list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1812 struct net_device *ol_dev = ipip_entry->ol_dev;
1813 struct net_device *ipip_ul_dev;
1816 ipip_ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1818 if (ipip_ul_dev == ul_dev)
1819 mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1823 int mlxsw_sp_netdevice_ipip_ol_event(struct mlxsw_sp *mlxsw_sp,
1824 struct net_device *ol_dev,
1825 unsigned long event,
1826 struct netdev_notifier_info *info)
1828 struct netdev_notifier_changeupper_info *chup;
1829 struct netlink_ext_ack *extack;
1832 mutex_lock(&mlxsw_sp->router->lock);
1834 case NETDEV_REGISTER:
1835 err = mlxsw_sp_netdevice_ipip_ol_reg_event(mlxsw_sp, ol_dev);
1837 case NETDEV_UNREGISTER:
1838 mlxsw_sp_netdevice_ipip_ol_unreg_event(mlxsw_sp, ol_dev);
1841 mlxsw_sp_netdevice_ipip_ol_up_event(mlxsw_sp, ol_dev);
1844 mlxsw_sp_netdevice_ipip_ol_down_event(mlxsw_sp, ol_dev);
1846 case NETDEV_CHANGEUPPER:
1847 chup = container_of(info, typeof(*chup), info);
1848 extack = info->extack;
1849 if (netif_is_l3_master(chup->upper_dev))
1850 err = mlxsw_sp_netdevice_ipip_ol_vrf_event(mlxsw_sp,
1855 extack = info->extack;
1856 err = mlxsw_sp_netdevice_ipip_ol_change_event(mlxsw_sp,
1859 case NETDEV_CHANGEMTU:
1860 err = mlxsw_sp_netdevice_ipip_ol_update_mtu(mlxsw_sp, ol_dev);
1863 mutex_unlock(&mlxsw_sp->router->lock);
1868 __mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1869 struct mlxsw_sp_ipip_entry *ipip_entry,
1870 struct net_device *ul_dev,
1872 unsigned long event,
1873 struct netdev_notifier_info *info)
1875 struct netdev_notifier_changeupper_info *chup;
1876 struct netlink_ext_ack *extack;
1879 case NETDEV_CHANGEUPPER:
1880 chup = container_of(info, typeof(*chup), info);
1881 extack = info->extack;
1882 if (netif_is_l3_master(chup->upper_dev))
1883 return mlxsw_sp_netdevice_ipip_ul_vrf_event(mlxsw_sp,
1891 return mlxsw_sp_netdevice_ipip_ul_up_event(mlxsw_sp, ipip_entry,
1894 return mlxsw_sp_netdevice_ipip_ul_down_event(mlxsw_sp,
1902 mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1903 struct net_device *ul_dev,
1904 unsigned long event,
1905 struct netdev_notifier_info *info)
1907 struct mlxsw_sp_ipip_entry *ipip_entry = NULL;
1910 mutex_lock(&mlxsw_sp->router->lock);
1911 while ((ipip_entry = mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp,
1914 struct mlxsw_sp_ipip_entry *prev;
1915 bool demote_this = false;
1917 err = __mlxsw_sp_netdevice_ipip_ul_event(mlxsw_sp, ipip_entry,
1918 ul_dev, &demote_this,
1921 mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(mlxsw_sp,
1927 if (list_is_first(&ipip_entry->ipip_list_node,
1928 &mlxsw_sp->router->ipip_list))
1931 /* This can't be cached from previous iteration,
1932 * because that entry could be gone now.
1934 prev = list_prev_entry(ipip_entry,
1936 mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1940 mutex_unlock(&mlxsw_sp->router->lock);
1945 int mlxsw_sp_router_nve_promote_decap(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
1946 enum mlxsw_sp_l3proto ul_proto,
1947 const union mlxsw_sp_l3addr *ul_sip,
1950 enum mlxsw_sp_fib_entry_type type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1951 struct mlxsw_sp_router *router = mlxsw_sp->router;
1952 struct mlxsw_sp_fib_entry *fib_entry;
1955 mutex_lock(&mlxsw_sp->router->lock);
1957 if (WARN_ON_ONCE(router->nve_decap_config.valid)) {
1962 router->nve_decap_config.ul_tb_id = ul_tb_id;
1963 router->nve_decap_config.tunnel_index = tunnel_index;
1964 router->nve_decap_config.ul_proto = ul_proto;
1965 router->nve_decap_config.ul_sip = *ul_sip;
1966 router->nve_decap_config.valid = true;
1968 /* It is valid to create a tunnel with a local IP and only later
1969 * assign this IP address to a local interface
1971 fib_entry = mlxsw_sp_router_ip2me_fib_entry_find(mlxsw_sp, ul_tb_id,
1977 fib_entry->decap.tunnel_index = tunnel_index;
1978 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
1980 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1982 goto err_fib_entry_update;
1986 err_fib_entry_update:
1987 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1988 mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1990 mutex_unlock(&mlxsw_sp->router->lock);
1994 void mlxsw_sp_router_nve_demote_decap(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
1995 enum mlxsw_sp_l3proto ul_proto,
1996 const union mlxsw_sp_l3addr *ul_sip)
1998 enum mlxsw_sp_fib_entry_type type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
1999 struct mlxsw_sp_router *router = mlxsw_sp->router;
2000 struct mlxsw_sp_fib_entry *fib_entry;
2002 mutex_lock(&mlxsw_sp->router->lock);
2004 if (WARN_ON_ONCE(!router->nve_decap_config.valid))
2007 router->nve_decap_config.valid = false;
2009 fib_entry = mlxsw_sp_router_ip2me_fib_entry_find(mlxsw_sp, ul_tb_id,
2015 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
2016 mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
2018 mutex_unlock(&mlxsw_sp->router->lock);
2021 static bool mlxsw_sp_router_nve_is_decap(struct mlxsw_sp *mlxsw_sp,
2023 enum mlxsw_sp_l3proto ul_proto,
2024 const union mlxsw_sp_l3addr *ul_sip)
2026 struct mlxsw_sp_router *router = mlxsw_sp->router;
2028 return router->nve_decap_config.valid &&
2029 router->nve_decap_config.ul_tb_id == ul_tb_id &&
2030 router->nve_decap_config.ul_proto == ul_proto &&
2031 !memcmp(&router->nve_decap_config.ul_sip, ul_sip,
2035 struct mlxsw_sp_neigh_key {
2036 struct neighbour *n;
2039 struct mlxsw_sp_neigh_entry {
2040 struct list_head rif_list_node;
2041 struct rhash_head ht_node;
2042 struct mlxsw_sp_neigh_key key;
2045 unsigned char ha[ETH_ALEN];
2046 struct list_head nexthop_list; /* list of nexthops using
2049 struct list_head nexthop_neighs_list_node;
2050 unsigned int counter_index;
2054 static const struct rhashtable_params mlxsw_sp_neigh_ht_params = {
2055 .key_offset = offsetof(struct mlxsw_sp_neigh_entry, key),
2056 .head_offset = offsetof(struct mlxsw_sp_neigh_entry, ht_node),
2057 .key_len = sizeof(struct mlxsw_sp_neigh_key),
2060 struct mlxsw_sp_neigh_entry *
2061 mlxsw_sp_rif_neigh_next(struct mlxsw_sp_rif *rif,
2062 struct mlxsw_sp_neigh_entry *neigh_entry)
2065 if (list_empty(&rif->neigh_list))
2068 return list_first_entry(&rif->neigh_list,
2069 typeof(*neigh_entry),
2072 if (list_is_last(&neigh_entry->rif_list_node, &rif->neigh_list))
2074 return list_next_entry(neigh_entry, rif_list_node);
2077 int mlxsw_sp_neigh_entry_type(struct mlxsw_sp_neigh_entry *neigh_entry)
2079 return neigh_entry->key.n->tbl->family;
2083 mlxsw_sp_neigh_entry_ha(struct mlxsw_sp_neigh_entry *neigh_entry)
2085 return neigh_entry->ha;
2088 u32 mlxsw_sp_neigh4_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
2090 struct neighbour *n;
2092 n = neigh_entry->key.n;
2093 return ntohl(*((__be32 *) n->primary_key));
2097 mlxsw_sp_neigh6_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
2099 struct neighbour *n;
2101 n = neigh_entry->key.n;
2102 return (struct in6_addr *) &n->primary_key;
2105 int mlxsw_sp_neigh_counter_get(struct mlxsw_sp *mlxsw_sp,
2106 struct mlxsw_sp_neigh_entry *neigh_entry,
2109 if (!neigh_entry->counter_valid)
2112 return mlxsw_sp_flow_counter_get(mlxsw_sp, neigh_entry->counter_index,
2116 static struct mlxsw_sp_neigh_entry *
2117 mlxsw_sp_neigh_entry_alloc(struct mlxsw_sp *mlxsw_sp, struct neighbour *n,
2120 struct mlxsw_sp_neigh_entry *neigh_entry;
2122 neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_KERNEL);
2126 neigh_entry->key.n = n;
2127 neigh_entry->rif = rif;
2128 INIT_LIST_HEAD(&neigh_entry->nexthop_list);
2133 static void mlxsw_sp_neigh_entry_free(struct mlxsw_sp_neigh_entry *neigh_entry)
2139 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
2140 struct mlxsw_sp_neigh_entry *neigh_entry)
2142 return rhashtable_insert_fast(&mlxsw_sp->router->neigh_ht,
2143 &neigh_entry->ht_node,
2144 mlxsw_sp_neigh_ht_params);
2148 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
2149 struct mlxsw_sp_neigh_entry *neigh_entry)
2151 rhashtable_remove_fast(&mlxsw_sp->router->neigh_ht,
2152 &neigh_entry->ht_node,
2153 mlxsw_sp_neigh_ht_params);
2157 mlxsw_sp_neigh_counter_should_alloc(struct mlxsw_sp *mlxsw_sp,
2158 struct mlxsw_sp_neigh_entry *neigh_entry)
2160 struct devlink *devlink;
2161 const char *table_name;
2163 switch (mlxsw_sp_neigh_entry_type(neigh_entry)) {
2165 table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST4;
2168 table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST6;
2175 devlink = priv_to_devlink(mlxsw_sp->core);
2176 return devlink_dpipe_table_counter_enabled(devlink, table_name);
2180 mlxsw_sp_neigh_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2181 struct mlxsw_sp_neigh_entry *neigh_entry)
2183 if (!mlxsw_sp_neigh_counter_should_alloc(mlxsw_sp, neigh_entry))
2186 if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &neigh_entry->counter_index))
2189 neigh_entry->counter_valid = true;
2193 mlxsw_sp_neigh_counter_free(struct mlxsw_sp *mlxsw_sp,
2194 struct mlxsw_sp_neigh_entry *neigh_entry)
2196 if (!neigh_entry->counter_valid)
2198 mlxsw_sp_flow_counter_free(mlxsw_sp,
2199 neigh_entry->counter_index);
2200 neigh_entry->counter_valid = false;
2203 static struct mlxsw_sp_neigh_entry *
2204 mlxsw_sp_neigh_entry_create(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
2206 struct mlxsw_sp_neigh_entry *neigh_entry;
2207 struct mlxsw_sp_rif *rif;
2210 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
2212 return ERR_PTR(-EINVAL);
2214 neigh_entry = mlxsw_sp_neigh_entry_alloc(mlxsw_sp, n, rif->rif_index);
2216 return ERR_PTR(-ENOMEM);
2218 err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
2220 goto err_neigh_entry_insert;
2222 mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2223 list_add(&neigh_entry->rif_list_node, &rif->neigh_list);
2227 err_neigh_entry_insert:
2228 mlxsw_sp_neigh_entry_free(neigh_entry);
2229 return ERR_PTR(err);
2233 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp *mlxsw_sp,
2234 struct mlxsw_sp_neigh_entry *neigh_entry)
2236 list_del(&neigh_entry->rif_list_node);
2237 mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2238 mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
2239 mlxsw_sp_neigh_entry_free(neigh_entry);
2242 static struct mlxsw_sp_neigh_entry *
2243 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
2245 struct mlxsw_sp_neigh_key key;
2248 return rhashtable_lookup_fast(&mlxsw_sp->router->neigh_ht,
2249 &key, mlxsw_sp_neigh_ht_params);
2253 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
2255 unsigned long interval;
2257 #if IS_ENABLED(CONFIG_IPV6)
2258 interval = min_t(unsigned long,
2259 NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME),
2260 NEIGH_VAR(&nd_tbl.parms, DELAY_PROBE_TIME));
2262 interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
2264 mlxsw_sp->router->neighs_update.interval = jiffies_to_msecs(interval);
2267 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
2271 struct net_device *dev;
2272 struct neighbour *n;
2277 mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
2279 if (!mlxsw_sp->router->rifs[rif]) {
2280 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
2285 dev = mlxsw_sp->router->rifs[rif]->dev;
2286 n = neigh_lookup(&arp_tbl, &dipn, dev);
2290 netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
2291 neigh_event_send(n, NULL);
2295 #if IS_ENABLED(CONFIG_IPV6)
2296 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2300 struct net_device *dev;
2301 struct neighbour *n;
2302 struct in6_addr dip;
2305 mlxsw_reg_rauhtd_ent_ipv6_unpack(rauhtd_pl, rec_index, &rif,
2308 if (!mlxsw_sp->router->rifs[rif]) {
2309 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
2313 dev = mlxsw_sp->router->rifs[rif]->dev;
2314 n = neigh_lookup(&nd_tbl, &dip, dev);
2318 netdev_dbg(dev, "Updating neighbour with IP=%pI6c\n", &dip);
2319 neigh_event_send(n, NULL);
2323 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2330 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
2337 num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2339 /* Hardware starts counting at 0, so add 1. */
2342 /* Each record consists of several neighbour entries. */
2343 for (i = 0; i < num_entries; i++) {
2346 ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
2347 mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
2353 static void mlxsw_sp_router_neigh_rec_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2357 /* One record contains one entry. */
2358 mlxsw_sp_router_neigh_ent_ipv6_process(mlxsw_sp, rauhtd_pl,
2362 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
2363 char *rauhtd_pl, int rec_index)
2365 switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
2366 case MLXSW_REG_RAUHTD_TYPE_IPV4:
2367 mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
2370 case MLXSW_REG_RAUHTD_TYPE_IPV6:
2371 mlxsw_sp_router_neigh_rec_ipv6_process(mlxsw_sp, rauhtd_pl,
2377 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
2379 u8 num_rec, last_rec_index, num_entries;
2381 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2382 last_rec_index = num_rec - 1;
2384 if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
2386 if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
2387 MLXSW_REG_RAUHTD_TYPE_IPV6)
2390 num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2392 if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
2398 __mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp,
2400 enum mlxsw_reg_rauhtd_type type)
2405 /* Ensure the RIF we read from the device does not change mid-dump. */
2406 mutex_lock(&mlxsw_sp->router->lock);
2408 mlxsw_reg_rauhtd_pack(rauhtd_pl, type);
2409 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
2412 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour table\n");
2415 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2416 for (i = 0; i < num_rec; i++)
2417 mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
2419 } while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
2420 mutex_unlock(&mlxsw_sp->router->lock);
2425 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
2427 enum mlxsw_reg_rauhtd_type type;
2431 rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
2435 type = MLXSW_REG_RAUHTD_TYPE_IPV4;
2436 err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2440 type = MLXSW_REG_RAUHTD_TYPE_IPV6;
2441 err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2447 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
2449 struct mlxsw_sp_neigh_entry *neigh_entry;
2451 mutex_lock(&mlxsw_sp->router->lock);
2452 list_for_each_entry(neigh_entry, &mlxsw_sp->router->nexthop_neighs_list,
2453 nexthop_neighs_list_node)
2454 /* If this neigh have nexthops, make the kernel think this neigh
2455 * is active regardless of the traffic.
2457 neigh_event_send(neigh_entry->key.n, NULL);
2458 mutex_unlock(&mlxsw_sp->router->lock);
2462 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
2464 unsigned long interval = mlxsw_sp->router->neighs_update.interval;
2466 mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw,
2467 msecs_to_jiffies(interval));
2470 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
2472 struct mlxsw_sp_router *router;
2475 router = container_of(work, struct mlxsw_sp_router,
2476 neighs_update.dw.work);
2477 err = mlxsw_sp_router_neighs_update_rauhtd(router->mlxsw_sp);
2479 dev_err(router->mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
2481 mlxsw_sp_router_neighs_update_nh(router->mlxsw_sp);
2483 mlxsw_sp_router_neighs_update_work_schedule(router->mlxsw_sp);
2486 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
2488 struct mlxsw_sp_neigh_entry *neigh_entry;
2489 struct mlxsw_sp_router *router;
2491 router = container_of(work, struct mlxsw_sp_router,
2492 nexthop_probe_dw.work);
2493 /* Iterate over nexthop neighbours, find those who are unresolved and
2494 * send arp on them. This solves the chicken-egg problem when
2495 * the nexthop wouldn't get offloaded until the neighbor is resolved
2496 * but it wouldn't get resolved ever in case traffic is flowing in HW
2497 * using different nexthop.
2499 mutex_lock(&router->lock);
2500 list_for_each_entry(neigh_entry, &router->nexthop_neighs_list,
2501 nexthop_neighs_list_node)
2502 if (!neigh_entry->connected)
2503 neigh_event_send(neigh_entry->key.n, NULL);
2504 mutex_unlock(&router->lock);
2506 mlxsw_core_schedule_dw(&router->nexthop_probe_dw,
2507 MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
2511 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
2512 struct mlxsw_sp_neigh_entry *neigh_entry,
2513 bool removing, bool dead);
2515 static enum mlxsw_reg_rauht_op mlxsw_sp_rauht_op(bool adding)
2517 return adding ? MLXSW_REG_RAUHT_OP_WRITE_ADD :
2518 MLXSW_REG_RAUHT_OP_WRITE_DELETE;
2522 mlxsw_sp_router_neigh_entry_op4(struct mlxsw_sp *mlxsw_sp,
2523 struct mlxsw_sp_neigh_entry *neigh_entry,
2524 enum mlxsw_reg_rauht_op op)
2526 struct neighbour *n = neigh_entry->key.n;
2527 u32 dip = ntohl(*((__be32 *) n->primary_key));
2528 char rauht_pl[MLXSW_REG_RAUHT_LEN];
2530 mlxsw_reg_rauht_pack4(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2532 if (neigh_entry->counter_valid)
2533 mlxsw_reg_rauht_pack_counter(rauht_pl,
2534 neigh_entry->counter_index);
2535 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2539 mlxsw_sp_router_neigh_entry_op6(struct mlxsw_sp *mlxsw_sp,
2540 struct mlxsw_sp_neigh_entry *neigh_entry,
2541 enum mlxsw_reg_rauht_op op)
2543 struct neighbour *n = neigh_entry->key.n;
2544 char rauht_pl[MLXSW_REG_RAUHT_LEN];
2545 const char *dip = n->primary_key;
2547 mlxsw_reg_rauht_pack6(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2549 if (neigh_entry->counter_valid)
2550 mlxsw_reg_rauht_pack_counter(rauht_pl,
2551 neigh_entry->counter_index);
2552 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2555 bool mlxsw_sp_neigh_ipv6_ignore(struct mlxsw_sp_neigh_entry *neigh_entry)
2557 struct neighbour *n = neigh_entry->key.n;
2559 /* Packets with a link-local destination address are trapped
2560 * after LPM lookup and never reach the neighbour table, so
2561 * there is no need to program such neighbours to the device.
2563 if (ipv6_addr_type((struct in6_addr *) &n->primary_key) &
2564 IPV6_ADDR_LINKLOCAL)
2570 mlxsw_sp_neigh_entry_update(struct mlxsw_sp *mlxsw_sp,
2571 struct mlxsw_sp_neigh_entry *neigh_entry,
2574 enum mlxsw_reg_rauht_op op = mlxsw_sp_rauht_op(adding);
2577 if (!adding && !neigh_entry->connected)
2579 neigh_entry->connected = adding;
2580 if (neigh_entry->key.n->tbl->family == AF_INET) {
2581 err = mlxsw_sp_router_neigh_entry_op4(mlxsw_sp, neigh_entry,
2585 } else if (neigh_entry->key.n->tbl->family == AF_INET6) {
2586 if (mlxsw_sp_neigh_ipv6_ignore(neigh_entry))
2588 err = mlxsw_sp_router_neigh_entry_op6(mlxsw_sp, neigh_entry,
2598 neigh_entry->key.n->flags |= NTF_OFFLOADED;
2600 neigh_entry->key.n->flags &= ~NTF_OFFLOADED;
2604 mlxsw_sp_neigh_entry_counter_update(struct mlxsw_sp *mlxsw_sp,
2605 struct mlxsw_sp_neigh_entry *neigh_entry,
2609 mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2611 mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2612 mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, true);
2615 struct mlxsw_sp_netevent_work {
2616 struct work_struct work;
2617 struct mlxsw_sp *mlxsw_sp;
2618 struct neighbour *n;
2621 static void mlxsw_sp_router_neigh_event_work(struct work_struct *work)
2623 struct mlxsw_sp_netevent_work *net_work =
2624 container_of(work, struct mlxsw_sp_netevent_work, work);
2625 struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2626 struct mlxsw_sp_neigh_entry *neigh_entry;
2627 struct neighbour *n = net_work->n;
2628 unsigned char ha[ETH_ALEN];
2629 bool entry_connected;
2632 /* If these parameters are changed after we release the lock,
2633 * then we are guaranteed to receive another event letting us
2636 read_lock_bh(&n->lock);
2637 memcpy(ha, n->ha, ETH_ALEN);
2638 nud_state = n->nud_state;
2640 read_unlock_bh(&n->lock);
2642 mutex_lock(&mlxsw_sp->router->lock);
2643 mlxsw_sp_span_respin(mlxsw_sp);
2645 entry_connected = nud_state & NUD_VALID && !dead;
2646 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
2647 if (!entry_connected && !neigh_entry)
2650 neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
2651 if (IS_ERR(neigh_entry))
2655 memcpy(neigh_entry->ha, ha, ETH_ALEN);
2656 mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, entry_connected);
2657 mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, !entry_connected,
2660 if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
2661 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2664 mutex_unlock(&mlxsw_sp->router->lock);
2669 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp);
2671 static void mlxsw_sp_router_mp_hash_event_work(struct work_struct *work)
2673 struct mlxsw_sp_netevent_work *net_work =
2674 container_of(work, struct mlxsw_sp_netevent_work, work);
2675 struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2677 mlxsw_sp_mp_hash_init(mlxsw_sp);
2681 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp);
2683 static void mlxsw_sp_router_update_priority_work(struct work_struct *work)
2685 struct mlxsw_sp_netevent_work *net_work =
2686 container_of(work, struct mlxsw_sp_netevent_work, work);
2687 struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2689 __mlxsw_sp_router_init(mlxsw_sp);
2693 static int mlxsw_sp_router_schedule_work(struct net *net,
2694 struct notifier_block *nb,
2695 void (*cb)(struct work_struct *))
2697 struct mlxsw_sp_netevent_work *net_work;
2698 struct mlxsw_sp_router *router;
2700 router = container_of(nb, struct mlxsw_sp_router, netevent_nb);
2701 if (!net_eq(net, mlxsw_sp_net(router->mlxsw_sp)))
2704 net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2708 INIT_WORK(&net_work->work, cb);
2709 net_work->mlxsw_sp = router->mlxsw_sp;
2710 mlxsw_core_schedule_work(&net_work->work);
2714 static int mlxsw_sp_router_netevent_event(struct notifier_block *nb,
2715 unsigned long event, void *ptr)
2717 struct mlxsw_sp_netevent_work *net_work;
2718 struct mlxsw_sp_port *mlxsw_sp_port;
2719 struct mlxsw_sp *mlxsw_sp;
2720 unsigned long interval;
2721 struct neigh_parms *p;
2722 struct neighbour *n;
2725 case NETEVENT_DELAY_PROBE_TIME_UPDATE:
2728 /* We don't care about changes in the default table. */
2729 if (!p->dev || (p->tbl->family != AF_INET &&
2730 p->tbl->family != AF_INET6))
2733 /* We are in atomic context and can't take RTNL mutex,
2734 * so use RCU variant to walk the device chain.
2736 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
2740 mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2741 interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
2742 mlxsw_sp->router->neighs_update.interval = interval;
2744 mlxsw_sp_port_dev_put(mlxsw_sp_port);
2746 case NETEVENT_NEIGH_UPDATE:
2749 if (n->tbl->family != AF_INET && n->tbl->family != AF_INET6)
2752 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(n->dev);
2756 net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2758 mlxsw_sp_port_dev_put(mlxsw_sp_port);
2762 INIT_WORK(&net_work->work, mlxsw_sp_router_neigh_event_work);
2763 net_work->mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2766 /* Take a reference to ensure the neighbour won't be
2767 * destructed until we drop the reference in delayed
2771 mlxsw_core_schedule_work(&net_work->work);
2772 mlxsw_sp_port_dev_put(mlxsw_sp_port);
2774 case NETEVENT_IPV4_MPATH_HASH_UPDATE:
2775 case NETEVENT_IPV6_MPATH_HASH_UPDATE:
2776 return mlxsw_sp_router_schedule_work(ptr, nb,
2777 mlxsw_sp_router_mp_hash_event_work);
2779 case NETEVENT_IPV4_FWD_UPDATE_PRIORITY_UPDATE:
2780 return mlxsw_sp_router_schedule_work(ptr, nb,
2781 mlxsw_sp_router_update_priority_work);
2787 static int mlxsw_sp_neigh_init(struct mlxsw_sp *mlxsw_sp)
2791 err = rhashtable_init(&mlxsw_sp->router->neigh_ht,
2792 &mlxsw_sp_neigh_ht_params);
2796 /* Initialize the polling interval according to the default
2799 mlxsw_sp_router_neighs_update_interval_init(mlxsw_sp);
2801 /* Create the delayed works for the activity_update */
2802 INIT_DELAYED_WORK(&mlxsw_sp->router->neighs_update.dw,
2803 mlxsw_sp_router_neighs_update_work);
2804 INIT_DELAYED_WORK(&mlxsw_sp->router->nexthop_probe_dw,
2805 mlxsw_sp_router_probe_unresolved_nexthops);
2806 mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw, 0);
2807 mlxsw_core_schedule_dw(&mlxsw_sp->router->nexthop_probe_dw, 0);
2811 static void mlxsw_sp_neigh_fini(struct mlxsw_sp *mlxsw_sp)
2813 cancel_delayed_work_sync(&mlxsw_sp->router->neighs_update.dw);
2814 cancel_delayed_work_sync(&mlxsw_sp->router->nexthop_probe_dw);
2815 rhashtable_destroy(&mlxsw_sp->router->neigh_ht);
2818 static void mlxsw_sp_neigh_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
2819 struct mlxsw_sp_rif *rif)
2821 struct mlxsw_sp_neigh_entry *neigh_entry, *tmp;
2823 list_for_each_entry_safe(neigh_entry, tmp, &rif->neigh_list,
2825 mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, false);
2826 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2830 enum mlxsw_sp_nexthop_type {
2831 MLXSW_SP_NEXTHOP_TYPE_ETH,
2832 MLXSW_SP_NEXTHOP_TYPE_IPIP,
2835 struct mlxsw_sp_nexthop_key {
2836 struct fib_nh *fib_nh;
2839 struct mlxsw_sp_nexthop {
2840 struct list_head neigh_list_node; /* member of neigh entry list */
2841 struct list_head rif_list_node;
2842 struct list_head router_list_node;
2843 struct mlxsw_sp_nexthop_group_info *nhgi; /* pointer back to the group
2844 * this nexthop belongs to
2846 struct rhash_head ht_node;
2847 struct neigh_table *neigh_tbl;
2848 struct mlxsw_sp_nexthop_key key;
2849 unsigned char gw_addr[sizeof(struct in6_addr)];
2853 int num_adj_entries;
2854 struct mlxsw_sp_rif *rif;
2855 u8 should_offload:1, /* set indicates this neigh is connected and
2856 * should be put to KVD linear area of this group.
2858 offloaded:1, /* set in case the neigh is actually put into
2859 * KVD linear area of this group.
2861 update:1; /* set indicates that MAC of this neigh should be
2864 enum mlxsw_sp_nexthop_type type;
2866 struct mlxsw_sp_neigh_entry *neigh_entry;
2867 struct mlxsw_sp_ipip_entry *ipip_entry;
2869 unsigned int counter_index;
2873 enum mlxsw_sp_nexthop_group_type {
2874 MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4,
2875 MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6,
2878 struct mlxsw_sp_nexthop_group_info {
2879 struct mlxsw_sp_nexthop_group *nh_grp;
2883 int sum_norm_weight;
2884 u8 adj_index_valid:1,
2885 gateway:1; /* routes using the group use a gateway */
2886 struct mlxsw_sp_nexthop nexthops[0];
2887 #define nh_rif nexthops[0].rif
2890 struct mlxsw_sp_nexthop_group {
2891 struct rhash_head ht_node;
2892 struct list_head fib_list; /* list of fib entries that use this group */
2895 struct fib_info *fi;
2898 struct mlxsw_sp_nexthop_group_info *nhgi;
2899 enum mlxsw_sp_nexthop_group_type type;
2902 void mlxsw_sp_nexthop_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2903 struct mlxsw_sp_nexthop *nh)
2905 struct devlink *devlink;
2907 devlink = priv_to_devlink(mlxsw_sp->core);
2908 if (!devlink_dpipe_table_counter_enabled(devlink,
2909 MLXSW_SP_DPIPE_TABLE_NAME_ADJ))
2912 if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &nh->counter_index))
2915 nh->counter_valid = true;
2918 void mlxsw_sp_nexthop_counter_free(struct mlxsw_sp *mlxsw_sp,
2919 struct mlxsw_sp_nexthop *nh)
2921 if (!nh->counter_valid)
2923 mlxsw_sp_flow_counter_free(mlxsw_sp, nh->counter_index);
2924 nh->counter_valid = false;
2927 int mlxsw_sp_nexthop_counter_get(struct mlxsw_sp *mlxsw_sp,
2928 struct mlxsw_sp_nexthop *nh, u64 *p_counter)
2930 if (!nh->counter_valid)
2933 return mlxsw_sp_flow_counter_get(mlxsw_sp, nh->counter_index,
2937 struct mlxsw_sp_nexthop *mlxsw_sp_nexthop_next(struct mlxsw_sp_router *router,
2938 struct mlxsw_sp_nexthop *nh)
2941 if (list_empty(&router->nexthop_list))
2944 return list_first_entry(&router->nexthop_list,
2945 typeof(*nh), router_list_node);
2947 if (list_is_last(&nh->router_list_node, &router->nexthop_list))
2949 return list_next_entry(nh, router_list_node);
2952 bool mlxsw_sp_nexthop_offload(struct mlxsw_sp_nexthop *nh)
2954 return nh->offloaded;
2957 unsigned char *mlxsw_sp_nexthop_ha(struct mlxsw_sp_nexthop *nh)
2961 return nh->neigh_entry->ha;
2964 int mlxsw_sp_nexthop_indexes(struct mlxsw_sp_nexthop *nh, u32 *p_adj_index,
2965 u32 *p_adj_size, u32 *p_adj_hash_index)
2967 struct mlxsw_sp_nexthop_group_info *nhgi = nh->nhgi;
2968 u32 adj_hash_index = 0;
2971 if (!nh->offloaded || !nhgi->adj_index_valid)
2974 *p_adj_index = nhgi->adj_index;
2975 *p_adj_size = nhgi->ecmp_size;
2977 for (i = 0; i < nhgi->count; i++) {
2978 struct mlxsw_sp_nexthop *nh_iter = &nhgi->nexthops[i];
2982 if (nh_iter->offloaded)
2983 adj_hash_index += nh_iter->num_adj_entries;
2986 *p_adj_hash_index = adj_hash_index;
2990 struct mlxsw_sp_rif *mlxsw_sp_nexthop_rif(struct mlxsw_sp_nexthop *nh)
2995 bool mlxsw_sp_nexthop_group_has_ipip(struct mlxsw_sp_nexthop *nh)
2997 struct mlxsw_sp_nexthop_group_info *nhgi = nh->nhgi;
3000 for (i = 0; i < nhgi->count; i++) {
3001 struct mlxsw_sp_nexthop *nh_iter = &nhgi->nexthops[i];
3003 if (nh_iter->type == MLXSW_SP_NEXTHOP_TYPE_IPIP)
3009 struct mlxsw_sp_nexthop_group_cmp_arg {
3010 enum mlxsw_sp_nexthop_group_type type;
3012 struct fib_info *fi;
3013 struct mlxsw_sp_fib6_entry *fib6_entry;
3018 mlxsw_sp_nexthop6_group_has_nexthop(const struct mlxsw_sp_nexthop_group *nh_grp,
3019 const struct in6_addr *gw, int ifindex,
3024 for (i = 0; i < nh_grp->nhgi->count; i++) {
3025 const struct mlxsw_sp_nexthop *nh;
3027 nh = &nh_grp->nhgi->nexthops[i];
3028 if (nh->ifindex == ifindex && nh->nh_weight == weight &&
3029 ipv6_addr_equal(gw, (struct in6_addr *) nh->gw_addr))
3037 mlxsw_sp_nexthop6_group_cmp(const struct mlxsw_sp_nexthop_group *nh_grp,
3038 const struct mlxsw_sp_fib6_entry *fib6_entry)
3040 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3042 if (nh_grp->nhgi->count != fib6_entry->nrt6)
3045 list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3046 struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3047 struct in6_addr *gw;
3048 int ifindex, weight;
3050 ifindex = fib6_nh->fib_nh_dev->ifindex;
3051 weight = fib6_nh->fib_nh_weight;
3052 gw = &fib6_nh->fib_nh_gw6;
3053 if (!mlxsw_sp_nexthop6_group_has_nexthop(nh_grp, gw, ifindex,
3062 mlxsw_sp_nexthop_group_cmp(struct rhashtable_compare_arg *arg, const void *ptr)
3064 const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = arg->key;
3065 const struct mlxsw_sp_nexthop_group *nh_grp = ptr;
3067 if (nh_grp->type != cmp_arg->type)
3070 switch (cmp_arg->type) {
3071 case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3072 return cmp_arg->fi != nh_grp->ipv4.fi;
3073 case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3074 return !mlxsw_sp_nexthop6_group_cmp(nh_grp,
3075 cmp_arg->fib6_entry);
3082 static u32 mlxsw_sp_nexthop_group_hash_obj(const void *data, u32 len, u32 seed)
3084 const struct mlxsw_sp_nexthop_group *nh_grp = data;
3085 const struct mlxsw_sp_nexthop *nh;
3086 struct fib_info *fi;
3090 switch (nh_grp->type) {
3091 case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3092 fi = nh_grp->ipv4.fi;
3093 return jhash(&fi, sizeof(fi), seed);
3094 case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3095 val = nh_grp->nhgi->count;
3096 for (i = 0; i < nh_grp->nhgi->count; i++) {
3097 nh = &nh_grp->nhgi->nexthops[i];
3098 val ^= jhash(&nh->ifindex, sizeof(nh->ifindex), seed);
3099 val ^= jhash(&nh->gw_addr, sizeof(nh->gw_addr), seed);
3101 return jhash(&val, sizeof(val), seed);
3109 mlxsw_sp_nexthop6_group_hash(struct mlxsw_sp_fib6_entry *fib6_entry, u32 seed)
3111 unsigned int val = fib6_entry->nrt6;
3112 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3114 list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3115 struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3116 struct net_device *dev = fib6_nh->fib_nh_dev;
3117 struct in6_addr *gw = &fib6_nh->fib_nh_gw6;
3119 val ^= jhash(&dev->ifindex, sizeof(dev->ifindex), seed);
3120 val ^= jhash(gw, sizeof(*gw), seed);
3123 return jhash(&val, sizeof(val), seed);
3127 mlxsw_sp_nexthop_group_hash(const void *data, u32 len, u32 seed)
3129 const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = data;
3131 switch (cmp_arg->type) {
3132 case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3133 return jhash(&cmp_arg->fi, sizeof(cmp_arg->fi), seed);
3134 case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3135 return mlxsw_sp_nexthop6_group_hash(cmp_arg->fib6_entry, seed);
3142 static const struct rhashtable_params mlxsw_sp_nexthop_group_ht_params = {
3143 .head_offset = offsetof(struct mlxsw_sp_nexthop_group, ht_node),
3144 .hashfn = mlxsw_sp_nexthop_group_hash,
3145 .obj_hashfn = mlxsw_sp_nexthop_group_hash_obj,
3146 .obj_cmpfn = mlxsw_sp_nexthop_group_cmp,
3149 static int mlxsw_sp_nexthop_group_insert(struct mlxsw_sp *mlxsw_sp,
3150 struct mlxsw_sp_nexthop_group *nh_grp)
3152 if (nh_grp->type == MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6 &&
3153 !nh_grp->nhgi->gateway)
3156 return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_group_ht,
3158 mlxsw_sp_nexthop_group_ht_params);
3161 static void mlxsw_sp_nexthop_group_remove(struct mlxsw_sp *mlxsw_sp,
3162 struct mlxsw_sp_nexthop_group *nh_grp)
3164 if (nh_grp->type == MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6 &&
3165 !nh_grp->nhgi->gateway)
3168 rhashtable_remove_fast(&mlxsw_sp->router->nexthop_group_ht,
3170 mlxsw_sp_nexthop_group_ht_params);
3173 static struct mlxsw_sp_nexthop_group *
3174 mlxsw_sp_nexthop4_group_lookup(struct mlxsw_sp *mlxsw_sp,
3175 struct fib_info *fi)
3177 struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
3179 cmp_arg.type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4;
3181 return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
3183 mlxsw_sp_nexthop_group_ht_params);
3186 static struct mlxsw_sp_nexthop_group *
3187 mlxsw_sp_nexthop6_group_lookup(struct mlxsw_sp *mlxsw_sp,
3188 struct mlxsw_sp_fib6_entry *fib6_entry)
3190 struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
3192 cmp_arg.type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6;
3193 cmp_arg.fib6_entry = fib6_entry;
3194 return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
3196 mlxsw_sp_nexthop_group_ht_params);
3199 static const struct rhashtable_params mlxsw_sp_nexthop_ht_params = {
3200 .key_offset = offsetof(struct mlxsw_sp_nexthop, key),
3201 .head_offset = offsetof(struct mlxsw_sp_nexthop, ht_node),
3202 .key_len = sizeof(struct mlxsw_sp_nexthop_key),
3205 static int mlxsw_sp_nexthop_insert(struct mlxsw_sp *mlxsw_sp,
3206 struct mlxsw_sp_nexthop *nh)
3208 return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_ht,
3209 &nh->ht_node, mlxsw_sp_nexthop_ht_params);
3212 static void mlxsw_sp_nexthop_remove(struct mlxsw_sp *mlxsw_sp,
3213 struct mlxsw_sp_nexthop *nh)
3215 rhashtable_remove_fast(&mlxsw_sp->router->nexthop_ht, &nh->ht_node,
3216 mlxsw_sp_nexthop_ht_params);
3219 static struct mlxsw_sp_nexthop *
3220 mlxsw_sp_nexthop_lookup(struct mlxsw_sp *mlxsw_sp,
3221 struct mlxsw_sp_nexthop_key key)
3223 return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_ht, &key,
3224 mlxsw_sp_nexthop_ht_params);
3227 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
3228 const struct mlxsw_sp_fib *fib,
3229 u32 adj_index, u16 ecmp_size,
3233 char raleu_pl[MLXSW_REG_RALEU_LEN];
3235 mlxsw_reg_raleu_pack(raleu_pl,
3236 (enum mlxsw_reg_ralxx_protocol) fib->proto,
3237 fib->vr->id, adj_index, ecmp_size, new_adj_index,
3239 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
3242 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
3243 struct mlxsw_sp_nexthop_group *nh_grp,
3244 u32 old_adj_index, u16 old_ecmp_size)
3246 struct mlxsw_sp_fib_entry *fib_entry;
3247 struct mlxsw_sp_fib *fib = NULL;
3250 list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
3251 struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
3253 if (fib == fib_entry->fib_node->fib)
3255 fib = fib_entry->fib_node->fib;
3256 err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, fib,
3267 static int __mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
3268 struct mlxsw_sp_nexthop *nh)
3270 struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
3271 char ratr_pl[MLXSW_REG_RATR_LEN];
3273 mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
3274 true, MLXSW_REG_RATR_TYPE_ETHERNET,
3275 adj_index, neigh_entry->rif);
3276 mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
3277 if (nh->counter_valid)
3278 mlxsw_reg_ratr_counter_pack(ratr_pl, nh->counter_index, true);
3280 mlxsw_reg_ratr_counter_pack(ratr_pl, 0, false);
3282 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
3285 int mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
3286 struct mlxsw_sp_nexthop *nh)
3290 for (i = 0; i < nh->num_adj_entries; i++) {
3293 err = __mlxsw_sp_nexthop_update(mlxsw_sp, adj_index + i, nh);
3301 static int __mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
3303 struct mlxsw_sp_nexthop *nh)
3305 const struct mlxsw_sp_ipip_ops *ipip_ops;
3307 ipip_ops = mlxsw_sp->router->ipip_ops_arr[nh->ipip_entry->ipipt];
3308 return ipip_ops->nexthop_update(mlxsw_sp, adj_index, nh->ipip_entry);
3311 static int mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
3313 struct mlxsw_sp_nexthop *nh)
3317 for (i = 0; i < nh->num_adj_entries; i++) {
3320 err = __mlxsw_sp_nexthop_ipip_update(mlxsw_sp, adj_index + i,
3330 mlxsw_sp_nexthop_group_update(struct mlxsw_sp *mlxsw_sp,
3331 struct mlxsw_sp_nexthop_group_info *nhgi,
3334 u32 adj_index = nhgi->adj_index; /* base */
3335 struct mlxsw_sp_nexthop *nh;
3338 for (i = 0; i < nhgi->count; i++) {
3339 nh = &nhgi->nexthops[i];
3341 if (!nh->should_offload) {
3346 if (nh->update || reallocate) {
3350 case MLXSW_SP_NEXTHOP_TYPE_ETH:
3351 err = mlxsw_sp_nexthop_update
3352 (mlxsw_sp, adj_index, nh);
3354 case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3355 err = mlxsw_sp_nexthop_ipip_update
3356 (mlxsw_sp, adj_index, nh);
3364 adj_index += nh->num_adj_entries;
3370 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
3371 struct mlxsw_sp_nexthop_group *nh_grp)
3373 struct mlxsw_sp_fib_entry *fib_entry;
3376 list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
3377 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
3384 static void mlxsw_sp_adj_grp_size_round_up(u16 *p_adj_grp_size)
3386 /* Valid sizes for an adjacency group are:
3387 * 1-64, 512, 1024, 2048 and 4096.
3389 if (*p_adj_grp_size <= 64)
3391 else if (*p_adj_grp_size <= 512)
3392 *p_adj_grp_size = 512;
3393 else if (*p_adj_grp_size <= 1024)
3394 *p_adj_grp_size = 1024;
3395 else if (*p_adj_grp_size <= 2048)
3396 *p_adj_grp_size = 2048;
3398 *p_adj_grp_size = 4096;
3401 static void mlxsw_sp_adj_grp_size_round_down(u16 *p_adj_grp_size,
3402 unsigned int alloc_size)
3404 if (alloc_size >= 4096)
3405 *p_adj_grp_size = 4096;
3406 else if (alloc_size >= 2048)
3407 *p_adj_grp_size = 2048;
3408 else if (alloc_size >= 1024)
3409 *p_adj_grp_size = 1024;
3410 else if (alloc_size >= 512)
3411 *p_adj_grp_size = 512;
3414 static int mlxsw_sp_fix_adj_grp_size(struct mlxsw_sp *mlxsw_sp,
3415 u16 *p_adj_grp_size)
3417 unsigned int alloc_size;
3420 /* Round up the requested group size to the next size supported
3421 * by the device and make sure the request can be satisfied.
3423 mlxsw_sp_adj_grp_size_round_up(p_adj_grp_size);
3424 err = mlxsw_sp_kvdl_alloc_count_query(mlxsw_sp,
3425 MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3426 *p_adj_grp_size, &alloc_size);
3429 /* It is possible the allocation results in more allocated
3430 * entries than requested. Try to use as much of them as
3433 mlxsw_sp_adj_grp_size_round_down(p_adj_grp_size, alloc_size);
3439 mlxsw_sp_nexthop_group_normalize(struct mlxsw_sp_nexthop_group_info *nhgi)
3441 int i, g = 0, sum_norm_weight = 0;
3442 struct mlxsw_sp_nexthop *nh;
3444 for (i = 0; i < nhgi->count; i++) {
3445 nh = &nhgi->nexthops[i];
3447 if (!nh->should_offload)
3450 g = gcd(nh->nh_weight, g);
3455 for (i = 0; i < nhgi->count; i++) {
3456 nh = &nhgi->nexthops[i];
3458 if (!nh->should_offload)
3460 nh->norm_nh_weight = nh->nh_weight / g;
3461 sum_norm_weight += nh->norm_nh_weight;
3464 nhgi->sum_norm_weight = sum_norm_weight;
3468 mlxsw_sp_nexthop_group_rebalance(struct mlxsw_sp_nexthop_group_info *nhgi)
3470 int i, weight = 0, lower_bound = 0;
3471 int total = nhgi->sum_norm_weight;
3472 u16 ecmp_size = nhgi->ecmp_size;
3474 for (i = 0; i < nhgi->count; i++) {
3475 struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
3478 if (!nh->should_offload)
3480 weight += nh->norm_nh_weight;
3481 upper_bound = DIV_ROUND_CLOSEST(ecmp_size * weight, total);
3482 nh->num_adj_entries = upper_bound - lower_bound;
3483 lower_bound = upper_bound;
3487 static struct mlxsw_sp_nexthop *
3488 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
3489 const struct mlxsw_sp_rt6 *mlxsw_sp_rt6);
3492 mlxsw_sp_nexthop4_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3493 struct mlxsw_sp_nexthop_group *nh_grp)
3497 for (i = 0; i < nh_grp->nhgi->count; i++) {
3498 struct mlxsw_sp_nexthop *nh = &nh_grp->nhgi->nexthops[i];
3501 nh->key.fib_nh->fib_nh_flags |= RTNH_F_OFFLOAD;
3503 nh->key.fib_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
3508 __mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp_nexthop_group *nh_grp,
3509 struct mlxsw_sp_fib6_entry *fib6_entry)
3511 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3513 list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3514 struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3515 struct mlxsw_sp_nexthop *nh;
3517 nh = mlxsw_sp_rt6_nexthop(nh_grp, mlxsw_sp_rt6);
3518 if (nh && nh->offloaded)
3519 fib6_nh->fib_nh_flags |= RTNH_F_OFFLOAD;
3521 fib6_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
3526 mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3527 struct mlxsw_sp_nexthop_group *nh_grp)
3529 struct mlxsw_sp_fib6_entry *fib6_entry;
3531 /* Unfortunately, in IPv6 the route and the nexthop are described by
3532 * the same struct, so we need to iterate over all the routes using the
3533 * nexthop group and set / clear the offload indication for them.
3535 list_for_each_entry(fib6_entry, &nh_grp->fib_list,
3536 common.nexthop_group_node)
3537 __mlxsw_sp_nexthop6_group_offload_refresh(nh_grp, fib6_entry);
3541 mlxsw_sp_nexthop_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3542 struct mlxsw_sp_nexthop_group *nh_grp)
3544 switch (nh_grp->type) {
3545 case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3546 mlxsw_sp_nexthop4_group_offload_refresh(mlxsw_sp, nh_grp);
3548 case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3549 mlxsw_sp_nexthop6_group_offload_refresh(mlxsw_sp, nh_grp);
3555 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
3556 struct mlxsw_sp_nexthop_group *nh_grp)
3558 struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
3559 u16 ecmp_size, old_ecmp_size;
3560 struct mlxsw_sp_nexthop *nh;
3561 bool offload_change = false;
3563 bool old_adj_index_valid;
3568 if (!nhgi->gateway) {
3569 mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3573 for (i = 0; i < nhgi->count; i++) {
3574 nh = &nhgi->nexthops[i];
3576 if (nh->should_offload != nh->offloaded) {
3577 offload_change = true;
3578 if (nh->should_offload)
3582 if (!offload_change) {
3583 /* Nothing was added or removed, so no need to reallocate. Just
3584 * update MAC on existing adjacency indexes.
3586 err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nhgi, false);
3588 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3593 mlxsw_sp_nexthop_group_normalize(nhgi);
3594 if (!nhgi->sum_norm_weight)
3595 /* No neigh of this group is connected so we just set
3596 * the trap and let everthing flow through kernel.
3600 ecmp_size = nhgi->sum_norm_weight;
3601 err = mlxsw_sp_fix_adj_grp_size(mlxsw_sp, &ecmp_size);
3603 /* No valid allocation size available. */
3606 err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3607 ecmp_size, &adj_index);
3609 /* We ran out of KVD linear space, just set the
3610 * trap and let everything flow through kernel.
3612 dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
3615 old_adj_index_valid = nhgi->adj_index_valid;
3616 old_adj_index = nhgi->adj_index;
3617 old_ecmp_size = nhgi->ecmp_size;
3618 nhgi->adj_index_valid = 1;
3619 nhgi->adj_index = adj_index;
3620 nhgi->ecmp_size = ecmp_size;
3621 mlxsw_sp_nexthop_group_rebalance(nhgi);
3622 err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nhgi, true);
3624 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3628 mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3630 if (!old_adj_index_valid) {
3631 /* The trap was set for fib entries, so we have to call
3632 * fib entry update to unset it and use adjacency index.
3634 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3636 dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
3642 err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
3643 old_adj_index, old_ecmp_size);
3644 mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3645 old_ecmp_size, old_adj_index);
3647 dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
3654 old_adj_index_valid = nhgi->adj_index_valid;
3655 nhgi->adj_index_valid = 0;
3656 for (i = 0; i < nhgi->count; i++) {
3657 nh = &nhgi->nexthops[i];
3660 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3662 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
3663 mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3664 if (old_adj_index_valid)
3665 mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3666 nhgi->ecmp_size, nhgi->adj_index);
3669 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
3673 nh->should_offload = 1;
3675 nh->should_offload = 0;
3680 mlxsw_sp_nexthop_dead_neigh_replace(struct mlxsw_sp *mlxsw_sp,
3681 struct mlxsw_sp_neigh_entry *neigh_entry)
3683 struct neighbour *n, *old_n = neigh_entry->key.n;
3684 struct mlxsw_sp_nexthop *nh;
3685 bool entry_connected;
3689 nh = list_first_entry(&neigh_entry->nexthop_list,
3690 struct mlxsw_sp_nexthop, neigh_list_node);
3692 n = neigh_lookup(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3694 n = neigh_create(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3697 neigh_event_send(n, NULL);
3700 mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
3701 neigh_entry->key.n = n;
3702 err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
3704 goto err_neigh_entry_insert;
3706 read_lock_bh(&n->lock);
3707 nud_state = n->nud_state;
3709 read_unlock_bh(&n->lock);
3710 entry_connected = nud_state & NUD_VALID && !dead;
3712 list_for_each_entry(nh, &neigh_entry->nexthop_list,
3714 neigh_release(old_n);
3716 __mlxsw_sp_nexthop_neigh_update(nh, !entry_connected);
3717 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
3724 err_neigh_entry_insert:
3725 neigh_entry->key.n = old_n;
3726 mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
3732 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
3733 struct mlxsw_sp_neigh_entry *neigh_entry,
3734 bool removing, bool dead)
3736 struct mlxsw_sp_nexthop *nh;
3738 if (list_empty(&neigh_entry->nexthop_list))
3744 err = mlxsw_sp_nexthop_dead_neigh_replace(mlxsw_sp,
3747 dev_err(mlxsw_sp->bus_info->dev, "Failed to replace dead neigh\n");
3751 list_for_each_entry(nh, &neigh_entry->nexthop_list,
3753 __mlxsw_sp_nexthop_neigh_update(nh, removing);
3754 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
3758 static void mlxsw_sp_nexthop_rif_init(struct mlxsw_sp_nexthop *nh,
3759 struct mlxsw_sp_rif *rif)
3765 list_add(&nh->rif_list_node, &rif->nexthop_list);
3768 static void mlxsw_sp_nexthop_rif_fini(struct mlxsw_sp_nexthop *nh)
3773 list_del(&nh->rif_list_node);
3777 static int mlxsw_sp_nexthop_neigh_init(struct mlxsw_sp *mlxsw_sp,
3778 struct mlxsw_sp_nexthop *nh)
3780 struct mlxsw_sp_neigh_entry *neigh_entry;
3781 struct neighbour *n;
3785 if (!nh->nhgi->gateway || nh->neigh_entry)
3788 /* Take a reference of neigh here ensuring that neigh would
3789 * not be destructed before the nexthop entry is finished.
3790 * The reference is taken either in neigh_lookup() or
3791 * in neigh_create() in case n is not found.
3793 n = neigh_lookup(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3795 n = neigh_create(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3798 neigh_event_send(n, NULL);
3800 neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
3802 neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
3803 if (IS_ERR(neigh_entry)) {
3805 goto err_neigh_entry_create;
3809 /* If that is the first nexthop connected to that neigh, add to
3810 * nexthop_neighs_list
3812 if (list_empty(&neigh_entry->nexthop_list))
3813 list_add_tail(&neigh_entry->nexthop_neighs_list_node,
3814 &mlxsw_sp->router->nexthop_neighs_list);
3816 nh->neigh_entry = neigh_entry;
3817 list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
3818 read_lock_bh(&n->lock);
3819 nud_state = n->nud_state;
3821 read_unlock_bh(&n->lock);
3822 __mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID && !dead));
3826 err_neigh_entry_create:
3831 static void mlxsw_sp_nexthop_neigh_fini(struct mlxsw_sp *mlxsw_sp,
3832 struct mlxsw_sp_nexthop *nh)
3834 struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
3835 struct neighbour *n;
3839 n = neigh_entry->key.n;
3841 __mlxsw_sp_nexthop_neigh_update(nh, true);
3842 list_del(&nh->neigh_list_node);
3843 nh->neigh_entry = NULL;
3845 /* If that is the last nexthop connected to that neigh, remove from
3846 * nexthop_neighs_list
3848 if (list_empty(&neigh_entry->nexthop_list))
3849 list_del(&neigh_entry->nexthop_neighs_list_node);
3851 if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
3852 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
3857 static bool mlxsw_sp_ipip_netdev_ul_up(struct net_device *ol_dev)
3859 struct net_device *ul_dev;
3863 ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
3864 is_up = ul_dev ? (ul_dev->flags & IFF_UP) : true;
3870 static void mlxsw_sp_nexthop_ipip_init(struct mlxsw_sp *mlxsw_sp,
3871 struct mlxsw_sp_nexthop *nh,
3872 struct mlxsw_sp_ipip_entry *ipip_entry)
3876 if (!nh->nhgi->gateway || nh->ipip_entry)
3879 nh->ipip_entry = ipip_entry;
3880 removing = !mlxsw_sp_ipip_netdev_ul_up(ipip_entry->ol_dev);
3881 __mlxsw_sp_nexthop_neigh_update(nh, removing);
3882 mlxsw_sp_nexthop_rif_init(nh, &ipip_entry->ol_lb->common);
3885 static void mlxsw_sp_nexthop_ipip_fini(struct mlxsw_sp *mlxsw_sp,
3886 struct mlxsw_sp_nexthop *nh)
3888 struct mlxsw_sp_ipip_entry *ipip_entry = nh->ipip_entry;
3893 __mlxsw_sp_nexthop_neigh_update(nh, true);
3894 nh->ipip_entry = NULL;
3897 static bool mlxsw_sp_nexthop4_ipip_type(const struct mlxsw_sp *mlxsw_sp,
3898 const struct fib_nh *fib_nh,
3899 enum mlxsw_sp_ipip_type *p_ipipt)
3901 struct net_device *dev = fib_nh->fib_nh_dev;
3904 fib_nh->nh_parent->fib_type == RTN_UNICAST &&
3905 mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, p_ipipt);
3908 static int mlxsw_sp_nexthop_type_init(struct mlxsw_sp *mlxsw_sp,
3909 struct mlxsw_sp_nexthop *nh,
3910 const struct net_device *dev)
3912 const struct mlxsw_sp_ipip_ops *ipip_ops;
3913 struct mlxsw_sp_ipip_entry *ipip_entry;
3914 struct mlxsw_sp_rif *rif;
3917 ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
3919 ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
3920 if (ipip_ops->can_offload(mlxsw_sp, dev)) {
3921 nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
3922 mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
3927 nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;
3928 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
3932 mlxsw_sp_nexthop_rif_init(nh, rif);
3933 err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
3935 goto err_neigh_init;
3940 mlxsw_sp_nexthop_rif_fini(nh);
3944 static void mlxsw_sp_nexthop_type_fini(struct mlxsw_sp *mlxsw_sp,
3945 struct mlxsw_sp_nexthop *nh)
3948 case MLXSW_SP_NEXTHOP_TYPE_ETH:
3949 mlxsw_sp_nexthop_neigh_fini(mlxsw_sp, nh);
3950 mlxsw_sp_nexthop_rif_fini(nh);
3952 case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3953 mlxsw_sp_nexthop_rif_fini(nh);
3954 mlxsw_sp_nexthop_ipip_fini(mlxsw_sp, nh);
3959 static int mlxsw_sp_nexthop4_init(struct mlxsw_sp *mlxsw_sp,
3960 struct mlxsw_sp_nexthop_group *nh_grp,
3961 struct mlxsw_sp_nexthop *nh,
3962 struct fib_nh *fib_nh)
3964 struct net_device *dev = fib_nh->fib_nh_dev;
3965 struct in_device *in_dev;
3968 nh->nhgi = nh_grp->nhgi;
3969 nh->key.fib_nh = fib_nh;
3970 #ifdef CONFIG_IP_ROUTE_MULTIPATH
3971 nh->nh_weight = fib_nh->fib_nh_weight;
3975 memcpy(&nh->gw_addr, &fib_nh->fib_nh_gw4, sizeof(fib_nh->fib_nh_gw4));
3976 nh->neigh_tbl = &arp_tbl;
3977 err = mlxsw_sp_nexthop_insert(mlxsw_sp, nh);
3981 mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
3982 list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
3986 nh->ifindex = dev->ifindex;
3989 in_dev = __in_dev_get_rcu(dev);
3990 if (in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
3991 fib_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
3997 err = mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, dev);
3999 goto err_nexthop_neigh_init;
4003 err_nexthop_neigh_init:
4004 mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
4008 static void mlxsw_sp_nexthop4_fini(struct mlxsw_sp *mlxsw_sp,
4009 struct mlxsw_sp_nexthop *nh)
4011 mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4012 list_del(&nh->router_list_node);
4013 mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
4014 mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
4017 static void mlxsw_sp_nexthop4_event(struct mlxsw_sp *mlxsw_sp,
4018 unsigned long event, struct fib_nh *fib_nh)
4020 struct mlxsw_sp_nexthop_key key;
4021 struct mlxsw_sp_nexthop *nh;
4023 if (mlxsw_sp->router->aborted)
4026 key.fib_nh = fib_nh;
4027 nh = mlxsw_sp_nexthop_lookup(mlxsw_sp, key);
4032 case FIB_EVENT_NH_ADD:
4033 mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, fib_nh->fib_nh_dev);
4035 case FIB_EVENT_NH_DEL:
4036 mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4040 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4043 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
4044 struct mlxsw_sp_rif *rif)
4046 struct mlxsw_sp_nexthop *nh;
4049 list_for_each_entry(nh, &rif->nexthop_list, rif_list_node) {
4051 case MLXSW_SP_NEXTHOP_TYPE_ETH:
4054 case MLXSW_SP_NEXTHOP_TYPE_IPIP:
4055 removing = !mlxsw_sp_ipip_netdev_ul_up(rif->dev);
4062 __mlxsw_sp_nexthop_neigh_update(nh, removing);
4063 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4067 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
4068 struct mlxsw_sp_rif *old_rif,
4069 struct mlxsw_sp_rif *new_rif)
4071 struct mlxsw_sp_nexthop *nh;
4073 list_splice_init(&old_rif->nexthop_list, &new_rif->nexthop_list);
4074 list_for_each_entry(nh, &new_rif->nexthop_list, rif_list_node)
4076 mlxsw_sp_nexthop_rif_update(mlxsw_sp, new_rif);
4079 static void mlxsw_sp_nexthop_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
4080 struct mlxsw_sp_rif *rif)
4082 struct mlxsw_sp_nexthop *nh, *tmp;
4084 list_for_each_entry_safe(nh, tmp, &rif->nexthop_list, rif_list_node) {
4085 mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4086 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4090 static bool mlxsw_sp_fi_is_gateway(const struct mlxsw_sp *mlxsw_sp,
4091 struct fib_info *fi)
4093 const struct fib_nh *nh = fib_info_nh(fi, 0);
4095 return nh->fib_nh_scope == RT_SCOPE_LINK ||
4096 mlxsw_sp_nexthop4_ipip_type(mlxsw_sp, nh, NULL);
4100 mlxsw_sp_nexthop4_group_info_init(struct mlxsw_sp *mlxsw_sp,
4101 struct mlxsw_sp_nexthop_group *nh_grp)
4103 unsigned int nhs = fib_info_num_path(nh_grp->ipv4.fi);
4104 struct mlxsw_sp_nexthop_group_info *nhgi;
4105 struct mlxsw_sp_nexthop *nh;
4108 nhgi = kzalloc(struct_size(nhgi, nexthops, nhs), GFP_KERNEL);
4111 nh_grp->nhgi = nhgi;
4112 nhgi->nh_grp = nh_grp;
4113 nhgi->gateway = mlxsw_sp_fi_is_gateway(mlxsw_sp, nh_grp->ipv4.fi);
4115 for (i = 0; i < nhgi->count; i++) {
4116 struct fib_nh *fib_nh;
4118 nh = &nhgi->nexthops[i];
4119 fib_nh = fib_info_nh(nh_grp->ipv4.fi, i);
4120 err = mlxsw_sp_nexthop4_init(mlxsw_sp, nh_grp, nh, fib_nh);
4122 goto err_nexthop4_init;
4124 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4129 for (i--; i >= 0; i--) {
4130 nh = &nhgi->nexthops[i];
4131 mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
4138 mlxsw_sp_nexthop4_group_info_fini(struct mlxsw_sp *mlxsw_sp,
4139 struct mlxsw_sp_nexthop_group *nh_grp)
4141 struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
4144 for (i = nhgi->count - 1; i >= 0; i--) {
4145 struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
4147 mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
4149 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4150 WARN_ON_ONCE(nhgi->adj_index_valid);
4154 static struct mlxsw_sp_nexthop_group *
4155 mlxsw_sp_nexthop4_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
4157 struct mlxsw_sp_nexthop_group *nh_grp;
4160 nh_grp = kzalloc(sizeof(*nh_grp), GFP_KERNEL);
4162 return ERR_PTR(-ENOMEM);
4163 INIT_LIST_HEAD(&nh_grp->fib_list);
4164 nh_grp->type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4;
4165 nh_grp->ipv4.fi = fi;
4168 err = mlxsw_sp_nexthop4_group_info_init(mlxsw_sp, nh_grp);
4170 goto err_nexthop_group_info_init;
4172 err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
4174 goto err_nexthop_group_insert;
4178 err_nexthop_group_insert:
4179 mlxsw_sp_nexthop4_group_info_fini(mlxsw_sp, nh_grp);
4180 err_nexthop_group_info_init:
4183 return ERR_PTR(err);
4187 mlxsw_sp_nexthop4_group_destroy(struct mlxsw_sp *mlxsw_sp,
4188 struct mlxsw_sp_nexthop_group *nh_grp)
4190 mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
4191 mlxsw_sp_nexthop4_group_info_fini(mlxsw_sp, nh_grp);
4192 fib_info_put(nh_grp->ipv4.fi);
4196 static int mlxsw_sp_nexthop4_group_get(struct mlxsw_sp *mlxsw_sp,
4197 struct mlxsw_sp_fib_entry *fib_entry,
4198 struct fib_info *fi)
4200 struct mlxsw_sp_nexthop_group *nh_grp;
4202 nh_grp = mlxsw_sp_nexthop4_group_lookup(mlxsw_sp, fi);
4204 nh_grp = mlxsw_sp_nexthop4_group_create(mlxsw_sp, fi);
4206 return PTR_ERR(nh_grp);
4208 list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
4209 fib_entry->nh_group = nh_grp;
4213 static void mlxsw_sp_nexthop4_group_put(struct mlxsw_sp *mlxsw_sp,
4214 struct mlxsw_sp_fib_entry *fib_entry)
4216 struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
4218 list_del(&fib_entry->nexthop_group_node);
4219 if (!list_empty(&nh_grp->fib_list))
4221 mlxsw_sp_nexthop4_group_destroy(mlxsw_sp, nh_grp);
4225 mlxsw_sp_fib4_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
4227 struct mlxsw_sp_fib4_entry *fib4_entry;
4229 fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4231 return !fib4_entry->tos;
4235 mlxsw_sp_fib_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
4237 struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
4239 switch (fib_entry->fib_node->fib->proto) {
4240 case MLXSW_SP_L3_PROTO_IPV4:
4241 if (!mlxsw_sp_fib4_entry_should_offload(fib_entry))
4244 case MLXSW_SP_L3_PROTO_IPV6:
4248 switch (fib_entry->type) {
4249 case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
4250 return !!nh_group->nhgi->adj_index_valid;
4251 case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
4252 return !!nh_group->nhgi->nh_rif;
4253 case MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE:
4254 case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4255 case MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP:
4262 static struct mlxsw_sp_nexthop *
4263 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
4264 const struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
4268 for (i = 0; i < nh_grp->nhgi->count; i++) {
4269 struct mlxsw_sp_nexthop *nh = &nh_grp->nhgi->nexthops[i];
4270 struct fib6_info *rt = mlxsw_sp_rt6->rt;
4272 if (nh->rif && nh->rif->dev == rt->fib6_nh->fib_nh_dev &&
4273 ipv6_addr_equal((const struct in6_addr *) &nh->gw_addr,
4274 &rt->fib6_nh->fib_nh_gw6))
4283 mlxsw_sp_fib4_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4284 struct mlxsw_sp_fib_entry *fib_entry)
4286 u32 *p_dst = (u32 *) fib_entry->fib_node->key.addr;
4287 int dst_len = fib_entry->fib_node->key.prefix_len;
4288 struct mlxsw_sp_fib4_entry *fib4_entry;
4289 struct fib_rt_info fri;
4290 bool should_offload;
4292 should_offload = mlxsw_sp_fib_entry_should_offload(fib_entry);
4293 fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4295 fri.fi = fib4_entry->fi;
4296 fri.tb_id = fib4_entry->tb_id;
4297 fri.dst = cpu_to_be32(*p_dst);
4298 fri.dst_len = dst_len;
4299 fri.tos = fib4_entry->tos;
4300 fri.type = fib4_entry->type;
4301 fri.offload = should_offload;
4302 fri.trap = !should_offload;
4303 fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
4307 mlxsw_sp_fib4_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4308 struct mlxsw_sp_fib_entry *fib_entry)
4310 u32 *p_dst = (u32 *) fib_entry->fib_node->key.addr;
4311 int dst_len = fib_entry->fib_node->key.prefix_len;
4312 struct mlxsw_sp_fib4_entry *fib4_entry;
4313 struct fib_rt_info fri;
4315 fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4317 fri.fi = fib4_entry->fi;
4318 fri.tb_id = fib4_entry->tb_id;
4319 fri.dst = cpu_to_be32(*p_dst);
4320 fri.dst_len = dst_len;
4321 fri.tos = fib4_entry->tos;
4322 fri.type = fib4_entry->type;
4323 fri.offload = false;
4325 fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
4329 mlxsw_sp_fib6_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4330 struct mlxsw_sp_fib_entry *fib_entry)
4332 struct mlxsw_sp_fib6_entry *fib6_entry;
4333 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4334 bool should_offload;
4336 should_offload = mlxsw_sp_fib_entry_should_offload(fib_entry);
4338 /* In IPv6 a multipath route is represented using multiple routes, so
4339 * we need to set the flags on all of them.
4341 fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
4343 list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list)
4344 fib6_info_hw_flags_set(mlxsw_sp_rt6->rt, should_offload,
4349 mlxsw_sp_fib6_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4350 struct mlxsw_sp_fib_entry *fib_entry)
4352 struct mlxsw_sp_fib6_entry *fib6_entry;
4353 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4355 fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
4357 list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list)
4358 fib6_info_hw_flags_set(mlxsw_sp_rt6->rt, false, false);
4362 mlxsw_sp_fib_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4363 struct mlxsw_sp_fib_entry *fib_entry)
4365 switch (fib_entry->fib_node->fib->proto) {
4366 case MLXSW_SP_L3_PROTO_IPV4:
4367 mlxsw_sp_fib4_entry_hw_flags_set(mlxsw_sp, fib_entry);
4369 case MLXSW_SP_L3_PROTO_IPV6:
4370 mlxsw_sp_fib6_entry_hw_flags_set(mlxsw_sp, fib_entry);
4376 mlxsw_sp_fib_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4377 struct mlxsw_sp_fib_entry *fib_entry)
4379 switch (fib_entry->fib_node->fib->proto) {
4380 case MLXSW_SP_L3_PROTO_IPV4:
4381 mlxsw_sp_fib4_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4383 case MLXSW_SP_L3_PROTO_IPV6:
4384 mlxsw_sp_fib6_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4390 mlxsw_sp_fib_entry_hw_flags_refresh(struct mlxsw_sp *mlxsw_sp,
4391 struct mlxsw_sp_fib_entry *fib_entry,
4392 enum mlxsw_sp_fib_entry_op op)
4395 case MLXSW_SP_FIB_ENTRY_OP_WRITE:
4396 case MLXSW_SP_FIB_ENTRY_OP_UPDATE:
4397 mlxsw_sp_fib_entry_hw_flags_set(mlxsw_sp, fib_entry);
4399 case MLXSW_SP_FIB_ENTRY_OP_DELETE:
4400 mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4407 struct mlxsw_sp_fib_entry_op_ctx_basic {
4408 char ralue_pl[MLXSW_REG_RALUE_LEN];
4412 mlxsw_sp_router_ll_basic_fib_entry_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4413 enum mlxsw_sp_l3proto proto,
4414 enum mlxsw_sp_fib_entry_op op,
4415 u16 virtual_router, u8 prefix_len,
4416 unsigned char *addr,
4417 struct mlxsw_sp_fib_entry_priv *priv)
4419 struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4420 enum mlxsw_reg_ralxx_protocol ralxx_proto;
4421 char *ralue_pl = op_ctx_basic->ralue_pl;
4422 enum mlxsw_reg_ralue_op ralue_op;
4424 ralxx_proto = (enum mlxsw_reg_ralxx_protocol) proto;
4427 case MLXSW_SP_FIB_ENTRY_OP_WRITE:
4428 case MLXSW_SP_FIB_ENTRY_OP_UPDATE:
4429 ralue_op = MLXSW_REG_RALUE_OP_WRITE_WRITE;
4431 case MLXSW_SP_FIB_ENTRY_OP_DELETE:
4432 ralue_op = MLXSW_REG_RALUE_OP_WRITE_DELETE;
4440 case MLXSW_SP_L3_PROTO_IPV4:
4441 mlxsw_reg_ralue_pack4(ralue_pl, ralxx_proto, ralue_op,
4442 virtual_router, prefix_len, (u32 *) addr);
4444 case MLXSW_SP_L3_PROTO_IPV6:
4445 mlxsw_reg_ralue_pack6(ralue_pl, ralxx_proto, ralue_op,
4446 virtual_router, prefix_len, addr);
4452 mlxsw_sp_router_ll_basic_fib_entry_act_remote_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4453 enum mlxsw_reg_ralue_trap_action trap_action,
4454 u16 trap_id, u32 adjacency_index, u16 ecmp_size)
4456 struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4458 mlxsw_reg_ralue_act_remote_pack(op_ctx_basic->ralue_pl, trap_action,
4459 trap_id, adjacency_index, ecmp_size);
4463 mlxsw_sp_router_ll_basic_fib_entry_act_local_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4464 enum mlxsw_reg_ralue_trap_action trap_action,
4465 u16 trap_id, u16 local_erif)
4467 struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4469 mlxsw_reg_ralue_act_local_pack(op_ctx_basic->ralue_pl, trap_action,
4470 trap_id, local_erif);
4474 mlxsw_sp_router_ll_basic_fib_entry_act_ip2me_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx)
4476 struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4478 mlxsw_reg_ralue_act_ip2me_pack(op_ctx_basic->ralue_pl);
4482 mlxsw_sp_router_ll_basic_fib_entry_act_ip2me_tun_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4485 struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4487 mlxsw_reg_ralue_act_ip2me_tun_pack(op_ctx_basic->ralue_pl, tunnel_ptr);
4491 mlxsw_sp_router_ll_basic_fib_entry_commit(struct mlxsw_sp *mlxsw_sp,
4492 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4493 bool *postponed_for_bulk)
4495 struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4497 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue),
4498 op_ctx_basic->ralue_pl);
4502 mlxsw_sp_router_ll_basic_fib_entry_is_committed(struct mlxsw_sp_fib_entry_priv *priv)
4507 static void mlxsw_sp_fib_entry_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4508 struct mlxsw_sp_fib_entry *fib_entry,
4509 enum mlxsw_sp_fib_entry_op op)
4511 struct mlxsw_sp_fib *fib = fib_entry->fib_node->fib;
4513 mlxsw_sp_fib_entry_op_ctx_priv_hold(op_ctx, fib_entry->priv);
4514 fib->ll_ops->fib_entry_pack(op_ctx, fib->proto, op, fib->vr->id,
4515 fib_entry->fib_node->key.prefix_len,
4516 fib_entry->fib_node->key.addr,
4520 int mlxsw_sp_fib_entry_commit(struct mlxsw_sp *mlxsw_sp,
4521 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4522 const struct mlxsw_sp_router_ll_ops *ll_ops)
4524 bool postponed_for_bulk = false;
4527 err = ll_ops->fib_entry_commit(mlxsw_sp, op_ctx, &postponed_for_bulk);
4528 if (!postponed_for_bulk)
4529 mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
4533 static int mlxsw_sp_adj_discard_write(struct mlxsw_sp *mlxsw_sp, u16 rif_index)
4535 enum mlxsw_reg_ratr_trap_action trap_action;
4536 char ratr_pl[MLXSW_REG_RATR_LEN];
4539 if (mlxsw_sp->router->adj_discard_index_valid)
4542 err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4543 &mlxsw_sp->router->adj_discard_index);
4547 trap_action = MLXSW_REG_RATR_TRAP_ACTION_DISCARD_ERRORS;
4548 mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY, true,
4549 MLXSW_REG_RATR_TYPE_ETHERNET,
4550 mlxsw_sp->router->adj_discard_index, rif_index);
4551 mlxsw_reg_ratr_trap_action_set(ratr_pl, trap_action);
4552 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
4554 goto err_ratr_write;
4556 mlxsw_sp->router->adj_discard_index_valid = true;
4561 mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4562 mlxsw_sp->router->adj_discard_index);
4566 static int mlxsw_sp_fib_entry_op_remote(struct mlxsw_sp *mlxsw_sp,
4567 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4568 struct mlxsw_sp_fib_entry *fib_entry,
4569 enum mlxsw_sp_fib_entry_op op)
4571 const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4572 struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
4573 struct mlxsw_sp_nexthop_group_info *nhgi = nh_group->nhgi;
4574 enum mlxsw_reg_ralue_trap_action trap_action;
4576 u32 adjacency_index = 0;
4580 /* In case the nexthop group adjacency index is valid, use it
4581 * with provided ECMP size. Otherwise, setup trap and pass
4582 * traffic to kernel.
4584 if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
4585 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4586 adjacency_index = nhgi->adj_index;
4587 ecmp_size = nhgi->ecmp_size;
4588 } else if (!nhgi->adj_index_valid && nhgi->count && nhgi->nh_rif) {
4589 err = mlxsw_sp_adj_discard_write(mlxsw_sp,
4590 nhgi->nh_rif->rif_index);
4593 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4594 adjacency_index = mlxsw_sp->router->adj_discard_index;
4597 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4598 trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
4601 mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4602 ll_ops->fib_entry_act_remote_pack(op_ctx, trap_action, trap_id,
4603 adjacency_index, ecmp_size);
4604 return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4607 static int mlxsw_sp_fib_entry_op_local(struct mlxsw_sp *mlxsw_sp,
4608 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4609 struct mlxsw_sp_fib_entry *fib_entry,
4610 enum mlxsw_sp_fib_entry_op op)
4612 const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4613 struct mlxsw_sp_rif *rif = fib_entry->nh_group->nhgi->nh_rif;
4614 enum mlxsw_reg_ralue_trap_action trap_action;
4618 if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
4619 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4620 rif_index = rif->rif_index;
4622 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4623 trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
4626 mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4627 ll_ops->fib_entry_act_local_pack(op_ctx, trap_action, trap_id, rif_index);
4628 return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4631 static int mlxsw_sp_fib_entry_op_trap(struct mlxsw_sp *mlxsw_sp,
4632 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4633 struct mlxsw_sp_fib_entry *fib_entry,
4634 enum mlxsw_sp_fib_entry_op op)
4636 const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4638 mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4639 ll_ops->fib_entry_act_ip2me_pack(op_ctx);
4640 return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4643 static int mlxsw_sp_fib_entry_op_blackhole(struct mlxsw_sp *mlxsw_sp,
4644 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4645 struct mlxsw_sp_fib_entry *fib_entry,
4646 enum mlxsw_sp_fib_entry_op op)
4648 const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4649 enum mlxsw_reg_ralue_trap_action trap_action;
4651 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_DISCARD_ERROR;
4652 mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4653 ll_ops->fib_entry_act_local_pack(op_ctx, trap_action, 0, 0);
4654 return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4658 mlxsw_sp_fib_entry_op_unreachable(struct mlxsw_sp *mlxsw_sp,
4659 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4660 struct mlxsw_sp_fib_entry *fib_entry,
4661 enum mlxsw_sp_fib_entry_op op)
4663 const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4664 enum mlxsw_reg_ralue_trap_action trap_action;
4667 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4668 trap_id = MLXSW_TRAP_ID_RTR_INGRESS1;
4670 mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4671 ll_ops->fib_entry_act_local_pack(op_ctx, trap_action, trap_id, 0);
4672 return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4676 mlxsw_sp_fib_entry_op_ipip_decap(struct mlxsw_sp *mlxsw_sp,
4677 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4678 struct mlxsw_sp_fib_entry *fib_entry,
4679 enum mlxsw_sp_fib_entry_op op)
4681 const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4682 struct mlxsw_sp_ipip_entry *ipip_entry = fib_entry->decap.ipip_entry;
4683 const struct mlxsw_sp_ipip_ops *ipip_ops;
4685 if (WARN_ON(!ipip_entry))
4688 ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
4689 return ipip_ops->fib_entry_op(mlxsw_sp, ll_ops, op_ctx, ipip_entry, op,
4690 fib_entry->decap.tunnel_index, fib_entry->priv);
4693 static int mlxsw_sp_fib_entry_op_nve_decap(struct mlxsw_sp *mlxsw_sp,
4694 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4695 struct mlxsw_sp_fib_entry *fib_entry,
4696 enum mlxsw_sp_fib_entry_op op)
4698 const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4700 mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4701 ll_ops->fib_entry_act_ip2me_tun_pack(op_ctx,
4702 fib_entry->decap.tunnel_index);
4703 return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4706 static int __mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
4707 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4708 struct mlxsw_sp_fib_entry *fib_entry,
4709 enum mlxsw_sp_fib_entry_op op)
4711 switch (fib_entry->type) {
4712 case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
4713 return mlxsw_sp_fib_entry_op_remote(mlxsw_sp, op_ctx, fib_entry, op);
4714 case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
4715 return mlxsw_sp_fib_entry_op_local(mlxsw_sp, op_ctx, fib_entry, op);
4716 case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
4717 return mlxsw_sp_fib_entry_op_trap(mlxsw_sp, op_ctx, fib_entry, op);
4718 case MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE:
4719 return mlxsw_sp_fib_entry_op_blackhole(mlxsw_sp, op_ctx, fib_entry, op);
4720 case MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE:
4721 return mlxsw_sp_fib_entry_op_unreachable(mlxsw_sp, op_ctx, fib_entry, op);
4722 case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4723 return mlxsw_sp_fib_entry_op_ipip_decap(mlxsw_sp, op_ctx, fib_entry, op);
4724 case MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP:
4725 return mlxsw_sp_fib_entry_op_nve_decap(mlxsw_sp, op_ctx, fib_entry, op);
4730 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
4731 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4732 struct mlxsw_sp_fib_entry *fib_entry,
4733 enum mlxsw_sp_fib_entry_op op)
4735 int err = __mlxsw_sp_fib_entry_op(mlxsw_sp, op_ctx, fib_entry, op);
4740 mlxsw_sp_fib_entry_hw_flags_refresh(mlxsw_sp, fib_entry, op);
4745 static int __mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
4746 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4747 struct mlxsw_sp_fib_entry *fib_entry,
4750 return mlxsw_sp_fib_entry_op(mlxsw_sp, op_ctx, fib_entry,
4751 is_new ? MLXSW_SP_FIB_ENTRY_OP_WRITE :
4752 MLXSW_SP_FIB_ENTRY_OP_UPDATE);
4755 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
4756 struct mlxsw_sp_fib_entry *fib_entry)
4758 struct mlxsw_sp_fib_entry_op_ctx *op_ctx = mlxsw_sp->router->ll_op_ctx;
4760 mlxsw_sp_fib_entry_op_ctx_clear(op_ctx);
4761 return __mlxsw_sp_fib_entry_update(mlxsw_sp, op_ctx, fib_entry, false);
4764 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
4765 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4766 struct mlxsw_sp_fib_entry *fib_entry)
4768 const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4770 if (!ll_ops->fib_entry_is_committed(fib_entry->priv))
4772 return mlxsw_sp_fib_entry_op(mlxsw_sp, op_ctx, fib_entry,
4773 MLXSW_SP_FIB_ENTRY_OP_DELETE);
4777 mlxsw_sp_fib4_entry_type_set(struct mlxsw_sp *mlxsw_sp,
4778 const struct fib_entry_notifier_info *fen_info,
4779 struct mlxsw_sp_fib_entry *fib_entry)
4781 struct net_device *dev = fib_info_nh(fen_info->fi, 0)->fib_nh_dev;
4782 union mlxsw_sp_l3addr dip = { .addr4 = htonl(fen_info->dst) };
4783 struct mlxsw_sp_router *router = mlxsw_sp->router;
4784 u32 tb_id = mlxsw_sp_fix_tb_id(fen_info->tb_id);
4785 struct mlxsw_sp_ipip_entry *ipip_entry;
4786 struct fib_info *fi = fen_info->fi;
4788 switch (fen_info->type) {
4790 ipip_entry = mlxsw_sp_ipip_entry_find_by_decap(mlxsw_sp, dev->ifindex,
4791 MLXSW_SP_L3_PROTO_IPV4, dip);
4792 if (ipip_entry && ipip_entry->ol_dev->flags & IFF_UP) {
4793 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
4794 return mlxsw_sp_fib_entry_decap_init(mlxsw_sp,
4798 if (mlxsw_sp_router_nve_is_decap(mlxsw_sp, tb_id,
4799 MLXSW_SP_L3_PROTO_IPV4,
4803 tunnel_index = router->nve_decap_config.tunnel_index;
4804 fib_entry->decap.tunnel_index = tunnel_index;
4805 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
4810 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
4813 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE;
4815 case RTN_UNREACHABLE:
4817 /* Packets hitting these routes need to be trapped, but
4818 * can do so with a lower priority than packets directed
4819 * at the host, so use action type local instead of trap.
4821 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE;
4824 if (mlxsw_sp_fi_is_gateway(mlxsw_sp, fi))
4825 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
4827 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
4835 mlxsw_sp_fib4_entry_type_unset(struct mlxsw_sp *mlxsw_sp,
4836 struct mlxsw_sp_fib_entry *fib_entry)
4838 switch (fib_entry->type) {
4839 case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4840 mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
4847 static struct mlxsw_sp_fib4_entry *
4848 mlxsw_sp_fib4_entry_create(struct mlxsw_sp *mlxsw_sp,
4849 struct mlxsw_sp_fib_node *fib_node,
4850 const struct fib_entry_notifier_info *fen_info)
4852 struct mlxsw_sp_fib4_entry *fib4_entry;
4853 struct mlxsw_sp_fib_entry *fib_entry;
4856 fib4_entry = kzalloc(sizeof(*fib4_entry), GFP_KERNEL);
4858 return ERR_PTR(-ENOMEM);
4859 fib_entry = &fib4_entry->common;
4861 fib_entry->priv = mlxsw_sp_fib_entry_priv_create(fib_node->fib->ll_ops);
4862 if (IS_ERR(fib_entry->priv)) {
4863 err = PTR_ERR(fib_entry->priv);
4864 goto err_fib_entry_priv_create;
4867 err = mlxsw_sp_fib4_entry_type_set(mlxsw_sp, fen_info, fib_entry);
4869 goto err_fib4_entry_type_set;
4871 err = mlxsw_sp_nexthop4_group_get(mlxsw_sp, fib_entry, fen_info->fi);
4873 goto err_nexthop4_group_get;
4875 fib4_entry->fi = fen_info->fi;
4876 fib_info_hold(fib4_entry->fi);
4877 fib4_entry->tb_id = fen_info->tb_id;
4878 fib4_entry->type = fen_info->type;
4879 fib4_entry->tos = fen_info->tos;
4881 fib_entry->fib_node = fib_node;
4885 err_nexthop4_group_get:
4886 mlxsw_sp_fib4_entry_type_unset(mlxsw_sp, fib_entry);
4887 err_fib4_entry_type_set:
4888 mlxsw_sp_fib_entry_priv_put(fib_entry->priv);
4889 err_fib_entry_priv_create:
4891 return ERR_PTR(err);
4894 static void mlxsw_sp_fib4_entry_destroy(struct mlxsw_sp *mlxsw_sp,
4895 struct mlxsw_sp_fib4_entry *fib4_entry)
4897 fib_info_put(fib4_entry->fi);
4898 mlxsw_sp_nexthop4_group_put(mlxsw_sp, &fib4_entry->common);
4899 mlxsw_sp_fib4_entry_type_unset(mlxsw_sp, &fib4_entry->common);
4900 mlxsw_sp_fib_entry_priv_put(fib4_entry->common.priv);
4904 static struct mlxsw_sp_fib4_entry *
4905 mlxsw_sp_fib4_entry_lookup(struct mlxsw_sp *mlxsw_sp,
4906 const struct fib_entry_notifier_info *fen_info)
4908 struct mlxsw_sp_fib4_entry *fib4_entry;
4909 struct mlxsw_sp_fib_node *fib_node;
4910 struct mlxsw_sp_fib *fib;
4911 struct mlxsw_sp_vr *vr;
4913 vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id);
4916 fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV4);
4918 fib_node = mlxsw_sp_fib_node_lookup(fib, &fen_info->dst,
4919 sizeof(fen_info->dst),
4924 fib4_entry = container_of(fib_node->fib_entry,
4925 struct mlxsw_sp_fib4_entry, common);
4926 if (fib4_entry->tb_id == fen_info->tb_id &&
4927 fib4_entry->tos == fen_info->tos &&
4928 fib4_entry->type == fen_info->type &&
4929 fib4_entry->fi == fen_info->fi)
4935 static const struct rhashtable_params mlxsw_sp_fib_ht_params = {
4936 .key_offset = offsetof(struct mlxsw_sp_fib_node, key),
4937 .head_offset = offsetof(struct mlxsw_sp_fib_node, ht_node),
4938 .key_len = sizeof(struct mlxsw_sp_fib_key),
4939 .automatic_shrinking = true,
4942 static int mlxsw_sp_fib_node_insert(struct mlxsw_sp_fib *fib,
4943 struct mlxsw_sp_fib_node *fib_node)
4945 return rhashtable_insert_fast(&fib->ht, &fib_node->ht_node,
4946 mlxsw_sp_fib_ht_params);
4949 static void mlxsw_sp_fib_node_remove(struct mlxsw_sp_fib *fib,
4950 struct mlxsw_sp_fib_node *fib_node)
4952 rhashtable_remove_fast(&fib->ht, &fib_node->ht_node,
4953 mlxsw_sp_fib_ht_params);
4956 static struct mlxsw_sp_fib_node *
4957 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
4958 size_t addr_len, unsigned char prefix_len)
4960 struct mlxsw_sp_fib_key key;
4962 memset(&key, 0, sizeof(key));
4963 memcpy(key.addr, addr, addr_len);
4964 key.prefix_len = prefix_len;
4965 return rhashtable_lookup_fast(&fib->ht, &key, mlxsw_sp_fib_ht_params);
4968 static struct mlxsw_sp_fib_node *
4969 mlxsw_sp_fib_node_create(struct mlxsw_sp_fib *fib, const void *addr,
4970 size_t addr_len, unsigned char prefix_len)
4972 struct mlxsw_sp_fib_node *fib_node;
4974 fib_node = kzalloc(sizeof(*fib_node), GFP_KERNEL);
4978 list_add(&fib_node->list, &fib->node_list);
4979 memcpy(fib_node->key.addr, addr, addr_len);
4980 fib_node->key.prefix_len = prefix_len;
4985 static void mlxsw_sp_fib_node_destroy(struct mlxsw_sp_fib_node *fib_node)
4987 list_del(&fib_node->list);
4991 static int mlxsw_sp_fib_lpm_tree_link(struct mlxsw_sp *mlxsw_sp,
4992 struct mlxsw_sp_fib_node *fib_node)
4994 struct mlxsw_sp_prefix_usage req_prefix_usage;
4995 struct mlxsw_sp_fib *fib = fib_node->fib;
4996 struct mlxsw_sp_lpm_tree *lpm_tree;
4999 lpm_tree = mlxsw_sp->router->lpm.proto_trees[fib->proto];
5000 if (lpm_tree->prefix_ref_count[fib_node->key.prefix_len] != 0)
5003 mlxsw_sp_prefix_usage_cpy(&req_prefix_usage, &lpm_tree->prefix_usage);
5004 mlxsw_sp_prefix_usage_set(&req_prefix_usage, fib_node->key.prefix_len);
5005 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
5007 if (IS_ERR(lpm_tree))
5008 return PTR_ERR(lpm_tree);
5010 err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
5012 goto err_lpm_tree_replace;
5015 lpm_tree->prefix_ref_count[fib_node->key.prefix_len]++;
5018 err_lpm_tree_replace:
5019 mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
5023 static void mlxsw_sp_fib_lpm_tree_unlink(struct mlxsw_sp *mlxsw_sp,
5024 struct mlxsw_sp_fib_node *fib_node)
5026 struct mlxsw_sp_lpm_tree *lpm_tree = fib_node->fib->lpm_tree;
5027 struct mlxsw_sp_prefix_usage req_prefix_usage;
5028 struct mlxsw_sp_fib *fib = fib_node->fib;
5031 if (--lpm_tree->prefix_ref_count[fib_node->key.prefix_len] != 0)
5033 /* Try to construct a new LPM tree from the current prefix usage
5034 * minus the unused one. If we fail, continue using the old one.
5036 mlxsw_sp_prefix_usage_cpy(&req_prefix_usage, &lpm_tree->prefix_usage);
5037 mlxsw_sp_prefix_usage_clear(&req_prefix_usage,
5038 fib_node->key.prefix_len);
5039 lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
5041 if (IS_ERR(lpm_tree))
5044 err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
5046 goto err_lpm_tree_replace;
5050 err_lpm_tree_replace:
5051 mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
5054 static int mlxsw_sp_fib_node_init(struct mlxsw_sp *mlxsw_sp,
5055 struct mlxsw_sp_fib_node *fib_node,
5056 struct mlxsw_sp_fib *fib)
5060 err = mlxsw_sp_fib_node_insert(fib, fib_node);
5063 fib_node->fib = fib;
5065 err = mlxsw_sp_fib_lpm_tree_link(mlxsw_sp, fib_node);
5067 goto err_fib_lpm_tree_link;
5071 err_fib_lpm_tree_link:
5072 fib_node->fib = NULL;
5073 mlxsw_sp_fib_node_remove(fib, fib_node);
5077 static void mlxsw_sp_fib_node_fini(struct mlxsw_sp *mlxsw_sp,
5078 struct mlxsw_sp_fib_node *fib_node)
5080 struct mlxsw_sp_fib *fib = fib_node->fib;
5082 mlxsw_sp_fib_lpm_tree_unlink(mlxsw_sp, fib_node);
5083 fib_node->fib = NULL;
5084 mlxsw_sp_fib_node_remove(fib, fib_node);
5087 static struct mlxsw_sp_fib_node *
5088 mlxsw_sp_fib_node_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id, const void *addr,
5089 size_t addr_len, unsigned char prefix_len,
5090 enum mlxsw_sp_l3proto proto)
5092 struct mlxsw_sp_fib_node *fib_node;
5093 struct mlxsw_sp_fib *fib;
5094 struct mlxsw_sp_vr *vr;
5097 vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, NULL);
5099 return ERR_CAST(vr);
5100 fib = mlxsw_sp_vr_fib(vr, proto);
5102 fib_node = mlxsw_sp_fib_node_lookup(fib, addr, addr_len, prefix_len);
5106 fib_node = mlxsw_sp_fib_node_create(fib, addr, addr_len, prefix_len);
5109 goto err_fib_node_create;
5112 err = mlxsw_sp_fib_node_init(mlxsw_sp, fib_node, fib);
5114 goto err_fib_node_init;
5119 mlxsw_sp_fib_node_destroy(fib_node);
5120 err_fib_node_create:
5121 mlxsw_sp_vr_put(mlxsw_sp, vr);
5122 return ERR_PTR(err);
5125 static void mlxsw_sp_fib_node_put(struct mlxsw_sp *mlxsw_sp,
5126 struct mlxsw_sp_fib_node *fib_node)
5128 struct mlxsw_sp_vr *vr = fib_node->fib->vr;
5130 if (fib_node->fib_entry)
5132 mlxsw_sp_fib_node_fini(mlxsw_sp, fib_node);
5133 mlxsw_sp_fib_node_destroy(fib_node);
5134 mlxsw_sp_vr_put(mlxsw_sp, vr);
5137 static int mlxsw_sp_fib_node_entry_link(struct mlxsw_sp *mlxsw_sp,
5138 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5139 struct mlxsw_sp_fib_entry *fib_entry)
5141 struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
5142 bool is_new = !fib_node->fib_entry;
5145 fib_node->fib_entry = fib_entry;
5147 err = __mlxsw_sp_fib_entry_update(mlxsw_sp, op_ctx, fib_entry, is_new);
5149 goto err_fib_entry_update;
5153 err_fib_entry_update:
5154 fib_node->fib_entry = NULL;
5158 static int __mlxsw_sp_fib_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
5159 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5160 struct mlxsw_sp_fib_entry *fib_entry)
5162 struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
5165 err = mlxsw_sp_fib_entry_del(mlxsw_sp, op_ctx, fib_entry);
5166 fib_node->fib_entry = NULL;
5170 static void mlxsw_sp_fib_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
5171 struct mlxsw_sp_fib_entry *fib_entry)
5173 struct mlxsw_sp_fib_entry_op_ctx *op_ctx = mlxsw_sp->router->ll_op_ctx;
5175 mlxsw_sp_fib_entry_op_ctx_clear(op_ctx);
5176 __mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, op_ctx, fib_entry);
5179 static bool mlxsw_sp_fib4_allow_replace(struct mlxsw_sp_fib4_entry *fib4_entry)
5181 struct mlxsw_sp_fib_node *fib_node = fib4_entry->common.fib_node;
5182 struct mlxsw_sp_fib4_entry *fib4_replaced;
5184 if (!fib_node->fib_entry)
5187 fib4_replaced = container_of(fib_node->fib_entry,
5188 struct mlxsw_sp_fib4_entry, common);
5189 if (fib4_entry->tb_id == RT_TABLE_MAIN &&
5190 fib4_replaced->tb_id == RT_TABLE_LOCAL)
5197 mlxsw_sp_router_fib4_replace(struct mlxsw_sp *mlxsw_sp,
5198 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5199 const struct fib_entry_notifier_info *fen_info)
5201 struct mlxsw_sp_fib4_entry *fib4_entry, *fib4_replaced;
5202 struct mlxsw_sp_fib_entry *replaced;
5203 struct mlxsw_sp_fib_node *fib_node;
5206 if (mlxsw_sp->router->aborted)
5209 fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, fen_info->tb_id,
5210 &fen_info->dst, sizeof(fen_info->dst),
5212 MLXSW_SP_L3_PROTO_IPV4);
5213 if (IS_ERR(fib_node)) {
5214 dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB node\n");
5215 return PTR_ERR(fib_node);
5218 fib4_entry = mlxsw_sp_fib4_entry_create(mlxsw_sp, fib_node, fen_info);
5219 if (IS_ERR(fib4_entry)) {
5220 dev_warn(mlxsw_sp->bus_info->dev, "Failed to create FIB entry\n");
5221 err = PTR_ERR(fib4_entry);
5222 goto err_fib4_entry_create;
5225 if (!mlxsw_sp_fib4_allow_replace(fib4_entry)) {
5226 mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
5227 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5231 replaced = fib_node->fib_entry;
5232 err = mlxsw_sp_fib_node_entry_link(mlxsw_sp, op_ctx, &fib4_entry->common);
5234 dev_warn(mlxsw_sp->bus_info->dev, "Failed to link FIB entry to node\n");
5235 goto err_fib_node_entry_link;
5238 /* Nothing to replace */
5242 mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, replaced);
5243 fib4_replaced = container_of(replaced, struct mlxsw_sp_fib4_entry,
5245 mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_replaced);
5249 err_fib_node_entry_link:
5250 fib_node->fib_entry = replaced;
5251 mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
5252 err_fib4_entry_create:
5253 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5257 static int mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
5258 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5259 struct fib_entry_notifier_info *fen_info)
5261 struct mlxsw_sp_fib4_entry *fib4_entry;
5262 struct mlxsw_sp_fib_node *fib_node;
5265 if (mlxsw_sp->router->aborted)
5268 fib4_entry = mlxsw_sp_fib4_entry_lookup(mlxsw_sp, fen_info);
5271 fib_node = fib4_entry->common.fib_node;
5273 err = __mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, op_ctx, &fib4_entry->common);
5274 mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
5275 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5279 static bool mlxsw_sp_fib6_rt_should_ignore(const struct fib6_info *rt)
5281 /* Multicast routes aren't supported, so ignore them. Neighbour
5282 * Discovery packets are specifically trapped.
5284 if (ipv6_addr_type(&rt->fib6_dst.addr) & IPV6_ADDR_MULTICAST)
5287 /* Cloned routes are irrelevant in the forwarding path. */
5288 if (rt->fib6_flags & RTF_CACHE)
5294 static struct mlxsw_sp_rt6 *mlxsw_sp_rt6_create(struct fib6_info *rt)
5296 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5298 mlxsw_sp_rt6 = kzalloc(sizeof(*mlxsw_sp_rt6), GFP_KERNEL);
5300 return ERR_PTR(-ENOMEM);
5302 /* In case of route replace, replaced route is deleted with
5303 * no notification. Take reference to prevent accessing freed
5306 mlxsw_sp_rt6->rt = rt;
5309 return mlxsw_sp_rt6;
5312 #if IS_ENABLED(CONFIG_IPV6)
5313 static void mlxsw_sp_rt6_release(struct fib6_info *rt)
5315 fib6_info_release(rt);
5318 static void mlxsw_sp_rt6_release(struct fib6_info *rt)
5323 static void mlxsw_sp_rt6_destroy(struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
5325 struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
5327 fib6_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
5328 mlxsw_sp_rt6_release(mlxsw_sp_rt6->rt);
5329 kfree(mlxsw_sp_rt6);
5332 static struct fib6_info *
5333 mlxsw_sp_fib6_entry_rt(const struct mlxsw_sp_fib6_entry *fib6_entry)
5335 return list_first_entry(&fib6_entry->rt6_list, struct mlxsw_sp_rt6,
5339 static struct mlxsw_sp_rt6 *
5340 mlxsw_sp_fib6_entry_rt_find(const struct mlxsw_sp_fib6_entry *fib6_entry,
5341 const struct fib6_info *rt)
5343 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5345 list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
5346 if (mlxsw_sp_rt6->rt == rt)
5347 return mlxsw_sp_rt6;
5353 static bool mlxsw_sp_nexthop6_ipip_type(const struct mlxsw_sp *mlxsw_sp,
5354 const struct fib6_info *rt,
5355 enum mlxsw_sp_ipip_type *ret)
5357 return rt->fib6_nh->fib_nh_dev &&
5358 mlxsw_sp_netdev_ipip_type(mlxsw_sp, rt->fib6_nh->fib_nh_dev, ret);
5361 static int mlxsw_sp_nexthop6_init(struct mlxsw_sp *mlxsw_sp,
5362 struct mlxsw_sp_nexthop_group *nh_grp,
5363 struct mlxsw_sp_nexthop *nh,
5364 const struct fib6_info *rt)
5366 struct net_device *dev = rt->fib6_nh->fib_nh_dev;
5368 nh->nhgi = nh_grp->nhgi;
5369 nh->nh_weight = rt->fib6_nh->fib_nh_weight;
5370 memcpy(&nh->gw_addr, &rt->fib6_nh->fib_nh_gw6, sizeof(nh->gw_addr));
5371 #if IS_ENABLED(CONFIG_IPV6)
5372 nh->neigh_tbl = &nd_tbl;
5374 mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
5376 list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
5380 nh->ifindex = dev->ifindex;
5382 return mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, dev);
5385 static void mlxsw_sp_nexthop6_fini(struct mlxsw_sp *mlxsw_sp,
5386 struct mlxsw_sp_nexthop *nh)
5388 mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
5389 list_del(&nh->router_list_node);
5390 mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
5393 static bool mlxsw_sp_rt6_is_gateway(const struct mlxsw_sp *mlxsw_sp,
5394 const struct fib6_info *rt)
5396 return rt->fib6_nh->fib_nh_gw_family ||
5397 mlxsw_sp_nexthop6_ipip_type(mlxsw_sp, rt, NULL);
5401 mlxsw_sp_nexthop6_group_info_init(struct mlxsw_sp *mlxsw_sp,
5402 struct mlxsw_sp_nexthop_group *nh_grp,
5403 struct mlxsw_sp_fib6_entry *fib6_entry)
5405 struct mlxsw_sp_nexthop_group_info *nhgi;
5406 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5407 struct mlxsw_sp_nexthop *nh;
5410 nhgi = kzalloc(struct_size(nhgi, nexthops, fib6_entry->nrt6),
5414 nh_grp->nhgi = nhgi;
5415 nhgi->nh_grp = nh_grp;
5416 mlxsw_sp_rt6 = list_first_entry(&fib6_entry->rt6_list,
5417 struct mlxsw_sp_rt6, list);
5418 nhgi->gateway = mlxsw_sp_rt6_is_gateway(mlxsw_sp, mlxsw_sp_rt6->rt);
5419 nhgi->count = fib6_entry->nrt6;
5420 for (i = 0; i < nhgi->count; i++) {
5421 struct fib6_info *rt = mlxsw_sp_rt6->rt;
5423 nh = &nhgi->nexthops[i];
5424 err = mlxsw_sp_nexthop6_init(mlxsw_sp, nh_grp, nh, rt);
5426 goto err_nexthop6_init;
5427 mlxsw_sp_rt6 = list_next_entry(mlxsw_sp_rt6, list);
5429 nh_grp->nhgi = nhgi;
5430 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
5435 for (i--; i >= 0; i--) {
5436 nh = &nhgi->nexthops[i];
5437 mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
5444 mlxsw_sp_nexthop6_group_info_fini(struct mlxsw_sp *mlxsw_sp,
5445 struct mlxsw_sp_nexthop_group *nh_grp)
5447 struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
5450 for (i = nhgi->count - 1; i >= 0; i--) {
5451 struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
5453 mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
5455 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
5456 WARN_ON_ONCE(nhgi->adj_index_valid);
5460 static struct mlxsw_sp_nexthop_group *
5461 mlxsw_sp_nexthop6_group_create(struct mlxsw_sp *mlxsw_sp,
5462 struct mlxsw_sp_fib6_entry *fib6_entry)
5464 struct mlxsw_sp_nexthop_group *nh_grp;
5467 nh_grp = kzalloc(sizeof(*nh_grp), GFP_KERNEL);
5469 return ERR_PTR(-ENOMEM);
5470 INIT_LIST_HEAD(&nh_grp->fib_list);
5471 nh_grp->type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6;
5473 err = mlxsw_sp_nexthop6_group_info_init(mlxsw_sp, nh_grp, fib6_entry);
5475 goto err_nexthop_group_info_init;
5477 err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
5479 goto err_nexthop_group_insert;
5483 err_nexthop_group_insert:
5484 mlxsw_sp_nexthop6_group_info_fini(mlxsw_sp, nh_grp);
5485 err_nexthop_group_info_init:
5487 return ERR_PTR(err);
5491 mlxsw_sp_nexthop6_group_destroy(struct mlxsw_sp *mlxsw_sp,
5492 struct mlxsw_sp_nexthop_group *nh_grp)
5494 mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
5495 mlxsw_sp_nexthop6_group_info_fini(mlxsw_sp, nh_grp);
5499 static int mlxsw_sp_nexthop6_group_get(struct mlxsw_sp *mlxsw_sp,
5500 struct mlxsw_sp_fib6_entry *fib6_entry)
5502 struct mlxsw_sp_nexthop_group *nh_grp;
5504 nh_grp = mlxsw_sp_nexthop6_group_lookup(mlxsw_sp, fib6_entry);
5506 nh_grp = mlxsw_sp_nexthop6_group_create(mlxsw_sp, fib6_entry);
5508 return PTR_ERR(nh_grp);
5511 list_add_tail(&fib6_entry->common.nexthop_group_node,
5513 fib6_entry->common.nh_group = nh_grp;
5515 /* The route and the nexthop are described by the same struct, so we
5516 * need to the update the nexthop offload indication for the new route.
5518 __mlxsw_sp_nexthop6_group_offload_refresh(nh_grp, fib6_entry);
5523 static void mlxsw_sp_nexthop6_group_put(struct mlxsw_sp *mlxsw_sp,
5524 struct mlxsw_sp_fib_entry *fib_entry)
5526 struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
5528 list_del(&fib_entry->nexthop_group_node);
5529 if (!list_empty(&nh_grp->fib_list))
5531 mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, nh_grp);
5534 static int mlxsw_sp_nexthop6_group_update(struct mlxsw_sp *mlxsw_sp,
5535 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5536 struct mlxsw_sp_fib6_entry *fib6_entry)
5538 struct mlxsw_sp_nexthop_group *old_nh_grp = fib6_entry->common.nh_group;
5541 fib6_entry->common.nh_group = NULL;
5542 list_del(&fib6_entry->common.nexthop_group_node);
5544 err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
5546 goto err_nexthop6_group_get;
5548 /* In case this entry is offloaded, then the adjacency index
5549 * currently associated with it in the device's table is that
5550 * of the old group. Start using the new one instead.
5552 err = __mlxsw_sp_fib_entry_update(mlxsw_sp, op_ctx,
5553 &fib6_entry->common, false);
5555 goto err_fib_entry_update;
5557 if (list_empty(&old_nh_grp->fib_list))
5558 mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, old_nh_grp);
5562 err_fib_entry_update:
5563 mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
5564 err_nexthop6_group_get:
5565 list_add_tail(&fib6_entry->common.nexthop_group_node,
5566 &old_nh_grp->fib_list);
5567 fib6_entry->common.nh_group = old_nh_grp;
5572 mlxsw_sp_fib6_entry_nexthop_add(struct mlxsw_sp *mlxsw_sp,
5573 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5574 struct mlxsw_sp_fib6_entry *fib6_entry,
5575 struct fib6_info **rt_arr, unsigned int nrt6)
5577 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5580 for (i = 0; i < nrt6; i++) {
5581 mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt_arr[i]);
5582 if (IS_ERR(mlxsw_sp_rt6)) {
5583 err = PTR_ERR(mlxsw_sp_rt6);
5584 goto err_rt6_create;
5587 list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
5591 err = mlxsw_sp_nexthop6_group_update(mlxsw_sp, op_ctx, fib6_entry);
5593 goto err_nexthop6_group_update;
5597 err_nexthop6_group_update:
5600 for (i--; i >= 0; i--) {
5602 mlxsw_sp_rt6 = list_last_entry(&fib6_entry->rt6_list,
5603 struct mlxsw_sp_rt6, list);
5604 list_del(&mlxsw_sp_rt6->list);
5605 mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5611 mlxsw_sp_fib6_entry_nexthop_del(struct mlxsw_sp *mlxsw_sp,
5612 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5613 struct mlxsw_sp_fib6_entry *fib6_entry,
5614 struct fib6_info **rt_arr, unsigned int nrt6)
5616 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5619 for (i = 0; i < nrt6; i++) {
5620 mlxsw_sp_rt6 = mlxsw_sp_fib6_entry_rt_find(fib6_entry,
5622 if (WARN_ON_ONCE(!mlxsw_sp_rt6))
5626 list_del(&mlxsw_sp_rt6->list);
5627 mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5630 mlxsw_sp_nexthop6_group_update(mlxsw_sp, op_ctx, fib6_entry);
5633 static void mlxsw_sp_fib6_entry_type_set(struct mlxsw_sp *mlxsw_sp,
5634 struct mlxsw_sp_fib_entry *fib_entry,
5635 const struct fib6_info *rt)
5637 if (rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
5638 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
5639 else if (rt->fib6_type == RTN_BLACKHOLE)
5640 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE;
5641 else if (rt->fib6_flags & RTF_REJECT)
5642 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE;
5643 else if (mlxsw_sp_rt6_is_gateway(mlxsw_sp, rt))
5644 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
5646 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
5650 mlxsw_sp_fib6_entry_rt_destroy_all(struct mlxsw_sp_fib6_entry *fib6_entry)
5652 struct mlxsw_sp_rt6 *mlxsw_sp_rt6, *tmp;
5654 list_for_each_entry_safe(mlxsw_sp_rt6, tmp, &fib6_entry->rt6_list,
5657 list_del(&mlxsw_sp_rt6->list);
5658 mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5662 static struct mlxsw_sp_fib6_entry *
5663 mlxsw_sp_fib6_entry_create(struct mlxsw_sp *mlxsw_sp,
5664 struct mlxsw_sp_fib_node *fib_node,
5665 struct fib6_info **rt_arr, unsigned int nrt6)
5667 struct mlxsw_sp_fib6_entry *fib6_entry;
5668 struct mlxsw_sp_fib_entry *fib_entry;
5669 struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5672 fib6_entry = kzalloc(sizeof(*fib6_entry), GFP_KERNEL);
5674 return ERR_PTR(-ENOMEM);
5675 fib_entry = &fib6_entry->common;
5677 fib_entry->priv = mlxsw_sp_fib_entry_priv_create(fib_node->fib->ll_ops);
5678 if (IS_ERR(fib_entry->priv)) {
5679 err = PTR_ERR(fib_entry->priv);
5680 goto err_fib_entry_priv_create;
5683 INIT_LIST_HEAD(&fib6_entry->rt6_list);
5685 for (i = 0; i < nrt6; i++) {
5686 mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt_arr[i]);
5687 if (IS_ERR(mlxsw_sp_rt6)) {
5688 err = PTR_ERR(mlxsw_sp_rt6);
5689 goto err_rt6_create;
5691 list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
5695 mlxsw_sp_fib6_entry_type_set(mlxsw_sp, fib_entry, rt_arr[0]);
5697 err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
5699 goto err_nexthop6_group_get;
5701 fib_entry->fib_node = fib_node;
5705 err_nexthop6_group_get:
5708 for (i--; i >= 0; i--) {
5710 mlxsw_sp_rt6 = list_last_entry(&fib6_entry->rt6_list,
5711 struct mlxsw_sp_rt6, list);
5712 list_del(&mlxsw_sp_rt6->list);
5713 mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5715 mlxsw_sp_fib_entry_priv_put(fib_entry->priv);
5716 err_fib_entry_priv_create:
5718 return ERR_PTR(err);
5721 static void mlxsw_sp_fib6_entry_destroy(struct mlxsw_sp *mlxsw_sp,
5722 struct mlxsw_sp_fib6_entry *fib6_entry)
5724 mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
5725 mlxsw_sp_fib6_entry_rt_destroy_all(fib6_entry);
5726 WARN_ON(fib6_entry->nrt6);
5727 mlxsw_sp_fib_entry_priv_put(fib6_entry->common.priv);
5731 static struct mlxsw_sp_fib6_entry *
5732 mlxsw_sp_fib6_entry_lookup(struct mlxsw_sp *mlxsw_sp,
5733 const struct fib6_info *rt)
5735 struct mlxsw_sp_fib6_entry *fib6_entry;
5736 struct mlxsw_sp_fib_node *fib_node;
5737 struct mlxsw_sp_fib *fib;
5738 struct fib6_info *cmp_rt;
5739 struct mlxsw_sp_vr *vr;
5741 vr = mlxsw_sp_vr_find(mlxsw_sp, rt->fib6_table->tb6_id);
5744 fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV6);
5746 fib_node = mlxsw_sp_fib_node_lookup(fib, &rt->fib6_dst.addr,
5747 sizeof(rt->fib6_dst.addr),
5752 fib6_entry = container_of(fib_node->fib_entry,
5753 struct mlxsw_sp_fib6_entry, common);
5754 cmp_rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
5755 if (rt->fib6_table->tb6_id == cmp_rt->fib6_table->tb6_id &&
5756 rt->fib6_metric == cmp_rt->fib6_metric &&
5757 mlxsw_sp_fib6_entry_rt_find(fib6_entry, rt))
5763 static bool mlxsw_sp_fib6_allow_replace(struct mlxsw_sp_fib6_entry *fib6_entry)
5765 struct mlxsw_sp_fib_node *fib_node = fib6_entry->common.fib_node;
5766 struct mlxsw_sp_fib6_entry *fib6_replaced;
5767 struct fib6_info *rt, *rt_replaced;
5769 if (!fib_node->fib_entry)
5772 fib6_replaced = container_of(fib_node->fib_entry,
5773 struct mlxsw_sp_fib6_entry,
5775 rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
5776 rt_replaced = mlxsw_sp_fib6_entry_rt(fib6_replaced);
5777 if (rt->fib6_table->tb6_id == RT_TABLE_MAIN &&
5778 rt_replaced->fib6_table->tb6_id == RT_TABLE_LOCAL)
5784 static int mlxsw_sp_router_fib6_replace(struct mlxsw_sp *mlxsw_sp,
5785 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5786 struct fib6_info **rt_arr, unsigned int nrt6)
5788 struct mlxsw_sp_fib6_entry *fib6_entry, *fib6_replaced;
5789 struct mlxsw_sp_fib_entry *replaced;
5790 struct mlxsw_sp_fib_node *fib_node;
5791 struct fib6_info *rt = rt_arr[0];
5794 if (mlxsw_sp->router->aborted)
5797 if (rt->fib6_src.plen)
5800 if (mlxsw_sp_fib6_rt_should_ignore(rt))
5803 fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->fib6_table->tb6_id,
5805 sizeof(rt->fib6_dst.addr),
5807 MLXSW_SP_L3_PROTO_IPV6);
5808 if (IS_ERR(fib_node))
5809 return PTR_ERR(fib_node);
5811 fib6_entry = mlxsw_sp_fib6_entry_create(mlxsw_sp, fib_node, rt_arr,
5813 if (IS_ERR(fib6_entry)) {
5814 err = PTR_ERR(fib6_entry);
5815 goto err_fib6_entry_create;
5818 if (!mlxsw_sp_fib6_allow_replace(fib6_entry)) {
5819 mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5820 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5824 replaced = fib_node->fib_entry;
5825 err = mlxsw_sp_fib_node_entry_link(mlxsw_sp, op_ctx, &fib6_entry->common);
5827 goto err_fib_node_entry_link;
5829 /* Nothing to replace */
5833 mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, replaced);
5834 fib6_replaced = container_of(replaced, struct mlxsw_sp_fib6_entry,
5836 mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_replaced);
5840 err_fib_node_entry_link:
5841 fib_node->fib_entry = replaced;
5842 mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5843 err_fib6_entry_create:
5844 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5848 static int mlxsw_sp_router_fib6_append(struct mlxsw_sp *mlxsw_sp,
5849 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5850 struct fib6_info **rt_arr, unsigned int nrt6)
5852 struct mlxsw_sp_fib6_entry *fib6_entry;
5853 struct mlxsw_sp_fib_node *fib_node;
5854 struct fib6_info *rt = rt_arr[0];
5857 if (mlxsw_sp->router->aborted)
5860 if (rt->fib6_src.plen)
5863 if (mlxsw_sp_fib6_rt_should_ignore(rt))
5866 fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->fib6_table->tb6_id,
5868 sizeof(rt->fib6_dst.addr),
5870 MLXSW_SP_L3_PROTO_IPV6);
5871 if (IS_ERR(fib_node))
5872 return PTR_ERR(fib_node);
5874 if (WARN_ON_ONCE(!fib_node->fib_entry)) {
5875 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5879 fib6_entry = container_of(fib_node->fib_entry,
5880 struct mlxsw_sp_fib6_entry, common);
5881 err = mlxsw_sp_fib6_entry_nexthop_add(mlxsw_sp, op_ctx, fib6_entry, rt_arr, nrt6);
5883 goto err_fib6_entry_nexthop_add;
5887 err_fib6_entry_nexthop_add:
5888 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5892 static int mlxsw_sp_router_fib6_del(struct mlxsw_sp *mlxsw_sp,
5893 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5894 struct fib6_info **rt_arr, unsigned int nrt6)
5896 struct mlxsw_sp_fib6_entry *fib6_entry;
5897 struct mlxsw_sp_fib_node *fib_node;
5898 struct fib6_info *rt = rt_arr[0];
5901 if (mlxsw_sp->router->aborted)
5904 if (mlxsw_sp_fib6_rt_should_ignore(rt))
5907 /* Multipath routes are first added to the FIB trie and only then
5908 * notified. If we vetoed the addition, we will get a delete
5909 * notification for a route we do not have. Therefore, do not warn if
5910 * route was not found.
5912 fib6_entry = mlxsw_sp_fib6_entry_lookup(mlxsw_sp, rt);
5916 /* If not all the nexthops are deleted, then only reduce the nexthop
5919 if (nrt6 != fib6_entry->nrt6) {
5920 mlxsw_sp_fib6_entry_nexthop_del(mlxsw_sp, op_ctx, fib6_entry, rt_arr, nrt6);
5924 fib_node = fib6_entry->common.fib_node;
5926 err = __mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, op_ctx, &fib6_entry->common);
5927 mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5928 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5932 static int __mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp,
5933 enum mlxsw_sp_l3proto proto,
5936 const struct mlxsw_sp_router_ll_ops *ll_ops = mlxsw_sp->router->proto_ll_ops[proto];
5937 enum mlxsw_reg_ralxx_protocol ralxx_proto =
5938 (enum mlxsw_reg_ralxx_protocol) proto;
5939 struct mlxsw_sp_fib_entry_priv *priv;
5940 char xralta_pl[MLXSW_REG_XRALTA_LEN];
5941 char xralst_pl[MLXSW_REG_XRALST_LEN];
5944 mlxsw_reg_xralta_pack(xralta_pl, true, ralxx_proto, tree_id);
5945 err = ll_ops->ralta_write(mlxsw_sp, xralta_pl);
5949 mlxsw_reg_xralst_pack(xralst_pl, 0xff, tree_id);
5950 err = ll_ops->ralst_write(mlxsw_sp, xralst_pl);
5954 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
5955 struct mlxsw_sp_fib_entry_op_ctx *op_ctx = mlxsw_sp->router->ll_op_ctx;
5956 struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
5957 char xraltb_pl[MLXSW_REG_XRALTB_LEN];
5959 mlxsw_sp_fib_entry_op_ctx_clear(op_ctx);
5960 mlxsw_reg_xraltb_pack(xraltb_pl, vr->id, ralxx_proto, tree_id);
5961 err = ll_ops->raltb_write(mlxsw_sp, xraltb_pl);
5965 priv = mlxsw_sp_fib_entry_priv_create(ll_ops);
5967 return PTR_ERR(priv);
5969 ll_ops->fib_entry_pack(op_ctx, proto, MLXSW_SP_FIB_ENTRY_OP_WRITE,
5970 vr->id, 0, NULL, priv);
5971 ll_ops->fib_entry_act_ip2me_pack(op_ctx);
5972 err = ll_ops->fib_entry_commit(mlxsw_sp, op_ctx, NULL);
5973 mlxsw_sp_fib_entry_priv_put(priv);
5981 static struct mlxsw_sp_mr_table *
5982 mlxsw_sp_router_fibmr_family_to_table(struct mlxsw_sp_vr *vr, int family)
5984 if (family == RTNL_FAMILY_IPMR)
5985 return vr->mr_table[MLXSW_SP_L3_PROTO_IPV4];
5987 return vr->mr_table[MLXSW_SP_L3_PROTO_IPV6];
5990 static int mlxsw_sp_router_fibmr_add(struct mlxsw_sp *mlxsw_sp,
5991 struct mfc_entry_notifier_info *men_info,
5994 struct mlxsw_sp_mr_table *mrt;
5995 struct mlxsw_sp_vr *vr;
5997 if (mlxsw_sp->router->aborted)
6000 vr = mlxsw_sp_vr_get(mlxsw_sp, men_info->tb_id, NULL);
6004 mrt = mlxsw_sp_router_fibmr_family_to_table(vr, men_info->info.family);
6005 return mlxsw_sp_mr_route_add(mrt, men_info->mfc, replace);
6008 static void mlxsw_sp_router_fibmr_del(struct mlxsw_sp *mlxsw_sp,
6009 struct mfc_entry_notifier_info *men_info)
6011 struct mlxsw_sp_mr_table *mrt;
6012 struct mlxsw_sp_vr *vr;
6014 if (mlxsw_sp->router->aborted)
6017 vr = mlxsw_sp_vr_find(mlxsw_sp, men_info->tb_id);
6021 mrt = mlxsw_sp_router_fibmr_family_to_table(vr, men_info->info.family);
6022 mlxsw_sp_mr_route_del(mrt, men_info->mfc);
6023 mlxsw_sp_vr_put(mlxsw_sp, vr);
6027 mlxsw_sp_router_fibmr_vif_add(struct mlxsw_sp *mlxsw_sp,
6028 struct vif_entry_notifier_info *ven_info)
6030 struct mlxsw_sp_mr_table *mrt;
6031 struct mlxsw_sp_rif *rif;
6032 struct mlxsw_sp_vr *vr;
6034 if (mlxsw_sp->router->aborted)
6037 vr = mlxsw_sp_vr_get(mlxsw_sp, ven_info->tb_id, NULL);
6041 mrt = mlxsw_sp_router_fibmr_family_to_table(vr, ven_info->info.family);
6042 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, ven_info->dev);
6043 return mlxsw_sp_mr_vif_add(mrt, ven_info->dev,
6044 ven_info->vif_index,
6045 ven_info->vif_flags, rif);
6049 mlxsw_sp_router_fibmr_vif_del(struct mlxsw_sp *mlxsw_sp,
6050 struct vif_entry_notifier_info *ven_info)
6052 struct mlxsw_sp_mr_table *mrt;
6053 struct mlxsw_sp_vr *vr;
6055 if (mlxsw_sp->router->aborted)
6058 vr = mlxsw_sp_vr_find(mlxsw_sp, ven_info->tb_id);
6062 mrt = mlxsw_sp_router_fibmr_family_to_table(vr, ven_info->info.family);
6063 mlxsw_sp_mr_vif_del(mrt, ven_info->vif_index);
6064 mlxsw_sp_vr_put(mlxsw_sp, vr);
6067 static int mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp)
6069 enum mlxsw_sp_l3proto proto = MLXSW_SP_L3_PROTO_IPV4;
6072 err = __mlxsw_sp_router_set_abort_trap(mlxsw_sp, proto,
6073 MLXSW_SP_LPM_TREE_MIN);
6077 /* The multicast router code does not need an abort trap as by default,
6078 * packets that don't match any routes are trapped to the CPU.
6081 proto = MLXSW_SP_L3_PROTO_IPV6;
6082 return __mlxsw_sp_router_set_abort_trap(mlxsw_sp, proto,
6083 MLXSW_SP_LPM_TREE_MIN + 1);
6086 static void mlxsw_sp_fib4_node_flush(struct mlxsw_sp *mlxsw_sp,
6087 struct mlxsw_sp_fib_node *fib_node)
6089 struct mlxsw_sp_fib4_entry *fib4_entry;
6091 fib4_entry = container_of(fib_node->fib_entry,
6092 struct mlxsw_sp_fib4_entry, common);
6093 mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, fib_node->fib_entry);
6094 mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
6095 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
6098 static void mlxsw_sp_fib6_node_flush(struct mlxsw_sp *mlxsw_sp,
6099 struct mlxsw_sp_fib_node *fib_node)
6101 struct mlxsw_sp_fib6_entry *fib6_entry;
6103 fib6_entry = container_of(fib_node->fib_entry,
6104 struct mlxsw_sp_fib6_entry, common);
6105 mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, fib_node->fib_entry);
6106 mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
6107 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
6110 static void mlxsw_sp_fib_node_flush(struct mlxsw_sp *mlxsw_sp,
6111 struct mlxsw_sp_fib_node *fib_node)
6113 switch (fib_node->fib->proto) {
6114 case MLXSW_SP_L3_PROTO_IPV4:
6115 mlxsw_sp_fib4_node_flush(mlxsw_sp, fib_node);
6117 case MLXSW_SP_L3_PROTO_IPV6:
6118 mlxsw_sp_fib6_node_flush(mlxsw_sp, fib_node);
6123 static void mlxsw_sp_vr_fib_flush(struct mlxsw_sp *mlxsw_sp,
6124 struct mlxsw_sp_vr *vr,
6125 enum mlxsw_sp_l3proto proto)
6127 struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
6128 struct mlxsw_sp_fib_node *fib_node, *tmp;
6130 list_for_each_entry_safe(fib_node, tmp, &fib->node_list, list) {
6131 bool do_break = &tmp->list == &fib->node_list;
6133 mlxsw_sp_fib_node_flush(mlxsw_sp, fib_node);
6139 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
6143 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
6144 struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
6146 if (!mlxsw_sp_vr_is_used(vr))
6149 for (j = 0; j < MLXSW_SP_L3_PROTO_MAX; j++)
6150 mlxsw_sp_mr_table_flush(vr->mr_table[j]);
6151 mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
6153 /* If virtual router was only used for IPv4, then it's no
6156 if (!mlxsw_sp_vr_is_used(vr))
6158 mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
6161 /* After flushing all the routes, it is not possible anyone is still
6162 * using the adjacency index that is discarding packets, so free it in
6163 * case it was allocated.
6165 if (!mlxsw_sp->router->adj_discard_index_valid)
6167 mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
6168 mlxsw_sp->router->adj_discard_index);
6169 mlxsw_sp->router->adj_discard_index_valid = false;
6172 static void mlxsw_sp_router_fib_abort(struct mlxsw_sp *mlxsw_sp)
6176 if (mlxsw_sp->router->aborted)
6178 dev_warn(mlxsw_sp->bus_info->dev, "FIB abort triggered. Note that FIB entries are no longer being offloaded to this device.\n");
6179 mlxsw_sp_router_fib_flush(mlxsw_sp);
6180 mlxsw_sp->router->aborted = true;
6181 err = mlxsw_sp_router_set_abort_trap(mlxsw_sp);
6183 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set abort trap.\n");
6186 struct mlxsw_sp_fib6_event {
6187 struct fib6_info **rt_arr;
6191 struct mlxsw_sp_fib_event {
6192 struct list_head list; /* node in fib queue */
6194 struct mlxsw_sp_fib6_event fib6_event;
6195 struct fib_entry_notifier_info fen_info;
6196 struct fib_rule_notifier_info fr_info;
6197 struct fib_nh_notifier_info fnh_info;
6198 struct mfc_entry_notifier_info men_info;
6199 struct vif_entry_notifier_info ven_info;
6201 struct mlxsw_sp *mlxsw_sp;
6202 unsigned long event;
6207 mlxsw_sp_router_fib6_event_init(struct mlxsw_sp_fib6_event *fib6_event,
6208 struct fib6_entry_notifier_info *fen6_info)
6210 struct fib6_info *rt = fen6_info->rt;
6211 struct fib6_info **rt_arr;
6212 struct fib6_info *iter;
6216 nrt6 = fen6_info->nsiblings + 1;
6218 rt_arr = kcalloc(nrt6, sizeof(struct fib6_info *), GFP_ATOMIC);
6222 fib6_event->rt_arr = rt_arr;
6223 fib6_event->nrt6 = nrt6;
6228 if (!fen6_info->nsiblings)
6231 list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
6232 if (i == fen6_info->nsiblings)
6235 rt_arr[i + 1] = iter;
6236 fib6_info_hold(iter);
6239 WARN_ON_ONCE(i != fen6_info->nsiblings);
6245 mlxsw_sp_router_fib6_event_fini(struct mlxsw_sp_fib6_event *fib6_event)
6249 for (i = 0; i < fib6_event->nrt6; i++)
6250 mlxsw_sp_rt6_release(fib6_event->rt_arr[i]);
6251 kfree(fib6_event->rt_arr);
6254 static void mlxsw_sp_router_fib4_event_process(struct mlxsw_sp *mlxsw_sp,
6255 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
6256 struct mlxsw_sp_fib_event *fib_event)
6260 mlxsw_sp_span_respin(mlxsw_sp);
6262 switch (fib_event->event) {
6263 case FIB_EVENT_ENTRY_REPLACE:
6264 err = mlxsw_sp_router_fib4_replace(mlxsw_sp, op_ctx, &fib_event->fen_info);
6266 mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6267 mlxsw_sp_router_fib_abort(mlxsw_sp);
6269 fib_info_put(fib_event->fen_info.fi);
6271 case FIB_EVENT_ENTRY_DEL:
6272 err = mlxsw_sp_router_fib4_del(mlxsw_sp, op_ctx, &fib_event->fen_info);
6274 mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6275 fib_info_put(fib_event->fen_info.fi);
6277 case FIB_EVENT_NH_ADD:
6278 case FIB_EVENT_NH_DEL:
6279 mlxsw_sp_nexthop4_event(mlxsw_sp, fib_event->event, fib_event->fnh_info.fib_nh);
6280 fib_info_put(fib_event->fnh_info.fib_nh->nh_parent);
6285 static void mlxsw_sp_router_fib6_event_process(struct mlxsw_sp *mlxsw_sp,
6286 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
6287 struct mlxsw_sp_fib_event *fib_event)
6291 mlxsw_sp_span_respin(mlxsw_sp);
6293 switch (fib_event->event) {
6294 case FIB_EVENT_ENTRY_REPLACE:
6295 err = mlxsw_sp_router_fib6_replace(mlxsw_sp, op_ctx, fib_event->fib6_event.rt_arr,
6296 fib_event->fib6_event.nrt6);
6298 mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6299 mlxsw_sp_router_fib_abort(mlxsw_sp);
6301 mlxsw_sp_router_fib6_event_fini(&fib_event->fib6_event);
6303 case FIB_EVENT_ENTRY_APPEND:
6304 err = mlxsw_sp_router_fib6_append(mlxsw_sp, op_ctx, fib_event->fib6_event.rt_arr,
6305 fib_event->fib6_event.nrt6);
6307 mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6308 mlxsw_sp_router_fib_abort(mlxsw_sp);
6310 mlxsw_sp_router_fib6_event_fini(&fib_event->fib6_event);
6312 case FIB_EVENT_ENTRY_DEL:
6313 err = mlxsw_sp_router_fib6_del(mlxsw_sp, op_ctx, fib_event->fib6_event.rt_arr,
6314 fib_event->fib6_event.nrt6);
6316 mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6317 mlxsw_sp_router_fib6_event_fini(&fib_event->fib6_event);
6322 static void mlxsw_sp_router_fibmr_event_process(struct mlxsw_sp *mlxsw_sp,
6323 struct mlxsw_sp_fib_event *fib_event)
6329 mutex_lock(&mlxsw_sp->router->lock);
6330 switch (fib_event->event) {
6331 case FIB_EVENT_ENTRY_REPLACE:
6332 case FIB_EVENT_ENTRY_ADD:
6333 replace = fib_event->event == FIB_EVENT_ENTRY_REPLACE;
6335 err = mlxsw_sp_router_fibmr_add(mlxsw_sp, &fib_event->men_info, replace);
6337 mlxsw_sp_router_fib_abort(mlxsw_sp);
6338 mr_cache_put(fib_event->men_info.mfc);
6340 case FIB_EVENT_ENTRY_DEL:
6341 mlxsw_sp_router_fibmr_del(mlxsw_sp, &fib_event->men_info);
6342 mr_cache_put(fib_event->men_info.mfc);
6344 case FIB_EVENT_VIF_ADD:
6345 err = mlxsw_sp_router_fibmr_vif_add(mlxsw_sp,
6346 &fib_event->ven_info);
6348 mlxsw_sp_router_fib_abort(mlxsw_sp);
6349 dev_put(fib_event->ven_info.dev);
6351 case FIB_EVENT_VIF_DEL:
6352 mlxsw_sp_router_fibmr_vif_del(mlxsw_sp, &fib_event->ven_info);
6353 dev_put(fib_event->ven_info.dev);
6356 mutex_unlock(&mlxsw_sp->router->lock);
6360 static void mlxsw_sp_router_fib_event_work(struct work_struct *work)
6362 struct mlxsw_sp_router *router = container_of(work, struct mlxsw_sp_router, fib_event_work);
6363 struct mlxsw_sp_fib_entry_op_ctx *op_ctx = router->ll_op_ctx;
6364 struct mlxsw_sp *mlxsw_sp = router->mlxsw_sp;
6365 struct mlxsw_sp_fib_event *next_fib_event;
6366 struct mlxsw_sp_fib_event *fib_event;
6367 int last_family = AF_UNSPEC;
6368 LIST_HEAD(fib_event_queue);
6370 spin_lock_bh(&router->fib_event_queue_lock);
6371 list_splice_init(&router->fib_event_queue, &fib_event_queue);
6372 spin_unlock_bh(&router->fib_event_queue_lock);
6374 /* Router lock is held here to make sure per-instance
6375 * operation context is not used in between FIB4/6 events
6378 mutex_lock(&router->lock);
6379 mlxsw_sp_fib_entry_op_ctx_clear(op_ctx);
6380 list_for_each_entry_safe(fib_event, next_fib_event,
6381 &fib_event_queue, list) {
6382 /* Check if the next entry in the queue exists and it is
6383 * of the same type (family and event) as the currect one.
6384 * In that case it is permitted to do the bulking
6385 * of multiple FIB entries to a single register write.
6387 op_ctx->bulk_ok = !list_is_last(&fib_event->list, &fib_event_queue) &&
6388 fib_event->family == next_fib_event->family &&
6389 fib_event->event == next_fib_event->event;
6391 /* In case family of this and the previous entry are different, context
6392 * reinitialization is going to be needed now, indicate that.
6393 * Note that since last_family is initialized to AF_UNSPEC, this is always
6394 * going to happen for the first entry processed in the work.
6396 if (fib_event->family != last_family)
6397 op_ctx->initialized = false;
6399 switch (fib_event->family) {
6401 mlxsw_sp_router_fib4_event_process(mlxsw_sp, op_ctx,
6405 mlxsw_sp_router_fib6_event_process(mlxsw_sp, op_ctx,
6408 case RTNL_FAMILY_IP6MR:
6409 case RTNL_FAMILY_IPMR:
6410 /* Unlock here as inside FIBMR the lock is taken again
6411 * under RTNL. The per-instance operation context
6412 * is not used by FIBMR.
6414 mutex_unlock(&router->lock);
6415 mlxsw_sp_router_fibmr_event_process(mlxsw_sp,
6417 mutex_lock(&router->lock);
6422 last_family = fib_event->family;
6426 WARN_ON_ONCE(!list_empty(&router->ll_op_ctx->fib_entry_priv_list));
6427 mutex_unlock(&router->lock);
6430 static void mlxsw_sp_router_fib4_event(struct mlxsw_sp_fib_event *fib_event,
6431 struct fib_notifier_info *info)
6433 struct fib_entry_notifier_info *fen_info;
6434 struct fib_nh_notifier_info *fnh_info;
6436 switch (fib_event->event) {
6437 case FIB_EVENT_ENTRY_REPLACE:
6438 case FIB_EVENT_ENTRY_DEL:
6439 fen_info = container_of(info, struct fib_entry_notifier_info,
6441 fib_event->fen_info = *fen_info;
6442 /* Take reference on fib_info to prevent it from being
6443 * freed while event is queued. Release it afterwards.
6445 fib_info_hold(fib_event->fen_info.fi);
6447 case FIB_EVENT_NH_ADD:
6448 case FIB_EVENT_NH_DEL:
6449 fnh_info = container_of(info, struct fib_nh_notifier_info,
6451 fib_event->fnh_info = *fnh_info;
6452 fib_info_hold(fib_event->fnh_info.fib_nh->nh_parent);
6457 static int mlxsw_sp_router_fib6_event(struct mlxsw_sp_fib_event *fib_event,
6458 struct fib_notifier_info *info)
6460 struct fib6_entry_notifier_info *fen6_info;
6463 switch (fib_event->event) {
6464 case FIB_EVENT_ENTRY_REPLACE:
6465 case FIB_EVENT_ENTRY_APPEND:
6466 case FIB_EVENT_ENTRY_DEL:
6467 fen6_info = container_of(info, struct fib6_entry_notifier_info,
6469 err = mlxsw_sp_router_fib6_event_init(&fib_event->fib6_event,
6480 mlxsw_sp_router_fibmr_event(struct mlxsw_sp_fib_event *fib_event,
6481 struct fib_notifier_info *info)
6483 switch (fib_event->event) {
6484 case FIB_EVENT_ENTRY_REPLACE:
6485 case FIB_EVENT_ENTRY_ADD:
6486 case FIB_EVENT_ENTRY_DEL:
6487 memcpy(&fib_event->men_info, info, sizeof(fib_event->men_info));
6488 mr_cache_hold(fib_event->men_info.mfc);
6490 case FIB_EVENT_VIF_ADD:
6491 case FIB_EVENT_VIF_DEL:
6492 memcpy(&fib_event->ven_info, info, sizeof(fib_event->ven_info));
6493 dev_hold(fib_event->ven_info.dev);
6498 static int mlxsw_sp_router_fib_rule_event(unsigned long event,
6499 struct fib_notifier_info *info,
6500 struct mlxsw_sp *mlxsw_sp)
6502 struct netlink_ext_ack *extack = info->extack;
6503 struct fib_rule_notifier_info *fr_info;
6504 struct fib_rule *rule;
6507 /* nothing to do at the moment */
6508 if (event == FIB_EVENT_RULE_DEL)
6511 if (mlxsw_sp->router->aborted)
6514 fr_info = container_of(info, struct fib_rule_notifier_info, info);
6515 rule = fr_info->rule;
6517 /* Rule only affects locally generated traffic */
6518 if (rule->iifindex == mlxsw_sp_net(mlxsw_sp)->loopback_dev->ifindex)
6521 switch (info->family) {
6523 if (!fib4_rule_default(rule) && !rule->l3mdev)
6527 if (!fib6_rule_default(rule) && !rule->l3mdev)
6530 case RTNL_FAMILY_IPMR:
6531 if (!ipmr_rule_default(rule) && !rule->l3mdev)
6534 case RTNL_FAMILY_IP6MR:
6535 if (!ip6mr_rule_default(rule) && !rule->l3mdev)
6541 NL_SET_ERR_MSG_MOD(extack, "FIB rules not supported");
6546 /* Called with rcu_read_lock() */
6547 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
6548 unsigned long event, void *ptr)
6550 struct mlxsw_sp_fib_event *fib_event;
6551 struct fib_notifier_info *info = ptr;
6552 struct mlxsw_sp_router *router;
6555 if ((info->family != AF_INET && info->family != AF_INET6 &&
6556 info->family != RTNL_FAMILY_IPMR &&
6557 info->family != RTNL_FAMILY_IP6MR))
6560 router = container_of(nb, struct mlxsw_sp_router, fib_nb);
6563 case FIB_EVENT_RULE_ADD:
6564 case FIB_EVENT_RULE_DEL:
6565 err = mlxsw_sp_router_fib_rule_event(event, info,
6567 return notifier_from_errno(err);
6568 case FIB_EVENT_ENTRY_ADD:
6569 case FIB_EVENT_ENTRY_REPLACE:
6570 case FIB_EVENT_ENTRY_APPEND:
6571 if (router->aborted) {
6572 NL_SET_ERR_MSG_MOD(info->extack, "FIB offload was aborted. Not configuring route");
6573 return notifier_from_errno(-EINVAL);
6575 if (info->family == AF_INET) {
6576 struct fib_entry_notifier_info *fen_info = ptr;
6578 if (fen_info->fi->fib_nh_is_v6) {
6579 NL_SET_ERR_MSG_MOD(info->extack, "IPv6 gateway with IPv4 route is not supported");
6580 return notifier_from_errno(-EINVAL);
6582 if (fen_info->fi->nh) {
6583 NL_SET_ERR_MSG_MOD(info->extack, "IPv4 route with nexthop objects is not supported");
6584 return notifier_from_errno(-EINVAL);
6586 } else if (info->family == AF_INET6) {
6587 struct fib6_entry_notifier_info *fen6_info;
6589 fen6_info = container_of(info,
6590 struct fib6_entry_notifier_info,
6592 if (fen6_info->rt->nh) {
6593 NL_SET_ERR_MSG_MOD(info->extack, "IPv6 route with nexthop objects is not supported");
6594 return notifier_from_errno(-EINVAL);
6600 fib_event = kzalloc(sizeof(*fib_event), GFP_ATOMIC);
6604 fib_event->mlxsw_sp = router->mlxsw_sp;
6605 fib_event->event = event;
6606 fib_event->family = info->family;
6608 switch (info->family) {
6610 mlxsw_sp_router_fib4_event(fib_event, info);
6613 err = mlxsw_sp_router_fib6_event(fib_event, info);
6617 case RTNL_FAMILY_IP6MR:
6618 case RTNL_FAMILY_IPMR:
6619 mlxsw_sp_router_fibmr_event(fib_event, info);
6623 /* Enqueue the event and trigger the work */
6624 spin_lock_bh(&router->fib_event_queue_lock);
6625 list_add_tail(&fib_event->list, &router->fib_event_queue);
6626 spin_unlock_bh(&router->fib_event_queue_lock);
6627 mlxsw_core_schedule_work(&router->fib_event_work);
6636 static struct mlxsw_sp_rif *
6637 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
6638 const struct net_device *dev)
6642 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
6643 if (mlxsw_sp->router->rifs[i] &&
6644 mlxsw_sp->router->rifs[i]->dev == dev)
6645 return mlxsw_sp->router->rifs[i];
6650 bool mlxsw_sp_rif_exists(struct mlxsw_sp *mlxsw_sp,
6651 const struct net_device *dev)
6653 struct mlxsw_sp_rif *rif;
6655 mutex_lock(&mlxsw_sp->router->lock);
6656 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6657 mutex_unlock(&mlxsw_sp->router->lock);
6662 u16 mlxsw_sp_rif_vid(struct mlxsw_sp *mlxsw_sp, const struct net_device *dev)
6664 struct mlxsw_sp_rif *rif;
6667 mutex_lock(&mlxsw_sp->router->lock);
6668 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6672 /* We only return the VID for VLAN RIFs. Otherwise we return an
6673 * invalid value (0).
6675 if (rif->ops->type != MLXSW_SP_RIF_TYPE_VLAN)
6678 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
6681 mutex_unlock(&mlxsw_sp->router->lock);
6685 static int mlxsw_sp_router_rif_disable(struct mlxsw_sp *mlxsw_sp, u16 rif)
6687 char ritr_pl[MLXSW_REG_RITR_LEN];
6690 mlxsw_reg_ritr_rif_pack(ritr_pl, rif);
6691 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6695 mlxsw_reg_ritr_enable_set(ritr_pl, false);
6696 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6699 static void mlxsw_sp_router_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
6700 struct mlxsw_sp_rif *rif)
6702 mlxsw_sp_router_rif_disable(mlxsw_sp, rif->rif_index);
6703 mlxsw_sp_nexthop_rif_gone_sync(mlxsw_sp, rif);
6704 mlxsw_sp_neigh_rif_gone_sync(mlxsw_sp, rif);
6708 mlxsw_sp_rif_should_config(struct mlxsw_sp_rif *rif, struct net_device *dev,
6709 unsigned long event)
6711 struct inet6_dev *inet6_dev;
6712 bool addr_list_empty = true;
6713 struct in_device *idev;
6720 idev = __in_dev_get_rcu(dev);
6721 if (idev && idev->ifa_list)
6722 addr_list_empty = false;
6724 inet6_dev = __in6_dev_get(dev);
6725 if (addr_list_empty && inet6_dev &&
6726 !list_empty(&inet6_dev->addr_list))
6727 addr_list_empty = false;
6730 /* macvlans do not have a RIF, but rather piggy back on the
6731 * RIF of their lower device.
6733 if (netif_is_macvlan(dev) && addr_list_empty)
6736 if (rif && addr_list_empty &&
6737 !netif_is_l3_slave(rif->dev))
6739 /* It is possible we already removed the RIF ourselves
6740 * if it was assigned to a netdev that is now a bridge
6749 static enum mlxsw_sp_rif_type
6750 mlxsw_sp_dev_rif_type(const struct mlxsw_sp *mlxsw_sp,
6751 const struct net_device *dev)
6753 enum mlxsw_sp_fid_type type;
6755 if (mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL))
6756 return MLXSW_SP_RIF_TYPE_IPIP_LB;
6758 /* Otherwise RIF type is derived from the type of the underlying FID. */
6759 if (is_vlan_dev(dev) && netif_is_bridge_master(vlan_dev_real_dev(dev)))
6760 type = MLXSW_SP_FID_TYPE_8021Q;
6761 else if (netif_is_bridge_master(dev) && br_vlan_enabled(dev))
6762 type = MLXSW_SP_FID_TYPE_8021Q;
6763 else if (netif_is_bridge_master(dev))
6764 type = MLXSW_SP_FID_TYPE_8021D;
6766 type = MLXSW_SP_FID_TYPE_RFID;
6768 return mlxsw_sp_fid_type_rif_type(mlxsw_sp, type);
6771 static int mlxsw_sp_rif_index_alloc(struct mlxsw_sp *mlxsw_sp, u16 *p_rif_index)
6775 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++) {
6776 if (!mlxsw_sp->router->rifs[i]) {
6785 static struct mlxsw_sp_rif *mlxsw_sp_rif_alloc(size_t rif_size, u16 rif_index,
6787 struct net_device *l3_dev)
6789 struct mlxsw_sp_rif *rif;
6791 rif = kzalloc(rif_size, GFP_KERNEL);
6795 INIT_LIST_HEAD(&rif->nexthop_list);
6796 INIT_LIST_HEAD(&rif->neigh_list);
6798 ether_addr_copy(rif->addr, l3_dev->dev_addr);
6799 rif->mtu = l3_dev->mtu;
6803 rif->rif_index = rif_index;
6808 struct mlxsw_sp_rif *mlxsw_sp_rif_by_index(const struct mlxsw_sp *mlxsw_sp,
6811 return mlxsw_sp->router->rifs[rif_index];
6814 u16 mlxsw_sp_rif_index(const struct mlxsw_sp_rif *rif)
6816 return rif->rif_index;
6819 u16 mlxsw_sp_ipip_lb_rif_index(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6821 return lb_rif->common.rif_index;
6824 u16 mlxsw_sp_ipip_lb_ul_vr_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6826 u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(lb_rif->common.dev);
6827 struct mlxsw_sp_vr *ul_vr;
6829 ul_vr = mlxsw_sp_vr_get(lb_rif->common.mlxsw_sp, ul_tb_id, NULL);
6830 if (WARN_ON(IS_ERR(ul_vr)))
6836 u16 mlxsw_sp_ipip_lb_ul_rif_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6838 return lb_rif->ul_rif_id;
6841 int mlxsw_sp_rif_dev_ifindex(const struct mlxsw_sp_rif *rif)
6843 return rif->dev->ifindex;
6846 const struct net_device *mlxsw_sp_rif_dev(const struct mlxsw_sp_rif *rif)
6851 static struct mlxsw_sp_rif *
6852 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
6853 const struct mlxsw_sp_rif_params *params,
6854 struct netlink_ext_ack *extack)
6856 u32 tb_id = l3mdev_fib_table(params->dev);
6857 const struct mlxsw_sp_rif_ops *ops;
6858 struct mlxsw_sp_fid *fid = NULL;
6859 enum mlxsw_sp_rif_type type;
6860 struct mlxsw_sp_rif *rif;
6861 struct mlxsw_sp_vr *vr;
6865 type = mlxsw_sp_dev_rif_type(mlxsw_sp, params->dev);
6866 ops = mlxsw_sp->rif_ops_arr[type];
6868 vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id ? : RT_TABLE_MAIN, extack);
6870 return ERR_CAST(vr);
6873 err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
6875 NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interfaces");
6876 goto err_rif_index_alloc;
6879 rif = mlxsw_sp_rif_alloc(ops->rif_size, rif_index, vr->id, params->dev);
6885 mlxsw_sp->router->rifs[rif_index] = rif;
6886 rif->mlxsw_sp = mlxsw_sp;
6890 fid = ops->fid_get(rif, extack);
6899 ops->setup(rif, params);
6901 err = ops->configure(rif);
6905 for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++) {
6906 err = mlxsw_sp_mr_rif_add(vr->mr_table[i], rif);
6908 goto err_mr_rif_add;
6911 mlxsw_sp_rif_counters_alloc(rif);
6916 for (i--; i >= 0; i--)
6917 mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
6918 ops->deconfigure(rif);
6921 mlxsw_sp_fid_put(fid);
6923 mlxsw_sp->router->rifs[rif_index] = NULL;
6927 err_rif_index_alloc:
6929 mlxsw_sp_vr_put(mlxsw_sp, vr);
6930 return ERR_PTR(err);
6933 static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif)
6935 const struct mlxsw_sp_rif_ops *ops = rif->ops;
6936 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6937 struct mlxsw_sp_fid *fid = rif->fid;
6938 struct mlxsw_sp_vr *vr;
6941 mlxsw_sp_router_rif_gone_sync(mlxsw_sp, rif);
6942 vr = &mlxsw_sp->router->vrs[rif->vr_id];
6944 mlxsw_sp_rif_counters_free(rif);
6945 for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
6946 mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
6947 ops->deconfigure(rif);
6949 /* Loopback RIFs are not associated with a FID. */
6950 mlxsw_sp_fid_put(fid);
6951 mlxsw_sp->router->rifs[rif->rif_index] = NULL;
6955 mlxsw_sp_vr_put(mlxsw_sp, vr);
6958 void mlxsw_sp_rif_destroy_by_dev(struct mlxsw_sp *mlxsw_sp,
6959 struct net_device *dev)
6961 struct mlxsw_sp_rif *rif;
6963 mutex_lock(&mlxsw_sp->router->lock);
6964 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6967 mlxsw_sp_rif_destroy(rif);
6969 mutex_unlock(&mlxsw_sp->router->lock);
6973 mlxsw_sp_rif_subport_params_init(struct mlxsw_sp_rif_params *params,
6974 struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
6976 struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
6978 params->vid = mlxsw_sp_port_vlan->vid;
6979 params->lag = mlxsw_sp_port->lagged;
6981 params->lag_id = mlxsw_sp_port->lag_id;
6983 params->system_port = mlxsw_sp_port->local_port;
6986 static struct mlxsw_sp_rif_subport *
6987 mlxsw_sp_rif_subport_rif(const struct mlxsw_sp_rif *rif)
6989 return container_of(rif, struct mlxsw_sp_rif_subport, common);
6992 static struct mlxsw_sp_rif *
6993 mlxsw_sp_rif_subport_get(struct mlxsw_sp *mlxsw_sp,
6994 const struct mlxsw_sp_rif_params *params,
6995 struct netlink_ext_ack *extack)
6997 struct mlxsw_sp_rif_subport *rif_subport;
6998 struct mlxsw_sp_rif *rif;
7000 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, params->dev);
7002 return mlxsw_sp_rif_create(mlxsw_sp, params, extack);
7004 rif_subport = mlxsw_sp_rif_subport_rif(rif);
7005 refcount_inc(&rif_subport->ref_count);
7009 static void mlxsw_sp_rif_subport_put(struct mlxsw_sp_rif *rif)
7011 struct mlxsw_sp_rif_subport *rif_subport;
7013 rif_subport = mlxsw_sp_rif_subport_rif(rif);
7014 if (!refcount_dec_and_test(&rif_subport->ref_count))
7017 mlxsw_sp_rif_destroy(rif);
7021 mlxsw_sp_port_vlan_router_join(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan,
7022 struct net_device *l3_dev,
7023 struct netlink_ext_ack *extack)
7025 struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
7026 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
7027 struct mlxsw_sp_rif_params params = {
7030 u16 vid = mlxsw_sp_port_vlan->vid;
7031 struct mlxsw_sp_rif *rif;
7032 struct mlxsw_sp_fid *fid;
7035 mlxsw_sp_rif_subport_params_init(¶ms, mlxsw_sp_port_vlan);
7036 rif = mlxsw_sp_rif_subport_get(mlxsw_sp, ¶ms, extack);
7038 return PTR_ERR(rif);
7040 /* FID was already created, just take a reference */
7041 fid = rif->ops->fid_get(rif, extack);
7042 err = mlxsw_sp_fid_port_vid_map(fid, mlxsw_sp_port, vid);
7044 goto err_fid_port_vid_map;
7046 err = mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, false);
7048 goto err_port_vid_learning_set;
7050 err = mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid,
7051 BR_STATE_FORWARDING);
7053 goto err_port_vid_stp_set;
7055 mlxsw_sp_port_vlan->fid = fid;
7059 err_port_vid_stp_set:
7060 mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
7061 err_port_vid_learning_set:
7062 mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
7063 err_fid_port_vid_map:
7064 mlxsw_sp_fid_put(fid);
7065 mlxsw_sp_rif_subport_put(rif);
7070 __mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
7072 struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
7073 struct mlxsw_sp_fid *fid = mlxsw_sp_port_vlan->fid;
7074 struct mlxsw_sp_rif *rif = mlxsw_sp_fid_rif(fid);
7075 u16 vid = mlxsw_sp_port_vlan->vid;
7077 if (WARN_ON(mlxsw_sp_fid_type(fid) != MLXSW_SP_FID_TYPE_RFID))
7080 mlxsw_sp_port_vlan->fid = NULL;
7081 mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid, BR_STATE_BLOCKING);
7082 mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
7083 mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
7084 mlxsw_sp_fid_put(fid);
7085 mlxsw_sp_rif_subport_put(rif);
7089 mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
7091 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port_vlan->mlxsw_sp_port->mlxsw_sp;
7093 mutex_lock(&mlxsw_sp->router->lock);
7094 __mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
7095 mutex_unlock(&mlxsw_sp->router->lock);
7098 static int mlxsw_sp_inetaddr_port_vlan_event(struct net_device *l3_dev,
7099 struct net_device *port_dev,
7100 unsigned long event, u16 vid,
7101 struct netlink_ext_ack *extack)
7103 struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(port_dev);
7104 struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan;
7106 mlxsw_sp_port_vlan = mlxsw_sp_port_vlan_find_by_vid(mlxsw_sp_port, vid);
7107 if (WARN_ON(!mlxsw_sp_port_vlan))
7112 return mlxsw_sp_port_vlan_router_join(mlxsw_sp_port_vlan,
7115 __mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
7122 static int mlxsw_sp_inetaddr_port_event(struct net_device *port_dev,
7123 unsigned long event,
7124 struct netlink_ext_ack *extack)
7126 if (netif_is_bridge_port(port_dev) ||
7127 netif_is_lag_port(port_dev) ||
7128 netif_is_ovs_port(port_dev))
7131 return mlxsw_sp_inetaddr_port_vlan_event(port_dev, port_dev, event,
7132 MLXSW_SP_DEFAULT_VID, extack);
7135 static int __mlxsw_sp_inetaddr_lag_event(struct net_device *l3_dev,
7136 struct net_device *lag_dev,
7137 unsigned long event, u16 vid,
7138 struct netlink_ext_ack *extack)
7140 struct net_device *port_dev;
7141 struct list_head *iter;
7144 netdev_for_each_lower_dev(lag_dev, port_dev, iter) {
7145 if (mlxsw_sp_port_dev_check(port_dev)) {
7146 err = mlxsw_sp_inetaddr_port_vlan_event(l3_dev,
7158 static int mlxsw_sp_inetaddr_lag_event(struct net_device *lag_dev,
7159 unsigned long event,
7160 struct netlink_ext_ack *extack)
7162 if (netif_is_bridge_port(lag_dev))
7165 return __mlxsw_sp_inetaddr_lag_event(lag_dev, lag_dev, event,
7166 MLXSW_SP_DEFAULT_VID, extack);
7169 static int mlxsw_sp_inetaddr_bridge_event(struct mlxsw_sp *mlxsw_sp,
7170 struct net_device *l3_dev,
7171 unsigned long event,
7172 struct netlink_ext_ack *extack)
7174 struct mlxsw_sp_rif_params params = {
7177 struct mlxsw_sp_rif *rif;
7181 rif = mlxsw_sp_rif_create(mlxsw_sp, ¶ms, extack);
7183 return PTR_ERR(rif);
7186 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
7187 mlxsw_sp_rif_destroy(rif);
7194 static int mlxsw_sp_inetaddr_vlan_event(struct mlxsw_sp *mlxsw_sp,
7195 struct net_device *vlan_dev,
7196 unsigned long event,
7197 struct netlink_ext_ack *extack)
7199 struct net_device *real_dev = vlan_dev_real_dev(vlan_dev);
7200 u16 vid = vlan_dev_vlan_id(vlan_dev);
7202 if (netif_is_bridge_port(vlan_dev))
7205 if (mlxsw_sp_port_dev_check(real_dev))
7206 return mlxsw_sp_inetaddr_port_vlan_event(vlan_dev, real_dev,
7207 event, vid, extack);
7208 else if (netif_is_lag_master(real_dev))
7209 return __mlxsw_sp_inetaddr_lag_event(vlan_dev, real_dev, event,
7211 else if (netif_is_bridge_master(real_dev) && br_vlan_enabled(real_dev))
7212 return mlxsw_sp_inetaddr_bridge_event(mlxsw_sp, vlan_dev, event,
7218 static bool mlxsw_sp_rif_macvlan_is_vrrp4(const u8 *mac)
7220 u8 vrrp4[ETH_ALEN] = { 0x00, 0x00, 0x5e, 0x00, 0x01, 0x00 };
7221 u8 mask[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x00 };
7223 return ether_addr_equal_masked(mac, vrrp4, mask);
7226 static bool mlxsw_sp_rif_macvlan_is_vrrp6(const u8 *mac)
7228 u8 vrrp6[ETH_ALEN] = { 0x00, 0x00, 0x5e, 0x00, 0x02, 0x00 };
7229 u8 mask[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x00 };
7231 return ether_addr_equal_masked(mac, vrrp6, mask);
7234 static int mlxsw_sp_rif_vrrp_op(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
7235 const u8 *mac, bool adding)
7237 char ritr_pl[MLXSW_REG_RITR_LEN];
7238 u8 vrrp_id = adding ? mac[5] : 0;
7241 if (!mlxsw_sp_rif_macvlan_is_vrrp4(mac) &&
7242 !mlxsw_sp_rif_macvlan_is_vrrp6(mac))
7245 mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
7246 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7250 if (mlxsw_sp_rif_macvlan_is_vrrp4(mac))
7251 mlxsw_reg_ritr_if_vrrp_id_ipv4_set(ritr_pl, vrrp_id);
7253 mlxsw_reg_ritr_if_vrrp_id_ipv6_set(ritr_pl, vrrp_id);
7255 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7258 static int mlxsw_sp_rif_macvlan_add(struct mlxsw_sp *mlxsw_sp,
7259 const struct net_device *macvlan_dev,
7260 struct netlink_ext_ack *extack)
7262 struct macvlan_dev *vlan = netdev_priv(macvlan_dev);
7263 struct mlxsw_sp_rif *rif;
7266 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, vlan->lowerdev);
7268 NL_SET_ERR_MSG_MOD(extack, "macvlan is only supported on top of router interfaces");
7272 err = mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
7273 mlxsw_sp_fid_index(rif->fid), true);
7277 err = mlxsw_sp_rif_vrrp_op(mlxsw_sp, rif->rif_index,
7278 macvlan_dev->dev_addr, true);
7280 goto err_rif_vrrp_add;
7282 /* Make sure the bridge driver does not have this MAC pointing at
7285 if (rif->ops->fdb_del)
7286 rif->ops->fdb_del(rif, macvlan_dev->dev_addr);
7291 mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
7292 mlxsw_sp_fid_index(rif->fid), false);
7296 static void __mlxsw_sp_rif_macvlan_del(struct mlxsw_sp *mlxsw_sp,
7297 const struct net_device *macvlan_dev)
7299 struct macvlan_dev *vlan = netdev_priv(macvlan_dev);
7300 struct mlxsw_sp_rif *rif;
7302 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, vlan->lowerdev);
7303 /* If we do not have a RIF, then we already took care of
7304 * removing the macvlan's MAC during RIF deletion.
7308 mlxsw_sp_rif_vrrp_op(mlxsw_sp, rif->rif_index, macvlan_dev->dev_addr,
7310 mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
7311 mlxsw_sp_fid_index(rif->fid), false);
7314 void mlxsw_sp_rif_macvlan_del(struct mlxsw_sp *mlxsw_sp,
7315 const struct net_device *macvlan_dev)
7317 mutex_lock(&mlxsw_sp->router->lock);
7318 __mlxsw_sp_rif_macvlan_del(mlxsw_sp, macvlan_dev);
7319 mutex_unlock(&mlxsw_sp->router->lock);
7322 static int mlxsw_sp_inetaddr_macvlan_event(struct mlxsw_sp *mlxsw_sp,
7323 struct net_device *macvlan_dev,
7324 unsigned long event,
7325 struct netlink_ext_ack *extack)
7329 return mlxsw_sp_rif_macvlan_add(mlxsw_sp, macvlan_dev, extack);
7331 __mlxsw_sp_rif_macvlan_del(mlxsw_sp, macvlan_dev);
7338 static int mlxsw_sp_router_port_check_rif_addr(struct mlxsw_sp *mlxsw_sp,
7339 struct net_device *dev,
7340 const unsigned char *dev_addr,
7341 struct netlink_ext_ack *extack)
7343 struct mlxsw_sp_rif *rif;
7346 /* A RIF is not created for macvlan netdevs. Their MAC is used to
7349 if (netif_is_macvlan(dev) || netif_is_l3_master(dev))
7352 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++) {
7353 rif = mlxsw_sp->router->rifs[i];
7354 if (rif && rif->ops &&
7355 rif->ops->type == MLXSW_SP_RIF_TYPE_IPIP_LB)
7357 if (rif && rif->dev && rif->dev != dev &&
7358 !ether_addr_equal_masked(rif->dev->dev_addr, dev_addr,
7359 mlxsw_sp->mac_mask)) {
7360 NL_SET_ERR_MSG_MOD(extack, "All router interface MAC addresses must have the same prefix");
7368 static int __mlxsw_sp_inetaddr_event(struct mlxsw_sp *mlxsw_sp,
7369 struct net_device *dev,
7370 unsigned long event,
7371 struct netlink_ext_ack *extack)
7373 if (mlxsw_sp_port_dev_check(dev))
7374 return mlxsw_sp_inetaddr_port_event(dev, event, extack);
7375 else if (netif_is_lag_master(dev))
7376 return mlxsw_sp_inetaddr_lag_event(dev, event, extack);
7377 else if (netif_is_bridge_master(dev))
7378 return mlxsw_sp_inetaddr_bridge_event(mlxsw_sp, dev, event,
7380 else if (is_vlan_dev(dev))
7381 return mlxsw_sp_inetaddr_vlan_event(mlxsw_sp, dev, event,
7383 else if (netif_is_macvlan(dev))
7384 return mlxsw_sp_inetaddr_macvlan_event(mlxsw_sp, dev, event,
7390 static int mlxsw_sp_inetaddr_event(struct notifier_block *nb,
7391 unsigned long event, void *ptr)
7393 struct in_ifaddr *ifa = (struct in_ifaddr *) ptr;
7394 struct net_device *dev = ifa->ifa_dev->dev;
7395 struct mlxsw_sp_router *router;
7396 struct mlxsw_sp_rif *rif;
7399 /* NETDEV_UP event is handled by mlxsw_sp_inetaddr_valid_event */
7400 if (event == NETDEV_UP)
7403 router = container_of(nb, struct mlxsw_sp_router, inetaddr_nb);
7404 mutex_lock(&router->lock);
7405 rif = mlxsw_sp_rif_find_by_dev(router->mlxsw_sp, dev);
7406 if (!mlxsw_sp_rif_should_config(rif, dev, event))
7409 err = __mlxsw_sp_inetaddr_event(router->mlxsw_sp, dev, event, NULL);
7411 mutex_unlock(&router->lock);
7412 return notifier_from_errno(err);
7415 int mlxsw_sp_inetaddr_valid_event(struct notifier_block *unused,
7416 unsigned long event, void *ptr)
7418 struct in_validator_info *ivi = (struct in_validator_info *) ptr;
7419 struct net_device *dev = ivi->ivi_dev->dev;
7420 struct mlxsw_sp *mlxsw_sp;
7421 struct mlxsw_sp_rif *rif;
7424 mlxsw_sp = mlxsw_sp_lower_get(dev);
7428 mutex_lock(&mlxsw_sp->router->lock);
7429 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7430 if (!mlxsw_sp_rif_should_config(rif, dev, event))
7433 err = mlxsw_sp_router_port_check_rif_addr(mlxsw_sp, dev, dev->dev_addr,
7438 err = __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, ivi->extack);
7440 mutex_unlock(&mlxsw_sp->router->lock);
7441 return notifier_from_errno(err);
7444 struct mlxsw_sp_inet6addr_event_work {
7445 struct work_struct work;
7446 struct mlxsw_sp *mlxsw_sp;
7447 struct net_device *dev;
7448 unsigned long event;
7451 static void mlxsw_sp_inet6addr_event_work(struct work_struct *work)
7453 struct mlxsw_sp_inet6addr_event_work *inet6addr_work =
7454 container_of(work, struct mlxsw_sp_inet6addr_event_work, work);
7455 struct mlxsw_sp *mlxsw_sp = inet6addr_work->mlxsw_sp;
7456 struct net_device *dev = inet6addr_work->dev;
7457 unsigned long event = inet6addr_work->event;
7458 struct mlxsw_sp_rif *rif;
7461 mutex_lock(&mlxsw_sp->router->lock);
7463 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7464 if (!mlxsw_sp_rif_should_config(rif, dev, event))
7467 __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, NULL);
7469 mutex_unlock(&mlxsw_sp->router->lock);
7472 kfree(inet6addr_work);
7475 /* Called with rcu_read_lock() */
7476 static int mlxsw_sp_inet6addr_event(struct notifier_block *nb,
7477 unsigned long event, void *ptr)
7479 struct inet6_ifaddr *if6 = (struct inet6_ifaddr *) ptr;
7480 struct mlxsw_sp_inet6addr_event_work *inet6addr_work;
7481 struct net_device *dev = if6->idev->dev;
7482 struct mlxsw_sp_router *router;
7484 /* NETDEV_UP event is handled by mlxsw_sp_inet6addr_valid_event */
7485 if (event == NETDEV_UP)
7488 inet6addr_work = kzalloc(sizeof(*inet6addr_work), GFP_ATOMIC);
7489 if (!inet6addr_work)
7492 router = container_of(nb, struct mlxsw_sp_router, inet6addr_nb);
7493 INIT_WORK(&inet6addr_work->work, mlxsw_sp_inet6addr_event_work);
7494 inet6addr_work->mlxsw_sp = router->mlxsw_sp;
7495 inet6addr_work->dev = dev;
7496 inet6addr_work->event = event;
7498 mlxsw_core_schedule_work(&inet6addr_work->work);
7503 int mlxsw_sp_inet6addr_valid_event(struct notifier_block *unused,
7504 unsigned long event, void *ptr)
7506 struct in6_validator_info *i6vi = (struct in6_validator_info *) ptr;
7507 struct net_device *dev = i6vi->i6vi_dev->dev;
7508 struct mlxsw_sp *mlxsw_sp;
7509 struct mlxsw_sp_rif *rif;
7512 mlxsw_sp = mlxsw_sp_lower_get(dev);
7516 mutex_lock(&mlxsw_sp->router->lock);
7517 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7518 if (!mlxsw_sp_rif_should_config(rif, dev, event))
7521 err = mlxsw_sp_router_port_check_rif_addr(mlxsw_sp, dev, dev->dev_addr,
7526 err = __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, i6vi->extack);
7528 mutex_unlock(&mlxsw_sp->router->lock);
7529 return notifier_from_errno(err);
7532 static int mlxsw_sp_rif_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
7533 const char *mac, int mtu)
7535 char ritr_pl[MLXSW_REG_RITR_LEN];
7538 mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
7539 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7543 mlxsw_reg_ritr_mtu_set(ritr_pl, mtu);
7544 mlxsw_reg_ritr_if_mac_memcpy_to(ritr_pl, mac);
7545 mlxsw_reg_ritr_op_set(ritr_pl, MLXSW_REG_RITR_RIF_CREATE);
7546 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7550 mlxsw_sp_router_port_change_event(struct mlxsw_sp *mlxsw_sp,
7551 struct mlxsw_sp_rif *rif)
7553 struct net_device *dev = rif->dev;
7557 fid_index = mlxsw_sp_fid_index(rif->fid);
7559 err = mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, false);
7563 err = mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, dev->dev_addr,
7568 err = mlxsw_sp_rif_fdb_op(mlxsw_sp, dev->dev_addr, fid_index, true);
7570 goto err_rif_fdb_op;
7572 if (rif->mtu != dev->mtu) {
7573 struct mlxsw_sp_vr *vr;
7576 /* The RIF is relevant only to its mr_table instance, as unlike
7577 * unicast routing, in multicast routing a RIF cannot be shared
7578 * between several multicast routing tables.
7580 vr = &mlxsw_sp->router->vrs[rif->vr_id];
7581 for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
7582 mlxsw_sp_mr_rif_mtu_update(vr->mr_table[i],
7586 ether_addr_copy(rif->addr, dev->dev_addr);
7587 rif->mtu = dev->mtu;
7589 netdev_dbg(dev, "Updated RIF=%d\n", rif->rif_index);
7594 mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, rif->addr, rif->mtu);
7596 mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, true);
7600 static int mlxsw_sp_router_port_pre_changeaddr_event(struct mlxsw_sp_rif *rif,
7601 struct netdev_notifier_pre_changeaddr_info *info)
7603 struct netlink_ext_ack *extack;
7605 extack = netdev_notifier_info_to_extack(&info->info);
7606 return mlxsw_sp_router_port_check_rif_addr(rif->mlxsw_sp, rif->dev,
7607 info->dev_addr, extack);
7610 int mlxsw_sp_netdevice_router_port_event(struct net_device *dev,
7611 unsigned long event, void *ptr)
7613 struct mlxsw_sp *mlxsw_sp;
7614 struct mlxsw_sp_rif *rif;
7617 mlxsw_sp = mlxsw_sp_lower_get(dev);
7621 mutex_lock(&mlxsw_sp->router->lock);
7622 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7627 case NETDEV_CHANGEMTU:
7628 case NETDEV_CHANGEADDR:
7629 err = mlxsw_sp_router_port_change_event(mlxsw_sp, rif);
7631 case NETDEV_PRE_CHANGEADDR:
7632 err = mlxsw_sp_router_port_pre_changeaddr_event(rif, ptr);
7637 mutex_unlock(&mlxsw_sp->router->lock);
7641 static int mlxsw_sp_port_vrf_join(struct mlxsw_sp *mlxsw_sp,
7642 struct net_device *l3_dev,
7643 struct netlink_ext_ack *extack)
7645 struct mlxsw_sp_rif *rif;
7647 /* If netdev is already associated with a RIF, then we need to
7648 * destroy it and create a new one with the new virtual router ID.
7650 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
7652 __mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_DOWN,
7655 return __mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_UP, extack);
7658 static void mlxsw_sp_port_vrf_leave(struct mlxsw_sp *mlxsw_sp,
7659 struct net_device *l3_dev)
7661 struct mlxsw_sp_rif *rif;
7663 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
7666 __mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_DOWN, NULL);
7669 int mlxsw_sp_netdevice_vrf_event(struct net_device *l3_dev, unsigned long event,
7670 struct netdev_notifier_changeupper_info *info)
7672 struct mlxsw_sp *mlxsw_sp = mlxsw_sp_lower_get(l3_dev);
7675 /* We do not create a RIF for a macvlan, but only use it to
7676 * direct more MAC addresses to the router.
7678 if (!mlxsw_sp || netif_is_macvlan(l3_dev))
7681 mutex_lock(&mlxsw_sp->router->lock);
7683 case NETDEV_PRECHANGEUPPER:
7685 case NETDEV_CHANGEUPPER:
7686 if (info->linking) {
7687 struct netlink_ext_ack *extack;
7689 extack = netdev_notifier_info_to_extack(&info->info);
7690 err = mlxsw_sp_port_vrf_join(mlxsw_sp, l3_dev, extack);
7692 mlxsw_sp_port_vrf_leave(mlxsw_sp, l3_dev);
7696 mutex_unlock(&mlxsw_sp->router->lock);
7701 static int __mlxsw_sp_rif_macvlan_flush(struct net_device *dev,
7702 struct netdev_nested_priv *priv)
7704 struct mlxsw_sp_rif *rif = (struct mlxsw_sp_rif *)priv->data;
7706 if (!netif_is_macvlan(dev))
7709 return mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, dev->dev_addr,
7710 mlxsw_sp_fid_index(rif->fid), false);
7713 static int mlxsw_sp_rif_macvlan_flush(struct mlxsw_sp_rif *rif)
7715 struct netdev_nested_priv priv = {
7716 .data = (void *)rif,
7719 if (!netif_is_macvlan_port(rif->dev))
7722 netdev_warn(rif->dev, "Router interface is deleted. Upper macvlans will not work\n");
7723 return netdev_walk_all_upper_dev_rcu(rif->dev,
7724 __mlxsw_sp_rif_macvlan_flush, &priv);
7727 static void mlxsw_sp_rif_subport_setup(struct mlxsw_sp_rif *rif,
7728 const struct mlxsw_sp_rif_params *params)
7730 struct mlxsw_sp_rif_subport *rif_subport;
7732 rif_subport = mlxsw_sp_rif_subport_rif(rif);
7733 refcount_set(&rif_subport->ref_count, 1);
7734 rif_subport->vid = params->vid;
7735 rif_subport->lag = params->lag;
7737 rif_subport->lag_id = params->lag_id;
7739 rif_subport->system_port = params->system_port;
7742 static int mlxsw_sp_rif_subport_op(struct mlxsw_sp_rif *rif, bool enable)
7744 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7745 struct mlxsw_sp_rif_subport *rif_subport;
7746 char ritr_pl[MLXSW_REG_RITR_LEN];
7748 rif_subport = mlxsw_sp_rif_subport_rif(rif);
7749 mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_SP_IF,
7750 rif->rif_index, rif->vr_id, rif->dev->mtu);
7751 mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
7752 mlxsw_reg_ritr_sp_if_pack(ritr_pl, rif_subport->lag,
7753 rif_subport->lag ? rif_subport->lag_id :
7754 rif_subport->system_port,
7757 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7760 static int mlxsw_sp_rif_subport_configure(struct mlxsw_sp_rif *rif)
7764 err = mlxsw_sp_rif_subport_op(rif, true);
7768 err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7769 mlxsw_sp_fid_index(rif->fid), true);
7771 goto err_rif_fdb_op;
7773 mlxsw_sp_fid_rif_set(rif->fid, rif);
7777 mlxsw_sp_rif_subport_op(rif, false);
7781 static void mlxsw_sp_rif_subport_deconfigure(struct mlxsw_sp_rif *rif)
7783 struct mlxsw_sp_fid *fid = rif->fid;
7785 mlxsw_sp_fid_rif_set(fid, NULL);
7786 mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7787 mlxsw_sp_fid_index(fid), false);
7788 mlxsw_sp_rif_macvlan_flush(rif);
7789 mlxsw_sp_rif_subport_op(rif, false);
7792 static struct mlxsw_sp_fid *
7793 mlxsw_sp_rif_subport_fid_get(struct mlxsw_sp_rif *rif,
7794 struct netlink_ext_ack *extack)
7796 return mlxsw_sp_fid_rfid_get(rif->mlxsw_sp, rif->rif_index);
7799 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_subport_ops = {
7800 .type = MLXSW_SP_RIF_TYPE_SUBPORT,
7801 .rif_size = sizeof(struct mlxsw_sp_rif_subport),
7802 .setup = mlxsw_sp_rif_subport_setup,
7803 .configure = mlxsw_sp_rif_subport_configure,
7804 .deconfigure = mlxsw_sp_rif_subport_deconfigure,
7805 .fid_get = mlxsw_sp_rif_subport_fid_get,
7808 static int mlxsw_sp_rif_vlan_fid_op(struct mlxsw_sp_rif *rif,
7809 enum mlxsw_reg_ritr_if_type type,
7810 u16 vid_fid, bool enable)
7812 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7813 char ritr_pl[MLXSW_REG_RITR_LEN];
7815 mlxsw_reg_ritr_pack(ritr_pl, enable, type, rif->rif_index, rif->vr_id,
7817 mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
7818 mlxsw_reg_ritr_fid_set(ritr_pl, type, vid_fid);
7820 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7823 u8 mlxsw_sp_router_port(const struct mlxsw_sp *mlxsw_sp)
7825 return mlxsw_core_max_ports(mlxsw_sp->core) + 1;
7828 static int mlxsw_sp_rif_fid_configure(struct mlxsw_sp_rif *rif)
7830 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7831 u16 fid_index = mlxsw_sp_fid_index(rif->fid);
7834 err = mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index,
7839 err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7840 mlxsw_sp_router_port(mlxsw_sp), true);
7842 goto err_fid_mc_flood_set;
7844 err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7845 mlxsw_sp_router_port(mlxsw_sp), true);
7847 goto err_fid_bc_flood_set;
7849 err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7850 mlxsw_sp_fid_index(rif->fid), true);
7852 goto err_rif_fdb_op;
7854 mlxsw_sp_fid_rif_set(rif->fid, rif);
7858 mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7859 mlxsw_sp_router_port(mlxsw_sp), false);
7860 err_fid_bc_flood_set:
7861 mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7862 mlxsw_sp_router_port(mlxsw_sp), false);
7863 err_fid_mc_flood_set:
7864 mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index, false);
7868 static void mlxsw_sp_rif_fid_deconfigure(struct mlxsw_sp_rif *rif)
7870 u16 fid_index = mlxsw_sp_fid_index(rif->fid);
7871 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7872 struct mlxsw_sp_fid *fid = rif->fid;
7874 mlxsw_sp_fid_rif_set(fid, NULL);
7875 mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7876 mlxsw_sp_fid_index(fid), false);
7877 mlxsw_sp_rif_macvlan_flush(rif);
7878 mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7879 mlxsw_sp_router_port(mlxsw_sp), false);
7880 mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7881 mlxsw_sp_router_port(mlxsw_sp), false);
7882 mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index, false);
7885 static struct mlxsw_sp_fid *
7886 mlxsw_sp_rif_fid_fid_get(struct mlxsw_sp_rif *rif,
7887 struct netlink_ext_ack *extack)
7889 return mlxsw_sp_fid_8021d_get(rif->mlxsw_sp, rif->dev->ifindex);
7892 static void mlxsw_sp_rif_fid_fdb_del(struct mlxsw_sp_rif *rif, const char *mac)
7894 struct switchdev_notifier_fdb_info info;
7895 struct net_device *dev;
7897 dev = br_fdb_find_port(rif->dev, mac, 0);
7903 call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_BRIDGE, dev, &info.info,
7907 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_fid_ops = {
7908 .type = MLXSW_SP_RIF_TYPE_FID,
7909 .rif_size = sizeof(struct mlxsw_sp_rif),
7910 .configure = mlxsw_sp_rif_fid_configure,
7911 .deconfigure = mlxsw_sp_rif_fid_deconfigure,
7912 .fid_get = mlxsw_sp_rif_fid_fid_get,
7913 .fdb_del = mlxsw_sp_rif_fid_fdb_del,
7916 static struct mlxsw_sp_fid *
7917 mlxsw_sp_rif_vlan_fid_get(struct mlxsw_sp_rif *rif,
7918 struct netlink_ext_ack *extack)
7920 struct net_device *br_dev;
7924 if (is_vlan_dev(rif->dev)) {
7925 vid = vlan_dev_vlan_id(rif->dev);
7926 br_dev = vlan_dev_real_dev(rif->dev);
7927 if (WARN_ON(!netif_is_bridge_master(br_dev)))
7928 return ERR_PTR(-EINVAL);
7930 err = br_vlan_get_pvid(rif->dev, &vid);
7931 if (err < 0 || !vid) {
7932 NL_SET_ERR_MSG_MOD(extack, "Couldn't determine bridge PVID");
7933 return ERR_PTR(-EINVAL);
7937 return mlxsw_sp_fid_8021q_get(rif->mlxsw_sp, vid);
7940 static void mlxsw_sp_rif_vlan_fdb_del(struct mlxsw_sp_rif *rif, const char *mac)
7942 u16 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
7943 struct switchdev_notifier_fdb_info info;
7944 struct net_device *br_dev;
7945 struct net_device *dev;
7947 br_dev = is_vlan_dev(rif->dev) ? vlan_dev_real_dev(rif->dev) : rif->dev;
7948 dev = br_fdb_find_port(br_dev, mac, vid);
7954 call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_BRIDGE, dev, &info.info,
7958 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_vlan_emu_ops = {
7959 .type = MLXSW_SP_RIF_TYPE_VLAN,
7960 .rif_size = sizeof(struct mlxsw_sp_rif),
7961 .configure = mlxsw_sp_rif_fid_configure,
7962 .deconfigure = mlxsw_sp_rif_fid_deconfigure,
7963 .fid_get = mlxsw_sp_rif_vlan_fid_get,
7964 .fdb_del = mlxsw_sp_rif_vlan_fdb_del,
7967 static struct mlxsw_sp_rif_ipip_lb *
7968 mlxsw_sp_rif_ipip_lb_rif(struct mlxsw_sp_rif *rif)
7970 return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
7974 mlxsw_sp_rif_ipip_lb_setup(struct mlxsw_sp_rif *rif,
7975 const struct mlxsw_sp_rif_params *params)
7977 struct mlxsw_sp_rif_params_ipip_lb *params_lb;
7978 struct mlxsw_sp_rif_ipip_lb *rif_lb;
7980 params_lb = container_of(params, struct mlxsw_sp_rif_params_ipip_lb,
7982 rif_lb = mlxsw_sp_rif_ipip_lb_rif(rif);
7983 rif_lb->lb_config = params_lb->lb_config;
7987 mlxsw_sp1_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif)
7989 struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
7990 u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
7991 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7992 struct mlxsw_sp_vr *ul_vr;
7995 ul_vr = mlxsw_sp_vr_get(mlxsw_sp, ul_tb_id, NULL);
7997 return PTR_ERR(ul_vr);
7999 err = mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr->id, 0, true);
8001 goto err_loopback_op;
8003 lb_rif->ul_vr_id = ul_vr->id;
8004 lb_rif->ul_rif_id = 0;
8009 mlxsw_sp_vr_put(mlxsw_sp, ul_vr);
8013 static void mlxsw_sp1_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
8015 struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
8016 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8017 struct mlxsw_sp_vr *ul_vr;
8019 ul_vr = &mlxsw_sp->router->vrs[lb_rif->ul_vr_id];
8020 mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr->id, 0, false);
8023 mlxsw_sp_vr_put(mlxsw_sp, ul_vr);
8026 static const struct mlxsw_sp_rif_ops mlxsw_sp1_rif_ipip_lb_ops = {
8027 .type = MLXSW_SP_RIF_TYPE_IPIP_LB,
8028 .rif_size = sizeof(struct mlxsw_sp_rif_ipip_lb),
8029 .setup = mlxsw_sp_rif_ipip_lb_setup,
8030 .configure = mlxsw_sp1_rif_ipip_lb_configure,
8031 .deconfigure = mlxsw_sp1_rif_ipip_lb_deconfigure,
8034 const struct mlxsw_sp_rif_ops *mlxsw_sp1_rif_ops_arr[] = {
8035 [MLXSW_SP_RIF_TYPE_SUBPORT] = &mlxsw_sp_rif_subport_ops,
8036 [MLXSW_SP_RIF_TYPE_VLAN] = &mlxsw_sp_rif_vlan_emu_ops,
8037 [MLXSW_SP_RIF_TYPE_FID] = &mlxsw_sp_rif_fid_ops,
8038 [MLXSW_SP_RIF_TYPE_IPIP_LB] = &mlxsw_sp1_rif_ipip_lb_ops,
8042 mlxsw_sp_rif_ipip_lb_ul_rif_op(struct mlxsw_sp_rif *ul_rif, bool enable)
8044 struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
8045 char ritr_pl[MLXSW_REG_RITR_LEN];
8047 mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
8048 ul_rif->rif_index, ul_rif->vr_id, IP_MAX_MTU);
8049 mlxsw_reg_ritr_loopback_protocol_set(ritr_pl,
8050 MLXSW_REG_RITR_LOOPBACK_GENERIC);
8052 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
8055 static struct mlxsw_sp_rif *
8056 mlxsw_sp_ul_rif_create(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr,
8057 struct netlink_ext_ack *extack)
8059 struct mlxsw_sp_rif *ul_rif;
8063 err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
8065 NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interfaces");
8066 return ERR_PTR(err);
8069 ul_rif = mlxsw_sp_rif_alloc(sizeof(*ul_rif), rif_index, vr->id, NULL);
8071 return ERR_PTR(-ENOMEM);
8073 mlxsw_sp->router->rifs[rif_index] = ul_rif;
8074 ul_rif->mlxsw_sp = mlxsw_sp;
8075 err = mlxsw_sp_rif_ipip_lb_ul_rif_op(ul_rif, true);
8082 mlxsw_sp->router->rifs[rif_index] = NULL;
8084 return ERR_PTR(err);
8087 static void mlxsw_sp_ul_rif_destroy(struct mlxsw_sp_rif *ul_rif)
8089 struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
8091 mlxsw_sp_rif_ipip_lb_ul_rif_op(ul_rif, false);
8092 mlxsw_sp->router->rifs[ul_rif->rif_index] = NULL;
8096 static struct mlxsw_sp_rif *
8097 mlxsw_sp_ul_rif_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
8098 struct netlink_ext_ack *extack)
8100 struct mlxsw_sp_vr *vr;
8103 vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, extack);
8105 return ERR_CAST(vr);
8107 if (refcount_inc_not_zero(&vr->ul_rif_refcnt))
8110 vr->ul_rif = mlxsw_sp_ul_rif_create(mlxsw_sp, vr, extack);
8111 if (IS_ERR(vr->ul_rif)) {
8112 err = PTR_ERR(vr->ul_rif);
8113 goto err_ul_rif_create;
8117 refcount_set(&vr->ul_rif_refcnt, 1);
8122 mlxsw_sp_vr_put(mlxsw_sp, vr);
8123 return ERR_PTR(err);
8126 static void mlxsw_sp_ul_rif_put(struct mlxsw_sp_rif *ul_rif)
8128 struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
8129 struct mlxsw_sp_vr *vr;
8131 vr = &mlxsw_sp->router->vrs[ul_rif->vr_id];
8133 if (!refcount_dec_and_test(&vr->ul_rif_refcnt))
8137 mlxsw_sp_ul_rif_destroy(ul_rif);
8138 mlxsw_sp_vr_put(mlxsw_sp, vr);
8141 int mlxsw_sp_router_ul_rif_get(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
8144 struct mlxsw_sp_rif *ul_rif;
8147 mutex_lock(&mlxsw_sp->router->lock);
8148 ul_rif = mlxsw_sp_ul_rif_get(mlxsw_sp, ul_tb_id, NULL);
8149 if (IS_ERR(ul_rif)) {
8150 err = PTR_ERR(ul_rif);
8153 *ul_rif_index = ul_rif->rif_index;
8155 mutex_unlock(&mlxsw_sp->router->lock);
8159 void mlxsw_sp_router_ul_rif_put(struct mlxsw_sp *mlxsw_sp, u16 ul_rif_index)
8161 struct mlxsw_sp_rif *ul_rif;
8163 mutex_lock(&mlxsw_sp->router->lock);
8164 ul_rif = mlxsw_sp->router->rifs[ul_rif_index];
8165 if (WARN_ON(!ul_rif))
8168 mlxsw_sp_ul_rif_put(ul_rif);
8170 mutex_unlock(&mlxsw_sp->router->lock);
8174 mlxsw_sp2_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif)
8176 struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
8177 u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
8178 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8179 struct mlxsw_sp_rif *ul_rif;
8182 ul_rif = mlxsw_sp_ul_rif_get(mlxsw_sp, ul_tb_id, NULL);
8184 return PTR_ERR(ul_rif);
8186 err = mlxsw_sp_rif_ipip_lb_op(lb_rif, 0, ul_rif->rif_index, true);
8188 goto err_loopback_op;
8190 lb_rif->ul_vr_id = 0;
8191 lb_rif->ul_rif_id = ul_rif->rif_index;
8196 mlxsw_sp_ul_rif_put(ul_rif);
8200 static void mlxsw_sp2_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
8202 struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
8203 struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8204 struct mlxsw_sp_rif *ul_rif;
8206 ul_rif = mlxsw_sp_rif_by_index(mlxsw_sp, lb_rif->ul_rif_id);
8207 mlxsw_sp_rif_ipip_lb_op(lb_rif, 0, lb_rif->ul_rif_id, false);
8208 mlxsw_sp_ul_rif_put(ul_rif);
8211 static const struct mlxsw_sp_rif_ops mlxsw_sp2_rif_ipip_lb_ops = {
8212 .type = MLXSW_SP_RIF_TYPE_IPIP_LB,
8213 .rif_size = sizeof(struct mlxsw_sp_rif_ipip_lb),
8214 .setup = mlxsw_sp_rif_ipip_lb_setup,
8215 .configure = mlxsw_sp2_rif_ipip_lb_configure,
8216 .deconfigure = mlxsw_sp2_rif_ipip_lb_deconfigure,
8219 const struct mlxsw_sp_rif_ops *mlxsw_sp2_rif_ops_arr[] = {
8220 [MLXSW_SP_RIF_TYPE_SUBPORT] = &mlxsw_sp_rif_subport_ops,
8221 [MLXSW_SP_RIF_TYPE_VLAN] = &mlxsw_sp_rif_vlan_emu_ops,
8222 [MLXSW_SP_RIF_TYPE_FID] = &mlxsw_sp_rif_fid_ops,
8223 [MLXSW_SP_RIF_TYPE_IPIP_LB] = &mlxsw_sp2_rif_ipip_lb_ops,
8226 static int mlxsw_sp_rifs_init(struct mlxsw_sp *mlxsw_sp)
8228 u64 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
8230 mlxsw_sp->router->rifs = kcalloc(max_rifs,
8231 sizeof(struct mlxsw_sp_rif *),
8233 if (!mlxsw_sp->router->rifs)
8239 static void mlxsw_sp_rifs_fini(struct mlxsw_sp *mlxsw_sp)
8243 for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
8244 WARN_ON_ONCE(mlxsw_sp->router->rifs[i]);
8246 kfree(mlxsw_sp->router->rifs);
8250 mlxsw_sp_ipip_config_tigcr(struct mlxsw_sp *mlxsw_sp)
8252 char tigcr_pl[MLXSW_REG_TIGCR_LEN];
8254 mlxsw_reg_tigcr_pack(tigcr_pl, true, 0);
8255 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tigcr), tigcr_pl);
8258 static int mlxsw_sp_ipips_init(struct mlxsw_sp *mlxsw_sp)
8262 mlxsw_sp->router->ipip_ops_arr = mlxsw_sp_ipip_ops_arr;
8263 INIT_LIST_HEAD(&mlxsw_sp->router->ipip_list);
8265 err = mlxsw_sp_ipip_ecn_encap_init(mlxsw_sp);
8268 err = mlxsw_sp_ipip_ecn_decap_init(mlxsw_sp);
8272 return mlxsw_sp_ipip_config_tigcr(mlxsw_sp);
8275 static void mlxsw_sp_ipips_fini(struct mlxsw_sp *mlxsw_sp)
8277 WARN_ON(!list_empty(&mlxsw_sp->router->ipip_list));
8280 static void mlxsw_sp_router_fib_dump_flush(struct notifier_block *nb)
8282 struct mlxsw_sp_router *router;
8284 /* Flush pending FIB notifications and then flush the device's
8285 * table before requesting another dump. The FIB notification
8286 * block is unregistered, so no need to take RTNL.
8288 mlxsw_core_flush_owq();
8289 router = container_of(nb, struct mlxsw_sp_router, fib_nb);
8290 mlxsw_sp_router_fib_flush(router->mlxsw_sp);
8293 #ifdef CONFIG_IP_ROUTE_MULTIPATH
8294 static void mlxsw_sp_mp_hash_header_set(char *recr2_pl, int header)
8296 mlxsw_reg_recr2_outer_header_enables_set(recr2_pl, header, true);
8299 static void mlxsw_sp_mp_hash_field_set(char *recr2_pl, int field)
8301 mlxsw_reg_recr2_outer_header_fields_enable_set(recr2_pl, field, true);
8304 static void mlxsw_sp_mp4_hash_init(struct mlxsw_sp *mlxsw_sp, char *recr2_pl)
8306 struct net *net = mlxsw_sp_net(mlxsw_sp);
8307 bool only_l3 = !net->ipv4.sysctl_fib_multipath_hash_policy;
8309 mlxsw_sp_mp_hash_header_set(recr2_pl,
8310 MLXSW_REG_RECR2_IPV4_EN_NOT_TCP_NOT_UDP);
8311 mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_IPV4_EN_TCP_UDP);
8312 mlxsw_reg_recr2_ipv4_sip_enable(recr2_pl);
8313 mlxsw_reg_recr2_ipv4_dip_enable(recr2_pl);
8316 mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_EN_IPV4);
8317 mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV4_PROTOCOL);
8318 mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_SPORT);
8319 mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_DPORT);
8322 static void mlxsw_sp_mp6_hash_init(struct mlxsw_sp *mlxsw_sp, char *recr2_pl)
8324 bool only_l3 = !ip6_multipath_hash_policy(mlxsw_sp_net(mlxsw_sp));
8326 mlxsw_sp_mp_hash_header_set(recr2_pl,
8327 MLXSW_REG_RECR2_IPV6_EN_NOT_TCP_NOT_UDP);
8328 mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_IPV6_EN_TCP_UDP);
8329 mlxsw_reg_recr2_ipv6_sip_enable(recr2_pl);
8330 mlxsw_reg_recr2_ipv6_dip_enable(recr2_pl);
8331 mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV6_NEXT_HEADER);
8333 mlxsw_sp_mp_hash_field_set(recr2_pl,
8334 MLXSW_REG_RECR2_IPV6_FLOW_LABEL);
8336 mlxsw_sp_mp_hash_header_set(recr2_pl,
8337 MLXSW_REG_RECR2_TCP_UDP_EN_IPV6);
8338 mlxsw_sp_mp_hash_field_set(recr2_pl,
8339 MLXSW_REG_RECR2_TCP_UDP_SPORT);
8340 mlxsw_sp_mp_hash_field_set(recr2_pl,
8341 MLXSW_REG_RECR2_TCP_UDP_DPORT);
8345 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
8347 char recr2_pl[MLXSW_REG_RECR2_LEN];
8350 seed = jhash(mlxsw_sp->base_mac, sizeof(mlxsw_sp->base_mac), 0);
8351 mlxsw_reg_recr2_pack(recr2_pl, seed);
8352 mlxsw_sp_mp4_hash_init(mlxsw_sp, recr2_pl);
8353 mlxsw_sp_mp6_hash_init(mlxsw_sp, recr2_pl);
8355 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(recr2), recr2_pl);
8358 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
8364 static int mlxsw_sp_dscp_init(struct mlxsw_sp *mlxsw_sp)
8366 char rdpm_pl[MLXSW_REG_RDPM_LEN];
8369 MLXSW_REG_ZERO(rdpm, rdpm_pl);
8371 /* HW is determining switch priority based on DSCP-bits, but the
8372 * kernel is still doing that based on the ToS. Since there's a
8373 * mismatch in bits we need to make sure to translate the right
8374 * value ToS would observe, skipping the 2 least-significant ECN bits.
8376 for (i = 0; i < MLXSW_REG_RDPM_DSCP_ENTRY_REC_MAX_COUNT; i++)
8377 mlxsw_reg_rdpm_pack(rdpm_pl, i, rt_tos2priority(i << 2));
8379 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rdpm), rdpm_pl);
8382 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
8384 struct net *net = mlxsw_sp_net(mlxsw_sp);
8385 bool usp = net->ipv4.sysctl_ip_fwd_update_priority;
8386 char rgcr_pl[MLXSW_REG_RGCR_LEN];
8389 if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_RIFS))
8391 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
8393 mlxsw_reg_rgcr_pack(rgcr_pl, true, true);
8394 mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, max_rifs);
8395 mlxsw_reg_rgcr_usp_set(rgcr_pl, usp);
8396 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
8399 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
8401 char rgcr_pl[MLXSW_REG_RGCR_LEN];
8403 mlxsw_reg_rgcr_pack(rgcr_pl, false, false);
8404 mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
8407 static const struct mlxsw_sp_router_ll_ops mlxsw_sp_router_ll_basic_ops = {
8408 .ralta_write = mlxsw_sp_router_ll_basic_ralta_write,
8409 .ralst_write = mlxsw_sp_router_ll_basic_ralst_write,
8410 .raltb_write = mlxsw_sp_router_ll_basic_raltb_write,
8411 .fib_entry_op_ctx_size = sizeof(struct mlxsw_sp_fib_entry_op_ctx_basic),
8412 .fib_entry_pack = mlxsw_sp_router_ll_basic_fib_entry_pack,
8413 .fib_entry_act_remote_pack = mlxsw_sp_router_ll_basic_fib_entry_act_remote_pack,
8414 .fib_entry_act_local_pack = mlxsw_sp_router_ll_basic_fib_entry_act_local_pack,
8415 .fib_entry_act_ip2me_pack = mlxsw_sp_router_ll_basic_fib_entry_act_ip2me_pack,
8416 .fib_entry_act_ip2me_tun_pack = mlxsw_sp_router_ll_basic_fib_entry_act_ip2me_tun_pack,
8417 .fib_entry_commit = mlxsw_sp_router_ll_basic_fib_entry_commit,
8418 .fib_entry_is_committed = mlxsw_sp_router_ll_basic_fib_entry_is_committed,
8421 static int mlxsw_sp_router_ll_op_ctx_init(struct mlxsw_sp_router *router)
8423 size_t max_size = 0;
8426 for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++) {
8427 size_t size = router->proto_ll_ops[i]->fib_entry_op_ctx_size;
8429 if (size > max_size)
8432 router->ll_op_ctx = kzalloc(sizeof(*router->ll_op_ctx) + max_size,
8434 if (!router->ll_op_ctx)
8436 INIT_LIST_HEAD(&router->ll_op_ctx->fib_entry_priv_list);
8440 static void mlxsw_sp_router_ll_op_ctx_fini(struct mlxsw_sp_router *router)
8442 WARN_ON(!list_empty(&router->ll_op_ctx->fib_entry_priv_list));
8443 kfree(router->ll_op_ctx);
8446 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp,
8447 struct netlink_ext_ack *extack)
8449 struct mlxsw_sp_router *router;
8452 router = kzalloc(sizeof(*mlxsw_sp->router), GFP_KERNEL);
8455 mutex_init(&router->lock);
8456 mlxsw_sp->router = router;
8457 router->mlxsw_sp = mlxsw_sp;
8459 router->proto_ll_ops[MLXSW_SP_L3_PROTO_IPV4] = &mlxsw_sp_router_ll_basic_ops;
8460 router->proto_ll_ops[MLXSW_SP_L3_PROTO_IPV6] = &mlxsw_sp_router_ll_basic_ops;
8462 err = mlxsw_sp_router_ll_op_ctx_init(router);
8464 goto err_ll_op_ctx_init;
8466 INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_neighs_list);
8467 err = __mlxsw_sp_router_init(mlxsw_sp);
8469 goto err_router_init;
8471 err = mlxsw_sp_rifs_init(mlxsw_sp);
8475 err = mlxsw_sp_ipips_init(mlxsw_sp);
8477 goto err_ipips_init;
8479 err = rhashtable_init(&mlxsw_sp->router->nexthop_ht,
8480 &mlxsw_sp_nexthop_ht_params);
8482 goto err_nexthop_ht_init;
8484 err = rhashtable_init(&mlxsw_sp->router->nexthop_group_ht,
8485 &mlxsw_sp_nexthop_group_ht_params);
8487 goto err_nexthop_group_ht_init;
8489 INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_list);
8490 err = mlxsw_sp_lpm_init(mlxsw_sp);
8494 err = mlxsw_sp_mr_init(mlxsw_sp, &mlxsw_sp_mr_tcam_ops);
8498 err = mlxsw_sp_vrs_init(mlxsw_sp);
8502 err = mlxsw_sp_neigh_init(mlxsw_sp);
8504 goto err_neigh_init;
8506 err = mlxsw_sp_mp_hash_init(mlxsw_sp);
8508 goto err_mp_hash_init;
8510 err = mlxsw_sp_dscp_init(mlxsw_sp);
8514 INIT_WORK(&router->fib_event_work, mlxsw_sp_router_fib_event_work);
8515 INIT_LIST_HEAD(&router->fib_event_queue);
8516 spin_lock_init(&router->fib_event_queue_lock);
8518 router->inetaddr_nb.notifier_call = mlxsw_sp_inetaddr_event;
8519 err = register_inetaddr_notifier(&router->inetaddr_nb);
8521 goto err_register_inetaddr_notifier;
8523 router->inet6addr_nb.notifier_call = mlxsw_sp_inet6addr_event;
8524 err = register_inet6addr_notifier(&router->inet6addr_nb);
8526 goto err_register_inet6addr_notifier;
8528 mlxsw_sp->router->netevent_nb.notifier_call =
8529 mlxsw_sp_router_netevent_event;
8530 err = register_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8532 goto err_register_netevent_notifier;
8534 mlxsw_sp->router->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
8535 err = register_fib_notifier(mlxsw_sp_net(mlxsw_sp),
8536 &mlxsw_sp->router->fib_nb,
8537 mlxsw_sp_router_fib_dump_flush, extack);
8539 goto err_register_fib_notifier;
8543 err_register_fib_notifier:
8544 unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8545 err_register_netevent_notifier:
8546 unregister_inet6addr_notifier(&router->inet6addr_nb);
8547 err_register_inet6addr_notifier:
8548 unregister_inetaddr_notifier(&router->inetaddr_nb);
8549 err_register_inetaddr_notifier:
8550 mlxsw_core_flush_owq();
8551 WARN_ON(!list_empty(&router->fib_event_queue));
8554 mlxsw_sp_neigh_fini(mlxsw_sp);
8556 mlxsw_sp_vrs_fini(mlxsw_sp);
8558 mlxsw_sp_mr_fini(mlxsw_sp);
8560 mlxsw_sp_lpm_fini(mlxsw_sp);
8562 rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
8563 err_nexthop_group_ht_init:
8564 rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
8565 err_nexthop_ht_init:
8566 mlxsw_sp_ipips_fini(mlxsw_sp);
8568 mlxsw_sp_rifs_fini(mlxsw_sp);
8570 __mlxsw_sp_router_fini(mlxsw_sp);
8572 mlxsw_sp_router_ll_op_ctx_fini(router);
8574 mutex_destroy(&mlxsw_sp->router->lock);
8575 kfree(mlxsw_sp->router);
8579 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
8581 unregister_fib_notifier(mlxsw_sp_net(mlxsw_sp),
8582 &mlxsw_sp->router->fib_nb);
8583 unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8584 unregister_inet6addr_notifier(&mlxsw_sp->router->inet6addr_nb);
8585 unregister_inetaddr_notifier(&mlxsw_sp->router->inetaddr_nb);
8586 mlxsw_core_flush_owq();
8587 WARN_ON(!list_empty(&mlxsw_sp->router->fib_event_queue));
8588 mlxsw_sp_neigh_fini(mlxsw_sp);
8589 mlxsw_sp_vrs_fini(mlxsw_sp);
8590 mlxsw_sp_mr_fini(mlxsw_sp);
8591 mlxsw_sp_lpm_fini(mlxsw_sp);
8592 rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
8593 rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
8594 mlxsw_sp_ipips_fini(mlxsw_sp);
8595 mlxsw_sp_rifs_fini(mlxsw_sp);
8596 __mlxsw_sp_router_fini(mlxsw_sp);
8597 mlxsw_sp_router_ll_op_ctx_fini(mlxsw_sp->router);
8598 mutex_destroy(&mlxsw_sp->router->lock);
8599 kfree(mlxsw_sp->router);